diff options
Diffstat (limited to 'erts')
114 files changed, 5652 insertions, 1255 deletions
diff --git a/erts/aclocal.m4 b/erts/aclocal.m4 index ed492d55ff..d78025b0be 100644 --- a/erts/aclocal.m4 +++ b/erts/aclocal.m4 @@ -1421,9 +1421,31 @@ case "$THR_LIB_NAME" in int z; AO_nop_full(); +#if defined(AO_HAVE_store) AO_store(&x, (AO_t) 0); +#elif defined(AO_HAVE_store_release) + AO_store_release(&x, (AO_t) 0); +#else +#error No store +#endif +#if defined(AO_HAVE_load) z = AO_load(&x); +#elif defined(AO_HAVE_load_acquire) + z = AO_load_acquire(&x); +#else +#error No load +#endif +#if defined(AO_HAVE_compare_and_swap_full) z = AO_compare_and_swap_full(&x, (AO_t) 0, (AO_t) 1); +#elif defined(AO_HAVE_compare_and_swap_release) + z = AO_compare_and_swap_release(&x, (AO_t) 0, (AO_t) 1); +#elif defined(AO_HAVE_compare_and_swap_acquire) + z = AO_compare_and_swap_acquire(&x, (AO_t) 0, (AO_t) 1); +#elif defined(AO_HAVE_compare_and_swap) + z = AO_compare_and_swap(&x, (AO_t) 0, (AO_t) 1); +#else +#error No compare_and_swap +#endif ], [ethr_have_native_atomics=yes ethr_have_libatomic_ops=yes]) diff --git a/erts/configure.in b/erts/configure.in index 40b335849c..b3fe48d62c 100644 --- a/erts/configure.in +++ b/erts/configure.in @@ -446,13 +446,13 @@ else fi AC_ARG_ENABLE(static-nifs, -AS_HELP_STRING([--enable-static-nifs], [link nifs statically. If yes then all nifs in all Erlang/OTP applications will be statically linked into the main binary. It is also possible to give a list of nifs that should be linked statically. The list should be a comma seperated and contain the absolute path to a .a archive for each nif that is to be statically linked. The name of the .a archive has to be the same as the name of the nif. Note that you have to link any external dependencies that the nifs have to the main binary, so for the crypto nif you want to pass LIBS=-lcrypto to configure.]), +AS_HELP_STRING([--enable-static-nifs], [link nifs statically. If yes then all nifs in all Erlang/OTP applications will be statically linked into the main binary. It is also possible to give a list of nifs that should be linked statically. The list should be a comma separated and contain the absolute path to a .a archive for each nif that is to be statically linked. The name of the .a archive has to be the same as the name of the nif. Note that you have to link any external dependencies that the nifs have to the main binary, so for the crypto nif you want to pass LIBS=-lcrypto to configure.]), STATIC_NIFS="$enableval", STATIC_NIFS=no) AC_SUBST(STATIC_NIFS) AC_ARG_ENABLE(static-drivers, -AS_HELP_STRING([--enable-static-drivers], [comma seperated list of linked-in drivers to link statically with the main binary. The list should contain the absolute path to a .a archive for each driver that is to be statically linked. The name of the .a archive has to be the same as the name of the driver.]), +AS_HELP_STRING([--enable-static-drivers], [comma separated list of linked-in drivers to link statically with the main binary. The list should contain the absolute path to a .a archive for each driver that is to be statically linked. The name of the .a archive has to be the same as the name of the driver.]), STATIC_DRIVERS="$enableval", STATIC_DRIVERS=no) AC_SUBST(STATIC_DRIVERS) @@ -1691,10 +1691,13 @@ systemd_daemon_save_LIBS=$LIBS LIBS= AC_SEARCH_LIBS(sd_listen_fds,[systemd systemd-daemon], [have_sd_listen_fds=yes],[have_sd_listen_fds=no],$systemd_daemon_save_LIBS) +AC_SEARCH_LIBS(sd_notify,[systemd systemd-daemon], + [have_sd_notify=yes],[have_sd_notify=no],$systemd_daemon_save_LIBS) AC_CHECK_HEADERS(systemd/sd-daemon.h, [have_systemd_sd_daemon_h=yes],[have_systemd_sd_daemon_h=no]) if test x"$have_sd_listen_fds" = x"yes" && \ + test x"$have_sd_notify" = x"yes" && \ test x"$have_systemd_sd_daemon_h" = x"yes"; then AC_DEFINE([HAVE_SYSTEMD_DAEMON],[1],[Define if you have systemd daemon]) SYSTEMD_DAEMON_LIBS=$LIBS @@ -1781,7 +1784,7 @@ 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, @@ -2109,6 +2112,17 @@ AC_CHECK_FUNCS([ieee_handler fpsetmask finite isnan isinf res_gethostbyname dlop flockfile fstat strlcpy strlcat setsid posix2time time2posix \ setlocale nl_langinfo poll mlockall]) +AC_MSG_CHECKING([for isfinite]) +AC_TRY_LINK([#include <math.h>], + [isfinite(0);], have_isfinite=yes, have_isfinite=no), + +if test $have_isfinite = yes; then + AC_DEFINE(HAVE_ISFINITE,[1], + [Define to 1 if you have the `isfinite' function.]) + AC_MSG_RESULT(yes) +else + AC_MSG_RESULT(no) +fi case X$erl_xcomp_posix_memalign in Xno) ;; @@ -3956,6 +3970,7 @@ if test "$enable_dtrace_test" = "yes" ; then DTRACE_ENABLED_2STEP=yes fi], []) + $RM -f foo-dtrace.h AS_IF([test "x$DTRACE_ENABLED_2STEP" = "xyes"], [AC_MSG_RESULT([yes])], [AC_MSG_RESULT([no])]) @@ -3988,7 +4003,7 @@ dnl If set to --with-ssl=PATH we use that path as the prefix, i.e. we dnl use "PATH/include" and "PATH/lib". AC_SUBST(SSL_INCLUDE) -AC_SUBST(SSL_ROOT) +AC_SUBST(SSL_INCDIR) AC_SUBST(SSL_LIBDIR) AC_SUBST(SSL_CRYPTO_LIBNAME) AC_SUBST(SSL_SSL_LIBNAME) @@ -4082,6 +4097,29 @@ AS_HELP_STRING([--with-ssl=PATH], [specify location of OpenSSL include and lib]) AS_HELP_STRING([--with-ssl], [use SSL (default)]) AS_HELP_STRING([--without-ssl], [don't use SSL])) +AC_ARG_WITH(ssl-incl, +AS_HELP_STRING([--with-ssl-incl=PATH], [location of OpenSSL include dir, if different than specified by --with-ssl=PATH]), +[ +case X$with_ssl in + X | Xyes | Xno) AC_MSG_ERROR([--with-ssl-incl=PATH set without --with-ssl=PATH]);; +esac +], +[with_ssl_incl=$with_ssl]) #default + +AC_ARG_WITH(ssl-rpath, +AS_HELP_STRING([--with-ssl-rpath=yes|no|PATHS], + [runtime library path for OpenSSL. Default is "yes", which equates to a + number of standard locations. If "no", then no runtime + library paths wil be used. Anything else should be a + comma separated list of paths.]), +[ +case X$with_ssl in + Xno) AC_MSG_ERROR([--with-ssl-rpath set without --with-ssl]);; +esac +], +[with_ssl_rpath=yes]) #default + + AC_ARG_ENABLE(dynamic-ssl-lib, AS_HELP_STRING([--disable-dynamic-ssl-lib], [disable using dynamic openssl libraries]), @@ -4196,7 +4234,7 @@ case "$erl_xcomp_without_sysroot-$with_ssl" in dir="$erl_xcomp_sysroot$rdir" if test -f "$erl_xcomp_isysroot$rdir/include/openssl/opensslv.h"; then is_real_ssl=yes - SSL_ROOT="$dir" + SSL_INCDIR="$dir" if test "x$MIXED_CYGWIN" = "xyes" -o "x$MIXED_MSYS" = "xyes"; then if test -f "$dir/lib/VC/libeay32.lib"; then SSL_RUNTIME_LIBDIR="$rdir/lib/VC" @@ -4326,8 +4364,8 @@ case "$erl_xcomp_without_sysroot-$with_ssl" in # Trust OpenBSD to have everything the in the correct locations. ssl_found=yes ssl_linkable=yes - SSL_ROOT="$erl_xcomp_sysroot/usr" - AC_MSG_RESULT([$SSL_ROOT]) + SSL_INCDIR="$erl_xcomp_sysroot/usr" + AC_MSG_RESULT([$SSL_INCDIR]) SSL_RUNTIME_LIB="/usr/lib" SSL_LIB="$erl_xcomp_sysroot/usr/lib" SSL_BINDIR="/usr/sbin" @@ -4394,7 +4432,10 @@ dnl so it is - be adoptable if test ! -d "$with_ssl" ; then AC_MSG_ERROR(Invalid path to option --with-ssl=PATH) fi - SSL_ROOT="$with_ssl" + if test ! -d "$with_ssl_incl" ; then + AC_MSG_ERROR(Invalid path to option --with-ssl-incl=PATH) + fi + SSL_INCDIR="$with_ssl_incl" SSL_CRYPTO_LIBNAME=crypto SSL_SSL_LIBNAME=ssl if test "x$MIXED_CYGWIN" = "xyes" -o "x$MIXED_MSYS" = "xyes" && test -d "$with_ssl/lib/VC"; then @@ -4444,12 +4485,12 @@ dnl so it is - be adoptable elif test '!' -f ${SSL_LIBDIR}/lib${SSL_CRYPTO_LIBNAME}.so -a '!' -f "$SSL_LIBDIR/lib${SSL_CRYPTO_LIBNAME}.dylib"; then SSL_STATIC_ONLY=yes fi - SSL_INCLUDE="-I$with_ssl/include" + SSL_INCLUDE="-I$with_ssl_incl/include" SSL_APP=ssl CRYPTO_APP=crypto SSH_APP=ssh if test "$cross_compiling" = "yes"; then - SSL_RUNTIME_LIBDIR=`echo "$SSL_LIBDIR" | sed -n "s|^$erl_xcomp_sysroot\(.*\)\$|\1|p"` + SSL_RUNTIME_LIBDIR=`echo "$SSL_LIBDIR" | sed -n "s|^$erl_xcomp_sysroot\(/*\)\(.*\)\$|/\2|p"` else SSL_RUNTIME_LIBDIR="$SSL_LIBDIR" fi @@ -4507,8 +4548,8 @@ if test "x$SSL_APP" != "x" ; then SSL_KRB5_INCLUDE= if test "x$ssl_krb5_enabled" = "xyes" ; then AC_MSG_CHECKING(for krb5.h in standard locations) - for dir in $extra_dir "$SSL_ROOT/include" "$SSL_ROOT/include/openssl" \ - "$SSL_ROOT/include/kerberos" \ + for dir in $extra_dir "$SSL_INCDIR/include" "$SSL_INCDIR/include/openssl" \ + "$SSL_INCDIR/include/kerberos" \ "$erl_xcomp_isysroot/cygdrive/c/kerberos/include" \ "$erl_xcomp_isysroot/usr/local/kerberos/include" \ "$erl_xcomp_isysroot/usr/kerberos/include" \ @@ -4544,7 +4585,12 @@ cc_rflg="$CFLAG_RUNTIME_LIBRARY_PATH" ld_rflg="$LDFLAG_RUNTIME_LIBRARY_PATH" ded_ld_rflg="$DED_LD_FLAG_RUNTIME_LIBRARY_PATH" -if test "$SSL_APP" != "" && test "$SSL_DYNAMIC_ONLY" = "yes" && \ + +case "$with_ssl_rpath" in + +yes) # Use standard lib locations for ssl runtime library path + + if test "$SSL_APP" != "" && test "$SSL_DYNAMIC_ONLY" = "yes" && \ { test "$cc_rflg" != "" || test "$ld_rflg" != "" || test "$ded_ld_rflg" != ""; } ; then AC_MSG_CHECKING(for ssl runtime library path to use) @@ -4628,7 +4674,25 @@ if test "$SSL_APP" != "" && test "$SSL_DYNAMIC_ONLY" = "yes" && \ AC_MSG_RESULT([$rpath]) test "$rpath" != "" || AC_MSG_WARN([Cannot set run path during linking]) -fi + fi + ;; + +no) # Use no ssl runtime library path + SSL_DED_LD_RUNTIME_LIBRARY_PATH= + ;; + +*) # Use ssl runtime library paths set by --with-ssl-rpath (without any check) + ded_ld_rpath= + delimit= + for dir in `echo $with_ssl_rpath | sed "s/,/ /g"`; do + ded_ld_rpath="$ded_ld_rpath$delimit$ded_ld_rflg$dir" + delimit=" " + done + SSL_DED_LD_RUNTIME_LIBRARY_PATH="$ded_ld_rpath" + ;; + +esac + #-------------------------------------------------------------------- # Os mon stuff. @@ -4805,7 +4869,7 @@ AH_BOTTOM([ #define HAVE_GETHRVTIME #endif -#ifndef HAVE_FINITE +#if !defined(HAVE_ISFINITE) && !defined(HAVE_FINITE) # if defined(HAVE_ISINF) && defined(HAVE_ISNAN) # define USE_ISINF_ISNAN # endif diff --git a/erts/doc/src/crash_dump.xml b/erts/doc/src/crash_dump.xml index c59741f250..2b5fc877c3 100644 --- a/erts/doc/src/crash_dump.xml +++ b/erts/doc/src/crash_dump.xml @@ -85,20 +85,22 @@ operating system.</p> <list type="bulleted"> <item>"<em><A></em>: Cannot allocate <em><N></em> - bytes of memory (of type "<em><T></em>")." - The system - has run out of memory. <A> is the allocator that failed - to allocate memory, <N> is the number of bytes that - <A> tried to allocate, and <T> is the memory block - type that the memory was needed for. The most common case is - that a process stores huge amounts of data. In this case - <T> is most often <c><![CDATA[heap]]></c>, <c><![CDATA[old_heap]]></c>, - <c><![CDATA[heap_frag]]></c>, or <c><![CDATA[binary]]></c>. For more information on - allocators see - <seealso marker="erts_alloc">erts_alloc(3)</seealso>.</item> + bytes of memory (of type "<em><T></em>", thread + <em><I></em>em>)." - The system has run out of memory. <A> + is the allocator that failed to allocate memory, <N> is the + number of bytes that <A> tried to allocate, <T> is the + memory block type that the memory was needed for, and <I> is the + thread identifier. The most common case is that a process stores huge + amounts of data. In this case <T> is most often + <c><![CDATA[heap]]></c>, <c><![CDATA[old_heap]]></c>, + <c><![CDATA[heap_frag]]></c>, or <c><![CDATA[binary]]></c>. + For more information on allocators see + <seealso marker="erts_alloc">erts_alloc(3)</seealso>.</item> <item>"<em><A></em>: Cannot reallocate <em><N></em> - bytes of memory (of type "<em><T></em>")." - Same as - above with the exception that memory was being reallocated - instead of being allocated when the system ran out of memory.</item> + bytes of memory (of type "<em><T></em>", thread + <em><I></em>em>)." - Same as above with the exception that memory + was being reallocated instead of being allocated when the system ran + out of memory.</item> <item>"Unexpected op code <em>N</em>" - Error in compiled code, <c><![CDATA[beam]]></c> file damaged or error in the compiler.</item> <item>"Module <em>Name</em> undefined" <c><![CDATA[|]]></c> "Function @@ -113,8 +115,9 @@ sockets/pipes can be used simultaneously by Erlang (due to limitations in the Unix <c><![CDATA[select]]></c> call). The number of open regular files is not affected by this.</item> - <item>"Received SIGUSR1" - The SIGUSR1 signal was sent to the - Erlang machine (Unix only).</item> + <item>"Received SIGUSR1" - Sending the SIGUSR1 signal to a + Erlang machine (Unix only) forces a crash dump. This slogan reflects + that the Erlang machine crash-dumped due to receiving that signal.</item> <item>"Kernel pid terminated (<em>Who</em>) (<em>Exit-reason</em>)" - The kernel supervisor has detected a failure, usually that the <c><![CDATA[application_controller]]></c> @@ -246,6 +249,9 @@ <tag><em>Last scheduled in for | Current call</em></tag> <item>The current function of the process. These fields will not always exist.</item> + <tag><em>Run queue</em></tag> + <item>The identifier of the scheduler run queue in which the process is + running.</item> <tag><em>Spawned by</em></tag> <item>The parent of the process, i.e. the process which executed <c><![CDATA[spawn]]></c> or <c><![CDATA[spawn_link]]></c>.</item> diff --git a/erts/doc/src/erl.xml b/erts/doc/src/erl.xml index 5bde285311..d11f6b0c6d 100644 --- a/erts/doc/src/erl.xml +++ b/erts/doc/src/erl.xml @@ -525,7 +525,8 @@ core dump and no crash dump if an internal error is detected.</p> <p>Calling <c>erlang:halt/1</c> with a string argument will still - produce a crash dump.</p> + produce a crash dump. On Unix systems, sending an emulator process + a SIGUSR1 signal will also force a crash dump.</p> </item> <tag><marker id="+e"><c><![CDATA[+e Number]]></c></marker></tag> <item> @@ -1141,6 +1142,23 @@ <p>For more information, see <seealso marker="erlang#system_info_cpu_topology">erlang:system_info(cpu_topology)</seealso>.</p> </item> + <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> + <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, + schedulers will be more eager to check for I/O when + <c>true</c> is passed. This however also implies that + execution of outstanding I/O operation will not be + prioritized to the same extent as when <c>false</c> is + passed.</p> + <p><seealso marker="erlang#system_info_eager_check_io"><c>erlang:system_info(eager_check_io)</c></seealso> + returns the value of this parameter used when starting the VM.</p> + </item> <tag><marker id="+sfwi"><c>+sfwi Interval</c></marker></tag> <item> <p>Set scheduler forced wakeup interval. All run queues will @@ -1186,7 +1204,7 @@ utilization. </p> </item> - <tag><marker id="+swct"><c>+sws very_eager|eager|medium|lazy|very_lazy</c></marker></tag> + <tag><marker id="+swct"><c>+swct very_eager|eager|medium|lazy|very_lazy</c></marker></tag> <item> <p> Set scheduler wake cleanup threshold. Default is <c>medium</c>. diff --git a/erts/doc/src/erl_driver.xml b/erts/doc/src/erl_driver.xml index ad37813ac0..77fc906aca 100644 --- a/erts/doc/src/erl_driver.xml +++ b/erts/doc/src/erl_driver.xml @@ -546,6 +546,7 @@ typedef struct ErlDrvSysInfo { int scheduler_threads; int nif_major_version; int nif_minor_version; + int dirty_scheduler_support; } ErlDrvSysInfo; </code> @@ -610,6 +611,10 @@ typedef struct ErlDrvSysInfo { <tag><c>nif_minor_version</c></tag> <item>The value of <c>ERL_NIF_MINOR_VERSION</c> when the runtime system was compiled. </item> + <tag><c>dirty_scheduler_support</c></tag> + <item>A value <c>!= 0</c> if the runtime system has support for dirty scheduler threads; + otherwise <c>0</c>. + </item> </taglist> </item> <tag><marker id="ErlDrvBinary"/>ErlDrvBinary</tag> @@ -2028,7 +2033,8 @@ ERL_DRV_MAP int sz entry function is called. If <c>ready_async</c> is null in the driver entry, the <c>async_free</c> function is called instead.</p> - <p>The return value is a handle to the asynchronous task.</p> + <p>The return value is -1 if the <c>driver_async</c> call + fails.</p> <note> <p>As of erts version 5.5.4.3 the default stack size for threads in the async-thread pool is 16 kilowords, diff --git a/erts/doc/src/erl_ext_dist.xml b/erts/doc/src/erl_ext_dist.xml index fa083db4c7..a6e7dddbed 100644 --- a/erts/doc/src/erl_ext_dist.xml +++ b/erts/doc/src/erl_ext_dist.xml @@ -126,9 +126,8 @@ However, only characters that can be encoded using Latin1 (ISO-8859-1) are currently supported in atoms. The support for UTF-8 encoded atoms in the external format has been implemented in order to be able to support - all Unicode characters in atoms in <em>some future release</em>. Full - support for Unicode atoms will not happen before OTP-R18, and might - be introduced even later than that. Until full Unicode support for + all Unicode characters in atoms in <em>some future release</em>. + Until full Unicode support for atoms has been introduced, it is an <em>error</em> to pass atoms containing characters that cannot be encoded in Latin1, and <em>the behavior is undefined</em>.</p> diff --git a/erts/doc/src/erl_nif.xml b/erts/doc/src/erl_nif.xml index 6b1f4cccf8..3de94be9ff 100644 --- a/erts/doc/src/erl_nif.xml +++ b/erts/doc/src/erl_nif.xml @@ -168,16 +168,18 @@ ok <p><marker id="lengthy_work"/> As mentioned in the <seealso marker="#WARNING">warning</seealso> text at the beginning of this document it is of vital importance that a native function - does return relatively fast. It is hard to give an exact maximum amount + return relatively quickly. It is hard to give an exact maximum amount of time that a native function is allowed to work, but as a rule of thumb - a well behaving native function should return to its caller before a + a well-behaving native function should return to its caller before a millisecond has passed. This can be achieved using different approaches. - If you have full control over the code that are to execute in the native + If you have full control over the code to execute in the native function, the best approach is to divide the work into multiple chunks of - work and call the native function multiple times. Function + work and call the native function multiple times, either directly from Erlang code + or by having a native function schedule a future NIF call via the + <seealso marker="#enif_schedule_nif"> enif_schedule_nif</seealso> function. Function <seealso marker="#enif_consume_timeslice">enif_consume_timeslice</seealso> can be - used this facilitate such work division. In some cases, however, this might not - be possible, e.g. when calling third party libraries. Then you typically want + used to help with such work division. In some cases, however, this might not + be possible, e.g. when calling third-party libraries. Then you typically want to dispatch the work to another thread, return from the native function, and wait for the result. The thread can send the result back to the calling thread using message passing. Information @@ -342,37 +344,38 @@ ok libraries might however fail if deprecated features are used. </p></item> - <tag>Dirty NIFs</tag> - <item><p><marker id="dirty_nifs"/><em>Note that the dirty NIF functionality - is experimental</em> and that you have to enable support for dirty - schedulers when building OTP in order to try the functionality out. Native functions + <tag>Long-running NIFs</tag> + <item><p><marker id="dirty_nifs"/>Native functions <seealso marker="#lengthy_work"> must normally run quickly</seealso>, as explained earlier in this document. They generally should execute for no more than a millisecond. But not all native functions can execute so quickly; for example, functions that encrypt large blocks of data or perform lengthy file system operations can often run for tens of seconds or more.</p> - <p>A NIF that cannot execute in a millisecond or less is called a "dirty NIF" since - it performs work that the Erlang runtime cannot handle cleanly. Applications - that make use of such functions must indicate to the runtime that the functions are + <p>If the functionality of a long-running NIF can be split so that its work can be + achieved through a series of shorter NIF calls, the application can either make that series + of NIF calls from the Erlang level, or it can call a NIF that first performs a chunk of the + work, then invokes the <seealso marker="#enif_schedule_nif">enif_schedule_nif</seealso> + function to schedule another NIF call to perform the next chunk. The final call scheduled + in this manner can then return the overall result. Breaking up a long-running function in + this manner enables the VM to regain control between calls to the NIFs, thereby avoiding + degraded responsiveness, scheduler load balancing problems, and other strange behaviours.</p> + <p>A NIF that cannot be split and cannot execute in a millisecond or less is called a "dirty NIF" + because it performs work that the Erlang runtime cannot handle cleanly. + <em>Note that the dirty NIF functionality described here is experimental</em> and that you have to + enable support for dirty schedulers when building OTP in order to try the functionality out. + Applications that make use of such functions must indicate to the runtime that the functions are dirty so they can be handled specially. To schedule a dirty NIF for execution, the - application calls <seealso marker="#enif_schedule_dirty_nif">enif_schedule_dirty_nif</seealso>, - passing to it a pointer to the dirty NIF to be executed and indicating with a flag + appropriate flags value can be set for the NIF in its <seealso marker="#ErlNifFunc">ErlNifFunc</seealso> + entry, or the application can call <seealso marker="#enif_schedule_nif">enif_schedule_nif</seealso>, + passing to it a pointer to the dirty NIF to be executed and indicating with the <c>flags</c> argument whether it expects the operation to be CPU-bound or I/O-bound.</p> - <p>All dirty NIFs must ultimately invoke the <seealso marker="#enif_schedule_dirty_nif_finalizer"> - enif_schedule_dirty_nif_finalizer</seealso> as their final action, passing to it the - result they wish to return to the original caller. A finalizer function can either - receive the result and return it directly, or it can return a different value instead. - For convenience, the NIF API provides the <seealso marker="#enif_dirty_nif_finalizer"> - enif_dirty_nif_finalizer</seealso> function that applications can use as a finalizer; - it simply returns its result argument.</p> <note><p>Dirty NIF support is available only when the emulator is configured with dirty schedulers enabled. This feature is currently disabled by default. To determine whether the dirty NIF API is available, native code can check to see if the C preprocessor macro <c>ERL_NIF_DIRTY_SCHEDULER_SUPPORT</c> is defined. Also, if the Erlang runtime was built without threading support, dirty schedulers are disabled. To check at runtime for the presence - of dirty scheduler threads, code can call the <seealso marker="#enif_have_dirty_schedulers"><c> - enif_have_dirty_schedulers()</c></seealso> API function, which returns true if dirty - scheduler threads are present, false otherwise.</p></note> + of dirty scheduler threads, code can use the <seealso marker="#enif_system_info"><c> + enif_system_info()</c></seealso> API function.</p></note> </item> </taglist> </section> @@ -498,6 +501,7 @@ typedef struct { const char* <em>name</em>; unsigned <em>arity</em>; ERL_NIF_TERM (*<em>fptr</em>)(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]); + unsigned flags; } ErlNifFunc; </code> <p>Describes a NIF by its name, arity and implementation. @@ -508,7 +512,17 @@ typedef struct { will thus denote the Nth argument to the NIF. Note that the <c>argc</c> argument allows for the same C function to implement several Erlang functions with different arity (but - same name probably).</p> + same name probably). For a regular NIF, <c>flags</c> is 0 (and + so its value can be omitted for statically initialized <c>ErlNifFunc</c> + instances), or it can be used to indicate that the NIF is a <seealso + marker="#dirty_nifs">dirty NIF</seealso> that should be executed + on a dirty scheduler thread (<em>note that the dirty NIF functionality + described here is experimental</em> and that you have to enable + support for dirty schedulers when building OTP in order to try the + functionality out). If the dirty NIF is expected to be + CPU-bound, its <c>flags</c> field should be set to + <c>ERL_NIF_DIRTY_JOB_CPU_BOUND</c>, or for I/O-bound jobs, + <c>ERL_NIF_DIRTY_JOB_IO_BOUND</c>.</p> </item> <tag><marker id="ErlNifBinary"/>ErlNifBinary</tag> <item> @@ -672,18 +686,6 @@ typedef enum { See also the <seealso marker="#WARNING">warning</seealso> text at the beginning of this document.</p> </desc> </func> - <func><name><ret>ERL_NIF_TERM</ret><nametext>enif_dirty_nif_finalizer(ErlNifEnv* env, ERL_NIF_TERM result)</nametext></name> - <fsummary>Simple dirty NIF result finalizer</fsummary> - <desc> - <p>A convenience function that a dirty NIF can use as a finalizer that simply - return its <c>result</c> argument as its return value. This function is provided - for dirty NIFs with results that should be returned directly to the original caller.</p> - <note><p>This function is available only when the emulator is configured with dirty - schedulers enabled. This feature is currently disabled by default. To determine whether - the dirty NIF API is available, native code can check to see if the C preprocessor macro - <c>ERL_NIF_DIRTY_SCHEDULER_SUPPORT</c> is defined.</p></note> - </desc> - </func> <func><name><ret>int</ret><nametext>enif_equal_tids(ErlNifTid tid1, ErlNifTid tid2)</nametext></name> <fsummary></fsummary> <desc><p>Same as <seealso marker="erl_driver#erl_drv_equal_tids">erl_drv_equal_tids</seealso>. @@ -804,22 +806,6 @@ typedef enum { and return true, or return false if <c>term</c> is not an unsigned integer or is outside the bounds of type <c>unsigned long</c>.</p></desc> </func> - <func><name><ret>int</ret><nametext>enif_have_dirty_schedulers()</nametext></name> - <fsummary>Runtime check for the presence of dirty scheduler threads</fsummary> - <desc> - <p>Check at runtime for the presence of dirty scheduler threads. If the emulator is - built with threading support, dirty scheduler threads are available and - <c>enif_have_dirty_schedulers()</c> returns true. If the emulator was built without - threading support, <c>enif_have_dirty_schedulers()</c> returns false.</p> - <p>If dirty scheduler threads are not available in the emulator, calls to - <c>enif_schedule_dirty_nif</c> and <c>enif_schedule_dirty_nif_finalizer</c> result in - the NIF and finalizer functions being called directly within the calling thread.</p> - <note><p>This function is available only when the emulator is configured with dirty - schedulers enabled. This feature is currently disabled by default. To determine whether - the dirty NIF API is available, native code can check to see if the C preprocessor macro - <c>ERL_NIF_DIRTY_SCHEDULER_SUPPORT</c> is defined.</p></note> - </desc> - </func> <func><name><ret>int</ret><nametext>enif_inspect_binary(ErlNifEnv* env, ERL_NIF_TERM bin_term, ErlNifBinary* bin)</nametext></name> <fsummary>Inspect the content of a binary</fsummary> <desc><p>Initialize the structure pointed to by <c>bin</c> with @@ -873,8 +859,8 @@ typedef enum { <p>Check to see if the current NIF is executing on a dirty scheduler thread. If the emulator is built with threading support, calling <c>enif_is_on_dirty_scheduler</c> from within a dirty NIF returns true. It returns false when the calling NIF is a regular - NIF or a NIF finalizer, both of which run on normal scheduler threads, or when the emulator - is built without threading support.</p> + NIF running on a normal scheduler thread, or when the emulator is built without threading + support.</p> <note><p>This function is available only when the emulator is configured with dirty schedulers enabled. This feature is currently disabled by default. To determine whether the dirty NIF API is available, native code can check to see if the C preprocessor macro @@ -1245,46 +1231,29 @@ typedef enum { <desc><p>Same as <seealso marker="erl_driver#erl_drv_rwlock_tryrwlock">erl_drv_rwlock_tryrwlock</seealso>. </p></desc> </func> - <func><name><ret>ERL_NIF_TERM</ret><nametext>enif_schedule_dirty_nif(ErlNifEnv* env, int flags, ERL_NIF_TERM (*fp)(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]), int argc, const ERL_NIF_TERM argv[])</nametext></name> - <fsummary>Schedule a dirty NIF for execution</fsummary> + <func><name><ret>ERL_NIF_TERM</ret><nametext>enif_schedule_nif(ErlNifEnv* env, const char* fun_name, int flags, ERL_NIF_TERM (*fp)(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]), int argc, const ERL_NIF_TERM argv[])</nametext></name> + <fsummary>Schedule a NIF for execution</fsummary> <desc> - <p>Schedule dirty NIF <c>fp</c> to execute a long-running operation. The <c>flags</c> - argument must be set to either <c>ERL_NIF_DIRTY_JOB_CPU_BOUND</c> if the job is expected to - be primarily CPU-bound, or <c>ERL_NIF_DIRTY_JOB_IO_BOUND</c> for jobs that will be - I/O-bound. The <c>argc</c> and <c>argv</c> arguments can either be the originals passed - into the calling NIF, or they can be values created by the calling NIF. The calling - NIF must use the return value of <c>enif_schedule_dirty_nif</c> as its own return value.</p> - <p>Be aware that <c>enif_schedule_dirty_nif</c>, as its name implies, only schedules the - dirty NIF for future execution. The calling NIF does not block waiting for the dirty NIF to - execute and return, which means that the calling NIF can't expect to receive the dirty NIF + <p>Schedule NIF <c>fp</c> to execute. This function allows an application to break up long-running + work into multiple regular NIF calls or to schedule a <seealso marker="#dirty_nifs">dirty NIF</seealso> + to execute on a dirty scheduler thread (<em>note that the dirty NIF functionality described here is + experimental</em> and that you have to enable support for dirty schedulers when building OTP in + order to try the functionality out).</p> + <p>The <c>fun_name</c> argument provides a name for the NIF being scheduled for execution. If it cannot + be converted to an atom, <c>enif_schedule_nif</c> returns a <c>badarg</c> exception.</p> + <p>The <c>flags</c> argument must be set to 0 for a regular NIF, or if the emulator was built the + experimental dirty scheduler support enabled, <c>flags</c> can be set to either <c>ERL_NIF_DIRTY_JOB_CPU_BOUND</c> + if the job is expected to be primarily CPU-bound, or <c>ERL_NIF_DIRTY_JOB_IO_BOUND</c> for jobs that will + be I/O-bound. If dirty scheduler threads are not available in the emulator, a try to schedule such a job + will result in a <c>badarg</c> exception.</p> + + <p>The <c>argc</c> and <c>argv</c> arguments can either be the originals passed into the calling NIF, or + they can be values created by the calling NIF.</p> + <p>The calling NIF must use the return value of <c>enif_schedule_nif</c> as its own return value.</p> + <p>Be aware that <c>enif_schedule_nif</c>, as its name implies, only schedules the + NIF for future execution. The calling NIF does not block waiting for the scheduled NIF to + execute and return, which means that the calling NIF can't expect to receive the scheduled NIF return value and use it for further operations.</p> - <p>A dirty NIF may not invoke the <seealso marker="#enif_make_badarg">enif_make_badarg</seealso> - to raise an exception. If it wishes to return an exception, the dirty NIF should pass a - regular result indicating the exception details to its finalizer, and allow the finalizer - to raise the exception on its behalf.</p> - <note><p>This function is available only when the emulator is configured with dirty schedulers - enabled. This feature is currently disabled by default. To determine whether the dirty NIF API - is available, native code can check to see if the C preprocessor macro - <c>ERL_NIF_DIRTY_SCHEDULER_SUPPORT</c> is defined.</p></note> - </desc> - </func> - <func><name><ret>ERL_NIF_TERM</ret><nametext>enif_schedule_dirty_nif_finalizer(ErlNifEnv* env, ERL_NIF_TERM result, ERL_NIF_TERM (*fp)(ErlNifEnv* env, ERL_NIF_TERM result))</nametext></name> - <fsummary>Schedule a dirty NIF finalizer</fsummary> - <desc> - <p>When a dirty NIF finishes executing, it must schedule a finalizer function to return - its result to the original NIF caller. The dirty NIF passes <c>result</c> as the value it - wants the finalizer to use as the return value. The <c>fp</c> argument is a pointer to the - finalizer function. The NIF API provides the <seealso marker="#enif_dirty_nif_finalizer"> - enif_dirty_nif_finalizer</seealso> function that can be used as a finalizer that simply - returns its <c>result</c> argument. You are also free to write your own custom finalizer - that uses <c>result</c> to derive a different return value, or ignores <c>result</c> - entirely and returns a completely different value.</p> - <p>Without exception, all dirty NIFs must invoke <c>enif_schedule_dirty_nif_finalizer</c> - to complete their execution.</p> - <note><p>This function is available only when the emulator is configured with dirty - schedulers enabled. This feature is currently disabled by default. To determine whether - the dirty NIF API is available, native code can check to see if the C preprocessor macro - <c>ERL_NIF_DIRTY_SCHEDULER_SUPPORT</c> is defined.</p></note> </desc> </func> <func><name><ret>ErlNifPid *</ret><nametext>enif_self(ErlNifEnv* caller_env, ErlNifPid* pid)</nametext></name> @@ -1384,4 +1353,3 @@ typedef enum { <p><seealso marker="erlang#load_nif-2">erlang:load_nif/2</seealso></p> </section> </cref> - diff --git a/erts/doc/src/erlang.xml b/erts/doc/src/erlang.xml index 84168397f6..cba2c07959 100644 --- a/erts/doc/src/erlang.xml +++ b/erts/doc/src/erlang.xml @@ -2840,7 +2840,7 @@ 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> @@ -5775,6 +5775,7 @@ ok <name name="system_info" arity="1" clause_i="52"/> <name name="system_info" arity="1" clause_i="53"/> <name name="system_info" arity="1" clause_i="54"/> + <name name="system_info" arity="1" clause_i="55"/> <fsummary>Information about the system</fsummary> <desc> <p>Returns various information about the current system @@ -5970,6 +5971,16 @@ ok The return value will always be <c>false</c> since the elib_malloc allocator has been removed.</p> </item> + <tag><marker id="system_info_eager_check_io"><c>eager_check_io</c></marker></tag> + <item> + <p> + Returns the value of the <c>erl</c> + <seealso marker="erl#+secio">+secio</seealso> command line + flag which is either <c>true</c> or <c>false</c>. See the + documentation of the command line flag for information about + the different values. + </p> + </item> <tag><c>ets_limit</c></tag> <item> <p>Returns the maximum number of ETS tables allowed. This limit @@ -6133,6 +6144,11 @@ ok <seealso marker="#system_info_multi_scheduling">erlang:system_info(multi_scheduling)</seealso>, and <seealso marker="#system_info_schedulers">erlang:system_info(schedulers)</seealso>.</p> </item> + <tag><c>nif_version</c></tag> + <item> + <p>Returns a string containing the erlang NIF version + used by the runtime system. It will be on the form "<major ver>.<minor ver>".</p> + </item> <tag><marker id="system_info_otp_release"><c>otp_release</c></marker></tag> <item> <p>Returns a string containing the OTP release number of the diff --git a/erts/doc/src/notes.xml b/erts/doc/src/notes.xml index 5c4bb3ed25..c896ee0cae 100644 --- a/erts/doc/src/notes.xml +++ b/erts/doc/src/notes.xml @@ -30,6 +30,453 @@ </header> <p>This document describes the changes made to the ERTS application.</p> +<section><title>Erts 6.3</title> + + <section><title>Fixed Bugs and Malfunctions</title> + <list> + <item> + <p> + Fix HiPE debug lock checking on OS X 64bit</p> + <p> + Position-independent code is mandatory on OS X. We use + r11 as an intermediate register to fill + BIF_P->hipe.bif_callee. This fixes the following error + when doing `make debug FLAVOR=smp`:</p> + <p> + clang -cc1as: fatal error: error in backend: 32-bit + absolute addressing is not supported in 64-bit mode</p> + <p> + Own Id: OTP-12188</p> + </item> + <item> + <p> + Fix race bug that could cause VM crash in + <c>erlang:port_get_data/1</c> if the port was closed by a + concurrent process. Also fix fatal bug if + <c>port_set_data/2</c> is called with a non-immediate + data term. Both bugs exist since R16B01.</p> + <p> + Own Id: OTP-12208</p> + </item> + <item> + <p> + Correct make variable SSL_DED_LD_RUNTIME_LIBRARY_PATH + when erl_xcomp_sysroot ends with a slash.</p> + <p> + Own Id: OTP-12216 Aux Id: seq12700 </p> + </item> + <item> + <p> + Fix two cases of unreachable code caused by false use of + assigment operators.</p> + <p> + Own Id: OTP-12222</p> + </item> + <item> + <p> + Fix bug when hipe compiled code makes tail call to a BIF + that disables GC while trapping (sush as binary_to_list, + list_to_binary, binary_to_term, term_to_binary).</p> + <p> + Own Id: OTP-12231</p> + </item> + <item> + <p> + Fix bug when a migrated empty memory carrier is reused + just before it should be destroyed by the thread that + created it.</p> + <p> + Own Id: OTP-12249</p> + </item> + <item> + <p> + Prevents compile-time errors in NIFs, when the compiler + is instructed to treat missing field initializers as + errors, by adding an initializer for the new options + field which was added to ErlNifEntry for 17.3.</p> + <p> + Own Id: OTP-12266</p> + </item> + <item> + <p> + Fixed CPU topology detection on FreeBSD systems where + Erlang/OTP is compiled by new C compilers (including, but + possibly not limited to, gcc 4.9 and clang).</p> + <p> + Own Id: OTP-12267</p> + </item> + <item> + <p> + Use C99 function isfinite() instead of finite() when + available on non GCC compilers.</p> + <p> + Own Id: OTP-12268</p> + </item> + <item> + <p> + Fix bug on windows where an incorrect number of links + could be returned when doing file:read_file_info on a + directory.</p> + <p> + Own Id: OTP-12269</p> + </item> + <item> + <p> + Fix rare bug when purging module on VM started with + +Meamin.</p> + <p> + Own Id: OTP-12273</p> + </item> + <item> + <p> + Repair run_erl terminal window size adjustment sent from + to_erl. This was broken in OTP 17.0 which could lead to + strange cursor behaviour in the to_erl shell.</p> + <p> + Own Id: OTP-12275 Aux Id: seq12739 </p> + </item> + <item> + <p> + Fixed bug on windows causing gen_tcp/udp to return an + error when given an fd to work with.</p> + <p> + Own Id: OTP-12289</p> + </item> + <item> + <p> + Fix various internal erts issues where negating a signed + integer in C would trigger undefined behavior. This fixes + issues when dividing with bignums and list_to_integer.</p> + <p> + Own Id: OTP-12290</p> + </item> + <item> + <p> + When flushing output to stdout on windows, the emulator + could sometimes hang indefinitely waiting for the flush + to complete. This has been fixed.</p> + <p> + Own Id: OTP-12291</p> + </item> + <item> + <p> + Fix so that non-smp emulators with dirty scheduler + support shows the correct number of dirty schedulers when + calling erlang:system_info(system_version).</p> + <p> + Own Id: OTP-12295</p> + </item> + <item> + <p> + Add <c>nif_version</c> to <c>erlang:system_info/1</c> in + order to get the NIF API version of the runtime system in + a way similar to <c>driver_version</c>.</p> + <p> + Own Id: OTP-12298</p> + </item> + <item> + <p> + Fix bug that could cause the return value from dirty NIF + with zero arity to be treated as garbage, leading to VM + crash.</p> + <p> + Own Id: OTP-12300</p> + </item> + <item> + <p> + Improve allocation carrier migration search logic. This + will reduce the risk of failed migrations that could lead + to excess memory consumption. It will also improve smp + performance due to reduced memory contention on the + migration pool.</p> + <p> + Own Id: OTP-12323</p> + </item> + </list> + </section> + + + <section><title>Improvements and New Features</title> + <list> + <item> + <p>Introduced support for eager check I/O.</p> + <p>By default eager check I/O will be disabled, but this + will most likely be changed in OTP 18. When eager check + I/O is enabled, schedulers will more frequently check for + I/O work. Outstanding I/O operations will however not be + prioritized to the same extent as when eager check I/O is + disabled.</p> + <p>Eager check I/O can be enabled using the <c>erl</c> + command line argument: <seealso + marker="erl#+secio"><c>+secio true</c></seealso></p> + <p>Characteristics impact when enabled:</p> <list> + <item>Lower latency and smoother management of externally + triggered I/O operations.</item> <item>A slightly reduced + priority of externally triggered I/O operations.</item> + </list> + <p> + Own Id: OTP-12117</p> + </item> + <item> + <p> + Fix erts .app-file</p> + <p> + Own Id: OTP-12189</p> + </item> + <item> + <p> + Add configure option --with-ssl-incl=PATH to support + OpenSSL installations with headers and libraries at + different places.</p> + <p> + Own Id: OTP-12215 Aux Id: seq12700 </p> + </item> + <item> + <p> + Optimization of atomic memory operations with release + barrier semantics on 32-bit PowerPC when using the + implementation included in OTP.</p> + <p> + Own Id: OTP-12250</p> + </item> + <item> + <p> + Minor adjustment of scheduler activation code making sure + that an activation of a scheduler is not prevented by its + run-queue being non-empty. (Thanks to Songlu Cai)</p> + <p> + Own Id: OTP-12287</p> + </item> + <item> + <p> + Improved support for atomic memory operations provided by + the <url + href="https://github.com/ivmai/libatomic_ops/"><c>libatomic_ops</c></url> + library. Most importantly support for use of native + double word atomics when implemented by + <c>libatomic_ops</c> (for example, implemented for ARM).</p> + <p> + The <seealso + marker="doc/installation_guide:INSTALL#Advanced-configuration-and-build-of-ErlangOTP_Configuring_Atomic-Memory-Operations-and-the-VM"><c>$ERL_TOP/HOWTO/INSTALL.md</c></seealso> + document now also more clearly describes when you want to + build together with a <c>libatomic_ops</c> installation.</p> + <p> + Own Id: OTP-12302</p> + </item> + <item> + <p> + Add configure option --with-ssl-rpath to control which + runtime library path to use for dynamic linkage toward + OpenSSL.</p> + <p> + Own Id: OTP-12316 Aux Id: seq12753 </p> + </item> + <item> + <p> + Added systemd notify support to epmd</p> + <p> + Own Id: OTP-12321</p> + </item> + </list> + </section> + +</section> + +<section><title>Erts 6.2.1</title> + + <section><title>Fixed Bugs and Malfunctions</title> + <list> + <item> + <p> + Fix bug when an migrated empty memory carrier is reused + just before it should be destroyed by the thread that + created it.</p> + <p> + Own Id: OTP-12249</p> + </item> + <item> + <p> + Repair run_erl terminal window size adjustment sent from + to_erl. This was broken in OTP 17.0 which could lead to + strange cursor behaviour in the to_erl shell.</p> + <p> + Own Id: OTP-12275 Aux Id: seq12739 </p> + </item> + </list> + </section> + +</section> + +<section><title>Erts 6.2</title> + + <section><title>Fixed Bugs and Malfunctions</title> + <list> + <item> + <p> + General documentation updates.</p> + <p> + Own Id: OTP-12052</p> + </item> + <item> + <p>A bug in the VM code implementing sending of signals + to ports could cause the receiving port queue to remain + in a busy state forever. When this state had been + reached, processes sending command signals to the port + either got suspended forever, or, if the <c>nosuspend</c> + feature was used, always failed to send to the port. This + bug was introduced in ERTS version 5.10.</p> + <p>In order for this bug to be triggered on a port, one + had to at least once utilize the <c>nosuspend</c> + functionality when passing a signal to the port. This by + either calling</p> <list> <item> <seealso + marker="erlang#port_command/3"><c>port_command(Port, + Data, [nosuspend | Options])</c></seealso>, </item> + <item> <seealso + marker="erlang#send/3"><c>erlang:send(Port, {PortOwner, + {command, Data}}, [nosuspend | Options])</c></seealso>, + </item> <item> <seealso + marker="erlang#send_nosuspend/2"><c>erlang:send_nosuspend(Port, + {PortOwner, {command, Data}})</c></seealso>, or </item> + <item> <seealso + marker="erlang#send_nosuspend/3"><c>erlang:send_nosuspend(Port, + {PortOwner, {command, Data}}, Options)</c></seealso>. + </item> </list> + <p>Thanks Vasily Demidenok for reporting the issue, and + Sergey Kudryashov for providing a testcase.</p> + <p> + Own Id: OTP-12082 Aux Id: OTP-10336 </p> + </item> + <item> + <p> + Fix size overflow bug at memory allocation. A memory + allocation call, with an insane size close to the entire + address space, could return successfully as if it had + allocated just a few bytes. (Thanks to Don A. Bailey for + reporting)</p> + <p> + Own Id: OTP-12091</p> + </item> + <item> + <p> + Fix various issues where negating a signed integer would + trigger undefined behaviour. This fixes issues in the + enif_make_int64 interface and some edge cases inside the + erlang runtime system.</p> + <p> + Own Id: OTP-12097</p> + </item> + <item> + <p> + The documentation erroneously listed the <seealso + marker="erl#+swct"><c>+swct</c></seealso> command line + argument under <c>+sws</c>.</p> + <p> + Own Id: OTP-12102 Aux Id: OTP-10994 </p> + </item> + <item> + <p> + Profiling messages could be delivered out of order when + profiling on <c>runnable_procs</c> and/or + <c>runnable_ports</c> using <seealso + marker="erlang#system_profile/2"><c>erlang:system_profile/2</c></seealso>. + This bug was introduced in ERTS version 5.10.</p> + <p> + Own Id: OTP-12105 Aux Id: OTP-10336 </p> + </item> + <item> + <p> + Various logging fixes, including: Add run queue index to + the process dump in crash dumps.<br/> Add thread index to + enomem slogan when crashing.<br/> Remove error logger + message for sending messages to old instances of the same + node.</p> + <p> + Own Id: OTP-12115</p> + </item> + <item> + <p> + Fix compiler warnings reported by LLVM</p> + <p> + Own Id: OTP-12138</p> + </item> + <item> + <p> + Correct conversion of <c>MIN_SMALL</c> by + <c>list_to_integer/1</c> and <c>binary_to_integer/1</c>. + The bug produced an unnormalized bignum which can cause + strange behavior such as comparing different to a correct + <c>MIN_SMALL</c> integer. The value <c>MIN_SMALL</c> is + <c>-(1 bsl 27) = -134217728</c> on a 32-bit VM and <c>-(1 + bsl 59) = -576460752303423488</c> on a 64-bit VM. (Thanks + to Jesper Louis Andersen, Mikael Pettersson and Anthony + Ramine for report, patch and optimization suggestion)</p> + <p> + Own Id: OTP-12140</p> + </item> + <item> + <p> + Fix bug in <c>term_to_binary</c> that reallocates binary + with inconsistent size information. Bug has never been + confirmed to be the cause of any faulty behavior.</p> + <p> + Own Id: OTP-12141</p> + </item> + <item> + <p> + Real_path method used while prim loading archive files + was not taking into account the fact that windows + directory symlinks can be across different drives.</p> + <p> + Own Id: OTP-12155</p> + </item> + </list> + </section> + + + <section><title>Improvements and New Features</title> + <list> + <item> + <p> + Add log2 histogram to lcnt for lock wait time</p> + <p> + Own Id: OTP-12059</p> + </item> + <item> + <p> + Introduced <seealso + marker="erl_nif#enif_schedule_nif"><c>enif_schedule_nif()</c></seealso> + to the NIF API.</p> + <p> + The <c>enif_schedule_nif()</c> function allows a + long-running NIF to be broken into separate NIF + invocations without the help of a wrapper function + written in Erlang. The NIF first executes part of the + long-running task, then calls <c>enif_schedule_nif()</c> + to schedule a NIF for later execution to continue the + task. Any number of NIFs can be scheduled in this manner, + one after another. Since the emulator regains control + between invocations, this helps avoid problems caused by + native code tying up scheduler threads for too long.</p> + <p> + The <c>enif_schedule_nif()</c> function also replaces the + <c>enif_schedule_dirty_nif()</c> in the experimental + dirty NIF API. Note that the only incompatible changes + made are in the experimental dirty NIF API.</p> + <p> + See the <seealso marker="erl_nif">NIF + documentation</seealso> for more information.</p> + <p> + Thanks to Steve Vinoski.</p> + <p> + *** POTENTIAL INCOMPATIBILITY ***</p> + <p> + Own Id: OTP-12128</p> + </item> + </list> + </section> + +</section> + <section><title>Erts 6.1.2</title> <section><title>Fixed Bugs and Malfunctions</title> diff --git a/erts/doc/src/zlib.xml b/erts/doc/src/zlib.xml index 11a7437f5a..da8ccdecdf 100644 --- a/erts/doc/src/zlib.xml +++ b/erts/doc/src/zlib.xml @@ -302,7 +302,7 @@ list_to_binary([B1,B2])</pre> <fsummary>Decompress data</fsummary> <desc> <p><c>inflate/2</c> decompresses as much data as possible. - It may some introduce some output latency (reading + 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>inflate/2</c> throws a diff --git a/erts/emulator/beam/beam_emu.c b/erts/emulator/beam/beam_emu.c index 1026e5f649..8bfb7d2ad2 100644 --- a/erts/emulator/beam/beam_emu.c +++ b/erts/emulator/beam/beam_emu.c @@ -3503,6 +3503,7 @@ get_map_elements_fail: * I[0]: &&call_nif * I[1]: Function pointer to NIF function * I[2]: Pointer to erl_module_nif + * I[3]: Function pointer to dirty NIF */ BifFunction vbf; @@ -3523,13 +3524,6 @@ get_map_elements_fail: reg[0] = r(0); nif_bif_result = (*fp)(&env, bif_nif_arity, reg); erts_post_nif(&env); -#ifdef ERTS_DIRTY_SCHEDULERS - if (is_non_value(nif_bif_result) && c_p->freason == TRAP) { - Export* ep = ERTS_PROC_GET_DIRTY_SCHED_TRAP_EXPORT(c_p); - ep->code[0] = I[-3]; - ep->code[1] = I[-2]; - } -#endif } ASSERT(!ERTS_PROC_IS_EXITING(c_p) || is_non_value(nif_bif_result)); PROCESS_MAIN_CHK_LOCKS(c_p); diff --git a/erts/emulator/beam/beam_load.c b/erts/emulator/beam/beam_load.c index e96177cfd9..cfc6146b0a 100644 --- a/erts/emulator/beam/beam_load.c +++ b/erts/emulator/beam/beam_load.c @@ -2363,7 +2363,11 @@ load_code(LoaderState* stp) if (stp->may_load_nif) { const int finfo_ix = ci - FUNC_INFO_SZ; - enum { MIN_FUNC_SZ = 3 }; +#ifdef ERTS_DIRTY_SCHEDULERS + enum { MIN_FUNC_SZ = 4 }; +#else + enum { MIN_FUNC_SZ = 3 }; +#endif if (finfo_ix - last_func_start < MIN_FUNC_SZ && last_func_start) { /* Must make room for call_nif op */ int pad = MIN_FUNC_SZ - (finfo_ix - last_func_start); diff --git a/erts/emulator/beam/beam_ranges.c b/erts/emulator/beam/beam_ranges.c index 0f2d5d0c2a..cb6470638f 100644 --- a/erts/emulator/beam/beam_ranges.c +++ b/erts/emulator/beam/beam_ranges.c @@ -282,7 +282,7 @@ find_range(BeamInstr* pc) while (low < high) { if (pc < mid->start) { high = mid; - } else if (pc > RANGE_END(mid)) { + } else if (pc >= RANGE_END(mid)) { low = mid + 1; } else { erts_smp_atomic_set_nob(&r[active].mid, (erts_aint_t) mid); diff --git a/erts/emulator/beam/bif.c b/erts/emulator/beam/bif.c index fcbeb6cf5c..42dd160e38 100644 --- a/erts/emulator/beam/bif.c +++ b/erts/emulator/beam/bif.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 @@ -1869,6 +1869,7 @@ do_send(Process *p, Eterm to, Eterm msg, int suspend, Eterm *refp) { } else if (is_external_pid(to)) { dep = external_pid_dist_entry(to); if(dep == erts_this_dist_entry) { +#if DEBUG erts_dsprintf_buf_t *dsbufp = erts_create_logger_dsbuf(); erts_dsprintf(dsbufp, "Discarding message %T from %T to %T in an old " @@ -1879,6 +1880,7 @@ do_send(Process *p, Eterm to, Eterm msg, int suspend, Eterm *refp) { external_pid_creation(to), erts_this_node->creation); erts_send_error_to_logger(p->group_leader, dsbufp); +#endif return 0; } return remote_send(p, dep, to, to, msg, suspend); @@ -1912,6 +1914,7 @@ do_send(Process *p, Eterm to, Eterm msg, int suspend, Eterm *refp) { } else if (is_external_port(to) && (external_port_dist_entry(to) == erts_this_dist_entry)) { +#if DEBUG erts_dsprintf_buf_t *dsbufp = erts_create_logger_dsbuf(); erts_dsprintf(dsbufp, "Discarding message %T from %T to %T in an old " @@ -1922,6 +1925,7 @@ do_send(Process *p, Eterm to, Eterm msg, int suspend, Eterm *refp) { external_port_creation(to), erts_this_node->creation); erts_send_error_to_logger(p->group_leader, dsbufp); +#endif return 0; } else if (is_internal_port(to)) { int ret_val; @@ -2768,6 +2772,7 @@ static int do_list_to_integer(Process *p, Eterm orig_list, Eterm *integer, Eterm *rest) { Sint i = 0; + Uint ui = 0; int skip = 0; int neg = 0; int n = 0; @@ -2821,8 +2826,8 @@ static int do_list_to_integer(Process *p, Eterm orig_list, unsigned_val(CAR(list_val(lst))) > '9') { break; } - i = i * 10; - i = i + unsigned_val(CAR(list_val(lst))) - '0'; + ui = ui * 10; + ui = ui + unsigned_val(CAR(list_val(lst))) - '0'; n++; lst = CDR(list_val(lst)); if (is_nil(lst)) { @@ -2846,7 +2851,8 @@ static int do_list_to_integer(Process *p, Eterm orig_list, */ if (n <= SMALL_DIGITS) { /* It must be small */ - if (neg) i = -i; + if (neg) i = -(Sint)ui; + else i = (Sint)ui; res = make_small(i); } else { lg2 = (n+1)*230/69+1; @@ -2887,9 +2893,6 @@ static int do_list_to_integer(Process *p, Eterm orig_list, res = big_plus_small(res, m, hp); } - if (is_big(res)) /* check if small */ - res = big_plus_small(res, 0, hp); /* includes conversion to small */ - if (neg) { if (is_small(res)) res = make_small(-signed_val(res)); @@ -2899,8 +2902,12 @@ static int do_list_to_integer(Process *p, Eterm orig_list, } } - if (is_big(res)) { - hp += (big_arity(res)+1); + if (is_not_small(res)) { + res = big_plus_small(res, 0, hp); /* includes conversion to small */ + + if (is_not_small(res)) { + hp += (big_arity(res)+1); + } } HRelease(p,hp_end,hp); } diff --git a/erts/emulator/beam/bif.tab b/erts/emulator/beam/bif.tab index 011e49f1fe..e68b8e6274 100644 --- a/erts/emulator/beam/bif.tab +++ b/erts/emulator/beam/bif.tab @@ -601,6 +601,10 @@ bif maps:values/1 bif erts_internal:cmp_term/2 # +# New in 17.1. +# +bif erlang:fun_info_mfa/1 +# # Obsolete # diff --git a/erts/emulator/beam/big.c b/erts/emulator/beam/big.c index e62caa6b22..de7d370938 100644 --- a/erts/emulator/beam/big.c +++ b/erts/emulator/beam/big.c @@ -274,10 +274,9 @@ _b = _b << _s; \ _vn1 = _b >> H_EXP; \ _vn0 = _b & LO_MASK; \ - /* Sometimes _s is 0 which triggers undefined behaviour for the \ - (_a0>>(D_EXP-_s)) shift, but this is ok because the \ - & -s will make it all to 0 later anyways. */ \ - _un32 = (_a1 << _s) | ((_a0>>(D_EXP-_s)) & (-_s >> (D_EXP-1))); \ + /* If needed to avoid undefined behaviour */ \ + if (_s) _un32 = (_a1 << _s) | ((_a0>>(D_EXP-_s)) & (-_s >> (D_EXP-1))); \ + else _un32 = _a1; \ _un10 = _a0 << _s; \ _un1 = _un10 >> H_EXP; \ _un0 = _un10 & LO_MASK; \ @@ -2675,9 +2674,6 @@ Eterm erts_chars_to_integer(Process *BIF_P, char *bytes, res = big_plus_small(res, m, hp); } - if (is_big(res)) /* check if small */ - res = big_plus_small(res, 0, hp); /* includes conversion to small */ - if (neg) { if (is_small(res)) res = make_small(-signed_val(res)); @@ -2687,8 +2683,12 @@ Eterm erts_chars_to_integer(Process *BIF_P, char *bytes, } } - if (is_big(res)) { - hp += (big_arity(res) + 1); + if (is_not_small(res)) { + res = big_plus_small(res, 0, hp); /* includes conversion to small */ + + if (is_not_small(res)) { + hp += (big_arity(res) + 1); + } } HRelease(BIF_P, hp_end, hp); goto bytebuf_to_integer_1_done; diff --git a/erts/emulator/beam/break.c b/erts/emulator/beam/break.c index 7d4f52ee23..08265b590d 100644 --- a/erts/emulator/beam/break.c +++ b/erts/emulator/beam/break.c @@ -256,6 +256,7 @@ print_process_info(int to, void *to_arg, Process *p) p->current[1], p->current[2]); } + erts_print(to, to_arg, "Run queue: %d\n", erts_get_runq_proc(p)->ix); erts_print(to, to_arg, "Spawned by: %T\n", p->parent); approx_started = (time_t) p->approx_started; diff --git a/erts/emulator/beam/erl_alloc.c b/erts/emulator/beam/erl_alloc.c index 05ac24e04d..90cd227fae 100644 --- a/erts/emulator/beam/erl_alloc.c +++ b/erts/emulator/beam/erl_alloc.c @@ -1873,8 +1873,8 @@ erts_alc_fatal_error(int error, int func, ErtsAlcType_t n, ...) size = va_arg(argp, Uint); va_end(argp); erl_exit(1, - "%s: Cannot %s %lu bytes of memory (of type \"%s\").\n", - allctr_str, op, size, t_str); + "%s: Cannot %s %lu bytes of memory (of type \"%s\", thread %d).\n", + allctr_str, op, size, t_str, ERTS_ALC_GET_THR_IX()); break; } case ERTS_ALC_E_NOALLCTR: diff --git a/erts/emulator/beam/erl_alloc.types b/erts/emulator/beam/erl_alloc.types index 17ac6316b7..21434eb117 100644 --- a/erts/emulator/beam/erl_alloc.types +++ b/erts/emulator/beam/erl_alloc.types @@ -267,7 +267,6 @@ type CODE_IX_LOCK_Q SHORT_LIVED SYSTEM code_ix_lock_q type PROC_INTERVAL LONG_LIVED SYSTEM process_interval type BUSY_CALLER_TAB SHORT_LIVED SYSTEM busy_caller_table type BUSY_CALLER SHORT_LIVED SYSTEM busy_caller -type PORT_DATA_HEAP STANDARD SYSTEM port_data_heap type PROC_SYS_TSK SHORT_LIVED PROCESSES proc_sys_task type PROC_SYS_TSK_QS SHORT_LIVED PROCESSES proc_sys_task_queues @@ -357,12 +356,14 @@ type DB_MS_PSDO_PROC LONG_LIVED_LOW ETS db_match_pseudo_proc type SCHDLR_DATA LONG_LIVED_LOW SYSTEM scheduler_data type LL_TEMP_TERM LONG_LIVED_LOW SYSTEM ll_temp_term +type NIF_TRAP_EXPORT STANDARD_LOW CODE nif_trap_export_entry type EXPORT LONG_LIVED_LOW CODE export_entry type MONITOR_SH STANDARD_LOW PROCESSES monitor_sh type NLINK_SH STANDARD_LOW PROCESSES nlink_sh type AINFO_REQ STANDARD_LOW SYSTEM alloc_info_request type SCHED_WTIME_REQ STANDARD_LOW SYSTEM sched_wall_time_request type GC_INFO_REQ STANDARD_LOW SYSTEM gc_info_request +type PORT_DATA_HEAP STANDARD_LOW SYSTEM port_data_heap +else # "fullword" @@ -375,12 +376,14 @@ type DB_MS_PSDO_PROC LONG_LIVED ETS db_match_pseudo_proc type SCHDLR_DATA LONG_LIVED SYSTEM scheduler_data type LL_TEMP_TERM LONG_LIVED SYSTEM ll_temp_term +type NIF_TRAP_EXPORT STANDARD CODE nif_trap_export_entry type EXPORT LONG_LIVED CODE export_entry type MONITOR_SH FIXED_SIZE PROCESSES monitor_sh type NLINK_SH FIXED_SIZE PROCESSES nlink_sh type AINFO_REQ SHORT_LIVED SYSTEM alloc_info_request type SCHED_WTIME_REQ SHORT_LIVED SYSTEM sched_wall_time_request type GC_INFO_REQ SHORT_LIVED SYSTEM gc_info_request +type PORT_DATA_HEAP STANDARD SYSTEM port_data_heap +endif @@ -395,6 +398,7 @@ type DRV_EV_STATE LONG_LIVED SYSTEM driver_event_state type DRV_EV_D_STATE FIXED_SIZE SYSTEM driver_event_data_state type DRV_SEL_D_STATE FIXED_SIZE SYSTEM driver_select_data_state type FD_LIST SHORT_LIVED SYSTEM fd_list +type ACTIVE_FD_ARR SHORT_LIVED SYSTEM active_fd_array type POLLSET LONG_LIVED SYSTEM pollset type POLLSET_UPDREQ SHORT_LIVED SYSTEM pollset_update_req type POLL_FDS LONG_LIVED SYSTEM poll_fds diff --git a/erts/emulator/beam/erl_alloc_util.c b/erts/emulator/beam/erl_alloc_util.c index a4e164bf51..e3172dc4fb 100644 --- a/erts/emulator/beam/erl_alloc_util.c +++ b/erts/emulator/beam/erl_alloc_util.c @@ -205,7 +205,7 @@ MBC after deallocating first block: ASSERT(((UWord)(F) & (~FLG_MASK|THIS_FREE_BLK_HDR_FLG|PREV_FREE_BLK_HDR_FLG)) == THIS_FREE_BLK_HDR_FLG), \ (B)->bhdr = ((Sz) | (F)), \ (B)->u.carrier = (C)) - + # define IS_MBC_FIRST_ABLK(AP,B) \ ((((UWord)(B) & ~ERTS_SACRR_UNIT_MASK) == MBC_HEADER_SIZE(AP)) \ && ((B)->bhdr & MBC_ABLK_OFFSET_MASK) == 0) @@ -378,9 +378,8 @@ do { \ #ifdef ERTS_SMP #define SBC_HEADER_SIZE \ - (UNIT_CEILING(sizeof(Carrier_t) \ - - sizeof(ErtsAlcCPoolData_t) \ - + ABLK_HDR_SZ) \ + (UNIT_CEILING(offsetof(Carrier_t, cpool) \ + + ABLK_HDR_SZ) \ - ABLK_HDR_SZ) #else #define SBC_HEADER_SIZE \ @@ -929,6 +928,88 @@ unlink_carrier(CarrierList_t *cl, Carrier_t *crr) #ifdef ERTS_SMP +#ifdef DEBUG +static int is_in_list(ErtsDoubleLink_t* sentinel, ErtsDoubleLink_t* node) +{ + ErtsDoubleLink_t* p; + + ASSERT(node != sentinel); + for (p = sentinel->next; p != sentinel; p = p->next) { + if (p == node) + return 1; + } + return 0; +} +#endif /* DEBUG */ + +static ERTS_INLINE void +link_edl_after(ErtsDoubleLink_t* after_me, ErtsDoubleLink_t* node) +{ + ErtsDoubleLink_t* before_me = after_me->next; + ASSERT(node != after_me && node != before_me); + node->next = before_me; + node->prev = after_me; + before_me->prev = node; + after_me->next = node; +} + +static ERTS_INLINE void +link_edl_before(ErtsDoubleLink_t* before_me, ErtsDoubleLink_t* node) +{ + ErtsDoubleLink_t* after_me = before_me->prev; + ASSERT(node != before_me && node != after_me); + node->next = before_me; + node->prev = after_me; + before_me->prev = node; + after_me->next = node; +} + +static ERTS_INLINE void +unlink_edl(ErtsDoubleLink_t* node) +{ + node->next->prev = node->prev; + node->prev->next = node->next; +} + +static ERTS_INLINE void +relink_edl_before(ErtsDoubleLink_t* before_me, ErtsDoubleLink_t* node) +{ + if (node != before_me && node != before_me->prev) { + unlink_edl(node); + link_edl_before(before_me, node); + } +} + +static ERTS_INLINE int is_abandoned(Carrier_t *crr) +{ + return crr->cpool.abandoned.next != NULL; +} + +static ERTS_INLINE void +link_abandoned_carrier(ErtsDoubleLink_t* list, Carrier_t *crr) +{ + ASSERT(!is_abandoned(crr)); + + link_edl_after(list, &crr->cpool.abandoned); + + ASSERT(crr->cpool.abandoned.next != &crr->cpool.abandoned); + ASSERT(crr->cpool.abandoned.prev != &crr->cpool.abandoned); +} + +static ERTS_INLINE void +unlink_abandoned_carrier(Carrier_t *crr) +{ + ASSERT(is_in_list(&crr->cpool.orig_allctr->cpool.pooled_list, + &crr->cpool.abandoned) || + is_in_list(&crr->cpool.orig_allctr->cpool.traitor_list, + &crr->cpool.abandoned)); + + unlink_edl(&crr->cpool.abandoned); + + crr->cpool.abandoned.next = NULL; + crr->cpool.abandoned.prev = NULL; +} + static ERTS_INLINE void clear_busy_pool_carrier(Allctr_t *allctr, Carrier_t *crr) { @@ -955,7 +1036,7 @@ clear_busy_pool_carrier(Allctr_t *allctr, Carrier_t *crr) } } -#endif +#endif /* ERTS_SMP */ #if 0 #define ERTS_DBG_CHK_FIX_LIST(A, FIX, IX, B) \ @@ -1775,6 +1856,18 @@ handle_delayed_dealloc(Allctr_t *allctr, * data has been overwritten by the queue. */ Carrier_t *crr = FIRST_BLK_TO_MBC(allctr, blk); + + /* Restore word overwritten by the dd-queue as it will be read + * 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*)); +#ifdef MBC_ABLK_OFFSET_BITS + blk->u.carrier = crr; +#else + blk->carrier = crr; +#endif + ERTS_ALC_CPOOL_ASSERT(ERTS_ALC_IS_CPOOL_ENABLED(allctr)); ERTS_ALC_CPOOL_ASSERT(allctr == crr->cpool.orig_allctr); ERTS_ALC_CPOOL_ASSERT(((erts_aint_t) allctr) @@ -2563,10 +2656,9 @@ mbc_realloc(Allctr_t *allctr, void *p, Uint size, Uint32 alcu_flgs, #ifdef ERTS_SMP #define ERTS_ALC_MAX_DEALLOC_CARRIER 10 -#define ERTS_ALC_CPOOL_MAX_FETCH_INSPECT 10 +#define ERTS_ALC_CPOOL_MAX_FETCH_INSPECT 20 +#define ERTS_ALC_CPOOL_MAX_TRAITOR_INSPECT 10 #define ERTS_ALC_CPOOL_CHECK_LIMIT_COUNT 100 -#define ERTS_ALC_CPOOL_MAX_NO_CARRIERS 5 -#define ERTS_ALC_CPOOL_INSERT_ALLOWED_OFFSET 100 #define ERTS_ALC_CPOOL_MAX_FAILED_STAT_READS 3 #define ERTS_ALC_CPOOL_PTR_MOD_MRK (((erts_aint_t) 1) << 0) @@ -2743,9 +2835,6 @@ cpool_insert(Allctr_t *allctr, Carrier_t *crr) (erts_aint_t) CARRIER_SZ(crr)); erts_atomic_inc_nob(&allctr->cpool.stat.no_carriers); - erts_smp_atomic_set_nob(&crr->allctr, - ((erts_aint_t) allctr)|ERTS_CRR_ALCTR_FLG_IN_POOL); - /* * We search in 'next' direction and begin by passing * one element before trying to insert. This in order to @@ -2804,6 +2893,9 @@ cpool_insert(Allctr_t *allctr, Carrier_t *crr) cpool_set_mod_marked(&cpd2p->prev, (erts_aint_t) &crr->cpool, (erts_aint_t) cpd1p); + + erts_smp_atomic_set_wb(&crr->allctr, + ((erts_aint_t) allctr)|ERTS_CRR_ALCTR_FLG_IN_POOL); } static void @@ -2904,59 +2996,163 @@ cpool_delete(Allctr_t *allctr, Allctr_t *prev_allctr, Carrier_t *crr) static Carrier_t * cpool_fetch(Allctr_t *allctr, UWord size) { - int i; + int i, i_stop, has_passed_sentinel; Carrier_t *crr; ErtsAlcCPoolData_t *cpdp; - ErtsAlcCPoolData_t *sentinel = &carrier_pool[allctr->alloc_no].sentinel; + ErtsAlcCPoolData_t *cpool_entrance; + ErtsAlcCPoolData_t *sentinel; + ErtsDoubleLink_t* dl; + ErtsDoubleLink_t* first_old_traitor; ERTS_ALC_CPOOL_ASSERT(allctr->alloc_no == ERTS_ALC_A_INVALID /* testcase */ || erts_thr_progress_is_managed_thread()); - i = 0; + i = ERTS_ALC_CPOOL_MAX_FETCH_INSPECT; + first_old_traitor = allctr->cpool.traitor_list.next; + cpool_entrance = NULL; - /* First; check our own pending dealloc carrier list... */ - crr = allctr->cpool.dc_list.last; - while (crr && i < ERTS_ALC_CPOOL_MAX_FETCH_INSPECT) { - if (erts_atomic_read_nob(&crr->cpool.max_size) >= size) { - unlink_carrier(&allctr->cpool.dc_list, crr); -#ifdef ERTS_ALC_CPOOL_DEBUG - ERTS_ALC_CPOOL_ASSERT(erts_smp_atomic_xchg_nob(&crr->allctr, - ((erts_aint_t) allctr)) - == (((erts_aint_t) allctr) & ~ERTS_CRR_ALCTR_FLG_MASK)); -#else - erts_smp_atomic_set_nob(&crr->allctr, ((erts_aint_t) allctr)); -#endif - return crr; + /* + * Search my own pooled_list, + * i.e my abandoned carriers that were in the pool last time I checked. + */ + + dl = allctr->cpool.pooled_list.next; + while(dl != &allctr->cpool.pooled_list) { + erts_aint_t exp, act; + crr = (Carrier_t *) (((char *) dl) - offsetof(Carrier_t, cpool.abandoned)); + + ASSERT(!is_in_list(&allctr->cpool.traitor_list, dl)); + ASSERT(crr->cpool.orig_allctr == allctr); + dl = dl->next; + exp = erts_smp_atomic_read_rb(&crr->allctr); + if ((exp & ERTS_CRR_ALCTR_FLG_MASK) == ERTS_CRR_ALCTR_FLG_IN_POOL + && erts_atomic_read_nob(&crr->cpool.max_size) >= size) { + /* Try to fetch it... */ + act = erts_smp_atomic_cmpxchg_mb(&crr->allctr, + (erts_aint_t) allctr, + exp); + if (act == exp) { + cpool_delete(allctr, ((Allctr_t *) (act & ~ERTS_CRR_ALCTR_FLG_MASK)), crr); + unlink_abandoned_carrier(crr); + + /* Move sentinel to continue next search from here */ + relink_edl_before(dl, &allctr->cpool.pooled_list); + return crr; + } + exp = act; + } + if (exp & ERTS_CRR_ALCTR_FLG_IN_POOL) { + if (!cpool_entrance) + cpool_entrance = &crr->cpool; + } + else { /* Not in pool, move to traitor_list */ + unlink_abandoned_carrier(crr); + link_abandoned_carrier(&allctr->cpool.traitor_list, crr); + } + if (--i <= 0) { + /* Move sentinel to continue next search from here */ + relink_edl_before(dl, &allctr->cpool.pooled_list); + return NULL; } - crr = crr->prev; - i++; } - /* ... then the pool ... */ + /* Now search traitor_list. + * i.e carriers employed by other allocators last time I checked. + * They might have been abandoned since then. + */ + + i_stop = (i < ERTS_ALC_CPOOL_MAX_TRAITOR_INSPECT ? + 0 : i - ERTS_ALC_CPOOL_MAX_TRAITOR_INSPECT); + dl = first_old_traitor; + while(dl != &allctr->cpool.traitor_list) { + erts_aint_t exp, act; + crr = (Carrier_t *) (((char *) dl) - offsetof(Carrier_t, cpool.abandoned)); + ASSERT(dl != &allctr->cpool.pooled_list); + ASSERT(crr->cpool.orig_allctr == allctr); + dl = dl->next; + exp = erts_smp_atomic_read_rb(&crr->allctr); + if (exp & ERTS_CRR_ALCTR_FLG_IN_POOL) { + if (!(exp & ERTS_CRR_ALCTR_FLG_BUSY) + && erts_atomic_read_nob(&crr->cpool.max_size) >= size) { + /* Try to fetch it... */ + act = erts_smp_atomic_cmpxchg_mb(&crr->allctr, + (erts_aint_t) allctr, + exp); + if (act == exp) { + cpool_delete(allctr, ((Allctr_t *) (act & ~ERTS_CRR_ALCTR_FLG_MASK)), crr); + unlink_abandoned_carrier(crr); + + /* Move sentinel to continue next search from here */ + relink_edl_before(dl, &allctr->cpool.traitor_list); + return crr; + } + exp = act; + } + if (exp & ERTS_CRR_ALCTR_FLG_IN_POOL) { + if (!cpool_entrance) + cpool_entrance = &crr->cpool; + + /* Move to pooled_list */ + unlink_abandoned_carrier(crr); + link_abandoned_carrier(&allctr->cpool.pooled_list, crr); + } + } + if (--i <= i_stop) { + /* Move sentinel to continue next search from here */ + relink_edl_before(dl, &allctr->cpool.traitor_list); + if (i > 0) + break; + else + return NULL; + } + } /* - * We search in 'prev' direction and begin by passing - * one element before trying to fetch. This in order to - * avoid contention with threads inserting elements. + * Finally search the shared pool and try employ foreign carriers */ - cpdp = cpool_aint2cpd(cpool_read(&sentinel->prev)); - if (cpdp == sentinel) - return NULL; + sentinel = &carrier_pool[allctr->alloc_no].sentinel; + if (cpool_entrance) { + /* We saw a pooled carried above, use it as entrance into the pool + */ + cpdp = cpool_entrance; + } + else { + /* No pooled carried seen above. Start search at cpool sentinel, + * but begin by passing one element before trying to fetch. + * This in order to avoid contention with threads inserting elements. + */ + cpool_entrance = sentinel; + cpdp = cpool_aint2cpd(cpool_read(&cpool_entrance->prev)); + if (cpdp == sentinel) + return NULL; + } - while (i < ERTS_ALC_CPOOL_MAX_FETCH_INSPECT) { + has_passed_sentinel = 0; + while (1) { erts_aint_t exp; cpdp = cpool_aint2cpd(cpool_read(&cpdp->prev)); - if (cpdp == sentinel) { + if (cpdp == cpool_entrance) { + if (cpool_entrance == sentinel) { + cpdp = cpool_aint2cpd(cpool_read(&cpdp->prev)); + if (cpdp == sentinel) + return NULL; + } + i = 0; /* Last one to inspect */ + } + else if (cpdp == sentinel) { + if (has_passed_sentinel) { + /* We been here before. cpool_entrance must have been removed */ + return NULL; + } cpdp = cpool_aint2cpd(cpool_read(&cpdp->prev)); if (cpdp == sentinel) return NULL; - i = ERTS_ALC_CPOOL_MAX_FETCH_INSPECT; /* Last one to inspect */ + has_passed_sentinel = 1; } - crr = (Carrier_t *) (((char *) cpdp) - offsetof(Carrier_t, cpool)); + crr = (Carrier_t *)(((char *)cpdp) - offsetof(Carrier_t, cpool)); exp = erts_smp_atomic_read_rb(&crr->allctr); - if (((exp & (ERTS_CRR_ALCTR_FLG_IN_POOL|ERTS_CRR_ALCTR_FLG_BUSY)) - == ERTS_CRR_ALCTR_FLG_IN_POOL) + if (((exp & (ERTS_CRR_ALCTR_FLG_MASK)) == ERTS_CRR_ALCTR_FLG_IN_POOL) && (erts_atomic_read_nob(&cpdp->max_size) >= size)) { erts_aint_t act; /* Try to fetch it... */ @@ -2965,11 +3161,35 @@ cpool_fetch(Allctr_t *allctr, UWord size) exp); if (act == exp) { cpool_delete(allctr, ((Allctr_t *) (act & ~ERTS_CRR_ALCTR_FLG_MASK)), crr); + if (crr->cpool.orig_allctr == allctr) { + unlink_abandoned_carrier(crr); + } return crr; } } - i++; + if (--i <= 0) + return NULL; + } + + /* Last; check our own pending dealloc carrier list... */ + crr = allctr->cpool.dc_list.last; + while (crr) { + if (erts_atomic_read_nob(&crr->cpool.max_size) >= size) { + unlink_carrier(&allctr->cpool.dc_list, crr); +#ifdef ERTS_ALC_CPOOL_DEBUG + ERTS_ALC_CPOOL_ASSERT(erts_smp_atomic_xchg_nob(&crr->allctr, + ((erts_aint_t) allctr)) + == (((erts_aint_t) allctr) & ~ERTS_CRR_ALCTR_FLG_MASK)); +#else + erts_smp_atomic_set_nob(&crr->allctr, ((erts_aint_t) allctr)); +#endif + return crr; + } + crr = crr->prev; + if (--i <= 0) + return NULL; } + return NULL; } @@ -3066,6 +3286,9 @@ schedule_dealloc_carrier(Allctr_t *allctr, Carrier_t *crr) return; } + if (is_abandoned(crr)) + unlink_abandoned_carrier(crr); + if (crr->cpool.thr_prgr == ERTS_THR_PRGR_INVALID || erts_thr_progress_has_reached(crr->cpool.thr_prgr)) { dealloc_carrier(allctr, crr, 1); @@ -3112,6 +3335,8 @@ cpool_init_carrier_data(Allctr_t *allctr, Carrier_t *crr) limit = (csz/100)*allctr->cpool.util_limit; crr->cpool.abandon_limit = limit; } + crr->cpool.abandoned.next = NULL; + crr->cpool.abandoned.prev = NULL; } static void @@ -3142,6 +3367,9 @@ abandon_carrier(Allctr_t *allctr, Carrier_t *crr) STAT_MBC_CPOOL_INSERT(allctr, crr); unlink_carrier(&allctr->mbc_list, crr); + if (crr->cpool.orig_allctr == allctr) { + link_abandoned_carrier(&allctr->cpool.pooled_list, crr); + } allctr->remove_mbc(allctr, crr); @@ -3649,6 +3877,11 @@ destroy_carrier(Allctr_t *allctr, Block_t *blk, Carrier_t **busy_pcrr_pp) if (busy_pcrr_pp && *busy_pcrr_pp) { ERTS_ALC_CPOOL_ASSERT(*busy_pcrr_pp == crr); *busy_pcrr_pp = NULL; + ERTS_ALC_CPOOL_ASSERT(erts_smp_atomic_read_nob(&crr->allctr) + == (((erts_aint_t) allctr) + | ERTS_CRR_ALCTR_FLG_IN_POOL + | ERTS_CRR_ALCTR_FLG_BUSY)); + erts_smp_atomic_set_nob(&crr->allctr, ((erts_aint_t) allctr)); cpool_delete(allctr, allctr, crr); } else @@ -5528,6 +5761,10 @@ erts_alcu_start(Allctr_t *allctr, AllctrInit_t *init) allctr->min_block_size = sz; } + allctr->cpool.pooled_list.next = &allctr->cpool.pooled_list; + allctr->cpool.pooled_list.prev = &allctr->cpool.pooled_list; + allctr->cpool.traitor_list.next = &allctr->cpool.traitor_list; + allctr->cpool.traitor_list.prev = &allctr->cpool.traitor_list; allctr->cpool.dc_list.first = NULL; allctr->cpool.dc_list.last = NULL; allctr->cpool.abandon_limit = 0; diff --git a/erts/emulator/beam/erl_alloc_util.h b/erts/emulator/beam/erl_alloc_util.h index 7be6b1ed9d..eee920e66c 100644 --- a/erts/emulator/beam/erl_alloc_util.h +++ b/erts/emulator/beam/erl_alloc_util.h @@ -268,6 +268,11 @@ typedef union {char c[ERTS_ALLOC_ALIGN_BYTES]; long l; double d;} Unit_t; #ifdef ERTS_SMP +typedef struct ErtsDoubleLink_t_ { + struct ErtsDoubleLink_t_ *next; + struct ErtsDoubleLink_t_ *prev; +}ErtsDoubleLink_t; + typedef struct { erts_atomic_t next; erts_atomic_t prev; @@ -277,6 +282,7 @@ typedef struct { UWord abandon_limit; UWord blocks; UWord blocks_size; + ErtsDoubleLink_t abandoned; /* node in pooled_list or traitor_list */ } ErtsAlcCPoolData_t; #endif @@ -500,7 +506,12 @@ struct Allctr_t_ { CarrierList_t sbc_list; #ifdef ERTS_SMP struct { - CarrierList_t dc_list; + /* pooled_list, traitor list and dc_list contain only + carriers _created_ by this allocator */ + ErtsDoubleLink_t pooled_list; + ErtsDoubleLink_t traitor_list; + CarrierList_t dc_list; + UWord abandon_limit; int disable_abandon; int check_limit_count; diff --git a/erts/emulator/beam/erl_bif_info.c b/erts/emulator/beam/erl_bif_info.c index 6915765dab..b90362d82c 100644 --- a/erts/emulator/beam/erl_bif_info.c +++ b/erts/emulator/beam/erl_bif_info.c @@ -27,6 +27,7 @@ #include "erl_process.h" #include "error.h" #include "erl_driver.h" +#include "erl_nif.h" #include "bif.h" #include "big.h" #include "erl_version.h" @@ -90,7 +91,7 @@ static char erts_system_version[] = ("Erlang/OTP " ERLANG_OTP_RELEASE " [smp:%beu:%beu]" #endif #ifdef USE_THREADS -#ifdef ERTS_DIRTY_SCHEDULERS +#if defined(ERTS_DIRTY_SCHEDULERS) && defined(ERTS_SMP) " [ds:%beu:%beu:%beu]" #endif " [async-threads:%d]" @@ -2459,6 +2460,13 @@ BIF_RETTYPE system_info_1(BIF_ALIST_1) ERL_DRV_EXTENDED_MINOR_VERSION); hp = HAlloc(BIF_P, 2*n); BIF_RET(buf_to_intlist(&hp, buf, n, NIL)); + } else if (ERTS_IS_ATOM_STR("nif_version", BIF_ARG_1)) { + char buf[42]; + int n = erts_snprintf(buf, 42, "%d.%d", + ERL_NIF_MAJOR_VERSION, + ERL_NIF_MINOR_VERSION); + hp = HAlloc(BIF_P, 2*n); + BIF_RET(buf_to_intlist(&hp, buf, n, NIL)); } else if (ERTS_IS_ATOM_STR("smp_support", BIF_ARG_1)) { #ifdef ERTS_SMP BIF_RET(am_true); @@ -2696,6 +2704,9 @@ BIF_RETTYPE system_info_1(BIF_ALIST_1) ? am_disabled : am_enabled); } + else if (ERTS_IS_ATOM_STR("eager_check_io",BIF_ARG_1)) { + BIF_RET(erts_eager_check_io ? am_true : am_false); + } BIF_ERROR(BIF_P, BADARG); } @@ -3055,6 +3066,25 @@ fun_info_2(BIF_ALIST_2) return TUPLE2(hp, what, val); } +BIF_RETTYPE +fun_info_mfa_1(BIF_ALIST_1) +{ + Process* p = BIF_P; + Eterm fun = BIF_ARG_1; + Eterm* hp; + + if (is_fun(fun)) { + ErlFunThing* funp = (ErlFunThing *) fun_val(fun); + hp = HAlloc(p, 4); + BIF_RET(TUPLE3(hp,funp->fe->module,funp->fe->address[-2],make_small(funp->arity))); + } else if (is_export(fun)) { + Export* exp = (Export *) ((UWord) (export_val(fun))[1]); + hp = HAlloc(p, 4); + BIF_RET(TUPLE3(hp,exp->code[0],exp->code[1],make_small(exp->code[2]))); + } + BIF_ERROR(p, BADARG); +} + BIF_RETTYPE is_process_alive_1(BIF_ALIST_1) { if(is_internal_pid(BIF_ARG_1)) { @@ -3285,17 +3315,38 @@ BIF_RETTYPE erts_debug_get_internal_state_1(BIF_ALIST_1) BIF_RET(make_small((Uint) words)); } else if (ERTS_IS_ATOM_STR("check_io_debug", BIF_ARG_1)) { - /* Used by (emulator) */ - int res; + /* Used by driver_SUITE (emulator) */ + Uint sz, *szp; + Eterm res, *hp, **hpp; + int no_errors; + ErtsCheckIoDebugInfo ciodi = {0}; #ifdef HAVE_ERTS_CHECK_IO_DEBUG erts_smp_proc_unlock(BIF_P,ERTS_PROC_LOCK_MAIN); - res = erts_check_io_debug(); + no_errors = erts_check_io_debug(&ciodi); erts_smp_proc_lock(BIF_P,ERTS_PROC_LOCK_MAIN); #else - res = 0; + no_errors = 0; #endif - ASSERT(res >= 0); - BIF_RET(erts_make_integer((Uint) res, BIF_P)); + sz = 0; + szp = &sz; + hpp = NULL; + while (1) { + res = erts_bld_tuple(hpp, szp, 4, + erts_bld_uint(hpp, szp, + (Uint) no_errors), + erts_bld_uint(hpp, szp, + (Uint) ciodi.no_used_fds), + erts_bld_uint(hpp, szp, + (Uint) ciodi.no_driver_select_structs), + erts_bld_uint(hpp, szp, + (Uint) ciodi.no_driver_event_structs)); + if (hpp) + break; + hp = HAlloc(BIF_P, sz); + szp = NULL; + hpp = &hp; + } + BIF_RET(res); } else if (ERTS_IS_ATOM_STR("process_info_args", BIF_ARG_1)) { /* Used by process_SUITE (emulator) */ diff --git a/erts/emulator/beam/erl_bif_port.c b/erts/emulator/beam/erl_bif_port.c index afb33c1cdb..7ce950e090 100644 --- a/erts/emulator/beam/erl_bif_port.c +++ b/erts/emulator/beam/erl_bif_port.c @@ -472,7 +472,7 @@ cleanup_old_port_data(erts_aint_t data) ErtsPortDataHeap *pdhp = (ErtsPortDataHeap *) data; size_t size; ERTS_SMP_DATA_DEPENDENCY_READ_MEMORY_BARRIER; - size = sizeof(ErtsPortDataHeap) + pdhp->hsize*(sizeof(Eterm) - 1); + size = sizeof(ErtsPortDataHeap) + (pdhp->hsize-1)*sizeof(Eterm); erts_schedule_thr_prgr_later_cleanup_op(free_port_data_heap, (void *) pdhp, &pdhp->later_op, @@ -493,8 +493,8 @@ void erts_cleanup_port_data(Port *prt) { ASSERT(erts_atomic32_read_nob(&prt->state) & ERTS_PORT_SFLGS_INVALID_LOOKUP); - cleanup_old_port_data(erts_smp_atomic_read_nob(&prt->data)); - erts_smp_atomic_set_nob(&prt->data, (erts_aint_t) THE_NON_VALUE); + cleanup_old_port_data(erts_smp_atomic_xchg_nob(&prt->data, + (erts_aint_t) NULL)); } Uint @@ -508,7 +508,7 @@ erts_port_data_size(Port *prt) } else { ErtsPortDataHeap *pdhp = (ErtsPortDataHeap *) data; - return (Uint) sizeof(ErtsPortDataHeap) + pdhp->hsize*(sizeof(Eterm)-1); + return (Uint) sizeof(ErtsPortDataHeap) + (pdhp->hsize-1)*sizeof(Eterm); } } @@ -550,10 +550,11 @@ BIF_RETTYPE port_set_data_2(BIF_ALIST_2) hsize = size_object(BIF_ARG_2); pdhp = erts_alloc(ERTS_ALC_T_PORT_DATA_HEAP, - sizeof(ErtsPortDataHeap) + hsize*(sizeof(Eterm)-1)); + sizeof(ErtsPortDataHeap) + (hsize-1)*sizeof(Eterm)); hp = &pdhp->heap[0]; pdhp->off_heap.first = NULL; pdhp->off_heap.overhead = 0; + pdhp->hsize = hsize; pdhp->data = copy_struct(BIF_ARG_2, hsize, &hp, &pdhp->off_heap); data = (erts_aint_t) pdhp; ASSERT((data & 0x3) == 0); @@ -561,8 +562,14 @@ BIF_RETTYPE port_set_data_2(BIF_ALIST_2) data = erts_smp_atomic_xchg_wb(&prt->data, data); + if (data == (erts_aint_t)NULL) { + /* Port terminated by racing thread */ + data = erts_smp_atomic_xchg_wb(&prt->data, data); + ASSERT(data != (erts_aint_t)NULL); + cleanup_old_port_data(data); + BIF_ERROR(BIF_P, BADARG); + } cleanup_old_port_data(data); - BIF_RET(am_true); } @@ -581,6 +588,8 @@ BIF_RETTYPE port_get_data_1(BIF_ALIST_1) BIF_ERROR(BIF_P, BADARG); data = erts_smp_atomic_read_ddrb(&prt->data); + if (data == (erts_aint_t)NULL) + BIF_ERROR(BIF_P, BADARG); /* Port terminated by racing thread */ if ((data & 0x3) != 0) { res = (Eterm) (UWord) data; diff --git a/erts/emulator/beam/erl_driver.h b/erts/emulator/beam/erl_driver.h index 5ced8c5ca0..f9938fc66c 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 0 +#define ERL_DRV_EXTENDED_MINOR_VERSION 1 /* * The emulator will refuse to load a driver with a major version diff --git a/erts/emulator/beam/erl_drv_nif.h b/erts/emulator/beam/erl_drv_nif.h index 3f829ea7ea..4e8c6dc68b 100644 --- a/erts/emulator/beam/erl_drv_nif.h +++ b/erts/emulator/beam/erl_drv_nif.h @@ -35,6 +35,7 @@ typedef struct { int scheduler_threads; int nif_major_version; int nif_minor_version; + int dirty_scheduler_support; } ErlDrvSysInfo; typedef struct { diff --git a/erts/emulator/beam/erl_gc.c b/erts/emulator/beam/erl_gc.c index aa15d2cc57..0db42d4325 100644 --- a/erts/emulator/beam/erl_gc.c +++ b/erts/emulator/beam/erl_gc.c @@ -2018,6 +2018,20 @@ setup_rootset(Process *p, Eterm *objv, int nobj, Rootset *rootset) roots[n].sz = 1; n++; } + + /* + * If a NIF has saved arguments, they need to be added + */ + if (ERTS_PROC_GET_NIF_TRAP_EXPORT(p)) { + Eterm* argv; + int argc; + if (erts_setup_nif_gc(p, &argv, &argc)) { + roots[n].v = argv; + roots[n].sz = argc; + n++; + } + } + ASSERT(n <= rootset->size); mp = p->msg.first; diff --git a/erts/emulator/beam/erl_init.c b/erts/emulator/beam/erl_init.c index 88c4006934..61f8385efc 100644 --- a/erts/emulator/beam/erl_init.c +++ b/erts/emulator/beam/erl_init.c @@ -548,6 +548,8 @@ void erts_usage(void) erts_fprintf(stderr, " see the erl(1) documentation for more info.\n"); erts_fprintf(stderr, "-sct cput set cpu topology,\n"); erts_fprintf(stderr, " see the erl(1) documentation for more info.\n"); + erts_fprintf(stderr, "-secio bool enable/disable eager check I/O scheduling,\n"); + erts_fprintf(stderr, " see the erl(1) documentation for more info.\n"); #if ERTS_HAVE_SCHED_UTIL_BALANCING_SUPPORT_OPT erts_fprintf(stderr, "-sub bool enable/disable scheduler utilization balancing,\n"); #else @@ -1674,6 +1676,22 @@ erl_start(int argc, char **argv) erts_usage(); } } + else if (has_prefix("ecio", sub_param)) { + arg = get_arg(sub_param+4, argv[i+1], &i); +#ifndef __OSE__ + if (sys_strcmp("true", arg) == 0) + erts_eager_check_io = 1; + else +#endif + if (sys_strcmp("false", arg) == 0) + erts_eager_check_io = 0; + else { + erts_fprintf(stderr, + "bad schedule eager check I/O value '%s'\n", + arg); + erts_usage(); + } + } else if (has_prefix("pp", sub_param)) { arg = get_arg(sub_param+2, argv[i+1], &i); if (sys_strcmp(arg, "true") == 0) diff --git a/erts/emulator/beam/erl_lock_check.c b/erts/emulator/beam/erl_lock_check.c index c665aa51a2..b105ece6f1 100644 --- a/erts/emulator/beam/erl_lock_check.c +++ b/erts/emulator/beam/erl_lock_check.c @@ -139,7 +139,6 @@ static erts_lc_lock_order_t erts_lock_order[] = { { "drv_tsd", NULL }, { "async_enq_mtx", NULL }, #ifdef ERTS_SMP - { "sys_msg_q", NULL }, { "atom_tab", NULL }, { "make_ref", NULL }, { "misc_op_list_pre_alloc_lock", "address" }, @@ -148,6 +147,7 @@ static erts_lc_lock_order_t erts_lock_order[] = { { "btm_pre_alloc_lock", NULL, }, { "dist_entry_out_queue", "address" }, { "port_sched_lock", "port_id" }, + { "sys_msg_q", NULL }, { "port_table", NULL }, #endif { "mtrace_op", NULL }, diff --git a/erts/emulator/beam/erl_message.c b/erts/emulator/beam/erl_message.c index 59a677a12c..8870fac7d9 100644 --- a/erts/emulator/beam/erl_message.c +++ b/erts/emulator/beam/erl_message.c @@ -415,7 +415,13 @@ erts_queue_dist_message(Process *rcvr, if (!(*rcvr_locks & ERTS_PROC_LOCK_MSGQ)) erts_smp_proc_unlock(rcvr, ERTS_PROC_LOCK_MSGQ); - erts_proc_notify_new_message(rcvr); + erts_proc_notify_new_message(rcvr, +#ifdef ERTS_SMP + *rcvr_locks +#else + 0 +#endif + ); } } @@ -542,7 +548,13 @@ queue_message(Process *c_p, if (locked_msgq) erts_smp_proc_unlock(receiver, ERTS_PROC_LOCK_MSGQ); - erts_proc_notify_new_message(receiver); + erts_proc_notify_new_message(receiver, +#ifdef ERTS_SMP + *receiver_locks +#else + 0 +#endif + ); #ifndef ERTS_SMP ERTS_HOLE_CHECK(receiver); diff --git a/erts/emulator/beam/erl_nif.c b/erts/emulator/beam/erl_nif.c index ff551ea3af..adc3520ebb 100644 --- a/erts/emulator/beam/erl_nif.c +++ b/erts/emulator/beam/erl_nif.c @@ -472,6 +472,18 @@ int enif_inspect_binary(ErlNifEnv* env, Eterm bin_term, ErlNifBinary* bin) struct enif_tmp_obj_t* tmp; byte* raw_ptr; }u; + + if (is_boxed(bin_term) && *binary_val(bin_term) == HEADER_SUB_BIN) { + ErlSubBin* sb = (ErlSubBin*) binary_val(bin_term); + if (sb->is_writable) { + ProcBin* pb = (ProcBin*) binary_val(sb->orig); + ASSERT(pb->thing_word == HEADER_PROC_BIN); + if (pb->flags) { + erts_emasculate_writable_binary(pb); + sb->is_writable = 0; + } + } + } u.tmp = NULL; bin->data = erts_get_aligned_binary_bytes_extra(bin_term, &u.raw_ptr, allocator, sizeof(struct enif_tmp_obj_t)); @@ -1513,72 +1525,264 @@ int enif_consume_timeslice(ErlNifEnv* env, int percent) return ERTS_BIF_REDS_LEFT(env->proc) == 0; } -#ifdef ERTS_DIRTY_SCHEDULERS - -/* NIFs exports need one more item than the Export struct provides, the - * erl_module_nif*, so the DirtyNifExport below adds that. The Export - * member must be first in the struct. +/* + * NIF exports need a few more items than the Export struct provides, + * including the erl_module_nif* and a NIF function pointer, so the + * NifExport below adds those. The Export member must be first in the + * struct. The saved_mfa, saved_argc, nif_level, alloced_argv_sz and argv + * members are used to track the MFA and arguments of the top NIF in case a + * chain of one or more enif_schedule_nif() calls results in an exception, + * since in that case the original MFA and registers have to be restored + * before returning to Erlang to ensure stacktrace information associated + * with the exception is correct. */ +typedef ERL_NIF_TERM (*NativeFunPtr)(ErlNifEnv*, int, const ERL_NIF_TERM[]); + typedef struct { Export exp; struct erl_module_nif* m; -} DirtyNifExport; + NativeFunPtr fp; + Eterm saved_mfa[3]; + int saved_argc; + int alloced_argv_sz; + Eterm argv[1]; +} NifExport; -static void -alloc_proc_psd(Process* proc, DirtyNifExport **ep) +/* + * If a process has saved arguments, they need to be part of the GC + * rootset. The function below is called from setup_rootset() in + * erl_gc.c. This function is declared in erl_process.h. + */ +int +erts_setup_nif_gc(Process* proc, Eterm** objv, int* nobj) { + NifExport* ep = (NifExport*) ERTS_PROC_GET_NIF_TRAP_EXPORT(proc); + int gc = (ep && ep->saved_argc > 0); + + if (gc) { + *objv = ep->argv; + *nobj = ep->saved_argc; + } + return gc; +} + +/* + * Allocate a NifExport and set it in proc specific data + */ +static NifExport* +allocate_nif_sched_data(Process* proc, int argc) +{ + NifExport* ep; + size_t argv_extra, total; int i; - if (!*ep) { - *ep = erts_alloc(ERTS_ALC_T_PSD, sizeof(DirtyNifExport)); - sys_memset((void*) *ep, 0, sizeof(DirtyNifExport)); - for (i=0; i<ERTS_NUM_CODE_IX; i++) { - (*ep)->exp.addressv[i] = &(*ep)->exp.code[3]; - } - (*ep)->exp.code[3] = (BeamInstr) em_call_nif; + + argv_extra = argc > 1 ? sizeof(Eterm)*(argc-1) : 0; + total = sizeof(NifExport) + argv_extra; + ep = erts_alloc(ERTS_ALC_T_NIF_TRAP_EXPORT, total); + sys_memset((void*) ep, 0, total); + ep->alloced_argv_sz = argc; + for (i=0; i<ERTS_NUM_CODE_IX; i++) { + ep->exp.addressv[i] = &ep->exp.code[3]; } - (void) ERTS_PROC_SET_DIRTY_SCHED_TRAP_EXPORT(proc, ERTS_PROC_LOCK_MAIN, &(*ep)->exp); + ep->exp.code[3] = (BeamInstr) em_call_nif; + (void) ERTS_PROC_SET_NIF_TRAP_EXPORT(proc, ERTS_PROC_LOCK_MAIN, ep); + return ep; +} + +static ERTS_INLINE void +destroy_nif_export(NifExport *nif_export) +{ + erts_free(ERTS_ALC_T_NIF_TRAP_EXPORT, (void *) nif_export); } +void +erts_destroy_nif_export(void *nif_export) +{ + destroy_nif_export((NifExport *) nif_export); +} + +/* + * Initialize a NifExport struct. Create it if needed and store it in the + * proc. The direct_fp function is what will be invoked by op_call_nif, and + * the indirect_fp function, if not NULL, is what the direct_fp function + * will call. If the allocated NifExport isn't enough to hold all of argv, + * allocate a larger one. Save MFA and registers only if the need_save + * parameter is true. + */ static ERL_NIF_TERM -execute_dirty_nif_finalizer(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]) +init_nif_sched_data(ErlNifEnv* env, NativeFunPtr direct_fp, NativeFunPtr indirect_fp, + int need_save, int argc, const ERL_NIF_TERM argv[]) { - Eterm* reg = ERTS_PROC_GET_SCHDATA(env->proc)->x_reg_array; - ERL_NIF_TERM result, dirty_result = (ERL_NIF_TERM) reg[0]; - typedef ERL_NIF_TERM (*FinalizerFP)(ErlNifEnv*, ERL_NIF_TERM); - FinalizerFP fp; -#if HAVE_INT64 && SIZEOF_LONG != 8 - ASSERT(sizeof(fp) <= sizeof(ErlNifUInt64)); - enif_get_uint64(env, reg[1], (ErlNifUInt64 *) &fp); -#else - ASSERT(sizeof(fp) <= sizeof(unsigned long)); - enif_get_ulong(env, reg[1], (unsigned long *) &fp); -#endif - result = (*fp)(env, dirty_result); - if (erts_refc_dectest(&env->mod_nif->rt_dtor_cnt, 0) == 0 - && env->mod_nif->mod == NULL) - close_lib(env->mod_nif); - return result; + Process* proc = env->proc; + Eterm* reg = ERTS_PROC_GET_SCHDATA(proc)->x_reg_array; + NifExport* ep; + int i; + + ep = (NifExport*) ERTS_PROC_GET_NIF_TRAP_EXPORT(proc); + if (!ep) + ep = allocate_nif_sched_data(proc, argc); + else if (need_save && ep->alloced_argv_sz < argc) { + NifExport* new_ep = allocate_nif_sched_data(proc, argc); + destroy_nif_export(ep); + ep = new_ep; + } + ERTS_VBUMP_ALL_REDS(proc); + for (i = 0; i < argc; i++) { + if (need_save) + ep->argv[i] = reg[i]; + reg[i] = (Eterm) argv[i]; + } + if (need_save) { + ep->saved_mfa[0] = proc->current[0]; + ep->saved_mfa[1] = proc->current[1]; + ep->saved_mfa[2] = proc->current[2]; + ep->saved_argc = argc; + } + proc->i = (BeamInstr*) ep->exp.addressv[0]; + ep->exp.code[0] = (BeamInstr) proc->current[0]; + ep->exp.code[1] = (BeamInstr) proc->current[1]; + ep->exp.code[2] = argc; + ep->exp.code[4] = (BeamInstr) direct_fp; + ep->m = env->mod_nif; + ep->fp = indirect_fp; + proc->freason = TRAP; + proc->arity = argc; + return THE_NON_VALUE; } -#endif /* ERTS_DIRTY_SCHEDULERS */ +/* + * Restore saved MFA and registers. Registers are restored only when the + * exception flag is true. + */ +static void +restore_nif_mfa(Process* proc, NifExport* ep, int exception) +{ + int i; + Eterm* reg = ERTS_PROC_GET_SCHDATA(proc)->x_reg_array; -#ifdef ERL_NIF_DIRTY_SCHEDULER_SUPPORT + proc->current[0] = ep->saved_mfa[0]; + proc->current[1] = ep->saved_mfa[1]; + proc->current[2] = ep->saved_mfa[2]; + if (exception) + for (i = 0; i < ep->saved_argc; i++) + reg[i] = ep->argv[i]; + ep->saved_argc = 0; + ep->saved_mfa[0] = THE_NON_VALUE; +} -ERL_NIF_TERM -enif_schedule_dirty_nif(ErlNifEnv* env, int flags, - ERL_NIF_TERM (*fp)(ErlNifEnv*, int, const ERL_NIF_TERM[]), - int argc, const ERL_NIF_TERM argv[]) +#ifdef ERTS_DIRTY_SCHEDULERS + +/* + * Finalize a dirty NIF call. This function is scheduled to cause the VM to + * switch the process off a dirty scheduler thread and back onto a regular + * scheduler thread, and then return the result from the dirty NIF. It also + * restores the original NIF MFA when necessary based on the value of + * ep->fp set by execute_dirty_nif via init_nif_sched_data -- non-NULL + * means restore, NULL means do not restore. + */ +static ERL_NIF_TERM +dirty_nif_finalizer(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]) +{ + Process* proc = env->proc; + NifExport* ep; + + ASSERT(argc == 1); + ASSERT(!ERTS_SCHEDULER_IS_DIRTY(env->proc->scheduler_data)); + ep = (NifExport*) ERTS_PROC_GET_NIF_TRAP_EXPORT(proc); + ASSERT(ep); + if (ep->fp) + restore_nif_mfa(proc, ep, 0); + return argv[0]; +} + +/* Finalize a dirty NIF call that raised an exception. Otherwise same as + * the dirty_nif_finalizer() function. + */ +static ERL_NIF_TERM +dirty_nif_exception(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]) +{ + Process* proc = env->proc; + NifExport* ep; + + ASSERT(!ERTS_SCHEDULER_IS_DIRTY(env->proc->scheduler_data)); + ep = (NifExport*) ERTS_PROC_GET_NIF_TRAP_EXPORT(proc); + ASSERT(ep); + if (ep->fp) + restore_nif_mfa(proc, ep, 1); + return enif_make_badarg(env); +} + +/* + * Dirty NIF execution wrapper function. Invoke an application's dirty NIF, + * then check the result and schedule the appropriate finalizer function + * where needed. Also restore the original NIF MFA when appropriate. + */ +static ERL_NIF_TERM +execute_dirty_nif(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]) +{ + Process* proc = env->proc; + NativeFunPtr fp = (NativeFunPtr) proc->current[6]; + NifExport* ep; + ERL_NIF_TERM result; + + ASSERT(ERTS_SCHEDULER_IS_DIRTY(env->proc->scheduler_data)); + + /* + * Set ep->fp to NULL before the native call so we know later whether it scheduled another NIF for execution + */ + ep = (NifExport*) ERTS_PROC_GET_NIF_TRAP_EXPORT(proc); + ASSERT(ep); + ep->fp = NULL; + result = (*fp)(env, argc, argv); + erts_smp_atomic32_read_band_mb(&proc->state, + ~(ERTS_PSFLG_DIRTY_CPU_PROC + |ERTS_PSFLG_DIRTY_IO_PROC + |ERTS_PSFLG_DIRTY_CPU_PROC_IN_Q + |ERTS_PSFLG_DIRTY_IO_PROC_IN_Q)); + if (erts_refc_dectest(&env->mod_nif->rt_dtor_cnt, 0) == 0 && env->mod_nif->mod == NULL) + close_lib(env->mod_nif); + /* + * If no more NIFs were scheduled by the native call via + * enif_schedule_nif(), then ep->fp will still be NULL as set above, in + * which case we need to restore the original NIF calling + * context. Reuse fp essentially as a boolean for this, passing it to + * init_nif_sched_data below. Both dirty_nif_exception and + * dirty_nif_finalizer then check ep->fp to decide whether or not to + * restore the original calling context. + */ + ep = (NifExport*) ERTS_PROC_GET_NIF_TRAP_EXPORT(proc); + ASSERT(ep); + if (ep->fp) + fp = NULL; + if (is_non_value(result)) { + if (proc->freason != TRAP) { + ASSERT(proc->freason == BADARG); + return init_nif_sched_data(env, dirty_nif_exception, fp, 0, argc, argv); + } else { + if (ep->fp == NULL) + restore_nif_mfa(proc, ep, 1); + return result; + } + } + else + return init_nif_sched_data(env, dirty_nif_finalizer, fp, 0, 1, &result); +} + +/* + * Dirty NIF scheduling wrapper function. Schedule a dirty NIF to execute + * via the execute_dirty_nif() wrapper function. The dirty scheduler thread + * type (CPU or I/O) is indicated in flags parameter. + */ +static ERTS_INLINE ERL_NIF_TERM +schedule_dirty_nif(ErlNifEnv* env, int flags, int argc, const ERL_NIF_TERM argv[]) { -#ifdef USE_THREADS erts_aint32_t state, n, a; Process* proc = env->proc; - Eterm* reg = ERTS_PROC_GET_SCHDATA(proc)->x_reg_array; - DirtyNifExport* ep = NULL; - int i; + NativeFunPtr fp = (NativeFunPtr) proc->current[6]; + NifExport* ep; + int need_save; - int chkflgs = (flags & (ERL_NIF_DIRTY_JOB_IO_BOUND|ERL_NIF_DIRTY_JOB_CPU_BOUND)); - if (chkflgs != ERL_NIF_DIRTY_JOB_IO_BOUND && chkflgs != ERL_NIF_DIRTY_JOB_CPU_BOUND) - return enif_make_badarg(env); + ASSERT(flags==ERL_NIF_DIRTY_JOB_IO_BOUND || flags==ERL_NIF_DIRTY_JOB_CPU_BOUND); a = erts_smp_atomic32_read_acqb(&proc->state); while (1) { @@ -1590,7 +1794,7 @@ enif_schedule_dirty_nif(ErlNifEnv* env, int flags, */ n &= ~(ERTS_PSFLG_DIRTY_CPU_PROC|ERTS_PSFLG_DIRTY_IO_PROC |ERTS_PSFLG_DIRTY_CPU_PROC_IN_Q|ERTS_PSFLG_DIRTY_IO_PROC_IN_Q); - if (chkflgs == ERL_NIF_DIRTY_JOB_CPU_BOUND) + if (flags == ERL_NIF_DIRTY_JOB_CPU_BOUND) n |= ERTS_PSFLG_DIRTY_CPU_PROC; else n |= ERTS_PSFLG_DIRTY_IO_PROC; @@ -1598,85 +1802,106 @@ enif_schedule_dirty_nif(ErlNifEnv* env, int flags, if (a == state) break; } - if (!(ep = (DirtyNifExport*) ERTS_PROC_GET_DIRTY_SCHED_TRAP_EXPORT(proc))) - alloc_proc_psd(proc, &ep); - ERTS_VBUMP_ALL_REDS(proc); - ep->exp.code[2] = argc; - for (i = 0; i < argc; i++) { - reg[i] = (Eterm) argv[i]; - } - proc->i = (BeamInstr*) ep->exp.addressv[0]; - ep->exp.code[4] = (BeamInstr) fp; - ep->m = env->mod_nif; - proc->freason = TRAP; - erts_refc_inc(&env->mod_nif->rt_dtor_cnt, 1); + ep = (NifExport*) ERTS_PROC_GET_NIF_TRAP_EXPORT(proc); + need_save = (ep == NULL || is_non_value(ep->saved_mfa[0])); + return init_nif_sched_data(env, execute_dirty_nif, fp, need_save, argc, argv); +} - return THE_NON_VALUE; -#else - return (*fp)(env, argc, argv); -#endif +static ERL_NIF_TERM +schedule_dirty_io_nif(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]) +{ + return schedule_dirty_nif(env, ERL_NIF_DIRTY_JOB_IO_BOUND, argc, argv); +} + +static ERL_NIF_TERM +schedule_dirty_cpu_nif(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]) +{ + return schedule_dirty_nif(env, ERL_NIF_DIRTY_JOB_CPU_BOUND, argc, argv); +} + +#endif /* ERTS_DIRTY_SCHEDULERS */ + +/* + * NIF execution wrapper used by enif_schedule_nif() for regular NIFs. It + * calls the actual NIF, restores original NIF MFA if necessary, and + * then returns the NIF result. + */ +static ERL_NIF_TERM +execute_nif(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]) +{ + Process* proc = env->proc; + NativeFunPtr fp = (NativeFunPtr) proc->current[6]; + NifExport* ep; + ERL_NIF_TERM result; + + ep = (NifExport*) ERTS_PROC_GET_NIF_TRAP_EXPORT(proc); + ASSERT(ep); + ep->fp = NULL; + result = (*fp)(env, argc, argv); + ep = (NifExport*) ERTS_PROC_GET_NIF_TRAP_EXPORT(proc); + ASSERT(ep); + /* + * If no NIFs were scheduled by the native call via + * enif_schedule_nif(), then ep->fp will still be NULL as set above, in + * which case we need to restore the original NIF MFA. + */ + if (ep->fp == NULL) + restore_nif_mfa(proc, ep, is_non_value(result) && proc->freason != TRAP); + return result; } ERL_NIF_TERM -enif_schedule_dirty_nif_finalizer(ErlNifEnv* env, ERL_NIF_TERM result, - ERL_NIF_TERM (*fp)(ErlNifEnv*, ERL_NIF_TERM)) +enif_schedule_nif(ErlNifEnv* env, const char* fun_name, int flags, + ERL_NIF_TERM (*fp)(ErlNifEnv*, int, const ERL_NIF_TERM[]), + int argc, const ERL_NIF_TERM argv[]) { -#ifdef USE_THREADS Process* proc = env->proc; - Eterm* reg = ERTS_PROC_GET_SCHDATA(proc)->x_reg_array; - DirtyNifExport* ep; + NifExport* ep; + ERL_NIF_TERM fun_name_atom, result; + int need_save; - erts_smp_atomic32_read_band_mb(&proc->state, - ~(ERTS_PSFLG_DIRTY_CPU_PROC - |ERTS_PSFLG_DIRTY_IO_PROC - |ERTS_PSFLG_DIRTY_CPU_PROC_IN_Q - |ERTS_PSFLG_DIRTY_IO_PROC_IN_Q)); - if (!(ep = (DirtyNifExport*) ERTS_PROC_GET_DIRTY_SCHED_TRAP_EXPORT(proc))) - alloc_proc_psd(proc, &ep); - ERTS_VBUMP_ALL_REDS(proc); - ep->exp.code[2] = 2; - reg[0] = (Eterm) result; -#if HAVE_INT64 && SIZEOF_LONG != 8 - ASSERT(sizeof(fp) <= sizeof(ErlNifUInt64)); - reg[1] = (Eterm) enif_make_uint64(env, (ErlNifUInt64) fp); -#else - ASSERT(sizeof(fp) <= sizeof(unsigned long)); - reg[1] = (Eterm) enif_make_ulong(env, (unsigned long) fp); -#endif - proc->i = (BeamInstr*) ep->exp.addressv[0]; - ep->exp.code[4] = (BeamInstr) execute_dirty_nif_finalizer; - proc->freason = TRAP; + if (argc > MAX_ARG) + return enif_make_badarg(env); + fun_name_atom = enif_make_atom(env, fun_name); + if (enif_is_exception(env, fun_name_atom)) + return fun_name_atom; - return THE_NON_VALUE; + ep = (NifExport*) ERTS_PROC_GET_NIF_TRAP_EXPORT(proc); + need_save = (ep == NULL || is_non_value(ep->saved_mfa[0])); + + if (flags) { +#ifdef ERTS_DIRTY_SCHEDULERS + NativeFunPtr sched_fun; + int chkflgs = (flags & (ERL_NIF_DIRTY_JOB_IO_BOUND|ERL_NIF_DIRTY_JOB_CPU_BOUND)); + if (chkflgs == ERL_NIF_DIRTY_JOB_IO_BOUND) + sched_fun = schedule_dirty_io_nif; + else if (chkflgs == ERL_NIF_DIRTY_JOB_CPU_BOUND) + sched_fun = schedule_dirty_cpu_nif; + else + return enif_make_badarg(env); + result = init_nif_sched_data(env, sched_fun, fp, need_save, argc, argv); #else - return (*fp)(env, result); + return enif_make_badarg(env); #endif -} + } + else + result = init_nif_sched_data(env, execute_nif, fp, need_save, argc, argv); -/* A simple finalizer that just returns its result argument */ -ERL_NIF_TERM -enif_dirty_nif_finalizer(ErlNifEnv* env, ERL_NIF_TERM result) -{ + ep = (NifExport*) ERTS_PROC_GET_NIF_TRAP_EXPORT(proc); + ASSERT(ep); + ep->exp.code[1] = (BeamInstr) fun_name_atom; return result; } +#ifdef ERL_NIF_DIRTY_SCHEDULER_SUPPORT + int enif_is_on_dirty_scheduler(ErlNifEnv* env) { return ERTS_SCHEDULER_IS_DIRTY(env->proc->scheduler_data); } -int -enif_have_dirty_schedulers() -{ -#ifdef USE_THREADS - return 1; -#else - return 0; -#endif -} - #endif /* ERL_NIF_DIRTY_SCHEDULER_SUPPORT */ /* Maps */ @@ -1977,6 +2202,35 @@ static Eterm load_nif_error(Process* p, const char* atom, const char* format, .. return ret; } +/* + * The function below is for looping through ErlNifFunc arrays, helping + * provide backwards compatibility across the version 2.7 change that added + * the "flags" field to ErlNifFunc. + */ +static ErlNifFunc* next_func(ErlNifEntry* entry, int* incrp, ErlNifFunc* func) +{ + ASSERT(incrp); + if (!*incrp) { + if (entry->major > 2 || (entry->major == 2 && entry->minor >= 7)) + *incrp = sizeof(ErlNifFunc); + else { + /* + * ErlNifFuncV1 below is what ErlNifFunc was before the + * addition of the flags field for 2.7, and is needed to handle + * backward compatibility. + */ + typedef struct { + const char* name; + unsigned arity; + ERL_NIF_TERM (*fptr)(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]); + }ErlNifFuncV1; + *incrp = sizeof(ErlNifFuncV1); + } + } + return (ErlNifFunc*) ((char*)func + *incrp); +} + + BIF_RETTYPE load_nif_2(BIF_ALIST_2) { static const char bad_lib[] = "bad_lib"; @@ -2086,22 +2340,48 @@ BIF_RETTYPE load_nif_2(BIF_ALIST_2) } else { /*erts_fprintf(stderr, "Found module %T\r\n", mod_atom);*/ - + + int maybe_dirty_nifs = ((entry->major > 2 || (entry->major == 2 && entry->minor >= 7)) + && (entry->options & ERL_NIF_DIRTY_NIF_OPTION)); + int incr = 0; + ErlNifFunc* f = entry->funcs; for (i=0; i < entry->num_of_funcs && ret==am_ok; i++) { BeamInstr** code_pp; - ErlNifFunc* f = &entry->funcs[i]; if (!erts_atom_get(f->name, sys_strlen(f->name), &f_atom, ERTS_ATOM_ENC_LATIN1) || (code_pp = get_func_pp(mod->curr.code, f_atom, f->arity))==NULL) { ret = load_nif_error(BIF_P,bad_lib,"Function not found %T:%s/%u", mod_atom, f->name, f->arity); - } - else if (code_pp[1] - code_pp[0] < (5+3)) { + } + else if (maybe_dirty_nifs && f->flags) { + /* + * If the flags field is non-zero and this emulator was + * built with dirty scheduler support, check that the flags + * value is legal. But if this emulator was built without + * dirty scheduler support, treat a non-zero flags field as + * a load error. + */ +#ifdef ERTS_DIRTY_SCHEDULERS + if (f->flags != ERL_NIF_DIRTY_JOB_IO_BOUND && f->flags != ERL_NIF_DIRTY_JOB_CPU_BOUND) + ret = load_nif_error(BIF_P, bad_lib, "Illegal flags field value %d for NIF %T:%s/%u", + f->flags, mod_atom, f->name, f->arity); +#else + ret = load_nif_error(BIF_P, bad_lib, "NIF %T:%s/%u requires a runtime with dirty scheduler support.", + mod_atom, f->name, f->arity); +#endif + } +#ifdef ERTS_DIRTY_SCHEDULERS + else if (code_pp[1] - code_pp[0] < (5+4)) +#else + else if (code_pp[1] - code_pp[0] < (5+3)) +#endif + { ret = load_nif_error(BIF_P,bad_lib,"No explicit call to load_nif" - " in module (%T:%s/%u to small)", - mod_atom, entry->funcs[i].name, entry->funcs[i].arity); + " in module (%T:%s/%u too small)", + mod_atom, f->name, f->arity); } /*erts_fprintf(stderr, "Found NIF %T:%s/%u\r\n", - mod_atom, entry->funcs[i].name, entry->funcs[i].arity);*/ + mod_atom, f->name, f->arity);*/ + f = next_func(entry, &incr, f); } } @@ -2127,7 +2407,8 @@ BIF_RETTYPE load_nif_2(BIF_ALIST_2) * is deprecated and was only ment as a development feature not to * be used in production systems. (See warning below) */ - int k; + int k, old_incr = 0; + ErlNifFunc* old_func; lib->priv_data = mod->curr.nif->priv_data; ASSERT(mod->curr.nif->entry != NULL); @@ -2136,13 +2417,16 @@ BIF_RETTYPE load_nif_2(BIF_ALIST_2) goto error; } /* Check that no NIF is removed */ + old_func = mod->curr.nif->entry->funcs; for (k=0; k < mod->curr.nif->entry->num_of_funcs; k++) { - ErlNifFunc* old_func = &mod->curr.nif->entry->funcs[k]; + int incr = 0; + ErlNifFunc* f = entry->funcs; for (i=0; i < entry->num_of_funcs; i++) { - if (old_func->arity == entry->funcs[i].arity - && sys_strcmp(old_func->name, entry->funcs[i].name) == 0) { + if (old_func->arity == f->arity + && sys_strcmp(old_func->name, f->name) == 0) { break; } + f = next_func(entry, &incr, f); } if (i == entry->num_of_funcs) { ret = load_nif_error(BIF_P,reload,"Reloaded library missing " @@ -2150,7 +2434,8 @@ BIF_RETTYPE load_nif_2(BIF_ALIST_2) old_func->name, old_func->arity); goto error; } - } + old_func = next_func(mod->curr.nif->entry, &old_incr, old_func); + } erts_pre_nif(&env, BIF_P, lib); veto = entry->reload(&env, &lib->priv_data, BIF_ARG_2); erts_post_nif(&env); @@ -2197,13 +2482,17 @@ BIF_RETTYPE load_nif_2(BIF_ALIST_2) /* ** Everything ok, patch the beam code with op_call_nif */ - mod->curr.nif = lib; + + int incr = 0; + ErlNifFunc* f = entry->funcs; + + mod->curr.nif = lib; for (i=0; i < entry->num_of_funcs; i++) { BeamInstr* code_ptr; - erts_atom_get(entry->funcs[i].name, sys_strlen(entry->funcs[i].name), &f_atom, ERTS_ATOM_ENC_LATIN1); - code_ptr = *get_func_pp(mod->curr.code, f_atom, entry->funcs[i].arity); - + erts_atom_get(f->name, sys_strlen(f->name), &f_atom, ERTS_ATOM_ENC_LATIN1); + code_ptr = *get_func_pp(mod->curr.code, f_atom, f->arity); + if (code_ptr[1] == 0) { code_ptr[5+0] = (BeamInstr) BeamOp(op_call_nif); } @@ -2211,10 +2500,21 @@ BIF_RETTYPE load_nif_2(BIF_ALIST_2) GenericBp* g = (GenericBp *) code_ptr[1]; ASSERT(code_ptr[5+0] == (BeamInstr) BeamOp(op_i_generic_breakpoint)); - g->orig_instr = (BeamInstr) BeamOp(op_call_nif); - } - code_ptr[5+1] = (BeamInstr) entry->funcs[i].fptr; + g->orig_instr = (BeamInstr) BeamOp(op_call_nif); + } + if ((entry->major > 2 || (entry->major == 2 && entry->minor >= 7)) + && (entry->options & ERL_NIF_DIRTY_NIF_OPTION) && f->flags) { +#ifdef ERL_NIF_DIRTY_SCHEDULER_SUPPORT + code_ptr[5+3] = (BeamInstr) f->fptr; + code_ptr[5+1] = (f->flags == ERL_NIF_DIRTY_JOB_IO_BOUND) ? + (BeamInstr) schedule_dirty_io_nif : + (BeamInstr) schedule_dirty_cpu_nif; +#endif + } + else + code_ptr[5+1] = (BeamInstr) f->fptr; code_ptr[5+2] = (BeamInstr) lib; + f = next_func(entry, &incr, f); } } else { diff --git a/erts/emulator/beam/erl_nif.h b/erts/emulator/beam/erl_nif.h index 5b93c2398e..849024453c 100644 --- a/erts/emulator/beam/erl_nif.h +++ b/erts/emulator/beam/erl_nif.h @@ -42,9 +42,13 @@ ** 2.5: R17 Maps API additions ** 2.6: R17 with maps ** R17 dirty schedulers +** 2.7: 17.3 add enif_schedule_nif +** remove enif_schedule_dirty_nif, enif_schedule_dirty_nif_finalizer, enif_dirty_nif_finalizer +** add ErlNifEntry options +** add ErlNifFunc flags */ #define ERL_NIF_MAJOR_VERSION 2 -#define ERL_NIF_MINOR_VERSION 6 +#define ERL_NIF_MINOR_VERSION 7 /* * The emulator will refuse to load a nif-lib with a major version @@ -125,8 +129,10 @@ typedef struct const char* name; unsigned arity; ERL_NIF_TERM (*fptr)(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]); + unsigned flags; }ErlNifFunc; + typedef struct enif_entry_t { int major; @@ -139,8 +145,11 @@ typedef struct enif_entry_t int (*upgrade)(ErlNifEnv*, void** priv_data, void** old_priv_data, ERL_NIF_TERM load_info); void (*unload) (ErlNifEnv*, void* priv_data); const char* vm_variant; + unsigned options; }ErlNifEntry; +/* Field bits for ErlNifEntry options */ +#define ERL_NIF_DIRTY_NIF_OPTION 1 typedef struct @@ -243,6 +252,11 @@ extern TWinDynNifCallbacks WinDynNifCallbacks; # endif #endif +#ifdef ERL_NIF_DIRTY_SCHEDULER_SUPPORT +# define ERL_NIF_ENTRY_OPTIONS ERL_NIF_DIRTY_NIF_OPTION +#else +# define ERL_NIF_ENTRY_OPTIONS 0 +#endif #ifdef __cplusplus } @@ -268,7 +282,8 @@ ERL_NIF_INIT_DECL(NAME) \ sizeof(FUNCS) / sizeof(*FUNCS), \ FUNCS, \ LOAD, RELOAD, UPGRADE, UNLOAD, \ - ERL_NIF_VM_VARIANT \ + ERL_NIF_VM_VARIANT, \ + ERL_NIF_ENTRY_OPTIONS \ }; \ ERL_NIF_INIT_BODY; \ return &entry; \ diff --git a/erts/emulator/beam/erl_nif_api_funcs.h b/erts/emulator/beam/erl_nif_api_funcs.h index d7c554e60b..630cefae93 100644 --- a/erts/emulator/beam/erl_nif_api_funcs.h +++ b/erts/emulator/beam/erl_nif_api_funcs.h @@ -22,7 +22,7 @@ #endif /* -** WARNING: add new ERL_NIF_API_FUNC_DECL entries at the bottom of the list +** WARNING: Add new ERL_NIF_API_FUNC_DECL entries at the bottom of the list ** to keep compatibility on Windows!!! ** ** And don't forget to increase ERL_NIF_MINOR_VERSION in erl_nif.h @@ -141,14 +141,6 @@ ERL_NIF_API_FUNC_DECL(int,enif_is_number,(ErlNifEnv*, ERL_NIF_TERM term)); ERL_NIF_API_FUNC_DECL(void*,enif_dlopen,(const char* lib, void (*err_handler)(void*,const char*), void* err_arg)); ERL_NIF_API_FUNC_DECL(void*,enif_dlsym,(void* handle, const char* symbol, void (*err_handler)(void*,const char*), void* err_arg)); ERL_NIF_API_FUNC_DECL(int,enif_consume_timeslice,(ErlNifEnv*, int percent)); -#ifdef ERL_NIF_DIRTY_SCHEDULER_SUPPORT -ERL_NIF_API_FUNC_DECL(ERL_NIF_TERM,enif_schedule_dirty_nif,(ErlNifEnv*,int,ERL_NIF_TERM (*)(ErlNifEnv*,int,const ERL_NIF_TERM[]),int,const ERL_NIF_TERM[])); -ERL_NIF_API_FUNC_DECL(ERL_NIF_TERM,enif_schedule_dirty_nif_finalizer,(ErlNifEnv*,ERL_NIF_TERM,ERL_NIF_TERM (*)(ErlNifEnv*,ERL_NIF_TERM))); -ERL_NIF_API_FUNC_DECL(ERL_NIF_TERM,enif_dirty_nif_finalizer,(ErlNifEnv*,ERL_NIF_TERM)); -ERL_NIF_API_FUNC_DECL(int,enif_is_on_dirty_scheduler,(ErlNifEnv*)); -ERL_NIF_API_FUNC_DECL(int,enif_have_dirty_schedulers,(void)); -#endif - ERL_NIF_API_FUNC_DECL(int, enif_is_map, (ErlNifEnv* env, ERL_NIF_TERM term)); ERL_NIF_API_FUNC_DECL(int, enif_get_map_size, (ErlNifEnv* env, ERL_NIF_TERM term, size_t *size)); ERL_NIF_API_FUNC_DECL(ERL_NIF_TERM, enif_make_new_map, (ErlNifEnv* env)); @@ -163,12 +155,22 @@ ERL_NIF_API_FUNC_DECL(int, enif_map_iterator_is_tail, (ErlNifEnv *env, ErlNifMap ERL_NIF_API_FUNC_DECL(int, enif_map_iterator_next, (ErlNifEnv *env, ErlNifMapIterator *iter)); ERL_NIF_API_FUNC_DECL(int, enif_map_iterator_prev, (ErlNifEnv *env, ErlNifMapIterator *iter)); ERL_NIF_API_FUNC_DECL(int, enif_map_iterator_get_pair, (ErlNifEnv *env, ErlNifMapIterator *iter, ERL_NIF_TERM *key, ERL_NIF_TERM *value)); - +ERL_NIF_API_FUNC_DECL(ERL_NIF_TERM,enif_schedule_nif,(ErlNifEnv*,const char*,int,ERL_NIF_TERM (*)(ErlNifEnv*,int,const ERL_NIF_TERM[]),int,const ERL_NIF_TERM[])); /* -** Add new entries here to keep compatibility on Windows!!! +** ADD NEW ENTRIES HERE (before this comment) !!! */ + + +/* + * Conditional EXPERIMENTAL stuff always last. + * Must be moved up and made unconditional to support binary backward + * compatibility on Windows. + */ +#ifdef ERL_NIF_DIRTY_SCHEDULER_SUPPORT +ERL_NIF_API_FUNC_DECL(int,enif_is_on_dirty_scheduler,(ErlNifEnv*)); #endif +#endif /* ERL_NIF_API_FUNC_DECL */ /* ** Please keep the ERL_NIF_API_FUNC_MACRO list below in the same order @@ -282,21 +284,12 @@ ERL_NIF_API_FUNC_DECL(int, enif_map_iterator_get_pair, (ErlNifEnv *env, ErlNifMa # define enif_make_int64 ERL_NIF_API_FUNC_MACRO(enif_make_int64) # define enif_make_uint64 ERL_NIF_API_FUNC_MACRO(enif_make_uint64) #endif - # define enif_is_exception ERL_NIF_API_FUNC_MACRO(enif_is_exception) # define enif_make_reverse_list ERL_NIF_API_FUNC_MACRO(enif_make_reverse_list) # define enif_is_number ERL_NIF_API_FUNC_MACRO(enif_is_number) # define enif_dlopen ERL_NIF_API_FUNC_MACRO(enif_dlopen) # define enif_dlsym ERL_NIF_API_FUNC_MACRO(enif_dlsym) # define enif_consume_timeslice ERL_NIF_API_FUNC_MACRO(enif_consume_timeslice) -#ifdef ERL_NIF_DIRTY_SCHEDULER_SUPPORT -# define enif_schedule_dirty_nif ERL_NIF_API_FUNC_MACRO(enif_schedule_dirty_nif) -# define enif_schedule_dirty_nif_finalizer ERL_NIF_API_FUNC_MACRO(enif_schedule_dirty_nif_finalizer) -# define enif_dirty_nif_finalizer ERL_NIF_API_FUNC_MACRO(enif_dirty_nif_finalizer) -# define enif_is_on_dirty_scheduler ERL_NIF_API_FUNC_MACRO(enif_is_on_dirty_scheduler) -# define enif_have_dirty_schedulers ERL_NIF_API_FUNC_MACRO(enif_have_dirty_schedulers) -#endif - # define enif_is_map ERL_NIF_API_FUNC_MACRO(enif_is_map) # define enif_get_map_size ERL_NIF_API_FUNC_MACRO(enif_get_map_size) # define enif_make_new_map ERL_NIF_API_FUNC_MACRO(enif_make_new_map) @@ -311,11 +304,21 @@ ERL_NIF_API_FUNC_DECL(int, enif_map_iterator_get_pair, (ErlNifEnv *env, ErlNifMa # define enif_map_iterator_next ERL_NIF_API_FUNC_MACRO(enif_map_iterator_next) # define enif_map_iterator_prev ERL_NIF_API_FUNC_MACRO(enif_map_iterator_prev) # define enif_map_iterator_get_pair ERL_NIF_API_FUNC_MACRO(enif_map_iterator_get_pair) +# define enif_schedule_nif ERL_NIF_API_FUNC_MACRO(enif_schedule_nif) /* -** Add new entries here +** ADD NEW ENTRIES HERE (before this comment) */ + +/* + * Conditional EXPERIMENTAL stuff always last + * Must be moved up and made unconditional to support binary backward + * compatibility on Windows. + */ +#ifdef ERL_NIF_DIRTY_SCHEDULER_SUPPORT +# define enif_is_on_dirty_scheduler ERL_NIF_API_FUNC_MACRO(enif_is_on_dirty_scheduler) #endif +#endif /* ERL_NIF_API_FUNC_MACRO */ #if defined(__GNUC__) && !(defined(__WIN32__) || defined(_WIN32) || defined(_WIN32_)) diff --git a/erts/emulator/beam/erl_port_task.c b/erts/emulator/beam/erl_port_task.c index 2fc95ed5d8..2aa0a27197 100644 --- a/erts/emulator/beam/erl_port_task.c +++ b/erts/emulator/beam/erl_port_task.c @@ -32,6 +32,7 @@ #include "global.h" #include "erl_port_task.h" #include "dist.h" +#include "erl_check_io.h" #include "dtrace-wrapper.h" #include <stdarg.h> @@ -68,6 +69,13 @@ static void chk_task_queues(Port *pp, ErtsPortTask *execq, int processing_busy_q #define DTRACE_DRIVER(PROBE_NAME, PP) do {} while(0) #endif +#define ERTS_SMP_LC_VERIFY_RQ(RQ, PP) \ + do { \ + ERTS_SMP_LC_ASSERT(erts_smp_lc_runq_is_locked(runq)); \ + ERTS_SMP_LC_ASSERT((RQ) == ((ErtsRunQueue *) \ + erts_smp_atomic_read_nob(&(PP)->run_queue))); \ + } while (0) + erts_smp_atomic_t erts_port_task_outstanding_io_tasks; #define ERTS_PT_STATE_SCHEDULED 0 @@ -543,6 +551,16 @@ reset_handle(ErtsPortTask *ptp) } static ERTS_INLINE void +reset_executed_io_task_handle(ErtsPortTask *ptp) +{ + if (ptp->u.alive.handle) { + ASSERT(ptp == handle2task(ptp->u.alive.handle)); + erts_io_notify_port_task_executed(ptp->u.alive.handle); + reset_port_task_handle(ptp->u.alive.handle); + } +} + +static ERTS_INLINE void set_handle(ErtsPortTask *ptp, ErtsPortTaskHandle *pthp) { ptp->u.alive.handle = pthp; @@ -992,6 +1010,7 @@ static ERTS_INLINE int finalize_exec(Port *pp, ErtsPortTask **execq, int processing_busy_q) { erts_aint32_t act; + unsigned int prof_runnable_ports; if (!processing_busy_q) pp->sched.taskq.local.first = *execq; @@ -1008,6 +1027,10 @@ finalize_exec(Port *pp, ErtsPortTask **execq, int processing_busy_q) if (act & ERTS_PTS_FLG_CHK_UNSET_BUSY_PORT_Q) act = check_unset_busy_port_q(pp, act, pp->sched.taskq.bpq); + prof_runnable_ports = erts_system_profile_flags.runnable_ports; + if (prof_runnable_ports) + erts_port_task_sched_lock(&pp->sched); + while (1) { erts_aint32_t new, exp; @@ -1019,12 +1042,24 @@ finalize_exec(Port *pp, ErtsPortTask **execq, int processing_busy_q) act = erts_smp_atomic32_cmpxchg_relb(&pp->sched.flags, new, exp); - ASSERT(!(act & ERTS_PTS_FLG_IN_RUNQ)); + ERTS_LC_ASSERT(!(act & ERTS_PTS_FLG_IN_RUNQ)); + ERTS_LC_ASSERT(!(act & ERTS_PTS_FLG_EXEC_IMM)); if (exp == act) break; } + if (prof_runnable_ports | IS_TRACED_FL(pp, F_TRACE_SCHED_PORTS)) { + /* trace port scheduling, out */ + if (IS_TRACED_FL(pp, F_TRACE_SCHED_PORTS)) + trace_sched_ports(pp, am_out); + if (prof_runnable_ports) { + if (!(act & (ERTS_PTS_FLG_EXEC_IMM|ERTS_PTS_FLG_HAVE_TASKS))) + profile_runnable_port(pp, am_inactive); + erts_port_task_sched_unlock(&pp->sched); + } + } + return (act & ERTS_PTS_FLG_HAVE_TASKS) != 0; } @@ -1370,11 +1405,9 @@ erts_port_task_schedule(Eterm id, Port *pp; ErtsPortTask *ptp = NULL; erts_aint32_t act, add_flags; + unsigned int prof_runnable_ports; - if (pthp && erts_port_task_is_scheduled(pthp)) { - ASSERT(0); - erts_port_task_abort(pthp); - } + ERTS_LC_ASSERT(!pthp || !erts_port_task_is_scheduled(pthp)); ASSERT(is_internal_port(id)); @@ -1458,6 +1491,10 @@ erts_port_task_schedule(Eterm id, if (ns_pthlp) add_flags |= ERTS_PTS_FLG_HAVE_NS_TASKS; + prof_runnable_ports = erts_system_profile_flags.runnable_ports; + if (prof_runnable_ports) + erts_port_task_sched_lock(&pp->sched); + while (1) { erts_aint32_t new, exp; @@ -1482,6 +1519,13 @@ erts_port_task_schedule(Eterm id, goto done; /* Died after our task insert... */ } + if (prof_runnable_ports) { + if (!(act & ERTS_PTS_FLG_EXEC_IMM)) + profile_runnable_port(pp, am_active); + erts_port_task_sched_unlock(&pp->sched); + prof_runnable_ports = 0; + } + /* Enqueue port on run-queue */ runq = erts_port_runq(pp); @@ -1490,8 +1534,10 @@ erts_port_task_schedule(Eterm id, #ifdef ERTS_SMP xrunq = erts_check_emigration_need(runq, ERTS_PORT_PRIO_LEVEL); + ERTS_SMP_LC_ASSERT(runq != xrunq); + ERTS_SMP_LC_VERIFY_RQ(runq, pp); if (xrunq) { - /* Port emigrated ... */ + /* Emigrate port ... */ erts_smp_atomic_set_nob(&pp->run_queue, (erts_aint_t) xrunq); erts_smp_runq_unlock(runq); runq = erts_port_runq(pp); @@ -1501,10 +1547,6 @@ erts_port_task_schedule(Eterm id, #endif enqueue_port(runq, pp); - - if (erts_system_profile_flags.runnable_ports) { - profile_runnable_port(pp, am_active); - } erts_smp_runq_unlock(runq); @@ -1512,6 +1554,9 @@ erts_port_task_schedule(Eterm id, done: + if (prof_runnable_ports) + erts_port_task_sched_unlock(&pp->sched); + #ifdef ERTS_SMP if (dhndl != ERTS_THR_PRGR_DHANDLE_MANAGED) erts_port_dec_refc(pp); @@ -1610,6 +1655,8 @@ erts_port_task_execute(ErtsRunQueue *runq, Port **curr_port_pp) goto done; } + ERTS_SMP_LC_VERIFY_RQ(runq, pp); + erts_smp_runq_unlock(runq); *curr_port_pp = pp; @@ -1660,8 +1707,6 @@ erts_port_task_execute(ErtsRunQueue *runq, Port **curr_port_pp) goto aborted_port_task; } - reset_handle(ptp); - if (erts_system_monitor_long_schedule != 0) { start_time = erts_timestamp_millis(); } @@ -1672,6 +1717,7 @@ erts_port_task_execute(ErtsRunQueue *runq, Port **curr_port_pp) switch (ptp->type) { case ERTS_PORT_TASK_TIMEOUT: + reset_handle(ptp); reds = ERTS_PORT_REDS_TIMEOUT; if (!(state & ERTS_PORT_SFLGS_DEAD)) { DTRACE_DRIVER(driver_timeout, pp); @@ -1686,6 +1732,7 @@ erts_port_task_execute(ErtsRunQueue *runq, Port **curr_port_pp) for input and output */ (*pp->drv_ptr->ready_input)((ErlDrvData) pp->drv_data, ptp->u.alive.td.io.event); + reset_executed_io_task_handle(ptp); io_tasks_executed++; break; case ERTS_PORT_TASK_OUTPUT: @@ -1694,6 +1741,7 @@ erts_port_task_execute(ErtsRunQueue *runq, Port **curr_port_pp) DTRACE_DRIVER(driver_ready_output, pp); (*pp->drv_ptr->ready_output)((ErlDrvData) pp->drv_data, ptp->u.alive.td.io.event); + reset_executed_io_task_handle(ptp); io_tasks_executed++; break; case ERTS_PORT_TASK_EVENT: @@ -1703,10 +1751,12 @@ erts_port_task_execute(ErtsRunQueue *runq, Port **curr_port_pp) (*pp->drv_ptr->event)((ErlDrvData) pp->drv_data, ptp->u.alive.td.io.event, ptp->u.alive.td.io.event_data); + reset_executed_io_task_handle(ptp); io_tasks_executed++; break; case ERTS_PORT_TASK_PROC_SIG: { ErtsProc2PortSigData *sigdp = &ptp->u.alive.td.psig.data; + reset_handle(ptp); ASSERT((state & ERTS_PORT_SFLGS_DEAD) == 0); if (!pp->sched.taskq.bpq) reds = ptp->u.alive.td.psig.callback(pp, @@ -1724,6 +1774,7 @@ erts_port_task_execute(ErtsRunQueue *runq, Port **curr_port_pp) break; } case ERTS_PORT_TASK_DIST_CMD: + reset_handle(ptp); reds = erts_dist_command(pp, CONTEXT_REDS - pp->reds); break; default: @@ -1766,10 +1817,6 @@ erts_port_task_execute(ErtsRunQueue *runq, Port **curr_port_pp) erts_unblock_fpe(fpe_was_unmasked); - /* trace port scheduling, out */ - if (IS_TRACED_FL(pp, F_TRACE_SCHED_PORTS)) { - trace_sched_ports(pp, am_out); - } if (io_tasks_executed) { ASSERT(erts_smp_atomic_read_nob(&erts_port_task_outstanding_io_tasks) @@ -1792,11 +1839,7 @@ erts_port_task_execute(ErtsRunQueue *runq, Port **curr_port_pp) erts_smp_runq_lock(runq); - if (!active) { - if (erts_system_profile_flags.runnable_ports) - profile_runnable_port(pp, am_inactive); - } - else { + if (active) { #ifdef ERTS_SMP ErtsRunQueue *xrunq; #endif @@ -1805,6 +1848,8 @@ erts_port_task_execute(ErtsRunQueue *runq, Port **curr_port_pp) #ifdef ERTS_SMP xrunq = erts_check_emigration_need(runq, ERTS_PORT_PRIO_LEVEL); + ERTS_SMP_LC_ASSERT(runq != xrunq); + ERTS_SMP_LC_VERIFY_RQ(runq, pp); if (!xrunq) { #endif enqueue_port(runq, pp); @@ -1812,7 +1857,7 @@ erts_port_task_execute(ErtsRunQueue *runq, Port **curr_port_pp) #ifdef ERTS_SMP } else { - /* Port emigrated ... */ + /* Emigrate port... */ erts_smp_atomic_set_nob(&pp->run_queue, (erts_aint_t) xrunq); erts_smp_runq_unlock(runq); diff --git a/erts/emulator/beam/erl_port_task.h b/erts/emulator/beam/erl_port_task.h index 1d30465ec9..406cd3c492 100644 --- a/erts/emulator/beam/erl_port_task.h +++ b/erts/emulator/beam/erl_port_task.h @@ -78,6 +78,7 @@ extern erts_smp_atomic_t erts_port_task_outstanding_io_tasks; #define ERTS_PTS_FLG_PARALLELISM (((erts_aint32_t) 1) << 9) #define ERTS_PTS_FLG_FORCE_SCHED (((erts_aint32_t) 1) << 10) #define ERTS_PTS_FLG_EXITING (((erts_aint32_t) 1) << 11) +#define ERTS_PTS_FLG_EXEC_IMM (((erts_aint32_t) 1) << 12) #define ERTS_PTS_FLGS_BUSY \ (ERTS_PTS_FLG_BUSY_PORT | ERTS_PTS_FLG_BUSY_PORT_Q) @@ -87,6 +88,7 @@ extern erts_smp_atomic_t erts_port_task_outstanding_io_tasks; | ERTS_PTS_FLG_HAVE_BUSY_TASKS \ | ERTS_PTS_FLG_HAVE_TASKS \ | ERTS_PTS_FLG_EXEC \ + | ERTS_PTS_FLG_EXEC_IMM \ | ERTS_PTS_FLG_FORCE_SCHED \ | ERTS_PTS_FLG_EXITING) @@ -154,7 +156,7 @@ erts_port_task_handle_init(ErtsPortTaskHandle *pthp) ERTS_GLB_INLINE int erts_port_task_is_scheduled(ErtsPortTaskHandle *pthp) { - return ((void *) erts_smp_atomic_read_nob(pthp)) != NULL; + return ((void *) erts_smp_atomic_read_acqb(pthp)) != NULL; } ERTS_GLB_INLINE void erts_port_task_pre_init_sched(ErtsPortTaskSched *ptsp, diff --git a/erts/emulator/beam/erl_process.c b/erts/emulator/beam/erl_process.c index 1606ad119d..ea63d20dfa 100644 --- a/erts/emulator/beam/erl_process.c +++ b/erts/emulator/beam/erl_process.c @@ -148,6 +148,12 @@ extern BeamInstr beam_apply[]; extern BeamInstr beam_exit[]; extern BeamInstr beam_continue_exit[]; +#ifdef __OSE__ +/* Eager check I/O not supported on OSE yet. */ +int erts_eager_check_io = 0; +#else +int erts_eager_check_io = 0; +#endif int erts_sched_compact_load; int erts_sched_balance_util = 0; Uint erts_no_schedulers; @@ -590,12 +596,10 @@ erts_pre_init_process(void) erts_psd_required_locks[ERTS_PSD_DELAYED_GC_TASK_QS].set_locks = ERTS_PSD_DELAYED_GC_TASK_QS_SET_LOCKS; -#ifdef ERTS_DIRTY_SCHEDULERS - erts_psd_required_locks[ERTS_PSD_DIRTY_SCHED_TRAP_EXPORT].get_locks - = ERTS_PSD_DIRTY_SCHED_TRAP_EXPORT_GET_LOCKS; - erts_psd_required_locks[ERTS_PSD_DIRTY_SCHED_TRAP_EXPORT].set_locks - = ERTS_PSD_DIRTY_SCHED_TRAP_EXPORT_SET_LOCKS; -#endif + erts_psd_required_locks[ERTS_PSD_NIF_TRAP_EXPORT].get_locks + = ERTS_PSD_NIF_TRAP_EXPORT_GET_LOCKS; + erts_psd_required_locks[ERTS_PSD_NIF_TRAP_EXPORT].set_locks + = ERTS_PSD_NIF_TRAP_EXPORT_SET_LOCKS; /* Check that we have locks for all entries */ for (ix = 0; ix < ERTS_PSD_SIZE; ix++) { @@ -712,72 +716,24 @@ sched_wall_time_ts(void) #if ERTS_HAVE_SCHED_UTIL_BALANCING_SUPPORT -#ifdef ARCH_64 - -static ERTS_INLINE Uint64 -aschedtime_read(ErtsAtomicSchedTime *var) -{ - return (Uint64) erts_atomic_read_nob((erts_atomic_t *) var); -} - -static ERTS_INLINE void -aschedtime_set(ErtsAtomicSchedTime *var, Uint64 val) -{ - erts_atomic_set_nob((erts_atomic_t *) var, (erts_aint_t) val); -} - -static ERTS_INLINE void -aschedtime_init(ErtsAtomicSchedTime *var) -{ - erts_atomic_init_nob((erts_atomic_t *) var, (erts_aint_t) 0); -} - -#elif defined(ARCH_32) - static ERTS_INLINE Uint64 aschedtime_read(ErtsAtomicSchedTime *var) { - erts_dw_aint_t dw; - erts_dw_atomic_read_nob((erts_dw_atomic_t *) var, &dw); -#ifdef ETHR_SU_DW_NAINT_T__ - return (Uint64) dw.dw_sint; -#else - { - Uint64 res; - res = (Uint64) ((Uint32) dw.sint[ERTS_DW_AINT_HIGH_WORD]); - res <<= 32; - res |= (Uint64) ((Uint32) dw.sint[ERTS_DW_AINT_LOW_WORD]); - return res; - } -#endif + return (Uint64) erts_atomic64_read_nob((erts_atomic64_t *) var); } static ERTS_INLINE void aschedtime_set(ErtsAtomicSchedTime *var, Uint64 val) { - erts_dw_aint_t dw; -#ifdef ETHR_SU_DW_NAINT_T__ - dw.dw_sint = (ETHR_SU_DW_NAINT_T__) val; -#else - dw.sint[ERTS_DW_AINT_LOW_WORD] = (erts_aint_t) (val & 0xffffffff); - dw.sint[ERTS_DW_AINT_HIGH_WORD] = (erts_aint_t) ((val >> 32) & 0xffffffff); -#endif - erts_dw_atomic_set_nob((erts_dw_atomic_t *) var, &dw); + erts_atomic64_set_nob((erts_atomic64_t *) var, (erts_aint64_t) val); } static ERTS_INLINE void aschedtime_init(ErtsAtomicSchedTime *var) { - erts_dw_aint_t dw; - dw.sint[ERTS_DW_AINT_LOW_WORD] = (erts_aint_t) 0; - dw.sint[ERTS_DW_AINT_HIGH_WORD] = (erts_aint_t) 0; - erts_dw_atomic_init_nob((erts_dw_atomic_t *) var, &dw); + erts_atomic64_init_nob((erts_atomic64_t *) var, (erts_aint64_t) 0); } -#else -# error :-/ -#endif - #define ERTS_GET_AVG_MAX_UNLOCKED_TRY 50 #define ERTS_SCHED_AVG_UTIL_WRITE_MARKER (~((Uint64) 0)) @@ -2383,29 +2339,47 @@ try_set_sys_scheduling(void) #endif static ERTS_INLINE int -prepare_for_sys_schedule(ErtsSchedulerData *esdp) +prepare_for_sys_schedule(ErtsSchedulerData *esdp, int non_blocking) { + if (non_blocking && erts_eager_check_io) { #ifdef ERTS_SMP - while (!erts_port_task_have_outstanding_io_tasks() - && try_set_sys_scheduling()) { #ifdef ERTS_SCHED_ONLY_POLL_SCHED_1 - if (esdp->no != 1) { - /* If we are not scheduler 1 and ERTS_SCHED_ONLY_POLL_SCHED_1 is used - then we make sure to wake scheduler 1 */ - ErtsRunQueue *rq = ERTS_RUNQ_IX(0); - clear_sys_scheduling(); - wake_scheduler(rq); - return 0; - } + if (esdp->no != 1) { + /* If we are not scheduler 1 and ERTS_SCHED_ONLY_POLL_SCHED_1 is used + then we make sure to wake scheduler 1 */ + ErtsRunQueue *rq = ERTS_RUNQ_IX(0); + wake_scheduler(rq); + return 0; + } #endif - if (!erts_port_task_have_outstanding_io_tasks()) + return try_set_sys_scheduling(); +#else return 1; - clear_sys_scheduling(); +#endif } - return 0; + else { +#ifdef ERTS_SMP + while (!erts_port_task_have_outstanding_io_tasks() + && try_set_sys_scheduling()) { +#ifdef ERTS_SCHED_ONLY_POLL_SCHED_1 + if (esdp->no != 1) { + /* If we are not scheduler 1 and ERTS_SCHED_ONLY_POLL_SCHED_1 is used + then we make sure to wake scheduler 1 */ + ErtsRunQueue *rq = ERTS_RUNQ_IX(0); + clear_sys_scheduling(); + wake_scheduler(rq); + return 0; + } +#endif + if (!erts_port_task_have_outstanding_io_tasks()) + return 1; + clear_sys_scheduling(); + } + return 0; #else - return !erts_port_task_have_outstanding_io_tasks(); + return !erts_port_task_have_outstanding_io_tasks(); #endif + } } #ifdef ERTS_SMP @@ -2782,7 +2756,7 @@ scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq) * be waiting in erl_sys_schedule() */ - if (ERTS_SCHEDULER_IS_DIRTY(esdp) || !prepare_for_sys_schedule(esdp)) { + if (ERTS_SCHEDULER_IS_DIRTY(esdp) || !prepare_for_sys_schedule(esdp, 0)) { sched_waiting(esdp->no, rq); @@ -2946,7 +2920,7 @@ scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq) * Got to check that we still got I/O tasks; otherwise * we have to continue checking for I/O... */ - if (!prepare_for_sys_schedule(esdp)) { + if (!prepare_for_sys_schedule(esdp, 0)) { spincount *= ERTS_SCHED_TSE_SLEEP_SPINCOUNT_FACT; goto tse_wait; } @@ -2968,7 +2942,7 @@ scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq) * Got to check that we still got I/O tasks; otherwise * we have to wait in erl_sys_schedule() after all... */ - if (!prepare_for_sys_schedule(esdp)) { + if (!prepare_for_sys_schedule(esdp, 0)) { /* * Not allowed to wait in erl_sys_schedule; * do tse wait instead... @@ -3222,11 +3196,11 @@ chk_wake_sched(ErtsRunQueue *crq, int ix, int activate) return 0; wrq = ERTS_RUNQ_IX(ix); flags = ERTS_RUNQ_FLGS_GET(wrq); + if (activate && !(flags & ERTS_RUNQ_FLG_SUSPENDED)) { + if (try_inc_no_active_runqs(ix+1)) + (void) ERTS_RUNQ_FLGS_UNSET(wrq, ERTS_RUNQ_FLG_INACTIVE); + } if (!(flags & (ERTS_RUNQ_FLG_SUSPENDED|ERTS_RUNQ_FLG_NONEMPTY))) { - if (activate) { - if (try_inc_no_active_runqs(ix+1)) - (void) ERTS_RUNQ_FLGS_UNSET(wrq, ERTS_RUNQ_FLG_INACTIVE); - } wake_scheduler(wrq); return 1; } @@ -3758,17 +3732,25 @@ evacuate_run_queue(ErtsRunQueue *rq, } #ifdef ERTS_DIRTY_SCHEDULERS else if (state & ERTS_PSFLG_DIRTY_CPU_PROC_IN_Q) { - erts_aint32_t old; - old = erts_smp_atomic32_read_band_nob(&proc->state, - ~(ERTS_PSFLG_DIRTY_CPU_PROC - | ERTS_PSFLG_DIRTY_CPU_PROC_IN_Q)); +#ifdef DEBUG + erts_aint32_t old = +#else + (void) +#endif + erts_smp_atomic32_read_band_nob(&proc->state, + ~(ERTS_PSFLG_DIRTY_CPU_PROC + | ERTS_PSFLG_DIRTY_CPU_PROC_IN_Q)); /* assert that no other dirty flags are set */ ASSERT(!(old & (ERTS_PSFLG_DIRTY_IO_PROC|ERTS_PSFLG_DIRTY_IO_PROC_IN_Q))); } else if (state & ERTS_PSFLG_DIRTY_IO_PROC_IN_Q) { - erts_aint32_t old; - old = erts_smp_atomic32_read_band_nob(&proc->state, - ~(ERTS_PSFLG_DIRTY_IO_PROC - | ERTS_PSFLG_DIRTY_IO_PROC_IN_Q)); +#ifdef DEBUG + erts_aint32_t old = +#else + (void) +#endif + erts_smp_atomic32_read_band_nob(&proc->state, + ~(ERTS_PSFLG_DIRTY_IO_PROC + | ERTS_PSFLG_DIRTY_IO_PROC_IN_Q)); /* assert that no other dirty flags are set */ ASSERT(!(old & (ERTS_PSFLG_DIRTY_CPU_PROC|ERTS_PSFLG_DIRTY_CPU_PROC_IN_Q))); } @@ -5877,6 +5859,9 @@ schedule_out_process(ErtsRunQueue *c_rq, erts_aint32_t state, Process *p, Proces case ERTS_ENQUEUE_NOT: if (erts_system_profile_flags.runnable_procs) { + /* Status lock prevents out of order "runnable proc" trace msgs */ + ERTS_SMP_LC_ASSERT(ERTS_PROC_LOCK_STATUS & erts_proc_lc_my_proc_locks(p)); + if (!(a & ERTS_PSFLG_ACTIVE_SYS) && (!(a & ERTS_PSFLG_ACTIVE) || (a & ERTS_PSFLG_SUSPENDED))) { @@ -5990,7 +5975,8 @@ change_proc_schedule_state(Process *p, erts_aint32_t clear_state_flags, erts_aint32_t set_state_flags, erts_aint32_t *statep, - erts_aint32_t *enq_prio_p) + erts_aint32_t *enq_prio_p, + ErtsProcLocks locks) { /* * NOTE: ERTS_PSFLG_RUNNING, ERTS_PSFLG_RUNNING_SYS and @@ -5999,6 +5985,11 @@ change_proc_schedule_state(Process *p, */ erts_aint32_t a = *statep, n; int enqueue; /* < 0 -> use proxy */ + unsigned int prof_runnable_procs = erts_system_profile_flags.runnable_procs; + unsigned int lock_status = (prof_runnable_procs + && !(locks & ERTS_PROC_LOCK_STATUS)); + + ERTS_SMP_LC_ASSERT(locks == erts_proc_lc_my_proc_locks(p)); ASSERT(!(a & ERTS_PSFLG_PROXY)); ASSERT((clear_state_flags & (ERTS_PSFLG_RUNNING @@ -6008,6 +5999,9 @@ change_proc_schedule_state(Process *p, | ERTS_PSFLG_RUNNING_SYS | ERTS_PSFLG_ACTIVE_SYS)) == 0); + if (lock_status) + erts_smp_proc_lock(p, ERTS_PROC_LOCK_STATUS); + while (1) { erts_aint32_t e; n = e = a; @@ -6043,7 +6037,9 @@ change_proc_schedule_state(Process *p, break; } - if (erts_system_profile_flags.runnable_procs) { + if (prof_runnable_procs) { + + /* Status lock prevents out of order "runnable proc" trace msgs */ if (((n & (ERTS_PSFLG_SUSPENDED | ERTS_PSFLG_ACTIVE)) == ERTS_PSFLG_ACTIVE) @@ -6056,15 +6052,18 @@ change_proc_schedule_state(Process *p, profile_runnable_proc(p, am_active); } + if (lock_status) + erts_smp_proc_unlock(p, ERTS_PROC_LOCK_STATUS); } + *statep = a; return enqueue; } static ERTS_INLINE void -schedule_process(Process *p, erts_aint32_t in_state) +schedule_process(Process *p, erts_aint32_t in_state, ErtsProcLocks locks) { erts_aint32_t enq_prio = -1; erts_aint32_t state = in_state; @@ -6072,7 +6071,8 @@ schedule_process(Process *p, erts_aint32_t in_state) 0, ERTS_PSFLG_ACTIVE, &state, - &enq_prio); + &enq_prio, + locks); if (enqueue != ERTS_ENQUEUE_NOT) add2runq(enqueue > 0 ? p : make_proxy_proc(NULL, p, enq_prio), state, @@ -6080,16 +6080,27 @@ schedule_process(Process *p, erts_aint32_t in_state) } void -erts_schedule_process(Process *p, erts_aint32_t state) +erts_schedule_process(Process *p, erts_aint32_t state, ErtsProcLocks locks) { - schedule_process(p, state); + schedule_process(p, state, locks); } static void schedule_process_sys_task(Process *p, erts_aint32_t state, Process *proxy) { + /* + * Expects status lock to be locked when called, and + * returns with status lock unlocked... + */ erts_aint32_t a = state, n, enq_prio = -1; int enqueue; /* < 0 -> use proxy */ + unsigned int prof_runnable_procs = erts_system_profile_flags.runnable_procs; + + /* Status lock prevents out of order "runnable proc" trace msgs */ + ERTS_SMP_LC_ASSERT(ERTS_PROC_LOCK_STATUS & erts_proc_lc_my_proc_locks(p)); + + if (!prof_runnable_procs) + erts_smp_proc_unlock(p, ERTS_PROC_LOCK_STATUS); ASSERT(!(state & ERTS_PSFLG_PROXY)); @@ -6098,7 +6109,7 @@ schedule_process_sys_task(Process *p, erts_aint32_t state, Process *proxy) n = e = a; if (a & ERTS_PSFLG_FREE) - return; /* We don't want to schedule free processes... */ + goto cleanup; /* We don't want to schedule free processes... */ enqueue = ERTS_ENQUEUE_NOT; n |= ERTS_PSFLG_ACTIVE_SYS; @@ -6111,7 +6122,7 @@ schedule_process_sys_task(Process *p, erts_aint32_t state, Process *proxy) goto cleanup; } - if (erts_system_profile_flags.runnable_procs) { + if (prof_runnable_procs) { if (!(a & (ERTS_PSFLG_ACTIVE_SYS | ERTS_PSFLG_RUNNING @@ -6121,6 +6132,8 @@ schedule_process_sys_task(Process *p, erts_aint32_t state, Process *proxy) profile_runnable_proc(p, am_active); } + erts_smp_proc_unlock(p, ERTS_PROC_LOCK_STATUS); + prof_runnable_procs = 0; } if (enqueue != ERTS_ENQUEUE_NOT) { @@ -6135,8 +6148,14 @@ schedule_process_sys_task(Process *p, erts_aint32_t state, Process *proxy) } cleanup: + + if (prof_runnable_procs) + erts_smp_proc_unlock(p, ERTS_PROC_LOCK_STATUS); + if (proxy) free_proxy_proc(proxy); + + ERTS_SMP_LC_ASSERT(!(ERTS_PROC_LOCK_STATUS & erts_proc_lc_my_proc_locks(p))); } static ERTS_INLINE int @@ -6203,7 +6222,7 @@ suspend_process(Process *c_p, Process *p) } static ERTS_INLINE void -resume_process(Process *p) +resume_process(Process *p, ErtsProcLocks locks) { erts_aint32_t state, enq_prio = -1; int enqueue; @@ -6220,7 +6239,8 @@ resume_process(Process *p) ERTS_PSFLG_SUSPENDED, 0, &state, - &enq_prio); + &enq_prio, + locks); if (enqueue) add2runq(enqueue > 0 ? p : make_proxy_proc(NULL, p, enq_prio), state, @@ -8036,7 +8056,8 @@ handle_pend_sync_suspend(Process *suspendee, } /* suspender is suspended waiting for suspendee to suspend; resume suspender */ - resume_process(suspender); + ASSERT(suspendee != suspender); + resume_process(suspender, ERTS_PROC_LOCK_STATUS); erts_smp_proc_unlock(suspender, ERTS_PROC_LOCK_STATUS); } } @@ -8071,7 +8092,7 @@ pid2proc_not_running(Process *c_p, ErtsProcLocks c_p_locks, ASSERT(c_p->flags & F_P2PNR_RESCHED); c_p->flags &= ~F_P2PNR_RESCHED; if (!suspend && rp) - resume_process(rp); + resume_process(rp, rp_locks); } else { @@ -8229,7 +8250,8 @@ handle_pend_bif_sync_suspend(Process *suspendee, } /* suspender is suspended waiting for suspendee to suspend; resume suspender */ - resume_process(suspender); + ASSERT(suspender != suspendee); + resume_process(suspender, ERTS_PROC_LOCK_LINK|ERTS_PROC_LOCK_STATUS); erts_smp_proc_unlock(suspender, ERTS_PROC_LOCK_LINK|ERTS_PROC_LOCK_STATUS); } @@ -8589,7 +8611,8 @@ resume_process_1(BIF_ALIST_1) ASSERT(ERTS_PSFLG_SUSPENDED & erts_smp_atomic32_read_nob(&suspendee->state)); - resume_process(suspendee); + ASSERT(BIF_P != suspendee); + resume_process(suspendee, ERTS_PROC_LOCK_STATUS); erts_smp_proc_unlock(suspendee, ERTS_PROC_LOCK_STATUS); } @@ -8719,7 +8742,7 @@ erts_resume(Process* process, ErtsProcLocks process_locks) ERTS_SMP_LC_ASSERT(process_locks == erts_proc_lc_my_proc_locks(process)); if (!(process_locks & ERTS_PROC_LOCK_STATUS)) erts_smp_proc_lock(process, ERTS_PROC_LOCK_STATUS); - resume_process(process); + resume_process(process, process_locks|ERTS_PROC_LOCK_STATUS); if (!(process_locks & ERTS_PROC_LOCK_STATUS)) erts_smp_proc_unlock(process, ERTS_PROC_LOCK_STATUS); } @@ -8738,7 +8761,7 @@ erts_resume_processes(ErtsProcList *list) proc = erts_pid2proc(NULL, 0, plp->pid, ERTS_PROC_LOCK_STATUS); if (proc) { if (erts_proclist_same(plp, proc)) { - resume_process(proc); + resume_process(proc, ERTS_PROC_LOCK_STATUS); nresumed++; } erts_smp_proc_unlock(proc, ERTS_PROC_LOCK_STATUS); @@ -9153,7 +9176,7 @@ Process *schedule(Process *p, int calls) } else if (!ERTS_SCHEDULER_IS_DIRTY(esdp) && (fcalls > input_reductions && - prepare_for_sys_schedule(esdp))) { + prepare_for_sys_schedule(esdp, !0))) { /* * Schedule system-level activities. */ @@ -9161,8 +9184,6 @@ Process *schedule(Process *p, int calls) erts_smp_atomic32_set_relb(&function_calls, 0); fcalls = 0; - ASSERT(!erts_port_task_have_outstanding_io_tasks()); - #if 0 /* Not needed since we wont wait in sys schedule */ erts_sys_schedule_interrupt(0); #endif @@ -9194,7 +9215,9 @@ Process *schedule(Process *p, int calls) if (RUNQ_READ_LEN(&rq->ports.info.len)) { int have_outstanding_io; have_outstanding_io = erts_port_task_execute(rq, &esdp->current_port); - if ((have_outstanding_io && fcalls > 2*input_reductions) + if ((!erts_eager_check_io + && have_outstanding_io + && fcalls > 2*input_reductions) || rq->halt_in_progress) { /* * If we have performed more than 2*INPUT_REDUCTIONS since @@ -9974,8 +9997,10 @@ erts_internal_request_system_task_3(BIF_ALIST_3) rp_state = n; } - erts_smp_proc_unlock(rp, ERTS_PROC_LOCK_STATUS); - + /* + * schedule_process_sys_task() unlocks status + * lock on process. + */ schedule_process_sys_task(rp, rp_state, NULL); if (free_stqs) @@ -10720,7 +10745,7 @@ erl_create_process(Process* parent, /* Parent of process (default group leader). * Schedule process for execution. */ - schedule_process(p, state); + schedule_process(p, state, 0); VERBOSE(DEBUG_PROCESSES, ("Created a new process: %T\n",p->common.id)); @@ -11041,7 +11066,8 @@ set_proc_exiting(Process *p, ERTS_PSFLG_SUSPENDED|ERTS_PSFLG_PENDING_EXIT, ERTS_PSFLG_EXITING|ERTS_PSFLG_ACTIVE, &state, - &enq_prio); + &enq_prio, + ERTS_PROC_LOCKS_ALL); p->fvalue = reason; if (bp) @@ -11082,7 +11108,8 @@ set_proc_self_exiting(Process *c_p) ERTS_PSFLG_SUSPENDED|ERTS_PSFLG_PENDING_EXIT, ERTS_PSFLG_EXITING|ERTS_PSFLG_ACTIVE, &state, - &enq_prio); + &enq_prio, + ERTS_PROC_LOCKS_ALL); ASSERT(!enqueue); return state; @@ -11727,8 +11754,9 @@ resume_suspend_monitor(ErtsSuspendMonitor *smon, void *vc_p) Process *suspendee = erts_pid2proc((Process *) vc_p, ERTS_PROC_LOCK_MAIN, smon->pid, ERTS_PROC_LOCK_STATUS); if (suspendee) { + ASSERT(suspendee != vc_p); if (smon->active) - resume_process(suspendee); + resume_process(suspendee, ERTS_PROC_LOCK_STATUS); erts_smp_proc_unlock(suspendee, ERTS_PROC_LOCK_STATUS); } erts_destroy_suspend_monitor(smon); @@ -11820,6 +11848,7 @@ erts_continue_exit_process(Process *p) struct saved_calls *scb; process_breakpoint_time_t *pbt; erts_aint32_t state; + void *nif_export; #ifdef DEBUG int yield_allowed = 1; @@ -11970,6 +11999,7 @@ erts_continue_exit_process(Process *p) : NULL); scb = ERTS_PROC_SET_SAVED_CALLS_BUF(p, ERTS_PROC_LOCKS_ALL, NULL); pbt = ERTS_PROC_SET_CALL_TIME(p, ERTS_PROC_LOCKS_ALL, NULL); + nif_export = ERTS_PROC_SET_NIF_TRAP_EXPORT(p, ERTS_PROC_LOCKS_ALL, NULL); erts_smp_proc_unlock(p, ERTS_PROC_LOCKS_ALL); #ifdef BM_COUNTERS @@ -12017,6 +12047,9 @@ erts_continue_exit_process(Process *p) if (pbt) erts_free(ERTS_ALC_T_BPD, (void *) pbt); + if (nif_export) + erts_destroy_nif_export(nif_export); + delete_process(p); #ifdef ERTS_SMP @@ -12061,7 +12094,7 @@ timeout_proc(Process* p) state = erts_smp_atomic32_read_acqb(&p->state); if (!(state & ERTS_PSFLG_ACTIVE)) - schedule_process(p, state); + schedule_process(p, state, ERTS_PROC_LOCK_MAIN|ERTS_PROC_LOCK_STATUS); } diff --git a/erts/emulator/beam/erl_process.h b/erts/emulator/beam/erl_process.h index ed6dadbffa..f50b217d4a 100644 --- a/erts/emulator/beam/erl_process.h +++ b/erts/emulator/beam/erl_process.h @@ -104,6 +104,7 @@ struct saved_calls { }; extern Export exp_send, exp_receive, exp_timeout; +extern int erts_eager_check_io; extern int erts_sched_compact_load; extern int erts_sched_balance_util; extern Uint erts_no_schedulers; @@ -350,13 +351,7 @@ typedef struct { #undef ERTS_HAVE_SCHED_UTIL_BALANCING_SUPPORT #define ERTS_HAVE_SCHED_UTIL_BALANCING_SUPPORT ERTS_HAVE_SCHED_UTIL_BALANCING_SUPPORT_OPT -#ifdef ARCH_64 -typedef erts_atomic_t ErtsAtomicSchedTime; -#elif defined(ARCH_32) -typedef erts_dw_atomic_t ErtsAtomicSchedTime; -#else -# error :-/ -#endif +typedef erts_atomic64_t ErtsAtomicSchedTime; #if ERTS_HAVE_SCHED_UTIL_BALANCING_SUPPORT typedef struct { @@ -734,13 +729,9 @@ erts_smp_reset_max_len(ErtsRunQueue *rq, ErtsRunQueueInfo *rqi) #define ERTS_PSD_DIST_ENTRY 3 #define ERTS_PSD_CALL_TIME_BP 4 #define ERTS_PSD_DELAYED_GC_TASK_QS 5 -#ifdef ERTS_DIRTY_SCHEDULERS -#define ERTS_PSD_DIRTY_SCHED_TRAP_EXPORT 6 +#define ERTS_PSD_NIF_TRAP_EXPORT 6 #define ERTS_PSD_SIZE 7 -#else -#define ERTS_PSD_SIZE 6 -#endif typedef struct { void *data[ERTS_PSD_SIZE]; @@ -767,10 +758,8 @@ typedef struct { #define ERTS_PSD_DELAYED_GC_TASK_QS_GET_LOCKS ERTS_PROC_LOCK_MAIN #define ERTS_PSD_DELAYED_GC_TASK_QS_SET_LOCKS ERTS_PROC_LOCK_MAIN -#ifdef ERTS_DIRTY_SCHEDULERS -#define ERTS_PSD_DIRTY_SCHED_TRAP_EXPORT_GET_LOCKS ERTS_PROC_LOCK_MAIN -#define ERTS_PSD_DIRTY_SCHED_TRAP_EXPORT_SET_LOCKS ERTS_PROC_LOCK_MAIN -#endif +#define ERTS_PSD_NIF_TRAP_EXPORT_GET_LOCKS ERTS_PROC_LOCK_MAIN +#define ERTS_PSD_NIF_TRAP_EXPORT_SET_LOCKS ERTS_PROC_LOCK_MAIN typedef struct { ErtsProcLocks get_locks; @@ -1367,6 +1356,9 @@ Uint64 erts_get_proc_interval(void); Uint64 erts_ensure_later_proc_interval(Uint64); Uint64 erts_step_proc_interval(void); +int erts_setup_nif_gc(Process* proc, Eterm** objv, int* nobj); /* see erl_nif.c */ +void erts_destroy_nif_export(void *); /* see erl_nif.c */ + ErtsProcList *erts_proclist_create(Process *); void erts_proclist_destroy(ErtsProcList *); @@ -1704,17 +1696,17 @@ ErtsSchedulerData *erts_get_scheduler_data(void) #endif #endif -void erts_schedule_process(Process *, erts_aint32_t); +void erts_schedule_process(Process *, erts_aint32_t, ErtsProcLocks); -ERTS_GLB_INLINE void erts_proc_notify_new_message(Process *p); +ERTS_GLB_INLINE void erts_proc_notify_new_message(Process *p, ErtsProcLocks locks); #if ERTS_GLB_INLINE_INCL_FUNC_DEF ERTS_GLB_INLINE void -erts_proc_notify_new_message(Process *p) +erts_proc_notify_new_message(Process *p, ErtsProcLocks locks) { /* No barrier needed, due to msg lock */ erts_aint32_t state = erts_smp_atomic32_read_nob(&p->state); if (!(state & ERTS_PSFLG_ACTIVE)) - erts_schedule_process(p, state); + erts_schedule_process(p, state, locks); } #endif @@ -1817,12 +1809,10 @@ erts_psd_set(Process *p, ErtsProcLocks plocks, int ix, void *data) #define ERTS_PROC_SET_DELAYED_GC_TASK_QS(P, L, PBT) \ ((ErtsProcSysTaskQs *) erts_psd_set((P), (L), ERTS_PSD_DELAYED_GC_TASK_QS, (void *) (PBT))) -#ifdef ERTS_DIRTY_SCHEDULERS -#define ERTS_PROC_GET_DIRTY_SCHED_TRAP_EXPORT(P) \ - ((Export *) erts_psd_get((P), ERTS_PSD_DIRTY_SCHED_TRAP_EXPORT)) -#define ERTS_PROC_SET_DIRTY_SCHED_TRAP_EXPORT(P, L, DSTE) \ - ((Export *) erts_psd_set((P), (L), ERTS_PSD_DIRTY_SCHED_TRAP_EXPORT, (void *) (DSTE))) -#endif +#define ERTS_PROC_GET_NIF_TRAP_EXPORT(P) \ + erts_psd_get((P), ERTS_PSD_NIF_TRAP_EXPORT) +#define ERTS_PROC_SET_NIF_TRAP_EXPORT(P, L, NTE) \ + erts_psd_set((P), (L), ERTS_PSD_NIF_TRAP_EXPORT, (void *) (NTE)) ERTS_GLB_INLINE Eterm erts_proc_get_error_handler(Process *p); diff --git a/erts/emulator/beam/erl_ptab.c b/erts/emulator/beam/erl_ptab.c index eabf016081..02943ee683 100644 --- a/erts/emulator/beam/erl_ptab.c +++ b/erts/emulator/beam/erl_ptab.c @@ -280,124 +280,38 @@ struct ErtsPTabListBifData_ { }; -#ifdef ARCH_32 - -static ERTS_INLINE Uint64 -dw_aint_to_uint64(erts_dw_aint_t *dw) -{ -#ifdef ETHR_SU_DW_NAINT_T__ - return (Uint64) dw->dw_sint; -#else - Uint64 res; - res = (Uint64) ((Uint32) dw->sint[ERTS_DW_AINT_HIGH_WORD]); - res <<= 32; - res |= (Uint64) ((Uint32) dw->sint[ERTS_DW_AINT_LOW_WORD]); - return res; -#endif -} - -static void -unint64_to_dw_aint(erts_dw_aint_t *dw, Uint64 val) -{ -#ifdef ETHR_SU_DW_NAINT_T__ - dw->dw_sint = (ETHR_SU_DW_NAINT_T__) val; -#else - dw->sint[ERTS_DW_AINT_LOW_WORD] = (erts_aint_t) (val & 0xffffffff); - dw->sint[ERTS_DW_AINT_HIGH_WORD] = (erts_aint_t) ((val >> 32) & 0xffffffff); -#endif -} - static ERTS_INLINE void last_data_init_nob(ErtsPTab *ptab, Uint64 val) { - erts_dw_aint_t dw; - unint64_to_dw_aint(&dw, val); - erts_smp_dw_atomic_init_nob(&ptab->vola.tile.last_data, &dw); + erts_smp_atomic64_init_nob(&ptab->vola.tile.last_data, (erts_aint64_t) val); } static ERTS_INLINE void last_data_set_relb(ErtsPTab *ptab, Uint64 val) { - erts_dw_aint_t dw; - unint64_to_dw_aint(&dw, val); - erts_smp_dw_atomic_set_relb(&ptab->vola.tile.last_data, &dw); + erts_smp_atomic64_set_relb(&ptab->vola.tile.last_data, (erts_aint64_t) val); } static ERTS_INLINE Uint64 last_data_read_nob(ErtsPTab *ptab) { - erts_dw_aint_t dw; - erts_smp_dw_atomic_read_nob(&ptab->vola.tile.last_data, &dw); - return dw_aint_to_uint64(&dw); + return (Uint64) erts_smp_atomic64_read_nob(&ptab->vola.tile.last_data); } static ERTS_INLINE Uint64 last_data_read_acqb(ErtsPTab *ptab) { - erts_dw_aint_t dw; - erts_smp_dw_atomic_read_acqb(&ptab->vola.tile.last_data, &dw); - return dw_aint_to_uint64(&dw); + return (Uint64) erts_smp_atomic64_read_acqb(&ptab->vola.tile.last_data); } static ERTS_INLINE Uint64 last_data_cmpxchg_relb(ErtsPTab *ptab, Uint64 new, Uint64 exp) { - erts_dw_aint_t dw_new, dw_xchg; - - unint64_to_dw_aint(&dw_new, new); - unint64_to_dw_aint(&dw_xchg, exp); - - if (erts_smp_dw_atomic_cmpxchg_relb(&ptab->vola.tile.last_data, - &dw_new, - &dw_xchg)) - return exp; - else - return dw_aint_to_uint64(&dw_xchg); -} - -#elif defined(ARCH_64) - -union { - erts_smp_atomic_t pid_data; - char align[ERTS_CACHE_LINE_SIZE]; -} last erts_align_attribute(ERTS_CACHE_LINE_SIZE); - -static ERTS_INLINE void -last_data_init_nob(ErtsPTab *ptab, Uint64 val) -{ - erts_smp_atomic_init_nob(&ptab->vola.tile.last_data, (erts_aint_t) val); + return (Uint64) erts_smp_atomic64_cmpxchg_relb(&ptab->vola.tile.last_data, + (erts_aint64_t) new, + (erts_aint64_t) exp); } -static ERTS_INLINE void -last_data_set_relb(ErtsPTab *ptab, Uint64 val) -{ - erts_smp_atomic_set_relb(&ptab->vola.tile.last_data, (erts_aint_t) val); -} - -static ERTS_INLINE Uint64 -last_data_read_nob(ErtsPTab *ptab) -{ - return (Uint64) erts_smp_atomic_read_nob(&ptab->vola.tile.last_data); -} - -static ERTS_INLINE Uint64 -last_data_read_acqb(ErtsPTab *ptab) -{ - return (Uint64) erts_smp_atomic_read_acqb(&ptab->vola.tile.last_data); -} - -static ERTS_INLINE Uint64 -last_data_cmpxchg_relb(ErtsPTab *ptab, Uint64 new, Uint64 exp) -{ - return (Uint64) erts_smp_atomic_cmpxchg_relb(&ptab->vola.tile.last_data, - (erts_aint_t) new, - (erts_aint_t) exp); -} - -#else -# error "Not 64-bit, nor 32-bit architecture..." -#endif - static ERTS_INLINE int last_data_cmp(Uint64 ld1, Uint64 ld2) { diff --git a/erts/emulator/beam/erl_ptab.h b/erts/emulator/beam/erl_ptab.h index e3e05f14af..876241159b 100644 --- a/erts/emulator/beam/erl_ptab.h +++ b/erts/emulator/beam/erl_ptab.h @@ -88,11 +88,7 @@ typedef struct { } ErtsPTabListData; typedef struct { -#ifdef ARCH_32 - erts_smp_dw_atomic_t last_data; -#else - erts_smp_atomic_t last_data; -#endif + erts_smp_atomic64_t last_data; erts_smp_atomic32_t count; erts_smp_atomic32_t aid_ix; erts_smp_atomic32_t fid_ix; diff --git a/erts/emulator/beam/erl_smp.h b/erts/emulator/beam/erl_smp.h index c38ef47d87..6c40edeb3e 100644 --- a/erts/emulator/beam/erl_smp.h +++ b/erts/emulator/beam/erl_smp.h @@ -60,6 +60,7 @@ typedef erts_tsd_key_t erts_smp_tsd_key_t; #define erts_smp_dw_atomic_t erts_dw_atomic_t #define erts_smp_atomic_t erts_atomic_t #define erts_smp_atomic32_t erts_atomic32_t +#define erts_smp_atomic64_t erts_atomic64_t typedef erts_spinlock_t erts_smp_spinlock_t; typedef erts_rwlock_t erts_smp_rwlock_t; void erts_thr_fatal_error(int, char *); /* implemented in erl_init.c */ @@ -95,6 +96,7 @@ typedef int erts_smp_tsd_key_t; #define erts_smp_dw_atomic_t erts_no_dw_atomic_t #define erts_smp_atomic_t erts_no_atomic_t #define erts_smp_atomic32_t erts_no_atomic32_t +#define erts_smp_atomic64_t erts_no_atomic64_t #if __GNUC__ > 2 typedef struct { } erts_smp_spinlock_t; typedef struct { } erts_smp_rwlock_t; @@ -489,6 +491,116 @@ ERTS_GLB_INLINE void erts_smp_thr_sigwait(const sigset_t *set, int *sig); #define erts_smp_atomic32_set_dirty erts_atomic32_set_dirty #define erts_smp_atomic32_read_dirty erts_atomic32_read_dirty +/* 64-bit atomics */ + +#define erts_smp_atomic64_init_nob erts_atomic64_init_nob +#define erts_smp_atomic64_set_nob erts_atomic64_set_nob +#define erts_smp_atomic64_read_nob erts_atomic64_read_nob +#define erts_smp_atomic64_inc_read_nob erts_atomic64_inc_read_nob +#define erts_smp_atomic64_dec_read_nob erts_atomic64_dec_read_nob +#define erts_smp_atomic64_inc_nob erts_atomic64_inc_nob +#define erts_smp_atomic64_dec_nob erts_atomic64_dec_nob +#define erts_smp_atomic64_add_read_nob erts_atomic64_add_read_nob +#define erts_smp_atomic64_add_nob erts_atomic64_add_nob +#define erts_smp_atomic64_read_bor_nob erts_atomic64_read_bor_nob +#define erts_smp_atomic64_read_band_nob erts_atomic64_read_band_nob +#define erts_smp_atomic64_xchg_nob erts_atomic64_xchg_nob +#define erts_smp_atomic64_cmpxchg_nob erts_atomic64_cmpxchg_nob +#define erts_smp_atomic64_read_bset_nob erts_atomic64_read_bset_nob + +#define erts_smp_atomic64_init_mb erts_atomic64_init_mb +#define erts_smp_atomic64_set_mb erts_atomic64_set_mb +#define erts_smp_atomic64_read_mb erts_atomic64_read_mb +#define erts_smp_atomic64_inc_read_mb erts_atomic64_inc_read_mb +#define erts_smp_atomic64_dec_read_mb erts_atomic64_dec_read_mb +#define erts_smp_atomic64_inc_mb erts_atomic64_inc_mb +#define erts_smp_atomic64_dec_mb erts_atomic64_dec_mb +#define erts_smp_atomic64_add_read_mb erts_atomic64_add_read_mb +#define erts_smp_atomic64_add_mb erts_atomic64_add_mb +#define erts_smp_atomic64_read_bor_mb erts_atomic64_read_bor_mb +#define erts_smp_atomic64_read_band_mb erts_atomic64_read_band_mb +#define erts_smp_atomic64_xchg_mb erts_atomic64_xchg_mb +#define erts_smp_atomic64_cmpxchg_mb erts_atomic64_cmpxchg_mb +#define erts_smp_atomic64_read_bset_mb erts_atomic64_read_bset_mb + +#define erts_smp_atomic64_init_acqb erts_atomic64_init_acqb +#define erts_smp_atomic64_set_acqb erts_atomic64_set_acqb +#define erts_smp_atomic64_read_acqb erts_atomic64_read_acqb +#define erts_smp_atomic64_inc_read_acqb erts_atomic64_inc_read_acqb +#define erts_smp_atomic64_dec_read_acqb erts_atomic64_dec_read_acqb +#define erts_smp_atomic64_inc_acqb erts_atomic64_inc_acqb +#define erts_smp_atomic64_dec_acqb erts_atomic64_dec_acqb +#define erts_smp_atomic64_add_read_acqb erts_atomic64_add_read_acqb +#define erts_smp_atomic64_add_acqb erts_atomic64_add_acqb +#define erts_smp_atomic64_read_bor_acqb erts_atomic64_read_bor_acqb +#define erts_smp_atomic64_read_band_acqb erts_atomic64_read_band_acqb +#define erts_smp_atomic64_xchg_acqb erts_atomic64_xchg_acqb +#define erts_smp_atomic64_cmpxchg_acqb erts_atomic64_cmpxchg_acqb +#define erts_smp_atomic64_read_bset_acqb erts_atomic64_read_bset_acqb + +#define erts_smp_atomic64_init_relb erts_atomic64_init_relb +#define erts_smp_atomic64_set_relb erts_atomic64_set_relb +#define erts_smp_atomic64_read_relb erts_atomic64_read_relb +#define erts_smp_atomic64_inc_read_relb erts_atomic64_inc_read_relb +#define erts_smp_atomic64_dec_read_relb erts_atomic64_dec_read_relb +#define erts_smp_atomic64_inc_relb erts_atomic64_inc_relb +#define erts_smp_atomic64_dec_relb erts_atomic64_dec_relb +#define erts_smp_atomic64_add_read_relb erts_atomic64_add_read_relb +#define erts_smp_atomic64_add_relb erts_atomic64_add_relb +#define erts_smp_atomic64_read_bor_relb erts_atomic64_read_bor_relb +#define erts_smp_atomic64_read_band_relb erts_atomic64_read_band_relb +#define erts_smp_atomic64_xchg_relb erts_atomic64_xchg_relb +#define erts_smp_atomic64_cmpxchg_relb erts_atomic64_cmpxchg_relb +#define erts_smp_atomic64_read_bset_relb erts_atomic64_read_bset_relb + +#define erts_smp_atomic64_init_ddrb erts_atomic64_init_ddrb +#define erts_smp_atomic64_set_ddrb erts_atomic64_set_ddrb +#define erts_smp_atomic64_read_ddrb erts_atomic64_read_ddrb +#define erts_smp_atomic64_inc_read_ddrb erts_atomic64_inc_read_ddrb +#define erts_smp_atomic64_dec_read_ddrb erts_atomic64_dec_read_ddrb +#define erts_smp_atomic64_inc_ddrb erts_atomic64_inc_ddrb +#define erts_smp_atomic64_dec_ddrb erts_atomic64_dec_ddrb +#define erts_smp_atomic64_add_read_ddrb erts_atomic64_add_read_ddrb +#define erts_smp_atomic64_add_ddrb erts_atomic64_add_ddrb +#define erts_smp_atomic64_read_bor_ddrb erts_atomic64_read_bor_ddrb +#define erts_smp_atomic64_read_band_ddrb erts_atomic64_read_band_ddrb +#define erts_smp_atomic64_xchg_ddrb erts_atomic64_xchg_ddrb +#define erts_smp_atomic64_cmpxchg_ddrb erts_atomic64_cmpxchg_ddrb +#define erts_smp_atomic64_read_bset_ddrb erts_atomic64_read_bset_ddrb + +#define erts_smp_atomic64_init_rb erts_atomic64_init_rb +#define erts_smp_atomic64_set_rb erts_atomic64_set_rb +#define erts_smp_atomic64_read_rb erts_atomic64_read_rb +#define erts_smp_atomic64_inc_read_rb erts_atomic64_inc_read_rb +#define erts_smp_atomic64_dec_read_rb erts_atomic64_dec_read_rb +#define erts_smp_atomic64_inc_rb erts_atomic64_inc_rb +#define erts_smp_atomic64_dec_rb erts_atomic64_dec_rb +#define erts_smp_atomic64_add_read_rb erts_atomic64_add_read_rb +#define erts_smp_atomic64_add_rb erts_atomic64_add_rb +#define erts_smp_atomic64_read_bor_rb erts_atomic64_read_bor_rb +#define erts_smp_atomic64_read_band_rb erts_atomic64_read_band_rb +#define erts_smp_atomic64_xchg_rb erts_atomic64_xchg_rb +#define erts_smp_atomic64_cmpxchg_rb erts_atomic64_cmpxchg_rb +#define erts_smp_atomic64_read_bset_rb erts_atomic64_read_bset_rb + +#define erts_smp_atomic64_init_wb erts_atomic64_init_wb +#define erts_smp_atomic64_set_wb erts_atomic64_set_wb +#define erts_smp_atomic64_read_wb erts_atomic64_read_wb +#define erts_smp_atomic64_inc_read_wb erts_atomic64_inc_read_wb +#define erts_smp_atomic64_dec_read_wb erts_atomic64_dec_read_wb +#define erts_smp_atomic64_inc_wb erts_atomic64_inc_wb +#define erts_smp_atomic64_dec_wb erts_atomic64_dec_wb +#define erts_smp_atomic64_add_read_wb erts_atomic64_add_read_wb +#define erts_smp_atomic64_add_wb erts_atomic64_add_wb +#define erts_smp_atomic64_read_bor_wb erts_atomic64_read_bor_wb +#define erts_smp_atomic64_read_band_wb erts_atomic64_read_band_wb +#define erts_smp_atomic64_xchg_wb erts_atomic64_xchg_wb +#define erts_smp_atomic64_cmpxchg_wb erts_atomic64_cmpxchg_wb +#define erts_smp_atomic64_read_bset_wb erts_atomic64_read_bset_wb + +#define erts_smp_atomic64_set_dirty erts_atomic64_set_dirty +#define erts_smp_atomic64_read_dirty erts_atomic64_read_dirty + #else /* !ERTS_SMP */ /* Double word size atomics */ @@ -751,6 +863,116 @@ ERTS_GLB_INLINE void erts_smp_thr_sigwait(const sigset_t *set, int *sig); #define erts_smp_atomic32_set_dirty erts_no_atomic32_set #define erts_smp_atomic32_read_dirty erts_no_atomic32_read +/* 64-bit atomics */ + +#define erts_smp_atomic64_init_nob erts_no_atomic64_set +#define erts_smp_atomic64_set_nob erts_no_atomic64_set +#define erts_smp_atomic64_read_nob erts_no_atomic64_read +#define erts_smp_atomic64_inc_read_nob erts_no_atomic64_inc_read +#define erts_smp_atomic64_dec_read_nob erts_no_atomic64_dec_read +#define erts_smp_atomic64_inc_nob erts_no_atomic64_inc +#define erts_smp_atomic64_dec_nob erts_no_atomic64_dec +#define erts_smp_atomic64_add_read_nob erts_no_atomic64_add_read +#define erts_smp_atomic64_add_nob erts_no_atomic64_add +#define erts_smp_atomic64_read_bor_nob erts_no_atomic64_read_bor +#define erts_smp_atomic64_read_band_nob erts_no_atomic64_read_band +#define erts_smp_atomic64_xchg_nob erts_no_atomic64_xchg +#define erts_smp_atomic64_cmpxchg_nob erts_no_atomic64_cmpxchg +#define erts_smp_atomic64_read_bset_nob erts_no_atomic64_read_bset + +#define erts_smp_atomic64_init_mb erts_no_atomic64_set +#define erts_smp_atomic64_set_mb erts_no_atomic64_set +#define erts_smp_atomic64_read_mb erts_no_atomic64_read +#define erts_smp_atomic64_inc_read_mb erts_no_atomic64_inc_read +#define erts_smp_atomic64_dec_read_mb erts_no_atomic64_dec_read +#define erts_smp_atomic64_inc_mb erts_no_atomic64_inc +#define erts_smp_atomic64_dec_mb erts_no_atomic64_dec +#define erts_smp_atomic64_add_read_mb erts_no_atomic64_add_read +#define erts_smp_atomic64_add_mb erts_no_atomic64_add +#define erts_smp_atomic64_read_bor_mb erts_no_atomic64_read_bor +#define erts_smp_atomic64_read_band_mb erts_no_atomic64_read_band +#define erts_smp_atomic64_xchg_mb erts_no_atomic64_xchg +#define erts_smp_atomic64_cmpxchg_mb erts_no_atomic64_cmpxchg +#define erts_smp_atomic64_read_bset_mb erts_no_atomic64_read_bset + +#define erts_smp_atomic64_init_acqb erts_no_atomic64_set +#define erts_smp_atomic64_set_acqb erts_no_atomic64_set +#define erts_smp_atomic64_read_acqb erts_no_atomic64_read +#define erts_smp_atomic64_inc_read_acqb erts_no_atomic64_inc_read +#define erts_smp_atomic64_dec_read_acqb erts_no_atomic64_dec_read +#define erts_smp_atomic64_inc_acqb erts_no_atomic64_inc +#define erts_smp_atomic64_dec_acqb erts_no_atomic64_dec +#define erts_smp_atomic64_add_read_acqb erts_no_atomic64_add_read +#define erts_smp_atomic64_add_acqb erts_no_atomic64_add +#define erts_smp_atomic64_read_bor_acqb erts_no_atomic64_read_bor +#define erts_smp_atomic64_read_band_acqb erts_no_atomic64_read_band +#define erts_smp_atomic64_xchg_acqb erts_no_atomic64_xchg +#define erts_smp_atomic64_cmpxchg_acqb erts_no_atomic64_cmpxchg +#define erts_smp_atomic64_read_bset_acqb erts_no_atomic64_read_bset + +#define erts_smp_atomic64_init_relb erts_no_atomic64_set +#define erts_smp_atomic64_set_relb erts_no_atomic64_set +#define erts_smp_atomic64_read_relb erts_no_atomic64_read +#define erts_smp_atomic64_inc_read_relb erts_no_atomic64_inc_read +#define erts_smp_atomic64_dec_read_relb erts_no_atomic64_dec_read +#define erts_smp_atomic64_inc_relb erts_no_atomic64_inc +#define erts_smp_atomic64_dec_relb erts_no_atomic64_dec +#define erts_smp_atomic64_add_read_relb erts_no_atomic64_add_read +#define erts_smp_atomic64_add_relb erts_no_atomic64_add +#define erts_smp_atomic64_read_bor_relb erts_no_atomic64_read_bor +#define erts_smp_atomic64_read_band_relb erts_no_atomic64_read_band +#define erts_smp_atomic64_xchg_relb erts_no_atomic64_xchg +#define erts_smp_atomic64_cmpxchg_relb erts_no_atomic64_cmpxchg +#define erts_smp_atomic64_read_bset_relb erts_no_atomic64_read_bset + +#define erts_smp_atomic64_init_ddrb erts_no_atomic64_set +#define erts_smp_atomic64_set_ddrb erts_no_atomic64_set +#define erts_smp_atomic64_read_ddrb erts_no_atomic64_read +#define erts_smp_atomic64_inc_read_ddrb erts_no_atomic64_inc_read +#define erts_smp_atomic64_dec_read_ddrb erts_no_atomic64_dec_read +#define erts_smp_atomic64_inc_ddrb erts_no_atomic64_inc +#define erts_smp_atomic64_dec_ddrb erts_no_atomic64_dec +#define erts_smp_atomic64_add_read_ddrb erts_no_atomic64_add_read +#define erts_smp_atomic64_add_ddrb erts_no_atomic64_add +#define erts_smp_atomic64_read_bor_ddrb erts_no_atomic64_read_bor +#define erts_smp_atomic64_read_band_ddrb erts_no_atomic64_read_band +#define erts_smp_atomic64_xchg_ddrb erts_no_atomic64_xchg +#define erts_smp_atomic64_cmpxchg_ddrb erts_no_atomic64_cmpxchg +#define erts_smp_atomic64_read_bset_ddrb erts_no_atomic64_read_bset + +#define erts_smp_atomic64_init_rb erts_no_atomic64_set +#define erts_smp_atomic64_set_rb erts_no_atomic64_set +#define erts_smp_atomic64_read_rb erts_no_atomic64_read +#define erts_smp_atomic64_inc_read_rb erts_no_atomic64_inc_read +#define erts_smp_atomic64_dec_read_rb erts_no_atomic64_dec_read +#define erts_smp_atomic64_inc_rb erts_no_atomic64_inc +#define erts_smp_atomic64_dec_rb erts_no_atomic64_dec +#define erts_smp_atomic64_add_read_rb erts_no_atomic64_add_read +#define erts_smp_atomic64_add_rb erts_no_atomic64_add +#define erts_smp_atomic64_read_bor_rb erts_no_atomic64_read_bor +#define erts_smp_atomic64_read_band_rb erts_no_atomic64_read_band +#define erts_smp_atomic64_xchg_rb erts_no_atomic64_xchg +#define erts_smp_atomic64_cmpxchg_rb erts_no_atomic64_cmpxchg +#define erts_smp_atomic64_read_bset_rb erts_no_atomic64_read_bset + +#define erts_smp_atomic64_init_wb erts_no_atomic64_set +#define erts_smp_atomic64_set_wb erts_no_atomic64_set +#define erts_smp_atomic64_read_wb erts_no_atomic64_read +#define erts_smp_atomic64_inc_read_wb erts_no_atomic64_inc_read +#define erts_smp_atomic64_dec_read_wb erts_no_atomic64_dec_read +#define erts_smp_atomic64_inc_wb erts_no_atomic64_inc +#define erts_smp_atomic64_dec_wb erts_no_atomic64_dec +#define erts_smp_atomic64_add_read_wb erts_no_atomic64_add_read +#define erts_smp_atomic64_add_wb erts_no_atomic64_add +#define erts_smp_atomic64_read_bor_wb erts_no_atomic64_read_bor +#define erts_smp_atomic64_read_band_wb erts_no_atomic64_read_band +#define erts_smp_atomic64_xchg_wb erts_no_atomic64_xchg +#define erts_smp_atomic64_cmpxchg_wb erts_no_atomic64_cmpxchg +#define erts_smp_atomic64_read_bset_wb erts_no_atomic64_read_bset + +#define erts_smp_atomic64_set_dirty erts_no_atomic64_set +#define erts_smp_atomic64_read_dirty erts_no_atomic64_read + #endif /* !ERTS_SMP */ #if ERTS_GLB_INLINE_INCL_FUNC_DEF diff --git a/erts/emulator/beam/erl_thr_progress.c b/erts/emulator/beam/erl_thr_progress.c index 545a0343d0..664c479eb6 100644 --- a/erts/emulator/beam/erl_thr_progress.c +++ b/erts/emulator/beam/erl_thr_progress.c @@ -115,70 +115,24 @@ #undef read_nob #define read_nob erts_thr_prgr_read_nob__ -#ifdef ARCH_64 - static ERTS_INLINE void set_mb(ERTS_THR_PRGR_ATOMIC *atmc, ErtsThrPrgrVal val) { - erts_atomic_set_mb(atmc, val); + erts_atomic64_set_mb(atmc, (erts_aint64_t) val); } static ERTS_INLINE void set_nob(ERTS_THR_PRGR_ATOMIC *atmc, ErtsThrPrgrVal val) { - erts_atomic_set_nob(atmc, val); + erts_atomic64_set_nob(atmc, (erts_aint64_t) val); } static ERTS_INLINE void init_nob(ERTS_THR_PRGR_ATOMIC *atmc, ErtsThrPrgrVal val) { - erts_atomic_init_nob(atmc, val); -} - -#else - -#undef dw_aint_to_val -#define dw_aint_to_val erts_thr_prgr_dw_aint_to_val__ - -static void -val_to_dw_aint(erts_dw_aint_t *dw_aint, ErtsThrPrgrVal val) -{ -#ifdef ETHR_SU_DW_NAINT_T__ - dw_aint->dw_sint = (ETHR_SU_DW_NAINT_T__) val; -#else - dw_aint->sint[ERTS_DW_AINT_LOW_WORD] - = (erts_aint_t) (val & 0xffffffff); - dw_aint->sint[ERTS_DW_AINT_HIGH_WORD] - = (erts_aint_t) ((val >> 32) & 0xffffffff); -#endif + erts_atomic64_init_nob(atmc, (erts_aint64_t) val); } -static ERTS_INLINE void -set_mb(ERTS_THR_PRGR_ATOMIC *atmc, ErtsThrPrgrVal val) -{ - erts_dw_aint_t dw_aint; - val_to_dw_aint(&dw_aint, val); - erts_dw_atomic_set_mb(atmc, &dw_aint); -} - -static ERTS_INLINE void -set_nob(ERTS_THR_PRGR_ATOMIC *atmc, ErtsThrPrgrVal val) -{ - erts_dw_aint_t dw_aint; - val_to_dw_aint(&dw_aint, val); - erts_dw_atomic_set_nob(atmc, &dw_aint); -} - -static ERTS_INLINE void -init_nob(ERTS_THR_PRGR_ATOMIC *atmc, ErtsThrPrgrVal val) -{ - erts_dw_aint_t dw_aint; - val_to_dw_aint(&dw_aint, val); - erts_dw_atomic_init_nob(atmc, &dw_aint); -} - -#endif - /* #define ERTS_THR_PROGRESS_STATE_DEBUG */ #ifdef ERTS_THR_PROGRESS_STATE_DEBUG diff --git a/erts/emulator/beam/erl_thr_progress.h b/erts/emulator/beam/erl_thr_progress.h index 5f392944c2..03ddbd467c 100644 --- a/erts/emulator/beam/erl_thr_progress.h +++ b/erts/emulator/beam/erl_thr_progress.h @@ -115,11 +115,7 @@ struct ErtsThrPrgrLaterOp_ { extern erts_tsd_key_t erts_thr_prgr_data_key__; -#ifdef ARCH_64 -# define ERTS_THR_PRGR_ATOMIC erts_atomic_t -#else /* ARCH_32 */ -# define ERTS_THR_PRGR_ATOMIC erts_dw_atomic_t -#endif +#define ERTS_THR_PRGR_ATOMIC erts_atomic64_t typedef struct { void *arg; @@ -158,10 +154,6 @@ void erts_thr_progress_unmanaged_continue__(int umrefc_ix); void erts_thr_progress_dbg_print_state(void); -#ifdef ARCH_32 -#define ERTS_THR_PRGR_ATOMIC erts_dw_atomic_t -ERTS_GLB_INLINE ErtsThrPrgrVal erts_thr_prgr_dw_aint_to_val__(erts_dw_aint_t *dw_aint); -#endif ERTS_GLB_INLINE ErtsThrPrgrVal erts_thr_prgr_read_nob__(ERTS_THR_PRGR_ATOMIC *atmc); ERTS_GLB_INLINE ErtsThrPrgrVal erts_thr_prgr_read_acqb__(ERTS_THR_PRGR_ATOMIC *atmc); ERTS_GLB_INLINE ErtsThrPrgrVal erts_thr_prgr_read_mb__(ERTS_THR_PRGR_ATOMIC *atmc); @@ -184,68 +176,24 @@ ERTS_GLB_INLINE int erts_thr_progress_has_reached(ErtsThrPrgrVal val); #if ERTS_GLB_INLINE_INCL_FUNC_DEF -#ifdef ARCH_64 - -ERTS_GLB_INLINE ErtsThrPrgrVal -erts_thr_prgr_read_nob__(ERTS_THR_PRGR_ATOMIC *atmc) -{ - return (ErtsThrPrgrVal) erts_atomic_read_nob(atmc); -} - -ERTS_GLB_INLINE ErtsThrPrgrVal -erts_thr_prgr_read_acqb__(ERTS_THR_PRGR_ATOMIC *atmc) -{ - return (ErtsThrPrgrVal) erts_atomic_read_acqb(atmc); -} - -ERTS_GLB_INLINE ErtsThrPrgrVal -erts_thr_prgr_read_mb__(ERTS_THR_PRGR_ATOMIC *atmc) -{ - return (ErtsThrPrgrVal) erts_atomic_read_mb(atmc); -} - -#else /* ARCH_32 */ - -ERTS_GLB_INLINE ErtsThrPrgrVal -erts_thr_prgr_dw_aint_to_val__(erts_dw_aint_t *dw_aint) -{ -#ifdef ETHR_SU_DW_NAINT_T__ - return (ErtsThrPrgrVal) dw_aint->dw_sint; -#else - ErtsThrPrgrVal res; - res = (ErtsThrPrgrVal) ((Uint32) dw_aint->sint[ERTS_DW_AINT_HIGH_WORD]); - res <<= 32; - res |= (ErtsThrPrgrVal) ((Uint32) dw_aint->sint[ERTS_DW_AINT_LOW_WORD]); - return res; -#endif -} - ERTS_GLB_INLINE ErtsThrPrgrVal erts_thr_prgr_read_nob__(ERTS_THR_PRGR_ATOMIC *atmc) { - erts_dw_aint_t dw_aint; - erts_dw_atomic_read_nob(atmc, &dw_aint); - return erts_thr_prgr_dw_aint_to_val__(&dw_aint); + return (ErtsThrPrgrVal) erts_atomic64_read_nob(atmc); } ERTS_GLB_INLINE ErtsThrPrgrVal erts_thr_prgr_read_acqb__(ERTS_THR_PRGR_ATOMIC *atmc) { - erts_dw_aint_t dw_aint; - erts_dw_atomic_read_acqb(atmc, &dw_aint); - return erts_thr_prgr_dw_aint_to_val__(&dw_aint); + return (ErtsThrPrgrVal) erts_atomic64_read_acqb(atmc); } ERTS_GLB_INLINE ErtsThrPrgrVal erts_thr_prgr_read_mb__(ERTS_THR_PRGR_ATOMIC *atmc) { - erts_dw_aint_t dw_aint; - erts_dw_atomic_read_mb(atmc, &dw_aint); - return erts_thr_prgr_dw_aint_to_val__(&dw_aint); + return (ErtsThrPrgrVal) erts_atomic64_read_mb(atmc); } -#endif - ERTS_GLB_INLINE int erts_thr_progress_is_managed_thread(void) { diff --git a/erts/emulator/beam/erl_threads.h b/erts/emulator/beam/erl_threads.h index 80026104db..7214f3ea33 100644 --- a/erts/emulator/beam/erl_threads.h +++ b/erts/emulator/beam/erl_threads.h @@ -344,6 +344,16 @@ typedef ethr_ts_event erts_tse_t; #define erts_aint32_t ethr_sint32_t #define erts_atomic32_t ethr_atomic32_t +#if defined(ARCH_32) +# define erts_atomic64_t ethr_dw_atomic_t +# define erts_aint64_t ethr_sint64_t +#elif defined(ARCH_64) +# define erts_atomic64_t ethr_atomic_t +# define erts_aint64_t ethr_sint_t +#else +# error "Not supported architecture" +#endif + #define ERTS_DW_AINT_HIGH_WORD ETHR_DW_SINT_HIGH_WORD #define ERTS_DW_AINT_LOW_WORD ETHR_DW_SINT_LOW_WORD @@ -414,10 +424,12 @@ typedef int erts_tse_t; typedef struct { SWord sint[2]; } erts_dw_aint_t; typedef SWord erts_aint_t; typedef Sint32 erts_aint32_t; +typedef Sint64 erts_aint64_t; #define erts_dw_atomic_t erts_dw_aint_t #define erts_atomic_t erts_aint_t #define erts_atomic32_t erts_aint32_t +#define erts_atomic64_t erts_aint64_t #if __GNUC__ > 2 typedef struct { } erts_spinlock_t; @@ -446,6 +458,7 @@ typedef struct { int gcc_is_buggy; } erts_rwlock_t; #define erts_no_dw_atomic_t erts_dw_aint_t #define erts_no_atomic_t erts_aint_t #define erts_no_atomic32_t erts_aint32_t +#define erts_no_atomic64_t erts_aint64_t #define ERTS_AINT_NULL ((erts_aint_t) NULL) @@ -570,6 +583,29 @@ ERTS_GLB_INLINE erts_aint32_t erts_no_atomic32_cmpxchg(erts_no_atomic32_t *xchgp ERTS_GLB_INLINE erts_aint32_t erts_no_atomic32_read_bset(erts_no_atomic32_t *var, erts_aint32_t mask, erts_aint32_t set); +ERTS_GLB_INLINE void erts_no_atomic64_set(erts_no_atomic64_t *var, + erts_aint64_t i); +ERTS_GLB_INLINE erts_aint64_t erts_no_atomic64_read(erts_no_atomic64_t *var); +ERTS_GLB_INLINE erts_aint64_t erts_no_atomic64_inc_read(erts_no_atomic64_t *incp); +ERTS_GLB_INLINE erts_aint64_t erts_no_atomic64_dec_read(erts_no_atomic64_t *decp); +ERTS_GLB_INLINE void erts_no_atomic64_inc(erts_no_atomic64_t *incp); +ERTS_GLB_INLINE void erts_no_atomic64_dec(erts_no_atomic64_t *decp); +ERTS_GLB_INLINE erts_aint64_t erts_no_atomic64_add_read(erts_no_atomic64_t *addp, + erts_aint64_t i); +ERTS_GLB_INLINE void erts_no_atomic64_add(erts_no_atomic64_t *addp, + erts_aint64_t i); +ERTS_GLB_INLINE erts_aint64_t erts_no_atomic64_read_bor(erts_no_atomic64_t *var, + erts_aint64_t mask); +ERTS_GLB_INLINE erts_aint64_t erts_no_atomic64_read_band(erts_no_atomic64_t *var, + erts_aint64_t mask); +ERTS_GLB_INLINE erts_aint64_t erts_no_atomic64_xchg(erts_no_atomic64_t *xchgp, + erts_aint64_t new); +ERTS_GLB_INLINE erts_aint64_t erts_no_atomic64_cmpxchg(erts_no_atomic64_t *xchgp, + erts_aint64_t new, + erts_aint64_t expected); +ERTS_GLB_INLINE erts_aint64_t erts_no_atomic64_read_bset(erts_no_atomic64_t *var, + erts_aint64_t mask, + erts_aint64_t set); ERTS_GLB_INLINE void erts_spinlock_init_x_opt(erts_spinlock_t *lock, char *name, @@ -1200,6 +1236,441 @@ erts_atomic32_read_dirty(erts_atomic32_t *var) #endif +/* 64-bit atomics */ + +#if defined(ARCH_64) + +#define erts_atomic64_init_nob ethr_atomic_init +#define erts_atomic64_set_nob ethr_atomic_set +#define erts_atomic64_read_nob ethr_atomic_read +#define erts_atomic64_inc_read_nob ethr_atomic_inc_read +#define erts_atomic64_dec_read_nob ethr_atomic_dec_read +#define erts_atomic64_inc_nob ethr_atomic_inc +#define erts_atomic64_dec_nob ethr_atomic_dec +#define erts_atomic64_add_read_nob ethr_atomic_add_read +#define erts_atomic64_add_nob ethr_atomic_add +#define erts_atomic64_read_bor_nob ethr_atomic_read_bor +#define erts_atomic64_read_band_nob ethr_atomic_read_band +#define erts_atomic64_xchg_nob ethr_atomic_xchg +#define erts_atomic64_cmpxchg_nob ethr_atomic_cmpxchg +#define erts_atomic64_read_bset_nob erts_atomic_read_bset_nob + +#define erts_atomic64_init_mb ethr_atomic_init_mb +#define erts_atomic64_set_mb ethr_atomic_set_mb +#define erts_atomic64_read_mb ethr_atomic_read_mb +#define erts_atomic64_inc_read_mb ethr_atomic_inc_read_mb +#define erts_atomic64_dec_read_mb ethr_atomic_dec_read_mb +#define erts_atomic64_inc_mb ethr_atomic_inc_mb +#define erts_atomic64_dec_mb ethr_atomic_dec_mb +#define erts_atomic64_add_read_mb ethr_atomic_add_read_mb +#define erts_atomic64_add_mb ethr_atomic_add_mb +#define erts_atomic64_read_bor_mb ethr_atomic_read_bor_mb +#define erts_atomic64_read_band_mb ethr_atomic_read_band_mb +#define erts_atomic64_xchg_mb ethr_atomic_xchg_mb +#define erts_atomic64_cmpxchg_mb ethr_atomic_cmpxchg_mb +#define erts_atomic64_read_bset_mb erts_atomic_read_bset_mb + +#define erts_atomic64_init_acqb ethr_atomic_init_acqb +#define erts_atomic64_set_acqb ethr_atomic_set_acqb +#define erts_atomic64_read_acqb ethr_atomic_read_acqb +#define erts_atomic64_inc_read_acqb ethr_atomic_inc_read_acqb +#define erts_atomic64_dec_read_acqb ethr_atomic_dec_read_acqb +#define erts_atomic64_inc_acqb ethr_atomic_inc_acqb +#define erts_atomic64_dec_acqb ethr_atomic_dec_acqb +#define erts_atomic64_add_read_acqb ethr_atomic_add_read_acqb +#define erts_atomic64_add_acqb ethr_atomic_add_acqb +#define erts_atomic64_read_bor_acqb ethr_atomic_read_bor_acqb +#define erts_atomic64_read_band_acqb ethr_atomic_read_band_acqb +#define erts_atomic64_xchg_acqb ethr_atomic_xchg_acqb +#define erts_atomic64_cmpxchg_acqb ethr_atomic_cmpxchg_acqb +#define erts_atomic64_read_bset_acqb erts_atomic_read_bset_acqb + +#define erts_atomic64_init_relb ethr_atomic_init_relb +#define erts_atomic64_set_relb ethr_atomic_set_relb +#define erts_atomic64_read_relb ethr_atomic_read_relb +#define erts_atomic64_inc_read_relb ethr_atomic_inc_read_relb +#define erts_atomic64_dec_read_relb ethr_atomic_dec_read_relb +#define erts_atomic64_inc_relb ethr_atomic_inc_relb +#define erts_atomic64_dec_relb ethr_atomic_dec_relb +#define erts_atomic64_add_read_relb ethr_atomic_add_read_relb +#define erts_atomic64_add_relb ethr_atomic_add_relb +#define erts_atomic64_read_bor_relb ethr_atomic_read_bor_relb +#define erts_atomic64_read_band_relb ethr_atomic_read_band_relb +#define erts_atomic64_xchg_relb ethr_atomic_xchg_relb +#define erts_atomic64_cmpxchg_relb ethr_atomic_cmpxchg_relb +#define erts_atomic64_read_bset_relb erts_atomic_read_bset_relb + +#define erts_atomic64_init_ddrb ethr_atomic_init_ddrb +#define erts_atomic64_set_ddrb ethr_atomic_set_ddrb +#define erts_atomic64_read_ddrb ethr_atomic_read_ddrb +#define erts_atomic64_inc_read_ddrb ethr_atomic_inc_read_ddrb +#define erts_atomic64_dec_read_ddrb ethr_atomic_dec_read_ddrb +#define erts_atomic64_inc_ddrb ethr_atomic_inc_ddrb +#define erts_atomic64_dec_ddrb ethr_atomic_dec_ddrb +#define erts_atomic64_add_read_ddrb ethr_atomic_add_read_ddrb +#define erts_atomic64_add_ddrb ethr_atomic_add_ddrb +#define erts_atomic64_read_bor_ddrb ethr_atomic_read_bor_ddrb +#define erts_atomic64_read_band_ddrb ethr_atomic_read_band_ddrb +#define erts_atomic64_xchg_ddrb ethr_atomic_xchg_ddrb +#define erts_atomic64_cmpxchg_ddrb ethr_atomic_cmpxchg_ddrb +#define erts_atomic64_read_bset_ddrb erts_atomic_read_bset_ddrb + +#define erts_atomic64_init_rb ethr_atomic_init_rb +#define erts_atomic64_set_rb ethr_atomic_set_rb +#define erts_atomic64_read_rb ethr_atomic_read_rb +#define erts_atomic64_inc_read_rb ethr_atomic_inc_read_rb +#define erts_atomic64_dec_read_rb ethr_atomic_dec_read_rb +#define erts_atomic64_inc_rb ethr_atomic_inc_rb +#define erts_atomic64_dec_rb ethr_atomic_dec_rb +#define erts_atomic64_add_read_rb ethr_atomic_add_read_rb +#define erts_atomic64_add_rb ethr_atomic_add_rb +#define erts_atomic64_read_bor_rb ethr_atomic_read_bor_rb +#define erts_atomic64_read_band_rb ethr_atomic_read_band_rb +#define erts_atomic64_xchg_rb ethr_atomic_xchg_rb +#define erts_atomic64_cmpxchg_rb ethr_atomic_cmpxchg_rb +#define erts_atomic64_read_bset_rb erts_atomic_read_bset_rb + +#define erts_atomic64_init_wb ethr_atomic_init_wb +#define erts_atomic64_set_wb ethr_atomic_set_wb +#define erts_atomic64_read_wb ethr_atomic_read_wb +#define erts_atomic64_inc_read_wb ethr_atomic_inc_read_wb +#define erts_atomic64_dec_read_wb ethr_atomic_dec_read_wb +#define erts_atomic64_inc_wb ethr_atomic_inc_wb +#define erts_atomic64_dec_wb ethr_atomic_dec_wb +#define erts_atomic64_add_read_wb ethr_atomic_add_read_wb +#define erts_atomic64_add_wb ethr_atomic_add_wb +#define erts_atomic64_read_bor_wb ethr_atomic_read_bor_wb +#define erts_atomic64_read_band_wb ethr_atomic_read_band_wb +#define erts_atomic64_xchg_wb ethr_atomic_xchg_wb +#define erts_atomic64_cmpxchg_wb ethr_atomic_cmpxchg_wb +#define erts_atomic64_read_bset_wb erts_atomic_read_bset_wb + +#define erts_atomic64_set_dirty erts_atomic_set_dirty +#define erts_atomic64_read_dirty erts_atomic_read_dirty + +#elif defined(ARCH_32) + +#undef ERTS_ATOMIC64_OPS_DECL__ + +#define ERTS_ATOMIC64_OPS_DECL__(BARRIER) \ +ERTS_GLB_INLINE void \ +erts_atomic64_init_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val); \ +ERTS_GLB_INLINE void \ +erts_atomic64_set_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val); \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_read_ ## BARRIER(erts_atomic64_t *var); \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_inc_read_ ## BARRIER(erts_atomic64_t *var); \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_dec_read_ ## BARRIER(erts_atomic64_t *var); \ +ERTS_GLB_INLINE void \ +erts_atomic64_inc_ ## BARRIER(erts_atomic64_t *var); \ +ERTS_GLB_INLINE void \ +erts_atomic64_dec_ ## BARRIER(erts_atomic64_t *var); \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_add_read_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val); \ +ERTS_GLB_INLINE void \ +erts_atomic64_add_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val); \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_read_bor_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val); \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_read_band_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val); \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_xchg_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val); \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_cmpxchg_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t new, \ + erts_aint64_t exp); \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_read_bset_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t mask, \ + erts_aint64_t set) + +ERTS_ATOMIC64_OPS_DECL__(nob); +ERTS_ATOMIC64_OPS_DECL__(mb); +ERTS_ATOMIC64_OPS_DECL__(acqb); +ERTS_ATOMIC64_OPS_DECL__(relb); +ERTS_ATOMIC64_OPS_DECL__(ddrb); +ERTS_ATOMIC64_OPS_DECL__(rb); +ERTS_ATOMIC64_OPS_DECL__(wb); + +#undef ERTS_ATOMIC64_OPS_DECL__ + +ERTS_GLB_INLINE void +erts_atomic64_set_dirty(erts_atomic64_t *var, erts_aint64_t val); +ERTS_GLB_INLINE erts_aint64_t +erts_atomic64_read_dirty(erts_atomic64_t *var); + +#if ERTS_GLB_INLINE_INCL_FUNC_DEF + +/* + * The ethr_dw_atomic_*_nob() functions below + * are here to make it possible for the + * ERTS_ATOMIC64_OPS_IMPL__() to map erts + * barriers to ethread barriers... + */ +static ERTS_INLINE void +ethr_dw_atomic_init_nob(ethr_dw_atomic_t *var, + ethr_dw_sint_t *val) +{ + ethr_dw_atomic_init(var, val); +} + +static ERTS_INLINE void +ethr_dw_atomic_set_nob(ethr_dw_atomic_t *var, + ethr_dw_sint_t *val) +{ + ethr_dw_atomic_set(var, val); +} + +static ERTS_INLINE void +ethr_dw_atomic_read_nob(ethr_dw_atomic_t *var, + ethr_dw_sint_t *val) +{ + ethr_dw_atomic_read(var, val); +} + +static ERTS_INLINE int +ethr_dw_atomic_cmpxchg_nob(ethr_dw_atomic_t *var, + ethr_dw_sint_t *new, + ethr_dw_sint_t *xchg) +{ + return ethr_dw_atomic_cmpxchg(var, new, xchg); +} + +#undef ERTS_ATOMIC64_OPS_IMPL__ +#undef ERTS_ATOMIC64_DW_CMPXCHG_IMPL__ +#undef ERTS_DW_SINT_TO_AINT64__ +#undef ERTS_AINT64_TO_DW_SINT__ + +#ifdef ETHR_SU_DW_NAINT_T__ +#define ERTS_DW_SINT_TO_AINT64__(DW) \ + ((erts_aint64_t) DW.dw_sint) +#define ERTS_AINT64_TO_DW_SINT__(DW, AINT64) \ + (DW.dw_sint = (ETHR_SU_DW_NAINT_T__) AINT64) +#else /* !ETHR_SU_DW_NAINT_T__ */ +#define ERTS_DW_SINT_TO_AINT64__(DW) \ + ((((erts_aint64_t) DW.sint[ETHR_DW_SINT_HIGH_WORD]) << 32) \ + | (((erts_aint64_t) DW.sint[ETHR_DW_SINT_LOW_WORD]) \ + & ((erts_aint64_t) 0xffffffff))) +#define ERTS_AINT64_TO_DW_SINT__(DW, AINT64) \ + do { \ + DW.sint[ETHR_DW_SINT_LOW_WORD] = \ + (ethr_sint_t) (AINT64 & 0xffffffff); \ + DW.sint[ETHR_DW_SINT_HIGH_WORD] = \ + (ethr_sint_t) ((AINT64 >> 32) & 0xffffffff); \ + } while (0) +#endif /* !ETHR_SU_DW_NAINT_T__ */ + +#define ERTS_ATOMIC64_DW_CMPXCHG_IMPL__(CmpXchgOp, \ + AVarP, XchgVar, NewVar, \ + ModificationCode) \ +do { \ + ethr_dw_sint_t dw_xchg__, dw_new__; \ + ethr_dw_atomic_read(AVarP, &dw_xchg__); \ + do { \ + XchgVar = ERTS_DW_SINT_TO_AINT64__(dw_xchg__); \ + { \ + ModificationCode; \ + } \ + ERTS_AINT64_TO_DW_SINT__(dw_new__, NewVar); \ + } while (!CmpXchgOp((AVarP), &dw_new__, &dw_xchg__)); \ +} while (0) + +#define ERTS_ATOMIC64_OPS_IMPL__(BARRIER) \ + \ +ERTS_GLB_INLINE void \ +erts_atomic64_init_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val) \ +{ \ + ethr_dw_sint_t dw; \ + ERTS_AINT64_TO_DW_SINT__(dw, val); \ + ethr_dw_atomic_init_ ## BARRIER(var, &dw); \ +} \ + \ +ERTS_GLB_INLINE void \ +erts_atomic64_set_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val) \ +{ \ + ethr_dw_sint_t dw; \ + ERTS_AINT64_TO_DW_SINT__(dw, val); \ + ethr_dw_atomic_set_ ## BARRIER(var, &dw); \ +} \ + \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_read_ ## BARRIER(erts_atomic64_t *var) \ +{ \ + ethr_dw_sint_t dw; \ + ethr_dw_atomic_read_ ## BARRIER(var, &dw); \ + return ERTS_DW_SINT_TO_AINT64__(dw); \ +} \ + \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_inc_read_ ## BARRIER(erts_atomic64_t *var) \ +{ \ + erts_aint64_t xchg, new; \ + ERTS_ATOMIC64_DW_CMPXCHG_IMPL__(ethr_dw_atomic_cmpxchg_ ## BARRIER, \ + var, xchg, new, \ + new = xchg + 1); \ + return new; \ +} \ + \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_dec_read_ ## BARRIER(erts_atomic64_t *var) \ +{ \ + erts_aint64_t xchg, new; \ + ERTS_ATOMIC64_DW_CMPXCHG_IMPL__(ethr_dw_atomic_cmpxchg_ ## BARRIER, \ + var, xchg, new, \ + new = xchg - 1); \ + return new; \ +} \ + \ +ERTS_GLB_INLINE void \ +erts_atomic64_inc_ ## BARRIER(erts_atomic64_t *var) \ +{ \ + erts_aint64_t xchg, new; \ + ERTS_ATOMIC64_DW_CMPXCHG_IMPL__(ethr_dw_atomic_cmpxchg_ ## BARRIER, \ + var, xchg, new, \ + new = xchg + 1); \ +} \ + \ +ERTS_GLB_INLINE void \ +erts_atomic64_dec_ ## BARRIER(erts_atomic64_t *var) \ +{ \ + erts_aint64_t xchg, new; \ + ERTS_ATOMIC64_DW_CMPXCHG_IMPL__(ethr_dw_atomic_cmpxchg_ ## BARRIER, \ + var, xchg, new, \ + new = xchg - 1); \ +} \ + \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_add_read_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val) \ +{ \ + erts_aint64_t xchg, new; \ + ERTS_ATOMIC64_DW_CMPXCHG_IMPL__(ethr_dw_atomic_cmpxchg_ ## BARRIER, \ + var, xchg, new, \ + new = xchg + val); \ + return new; \ +} \ + \ +ERTS_GLB_INLINE void \ +erts_atomic64_add_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val) \ +{ \ + erts_aint64_t xchg, new; \ + ERTS_ATOMIC64_DW_CMPXCHG_IMPL__(ethr_dw_atomic_cmpxchg_ ## BARRIER, \ + var, xchg, new, \ + new = xchg + val); \ +} \ + \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_read_bor_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val) \ +{ \ + erts_aint64_t xchg, new; \ + ERTS_ATOMIC64_DW_CMPXCHG_IMPL__(ethr_dw_atomic_cmpxchg_ ## BARRIER, \ + var, xchg, new, \ + new = xchg | val); \ + return xchg; \ +} \ + \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_read_band_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val) \ +{ \ + erts_aint64_t xchg, new; \ + ERTS_ATOMIC64_DW_CMPXCHG_IMPL__(ethr_dw_atomic_cmpxchg_ ## BARRIER, \ + var, xchg, new, \ + new = xchg & val); \ + return xchg; \ +} \ + \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_xchg_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val) \ +{ \ + erts_aint64_t xchg, new; \ + ERTS_ATOMIC64_DW_CMPXCHG_IMPL__(ethr_dw_atomic_cmpxchg_ ## BARRIER, \ + var, xchg, new, \ + new = val); \ + return xchg; \ +} \ + \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_cmpxchg_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t new, \ + erts_aint64_t exp) \ +{ \ + ethr_dw_sint_t dw_xchg, dw_new; \ + ERTS_AINT64_TO_DW_SINT__(dw_xchg, exp); \ + ERTS_AINT64_TO_DW_SINT__(dw_new, new); \ + if (ethr_dw_atomic_cmpxchg_ ## BARRIER(var, &dw_new, &dw_xchg)) \ + return exp; \ + return ERTS_DW_SINT_TO_AINT64__(dw_xchg); \ +} \ + \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_read_bset_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t mask, \ + erts_aint64_t set) \ +{ \ + erts_aint64_t xchg, new; \ + ERTS_ATOMIC64_DW_CMPXCHG_IMPL__(ethr_dw_atomic_cmpxchg_ ## BARRIER, \ + var, xchg, new, \ + { \ + new = xchg & ~mask; \ + new |= mask & set; \ + }); \ + return xchg; \ +} + +ERTS_ATOMIC64_OPS_IMPL__(nob) +ERTS_ATOMIC64_OPS_IMPL__(mb) +ERTS_ATOMIC64_OPS_IMPL__(acqb) +ERTS_ATOMIC64_OPS_IMPL__(relb) +ERTS_ATOMIC64_OPS_IMPL__(ddrb) +ERTS_ATOMIC64_OPS_IMPL__(rb) +ERTS_ATOMIC64_OPS_IMPL__(wb) + +#undef ERTS_ATOMIC64_OPS_IMPL__ +#undef ERTS_ATOMIC64_DW_CMPXCHG_IMPL__ + +ERTS_GLB_INLINE void +erts_atomic64_set_dirty(erts_atomic64_t *var, erts_aint64_t val) +{ + ethr_sint_t *sint = ethr_dw_atomic_addr(var); + ethr_dw_sint_t dw; + ERTS_AINT64_TO_DW_SINT__(dw, val); + sint[0] = dw.sint[0]; + sint[1] = dw.sint[1]; +} + +ERTS_GLB_INLINE erts_aint64_t +erts_atomic64_read_dirty(erts_atomic64_t *var) +{ + ethr_sint_t *sint; + ethr_dw_sint_t dw; + sint = ethr_dw_atomic_addr(var); + dw.sint[0] = sint[0]; + dw.sint[1] = sint[1]; + return ERTS_DW_SINT_TO_AINT64__(dw); +} + +#undef ERTS_DW_SINT_TO_AINT64__ +#undef ERTS_AINT64_TO_DW_SINT__ + +#endif /* ERTS_GLB_INLINE_INCL_FUNC_DEF */ + +#endif /* ARCH_32 */ + #else /* !USE_THREADS */ /* Double word size atomics */ @@ -1462,6 +1933,116 @@ erts_atomic32_read_dirty(erts_atomic32_t *var) #define erts_atomic32_set_dirty erts_no_atomic32_set #define erts_atomic32_read_dirty erts_no_atomic32_read +/* 64-bit atomics */ + +#define erts_atomic64_init_nob erts_no_atomic64_set +#define erts_atomic64_set_nob erts_no_atomic64_set +#define erts_atomic64_read_nob erts_no_atomic64_read +#define erts_atomic64_inc_read_nob erts_no_atomic64_inc_read +#define erts_atomic64_dec_read_nob erts_no_atomic64_dec_read +#define erts_atomic64_inc_nob erts_no_atomic64_inc +#define erts_atomic64_dec_nob erts_no_atomic64_dec +#define erts_atomic64_add_read_nob erts_no_atomic64_add_read +#define erts_atomic64_add_nob erts_no_atomic64_add +#define erts_atomic64_read_bor_nob erts_no_atomic64_read_bor +#define erts_atomic64_read_band_nob erts_no_atomic64_read_band +#define erts_atomic64_xchg_nob erts_no_atomic64_xchg +#define erts_atomic64_cmpxchg_nob erts_no_atomic64_cmpxchg +#define erts_atomic64_read_bset_nob erts_no_atomic64_read_bset + +#define erts_atomic64_init_mb erts_no_atomic64_set +#define erts_atomic64_set_mb erts_no_atomic64_set +#define erts_atomic64_read_mb erts_no_atomic64_read +#define erts_atomic64_inc_read_mb erts_no_atomic64_inc_read +#define erts_atomic64_dec_read_mb erts_no_atomic64_dec_read +#define erts_atomic64_inc_mb erts_no_atomic64_inc +#define erts_atomic64_dec_mb erts_no_atomic64_dec +#define erts_atomic64_add_read_mb erts_no_atomic64_add_read +#define erts_atomic64_add_mb erts_no_atomic64_add +#define erts_atomic64_read_bor_mb erts_no_atomic64_read_bor +#define erts_atomic64_read_band_mb erts_no_atomic64_read_band +#define erts_atomic64_xchg_mb erts_no_atomic64_xchg +#define erts_atomic64_cmpxchg_mb erts_no_atomic64_cmpxchg +#define erts_atomic64_read_bset_mb erts_no_atomic64_read_bset + +#define erts_atomic64_init_acqb erts_no_atomic64_set +#define erts_atomic64_set_acqb erts_no_atomic64_set +#define erts_atomic64_read_acqb erts_no_atomic64_read +#define erts_atomic64_inc_read_acqb erts_no_atomic64_inc_read +#define erts_atomic64_dec_read_acqb erts_no_atomic64_dec_read +#define erts_atomic64_inc_acqb erts_no_atomic64_inc +#define erts_atomic64_dec_acqb erts_no_atomic64_dec +#define erts_atomic64_add_read_acqb erts_no_atomic64_add_read +#define erts_atomic64_add_acqb erts_no_atomic64_add +#define erts_atomic64_read_bor_acqb erts_no_atomic64_read_bor +#define erts_atomic64_read_band_acqb erts_no_atomic64_read_band +#define erts_atomic64_xchg_acqb erts_no_atomic64_xchg +#define erts_atomic64_cmpxchg_acqb erts_no_atomic64_cmpxchg +#define erts_atomic64_read_bset_acqb erts_no_atomic64_read_bset + +#define erts_atomic64_init_relb erts_no_atomic64_set +#define erts_atomic64_set_relb erts_no_atomic64_set +#define erts_atomic64_read_relb erts_no_atomic64_read +#define erts_atomic64_inc_read_relb erts_no_atomic64_inc_read +#define erts_atomic64_dec_read_relb erts_no_atomic64_dec_read +#define erts_atomic64_inc_relb erts_no_atomic64_inc +#define erts_atomic64_dec_relb erts_no_atomic64_dec +#define erts_atomic64_add_read_relb erts_no_atomic64_add_read +#define erts_atomic64_add_relb erts_no_atomic64_add +#define erts_atomic64_read_bor_relb erts_no_atomic64_read_bor +#define erts_atomic64_read_band_relb erts_no_atomic64_read_band +#define erts_atomic64_xchg_relb erts_no_atomic64_xchg +#define erts_atomic64_cmpxchg_relb erts_no_atomic64_cmpxchg +#define erts_atomic64_read_bset_relb erts_no_atomic64_read_bset + +#define erts_atomic64_init_ddrb erts_no_atomic64_set +#define erts_atomic64_set_ddrb erts_no_atomic64_set +#define erts_atomic64_read_ddrb erts_no_atomic64_read +#define erts_atomic64_inc_read_ddrb erts_no_atomic64_inc_read +#define erts_atomic64_dec_read_ddrb erts_no_atomic64_dec_read +#define erts_atomic64_inc_ddrb erts_no_atomic64_inc +#define erts_atomic64_dec_ddrb erts_no_atomic64_dec +#define erts_atomic64_add_read_ddrb erts_no_atomic64_add_read +#define erts_atomic64_add_ddrb erts_no_atomic64_add +#define erts_atomic64_read_bor_ddrb erts_no_atomic64_read_bor +#define erts_atomic64_read_band_ddrb erts_no_atomic64_read_band +#define erts_atomic64_xchg_ddrb erts_no_atomic64_xchg +#define erts_atomic64_cmpxchg_ddrb erts_no_atomic64_cmpxchg +#define erts_atomic64_read_bset_ddrb erts_no_atomic64_read_bset + +#define erts_atomic64_init_rb erts_no_atomic64_set +#define erts_atomic64_set_rb erts_no_atomic64_set +#define erts_atomic64_read_rb erts_no_atomic64_read +#define erts_atomic64_inc_read_rb erts_no_atomic64_inc_read +#define erts_atomic64_dec_read_rb erts_no_atomic64_dec_read +#define erts_atomic64_inc_rb erts_no_atomic64_inc +#define erts_atomic64_dec_rb erts_no_atomic64_dec +#define erts_atomic64_add_read_rb erts_no_atomic64_add_read +#define erts_atomic64_add_rb erts_no_atomic64_add +#define erts_atomic64_read_bor_rb erts_no_atomic64_read_bor +#define erts_atomic64_read_band_rb erts_no_atomic64_read_band +#define erts_atomic64_xchg_rb erts_no_atomic64_xchg +#define erts_atomic64_cmpxchg_rb erts_no_atomic64_cmpxchg +#define erts_atomic64_read_bset_rb erts_no_atomic64_read_bset + +#define erts_atomic64_init_wb erts_no_atomic64_set +#define erts_atomic64_set_wb erts_no_atomic64_set +#define erts_atomic64_read_wb erts_no_atomic64_read +#define erts_atomic64_inc_read_wb erts_no_atomic64_inc_read +#define erts_atomic64_dec_read_wb erts_no_atomic64_dec_read +#define erts_atomic64_inc_wb erts_no_atomic64_inc +#define erts_atomic64_dec_wb erts_no_atomic64_dec +#define erts_atomic64_add_read_wb erts_no_atomic64_add_read +#define erts_atomic64_add_wb erts_no_atomic64_add +#define erts_atomic64_read_bor_wb erts_no_atomic64_read_bor +#define erts_atomic64_read_band_wb erts_no_atomic64_read_band +#define erts_atomic64_xchg_wb erts_no_atomic64_xchg +#define erts_atomic64_cmpxchg_wb erts_no_atomic64_cmpxchg +#define erts_atomic64_read_bset_wb erts_no_atomic64_read_bset + +#define erts_atomic64_set_dirty erts_no_atomic64_set +#define erts_atomic64_read_dirty erts_no_atomic64_read + #endif /* !USE_THREADS */ #if ERTS_GLB_INLINE_INCL_FUNC_DEF @@ -2383,6 +2964,104 @@ erts_no_atomic32_read_bset(erts_no_atomic32_t *var, return old; } +/* atomic64 */ + +ERTS_GLB_INLINE void +erts_no_atomic64_set(erts_no_atomic64_t *var, erts_aint64_t i) +{ + *var = i; +} + +ERTS_GLB_INLINE erts_aint64_t +erts_no_atomic64_read(erts_no_atomic64_t *var) +{ + return *var; +} + +ERTS_GLB_INLINE erts_aint64_t +erts_no_atomic64_inc_read(erts_no_atomic64_t *incp) +{ + return ++(*incp); +} + +ERTS_GLB_INLINE erts_aint64_t +erts_no_atomic64_dec_read(erts_no_atomic64_t *decp) +{ + return --(*decp); +} + +ERTS_GLB_INLINE void +erts_no_atomic64_inc(erts_no_atomic64_t *incp) +{ + ++(*incp); +} + +ERTS_GLB_INLINE void +erts_no_atomic64_dec(erts_no_atomic64_t *decp) +{ + --(*decp); +} + +ERTS_GLB_INLINE erts_aint64_t +erts_no_atomic64_add_read(erts_no_atomic64_t *addp, erts_aint64_t i) +{ + return *addp += i; +} + +ERTS_GLB_INLINE void +erts_no_atomic64_add(erts_no_atomic64_t *addp, erts_aint64_t i) +{ + *addp += i; +} + +ERTS_GLB_INLINE erts_aint64_t +erts_no_atomic64_read_bor(erts_no_atomic64_t *var, erts_aint64_t mask) +{ + erts_aint64_t old; + old = *var; + *var |= mask; + return old; +} + +ERTS_GLB_INLINE erts_aint64_t +erts_no_atomic64_read_band(erts_no_atomic64_t *var, erts_aint64_t mask) +{ + erts_aint64_t old; + old = *var; + *var &= mask; + return old; +} + +ERTS_GLB_INLINE erts_aint64_t +erts_no_atomic64_xchg(erts_no_atomic64_t *xchgp, erts_aint64_t new) +{ + erts_aint64_t old = *xchgp; + *xchgp = new; + return old; +} + +ERTS_GLB_INLINE erts_aint64_t +erts_no_atomic64_cmpxchg(erts_no_atomic64_t *xchgp, + erts_aint64_t new, + erts_aint64_t expected) +{ + erts_aint64_t old = *xchgp; + if (old == expected) + *xchgp = new; + return old; +} + +ERTS_GLB_INLINE erts_aint64_t +erts_no_atomic64_read_bset(erts_no_atomic64_t *var, + erts_aint64_t mask, + erts_aint64_t set) +{ + erts_aint64_t old = *var; + *var &= ~mask; + *var |= (mask & set); + return old; +} + /* spinlock */ ERTS_GLB_INLINE void diff --git a/erts/emulator/beam/erl_utils.h b/erts/emulator/beam/erl_utils.h index 0807649ea1..c32f8fd61c 100644 --- a/erts/emulator/beam/erl_utils.h +++ b/erts/emulator/beam/erl_utils.h @@ -32,11 +32,7 @@ typedef struct { #endif union { Uint64 not_atomic; -#ifdef ARCH_64 - erts_atomic_t atomic; -#else - erts_dw_atomic_t atomic; -#endif + erts_atomic64_t atomic; } counter; } erts_interval_t; @@ -50,9 +46,6 @@ Uint64 erts_ensure_later_interval_nob(erts_interval_t *, Uint64); Uint64 erts_ensure_later_interval_acqb(erts_interval_t *, Uint64); Uint64 erts_smp_ensure_later_interval_nob(erts_interval_t *, Uint64); Uint64 erts_smp_ensure_later_interval_acqb(erts_interval_t *, Uint64); -#ifdef ARCH_32 -ERTS_GLB_INLINE Uint64 erts_interval_dw_aint_to_val__(erts_dw_aint_t *); -#endif ERTS_GLB_INLINE Uint64 erts_current_interval_nob__(erts_interval_t *); ERTS_GLB_INLINE Uint64 erts_current_interval_acqb__(erts_interval_t *); ERTS_GLB_INLINE Uint64 erts_current_interval_nob(erts_interval_t *); @@ -62,46 +55,16 @@ ERTS_GLB_INLINE Uint64 erts_smp_current_interval_acqb(erts_interval_t *); #if ERTS_GLB_INLINE_INCL_FUNC_DEF -#ifdef ARCH_32 - -ERTS_GLB_INLINE Uint64 -erts_interval_dw_aint_to_val__(erts_dw_aint_t *dw) -{ -#ifdef ETHR_SU_DW_NAINT_T__ - return (Uint64) dw->dw_sint; -#else - Uint64 res; - res = (Uint64) ((Uint32) dw->sint[ERTS_DW_AINT_HIGH_WORD]); - res <<= 32; - res |= (Uint64) ((Uint32) dw->sint[ERTS_DW_AINT_LOW_WORD]); - return res; -#endif -} - -#endif - ERTS_GLB_INLINE Uint64 erts_current_interval_nob__(erts_interval_t *icp) { -#ifdef ARCH_64 - return (Uint64) erts_atomic_read_nob(&icp->counter.atomic); -#else - erts_dw_aint_t dw; - erts_dw_atomic_read_nob(&icp->counter.atomic, &dw); - return erts_interval_dw_aint_to_val__(&dw); -#endif + return (Uint64) erts_atomic64_read_nob(&icp->counter.atomic); } ERTS_GLB_INLINE Uint64 erts_current_interval_acqb__(erts_interval_t *icp) { -#ifdef ARCH_64 - return (Uint64) erts_atomic_read_acqb(&icp->counter.atomic); -#else - erts_dw_aint_t dw; - erts_dw_atomic_read_acqb(&icp->counter.atomic, &dw); - return erts_interval_dw_aint_to_val__(&dw); -#endif + return (Uint64) erts_atomic64_read_acqb(&icp->counter.atomic); } ERTS_GLB_INLINE Uint64 diff --git a/erts/emulator/beam/external.c b/erts/emulator/beam/external.c index 8d240355b0..9b9b4b2a62 100644 --- a/erts/emulator/beam/external.c +++ b/erts/emulator/beam/external.c @@ -1925,6 +1925,7 @@ 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 */ @@ -2004,6 +2005,7 @@ 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; diff --git a/erts/emulator/beam/io.c b/erts/emulator/beam/io.c index 0f86d8e41d..9ae973e108 100644 --- a/erts/emulator/beam/io.c +++ b/erts/emulator/beam/io.c @@ -1218,9 +1218,10 @@ typedef struct { static ERTS_INLINE ErtsTryImmDrvCallResult try_imm_drv_call(ErtsTryImmDrvCallState *sp) { + unsigned int prof_runnable_ports; ErtsTryImmDrvCallResult res; int reds_left_in; - erts_aint32_t invalid_state, invalid_sched_flags; + erts_aint32_t act, exp, invalid_state, invalid_sched_flags; Port *prt = sp->port; Process *c_p = sp->c_p; @@ -1247,18 +1248,39 @@ try_imm_drv_call(ErtsTryImmDrvCallState *sp) goto locked_fail; } - sp->sched_flags = erts_smp_atomic32_read_nob(&prt->sched.flags); - if (sp->sched_flags & invalid_sched_flags) { - res = ERTS_TRY_IMM_DRV_CALL_INVALID_SCHED_FLAGS; - goto locked_fail; - } + prof_runnable_ports = erts_system_profile_flags.runnable_ports; + if (prof_runnable_ports) + erts_port_task_sched_lock(&prt->sched); + + act = erts_smp_atomic32_read_nob(&prt->sched.flags); + do { + erts_aint32_t new; + + if (act & invalid_sched_flags) { + res = ERTS_TRY_IMM_DRV_CALL_INVALID_SCHED_FLAGS; + sp->sched_flags = act; + goto locked_fail; + } + exp = act; + new = act | ERTS_PTS_FLG_EXEC_IMM; + act = erts_smp_atomic32_cmpxchg_mb(&prt->sched.flags, new, exp); + } while (act != exp); + + sp->sched_flags = act; if (!c_p) reds_left_in = CONTEXT_REDS/10; else { if (IS_TRACED_FL(c_p, F_TRACE_SCHED_PROCS)) trace_virtual_sched(c_p, am_out); + /* + * No status lock held while sending runnable + * proc trace messages. It is however not needed + * in this case, since only this thread can send + * such messages for this process until the process + * has been scheduled out. + */ if (erts_system_profile_flags.runnable_procs && erts_system_profile_flags.exclusive) profile_runnable_proc(c_p, am_inactive); @@ -1273,11 +1295,14 @@ try_imm_drv_call(ErtsTryImmDrvCallState *sp) ERTS_SMP_CHK_NO_PROC_LOCKS; - if (IS_TRACED_FL(prt, F_TRACE_SCHED_PORTS)) - trace_sched_ports_where(prt, am_in, sp->port_op); - if (erts_system_profile_flags.runnable_ports - && !erts_port_is_scheduled(prt)) - profile_runnable_port(prt, am_active); + if (prof_runnable_ports | IS_TRACED_FL(prt, F_TRACE_SCHED_PORTS)) { + if (prof_runnable_ports && !(act & (ERTS_PTS_FLG_IN_RUNQ|ERTS_PTS_FLG_EXEC))) + profile_runnable_port(prt, am_active); + if (IS_TRACED_FL(prt, F_TRACE_SCHED_PORTS)) + trace_sched_ports_where(prt, am_in, sp->port_op); + if (prof_runnable_ports) + erts_port_task_sched_unlock(&prt->sched); + } sp->fpe_was_unmasked = erts_block_fpe(); @@ -1294,17 +1319,31 @@ finalize_imm_drv_call(ErtsTryImmDrvCallState *sp) int reds; Port *prt = sp->port; Process *c_p = sp->c_p; + erts_aint32_t act; + unsigned int prof_runnable_ports; reds = prt->reds; reds += erts_port_driver_callback_epilogue(prt, NULL); erts_unblock_fpe(sp->fpe_was_unmasked); - if (IS_TRACED_FL(prt, F_TRACE_SCHED_PORTS)) - trace_sched_ports_where(prt, am_out, sp->port_op); - if (erts_system_profile_flags.runnable_ports - && !erts_port_is_scheduled(prt)) - profile_runnable_port(prt, am_inactive); + prof_runnable_ports = erts_system_profile_flags.runnable_ports; + if (prof_runnable_ports) + erts_port_task_sched_lock(&prt->sched); + + act = erts_smp_atomic32_read_band_mb(&prt->sched.flags, + ~ERTS_PTS_FLG_EXEC_IMM); + ERTS_SMP_LC_ASSERT(act & ERTS_PTS_FLG_EXEC_IMM); + + if (prof_runnable_ports | IS_TRACED_FL(prt, F_TRACE_SCHED_PORTS)) { + if (IS_TRACED_FL(prt, F_TRACE_SCHED_PORTS)) + trace_sched_ports_where(prt, am_out, sp->port_op); + if (prof_runnable_ports) { + if (!(act & (ERTS_PTS_FLG_IN_RUNQ|ERTS_PTS_FLG_EXEC))) + profile_runnable_port(prt, am_inactive); + erts_port_task_sched_unlock(&prt->sched); + } + } erts_port_release(prt); @@ -1319,6 +1358,13 @@ finalize_imm_drv_call(ErtsTryImmDrvCallState *sp) if (IS_TRACED_FL(c_p, F_TRACE_SCHED_PROCS)) trace_virtual_sched(c_p, am_in); + /* + * No status lock held while sending runnable + * proc trace messages. It is however not needed + * in this case, since only this thread can send + * such messages for this process until the process + * has been scheduled out. + */ if (erts_system_profile_flags.runnable_procs && erts_system_profile_flags.exclusive) profile_runnable_proc(c_p, am_active); @@ -7228,6 +7274,18 @@ driver_system_info(ErlDrvSysInfo *sip, size_t si_size) sip->nif_major_version = ERL_NIF_MAJOR_VERSION; sip->nif_minor_version = ERL_NIF_MINOR_VERSION; } + /* + * 'dirty_scheduler_support' is the last field in the 4th version + * (driver version 3.1, NIF version 2.7) + */ + if (si_size >= ERL_DRV_SYS_INFO_SIZE(dirty_scheduler_support)) { +#if defined(ERL_NIF_DIRTY_SCHEDULER_SUPPORT) && defined(USE_THREADS) + sip->dirty_scheduler_support = 1; +#else + sip->dirty_scheduler_support = 0; +#endif + } + } diff --git a/erts/emulator/beam/sys.h b/erts/emulator/beam/sys.h index 3d8dd9c6d0..c29d4b3777 100644 --- a/erts/emulator/beam/sys.h +++ b/erts/emulator/beam/sys.h @@ -66,8 +66,12 @@ */ #ifndef ERTS_SYS_FD_TYPE +#define ERTS_SYS_FD_INVALID ((ErtsSysFdType) -1) typedef int ErtsSysFdType; #else +#ifndef ERTS_SYS_FD_INVALID +# error missing ERTS_SYS_FD_INVALID +#endif typedef ERTS_SYS_FD_TYPE ErtsSysFdType; #endif @@ -501,7 +505,7 @@ extern volatile int erts_writing_erl_crash_dump; # define NO_ERF # define NO_ERFC /* This definition doesn't take NaN into account, but matherr() gets those */ -# define finite(x) (fabs(x) != HUGE_VAL) +# define isfinite(x) (fabs(x) != HUGE_VAL) # define USE_MATHERR # define HAVE_FINITE #endif @@ -744,6 +748,14 @@ void init_getenv_state(GETENV_STATE *); char * getenv_string(GETENV_STATE *); void fini_getenv_state(GETENV_STATE *); +#define HAVE_ERTS_CHECK_IO_DEBUG +typedef struct { + int no_used_fds; + int no_driver_select_structs; + int no_driver_event_structs; +} ErtsCheckIoDebugInfo; +int erts_check_io_debug(ErtsCheckIoDebugInfo *ip); + /* xxxP */ #define SYS_DEFAULT_FLOAT_DECIMALS 20 void init_sys_float(void); diff --git a/erts/emulator/beam/utils.c b/erts/emulator/beam/utils.c index 55f9e68e78..c505c44905 100644 --- a/erts/emulator/beam/utils.c +++ b/erts/emulator/beam/utils.c @@ -48,6 +48,7 @@ #include "erl_sched_spec_pre_alloc.h" #include "beam_bp.h" #include "erl_ptab.h" +#include "erl_check_io.h" #undef M_TRIM_THRESHOLD #undef M_TOP_PAD @@ -4215,19 +4216,7 @@ void erts_silence_warn_unused_result(long unused) void erts_interval_init(erts_interval_t *icp) { -#ifdef ARCH_64 - erts_atomic_init_nob(&icp->counter.atomic, 0); -#else - erts_dw_aint_t dw; -#ifdef ETHR_SU_DW_NAINT_T__ - dw.dw_sint = 0; -#else - dw.sint[ERTS_DW_AINT_HIGH_WORD] = 0; - dw.sint[ERTS_DW_AINT_LOW_WORD] = 0; -#endif - erts_dw_atomic_init_nob(&icp->counter.atomic, &dw); - -#endif + erts_atomic64_init_nob(&icp->counter.atomic, 0); #ifdef DEBUG icp->smp_api = 0; #endif @@ -4249,55 +4238,13 @@ erts_smp_interval_init(erts_interval_t *icp) static ERTS_INLINE Uint64 step_interval_nob(erts_interval_t *icp) { -#ifdef ARCH_64 - return (Uint64) erts_atomic_inc_read_nob(&icp->counter.atomic); -#else - erts_dw_aint_t exp; - - erts_dw_atomic_read_nob(&icp->counter.atomic, &exp); - while (1) { - erts_dw_aint_t new = exp; - -#ifdef ETHR_SU_DW_NAINT_T__ - new.dw_sint++; -#else - new.sint[ERTS_DW_AINT_LOW_WORD]++; - if (new.sint[ERTS_DW_AINT_LOW_WORD] == 0) - new.sint[ERTS_DW_AINT_HIGH_WORD]++; -#endif - - if (erts_dw_atomic_cmpxchg_nob(&icp->counter.atomic, &new, &exp)) - return erts_interval_dw_aint_to_val__(&new); - - } -#endif + return (Uint64) erts_atomic64_inc_read_nob(&icp->counter.atomic); } static ERTS_INLINE Uint64 step_interval_relb(erts_interval_t *icp) { -#ifdef ARCH_64 - return (Uint64) erts_atomic_inc_read_relb(&icp->counter.atomic); -#else - erts_dw_aint_t exp; - - erts_dw_atomic_read_nob(&icp->counter.atomic, &exp); - while (1) { - erts_dw_aint_t new = exp; - -#ifdef ETHR_SU_DW_NAINT_T__ - new.dw_sint++; -#else - new.sint[ERTS_DW_AINT_LOW_WORD]++; - if (new.sint[ERTS_DW_AINT_LOW_WORD] == 0) - new.sint[ERTS_DW_AINT_HIGH_WORD]++; -#endif - - if (erts_dw_atomic_cmpxchg_relb(&icp->counter.atomic, &new, &exp)) - return erts_interval_dw_aint_to_val__(&new); - - } -#endif + return (Uint64) erts_atomic64_inc_read_relb(&icp->counter.atomic); } @@ -4305,38 +4252,10 @@ static ERTS_INLINE Uint64 ensure_later_interval_nob(erts_interval_t *icp, Uint64 ic) { Uint64 curr_ic; -#ifdef ARCH_64 - curr_ic = (Uint64) erts_atomic_read_nob(&icp->counter.atomic); + curr_ic = (Uint64) erts_atomic64_read_nob(&icp->counter.atomic); if (curr_ic > ic) return curr_ic; - return (Uint64) erts_atomic_inc_read_nob(&icp->counter.atomic); -#else - erts_dw_aint_t exp; - - erts_dw_atomic_read_nob(&icp->counter.atomic, &exp); - curr_ic = erts_interval_dw_aint_to_val__(&exp); - if (curr_ic > ic) - return curr_ic; - - while (1) { - erts_dw_aint_t new = exp; - -#ifdef ETHR_SU_DW_NAINT_T__ - new.dw_sint++; -#else - new.sint[ERTS_DW_AINT_LOW_WORD]++; - if (new.sint[ERTS_DW_AINT_LOW_WORD] == 0) - new.sint[ERTS_DW_AINT_HIGH_WORD]++; -#endif - - if (erts_dw_atomic_cmpxchg_nob(&icp->counter.atomic, &new, &exp)) - return erts_interval_dw_aint_to_val__(&new); - - curr_ic = erts_interval_dw_aint_to_val__(&exp); - if (curr_ic > ic) - return curr_ic; - } -#endif + return (Uint64) erts_atomic64_inc_read_nob(&icp->counter.atomic); } @@ -4344,38 +4263,10 @@ static ERTS_INLINE Uint64 ensure_later_interval_acqb(erts_interval_t *icp, Uint64 ic) { Uint64 curr_ic; -#ifdef ARCH_64 - curr_ic = (Uint64) erts_atomic_read_acqb(&icp->counter.atomic); + curr_ic = (Uint64) erts_atomic64_read_acqb(&icp->counter.atomic); if (curr_ic > ic) return curr_ic; - return (Uint64) erts_atomic_inc_read_acqb(&icp->counter.atomic); -#else - erts_dw_aint_t exp; - - erts_dw_atomic_read_acqb(&icp->counter.atomic, &exp); - curr_ic = erts_interval_dw_aint_to_val__(&exp); - if (curr_ic > ic) - return curr_ic; - - while (1) { - erts_dw_aint_t new = exp; - -#ifdef ETHR_SU_DW_NAINT_T__ - new.dw_sint++; -#else - new.sint[ERTS_DW_AINT_LOW_WORD]++; - if (new.sint[ERTS_DW_AINT_LOW_WORD] == 0) - new.sint[ERTS_DW_AINT_HIGH_WORD]++; -#endif - - if (erts_dw_atomic_cmpxchg_acqb(&icp->counter.atomic, &new, &exp)) - return erts_interval_dw_aint_to_val__(&new); - - curr_ic = erts_interval_dw_aint_to_val__(&exp); - if (curr_ic > ic) - return curr_ic; - } -#endif + return (Uint64) erts_atomic64_inc_read_acqb(&icp->counter.atomic); } Uint64 diff --git a/erts/emulator/drivers/common/inet_drv.c b/erts/emulator/drivers/common/inet_drv.c index 09d90f4984..3fe5dac813 100644 --- a/erts/emulator/drivers/common/inet_drv.c +++ b/erts/emulator/drivers/common/inet_drv.c @@ -2890,6 +2890,9 @@ 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 +3922,10 @@ 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); @@ -4542,11 +4548,13 @@ static ErlDrvSSizeT inet_ctl_fdopen(inet_descriptor* desc, int domain, int type, inet_address name; unsigned int sz = sizeof(name); - /* check that it is a socket and that the socket is bound */ - if (IS_SOCKET_ERROR(sock_name(s, (struct sockaddr*) &name, &sz))) - return ctl_error(sock_errno(), rbuf, rsize); - if (name.sa.sa_family != domain) - return ctl_error(EINVAL, rbuf, rsize); + if (bound) { + /* check that it is a socket and that the socket is bound */ + if (IS_SOCKET_ERROR(sock_name(s, (struct sockaddr*) &name, &sz))) + return ctl_error(sock_errno(), rbuf, rsize); + if (name.sa.sa_family != domain) + return ctl_error(EINVAL, rbuf, rsize); + } #ifdef __OSE__ /* for fdopen duplicating the sd will allow to uniquely identify the signal from OSE with erlang port */ @@ -5778,7 +5786,7 @@ done: ia_p->Ipv6IfIndex && ia_p->Ipv6IfIndex != index) { - /* Oops, there was an other interface for IPv6. Possible? XXX */ + /* Oops, there was another interface for IPv6. Possible? XXX */ index = ia_p->Ipv6IfIndex; goto index; } @@ -7336,8 +7344,13 @@ 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); diff --git a/erts/emulator/drivers/unix/multi_drv.c b/erts/emulator/drivers/unix/multi_drv.c index 822c96730c..724d325ed5 100644 --- a/erts/emulator/drivers/unix/multi_drv.c +++ b/erts/emulator/drivers/unix/multi_drv.c @@ -20,7 +20,7 @@ /* Purpose: Multidriver interface This is an example of a driver which allows multiple instances of itself. I.e have one erlang process execute open_port(multi......) and - at the same time have an other erlang process open an other port + at the same time have another erlang process open another port running multi there as well. */ diff --git a/erts/emulator/drivers/win32/win_efile.c b/erts/emulator/drivers/win32/win_efile.c index a321bb9641..7e4043fc1b 100644 --- a/erts/emulator/drivers/win32/win_efile.c +++ b/erts/emulator/drivers/win32/win_efile.c @@ -1288,6 +1288,10 @@ do_fileinfo(Efile_call_state* state, Efile_info* pInfo, { HANDLE handle; /* Handle returned by CreateFile() */ BY_HANDLE_FILE_INFORMATION fileInfo; /* from CreateFile() */ + + /* We initialise nNumberOfLinks as GetFileInformationByHandle + does not always initialise this field */ + fileInfo.nNumberOfLinks = 1; if (handle = CreateFileW(name, GENERIC_READ, FILE_SHARE_FLAGS, NULL, OPEN_EXISTING, 0, NULL)) { GetFileInformationByHandle(handle, &fileInfo); diff --git a/erts/emulator/hipe/hipe_amd64_bifs.m4 b/erts/emulator/hipe/hipe_amd64_bifs.m4 index 0de69a617f..a3219c7586 100644 --- a/erts/emulator/hipe/hipe_amd64_bifs.m4 +++ b/erts/emulator/hipe/hipe_amd64_bifs.m4 @@ -39,7 +39,10 @@ define(HANDLE_GOT_MBUF,` jmp 2b') `#if defined(ERTS_ENABLE_LOCK_CHECK) && defined(ERTS_SMP) -# define CALL_BIF(F) movq $CSYM(F), P_BIF_CALLEE(P); call CSYM(hipe_debug_bif_wrapper) +# define CALL_BIF(F) \ + movq CSYM(F)@GOTPCREL(%rip), %r11; \ + movq %r11, P_BIF_CALLEE(P); \ + call CSYM(hipe_debug_bif_wrapper) #else # define CALL_BIF(F) call CSYM(F) #endif' diff --git a/erts/emulator/hipe/hipe_arm_bifs.m4 b/erts/emulator/hipe/hipe_arm_bifs.m4 index bd8bc5ab6b..3ca9a1bcdb 100644 --- a/erts/emulator/hipe/hipe_arm_bifs.m4 +++ b/erts/emulator/hipe/hipe_arm_bifs.m4 @@ -25,6 +25,7 @@ include(`hipe/hipe_arm_asm.m4') .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 +392,14 @@ $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 +415,14 @@ $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 +439,14 @@ $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 +464,14 @@ $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 +495,14 @@ $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 +524,16 @@ 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 +548,14 @@ $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 +571,14 @@ $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 +595,14 @@ $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 +622,14 @@ $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..5723afac12 100644 --- a/erts/emulator/hipe/hipe_arm_glue.S +++ b/erts/emulator/hipe/hipe_arm_glue.S @@ -25,6 +25,7 @@ .text .p2align 2 + .arm /* * Enter Erlang from C. @@ -70,6 +71,7 @@ * 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 +87,7 @@ 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 +98,7 @@ 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 +115,7 @@ 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 +130,7 @@ 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 */ @@ -142,6 +148,7 @@ hipe_arm_throw_to_native: * 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 r0, [P, #P_ARITY] @@ -153,6 +160,7 @@ nbif_callemu: * nbif_apply */ .global nbif_apply + .type nbif_apply, %function nbif_apply: STORE_ARG_REGS mov r0, #HIPE_MODE_SWITCH_RES_APPLY @@ -169,6 +177,7 @@ 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 +190,7 @@ 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 +203,7 @@ 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 +216,7 @@ 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 +229,7 @@ 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 +242,7 @@ 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 +254,7 @@ 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 +269,7 @@ 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 +278,7 @@ 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 +289,7 @@ 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 +304,31 @@ 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 +356,25 @@ 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 +418,7 @@ 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 +435,7 @@ 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 2497d51df1..c9eee2acf2 100644 --- a/erts/emulator/hipe/hipe_bif0.c +++ b/erts/emulator/hipe/hipe_bif0.c @@ -1022,7 +1022,7 @@ BIF_RETTYPE hipe_conv_big_to_float(BIF_ALIST_1) */ void hipe_emulate_fpe(Process* p) { - if (!finite(p->hipe.float_result)) { + if (!isfinite(p->hipe.float_result)) { p->fp_exception = 1; } } diff --git a/erts/emulator/hipe/hipe_mode_switch.c b/erts/emulator/hipe/hipe_mode_switch.c index 4ddc2790b1..1ae1d17b7f 100644 --- a/erts/emulator/hipe/hipe_mode_switch.c +++ b/erts/emulator/hipe/hipe_mode_switch.c @@ -2,7 +2,7 @@ * %CopyrightBegin% * - * Copyright Ericsson AB 2001-2013. All Rights Reserved. + * Copyright Ericsson AB 2001-2014. 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 @@ -187,6 +187,9 @@ void hipe_set_call_trap(Uint *bfun, void *nfun, int is_closure) 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"); @@ -195,25 +198,26 @@ void hipe_reserve_beam_trap_frame(Process *p, Eterm reg[], unsigned arity) } p->stop -= 2; p->stop[0] = NIL; - p->stop[1] = NIL; + p->stop[1] = hipe_beam_catch_throw; } static __inline__ void hipe_push_beam_trap_frame(Process *p, Eterm reg[], unsigned arity) { - if (p->flags & F_DISABLE_GC) { + if (&p->stop[1] < p->hend && p->stop[1] == hipe_beam_catch_throw) { /* Trap frame already reserved */ - ASSERT(p->stop[0] == NIL && p->stop[1] == NIL); + ASSERT(p->stop[0] == NIL); } else { + ASSERT(!(p->flags & F_DISABLE_GC)); if ((p->stop - 2) < p->htop) { DPRINTF("calling gc to increase BEAM stack size"); p->fcalls -= erts_garbage_collect(p, 2, reg, arity); ASSERT(!((p->stop - 2) < p->htop)); } p->stop -= 2; + p->stop[1] = hipe_beam_catch_throw; } - p->stop[1] = hipe_beam_catch_throw; p->stop[0] = make_cp(p->cp); ++p->catches; p->cp = hipe_beam_pc_return; @@ -221,12 +225,16 @@ hipe_push_beam_trap_frame(Process *p, Eterm reg[], unsigned arity) void hipe_unreserve_beam_trap_frame(Process *p) { - ASSERT(p->stop[0] == NIL && p->stop[1] == NIL); + 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); p->cp = cp_val(p->stop[0]); --p->catches; p->stop += 2; diff --git a/erts/emulator/hipe/hipe_risc_glue.h b/erts/emulator/hipe/hipe_risc_glue.h index cc2671c016..dbb7086dae 100644 --- a/erts/emulator/hipe/hipe_risc_glue.h +++ b/erts/emulator/hipe/hipe_risc_glue.h @@ -214,6 +214,14 @@ hipe_trap_from_native_is_recursive(Process *p) return 0; } +/* Native called BIF. Is it a recursive call? + i.e should we return back to native when BIF is done? */ +static __inline__ int +hipe_bifcall_from_native_is_recursive(Process *p) +{ + return (p->hipe.nra != (void(*)(void))&nbif_return); +} + /* Native makes a call which needs to unload the parameters. This differs from hipe_call_from_native_is_recursive() in diff --git a/erts/emulator/hipe/hipe_x86_glue.h b/erts/emulator/hipe/hipe_x86_glue.h index 63ad250d60..4b6e495b9a 100644 --- a/erts/emulator/hipe/hipe_x86_glue.h +++ b/erts/emulator/hipe/hipe_x86_glue.h @@ -207,6 +207,14 @@ hipe_trap_from_native_is_recursive(Process *p) return 0; } +/* Native called BIF. Is it a recursive call? + i.e should we return back to native when BIF is done? */ +static __inline__ int +hipe_bifcall_from_native_is_recursive(Process *p) +{ + return (*p->hipe.nsp != (Eterm)nbif_return); +} + /* Native makes a call which needs to unload the parameters. This differs from hipe_call_from_native_is_recursive() in diff --git a/erts/emulator/internal_doc/CarrierMigration.md b/erts/emulator/internal_doc/CarrierMigration.md index b93c11c6ec..2a9594db25 100644 --- a/erts/emulator/internal_doc/CarrierMigration.md +++ b/erts/emulator/internal_doc/CarrierMigration.md @@ -16,12 +16,12 @@ When a carrier is empty, i.e. contains only one large free block, it is deallocated. Since multiblock carriers can contain both allocated blocks and free blocks at the same time, an allocator instance might be stuck with a large amount of poorly utilized carriers if the memory -load decrease. After a peak in memory usage it is expected that not -all memory can be returned since the blocks still allocated is likely +load decreases. After a peak in memory usage it is expected that not +all memory can be returned since the blocks still allocated are likely to be dispersed over multiple carriers. Such poorly utilized carriers -can usually be reused if the memory load increase again. However, +can usually be reused if the memory load increases again. However, since each scheduler thread manages its own set of allocator -instances, and memory load is not necessarily connected to CPU load we +instances, and memory load is not necessarily correlated to CPU load, we might get into a situation where there are lots of poorly utilized multiblock carriers on some allocator instances while we need to allocate new multiblock carriers on other allocator instances. In @@ -50,13 +50,13 @@ the allocator instance manages. Free blocks in one specific carrier can be referred to from potentially every other carrier that is managed, and the amount of such references can be huge. That is, the work of removing the free blocks of such a carrier from the search -tree will be huge. One way of solving this could be to not migrate +tree will be huge. One way of solving this could be not to migrate carriers that contain lots of free blocks, but this would prevent us -from migrating carriers that potentially needs to be migrated in order +from migrating carriers that potentially need to be migrated in order to solve the problem we set out to solve. By using one data structure of free blocks in each carrier and an -allocator instance wide data structure of carriers managed by the +allocator instance-wide data structure of carriers managed by the allocator instance, the work needed in order to remove and add carriers can be kept to a minimum. When migration of carriers is enabled on a specific allocator type, we require that an allocation @@ -76,9 +76,9 @@ through a pool of carriers. In order for a carrier migration to complete, one scheduler needs to move the carrier into the pool, and another scheduler needs to take the carrier out of the pool. -The pool is implemented as a lock free, circular, double linked, +The pool is implemented as a lock-free, circular, double linked, list. The list contains a sentinel which is used as the starting point -when inserting to, or fetching from the pool. Carriers in the pool are +when inserting to, or fetching from, the pool. Carriers in the pool are elements in this list. The list can be modified by all scheduler threads @@ -108,19 +108,19 @@ all search operations need to read the content of the sentinel. If we were to modify the sentinel, the cache line containing the sentinel would unnecessarily be bounced between processors. -The `prev`, and `next` fields in the elements of the list contains the +The `prev` and `next` fields in the elements of the list contain the value of the pointer, a modification marker, and a deleted marker. Memory operations on these fields are done using atomic memory operations. When a thread has set the modification marker in a field, no-one except the thread that set the marker is allowed to modify the -field. If multiple modification markers needs to be set, we always +field. If multiple modification markers need to be set, we always begin with `next` fields followed by `prev` fields in the order following the actual pointers. This guarantees that no deadlocks will occur. When a carrier is being removed from a pool, we mark it with a thread progress value that needs to be reached before we are allowed to -modify the `next`, and `prev` fields. That is, until we reach this +modify the `next` and `prev` fields. That is, until we reach this thread progress we are not allowed to insert the carrier into the pool again, and we are not allowed to deallocate the carrier. This ensures that threads inspecting the pool always will be able to traverse the @@ -130,12 +130,12 @@ threads may have references to it via the pool. ### Migration ### -There exist one pool for each allocator type enabling migration of +There exists one pool for each allocator type enabling migration of carriers between scheduler specific allocator instances of the same allocator type. Each allocator instance keeps track of the current utilization of its -multiblock carriers. When the utilization falls below the "abandon +multiblock carriers. When the total utilization falls below the "abandon carrier utilization limit" it starts to inspect the utilization of the current carrier when deallocations are made. If also the utilization of the carrier falls below the "abandon carrier utilization limit" it @@ -146,28 +146,53 @@ Since the carrier has been unlinked from the data structure of available free blocks, no more allocations will be made in the carrier. The allocator instance putting the carrier into the pool, however, still has the responsibility of performing deallocations in -it while it remains in the pool. +it while it remains in the pool. The allocator instance with this +deallocation responsibility is here called the **employer**. -Each carrier has a flag field containing information about allocator -instance owning the carrier, a flag indicating if the carrier is in +Each carrier has a flag field containing information about the +employing allocator instance, a flag indicating if the carrier is in the pool or not, and a flag indicating if it is busy or not. When the -carrier is in the pool, the owning allocator instance needs to mark it +carrier is in the pool, the employing allocator instance needs to mark it as busy while operating on it. If another thread inspects it in order -to try to fetch it from the pool, it will abort the fetch if it is -busy. When fetching the carrier from the pool, ownership will changed -and further deallocations in the carrier will be redirected to the new -owner using the delayed dealloc functionality. +to try to fetch it from the pool, it will skip it if it is busy. When +fetching the carrier from the pool, employment will change and further +deallocations in the carrier will be redirected to the new +employer using the delayed dealloc functionality. If a carrier in the pool becomes empty, it will be withdrawn from the pool. All carriers that become empty are also always passed to its -originating allocator instance for deallocation using the delayed +**owning** allocator instance for deallocation using the delayed dealloc functionality. Since carriers this way always will be -deallocated by the allocator instance that allocated the carrier the +deallocated by the owner that allocated the carrier, the underlying functionality of allocating and deallocating carriers can remain simple and doesn't have to bother about multiple threads. In a NUMA system we will also not mix carriers originating from multiple NUMA nodes. +In short: + +* The allocator instance that created a carrier **owns** it. +* An empty carrier is always deallocated by its **owner**. +* **Ownership** never changes. +* The allocator instance that uses a carrier **employs** it. +* An **employer** can abandon a carrier into the pool. +* Pooled carriers are not allocated from. +* Deallocation in a pooled carrier is still performed by its **employer**. +* **Employment** can only change when a carrier is fetched from the pool. + +### Searching the pool ### + +To harbor real time characteristics, searching the pool is +limited. We only inspect a limited number of carriers. If none of +those carriers had a free block large enough to satisfy the allocation +request, the search will fail. A carrier in the pool can also be busy +if another thread is currently doing block deallocation work on the +carrier. A busy carrier will also be skipped by the search as it can +not satisfy the request. The pool is lock-free and we do not want to +block, waiting for the other thread to finish. + +#### Before OTP 17.4 #### + When an allocator instance needs more carrier space, it always begins by inspecting its own carriers that are waiting for thread progress before they can be deallocated. If no such carrier could be found, it @@ -176,10 +201,69 @@ it will allocate a new carrier. Regardless of where the allocator instance gets the carrier from it the just links in the carrier into its data structure of free blocks. +#### After OTP 17.4 #### + +The old search algorithm had a problem as the search always started at +the same position in the pool, the sentinel. This could lead to +contention from concurrent searching processes. But even worse, it +could lead to a "bad" state when searches fail with a high rate +leading to new carriers instead being allocated. These new carriers +may later be inserted into the pool due to bad utilization. If the +frequency of insertions into the pool is higher than successful +fetching from the pool, memory will eventually get exhausted. + +This "bad" state consists of a cluster of small and/or highly +fragmented carriers located at the sentinel in the pool. The largest free +block in such a "bad" carrier is rather small, making it unable to satisfy +most allocation requests. As the search always started at the +sentinel, any such "bad" carriers that had been left in the pool would +eventually cluster together at the sentinel. All searches first +have to skip past this cluster of "bad" carriers to reach a "good" +carrier. When the cluster gets to the same size as the search limit, +all searches will essentially fail. + +To counter the "bad cluster" problem and also ease the contention, the +search will now always start by first looking at the allocators **own** +carriers. That is, carriers that were initially created by the +allocator itself and later had been abandoned to the pool. If none of +our own abandoned carrier would do, then the search continues into the +pool, as before, to look for carriers created by other +allocators. However, if we have at least one abandoned carrier of our +own that could not satisfy the request, we can use that as entry point +into the pool. + +The result is that we prefer carriers created by the thread itself, +which is good for NUMA performance. And we get more entry points when +searching the pool, which will ease contention and clustering. + +To do the first search among own carriers, every allocator instance +has two new lists: `pooled_list` and `traitor_list`. These lists are only +accessed by the allocator itself and they only contain the allocator's +own carriers. When an owned carrier is abandoned and put in the +pool, it is also linked into `pooled_list`. When we search our +`pooled_list` and find a carrier that is no longer in the pool, we +move that carrier from `pooled_list` to `traitor_list` as it is now +employed by another allocator. If searching `pooled_list` fails, we +also do a limited search of `traitor_list`. When finding an abandoned +carrier in `traitor_list` it is either employed or moved back to +`pooled_list` if it could not satisfy the allocation request. + +When searching `pooled_list` and `traitor_list` we always start at the +point where the last search ended. This to avoid clustering +problems and increase the probability to find a "good" carrier. As +`pooled_list` and `traitor_list` are only accessed by the owning +allocator instance, they need no thread synchronization at all. + +Furthermore, the search for own carriers that are scheduled +for deallocation is now done as the last search option. The idea is +that it is better to reuse a poorly utilized carrier than to +resurrect an empty carrier that was just about to be released back to +the OS. + ### Result ### The use of this strategy of abandoning carriers with poor utilization -and reusing these in allocator instances with an increased carrier +and reusing them in allocator instances with an increased carrier demand is extremely effective and completely eliminates the problems that otherwise sometimes occurred when CPU load dropped while memory load did not. diff --git a/erts/emulator/internal_doc/SuperCarrier.md b/erts/emulator/internal_doc/SuperCarrier.md new file mode 100644 index 0000000000..0ad6af41de --- /dev/null +++ b/erts/emulator/internal_doc/SuperCarrier.md @@ -0,0 +1,191 @@ +Super Carrier +============= + +A super carrier is large memory area, allocated at VM start, which can +be used during runtime to allocate normal carriers from. + +The super carrier feature was introduced in OTP R16B03. It is +enabled with command line option +MMscs <size in Mb> +and can be configured with other options. + +Problem +------- + +The initial motivation for this feature was customers asking for a way +to pre-allocate physcial memory at VM start for it to use. + +Other problems were different experienced limitations of the OS +implementation of mmap: + +* Increasingly bad performance of mmap/munmap as the number of mmap'ed areas grow. +* Fragmentation problem between mmap'ed areas. + +A third problem was management of low memory in the halfword +emulator. The implementation used a naive linear search structure to +hold free segments which would lead to poor performance when +fragmentation increased. + + +Solution +-------- + +Allocate one large continious area of address space at VM start and +then use that area to satisfy our dynamic memory need during +runtime. In other words: implement our own mmap. + +### Use cases ### + +If command line option +MMscrpm (Reserve Physical Memory) is set to +false, only virtual space is allocated for the super carrier from +start. The super carrier then acts as an "alternative mmap" implementation +without changing the consumption of physical memory pages. Physical +pages will be reserved on demand when an allocation is done from the super +carrier and be unreserved when the memory is released back to the +super carrier. + +If +MMscrpm is set to true, which is default, the initial allocation +will reserve physical memory for the entire super carrier. This can be +used by users that want to ensure a certain *minimum* amount of +physical memory for the VM. + +However, what reservation of physical memory actually means highly +depends on the operating system, and how it is configured. For +example, different memory overcommit settings on Linux drastically +change the behaviour. + +A third feature is to have the super carrier limit the *maximum* +amount of memory used by the VM. If +MMsco (Super Carrier Only) is set +to true, which is default, allocations will only be done from the +super carrier. When the super carrier gets full, the VM will fail due +to out of memory. +If +MMsco is false, allocations will use mmap directly if the super +carrier is full. + + + +### Implementation ### + +The entire super carrier implementation is kept in erl_mmap.c. The +name suggest that it can be viewed as our own mmap implementation. + +A super carrier needs to satisfy two slightly different kinds of +allocation requests; multi block carriers (MBC) and single block +carriers (SBC). They are both rather large blocks of continious +memory, but MBCs and SBCs have different demands on alignment and +size. + +SBCs can have arbitrary size and do only need minimum 8-byte +alignment. + +MBCs are more restricted. They can only have a number of fixed +sizes that are powers of 2. The start address need to have a very +large aligment (currently 256 kb, called "super alignment"). This is a +design choice that allows very low overhead per allocated block in the +MBC. + +To reduce fragmentation within the super carrier, it is good to keep SBCs +and MBCs apart. MBCs with their uniform alignment and sizes can be +packed very efficiently together. SBCs without demand for aligment can +also be allocated quite efficiently together. But mixing them can lead +to a lot of memory wasted when we need to create large holes of +padding to the next alignment limit. + +The super carrier thus contains two areas. One area for MBCs growing from +the bottom and up. And one area for SBCs growing from the top and +down. Like a process with a heap and a stack growing towards each +other. + + +### Data structures ### + +The MBC area is called **sa** as in super aligned and the SBC area is +called **sua** as in super un-aligned. + +Note that the "super" in super alignment and the "super" in super +carrier has nothing to do with each other. We could have choosen +another naming to avoid confusion, such as "meta" carrier or "giant" +aligment. + + +-------+ <---- sua.top + | sua | + | | + |-------| <---- sua.bot + | | + | | + | | + |-------| <---- sa.top + | | + | sa | + | | + +-------+ <---- sa.bot + + +When a carrier is deallocated a free memory segment will be created +inside the corresponding area, unless the carrier was at the very top +(in `sa`) or bottom (in `sua`) in which case the area will just shrink +down or up. + +We need to keep track of all the free segments in order to reuse them +for new carrier allocations. One initial idea was to use the same +mechanism that is used to keep track of free blocks within MBCs +(alloc_util and the different strategies). However, that would not be +as straight forward as one can think and can also waste quite a lot of +memory as it uses prepended block headers. The granularity of the +super carrier is one memory page (usually 4kb). We want to allocate +and free entire pages and we don't want to waste an entire page just +to hold the block header of the following pages. + +Instead we store the meta information about all the free segments in a +dedicated area apart from the `sa` and `sua` areas. Every free segment is +represented by a descriptor struct (`ErtsFreeSegDesc`). + + typedef struct { + RBTNode snode; /* node in 'stree' */ + RBTNode anode; /* node in 'atree' */ + char* start; + char* end; + }ErtsFreeSegDesc; + +To find the smallest free segment that will satisfy a carrier allocation +(best fit), the free segments are organized in a tree sorted by +size (`stree`). We search in this tree at allocation. If no free segment of +sufficient size was found, the area (`sa` or `sua`) is instead expanded. +If two or more free segments with equal size exist, the one at lowest +address is choosen for `sa` and highest address for `sua`. + +At carrier deallocation, we want to coalesce with any adjacent free +segments, to form one large free segment. To do that, all free +segments are also organized in a tree sorted in address order (`atree`). + +So, in total we keep four trees of free descriptors for the super +carrier; two for `sa` and two for `sua`. They all use the same +red-black-tree implementation that support the different sorting +orders used. + +When allocating a new MBC we first search after a free segment in `sa`, +then try to raise `sa.top`, and then as a fallback try to search after a +free segment in `sua`. When an MBC is allocated in `sua`, a larger segment +is allocated which is then trimmed to obtain the right +alignment. Allocation search for an SBC is done in reverse order. When +an SBC is allocated in `sa`, the size is aligned up to super aligned +size. + +### The free descriptor area ### + +As mentioned above, the descriptors for the free segments are +allocated in a separate area. This area has a constant configurable +size (+MMscrfsd) that defaults to 65536 descriptors. This should be +more than enough in most cases. If the descriptors area should fill up, +new descriptor areas will be allocated first directly from the OS, and +then from `sua` and `sa` in the super carrier, and lastly from the memory +segment itself which is being deallocated. Allocating free descriptor +areas from the super carrier is only a last resort, and should be +avoided, as it creates fragmentation. + +### Halfword emulator ### + +The halfword emulator uses the super carrier implementation to manage +its low memory mappings thar are needed for all term storage. The +super carrier can here not be configured by command line options. One +could imagine a second configurable instance of the super carrier used +by high memory allocation, but that has not been implemented. diff --git a/erts/emulator/sys/common/erl_check_io.c b/erts/emulator/sys/common/erl_check_io.c index 1db673e7f3..0051b45b31 100644 --- a/erts/emulator/sys/common/erl_check_io.c +++ b/erts/emulator/sys/common/erl_check_io.c @@ -52,8 +52,17 @@ typedef char EventStateType; #define ERTS_EV_TYPE_STOP_USE ((EventStateType) 3) /* pending stop_select */ typedef char EventStateFlags; -#define ERTS_EV_FLAG_USED ((EventStateFlags) 1) /* ERL_DRV_USE has been turned on */ +#define ERTS_EV_FLAG_USED ((EventStateFlags) 1) /* ERL_DRV_USE has been turned on */ +#define ERTS_EV_FLAG_DEFER_IN_EV ((EventStateFlags) 2) +#define ERTS_EV_FLAG_DEFER_OUT_EV ((EventStateFlags) 4) +#ifdef DEBUG +# define ERTS_ACTIVE_FD_INC 2 +#else +# define ERTS_ACTIVE_FD_INC 128 +#endif + +#define ERTS_CHECK_IO_POLL_RES_LEN 512 #if defined(ERTS_KERNEL_POLL_VERSION) # define ERTS_CIO_EXPORT(FUNC) FUNC ## _kp @@ -67,6 +76,7 @@ typedef char EventStateFlags; (ERTS_POLL_USE_POLL && !ERTS_POLL_USE_KERNEL_POLL) #define ERTS_CIO_POLL_CTL ERTS_POLL_EXPORT(erts_poll_control) +#define ERTS_CIO_POLL_CTLV ERTS_POLL_EXPORT(erts_poll_controlv) #define ERTS_CIO_POLL_WAIT ERTS_POLL_EXPORT(erts_poll_wait) #ifdef ERTS_POLL_NEED_ASYNC_INTERRUPT_SUPPORT #define ERTS_CIO_POLL_AS_INTR ERTS_POLL_EXPORT(erts_poll_async_sig_interrupt) @@ -85,6 +95,13 @@ static struct pollset_info { ErtsPollSet ps; erts_smp_atomic_t in_poll_wait; /* set while doing poll */ + struct { + int six; /* start index */ + int eix; /* end index */ + erts_smp_atomic32_t no; + int size; + ErtsSysFdType *array; + } active_fd; #ifdef ERTS_SMP struct removed_fd* removed_list; /* list of deselected fd's*/ erts_smp_spinlock_t removed_list_lock; @@ -97,9 +114,11 @@ typedef struct { SafeHashBucket hb; #endif ErtsSysFdType fd; - union { - ErtsDrvEventDataState *event; /* ERTS_EV_TYPE_DRV_EV */ + struct { ErtsDrvSelectDataState *select; /* ERTS_EV_TYPE_DRV_SEL */ +#if ERTS_CIO_HAVE_DRV_EVENT + ErtsDrvEventDataState *event; /* ERTS_EV_TYPE_DRV_EV */ +#endif erts_driver_t* drv_ptr; /* ERTS_EV_TYPE_STOP_USE */ } driver; ErtsPollEvents events; @@ -169,6 +188,10 @@ static ERTS_INLINE ErtsDrvEventState* hash_new_drv_ev_state(ErtsSysFdType fd) ErtsDrvEventState tmpl; tmpl.fd = fd; tmpl.driver.select = NULL; +#if ERTS_CIO_HAVE_DRV_EVENT + tmpl.driver.event = NULL; +#endif + tmpl.driver.drv_ptr = NULL; tmpl.events = 0; tmpl.remove_cnt = 0; tmpl.type = ERTS_EV_TYPE_NONE; @@ -209,6 +232,69 @@ ERTS_SCHED_PREF_QUICK_ALLOC_IMPL(removed_fd, struct removed_fd, 64, ERTS_ALC_T_F #endif static ERTS_INLINE void +init_iotask(ErtsIoTask *io_task) +{ + erts_port_task_handle_init(&io_task->task); + erts_smp_atomic_init_nob(&io_task->executed_time, ~((erts_aint_t) 0)); +} + +static ERTS_INLINE int +is_iotask_active(ErtsIoTask *io_task, erts_aint_t current_cio_time) +{ + if (erts_port_task_is_scheduled(&io_task->task)) + return 1; + if (erts_smp_atomic_read_nob(&io_task->executed_time) == current_cio_time) + return 1; + return 0; +} + +static ERTS_INLINE ErtsDrvSelectDataState * +alloc_drv_select_data(void) +{ + ErtsDrvSelectDataState *dsp = erts_alloc(ERTS_ALC_T_DRV_SEL_D_STATE, + sizeof(ErtsDrvSelectDataState)); + dsp->inport = NIL; + dsp->outport = NIL; + init_iotask(&dsp->iniotask); + init_iotask(&dsp->outiotask); + return dsp; +} + +static ERTS_INLINE void +free_drv_select_data(ErtsDrvSelectDataState *dsp) +{ + ASSERT(!erts_port_task_is_scheduled(&dsp->iniotask.task)); + ASSERT(!erts_port_task_is_scheduled(&dsp->outiotask.task)); + erts_free(ERTS_ALC_T_DRV_SEL_D_STATE, dsp); +} + +#if ERTS_CIO_HAVE_DRV_EVENT + +static ERTS_INLINE ErtsDrvEventDataState * +alloc_drv_event_data(void) +{ + ErtsDrvEventDataState *dep = erts_alloc(ERTS_ALC_T_DRV_EV_D_STATE, + sizeof(ErtsDrvEventDataState)); + dep->port = NIL; + dep->data = NULL; + dep->removed_events = 0; +#if ERTS_CIO_DEFER_ACTIVE_EVENTS + dep->deferred_events = 0; +#endif + init_iotask(&dep->iotask); + return dep; +} + +static ERTS_INLINE void +free_drv_event_data(ErtsDrvEventDataState *dep) +{ + ASSERT(!erts_port_task_is_scheduled(&dep->iotask.task)); + erts_free(ERTS_ALC_T_DRV_EV_D_STATE, dep); +} + +#endif /* ERTS_CIO_HAVE_DRV_EVENT */ + +static ERTS_INLINE void remember_removed(ErtsDrvEventState *state, struct pollset_info* psi) { #ifdef ERTS_SMP @@ -288,7 +374,7 @@ forget_removed(struct pollset_info* psi) drv_ptr = state->driver.drv_ptr; ASSERT(drv_ptr); state->type = ERTS_EV_TYPE_NONE; - state->flags = 0; + state->flags &= ~ERTS_EV_FLAG_USED; state->driver.drv_ptr = NULL; /* Fall through */ case ERTS_EV_TYPE_NONE: @@ -345,6 +431,10 @@ grow_drv_ev_state(int min_ix) for (i = erts_smp_atomic_read_nob(&drv_ev_state_len); i < new_len; i++) { drv_ev_state[i].fd = (ErtsSysFdType) i; drv_ev_state[i].driver.select = NULL; +#if ERTS_CIO_HAVE_DRV_EVENT + drv_ev_state[i].driver.event = NULL; +#endif + drv_ev_state[i].driver.drv_ptr = NULL; drv_ev_state[i].events = 0; drv_ev_state[i].remove_cnt = 0; drv_ev_state[i].type = ERTS_EV_TYPE_NONE; @@ -365,11 +455,7 @@ grow_drv_ev_state(int min_ix) static ERTS_INLINE void abort_task(Eterm id, ErtsPortTaskHandle *pthp, EventStateType type) { - if (is_nil(id)) { - ASSERT(type == ERTS_EV_TYPE_NONE - || !erts_port_task_is_scheduled(pthp)); - } - else if (erts_port_task_is_scheduled(pthp)) { + if (is_not_nil(id) && erts_port_task_is_scheduled(pthp)) { erts_port_task_abort(pthp); ASSERT(erts_is_port_alive(id)); } @@ -384,7 +470,7 @@ abort_tasks(ErtsDrvEventState *state, int mode) #if ERTS_CIO_HAVE_DRV_EVENT case ERTS_EV_TYPE_DRV_EV: abort_task(state->driver.event->port, - &state->driver.event->task, + &state->driver.event->iotask.task, ERTS_EV_TYPE_DRV_EV); return; #endif @@ -398,14 +484,14 @@ abort_tasks(ErtsDrvEventState *state, int mode) case ERL_DRV_WRITE: ASSERT(state->type == ERTS_EV_TYPE_DRV_SEL); abort_task(state->driver.select->outport, - &state->driver.select->outtask, + &state->driver.select->outiotask.task, state->type); if (mode == ERL_DRV_WRITE) break; case ERL_DRV_READ: ASSERT(state->type == ERTS_EV_TYPE_DRV_SEL); abort_task(state->driver.select->inport, - &state->driver.select->intask, + &state->driver.select->iniotask.task, state->type); break; default: @@ -443,16 +529,14 @@ deselect(ErtsDrvEventState *state, int mode) if (!(state->events)) { switch (state->type) { case ERTS_EV_TYPE_DRV_SEL: - ASSERT(!erts_port_task_is_scheduled(&state->driver.select->intask)); - ASSERT(!erts_port_task_is_scheduled(&state->driver.select->outtask)); - erts_free(ERTS_ALC_T_DRV_SEL_D_STATE, - state->driver.select); + state->driver.select->inport = NIL; + state->driver.select->outport = NIL; break; #if ERTS_CIO_HAVE_DRV_EVENT case ERTS_EV_TYPE_DRV_EV: - ASSERT(!erts_port_task_is_scheduled(&state->driver.event->task)); - erts_free(ERTS_ALC_T_DRV_EV_D_STATE, - state->driver.event); + state->driver.event->port = NIL; + state->driver.event->data = NULL; + state->driver.event->removed_events = (ErtsPollEvents) 0; break; #endif case ERTS_EV_TYPE_NONE: @@ -462,20 +546,297 @@ deselect(ErtsDrvEventState *state, int mode) break; } - state->driver.select = NULL; state->type = ERTS_EV_TYPE_NONE; - state->flags = 0; + state->flags &= ~ERTS_EV_FLAG_USED; remember_removed(state, &pollset); } } - #ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS # define IS_FD_UNKNOWN(state) ((state)->type == ERTS_EV_TYPE_NONE && (state)->remove_cnt == 0) #else # define IS_FD_UNKNOWN(state) ((state) == NULL) #endif +static ERTS_INLINE void +check_fd_cleanup(ErtsDrvEventState *state, +#if ERTS_CIO_HAVE_DRV_EVENT + ErtsDrvEventDataState **free_event, +#endif + ErtsDrvSelectDataState **free_select) +{ + erts_aint_t current_cio_time; + + ERTS_SMP_LC_ASSERT(erts_smp_lc_mtx_is_locked(fd_mtx(state->fd))); + + current_cio_time = erts_smp_atomic_read_acqb(&erts_check_io_time); + *free_select = NULL; + if (state->driver.select + && (state->type != ERTS_EV_TYPE_DRV_SEL) + && !is_iotask_active(&state->driver.select->iniotask, current_cio_time) + && !is_iotask_active(&state->driver.select->outiotask, current_cio_time)) { + + *free_select = state->driver.select; + state->driver.select = NULL; + } + +#if ERTS_CIO_HAVE_DRV_EVENT + *free_event = NULL; + if (state->driver.event + && (state->type != ERTS_EV_TYPE_DRV_EV) + && !is_iotask_active(&state->driver.event->iotask, current_cio_time)) { + + *free_event = state->driver.event; + state->driver.event = NULL; + } +#endif + +#ifndef ERTS_SYS_CONTINOUS_FD_NUMBERS + if (((state->type != ERTS_EV_TYPE_NONE) + | state->remove_cnt +#if ERTS_CIO_HAVE_DRV_EVENT + | (state->driver.event != NULL) +#endif + | (state->driver.select != NULL)) == 0) { + + hash_erase_drv_ev_state(state); + + } +#endif +} + +static ERTS_INLINE int +check_cleanup_active_fd(ErtsSysFdType fd, +#if ERTS_CIO_DEFER_ACTIVE_EVENTS + ErtsPollControlEntry *pce, + int *pce_ix, +#endif + erts_aint_t current_cio_time) +{ + ErtsDrvEventState *state; + int active = 0; + erts_smp_mtx_t *mtx = fd_mtx(fd); + void *free_select = NULL; +#if ERTS_CIO_HAVE_DRV_EVENT + void *free_event = NULL; +#endif +#if ERTS_CIO_DEFER_ACTIVE_EVENTS + ErtsPollEvents evon = 0, evoff = 0; +#endif + + erts_smp_mtx_lock(mtx); + +#ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS + state = &drv_ev_state[(int) fd]; +#else + state = hash_get_drv_ev_state(fd); /* may be NULL! */ + if (state) +#endif + { + if (state->driver.select) { +#if ERTS_CIO_DEFER_ACTIVE_EVENTS + if (is_iotask_active(&state->driver.select->iniotask, current_cio_time)) { + active = 1; + if ((state->events & ERTS_POLL_EV_IN) + && !(state->flags & ERTS_EV_FLAG_DEFER_IN_EV)) { + evoff |= ERTS_POLL_EV_IN; + state->flags |= ERTS_EV_FLAG_DEFER_IN_EV; + } + } + else if (state->flags & ERTS_EV_FLAG_DEFER_IN_EV) { + if (state->events & ERTS_POLL_EV_IN) + evon |= ERTS_POLL_EV_IN; + state->flags &= ~ERTS_EV_FLAG_DEFER_IN_EV; + } + if (is_iotask_active(&state->driver.select->outiotask, current_cio_time)) { + active = 1; + if ((state->events & ERTS_POLL_EV_OUT) + && !(state->flags & ERTS_EV_FLAG_DEFER_OUT_EV)) { + evoff |= ERTS_POLL_EV_OUT; + state->flags |= ERTS_EV_FLAG_DEFER_OUT_EV; + } + } + else if (state->flags & ERTS_EV_FLAG_DEFER_OUT_EV) { + if (state->events & ERTS_POLL_EV_OUT) + evon |= ERTS_POLL_EV_OUT; + state->flags &= ~ERTS_EV_FLAG_DEFER_OUT_EV; + } + if (active) + (void) 0; + else +#else + if (is_iotask_active(&state->driver.select->iniotask, current_cio_time) + || is_iotask_active(&state->driver.select->outiotask, current_cio_time)) + active = 1; + else +#endif + if (state->type != ERTS_EV_TYPE_DRV_SEL) { + free_select = state->driver.select; + state->driver.select = NULL; + } + } + +#if ERTS_CIO_HAVE_DRV_EVENT + if (state->driver.event) { + if (is_iotask_active(&state->driver.event->iotask, current_cio_time)) { +#if ERTS_CIO_DEFER_ACTIVE_EVENTS + ErtsPollEvents evs = state->events & ~state->driver.event->deferred_events; + if (evs) { + evoff |= evs; + state->driver.event->deferred_events |= evs; + } +#endif + active = 1; + } + else if (state->type != ERTS_EV_TYPE_DRV_EV) { + free_event = state->driver.event; + state->driver.event = NULL; + } +#if ERTS_CIO_DEFER_ACTIVE_EVENTS + else { + ErtsPollEvents evs = state->events & state->driver.event->deferred_events; + if (evs) { + evon |= evs; + state->driver.event->deferred_events = 0; + } + } +#endif + + } +#endif + +#ifndef ERTS_SYS_CONTINOUS_FD_NUMBERS + if (((state->type != ERTS_EV_TYPE_NONE) | state->remove_cnt | active) == 0) + hash_erase_drv_ev_state(state); +#endif + + } + + erts_smp_mtx_unlock(mtx); + + if (free_select) + free_drv_select_data(free_select); +#if ERTS_CIO_HAVE_DRV_EVENT + if (free_event) + free_drv_event_data(free_event); +#endif + +#if ERTS_CIO_DEFER_ACTIVE_EVENTS + if (evoff) { + ErtsPollControlEntry *pcep = &pce[(*pce_ix)++]; + pcep->fd = fd; + pcep->events = evoff; + pcep->on = 0; + } + if (evon) { + ErtsPollControlEntry *pcep = &pce[(*pce_ix)++]; + pcep->fd = fd; + pcep->events = evon; + pcep->on = 1; + } +#endif + + return active; +} + +static void +check_cleanup_active_fds(erts_aint_t current_cio_time) +{ + int six = pollset.active_fd.six; + int eix = pollset.active_fd.eix; + erts_aint32_t no = erts_smp_atomic32_read_dirty(&pollset.active_fd.no); + int size = pollset.active_fd.size; + int ix = six; +#if ERTS_CIO_DEFER_ACTIVE_EVENTS + /* every fd might add two entries */ + Uint pce_sz = 2*sizeof(ErtsPollControlEntry)*no; + ErtsPollControlEntry *pctrl_entries = (pce_sz + ? erts_alloc(ERTS_ALC_T_TMP, pce_sz) + : NULL); + int pctrl_ix = 0; +#endif + + while (ix != eix) { + ErtsSysFdType fd = pollset.active_fd.array[ix]; + int nix = ix + 1; + if (nix >= size) + nix = 0; + ASSERT(fd != ERTS_SYS_FD_INVALID); + if (!check_cleanup_active_fd(fd, +#if ERTS_CIO_DEFER_ACTIVE_EVENTS + pctrl_entries, + &pctrl_ix, +#endif + current_cio_time)) { + no--; + if (ix == six) { +#ifdef DEBUG + pollset.active_fd.array[ix] = ERTS_SYS_FD_INVALID; +#endif + six = nix; + } + else { + pollset.active_fd.array[ix] = pollset.active_fd.array[six]; +#ifdef DEBUG + pollset.active_fd.array[six] = ERTS_SYS_FD_INVALID; +#endif + six++; + if (six >= size) + six = 0; + } + } + ix = nix; + } + +#if ERTS_CIO_DEFER_ACTIVE_EVENTS + ASSERT(pctrl_ix <= pce_sz/sizeof(ErtsPollControlEntry)); + if (pctrl_ix) + ERTS_CIO_POLL_CTLV(pollset.ps, pctrl_entries, pctrl_ix); + if (pctrl_entries) + erts_free(ERTS_ALC_T_TMP, pctrl_entries); +#endif + + pollset.active_fd.six = six; + pollset.active_fd.eix = eix; + erts_smp_atomic32_set_relb(&pollset.active_fd.no, no); +} + +static ERTS_INLINE void +add_active_fd(ErtsSysFdType fd) +{ + int eix = pollset.active_fd.eix; + int size = pollset.active_fd.size; + + + pollset.active_fd.array[eix] = fd; + + erts_smp_atomic32_set_relb(&pollset.active_fd.no, + (erts_smp_atomic32_read_dirty(&pollset.active_fd.no) + + 1)); + + eix++; + if (eix >= size) + eix = 0; + if (pollset.active_fd.six == eix) { + pollset.active_fd.six = 0; + eix = size; + size += ERTS_ACTIVE_FD_INC; + pollset.active_fd.array = erts_realloc(ERTS_ALC_T_ACTIVE_FD_ARR, + pollset.active_fd.array, + sizeof(ErtsSysFdType)*size); + pollset.active_fd.size = size; +#ifdef DEBUG + { + int i; + for (i = eix + 1; i < size; i++) + pollset.active_fd.array[i] = ERTS_SYS_FD_INVALID; + } +#endif + + } + + pollset.active_fd.eix = eix; +} int ERTS_CIO_EXPORT(driver_select)(ErlDrvPort ix, @@ -492,6 +853,10 @@ ERTS_CIO_EXPORT(driver_select)(ErlDrvPort ix, ErtsDrvEventState *state; int wake_poller; int ret; +#if ERTS_CIO_HAVE_DRV_EVENT + ErtsDrvEventDataState *free_event = NULL; +#endif + ErtsDrvSelectDataState *free_select = NULL; #ifdef USE_VM_PROBES DTRACE_CHARBUF(name, 64); #endif @@ -593,9 +958,9 @@ ERTS_CIO_EXPORT(driver_select)(ErlDrvPort ix, if (new_events & (ERTS_POLL_EV_ERR|ERTS_POLL_EV_NVAL)) { if (state->type == ERTS_EV_TYPE_DRV_SEL && !state->events) { state->type = ERTS_EV_TYPE_NONE; - state->flags = 0; - erts_free(ERTS_ALC_T_DRV_SEL_D_STATE, state->driver.select); - state->driver.select = NULL; + state->flags &= ~ERTS_EV_FLAG_USED; + state->driver.select->inport = NIL; + state->driver.select->outport = NIL; } ret = -1; goto done; @@ -613,18 +978,10 @@ ERTS_CIO_EXPORT(driver_select)(ErlDrvPort ix, state->events = new_events; if (ctl_events) { if (on) { - if (state->type == ERTS_EV_TYPE_NONE) { - ErtsDrvSelectDataState *dsdsp - = erts_alloc(ERTS_ALC_T_DRV_SEL_D_STATE, - sizeof(ErtsDrvSelectDataState)); - dsdsp->inport = NIL; - dsdsp->outport = NIL; - erts_port_task_handle_init(&dsdsp->intask); - erts_port_task_handle_init(&dsdsp->outtask); - ASSERT(state->driver.select == NULL); - state->driver.select = dsdsp; + if (!state->driver.select) + state->driver.select = alloc_drv_select_data(); + if (state->type == ERTS_EV_TYPE_NONE) state->type = ERTS_EV_TYPE_DRV_SEL; - } ASSERT(state->type == ERTS_EV_TYPE_DRV_SEL); if (ctl_events & ERTS_POLL_EV_IN) state->driver.select->inport = id; @@ -645,17 +1002,12 @@ ERTS_CIO_EXPORT(driver_select)(ErlDrvPort ix, state->driver.select->outport = NIL; } if (new_events == 0) { - ASSERT(!erts_port_task_is_scheduled(&state->driver.select->intask)); - ASSERT(!erts_port_task_is_scheduled(&state->driver.select->outtask)); if (old_events != 0) { remember_removed(state, &pollset); } if ((mode & ERL_DRV_USE) || !(state->flags & ERTS_EV_FLAG_USED)) { state->type = ERTS_EV_TYPE_NONE; - state->flags = 0; - erts_free(ERTS_ALC_T_DRV_SEL_D_STATE, - state->driver.select); - state->driver.select = NULL; + state->flags &= ~ERTS_EV_FLAG_USED; } /*else keep it, as fd will probably be selected upon again */ } @@ -686,13 +1038,15 @@ ERTS_CIO_EXPORT(driver_select)(ErlDrvPort ix, ret = 0; -done:; -#ifndef ERTS_SYS_CONTINOUS_FD_NUMBERS - if (state->type == ERTS_EV_TYPE_NONE && state->remove_cnt == 0) { - hash_erase_drv_ev_state(state); - } +done: + + check_fd_cleanup(state, +#if ERTS_CIO_HAVE_DRV_EVENT + &free_event, #endif -done_unknown: + &free_select); + +done_unknown: erts_smp_mtx_unlock(fd_mtx(fd)); if (stop_select_fn) { int was_unmasked = erts_block_fpe(); @@ -700,6 +1054,12 @@ done_unknown: (*stop_select_fn)(e, NULL); erts_unblock_fpe(was_unmasked); } + if (free_select) + free_drv_select_data(free_select); +#if ERTS_CIO_HAVE_DRV_EVENT + if (free_event) + free_drv_event_data(free_event); +#endif return ret; } @@ -719,6 +1079,10 @@ ERTS_CIO_EXPORT(driver_event)(ErlDrvPort ix, ErtsDrvEventState *state; int do_wake = 0; int ret; +#if ERTS_CIO_HAVE_DRV_EVENT + ErtsDrvEventDataState *free_event; +#endif + ErtsDrvSelectDataState *free_select; Port *prt = erts_drvport2port(ix); if (prt == ERTS_INVALID_ERL_DRV_PORT) @@ -799,10 +1163,8 @@ ERTS_CIO_EXPORT(driver_event)(ErlDrvPort ix, state->driver.event->removed_events |= remove_events; } else { - state->driver.event - = erts_alloc(ERTS_ALC_T_DRV_EV_D_STATE, - sizeof(ErtsDrvEventDataState)); - erts_port_task_handle_init(&state->driver.event->task); + if (!state->driver.event) + state->driver.event = alloc_drv_event_data(); state->driver.event->port = id; state->driver.event->removed_events = (ErtsPollEvents) 0; state->type = ERTS_EV_TYPE_DRV_EV; @@ -812,10 +1174,10 @@ ERTS_CIO_EXPORT(driver_event)(ErlDrvPort ix, else { if (state->type == ERTS_EV_TYPE_DRV_EV) { abort_tasks(state, 0); - erts_free(ERTS_ALC_T_DRV_EV_D_STATE, - state->driver.event); + state->driver.event->port = NIL; + state->driver.event->data = NULL; + state->driver.event->removed_events = (ErtsPollEvents) 0; } - state->driver.select = NULL; state->type = ERTS_EV_TYPE_NONE; remember_removed(state, &pollset); } @@ -825,12 +1187,22 @@ ERTS_CIO_EXPORT(driver_event)(ErlDrvPort ix, ret = 0; done: -#ifndef ERTS_SYS_CONTINOUS_FD_NUMBERS - if (state->type == ERTS_EV_TYPE_NONE && state->remove_cnt == 0) { - hash_erase_drv_ev_state(state); - } + + check_fd_cleanup(state, +#if ERTS_CIO_HAVE_DRV_EVENT + &free_event, #endif + &free_select); + erts_smp_mtx_unlock(fd_mtx(fd)); + + if (free_select) + free_drv_select_data(free_select); +#if ERTS_CIO_HAVE_DRV_EVENT + if (free_event) + free_drv_event_data(free_event); +#endif + return ret; #endif } @@ -1027,7 +1399,7 @@ steal_pending_stop_select(erts_dsprintf_buf_t *dsbufp, ErlDrvPort ix, * In either case stop_select should not be called. */ state->type = ERTS_EV_TYPE_NONE; - state->flags = 0; + state->flags &= ~ERTS_EV_FLAG_USED; if (state->driver.drv_ptr->handle) { erts_ddll_dereference_driver(state->driver.drv_ptr->handle); } @@ -1099,38 +1471,103 @@ event_large_fd_error(ErlDrvPort ix, ErtsSysFdType fd, ErlDrvEventData event_data #endif #endif +static ERTS_INLINE int +io_task_schedule_allowed(ErtsDrvEventState *state, + ErtsPortTaskType type, + erts_aint_t current_cio_time) +{ + ErtsIoTask *io_task; + + switch (type) { + case ERTS_PORT_TASK_INPUT: + if (!state->driver.select) + return 0; +#if ERTS_CIO_HAVE_DRV_EVENT + if (state->driver.event) + return 0; +#endif + io_task = &state->driver.select->iniotask; + break; + case ERTS_PORT_TASK_OUTPUT: + if (!state->driver.select) + return 0; +#if ERTS_CIO_HAVE_DRV_EVENT + if (state->driver.event) + return 0; +#endif + io_task = &state->driver.select->outiotask; + break; +#if ERTS_CIO_HAVE_DRV_EVENT + case ERTS_PORT_TASK_EVENT: + if (!state->driver.event) + return 0; + if (state->driver.select) + return 0; + io_task = &state->driver.event->iotask; + break; +#endif + default: + ERTS_INTERNAL_ERROR("Invalid I/O-task type"); + return 0; + } + + return !is_iotask_active(io_task, current_cio_time); +} + static ERTS_INLINE void -iready(Eterm id, ErtsDrvEventState *state) +iready(Eterm id, ErtsDrvEventState *state, erts_aint_t current_cio_time) { - if (erts_port_task_schedule(id, - &state->driver.select->intask, - ERTS_PORT_TASK_INPUT, - (ErlDrvEvent) state->fd) != 0) { - stale_drv_select(id, state, ERL_DRV_READ); + if (io_task_schedule_allowed(state, + ERTS_PORT_TASK_INPUT, + current_cio_time)) { + ErtsIoTask *iotask = &state->driver.select->iniotask; + erts_smp_atomic_set_nob(&iotask->executed_time, current_cio_time); + if (erts_port_task_schedule(id, + &iotask->task, + ERTS_PORT_TASK_INPUT, + (ErlDrvEvent) state->fd) != 0) { + stale_drv_select(id, state, ERL_DRV_READ); + } + add_active_fd(state->fd); } } static ERTS_INLINE void -oready(Eterm id, ErtsDrvEventState *state) +oready(Eterm id, ErtsDrvEventState *state, erts_aint_t current_cio_time) { - if (erts_port_task_schedule(id, - &state->driver.select->outtask, - ERTS_PORT_TASK_OUTPUT, - (ErlDrvEvent) state->fd) != 0) { - stale_drv_select(id, state, ERL_DRV_WRITE); + if (io_task_schedule_allowed(state, + ERTS_PORT_TASK_OUTPUT, + current_cio_time)) { + ErtsIoTask *iotask = &state->driver.select->outiotask; + erts_smp_atomic_set_nob(&iotask->executed_time, current_cio_time); + if (erts_port_task_schedule(id, + &iotask->task, + ERTS_PORT_TASK_OUTPUT, + (ErlDrvEvent) state->fd) != 0) { + stale_drv_select(id, state, ERL_DRV_WRITE); + } + add_active_fd(state->fd); } } #if ERTS_CIO_HAVE_DRV_EVENT static ERTS_INLINE void -eready(Eterm id, ErtsDrvEventState *state, ErlDrvEventData event_data) +eready(Eterm id, ErtsDrvEventState *state, ErlDrvEventData event_data, + erts_aint_t current_cio_time) { - if (erts_port_task_schedule(id, - &state->driver.event->task, - ERTS_PORT_TASK_EVENT, - (ErlDrvEvent) state->fd, - event_data) != 0) { - stale_drv_select(id, state, 0); + if (io_task_schedule_allowed(state, + ERTS_PORT_TASK_EVENT, + current_cio_time)) { + ErtsIoTask *iotask = &state->driver.event->iotask; + erts_smp_atomic_set_nob(&iotask->executed_time, current_cio_time); + if (erts_port_task_schedule(id, + &iotask->task, + ERTS_PORT_TASK_EVENT, + (ErlDrvEvent) state->fd, + event_data) != 0) { + stale_drv_select(id, state, 0); + } + add_active_fd(state->fd); } } #endif @@ -1161,10 +1598,11 @@ ERTS_CIO_EXPORT(erts_check_io_interrupt_timed)(int set, void ERTS_CIO_EXPORT(erts_check_io)(int do_wait) { - ErtsPollResFd pollres[256]; + ErtsPollResFd *pollres; int pollres_len; SysTimeval wait_time; int poll_ret, i; + erts_aint_t current_cio_time; restart: @@ -1181,10 +1619,24 @@ ERTS_CIO_EXPORT(erts_check_io)(int do_wait) wait_time.tv_usec = 0; } + /* + * No need for an atomic inc op when incrementing + * erts_check_io_time, since only one thread can + * check io at a time. + */ + current_cio_time = erts_smp_atomic_read_dirty(&erts_check_io_time); + current_cio_time++; + erts_smp_atomic_set_relb(&erts_check_io_time, current_cio_time); + + check_cleanup_active_fds(current_cio_time); + #ifdef ERTS_ENABLE_LOCK_CHECK erts_lc_check_exact(NULL, 0); /* No locks should be locked */ #endif - pollres_len = sizeof(pollres)/sizeof(ErtsPollResFd); + + pollres_len = erts_smp_atomic32_read_dirty(&pollset.active_fd.no) + ERTS_CHECK_IO_POLL_RES_LEN; + + pollres = erts_alloc(ERTS_ALC_T_TMP, sizeof(ErtsPollResFd)*pollres_len); erts_smp_atomic_set_nob(&pollset.in_poll_wait, 1); @@ -1204,6 +1656,7 @@ ERTS_CIO_EXPORT(erts_check_io)(int do_wait) if (poll_ret != 0) { erts_smp_atomic_set_nob(&pollset.in_poll_wait, 0); forget_removed(&pollset); + erts_free(ERTS_ALC_T_TMP, pollres); if (poll_ret == EAGAIN) { goto restart; } @@ -1263,15 +1716,15 @@ ERTS_CIO_EXPORT(erts_check_io)(int do_wait) if ((revents & ERTS_POLL_EV_IN) || (!(revents & ERTS_POLL_EV_OUT) && state->events & ERTS_POLL_EV_IN)) { - iready(state->driver.select->inport, state); + iready(state->driver.select->inport, state, current_cio_time); } else if (state->events & ERTS_POLL_EV_OUT) { - oready(state->driver.select->outport, state); + oready(state->driver.select->outport, state, current_cio_time); } } else if (revents & (ERTS_POLL_EV_IN|ERTS_POLL_EV_OUT)) { if (revents & ERTS_POLL_EV_OUT) { - oready(state->driver.select->outport, state); + oready(state->driver.select->outport, state, current_cio_time); } /* Someone might have deselected input since revents was read (true also on the non-smp emulator since @@ -1279,7 +1732,7 @@ ERTS_CIO_EXPORT(erts_check_io)(int do_wait) revents... */ revents &= ~(~state->events & ERTS_POLL_EV_IN); if (revents & ERTS_POLL_EV_IN) { - iready(state->driver.select->inport, state); + iready(state->driver.select->inport, state, current_cio_time); } } else if (revents & ERTS_POLL_EV_NVAL) { @@ -1287,6 +1740,7 @@ ERTS_CIO_EXPORT(erts_check_io)(int do_wait) state->driver.select->inport, state->driver.select->outport, state->events); + add_active_fd(state->fd); } break; } @@ -1304,8 +1758,7 @@ ERTS_CIO_EXPORT(erts_check_io)(int do_wait) if (revents) { event_data->events = state->events; event_data->revents = revents; - - eready(state->driver.event->port, state, event_data); + eready(state->driver.event->port, state, event_data, current_cio_time); } break; } @@ -1323,6 +1776,7 @@ ERTS_CIO_EXPORT(erts_check_io)(int do_wait) (int) state->type); ASSERT(0); deselect(state, 0); + add_active_fd(state->fd); break; } } @@ -1334,6 +1788,7 @@ ERTS_CIO_EXPORT(erts_check_io)(int do_wait) } erts_smp_atomic_set_nob(&pollset.in_poll_wait, 0); + erts_free(ERTS_ALC_T_TMP, pollres); forget_removed(&pollset); } @@ -1469,10 +1924,27 @@ static void drv_ev_state_free(void *des) void ERTS_CIO_EXPORT(erts_init_check_io)(void) { + erts_smp_atomic_init_nob(&erts_check_io_time, 0); erts_smp_atomic_init_nob(&pollset.in_poll_wait, 0); + ERTS_CIO_POLL_INIT(); pollset.ps = ERTS_CIO_NEW_POLLSET(); + pollset.active_fd.six = 0; + pollset.active_fd.eix = 0; + erts_smp_atomic32_init_nob(&pollset.active_fd.no, 0); + pollset.active_fd.size = ERTS_ACTIVE_FD_INC; + pollset.active_fd.array = erts_alloc(ERTS_ALC_T_ACTIVE_FD_ARR, + sizeof(ErtsSysFdType)*ERTS_ACTIVE_FD_INC); +#ifdef DEBUG + { + int i; + for (i = 0; i < ERTS_ACTIVE_FD_INC; i++) + pollset.active_fd.array[i] = ERTS_SYS_FD_INVALID; + } +#endif + + #ifdef ERTS_SMP init_removed_fd_alloc(); pollset.removed_list = NULL; @@ -1548,12 +2020,27 @@ Eterm ERTS_CIO_EXPORT(erts_check_io_info)(void *proc) { Process *p = (Process *) proc; - Eterm tags[15], values[15], res; + Eterm tags[16], values[16], res; Uint sz, *szp, *hp, **hpp, memory_size; Sint i; ErtsPollInfo pi; - - ERTS_CIO_POLL_INFO(pollset.ps, &pi); + erts_aint_t cio_time = erts_smp_atomic_read_acqb(&erts_check_io_time); + int active_fds = (int) erts_smp_atomic32_read_acqb(&pollset.active_fd.no); + + while (1) { + erts_aint_t post_cio_time; + int post_active_fds; + + ERTS_CIO_POLL_INFO(pollset.ps, &pi); + + post_cio_time = erts_smp_atomic_read_mb(&erts_check_io_time); + post_active_fds = (int) erts_smp_atomic32_read_acqb(&pollset.active_fd.no); + if (cio_time == post_cio_time && active_fds == post_active_fds) + break; + cio_time = post_cio_time; + active_fds = post_active_fds; + } + memory_size = pi.memory_size; #ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS memory_size += sizeof(ErtsDrvEventState) * erts_smp_atomic_read_nob(&drv_ev_state_len); @@ -1617,6 +2104,9 @@ ERTS_CIO_EXPORT(erts_check_io_info)(void *proc) tags[i] = erts_bld_atom(hpp, szp, "max_fds"); values[i++] = erts_bld_uint(hpp, szp, (Uint) pi.max_fds); + tags[i] = erts_bld_atom(hpp, szp, "active_fds"); + values[i++] = erts_bld_uint(hpp, szp, (Uint) active_fds); + #ifdef ERTS_POLL_COUNT_AVOIDED_WAKEUPS tags[i] = erts_bld_atom(hpp, szp, "no_avoided_wakeups"); values[i++] = erts_bld_uint(hpp, szp, (Uint) pi.no_avoided_wakeups); @@ -1671,6 +2161,8 @@ print_events(ErtsPollEvents ev) typedef struct { int used_fds; int num_errors; + int no_driver_select_structs; + int no_driver_event_structs; #ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS int internal_fds; ErtsPollEvents *epep; @@ -1693,6 +2185,13 @@ static void doit_erts_check_io_debug(void *vstate, void *vcounters) struct stat stat_buf; #endif + if (state->driver.select) + counters->no_driver_select_structs++; +#if ERTS_CIO_HAVE_DRV_EVENT + if (state->driver.event) + counters->no_driver_event_structs++; +#endif + #ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS if (state->events || ep_events) { if (ep_events & ERTS_POLL_EV_NVAL) { @@ -1831,6 +2330,7 @@ static void doit_erts_check_io_debug(void *vstate, void *vcounters) } } } +#if ERTS_CIO_HAVE_DRV_EVENT else if (state->type == ERTS_EV_TYPE_DRV_EV) { Eterm id; erts_printf("driver_event "); @@ -1866,6 +2366,7 @@ static void doit_erts_check_io_debug(void *vstate, void *vcounters) erts_free_port_names(pnp); } } +#endif #ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS else if (internal) { erts_printf("internal "); @@ -1905,7 +2406,7 @@ static void doit_erts_check_io_debug(void *vstate, void *vcounters) } int -ERTS_CIO_EXPORT(erts_check_io_debug)(void) +ERTS_CIO_EXPORT(erts_check_io_debug)(ErtsCheckIoDebugInfo *ciodip) { #ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS int fd, len; @@ -1915,6 +2416,10 @@ ERTS_CIO_EXPORT(erts_check_io_debug)(void) ErtsDrvEventState null_des; null_des.driver.select = NULL; +#if ERTS_CIO_HAVE_DRV_EVENT + null_des.driver.event = NULL; +#endif + null_des.driver.drv_ptr = NULL; null_des.events = 0; null_des.remove_cnt = 0; null_des.type = ERTS_EV_TYPE_NONE; @@ -1935,6 +2440,8 @@ ERTS_CIO_EXPORT(erts_check_io_debug)(void) #endif counters.used_fds = 0; counters.num_errors = 0; + counters.no_driver_select_structs = 0; + counters.no_driver_event_structs = 0; #ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS len = erts_smp_atomic_read_nob(&drv_ev_state_len); @@ -1951,8 +2458,16 @@ ERTS_CIO_EXPORT(erts_check_io_debug)(void) erts_smp_thr_progress_unblock(); + ciodip->no_used_fds = counters.used_fds; + ciodip->no_driver_select_structs = counters.no_driver_select_structs; + ciodip->no_driver_event_structs = counters.no_driver_event_structs; + erts_printf("\n"); erts_printf("used fds=%d\n", counters.used_fds); + erts_printf("Number of driver_select() structures=%d\n", counters.no_driver_select_structs); +#if ERTS_CIO_HAVE_DRV_EVENT + erts_printf("Number of driver_event() structures=%d\n", counters.no_driver_event_structs); +#endif #ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS erts_printf("internal fds=%d\n", counters.internal_fds); #endif @@ -1961,6 +2476,7 @@ ERTS_CIO_EXPORT(erts_check_io_debug)(void) #ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS erts_free(ERTS_ALC_T_TMP, (void *) counters.epep); #endif + return counters.num_errors; } diff --git a/erts/emulator/sys/common/erl_check_io.h b/erts/emulator/sys/common/erl_check_io.h index edab7947ba..d01297d55c 100644 --- a/erts/emulator/sys/common/erl_check_io.h +++ b/erts/emulator/sys/common/erl_check_io.h @@ -26,6 +26,7 @@ #ifndef ERL_CHECK_IO_H__ #define ERL_CHECK_IO_H__ +#include "sys.h" #include "erl_sys_driver.h" #ifdef ERTS_ENABLE_KERNEL_POLL @@ -52,8 +53,8 @@ void erts_check_io_kp(int); void erts_check_io_nkp(int); void erts_init_check_io_kp(void); void erts_init_check_io_nkp(void); -int erts_check_io_debug_kp(void); -int erts_check_io_debug_nkp(void); +int erts_check_io_debug_kp(ErtsCheckIoDebugInfo *); +int erts_check_io_debug_nkp(ErtsCheckIoDebugInfo *); #else /* !ERTS_ENABLE_KERNEL_POLL */ @@ -70,6 +71,27 @@ void erts_init_check_io(void); #endif +extern erts_smp_atomic_t erts_check_io_time; + +typedef struct { + ErtsPortTaskHandle task; + erts_smp_atomic_t executed_time; +} ErtsIoTask; + +ERTS_GLB_INLINE void erts_io_notify_port_task_executed(ErtsPortTaskHandle *pthp); + +#if ERTS_GLB_INLINE_INCL_FUNC_DEF + +ERTS_GLB_INLINE void +erts_io_notify_port_task_executed(ErtsPortTaskHandle *pthp) +{ + ErtsIoTask *itp = (ErtsIoTask *) (((char *) pthp) - offsetof(ErtsIoTask, task)); + erts_aint_t ci_time = erts_smp_atomic_read_acqb(&erts_check_io_time); + erts_smp_atomic_set_relb(&itp->executed_time, ci_time); +} + +#endif + #endif /* ERL_CHECK_IO_H__ */ #if !defined(ERL_CHECK_IO_C__) && !defined(ERTS_ALLOC_C__) @@ -81,6 +103,16 @@ void erts_init_check_io(void); #include "erl_poll.h" #include "erl_port_task.h" +#ifdef __WIN32__ +/* + * Current erts_poll implementation for Windows cannot handle + * active events in the set of events polled. + */ +# define ERTS_CIO_DEFER_ACTIVE_EVENTS 1 +#else +# define ERTS_CIO_DEFER_ACTIVE_EVENTS 0 +#endif + /* * ErtsDrvEventDataState is used by driver_event() which is almost never * used. We allocate ErtsDrvEventDataState separate since we dont wan't @@ -91,13 +123,16 @@ typedef struct { Eterm port; ErlDrvEventData data; ErtsPollEvents removed_events; - ErtsPortTaskHandle task; +#if ERTS_CIO_DEFER_ACTIVE_EVENTS + ErtsPollEvents deferred_events; +#endif + ErtsIoTask iotask; } ErtsDrvEventDataState; typedef struct { Eterm inport; Eterm outport; - ErtsPortTaskHandle intask; - ErtsPortTaskHandle outtask; + ErtsIoTask iniotask; + ErtsIoTask outiotask; } ErtsDrvSelectDataState; #endif /* #ifndef ERL_CHECK_IO_INTERNAL__ */ diff --git a/erts/emulator/sys/common/erl_poll.c b/erts/emulator/sys/common/erl_poll.c index 0a58a625b2..aa412a20c8 100644 --- a/erts/emulator/sys/common/erl_poll.c +++ b/erts/emulator/sys/common/erl_poll.c @@ -2157,7 +2157,7 @@ ERTS_POLL_EXPORT(erts_poll_wait)(ErtsPollSet ps, #ifdef ERTS_POLL_DEBUG_PRINT erts_printf("Entering erts_poll_wait(), timeout=%d\n", - (int) tv->tv_sec*1000 + tv->tv_usec/1000); + (int) tvp->tv_sec*1000 + tvp->tv_usec/1000); #endif if (ERTS_POLLSET_SET_POLLED_CHK(ps)) { diff --git a/erts/emulator/sys/common/erl_sys_common_misc.c b/erts/emulator/sys/common/erl_sys_common_misc.c index e3ba741058..e63f0bda54 100644 --- a/erts/emulator/sys/common/erl_sys_common_misc.c +++ b/erts/emulator/sys/common/erl_sys_common_misc.c @@ -44,6 +44,14 @@ #endif #endif +/* + * erts_check_io_time is used by the erl_check_io implementation. The + * global erts_check_io_time variable is declared here since there + * (often) exist two versions of erl_check_io (kernel-poll and + * non-kernel-poll), and we dont want two versions of this variable. + */ +erts_smp_atomic_t erts_check_io_time; + /* Written once and only once */ static int filename_encoding = ERL_FILENAME_UNKNOWN; diff --git a/erts/emulator/sys/unix/erl_unix_sys.h b/erts/emulator/sys/unix/erl_unix_sys.h index 176fc049a7..26ed2fb558 100644 --- a/erts/emulator/sys/unix/erl_unix_sys.h +++ b/erts/emulator/sys/unix/erl_unix_sys.h @@ -135,9 +135,6 @@ /* File descriptors are numbers anc consecutively allocated on Unix */ #define ERTS_SYS_CONTINOUS_FD_NUMBERS -#define HAVE_ERTS_CHECK_IO_DEBUG -int erts_check_io_debug(void); - #ifndef ERTS_SMP # undef ERTS_POLL_NEED_ASYNC_INTERRUPT_SUPPORT # define ERTS_POLL_NEED_ASYNC_INTERRUPT_SUPPORT @@ -230,8 +227,24 @@ extern void sys_stop_cat(void); */ #ifdef USE_ISINF_ISNAN /* simulate finite() */ -# define finite(f) (!isinf(f) && !isnan(f)) -# define HAVE_FINITE +# define isfinite(f) (!isinf(f) && !isnan(f)) +# define HAVE_ISFINITE +#elif (defined(__GNUC__) && !defined(__llvm__)) && defined(HAVE_FINITE) +/* We use finite in gcc as it emits assembler instead of + the function call that isfinite emits. The assembler is + significantly faster. */ +# ifdef isfinite +# undef isfinite +# endif +# define isfinite finite +# ifndef HAVE_ISFINITE +# define HAVE_ISFINITE +# endif +#elif defined(isfinite) && !defined(HAVE_ISFINITE) +# define HAVE_ISFINITE +#elif !defined(HAVE_ISFINITE) && defined(HAVE_FINITE) +# define isfinite finite +# define HAVE_ISFINITE #endif #ifdef NO_FPE_SIGNALS @@ -241,7 +254,7 @@ extern void sys_stop_cat(void); #define erts_thread_init_fp_exception() do{}while(0) #endif # define __ERTS_FP_CHECK_INIT(fpexnp) do {} while (0) -# define __ERTS_FP_ERROR(fpexnp, f, Action) if (!finite(f)) { Action; } else {} +# define __ERTS_FP_ERROR(fpexnp, f, Action) if (!isfinite(f)) { Action; } else {} # define __ERTS_FP_ERROR_THOROUGH(fpexnp, f, Action) __ERTS_FP_ERROR(fpexnp, f, Action) # define __ERTS_SAVE_FP_EXCEPTION(fpexnp) # define __ERTS_RESTORE_FP_EXCEPTION(fpexnp) @@ -305,7 +318,7 @@ static __inline__ void __ERTS_FP_CHECK_INIT(volatile unsigned long *fp_exception code to always throw floating-point exceptions on errors. */ static __inline__ int erts_check_fpe_thorough(volatile unsigned long *fp_exception, double f) { - return erts_check_fpe(fp_exception, f) || !finite(f); + return erts_check_fpe(fp_exception, f) || !isfinite(f); } # define __ERTS_FP_ERROR_THOROUGH(fpexnp, f, Action) \ do { if (erts_check_fpe_thorough((fpexnp),(f))) { Action; } } while (0) diff --git a/erts/emulator/sys/unix/sys.c b/erts/emulator/sys/unix/sys.c index c3d7440409..0d677d5f34 100644 --- a/erts/emulator/sys/unix/sys.c +++ b/erts/emulator/sys/unix/sys.c @@ -284,7 +284,7 @@ struct { void (*check_io)(int); Uint (*size)(void); Eterm (*info)(void *); - int (*check_io_debug)(void); + int (*check_io_debug)(ErtsCheckIoDebugInfo *); } io_func = {0}; @@ -306,9 +306,9 @@ Eterm erts_check_io_info(void *p) } int -erts_check_io_debug(void) +erts_check_io_debug(ErtsCheckIoDebugInfo *ip) { - return (*io_func.check_io_debug)(); + return (*io_func.check_io_debug)(ip); } diff --git a/erts/emulator/sys/win32/erl_poll.c b/erts/emulator/sys/win32/erl_poll.c index 7a1d129cd5..972170d465 100644 --- a/erts/emulator/sys/win32/erl_poll.c +++ b/erts/emulator/sys/win32/erl_poll.c @@ -1085,7 +1085,7 @@ void erts_poll_controlv(ErtsPollSet ps, pcev[i].events, pcev[i].on); } - ERTS_POLLSET_LOCK(ps); + ERTS_POLLSET_UNLOCK(ps); HARDTRACEF(("Out erts_poll_controlv")); } diff --git a/erts/emulator/sys/win32/erl_win_sys.h b/erts/emulator/sys/win32/erl_win_sys.h index a78dbf64af..838f0c61eb 100644 --- a/erts/emulator/sys/win32/erl_win_sys.h +++ b/erts/emulator/sys/win32/erl_win_sys.h @@ -113,12 +113,10 @@ /* * Our own type of "FD's" */ +#define ERTS_SYS_FD_INVALID INVALID_HANDLE_VALUE #define ERTS_SYS_FD_TYPE HANDLE #define NO_FSTAT_ON_SYS_FD_TYPE 1 /* They are events, not files */ -#define HAVE_ERTS_CHECK_IO_DEBUG -int erts_check_io_debug(void); - /* * For erl_time_sup */ diff --git a/erts/emulator/test/a_SUITE.erl b/erts/emulator/test/a_SUITE.erl index 195c9c0a5f..17579be416 100644 --- a/erts/emulator/test/a_SUITE.erl +++ b/erts/emulator/test/a_SUITE.erl @@ -97,23 +97,13 @@ display_check_io(ChkIo) -> catch erlang:display('--- CHECK IO INFO ---'), catch erlang:display(ChkIo), catch erts_debug:set_internal_state(available_internal_state, true), - NoOfErrorFds = (catch erts_debug:get_internal_state(check_io_debug)), + NoOfErrorFds = (catch element(1, erts_debug:get_internal_state(check_io_debug))), catch erlang:display({'NoOfErrorFds', NoOfErrorFds}), catch erts_debug:set_internal_state(available_internal_state, false), catch erlang:display('--- CHECK IO INFO ---'), ok. get_check_io_info() -> - ChkIo = erlang:system_info(check_io), - case lists:keysearch(pending_updates, 1, ChkIo) of - {value, {pending_updates, 0}} -> - display_check_io(ChkIo), - ChkIo; - false -> - ChkIo; - _ -> - receive after 10 -> ok end, - get_check_io_info() - end. + z_SUITE:get_check_io_info(). diff --git a/erts/emulator/test/busy_port_SUITE.erl b/erts/emulator/test/busy_port_SUITE.erl index 4b4af0babe..2ed5aaa0d0 100644 --- a/erts/emulator/test/busy_port_SUITE.erl +++ b/erts/emulator/test/busy_port_SUITE.erl @@ -98,8 +98,10 @@ generator(0, Writer, _Data) -> %% Calling process_info(Pid, current_function) on a suspended process %% used to crash Beam. - {current_function, {erlang, send, 2}} = - process_info(Writer, current_function), + case process_info(Writer, [status,current_function]) of + [{status,suspended},{current_function,{erlang,send,2}}] -> ok; + [{status,suspended},{current_function,{erlang,bif_return_trap,_}}] -> ok + end, unlock_slave(); generator(N, Writer, Data) -> Writer ! {exec, Data}, diff --git a/erts/emulator/test/driver_SUITE.erl b/erts/emulator/test/driver_SUITE.erl index 344bde7c91..623d62f876 100644 --- a/erts/emulator/test/driver_SUITE.erl +++ b/erts/emulator/test/driver_SUITE.erl @@ -31,8 +31,9 @@ end_per_suite/1, init_per_group/2,end_per_group/2, init_per_testcase/2, end_per_testcase/2, + + a_test/1, outputv_echo/1, - timer_measure/1, timer_cancel/1, timer_change/1, @@ -79,7 +80,8 @@ thr_free_drv/1, async_blast/1, thr_msg_blast/1, - consume_timeslice/1]). + consume_timeslice/1, + z_test/1]). -export([bin_prefix/2]). @@ -122,19 +124,19 @@ init_per_testcase(Case, Config) when is_atom(Case), is_list(Config) -> _ -> erts_debug:set_internal_state(available_internal_state, true) end, erlang:display({init_per_testcase, Case}), - ?line 0 = erts_debug:get_internal_state(check_io_debug), + ?line 0 = element(1, erts_debug:get_internal_state(check_io_debug)), [{watchdog, Dog},{testcase, Case}|Config]. end_per_testcase(Case, Config) -> Dog = ?config(watchdog, Config), erlang:display({end_per_testcase, Case}), - ?line 0 = erts_debug:get_internal_state(check_io_debug), + ?line 0 = element(1, erts_debug:get_internal_state(check_io_debug)), ?t:timetrap_cancel(Dog). suite() -> [{ct_hooks,[ts_install_cth]}]. -all() -> - [outputv_errors, outputv_echo, queue_echo, {group, timer}, +all() -> %% Keep a_test first and z_test last... + [a_test, outputv_errors, outputv_echo, queue_echo, {group, timer}, driver_unloaded, io_ready_exit, use_fallback_pollset, bad_fd_in_pollset, driver_event, fd_change, steal_control, otp_6602, driver_system_info_base_ver, @@ -151,7 +153,8 @@ all() -> thr_free_drv, async_blast, thr_msg_blast, - consume_timeslice]. + consume_timeslice, + z_test]. groups() -> [{timer, [], @@ -917,8 +920,7 @@ steal_control_test(Hndl = {erts_poll_info, Before}) -> end. chkio_test_init(Config) when is_list(Config) -> - ?line wait_until_no_pending_updates(), - ?line ChkIo = erlang:system_info(check_io), + ?line ChkIo = get_stable_check_io_info(), ?line case catch lists:keysearch(name, 1, ChkIo) of {value, {name, erts_poll}} -> ?line ?t:format("Before test: ~p~n", [ChkIo]), @@ -937,8 +939,7 @@ chkio_test_fini({skipped, _} = Res) -> chkio_test_fini({chkio_test_result, Res, Before}) -> ?line ok = erl_ddll:unload_driver('chkio_drv'), ?line ok = erl_ddll:stop(), - ?line wait_until_no_pending_updates(), - ?line After = erlang:system_info(check_io), + ?line After = get_stable_check_io_info(), ?line ?t:format("After test: ~p~n", [After]), ?line verify_chkio_state(Before, After), ?line Res. @@ -985,7 +986,7 @@ chkio_test({erts_poll_info, Before}, ?line Fun(), ?line During = erlang:system_info(check_io), ?line erlang:display(During), - ?line 0 = erts_debug:get_internal_state(check_io_debug), + ?line 0 = element(1, erts_debug:get_internal_state(check_io_debug)), ?line ?t:format("During test: ~p~n", [During]), ?line chk_chkio_port(Port), ?line case erlang:port_control(Port, ?CHKIO_STOP, "") of @@ -1034,18 +1035,22 @@ verify_chkio_state(Before, After) -> After) end, ?line ok. - - -wait_until_no_pending_updates() -> - case lists:keysearch(pending_updates, 1, erlang:system_info(check_io)) of - {value, {pending_updates, 0}} -> - ok; - false -> - ok; +get_stable_check_io_info() -> + ChkIo = erlang:system_info(check_io), + PendUpdNo = case lists:keysearch(pending_updates, 1, ChkIo) of + {value, {pending_updates, PendNo}} -> + PendNo; + false -> + 0 + end, + {value, {active_fds, ActFds}} = lists:keysearch(active_fds, 1, ChkIo), + case {PendUpdNo, ActFds} of + {0, 0} -> + ChkIo; _ -> receive after 10 -> ok end, - wait_until_no_pending_updates() + get_stable_check_io_info() end. otp_6602(doc) -> ["Missed port lock when stealing control of fd from a " @@ -1084,7 +1089,15 @@ otp_6602(Config) when is_list(Config) -> ["async_thrs", "sched_thrs"])). --define(EXPECTED_SYSTEM_INFO_NAMES, ?EXPECTED_SYSTEM_INFO_NAMES2). +-define(EXPECTED_SYSTEM_INFO_NAMES3, + (?EXPECTED_SYSTEM_INFO_NAMES2 ++ + ["emu_nif_vsn"])). + +-define(EXPECTED_SYSTEM_INFO_NAMES4, + (?EXPECTED_SYSTEM_INFO_NAMES3 ++ + ["dirty_sched"])). + +-define(EXPECTED_SYSTEM_INFO_NAMES, ?EXPECTED_SYSTEM_INFO_NAMES4). 'driver_system_info_base_ver'(doc) -> []; @@ -1132,16 +1145,18 @@ check_driver_system_info_result(Result) -> drv_vsn_str2tup(erlang:system_info(driver_version))} of {DDVSN, DDVSN} -> ?line [] = Ns; - {{1, 0}, _} -> - ?line ExpNs = lists:sort(?EXPECTED_SYSTEM_INFO_NAMES - -- ?EXPECTED_SYSTEM_INFO_NAMES1), - ?line ExpNs = lists:sort(Ns); - {{1, 1}, _} -> + %% {{1, 0}, _} -> + %% ?line ExpNs = lists:sort(?EXPECTED_SYSTEM_INFO_NAMES + %% -- ?EXPECTED_SYSTEM_INFO_NAMES1), + %% ?line ExpNs = lists:sort(Ns); + %% {{1, 1}, _} -> + %% ?line ExpNs = lists:sort(?EXPECTED_SYSTEM_INFO_NAMES + %% -- ?EXPECTED_SYSTEM_INFO_NAMES2), + %% ?line ExpNs = lists:sort(Ns); + {{3, 0}, _} -> ?line ExpNs = lists:sort(?EXPECTED_SYSTEM_INFO_NAMES - -- ?EXPECTED_SYSTEM_INFO_NAMES2), - ?line ExpNs = lists:sort(Ns); - {{2, 0}, _} -> - ?line [] = Ns + -- ?EXPECTED_SYSTEM_INFO_NAMES3), + ?line ExpNs = lists:sort(Ns) end. chk_sis(SIs, Ns) -> @@ -1188,6 +1203,14 @@ check_si_res(["async_thrs", Value]) -> check_si_res(["sched_thrs", Value]) -> ?line Value = integer_to_list(erlang:system_info(schedulers)); +%% Data added in 3rd version of driver_system_info() (driver version 1.5) +check_si_res(["emu_nif_vsn", Value]) -> + ?line Value = erlang:system_info(nif_version); + +%% Data added in 4th version of driver_system_info() (driver version 3.1) +check_si_res(["dirty_sched", _Value]) -> + true; + check_si_res(Unexpected) -> ?line ?t:fail({unexpected_result, Unexpected}). @@ -2369,10 +2392,25 @@ count_proc_sched(Ps, PNs) -> PNs end. +a_test(Config) when is_list(Config) -> + check_io_debug(). + +z_test(Config) when is_list(Config) -> + check_io_debug(). + %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %% Utilities %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% +check_io_debug() -> + get_stable_check_io_info(), + {NoErrorFds, NoUsedFds, NoDrvSelStructs, NoDrvEvStructs} + = erts_debug:get_internal_state(check_io_debug), + 0 = NoErrorFds, + NoUsedFds = NoDrvSelStructs, + 0 = NoDrvEvStructs, + ok. + %flush_msgs() -> % receive % M -> diff --git a/erts/emulator/test/driver_SUITE_data/sys_info_base_drv.c b/erts/emulator/test/driver_SUITE_data/sys_info_base_drv.c index e44c7dbd5e..964034f5a6 100644 --- a/erts/emulator/test/driver_SUITE_data/sys_info_base_drv.c +++ b/erts/emulator/test/driver_SUITE_data/sys_info_base_drv.c @@ -41,7 +41,9 @@ "thread=%s " \ "smp=%s " \ "async_thrs=%d " \ - "sched_thrs=%d" + "sched_thrs=%d " \ + "emu_nif_vsn=%d.%d" + static size_t sys_info_drv_max_res_len(ErlDrvSysInfo *sip) @@ -55,6 +57,7 @@ sys_info_drv_max_res_len(ErlDrvSysInfo *sip) slen += 5; /* smp */ slen += 20; /* async_thrs */ slen += 20; /* sched_thrs */ + slen += 2*20; /* emu_nif_vsn */ return slen; } @@ -72,7 +75,9 @@ sys_info_drv_sprintf_sys_info(ErlDrvSysInfo *sip, char *str) sip->thread_support ? "true" : "false", sip->smp_support ? "true" : "false", sip->async_threads, - sip->scheduler_threads); + sip->scheduler_threads, + sip->nif_major_version, + sip->nif_minor_version); } #include "sys_info_drv_impl.c" diff --git a/erts/emulator/test/driver_SUITE_data/sys_info_curr_drv.c b/erts/emulator/test/driver_SUITE_data/sys_info_curr_drv.c index 5bbc966932..6d2c47fdaf 100644 --- a/erts/emulator/test/driver_SUITE_data/sys_info_curr_drv.c +++ b/erts/emulator/test/driver_SUITE_data/sys_info_curr_drv.c @@ -40,7 +40,9 @@ "thread=%s " \ "smp=%s " \ "async_thrs=%d " \ - "sched_thrs=%d" + "sched_thrs=%d " \ + "emu_nif_vsn=%d.%d " \ + "dirty_sched=%s" static size_t sys_info_drv_max_res_len(ErlDrvSysInfo *sip) @@ -54,6 +56,8 @@ sys_info_drv_max_res_len(ErlDrvSysInfo *sip) slen += 5; /* smp */ slen += 20; /* async_thrs */ slen += 20; /* sched_thrs */ + slen += 2*20; /* emu_nif_vsn */ + slen += 5; /* dirty_sched */ return slen; } @@ -71,7 +75,10 @@ sys_info_drv_sprintf_sys_info(ErlDrvSysInfo *sip, char *str) sip->thread_support ? "true" : "false", sip->smp_support ? "true" : "false", sip->async_threads, - sip->scheduler_threads); + sip->scheduler_threads, + sip->nif_major_version, + sip->nif_minor_version, + sip->dirty_scheduler_support ? "true" : "false"); } #include "sys_info_drv_impl.c" diff --git a/erts/emulator/test/driver_SUITE_data/sys_info_prev_drv.c b/erts/emulator/test/driver_SUITE_data/sys_info_prev_drv.c index 63c69f751c..2271d7027b 100644 --- a/erts/emulator/test/driver_SUITE_data/sys_info_prev_drv.c +++ b/erts/emulator/test/driver_SUITE_data/sys_info_prev_drv.c @@ -41,7 +41,9 @@ "thread=%s " \ "smp=%s " \ "async_thrs=%d " \ - "sched_thrs=%d" + "sched_thrs=%d " \ + "emu_nif_vsn=%d.%d" + static size_t sys_info_drv_max_res_len(ErlDrvSysInfo *sip) @@ -55,6 +57,7 @@ sys_info_drv_max_res_len(ErlDrvSysInfo *sip) slen += 5; /* smp */ slen += 20; /* async_thrs */ slen += 20; /* sched_thrs */ + slen += 2*20; /* emu_nif_vsn */ return slen; } @@ -72,7 +75,9 @@ sys_info_drv_sprintf_sys_info(ErlDrvSysInfo *sip, char *str) sip->thread_support ? "true" : "false", sip->smp_support ? "true" : "false", sip->async_threads, - sip->scheduler_threads); + sip->scheduler_threads, + sip->nif_major_version, + sip->nif_minor_version); } #include "sys_info_drv_impl.c" diff --git a/erts/emulator/test/float_SUITE_data/fp_drv.c b/erts/emulator/test/float_SUITE_data/fp_drv.c index b80385c3f9..82d18d6440 100644 --- a/erts/emulator/test/float_SUITE_data/fp_drv.c +++ b/erts/emulator/test/float_SUITE_data/fp_drv.c @@ -29,9 +29,14 @@ #if defined (__GNUC__) int _finite(double x); #endif -#ifndef finite -#define finite _finite +#ifndef isfinite +#define isfinite _finite #endif +#elif !defined(HAVE_ISFINITE) && defined(HAVE_FINITE) +/* If not windows and we do not have isfinite */ +#define isfinite finite +#elif !defined(HAVE_ISFINITE) +# error "No finite function found!" #endif #include "erl_driver.h" @@ -79,21 +84,21 @@ do_test(void *unused) x = 3.23e133; y = 3.57e257; z = x*y; - if (finite(z)) + if (isfinite(z)) return "is finite (1)"; x = 5.0; y = 0.0; z = x/y; - if (finite(z)) + if (isfinite(z)) return "is finite (2)"; z = log(-1.0); - if (finite(z)) + if (isfinite(z)) return "is finite (3)"; z = log(0.0); - if (finite(z)) + if (isfinite(z)) return "is finite (4)"; return "ok"; diff --git a/erts/emulator/test/fun_SUITE.erl b/erts/emulator/test/fun_SUITE.erl index 8ad5f290ed..2968f5bebb 100644 --- a/erts/emulator/test/fun_SUITE.erl +++ b/erts/emulator/test/fun_SUITE.erl @@ -30,7 +30,7 @@ fun_to_port/1,t_hash/1,t_phash/1,t_phash2/1,md5/1, refc/1,refc_ets/1,refc_dist/1, const_propagation/1,t_arity/1,t_is_function2/1, - t_fun_info/1]). + t_fun_info/1,t_fun_info_mfa/1]). -export([nothing/0]). @@ -42,7 +42,8 @@ all() -> [bad_apply, bad_fun_call, badarity, ext_badarity, equality, ordering, fun_to_port, t_hash, t_phash, t_phash2, md5, refc, refc_ets, refc_dist, - const_propagation, t_arity, t_is_function2, t_fun_info]. + const_propagation, t_arity, t_is_function2, t_fun_info, + t_fun_info_mfa]. groups() -> []. @@ -824,6 +825,24 @@ t_fun_info(Config) when is_list(Config) -> ?line bad_info(<<1,2>>), ok. +t_fun_info_mfa(Config) when is_list(Config) -> + Fun1 = fun spawn_call/2, + {module,M1} = erlang:fun_info(Fun1, module), + {name,F1} = erlang:fun_info(Fun1, name), + {arity,A1} = erlang:fun_info(Fun1, arity), + {M1,F1,A1=2} = erlang:fun_info_mfa(Fun1), + %% Module fun. + Fun2 = fun ?MODULE:t_fun_info/1, + {module,M2} = erlang:fun_info(Fun2, module), + {name,F2} = erlang:fun_info(Fun2, name), + {arity,A2} = erlang:fun_info(Fun2, arity), + {M2,F2,A2=1} = erlang:fun_info_mfa(Fun2), + + %% Not fun. + {'EXIT',_} = (catch erlang:fun_info_mfa(id(d))), + ok. + + bad_info(Term) -> try erlang:fun_info(Term, module) of Any -> diff --git a/erts/emulator/test/nif_SUITE.erl b/erts/emulator/test/nif_SUITE.erl index b2da6f58af..4560077a51 100644 --- a/erts/emulator/test/nif_SUITE.erl +++ b/erts/emulator/test/nif_SUITE.erl @@ -37,7 +37,9 @@ threading/1, send/1, send2/1, send3/1, send_threaded/1, neg/1, is_checks/1, get_length/1, make_atom/1, make_string/1, reverse_list_test/1, - otp_9668/1, consume_timeslice/1, dirty_nif/1, dirty_nif_send/1 + otp_9828/1, + otp_9668/1, consume_timeslice/1, dirty_nif/1, dirty_nif_send/1, + dirty_nif_exception/1, nif_schedule/1 ]). -export([many_args_100/100]). @@ -64,7 +66,9 @@ all() -> resource_takeover, threading, send, send2, send3, send_threaded, neg, is_checks, get_length, make_atom, make_string,reverse_list_test, - otp_9668, consume_timeslice, dirty_nif, dirty_nif_send + otp_9828, + otp_9668, consume_timeslice, + nif_schedule, dirty_nif, dirty_nif_send, dirty_nif_exception ]. groups() -> @@ -1440,6 +1444,20 @@ otp_9668(Config) -> ?line verify_tmpmem(TmpMem), ok. +otp_9828(doc) -> ["Copy of writable binary"]; +otp_9828(Config) -> + ensure_lib_loaded(Config, 1), + + otp_9828_loop(<<"I'm alive!">>, 1000). + +otp_9828_loop(Bin, 0) -> + ok; +otp_9828_loop(Bin, Val) -> + WrtBin = <<Bin/binary, Val:32>>, + ok = otp_9828_nif(WrtBin), + otp_9828_loop(WrtBin, Val-1). + + consume_timeslice(Config) when is_list(Config) -> CONTEXT_REDS = 2000, Me = self(), @@ -1524,6 +1542,20 @@ consume_timeslice(Config) when is_list(Config) -> ok. +nif_schedule(Config) when is_list(Config) -> + ensure_lib_loaded(Config), + A = "this is a string", + B = {this,is,a,tuple}, + {B,A} = call_nif_schedule(A, B), + ok = try call_nif_schedule(1, 2) + catch + error:badarg -> + [{?MODULE,call_nif_schedule,[1,2],_}|_] = + erlang:get_stacktrace(), + ok + end, + ok. + dirty_nif(Config) when is_list(Config) -> try erlang:system_info(dirty_cpu_schedulers) of N when is_integer(N) -> @@ -1532,6 +1564,8 @@ dirty_nif(Config) when is_list(Config) -> Val2 = "Erlang", Val3 = list_to_binary([Val2, 0]), {Val1, Val2, Val3} = call_dirty_nif(Val1, Val2, Val3), + LargeArray = lists:duplicate(1000, ok), + LargeArray = call_dirty_nif_zero_args(), ok catch error:badarg -> @@ -1556,6 +1590,24 @@ dirty_nif_send(Config) when is_list(Config) -> {skipped,"No dirty scheduler support"} end. +dirty_nif_exception(Config) when is_list(Config) -> + try erlang:system_info(dirty_cpu_schedulers) of + N when is_integer(N) -> + ensure_lib_loaded(Config), + try + call_dirty_nif_exception(), + ?t:fail(expected_badarg) + catch + error:badarg -> + [{?MODULE,call_dirty_nif_exception,[],_}|_] = + erlang:get_stacktrace(), + ok + end + catch + error:badarg -> + {skipped,"No dirty scheduler support"} + end. + next_msg(_Pid) -> receive M -> M @@ -1684,9 +1736,13 @@ reverse_list(_) -> ?nif_stub. echo_int(_) -> ?nif_stub. type_sizes() -> ?nif_stub. otp_9668_nif(_) -> ?nif_stub. +otp_9828_nif(_) -> ?nif_stub. consume_timeslice_nif(_,_) -> ?nif_stub. +call_nif_schedule(_,_) -> ?nif_stub. call_dirty_nif(_,_,_) -> ?nif_stub. send_from_dirty_nif(_) -> ?nif_stub. +call_dirty_nif_exception() -> ?nif_stub. +call_dirty_nif_zero_args() -> ?nif_stub. %% maps is_map_nif(_) -> ?nif_stub. diff --git a/erts/emulator/test/nif_SUITE_data/nif_SUITE.c b/erts/emulator/test/nif_SUITE_data/nif_SUITE.c index 955dc64189..85544db2ab 100644 --- a/erts/emulator/test/nif_SUITE_data/nif_SUITE.c +++ b/erts/emulator/test/nif_SUITE_data/nif_SUITE.c @@ -1473,6 +1473,26 @@ static ERL_NIF_TERM otp_9668_nif(ErlNifEnv* env, int argc, const ERL_NIF_TERM ar return atom_ok; } +static ERL_NIF_TERM otp_9828_nif(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]) +{ + /* copy a writable binary could reallocate it due to "emasculation" + and thereby render a previous inspection invalid. + */ + ErlNifBinary bin1; + ErlNifEnv* myenv; + + if (!enif_inspect_binary(env, argv[0], &bin1)) { + return enif_make_badarg(env); + } + + myenv = enif_alloc_env(); + enif_make_copy(myenv, argv[0]); + enif_free_env(myenv); + + return memcmp(bin1.data, "I'm alive!", 10)==0 ? atom_ok : atom_false; +} + + static ERL_NIF_TERM consume_timeslice_nif(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]) { int percent; @@ -1493,25 +1513,57 @@ static ERL_NIF_TERM consume_timeslice_nif(ErlNifEnv* env, int argc, const ERL_NI } } +static ERL_NIF_TERM nif_sched2(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]) +{ + char s[64]; + if (!enif_get_string(env, argv[2], s, sizeof s, ERL_NIF_LATIN1)) + return enif_make_badarg(env); + return enif_make_tuple2(env, argv[3], argv[2]); +} + +static ERL_NIF_TERM nif_sched1(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]) +{ + ERL_NIF_TERM new_argv[4]; + new_argv[0] = enif_make_atom(env, "garbage0"); + new_argv[1] = enif_make_atom(env, "garbage1"); + new_argv[2] = argv[0]; + new_argv[3] = argv[1]; + return enif_schedule_nif(env, "nif_sched2", 0, nif_sched2, 4, new_argv); +} + +static ERL_NIF_TERM call_nif_schedule(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]) +{ + if (argc != 2) + return enif_make_atom(env, "false"); + return enif_schedule_nif(env, "nif_sched1", 0, nif_sched1, argc, argv); +} + #ifdef ERL_NIF_DIRTY_SCHEDULER_SUPPORT + +static int have_dirty_schedulers(void) +{ + ErlNifSysInfo si; + enif_system_info(&si, sizeof(si)); + return si.dirty_scheduler_support; +} + static ERL_NIF_TERM dirty_nif(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]) { int n; char s[10]; ErlNifBinary b; ERL_NIF_TERM result; - if (enif_have_dirty_schedulers()) { + if (have_dirty_schedulers()) { assert(enif_is_on_dirty_scheduler(env)); } assert(argc == 3); enif_get_int(env, argv[0], &n); enif_get_string(env, argv[1], s, sizeof s, ERL_NIF_LATIN1); enif_inspect_binary(env, argv[2], &b); - result = enif_make_tuple3(env, - enif_make_int(env, n), - enif_make_string(env, s, ERL_NIF_LATIN1), - enif_make_binary(env, &b)); - return enif_schedule_dirty_nif_finalizer(env, result, enif_dirty_nif_finalizer); + return enif_make_tuple3(env, + enif_make_int(env, n), + enif_make_string(env, s, ERL_NIF_LATIN1), + enif_make_binary(env, &b)); } static ERL_NIF_TERM call_dirty_nif(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]) @@ -1522,11 +1574,11 @@ static ERL_NIF_TERM call_dirty_nif(ErlNifEnv* env, int argc, const ERL_NIF_TERM assert(!enif_is_on_dirty_scheduler(env)); if (argc != 3) return enif_make_badarg(env); - if (enif_have_dirty_schedulers()) { + if (have_dirty_schedulers()) { if (enif_get_int(env, argv[0], &n) && enif_get_string(env, argv[1], s, sizeof s, ERL_NIF_LATIN1) && enif_inspect_binary(env, argv[2], &b)) - return enif_schedule_dirty_nif(env, ERL_NIF_DIRTY_JOB_CPU_BOUND, dirty_nif, argc, argv); + return enif_schedule_nif(env, "call_dirty_nif", ERL_NIF_DIRTY_JOB_CPU_BOUND, dirty_nif, argc, argv); else return enif_make_badarg(env); } else { @@ -1534,35 +1586,54 @@ static ERL_NIF_TERM call_dirty_nif(ErlNifEnv* env, int argc, const ERL_NIF_TERM } } -static ERL_NIF_TERM dirty_sender(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]) +static ERL_NIF_TERM send_from_dirty_nif(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]) { ERL_NIF_TERM result; ErlNifPid pid; ErlNifEnv* menv; int res; - enif_get_local_pid(env, argv[0], &pid); + if (!enif_get_local_pid(env, argv[0], &pid)) + return enif_make_badarg(env); result = enif_make_tuple2(env, enif_make_atom(env, "ok"), enif_make_pid(env, &pid)); menv = enif_alloc_env(); res = enif_send(env, &pid, menv, result); enif_free_env(menv); if (!res) - /* Note the next line will crash, since dirty nifs can't return exceptions. - * This is intentional, since enif_send should not fail if the test succeeds. - */ - return enif_schedule_dirty_nif_finalizer(env, enif_make_badarg(env), enif_dirty_nif_finalizer); + return enif_make_badarg(env); else - return enif_schedule_dirty_nif_finalizer(env, result, enif_dirty_nif_finalizer); + return result; } -static ERL_NIF_TERM send_from_dirty_nif(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]) +static ERL_NIF_TERM call_dirty_nif_exception(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]) { - ERL_NIF_TERM result; - ErlNifPid pid; - - if (!enif_get_local_pid(env, argv[0], &pid)) + switch (argc) { + case 0: { + ERL_NIF_TERM args[255]; + int i; + for (i = 0; i < 255; i++) + args[i] = enif_make_int(env, i); + return enif_schedule_nif(env, "call_dirty_nif_exception", ERL_NIF_DIRTY_JOB_CPU_BOUND, + call_dirty_nif_exception, 255, argv); + } + case 1: return enif_make_badarg(env); - return enif_schedule_dirty_nif(env, ERL_NIF_DIRTY_JOB_CPU_BOUND, dirty_sender, argc, argv); + default: + return enif_schedule_nif(env, "call_dirty_nif_exception", ERL_NIF_DIRTY_JOB_CPU_BOUND, + call_dirty_nif_exception, argc-1, argv); + } +} + +static ERL_NIF_TERM call_dirty_nif_zero_args(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]) +{ + int i; + ERL_NIF_TERM result[1000]; + ERL_NIF_TERM ok = enif_make_atom(env, "ok"); + assert(argc == 0); + for (i = 0; i < sizeof(result)/sizeof(*result); i++) { + result[i] = ok; + } + return enif_make_list_from_array(env, result, i); } #endif @@ -1741,10 +1812,14 @@ static ErlNifFunc nif_funcs[] = {"echo_int", 1, echo_int}, {"type_sizes", 0, type_sizes}, {"otp_9668_nif", 1, otp_9668_nif}, + {"otp_9828_nif", 1, otp_9828_nif}, {"consume_timeslice_nif", 2, consume_timeslice_nif}, + {"call_nif_schedule", 2, call_nif_schedule}, #ifdef ERL_NIF_DIRTY_SCHEDULER_SUPPORT {"call_dirty_nif", 3, call_dirty_nif}, - {"send_from_dirty_nif", 1, send_from_dirty_nif}, + {"send_from_dirty_nif", 1, send_from_dirty_nif, ERL_NIF_DIRTY_JOB_CPU_BOUND}, + {"call_dirty_nif_exception", 0, call_dirty_nif_exception, ERL_NIF_DIRTY_JOB_IO_BOUND}, + {"call_dirty_nif_zero_args", 0, call_dirty_nif_zero_args, ERL_NIF_DIRTY_JOB_CPU_BOUND}, #endif {"is_map_nif", 1, is_map_nif}, {"get_map_size_nif", 1, get_map_size_nif}, diff --git a/erts/emulator/test/num_bif_SUITE.erl b/erts/emulator/test/num_bif_SUITE.erl index ff8d18eef8..8cf8377c30 100644 --- a/erts/emulator/test/num_bif_SUITE.erl +++ b/erts/emulator/test/num_bif_SUITE.erl @@ -1,7 +1,7 @@ %% %% %CopyrightBegin% %% -%% Copyright Ericsson AB 1997-2013. All Rights Reserved. +%% Copyright Ericsson AB 1997-2014. 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 @@ -394,18 +394,15 @@ t_string_to_integer(Config) when is_list(Config) -> test_sti(268435455), test_sti(-268435455), - %% 1 bsl 28 - 1, just before 32 bit bignum - test_sti(1 bsl 28 - 1), - %% 1 bsl 28, just beyond 32 bit small - test_sti(1 bsl 28), - %% 1 bsl 33, just beyond 32 bit - test_sti(1 bsl 33), - %% 1 bsl 60 - 1, just before 64 bit bignum - test_sti(1 bsl 60 - 1), - %% 1 bsl 60, just beyond 64 bit small - test_sti(1 bsl 60), - %% 1 bsl 65, just beyond 64 bit - test_sti(1 bsl 65), + % Interesting values around 2-pows, such as MIN_SMALL and MAX_SMALL. + lists:foreach(fun(Bits) -> + N = 1 bsl Bits, + test_sti(N - 1), + test_sti(N), + test_sti(N + 1) + end, + lists:seq(16, 130)), + %% Bignums. test_sti(123456932798748738738,16), test_sti(list_to_integer(lists:duplicate(2000, $1))), @@ -454,10 +451,11 @@ test_sti(Num) -> end|| Base <- lists:seq(2,36)]. test_sti(Num,Base) -> - Num = list_to_integer(int2list(Num,Base),Base), - Num = -1*list_to_integer(int2list(Num*-1,Base),Base), - Num = binary_to_integer(int2bin(Num,Base),Base), - Num = -1*binary_to_integer(int2bin(Num*-1,Base),Base). + Neg = -Num, + Num = list_to_integer(int2list(Num,Base),Base), + Neg = list_to_integer(int2list(Num*-1,Base),Base), + Num = binary_to_integer(int2bin(Num,Base),Base), + Neg = binary_to_integer(int2bin(Num*-1,Base),Base). % Calling this function (which is not supposed to be inlined) prevents % the compiler from calculating the answer, so we don't test the compiler diff --git a/erts/emulator/test/port_SUITE.erl b/erts/emulator/test/port_SUITE.erl index e01b2f253b..9083545060 100644 --- a/erts/emulator/test/port_SUITE.erl +++ b/erts/emulator/test/port_SUITE.erl @@ -90,6 +90,7 @@ mix_up_ports/1, otp_5112/1, otp_5119/1, otp_6224/1, exit_status_multi_scheduling_block/1, ports/1, spawn_driver/1, spawn_executable/1, close_deaf_port/1, + port_setget_data/1, unregister_name/1, parallelism_option/1]). -export([do_iter_max_ports/2]). @@ -115,6 +116,7 @@ all() -> mix_up_ports, otp_5112, otp_5119, exit_status_multi_scheduling_block, ports, spawn_driver, spawn_executable, close_deaf_port, unregister_name, + port_setget_data, parallelism_option]. groups() -> @@ -2333,6 +2335,61 @@ close_deaf_port_1(N, Cmd) -> {comment, "Could not spawn more than " ++ integer_to_list(N) ++ " OS processes."} end. +%% Test undocumented port_set_data/2 and port_get_data/1 +%% Hammer from multiple processes a while +%% and then abrubtly close the port (OTP-12208). +port_setget_data(Config) when is_list(Config) -> + ok = load_driver(?config(data_dir, Config), "echo_drv"), + Port = erlang:open_port({spawn_driver, "echo_drv"}, []), + + NSched = erlang:system_info(schedulers_online), + HeapData = {1,2,3,<<"A heap binary">>,fun()->"This is fun"end, + list_to_binary(lists:seq(1,100))}, + PRs = lists:map(fun(I) -> + spawn_opt(fun() -> port_setget_data_hammer(Port,HeapData,false,1) end, + [monitor, {scheduler, I rem NSched}]) + end, + lists:seq(1,10)), + receive after 100 -> ok end, + Papa = self(), + lists:foreach(fun({Pid,_}) -> Pid ! {Papa,prepare_for_close} end, PRs), + lists:foreach(fun({Pid,_}) -> + receive {Pid,prepare_for_close} -> ok end + end, + PRs), + port_close(Port), + lists:foreach(fun({Pid,Ref}) -> + receive {'DOWN', Ref, process, Pid, normal} -> ok end + end, + PRs), + ok. + +port_setget_data_hammer(Port, HeapData, IsSet0, N) -> + Rand = random:uniform(3), + IsSet1 = try case Rand of + 1 -> true = erlang:port_set_data(Port, atom), true; + 2 -> true = erlang:port_set_data(Port, HeapData), true; + 3 -> case erlang:port_get_data(Port) of + atom -> true; + HeapData -> true; + undefined -> false=IsSet0 + end + end + catch + error:badarg -> + true = get(prepare_for_close), + io:format("~p did ~p rounds before port closed\n", [self(), N]), + exit(normal) + end, + receive {Papa, prepare_for_close} -> + put(prepare_for_close, true), + Papa ! {self(),prepare_for_close} + after 0 -> + ok + end, + port_setget_data_hammer(Port, HeapData, IsSet1, N+1). + + wait_until(Fun) -> case catch Fun() of true -> diff --git a/erts/emulator/test/tuple_SUITE.erl b/erts/emulator/test/tuple_SUITE.erl index 46ece41096..f627eea07f 100644 --- a/erts/emulator/test/tuple_SUITE.erl +++ b/erts/emulator/test/tuple_SUITE.erl @@ -21,8 +21,9 @@ init_per_group/2,end_per_group/2, t_size/1, t_tuple_size/1, t_element/1, t_setelement/1, t_insert_element/1, t_delete_element/1, - t_list_to_tuple/1, t_tuple_to_list/1, - t_make_tuple_2/1, t_make_tuple_3/1, t_append_element/1, + t_list_to_tuple/1, t_list_to_upper_boundry_tuple/1, t_tuple_to_list/1, + t_make_tuple_2/1, t_make_upper_boundry_tuple_2/1, t_make_tuple_3/1, + t_append_element/1, t_append_element_upper_boundry/1, build_and_match/1, tuple_with_case/1, tuple_in_guard/1]). -include_lib("test_server/include/test_server.hrl"). @@ -40,8 +41,10 @@ suite() -> [{ct_hooks,[ts_install_cth]}]. all() -> [build_and_match, t_size, t_tuple_size, t_list_to_tuple, + t_list_to_upper_boundry_tuple, t_tuple_to_list, t_element, t_setelement, - t_make_tuple_2, t_make_tuple_3, t_append_element, + t_make_tuple_2, t_make_upper_boundry_tuple_2, t_make_tuple_3, + t_append_element, t_append_element_upper_boundry, t_insert_element, t_delete_element, tuple_with_case, tuple_in_guard]. @@ -49,11 +52,21 @@ groups() -> []. init_per_suite(Config) -> + A0 = case application:start(sasl) of + ok -> [sasl]; + _ -> [] + end, + A = case application:start(os_mon) of + ok -> [os_mon|A0]; + _ -> A0 + end, + [{started_apps, A}|Config]. + +end_per_suite(Config) -> + As = ?config(started_apps, Config), + lists:foreach(fun (A) -> application:stop(A) end, As), Config. -end_per_suite(_Config) -> - ok. - init_per_group(_GroupName, Config) -> Config. @@ -176,14 +189,19 @@ t_list_to_tuple(Config) when is_list(Config) -> {'EXIT', {badarg, _}} = (catch list_to_tuple(id([a|b]))), {'EXIT', {badarg, _}} = (catch list_to_tuple(id([a|b]))), - % test upper boundry, 16777215 elements - MaxSize = 1 bsl 24 - 1, - MaxTuple = list_to_tuple(lists:seq(1, MaxSize)), - MaxSize = size(MaxTuple), - {'EXIT', {badarg,_}} = (catch list_to_tuple(lists:seq(1, 1 bsl 24))), ok. +t_list_to_upper_boundry_tuple(Config) when is_list(Config) -> + sys_mem_cond_run(2048, + fun () -> + %% test upper boundry, 16777215 elements + MaxSize = 1 bsl 24 - 1, + MaxTuple = list_to_tuple(lists:seq(1, MaxSize)), + MaxSize = size(MaxTuple), + ok + end). + %% Tests tuple_to_list/1. t_tuple_to_list(Config) when is_list(Config) -> @@ -214,8 +232,6 @@ t_make_tuple_2(Config) when is_list(Config) -> t_make_tuple1({a}), t_make_tuple1(erlang:make_tuple(400, [])), - % test upper boundry, 16777215 elements - t_make_tuple(1 bsl 24 - 1, a), {'EXIT', {badarg,_}} = (catch erlang:make_tuple(1 bsl 24, a)), {'EXIT', {badarg,_}} = (catch erlang:make_tuple(-1, a)), @@ -225,6 +241,13 @@ t_make_tuple_2(Config) when is_list(Config) -> {'EXIT', {badarg,_}} = (catch erlang:make_tuple(1 bsl 65 + 3, a)), ok. +t_make_upper_boundry_tuple_2(Config) when is_list(Config) -> + sys_mem_cond_run(2048, + fun () -> + %% test upper boundry, 16777215 elements + t_make_tuple(1 bsl 24 - 1, a) + end). + t_make_tuple1(Element) -> lists:foreach(fun(Size) -> t_make_tuple(Size, Element) end, [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 255, 256, 511, 512, 999, @@ -309,13 +332,17 @@ t_delete_element(Config) when is_list(Config) -> %% Tests the append_element/2 BIF. t_append_element(Config) when is_list(Config) -> - ok = t_append_element({}, 2048, 2048), - - % test upper boundry, 16777215 elements - MaxSize = 1 bsl 24 - 1, - MaxTuple = list_to_tuple(lists:seq(1, MaxSize)), - {'EXIT',{badarg,_}} = (catch erlang:append_element(MaxTuple, a)), - ok. + ok = t_append_element({}, 2048, 2048). + +t_append_element_upper_boundry(Config) when is_list(Config) -> + sys_mem_cond_run(2048, + fun () -> + %% test upper boundry, 16777215 elements + MaxSize = 1 bsl 24 - 1, + MaxTuple = list_to_tuple(lists:seq(1, MaxSize)), + {'EXIT',{badarg,_}} = (catch erlang:append_element(MaxTuple, a)), + ok + end). t_append_element(_Tuple, 0, _High) -> ok; t_append_element(Tuple, N, High) -> @@ -371,3 +398,31 @@ tuple_in_guard(Config) when is_list(Config) -> %% Use this function to avoid compile-time evaluation of an expression. id(I) -> I. + +sys_mem_cond_run(ReqSizeMB, TestFun) when is_integer(ReqSizeMB) -> + case total_memory() of + TotMem when is_integer(TotMem), TotMem >= ReqSizeMB -> + TestFun(); + TotMem when is_integer(TotMem) -> + {skipped, "Not enough memory ("++integer_to_list(TotMem)++" MB)"}; + undefined -> + {skipped, "Could not retrieve memory information"} + end. + + +total_memory() -> + %% Totat memory in MB. + try + MemoryData = memsup:get_system_memory_data(), + case lists:keysearch(total_memory, 1, MemoryData) of + {value, {total_memory, TM}} -> + TM div (1024*1024); + false -> + {value, {system_total_memory, STM}} = + lists:keysearch(system_total_memory, 1, MemoryData), + STM div (1024*1024) + end + catch + _ : _ -> + undefined + end. diff --git a/erts/emulator/test/z_SUITE.erl b/erts/emulator/test/z_SUITE.erl index 4b3075a164..b0c6224dfe 100644 --- a/erts/emulator/test/z_SUITE.erl +++ b/erts/emulator/test/z_SUITE.erl @@ -38,7 +38,7 @@ -export([schedulers_alive/1, node_container_refc_check/1, long_timers/1, pollset_size/1, - check_io_debug/1]). + check_io_debug/1, get_check_io_info/0]). -define(DEFAULT_TIMEOUT, ?t:minutes(5)). @@ -288,11 +288,14 @@ check_io_debug(Config) when is_list(Config) -> end. check_io_debug_test() -> + ?line erlang:display(get_check_io_info()), ?line erts_debug:set_internal_state(available_internal_state, true), - ?line erlang:display(erlang:system_info(check_io)), - ?line NoOfErrorFds = erts_debug:get_internal_state(check_io_debug), + ?line {NoErrorFds, NoUsedFds, NoDrvSelStructs, NoDrvEvStructs} + = erts_debug:get_internal_state(check_io_debug), ?line erts_debug:set_internal_state(available_internal_state, false), - ?line 0 = NoOfErrorFds, + ?line 0 = NoErrorFds, + ?line NoUsedFds = NoDrvSelStructs, + ?line 0 = NoDrvEvStructs, ?line ok. @@ -305,7 +308,7 @@ display_check_io(ChkIo) -> catch erlang:display('--- CHECK IO INFO ---'), catch erlang:display(ChkIo), catch erts_debug:set_internal_state(available_internal_state, true), - NoOfErrorFds = (catch erts_debug:get_internal_state(check_io_debug)), + NoOfErrorFds = (catch element(1, erts_debug:get_internal_state(check_io_debug))), catch erlang:display({'NoOfErrorFds', NoOfErrorFds}), catch erts_debug:set_internal_state(available_internal_state, false), catch erlang:display('--- CHECK IO INFO ---'), @@ -313,14 +316,19 @@ display_check_io(ChkIo) -> get_check_io_info() -> ChkIo = erlang:system_info(check_io), - case lists:keysearch(pending_updates, 1, ChkIo) of - {value, {pending_updates, 0}} -> + PendUpdNo = case lists:keysearch(pending_updates, 1, ChkIo) of + {value, {pending_updates, PendNo}} -> + PendNo; + false -> + 0 + end, + {value, {active_fds, ActFds}} = lists:keysearch(active_fds, 1, ChkIo), + case {PendUpdNo, ActFds} of + {0, 0} -> display_check_io(ChkIo), ChkIo; - false -> - ChkIo; _ -> - receive after 10 -> ok end, + receive after 100 -> ok end, get_check_io_info() end. diff --git a/erts/emulator/utils/make_compiler_flags b/erts/emulator/utils/make_compiler_flags index cebe8cd0c5..ca1bc47113 100755 --- a/erts/emulator/utils/make_compiler_flags +++ b/erts/emulator/utils/make_compiler_flags @@ -70,7 +70,7 @@ my($prog) = $prog[$#prog]; print "/* Warning: Do not edit this file.\n"; print " Auto-generated by '$prog'.*/\n"; -foreach(keys %constants) { +foreach (sort(keys %constants)) { print "const char* erts_build_flags_$_ = \"$constants{$_}\";\n" } diff --git a/erts/epmd/src/epmd.c b/erts/epmd/src/epmd.c index 3cfa7a782f..9699491526 100644 --- a/erts/epmd/src/epmd.c +++ b/erts/epmd/src/epmd.c @@ -175,9 +175,9 @@ int main(int argc, char** argv) g->nodes.reg = g->nodes.unreg = g->nodes.unreg_tail = NULL; g->nodes.unreg_count = 0; g->active_conn = 0; -#ifdef HAVE_SYSTEMD_SD_DAEMON_H +#ifdef HAVE_SYSTEMD_DAEMON g->is_systemd = 0; -#endif +#endif /* HAVE_SYSTEMD_DAEMON */ for (i = 0; i < MAX_LISTEN_SOCKETS; i++) g->listenfd[i] = -1; @@ -251,11 +251,11 @@ int main(int argc, char** argv) else usage(g); epmd_cleanup_exit(g,0); -#ifdef HAVE_SYSTEMD_SD_DAEMON_H +#ifdef HAVE_SYSTEMD_DAEMON } else if (strcmp(argv[0], "-systemd") == 0) { g->is_systemd = 1; argv++; argc--; -#endif +#endif /* HAVE_SYSTEMD_DAEMON */ } else usage(g); } @@ -461,11 +461,11 @@ static void usage(EpmdVars *g) fprintf(stderr, " Forcibly unregisters a name with epmd\n"); fprintf(stderr, " (only allowed if -relaxed_command_check was given when \n"); fprintf(stderr, " epmd was started).\n"); -#ifdef HAVE_SYSTEMD_SD_DAEMON_H +#ifdef HAVE_SYSTEMD_DAEMON fprintf(stderr, " -systemd\n"); fprintf(stderr, " Wait for socket from systemd. The option makes sense\n"); fprintf(stderr, " when started from .socket unit.\n"); -#endif +#endif /* HAVE_SYSTEMD_DAEMON */ epmd_cleanup_exit(g,1); } @@ -498,7 +498,11 @@ static void dbg_gen_printf(int onsyslog,int perr,int from_level, #ifdef HAVE_SYSLOG_H if (onsyslog) { - erts_vsnprintf(buf, DEBUG_BUFFER_SIZE, format, args); + int len; + len = erts_vsnprintf(buf, DEBUG_BUFFER_SIZE, format, args); + if (perr != 0 && len < sizeof(buf)) { + erts_snprintf(buf+len, sizeof(buf)-len, ": %s", strerror(perr)); + } syslog(LOG_ERR,"epmd: %s",buf); } #endif @@ -589,9 +593,11 @@ void epmd_cleanup_exit(EpmdVars *g, int exitval) for(i=0; g->argv[i] != NULL; ++i) free(g->argv[i]); free(g->argv); - } - - + } +#ifdef HAVE_SYSTEMD_DAEMON + sd_notifyf(0, "STATUS=Exited.\n" + "ERRNO=%i", exitval); +#endif /* HAVE_SYSTEMD_DAEMON */ exit(exitval); } diff --git a/erts/epmd/src/epmd_int.h b/erts/epmd/src/epmd_int.h index c8f2192f7f..52badd7086 100644 --- a/erts/epmd/src/epmd_int.h +++ b/erts/epmd/src/epmd_int.h @@ -125,9 +125,9 @@ # include "sys/select.h" #endif -#ifdef HAVE_SYSTEMD_SD_DAEMON_H +#ifdef HAVE_SYSTEMD_DAEMON # include <systemd/sd-daemon.h> -#endif +#endif /* HAVE_SYSTEMD_DAEMON */ /* ************************************************************************ */ /* Replace some functions by others by making the function name a macro */ @@ -340,9 +340,9 @@ typedef struct { int listenfd[MAX_LISTEN_SOCKETS]; char *addresses; char **argv; -#ifdef HAVE_SYSTEMD_SD_DAEMON_H +#ifdef HAVE_SYSTEMD_DAEMON int is_systemd; -#endif +#endif /* HAVE_SYSTEMD_DAEMON */ } EpmdVars; void dbg_printf(EpmdVars*,int,const char*,...); diff --git a/erts/epmd/src/epmd_srv.c b/erts/epmd/src/epmd_srv.c index 48fd7a5f9c..26e42adb19 100644 --- a/erts/epmd/src/epmd_srv.c +++ b/erts/epmd/src/epmd_srv.c @@ -244,7 +244,7 @@ void run(EpmdVars *g) } else { -#endif +#endif /* HAVE_SYSTEMD_DAEMON */ dbg_printf(g,2,"try to initiate listening port %d", g->port); @@ -312,7 +312,7 @@ void run(EpmdVars *g) } #ifdef HAVE_SYSTEMD_DAEMON } -#endif +#endif /* HAVE_SYSTEMD_DAEMON */ #if !defined(__WIN32__) && !defined(__OSE__) /* We ignore the SIGPIPE signal that is raised when we call write @@ -330,13 +330,13 @@ void run(EpmdVars *g) FD_ZERO(&g->orig_read_mask); g->select_fd_top = 0; -#ifdef HAVE_SYSTEMD_SD_DAEMON_H +#ifdef HAVE_SYSTEMD_DAEMON if (g->is_systemd) for (i = 0; i < num_sockets; i++) select_fd_set(g, listensock[i]); else { -#endif +#endif /* HAVE_SYSTEMD_DAEMON */ for (i = 0; i < num_sockets; i++) { if ((listensock[i] = socket(FAMILY,SOCK_STREAM,0)) < 0) @@ -399,9 +399,12 @@ void run(EpmdVars *g) } select_fd_set(g, listensock[i]); } -#ifdef HAVE_SYSTEMD_SD_DAEMON_H +#ifdef HAVE_SYSTEMD_DAEMON } -#endif + sd_notifyf(0, "READY=1\n" + "STATUS=Processing port mapping requests...\n" + "MAINPID=%lu", (unsigned long) getpid()); +#endif /* HAVE_SYSTEMD_DAEMON */ dbg_tty_printf(g,2,"entering the main select() loop"); diff --git a/erts/etc/common/erlexec.c b/erts/etc/common/erlexec.c index 709c6f02d1..5ebde8ca3c 100644 --- a/erts/etc/common/erlexec.c +++ b/erts/etc/common/erlexec.c @@ -128,6 +128,7 @@ static char *pluss_val_switches[] = { "bwt", "cl", "ct", + "ecio", "fwi", "tbt", "wct", diff --git a/erts/etc/common/run_erl_common.c b/erts/etc/common/run_erl_common.c index dc55c2bea4..20b78eb05e 100644 --- a/erts/etc/common/run_erl_common.c +++ b/erts/etc/common/run_erl_common.c @@ -36,6 +36,10 @@ # include <syslog.h> #endif +#ifdef HAVE_SYS_IOCTL_H +# include <sys/ioctl.h> +#endif + #ifdef __OSE__ # include "ramlog.h" #endif @@ -74,15 +78,6 @@ * run_erl multiple times with different global variables without them * effecting eachother. */ -typedef struct run_erl_ run_erl; - -#ifdef __OSE__ -static OSPPDKEY run_erl_pp_key; -#define RE_DATA (*(run_erl**)ose_get_ppdata(run_erl_pp_key)) -#else -static run_erl re; -#define RE_DATA (&re) -#endif #define STATUSFILE (RE_DATA->statusfile) #define LOG_DIR (RE_DATA->log_dir) @@ -116,6 +111,16 @@ struct run_erl_ { unsigned protocol_ver; }; +typedef struct run_erl_ run_erl; + +#ifdef __OSE__ +static OSPPDKEY run_erl_pp_key; +#define RE_DATA (*(run_erl**)ose_get_ppdata(run_erl_pp_key)) +#else +static run_erl re; +#define RE_DATA (&re) +#endif + /* prototypes */ static int next_log(int log_num); @@ -636,7 +641,7 @@ int erts_run_erl_open_fifo(char *pipename,char *w_pipename,char *r_pipename) { /* Extract any control sequences that are ment only for run_erl * and should not be forwarded to the pty. */ -int erts_run_erl_extract_ctrl_seq(char* buf, int len) +int erts_run_erl_extract_ctrl_seq(char* buf, int len, int mfd) { static const char prefix[] = "\033_"; static const char suffix[] = "\033\\"; @@ -661,7 +666,7 @@ int erts_run_erl_extract_ctrl_seq(char* buf, int len) struct winsize ws; ws.ws_col = col; ws.ws_row = row; - if (ioctl(MFD, TIOCSWINSZ, &ws) < 0) { + if (ioctl(mfd, TIOCSWINSZ, &ws) < 0) { ERRNO_ERR0(LOG_ERR,"Failed to set window size"); } #endif diff --git a/erts/etc/common/run_erl_common.h b/erts/etc/common/run_erl_common.h index c47a0db054..14207ee4de 100644 --- a/erts/etc/common/run_erl_common.h +++ b/erts/etc/common/run_erl_common.h @@ -40,7 +40,7 @@ void erts_run_erl_log_error(int priority, int line, const char *format,...); int erts_run_erl_open_fifo(char *pipename,char *w_pipename,char *r_pipename); int erts_run_erl_log_alive_minutes(void); -int erts_run_erl_extract_ctrl_seq(char* buf, int len); +int erts_run_erl_extract_ctrl_seq(char* buf, int len, int mfd); /* File operations */ ssize_t sf_read(int fd, void *buffer, size_t len); diff --git a/erts/etc/ose/run_erl.c b/erts/etc/ose/run_erl.c index 6bb59b7f7e..8bc49a485e 100644 --- a/erts/etc/ose/run_erl.c +++ b/erts/etc/ose/run_erl.c @@ -495,7 +495,7 @@ int pass_on(ProgramState *s) { #ifdef DEBUG erts_run_erl_log_status("Pty master write; "); #endif - len = erts_run_erl_extract_ctrl_seq(buffer,len); + len = erts_run_erl_extract_ctrl_seq(buffer,len, s->ofd); if (len > 0) { int wlen = erts_run_erl_write_all(s->ofd, buffer, len); diff --git a/erts/etc/unix/etp-commands.in b/erts/etc/unix/etp-commands.in index bf6eb00314..141d51824f 100644 --- a/erts/etc/unix/etp-commands.in +++ b/erts/etc/unix/etp-commands.in @@ -1130,6 +1130,39 @@ document etp-cp %--------------------------------------------------------------------------- end +define etp-check-beam-ranges + set $etp_ci = 0 + while $etp_ci < 3 + printf "Checking code index %i...\n", $etp_ci + set $etp_j = 0 + while $etp_j < r[$etp_ci].n + set $etp_p = &r[$etp_ci].modules[$etp_j] + if $etp_j > 0 && $etp_p->start < (Range*)$etp_p[-1].end.counter + printf "r[%i].modules[%i]: ERROR start < previous\n", $etp_ci, $etp_j + end + if $etp_p->start > (Range*)$etp_p->end.counter + printf "r[%i].modules[%i]: ERROR start > end\n", $etp_ci, $etp_j + else + if $etp_p->start == (Range*)$etp_p->end.counter + printf "r[%i].modules[%i]: Purged\n", $etp_ci, $etp_j + end + end + set $etp_j = $etp_j + 1 + end + set $etp_ci = $etp_ci + 1 + end +end + +document etp-check-beam-ranges +%--------------------------------------------------------------------------- +% etp-check-beam-ranges +% +% Do consistency check of beam_ranges data structure +% and print errors and empty slots from purged modules. +%--------------------------------------------------------------------------- +end + + ############################################################################ # Commands for special term bunches. # @@ -2401,7 +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 @@ -3552,6 +3585,39 @@ document etp-carrier-blocks %--------------------------------------------------------------------------- end +define etp-address-to-beam-opcode + set $etp_i = 0 + set $etp_min_diff = ((UWord)1 << (sizeof(UWord)*8 - 1)) + set $etp_min_opcode = -1 + set $etp_addr = (UWord) ($arg0) + + while $etp_i < num_instructions && $etp_min_diff > 0 + if ($etp_addr - (UWord)beam_ops[$etp_i]) < $etp_min_diff + set $etp_min_diff = $etp_addr - (UWord)beam_ops[$etp_i] + set $etp_min_opcode = $etp_i + end + set $etp_i = $etp_i + 1 + end + if $etp_min_diff == 0 + printf "Address %p is start of '%s'\n", $etp_addr, opc[$etp_min_opcode].name + else + if $etp_min_opcode >= 0 + printf "Address is %ld bytes into opcode '%s' at %p\n", $etp_min_diff, opc[$etp_min_opcode].name, beam_ops[$etp_min_opcode] + else + printf "Invalid opcode address\n" + end + end +end + +document etp-address-to-beam-opcode +%--------------------------------------------------------------------------- +% Get beam opcode from a native instruction address (within process_main()) +% Arg: Instructon pointer value +% +% Does not work with NO_JUMP_TABLE +%--------------------------------------------------------------------------- +end + ############################################################################ # Toolbox parameter handling diff --git a/erts/etc/unix/run_erl.c b/erts/etc/unix/run_erl.c index a6fc4c2bf5..049e83f9e4 100644 --- a/erts/etc/unix/run_erl.c +++ b/erts/etc/unix/run_erl.c @@ -40,9 +40,13 @@ #ifdef HAVE_CONFIG_H # include "config.h" #endif + #ifdef HAVE_WORKING_POSIX_OPENPT +#ifndef _XOPEN_SOURCE #define _XOPEN_SOURCE 600 #endif +#endif + #include <sys/types.h> #include <sys/wait.h> #include <sys/stat.h> @@ -486,7 +490,7 @@ static void pass_on(pid_t childpid) #ifdef DEBUG erts_run_erl_log_status("Pty master write; "); #endif - len = erts_run_erl_extract_ctrl_seq(buf, len); + len = erts_run_erl_extract_ctrl_seq(buf, len, mfd); if(len==1 && buf[0] == '\003') { kill(childpid,SIGINT); diff --git a/erts/etc/win32/Install.c b/erts/etc/win32/Install.c index 500fd166f8..9d85d642ab 100644 --- a/erts/etc/win32/Install.c +++ b/erts/etc/win32/Install.c @@ -80,7 +80,7 @@ int wmain(int argc, wchar_t **argv) } } if (root == NULL) { - if (module = NULL) { + if (module == NULL) { fprintf(stderr, "Cannot GetModuleHandle()\n"); exit(1); } diff --git a/erts/etc/win32/erl.c b/erts/etc/win32/erl.c index 1d116bf36e..772b668586 100644 --- a/erts/etc/win32/erl.c +++ b/erts/etc/win32/erl.c @@ -264,7 +264,7 @@ static void get_parameters(void) int len; - if (module = NULL) { + if (module == NULL) { error("Cannot GetModuleHandle()"); } diff --git a/erts/include/internal/ethread.h b/erts/include/internal/ethread.h index 72c054b588..ad5d05704c 100644 --- a/erts/include/internal/ethread.h +++ b/erts/include/internal/ethread.h @@ -364,8 +364,8 @@ extern ethr_runtime_t ethr_runtime__; # include "sparc64/ethread.h" # endif # endif -# include "gcc/ethread.h" # include "libatomic_ops/ethread.h" +# include "gcc/ethread.h" # endif # elif defined(ETHR_HAVE_LIBATOMIC_OPS) # include "libatomic_ops/ethread.h" diff --git a/erts/include/internal/libatomic_ops/ethr_atomic.h b/erts/include/internal/libatomic_ops/ethr_atomic.h index fb1288c330..734cdf0890 100644 --- a/erts/include/internal/libatomic_ops/ethr_atomic.h +++ b/erts/include/internal/libatomic_ops/ethr_atomic.h @@ -1,7 +1,7 @@ /* * %CopyrightBegin% * - * Copyright Ericsson AB 2010-2011. All Rights Reserved. + * Copyright Ericsson AB 2010-2014. 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 @@ -32,22 +32,23 @@ * These operations need to be defined by libatomic_ops; * otherwise, we won't compile: * - AO_nop_full() - * - AO_load() - * - AO_store() - * - AO_compare_and_swap() + * - AO_load() || AO_load_aquire() + * - AO_store() || AO_store_release() + * - AO_compare_and_swap() || AO_compare_and_swap_acquire() + * || AO_compare_and_swap_release() || AO_compare_and_swap_full() * */ #if ETHR_SIZEOF_AO_T == 4 #define ETHR_HAVE_NATIVE_ATOMIC32 1 -#define ETHR_NATIVE_ATOMIC32_IMPL "libatomic_ops" +#define ETHR_NATIVE_ATOMIC32_IMPL ETHR_NATIVE_IMPL__ #define ETHR_NATMC_FUNC__(X) ethr_native_atomic32_ ## X #define ETHR_ATMC_T__ ethr_native_atomic32_t #define ETHR_AINT_T__ ethr_sint32_t #define ETHR_AINT_SUFFIX__ "l" #elif ETHR_SIZEOF_AO_T == 8 #define ETHR_HAVE_NATIVE_ATOMIC64 1 -#define ETHR_NATIVE_ATOMIC64_IMPL "libatomic_ops" +#define ETHR_NATIVE_ATOMIC64_IMPL ETHR_NATIVE_IMPL__ #define ETHR_NATMC_FUNC__(X) ethr_native_atomic64_ ## X #define ETHR_ATMC_T__ ethr_native_atomic64_t #define ETHR_AINT_T__ ethr_sint64_t @@ -74,6 +75,8 @@ ETHR_NATMC_FUNC__(addr)(ETHR_ATMC_T__ *var) return (ETHR_AINT_T__ *) &var->counter; } +#ifdef AO_HAVE_store + #if ETHR_SIZEOF_AO_T == 4 # define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_SET 1 #else @@ -86,6 +89,24 @@ ETHR_NATMC_FUNC__(set)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ value) AO_store(&var->counter, (AO_t) value); } +#endif + +#ifdef AO_HAVE_store_write + +#if ETHR_SIZEOF_AO_T == 4 +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_SET_WB 1 +#else +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC64_SET_WB 1 +#endif + +static ETHR_INLINE void +ETHR_NATMC_FUNC__(set_wb)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ value) +{ + AO_store_write(&var->counter, (AO_t) value); +} + +#endif + #ifdef AO_HAVE_store_release #if ETHR_SIZEOF_AO_T == 4 @@ -102,6 +123,24 @@ ETHR_NATMC_FUNC__(set_relb)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ value) #endif +#ifdef AO_HAVE_store_full + +#if ETHR_SIZEOF_AO_T == 4 +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_SET_MB 1 +#else +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC64_SET_MB 1 +#endif + +static ETHR_INLINE void +ETHR_NATMC_FUNC__(set_mb)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ value) +{ + AO_store_full(&var->counter, (AO_t) value); +} + +#endif + +#ifdef AO_HAVE_load + #if ETHR_SIZEOF_AO_T == 4 # define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_READ 1 #else @@ -114,6 +153,24 @@ ETHR_NATMC_FUNC__(read)(ETHR_ATMC_T__ *var) return (ETHR_AINT_T__) AO_load(&var->counter); } +#endif + +#ifdef AO_HAVE_load_read + +#if ETHR_SIZEOF_AO_T == 4 +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_READ_RB 1 +#else +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC64_READ_RB 1 +#endif + +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(read_rb)(ETHR_ATMC_T__ *var) +{ + return (ETHR_AINT_T__) AO_load_read(&var->counter); +} + +#endif + #ifdef AO_HAVE_load_acquire #if ETHR_SIZEOF_AO_T == 4 @@ -130,6 +187,22 @@ ETHR_NATMC_FUNC__(read_acqb)(ETHR_ATMC_T__ *var) #endif +#ifdef AO_HAVE_load_full + +#if ETHR_SIZEOF_AO_T == 4 +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_READ_MB 1 +#else +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC64_READ_MB 1 +#endif + +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(read_mb)(ETHR_ATMC_T__ *var) +{ + return (ETHR_AINT_T__) AO_load_full(&var->counter); +} + +#endif + #ifdef AO_HAVE_fetch_and_add #if ETHR_SIZEOF_AO_T == 4 @@ -146,6 +219,54 @@ ETHR_NATMC_FUNC__(add_return)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ incr) #endif +#ifdef AO_HAVE_fetch_and_add_acquire + +#if ETHR_SIZEOF_AO_T == 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 ((ETHR_AINT_T__) AO_fetch_and_add_acquire(&var->counter, (AO_t) incr)) + incr; +} + +#endif + +#ifdef AO_HAVE_fetch_and_add_release + +#if ETHR_SIZEOF_AO_T == 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 ((ETHR_AINT_T__) AO_fetch_and_add_release(&var->counter, (AO_t) incr)) + incr; +} + +#endif + +#ifdef AO_HAVE_fetch_and_add_full + +#if ETHR_SIZEOF_AO_T == 4 +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_ADD_RETURN_MB 1 +#else +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC64_ADD_RETURN_MB 1 +#endif + +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(add_return_mb)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ incr) +{ + return ((ETHR_AINT_T__) AO_fetch_and_add_full(&var->counter, (AO_t) incr)) + incr; +} + +#endif + #ifdef AO_HAVE_fetch_and_add1 #if ETHR_SIZEOF_AO_T == 4 @@ -178,6 +299,38 @@ ETHR_NATMC_FUNC__(inc_return_acqb)(ETHR_ATMC_T__ *var) #endif +#ifdef AO_HAVE_fetch_and_add1_release + +#if ETHR_SIZEOF_AO_T == 4 +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_INC_RETURN_RELB 1 +#else +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC64_INC_RETURN_RELB 1 +#endif + +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(inc_return_relb)(ETHR_ATMC_T__ *var) +{ + return ((ETHR_AINT_T__) AO_fetch_and_add1_release(&var->counter)) + 1; +} + +#endif + +#ifdef AO_HAVE_fetch_and_add1_full + +#if ETHR_SIZEOF_AO_T == 4 +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_INC_RETURN_MB 1 +#else +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC64_INC_RETURN_MB 1 +#endif + +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(inc_return_mb)(ETHR_ATMC_T__ *var) +{ + return ((ETHR_AINT_T__) AO_fetch_and_add1_full(&var->counter)) + 1; +} + +#endif + #ifdef AO_HAVE_fetch_and_sub1 #if ETHR_SIZEOF_AO_T == 4 @@ -194,6 +347,22 @@ ETHR_NATMC_FUNC__(dec_return)(ETHR_ATMC_T__ *var) #endif +#ifdef AO_HAVE_fetch_and_sub1_acquire + +#if ETHR_SIZEOF_AO_T == 4 +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_DEC_RETURN_ACQB 1 +#else +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC64_DEC_RETURN_ACQB 1 +#endif + +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(dec_return_acqb)(ETHR_ATMC_T__ *var) +{ + return ((ETHR_AINT_T__) AO_fetch_and_sub1_acquire(&var->counter)) - 1; +} + +#endif + #ifdef AO_HAVE_fetch_and_sub1_release #if ETHR_SIZEOF_AO_T == 4 @@ -210,7 +379,60 @@ ETHR_NATMC_FUNC__(dec_return_relb)(ETHR_ATMC_T__ *var) #endif -#ifdef AO_HAVE_compare_and_swap +#ifdef AO_HAVE_fetch_and_sub1_full + +#if ETHR_SIZEOF_AO_T == 4 +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_DEC_RETURN_MB 1 +#else +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC64_DEC_RETURN_MB 1 +#endif + +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(dec_return_mb)(ETHR_ATMC_T__ *var) +{ + return ((ETHR_AINT_T__) AO_fetch_and_sub1_full(&var->counter)) - 1; +} + +#endif + +#if defined(AO_HAVE_compare_and_swap_full) || defined(AO_HAVE_fetch_compare_and_swap_full) + +#if ETHR_SIZEOF_AO_T == 4 +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_CMPXCHG_MB 1 +#else +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC64_CMPXCHG_MB 1 +#endif + +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(cmpxchg_mb)(ETHR_ATMC_T__ *var, + ETHR_AINT_T__ new, + ETHR_AINT_T__ exp) +{ +#if defined(AO_HAVE_fetch_compare_and_swap_full) + return (ETHR_AINT_T__) AO_fetch_compare_and_swap_full(&var->counter, + (AO_t) exp, + (AO_t) new); +#else + ETHR_AINT_T__ act; + do { + if (AO_compare_and_swap_full(&var->counter, (AO_t) exp, (AO_t) new)) + return exp; +#ifdef AO_HAVE_load_acquire + act = (ETHR_AINT_T__) AO_load_acquire(&var->counter); +#else + act = (ETHR_AINT_T__) AO_load(&var->counter); +#endif + } while (act == exp); +#ifndef AO_HAVE_load_acquire + AO_nop_full(); +#endif + return act; +#endif +} + +#endif + +#if defined(AO_HAVE_compare_and_swap) || defined(AO_HAVE_fetch_compare_and_swap) #if ETHR_SIZEOF_AO_T == 4 # define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_CMPXCHG 1 @@ -223,18 +445,28 @@ ETHR_NATMC_FUNC__(cmpxchg)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ new, ETHR_AINT_T__ exp) { +#if defined(AO_HAVE_fetch_compare_and_swap) + return (ETHR_AINT_T__) AO_fetch_compare_and_swap(&var->counter, + (AO_t) exp, + (AO_t) new); +#else ETHR_AINT_T__ act; do { if (AO_compare_and_swap(&var->counter, (AO_t) exp, (AO_t) new)) return exp; +#ifdef AO_HAVE_load act = (ETHR_AINT_T__) AO_load(&var->counter); +#else + act = (ETHR_AINT_T__) AO_load_aquire(&var->counter); +#endif } while (act == exp); return act; +#endif } #endif -#ifdef AO_HAVE_compare_and_swap_acquire +#if defined(AO_HAVE_compare_and_swap_acquire) || defined(AO_HAVE_fetch_compare_and_swap_acquire) #if ETHR_SIZEOF_AO_T == 4 # define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_CMPXCHG_ACQB 1 @@ -247,6 +479,11 @@ ETHR_NATMC_FUNC__(cmpxchg_acqb)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ new, ETHR_AINT_T__ exp) { +#if defined(AO_HAVE_fetch_compare_and_swap_acquire) + return (ETHR_AINT_T__) AO_fetch_compare_and_swap_acquire(&var->counter, + (AO_t) exp, + (AO_t) new); +#else ETHR_AINT_T__ act; do { if (AO_compare_and_swap_acquire(&var->counter, (AO_t) exp, (AO_t) new)) @@ -261,11 +498,55 @@ ETHR_NATMC_FUNC__(cmpxchg_acqb)(ETHR_ATMC_T__ *var, AO_nop_full(); #endif return act; +#endif +} + +#endif + +#if defined(AO_HAVE_compare_and_swap_read) || defined(AO_HAVE_fetch_compare_and_swap_read) + +#if ETHR_SIZEOF_AO_T == 4 +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_CMPXCHG_RB 1 +#else +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC64_CMPXCHG_RB 1 +#endif + +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(cmpxchg_rb)(ETHR_ATMC_T__ *var, + ETHR_AINT_T__ new, + ETHR_AINT_T__ exp) +{ +#if defined(AO_HAVE_fetch_compare_and_swap_read) + return (ETHR_AINT_T__) AO_fetch_compare_and_swap_read(&var->counter, + (AO_t) exp, + (AO_t) new); +#else + ETHR_AINT_T__ act; + do { + if (AO_compare_and_swap_read(&var->counter, (AO_t) exp, (AO_t) new)) + return exp; +#if defined(AO_HAVE_load_read) + act = (ETHR_AINT_T__) AO_load_read(&var->counter); +#elif defined(AO_HAVE_load) + act = (ETHR_AINT_T__) AO_load(&var->counter); +#else + act = (ETHR_AINT_T__) AO_load_acquire(&var->counter); +#endif + } while (act == exp); +#ifndef AO_HAVE_load_read +#ifdef AO_HAVE_nop_read + AO_nop_read(); +#else + AO_nop_full(); +#endif +#endif + return act; +#endif } #endif -#ifdef AO_HAVE_compare_and_swap_release +#if defined(AO_HAVE_compare_and_swap_release) || defined(AO_HAVE_fetch_compare_and_swap_release) #if ETHR_SIZEOF_AO_T == 4 # define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_CMPXCHG_RELB 1 @@ -278,13 +559,57 @@ ETHR_NATMC_FUNC__(cmpxchg_relb)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ new, ETHR_AINT_T__ exp) { +#if defined(AO_HAVE_fetch_compare_and_swap_release) + return (ETHR_AINT_T__) AO_fetch_compare_and_swap_release(&var->counter, + (AO_t) exp, + (AO_t) new); +#else ETHR_AINT_T__ act; do { if (AO_compare_and_swap_release(&var->counter, (AO_t) exp, (AO_t) new)) return exp; +#ifdef AO_HAVE_load act = (ETHR_AINT_T__) AO_load(&var->counter); +#else + act = (ETHR_AINT_T__) AO_load_acquire(&var->counter); +#endif + } while (act == exp); + return act; +#endif +} + +#endif + +#if defined(AO_HAVE_compare_and_swap_write) || defined(AO_HAVE_fetch_compare_and_swap_write) + +#if ETHR_SIZEOF_AO_T == 4 +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_CMPXCHG_WB 1 +#else +# define ETHR_HAVE_ETHR_NATIVE_ATOMIC64_CMPXCHG_WB 1 +#endif + +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(cmpxchg_wb)(ETHR_ATMC_T__ *var, + ETHR_AINT_T__ new, + ETHR_AINT_T__ exp) +{ +#if defined(AO_HAVE_fetch_compare_and_swap_write) + return (ETHR_AINT_T__) AO_fetch_compare_and_swap_write(&var->counter, + (AO_t) exp, + (AO_t) new); +#else + ETHR_AINT_T__ act; + do { + if (AO_compare_and_swap_write(&var->counter, (AO_t) exp, (AO_t) new)) + return exp; +#ifdef AO_HAVE_load + act = (ETHR_AINT_T__) AO_load(&var->counter); +#else + act = (ETHR_AINT_T__) AO_load_acquire(&var->counter); +#endif } while (act == exp); return act; +#endif } #endif diff --git a/erts/include/internal/libatomic_ops/ethr_dw_atomic.h b/erts/include/internal/libatomic_ops/ethr_dw_atomic.h new file mode 100644 index 0000000000..4dd9f41e96 --- /dev/null +++ b/erts/include/internal/libatomic_ops/ethr_dw_atomic.h @@ -0,0 +1,567 @@ +/* + * %CopyrightBegin% + * + * Copyright Ericsson AB 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 + * 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% + */ + +/* + * Description: Native double word atomics using libatomic_ops + * Author: Rickard Green + */ + +#ifndef ETHR_LIBATOMIC_OPS_DW_ATOMIC_H__ +#define ETHR_LIBATOMIC_OPS_DW_ATOMIC_H__ + +#if defined(AO_HAVE_double_t) \ + && (defined(AO_HAVE_double_load_acquire) \ + || defined(AO_HAVE_double_load)) \ + && (defined(AO_HAVE_compare_double_and_swap_double) \ + || defined(AO_HAVE_compare_double_and_swap_double_full) \ + || defined(AO_HAVE_compare_double_and_swap_double_acquire) \ + || defined(AO_HAVE_compare_double_and_swap_double_release) \ + || defined(AO_HAVE_double_compare_and_swap) \ + || defined(AO_HAVE_double_compare_and_swap_full) \ + || defined(AO_HAVE_double_compare_and_swap_acquire) \ + || defined(AO_HAVE_double_compare_and_swap_release)) + +#if ETHR_SIZEOF_PTR == 4 +# define ETHR_NATIVE_SU_DW_SINT_T ethr_sint64_t +#elif ETHR_SIZEOF_PTR == 8 && defined(ETHR_HAVE_INT128_T) +# define ETHR_NATIVE_SU_DW_SINT_T ethr_sint128_t +#endif + +typedef union { + volatile AO_double_t dw_mem; +#if defined(ETHR_NATIVE_SU_DW_SINT_T) + ETHR_NATIVE_SU_DW_SINT_T su_dw_sint; +#endif +} ethr_native_dw_atomic_t; + +#if defined(ETHR_NATIVE_SU_DW_SINT_T) +# define ETHR_HAVE_NATIVE_SU_DW_ATOMIC +#else +# define ETHR_HAVE_NATIVE_DW_ATOMIC +#endif + +#define ETHR_NATIVE_DW_ATOMIC_IMPL ETHR_NATIVE_IMPL__ + +#if defined(ETHR_TRY_INLINE_FUNCS) || defined(ETHR_ATOMIC_IMPL__) + + +#if defined(ETHR_NATIVE_SU_DW_SINT_T) +# define ETHR_NDWA_FUNC__(Func) ethr_native_su_dw_atomic_ ## Func +# define ETHR_NDWA_RET_3_TYPE__ ETHR_NATIVE_SU_DW_SINT_T +# define ETHR_NDWA_RET_2_TYPE__ ETHR_NATIVE_SU_DW_SINT_T +# define ETHR_NDWA_VAL_ARG_TYPE__ ETHR_NATIVE_SU_DW_SINT_T +# define ETHR_NDWA_DECL_ARG__(Arg) +# if defined(AO_HAVE_DOUBLE_PTR_STORAGE) +# define ETHR_NDWA_VAL2AOVAL__(AOV, V) \ + ((AOV).AO_whole = (double_ptr_storage) (V)) +# define ETHR_NDWA_AOVAL2VAL__(AOV, V) \ + ((V) = (ETHR_NATIVE_SU_DW_SINT_T) (AOV).AO_whole) +# define ETHR_NDWA_RETURN_VAL_3__(SUCCESS, AOVAL, VAL) \ + do { \ + return (ETHR_NATIVE_SU_DW_SINT_T) (AOVAL).AO_whole; \ + } while (0) +# define ETHR_NDWA_RETURN_VAL_2__(AOVAL, VAL) \ + do { \ + return (ETHR_NATIVE_SU_DW_SINT_T) (AOVAL).AO_whole; \ + } while (0) +# define ETHR_NDWA_AOVAL_EQ__(AOV1, AOV2) \ + ((AOV1).AO_whole == (AOV2).AO_whole) +# else +typedef union { + ethr_sint_t sint[2]; + ETHR_NATIVE_SU_DW_SINT_T dw_sint; +} ethr_dw_splitter_t; +# define ETHR_NDWA_VAL2AOVAL__(AOV, V) \ + do { \ + ethr_dw_splitter_t tmp__; \ + tmp__.dw_sint = (V); \ + (AOV).AO_val1 = (AO_t) tmp__.sint[0]; \ + (AOV).AO_val2 = (AO_t) tmp__.sint[1]; \ + } while (0) +# define ETHR_NDWA_AOVAL2VAL__(AOV, V) \ + do { \ + ethr_dw_splitter_t tmp__; \ + tmp__.sint[0] = (ethr_sint_t) (AOV).AO_val1; \ + tmp__.sint[1] = (ethr_sint_t) (AOV).AO_val2; \ + (V) = tmp__.dw_sint; \ + } while (0) +# define ETHR_NDWA_RETURN_VAL_3__(SUCCESS, AOVAL, VAL) \ + do { \ + ethr_dw_splitter_t tmp__; \ + tmp__.sint[0] = (ethr_sint_t) (AOVAL).AO_val1; \ + tmp__.sint[1] = (ethr_sint_t) (AOVAL).AO_val2; \ + return tmp__.dw_sint; \ + } while (0) +# define ETHR_NDWA_AOVAL_EQ__(AOV1, AOV2) \ + ((AOV1).AO_val1 == (AOV2).AO_val1 \ + && (AOV1).AO_val2 == (AOV2).AO_val2) +# endif +#else +# define ETHR_NDWA_FUNC__(Func) ethr_native_dw_atomic_ ## Func +# define ETHR_NDWA_RET_3_TYPE__ int +# define ETHR_NDWA_RET_2_TYPE__ void +# define ETHR_NDWA_VAL_ARG_TYPE__ ethr_sint_t * +# define ETHR_NDWA_DECL_ARG__(Arg) , ETHR_NDWA_VAL_ARG_TYPE__ Arg +# define ETHR_NDWA_VAL2AOVAL__(AOV, V) \ + do { \ + (AOV).AO_val1 = (AO_t) (V)[0]; \ + (AOV).AO_val2 = (AO_t) (V)[1]; \ + } while (0) +# define ETHR_NDWA_AOVAL2VAL__(AOV, V) \ + do { \ + ethr_dw_splitter_t tmp__; \ + (V)[0] = (ethr_sint_t) (AOV).AO_val1; \ + (V)[1] = (ethr_sint_t) (AOV).AO_val2; \ + } while (0) +# define ETHR_NDWA_RETURN_VAL_3__(SUCCESS, AOVAL, VAL) \ + do { \ + (VAL)[0] = (ethr_sint_t) (AOVAL).AO_val1; \ + (VAL)[1] = (ethr_sint_t) (AOVAL).AO_val2; \ + return (SUCCESS); \ + } while (0) +# define ETHR_NDWA_RETURN_VAL_2__(AOVAL, VAL) \ + do { \ + (VAL)[0] = (ethr_sint_t) (AOVAL).AO_val1; \ + (VAL)[1] = (ethr_sint_t) (AOVAL).AO_val2; \ + return; \ + } while (0) +# if defined(AO_HAVE_DOUBLE_PTR_STORAGE) +# define ETHR_NDWA_AOVAL_EQ__(AOV1, AOV2) \ + ((AOV1).AO_whole == (AOV2).AO_whole) +# else +# define ETHR_NDWA_AOVAL_EQ__(AOV1, AOV2) \ + ((AOV1).AO_val1 == (AOV2).AO_val1 \ + && (AOV1).AO_val2 == (AOV2).AO_val2) +# endif +#endif + +#define ETHR_HAVE_ETHR_NATIVE_DW_ATOMIC_ADDR +static ETHR_INLINE ethr_sint_t * +ethr_native_dw_atomic_addr(ethr_native_dw_atomic_t *var) +{ + return (ethr_sint_t *) &var->dw_mem; +} + +#ifdef AO_HAVE_double_load + +#if defined(ETHR_NATIVE_SU_DW_SINT_T) +# define ETHR_HAVE_ETHR_NATIVE_SU_DW_ATOMIC_READ +#else +# define ETHR_HAVE_ETHR_NATIVE_DW_ATOMIC_READ +#endif + +static ETHR_INLINE ETHR_NDWA_RET_2_TYPE__ +ETHR_NDWA_FUNC__(read)(ethr_native_dw_atomic_t *var + ETHR_NDWA_DECL_ARG__(val)) +{ + AO_double_t act = AO_double_load(&var->dw_mem); + ETHR_NDWA_RETURN_VAL_2__(act, val); +} + +#endif + +#ifdef AO_HAVE_double_load_read + +#if defined(ETHR_NATIVE_SU_DW_SINT_T) +# define ETHR_HAVE_ETHR_NATIVE_SU_DW_ATOMIC_READ_RB +#else +# define ETHR_HAVE_ETHR_NATIVE_DW_ATOMIC_READ_RB +#endif + +static ETHR_INLINE ETHR_NDWA_RET_2_TYPE__ +ETHR_NDWA_FUNC__(read_rb)(ethr_native_dw_atomic_t *var + ETHR_NDWA_DECL_ARG__(val)) +{ + AO_double_t act = AO_double_load_read(&var->dw_mem); + ETHR_NDWA_RETURN_VAL_2__(act, val); +} + +#endif + +#ifdef AO_HAVE_double_load_acquire + +#if defined(ETHR_NATIVE_SU_DW_SINT_T) +# define ETHR_HAVE_ETHR_NATIVE_SU_DW_ATOMIC_READ_ACQB +#else +# define ETHR_HAVE_ETHR_NATIVE_DW_ATOMIC_READ_ACQB +#endif + +static ETHR_INLINE ETHR_NDWA_RET_2_TYPE__ +ETHR_NDWA_FUNC__(read_acqb)(ethr_native_dw_atomic_t *var + ETHR_NDWA_DECL_ARG__(val)) +{ + AO_double_t act = AO_double_load_acquire(&var->dw_mem); + ETHR_NDWA_RETURN_VAL_2__(act, val); +} + +#endif + +#ifdef AO_HAVE_double_store + +#if defined(ETHR_NATIVE_SU_DW_SINT_T) +# define ETHR_HAVE_ETHR_NATIVE_SU_DW_ATOMIC_SET +#else +# define ETHR_HAVE_ETHR_NATIVE_DW_ATOMIC_SET +#endif + +static ETHR_INLINE void +ETHR_NDWA_FUNC__(set)(ethr_native_dw_atomic_t *var, + ETHR_NDWA_VAL_ARG_TYPE__ val) +{ + AO_double_t new; + ETHR_NDWA_VAL2AOVAL__(new, val); + AO_double_store(&var->dw_mem, new); +} + +#endif + +#ifdef AO_HAVE_double_store_write + +#if defined(ETHR_NATIVE_SU_DW_SINT_T) +# define ETHR_HAVE_ETHR_NATIVE_SU_DW_ATOMIC_SET_WB +#else +# define ETHR_HAVE_ETHR_NATIVE_DW_ATOMIC_SET_WB +#endif + +static ETHR_INLINE void +ETHR_NDWA_FUNC__(set_wb)(ethr_native_dw_atomic_t *var, + ETHR_NDWA_VAL_ARG_TYPE__ val) +{ + AO_double_t new; + ETHR_NDWA_VAL2AOVAL__(new, val); + AO_double_store_write(&var->dw_mem, new); +} + +#endif + +#ifdef AO_HAVE_double_store_release + +#if defined(ETHR_NATIVE_SU_DW_SINT_T) +# define ETHR_HAVE_ETHR_NATIVE_SU_DW_ATOMIC_SET_RELB +#else +# define ETHR_HAVE_ETHR_NATIVE_DW_ATOMIC_SET_RELB +#endif + +static ETHR_INLINE void +ETHR_NDWA_FUNC__(set_relb)(ethr_native_dw_atomic_t *var, + ETHR_NDWA_VAL_ARG_TYPE__ val) +{ + AO_double_t new; + ETHR_NDWA_VAL2AOVAL__(new, val); + AO_double_store_release(&var->dw_mem, new); +} + +#endif + +#if defined(AO_HAVE_double_compare_and_swap_full) || defined(AO_HAVE_compare_double_and_swap_double_full) + +#if defined(ETHR_NATIVE_SU_DW_SINT_T) +# define ETHR_HAVE_ETHR_NATIVE_SU_DW_ATOMIC_CMPXCHG_MB +#else +# define ETHR_HAVE_ETHR_NATIVE_DW_ATOMIC_CMPXCHG_MB +#endif + +static ETHR_INLINE ETHR_NDWA_RET_3_TYPE__ +ETHR_NDWA_FUNC__(cmpxchg_mb)(ethr_native_dw_atomic_t *var, + ETHR_NDWA_VAL_ARG_TYPE__ new, + ETHR_NDWA_VAL_ARG_TYPE__ exp) +{ + AO_double_t ao_act, ao_new, ao_exp; + + ETHR_NDWA_VAL2AOVAL__(ao_exp, exp); + ETHR_NDWA_VAL2AOVAL__(ao_new, new); + + do { + int xchgd; +#if defined(AO_HAVE_double_compare_and_swap_full) + xchgd = AO_double_compare_and_swap_full(&var->dw_mem, ao_exp, ao_new); +#elif defined(AO_HAVE_compare_double_and_swap_double_full) + xchgd = AO_compare_double_and_swap_double_full(&var->dw_mem, + ao_exp.AO_val1, + ao_exp.AO_val2, + ao_new.AO_val1, + ao_new.AO_val2); +#endif + + if (xchgd) + ETHR_NDWA_RETURN_VAL_3__(1, ao_exp, exp); + +#ifdef AO_HAVE_double_load_acquire + ao_act = AO_double_load_acquire(&var->dw_mem); +#else + ao_act = AO_double_load(&var->dw_mem); +#endif + + } while (ETHR_NDWA_AOVAL_EQ__(ao_exp, ao_act)); + +#ifndef AO_HAVE_double_load_acquire + AO_nop_full(); +#endif + + ETHR_NDWA_RETURN_VAL_3__(1, ao_act, exp); +} + +#endif + +#if defined(AO_HAVE_double_compare_and_swap) || defined(AO_HAVE_compare_double_and_swap_double) + +#if defined(ETHR_NATIVE_SU_DW_SINT_T) +# define ETHR_HAVE_ETHR_NATIVE_SU_DW_ATOMIC_CMPXCHG +#else +# define ETHR_HAVE_ETHR_NATIVE_DW_ATOMIC_CMPXCHG +#endif + +static ETHR_INLINE ETHR_NDWA_RET_3_TYPE__ +ETHR_NDWA_FUNC__(cmpxchg)(ethr_native_dw_atomic_t *var, + ETHR_NDWA_VAL_ARG_TYPE__ new, + ETHR_NDWA_VAL_ARG_TYPE__ exp) +{ + AO_double_t ao_act, ao_new, ao_exp; + + ETHR_NDWA_VAL2AOVAL__(ao_exp, exp); + ETHR_NDWA_VAL2AOVAL__(ao_new, new); + + do { + int xchgd; +#if defined(AO_HAVE_double_compare_and_swap) + xchgd = AO_double_compare_and_swap(&var->dw_mem, ao_exp, ao_new); +#elif defined(AO_HAVE_compare_double_and_swap_double) + xchgd = AO_compare_double_and_swap_double(&var->dw_mem, + ao_exp.AO_val1, + ao_exp.AO_val2, + ao_new.AO_val1, + ao_new.AO_val2); +#endif + + if (xchgd) + ETHR_NDWA_RETURN_VAL_3__(1, ao_exp, exp); + +#ifdef AO_HAVE_double_load + ao_act = AO_double_load(&var->dw_mem); +#else + ao_act = AO_double_load_acquire(&var->dw_mem); +#endif + + } while (ETHR_NDWA_AOVAL_EQ__(ao_exp, ao_act)); + + ETHR_NDWA_RETURN_VAL_3__(1, ao_act, exp); +} + +#endif + +#if defined(AO_HAVE_double_compare_and_swap_read) || defined(AO_HAVE_compare_double_and_swap_double_read) + +#if defined(ETHR_NATIVE_SU_DW_SINT_T) +# define ETHR_HAVE_ETHR_NATIVE_SU_DW_ATOMIC_CMPXCHG_RB +#else +# define ETHR_HAVE_ETHR_NATIVE_DW_ATOMIC_CMPXCHG_RB +#endif + +static ETHR_INLINE ETHR_NDWA_RET_3_TYPE__ +ETHR_NDWA_FUNC__(cmpxchg_rb)(ethr_native_dw_atomic_t *var, + ETHR_NDWA_VAL_ARG_TYPE__ new, + ETHR_NDWA_VAL_ARG_TYPE__ exp) +{ + AO_double_t ao_act, ao_new, ao_exp; + + ETHR_NDWA_VAL2AOVAL__(ao_exp, exp); + ETHR_NDWA_VAL2AOVAL__(ao_new, new); + + do { + int xchgd; +#if defined(AO_HAVE_double_compare_and_swap_read) + xchgd = AO_double_compare_and_swap_read(&var->dw_mem, ao_exp, ao_new); +#elif defined(AO_HAVE_compare_double_and_swap_double_read) + xchgd = AO_compare_double_and_swap_double_read(&var->dw_mem, + ao_exp.AO_val1, + ao_exp.AO_val2, + ao_new.AO_val1, + ao_new.AO_val2); +#endif + + if (xchgd) + ETHR_NDWA_RETURN_VAL_3__(1, ao_exp, exp); + +#if defined(AO_HAVE_double_load_read) + ao_act = AO_double_load_read(&var->dw_mem); +#elif defined(AO_HAVE_double_load) + ao_act = AO_double_load(&var->dw_mem); +#else + ao_act = AO_double_load_acquire(&var->dw_mem); +#endif + + } while (ETHR_NDWA_AOVAL_EQ__(ao_exp, ao_act)); + +#ifndef AO_HAVE_double_load_read +#ifdef AO_HAVE_nop_read + AO_nop_read(); +#else + AO_nop_full(); +#endif +#endif + + ETHR_NDWA_RETURN_VAL_3__(1, ao_act, exp); +} + +#endif + +#if defined(AO_HAVE_double_compare_and_swap_acquire) || defined(AO_HAVE_compare_double_and_swap_double_acquire) + +#if defined(ETHR_NATIVE_SU_DW_SINT_T) +# define ETHR_HAVE_ETHR_NATIVE_SU_DW_ATOMIC_CMPXCHG_ACQB +#else +# define ETHR_HAVE_ETHR_NATIVE_DW_ATOMIC_CMPXCHG_ACQB +#endif + +static ETHR_INLINE ETHR_NDWA_RET_3_TYPE__ +ETHR_NDWA_FUNC__(cmpxchg_acqb)(ethr_native_dw_atomic_t *var, + ETHR_NDWA_VAL_ARG_TYPE__ new, + ETHR_NDWA_VAL_ARG_TYPE__ exp) +{ + AO_double_t ao_act, ao_new, ao_exp; + + ETHR_NDWA_VAL2AOVAL__(ao_exp, exp); + ETHR_NDWA_VAL2AOVAL__(ao_new, new); + + do { + int xchgd; +#if defined(AO_HAVE_double_compare_and_swap_acquire) + xchgd = AO_double_compare_and_swap_acquire(&var->dw_mem, ao_exp, ao_new); +#elif defined(AO_HAVE_compare_double_and_swap_double_acquire) + xchgd = AO_compare_double_and_swap_double_acquire(&var->dw_mem, + ao_exp.AO_val1, + ao_exp.AO_val2, + ao_new.AO_val1, + ao_new.AO_val2); +#endif + + if (xchgd) + ETHR_NDWA_RETURN_VAL_3__(1, ao_exp, exp); + +#ifdef AO_HAVE_double_load_acquire + ao_act = AO_double_load_acquire(&var->dw_mem); +#else + ao_act = AO_double_load(&var->dw_mem); +#endif + + } while (ETHR_NDWA_AOVAL_EQ__(ao_exp, ao_act)); + +#ifndef AO_HAVE_double_load_acquire + AO_nop_full(); +#endif + + ETHR_NDWA_RETURN_VAL_3__(1, ao_act, exp); +} + +#endif + +#if defined(AO_HAVE_double_compare_and_swap_write) || defined(AO_HAVE_compare_double_and_swap_double_write) + +#if defined(ETHR_NATIVE_SU_DW_SINT_T) +# define ETHR_HAVE_ETHR_NATIVE_SU_DW_ATOMIC_CMPXCHG_WB +#else +# define ETHR_HAVE_ETHR_NATIVE_DW_ATOMIC_CMPXCHG_WB +#endif + +static ETHR_INLINE ETHR_NDWA_RET_3_TYPE__ +ETHR_NDWA_FUNC__(cmpxchg_wb)(ethr_native_dw_atomic_t *var, + ETHR_NDWA_VAL_ARG_TYPE__ new, + ETHR_NDWA_VAL_ARG_TYPE__ exp) +{ + AO_double_t ao_act, ao_new, ao_exp; + + ETHR_NDWA_VAL2AOVAL__(ao_exp, exp); + ETHR_NDWA_VAL2AOVAL__(ao_new, new); + + do { + int xchgd; +#if defined(AO_HAVE_double_compare_and_swap_write) + xchgd = AO_double_compare_and_swap_write(&var->dw_mem, ao_exp, ao_new); +#elif defined(AO_HAVE_compare_double_and_swap_double_write) + xchgd = AO_compare_double_and_swap_double_write(&var->dw_mem, + ao_exp.AO_val1, + ao_exp.AO_val2, + ao_new.AO_val1, + ao_new.AO_val2); +#endif + + if (xchgd) + ETHR_NDWA_RETURN_VAL_3__(1, ao_exp, exp); + +#ifdef AO_HAVE_double_load + ao_act = AO_double_load(&var->dw_mem); +#else + ao_act = AO_double_load_acquire(&var->dw_mem); +#endif + + } while (ETHR_NDWA_AOVAL_EQ__(ao_exp, ao_act)); + + ETHR_NDWA_RETURN_VAL_3__(1, ao_act, exp); +} + +#endif + +#if defined(AO_HAVE_double_compare_and_swap_release) || defined(AO_HAVE_compare_double_and_swap_double_release) + +#if defined(ETHR_NATIVE_SU_DW_SINT_T) +# define ETHR_HAVE_ETHR_NATIVE_SU_DW_ATOMIC_CMPXCHG_RELB +#else +# define ETHR_HAVE_ETHR_NATIVE_DW_ATOMIC_CMPXCHG_RELB +#endif + +static ETHR_INLINE ETHR_NDWA_RET_3_TYPE__ +ETHR_NDWA_FUNC__(cmpxchg_relb)(ethr_native_dw_atomic_t *var, + ETHR_NDWA_VAL_ARG_TYPE__ new, + ETHR_NDWA_VAL_ARG_TYPE__ exp) +{ + AO_double_t ao_act, ao_new, ao_exp; + + ETHR_NDWA_VAL2AOVAL__(ao_exp, exp); + ETHR_NDWA_VAL2AOVAL__(ao_new, new); + + do { + int xchgd; +#if defined(AO_HAVE_double_compare_and_swap_release) + xchgd = AO_double_compare_and_swap_release(&var->dw_mem, ao_exp, ao_new); +#elif defined(AO_HAVE_compare_double_and_swap_double_release) + xchgd = AO_compare_double_and_swap_double_release(&var->dw_mem, + ao_exp.AO_val1, + ao_exp.AO_val2, + ao_new.AO_val1, + ao_new.AO_val2); +#endif + + if (xchgd) + ETHR_NDWA_RETURN_VAL_3__(1, ao_exp, exp); + + ao_act = AO_double_load(&var->dw_mem); + + } while (ETHR_NDWA_AOVAL_EQ__(ao_exp, ao_act)); + + ETHR_NDWA_RETURN_VAL_3__(1, ao_act, exp); +} + +#endif + +#endif /* defined(ETHR_TRY_INLINE_FUNCS) || defined(ETHR_ATOMIC_IMPL__) */ + +#endif /* Have AO double functionality ... */ + +#endif /* ETHR_LIBATOMIC_OPS_DW_ATOMIC_H__ */ + diff --git a/erts/include/internal/libatomic_ops/ethread.h b/erts/include/internal/libatomic_ops/ethread.h index e1fdd588bb..d65ee19b04 100644 --- a/erts/include/internal/libatomic_ops/ethread.h +++ b/erts/include/internal/libatomic_ops/ethread.h @@ -33,9 +33,12 @@ #define AO_USE_PENTIUM4_INSTRS #endif +#define ETHR_NATIVE_IMPL__ "libatomic_ops" + #include "atomic_ops.h" #include "ethr_membar.h" #include "ethr_atomic.h" +#include "ethr_dw_atomic.h" #endif diff --git a/erts/include/internal/ppc32/atomic.h b/erts/include/internal/ppc32/atomic.h index 6001620677..b558626b09 100644 --- a/erts/include/internal/ppc32/atomic.h +++ b/erts/include/internal/ppc32/atomic.h @@ -91,6 +91,20 @@ ethr_native_atomic32_add_return_acqb(ethr_native_atomic32_t *var, ethr_sint32_t return res; } + +#ifndef ETHR_PPC_HAVE_NO_LWSYNC + +#define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_ADD_RETURN_RELB 1 + +static ETHR_INLINE ethr_sint32_t +ethr_native_atomic32_add_return_relb(ethr_native_atomic32_t *var, ethr_sint32_t incr) +{ + ethr_lwsync__(); + return ethr_native_atomic32_add_return(var, incr); +} + +#endif + #define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_INC_RETURN 1 static ETHR_INLINE ethr_sint32_t @@ -120,7 +134,19 @@ ethr_native_atomic32_inc_return_acqb(ethr_native_atomic32_t *var) __asm__ __volatile("isync\n\t" : : : "memory"); return res; } - + +#ifndef ETHR_PPC_HAVE_NO_LWSYNC + +#define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_INC_RETURN_RELB 1 + +static ETHR_INLINE ethr_sint32_t +ethr_native_atomic32_inc_return_relb(ethr_native_atomic32_t *var) +{ + ethr_lwsync__(); + return ethr_native_atomic32_inc_return(var); +} + +#endif #define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_DEC_RETURN 1 @@ -152,6 +178,19 @@ ethr_native_atomic32_dec_return_acqb(ethr_native_atomic32_t *var) return res; } +#ifndef ETHR_PPC_HAVE_NO_LWSYNC + +#define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_DEC_RETURN_RELB 1 + +static ETHR_INLINE ethr_sint32_t +ethr_native_atomic32_dec_return_relb(ethr_native_atomic32_t *var) +{ + ethr_lwsync__(); + return ethr_native_atomic32_dec_return(var); +} + +#endif + #define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_AND_RETOLD 1 static ETHR_INLINE ethr_sint32_t @@ -182,6 +221,19 @@ ethr_native_atomic32_and_retold_acqb(ethr_native_atomic32_t *var, ethr_sint32_t return res; } +#ifndef ETHR_PPC_HAVE_NO_LWSYNC + +#define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_AND_RETOLD_RELB 1 + +static ETHR_INLINE ethr_sint32_t +ethr_native_atomic32_and_retold_relb(ethr_native_atomic32_t *var, ethr_sint32_t mask) +{ + ethr_lwsync__(); + return ethr_native_atomic32_and_retold(var, mask); +} + +#endif + #define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_OR_RETOLD 1 static ETHR_INLINE ethr_sint32_t @@ -212,6 +264,18 @@ ethr_native_atomic32_or_retold_acqb(ethr_native_atomic32_t *var, ethr_sint32_t m return res; } +#ifndef ETHR_PPC_HAVE_NO_LWSYNC + +#define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_OR_RETOLD_RELB 1 + +static ETHR_INLINE ethr_sint32_t +ethr_native_atomic32_or_retold_relb(ethr_native_atomic32_t *var, ethr_sint32_t mask) +{ + ethr_lwsync__(); + return ethr_native_atomic32_or_retold(var, mask); +} + +#endif #define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_XCHG 1 @@ -242,6 +306,19 @@ ethr_native_atomic32_xchg_acqb(ethr_native_atomic32_t *var, ethr_sint32_t val) return res; } +#ifndef ETHR_PPC_HAVE_NO_LWSYNC + +#define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_XCHG_RELB 1 + +static ETHR_INLINE ethr_sint32_t +ethr_native_atomic32_xchg_relb(ethr_native_atomic32_t *var, ethr_sint32_t val) +{ + ethr_lwsync__(); + return ethr_native_atomic32_xchg(var, val); +} + +#endif + #define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_CMPXCHG 1 static ETHR_INLINE ethr_sint32_t @@ -291,6 +368,73 @@ ethr_native_atomic32_cmpxchg_acqb(ethr_native_atomic32_t *var, return old; } +#if !defined(ETHR_DISABLE_LWSYNC_FOR_CMPXCHG_RELB) && !defined(ETHR_PPC_HAVE_NO_LWSYNC) + +#define ETHR_HAVE_ETHR_NATIVE_ATOMIC32_CMPXCHG_RELB 1 + +static ETHR_INLINE ethr_sint32_t +ethr_native_atomic32_cmpxchg_relb(ethr_native_atomic32_t *var, + ethr_sint32_t new, + ethr_sint32_t expected) +{ + ethr_sint32_t actual; + + /* + * We want to implement the release barrier using the + * 'lwsync' instruction instead of using the more + * expensive 'sync' instruction. + * + * cmpxchg looks something like this: + * + * lwarx # Load + * ... + * if (fail) + * goto done; + * stwcx # Store + * if (fail) + * goto done; + * ... + * + * In the case we succeeded, 'lwsync' will have + * ordered all previously issued loads and stores + * against the successful store to this variable. + * That is everything is fine! + * + * In the case we did not succeed, we need to order + * all previously issued loads and stores against + * the load of this variable. 'lwsync' does not + * guarantee this. In order to solve this we issue + * a 'sync' and redo the load. If the value has + * changed to what the user passed as expected value + * we need to try the cmpxchg operation again, since + * this value indicates success. + */ + + ethr_lwsync__(); + + actual = ethr_native_atomic32_cmpxchg(var, new, expected); + +#ifndef ETHR_PPC_HAVE_LWSYNC + /* We checked for lwsync support in runtime... */ + if (ETHR_PPC_RUNTIME_CONF_HAVE_NO_LWSYNC__) + return actual; /* No need to; ethr_lwsync__() issued a sync... */ +#endif + + /* ethr_lwsync__() issued an lwsync... */ + if (actual == expected) + return actual; /* Successful operation */ + + /* Failure... need to issue a sync... */ + ethr_sync__(); + actual = ethr_native_atomic32_read(var); + if (actual != expected) + return actual; /* Fail... */ + /* Try again... */ + return ethr_native_atomic32_cmpxchg(var, new, expected); +} + +#endif + #endif /* ETHR_TRY_INLINE_FUNCS */ #endif /* ETHREAD_PPC_ATOMIC_H */ diff --git a/erts/lib_src/common/erl_misc_utils.c b/erts/lib_src/common/erl_misc_utils.c index d58a28b5cb..7833dd8219 100644 --- a/erts/lib_src/common/erl_misc_utils.c +++ b/erts/lib_src/common/erl_misc_utils.c @@ -1515,7 +1515,7 @@ const char* parse_topology_spec_group(erts_cpu_info_t *cpuinfo, const char* xml, if (is_thread_group) { thread++; } else { - *core_p = (*core_p)++; + *core_p = (*core_p) + 1; } index_procs++; } @@ -1535,9 +1535,9 @@ const char* parse_topology_spec_group(erts_cpu_info_t *cpuinfo, const char* xml, if (parentCacheLevel == 0) { *core_p = 0; - *processor_p = (*processor_p)++; + *processor_p = (*processor_p) + 1; } else { - *core_p = (*core_p)++; + *core_p = (*core_p) + 1; } if (error) diff --git a/erts/preloaded/ebin/erl_prim_loader.beam b/erts/preloaded/ebin/erl_prim_loader.beam Binary files differindex 5f2b619322..10138cab22 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 260badbcb3..7a3d94dd16 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 7dc7407a81..9b364d8b21 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 5c139c4550..a190129b2c 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 cf32b79e8d..621f036f33 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 37ed8d0365..6695f4cd15 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 d49578abfa..5e9230826d 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 8420052533..83230d42ae 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 8dc8cb961b..b3abba3223 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 7507efb076..7006764d96 100644 --- a/erts/preloaded/ebin/zlib.beam +++ b/erts/preloaded/ebin/zlib.beam diff --git a/erts/preloaded/src/erl_prim_loader.erl b/erts/preloaded/src/erl_prim_loader.erl index 466e0b0020..6b86a427ba 100644 --- a/erts/preloaded/src/erl_prim_loader.erl +++ b/erts/preloaded/src/erl_prim_loader.erl @@ -1507,7 +1507,14 @@ real_path(Name,[Path|Paths],Acc,Links) -> [""|_] = LinkPaths -> real_path(Name,LinkPaths++Paths,[],[ThisFile|Links]); LinkPaths -> - real_path(Name,LinkPaths++Paths,Acc,[ThisFile|Links]) + % windows currently does not allow creation of relative symlinks + % across different drives + case erlang:system_info(os_type) of + {win32, _} -> + real_path(Name,LinkPaths++Paths,[],[ThisFile|Links]); + _ -> + real_path(Name,LinkPaths++Paths,Acc,[ThisFile|Links]) + end end; _ -> real_path(Name,Paths,This,Links) diff --git a/erts/preloaded/src/erlang.erl b/erts/preloaded/src/erlang.erl index 4ff0513321..611a568014 100644 --- a/erts/preloaded/src/erlang.erl +++ b/erts/preloaded/src/erlang.erl @@ -91,7 +91,7 @@ -export([external_size/2, finish_after_on_load/2, finish_loading/1, float/1]). -export([float_to_binary/1, float_to_binary/2, float_to_list/1, float_to_list/2]). --export([fun_info/2, fun_to_list/1, function_exported/3]). +-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([get_module_info/1, get_stacktrace/0, group_leader/0]). @@ -827,6 +827,15 @@ float_to_list(_Float, _Options) -> fun_info(_Fun, _Item) -> erlang:nif_error(undefined). +%% fun_info_mfa/1 +-spec erlang:fun_info_mfa(Fun) -> {Mod, Name, Arity} when + Fun :: function(), + Mod :: atom(), + Name :: atom(), + Arity :: non_neg_integer(). +fun_info_mfa(_Fun) -> + erlang:nif_error(undefined). + %% fun_to_list/1 -spec erlang:fun_to_list(Fun) -> string() when Fun :: function(). @@ -2224,6 +2233,7 @@ tuple_to_list(_Tuple) -> (dynamic_trace) -> none | dtrace | systemtap; (dynamic_trace_probes) -> boolean(); (elib_malloc) -> false; + (eager_check_io) -> boolean(); (ets_limit) -> pos_integer(); (fullsweep_after) -> {fullsweep_after, non_neg_integer()}; (garbage_collection) -> [{atom(), integer()}]; @@ -2242,6 +2252,7 @@ tuple_to_list(_Tuple) -> (modified_timing_level) -> integer() | undefined; (multi_scheduling) -> disabled | blocked | enabled; (multi_scheduling_blockers) -> [PID :: pid()]; + (nif_version) -> string(); (otp_release) -> string(); (port_count) -> non_neg_integer(); (port_limit) -> pos_integer(); diff --git a/erts/preloaded/src/erts.app.src b/erts/preloaded/src/erts.app.src index a15da3a421..345a6ae3be 100644 --- a/erts/preloaded/src/erts.app.src +++ b/erts/preloaded/src/erts.app.src @@ -35,7 +35,6 @@ {registered, []}, {applications, []}, {env, []}, - {mod, {erts, []}}, {runtime_dependencies, ["stdlib-2.0", "kernel-3.0", "sasl-2.4"]} ]}. diff --git a/erts/test/upgrade_SUITE.erl b/erts/test/upgrade_SUITE.erl index d5a920e03d..217056971e 100644 --- a/erts/test/upgrade_SUITE.erl +++ b/erts/test/upgrade_SUITE.erl @@ -37,8 +37,9 @@ %% - hipe does not support any upgrade at all %% - dialyzer requires hipe (in the .app file) %% - typer requires hipe (in the .app file) +%% - erl_interface, jinterface support no upgrade -define(appup_exclude, - [dialyzer,hipe,typer]). + [dialyzer,hipe,typer,erl_interface,jinterface]). init_per_suite(Config) -> %% Check that a real release is running, not e.g. cerl diff --git a/erts/vsn.mk b/erts/vsn.mk index 0db4370ea8..d0dc8f7243 100644 --- a/erts/vsn.mk +++ b/erts/vsn.mk @@ -17,7 +17,7 @@ # %CopyrightEnd% # -VSN = 6.1.2 +VSN = 6.3 # Port number 4365 in 4.2 # Port number 4366 in 4.3 |