diff options
Diffstat (limited to 'erts')
173 files changed, 11601 insertions, 3584 deletions
diff --git a/erts/aclocal.m4 b/erts/aclocal.m4 index 1e449bf874..b2d30449cd 100644 --- a/erts/aclocal.m4 +++ b/erts/aclocal.m4 @@ -1,7 +1,7 @@ dnl dnl %CopyrightBegin% dnl -dnl Copyright Ericsson AB 1998-2013. All Rights Reserved. +dnl Copyright Ericsson AB 1998-2015. All Rights Reserved. dnl dnl The contents of this file are subject to the Erlang Public License, dnl Version 1.1, (the "License"); you may not use this file except in @@ -246,31 +246,31 @@ lbl1: return 1; lbl2: return 2; -],ac_cv_prog_emu_cc=$CC,ac_cv_prog_emu_cc=no) +],ac_cv_prog_emu_cc="$CC",ac_cv_prog_emu_cc=no) -if test $ac_cv_prog_emu_cc = no; then +if test "$ac_cv_prog_emu_cc" = no; then for ac_progname in emu_cc.sh gcc-4.2 gcc; do IFS="${IFS= }"; ac_save_ifs="$IFS"; IFS=":" ac_dummy="$PATH" for ac_dir in $ac_dummy; do test -z "$ac_dir" && ac_dir=. - if test -f $ac_dir/$ac_progname; then - ac_cv_prog_emu_cc=$ac_dir/$ac_progname + if test -f "$ac_dir/$ac_progname"; then + ac_cv_prog_emu_cc="$ac_dir/$ac_progname" break fi done IFS="$ac_save_ifs" - if test $ac_cv_prog_emu_cc != no; then + if test "$ac_cv_prog_emu_cc" != no; then break fi done fi -if test $ac_cv_prog_emu_cc != no; then - save_CC=$CC +if test "$ac_cv_prog_emu_cc" != no; then + save_CC="$CC" save_CFLAGS=$CFLAGS save_CPPFLAGS=$CPPFLAGS - CC=$ac_cv_prog_emu_cc + CC="$ac_cv_prog_emu_cc" CFLAGS="" CPPFLAGS="" AC_TRY_COMPILE([],[ @@ -291,17 +291,17 @@ if test $ac_cv_prog_emu_cc != no; then return 1; lbl2: return 2; - ],ac_cv_prog_emu_cc=$CC,ac_cv_prog_emu_cc=no) + ],ac_cv_prog_emu_cc="$CC",ac_cv_prog_emu_cc=no) CC=$save_CC CFLAGS=$save_CFLAGS CPPFLAGS=$save_CPPFLAGS fi ]) -if test $ac_cv_prog_emu_cc = no; then +if test "$ac_cv_prog_emu_cc" = no; then AC_DEFINE(NO_JUMP_TABLE,[],[Defined if no found C compiler can handle jump tables]) - EMU_CC=$CC + EMU_CC="$CC" else - EMU_CC=$ac_cv_prog_emu_cc + EMU_CC="$ac_cv_prog_emu_cc" fi AC_SUBST(EMU_CC) ]) @@ -1019,24 +1019,226 @@ AC_SUBST(ERTS_INTERNAL_X_LIBS) ]) -AC_DEFUN(ETHR_CHK_SYNC_OP, +AC_DEFUN(ETHR_CHK_GCC_ATOMIC_OP__, [ - AC_MSG_CHECKING([for $3-bit $1()]) - case "$2" in - "1") sync_call="$1(&var);";; - "2") sync_call="$1(&var, ($4) 0);";; - "3") sync_call="$1(&var, ($4) 0, ($4) 0);";; + # $1 - atomic_op + + for atomic_bit_size in 32 64 128; do + case $atomic_bit_size in + 32) gcc_atomic_type="$gcc_atomic_type32";; + 64) gcc_atomic_type="$gcc_atomic_type64";; + 128) gcc_atomic_type="$gcc_atomic_type128";; + esac + gcc_atomic_lockfree="int x[[(2*__atomic_always_lock_free(sizeof($gcc_atomic_type), 0))-1]]" + case $1 in + __sync_add_and_fetch | __sync_fetch_and_and | __sync_fetch_and_or) + atomic_call="volatile $gcc_atomic_type var; $gcc_atomic_type res = $1(&var, ($gcc_atomic_type) 0);" + ;; + __sync_val_compare_and_swap) + atomic_call="volatile $gcc_atomic_type var; $gcc_atomic_type res = $1(&var, ($gcc_atomic_type) 0, ($gcc_atomic_type) 0);" + ;; + __atomic_store_n) + atomic_call="$gcc_atomic_lockfree; volatile $gcc_atomic_type var; $1(&var, ($gcc_atomic_type) 0, __ATOMIC_RELAXED); $1(&var, ($gcc_atomic_type) 0, __ATOMIC_RELEASE);" + ;; + __atomic_load_n) + atomic_call="$gcc_atomic_lockfree; volatile $gcc_atomic_type var; $gcc_atomic_type res = $1(&var, __ATOMIC_RELAXED); res = $1(&var, __ATOMIC_ACQUIRE);" + ;; + __atomic_add_fetch| __atomic_fetch_and | __atomic_fetch_or) + atomic_call="$gcc_atomic_lockfree; volatile $gcc_atomic_type var; $gcc_atomic_type res = $1(&var, ($gcc_atomic_type) 0, __ATOMIC_RELAXED); res = $1(&var, ($gcc_atomic_type) 0, __ATOMIC_ACQUIRE); res = $1(&var, ($gcc_atomic_type) 0, __ATOMIC_RELEASE);" + ;; + __atomic_compare_exchange_n) + atomic_call="$gcc_atomic_lockfree; volatile $gcc_atomic_type var; $gcc_atomic_type val; int res = $1(&var, &val, ($gcc_atomic_type) 0, 0, __ATOMIC_RELAXED, __ATOMIC_RELAXED); res = $1(&var, &val, ($gcc_atomic_type) 0, 0, __ATOMIC_ACQUIRE, __ATOMIC_ACQUIRE);" + ;; + *) + AC_MSG_ERROR([Internal error: missing implementation for $1]) + ;; + esac + eval atomic${atomic_bit_size}_call=\"$atomic_call\" + done + + AC_CACHE_CHECK([for 32-bit $1()], ethr_cv_32bit_$1, + [ + ethr_cv_32bit_$1=no + AC_TRY_LINK([], [$atomic32_call], [ethr_cv_32bit_$1=yes]) + ]) + AC_CACHE_CHECK([for 64-bit $1()], ethr_cv_64bit_$1, + [ + ethr_cv_64bit_$1=no + AC_TRY_LINK([], [$atomic64_call], [ethr_cv_64bit_$1=yes]) + ]) + AC_CACHE_CHECK([for 128-bit $1()], ethr_cv_128bit_$1, + [ + ethr_cv_128bit_$1=no + AC_TRY_LINK([], [$atomic128_call], [ethr_cv_128bit_$1=yes]) + ]) + + case $ethr_cv_128bit_$1-$ethr_cv_64bit_$1-$ethr_cv_32bit_$1 in + no-no-no) + have_atomic_ops=0;; + no-no-yes) + have_atomic_ops=4;; + no-yes-no) + have_atomic_ops=8;; + no-yes-yes) + have_atomic_ops=12;; + yes-no-no) + have_atomic_ops=16;; + yes-no-yes) + have_atomic_ops=20;; + yes-yes-no) + have_atomic_ops=24;; + yes-yes-yes) + have_atomic_ops=28;; + esac + AC_DEFINE_UNQUOTED([ETHR_HAVE_$1], [$have_atomic_ops], [Define as a bitmask corresponding to the word sizes that $1() can handle on your system]) +]) + +AC_DEFUN(ETHR_CHK_IF_NOOP, +[ + ethr_test_filename="chk_if_$1$3_noop_config1test.$$" + cat > "${ethr_test_filename}.c" <<EOF +int +my_test(void) +{ + $1$2; + return 0; +} +EOF + $CC -O3 $ETHR_DEFS -c "${ethr_test_filename}.c" -o "${ethr_test_filename}1.o" + cat > "${ethr_test_filename}.c" <<EOF +int +my_test(void) +{ + ; + return 0; +} +EOF + $CC -O3 $ETHR_DEFS -c "${ethr_test_filename}.c" -o "${ethr_test_filename}2.o" + if diff "${ethr_test_filename}1.o" "${ethr_test_filename}2.o" >/dev/null 2>&1; then + ethr_$1$3_noop=yes + else + ethr_$1$3_noop=no + fi + rm -f "${ethr_test_filename}.c" "${ethr_test_filename}1.o" "${ethr_test_filename}2.o" +]) + +AC_DEFUN(ETHR_CHK_GCC_ATOMIC_OPS, +[ + AC_CHECK_SIZEOF(short) + AC_CHECK_SIZEOF(int) + AC_CHECK_SIZEOF(long) + AC_CHECK_SIZEOF(long long) + AC_CHECK_SIZEOF(__int128_t) + + if test "$ac_cv_sizeof_short" = "4"; then + gcc_atomic_type32="short" + elif test "$ac_cv_sizeof_int" = "4"; then + gcc_atomic_type32="int" + elif test "$ac_cv_sizeof_long" = "4"; then + gcc_atomic_type32="long" + else + AC_MSG_ERROR([No 32-bit type found]) + fi + + if test "$ac_cv_sizeof_int" = "8"; then + gcc_atomic_type64="int" + elif test "$ac_cv_sizeof_long" = "8"; then + gcc_atomic_type64="long" + elif test "$ac_cv_sizeof_long_long" = "8"; then + gcc_atomic_type64="long long" + else + AC_MSG_ERROR([No 64-bit type found]) + fi + + if test "$ac_cv_sizeof___int128_t" = "16"; then + gcc_atomic_type128="__int128_t" + else + gcc_atomic_type128="#error " + fi + AC_CACHE_CHECK([for a working __sync_synchronize()], ethr_cv___sync_synchronize, + [ + ethr_cv___sync_synchronize=no + AC_TRY_LINK([], + [ __sync_synchronize(); ], + [ethr_cv___sync_synchronize=yes]) + if test $ethr_cv___sync_synchronize = yes; then + # + # Old gcc versions on at least x86 have a buggy + # __sync_synchronize() which does not emit a + # memory barrier. We try to detect this by + # compiling to assembly with and without + # __sync_synchronize() and compare the results. + # + ETHR_CHK_IF_NOOP(__sync_synchronize, [()], []) + if test $ethr___sync_synchronize_noop = yes; then + # Got a buggy implementation of + # __sync_synchronize... + ethr_cv___sync_synchronize="no; buggy implementation" + fi + fi + ]) + + if test "$ethr_cv___sync_synchronize" = "yes"; then + have_sync_synchronize_value="~0" + else + have_sync_synchronize_value="0" + fi + AC_DEFINE_UNQUOTED([ETHR_HAVE___sync_synchronize], [$have_sync_synchronize_value], [Define as a bitmask corresponding to the word sizes that __sync_synchronize() can handle on your system]) + + ETHR_CHK_GCC_ATOMIC_OP__(__sync_add_and_fetch) + ETHR_CHK_GCC_ATOMIC_OP__(__sync_fetch_and_and) + ETHR_CHK_GCC_ATOMIC_OP__(__sync_fetch_and_or) + ETHR_CHK_GCC_ATOMIC_OP__(__sync_val_compare_and_swap) + + ETHR_CHK_GCC_ATOMIC_OP__(__atomic_store_n) + ETHR_CHK_GCC_ATOMIC_OP__(__atomic_load_n) + ETHR_CHK_GCC_ATOMIC_OP__(__atomic_add_fetch) + ETHR_CHK_GCC_ATOMIC_OP__(__atomic_fetch_and) + ETHR_CHK_GCC_ATOMIC_OP__(__atomic_fetch_or) + ETHR_CHK_GCC_ATOMIC_OP__(__atomic_compare_exchange_n) + + ethr_have_gcc_native_atomics=no + ethr_arm_dbm_instr_val=0 + case "$GCC-$host_cpu" in + yes-arm*) + AC_CACHE_CHECK([for ARM DMB instruction], ethr_cv_arm_dbm_instr, + [ + ethr_cv_arm_dbm_instr=no + AC_TRY_LINK([], + [ + __asm__ __volatile__("dmb sy" : : : "memory"); + __asm__ __volatile__("dmb st" : : : "memory"); + ], + [ethr_cv_arm_dbm_instr=yes]) + ]) + if test $ethr_cv_arm_dbm_instr = yes; then + ethr_arm_dbm_instr_val=1 + test $ethr_cv_64bit___atomic_compare_exchange_n = yes && + ethr_have_gcc_native_atomics=yes + fi;; + *) + ;; esac - have_sync_op=no - AC_TRY_LINK([], - [ - $4 res; - volatile $4 var; - res = $sync_call - ], - [have_sync_op=yes]) - test $have_sync_op = yes && $5 - AC_MSG_RESULT([$have_sync_op]) + AC_DEFINE_UNQUOTED([ETHR_HAVE_GCC_ASM_ARM_DMB_INSTRUCTION], [$ethr_arm_dbm_instr_val], [Define as a boolean indicating whether you have a gcc compatible compiler capable of generating the ARM DMB instruction, and are compiling for an ARM processor with ARM DMB instruction support, or not]) + test $ethr_cv_32bit___sync_val_compare_and_swap = yes && + ethr_have_gcc_native_atomics=yes + test $ethr_cv_64bit___sync_val_compare_and_swap = yes && + ethr_have_gcc_native_atomics=yes + if test "$ethr_cv___sync_synchronize" = "yes"; then + test $ethr_cv_64bit___atomic_compare_exchange_n = yes && + ethr_have_gcc_native_atomics=yes + test $ethr_cv_32bit___atomic_compare_exchange_n = yes && + ethr_have_gcc_native_atomics=yes + fi + ethr_have_gcc_atomic_builtins=0 + if test $ethr_have_gcc_native_atomics = yes; then + ethr_native_atomic_implementation=gcc_sync + test $ethr_cv_32bit___atomic_compare_exchange_n = yes && ethr_have_gcc_atomic_builtins=1 + test $ethr_cv_64bit___atomic_compare_exchange_n = yes && ethr_have_gcc_atomic_builtins=1 + test $ethr_have_gcc_atomic_builtins = 1 && ethr_native_atomic_implementation=gcc_atomic_sync + fi + AC_DEFINE_UNQUOTED([ETHR_HAVE_GCC___ATOMIC_BUILTINS], [$ethr_have_gcc_atomic_builtins], [Define as a boolean indicating whether you have a gcc __atomic builtins or not]) + test $ethr_have_gcc_native_atomics = yes && ethr_have_native_atomics=yes ]) AC_DEFUN(ETHR_CHK_INTERLOCKED, @@ -1116,6 +1318,16 @@ AC_ARG_ENABLE(prefer-gcc-native-ethr-impls, test $enable_prefer_gcc_native_ethr_impls = yes && AC_DEFINE(ETHR_PREFER_GCC_NATIVE_IMPLS, 1, [Define if you prefer gcc native ethread implementations]) +AC_ARG_ENABLE(trust-gcc-atomic-builtins-memory-barriers, + AS_HELP_STRING([--enable-trust-gcc-atomic-builtins-memory-barriers], + [trust gcc atomic builtins memory barriers]), +[ case "$enableval" in + yes) trust_gcc_atomic_builtins_mbs=1 ;; + *) trust_gcc_atomic_builtins_mbs=0 ;; + esac ], trust_gcc_atomic_builtins_mbs=0) + +AC_DEFINE_UNQUOTED(ETHR_TRUST_GCC_ATOMIC_BUILTINS_MEMORY_BARRIERS, [$trust_gcc_atomic_builtins_mbs], [Define as a boolean indicating whether you trust gcc's __atomic_* builtins memory barrier implementations, or not]) + AC_ARG_WITH(libatomic_ops, AS_HELP_STRING([--with-libatomic_ops=PATH], [specify and prefer usage of libatomic_ops in the ethread library])) @@ -1147,6 +1359,7 @@ if test "x$erl_monotonic_clock_id" != "x"; then AC_DEFINE_UNQUOTED(ETHR_MONOTONIC_CLOCK_ID, [$erl_monotonic_clock_id], [Define to the monotonic clock id to use]) fi +ethr_native_atomic_implementation=none ethr_have_native_atomics=no ethr_have_native_spinlock=no ETHR_THR_LIB_BASE="$THR_LIB_NAME" @@ -1231,7 +1444,10 @@ case "$THR_LIB_NAME" in ETHR_CHK_INTERLOCKED([_InterlockedCompareExchange128], [4], [__int64], AC_DEFINE_UNQUOTED(ETHR_HAVE__INTERLOCKEDCOMPAREEXCHANGE128, 1, [Define if you have _InterlockedCompareExchange128()])) fi - test "$ethr_have_native_atomics" = "yes" && ethr_have_native_spinlock=yes + if test "$ethr_have_native_atomics" = "yes"; then + ethr_native_atomic_implementation=windows + ethr_have_native_spinlock=yes + fi ;; pthread|ose_threads) @@ -1524,56 +1740,62 @@ case "$THR_LIB_NAME" in AC_MSG_RESULT([$linux_futex]) test $linux_futex = yes && AC_DEFINE(ETHR_HAVE_LINUX_FUTEX, 1, [Define if you have a linux futex implementation.]) - fi - - AC_CHECK_SIZEOF(int) - AC_CHECK_SIZEOF(long) - AC_CHECK_SIZEOF(long long) - AC_CHECK_SIZEOF(__int128_t) - - if test "$ac_cv_sizeof_int" = "4"; then - int32="int" - elif test "$ac_cv_sizeof_long" = "4"; then - int32="long" - elif test "$ac_cv_sizeof_long_long" = "4"; then - int32="long long" - else - AC_MSG_ERROR([No 32-bit type found]) - fi + pthread_setname=no + AC_MSG_CHECKING([for pthread_setname_np]) + old_CFLAGS=$CFLAGS + CFLAGS="$CFLAGS -Werror" + AC_TRY_LINK([#define __USE_GNU + #include <pthread.h>], + [pthread_setname_np(pthread_self(), "name");], + pthread_setname=linux) + AC_TRY_LINK([#define __USE_GNU + #include <pthread.h>], + [pthread_set_name_np(pthread_self(), "name");], + pthread_setname=bsd) + AC_TRY_LINK([#define _DARWIN_C_SOURCE + #include <pthread.h>], + [pthread_setname_np("name");], + pthread_setname=darwin) + AC_MSG_RESULT([$pthread_setname]) + case $pthread_setname in + linux) AC_DEFINE(ETHR_HAVE_PTHREAD_SETNAME_NP_2, 1, + [Define if you have linux style pthread_setname_np]);; + bsd) AC_DEFINE(ETHR_HAVE_PTHREAD_SET_NAME_NP_2, 1, + [Define if you have bsd style pthread_set_name_np]);; + darwin) AC_DEFINE(ETHR_HAVE_PTHREAD_SETNAME_NP_1, 1, + [Define if you have darwin style pthread_setname_np]);; + *) ;; + esac - if test "$ac_cv_sizeof_int" = "8"; then - int64="int" - elif test "$ac_cv_sizeof_long" = "8"; then - int64="long" - elif test "$ac_cv_sizeof_long_long" = "8"; then - int64="long long" - else - AC_MSG_ERROR([No 64-bit type found]) - fi + pthread_getname=no + AC_MSG_CHECKING([for pthread_getname_np]) + AC_TRY_LINK([#define __USE_GNU + #define _DARWIN_C_SOURCE + #include <pthread.h>], + [char buff[256]; pthread_getname_np(pthread_self(), buff, 256);], + pthread_getname=normal) + AC_TRY_LINK([#define __USE_GNU + #define _DARWIN_C_SOURCE + #include <pthread.h>], + [char buff[256]; pthread_getname_np(pthread_self(), buff);], + pthread_getname=ibm) + AC_MSG_RESULT([$pthread_getname]) + case $pthread_getname in + linux) AC_DEFINE(ETHR_HAVE_PTHREAD_GETNAME_NP_3, 1, + [Define if you have linux style pthread_getname_np]);; + ibm) AC_DEFINE(ETHR_HAVE_PTHREAD_GETNAME_NP_2, 1, + [Define if you have ibm style pthread_getname_np]);; + *) ;; + esac + CFLAGS=$old_CFLAGS - int128=no - if test "$ac_cv_sizeof___int128_t" = "16"; then - int128="__int128_t" - fi + fi ## test "x$THR_LIB_NAME" = "xpthread" if test "X$disable_native_ethr_impls" = "Xyes"; then ethr_have_native_atomics=no else - ETHR_CHK_SYNC_OP([__sync_val_compare_and_swap], [3], [32], [$int32], AC_DEFINE(ETHR_HAVE___SYNC_VAL_COMPARE_AND_SWAP32, 1, [Define if you have __sync_val_compare_and_swap() for 32-bit integers])) - test "$have_sync_op" = "yes" && ethr_have_native_atomics=yes - ETHR_CHK_SYNC_OP([__sync_add_and_fetch], [2], [32], [$int32], AC_DEFINE(ETHR_HAVE___SYNC_ADD_AND_FETCH32, 1, [Define if you have __sync_add_and_fetch() for 32-bit integers])) - ETHR_CHK_SYNC_OP([__sync_fetch_and_and], [2], [32], [$int32], AC_DEFINE(ETHR_HAVE___SYNC_FETCH_AND_AND32, 1, [Define if you have __sync_fetch_and_and() for 32-bit integers])) - ETHR_CHK_SYNC_OP([__sync_fetch_and_or], [2], [32], [$int32], AC_DEFINE(ETHR_HAVE___SYNC_FETCH_AND_OR32, 1, [Define if you have __sync_fetch_and_or() for 32-bit integers])) - - ETHR_CHK_SYNC_OP([__sync_val_compare_and_swap], [3], [64], [$int64], AC_DEFINE(ETHR_HAVE___SYNC_VAL_COMPARE_AND_SWAP64, 1, [Define if you have __sync_val_compare_and_swap() for 64-bit integers])) - test "$have_sync_op" = "yes" && ethr_have_native_atomics=yes - ETHR_CHK_SYNC_OP([__sync_add_and_fetch], [2], [64], [$int64], AC_DEFINE(ETHR_HAVE___SYNC_ADD_AND_FETCH64, 1, [Define if you have __sync_add_and_fetch() for 64-bit integers])) - ETHR_CHK_SYNC_OP([__sync_fetch_and_and], [2], [64], [$int64], AC_DEFINE(ETHR_HAVE___SYNC_FETCH_AND_AND64, 1, [Define if you have __sync_fetch_and_and() for 64-bit integers])) - ETHR_CHK_SYNC_OP([__sync_fetch_and_or], [2], [64], [$int64], AC_DEFINE(ETHR_HAVE___SYNC_FETCH_AND_OR64, 1, [Define if you have __sync_fetch_and_or() for 64-bit integers])) - - if test $int128 != no; then - ETHR_CHK_SYNC_OP([__sync_val_compare_and_swap], [3], [128], [$int128], AC_DEFINE(ETHR_HAVE___SYNC_VAL_COMPARE_AND_SWAP128, 1, [Define if you have __sync_val_compare_and_swap() for 128-bit integers])) - fi + + ETHR_CHK_GCC_ATOMIC_OPS([]) AC_MSG_CHECKING([for a usable libatomic_ops implementation]) case "x$with_libatomic_ops" in @@ -1623,6 +1845,7 @@ case "$THR_LIB_NAME" in #endif ], [ethr_have_native_atomics=yes + ethr_native_atomic_implementation=libatomic_ops ethr_have_libatomic_ops=yes]) AC_MSG_RESULT([$ethr_have_libatomic_ops]) if test $ethr_have_libatomic_ops = yes; then @@ -1654,15 +1877,19 @@ case "$THR_LIB_NAME" in *) AC_MSG_ERROR([Unsupported Sparc memory order: $with_sparc_memory_order]);; esac + ethr_native_atomic_implementation=ethread ethr_have_native_atomics=yes;; i86pc | i*86 | x86_64 | amd64) if test "$enable_x86_out_of_order" = "yes"; then AC_DEFINE(ETHR_X86_OUT_OF_ORDER, 1, [Define if x86/x86_64 out of order instructions should be synchronized]) fi + ethr_native_atomic_implementation=ethread ethr_have_native_atomics=yes;; macppc | ppc | powerpc | "Power Macintosh") + ethr_native_atomic_implementation=ethread ethr_have_native_atomics=yes;; tile) + ethr_native_atomic_implementation=ethread ethr_have_native_atomics=yes;; *) ;; @@ -1996,35 +2223,28 @@ case $erl_gethrvtime in exit(0); return 0; } ], - erl_clock_gettime=yes, - erl_clock_gettime=no, + erl_clock_gettime_cpu_time=yes, + erl_clock_gettime_cpu_time=no, [ case X$erl_xcomp_clock_gettime_cpu_time in - X) erl_clock_gettime=cross;; - Xyes|Xno) erl_clock_gettime=$erl_xcomp_clock_gettime_cpu_time;; + X) erl_clock_gettime_cpu_time=cross;; + Xyes|Xno) erl_clock_gettime_cpu_time=$erl_xcomp_clock_gettime_cpu_time;; *) AC_MSG_ERROR([Bad erl_xcomp_clock_gettime_cpu_time value: $erl_xcomp_clock_gettime_cpu_time]);; esac ]) LIBS=$save_libs - case $host_os in - linux*) - AC_MSG_RESULT([no; not stable]) + AC_MSG_RESULT($erl_clock_gettime_cpu_time) + case $erl_clock_gettime_cpu_time in + yes) + AC_DEFINE(HAVE_CLOCK_GETTIME_CPU_TIME,[], + [define if clock_gettime() works for getting process time]) + LIBRT=-lrt + ;; + cross) + erl_clock_gettime_cpu_time=no + AC_MSG_WARN([result no guessed because of cross compilation]) ;; *) - AC_MSG_RESULT($erl_clock_gettime) - case $erl_clock_gettime in - yes) - AC_DEFINE(HAVE_CLOCK_GETTIME,[], - [define if clock_gettime() works for getting process time]) - LIBRT=-lrt - ;; - cross) - erl_clock_gettime=no - AC_MSG_WARN([result no guessed because of cross compilation]) - ;; - *) - ;; - esac ;; esac ;; diff --git a/erts/autoconf/vxworks/sed.general b/erts/autoconf/vxworks/sed.general index efa4e99054..c82b7348cf 100644 --- a/erts/autoconf/vxworks/sed.general +++ b/erts/autoconf/vxworks/sed.general @@ -78,8 +78,6 @@ s|@WITH_SCTP@|| # HiPE s|@HIPE_ENABLED@|| -s|@PERFCTR_PATH@|| -s|@USE_PERFCTR@|| # m4 s|@OPSYS@|noopsys| diff --git a/erts/configure.in b/erts/configure.in index 1676d3d216..481dfe405e 100644 --- a/erts/configure.in +++ b/erts/configure.in @@ -2,7 +2,7 @@ dnl Process this file with autoconf to produce a configure script. -*-m4-*- dnl %CopyrightBegin% dnl -dnl Copyright Ericsson AB 1997-2014. All Rights Reserved. +dnl Copyright Ericsson AB 1997-2015. All Rights Reserved. dnl dnl The contents of this file are subject to the Erlang Public License, dnl Version 1.1, (the "License"); you may not use this file except in @@ -264,24 +264,6 @@ AS_HELP_STRING([--enable-m32-build], esac ],enable_m32_build=no) -AC_SUBST(PERFCTR_PATH) -AC_ARG_WITH(perfctr, -AS_HELP_STRING([--with-perfctr=PATH], - [specify location of perfctr include and lib]) -AS_HELP_STRING([--without-perfctr], [don't use perfctr (default)])) - -if test "x$with_perfctr" = "xno" -o "x$with_perfctr" = "x" ; then - PERFCTR_PATH= -else - if test ! -f "$with_perfctr/usr.lib/libperfctr.a" ; then - AC_MSG_ERROR(Invalid path to option --with-perfctr=PATH) - fi - PERFCTR_PATH="$with_perfctr" - AC_DEFINE(USE_PERFCTR,[1], - [Define to enable hrvtime() on Linux systems with perfctr extension]) -fi - - AC_ARG_WITH(dynamic-trace, AS_HELP_STRING([--with-dynamic-trace={dtrace|systemtap}], [specify use of dynamic trace framework, dtrace or systemtap]) @@ -1123,10 +1105,31 @@ if test $ERTS_BUILD_SMP_EMU = yes; then case "$ethr_have_native_atomics-$smp_require_native_atomics-$ethr_have_native_spinlock" in yes-*) + if test "$ethr_native_atomic_implementation" = "gcc_sync"; then + test -f "$ERL_TOP/erts/CONF_INFO" || + echo "" > "$ERL_TOP/erts/CONF_INFO" + cat >> $ERL_TOP/erts/CONF_INFO <<EOF + + WARNING: + Only gcc's __sync_* builtins available for + atomic memory access. This will cause lots + of expensive and unnecessary memory barrier + instructions to be issued which will make + the performance of the runtime system + suffer. You are *strongly* advised to + upgrade to a gcc version that supports the + __atomic_* builtins (at least gcc version + 4.7) or build with libatomic_ops. See the + "Atomic Memory Operations and the VM" + chapter of \$ERL_TOP/HOWTO/INSTALL.md for + more information. + +EOF + fi ;; no-yes-*) - AC_MSG_ERROR([No native atomic implementation found. See Configuring section in INSTALL.md for more information.]) + AC_MSG_ERROR([No native atomic implementation found. See the \"Atomic Memory Operations and the VM\" chapter of \$ERL_TOP/HOWTO/INSTALL.md for more information.]) ;; no-no-yes) @@ -1784,7 +1787,7 @@ if test x"$ac_cv_header_netinet_sctp_h" = x"yes"; then AC_CHECK_DECLS([SCTP_UNORDERED, SCTP_ADDR_OVER, SCTP_ABORT, SCTP_EOF, SCTP_SENDALL, SCTP_ADDR_CONFIRMED, SCTP_DELAYED_ACK_TIME, - SCTP_EMPTY, + SCTP_EMPTY, SCTP_UNCONFIRMED, SCTP_CLOSED, SCTPS_IDLE, SCTP_BOUND, SCTPS_BOUND, SCTP_LISTEN, SCTPS_LISTEN, @@ -3503,19 +3506,6 @@ fi AC_SUBST(NATIVE_LIBS_ENABLED) # -# Check if HiPE should use a standard installation of perfctr. -# -AC_SUBST(USE_PERFCTR) -if test "x$HIPE_ENABLED" = "xyes" ; then - if test "x$with_perfctr" = "x" ; then - AC_CHECK_LIB(perfctr, vperfctr_info, [USE_PERFCTR=1 - AC_DEFINE(USE_PERFCTR,[1],[Define to enable hrvtime() on Linux systems with perfctr extension])]) - elif test "x$with_perfctr" != "xno" ; then - USE_PERFCTR=1 - fi -fi - -# # Check for working poll(). # AC_MSG_CHECKING([for working poll()]) @@ -4905,6 +4895,11 @@ LDFLAGS="$LDFLAGS $sanitizers" ]) dnl ---------------------------------------------------------------------- +dnl Check for log2 +dnl ---------------------------------------------------------------------- +AC_CHECK_FUNCS([log2]) + +dnl ---------------------------------------------------------------------- dnl Output the result. dnl ---------------------------------------------------------------------- diff --git a/erts/doc/src/crash_dump.xml b/erts/doc/src/crash_dump.xml index 2b5fc877c3..8291bf38b7 100644 --- a/erts/doc/src/crash_dump.xml +++ b/erts/doc/src/crash_dump.xml @@ -55,10 +55,12 @@ emulator or the operating system can be reconfigured to avoid the crash, which is why interpreting the crash dump correctly is important.</p> + <p>On systems that support OS signals, it is also possible to stop + the runtime system and generate a crash dump by sending the SIGUSR1.</p> <p>The erlang crash dump is a readable text file, but it might not be very easy to read. Using the Crashdump Viewer tool in the <c><![CDATA[observer]]></c> application will simplify the task. This is an - HTML based tool for browsing Erlang crash dumps.</p> + wx-widget based tool for browsing Erlang crash dumps.</p> <section> <marker id="general_info"></marker> @@ -66,8 +68,9 @@ <p>The first part of the dump shows the creation time for the dump, a slogan indicating the reason for the dump, the system version, of the node from which the dump originates, the compile time of - the emulator running the originating node and the number of - atoms in the atom table. + the emulator running the originating node, the number of + atoms in the atom table and the runtime system thread that caused + the crash dump to happen. </p> <section> @@ -170,6 +173,60 @@ </section> <section> + <marker id="scheduler"></marker> + <title>Scheduler information</title> + <p>Under the tag <em>=scheduler</em> information about the current state + and statistics of the schedulers in the runtime system is displayed. + On OSs that do allow instant suspension of other threads, the data within + this section will reflect what the runtime system looks like at the moment + when the crash happens.</p> + <p>The following fields can exist for a process:</p> + <taglist> + <tag><em>=scheduler:id</em></tag> + <item>Header, states the scheduler identifier.</item> + <tag><em>Scheduler Sleep Info Flags</em></tag> + <item>If empty the scheduler was doing some work. + If not empty the scheduler is either in some state of sleep, + or suspended. This entry is only present in a SMP enabled emulator</item> + <tag><em>Scheduler Sleep Info Aux Work</em></tag> + <item>If not empty, a scheduler internal auxiliary work is scheduled + to be done.</item> + <tag><em>Current Port</em></tag> + <item>The port identifier of the port that is currently being + executed by the scheduler.</item> + <tag><em>Current Process</em></tag> + <item>The process identifier of the process that is currently being + executed by the scheduler. If there is such a process this entry is + followed by the <em>State</em>,<em>Internal State</em>, + <em>Program Counter</em>, <em>CP</em> of that same process. See + <seealso marker="#processes">Process Information</seealso> for a + description what the different entries mean. Keep in mind that + this is a snapshot of what the entries are exactly when the crash + dump is starting to be generated. Therefore they will most likely + be different (and more telling) then the entries for the same + processes found in the <em>=proc</em> section. If there is no currently + running process, only the <em>Current Process</em> entry will be printed. + </item> + <tag><em>Current Process Limited Stack Trace</em></tag> + <item>This entry only shows up if there is a current process. It is very + similar to <seealso marker="#proc_data"><em>=proc_stack</em></seealso>, + except that only the function frames are printed (i.e. the stack variables + are omited). It is also limited to only print the top and bottom part + of the stack. If the stack is small (less that 512 slots) then the + entire stack will be printed. If not, an entry stating + <code>skipping ## slots</code> will be printed where ## is + replaced by the number of slots that has been skipped.</item> + <tag><em>Run Queue</em></tag> + <item>Displays statistics about how many processes and ports + of different priorities are scheduled on this scheduler.</item> + <tag><em>** crashed **</em></tag> + <item>This entry is normally not printed. It signifies that getting + the rest of the information about this scheduler failed for some reason. + </item> + </taglist> + </section> + + <section> <marker id="memory"></marker> <title>Memory information</title> <p>Under the tag <em>=memory</em> you will find information similar @@ -314,6 +371,9 @@ <item>The number of live argument registers. The argument registers, if any are live, will follow. These may contain the arguments of the function if they are not yet moved to the stack.</item> + <item><em>Internal State</em></item> + <item>A more detailed internal represantation of the state of + this process.</item> </taglist> <p>See also the section about <seealso marker="#proc_data">process data</seealso>.</p> </section> @@ -339,18 +399,38 @@ <tag><em>Name</em></tag> <item>The name of the table, regardless of whether it is a <c><![CDATA[named_table]]></c> or not.</item> - <tag><em>Buckets</em></tag> + <tag><em>Hash table, Buckets</em></tag> <item>This occurs if the table is a hash table, i.e. if it is not an <c><![CDATA[ordered_set]]></c>.</item> + <tag><em>Hash table, Chain Length</em></tag> + <item>Only applicable for hash tables. Contains statistics about the + hash table, such as the max, min and avg chain length. Having a max much + larger than the avg, and a std dev much larger that + the expected std dev is a sign that the hashing of the terms is + behaving badly for some reason.</item> <tag><em>Ordered set (AVL tree), Elements</em></tag> <item>This occurs only if the table is an <c><![CDATA[ordered_set]]></c>. (The number of elements is the same as the number of objects in the table.)</item> + <tag><em>Fixed</em></tag> + <item>If the table is fixed using ets:safe_fixtable or some internal + mechanism.</item> <tag><em>Objects</em></tag> <item>The number of objects in the table</item> <tag><em>Words</em></tag> <item>The number of words (usually 4 bytes/word) allocated to data in the table.</item> + <tag><em>Type</em></tag> + <item>The type of the table, i.e. <c>set</c>, <c>bag</c>, + <c>dublicate_bag</c> or <c>ordered_set</c>.</item> + <tag><em>Compressed</em></tag> + <item>If this table was compressed.</item> + <tag><em>Protection</em></tag> + <item>The protection of this table.</item> + <tag><em>Write Concurrency</em></tag> + <item>If write_concurrency was enabled for this table.</item> + <tag><em>Read Concurrency</em></tag> + <item>If read_concurrency was enabled for this table.</item> </taglist> </section> diff --git a/erts/doc/src/erl.xml b/erts/doc/src/erl.xml index 19a8e1f789..ea94a4e82b 100644 --- a/erts/doc/src/erl.xml +++ b/erts/doc/src/erl.xml @@ -599,6 +599,11 @@ <p>Sets the default binary virtual heap size of processes to the size <c><![CDATA[Size]]></c>.</p> </item> + <tag><c><![CDATA[+hpds Size]]></c></tag> + <item> + <p>Sets the initial process dictionary size of processes to the size + <c><![CDATA[Size]]></c>.</p> + </item> <tag><c><![CDATA[+K true | false]]></c></tag> <item> <p>Enables or disables the kernel poll functionality if @@ -1156,9 +1161,9 @@ <tag><marker id="+secio"><c>+secio true|false</c></marker></tag> <item> <p>Enable or disable eager check I/O scheduling. The default - is currently <c>false</c>, but will most likely be changed - to <c>true</c> in OTP 18. The behaviour before this flag - was introduced corresponds to <c>+secio false</c>.</p> + is currently <c>true</c>. The default was changed from <c>false</c> + to <c>true</c> as of erts version 7.0. The behaviour before this + flag was introduced corresponds to <c>+secio false</c>.</p> <p>The flag effects when schedulers will check for I/O operations possible to execute, and when such I/O operations will execute. As the name of the parameter implies, diff --git a/erts/doc/src/erlang.xml b/erts/doc/src/erlang.xml index 3cbfd372ce..880c294c73 100644 --- a/erts/doc/src/erlang.xml +++ b/erts/doc/src/erlang.xml @@ -1332,10 +1332,11 @@ true <fsummary>Check if a function is exported and loaded</fsummary> <desc> <p>Returns <c>true</c> if the module <c><anno>Module</anno></c> is loaded - and contains an exported function <c><anno>Function</anno>/<anno>Arity</anno></c>; - otherwise <c>false</c>.</p> - <p>Returns <c>false</c> for any BIF (functions implemented in C - rather than in Erlang).</p> + and contains an exported function <c><anno>Function</anno>/<anno>Arity</anno></c>, + or if there is a BIF (a built-in function implemented in C) + with the given name; otherwise returns <c>false</c>.</p> + <note><p>This function used to return false for built-in + functions before the 18.0 release.</p></note> </desc> </func> <func> @@ -1436,7 +1437,7 @@ true <name name="get" arity="1"/> <fsummary>Return a value from the process dictionary</fsummary> <desc> - <p>Returns the value <c><anno>Val</anno></c>associated with <c><anno>Key</anno></c> in + <p>Returns the value <c><anno>Val</anno></c> associated with <c><anno>Key</anno></c> in the process dictionary, or <c>undefined</c> if <c><anno>Key</anno></c> does not exist.</p> <pre> @@ -1456,6 +1457,19 @@ true </desc> </func> <func> + <name name="get_keys" arity="0"/> + <fsummary>Return a list of all keys from the process dictionary</fsummary> + <desc> + <p>Returns a list of keys all keys present in the process dictionary.</p> + <pre> +> <input>put(dog, {animal,1}),</input> +<input>put(cow, {animal,2}),</input> +<input>put(lamb, {animal,3}),</input> +<input>get_keys().</input> +[dog,cow,lamb]</pre> + </desc> + </func> + <func> <name name="get_keys" arity="1"/> <fsummary>Return a list of keys from the process dictionary</fsummary> <desc> @@ -3050,7 +3064,7 @@ os_prompt% </pre> raised. The error reason may differ between operating systems. Typically the error <c>enoent</c> is raised when one tries to run a program that is not found and - <c>eaccess</c> is raised when the given file is not + <c>eacces</c> is raised when the given file is not executable.</p> </item> <tag><c>{fd, <anno>In</anno>, <anno>Out</anno>}</c></tag> @@ -7290,7 +7304,9 @@ timestamp() -> only allowed with <c>PidSpec==all</c>. If the host machine operating system does not support high resolution CPU time measurements, <c>trace/3</c> exits with - <c>badarg</c>.</p> + <c>badarg</c>. Note that most operating systems do + not synchronize this value across cores, so be prepared + that time might seem to go backwards when using this option.</p> </item> <tag><c>arity</c></tag> <item> diff --git a/erts/doc/src/match_spec.xml b/erts/doc/src/match_spec.xml index 334b47d34c..b4cc8e9f78 100644 --- a/erts/doc/src/match_spec.xml +++ b/erts/doc/src/match_spec.xml @@ -76,22 +76,26 @@ { GuardFunction, ConditionExpression, ... } </item> <item>BoolFunction ::= <c><![CDATA[is_atom]]></c> | - <c><![CDATA[is_float]]></c> | <c><![CDATA[is_integer]]></c> | <c><![CDATA[is_list]]></c> | - <c><![CDATA[is_number]]></c> | <c><![CDATA[is_pid]]></c> | <c><![CDATA[is_port]]></c> | - <c><![CDATA[is_reference]]></c> | <c><![CDATA[is_tuple]]></c> | <c><![CDATA[is_binary]]></c> | - <c><![CDATA[is_function]]></c> | <c><![CDATA[is_record]]></c> | <c><![CDATA[is_seq_trace]]></c> | - <c><![CDATA['and']]></c> | <c><![CDATA['or']]></c> | <c><![CDATA['not']]></c> | <c><![CDATA['xor']]></c> | - <c><![CDATA[andalso]]></c> | <c><![CDATA[orelse]]></c></item> + <c><![CDATA[is_float]]></c> | <c><![CDATA[is_integer]]></c> | + <c><![CDATA[is_list]]></c> | <c><![CDATA[is_number]]></c> | + <c><![CDATA[is_pid]]></c> | <c><![CDATA[is_port]]></c> | + <c><![CDATA[is_reference]]></c> | <c><![CDATA[is_tuple]]></c> | + <c><![CDATA[is_map]]></c> | <c><![CDATA[is_binary]]></c> | + <c><![CDATA[is_function]]></c> | <c><![CDATA[is_record]]></c> | + <c><![CDATA[is_seq_trace]]></c> | <c><![CDATA['and']]></c> | + <c><![CDATA['or']]></c> | <c><![CDATA['not']]></c> | + <c><![CDATA['xor']]></c> | <c><![CDATA[andalso]]></c> | + <c><![CDATA[orelse]]></c></item> <item>ConditionExpression ::= ExprMatchVariable | { GuardFunction } | { GuardFunction, ConditionExpression, ... } | TermConstruct </item> <item>ExprMatchVariable ::= MatchVariable (bound in the MatchHead) | <c><![CDATA['$_']]></c> | <c><![CDATA['$$']]></c></item> - <item>TermConstruct = {{}} | {{ ConditionExpression, ... }} | - <c><![CDATA[[]]]></c> | [ConditionExpression, ...] | NonCompositeTerm | Constant - </item> - <item>NonCompositeTerm ::= term() (not list or tuple) - </item> + <item>TermConstruct = {{}} | {{ ConditionExpression, ... }} | + <c><![CDATA[[]]]></c> | [ConditionExpression, ...] | + <c><![CDATA[#{}]]></c> | #{term() => ConditionExpression, ...} | + NonCompositeTerm | Constant</item> + <item>NonCompositeTerm ::= term() (not list or tuple or map)</item> <item>Constant ::= {<c><![CDATA[const]]></c>, term()} </item> <item>GuardFunction ::= BoolFunction | <c><![CDATA[abs]]></c> | @@ -134,22 +138,26 @@ { GuardFunction, ConditionExpression, ... } </item> <item>BoolFunction ::= <c><![CDATA[is_atom]]></c> | - <c><![CDATA[is_float]]></c> | <c><![CDATA[is_integer]]></c> | <c><![CDATA[is_list]]></c> | - <c><![CDATA[is_number]]></c> | <c><![CDATA[is_pid]]></c> | <c><![CDATA[is_port]]></c> | - <c><![CDATA[is_reference]]></c> | <c><![CDATA[is_tuple]]></c> | <c><![CDATA[is_binary]]></c> | - <c><![CDATA[is_function]]></c> | <c><![CDATA[is_record]]></c> | <c><![CDATA[is_seq_trace]]></c> | - <c><![CDATA['and']]></c> | <c><![CDATA['or']]></c> | <c><![CDATA['not']]></c> | <c><![CDATA['xor']]></c> | - <c><![CDATA[andalso]]></c> | <c><![CDATA[orelse]]></c></item> + <c><![CDATA[is_float]]></c> | <c><![CDATA[is_integer]]></c> | + <c><![CDATA[is_list]]></c> | <c><![CDATA[is_number]]></c> | + <c><![CDATA[is_pid]]></c> | <c><![CDATA[is_port]]></c> | + <c><![CDATA[is_reference]]></c> | <c><![CDATA[is_tuple]]></c> | + <c><![CDATA[is_map]]></c> | <c><![CDATA[is_binary]]></c> | + <c><![CDATA[is_function]]></c> | <c><![CDATA[is_record]]></c> | + <c><![CDATA[is_seq_trace]]></c> | <c><![CDATA['and']]></c> | + <c><![CDATA['or']]></c> | <c><![CDATA['not']]></c> | + <c><![CDATA['xor']]></c> | <c><![CDATA[andalso]]></c> | + <c><![CDATA[orelse]]></c></item> <item>ConditionExpression ::= ExprMatchVariable | { GuardFunction } | { GuardFunction, ConditionExpression, ... } | TermConstruct </item> <item>ExprMatchVariable ::= MatchVariable (bound in the MatchHead) | <c><![CDATA['$_']]></c> | <c><![CDATA['$$']]></c></item> <item>TermConstruct = {{}} | {{ ConditionExpression, ... }} | - <c><![CDATA[[]]]></c> | [ConditionExpression, ...] | NonCompositeTerm | Constant - </item> - <item>NonCompositeTerm ::= term() (not list or tuple) - </item> + <c><![CDATA[[]]]></c> | [ConditionExpression, ...] | #{} | + #{term() => ConditionExpression, ...} | NonCompositeTerm | + Constant</item> + <item>NonCompositeTerm ::= term() (not list or tuple or map)</item> <item>Constant ::= {<c><![CDATA[const]]></c>, term()} </item> <item>GuardFunction ::= BoolFunction | <c><![CDATA[abs]]></c> | @@ -172,9 +180,10 @@ <title>Functions allowed in all types of match specifications</title> <p>The different functions allowed in <c><![CDATA[match_spec]]></c> work like this: </p> - <p><em>is_atom, is_float, is_integer, is_list, is_number, is_pid, is_port, is_reference, is_tuple, is_binary, is_function: </em> Like the corresponding guard tests in - Erlang, return <c><![CDATA[true]]></c> or <c><![CDATA[false]]></c>. - </p> + <p><em>is_atom, is_float, is_integer, is_list, is_number, is_pid, is_port, + is_reference, is_tuple, is_map, is_binary, is_function:</em> Like the + corresponding guard tests in Erlang, return <c><![CDATA[true]]></c> or + <c><![CDATA[false]]></c>.</p> <p><em>is_record: </em>Takes an additional parameter, which SHALL be the result of <c><![CDATA[record_info(size, <record_type>)]]></c>, like in <c><![CDATA[{is_record, '$1', rectype, record_info(size, rectype)}]]></c>. diff --git a/erts/doc/src/notes.xml b/erts/doc/src/notes.xml index c896ee0cae..af0d4d7377 100644 --- a/erts/doc/src/notes.xml +++ b/erts/doc/src/notes.xml @@ -30,6 +30,65 @@ </header> <p>This document describes the changes made to the ERTS application.</p> +<section><title>Erts 6.3.1</title> + + <section><title>Fixed Bugs and Malfunctions</title> + <list> + <item> + <p> + Fix getifaddrs realloc pointer error</p> + <p> + When a buffer was exhausted and subsequently reallocated, + we could get an unsafe pointer pointing to faulty memory.</p> + <p> + For this to occur we would need to have a large number of + interfaces and a reallocation of memory to a lower + addresses.</p> + <p> + The symptom would be garbage returned from + erlang:port_control(Port, 25, []) + (prim_inet:getifaddrs(Port) resulting in a badarg) or a + segmentation fault.</p> + <p> + Own Id: OTP-12445</p> + </item> + <item> + <p> + Don't close all file descriptors twice in child_setup</p> + <p> + The commit c2b4eab25c907f453a394d382c04cd04e6c06b49 + introduced an error in which child_setup erroneously + tried to close all file descriptors twice.</p> + <p> + Use closefrom() if available when closing all file + descriptors.</p> + <p> + The function closefrom() was only used in the vfork() + case before but is now also used in the fork() case if + available.</p> + <p> + Own Id: OTP-12446</p> + </item> + <item> + <p> + During a crashdump all file descriptors are closed to + ensure the closing of the epmd port and to reserve a file + descriptor for the crashdump file.</p> + <p> + If a driver (third party library) cannot handle closing + of sockets this could result in a segmentation fault in + which case a crashdump would not be produced. This is now + fixed by only closing inets sockets via an emergency + close callback to the driver and thus closing the epmd + socket.</p> + <p> + Own Id: OTP-12447</p> + </item> + </list> + </section> + +</section> + <section><title>Erts 6.3</title> <section><title>Fixed Bugs and Malfunctions</title> diff --git a/erts/doc/src/zlib.xml b/erts/doc/src/zlib.xml index da8ccdecdf..1f10ddef6d 100644 --- a/erts/doc/src/zlib.xml +++ b/erts/doc/src/zlib.xml @@ -99,7 +99,7 @@ list_to_binary([Compressed|Last])</pre> <datatype> <name name="zwindowbits"/> <desc> - <p>Normally in the range <c>-15..-9 | 9..15</c>.</p> + <p>Normally in the range <c>-15..-8 | 8..15</c>.</p> </desc> </datatype> </datatypes> @@ -149,7 +149,7 @@ list_to_binary([Compressed|Last])</pre> currently the only supported method is <c>deflated</c>.</p> <p>The <c><anno>WindowBits</anno></c> parameter is the base two logarithm of the window size (the size of the history buffer). It - should be in the range 9 through 15. Larger values + should be in the range 8 through 15. Larger values of this parameter result in better compression at the expense of memory usage. The default value is 15 if <c>deflateInit/2</c>. A negative <c><anno>WindowBits</anno></c> @@ -288,7 +288,7 @@ list_to_binary([B1,B2])</pre> <p>Initialize decompression session on zlib stream.</p> <p>The <c><anno>WindowBits</anno></c> parameter is the base two logarithm of the maximum window size (the size of the history buffer). - It should be in the range 9 through 15. + It should be in the range 8 through 15. The default value is 15 if <c>inflateInit/1</c> is used. If a compressed stream with a larger window size is given as input, inflate() will throw the <c>data_error</c> @@ -312,6 +312,53 @@ list_to_binary([B1,B2])</pre> </desc> </func> <func> + <name name="inflateChunk" arity="2"/> + <fsummary>Decompress data with limited output size</fsummary> + <desc> + <p>Like <c>inflate/2</c>, but decompress no more data than + will fit in the buffer configured via <c>setBufSize/2</c>. + Is is useful when decompressing a stream with a high compression + ratio such that a small amount of compressed input may expand up to + 1000 times. + It returns <c>{more, Decompressed}</c>, when there is more output + available, and <c>inflateChunk/1</c> should be used to read it. + It may introduce some output latency (reading + input without producing any output).</p> + <p>If a preset dictionary is needed at this point (see + <c>inflateSetDictionary</c> below), <c>inflateChunk/2</c> throws a + <c>{need_dictionary,Adler}</c> exception where <c>Adler</c> is + the adler32 checksum of the dictionary chosen by the + compressor.</p> + + <pre> +walk(Compressed, Handler) -> + Z = zlib:open(), + zlib:inflateInit(Z), + % Limit single uncompressed chunk size to 512kb + zlib:setBufSize(Z, 512 * 1024), + loop(Z, Handler, zlib:inflateChunk(Z, Compressed)), + zlib:inflateEnd(Z), + zlib:close(Z). + +loop(Z, Handler, {more, Uncompressed}) -> + Handler(Uncompressed), + loop(Z, Handler, zlib:inflateChunk(Z)); +loop(Z, Handler, Uncompressed) -> + Handler(Uncompressed). + </pre> + </desc> + </func> + <func> + <name name="inflateChunk" arity="1"/> + <fsummary>Read next uncompressed chunk</fsummary> + <desc> + <p>Read next chunk of uncompressed data, initialized by + <c>inflateChunk/2</c>.</p> + <p>This function should be repeatedly called, while it returns + <c>{more, Decompressed}</c>.</p> + </desc> + </func> + <func> <name name="inflateSetDictionary" arity="2"/> <fsummary>Initialize the decompression dictionary</fsummary> <desc> diff --git a/erts/emulator/Makefile.in b/erts/emulator/Makefile.in index 560cab396d..1429a6cf2c 100644 --- a/erts/emulator/Makefile.in +++ b/erts/emulator/Makefile.in @@ -112,18 +112,24 @@ NO_INLINE_FUNCTIONS=true else ifeq ($(TYPE),lcnt) -PURIFY = +PURIFY = TYPEMARKER = .lcnt TYPE_FLAGS = @CFLAGS@ -DERTS_ENABLE_LOCK_COUNT else ifeq ($(TYPE),frmptr) -PURIFY = +PURIFY = OMIT_OMIT_FP=yes TYPEMARKER = .frmptr TYPE_FLAGS = @CFLAGS@ -DERTS_FRMPTR else +ifeq ($(TYPE),icount) +PURIFY = +TYPEMARKER = .icount +TYPE_FLAGS = @CFLAGS@ -DERTS_OPCODE_COUNTER_SUPPORT +else + # If type isn't one of the above, it *is* opt type... override TYPE=opt PURIFY = @@ -138,6 +144,7 @@ endif endif endif endif +endif comma:=, space:= @@ -342,16 +349,6 @@ endif EPCRE_LIB = $(ERL_TOP)/erts/emulator/pcre/obj/$(TARGET)/$(TYPE)/$(LIB_PREFIX)epcre$(LIB_SUFFIX) DEPLIBS += $(EPCRE_LIB) -PERFCTR_PATH=@PERFCTR_PATH@ -USE_PERFCTR=@USE_PERFCTR@ -ifdef PERFCTR_PATH -LIBS += $(PERFCTR_PATH)/usr.lib/libperfctr.a -else -ifdef USE_PERFCTR -LIBS += -lperfctr -endif -endif - LIBSCTP = @LIBSCTP@ ORG_THR_LIBS=@EMU_THR_LIBS@ @@ -559,9 +556,6 @@ HIPE_ppc64_TAB=hipe/hipe_ppc64.tab $(HIPE_ARCH64_TAB) HIPE_arm_TAB=hipe/hipe_arm.tab HIPE_ARCH_TAB=$(HIPE_$(ARCH)_TAB) BIFS += hipe/hipe_bif0.tab hipe/hipe_bif1.tab hipe/hipe_bif2.tab $(HIPE_ARCH_TAB) -ifdef USE_PERFCTR -BIFS += hipe/hipe_perfctr.tab -endif endif $(TARGET)/erl_bif_table.c \ @@ -657,10 +651,6 @@ COMMON_INCLUDES += -I../include/internal -I../include/internal/$(TARGET) INCLUDES = -I$(TTF_DIR) $(COMMON_INCLUDES) -ifdef PERFCTR_PATH -INCLUDES += -I$(PERFCTR_PATH)/usr.lib -I$(PERFCTR_PATH)/linux/include -endif - ifeq ($(TARGET),win32) $(OBJDIR)/dll_sys.o: sys/$(ERLANG_OSTYPE)/sys.c $(V_CC) $(CFLAGS) -DERL_RUN_SHARED_LIB=1 $(INCLUDES) -c $< -o $@ @@ -921,14 +911,11 @@ HIPE_OBJS= \ $(OBJDIR)/hipe_mode_switch.o \ $(OBJDIR)/hipe_native_bif.o \ $(OBJDIR)/hipe_stack.o $(HIPE_ARCH_OBJS) -ifdef USE_PERFCTR -HIPE_OBJS += $(OBJDIR)/hipe_perfctr.o -endif ifdef HIPE_ENABLED EXTRA_BASE_OBJS += $(HIPE_OBJS) endif -BASE_OBJS = $(RUN_OBJS) $(EMU_OBJS) $(OS_OBJS) $(EXTRA_BASE_OBJS) +BASE_OBJS = $(EMU_OBJS) $(RUN_OBJS) $(OS_OBJS) $(EXTRA_BASE_OBJS) before_DTrace_OBJS = $(BASE_OBJS) $(DRV_OBJS) diff --git a/erts/emulator/beam/atom.names b/erts/emulator/beam/atom.names index e2760c18b4..ae3f30d82f 100644 --- a/erts/emulator/beam/atom.names +++ b/erts/emulator/beam/atom.names @@ -202,6 +202,7 @@ atom dotall atom driver atom driver_options atom dsend +atom dsend_continue_trap atom dunlink atom duplicate_bag atom dupnames @@ -283,7 +284,6 @@ atom http httph https http_response http_request http_header http_eoh http_error atom id atom if_clause atom ignore -atom imports atom in atom in_exiting atom inactive @@ -339,6 +339,7 @@ atom max atom maximum atom max_tables max_processes atom mbuf_size +atom md5 atom memory atom memory_internal atom memory_types diff --git a/erts/emulator/beam/beam_debug.c b/erts/emulator/beam/beam_debug.c index a3cd08834f..6bb987985d 100644 --- a/erts/emulator/beam/beam_debug.c +++ b/erts/emulator/beam/beam_debug.c @@ -579,9 +579,29 @@ print_op(int to, void *to_arg, int op, int size, BeamInstr* addr) unpacked = ap; ap = addr + size; switch (op) { - case op_i_select_val_rfI: - case op_i_select_val_xfI: - case op_i_select_val_yfI: + case op_i_select_val_lins_rfI: + case op_i_select_val_lins_xfI: + case op_i_select_val_lins_yfI: + { + int n = ap[-1]; + int ix = n; + + while (ix--) { + erts_print(to, to_arg, "%T ", (Eterm) ap[0]); + ap++; + size++; + } + ix = n; + while (ix--) { + erts_print(to, to_arg, "f(" HEXF ") ", (Eterm) ap[0]); + ap++; + size++; + } + } + break; + case op_i_select_val_bins_rfI: + case op_i_select_val_bins_xfI: + case op_i_select_val_bins_yfI: { int n = ap[-1]; @@ -598,13 +618,19 @@ print_op(int to, void *to_arg, int op, int size, BeamInstr* addr) case op_i_select_tuple_arity_yfI: { int n = ap[-1]; + int ix = n; - while (n > 0) { + while (ix--) { Uint arity = arityval(ap[0]); - erts_print(to, to_arg, " {%d} f(" HEXF ")", arity, ap[1]); - ap += 2; - size += 2; - n--; + erts_print(to, to_arg, "{%d} ", arity, ap[1]); + ap++; + size++; + } + ix = n; + while (ix--) { + erts_print(to, to_arg, "f(" HEXF ") ", ap[0]); + ap++; + size++; } } break; diff --git a/erts/emulator/beam/beam_emu.c b/erts/emulator/beam/beam_emu.c index 8bfb7d2ad2..6526e87e4c 100644 --- a/erts/emulator/beam/beam_emu.c +++ b/erts/emulator/beam/beam_emu.c @@ -241,10 +241,6 @@ BeamInstr beam_return_time_trace[1]; /* OpCode(i_return_time_trace) */ void** beam_ops; #endif -#ifndef ERTS_SMP /* Not supported with smp emulator */ -extern int count_instructions; -#endif - #define SWAPIN \ HTOP = HEAP_TOP(c_p); \ E = c_p->stop @@ -703,7 +699,7 @@ extern int count_instructions; Fail; \ } -#define IsMap(Src, Fail) if (is_not_map(Src)) { Fail; } +#define IsMap(Src, Fail) if (!is_map(Src)) { Fail; } #define HasMapField(Src, Key, Fail) if (has_not_map_field(Src, Key)) { Fail; } @@ -1163,14 +1159,15 @@ void process_main(void) Eterm (*arith_func)(Process* p, Eterm* reg, Uint live); -#ifndef NO_JUMP_TABLE - static void* opcodes[] = { DEFINE_OPCODES }; #ifdef ERTS_OPCODE_COUNTER_SUPPORT static void* counting_opcodes[] = { DEFINE_COUNTING_OPCODES }; -#endif +#else +#ifndef NO_JUMP_TABLE + static void* opcodes[] = { DEFINE_OPCODES }; #else int Go; #endif +#endif Uint temp_bits; /* Temporary used by BsSkipBits2 & BsGetInteger2 */ @@ -2141,19 +2138,18 @@ void process_main(void) NextPF(0, next); } - { Eterm select_val2; - OpCase(i_select_tuple_arity2_yfAfAf): + OpCase(i_select_tuple_arity2_yfAAff): select_val2 = yb(Arg(0)); goto do_select_tuple_arity2; - OpCase(i_select_tuple_arity2_xfAfAf): + OpCase(i_select_tuple_arity2_xfAAff): select_val2 = xb(Arg(0)); goto do_select_tuple_arity2; - OpCase(i_select_tuple_arity2_rfAfAf): + OpCase(i_select_tuple_arity2_rfAAff): select_val2 = r(0); I--; @@ -2164,22 +2160,22 @@ void process_main(void) select_val2 = *tuple_val(select_val2); goto do_select_val2; - OpCase(i_select_val2_yfcfcf): + OpCase(i_select_val2_yfccff): select_val2 = yb(Arg(0)); goto do_select_val2; - OpCase(i_select_val2_xfcfcf): + OpCase(i_select_val2_xfccff): select_val2 = xb(Arg(0)); goto do_select_val2; - OpCase(i_select_val2_rfcfcf): + OpCase(i_select_val2_rfccff): select_val2 = r(0); I--; do_select_val2: if (select_val2 == Arg(2)) { - I += 2; - } else if (select_val2 == Arg(4)) { + I += 3; + } else if (select_val2 == Arg(3)) { I += 4; } @@ -2206,20 +2202,50 @@ void process_main(void) do_select_tuple_arity: if (is_tuple(select_val)) { select_val = *tuple_val(select_val); - goto do_binary_search; + goto do_linear_search; + } + SET_I((BeamInstr *) Arg(1)); + Goto(*I); + + OpCase(i_select_val_lins_xfI): + select_val = xb(Arg(0)); + goto do_linear_search; + + OpCase(i_select_val_lins_yfI): + select_val = yb(Arg(0)); + goto do_linear_search; + + OpCase(i_select_val_lins_rfI): + select_val = r(0); + I--; + + do_linear_search: { + BeamInstr *vs = &Arg(3); + int ix = 0; + + for(;;) { + if (vs[ix+0] >= select_val) { ix += 0; break; } + if (vs[ix+1] >= select_val) { ix += 1; break; } + ix += 2; } + + if (vs[ix] == select_val) { + I += ix + Arg(2) + 2; + } + SET_I((BeamInstr *) Arg(1)); Goto(*I); + } - OpCase(i_select_val_xfI): + OpCase(i_select_val_bins_xfI): select_val = xb(Arg(0)); goto do_binary_search; - OpCase(i_select_val_yfI): + OpCase(i_select_val_bins_yfI): select_val = yb(Arg(0)); goto do_binary_search; - OpCase(i_select_val_rfI): + OpCase(i_select_val_bins_rfI): select_val = r(0); I--; @@ -2366,7 +2392,7 @@ void process_main(void) } OpCase(i_has_map_fields_fsI): { - map_t* mp; + flatmap_t* mp; Eterm map; Eterm field; Eterm *ks; @@ -2374,22 +2400,34 @@ void process_main(void) Uint sz,n; GetArg1(1, map); + n = (Uint)Arg(2); + fs = &Arg(3); /* pattern fields */ - /* this instruction assumes Arg1 is a map, - * i.e. that it follows a test is_map if needed. - */ + /* get term from field? */ + if (is_hashmap(map)) { + Uint32 hx; + while(n--) { + field = *fs++; + hx = hashmap_make_hash(field); + if (!erts_hashmap_get(hx,field,map)) { + SET_I((BeamInstr *) Arg(0)); + goto has_map_fields_fail; + } + } + goto has_map_fields_ok; + } + + ASSERT(is_flatmap(map)); - mp = (map_t *)map_val(map); - sz = map_get_size(mp); + mp = (flatmap_t *)flatmap_val(map); + sz = flatmap_get_size(mp); if (sz == 0) { SET_I((BeamInstr *) Arg(0)); goto has_map_fields_fail; } - ks = map_get_keys(mp); - n = (Uint)Arg(2); - fs = &Arg(3); /* pattern fields */ + ks = flatmap_get_keys(mp); ASSERT(n>0); @@ -2407,7 +2445,7 @@ void process_main(void) SET_I((BeamInstr *) Arg(0)); goto has_map_fields_fail; } - +has_map_fields_ok: I += 4 + Arg(2); has_map_fields_fail: ASSERT(VALID_INSTR(*I)); @@ -2434,12 +2472,8 @@ do { \ OpCase(i_get_map_elements_fsI): { Eterm map; - map_t *mp; - Eterm field; - Eterm *ks; - Eterm *vs; BeamInstr *fs; - Uint sz,n; + Uint sz, n; GetArg1(1, map); @@ -2447,36 +2481,56 @@ do { \ * i.e. that it follows a test is_map if needed. */ - mp = (map_t *)map_val(map); - sz = map_get_size(mp); - - if (sz == 0) { - SET_I((BeamInstr *) Arg(0)); - goto get_map_elements_fail; - } - n = (Uint)Arg(2) / 2; fs = &Arg(3); /* pattern fields and target registers */ - ks = map_get_keys(mp); - vs = map_get_values(mp); - while(sz) { - field = (Eterm)*fs; - if (EQ(field,*ks)) { - PUT_TERM_REG(*vs, fs[1]); - n--; + if (is_flatmap(map)) { + flatmap_t *mp; + Eterm *ks; + Eterm *vs; + + mp = (flatmap_t *)flatmap_val(map); + sz = flatmap_get_size(mp); + + if (sz == 0) { + SET_I((BeamInstr *) Arg(0)); + goto get_map_elements_fail; + } + + ks = flatmap_get_keys(mp); + vs = flatmap_get_values(mp); + + while(sz) { + if (EQ((Eterm)*fs,*ks)) { + PUT_TERM_REG(*vs, fs[1]); + n--; + fs += 2; + /* no more values to fetch, we are done */ + if (n == 0) break; + } + ks++; sz--; + vs++; + } + + if (n) { + SET_I((BeamInstr *) Arg(0)); + goto get_map_elements_fail; + } + } else { + const Eterm *v; + Uint32 hx; + ASSERT(is_hashmap(map)); + while(n--) { + hx = hashmap_make_hash((Eterm)*fs); + if ((v = erts_hashmap_get(hx,(Eterm)*fs, map)) == NULL) { + SET_I((BeamInstr *) Arg(0)); + goto get_map_elements_fail; + } + PUT_TERM_REG(*v, fs[1]); fs += 2; - /* no more values to fetch, we are done */ - if (n == 0) break; } - ks++; sz--; - vs++; } - if (n) { - SET_I((BeamInstr *) Arg(0)); - goto get_map_elements_fail; - } I += 4 + Arg(2); get_map_elements_fail: @@ -2775,6 +2829,7 @@ get_map_elements_fail: } PreFetch(1, next); ASSERT(!ERTS_PROC_IS_EXITING(c_p)); + ERTS_VERIFY_UNUSED_TEMP_ALLOC(c_p); reg[0] = r(0); result = (*bf)(c_p, reg, I); ASSERT(!ERTS_PROC_IS_EXITING(c_p) || is_non_value(result)); @@ -3556,7 +3611,7 @@ get_map_elements_fail: vbf = (BifFunction) Arg(0); PROCESS_MAIN_CHK_LOCKS(c_p); bif_nif_arity = I[-1]; - ASSERT(bif_nif_arity <= 3); + ASSERT(bif_nif_arity <= 4); ERTS_SMP_UNREQ_PROC_MAIN_LOCK(c_p); reg[0] = r(0); { @@ -3782,8 +3837,6 @@ get_map_elements_fail: * Allocate the binary struct itself. */ bptr = erts_bin_nrml_alloc(num_bytes); - bptr->flags = 0; - bptr->orig_size = num_bytes; erts_refc_init(&bptr->refc, 1); erts_current_bin = (byte *) bptr->orig_bytes; @@ -3883,8 +3936,6 @@ get_map_elements_fail: * Allocate the binary struct itself. */ bptr = erts_bin_nrml_alloc(tmp_arg1); - bptr->flags = 0; - bptr->orig_size = tmp_arg1; erts_refc_init(&bptr->refc, 1); erts_current_bin = (byte *) bptr->orig_bytes; @@ -4987,14 +5038,14 @@ get_map_elements_fail: * ... remainder of original BEAM code */ ASSERT(I[-5] == (Uint) OpCode(i_func_info_IaaI)); - c_p->hipe.ncallee = (void(*)(void)) I[-4]; + c_p->hipe.u.ncallee = (void(*)(void)) I[-4]; cmd = HIPE_MODE_SWITCH_CMD_CALL | (I[-1] << 8); ++hipe_trap_count; goto L_hipe_mode_switch; } OpCase(hipe_trap_call_closure): { ASSERT(I[-5] == (Uint) OpCode(i_func_info_IaaI)); - c_p->hipe.ncallee = (void(*)(void)) I[-4]; + c_p->hipe.u.ncallee = (void(*)(void)) I[-4]; cmd = HIPE_MODE_SWITCH_CMD_CALL_CLOSURE | (I[-1] << 8); ++hipe_trap_count; goto L_hipe_mode_switch; @@ -5028,7 +5079,10 @@ get_map_elements_fail: case HIPE_MODE_SWITCH_RES_RETURN: ASSERT(is_value(reg[0])); MoveReturn(reg[0], r(0)); - case HIPE_MODE_SWITCH_RES_CALL: + case HIPE_MODE_SWITCH_RES_CALL_EXPORTED: + c_p->i = c_p->hipe.u.callee_exp->addressv[erts_active_code_ix()]; + /*fall through*/ + case HIPE_MODE_SWITCH_RES_CALL_BEAM: SET_I(c_p->i); r(0) = reg[0]; Dispatch(); @@ -5145,22 +5199,16 @@ get_map_elements_fail: #ifndef NO_JUMP_TABLE #ifdef ERTS_OPCODE_COUNTER_SUPPORT - /* Are tables correctly generated by beam_makeops? */ - ASSERT(sizeof(counting_opcodes) == sizeof(opcodes)); - - if (count_instructions) { + ERTS_CT_ASSERT(sizeof(counting_opcodes) == sizeof(opcodes)); #ifdef DEBUG - counting_opcodes[op_catch_end_y] = LabelAddr(lb_catch_end_y); + counting_opcodes[op_catch_end_y] = LabelAddr(lb_catch_end_y); #endif - counting_opcodes[op_i_func_info_IaaI] = LabelAddr(lb_i_func_info_IaaI); - beam_ops = counting_opcodes; - } - else -#endif /* #ifndef ERTS_OPCODE_COUNTER_SUPPORT */ - { - beam_ops = opcodes; - } + counting_opcodes[op_i_func_info_IaaI] = LabelAddr(lb_i_func_info_IaaI); + beam_ops = counting_opcodes; +#else /* #ifndef ERTS_OPCODE_COUNTER_SUPPORT */ + beam_ops = opcodes; +#endif /* ERTS_OPCODE_COUNTER_SUPPORT */ #endif /* NO_JUMP_TABLE */ em_call_error_handler = OpCode(call_error_handler); @@ -6424,55 +6472,69 @@ new_fun(Process* p, Eterm* reg, ErlFunEntry* fe, int num_free) static int has_not_map_field(Eterm map, Eterm key) { - map_t* mp; - Eterm* keys; - Uint i; - Uint n; - - mp = (map_t *)map_val(map); - keys = map_get_keys(mp); - n = map_get_size(mp); - if (is_immed(key)) { - for (i = 0; i < n; i++) { - if (keys[i] == key) { - return 0; + Uint32 hx; + if (is_flatmap(map)) { + flatmap_t* mp; + Eterm* keys; + Uint i; + Uint n; + + mp = (flatmap_t *)flatmap_val(map); + keys = flatmap_get_keys(mp); + n = flatmap_get_size(mp); + if (is_immed(key)) { + for (i = 0; i < n; i++) { + if (keys[i] == key) { + return 0; + } } - } - } else { - for (i = 0; i < n; i++) { - if (EQ(keys[i], key)) { - return 0; + } else { + for (i = 0; i < n; i++) { + if (EQ(keys[i], key)) { + return 0; + } } } + return 1; } - return 1; + ASSERT(is_hashmap(map)); + hx = hashmap_make_hash(key); + return erts_hashmap_get(hx,key,map) ? 0 : 1; } static Eterm get_map_element(Eterm map, Eterm key) { - map_t *mp; - Eterm* ks, *vs; - Uint i; - Uint n; - - mp = (map_t *)map_val(map); - ks = map_get_keys(mp); - vs = map_get_values(mp); - n = map_get_size(mp); - if (is_immed(key)) { - for (i = 0; i < n; i++) { - if (ks[i] == key) { - return vs[i]; + Uint32 hx; + const Eterm *vs; + if (is_flatmap(map)) { + flatmap_t *mp; + Eterm *ks; + Uint i; + Uint n; + + mp = (flatmap_t *)flatmap_val(map); + ks = flatmap_get_keys(mp); + vs = flatmap_get_values(mp); + n = flatmap_get_size(mp); + if (is_immed(key)) { + for (i = 0; i < n; i++) { + if (ks[i] == key) { + return vs[i]; + } } - } - } else { - for (i = 0; i < n; i++) { - if (EQ(ks[i], key)) { - return vs[i]; + } else { + for (i = 0; i < n; i++) { + if (EQ(ks[i], key)) { + return vs[i]; + } } } + return THE_NON_VALUE; } - return THE_NON_VALUE; + ASSERT(is_hashmap(map)); + hx = hashmap_make_hash(key); + vs = erts_hashmap_get(hx,key,map); + return vs ? *vs : THE_NON_VALUE; } #define GET_TERM(term, dest) \ @@ -6505,7 +6567,30 @@ new_map(Process* p, Eterm* reg, BeamInstr* I) Eterm *mhp,*thp; Eterm *E; BeamInstr *ptr; - map_t *mp; + flatmap_t *mp; + ErtsHeapFactory factory; + + ptr = &Arg(4); + + if (n > 2*MAP_SMALL_MAP_LIMIT) { + if (HeapWordsLeft(p) < n) { + erts_garbage_collect(p, n, reg, Arg(2)); + } + + mhp = p->htop; + thp = p->htop; + E = p->stop; + + for (i = 0; i < n/2; i++) { + GET_TERM(*ptr++, *mhp++); + GET_TERM(*ptr++, *mhp++); + } + + p->htop = mhp; + + factory.p = p; + return erts_hashmap_from_array(&factory, thp, n/2, 0); + } if (HeapWordsLeft(p) < need) { erts_garbage_collect(p, need, reg, Arg(2)); @@ -6514,11 +6599,10 @@ new_map(Process* p, Eterm* reg, BeamInstr* I) thp = p->htop; mhp = thp + 1 + n/2; E = p->stop; - ptr = &Arg(4); keys = make_tuple(thp); *thp++ = make_arityval(n/2); - mp = (map_t *)mhp; mhp += MAP_HEADER_SIZE; + mp = (flatmap_t *)mhp; mhp += MAP_HEADER_SIZE; mp->thing_word = MAP_HEADER; mp->size = n/2; mp->keys = keys; @@ -6528,7 +6612,7 @@ new_map(Process* p, Eterm* reg, BeamInstr* I) GET_TERM(*ptr++, *mhp++); } p->htop = mhp; - return make_map(mp); + return make_flatmap(mp); } static Eterm @@ -6538,7 +6622,7 @@ update_map_assoc(Process* p, Eterm* reg, Eterm map, BeamInstr* I) Uint num_old; Uint num_updates; Uint need; - map_t *old_mp, *mp; + flatmap_t *old_mp, *mp; Eterm res; Eterm* hp; Eterm* E; @@ -6548,12 +6632,44 @@ update_map_assoc(Process* p, Eterm* reg, Eterm map, BeamInstr* I) Eterm new_key; Eterm* kp; - if (is_not_map(map)) { - return THE_NON_VALUE; + new_p = &Arg(5); + num_updates = Arg(4) / 2; + + if (is_not_flatmap(map)) { + Uint32 hx; + Eterm val; + + /* apparently the compiler does not emit is_map instructions, + * bad compiler */ + + if (is_not_hashmap(map)) + return THE_NON_VALUE; + + res = map; + E = p->stop; + while(num_updates--) { + /* assoc can't fail */ + GET_TERM(new_p[0], new_key); + GET_TERM(new_p[1], val); + hx = hashmap_make_hash(new_key); + + res = erts_hashmap_insert(p, hx, new_key, val, res, 0); + if (p->mbuf) { + Uint live = Arg(3); + reg[live] = res; + erts_garbage_collect(p, 0, reg, live+1); + res = reg[live]; + } + + E = p->stop; + + new_p += 2; + } + return res; } - old_mp = (map_t *) map_val(map); - num_old = map_get_size(old_mp); + old_mp = (flatmap_t *) flatmap_val(map); + num_old = flatmap_get_size(old_mp); /* * If the old map is empty, create a new map. @@ -6568,14 +6684,13 @@ update_map_assoc(Process* p, Eterm* reg, Eterm map, BeamInstr* I) * update list are new). */ - num_updates = Arg(4) / 2; need = 2*(num_old+num_updates) + 1 + MAP_HEADER_SIZE; if (HeapWordsLeft(p) < need) { Uint live = Arg(3); reg[live] = map; erts_garbage_collect(p, need, reg, live+1); map = reg[live]; - old_mp = (map_t *)map_val(map); + old_mp = (flatmap_t *)flatmap_val(map); } /* @@ -6606,16 +6721,15 @@ update_map_assoc(Process* p, Eterm* reg, Eterm map, BeamInstr* I) kp = p->htop + 1; /* Point to first key */ hp = kp + num_old + num_updates; - res = make_map(hp); - mp = (map_t *)hp; + res = make_flatmap(hp); + mp = (flatmap_t *)hp; hp += MAP_HEADER_SIZE; mp->thing_word = MAP_HEADER; mp->keys = make_tuple(kp-1); - old_vals = map_get_values(old_mp); - old_keys = map_get_keys(old_mp); + old_vals = flatmap_get_values(old_mp); + old_keys = flatmap_get_keys(old_mp); - new_p = &Arg(5); GET_TERM(*new_p, new_key); n = num_updates; @@ -6701,8 +6815,19 @@ update_map_assoc(Process* p, Eterm* reg, Eterm map, BeamInstr* I) n = kp - p->htop - 1; /* Actual number of keys/values */ *p->htop = make_arityval(n); + p->htop = hp; mp->size = n; - p->htop = hp; + + /* The expensive case, need to build a hashmap */ + if (n > MAP_SMALL_MAP_LIMIT) { + res = erts_hashmap_from_ks_and_vs(p,flatmap_get_keys(mp),flatmap_get_values(mp),n); + if (p->mbuf) { + Uint live = Arg(3); + reg[live] = res; + erts_garbage_collect(p, 0, reg, live+1); + res = reg[live]; + } + } return res; } @@ -6717,7 +6842,7 @@ update_map_exact(Process* p, Eterm* reg, Eterm map, BeamInstr* I) Uint i; Uint num_old; Uint need; - map_t *old_mp, *mp; + flatmap_t *old_mp, *mp; Eterm res; Eterm* hp; Eterm* E; @@ -6726,12 +6851,48 @@ update_map_exact(Process* p, Eterm* reg, Eterm map, BeamInstr* I) BeamInstr* new_p; Eterm new_key; - if (is_not_map(map)) { - return THE_NON_VALUE; + new_p = &Arg(5); + n = Arg(4) / 2; /* Number of values to be updated */ + ASSERT(n > 0); + + if (is_not_flatmap(map)) { + Uint32 hx; + Eterm val; + + /* apparently the compiler does not emit is_map instructions, + * bad compiler */ + + if (is_not_hashmap(map)) + return THE_NON_VALUE; + + res = map; + E = p->stop; + while(n--) { + /* assoc can't fail */ + GET_TERM(new_p[0], new_key); + GET_TERM(new_p[1], val); + hx = hashmap_make_hash(new_key); + + res = erts_hashmap_insert(p, hx, new_key, val, res, 1); + if (is_non_value(res)) + return res; + + if (p->mbuf) { + Uint live = Arg(3); + reg[live] = res; + erts_garbage_collect(p, 0, reg, live+1); + res = reg[live]; + } + + E = p->stop; + + new_p += 2; + } + return res; } - old_mp = (map_t *) map_val(map); - num_old = map_get_size(old_mp); + old_mp = (flatmap_t *) flatmap_val(map); + num_old = flatmap_get_size(old_mp); /* * If the old map is empty, create a new map. @@ -6751,7 +6912,7 @@ update_map_exact(Process* p, Eterm* reg, Eterm map, BeamInstr* I) reg[live] = map; erts_garbage_collect(p, need, reg, live+1); map = reg[live]; - old_mp = (map_t *)map_val(map); + old_mp = (flatmap_t *)flatmap_val(map); } /* @@ -6761,23 +6922,20 @@ update_map_exact(Process* p, Eterm* reg, Eterm map, BeamInstr* I) hp = p->htop; E = p->stop; - old_vals = map_get_values(old_mp); - old_keys = map_get_keys(old_mp); + old_vals = flatmap_get_values(old_mp); + old_keys = flatmap_get_keys(old_mp); - res = make_map(hp); - mp = (map_t *)hp; + res = make_flatmap(hp); + mp = (flatmap_t *)hp; hp += MAP_HEADER_SIZE; mp->thing_word = MAP_HEADER; mp->size = num_old; mp->keys = old_mp->keys; /* Get array of key/value pairs to be updated */ - new_p = &Arg(5); GET_TERM(*new_p, new_key); /* Update all values */ - n = Arg(4) / 2; /* Number of values to be updated */ - ASSERT(n > 0); for (i = 0; i < num_old; i++) { if (!EQ(*old_keys, new_key)) { /* Not same keys */ diff --git a/erts/emulator/beam/beam_load.c b/erts/emulator/beam/beam_load.c index cfc6146b0a..02689e5b19 100644 --- a/erts/emulator/beam/beam_load.c +++ b/erts/emulator/beam/beam_load.c @@ -245,7 +245,7 @@ typedef struct { /* * This structure contains all information about the module being loaded. */ - +#define MD5_SIZE 16 typedef struct LoaderState { /* * The current logical file within the binary. @@ -292,7 +292,7 @@ typedef struct LoaderState { StringPatch* string_patches; /* Linked list of position into string table to patch. */ BeamInstr catches; /* Linked list of catch_yf instructions. */ unsigned loaded_size; /* Final size of code when loaded. */ - byte mod_md5[16]; /* MD5 for module code. */ + byte mod_md5[MD5_SIZE]; /* MD5 for module code. */ int may_load_nif; /* true if NIFs may later be loaded for this module */ int on_load; /* Index in the code for the on_load function * (or 0 if there is no on_load function) @@ -528,6 +528,7 @@ static Eterm exported_from_module(Process* p, Eterm mod); static Eterm functions_in_module(Process* p, Eterm mod); static Eterm attributes_for_module(Process* p, Eterm mod); static Eterm compilation_info_for_module(Process* p, Eterm mod); +static Eterm md5_of_module(Process* p, Eterm mod); static Eterm native_addresses(Process* p, Eterm mod); int patch_funentries(Eterm Patchlist); int patch(Eterm Addresses, Uint fe); @@ -648,6 +649,7 @@ erts_prepare_loading(Binary* magic, Process *c_p, Eterm group_leader, stp->code[MI_COMPILE_PTR] = 0; stp->code[MI_COMPILE_SIZE] = 0; stp->code[MI_COMPILE_SIZE_ON_HEAP] = 0; + stp->code[MI_MD5_PTR] = 0; /* * Read the atom table. @@ -3319,9 +3321,10 @@ gen_select_tuple_arity(LoaderState* stp, GenOpArg S, GenOpArg Fail, { GenOp* op; + GenOpArg *tmp; int arity = Size.val + 3; int size = Size.val / 2; - int i; + int i, j, align = 0; /* * Verify the validity of the list. @@ -3336,9 +3339,37 @@ gen_select_tuple_arity(LoaderState* stp, GenOpArg S, GenOpArg Fail, } /* + * Use a special-cased instruction if there are only two values. + */ + if (size == 2) { + NEW_GENOP(stp, op); + op->next = NULL; + op->op = genop_i_select_tuple_arity2_6; + GENOP_ARITY(op, arity - 1); + op->a[0] = S; + op->a[1] = Fail; + op->a[2].type = TAG_u; + op->a[2].val = Rest[0].val; + op->a[3].type = TAG_u; + op->a[3].val = Rest[2].val; + op->a[4] = Rest[1]; + op->a[5] = Rest[3]; + + return op; + } + + /* * Generate the generic instruction. + * Assumption: + * Few different tuple arities to select on (fewer than 20). + * Use linear scan approach. */ + align = 1; + + arity += 2*align; + size += align; + NEW_GENOP(stp, op); op->next = NULL; op->op = genop_i_select_tuple_arity_3; @@ -3346,39 +3377,36 @@ gen_select_tuple_arity(LoaderState* stp, GenOpArg S, GenOpArg Fail, op->a[0] = S; op->a[1] = Fail; op->a[2].type = TAG_u; - op->a[2].val = Size.val / 2; - for (i = 0; i < Size.val; i += 2) { - op->a[i+3].type = TAG_v; - op->a[i+3].val = make_arityval(Rest[i].val); - op->a[i+4] = Rest[i+1]; - } + op->a[2].val = size; - /* - * Sort the values to make them useful for a binary search. - */ + tmp = (GenOpArg *) erts_alloc(ERTS_ALC_T_LOADER_TMP, sizeof(GenOpArg)*(arity-2*align)); - qsort(op->a+3, size, 2*sizeof(GenOpArg), - (int (*)(const void *, const void *)) genopargcompare); -#ifdef DEBUG - for (i = 3; i < arity-2; i += 2) { - ASSERT(op->a[i].val < op->a[i+2].val); + for (i = 3; i < arity - 2*align; i+=2) { + tmp[i-3].type = TAG_v; + tmp[i-3].val = make_arityval(Rest[i-3].val); + tmp[i-2] = Rest[i-2]; } -#endif /* - * Use a special-cased instruction if there are only two values. + * Sort the values to make them useful for a sentinel search */ - if (size == 2) { - op->op = genop_i_select_tuple_arity2_6; - op->arity--; - op->a[2].type = TAG_u; - op->a[2].val = arityval(op->a[3].val); - op->a[3] = op->a[4]; - op->a[4].type = TAG_u; - op->a[4].val = arityval(op->a[5].val); - op->a[5] = op->a[6]; + + qsort(tmp, size - align, 2*sizeof(GenOpArg), + (int (*)(const void *, const void *)) genopargcompare); + + j = 3; + for (i = 3; i < arity - 2*align; i += 2) { + op->a[j] = tmp[i-3]; + op->a[j + size] = tmp[i-2]; + j++; } + erts_free(ERTS_ALC_T_LOADER_TMP, (void *) tmp); + + op->a[j].type = TAG_u; + op->a[j].val = ~((BeamInstr)0); + op->a[j+size] = Fail; + return op; } @@ -3600,45 +3628,109 @@ gen_select_val(LoaderState* stp, GenOpArg S, GenOpArg Fail, GenOpArg Size, GenOpArg* Rest) { GenOp* op; + GenOpArg *tmp; int arity = Size.val + 3; int size = Size.val / 2; - int i; + int i, j, align = 0; + + if (size == 2) { + + /* + * Use a special-cased instruction if there are only two values. + */ + + NEW_GENOP(stp, op); + op->next = NULL; + op->op = genop_i_select_val2_6; + GENOP_ARITY(op, arity - 1); + op->a[0] = S; + op->a[1] = Fail; + op->a[2] = Rest[0]; + op->a[3] = Rest[2]; + op->a[4] = Rest[1]; + op->a[5] = Rest[3]; + + return op; + + } else if (size > 10) { + + /* binary search instruction */ + + NEW_GENOP(stp, op); + op->next = NULL; + op->op = genop_i_select_val_bins_3; + GENOP_ARITY(op, arity); + op->a[0] = S; + op->a[1] = Fail; + op->a[2].type = TAG_u; + op->a[2].val = size; + for (i = 3; i < arity; i++) { + op->a[i] = Rest[i-3]; + } + + /* + * Sort the values to make them useful for a binary search. + */ + + qsort(op->a+3, size, 2*sizeof(GenOpArg), + (int (*)(const void *, const void *)) genopargcompare); +#ifdef DEBUG + for (i = 3; i < arity-2; i += 2) { + ASSERT(op->a[i].val < op->a[i+2].val); + } +#endif + return op; + } + + /* linear search instruction */ + + align = 1; + + arity += 2*align; + size += align; NEW_GENOP(stp, op); op->next = NULL; - op->op = genop_i_select_val_3; + op->op = genop_i_select_val_lins_3; GENOP_ARITY(op, arity); op->a[0] = S; op->a[1] = Fail; op->a[2].type = TAG_u; op->a[2].val = size; - for (i = 3; i < arity; i++) { - op->a[i] = Rest[i-3]; + + tmp = (GenOpArg *) erts_alloc(ERTS_ALC_T_LOADER_TMP, sizeof(GenOpArg)*(arity-2*align)); + + for (i = 3; i < arity - 2*align; i++) { + tmp[i-3] = Rest[i-3]; } /* - * Sort the values to make them useful for a binary search. + * Sort the values to make them useful for a sentinel search */ - qsort(op->a+3, size, 2*sizeof(GenOpArg), - (int (*)(const void *, const void *)) genopargcompare); -#ifdef DEBUG - for (i = 3; i < arity-2; i += 2) { - ASSERT(op->a[i].val < op->a[i+2].val); + qsort(tmp, size - align, 2*sizeof(GenOpArg), + (int (*)(const void *, const void *)) genopargcompare); + + j = 3; + for (i = 3; i < arity - 2*align; i += 2) { + op->a[j] = tmp[i-3]; + op->a[j+size] = tmp[i-2]; + j++; } -#endif - /* - * Use a special-cased instruction if there are only two values. - */ - if (size == 2) { - op->op = genop_i_select_val2_6; - op->arity--; - op->a[2] = op->a[3]; - op->a[3] = op->a[4]; - op->a[4] = op->a[5]; - op->a[5] = op->a[6]; + erts_free(ERTS_ALC_T_LOADER_TMP, (void *) tmp); + + /* add sentinel */ + + op->a[j].type = TAG_u; + op->a[j].val = ~((BeamInstr)0); + op->a[j+size] = Fail; + +#ifdef DEBUG + for (i = 0; i < size - 1; i++) { + ASSERT(op->a[i+3].val <= op->a[i+4].val); } +#endif return op; } @@ -4042,7 +4134,7 @@ freeze_code(LoaderState* stp) } size = (stp->ci * sizeof(BeamInstr)) + (stp->total_literal_size * sizeof(Eterm)) + - strtab_size + attr_size + compile_size + line_size; + strtab_size + attr_size + compile_size + MD5_SIZE + line_size; /* * Move the code to its final location. @@ -4251,11 +4343,20 @@ freeze_code(LoaderState* stp) code[MI_COMPILE_SIZE_ON_HEAP] = decoded_size; } CHKBLK(ERTS_ALC_T_CODE,code); + { + byte* md5_sum = str_table + strtab_size + attr_size + compile_size; + CHKBLK(ERTS_ALC_T_CODE,code); + sys_memcpy(md5_sum, stp->mod_md5, MD5_SIZE); + CHKBLK(ERTS_ALC_T_CODE,code); + code[MI_MD5_PTR] = (BeamInstr) md5_sum; + CHKBLK(ERTS_ALC_T_CODE,code); + } + CHKBLK(ERTS_ALC_T_CODE,code); /* * Make sure that we have not overflowed the allocated code space. */ - ASSERT(str_table + strtab_size + attr_size + compile_size == + ASSERT(str_table + strtab_size + attr_size + compile_size + MD5_SIZE == ((byte *) code) + size); /* @@ -4971,7 +5072,8 @@ get_tag_and_value(LoaderState* stp, Uint len_code, arity = count/sizeof(Eterm); *result = new_literal(stp, &hp, arity+1); - (void) bytes_to_big(bigbuf, count, neg, hp); + if (is_nil(bytes_to_big(bigbuf, count, neg, hp))) + goto load_error; if (bigbuf != default_buf) { erts_free(ERTS_ALC_T_LOADER_TMP, (void *) bigbuf); @@ -5107,10 +5209,11 @@ erts_module_info_0(Process* p, Eterm module) hp += 3; \ list = CONS(hp, tup, list) + BUILD_INFO(am_md5); BUILD_INFO(am_compile); BUILD_INFO(am_attributes); - BUILD_INFO(am_imports); BUILD_INFO(am_exports); + BUILD_INFO(am_module); #undef BUILD_INFO return list; } @@ -5120,8 +5223,8 @@ erts_module_info_1(Process* p, Eterm module, Eterm what) { if (what == am_module) { return module; - } else if (what == am_imports) { - return NIL; + } else if (what == am_md5) { + return md5_of_module(p, module); } else if (what == am_exports) { return exported_from_module(p, module); } else if (what == am_functions) { @@ -5310,7 +5413,7 @@ attributes_for_module(Process* p, /* Process whose heap to use. */ Eterm result = NIL; Eterm* end; - if (is_not_atom(mod) || (is_not_list(result) && is_not_nil(result))) { + if (is_not_atom(mod)) { return THE_NON_VALUE; } @@ -5349,7 +5452,7 @@ compilation_info_for_module(Process* p, /* Process whose heap to use. */ Eterm result = NIL; Eterm* end; - if (is_not_atom(mod) || (is_not_list(result) && is_not_nil(result))) { + if (is_not_atom(mod)) { return THE_NON_VALUE; } @@ -5372,6 +5475,33 @@ compilation_info_for_module(Process* p, /* Process whose heap to use. */ } /* + * Returns the MD5 checksum for a module + * + * Returns a tagged term, or 0 on error. + */ + +Eterm +md5_of_module(Process* p, /* Process whose heap to use. */ + Eterm mod) /* Tagged atom for module. */ +{ + Module* modp; + BeamInstr* code; + Eterm res = NIL; + + if (is_not_atom(mod)) { + return THE_NON_VALUE; + } + + modp = erts_get_module(mod, erts_active_code_ix()); + if (modp == NULL) { + return THE_NON_VALUE; + } + code = modp->curr.code; + res = new_binary(p, (byte *) code[MI_MD5_PTR], MD5_SIZE); + return res; +} + +/* * Build a single {M,F,A,Loction} item to be part of * a stack trace. */ @@ -5547,7 +5677,7 @@ code_module_md5_1(BIF_ALIST_1) res = am_undefined; goto done; } - res = new_binary(p, stp->mod_md5, sizeof(stp->mod_md5)); + res = new_binary(p, stp->mod_md5, MD5_SIZE); done: erts_free_aligned_binary_bytes(temp_alloc); @@ -5943,6 +6073,7 @@ erts_make_stub_module(Process* p, Eterm Mod, Eterm Beam, Eterm Info) code[MI_LITERALS_END] = 0; code[MI_LITERALS_OFF_HEAP] = 0; code[MI_ON_LOAD_FUNCTION_PTR] = 0; + code[MI_MD5_PTR] = 0; ci = MI_FUNCTIONS + n + 1; /* diff --git a/erts/emulator/beam/beam_load.h b/erts/emulator/beam/beam_load.h index bd22b0c4de..0e3ca0bdb0 100644 --- a/erts/emulator/beam/beam_load.h +++ b/erts/emulator/beam/beam_load.h @@ -91,7 +91,6 @@ extern Uint erts_total_code_size; #define MI_LITERALS_END 8 #define MI_LITERALS_OFF_HEAP 9 - /* * Pointer to the on_load function (or NULL if none). */ @@ -103,6 +102,11 @@ extern Uint erts_total_code_size; #define MI_LINE_TABLE 11 /* + * Pointer to the module MD5 sum (16 bytes) + */ +#define MI_MD5_PTR 12 + +/* * Start of function pointer table. This table contains pointers to * all functions in the module plus an additional pointer just beyond * the end of the last function. @@ -111,7 +115,7 @@ extern Uint erts_total_code_size; * this table. */ -#define MI_FUNCTIONS 12 +#define MI_FUNCTIONS 13 /* * Layout of the line table. diff --git a/erts/emulator/beam/benchmark.c b/erts/emulator/beam/benchmark.c index 8613131176..b16fe6b271 100644 --- a/erts/emulator/beam/benchmark.c +++ b/erts/emulator/beam/benchmark.c @@ -37,37 +37,9 @@ unsigned long long major_gc; #ifdef BM_TIMERS -#if (defined(__i386__) || defined(__x86_64__)) && USE_PERFCTR - -#include "libperfctr.h" -struct vperfctr *system_clock; -double cpu_khz; -BM_NEW_TIMER(start); - -static double get_hrvtime(void) -{ - unsigned long long ticks; - double milli_seconds; - - ticks = vperfctr_read_tsc(system_clock); - milli_seconds = (double)ticks / cpu_khz; - return milli_seconds; -} - -static void stop_hrvtime(void) -{ - if(system_clock) - { - vperfctr_stop(system_clock); - vperfctr_close(system_clock); - system_clock = NULL; - } -} - -#else /* not perfctr, asuming Solaris */ +/* assuming Solaris */ #include <time.h> BM_TIMER_T system_clock; -#endif unsigned long local_pause_times[MAX_PAUSE_TIME]; unsigned long pause_times[MAX_PAUSE_TIME]; @@ -117,40 +89,6 @@ unsigned long long message_sizes[1000]; void init_benchmarking() { #ifdef BM_TIMERS -#if (defined(__i386__) || defined(__x86_64__)) && USE_PERFCTR - /* pass `--with-perfctr=/path/to/perfctr' when configuring */ - struct perfctr_info info; - struct vperfctr_control control; - int i; - - system_clock = vperfctr_open(); - if (system_clock != NULL) - { - if (vperfctr_info(system_clock,&info) >= 0) - { - cpu_khz = (double)info.cpu_khz; - if (info.cpu_features & PERFCTR_FEATURE_RDTSC) - { - memset(&control,0,sizeof control); - control.cpu_control.tsc_on = 1; - } - } - if (vperfctr_control(system_clock,&control) < 0) - { - vperfctr_close(system_clock); - system_clock = NULL; - } - } - - for (i = 0; i < 1000; i++) - { - BM_START_TIMER(system); - BM_STOP_TIMER(system); - } - - timer_time = system_time / 1000; - start_time = 0; -#else int i; for (i = 0; i < 1000; i++) { @@ -158,7 +96,6 @@ void init_benchmarking() BM_STOP_TIMER(system); } timer_time = system_time / 1000; -#endif for (i = 0; i < MAX_PAUSE_TIME; i++) { local_pause_times[i] = 0; diff --git a/erts/emulator/beam/benchmark.h b/erts/emulator/beam/benchmark.h index 7f267b7201..7fc3933f3d 100644 --- a/erts/emulator/beam/benchmark.h +++ b/erts/emulator/beam/benchmark.h @@ -37,10 +37,7 @@ /* BM_TIMERS keeps track of the time spent in diferent parts of the * system. It only measures accual active time, not time spent in idle - * mode. These timers requires hardware support. For Linux, use the - * package perfctr from user.it.uu.se/~mikpe/linux/perfctr. If this - * package is not specified when configuring the system - * (--with-perfctr=PATH), the Solaris hrtime_t will be used. + * mode. Currently, the Solaris hrtime_t will be used. * To add new timers look below. */ #define BM_TIMERS @@ -142,38 +139,7 @@ extern unsigned long long major_gc; * meassure (send time in shared heap for instance). */ -#if (defined(__i386__) || defined(__x86_64__)) && USE_PERFCTR -#include "libperfctr.h" - -#define BM_TIMER_T double - -extern struct vperfctr *system_clock; -extern double cpu_khz; -extern BM_TIMER_T start_time; - -#define BM_START_TIMER(t) start_time = \ - (BM_TIMER_T)vperfctr_read_tsc(system_clock) / \ - cpu_khz; - -#define BM_STOP_TIMER(t) do { \ - BM_TIMER_T tmp = ((BM_TIMER_T)vperfctr_read_tsc(system_clock) / cpu_khz); \ - tmp -= (start_time + timer_time); \ - t##_time += (tmp > 0 ? tmp : 0); \ -} while(0) - -#define BM_TIME_PRINTER(str,time) do { \ - int min,sec,milli,micro; \ - BM_TIMER_T tmp = (time) * 1000; \ - micro = (uint)(tmp - ((int)(tmp / 1000)) * 1000); \ - tmp /= 1000; \ - milli = (uint)(tmp - ((int)(tmp / 1000)) * 1000); \ - tmp /= 1000; \ - sec = (uint)(tmp - ((int)(tmp / 60)) * 60); \ - min = (uint)tmp / 60; \ - erts_fprintf(file,str": %d:%02d.%03d %03d\n",min,sec,milli,micro); \ -} while(0) - -#else /* !USE_PERFCTR (Assuming Solaris) */ +/* (Assuming Solaris) */ #define BM_TIMER_T ErtsMonotonicTime #define BM_START_TIMER(t) system_clock = ERTS_MONOTONIC_TO_NSEC(erts_os_monotonic_time()) @@ -196,7 +162,6 @@ extern BM_TIMER_T start_time; } while(0) extern BM_TIMER_T system_clock; -#endif /* USE_PERFCTR */ extern BM_TIMER_T timer_time; extern BM_TIMER_T system_time; diff --git a/erts/emulator/beam/bif.c b/erts/emulator/beam/bif.c index a48f65d3a2..af02a55b69 100644 --- a/erts/emulator/beam/bif.c +++ b/erts/emulator/beam/bif.c @@ -28,7 +28,9 @@ #include "global.h" #include "erl_process.h" #include "error.h" +#define ERL_WANT_HIPE_BIF_WRAPPER__ #include "bif.h" +#undef ERL_WANT_HIPE_BIF_WRAPPER__ #include "big.h" #include "dist.h" #include "erl_version.h" @@ -47,6 +49,7 @@ static Export* set_cpu_topology_trap = NULL; static Export* await_proc_exit_trap = NULL; static Export* await_port_send_result_trap = NULL; Export* erts_format_cpu_topology_trap = NULL; +static Export dsend_continue_trap_export; Export *erts_convert_time_unit_trap = NULL; static Export *await_sched_wall_time_mod_trap; @@ -1835,6 +1838,8 @@ BIF_RETTYPE whereis_1(BIF_ALIST_1) * erlang:'!'/2 */ +HIPE_WRAPPER_BIF_DISABLE_GC(ebif_bang, 2) + BIF_RETTYPE ebif_bang_2(BIF_ALIST_2) { @@ -1853,34 +1858,36 @@ ebif_bang_2(BIF_ALIST_2) #define SEND_USER_ERROR (-5) #define SEND_INTERNAL_ERROR (-6) #define SEND_AWAIT_RESULT (-7) +#define SEND_YIELD_CONTINUE (-8) + -Sint do_send(Process *p, Eterm to, Eterm msg, int suspend, Eterm *refp); +Sint do_send(Process *p, Eterm to, Eterm msg, Eterm *refp, ErtsSendContext*); static Sint remote_send(Process *p, DistEntry *dep, - Eterm to, Eterm full_to, Eterm msg, int suspend) + Eterm to, Eterm full_to, Eterm msg, + ErtsSendContext* ctx) { Sint res; int code; - ErtsDSigData dsd; ASSERT(is_atom(to) || is_external_pid(to)); - code = erts_dsig_prepare(&dsd, dep, p, ERTS_DSP_NO_LOCK, !suspend); + code = erts_dsig_prepare(&ctx->dsd, dep, p, ERTS_DSP_NO_LOCK, !ctx->suspend); switch (code) { case ERTS_DSIG_PREP_NOT_ALIVE: case ERTS_DSIG_PREP_NOT_CONNECTED: res = SEND_TRAP; break; case ERTS_DSIG_PREP_WOULD_SUSPEND: - ASSERT(!suspend); + ASSERT(!ctx->suspend); res = SEND_YIELD; break; case ERTS_DSIG_PREP_CONNECTED: { if (is_atom(to)) - code = erts_dsig_send_reg_msg(&dsd, to, msg); + code = erts_dsig_send_reg_msg(to, msg, ctx); else - code = erts_dsig_send_msg(&dsd, to, msg); + code = erts_dsig_send_msg(to, msg, ctx); /* * Note that reductions have been bumped on calling * process by erts_dsig_send_reg_msg() or @@ -1888,6 +1895,8 @@ static Sint remote_send(Process *p, DistEntry *dep, */ if (code == ERTS_DSIG_SEND_YIELD) res = SEND_YIELD_RETURN; + else if (code == ERTS_DSIG_SEND_CONTINUE) + res = SEND_YIELD_CONTINUE; else res = 0; break; @@ -1908,7 +1917,8 @@ static Sint remote_send(Process *p, DistEntry *dep, } Sint -do_send(Process *p, Eterm to, Eterm msg, int suspend, Eterm *refp) { +do_send(Process *p, Eterm to, Eterm msg, Eterm *refp, ErtsSendContext* ctx) +{ Eterm portid; Port *pt; Process* rp; @@ -1941,7 +1951,7 @@ do_send(Process *p, Eterm to, Eterm msg, int suspend, Eterm *refp) { #endif return 0; } - return remote_send(p, dep, to, to, msg, suspend); + return remote_send(p, dep, to, to, msg, ctx); } else if (is_atom(to)) { Eterm id = erts_whereis_name_to_id(p, to); @@ -1998,7 +2008,7 @@ do_send(Process *p, Eterm to, Eterm msg, int suspend, Eterm *refp) { ret_val = 0; if (pt) { - int ps_flags = suspend ? 0 : ERTS_PORT_SIG_FLG_NOSUSPEND; + int ps_flags = ctx->suspend ? 0 : ERTS_PORT_SIG_FLG_NOSUSPEND; *refp = NIL; switch (erts_port_command(p, ps_flags, pt, msg, refp)) { @@ -2007,12 +2017,12 @@ do_send(Process *p, Eterm to, Eterm msg, int suspend, Eterm *refp) { return SEND_USER_ERROR; case ERTS_PORT_OP_BUSY: /* Nothing has been sent */ - if (suspend) + if (ctx->suspend) erts_suspend(p, ERTS_PROC_LOCK_MAIN, pt); return SEND_YIELD; case ERTS_PORT_OP_BUSY_SCHEDULED: /* Message was sent */ - if (suspend) { + if (ctx->suspend) { erts_suspend(p, ERTS_PROC_LOCK_MAIN, pt); ret_val = SEND_YIELD_RETURN; break; @@ -2092,9 +2102,14 @@ do_send(Process *p, Eterm to, Eterm msg, int suspend, Eterm *refp) { return 0; } - ret = remote_send(p, dep, tp[1], to, msg, suspend); - if (dep) - erts_deref_dist_entry(dep); + ret = remote_send(p, dep, tp[1], to, msg, ctx); + if (ret != SEND_YIELD_CONTINUE) { + if (dep) { + erts_deref_dist_entry(dep); + } + } else { + ctx->dep_to_deref = dep; + } return ret; } else { if (IS_TRACED(p)) /* XXX Is this really neccessary ??? */ @@ -2125,9 +2140,11 @@ do_send(Process *p, Eterm to, Eterm msg, int suspend, Eterm *refp) { } } +HIPE_WRAPPER_BIF_DISABLE_GC(send, 3) BIF_RETTYPE send_3(BIF_ALIST_3) { + BIF_RETTYPE retval; Eterm ref; Process *p = BIF_P; Eterm to = BIF_ARG_1; @@ -2135,34 +2152,44 @@ BIF_RETTYPE send_3(BIF_ALIST_3) Eterm opts = BIF_ARG_3; int connect = !0; - int suspend = !0; Eterm l = opts; Sint result; - + DeclareTypedTmpHeap(ErtsSendContext, ctx, BIF_P); + UseTmpHeap(sizeof(ErtsSendContext)/sizeof(Eterm), BIF_P); + + ctx->suspend = !0; + ctx->dep_to_deref = NULL; + ctx->return_term = am_ok; + ctx->dss.reds = (Sint) (ERTS_BIF_REDS_LEFT(p) * TERM_TO_BINARY_LOOP_FACTOR); + ctx->dss.phase = ERTS_DSIG_SEND_PHASE_INIT; + while (is_list(l)) { if (CAR(list_val(l)) == am_noconnect) { connect = 0; } else if (CAR(list_val(l)) == am_nosuspend) { - suspend = 0; + ctx->suspend = 0; } else { - BIF_ERROR(p, BADARG); + ERTS_BIF_PREP_ERROR(retval, p, BADARG); + goto done; } l = CDR(list_val(l)); } if(!is_nil(l)) { - BIF_ERROR(p, BADARG); + ERTS_BIF_PREP_ERROR(retval, p, BADARG); + goto done; } #ifdef DEBUG ref = NIL; #endif - result = do_send(p, to, msg, suspend, &ref); + result = do_send(p, to, msg, &ref, ctx); if (result > 0) { ERTS_VBUMP_REDS(p, result); if (ERTS_IS_PROC_OUT_OF_REDS(p)) goto yield_return; - BIF_RET(am_ok); + ERTS_BIF_PREP_RET(retval, am_ok); + goto done; } switch (result) { @@ -2170,68 +2197,127 @@ BIF_RETTYPE send_3(BIF_ALIST_3) /* May need to yield even though we do not bump reds here... */ if (ERTS_IS_PROC_OUT_OF_REDS(p)) goto yield_return; - BIF_RET(am_ok); + ERTS_BIF_PREP_RET(retval, am_ok); break; case SEND_TRAP: if (connect) { - BIF_TRAP3(dsend3_trap, p, to, msg, opts); + ERTS_BIF_PREP_TRAP3(retval, dsend3_trap, p, to, msg, opts); } else { - BIF_RET(am_noconnect); + ERTS_BIF_PREP_RET(retval, am_noconnect); } break; case SEND_YIELD: - if (suspend) { - ERTS_BIF_YIELD3(bif_export[BIF_send_3], p, to, msg, opts); + if (ctx->suspend) { + ERTS_BIF_PREP_YIELD3(retval, + bif_export[BIF_send_3], p, to, msg, opts); } else { - BIF_RET(am_nosuspend); + ERTS_BIF_PREP_RET(retval, am_nosuspend); } break; case SEND_YIELD_RETURN: - if (!suspend) - BIF_RET(am_nosuspend); + if (!ctx->suspend) { + ERTS_BIF_PREP_RET(retval, am_nosuspend); + break; + } yield_return: - ERTS_BIF_YIELD_RETURN(p, am_ok); + ERTS_BIF_PREP_YIELD_RETURN(retval, p, am_ok); + break; case SEND_AWAIT_RESULT: ASSERT(is_internal_ref(ref)); - BIF_TRAP3(await_port_send_result_trap, p, ref, am_nosuspend, am_ok); + ERTS_BIF_PREP_TRAP3(retval, await_port_send_result_trap, p, ref, am_nosuspend, am_ok); + break; case SEND_BADARG: - BIF_ERROR(p, BADARG); + ERTS_BIF_PREP_ERROR(retval, p, BADARG); break; case SEND_USER_ERROR: - BIF_ERROR(p, EXC_ERROR); + ERTS_BIF_PREP_ERROR(retval, p, EXC_ERROR); break; case SEND_INTERNAL_ERROR: - BIF_ERROR(p, EXC_INTERNAL_ERROR); + ERTS_BIF_PREP_ERROR(retval, p, EXC_INTERNAL_ERROR); + break; + case SEND_YIELD_CONTINUE: + BUMP_ALL_REDS(p); + erts_set_gc_state(p, 0); + ERTS_BIF_PREP_TRAP1(retval, &dsend_continue_trap_export, p, + erts_dsend_export_trap_context(p, ctx)); break; default: - ASSERT(! "Illegal send result"); + erl_exit(ERTS_ABORT_EXIT, "send_3 invalid result %d\n", (int)result); break; } - ASSERT(! "Can not arrive here"); - BIF_ERROR(p, BADARG); + +done: + UnUseTmpHeap(sizeof(ErtsSendContext)/sizeof(Eterm), BIF_P); + return retval; } +HIPE_WRAPPER_BIF_DISABLE_GC(send, 2) + BIF_RETTYPE send_2(BIF_ALIST_2) { return erl_send(BIF_P, BIF_ARG_1, BIF_ARG_2); } +static BIF_RETTYPE dsend_continue_trap_1(BIF_ALIST_1) +{ + Binary* bin = ((ProcBin*) binary_val(BIF_ARG_1))->val; + ErtsSendContext* ctx = (ErtsSendContext*) ERTS_MAGIC_BIN_DATA(bin); + Sint initial_reds = (Sint) (ERTS_BIF_REDS_LEFT(BIF_P) * TERM_TO_BINARY_LOOP_FACTOR); + int result; + + ASSERT(ERTS_MAGIC_BIN_DESTRUCTOR(bin) == erts_dsend_context_dtor); + + ctx->dss.reds = initial_reds; + result = erts_dsig_send(&ctx->dsd, &ctx->dss); + + switch (result) { + case ERTS_DSIG_SEND_OK: + erts_set_gc_state(BIF_P, 1); + BIF_RET(ctx->return_term); + break; + case ERTS_DSIG_SEND_YIELD: /*SEND_YIELD_RETURN*/ + erts_set_gc_state(BIF_P, 1); + if (!ctx->suspend) + BIF_RET(am_nosuspend); + ERTS_BIF_YIELD_RETURN(BIF_P, ctx->return_term); + + case ERTS_DSIG_SEND_CONTINUE: { /*SEND_YIELD_CONTINUE*/ + BUMP_ALL_REDS(BIF_P); + BIF_TRAP1(&dsend_continue_trap_export, BIF_P, BIF_ARG_1); + } + default: + erl_exit(ERTS_ABORT_EXIT, "dsend_continue_trap invalid result %d\n", (int)result); + break; + } + ASSERT(! "Can not arrive here"); + BIF_ERROR(BIF_P, BADARG); +} + Eterm erl_send(Process *p, Eterm to, Eterm msg) { + Eterm retval; Eterm ref; Sint result; + DeclareTypedTmpHeap(ErtsSendContext, ctx, p); + UseTmpHeap(sizeof(ErtsSendContext)/sizeof(Eterm), p); #ifdef DEBUG ref = NIL; #endif + ctx->suspend = !0; + ctx->dep_to_deref = NULL; + ctx->return_term = msg; + ctx->dss.reds = (Sint) (ERTS_BIF_REDS_LEFT(p) * TERM_TO_BINARY_LOOP_FACTOR); + ctx->dss.phase = ERTS_DSIG_SEND_PHASE_INIT; - result = do_send(p, to, msg, !0, &ref); + result = do_send(p, to, msg, &ref, ctx); if (result > 0) { ERTS_VBUMP_REDS(p, result); if (ERTS_IS_PROC_OUT_OF_REDS(p)) goto yield_return; - BIF_RET(msg); + ERTS_BIF_PREP_RET(retval, msg); + goto done; } switch (result) { @@ -2239,35 +2325,46 @@ Eterm erl_send(Process *p, Eterm to, Eterm msg) /* May need to yield even though we do not bump reds here... */ if (ERTS_IS_PROC_OUT_OF_REDS(p)) goto yield_return; - BIF_RET(msg); + ERTS_BIF_PREP_RET(retval, msg); break; case SEND_TRAP: - BIF_TRAP2(dsend2_trap, p, to, msg); + ERTS_BIF_PREP_TRAP2(retval, dsend2_trap, p, to, msg); break; case SEND_YIELD: - ERTS_BIF_YIELD2(bif_export[BIF_send_2], p, to, msg); + ERTS_BIF_PREP_YIELD2(retval, bif_export[BIF_send_2], p, to, msg); break; case SEND_YIELD_RETURN: yield_return: - ERTS_BIF_YIELD_RETURN(p, msg); + ERTS_BIF_PREP_YIELD_RETURN(retval, p, msg); + break; case SEND_AWAIT_RESULT: ASSERT(is_internal_ref(ref)); - BIF_TRAP3(await_port_send_result_trap, p, ref, msg, msg); + ERTS_BIF_PREP_TRAP3(retval, + await_port_send_result_trap, p, ref, msg, msg); + break; case SEND_BADARG: - BIF_ERROR(p, BADARG); + ERTS_BIF_PREP_ERROR(retval, p, BADARG); break; case SEND_USER_ERROR: - BIF_ERROR(p, EXC_ERROR); + ERTS_BIF_PREP_ERROR(retval, p, EXC_ERROR); break; case SEND_INTERNAL_ERROR: - BIF_ERROR(p, EXC_INTERNAL_ERROR); + ERTS_BIF_PREP_ERROR(retval, p, EXC_INTERNAL_ERROR); + break; + case SEND_YIELD_CONTINUE: + BUMP_ALL_REDS(p); + erts_set_gc_state(p, 0); + ERTS_BIF_PREP_TRAP1(retval, &dsend_continue_trap_export, p, + erts_dsend_export_trap_context(p, ctx)); break; default: - ASSERT(! "Illegal send result"); + erl_exit(ERTS_ABORT_EXIT, "invalid send result %d\n", (int)result); break; } - ASSERT(! "Can not arrive here"); - BIF_ERROR(p, BADARG); + +done: + UnUseTmpHeap(sizeof(ErtsSendContext)/sizeof(Eterm), p); + return retval; } /**********************************************************************/ @@ -3969,16 +4066,19 @@ BIF_RETTYPE halt_2(BIF_ALIST_2) BIF_RETTYPE function_exported_3(BIF_ALIST_3) { + int arity; if (is_not_atom(BIF_ARG_1) || is_not_atom(BIF_ARG_2) || is_not_small(BIF_ARG_3)) { BIF_ERROR(BIF_P, BADARG); } - if (erts_find_function(BIF_ARG_1, BIF_ARG_2, signed_val(BIF_ARG_3), - erts_active_code_ix()) == NULL) { - BIF_RET(am_false); + arity = signed_val(BIF_ARG_3); + if (erts_find_function(BIF_ARG_1, BIF_ARG_2, arity, + erts_active_code_ix()) != NULL || + erts_is_builtin(BIF_ARG_1, BIF_ARG_2, arity)) { + BIF_RET(am_true); } - BIF_RET(am_true); + BIF_RET(am_false); } /**********************************************************************/ @@ -4806,6 +4906,10 @@ void erts_init_bif(void) #endif , &bif_return_trap); + erts_init_trap_export(&dsend_continue_trap_export, + am_erts_internal, am_dsend_continue_trap, 1, + dsend_continue_trap_1); + flush_monitor_messages_trap = erts_export_put(am_erts_internal, am_flush_monitor_messages, 3); diff --git a/erts/emulator/beam/bif.h b/erts/emulator/beam/bif.h index bd0f8cda2b..d461c3f479 100644 --- a/erts/emulator/beam/bif.h +++ b/erts/emulator/beam/bif.h @@ -31,10 +31,12 @@ extern Export *erts_convert_time_unit_trap; #define BIF_ALIST_1 Process* A__p, Eterm* BIF__ARGS #define BIF_ALIST_2 Process* A__p, Eterm* BIF__ARGS #define BIF_ALIST_3 Process* A__p, Eterm* BIF__ARGS +#define BIF_ALIST_4 Process* A__p, Eterm* BIF__ARGS #define BIF_ARG_1 (BIF__ARGS[0]) #define BIF_ARG_2 (BIF__ARGS[1]) #define BIF_ARG_3 (BIF__ARGS[2]) +#define BIF_ARG_4 (BIF__ARGS[3]) #define ERTS_IS_PROC_OUT_OF_REDS(p) \ ((p)->fcalls > 0 \ @@ -466,6 +468,8 @@ erts_bif_prep_await_proc_exit_apply_trap(Process *c_p, Eterm args[], int nargs); +#ifdef ERL_WANT_HIPE_BIF_WRAPPER__ + #ifndef HIPE #define HIPE_WRAPPER_BIF_DISABLE_GC(BIF_NAME, ARITY) @@ -510,6 +514,7 @@ BIF_RETTYPE hipe_wrapper_ ## BIF_NAME ## _ ## ARITY (Process* c_p, \ #endif +#endif /* ERL_WANT_HIPE_BIF_WRAPPER__ */ #include "erl_bif_table.h" diff --git a/erts/emulator/beam/bif.tab b/erts/emulator/beam/bif.tab index 8ac1fba733..471f687101 100644 --- a/erts/emulator/beam/bif.tab +++ b/erts/emulator/beam/bif.tab @@ -166,6 +166,8 @@ bif erts_internal:request_system_task/3 bif erts_internal:check_process_code/2 bif erts_internal:map_to_tuple_keys/1 +bif erts_internal:map_type/1 +bif erts_internal:map_hashmap_children/1 bif erts_internal:time_unit/0 @@ -212,6 +214,7 @@ bif math:erf/1 bif math:erfc/1 bif math:exp/1 bif math:log/1 +bif math:log2/1 bif math:log10/1 bif math:sqrt/1 bif math:atan2/2 @@ -593,7 +596,7 @@ bif io:printable_range/0 bif os:unsetenv/1 # -# New in R17A +# New in 17.0 # bif re:inspect/2 @@ -615,10 +618,21 @@ bif maps:values/1 bif erts_internal:cmp_term/2 +bif ets:take/2 + # -# New in 17.1. +# New in 17.1 # + bif erlang:fun_info_mfa/1 + +# New in 18.0 +# + +bif erlang:get_keys/0 +bif ets:update_counter/4 +bif erts_debug:map_info/1 + # # Obsolete # diff --git a/erts/emulator/beam/big.c b/erts/emulator/beam/big.c index 30e6a2c522..a4ea9c59ca 100644 --- a/erts/emulator/beam/big.c +++ b/erts/emulator/beam/big.c @@ -1940,6 +1940,8 @@ Eterm bytes_to_big(byte *xp, dsize_t xsz, int xsgn, Eterm *r) *rwp = d; rwp++; } + if (rsz > BIG_ARITY_MAX) + return NIL; if (xsgn) { *r = make_neg_bignum_header(rsz); } diff --git a/erts/emulator/beam/binary.c b/erts/emulator/beam/binary.c index f50d484576..cc0b3b9b6c 100644 --- a/erts/emulator/beam/binary.c +++ b/erts/emulator/beam/binary.c @@ -26,7 +26,9 @@ #include "global.h" #include "erl_process.h" #include "error.h" +#define ERL_WANT_HIPE_BIF_WRAPPER__ #include "bif.h" +#undef ERL_WANT_HIPE_BIF_WRAPPER__ #include "big.h" #include "erl_binary.h" #include "erl_bits.h" @@ -83,8 +85,6 @@ new_binary(Process *p, byte *buf, Uint len) * Allocate the binary struct itself. */ bptr = erts_bin_nrml_alloc(len); - bptr->flags = 0; - bptr->orig_size = len; erts_refc_init(&bptr->refc, 1); if (buf != NULL) { sys_memcpy(bptr->orig_bytes, buf, len); @@ -122,8 +122,6 @@ Eterm erts_new_mso_binary(Process *p, byte *buf, int len) * Allocate the binary struct itself. */ bptr = erts_bin_nrml_alloc(len); - bptr->flags = 0; - bptr->orig_size = len; erts_refc_init(&bptr->refc, 1); if (buf != NULL) { sys_memcpy(bptr->orig_bytes, buf, len); @@ -177,7 +175,6 @@ erts_realloc_binary(Eterm bin, size_t size) } else { /* REFC */ ProcBin* pb = (ProcBin *) bval; Binary* newbin = erts_bin_realloc(pb->val, size); - newbin->orig_size = size; pb->val = newbin; pb->size = size; pb->bytes = (byte*) newbin->orig_bytes; diff --git a/erts/emulator/beam/break.c b/erts/emulator/beam/break.c index 08265b590d..e2fa572546 100644 --- a/erts/emulator/beam/break.c +++ b/erts/emulator/beam/break.c @@ -181,7 +181,6 @@ static void doit_print_monitor(ErtsMonitor *mon, void *vpcontext) ASSERT(is_node_name_atom(mon->pid)); erts_print(to, to_arg, "%s{to,{%T,%T},%T}", prefix, mon->name, mon->pid, mon->ref); - erts_print(to, to_arg,"}"); } else if (is_atom(mon->name)){ /* local by name */ erts_print(to, to_arg, "%s{to,{%T,%T},%T}", prefix, mon->name, erts_this_dist_entry->sysname, mon->ref); @@ -210,25 +209,12 @@ print_process_info(int to, void *to_arg, Process *p) erts_print(to, to_arg, "State: "); state = erts_smp_atomic32_read_acqb(&p->state); - if (state & ERTS_PSFLG_FREE) - erts_print(to, to_arg, "Non Existing\n"); /* Should never happen */ - else if (state & ERTS_PSFLG_EXITING) - erts_print(to, to_arg, "Exiting\n"); - else if (state & ERTS_PSFLG_GC) { - garbing = 1; - running = 1; - erts_print(to, to_arg, "Garbing\n"); - } - else if (state & ERTS_PSFLG_SUSPENDED) - erts_print(to, to_arg, "Suspended\n"); - else if (state & ERTS_PSFLG_RUNNING) { - running = 1; - erts_print(to, to_arg, "Running\n"); - } - else if (state & ERTS_PSFLG_ACTIVE) - erts_print(to, to_arg, "Scheduled\n"); - else - erts_print(to, to_arg, "Waiting\n"); + erts_dump_process_state(to, to_arg, state); + if (state & ERTS_PSFLG_GC) { + garbing = 1; + running = 1; + } else if (state & ERTS_PSFLG_RUNNING) + running = 1; /* * If the process is registered as a global process, display the @@ -352,6 +338,10 @@ print_process_info(int to, void *to_arg, Process *p) #endif erts_stack_dump(to, to_arg, p); } + + /* Display all states */ + erts_print(to, to_arg, "Internal State: "); + erts_dump_extended_process_state(to, to_arg, state); } static void @@ -681,27 +671,39 @@ erl_crash_dump_v(char *file, int line, char* fmt, va_list args) char* dumpname; int secs; int env_erl_crash_dump_seconds_set = 1; + int i; if (ERTS_SOMEONE_IS_CRASH_DUMPING) return; #ifdef ERTS_SMP + /* Order all managed threads to block, this has to be done + first to guarantee that this is the only thread to generate + crash dump. */ + erts_thr_progress_fatal_error_block(&tpd_buf); + +#ifdef ERTS_THR_HAVE_SIG_FUNCS /* - * Wait for all managed threads to block. If all threads haven't blocked - * after a minute, we go anyway and hope for the best... - * - * We do not release system again. We expect an exit() or abort() after - * dump has been written. + * We suspend all scheduler threads so that we can dump some + * data about the currently running processes and scheduler data. + * We have to be very very careful when doing this as the schedulers + * could be anywhere. */ - erts_thr_progress_fatal_error_block(60000, &tpd_buf); - /* Either worked or not... */ + for (i = 0; i < erts_no_schedulers; i++) { + erts_tid_t tid = ERTS_SCHEDULER_IX(i)->tid; + if (!erts_equal_tids(tid,erts_thr_self())) + sys_thr_suspend(tid); + } + +#endif /* Allow us to pass certain places without locking... */ erts_smp_atomic32_set_mb(&erts_writing_erl_crash_dump, 1); erts_smp_tsd_set(erts_is_crash_dumping_key, (void *) 1); -#else + +#else /* !ERTS_SMP */ erts_writing_erl_crash_dump = 1; -#endif +#endif /* ERTS_SMP */ envsz = sizeof(env); /* ERL_CRASH_DUMP_SECONDS not set @@ -754,11 +756,12 @@ erl_crash_dump_v(char *file, int line, char* fmt, va_list args) dumpname = "erl_crash.dump"; else dumpname = &dumpnamebuf[0]; + + erts_fprintf(stderr,"\nCrash dump is being written to: %s...", dumpname); fd = open(dumpname,O_WRONLY | O_CREAT | O_TRUNC,0640); - if (fd < 0) + if (fd < 0) return; /* Can't create the crash dump, skip it */ - time(&now); erts_fdprintf(fd, "=erl_crash_dump:0.3\n%s", ctime(&now)); @@ -772,9 +775,74 @@ erl_crash_dump_v(char *file, int line, char* fmt, va_list args) erts_fdprintf(fd, "System version: "); erts_print_system_version(fd, NULL, NULL); erts_fdprintf(fd, "%s\n", "Compiled: " ERLANG_COMPILE_DATE); + erts_fdprintf(fd, "Taints: "); erts_print_nif_taints(fd, NULL); erts_fdprintf(fd, "Atoms: %d\n", atom_table_size()); + +#ifdef USE_THREADS + /* We want to note which thread it was that called erl_exit */ + if (erts_get_scheduler_data()) { + erts_fdprintf(fd, "Calling Thread: scheduler:%d\n", + erts_get_scheduler_data()->no); + } else { + if (!erts_thr_getname(erts_thr_self(), dumpnamebuf, MAXPATHLEN)) + erts_fdprintf(fd, "Calling Thread: %s\n", dumpnamebuf); + else + erts_fdprintf(fd, "Calling Thread: %p\n", erts_thr_self()); + } +#else + erts_fdprintf(fd, "Calling Thread: scheduler:1\n"); +#endif + +#if defined(ERTS_HAVE_TRY_CATCH) + + /* + * erts_print_scheduler_info is not guaranteed to be safe to call + * here for all schedulers as we may have suspended a scheduler + * in the middle of updating the STACK_TOP and STACK_START + * variables and thus when scanning the stack we could get + * segmentation faults. We protect against this very unlikely + * scenario by using the ERTS_SYS_TRY_CATCH. + */ + for (i = 0; i < erts_no_schedulers; i++) { + ERTS_SYS_TRY_CATCH( + erts_print_scheduler_info(fd, NULL, ERTS_SCHEDULER_IX(i)), + erts_fdprintf(fd, "** crashed **\n")); + } +#endif + +#ifdef ERTS_SMP + +#if defined(ERTS_THR_HAVE_SIG_FUNCS) + + /* We resume all schedulers so that we are in a known safe state + when we write the rest of the crash dump */ + for (i = 0; i < erts_no_schedulers; i++) { + erts_tid_t tid = ERTS_SCHEDULER_IX(i)->tid; + if (!erts_equal_tids(tid,erts_thr_self())) + sys_thr_resume(tid); + } +#endif + + /* + * Wait for all managed threads to block. If all threads haven't blocked + * after a minute, we go anyway and hope for the best... + * + * We do not release system again. We expect an exit() or abort() after + * dump has been written. + */ + erts_thr_progress_fatal_error_wait(60000); + /* Either worked or not... */ +#endif + +#ifndef ERTS_HAVE_TRY_CATCH + /* This is safe to call here, as all schedulers are blocked */ + for (i = 0; i < erts_no_schedulers; i++) { + erts_print_scheduler_info(fd, NULL, ERTS_SCHEDULER_IX(i)); + } +#endif + info(fd, NULL); /* General system info */ if (erts_ptab_initialized(&erts_proc)) process_info(fd, NULL); /* Info about each process and port */ @@ -804,7 +872,7 @@ erl_crash_dump_v(char *file, int line, char* fmt, va_list args) erts_fdprintf(fd, "=end\n"); close(fd); - erts_fprintf(stderr,"\nCrash dump was written to: %s\n", dumpname); + erts_fprintf(stderr,"done\n"); } void diff --git a/erts/emulator/beam/copy.c b/erts/emulator/beam/copy.c index 50548850eb..027b85b079 100644 --- a/erts/emulator/beam/copy.c +++ b/erts/emulator/beam/copy.c @@ -21,6 +21,8 @@ # include "config.h" #endif +#define ERL_WANT_GC_INTERNALS__ + #include "sys.h" #include "erl_vm.h" #include "global.h" @@ -125,6 +127,35 @@ Uint size_object(Eterm obj) obj = *bptr; break; } + case HASHMAP_SUBTAG: + switch (MAP_HEADER_TYPE(hdr)) { + case MAP_HEADER_TAG_HAMT_HEAD_BITMAP : + case MAP_HEADER_TAG_HAMT_HEAD_ARRAY : + case MAP_HEADER_TAG_HAMT_NODE_BITMAP : + { + Eterm *head; + Uint sz; + head = hashmap_val_rel(obj, base); + sz = hashmap_bitcount(MAP_HEADER_VAL(hdr)); + sum += 1 + sz + header_arity(hdr); + head += 1 + header_arity(hdr); + + if (sz == 0) { + goto pop_next; + } + while(sz-- > 1) { + obj = head[sz]; + if (!IS_CONST(obj)) { + ESTACK_PUSH(s, obj); + } + } + obj = head[0]; + } + break; + default: + erl_exit(ERTS_ABORT_EXIT, "size_object: bad hashmap type %d\n", MAP_HEADER_TYPE(hdr)); + } + break; case SUB_BINARY_SUBTAG: { Eterm real_bin; @@ -155,10 +186,10 @@ Uint size_object(Eterm obj) case MAP_SUBTAG: { Uint n; - map_t *mp; - mp = (map_t*)map_val_rel(obj,base); + flatmap_t *mp; + mp = (flatmap_t*)flatmap_val_rel(obj,base); ptr = (Eterm *)mp; - n = map_get_size(mp) + 1; + n = flatmap_get_size(mp) + 1; sum += n + 2; ptr += 2; /* hdr + size words */ while (n--) { @@ -340,8 +371,8 @@ Eterm copy_struct(Eterm obj, Uint sz, Eterm** hpp, ErlOffHeap* off_heap) break; case MAP_SUBTAG: { - i = map_get_size(objp) + 3; - *argp = make_map_rel(htop, dst_base); + i = flatmap_get_size(objp) + 3; + *argp = make_flatmap_rel(htop, dst_base); while (i--) { *htop++ = *objp++; } @@ -457,7 +488,7 @@ Eterm copy_struct(Eterm obj, Uint sz, Eterm** hpp, ErlOffHeap* off_heap) { ExternalThing *etp = (ExternalThing *) htop; - i = thing_arityval(hdr) + 1; + i = thing_arityval(hdr) + 1; tp = htop; while (i--) { @@ -471,6 +502,21 @@ Eterm copy_struct(Eterm obj, Uint sz, Eterm** hpp, ErlOffHeap* off_heap) *argp = make_external_rel(tp, dst_base); } break; + case HASHMAP_SUBTAG: + tp = htop; + switch (MAP_HEADER_TYPE(hdr)) { + case MAP_HEADER_TAG_HAMT_HEAD_BITMAP : + case MAP_HEADER_TAG_HAMT_HEAD_ARRAY : + *htop++ = *objp++; + case MAP_HEADER_TAG_HAMT_NODE_BITMAP : + i = 1 + hashmap_bitcount(MAP_HEADER_VAL(hdr)); + while (i--) { *htop++ = *objp++; } + *argp = make_hashmap_rel(tp, dst_base); + break; + default: + erl_exit(ERTS_ABORT_EXIT, "copy_struct: bad hashmap type %d\n", MAP_HEADER_TYPE(hdr)); + } + break; case BIN_MATCHSTATE_SUBTAG: erl_exit(ERTS_ABORT_EXIT, "copy_struct: matchstate term not allowed"); diff --git a/erts/emulator/beam/dist.c b/erts/emulator/beam/dist.c index ec07ddcd9c..bfecac1612 100644 --- a/erts/emulator/beam/dist.c +++ b/erts/emulator/beam/dist.c @@ -1,7 +1,7 @@ /* * %CopyrightBegin% * - * Copyright Ericsson AB 1996-2013. All Rights Reserved. + * Copyright Ericsson AB 1996-2014. All Rights Reserved. * * The contents of this file are subject to the Erlang Public License, * Version 1.1, (the "License"); you may not use this file except in @@ -119,7 +119,7 @@ Export* dmonitor_p_trap = NULL; /* forward declarations */ static void clear_dist_entry(DistEntry*); -static int dsig_send(ErtsDSigData *, Eterm, Eterm, int); +static int dsig_send_ctl(ErtsDSigData* dsdp, Eterm ctl, int force_busy); static void send_nodes_mon_msgs(Process *, Eterm, Eterm, Eterm, Eterm); static void init_nodes_monitors(void); @@ -622,9 +622,7 @@ alloc_dist_obuf(Uint size) ErtsDistOutputBuf *obuf; Uint obuf_size = sizeof(ErtsDistOutputBuf)+sizeof(byte)*(size-1); Binary *bin = erts_bin_drv_alloc(obuf_size); - bin->flags = BIN_FLAG_DRV; erts_refc_init(&bin->refc, 1); - bin->orig_size = (SWord) obuf_size; obuf = (ErtsDistOutputBuf *) &bin->orig_bytes[0]; #ifdef DEBUG obuf->dbg_pattern = ERTS_DIST_OUTPUT_BUF_DBG_PATTERN; @@ -709,6 +707,55 @@ static void clear_dist_entry(DistEntry *dep) } } +void erts_dsend_context_dtor(Binary* ctx_bin) +{ + ErtsSendContext* ctx = ERTS_MAGIC_BIN_DATA(ctx_bin); + switch (ctx->dss.phase) { + case ERTS_DSIG_SEND_PHASE_MSG_SIZE: + DESTROY_SAVED_ESTACK(&ctx->dss.u.sc.estack); + break; + case ERTS_DSIG_SEND_PHASE_MSG_ENCODE: + DESTROY_SAVED_WSTACK(&ctx->dss.u.ec.wstack); + break; + default:; + } + if (ctx->dss.phase >= ERTS_DSIG_SEND_PHASE_ALLOC && ctx->dss.obuf) { + free_dist_obuf(ctx->dss.obuf); + } + if (ctx->dep_to_deref) + erts_deref_dist_entry(ctx->dep_to_deref); +} + +Eterm erts_dsend_export_trap_context(Process* p, ErtsSendContext* ctx) +{ + struct exported_ctx { + ErtsSendContext ctx; + ErtsAtomCacheMap acm; + }; + Binary* ctx_bin = erts_create_magic_binary(sizeof(struct exported_ctx), + erts_dsend_context_dtor); + struct exported_ctx* dst = ERTS_MAGIC_BIN_DATA(ctx_bin); + Uint ctl_size = !HALFWORD_HEAP ? 0 : (arityval(ctx->ctl_heap[0]) + 1); + Eterm* hp = HAlloc(p, ctl_size + PROC_BIN_SIZE); + + sys_memcpy(&dst->ctx, ctx, sizeof(ErtsSendContext)); + ASSERT(ctx->dss.ctl == make_tuple(ctx->ctl_heap)); +#if !HALFWORD_HEAP + dst->ctx.dss.ctl = make_tuple(dst->ctx.ctl_heap); +#else + /* Must put control tuple in low mem */ + sys_memcpy(hp, ctx->ctl_heap, ctl_size*sizeof(Eterm)); + dst->ctx.dss.ctl = make_tuple(hp); + hp += ctl_size; +#endif + if (ctx->dss.acmp) { + sys_memcpy(&dst->acm, ctx->dss.acmp, sizeof(ErtsAtomCacheMap)); + dst->ctx.dss.acmp = &dst->acm; + } + return erts_mk_magic_binary_term(&hp, &MSO(p), ctx_bin); +} + + /* * The erts_dsig_send_*() functions implemented below, sends asynchronous * distributed signals to other Erlang nodes. Before sending a distributed @@ -731,7 +778,7 @@ erts_dsig_send_link(ErtsDSigData *dsdp, Eterm local, Eterm remote) int res; UseTmpHeapNoproc(4); - res = dsig_send(dsdp, ctl, THE_NON_VALUE, 0); + res = dsig_send_ctl(dsdp, ctl, 0); UnUseTmpHeapNoproc(4); return res; } @@ -744,7 +791,7 @@ erts_dsig_send_unlink(ErtsDSigData *dsdp, Eterm local, Eterm remote) int res; UseTmpHeapNoproc(4); - res = dsig_send(dsdp, ctl, THE_NON_VALUE, 0); + res = dsig_send_ctl(dsdp, ctl, 0); UnUseTmpHeapNoproc(4); return res; } @@ -772,7 +819,7 @@ erts_dsig_send_m_exit(ErtsDSigData *dsdp, Eterm watcher, Eterm watched, erts_smp_de_links_unlock(dsdp->dep); #endif - res = dsig_send(dsdp, ctl, THE_NON_VALUE, 1); + res = dsig_send_ctl(dsdp, ctl, 1); UnUseTmpHeapNoproc(6); return res; } @@ -793,7 +840,7 @@ erts_dsig_send_monitor(ErtsDSigData *dsdp, Eterm watcher, Eterm watched, make_small(DOP_MONITOR_P), watcher, watched, ref); - res = dsig_send(dsdp, ctl, THE_NON_VALUE, 0); + res = dsig_send_ctl(dsdp, ctl, 0); UnUseTmpHeapNoproc(5); return res; } @@ -815,18 +862,17 @@ erts_dsig_send_demonitor(ErtsDSigData *dsdp, Eterm watcher, make_small(DOP_DEMONITOR_P), watcher, watched, ref); - res = dsig_send(dsdp, ctl, THE_NON_VALUE, force); + res = dsig_send_ctl(dsdp, ctl, force); UnUseTmpHeapNoproc(5); return res; } int -erts_dsig_send_msg(ErtsDSigData *dsdp, Eterm remote, Eterm message) +erts_dsig_send_msg(Eterm remote, Eterm message, ErtsSendContext* ctx) { Eterm ctl; - DeclareTmpHeapNoproc(ctl_heap,5); Eterm token = NIL; - Process *sender = dsdp->proc; + Process *sender = ctx->dsd.proc; int res; #ifdef USE_VM_PROBES Sint tok_label = 0; @@ -838,8 +884,7 @@ erts_dsig_send_msg(ErtsDSigData *dsdp, Eterm remote, Eterm message) DTRACE_CHARBUF(receiver_name, 64); #endif - UseTmpHeapNoproc(5); - if (SEQ_TRACE_TOKEN(sender) != NIL + if (SEQ_TRACE_TOKEN(sender) != NIL #ifdef USE_VM_PROBES && SEQ_TRACE_TOKEN(sender) != am_have_dt_utag #endif @@ -852,7 +897,7 @@ erts_dsig_send_msg(ErtsDSigData *dsdp, Eterm remote, Eterm message) *node_name = *sender_name = *receiver_name = '\0'; if (DTRACE_ENABLED(message_send) || DTRACE_ENABLED(message_send_remote)) { erts_snprintf(node_name, sizeof(DTRACE_CHARBUF_NAME(node_name)), - "%T", dsdp->dep->sysname); + "%T", ctx->dsd.dep->sysname); erts_snprintf(sender_name, sizeof(DTRACE_CHARBUF_NAME(sender_name)), "%T", sender->common.id); erts_snprintf(receiver_name, sizeof(DTRACE_CHARBUF_NAME(receiver_name)), @@ -867,26 +912,28 @@ erts_dsig_send_msg(ErtsDSigData *dsdp, Eterm remote, Eterm message) #endif if (token != NIL) - ctl = TUPLE4(&ctl_heap[0], + ctl = TUPLE4(&ctx->ctl_heap[0], make_small(DOP_SEND_TT), am_Cookie, remote, token); else - ctl = TUPLE3(&ctl_heap[0], make_small(DOP_SEND), am_Cookie, remote); + ctl = TUPLE3(&ctx->ctl_heap[0], make_small(DOP_SEND), am_Cookie, remote); DTRACE6(message_send, sender_name, receiver_name, msize, tok_label, tok_lastcnt, tok_serial); DTRACE7(message_send_remote, sender_name, node_name, receiver_name, msize, tok_label, tok_lastcnt, tok_serial); - res = dsig_send(dsdp, ctl, message, 0); - UnUseTmpHeapNoproc(5); + ctx->dss.ctl = ctl; + ctx->dss.msg = message; + ctx->dss.force_busy = 0; + res = erts_dsig_send(&ctx->dsd, &ctx->dss); return res; } int -erts_dsig_send_reg_msg(ErtsDSigData *dsdp, Eterm remote_name, Eterm message) +erts_dsig_send_reg_msg(Eterm remote_name, Eterm message, + ErtsSendContext* ctx) { Eterm ctl; - DeclareTmpHeapNoproc(ctl_heap,6); Eterm token = NIL; - Process *sender = dsdp->proc; + Process *sender = ctx->dsd.proc; int res; #ifdef USE_VM_PROBES Sint tok_label = 0; @@ -898,7 +945,6 @@ erts_dsig_send_reg_msg(ErtsDSigData *dsdp, Eterm remote_name, Eterm message) DTRACE_CHARBUF(receiver_name, 128); #endif - UseTmpHeapNoproc(6); if (SEQ_TRACE_TOKEN(sender) != NIL #ifdef USE_VM_PROBES && SEQ_TRACE_TOKEN(sender) != am_have_dt_utag @@ -912,7 +958,7 @@ erts_dsig_send_reg_msg(ErtsDSigData *dsdp, Eterm remote_name, Eterm message) *node_name = *sender_name = *receiver_name = '\0'; if (DTRACE_ENABLED(message_send) || DTRACE_ENABLED(message_send_remote)) { erts_snprintf(node_name, sizeof(DTRACE_CHARBUF_NAME(node_name)), - "%T", dsdp->dep->sysname); + "%T", ctx->dsd.dep->sysname); erts_snprintf(sender_name, sizeof(DTRACE_CHARBUF_NAME(sender_name)), "%T", sender->common.id); erts_snprintf(receiver_name, sizeof(DTRACE_CHARBUF_NAME(receiver_name)), @@ -927,17 +973,19 @@ erts_dsig_send_reg_msg(ErtsDSigData *dsdp, Eterm remote_name, Eterm message) #endif if (token != NIL) - ctl = TUPLE5(&ctl_heap[0], make_small(DOP_REG_SEND_TT), + ctl = TUPLE5(&ctx->ctl_heap[0], make_small(DOP_REG_SEND_TT), sender->common.id, am_Cookie, remote_name, token); else - ctl = TUPLE4(&ctl_heap[0], make_small(DOP_REG_SEND), + ctl = TUPLE4(&ctx->ctl_heap[0], make_small(DOP_REG_SEND), sender->common.id, am_Cookie, remote_name); DTRACE6(message_send, sender_name, receiver_name, msize, tok_label, tok_lastcnt, tok_serial); DTRACE7(message_send_remote, sender_name, node_name, receiver_name, msize, tok_label, tok_lastcnt, tok_serial); - res = dsig_send(dsdp, ctl, message, 0); - UnUseTmpHeapNoproc(6); + ctx->dss.ctl = ctl; + ctx->dss.msg = message; + ctx->dss.force_busy = 0; + res = erts_dsig_send(&ctx->dsd, &ctx->dss); return res; } @@ -994,7 +1042,7 @@ erts_dsig_send_exit_tt(ErtsDSigData *dsdp, Eterm local, Eterm remote, DTRACE7(process_exit_signal_remote, sender_name, node_name, remote_name, reason_str, tok_label, tok_lastcnt, tok_serial); /* forced, i.e ignore busy */ - res = dsig_send(dsdp, ctl, THE_NON_VALUE, 1); + res = dsig_send_ctl(dsdp, ctl, 1); UnUseTmpHeapNoproc(6); return res; } @@ -1010,7 +1058,7 @@ erts_dsig_send_exit(ErtsDSigData *dsdp, Eterm local, Eterm remote, Eterm reason) ctl = TUPLE4(&ctl_heap[0], make_small(DOP_EXIT), local, remote, reason); /* forced, i.e ignore busy */ - res = dsig_send(dsdp, ctl, THE_NON_VALUE, 1); + res = dsig_send_ctl(dsdp, ctl, 1); UnUseTmpHeapNoproc(5); return res; } @@ -1026,7 +1074,7 @@ erts_dsig_send_exit2(ErtsDSigData *dsdp, Eterm local, Eterm remote, Eterm reason ctl = TUPLE4(&ctl_heap[0], make_small(DOP_EXIT2), local, remote, reason); - res = dsig_send(dsdp, ctl, THE_NON_VALUE, 0); + res = dsig_send_ctl(dsdp, ctl, 0); UnUseTmpHeapNoproc(5); return res; } @@ -1043,7 +1091,7 @@ erts_dsig_send_group_leader(ErtsDSigData *dsdp, Eterm leader, Eterm remote) ctl = TUPLE3(&ctl_heap[0], make_small(DOP_GROUP_LEADER), leader, remote); - res = dsig_send(dsdp, ctl, THE_NON_VALUE, 0); + res = dsig_send_ctl(dsdp, ctl, 0); UnUseTmpHeapNoproc(4); return res; } @@ -1693,194 +1741,235 @@ int erts_net_message(Port *prt, return -1; } -static int -dsig_send(ErtsDSigData *dsdp, Eterm ctl, Eterm msg, int force_busy) +static int dsig_send_ctl(ErtsDSigData* dsdp, Eterm ctl, int force_busy) { + struct erts_dsig_send_context ctx; + int ret; + ctx.ctl = ctl; + ctx.msg = THE_NON_VALUE; + ctx.force_busy = force_busy; + ctx.phase = ERTS_DSIG_SEND_PHASE_INIT; +#ifdef DEBUG + ctx.reds = 1; /* provoke assert below (no reduction count without msg) */ +#endif + ret = erts_dsig_send(dsdp, &ctx); + ASSERT(ret != ERTS_DSIG_SEND_CONTINUE); + return ret; +} + +int +erts_dsig_send(ErtsDSigData *dsdp, struct erts_dsig_send_context* ctx) +{ + int retval; + Sint initial_reds = ctx->reds; Eterm cid; - int suspended = 0; - int resume = 0; - Uint32 pass_through_size; - Uint data_size, dhdr_ext_size; - ErtsAtomCacheMap *acmp; - ErtsDistOutputBuf *obuf; - DistEntry *dep = dsdp->dep; - Uint32 flags = dep->flags; - Process *c_p = dsdp->proc; - if (!c_p || dsdp->no_suspend) - force_busy = 1; + while (1) { + switch (ctx->phase) { + case ERTS_DSIG_SEND_PHASE_INIT: + ctx->flags = dsdp->dep->flags; + ctx->c_p = dsdp->proc; - ERTS_SMP_LC_ASSERT(!c_p - || (ERTS_PROC_LOCK_MAIN - == erts_proc_lc_my_proc_locks(c_p))); + if (!ctx->c_p || dsdp->no_suspend) + ctx->force_busy = 1; - if (!erts_is_alive) - return ERTS_DSIG_SEND_OK; + ERTS_SMP_LC_ASSERT(!ctx->c_p + || (ERTS_PROC_LOCK_MAIN + == erts_proc_lc_my_proc_locks(ctx->c_p))); - if (flags & DFLAG_DIST_HDR_ATOM_CACHE) { - acmp = erts_get_atom_cache_map(c_p); - pass_through_size = 0; - } - else { - acmp = NULL; - pass_through_size = 1; - } + if (!erts_is_alive) + return ERTS_DSIG_SEND_OK; -#ifdef ERTS_DIST_MSG_DBG - erts_fprintf(stderr, ">>%s CTL: %T\n", pass_through_size ? "P" : " ", ctl); - if (is_value(msg)) - erts_fprintf(stderr, " MSG: %T\n", msg); -#endif + if (ctx->flags & DFLAG_DIST_HDR_ATOM_CACHE) { + ctx->acmp = erts_get_atom_cache_map(ctx->c_p); + ctx->pass_through_size = 0; + } + else { + ctx->acmp = NULL; + ctx->pass_through_size = 1; + } - data_size = pass_through_size; - erts_reset_atom_cache_map(acmp); - data_size += erts_encode_dist_ext_size(ctl, flags, acmp); - if (is_value(msg)) - data_size += erts_encode_dist_ext_size(msg, flags, acmp); - erts_finalize_atom_cache_map(acmp, flags); + #ifdef ERTS_DIST_MSG_DBG + erts_fprintf(stderr, ">>%s CTL: %T\n", ctx->pass_through_size ? "P" : " ", ctx->ctl); + if (is_value(msg)) + erts_fprintf(stderr, " MSG: %T\n", msg); + #endif + + ctx->data_size = ctx->pass_through_size; + erts_reset_atom_cache_map(ctx->acmp); + erts_encode_dist_ext_size(ctx->ctl, ctx->flags, ctx->acmp, &ctx->data_size); + + if (is_value(ctx->msg)) { + ctx->u.sc.estack.start = NULL; + ctx->u.sc.flags = ctx->flags; + ctx->u.sc.level = 0; + ctx->phase = ERTS_DSIG_SEND_PHASE_MSG_SIZE; + } else { + ctx->phase = ERTS_DSIG_SEND_PHASE_ALLOC; + } + break; - dhdr_ext_size = erts_encode_ext_dist_header_size(acmp); - data_size += dhdr_ext_size; + case ERTS_DSIG_SEND_PHASE_MSG_SIZE: + if (erts_encode_dist_ext_size_int(ctx->msg, ctx, &ctx->data_size)) { + retval = ERTS_DSIG_SEND_CONTINUE; + goto done; + } - obuf = alloc_dist_obuf(data_size); - obuf->ext_endp = &obuf->data[0] + pass_through_size + dhdr_ext_size; + ctx->phase = ERTS_DSIG_SEND_PHASE_ALLOC; + case ERTS_DSIG_SEND_PHASE_ALLOC: + erts_finalize_atom_cache_map(ctx->acmp, ctx->flags); + + ctx->dhdr_ext_size = erts_encode_ext_dist_header_size(ctx->acmp); + ctx->data_size += ctx->dhdr_ext_size; + + ctx->obuf = alloc_dist_obuf(ctx->data_size); + ctx->obuf->ext_endp = &ctx->obuf->data[0] + ctx->pass_through_size + ctx->dhdr_ext_size; + + /* Encode internal version of dist header */ + ctx->obuf->extp = erts_encode_ext_dist_header_setup(ctx->obuf->ext_endp, ctx->acmp); + /* Encode control message */ + erts_encode_dist_ext(ctx->ctl, &ctx->obuf->ext_endp, ctx->flags, ctx->acmp, NULL, NULL); + if (is_value(ctx->msg)) { + ctx->u.ec.flags = ctx->flags; + ctx->u.ec.level = 0; + ctx->u.ec.wstack.wstart = NULL; + ctx->phase = ERTS_DSIG_SEND_PHASE_MSG_ENCODE; + } else { + ctx->phase = ERTS_DSIG_SEND_PHASE_FIN; + } + break; - /* Encode internal version of dist header */ - obuf->extp = erts_encode_ext_dist_header_setup(obuf->ext_endp, acmp); - /* Encode control message */ - erts_encode_dist_ext(ctl, &obuf->ext_endp, flags, acmp); - if (is_value(msg)) { - /* Encode message */ - erts_encode_dist_ext(msg, &obuf->ext_endp, flags, acmp); - } + case ERTS_DSIG_SEND_PHASE_MSG_ENCODE: + if (erts_encode_dist_ext(ctx->msg, &ctx->obuf->ext_endp, ctx->flags, ctx->acmp, &ctx->u.ec, &ctx->reds)) { + retval = ERTS_DSIG_SEND_CONTINUE; + goto done; + } - ASSERT(obuf->extp < obuf->ext_endp); - ASSERT(&obuf->data[0] <= obuf->extp - pass_through_size); - ASSERT(obuf->ext_endp <= &obuf->data[0] + data_size); + ctx->phase = ERTS_DSIG_SEND_PHASE_FIN; + case ERTS_DSIG_SEND_PHASE_FIN: { + DistEntry *dep = dsdp->dep; + int suspended = 0; + int resume = 0; - data_size = obuf->ext_endp - obuf->extp; + ASSERT(ctx->obuf->extp < ctx->obuf->ext_endp); + ASSERT(&ctx->obuf->data[0] <= ctx->obuf->extp - ctx->pass_through_size); + ASSERT(ctx->obuf->ext_endp <= &ctx->obuf->data[0] + ctx->data_size); - /* - * Signal encoded; now verify that the connection still exists, - * and if so enqueue the signal and schedule it for send. - */ - obuf->next = NULL; - erts_smp_de_rlock(dep); - cid = dep->cid; - if (cid != dsdp->cid - || dep->connection_id != dsdp->connection_id - || dep->status & ERTS_DE_SFLG_EXITING) { - /* Not the same connection as when we started; drop message... */ - erts_smp_de_runlock(dep); - free_dist_obuf(obuf); - } - else { - ErtsProcList *plp = NULL; - erts_smp_mtx_lock(&dep->qlock); - dep->qsize += size_obuf(obuf); - if (dep->qsize >= erts_dist_buf_busy_limit) - dep->qflgs |= ERTS_DE_QFLG_BUSY; - if (!force_busy && (dep->qflgs & ERTS_DE_QFLG_BUSY)) { - erts_smp_mtx_unlock(&dep->qlock); + ctx->data_size = ctx->obuf->ext_endp - ctx->obuf->extp; - plp = erts_proclist_create(c_p); - erts_suspend(c_p, ERTS_PROC_LOCK_MAIN, NULL); - suspended = 1; - erts_smp_mtx_lock(&dep->qlock); - } + /* + * Signal encoded; now verify that the connection still exists, + * and if so enqueue the signal and schedule it for send. + */ + ctx->obuf->next = NULL; + erts_smp_de_rlock(dep); + cid = dep->cid; + if (cid != dsdp->cid + || dep->connection_id != dsdp->connection_id + || dep->status & ERTS_DE_SFLG_EXITING) { + /* Not the same connection as when we started; drop message... */ + erts_smp_de_runlock(dep); + free_dist_obuf(ctx->obuf); + } + else { + ErtsProcList *plp = NULL; + erts_smp_mtx_lock(&dep->qlock); + dep->qsize += size_obuf(ctx->obuf); + if (dep->qsize >= erts_dist_buf_busy_limit) + dep->qflgs |= ERTS_DE_QFLG_BUSY; + if (!ctx->force_busy && (dep->qflgs & ERTS_DE_QFLG_BUSY)) { + erts_smp_mtx_unlock(&dep->qlock); + + plp = erts_proclist_create(ctx->c_p); + erts_suspend(ctx->c_p, ERTS_PROC_LOCK_MAIN, NULL); + suspended = 1; + erts_smp_mtx_lock(&dep->qlock); + } - /* Enqueue obuf on dist entry */ - if (dep->out_queue.last) - dep->out_queue.last->next = obuf; - else - dep->out_queue.first = obuf; - dep->out_queue.last = obuf; + /* Enqueue obuf on dist entry */ + if (dep->out_queue.last) + dep->out_queue.last->next = ctx->obuf; + else + dep->out_queue.first = ctx->obuf; + dep->out_queue.last = ctx->obuf; + + if (!ctx->force_busy) { + if (!(dep->qflgs & ERTS_DE_QFLG_BUSY)) { + if (suspended) + resume = 1; /* was busy when we started, but isn't now */ + #ifdef USE_VM_PROBES + if (resume && DTRACE_ENABLED(dist_port_not_busy)) { + DTRACE_CHARBUF(port_str, 64); + DTRACE_CHARBUF(remote_str, 64); + + erts_snprintf(port_str, sizeof(DTRACE_CHARBUF_NAME(port_str)), + "%T", cid); + erts_snprintf(remote_str, sizeof(DTRACE_CHARBUF_NAME(remote_str)), + "%T", dep->sysname); + DTRACE3(dist_port_not_busy, erts_this_node_sysname, + port_str, remote_str); + } + #endif + } + else { + /* Enqueue suspended process on dist entry */ + ASSERT(plp); + erts_proclist_store_last(&dep->suspended, plp); + } + } - if (!force_busy) { - if (!(dep->qflgs & ERTS_DE_QFLG_BUSY)) { - if (suspended) - resume = 1; /* was busy when we started, but isn't now */ -#ifdef USE_VM_PROBES - if (resume && DTRACE_ENABLED(dist_port_not_busy)) { - DTRACE_CHARBUF(port_str, 64); - DTRACE_CHARBUF(remote_str, 64); - - erts_snprintf(port_str, sizeof(DTRACE_CHARBUF_NAME(port_str)), - "%T", cid); - erts_snprintf(remote_str, sizeof(DTRACE_CHARBUF_NAME(remote_str)), - "%T", dep->sysname); - DTRACE3(dist_port_not_busy, erts_this_node_sysname, - port_str, remote_str); - } -#endif + erts_smp_mtx_unlock(&dep->qlock); + erts_schedule_dist_command(NULL, dep); + erts_smp_de_runlock(dep); + + if (resume) { + erts_resume(ctx->c_p, ERTS_PROC_LOCK_MAIN); + erts_proclist_destroy(plp); + /* + * Note that the calling process still have to yield as if it + * suspended. If not, the calling process could later be + * erroneously scheduled when it shouldn't be. + */ + } } - else { - /* Enqueue suspended process on dist entry */ - ASSERT(plp); - erts_proclist_store_last(&dep->suspended, plp); + ctx->obuf = NULL; + + if (suspended) { + #ifdef USE_VM_PROBES + if (!resume && DTRACE_ENABLED(dist_port_busy)) { + DTRACE_CHARBUF(port_str, 64); + DTRACE_CHARBUF(remote_str, 64); + DTRACE_CHARBUF(pid_str, 16); + + erts_snprintf(port_str, sizeof(DTRACE_CHARBUF_NAME(port_str)), "%T", cid); + erts_snprintf(remote_str, sizeof(DTRACE_CHARBUF_NAME(remote_str)), + "%T", dep->sysname); + erts_snprintf(pid_str, sizeof(DTRACE_CHARBUF_NAME(pid_str)), + "%T", ctx->c_p->common.id); + DTRACE4(dist_port_busy, erts_this_node_sysname, + port_str, remote_str, pid_str); + } + #endif + if (!resume && erts_system_monitor_flags.busy_dist_port) + monitor_generic(ctx->c_p, am_busy_dist_port, cid); + retval = ERTS_DSIG_SEND_YIELD; + } else { + retval = ERTS_DSIG_SEND_OK; } + goto done; } - - erts_smp_mtx_unlock(&dep->qlock); - erts_schedule_dist_command(NULL, dep); - erts_smp_de_runlock(dep); - - if (resume) { - erts_resume(c_p, ERTS_PROC_LOCK_MAIN); - erts_proclist_destroy(plp); - /* - * Note that the calling process still have to yield as if it - * suspended. If not, the calling process could later be - * erroneously scheduled when it shouldn't be. - */ + default: + erl_exit(ERTS_ABORT_EXIT, "dsig_send invalid phase (%d)\n", (int)ctx->phase); } } - if (c_p) { - int reds; - /* - * Bump reductions on calling process. - * - * This is the reduction cost: Always a base cost of 8 reductions - * plus 16 reductions per kilobyte generated external data. - */ - - data_size >>= (10-4); -#if defined(ARCH_64) && !HALFWORD_HEAP - data_size &= 0x003fffffffffffff; -#elif defined(ARCH_32) || HALFWORD_HEAP - data_size &= 0x003fffff; -#else -# error "Ohh come on ... !?!" -#endif - reds = 8 + ((int) data_size > 1000000 ? 1000000 : (int) data_size); - BUMP_REDS(c_p, reds); - } - - if (suspended) { -#ifdef USE_VM_PROBES - if (!resume && DTRACE_ENABLED(dist_port_busy)) { - DTRACE_CHARBUF(port_str, 64); - DTRACE_CHARBUF(remote_str, 64); - DTRACE_CHARBUF(pid_str, 16); - - erts_snprintf(port_str, sizeof(DTRACE_CHARBUF_NAME(port_str)), "%T", cid); - erts_snprintf(remote_str, sizeof(DTRACE_CHARBUF_NAME(remote_str)), - "%T", dep->sysname); - erts_snprintf(pid_str, sizeof(DTRACE_CHARBUF_NAME(pid_str)), - "%T", c_p->common.id); - DTRACE4(dist_port_busy, erts_this_node_sysname, - port_str, remote_str, pid_str); - } -#endif - if (!resume && erts_system_monitor_flags.busy_dist_port) - monitor_generic(c_p, am_busy_dist_port, cid); - return ERTS_DSIG_SEND_YIELD; +done: + if (ctx->msg && ctx->c_p) { + BUMP_REDS(ctx->c_p, (initial_reds - ctx->reds) / TERM_TO_BINARY_LOOP_FACTOR); } - return ERTS_DSIG_SEND_OK; + return retval; } - static Uint dist_port_command(Port *prt, ErtsDistOutputBuf *obuf) { diff --git a/erts/emulator/beam/dist.h b/erts/emulator/beam/dist.h index f32b999198..2a2ba0c83f 100644 --- a/erts/emulator/beam/dist.h +++ b/erts/emulator/beam/dist.h @@ -22,6 +22,7 @@ #include "erl_process.h" #include "erl_node_tables.h" +#include "zlib.h" #define DFLAG_PUBLISHED 0x01 #define DFLAG_ATOM_CACHE 0x02 @@ -264,17 +265,105 @@ erts_destroy_dist_link(ErtsDistLinkData *dldp) #endif + + +/* Define for testing */ +/* #define EXTREME_TTB_TRAPPING 1 */ + +#ifndef EXTREME_TTB_TRAPPING +#define TERM_TO_BINARY_LOOP_FACTOR 32 +#else +#define TERM_TO_BINARY_LOOP_FACTOR 1 +#endif + +typedef enum { TTBSize, TTBEncode, TTBCompress } TTBState; +typedef struct TTBSizeContext_ { + Uint flags; + int level; + Uint result; + Eterm obj; + ErtsEStack estack; +} TTBSizeContext; + +typedef struct TTBEncodeContext_ { + Uint flags; + int level; + byte* ep; + Eterm obj; + ErtsWStack wstack; + Binary *result_bin; +} TTBEncodeContext; + +typedef struct { + Uint real_size; + Uint dest_len; + byte *dbytes; + Binary *result_bin; + Binary *destination_bin; + z_stream stream; +} TTBCompressContext; + +typedef struct { + int alive; + TTBState state; + union { + TTBSizeContext sc; + TTBEncodeContext ec; + TTBCompressContext cc; + } s; +} TTBContext; + +enum erts_dsig_send_phase { + ERTS_DSIG_SEND_PHASE_INIT, + ERTS_DSIG_SEND_PHASE_MSG_SIZE, + ERTS_DSIG_SEND_PHASE_ALLOC, + ERTS_DSIG_SEND_PHASE_MSG_ENCODE, + ERTS_DSIG_SEND_PHASE_FIN +}; + +struct erts_dsig_send_context { + enum erts_dsig_send_phase phase; + Sint reds; + + Eterm ctl; + Eterm msg; + int force_busy; + Uint32 pass_through_size; + Uint data_size, dhdr_ext_size; + ErtsAtomCacheMap *acmp; + ErtsDistOutputBuf *obuf; + Uint32 flags; + Process *c_p; + union { + TTBSizeContext sc; + TTBEncodeContext ec; + }u; +}; + +typedef struct { + int suspend; + + Eterm ctl_heap[6]; + ErtsDSigData dsd; + DistEntry* dep_to_deref; + struct erts_dsig_send_context dss; + + Eterm return_term; +}ErtsSendContext; + + /* * erts_dsig_send_* return values. */ #define ERTS_DSIG_SEND_OK 0 #define ERTS_DSIG_SEND_YIELD 1 +#define ERTS_DSIG_SEND_CONTINUE 2 extern int erts_dsig_send_link(ErtsDSigData *, Eterm, Eterm); -extern int erts_dsig_send_msg(ErtsDSigData *, Eterm, Eterm); +extern int erts_dsig_send_msg(Eterm, Eterm, ErtsSendContext*); extern int erts_dsig_send_exit_tt(ErtsDSigData *, Eterm, Eterm, Eterm, Eterm); extern int erts_dsig_send_unlink(ErtsDSigData *, Eterm, Eterm); -extern int erts_dsig_send_reg_msg(ErtsDSigData *, Eterm, Eterm); +extern int erts_dsig_send_reg_msg(Eterm, Eterm, ErtsSendContext*); extern int erts_dsig_send_group_leader(ErtsDSigData *, Eterm, Eterm); extern int erts_dsig_send_exit(ErtsDSigData *, Eterm, Eterm, Eterm); extern int erts_dsig_send_exit2(ErtsDSigData *, Eterm, Eterm, Eterm); @@ -282,6 +371,10 @@ extern int erts_dsig_send_demonitor(ErtsDSigData *, Eterm, Eterm, Eterm, int); extern int erts_dsig_send_monitor(ErtsDSigData *, Eterm, Eterm, Eterm); extern int erts_dsig_send_m_exit(ErtsDSigData *, Eterm, Eterm, Eterm, Eterm); +extern int erts_dsig_send(ErtsDSigData *dsdp, struct erts_dsig_send_context* ctx); +extern void erts_dsend_context_dtor(Binary*); +extern Eterm erts_dsend_export_trap_context(Process* p, ErtsSendContext* ctx); + extern int erts_dist_command(Port *prt, int reds); extern void erts_dist_port_not_busy(Port *prt); extern void erts_kill_dist_connection(DistEntry *dep, Uint32); diff --git a/erts/emulator/beam/erl_alloc.c b/erts/emulator/beam/erl_alloc.c index 90cd227fae..f2bceff4eb 100644 --- a/erts/emulator/beam/erl_alloc.c +++ b/erts/emulator/beam/erl_alloc.c @@ -3939,7 +3939,7 @@ static Uint install_debug_functions(void) { int i; - ASSERT(sizeof(erts_allctrs) == sizeof(real_allctrs)); + ERTS_CT_ASSERT(sizeof(erts_allctrs) == sizeof(real_allctrs)); sys_memcpy((void *)real_allctrs,(void *)erts_allctrs,sizeof(erts_allctrs)); diff --git a/erts/emulator/beam/erl_alloc.types b/erts/emulator/beam/erl_alloc.types index 43279715b8..e2f8da38b9 100644 --- a/erts/emulator/beam/erl_alloc.types +++ b/erts/emulator/beam/erl_alloc.types @@ -418,6 +418,8 @@ type CS_PROG_PATH LONG_LIVED SYSTEM cs_prog_path type ENVIRONMENT TEMPORARY SYSTEM environment type PUTENV_STR SYSTEM SYSTEM putenv_string type PRT_REP_EXIT STANDARD SYSTEM port_report_exit +type SYS_BLOCKING STANDARD SYSTEM sys_blocking +type SYS_WRITE_BUF TEMPORARY SYSTEM sys_write_buf +endif diff --git a/erts/emulator/beam/erl_alloc_util.c b/erts/emulator/beam/erl_alloc_util.c index e3172dc4fb..2f277690e4 100644 --- a/erts/emulator/beam/erl_alloc_util.c +++ b/erts/emulator/beam/erl_alloc_util.c @@ -1442,7 +1442,7 @@ get_pref_allctr(void *extra) pref_ix = ERTS_ALC_GET_THR_IX(); - ASSERT(sizeof(UWord) == sizeof(Allctr_t *)); + ERTS_CT_ASSERT(sizeof(UWord) == sizeof(Allctr_t *)); ASSERT(0 <= pref_ix && pref_ix < tspec->size); return tspec->allctr[pref_ix]; @@ -1861,7 +1861,7 @@ handle_delayed_dealloc(Allctr_t *allctr, * if this carrier is pulled from dc_list by cpool_fetch() */ ERTS_ALC_CPOOL_ASSERT(FBLK_TO_MBC(blk) != crr); - ERTS_ALC_CPOOL_ASSERT(sizeof(ErtsAllctrDDBlock_t) == sizeof(void*)); + ERTS_CT_ASSERT(sizeof(ErtsAllctrDDBlock_t) == sizeof(void*)); #ifdef MBC_ABLK_OFFSET_BITS blk->u.carrier = crr; #else @@ -5942,7 +5942,7 @@ erts_alcu_init(AlcUInit_t *init) erts_atomic_init_nob(&sentinel->prev, (erts_aint_t) sentinel); } #endif - ASSERT(SBC_BLK_SZ_MASK == MBC_FBLK_SZ_MASK); /* see BLK_SZ */ + ERTS_CT_ASSERT(SBC_BLK_SZ_MASK == MBC_FBLK_SZ_MASK); /* see BLK_SZ */ #if HAVE_ERTS_MSEG ASSERT(erts_mseg_unit_size() == ERTS_SACRR_UNIT_SZ); max_mseg_carriers = init->mmc; diff --git a/erts/emulator/beam/erl_async.c b/erts/emulator/beam/erl_async.c index decae6b2ca..bc06d41720 100644 --- a/erts/emulator/beam/erl_async.c +++ b/erts/emulator/beam/erl_async.c @@ -176,7 +176,7 @@ erts_init_async(void) ErtsThrQInit_t qinit = ERTS_THR_Q_INIT_DEFAULT; #endif erts_thr_opts_t thr_opts = ERTS_THR_OPTS_DEFAULT_INITER; - char *ptr; + char *ptr, thr_name[16]; size_t tot_size = 0; int i; @@ -227,23 +227,16 @@ erts_init_async(void) thr_opts.suggested_stack_size = erts_async_thread_suggested_stack_size; -#ifdef ETHR_HAVE_THREAD_NAMES - thr_opts.name = malloc(sizeof(char)*(strlen("async_XXXX")+1)); -#endif + thr_opts.name = thr_name; for (i = 0; i < erts_async_max_threads; i++) { ErtsAsyncQ *aq = async_q(i); -#ifdef ETHR_HAVE_THREAD_NAMES - sprintf(thr_opts.name, "async_%d", i+1); -#endif + erts_snprintf(thr_opts.name, 16, "async_%d", i+1); erts_thr_create(&aq->thr_id, async_main, (void*) aq, &thr_opts); } -#ifdef ETHR_HAVE_THREAD_NAMES - free(thr_opts.name); -#endif /* Wait for async threads to initialize... */ erts_mtx_lock(&async->init.data.mtx); diff --git a/erts/emulator/beam/erl_bif_binary.c b/erts/emulator/beam/erl_bif_binary.c index 68004a7725..934904d58e 100644 --- a/erts/emulator/beam/erl_bif_binary.c +++ b/erts/emulator/beam/erl_bif_binary.c @@ -32,7 +32,9 @@ #include "global.h" #include "erl_process.h" #include "error.h" +#define ERL_WANT_HIPE_BIF_WRAPPER__ #include "bif.h" +#undef ERL_WANT_HIPE_BIF_WRAPPER__ #include "big.h" #include "erl_binary.h" #include "erl_bits.h" @@ -2425,8 +2427,6 @@ static BIF_RETTYPE do_binary_copy(Process *p, Eterm bin, Eterm en) } cbs->result = erts_bin_nrml_alloc(target_size); /* Always offheap if trapping */ - cbs->result->flags = 0; - cbs->result->orig_size = target_size; erts_refc_init(&(cbs->result->refc), 1); t = (byte *) cbs->result->orig_bytes; /* No offset or anything */ pos = 0; diff --git a/erts/emulator/beam/erl_bif_guard.c b/erts/emulator/beam/erl_bif_guard.c index bbd8aa31d9..e7d84ebda1 100644 --- a/erts/emulator/beam/erl_bif_guard.c +++ b/erts/emulator/beam/erl_bif_guard.c @@ -459,23 +459,25 @@ Eterm erts_gc_byte_size_1(Process* p, Eterm* reg, Uint live) Eterm erts_gc_map_size_1(Process* p, Eterm* reg, Uint live) { Eterm arg = reg[live]; - if (is_map(arg)) { - map_t *mp = (map_t*)map_val(arg); - Uint size = map_get_size(mp); - if (IS_USMALL(0, size)) { - return make_small(size); - } else { - Eterm* hp; - if (ERTS_NEED_GC(p, BIG_UINT_HEAP_SIZE)) { - erts_garbage_collect(p, BIG_UINT_HEAP_SIZE, reg, live); - } - hp = p->htop; - p->htop += BIG_UINT_HEAP_SIZE; - return uint_to_big(size, hp); - } + Eterm* hp; + Uint size; + if (is_flatmap(arg)) { + flatmap_t *mp = (flatmap_t*)flatmap_val(arg); + size = flatmap_get_size(mp); + } else if (is_hashmap(arg)) { + size = hashmap_size(arg); } else { BIF_ERROR(p, BADARG); } + if (IS_USMALL(0, size)) { + return make_small(size); + } + if (ERTS_NEED_GC(p, BIG_UINT_HEAP_SIZE)) { + erts_garbage_collect(p, BIG_UINT_HEAP_SIZE, reg, live); + } + hp = p->htop; + p->htop += BIG_UINT_HEAP_SIZE; + return uint_to_big(size, hp); } Eterm erts_gc_abs_1(Process* p, Eterm* reg, Uint live) diff --git a/erts/emulator/beam/erl_bif_info.c b/erts/emulator/beam/erl_bif_info.c index 70062b2305..2eeebab9a3 100644 --- a/erts/emulator/beam/erl_bif_info.c +++ b/erts/emulator/beam/erl_bif_info.c @@ -117,6 +117,9 @@ static char erts_system_version[] = ("Erlang/OTP " ERLANG_OTP_RELEASE #ifdef ERTS_ENABLE_LOCK_COUNT " [lock-counting]" #endif +#ifdef ERTS_OPCODE_COUNTER_SUPPORT + " [instruction-counting]" +#endif #ifdef PURIFY " [purify-compiled]" #endif @@ -2343,7 +2346,7 @@ BIF_RETTYPE system_info_1(BIF_ALIST_1) for (i = num_instructions-1; i >= 0; i--) { res = erts_bld_cons(hpp, hszp, erts_bld_tuple(hpp, hszp, 2, - erts_atom_put(opc[i].name, + erts_atom_put((byte *)opc[i].name, strlen(opc[i].name), ERTS_ATOM_ENC_LATIN1, 1), @@ -3661,6 +3664,26 @@ BIF_RETTYPE erts_debug_get_internal_state_1(BIF_ALIST_1) BIF_RET(erts_debug_reader_groups_map(BIF_P, (int) groups)); } + else if (ERTS_IS_ATOM_STR("internal_hash", tp[1])) { + Uint hash = (Uint) make_internal_hash(tp[2]); + Uint hsz = 0; + Eterm* hp; + erts_bld_uint(NULL, &hsz, hash); + hp = HAlloc(BIF_P,hsz); + return erts_bld_uint(&hp, NULL, hash); + } + else if (ERTS_IS_ATOM_STR("atom", tp[1])) { + Uint ix; + if (!term_to_Uint(tp[2], &ix)) + BIF_ERROR(BIF_P, BADARG); + while (ix >= atom_table_size()) { + char tmp[20]; + erts_snprintf(tmp, sizeof(tmp), "am%x", atom_table_size()); + erts_atom_put((byte *) tmp, strlen(tmp), ERTS_ATOM_ENC_LATIN1, 1); + } + return make_atom(ix); + } + break; } case 3: { @@ -4014,6 +4037,13 @@ BIF_RETTYPE erts_debug_set_internal_state_2(BIF_ALIST_2) } } } + else if (ERTS_IS_ATOM_STR("broken_halt", BIF_ARG_1)) { + /* Ugly ugly code used by bif_SUITE:erlang_halt/1 */ +#if defined(ERTS_HAVE_TRY_CATCH) + erts_get_scheduler_data()->run_queue = NULL; +#endif + erl_exit(ERTS_DUMP_EXIT, "%T", BIF_ARG_2); + } else if (ERTS_IS_ATOM_STR("unique_monotonic_integer_state", BIF_ARG_1)) { int res = erts_debug_set_unique_monotonic_integer_state(BIF_ARG_2); BIF_RET(res ? am_true : am_false); diff --git a/erts/emulator/beam/erl_bif_trace.c b/erts/emulator/beam/erl_bif_trace.c index 08796df912..ac57205c47 100644 --- a/erts/emulator/beam/erl_bif_trace.c +++ b/erts/emulator/beam/erl_bif_trace.c @@ -652,7 +652,7 @@ Eterm trace_3(BIF_ALIST_3) if (pid_spec == am_all) { if (on) { if (!erts_cpu_timestamp) { -#ifdef HAVE_CLOCK_GETTIME +#ifdef HAVE_CLOCK_GETTIME_CPU_TIME /* Perhaps clock_gettime was found during config on a different machine than this. We check @@ -679,7 +679,7 @@ Eterm trace_3(BIF_ALIST_3) if (erts_start_now_cpu() < 0) { goto error; } -#endif /* HAVE_CLOCK_GETTIME */ +#endif /* HAVE_CLOCK_GETTIME_CPU_TIME */ erts_cpu_timestamp = !0; } } diff --git a/erts/emulator/beam/erl_binary.h b/erts/emulator/beam/erl_binary.h index 06dfeb1260..8d264d166e 100644 --- a/erts/emulator/beam/erl_binary.h +++ b/erts/emulator/beam/erl_binary.h @@ -231,41 +231,58 @@ erts_free_aligned_binary_bytes(byte* buf) # define CHICKEN_PAD (sizeof(void*) - 1) #endif +/* Caller must initialize 'refc' +*/ ERTS_GLB_INLINE Binary * erts_bin_drv_alloc_fnf(Uint size) { Uint bsize = ERTS_SIZEOF_Binary(size) + CHICKEN_PAD; - void *res; + Binary *res; + if (bsize < size) /* overflow */ return NULL; res = erts_alloc_fnf(ERTS_ALC_T_DRV_BINARY, bsize); ERTS_CHK_BIN_ALIGNMENT(res); - return (Binary *) res; + if (res) { + res->orig_size = size; + res->flags = BIN_FLAG_DRV; + } + return res; } +/* Caller must initialize 'refc' +*/ ERTS_GLB_INLINE Binary * erts_bin_drv_alloc(Uint size) { Uint bsize = ERTS_SIZEOF_Binary(size) + CHICKEN_PAD; - void *res; + Binary *res; + if (bsize < size) /* overflow */ erts_alloc_enomem(ERTS_ALC_T_DRV_BINARY, size); res = erts_alloc(ERTS_ALC_T_DRV_BINARY, bsize); ERTS_CHK_BIN_ALIGNMENT(res); - return (Binary *) res; + res->orig_size = size; + res->flags = BIN_FLAG_DRV; + return res; } +/* Caller must initialize 'refc' +*/ ERTS_GLB_INLINE Binary * erts_bin_nrml_alloc(Uint size) { Uint bsize = ERTS_SIZEOF_Binary(size) + CHICKEN_PAD; - void *res; + Binary *res; + if (bsize < size) /* overflow */ erts_alloc_enomem(ERTS_ALC_T_BINARY, size); res = erts_alloc(ERTS_ALC_T_BINARY, bsize); ERTS_CHK_BIN_ALIGNMENT(res); - return (Binary *) res; + res->orig_size = size; + res->flags = 0; + return res; } ERTS_GLB_INLINE Binary * @@ -280,6 +297,8 @@ erts_bin_realloc_fnf(Binary *bp, Uint size) return NULL; nbp = erts_realloc_fnf(type, (void *) bp, bsize); ERTS_CHK_BIN_ALIGNMENT(nbp); + if (nbp) + nbp->orig_size = size; return nbp; } @@ -297,6 +316,7 @@ erts_bin_realloc(Binary *bp, Uint size) if (!nbp) erts_realloc_enomem(type, bp, bsize); ERTS_CHK_BIN_ALIGNMENT(nbp); + nbp->orig_size = size; return nbp; } @@ -329,4 +349,4 @@ erts_create_magic_binary(Uint size, void (*destructor)(Binary *)) #endif /* #if ERTS_GLB_INLINE_INCL_FUNC_DEF */ -#endif +#endif /* !__ERL_BINARY_H */ diff --git a/erts/emulator/beam/erl_bits.c b/erts/emulator/beam/erl_bits.c index 73765772c8..5cc0a23dc9 100644 --- a/erts/emulator/beam/erl_bits.c +++ b/erts/emulator/beam/erl_bits.c @@ -403,7 +403,10 @@ erts_bs_get_integer_2(Process *p, Uint num_bits, unsigned flags, ErlBinMatchBuff words_needed = 1+WSIZE(bytes); hp = HeapOnlyAlloc(p, words_needed); res = bytes_to_big(LSB, bytes, sgn, hp); - if (is_small(res)) { + if (is_nil(res)) { + p->htop = hp; + res = THE_NON_VALUE; + } else if (is_small(res)) { p->htop = hp; } else if ((actual = bignum_header_arity(*hp)+1) < words_needed) { p->htop = hp + actual; @@ -1299,7 +1302,6 @@ erts_bs_append(Process* c_p, Eterm* reg, Uint live, Eterm build_size_term, if (binp->orig_size < pb->size) { Uint new_size = 2*pb->size; binp = erts_bin_realloc(binp, new_size); - binp->orig_size = new_size; pb->val = binp; pb->bytes = (byte *) binp->orig_bytes; } @@ -1371,8 +1373,6 @@ erts_bs_append(Process* c_p, Eterm* reg, Uint live, Eterm build_size_term, * Allocate the binary data struct itself. */ bptr = erts_bin_nrml_alloc(bin_size); - bptr->flags = 0; - bptr->orig_size = bin_size; erts_refc_init(&bptr->refc, 1); erts_current_bin = (byte *) bptr->orig_bytes; @@ -1475,7 +1475,6 @@ erts_bs_private_append(Process* p, Eterm bin, Eterm build_size_term, Uint unit) * is safe to reallocate it. */ binp = erts_bin_realloc(binp, new_size); - binp->orig_size = new_size; pb->val = binp; pb->bytes = (byte *) binp->orig_bytes; } else { @@ -1488,8 +1487,6 @@ erts_bs_private_append(Process* p, Eterm bin, Eterm build_size_term, Uint unit) * binary and copy the contents of the old binary into it. */ Binary* bptr = erts_bin_nrml_alloc(new_size); - bptr->flags = 0; - bptr->orig_size = new_size; erts_refc_init(&bptr->refc, 1); sys_memcpy(bptr->orig_bytes, binp->orig_bytes, binp->orig_size); pb->flags |= PB_IS_WRITABLE | PB_ACTIVE_WRITER; @@ -1537,8 +1534,6 @@ erts_bs_init_writable(Process* p, Eterm sz) * Allocate the binary data struct itself. */ bptr = erts_bin_nrml_alloc(bin_size); - bptr->flags = 0; - bptr->orig_size = bin_size; erts_refc_init(&bptr->refc, 1); /* @@ -1585,9 +1580,7 @@ erts_emasculate_writable_binary(ProcBin* pb) /* Our allocators are 8 byte aligned, i.e., shrinking with less than 8 bytes will have no real effect */ if (unused >= 8) { - Uint new_size = pb->size; binp = erts_bin_realloc(binp, pb->size); - binp->orig_size = new_size; pb->val = binp; pb->bytes = (byte *) binp->orig_bytes; } diff --git a/erts/emulator/beam/erl_db.c b/erts/emulator/beam/erl_db.c index 8f246ffa07..fff892ae54 100644 --- a/erts/emulator/beam/erl_db.c +++ b/erts/emulator/beam/erl_db.c @@ -753,6 +753,31 @@ BIF_RETTYPE ets_prev_2(BIF_ALIST_2) BIF_RET(ret); } +/* +** take(Tab, Key) +*/ +BIF_RETTYPE ets_take_2(BIF_ALIST_2) +{ + DbTable* tb; +#ifdef DEBUG + int cret; +#endif + Eterm ret; + CHECK_TABLES(); + + tb = db_get_table(BIF_P, BIF_ARG_1, DB_WRITE, LCK_WRITE_REC); + if (!tb) { + BIF_ERROR(BIF_P, BADARG); + } +#ifdef DEBUG + cret = +#endif + tb->common.meth->db_take(BIF_P, tb, BIF_ARG_2, &ret); + ASSERT(cret == DB_ERROR_NONE); + db_unlock(tb, LCK_WRITE_REC); + BIF_RET(ret); +} + /* ** update_element(Tab, Key, {Pos, Value}) ** update_element(Tab, Key, [{Pos, Value}]) @@ -780,7 +805,7 @@ BIF_RETTYPE ets_update_element_3(BIF_ALIST_3) list = BIF_ARG_3; } - if (!tb->common.meth->db_lookup_dbterm(tb, BIF_ARG_2, &handle)) { + if (!tb->common.meth->db_lookup_dbterm(BIF_P, tb, BIF_ARG_2, THE_NON_VALUE, &handle)) { cret = DB_ERROR_BADKEY; goto bail_out; } @@ -819,7 +844,7 @@ BIF_RETTYPE ets_update_element_3(BIF_ALIST_3) } finalize: - tb->common.meth->db_finalize_dbterm(&handle); + tb->common.meth->db_finalize_dbterm(cret, &handle); bail_out: UnUseTmpHeap(2,BIF_P); @@ -838,14 +863,8 @@ bail_out: } } -/* -** update_counter(Tab, Key, Incr) -** update_counter(Tab, Key, {Upop}) -** update_counter(Tab, Key, [{Upop}]) -** Upop = {Pos,Incr} | {Pos,Incr,Threshold,WarpTo} -** Returns new value(s) (integer or [integer]) -*/ -BIF_RETTYPE ets_update_counter_3(BIF_ALIST_3) +static BIF_RETTYPE +do_update_counter(Process *p, Eterm arg1, Eterm arg2, Eterm arg3, Eterm arg4) { DbTable* tb; int cret = DB_ERROR_BADITEM; @@ -855,7 +874,7 @@ BIF_RETTYPE ets_update_counter_3(BIF_ALIST_3) Eterm* ret_list_currp = NULL; Eterm* ret_list_prevp = NULL; Eterm iter; - DeclareTmpHeap(cell,5,BIF_P); + DeclareTmpHeap(cell, 5, p); Eterm *tuple = cell+2; DbUpdateHandle handle; Uint halloc_size = 0; /* overestimated heap usage */ @@ -863,28 +882,29 @@ BIF_RETTYPE ets_update_counter_3(BIF_ALIST_3) Eterm* hstart; Eterm* hend; - if ((tb = db_get_table(BIF_P, BIF_ARG_1, DB_WRITE, LCK_WRITE_REC)) == NULL) { - BIF_ERROR(BIF_P, BADARG); + if ((tb = db_get_table(p, arg1, DB_WRITE, LCK_WRITE_REC)) == NULL) { + BIF_ERROR(p, BADARG); } - UseTmpHeap(5,BIF_P); + UseTmpHeap(5, p); if (!(tb->common.status & (DB_SET | DB_ORDERED_SET))) { goto bail_out; } - if (is_integer(BIF_ARG_3)) { /* Incr */ - upop_list = CONS(cell, TUPLE2(tuple, make_small(tb->common.keypos+1), - BIF_ARG_3), NIL); + if (is_integer(arg3)) { /* Incr */ + upop_list = CONS(cell, + TUPLE2(tuple, make_small(tb->common.keypos+1), arg3), + NIL); } - else if (is_tuple(BIF_ARG_3)) { /* {Upop} */ - upop_list = CONS(cell, BIF_ARG_3, NIL); + else if (is_tuple(arg3)) { /* {Upop} */ + upop_list = CONS(cell, arg3, NIL); } else { /* [{Upop}] (probably) */ - upop_list = BIF_ARG_3; + upop_list = arg3; ret_list_prevp = &ret; } - if (!tb->common.meth->db_lookup_dbterm(tb, BIF_ARG_2, &handle)) { + if (!tb->common.meth->db_lookup_dbterm(p, tb, arg2, arg4, &handle)) { goto bail_out; /* key not found */ } @@ -957,13 +977,13 @@ BIF_RETTYPE ets_update_counter_3(BIF_ALIST_3) if (ret_list_prevp) { /* Prepare to return a list */ ret = NIL; halloc_size += list_size; - hstart = HAlloc(BIF_P, halloc_size); + hstart = HAlloc(p, halloc_size); ret_list_currp = hstart; htop = hstart + list_size; hend = hstart + halloc_size; } else { - hstart = htop = HAlloc(BIF_P, halloc_size); + hstart = htop = HAlloc(p, halloc_size); } hend = hstart + halloc_size; @@ -1010,26 +1030,54 @@ BIF_RETTYPE ets_update_counter_3(BIF_ALIST_3) (is_list(ret) && (list_val(ret)+list_size)==ret_list_currp)); ASSERT(htop <= hend); - HRelease(BIF_P,hend,htop); + HRelease(p, hend, htop); finalize: - tb->common.meth->db_finalize_dbterm(&handle); + tb->common.meth->db_finalize_dbterm(cret, &handle); bail_out: - UnUseTmpHeap(5,BIF_P); + UnUseTmpHeap(5, p); db_unlock(tb, LCK_WRITE_REC); switch (cret) { case DB_ERROR_NONE: BIF_RET(ret); case DB_ERROR_SYSRES: - BIF_ERROR(BIF_P, SYSTEM_LIMIT); + BIF_ERROR(p, SYSTEM_LIMIT); default: - BIF_ERROR(BIF_P, BADARG); + BIF_ERROR(p, BADARG); break; } } +/* +** update_counter(Tab, Key, Incr) +** update_counter(Tab, Key, Upop) +** update_counter(Tab, Key, [{Upop}]) +** Upop = {Pos,Incr} | {Pos,Incr,Threshold,WarpTo} +** Returns new value(s) (integer or [integer]) +*/ +BIF_RETTYPE ets_update_counter_3(BIF_ALIST_3) +{ + return do_update_counter(BIF_P, BIF_ARG_1, BIF_ARG_2, BIF_ARG_3, THE_NON_VALUE); +} + +/* +** update_counter(Tab, Key, Incr, Default) +** update_counter(Tab, Key, Upop, Default) +** update_counter(Tab, Key, [{Upop}], Default) +** Upop = {Pos,Incr} | {Pos,Incr,Threshold,WarpTo} +** Returns new value(s) (integer or [integer]) +*/ +BIF_RETTYPE ets_update_counter_4(BIF_ALIST_4) +{ + if (is_not_tuple(BIF_ARG_4)) { + BIF_ERROR(BIF_P, BADARG); + } + return do_update_counter(BIF_P, BIF_ARG_1, BIF_ARG_2, BIF_ARG_3, BIF_ARG_4); +} + + /* ** The put BIF */ @@ -2643,7 +2691,9 @@ BIF_RETTYPE ets_match_object_3(BIF_ALIST_3) BIF_RETTYPE ets_info_1(BIF_ALIST_1) { static Eterm fields[] = {am_protection, am_keypos, am_type, am_named_table, - am_node, am_size, am_name, am_heir, am_owner, am_memory, am_compressed}; + am_node, am_size, am_name, am_heir, am_owner, am_memory, am_compressed, + am_write_concurrency, + am_read_concurrency}; Eterm results[sizeof(fields)/sizeof(Eterm)]; DbTable* tb; Eterm res; @@ -3670,6 +3720,10 @@ static Eterm table_info(Process* p, DbTable* tb, Eterm What) ret = am_protected; else if (tb->common.status & DB_PUBLIC) ret = am_public; + } else if (What == am_write_concurrency) { + ret = tb->common.status & DB_FINE_LOCKED ? am_true : am_false; + } else if (What == am_read_concurrency) { + ret = tb->common.status & DB_FREQ_READ ? am_true : am_false; } else if (What == am_name) { ret = tb->common.the_name; } else if (What == am_keypos) { @@ -3752,7 +3806,7 @@ static Eterm table_info(Process* p, DbTable* tb, Eterm What) avg, std_dev_real, std_dev_exp, make_small(stats.min_chain_len), make_small(stats.max_chain_len), - make_small(db_kept_items_hash(&tb->hash))); + make_small(stats.kept_items)); } else { ret = am_false; @@ -3774,6 +3828,11 @@ static void print_table(int to, void *to_arg, int show, DbTable* tb) + sizeof(Uint) - 1) / sizeof(Uint))); + erts_print(to, to_arg, "Type: %T\n", table_info(NULL, tb, am_type)); + erts_print(to, to_arg, "Protection: %T\n", table_info(NULL, tb, am_protection)); + erts_print(to, to_arg, "Compressed: %T\n", table_info(NULL, tb, am_compressed)); + erts_print(to, to_arg, "Write Concurrency: %T\n", table_info(NULL, tb, am_write_concurrency)); + erts_print(to, to_arg, "Read Concurrency: %T\n", table_info(NULL, tb, am_read_concurrency)); } void db_info(int to, void *to_arg, int show) /* Called by break handler */ diff --git a/erts/emulator/beam/erl_db_hash.c b/erts/emulator/beam/erl_db_hash.c index 063808eb79..045c8ae135 100644 --- a/erts/emulator/beam/erl_db_hash.c +++ b/erts/emulator/beam/erl_db_hash.c @@ -171,53 +171,10 @@ static ERTS_INLINE void add_fixed_deletion(DbTableHash* tb, int ix) #define MAX_HASH 0xEFFFFFFFUL #define INVALID_HASH 0xFFFFFFFFUL -static ERTS_INLINE HashValue -MAKE_HASH(Eterm term) -{ - if (is_atom(term)) { - /* - * optimised version of make_hash, although poor hashvalue - * (normal case? atomic key) - */ - return atom_tab(atom_val(term))->slot.bucket.hvalue; - } - if (is_ref(term)) { - /* - * make_hash2() produce poor hash values - * for refs. - */ - int no; - Uint32 *ref; - HashValue hval; - if (is_internal_ref(term)) { - no = (int) internal_ref_no_of_numbers(term); - ref = internal_ref_numbers(term); - } - else { - no = (int) external_ref_no_of_numbers(term); - ref = external_ref_numbers(term); - } - switch (no) { - case 3: - ref_limit: - if (!ref[2]) - no = 2; - case 2: - if (!ref[1]) - no = 1; - case 1: - break; - default: - no = 3; - goto ref_limit; - } - hval = (HashValue) block_hash((byte *) ref, - no * sizeof(Uint32), - 0x08d12e65); - return hval % MAX_HASH; - } - return make_hash2(term) % MAX_HASH; -} +/* optimised version of make_hash (normal case? atomic key) */ +#define MAKE_HASH(term) \ + ((is_atom(term) ? (atom_tab(atom_val(term))->slot.bucket.hvalue) : \ + make_internal_hash(term)) % MAX_HASH) #ifdef ERTS_SMP # define DB_HASH_LOCK_MASK (DB_HASH_LOCK_CNT-1) @@ -425,7 +382,7 @@ static HashDbTerm* search_list(DbTableHash* tb, Eterm key, static void shrink(DbTableHash* tb, int nactive); static void grow(DbTableHash* tb, int nactive); static Eterm build_term_list(Process* p, HashDbTerm* ptr1, HashDbTerm* ptr2, - DbTableHash*); + Uint sz, DbTableHash*); static int analyze_pattern(DbTableHash *tb, Eterm pattern, struct mp_info *mpi); @@ -469,6 +426,7 @@ static int db_select_count_continue_hash(Process *p, DbTable *tbl, static int db_select_delete_continue_hash(Process *p, DbTable *tbl, Eterm continuation, Eterm *ret); +static int db_take_hash(Process *, DbTable *, Eterm, Eterm *); static void db_print_hash(int to, void *to_arg, int show, @@ -486,8 +444,11 @@ static int db_delete_all_objects_hash(Process* p, DbTable* tbl); #ifdef HARDDEBUG static void db_check_table_hash(DbTableHash *tb); #endif -static int db_lookup_dbterm_hash(DbTable *tbl, Eterm key, DbUpdateHandle* handle); -static void db_finalize_dbterm_hash(DbUpdateHandle* handle); +static int +db_lookup_dbterm_hash(Process *p, DbTable *tbl, Eterm key, Eterm obj, + DbUpdateHandle* handle); +static void +db_finalize_dbterm_hash(int cret, DbUpdateHandle* handle); static ERTS_INLINE void try_shrink(DbTableHash* tb) { @@ -579,6 +540,7 @@ DbTableMethod db_hash = db_select_delete_continue_hash, db_select_count_hash, db_select_count_continue_hash, + db_take_hash, db_delete_all_objects_hash, db_free_table_hash, db_free_table_continue_hash, @@ -689,25 +651,6 @@ restart: /* ToDo: Maybe try grow/shrink the table as well */ } -/* Only used by tests -*/ -Uint db_kept_items_hash(DbTableHash *tb) -{ - Uint kept_items = 0; - Uint ix = 0; - erts_smp_rwmtx_t* lck = RLOCK_HASH(tb,ix); - HashDbTerm* b; - do { - for (b = BUCKET(tb, ix); b != NULL; b = b->next) { - if (b->hvalue == INVALID_HASH) { - ++kept_items; - } - } - ix = next_slot(tb, ix, &lck); - }while (ix); - return kept_items; -} - int db_create_hash(Process *p, DbTable *tbl) { DbTableHash *tb = &tbl->hash; @@ -922,34 +865,49 @@ Ldone: return ret; } +static Eterm +get_term_list(Process *p, DbTableHash *tb, Eterm key, HashValue hval, + HashDbTerm *b1, HashDbTerm **bend) +{ + HashDbTerm* b2 = b1->next; + Eterm copy; + Uint sz = b1->dbterm.size + 2; + + if (tb->common.status & (DB_BAG | DB_DUPLICATE_BAG)) { + while (b2 && has_key(tb, b2, key, hval)) { + if (b2->hvalue != INVALID_HASH) + sz += b2->dbterm.size + 2; + + b2 = b2->next; + } + } + copy = build_term_list(p, b1, b2, sz, tb); + CHECK_TABLES(); + if (bend) { + *bend = b2; + } + return copy; +} + int db_get_hash(Process *p, DbTable *tbl, Eterm key, Eterm *ret) { DbTableHash *tb = &tbl->hash; HashValue hval; int ix; - HashDbTerm* b1; + HashDbTerm* b; erts_smp_rwmtx_t* lck; hval = MAKE_HASH(key); lck = RLOCK_HASH(tb,hval); ix = hash_to_ix(tb, hval); - b1 = BUCKET(tb, ix); + b = BUCKET(tb, ix); - while(b1 != 0) { - if (has_live_key(tb,b1,key,hval)) { - HashDbTerm* b2 = b1->next; - Eterm copy; - - if (tb->common.status & (DB_BAG | DB_DUPLICATE_BAG)) { - while(b2 != NULL && has_key(tb,b2,key,hval)) - b2 = b2->next; - } - copy = build_term_list(p, b1, b2, tb); - CHECK_TABLES(); - *ret = copy; + while(b != 0) { + if (has_live_key(tb, b, key, hval)) { + *ret = get_term_list(p, tb, key, hval, b, NULL); goto done; } - b1 = b1->next; + b = b->next; } *ret = NIL; done: @@ -1283,7 +1241,7 @@ static int db_slot_hash(Process *p, DbTable *tbl, Eterm slot_term, Eterm *ret) lck = RLOCK_HASH(tb, slot); nactive = NACTIVE(tb); if (slot < nactive) { - *ret = build_term_list(p, BUCKET(tb, slot), 0, tb); + *ret = build_term_list(p, BUCKET(tb, slot), NULL, 0, tb); retval = DB_ERROR_NONE; } else if (slot == nactive) { @@ -2112,6 +2070,46 @@ trap: } +static int db_take_hash(Process *p, DbTable *tbl, Eterm key, Eterm *ret) +{ + DbTableHash *tb = &tbl->hash; + HashDbTerm **bp, *b; + HashValue hval = MAKE_HASH(key); + erts_smp_rwmtx_t *lck = WLOCK_HASH(tb, hval); + int ix = hash_to_ix(tb, hval); + int nitems_diff = 0; + + *ret = NIL; + for (bp = &BUCKET(tb, ix), b = *bp; b; bp = &b->next, b = b->next) { + if (has_live_key(tb, b, key, hval)) { + HashDbTerm *bend; + + *ret = get_term_list(p, tb, key, hval, b, &bend); + while (b != bend) { + --nitems_diff; + if (nitems_diff == -1 && IS_FIXED(tb)) { + /* Pseudo remove (no need to keep several of same key) */ + add_fixed_deletion(tb, ix); + bp = &b->next; + b->hvalue = INVALID_HASH; + b = b->next; + } else { + *bp = b->next; + free_term(tb, b); + b = *bp; + } + } + break; + } + } + WUNLOCK_HASH(lck); + if (nitems_diff) { + erts_smp_atomic_add_nob(&tb->common.nitems, nitems_diff); + try_shrink(tb); + } + return DB_ERROR_NONE; +} + /* ** Other interface routines (not directly coupled to one bif) */ @@ -2147,10 +2145,38 @@ int db_mark_all_deleted_hash(DbTable *tbl) static void db_print_hash(int to, void *to_arg, int show, DbTable *tbl) { DbTableHash *tb = &tbl->hash; + DbHashStats stats; int i; erts_print(to, to_arg, "Buckets: %d\n", NACTIVE(tb)); - + +#ifdef ERTS_SMP + i = tbl->common.is_thread_safe; + /* If crash dumping we set table to thread safe in order to + avoid taking any locks */ + if (ERTS_IS_CRASH_DUMPING) + tbl->common.is_thread_safe = 1; + + db_calc_stats_hash(&tbl->hash, &stats); + + tbl->common.is_thread_safe = i; +#else + db_calc_stats_hash(&tbl->hash, &stats); +#endif + + erts_print(to, to_arg, "Chain Length Avg: %f\n", stats.avg_chain_len); + erts_print(to, to_arg, "Chain Length Max: %d\n", stats.max_chain_len); + erts_print(to, to_arg, "Chain Length Min: %d\n", stats.min_chain_len); + erts_print(to, to_arg, "Chain Length Std Dev: %f\n", + stats.std_dev_chain_len); + erts_print(to, to_arg, "Chain Length Expected Std Dev: %f\n", + stats.std_dev_expected); + + if (IS_FIXED(tb)) + erts_print(to, to_arg, "Fixed: %d\n", stats.kept_items); + else + erts_print(to, to_arg, "Fixed: false\n"); + if (show) { for (i = 0; i < NACTIVE(tb); i++) { HashDbTerm* list = BUCKET(tb,i); @@ -2526,23 +2552,23 @@ static int free_seg(DbTableHash *tb, int free_records) ** Copy terms from ptr1 until ptr2 ** works for ptr1 == ptr2 == 0 => [] ** or ptr2 == 0 +** sz is either precalculated heap size or 0 if not known */ static Eterm build_term_list(Process* p, HashDbTerm* ptr1, HashDbTerm* ptr2, - DbTableHash* tb) + Uint sz, DbTableHash* tb) { - int sz = 0; HashDbTerm* ptr; Eterm list = NIL; Eterm copy; Eterm *hp, *hend; - ptr = ptr1; - while(ptr != ptr2) { - - if (ptr->hvalue != INVALID_HASH) - sz += ptr->dbterm.size + 2; - - ptr = ptr->next; + if (!sz) { + ptr = ptr1; + while(ptr != ptr2) { + if (ptr->hvalue != INVALID_HASH) + sz += ptr->dbterm.size + 2; + ptr = ptr->next; + } } hp = HAlloc(p, sz); @@ -2773,59 +2799,129 @@ static HashDbTerm* next(DbTableHash *tb, Uint *iptr, erts_smp_rwmtx_t** lck_ptr, return NULL; } -static int db_lookup_dbterm_hash(DbTable *tbl, Eterm key, DbUpdateHandle* handle) +static int +db_lookup_dbterm_hash(Process *p, DbTable *tbl, Eterm key, Eterm obj, + DbUpdateHandle* handle) { DbTableHash *tb = &tbl->hash; - HashDbTerm* b; - HashDbTerm** prevp; - int ix; HashValue hval; + HashDbTerm **bp, *b; erts_smp_rwmtx_t* lck; + int flags = 0; + + ASSERT(tb->common.status & DB_SET); hval = MAKE_HASH(key); - lck = WLOCK_HASH(tb,hval); - ix = hash_to_ix(tb, hval); - prevp = &BUCKET(tb, ix); - b = *prevp; + lck = WLOCK_HASH(tb, hval); + bp = &BUCKET(tb, hash_to_ix(tb, hval)); + b = *bp; - while (b != 0) { - if (has_live_key(tb,b,key,hval)) { - handle->tb = tbl; - handle->bp = (void**) prevp; - handle->dbterm = &b->dbterm; - handle->mustResize = 0; - handle->new_size = b->dbterm.size; - #if HALFWORD_HEAP - handle->abs_vec = NULL; - #endif - handle->lck = lck; - /* KEEP hval WLOCKED, db_finalize_dbterm_hash will WUNLOCK */ - return 1; - } - prevp = &b->next; - b = *prevp; + for (;;) { + if (b == NULL) { + break; + } + if (has_key(tb, b, key, hval)) { + if (b->hvalue != INVALID_HASH) { + goto Ldone; + } + break; + } + bp = &b->next; + b = *bp; } - WUNLOCK_HASH(lck); - return 0; + + if (obj == THE_NON_VALUE) { + WUNLOCK_HASH(lck); + return 0; + } + + { + Eterm *objp = tuple_val(obj); + int arity = arityval(*objp); + Eterm *htop, *hend; + + ASSERT(arity >= tb->common.keypos); + htop = HAlloc(p, arity + 1); + hend = htop + arity + 1; + sys_memcpy(htop, objp, sizeof(Eterm) * (arity + 1)); + htop[tb->common.keypos] = key; + obj = make_tuple(htop); + + if (b == NULL) { + HashDbTerm *q = new_dbterm(tb, obj); + + q->hvalue = hval; + q->next = NULL; + *bp = b = q; + + { + int nitems = erts_smp_atomic_inc_read_nob(&tb->common.nitems); + int nactive = NACTIVE(tb); + + if (nitems > nactive * (CHAIN_LEN + 1) && !IS_FIXED(tb)) { + grow(tb, nactive); + } + } + } else { + HashDbTerm *q, *next = b->next; + + ASSERT(b->hvalue == INVALID_HASH); + q = replace_dbterm(tb, b, obj); + q->next = next; + q->hvalue = hval; + *bp = b = q; + erts_smp_atomic_inc_nob(&tb->common.nitems); + } + + HRelease(p, hend, htop); + flags |= DB_NEW_OBJECT; + } + +Ldone: + handle->tb = tbl; + handle->bp = (void **)bp; + handle->dbterm = &b->dbterm; + handle->flags = flags; + handle->new_size = b->dbterm.size; +#if HALFWORD_HEAP + handle->abs_vec = NULL; +#endif + handle->lck = lck; + return 1; } /* Must be called after call to db_lookup_dbterm */ -static void db_finalize_dbterm_hash(DbUpdateHandle* handle) +static void +db_finalize_dbterm_hash(int cret, DbUpdateHandle* handle) { DbTable* tbl = handle->tb; - HashDbTerm* oldp = (HashDbTerm*) *(handle->bp); + DbTableHash *tb = &tbl->hash; + HashDbTerm **bp = (HashDbTerm **) handle->bp; + HashDbTerm *b = *bp; erts_smp_rwmtx_t* lck = (erts_smp_rwmtx_t*) handle->lck; - ERTS_SMP_LC_ASSERT(IS_HASH_WLOCKED(&tbl->hash,lck)); /* locked by db_lookup_dbterm_hash */ + ERTS_SMP_LC_ASSERT(IS_HASH_WLOCKED(tb, lck)); /* locked by db_lookup_dbterm_hash */ + + ASSERT((&b->dbterm == handle->dbterm) == !(tb->common.compress && handle->flags & DB_MUST_RESIZE)); - ASSERT((&oldp->dbterm == handle->dbterm) == !(tbl->common.compress && handle->mustResize)); + if (handle->flags & DB_NEW_OBJECT && cret != DB_ERROR_NONE) { + if (IS_FIXED(tb)) { + add_fixed_deletion(tb, hash_to_ix(tb, b->hvalue)); + b->hvalue = INVALID_HASH; + } else { + *bp = b->next; + free_term(tb, b); + } - if (handle->mustResize) { + WUNLOCK_HASH(lck); + erts_smp_atomic_dec_nob(&tb->common.nitems); + try_shrink(tb); + } else if (handle->flags & DB_MUST_RESIZE) { db_finalize_resize(handle, offsetof(HashDbTerm,dbterm)); WUNLOCK_HASH(lck); - free_term(&tbl->hash, oldp); + free_term(tb, b); } else { WUNLOCK_HASH(lck); @@ -2876,6 +2972,7 @@ void db_calc_stats_hash(DbTableHash* tb, DbHashStats* stats) erts_smp_rwmtx_t* lck; int sum = 0; int sq_sum = 0; + int kept_items = 0; int ix; int len; @@ -2887,6 +2984,8 @@ void db_calc_stats_hash(DbTableHash* tb, DbHashStats* stats) len = 0; for (b = BUCKET(tb,ix); b!=NULL; b=b->next) { len++; + if (b->hvalue == INVALID_HASH) + ++kept_items; } sum += len; sq_sum += len*len; @@ -2898,7 +2997,8 @@ void db_calc_stats_hash(DbTableHash* tb, DbHashStats* stats) stats->std_dev_chain_len = sqrt((sq_sum - stats->avg_chain_len*sum) / NACTIVE(tb)); /* Expected standard deviation from a good uniform hash function, ie binomial distribution (not taking the linear hashing into acount) */ - stats->std_dev_expected = sqrt(stats->avg_chain_len * (1 - 1.0/NACTIVE(tb))); + stats->std_dev_expected = sqrt(stats->avg_chain_len * (1 - 1.0/NACTIVE(tb))); + stats->kept_items = kept_items; } #ifdef HARDDEBUG diff --git a/erts/emulator/beam/erl_db_hash.h b/erts/emulator/beam/erl_db_hash.h index 908cec11d4..f12cd363b0 100644 --- a/erts/emulator/beam/erl_db_hash.h +++ b/erts/emulator/beam/erl_db_hash.h @@ -1,7 +1,7 @@ /* * %CopyrightBegin% * - * Copyright Ericsson AB 1998-2013. All Rights Reserved. + * Copyright Ericsson AB 1998-2014. All Rights Reserved. * * The contents of this file are subject to the Erlang Public License, * Version 1.1, (the "License"); you may not use this file except in @@ -42,7 +42,7 @@ typedef struct hash_db_term { typedef struct db_table_hash_fine_locks { union { erts_smp_rwmtx_t lck; - byte _cache_line_alignment[64]; + byte _cache_line_alignment[ERTS_ALC_CACHE_LINE_ALIGN_SIZE(sizeof(erts_smp_rwmtx_t))]; }lck_vec[DB_HASH_LOCK_CNT]; } DbTableHashFineLocks; @@ -104,6 +104,7 @@ typedef struct { float std_dev_expected; int max_chain_len; int min_chain_len; + int kept_items; }DbHashStats; void db_calc_stats_hash(DbTableHash* tb, DbHashStats*); diff --git a/erts/emulator/beam/erl_db_tree.c b/erts/emulator/beam/erl_db_tree.c index a62a83a928..577da35b75 100644 --- a/erts/emulator/beam/erl_db_tree.c +++ b/erts/emulator/beam/erl_db_tree.c @@ -383,6 +383,7 @@ static int db_select_delete_tree(Process *p, DbTable *tbl, Eterm pattern, Eterm *ret); static int db_select_delete_continue_tree(Process *p, DbTable *tbl, Eterm continuation, Eterm *ret); +static int db_take_tree(Process *, DbTable *, Eterm, Eterm *); static void db_print_tree(int to, void *to_arg, int show, DbTable *tbl); static int db_free_table_tree(DbTable *tbl); @@ -398,8 +399,11 @@ static int db_delete_all_objects_tree(Process* p, DbTable* tbl); #ifdef HARDDEBUG static void db_check_table_tree(DbTable *tbl); #endif -static int db_lookup_dbterm_tree(DbTable *, Eterm key, DbUpdateHandle*); -static void db_finalize_dbterm_tree(DbUpdateHandle*); +static int +db_lookup_dbterm_tree(Process *, DbTable *, Eterm key, Eterm obj, + DbUpdateHandle*); +static void +db_finalize_dbterm_tree(int cret, DbUpdateHandle *); /* ** Static variables @@ -431,6 +435,7 @@ DbTableMethod db_tree = db_select_delete_continue_tree, db_select_count_tree, db_select_count_continue_tree, + db_take_tree, db_delete_all_objects_tree, db_free_table_tree, db_free_table_continue_tree, @@ -1722,6 +1727,28 @@ static int db_select_delete_tree(Process *p, DbTable *tbl, } +static int db_take_tree(Process *p, DbTable *tbl, Eterm key, Eterm *ret) +{ + DbTableTree *tb = &tbl->tree; + TreeDbTerm *this; + + *ret = NIL; + this = linkout_tree(tb, key, NULL); + if (this) { + Eterm copy, *hp, *hend; + + hp = HAlloc(p, this->dbterm.size + 2); + hend = hp + this->dbterm.size + 2; + copy = db_copy_object_from_ets(&tb->common, + &this->dbterm, &hp, &MSO(p)); + *ret = CONS(hp, copy, NIL); + hp += 2; + HRelease(p, hend, hp); + free_term(tb, this); + } + return DB_ERROR_NONE; +} + /* ** Other interface routines (not directly coupled to one bif) */ @@ -2522,16 +2549,43 @@ static TreeDbTerm **find_node2(DbTableTree *tb, Eterm key) return this; } -static int db_lookup_dbterm_tree(DbTable *tbl, Eterm key, DbUpdateHandle* handle) +static int +db_lookup_dbterm_tree(Process *p, DbTable *tbl, Eterm key, Eterm obj, + DbUpdateHandle* handle) { DbTableTree *tb = &tbl->tree; TreeDbTerm **pp = find_node2(tb, key); - - if (pp == NULL) return 0; + int flags = 0; + + if (pp == NULL) { + if (obj == THE_NON_VALUE) { + return 0; + } else { + Eterm *objp = tuple_val(obj); + int arity = arityval(*objp); + Eterm *htop, *hend; + + ASSERT(arity >= tb->common.keypos); + htop = HAlloc(p, arity + 1); + hend = htop + arity + 1; + sys_memcpy(htop, objp, sizeof(Eterm) * (arity + 1)); + htop[tb->common.keypos] = key; + obj = make_tuple(htop); + + if (db_put_tree(tbl, obj, 1) != DB_ERROR_NONE) { + return 0; + } + + pp = find_node2(tb, key); + ASSERT(pp != NULL); + HRelease(p, hend, htop); + flags |= DB_NEW_OBJECT; + } + } handle->tb = tbl; handle->dbterm = &(*pp)->dbterm; - handle->mustResize = 0; + handle->flags = flags; handle->bp = (void**) pp; handle->new_size = (*pp)->dbterm.size; #if HALFWORD_HEAP @@ -2540,15 +2594,21 @@ static int db_lookup_dbterm_tree(DbTable *tbl, Eterm key, DbUpdateHandle* handle return 1; } -static void db_finalize_dbterm_tree(DbUpdateHandle* handle) +static void +db_finalize_dbterm_tree(int cret, DbUpdateHandle *handle) { - if (handle->mustResize) { - TreeDbTerm* oldp = (TreeDbTerm*) *handle->bp; + DbTable *tbl = handle->tb; + DbTableTree *tb = &tbl->tree; + TreeDbTerm *bp = (TreeDbTerm *) *handle->bp; + if (handle->flags & DB_NEW_OBJECT && cret != DB_ERROR_NONE) { + Eterm ret; + db_erase_tree(tbl, GETKEY(tb, bp->dbterm.tpl), &ret); + } else if (handle->flags & DB_MUST_RESIZE) { db_finalize_resize(handle, offsetof(TreeDbTerm,dbterm)); - reset_static_stack(&handle->tb->tree); + reset_static_stack(tb); - free_term(&handle->tb->tree, oldp); + free_term(tb, bp); } #ifdef DEBUG handle->dbterm = 0; diff --git a/erts/emulator/beam/erl_db_util.c b/erts/emulator/beam/erl_db_util.c index 3927615e04..9d699d4b22 100644 --- a/erts/emulator/beam/erl_db_util.c +++ b/erts/emulator/beam/erl_db_util.c @@ -198,11 +198,6 @@ set_match_trace(Process *tracee_p, Eterm fail_term, Eterm tracer, return ret; } - -/* Type checking... */ - -#define BOXED_IS_TUPLE(Boxed) is_arity_value(*boxed_val((Boxed))) - /* ** ** Types and enum's (compiled matches) @@ -218,7 +213,9 @@ typedef enum { matchTuple, matchPushT, matchPushL, + matchPushM, matchPop, + matchSwap, matchBind, matchCmp, matchEqBin, @@ -227,11 +224,15 @@ typedef enum { matchEqRef, matchEq, matchList, + matchMap, + matchKey, matchSkip, matchPushC, matchConsA, /* Car is below Cdr */ matchConsB, /* Cdr is below Car (unusual) */ matchMkTuple, + matchMkFlatMap, + matchMkHashMap, matchCall0, matchCall1, matchCall2, @@ -856,6 +857,13 @@ static int match_compact(ErlHeapFragment *expr, DMCErrInfo *err_info); static Uint my_size_object(Eterm t); static Eterm my_copy_struct(Eterm t, Eterm **hp, ErlOffHeap* off_heap); +/* Guard subroutines */ +static void +dmc_rearrange_constants(DMCContext *context, DMC_STACK_TYPE(UWord) *text, + int textpos, Eterm *p, Uint nelems); +static DMCRet +dmc_array(DMCContext *context, DMCHeap *heap, DMC_STACK_TYPE(UWord) *text, + Eterm *p, Uint nelems, int *constant); /* Guard compilation */ static void do_emit_constant(DMCContext *context, DMC_STACK_TYPE(UWord) *text, Eterm t); @@ -869,6 +877,9 @@ static DMCRet dmc_tuple(DMCContext *context, DMC_STACK_TYPE(UWord) *text, Eterm t, int *constant); +static DMCRet +dmc_map(DMCContext *context, DMCHeap *heap, DMC_STACK_TYPE(UWord) *text, + Eterm t, int *constant); static DMCRet dmc_variable(DMCContext *context, DMCHeap *heap, DMC_STACK_TYPE(UWord) *text, @@ -888,12 +899,14 @@ static DMCRet compile_guard_expr(DMCContext *context, DMCHeap *heap, DMC_STACK_TYPE(UWord) *text, Eterm t); -/* match expression subroutine */ +/* match expression subroutines */ static DMCRet dmc_one_term(DMCContext *context, DMCHeap *heap, DMC_STACK_TYPE(Eterm) *stack, DMC_STACK_TYPE(UWord) *text, Eterm c); +static Eterm +dmc_private_copy(DMCContext *context, Eterm c); #ifdef DMC_DEBUG @@ -1364,7 +1377,112 @@ restart: for (;;) { switch (t & _TAG_PRIMARY_MASK) { case TAG_PRIMARY_BOXED: - if (!BOXED_IS_TUPLE(t)) { + if (is_flatmap(t)) { + num_iters = flatmap_get_size(flatmap_val(t)); + if (!structure_checked) { + DMC_PUSH(text, matchMap); + DMC_PUSH(text, num_iters); + } + structure_checked = 0; + for (i = 0; i < num_iters; ++i) { + Eterm key = flatmap_get_keys(flatmap_val(t))[i]; + if (db_is_variable(key) >= 0) { + if (context.err_info) { + add_dmc_err(context.err_info, + "Variable found in map key.", + -1, 0UL, dmcError); + } + goto error; + } else if (key == am_Underscore) { + if (context.err_info) { + add_dmc_err(context.err_info, + "Underscore found in map key.", + -1, 0UL, dmcError); + } + goto error; + } + DMC_PUSH(text, matchKey); + DMC_PUSH(text, dmc_private_copy(&context, key)); + { + int old_stack = ++(context.stack_used); + Eterm value = flatmap_get_values(flatmap_val(t))[i]; + res = dmc_one_term(&context, &heap, &stack, &text, + value); + ASSERT(res != retFail); + if (res == retRestart) { + goto restart; + } + if (old_stack != context.stack_used) { + ASSERT(old_stack + 1 == context.stack_used); + DMC_PUSH(text, matchSwap); + } + if (context.stack_used > context.stack_need) { + context.stack_need = context.stack_used; + } + DMC_PUSH(text, matchPop); + --(context.stack_used); + } + } + break; + } + if (is_hashmap(t)) { + DECLARE_WSTACK(wstack); + Eterm *kv; + num_iters = hashmap_size(t); + if (!structure_checked) { + DMC_PUSH(text, matchMap); + DMC_PUSH(text, num_iters); + } + structure_checked = 0; + + hashmap_iterator_init(&wstack, t, 0); + + while ((kv=hashmap_iterator_next(&wstack)) != NULL) { + Eterm key = CAR(kv); + Eterm value = CDR(kv); + if (db_is_variable(key) >= 0) { + if (context.err_info) { + add_dmc_err(context.err_info, + "Variable found in map key.", + -1, 0UL, dmcError); + } + DESTROY_WSTACK(wstack); + goto error; + } else if (key == am_Underscore) { + if (context.err_info) { + add_dmc_err(context.err_info, + "Underscore found in map key.", + -1, 0UL, dmcError); + } + DESTROY_WSTACK(wstack); + goto error; + } + DMC_PUSH(text, matchKey); + DMC_PUSH(text, dmc_private_copy(&context, key)); + { + int old_stack = ++(context.stack_used); + res = dmc_one_term(&context, &heap, &stack, &text, + value); + ASSERT(res != retFail); + if (res == retRestart) { + DESTROY_WSTACK(wstack); + goto restart; + } + if (old_stack != context.stack_used) { + ASSERT(old_stack + 1 == context.stack_used); + DMC_PUSH(text, matchSwap); + } + if (context.stack_used > context.stack_need) { + context.stack_need = context.stack_used; + } + DMC_PUSH(text, matchPop); + --(context.stack_used); + } + } + DESTROY_WSTACK(wstack); + break; + } + if (!is_tuple(t)) { goto simple_term; } num_iters = arityval(*tuple_val(t)); @@ -1715,10 +1833,8 @@ Eterm db_prog_match(Process *c_p, Binary *bprog, Uint32 *return_flags) { MatchProg *prog = Binary2MatchProg(bprog); - Eterm *ep; - Eterm *tp; + const Eterm *ep, *tp, **sp; Eterm t; - Eterm **sp; Eterm *esp; MatchVariable* variables; BeamInstr *cp; @@ -1808,7 +1924,7 @@ Eterm db_prog_match(Process *c_p, Binary *bprog, restart: ep = &term; esp = (Eterm*)((char*)mpsp->u.heap + prog->stack_offset); - sp = (Eterm **) esp; + sp = (const Eterm **)esp; ret = am_true; do_catch = 0; fail_label = -1; @@ -1887,9 +2003,57 @@ restart: *sp++ = list_val_rel(*ep,base); ++ep; break; + case matchMap: + if (!is_map_rel(*ep, base)) { + FAIL(); + } + n = *pc++; + if (is_flatmap_rel(*ep,base)) { + if (flatmap_get_size(flatmap_val_rel(*ep, base)) < n) { + FAIL(); + } + } else { + ASSERT(is_hashmap_rel(*ep,base)); + if (hashmap_size_rel(*ep, base) < n) { + FAIL(); + } + } + ep = flatmap_val_rel(*ep, base); + break; + case matchPushM: + if (!is_map_rel(*ep, base)) { + FAIL(); + } + n = *pc++; + if (is_flatmap_rel(*ep,base)) { + if (flatmap_get_size(flatmap_val_rel(*ep, base)) < n) { + FAIL(); + } + } else { + ASSERT(is_hashmap_rel(*ep,base)); + if (hashmap_size_rel(*ep, base) < n) { + FAIL(); + } + } + *sp++ = flatmap_val_rel(*ep++, base); + break; + case matchKey: + t = (Eterm) *pc++; + tp = erts_maps_get_rel(t, make_flatmap_rel(ep, base), base); + if (!tp) { + FAIL(); + } + *sp++ = ep; + ep = tp; + break; case matchPop: ep = *(--sp); break; + case matchSwap: + tp = sp[-1]; + sp[-1] = sp[-2]; + sp[-2] = tp; + break; case matchBind: n = *pc++; variables[n].term = *ep++; @@ -1987,6 +2151,38 @@ restart: } *esp++ = t; break; + case matchMkFlatMap: + n = *pc++; + ehp = HAllocX(build_proc, 1 + MAP_HEADER_SIZE + n, HEAP_XTRA); + t = *ehp++ = *--esp; + { + flatmap_t *m = (flatmap_t *)ehp; + m->thing_word = MAP_HEADER; + m->size = n; + m->keys = t; + } + t = make_flatmap(ehp); + ehp += MAP_HEADER_SIZE; + while (n--) { + *ehp++ = *--esp; + } + *esp++ = t; + break; + case matchMkHashMap: + n = *pc++; + esp -= 2*n; + ehp = HAllocX(build_proc, 2*n, HEAP_XTRA); + { + ErtsHeapFactory factory; + Uint ix; + factory.p = build_proc; + for (ix = 0; ix < 2*n; ix++){ + ehp[ix] = esp[ix]; + } + t = erts_hashmap_from_array(&factory, ehp, n, 0); + } + *esp++ = t; + break; case matchCall0: bif = (Eterm (*)(Process*, ...)) *pc++; t = (*bif)(build_proc, bif_args); @@ -2601,10 +2797,10 @@ Wterm db_do_read_element(DbUpdateHandle* handle, Sint position) } ASSERT(((DbTableCommon*)handle->tb)->compress); - ASSERT(!handle->mustResize); + ASSERT(!(handle->flags & DB_MUST_RESIZE)); handle->dbterm = db_alloc_tmp_uncompressed(&handle->tb->common, handle->dbterm); - handle->mustResize = 1; + handle->flags |= DB_MUST_RESIZE; return handle->dbterm->tpl[position]; } @@ -2637,11 +2833,11 @@ void db_do_update_element(DbUpdateHandle* handle, #endif return; } - if (!handle->mustResize) { + if (!(handle->flags & DB_MUST_RESIZE)) { if (handle->tb->common.compress) { handle->dbterm = db_alloc_tmp_uncompressed(&handle->tb->common, handle->dbterm); - handle->mustResize = 1; + handle->flags |= DB_MUST_RESIZE; oldval = handle->dbterm->tpl[position]; #if HALFWORD_HEAP old_base = NULL; @@ -2701,7 +2897,7 @@ both_size_set: /* write new value in old dbterm, finalize will make a flat copy */ handle->dbterm->tpl[position] = newval; - handle->mustResize = 1; + handle->flags |= DB_MUST_RESIZE; #if HALFWORD_HEAP if (old_base && newval_sz > 0) { @@ -3156,34 +3352,45 @@ int db_is_variable(Eterm obj) /* return 1 if obj contains a variable or underscore */ /* return 0 if obj is fully ground */ -int db_has_variable(Eterm obj) -{ - switch(obj & _TAG_PRIMARY_MASK) { - case TAG_PRIMARY_LIST: { - while (is_list(obj)) { - if (db_has_variable(CAR(list_val(obj)))) +int db_has_variable(Eterm node) { + DECLARE_ESTACK(s); + + ESTACK_PUSH(s,node); + while (!ESTACK_ISEMPTY(s)) { + node = ESTACK_POP(s); + switch(node & _TAG_PRIMARY_MASK) { + case TAG_PRIMARY_LIST: + while (is_list(node)) { + ESTACK_PUSH(s,CAR(list_val(node))); + node = CDR(list_val(node)); + } + ESTACK_PUSH(s,node); /* Non wellformed list or [] */ + break; + case TAG_PRIMARY_BOXED: + if (is_tuple(node)) { + Eterm *tuple = tuple_val(node); + int arity = arityval(*tuple); + while(arity--) { + ESTACK_PUSH(s,*(++tuple)); + } + } else if (is_flatmap(node)) { + Eterm *values = flatmap_get_values(flatmap_val(node)); + Uint size = flatmap_get_size(flatmap_val(node)); + ESTACK_PUSH(s, ((flatmap_t *) flatmap_val(node))->keys); + while (size--) { + ESTACK_PUSH(s, *(values++)); + } + } + break; + case TAG_PRIMARY_IMMED1: + if (node == am_Underscore || db_is_variable(node) >= 0) { + DESTROY_ESTACK(s); return 1; - obj = CDR(list_val(obj)); - } - return(db_has_variable(obj)); /* Non wellformed list or [] */ - } - case TAG_PRIMARY_BOXED: - if (!BOXED_IS_TUPLE(obj)) { - return 0; - } else { - Eterm *tuple = tuple_val(obj); - int arity = arityval(*tuple++); - while(arity--) { - if (db_has_variable(*tuple)) - return 1; - tuple++; } - return(0); + break; } - case TAG_PRIMARY_IMMED1: - if (obj == am_Underscore || db_is_variable(obj) >= 0) - return 1; } + DESTROY_ESTACK(s); return 0; } @@ -3243,11 +3450,9 @@ static DMCRet dmc_one_term(DMCContext *context, { Sint n; Eterm *hp; - ErlHeapFragment *tmp_mb; Uint sz, sz2, sz3; Uint i, j; - switch (c & _TAG_PRIMARY_MASK) { case TAG_PRIMARY_IMMED1: if ((n = db_is_variable(c)) >= 0) { /* variable */ @@ -3334,6 +3539,20 @@ static DMCRet dmc_one_term(DMCContext *context, DMC_PUSH(*text, n); DMC_PUSH(*stack, c); break; + case (_TAG_HEADER_MAP >> _TAG_PRIMARY_SIZE): + n = flatmap_get_size(flatmap_val(c)); + DMC_PUSH(*text, matchPushM); + ++(context->stack_used); + DMC_PUSH(*text, n); + DMC_PUSH(*stack, c); + break; + case (_TAG_HEADER_HASHMAP >> _TAG_PRIMARY_SIZE): + n = hashmap_size(c); + DMC_PUSH(*text, matchPushM); + ++(context->stack_used); + DMC_PUSH(*text, n); + DMC_PUSH(*stack, c); + break; case (_TAG_HEADER_REF >> _TAG_PRIMARY_SIZE): { Eterm* ref_val = internal_ref_val(c); @@ -3415,16 +3634,8 @@ static DMCRet dmc_one_term(DMCContext *context, #endif break; default: /* BINARY, FUN, VECTOR, or EXTERNAL */ - /* - ** Make a private copy... - */ - n = size_object(c); - tmp_mb = new_message_buffer(n); - hp = tmp_mb->mem; DMC_PUSH(*text, matchEqBin); - DMC_PUSH(*text, copy_struct(c, n, &hp, &(tmp_mb->off_heap))); - tmp_mb->next = context->save; - context->save = tmp_mb; + DMC_PUSH(*text, dmc_private_copy(context, c)); break; } break; @@ -3437,6 +3648,22 @@ static DMCRet dmc_one_term(DMCContext *context, } /* +** Make a private copy of a term in a context. +*/ + +static Eterm +dmc_private_copy(DMCContext *context, Eterm c) +{ + Uint n = size_object(c); + ErlHeapFragment *tmp_mb = new_message_buffer(n); + Eterm *hp = tmp_mb->mem; + Eterm copy = copy_struct(c, n, &hp, &(tmp_mb->off_heap)); + tmp_mb->next = context->save; + context->save = tmp_mb; + return copy; +} + +/* ** Match guard compilation */ @@ -3527,57 +3754,78 @@ static DMCRet dmc_list(DMCContext *context, return retOk; } -static DMCRet dmc_tuple(DMCContext *context, - DMCHeap *heap, - DMC_STACK_TYPE(UWord) *text, - Eterm t, - int *constant) +static void +dmc_rearrange_constants(DMCContext *context, DMC_STACK_TYPE(UWord) *text, + int textpos, Eterm *p, Uint nelems) { DMC_STACK_TYPE(UWord) instr_save; + Uint i; + + DMC_INIT_STACK(instr_save); + while (DMC_STACK_NUM(*text) > textpos) { + DMC_PUSH(instr_save, DMC_POP(*text)); + } + for (i = nelems; i--;) { + do_emit_constant(context, text, p[i]); + } + while(!DMC_EMPTY(instr_save)) { + DMC_PUSH(*text, DMC_POP(instr_save)); + } + DMC_FREE(instr_save); +} + +static DMCRet +dmc_array(DMCContext *context, DMCHeap *heap, DMC_STACK_TYPE(UWord) *text, + Eterm *p, Uint nelems, int *constant) +{ int all_constant = 1; int textpos = DMC_STACK_NUM(*text); - Eterm *p = tuple_val(t); - Uint nelems = arityval(*p); Uint i; - int c; - DMCRet ret; /* - ** We remember where we started to layout code, + ** We remember where we started to layout code, ** assume all is constant and back up and restart if not so. - ** The tuple should be laid out with the last element first, - ** so we can memcpy the tuple to the eheap. + ** The array should be laid out with the last element first, + ** so we can memcpy it to the eheap. */ - for (i = nelems; i > 0; --i) { - if ((ret = dmc_expr(context, heap, text, p[i], &c)) != retOk) - return ret; - if (!c && all_constant) { - all_constant = 0; - if (i < nelems) { - Uint j; + for (i = nelems; i--;) { + DMCRet ret; + int c; - /* - * Oops, we need to relayout the constants. - * Save the already laid out instructions. - */ - DMC_INIT_STACK(instr_save); - while (DMC_STACK_NUM(*text) > textpos) - DMC_PUSH(instr_save, DMC_POP(*text)); - for (j = nelems; j > i; --j) - do_emit_constant(context, text, p[j]); - while(!DMC_EMPTY(instr_save)) - DMC_PUSH(*text, DMC_POP(instr_save)); - DMC_FREE(instr_save); - } - } else if (c && !all_constant) { - /* push a constant */ - do_emit_constant(context, text, p[i]); - } + ret = dmc_expr(context, heap, text, p[i], &c); + if (ret != retOk) { + return ret; + } + if (!c && all_constant) { + all_constant = 0; + if (i < nelems - 1) { + dmc_rearrange_constants(context, text, textpos, + p + i + 1, nelems - i - 1); + } + } else if (c && !all_constant) { + do_emit_constant(context, text, p[i]); + } + } + *constant = all_constant; + return retOk; +} + +static DMCRet +dmc_tuple(DMCContext *context, DMCHeap *heap, DMC_STACK_TYPE(UWord) *text, + Eterm t, int *constant) +{ + int all_constant; + Eterm *p = tuple_val(t); + Uint nelems = arityval(*p); + DMCRet ret; + + ret = dmc_array(context, heap, text, p + 1, nelems, &all_constant); + if (ret != retOk) { + return ret; } - if (all_constant) { - *constant = 1; - return retOk; + *constant = 1; + return retOk; } DMC_PUSH(*text, matchMkTuple); DMC_PUSH(*text, nelems); @@ -3586,6 +3834,93 @@ static DMCRet dmc_tuple(DMCContext *context, return retOk; } +static DMCRet +dmc_map(DMCContext *context, DMCHeap *heap, DMC_STACK_TYPE(UWord) *text, + Eterm t, int *constant) +{ + int nelems; + int constant_values; + DMCRet ret; + if (is_flatmap(t)) { + flatmap_t *m = (flatmap_t *)flatmap_val(t); + Eterm *values = flatmap_get_values(m); + + nelems = flatmap_get_size(m); + ret = dmc_array(context, heap, text, values, nelems, &constant_values); + + if (ret != retOk) { + return ret; + } + if (constant_values) { + *constant = 1; + return retOk; + } + DMC_PUSH(*text, matchPushC); + DMC_PUSH(*text, dmc_private_copy(context, m->keys)); + if (++context->stack_used > context->stack_need) { + context->stack_need = context->stack_used; + } + DMC_PUSH(*text, matchMkFlatMap); + DMC_PUSH(*text, nelems); + context->stack_used -= nelems; + *constant = 0; + return retOk; + } else { + DECLARE_WSTACK(wstack); + Eterm *kv; + int c; + + ASSERT(is_hashmap(t)); + + hashmap_iterator_init(&wstack, t, 1); + constant_values = 1; + nelems = hashmap_size(t); + + while ((kv=hashmap_iterator_prev(&wstack)) != NULL) { + if ((ret = dmc_expr(context, heap, text, CDR(kv), &c)) != retOk) { + DESTROY_WSTACK(wstack); + return ret; + } + if (!c) + constant_values = 0; + } + + if (constant_values) { + *constant = 1; + DESTROY_WSTACK(wstack); + return retOk; + } + + *constant = 0; + + hashmap_iterator_init(&wstack, t, 1); + + while ((kv=hashmap_iterator_prev(&wstack)) != NULL) { + /* push key */ + if ((ret = dmc_expr(context, heap, text, CAR(kv), &c)) != retOk) { + DESTROY_WSTACK(wstack); + return ret; + } + if (c) { + do_emit_constant(context, text, CAR(kv)); + } + /* push value */ + if ((ret = dmc_expr(context, heap, text, CDR(kv), &c)) != retOk) { + DESTROY_WSTACK(wstack); + return ret; + } + if (c) { + do_emit_constant(context, text, CDR(kv)); + } + } + DMC_PUSH(*text, matchMkHashMap); + DMC_PUSH(*text, nelems); + context->stack_used -= nelems; + DESTROY_WSTACK(wstack); + return retOk; + } +} + static DMCRet dmc_whole_expression(DMCContext *context, DMCHeap *heap, DMC_STACK_TYPE(UWord) *text, @@ -4580,7 +4915,10 @@ static DMCRet dmc_expr(DMCContext *context, return ret; break; case TAG_PRIMARY_BOXED: - if (!BOXED_IS_TUPLE(t)) { + if (is_map(t)) { + return dmc_map(context, heap, text, t, constant); + } + if (!is_tuple(t)) { goto simple_term; } p = tuple_val(t); @@ -4855,7 +5193,7 @@ static Eterm my_copy_struct(Eterm t, Eterm **hp, ErlOffHeap* off_heap) *hp += 2; break; case TAG_PRIMARY_BOXED: - if (BOXED_IS_TUPLE(t)) { + if (is_tuple(t)) { if (arityval(*tuple_val(t)) == 1 && is_tuple(a = tuple_val(t)[1])) { Uint i,n; @@ -5126,6 +5464,18 @@ void db_match_dis(Binary *bp) ++t; erts_printf("Tuple\t%beu\n", n); break; + case matchMap: + ++t; + n = *t; + ++t; + erts_printf("Map\t%beu\n", n); + break; + case matchKey: + ++t; + p = (Eterm) *t; + ++t; + erts_printf("Key\t%p (%T)\n", t, p); + break; case matchPushT: ++t; n = *t; @@ -5136,10 +5486,20 @@ void db_match_dis(Binary *bp) ++t; erts_printf("PushL\n"); break; + case matchPushM: + ++t; + n = *t; + ++t; + erts_printf("PushM\t%beu\n", n); + break; case matchPop: ++t; erts_printf("Pop\n"); break; + case matchSwap: + ++t; + erts_printf("Swap\n"); + break; case matchBind: ++t; n = *t; @@ -5252,6 +5612,18 @@ void db_match_dis(Binary *bp) ++t; erts_printf("MkTuple\t%beu\n", n); break; + case matchMkFlatMap: + ++t; + n = *t; + ++t; + erts_printf("MkFlatMap\t%beu\n", n); + break; + case matchMkHashMap: + ++t; + n = *t; + ++t; + erts_printf("MkHashMap\t%beu\n", n); + break; case matchOr: ++t; n = *t; diff --git a/erts/emulator/beam/erl_db_util.h b/erts/emulator/beam/erl_db_util.h index 328b19dfc9..ca206c7f58 100644 --- a/erts/emulator/beam/erl_db_util.h +++ b/erts/emulator/beam/erl_db_util.h @@ -76,6 +76,9 @@ typedef struct db_term { union db_table; typedef union db_table DbTable; +#define DB_MUST_RESIZE 1 +#define DB_NEW_OBJECT 2 + /* Info about a database entry while it's being updated * (by update_counter or update_element) */ @@ -84,7 +87,7 @@ typedef struct { DbTerm* dbterm; void** bp; /* {Hash|Tree}DbTerm** */ Uint new_size; - int mustResize; + int flags; void* lck; #if HALFWORD_HEAP unsigned char* abs_vec; /* [i] true if dbterm->tpl[i] is absolute Eterm */ @@ -165,6 +168,7 @@ typedef struct db_table_method DbTable* tb, /* [in out] */ Eterm continuation, Eterm* ret); + int (*db_take)(Process *, DbTable *, Eterm, Eterm *); int (*db_delete_all_objects)(Process* p, DbTable* db /* [in out] */ ); @@ -182,15 +186,14 @@ typedef struct db_table_method void *arg); void (*db_check_table)(DbTable* tb); - /* Lookup a dbterm for updating. Return false if not found. - */ - int (*db_lookup_dbterm)(DbTable*, Eterm key, - DbUpdateHandle* handle); /* [out] */ + /* Lookup a dbterm for updating. Return false if not found. */ + int (*db_lookup_dbterm)(Process *, DbTable *, Eterm key, Eterm obj, + DbUpdateHandle* handle); - /* Must be called for each db_lookup_dbterm that returned true, - ** even if dbterm was not updated. - */ - void (*db_finalize_dbterm)(DbUpdateHandle* handle); + /* Must be called for each db_lookup_dbterm that returned true, even if + ** dbterm was not updated. If the handle was of a new object and cret is + ** not DB_ERROR_NONE, the object is removed from the table. */ + void (*db_finalize_dbterm)(int cret, DbUpdateHandle* handle); } DbTableMethod; diff --git a/erts/emulator/beam/erl_driver.h b/erts/emulator/beam/erl_driver.h index f9938fc66c..e498ac70ec 100644 --- a/erts/emulator/beam/erl_driver.h +++ b/erts/emulator/beam/erl_driver.h @@ -133,7 +133,7 @@ typedef struct { #define ERL_DRV_EXTENDED_MARKER (0xfeeeeeed) #define ERL_DRV_EXTENDED_MAJOR_VERSION 3 -#define ERL_DRV_EXTENDED_MINOR_VERSION 1 +#define ERL_DRV_EXTENDED_MINOR_VERSION 2 /* * The emulator will refuse to load a driver with a major version @@ -361,6 +361,9 @@ typedef struct erl_drv_entry { /* Called on behalf of driver_select when it is safe to release 'event'. A typical unix driver would call close(event) */ + void (*emergency_close)(ErlDrvData drv_data); + /* called when the port is closed abruptly. + specifically when erl_crash_dump is called. */ /* When adding entries here, dont forget to pad in obsolete/driver.h */ } ErlDrvEntry; diff --git a/erts/emulator/beam/erl_drv_thread.c b/erts/emulator/beam/erl_drv_thread.c index 147249f751..31b05d22af 100644 --- a/erts/emulator/beam/erl_drv_thread.c +++ b/erts/emulator/beam/erl_drv_thread.c @@ -601,17 +601,14 @@ erl_drv_thread_create(char *name, #ifdef USE_THREADS int res; struct ErlDrvTid_ *dtid; - ethr_thr_opts ethr_opts; + ethr_thr_opts ethr_opts = ETHR_THR_OPTS_DEFAULT_INITER; ethr_thr_opts *use_opts; - ethr_thr_opts def_ethr_opts = ETHR_THR_OPTS_DEFAULT_INITER; if (!opts) use_opts = NULL; else { - sys_memcpy((void *) ðr_opts, - (void *) &def_ethr_opts, - sizeof(ethr_thr_opts)); ethr_opts.suggested_stack_size = opts->suggested_stack_size; + ethr_opts.name = name; use_opts = ðr_opts; } diff --git a/erts/emulator/beam/erl_gc.c b/erts/emulator/beam/erl_gc.c index 9a05e5b23a..4a116c0740 100644 --- a/erts/emulator/beam/erl_gc.c +++ b/erts/emulator/beam/erl_gc.c @@ -20,6 +20,8 @@ # include "config.h" #endif +#define ERL_WANT_GC_INTERNALS__ + #include "sys.h" #include "erl_vm.h" #include "global.h" @@ -174,15 +176,15 @@ erts_init_gc(void) int i = 0, ix; Sint max_heap_size = 0; - ASSERT(offsetof(ProcBin,thing_word) == offsetof(struct erl_off_heap_header,thing_word)); - ASSERT(offsetof(ProcBin,thing_word) == offsetof(ErlFunThing,thing_word)); - ASSERT(offsetof(ProcBin,thing_word) == offsetof(ExternalThing,header)); - ASSERT(offsetof(ProcBin,size) == offsetof(struct erl_off_heap_header,size)); - ASSERT(offsetof(ProcBin,size) == offsetof(ErlSubBin,size)); - ASSERT(offsetof(ProcBin,size) == offsetof(ErlHeapBin,size)); - ASSERT(offsetof(ProcBin,next) == offsetof(struct erl_off_heap_header,next)); - ASSERT(offsetof(ProcBin,next) == offsetof(ErlFunThing,next)); - ASSERT(offsetof(ProcBin,next) == offsetof(ExternalThing,next)); + ERTS_CT_ASSERT(offsetof(ProcBin,thing_word) == offsetof(struct erl_off_heap_header,thing_word)); + ERTS_CT_ASSERT(offsetof(ProcBin,thing_word) == offsetof(ErlFunThing,thing_word)); + ERTS_CT_ASSERT(offsetof(ProcBin,thing_word) == offsetof(ExternalThing,header)); + ERTS_CT_ASSERT(offsetof(ProcBin,size) == offsetof(struct erl_off_heap_header,size)); + ERTS_CT_ASSERT(offsetof(ProcBin,size) == offsetof(ErlSubBin,size)); + ERTS_CT_ASSERT(offsetof(ProcBin,size) == offsetof(ErlHeapBin,size)); + ERTS_CT_ASSERT(offsetof(ProcBin,next) == offsetof(struct erl_off_heap_header,next)); + ERTS_CT_ASSERT(offsetof(ProcBin,next) == offsetof(ErlFunThing,next)); + ERTS_CT_ASSERT(offsetof(ProcBin,next) == offsetof(ExternalThing,next)); erts_test_long_gc_sleep = 0; @@ -2410,7 +2412,6 @@ sweep_off_heap(Process *p, int fullsweep) } pb->val = erts_bin_realloc(pb->val, new_size); - pb->val->orig_size = new_size; pb->bytes = (byte *) pb->val->orig_bytes; } } diff --git a/erts/emulator/beam/erl_gc.h b/erts/emulator/beam/erl_gc.h index 5203dda263..8afcb060a1 100644 --- a/erts/emulator/beam/erl_gc.h +++ b/erts/emulator/beam/erl_gc.h @@ -20,10 +20,12 @@ #ifndef __ERL_GC_H__ #define __ERL_GC_H__ -#include "erl_map.h" +#if defined(ERL_WANT_GC_INTERNALS__) || defined(ERTS_DO_INCL_GLB_INLINE_FUNC_DEF) /* GC declarations shared by beam/erl_gc.c and hipe/hipe_gc.c */ +#include "erl_map.h" + #if defined(DEBUG) && !ERTS_GLB_INLINE_INCL_FUNC_DEF # define HARDDEBUG 1 #endif @@ -53,7 +55,9 @@ do { \ nelts = header_arity(HDR); \ switch ((HDR) & _HEADER_SUBTAG_MASK) { \ case SUB_BINARY_SUBTAG: nelts++; break; \ - case MAP_SUBTAG: nelts+=map_get_size(PTR) + 1; break; \ + case MAP_SUBTAG: nelts+=flatmap_get_size(PTR) + 1; break; \ + case HASHMAP_SUBTAG: nelts=hashmap_bitcount(MAP_HEADER_VAL(HDR)); \ + nelts += is_hashmap_header_head(HDR) ? 1 : 0; break; \ case FUN_SUBTAG: nelts+=((ErlFunThing*)(PTR))->num_free+1; break; \ } \ gval = make_boxed(HTOP); \ @@ -61,14 +65,11 @@ do { \ *HTOP++ = HDR; \ *PTR++ = gval; \ while (nelts--) *HTOP++ = *PTR++; \ - \ } while(0) #define in_area(ptr,start,nbytes) \ ((UWord)((char*)(ptr) - (char*)(start)) < (nbytes)) -extern Uint erts_test_long_gc_sleep; - #if defined(DEBUG) || defined(ERTS_OFFHEAP_DEBUG) int within(Eterm *ptr, Process *p); #endif @@ -97,4 +98,33 @@ ERTS_GLB_INLINE Eterm follow_moved(Eterm term) } #endif +#endif /* ERL_GC_C__ || HIPE_GC_C__ */ + +/* + * Global exported + */ + +extern Uint erts_test_long_gc_sleep; + +typedef struct { + Uint64 reclaimed; + Uint64 garbage_cols; +} ErtsGCInfo; + +void erts_gc_info(ErtsGCInfo *gcip); +void erts_init_gc(void); +int erts_garbage_collect(struct process*, int, Eterm*, int); +void erts_garbage_collect_hibernate(struct process* p); +Eterm erts_gc_after_bif_call(struct process* p, Eterm result, Eterm* regs, Uint arity); +void erts_garbage_collect_literals(struct process* p, Eterm* literals, + Uint lit_size, + struct erl_off_heap_header* oh); +Uint erts_next_heap_size(Uint, Uint); +Eterm erts_heap_sizes(struct process* p); + +void erts_offset_off_heap(struct erl_off_heap*, Sint, Eterm*, Eterm*); +void erts_offset_heap_ptr(Eterm*, Uint, Sint, Eterm*, Eterm*); +void erts_offset_heap(Eterm*, Uint, Sint, Eterm*, Eterm*); +void erts_free_heap_frags(struct process* p); + #endif /* __ERL_GC_H__ */ diff --git a/erts/emulator/beam/erl_init.c b/erts/emulator/beam/erl_init.c index 8d38ff9c2c..86d3416423 100644 --- a/erts/emulator/beam/erl_init.c +++ b/erts/emulator/beam/erl_init.c @@ -164,9 +164,6 @@ int H_MIN_SIZE; /* The minimum heap grain */ int BIN_VH_MIN_SIZE; /* The minimum binary virtual*/ Uint32 erts_debug_flags; /* Debug flags. */ -#ifdef ERTS_OPCODE_COUNTER_SUPPORT -int count_instructions; -#endif int erts_backtrace_depth; /* How many functions to show in a backtrace * in error codes. */ @@ -193,6 +190,8 @@ Uint32 verbose; /* See erl_debug.h for information about verbose */ int erts_atom_table_size = ATOM_LIMIT; /* Maximum number of atoms */ +int erts_pd_initial_size = 10; + int erts_modified_timing_level; int erts_no_crash_dump = 0; /* Use -d to suppress crash dump. */ @@ -546,6 +545,8 @@ void erts_usage(void) H_DEFAULT_SIZE); erts_fprintf(stderr, "-hmbs size set minimum binary virtual heap size in words (default %d)\n", VH_DEFAULT_SIZE); + erts_fprintf(stderr, "-hpds size initial process dictionary size (default %d)\n", + erts_pd_initial_size); /* erts_fprintf(stderr, "-i module set the boot module (default init)\n"); */ @@ -1438,6 +1439,7 @@ erl_start(int argc, char **argv) * * h|ms - min_heap_size * h|mbs - min_bin_vheap_size + * h|pds - erts_pd_initial_size * */ if (has_prefix("mbs", sub_param)) { @@ -1455,6 +1457,14 @@ erl_start(int argc, char **argv) erts_usage(); } VERBOSE(DEBUG_SYSTEM, ("using minimum heap size %d\n", H_MIN_SIZE)); + } else if (has_prefix("pds", sub_param)) { + arg = get_arg(sub_param+3, argv[i+1], &i); + if ((erts_pd_initial_size = atoi(arg)) <= 0) { + erts_fprintf(stderr, "bad initial process dictionary size %s\n", arg); + erts_usage(); + } + VERBOSE(DEBUG_SYSTEM, ("using initial process dictionary size %d\n", + erts_pd_initial_size)); } else { /* backward compatibility */ arg = get_arg(argv[i]+2, argv[i+1], &i); diff --git a/erts/emulator/beam/erl_instrument.c b/erts/emulator/beam/erl_instrument.c index df7c443387..da85b86c87 100644 --- a/erts/emulator/beam/erl_instrument.c +++ b/erts/emulator/beam/erl_instrument.c @@ -1226,7 +1226,7 @@ erts_instr_init(int stat, int map_stat) mem_anchor = NULL; /* Install instrumentation functions */ - ASSERT(sizeof(erts_allctrs) == sizeof(real_allctrs)); + ERTS_CT_ASSERT(sizeof(erts_allctrs) == sizeof(real_allctrs)); sys_memcpy((void *)real_allctrs,(void *)erts_allctrs,sizeof(erts_allctrs)); diff --git a/erts/emulator/beam/erl_map.c b/erts/emulator/beam/erl_map.c index 5e740aacdd..bcbda65da0 100644 --- a/erts/emulator/beam/erl_map.c +++ b/erts/emulator/beam/erl_map.c @@ -16,6 +16,9 @@ * * %CopyrightEnd% * + * hashmaps are an adaption of Rich Hickeys Persistent HashMaps + * which were an adaption of Phil Bagwells - Hash Array Mapped Tries + * * Author: Björn-Egil Dahlberg */ @@ -62,39 +65,78 @@ * - erts_internal:map_to_tuple_keys/1 */ +#ifndef DECL_AM +#define DECL_AM(S) Eterm AM_ ## S = am_atom_put(#S, sizeof(#S) - 1) +#endif + +/* for hashmap_from_list/1 */ +typedef struct { + Uint32 hx; + Uint32 skip; + Uint i; + Eterm val; +} hxnode_t; + + +static Eterm flatmap_merge(Process *p, Eterm nodeA, Eterm nodeB); +static Eterm map_merge_mixed(Process *p, Eterm flat, Eterm tree, int swap_args); +static Eterm hashmap_merge(Process *p, Eterm nodeA, Eterm nodeB); +static Eterm hashmap_to_list(Process *p, Eterm map); +static Eterm hashmap_keys(Process *p, Eterm map); +static Eterm hashmap_values(Process *p, Eterm map); +static Eterm hashmap_delete(Process *p, Uint32 hx, Eterm key, Eterm node); +static Eterm map_from_validated_list(Process *p, Eterm list, Uint size); +static Eterm hashmap_from_validated_list(Process *p, Eterm list, Uint size); +static Eterm hashmap_from_unsorted_array(ErtsHeapFactory*, hxnode_t *hxns, Uint n, int reject_dupkeys); +static Eterm hashmap_from_sorted_unique_array(ErtsHeapFactory*, hxnode_t *hxns, Uint n, int is_root); +static Eterm hashmap_from_chunked_array(ErtsHeapFactory*, hxnode_t *hxns, Uint n, int is_root); +static Eterm hashmap_info(Process *p, Eterm node); +static Eterm hashmap_bld_tuple_uint(Uint **hpp, Uint *szp, Uint n, Uint nums[]); +static int hxnodecmp(hxnode_t* a, hxnode_t* b); +static int hxnodecmpkey(hxnode_t* a, hxnode_t* b); + /* erlang:map_size/1 * the corresponding instruction is implemented in: * beam/erl_bif_guard.c */ BIF_RETTYPE map_size_1(BIF_ALIST_1) { - if (is_map(BIF_ARG_1)) { + if (is_flatmap(BIF_ARG_1)) { Eterm *hp; Uint hsz = 0; - map_t *mp = (map_t*)map_val(BIF_ARG_1); - Uint n = map_get_size(mp); + flatmap_t *mp = (flatmap_t*)flatmap_val(BIF_ARG_1); + Uint n = flatmap_get_size(mp); erts_bld_uint(NULL, &hsz, n); hp = HAlloc(BIF_P, hsz); BIF_RET(erts_bld_uint(&hp, NULL, n)); + } else if (is_hashmap(BIF_ARG_1)) { + Eterm *head, *hp, res; + Uint size, hsz=0; + + head = hashmap_val(BIF_ARG_1); + size = head[1]; + (void) erts_bld_uint(NULL, &hsz, size); + hp = HAlloc(BIF_P, hsz); + res = erts_bld_uint(&hp, NULL, size); + BIF_RET(res); } BIF_ERROR(BIF_P, BADARG); } -/* maps:to_list/1 - */ +/* maps:to_list/1 */ BIF_RETTYPE maps_to_list_1(BIF_ALIST_1) { - if (is_map(BIF_ARG_1)) { + if (is_flatmap(BIF_ARG_1)) { Uint n; Eterm* hp; Eterm *ks,*vs, res, tup; - map_t *mp = (map_t*)map_val(BIF_ARG_1); + flatmap_t *mp = (flatmap_t*)flatmap_val(BIF_ARG_1); - ks = map_get_keys(mp); - vs = map_get_values(mp); - n = map_get_size(mp); + ks = flatmap_get_keys(mp); + vs = flatmap_get_values(mp); + n = flatmap_get_size(mp); hp = HAlloc(BIF_P, (2 + 3) * n); res = NIL; @@ -104,6 +146,8 @@ BIF_RETTYPE maps_to_list_1(BIF_ALIST_1) { } BIF_RET(res); + } else if (is_hashmap(BIF_ARG_1)) { + return hashmap_to_list(BIF_P, BIF_ARG_1); } BIF_ERROR(BIF_P, BADARG); @@ -113,89 +157,84 @@ BIF_RETTYPE maps_to_list_1(BIF_ALIST_1) { * return value if key *matches* a key in the map */ -int erts_maps_find(Eterm key, Eterm map, Eterm *value) { +const Eterm * +#if HALFWORD_HEAP +erts_maps_get_rel(Eterm key, Eterm map, Eterm *map_base) +#else +erts_maps_get(Eterm key, Eterm map) +#endif +{ + Uint32 hx; + if (is_flatmap_rel(map, map_base)) { + Eterm *ks, *vs; + flatmap_t *mp; + Uint n, i; - Eterm *ks,*vs; - map_t *mp; - Uint n,i; + mp = (flatmap_t *)flatmap_val_rel(map, map_base); + n = flatmap_get_size(mp); - mp = (map_t*)map_val(map); - n = map_get_size(mp); - ks = map_get_keys(mp); - vs = map_get_values(mp); + if (n == 0) { + return NULL; + } - for( i = 0; i < n; i++) { - if (EQ(ks[i], key)) { - *value = vs[i]; - return 1; + ks = (Eterm *)tuple_val_rel(mp->keys, map_base) + 1; + vs = flatmap_get_values(mp); + + if (is_immed(key)) { + for (i = 0; i < n; i++) { + if (ks[i] == key) { + return &vs[i]; + } + } + } + + for (i = 0; i < n; i++) { + if (eq_rel(ks[i], map_base, key, NULL)) { + return &vs[i]; + } } + return NULL; } - return 0; + ASSERT(is_hashmap_rel(map, map_base)); + hx = hashmap_make_hash(key); + + return erts_hashmap_get_rel(hx, key, map, map_base); } BIF_RETTYPE maps_find_2(BIF_ALIST_2) { if (is_map(BIF_ARG_2)) { - Eterm *hp, value,res; + Eterm *hp, res; + const Eterm *value; - if (erts_maps_find(BIF_ARG_1, BIF_ARG_2, &value)) { + value = erts_maps_get(BIF_ARG_1, BIF_ARG_2); + if (value) { hp = HAlloc(BIF_P, 3); res = make_tuple(hp); *hp++ = make_arityval(2); *hp++ = am_ok; - *hp++ = value; + *hp++ = *value; BIF_RET(res); } - BIF_RET(am_error); } BIF_ERROR(BIF_P, BADARG); } + /* maps:get/2 * return value if key *matches* a key in the map * exception bad_key if none matches */ - -int erts_maps_get(Eterm key, Eterm map, Eterm *value) { - Eterm *ks,*vs; - map_t *mp; - Uint n,i; - - mp = (map_t*)map_val(map); - n = map_get_size(mp); - - if (n == 0) - return 0; - - ks = map_get_keys(mp); - vs = map_get_values(mp); - - if (is_immed(key)) { - for( i = 0; i < n; i++) { - if (ks[i] == key) { - *value = vs[i]; - return 1; - } - } - } - - for( i = 0; i < n; i++) { - if (EQ(ks[i], key)) { - *value = vs[i]; - return 1; - } - } - return 0; -} - BIF_RETTYPE maps_get_2(BIF_ALIST_2) { if (is_map(BIF_ARG_2)) { Eterm *hp; - Eterm value, error; + Eterm error; + const Eterm *value; char *s_error; - if (erts_maps_get(BIF_ARG_1, BIF_ARG_2, &value)) { - BIF_RET(value); + value = erts_maps_get(BIF_ARG_1, BIF_ARG_2); + if (value) { + BIF_RET(*value); } s_error = "bad_key"; @@ -213,13 +252,8 @@ BIF_RETTYPE maps_get_2(BIF_ALIST_2) { */ BIF_RETTYPE maps_from_list_1(BIF_ALIST_1) { - Eterm *kv, item = BIF_ARG_1; - Eterm *hp, *thp,*vs, *ks, keys, res; - map_t *mp; - Uint size = 0, unused_size = 0; - Sint c = 0; - Sint idx = 0; - + Eterm item = BIF_ARG_1, res, *kv; + Uint size = 0; if (is_list(item) || is_nil(item)) { /* Calculate size and check validity */ @@ -240,368 +274,1091 @@ BIF_RETTYPE maps_from_list_1(BIF_ALIST_1) { if (is_not_nil(item)) goto error; - hp = HAlloc(BIF_P, 3 + 1 + (2 * size)); - thp = hp; + if (size > MAP_SMALL_MAP_LIMIT) { + BIF_RET(hashmap_from_validated_list(BIF_P, BIF_ARG_1, size)); + } else { + BIF_RET(map_from_validated_list(BIF_P, BIF_ARG_1, size)); + } + } + +error: + + BIF_ERROR(BIF_P, BADARG); +} + +static Eterm map_from_validated_list(Process *p, Eterm list, Uint size) { + Eterm *kv, item = list; + Eterm *hp, *thp,*vs, *ks, keys, res; + flatmap_t *mp; + Uint unused_size = 0; + Sint c = 0; + Sint idx = 0; + + + hp = HAlloc(p, 3 + 1 + (2 * size)); + thp = hp; + keys = make_tuple(hp); + *hp++ = make_arityval(size); + ks = hp; + hp += size; + mp = (flatmap_t*)hp; + res = make_flatmap(mp); + hp += MAP_HEADER_SIZE; + vs = hp; + + mp->thing_word = MAP_HEADER; + mp->size = size; /* set later, might shrink*/ + mp->keys = keys; + + if (size == 0) + return res; + + /* first entry */ + kv = tuple_val(CAR(list_val(item))); + ks[0] = kv[1]; + vs[0] = kv[2]; + size = 1; + item = CDR(list_val(item)); + + /* insert sort key/value pairs */ + while(is_list(item)) { + + kv = tuple_val(CAR(list_val(item))); + + /* compare ks backwards + * idx represent word index to be written (hole position). + * We cannot copy the elements when searching since we might + * have an equal key. So we search for just the index first =( + * + * It is perhaps faster to move the values in the first pass. + * Check for uniqueness during insert phase and then have a + * second phace compacting the map if duplicates are found + * during insert. .. or do someother sort .. shell-sort perhaps. + */ + + idx = size; + + while(idx > 0 && (c = CMP_TERM(kv[1],ks[idx-1])) < 0) { idx--; } + + if (c == 0) { + /* last compare was equal, + * i.e. we have to release memory + * and overwrite that key/value + */ + ks[idx-1] = kv[1]; + vs[idx-1] = kv[2]; + unused_size++; + } else { + Uint i = size; + while(i > idx) { + ks[i] = ks[i-1]; + vs[i] = vs[i-1]; + i--; + } + ks[idx] = kv[1]; + vs[idx] = kv[2]; + size++; + } + item = CDR(list_val(item)); + } + + if (unused_size) { + /* the key tuple is embedded in the heap + * write a bignum to clear it. + */ + /* release values as normal since they are on the top of the heap */ + + ks[size] = make_pos_bignum_header(unused_size - 1); + HRelease(p, vs + size + unused_size, vs + size); + } + + *thp = make_arityval(size); + mp->size = size; + return res; +} + +#define swizzle32(D,S) \ + do { \ + (D) = ((S) & 0x0000000f) << 28 | ((S) & 0x000000f0) << 20 \ + | ((S) & 0x00000f00) << 12 | ((S) & 0x0000f000) << 4 \ + | ((S) & 0x000f0000) >> 4 | ((S) & 0x00f00000) >> 12 \ + | ((S) & 0x0f000000) >> 20 | ((S) & 0xf0000000) >> 28; \ + } while(0) + +#define maskval(V,L) (((V) >> ((7 - (L))*4)) & 0xf) +#define cdepth(V1,V2) (hashmap_clz((V1) ^ (V2)) >> 2) + +static Eterm hashmap_from_validated_list(Process *p, Eterm list, Uint size) { + Eterm item = list; + Eterm *hp; + Eterm *kv, res; + Eterm tmp[2]; + Uint32 sw, hx; + Uint ix = 0; + hxnode_t *hxns; + ErtsHeapFactory factory; + + ASSERT(size > 0); + + hp = HAlloc(p, (2 * size)); + + /* create tmp hx values and leaf ptrs */ + hxns = (hxnode_t *)erts_alloc(ERTS_ALC_T_TMP, size * sizeof(hxnode_t)); + + while(is_list(item)) { + res = CAR(list_val(item)); + kv = tuple_val(res); + hx = hashmap_restore_hash(tmp,0,kv[1]); + swizzle32(sw,hx); + hxns[ix].hx = sw; + hxns[ix].val = CONS(hp, kv[1], kv[2]); hp += 2; + hxns[ix].skip = 1; /* will be reassigned in from_array */ + hxns[ix].i = ix; + ix++; + item = CDR(list_val(item)); + } + + factory.p = p; + res = hashmap_from_unsorted_array(&factory, hxns, size, 0); + + erts_free(ERTS_ALC_T_TMP, (void *) hxns); + ERTS_VERIFY_UNUSED_TEMP_ALLOC(p); + + if (hashmap_size(res) <= MAP_SMALL_MAP_LIMIT) { + DECLARE_WSTACK(wstack); + Eterm *kv, *ks, *vs; + flatmap_t *mp; + Eterm keys; + Uint n = hashmap_size(res); + + /* build flat structure */ + hp = HAlloc(p, 3 + 1 + (2 * n)); keys = make_tuple(hp); - *hp++ = make_arityval(size); + *hp++ = make_arityval(n); ks = hp; - hp += size; - mp = (map_t*)hp; - res = make_map(mp); + hp += n; + mp = (flatmap_t*)hp; hp += MAP_HEADER_SIZE; vs = hp; mp->thing_word = MAP_HEADER; - mp->size = size; /* set later, might shrink*/ + mp->size = n; mp->keys = keys; - if (size == 0) - BIF_RET(res); + hashmap_iterator_init(&wstack, res, 0); - item = BIF_ARG_1; + while ((kv=hashmap_iterator_next(&wstack)) != NULL) { + *ks++ = CAR(kv); + *vs++ = CDR(kv); + } - /* first entry */ - kv = tuple_val(CAR(list_val(item))); - ks[0] = kv[1]; - vs[0] = kv[2]; - size = 1; - item = CDR(list_val(item)); + /* it cannot have multiple keys */ + erts_validate_and_sort_flatmap(mp); - /* insert sort key/value pairs */ - while(is_list(item)) { + DESTROY_WSTACK(wstack); + return make_flatmap(mp); + } - kv = tuple_val(CAR(list_val(item))); - - /* compare ks backwards - * idx represent word index to be written (hole position). - * We cannot copy the elements when searching since we might - * have an equal key. So we search for just the index first =( - * - * It is perhaps faster to move the values in the first pass. - * Check for uniqueness during insert phase and then have a - * second phace compacting the map if duplicates are found - * during insert. .. or do someother sort .. shell-sort perhaps. - */ + return res; +} - idx = size; +Eterm erts_hashmap_from_array(ErtsHeapFactory* factory, Eterm *leafs, Uint n, + int reject_dupkeys) { + Uint32 sw, hx; + Uint ix; + hxnode_t *hxns; + Eterm res; + + /* create tmp hx values and leaf ptrs */ + hxns = (hxnode_t *)erts_alloc(ERTS_ALC_T_TMP, n * sizeof(hxnode_t)); + + for (ix = 0; ix < n; ix++) { + hx = hashmap_make_hash(*leafs); + swizzle32(sw,hx); + hxns[ix].hx = sw; + hxns[ix].val = make_list(leafs); + hxns[ix].skip = 1; + hxns[ix].i = ix; + leafs += 2; + } - while(idx > 0 && (c = CMP_TERM(kv[1],ks[idx-1])) < 0) { idx--; } + res = hashmap_from_unsorted_array(factory, hxns, n, reject_dupkeys); - if (c == 0) { - /* last compare was equal, - * i.e. we have to release memory - * and overwrite that key/value - */ - ks[idx-1] = kv[1]; - vs[idx-1] = kv[2]; - unused_size++; - } else { - Uint i = size; - while(i > idx) { - ks[i] = ks[i-1]; - vs[i] = vs[i-1]; - i--; + erts_free(ERTS_ALC_T_TMP, (void *) hxns); + + return res; +} + + +Eterm erts_hashmap_from_ks_and_vs_extra(Process *p, Eterm *ks, Eterm *vs, Uint n, + Eterm key, Eterm value) { + Uint32 sw, hx; + Uint i,sz; + hxnode_t *hxns; + ErtsHeapFactory factory; + Eterm *hp, res; + + sz = (key == THE_NON_VALUE) ? n : (n + 1); + ASSERT(sz > MAP_SMALL_MAP_LIMIT); + hp = HAlloc(p, 2 * sz); + + /* create tmp hx values and leaf ptrs */ + hxns = (hxnode_t *)erts_alloc(ERTS_ALC_T_TMP, sz * sizeof(hxnode_t)); + + for(i = 0; i < n; i++) { + hx = hashmap_make_hash(ks[i]); + swizzle32(sw,hx); + hxns[i].hx = sw; + hxns[i].val = CONS(hp, ks[i], vs[i]); hp += 2; + hxns[i].skip = 1; /* will be reassigned in from_array */ + hxns[i].i = i; + } + + if (key != THE_NON_VALUE) { + hx = hashmap_make_hash(key); + swizzle32(sw,hx); + hxns[i].hx = sw; + hxns[i].val = CONS(hp, key, value); hp += 2; + hxns[i].skip = 1; + hxns[i].i = i; + } + + factory.p = p; + res = hashmap_from_unsorted_array(&factory, hxns, sz, 0); + + erts_free(ERTS_ALC_T_TMP, (void *) hxns); + ERTS_VERIFY_UNUSED_TEMP_ALLOC(p); + + return res; +} + +static Eterm hashmap_from_unsorted_array(ErtsHeapFactory* factory, + hxnode_t *hxns, Uint n, + int reject_dupkeys) { + Uint jx = 0, ix = 0, lx, cx; + Eterm res; + + if (n == 0) { + Eterm *hp; + hp = erts_produce_heap(factory, 2, 0); + hp[0] = MAP_HEADER_HAMT_HEAD_BITMAP(0); + hp[1] = 0; + + return make_hashmap(hp); + } + + /* sort and compact array (remove non-unique entries) */ + qsort(hxns, n, sizeof(hxnode_t), (int (*)(const void *, const void *)) hxnodecmp); + + ix = 0, cx = 0; + while(ix < n - 1) { + if (hxns[ix].hx == hxns[ix+1].hx) { + + /* find region of equal hash values */ + jx = ix + 1; + while(jx < n && hxns[ix].hx == hxns[jx].hx) { jx++; } + /* find all correct keys from region + * (last in list but now hash sorted so we check highest id instead) */ + + /* resort with keys instead of hash value within region */ + + qsort(&hxns[ix], jx - ix, sizeof(hxnode_t), + (int (*)(const void *, const void *)) hxnodecmpkey); + + while(ix < jx) { + lx = ix; + while(ix < jx && EQ(CAR(list_val(hxns[ix].val)), CAR(list_val(hxns[lx].val)))) { + if (reject_dupkeys) + return THE_NON_VALUE; + + if (hxns[ix].i > hxns[lx].i) { + lx = ix; + } + ix++; } - ks[idx] = kv[1]; - vs[idx] = kv[2]; - size++; + hxns[cx].hx = hxns[lx].hx; + hxns[cx].val = hxns[lx].val; + cx++; } - item = CDR(list_val(item)); + ix = jx; + continue; + } + if (ix > cx) { + hxns[cx].hx = hxns[ix].hx; + hxns[cx].val = hxns[ix].val; } + cx++; + ix++; + } - if (unused_size) { - /* the key tuple is embedded in the heap - * write a bignum to clear it. - */ - /* release values as normal since they are on the top of the heap */ + if (ix < n) { + hxns[cx].hx = hxns[ix].hx; + hxns[cx].val = hxns[ix].val; + cx++; + } - ks[size] = make_pos_bignum_header(unused_size - 1); - HRelease(BIF_P, vs + size + unused_size, vs + size); - } + if (cx > 1) { + /* recursive decompose array */ + res = hashmap_from_sorted_unique_array(factory, hxns, cx, 0); + } else { + Eterm *hp; - *thp = make_arityval(size); - mp->size = size; - BIF_RET(res); + /* we only have one item, either because n was 1 or + * because we hade multiples of the same key. + * + * hash value has been swizzled, need to drag it down to get the + * correct slot. */ + + hp = erts_produce_heap(factory, HAMT_HEAD_BITMAP_SZ(1), 0); + hp[0] = MAP_HEADER_HAMT_HEAD_BITMAP(1 << ((hxns[0].hx >> 0x1c) & 0xf)); + hp[1] = 1; + hp[2] = hxns[0].val; + res = make_hashmap(hp); } -error: + return res; +} - BIF_ERROR(BIF_P, BADARG); +static Eterm hashmap_from_sorted_unique_array(ErtsHeapFactory* factory, + hxnode_t *hxns, Uint n, int lvl) { + Eterm res = NIL; + Uint i,ix,jx,elems; + Uint32 sw, hx; + Eterm val; + Eterm th[2]; + hxnode_t *tmp; + + ASSERT(lvl < 32); + ix = 0; + elems = 1; + while (ix < n - 1) { + if (hxns[ix].hx == hxns[ix+1].hx) { + jx = ix + 1; + while (jx < n && hxns[ix].hx == hxns[jx].hx) { jx++; } + tmp = (hxnode_t *)erts_alloc(ERTS_ALC_T_TMP, ((jx - ix)) * sizeof(hxnode_t)); + + for(i = 0; i < jx - ix; i++) { + val = hxns[i + ix].val; + hx = hashmap_restore_hash(th, lvl + 8, CAR(list_val(val))); + swizzle32(sw,hx); + tmp[i].hx = sw; + tmp[i].val = val; + tmp[i].i = i; + tmp[i].skip = 1; + } + + qsort(tmp, jx - ix, sizeof(hxnode_t), (int (*)(const void *, const void *)) hxnodecmp); + + hxns[ix].skip = jx - ix; + hxns[ix].val = hashmap_from_sorted_unique_array(factory, tmp, jx - ix, lvl + 8); + erts_free(ERTS_ALC_T_TMP, (void *) tmp); + ix = jx; + if (ix < n) { elems++; } + continue; + } + hxns[ix].skip = 1; + elems++; + ix++; + } + + res = hashmap_from_chunked_array(factory, hxns, elems, !lvl); + + ERTS_FACTORY_HOLE_CHECK(factory); + + return res; } -/* maps:is_key/2 - */ +#define HALLOC_EXTRA 200 +static Eterm hashmap_from_chunked_array(ErtsHeapFactory *factory, + hxnode_t *hxns, Uint n, int is_root) { + Uint ix, d, dn, dc, slot, elems; + Uint32 v, vp, vn, hdr; + Uint bp, sz; + DECLARE_ESTACK(stack); + Eterm res = NIL, *hp = NULL, *nhp; -BIF_RETTYPE maps_is_key_2(BIF_ALIST_2) { - if (is_map(BIF_ARG_2)) { - Eterm *ks, key; - map_t *mp; - Uint n,i; + ASSERT(n > 1); - mp = (map_t*)map_val(BIF_ARG_2); - key = BIF_ARG_1; - n = map_get_size(mp); - ks = map_get_keys(mp); + /* push initial nodes on the stack, + * this is the starting depth */ - if (n == 0) - BIF_RET(am_false); + ix = 0; + d = 0; + vp = hxns[ix].hx; + v = hxns[ix + hxns[ix].skip].hx; - if (is_immed(key)) { - for( i = 0; i < n; i++) { - if (ks[i] == key) { - BIF_RET(am_true); - } + ASSERT(vp > v); + slot = maskval(vp,d); + + while(slot == maskval(v,d)) { + ESTACK_PUSH(stack, 1 << slot); + d++; + slot = maskval(vp,d); + } + + res = hxns[ix].val; + + if (hxns[ix].skip > 1) { + dc = 7; + /* build collision nodes */ + while (dc > d) { + hp = erts_produce_heap(factory, HAMT_NODE_BITMAP_SZ(1), HALLOC_EXTRA); + hp[0] = MAP_HEADER_HAMT_NODE_BITMAP(1 << maskval(vp,dc)); + hp[1] = res; + res = make_hashmap(hp); + dc--; + } + } + + ESTACK_PUSH(stack, res); + ESTACK_PUSH(stack, 1 << slot); + + /* all of the other nodes .. */ + elems = n - 2; /* remove first and last elements */ + while(elems--) { + hdr = ESTACK_POP(stack); + ix = ix + hxns[ix].skip; + + /* determine if node or subtree should be built by looking + * at the next value. */ + + vn = hxns[ix + hxns[ix].skip].hx; + dn = cdepth(v,vn); + ASSERT(v > vn); + + res = hxns[ix].val; + + if (hxns[ix].skip > 1) { + int wat = (d > dn) ? d : dn; + dc = 7; + /* build collision nodes */ + while (dc > wat) { + hp = erts_produce_heap(factory, HAMT_NODE_BITMAP_SZ(1), HALLOC_EXTRA); + hp[0] = MAP_HEADER_HAMT_NODE_BITMAP(1 << maskval(v,dc)); + hp[1] = res; + res = make_hashmap(hp); + dc--; } } - for( i = 0; i < n; i++) { - if (EQ(ks[i], key)) { - BIF_RET(am_true); + /* next depth is higher (implies collision) */ + if (d < dn) { + /* hdr is the popped one initially */ + while(d < dn) { + slot = maskval(v, d); + bp = 1 << slot; + ESTACK_PUSH(stack, hdr | bp); + d++; + hdr = 0; /* clear hdr for all other collisions */ } + + slot = maskval(v, d); + bp = 1 << slot; + /* no more collisions */ + ESTACK_PUSH(stack,res); + ESTACK_PUSH(stack,bp); + } else if (d == dn) { + /* no collisions at all */ + slot = maskval(v, d); + bp = 1 << slot; + ESTACK_PUSH(stack,res); + ESTACK_PUSH(stack,hdr | bp); + } else { + /* dn < n, we have a drop and we are done + * build nodes and subtree */ + while (dn != d) { + slot = maskval(v, d); + bp = 1 << slot; + /* OR bitposition before sz calculation to handle + * redundant collisions */ + hdr |= bp; + sz = hashmap_bitcount(hdr); + hp = erts_produce_heap(factory, HAMT_NODE_BITMAP_SZ(sz), HALLOC_EXTRA); + nhp = hp; + *hp++ = (hdr == 0xffff) ? MAP_HEADER_HAMT_NODE_ARRAY : MAP_HEADER_HAMT_NODE_BITMAP(hdr); + *hp++ = res; sz--; + while (sz--) { *hp++ = ESTACK_POP(stack); } + ASSERT((hp - nhp) < 18); + res = make_hashmap(nhp); + + /* we need to pop the next hdr and push if we don't need it */ + + hdr = ESTACK_POP(stack); + d--; + } + ESTACK_PUSH(stack, res); + ESTACK_PUSH(stack, hdr); } - BIF_RET(am_false); + + vp = v; + v = vn; + d = dn; + ERTS_FACTORY_HOLE_CHECK(factory); + } + + /* v and vp are reused from above */ + dn = cdepth(vp,v); + ix = ix + hxns[ix].skip; + res = hxns[ix].val; + + if (hxns[ix].skip > 1) { + dc = 7; + /* build collision nodes */ + while (dc > dn) { + hp = erts_produce_heap(factory, HAMT_NODE_BITMAP_SZ(1), HALLOC_EXTRA); + hp[0] = MAP_HEADER_HAMT_NODE_BITMAP(1 << maskval(v,dc)); + hp[1] = res; + res = make_hashmap(hp); + dc--; + } + } + + hdr = ESTACK_POP(stack); + /* pop remaining subtree if any */ + while (dn) { + slot = maskval(v, dn); + bp = 1 << slot; + /* OR bitposition before sz calculation to handle + * redundant collisions */ + hdr |= bp; + sz = hashmap_bitcount(hdr); + hp = erts_produce_heap(factory, HAMT_NODE_BITMAP_SZ(sz), HALLOC_EXTRA); + nhp = hp; + *hp++ = (hdr == 0xffff) ? MAP_HEADER_HAMT_NODE_ARRAY : MAP_HEADER_HAMT_NODE_BITMAP(hdr); + *hp++ = res; sz--; + + while (sz--) { *hp++ = ESTACK_POP(stack); } + res = make_hashmap(nhp); + hdr = ESTACK_POP(stack); + dn--; + } + + /* and finally the root .. */ + + slot = maskval(v, dn); + bp = 1 << slot; + hdr |= bp; + sz = hashmap_bitcount(hdr); + hp = erts_produce_heap(factory, sz + /* hdr + item */ (is_root ? 2 : 1), 0); + nhp = hp; + + if (is_root) { + *hp++ = (hdr == 0xffff) ? MAP_HEADER_HAMT_HEAD_ARRAY : MAP_HEADER_HAMT_HEAD_BITMAP(hdr); + *hp++ = n; + } else { + *hp++ = (hdr == 0xffff) ? MAP_HEADER_HAMT_NODE_ARRAY : MAP_HEADER_HAMT_NODE_BITMAP(hdr); + } + + *hp++ = res; sz--; + while (sz--) { *hp++ = ESTACK_POP(stack); } + + res = make_hashmap(nhp); + + ASSERT(ESTACK_COUNT(stack) == 0); + DESTROY_ESTACK(stack); + ERTS_FACTORY_HOLE_CHECK(factory); + return res; +} +#undef HALLOC_EXTRA + +static int hxnodecmpkey(hxnode_t *a, hxnode_t *b) { + return CMP_TERM(CAR(list_val(a->val)), CAR(list_val(b->val))); +} + +static int hxnodecmp(hxnode_t *a, hxnode_t *b) { + if (a->hx < b->hx) + return 1; + else if (a->hx == b->hx) + return 0; + else + return -1; +} + +/* maps:is_key/2 */ + +BIF_RETTYPE maps_is_key_2(BIF_ALIST_2) { + if (is_map(BIF_ARG_2)) { + BIF_RET(erts_maps_get(BIF_ARG_1, BIF_ARG_2) ? am_true : am_false); } BIF_ERROR(BIF_P, BADARG); } -/* maps:keys/1 - */ +/* maps:keys/1 */ BIF_RETTYPE maps_keys_1(BIF_ALIST_1) { - if (is_map(BIF_ARG_1)) { + if (is_flatmap(BIF_ARG_1)) { Eterm *hp, *ks, res = NIL; - map_t *mp; + flatmap_t *mp; Uint n; - mp = (map_t*)map_val(BIF_ARG_1); - n = map_get_size(mp); + mp = (flatmap_t*)flatmap_val(BIF_ARG_1); + n = flatmap_get_size(mp); if (n == 0) BIF_RET(res); hp = HAlloc(BIF_P, (2 * n)); - ks = map_get_keys(mp); + ks = flatmap_get_keys(mp); while(n--) { res = CONS(hp, ks[n], res); hp += 2; } BIF_RET(res); + } else if (is_hashmap(BIF_ARG_1)) { + BIF_RET(hashmap_keys(BIF_P, BIF_ARG_1)); } BIF_ERROR(BIF_P, BADARG); } -/* maps:merge/2 - */ +/* maps:merge/2 */ BIF_RETTYPE maps_merge_2(BIF_ALIST_2) { - if (is_map(BIF_ARG_1) && is_map(BIF_ARG_2)) { - Eterm *hp,*thp; - Eterm tup; - Eterm *ks,*vs,*ks1,*vs1,*ks2,*vs2; - map_t *mp1,*mp2,*mp_new; - Uint n1,n2,i1,i2,need,unused_size=0; - int c = 0; - - mp1 = (map_t*)map_val(BIF_ARG_1); - mp2 = (map_t*)map_val(BIF_ARG_2); - n1 = map_get_size(mp1); - n2 = map_get_size(mp2); - - need = MAP_HEADER_SIZE + 1 + 2*(n1 + n2); - - hp = HAlloc(BIF_P, need); - thp = hp; - tup = make_tuple(thp); - ks = hp + 1; hp += 1 + n1 + n2; - mp_new = (map_t*)hp; hp += MAP_HEADER_SIZE; - vs = hp; hp += n1 + n2; - - mp_new->thing_word = MAP_HEADER; - mp_new->size = 0; - mp_new->keys = tup; - - i1 = 0; i2 = 0; - ks1 = map_get_keys(mp1); - vs1 = map_get_values(mp1); - ks2 = map_get_keys(mp2); - vs2 = map_get_values(mp2); - - while(i1 < n1 && i2 < n2) { - c = CMP_TERM(ks1[i1],ks2[i2]); - if ( c == 0) { - /* use righthand side arguments map value, - * but advance both maps */ - *ks++ = ks2[i2]; - *vs++ = vs2[i2]; - i1++, i2++, unused_size++; - } else if ( c < 0) { - *ks++ = ks1[i1]; - *vs++ = vs1[i1]; - i1++; - } else { - *ks++ = ks2[i2]; - *vs++ = vs2[i2]; - i2++; - } + if (is_flatmap(BIF_ARG_1)) { + if (is_flatmap(BIF_ARG_2)) { + BIF_RET(flatmap_merge(BIF_P, BIF_ARG_1, BIF_ARG_2)); + } else if (is_hashmap(BIF_ARG_2)) { + /* Will always become a tree */ + BIF_RET(map_merge_mixed(BIF_P, BIF_ARG_1, BIF_ARG_2, 0)); + } + } else if (is_hashmap(BIF_ARG_1)) { + if (is_hashmap(BIF_ARG_2)) { + BIF_RET(hashmap_merge(BIF_P, BIF_ARG_1, BIF_ARG_2)); + } else if (is_flatmap(BIF_ARG_2)) { + /* Will always become a tree */ + BIF_RET(map_merge_mixed(BIF_P, BIF_ARG_2, BIF_ARG_1, 1)); } + } + BIF_ERROR(BIF_P, BADARG); +} + +static Eterm flatmap_merge(Process *p, Eterm nodeA, Eterm nodeB) { + Eterm *hp,*thp; + Eterm tup; + Eterm *ks,*vs,*ks1,*vs1,*ks2,*vs2; + flatmap_t *mp1,*mp2,*mp_new; + Uint n,n1,n2,i1,i2,need,unused_size=0; + int c = 0; + + mp1 = (flatmap_t*)flatmap_val(nodeA); + mp2 = (flatmap_t*)flatmap_val(nodeB); + n1 = flatmap_get_size(mp1); + n2 = flatmap_get_size(mp2); + + need = MAP_HEADER_SIZE + 1 + 2*(n1 + n2); - /* copy remaining */ - while (i1 < n1) { + hp = HAlloc(p, need); + thp = hp; + tup = make_tuple(thp); + ks = hp + 1; hp += 1 + n1 + n2; + mp_new = (flatmap_t*)hp; hp += MAP_HEADER_SIZE; + vs = hp; hp += n1 + n2; + + mp_new->thing_word = MAP_HEADER; + mp_new->size = 0; + mp_new->keys = tup; + + i1 = 0; i2 = 0; + ks1 = flatmap_get_keys(mp1); + vs1 = flatmap_get_values(mp1); + ks2 = flatmap_get_keys(mp2); + vs2 = flatmap_get_values(mp2); + + while(i1 < n1 && i2 < n2) { + c = CMP_TERM(ks1[i1],ks2[i2]); + if (c == 0) { + /* use righthand side arguments map value, + * but advance both maps */ + *ks++ = ks2[i2]; + *vs++ = vs2[i2]; + i1++, i2++, unused_size++; + } else if (c < 0) { *ks++ = ks1[i1]; *vs++ = vs1[i1]; i1++; - } - - while (i2 < n2) { + } else { *ks++ = ks2[i2]; *vs++ = vs2[i2]; i2++; } + } - if (unused_size) { - /* the key tuple is embedded in the heap, write a bignum to clear it. - * - * release values as normal since they are on the top of the heap - * size = n1 + n1 - unused_size - */ + /* copy remaining */ + while (i1 < n1) { + *ks++ = ks1[i1]; + *vs++ = vs1[i1]; + i1++; + } - *ks = make_pos_bignum_header(unused_size - 1); - HRelease(BIF_P, vs + unused_size, vs); - } + while (i2 < n2) { + *ks++ = ks2[i2]; + *vs++ = vs2[i2]; + i2++; + } - mp_new->size = n1 + n2 - unused_size; - *thp = make_arityval(n1 + n2 - unused_size); + if (unused_size) { + /* the key tuple is embedded in the heap, write a bignum to clear it. + * + * release values as normal since they are on the top of the heap + * size = n1 + n1 - unused_size + */ - BIF_RET(make_map(mp_new)); + *ks = make_pos_bignum_header(unused_size - 1); + HRelease(p, vs + unused_size, vs); } - BIF_ERROR(BIF_P, BADARG); -} -/* maps:new/2 - */ -BIF_RETTYPE maps_new_0(BIF_ALIST_0) { - Eterm* hp; - Eterm tup; - map_t *mp; + n = n1 + n2 - unused_size; + *thp = make_arityval(n); - hp = HAlloc(BIF_P, (MAP_HEADER_SIZE + 1)); - tup = make_tuple(hp); - *hp++ = make_arityval(0); + /* Reshape map to a hashmap if the map exceeds the limit */ - mp = (map_t*)hp; - mp->thing_word = MAP_HEADER; - mp->size = 0; - mp->keys = tup; + if (n > MAP_SMALL_MAP_LIMIT) { + Uint32 hx,sw; + Uint i; + Eterm res; + hxnode_t *hxns; + ErtsHeapFactory factory; - BIF_RET(make_map(mp)); -} + ks = flatmap_get_keys(mp_new); + vs = flatmap_get_values(mp_new); -/* maps:put/3 - */ + hp = HAlloc(p, 2 * n); -Eterm erts_maps_put(Process *p, Eterm key, Eterm value, Eterm map) { - Sint n,i; - Sint c = 0; - Eterm* hp, *shp; - Eterm *ks,*vs, res, tup; - map_t *mp = (map_t*)map_val(map); + hxns = (hxnode_t *)erts_alloc(ERTS_ALC_T_TMP,n * sizeof(hxnode_t)); - n = map_get_size(mp); + for (i = 0; i < n; i++) { + hx = hashmap_make_hash(ks[i]); + swizzle32(sw,hx); + hxns[i].hx = sw; + hxns[i].val = CONS(hp, ks[i], vs[i]); hp += 2; + hxns[i].skip = 1; + hxns[i].i = i; + } - if (n == 0) { - hp = HAlloc(p, MAP_HEADER_SIZE + 1 + 2); - tup = make_tuple(hp); - *hp++ = make_arityval(1); - *hp++ = key; - res = make_map(hp); - *hp++ = MAP_HEADER; - *hp++ = 1; - *hp++ = tup; - *hp++ = value; + factory.p = p; + res = hashmap_from_unsorted_array(&factory, hxns, n, 0); + + erts_free(ERTS_ALC_T_TMP, (void *) hxns); + ERTS_VERIFY_UNUSED_TEMP_ALLOC(p); return res; } - ks = map_get_keys(mp); - vs = map_get_values(mp); + mp_new->size = n; + + return make_flatmap(mp_new); +} + +static Eterm map_merge_mixed(Process *p, Eterm flat, Eterm tree, int swap_args) { + Eterm *ks, *vs, *hp, res; + flatmap_t *mp; + Uint n, i; + hxnode_t *hxns; + Uint32 sw, hx; + ErtsHeapFactory factory; + + /* convert flat to tree */ + + ASSERT(is_flatmap(flat)); + ASSERT(is_hashmap(tree)); + + mp = (flatmap_t*)flatmap_val(flat); + n = flatmap_get_size(mp); + + ks = flatmap_get_keys(mp); + vs = flatmap_get_values(mp); - /* only allocate for values, - * assume key-tuple will be intact + hp = HAlloc(p, 2 * n); + + hxns = (hxnode_t *)erts_alloc(ERTS_ALC_T_TMP, n * sizeof(hxnode_t)); + + for (i = 0; i < n; i++) { + hx = hashmap_make_hash(ks[i]); + swizzle32(sw,hx); + hxns[i].hx = sw; + hxns[i].val = CONS(hp, ks[i], vs[i]); hp += 2; + hxns[i].skip = 1; + hxns[i].i = i; + } + + factory.p = p; + res = hashmap_from_unsorted_array(&factory, hxns, n, 0); + + erts_free(ERTS_ALC_T_TMP, (void *) hxns); + ERTS_VERIFY_UNUSED_TEMP_ALLOC(p); + + return swap_args ? hashmap_merge(p, tree, res) : hashmap_merge(p, res, tree); +} + +#define HALLOC_EXTRA 200 + +static Eterm hashmap_merge(Process *p, Eterm nodeA, Eterm nodeB) { +#define PSTACK_TYPE struct HashmapMergePStackType + struct HashmapMergePStackType { + Eterm *srcA, *srcB; + Uint32 abm, bbm, rbm; /* node bitmaps */ + int keepA; + int ix; + Eterm array[16]; + }; + PSTACK_DECLARE(s, 4); + struct HashmapMergePStackType* sp = PSTACK_PUSH(s); + Eterm *hp, *nhp; + Eterm hdrA, hdrB; + Eterm th[2]; + Uint32 ahx, bhx; + Uint size; /* total key-value counter */ + int keepA = 0; + unsigned lvl = 0; + Eterm res = THE_NON_VALUE; + + /* + * Strategy: Do depth-first traversal of both trees (at the same time) + * and merge each pair of nodes. */ - hp = HAlloc(p, MAP_HEADER_SIZE + n); - shp = hp; /* save hp, used if optimistic update fails */ - res = make_map(hp); - *hp++ = MAP_HEADER; - *hp++ = n; - *hp++ = mp->keys; - - if (is_immed(key)) { - for( i = 0; i < n; i ++) { - if (ks[i] == key) { - *hp++ = value; - vs++; - c = 1; + { + hashmap_head_t* a = (hashmap_head_t*) hashmap_val(nodeA); + hashmap_head_t* b = (hashmap_head_t*) hashmap_val(nodeB); + size = a->size + b->size; + } + +recurse: + + if (primary_tag(nodeA) == TAG_PRIMARY_BOXED && + primary_tag(nodeB) == TAG_PRIMARY_LIST) { + /* Avoid implementing this combination by switching places */ + Eterm tmp = nodeA; + nodeA = nodeB; + nodeB = tmp; + keepA = !keepA; + } + + switch (primary_tag(nodeA)) { + case TAG_PRIMARY_LIST: { + sp->srcA = list_val(nodeA); + switch (primary_tag(nodeB)) { + case TAG_PRIMARY_LIST: { /* LEAF + LEAF */ + sp->srcB = list_val(nodeB); + + if (EQ(CAR(sp->srcA), CAR(sp->srcB))) { + --size; + res = keepA ? nodeA : nodeB; } else { - *hp++ = *vs++; + ahx = hashmap_restore_hash(th, lvl, CAR(sp->srcA)); + bhx = hashmap_restore_hash(th, lvl, CAR(sp->srcB)); + sp->abm = 1 << hashmap_index(ahx); + sp->bbm = 1 << hashmap_index(bhx); + + sp->srcA = &nodeA; + sp->srcB = &nodeB; } + break; } - } else { - for( i = 0; i < n; i ++) { - if (EQ(ks[i], key)) { - *hp++ = value; - vs++; - c = 1; - } else { - *hp++ = *vs++; + case TAG_PRIMARY_BOXED: { /* LEAF + NODE */ + sp->srcB = boxed_val(nodeB); + ASSERT(is_header(*sp->srcB)); + hdrB = *sp->srcB++; + + ahx = hashmap_restore_hash(th, lvl, CAR(sp->srcA)); + sp->abm = 1 << hashmap_index(ahx); + sp->srcA = &nodeA; + switch(hdrB & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_HEAD_ARRAY: sp->srcB++; + case HAMT_SUBTAG_NODE_ARRAY: + sp->bbm = 0xffff; + break; + + case HAMT_SUBTAG_HEAD_BITMAP: sp->srcB++; + case HAMT_SUBTAG_NODE_BITMAP: + sp->bbm = MAP_HEADER_VAL(hdrB); + break; + + default: + erl_exit(1, "bad header tag %ld\r\n", *sp->srcB & _HEADER_MAP_SUBTAG_MASK); + break; } + break; + } + default: + erl_exit(1, "bad primary tag %ld\r\n", nodeB); } + break; } + case TAG_PRIMARY_BOXED: { /* NODE + NODE */ + sp->srcA = boxed_val(nodeA); + hdrA = *sp->srcA++; + ASSERT(is_header(hdrA)); + switch (hdrA & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_HEAD_ARRAY: sp->srcA++; + case HAMT_SUBTAG_NODE_ARRAY: { + ASSERT(primary_tag(nodeB) == TAG_PRIMARY_BOXED); + sp->abm = 0xffff; + sp->srcB = boxed_val(nodeB); + hdrB = *sp->srcB++; + ASSERT(is_header(hdrB)); + switch (hdrB & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_HEAD_ARRAY: sp->srcB++; + case HAMT_SUBTAG_NODE_ARRAY: + sp->bbm = 0xffff; + break; + case HAMT_SUBTAG_HEAD_BITMAP: sp->srcB++; + case HAMT_SUBTAG_NODE_BITMAP: + sp->bbm = MAP_HEADER_VAL(hdrB); + break; + default: + erl_exit(1, "bad header tag %ld\r\n", *sp->srcB & _HEADER_MAP_SUBTAG_MASK); + } + break; + } + case HAMT_SUBTAG_HEAD_BITMAP: sp->srcA++; + case HAMT_SUBTAG_NODE_BITMAP: { + ASSERT(primary_tag(nodeB) == TAG_PRIMARY_BOXED); + sp->abm = MAP_HEADER_VAL(hdrA); + sp->srcB = boxed_val(nodeB); + hdrB = *sp->srcB++; + ASSERT(is_header(hdrB)); + switch (hdrB & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_HEAD_ARRAY: sp->srcB++; + case HAMT_SUBTAG_NODE_ARRAY: + sp->bbm = 0xffff; + break; + case HAMT_SUBTAG_HEAD_BITMAP: sp->srcB++; + case HAMT_SUBTAG_NODE_BITMAP: + sp->bbm = MAP_HEADER_VAL(hdrB); + break; - if (c) - return res; + default: + erl_exit(1, "bad header tag %ld\r\n", *sp->srcB & _HEADER_MAP_SUBTAG_MASK); + } + break; + } + default: + erl_exit(1, "bad primary tag %ld\r\n", nodeA); + } + break; + } + default: + erl_exit(1, "bad primary tag %ld\r\n", nodeA); + } - /* need to make a new tuple, - * use old hp since it needs to be recreated anyway. - */ - tup = make_tuple(shp); - *shp++ = make_arityval(n+1); + for (;;) { + if (is_value(res)) { /* We have a complete (sub-)tree or leaf */ + if (lvl == 0) + break; - hp = HAlloc(p, 3 + n + 1); - res = make_map(hp); - *hp++ = MAP_HEADER; - *hp++ = n + 1; - *hp++ = tup; + /* Pop from stack and continue build parent node */ + lvl--; + sp = PSTACK_POP(s); + sp->array[sp->ix++] = res; + res = THE_NON_VALUE; + if (sp->rbm) { + sp->srcA++; + sp->srcB++; + keepA = sp->keepA; + } + } else { /* Start build a node */ + sp->ix = 0; + sp->rbm = sp->abm | sp->bbm; + ASSERT(!(sp->rbm == 0 && lvl > 0)); + } - ks = map_get_keys(mp); - vs = map_get_values(mp); + while (sp->rbm) { + Uint32 next = sp->rbm & (sp->rbm-1); + Uint32 bit = sp->rbm ^ next; + sp->rbm = next; + if (sp->abm & bit) { + if (sp->bbm & bit) { + /* Bit clash. Push and resolve by recursive merge */ + if (sp->rbm) { + sp->keepA = keepA; + } + nodeA = *sp->srcA; + nodeB = *sp->srcB; + lvl++; + sp = PSTACK_PUSH(s); + goto recurse; + } else { + sp->array[sp->ix++] = *sp->srcA++; + } + } else { + ASSERT(sp->bbm & bit); + sp->array[sp->ix++] = *sp->srcB++; + } + } + + ASSERT(sp->ix == hashmap_bitcount(sp->abm | sp->bbm)); + if (lvl == 0) { + nhp = HAllocX(p, HAMT_HEAD_BITMAP_SZ(sp->ix), HALLOC_EXTRA); + hp = nhp; + *hp++ = (sp->ix == 16 ? MAP_HEADER_HAMT_HEAD_ARRAY + : MAP_HEADER_HAMT_HEAD_BITMAP(sp->abm | sp->bbm)); + *hp++ = size; + } else { + nhp = HAllocX(p, HAMT_NODE_BITMAP_SZ(sp->ix), HALLOC_EXTRA); + hp = nhp; + *hp++ = (sp->ix == 16 ? make_arityval(16) + : MAP_HEADER_HAMT_NODE_BITMAP(sp->abm | sp->bbm)); + } + memcpy(hp, sp->array, sp->ix * sizeof(Eterm)); + res = make_boxed(nhp); + } + PSTACK_DESTROY(s); + return res; +} - ASSERT(n >= 0); +static int hash_cmp(Uint32 ha, Uint32 hb) +{ + int i; + for (i=0; i<8; i++) { + int cmp = (int)(ha & 0xF) - (int)(hb & 0xF); + if (cmp) + return cmp; + ha >>= 4; + hb >>= 4; + } + return 0; +} - /* copy map in order */ - while (n && ((c = CMP_TERM(*ks, key)) < 0)) { - *shp++ = *ks++; - *hp++ = *vs++; - n--; +int hashmap_key_hash_cmp(Eterm* ap, Eterm* bp) +{ + Eterm th[2]; + unsigned lvl = 0; + + if (ap && bp) { + ASSERT(CMP_TERM(CAR(ap), CAR(bp)) != 0); + for (;;) { + Uint32 ha = hashmap_restore_hash(th, lvl, CAR(ap)); + Uint32 hb = hashmap_restore_hash(th, lvl, CAR(bp)); + int cmp = hash_cmp(ha, hb); + if (cmp) + return cmp; + lvl += 8; + } } + return ap ? -1 : 1; +} - *shp++ = key; - *hp++ = value; +/* maps:new/0 */ - ASSERT(n >= 0); +BIF_RETTYPE maps_new_0(BIF_ALIST_0) { + Eterm* hp; + Eterm tup; + flatmap_t *mp; - while(n--) { - *shp++ = *ks++; - *hp++ = *vs++; - } - /* we have one word remaining - * this will work out fine once we get the size word - * in the header. - */ - *shp = make_pos_bignum_header(0); - return res; + hp = HAlloc(BIF_P, (MAP_HEADER_SIZE + 1)); + tup = make_tuple(hp); + *hp++ = make_arityval(0); + + mp = (flatmap_t*)hp; + mp->thing_word = MAP_HEADER; + mp->size = 0; + mp->keys = tup; + + BIF_RET(make_flatmap(mp)); } +/* maps:put/3 */ + BIF_RETTYPE maps_put_3(BIF_ALIST_3) { if (is_map(BIF_ARG_3)) { BIF_RET(erts_maps_put(BIF_P, BIF_ARG_1, BIF_ARG_2, BIF_ARG_3)); @@ -609,81 +1366,87 @@ BIF_RETTYPE maps_put_3(BIF_ALIST_3) { BIF_ERROR(BIF_P, BADARG); } -/* maps:remove/3 - */ +/* maps:remove/3 */ int erts_maps_remove(Process *p, Eterm key, Eterm map, Eterm *res) { - Sint n; - Uint need; - Eterm *hp_start; - Eterm *thp, *mhp; - Eterm *ks, *vs, tup; - map_t *mp = (map_t*)map_val(map); + Uint32 hx; + if (is_flatmap(map)) { + Sint n; + Uint need; + Eterm *hp_start; + Eterm *thp, *mhp; + Eterm *ks, *vs, tup; + flatmap_t *mp = (flatmap_t*)flatmap_val(map); + + n = flatmap_get_size(mp); + + if (n == 0) { + *res = map; + return 1; + } - n = map_get_size(mp); + ks = flatmap_get_keys(mp); + vs = flatmap_get_values(mp); - if (n == 0) { - *res = map; - return 1; - } + /* Assume key exists. + * Release allocated if it didn't. + * Allocate key tuple first. + */ - ks = map_get_keys(mp); - vs = map_get_values(mp); + need = n + 1 - 1 + 3 + n - 1; /* tuple - 1 + map - 1 */ + hp_start = HAlloc(p, need); + thp = hp_start; + mhp = thp + n; /* offset with tuple heap size */ - /* Assume key exists. - * Release allocated if it didn't. - * Allocate key tuple first. - */ + tup = make_tuple(thp); + *thp++ = make_arityval(n - 1); - need = n + 1 - 1 + 3 + n - 1; /* tuple - 1 + map - 1 */ - hp_start = HAlloc(p, need); - thp = hp_start; - mhp = thp + n; /* offset with tuple heap size */ + *res = make_flatmap(mhp); + *mhp++ = MAP_HEADER; + *mhp++ = n - 1; + *mhp++ = tup; - tup = make_tuple(thp); - *thp++ = make_arityval(n - 1); - - *res = make_map(mhp); - *mhp++ = MAP_HEADER; - *mhp++ = n - 1; - *mhp++ = tup; - - if (is_immed(key)) { - while (1) { - if (*ks == key) { - goto found_key; - } else if (--n) { - *mhp++ = *vs++; - *thp++ = *ks++; - } else - break; - } - } else { - while(1) { - if (EQ(*ks, key)) { - goto found_key; - } else if (--n) { - *mhp++ = *vs++; - *thp++ = *ks++; - } else - break; + if (is_immed(key)) { + while (1) { + if (*ks == key) { + goto found_key; + } else if (--n) { + *mhp++ = *vs++; + *thp++ = *ks++; + } else + break; + } + } else { + while(1) { + if (EQ(*ks, key)) { + goto found_key; + } else if (--n) { + *mhp++ = *vs++; + *thp++ = *ks++; + } else + break; + } } - } - /* Not found, remove allocated memory - * and return previous map. - */ - HRelease(p, hp_start + need, hp_start); + /* Not found, remove allocated memory + * and return previous map. + */ + HRelease(p, hp_start + need, hp_start); - *res = map; - return 1; + *res = map; + return 1; found_key: - /* Copy rest of keys and values */ - if (--n) { - sys_memcpy(mhp, vs+1, n*sizeof(Eterm)); - sys_memcpy(thp, ks+1, n*sizeof(Eterm)); + /* Copy rest of keys and values */ + if (--n) { + sys_memcpy(mhp, vs+1, n*sizeof(Eterm)); + sys_memcpy(thp, ks+1, n*sizeof(Eterm)); + } + return 1; } + ASSERT(is_hashmap(map)); + hx = hashmap_make_hash(key); + *res = hashmap_delete(p, hx, key, map); return 1; } @@ -697,21 +1460,20 @@ BIF_RETTYPE maps_remove_2(BIF_ALIST_2) { BIF_ERROR(BIF_P, BADARG); } -/* maps:update/3 - */ - int erts_maps_update(Process *p, Eterm key, Eterm value, Eterm map, Eterm *res) { + Uint32 hx; + if (is_flatmap(map)) { Sint n,i; Eterm* hp,*shp; Eterm *ks,*vs; - map_t *mp = (map_t*)map_val(map); + flatmap_t *mp = (flatmap_t*)flatmap_val(map); - if ((n = map_get_size(mp)) == 0) { + if ((n = flatmap_get_size(mp)) == 0) { return 0; } - ks = map_get_keys(mp); - vs = map_get_values(mp); + ks = flatmap_get_keys(mp); + vs = flatmap_get_values(mp); /* only allocate for values, * assume key-tuple will be intact @@ -749,10 +1511,147 @@ found_key: vs++; if (++i < n) sys_memcpy(hp, vs, (n - i)*sizeof(Eterm)); - *res = make_map(shp); + *res = make_flatmap(shp); return 1; + } + + ASSERT(is_hashmap(map)); + hx = hashmap_make_hash(key); + *res = erts_hashmap_insert(p, hx, key, value, map, 1); + if (is_value(*res)) + return 1; + + return 0; } +Eterm erts_maps_put(Process *p, Eterm key, Eterm value, Eterm map) { + Uint32 hx; + Eterm res; + if (is_flatmap(map)) { + Sint n,i; + Sint c = 0; + Eterm* hp, *shp; + Eterm *ks, *vs, tup; + flatmap_t *mp = (flatmap_t*)flatmap_val(map); + + n = flatmap_get_size(mp); + + if (n == 0) { + hp = HAlloc(p, MAP_HEADER_SIZE + 1 + 2); + tup = make_tuple(hp); + *hp++ = make_arityval(1); + *hp++ = key; + res = make_flatmap(hp); + *hp++ = MAP_HEADER; + *hp++ = 1; + *hp++ = tup; + *hp++ = value; + + return res; + } + + ks = flatmap_get_keys(mp); + vs = flatmap_get_values(mp); + + /* only allocate for values, + * assume key-tuple will be intact + */ + + hp = HAlloc(p, MAP_HEADER_SIZE + n); + shp = hp; /* save hp, used if optimistic update fails */ + res = make_flatmap(hp); + *hp++ = MAP_HEADER; + *hp++ = n; + *hp++ = mp->keys; + + if (is_immed(key)) { + for( i = 0; i < n; i ++) { + if (ks[i] == key) { + *hp++ = value; + vs++; + c = 1; + } else { + *hp++ = *vs++; + } + } + } else { + for( i = 0; i < n; i ++) { + if (EQ(ks[i], key)) { + *hp++ = value; + vs++; + c = 1; + } else { + *hp++ = *vs++; + } + } + } + + if (c) + return res; + + /* the map will grow */ + + if (n >= MAP_SMALL_MAP_LIMIT) { + HRelease(p, shp + MAP_HEADER_SIZE + n, shp); + ks = flatmap_get_keys(mp); + vs = flatmap_get_values(mp); + + res = erts_hashmap_from_ks_and_vs_extra(p,ks,vs,n,key,value); + + return res; + } + + /* still a small map. need to make a new tuple, + * use old hp since it needs to be recreated anyway. */ + + tup = make_tuple(shp); + *shp++ = make_arityval(n+1); + + hp = HAlloc(p, 3 + n + 1); + res = make_flatmap(hp); + *hp++ = MAP_HEADER; + *hp++ = n + 1; + *hp++ = tup; + + ks = flatmap_get_keys(mp); + vs = flatmap_get_values(mp); + + ASSERT(n >= 0); + + /* copy map in order */ + while (n && ((c = CMP_TERM(*ks, key)) < 0)) { + *shp++ = *ks++; + *hp++ = *vs++; + n--; + } + + *shp++ = key; + *hp++ = value; + + ASSERT(n >= 0); + + while(n--) { + *shp++ = *ks++; + *hp++ = *vs++; + } + /* we have one word remaining + * this will work out fine once we get the size word + * in the header. + */ + *shp = make_pos_bignum_header(0); + return res; + } + ASSERT(is_hashmap(map)); + + hx = hashmap_make_hash(key); + res = erts_hashmap_insert(p, hx, key, value, map, 0); + ASSERT(is_hashmap(res)); + + return res; +} + +/* maps:update/3 */ + BIF_RETTYPE maps_update_3(BIF_ALIST_3) { if (is_map(BIF_ARG_3)) { Eterm res; @@ -764,38 +1663,845 @@ BIF_RETTYPE maps_update_3(BIF_ALIST_3) { } -/* maps:values/1 - */ +/* maps:values/1 */ BIF_RETTYPE maps_values_1(BIF_ALIST_1) { - if (is_map(BIF_ARG_1)) { + if (is_flatmap(BIF_ARG_1)) { Eterm *hp, *vs, res = NIL; - map_t *mp; + flatmap_t *mp; Uint n; - mp = (map_t*)map_val(BIF_ARG_1); - n = map_get_size(mp); + mp = (flatmap_t*)flatmap_val(BIF_ARG_1); + n = flatmap_get_size(mp); if (n == 0) BIF_RET(res); hp = HAlloc(BIF_P, (2 * n)); - vs = map_get_values(mp); + vs = flatmap_get_values(mp); while(n--) { res = CONS(hp, vs[n], res); hp += 2; } BIF_RET(res); + } else if (is_hashmap(BIF_ARG_1)) { + BIF_RET(hashmap_values(BIF_P, BIF_ARG_1)); } BIF_ERROR(BIF_P, BADARG); } -int erts_validate_and_sort_map(map_t* mp) +static Eterm hashmap_to_list(Process *p, Eterm node) { + DECLARE_WSTACK(stack); + Eterm *hp, *kv; + Eterm res = NIL; + + hp = HAlloc(p, hashmap_size(node) * (2 + 3)); + hashmap_iterator_init(&stack, node, 0); + while ((kv=hashmap_iterator_next(&stack)) != NULL) { + Eterm tup = TUPLE2(hp, CAR(kv), CDR(kv)); + hp += 3; + res = CONS(hp, tup, res); + hp += 2; + } + DESTROY_WSTACK(stack); + return res; +} + +void hashmap_iterator_init(ErtsWStack* s, Eterm node, int reverse) { + Eterm hdr = *hashmap_val(node); + Uint sz; + + switch(hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_HEAD_ARRAY: + sz = 16; + break; + case HAMT_SUBTAG_HEAD_BITMAP: + sz = hashmap_bitcount(MAP_HEADER_VAL(hdr)); + break; + default: + erl_exit(1, "bad header"); + } + + WSTACK_PUSH3((*s), (UWord)THE_NON_VALUE, /* end marker */ + (UWord)(!reverse ? 0 : sz+1), + (UWord)node); +} + +Eterm* hashmap_iterator_next(ErtsWStack* s) { + Eterm node, *ptr, hdr; + Uint32 sz; + Uint idx; + + for (;;) { + ASSERT(!WSTACK_ISEMPTY((*s))); + node = (Eterm) WSTACK_POP((*s)); + if (is_non_value(node)) { + return NULL; + } + idx = (Uint) WSTACK_POP((*s)); + for (;;) { + ASSERT(is_boxed(node)); + ptr = boxed_val(node); + hdr = *ptr; + ASSERT(is_header(hdr)); + switch(hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_HEAD_ARRAY: + ptr++; + case HAMT_SUBTAG_NODE_ARRAY: + sz = 16; + break; + case HAMT_SUBTAG_HEAD_BITMAP: + ptr++; + case HAMT_SUBTAG_NODE_BITMAP: + sz = hashmap_bitcount(MAP_HEADER_VAL(hdr)); + ASSERT(sz < 17); + break; + default: + erl_exit(1, "bad header"); + } + + idx++; + + if (idx <= sz) { + WSTACK_PUSH2((*s), (UWord)idx, (UWord)node); + + if (is_list(ptr[idx])) { + return list_val(ptr[idx]); + } + ASSERT(is_boxed(ptr[idx])); + node = ptr[idx]; + idx = 0; + } + else + break; /* and pop parent node */ + } + } +} + +Eterm* hashmap_iterator_prev(ErtsWStack* s) { + Eterm node, *ptr, hdr; + Uint32 sz; + Uint idx; + + for (;;) { + ASSERT(!WSTACK_ISEMPTY((*s))); + node = (Eterm) WSTACK_POP((*s)); + if (is_non_value(node)) { + return NULL; + } + idx = (Uint) WSTACK_POP((*s)); + for (;;) { + ASSERT(is_boxed(node)); + ptr = boxed_val(node); + hdr = *ptr; + ASSERT(is_header(hdr)); + switch(hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_HEAD_ARRAY: + ptr++; + case HAMT_SUBTAG_NODE_ARRAY: + sz = 16; + break; + case HAMT_SUBTAG_HEAD_BITMAP: + ptr++; + case HAMT_SUBTAG_NODE_BITMAP: + sz = hashmap_bitcount(MAP_HEADER_VAL(hdr)); + ASSERT(sz < 17); + break; + default: + erl_exit(1, "bad header"); + } + + if (idx > sz) + idx = sz; + else + idx--; + + if (idx >= 1) { + WSTACK_PUSH2((*s), (UWord)idx, (UWord)node); + + if (is_list(ptr[idx])) { + return list_val(ptr[idx]); + } + ASSERT(is_boxed(ptr[idx])); + node = ptr[idx]; + idx = 17; + } + else + break; /* and pop parent node */ + } + } +} + +const Eterm * +#if HALFWORD_HEAP +erts_hashmap_get_rel(Uint32 hx, Eterm key, Eterm node, Eterm *map_base) +#else +erts_hashmap_get(Uint32 hx, Eterm key, Eterm node) +#endif +{ + Eterm *ptr, hdr; + Uint ix,slot, lvl = 0; + Uint32 hval,bp; + DeclareTmpHeapNoproc(th,2); + UseTmpHeapNoproc(2); + + for (;;) { + switch(primary_tag(node)) { + case TAG_PRIMARY_LIST: /* LEAF NODE [K|V] */ + ptr = list_val(node); + UnUseTmpHeapNoproc(2); + + if (eq_rel(CAR(ptr), map_base, key, NULL)) { + return &(CDR(ptr)); + } + return NULL; + case TAG_PRIMARY_BOXED: + ptr = boxed_val(node); + hdr = *ptr; + ASSERT(is_header(hdr)); + + switch(hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_NODE_ARRAY: + ix = hashmap_index(hx); + hx = hashmap_shift_hash(th,hx,lvl,key); + node = ptr[ix+1]; + break; + case HAMT_SUBTAG_HEAD_ARRAY: + ix = hashmap_index(hx); + hx = hashmap_shift_hash(th,hx,lvl,key); + node = ptr[ix+2]; + break; + case HAMT_SUBTAG_NODE_BITMAP: + hval = MAP_HEADER_VAL(hdr); + ix = hashmap_index(hx); + bp = 1 << ix; + slot = hashmap_bitcount(hval & (bp - 1)); + + /* occupied */ + if (bp & hval) { + hx = hashmap_shift_hash(th,hx,lvl,key); + node = ptr[slot+1]; + break; + } + /* not occupied */ + UnUseTmpHeapNoproc(2); + return NULL; + case HAMT_SUBTAG_HEAD_BITMAP: + hval = MAP_HEADER_VAL(hdr); + ix = hashmap_index(hx); + bp = 1 << ix; + slot = hashmap_bitcount(hval & (bp - 1)); + + /* occupied */ + if (bp & hval) { + hx = hashmap_shift_hash(th,hx,lvl,key); + node = ptr[slot+2]; + break; + } + /* not occupied */ + UnUseTmpHeapNoproc(2); + return NULL; + default: + erl_exit(1, "bad header tag %ld\r\n", hdr & _HEADER_MAP_SUBTAG_MASK); + break; + } + break; + default: + erl_exit(1, "bad primary tag %p\r\n", node); + break; + } + } + UnUseTmpHeapNoproc(2); + return NULL; +} + +Eterm erts_hashmap_insert(Process *p, Uint32 hx, Eterm key, Eterm value, + Eterm map, int is_update) { + Uint size, upsz; + Eterm *hp, res = THE_NON_VALUE; + DECLARE_ESTACK(stack); + if (erts_hashmap_insert_down(hx, key, map, &size, &upsz, &stack, is_update)) { + hp = HAlloc(p, size); + res = erts_hashmap_insert_up(hp, key, value, &upsz, &stack); + } + + DESTROY_ESTACK(stack); + ERTS_VERIFY_UNUSED_TEMP_ALLOC(p); + ERTS_HOLE_CHECK(p); + + return res; +} + + +int erts_hashmap_insert_down(Uint32 hx, Eterm key, Eterm node, Uint *sz, + Uint *update_size, ErtsEStack *sp, int is_update) { + Eterm *ptr; + Eterm hdr, ckey; + Eterm th[2]; + Uint32 ix, cix, bp, hval, chx; + Uint slot, lvl = 0, clvl; + Uint size = 0, n = 0; + + *update_size = 1; + + for (;;) { + switch(primary_tag(node)) { + case TAG_PRIMARY_LIST: /* LEAF NODE [K|V] */ + ptr = list_val(node); + ckey = CAR(ptr); + if (EQ(ckey, key)) { + *update_size = 0; + goto unroll; + } + if (is_update) { + return 0; + } + goto insert_subnodes; + case TAG_PRIMARY_BOXED: + ptr = boxed_val(node); + hdr = *ptr; + ASSERT(is_header(hdr)); + + switch(hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_NODE_ARRAY: + ix = hashmap_index(hx); + hx = hashmap_shift_hash(th,hx,lvl,key); + size += HAMT_NODE_ARRAY_SZ; + ESTACK_PUSH2(*sp, ix, node); + node = ptr[ix+1]; + break; + case HAMT_SUBTAG_HEAD_ARRAY: + ix = hashmap_index(hx); + hx = hashmap_shift_hash(th,hx,lvl,key); + size += HAMT_HEAD_ARRAY_SZ; + ESTACK_PUSH2(*sp, ix, node); + node = ptr[ix+2]; + break; + case HAMT_SUBTAG_NODE_BITMAP: + hval = MAP_HEADER_VAL(hdr); + ix = hashmap_index(hx); + bp = 1 << ix; + slot = hashmap_bitcount(hval & (bp - 1)); + n = hashmap_bitcount(hval); + + ESTACK_PUSH(*sp, n); + ESTACK_PUSH3(*sp, bp, slot, node); + + /* occupied */ + if (bp & hval) { + hx = hashmap_shift_hash(th,hx,lvl,key); + node = ptr[slot+1]; + ASSERT(HAMT_NODE_BITMAP_SZ(n) <= 17); + size += HAMT_NODE_BITMAP_SZ(n); + break; + } + /* not occupied */ + if (is_update) { + return 0; + } + size += HAMT_NODE_BITMAP_SZ(n+1); + goto unroll; + case HAMT_SUBTAG_HEAD_BITMAP: + hval = MAP_HEADER_VAL(hdr); + ix = hashmap_index(hx); + bp = 1 << ix; + slot = hashmap_bitcount(hval & (bp - 1)); + n = hashmap_bitcount(hval); + + ESTACK_PUSH(*sp, n); + ESTACK_PUSH3(*sp, bp, slot, node); + + /* occupied */ + if (bp & hval) { + hx = hashmap_shift_hash(th,hx,lvl,key); + node = ptr[slot+2]; + ASSERT(HAMT_HEAD_BITMAP_SZ(n) <= 18); + size += HAMT_HEAD_BITMAP_SZ(n); + break; + } + /* not occupied */ + if (is_update) { + return 0; + } + size += HAMT_HEAD_BITMAP_SZ(n+1); + goto unroll; + default: + erl_exit(1, "bad header tag %ld\r\n", hdr & _HEADER_MAP_SUBTAG_MASK); + break; + } + break; + default: + erl_exit(1, "bad primary tag %p\r\n", node); + break; + } + } +insert_subnodes: + clvl = lvl; + chx = hashmap_restore_hash(th,clvl,ckey); + size += HAMT_NODE_BITMAP_SZ(2); + ix = hashmap_index(hx); + cix = hashmap_index(chx); + + while (cix == ix) { + ESTACK_PUSH(*sp, 0); + ESTACK_PUSH3(*sp, 1 << ix, 0, MAP_HEADER_HAMT_NODE_BITMAP(0)); + size += HAMT_NODE_BITMAP_SZ(1); + hx = hashmap_shift_hash(th,hx,lvl,key); + chx = hashmap_shift_hash(th,chx,clvl,ckey); + ix = hashmap_index(hx); + cix = hashmap_index(chx); + } + ESTACK_PUSH3(*sp, cix, ix, node); + +unroll: + *sz = size + /* res cons */ 2; + return 1; +} + +Eterm erts_hashmap_insert_up(Eterm *hp, Eterm key, Eterm value, + Uint *update_size, ErtsEStack *sp) { + Eterm node, *ptr, hdr; + Eterm res; + Eterm *nhp = NULL; + Uint32 ix, cix, bp, hval; + Uint slot, n; + /* Needed for halfword */ + DeclareTmpHeapNoproc(fake,1); + UseTmpHeapNoproc(1); + + res = CONS(hp, key, value); hp += 2; + + do { + node = ESTACK_POP(*sp); + switch(primary_tag(node)) { + case TAG_PRIMARY_LIST: + ix = (Uint32) ESTACK_POP(*sp); + cix = (Uint32) ESTACK_POP(*sp); + + nhp = hp; + *hp++ = MAP_HEADER_HAMT_NODE_BITMAP((1 << ix) | (1 << cix)); + if (ix < cix) { + *hp++ = res; + *hp++ = node; + } else { + *hp++ = node; + *hp++ = res; + } + res = make_hashmap(nhp); + break; + case TAG_PRIMARY_HEADER: + /* subnodes, fake it */ + *fake = node; + node = make_boxed(fake); + case TAG_PRIMARY_BOXED: + ptr = boxed_val(node); + hdr = *ptr; + ASSERT(is_header(hdr)); + + switch(hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_NODE_ARRAY: + slot = (Uint) ESTACK_POP(*sp); + nhp = hp; + n = HAMT_NODE_ARRAY_SZ; + while(n--) { *hp++ = *ptr++; } + nhp[slot+1] = res; + res = make_hashmap(nhp); + break; + case HAMT_SUBTAG_HEAD_ARRAY: + slot = (Uint) ESTACK_POP(*sp); + nhp = hp; + n = HAMT_HEAD_ARRAY_SZ - 2; + *hp++ = MAP_HEADER_HAMT_HEAD_ARRAY; ptr++; + *hp++ = (*ptr++) + *update_size; + while(n--) { *hp++ = *ptr++; } + nhp[slot+2] = res; + res = make_hashmap(nhp); + break; + case HAMT_SUBTAG_NODE_BITMAP: + slot = (Uint) ESTACK_POP(*sp); + bp = (Uint32) ESTACK_POP(*sp); + n = (Uint32) ESTACK_POP(*sp); + hval = MAP_HEADER_VAL(hdr); + nhp = hp; + *hp++ = MAP_HEADER_HAMT_NODE_BITMAP(hval | bp); ptr++; + + n -= slot; + while(slot--) { *hp++ = *ptr++; } + *hp++ = res; + if (hval & bp) { ptr++; n--; } + while(n--) { *hp++ = *ptr++; } + + if ((hval | bp) == 0xffff) { + *nhp = make_arityval(16); + } + res = make_hashmap(nhp); + break; + case HAMT_SUBTAG_HEAD_BITMAP: + slot = (Uint) ESTACK_POP(*sp); + bp = (Uint32) ESTACK_POP(*sp); + n = (Uint32) ESTACK_POP(*sp); + hval = MAP_HEADER_VAL(hdr); + nhp = hp; + *hp++ = MAP_HEADER_HAMT_HEAD_BITMAP(hval | bp); ptr++; + *hp++ = (*ptr++) + *update_size; + + n -= slot; + while(slot--) { *hp++ = *ptr++; } + *hp++ = res; + if (hval & bp) { ptr++; n--; } + while(n--) { *hp++ = *ptr++; } + + if ((hval | bp) == 0xffff) { + *nhp = MAP_HEADER_HAMT_HEAD_ARRAY; + } + res = make_hashmap(nhp); + break; + default: + erl_exit(1, "bad header tag %x\r\n", hdr & _HEADER_MAP_SUBTAG_MASK); + break; + } + break; + default: + erl_exit(1, "bad primary tag %x\r\n", primary_tag(node)); + break; + } + + } while(!ESTACK_ISEMPTY(*sp)); + + UnUseTmpHeapNoproc(1); + return res; +} + +static Eterm hashmap_keys(Process* p, Eterm node) { + DECLARE_WSTACK(stack); + hashmap_head_t* root; + Eterm *hp, *kv; + Eterm res = NIL; + + root = (hashmap_head_t*) boxed_val(node); + hp = HAlloc(p, root->size * 2); + hashmap_iterator_init(&stack, node, 0); + while ((kv=hashmap_iterator_next(&stack)) != NULL) { + res = CONS(hp, CAR(kv), res); + hp += 2; + } + DESTROY_WSTACK(stack); + return res; +} + +static Eterm hashmap_values(Process* p, Eterm node) { + DECLARE_WSTACK(stack); + hashmap_head_t* root; + Eterm *hp, *kv; + Eterm res = NIL; + + root = (hashmap_head_t*) boxed_val(node); + hp = HAlloc(p, root->size * 2); + hashmap_iterator_init(&stack, node, 0); + while ((kv=hashmap_iterator_next(&stack)) != NULL) { + res = CONS(hp, CDR(kv), res); + hp += 2; + } + DESTROY_WSTACK(stack); + return res; +} + +static Eterm hashmap_delete(Process *p, Uint32 hx, Eterm key, Eterm map) { + Eterm *hp = NULL, *nhp = NULL, *hp_end = NULL; + Eterm th[2]; + Eterm *ptr; + Eterm hdr, res = map, node = map; + Uint32 ix, bp, hval; + Uint slot, lvl = 0; + Uint size = 0, n = 0; + DECLARE_ESTACK(stack); + + for (;;) { + switch(primary_tag(node)) { + case TAG_PRIMARY_LIST: + if (EQ(CAR(list_val(node)), key)) { + goto unroll; + } + goto not_found; + case TAG_PRIMARY_BOXED: + ptr = boxed_val(node); + hdr = *ptr; + ASSERT(is_header(hdr)); + + switch(hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_NODE_ARRAY: + ix = hashmap_index(hx); + hx = hashmap_shift_hash(th,hx,lvl,key); + size += HAMT_NODE_ARRAY_SZ; + ESTACK_PUSH2(stack, ix, node); + node = ptr[ix+1]; + break; + case HAMT_SUBTAG_HEAD_ARRAY: + ix = hashmap_index(hx); + hx = hashmap_shift_hash(th,hx,lvl,key); + size += HAMT_HEAD_ARRAY_SZ; + ESTACK_PUSH2(stack, ix, node); + node = ptr[ix+2]; + break; + case HAMT_SUBTAG_NODE_BITMAP: + hval = MAP_HEADER_VAL(hdr); + ix = hashmap_index(hx); + bp = 1 << ix; + slot = hashmap_bitcount(hval & (bp - 1)); + n = hashmap_bitcount(hval); + + ESTACK_PUSH(stack, n); + ESTACK_PUSH3(stack, bp, slot, node); + + /* occupied */ + if (bp & hval) { + hx = hashmap_shift_hash(th,hx,lvl,key); + node = ptr[slot+1]; + ASSERT(HAMT_NODE_BITMAP_SZ(n) <= 17); + size += HAMT_NODE_BITMAP_SZ(n); + break; + } + /* not occupied */ + goto not_found; + case HAMT_SUBTAG_HEAD_BITMAP: + hval = MAP_HEADER_VAL(hdr); + ix = hashmap_index(hx); + bp = 1 << ix; + slot = hashmap_bitcount(hval & (bp - 1)); + n = hashmap_bitcount(hval); + + ESTACK_PUSH(stack, n); + ESTACK_PUSH3(stack, bp, slot, node); + + /* occupied */ + if (bp & hval) { + hx = hashmap_shift_hash(th,hx,lvl,key); + node = ptr[slot+2]; + ASSERT(HAMT_HEAD_BITMAP_SZ(n) <= 18); + size += HAMT_HEAD_BITMAP_SZ(n); + break; + } + /* not occupied */ + goto not_found; + default: + erl_exit(1, "bad header tag %ld\r\n", hdr & _HEADER_MAP_SUBTAG_MASK); + break; + } + break; + default: + erl_exit(1, "bad primary tag %p\r\n", node); + break; + } + } + +unroll: + /* the size is bounded and atleast one less than the previous size */ + size -= 1; + n = hashmap_size(map) - 1; + + if (n <= MAP_SMALL_MAP_LIMIT) { + DECLARE_WSTACK(wstack); + Eterm *kv, *ks, *vs; + flatmap_t *mp; + Eterm keys; + + DESTROY_ESTACK(stack); + + /* build flat structure */ + hp = HAlloc(p, 3 + 1 + (2 * n)); + keys = make_tuple(hp); + *hp++ = make_arityval(n); + ks = hp; + hp += n; + mp = (flatmap_t*)hp; + hp += MAP_HEADER_SIZE; + vs = hp; + + mp->thing_word = MAP_HEADER; + mp->size = n; + mp->keys = keys; + + hashmap_iterator_init(&wstack, map, 0); + + while ((kv=hashmap_iterator_next(&wstack)) != NULL) { + if (EQ(CAR(kv),key)) + continue; + *ks++ = CAR(kv); + *vs++ = CDR(kv); + } + + /* it cannot have multiple keys */ + erts_validate_and_sort_flatmap(mp); + + DESTROY_WSTACK(wstack); + return make_flatmap(mp); + } + + hp = HAlloc(p, size); + hp_end = hp + size; + res = THE_NON_VALUE; + + do { + node = ESTACK_POP(stack); + + /* all nodes are things */ + ptr = boxed_val(node); + hdr = *ptr; + ASSERT(is_header(hdr)); + + switch(hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_NODE_ARRAY: + ix = (Uint) ESTACK_POP(stack); + nhp = hp; + if (res == THE_NON_VALUE) { + *hp++ = MAP_HEADER_HAMT_NODE_BITMAP(0xffff ^ (1 << ix)); ptr++; + n = 16; + n -= ix; + while(ix--) { *hp++ = *ptr++; } + ptr++; n--; + while(n--) { *hp++ = *ptr++; } + res = make_hashmap(nhp); + } else { + n = HAMT_NODE_ARRAY_SZ; + while(n--) { *hp++ = *ptr++; } + nhp[ix+1] = res; + res = make_hashmap(nhp); + } + break; + case HAMT_SUBTAG_HEAD_ARRAY: + ix = (Uint) ESTACK_POP(stack); + nhp = hp; + if (res == THE_NON_VALUE) { + n = 16; + n -= ix; + *hp++ = MAP_HEADER_HAMT_HEAD_BITMAP(0xffff ^ (1 << ix)); ptr++; + *hp++ = (*ptr++) - 1; + while(ix--) { *hp++ = *ptr++; } + ptr++; n--; + while(n--) { *hp++ = *ptr++; } + res = make_hashmap(nhp); + } else { + n = 16; + *hp++ = MAP_HEADER_HAMT_HEAD_ARRAY; ptr++; + *hp++ = (*ptr++) - 1; + while(n--) { *hp++ = *ptr++; } + nhp[ix+2] = res; + res = make_hashmap(nhp); + } + break; + case HAMT_SUBTAG_NODE_BITMAP: + slot = (Uint) ESTACK_POP(stack); + bp = (Uint32) ESTACK_POP(stack); + n = (Uint32) ESTACK_POP(stack); + nhp = hp; + + /* bitmap change matrix + * res | none leaf bitmap + * ---------------------------- + * n=1 | remove remove keep + * n=2 | other keep keep + * n>2 | shrink keep keep + * + * other: (remember, n is 2) + * shrink if the other bitmap value is a bitmap node + * remove if the other bitmap value is a leaf + * + * remove: + * this bitmap node is removed, res is moved up in tree (could be none) + * this is a special case of shrink + * + * keep: + * the current path index is still used down in the tree, need to keep it + * copy as usual with the updated res + * + * shrink: + * the current path index is no longer used down in the tree, remove it (shrink) + */ + if (res == THE_NON_VALUE) { + if (n == 1) { + break; + } else if (n == 2) { + if (slot == 0) { + ix = 2; /* off by one 'cause hdr */ + } else { + ix = 1; /* off by one 'cause hdr */ + } + if (primary_tag(ptr[ix]) == TAG_PRIMARY_LIST) { + res = ptr[ix]; + } else { + hval = MAP_HEADER_VAL(hdr); + *hp++ = MAP_HEADER_HAMT_NODE_BITMAP(hval ^ bp); + *hp++ = ptr[ix]; + res = make_hashmap(nhp); + } + } else { + /* n > 2 */ + hval = MAP_HEADER_VAL(hdr); + *hp++ = MAP_HEADER_HAMT_NODE_BITMAP(hval ^ bp); ptr++; + n -= slot; + while(slot--) { *hp++ = *ptr++; } + ptr++; n--; + while(n--) { *hp++ = *ptr++; } + res = make_hashmap(nhp); + } + } else if (primary_tag(res) == TAG_PRIMARY_LIST && n == 1) { + break; + } else { + /* res is bitmap or leaf && n > 1, keep */ + n -= slot; + *hp++ = *ptr++; + while(slot--) { *hp++ = *ptr++; } + *hp++ = res; + ptr++; n--; + while(n--) { *hp++ = *ptr++; } + res = make_hashmap(nhp); + } + break; + case HAMT_SUBTAG_HEAD_BITMAP: + slot = (Uint) ESTACK_POP(stack); + bp = (Uint32) ESTACK_POP(stack); + n = (Uint32) ESTACK_POP(stack); + nhp = hp; + + if (res != THE_NON_VALUE) { + *hp++ = *ptr++; + *hp++ = (*ptr++) - 1; + n -= slot; + while(slot--) { *hp++ = *ptr++; } + *hp++ = res; + ptr++; n--; + while(n--) { *hp++ = *ptr++; } + } else { + hval = MAP_HEADER_VAL(hdr); + *hp++ = MAP_HEADER_HAMT_HEAD_BITMAP(hval ^ bp); ptr++; + *hp++ = (*ptr++) - 1; + n -= slot; + while(slot--) { *hp++ = *ptr++; } + ptr++; n--; + while(n--) { *hp++ = *ptr++; } + } + res = make_hashmap(nhp); + break; + default: + erl_exit(1, "bad header tag %x\r\n", hdr & _HEADER_MAP_SUBTAG_MASK); + break; + } + } while(!ESTACK_ISEMPTY(stack)); + HRelease(p, hp_end, hp); +not_found: + DESTROY_ESTACK(stack); + ERTS_VERIFY_UNUSED_TEMP_ALLOC(p); + ERTS_HOLE_CHECK(p); + return res; +} + + +int erts_validate_and_sort_flatmap(flatmap_t* mp) { - Eterm *ks = map_get_keys(mp); - Eterm *vs = map_get_values(mp); - Uint sz = map_get_size(mp); + Eterm *ks = flatmap_get_keys(mp); + Eterm *vs = flatmap_get_values(mp); + Uint sz = flatmap_get_size(mp); Uint ix,jx; Eterm tmp; int c; @@ -822,6 +2528,55 @@ int erts_validate_and_sort_map(map_t* mp) return 1; } +/* Really rough estimate of sqrt(x) + * Guaranteed not to be less than sqrt(x) + */ +static int int_sqrt_ceiling(Uint x) +{ + int n; + + if (x <= 2) + return x; + + n = erts_fit_in_bits_uint(x-1); + if (n & 1) { + /* Calc: sqrt(2^n) = 2^(n/2) * sqrt(2) ~= 2^(n/2) * 3 / 2 */ + return (1 << (n/2 - 1)) * 3; + } + else { + /* Calc: sqrt(2^n) = 2^(n/2) */ + return 1 << (n / 2); + } +} + +Uint hashmap_over_estimated_heap_size(Uint k) +{ + /* k is nr of key-value pairs. + N(k) is expected nr of nodes in hamt. + + Observation: + For uniformly distributed hash values, average of N varies between + 0.3*k and 0.4*k (with a beautiful sine curve) + and standard deviation of N is about sqrt(k)/3. + + Assuming normal probability distribution, we overestimate nr of nodes + by 15 std.devs above the average, which gives a probability for overrun + less than 1.0e-49 (same magnitude as a git SHA1 collision). + */ + Uint max_nodes = 2*k/5 + (15/3)*int_sqrt_ceiling(k); + return (k*2 + /* leaf cons cells */ + k + /* leaf list terms */ + max_nodes*2); /* headers + parent boxed terms */ +} + + +BIF_RETTYPE erts_debug_map_info_1(BIF_ALIST_1) { + if (is_hashmap(BIF_ARG_1)) { + BIF_RET(hashmap_info(BIF_P,BIF_ARG_1)); + } + BIF_ERROR(BIF_P, BADARG); +} + /* * erts_internal:map_to_tuple_keys/1 * @@ -829,9 +2584,233 @@ int erts_validate_and_sort_map(map_t* mp) */ BIF_RETTYPE erts_internal_map_to_tuple_keys_1(BIF_ALIST_1) { - if (is_map(BIF_ARG_1)) { - map_t *mp = (map_t*)map_val(BIF_ARG_1); + if (is_flatmap(BIF_ARG_1)) { + flatmap_t *mp = (flatmap_t*)flatmap_val(BIF_ARG_1); BIF_RET(mp->keys); } BIF_ERROR(BIF_P, BADARG); } + +/* + * erts_internal:map_type/1 + * + * Used in erts_debug:size/1 + */ + +BIF_RETTYPE erts_internal_map_type_1(BIF_ALIST_1) { + DECL_AM(hashmap); + DECL_AM(hashmap_node); + DECL_AM(flatmap); + if (is_flatmap(BIF_ARG_1)) { + BIF_RET(AM_flatmap); + } else if (is_hashmap(BIF_ARG_1)) { + Eterm hdr = *(boxed_val(BIF_ARG_1)); + ASSERT(is_header(hdr)); + switch (hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_HEAD_ARRAY: + case HAMT_SUBTAG_HEAD_BITMAP: + BIF_RET(AM_hashmap); + case HAMT_SUBTAG_NODE_ARRAY: + case HAMT_SUBTAG_NODE_BITMAP: + BIF_RET(AM_hashmap_node); + default: + erl_exit(1, "bad header"); + } + } + BIF_ERROR(BIF_P, BADARG); +} + +/* + * erts_internal:map_hashmap_children/1 + * + * Used in erts_debug:size/1 + */ + +BIF_RETTYPE erts_internal_map_hashmap_children_1(BIF_ALIST_1) { + if (is_hashmap(BIF_ARG_1)) { + Eterm node = BIF_ARG_1; + Eterm *ptr, hdr, *hp, res = NIL; + Uint sz = 0; + ptr = boxed_val(node); + hdr = *ptr; + + ASSERT(is_header(hdr)); + + switch(hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_NODE_ARRAY: + sz = 16; + ptr += 1; + break; + case HAMT_SUBTAG_NODE_BITMAP: + sz = hashmap_bitcount(MAP_HEADER_VAL(hdr)); + ptr += 1; + break; + case HAMT_SUBTAG_HEAD_BITMAP: + sz = hashmap_bitcount(MAP_HEADER_VAL(hdr)); + ptr += 2; + break; + case HAMT_SUBTAG_HEAD_ARRAY: + sz = 16; + ptr += 2; + break; + default: + erl_exit(1, "bad header\r\n"); + break; + } + ASSERT(sz < 17); + hp = HAlloc(BIF_P, 2*sz); + while(sz--) { res = CONS(hp, *ptr++, res); hp += 2; } + BIF_RET(res); + } + BIF_ERROR(BIF_P, BADARG); +} + + +static Eterm hashmap_info(Process *p, Eterm node) { + Eterm *hp; + Eterm res = NIL, info = NIL; + Eterm *ptr, tup, hdr; + Uint sz; + DECL_AM(depth); + DECL_AM(leafs); + DECL_AM(bitmaps); + DECL_AM(arrays); + Uint nleaf=0, nbitmap=0, narray=0; + Uint bitmap_usage[16], leaf_usage[16]; + Uint lvl = 0, clvl; + DECLARE_ESTACK(stack); + + for (sz = 0; sz < 16; sz++) { + bitmap_usage[sz] = 0; + leaf_usage[sz] = 0; + } + + ptr = boxed_val(node); + ESTACK_PUSH(stack, 0); + ESTACK_PUSH(stack, node); + do { + node = ESTACK_POP(stack); + clvl = ESTACK_POP(stack); + if (lvl < clvl) + lvl = clvl; + switch(primary_tag(node)) { + case TAG_PRIMARY_LIST: + nleaf++; + leaf_usage[clvl] += 1; + break; + case TAG_PRIMARY_BOXED: + ptr = boxed_val(node); + hdr = *ptr; + ASSERT(is_header(hdr)); + switch(hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_NODE_ARRAY: + narray++; + sz = 16; + while(sz--) { + ESTACK_PUSH(stack, clvl + 1); + ESTACK_PUSH(stack, ptr[sz+1]); + } + break; + case HAMT_SUBTAG_NODE_BITMAP: + nbitmap++; + sz = hashmap_bitcount(MAP_HEADER_VAL(hdr)); + ASSERT(sz < 17); + bitmap_usage[sz-1] += 1; + while(sz--) { + ESTACK_PUSH(stack, clvl + 1); + ESTACK_PUSH(stack, ptr[sz+1]); + } + break; + case HAMT_SUBTAG_HEAD_BITMAP: + nbitmap++; + sz = hashmap_bitcount(MAP_HEADER_VAL(hdr)); + bitmap_usage[sz-1] += 1; + while(sz--) { + ESTACK_PUSH(stack, clvl + 1); + ESTACK_PUSH(stack, ptr[sz+2]); + } + break; + case HAMT_SUBTAG_HEAD_ARRAY: + narray++; + sz = 16; + while(sz--) { + ESTACK_PUSH(stack, clvl + 1); + ESTACK_PUSH(stack, ptr[sz+2]); + } + break; + default: + erl_exit(1, "bad header\r\n"); + break; + } + } + } while(!ESTACK_ISEMPTY(stack)); + + + /* size */ + sz = 0; + hashmap_bld_tuple_uint(NULL,&sz,16,leaf_usage); + hashmap_bld_tuple_uint(NULL,&sz,16,bitmap_usage); + + /* alloc */ + hp = HAlloc(p, 2+3 + 3*(2+4) + sz); + + info = hashmap_bld_tuple_uint(&hp,NULL,16,leaf_usage); + tup = TUPLE3(hp, AM_leafs, make_small(nleaf),info); hp += 4; + res = CONS(hp, tup, res); hp += 2; + + info = hashmap_bld_tuple_uint(&hp,NULL,16,bitmap_usage); + tup = TUPLE3(hp, AM_bitmaps, make_small(nbitmap), info); hp += 4; + res = CONS(hp, tup, res); hp += 2; + + tup = TUPLE3(hp, AM_arrays, make_small(narray),NIL); hp += 4; + res = CONS(hp, tup, res); hp += 2; + + tup = TUPLE2(hp, AM_depth, make_small(lvl)); hp += 3; + res = CONS(hp, tup, res); hp += 2; + + DESTROY_ESTACK(stack); + ERTS_HOLE_CHECK(p); + return res; +} + +static Eterm hashmap_bld_tuple_uint(Uint **hpp, Uint *szp, Uint n, Uint nums[]) { + Eterm res = THE_NON_VALUE; + Eterm *ts = (Eterm *)erts_alloc(ERTS_ALC_T_TMP, n * sizeof(Eterm)); + Uint i; + + for (i = 0; i < n; i++) { + ts[i] = erts_bld_uint(hpp, szp, nums[i]); + } + res = erts_bld_tuplev(hpp, szp, n, ts); + erts_free(ERTS_ALC_T_TMP, (void *) ts); + return res; +} + + +/* implementation of builtin emulations */ + +#if !ERTS_AT_LEAST_GCC_VSN__(3, 4, 0) +/* Count leading zeros emulation */ +Uint32 hashmap_clz(Uint32 x) { + Uint32 y; + int n = 32; + y = x >>16; if (y != 0) {n = n -16; x = y;} + y = x >> 8; if (y != 0) {n = n - 8; x = y;} + y = x >> 4; if (y != 0) {n = n - 4; x = y;} + y = x >> 2; if (y != 0) {n = n - 2; x = y;} + y = x >> 1; if (y != 0) return n - 2; + return n - x; +} + +const Uint32 SK5 = 0x55555555, SK3 = 0x33333333; +const Uint32 SKF0 = 0xF0F0F0F, SKFF = 0xFF00FF; + +/* CTPOP emulation */ +Uint32 hashmap_bitcount(Uint32 x) { + x -= ((x >> 1 ) & SK5); + x = (x & SK3 ) + ((x >> 2 ) & SK3 ); + x = (x & SKF0) + ((x >> 4 ) & SKF0); + x += x >> 8; + return (x + (x >> 16)) & 0x3F; +} +#endif diff --git a/erts/emulator/beam/erl_map.h b/erts/emulator/beam/erl_map.h index cfacb2ec28..1333a734a8 100644 --- a/erts/emulator/beam/erl_map.h +++ b/erts/emulator/beam/erl_map.h @@ -22,13 +22,23 @@ #define __ERL_MAP_H__ #include "sys.h" + +/* instrinsic wrappers */ +#if ERTS_AT_LEAST_GCC_VSN__(3, 4, 0) +#define hashmap_clz(x) ((Uint32) __builtin_clz((unsigned int)(x))) +#define hashmap_bitcount(x) ((Uint32) __builtin_popcount((unsigned int) (x))) +#else +Uint32 hashmap_clz(Uint32 x); +Uint32 hashmap_bitcount(Uint32 x); +#endif + /* MAP */ -typedef struct map_s { +typedef struct flatmap_s { Eterm thing_word; Uint size; Eterm keys; /* tuple */ -} map_t; +} flatmap_t; /* map node * * ----------- @@ -42,31 +52,152 @@ typedef struct map_s { * ----------- */ +/* the head-node is a bitmap or array with an untagged size */ + + +#define hashmap_size(x) (((hashmap_head_t*) hashmap_val(x))->size) +#define hashmap_size_rel(RTERM, BASE) hashmap_size(rterm2wterm(RTERM, BASE)) +#define hashmap_make_hash(Key) make_internal_hash(Key) + +#define hashmap_restore_hash(Heap,Lvl,Key) \ + (((Lvl) < 8) ? hashmap_make_hash(Key) >> (4*(Lvl)) : hashmap_make_hash(CONS(Heap, make_small((Lvl)>>3), (Key))) >> (4*((Lvl) & 7))) +#define hashmap_shift_hash(Heap,Hx,Lvl,Key) \ + (((++(Lvl)) & 7) ? (Hx) >> 4 : hashmap_make_hash(CONS(Heap, make_small((Lvl)>>3), Key))) /* erl_term.h stuff */ -#define make_map(x) make_boxed((Eterm*)(x)) -#define make_map_rel(x, BASE) make_boxed_rel((Eterm*)(x),(BASE)) -#define is_map(x) (is_boxed((x)) && is_map_header(*boxed_val((x)))) -#define is_map_rel(RTERM,BASE) is_map(rterm2wterm(RTERM,BASE)) -#define is_not_map(x) (!is_map((x))) -#define is_map_header(x) (((x) & (_TAG_HEADER_MASK)) == _TAG_HEADER_MAP) -#define header_is_map(x) ((((x) & (_HEADER_SUBTAG_MASK)) == MAP_SUBTAG)) -#define map_val(x) (_unchecked_boxed_val((x))) -#define map_val_rel(RTERM, BASE) map_val(rterm2wterm(RTERM, BASE)) - -#define map_get_values(x) (((Eterm *)(x)) + 3) -#define map_get_keys(x) (((Eterm *)tuple_val(((map_t *)(x))->keys)) + 1) -#define map_get_size(x) (((map_t*)(x))->size) +#define make_flatmap(x) make_boxed((Eterm*)(x)) +#define make_flatmap_rel(x, BASE) make_boxed_rel((Eterm*)(x),(BASE)) +#define is_flatmap(x) (is_boxed((x)) && is_flatmap_header(*boxed_val((x)))) +#define is_flatmap_rel(RTERM,BASE) is_flatmap(rterm2wterm(RTERM,BASE)) +#define is_not_flatmap(x) (!is_flatmap((x))) +#define is_flatmap_header(x) (((x) & (_TAG_HEADER_MASK)) == _TAG_HEADER_MAP) +#define header_is_flatmap(x) ((((x) & (_HEADER_SUBTAG_MASK)) == MAP_SUBTAG)) +#define flatmap_val(x) (_unchecked_boxed_val((x))) +#define flatmap_val_rel(RTERM, BASE) flatmap_val(rterm2wterm(RTERM, BASE)) +#define flatmap_get_values(x) (((Eterm *)(x)) + 3) +#define flatmap_get_keys(x) (((Eterm *)tuple_val(((flatmap_t *)(x))->keys)) + 1) +#define flatmap_get_size(x) (((flatmap_t*)(x))->size) + +#ifdef DEBUG +#define MAP_SMALL_MAP_LIMIT (3) +#else +#define MAP_SMALL_MAP_LIMIT (32) +#endif #define MAP_HEADER _make_header(1,_TAG_HEADER_MAP) -#define MAP_HEADER_SIZE (sizeof(map_t) / sizeof(Eterm)) - -Eterm erts_maps_put(Process *p, Eterm key, Eterm value, Eterm map); -int erts_maps_update(Process *p, Eterm key, Eterm value, Eterm map, Eterm *res); -int erts_maps_find(Eterm key, Eterm map, Eterm *value); -int erts_maps_get(Eterm key, Eterm map, Eterm *value); -int erts_maps_remove(Process *p, Eterm key, Eterm map, Eterm *res); -int erts_validate_and_sort_map(map_t* map); +#define MAP_HEADER_SIZE (sizeof(flatmap_t) / sizeof(Eterm)) + +struct ErtsWStack_; +struct ErtsEStack_; + +Eterm erts_maps_put(Process *p, Eterm key, Eterm value, Eterm map); +int erts_maps_update(Process *p, Eterm key, Eterm value, Eterm map, Eterm *res); +int erts_maps_remove(Process *p, Eterm key, Eterm map, Eterm *res); + +Eterm erts_hashmap_insert(Process *p, Uint32 hx, Eterm key, Eterm value, + Eterm node, int is_update); +int erts_hashmap_insert_down(Uint32 hx, Eterm key, Eterm node, Uint *sz, + Uint *upsz, struct ErtsEStack_ *sp, int is_update); +Eterm erts_hashmap_insert_up(Eterm *hp, Eterm key, Eterm value, + Uint *upsz, struct ErtsEStack_ *sp); + +int erts_validate_and_sort_flatmap(flatmap_t* map); +Uint hashmap_over_estimated_heap_size(Uint n); +void hashmap_iterator_init(struct ErtsWStack_* s, Eterm node, int reverse); +Eterm* hashmap_iterator_next(struct ErtsWStack_* s); +Eterm* hashmap_iterator_prev(struct ErtsWStack_* s); +int hashmap_key_hash_cmp(Eterm* ap, Eterm* bp); +Eterm erts_hashmap_from_array(ErtsHeapFactory*, Eterm *leafs, Uint n, int reject_dupkeys); + +#define erts_hashmap_from_ks_and_vs(P, KS, VS, N) \ + erts_hashmap_from_ks_and_vs_extra((P), (KS), (VS), (N), THE_NON_VALUE, THE_NON_VALUE); + +Eterm erts_hashmap_from_ks_and_vs_extra(Process *p, Eterm *ks, Eterm *vs, Uint n, + Eterm k, Eterm v); + +const Eterm * +#if HALFWORD_HEAP +erts_maps_get_rel(Eterm key, Eterm map, Eterm *map_base); +# define erts_maps_get(A, B) erts_maps_get_rel(A, B, NULL) +#else +erts_maps_get(Eterm key, Eterm map); +# define erts_maps_get_rel(A, B, B_BASE) erts_maps_get(A, B) +#endif + +const Eterm * +#if HALFWORD_HEAP +erts_hashmap_get_rel(Uint32 hx, Eterm key, Eterm node, Eterm *map_base); +# define erts_hashmap_get(Hx, K, M) erts_hashmap_get_rel(Hx, K, M, NULL) +#else +erts_hashmap_get(Uint32 hx, Eterm key, Eterm map); +# define erts_hashmap_get_rel(Hx, K, M, M_BASE) erts_hashmap_get(Hx, K, M) #endif +/* hamt nodes v2.0 + * + * node :: leaf | array | bitmap + * head + */ +typedef struct hashmap_head_s { + Eterm thing_word; + Uint size; + Eterm items[1]; +} hashmap_head_t; + +/* thing_word tagscheme + * Need two bits for map subtags + * + * Original HEADER representation: + * + * aaaaaaaaaaaaaaaa aaaaaaaaaatttt00 arity:26, tag:4 + * + * For maps we have: + * + * vvvvvvvvvvvvvvvv aaaaaaaamm111100 val:16, arity:8, mtype:2 + * + * unsure about trailing zeros + * + * map-tag: + * 00 - flat map tag (non-hamt) -> val:16 = #items + * 01 - map-node bitmap tag -> val:16 = bitmap + * 10 - map-head (array-node) -> val:16 = 0xffff + * 11 - map-head (bitmap-node) -> val:16 = bitmap + */ + +/* erl_map.h stuff */ + +#define is_hashmap_header_head(x) ((MAP_HEADER_TYPE(x) & (0x2))) + +#define MAKE_MAP_HEADER(Type,Arity,Val) \ + (_make_header(((((Uint16)(Val)) << MAP_HEADER_ARITY_SZ) | (Arity)) << MAP_HEADER_TAG_SZ | (Type) , _TAG_HEADER_HASHMAP)) + +#define MAP_HEADER_HAMT_HEAD_ARRAY \ + MAKE_MAP_HEADER(MAP_HEADER_TAG_HAMT_HEAD_ARRAY,0x1,0xffff) + +#define MAP_HEADER_HAMT_HEAD_BITMAP(Bmp) \ + MAKE_MAP_HEADER(MAP_HEADER_TAG_HAMT_HEAD_BITMAP,0x1,Bmp) + +#define MAP_HEADER_HAMT_NODE_ARRAY \ + make_arityval(16) + +#define MAP_HEADER_HAMT_NODE_BITMAP(Bmp) \ + MAKE_MAP_HEADER(MAP_HEADER_TAG_HAMT_NODE_BITMAP,0x0,Bmp) + +#define HAMT_HEAD_EMPTY_SZ (2) +#define HAMT_NODE_ARRAY_SZ (17) +#define HAMT_HEAD_ARRAY_SZ (18) +#define HAMT_NODE_BITMAP_SZ(n) (1 + n) +#define HAMT_HEAD_BITMAP_SZ(n) (2 + n) + +#define _HEADER_MAP_SUBTAG_MASK (0xfc) /* 2 bits maps tag + 4 bits subtag + 2 ignore bits */ +/* SUBTAG_NODE_ARRAY is in fact a tuple with 16 elements */ +#define HAMT_SUBTAG_NODE_ARRAY (((16 << _HEADER_ARITY_OFFS) | ARITYVAL_SUBTAG) & _HEADER_MAP_SUBTAG_MASK) +#define HAMT_SUBTAG_NODE_BITMAP ((MAP_HEADER_TAG_HAMT_NODE_BITMAP << _HEADER_ARITY_OFFS) | HASHMAP_SUBTAG) +#define HAMT_SUBTAG_HEAD_ARRAY ((MAP_HEADER_TAG_HAMT_HEAD_ARRAY << _HEADER_ARITY_OFFS) | HASHMAP_SUBTAG) +#define HAMT_SUBTAG_HEAD_BITMAP ((MAP_HEADER_TAG_HAMT_HEAD_BITMAP << _HEADER_ARITY_OFFS) | HASHMAP_SUBTAG) + +#define hashmap_index(hash) (((Uint32)hash) & 0xf) + + +#endif diff --git a/erts/emulator/beam/erl_math.c b/erts/emulator/beam/erl_math.c index 16d4fdc09c..9b864628db 100644 --- a/erts/emulator/beam/erl_math.c +++ b/erts/emulator/beam/erl_math.c @@ -207,6 +207,24 @@ BIF_RETTYPE math_log_1(BIF_ALIST_1) return math_call_1(BIF_P, log, BIF_ARG_1); } +#ifdef HAVE_LOG2 +static double +log2_wrapper(double x) +{ + return log2(x); +} +#else +static double +log2_wrapper(double x) +{ + return log(x) / 0.6931471805599453; /* log(2.0); */ +} +#endif + +BIF_RETTYPE math_log2_1(BIF_ALIST_1) +{ + return math_call_1(BIF_P, log2_wrapper, BIF_ARG_1); +} BIF_RETTYPE math_log10_1(BIF_ALIST_1) { diff --git a/erts/emulator/beam/erl_message.c b/erts/emulator/beam/erl_message.c index ce91acb6a4..22cbae10d1 100644 --- a/erts/emulator/beam/erl_message.c +++ b/erts/emulator/beam/erl_message.c @@ -1146,3 +1146,15 @@ erts_deliver_exit_message(Eterm from, Process *to, ErtsProcLocks *to_locksp, } } +Eterm* erts_produce_heap(ErtsHeapFactory* factory, Uint need, Uint xtra) +{ + Eterm* res; + if (factory->p) { + res = HAllocX(factory->p, need, xtra); + } else { + res = factory->hp; + factory->hp += need; + } + return res; +} + diff --git a/erts/emulator/beam/erl_message.h b/erts/emulator/beam/erl_message.h index a2a7193ea9..8713941769 100644 --- a/erts/emulator/beam/erl_message.h +++ b/erts/emulator/beam/erl_message.h @@ -68,6 +68,21 @@ struct erl_heap_fragment { Eterm mem[1]; /* Data */ }; +typedef struct { + Process* p; + Eterm* hp; +} ErtsHeapFactory; + +Eterm* erts_produce_heap(ErtsHeapFactory*, Uint need, Uint xtra); +#ifdef CHECK_FOR_HOLES +# define ERTS_FACTORY_HOLE_CHECK(f) do { \ + if ((f)->p) erts_check_for_holes((f)->p); \ + } while (0) +#else +# define ERTS_FACTORY_HOLE_CHECK(p) +#endif + + typedef struct erl_mesg { struct erl_mesg* next; /* Next message */ union { diff --git a/erts/emulator/beam/erl_mtrace.c b/erts/emulator/beam/erl_mtrace.c index c8bb126687..fa1bde1c87 100644 --- a/erts/emulator/beam/erl_mtrace.c +++ b/erts/emulator/beam/erl_mtrace.c @@ -627,7 +627,7 @@ erts_mtrace_install_wrapper_functions(void) if (erts_mtrace_enabled) { int i; /* Install trace functions */ - ASSERT(sizeof(erts_allctrs) == sizeof(real_allctrs)); + ERTS_CT_ASSERT(sizeof(erts_allctrs) == sizeof(real_allctrs)); sys_memcpy((void *) real_allctrs, (void *) erts_allctrs, diff --git a/erts/emulator/beam/erl_nif.c b/erts/emulator/beam/erl_nif.c index 47d0af16bc..c7c8b3fee3 100644 --- a/erts/emulator/beam/erl_nif.c +++ b/erts/emulator/beam/erl_nif.c @@ -552,9 +552,7 @@ int enif_alloc_binary(size_t size, ErlNifBinary* bin) if (refbin == NULL) { return 0; /* The NIF must take action */ } - refbin->flags = BIN_FLAG_DRV; /* BUGBUG: Flag? */ erts_refc_init(&refbin->refc, 1); - refbin->orig_size = (SWord) size; bin->size = size; bin->data = (unsigned char*) refbin->orig_bytes; @@ -574,7 +572,6 @@ int enif_realloc_binary(ErlNifBinary* bin, size_t size) if (!newbin) { return 0; } - newbin->orig_size = size; bin->ref_bin = newbin; bin->data = (unsigned char*) newbin->orig_bytes; bin->size = size; @@ -1914,12 +1911,16 @@ int enif_is_map(ErlNifEnv* env, ERL_NIF_TERM term) int enif_get_map_size(ErlNifEnv* env, ERL_NIF_TERM term, size_t *size) { - if (is_map(term)) { - map_t *mp; - mp = (map_t*)map_val(term); - *size = map_get_size(mp); + if (is_flatmap(term)) { + flatmap_t *mp; + mp = (flatmap_t*)flatmap_val(term); + *size = flatmap_get_size(mp); return 1; } + else if (is_hashmap(term)) { + *size = hashmap_size(term); + return 1; + } return 0; } @@ -1927,16 +1928,16 @@ ERL_NIF_TERM enif_make_new_map(ErlNifEnv* env) { Eterm* hp = alloc_heap(env,MAP_HEADER_SIZE+1); Eterm tup; - map_t *mp; + flatmap_t *mp; tup = make_tuple(hp); *hp++ = make_arityval(0); - mp = (map_t*)hp; + mp = (flatmap_t*)hp; mp->thing_word = MAP_HEADER; mp->size = 0; mp->keys = tup; - return make_map(mp); + return make_flatmap(mp); } int enif_make_map_put(ErlNifEnv* env, @@ -1945,7 +1946,7 @@ int enif_make_map_put(ErlNifEnv* env, Eterm value, Eterm *map_out) { - if (is_not_map(map_in)) { + if (!is_map(map_in)) { return 0; } flush_env(env); @@ -1959,10 +1960,16 @@ int enif_get_map_value(ErlNifEnv* env, Eterm key, Eterm *value) { - if (is_not_map(map)) { + const Eterm *ret; + if (!is_map(map)) { return 0; } - return erts_maps_get(key, map, value); + ret = erts_maps_get(key, map); + if (ret) { + *value = *ret; + return 1; + } + return 0; } int enif_make_map_update(ErlNifEnv* env, @@ -1972,7 +1979,7 @@ int enif_make_map_update(ErlNifEnv* env, Eterm *map_out) { int res; - if (is_not_map(map_in)) { + if (!is_map(map_in)) { return 0; } @@ -1988,7 +1995,7 @@ int enif_make_map_remove(ErlNifEnv* env, Eterm *map_out) { int res; - if (is_not_map(map_in)) { + if (!is_map(map_in)) { return 0; } flush_env(env); @@ -2002,13 +2009,13 @@ int enif_map_iterator_create(ErlNifEnv *env, ErlNifMapIterator *iter, ErlNifMapIteratorEntry entry) { - if (is_map(map)) { - map_t *mp = (map_t*)map_val(map); + if (is_flatmap(map)) { + flatmap_t *mp = (flatmap_t*)flatmap_val(map); size_t offset; switch (entry) { case ERL_NIF_MAP_ITERATOR_HEAD: offset = 0; break; - case ERL_NIF_MAP_ITERATOR_TAIL: offset = map_get_size(mp) - 1; break; + case ERL_NIF_MAP_ITERATOR_TAIL: offset = flatmap_get_size(mp) - 1; break; default: goto error; } @@ -2017,14 +2024,37 @@ int enif_map_iterator_create(ErlNifEnv *env, */ iter->map = map; - iter->ks = ((Eterm *)map_get_keys(mp)) + offset; - iter->vs = ((Eterm *)map_get_values(mp)) + offset; - iter->t_limit = map_get_size(mp) + 1; + iter->u.flat.ks = ((Eterm *)flatmap_get_keys(mp)) + offset; + iter->u.flat.vs = ((Eterm *)flatmap_get_values(mp)) + offset; + iter->size = flatmap_get_size(mp); iter->idx = offset + 1; return 1; } - + else if (is_hashmap(map)) { + iter->map = map; + iter->size = hashmap_size(map); + iter->u.hash.wstack = erts_alloc(ERTS_ALC_T_NIF, sizeof(ErtsDynamicWStack)); + WSTACK_INIT(iter->u.hash.wstack, ERTS_ALC_T_NIF); + + switch (entry) { + case ERL_NIF_MAP_ITERATOR_HEAD: + iter->idx = 1; + hashmap_iterator_init(&iter->u.hash.wstack->ws, map, 0); + iter->u.hash.kv = hashmap_iterator_next(&iter->u.hash.wstack->ws); + break; + case ERL_NIF_MAP_ITERATOR_TAIL: + iter->idx = hashmap_size(map); + hashmap_iterator_init(&iter->u.hash.wstack->ws, map, 1); + iter->u.hash.kv = hashmap_iterator_prev(&iter->u.hash.wstack->ws); + break; + default: + goto error; + } + ASSERT(!!iter->u.hash.kv == (iter->idx >= 1 && + iter->idx <= iter->size)); + return 1; + } error: #ifdef DEBUG iter->map = THE_NON_VALUE; @@ -2034,48 +2064,97 @@ error: void enif_map_iterator_destroy(ErlNifEnv *env, ErlNifMapIterator *iter) { - /* not used */ + if (is_hashmap(iter->map)) { + WSTACK_DESTROY(iter->u.hash.wstack->ws); + erts_free(ERTS_ALC_T_NIF, iter->u.hash.wstack); + } + else + ASSERT(is_flatmap(iter->map)); + #ifdef DEBUG iter->map = THE_NON_VALUE; #endif - } int enif_map_iterator_is_tail(ErlNifEnv *env, ErlNifMapIterator *iter) { - ASSERT(iter && is_map(iter->map)); - ASSERT(iter->idx >= 0 && (iter->idx <= map_get_size(map_val(iter->map)) + 1)); - return (iter->t_limit == 1 || iter->idx == iter->t_limit); + ASSERT(iter); + if (is_flatmap(iter->map)) { + ASSERT(iter->idx >= 0); + ASSERT(iter->idx <= flatmap_get_size(flatmap_val(iter->map)) + 1); + return (iter->size == 0 || iter->idx > iter->size); + } + else { + ASSERT(is_hashmap(iter->map)); + return iter->idx > iter->size; + } } int enif_map_iterator_is_head(ErlNifEnv *env, ErlNifMapIterator *iter) { - ASSERT(iter && is_map(iter->map)); - ASSERT(iter->idx >= 0 && (iter->idx <= map_get_size(map_val(iter->map)) + 1)); - return (iter->t_limit == 1 || iter->idx == 0); + ASSERT(iter); + if (is_flatmap(iter->map)) { + ASSERT(iter->idx >= 0); + ASSERT(iter->idx <= flatmap_get_size(flatmap_val(iter->map)) + 1); + return (iter->size == 0 || iter->idx == 0); + } + else { + ASSERT(is_hashmap(iter->map)); + return iter->idx == 0; + } } int enif_map_iterator_next(ErlNifEnv *env, ErlNifMapIterator *iter) { - ASSERT(iter && is_map(iter->map)); - if (iter->idx < iter->t_limit) { - iter->idx++; - iter->ks++; - iter->vs++; + ASSERT(iter); + if (is_flatmap(iter->map)) { + if (iter->idx <= iter->size) { + iter->idx++; + iter->u.flat.ks++; + iter->u.flat.vs++; + } + return (iter->idx <= iter->size); + } + else { + ASSERT(is_hashmap(iter->map)); + + if (iter->idx <= hashmap_size(iter->map)) { + if (iter->idx < 1) { + hashmap_iterator_init(&iter->u.hash.wstack->ws, iter->map, 0); + } + iter->u.hash.kv = hashmap_iterator_next(&iter->u.hash.wstack->ws); + iter->idx++; + ASSERT(!!iter->u.hash.kv == (iter->idx <= iter->size)); + } + return iter->idx <= iter->size; } - return (iter->idx != iter->t_limit); } int enif_map_iterator_prev(ErlNifEnv *env, ErlNifMapIterator *iter) { - ASSERT(iter && is_map(iter->map)); - if (iter->idx > 0) { - iter->idx--; - iter->ks--; - iter->vs--; + ASSERT(iter); + if (is_flatmap(iter->map)) { + if (iter->idx > 0) { + iter->idx--; + iter->u.flat.ks--; + iter->u.flat.vs--; + } + return iter->idx > 0; + } + else { + ASSERT(is_hashmap(iter->map)); + + if (iter->idx > 0) { + if (iter->idx > iter->size) { + hashmap_iterator_init(&iter->u.hash.wstack->ws, iter->map, 1); + } + iter->u.hash.kv = hashmap_iterator_prev(&iter->u.hash.wstack->ws); + iter->idx--; + ASSERT(!!iter->u.hash.kv == (iter->idx > 0)); + } + return iter->idx > 0; } - return (iter->idx > 0); } int enif_map_iterator_get_pair(ErlNifEnv *env, @@ -2083,15 +2162,25 @@ int enif_map_iterator_get_pair(ErlNifEnv *env, Eterm *key, Eterm *value) { - ASSERT(iter && is_map(iter->map)); - if (iter->idx > 0 && iter->idx < iter->t_limit) { - ASSERT(iter->ks >= map_get_keys(map_val(iter->map)) && - iter->ks < (map_get_keys(map_val(iter->map)) + map_get_size(map_val(iter->map)))); - ASSERT(iter->vs >= map_get_values(map_val(iter->map)) && - iter->vs < (map_get_values(map_val(iter->map)) + map_get_size(map_val(iter->map)))); - *key = *(iter->ks); - *value = *(iter->vs); - return 1; + ASSERT(iter); + if (is_flatmap(iter->map)) { + if (iter->idx > 0 && iter->idx <= iter->size) { + ASSERT(iter->u.flat.ks >= flatmap_get_keys(flatmap_val(iter->map)) && + iter->u.flat.ks < (flatmap_get_keys(flatmap_val(iter->map)) + flatmap_get_size(flatmap_val(iter->map)))); + ASSERT(iter->u.flat.vs >= flatmap_get_values(flatmap_val(iter->map)) && + iter->u.flat.vs < (flatmap_get_values(flatmap_val(iter->map)) + flatmap_get_size(flatmap_val(iter->map)))); + *key = *(iter->u.flat.ks); + *value = *(iter->u.flat.vs); + return 1; + } + } + else { + ASSERT(is_hashmap(iter->map)); + if (iter->idx > 0 && iter->idx <= iter->size) { + *key = CAR(iter->u.hash.kv); + *value = CDR(iter->u.hash.kv); + return 1; + } } return 0; } diff --git a/erts/emulator/beam/erl_nif.h b/erts/emulator/beam/erl_nif.h index 849024453c..9b2b90c82d 100644 --- a/erts/emulator/beam/erl_nif.h +++ b/erts/emulator/beam/erl_nif.h @@ -201,10 +201,18 @@ typedef enum typedef struct /* All fields all internal and may change */ { ERL_NIF_TERM map; - ERL_NIF_UINT t_limit; + ERL_NIF_UINT size; ERL_NIF_UINT idx; - ERL_NIF_TERM *ks; - ERL_NIF_TERM *vs; + union { + struct { + ERL_NIF_TERM *ks; + ERL_NIF_TERM *vs; + }flat; + struct { + struct ErtsDynamicWStack_* wstack; + ERL_NIF_TERM* kv; + }hash; + }u; void* __spare__[2]; /* for future additions to be ABI compatible (same struct size) */ } ErlNifMapIterator; diff --git a/erts/emulator/beam/erl_printf_term.c b/erts/emulator/beam/erl_printf_term.c index d18760dc43..ac5b139f8d 100644 --- a/erts/emulator/beam/erl_printf_term.c +++ b/erts/emulator/beam/erl_printf_term.c @@ -1,7 +1,7 @@ /* * %CopyrightBegin% * - * Copyright Ericsson AB 2005-2013. All Rights Reserved. + * Copyright Ericsson AB 2005-2014. All Rights Reserved. * * The contents of this file are subject to the Erlang Public License, * Version 1.1, (the "License"); you may not use this file except in @@ -25,6 +25,7 @@ #include "sys.h" #include "big.h" #include "erl_map.h" +#include "erl_binary.h" #define PRINT_CHAR(CNT, FN, ARG, C) \ do { \ @@ -138,6 +139,25 @@ is_printable_string(Eterm list, Eterm* base) return 0; } +static int is_printable_ascii(byte* bytep, Uint bytesize, Uint bitoffs) +{ + if (!bitoffs) { + while (bytesize--) { + if (*bytep < ' ' || *bytep >= 127) + return 0; + bytep++; + } + } else { + while (bytesize--) { + byte octet = (bytep[0] << bitoffs) | (bytep[1] >> (8-bitoffs)); + if (octet < ' ' || octet >= 127) + return 0; + bytep++; + } + } + return 1; +} + /* print a atom doing what quoting is necessary */ static int print_atom_name(fmtfn_t fn, void* arg, Eterm atom, long *dcount) { @@ -227,6 +247,17 @@ static int print_atom_name(fmtfn_t fn, void* arg, Eterm atom, long *dcount) #define PRT_PATCH_FUN_SIZE ((Eterm) 7) #define PRT_LAST_ARRAY_ELEMENT ((Eterm) 8) /* Note! Must be last... */ +#if 0 +static char *format_binary(Uint16 x, char *b) { + int z; + b[16] = '\0'; + for (z = 0; z < 16; z++) { + b[15-z] = ((x>>z) & 0x1) ? '1' : '0'; + } + return b; +} +#endif + static int print_term(fmtfn_t fn, void* arg, Eterm obj, long *dcount, Eterm* obj_base) /* ignored if !HALFWORD_HEAP */ @@ -283,13 +314,9 @@ print_term(fmtfn_t fn, void* arg, Eterm obj, long *dcount, tl = CDR(cons); if (is_not_nil(tl)) { if (is_list(tl)) { - WSTACK_PUSH(s, tl); - WSTACK_PUSH(s, PRT_ONE_CONS); - WSTACK_PUSH(s, PRT_COMMA); + WSTACK_PUSH3(s, tl, PRT_ONE_CONS, PRT_COMMA); } else { - WSTACK_PUSH(s, tl); - WSTACK_PUSH(s, PRT_TERM); - WSTACK_PUSH(s, PRT_BAR); + WSTACK_PUSH3(s, tl, PRT_TERM, PRT_BAR); } } } @@ -299,9 +326,7 @@ print_term(fmtfn_t fn, void* arg, Eterm obj, long *dcount, break; default: /* PRT_LAST_ARRAY_ELEMENT+1 and upwards */ obj = *popped.ptr; - WSTACK_PUSH(s, (UWord) (popped.ptr + 1)); - WSTACK_PUSH(s, val-1); - WSTACK_PUSH(s, PRT_COMMA); + WSTACK_PUSH3(s, (UWord) (popped.ptr + 1), val-1, PRT_COMMA); break; } break; @@ -431,8 +456,7 @@ print_term(fmtfn_t fn, void* arg, Eterm obj, long *dcount, WSTACK_PUSH(s,PRT_CLOSE_TUPLE); ++nobj; if (i > 0) { - WSTACK_PUSH(s, (UWord) nobj); - WSTACK_PUSH(s, PRT_LAST_ARRAY_ELEMENT+i-1); + WSTACK_PUSH2(s, (UWord) nobj, PRT_LAST_ARRAY_ELEMENT+i-1); } break; case FLOAT_DEF: { @@ -446,13 +470,65 @@ print_term(fmtfn_t fn, void* arg, Eterm obj, long *dcount, PRINT_STRING(res, fn, arg, "#MatchState"); } else { - ProcBin* pb = (ProcBin *) binary_val(wobj); - if (pb->size == 1) - PRINT_STRING(res, fn, arg, "<<1 byte>>"); - else { + byte* bytep; + Uint bytesize = binary_size_rel(obj,obj_base); + Uint bitoffs; + Uint bitsize; + byte octet; + ERTS_GET_BINARY_BYTES_REL(obj, bytep, bitoffs, bitsize, obj_base); + + if (bitsize || !bytesize + || !is_printable_ascii(bytep, bytesize, bitoffs)) { + int is_first = 1; PRINT_STRING(res, fn, arg, "<<"); - PRINT_UWORD(res, fn, arg, 'u', 0, 1, (ErlPfUWord) pb->size); - PRINT_STRING(res, fn, arg, " bytes>>"); + while (bytesize) { + if (is_first) + is_first = 0; + else + PRINT_CHAR(res, fn, arg, ','); + if (bitoffs) + octet = (bytep[0] << bitoffs) | (bytep[1] >> (8-bitoffs)); + else + octet = bytep[0]; + PRINT_UWORD(res, fn, arg, 'u', 0, 1, octet); + ++bytep; + --bytesize; + } + if (bitsize) { + Uint bits = bitoffs + bitsize; + octet = bytep[0]; + if (bits < 8) + octet >>= 8 - bits; + else if (bits > 8) { + bits -= 8; /* bits in last byte */ + octet <<= bits; + octet |= bytep[1] >> (8 - bits); + } + octet &= (1 << bitsize) - 1; + if (is_first) + is_first = 0; + else + PRINT_CHAR(res, fn, arg, ','); + PRINT_UWORD(res, fn, arg, 'u', 0, 1, octet); + PRINT_CHAR(res, fn, arg, ':'); + PRINT_UWORD(res, fn, arg, 'u', 0, 1, bitsize); + } + PRINT_STRING(res, fn, arg, ">>"); + } + else { + PRINT_STRING(res, fn, arg, "<<\""); + while (bytesize) { + if (bitoffs) + octet = (bytep[0] << bitoffs) | (bytep[1] >> (8-bitoffs)); + else + octet = bytep[0]; + if (octet == '"') + PRINT_CHAR(res, fn, arg, '\\'); + PRINT_CHAR(res, fn, arg, octet); + ++bytep; + --bytesize; + } + PRINT_STRING(res, fn, arg, "\">>"); } } break; @@ -492,33 +568,73 @@ print_term(fmtfn_t fn, void* arg, Eterm obj, long *dcount, { Uint n; Eterm *ks, *vs; - map_t *mp = (map_t *)map_val(wobj); - n = map_get_size(mp); - ks = map_get_keys(mp); - vs = map_get_values(mp); + flatmap_t *mp = (flatmap_t *)flatmap_val(wobj); + n = flatmap_get_size(mp); + ks = flatmap_get_keys(mp); + vs = flatmap_get_values(mp); PRINT_CHAR(res, fn, arg, '#'); PRINT_CHAR(res, fn, arg, '{'); WSTACK_PUSH(s, PRT_CLOSE_TUPLE); if (n > 0) { n--; - WSTACK_PUSH(s, vs[n]); - WSTACK_PUSH(s, PRT_TERM); - WSTACK_PUSH(s, PRT_ASSOC); - WSTACK_PUSH(s, ks[n]); - WSTACK_PUSH(s, PRT_TERM); - + WSTACK_PUSH5(s, vs[n], PRT_TERM, PRT_ASSOC, ks[n], PRT_TERM); while (n--) { - WSTACK_PUSH(s, PRT_COMMA); - WSTACK_PUSH(s, vs[n]); - WSTACK_PUSH(s, PRT_TERM); - WSTACK_PUSH(s, PRT_ASSOC); - WSTACK_PUSH(s, ks[n]); - WSTACK_PUSH(s, PRT_TERM); + WSTACK_PUSH6(s, PRT_COMMA, vs[n], PRT_TERM, PRT_ASSOC, + ks[n], PRT_TERM); } } } break; + case HASHMAP_DEF: + { + Uint n,mapval; + Eterm *head; + head = hashmap_val(wobj); + mapval = MAP_HEADER_VAL(*head); + switch (MAP_HEADER_TYPE(*head)) { + case MAP_HEADER_TAG_HAMT_HEAD_ARRAY: + case MAP_HEADER_TAG_HAMT_HEAD_BITMAP: + PRINT_STRING(res, fn, arg, "#<"); + PRINT_UWORD(res, fn, arg, 'x', 0, 1, mapval); + PRINT_STRING(res, fn, arg, ">{"); + WSTACK_PUSH(s,PRT_CLOSE_TUPLE); + n = hashmap_bitcount(mapval); + ASSERT(n < 17); + head += 2; + if (n > 0) { + n--; + WSTACK_PUSH(s, head[n]); + WSTACK_PUSH(s, PRT_TERM); + while (n--) { + WSTACK_PUSH(s, PRT_COMMA); + WSTACK_PUSH(s, head[n]); + WSTACK_PUSH(s, PRT_TERM); + } + } + break; + case MAP_HEADER_TAG_HAMT_NODE_BITMAP: + n = hashmap_bitcount(mapval); + head++; + PRINT_CHAR(res, fn, arg, '<'); + PRINT_UWORD(res, fn, arg, 'x', 0, 1, mapval); + PRINT_STRING(res, fn, arg, ">{"); + WSTACK_PUSH(s,PRT_CLOSE_TUPLE); + ASSERT(n < 17); + if (n > 0) { + n--; + WSTACK_PUSH(s, head[n]); + WSTACK_PUSH(s, PRT_TERM); + while (n--) { + WSTACK_PUSH(s, PRT_COMMA); + WSTACK_PUSH(s, head[n]); + WSTACK_PUSH(s, PRT_TERM); + } + } + break; + } + } + break; default: PRINT_STRING(res, fn, arg, "<unknown:"); PRINT_POINTER(res, fn, arg, wobj); @@ -528,17 +644,17 @@ print_term(fmtfn_t fn, void* arg, Eterm obj, long *dcount, } L_done: - DESTROY_WSTACK(s); return res; } + int erts_printf_term(fmtfn_t fn, void* arg, ErlPfEterm term, long precision, ErlPfEterm* term_base) { int res; - ASSERT(sizeof(ErlPfEterm) == sizeof(Eterm)); + ERTS_CT_ASSERT(sizeof(ErlPfEterm) == sizeof(Eterm)); res = print_term(fn, arg, (Eterm)term, &precision, (Eterm*)term_base); if (res < 0) diff --git a/erts/emulator/beam/erl_process.c b/erts/emulator/beam/erl_process.c index 45b6fc5fb3..f74a2ee54c 100644 --- a/erts/emulator/beam/erl_process.c +++ b/erts/emulator/beam/erl_process.c @@ -153,7 +153,7 @@ extern BeamInstr beam_continue_exit[]; /* Eager check I/O not supported on OSE yet. */ int erts_eager_check_io = 0; #else -int erts_eager_check_io = 0; +int erts_eager_check_io = 1; #endif int erts_sched_compact_load; int erts_sched_balance_util = 0; @@ -458,8 +458,7 @@ do { \ static void exec_misc_ops(ErtsRunQueue *); static void print_function_from_pc(int to, void *to_arg, BeamInstr* x); -static int stack_element_dump(int to, void *to_arg, Process* p, Eterm* sp, - int yreg); +static int stack_element_dump(int to, void *to_arg, Eterm* sp, int yreg); static void aux_work_timeout(void *unused); static void aux_work_timeout_early_init(int no_schedulers); @@ -2318,7 +2317,6 @@ erts_active_schedulers(void) ERTS_ATOMIC_FOREACH_RUNQ(rq, as -= abs(rq->waiting)); - ASSERT(as >= 0); return as; } @@ -7969,23 +7967,17 @@ erts_start_schedulers(void) Uint actual; Uint wanted = erts_no_schedulers; Uint wanted_no_schedulers = erts_no_schedulers; + char name[16]; ethr_thr_opts opts = ETHR_THR_OPTS_DEFAULT_INITER; opts.detached = 1; -#ifdef ETHR_HAVE_THREAD_NAMES - opts.name = malloc(80); - if (!opts.name) { - ERTS_INTERNAL_ERROR("malloc failed to allocate memory!"); - } -#endif + opts.name = name; #ifdef ERTS_SMP if (erts_runq_supervision_interval) { opts.suggested_stack_size = 16; -#ifdef ETHR_HAVE_THREAD_NAMES - sprintf(opts.name, "runq_supervisor"); -#endif + erts_snprintf(opts.name, 16, "runq_supervisor"); erts_atomic_init_nob(&runq_supervisor_sleeping, 0); if (0 != ethr_event_init(&runq_supervision_event)) erl_exit(1, "Failed to create run-queue supervision event\n"); @@ -8012,9 +8004,7 @@ erts_start_schedulers(void) ASSERT(actual == esdp->no - 1); -#ifdef ETHR_HAVE_THREAD_NAMES - sprintf(opts.name, "scheduler_%d", actual + 1); -#endif + erts_snprintf(opts.name, 16, "%lu_scheduler", actual + 1); #ifdef __OSE__ /* This should be done in the bind strategy */ @@ -8036,18 +8026,14 @@ erts_start_schedulers(void) int ix; for (ix = 0; ix < erts_no_dirty_cpu_schedulers; ix++) { ErtsSchedulerData *esdp = ERTS_DIRTY_CPU_SCHEDULER_IX(ix); -#ifdef ETHR_HAVE_THREAD_NAMES - sprintf(opts.name,"dirty_cpu_scheduler_%d", ix + 1); -#endif + erts_snprintf(opts.name, 16, "%d_dirty_cpu_scheduler", ix + 1); res = ethr_thr_create(&esdp->tid,sched_dirty_cpu_thread_func,(void*)esdp,&opts); if (res != 0) erl_exit(1, "Failed to create dirty cpu scheduler thread %d\n", ix); } for (ix = 0; ix < erts_no_dirty_io_schedulers; ix++) { ErtsSchedulerData *esdp = ERTS_DIRTY_IO_SCHEDULER_IX(ix); -#ifdef ETHR_HAVE_THREAD_NAMES - sprintf(opts.name,"dirty_io_scheduler_%d", ix + 1); -#endif + erts_snprintf(opts.name, 16, "%d_dirty_io_scheduler", ix + 1); res = ethr_thr_create(&esdp->tid,sched_dirty_io_thread_func,(void*)esdp,&opts); if (res != 0) erl_exit(1, "Failed to create dirty io scheduler thread %d\n", ix); @@ -8058,9 +8044,7 @@ erts_start_schedulers(void) ERTS_THR_MEMORY_BARRIER; -#ifdef ETHR_HAVE_THREAD_NAMES - sprintf(opts.name, "aux"); -#endif + erts_snprintf(opts.name, 16, "aux"); #ifdef __OSE__ opts.coreNo = 0; @@ -8086,9 +8070,6 @@ erts_start_schedulers(void) erts_send_error_to_logger_nogl(dsbufp); } -#ifdef ETHR_HAVE_THREAD_NAMES - free(opts.name); -#endif } #endif /* ERTS_SMP */ @@ -10574,7 +10555,7 @@ alloc_process(ErtsRunQueue *rq, erts_aint32_t state) init_arg.run_queue = rq; init_arg.state = state; - ASSERT(((char *) p) == ((char *) &p->common)); + ERTS_CT_ASSERT(offsetof(Process,common) == 0); if (!erts_ptab_new_element(&erts_proc, &p->common, @@ -12327,7 +12308,7 @@ erts_stack_dump(int to, void *to_arg, Process *p) } erts_program_counter_info(to, to_arg, p); for (sp = p->stop; sp < STACK_START(p); sp++) { - yreg = stack_element_dump(to, to_arg, p, sp, yreg); + yreg = stack_element_dump(to, to_arg, sp, yreg); } } @@ -12384,7 +12365,7 @@ print_function_from_pc(int to, void *to_arg, BeamInstr* x) } static int -stack_element_dump(int to, void *to_arg, Process* p, Eterm* sp, int yreg) +stack_element_dump(int to, void *to_arg, Eterm* sp, int yreg) { Eterm x = *sp; @@ -12413,6 +12394,214 @@ stack_element_dump(int to, void *to_arg, Process* p, Eterm* sp, int yreg) } /* + * Print scheduler information + */ +void +erts_print_scheduler_info(int to, void *to_arg, ErtsSchedulerData *esdp) { + int i; + erts_aint32_t flg; + Process *p; + + erts_print(to, to_arg, "=scheduler:%u\n", esdp->no); + +#ifdef ERTS_SMP + flg = erts_smp_atomic32_read_dirty(&esdp->ssi->flags); + erts_print(to, to_arg, "Scheduler Sleep Info Flags: "); + for (i = 0; i < ERTS_SSI_FLGS_MAX && flg; i++) { + erts_aint32_t chk = (1 << i); + if (flg & chk) { + switch (chk) { + case ERTS_SSI_FLG_SLEEPING: + erts_print(to, to_arg, "SLEEPING"); break; + case ERTS_SSI_FLG_POLL_SLEEPING: + erts_print(to, to_arg, "POLL_SLEEPING"); break; + case ERTS_SSI_FLG_TSE_SLEEPING: + erts_print(to, to_arg, "TSE_SLEEPING"); break; + case ERTS_SSI_FLG_WAITING: + erts_print(to, to_arg, "WAITING"); break; + case ERTS_SSI_FLG_SUSPENDED: + erts_print(to, to_arg, "SUSPENDED"); break; + default: + erts_print(to, to_arg, "UNKNOWN(%d)", flg); break; + } + if (flg > chk) + erts_print(to, to_arg, " | "); + flg -= chk; + } + } + erts_print(to, to_arg, "\n"); +#endif + + flg = erts_atomic32_read_dirty(&esdp->ssi->aux_work); + erts_print(to, to_arg, "Scheduler Sleep Info Aux Work: "); + for (i = 0; i < ERTS_SSI_AUX_WORK_MAX && flg; i++) { + erts_aint32_t chk = (1 << i); + if (flg & chk) { + switch (chk) { + case ERTS_SSI_AUX_WORK_DELAYED_AW_WAKEUP: + erts_print(to, to_arg, "DELAYED_AW_WAKEUP"); break; + case ERTS_SSI_AUX_WORK_DD: + erts_print(to, to_arg, "DELAYED_DEALLOC"); break; + case ERTS_SSI_AUX_WORK_DD_THR_PRGR: + erts_print(to, to_arg, "DELAYED_DEALLOC_THR_PRGR"); break; + case ERTS_SSI_AUX_WORK_FIX_ALLOC_DEALLOC: + erts_print(to, to_arg, "FIX_ALLOC_DEALLOC"); break; + case ERTS_SSI_AUX_WORK_FIX_ALLOC_LOWER_LIM: + erts_print(to, to_arg, "FIX_ALLOC_LOWER_LIM"); break; + case ERTS_SSI_AUX_WORK_THR_PRGR_LATER_OP: + erts_print(to, to_arg, "THR_PRGR_LATER_OP"); break; + case ERTS_SSI_AUX_WORK_ASYNC_READY: + erts_print(to, to_arg, "ASYNC_READY"); break; + case ERTS_SSI_AUX_WORK_ASYNC_READY_CLEAN: + erts_print(to, to_arg, "ASYNC_READY_CLEAN"); break; + case ERTS_SSI_AUX_WORK_MISC_THR_PRGR: + erts_print(to, to_arg, "MISC_THR_PRGR"); break; + case ERTS_SSI_AUX_WORK_MISC: + erts_print(to, to_arg, "MISC"); break; + case ERTS_SSI_AUX_WORK_CHECK_CHILDREN: + erts_print(to, to_arg, "CHECK_CHILDREN"); break; + case ERTS_SSI_AUX_WORK_SET_TMO: + erts_print(to, to_arg, "SET_TMO"); break; + case ERTS_SSI_AUX_WORK_MSEG_CACHE_CHECK: + erts_print(to, to_arg, "MSEG_CACHE_CHECK"); break; + case ERTS_SSI_AUX_WORK_REAP_PORTS: + erts_print(to, to_arg, "REAP_PORTS"); break; + default: + erts_print(to, to_arg, "UNKNOWN(%d)", flg); break; + } + if (flg > chk) + erts_print(to, to_arg, " | "); + flg -= chk; + } + } + erts_print(to, to_arg, "\n"); + + erts_print(to, to_arg, "Current Port: "); + if (esdp->current_port) + erts_print(to, to_arg, "%T", esdp->current_port->common.id); + erts_print(to, to_arg, "\n"); + + p = esdp->current_process; + erts_print(to, to_arg, "Current Process: "); + if (esdp->current_process && !(ERTS_TRACE_FLAGS(p) & F_SENSITIVE)) { + flg = erts_smp_atomic32_read_dirty(&p->state); + erts_print(to, to_arg, "%T\n", p->common.id); + + erts_print(to, to_arg, "Current Process State: "); + erts_dump_process_state(to, to_arg, flg); + + erts_print(to, to_arg, "Current Process Internal State: "); + erts_dump_extended_process_state(to, to_arg, flg); + + erts_print(to, to_arg, "Current Process Program counter: %p (", p->i); + print_function_from_pc(to, to_arg, p->i); + erts_print(to, to_arg, ")\n"); + erts_print(to, to_arg, "Current Process CP: %p (", p->cp); + print_function_from_pc(to, to_arg, p->cp); + erts_print(to, to_arg, ")\n"); + + /* Getting this stacktrace can segfault if we are very very + unlucky if called while a process is being garbage collected. + Therefore we only call this on other schedulers if we either + have protection against segfaults, or we know that the process + is not garbage collecting. It *should* always be safe to call + on a process owned by us, even if it is currently being garbage + collected. + */ + erts_print(to, to_arg, "Current Process Limited Stack Trace:\n"); + erts_limited_stack_trace(to, to_arg, p); + } else + erts_print(to, to_arg, "\n"); + + for (i = 0; i < ERTS_NO_PROC_PRIO_LEVELS; i++) { + erts_print(to, to_arg, "Run Queue "); + switch (i) { + case PRIORITY_MAX: + erts_print(to, to_arg, "Max "); + break; + case PRIORITY_HIGH: + erts_print(to, to_arg, "High "); + break; + case PRIORITY_NORMAL: + erts_print(to, to_arg, "Normal "); + break; + case PRIORITY_LOW: + erts_print(to, to_arg, "Low "); + break; + default: + erts_print(to, to_arg, "Unknown "); + break; + } + erts_print(to, to_arg, "Length: %d\n", + erts_smp_atomic32_read_dirty(&esdp->run_queue->procs.prio_info[i].len)); + } + erts_print(to, to_arg, "Run Queue Port Length: %d\n", + erts_smp_atomic32_read_dirty(&esdp->run_queue->ports.info.len)); + + flg = erts_smp_atomic32_read_dirty(&esdp->run_queue->flags); + erts_print(to, to_arg, "Run Queue Flags: "); + for (i = 0; i < ERTS_RUNQ_FLG_MAX && flg; i++) { + erts_aint32_t chk = (1 << i); + if (flg & chk) { + switch (chk) { + case (1 << PRIORITY_MAX): + erts_print(to, to_arg, "NONEMPTY_MAX"); break; + case (1 << PRIORITY_HIGH): + erts_print(to, to_arg, "NONEMPTY_HIGH"); break; + case (1 << PRIORITY_NORMAL): + erts_print(to, to_arg, "NONEMPTY_NORMAL"); break; + case (1 << PRIORITY_LOW): + erts_print(to, to_arg, "NONEMPTY_LOW"); break; + case (1 << (PRIORITY_MAX + ERTS_RUNQ_FLGS_EMIGRATE_SHFT)): + erts_print(to, to_arg, "EMIGRATE_MAX"); break; + case (1 << (PRIORITY_HIGH + ERTS_RUNQ_FLGS_EMIGRATE_SHFT)): + erts_print(to, to_arg, "EMIGRATE_HIGH"); break; + case (1 << (PRIORITY_NORMAL + ERTS_RUNQ_FLGS_EMIGRATE_SHFT)): + erts_print(to, to_arg, "EMIGRATE_NORMAL"); break; + case (1 << (PRIORITY_LOW + ERTS_RUNQ_FLGS_EMIGRATE_SHFT)): + erts_print(to, to_arg, "EMIGRATE_LOW"); break; + case (1 << (PRIORITY_MAX + ERTS_RUNQ_FLGS_IMMIGRATE_SHFT)): + erts_print(to, to_arg, "IMMIGRATE_MAX"); break; + case (1 << (PRIORITY_HIGH + ERTS_RUNQ_FLGS_IMMIGRATE_SHFT)): + erts_print(to, to_arg, "IMMIGRATE_HIGH"); break; + case (1 << (PRIORITY_NORMAL + ERTS_RUNQ_FLGS_IMMIGRATE_SHFT)): + erts_print(to, to_arg, "IMMIGRATE_NORMAL"); break; + case (1 << (PRIORITY_LOW + ERTS_RUNQ_FLGS_IMMIGRATE_SHFT)): + erts_print(to, to_arg, "IMMIGRATE_LOW"); break; + case (1 << (PRIORITY_MAX + ERTS_RUNQ_FLGS_EVACUATE_SHFT)): + erts_print(to, to_arg, "EVACUATE_MAX"); break; + case (1 << (PRIORITY_HIGH + ERTS_RUNQ_FLGS_EVACUATE_SHFT)): + erts_print(to, to_arg, "EVACUATE_HIGH"); break; + case (1 << (PRIORITY_NORMAL + ERTS_RUNQ_FLGS_EVACUATE_SHFT)): + erts_print(to, to_arg, "EVACUATE_NORMAL"); break; + case (1 << (PRIORITY_LOW + ERTS_RUNQ_FLGS_EVACUATE_SHFT)): + erts_print(to, to_arg, "EVACUATE_LOW"); break; + case ERTS_RUNQ_FLG_OUT_OF_WORK: + erts_print(to, to_arg, "OUT_OF_WORK"); break; + case ERTS_RUNQ_FLG_HALFTIME_OUT_OF_WORK: + erts_print(to, to_arg, "HALFTIME_OUT_OF_WORK"); break; + case ERTS_RUNQ_FLG_SUSPENDED: + erts_print(to, to_arg, "SUSPENDED"); break; + case ERTS_RUNQ_FLG_CHK_CPU_BIND: + erts_print(to, to_arg, "CHK_CPU_BIND"); break; + case ERTS_RUNQ_FLG_INACTIVE: + erts_print(to, to_arg, "INACTIVE"); break; + case ERTS_RUNQ_FLG_NONEMPTY: + erts_print(to, to_arg, "NONEMPTY"); break; + case ERTS_RUNQ_FLG_PROTECTED: + erts_print(to, to_arg, "PROTECTED"); break; + default: + erts_print(to, to_arg, "UNKNOWN(%d)", flg); break; + } + if (flg > chk) + erts_print(to, to_arg, " | "); + flg -= chk; + } + } + erts_print(to, to_arg, "\n"); +} + +/* * A nice system halt closing all open port goes as follows: * 1) This function schedules the aux work ERTS_SSI_AUX_WORK_REAP_PORTS * on all schedulers, then schedules itself out. diff --git a/erts/emulator/beam/erl_process.h b/erts/emulator/beam/erl_process.h index 77a4b45b09..743711cc3b 100644 --- a/erts/emulator/beam/erl_process.h +++ b/erts/emulator/beam/erl_process.h @@ -58,6 +58,7 @@ typedef struct process Process; #include "external.h" #include "erl_mseg.h" #include "erl_async.h" +#include "erl_gc.h" #ifdef HIPE #include "hipe_process.h" @@ -169,6 +170,8 @@ extern int erts_sched_thread_suggested_stack_size; #define ERTS_RUNQ_FLG_PROTECTED \ (((Uint32) 1) << (ERTS_RUNQ_FLG_BASE2 + 6)) +#define ERTS_RUNQ_FLG_MAX (ERTS_RUNQ_FLG_BASE2 + 7) + #define ERTS_RUNQ_FLGS_MIGRATION_QMASKS \ (ERTS_RUNQ_FLGS_EMIGRATE_QMASK \ | ERTS_RUNQ_FLGS_IMMIGRATE_QMASK \ @@ -251,6 +254,8 @@ typedef enum { #define ERTS_SSI_FLG_WAITING (((erts_aint32_t) 1) << 3) #define ERTS_SSI_FLG_SUSPENDED (((erts_aint32_t) 1) << 4) +#define ERTS_SSI_FLGS_MAX 5 + #define ERTS_SSI_FLGS_SLEEP_TYPE \ (ERTS_SSI_FLG_TSE_SLEEPING|ERTS_SSI_FLG_POLL_SLEEPING) @@ -282,6 +287,8 @@ typedef enum { #define ERTS_SSI_AUX_WORK_MSEG_CACHE_CHECK (((erts_aint32_t) 1) << 12) #define ERTS_SSI_AUX_WORK_REAP_PORTS (((erts_aint32_t) 1) << 13) +#define ERTS_SSI_AUX_WORK_MAX 14 + typedef struct ErtsSchedulerSleepInfo_ ErtsSchedulerSleepInfo; #ifdef ERTS_DIRTY_SCHEDULERS @@ -483,11 +490,6 @@ typedef struct { } ErtsSchedWallTime; typedef struct { - Uint64 reclaimed; - Uint64 garbage_cols; -} ErtsGCInfo; - -typedef struct { int sched; erts_aint32_t aux_work; } ErtsDelayedAuxWorkWakeupJob; @@ -1089,6 +1091,9 @@ void erts_check_for_holes(Process* p); #define ERTS_PSFLG_DIRTY_IO_PROC ERTS_PSFLG_BIT(20) #define ERTS_PSFLG_DIRTY_CPU_PROC_IN_Q ERTS_PSFLG_BIT(21) #define ERTS_PSFLG_DIRTY_IO_PROC_IN_Q ERTS_PSFLG_BIT(22) +#define ERTS_PSFLG_MAX (ERTS_PSFLGS_ZERO_BIT_OFFSET + 23) +#else +#define ERTS_PSFLG_MAX (ERTS_PSFLGS_ZERO_BIT_OFFSET + 19) #endif #define ERTS_PSFLGS_IN_PRQ_MASK (ERTS_PSFLG_IN_PRQ_MAX \ @@ -1634,7 +1639,11 @@ void erts_cleanup_empty_process(Process* p); void erts_debug_verify_clean_empty_process(Process* p); #endif void erts_stack_dump(int to, void *to_arg, Process *); +void erts_limited_stack_trace(int to, void *to_arg, Process *); void erts_program_counter_info(int to, void *to_arg, Process *); +void erts_print_scheduler_info(int to, void *to_arg, ErtsSchedulerData *esdp); +void erts_dump_extended_process_state(int to, void *to_arg, erts_aint32_t psflg); +void erts_dump_process_state(int to, void *to_arg, erts_aint32_t psflg); Eterm erts_get_process_priority(Process *p); Eterm erts_set_process_priority(Process *p, Eterm prio); diff --git a/erts/emulator/beam/erl_process_dict.c b/erts/emulator/beam/erl_process_dict.c index 23e5bf737f..00761f2d0e 100644 --- a/erts/emulator/beam/erl_process_dict.c +++ b/erts/emulator/beam/erl_process_dict.c @@ -47,7 +47,7 @@ /* Hash constant macros */ #define MAX_HASH 1342177280UL -#define INITIAL_SIZE 10 +#define INITIAL_SIZE (erts_pd_initial_size) /* Hash utility macros */ #define HASH_RANGE(PDict) ((PDict)->homeSize + (PDict)->splitPosition) @@ -82,6 +82,7 @@ static void pd_hash_erase(Process *p, Eterm id, Eterm *ret); static void pd_hash_erase_all(Process *p); static Eterm pd_hash_get_keys(Process *p, Eterm value); +static Eterm pd_hash_get_all_keys(Process *p, ProcDict *pd); static Eterm pd_hash_get_all(Process *p, ProcDict *pd); static Eterm pd_hash_put(Process *p, Eterm id, Eterm value); @@ -275,6 +276,16 @@ BIF_RETTYPE get_1(BIF_ALIST_1) BIF_RET(ret); } +BIF_RETTYPE get_keys_0(BIF_ALIST_0) +{ + Eterm ret; + + PD_CHECK(BIF_P->dictionary); + ret = pd_hash_get_all_keys(BIF_P,BIF_P->dictionary); + PD_CHECK(BIF_P->dictionary); + BIF_RET(ret); +} + BIF_RETTYPE get_keys_1(BIF_ALIST_1) { Eterm ret; @@ -412,6 +423,47 @@ Eterm erts_pd_hash_get(Process *p, Eterm id) return am_undefined; } +#define PD_GET_TKEY(Dst,Src) \ +do { \ + ASSERT(is_tuple((Src))); \ + ASSERT(arityval(*((Eterm*)tuple_val((Src)))) == 2); \ + (Dst) = ((Eterm*)tuple_val((Src)))[1]; \ +} while(0) + +static Eterm pd_hash_get_all_keys(Process *p, ProcDict *pd) { + Eterm* hp; + Eterm res = NIL; + Eterm tmp, tmp2; + unsigned int i; + unsigned int num; + + if (pd == NULL) { + return res; + } + + num = HASH_RANGE(pd); + hp = HAlloc(p, pd->numElements * 2); + + for (i = 0; i < num; ++i) { + tmp = ARRAY_GET(pd, i); + if (is_boxed(tmp)) { + PD_GET_TKEY(tmp,tmp); + res = CONS(hp, tmp, res); + hp += 2; + } else if (is_list(tmp)) { + while (tmp != NIL) { + tmp2 = TCAR(tmp); + PD_GET_TKEY(tmp2,tmp2); + res = CONS(hp, tmp2, res); + hp += 2; + tmp = TCDR(tmp); + } + } + } + return res; +} +#undef PD_GET_TKEY + static Eterm pd_hash_get_keys(Process *p, Eterm value) { Eterm *hp; diff --git a/erts/emulator/beam/erl_process_dump.c b/erts/emulator/beam/erl_process_dump.c index 2f3cf23b00..36bb6b2f0e 100644 --- a/erts/emulator/beam/erl_process_dump.c +++ b/erts/emulator/beam/erl_process_dump.c @@ -43,8 +43,9 @@ static void dump_process_info(int to, void *to_arg, Process *p); static void dump_element(int to, void *to_arg, Eterm x); static void dump_dist_ext(int to, void *to_arg, ErtsDistExternal *edep); static void dump_element_nl(int to, void *to_arg, Eterm x); -static int stack_element_dump(int to, void *to_arg, Process* p, Eterm* sp, +static int stack_element_dump(int to, void *to_arg, Eterm* sp, int yreg); +static void stack_trace_dump(int to, void *to_arg, Eterm* sp); static void print_function_from_pc(int to, void *to_arg, BeamInstr* x); static void heap_dump(int to, void *to_arg, Eterm x); static void dump_binaries(int to, void *to_arg, Binary* root); @@ -148,7 +149,7 @@ dump_process_info(int to, void *to_arg, Process *p) if ((ERTS_TRACE_FLAGS(p) & F_SENSITIVE) == 0) { erts_print(to, to_arg, "=proc_stack:%T\n", p->common.id); for (sp = p->stop; sp < STACK_START(p); sp++) { - yreg = stack_element_dump(to, to_arg, p, sp, yreg); + yreg = stack_element_dump(to, to_arg, sp, yreg); } erts_print(to, to_arg, "=proc_heap:%T\n", p->common.id); @@ -243,9 +244,65 @@ dump_element_nl(int to, void *to_arg, Eterm x) erts_putc(to, to_arg, '\n'); } +static void +stack_trace_dump(int to, void *to_arg, Eterm *sp) { + Eterm x = *sp; + if (is_CP(x)) { + erts_print(to, to_arg, "%p:", sp); + erts_print(to, to_arg, "SReturn addr 0x%X (", cp_val(x)); + print_function_from_pc(to, to_arg, cp_val(x)); + erts_print(to, to_arg, ")\n"); + } +} + +void +erts_limited_stack_trace(int to, void *to_arg, Process *p) +{ + Eterm* sp; + + + if (ERTS_TRACE_FLAGS(p) & F_SENSITIVE) { + return; + } + + if (STACK_START(p) < STACK_TOP(p)) { + return; + } + + if ((STACK_START(p) - STACK_TOP(p)) < 512) { + if (erts_sys_is_area_readable((char*)STACK_TOP(p), + (char*)STACK_START(p))) + for (sp = STACK_TOP(p); sp < STACK_START(p); sp++) + stack_trace_dump(to, to_arg, sp); + else + erts_print(to, to_arg, "Could not read from stack memory: %p - %p\n", + STACK_TOP(p), STACK_START(p)); + } else { + sp = STACK_TOP(p); + if (erts_sys_is_area_readable((char*)STACK_TOP(p), + (char*)(STACK_TOP(p) + 25))) + for (; sp < (STACK_TOP(p) + 256); sp++) + stack_trace_dump(to, to_arg, sp); + else + erts_print(to, to_arg, "Could not read from stack memory: %p - %p\n", + STACK_TOP(p), STACK_TOP(p) + 256); + + erts_print(to, to_arg, "%p: skipping %d frames\n", + sp, STACK_START(p) - STACK_TOP(p) - 512); + + if (erts_sys_is_area_readable((char*)(STACK_START(p) - 256), + (char*)STACK_START(p))) + for (sp = STACK_START(p) - 256; sp < STACK_START(p); sp++) + stack_trace_dump(to, to_arg, sp); + else + erts_print(to, to_arg, "Could not read from stack memory: %p - %p\n", + STACK_START(p) - 256, STACK_START(p)); + } + +} static int -stack_element_dump(int to, void *to_arg, Process* p, Eterm* sp, int yreg) +stack_element_dump(int to, void *to_arg, Eterm* sp, int yreg) { Eterm x = *sp; @@ -508,3 +565,114 @@ dump_externally(int to, void *to_arg, Eterm term) erts_print(to, to_arg, "%02X", *s++); } } + +void erts_dump_process_state(int to, void *to_arg, erts_aint32_t psflg) { + if (psflg & ERTS_PSFLG_FREE) + erts_print(to, to_arg, "Non Existing\n"); /* Should never happen */ + else if (psflg & ERTS_PSFLG_EXITING) + erts_print(to, to_arg, "Exiting\n"); + else if (psflg & ERTS_PSFLG_GC) { + erts_print(to, to_arg, "Garbing\n"); + } + else if (psflg & ERTS_PSFLG_SUSPENDED) + erts_print(to, to_arg, "Suspended\n"); + else if (psflg & ERTS_PSFLG_RUNNING) { + erts_print(to, to_arg, "Running\n"); + } + else if (psflg & ERTS_PSFLG_ACTIVE) + erts_print(to, to_arg, "Scheduled\n"); + else + erts_print(to, to_arg, "Waiting\n"); +} + +void +erts_dump_extended_process_state(int to, void *to_arg, erts_aint32_t psflg) { + + int i; + + switch (ERTS_PSFLGS_GET_ACT_PRIO(psflg)) { + case PRIORITY_MAX: erts_print(to, to_arg, "ACT_PRIO_MAX | "); break; + case PRIORITY_HIGH: erts_print(to, to_arg, "ACT_PRIO_HIGH | "); break; + case PRIORITY_NORMAL: erts_print(to, to_arg, "ACT_PRIO_NORMAL | "); break; + case PRIORITY_LOW: erts_print(to, to_arg, "ACT_PRIO_LOW | "); break; + } + switch (ERTS_PSFLGS_GET_USR_PRIO(psflg)) { + case PRIORITY_MAX: erts_print(to, to_arg, "USR_PRIO_MAX | "); break; + case PRIORITY_HIGH: erts_print(to, to_arg, "USR_PRIO_HIGH | "); break; + case PRIORITY_NORMAL: erts_print(to, to_arg, "USR_PRIO_NORMAL | "); break; + case PRIORITY_LOW: erts_print(to, to_arg, "USR_PRIO_LOW | "); break; + } + switch (ERTS_PSFLGS_GET_PRQ_PRIO(psflg)) { + case PRIORITY_MAX: erts_print(to, to_arg, "PRQ_PRIO_MAX"); break; + case PRIORITY_HIGH: erts_print(to, to_arg, "PRQ_PRIO_HIGH"); break; + case PRIORITY_NORMAL: erts_print(to, to_arg, "PRQ_PRIO_NORMAL"); break; + case PRIORITY_LOW: erts_print(to, to_arg, "PRQ_PRIO_LOW"); break; + } + + psflg &= ~(ERTS_PSFLGS_ACT_PRIO_MASK | + ERTS_PSFLGS_USR_PRIO_MASK | + ERTS_PSFLGS_PRQ_PRIO_MASK); + + if (psflg) + erts_print(to, to_arg, " | "); + + for (i = 0; i < ERTS_PSFLG_MAX && psflg; i++) { + erts_aint32_t chk = (1 << i); + if (psflg & chk) { + switch (chk) { + case ERTS_PSFLG_IN_PRQ_MAX: + erts_print(to, to_arg, "IN_PRQ_MAX"); break; + case ERTS_PSFLG_IN_PRQ_HIGH: + erts_print(to, to_arg, "IN_PRQ_HIGH"); break; + case ERTS_PSFLG_IN_PRQ_NORMAL: + erts_print(to, to_arg, "IN_PRQ_NORMAL"); break; + case ERTS_PSFLG_IN_PRQ_LOW: + erts_print(to, to_arg, "IN_PRQ_LOW"); break; + case ERTS_PSFLG_FREE: + erts_print(to, to_arg, "FREE"); break; + case ERTS_PSFLG_EXITING: + erts_print(to, to_arg, "EXITING"); break; + case ERTS_PSFLG_PENDING_EXIT: + erts_print(to, to_arg, "PENDING_EXIT"); break; + case ERTS_PSFLG_ACTIVE: + erts_print(to, to_arg, "ACTIVE"); break; + case ERTS_PSFLG_IN_RUNQ: + erts_print(to, to_arg, "IN_RUNQ"); break; + case ERTS_PSFLG_RUNNING: + erts_print(to, to_arg, "RUNNING"); break; + case ERTS_PSFLG_SUSPENDED: + erts_print(to, to_arg, "SUSPENDED"); break; + case ERTS_PSFLG_GC: + erts_print(to, to_arg, "GC"); break; + case ERTS_PSFLG_BOUND: + erts_print(to, to_arg, "BOUND"); break; + case ERTS_PSFLG_TRAP_EXIT: + erts_print(to, to_arg, "TRAP_EXIT"); break; + case ERTS_PSFLG_ACTIVE_SYS: + erts_print(to, to_arg, "ACTIVE_SYS"); break; + case ERTS_PSFLG_RUNNING_SYS: + erts_print(to, to_arg, "RUNNING_SYS"); break; + case ERTS_PSFLG_PROXY: + erts_print(to, to_arg, "PROXY"); break; + case ERTS_PSFLG_DELAYED_SYS: + erts_print(to, to_arg, "DELAYED_SYS"); break; +#ifdef ERTS_DIRTY_SCHEDULERS + case ERTS_PSFLG_DIRTY_CPU_PROC: + erts_print(to, to_arg, "DIRTY_CPU_PROC"); break; + case ERTS_PSFLG_DIRTY_IO_PROC: + erts_print(to, to_arg, "DIRTY_IO_PROC"); break; + case ERTS_PSFLG_DIRTY_CPU_PROC_IN_Q: + erts_print(to, to_arg, "DIRTY_CPU_PROC_IN_Q"); break; + case ERTS_PSFLG_DIRTY_IO_PROC_IN_Q: + erts_print(to, to_arg, "DIRTY_IO_PROC_IN_Q"); break; +#endif + default: + erts_print(to, to_arg, "UNKNOWN(%d)", chk); break; + } + if (psflg > chk) + erts_print(to, to_arg, " | "); + psflg -= chk; + } + } + erts_print(to, to_arg, "\n"); +} diff --git a/erts/emulator/beam/erl_term.c b/erts/emulator/beam/erl_term.c index 28cbe7004f..d6fb88ea61 100644 --- a/erts/emulator/beam/erl_term.c +++ b/erts/emulator/beam/erl_term.c @@ -86,11 +86,13 @@ unsigned tag_val_def(Wterm x) case (_TAG_HEADER_EXTERNAL_PID >> _TAG_PRIMARY_SIZE): return EXTERNAL_PID_DEF; case (_TAG_HEADER_EXTERNAL_PORT >> _TAG_PRIMARY_SIZE): return EXTERNAL_PORT_DEF; case (_TAG_HEADER_EXTERNAL_REF >> _TAG_PRIMARY_SIZE): return EXTERNAL_REF_DEF; + case (_TAG_HEADER_MAP >> _TAG_PRIMARY_SIZE): return MAP_DEF; case (_TAG_HEADER_REFC_BIN >> _TAG_PRIMARY_SIZE): return BINARY_DEF; case (_TAG_HEADER_HEAP_BIN >> _TAG_PRIMARY_SIZE): return BINARY_DEF; case (_TAG_HEADER_SUB_BIN >> _TAG_PRIMARY_SIZE): return BINARY_DEF; - case (_TAG_HEADER_MAP >> _TAG_PRIMARY_SIZE): return MAP_DEF; + case (_TAG_HEADER_HASHMAP >> _TAG_PRIMARY_SIZE): return HASHMAP_DEF; } + break; } case TAG_PRIMARY_IMMED1: { diff --git a/erts/emulator/beam/erl_term.h b/erts/emulator/beam/erl_term.h index 37014ccf94..1625a4ec15 100644 --- a/erts/emulator/beam/erl_term.h +++ b/erts/emulator/beam/erl_term.h @@ -141,6 +141,7 @@ struct erl_node_; /* Declared in erl_node_tables.h */ #define HEAP_BINARY_SUBTAG (0x9 << _TAG_PRIMARY_SIZE) /* BINARY */ #define SUB_BINARY_SUBTAG (0xA << _TAG_PRIMARY_SIZE) /* BINARY */ /* _BINARY_XXX_MASK depends on 0xB being unused */ +#define HASHMAP_SUBTAG (0xB << _TAG_PRIMARY_SIZE) /* HASHMAP */ #define EXTERNAL_PID_SUBTAG (0xC << _TAG_PRIMARY_SIZE) /* EXTERNAL_PID */ #define EXTERNAL_PORT_SUBTAG (0xD << _TAG_PRIMARY_SIZE) /* EXTERNAL_PORT */ #define EXTERNAL_REF_SUBTAG (0xE << _TAG_PRIMARY_SIZE) /* EXTERNAL_REF */ @@ -162,6 +163,7 @@ struct erl_node_; /* Declared in erl_node_tables.h */ #define _TAG_HEADER_EXTERNAL_REF (TAG_PRIMARY_HEADER|EXTERNAL_REF_SUBTAG) #define _TAG_HEADER_BIN_MATCHSTATE (TAG_PRIMARY_HEADER|BIN_MATCHSTATE_SUBTAG) #define _TAG_HEADER_MAP (TAG_PRIMARY_HEADER|MAP_SUBTAG) +#define _TAG_HEADER_HASHMAP (TAG_PRIMARY_HEADER|HASHMAP_SUBTAG) #define _TAG_HEADER_MASK 0x3F @@ -296,9 +298,11 @@ _ET_DECLARE_CHECKED(Uint,atom_val,Eterm) #define atom_val(x) _ET_APPLY(atom_val,(x)) /* header (arityval or thing) access methods */ -#define _make_header(sz,tag) ((Uint)(((sz) << _HEADER_ARITY_OFFS) + (tag))) +#define _make_header(sz,tag) ((Uint)(((Uint)(sz) << _HEADER_ARITY_OFFS) + (tag))) #define is_header(x) (((x) & _TAG_PRIMARY_MASK) == TAG_PRIMARY_HEADER) -#define _unchecked_header_arity(x) ((x) >> _HEADER_ARITY_OFFS) +//#define _unchecked_header_arity(x) ((x) >> _HEADER_ARITY_OFFS) +#define _unchecked_header_arity(x) \ + (is_hashmap_header(x) ? MAP_HEADER_ARITY(x) : ((x) >> _HEADER_ARITY_OFFS)) _ET_DECLARE_CHECKED(Uint,header_arity,Eterm) #define header_arity(x) _ET_APPLY(header_arity,(x)) @@ -361,6 +365,7 @@ _ET_DECLARE_CHECKED(Uint,thing_subtag,Eterm) ((((x) & (_TAG_HEADER_MASK)) == _TAG_HEADER_REFC_BIN) || \ (((x) & (_TAG_HEADER_MASK)) == _TAG_HEADER_HEAP_BIN) || \ (((x) & (_TAG_HEADER_MASK)) == _TAG_HEADER_SUB_BIN)) + #define make_binary(x) make_boxed((Eterm*)(x)) #define is_binary(x) (is_boxed((x)) && is_binary_header(*boxed_val((x)))) #define is_not_binary(x) (!is_binary((x))) @@ -990,6 +995,33 @@ _ET_DECLARE_CHECKED(Uint32*,external_ref_data,Wterm) _ET_DECLARE_CHECKED(struct erl_node_*,external_ref_node,Eterm) #define external_ref_node(x) _ET_APPLY(external_ref_node,(x)) +/* maps */ + +#define MAP_HEADER_TAG_SZ (2) +#define MAP_HEADER_ARITY_SZ (8) +#define MAP_HEADER_VAL_SZ (16) + +#define MAP_HEADER_TAG_FLAT (0x0) +#define MAP_HEADER_TAG_HAMT_NODE_BITMAP (0x1) +#define MAP_HEADER_TAG_HAMT_HEAD_ARRAY (0x2) +#define MAP_HEADER_TAG_HAMT_HEAD_BITMAP (0x3) + +#define MAP_HEADER_TYPE(Hdr) (((Hdr) >> (_HEADER_ARITY_OFFS)) & (0x3)) +#define MAP_HEADER_ARITY(Hdr) (((Hdr) >> (_HEADER_ARITY_OFFS + MAP_HEADER_TAG_SZ)) & (0xff)) +#define MAP_HEADER_VAL(Hdr) (((Hdr) >> (_HEADER_ARITY_OFFS + MAP_HEADER_TAG_SZ + MAP_HEADER_ARITY_SZ)) & (0xffff)) + +#define make_hashmap(x) make_boxed((Eterm*)(x)) +#define make_hashmap_rel make_boxed_rel +#define is_hashmap(x) (is_boxed((x)) && is_hashmap_header(*boxed_val((x)))) +#define is_not_hashmap(x) (!is_hashmap(x)) +#define is_hashmap_rel(RTERM,BASE) is_hashmap(rterm2wterm(RTERM,BASE)) +#define is_hashmap_header(x) (((x) & (_TAG_HEADER_MASK)) == _TAG_HEADER_HASHMAP) +#define hashmap_val(x) _unchecked_boxed_val((x)) +#define hashmap_val_rel(RTERM, BASE) hashmap_val(rterm2wterm(RTERM, BASE)) + +#define is_map(x) (is_flatmap(x) || is_hashmap(x)) +#define is_map_rel(x,BASE) (is_flatmap_rel(x,BASE) || is_hashmap_rel(x,BASE)) + /* number tests */ #define is_integer(x) (is_small(x) || is_big(x)) @@ -1081,20 +1113,23 @@ _ET_DECLARE_CHECKED(Uint,y_reg_index,Uint) #define BINARY_DEF 0x0 #define LIST_DEF 0x1 #define NIL_DEF 0x2 -#define MAP_DEF 0x3 -#define TUPLE_DEF 0x4 -#define PID_DEF 0x5 -#define EXTERNAL_PID_DEF 0x6 -#define PORT_DEF 0x7 -#define EXTERNAL_PORT_DEF 0x8 -#define EXPORT_DEF 0x9 -#define FUN_DEF 0xa -#define REF_DEF 0xb -#define EXTERNAL_REF_DEF 0xc -#define ATOM_DEF 0xd -#define FLOAT_DEF 0xe -#define BIG_DEF 0xf -#define SMALL_DEF 0x10 +#define HASHMAP_DEF 0x3 +#define MAP_DEF 0x4 +#define TUPLE_DEF 0x5 +#define PID_DEF 0x6 +#define EXTERNAL_PID_DEF 0x7 +#define PORT_DEF 0x8 +#define EXTERNAL_PORT_DEF 0x9 +#define EXPORT_DEF 0xa +#define FUN_DEF 0xb +#define REF_DEF 0xc +#define EXTERNAL_REF_DEF 0xd +#define ATOM_DEF 0xe +#define FLOAT_DEF 0xf +#define BIG_DEF 0x10 +#define SMALL_DEF 0x11 + +#define FIRST_VACANT_TAG_DEF 0x12 #if ET_DEBUG extern unsigned tag_val_def_debug(Wterm, const char*, unsigned); diff --git a/erts/emulator/beam/erl_thr_progress.c b/erts/emulator/beam/erl_thr_progress.c index f111846041..4c9b00d2ee 100644 --- a/erts/emulator/beam/erl_thr_progress.c +++ b/erts/emulator/beam/erl_thr_progress.c @@ -1335,23 +1335,10 @@ erts_thr_progress_block(void) thr_progress_block(tmp_thr_prgr_data(NULL), 1); } -void -erts_thr_progress_fatal_error_block(SWord timeout, - ErtsThrPrgrData *tmp_tpd_bufp) +int +erts_thr_progress_fatal_error_block(ErtsThrPrgrData *tmp_tpd_bufp) { ErtsThrPrgrData *tpd = perhaps_thr_prgr_data(NULL); - erts_aint32_t bc; - SWord time_left = timeout; - ErtsMonotonicTime timeout_time; - - /* - * Counting poll intervals may give us a too long timeout - * if cpu is busy. We use timeout time to try to prevent - * this. In case we havn't got time correction this may - * however fail too... - */ - timeout_time = erts_get_monotonic_time(); - timeout_time += ERTS_MSEC_TO_MONOTONIC((ErtsMonotonicTime) timeout); if (!tpd) { /* @@ -1364,9 +1351,24 @@ erts_thr_progress_fatal_error_block(SWord timeout, init_tmp_thr_prgr_data(tpd); } - bc = thr_progress_block(tpd, 0); - if (bc == 0) - return; /* Succefully blocked all managed threads */ + /* Returns number of threads that have not yes been blocked */ + return thr_progress_block(tpd, 0); +} + +void +erts_thr_progress_fatal_error_wait(SWord timeout) { + erts_aint32_t bc; + SWord time_left = timeout; + ErtsMonotonicTime timeout_time; + + /* + * Counting poll intervals may give us a too long timeout + * if cpu is busy. We use timeout time to try to prevent + * this. In case we havn't got time correction this may + * however fail too... + */ + timeout_time = erts_get_monotonic_time(); + timeout_time += ERTS_MSEC_TO_MONOTONIC((ErtsMonotonicTime) timeout); while (1) { if (erts_milli_sleep(ERTS_THR_PRGR_FTL_ERR_BLCK_POLL_INTERVAL) == 0) diff --git a/erts/emulator/beam/erl_thr_progress.h b/erts/emulator/beam/erl_thr_progress.h index 03ddbd467c..cf11c4e114 100644 --- a/erts/emulator/beam/erl_thr_progress.h +++ b/erts/emulator/beam/erl_thr_progress.h @@ -83,8 +83,8 @@ typedef struct { ErtsThrPrgrLeaderState leader_state; } ErtsThrPrgrData; -void erts_thr_progress_fatal_error_block(SWord timeout, - ErtsThrPrgrData *tmp_tpd_bufp); +int erts_thr_progress_fatal_error_block(ErtsThrPrgrData *tmp_tpd_bufp); +void erts_thr_progress_fatal_error_wait(SWord timeout); #endif /* ERTS_SMP */ diff --git a/erts/emulator/beam/erl_threads.h b/erts/emulator/beam/erl_threads.h index b9f34d4c12..dc20ac207f 100644 --- a/erts/emulator/beam/erl_threads.h +++ b/erts/emulator/beam/erl_threads.h @@ -476,6 +476,7 @@ ERTS_GLB_INLINE void erts_thr_detach(erts_tid_t tid); ERTS_GLB_INLINE void erts_thr_exit(void *res); ERTS_GLB_INLINE void erts_thr_install_exit_handler(void (*exit_handler)(void)); ERTS_GLB_INLINE erts_tid_t erts_thr_self(void); +ERTS_GLB_INLINE int erts_thr_getname(erts_tid_t tid, char *buf, size_t len); ERTS_GLB_INLINE int erts_equal_tids(erts_tid_t x, erts_tid_t y); ERTS_GLB_INLINE void erts_mtx_init_x(erts_mtx_t *mtx, char *name, Eterm extra, int enable_lcnt); @@ -658,11 +659,17 @@ ERTS_GLB_INLINE void erts_thr_set_main_status(int, int); ERTS_GLB_INLINE int erts_thr_get_main_status(void); ERTS_GLB_INLINE void erts_thr_yield(void); + #ifdef ETHR_HAVE_ETHR_SIG_FUNCS #define ERTS_THR_HAVE_SIG_FUNCS 1 ERTS_GLB_INLINE void erts_thr_sigmask(int how, const sigset_t *set, sigset_t *oset); ERTS_GLB_INLINE void erts_thr_sigwait(const sigset_t *set, int *sig); + +#ifdef USE_THREADS +ERTS_GLB_INLINE void erts_thr_kill(erts_tid_t tid, int sig); +#endif + #endif /* #ifdef HAVE_ETHR_SIG_FUNCS */ #ifdef USE_THREADS @@ -2131,6 +2138,16 @@ erts_thr_self(void) #endif } +ERTS_GLB_INLINE int +erts_thr_getname(erts_tid_t tid, char *buf, size_t len) +{ +#ifdef USE_THREADS + return ethr_getname(tid, buf, len); +#else + return -1; +#endif +} + ERTS_GLB_INLINE int erts_equal_tids(erts_tid_t x, erts_tid_t y) @@ -3540,6 +3557,15 @@ ERTS_GLB_INLINE void erts_thr_yield(void) #ifdef ETHR_HAVE_ETHR_SIG_FUNCS ERTS_GLB_INLINE void +erts_thr_kill(erts_tid_t tid, int sig) { +#ifdef USE_THREADS + int res = ethr_kill((ethr_tid)tid, sig); + if (res) + erts_thr_fatal_error(res, "killing thread"); +#endif +} + +ERTS_GLB_INLINE void erts_thr_sigmask(int how, const sigset_t *set, sigset_t *oset) { #ifdef USE_THREADS diff --git a/erts/emulator/beam/erl_time.h b/erts/emulator/beam/erl_time.h index 924b0b6091..c9cda4d10e 100644 --- a/erts/emulator/beam/erl_time.h +++ b/erts/emulator/beam/erl_time.h @@ -117,7 +117,7 @@ ERTS_GLB_INLINE ErtsMonotonicTime erts_next_timeout_time(ErtsNextTimeoutRef nxt_ /* time_sup */ -#if (defined(HAVE_GETHRVTIME) || defined(HAVE_CLOCK_GETTIME)) +#if (defined(HAVE_GETHRVTIME) || defined(HAVE_CLOCK_GETTIME_CPU_TIME)) # ifndef HAVE_ERTS_NOW_CPU # define HAVE_ERTS_NOW_CPU # ifdef HAVE_GETHRVTIME diff --git a/erts/emulator/beam/erl_trace.c b/erts/emulator/beam/erl_trace.c index ea5c850a30..2f9969b0e7 100644 --- a/erts/emulator/beam/erl_trace.c +++ b/erts/emulator/beam/erl_trace.c @@ -2225,7 +2225,7 @@ trace_gc(Process *p, Eterm what) Eterm* limit; #endif - ASSERT(sizeof(values)/sizeof(*values) == sizeof(tags)/sizeof(Eterm)); + ERTS_CT_ASSERT(sizeof(values)/sizeof(*values) == sizeof(tags)/sizeof(Eterm)); UseTmpHeap(LOCAL_HEAP_SIZE,p); @@ -3492,16 +3492,13 @@ init_sys_msg_dispatcher(void) thr_opts.coreNo = 0; #endif thr_opts.detached = 1; + thr_opts.name = "sys_msg_dispatcher"; init_smq_element_alloc(); sys_message_queue = NULL; sys_message_queue_end = NULL; erts_smp_cnd_init(&smq_cnd); erts_smp_mtx_init(&smq_mtx, "sys_msg_q"); -#ifdef ETHR_HAVE_THREAD_NAMES - thr_opts.name = "sys_msg_dispatcher"; -#endif - erts_smp_thr_create(&sys_msg_dispatcher_tid, sys_msg_dispatcher_func, NULL, diff --git a/erts/emulator/beam/erl_utils.h b/erts/emulator/beam/erl_utils.h index c32f8fd61c..7cb8972e29 100644 --- a/erts/emulator/beam/erl_utils.h +++ b/erts/emulator/beam/erl_utils.h @@ -113,12 +113,14 @@ void erts_silence_warn_unused_result(long unused); int erts_fit_in_bits_int64(Sint64); int erts_fit_in_bits_int32(Sint32); +int erts_fit_in_bits_uint(Uint); int erts_list_length(Eterm); int erts_is_builtin(Eterm, Eterm, int); Uint32 make_broken_hash(Eterm); Uint32 block_hash(byte *, unsigned, Uint32); Uint32 make_hash2(Eterm); Uint32 make_hash(Eterm); +Uint32 make_internal_hash(Eterm); void erts_save_emu_args(int argc, char **argv); Eterm erts_get_emu_args(struct process *c_p); diff --git a/erts/emulator/beam/erl_vm.h b/erts/emulator/beam/erl_vm.h index b7de8208ad..3a9fb1e07b 100644 --- a/erts/emulator/beam/erl_vm.h +++ b/erts/emulator/beam/erl_vm.h @@ -20,8 +20,6 @@ #ifndef __ERL_VM_H__ #define __ERL_VM_H__ -/* #define ERTS_OPCODE_COUNTER_SUPPORT */ - /* FORCE_HEAP_FRAGS: * Debug provocation to make HAlloc always create heap fragments (if allowed) * even if there is room on heap. @@ -119,9 +117,9 @@ #if defined(DEBUG) || defined(CHECK_FOR_HOLES) #if HALFWORD_HEAP -# define ERTS_HOLE_MARKER (0xaf5e78ccU) +# define ERTS_HOLE_MARKER (0xdeadbeef) #else -# define ERTS_HOLE_MARKER (((0xaf5e78ccUL << 24) << 8) | 0xaf5e78ccUL) +# define ERTS_HOLE_MARKER (((0xdeadbeef << 24) << 8) | 0xdeadbeef) #endif #endif @@ -174,6 +172,7 @@ extern int H_MIN_SIZE; /* minimum (heap + stack) */ extern int BIN_VH_MIN_SIZE; /* minimum virtual (bin) heap */ extern int erts_atom_table_size;/* Atom table size */ +extern int erts_pd_initial_size;/* Initial Process dictionary table size */ #define ORIG_CREATION 0 #define INTERNAL_CREATION 255 diff --git a/erts/emulator/beam/external.c b/erts/emulator/beam/external.c index 9b9b4b2a62..82c60840e5 100644 --- a/erts/emulator/beam/external.c +++ b/erts/emulator/beam/external.c @@ -36,7 +36,9 @@ #include "erl_process.h" #include "error.h" #include "external.h" +#define ERL_WANT_HIPE_BIF_WRAPPER__ #include "bif.h" +#undef ERL_WANT_HIPE_BIF_WRAPPER__ #include "big.h" #include "dist.h" #include "erl_binary.h" @@ -498,15 +500,37 @@ byte *erts_encode_ext_dist_header_finalize(byte *ext, ErtsAtomCache *cache, Uint return ep; } -Uint erts_encode_dist_ext_size(Eterm term, Uint32 flags, ErtsAtomCacheMap *acmp) +int erts_encode_dist_ext_size(Eterm term, Uint32 flags, ErtsAtomCacheMap *acmp, + Uint* szp) { - Uint sz = 0; + Uint sz; + if (encode_size_struct_int(NULL, acmp, term, flags, NULL, &sz)) { + return -1; + } else { #ifndef ERTS_DEBUG_USE_DIST_SEP - if (!(flags & DFLAG_DIST_HDR_ATOM_CACHE)) + if (!(flags & DFLAG_DIST_HDR_ATOM_CACHE)) #endif - sz++ /* VERSION_MAGIC */; - sz += encode_size_struct2(acmp, term, flags); - return sz; + sz++ /* VERSION_MAGIC */; + + *szp += sz; + return 0; + } +} + +int erts_encode_dist_ext_size_int(Eterm term, struct erts_dsig_send_context* ctx, Uint* szp) +{ + Uint sz; + if (encode_size_struct_int(&ctx->u.sc, ctx->acmp, term, ctx->flags, &ctx->reds, &sz)) { + return -1; + } else { +#ifndef ERTS_DEBUG_USE_DIST_SEP + if (!(ctx->flags & DFLAG_DIST_HDR_ATOM_CACHE)) +#endif + sz++ /* VERSION_MAGIC */; + + *szp += sz; + return 0; + } } Uint erts_encode_ext_size(Eterm term) @@ -527,19 +551,16 @@ Uint erts_encode_ext_size_ets(Eterm term) } -void erts_encode_dist_ext(Eterm term, byte **ext, Uint32 flags, ErtsAtomCacheMap *acmp) +int erts_encode_dist_ext(Eterm term, byte **ext, Uint32 flags, ErtsAtomCacheMap *acmp, + TTBEncodeContext* ctx, Sint* reds) { - byte *ep = *ext; -#ifndef ERTS_DEBUG_USE_DIST_SEP - if (!(flags & DFLAG_DIST_HDR_ATOM_CACHE)) -#endif - *ep++ = VERSION_MAGIC; - ep = enc_term(acmp, term, ep, flags, NULL); - if (!ep) - erl_exit(ERTS_ABORT_EXIT, - "%s:%d:erts_encode_dist_ext(): Internal data structure error\n", - __FILE__, __LINE__); - *ext = ep; + if (!ctx || !ctx->wstack.wstart) { + #ifndef ERTS_DEBUG_USE_DIST_SEP + if (!(flags & DFLAG_DIST_HDR_ATOM_CACHE)) + #endif + *(*ext)++ = VERSION_MAGIC; + } + return enc_term_int(ctx, acmp, term, *ext, flags, NULL, reds, ext); } void erts_encode_ext(Eterm term, byte **ext) @@ -1161,7 +1182,8 @@ typedef struct { Eterm* hp_end; int remaining_n; char* remaining_bytes; - Eterm* maps_head; + Eterm* maps_list; + struct dec_term_hamt_placeholder* hamt_list; } B2TDecodeContext; typedef struct { @@ -1487,7 +1509,8 @@ static BIF_RETTYPE binary_to_term_int(Process* p, Uint32 flags, Eterm bin, Binar ctx->u.dc.hp_start = HAlloc(p, ctx->heap_size); ctx->u.dc.hp = ctx->u.dc.hp_start; ctx->u.dc.hp_end = ctx->u.dc.hp_start + ctx->heap_size; - ctx->u.dc.maps_head = NULL; + ctx->u.dc.maps_list = NULL; + ctx->u.dc.hamt_list = NULL; ctx->state = B2TDecode; /*fall through*/ case B2TDecode: @@ -1740,54 +1763,14 @@ erts_term_to_binary(Process* p, Eterm Term, int level, Uint flags) { return erts_term_to_binary_simple(p, Term, size, level, flags); } -/* Define for testing */ -/* #define EXTREME_TTB_TRAPPING 1 */ +/* Define EXTREME_TTB_TRAPPING for testing in dist.h */ #ifndef EXTREME_TTB_TRAPPING -#define TERM_TO_BINARY_LOOP_FACTOR 32 #define TERM_TO_BINARY_COMPRESS_CHUNK (1 << 18) #else -#define TERM_TO_BINARY_LOOP_FACTOR 1 #define TERM_TO_BINARY_COMPRESS_CHUNK 10 #endif - - -typedef enum { TTBSize, TTBEncode, TTBCompress } TTBState; -typedef struct TTBSizeContext_ { - Uint flags; - int level; - Uint result; - Eterm obj; - ErtsEStack estack; -} TTBSizeContext; - -typedef struct TTBEncodeContext_ { - Uint flags; - int level; - byte* ep; - Eterm obj; - ErtsWStack wstack; - Binary *result_bin; -} TTBEncodeContext; - -typedef struct { - Uint real_size; - Uint dest_len; - byte *dbytes; - Binary *result_bin; - Binary *destination_bin; - z_stream stream; -} TTBCompressContext; - -typedef struct { - int alive; - TTBState state; - union { - TTBSizeContext sc; - TTBEncodeContext ec; - TTBCompressContext cc; - } s; -} TTBContext; +#define TERM_TO_BINARY_MEMCPY_FACTOR 8 static void ttb_context_destructor(Binary *context_bin) { @@ -1899,8 +1882,6 @@ static Eterm erts_term_to_binary_int(Process* p, Eterm Term, int level, Uint fla } result_bin = erts_bin_nrml_alloc(size); - result_bin->flags = 0; - result_bin->orig_size = size; erts_refc_init(&result_bin->refc, 0); result_bin->orig_bytes[0] = VERSION_MAGIC; /* Next state immediately, no need to export context */ @@ -1925,7 +1906,6 @@ static Eterm erts_term_to_binary_int(Process* p, Eterm Term, int level, Uint fla } real_size = endp - bytes; result_bin = erts_bin_realloc(context->s.ec.result_bin,real_size); - result_bin->orig_size = real_size; level = context->s.ec.level; BUMP_REDS(p, (initial_reds - reds) / TERM_TO_BINARY_LOOP_FACTOR); if (level == 0 || real_size < 6) { /* We are done */ @@ -1962,8 +1942,6 @@ static Eterm erts_term_to_binary_int(Process* p, Eterm Term, int level, Uint fla context->s.cc.result_bin = result_bin; result_bin = erts_bin_nrml_alloc(real_size); - result_bin->flags = 0; - result_bin->orig_size = real_size; erts_refc_init(&result_bin->refc, 0); result_bin->orig_bytes[0] = VERSION_MAGIC; @@ -2005,7 +1983,6 @@ static Eterm erts_term_to_binary_int(Process* p, Eterm Term, int level, Uint fla erl_zlib_deflate_finish(&(context->s.cc.stream)); result_bin = erts_bin_realloc(context->s.cc.destination_bin, context->s.cc.dest_len+6); - result_bin->orig_size = context->s.cc.dest_len+6; context->s.cc.destination_bin = NULL; pb = (ProcBin *) HAlloc(p, PROC_BIN_SIZE); pb->thing_word = HEADER_PROC_BIN; @@ -2327,8 +2304,10 @@ dec_pid(ErtsDistExternal *edep, Eterm** hpp, byte* ep, ErlOffHeap* off_heap, Ete #define ENC_TERM ((Eterm) 0) #define ENC_ONE_CONS ((Eterm) 1) #define ENC_PATCH_FUN_SIZE ((Eterm) 2) -#define ENC_LAST_ARRAY_ELEMENT ((Eterm) 3) - +#define ENC_BIN_COPY ((Eterm) 3) +#define ENC_MAP_PAIR ((Eterm) 4) +#define ENC_HASHMAP_NODE ((Eterm) 5) +#define ENC_LAST_ARRAY_ELEMENT ((Eterm) 6) static byte* enc_term(ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, Uint32 dflags, @@ -2364,6 +2343,9 @@ enc_term_int(TTBEncodeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, WSTACK_RESTORE(s, &ctx->wstack); ep = ctx->ep; obj = ctx->obj; + if (is_non_value(obj)) { + goto outer_loop; + } } } @@ -2387,8 +2369,8 @@ enc_term_int(TTBEncodeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, obj = CAR(cons); tl = CDR(cons); - WSTACK_PUSH(s, is_list(tl) ? ENC_ONE_CONS : ENC_TERM); - WSTACK_PUSH(s, tl); + WSTACK_PUSH2(s, (is_list(tl) ? ENC_ONE_CONS : ENC_TERM), + tl); } break; case ENC_PATCH_FUN_SIZE: @@ -2401,6 +2383,46 @@ enc_term_int(TTBEncodeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, put_int32(ep - size_p, size_p); } goto outer_loop; + case ENC_BIN_COPY: { + Uint bits = (Uint)obj; + Uint bitoffs = WSTACK_POP(s); + byte* bytes = (byte*) WSTACK_POP(s); + byte* dst = (byte*) WSTACK_POP(s); + if (bits > r * (TERM_TO_BINARY_MEMCPY_FACTOR * 8)) { + Uint n = r * TERM_TO_BINARY_MEMCPY_FACTOR; + WSTACK_PUSH5(s, (UWord)(dst + n), (UWord)(bytes + n), bitoffs, + ENC_BIN_COPY, bits - 8*n); + bits = 8*n; + copy_binary_to_buffer(dst, 0, bytes, bitoffs, bits); + obj = THE_NON_VALUE; + r = 0; /* yield */ + break; + } else { + copy_binary_to_buffer(dst, 0, bytes, bitoffs, bits); + r -= bits / (TERM_TO_BINARY_MEMCPY_FACTOR * 8); + goto outer_loop; + } + } + case ENC_MAP_PAIR: { + Uint pairs_left = obj; + Eterm *vptr = (Eterm*) WSTACK_POP(s); + Eterm *kptr = (Eterm*) WSTACK_POP(s); + + obj = *kptr; + if (--pairs_left > 0) { + WSTACK_PUSH4(s, (UWord)(kptr+1), (UWord)(vptr+1), + ENC_MAP_PAIR, pairs_left); + } + WSTACK_PUSH2(s, ENC_TERM, *vptr); + break; + } + case ENC_HASHMAP_NODE: + if (is_list(obj)) { /* leaf node [K|V] */ + ptr = list_val(obj); + WSTACK_PUSH2(s, ENC_TERM, CDR(ptr)); + obj = CAR(ptr); + } + break; case ENC_LAST_ARRAY_ELEMENT: /* obj is the tuple */ { @@ -2419,17 +2441,16 @@ enc_term_int(TTBEncodeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, #else Eterm* ptr = (Eterm *) obj; #endif - WSTACK_PUSH(s, val-1); obj = *ptr++; - WSTACK_PUSH(s, (UWord)ptr); + WSTACK_PUSH2(s, val-1, (UWord)ptr); } break; } L_jump_start: - if (ctx && --r == 0) { - *reds = r; + if (ctx && --r <= 0) { + *reds = 0; ctx->obj = obj; ctx->ep = ep; WSTACK_SAVE(s, &ctx->wstack); @@ -2578,35 +2599,62 @@ enc_term_int(TTBEncodeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, ep += 4; } if (i > 0) { - WSTACK_PUSH(s, ENC_LAST_ARRAY_ELEMENT+i-1); - WSTACK_PUSH(s, (UWord)ptr); + WSTACK_PUSH2(s, ENC_LAST_ARRAY_ELEMENT+i-1, (UWord)ptr); } break; case MAP_DEF: { - map_t *mp = (map_t*)map_val(obj); - Uint size = map_get_size(mp); + flatmap_t *mp = (flatmap_t*)flatmap_val(obj); + Uint size = flatmap_get_size(mp); *ep++ = MAP_EXT; put_int32(size, ep); ep += 4; if (size > 0) { - Eterm *kptr = map_get_keys(mp); - Eterm *vptr = map_get_values(mp); - - for (i = size-1; i >= 1; i--) { - WSTACK_PUSH(s, ENC_TERM); - WSTACK_PUSH(s, (UWord) vptr[i]); - WSTACK_PUSH(s, ENC_TERM); - WSTACK_PUSH(s, (UWord) kptr[i]); - } + Eterm *kptr = flatmap_get_keys(mp); + Eterm *vptr = flatmap_get_values(mp); - WSTACK_PUSH(s, ENC_TERM); - WSTACK_PUSH(s, (UWord) vptr[0]); + WSTACK_PUSH4(s, (UWord)kptr, (UWord)vptr, + ENC_MAP_PAIR, size); + } + } + break; - obj = kptr[0]; - goto L_jump_start; + case HASHMAP_DEF: + { + Eterm hdr; + Uint node_sz; + ptr = boxed_val(obj); + hdr = *ptr; + ASSERT(is_header(hdr)); + switch(hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_HEAD_ARRAY: + *ep++ = MAP_EXT; + ptr++; + put_int32(*ptr, ep); ep += 4; + /*fall through*/ + case HAMT_SUBTAG_NODE_ARRAY: + node_sz = 16; + break; + case HAMT_SUBTAG_HEAD_BITMAP: + *ep++ = MAP_EXT; + ptr++; + put_int32(*ptr, ep); ep += 4; + /*fall through*/ + case HAMT_SUBTAG_NODE_BITMAP: + node_sz = hashmap_bitcount(MAP_HEADER_VAL(hdr)); + ASSERT(node_sz < 17); + break; + default: + erl_exit(1, "bad header\r\n"); + } + + ptr++; + WSTACK_RESERVE(s, node_sz*2); + while(node_sz--) { + WSTACK_FAST_PUSH(s, ENC_HASHMAP_NODE); + WSTACK_FAST_PUSH(s, *ptr++); } } break; @@ -2644,6 +2692,7 @@ enc_term_int(TTBEncodeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, Uint bitoffs; Uint bitsize; byte* bytes; + byte* data_dst; ERTS_GET_BINARY_BYTES(obj, bytes, bitoffs, bitsize); if (dflags & DFLAG_INTERNAL_TAGS) { @@ -2689,7 +2738,7 @@ enc_term_int(TTBEncodeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, j = binary_size(obj); put_int32(j, ep); ep += 4; - copy_binary_to_buffer(ep, 0, bytes, bitoffs, 8*j); + data_dst = ep; ep += j; } else if (dflags & DFLAG_BIT_BINARIES) { /* Bit-level binary. */ @@ -2699,7 +2748,7 @@ enc_term_int(TTBEncodeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, ep += 4; *ep++ = bitsize; ep[j] = 0; /* Zero unused bits at end of binary */ - copy_binary_to_buffer(ep, 0, bytes, bitoffs, 8*j+bitsize); + data_dst = ep; ep += j + 1; } else { /* @@ -2713,11 +2762,18 @@ enc_term_int(TTBEncodeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, put_int32((j+1), ep); ep += 4; ep[j] = 0; /* Zero unused bits at end of binary */ - copy_binary_to_buffer(ep, 0, bytes, bitoffs, 8*j+bitsize); + data_dst = ep; ep += j+1; *ep++ = SMALL_INTEGER_EXT; *ep++ = bitsize; } + if (ctx && j > r * TERM_TO_BINARY_MEMCPY_FACTOR) { + WSTACK_PUSH5(s, (UWord)data_dst, (UWord)bytes, bitoffs, + ENC_BIN_COPY, 8*j + bitsize); + } else { + copy_binary_to_buffer(data_dst, 0, bytes, bitoffs, + 8 * j + bitsize); + } } break; case EXPORT_DEF: @@ -2746,13 +2802,12 @@ enc_term_int(TTBEncodeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, case FUN_DEF: { ErlFunThing* funp = (ErlFunThing *) fun_val(obj); + int ei; if ((dflags & DFLAG_NEW_FUN_TAGS) != 0) { - int ei; - *ep++ = NEW_FUN_EXT; - WSTACK_PUSH(s, ENC_PATCH_FUN_SIZE); - WSTACK_PUSH(s, (UWord) ep); /* Position for patching in size */ + WSTACK_PUSH2(s, ENC_PATCH_FUN_SIZE, + (UWord) ep); /* Position for patching in size */ ep += 4; *ep = funp->arity; ep += 1; @@ -2766,16 +2821,6 @@ enc_term_int(TTBEncodeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, ep = enc_term(acmp, make_small(funp->fe->old_index), ep, dflags, off_heap); ep = enc_term(acmp, make_small(funp->fe->old_uniq), ep, dflags, off_heap); ep = enc_pid(acmp, funp->creator, ep, dflags); - - fun_env: - for (ei = funp->num_free-1; ei > 0; ei--) { - WSTACK_PUSH(s, ENC_TERM); - WSTACK_PUSH(s, (UWord) funp->env[ei]); - } - if (funp->num_free != 0) { - obj = funp->env[0]; - goto L_jump_start; - } } else { /* * Communicating with an obsolete erl_interface or @@ -2807,7 +2852,13 @@ enc_term_int(TTBEncodeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, *ep++ = SMALL_TUPLE_EXT; put_int8(funp->num_free, ep); ep += 1; - goto fun_env; + } + for (ei = funp->num_free-1; ei > 0; ei--) { + WSTACK_PUSH2(s, ENC_TERM, (UWord) funp->env[ei]); + } + if (funp->num_free != 0) { + obj = funp->env[0]; + goto L_jump_start; } } break; @@ -2889,9 +2940,19 @@ undo_offheap_in_area(ErlOffHeap* off_heap, Eterm* start, Eterm* end) #endif /* DEBUG */ } +struct dec_term_hamt_placeholder +{ + struct dec_term_hamt_placeholder* next; + Eterm* objp; /* write result here */ + Uint size; /* nr of leafs */ + Eterm leafs[1]; +}; + +#define DEC_TERM_HAMT_PLACEHOLDER_SIZE \ + (offsetof(struct dec_term_hamt_placeholder, leafs) / sizeof(Eterm)) /* Decode term from external format into *objp. -** On failure return NULL and (R13B04) *hpp will be unchanged. +** On failure return NULL and *hpp will be unchanged. */ static byte* dec_term(ErtsDistExternal *edep, Eterm** hpp, byte* ep, ErlOffHeap* off_heap, @@ -2901,7 +2962,8 @@ dec_term(ErtsDistExternal *edep, Eterm** hpp, byte* ep, ErlOffHeap* off_heap, int n; ErtsAtomEncoding char_enc; register Eterm* hp; /* Please don't take the address of hp */ - Eterm *maps_head; /* for validation of maps */ + Eterm *maps_list; /* for preprocessing of small maps */ + struct dec_term_hamt_placeholder* hamt_list; /* for preprocessing of big maps */ Eterm* next; SWord reds; @@ -2911,7 +2973,8 @@ dec_term(ErtsDistExternal *edep, Eterm** hpp, byte* ep, ErlOffHeap* off_heap, next = ctx->u.dc.next; ep = ctx->u.dc.ep; hpp = &ctx->u.dc.hp; - maps_head = ctx->u.dc.maps_head; + maps_list = ctx->u.dc.maps_list; + hamt_list = ctx->u.dc.hamt_list; if (ctx->state != B2TDecode) { int n_limit = reds; @@ -2992,7 +3055,8 @@ dec_term(ErtsDistExternal *edep, Eterm** hpp, byte* ep, ErlOffHeap* off_heap, reds = ERTS_SWORD_MAX; next = objp; *next = (Eterm) (UWord) NULL; - maps_head = NULL; + maps_list = NULL; + hamt_list = NULL; } hp = *hpp; @@ -3056,6 +3120,8 @@ dec_term(ErtsDistExternal *edep, Eterm** hpp, byte* ep, ErlOffHeap* off_heap, big = make_small(0); } else { big = bytes_to_big(first, n, neg, hp); + if (is_nil(big)) + goto error; if (is_big(big)) { hp += big_arity(big) + 1; } @@ -3387,8 +3453,6 @@ dec_term_atom_common: } else { Binary* dbin = erts_bin_nrml_alloc(n); ProcBin* pb; - dbin->flags = 0; - dbin->orig_size = n; erts_refc_init(&dbin->refc, 1); pb = (ProcBin *) hp; hp += PROC_BIN_SIZE; @@ -3396,6 +3460,7 @@ dec_term_atom_common: pb->size = n; pb->next = off_heap->first; off_heap->first = (struct erl_off_heap_header*)pb; + OH_OVERHEAD(off_heap, pb->size / sizeof(Eterm)); pb->val = dbin; pb->bytes = (byte*) dbin->orig_bytes; pb->flags = 0; @@ -3441,14 +3506,13 @@ dec_term_atom_common: Binary* dbin = erts_bin_nrml_alloc(n); ProcBin* pb; - dbin->flags = 0; - dbin->orig_size = n; erts_refc_init(&dbin->refc, 1); pb = (ProcBin *) hp; pb->thing_word = HEADER_PROC_BIN; pb->size = n; pb->next = off_heap->first; off_heap->first = (struct erl_off_heap_header*)pb; + OH_OVERHEAD(off_heap, pb->size / sizeof(Eterm)); pb->val = dbin; pb->bytes = (byte*) dbin->orig_bytes; pb->flags = 0; @@ -3530,46 +3594,67 @@ dec_term_atom_common: break; case MAP_EXT: { - map_t *mp; Uint32 size,n; Eterm *kptr,*vptr; Eterm keys; size = get_int32(ep); ep += 4; - keys = make_tuple(hp); - *hp++ = make_arityval(size); - hp += size; - kptr = hp - 1; - - mp = (map_t*)hp; - hp += MAP_HEADER_SIZE; - hp += size; - vptr = hp - 1; - - /* kptr, last word for keys - * vptr, last word for values - */ - - /* - * Use thing_word to link through decoded maps. - * The list of maps is for later validation. - */ - - mp->thing_word = (Eterm) COMPRESS_POINTER(maps_head); - maps_head = (Eterm *) mp; - - mp->size = size; - mp->keys = keys; - *objp = make_map(mp); - - for (n = size; n; n--) { - *vptr = (Eterm) COMPRESS_POINTER(next); - *kptr = (Eterm) COMPRESS_POINTER(vptr); - next = kptr; - vptr--; - kptr--; - } + if (size <= MAP_SMALL_MAP_LIMIT) { + flatmap_t *mp; + + keys = make_tuple(hp); + *hp++ = make_arityval(size); + hp += size; + kptr = hp - 1; + + mp = (flatmap_t*)hp; + hp += MAP_HEADER_SIZE; + hp += size; + vptr = hp - 1; + + /* kptr, last word for keys + * vptr, last word for values + */ + + /* + * Use thing_word to link through decoded maps. + * The list of maps is for later validation. + */ + + mp->thing_word = (Eterm) COMPRESS_POINTER(maps_list); + maps_list = (Eterm *) mp; + + mp->size = size; + mp->keys = keys; + *objp = make_flatmap(mp); + + for (n = size; n; n--) { + *vptr = (Eterm) COMPRESS_POINTER(next); + *kptr = (Eterm) COMPRESS_POINTER(vptr); + next = kptr; + vptr--; + kptr--; + } + } + else { /* Make hamt */ + struct dec_term_hamt_placeholder* holder = + (struct dec_term_hamt_placeholder*) hp; + + holder->next = hamt_list; + hamt_list = holder; + holder->objp = objp; + holder->size = size; + + hp += DEC_TERM_HAMT_PLACEHOLDER_SIZE; + + for (n = size; n; n--) { + CDR(hp) = (Eterm) COMPRESS_POINTER(next); + CAR(hp) = (Eterm) COMPRESS_POINTER(&CDR(hp)); + next = &CAR(hp); + hp += 2; + } + } } break; case NEW_FUN_EXT: @@ -3747,6 +3832,7 @@ dec_term_atom_common: hp += PROC_BIN_SIZE; pb->next = off_heap->first; off_heap->first = (struct erl_off_heap_header*)pb; + OH_OVERHEAD(off_heap, pb->size / sizeof(Eterm)); pb->flags = 0; *objp = make_binary(pb); break; @@ -3764,6 +3850,7 @@ dec_term_atom_common: hp += PROC_BIN_SIZE; pb->next = off_heap->first; off_heap->first = (struct erl_off_heap_header*)pb; + OH_OVERHEAD(off_heap, pb->size / sizeof(Eterm)); pb->flags = 0; sub = (ErlSubBin*)hp; @@ -3790,7 +3877,7 @@ dec_term_atom_common: ctx->u.dc.ep = ep; ctx->u.dc.next = next; ctx->u.dc.hp = hp; - ctx->u.dc.maps_head = maps_head; + ctx->u.dc.maps_list = maps_list; ctx->reds = 0; return NULL; } @@ -3805,12 +3892,40 @@ dec_term_atom_common: * - done here for when we know it is complete. */ - while (maps_head) { - next = (Eterm *)(EXPAND_POINTER(*maps_head)); - *maps_head = MAP_HEADER; - if (!erts_validate_and_sort_map((map_t*)maps_head)) + while (maps_list) { + next = (Eterm *)(EXPAND_POINTER(*maps_list)); + *maps_list = MAP_HEADER; + if (!erts_validate_and_sort_flatmap((flatmap_t*)maps_list)) goto error; - maps_head = next; + maps_list = next; + } + + /* Iterate through all the hamts and build tree nodes. + */ + if (hamt_list) { + ErtsHeapFactory factory; + + factory.p = NULL; + factory.hp = hp; + /* We assume heap will suffice (see hashmap_over_estimated_heap_size) */ + + do { + struct dec_term_hamt_placeholder* hamt = hamt_list; + *hamt->objp = erts_hashmap_from_array(&factory, + hamt->leafs, + hamt->size, + 1); + if (is_non_value(*hamt->objp)) + goto error; + + hamt_list = hamt->next; + + /* Yes, we waste a couple of heap words per hamt + for the temporary placeholder */ + *(Eterm*)hamt = make_pos_bignum_header(DEC_TERM_HAMT_PLACEHOLDER_SIZE-1); + } while (hamt_list); + + hp = factory.hp; } if (ctx) { @@ -4008,15 +4123,15 @@ encode_size_struct_int(TTBSizeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, break; case MAP_DEF: { - map_t *mp = (map_t*)map_val(obj); - Uint size = map_get_size(mp); + flatmap_t *mp = (flatmap_t*)flatmap_val(obj); + Uint size = flatmap_get_size(mp); Uint i; Eterm *ptr; result += 1 + 4; /* tag + 4 bytes size */ /* push values first */ - ptr = map_get_values(mp); + ptr = flatmap_get_values(mp); i = size; while(i--) { if (is_list(*ptr)) { @@ -4030,7 +4145,7 @@ encode_size_struct_int(TTBSizeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, ++ptr; } - ptr = map_get_keys(mp); + ptr = flatmap_get_keys(mp); i = size; while(i--) { if (is_list(*ptr)) { @@ -4046,6 +4161,38 @@ encode_size_struct_int(TTBSizeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, goto outer_loop; } break; + + case HASHMAP_DEF: + { + Eterm *ptr; + Eterm hdr; + Uint node_sz; + ptr = boxed_val(obj); + hdr = *ptr; + ASSERT(is_header(hdr)); + switch(hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_HEAD_ARRAY: ptr++; + case HAMT_SUBTAG_NODE_ARRAY: + node_sz = 16; + break; + case HAMT_SUBTAG_HEAD_BITMAP: ptr++; + case HAMT_SUBTAG_NODE_BITMAP: + node_sz = hashmap_bitcount(MAP_HEADER_VAL(hdr)); + ASSERT(node_sz < 17); + break; + default: + erl_exit(1, "bad header\r\n"); + } + + ptr++; + ESTACK_RESERVE(s, node_sz); + while(node_sz--) { + ESTACK_FAST_PUSH(s, *ptr++); + } + result += 1 + 4; /* tag + 4 bytes size */ + } + + break; case FLOAT_DEF: if (dflags & DFLAG_NEW_FLOATS) { result += 9; @@ -4148,6 +4295,8 @@ encode_size_struct_int(TTBSizeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, return 0; } + + static Sint decoded_size(byte *ep, byte* endp, int internal_tags, B2TContext* ctx) { @@ -4341,7 +4490,11 @@ init_done: n = get_int32(ep); ep += 4; ADDTERMS(2*n); - heap_size += 3 + n + 1 + n; + if (n <= MAP_SMALL_MAP_LIMIT) { + heap_size += 3 + n + 1 + n; + } else { + heap_size += hashmap_over_estimated_heap_size(n); + } break; case STRING_EXT: CHKSIZE(2); diff --git a/erts/emulator/beam/external.h b/erts/emulator/beam/external.h index bf00958eb1..50fcfa04d6 100644 --- a/erts/emulator/beam/external.h +++ b/erts/emulator/beam/external.h @@ -1,7 +1,7 @@ /* * %CopyrightBegin% * - * Copyright Ericsson AB 1996-2013. All Rights Reserved. + * Copyright Ericsson AB 1996-2014. All Rights Reserved. * * The contents of this file are subject to the Erlang Public License, * Version 1.1, (the "License"); you may not use this file except in @@ -150,19 +150,23 @@ typedef struct { Uint extsize; } ErtsBinary2TermState; + /* -------------------------------------------------------------------------- */ void erts_init_atom_cache_map(ErtsAtomCacheMap *); void erts_reset_atom_cache_map(ErtsAtomCacheMap *); void erts_destroy_atom_cache_map(ErtsAtomCacheMap *); void erts_finalize_atom_cache_map(ErtsAtomCacheMap *, Uint32); -Uint erts_encode_ext_dist_header_size(ErtsAtomCacheMap *); Uint erts_encode_ext_dist_header_size(ErtsAtomCacheMap *); byte *erts_encode_ext_dist_header_setup(byte *, ErtsAtomCacheMap *); byte *erts_encode_ext_dist_header_finalize(byte *, ErtsAtomCache *, Uint32); -Uint erts_encode_dist_ext_size(Eterm, Uint32, ErtsAtomCacheMap *); -void erts_encode_dist_ext(Eterm, byte **, Uint32, ErtsAtomCacheMap *); +struct erts_dsig_send_context; +int erts_encode_dist_ext_size(Eterm, Uint32, ErtsAtomCacheMap*, Uint* szp); +int erts_encode_dist_ext_size_int(Eterm term, struct erts_dsig_send_context* ctx, Uint* szp); +struct TTBEncodeContext_; +int erts_encode_dist_ext(Eterm, byte **, Uint32, ErtsAtomCacheMap *, + struct TTBEncodeContext_ *, Sint* reds); Uint erts_encode_ext_size(Eterm); Uint erts_encode_ext_size_2(Eterm, unsigned); diff --git a/erts/emulator/beam/global.h b/erts/emulator/beam/global.h index 96c21d5320..634fe533d0 100644 --- a/erts/emulator/beam/global.h +++ b/erts/emulator/beam/global.h @@ -41,6 +41,7 @@ #include "error.h" #include "erl_utils.h" #include "erl_port.h" +#include "erl_gc.h" struct enif_environment_t /* ErlNifEnv */ { @@ -160,6 +161,7 @@ struct erts_driver_t_ { void (*ready_async)(ErlDrvData drv_data, ErlDrvThreadData thread_data); /* Might be NULL */ void (*process_exit)(ErlDrvData drv_data, ErlDrvMonitor *monitor); void (*stop_select)(ErlDrvEvent event, void*); /* Might be NULL */ + void (*emergency_close)(ErlDrvData drv_data); /* Might be NULL */ }; extern erts_driver_t *driver_list; @@ -368,16 +370,17 @@ extern int stackdump_on_exit; * DESTROY_ESTACK(Stack) */ -typedef struct { +typedef struct ErtsEStack_ { Eterm* start; Eterm* sp; Eterm* end; + Eterm* edefault; ErtsAlcType_t alloc_type; }ErtsEStack; #define DEF_ESTACK_SIZE (16) -void erl_grow_estack(ErtsEStack*, Eterm* def_stack); +void erl_grow_estack(ErtsEStack*, Uint need); #define ESTK_CONCAT(a,b) a##b #define ESTK_DEF_STACK(s) ESTK_CONCAT(s,_default_estack) @@ -387,22 +390,23 @@ void erl_grow_estack(ErtsEStack*, Eterm* def_stack); ESTK_DEF_STACK(s), /* start */ \ ESTK_DEF_STACK(s), /* sp */ \ ESTK_DEF_STACK(s) + DEF_ESTACK_SIZE, /* end */ \ + ESTK_DEF_STACK(s), /* default */ \ ERTS_ALC_T_ESTACK /* alloc_type */ \ } #define ESTACK_CHANGE_ALLOCATOR(s,t) \ do { \ - if (s.start != ESTK_DEF_STACK(s)) { \ + if ((s).start != ESTK_DEF_STACK(s)) { \ erl_exit(1, "Internal error - trying to change allocator " \ "type of active estack\n"); \ } \ - s.alloc_type = (t); \ + (s).alloc_type = (t); \ } while (0) #define DESTROY_ESTACK(s) \ do { \ - if (s.start != ESTK_DEF_STACK(s)) { \ - erts_free(s.alloc_type, s.start); \ + if ((s).start != ESTK_DEF_STACK(s)) { \ + erts_free((s).alloc_type, (s).start); \ } \ } while(0) @@ -413,16 +417,17 @@ do { \ */ #define ESTACK_SAVE(s,dst)\ do {\ - if (s.start == ESTK_DEF_STACK(s)) {\ + if ((s).start == ESTK_DEF_STACK(s)) {\ UWord _wsz = ESTACK_COUNT(s);\ - (dst)->start = erts_alloc(s.alloc_type,\ + (dst)->start = erts_alloc((s).alloc_type,\ DEF_ESTACK_SIZE * sizeof(Eterm));\ - memcpy((dst)->start, s.start,_wsz*sizeof(Eterm));\ + memcpy((dst)->start, (s).start,_wsz*sizeof(Eterm));\ (dst)->sp = (dst)->start + _wsz;\ (dst)->end = (dst)->start + DEF_ESTACK_SIZE;\ - (dst)->alloc_type = s.alloc_type;\ + (dst)->edefault = NULL;\ + (dst)->alloc_type = (s).alloc_type;\ } else\ - *(dst) = s;\ + *(dst) = (s);\ } while (0) #define DESTROY_SAVED_ESTACK(estack)\ @@ -441,72 +446,114 @@ do {\ */ #define ESTACK_RESTORE(s, src) \ do { \ - ASSERT(s.start == ESTK_DEF_STACK(s)); \ - s = *(src); /* struct copy */ \ + ASSERT((s).start == ESTK_DEF_STACK(s)); \ + (s) = *(src); /* struct copy */ \ (src)->start = NULL; \ - ASSERT(s.sp >= s.start); \ - ASSERT(s.sp <= s.end); \ + ASSERT((s).sp >= (s).start); \ + ASSERT((s).sp <= (s).end); \ } while (0) -#define ESTACK_IS_STATIC(s) (s.start == ESTK_DEF_STACK(s))) +#define ESTACK_IS_STATIC(s) ((s).start == ESTK_DEF_STACK(s)) -#define ESTACK_PUSH(s, x) \ -do { \ - if (s.sp == s.end) { \ - erl_grow_estack(&s, ESTK_DEF_STACK(s)); \ - } \ - *s.sp++ = (x); \ +#define ESTACK_PUSH(s, x) \ +do { \ + if ((s).sp == (s).end) { \ + erl_grow_estack(&(s), 1); \ + } \ + *(s).sp++ = (x); \ } while(0) #define ESTACK_PUSH2(s, x, y) \ do { \ - if (s.sp > s.end - 2) { \ - erl_grow_estack(&s, ESTK_DEF_STACK(s)); \ + if ((s).sp > (s).end - 2) { \ + erl_grow_estack(&(s), 2); \ } \ - *s.sp++ = (x); \ - *s.sp++ = (y); \ + *(s).sp++ = (x); \ + *(s).sp++ = (y); \ } while(0) #define ESTACK_PUSH3(s, x, y, z) \ do { \ - if (s.sp > s.end - 3) { \ - erl_grow_estack(&s, ESTK_DEF_STACK(s)); \ + if ((s).sp > (s).end - 3) { \ + erl_grow_estack(&s, 3); \ } \ - *s.sp++ = (x); \ - *s.sp++ = (y); \ - *s.sp++ = (z); \ + *(s).sp++ = (x); \ + *(s).sp++ = (y); \ + *(s).sp++ = (z); \ +} while(0) + +#define ESTACK_PUSH4(s, E1, E2, E3, E4) \ +do { \ + if ((s).sp > (s).end - 4) { \ + erl_grow_estack(&s, 4); \ + } \ + *(s).sp++ = (E1); \ + *(s).sp++ = (E2); \ + *(s).sp++ = (E3); \ + *(s).sp++ = (E4); \ +} while(0) + +#define ESTACK_RESERVE(s, push_cnt) \ +do { \ + if ((s).sp > (s).end - (push_cnt)) { \ + erl_grow_estack(&(s), (push_cnt)); \ + } \ +} while(0) + +/* Must be preceded by ESTACK_RESERVE */ +#define ESTACK_FAST_PUSH(s, x) \ +do { \ + ASSERT((s).sp < (s).end); \ + *s.sp++ = (x); \ } while(0) -#define ESTACK_COUNT(s) (s.sp - s.start) -#define ESTACK_ISEMPTY(s) (s.sp == s.start) -#define ESTACK_POP(s) (*(--s.sp)) +#define ESTACK_COUNT(s) ((s).sp - (s).start) +#define ESTACK_ISEMPTY(s) ((s).sp == (s).start) +#define ESTACK_POP(s) (*(--(s).sp)) /* * WSTACK: same as ESTACK but with UWord instead of Eterm */ -typedef struct { +typedef struct ErtsWStack_ { UWord* wstart; UWord* wsp; UWord* wend; + UWord* wdefault; ErtsAlcType_t alloc_type; }ErtsWStack; #define DEF_WSTACK_SIZE (16) -void erl_grow_wstack(ErtsWStack*, UWord* def_stack); +void erl_grow_wstack(ErtsWStack*, Uint need); #define WSTK_CONCAT(a,b) a##b #define WSTK_DEF_STACK(s) WSTK_CONCAT(s,_default_wstack) -#define DECLARE_WSTACK(s) \ +#define WSTACK_DECLARE(s) \ UWord WSTK_DEF_STACK(s)[DEF_WSTACK_SIZE]; \ ErtsWStack s = { \ WSTK_DEF_STACK(s), /* wstart */ \ WSTK_DEF_STACK(s), /* wsp */ \ WSTK_DEF_STACK(s) + DEF_WSTACK_SIZE, /* wend */ \ + WSTK_DEF_STACK(s), /* wdflt */ \ ERTS_ALC_T_ESTACK /* alloc_type */ \ } +#define DECLARE_WSTACK WSTACK_DECLARE + +typedef struct ErtsDynamicWStack_ { + UWord default_stack[DEF_WSTACK_SIZE]; + ErtsWStack ws; +}ErtsDynamicWStack; + +#define WSTACK_INIT(dwsp, ALC_TYPE) \ +do { \ + (dwsp)->ws.wstart = (dwsp)->default_stack; \ + (dwsp)->ws.wsp = (dwsp)->default_stack; \ + (dwsp)->ws.wend = (dwsp)->default_stack + DEF_WSTACK_SIZE;\ + (dwsp)->ws.wdefault = (dwsp)->default_stack; \ + (dwsp)->ws.alloc_type = ALC_TYPE; \ +} while (0) #define WSTACK_CHANGE_ALLOCATOR(s,t) \ do { \ @@ -517,13 +564,20 @@ do { \ s.alloc_type = (t); \ } while (0) -#define DESTROY_WSTACK(s) \ +#define WSTACK_DESTROY(s) \ do { \ - if (s.wstart != WSTK_DEF_STACK(s)) { \ + if (s.wstart != s.wdefault) { \ erts_free(s.alloc_type, s.wstart); \ } \ } while(0) +#define DESTROY_WSTACK WSTACK_DESTROY +#define WSTACK_DEBUG(s) \ + do { \ + fprintf(stderr, "wstack size = %ld\r\n", s.wsp - s.wstart); \ + fprintf(stderr, "wstack wstart = %p\r\n", s.wstart); \ + fprintf(stderr, "wstack wsp = %p\r\n", s.wsp); \ + } while(0) /* * Do not free the stack after this, it may have pointers into what @@ -538,6 +592,7 @@ do {\ memcpy((dst)->wstart, s.wstart,_wsz*sizeof(UWord));\ (dst)->wsp = (dst)->wstart + _wsz;\ (dst)->wend = (dst)->wstart + DEF_WSTACK_SIZE;\ + (dst)->wdefault = NULL;\ (dst)->alloc_type = s.alloc_type;\ } else\ *(dst) = s;\ @@ -566,12 +621,12 @@ do { \ ASSERT(s.wsp <= s.wend); \ } while (0) -#define WSTACK_IS_STATIC(s) (s.wstart == WSTK_DEF_STACK(s))) +#define WSTACK_IS_STATIC(s) (s.wstart == WSTK_DEF_STACK(s)) #define WSTACK_PUSH(s, x) \ do { \ if (s.wsp == s.wend) { \ - erl_grow_wstack(&s, WSTK_DEF_STACK(s)); \ + erl_grow_wstack(&s, 1); \ } \ *s.wsp++ = (x); \ } while(0) @@ -579,7 +634,7 @@ do { \ #define WSTACK_PUSH2(s, x, y) \ do { \ if (s.wsp > s.wend - 2) { \ - erl_grow_wstack(&s, WSTK_DEF_STACK(s)); \ + erl_grow_wstack(&s, 2); \ } \ *s.wsp++ = (x); \ *s.wsp++ = (y); \ @@ -587,17 +642,129 @@ do { \ #define WSTACK_PUSH3(s, x, y, z) \ do { \ - if (s.wsp > s.wend - 3) { \ - erl_grow_wstack(&s, WSTK_DEF_STACK(s)); \ + if (s.wsp > s.wend - 3) { \ + erl_grow_wstack(&s, 3); \ } \ *s.wsp++ = (x); \ *s.wsp++ = (y); \ *s.wsp++ = (z); \ } while(0) +#define WSTACK_PUSH4(s, A1, A2, A3, A4) \ +do { \ + if (s.wsp > s.wend - 4) { \ + erl_grow_wstack(&s, 4); \ + } \ + *s.wsp++ = (A1); \ + *s.wsp++ = (A2); \ + *s.wsp++ = (A3); \ + *s.wsp++ = (A4); \ +} while(0) + +#define WSTACK_PUSH5(s, A1, A2, A3, A4, A5) \ +do { \ + if (s.wsp > s.wend - 5) { \ + erl_grow_wstack(&s, 5); \ + } \ + *s.wsp++ = (A1); \ + *s.wsp++ = (A2); \ + *s.wsp++ = (A3); \ + *s.wsp++ = (A4); \ + *s.wsp++ = (A5); \ +} while(0) + +#define WSTACK_PUSH6(s, A1, A2, A3, A4, A5, A6) \ +do { \ + if (s.wsp > s.wend - 6) { \ + erl_grow_wstack(&s, 6); \ + } \ + *s.wsp++ = (A1); \ + *s.wsp++ = (A2); \ + *s.wsp++ = (A3); \ + *s.wsp++ = (A4); \ + *s.wsp++ = (A5); \ + *s.wsp++ = (A6); \ +} while(0) + +#define WSTACK_RESERVE(s, push_cnt) \ +do { \ + if (s.wsp > s.wend - (push_cnt)) { \ + erl_grow_wstack(&s, (push_cnt)); \ + } \ +} while(0) + +/* Must be preceded by WSTACK_RESERVE */ +#define WSTACK_FAST_PUSH(s, x) \ +do { \ + ASSERT(s.wsp < s.wend); \ + *s.wsp++ = (x); \ +} while(0) + #define WSTACK_COUNT(s) (s.wsp - s.wstart) #define WSTACK_ISEMPTY(s) (s.wsp == s.wstart) -#define WSTACK_POP(s) (*(--s.wsp)) +#define WSTACK_POP(s) ((ASSERT(s.wsp > s.wstart)),*(--s.wsp)) + +#define WSTACK_ROLLBACK(s, count) (ASSERT(WSTACK_COUNT(s) >= (count)), \ + s.wsp = s.wstart + (count)) + +/* PSTACK - Stack of any type. + * Usage: + * { + * #define PSTACK_TYPE MyType + * PSTACK_DECLARE(s,16); + * MyType *sp = PSTACK_PUSH(s); + * + * sp->x = .... + * sp->y = .... + * sp = PSTACK_PUSH(s); + * ... + * sp = PSTACK_POP(s); + * if (PSTACK_IS_EMPTY(s)) { + * // sp is invalid when stack is empty after pop + * } + * + * PSTACK_DESTROY(s); + * } + */ + + +typedef struct ErtsPStack_ { + byte* pstart; + byte* psp; + byte* pend; + ErtsAlcType_t alloc_type; +}ErtsPStack; + +void erl_grow_pstack(ErtsPStack* s, void* default_pstack, unsigned need_bytes); +#define PSTK_CONCAT(a,b) a##b +#define PSTK_DEF_STACK(s) PSTK_CONCAT(s,_default_pstack) + +#define PSTACK_DECLARE(s, DEF_PSTACK_SIZE) \ +PSTACK_TYPE PSTK_DEF_STACK(s)[DEF_PSTACK_SIZE]; \ +ErtsPStack s = { (byte*)PSTK_DEF_STACK(s), /* pstart */ \ + (byte*)(PSTK_DEF_STACK(s) - 1), /* psp */ \ + (byte*)(PSTK_DEF_STACK(s) + (DEF_PSTACK_SIZE)), /* pend */\ + ERTS_ALC_T_ESTACK /* alloc_type */ \ +} + +#define PSTACK_DESTROY(s) \ +do { \ + if (s.pstart != (byte*)PSTK_DEF_STACK(s)) { \ + erts_free(s.alloc_type, s.pstart); \ + } \ +} while(0) + +#define PSTACK_IS_EMPTY(s) (s.psp < s.pstart) + +#define PSTACK_TOP(s) (ASSERT(!PSTACK_IS_EMPTY(s)), (PSTACK_TYPE*)(s.psp)) + +#define PSTACK_PUSH(s) \ + (s.psp += sizeof(PSTACK_TYPE), \ + ((s.psp == s.pend) ? erl_grow_pstack(&s, PSTK_DEF_STACK(s), \ + sizeof(PSTACK_TYPE)) : (void)0), \ + ((PSTACK_TYPE*) s.psp)) + +#define PSTACK_POP(s) ((PSTACK_TYPE*) (s.psp -= sizeof(PSTACK_TYPE))) /* binary.c */ @@ -804,23 +971,6 @@ void MD5Init(MD5_CTX *); void MD5Update(MD5_CTX *, unsigned char *, unsigned int); void MD5Final(unsigned char [16], MD5_CTX *); -/* ggc.c */ - -void erts_gc_info(ErtsGCInfo *gcip); -void erts_init_gc(void); -int erts_garbage_collect(Process*, int, Eterm*, int); -void erts_garbage_collect_hibernate(Process* p); -Eterm erts_gc_after_bif_call(Process* p, Eterm result, Eterm* regs, Uint arity); -void erts_garbage_collect_literals(Process* p, Eterm* literals, - Uint lit_size, - struct erl_off_heap_header* oh); -Uint erts_next_heap_size(Uint, Uint); -Eterm erts_heap_sizes(Process* p); - -void erts_offset_off_heap(ErlOffHeap *, Sint, Eterm*, Eterm*); -void erts_offset_heap_ptr(Eterm*, Uint, Sint, Eterm*, Eterm*); -void erts_offset_heap(Eterm*, Uint, Sint, Eterm*, Eterm*); -void erts_free_heap_frags(Process* p); /* io.c */ @@ -847,6 +997,7 @@ Uint erts_port_ioq_size(Port *pp); void erts_stale_drv_select(Eterm, ErlDrvPort, ErlDrvEvent, int, int); Port *erts_get_heart_port(void); +void erts_emergency_close_ports(void); #if defined(ERTS_SMP) && defined(ERTS_ENABLE_LOCK_COUNT) void erts_lcnt_enable_io_lock_count(int enable); diff --git a/erts/emulator/beam/io.c b/erts/emulator/beam/io.c index c0e44b239b..8ffe718334 100644 --- a/erts/emulator/beam/io.c +++ b/erts/emulator/beam/io.c @@ -392,7 +392,7 @@ static Port *create_port(char *name, /* Set default tracing */ erts_get_default_tracing(&ERTS_TRACE_FLAGS(prt), &ERTS_TRACER_PROC(prt)); - ASSERT(((char *) prt) == ((char *) &prt->common)); + ERTS_CT_ASSERT(offsetof(Port,common) == 0); #if !ERTS_PORT_INIT_INSTR_NEED_ID /* @@ -3154,8 +3154,6 @@ static void deliver_read_message(Port* prt, erts_aint32_t state, Eterm to, Binary* bptr; bptr = erts_bin_nrml_alloc(len); - bptr->flags = 0; - bptr->orig_size = len; erts_refc_init(&bptr->refc, 1); sys_memcpy(bptr->orig_bytes, buf, len); @@ -4088,6 +4086,9 @@ erts_port_control(Process* c_p, size, &resp_bufp, &resp_size); + + control_flags = prt->control_flags; + finalize_imm_drv_call(&try_call_state); if (tmp_alloced) erts_free(ERTS_ALC_T_TMP, bufp); @@ -4095,8 +4096,6 @@ erts_port_control(Process* c_p, return ERTS_PORT_OP_BADARG; } - control_flags = prt->control_flags; - hsz = port_control_result_size(control_flags, resp_bufp, &resp_size, @@ -5352,7 +5351,11 @@ driver_deliver_term(Eterm to, ErlDrvTermData* data, int len) case ERL_DRV_MAP: { /* int */ ERTS_DDT_CHK_ENOUGH_ARGS(1); if ((int) ptr[0] < 0) ERTS_DDT_FAIL; - need += MAP_HEADER_SIZE + 1 + 2*ptr[0]; + if (ptr[0] > MAP_SMALL_MAP_LIMIT) { + need += hashmap_over_estimated_heap_size(ptr[0]); + } else { + need += MAP_HEADER_SIZE + 1 + 2*ptr[0]; + } depth -= 2*ptr[0]; if (depth < 0) ERTS_DDT_FAIL; ptr++; @@ -5507,8 +5510,6 @@ driver_deliver_term(Eterm to, ErlDrvTermData* data, int len) ProcBin* pbp; Binary* bp = erts_bin_nrml_alloc(size); ASSERT(bufp); - bp->flags = 0; - bp->orig_size = (SWord) size; erts_refc_init(&bp->refc, 1); sys_memcpy((void *) bp->orig_bytes, (void *) bufp, size); pbp = (ProcBin *) hp; @@ -5598,31 +5599,52 @@ driver_deliver_term(Eterm to, ErlDrvTermData* data, int len) case ERL_DRV_MAP: { /* int */ int size = (int)ptr[0]; - Eterm* tp = hp; - Eterm* vp; - map_t *mp; - - *tp = make_arityval(size); - - hp += 1 + size; - mp = (map_t*)hp; - mp->thing_word = MAP_HEADER; - mp->size = size; - mp->keys = make_tuple(tp); - mess = make_map(mp); - - hp += MAP_HEADER_SIZE + size; /* advance "heap" pointer */ - - tp += size; /* point at last key */ - vp = hp - 1; /* point at last value */ - - while(size--) { - *vp-- = ESTACK_POP(stack); - *tp-- = ESTACK_POP(stack); - } - if (!erts_validate_and_sort_map(mp)) - ERTS_DDT_FAIL; - ptr++; + if (size > MAP_SMALL_MAP_LIMIT) { + int ix = 2*size; + ErtsHeapFactory factory; + Eterm* leafs = hp; + + hp += 2*size; + while(ix--) { *--hp = ESTACK_POP(stack); } + + hp += 2*size; + factory.p = NULL; + factory.hp = hp; + /* We assume heap will suffice (see hashmap_over_estimated_heap_size) */ + + mess = erts_hashmap_from_array(&factory, leafs, size, 1); + + if (is_non_value(mess)) + ERTS_DDT_FAIL; + + hp = factory.hp; + } else { + Eterm* tp = hp; + Eterm* vp; + flatmap_t *mp; + + *tp = make_arityval(size); + + hp += 1 + size; + mp = (flatmap_t*)hp; + mp->thing_word = MAP_HEADER; + mp->size = size; + mp->keys = make_tuple(tp); + mess = make_flatmap(mp); + + hp += MAP_HEADER_SIZE + size; /* advance "heap" pointer */ + + tp += size; /* point at last key */ + vp = hp - 1; /* point at last value */ + + while(size--) { + *vp-- = ESTACK_POP(stack); + *tp-- = ESTACK_POP(stack); + } + if (!erts_validate_and_sort_flatmap(mp)) + ERTS_DDT_FAIL; + } + ptr++; break; } @@ -6000,9 +6022,7 @@ driver_alloc_binary(ErlDrvSizeT size) bin = erts_bin_drv_alloc_fnf((Uint) size); if (!bin) return NULL; /* The driver write must take action */ - bin->flags = BIN_FLAG_DRV; erts_refc_init(&bin->refc, 1); - bin->orig_size = (SWord) size; return Binary2ErlDrvBinary(bin); } @@ -6032,7 +6052,6 @@ ErlDrvBinary* driver_realloc_binary(ErlDrvBinary* bin, ErlDrvSizeT size) if (!newbin) return NULL; - newbin->orig_size = size; return Binary2ErlDrvBinary(newbin); } @@ -6706,7 +6725,7 @@ static void ref_to_driver_monitor(Eterm ref, ErlDrvMonitor *mon) { RefThing *refp; ASSERT(is_internal_ref(ref)); - ASSERT(sizeof(RefThing) <= sizeof(ErlDrvMonitor)); + ERTS_CT_ASSERT(sizeof(RefThing) <= sizeof(ErlDrvMonitor)); refp = ref_thing_ptr(ref); memset(mon,0,sizeof(ErlDrvMonitor)); memcpy(mon,refp,sizeof(RefThing)); @@ -7350,6 +7369,8 @@ no_stop_select_callback(ErlDrvEvent event, void* private) erts_send_error_to_logger_nogl(dsbufp); } +#define IS_DRIVER_VERSION_GE(DE,MAJOR,MINOR) \ + ((DE)->major_version >= (MAJOR) && (DE)->minor_version >= (MINOR)) static int init_driver(erts_driver_t *drv, ErlDrvEntry *de, DE_Handle *handle) @@ -7397,6 +7418,7 @@ init_driver(erts_driver_t *drv, ErlDrvEntry *de, DE_Handle *handle) drv->timeout = de->timeout ? de->timeout : no_timeout_callback; drv->ready_async = de->ready_async; drv->process_exit = de->process_exit; + drv->emergency_close = IS_DRIVER_VERSION_GE(de,3,2) ? de->emergency_close : NULL; if (de->stop_select) drv->stop_select = de->stop_select; else @@ -7415,6 +7437,8 @@ init_driver(erts_driver_t *drv, ErlDrvEntry *de, DE_Handle *handle) } } +#undef IS_DRIVER_VERSION_GE + void erts_destroy_driver(erts_driver_t *drv) { @@ -7558,7 +7582,7 @@ Port *erts_get_heart_port(void) if (!port) continue; /* only examine undead or alive ports */ - if (erts_atomic32_read_nob(&port->state) & ERTS_PORT_SFLGS_DEAD) + if (erts_atomic32_read_nob(&port->state) & ERTS_PORT_SFLGS_INVALID_DRIVER_LOOKUP) continue; /* immediate atom compare */ reg = port->common.u.alive.reg; @@ -7569,3 +7593,23 @@ Port *erts_get_heart_port(void) return NULL; } + +void erts_emergency_close_ports(void) +{ + int ix, max = erts_ptab_max(&erts_port); + + for (ix = 0; ix < max; ix++) { + Port *port = erts_pix2port(ix); + + if (!port) + continue; + /* only examine undead or alive ports */ + if (erts_atomic32_read_nob(&port->state) & ERTS_PORT_SFLGS_INVALID_DRIVER_LOOKUP) + continue; + + /* emergency close socket */ + if (port->drv_ptr->emergency_close) { + port->drv_ptr->emergency_close((ErlDrvData) port->drv_data); + } + } +} diff --git a/erts/emulator/beam/ops.tab b/erts/emulator/beam/ops.tab index 68fcc177ae..d3649080dc 100644 --- a/erts/emulator/beam/ops.tab +++ b/erts/emulator/beam/ops.tab @@ -166,22 +166,26 @@ is_tuple Fail=f S | select_tuple_arity S=d Fail=f Size=u Rest=* => \ select_tuple_arity S=d Fail=f Size=u Rest=* => \ gen_select_tuple_arity(S, Fail, Size, Rest) -i_select_val r f I -i_select_val x f I -i_select_val y f I +i_select_val_bins r f I +i_select_val_bins x f I +i_select_val_bins y f I -i_select_val2 r f c f c f -i_select_val2 x f c f c f -i_select_val2 y f c f c f +i_select_val_lins r f I +i_select_val_lins x f I +i_select_val_lins y f I -i_select_tuple_arity2 r f A f A f -i_select_tuple_arity2 x f A f A f -i_select_tuple_arity2 y f A f A f +i_select_val2 r f c c f f +i_select_val2 x f c c f f +i_select_val2 y f c c f f i_select_tuple_arity r f I i_select_tuple_arity x f I i_select_tuple_arity y f I +i_select_tuple_arity2 r f A A f f +i_select_tuple_arity2 x f A A f f +i_select_tuple_arity2 y f A A f f + i_jump_on_val_zero r f I i_jump_on_val_zero x f I i_jump_on_val_zero y f I diff --git a/erts/emulator/beam/sys.h b/erts/emulator/beam/sys.h index a108e819c6..a0f35fef1b 100644 --- a/erts/emulator/beam/sys.h +++ b/erts/emulator/beam/sys.h @@ -191,6 +191,32 @@ __decl_noreturn void __noreturn erl_assert_error(const char* expr, const char *f # define ASSERT(e) ((void) 1) #endif +/* ERTS_UNDEF can be used to silence false warnings about + * "variable may be used uninitialized" while keeping the variable + * marked as undefined by valgrind. + */ +#ifdef VALGRIND +# define ERTS_UNDEF(V,I) +#else +# define ERTS_UNDEF(V,I) V = I +#endif + +/* + * Compile time assert + * (the actual compiler error msg can be a bit confusing) + */ +#if ERTS_AT_LEAST_GCC_VSN__(3,1,1) +# define ERTS_CT_ASSERT(e) \ + do { \ + enum { compile_time_assert__ = __builtin_choose_expr((e),0,(void)0) }; \ + } while(0) +#else +# define ERTS_CT_ASSERT(e) \ + do { \ + enum { compile_time_assert__ = 1/(e) }; \ + } while (0) +#endif + /* * Microsoft C/C++: We certainly want to use stdarg.h and prototypes. * But MSC doesn't define __STDC__, unless we compile with the -Za @@ -810,6 +836,8 @@ typedef struct { } ErtsCheckIoDebugInfo; int erts_check_io_debug(ErtsCheckIoDebugInfo *ip); +int erts_sys_is_area_readable(char *start, char *stop); + /* xxxP */ #define SYS_DEFAULT_FLOAT_DECIMALS 20 void init_sys_float(void); @@ -838,6 +866,11 @@ int erts_sys_unsetenv(char *key); char *erts_read_env(char *key); void erts_free_read_env(void *value); +#if defined(ERTS_THR_HAVE_SIG_FUNCS) && !defined(ETHR_UNUSABLE_SIGUSRX) +extern void sys_thr_resume(erts_tid_t tid); +extern void sys_thr_suspend(erts_tid_t tid); +#endif + /* utils.c */ /* Options to sys_alloc_opt */ diff --git a/erts/emulator/beam/utils.c b/erts/emulator/beam/utils.c index da03960b59..127f1e4a6a 100644 --- a/erts/emulator/beam/utils.c +++ b/erts/emulator/beam/utils.c @@ -50,6 +50,9 @@ #include "erl_ptab.h" #include "erl_check_io.h" #include "erl_bif_unique.h" +#ifdef HIPE +# include "hipe_mode_switch.h" +#endif #undef M_TRIM_THRESHOLD #undef M_TOP_PAD @@ -188,12 +191,18 @@ erts_set_hole_marker(Eterm* ptr, Uint sz) * Helper function for the ESTACK macros defined in global.h. */ void -erl_grow_estack(ErtsEStack* s, Eterm* default_estack) +erl_grow_estack(ErtsEStack* s, Uint need) { Uint old_size = (s->end - s->start); - Uint new_size = old_size * 2; + Uint new_size; Uint sp_offs = s->sp - s->start; - if (s->start != default_estack) { + + if (need < old_size) + new_size = 2*old_size; + else + new_size = ((need / old_size) + 2) * old_size; + + if (s->start != s->edefault) { s->start = erts_realloc(s->alloc_type, s->start, new_size*sizeof(Eterm)); } else { @@ -208,12 +217,18 @@ erl_grow_estack(ErtsEStack* s, Eterm* default_estack) * Helper function for the WSTACK macros defined in global.h. */ void -erl_grow_wstack(ErtsWStack* s, UWord* default_wstack) +erl_grow_wstack(ErtsWStack* s, Uint need) { Uint old_size = (s->wend - s->wstart); - Uint new_size = old_size * 2; + Uint new_size; Uint sp_offs = s->wsp - s->wstart; - if (s->wstart != default_wstack) { + + if (need < old_size) + new_size = 2 * old_size; + else + new_size = ((need / old_size) + 2) * old_size; + + if (s->wstart != s->wdefault) { s->wstart = erts_realloc(s->alloc_type, s->wstart, new_size*sizeof(UWord)); } else { @@ -225,6 +240,32 @@ erl_grow_wstack(ErtsWStack* s, UWord* default_wstack) s->wsp = s->wstart + sp_offs; } +/* + * Helper function for the PSTACK macros defined in global.h. + */ +void +erl_grow_pstack(ErtsPStack* s, void* default_pstack, unsigned need_bytes) +{ + Uint old_size = s->pend - s->pstart; + Uint new_size; + Uint sp_offs = s->psp - s->pstart; + + if (need_bytes < old_size) + new_size = 2 * old_size; + else + new_size = ((need_bytes / old_size) + 2) * old_size; + + if (s->pstart != default_pstack) { + s->pstart = erts_realloc(s->alloc_type, s->pstart, new_size); + } else { + byte* new_ptr = erts_alloc(s->alloc_type, new_size); + sys_memcpy(new_ptr, s->pstart, old_size); + s->pstart = new_ptr; + } + s->pend = s->pstart + new_size; + s->psp = s->pstart + sp_offs; +} + /* CTYPE macros */ #define LATIN1 @@ -312,6 +353,17 @@ int erts_fit_in_bits_int32(Sint32 value) return fit_in_bits((Sint64) (Uint32) value, 4); } +int erts_fit_in_bits_uint(Uint value) +{ +#if ERTS_SIZEOF_ETERM == 4 + return fit_in_bits((Sint64) (Uint32) value, 4); +#elif ERTS_SIZEOF_ETERM == 8 + return fit_in_bits(value, 5); +#else +# error "No way, Jose" +#endif +} + int erts_print(int to, void *arg, char *format, ...) { @@ -708,7 +760,7 @@ erts_bld_atom_2uint_3tup_list(Uint **hpp, Uint *szp, Sint length, ** If N < 0, Y = FUNNY_NUMBER4 else Y = FUNNY_NUMBER3. ** The hash value is Y*h(J) mod 2^32 where h(J) is calculated like ** h(0) = <initial hash> -** h(i) = h(i-i)*X + B(i-1) +** h(i) = h(i-1)*X + B(i-1) ** The above should hold regardless of internal representation. ** Pids are hashed like small numbers but with differrent constants, as are ** ports. @@ -790,10 +842,10 @@ Uint32 make_hash(Eterm term_arg) unsigned op; /* Must not collide with the real tag_val_def's: */ -#define MAKE_HASH_TUPLE_OP 0x11 -#define MAKE_HASH_TERM_ARRAY_OP 0x12 -#define MAKE_HASH_CDR_PRE_OP 0x13 -#define MAKE_HASH_CDR_POST_OP 0x14 +#define MAKE_HASH_TUPLE_OP (FIRST_VACANT_TAG_DEF) +#define MAKE_HASH_TERM_ARRAY_OP (FIRST_VACANT_TAG_DEF+1) +#define MAKE_HASH_CDR_PRE_OP (FIRST_VACANT_TAG_DEF+2) +#define MAKE_HASH_CDR_POST_OP (FIRST_VACANT_TAG_DEF+3) /* ** Convenience macro for calculating a bytewise hash on an unsigned 32 bit @@ -973,21 +1025,9 @@ tail_recur: break; } case MAP_DEF: + case HASHMAP_DEF: { - map_t *mp = (map_t *)map_val(term); - int size = map_get_size(mp); - Eterm *ks = map_get_keys(mp); - Eterm *vs = map_get_values(mp); - - /* Use a prime with size to remedy some of - * the {} and <<>> hash problems */ - hash = hash*FUNNY_NUMBER13 + FUNNY_NUMBER14 + size; - if (size == 0) - break; - - /* push values first */ - WSTACK_PUSH3(stack, (UWord)vs, (UWord) size, MAKE_HASH_TERM_ARRAY_OP); - WSTACK_PUSH3(stack, (UWord)ks, (UWord) size, MAKE_HASH_TERM_ARRAY_OP); + hash = hash*FUNNY_NUMBER13 + FUNNY_NUMBER14 + make_hash2(term); break; } case TUPLE_DEF: @@ -1093,10 +1133,11 @@ Uint32 make_hash2(Eterm term) { Uint32 hash; - Uint32 hash_xor_keys = 0; - Uint32 hash_xor_values = 0; + Uint32 hash_xor_pairs; DeclareTmpHeapNoproc(tmp_big,2); + ERTS_UNDEF(hash_xor_pairs, 0); + /* (HCONST * {2, ..., 16}) mod 2^32 */ #define HCONST_2 0x3c6ef372UL #define HCONST_3 0xdaa66d2bUL @@ -1113,10 +1154,15 @@ make_hash2(Eterm term) #define HCONST_14 0xa708a81eUL #define HCONST_15 0x454021d7UL #define HCONST_16 0xe3779b90UL +#define HCONST_17 0x81af1549UL +#define HCONST_18 0x1fe68f02UL +#define HCONST_19 0xbe1e08bbUL +#define HCONST_20 0x5c558274UL +#define HCONST_21 0xfa8cfc2dUL #define HASH_MAP_TAIL (_make_header(1,_TAG_HEADER_REF)) -#define HASH_MAP_KEY (_make_header(2,_TAG_HEADER_REF)) -#define HASH_MAP_VAL (_make_header(3,_TAG_HEADER_REF)) +#define HASH_MAP_PAIR (_make_header(2,_TAG_HEADER_REF)) +#define HASH_CDR (_make_header(3,_TAG_HEADER_REF)) #define UINT32_HASH_2(Expr1, Expr2, AConst) \ do { \ @@ -1139,6 +1185,13 @@ make_hash2(Eterm term) } while(0) #define IS_SSMALL28(x) (((Uint) (((x) >> (28-1)) + 1)) < 2) + +#ifdef ARCH_64 +# define POINTER_HASH(Ptr, AConst) UINT32_HASH_2((Uint32)(UWord)(Ptr), (((UWord)(Ptr)) >> 32), AConst) +#else +# define POINTER_HASH(Ptr, AConst) UINT32_HASH(Ptr, AConst) +#endif + /* Optimization. Simple cases before declaration of estack. */ if (primary_tag(term) == TAG_PRIMARY_IMMED1) { switch (term & _TAG_IMMED1_MASK) { @@ -1193,9 +1246,9 @@ make_hash2(Eterm term) if (c > 0) UINT32_HASH(sh, HCONST_4); if (is_list(term)) { - term = *ptr; - tmp = *++ptr; - ESTACK_PUSH(s, tmp); + tmp = CDR(ptr); + ESTACK_PUSH(s, tmp); + term = CAR(ptr); } } break; @@ -1212,51 +1265,92 @@ make_hash2(Eterm term) UINT32_HASH(arity, HCONST_9); if (arity == 0) /* Empty tuple */ goto hash2_common; - for (i = arity; i >= 1; i--) { - tmp = elem[i]; - ESTACK_PUSH(s, tmp); + for (i = arity; ; i--) { + term = elem[i]; + if (i == 1) + break; + ESTACK_PUSH(s, term); } - goto hash2_common; } break; case MAP_SUBTAG: { - map_t *mp = (map_t *)map_val(term); + flatmap_t *mp = (flatmap_t *)flatmap_val(term); int i; - int size = map_get_size(mp); - Eterm *ks = map_get_keys(mp); - Eterm *vs = map_get_values(mp); + int size = flatmap_get_size(mp); + Eterm *ks = flatmap_get_keys(mp); + Eterm *vs = flatmap_get_values(mp); UINT32_HASH(size, HCONST_16); if (size == 0) { goto hash2_common; } - ESTACK_PUSH(s, hash_xor_values); - ESTACK_PUSH(s, hash_xor_keys); + /* We want a portable hash function that is *independent* of + * the order in which keys and values are encountered. + * We therefore calculate context independent hashes for all . + * key-value pairs and then xor them together. + */ + ESTACK_PUSH(s, hash_xor_pairs); ESTACK_PUSH(s, hash); ESTACK_PUSH(s, HASH_MAP_TAIL); hash = 0; - hash_xor_keys = 0; - hash_xor_values = 0; - for (i = size - 1; i >= 0; i--) { - tmp = vs[i]; - ESTACK_PUSH(s, HASH_MAP_VAL); - ESTACK_PUSH(s, tmp); - } - /* We do not want to expose the tuple representation. - * Do not push the keys as a tuple. - */ + hash_xor_pairs = 0; for (i = size - 1; i >= 0; i--) { - tmp = ks[i]; - ESTACK_PUSH(s, HASH_MAP_KEY); - ESTACK_PUSH(s, tmp); + ESTACK_PUSH(s, HASH_MAP_PAIR); + ESTACK_PUSH(s, vs[i]); + ESTACK_PUSH(s, ks[i]); } goto hash2_common; } break; + case HASHMAP_SUBTAG: + { + Eterm* ptr = boxed_val(term) + 1; + Uint size; + int i; + switch (hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_HEAD_ARRAY: + case HAMT_SUBTAG_HEAD_BITMAP: + size = *ptr++; + UINT32_HASH(size, HCONST_16); + if (size == 0) + goto hash2_common; + ESTACK_PUSH(s, hash_xor_pairs); + ESTACK_PUSH(s, hash); + ESTACK_PUSH(s, HASH_MAP_TAIL); + hash = 0; + hash_xor_pairs = 0; + } + switch (hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_HEAD_ARRAY: + case HAMT_SUBTAG_NODE_ARRAY: + i = 16; + break; + case HAMT_SUBTAG_HEAD_BITMAP: + case HAMT_SUBTAG_NODE_BITMAP: + i = hashmap_bitcount(MAP_HEADER_VAL(hdr)); + break; + default: + erl_exit(1, "bad header"); + } + while (i) { + if (is_list(*ptr)) { + Eterm* cons = list_val(*ptr); + ESTACK_PUSH(s, HASH_MAP_PAIR); + ESTACK_PUSH(s, CDR(cons)); + ESTACK_PUSH(s, CAR(cons)); + } + else { + ASSERT(is_boxed(*ptr)); + ESTACK_PUSH(s, *ptr); + } + i--; ptr++; + } + goto hash2_common; + } + break; case EXPORT_SUBTAG: { Export* ep = *((Export **) (export_val(term) + 1)); - UINT32_HASH_2 (ep->code[2], atom_tab(atom_val(ep->code[0]))->slot.bucket.hvalue, @@ -1271,7 +1365,6 @@ make_hash2(Eterm term) { ErlFunThing* funp = (ErlFunThing *) fun_val(term); Uint num_free = funp->num_free; - UINT32_HASH_2 (num_free, atom_tab(atom_val(funp->fe->module))->slot.bucket.hvalue, @@ -1352,7 +1445,8 @@ make_hash2(Eterm term) do { Uint t; Uint32 x, y; - t = i < n ? BIG_DIGIT(ptr, i++) : 0; + ASSERT(i < n); + t = BIG_DIGIT(ptr, i++); x = t & 0xffffffff; y = t >> 32; UINT32_HASH_2(x, y, con); @@ -1460,20 +1554,397 @@ make_hash2(Eterm term) switch (term) { case HASH_MAP_TAIL: { hash = (Uint32) ESTACK_POP(s); - UINT32_HASH(hash_xor_keys, HCONST_16); - UINT32_HASH(hash_xor_values, HCONST_16); - hash_xor_keys = (Uint32) ESTACK_POP(s); - hash_xor_values = (Uint32) ESTACK_POP(s); + UINT32_HASH(hash_xor_pairs, HCONST_19); + hash_xor_pairs = (Uint32) ESTACK_POP(s); goto hash2_common; } - case HASH_MAP_KEY: - hash_xor_keys ^= hash; + case HASH_MAP_PAIR: + hash_xor_pairs ^= hash; hash = 0; goto hash2_common; - case HASH_MAP_VAL: - hash_xor_values ^= hash; + default: + break; + } + } + } + } +} + +/* Term hash function for internal use. + * + * Limitation #1: Is not "portable" in any way between different VM instances. + * + * Limitation #2: The hash value is only valid as long as the term exists + * somewhere in the VM. Why? Because external pids, ports and refs are hashed + * by mixing the node *pointer* value. If a node disappears and later reappears + * with a new ErlNode struct, externals from that node will hash different than + * before. + * + * One IMPORTANT property must hold (for hamt). + * EVERY BIT of the term that is significant for equality (see EQ) + * MUST BE USED AS INPUT FOR THE HASH. Two different terms must always have a + * chance of hashing different when salted: hash([Salt|A]) vs hash([Salt|B]). + * + * This is why we can not use cached hash values for atoms for example. + * + */ + +#define CONST_HASH(AConst) \ +do { /* Lightweight mixing of constant (type info) */ \ + hash ^= AConst; \ + hash = (hash << 17) ^ (hash >> (32-17)); \ +} while (0) + +Uint32 +make_internal_hash(Eterm term) +{ + Uint32 hash; + Uint32 hash_xor_pairs; + + ERTS_UNDEF(hash_xor_pairs, 0); + + /* Optimization. Simple cases before declaration of estack. */ + if (primary_tag(term) == TAG_PRIMARY_IMMED1) { + hash = 0; + #if ERTS_SIZEOF_ETERM == 8 + UINT32_HASH_2((Uint32)term, (Uint32)(term >> 32), HCONST); + #elif ERTS_SIZEOF_ETERM == 4 + UINT32_HASH(term, HCONST); + #else + # error "No you don't" + #endif + return hash; + } + { + Eterm tmp; + DECLARE_ESTACK(s); + + UseTmpHeapNoproc(2); + hash = 0; + for (;;) { + switch (primary_tag(term)) { + case TAG_PRIMARY_LIST: + { + int c = 0; + Uint32 sh = 0; + Eterm* ptr = list_val(term); + while (is_byte(*ptr)) { + /* Optimization for strings. */ + sh = (sh << 8) + unsigned_val(*ptr); + if (c == 3) { + UINT32_HASH(sh, HCONST_4); + c = sh = 0; + } else { + c++; + } + term = CDR(ptr); + if (is_not_list(term)) + break; + ptr = list_val(term); + } + if (c > 0) + UINT32_HASH(sh, HCONST_4); + if (is_list(term)) { + tmp = CDR(ptr); + CONST_HASH(HCONST_17); /* Hash CAR in cons cell */ + ESTACK_PUSH(s, tmp); + if (is_not_list(tmp)) { + ESTACK_PUSH(s, HASH_CDR); + } + term = CAR(ptr); + } + } + break; + case TAG_PRIMARY_BOXED: + { + Eterm hdr = *boxed_val(term); + ASSERT(is_header(hdr)); + switch (hdr & _TAG_HEADER_MASK) { + case ARITYVAL_SUBTAG: + { + int i; + int arity = header_arity(hdr); + Eterm* elem = tuple_val(term); + UINT32_HASH(arity, HCONST_9); + if (arity == 0) /* Empty tuple */ + goto pop_next; + for (i = arity; ; i--) { + term = elem[i]; + if (i == 1) + break; + ESTACK_PUSH(s, term); + } + } + break; + case MAP_SUBTAG: + { + flatmap_t *mp = (flatmap_t *)flatmap_val(term); + int i; + int size = flatmap_get_size(mp); + Eterm *ks = flatmap_get_keys(mp); + Eterm *vs = flatmap_get_values(mp); + UINT32_HASH(size, HCONST_16); + if (size == 0) { + goto pop_next; + } + /* We want a hash function that is *independent* of + * the order in which keys and values are encountered. + * We therefore calculate context independent hashes for all . + * key-value pairs and then xor them together. + */ + ESTACK_PUSH(s, hash_xor_pairs); + ESTACK_PUSH(s, hash); + ESTACK_PUSH(s, HASH_MAP_TAIL); + hash = 0; + hash_xor_pairs = 0; + for (i = size - 1; i >= 0; i--) { + ESTACK_PUSH(s, HASH_MAP_PAIR); + ESTACK_PUSH(s, vs[i]); + ESTACK_PUSH(s, ks[i]); + } + goto pop_next; + } + break; + case HASHMAP_SUBTAG: + { + Eterm* ptr = boxed_val(term) + 1; + Uint size; + int i; + switch (hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_HEAD_ARRAY: + case HAMT_SUBTAG_HEAD_BITMAP: + size = *ptr++; + UINT32_HASH(size, HCONST_16); + if (size == 0) + goto pop_next; + ESTACK_PUSH(s, hash_xor_pairs); + ESTACK_PUSH(s, hash); + ESTACK_PUSH(s, HASH_MAP_TAIL); + hash = 0; + hash_xor_pairs = 0; + } + switch (hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_HEAD_ARRAY: + case HAMT_SUBTAG_NODE_ARRAY: + i = 16; + break; + case HAMT_SUBTAG_HEAD_BITMAP: + case HAMT_SUBTAG_NODE_BITMAP: + i = hashmap_bitcount(MAP_HEADER_VAL(hdr)); + break; + default: + erl_exit(1, "bad header"); + } + while (i) { + if (is_list(*ptr)) { + Eterm* cons = list_val(*ptr); + ESTACK_PUSH(s, HASH_MAP_PAIR); + ESTACK_PUSH(s, CDR(cons)); + ESTACK_PUSH(s, CAR(cons)); + } + else { + ASSERT(is_boxed(*ptr)); + ESTACK_PUSH(s, *ptr); + } + i--; ptr++; + } + goto pop_next; + } + break; + case EXPORT_SUBTAG: + { + Export* ep = *((Export **) (export_val(term) + 1)); + /* Assumes Export entries never moves */ + POINTER_HASH(ep, HCONST_14); + goto pop_next; + } + + case FUN_SUBTAG: + { + ErlFunThing* funp = (ErlFunThing *) fun_val(term); + Uint num_free = funp->num_free; + UINT32_HASH_2(num_free, funp->fe->module, HCONST_20); + UINT32_HASH_2(funp->fe->old_index, funp->fe->old_uniq, HCONST_21); + if (num_free == 0) { + goto pop_next; + } else { + Eterm* bptr = funp->env + num_free - 1; + while (num_free-- > 1) { + term = *bptr--; + ESTACK_PUSH(s, term); + } + term = *bptr; + } + } + break; + case REFC_BINARY_SUBTAG: + case HEAP_BINARY_SUBTAG: + case SUB_BINARY_SUBTAG: + { + byte* bptr; + unsigned sz = binary_size(term); + Uint32 con = HCONST_13 + hash; + Uint bitoffs; + Uint bitsize; + + ERTS_GET_BINARY_BYTES(term, bptr, bitoffs, bitsize); + if (sz == 0 && bitsize == 0) { + hash = con; + } else { + if (bitoffs == 0) { + hash = block_hash(bptr, sz, con); + if (bitsize > 0) { + UINT32_HASH_2(bitsize, (bptr[sz] >> (8 - bitsize)), + HCONST_15); + } + } else { + byte* buf = (byte *) erts_alloc(ERTS_ALC_T_TMP, + sz + (bitsize != 0)); + erts_copy_bits(bptr, bitoffs, 1, buf, 0, 1, sz*8+bitsize); + hash = block_hash(buf, sz, con); + if (bitsize > 0) { + UINT32_HASH_2(bitsize, (buf[sz] >> (8 - bitsize)), + HCONST_15); + } + erts_free(ERTS_ALC_T_TMP, (void *) buf); + } + } + goto pop_next; + } + break; + case POS_BIG_SUBTAG: + case NEG_BIG_SUBTAG: + { + Eterm* ptr = big_val(term); + Uint i = 0; + Uint n = BIG_SIZE(ptr); + Uint32 con = BIG_SIGN(ptr) ? HCONST_10 : HCONST_11; +#if D_EXP == 16 + do { + Uint32 x, y; + x = i < n ? BIG_DIGIT(ptr, i++) : 0; + x += (Uint32)(i < n ? BIG_DIGIT(ptr, i++) : 0) << 16; + y = i < n ? BIG_DIGIT(ptr, i++) : 0; + y += (Uint32)(i < n ? BIG_DIGIT(ptr, i++) : 0) << 16; + UINT32_HASH_2(x, y, con); + } while (i < n); +#elif D_EXP == 32 + do { + Uint32 x, y; + x = i < n ? BIG_DIGIT(ptr, i++) : 0; + y = i < n ? BIG_DIGIT(ptr, i++) : 0; + UINT32_HASH_2(x, y, con); + } while (i < n); +#elif D_EXP == 64 + do { + Uint t; + Uint32 x, y; + ASSERT(i < n); + t = BIG_DIGIT(ptr, i++); + x = t & 0xffffffff; + y = t >> 32; + UINT32_HASH_2(x, y, con); + } while (i < n); +#else +#error "unsupported D_EXP size" +#endif + goto pop_next; + } + break; + case REF_SUBTAG: + UINT32_HASH(internal_ref_numbers(term)[0], HCONST_7); + ASSERT(internal_ref_no_of_numbers(term) == 3); + UINT32_HASH_2(internal_ref_numbers(term)[1], + internal_ref_numbers(term)[2], HCONST_8); + goto pop_next; + + case EXTERNAL_REF_SUBTAG: + { + ExternalThing* thing = external_thing_ptr(term); + + ASSERT(external_thing_ref_no_of_numbers(thing) == 3); + /* See limitation #2 */ + #ifdef ARCH_64 + POINTER_HASH(thing->node, HCONST_7); + UINT32_HASH(external_thing_ref_numbers(thing)[0], HCONST_7); + #else + UINT32_HASH_2(thing->node, + external_thing_ref_numbers(thing)[0], HCONST_7); + #endif + UINT32_HASH_2(external_thing_ref_numbers(thing)[1], + external_thing_ref_numbers(thing)[2], HCONST_8); + goto pop_next; + } + case EXTERNAL_PID_SUBTAG: { + ExternalThing* thing = external_thing_ptr(term); + /* See limitation #2 */ + #ifdef ARCH_64 + POINTER_HASH(thing->node, HCONST_5); + UINT32_HASH(thing->data.ui[0], HCONST_5); + #else + UINT32_HASH_2(thing->node, thing->data.ui[0], HCONST_5); + #endif + goto pop_next; + } + case EXTERNAL_PORT_SUBTAG: { + ExternalThing* thing = external_thing_ptr(term); + /* See limitation #2 */ + #ifdef ARCH_64 + POINTER_HASH(thing->node, HCONST_6); + UINT32_HASH(thing->data.ui[0], HCONST_6); + #else + UINT32_HASH_2(thing->node, thing->data.ui[0], HCONST_6); + #endif + goto pop_next; + } + case FLOAT_SUBTAG: + { + FloatDef ff; + GET_DOUBLE(term, ff); + UINT32_HASH_2(ff.fw[0], ff.fw[1], HCONST_12); + goto pop_next; + } + + default: + erl_exit(1, "Invalid tag in make_hash2(0x%X)\n", term); + } + } + break; + case TAG_PRIMARY_IMMED1: + #if ERTS_SIZEOF_ETERM == 8 + UINT32_HASH_2((Uint32)term, (Uint32)(term >> 32), HCONST); + #else + UINT32_HASH(term, HCONST); + #endif + goto pop_next; + + default: + erl_exit(1, "Invalid tag in make_hash2(0x%X)\n", term); + + pop_next: + if (ESTACK_ISEMPTY(s)) { + DESTROY_ESTACK(s); + UnUseTmpHeapNoproc(2); + return hash; + } + + term = ESTACK_POP(s); + + switch (term) { + case HASH_MAP_TAIL: { + hash = (Uint32) ESTACK_POP(s); + UINT32_HASH(hash_xor_pairs, HCONST_19); + hash_xor_pairs = (Uint32) ESTACK_POP(s); + goto pop_next; + } + case HASH_MAP_PAIR: + hash_xor_pairs ^= hash; hash = 0; - goto hash2_common; + goto pop_next; + + case HASH_CDR: + CONST_HASH(HCONST_18); /* Hash CDR i cons cell */ + goto pop_next; default: break; } @@ -1481,9 +1952,10 @@ make_hash2(Eterm term) } } +#undef CONST_HASH #undef HASH_MAP_TAIL -#undef HASH_MAP_KEY -#undef HASH_MAP_VAL +#undef HASH_MAP_PAIR +#undef HASH_CDR #undef UINT32_HASH_2 #undef UINT32_HASH @@ -1700,21 +2172,9 @@ tail_recur: break; case MAP_DEF: + case HASHMAP_DEF: { - map_t *mp = (map_t *)map_val(term); - int size = map_get_size(mp); - Eterm *ks = map_get_keys(mp); - Eterm *vs = map_get_values(mp); - - /* Use a prime with size to remedy some of - * the {} and <<>> hash problems */ - hash = hash*FUNNY_NUMBER13 + FUNNY_NUMBER14 + size; - if (size == 0) - break; - - /* push values first */ - WSTACK_PUSH3(stack, (UWord)vs, (UWord) size, MAKE_HASH_TERM_ARRAY_OP); - WSTACK_PUSH3(stack, (UWord)ks, (UWord) size, MAKE_HASH_TERM_ARRAY_OP); + hash = hash*FUNNY_NUMBER13 + FUNNY_NUMBER14 + make_hash2(term); break; } case TUPLE_DEF: @@ -2123,13 +2583,13 @@ tailrecur_ne: } case MAP_SUBTAG: { - aa = map_val_rel(a, a_base); + aa = flatmap_val_rel(a, a_base); if (!is_boxed(b) || *boxed_val_rel(b,b_base) != *aa) goto not_equal; - bb = map_val_rel(b,b_base); - sz = map_get_size((map_t*)aa); + bb = flatmap_val_rel(b,b_base); + sz = flatmap_get_size((flatmap_t*)aa); - if (sz != map_get_size((map_t*)bb)) goto not_equal; + if (sz != flatmap_get_size((flatmap_t*)bb)) goto not_equal; if (sz == 0) goto pop_next; aa += 2; @@ -2328,6 +2788,32 @@ tailrecur_ne: } break; /* not equal */ } + case HASHMAP_SUBTAG: + { + if (!is_boxed(b) || *boxed_val_rel(b,b_base) != hdr) + goto not_equal; + + aa = hashmap_val_rel(a, a_base) + 1; + bb = hashmap_val_rel(b, b_base) + 1; + switch (hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_HEAD_ARRAY: + aa++; bb++; + case HAMT_SUBTAG_NODE_ARRAY: + sz = 16; + break; + case HAMT_SUBTAG_HEAD_BITMAP: + aa++; bb++; + case HAMT_SUBTAG_NODE_BITMAP: + sz = hashmap_bitcount(MAP_HEADER_VAL(hdr)); + ASSERT(sz > 0 && sz < 16); + break; + default: + erl_exit(1, "Unknown hashmap subsubtag\n"); + } + goto term_array; + } + default: + ASSERT(!"Unknown boxed subtab in EQ"); } break; } @@ -2451,7 +2937,18 @@ Sint erts_cmp_rel_opt(Eterm a, Eterm* a_base, Eterm b, Eterm* b_base, int exact) Sint erts_cmp(Eterm a, Eterm b, int exact) #endif { - DECLARE_WSTACK(stack); +#define PSTACK_TYPE struct erts_cmp_hashmap_state + struct erts_cmp_hashmap_state { + Sint wstack_rollback; + int was_exact; + Eterm *ap; + Eterm *bp; + Eterm min_key; + Sint cmp_res; /* result so far -1,0,+1 */ + }; + PSTACK_DECLARE(hmap_stack, 1); + WSTACK_DECLARE(stack); + WSTACK_DECLARE(b_stack); /* only used by hashmaps */ Eterm* aa; Eterm* bb; int i; @@ -2467,6 +2964,26 @@ Sint erts_cmp(Eterm a, Eterm b, int exact) Uint32 *anum; Uint32 *bnum; +/* The WSTACK contains naked Eterms and Operations marked with header-tags */ +#define OP_BITS 4 +#define OP_MASK 0xF +#define TERM_ARRAY_OP 0 +#define SWITCH_EXACT_OFF_OP 1 +#define HASHMAP_PHASE1_ARE_KEYS_EQUAL 2 +#define HASHMAP_PHASE1_IS_MIN_KEY 3 +#define HASHMAP_PHASE1_CMP_VALUES 4 +#define HASHMAP_PHASE2_ARE_KEYS_EQUAL 5 +#define HASHMAP_PHASE2_IS_MIN_KEY_A 6 +#define HASHMAP_PHASE2_IS_MIN_KEY_B 7 + + +#define OP_WORD(OP) (((OP) << _TAG_PRIMARY_SIZE) | TAG_PRIMARY_HEADER) +#define TERM_ARRAY_OP_WORD(SZ) OP_WORD(((SZ) << OP_BITS) | TERM_ARRAY_OP) + +#define GET_OP(WORD) (ASSERT(is_header(WORD)), ((WORD) >> _TAG_PRIMARY_SIZE) & OP_MASK) +#define GET_OP_ARG(WORD) (ASSERT(is_header(WORD)), ((WORD) >> (OP_BITS + _TAG_PRIMARY_SIZE))) + + #define RETURN_NEQ(cmp) { j=(cmp); ASSERT(j != 0); goto not_equal; } #define ON_CMP_GOTO(cmp) if ((j=(cmp)) == 0) goto pop_next; else goto not_equal @@ -2482,6 +2999,8 @@ Sint erts_cmp(Eterm a, Eterm b, int exact) } while (0) +bodyrecur: + j = 0; tailrecur: if (is_same(a,a_base,b,b_base)) { /* Equal values or pointers. */ goto pop_next; @@ -2609,24 +3128,83 @@ tailrecur_ne: ++bb; goto term_array; case (_TAG_HEADER_MAP >> _TAG_PRIMARY_SIZE) : - if (!is_map_rel(b,b_base)) { + if (!is_flatmap_rel(b,b_base)) { a_tag = MAP_DEF; goto mixed_types; } - aa = (Eterm *)map_val_rel(a,a_base); - bb = (Eterm *)map_val_rel(b,b_base); + aa = (Eterm *)flatmap_val_rel(a,a_base); + bb = (Eterm *)flatmap_val_rel(b,b_base); - i = map_get_size((map_t*)aa); - if (i != map_get_size((map_t*)bb)) { - RETURN_NEQ((int)(i - map_get_size((map_t*)bb))); + i = flatmap_get_size((flatmap_t*)aa); + if (i != flatmap_get_size((flatmap_t*)bb)) { + RETURN_NEQ((int)(i - flatmap_get_size((flatmap_t*)bb))); } if (i == 0) { goto pop_next; } aa += 2; bb += 2; - i += 1; /* increment for tuple-keys */ - goto term_array; + if (exact) { + i += 1; /* increment for tuple-keys */ + goto term_array; + } + else { + /* Value array */ + WSTACK_PUSH3(stack, (UWord)(bb+1), (UWord)(aa+1), TERM_ARRAY_OP_WORD(i)); + /* Switch back from 'exact' key compare */ + WSTACK_PUSH(stack, OP_WORD(SWITCH_EXACT_OFF_OP)); + /* Now do 'exact' compare of key tuples */ + a = *aa; + b = *bb; + exact = 1; + goto bodyrecur; + } + + case (_TAG_HEADER_HASHMAP >> _TAG_PRIMARY_SIZE) : + { + struct erts_cmp_hashmap_state* sp; + if (!is_hashmap_rel(b,b_base)) { + a_tag = HASHMAP_DEF; + goto mixed_types; + } + i = hashmap_size_rel(a,a_base) - hashmap_size_rel(b,b_base); + if (i) { + RETURN_NEQ(i); + } + if (hashmap_size_rel(a,a_base) == 0) { + goto pop_next; + } + + /* Hashmap compare strategy: + Phase 1. While keys are identical + Do synchronous stepping through leafs of both trees in hash + order. Maintain value compare result of minimal key. + + Phase 2. If key diff was found in phase 1 + Ignore values from now on. + Continue iterate trees by always advancing the one + lagging behind hash-wise. Identical keys are skipped. + A minimal key can only be candidate as tie-breaker if we + have passed that hash value in the other tree (which means + the key did not exist in the other tree). + */ + + sp = PSTACK_PUSH(hmap_stack); + hashmap_iterator_init(&stack, a, 0); + hashmap_iterator_init(&b_stack, b, 0); + sp->ap = hashmap_iterator_next(&stack); + sp->bp = hashmap_iterator_next(&b_stack); + sp->cmp_res = 0; + ASSERT(sp->ap && sp->bp); + + a = CAR(sp->ap); + b = CAR(sp->bp); + sp->was_exact = exact; + exact = 1; + WSTACK_PUSH(stack, OP_WORD(HASHMAP_PHASE1_ARE_KEYS_EQUAL)); + sp->wstack_rollback = WSTACK_COUNT(stack); + goto bodyrecur; + } case (_TAG_HEADER_FLOAT >> _TAG_PRIMARY_SIZE): if (!is_float_rel(b,b_base)) { a_tag = FLOAT_DEF; @@ -2986,8 +3564,7 @@ term_array: /* arrays in 'aa' and 'bb', length in 'i' */ goto not_equal; } } else { - /* (ab)Use TAG_PRIMARY_HEADER to recognize a term_array */ - WSTACK_PUSH3(stack, i, (UWord)bb, (UWord)aa | TAG_PRIMARY_HEADER); + WSTACK_PUSH3(stack, (UWord)bb, (UWord)aa, TERM_ARRAY_OP_WORD(i)); goto tailrecur_ne; } } @@ -2999,22 +3576,179 @@ term_array: /* arrays in 'aa' and 'bb', length in 'i' */ pop_next: if (!WSTACK_ISEMPTY(stack)) { UWord something = WSTACK_POP(stack); - if (primary_tag((Eterm) something) == TAG_PRIMARY_HEADER) { /* a term_array */ - aa = (Eterm*) something; - bb = (Eterm*) WSTACK_POP(stack); - i = WSTACK_POP(stack); - goto term_array; + struct erts_cmp_hashmap_state* sp; + if (primary_tag((Eterm) something) == TAG_PRIMARY_HEADER) { /* an operation */ + switch (GET_OP(something)) { + case TERM_ARRAY_OP: + i = GET_OP_ARG(something); + aa = (Eterm*)WSTACK_POP(stack); + bb = (Eterm*) WSTACK_POP(stack); + goto term_array; + + case SWITCH_EXACT_OFF_OP: + /* Done with exact compare of map keys, switch back */ + ASSERT(exact); + exact = 0; + goto pop_next; + + case HASHMAP_PHASE1_ARE_KEYS_EQUAL: { + sp = PSTACK_TOP(hmap_stack); + if (j) { + /* Key diff found, enter phase 2 */ + if (hashmap_key_hash_cmp(sp->ap, sp->bp) < 0) { + sp->min_key = CAR(sp->ap); + sp->cmp_res = -1; + sp->ap = hashmap_iterator_next(&stack); + } + else { + sp->min_key = CAR(sp->bp); + sp->cmp_res = 1; + sp->bp = hashmap_iterator_next(&b_stack); + } + exact = 1; /* only exact key compares in phase 2 */ + goto case_HASHMAP_PHASE2_LOOP; + } + + /* No key diff found so far, compare values if min key */ + + if (sp->cmp_res) { + a = CAR(sp->ap); + b = sp->min_key; + exact = 1; + WSTACK_PUSH(stack, OP_WORD(HASHMAP_PHASE1_IS_MIN_KEY)); + sp->wstack_rollback = WSTACK_COUNT(stack); + goto bodyrecur; + } + /* no min key-value found yet */ + a = CDR(sp->ap); + b = CDR(sp->bp); + exact = sp->was_exact; + WSTACK_PUSH(stack, OP_WORD(HASHMAP_PHASE1_CMP_VALUES)); + sp->wstack_rollback = WSTACK_COUNT(stack); + goto bodyrecur; + } + case HASHMAP_PHASE1_IS_MIN_KEY: + sp = PSTACK_TOP(hmap_stack); + if (j < 0) { + a = CDR(sp->ap); + b = CDR(sp->bp); + exact = sp->was_exact; + WSTACK_PUSH(stack, OP_WORD(HASHMAP_PHASE1_CMP_VALUES)); + sp->wstack_rollback = WSTACK_COUNT(stack); + goto bodyrecur; + } + goto case_HASHMAP_PHASE1_LOOP; + + case HASHMAP_PHASE1_CMP_VALUES: + sp = PSTACK_TOP(hmap_stack); + if (j) { + sp->cmp_res = j; + sp->min_key = CAR(sp->ap); + } + case_HASHMAP_PHASE1_LOOP: + sp->ap = hashmap_iterator_next(&stack); + sp->bp = hashmap_iterator_next(&b_stack); + if (!sp->ap) { + /* end of maps with identical keys */ + ASSERT(!sp->bp); + j = sp->cmp_res; + exact = sp->was_exact; + (void) PSTACK_POP(hmap_stack); + ON_CMP_GOTO(j); + } + a = CAR(sp->ap); + b = CAR(sp->bp); + exact = 1; + WSTACK_PUSH(stack, OP_WORD(HASHMAP_PHASE1_ARE_KEYS_EQUAL)); + sp->wstack_rollback = WSTACK_COUNT(stack); + goto bodyrecur; + + case_HASHMAP_PHASE2_LOOP: + if (sp->ap && sp->bp) { + a = CAR(sp->ap); + b = CAR(sp->bp); + ASSERT(exact); + WSTACK_PUSH(stack, OP_WORD(HASHMAP_PHASE2_ARE_KEYS_EQUAL)); + sp->wstack_rollback = WSTACK_COUNT(stack); + goto bodyrecur; + } + goto case_HASHMAP_PHASE2_NEXT_STEP; + + case HASHMAP_PHASE2_ARE_KEYS_EQUAL: + sp = PSTACK_TOP(hmap_stack); + if (j == 0) { + /* keys are equal, skip them */ + sp->ap = hashmap_iterator_next(&stack); + sp->bp = hashmap_iterator_next(&b_stack); + goto case_HASHMAP_PHASE2_LOOP; + } + /* fall through */ + case_HASHMAP_PHASE2_NEXT_STEP: + if (sp->ap || sp->bp) { + if (hashmap_key_hash_cmp(sp->ap, sp->bp) < 0) { + ASSERT(sp->ap); + a = CAR(sp->ap); + b = sp->min_key; + ASSERT(exact); + WSTACK_PUSH(stack, OP_WORD(HASHMAP_PHASE2_IS_MIN_KEY_A)); + } + else { /* hash_cmp > 0 */ + ASSERT(sp->bp); + a = CAR(sp->bp); + b = sp->min_key; + ASSERT(exact); + WSTACK_PUSH(stack, OP_WORD(HASHMAP_PHASE2_IS_MIN_KEY_B)); + } + sp->wstack_rollback = WSTACK_COUNT(stack); + goto bodyrecur; + } + /* End of both maps */ + j = sp->cmp_res; + exact = sp->was_exact; + (void) PSTACK_POP(hmap_stack); + ON_CMP_GOTO(j); + + case HASHMAP_PHASE2_IS_MIN_KEY_A: + sp = PSTACK_TOP(hmap_stack); + if (j < 0) { + sp->min_key = CAR(sp->ap); + sp->cmp_res = -1; + } + sp->ap = hashmap_iterator_next(&stack); + goto case_HASHMAP_PHASE2_LOOP; + + case HASHMAP_PHASE2_IS_MIN_KEY_B: + sp = PSTACK_TOP(hmap_stack); + if (j < 0) { + sp->min_key = CAR(sp->bp); + sp->cmp_res = 1; + } + sp->bp = hashmap_iterator_next(&b_stack); + goto case_HASHMAP_PHASE2_LOOP; + + default: + ASSERT(!"Invalid cmp op"); + } /* switch */ } a = (Eterm) something; b = (Eterm) WSTACK_POP(stack); goto tailrecur; } - DESTROY_WSTACK(stack); + ASSERT(PSTACK_IS_EMPTY(hmap_stack)); + PSTACK_DESTROY(hmap_stack); + WSTACK_DESTROY(stack); + WSTACK_DESTROY(b_stack); return 0; not_equal: - DESTROY_WSTACK(stack); + if (!PSTACK_IS_EMPTY(hmap_stack)) { + WSTACK_ROLLBACK(stack, PSTACK_TOP(hmap_stack)->wstack_rollback); + goto pop_next; + } + PSTACK_DESTROY(hmap_stack); + WSTACK_DESTROY(stack); + WSTACK_DESTROY(b_stack); return j; #undef CMP_NODES diff --git a/erts/emulator/drivers/common/efile_drv.c b/erts/emulator/drivers/common/efile_drv.c index b62e9a0306..3b8e7acb6e 100644 --- a/erts/emulator/drivers/common/efile_drv.c +++ b/erts/emulator/drivers/common/efile_drv.c @@ -264,23 +264,6 @@ dt_private *get_dt_private(int); -#define GET_TIME(i, b) \ - (i).year = get_int32((b) + 0 * 4); \ - (i).month = get_int32((b) + 1 * 4); \ - (i).day = get_int32((b) + 2 * 4); \ - (i).hour = get_int32((b) + 3 * 4); \ - (i).minute = get_int32((b) + 4 * 4); \ - (i).second = get_int32((b) + 5 * 4) - -#define PUT_TIME(i, b) \ - put_int32((i).year, (b) + 0 * 4); \ - put_int32((i).month, (b) + 1 * 4); \ - put_int32((i).day, (b) + 2 * 4); \ - put_int32((i).hour, (b) + 3 * 4); \ - put_int32((i).minute,(b) + 4 * 4); \ - put_int32((i).second,(b) + 5 * 4) - - #if ALWAYS_READ_LINE_AHEAD #define DEFAULT_LINEBUF_SIZE 2048 #else diff --git a/erts/emulator/drivers/common/inet_drv.c b/erts/emulator/drivers/common/inet_drv.c index db8a251fdd..5196eb51c6 100644 --- a/erts/emulator/drivers/common/inet_drv.c +++ b/erts/emulator/drivers/common/inet_drv.c @@ -268,14 +268,13 @@ static BOOL (WINAPI *fpSetHandleInformation)(HANDLE,DWORD,DWORD); #define sock_htonl(x) htonl((x)) #define sock_send(s,buf,len,flag) send((s),(buf),(len),(flag)) #define sock_sendv(s, vec, size, np, flag) \ - WSASend((s),(WSABUF*)(vec),\ - (size),(np),(flag),NULL,NULL) + WSASend((s),(WSABUF*)(vec),(size),(np),(flag),NULL,NULL) #define sock_recv(s,buf,len,flag) recv((s),(buf),(len),(flag)) #define sock_recvfrom(s,buf,blen,flag,addr,alen) \ - recvfrom((s),(buf),(blen),(flag),(addr),(alen)) + recvfrom((s),(buf),(blen),(flag),(addr),(alen)) #define sock_sendto(s,buf,blen,flag,addr,alen) \ - sendto((s),(buf),(blen),(flag),(addr),(alen)) + sendto((s),(buf),(blen),(flag),(addr),(alen)) #define sock_hostname(buf, len) gethostname((buf), (len)) #define sock_getservbyname(name,proto) getservbyname((name),(proto)) @@ -360,9 +359,9 @@ static ssize_t writev_fallback(int fd, const struct iovec *iov, int iovcnt, int #define sock_accept(s, addr, len) accept((s), (addr), (len)) #define sock_send(s,buf,len,flag) inet_send((s),(buf),(len),(flag)) #define sock_sendto(s,buf,blen,flag,addr,alen) \ - sendto((s),(buf),(blen),(flag),(addr),(alen)) + sendto((s),(buf),(blen),(flag),(addr),(alen)) #define sock_sendv(s, vec, size, np, flag) \ - (*(np) = writev_fallback((s), (struct iovec*)(vec), (size), (*(np)))) + (*(np) = writev_fallback((s), (struct iovec*)(vec), (size), (*(np)))) #define sock_sendmsg(s,msghdr,flag) sendmsg((s),(msghdr),(flag)) #define sock_open(af, type, proto) socket((af), (type), (proto)) @@ -1178,6 +1177,7 @@ static ErlDrvSSizeT tcp_inet_ctl(ErlDrvData, unsigned int, static void tcp_inet_timeout(ErlDrvData); static void tcp_inet_process_exit(ErlDrvData, ErlDrvMonitor *); static void inet_stop_select(ErlDrvEvent, void*); +static void inet_emergency_close(ErlDrvData); #ifdef __WIN32__ static void tcp_inet_event(ErlDrvData, ErlDrvEvent); static void find_dynamic_functions(void); @@ -1288,7 +1288,8 @@ static struct erl_drv_entry tcp_inet_driver_entry = ERL_DRV_FLAG_USE_PORT_LOCKING|ERL_DRV_FLAG_SOFT_BUSY, NULL, tcp_inet_process_exit, - inet_stop_select + inet_stop_select, + inet_emergency_close }; @@ -1341,7 +1342,8 @@ static struct erl_drv_entry udp_inet_driver_entry = ERL_DRV_FLAG_USE_PORT_LOCKING, NULL, NULL, - inet_stop_select + inet_stop_select, + inet_emergency_close }; #endif @@ -1375,7 +1377,8 @@ static struct erl_drv_entry sctp_inet_driver_entry = ERL_DRV_FLAG_USE_PORT_LOCKING, NULL, NULL, /* process_exit */ - inet_stop_select + inet_stop_select, + inet_emergency_close }; #endif @@ -1421,7 +1424,7 @@ static int packet_inet_input(udp_descriptor* udesc, HANDLE event); static int packet_inet_output(udp_descriptor* udesc, HANDLE event); #endif -/* convert descriptor poiner to inet_descriptor pointer */ +/* convert descriptor pointer to inet_descriptor pointer */ #define INETP(d) (&(d)->inet) #ifdef __OSE__ @@ -2890,6 +2893,9 @@ static ErlDrvTermData am_sctp_rtoinfo, /* Option names */ /* For #sctp_paddrinfo{}: */ am_active, am_inactive, +# if HAVE_DECL_SCTP_UNCONFIRMED + am_unconfirmed, +# endif /* For #sctp_status{}: */ # if HAVE_DECL_SCTP_EMPTY @@ -3919,7 +3925,10 @@ static void inet_init_sctp(void) { /* For #sctp_paddrinfo{}: */ INIT_ATOM(active); INIT_ATOM(inactive); - +# if HAVE_DECL_SCTP_UNCONFIRMED + INIT_ATOM(unconfirmed); +# endif + /* For #sctp_status{}: */ # if HAVE_DECL_SCTP_EMPTY INIT_ATOM(empty); @@ -3948,9 +3957,9 @@ static int inet_init() if (0 != erl_drv_tsd_key_create("inet_buffer_stack_key", &buffer_stack_key)) goto error; - ASSERT(sizeof(struct in_addr) == 4); + ERTS_CT_ASSERT(sizeof(struct in_addr) == 4); # if defined(HAVE_IN6) && defined(AF_INET6) - ASSERT(sizeof(struct in6_addr) == 16); + ERTS_CT_ASSERT(sizeof(struct in6_addr) == 16); # endif INIT_ATOM(ok); @@ -3996,7 +4005,7 @@ static int inet_init() #ifdef HAVE_SCTP /* Check the size of SCTP AssocID -- currently both this driver and the Erlang part require 32 bit: */ - ASSERT(sizeof(sctp_assoc_t)==ASSOC_ID_LEN); + ERTS_CT_ASSERT(sizeof(sctp_assoc_t)==ASSOC_ID_LEN); # if defined(HAVE_SCTP_BINDX) p_sctp_bindx = sctp_bindx; # if defined(HAVE_SCTP_PEELOFF) @@ -4721,6 +4730,36 @@ static char* sockaddr_to_buf(struct sockaddr* addr, char* ptr, char* end) return NULL; } +/* sockaddr_bufsz_need + * Returns the number of bytes needed to store the information + * through sockaddr_to_buf + */ + +static size_t sockaddr_bufsz_need(struct sockaddr* addr) +{ + if (addr->sa_family == AF_INET || addr->sa_family == 0) { + return 1 + sizeof(struct in_addr); + } +#if defined(HAVE_IN6) && defined(AF_INET6) + else if (addr->sa_family == AF_INET6) { + return 1 + sizeof(struct in6_addr); + } +#endif +#if defined(AF_LINK) + if (addr->sa_family == AF_LINK) { + struct sockaddr_dl *sdl_p = (struct sockaddr_dl*) addr; + return 2 + sdl_p->sdl_alen; + } +#endif +#if defined(AF_PACKET) && defined(HAVE_NETPACKET_PACKET_H) + else if(addr->sa_family == AF_PACKET) { + struct sockaddr_ll *sll_p = (struct sockaddr_ll*) addr; + return 2 + sll_p->sll_halen; + } +#endif + return 0; +} + static char* buf_to_sockaddr(char* ptr, char* end, struct sockaddr* addr) { buf_check(ptr,end,1); @@ -5799,6 +5838,11 @@ done: } #elif defined(HAVE_GETIFADDRS) +#ifdef DEBUG +#define GETIFADDRS_BUFSZ (1) +#else +#define GETIFADDRS_BUFSZ (512) +#endif static ErlDrvSSizeT inet_ctl_getifaddrs(inet_descriptor* desc_p, char **rbuf_pp, ErlDrvSizeT rsize) @@ -5809,15 +5853,15 @@ static ErlDrvSSizeT inet_ctl_getifaddrs(inet_descriptor* desc_p, char *buf_p; char *buf_alloc_p; - buf_size = 512; - buf_alloc_p = ALLOC(buf_size); + buf_size = GETIFADDRS_BUFSZ; + buf_alloc_p = ALLOC(GETIFADDRS_BUFSZ); buf_p = buf_alloc_p; # define BUF_ENSURE(Size) \ do { \ int NEED_, GOT_ = buf_p - buf_alloc_p; \ NEED_ = GOT_ + (Size); \ if (NEED_ > buf_size) { \ - buf_size = NEED_ + 512; \ + buf_size = NEED_ + GETIFADDRS_BUFSZ; \ buf_alloc_p = REALLOC(buf_alloc_p, buf_size); \ buf_p = buf_alloc_p + GOT_; \ } \ @@ -5830,7 +5874,7 @@ static ErlDrvSSizeT inet_ctl_getifaddrs(inet_descriptor* desc_p, while (! (P_ = sockaddr_to_buf((sa), buf_p, \ buf_alloc_p+buf_size))) { \ int GOT_ = buf_p - buf_alloc_p; \ - buf_size += 512; \ + buf_size += GETIFADDRS_BUFSZ; \ buf_alloc_p = REALLOC(buf_alloc_p, buf_size); \ buf_p = buf_alloc_p + GOT_; \ } \ @@ -5887,10 +5931,11 @@ static ErlDrvSSizeT inet_ctl_getifaddrs(inet_descriptor* desc_p, || ifa_p->ifa_addr->sa_family == AF_PACKET #endif ) { - char *bp = buf_p; - BUF_ENSURE(1); - SOCKADDR_TO_BUF(INET_IFOPT_HWADDR, ifa_p->ifa_addr); - if (buf_p - bp < 4) buf_p = bp; /* Empty hwaddr */ + size_t need = sockaddr_bufsz_need(ifa_p->ifa_addr); + if (need > 3) { + BUF_ENSURE(1 + need); + SOCKADDR_TO_BUF(INET_IFOPT_HWADDR, ifa_p->ifa_addr); + } } #endif } @@ -5905,6 +5950,7 @@ static ErlDrvSSizeT inet_ctl_getifaddrs(inet_descriptor* desc_p, return buf_size; # undef BUF_ENSURE } +#undef GETIFADDRS_BUFSZ #else @@ -7338,8 +7384,13 @@ static int load_paddrinfo (ErlDrvTermData * spec, int i, case SCTP_INACTIVE: i = LOAD_ATOM (spec, i, am_inactive); break; +# if HAVE_DECL_SCTP_UNCONFIRMED + case SCTP_UNCONFIRMED: + i = LOAD_ATOM (spec, i, am_unconfirmed); + break; +# endif default: - ASSERT(0); /* NB: SCTP_UNCONFIRMED modifier not yet supported */ + i = LOAD_ATOM (spec, i, am_undefined); } i = LOAD_INT (spec, i, pai->spinfo_cwnd); i = LOAD_INT (spec, i, pai->spinfo_srtt); @@ -8204,6 +8255,19 @@ static void inet_stop(inet_descriptor* desc) FREE(desc); } +static void inet_emergency_close(ErlDrvData data) +{ + /* valid for any (UDP, TCP or SCTP) descriptor */ + tcp_descriptor* tcp_desc = (tcp_descriptor*)data; + inet_descriptor* desc = INETP(tcp_desc); + DEBUGF(("inet_emergency_close(%ld) {s=%d\r\n", + (long)desc->port, desc->s)); + if (desc->s != INVALID_SOCKET) { + sock_close(desc->s); + } +} + + static void set_default_msgq_limits(ErlDrvPort port) { ErlDrvSizeT q_high = INET_HIGH_MSGQ_WATERMARK; diff --git a/erts/emulator/drivers/common/zlib_drv.c b/erts/emulator/drivers/common/zlib_drv.c index 3143e4511d..f7b2d91d23 100644 --- a/erts/emulator/drivers/common/zlib_drv.c +++ b/erts/emulator/drivers/common/zlib_drv.c @@ -62,8 +62,17 @@ #define CRC32_COMBINE 23 #define ADLER32_COMBINE 24 +#define INFLATE_CHUNK 25 + + #define DEFAULT_BUFSZ 4000 +/* This flag is used in the same places, where zlib return codes + * (Z_OK, Z_STREAM_END, Z_NEED_DICT) are. So, we need to set it to + * relatively large value to avoid possible value clashes in future. + * */ +#define INFLATE_HAS_MORE 100 + static int zlib_init(void); static ErlDrvData zlib_start(ErlDrvPort port, char* buf); static void zlib_stop(ErlDrvData e); @@ -295,6 +304,58 @@ static int zlib_inflate(ZLibData* d, int flush) return res; } +static int zlib_inflate_chunk(ZLibData* d) +{ + int res = Z_OK; + + if ((d->bin == NULL) && (zlib_output_init(d) < 0)) { + errno = ENOMEM; + return Z_ERRNO; + } + + while ((driver_sizeq(d->port) > 0) && (d->s.avail_out > 0) && + (res != Z_STREAM_END)) { + int vlen; + SysIOVec* iov = driver_peekq(d->port, &vlen); + int len; + + d->s.next_in = iov[0].iov_base; + d->s.avail_in = iov[0].iov_len; + while((d->s.avail_in > 0) && (d->s.avail_out > 0) && (res != Z_STREAM_END)) { + res = inflate(&d->s, Z_NO_FLUSH); + if (res == Z_NEED_DICT) { + /* Essential to eat the header bytes that zlib has looked at */ + len = iov[0].iov_len - d->s.avail_in; + driver_deq(d->port, len); + return res; + } + if (res == Z_BUF_ERROR) { + /* Was possible more output, but actually not */ + res = Z_OK; + } + else if (res < 0) { + return res; + } + } + len = iov[0].iov_len - d->s.avail_in; + driver_deq(d->port, len); + } + + /* We are here because all input was consumed or EOS reached or output + * buffer is full */ + if (d->want_crc) { + d->crc = crc32(d->crc, (unsigned char*) d->bin->orig_bytes, + d->binsz - d->s.avail_out); + } + zlib_output(d); + if ((res == Z_OK) && (d->s.avail_in > 0)) + res = INFLATE_HAS_MORE; + else if (res == Z_STREAM_END) { + d->inflate_eos_seen = 1; + } + return res; +} + static int zlib_deflate(ZLibData* d, int flush) { int res = Z_OK; @@ -568,6 +629,18 @@ static ErlDrvSSizeT zlib_ctl(ErlDrvData drv_data, unsigned int command, char *bu return zlib_return(res, rbuf, rlen); } + case INFLATE_CHUNK: + if (d->state != ST_INFLATE) goto badarg; + if (len != 0) goto badarg; + res = zlib_inflate_chunk(d); + if (res == INFLATE_HAS_MORE) { + return zlib_value2(4, 0, rbuf, rlen); + } else if (res == Z_NEED_DICT) { + return zlib_value2(3, d->s.adler, rbuf, rlen); + } else { + return zlib_return(res, rbuf, rlen); + } + case GET_QSIZE: return zlib_value(driver_sizeq(d->port), rbuf, rlen); diff --git a/erts/emulator/drivers/unix/ttsl_drv.c b/erts/emulator/drivers/unix/ttsl_drv.c index 491e0a090e..a5960716f2 100644 --- a/erts/emulator/drivers/unix/ttsl_drv.c +++ b/erts/emulator/drivers/unix/ttsl_drv.c @@ -32,6 +32,10 @@ static ErlDrvData ttysl_start(ErlDrvPort, char*); #ifdef HAVE_TERMCAP /* else make an empty driver that can not be opened */ +#ifndef WANT_NONBLOCKING +#define WANT_NONBLOCKING +#endif + #include "sys.h" #include <ctype.h> #include <stdlib.h> @@ -39,6 +43,7 @@ static ErlDrvData ttysl_start(ErlDrvPort, char*); #include <string.h> #include <signal.h> #include <fcntl.h> +#include <limits.h> #include <locale.h> #include <unistd.h> #include <termios.h> @@ -57,6 +62,14 @@ static ErlDrvData ttysl_start(ErlDrvPort, char*); #include <langinfo.h> #endif +#if defined IOV_MAX +#define MAXIOV IOV_MAX +#elif defined UIO_MAXIOV +#define MAXIOV UIO_MAXIOV +#else +#define MAXIOV 16 +#endif + #define TRUE 1 #define FALSE 0 @@ -80,12 +93,15 @@ static volatile int cols_needs_update = FALSE; #define OP_INSC 2 #define OP_DELC 3 #define OP_BEEP 4 +#define OP_PUTC_SYNC 5 /* Control op */ #define CTRL_OP_GET_WINSIZE 100 #define CTRL_OP_GET_UNICODE_STATE 101 #define CTRL_OP_SET_UNICODE_STATE 102 - +/* We use 1024 as the buf size as that was the default buf size of FILE streams + on all platforms that I checked. */ +#define TTY_BUFFSIZE 1024 static int lbuf_size = BUFSIZ; static Uint32 *lbuf; /* The current line buffer */ @@ -113,13 +129,19 @@ static int lpos; /* The current "cursor position" in the line buf /* Main interface functions. */ static void ttysl_stop(ErlDrvData); static void ttysl_from_erlang(ErlDrvData, char*, ErlDrvSizeT); +static void ttysl_to_tty(ErlDrvData, ErlDrvEvent); +static void ttysl_flush_tty(ErlDrvData); static void ttysl_from_tty(ErlDrvData, ErlDrvEvent); static void ttysl_stop_select(ErlDrvEvent, void*); static Sint16 get_sint16(char*); static ErlDrvPort ttysl_port; static int ttysl_fd; -static FILE *ttysl_out; +static int ttysl_terminate = 0; +static int ttysl_send_ok = 0; +static ErlDrvBinary *putcbuf; +static int putcpos; +static int putclen; /* Functions that work on the line buffer. */ static int start_lbuf(void); @@ -201,22 +223,22 @@ struct erl_drv_entry ttsl_driver_entry = { IF_IMPL(ttysl_stop), IF_IMPL(ttysl_from_erlang), IF_IMPL(ttysl_from_tty), - NULL, - "tty_sl", - NULL, - NULL, + IF_IMPL(ttysl_to_tty), + "tty_sl", /* driver_name */ + NULL, /* finish */ + NULL, /* handle */ IF_IMPL(ttysl_control), NULL, /* timeout */ NULL, /* outputv */ NULL, /* ready_async */ - NULL, /* flush */ + IF_IMPL(ttysl_flush_tty), NULL, /* call */ NULL, /* event */ ERL_DRV_EXTENDED_MARKER, ERL_DRV_EXTENDED_MAJOR_VERSION, ERL_DRV_EXTENDED_MINOR_VERSION, 0, /* ERL_DRV_FLAGs */ - NULL, + NULL, /* handle2 */ NULL, /* process_exit */ IF_IMPL(ttysl_stop_select) }; @@ -296,8 +318,7 @@ static ErlDrvData ttysl_start(ErlDrvPort port, char* buf) return ERL_DRV_ERROR_GENERAL; } - /* Open the terminal and set the terminal */ - ttysl_out = fdopen(ttysl_fd, "w"); + SET_NONBLOCKING(ttysl_fd); #ifdef PRIMITIVE_UTF8_CHECK setlocale(LC_CTYPE, ""); /* Set international environment, @@ -317,8 +338,8 @@ static ErlDrvData ttysl_start(ErlDrvPort port, char* buf) } #endif DEBUGLOG(("utf8_mode is %s\n",(utf8_mode) ? "on" : "off")); - sys_sigset(SIGCONT, cont); - sys_sigset(SIGWINCH, winch); + sys_signal(SIGCONT, cont); + sys_signal(SIGWINCH, winch); driver_select(port, (ErlDrvEvent)(UWord)ttysl_fd, ERL_DRV_READ|ERL_DRV_USE, 1); ttysl_port = port; @@ -400,12 +421,14 @@ static void ttysl_stop(ErlDrvData ttysl_data) stop_lbuf(); stop_termcap(); tty_reset(ttysl_fd); - driver_select(ttysl_port, (ErlDrvEvent)(UWord)ttysl_fd, ERL_DRV_READ|ERL_DRV_USE, 0); - sys_sigset(SIGCONT, SIG_DFL); - sys_sigset(SIGWINCH, SIG_DFL); + driver_select(ttysl_port, (ErlDrvEvent)(UWord)ttysl_fd, + ERL_DRV_WRITE|ERL_DRV_READ|ERL_DRV_USE, 0); + sys_signal(SIGCONT, SIG_DFL); + sys_signal(SIGWINCH, SIG_DFL); } ttysl_port = (ErlDrvPort)-1; ttysl_fd = -1; + ttysl_terminate = 0; /* return TRUE; */ } @@ -650,10 +673,26 @@ static int check_buf_size(byte *s, int n) static void ttysl_from_erlang(ErlDrvData ttysl_data, char* buf, ErlDrvSizeT count) { + ErlDrvSizeT sz; + + sz = driver_sizeq(ttysl_port); + + putclen = count > TTY_BUFFSIZE ? TTY_BUFFSIZE : count; + putcbuf = driver_alloc_binary(putclen); + putcpos = 0; + if (lpos > MAXSIZE) put_chars((byte*)"\n", 1); switch (buf[0]) { + case OP_PUTC_SYNC: + /* Using sync means that we have to send an ok to the + controlling process for each command call. We delay + sending ok if the driver queue exceeds a certain size. + We do not set ourselves as a busy port, as this + could be very bad for user_drv, if it gets blocked on + the port_command. */ + /* fall through */ case OP_PUTC: DEBUGLOG(("OP: Putc(%lu)",(unsigned long) count-1)); if (check_buf_size((byte*)buf+1, count-1) == 0) @@ -678,10 +717,104 @@ static void ttysl_from_erlang(ErlDrvData ttysl_data, char* buf, ErlDrvSizeT coun /* Unknown op, just ignore. */ break; } - fflush(ttysl_out); + + driver_enq_bin(ttysl_port,putcbuf,0,putcpos); + + if (sz == 0) { + for (;;) { + int written, qlen; + SysIOVec *iov; + + iov = driver_peekq(ttysl_port,&qlen); + if (iov) + written = writev(ttysl_fd, iov, qlen > MAXIOV ? MAXIOV : qlen); + else + written = 0; + if (written < 0) { + if (errno == EAGAIN) { + driver_select(ttysl_port,(ErlDrvEvent)(long)ttysl_fd, + ERL_DRV_USE|ERL_DRV_WRITE,1); + break; + } else { + /* we ignore all other errors */ + break; + } + } else { + if (driver_deq(ttysl_port, written) == 0) + break; + } + } + } + + if (buf[0] == OP_PUTC_SYNC) { + if (driver_sizeq(ttysl_port) > TTY_BUFFSIZE && !ttysl_terminate) { + /* We delay sending the ack until the buffer has been consumed */ + ttysl_send_ok = 1; + } else { + ErlDrvTermData spec[] = { + ERL_DRV_PORT, driver_mk_port(ttysl_port), + ERL_DRV_ATOM, driver_mk_atom("ok"), + ERL_DRV_TUPLE, 2 + }; + ASSERT(ttysl_send_ok == 0); + erl_drv_output_term(driver_mk_port(ttysl_port), spec, + sizeof(spec) / sizeof(spec[0])); + } + } + return; /* TRUE; */ } +static void ttysl_to_tty(ErlDrvData ttysl_data, ErlDrvEvent fd) { + for (;;) { + int written, qlen; + SysIOVec *iov; + ErlDrvSizeT sz; + + iov = driver_peekq(ttysl_port,&qlen); + if (iov) + written = writev(ttysl_fd, iov, qlen > MAXIOV ? MAXIOV : qlen); + else + written = 0; + if (written < 0) { + if (errno == EAGAIN) { + break; + } else { + /* we ignore all other errors */ + } + } else { + sz = driver_deq(ttysl_port, written); + if (sz < TTY_BUFFSIZE && ttysl_send_ok) { + ErlDrvTermData spec[] = { + ERL_DRV_PORT, driver_mk_port(ttysl_port), + ERL_DRV_ATOM, driver_mk_atom("ok"), + ERL_DRV_TUPLE, 2 + }; + ttysl_send_ok = 0; + erl_drv_output_term(driver_mk_port(ttysl_port), spec, + sizeof(spec) / sizeof(spec[0])); + } + if (sz == 0) { + driver_select(ttysl_port,(ErlDrvEvent)(long)ttysl_fd, + ERL_DRV_WRITE,0); + if (ttysl_terminate) + /* flush has been called, which means we should terminate + when queue is empty. This will not send any exit + message */ + driver_failure_atom(ttysl_port, "normal"); + break; + } + } + } + + return; +} + +static void ttysl_flush_tty(ErlDrvData ttysl_data) { + ttysl_terminate = 1; + return; +} + static void ttysl_from_tty(ErlDrvData ttysl_data, ErlDrvEvent fd) { byte b[1024]; @@ -1070,7 +1203,14 @@ static int write_buf(Uint32 *s, int n) /* The basic procedure for outputting one character. */ static int outc(int c) { - return (int)putc(c, ttysl_out); + putcbuf->orig_bytes[putcpos++] = c; + if (putcpos == putclen) { + driver_enq_bin(ttysl_port,putcbuf,0,putclen); + putcpos = 0; + putclen = TTY_BUFFSIZE; + putcbuf = driver_alloc_binary(BUFSIZ); + } + return 1; } static int move_cursor(int from, int to) @@ -1318,11 +1458,11 @@ static RETSIGTYPE suspend(int sig) exit(1); } - sys_sigset(sig, SIG_DFL); /* Set signal handler to default */ + sys_signal(sig, SIG_DFL); /* Set signal handler to default */ sys_sigrelease(sig); /* Allow 'sig' to come through */ kill(getpid(), sig); /* Send ourselves the signal */ sys_sigblock(sig); /* Reset to old mask */ - sys_sigset(sig, suspend); /* Reset signal handler */ + sys_signal(sig, suspend); /* Reset signal handler */ if (tty_set(ttysl_fd) < 0) { fprintf(stderr,"Can't set tty raw \n"); diff --git a/erts/emulator/drivers/win32/ttsl_drv.c b/erts/emulator/drivers/win32/ttsl_drv.c index 502cb58dfa..851c336a11 100644 --- a/erts/emulator/drivers/win32/ttsl_drv.c +++ b/erts/emulator/drivers/win32/ttsl_drv.c @@ -46,6 +46,7 @@ static int rows; /* Number of rows available. */ #define OP_INSC 2 #define OP_DELC 3 #define OP_BEEP 4 +#define OP_PUTC_SYNC 5 /* Control op */ #define CTRL_OP_GET_WINSIZE 100 @@ -458,6 +459,7 @@ static void ttysl_from_erlang(ErlDrvData ttysl_data, char* buf, ErlDrvSizeT coun switch (buf[0]) { case OP_PUTC: + case OP_PUTC_SYNC: DEBUGLOG(("OP: Putc(%I64u)",(unsigned long long)count-1)); if (check_buf_size((byte*)buf+1, count-1) == 0) return; @@ -481,6 +483,20 @@ static void ttysl_from_erlang(ErlDrvData ttysl_data, char* buf, ErlDrvSizeT coun /* Unknown op, just ignore. */ break; } + + if (buf[0] == OP_PUTC_SYNC) { + /* On windows we do a blocking write to the tty so we just + send the ack immidiately. If at some point in the future + someone has a problem with tty output being blocking + this has to be changed. */ + ErlDrvTermData spec[] = { + ERL_DRV_PORT, driver_mk_port(ttysl_port), + ERL_DRV_ATOM, driver_mk_atom("ok"), + ERL_DRV_TUPLE, 2 + }; + erl_drv_output_term(driver_mk_port(ttysl_port), spec, + sizeof(spec) / sizeof(spec[0])); + } return; } diff --git a/erts/emulator/hipe/hipe_amd64.c b/erts/emulator/hipe/hipe_amd64.c index b5dff06987..63646825b2 100644 --- a/erts/emulator/hipe/hipe_amd64.c +++ b/erts/emulator/hipe/hipe_amd64.c @@ -125,7 +125,7 @@ static void atexit_alloc_code_stats(void) #define MAP_ANONYMOUS MAP_ANON #endif -static void morecore(unsigned int alloc_bytes) +static int morecore(unsigned int alloc_bytes) { unsigned int map_bytes; char *map_hint, *map_start; @@ -174,10 +174,9 @@ static void morecore(unsigned int alloc_bytes) abort(); } #endif - if (map_start == MAP_FAILED) { - perror("mmap"); - abort(); - } + if (map_start == MAP_FAILED) + return -1; + ALLOC_CODE_STATS(total_mapped += map_bytes); /* Merge adjacent mappings, so the trailing portion of the previous @@ -197,6 +196,8 @@ static void morecore(unsigned int alloc_bytes) } ALLOC_CODE_STATS(atexit_alloc_code_stats()); + + return 0; } static void *alloc_code(unsigned int alloc_bytes) @@ -206,8 +207,8 @@ static void *alloc_code(unsigned int alloc_bytes) /* Align function entries. */ alloc_bytes = (alloc_bytes + 3) & ~3; - if (code_bytes < alloc_bytes) - morecore(alloc_bytes); + if (code_bytes < alloc_bytes && morecore(alloc_bytes) != 0) + return NULL; ALLOC_CODE_STATS(++nr_allocs); ALLOC_CODE_STATS(total_alloc += alloc_bytes); res = code_next; @@ -224,18 +225,18 @@ void *hipe_alloc_code(Uint nrbytes, Eterm callees, Eterm *trampolines, Process * return alloc_code(nrbytes); } -/* called from hipe_bif0.c:hipe_bifs_make_native_stub_2() - and hipe_bif0.c:hipe_make_stub() */ -void *hipe_make_native_stub(void *beamAddress, unsigned int beamArity) +/* Make stub for native code calling exported beam function. +*/ +void *hipe_make_native_stub(void *callee_exp, unsigned int beamArity) { /* * This creates a native code stub with the following contents: * - * movq $Address, P_BEAM_IP(%ebp) %% Actually two movl + * movq $Address, P_CALLEE_EXP(%ebp) %% Actually two movl * movb $Arity, P_ARITY(%ebp) * jmp callemu * - * The stub has variable size, depending on whether the P_BEAM_IP + * The stub has variable size, depending on whether the P_CALLEE_EXP * and P_ARITY offsets fit in 8-bit signed displacements or not. * The rel32 offset in the final jmp depends on its actual location, * which also depends on the size of the previous instructions. @@ -248,49 +249,51 @@ void *hipe_make_native_stub(void *beamAddress, unsigned int beamArity) codeSize = /* 23, 26, 29, or 32 bytes */ 23 + /* 23 when all offsets are 8-bit */ - (P_BEAM_IP >= 128 ? 3 : 0) + - ((P_BEAM_IP + 4) >= 128 ? 3 : 0) + + (P_CALLEE_EXP >= 128 ? 3 : 0) + + ((P_CALLEE_EXP + 4) >= 128 ? 3 : 0) + (P_ARITY >= 128 ? 3 : 0); codep = code = alloc_code(codeSize); + if (!code) + return NULL; - /* movl $beamAddress, P_BEAM_IP(%ebp); 3 or 6 bytes, plus 4 */ + /* movl $callee_exp, P_CALLEE_EXP(%ebp); 3 or 6 bytes, plus 4 */ codep[0] = 0xc7; -#if P_BEAM_IP >= 128 +#if P_CALLEE_EXP >= 128 codep[1] = 0x85; /* disp32[EBP] */ - codep[2] = P_BEAM_IP & 0xFF; - codep[3] = (P_BEAM_IP >> 8) & 0xFF; - codep[4] = (P_BEAM_IP >> 16) & 0xFF; - codep[5] = (P_BEAM_IP >> 24) & 0xFF; + codep[2] = P_CALLEE_EXP & 0xFF; + codep[3] = (P_CALLEE_EXP >> 8) & 0xFF; + codep[4] = (P_CALLEE_EXP >> 16) & 0xFF; + codep[5] = (P_CALLEE_EXP >> 24) & 0xFF; codep += 6; #else codep[1] = 0x45; /* disp8[EBP] */ - codep[2] = P_BEAM_IP; + codep[2] = P_CALLEE_EXP; codep += 3; #endif - codep[0] = ((unsigned long)beamAddress ) & 0xFF; - codep[1] = ((unsigned long)beamAddress >> 8) & 0xFF; - codep[2] = ((unsigned long)beamAddress >> 16) & 0xFF; - codep[3] = ((unsigned long)beamAddress >> 24) & 0xFF; + codep[0] = ((unsigned long)callee_exp ) & 0xFF; + codep[1] = ((unsigned long)callee_exp >> 8) & 0xFF; + codep[2] = ((unsigned long)callee_exp >> 16) & 0xFF; + codep[3] = ((unsigned long)callee_exp >> 24) & 0xFF; codep += 4; - /* movl (shl 32 $beamAddress), P_BEAM_IP+4(%ebp); 3 or 6 bytes, plus 4 */ + /* movl (shl 32 $callee_exp), P_CALLEE_EXP+4(%ebp); 3 or 6 bytes, plus 4 */ codep[0] = 0xc7; -#if P_BEAM_IP+4 >= 128 +#if P_CALLEE_EXP+4 >= 128 codep[1] = 0x85; /* disp32[EBP] */ - codep[2] = (P_BEAM_IP+4) & 0xFF; - codep[3] = ((P_BEAM_IP+4) >> 8) & 0xFF; - codep[4] = ((P_BEAM_IP+4) >> 16) & 0xFF; - codep[5] = ((P_BEAM_IP+4) >> 24) & 0xFF; + codep[2] = (P_CALLEE_EXP+4) & 0xFF; + codep[3] = ((P_CALLEE_EXP+4) >> 8) & 0xFF; + codep[4] = ((P_CALLEE_EXP+4) >> 16) & 0xFF; + codep[5] = ((P_CALLEE_EXP+4) >> 24) & 0xFF; codep += 6; #else codep[1] = 0x45; /* disp8[EBP] */ - codep[2] = (P_BEAM_IP+4); + codep[2] = (P_CALLEE_EXP+4); codep += 3; #endif - codep[0] = ((unsigned long)beamAddress >> 32) & 0xFF; - codep[1] = ((unsigned long)beamAddress >> 40) & 0xFF; - codep[2] = ((unsigned long)beamAddress >> 48) & 0xFF; - codep[3] = ((unsigned long)beamAddress >> 56) & 0xFF; + codep[0] = ((unsigned long)callee_exp >> 32) & 0xFF; + codep[1] = ((unsigned long)callee_exp >> 40) & 0xFF; + codep[2] = ((unsigned long)callee_exp >> 48) & 0xFF; + codep[3] = ((unsigned long)callee_exp >> 56) & 0xFF; codep += 4; /* movb $beamArity, P_ARITY(%ebp); 3 or 6 bytes */ diff --git a/erts/emulator/hipe/hipe_amd64_abi.txt b/erts/emulator/hipe/hipe_amd64_abi.txt index 8a34bfa67f..72aed13995 100644 --- a/erts/emulator/hipe/hipe_amd64_abi.txt +++ b/erts/emulator/hipe/hipe_amd64_abi.txt @@ -45,7 +45,7 @@ The first return value from a function is placed in %rax, the second (if any) is placed in %rdx. Notes: -- Currently, NR_ARG_REGS==0. +- Currently, NR_ARG_REGS == 4. - C BIFs expect P in C parameter register 1: %rdi. By making Erlang parameter registers 1-5 coincide with C parameter registers 2-6, our BIF wrappers can simply move P to %rdi without having to shift diff --git a/erts/emulator/hipe/hipe_amd64_asm.m4 b/erts/emulator/hipe/hipe_amd64_asm.m4 index 7c81040b8b..ca55d5bf3b 100644 --- a/erts/emulator/hipe/hipe_amd64_asm.m4 +++ b/erts/emulator/hipe/hipe_amd64_asm.m4 @@ -33,7 +33,35 @@ define(HEAP_LIMIT_IN_REGISTER,0)dnl global for HL define(SIMULATE_NSP,0)dnl change to 1 to simulate call/ret insns `#define AMD64_LEAF_WORDS 'LEAF_WORDS -`#define LEAF_WORDS 'LEAF_WORDS +`#define LEAF_WORDS 'LEAF_WORDS +`#define AMD64_NR_ARG_REGS 'NR_ARG_REGS +`#define NR_ARG_REGS 'NR_ARG_REGS + +`#define AMD64_HP_IN_REGISTER 'HP_IN_REGISTER +`#if AMD64_HP_IN_REGISTER' +`#define AMD64_HEAP_POINTER 15' +define(HP,%r15)dnl Only change this together with above +`#endif' + +`#define AMD64_FCALLS_IN_REGISTER 'FCALLS_IN_REGISTER +`#if AMD64_FCALLS_IN_REGISTER' +`#define AMD64_FCALLS_REGISTER 11' +define(FCALLS,%r11)dnl This goes together with line above +`#endif' + +`#define AMD64_HEAP_LIMIT_IN_REGISTER 'HEAP_LIMIT_IN_REGISTER +`#if AMD64_HEAP_LIMIT_IN_REGISTER' +`#define AMD64_HEAP_LIMIT_REGISTER 12' +define(HEAP_LIMIT,%r12)dnl Change this together with line above +`#endif' + +`#define AMD64_SIMULATE_NSP 'SIMULATE_NSP + + +`#ifdef ASM' +/* + * Only assembler stuff from here on (when included from *.S) + */ /* * Workarounds for Darwin. @@ -63,33 +91,24 @@ ifelse(OPSYS,darwin,`` */ `#define P %rbp' -`#define AMD64_HP_IN_REGISTER 'HP_IN_REGISTER `#if AMD64_HP_IN_REGISTER -#define AMD64_HEAP_POINTER 15' -define(HP,%r15)dnl Only change this together with above -`#define SAVE_HP movq 'HP`, P_HP(P) +#define SAVE_HP movq 'HP`, P_HP(P) #define RESTORE_HP movq P_HP(P), 'HP` #else #define SAVE_HP /*empty*/ #define RESTORE_HP /*empty*/ #endif' -`#define AMD64_FCALLS_IN_REGISTER 'FCALLS_IN_REGISTER `#if AMD64_FCALLS_IN_REGISTER -#define AMD64_FCALLS_REGISTER 11' -define(FCALLS,%r11)dnl This goes together with line above -`#define SAVE_FCALLS movq 'FCALLS`, P_FCALLS(P) +#define SAVE_FCALLS movq 'FCALLS`, P_FCALLS(P) #define RESTORE_FCALLS movq P_FCALLS(P), 'FCALLS` #else #define SAVE_FCALLS /*empty*/ #define RESTORE_FCALLS /*empty*/ #endif' -`#define AMD64_HEAP_LIMIT_IN_REGISTER 'HEAP_LIMIT_IN_REGISTER `#if AMD64_HEAP_LIMIT_IN_REGISTER -#define AMD64_HEAP_LIMIT_REGISTER 12' -define(HEAP_LIMIT,%r12)dnl Change this together with line above -`#define RESTORE_HEAP_LIMIT movq P_HP_LIMIT(P), 'HEAP_LIMIT` +#define RESTORE_HEAP_LIMIT movq P_HP_LIMIT(P), 'HEAP_LIMIT` #else #define RESTORE_HEAP_LIMIT /*empty*/ #endif' @@ -99,7 +118,6 @@ define(NSP,%rsp)dnl `#define SAVE_CSP movq %rsp, P_CSP(P) #define RESTORE_CSP movq P_CSP(P), %rsp' -`#define AMD64_SIMULATE_NSP 'SIMULATE_NSP /* * Context switching macros. @@ -132,8 +150,6 @@ define(NSP,%rsp)dnl /* * Argument (parameter) registers. */ -`#define AMD64_NR_ARG_REGS 'NR_ARG_REGS -`#define NR_ARG_REGS 'NR_ARG_REGS define(defarg,`define(ARG$1,`$2')dnl #`define ARG'$1 $2' @@ -237,6 +253,10 @@ define(NBIF_ARG,`ifelse(eval($3 >= NR_ARG_REGS),0,`NBIF_REG_ARG($1,$3)',`NBIF_ST `/* #define NBIF_ARG_3_0 'NBIF_ARG(%rsi,3,0)` */' `/* #define NBIF_ARG_3_1 'NBIF_ARG(%rdx,3,1)` */' `/* #define NBIF_ARG_3_2 'NBIF_ARG(%rcx,3,2)` */' +`/* #define NBIF_ARG_4_0 'NBIF_ARG(%rsi,4,0)` */' +`/* #define NBIF_ARG_4_1 'NBIF_ARG(%rdx,4,1)` */' +`/* #define NBIF_ARG_4_2 'NBIF_ARG(%rcx,4,2)` */' +`/* #define NBIF_ARG_4_3 'NBIF_ARG(%r8,4,3)` */' `/* #define NBIF_ARG_5_0 'NBIF_ARG(%rsi,5,0)` */' `/* #define NBIF_ARG_5_1 'NBIF_ARG(%rdx,5,1)` */' `/* #define NBIF_ARG_5_2 'NBIF_ARG(%rcx,5,2)` */' @@ -261,6 +281,9 @@ define(NBIF_RET,`NBIF_RET_N(eval(RET_POP($1)))')dnl `/* #define NBIF_RET_1 'NBIF_RET(1)` */' `/* #define NBIF_RET_2 'NBIF_RET(2)` */' `/* #define NBIF_RET_3 'NBIF_RET(3)` */' +`/* #define NBIF_RET_4 'NBIF_RET(4)` */' `/* #define NBIF_RET_5 'NBIF_RET(5)` */' +`#endif /* ASM */' + `#endif /* HIPE_AMD64_ASM_H */' diff --git a/erts/emulator/hipe/hipe_amd64_bifs.m4 b/erts/emulator/hipe/hipe_amd64_bifs.m4 index a3219c7586..7d94aa05b3 100644 --- a/erts/emulator/hipe/hipe_amd64_bifs.m4 +++ b/erts/emulator/hipe/hipe_amd64_bifs.m4 @@ -18,7 +18,7 @@ changecom(`/*', `*/')dnl * %CopyrightEnd% */ - +#`define ASM' include(`hipe/hipe_amd64_asm.m4') #`include' "config.h" #`include' "hipe_literals.h" @@ -51,9 +51,10 @@ define(HANDLE_GOT_MBUF,` * standard_bif_interface_1(nbif_name, cbif_name) * standard_bif_interface_2(nbif_name, cbif_name) * standard_bif_interface_3(nbif_name, cbif_name) + * standard_bif_interface_4(nbif_name, cbif_name) * standard_bif_interface_0(nbif_name, cbif_name) * - * Generate native interface for a BIF with 0-3 parameters and + * Generate native interface for a BIF with 0-4 parameters and * standard failure mode. */ define(standard_bif_interface_1, @@ -154,6 +155,43 @@ ASYM($1): TYPE_FUNCTION(ASYM($1)) #endif') +define(standard_bif_interface_4, +` +#ifndef HAVE_$1 +#`define' HAVE_$1 + TEXT + .align 4 + GLOBAL(ASYM($1)) +ASYM($1): + /* set up the parameters */ + movq P, %rdi + NBIF_ARG(%rsi,4,0) + NBIF_ARG(%rdx,4,1) + NBIF_ARG(%rcx,4,2) + NBIF_ARG(%r8,4,3) + + /* make the call on the C stack */ + SWITCH_ERLANG_TO_C + pushq %r8 + pushq %rcx + pushq %rdx + pushq %rsi + movq %rsp, %rsi /* Eterm* BIF__ARGS */ + sub $(8), %rsp /* stack frame 16-byte alignment */ + CALL_BIF($2) + add $(4*8 + 8), %rsp + TEST_GOT_MBUF + SWITCH_C_TO_ERLANG + + /* throw exception if failure, otherwise return */ + TEST_GOT_EXN + jz nbif_4_simple_exception + NBIF_RET(4) + HANDLE_GOT_MBUF(4) + SET_SIZE(ASYM($1)) + TYPE_FUNCTION(ASYM($1)) +#endif') + define(standard_bif_interface_0, ` #ifndef HAVE_$1 diff --git a/erts/emulator/hipe/hipe_amd64_glue.S b/erts/emulator/hipe/hipe_amd64_glue.S index 8816906870..3cb0a2875b 100644 --- a/erts/emulator/hipe/hipe_amd64_glue.S +++ b/erts/emulator/hipe/hipe_amd64_glue.S @@ -17,10 +17,9 @@ * %CopyrightEnd% */ - +#define ASM #include "hipe_amd64_asm.h" #include "hipe_literals.h" -#define ASM #include "hipe_mode_switch.h" /* @@ -109,7 +108,7 @@ ASYM(nbif_return): * stub (hipe_x86_loader.erl) which should look as follows: * * stub for f/N: - * movq $<f's BEAM code address>, P_BEAM_IP(P) + * movq $<f's export entry address>, P_CALLEE_EXP(P) * movb $<N>, P_ARITY(P) * jmp nbif_callemu * @@ -119,7 +118,7 @@ ASYM(nbif_return): GLOBAL(ASYM(nbif_callemu)) ASYM(nbif_callemu): STORE_ARG_REGS - movl $HIPE_MODE_SWITCH_RES_CALL, %eax + movl $HIPE_MODE_SWITCH_RES_CALL_EXPORTED, %eax jmp .suspend_exit /* @@ -322,6 +321,7 @@ ASYM(nbif_fail): GLOBAL(nbif_1_gc_after_bif) GLOBAL(nbif_2_gc_after_bif) GLOBAL(nbif_3_gc_after_bif) + GLOBAL(nbif_4_gc_after_bif) .align 4 nbif_0_gc_after_bif: xorl %edx, %edx @@ -337,6 +337,10 @@ nbif_2_gc_after_bif: .align 4 nbif_3_gc_after_bif: movl $3, %edx + jmp .gc_after_bif + .align 4 +nbif_4_gc_after_bif: + movl $4, %edx /*FALLTHROUGH*/ .align 4 .gc_after_bif: @@ -360,6 +364,7 @@ nbif_3_gc_after_bif: GLOBAL(nbif_1_simple_exception) GLOBAL(nbif_2_simple_exception) GLOBAL(nbif_3_simple_exception) + GLOBAL(nbif_4_simple_exception) .align 4 nbif_0_simple_exception: xorl %eax, %eax @@ -375,6 +380,10 @@ nbif_2_simple_exception: .align 4 nbif_3_simple_exception: movl $3, %eax + jmp .nbif_simple_exception + .align 4 +nbif_4_simple_exception: + movl $4, %eax /*FALLTHROUGH*/ .align 4 .nbif_simple_exception: diff --git a/erts/emulator/hipe/hipe_arch.h b/erts/emulator/hipe/hipe_arch.h index 04ed980126..b45209b3f7 100644 --- a/erts/emulator/hipe/hipe_arch.h +++ b/erts/emulator/hipe/hipe_arch.h @@ -29,6 +29,7 @@ extern void hipe_patch_load_fe(Uint *address, Uint value); extern int hipe_patch_insn(void *address, Uint value, Eterm type); extern int hipe_patch_call(void *callAddress, void *destAddress, void *trampoline); +extern void *hipe_alloc_code(Uint nrbytes, Eterm callees, Eterm *trampolines, Process *p); extern void *hipe_make_native_stub(void *beamAddress, unsigned int beamArity); #if defined(__sparc__) diff --git a/erts/emulator/hipe/hipe_arm.c b/erts/emulator/hipe/hipe_arm.c index 3db3ffe9b1..c0c6305c68 100644 --- a/erts/emulator/hipe/hipe_arm.c +++ b/erts/emulator/hipe/hipe_arm.c @@ -260,9 +260,9 @@ int hipe_patch_insn(void *address, Uint32 value, Eterm type) return 0; } -/* called from hipe_bif0.c:hipe_bifs_make_native_stub_2() - and hipe_bif0.c:hipe_make_stub() */ -void *hipe_make_native_stub(void *beamAddress, unsigned int beamArity) +/* Make stub for native code calling exported beam function +*/ +void *hipe_make_native_stub(void *callee_exp, unsigned int beamArity) { unsigned int *code; unsigned int *tramp_callemu; @@ -272,9 +272,9 @@ void *hipe_make_native_stub(void *beamAddress, unsigned int beamArity) * Native code calls BEAM via a stub looking as follows: * * mov r0, #beamArity - * ldr r8, [pc,#0] // beamAddress + * ldr r8, [pc,#0] // callee_exp * b nbif_callemu - * .long beamAddress + * .long callee_exp * * I'm using r0 and r8 since they aren't used for * parameter passing in native code. The branch to @@ -283,6 +283,8 @@ void *hipe_make_native_stub(void *beamAddress, unsigned int beamArity) */ code = alloc_stub(4, &tramp_callemu); + if (!code) + return NULL; callemu_offset = ((int)&nbif_callemu - ((int)&code[2] + 8)) >> 2; if (!(callemu_offset >= -0x00800000 && callemu_offset <= 0x007FFFFF)) { callemu_offset = ((int)tramp_callemu - ((int)&code[2] + 8)) >> 2; @@ -292,12 +294,12 @@ void *hipe_make_native_stub(void *beamAddress, unsigned int beamArity) /* mov r0, #beamArity */ code[0] = 0xE3A00000 | (beamArity & 0xFF); - /* ldr r8, [pc,#0] // beamAddress */ + /* ldr r8, [pc,#0] // callee_exp */ code[1] = 0xE59F8000; /* b nbif_callemu */ code[2] = 0xEA000000 | (callemu_offset & 0x00FFFFFF); - /* .long beamAddress */ - code[3] = (unsigned int)beamAddress; + /* .long callee_exp */ + code[3] = (unsigned int)callee_exp; hipe_flush_icache_range(code, 4*sizeof(int)); diff --git a/erts/emulator/hipe/hipe_arm.h b/erts/emulator/hipe/hipe_arm.h index 19f2a986cf..b9cd1a750c 100644 --- a/erts/emulator/hipe/hipe_arm.h +++ b/erts/emulator/hipe/hipe_arm.h @@ -40,8 +40,4 @@ static __inline__ int hipe_word32_address_ok(void *address) extern void hipe_arm_inc_stack(void); -/* for hipe_bifs_enter_code_2 */ -extern void *hipe_alloc_code(Uint nrbytes, Eterm callees, Eterm *trampolines, Process *p); -#define HIPE_ALLOC_CODE(n,c,t,p) hipe_alloc_code((n),(c),(t),(p)) - #endif /* HIPE_ARM_H */ diff --git a/erts/emulator/hipe/hipe_arm_asm.m4 b/erts/emulator/hipe/hipe_arm_asm.m4 index 85dc84973d..b2e3f83d1e 100644 --- a/erts/emulator/hipe/hipe_arm_asm.m4 +++ b/erts/emulator/hipe/hipe_arm_asm.m4 @@ -29,6 +29,14 @@ define(LEAF_WORDS,16)dnl number of stack words for leaf functions define(NR_ARG_REGS,3)dnl admissible values are 0 to 6, inclusive `#define ARM_LEAF_WORDS 'LEAF_WORDS +`#define ARM_NR_ARG_REGS 'NR_ARG_REGS +`#define NR_ARG_REGS 'NR_ARG_REGS + + +`#ifdef ASM' +/* + * Only assembler stuff from here on (when included from *.S) + */ /* * Reserved registers. @@ -77,8 +85,6 @@ define(NR_ARG_REGS,3)dnl admissible values are 0 to 6, inclusive /* * Argument (parameter) registers. */ -`#define ARM_NR_ARG_REGS 'NR_ARG_REGS -`#define NR_ARG_REGS 'NR_ARG_REGS define(defarg,`define(ARG$1,`$2')dnl #`define ARG'$1 $2' @@ -195,4 +201,6 @@ define(QUICK_CALL_RET,`NBIF_POP_N(eval(RET_POP($2)))b $1')dnl `/* #define QUICK_CALL_RET_F_3 'QUICK_CALL_RET(F,3)` */' `/* #define QUICK_CALL_RET_F_5 'QUICK_CALL_RET(F,5)` */' +`#endif /* ASM */' + `#endif /* HIPE_ARM_ASM_H */' diff --git a/erts/emulator/hipe/hipe_arm_bifs.m4 b/erts/emulator/hipe/hipe_arm_bifs.m4 index bd8bc5ab6b..884240be9c 100644 --- a/erts/emulator/hipe/hipe_arm_bifs.m4 +++ b/erts/emulator/hipe/hipe_arm_bifs.m4 @@ -19,12 +19,14 @@ changecom(`/*', `*/')dnl */ +#`define ASM' include(`hipe/hipe_arm_asm.m4') #`include' "config.h" #`include' "hipe_literals.h" .text .p2align 2 + .arm `#if defined(ERTS_ENABLE_LOCK_CHECK) && defined(ERTS_SMP) # define CALL_BIF(F) ldr r14, =F; str r14, [r0, #P_BIF_CALLEE]; bl hipe_debug_bif_wrapper @@ -391,7 +393,14 @@ $1: mov r0, P /* Perform a quick save;call;restore;ret sequence. */ +#ifdef __thumb__ + SAVE_CONTEXT_QUICK + bl $2 + RESTORE_CONTEXT_QUICK + NBIF_RET(0) +#else QUICK_CALL_RET($2,0) +#endif .size $1, .-$1 .type $1, %function #endif') @@ -407,7 +416,14 @@ $1: NBIF_ARG(r1,1,0) /* Perform a quick save;call;restore;ret sequence. */ +#ifdef __thumb__ + SAVE_CONTEXT_QUICK + bl $2 + RESTORE_CONTEXT_QUICK + NBIF_RET(1) +#else QUICK_CALL_RET($2,1) +#endif .size $1, .-$1 .type $1, %function #endif') @@ -424,7 +440,14 @@ $1: NBIF_ARG(r2,2,1) /* Perform a quick save;call;restore;ret sequence. */ +#ifdef __thumb__ + SAVE_CONTEXT_QUICK + bl $2 + RESTORE_CONTEXT_QUICK + NBIF_RET(2) +#else QUICK_CALL_RET($2,2) +#endif .size $1, .-$1 .type $1, %function #endif') @@ -442,7 +465,14 @@ $1: NBIF_ARG(r3,3,2) /* Perform a quick save;call;restore;ret sequence. */ +#ifdef __thumb__ + SAVE_CONTEXT_QUICK + bl $2 + RESTORE_CONTEXT_QUICK + NBIF_RET(3) +#else QUICK_CALL_RET($2,3) +#endif .size $1, .-$1 .type $1, %function #endif') @@ -466,7 +496,14 @@ $1: NBIF_ARG(r3,5,2) /* Perform a quick save;call;restore;ret sequence. */ +#ifdef __thumb__ + SAVE_CONTEXT_QUICK + bl $2 + RESTORE_CONTEXT_QUICK + NBIF_RET(5) +#else QUICK_CALL_RET($2,5) +#endif .size $1, .-$1 .type $1, %function #endif') @@ -488,9 +525,16 @@ define(noproc_primop_interface_0, #`define' HAVE_$1 .global $1 $1: - /* XXX: this case is always trivial; how to suppress the branch? */ /* Perform a quick save;call;restore;ret sequence. */ +#ifdef __thumb__ + SAVE_CONTEXT_QUICK + bl $2 + RESTORE_CONTEXT_QUICK + NBIF_RET(0) +#else + /* XXX: this case is always trivial; how to suppress the branch? */ QUICK_CALL_RET($2,0) +#endif .size $1, .-$1 .type $1, %function #endif') @@ -505,7 +549,14 @@ $1: NBIF_ARG(r0,1,0) /* Perform a quick save;call;restore;ret sequence. */ +#ifdef __thumb__ + SAVE_CONTEXT_QUICK + bl $2 + RESTORE_CONTEXT_QUICK + NBIF_RET(1) +#else QUICK_CALL_RET($2,1) +#endif .size $1, .-$1 .type $1, %function #endif') @@ -521,7 +572,14 @@ $1: NBIF_ARG(r1,2,1) /* Perform a quick save;call;restore;ret sequence. */ +#ifdef __thumb__ + SAVE_CONTEXT_QUICK + bl $2 + RESTORE_CONTEXT_QUICK + NBIF_RET(2) +#else QUICK_CALL_RET($2,2) +#endif .size $1, .-$1 .type $1, %function #endif') @@ -538,7 +596,14 @@ $1: NBIF_ARG(r2,3,2) /* Perform a quick save;call;restore;ret sequence. */ +#ifdef __thumb__ + SAVE_CONTEXT_QUICK + bl $2 + RESTORE_CONTEXT_QUICK + NBIF_RET(3) +#else QUICK_CALL_RET($2,3) +#endif .size $1, .-$1 .type $1, %function #endif') @@ -558,7 +623,14 @@ $1: str r4, [sp, #0] /* Perform a quick save;call;restore;ret sequence. */ +#ifdef __thumb__ + SAVE_CONTEXT_QUICK + bl $2 + RESTORE_CONTEXT_QUICK + NBIF_RET(5) +#else QUICK_CALL_RET($2,5) +#endif .size $1, .-$1 .type $1, %function #endif') diff --git a/erts/emulator/hipe/hipe_arm_glue.S b/erts/emulator/hipe/hipe_arm_glue.S index 2e2b8604a6..e7ff267606 100644 --- a/erts/emulator/hipe/hipe_arm_glue.S +++ b/erts/emulator/hipe/hipe_arm_glue.S @@ -17,14 +17,14 @@ * %CopyrightEnd% */ - +#define ASM #include "hipe_arm_asm.h" #include "hipe_literals.h" -#define ASM #include "hipe_mode_switch.h" .text .p2align 2 + .arm /* * Enter Erlang from C. @@ -70,6 +70,7 @@ * Emulated code recursively calls native code. */ .global hipe_arm_call_to_native + .type hipe_arm_call_to_native, %function hipe_arm_call_to_native: ENTER_FROM_C /* get argument registers */ @@ -85,6 +86,7 @@ hipe_arm_call_to_native: * This is where native code returns to emulated code. */ .global nbif_return + .type nbif_return, %function nbif_return: str r0, [P, #P_ARG0] /* save retval */ mov r0, #HIPE_MODE_SWITCH_RES_RETURN @@ -95,6 +97,7 @@ nbif_return: * Emulated code returns to its native code caller. */ .global hipe_arm_return_to_native + .type hipe_arm_return_to_native, %function hipe_arm_return_to_native: ENTER_FROM_C /* get return value */ @@ -111,6 +114,7 @@ hipe_arm_return_to_native: * Emulated code tailcalls native code. */ .global hipe_arm_tailcall_to_native + .type hipe_arm_tailcall_to_native, %function hipe_arm_tailcall_to_native: ENTER_FROM_C /* get argument registers */ @@ -125,6 +129,7 @@ hipe_arm_tailcall_to_native: * Emulated code throws an exception to its native code caller. */ .global hipe_arm_throw_to_native + .type hipe_arm_throw_to_native, %function hipe_arm_throw_to_native: ENTER_FROM_C /* invoke the handler */ @@ -135,24 +140,26 @@ hipe_arm_throw_to_native: * which should look as follows: * * stub for f/N: - * <set r8 to f's BEAM code address> + * <set r8 to f's export entry address> * <set r0 to N> * b nbif_callemu * * XXX: Different stubs for different number of register parameters? */ .global nbif_callemu + .type nbif_callemu, %function nbif_callemu: - str r8, [P, #P_BEAM_IP] + str r8, [P, #P_CALLEE_EXP] str r0, [P, #P_ARITY] STORE_ARG_REGS - mov r0, #HIPE_MODE_SWITCH_RES_CALL + mov r0, #HIPE_MODE_SWITCH_RES_CALL_EXPORTED b .suspend_exit /* * nbif_apply */ .global nbif_apply + .type nbif_apply, %function nbif_apply: STORE_ARG_REGS mov r0, #HIPE_MODE_SWITCH_RES_APPLY @@ -169,6 +176,7 @@ nbif_apply: */ #if NR_ARG_REGS >= 6 .global nbif_ccallemu6 + .type nbif_ccallemu6, %function nbif_ccallemu6: str ARG5, [P, #P_ARG5] #if NR_ARG_REGS > 6 @@ -181,6 +189,7 @@ nbif_ccallemu6: #if NR_ARG_REGS >= 5 .global nbif_ccallemu5 + .type nbif_ccallemu5, %function nbif_ccallemu5: str ARG4, [P, #P_ARG4] #if NR_ARG_REGS > 5 @@ -193,6 +202,7 @@ nbif_ccallemu5: #if NR_ARG_REGS >= 4 .global nbif_ccallemu4 + .type nbif_ccallemu4, %function nbif_ccallemu4: str ARG3, [P, #P_ARG3] #if NR_ARG_REGS > 4 @@ -205,6 +215,7 @@ nbif_ccallemu4: #if NR_ARG_REGS >= 3 .global nbif_ccallemu3 + .type nbif_ccallemu3, %function nbif_ccallemu3: str ARG2, [P, #P_ARG2] #if NR_ARG_REGS > 3 @@ -217,6 +228,7 @@ nbif_ccallemu3: #if NR_ARG_REGS >= 2 .global nbif_ccallemu2 + .type nbif_ccallemu2, %function nbif_ccallemu2: str ARG1, [P, #P_ARG1] #if NR_ARG_REGS > 2 @@ -229,6 +241,7 @@ nbif_ccallemu2: #if NR_ARG_REGS >= 1 .global nbif_ccallemu1 + .type nbif_ccallemu1, %function nbif_ccallemu1: str ARG0, [P, #P_ARG0] #if NR_ARG_REGS > 1 @@ -240,6 +253,7 @@ nbif_ccallemu1: #endif .global nbif_ccallemu0 + .type nbif_ccallemu0, %function nbif_ccallemu0: /* We use r1 not ARG0 here because ARG0 is not defined when NR_ARG_REGS == 0. */ @@ -254,6 +268,7 @@ nbif_ccallemu0: * This is where native code suspends. */ .global nbif_suspend_0 + .type nbif_suspend_0, %function nbif_suspend_0: mov r0, #HIPE_MODE_SWITCH_RES_SUSPEND b .suspend_exit @@ -262,6 +277,7 @@ nbif_suspend_0: * Suspend from a receive (waiting for a message) */ .global nbif_suspend_msg + .type nbif_suspend_msg, %function nbif_suspend_msg: mov r0, #HIPE_MODE_SWITCH_RES_WAIT b .suspend_exit @@ -272,6 +288,7 @@ nbif_suspend_msg: * else { return 0; } */ .global nbif_suspend_msg_timeout + .type nbif_suspend_msg_timeout, %function nbif_suspend_msg_timeout: ldr r1, [P, #P_FLAGS] mov r0, #HIPE_MODE_SWITCH_RES_WAIT_TIMEOUT @@ -286,23 +303,31 @@ nbif_suspend_msg_timeout: * This is the default exception handler for native code. */ .global nbif_fail + .type nbif_fail, %function nbif_fail: mov r0, #HIPE_MODE_SWITCH_RES_THROW b .flush_exit /* no need to save RA */ .global nbif_0_gc_after_bif - .global nbif_1_gc_after_bif - .global nbif_2_gc_after_bif - .global nbif_3_gc_after_bif + .type nbif_0_gc_after_bif, %function nbif_0_gc_after_bif: mov r1, #0 b .gc_after_bif + + .global nbif_1_gc_after_bif + .type nbif_1_gc_after_bif, %function nbif_1_gc_after_bif: mov r1, #1 b .gc_after_bif + + .global nbif_2_gc_after_bif + .type nbif_2_gc_after_bif, %function nbif_2_gc_after_bif: mov r1, #2 b .gc_after_bif + + .global nbif_3_gc_after_bif + .type nbif_3_gc_after_bif, %function nbif_3_gc_after_bif: mov r1, #3 /*FALLTHROUGH*/ @@ -330,18 +355,25 @@ nbif_3_gc_after_bif: * TEMP_LR contains a copy of LR */ .global nbif_0_simple_exception + .type nbif_0_simple_exception, %function nbif_0_simple_exception: mov r1, #0 b .nbif_simple_exception + .global nbif_1_simple_exception + .type nbif_1_simple_exception, %function nbif_1_simple_exception: mov r1, #1 b .nbif_simple_exception + .global nbif_2_simple_exception + .type nbif_2_simple_exception, %function nbif_2_simple_exception: mov r1, #2 b .nbif_simple_exception + .global nbif_3_simple_exception + .type nbif_3_simple_exception, %function nbif_3_simple_exception: mov r1, #3 /*FALLTHROUGH*/ @@ -385,6 +417,7 @@ nbif_3_simple_exception: * the gray/white stack boundary */ .global nbif_stack_trap_ra + .type nbif_stack_trap_ra, %function nbif_stack_trap_ra: /* a return address, not a function */ # This only handles a single return value. # If we have more, we need to save them in the PCB. @@ -401,6 +434,7 @@ nbif_stack_trap_ra: /* a return address, not a function */ * Caller saved its LR in TEMP_LR (== TEMP1) before calling us. */ .global hipe_arm_inc_stack + .type hipe_arm_inc_stack, %function hipe_arm_inc_stack: STORE_ARG_REGS mov TEMP_ARG0, lr diff --git a/erts/emulator/hipe/hipe_bif0.c b/erts/emulator/hipe/hipe_bif0.c index c9eee2acf2..6c1de05a4c 100644 --- a/erts/emulator/hipe/hipe_bif0.c +++ b/erts/emulator/hipe/hipe_bif0.c @@ -89,25 +89,6 @@ static Eterm address_to_term(const void *address, Process *p) /* * BIFs for reading and writing memory. Used internally by HiPE. */ -#if 0 /* XXX: unused */ -BIF_RETTYPE hipe_bifs_read_u8_1(BIF_ALIST_1) -{ - unsigned char *address = term_to_address(BIF_ARG_1); - if (!address) - BIF_ERROR(BIF_P, BADARG); - BIF_RET(make_small(*address)); -} -#endif - -#if 0 /* XXX: unused */ -BIF_RETTYPE hipe_bifs_read_u32_1(BIF_ALIST_1) -{ - Uint32 *address = term_to_address(BIF_ARG_1); - if (!address || !hipe_word32_address_ok(address)) - BIF_ERROR(BIF_P, BADARG); - BIF_RET(Uint_to_term(*address, BIF_P)); -} -#endif BIF_RETTYPE hipe_bifs_write_u8_2(BIF_ALIST_2) { @@ -120,22 +101,6 @@ BIF_RETTYPE hipe_bifs_write_u8_2(BIF_ALIST_2) BIF_RET(NIL); } -#if 0 /* XXX: unused */ -BIF_RETTYPE hipe_bifs_write_s32_2(BIF_ALIST_2) -{ - Sint32 *address; - Sint value; - - address = term_to_address(BIF_ARG_1); - if (!address || !hipe_word32_address_ok(address)) - BIF_ERROR(BIF_P, BADARG); - if (!term_to_Sint32(BIF_ARG_2, &value)) - BIF_ERROR(BIF_P, BADARG); - *address = value; - BIF_RET(NIL); -} -#endif - BIF_RETTYPE hipe_bifs_write_u32_2(BIF_ALIST_2) { Uint32 *address; @@ -432,15 +397,17 @@ BIF_RETTYPE hipe_bifs_enter_code_2(BIF_ALIST_2) ASSERT(bitoffs == 0); ASSERT(bitsize == 0); trampolines = NIL; -#ifdef HIPE_ALLOC_CODE - address = HIPE_ALLOC_CODE(nrbytes, BIF_ARG_2, &trampolines, BIF_P); - if (!address) - BIF_ERROR(BIF_P, BADARG); -#else - if (is_not_nil(BIF_ARG_2)) - BIF_ERROR(BIF_P, BADARG); - address = erts_alloc(ERTS_ALC_T_HIPE, nrbytes); -#endif + address = hipe_alloc_code(nrbytes, BIF_ARG_2, &trampolines, BIF_P); + if (!address) { + Uint nrcallees; + + if (is_tuple(BIF_ARG_2)) + nrcallees = arityval(tuple_val(BIF_ARG_2)[0]); + else + nrcallees = 0; + erl_exit(1, "%s: failed to allocate %lu bytes and %lu trampolines\r\n", + __func__, (unsigned long)nrbytes, (unsigned long)nrcallees); + } memcpy(address, bytes, nrbytes); hipe_flush_icache_range(address, nrbytes); hp = HAlloc(BIF_P, 3); @@ -639,33 +606,6 @@ BIF_RETTYPE hipe_bifs_fun_to_address_1(BIF_ALIST_1) BIF_RET(address_to_term(pc, BIF_P)); } -static void *hipe_get_emu_address(Eterm m, Eterm f, unsigned int arity, int is_remote) -{ - void *address = NULL; - if (!is_remote) - address = hipe_find_emu_address(m, f, arity); - if (!address) { - /* if not found, stub it via the export entry */ - /* no lock needed around erts_export_get_or_make_stub() */ - Export *export_entry = erts_export_get_or_make_stub(m, f, arity); - address = export_entry->addressv[erts_active_code_ix()]; - } - return address; -} - -#if 0 /* XXX: unused */ -BIF_RETTYPE hipe_bifs_get_emu_address_1(BIF_ALIST_1) -{ - struct mfa mfa; - void *address; - - if (!term_to_mfa(BIF_ARG_1, &mfa)) - BIF_ERROR(BIF_P, BADARG); - address = hipe_get_emu_address(mfa.mod, mfa.fun, mfa.ari); - BIF_RET(address_to_term(address, BIF_P)); -} -#endif - BIF_RETTYPE hipe_bifs_set_native_address_3(BIF_ALIST_3) { Eterm *pc; @@ -713,33 +653,6 @@ BIF_RETTYPE hipe_bifs_set_native_address_3(BIF_ALIST_3) BIF_RET(am_false); } -#if 0 /* XXX: unused */ -/* - * hipe_bifs_address_to_fun(Address) - * - Address is the address of the start of a emu function's code - * - returns {Module, Function, Arity} - */ -BIF_RETTYPE hipe_bifs_address_to_fun_1(BIF_ALIST_1) -{ - Eterm *pc; - Eterm *funcinfo; - Eterm *hp; - - pc = term_to_address(BIF_ARG_1); - if (!pc) - BIF_ERROR(BIF_P, BADARG); - funcinfo = find_function_from_pc(pc); - if (!funcinfo) - BIF_RET(am_false); - hp = HAlloc(BIF_P, 4); - hp[0] = make_arityval(3); - hp[1] = funcinfo[0]; - hp[2] = funcinfo[1]; - hp[3] = make_small(funcinfo[2]); - BIF_RET(make_tuple(hp)); -} -#endif - BIF_RETTYPE hipe_bifs_enter_sdesc_1(BIF_ALIST_1) { struct sdesc *sdesc; @@ -948,37 +861,6 @@ BIF_RETTYPE hipe_bifs_primop_address_1(BIF_ALIST_1) BIF_RET(address_to_term(primop->address, BIF_P)); } -#if 0 /* XXX: unused */ -/* - * hipe_bifs_gbif_address(F,A) -> address or false - */ -#define GBIF_LIST(ATOM,ARY,CFUN) extern Eterm gbif_##CFUN(void); -#include "hipe_gbif_list.h" -#undef GBIF_LIST - -BIF_RETTYPE hipe_bifs_gbif_address_2(BIF_ALIST_2) -{ - Uint arity; - void *address; - - if (is_not_atom(BIF_ARG_1) || is_not_small(BIF_ARG_2)) - BIF_RET(am_false); /* error or false, does it matter? */ - arity = signed_val(BIF_ARG_2); - /* XXX: replace with a hash table later */ - do { /* trick to let us use 'break' instead of 'goto' */ -#define GBIF_LIST(ATOM,ARY,CFUN) if (BIF_ARG_1 == ATOM && arity == ARY) { address = CFUN; break; } -#include "hipe_gbif_list.h" -#undef GBIF_LIST - printf("\r\n%s: guard BIF ", __FUNCTION__); - fflush(stdout); - erts_printf("%T", BIF_ARG_1); - printf("/%lu isn't listed in hipe_gbif_list.h\r\n", arity); - BIF_RET(am_false); - } while (0); - BIF_RET(address_to_term(address, BIF_P)); -} -#endif - BIF_RETTYPE hipe_bifs_atom_to_word_1(BIF_ALIST_1) { if (is_not_atom(BIF_ARG_1)) @@ -1028,77 +910,6 @@ void hipe_emulate_fpe(Process* p) } #endif -#if 0 /* XXX: unused */ -/* - * At least parts of this should be inlined in native code. - * The rest could be made a primop used by both the emulator and - * native code... - */ -BIF_RETTYPE hipe_bifs_make_fun_3(BIF_ALIST_3) -{ - Eterm free_vars; - Eterm mod; - Eterm *tp; - Uint index; - Uint uniq; - Uint num_free; - Eterm tmp_var; - Uint *tmp_ptr; - unsigned needed; - ErlFunThing *funp; - Eterm *hp; - int i; - - if (is_not_list(BIF_ARG_1) && is_not_nil(BIF_ARG_1)) - BIF_ERROR(BIF_P, BADARG); - free_vars = BIF_ARG_1; - - if (is_not_atom(BIF_ARG_2)) - BIF_ERROR(BIF_P, BADARG); - mod = BIF_ARG_2; - - if (is_not_tuple(BIF_ARG_3) || - (arityval(*tuple_val(BIF_ARG_3)) != 3)) - BIF_ERROR(BIF_P, BADARG); - tp = tuple_val(BIF_ARG_3); - - if (term_to_Uint(tp[1], &index) == 0) - BIF_ERROR(BIF_P, BADARG); - if (term_to_Uint(tp[2], &uniq) == 0) - BIF_ERROR(BIF_P, BADARG); - if (term_to_Uint(tp[3], &num_free) == 0) - BIF_ERROR(BIF_P, BADARG); - - needed = ERL_FUN_SIZE + num_free; - funp = (ErlFunThing *) HAlloc(BIF_P, needed); - hp = funp->env; - - funp->thing_word = HEADER_FUN; - - /* Need a ErlFunEntry *fe - * fe->refc++; - * funp->fe = fe; - */ - - funp->num_free = num_free; - funp->creator = BIF_P->id; - for (i = 0; i < num_free; i++) { - if (is_nil(free_vars)) - BIF_ERROR(BIF_P, BADARG); - tmp_ptr = list_val(free_vars); - tmp_var = CAR(tmp_ptr); - free_vars = CDR(tmp_ptr); - *hp++ = tmp_var; - } - if (is_not_nil(free_vars)) - BIF_ERROR(BIF_P, BADARG); - - funp->next = MSO(BIF_P).funs; - MSO(BIF_P).funs = funp; - - BIF_RET(make_fun(funp)); -} -#endif /* * args: Module, {Uniq, Index, BeamAddress} @@ -1163,22 +974,6 @@ BIF_RETTYPE hipe_bifs_set_native_address_in_fe_2(BIF_ALIST_2) BIF_RET(am_true); } -#if 0 /* XXX: unused */ -BIF_RETTYPE hipe_bifs_make_native_stub_2(BIF_ALIST_2) -{ - void *beamAddress; - Uint beamArity; - void *stubAddress; - - if ((beamAddress = term_to_address(BIF_ARG_1)) == 0 || - is_not_small(BIF_ARG_2) || - (beamArity = unsigned_val(BIF_ARG_2)) >= 256) - BIF_ERROR(BIF_P, BADARG); - stubAddress = hipe_make_native_stub(beamAddress, beamArity); - BIF_RET(address_to_term(stubAddress, BIF_P)); -} -#endif - /* * MFA info hash table: * - maps MFA to native code entry point @@ -1217,22 +1012,32 @@ static struct { * they create a new stub for the mfa, which forces locking. * XXX: Redesign apply et al to avoid those updates. */ - erts_smp_mtx_t lock; + erts_smp_rwmtx_t lock; } hipe_mfa_info_table; static inline void hipe_mfa_info_table_init_lock(void) { - erts_smp_mtx_init(&hipe_mfa_info_table.lock, "hipe_mfait_lock"); + erts_smp_rwmtx_init(&hipe_mfa_info_table.lock, "hipe_mfait_lock"); } -static inline void hipe_mfa_info_table_lock(void) +static inline void hipe_mfa_info_table_rlock(void) { - erts_smp_mtx_lock(&hipe_mfa_info_table.lock); + erts_smp_rwmtx_rlock(&hipe_mfa_info_table.lock); } -static inline void hipe_mfa_info_table_unlock(void) +static inline void hipe_mfa_info_table_runlock(void) { - erts_smp_mtx_unlock(&hipe_mfa_info_table.lock); + erts_smp_rwmtx_runlock(&hipe_mfa_info_table.lock); +} + +static inline void hipe_mfa_info_table_rwlock(void) +{ + erts_smp_rwmtx_rwlock(&hipe_mfa_info_table.lock); +} + +static inline void hipe_mfa_info_table_rwunlock(void) +{ + erts_smp_rwmtx_rwunlock(&hipe_mfa_info_table.lock); } #define HIPE_MFA_HASH(M,F,A) ((M) * (F) + (A)) @@ -1333,7 +1138,7 @@ void *hipe_mfa_find_na(Eterm m, Eterm f, unsigned int arity) } #endif -static struct hipe_mfa_info *hipe_mfa_info_table_put_locked(Eterm m, Eterm f, unsigned int arity) +static struct hipe_mfa_info *hipe_mfa_info_table_put_rwlocked(Eterm m, Eterm f, unsigned int arity) { unsigned long h; unsigned int i; @@ -1362,8 +1167,8 @@ static void hipe_mfa_set_na(Eterm m, Eterm f, unsigned int arity, void *address, { struct hipe_mfa_info *p; - hipe_mfa_info_table_lock(); - p = hipe_mfa_info_table_put_locked(m, f, arity); + hipe_mfa_info_table_rwlock(); + p = hipe_mfa_info_table_put_rwlocked(m, f, arity); #ifdef DEBUG_LINKER printf("%s: ", __FUNCTION__); print_mfa(m, f, arity); @@ -1372,7 +1177,7 @@ static void hipe_mfa_set_na(Eterm m, Eterm f, unsigned int arity, void *address, p->local_address = address; if (is_exported) p->remote_address = address; - hipe_mfa_info_table_unlock(); + hipe_mfa_info_table_rwunlock(); } #if defined(__powerpc__) || defined(__ppc__) || defined(__powerpc64__) || defined(__arm__) @@ -1381,10 +1186,10 @@ void *hipe_mfa_get_trampoline(Eterm m, Eterm f, unsigned int arity) struct hipe_mfa_info *p; void *trampoline; - hipe_mfa_info_table_lock(); - p = hipe_mfa_info_table_put_locked(m, f, arity); - trampoline = p->trampoline; - hipe_mfa_info_table_unlock(); + hipe_mfa_info_table_rlock(); + p = hipe_mfa_info_table_get_locked(m, f, arity); + trampoline = p ? p->trampoline : NULL; + hipe_mfa_info_table_runlock(); return trampoline; } @@ -1392,10 +1197,10 @@ void hipe_mfa_set_trampoline(Eterm m, Eterm f, unsigned int arity, void *trampol { struct hipe_mfa_info *p; - hipe_mfa_info_table_lock(); - p = hipe_mfa_info_table_put_locked(m, f, arity); + hipe_mfa_info_table_rwlock(); + p = hipe_mfa_info_table_put_rwlocked(m, f, arity); p->trampoline = trampoline; - hipe_mfa_info_table_unlock(); + hipe_mfa_info_table_rwunlock(); } #endif @@ -1426,7 +1231,7 @@ BIF_RETTYPE hipe_bifs_invalidate_funinfo_native_addresses_1(BIF_ALIST_1) struct mfa mfa; struct hipe_mfa_info *p; - hipe_mfa_info_table_lock(); + hipe_mfa_info_table_rwlock(); lst = BIF_ARG_1; while (is_list(lst)) { if (!term_to_mfa(CAR(list_val(lst)), &mfa)) @@ -1455,7 +1260,7 @@ BIF_RETTYPE hipe_bifs_invalidate_funinfo_native_addresses_1(BIF_ALIST_1) } } } - hipe_mfa_info_table_unlock(); + hipe_mfa_info_table_rwunlock(); if (is_not_nil(lst)) BIF_ERROR(BIF_P, BADARG); BIF_RET(NIL); @@ -1469,8 +1274,8 @@ void hipe_mfa_save_orig_beam_op(Eterm mod, Eterm fun, unsigned int ari, Eterm *p orig_beam_op = pc[0]; if (orig_beam_op != BeamOpCode(op_hipe_trap_call_closure) && orig_beam_op != BeamOpCode(op_hipe_trap_call)) { - hipe_mfa_info_table_lock(); - p = hipe_mfa_info_table_put_locked(mod, fun, ari); + hipe_mfa_info_table_rwlock(); + p = hipe_mfa_info_table_put_rwlocked(mod, fun, ari); #ifdef DEBUG_LINKER printf("%s: ", __FUNCTION__); print_mfa(mod, fun, ari); @@ -1478,7 +1283,7 @@ void hipe_mfa_save_orig_beam_op(Eterm mod, Eterm fun, unsigned int ari, Eterm *p #endif p->beam_code = pc; p->orig_beam_op = orig_beam_op; - hipe_mfa_info_table_unlock(); + hipe_mfa_info_table_rwunlock(); } else { #ifdef DEBUG_LINKER printf("%s: ", __FUNCTION__); @@ -1490,22 +1295,19 @@ void hipe_mfa_save_orig_beam_op(Eterm mod, Eterm fun, unsigned int ari, Eterm *p static void *hipe_make_stub(Eterm m, Eterm f, unsigned int arity, int is_remote) { - void *BEAMAddress; + Export *export_entry; void *StubAddress; -#if 0 - if (is_not_atom(m) || is_not_atom(f) || arity > 255) - return NULL; -#endif - BEAMAddress = hipe_get_emu_address(m, f, arity, is_remote); - StubAddress = hipe_make_native_stub(BEAMAddress, arity); -#if 0 - hipe_mfa_set_na(m, f, arity, StubAddress); -#endif + ASSERT(is_remote); + + export_entry = erts_export_get_or_make_stub(m, f, arity); + StubAddress = hipe_make_native_stub(export_entry, arity); + if (!StubAddress) + erl_exit(1, "hipe_make_stub: code allocation failed\r\n"); return StubAddress; } -static void *hipe_get_na_nofail_locked(Eterm m, Eterm f, unsigned int a, int is_remote) +static void *hipe_get_na_try_locked(Eterm m, Eterm f, unsigned int a, int is_remote, struct hipe_mfa_info **pp) { struct hipe_mfa_info *p; void *address; @@ -1523,22 +1325,53 @@ static void *hipe_get_na_nofail_locked(Eterm m, Eterm f, unsigned int a, int is_ address = p->remote_address; if (address) return address; - } else - p = hipe_mfa_info_table_put_locked(m, f, a); + } + /* Caller must take the slow path with the write lock held, but allow + it to avoid some work if it already holds the write lock. */ + if (pp) + *pp = p; + return NULL; +} + +static void *hipe_get_na_slow_rwlocked(Eterm m, Eterm f, unsigned int a, int is_remote, struct hipe_mfa_info *p) +{ + void *address; + + if (!p) + p = hipe_mfa_info_table_put_rwlocked(m, f, a); address = hipe_make_stub(m, f, a, is_remote); /* XXX: how to tell if a BEAM MFA is exported or not? */ p->remote_address = address; return address; } +static void *hipe_get_na_nofail_rwlocked(Eterm m, Eterm f, unsigned int a, int is_remote) +{ + struct hipe_mfa_info *p; + void *address; + + address = hipe_get_na_try_locked(m, f, a, is_remote, &p); + if (address) + return address; + + address = hipe_get_na_slow_rwlocked(m, f, a, is_remote, p); + return address; +} + static void *hipe_get_na_nofail(Eterm m, Eterm f, unsigned int a, int is_remote) { - void *p; + void *address; - hipe_mfa_info_table_lock(); - p = hipe_get_na_nofail_locked(m, f, a, is_remote); - hipe_mfa_info_table_unlock(); - return p; + hipe_mfa_info_table_rlock(); + address = hipe_get_na_try_locked(m, f, a, is_remote, NULL); + hipe_mfa_info_table_runlock(); + if (address) + return address; + + hipe_mfa_info_table_rwlock(); + address = hipe_get_na_slow_rwlocked(m, f, a, is_remote, NULL); + hipe_mfa_info_table_rwunlock(); + return address; } /* used for apply/3 in hipe_mode_switch */ @@ -1617,7 +1450,7 @@ int hipe_find_mfa_from_ra(const void *ra, Eterm *m, Eterm *f, unsigned int *a) /* Note about locking: the table is only updated from the loader, which runs with the rest of the system suspended. */ /* XXX: alas not true; see comment at hipe_mfa_info_table.lock */ - hipe_mfa_info_table_lock(); + hipe_mfa_info_table_rlock(); bucket = hipe_mfa_info_table.bucket; nrbuckets = 1 << hipe_mfa_info_table.log2size; mfa = NULL; @@ -1638,7 +1471,7 @@ int hipe_find_mfa_from_ra(const void *ra, Eterm *m, Eterm *f, unsigned int *a) *f = mfa->f; *a = mfa->a; } - hipe_mfa_info_table_unlock(); + hipe_mfa_info_table_runlock(); return mfa ? 1 : 0; } @@ -1715,9 +1548,9 @@ BIF_RETTYPE hipe_bifs_add_ref_2(BIF_ALIST_2) default: goto badarg; } - hipe_mfa_info_table_lock(); - callee_mfa = hipe_mfa_info_table_put_locked(callee.mod, callee.fun, callee.ari); - caller_mfa = hipe_mfa_info_table_put_locked(caller.mod, caller.fun, caller.ari); + hipe_mfa_info_table_rwlock(); + callee_mfa = hipe_mfa_info_table_put_rwlocked(callee.mod, callee.fun, callee.ari); + caller_mfa = hipe_mfa_info_table_put_rwlocked(caller.mod, caller.fun, caller.ari); refers_to = erts_alloc(ERTS_ALC_T_HIPE, sizeof(*refers_to)); refers_to->mfa = callee_mfa; @@ -1731,7 +1564,7 @@ BIF_RETTYPE hipe_bifs_add_ref_2(BIF_ALIST_2) ref->flags = flags; ref->next = callee_mfa->referred_from; callee_mfa->referred_from = ref; - hipe_mfa_info_table_unlock(); + hipe_mfa_info_table_rwunlock(); BIF_RET(NIL); @@ -1751,12 +1584,12 @@ BIF_RETTYPE hipe_bifs_mark_referred_from_1(BIF_ALIST_1) /* get_refs_from */ if (!term_to_mfa(BIF_ARG_1, &mfa)) BIF_ERROR(BIF_P, BADARG); - hipe_mfa_info_table_lock(); + hipe_mfa_info_table_rwlock(); p = hipe_mfa_info_table_get_locked(mfa.mod, mfa.fun, mfa.ari); if (p) for (ref = p->referred_from; ref != NULL; ref = ref->next) ref->flags |= REF_FLAG_PENDING_REDIRECT; - hipe_mfa_info_table_unlock(); + hipe_mfa_info_table_rwunlock(); BIF_RET(NIL); } @@ -1770,7 +1603,7 @@ static void hipe_purge_all_refs(void) struct hipe_mfa_info **bucket; unsigned int i, nrbuckets; - hipe_mfa_info_table_lock(); + hipe_mfa_info_table_rwlock(); bucket = hipe_mfa_info_table.bucket; nrbuckets = 1 << hipe_mfa_info_table.log2size; @@ -1792,7 +1625,7 @@ static void hipe_purge_all_refs(void) erts_free(ERTS_ALC_T_HIPE, mfa); } } - hipe_mfa_info_table_unlock(); + hipe_mfa_info_table_rwunlock(); } BIF_RETTYPE hipe_bifs_remove_refs_from_1(BIF_ALIST_1) @@ -1809,7 +1642,7 @@ BIF_RETTYPE hipe_bifs_remove_refs_from_1(BIF_ALIST_1) if (!term_to_mfa(BIF_ARG_1, &mfa)) BIF_ERROR(BIF_P, BADARG); - hipe_mfa_info_table_lock(); + hipe_mfa_info_table_rwlock(); caller_mfa = hipe_mfa_info_table_get_locked(mfa.mod, mfa.fun, mfa.ari); if (caller_mfa) { refers_to = caller_mfa->refers_to; @@ -1840,7 +1673,7 @@ BIF_RETTYPE hipe_bifs_remove_refs_from_1(BIF_ALIST_1) } caller_mfa->refers_to = NULL; } - hipe_mfa_info_table_unlock(); + hipe_mfa_info_table_rwunlock(); BIF_RET(am_ok); } @@ -1859,7 +1692,7 @@ BIF_RETTYPE hipe_bifs_redirect_referred_from_1(BIF_ALIST_1) if (!term_to_mfa(BIF_ARG_1, &mfa)) BIF_ERROR(BIF_P, BADARG); - hipe_mfa_info_table_lock(); + hipe_mfa_info_table_rwlock(); p = hipe_mfa_info_table_get_locked(mfa.mod, mfa.fun, mfa.ari); if (p) { prev = &p->referred_from; @@ -1867,7 +1700,7 @@ BIF_RETTYPE hipe_bifs_redirect_referred_from_1(BIF_ALIST_1) while (ref) { if (ref->flags & REF_FLAG_PENDING_REDIRECT) { is_remote = ref->flags & REF_FLAG_IS_REMOTE; - new_address = hipe_get_na_nofail_locked(p->m, p->f, p->a, is_remote); + new_address = hipe_get_na_nofail_rwlocked(p->m, p->f, p->a, is_remote); if (ref->flags & REF_FLAG_IS_LOAD_MFA) res = hipe_patch_insn(ref->address, (Uint)new_address, am_load_mfa); else @@ -1890,7 +1723,7 @@ BIF_RETTYPE hipe_bifs_redirect_referred_from_1(BIF_ALIST_1) } } } - hipe_mfa_info_table_unlock(); + hipe_mfa_info_table_rwunlock(); BIF_RET(NIL); } diff --git a/erts/emulator/hipe/hipe_bif0.tab b/erts/emulator/hipe/hipe_bif0.tab index 2514b1c3a5..d715a0914b 100644 --- a/erts/emulator/hipe/hipe_bif0.tab +++ b/erts/emulator/hipe/hipe_bif0.tab @@ -49,7 +49,6 @@ bif hipe_bifs:constants_size/0 bif hipe_bifs:merge_term/1 bif hipe_bifs:fun_to_address/1 -#bif hipe_bifs:get_emu_address/1 bif hipe_bifs:set_native_address/3 #bif hipe_bifs:address_to_fun/1 @@ -72,7 +71,6 @@ bif hipe_bifs:term_to_word/1 bif hipe_bifs:get_fe/2 bif hipe_bifs:set_native_address_in_fe/2 -#bif hipe_bifs:make_native_stub/2 bif hipe_bifs:find_na_or_make_stub/2 bif hipe_bifs:check_crc/1 diff --git a/erts/emulator/hipe/hipe_bif1.c b/erts/emulator/hipe/hipe_bif1.c index 56767ef04b..ecb34df412 100644 --- a/erts/emulator/hipe/hipe_bif1.c +++ b/erts/emulator/hipe/hipe_bif1.c @@ -574,22 +574,6 @@ BIF_RETTYPE hipe_bifs_pause_times_0(BIF_ALIST_0) /* XXX: these macros have free variables */ #ifdef BM_TIMERS -#if USE_PERFCTR -#define MAKE_TIME(_timer_) { \ - BM_TIMER_T tmp = _timer_##_time; \ - milli = (uint)(tmp - ((int)(tmp / 1000)) * 1000); \ - tmp /= 1000; \ - sec = (uint)(tmp - ((int)(tmp / 60)) * 60); \ - min = (uint)tmp / 60; } - -#define MAKE_MICRO_TIME(_timer_) { \ - BM_TIMER_T tmp = _timer_##_time * 1000; \ - micro = (uint)(tmp - ((int)(tmp / 1000)) * 1000); \ - tmp /= 1000; \ - milli = (uint)(tmp - ((int)(tmp / 1000)) * 1000); \ - sec = (uint)tmp / 1000; } - -#else #define MAKE_TIME(_timer_) { \ BM_TIMER_T tmp = _timer_##_time / 1000000; \ milli = tmp % 1000; \ @@ -604,7 +588,6 @@ BIF_RETTYPE hipe_bifs_pause_times_0(BIF_ALIST_0) milli = tmp % 1000; \ sec = tmp / 1000; } -#endif #else #define MAKE_TIME(_timer_) #define MAKE_MICRO_TIME(_timer_) @@ -852,9 +835,6 @@ BIF_RETTYPE hipe_bifs_misc_timer_clear_0(BIF_ALIST_0) /* * HiPE hrvtime(). * These implementations are currently available: - * + On Linux with the perfctr extension we can use the process' - * virtualised time-stamp counter. To enable this mode you must - * pass `--with-perfctr=/path/to/perfctr' when configuring. * + The fallback, which is the same as {X,_} = runtime(statistics). */ @@ -866,37 +846,6 @@ static double fallback_get_hrvtime(void) return (double)ms_user; } -#if USE_PERFCTR - -#include "hipe_perfctr.h" -static int hrvtime_started; /* 0: closed, +1: perfctr, -1: fallback */ -#define hrvtime_is_started() (hrvtime_started != 0) - -static void start_hrvtime(void) -{ - if (hipe_perfctr_hrvtime_open() >= 0) - hrvtime_started = 1; - else - hrvtime_started = -1; -} - -static void stop_hrvtime(void) -{ - if (hrvtime_started > 0) - hipe_perfctr_hrvtime_close(); - hrvtime_started = 0; -} - -static double get_hrvtime(void) -{ - if (hrvtime_started > 0) - return hipe_perfctr_hrvtime_get(); - else - return fallback_get_hrvtime(); -} - -#else /* !USE_PERFCTR */ - /* * Fallback, if nothing better exists. * This is the same as {X,_} = statistics(runtime), which uses @@ -908,8 +857,6 @@ static double get_hrvtime(void) #define stop_hrvtime() do{}while(0) #define get_hrvtime() fallback_get_hrvtime() -#endif /* !USE_PERFCTR */ - BIF_RETTYPE hipe_bifs_get_hrvtime_0(BIF_ALIST_0) { Eterm *hp; diff --git a/erts/emulator/hipe/hipe_bif_list.m4 b/erts/emulator/hipe/hipe_bif_list.m4 index 5f92b6bac4..96a849621f 100644 --- a/erts/emulator/hipe/hipe_bif_list.m4 +++ b/erts/emulator/hipe/hipe_bif_list.m4 @@ -277,7 +277,10 @@ ifelse($1,list_to_binary_1,hipe_wrapper_list_to_binary_1, ifelse($1,iolist_to_binary_1,hipe_wrapper_iolist_to_binary_1, ifelse($1,binary_list_to_bin_1,hipe_wrapper_binary_list_to_bin_1, ifelse($1,list_to_bitstring_1,hipe_wrapper_list_to_bitstring_1, -$1)))))))))))') +ifelse($1,send_2,hipe_wrapper_send_2, +ifelse($1,send_3,hipe_wrapper_send_3, +ifelse($1,ebif_bang_2,hipe_wrapper_ebif_bang_2, +$1))))))))))))))') define(BIF_LIST,`standard_bif_interface_$3(nbif_$4, CFUN($4))') include(TARGET/`erl_bif_list.h') diff --git a/erts/emulator/hipe/hipe_debug.c b/erts/emulator/hipe/hipe_debug.c index 32694a8f97..61406b92af 100644 --- a/erts/emulator/hipe/hipe_debug.c +++ b/erts/emulator/hipe/hipe_debug.c @@ -172,8 +172,10 @@ void hipe_print_pcb(Process *p) printf("P: 0x%0*lx\r\n", 2*(int)sizeof(long), (unsigned long)p); printf("-----------------------------------------------\r\n"); printf("Offset| Name | Value | *Value |\r\n"); +#undef U #define U(n,x) \ printf(" % 4d | %s | 0x%0*lx | |\r\n", (int)offsetof(Process,x), n, 2*(int)sizeof(long), (unsigned long)p->x) +#undef P #define P(n,x) \ printf(" % 4d | %s | 0x%0*lx | 0x%0*lx |\r\n", (int)offsetof(Process,x), n, 2*(int)sizeof(long), (unsigned long)p->x, 2*(int)sizeof(long), p->x ? (unsigned long)*(p->x) : -1UL) @@ -231,7 +233,7 @@ void hipe_print_pcb(Process *p) U("nsp ", hipe.nsp); U("nstack ", hipe.nstack); U("nstend ", hipe.nstend); - U("ncallee ", hipe.ncallee); + U("ncallee ", hipe.u.ncallee); hipe_arch_print_pcb(&p->hipe); #endif /* HIPE */ #undef U diff --git a/erts/emulator/hipe/hipe_gc.c b/erts/emulator/hipe/hipe_gc.c index 86c4068072..b10263f6e2 100644 --- a/erts/emulator/hipe/hipe_gc.c +++ b/erts/emulator/hipe/hipe_gc.c @@ -22,6 +22,9 @@ #ifdef HAVE_CONFIG_H #include "config.h" #endif + +#define ERL_WANT_GC_INTERNALS__ + #include "global.h" #include "erl_gc.h" diff --git a/erts/emulator/hipe/hipe_mkliterals.c b/erts/emulator/hipe/hipe_mkliterals.c index 0e287908b1..ed355ce264 100644 --- a/erts/emulator/hipe/hipe_mkliterals.c +++ b/erts/emulator/hipe/hipe_mkliterals.c @@ -498,8 +498,8 @@ static const struct rts_param rts_params[] = { { 38, "P_ARG4", 1, offsetof(struct process, def_arg_reg[4]) }, { 39, "P_ARG5", 1, offsetof(struct process, def_arg_reg[5]) }, { 40, "P_NSP", 1, offsetof(struct process, hipe.nsp) }, - { 41, "P_NCALLEE", 1, offsetof(struct process, hipe.ncallee) }, - { 42, "P_CLOSURE", 1, offsetof(struct process, hipe.closure) }, + { 41, "P_NCALLEE", 1, offsetof(struct process, hipe.u.ncallee) }, + { 42, "P_CLOSURE", 1, offsetof(struct process, hipe.u.closure) }, { 43, "P_NSP_LIMIT", 1, offsetof(struct process, hipe.nstack) }, { 44, "P_CSP", #if defined(__i386__) || defined(__x86_64__) @@ -524,6 +524,7 @@ static const struct rts_param rts_params[] = { }, { 49, "P_MSG_FIRST", 1, offsetof(struct process, msg.first) }, { 50, "P_MSG_SAVE", 1, offsetof(struct process, msg.save) }, + { 51, "P_CALLEE_EXP", 1, offsetof(struct process, hipe.u.callee_exp) }, }; #define NR_PARAMS ARRAY_SIZE(rts_params) diff --git a/erts/emulator/hipe/hipe_mode_switch.c b/erts/emulator/hipe/hipe_mode_switch.c index 1ae1d17b7f..8c73312d45 100644 --- a/erts/emulator/hipe/hipe_mode_switch.c +++ b/erts/emulator/hipe/hipe_mode_switch.c @@ -140,7 +140,6 @@ void hipe_check_pcb(Process *p, const char *file, unsigned line) #endif /* HIPE_DEBUG > 0 */ /* ensure that at least nwords words are available on the native stack */ -static void hipe_check_nstack(Process *p, unsigned nwords); #if defined(__sparc__) #include "hipe_sparc_glue.h" @@ -159,7 +158,7 @@ static void hipe_check_nstack(Process *p, unsigned nwords); Uint hipe_beam_pc_return[1]; /* needed in hipe_debug.c */ Uint hipe_beam_pc_throw[1]; /* needed in hipe_debug.c */ Uint hipe_beam_pc_resume[1]; /* needed by hipe_set_timeout() */ -static Eterm hipe_beam_catch_throw; +Eterm hipe_beam_catch_throw; void hipe_mode_switch_init(void) { @@ -185,22 +184,6 @@ void hipe_set_call_trap(Uint *bfun, void *nfun, int is_closure) bfun[-4] = (Uint)nfun; } -void hipe_reserve_beam_trap_frame(Process *p, Eterm reg[], unsigned arity) -{ - if (!hipe_bifcall_from_native_is_recursive(p)) - return; - - /* ensure that at least 2 words are available on the BEAM stack */ - if ((p->stop - 2) < p->htop) { - DPRINTF("calling gc to reserve BEAM stack size"); - p->fcalls -= erts_garbage_collect(p, 2, reg, arity); - ASSERT(!((p->stop - 2) < p->htop)); - } - p->stop -= 2; - p->stop[0] = NIL; - p->stop[1] = hipe_beam_catch_throw; -} - static __inline__ void hipe_push_beam_trap_frame(Process *p, Eterm reg[], unsigned arity) { @@ -223,15 +206,6 @@ hipe_push_beam_trap_frame(Process *p, Eterm reg[], unsigned arity) p->cp = hipe_beam_pc_return; } -void hipe_unreserve_beam_trap_frame(Process *p) -{ - if (!hipe_bifcall_from_native_is_recursive(p)) - return; - - ASSERT(p->stop[0] == NIL && p->stop[1] == hipe_beam_catch_throw); - p->stop += 2; -} - static __inline__ void hipe_pop_beam_trap_frame(Process *p) { ASSERT(p->stop[1] == hipe_beam_catch_throw); @@ -265,14 +239,14 @@ Process *hipe_mode_switch(Process *p, unsigned cmd, Eterm reg[]) /* BEAM calls a native code function */ unsigned arity = cmd >> 8; - /* p->hipe.ncallee set in beam_emu */ + /* p->hipe.u.ncallee set in beam_emu */ if (p->cp == hipe_beam_pc_return) { /* Native called BEAM, which now tailcalls native. */ hipe_pop_beam_trap_frame(p); result = hipe_tailcall_to_native(p, arity, reg); break; } - DPRINTF("calling %#lx/%u", (long)p->hipe.ncallee, arity); + DPRINTF("calling %#lx/%u", (long)p->hipe.u.ncallee, arity); result = hipe_call_to_native(p, arity, reg); break; } @@ -290,18 +264,18 @@ Process *hipe_mode_switch(Process *p, unsigned cmd, Eterm reg[]) arity -= funp->num_free; /* arity == #formals */ reg[arity] = fun; ++arity; /* correct for having added the closure */ - /* HIPE_ASSERT(p->hipe.ncallee == (void(*)(void))funp->native_address); */ + /* HIPE_ASSERT(p->hipe.u.ncallee == (void(*)(void))funp->native_address); */ /* just like a normal call from now on */ - /* p->hipe.ncallee set in beam_emu */ + /* p->hipe.u.ncallee set in beam_emu */ if (p->cp == hipe_beam_pc_return) { /* Native called BEAM, which now tailcalls native. */ hipe_pop_beam_trap_frame(p); result = hipe_tailcall_to_native(p, arity, reg); break; } - DPRINTF("calling %#lx/%u", (long)p->hipe.ncallee, arity); + DPRINTF("calling %#lx/%u", (long)p->hipe.u.ncallee, arity); result = hipe_call_to_native(p, arity, reg); break; } @@ -404,13 +378,13 @@ Process *hipe_mode_switch(Process *p, unsigned cmd, Eterm reg[]) if (is_recursive) hipe_push_beam_trap_frame(p, reg, p->arity); - result = HIPE_MODE_SWITCH_RES_CALL; + result = HIPE_MODE_SWITCH_RES_CALL_BEAM; break; } - case HIPE_MODE_SWITCH_RES_CALL: { + case HIPE_MODE_SWITCH_RES_CALL_EXPORTED: { /* Native code calls or tailcalls BEAM. * - * p->i is the callee's BEAM code + * p->hipe.u.callee_exp is the callee's export entry * p->arity is the callee's arity * p->def_arg_reg[] contains the register parameters * p->hipe.nsp[] contains the stacked parameters @@ -430,15 +404,15 @@ Process *hipe_mode_switch(Process *p, unsigned cmd, Eterm reg[]) * F(A1, ..., AN, FV1, ..., FVM, Closure) * (Where Ai is argument i and FVj is free variable j) * - * p->hipe.closure contains the closure + * p->hipe.u.closure contains the closure * p->def_arg_reg[] contains the register parameters * p->hipe.nsp[] contains the stacked parameters */ ErlFunThing *closure; unsigned num_free, arity, i, is_recursive; - HIPE_ASSERT(is_fun(p->hipe.closure)); - closure = (ErlFunThing*)fun_val(p->hipe.closure); + HIPE_ASSERT(is_fun(p->hipe.u.closure)); + closure = (ErlFunThing*)fun_val(p->hipe.u.closure); num_free = closure->num_free; arity = closure->fe->arity; @@ -468,10 +442,10 @@ Process *hipe_mode_switch(Process *p, unsigned cmd, Eterm reg[]) p->i = closure->fe->address; /* Change result code to the faster plain CALL type. */ - result = HIPE_MODE_SWITCH_RES_CALL; + result = HIPE_MODE_SWITCH_RES_CALL_BEAM; } /* Append the closure as the last parameter. Don't increment arity. */ - reg[arity] = p->hipe.closure; + reg[arity] = p->hipe.u.closure; if (is_recursive) { /* BEAM called native, which now calls BEAM. @@ -549,7 +523,7 @@ Process *hipe_mode_switch(Process *p, unsigned cmd, Eterm reg[]) } } HIPE_CHECK_PCB(p); - result = HIPE_MODE_SWITCH_RES_CALL; + result = HIPE_MODE_SWITCH_RES_CALL_BEAM; p->def_arg_reg[3] = result; return p; } @@ -577,7 +551,7 @@ Process *hipe_mode_switch(Process *p, unsigned cmd, Eterm reg[]) address = hipe_get_remote_na(mfa[0], mfa[1], arity); if (!address) goto do_apply_fail; - p->hipe.ncallee = (void(*)(void)) address; + p->hipe.u.ncallee = (void(*)(void)) address; result = hipe_tailcall_to_native(p, arity, reg); goto do_return_from_native; do_apply_fail: @@ -607,7 +581,6 @@ static unsigned hipe_next_nstack_size(unsigned size) } #if 0 && defined(HIPE_NSTACK_GROWS_UP) -#define hipe_nstack_avail(p) ((p)->hipe.nstend - (p)->hipe.nsp) void hipe_inc_nstack(Process *p) { Eterm *old_nstack = p->hipe.nstack; @@ -631,7 +604,6 @@ void hipe_inc_nstack(Process *p) #endif #if defined(HIPE_NSTACK_GROWS_DOWN) -#define hipe_nstack_avail(p) ((unsigned)((p)->hipe.nsp - (p)->hipe.nstack)) void hipe_inc_nstack(Process *p) { unsigned old_size = p->hipe.nstend - p->hipe.nstack; @@ -663,12 +635,6 @@ void hipe_empty_nstack(Process *p) p->hipe.nstend = NULL; } -static void hipe_check_nstack(Process *p, unsigned nwords) -{ - while (hipe_nstack_avail(p) < nwords) - hipe_inc_nstack(p); -} - void hipe_set_closure_stub(ErlFunEntry *fe, unsigned num_free) { unsigned arity; diff --git a/erts/emulator/hipe/hipe_mode_switch.h b/erts/emulator/hipe/hipe_mode_switch.h index 06721e3c04..b8de12fcbb 100644 --- a/erts/emulator/hipe/hipe_mode_switch.h +++ b/erts/emulator/hipe/hipe_mode_switch.h @@ -31,7 +31,7 @@ /* result codes for beam_emu <- hipe_mode_switch() return */ #define HIPE_MODE_SWITCH_RES_RETURN 4 -#define HIPE_MODE_SWITCH_RES_CALL 5 +#define HIPE_MODE_SWITCH_RES_CALL_EXPORTED 5 #define HIPE_MODE_SWITCH_RES_THROW 6 /* additional result codes for hipe_mode_switch() <- native return */ @@ -45,6 +45,8 @@ #define HIPE_MODE_SWITCH_RES_APPLY 13 /* mode_switch <- native */ +#define HIPE_MODE_SWITCH_RES_CALL_BEAM 14 + #ifndef ASM #include "error.h" @@ -59,13 +61,58 @@ void hipe_empty_nstack(Process *p); void hipe_set_closure_stub(ErlFunEntry *fe, unsigned num_free); Eterm hipe_build_stacktrace(Process *p, struct StackTrace *s); -void hipe_reserve_beam_trap_frame(Process*, Eterm reg[], unsigned arity); -void hipe_unreserve_beam_trap_frame(Process*); +ERTS_GLB_INLINE void hipe_reserve_beam_trap_frame(Process*, Eterm reg[], unsigned arity); +ERTS_GLB_INLINE void hipe_unreserve_beam_trap_frame(Process*); extern Uint hipe_beam_pc_return[]; extern Uint hipe_beam_pc_throw[]; extern Uint hipe_beam_pc_resume[]; +#if ERTS_GLB_INLINE_INCL_FUNC_DEF + +#include "erl_gc.h" +#include "hipe_stack.h" + +#if defined(__sparc__) +#include "hipe_sparc_glue.h" +#elif defined(__i386__) +#include "hipe_x86_glue.h" +#elif defined(__x86_64__) +#include "hipe_amd64_glue.h" +#elif defined(__powerpc__) || defined(__ppc__) || defined(__powerpc64__) +#include "hipe_ppc_glue.h" +#elif defined(__arm__) +#include "hipe_arm_glue.h" +#endif + +extern Eterm hipe_beam_catch_throw; + +ERTS_GLB_INLINE void hipe_reserve_beam_trap_frame(Process *p, Eterm reg[], unsigned arity) +{ + if (!hipe_bifcall_from_native_is_recursive(p)) + return; + + /* ensure that at least 2 words are available on the BEAM stack */ + if ((p->stop - 2) < p->htop) { + p->fcalls -= erts_garbage_collect(p, 2, reg, arity); + ASSERT(!((p->stop - 2) < p->htop)); + } + p->stop -= 2; + p->stop[0] = NIL; + p->stop[1] = hipe_beam_catch_throw; +} + +ERTS_GLB_INLINE void hipe_unreserve_beam_trap_frame(Process *p) +{ + if (!hipe_bifcall_from_native_is_recursive(p)) + return; + + ASSERT(p->stop[0] == NIL && p->stop[1] == hipe_beam_catch_throw); + p->stop += 2; +} + +#endif /* ERTS_GLB_INLINE_INCL_FUNC_DEF */ + #endif /* ASM */ #endif /* HIPE_MODE_SWITCH_H */ diff --git a/erts/emulator/hipe/hipe_native_bif.c b/erts/emulator/hipe/hipe_native_bif.c index 7d343dd91e..7e8632b50d 100644 --- a/erts/emulator/hipe/hipe_native_bif.c +++ b/erts/emulator/hipe/hipe_native_bif.c @@ -330,8 +330,6 @@ char *hipe_bs_allocate(int len) Binary *bptr; bptr = erts_bin_nrml_alloc(len); - bptr->flags = 0; - bptr->orig_size = len; erts_smp_atomic_init_nob(&bptr->refc, 1); return bptr->orig_bytes; } @@ -341,7 +339,6 @@ Binary *hipe_bs_reallocate(Binary* oldbptr, int newsize) Binary *bptr; bptr = erts_bin_realloc(oldbptr, newsize); - bptr->orig_size = newsize; return bptr; } diff --git a/erts/emulator/hipe/hipe_perfctr.c b/erts/emulator/hipe/hipe_perfctr.c deleted file mode 100644 index 371b3fb097..0000000000 --- a/erts/emulator/hipe/hipe_perfctr.c +++ /dev/null @@ -1,229 +0,0 @@ -/* - * %CopyrightBegin% - * - * Copyright Ericsson AB 2004-2011. All Rights Reserved. - * - * The contents of this file are subject to the Erlang Public License, - * Version 1.1, (the "License"); you may not use this file except in - * compliance with the License. You should have received a copy of the - * Erlang Public License along with this software. If not, it can be - * retrieved online at http://www.erlang.org/. - * - * Software distributed under the License is distributed on an "AS IS" - * basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See - * the License for the specific language governing rights and limitations - * under the License. - * - * %CopyrightEnd% - */ - - -#ifdef HAVE_CONFIG_H -#include "config.h" -#endif -#include "sys.h" -#include "error.h" -#include "global.h" -#include "bif.h" -#include "big.h" -#include "erl_binary.h" -#include "hipe_perfctr.h" -#include "libperfctr.h" - -static struct vperfctr *vperfctr; -static unsigned int have_rdtsc; -static double tsc_to_ms; -static unsigned int tsc_on; /* control calls must set tsc_on if have_rdtsc is true */ -static unsigned int nractrs; -static unsigned int users; -#define USER_BIFS (1<<0) -#define USER_HRVTIME (1<<1) - -static int hipe_perfctr_open(unsigned int user) -{ - struct perfctr_info info; - - if (!vperfctr) { - vperfctr = vperfctr_open(); - if (!vperfctr) - return -1; - if (vperfctr_info(vperfctr, &info) >= 0) { - tsc_to_ms = (double)(info.tsc_to_cpu_mult ? : 1) / (double)info.cpu_khz; - have_rdtsc = (info.cpu_features & PERFCTR_FEATURE_RDTSC) ? 1 : 0; - } - tsc_on = 0; - nractrs = 0; - } - users |= user; - return 0; -} - -static void hipe_perfctr_reset(void) -{ - struct vperfctr_control control; - - memset(&control, 0, sizeof control); - if (have_rdtsc) - control.cpu_control.tsc_on = 1; - nractrs = 0; - if (vperfctr_control(vperfctr, &control) >= 0) - tsc_on = 1; -} - -static void hipe_perfctr_close(unsigned int user) -{ - if (!vperfctr) - return; - users &= ~user; - switch (users) { - case 0: - vperfctr_unlink(vperfctr); - vperfctr_close(vperfctr); - vperfctr = NULL; - tsc_on = 0; - nractrs = 0; - break; - case USER_HRVTIME: - hipe_perfctr_reset(); - } -} - -/* - * Interface for HiPE's hrvtime code. - */ - -int hipe_perfctr_hrvtime_open(void) -{ - if (hipe_perfctr_open(USER_HRVTIME) < 0) - return -1; - if (have_rdtsc) { - if (!tsc_on) - hipe_perfctr_reset(); /* note: updates tsc_on */ - if (tsc_on) - return 0; - } - hipe_perfctr_hrvtime_close(); - return -1; -} - -void hipe_perfctr_hrvtime_close(void) -{ - hipe_perfctr_close(USER_HRVTIME); -} - -double hipe_perfctr_hrvtime_get(void) -{ - return (double)vperfctr_read_tsc(vperfctr) * tsc_to_ms; -} - -/* - * BIF interface for user-programmable performance counters. - */ - -BIF_RETTYPE hipe_bifs_vperfctr_open_0(BIF_ALIST_0) -{ - if (hipe_perfctr_open(USER_BIFS) < 0) - BIF_RET(am_false); /* arity 0 BIFs can't fail :-( */ - BIF_RET(am_true); -} - -BIF_RETTYPE hipe_bifs_vperfctr_close_0(BIF_ALIST_0) -{ - hipe_perfctr_close(USER_BIFS); - BIF_RET(NIL); -} - -static Eterm ull_to_integer(unsigned long long x, Process *p) -{ - unsigned long long tmpx; - unsigned int ds, i; - size_t sz; - Eterm *hp; - ErtsDigit *xp; - - if (x <= (unsigned long long)MAX_SMALL) - return make_small(x); - - /* Calculate number of digits. */ - ds = 0; - tmpx = x; - do { - ++ds; - tmpx = (tmpx >> (D_EXP / 2)) >> (D_EXP / 2); - } while (tmpx != 0); - - sz = BIG_NEED_SIZE(ds); /* number of words including arity */ - hp = HAlloc(p, sz); - *hp = make_pos_bignum_header(sz-1); - - xp = (ErtsDigit*)(hp+1); - i = 0; - do { - xp[i++] = (ErtsDigit)x; - x = (x >> (D_EXP / 2)) >> (D_EXP / 2); - } while (i < ds); - while (i & (BIG_DIGITS_PER_WORD-1)) - xp[i++] = 0; - - return make_big(hp); -} - -BIF_RETTYPE hipe_bifs_vperfctr_info_0(BIF_ALIST_0) -{ - struct perfctr_info info; - - if (!vperfctr || vperfctr_info(vperfctr, &info) < 0) - BIF_RET(am_false); /* arity 0 BIFs can't fail :-( */ - BIF_RET(new_binary(BIF_P, (void*)&info, sizeof info)); -} - -BIF_RETTYPE hipe_bifs_vperfctr_read_tsc_0(BIF_ALIST_0) -{ - unsigned long long val; - - if (!vperfctr || !tsc_on) - BIF_RET(am_false); /* arity 0 BIFs can't fail :-( */ - val = vperfctr_read_tsc(vperfctr); - BIF_RET(ull_to_integer(val, BIF_P)); -} - -BIF_RETTYPE hipe_bifs_vperfctr_read_pmc_1(BIF_ALIST_1) -{ - Uint pmc; - unsigned long long val; - - if (!vperfctr || - is_not_small(BIF_ARG_1) || - (pmc = unsigned_val(BIF_ARG_1), pmc >= nractrs)) - BIF_RET(am_false); /* for consistency with the arity 0 BIFs */ - val = vperfctr_read_pmc(vperfctr, pmc); - BIF_RET(ull_to_integer(val, BIF_P)); -} - -BIF_RETTYPE hipe_bifs_vperfctr_control_1(BIF_ALIST_1) -{ - void *bytes; - struct vperfctr_control control; - Uint bitoffs; - Uint bitsize; - - if (!vperfctr) - BIF_ERROR(BIF_P, BADARG); - if (is_not_binary(BIF_ARG_1)) - BIF_ERROR(BIF_P, BADARG); - if (binary_size(BIF_ARG_1) != sizeof control) - BIF_ERROR(BIF_P, BADARG); - ERTS_GET_BINARY_BYTES(BIF_ARG_1, bytes, bitoffs, bitsize); - ASSERT(bitoffs == 0); - ASSERT(bitsize == 0); - memcpy(&control, bytes, sizeof control); - if (have_rdtsc) - control.cpu_control.tsc_on = 1; - if (vperfctr_control(vperfctr, &control) < 0) { - hipe_perfctr_reset(); - BIF_ERROR(BIF_P, BADARG); - } - tsc_on = control.cpu_control.tsc_on; - nractrs = control.cpu_control.nractrs; - BIF_RET(NIL); -} diff --git a/erts/emulator/hipe/hipe_perfctr.h b/erts/emulator/hipe/hipe_perfctr.h deleted file mode 100644 index 8fbf9ecf35..0000000000 --- a/erts/emulator/hipe/hipe_perfctr.h +++ /dev/null @@ -1,23 +0,0 @@ -/* - * %CopyrightBegin% - * - * Copyright Ericsson AB 2004-2011. All Rights Reserved. - * - * The contents of this file are subject to the Erlang Public License, - * Version 1.1, (the "License"); you may not use this file except in - * compliance with the License. You should have received a copy of the - * Erlang Public License along with this software. If not, it can be - * retrieved online at http://www.erlang.org/. - * - * Software distributed under the License is distributed on an "AS IS" - * basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See - * the License for the specific language governing rights and limitations - * under the License. - * - * %CopyrightEnd% - */ - - -extern int hipe_perfctr_hrvtime_open(void); -extern void hipe_perfctr_hrvtime_close(void); -extern double hipe_perfctr_hrvtime_get(void); diff --git a/erts/emulator/hipe/hipe_perfctr.tab b/erts/emulator/hipe/hipe_perfctr.tab deleted file mode 100644 index eaecea4651..0000000000 --- a/erts/emulator/hipe/hipe_perfctr.tab +++ /dev/null @@ -1,25 +0,0 @@ -# -# %CopyrightBegin% -# -# Copyright Ericsson AB 2004-2011. All Rights Reserved. -# -# The contents of this file are subject to the Erlang Public License, -# Version 1.1, (the "License"); you may not use this file except in -# compliance with the License. You should have received a copy of the -# Erlang Public License along with this software. If not, it can be -# retrieved online at http://www.erlang.org/. -# -# Software distributed under the License is distributed on an "AS IS" -# basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See -# the License for the specific language governing rights and limitations -# under the License. -# -# %CopyrightEnd% -# - -bif hipe_bifs:vperfctr_open/0 -bif hipe_bifs:vperfctr_close/0 -bif hipe_bifs:vperfctr_info/0 -bif hipe_bifs:vperfctr_read_tsc/0 -bif hipe_bifs:vperfctr_read_pmc/1 -bif hipe_bifs:vperfctr_control/1 diff --git a/erts/emulator/hipe/hipe_ppc.c b/erts/emulator/hipe/hipe_ppc.c index 2d8fd61e1e..1eaa9f6855 100644 --- a/erts/emulator/hipe/hipe_ppc.c +++ b/erts/emulator/hipe/hipe_ppc.c @@ -285,7 +285,7 @@ int hipe_patch_insn(void *address, Uint64 value, Eterm type) } } -void *hipe_make_native_stub(void *beamAddress, unsigned int beamArity) +void *hipe_make_native_stub(void *callee_exp, unsigned int beamArity) { unsigned int *code; @@ -293,17 +293,19 @@ void *hipe_make_native_stub(void *beamAddress, unsigned int beamArity) abort(); code = alloc_stub(7); + if (!code) + return NULL; - /* addis r12,0,beamAddress@highest */ - code[0] = 0x3d800000 | (((unsigned long)beamAddress >> 48) & 0xffff); - /* ori r12,r12,beamAddress@higher */ - code[1] = 0x618c0000 | (((unsigned long)beamAddress >> 32) & 0xffff); + /* addis r12,0,callee_exp@highest */ + code[0] = 0x3d800000 | (((unsigned long)callee_exp >> 48) & 0xffff); + /* ori r12,r12,callee_exp@higher */ + code[1] = 0x618c0000 | (((unsigned long)callee_exp >> 32) & 0xffff); /* sldi r12,r12,32 (rldicr r12,r12,32,31) */ code[2] = 0x798c07c6; - /* oris r12,r12,beamAddress@h */ - code[3] = 0x658c0000 | (((unsigned long)beamAddress >> 16) & 0xffff); - /* ori r12,r12,beamAddress@l */ - code[4] = 0x618c0000 | ((unsigned long)beamAddress & 0xffff); + /* oris r12,r12,callee_exp@h */ + code[3] = 0x658c0000 | (((unsigned long)callee_exp >> 16) & 0xffff); + /* ori r12,r12,callee_exp@l */ + code[4] = 0x618c0000 | ((unsigned long)callee_exp & 0xffff); /* addi r0,0,beamArity */ code[5] = 0x38000000 | (beamArity & 0x7FFF); /* ba nbif_callemu */ @@ -355,18 +357,16 @@ int hipe_patch_insn(void *address, Uint32 value, Eterm type) return 0; } -/* called from hipe_bif0.c:hipe_bifs_make_native_stub_2() - and hipe_bif0.c:hipe_make_stub() */ -void *hipe_make_native_stub(void *beamAddress, unsigned int beamArity) +void *hipe_make_native_stub(void *callee_exp, unsigned int beamArity) { unsigned int *code; /* * Native code calls BEAM via a stub looking as follows: * - * addi r12,0,beamAddress@l + * addi r12,0,callee_exp@l * addi r0,0,beamArity - * addis r12,r12,beamAddress@ha + * addis r12,r12,callee_exp@ha * ba nbif_callemu * * I'm using r0 and r12 since the standard SVR4 ABI allows @@ -383,13 +383,15 @@ void *hipe_make_native_stub(void *beamAddress, unsigned int beamArity) abort(); code = alloc_stub(4); + if (!code) + return NULL; - /* addi r12,0,beamAddress@l */ - code[0] = 0x39800000 | ((unsigned long)beamAddress & 0xFFFF); + /* addi r12,0,callee_exp@l */ + code[0] = 0x39800000 | ((unsigned long)callee_exp & 0xFFFF); /* addi r0,0,beamArity */ code[1] = 0x38000000 | (beamArity & 0x7FFF); - /* addis r12,r12,beamAddress@ha */ - code[2] = 0x3D8C0000 | at_ha((unsigned long)beamAddress); + /* addis r12,r12,callee_exp@ha */ + code[2] = 0x3D8C0000 | at_ha((unsigned long)callee_exp); /* ba nbif_callemu */ code[3] = 0x48000002 | (unsigned long)&nbif_callemu; diff --git a/erts/emulator/hipe/hipe_ppc.h b/erts/emulator/hipe/hipe_ppc.h index 66000c1846..e9d3e6564b 100644 --- a/erts/emulator/hipe/hipe_ppc.h +++ b/erts/emulator/hipe/hipe_ppc.h @@ -64,10 +64,6 @@ AEXTERN(void,hipe_ppc_inc_stack,(void)); extern void hipe_ppc_inc_stack(void); /* we don't have the AEXTERN() fallback :-( */ #endif -/* for hipe_bifs_enter_code_2 */ -extern void *hipe_alloc_code(Uint nrbytes, Eterm callees, Eterm *trampolines, Process *p); -#define HIPE_ALLOC_CODE(n,c,t,p) hipe_alloc_code((n),(c),(t),(p)) - #if !defined(__powerpc64__) extern const unsigned int fconv_constant[]; #endif diff --git a/erts/emulator/hipe/hipe_ppc_asm.m4 b/erts/emulator/hipe/hipe_ppc_asm.m4 index 343402f9f0..e5a56de687 100644 --- a/erts/emulator/hipe/hipe_ppc_asm.m4 +++ b/erts/emulator/hipe/hipe_ppc_asm.m4 @@ -23,6 +23,22 @@ changecom(`/*', `*/')dnl #define HIPE_PPC_ASM_H' /* + * Tunables. + */ +define(LEAF_WORDS,16)dnl number of stack words for leaf functions +define(NR_ARG_REGS,4)dnl admissible values are 0 to 6, inclusive + +`#define PPC_LEAF_WORDS 'LEAF_WORDS +`#define PPC_NR_ARG_REGS 'NR_ARG_REGS +`#define NR_ARG_REGS 'NR_ARG_REGS + + +`#ifdef ASM' +/* + * Only assembler stuff from here on (when included from *.S) + */ + +/* * Handle 32 vs 64-bit. */ ifelse(ARCH,ppc64,` @@ -53,13 +69,6 @@ define(WSIZE,4)dnl `#define STORE 'STORE `#define CMPI 'CMPI -/* - * Tunables. - */ -define(LEAF_WORDS,16)dnl number of stack words for leaf functions -define(NR_ARG_REGS,4)dnl admissible values are 0 to 6, inclusive - -`#define PPC_LEAF_WORDS 'LEAF_WORDS /* * Workarounds for Darwin. @@ -193,8 +202,6 @@ NAME: \ /* * Argument (parameter) registers. */ -`#define PPC_NR_ARG_REGS 'NR_ARG_REGS -`#define NR_ARG_REGS 'NR_ARG_REGS define(defarg,`define(ARG$1,`$2')dnl #`define ARG'$1 $2' @@ -273,6 +280,10 @@ define(NBIF_ARG,`ifelse(eval($3 >= NR_ARG_REGS),0,`NBIF_REG_ARG($1,$3)',`NBIF_ST `/* #define NBIF_ARG_3_0 'NBIF_ARG(r3,3,0)` */' `/* #define NBIF_ARG_3_1 'NBIF_ARG(r3,3,1)` */' `/* #define NBIF_ARG_3_2 'NBIF_ARG(r3,3,2)` */' +`/* #define NBIF_ARG_4_0 'NBIF_ARG(r3,4,0)` */' +`/* #define NBIF_ARG_4_1 'NBIF_ARG(r3,4,1)` */' +`/* #define NBIF_ARG_4_2 'NBIF_ARG(r3,4,2)` */' +`/* #define NBIF_ARG_4_3 'NBIF_ARG(r3,4,3)` */' `/* #define NBIF_ARG_5_0 'NBIF_ARG(r3,5,0)` */' `/* #define NBIF_ARG_5_1 'NBIF_ARG(r3,5,1)` */' `/* #define NBIF_ARG_5_2 'NBIF_ARG(r3,5,2)` */' @@ -294,6 +305,7 @@ define(NBIF_RET,`NBIF_RET_N(eval(RET_POP($1)))')dnl `/* #define NBIF_RET_1 'NBIF_RET(1)` */' `/* #define NBIF_RET_2 'NBIF_RET(2)` */' `/* #define NBIF_RET_3 'NBIF_RET(3)` */' +`/* #define NBIF_RET_4 'NBIF_RET(4)` */' `/* #define NBIF_RET_5 'NBIF_RET(5)` */' dnl @@ -309,4 +321,6 @@ define(QUICK_CALL_RET,`NBIF_POP_N(eval(RET_POP($2)))b $1')dnl `/* #define QUICK_CALL_RET_F_3 'QUICK_CALL_RET(F,3)` */' `/* #define QUICK_CALL_RET_F_5 'QUICK_CALL_RET(F,5)` */' +`#endif /* ASM */' + `#endif /* HIPE_PPC_ASM_H */' diff --git a/erts/emulator/hipe/hipe_ppc_bifs.m4 b/erts/emulator/hipe/hipe_ppc_bifs.m4 index 7cc2b5c7b6..b173b896b8 100644 --- a/erts/emulator/hipe/hipe_ppc_bifs.m4 +++ b/erts/emulator/hipe/hipe_ppc_bifs.m4 @@ -19,6 +19,7 @@ changecom(`/*', `*/')dnl */ +#`define ASM' include(`hipe/hipe_ppc_asm.m4') #`include' "config.h" #`include' "hipe_literals.h" @@ -45,9 +46,10 @@ define(HANDLE_GOT_MBUF,` * standard_bif_interface_1(nbif_name, cbif_name) * standard_bif_interface_2(nbif_name, cbif_name) * standard_bif_interface_3(nbif_name, cbif_name) + * standard_bif_interface_4(nbif_name, cbif_name) * standard_bif_interface_0(nbif_name, cbif_name) * - * Generate native interface for a BIF with 0-3 parameters and + * Generate native interface for a BIF with 0-4 parameters and * standard failure mode. */ define(standard_bif_interface_1, @@ -143,6 +145,41 @@ ASYM($1): TYPE_FUNCTION(ASYM($1)) #endif') +define(standard_bif_interface_4, +` +#ifndef HAVE_$1 +#`define' HAVE_$1 + GLOBAL(ASYM($1)) +ASYM($1): + /* Set up C argument registers. */ + mr r3, P + NBIF_ARG(r4,4,0) + NBIF_ARG(r5,4,1) + NBIF_ARG(r6,4,2) + NBIF_ARG(r7,4,3) + + /* Save caller-save registers and call the C function. */ + SAVE_CONTEXT_BIF + STORE r4, P_ARG0(r3) /* Store BIF__ARGS in def_arg_reg[] */ + STORE r5, P_ARG1(r3) + STORE r6, P_ARG2(r3) + STORE r7, P_ARG3(r3) + addi r4, r3, P_ARG0 + CALL_BIF($2) + TEST_GOT_MBUF + + /* Restore registers. Check for exception. */ + CMPI r3, THE_NON_VALUE + RESTORE_CONTEXT_BIF + beq- 1f + NBIF_RET(4) +1: /* workaround for bc:s small offset operand */ + b CSYM(nbif_4_simple_exception) + HANDLE_GOT_MBUF(4) + SET_SIZE(ASYM($1)) + TYPE_FUNCTION(ASYM($1)) +#endif') + define(standard_bif_interface_0, ` #ifndef HAVE_$1 diff --git a/erts/emulator/hipe/hipe_ppc_glue.S b/erts/emulator/hipe/hipe_ppc_glue.S index 6f0217c738..b07f4bc9c8 100644 --- a/erts/emulator/hipe/hipe_ppc_glue.S +++ b/erts/emulator/hipe/hipe_ppc_glue.S @@ -17,10 +17,9 @@ * %CopyrightEnd% */ - +#define ASM #include "hipe_ppc_asm.h" #include "hipe_literals.h" -#define ASM #include "hipe_mode_switch.h" .text @@ -296,7 +295,7 @@ CSYM(hipe_ppc_throw_to_native): * which should look as follows: * * stub for f/N: - * <set r12 to f's BEAM code address> + * <set r12 to f's export entry address> * <set r0 to N> * b nbif_callemu * @@ -312,10 +311,10 @@ CSYM(hipe_ppc_throw_to_native): */ GLOBAL(ASYM(nbif_callemu)) ASYM(nbif_callemu): - STORE r12, P_BEAM_IP(P) + STORE r12, P_CALLEE_EXP(P) STORE r0, P_ARITY(P) STORE_ARG_REGS - li r3, HIPE_MODE_SWITCH_RES_CALL + li r3, HIPE_MODE_SWITCH_RES_CALL_EXPORTED b .suspend_exit /* @@ -463,10 +462,12 @@ ASYM(nbif_fail): OPD(nbif_1_gc_after_bif) OPD(nbif_2_gc_after_bif) OPD(nbif_3_gc_after_bif) + OPD(nbif_4_gc_after_bif) GLOBAL(CSYM(nbif_0_gc_after_bif)) GLOBAL(CSYM(nbif_1_gc_after_bif)) GLOBAL(CSYM(nbif_2_gc_after_bif)) GLOBAL(CSYM(nbif_3_gc_after_bif)) + GLOBAL(CSYM(nbif_4_gc_after_bif)) CSYM(nbif_0_gc_after_bif): li r4, 0 b .gc_after_bif @@ -478,6 +479,9 @@ CSYM(nbif_2_gc_after_bif): b .gc_after_bif CSYM(nbif_3_gc_after_bif): li r4, 3 + b .gc_after_bif +CSYM(nbif_4_gc_after_bif): + li r4, 4 /*FALLTHROUGH*/ .gc_after_bif: stw r4, P_NARITY(P) /* Note: narity is a 32-bit field */ @@ -520,6 +524,11 @@ CSYM(nbif_2_simple_exception): GLOBAL(CSYM(nbif_3_simple_exception)) CSYM(nbif_3_simple_exception): li r4, 3 + b .nbif_simple_exception + OPD(nbif_3_simple_exception) + GLOBAL(CSYM(nbif_4_simple_exception)) +CSYM(nbif_4_simple_exception): + li r4, 4 /*FALLTHROUGH*/ .nbif_simple_exception: LOAD r3, P_FREASON(P) diff --git a/erts/emulator/hipe/hipe_process.h b/erts/emulator/hipe/hipe_process.h index 4ee99d78a2..86655ad42c 100644 --- a/erts/emulator/hipe/hipe_process.h +++ b/erts/emulator/hipe/hipe_process.h @@ -23,14 +23,17 @@ #define HIPE_PROCESS_H #include "erl_alloc.h" +#include "export.h" struct hipe_process_state { Eterm *nsp; /* Native stack pointer. */ Eterm *nstack; /* Native stack block start. */ Eterm *nstend; /* Native stack block end (start+size). */ - /* XXX: ncallee and closure could share space in a union */ - void (*ncallee)(void); /* Native code callee (label) to invoke. */ - Eterm closure; /* Used to pass a closure from native code. */ + union { + void (*ncallee)(void); /* Native code callee (label) to invoke. */ + Eterm closure; /* Used to pass a closure from native code. */ + Export* callee_exp; /* Used to pass export entry from native code */ + }u; Eterm *nstgraylim; /* Gray/white stack boundary. */ Eterm *nstblacklim; /* Black/gray stack boundary. Must exist if graylim exists. Ignored if no graylim. */ diff --git a/erts/emulator/hipe/hipe_risc_stack.c b/erts/emulator/hipe/hipe_risc_stack.c index 1183856c7e..bea3a0fecd 100644 --- a/erts/emulator/hipe/hipe_risc_stack.c +++ b/erts/emulator/hipe/hipe_risc_stack.c @@ -226,7 +226,7 @@ void (*hipe_handle_stack_trap(Process *p))(void) * The native stack MUST contain a stack frame as it appears on * entry to a function (actuals, caller's frame, caller's return address). * p->hipe.narity MUST contain the arity (number of actuals). - * On exit, p->hipe.ncallee is set to the handler's PC and p->hipe.nsp + * On exit, p->hipe.u.ncallee is set to the handler's PC and p->hipe.nsp * is set to its SP (low address of its stack frame). */ void hipe_find_handler(Process *p) @@ -254,7 +254,7 @@ void hipe_find_handler(Process *p) if ((exnra = sdesc_exnra(sdesc)) != 0 && (p->catches >= 0 || exnra == (unsigned long)&nbif_fail)) { - p->hipe.ncallee = (void(*)(void)) exnra; + p->hipe.u.ncallee = (void(*)(void)) exnra; p->hipe.nsp = nsp; p->hipe.narity = 0; /* update the gray/white boundary if we threw past it */ diff --git a/erts/emulator/hipe/hipe_sparc.c b/erts/emulator/hipe/hipe_sparc.c index 49d4da7bab..fea3b623a9 100644 --- a/erts/emulator/hipe/hipe_sparc.c +++ b/erts/emulator/hipe/hipe_sparc.c @@ -130,7 +130,7 @@ static void atexit_alloc_code_stats(void) #define ALLOC_CODE_STATS(X) do{}while(0) #endif -static void morecore(unsigned int alloc_bytes) +static int morecore(unsigned int alloc_bytes) { unsigned int map_bytes; char *map_hint, *map_start; @@ -158,10 +158,9 @@ static void morecore(unsigned int alloc_bytes) #endif , -1, 0); - if (map_start == MAP_FAILED) { - perror("mmap"); - abort(); - } + if (map_start == MAP_FAILED) + return -1; + ALLOC_CODE_STATS(total_mapped += map_bytes); /* Merge adjacent mappings, so the trailing portion of the previous @@ -177,6 +176,8 @@ static void morecore(unsigned int alloc_bytes) } ALLOC_CODE_STATS(atexit_alloc_code_stats()); + + return 0; } static void *alloc_code(unsigned int alloc_bytes) @@ -186,8 +187,8 @@ static void *alloc_code(unsigned int alloc_bytes) /* Align function entries. */ alloc_bytes = (alloc_bytes + 3) & ~3; - if (code_bytes < alloc_bytes) - morecore(alloc_bytes); + if (code_bytes < alloc_bytes && morecore(alloc_bytes) != 0) + return NULL; ALLOC_CODE_STATS(++nr_allocs); ALLOC_CODE_STATS(total_alloc += alloc_bytes); res = code_next; @@ -204,22 +205,22 @@ void *hipe_alloc_code(Uint nrbytes, Eterm callees, Eterm *trampolines, Process * return alloc_code(nrbytes); } -/* called from hipe_bif0.c:hipe_bifs_make_native_stub_2() - and hipe_bif0.c:hipe_make_stub() */ -void *hipe_make_native_stub(void *beamAddress, unsigned int beamArity) +void *hipe_make_native_stub(void *callee_exp, unsigned int beamArity) { unsigned int *code; unsigned int callEmuOffset; int i; code = alloc_code(5*sizeof(int)); + if (!code) + return NULL; /* sethi %hi(Address), %i4 */ - code[0] = 0x39000000 | (((unsigned int)beamAddress >> 10) & 0x3FFFFF); + code[0] = 0x39000000 | (((unsigned int)callee_exp >> 10) & 0x3FFFFF); /* or %g0, %o7, %i3 ! mov %o7, %i3 */ code[1] = 0xB610000F; /* or %i4, %lo(Address), %i4 */ - code[2] = 0xB8172000 | ((unsigned int)beamAddress & 0x3FF); + code[2] = 0xB8172000 | ((unsigned int)callee_exp & 0x3FF); /* call callemu */ callEmuOffset = (char*)nbif_callemu - (char*)&code[3]; code[3] = (1 << 30) | ((callEmuOffset >> 2) & 0x3FFFFFFF); diff --git a/erts/emulator/hipe/hipe_sparc.h b/erts/emulator/hipe/hipe_sparc.h index 1134b86004..2d92ca3ca8 100644 --- a/erts/emulator/hipe/hipe_sparc.h +++ b/erts/emulator/hipe/hipe_sparc.h @@ -47,8 +47,4 @@ static __inline__ int hipe_word32_address_ok(void *address) extern void hipe_sparc_inc_stack(void); -/* for hipe_bifs_enter_code_2 */ -extern void *hipe_alloc_code(Uint nrbytes, Eterm callees, Eterm *trampolines, Process *p); -#define HIPE_ALLOC_CODE(n,c,t,p) hipe_alloc_code((n),(c),(t),(p)) - #endif /* HIPE_SPARC_H */ diff --git a/erts/emulator/hipe/hipe_sparc_asm.m4 b/erts/emulator/hipe/hipe_sparc_asm.m4 index 227d10ed80..8020104e40 100644 --- a/erts/emulator/hipe/hipe_sparc_asm.m4 +++ b/erts/emulator/hipe/hipe_sparc_asm.m4 @@ -29,6 +29,14 @@ define(LEAF_WORDS,16)dnl number of stack words for leaf functions define(NR_ARG_REGS,4)dnl admissible values are 0 to 6, inclusive `#define SPARC_LEAF_WORDS 'LEAF_WORDS +`#define SPARC_NR_ARG_REGS 'NR_ARG_REGS +`#define NR_ARG_REGS 'NR_ARG_REGS + + +`#ifdef ASM' +/* + * Only assembler stuff from here on (when included from *.S) + */ /* * Reserved registers. @@ -80,9 +88,6 @@ define(NR_ARG_REGS,4)dnl admissible values are 0 to 6, inclusive /* * Argument (parameter) registers. */ -`#define SPARC_NR_ARG_REGS 'NR_ARG_REGS -`#define NR_ARG_REGS 'NR_ARG_REGS - define(defarg,`define(ARG$1,`$2')dnl #`define ARG'$1 $2' )dnl @@ -171,6 +176,10 @@ define(NBIF_ARG,`ifelse(eval($3 >= NR_ARG_REGS),0,`NBIF_REG_ARG($1,$3)',`NBIF_ST `/* #define NBIF_ARG_3_0 'NBIF_ARG(r1,3,0)` */' `/* #define NBIF_ARG_3_1 'NBIF_ARG(r2,3,1)` */' `/* #define NBIF_ARG_3_2 'NBIF_ARG(r3,3,2)` */' +`/* #define NBIF_ARG_4_0 'NBIF_ARG(r1,4,0)` */' +`/* #define NBIF_ARG_4_1 'NBIF_ARG(r2,4,1)` */' +`/* #define NBIF_ARG_4_2 'NBIF_ARG(r3,4,2)` */' +`/* #define NBIF_ARG_4_3 'NBIF_ARG(r3,4,3)` */' `/* #define NBIF_ARG_5_0 'NBIF_ARG(r1,5,0)` */' `/* #define NBIF_ARG_5_1 'NBIF_ARG(r2,5,1)` */' `/* #define NBIF_ARG_5_2 'NBIF_ARG(r3,5,2)` */' @@ -195,6 +204,7 @@ define(NBIF_RET,`NBIF_RET_N(eval(RET_POP($1)))')dnl `/* #define NBIF_RET_1 'NBIF_RET(1)` */' `/* #define NBIF_RET_2 'NBIF_RET(2)` */' `/* #define NBIF_RET_3 'NBIF_RET(3)` */' +`/* #define NBIF_RET_4 'NBIF_RET(4)` */' `/* #define NBIF_RET_5 'NBIF_RET(5)` */' dnl @@ -210,4 +220,6 @@ define(QUICK_CALL_RET,`ba $1; NBIF_POP_N(eval(RET_POP($2)))')dnl `/* #define QUICK_CALL_RET_F_3 'QUICK_CALL_RET(F,3)` */' `/* #define QUICK_CALL_RET_F_5 'QUICK_CALL_RET(F,5)` */' +`#endif /* ASM */' + `#endif /* HIPE_SPARC_ASM_H */' diff --git a/erts/emulator/hipe/hipe_sparc_bifs.m4 b/erts/emulator/hipe/hipe_sparc_bifs.m4 index ca5af45d58..8dfb28c8e0 100644 --- a/erts/emulator/hipe/hipe_sparc_bifs.m4 +++ b/erts/emulator/hipe/hipe_sparc_bifs.m4 @@ -19,6 +19,7 @@ changecom(`/*', `*/')dnl */ +#`define ASM' include(`hipe/hipe_sparc_asm.m4') #`include' "config.h" #`include' "hipe_literals.h" @@ -53,9 +54,10 @@ define(HANDLE_GOT_MBUF,` * standard_bif_interface_1(nbif_name, cbif_name) * standard_bif_interface_2(nbif_name, cbif_name) * standard_bif_interface_3(nbif_name, cbif_name) + * standard_bif_interface_3(nbif_name, cbif_name) * standard_bif_interface_0(nbif_name, cbif_name) * - * Generate native interface for a BIF with 0-3 parameters and + * Generate native interface for a BIF with 0-4 parameters and * standard failure mode. */ define(standard_bif_interface_1, @@ -145,6 +147,39 @@ $1: .type $1, #function #endif') +define(standard_bif_interface_4, +` +#ifndef HAVE_$1 +#`define' HAVE_$1 + .global $1 +$1: + /* Set up C argument registers. */ + mov P, %o0 + NBIF_ARG(%o1,4,0) + NBIF_ARG(%o2,4,1) + NBIF_ARG(%o3,4,2) + NBIF_ARG(%o4,4,3) + + /* Save caller-save registers and call the C function. */ + SAVE_CONTEXT_BIF + st %o1, [%o0+P_ARG0] ! Store BIF__ARGS in def_arg_reg + st %o2, [%o0+P_ARG1] + st %o3, [%o0+P_ARG2] + st %o4, [%o0+P_ARG3] + add %o0, P_ARG0, %o1 + CALL_BIF($2) + nop + TEST_GOT_MBUF + + /* Restore registers. Check for exception. */ + TEST_GOT_EXN(4) + RESTORE_CONTEXT_BIF + NBIF_RET(4) + HANDLE_GOT_MBUF(4) + .size $1, .-$1 + .type $1, #function +#endif') + define(standard_bif_interface_0, ` #ifndef HAVE_$1 diff --git a/erts/emulator/hipe/hipe_sparc_glue.S b/erts/emulator/hipe/hipe_sparc_glue.S index 44bdf1bc7e..094a87fd58 100644 --- a/erts/emulator/hipe/hipe_sparc_glue.S +++ b/erts/emulator/hipe/hipe_sparc_glue.S @@ -18,10 +18,9 @@ * %CopyrightEnd% */ - +#define ASM #include "hipe_sparc_asm.h" #include "hipe_literals.h" -#define ASM #include "hipe_mode_switch.h" .section ".text" @@ -155,9 +154,9 @@ hipe_sparc_throw_to_native: * which should look as follows: * * stub for f/N: - * sethi %hi(f's BEAM code address), TEMP_ARG0 + * sethi %hi(f's export entry address), TEMP_ARG0 * mov RA, TEMP_RA ! because the call below clobbers RA (%o7) - * or TEMP_ARG0, %lo(f's BEAM code address), TEMP_ARG0 + * or TEMP_ARG0, %lo(f's export entry address), TEMP_ARG0 * call nbif_callemu ! clobbers RA! * mov N, TEMP_ARG1 ! delay slot: TEMP_ARG1 := ARITY * @@ -165,12 +164,12 @@ hipe_sparc_throw_to_native: */ .global nbif_callemu nbif_callemu: - st TEMP_ARG0, [P+P_BEAM_IP] + st TEMP_ARG0, [P+P_CALLEE_EXP] st TEMP_ARG1, [P+P_ARITY] st TEMP_RA, [P+P_NRA] STORE_ARG_REGS ba .flush_exit - mov HIPE_MODE_SWITCH_RES_CALL, %o0 + mov HIPE_MODE_SWITCH_RES_CALL_EXPORTED, %o0 /* * nbif_apply @@ -317,6 +316,7 @@ nbif_fail: .global nbif_1_gc_after_bif .global nbif_2_gc_after_bif .global nbif_3_gc_after_bif + .global nbif_4_gc_after_bif nbif_0_gc_after_bif: ba .gc_after_bif mov 0, %o1 /* delay slot */ @@ -327,7 +327,10 @@ nbif_2_gc_after_bif: ba .gc_after_bif mov 2, %o1 /* delay slot */ nbif_3_gc_after_bif: - mov 3, %o1 + ba .gc_after_bif + mov 3, %o1 /* delay slot */ +nbif_4_gc_after_bif: + mov 4, %o1 /*FALLTHROUGH*/ .gc_after_bif: st %o1, [P+P_NARITY] @@ -365,7 +368,11 @@ nbif_2_simple_exception: mov 2, %o1 /* delay slot */ .global nbif_3_simple_exception nbif_3_simple_exception: - mov 3, %o1 + ba .nbif_simple_exception + mov 3, %o1 /* delay slot */ + .global nbif_4_simple_exception +nbif_4_simple_exception: + mov 4, %o1 /*FALLTHROUGH*/ .nbif_simple_exception: ld [P+P_FREASON], %o0 diff --git a/erts/emulator/hipe/hipe_stack.h b/erts/emulator/hipe/hipe_stack.h index 66f9f04c73..4cfdb54dd8 100644 --- a/erts/emulator/hipe/hipe_stack.h +++ b/erts/emulator/hipe/hipe_stack.h @@ -108,12 +108,23 @@ extern int hipe_fill_stacktrace(Process*, int, Eterm**); #if 0 && defined(HIPE_NSTACK_GROWS_UP) #define hipe_nstack_start(p) ((p)->hipe.nstack) #define hipe_nstack_used(p) ((p)->hipe.nsp - (p)->hipe.nstack) +#define hipe_nstack_avail(p) ((p)->hipe.nstend - (p)->hipe.nsp) #endif #if defined(HIPE_NSTACK_GROWS_DOWN) #define hipe_nstack_start(p) ((p)->hipe.nsp) #define hipe_nstack_used(p) ((p)->hipe.nstend - (p)->hipe.nsp) +#define hipe_nstack_avail(p) ((unsigned)((p)->hipe.nsp - (p)->hipe.nstack)) #endif +/* ensure that at least nwords words are available on the native stack */ +static __inline__ void hipe_check_nstack(Process *p, unsigned nwords) +{ + extern void hipe_inc_nstack(Process *p); + + while (hipe_nstack_avail(p) < nwords) + hipe_inc_nstack(p); +} + /* * GC support procedures */ diff --git a/erts/emulator/hipe/hipe_x86.c b/erts/emulator/hipe/hipe_x86.c index 327c74e9aa..998905ea63 100644 --- a/erts/emulator/hipe/hipe_x86.c +++ b/erts/emulator/hipe/hipe_x86.c @@ -108,7 +108,7 @@ static void atexit_alloc_code_stats(void) #define MAP_ANONYMOUS MAP_ANON #endif -static void morecore(unsigned int alloc_bytes) +static int morecore(unsigned int alloc_bytes) { unsigned int map_bytes; char *map_hint, *map_start; @@ -136,10 +136,9 @@ static void morecore(unsigned int alloc_bytes) #endif , -1, 0); - if (map_start == MAP_FAILED) { - perror("mmap"); - abort(); - } + if (map_start == MAP_FAILED) + return -1; + ALLOC_CODE_STATS(total_mapped += map_bytes); /* Merge adjacent mappings, so the trailing portion of the previous @@ -155,6 +154,8 @@ static void morecore(unsigned int alloc_bytes) } ALLOC_CODE_STATS(atexit_alloc_code_stats()); + + return 0; } static void *alloc_code(unsigned int alloc_bytes) @@ -164,8 +165,8 @@ static void *alloc_code(unsigned int alloc_bytes) /* Align function entries. */ alloc_bytes = (alloc_bytes + 3) & ~3; - if (code_bytes < alloc_bytes) - morecore(alloc_bytes); + if (code_bytes < alloc_bytes && morecore(alloc_bytes) != 0) + return NULL; ALLOC_CODE_STATS(++nr_allocs); ALLOC_CODE_STATS(total_alloc += alloc_bytes); res = code_next; @@ -182,18 +183,16 @@ void *hipe_alloc_code(Uint nrbytes, Eterm callees, Eterm *trampolines, Process * return alloc_code(nrbytes); } -/* called from hipe_bif0.c:hipe_bifs_make_native_stub_2() - and hipe_bif0.c:hipe_make_stub() */ -void *hipe_make_native_stub(void *beamAddress, unsigned int beamArity) +void *hipe_make_native_stub(void *callee_exp, unsigned int beamArity) { /* * This creates a native code stub with the following contents: * - * movl $Address, P_BEAM_IP(%ebp) + * movl $Address, P_CALLEE_EXP(%ebp) * movb $Arity, P_ARITY(%ebp) * jmp callemu * - * The stub has variable size, depending on whether the P_BEAM_IP + * The stub has variable size, depending on whether the P_CALLEE_EXP * and P_ARITY offsets fit in 8-bit signed displacements or not. * The rel32 offset in the final jmp depends on its actual location, * which also depends on the size of the previous instructions. @@ -206,28 +205,30 @@ void *hipe_make_native_stub(void *beamAddress, unsigned int beamArity) codeSize = /* 16, 19, or 22 bytes */ 16 + /* 16 when both offsets are 8-bit */ - (P_BEAM_IP >= 128 ? 3 : 0) + + (P_CALLEE_EXP >= 128 ? 3 : 0) + (P_ARITY >= 128 ? 3 : 0); codep = code = alloc_code(codeSize); + if (!code) + return NULL; - /* movl $beamAddress, P_BEAM_IP(%ebp); 3 or 6 bytes, plus 4 */ + /* movl $beamAddress, P_CALLEE_EXP(%ebp); 3 or 6 bytes, plus 4 */ codep[0] = 0xc7; -#if P_BEAM_IP >= 128 +#if P_CALLEE_EXP >= 128 codep[1] = 0x85; /* disp32[EBP] */ - codep[2] = P_BEAM_IP & 0xFF; - codep[3] = (P_BEAM_IP >> 8) & 0xFF; - codep[4] = (P_BEAM_IP >> 16) & 0xFF; - codep[5] = (P_BEAM_IP >> 24) & 0xFF; + codep[2] = P_CALLEE_EXP & 0xFF; + codep[3] = (P_CALLEE_EXP >> 8) & 0xFF; + codep[4] = (P_CALLEE_EXP >> 16) & 0xFF; + codep[5] = (P_CALLEE_EXP >> 24) & 0xFF; codep += 6; #else codep[1] = 0x45; /* disp8[EBP] */ - codep[2] = P_BEAM_IP; + codep[2] = P_CALLEE_EXP; codep += 3; #endif - codep[0] = ((unsigned int)beamAddress) & 0xFF; - codep[1] = ((unsigned int)beamAddress >> 8) & 0xFF; - codep[2] = ((unsigned int)beamAddress >> 16) & 0xFF; - codep[3] = ((unsigned int)beamAddress >> 24) & 0xFF; + codep[0] = ((unsigned int)callee_exp) & 0xFF; + codep[1] = ((unsigned int)callee_exp >> 8) & 0xFF; + codep[2] = ((unsigned int)callee_exp >> 16) & 0xFF; + codep[3] = ((unsigned int)callee_exp >> 24) & 0xFF; codep += 4; /* movb $beamArity, P_ARITY(%ebp); 3 or 6 bytes */ diff --git a/erts/emulator/hipe/hipe_x86.h b/erts/emulator/hipe/hipe_x86.h index 97f09e38cd..f29117d0c4 100644 --- a/erts/emulator/hipe/hipe_x86.h +++ b/erts/emulator/hipe/hipe_x86.h @@ -53,8 +53,4 @@ extern void nbif_inc_stack_0(void); extern void nbif_handle_fp_exception(void); #endif -/* for hipe_bifs_enter_code_2 */ -extern void *hipe_alloc_code(Uint nrbytes, Eterm callees, Eterm *trampolines, Process *p); -#define HIPE_ALLOC_CODE(n,c,t,p) hipe_alloc_code((n),(c),(t),(p)) - #endif /* HIPE_X86_H */ diff --git a/erts/emulator/hipe/hipe_x86_asm.m4 b/erts/emulator/hipe/hipe_x86_asm.m4 index 020ccf8d4b..436feca506 100644 --- a/erts/emulator/hipe/hipe_x86_asm.m4 +++ b/erts/emulator/hipe/hipe_x86_asm.m4 @@ -33,6 +33,18 @@ define(SIMULATE_NSP,0)dnl change to 1 to simulate call/ret insns `#define X86_LEAF_WORDS 'LEAF_WORDS `#define LEAF_WORDS 'LEAF_WORDS +`#define X86_NR_ARG_REGS 'NR_ARG_REGS +`#define NR_ARG_REGS 'NR_ARG_REGS + +`#define X86_HP_IN_ESI 'HP_IN_ESI +`#define X86_SIMULATE_NSP 'SIMULATE_NSP + + +`#ifdef ASM' +/* + * Only assembler stuff from here on (when included from *.S) + */ + /* * Workarounds for Darwin. */ @@ -60,7 +72,6 @@ ifelse(OPSYS,darwin,`` */ `#define P %ebp' -`#define X86_HP_IN_ESI 'HP_IN_ESI `#if X86_HP_IN_ESI #define SAVE_HP movl %esi, P_HP(P) #define RESTORE_HP movl P_HP(P), %esi @@ -73,7 +84,6 @@ ifelse(OPSYS,darwin,`` #define SAVE_CSP movl %esp, P_CSP(P) #define RESTORE_CSP movl P_CSP(P), %esp' -`#define X86_SIMULATE_NSP 'SIMULATE_NSP /* * Context switching macros. @@ -100,12 +110,10 @@ ifelse(OPSYS,darwin,`` SAVE_CACHED_STATE; \ SWITCH_ERLANG_TO_C_QUICK' + /* * Argument (parameter) registers. */ -`#define X86_NR_ARG_REGS 'NR_ARG_REGS -`#define NR_ARG_REGS 'NR_ARG_REGS - ifelse(eval(NR_ARG_REGS >= 1),0,, ``#define ARG0 %eax '')dnl @@ -204,6 +212,7 @@ define(NBIF_COPY_NSP,`ifelse(eval($1 > NR_ARG_REGS),0,,`movl %esp, TEMP_NSP')')d `/* #define NBIF_COPY_NSP_1 'NBIF_COPY_NSP(1)` */' `/* #define NBIF_COPY_NSP_2 'NBIF_COPY_NSP(2)` */' `/* #define NBIF_COPY_NSP_3 'NBIF_COPY_NSP(3)` */' +`/* #define NBIF_COPY_NSP_4 'NBIF_COPY_NSP(4)` */' `/* #define NBIF_COPY_NSP_5 'NBIF_COPY_NSP(5)` */' dnl @@ -227,6 +236,10 @@ define(NBIF_ARG_OPND,`ifelse(eval($2 >= NR_ARG_REGS),0,`ARG'$2,BASE_OFFSET(eval( `/* #define NBIF_ARG_OPND_3_0 'NBIF_ARG_OPND(3,0)` */' `/* #define NBIF_ARG_OPND_3_1 'NBIF_ARG_OPND(3,1)` */' `/* #define NBIF_ARG_OPND_3_2 'NBIF_ARG_OPND(3,2)` */' +`/* #define NBIF_ARG_OPND_4_0 'NBIF_ARG_OPND(4,0)` */' +`/* #define NBIF_ARG_OPND_4_1 'NBIF_ARG_OPND(4,1)` */' +`/* #define NBIF_ARG_OPND_4_2 'NBIF_ARG_OPND(4,2)` */' +`/* #define NBIF_ARG_OPND_4_3 'NBIF_ARG_OPND(4,3)` */' `/* #define NBIF_ARG_OPND_5_0 'NBIF_ARG_OPND(5,0)` */' `/* #define NBIF_ARG_OPND_5_1 'NBIF_ARG_OPND(5,1)` */' `/* #define NBIF_ARG_OPND_5_2 'NBIF_ARG_OPND(5,2)` */' @@ -266,6 +279,7 @@ define(NBIF_RET,`NBIF_RET_N(eval(RET_POP($1)))')dnl `/* #define NBIF_RET_1 'NBIF_RET(1)` */' `/* #define NBIF_RET_2 'NBIF_RET(2)` */' `/* #define NBIF_RET_3 'NBIF_RET(3)` */' +`/* #define NBIF_RET_4 'NBIF_RET(4)` */' `/* #define NBIF_RET_5 'NBIF_RET(5)` */' dnl @@ -282,4 +296,6 @@ define(LOAD_CALLER_SAVE,`LAR_N(eval(NR_CALLER_SAVE-1))')dnl `#define STORE_CALLER_SAVE 'STORE_CALLER_SAVE `#define LOAD_CALLER_SAVE 'LOAD_CALLER_SAVE +`#endif /* ASM */' + `#endif /* HIPE_X86_ASM_H */' diff --git a/erts/emulator/hipe/hipe_x86_bifs.m4 b/erts/emulator/hipe/hipe_x86_bifs.m4 index dd6980f555..b0064ee628 100644 --- a/erts/emulator/hipe/hipe_x86_bifs.m4 +++ b/erts/emulator/hipe/hipe_x86_bifs.m4 @@ -19,6 +19,7 @@ changecom(`/*', `*/')dnl */ +#`define ASM' include(`hipe/hipe_x86_asm.m4') #`include' "config.h" #`include' "hipe_literals.h" @@ -47,6 +48,7 @@ define(HANDLE_GOT_MBUF,` * standard_bif_interface_1(nbif_name, cbif_name) * standard_bif_interface_2(nbif_name, cbif_name) * standard_bif_interface_3(nbif_name, cbif_name) + * standard_bif_interface_4(nbif_name, cbif_name) * standard_bif_interface_0(nbif_name, cbif_name) * * Generate native interface for a BIF with 0-3 parameters and @@ -157,6 +159,43 @@ ASYM($1): TYPE_FUNCTION(ASYM($1)) #endif') +define(standard_bif_interface_4, +` +#ifndef HAVE_$1 +#`define' HAVE_$1 + TEXT + .align 4 + GLOBAL(ASYM($1)) +ASYM($1): + /* copy native stack pointer */ + NBIF_COPY_NSP(4) + + /* switch to C stack */ + SWITCH_ERLANG_TO_C + + /* make the call on the C stack */ + NBIF_ARG_REG(0,P) + NBIF_ARG(2,4,0) + NBIF_ARG(3,4,1) + NBIF_ARG(4,4,2) + NBIF_ARG(5,4,3) + lea 8(%esp), %eax + NBIF_ARG_REG(1,%eax) /* BIF__ARGS */ + CALL_BIF($2) + TEST_GOT_MBUF + + /* switch to native stack */ + SWITCH_C_TO_ERLANG + + /* throw exception if failure, otherwise return */ + TEST_GOT_EXN + jz nbif_4_simple_exception + NBIF_RET(4) + HANDLE_GOT_MBUF(4) + SET_SIZE(ASYM($1)) + TYPE_FUNCTION(ASYM($1)) +#endif') + define(standard_bif_interface_0, ` #ifndef HAVE_$1 diff --git a/erts/emulator/hipe/hipe_x86_glue.S b/erts/emulator/hipe/hipe_x86_glue.S index 88b86f4de7..f124e36a26 100644 --- a/erts/emulator/hipe/hipe_x86_glue.S +++ b/erts/emulator/hipe/hipe_x86_glue.S @@ -18,10 +18,9 @@ * %CopyrightEnd% */ - +#define ASM #include "hipe_x86_asm.h" #include "hipe_literals.h" -#define ASM #include "hipe_mode_switch.h" /* @@ -104,7 +103,7 @@ ASYM(nbif_return): * stub (hipe_x86_loader.erl) which should look as follows: * * stub for f/N: - * movl $<f's BEAM code address>, P_BEAM_IP(P) + * movl $<f's export entry address>, P_CALLEE_EXP(P) * movb $<N>, P_ARITY(P) * jmp nbif_callemu * @@ -114,7 +113,7 @@ ASYM(nbif_return): GLOBAL(ASYM(nbif_callemu)) ASYM(nbif_callemu): STORE_ARG_REGS - movl $HIPE_MODE_SWITCH_RES_CALL, %eax + movl $HIPE_MODE_SWITCH_RES_CALL_EXPORTED, %eax jmp .suspend_exit /* @@ -300,6 +299,7 @@ ASYM(nbif_fail): GLOBAL(nbif_1_gc_after_bif) GLOBAL(nbif_2_gc_after_bif) GLOBAL(nbif_3_gc_after_bif) + GLOBAL(nbif_4_gc_after_bif) .align 4 nbif_0_gc_after_bif: xorl %edx, %edx @@ -315,6 +315,10 @@ nbif_2_gc_after_bif: .align 4 nbif_3_gc_after_bif: movl $3, %edx + jmp .gc_after_bif + .align 4 +nbif_4_gc_after_bif: + movl $4, %edx /*FALLTHROUGH*/ .align 4 .gc_after_bif: @@ -338,6 +342,7 @@ nbif_3_gc_after_bif: GLOBAL(nbif_1_simple_exception) GLOBAL(nbif_2_simple_exception) GLOBAL(nbif_3_simple_exception) + GLOBAL(nbif_4_simple_exception) .align 4 nbif_0_simple_exception: xorl %eax, %eax @@ -353,6 +358,10 @@ nbif_2_simple_exception: .align 4 nbif_3_simple_exception: movl $3, %eax + jmp .nbif_simple_exception + .align 4 +nbif_4_simple_exception: + movl $4, %eax /*FALLTHROUGH*/ .align 4 .nbif_simple_exception: diff --git a/erts/emulator/hipe/hipe_x86_stack.c b/erts/emulator/hipe/hipe_x86_stack.c index 9ad3fa9d31..7f1c2f7d41 100644 --- a/erts/emulator/hipe/hipe_x86_stack.c +++ b/erts/emulator/hipe/hipe_x86_stack.c @@ -209,7 +209,7 @@ void (*hipe_handle_stack_trap(Process *p))(void) * The native stack MUST contain a stack frame as it appears on * entry to a function (return address, actuals, caller's frame). * p->hipe.narity MUST contain the arity (number of actuals). - * On exit, p->hipe.ncallee is set to the handler's PC and p->hipe.nsp + * On exit, p->hipe.u.ncallee is set to the handler's PC and p->hipe.nsp * is set to its SP (low address of its stack frame). */ void hipe_find_handler(Process *p) @@ -240,7 +240,7 @@ void hipe_find_handler(Process *p) if ((exnra = sdesc_exnra(sdesc)) != 0 && (p->catches >= 0 || exnra == (unsigned long)nbif_fail)) { - p->hipe.ncallee = (void(*)(void)) exnra; + p->hipe.u.ncallee = (void(*)(void)) exnra; p->hipe.nsp = nsp; p->hipe.narity = 0; /* update the gray/white boundary if we threw past it */ diff --git a/erts/emulator/sys/unix/erl_child_setup.c b/erts/emulator/sys/unix/erl_child_setup.c index 94eb6b1547..5ad92dad02 100644 --- a/erts/emulator/sys/unix/erl_child_setup.c +++ b/erts/emulator/sys/unix/erl_child_setup.c @@ -101,7 +101,9 @@ main(int argc, char *argv[]) if (sscanf(argv[CS_ARGV_FD_CR_IX], "%d:%d", &from, &to) != 2) return 1; -#if defined(__ANDROID__) +#if defined(HAVE_CLOSEFROM) + closefrom(from); +#elif defined(__ANDROID__) for (i = from; i <= to; i++) { if (i!=__system_properties_fd) (void) close(i); @@ -109,13 +111,6 @@ main(int argc, char *argv[]) #else for (i = from; i <= to; i++) (void) close(i); -#endif /* __ANDROID__ */ - -#if defined(HAVE_CLOSEFROM) - closefrom(from); -#else - for (i = from; i <= to; i++) - (void) close(i); #endif if (!(argv[CS_ARGV_WD_IX][0] == '.' && argv[CS_ARGV_WD_IX][1] == '\0') @@ -147,8 +142,6 @@ main(int argc, char *argv[]) return 1; } - - #if defined(__ANDROID__) int __system_properties_fd(void) { diff --git a/erts/emulator/sys/unix/erl_unix_sys.h b/erts/emulator/sys/unix/erl_unix_sys.h index 5417bb2687..46d0b8dc9b 100644 --- a/erts/emulator/sys/unix/erl_unix_sys.h +++ b/erts/emulator/sys/unix/erl_unix_sys.h @@ -45,7 +45,7 @@ #include <fcntl.h> #include "erl_errno.h" #include <signal.h> - +#include <setjmp.h> #if HAVE_SYS_SOCKETIO_H # include <sys/socketio.h> @@ -262,7 +262,7 @@ erts_os_monotonic_time(void) * */ -#if (defined(HAVE_GETHRVTIME) || defined(HAVE_CLOCK_GETTIME)) +#if (defined(HAVE_GETHRVTIME) || defined(HAVE_CLOCK_GETTIME_CPU_TIME)) typedef long long SysCpuTime; typedef struct timespec SysTimespec; @@ -274,7 +274,7 @@ typedef struct timespec SysTimespec; int sys_start_hrvtime(void); int sys_stop_hrvtime(void); -#elif defined(HAVE_CLOCK_GETTIME) +#elif defined(HAVE_CLOCK_GETTIME_CPU_TIME) #define sys_clock_gettime(cid,tp) clock_gettime((cid),&(tp)) #define sys_get_proc_cputime(t,tp) sys_clock_gettime(CLOCK_PROCESS_CPUTIME_ID,(tp)) @@ -285,13 +285,8 @@ int sys_stop_hrvtime(void); #define SYS_CLOCK_RESOLUTION 1 /* These are defined in sys.c */ -#if defined(SIG_SIGSET) /* Old SysV */ -RETSIGTYPE (*sys_sigset())(); -#elif defined(SIG_SIGNAL) /* Old BSD */ -RETSIGTYPE (*sys_sigset())(); -#else -RETSIGTYPE (*sys_sigset(int, RETSIGTYPE (*func)(int)))(int); -#endif +typedef void (*SIGFUNC)(int); +extern SIGFUNC sys_signal(int, SIGFUNC); extern void sys_sigrelease(int); extern void sys_sigblock(int); extern void sys_stop_cat(void); @@ -428,4 +423,28 @@ extern int exit_async(void); #define ERTS_EXIT_AFTER_DUMP _exit +#if !defined(__APPLE__) && !defined(__MACH__) +/* Some OS X versions do not allow (ab)using signal handlers like this */ +#define ERTS_HAVE_TRY_CATCH 1 + +/* We try to simulate a try catch in C with the help of signal handlers. + * Only use this as a very last resort, as it is not very portable and + * quite unstable. It is also not thread safe, so make sure that only + * one thread can call this at a time! + */ +extern void erts_sys_sigsegv_handler(int); +extern jmp_buf erts_sys_sigsegv_jmp; +#define ERTS_SYS_TRY_CATCH(EXPR,CATCH) \ + do { \ + SIGFUNC prev_handler = sys_signal(SIGSEGV, \ + erts_sys_sigsegv_handler); \ + if (!setjmp(erts_sys_sigsegv_jmp)) { \ + EXPR; \ + } else { \ + CATCH; \ + } \ + sys_signal(SIGSEGV,prev_handler); \ + } while(0) +#endif + #endif /* #ifndef _ERL_UNIX_SYS_H */ diff --git a/erts/emulator/sys/unix/sys.c b/erts/emulator/sys/unix/sys.c index 58b01c094e..f1c785890c 100644 --- a/erts/emulator/sys/unix/sys.c +++ b/erts/emulator/sys/unix/sys.c @@ -34,6 +34,7 @@ #include <termios.h> #include <ctype.h> #include <sys/utsname.h> +#include <sys/select.h> #ifdef ISC32 #include <sys/bsdtypes.h> @@ -85,14 +86,24 @@ static erts_smp_rwmtx_t environ_rwmtx; #define DISABLE_VFORK 0 #endif +#if defined IOV_MAX +#define MAXIOV IOV_MAX +#elif defined UIO_MAXIOV +#define MAXIOV UIO_MAXIOV +#else +#define MAXIOV 16 +#endif + #ifdef USE_THREADS # ifdef ENABLE_CHILD_WAITER_THREAD # define CHLDWTHR ENABLE_CHILD_WAITER_THREAD # else # define CHLDWTHR 0 # endif +# define FDBLOCK 1 #else # define CHLDWTHR 0 +# define FDBLOCK 0 #endif /* * [OTP-3906] @@ -121,6 +132,15 @@ struct ErtsSysReportExit_ { #endif }; +/* Used by the fd driver iff the fd could not be set to non-blocking */ +typedef struct ErtsSysBlocking_ { + ErlDrvPDL pdl; + int res; + int err; + unsigned int pkey; +} ErtsSysBlocking; + + /* This data is shared by these drivers - initialized by spawn_init() */ static struct driver_data { ErlDrvPort port_num; @@ -129,6 +149,8 @@ static struct driver_data { int pid; int alive; int status; + int terminating; + ErtsSysBlocking *blocking; } *driver_data; /* indexed by fd */ static ErtsSysReportExit *report_exit_list; @@ -202,10 +224,14 @@ static erts_smp_atomic_t sys_misc_mem_sz; #if defined(ERTS_SMP) static void smp_sig_notify(char c); static int sig_notify_fds[2] = {-1, -1}; -#elif defined(USE_THREADS) -static int async_fd[2]; + +static int sig_suspend_fds[2] = {-1, -1}; +#define ERTS_SYS_SUSPEND_SIGNAL SIGUSR2 + #endif +jmp_buf erts_sys_sigsegv_jmp; + #if CHLDWTHR || defined(ERTS_SMP) erts_mtx_t chld_stat_mtx; #endif @@ -246,6 +272,8 @@ static void note_child_death(int, int); static void* child_waiter(void *); #endif +static int crashdump_companion_cube_fd = -1; + /********************* General functions ****************************/ /* This is used by both the drivers and general I/O, must be set early */ @@ -586,6 +614,14 @@ erts_sys_pre_init(void) close(fd); } + /* We need a file descriptor to close in the crashdump creation. + * We close this one to be sure we can get a fd for our real file ... + * so, we create one here ... a stone to carry all the way home. + */ + + crashdump_companion_cube_fd = open("/dev/null", O_RDONLY); + + /* don't lose it, there will be cake */ } void @@ -643,39 +679,7 @@ erl_sys_init(void) /* signal handling */ -#ifdef SIG_SIGSET /* Old SysV */ -RETSIGTYPE (*sys_sigset(sig, func))() -int sig; -RETSIGTYPE (*func)(); -{ - return(sigset(sig, func)); -} -void sys_sigblock(int sig) -{ - sighold(sig); -} -void sys_sigrelease(int sig) -{ - sigrelse(sig); -} -#else /* !SIG_SIGSET */ -#ifdef SIG_SIGNAL /* Old BSD */ -RETSIGTYPE (*sys_sigset(sig, func))(int, int) -int sig; -RETSIGTYPE (*func)(); -{ - return(signal(sig, func)); -} -sys_sigblock(int sig) -{ - sigblock(sig); -} -sys_sigrelease(int sig) -{ - sigsetmask(sigblock(0) & ~sigmask(sig)); -} -#else /* !SIG_SIGNAL */ /* The True Way - POSIX!:-) */ -RETSIGTYPE (*sys_sigset(int sig, RETSIGTYPE (*func)(int)))(int) +SIGFUNC sys_signal(int sig, SIGFUNC func) { struct sigaction act, oact; @@ -708,36 +712,47 @@ void sys_sigrelease(int sig) sigaddset(&mask, sig); sigprocmask(SIG_UNBLOCK, &mask, (sigset_t *)NULL); } -#endif /* !SIG_SIGNAL */ -#endif /* !SIG_SIGSET */ -#if (0) /* not used? -- gordon */ -static void (*break_func)(); -static RETSIGTYPE break_handler(int sig) -{ -#ifdef QNX - /* Turn off SIGCHLD during break processing */ - sys_sigblock(SIGCHLD); -#endif - (*break_func)(); -#ifdef QNX - sys_sigrelease(SIGCHLD); -#endif +void erts_sys_sigsegv_handler(int signo) { + if (signo == SIGSEGV) { + longjmp(erts_sys_sigsegv_jmp, 1); + } +} + +/* + * Function returns 1 if we can read from all values in between + * start and stop. + */ +int +erts_sys_is_area_readable(char *start, char *stop) { + int fds[2]; + if (!pipe(fds)) { + /* We let write try to figure out if the pointers are readable */ + int res = write(fds[1], start, (char*)stop - (char*)start); + if (res == -1) { + close(fds[0]); + close(fds[1]); + return 0; + } + close(fds[0]); + close(fds[1]); + return 1; + } + return 0; + } -#endif /* 0 */ static ERTS_INLINE int prepare_crash_dump(int secs) { #define NUFBUF (3) - int i, max; + int i; char env[21]; /* enough to hold any 64-bit integer */ size_t envsz; DeclareTmpHeapNoproc(heap,NUFBUF); Port *heart_port; Eterm *hp = heap; Eterm list = NIL; - int heart_fd[2] = {-1,-1}; int has_heart = 0; UseTmpHeapNoproc(NUFBUF); @@ -760,43 +775,22 @@ prepare_crash_dump(int secs) alarm((unsigned int)secs); } - if (heart_port) { - /* hearts input fd - * We "know" drv_data is the in_fd since the port is started with read|write - */ - heart_fd[0] = (int)heart_port->drv_data; - heart_fd[1] = (int)driver_data[heart_fd[0]].ofd; - has_heart = 1; + /* close all viable sockets via emergency close callbacks. + * Specifically we want to close epmd sockets. + */ - list = CONS(hp, make_small(8), list); hp += 2; + erts_emergency_close_ports(); + if (heart_port) { + has_heart = 1; + list = CONS(hp, make_small(8), list); hp += 2; /* send to heart port, CMD = 8, i.e. prepare crash dump =o */ erts_port_output(NULL, ERTS_PORT_SIG_FLG_FORCE_IMM_CALL, heart_port, heart_port->common.id, list, NULL); } - /* Make sure we unregister at epmd (unknown fd) and get at least - one free filedescriptor (for erl_crash.dump) */ - - max = max_files; - if (max < 1024) - max = 1024; - for (i = 3; i < max; i++) { -#if defined(ERTS_SMP) - /* We don't want to close the signal notification pipe... */ - if (i == sig_notify_fds[0] || i == sig_notify_fds[1]) - continue; -#elif defined(USE_THREADS) - /* We don't want to close the async notification pipe... */ - if (i == async_fd[0] || i == async_fd[1]) - continue; -#endif - /* We don't want to close our heart yet ... */ - if (i == heart_fd[0] || i == heart_fd[1]) - continue; - - close(i); - } + /* Make sure we have a fd for our crashdump file. */ + close(crashdump_companion_cube_fd); envsz = sizeof(env); i = erts_sys_getenv__("ERL_CRASH_DUMP_NICE", env, &envsz); @@ -874,9 +868,23 @@ sigusr1_exit(void) #ifdef ETHR_UNUSABLE_SIGUSRX #warning "Unusable SIGUSR1 & SIGUSR2. Disabling use of these signals" -#endif -#ifndef ETHR_UNUSABLE_SIGUSRX +#else + +#ifdef ERTS_SMP +void +sys_thr_suspend(erts_tid_t tid) { + erts_thr_kill(tid, ERTS_SYS_SUSPEND_SIGNAL); +} + +void +sys_thr_resume(erts_tid_t tid) { + int i = 0, res; + do { + res = write(sig_suspend_fds[1],&i,sizeof(i)); + } while (res < 0 && errno == EAGAIN); +} +#endif #if (defined(SIG_SIGSET) || defined(SIG_SIGNAL)) static RETSIGTYPE user_signal1(void) @@ -891,20 +899,20 @@ static RETSIGTYPE user_signal1(int signum) #endif } -#ifdef QUANTIFY +#ifdef ERTS_SMP #if (defined(SIG_SIGSET) || defined(SIG_SIGNAL)) -static RETSIGTYPE user_signal2(void) +static RETSIGTYPE suspend_signal(void) #else -static RETSIGTYPE user_signal2(int signum) +static RETSIGTYPE suspend_signal(int signum) #endif { -#ifdef ERTS_SMP - smp_sig_notify('2'); -#else - quantify_save_data(); -#endif + int res; + int buf[1]; + do { + res = read(sig_suspend_fds[0], buf, sizeof(int)); + } while (res < 0 && errno == EINTR); } -#endif +#endif /* #ifdef ERTS_SMP */ #endif /* #ifndef ETHR_UNUSABLE_SIGUSRX */ @@ -929,9 +937,9 @@ static RETSIGTYPE do_quit(int signum) /* Disable break */ void erts_set_ignore_break(void) { - sys_sigset(SIGINT, SIG_IGN); - sys_sigset(SIGQUIT, SIG_IGN); - sys_sigset(SIGTSTP, SIG_IGN); + sys_signal(SIGINT, SIG_IGN); + sys_signal(SIGQUIT, SIG_IGN); + sys_signal(SIGTSTP, SIG_IGN); } /* Don't use ctrl-c for break handler but let it be @@ -954,14 +962,14 @@ void erts_replace_intr(void) { void init_break_handler(void) { - sys_sigset(SIGINT, request_break); + sys_signal(SIGINT, request_break); #ifndef ETHR_UNUSABLE_SIGUSRX - sys_sigset(SIGUSR1, user_signal1); -#ifdef QUANTIFY - sys_sigset(SIGUSR2, user_signal2); -#endif + sys_signal(SIGUSR1, user_signal1); +#ifdef ERTS_SMP + sys_signal(ERTS_SYS_SUSPEND_SIGNAL, suspend_signal); +#endif /* #ifdef ERTS_SMP */ #endif /* #ifndef ETHR_UNUSABLE_SIGUSRX */ - sys_sigset(SIGQUIT, do_quit); + sys_signal(SIGQUIT, do_quit); } int sys_max_files(void) @@ -978,8 +986,13 @@ static void block_signals(void) sys_sigblock(SIGINT); #ifndef ETHR_UNUSABLE_SIGUSRX sys_sigblock(SIGUSR1); +#endif /* #ifndef ETHR_UNUSABLE_SIGUSRX */ +#endif /* #ifndef ERTS_SMP */ + +#if defined(ERTS_SMP) && !defined(ETHR_UNUSABLE_SIGUSRX) + sys_sigblock(ERTS_SYS_SUSPEND_SIGNAL); #endif -#endif + } static void unblock_signals(void) @@ -993,7 +1006,12 @@ static void unblock_signals(void) #ifndef ETHR_UNUSABLE_SIGUSRX sys_sigrelease(SIGUSR1); #endif /* #ifndef ETHR_UNUSABLE_SIGUSRX */ +#endif /* #ifndef ERTS_SMP */ + +#if defined(ERTS_SMP) && !defined(ETHR_UNUSABLE_SIGUSRX) + sys_sigrelease(ERTS_SYS_SUSPEND_SIGNAL); #endif + } /************************** OS info *******************************/ @@ -1101,11 +1119,16 @@ void fini_getenv_state(GETENV_STATE *state) /* Driver interfaces */ static ErlDrvData spawn_start(ErlDrvPort, char*, SysDriverOpts*); static ErlDrvData fd_start(ErlDrvPort, char*, SysDriverOpts*); +#if FDBLOCK +static void fd_async(void *); +static void fd_ready_async(ErlDrvData drv_data, ErlDrvThreadData thread_data); +#endif static ErlDrvSSizeT fd_control(ErlDrvData, unsigned int, char *, ErlDrvSizeT, char **, ErlDrvSizeT); static ErlDrvData vanilla_start(ErlDrvPort, char*, SysDriverOpts*); static int spawn_init(void); static void fd_stop(ErlDrvData); +static void fd_flush(ErlDrvData); static void stop(ErlDrvData); static void ready_input(ErlDrvData, ErlDrvEvent); static void ready_output(ErlDrvData, ErlDrvEvent); @@ -1150,8 +1173,12 @@ struct erl_drv_entry fd_driver_entry = { fd_control, NULL, outputv, - NULL, /* ready_async */ - NULL, /* flush */ +#if FDBLOCK + fd_ready_async, /* ready_async */ +#else + NULL, +#endif + fd_flush, /* flush */ NULL, /* call */ NULL, /* event */ ERL_DRV_EXTENDED_MARKER, @@ -1205,13 +1232,28 @@ static RETSIGTYPE onchld(int signum) #endif } +static int set_blocking_data(struct driver_data *dd) { + + dd->blocking = erts_alloc(ERTS_ALC_T_SYS_BLOCKING, sizeof(ErtsSysBlocking)); + + erts_smp_atomic_add_nob(&sys_misc_mem_sz, sizeof(ErtsSysBlocking)); + + dd->blocking->pdl = driver_pdl_create(dd->port_num); + dd->blocking->res = 0; + dd->blocking->err = 0; + dd->blocking->pkey = driver_async_port_key(dd->port_num); + + return 1; +} + static int set_driver_data(ErlDrvPort port_num, int ifd, int ofd, int packet_bytes, int read_write, int exit_status, - int pid) + int pid, + int is_blocking) { Port *prt; ErtsSysReportExit *report_exit; @@ -1243,8 +1285,13 @@ static int set_driver_data(ErlDrvPort port_num, driver_data[ifd].pid = pid; driver_data[ifd].alive = 1; driver_data[ifd].status = 0; + driver_data[ifd].terminating = 0; + driver_data[ifd].blocking = NULL; if (read_write & DO_WRITE) { driver_data[ifd].ofd = ofd; + if (is_blocking && FDBLOCK) + if (!set_blocking_data(driver_data+ifd)) + return -1; if (ifd != ofd) driver_data[ofd] = driver_data[ifd]; /* structure copy */ } else { /* DO_READ only */ @@ -1260,6 +1307,11 @@ static int set_driver_data(ErlDrvPort port_num, driver_data[ofd].pid = pid; driver_data[ofd].alive = 1; driver_data[ofd].status = 0; + driver_data[ofd].terminating = 0; + driver_data[ofd].blocking = NULL; + if (is_blocking && FDBLOCK) + if (!set_blocking_data(driver_data+ofd)) + return -1; return(ofd); } } @@ -1269,11 +1321,13 @@ static int spawn_init() int i; #if CHLDWTHR erts_thr_opts_t thr_opts = ERTS_THR_OPTS_DEFAULT_INITER; + thr_opts.detached = 0; thr_opts.suggested_stack_size = 0; /* Smallest possible */ + thr_opts.name = "child_waiter"; #endif - sys_sigset(SIGPIPE, SIG_IGN); /* Ignore - we'll handle the write failure */ + sys_signal(SIGPIPE, SIG_IGN); /* Ignore - we'll handle the write failure */ driver_data = (struct driver_data *) erts_alloc(ERTS_ALC_T_DRV_TAB, max_files * sizeof(struct driver_data)); erts_smp_atomic_add_nob(&sys_misc_mem_sz, @@ -1286,7 +1340,7 @@ static int spawn_init() sys_sigblock(SIGCHLD); #endif - sys_sigset(SIGCHLD, onchld); /* Reap children */ + sys_signal(SIGCHLD, onchld); /* Reap children */ #if CHLDWTHR erts_thr_create(&child_waiter_tid, child_waiter, NULL, &thr_opts); @@ -1567,9 +1621,13 @@ static ErlDrvData spawn_start(ErlDrvPort port_num, char* name, SysDriverOpts* op goto child_error; } +#if defined(HAVE_CLOSEFROM) + closefrom(opts->use_stdio ? 3 : 5); +#else for (i = opts->use_stdio ? 3 : 5; i < max_files; i++) (void) close(i); - +#endif + if (opts->wd && chdir(opts->wd) < 0) goto child_error; @@ -1748,7 +1806,7 @@ static ErlDrvData spawn_start(ErlDrvPort port_num, char* name, SysDriverOpts* op } res = set_driver_data(port_num, ifd[0], ofd[1], opts->packet_bytes, - opts->read_write, opts->exit_status, pid); + opts->read_write, opts->exit_status, pid, 0); /* Don't unblock SIGCHLD until now, since the call above must first complete putting away the info about our new subprocess. */ unblock_signals(); @@ -1833,6 +1891,7 @@ static ErlDrvData fd_start(ErlDrvPort port_num, char* name, SysDriverOpts* opts) { ErlDrvData res; + int non_blocking = 0; if (((opts->read_write & DO_READ) && opts->ifd >= max_files) || ((opts->read_write & DO_WRITE) && opts->ofd >= max_files)) @@ -1905,6 +1964,20 @@ static ErlDrvData fd_start(ErlDrvPort port_num, char* name, * case - it can be called with any old pre-existing file descriptors, * the relations between which (if they're even two) we can only guess * at - still, we try our best... + * + * Added note OTP 18: Some systems seem to use stdout/stderr to log data + * using unix pipes, so we cannot allow the system to block on a write. + * Therefore we use an async thread to write the data to fd's that could + * not be set to non-blocking. When no async threads are available we + * fall back on the old behaviour. + * + * Also the guarantee about what is delivered to the OS has changed. + * Pre 18 the fd driver did no flushing of data before terminating. + * Now it does. This is because we want to be able to guarantee that things + * such as escripts and friends really have outputted all data before + * terminating. This could potentially block the termination of the system + * for a very long time, but if the user wants to terminate fast she should + * use erlang:halt with flush=false. */ if (opts->read_write & DO_READ) { @@ -1927,6 +2000,7 @@ static ErlDrvData fd_start(ErlDrvPort port_num, char* name, imagine a scenario where setting non-blocking mode here would cause problems - go ahead and do it. */ + non_blocking = 1; SET_NONBLOCKING(opts->ofd); } else { /* output fd is a tty, input fd isn't */ @@ -1969,6 +2043,7 @@ static ErlDrvData fd_start(ErlDrvPort port_num, char* name, (nfd = open(tty, O_WRONLY)) != -1) { dup2(nfd, opts->ofd); close(nfd); + non_blocking = 1; SET_NONBLOCKING(opts->ofd); } } @@ -1977,8 +2052,9 @@ static ErlDrvData fd_start(ErlDrvPort port_num, char* name, } CHLD_STAT_LOCK; res = (ErlDrvData)(long)set_driver_data(port_num, opts->ifd, opts->ofd, - opts->packet_bytes, - opts->read_write, 0, -1); + opts->packet_bytes, + opts->read_write, 0, -1, + !non_blocking); CHLD_STAT_UNLOCK; return res; } @@ -2004,14 +2080,30 @@ static void nbio_stop_fd(ErlDrvPort prt, int fd) SET_BLOCKING(fd); } -static void fd_stop(ErlDrvData fd) /* Does not close the fds */ +static void fd_stop(ErlDrvData ev) /* Does not close the fds */ { int ofd; + int fd = (int)(long)ev; + ErlDrvPort prt = driver_data[fd].port_num; - nbio_stop_fd(driver_data[(int)(long)fd].port_num, (int)(long)fd); - ofd = driver_data[(int)(long)fd].ofd; - if (ofd != (int)(long)fd && ofd != -1) - nbio_stop_fd(driver_data[(int)(long)fd].port_num, (int)(long)ofd); +#if FDBLOCK + if (driver_data[fd].blocking) { + erts_free(ERTS_ALC_T_SYS_BLOCKING,driver_data[fd].blocking); + driver_data[fd].blocking = NULL; + erts_smp_atomic_add_nob(&sys_misc_mem_sz, -1*sizeof(ErtsSysBlocking)); + } +#endif + + nbio_stop_fd(prt, fd); + ofd = driver_data[fd].ofd; + if (ofd != fd && ofd != -1) + nbio_stop_fd(prt, ofd); +} + +static void fd_flush(ErlDrvData fd) +{ + if (!driver_data[(int)(long)fd].terminating) + driver_data[(int)(long)fd].terminating = 1; } static ErlDrvData vanilla_start(ErlDrvPort port_num, char* name, @@ -2034,8 +2126,8 @@ static ErlDrvData vanilla_start(ErlDrvPort port_num, char* name, CHLD_STAT_LOCK; res = (ErlDrvData)(long)set_driver_data(port_num, fd, fd, - opts->packet_bytes, - opts->read_write, 0, -1); + opts->packet_bytes, + opts->read_write, 0, -1, 0); CHLD_STAT_UNLOCK; return res; } @@ -2072,6 +2164,7 @@ static void stop(ErlDrvData fd) } } +/* used by fd_driver */ static void outputv(ErlDrvData e, ErlIOVec* ev) { int fd = (int)(long)e; @@ -2097,12 +2190,21 @@ static void outputv(ErlDrvData e, ErlIOVec* ev) ev->iov[0].iov_base = lbp; ev->iov[0].iov_len = pb; ev->size += pb; + + if (driver_data[fd].blocking && FDBLOCK) + driver_pdl_lock(driver_data[fd].blocking->pdl); + if ((sz = driver_sizeq(ix)) > 0) { driver_enqv(ix, ev, 0); + + if (driver_data[fd].blocking && FDBLOCK) + driver_pdl_unlock(driver_data[fd].blocking->pdl); + if (sz + ev->size >= (1 << 13)) set_busy_port(ix, 1); } - else { + else if (!driver_data[fd].blocking || !FDBLOCK) { + /* We try to write directly if the fd in non-blocking */ int vsize = ev->vsize > MAX_VSIZE ? MAX_VSIZE : ev->vsize; n = writev(ofd, (const void *) (ev->iov), vsize); @@ -2118,10 +2220,22 @@ static void outputv(ErlDrvData e, ErlIOVec* ev) driver_enqv(ix, ev, n); /* n is the skip value */ driver_select(ix, ofd, ERL_DRV_WRITE|ERL_DRV_USE, 1); } +#if FDBLOCK + else { + if (ev->size != 0) { + driver_enqv(ix, ev, 0); + driver_pdl_unlock(driver_data[fd].blocking->pdl); + driver_async(ix, &driver_data[fd].blocking->pkey, + fd_async, driver_data+fd, NULL); + } else { + driver_pdl_unlock(driver_data[fd].blocking->pdl); + } + } +#endif /* return 0;*/ } - +/* Used by spawn_driver and vanilla driver */ static void output(ErlDrvData e, char* buf, ErlDrvSizeT len) { int fd = (int)(long)e; @@ -2184,6 +2298,23 @@ static int port_inp_failure(ErlDrvPort port_num, int ready_fd, int res) ASSERT(res <= 0); (void) driver_select(port_num, ready_fd, ERL_DRV_READ|ERL_DRV_WRITE, 0); clear_fd_data(ready_fd); + + if (driver_data[ready_fd].blocking && FDBLOCK) { + driver_pdl_lock(driver_data[ready_fd].blocking->pdl); + if (driver_sizeq(driver_data[ready_fd].port_num) > 0) { + driver_pdl_unlock(driver_data[ready_fd].blocking->pdl); + /* We have stuff in the output queue, so we just + set the state to terminating and wait for fd_async_ready + to terminate the port */ + if (res == 0) + driver_data[ready_fd].terminating = 2; + else + driver_data[ready_fd].terminating = -err; + return 0; + } + driver_pdl_unlock(driver_data[ready_fd].blocking->pdl); + } + if (res == 0) { if (driver_data[ready_fd].report_exit) { CHLD_STAT_LOCK; @@ -2234,6 +2365,7 @@ static void ready_input(ErlDrvData e, ErlDrvEvent ready_fd) port_num = driver_data[fd].port_num; packet_bytes = driver_data[fd].packet_bytes; + if (packet_bytes == 0) { byte *read_buf = (byte *) erts_alloc(ERTS_ALC_T_SYS_READ_BUF, ERTS_SYS_READ_BUF_SZ); @@ -2357,6 +2489,8 @@ static void ready_output(ErlDrvData e, ErlDrvEvent ready_fd) if ((iv = (struct iovec*) driver_peekq(ix, &vsize)) == NULL) { driver_select(ix, ready_fd, ERL_DRV_WRITE, 0); + if (driver_data[fd].terminating) + driver_failure_atom(driver_data[fd].port_num,"normal"); return; /* 0; */ } vsize = vsize > MAX_VSIZE ? MAX_VSIZE : vsize; @@ -2382,6 +2516,78 @@ static void stop_select(ErlDrvEvent fd, void* _) close((int)fd); } +#if FDBLOCK + +static void +fd_async(void *async_data) +{ + int res; + struct driver_data *dd = (struct driver_data*)async_data; + SysIOVec *iov0; + SysIOVec *iov; + int iovlen; + int err; + /* much of this code is stolen from efile_drv:invoke_writev */ + driver_pdl_lock(dd->blocking->pdl); + iov0 = driver_peekq(dd->port_num, &iovlen); + iovlen = iovlen < MAXIOV ? iovlen : MAXIOV; + iov = erts_alloc_fnf(ERTS_ALC_T_SYS_WRITE_BUF, + sizeof(SysIOVec)*iovlen); + if (!iov) { + res = -1; + err = ENOMEM; + driver_pdl_unlock(dd->blocking->pdl); + } else { + memcpy(iov,iov0,iovlen*sizeof(SysIOVec)); + driver_pdl_unlock(dd->blocking->pdl); + + res = writev(dd->ofd, iov, iovlen); + err = errno; + + erts_free(ERTS_ALC_T_SYS_WRITE_BUF, iov); + } + dd->blocking->res = res; + dd->blocking->err = err; +} + +void fd_ready_async(ErlDrvData drv_data, + ErlDrvThreadData thread_data) { + struct driver_data *dd = (struct driver_data *)thread_data; + ErlDrvPort port_num = dd->port_num; + + ASSERT(dd->blocking); + ASSERT(dd == (driver_data + (int)(long)drv_data)); + + if (dd->blocking->res > 0) { + driver_pdl_lock(dd->blocking->pdl); + if (driver_deq(port_num, dd->blocking->res) == 0) { + driver_pdl_unlock(dd->blocking->pdl); + set_busy_port(port_num, 0); + if (dd->terminating) { + /* The port is has been ordered to terminate + from either fd_flush or port_inp_failure */ + if (dd->terminating == 1) + driver_failure_atom(port_num, "normal"); + else if (dd->terminating == 2) + driver_failure_eof(port_num); + else if (dd->terminating < 0) + driver_failure_posix(port_num, -dd->terminating); + return; /* -1; */ + } + } else { + driver_pdl_unlock(dd->blocking->pdl); + /* still data left to write in queue */ + driver_async(port_num, &dd->blocking->pkey, fd_async, dd, NULL); + return /* 0; */; + } + } else if (dd->blocking->res < 0) { + driver_failure_posix(port_num, dd->blocking->err); + return; /* -1; */ + } + return; /* 0; */ +} + +#endif void erts_do_break_handling(void) { @@ -2651,18 +2857,30 @@ void sys_preload_end(Preload* p) /* Nothing */ } -/* Read a key from console (?) */ - +/* Read a key from console, used by break.c + Here we assume that all schedulers are stopped so that erl_poll + does not interfere with the select below. +*/ int sys_get_key(fd) int fd; { - int c; + int c, ret; unsigned char rbuf[64]; + fd_set fds; fflush(stdout); /* Flush query ??? */ - if ((c = read(fd,rbuf,64)) <= 0) { - return c; + FD_ZERO(&fds); + FD_SET(fd,&fds); + + ret = select(fd+1, &fds, NULL, NULL, NULL); + + if (ret == 1) { + do { + c = read(fd,rbuf,64); + } while (c < 0 && errno == EAGAIN); + if (c <= 0) + return c; } return rbuf[0]; @@ -2986,13 +3204,6 @@ signal_dispatcher_thread_func(void *unused) case '1': /* SIGUSR1 */ sigusr1_exit(); break; -#ifdef QUANTIFY - case '2': /* SIGUSR2 */ - quantify_save_data(); /* Might take a substantial amount of - time, but this is a test/debug - build */ - break; -#endif default: erl_exit(ERTS_ABORT_EXIT, "signal-dispatcher thread received unknown " @@ -3010,6 +3221,7 @@ init_smp_sig_notify(void) { erts_smp_thr_opts_t thr_opts = ERTS_SMP_THR_OPTS_DEFAULT_INITER; thr_opts.detached = 1; + thr_opts.name = "sys_sig_dispatcher"; if (pipe(sig_notify_fds) < 0) { erl_exit(ERTS_ABORT_EXIT, @@ -3024,6 +3236,17 @@ init_smp_sig_notify(void) NULL, &thr_opts); } + +static void +init_smp_sig_suspend(void) { + if (pipe(sig_suspend_fds) < 0) { + erl_exit(ERTS_ABORT_EXIT, + "Failed to create sig_suspend pipe: %s (%d)\n", + erl_errno_id(errno), + errno); + } +} + #ifdef __DARWIN__ int erts_darwin_main_thread_pipe[2]; @@ -3051,9 +3274,11 @@ erts_sys_main_thread(void) #endif smp_sig_notify(0); /* Notify initialized */ - while (1) { - /* Wait for a signal to arrive... */ + + /* Wait for a signal to arrive... */ + #ifdef __DARWIN__ + while (1) { /* * The wx driver needs to be able to steal the main thread for Cocoa to * work properly. @@ -3068,12 +3293,24 @@ erts_sys_main_thread(void) void* (*func)(void*); void* arg; void *resp; - read(erts_darwin_main_thread_pipe[0],&func,sizeof(void* (*)(void*))); - read(erts_darwin_main_thread_pipe[0],&arg, sizeof(void*)); + res = read(erts_darwin_main_thread_pipe[0],&func,sizeof(void* (*)(void*))); + if (res != sizeof(void* (*)(void*))) + break; + res = read(erts_darwin_main_thread_pipe[0],&arg,sizeof(void*)); + if (res != sizeof(void*)) + break; resp = (*func)(arg); write(erts_darwin_main_thread_result_pipe[1],&resp,sizeof(void *)); } -#else + + if (res == -1 && errno != EINTR) + break; + } + /* Something broke with the main thread pipe, so we ignore it for now. + Most probably erts has closed this pipe and is about to exit. */ +#endif /* #ifdef __DARWIN__ */ + + while (1) { #ifdef DEBUG int res = #else @@ -3082,7 +3319,6 @@ erts_sys_main_thread(void) select(0, NULL, NULL, NULL, NULL); ASSERT(res < 0); ASSERT(errno == EINTR); -#endif } } @@ -3174,6 +3410,7 @@ erl_sys_args(int* argc, char** argv) #ifdef ERTS_SMP init_smp_sig_notify(); + init_smp_sig_suspend(); #endif /* Handled arguments have been marked with NULL. Slide arguments diff --git a/erts/emulator/sys/unix/sys_float.c b/erts/emulator/sys/unix/sys_float.c index cafeab547e..2ffa649767 100644 --- a/erts/emulator/sys/unix/sys_float.c +++ b/erts/emulator/sys/unix/sys_float.c @@ -32,7 +32,7 @@ void erts_sys_init_float(void) { # ifdef SIGFPE - sys_sigset(SIGFPE, SIG_IGN); /* Ignore so we can test for NaN and Inf */ + sys_signal(SIGFPE, SIG_IGN); /* Ignore so we can test for NaN and Inf */ # endif } @@ -667,7 +667,7 @@ static void fpe_sig_handler(int sig) static void erts_thread_catch_fp_exceptions(void) { - sys_sigset(SIGFPE, fpe_sig_handler); + sys_signal(SIGFPE, fpe_sig_handler); unmask_fpe(); } diff --git a/erts/emulator/sys/win32/erl_win_sys.h b/erts/emulator/sys/win32/erl_win_sys.h index f04bb6a0e5..33aa88ab5f 100644 --- a/erts/emulator/sys/win32/erl_win_sys.h +++ b/erts/emulator/sys/win32/erl_win_sys.h @@ -273,4 +273,16 @@ typedef long ssize_t; int init_async(int); int exit_async(void); #endif + +#define ERTS_HAVE_TRY_CATCH 1 + +#define ERTS_SYS_TRY_CATCH(EXPR,CATCH) \ + __try { \ + EXPR; \ + } \ + __except(GetExceptionCode() == EXCEPTION_ACCESS_VIOLATION ? EXCEPTION_EXECUTE_HANDLER : EXCEPTION_CONTINUE_SEARCH) \ + { \ + CATCH; \ + } + #endif diff --git a/erts/emulator/sys/win32/sys.c b/erts/emulator/sys/win32/sys.c index ad636b7249..cf587af4ac 100644 --- a/erts/emulator/sys/win32/sys.c +++ b/erts/emulator/sys/win32/sys.c @@ -247,6 +247,27 @@ void erl_sys_args(int* argc, char** argv) #endif } +/* + * Function returns 1 if we can read from all values in between + * start and stop. + */ +int +erts_sys_is_area_readable(char *start, char *stop) { + volatile char tmp; + __try + { + while(start < stop) { + tmp = *start; + start++; + } + } + __except(EXCEPTION_EXECUTE_HANDLER) + { + return 0; + } + return 1; +} + int erts_sys_prepare_crash_dump(int secs) { Port *heart_port; @@ -1392,39 +1413,46 @@ int parse_command(wchar_t* cmd){ return i; } -static BOOL need_quotes(wchar_t *str) -{ - int in_quote = 0; - int backslashed = 0; - int naked_space = 0; - while (*str != L'\0') { - switch (*str) { - case L'\\' : - backslashed = !backslashed; - break; - case L'"': - if (backslashed) { - backslashed=0; - } else { - in_quote = !in_quote; - } - break; - case L' ': - backslashed = 0; - if (!(backslashed || in_quote)) { - naked_space++; - } - break; - default: - backslashed = 0; +/* + * Translating of command line arguments to correct format. In the examples + * below the '' are not part of the actual string. + * 'io:format("hello").' -> 'io:format(\"hello\").' + * 'io:format("is anybody in there?").' -> '"io:format(\"is anybody in there?\")."' + * 'Just nod if you can hear me.' -> '"Just nod if you can hear me."' + * 'Is there ""anyone at home?' -> '"Is there \"\"anyone at home?"' + * 'Relax."' -> 'Relax.\"' + * + * If new == NULL we just calculate the length. + * + * The reason for having to quote all of the is becasue CreateProcessW removes + * one level of escaping since it takes a single long command line rather + * than the argument chunks that unix uses. + */ +static int escape_and_quote(wchar_t *str, wchar_t *new, BOOL *quoted) { + int i, j = 0; + if (new == NULL) + *quoted = FALSE; + else if (*quoted) + new[j++] = L'"'; + for ( i = 0; str[i] != L'\0'; i++,j++) { + if (str[i] == L' ' && new == NULL && *quoted == FALSE) { + *quoted = TRUE; + j++; } - ++str; + /* check if we have to escape quotes */ + if (str[i] == L'"') { + if (new) new[j] = L'\\'; + j++; + } + if (new) new[j] = str[i]; + } + if (*quoted) { + if (new) new[j] = L'"'; + j++; } - return (naked_space > 0); + return j; } - - /* *---------------------------------------------------------------------- @@ -1585,31 +1613,24 @@ create_child_process wcscpy(appname, execPath); } if (argv == NULL) { - BOOL orig_need_q = need_quotes(execPath); + BOOL orig_need_q; wchar_t *ptr; - int ocl = wcslen(execPath); + int ocl = escape_and_quote(execPath, NULL, &orig_need_q); if (run_cmd) { newcmdline = (wchar_t *) erts_alloc(ERTS_ALC_T_TMP, - (ocl + ((orig_need_q) ? 3 : 1) - + 11)*sizeof(wchar_t)); + (ocl + 1 + 11)*sizeof(wchar_t)); memcpy(newcmdline,L"cmd.exe /c ",11*sizeof(wchar_t)); ptr = newcmdline + 11; } else { newcmdline = (wchar_t *) erts_alloc(ERTS_ALC_T_TMP, - (ocl + ((orig_need_q) ? 3 : 1))*sizeof(wchar_t)); + (ocl + 1)*sizeof(wchar_t)); ptr = (wchar_t *) newcmdline; } - if (orig_need_q) { - *ptr++ = L'"'; - } - memcpy(ptr,execPath,ocl*sizeof(wchar_t)); - ptr += ocl; - if (orig_need_q) { - *ptr++ = L'"'; - } - *ptr = L'\0'; + ptr += escape_and_quote(execPath, ptr, &orig_need_q); + ptr[0] = L'\0'; } else { - int sum = 1; /* '\0' */ + int sum = 0; + BOOL *qte = NULL; wchar_t **ar = argv; wchar_t *n; wchar_t *save_arg0 = NULL; @@ -1620,11 +1641,13 @@ create_child_process if (run_cmd) { sum += 11; /* cmd.exe /c */ } + + while (*ar != NULL) ar++; + qte = erts_alloc(ERTS_ALC_T_TMP, (ar - argv)*sizeof(BOOL)); + + ar = argv; while (*ar != NULL) { - sum += wcslen(*ar); - if (need_quotes(*ar)) { - sum += 2; /* quotes */ - } + sum += escape_and_quote(*ar,NULL,qte+(ar - argv)); sum++; /* space */ ++ar; } @@ -1636,26 +1659,18 @@ create_child_process n += 11; } while (*ar != NULL) { - int q = need_quotes(*ar); - sum = wcslen(*ar); - if (q) { - *n++ = L'"'; - } - memcpy(n,*ar,sum*sizeof(wchar_t)); - n += sum; - if (q) { - *n++ = L'"'; - } + n += escape_and_quote(*ar,n,qte+(ar - argv)); *n++ = L' '; ++ar; } - *(n-1) = L'\0'; + *(n-1) = L'\0'; /* overwrite last space with '\0' */ if (save_arg0 != NULL) { argv[0] = save_arg0; } + erts_free(ERTS_ALC_T_TMP, qte); } - DEBUGF(("Creating child process: %s, createFlags = %d\n", newcmdline, createFlags)); + DEBUGF((stderr,"Creating child process: %S, createFlags = %d\n", newcmdline, createFlags)); ok = CreateProcessW((wchar_t *) appname, (wchar_t *) newcmdline, NULL, @@ -2190,7 +2205,7 @@ static void fd_stop(ErlDrvData data) ASSERT(dp->out.flushEvent); SetEvent(dp->out.flushEvent); } while (WaitForSingleObject(dp->out.flushReplyEvent, 10) == WAIT_TIMEOUT - || !(dp->out.flags & DF_THREAD_FLUSHED)); + && !(dp->out.flags & DF_THREAD_FLUSHED)); } } diff --git a/erts/emulator/test/bif_SUITE.erl b/erts/emulator/test/bif_SUITE.erl index fbc229bc53..fc9bdae0a0 100644 --- a/erts/emulator/test/bif_SUITE.erl +++ b/erts/emulator/test/bif_SUITE.erl @@ -20,6 +20,7 @@ -module(bif_SUITE). -include_lib("test_server/include/test_server.hrl"). +-include_lib("kernel/include/file.hrl"). -export([all/0, suite/0,groups/0,init_per_suite/1, end_per_suite/1, init_per_group/2,end_per_group/2, @@ -681,8 +682,38 @@ erlang_halt(Config) when is_list(Config) -> {badrpc,nodedown} = rpc:call(N2, erlang, halt, [0]), {ok,N3} = slave:start(H, halt_node3), {badrpc,nodedown} = rpc:call(N3, erlang, halt, [0,[]]), - ok. + % This test triggers a segfault when dumping a crash dump + % to make sure that we can handle it properly. + {ok,N4} = slave:start(H, halt_node4), + CrashDump = filename:join(proplists:get_value(priv_dir,Config), + "segfault_erl_crash.dump"), + true = rpc:call(N4, os, putenv, ["ERL_CRASH_DUMP",CrashDump]), + false = rpc:call(N4, erts_debug, set_internal_state, + [available_internal_state, true]), + {badrpc,nodedown} = rpc:call(N4, erts_debug, set_internal_state, + [broken_halt, "Validate correct crash dump"]), + ok = wait_until_stable_size(CrashDump,-1), + {ok, Bin} = file:read_file(CrashDump), + case {string:str(binary_to_list(Bin),"\n=end\n"), + string:str(binary_to_list(Bin),"\r\n=end\r\n")} of + {0,0} -> ct:fail("Could not find end marker in crash dump"); + _ -> ok + end. + +wait_until_stable_size(_File,-10) -> + {error,enoent}; +wait_until_stable_size(File,PrevSz) -> + timer:sleep(250), + case file:read_file_info(File) of + {error,enoent} -> + wait_until_stable_size(File,PrevSz-1); + {ok,#file_info{size = PrevSz }} when PrevSz /= -1 -> + io:format("Crashdump file size was: ~p (~s)~n",[PrevSz,File]), + ok; + {ok,#file_info{size = NewSz }} -> + wait_until_stable_size(File,NewSz) + end. %% Helpers diff --git a/erts/emulator/test/big_SUITE.erl b/erts/emulator/test/big_SUITE.erl index 413bd3bcae..3193d56e2a 100644 --- a/erts/emulator/test/big_SUITE.erl +++ b/erts/emulator/test/big_SUITE.erl @@ -23,7 +23,7 @@ init_per_group/2,end_per_group/2]). -export([t_div/1, eq_28/1, eq_32/1, eq_big/1, eq_math/1, big_literals/1, borders/1, negative/1, big_float_1/1, big_float_2/1, - shift_limit_1/1, powmod/1, system_limit/1, otp_6692/1]). + shift_limit_1/1, powmod/1, system_limit/1, toobig/1, otp_6692/1]). %% Internal exports. -export([eval/1]). @@ -40,7 +40,7 @@ suite() -> [{ct_hooks,[ts_install_cth]}]. all() -> [t_div, eq_28, eq_32, eq_big, eq_math, big_literals, borders, negative, {group, big_float}, shift_limit_1, - powmod, system_limit, otp_6692]. + powmod, system_limit, toobig, otp_6692]. groups() -> [{big_float, [], [big_float_1, big_float_2]}]. @@ -370,6 +370,16 @@ maxbig() -> id(I) -> I. +toobig(Config) when is_list(Config) -> + ?line {'EXIT',{{badmatch,_},_}} = (catch toobig()), + ok. + +toobig() -> + A = erlang:term_to_binary(lists:seq(1000000, 2200000)), + ASize = erlang:bit_size(A), + <<ANr:ASize>> = A, % should fail + ANr band ANr. + otp_6692(suite) -> []; otp_6692(doc) -> diff --git a/erts/emulator/test/map_SUITE.erl b/erts/emulator/test/map_SUITE.erl index 888ed8e272..1da08beb8b 100644 --- a/erts/emulator/test/map_SUITE.erl +++ b/erts/emulator/test/map_SUITE.erl @@ -30,7 +30,7 @@ t_list_comprehension/1, t_map_sort_literals/1, t_map_equal/1, - %t_size/1, + t_map_compare/1, t_map_size/1, %% Specific Map BIFs @@ -51,6 +51,10 @@ t_erlang_hash/1, t_map_encode_decode/1, + %% non specific BIF related + t_bif_build_and_check/1, + t_bif_merge_and_check/1, + %% maps module not bifs t_maps_fold/1, t_maps_map/1, @@ -58,6 +62,7 @@ t_maps_without/1, %% misc + t_hashmap_balance/1, t_pdict/1, t_ets/1, t_dets/1, @@ -66,6 +71,13 @@ -include_lib("stdlib/include/ms_transform.hrl"). +-define(CHECK(Cond,Term), + case (catch (Cond)) of + true -> true; + _ -> io:format("###### CHECK FAILED ######~nINPUT: ~p~n", [Term]), + exit(Term) + end). + suite() -> []. all() -> [ @@ -75,7 +87,7 @@ all() -> [ t_update_assoc,t_update_exact, t_guard_bifs, t_guard_sequence, t_guard_update, t_guard_receive,t_guard_fun, t_list_comprehension, - t_map_equal, + t_map_equal, t_map_compare, t_map_sort_literals, %% Specific Map BIFs @@ -90,12 +102,17 @@ all() -> [ t_erlang_hash, t_map_encode_decode, t_map_size, + %% non specific BIF related + t_bif_build_and_check, + t_bif_merge_and_check, + %% maps module t_maps_fold, t_maps_map, t_maps_size, t_maps_without, %% Other functions + t_hashmap_balance, t_pdict, t_ets, t_tracing @@ -146,17 +163,6 @@ t_build_and_match_literals(Config) when is_list(Config) -> ok. -%% Tests size(Map). -%% not implemented, perhaps it shouldn't be either - -%t_size(Config) when is_list(Config) -> -% 0 = size(#{}), -% 1 = size(#{a=>1}), -% 1 = size(#{a=>#{a=>1}}), -% 2 = size(#{a=>1, b=>2}), -% 3 = size(#{a=>1, b=>2, b=>"3"}), -% ok. - t_map_size(Config) when is_list(Config) -> 0 = map_size(id(#{})), 1 = map_size(id(#{a=>1})), @@ -172,12 +178,23 @@ t_map_size(Config) when is_list(Config) -> true = map_is_size(M#{ "a" => 2}, 2), false = map_is_size(M#{ "c" => 2}, 2), + Ks = [build_key(fun(K) -> <<1,K:32,1>> end,I)||I<-lists:seq(1,100)], + ok = build_and_check_size(Ks,0,#{}), + %% Error cases. {'EXIT',{badarg,_}} = (catch map_size([])), {'EXIT',{badarg,_}} = (catch map_size(<<1,2,3>>)), {'EXIT',{badarg,_}} = (catch map_size(1)), ok. +build_and_check_size([K|Ks],N,M0) -> + N = map_size(M0), + M1 = M0#{ K => K }, + build_and_check_size(Ks,N + 1,M1); +build_and_check_size([],N,M) -> + N = map_size(M), + ok. + map_is_size(M,N) when map_size(M) =:= N -> true; map_is_size(_,_) -> false. @@ -432,7 +449,7 @@ t_map_sort_literals(Config) when is_list(Config) -> true = #{ c => 1, b => 1, a => 1 } < id(#{ b => 1, c => 1, d => 1}), true = #{ "a" => 1 } < id(#{ <<"a">> => 1}), false = #{ <<"a">> => 1 } < id(#{ "a" => 1}), - false = #{ 1 => 1 } < id(#{ 1.0 => 1}), + true = #{ 1 => 1 } < id(#{ 1.0 => 1}), false = #{ 1.0 => 1 } < id(#{ 1 => 1}), %% value order @@ -440,16 +457,47 @@ t_map_sort_literals(Config) when is_list(Config) -> false = #{ a => 2 } < id(#{ a => 1}), false = #{ a => 2, b => 1 } < id(#{ a => 1, b => 3}), true = #{ a => 1, b => 1 } < id(#{ a => 1, b => 3}), + false = #{ a => 1 } < id(#{ a => 1.0}), + false = #{ a => 1.0 } < id(#{ a => 1}), true = #{ "a" => "hi", b => 134 } == id(#{ b => 134,"a" => "hi"}), + %% large maps + + M = maps:from_list([{I,I}||I <- lists:seq(1,500)]), + + %% size order + true = M#{ a => 1, b => 2} < id(M#{ a => 1, b => 1, c => 1}), + true = M#{ b => 1, a => 1} < id(M#{ c => 1, a => 1, b => 1}), + false = M#{ c => 1, b => 1, a => 1} < id(M#{ c => 1, a => 1}), + + %% key order + true = M#{ a => 1 } < id(M#{ b => 1}), + false = M#{ b => 1 } < id(M#{ a => 1}), + true = M#{ a => 1, b => 1, c => 1 } < id(M#{ b => 1, c => 1, d => 1}), + true = M#{ b => 1, c => 1, d => 1 } > id(M#{ a => 1, b => 1, c => 1}), + true = M#{ c => 1, b => 1, a => 1 } < id(M#{ b => 1, c => 1, d => 1}), + true = M#{ "a" => 1 } < id(M#{ <<"a">> => 1}), + false = M#{ <<"a">> => 1 } < id(#{ "a" => 1}), + true = M#{ 1 => 1 } < id(maps:remove(1,M#{ 1.0 => 1})), + false = M#{ 1.0 => 1 } < id(M#{ 1 => 1}), + + %% value order + true = M#{ a => 1 } < id(M#{ a => 2}), + false = M#{ a => 2 } < id(M#{ a => 1}), + false = M#{ a => 2, b => 1 } < id(M#{ a => 1, b => 3}), + true = M#{ a => 1, b => 1 } < id(M#{ a => 1, b => 3}), + false = M#{ a => 1 } < id(M#{ a => 1.0}), + false = M#{ a => 1.0 } < id(M#{ a => 1}), + + true = M#{ "a" => "hi", b => 134 } == id(M#{ b => 134,"a" => "hi"}), + %% lists:sort SortVs = [#{"a"=>1},#{a=>2},#{1=>3},#{<<"a">>=>4}], [#{1:=ok},#{a:=ok},#{"a":=ok},#{<<"a">>:=ok}] = lists:sort([#{"a"=>ok},#{a=>ok},#{1=>ok},#{<<"a">>=>ok}]), [#{1:=3},#{a:=2},#{"a":=1},#{<<"a">>:=4}] = lists:sort(SortVs), [#{1:=3},#{a:=2},#{"a":=1},#{<<"a">>:=4}] = lists:sort(lists:reverse(SortVs)), - ok. t_map_equal(Config) when is_list(Config) -> @@ -469,27 +517,285 @@ t_map_equal(Config) when is_list(Config) -> true = id(#{ a => 1, b => 3, c => <<"wat">> }) =:= id(#{ a => 1, b => 3, c=><<"wat">>}), ok. + +t_map_compare(Config) when is_list(Config) -> + Seed = erlang:now(), + io:format("seed = ~p\n", [Seed]), + random:seed(Seed), + repeat(100, fun(_) -> float_int_compare() end, []), + repeat(100, fun(_) -> recursive_compare() end, []), + ok. + +float_int_compare() -> + Terms = numeric_keys(3), + %%io:format("Keys to use: ~p\n", [Terms]), + Pairs = lists:map(fun(K) -> list_to_tuple([{K,V} || V <- Terms]) end, Terms), + lists:foreach(fun(Size) -> + MapGen = fun() -> map_gen(list_to_tuple(Pairs), Size) end, + repeat(100, fun do_compare/1, [MapGen, MapGen]) + end, + lists:seq(1,length(Terms))), + ok. + +numeric_keys(N) -> + lists:foldl(fun(_,Acc) -> + Int = random:uniform(N*4) - N*2, + Float = float(Int), + [Int, Float, Float * 0.99, Float * 1.01 | Acc] + end, + [], + lists:seq(1,N)). + + +repeat(0, _, _) -> + ok; +repeat(N, Fun, Arg) -> + Fun(Arg), + repeat(N-1, Fun, Arg). + +copy_term(T) -> + Papa = self(), + P = spawn_link(fun() -> receive Msg -> Papa ! Msg end end), + P ! T, + receive R -> R end. + +do_compare([Gen1, Gen2]) -> + M1 = Gen1(), + M2 = Gen2(), + %%io:format("Maps to compare: ~p AND ~p\n", [M1, M2]), + C = (M1 < M2), + Erlang = maps_lessthan(M1, M2), + C = Erlang, + ?CHECK(M1==M1, M1), + + %% Change one key from int to float (or vice versa) and check compare + ML1 = maps:to_list(M1), + {K1,V1} = lists:nth(random:uniform(length(ML1)), ML1), + case K1 of + I when is_integer(I) -> + case maps:find(float(I),M1) of + error -> + M1f = maps:remove(I, maps:put(float(I), V1, M1)), + ?CHECK(M1f > M1, [M1f, M1]); + _ -> ok + end; + + F when is_float(F), round(F) == F -> + case maps:find(round(F),M1) of + error -> + M1i = maps:remove(F, maps:put(round(F), V1, M1)), + ?CHECK(M1i < M1, [M1i, M1]); + _ -> ok + end; + + _ -> ok % skip floats with decimals + end, + + ?CHECK(M2 == M2, [M2]). + + +maps_lessthan(M1, M2) -> + case {maps:size(M1),maps:size(M2)} of + {_S,_S} -> + {K1,V1} = lists:unzip(term_sort(maps:to_list(M1))), + {K2,V2} = lists:unzip(term_sort(maps:to_list(M2))), + + case erts_internal:cmp_term(K1,K2) of + -1 -> true; + 0 -> (V1 < V2); + 1 -> false + end; + + {S1, S2} -> + S1 < S2 + end. + +term_sort(L) -> + lists:sort(fun(A,B) -> erts_internal:cmp_term(A,B) =< 0 end, + L). + + +cmp(T1, T2, Exact) when is_tuple(T1) and is_tuple(T2) -> + case {size(T1),size(T2)} of + {_S,_S} -> cmp(tuple_to_list(T1), tuple_to_list(T2), Exact); + {S1,S2} when S1 < S2 -> -1; + {S1,S2} when S1 > S2 -> 1 + end; + +cmp([H1|T1], [H2|T2], Exact) -> + case cmp(H1,H2, Exact) of + 0 -> cmp(T1,T2, Exact); + C -> C + end; + +cmp(M1, M2, Exact) when is_map(M1) andalso is_map(M2) -> + cmp_maps(M1,M2,Exact); +cmp(M1, M2, Exact) -> + cmp_others(M1, M2, Exact). + +cmp_maps(M1, M2, Exact) -> + case {maps:size(M1),maps:size(M2)} of + {_S,_S} -> + {K1,V1} = lists:unzip(term_sort(maps:to_list(M1))), + {K2,V2} = lists:unzip(term_sort(maps:to_list(M2))), + + case cmp(K1, K2, true) of + 0 -> cmp(V1, V2, Exact); + C -> C + end; + + {S1,S2} when S1 < S2 -> -1; + {S1,S2} when S1 > S2 -> 1 + end. + +cmp_others(I, F, true) when is_integer(I), is_float(F) -> + -1; +cmp_others(F, I, true) when is_float(F), is_integer(I) -> + 1; +cmp_others(T1, T2, _) -> + case {T1<T2, T1==T2} of + {true,false} -> -1; + {false,true} -> 0; + {false,false} -> 1 + end. + +map_gen(Pairs, Size) -> + {_,L} = lists:foldl(fun(_, {Keys, Acc}) -> + KI = random:uniform(size(Keys)), + K = element(KI,Keys), + KV = element(random:uniform(size(K)), K), + {erlang:delete_element(KI,Keys), [KV | Acc]} + end, + {Pairs, []}, + lists:seq(1,Size)), + + maps:from_list(L). + + +recursive_compare() -> + Leafs = {atom, 17, 16.9, 17.1, [], self(), spawn(fun() -> ok end), make_ref(), make_ref()}, + {A, B} = term_gen_recursive(Leafs, 0, 0), + %%io:format("Recursive term A = ~p\n", [A]), + %%io:format("Recursive term B = ~p\n", [B]), + + ?CHECK({true,false} =:= case do_cmp(A, B, false) of + -1 -> {A<B, A>=B}; + 0 -> {A==B, A/=B}; + 1 -> {A>B, A=<B} + end, + {A,B}), + A2 = copy_term(A), + ?CHECK(A == A2, {A,A2}), + ?CHECK(0 =:= cmp(A, A2, false), {A,A2}), + + B2 = copy_term(B), + ?CHECK(B == B2, {B,B2}), + ?CHECK(0 =:= cmp(B, B2, false), {B,B2}), + ok. + +do_cmp(A, B, Exact) -> + C = cmp(A, B, Exact), + C. + +%% Generate two terms {A,B} that may only differ +%% at float vs integer types. +term_gen_recursive(Leafs, Flags, Depth) -> + MaxDepth = 10, + Rnd = case {Flags, Depth} of + {_, MaxDepth} -> % Only leafs + random:uniform(size(Leafs)) + 3; + {0, 0} -> % Only containers + random:uniform(3); + {0,_} -> % Anything + random:uniform(size(Leafs)+3) + end, + case Rnd of + 1 -> % Make map + Size = random:uniform(size(Leafs)), + lists:foldl(fun(_, {Acc1,Acc2}) -> + {K1,K2} = term_gen_recursive(Leafs, Flags, + Depth+1), + {V1,V2} = term_gen_recursive(Leafs, Flags, Depth+1), + %%ok = check_keys(K1,K2, 0), + {maps:put(K1,V1, Acc1), maps:put(K2,V2, Acc2)} + end, + {maps:new(), maps:new()}, + lists:seq(1,Size)); + 2 -> % Make cons + {Car1,Car2} = term_gen_recursive(Leafs, Flags, Depth+1), + {Cdr1,Cdr2} = term_gen_recursive(Leafs, Flags, Depth+1), + {[Car1 | Cdr1], [Car2 | Cdr2]}; + 3 -> % Make tuple + Size = random:uniform(size(Leafs)), + L = lists:map(fun(_) -> term_gen_recursive(Leafs, Flags, Depth+1) end, + lists:seq(1,Size)), + {L1, L2} = lists:unzip(L), + {list_to_tuple(L1), list_to_tuple(L2)}; + + N -> % Make leaf + case element(N-3, Leafs) of + I when is_integer(I) -> + case random:uniform(4) of + 1 -> {I, float(I)}; + 2 -> {float(I), I}; + _ -> {I,I} + end; + T -> {T,T} + end + end. + +check_keys(K1, K2, _) when K1 =:= K2 -> + case erlang:phash3(K1) =:= erlang:phash3(K2) of + true -> ok; + false -> + io:format("Same keys with different hash values !!!\nK1 = ~p\nK2 = ~p\n", [K1,K2]), + error + end; +check_keys(K1, K2, 0) -> + case {erlang:phash3(K1), erlang:phash3(K2)} of + {H,H} -> check_keys(K1, K2, 1); + {_,_} -> ok + end; +check_keys(K1, K2, L) when L < 10 -> + case {erlang:phash3([L|K1]), erlang:phash3([L|K2])} of + {H,H} -> check_keys(K1, K2, L+1); + {_,_} -> ok + end; +check_keys(K1, K2, L) -> + io:format("Same hash value at level ~p !!!\nK1 = ~p\nK2 = ~p\n", [L,K1,K2]), + error. + %% BIFs t_bif_map_get(Config) when is_list(Config) -> - + %% small map 1 = maps:get(a, #{ a=> 1}), 2 = maps:get(b, #{ a=> 1, b => 2}), "hi" = maps:get("hello", #{ a=>1, "hello" => "hi"}), "tuple hi" = maps:get({1,1.0}, #{ a=>a, {1,1.0} => "tuple hi"}), - M = id(#{ k1=>"v1", <<"k2">> => <<"v3">> }), - "v4" = maps:get(<<"k2">>, M#{ <<"k2">> => "v4" }), + M0 = id(#{ k1=>"v1", <<"k2">> => <<"v3">> }), + "v4" = maps:get(<<"k2">>, M0#{<<"k2">> => "v4"}), + + %% large map + M1 = maps:from_list([{I,I}||I<-lists:seq(1,100)] ++ + [{a,1},{b,2},{"hello","hi"},{{1,1.0},"tuple hi"}, + {k1,"v1"},{<<"k2">>,"v3"}]), + 1 = maps:get(a, M1), + 2 = maps:get(b, M1), + "hi" = maps:get("hello", M1), + "tuple hi" = maps:get({1,1.0}, M1), + "v3" = maps:get(<<"k2">>, M1), %% error case {'EXIT',{badarg, [{maps,get,_,_}|_]}} = (catch maps:get(a,[])), {'EXIT',{badarg, [{maps,get,_,_}|_]}} = (catch maps:get(a,<<>>)), {'EXIT',{bad_key,[{maps,get,_,_}|_]}} = (catch maps:get({1,1}, #{{1,1.0} => "tuple"})), {'EXIT',{bad_key,[{maps,get,_,_}|_]}} = (catch maps:get(a,#{})), - {'EXIT',{bad_key,[{maps,get,_,_}|_]}} = (catch maps:get(a,#{ b=>1, c=>2})), + {'EXIT',{bad_key,[{maps,get,_,_}|_]}} = (catch maps:get(a,#{b=>1, c=>2})), ok. t_bif_map_find(Config) when is_list(Config) -> - + %% small map {ok, 1} = maps:find(a, #{ a=> 1}), {ok, 2} = maps:find(b, #{ a=> 1, b => 2}), {ok, "int"} = maps:find(1, #{ 1 => "int"}), @@ -498,8 +804,18 @@ t_bif_map_find(Config) when is_list(Config) -> {ok, "hi"} = maps:find("hello", #{ a=>1, "hello" => "hi"}), {ok, "tuple hi"} = maps:find({1,1.0}, #{ a=>a, {1,1.0} => "tuple hi"}), - M = id(#{ k1=>"v1", <<"k2">> => <<"v3">> }), - {ok, "v4"} = maps:find(<<"k2">>, M#{ <<"k2">> => "v4" }), + M0 = id(#{ k1=>"v1", <<"k2">> => <<"v3">> }), + {ok, "v4"} = maps:find(<<"k2">>, M0#{ <<"k2">> => "v4" }), + + %% large map + M1 = maps:from_list([{I,I}||I<-lists:seq(1,100)] ++ + [{a,1},{b,2},{"hello","hi"},{{1,1.0},"tuple hi"}, + {k1,"v1"},{<<"k2">>,"v3"}]), + {ok, 1} = maps:find(a, M1), + {ok, 2} = maps:find(b, M1), + {ok, "hi"} = maps:find("hello", M1), + {ok, "tuple hi"} = maps:find({1,1.0}, M1), + {ok, "v3"} = maps:find(<<"k2">>, M1), %% error case error = maps:find(a,#{}), @@ -508,7 +824,6 @@ t_bif_map_find(Config) when is_list(Config) -> error = maps:find(1, #{ 1.0 => "float"}), error = maps:find({1.0,1}, #{ a=>a, {1,1.0} => "tuple hi"}), % reverse types in tuple key - {'EXIT',{badarg,[{maps,find,_,_}|_]}} = (catch maps:find(a,id([]))), {'EXIT',{badarg,[{maps,find,_,_}|_]}} = (catch maps:find(a,id(<<>>))), ok. @@ -542,12 +857,12 @@ t_bif_map_is_key(Config) when is_list(Config) -> t_bif_map_keys(Config) when is_list(Config) -> [] = maps:keys(#{}), - [1,2,3,4,5] = maps:keys(#{ 1 => a, 2 => b, 3 => c, 4 => d, 5 => e}), - [1,2,3,4,5] = maps:keys(#{ 4 => d, 5 => e, 1 => a, 2 => b, 3 => c}), + [1,2,3,4,5] = lists:sort(maps:keys(#{ 1 => a, 2 => b, 3 => c, 4 => d, 5 => e})), + [1,2,3,4,5] = lists:sort(maps:keys(#{ 4 => d, 5 => e, 1 => a, 2 => b, 3 => c})), % values in key order: [4,int,"hi",<<"key">>] M1 = #{ "hi" => "hello", int => 3, <<"key">> => <<"value">>, 4 => number}, - [4,int,"hi",<<"key">>] = maps:keys(M1), + [4,int,"hi",<<"key">>] = lists:sort(maps:keys(M1)), %% error case {'EXIT',{badarg,[{maps,keys,_,_}|_]}} = (catch maps:keys(1 bsl 65 + 3)), @@ -596,33 +911,33 @@ t_bif_map_put(Config) when is_list(Config) -> M1 = #{ "hi" := "hello"} = maps:put("hi", "hello", #{}), - ["hi"] = maps:keys(M1), - ["hello"] = maps:values(M1), + true = is_members(["hi"],maps:keys(M1)), + true = is_members(["hello"],maps:values(M1)), M2 = #{ int := 3 } = maps:put(int, 3, M1), - [int,"hi"] = maps:keys(M2), - [3,"hello"] = maps:values(M2), + true = is_members([int,"hi"],maps:keys(M2)), + true = is_members([3,"hello"],maps:values(M2)), M3 = #{ <<"key">> := <<"value">> } = maps:put(<<"key">>, <<"value">>, M2), - [int,"hi",<<"key">>] = maps:keys(M3), - [3,"hello",<<"value">>] = maps:values(M3), + true = is_members([int,"hi",<<"key">>],maps:keys(M3)), + true = is_members([3,"hello",<<"value">>],maps:values(M3)), M4 = #{ 18446744073709551629 := wat } = maps:put(18446744073709551629, wat, M3), - [18446744073709551629,int,"hi",<<"key">>] = maps:keys(M4), - [wat,3,"hello",<<"value">>] = maps:values(M4), + true = is_members([18446744073709551629,int,"hi",<<"key">>],maps:keys(M4)), + true = is_members([wat,3,"hello",<<"value">>],maps:values(M4)), M0 = #{ 4 := number } = M5 = maps:put(4, number, M4), - [4,18446744073709551629,int,"hi",<<"key">>] = maps:keys(M5), - [number,wat,3,"hello",<<"value">>] = maps:values(M5), + true = is_members([4,18446744073709551629,int,"hi",<<"key">>],maps:keys(M5)), + true = is_members([number,wat,3,"hello",<<"value">>],maps:values(M5)), M6 = #{ <<"key">> := <<"other value">> } = maps:put(<<"key">>, <<"other value">>, M5), - [4,18446744073709551629,int,"hi",<<"key">>] = maps:keys(M6), - [number,wat,3,"hello",<<"other value">>] = maps:values(M6), + true = is_members([4,18446744073709551629,int,"hi",<<"key">>],maps:keys(M6)), + true = is_members([number,wat,3,"hello",<<"other value">>],maps:values(M6)), %% error case {'EXIT',{badarg,[{maps,put,_,_}|_]}} = (catch maps:put(1,a,1 bsl 65 + 3)), @@ -630,7 +945,15 @@ t_bif_map_put(Config) when is_list(Config) -> {'EXIT',{badarg,[{maps,put,_,_}|_]}} = (catch maps:put(1,a,atom)), {'EXIT',{badarg,[{maps,put,_,_}|_]}} = (catch maps:put(1,a,[])), {'EXIT',{badarg,[{maps,put,_,_}|_]}} = (catch maps:put(1,a,<<>>)), - ok. + ok. + +is_members(Ks,Ls) when length(Ks) =/= length(Ls) -> false; +is_members(Ks,Ls) -> is_members_do(Ks,Ls). + +is_members_do([],[]) -> true; +is_members_do([],_) -> false; +is_members_do([K|Ks],Ls) -> + is_members_do(Ks, lists:delete(K,Ls)). t_bif_map_remove(Config) when is_list(Config) -> 0 = erlang:map_size(maps:remove(some_key, #{})), @@ -639,20 +962,20 @@ t_bif_map_remove(Config) when is_list(Config) -> 4 => number, 18446744073709551629 => wat}, M1 = maps:remove("hi", M0), - [4,18446744073709551629,int,<<"key">>] = maps:keys(M1), - [number,wat,3,<<"value">>] = maps:values(M1), + true = is_members([4,18446744073709551629,int,<<"key">>],maps:keys(M1)), + true = is_members([number,wat,3,<<"value">>],maps:values(M1)), M2 = maps:remove(int, M1), - [4,18446744073709551629,<<"key">>] = maps:keys(M2), - [number,wat,<<"value">>] = maps:values(M2), + true = is_members([4,18446744073709551629,<<"key">>],maps:keys(M2)), + true = is_members([number,wat,<<"value">>],maps:values(M2)), M3 = maps:remove(<<"key">>, M2), - [4,18446744073709551629] = maps:keys(M3), - [number,wat] = maps:values(M3), + true = is_members([4,18446744073709551629],maps:keys(M3)), + true = is_members([number,wat],maps:values(M3)), M4 = maps:remove(18446744073709551629, M3), - [4] = maps:keys(M4), - [number] = maps:values(M4), + true = is_members([4],maps:keys(M4)), + true = is_members([number],maps:values(M4)), M5 = maps:remove(4, M4), [] = maps:keys(M5), @@ -702,15 +1025,15 @@ t_bif_map_update(Config) when is_list(Config) -> t_bif_map_values(Config) when is_list(Config) -> [] = maps:values(#{}), + [1] = maps:values(#{a=>1}), - [a,b,c,d,e] = maps:values(#{ 1 => a, 2 => b, 3 => c, 4 => d, 5 => e}), - [a,b,c,d,e] = maps:values(#{ 4 => d, 5 => e, 1 => a, 2 => b, 3 => c}), + true = is_members([a,b,c,d,e],maps:values(#{ 1 => a, 2 => b, 3 => c, 4 => d, 5 => e})), + true = is_members([a,b,c,d,e],maps:values(#{ 4 => d, 5 => e, 1 => a, 2 => b, 3 => c})), - % values in key order: [4,int,"hi",<<"key">>] M1 = #{ "hi" => "hello", int => 3, <<"key">> => <<"value">>, 4 => number}, M2 = M1#{ "hi" => "hello2", <<"key">> => <<"value2">> }, - [number,3,"hello2",<<"value2">>] = maps:values(M2), - [number,3,"hello",<<"value">>] = maps:values(M1), + true = is_members([number,3,"hello2",<<"value2">>],maps:values(M2)), + true = is_members([number,3,"hello",<<"value">>],maps:values(M1)), %% error case {'EXIT',{badarg,[{maps,values,_,_}|_]}} = (catch maps:values(1 bsl 65 + 3)), @@ -730,61 +1053,61 @@ t_erlang_hash(Config) when is_list(Config) -> t_bif_erlang_phash2() -> 39679005 = erlang:phash2(#{}), - 78942764 = erlang:phash2(#{ a => 1, "a" => 2, <<"a">> => 3, {a,b} => 4 }), - 37338230 = erlang:phash2(#{ 1 => a, 2 => "a", 3 => <<"a">>, 4 => {a,b} }), - 14363616 = erlang:phash2(#{ 1 => a }), - 51612236 = erlang:phash2(#{ a => 1 }), + 33667975 = erlang:phash2(#{ a => 1, "a" => 2, <<"a">> => 3, {a,b} => 4 }), % 78942764 + 95332690 = erlang:phash2(#{ 1 => a, 2 => "a", 3 => <<"a">>, 4 => {a,b} }), % 37338230 + 108954384 = erlang:phash2(#{ 1 => a }), % 14363616 + 59617982 = erlang:phash2(#{ a => 1 }), % 51612236 - 37468437 = erlang:phash2(#{{} => <<>>}), - 44049159 = erlang:phash2(#{<<>> => {}}), + 42770201 = erlang:phash2(#{{} => <<>>}), % 37468437 + 71687700 = erlang:phash2(#{<<>> => {}}), % 44049159 M0 = #{ a => 1, "key" => <<"value">> }, M1 = maps:remove("key",M0), M2 = M1#{ "key" => <<"value">> }, - 118679416 = erlang:phash2(M0), - 51612236 = erlang:phash2(M1), - 118679416 = erlang:phash2(M2), + 70249457 = erlang:phash2(M0), % 118679416 + 59617982 = erlang:phash2(M1), % 51612236 + 70249457 = erlang:phash2(M2), % 118679416 ok. t_bif_erlang_phash() -> Sz = 1 bsl 32, - 268440612 = erlang:phash(#{},Sz), - 1196461908 = erlang:phash(#{ a => 1, "a" => 2, <<"a">> => 3, {a,b} => 4 },Sz), - 3944426064 = erlang:phash(#{ 1 => a, 2 => "a", 3 => <<"a">>, 4 => {a,b} },Sz), - 1394238263 = erlang:phash(#{ 1 => a },Sz), - 4066388227 = erlang:phash(#{ a => 1 },Sz), + 1113425985 = erlang:phash(#{},Sz), % 268440612 + 1510068139 = erlang:phash(#{ a => 1, "a" => 2, <<"a">> => 3, {a,b} => 4 },Sz), % 1196461908 + 3182345590 = erlang:phash(#{ 1 => a, 2 => "a", 3 => <<"a">>, 4 => {a,b} },Sz), % 3944426064 + 2927531828 = erlang:phash(#{ 1 => a },Sz), % 1394238263 + 1670235874 = erlang:phash(#{ a => 1 },Sz), % 4066388227 - 1578050717 = erlang:phash(#{{} => <<>>},Sz), - 1578050717 = erlang:phash(#{<<>> => {}},Sz), % yep, broken + 3935089469 = erlang:phash(#{{} => <<>>},Sz), % 1578050717 + 71692856 = erlang:phash(#{<<>> => {}},Sz), % 1578050717 M0 = #{ a => 1, "key" => <<"value">> }, M1 = maps:remove("key",M0), M2 = M1#{ "key" => <<"value">> }, - 3590546636 = erlang:phash(M0,Sz), - 4066388227 = erlang:phash(M1,Sz), - 3590546636 = erlang:phash(M2,Sz), + 2620391445 = erlang:phash(M0,Sz), % 3590546636 + 1670235874 = erlang:phash(M1,Sz), % 4066388227 + 2620391445 = erlang:phash(M2,Sz), % 3590546636 ok. t_bif_erlang_hash() -> Sz = 1 bsl 27 - 1, - 5158 = erlang:hash(#{},Sz), - 71555838 = erlang:hash(#{ a => 1, "a" => 2, <<"a">> => 3, {a,b} => 4 },Sz), - 5497225 = erlang:hash(#{ 1 => a, 2 => "a", 3 => <<"a">>, 4 => {a,b} },Sz), - 126071654 = erlang:hash(#{ 1 => a },Sz), - 126426236 = erlang:hash(#{ a => 1 },Sz), + 39684169 = erlang:hash(#{},Sz), % 5158 + 33673142 = erlang:hash(#{ a => 1, "a" => 2, <<"a">> => 3, {a,b} => 4 },Sz), % 71555838 + 95337869 = erlang:hash(#{ 1 => a, 2 => "a", 3 => <<"a">>, 4 => {a,b} },Sz), % 5497225 + 108959561 = erlang:hash(#{ 1 => a },Sz), % 126071654 + 59623150 = erlang:hash(#{ a => 1 },Sz), % 126426236 - 101655720 = erlang:hash(#{{} => <<>>},Sz), - 101655720 = erlang:hash(#{<<>> => {}},Sz), % yep, broken + 42775386 = erlang:hash(#{{} => <<>>},Sz), % 101655720 + 71692856 = erlang:hash(#{<<>> => {}},Sz), % 101655720 M0 = #{ a => 1, "key" => <<"value">> }, M1 = maps:remove("key",M0), M2 = M1#{ "key" => <<"value">> }, - 38260486 = erlang:hash(M0,Sz), - 126426236 = erlang:hash(M1,Sz), - 38260486 = erlang:hash(M2,Sz), + 70254632 = erlang:hash(M0,Sz), % 38260486 + 59623150 = erlang:hash(M1,Sz), % 126426236 + 70254632 = erlang:hash(M2,Sz), % 38260486 ok. @@ -818,14 +1141,21 @@ t_map_encode_decode(Config) when is_list(Config) -> %% literally #{ "hi" => "value", a=>33, b=>55 } in the internal order #{ a:=33, b:=55, "hi" := "value"} = erlang:binary_to_term(<<131,116,0,0,0,3, - 107,0,2,104,105, % "hi" :: list() + 107,0,2,104,105, % "hi" :: list() 107,0,5,118,97,108,117,101, % "value" :: list() - 100,0,1,97, % a :: atom() - 97,33, % 33 :: integer() - 100,0,1,98, % b :: atom() - 97,55 % 55 :: integer() + 100,0,1,97, % a :: atom() + 97,33, % 33 :: integer() + 100,0,1,98, % b :: atom() + 97,55 % 55 :: integer() >>), + %% many maps in same binary + MapList = lists:foldl(fun(K, [M|_]=Acc) -> [M#{K => K} | Acc] end, + [#{}], + lists:seq(1,100)), + MapList = binary_to_term(term_to_binary(MapList)), + MapListR = lists:reverse(MapList), + MapListR = binary_to_term(term_to_binary(MapListR)), %% error cases %% template: <<131,116,0,0,0,2,100,0,1,97,100,0,1,98,97,1,97,1>> @@ -856,39 +1186,42 @@ t_map_encode_decode(Config) when is_list(Config) -> map_encode_decode_and_match([{K,V}|Pairs], EncodedPairs, M0) -> M1 = maps:put(K,V,M0), B0 = erlang:term_to_binary(M1), - Ls = lists:sort(fun(A,B) -> erts_internal:cmp_term(A,B) < 0 end, [{K, erlang:term_to_binary(K), erlang:term_to_binary(V)}|EncodedPairs]), - %% sort Ks and Vs according to term spec, then match it - KVbins = lists:foldr(fun({_,Kbin,Vbin}, Acc) -> [Kbin,Vbin | Acc] end, [], Ls), - ok = match_encoded_map(B0, length(Ls), KVbins), + Ls = [{erlang:term_to_binary(K), erlang:term_to_binary(V)}|EncodedPairs], + ok = match_encoded_map(B0, length(Ls), Ls), %% decode and match it M1 = erlang:binary_to_term(B0), map_encode_decode_and_match(Pairs,Ls,M1); map_encode_decode_and_match([],_,_) -> ok. match_encoded_map(<<131,116,Size:32,Encoded/binary>>,Size,Items) -> - match_encoded_map(Encoded,Items); + match_encoded_map_stripped_size(Encoded,Items,Items); match_encoded_map(_,_,_) -> no_match_size. -match_encoded_map(<<>>,[]) -> ok; -match_encoded_map(Bin,[<<131,Item/binary>>|Items]) -> - Size = erlang:byte_size(Item), - <<EncodedTerm:Size/binary, Bin1/binary>> = Bin, - EncodedTerm = Item, %% Asssert - match_encoded_map(Bin1,Items). +match_encoded_map_stripped_size(<<>>,_,_) -> ok; +match_encoded_map_stripped_size(B0,[{<<131,K/binary>>,<<131,V/binary>>}|Items],Ls) -> + Ksz = byte_size(K), + Vsz = byte_size(V), + case B0 of + <<K:Ksz/binary,V:Vsz/binary,B1/binary>> -> + match_encoded_map_stripped_size(B1,Ls,Ls); + _ -> + match_encoded_map_stripped_size(B0,Items,Ls) + end; +match_encoded_map_stripped_size(_,[],_) -> fail. t_bif_map_to_list(Config) when is_list(Config) -> [] = maps:to_list(#{}), - [{a,1},{b,2}] = maps:to_list(#{a=>1,b=>2}), - [{a,1},{b,2},{c,3}] = maps:to_list(#{c=>3,a=>1,b=>2}), - [{a,1},{b,2},{g,3}] = maps:to_list(#{g=>3,a=>1,b=>2}), - [{a,1},{b,2},{g,3},{"c",4}] = maps:to_list(#{g=>3,a=>1,b=>2,"c"=>4}), - [{3,v2},{hi,v4},{{hi,3},v5},{"hi",v3},{<<"hi">>,v1}] = maps:to_list(#{ - <<"hi">>=>v1,3=>v2,"hi"=>v3,hi=>v4,{hi,3}=>v5}), + [{a,1},{b,2}] = lists:sort(maps:to_list(#{a=>1,b=>2})), + [{a,1},{b,2},{c,3}] = lists:sort(maps:to_list(#{c=>3,a=>1,b=>2})), + [{a,1},{b,2},{g,3}] = lists:sort(maps:to_list(#{g=>3,a=>1,b=>2})), + [{a,1},{b,2},{g,3},{"c",4}] = lists:sort(maps:to_list(#{g=>3,a=>1,b=>2,"c"=>4})), + [{3,v2},{hi,v4},{{hi,3},v5},{"hi",v3},{<<"hi">>,v1}] = + lists:sort(maps:to_list(#{<<"hi">>=>v1,3=>v2,"hi"=>v3,hi=>v4,{hi,3}=>v5})), - [{3,v7},{hi,v9},{{hi,3},v10},{"hi",v8},{<<"hi">>,v6}] = maps:to_list(#{ - <<"hi">>=>v1,3=>v2,"hi"=>v3,hi=>v4,{hi,3}=>v5, - <<"hi">>=>v6,3=>v7,"hi"=>v8,hi=>v9,{hi,3}=>v10}), + [{3,v7},{hi,v9},{{hi,3},v10},{"hi",v8},{<<"hi">>,v6}] = + lists:sort(maps:to_list(#{<<"hi">>=>v1,3=>v2,"hi"=>v3,hi=>v4,{hi,3}=>v5, + <<"hi">>=>v6,3=>v7,"hi"=>v8,hi=>v9,{hi,3}=>v10})), %% error cases {'EXIT', {badarg,_}} = (catch maps:to_list(id(a))), @@ -901,7 +1234,7 @@ t_bif_map_from_list(Config) when is_list(Config) -> A = maps:from_list([]), 0 = erlang:map_size(A), - #{a:=1,b:=2} = maps:from_list([{a,1},{b,2}]), + #{a:=1,b:=2} = maps:from_list([{a,1},{b,2}]), #{c:=3,a:=1,b:=2} = maps:from_list([{a,1},{b,2},{c,3}]), #{g:=3,a:=1,b:=2} = maps:from_list([{a,1},{b,2},{g,3}]), @@ -923,6 +1256,136 @@ t_bif_map_from_list(Config) when is_list(Config) -> {'EXIT', {badarg,_}} = (catch maps:from_list(id(42))), ok. +t_bif_build_and_check(Config) when is_list(Config) -> + ok = check_build_and_remove(750,[ + fun(K) -> [K,K] end, + fun(K) -> [float(K),K] end, + fun(K) -> K end, + fun(K) -> {1,K} end, + fun(K) -> {K} end, + fun(K) -> [K|K] end, + fun(K) -> [K,1,2,3,4] end, + fun(K) -> {K,atom} end, + fun(K) -> float(K) end, + fun(K) -> integer_to_list(K) end, + fun(K) -> list_to_atom(integer_to_list(K)) end, + fun(K) -> [K,{K,[K,{K,[K]}]}] end, + fun(K) -> <<K:32>> end + ]), + + ok. + +check_build_and_remove(_,[]) -> ok; +check_build_and_remove(N,[F|Fs]) -> + {M,Ks} = build_and_check(N, maps:new(), F, []), + ok = remove_and_check(Ks,M), + check_build_and_remove(N,Fs). + +build_and_check(0, M0, _, Ks) -> {M0, Ks}; +build_and_check(N, M0, F, Ks) -> + K = build_key(F,N), + M1 = maps:put(K,K,M0), + ok = check_keys_exist([K|Ks], M1), + M2 = maps:update(K,v,M1), + v = maps:get(K,M2), + build_and_check(N-1,M1,F,[K|Ks]). + +remove_and_check([],_) -> ok; +remove_and_check([K|Ks], M0) -> + K = maps:get(K,M0), + true = maps:is_key(K,M0), + M1 = maps:remove(K,M0), + false = maps:is_key(K,M1), + true = maps:is_key(K,M0), + ok = check_keys_exist(Ks,M1), + error = maps:find(K,M1), + remove_and_check(Ks, M1). + +build_key(F,N) when N rem 3 =:= 0 -> F(N); +build_key(F,N) when N rem 3 =:= 1 -> K = F(N), {K,K}; +build_key(F,N) when N rem 3 =:= 2 -> K = F(N), [K,K]. + +check_keys_exist([], _) -> ok; +check_keys_exist([K|Ks],M) -> + true = maps:is_key(K,M), + check_keys_exist(Ks,M). + +t_bif_merge_and_check(Config) when is_list(Config) -> + %% simple disjunct ones + %% make sure all keys are unique + Kss = [[a,b,c,d], + [1,2,3,4], + [], + ["hi"], + [e], + [build_key(fun(K) -> {small,K} end, I) || I <- lists:seq(1,32)], + lists:seq(5, 28), + lists:seq(29, 59), + [build_key(fun(K) -> integer_to_list(K) end, I) || I <- lists:seq(2000,10000)], + [build_key(fun(K) -> <<K:32>> end, I) || I <- lists:seq(1,80)], + [build_key(fun(K) -> {<<K:32>>} end, I) || I <- lists:seq(100,1000)]], + + + KsMs = build_keys_map_pairs(Kss), + Cs = [{CKs1,CM1,CKs2,CM2} || {CKs1,CM1} <- KsMs, {CKs2,CM2} <- KsMs], + ok = merge_and_check_combo(Cs), + + %% overlapping ones + + KVs1 = [{a,1},{b,2},{c,3}], + KVs2 = [{b,3},{c,4},{d,5}], + KVs = [{I,I} || I <- lists:seq(1,32)], + KVs3 = KVs1 ++ KVs, + KVs4 = KVs2 ++ KVs, + + M1 = maps:from_list(KVs1), + M2 = maps:from_list(KVs2), + M3 = maps:from_list(KVs3), + M4 = maps:from_list(KVs4), + + M12 = maps:merge(M1,M2), + ok = check_key_values(KVs2 ++ [{a,1}], M12), + M21 = maps:merge(M2,M1), + ok = check_key_values(KVs1 ++ [{d,5}], M21), + + M34 = maps:merge(M3,M4), + ok = check_key_values(KVs4 ++ [{a,1}], M34), + M43 = maps:merge(M4,M3), + ok = check_key_values(KVs3 ++ [{d,5}], M43), + + M14 = maps:merge(M1,M4), + ok = check_key_values(KVs4 ++ [{a,1}], M14), + M41 = maps:merge(M4,M1), + ok = check_key_values(KVs1 ++ [{d,5}] ++ KVs, M41), + + ok. + +check_key_values([],_) -> ok; +check_key_values([{K,V}|KVs],M) -> + V = maps:get(K,M), + check_key_values(KVs,M). + +merge_and_check_combo([]) -> ok; +merge_and_check_combo([{Ks1,M1,Ks2,M2}|Cs]) -> + M12 = maps:merge(M1,M2), + ok = check_keys_exist(Ks1 ++ Ks2, M12), + M21 = maps:merge(M2,M1), + ok = check_keys_exist(Ks1 ++ Ks2, M21), + + true = M12 =:= M21, + M12 = M21, + + merge_and_check_combo(Cs). + +build_keys_map_pairs([]) -> []; +build_keys_map_pairs([Ks|Kss]) -> + M = maps:from_list(keys_to_pairs(Ks)), + ok = check_keys_exist(Ks, M), + [{Ks,M}|build_keys_map_pairs(Kss)]. + +keys_to_pairs(Ks) -> [{K,K} || K <- Ks]. + + %% Maps module, not BIFs t_maps_fold(_Config) -> Vs = lists:seq(1,100), @@ -960,6 +1423,75 @@ t_maps_without(_Config) -> %% MISC + +%% Verify that the the number of nodes in hashmaps +%% of different types and sizes does not deviate too +%% much from the theoretical model. +t_hashmap_balance(_Config) -> + io:format("Integer keys\n", []), + hashmap_balance(fun(I) -> I end), + io:format("Float keys\n", []), + hashmap_balance(fun(I) -> float(I) end), + io:format("String keys\n", []), + hashmap_balance(fun(I) -> integer_to_list(I) end), + io:format("Binary (big) keys\n", []), + hashmap_balance(fun(I) -> <<I:16/big>> end), + io:format("Binary (little) keys\n", []), + hashmap_balance(fun(I) -> <<I:16/little>> end), + io:format("Atom keys\n", []), + erts_debug:set_internal_state(available_internal_state, true), + hashmap_balance(fun(I) -> erts_debug:get_internal_state({atom,I}) end), + erts_debug:set_internal_state(available_internal_state, false), + + ok. + +hashmap_balance(KeyFun) -> + F = fun(I, {M0,Max0}) -> + Key = KeyFun(I), + M1 = M0#{Key => Key}, + Max1 = case erts_internal:map_type(M1) of + hashmap -> + Nodes = hashmap_nodes(M1), + Avg = maps:size(M1) * 0.4, + StdDev = math:sqrt(maps:size(M1)) / 3, + SD_diff = abs(Nodes - Avg) / StdDev, + %%io:format("~p keys: ~p nodes avg=~p SD_diff=~p\n", + %% [maps:size(M1), Nodes, Avg, SD_diff]), + {MaxDiff0, _} = Max0, + case {Nodes > Avg, SD_diff > MaxDiff0} of + {true, true} -> {SD_diff, M1}; + _ -> Max0 + end; + + flatmap -> Max0 + end, + {M1, Max1} + end, + + {_,{MaxDiff,MaxMap}} = lists:foldl(F, + {#{}, {0, 0}}, + lists:seq(1,10000)), + io:format("Max std dev diff ~p for map of size ~p (nodes=~p, flatsize=~p)\n", + [MaxDiff, maps:size(MaxMap), hashmap_nodes(MaxMap), erts_debug:flat_size(MaxMap)]), + + true = (MaxDiff < 6), % The probability of this line failing is about 0.000000001 + % for a uniform hash. I've set the probability this "high" for now + % to detect flaws in our make_internal_hash. + % Hard limit is 15 (see hashmap_over_estimated_heap_size). + ok. + +hashmap_nodes(M) -> + Info = erts_debug:map_info(M), + lists:foldl(fun(Tpl,Acc) -> + case element(1,Tpl) of + bitmaps -> Acc + element(2,Tpl); + arrays -> Acc + element(2,Tpl); + _ -> Acc + end + end, + 0, + Info). + t_pdict(_Config) -> put(#{ a => b, b => a},#{ c => d}), diff --git a/erts/emulator/test/match_spec_SUITE.erl b/erts/emulator/test/match_spec_SUITE.erl index fdce157abc..d3c884689f 100644 --- a/erts/emulator/test/match_spec_SUITE.erl +++ b/erts/emulator/test/match_spec_SUITE.erl @@ -30,6 +30,7 @@ -export([fpe/1]). -export([otp_9422/1]). -export([faulty_seq_trace/1, do_faulty_seq_trace/0]). +-export([maps/1]). -export([runner/2, loop_runner/3]). -export([f1/1, f2/2, f3/2, fn/1, fn/2, fn/3]). -export([do_boxed_and_small/0]). @@ -62,7 +63,8 @@ all() -> moving_labels, faulty_seq_trace, empty_list, - otp_9422]; + otp_9422, + maps]; true -> [not_run] end. @@ -899,6 +901,74 @@ fpe(Config) when is_list(Config) -> _ -> ok end. +maps(Config) when is_list(Config) -> + {ok,#{},[],[]} = erlang:match_spec_test(#{}, [{'_',[],['$_']}], table), + {ok,#{},[],[]} = erlang:match_spec_test(#{}, [{#{},[],['$_']}], table), + {ok,false,[],[]} = + erlang:match_spec_test(#{}, [{not_a_map,[],['$_']}], table), + {ok,bar,[],[]} = + erlang:match_spec_test(#{foo => bar}, + [{#{foo => '$1'},[],['$1']}], + table), + {ok,false,[],[]} = + erlang:match_spec_test(#{foo => bar}, + [{#{foo => qux},[],[qux]}], + table), + {ok,false,[],[]} = + erlang:match_spec_test(#{}, [{#{foo => '_'},[],[foo]}], table), + {error,_} = + erlang:match_spec_test(#{}, [{#{'$1' => '_'},[],[foo]}], table), + {ok,bar,[],[]} = + erlang:match_spec_test({#{foo => bar}}, + [{{#{foo => '$1'}},[],['$1']}], + table), + {ok,#{foo := 3},[],[]} = + erlang:match_spec_test({}, [{{},[],[#{foo => {'+',1,2}}]}], table), + + {ok,"camembert",[],[]} = + erlang:match_spec_test(#{b => "camembert",c => "cabécou"}, + [{#{b => '$1',c => "cabécou"},[],['$1']}], table), + + {ok,#{a :="camembert",b := "hi"},[],[]} = + erlang:match_spec_test(#{<<"b">> =>"camembert","c"=>"cabécou", "wat"=>"hi", b=><<"other">>}, + [{#{<<"b">> => '$1',"wat" => '$2'},[],[#{a=>'$1',b=>'$2'}]}], + table), + %% large maps + + Ls0 = [{I,<<I:32>>}||I <- lists:seq(1,415)], + M0 = maps:from_list(Ls0), + M1 = #{a=>1,b=>2,c=>3,d=>4}, + + R1 = M0#{263 := #{ a=> 3 }}, + Ms1 = [{M1#{c:='$1'},[],[M0#{263 := #{a => '$1'}}]}], + + {ok,R1,[],[]} = erlang:match_spec_test(M1,Ms1,table), + + Ms2 = [{M0#{63:='$1', 19:='$2'},[],[M0#{19:='$1', 63:='$2'}]}], + R2 = M0#{63 := maps:get(19,M0), 19 := maps:get(63,M0) }, + {ok,R2,[],[]} = erlang:match_spec_test(M0,Ms2,table), + + ok = maps_check_loop(M1), + ok = maps_check_loop(M0), + M2 = maps:from_list([{integer_to_list(K),V} || {K,V} <- Ls0]), + ok = maps_check_loop(M2), + ok. + +maps_check_loop(M) -> + Ks = maps:keys(M), + maps_check_loop(M,M,M,M,Ks,lists:reverse(Ks),1). + +maps_check_loop(Orig,M0,MsM0,Rm0,[K|Ks],[Rk|Rks],Ix) -> + MsK = list_to_atom([$$]++integer_to_list(Ix)), + MsM1 = MsM0#{K := MsK}, + Rm1 = Rm0#{Rk := MsK}, + M1 = M0#{Rk := maps:get(K,MsM0)}, + Ms = [{MsM1,[],[Rm1]}], + {ok,M1,[],[]} = erlang:match_spec_test(Orig,Ms,table), + maps_check_loop(Orig,M1,MsM1,Rm1,Ks,Rks,Ix+1); +maps_check_loop(_,_,_,_,[],[],_) -> ok. + + empty_list(Config) when is_list(Config) -> Val=[{'$1',[], [{message,'$1'},{message,{caller}},{return_trace}]}], %% Did crash debug VM in faulty assert: diff --git a/erts/emulator/test/module_info_SUITE.erl b/erts/emulator/test/module_info_SUITE.erl index 8a63d9fe3e..f3986f0c4f 100644 --- a/erts/emulator/test/module_info_SUITE.erl +++ b/erts/emulator/test/module_info_SUITE.erl @@ -24,7 +24,7 @@ -export([all/0, suite/0,groups/0,init_per_suite/1, end_per_suite/1, init_per_group/2,end_per_group/2, init_per_testcase/2,end_per_testcase/2, - exports/1,functions/1,native/1]). + exports/1,functions/1,native/1,info/1]). %%-compile(native). @@ -52,8 +52,8 @@ end_per_group(_GroupName, Config) -> Config. -modules() -> - [exports, functions, native]. +modules() -> + [exports, functions, native, info]. init_per_testcase(Func, Config) when is_atom(Func), is_list(Config) -> Dog = ?t:timetrap(?t:minutes(3)), @@ -122,6 +122,22 @@ native_proj({Name,Arity,Addr}) -> native_filter(Set) -> sofs:no_elements(Set) =/= 1. +%% Test that the module info of this module is correct. Use +%% erlang:get_module_info(?MODULE) to avoid compiler optimization tricks. +info(Config) when is_list(Config) -> + Info = erlang:get_module_info(?MODULE), + All = all_exported(), + {ok,{?MODULE,MD5}} = beam_lib:md5(code:which(?MODULE)), + {module, ?MODULE} = lists:keyfind(module, 1, Info), + {md5, MD5} = lists:keyfind(md5, 1, Info), + {exports, Exports} = lists:keyfind(exports, 1, Info), + All = lists:sort(Exports), + {attributes, Attrs} = lists:keyfind(attributes, 1, Info), + {vsn,_} = lists:keyfind(vsn, 1, Attrs), + {compile, Compile} = lists:keyfind(compile, 1, Info), + {options,_} = lists:keyfind(options, 1, Compile), + ok. + %% Helper functions (local). add_arity(L) -> diff --git a/erts/emulator/test/nif_SUITE.erl b/erts/emulator/test/nif_SUITE.erl index 4560077a51..b0624fb8c1 100644 --- a/erts/emulator/test/nif_SUITE.erl +++ b/erts/emulator/test/nif_SUITE.erl @@ -451,7 +451,7 @@ maps(Config) when is_list(Config) -> M = maps_from_list_nif(Pairs), R = {RIs,Is} = sorted_list_from_maps_nif(M), io:format("Pairs: ~p~nMap: ~p~nReturned: ~p~n", [lists:sort(Pairs),M,R]), - Is = lists:sort(Pairs), + true = (lists:sort(Is) =:= lists:sort(Pairs)), Is = lists:reverse(RIs), #{} = maps_from_list_nif([]), diff --git a/erts/emulator/test/nif_SUITE_data/nif_SUITE.c b/erts/emulator/test/nif_SUITE_data/nif_SUITE.c index 85544db2ab..5a3be84825 100644 --- a/erts/emulator/test/nif_SUITE_data/nif_SUITE.c +++ b/erts/emulator/test/nif_SUITE_data/nif_SUITE.c @@ -1717,8 +1717,9 @@ static ERL_NIF_TERM sorted_list_from_maps_nif(ErlNifEnv* env, int argc, const ER return enif_make_int(env, __LINE__); cnt = 0; + next_ret = 1; while(enif_map_iterator_get_pair(env,&iter_f,&key,&value)) { - if (cnt && !next_ret) + if (!next_ret) return enif_make_int(env, __LINE__); list_f = enif_make_list_cell(env, enif_make_tuple2(env, key, value), list_f); next_ret = enif_map_iterator_next(env,&iter_f); @@ -1731,8 +1732,9 @@ static ERL_NIF_TERM sorted_list_from_maps_nif(ErlNifEnv* env, int argc, const ER return enif_make_int(env, __LINE__); cnt = 0; + prev_ret = 1; while(enif_map_iterator_get_pair(env,&iter_b,&key,&value)) { - if (cnt && !prev_ret) + if (!prev_ret) return enif_make_int(env, __LINE__); /* Test that iter_f can step "backwards" */ @@ -1744,6 +1746,7 @@ static ERL_NIF_TERM sorted_list_from_maps_nif(ErlNifEnv* env, int argc, const ER list_b = enif_make_list_cell(env, enif_make_tuple2(env, key, value), list_b); prev_ret = enif_map_iterator_prev(env,&iter_b); + cnt++; } if (cnt) { diff --git a/erts/emulator/test/port_SUITE.erl b/erts/emulator/test/port_SUITE.erl index 9083545060..6bbf93b7d7 100644 --- a/erts/emulator/test/port_SUITE.erl +++ b/erts/emulator/test/port_SUITE.erl @@ -1407,6 +1407,12 @@ spawn_executable(Config) when is_list(Config) -> run_echo_args(SpaceDir,[ExactFile2,"hello world","dlrow olleh"]), [ExactFile2,"hello world","dlrow olleh"] = run_echo_args(SpaceDir,[binary, ExactFile2,"hello world","dlrow olleh"]), + + [ExactFile2,"hello \"world\"","\"dlrow\" olleh"] = + run_echo_args(SpaceDir,[binary, ExactFile2,"hello \"world\"","\"dlrow\" olleh"]), + [ExactFile2,"hello \"world\"","\"dlrow\" olleh"] = + run_echo_args(SpaceDir,[binary, ExactFile2,"hello \"world\"","\"dlrow\" olleh"]), + [ExactFile2] = run_echo_args(SpaceDir,[default]), [ExactFile2,"hello world","dlrow olleh"] = run_echo_args(SpaceDir,[switch_order,ExactFile2,"hello world", diff --git a/erts/emulator/test/trace_bif_SUITE.erl b/erts/emulator/test/trace_bif_SUITE.erl index 2c78aa394f..063e348836 100644 --- a/erts/emulator/test/trace_bif_SUITE.erl +++ b/erts/emulator/test/trace_bif_SUITE.erl @@ -260,7 +260,9 @@ bif_process() -> apply(erlang, Name, Args), bif_process(); {do_time_bif} -> - _ = time(), %Assignment tells compiler to keep call. + %% Match the return value to ensure that the time() call + %% is not optimized away. + {_,_,_} = time(), bif_process(); {do_statistics_bif} -> statistics(runtime), diff --git a/erts/emulator/valgrind/suppress.patched.3.6.0 b/erts/emulator/valgrind/suppress.patched.3.6.0 index b3507bdba7..f79e3ff634 100644 --- a/erts/emulator/valgrind/suppress.patched.3.6.0 +++ b/erts/emulator/valgrind/suppress.patched.3.6.0 @@ -273,6 +273,11 @@ obj:*/ssleay.* fun:AES_cbc_encrypt ... } +{ + crypto RC4 can do harmless word aligned read past end of input + Memcheck:Addr8 + fun:RC4 +} { erts_bits_init_state; Why is this needed? diff --git a/erts/emulator/valgrind/suppress.standard b/erts/emulator/valgrind/suppress.standard index a4da31a61d..b3c77119fb 100644 --- a/erts/emulator/valgrind/suppress.standard +++ b/erts/emulator/valgrind/suppress.standard @@ -260,6 +260,11 @@ obj:*/ssleay.* fun:AES_cbc_encrypt ... } +{ + crypto RC4 can do harmless word aligned read past end of input + Memcheck:Addr8 + fun:RC4 +} { Prebuilt constant terms in os_info_init (PossiblyLost) diff --git a/erts/epmd/src/epmd.c b/erts/epmd/src/epmd.c index 9699491526..447aae47aa 100644 --- a/erts/epmd/src/epmd.c +++ b/erts/epmd/src/epmd.c @@ -345,7 +345,7 @@ static void run_daemon(EpmdVars *g) inform it of that the log is closed. */ closelog(); - /* These chouldn't be needed but for safety... */ + /* These shouldn't be needed but for safety... */ open("/dev/null", O_RDONLY); /* Order is important! */ open("/dev/null", O_WRONLY); @@ -386,7 +386,7 @@ static void run_daemon(EpmdVars *g) close(1); close(2); - /* These chouldn't be needed but for safety... */ + /* These shouldn't be needed but for safety... */ open("nul", O_RDONLY); open("nul", O_WRONLY); diff --git a/erts/etc/common/erlexec.c b/erts/etc/common/erlexec.c index d33ea59c1d..23226909a7 100644 --- a/erts/etc/common/erlexec.c +++ b/erts/etc/common/erlexec.c @@ -143,6 +143,7 @@ static char *pluss_val_switches[] = { static char *plush_val_switches[] = { "ms", "mbs", + "pds", "", NULL }; diff --git a/erts/etc/unix/cerl.src b/erts/etc/unix/cerl.src index 6f54475e5a..59cf29d381 100644 --- a/erts/etc/unix/cerl.src +++ b/erts/etc/unix/cerl.src @@ -43,6 +43,7 @@ # -gcov Run emulator compiled for gcov # -valgrind Run emulator compiled for valgrind # -lcnt Run emulator compiled for lock counting +# -icount Run emulator compiled for instruction counting # -nox Unset the DISPLAY variable to disable us of X Windows # # FIXME For GDB you can also set the break point using "-break FUNCTION". @@ -180,6 +181,11 @@ while [ $# -gt 0 ]; do cargs="$cargs -frmptr" TYPE=.frmptr ;; + "-icount") + shift + cargs="$cargs -icount" + TYPE=.icount + ;; "-dump") shift GDB=dump diff --git a/erts/etc/unix/etp-commands.in b/erts/etc/unix/etp-commands.in index 0190ea613e..c117a62a21 100644 --- a/erts/etc/unix/etp-commands.in +++ b/erts/etc/unix/etp-commands.in @@ -1065,8 +1065,8 @@ define etp-cp-1 set $etp_cp_mid = $etp_cp_low + ($etp_cp_high-$etp_cp_low)/2 end if $etp_cp_p - # 12 = MI_FUNCTIONS - set $etp_cp_low = (Eterm**)($etp_cp_p->start + 12) + # 13 = MI_FUNCTIONS + set $etp_cp_low = (Eterm**)($etp_cp_p->start + 13) # 0 = MI_NUM_FUNCTIONS set $etp_cp_high = $etp_cp_low +$etp_cp_p->start[0] set $etp_cp_p = 0 @@ -2434,7 +2434,7 @@ define etp-system-info printf "ERTS version: %s\n", etp_erts_version printf "Compile date: %s\n", etp_compile_date printf "Arch: %s\n", etp_arch - printf "Endianess: " + printf "Endianness: " if (etp_big_endian) printf "Big\n" else @@ -3514,11 +3514,13 @@ end define etp-carrier-blocks set $etp_crr = (Carrier_t*) $arg0 set $etp_alc = (Allctr_t*)($etp_crr->allctr.counter & ~7) + set $etp_crr_end = ((char*)$etp_crr + ($etp_crr->chdr & ~7) - (sizeof(void*) & ~8)) set $etp_blk = (Block_t*) ((char*)$etp_crr + $etp_alc->mbc_header_size) set $etp_prev_blk = 0 set $etp_error_cnt = 0 set $etp_ablk_cnt = 0 set $etp_fblk_cnt = 0 + set $etp_aborted = 0 if $argc == 2 set $etp_be_silent = $arg1 @@ -3565,14 +3567,21 @@ define etp-carrier-blocks end set $etp_prev_blk = $etp_blk set $etp_blk = (Block_t*) ((char*)$etp_blk + $etp_blk_sz) + if $etp_blk < (Block_t*) ((char*)$etp_prev_blk + $etp_alc->min_block_size) || $etp_blk >= $etp_crr_end + printf "ERROR: Invalid size of block at %#lx. ABORTING\n", $etp_prev_blk + set $etp_aborted = 1 + loop_break + end end - if ((char*)$etp_blk + $etp_blk_sz) != ((char*)$etp_crr + ($etp_crr->chdr & ~7)) - printf "ERROR: Last block not at end of carrier\n" - set $etp_error_cnt = $etp_error_cnt + 1 + if !$etp_aborted + if ((char*)$etp_blk + $etp_blk_sz) != $etp_crr_end + printf "ERROR: Last block not at end of carrier\n" + set $etp_error_cnt = $etp_error_cnt + 1 + end + printf "Allocated blocks: %u\n", $etp_ablk_cnt + printf "Free blocks: %u\n", $etp_fblk_cnt end - printf "Allocated blocks: %u\n", $etp_ablk_cnt - printf "Free blocks: %u\n", $etp_fblk_cnt if $etp_error_cnt printf "%u ERRORs reported above\n", $etp-error-cnt end diff --git a/erts/include/internal/ethread.h b/erts/include/internal/ethread.h index ad5d05704c..e598017ada 100644 --- a/erts/include/internal/ethread.h +++ b/erts/include/internal/ethread.h @@ -1,7 +1,7 @@ /* * %CopyrightBegin% * - * Copyright Ericsson AB 2004-2013. All Rights Reserved. + * Copyright Ericsson AB 2004-2015. All Rights Reserved. * * The contents of this file are subject to the Erlang Public License, * Version 1.1, (the "License"); you may not use this file except in @@ -214,8 +214,6 @@ typedef OSPPDKEY ethr_tsd_key; /* Out own RW mutexes are probably faster, but use OSEs mutexes */ #define ETHR_USE_OWN_RWMTX_IMPL__ -#define ETHR_HAVE_THREAD_NAMES - #else /* No supported thread lib found */ #ifdef ETHR_NO_SUPP_THR_LIB_NOT_FATAL @@ -364,6 +362,9 @@ extern ethr_runtime_t ethr_runtime__; # include "sparc64/ethread.h" # endif # endif +# if ETHR_HAVE_GCC___ATOMIC_BUILTINS +# include "gcc/ethread.h" +# endif # include "libatomic_ops/ethread.h" # include "gcc/ethread.h" # endif @@ -501,21 +502,18 @@ typedef struct { typedef struct { int detached; /* boolean (default false) */ int suggested_stack_size; /* kilo words (default sys dependent) */ + char *name; /* max 14 char long (default no-name) */ #ifdef ETHR_OSE_THREADS - char *name; U32 coreNo; #endif } ethr_thr_opts; #if defined(ETHR_OSE_THREADS) -/* Default ethr name is big as we want to be able to sprint stuff in there */ -#define ETHR_THR_OPTS_DEFAULT_INITER \ - {0, -1, "ethread", 0} +#define ETHR_THR_OPTS_DEFAULT_INITER {0, -1, NULL, 0} #else -#define ETHR_THR_OPTS_DEFAULT_INITER {0, -1} +#define ETHR_THR_OPTS_DEFAULT_INITER {0, -1, NULL} #endif - #if !defined(ETHR_TRY_INLINE_FUNCS) || defined(ETHR_AUX_IMPL__) # define ETHR_NEED_SPINLOCK_PROTOTYPES__ # define ETHR_NEED_RWSPINLOCK_PROTOTYPES__ @@ -529,6 +527,8 @@ int ethr_thr_join(ethr_tid, void **); int ethr_thr_detach(ethr_tid); void ethr_thr_exit(void *); ethr_tid ethr_self(void); +int ethr_getname(ethr_tid, char *, size_t); +void ethr_setname(char *); int ethr_equal_tids(ethr_tid, ethr_tid); int ethr_tsd_key_create(ethr_tsd_key *,char *); @@ -540,6 +540,7 @@ void *ethr_tsd_get(ethr_tsd_key); #include <signal.h> int ethr_sigmask(int how, const sigset_t *set, sigset_t *oset); int ethr_sigwait(const sigset_t *set, int *sig); +int ethr_kill(const ethr_tid tid, const int sig); #endif void ethr_compiler_barrier(void); diff --git a/erts/include/internal/ethread_header_config.h.in b/erts/include/internal/ethread_header_config.h.in index 618fcba380..a9727568a2 100644 --- a/erts/include/internal/ethread_header_config.h.in +++ b/erts/include/internal/ethread_header_config.h.in @@ -1,7 +1,7 @@ /* * %CopyrightBegin% * - * Copyright Ericsson AB 2004-2011. All Rights Reserved. + * Copyright Ericsson AB 2004-2015. All Rights Reserved. * * The contents of this file are subject to the Erlang Public License, * Version 1.1, (the "License"); you may not use this file except in @@ -84,32 +84,62 @@ /* Define if run in Sparc RMO, PSO, or TSO mode */ #undef ETHR_SPARC_RMO -/* Define if you have __sync_add_and_fetch() for 32-bit integers */ -#undef ETHR_HAVE___SYNC_ADD_AND_FETCH32 +/* Define as a boolean indicating whether you have a gcc compatible compiler + capable of generating the ARM DMB instruction, and are compiling for an ARM + processor with ARM DMB instruction support, or not */ +#undef ETHR_HAVE_GCC_ASM_ARM_DMB_INSTRUCTION -/* Define if you have __sync_add_and_fetch() for 64-bit integers */ -#undef ETHR_HAVE___SYNC_ADD_AND_FETCH64 +/* Define as a bitmask corresponding to the word sizes that + __sync_synchronize() can handle on your system */ +#undef ETHR_HAVE___sync_synchronize -/* Define if you have __sync_fetch_and_and() for 32-bit integers */ -#undef ETHR_HAVE___SYNC_FETCH_AND_AND32 +/* Define as a bitmask corresponding to the word sizes that + __sync_add_and_fetch() can handle on your system */ +#undef ETHR_HAVE___sync_add_and_fetch -/* Define if you have __sync_fetch_and_and() for 64-bit integers */ -#undef ETHR_HAVE___SYNC_FETCH_AND_AND64 +/* Define as a bitmask corresponding to the word sizes that + __sync_fetch_and_and() can handle on your system */ +#undef ETHR_HAVE___sync_fetch_and_and -/* Define if you have __sync_fetch_and_or() for 32-bit integers */ -#undef ETHR_HAVE___SYNC_FETCH_AND_OR32 +/* Define as a bitmask corresponding to the word sizes that + __sync_fetch_and_or() can handle on your system */ +#undef ETHR_HAVE___sync_fetch_and_or -/* Define if you have __sync_fetch_and_or() for 64-bit integers */ -#undef ETHR_HAVE___SYNC_FETCH_AND_OR64 +/* Define as a bitmask corresponding to the word sizes that + __sync_val_compare_and_swap() can handle on your system */ +#undef ETHR_HAVE___sync_val_compare_and_swap -/* Define if you have __sync_val_compare_and_swap() for 32-bit integers */ -#undef ETHR_HAVE___SYNC_VAL_COMPARE_AND_SWAP32 +/* Define as a boolean indicating whether you have a gcc __atomic builtins or + not */ +#undef ETHR_HAVE_GCC___ATOMIC_BUILTINS -/* Define if you have __sync_val_compare_and_swap() for 64-bit integers */ -#undef ETHR_HAVE___SYNC_VAL_COMPARE_AND_SWAP64 +/* Define as a boolean indicating whether you trust gcc's __atomic_* builtins + memory barrier implementations, or not */ +#undef ETHR_TRUST_GCC_ATOMIC_BUILTINS_MEMORY_BARRIERS -/* Define if you have __sync_val_compare_and_swap() for 128-bit integers */ -#undef ETHR_HAVE___SYNC_VAL_COMPARE_AND_SWAP128 +/* Define as a bitmask corresponding to the word sizes that __atomic_store_n() + can handle on your system */ +#undef ETHR_HAVE___atomic_store_n + +/* Define as a bitmask corresponding to the word sizes that __atomic_load_n() + can handle on your system */ +#undef ETHR_HAVE___atomic_load_n + +/* Define as a bitmask corresponding to the word sizes that + __atomic_add_fetch() can handle on your system */ +#undef ETHR_HAVE___atomic_add_fetch + +/* Define as a bitmask corresponding to the word sizes that + __atomic_fetch_and() can handle on your system */ +#undef ETHR_HAVE___atomic_fetch_and + +/* Define as a bitmask corresponding to the word sizes that + __atomic_fetch_or() can handle on your system */ +#undef ETHR_HAVE___atomic_fetch_or + +/* Define as a bitmask corresponding to the word sizes that + __atomic_compare_exchange_n() can handle on your system */ +#undef ETHR_HAVE___atomic_compare_exchange_n /* Define if you prefer gcc native ethread implementations */ #undef ETHR_PREFER_GCC_NATIVE_IMPLS diff --git a/erts/include/internal/ethread_inline.h b/erts/include/internal/ethread_inline.h index ffb756c84f..c09a67619a 100644 --- a/erts/include/internal/ethread_inline.h +++ b/erts/include/internal/ethread_inline.h @@ -20,6 +20,29 @@ #ifndef ETHREAD_INLINE_H__ #define ETHREAD_INLINE_H__ +#define ETHR_GCC_COMPILER_FALSE 0 /* Not a gcc compatible compiler */ +#define ETHR_GCC_COMPILER_TRUE 1 /* The GNU gcc compiler */ +/* Negative integers for gcc compatible compilers */ +#define ETHR_GCC_COMPILER_CLANG -1 /* The Clang gcc compatible compiler */ +#define ETHR_GCC_COMPILER_ICC -2 /* The Intel gcc compatible compiler */ +/* Line them up... */ + +/* + * Unfortunately there is no easy and certain way of + * detecting a true gcc compiler, since the compatible + * ones all define the same defines as the true gnu-gcc... + */ +#if !defined(__GNUC__) && !defined(__GNUG__) +# define ETHR_GCC_COMPILER ETHR_GCC_COMPILER_FALSE +#elif defined(__clang__) +# define ETHR_GCC_COMPILER ETHR_GCC_COMPILER_CLANG +#elif defined(__ICC) || defined(__INTEL_COMPILER) +# define ETHR_GCC_COMPILER ETHR_GCC_COMPILER_ICC +#else +/* Seems to be the true gnu-gcc... */ +# define ETHR_GCC_COMPILER ETHR_GCC_COMPILER_TRUE +#endif + #if !defined(__GNUC__) # define ETHR_AT_LEAST_GCC_VSN__(MAJ, MIN, PL) 0 #elif !defined(__GNUC_MINOR__) diff --git a/erts/include/internal/gcc/ethr_atomic.h b/erts/include/internal/gcc/ethr_atomic.h index f598f8537b..62eed78f76 100644 --- a/erts/include/internal/gcc/ethr_atomic.h +++ b/erts/include/internal/gcc/ethr_atomic.h @@ -1,7 +1,7 @@ /* * %CopyrightBegin% * - * Copyright Ericsson AB 2010-2011. All Rights Reserved. + * Copyright Ericsson AB 2010-2015. All Rights Reserved. * * The contents of this file are subject to the Erlang Public License, * Version 1.1, (the "License"); you may not use this file except in @@ -18,78 +18,81 @@ */ /* - * Description: Native atomics ethread support using gcc's builtins + * Description: Native atomics ethread support using gcc's __atomic + * and __sync builtins * Author: Rickard Green + * + * Note: The C11 memory model implemented by gcc's __atomic + * builtins does not match the ethread API very well. + * + * Due to this we cannot use the __ATOMIC_SEQ_CST + * memory model. For more information see the comment + * in the begining of ethr_membar.h in this directory. */ #undef ETHR_INCLUDE_ATOMIC_IMPL__ -#if !defined(ETHR_GCC_ATOMIC32_H__) && defined(ETHR_ATOMIC_WANT_32BIT_IMPL__) -#define ETHR_GCC_ATOMIC32_H__ -#if defined(ETHR_HAVE___SYNC_VAL_COMPARE_AND_SWAP32) -# define ETHR_INCLUDE_ATOMIC_IMPL__ 4 -#endif +#if !defined(ETHR_GCC_ATOMIC_ATOMIC32_H__) \ + && defined(ETHR_ATOMIC_WANT_32BIT_IMPL__) \ + && ((ETHR_HAVE___sync_val_compare_and_swap & 4) \ + || (ETHR_HAVE___atomic_compare_exchange_n & 4)) + +#define ETHR_GCC_ATOMIC_ATOMIC32_H__ +#define ETHR_INCLUDE_ATOMIC_IMPL__ 4 #undef ETHR_ATOMIC_WANT_32BIT_IMPL__ -#elif !defined(ETHR_GCC_ATOMIC64_H__) && defined(ETHR_ATOMIC_WANT_64BIT_IMPL__) -#define ETHR_GCC_ATOMIC64_H__ -#if defined(ETHR_HAVE___SYNC_VAL_COMPARE_AND_SWAP64) -# define ETHR_INCLUDE_ATOMIC_IMPL__ 8 -#endif -#undef ETHR_ATOMIC_WANT_64BIT_IMPL__ -#endif -#ifdef ETHR_INCLUDE_ATOMIC_IMPL__ +#elif !defined(ETHR_GCC_ATOMIC64_H__) \ + && defined(ETHR_ATOMIC_WANT_64BIT_IMPL__) \ + && ((ETHR_HAVE___sync_val_compare_and_swap & 8) \ + || (ETHR_HAVE___atomic_compare_exchange_n & 8)) -#ifndef ETHR_GCC_ATOMIC_COMMON__ -#define ETHR_GCC_ATOMIC_COMMON__ +#define ETHR_GCC_ATOMIC64_H__ +#define ETHR_INCLUDE_ATOMIC_IMPL__ 8 +#undef ETHR_ATOMIC_WANT_64BIT_IMPL__ -#define ETHR_READ_AND_SET_WITHOUT_SYNC_OP__ 0 -#if defined(__i386__) || defined(__x86_64__) || defined(__sparc__) \ - || defined(__powerpc__) || defined(__ppc__) || defined(__mips__) -# undef ETHR_READ_AND_SET_WITHOUT_SYNC_OP__ -# define ETHR_READ_AND_SET_WITHOUT_SYNC_OP__ 1 #endif -#endif /* ETHR_GCC_ATOMIC_COMMON__ */ +#ifdef ETHR_INCLUDE_ATOMIC_IMPL__ #if ETHR_INCLUDE_ATOMIC_IMPL__ == 4 #define ETHR_HAVE_NATIVE_ATOMIC32 1 -#define ETHR_NATIVE_ATOMIC32_IMPL "gcc" #define ETHR_NATMC_FUNC__(X) ethr_native_atomic32_ ## X #define ETHR_ATMC_T__ ethr_native_atomic32_t #define ETHR_AINT_T__ ethr_sint32_t -#if defined(ETHR_HAVE___SYNC_ADD_AND_FETCH32) -# define ETHR_HAVE___SYNC_ADD_AND_FETCH -#endif -#if defined(ETHR_HAVE___SYNC_FETCH_AND_AND32) -# define ETHR_HAVE___SYNC_FETCH_AND_AND -#endif -#if defined(ETHR_HAVE___SYNC_FETCH_AND_OR32) -# define ETHR_HAVE___SYNC_FETCH_AND_OR +#if ((ETHR_HAVE___sync_val_compare_and_swap & 4) \ + && (ETHR_HAVE___atomic_compare_exchange_n & 4)) +# define ETHR_NATIVE_ATOMIC32_IMPL "gcc_atomic_and_sync_builtins" +#elif (ETHR_HAVE___atomic_compare_exchange_n & 4) +# define ETHR_NATIVE_ATOMIC32_IMPL "gcc_atomic_builtins" +#elif (ETHR_HAVE___sync_val_compare_and_swap & 4) +# define ETHR_NATIVE_ATOMIC32_IMPL "gcc_sync_builtins" +#else +# error "!?" #endif #elif ETHR_INCLUDE_ATOMIC_IMPL__ == 8 #define ETHR_HAVE_NATIVE_ATOMIC64 1 -#define ETHR_NATIVE_ATOMIC64_IMPL "gcc" #define ETHR_NATMC_FUNC__(X) ethr_native_atomic64_ ## X #define ETHR_ATMC_T__ ethr_native_atomic64_t #define ETHR_AINT_T__ ethr_sint64_t -#if defined(ETHR_HAVE___SYNC_ADD_AND_FETCH64) -# define ETHR_HAVE___SYNC_ADD_AND_FETCH -#endif -#if defined(ETHR_HAVE___SYNC_FETCH_AND_AND64) -# define ETHR_HAVE___SYNC_FETCH_AND_AND -#endif -#if defined(ETHR_HAVE___SYNC_FETCH_AND_OR64) -# define ETHR_HAVE___SYNC_FETCH_AND_OR +#if ((ETHR_HAVE___sync_val_compare_and_swap & 8) \ + && (ETHR_HAVE___atomic_compare_exchange_n & 8)) +# define ETHR_NATIVE_ATOMIC64_IMPL "gcc_atomic_and_sync_builtins" +#elif (ETHR_HAVE___atomic_compare_exchange_n & 8) +# define ETHR_NATIVE_ATOMIC64_IMPL "gcc_atomic_builtins" +#elif (ETHR_HAVE___sync_val_compare_and_swap & 8) +# define ETHR_NATIVE_ATOMIC64_IMPL "gcc_sync_builtins" +#else +# error "!?" #endif #else #error "Unsupported integer size" #endif +#undef ETHR_NATIVE_ATOMIC_IMPL__ + typedef struct { - volatile ETHR_AINT_T__ counter; + volatile ETHR_AINT_T__ value; } ETHR_ATMC_T__; - #if defined(ETHR_TRY_INLINE_FUNCS) || defined(ETHR_ATOMIC_IMPL__) #if ETHR_INCLUDE_ATOMIC_IMPL__ == 4 @@ -98,13 +101,19 @@ typedef struct { # define ETHR_HAVE_ETHR_NATIVE_ATOMIC64_ADDR 1 #endif + static ETHR_INLINE ETHR_AINT_T__ * ETHR_NATMC_FUNC__(addr)(ETHR_ATMC_T__ *var) { - return (ETHR_AINT_T__ *) &var->counter; + return (ETHR_AINT_T__ *) &var->value; } -#if ETHR_READ_AND_SET_WITHOUT_SYNC_OP__ +/* + * set() + */ +#if (ETHR_HAVE___atomic_store_n & ETHR_INCLUDE_ATOMIC_IMPL__) + +#if (ETHR_GCC_RELAXED_VERSIONS__ & ETHR_INCLUDE_ATOMIC_IMPL__) #if ETHR_INCLUDE_ATOMIC_IMPL__ == 4 # define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_SET 1 @@ -115,12 +124,109 @@ ETHR_NATMC_FUNC__(addr)(ETHR_ATMC_T__ *var) static ETHR_INLINE void ETHR_NATMC_FUNC__(set)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ value) { - var->counter = value; + __atomic_store_n(&var->value, value, __ATOMIC_RELAXED); } -#endif /* ETHR_READ_AND_SET_WITHOUT_SYNC_OP__ */ +#endif /* ETHR_GCC_RELAXED_VERSIONS__ */ -#if ETHR_READ_AND_SET_WITHOUT_SYNC_OP__ +#if (ETHR_GCC_RELB_VERSIONS__ & ETHR_INCLUDE_ATOMIC_IMPL__) + +#if ETHR_INCLUDE_ATOMIC_IMPL__ == 4 +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_SET_RELB 1 +#else +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC64_SET_RELB 1 +#endif + +static ETHR_INLINE void +ETHR_NATMC_FUNC__(set_relb)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ value) +{ + __atomic_store_n(&var->value, value, __ATOMIC_RELEASE); +} + +#endif /* ETHR_GCC_RELB_VERSIONS__ */ + +#elif (ETHR_GCC_VOLATILE_STORE_IS_ATOMIC_STORE__ & ETHR_INCLUDE_ATOMIC_IMPL__) + +#if (ETHR_GCC_RELAXED_VERSIONS__ & ETHR_INCLUDE_ATOMIC_IMPL__) + +#if ETHR_INCLUDE_ATOMIC_IMPL__ == 4 +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_SET 1 +#else +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC64_SET 1 +#endif + +static ETHR_INLINE void +ETHR_NATMC_FUNC__(set)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ value) +{ + var->value = value; +} + +#endif /* ETHR_GCC_RELAXED_VERSIONS__ */ + +#if (ETHR_GCC_VOLATILE_STORE_IS_ATOMIC_STORE_RELB__ & ETHR_INCLUDE_ATOMIC_IMPL__) + +#if (ETHR_GCC_RELB_VERSIONS__ & ETHR_INCLUDE_ATOMIC_IMPL__) + +#if ETHR_INCLUDE_ATOMIC_IMPL__ == 4 +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_SET_RELB 1 +#else +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC64_SET_RELB 1 +#endif + +static ETHR_INLINE void +ETHR_NATMC_FUNC__(set_relb)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ value) +{ + var->value = value; +} + +#endif /* ETHR_GCC_RELB_VERSIONS__ */ + +#endif /* ETHR_GCC_VOLATILE_STORE_IS_ATOMIC_STORE_RELB__ */ + +#endif /* ETHR_GCC_VOLATILE_STORE_IS_ATOMIC_STORE__ */ + +/* + * read() + */ + +#if (ETHR_HAVE___atomic_load_n & ETHR_INCLUDE_ATOMIC_IMPL__) + +#if (ETHR_GCC_RELAXED_VERSIONS__ & ETHR_INCLUDE_ATOMIC_IMPL__) + +#if ETHR_INCLUDE_ATOMIC_IMPL__ == 4 +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_READ 1 +#else +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC64_READ 1 +#endif + +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(read)(ETHR_ATMC_T__ *var) +{ + return __atomic_load_n(&var->value, __ATOMIC_RELAXED); +} + +#endif /* ETHR_GCC_RELAXED_VERSIONS__ */ + +#if ((ETHR_GCC_ACQB_VERSIONS__ & ETHR_INCLUDE_ATOMIC_IMPL__) \ + & ~ETHR___atomic_load_ACQUIRE_barrier_bug) + +#if ETHR_INCLUDE_ATOMIC_IMPL__ == 4 +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_READ_ACQB 1 +#else +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC64_READ_ACQB 1 +#endif + +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(read_acqb)(ETHR_ATMC_T__ *var) +{ + return __atomic_load_n(&var->value, __ATOMIC_ACQUIRE); +} + +#endif /* ETHR_GCC_ACQB_VERSIONS__ */ + +#elif (ETHR_GCC_VOLATILE_LOAD_IS_ATOMIC_LOAD__ & ETHR_INCLUDE_ATOMIC_IMPL__) + +#if (ETHR_GCC_RELAXED_VERSIONS__ & ETHR_INCLUDE_ATOMIC_IMPL__) #if ETHR_INCLUDE_ATOMIC_IMPL__ == 4 # define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_READ 1 @@ -131,12 +237,90 @@ ETHR_NATMC_FUNC__(set)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ value) static ETHR_INLINE ETHR_AINT_T__ ETHR_NATMC_FUNC__(read)(ETHR_ATMC_T__ *var) { - return var->counter; + return var->value; +} + +#endif /* ETHR_GCC_RELAXED_VERSIONS__ */ + +#if (ETHR_GCC_VOLATILE_LOAD_IS_ATOMIC_LOAD_ACQB__ & ETHR_INCLUDE_ATOMIC_IMPL__) + +#if (ETHR_GCC_ACQB_VERSIONS__ & ETHR_INCLUDE_ATOMIC_IMPL__) + +#if ETHR_INCLUDE_ATOMIC_IMPL__ == 4 +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_READ_ACQB 1 +#else +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC64_READ_ACQB 1 +#endif + +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(read_acqb)(ETHR_ATMC_T__ *var) +{ + return var->value; +} + +#endif /* ETHR_GCC_ACQB_VERSIONS__ */ + +#endif /* ETHR_GCC_VOLATILE_LOAD_IS_ATOMIC_LOAD_ACQB__ */ + +#endif /* ETHR_GCC_VOLATILE_LOAD_IS_ATOMIC_LOAD__ */ + +/* + * add_return() + */ +#if (ETHR_HAVE___atomic_add_fetch & ETHR_INCLUDE_ATOMIC_IMPL__) + +#if (ETHR_GCC_RELAXED_MOD_VERSIONS__ & ETHR_INCLUDE_ATOMIC_IMPL__) + +#if ETHR_INCLUDE_ATOMIC_IMPL__ == 4 +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_ADD_RETURN 1 +#else +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC64_ADD_RETURN 1 +#endif + +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(add_return)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ incr) +{ + return __atomic_add_fetch(&var->value, incr, __ATOMIC_RELAXED); } -#endif /* ETHR_READ_AND_SET_WITHOUT_SYNC_OP__ */ +#endif /* ETHR_GCC_RELAXED_MOD_VERSIONS__ */ -#if defined(ETHR_HAVE___SYNC_ADD_AND_FETCH) +#if (ETHR_GCC_ACQB_MOD_VERSIONS__ & ETHR_INCLUDE_ATOMIC_IMPL__) + +#if ETHR_INCLUDE_ATOMIC_IMPL__ == 4 +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_ADD_RETURN_ACQB 1 +#else +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC64_ADD_RETURN_ACQB 1 +#endif + +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(add_return_acqb)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ incr) +{ + return __atomic_add_fetch(&var->value, incr, __ATOMIC_ACQUIRE); +} + +#endif /* ETHR_GCC_ACQB_MOD_VERSIONS__ */ + +#if (ETHR_GCC_RELB_MOD_VERSIONS__ & ETHR_INCLUDE_ATOMIC_IMPL__) + +#if ETHR_INCLUDE_ATOMIC_IMPL__ == 4 +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_ADD_RETURN_RELB 1 +#else +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC64_ADD_RETURN_RELB 1 +#endif + +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(add_return_relb)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ incr) +{ + return __atomic_add_fetch(&var->value, incr, __ATOMIC_RELEASE); +} + +#endif /* ETHR_GCC_RELB_MOD_VERSIONS__ */ + +#endif /* ETHR_HAVE___atomic_add_fetch */ + +#if ((ETHR_HAVE___sync_add_and_fetch & ETHR_INCLUDE_ATOMIC_IMPL__) \ + & ETHR_GCC_MB_MOD_VERSIONS__) #if ETHR_INCLUDE_ATOMIC_IMPL__ == 4 # define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_ADD_RETURN_MB 1 @@ -147,12 +331,68 @@ ETHR_NATMC_FUNC__(read)(ETHR_ATMC_T__ *var) static ETHR_INLINE ETHR_AINT_T__ ETHR_NATMC_FUNC__(add_return_mb)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ incr) { - return __sync_add_and_fetch(&var->counter, incr); + return __sync_add_and_fetch(&var->value, incr); +} + +#endif /* ETHR_HAVE___sync_add_and_fetch */ + +/* + * and_retold() + */ +#if (ETHR_HAVE___atomic_fetch_and & ETHR_INCLUDE_ATOMIC_IMPL__) + +#if (ETHR_GCC_RELAXED_MOD_VERSIONS__ & ETHR_INCLUDE_ATOMIC_IMPL__) + +#if ETHR_INCLUDE_ATOMIC_IMPL__ == 4 +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_AND_RETOLD 1 +#else +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC64_AND_RETOLD 1 +#endif + +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(and_retold)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ mask) +{ + return __atomic_fetch_and(&var->value, mask, __ATOMIC_RELAXED); +} + +#endif /* ETHR_GCC_RELAXED_MOD_VERSIONS__ */ + +#if (ETHR_GCC_ACQB_MOD_VERSIONS__ & ETHR_INCLUDE_ATOMIC_IMPL__) + +#if ETHR_INCLUDE_ATOMIC_IMPL__ == 4 +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_AND_RETOLD_ACQB 1 +#else +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC64_AND_RETOLD_ACQB 1 +#endif + +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(and_retold_acqb)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ mask) +{ + return __atomic_fetch_and(&var->value, mask, __ATOMIC_ACQUIRE); } +#endif /* ETHR_GCC_ACQB_MOD_VERSIONS__ */ + +#if (ETHR_GCC_RELB_MOD_VERSIONS__ & ETHR_INCLUDE_ATOMIC_IMPL__) + +#if ETHR_INCLUDE_ATOMIC_IMPL__ == 4 +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_AND_RETOLD_RELB 1 +#else +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC64_AND_RETOLD_RELB 1 #endif -#if defined(ETHR_HAVE___SYNC_FETCH_AND_AND) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(and_retold_relb)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ mask) +{ + return __atomic_fetch_and(&var->value, mask, __ATOMIC_RELEASE); +} + +#endif /* ETHR_GCC_RELB_MOD_VERSIONS__ */ + +#endif /* ETHR_HAVE___atomic_fetch_and */ + +#if ((ETHR_HAVE___sync_fetch_and_and & ETHR_INCLUDE_ATOMIC_IMPL__) \ + & ETHR_GCC_MB_MOD_VERSIONS__) #if ETHR_INCLUDE_ATOMIC_IMPL__ == 4 # define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_AND_RETOLD_MB 1 @@ -163,12 +403,68 @@ ETHR_NATMC_FUNC__(add_return_mb)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ incr) static ETHR_INLINE ETHR_AINT_T__ ETHR_NATMC_FUNC__(and_retold_mb)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ mask) { - return __sync_fetch_and_and(&var->counter, mask); + return __sync_fetch_and_and(&var->value, mask); +} + +#endif /* ETHR_HAVE___sync_fetch_and_and */ + +/* + * or_retold() + */ +#if (ETHR_HAVE___atomic_fetch_or & ETHR_INCLUDE_ATOMIC_IMPL__) + +#if (ETHR_GCC_RELAXED_MOD_VERSIONS__ & ETHR_INCLUDE_ATOMIC_IMPL__) + +#if ETHR_INCLUDE_ATOMIC_IMPL__ == 4 +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_OR_RETOLD 1 +#else +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC64_OR_RETOLD 1 +#endif + +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(or_retold)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ mask) +{ + return __atomic_fetch_or(&var->value, mask, __ATOMIC_RELAXED); +} + +#endif /* ETHR_GCC_RELAXED_MOD_VERSIONS__ */ + +#if (ETHR_GCC_ACQB_MOD_VERSIONS__ & ETHR_INCLUDE_ATOMIC_IMPL__) + +#if ETHR_INCLUDE_ATOMIC_IMPL__ == 4 +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_OR_RETOLD_ACQB 1 +#else +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC64_OR_RETOLD_ACQB 1 +#endif + +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(or_retold_acqb)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ mask) +{ + return __atomic_fetch_or(&var->value, mask, __ATOMIC_ACQUIRE); } +#endif /* ETHR_GCC_ACQB_MOD_VERSIONS__ */ + +#if (ETHR_GCC_RELB_MOD_VERSIONS__ & ETHR_INCLUDE_ATOMIC_IMPL__) + +#if ETHR_INCLUDE_ATOMIC_IMPL__ == 4 +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_OR_RETOLD_RELB 1 +#else +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC64_OR_RETOLD_RELB 1 #endif -#if defined(ETHR_HAVE___SYNC_FETCH_AND_OR) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(or_retold_relb)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ mask) +{ + return __atomic_fetch_or(&var->value, mask, __ATOMIC_RELEASE); +} + +#endif /* ETHR_GCC_RELB_MOD_VERSIONS__ */ + +#endif /* ETHR_HAVE___atomic_fetch_or */ + +#if ((ETHR_HAVE___sync_fetch_and_or & ETHR_INCLUDE_ATOMIC_IMPL__) \ + & ETHR_GCC_MB_MOD_VERSIONS__) #if ETHR_INCLUDE_ATOMIC_IMPL__ == 4 # define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_OR_RETOLD_MB 1 @@ -179,11 +475,73 @@ ETHR_NATMC_FUNC__(and_retold_mb)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ mask) static ETHR_INLINE ETHR_AINT_T__ ETHR_NATMC_FUNC__(or_retold_mb)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ mask) { - return (ETHR_AINT_T__) __sync_fetch_and_or(&var->counter, mask); + return (ETHR_AINT_T__) __sync_fetch_and_or(&var->value, mask); +} + +#endif /* ETHR_HAVE___sync_fetch_and_or */ + +/* + * cmpxchg() + */ +#if (ETHR_HAVE___atomic_compare_exchange_n & ETHR_INCLUDE_ATOMIC_IMPL__) + +#if (ETHR_GCC_RELAXED_MOD_VERSIONS__ & ETHR_INCLUDE_ATOMIC_IMPL__) + +#if ETHR_INCLUDE_ATOMIC_IMPL__ == 4 +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_CMPXCHG 1 +#else +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC64_CMPXCHG 1 +#endif + +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(cmpxchg)(ETHR_ATMC_T__ *var, + ETHR_AINT_T__ new, + ETHR_AINT_T__ exp) +{ + ETHR_AINT_T__ xchg = exp; + if (__atomic_compare_exchange_n(&var->value, + &xchg, + new, + 0, /* No spurious failures, please */ + __ATOMIC_RELAXED, + __ATOMIC_RELAXED)) + return exp; + return xchg; } +#endif /* ETHR_GCC_RELAXED_MOD_VERSIONS__ */ + +#if (ETHR_GCC_ACQB_MOD_VERSIONS__ & ETHR_INCLUDE_ATOMIC_IMPL__) + +#if ETHR_INCLUDE_ATOMIC_IMPL__ == 4 +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_CMPXCHG_ACQB 1 +#else +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC64_CMPXCHG_ACQB 1 #endif +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(cmpxchg_acqb)(ETHR_ATMC_T__ *var, + ETHR_AINT_T__ new, + ETHR_AINT_T__ exp) +{ + ETHR_AINT_T__ xchg = exp; + if (__atomic_compare_exchange_n(&var->value, + &xchg, + new, + 0, /* No spurious failures, please */ + __ATOMIC_ACQUIRE, + __ATOMIC_ACQUIRE)) + return exp; + return xchg; +} + +#endif /* ETHR_GCC_ACQB_MOD_VERSIONS__ */ + +#endif /* ETHR_HAVE___atomic_compare_exchange_n */ + +#if ((ETHR_HAVE___sync_val_compare_and_swap & ETHR_INCLUDE_ATOMIC_IMPL__) \ + & ETHR_GCC_MB_MOD_VERSIONS__) + #if ETHR_INCLUDE_ATOMIC_IMPL__ == 4 # define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_CMPXCHG_MB 1 #else @@ -195,17 +553,16 @@ ETHR_NATMC_FUNC__(cmpxchg_mb)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ new, ETHR_AINT_T__ old) { - return __sync_val_compare_and_swap(&var->counter, old, new); + return __sync_val_compare_and_swap(&var->value, old, new); } +#endif /* ETHR_HAVE___sync_val_compare_and_swap */ + #endif /* ETHR_TRY_INLINE_FUNCS */ #undef ETHR_NATMC_FUNC__ #undef ETHR_ATMC_T__ #undef ETHR_AINT_T__ #undef ETHR_AINT_SUFFIX__ -#undef ETHR_HAVE___SYNC_ADD_AND_FETCH -#undef ETHR_HAVE___SYNC_FETCH_AND_AND -#undef ETHR_HAVE___SYNC_FETCH_AND_OR #endif diff --git a/erts/include/internal/gcc/ethr_dw_atomic.h b/erts/include/internal/gcc/ethr_dw_atomic.h index 6736f9c547..c2c8f85b7b 100644 --- a/erts/include/internal/gcc/ethr_dw_atomic.h +++ b/erts/include/internal/gcc/ethr_dw_atomic.h @@ -1,7 +1,7 @@ /* * %CopyrightBegin% * - * Copyright Ericsson AB 2011. All Rights Reserved. + * Copyright Ericsson AB 2011-2015. All Rights Reserved. * * The contents of this file are subject to the Erlang Public License, * Version 1.1, (the "License"); you may not use this file except in @@ -18,35 +18,39 @@ */ /* - * Description: Native double word atomics using gcc's builtins + * Description: Native double word atomics using gcc's __atomic + * and __sync builtins * Author: Rickard Green + * + * Note: The C11 memory model implemented by gcc's __atomic + * builtins does not match the ethread API very well. + * + * Due to this we cannot use the __ATOMIC_SEQ_CST + * memory model. For more information see the comment + * in the begining of ethr_membar.h in this directory. */ #undef ETHR_INCLUDE_DW_ATOMIC_IMPL__ -#ifndef ETHR_GCC_DW_ATOMIC_H__ -# define ETHR_GCC_DW_ATOMIC_H__ -# if ((ETHR_SIZEOF_PTR == 4 \ - && defined(ETHR_HAVE___SYNC_VAL_COMPARE_AND_SWAP64)) \ - || (ETHR_SIZEOF_PTR == 8 \ - && defined(ETHR_HAVE___SYNC_VAL_COMPARE_AND_SWAP128) \ - && defined(ETHR_HAVE_INT128_T))) -# define ETHR_INCLUDE_DW_ATOMIC_IMPL__ -# endif +#if !defined(ETHR_GCC_ATOMIC_DW_ATOMIC_H__) \ + && ((ETHR_HAVE___sync_val_compare_and_swap & (2*ETHR_SIZEOF_PTR)) \ + || (ETHR_HAVE___atomic_compare_exchange_n & (2*ETHR_SIZEOF_PTR))) +# define ETHR_GCC_ATOMIC_DW_ATOMIC_H__ +# define ETHR_INCLUDE_DW_ATOMIC_IMPL__ #endif #ifdef ETHR_INCLUDE_DW_ATOMIC_IMPL__ # define ETHR_HAVE_NATIVE_SU_DW_ATOMIC -# define ETHR_NATIVE_DW_ATOMIC_IMPL "gcc" -# if defined(__i386__) || defined(__x86_64__) -/* - * If ETHR_RTCHK_USE_NATIVE_DW_ATOMIC_IMPL__ is defined, it will be used - * at runtime in order to determine if native or fallback implementation - * should be used. - */ -# define ETHR_RTCHK_USE_NATIVE_DW_ATOMIC_IMPL__ \ - ETHR_X86_RUNTIME_CONF_HAVE_DW_CMPXCHG__ -# endif +#if ((ETHR_HAVE___sync_val_compare_and_swap & (2*ETHR_SIZEOF_PTR)) \ + && (ETHR_HAVE___atomic_compare_exchange_n & (2*ETHR_SIZEOF_PTR))) +# define ETHR_NATIVE_DW_ATOMIC_IMPL "gcc_atomic_and_sync_builtins" +#elif (ETHR_HAVE___atomic_compare_exchange_n & (2*ETHR_SIZEOF_PTR)) +# define ETHR_NATIVE_DW_ATOMIC_IMPL "gcc_atomic_builtins" +#elif (ETHR_HAVE___sync_val_compare_and_swap & (2*ETHR_SIZEOF_PTR)) +# define ETHR_NATIVE_DW_ATOMIC_IMPL "gcc_sync_builtins" +#else +# error "!?" +#endif # if ETHR_SIZEOF_PTR == 4 # define ETHR_DW_NATMC_ALIGN_MASK__ 0x7 @@ -89,15 +93,138 @@ typedef union { # define ETHR_DW_DBG_ALIGNED__(PTR) # endif -#define ETHR_HAVE_ETHR_NATIVE_DW_ATOMIC_ADDR + +#define ETHR_HAVE_ETHR_NATIVE_DW_ATOMIC_ADDR 1 + static ETHR_INLINE ethr_sint_t * ethr_native_dw_atomic_addr(ethr_native_dw_atomic_t *var) { return (ethr_sint_t *) ETHR_DW_NATMC_MEM__(var); } +#if (ETHR_HAVE___atomic_store_n & (2*ETHR_SIZEOF_PTR)) + +#if (ETHR_GCC_RELAXED_VERSIONS__ & (2*ETHR_SIZEOF_PTR)) + +#define ETHR_HAVE_ETHR_NATIVE_SU_DW_ATOMIC_SET 1 + +static ETHR_INLINE void +ethr_native_su_dw_atomic_set(ethr_native_dw_atomic_t *var, + ETHR_NATIVE_SU_DW_SINT_T value) +{ + ethr_native_dw_ptr_t p = (ethr_native_dw_ptr_t) ETHR_DW_NATMC_MEM__(var); + ETHR_DW_DBG_ALIGNED__(p); + __atomic_store_n(p, value, __ATOMIC_RELAXED); +} + +#endif /* ETHR_GCC_RELAXED_VERSIONS__ */ + +#if (ETHR_GCC_RELB_VERSIONS__ & (2*ETHR_SIZEOF_PTR)) -#define ETHR_HAVE_ETHR_NATIVE_SU_DW_ATOMIC_CMPXCHG_MB +#define ETHR_HAVE_ETHR_NATIVE_SU_DW_ATOMIC_SET_RELB 1 + +static ETHR_INLINE void +ethr_native_su_dw_atomic_set_relb(ethr_native_dw_atomic_t *var, + ETHR_NATIVE_SU_DW_SINT_T value) +{ + ethr_native_dw_ptr_t p = (ethr_native_dw_ptr_t) ETHR_DW_NATMC_MEM__(var); + ETHR_DW_DBG_ALIGNED__(p); + __atomic_store_n(p, value, __ATOMIC_RELEASE); +} + +#endif /* ETHR_GCC_RELB_VERSIONS__ */ + +#endif /* ETHR_HAVE___atomic_store_n */ + +#if (ETHR_HAVE___atomic_load_n & (2*ETHR_SIZEOF_PTR)) + +#if (ETHR_GCC_RELAXED_VERSIONS__ & (2*ETHR_SIZEOF_PTR)) + +#define ETHR_HAVE_ETHR_NATIVE_SU_DW_ATOMIC_READ 1 + +static ETHR_INLINE ETHR_NATIVE_SU_DW_SINT_T +ethr_native_su_dw_atomic_read(ethr_native_dw_atomic_t *var) +{ + ethr_native_dw_ptr_t p = (ethr_native_dw_ptr_t) ETHR_DW_NATMC_MEM__(var); + ETHR_DW_DBG_ALIGNED__(p); + return __atomic_load_n(p, __ATOMIC_RELAXED); +} + +#endif /* ETHR_GCC_RELAXED_VERSIONS__ */ + +#if ((ETHR_GCC_ACQB_VERSIONS__ & (2*ETHR_SIZEOF_PTR)) \ + & ~ETHR___atomic_load_ACQUIRE_barrier_bug) + +#define ETHR_HAVE_ETHR_NATIVE_SU_DW_ATOMIC_READ_ACQB 1 + +static ETHR_INLINE ETHR_NATIVE_SU_DW_SINT_T +ethr_native_su_dw_atomic_read_acqb(ethr_native_dw_atomic_t *var) +{ + ethr_native_dw_ptr_t p = (ethr_native_dw_ptr_t) ETHR_DW_NATMC_MEM__(var); + ETHR_DW_DBG_ALIGNED__(p); + return __atomic_load_n(p, __ATOMIC_ACQUIRE); +} + +#endif /* ETHR_GCC_ACQB_VERSIONS__ */ + +#endif /* ETHR_HAVE___atomic_load_n */ + +#if (ETHR_HAVE___atomic_compare_exchange_n & (2*ETHR_SIZEOF_PTR)) + +#if (ETHR_GCC_RELAXED_MOD_VERSIONS__ & (2*ETHR_SIZEOF_PTR)) + +#define ETHR_HAVE_ETHR_NATIVE_SU_DW_ATOMIC_CMPXCHG 1 + +static ETHR_INLINE ETHR_NATIVE_SU_DW_SINT_T +ethr_native_su_dw_atomic_cmpxchg(ethr_native_dw_atomic_t *var, + ETHR_NATIVE_SU_DW_SINT_T new, + ETHR_NATIVE_SU_DW_SINT_T exp) +{ + ethr_native_dw_ptr_t p = (ethr_native_dw_ptr_t) ETHR_DW_NATMC_MEM__(var); + ETHR_NATIVE_SU_DW_SINT_T xchg = exp; + ETHR_DW_DBG_ALIGNED__(p); + if (__atomic_compare_exchange_n(p, + &xchg, + new, + 0, + __ATOMIC_RELAXED, + __ATOMIC_RELAXED)) + return exp; + return xchg; +} + +#endif /* ETHR_GCC_RELAXED_MOD_VERSIONS__ */ + +#if (ETHR_GCC_ACQB_MOD_VERSIONS__ & (2*ETHR_SIZEOF_PTR)) + +#define ETHR_HAVE_ETHR_NATIVE_SU_DW_ATOMIC_CMPXCHG_ACQB 1 + +static ETHR_INLINE ETHR_NATIVE_SU_DW_SINT_T +ethr_native_su_dw_atomic_cmpxchg_acqb(ethr_native_dw_atomic_t *var, + ETHR_NATIVE_SU_DW_SINT_T new, + ETHR_NATIVE_SU_DW_SINT_T exp) +{ + ethr_native_dw_ptr_t p = (ethr_native_dw_ptr_t) ETHR_DW_NATMC_MEM__(var); + ETHR_NATIVE_SU_DW_SINT_T xchg = exp; + ETHR_DW_DBG_ALIGNED__(p); + if (__atomic_compare_exchange_n(p, + &xchg, + new, + 0, + __ATOMIC_ACQUIRE, + __ATOMIC_ACQUIRE)) + return exp; + return xchg; +} + +#endif /* ETHR_GCC_ACQB_MOD_VERSIONS__ */ + +#endif /* ETHR_HAVE___atomic_compare_exchange_n */ + +#if ((ETHR_HAVE___sync_val_compare_and_swap & (2*ETHR_SIZEOF_PTR)) \ + & ETHR_GCC_MB_MOD_VERSIONS__) + +#define ETHR_HAVE_ETHR_NATIVE_SU_DW_ATOMIC_CMPXCHG_MB 1 static ETHR_INLINE ETHR_NATIVE_SU_DW_SINT_T ethr_native_su_dw_atomic_cmpxchg_mb(ethr_native_dw_atomic_t *var, @@ -109,7 +236,8 @@ ethr_native_su_dw_atomic_cmpxchg_mb(ethr_native_dw_atomic_t *var, return __sync_val_compare_and_swap(p, old, new); } -#endif /* ETHR_TRY_INLINE_FUNCS */ +#endif /* ETHR_HAVE___sync_val_compare_and_swap */ -#endif /* ETHR_GCC_DW_ATOMIC_H__ */ +#endif /* ETHR_TRY_INLINE_FUNCS */ +#endif /* ETHR_INCLUDE_DW_ATOMIC_IMPL__ */ diff --git a/erts/include/internal/gcc/ethr_membar.h b/erts/include/internal/gcc/ethr_membar.h index 7d428fc68e..d2d36907f3 100644 --- a/erts/include/internal/gcc/ethr_membar.h +++ b/erts/include/internal/gcc/ethr_membar.h @@ -1,7 +1,7 @@ /* * %CopyrightBegin% * - * Copyright Ericsson AB 2011. All Rights Reserved. + * Copyright Ericsson AB 2011-2015. All Rights Reserved. * * The contents of this file are subject to the Erlang Public License, * Version 1.1, (the "License"); you may not use this file except in @@ -18,56 +18,196 @@ */ /* - * Description: Memory barriers when using gcc's builtins + * Description: Memory barriers when using gcc's __atomic and + * __sync builtins * Author: Rickard Green + * + * Note: The C11 memory model implemented by gcc's __atomic + * builtins does not match the ethread API very well. + * + * A function with a barrier postfix in the ethread atomic + * API needs to ensure that all stores and loads are + * ordered around it according to the semantics of the + * barrier specified. + * + * The C11 aproch is different. The __atomic builtins + * API takes a memory model parameter. Assuming that all + * memory syncronizations using the involved atomic + * variables are made using this API, the synchronizations + * will adhere to the memory models used. That is, you do + * *not* know how loads and stores will be ordered around + * a specific __atomic operation in the general case. You + * only know the total effect of the combination of + * operations issued will adhere to the model. + * + * This limits how we can use the __atomic builtins. What + * we cannot use: + * + * 1. We cannot rely on __atomic_thread_fence() to issue + * any specific memory barriers at all. This regardless + * of memory model parameter passed. That is, we cannot + * use the __atomic_thread_fence() builtin at all. + * + * Why is this? If all __atomic builtins accessing + * memory issue memory barriers, __atomic_thread_fence() + * does not have to issue memory barriers. The + * implementation for the Itanium architecture is an + * example of this. Even using the __ATOMIC_RELAXED + * memory model all __atomic builtins accessing memory + * will issue memory barriers. Due to this no memory + * barriers at all will be issued by + * __atomic_thread_fence() using either one of the + * __ATOMIC_CONSUME, __ATOMIC_ACQUIRE, or + * __ATOMIC_RELEASE memory models. + * + * 2. We cannot rely on any __atomic builtin with the + * __ATOMIC_SEQ_CST memory model parameters to + * issue any specific memory barriers. That is, we + * cannot use these memory models at all. + * + * Why is this? Since all synchronizations is expected + * to be made using the __atomic builtins, memory + * barriers only have to be issued by some of them, + * and you do not know which ones wont issue memory + * barriers. + * + * One can easily be fooled into believing that when + * using the __ATOMIC_SEQ_CST memory model on all + * operations, all operations will issue full memory + * barriers. This is however not the case. The + * implementation for the x86_64 architecture is an + * example of this. Since all operations except loads + * issue full memory barriers, no memory barriers at + * all is issued by loads. This could also be + * implemented by issuing a full memory barrier on + * loads, but no barrier at all on stores. + * + * What can be used then? + * 1. All (legacy) __sync builtins implying full memory + * barriers issued. + * 2. All __atomic builtins using the __ATOMIC_RELAXED + * memory model can, of course, be used. This since + * no ordering guarantees at all are made. + * 3. All __atomic builtins accessing memory using the + * __ATOMIC_ACQUIRE and __ATOMIC_RELEASE memory + * models. This since an __atomic builtin memory + * access using the __ATOMIC_ACQUIRE must at least + * issue an aquire memory barrier and an __atomic + * builtin memory acess with the __ATOMIC_RELEASE + * memory model must at least issue a release memory + * barrier. Otherwise the two can not be paired. + * 4. All __atomic builtins accessing memory using the + * __ATOMIC_CONSUME builtin can be used for the same + * reason __ATOMIC_ACQUIRE can be used. The ethread + * atomic framework implementing the ethread API + * using native implementations does not expect the + * native implementations to produce versions with + * data dependent read barriers, so until the + * framework is changed we haven't got any use for + * for it. + * + * For some architectures we have our own memory barrier + * implementations. We prefer to use these since they + * should be as fine grained as possible. For other + * architectures we use the __sync_synchronize() builtin + * which issue a full memory barrier. For these + * architectures we have to assume that all loads and + * stores can be reordered without limitation. That is, + * unnecessary memory barriers will be issued if such + * reordering actually cannot occur. */ -#ifndef ETHR_GCC_MEMBAR_H__ -#define ETHR_GCC_MEMBAR_H__ +/* + * We prefer to use our own memory barrier implementation if + * such exist instead of using __sync_synchronize()... + */ +#if defined(__i386__) || defined(__x86_64__) +# include "../i386/ethr_membar.h" +#elif defined(__sparc__) +# include "../sparc32/ethr_membar.h" +#elif defined(__powerpc__) || defined(__ppc__) || defined(__powerpc64__) +# include "../ppc32/ethr_membar.h" +#elif !defined(ETHR_GCC_ATOMIC_MEMBAR_H__) \ + && (ETHR_HAVE_GCC_ASM_ARM_DMB_INSTRUCTION \ + || ETHR_HAVE___sync_synchronize \ + || (ETHR_HAVE___sync_val_compare_and_swap & 12)) +#define ETHR_GCC_ATOMIC_MEMBAR_H__ #define ETHR_LoadLoad (1 << 0) #define ETHR_LoadStore (1 << 1) #define ETHR_StoreLoad (1 << 2) #define ETHR_StoreStore (1 << 3) +#define ETHR_COMPILER_BARRIER __asm__ __volatile__("" : : : "memory") + +#if ETHR_HAVE_GCC_ASM_ARM_DMB_INSTRUCTION + +static __inline__ __attribute__((__always_inline__)) void +ethr_full_fence__(void) +{ + __asm__ __volatile__("dmb sy" : : : "memory"); +} + +static __inline__ __attribute__((__always_inline__)) void +ethr_store_fence__(void) +{ + __asm__ __volatile__("dmb st" : : : "memory"); +} + +#define ETHR_MEMBAR(B) \ + ETHR_CHOOSE_EXPR((B) == ETHR_StoreStore, ethr_store_fence__(), ethr_full_fence__()) + +#elif ETHR_HAVE___sync_synchronize + +static __inline__ __attribute__((__always_inline__)) void +ethr_full_fence__(void) +{ + /* + * The compiler barriers are here to fix missing clobbers + * in __sync_synchronize() when using buggy LLVM + * implementation of __sync_synchronize(). They + * do not introduce any unnecessary overhead when used + * here, so we use them for all systems. + */ + ETHR_COMPILER_BARRIER; + __sync_synchronize(); + ETHR_COMPILER_BARRIER; +} + +#else /* !ETHR_HAVE___sync_synchronize */ + /* - * According to the documentation __sync_synchronize() will - * issue a full memory barrier. However, __sync_synchronize() - * is known to erroneously be a noop on at least some - * platforms with some gcc versions. This has suposedly been - * fixed in some gcc version, but we don't know from which - * version. Therefore, we only use it when it has been - * verified to work. Otherwise we use the workaround - * below. + * Buggy __sync_synchronize(); call __sync_val_compare_and_swap() + * instead which imply a full memory barrier (and hope that one + * isn't buggy too). */ -#if defined(ETHR_HAVE___SYNC_VAL_COMPARE_AND_SWAP32) +#if (ETHR_HAVE___sync_val_compare_and_swap & 4) # define ETHR_MB_T__ ethr_sint32_t -#elif defined(ETHR_HAVE___SYNC_VAL_COMPARE_AND_SWAP64) +#elif (ETHR_HAVE___sync_val_compare_and_swap & 8) # define ETHR_MB_T__ ethr_sint64_t -#else -# error "No __sync_val_compare_and_swap" #endif -#define ETHR_SYNC_SYNCHRONIZE_WORKAROUND__ \ -do { \ - volatile ETHR_MB_T__ x___ = 0; \ - (void) __sync_val_compare_and_swap(&x___, (ETHR_MB_T__) 0, (ETHR_MB_T__) 1); \ -} while (0) -#define ETHR_COMPILER_BARRIER __asm__ __volatile__("" : : : "memory") +static __inline__ __attribute__((__always_inline__)) void +ethr_full_fence__(void) +{ + volatile ETHR_MB_T__ x = 0; + (void) __sync_val_compare_and_swap(&x, (ETHR_MB_T__) 0, (ETHR_MB_T__) 1); +} -#if defined(__mips__) && ETHR_AT_LEAST_GCC_VSN__(4, 2, 0) -# define ETHR_MEMBAR(B) __sync_synchronize() -# define ETHR_READ_DEPEND_MEMORY_BARRIER __sync_synchronize() -#elif ((defined(__powerpc__) || defined(__ppc__)) \ - && ETHR_AT_LEAST_GCC_VSN__(4, 1, 2)) -# define ETHR_MEMBAR(B) __sync_synchronize() -#else /* Use workaround */ -# define ETHR_MEMBAR(B) \ - ETHR_SYNC_SYNCHRONIZE_WORKAROUND__ -# define ETHR_READ_DEPEND_MEMORY_BARRIER \ - ETHR_SYNC_SYNCHRONIZE_WORKAROUND__ +#endif /* !ETHR_HAVE___sync_synchronize */ + +#ifndef ETHR_MEMBAR +# define ETHR_MEMBAR(B) ethr_full_fence__() #endif +/* + * Define ETHR_READ_DEPEND_MEMORY_BARRIER for all architechtures + * not known to order data dependent loads + */ + +#if !defined(__ia64__) && !defined(__arm__) +# define ETHR_READ_DEPEND_MEMORY_BARRIER ETHR_MEMBAR(ETHR_LoadLoad) +#endif -#endif /* ETHR_GCC_MEMBAR_H__ */ +#endif /* ETHR_GCC_ATOMIC_MEMBAR_H__ */ diff --git a/erts/include/internal/gcc/ethread.h b/erts/include/internal/gcc/ethread.h index 365a3535cf..be3e1da90e 100644 --- a/erts/include/internal/gcc/ethread.h +++ b/erts/include/internal/gcc/ethread.h @@ -1,7 +1,7 @@ /* * %CopyrightBegin% * - * Copyright Ericsson AB 2010-2011. All Rights Reserved. + * Copyright Ericsson AB 2010-2015. All Rights Reserved. * * The contents of this file are subject to the Erlang Public License, * Version 1.1, (the "License"); you may not use this file except in @@ -18,20 +18,292 @@ */ /* - * Description: Native atomic ethread support when using gcc + * Description: Native atomic ethread support when using gcc's __atomic + * and __sync builtins * Author: Rickard Green */ -#ifndef ETHREAD_GCC_H__ -#define ETHREAD_GCC_H__ - -#if defined(ETHR_HAVE___SYNC_VAL_COMPARE_AND_SWAP32) \ - || defined(ETHR_HAVE___SYNC_VAL_COMPARE_AND_SWAP64) +#if !defined(ETHREAD_GCC_NATIVE_H__) && ETHR_GCC_COMPILER +#define ETHREAD_GCC_NATIVE_H__ #ifndef ETHR_MEMBAR # include "ethr_membar.h" #endif +#define ETHR_GCC_VERSIONS_MASK__ 28 + +#undef ETHR_GCC_VOLATILE_STORE_IS_ATOMIC_STORE__ +#undef ETHR_GCC_VOLATILE_STORE_IS_ATOMIC_STORE_RELB__ +#undef ETHR_GCC_VOLATILE_LOAD_IS_ATOMIC_LOAD__ +#undef ETHR_GCC_VOLATILE_LOAD_IS_ATOMIC_LOAD_ACQB__ +#undef ETHR_GCC_RELAXED_VERSIONS__ +#undef ETHR_GCC_RELAXED_MOD_VERSIONS__ +#undef ETHR_GCC_ACQB_VERSIONS__ +#undef ETHR_GCC_ACQB_MOD_VERSIONS__ +#undef ETHR_GCC_RELB_VERSIONS__ +#undef ETHR_GCC_RELB_MOD_VERSIONS__ +#undef ETHR_GCC_MB_MOD_VERSIONS__ + +/* + * True GNU GCCs before version 4.8 do not emit a memory barrier + * after the load in the __atomic_load_n(_, __ATOMIC_ACQUIRE) + * case (which is needed on most architectures). + */ +#undef ETHR___atomic_load_ACQUIRE_barrier_bug +#if ETHR_GCC_COMPILER != ETHR_GCC_COMPILER_TRUE +/* + * A gcc compatible compiler. We have no information + * about the existence of this bug, but we assume + * that it is not impossible that it could have + * been "inherited". Therefore, until we are certain + * that the bug does not exist, we assume that it + * does. + */ +# define ETHR___atomic_load_ACQUIRE_barrier_bug ETHR_GCC_VERSIONS_MASK__ +#elif !ETHR_AT_LEAST_GCC_VSN__(4, 8, 0) +/* True gcc of version < 4.8, i.e., bug exist... */ +# define ETHR___atomic_load_ACQUIRE_barrier_bug ETHR_GCC_VERSIONS_MASK__ +#else /* True gcc of version >= 4.8 */ +/* + * Sizes less than or equal to word size have been fixed, + * but double word size has not been fixed. + */ +# if ETHR_SIZEOF_PTR == 8 +# define ETHR___atomic_load_ACQUIRE_barrier_bug \ + (~(8|4) & ETHR_GCC_VERSIONS_MASK__) +# elif ETHR_SIZEOF_PTR == 4 +# define ETHR___atomic_load_ACQUIRE_barrier_bug \ + (~4 & ETHR_GCC_VERSIONS_MASK__) +# else +# error word size not supported +# endif +#endif + +#define ETHR_GCC_VOLATILE_STORE_IS_ATOMIC_STORE__ 0 +#define ETHR_GCC_VOLATILE_STORE_IS_ATOMIC_STORE_RELB__ 0 +#define ETHR_GCC_VOLATILE_LOAD_IS_ATOMIC_LOAD__ 0 +#define ETHR_GCC_VOLATILE_LOAD_IS_ATOMIC_LOAD_ACQB__ 0 +#define ETHR_GCC_RELAXED_VERSIONS__ ETHR_GCC_VERSIONS_MASK__ +#define ETHR_GCC_RELAXED_MOD_VERSIONS__ ETHR_GCC_VERSIONS_MASK__ + +#if ETHR_TRUST_GCC_ATOMIC_BUILTINS_MEMORY_BARRIERS +# define ETHR_GCC_ACQB_VERSIONS__ ETHR_GCC_VERSIONS_MASK__ +# define ETHR_GCC_ACQB_MOD_VERSIONS__ ETHR_GCC_VERSIONS_MASK__ +# define ETHR_GCC_RELB_VERSIONS__ ETHR_GCC_VERSIONS_MASK__ +# define ETHR_GCC_RELB_MOD_VERSIONS__ ETHR_GCC_VERSIONS_MASK__ +#else +/* + * This is currently the default (on most platforms) since + * we've seen too many memory barrier bugs produced by gcc... + */ +# define ETHR_GCC_ACQB_VERSIONS__ 0 +# define ETHR_GCC_ACQB_MOD_VERSIONS__ 0 +# define ETHR_GCC_RELB_VERSIONS__ 0 +# define ETHR_GCC_RELB_MOD_VERSIONS__ 0 +#endif +/* + * In the general case we do not want any full barrier versions + * if we can implement more relaxed ones (using __atomic_* builtins). + * This since the implementations normally need extra memory barrier + * instructions to implement these. The x86/x86_64 implementations + * are an exception see below. + */ +#define ETHR_GCC_MB_MOD_VERSIONS__ \ + (ETHR_GCC_VERSIONS_MASK__ & ~ETHR_HAVE___atomic_compare_exchange_n) + +#if ETHR_SIZEOF_PTR == 8 +# define ETHR_GCC_VOLATILE_BIT_MASK__ 12 +#elif ETHR_SIZEOF_PTR == 4 +# define ETHR_GCC_VOLATILE_BIT_MASK__ 4 +#endif + +#if defined(__i386__) || defined(__x86_64__) || defined(__sparc__) \ + || defined(__powerpc__) || defined(__ppc__) || defined(__mips__) \ + || defined(__alpha__) || defined(__ia64__) + +/* + * Aligned volatile stores and loads of data smaller + * than or equal to word size are atomic... + */ +# undef ETHR_GCC_VOLATILE_STORE_IS_ATOMIC_STORE__ +# define ETHR_GCC_VOLATILE_STORE_IS_ATOMIC_STORE__ ETHR_GCC_VOLATILE_BIT_MASK__ +# undef ETHR_GCC_VOLATILE_LOAD_IS_ATOMIC_LOAD__ +# define ETHR_GCC_VOLATILE_LOAD_IS_ATOMIC_LOAD__ ETHR_GCC_VOLATILE_BIT_MASK__ + +#elif defined(__arm__) + +/* volatile stores are problematic on some machines... */ +# undef ETHR_GCC_VOLATILE_LOAD_IS_ATOMIC_LOAD__ +# define ETHR_GCC_VOLATILE_LOAD_IS_ATOMIC_LOAD__ ETHR_GCC_VOLATILE_BIT_MASK__ + +#endif + +#if defined(__ia64__) + +/* Volatile stores produce stores with release barriers. */ +# undef ETHR_GCC_VOLATILE_STORE_IS_ATOMIC_STORE_RELB__ +# define ETHR_GCC_VOLATILE_STORE_IS_ATOMIC_STORE_RELB__ ETHR_GCC_VOLATILE_BIT_MASK__ + +/* Volatile loads produce loads with acquire barrier. */ +# undef ETHR_GCC_VOLATILE_LOAD_IS_ATOMIC_LOAD_ACQB__ +# define ETHR_GCC_VOLATILE_LOAD_IS_ATOMIC_LOAD_ACQB__ ETHR_GCC_VOLATILE_BIT_MASK__ + +/* + * We trust gcc to produce acquire/release barriers on itanium. + * Since all atomic ops also have at least acquire or release + * barriers (also when passed the relaxed memory model) it + * would be very inefficient not to use these as native + * barriers on Itanium. + */ +# undef ETHR_GCC_ACQB_VERSIONS__ +# define ETHR_GCC_ACQB_VERSIONS__ ETHR_GCC_VERSIONS_MASK__ +# undef ETHR_GCC_ACQB_MOD_VERSIONS__ +# define ETHR_GCC_ACQB_MOD_VERSIONS__ ETHR_GCC_VERSIONS_MASK__ +# undef ETHR_GCC_RELB_VERSIONS__ +# define ETHR_GCC_RELB_VERSIONS__ ETHR_GCC_VERSIONS_MASK__ +# undef ETHR_GCC_RELB_MOD_VERSIONS__ +# define ETHR_GCC_RELB_MOD_VERSIONS__ ETHR_GCC_VERSIONS_MASK__ + +/* + * Itanium is not effected by the load acquire + * bug since the barrier is part of the instruction + * on Itanium (ld.acq), and not a separate instruction + * as on most platforms. + */ +# undef ETHR___atomic_load_ACQUIRE_barrier_bug +# define ETHR___atomic_load_ACQUIRE_barrier_bug 0 + +/* + * No point exposing relaxed versions since they are + * implemended using either acquire or release + * barriers. + */ +# undef ETHR_GCC_RELAXED_VERSIONS__ +# define ETHR_GCC_RELAXED_VERSIONS__ 0 + +/* #endif defined(__ia64__) */ +#elif defined(__i386__) || defined(__x86_64__) + +/* + * Want full barrier versions of all modification + * operations since all of these are implemented + * using locked instructions implying full memory + * barriers. + */ +# undef ETHR_GCC_MB_MOD_VERSIONS__ +# define ETHR_GCC_MB_MOD_VERSIONS__ ETHR_HAVE___sync_val_compare_and_swap + +/* + * No point exposing acquire/release versions + * when we got full memory barrier versions + * of modification operations since all of these + * are implemented using locked instructions + * implying full memory barriers. + */ +# if ETHR_GCC_ACQB_MOD_VERSIONS__ +# undef ETHR_GCC_ACQB_MOD_VERSIONS__ +# define ETHR_GCC_ACQB_MOD_VERSIONS__ \ + (ETHR_GCC_VERSIONS_MASK__ & ~ETHR_HAVE___sync_val_compare_and_swap) +# endif +# if ETHR_GCC_RELB_MOD_VERSIONS__ +# undef ETHR_GCC_RELB_MOD_VERSIONS__ +# define ETHR_GCC_RELB_MOD_VERSIONS__ \ + (ETHR_GCC_VERSIONS_MASK__ & ~ETHR_HAVE___sync_val_compare_and_swap) +# endif + +# ifdef ETHR_X86_OUT_OF_ORDER + +/* See above... */ +# undef ETHR_GCC_RELAXED_MOD_VERSIONS__ +# define ETHR_GCC_RELAXED_MOD_VERSIONS__ 0 + +# else /* !ETHR_X86_OUT_OF_ORDER, i.e., we don't use any x86-OOO instructions... */ + +/* + * Not effected by the load acquire barrier bug, + * since no barrier at all is needed for a load + * acquire... + */ +# undef ETHR___atomic_load_ACQUIRE_barrier_bug +# define ETHR___atomic_load_ACQUIRE_barrier_bug 0 + +/* Stores imply release barriers semantics. */ +# undef ETHR_GCC_VOLATILE_STORE_IS_ATOMIC_STORE_RELB__ +# define ETHR_GCC_VOLATILE_STORE_IS_ATOMIC_STORE_RELB__ ETHR_GCC_VOLATILE_BIT_MASK__ + +/* Loads imply acquire barrier semantics. */ +# undef ETHR_GCC_VOLATILE_LOAD_IS_ATOMIC_LOAD_ACQB__ +# define ETHR_GCC_VOLATILE_LOAD_IS_ATOMIC_LOAD_ACQB__ ETHR_GCC_VOLATILE_BIT_MASK__ + +/* + * Trust load acquire and store release for sizes + * where volatile operation implies these barrier + * semantics since no barriers are needed. + */ +# if !ETHR_GCC_ACQB_VERSIONS__ +# undef ETHR_GCC_ACQB_VERSIONS__ +# define ETHR_GCC_ACQB_VERSIONS__ ETHR_GCC_VOLATILE_BIT_MASK__ +# endif +# if !ETHR_GCC_RELB_VERSIONS__ +# undef ETHR_GCC_RELB_VERSIONS__ +# define ETHR_GCC_RELB_VERSIONS__ ETHR_GCC_VOLATILE_BIT_MASK__ +# endif + +/* + * No point exposing relaxed versions at all since + * all mod operations are implemented with locked + * instructions implying full memory barriers and + * volatile store and load imply release and + * acquire barrier semantics. + */ +# undef ETHR_GCC_RELAXED_VERSIONS__ +# define ETHR_GCC_RELAXED_VERSIONS__ 0 + +# endif /* !ETHR_X86_OUT_OF_ORDER */ + +/* #endif defined(__i386__) || defined(__x86_64__) */ +#elif defined(__powerpc__) || defined(__ppc__) + +# if !defined(ETHR_PPC_HAVE_LWSYNC) +/* + * Release barriers are typically implemented using + * the lwsync instruction. We want our runtime + * configure test to determine if the lwsync + * instruction is available on the system or not + * before we use it. Therefore, do not implement any + * native ops using the __ATOMIC_RELEASE model. + */ +# undef ETHR_GCC_RELB_VERSIONS__ +# define ETHR_GCC_RELB_VERSIONS__ 0 +# if defined(ETHR_GCC_IMPLEMENT_ACQB_USING_LWSYNC) +/* + * Acquire barriers are usually implemented by other means + * than lwsync, but can be implemented using lwsync. Define + * ETHR_GCC_IMPLEMENT_ACQB_USING_LWSYNC if acquire barriers + * are implemented using lwsync. + */ +# undef ETHR_GCC_ACQB_VERSIONS__ +# define ETHR_GCC_ACQB_VERSIONS__ 0 +# endif +# endif + +#endif /* defined(__powerpc__) || defined(__ppc__) */ + +#if !ETHR_GCC_RELAXED_VERSIONS__ +# undef ETHR_GCC_RELAXED_MOD_VERSIONS__ +# define ETHR_GCC_RELAXED_MOD_VERSIONS__ 0 +#endif + +#if !ETHR_GCC_ACQB_VERSIONS__ +# undef ETHR_GCC_ACQB_MOD_VERSIONS__ +# define ETHR_GCC_ACQB_MOD_VERSIONS__ 0 +#endif + +#if !ETHR_GCC_RELB_VERSIONS__ +# undef ETHR_GCC_RELB_MOD_VERSIONS__ +# define ETHR_GCC_RELB_MOD_VERSIONS__ 0 +#endif + #if !defined(ETHR_HAVE_NATIVE_ATOMIC32) # define ETHR_ATOMIC_WANT_32BIT_IMPL__ # include "ethr_atomic.h" @@ -42,12 +314,51 @@ # include "ethr_atomic.h" #endif +#if defined(__x86_64__) +/* + * No instructions available for native implementation + * of these for dw-atomics... + */ +# undef ETHR_GCC_RELAXED_VERSIONS__ +# define ETHR_GCC_RELAXED_VERSIONS__ 0 +# undef ETHR_GCC_ACQB_VERSIONS__ +# define ETHR_GCC_ACQB_VERSIONS__ 0 +# undef ETHR_GCC_RELB_VERSIONS__ +# define ETHR_GCC_RELB_VERSIONS__ 0 +#endif + +#if !ETHR_GCC_RELAXED_VERSIONS__ +# undef ETHR_GCC_RELAXED_MOD_VERSIONS__ +# define ETHR_GCC_RELAXED_MOD_VERSIONS__ 0 +#endif + +#if !ETHR_GCC_ACQB_VERSIONS__ +# undef ETHR_GCC_ACQB_MOD_VERSIONS__ +# define ETHR_GCC_ACQB_MOD_VERSIONS__ 0 +#endif + +#if !ETHR_GCC_RELB_VERSIONS__ +# undef ETHR_GCC_RELB_MOD_VERSIONS__ +# define ETHR_GCC_RELB_MOD_VERSIONS__ 0 +#endif + #if (!defined(ETHR_HAVE_NATIVE_DW_ATOMIC) \ && !(ETHR_SIZEOF_PTR == 4 && defined(ETHR_HAVE_NATIVE_ATOMIC64)) \ && !(ETHR_SIZEOF_PTR == 8 && defined(ETHR_HAVE_NATIVE_ATOMIC128))) # include "ethr_dw_atomic.h" #endif -#endif +#undef ETHR___atomic_load_ACQUIRE_barrier_bug +#undef ETHR_GCC_VOLATILE_STORE_IS_ATOMIC_STORE__ +#undef ETHR_GCC_VOLATILE_STORE_IS_ATOMIC_STORE_RELB__ +#undef ETHR_GCC_VOLATILE_LOAD_IS_ATOMIC_LOAD__ +#undef ETHR_GCC_VOLATILE_LOAD_IS_ATOMIC_LOAD_ACQB__ +#undef ETHR_GCC_RELAXED_VERSIONS__ +#undef ETHR_GCC_RELB_VERSIONS__ +#undef ETHR_GCC_RELB_VERSIONS__ +#undef ETHR_GCC_RELAXED_MOD_VERSIONS__ +#undef ETHR_GCC_ACQB_MOD_VERSIONS__ +#undef ETHR_GCC_RELB_MOD_VERSIONS__ +#undef ETHR_GCC_MB_MOD_VERSIONS__ -#endif +#endif /* ETHREAD_GCC_NATIVE_H__ */ diff --git a/erts/lib_src/Makefile.in b/erts/lib_src/Makefile.in index b680c03b1d..d0ebab49d8 100644 --- a/erts/lib_src/Makefile.in +++ b/erts/lib_src/Makefile.in @@ -92,6 +92,11 @@ CFLAGS += -DERTS_FRMPTR OMIT_OMIT_FP=yes PRE_LD= else +ifeq ($(TYPE),icount) +TYPE_SUFFIX = .icount +CFLAGS += -DERTS_OPCODE_COUNTER_SUPPORT +PRE_LD= +else override TYPE=opt OMIT_FP=true TYPE_SUFFIX= @@ -105,6 +110,7 @@ endif endif endif endif +endif OPSYS=@OPSYS@ sol2CFLAGS= diff --git a/erts/lib_src/common/erl_printf.c b/erts/lib_src/common/erl_printf.c index 2c177ee5ac..a38017b62f 100644 --- a/erts/lib_src/common/erl_printf.c +++ b/erts/lib_src/common/erl_printf.c @@ -87,6 +87,41 @@ void (*erts_printf_unblock_fpe)(int) = NULL; # define FWRITE fwrite #endif +/* We use write for stdout and stderr as they could be + set to non-blocking by shell drivers, and non-blocking + FILE * functions work unpredictably as best */ +static int +printf_putc(int c, FILE *stream) { + if ((FILE*)stream == stdout || (FILE*)stream == stderr) { + int fd = stream == stdout ? fileno(stdout) : fileno(stderr); + /* cast to a char here, because write expects bytes. */ + unsigned char buf[1] = { c }; + int res; + do { + res = write(fd, buf, 1); + } while (res == -1 && (errno == EAGAIN || errno == EINTR)); + if (res == -1) return EOF; + return res; + } + + return PUTC(c, stream); +} + +static size_t +printf_fwrite(const void *ptr, size_t size, size_t nitems, + FILE *stream) { + if ((FILE*)stream == stdout || (FILE*)stream == stderr) { + int fd = stream == stdout ? fileno(stdout) : fileno(stderr); + int res; + do { + res = write(fd, ptr, size*nitems); + } while (res == -1 && (errno == EAGAIN || errno == EINTR)); + if (res == -1) return 0; + return res; + } + return FWRITE(ptr, size, nitems, stream); +} + static int get_error_result(void) { @@ -103,10 +138,10 @@ write_f_add_cr(void *vfp, char* buf, size_t len) size_t i; ASSERT(vfp); for (i = 0; i < len; i++) { - if (buf[i] == '\n' && PUTC('\r', (FILE *) vfp) == EOF) - return get_error_result(); - if (PUTC(buf[i], (FILE *) vfp) == EOF) - return get_error_result(); + if (buf[i] == '\n' && printf_putc('\r', (FILE *) vfp) == EOF) + return get_error_result(); + if (printf_putc(buf[i], (FILE *) vfp) == EOF) + return get_error_result(); } return len; } @@ -119,12 +154,12 @@ write_f(void *vfp, char* buf, size_t len) if (len <= 64) { /* Try to optimize writes of small bufs. */ int i; for (i = 0; i < len; i++) - if (PUTC(buf[i], (FILE *) vfp) == EOF) + if (printf_putc(buf[i], (FILE *) vfp) == EOF) return get_error_result(); } else #endif - if (FWRITE((void *) buf, sizeof(char), len, (FILE *) vfp) != len) + if (printf_fwrite((void *) buf, sizeof(char), len, (FILE *) vfp) != len) return get_error_result(); return len; } diff --git a/erts/lib_src/pthread/ethread.c b/erts/lib_src/pthread/ethread.c index 42d2abd3f1..c0b1dad0b6 100644 --- a/erts/lib_src/pthread/ethread.c +++ b/erts/lib_src/pthread/ethread.c @@ -42,6 +42,7 @@ #include <sys/types.h> #include <unistd.h> #include <signal.h> +#include <string.h> #include <limits.h> @@ -79,6 +80,8 @@ typedef struct { void *(*thr_func)(void *); void *arg; void *prep_func_res; + char *name; + char name_buff[16]; } ethr_thr_wrap_data__; static void *thr_wrapper(void *vtwd) @@ -100,6 +103,8 @@ static void *thr_wrapper(void *vtwd) tsep = twd->tse; /* We aren't allowed to follow twd after result has been set! */ + if (twd->name) + ethr_setname(twd->name); ethr_atomic32_set(&twd->result, result); @@ -325,6 +330,12 @@ ethr_thr_create(ethr_tid *tid, void * (*func)(void *), void *arg, twd.thr_func = func; twd.arg = arg; + if (opts && opts->name) { + snprintf(twd.name_buff, 16, "%s", opts->name); + twd.name = twd.name_buff; + } else + twd.name = NULL; + res = pthread_attr_init(&attr); if (res != 0) return res; @@ -455,6 +466,30 @@ ethr_self(void) } int +ethr_getname(ethr_tid tid, char *buf, size_t len) +{ +#if defined(ETHR_HAVE_PTHREAD_GETNAME_NP_3) + return pthread_getname_np((pthread_t) tid, buf, len); +#elif defined(ETHR_HAVE_PTHREAD_GETNAME_NP_2) + return pthread_getname_np((pthread_t) tid, buf); +#else + return ENOSYS; +#endif +} + +void +ethr_setname(char *name) +{ +#if defined(ETHR_HAVE_PTHREAD_SETNAME_NP_2) + pthread_setname_np(ethr_self(), name); +#elif defined(ETHR_HAVE_PTHREAD_SET_NAME_NP_2) + pthread_set_name_np(ethr_self(), name); +#elif defined(ETHR_HAVE_PTHREAD_SETNAME_NP_1) + pthread_setname_np(name); +#endif +} + +int ethr_equal_tids(ethr_tid tid1, ethr_tid tid2) { return pthread_equal((pthread_t) tid1, (pthread_t) tid2); @@ -575,6 +610,17 @@ int ethr_sigwait(const sigset_t *set, int *sig) return 0; } +int ethr_kill(const ethr_tid tid, const int sig) +{ +#if ETHR_XCHK + if (ethr_not_inited__) { + ETHR_ASSERT(0); + return EACCES; + } +#endif + return pthread_kill((const pthread_t)tid, sig); +} + #endif /* #if ETHR_HAVE_ETHR_SIG_FUNCS */ #ifdef ETHR_HAVE_ETHR_GET_MONOTONIC_TIME diff --git a/erts/lib_src/win/ethread.c b/erts/lib_src/win/ethread.c index 14d0b6deff..fe5d4a327f 100644 --- a/erts/lib_src/win/ethread.c +++ b/erts/lib_src/win/ethread.c @@ -508,6 +508,19 @@ ethr_self(void) return *tid; } +/* getname and setname are not available on windows */ +int +ethr_getname(ethr_tid tid, char *buf, size_t len) +{ + return ENOSYS; +} + +void +ethr_setname(char *name) +{ + return; +} + int ethr_equal_tids(ethr_tid tid1, ethr_tid tid2) { diff --git a/erts/preloaded/ebin/erl_prim_loader.beam b/erts/preloaded/ebin/erl_prim_loader.beam Binary files differindex 10138cab22..c8ec111e57 100644 --- a/erts/preloaded/ebin/erl_prim_loader.beam +++ b/erts/preloaded/ebin/erl_prim_loader.beam diff --git a/erts/preloaded/ebin/erlang.beam b/erts/preloaded/ebin/erlang.beam Binary files differindex 303f3f47b6..7238dabf40 100644 --- a/erts/preloaded/ebin/erlang.beam +++ b/erts/preloaded/ebin/erlang.beam diff --git a/erts/preloaded/ebin/erts_internal.beam b/erts/preloaded/ebin/erts_internal.beam Binary files differindex 68fb357eb4..ad43962b8b 100644 --- a/erts/preloaded/ebin/erts_internal.beam +++ b/erts/preloaded/ebin/erts_internal.beam diff --git a/erts/preloaded/ebin/init.beam b/erts/preloaded/ebin/init.beam Binary files differindex 7f79493b4d..f196952ef2 100644 --- a/erts/preloaded/ebin/init.beam +++ b/erts/preloaded/ebin/init.beam diff --git a/erts/preloaded/ebin/otp_ring0.beam b/erts/preloaded/ebin/otp_ring0.beam Binary files differindex 621f036f33..4d22d8bace 100644 --- a/erts/preloaded/ebin/otp_ring0.beam +++ b/erts/preloaded/ebin/otp_ring0.beam diff --git a/erts/preloaded/ebin/prim_eval.beam b/erts/preloaded/ebin/prim_eval.beam Binary files differindex 6695f4cd15..efc8347b6e 100644 --- a/erts/preloaded/ebin/prim_eval.beam +++ b/erts/preloaded/ebin/prim_eval.beam diff --git a/erts/preloaded/ebin/prim_file.beam b/erts/preloaded/ebin/prim_file.beam Binary files differindex 5e9230826d..6c49b5185e 100644 --- a/erts/preloaded/ebin/prim_file.beam +++ b/erts/preloaded/ebin/prim_file.beam diff --git a/erts/preloaded/ebin/prim_inet.beam b/erts/preloaded/ebin/prim_inet.beam Binary files differindex 83230d42ae..f58ee4b4d5 100644 --- a/erts/preloaded/ebin/prim_inet.beam +++ b/erts/preloaded/ebin/prim_inet.beam diff --git a/erts/preloaded/ebin/prim_zip.beam b/erts/preloaded/ebin/prim_zip.beam Binary files differindex b3abba3223..73be297bbb 100644 --- a/erts/preloaded/ebin/prim_zip.beam +++ b/erts/preloaded/ebin/prim_zip.beam diff --git a/erts/preloaded/ebin/zlib.beam b/erts/preloaded/ebin/zlib.beam Binary files differindex 7006764d96..9eaf8b9e59 100644 --- a/erts/preloaded/ebin/zlib.beam +++ b/erts/preloaded/ebin/zlib.beam diff --git a/erts/preloaded/src/erlang.erl b/erts/preloaded/src/erlang.erl index 83010b17d2..425b59790e 100644 --- a/erts/preloaded/src/erlang.erl +++ b/erts/preloaded/src/erlang.erl @@ -101,7 +101,7 @@ float_to_list/1, float_to_list/2]). -export([fun_info/2, fun_info_mfa/1, fun_to_list/1, function_exported/3]). -export([garbage_collect/0, garbage_collect/1, garbage_collect/2]). --export([garbage_collect_message_area/0, get/0, get/1, get_keys/1]). +-export([garbage_collect_message_area/0, get/0, get/1, get_keys/0, get_keys/1]). -export([get_module_info/1, get_stacktrace/0, group_leader/0]). -export([group_leader/2, halt/0, halt/1, halt/2, hash/2, hibernate/3]). -export([insert_element/3]). @@ -1014,6 +1014,12 @@ get() -> get(_Key) -> erlang:nif_error(undefined). +%% get_keys/0 +-spec get_keys() -> [Key] when + Key :: term(). +get_keys() -> + erlang:nif_error(undefined). + %% get_keys/1 -spec get_keys(Val) -> [Key] when Val :: term(), @@ -1925,7 +1931,7 @@ element(_N, _Tuple) -> %% Not documented -spec erlang:get_module_info(Module, Item) -> ModuleInfo when Module :: atom(), - Item :: module | imports | exports | functions | attributes | compile | native_addresses, + Item :: module | imports | exports | functions | attributes | compile | native_addresses | md5, ModuleInfo :: atom() | [] | [{atom(), arity()}] | [{atom(), term()}] | [{atom(), arity(), integer()}]. get_module_info(_Module, _Item) -> erlang:nif_error(undefined). diff --git a/erts/preloaded/src/erts_internal.erl b/erts/preloaded/src/erts_internal.erl index 2c701d75e4..e2c3d40f61 100644 --- a/erts/preloaded/src/erts_internal.erl +++ b/erts/preloaded/src/erts_internal.erl @@ -30,7 +30,7 @@ -export([await_port_send_result/3]). -export([cmp_term/2]). --export([map_to_tuple_keys/1]). +-export([map_to_tuple_keys/1, map_type/1, map_hashmap_children/1]). -export([port_command/3, port_connect/2, port_close/1, port_control/3, port_call/3, port_info/1, port_info/2]). @@ -191,6 +191,23 @@ cmp_term(_A,_B) -> map_to_tuple_keys(_M) -> erlang:nif_error(undefined). +%% return the internal map type +-spec map_type(M) -> Type when + M :: map(), + Type :: 'flatmap' | 'hashmap' | 'hashmap_node'. + +map_type(_M) -> + erlang:nif_error(undefined). + +%% return the internal hashmap sub-nodes from +%% a hashmap node +-spec map_hashmap_children(M) -> Children when + M :: map(), %% hashmap node + Children :: [map() | nonempty_improper_list(term(),term())]. + +map_hashmap_children(_M) -> + erlang:nif_error(undefined). + -spec erts_internal:flush_monitor_messages(Ref, Multi, Res) -> term() when Ref :: reference(), Multi :: boolean(), diff --git a/erts/preloaded/src/zlib.erl b/erts/preloaded/src/zlib.erl index df7b2e6198..5d9f90ec58 100644 --- a/erts/preloaded/src/zlib.erl +++ b/erts/preloaded/src/zlib.erl @@ -24,13 +24,14 @@ deflate/2,deflate/3,deflateEnd/1, inflateInit/1,inflateInit/2,inflateSetDictionary/2, inflateSync/1,inflateReset/1,inflate/2,inflateEnd/1, + inflateChunk/1, inflateChunk/2, setBufSize/2,getBufSize/1, crc32/1,crc32/2,crc32/3,adler32/2,adler32/3,getQSize/1, crc32_combine/4,adler32_combine/4, compress/1,uncompress/1,zip/1,unzip/1, gzip/1,gunzip/1]). --export_type([zstream/0]). +-export_type([zstream/0, zlevel/0, zwindowbits/0, zmemlevel/0, zstrategy/0]). %% flush argument encoding -define(Z_NO_FLUSH, 0). @@ -100,6 +101,7 @@ -define(INFLATE_RESET, 12). -define(INFLATE_END, 13). -define(INFLATE, 14). +-define(INFLATE_CHUNK, 25). -define(CRC32_0, 15). -define(CRC32_1, 16). @@ -124,7 +126,7 @@ -type zlevel() :: 'none' | 'default' | 'best_compression' | 'best_speed' | 0..9. -type zmethod() :: 'deflated'. --type zwindowbits() :: -15..-9 | 9..47. +-type zwindowbits() :: -15..-8 | 8..47. -type zmemlevel() :: 1..9. -type zstrategy() :: 'default' | 'filtered' | 'huffman_only' | 'rle'. @@ -263,6 +265,39 @@ inflate(Z, Data) -> erlang:error(badarg) end. +-spec inflateChunk(Z, Data) -> Decompressed | {more, Decompressed} when + Z :: zstream(), + Data :: iodata(), + Decompressed :: iolist(). +inflateChunk(Z, Data) -> + try port_command(Z, Data) of + true -> + inflateChunk(Z) + catch + error:_Err -> + flush(Z), + erlang:error(badarg) + end. + +-spec inflateChunk(Z) -> Decompressed | {more, Decompressed} when + Z :: zstream(), + Decompressed :: iolist(). +inflateChunk(Z) -> + Status = call(Z, ?INFLATE_CHUNK, []), + Data = receive + {Z, {data, Bin}} -> + Bin + after 0 -> + [] + end, + + case Status of + Good when (Good == ok) orelse (Good == stream_end) -> + Data; + inflate_has_more -> + {more, Data} + end. + -spec inflateEnd(Z) -> 'ok' when Z :: zstream(). inflateEnd(Z) -> @@ -496,8 +531,8 @@ arg_method(_) -> erlang:error(badarg). -spec arg_bitsz(zwindowbits()) -> zwindowbits(). arg_bitsz(Bits) when is_integer(Bits) andalso - ((8 < Bits andalso Bits < 48) orelse - (-15 =< Bits andalso Bits < -8)) -> + ((8 =< Bits andalso Bits < 48) orelse + (-15 =< Bits andalso Bits =< -8)) -> Bits; arg_bitsz(_) -> erlang:error(badarg). @@ -514,7 +549,9 @@ call(Z, Cmd, Arg) -> [2,A,B,C,D] -> (A bsl 24)+(B bsl 16)+(C bsl 8)+D; [3,A,B,C,D] -> - erlang:error({need_dictionary,(A bsl 24)+(B bsl 16)+(C bsl 8)+D}) + erlang:error({need_dictionary,(A bsl 24)+(B bsl 16)+(C bsl 8)+D}); + [4, _, _, _, _] -> + inflate_has_more catch error:badarg -> %% Rethrow loses port_control from stacktrace. erlang:error(badarg) diff --git a/erts/test/erlc_SUITE_data/src/start_ok.script b/erts/test/erlc_SUITE_data/src/start_ok.script index 4cd89f0439..7ef97dc3f3 100644 --- a/erts/test/erlc_SUITE_data/src/start_ok.script +++ b/erts/test/erlc_SUITE_data/src/start_ok.script @@ -52,7 +52,6 @@ shell_default, timer, gen_fsm, - pg, unix, dict, pool, @@ -156,7 +155,6 @@ {timer,1}, {gen_fsm,1}, {io_lib_pretty,1}, - {pg,1}, {slave,1}, {unix,1}, {dict,1}, diff --git a/erts/test/otp_SUITE.erl b/erts/test/otp_SUITE.erl index ebfdf8a830..ba71100081 100644 --- a/erts/test/otp_SUITE.erl +++ b/erts/test/otp_SUITE.erl @@ -1,7 +1,7 @@ %% %% %CopyrightBegin% %% -%% Copyright Ericsson AB 2000-2013. All Rights Reserved. +%% Copyright Ericsson AB 2000-2015. All Rights Reserved. %% %% The contents of this file are subject to the Erlang Public License, %% Version 1.1, (the "License"); you may not use this file except in @@ -94,7 +94,9 @@ undefined_functions(Config) when is_list(Config) -> Undef4 = eunit_filter(Undef3), Undef5 = dialyzer_filter(Undef4), Undef6 = wx_filter(Undef5), - Undef = gs_filter(Undef6), + Undef7 = gs_filter(Undef6), + Undef8 = diameter_filter(Undef7), + Undef = ssh_filter(Undef8), case Undef of [] -> ok; @@ -217,6 +219,28 @@ gs_filter(Undef) -> _ -> Undef end. +diameter_filter(Undef) -> + %% Filter away function calls that are catched for OTP 18 time API + filter(fun({{diameter_lib,_,_},{erlang,convert_time_resolution,3}}) -> + false; + ({{diameter_lib,_,_},{erlang,monotonic_time,0}}) -> + false; + ({{diameter_lib,_,_},{erlang,time_resolution,0}}) -> + false; + ({{diameter_lib,_,_},{erlang,unique_integer,0}}) -> + false; + ({{diameter_lib,_,_},{erlang,time_offset,0}}) -> + false; + (_) -> true + end, Undef). + +ssh_filter(Undef) -> + %% Filter away function calls that are catched for OTP 18 time API + filter(fun({{ssh_info,_,_},{erlang,timestamp,0}}) -> + false; + (_) -> true + end, Undef). + deprecated_not_in_obsolete(Config) when is_list(Config) -> ?line Server = ?config(xref_server, Config), ?line {ok,DeprecatedFunctions} = xref:q(Server, "DF"), diff --git a/erts/test/upgrade_SUITE.erl b/erts/test/upgrade_SUITE.erl index 217056971e..4858bfdfb7 100644 --- a/erts/test/upgrade_SUITE.erl +++ b/erts/test/upgrade_SUITE.erl @@ -238,7 +238,10 @@ do_upgrade(FromVsn,FromApps,ToRel,ToApps,InstallDir) -> [{"OTP upgrade test",FromVsn,_,permanent}] = rpc:call(Node,release_handler,which_releases,[]), - {ok,ToVsn} = rpc:call(Node,release_handler,unpack_release,[ToRel]), + ToRelName = filename:basename(ToRel), + copy_file(ToRel++".tar.gz", + filename:join([InstallDir,releases,ToRelName++".tar.gz"])), + {ok,ToVsn} = rpc:call(Node,release_handler,unpack_release,[ToRelName]), [{"OTP upgrade test",ToVsn,_,unpacked}, {"OTP upgrade test",FromVsn,_,permanent}] = rpc:call(Node,release_handler,which_releases,[]), |