diff options
Diffstat (limited to 'erts')
103 files changed, 6403 insertions, 3701 deletions
diff --git a/erts/aclocal.m4 b/erts/aclocal.m4 index 443d8622bf..a1211bbf0c 100644 --- a/erts/aclocal.m4 +++ b/erts/aclocal.m4 @@ -747,9 +747,124 @@ case "$THR_LIB_NAME" in if test $found_win32_winnt = no; then AC_MSG_ERROR([-D_WIN32_WINNT missing in CPPFLAGS]) fi - ethr_have_native_atomics=yes - ethr_have_native_spinlock=yes + AC_DEFINE(ETHR_WIN32_THREADS, 1, [Define if you have win32 threads]) + + have_ilckd=no + AC_MSG_CHECKING([for _InterlockedCompareExchange64()]) + AC_TRY_LINK([ + #define WIN32_LEAN_AND_MEAN + #include <windows.h> + ], + [ + volatile __int64 *var; + _InterlockedCompareExchange64(var, (__int64) 1, (__int64) 0); + return 0; + ], + have_ilckd=yes) + AC_MSG_RESULT([$have_ilckd]) + test $have_ilckd = yes && AC_DEFINE(ETHR_HAVE__INTERLOCKEDCOMPAREEXCHANGE64, 1, [Define if you have _InterlockedCompareExchange64()]) + + AC_CHECK_SIZEOF(void *) + case "$ac_cv_sizeof_void_p-$have_ilckd" in + 8-no) + ethr_have_native_atomics=no + ethr_have_native_spinlock=no;; + *) + ethr_have_native_atomics=yes + ethr_have_native_spinlock=yes;; + esac + + have_ilckd=no + AC_MSG_CHECKING([for _InterlockedDecrement64()]) + AC_TRY_LINK([ + #define WIN32_LEAN_AND_MEAN + #include <windows.h> + ], + [ + volatile __int64 *var; + _InterlockedDecrement64(var); + return 0; + ], + have_ilckd=yes) + AC_MSG_RESULT([$have_ilckd]) + test $have_ilckd = yes && AC_DEFINE(ETHR_HAVE__INTERLOCKEDDECREMENT64, 1, [Define if you have _InterlockedDecrement64()]) + + have_ilckd=no + AC_MSG_CHECKING([for _InterlockedIncrement64()]) + AC_TRY_LINK([ + #define WIN32_LEAN_AND_MEAN + #include <windows.h> + ], + [ + volatile __int64 *var; + _InterlockedIncrement64(var); + return 0; + ], + have_ilckd=yes) + AC_MSG_RESULT([$have_ilckd]) + test $have_ilckd = yes && AC_DEFINE(ETHR_HAVE__INTERLOCKEDINCREMENT64, 1, [Define if you have _InterlockedIncrement64()]) + + have_ilckd=no + AC_MSG_CHECKING([for _InterlockedExchangeAdd64()]) + AC_TRY_LINK([ + #define WIN32_LEAN_AND_MEAN + #include <windows.h> + ], + [ + volatile __int64 *var; + _InterlockedExchangeAdd64(var, (__int64) 1); + return 0; + ], + have_ilckd=yes) + AC_MSG_RESULT([$have_ilckd]) + test $have_ilckd = yes && AC_DEFINE(ETHR_HAVE__INTERLOCKEDEXCHANGEADD64, 1, [Define if you have _InterlockedExchangeAdd64()]) + + have_ilckd=no + AC_MSG_CHECKING([for _InterlockedExchange64()]) + AC_TRY_LINK([ + #define WIN32_LEAN_AND_MEAN + #include <windows.h> + ], + [ + volatile __int64 *var; + _InterlockedExchange64(var, (__int64) 1); + return 0; + ], + have_ilckd=yes) + AC_MSG_RESULT([$have_ilckd]) + test $have_ilckd = yes && AC_DEFINE(ETHR_HAVE__INTERLOCKEDEXCHANGE64, 1, [Define if you have _InterlockedExchange64()]) + + have_ilckd=no + AC_MSG_CHECKING([for _InterlockedAnd64()]) + AC_TRY_LINK([ + #define WIN32_LEAN_AND_MEAN + #include <windows.h> + ], + [ + volatile __int64 *var; + _InterlockedAnd64(var, (__int64) 1); + return 0; + ], + have_ilckd=yes) + AC_MSG_RESULT([$have_ilckd]) + test $have_ilckd = yes && AC_DEFINE(ETHR_HAVE__INTERLOCKEDAND64, 1, [Define if you have _InterlockedAnd64()]) + + have_ilckd=no + AC_MSG_CHECKING([for _InterlockedOr64()]) + AC_TRY_LINK([ + #define WIN32_LEAN_AND_MEAN + #include <windows.h> + ], + [ + volatile __int64 *var; + _InterlockedOr64(var, (__int64) 1); + return 0; + ], + have_ilckd=yes) + AC_MSG_RESULT([$have_ilckd]) + test $have_ilckd = yes && AC_DEFINE(ETHR_HAVE__INTERLOCKEDOR64, 1, [Define if you have _InterlockedOr64()]) + ;; pthread) @@ -1087,6 +1202,28 @@ fi AC_CHECK_SIZEOF(void *) AC_DEFINE_UNQUOTED(ETHR_SIZEOF_PTR, $ac_cv_sizeof_void_p, [Define to the size of pointers]) +AC_CHECK_SIZEOF(int) +AC_DEFINE_UNQUOTED(ETHR_SIZEOF_INT, $ac_cv_sizeof_int, [Define to the size of int]) +AC_CHECK_SIZEOF(long) +AC_DEFINE_UNQUOTED(ETHR_SIZEOF_LONG, $ac_cv_sizeof_long, [Define to the size of long]) +AC_CHECK_SIZEOF(long long) +AC_DEFINE_UNQUOTED(ETHR_SIZEOF_LONG_LONG, $ac_cv_sizeof_long_long, [Define to the size of long long]) +AC_CHECK_SIZEOF(__int64) +AC_DEFINE_UNQUOTED(ETHR_SIZEOF___INT64, $ac_cv_sizeof___int64, [Define to the size of __int64]) + + +case X$erl_xcomp_bigendian in + X) ;; + Xyes|Xno) ac_cv_c_bigendian=$erl_xcomp_bigendian;; + *) AC_MSG_ERROR([Bad erl_xcomp_bigendian value: $erl_xcomp_bigendian]);; +esac + +AC_C_BIGENDIAN + +if test "$ac_cv_c_bigendian" = "yes"; then + AC_DEFINE(ETHR_BIGENDIAN, 1, [Define if bigendian]) +fi + AC_ARG_ENABLE(native-ethr-impls, AS_HELP_STRING([--disable-native-ethr-impls], [disable native ethread implementations]), diff --git a/erts/configure.in b/erts/configure.in index 6e983a07b0..627f734409 100644 --- a/erts/configure.in +++ b/erts/configure.in @@ -580,6 +580,11 @@ AC_SUBST(WFLAGS) AC_SUBST(CFLAG_RUNTIME_LIBRARY_PATH) AC_CHECK_SIZEOF(void *) # Needed for ARCH and smp checks below +if test "x$ac_cv_sizeof_void_p" = x8; then + AC_SUBST(EXTERNAL_WORD_SIZE, 64) +else + AC_SUBST(EXTERNAL_WORD_SIZE, 32) +fi dnl dnl Figure out operating system and cpu architecture diff --git a/erts/emulator/Makefile.in b/erts/emulator/Makefile.in index 6c33e2ca16..f04df354a8 100644 --- a/erts/emulator/Makefile.in +++ b/erts/emulator/Makefile.in @@ -505,8 +505,10 @@ ifdef HIPE_ENABLED OPCODE_TABLES += hipe/hipe_ops.tab endif -$(TTF_DIR)/beam_opcodes.h $(TTF_DIR)/beam_opcodes.c: $(OPCODE_TABLES) - LANG=C $(PERL) utils/beam_makeops -outdir $(TTF_DIR) \ +$(TTF_DIR)/beam_opcodes.h $(TTF_DIR)/beam_opcodes.c: $(OPCODE_TABLES) utils/beam_makeops + LANG=C $(PERL) utils/beam_makeops \ + -wordsize @EXTERNAL_WORD_SIZE@ \ + -outdir $(TTF_DIR) \ -emulator $(OPCODE_TABLES) # bif and atom table diff --git a/erts/emulator/beam/beam_bp.c b/erts/emulator/beam/beam_bp.c index 682f31b83f..31910888d1 100644 --- a/erts/emulator/beam/beam_bp.c +++ b/erts/emulator/beam/beam_bp.c @@ -950,8 +950,8 @@ static int set_function_break(Module *modp, BeamInstr *pc, int bif, MatchSetUnref(old_match_spec); } else { BpDataCount *bdc = (BpDataCount *) bd; - long count = 0; - long res = 0; + erts_aint_t count = 0; + erts_aint_t res = 0; ASSERT(! match_spec); ASSERT(is_nil(tracer_pid)); diff --git a/erts/emulator/beam/beam_bp.h b/erts/emulator/beam/beam_bp.h index ebc171078d..bd8a7249a7 100644 --- a/erts/emulator/beam/beam_bp.h +++ b/erts/emulator/beam/beam_bp.h @@ -157,7 +157,7 @@ do { \ BpData **bds = (BpData **) (pc)[-4]; \ BpDataCount *bdc = NULL; \ Uint ix = bp_sched2ix_proc( (p) ); \ - long count = 0; \ + erts_aint_t count = 0; \ \ ASSERT((pc)[-5] == (BeamInstr) BeamOp(op_i_func_info_IaaI)); \ ASSERT(bds); \ diff --git a/erts/emulator/beam/beam_debug.c b/erts/emulator/beam/beam_debug.c index b0bf14b94f..2855241b91 100644 --- a/erts/emulator/beam/beam_debug.c +++ b/erts/emulator/beam/beam_debug.c @@ -48,7 +48,6 @@ void dbg_bt(Process* p, Eterm* sp); void dbg_where(BeamInstr* addr, Eterm x0, Eterm* reg); -static void print_big(int to, void *to_arg, Eterm* addr); static int print_op(int to, void *to_arg, int op, int size, BeamInstr* addr); Eterm erts_debug_same_2(Process* p, Eterm term1, Eterm term2) @@ -157,6 +156,25 @@ void debug_dump_code(BeamInstr *I, int num) } #endif +BIF_RETTYPE +erts_debug_instructions_0(BIF_ALIST_0) +{ + int i = 0; + Uint needed = num_instructions * 2; + Eterm* hp; + Eterm res = NIL; + + for (i = 0; i < num_instructions; i++) { + needed += 2*strlen(opc[i].name); + } + hp = HAlloc(BIF_P, needed); + for (i = num_instructions-1; i >= 0; i--) { + Eterm s = erts_bld_string_n(&hp, 0, opc[i].name, strlen(opc[i].name)); + res = erts_bld_cons(&hp, 0, s, res); + } + return res; +} + Eterm erts_debug_disassemble_1(Process* p, Eterm addr) { @@ -312,6 +330,7 @@ print_op(int to, void *to_arg, int op, int size, BeamInstr* addr) BeamInstr packed = 0; /* Accumulator for packed operations. */ BeamInstr args[8]; /* Arguments for this instruction. */ BeamInstr* ap; /* Pointer to arguments. */ + BeamInstr* unpacked; /* Unpacked arguments */ start_prog = opc[op].pack; @@ -360,6 +379,12 @@ print_op(int to, void *to_arg, int op, int size, BeamInstr* addr) *ap++ = packed & BEAM_LOOSE_MASK; packed >>= BEAM_LOOSE_SHIFT; break; +#ifdef ARCH_64 + case 'w': /* Shift 32 steps */ + *ap++ = packed & BEAM_WIDE_MASK; + packed >>= BEAM_WIDE_SHIFT; + break; +#endif case 'p': *sp++ = *--ap; break; @@ -386,7 +411,7 @@ print_op(int to, void *to_arg, int op, int size, BeamInstr* addr) break; case 'x': /* x(N) */ if (reg_index(ap[0]) == 0) { - erts_print(to, to_arg, "X[0]"); + erts_print(to, to_arg, "x[0]"); } else { erts_print(to, to_arg, "x(%d)", reg_index(ap[0])); } @@ -506,6 +531,7 @@ print_op(int to, void *to_arg, int op, int size, BeamInstr* addr) ap++; break; case 'P': /* Byte offset into tuple (see beam_load.c) */ + case 'Q': /* Like 'P', but packable */ erts_print(to, to_arg, "%d", (*ap / sizeof(Eterm)) - 1); ap++; break; @@ -526,9 +552,12 @@ print_op(int to, void *to_arg, int op, int size, BeamInstr* addr) * Print more information about certain instructions. */ + unpacked = ap; ap = addr + size; switch (op) { - case op_i_select_val_sfI: + case op_i_select_val_rfI: + case op_i_select_val_xfI: + case op_i_select_val_yfI: { int n = ap[-1]; @@ -540,7 +569,24 @@ print_op(int to, void *to_arg, int op, int size, BeamInstr* addr) } } break; - case op_i_jump_on_val_sfII: + case op_i_select_tuple_arity_rfI: + case op_i_select_tuple_arity_xfI: + case op_i_select_tuple_arity_yfI: + { + int n = ap[-1]; + + while (n > 0) { + Uint arity = arityval(ap[0]); + erts_print(to, to_arg, " {%d} f(" HEXF ")", arity, ap[1]); + ap += 2; + size += 2; + n--; + } + } + break; + case op_i_jump_on_val_rfII: + case op_i_jump_on_val_xfII: + case op_i_jump_on_val_yfII: { int n; for (n = ap[-2]; n > 0; n--) { @@ -550,39 +596,46 @@ print_op(int to, void *to_arg, int op, int size, BeamInstr* addr) } } break; - case op_i_select_big_sf: - while (ap[0]) { - Eterm *bigp = (Eterm *) ap; - int arity = thing_arityval(*bigp); - print_big(to, to_arg, bigp); - size += TermWords(arity+1); - ap += TermWords(arity+1); - erts_print(to, to_arg, " f(" HEXF ") ", ap[0]); - ap++; - size++; + case op_i_jump_on_val_zero_rfI: + case op_i_jump_on_val_zero_xfI: + case op_i_jump_on_val_zero_yfI: + { + int n; + for (n = ap[-1]; n > 0; n--) { + erts_print(to, to_arg, "f(" HEXF ") ", ap[0]); + ap++; + size++; + } + } + break; + case op_i_put_tuple_rI: + case op_i_put_tuple_xI: + case op_i_put_tuple_yI: + { + int n = unpacked[-1]; + + while (n > 0) { + if (!is_header(ap[0])) { + erts_print(to, to_arg, " %T", (Eterm) ap[0]); + } else { + switch ((ap[0] >> 2) & 0x03) { + case R_REG_DEF: + erts_print(to, to_arg, " x(0)"); + break; + case X_REG_DEF: + erts_print(to, to_arg, " x(%d)", ap[0] >> 4); + break; + case Y_REG_DEF: + erts_print(to, to_arg, " y(%d)", ap[0] >> 4); + break; + } + } + ap++, size++, n--; + } } - ap++; - size++; break; } erts_print(to, to_arg, "\n"); return size; } - -static void -print_big(int to, void *to_arg, Eterm* addr) -{ - int i; - int k; - - i = BIG_SIZE(addr); - if (BIG_SIGN(addr)) - erts_print(to, to_arg, "-#integer(%d) = {", i); - else - erts_print(to, to_arg, "#integer(%d) = {", i); - erts_print(to, to_arg, "0x%x", BIG_DIGIT(addr, 0)); - for (k = 1; k < i; k++) - erts_print(to, to_arg, ",0x%x", BIG_DIGIT(addr, k)); - erts_print(to, to_arg, "}"); -} diff --git a/erts/emulator/beam/beam_emu.c b/erts/emulator/beam/beam_emu.c index 8a0e12dd4f..16741aa2d7 100644 --- a/erts/emulator/beam/beam_emu.c +++ b/erts/emulator/beam/beam_emu.c @@ -344,6 +344,8 @@ extern int count_instructions; #define xb(N) (*(Eterm *) (((unsigned char *)reg) + (N))) #define yb(N) (*(Eterm *) (((unsigned char *)E) + (N))) #define fb(N) (*(double *) (((unsigned char *)&(freg[0].fd)) + (N))) +#define Qb(N) (N) +#define Ib(N) (N) #define x(N) reg[N] #define y(N) E[N] #define r(N) x##N @@ -472,6 +474,13 @@ extern int count_instructions; HEAP_SPACE_VERIFIED(need); \ } while (0) +#define TestHeapPutList(Need, Reg) \ + do { \ + TestHeap((Need), 1); \ + PutList(Reg, r(0), r(0), StoreSimpleDest); \ + CHECK_TERM(r(0)); \ + } while (0) + #ifdef HYBRID #ifdef INCREMENTAL #define TestGlobalHeap(Nh, Live, hp) \ @@ -516,6 +525,11 @@ extern int count_instructions; SWAPIN; \ } while (0) +#define PutTuple(Dst, Arity) \ + do { \ + Dst = make_tuple(HTOP); \ + pt_arity = (Arity); \ + } while (0) /* * Check that we haven't used the reductions and jump to function pointed to by @@ -674,6 +688,11 @@ extern int count_instructions; SET_I((BeamInstr *) CallDest); \ Dispatch(); +#define MoveJump(Src) \ + r(0) = (Src); \ + SET_I((BeamInstr *) Arg(0)); \ + Goto(*I); + #define GetList(Src, H, T) do { \ Eterm* tmp_ptr = list_val(Src); \ H = CAR(tmp_ptr); \ @@ -723,16 +742,8 @@ extern int count_instructions; (Dest) = (* (Eterm *) EXPAND_POINTER(tmp_arg1)); \ } while (0) -#define PutTuple(Arity, Src, Dest) \ - ASSERT(is_arity_value(Arity)); \ - Dest = make_tuple(HTOP); \ - HTOP[0] = (Arity); \ - HTOP[1] = (Src); \ - HTOP += 2 - -#define Put(Word) *HTOP++ = (Word) - #define EqualImmed(X, Y, Action) if (X != Y) { Action; } +#define NotEqualImmed(X, Y, Action) if (X == Y) { Action; } #define IsFloat(Src, Fail) if (is_not_float(Src)) { Fail; } @@ -984,8 +995,41 @@ extern int count_instructions; #define IsPid(Src, Fail) if (is_not_pid(Src)) { Fail; } #define IsRef(Src, Fail) if (is_not_ref(Src)) { Fail; } -static BifFunction translate_gc_bif(void* gcf); -static BeamInstr* handle_error(Process* c_p, BeamInstr* pc, Eterm* reg, BifFunction bf); +/* + * process_main() is already huge, so we want to avoid inlining + * into it. Especially functions that are seldom used. + */ +#ifdef __GNUC__ +# define NOINLINE __attribute__((__noinline__)) +#else +# define NOINLINE +#endif + +/* + * The following functions are called directly by process_main(). + * Don't inline them. + */ +static BifFunction translate_gc_bif(void* gcf) NOINLINE; +static BeamInstr* handle_error(Process* c_p, BeamInstr* pc, + Eterm* reg, BifFunction bf) NOINLINE; +static BeamInstr* call_error_handler(Process* p, BeamInstr* ip, + Eterm* reg, Eterm func) NOINLINE; +static BeamInstr* fixed_apply(Process* p, Eterm* reg, Uint arity) NOINLINE; +static BeamInstr* apply(Process* p, Eterm module, Eterm function, + Eterm args, Eterm* reg) NOINLINE; +static int hibernate(Process* c_p, Eterm module, Eterm function, + Eterm args, Eterm* reg) NOINLINE; +static BeamInstr* call_fun(Process* p, int arity, + Eterm* reg, Eterm args) NOINLINE; +static BeamInstr* apply_fun(Process* p, Eterm fun, + Eterm args, Eterm* reg) NOINLINE; +static Eterm new_fun(Process* p, Eterm* reg, + ErlFunEntry* fe, int num_free) NOINLINE; + + +/* + * Functions not directly called by process_main(). OK to inline. + */ static BeamInstr* next_catch(Process* c_p, Eterm *reg); static void terminate_proc(Process* c_p, Eterm Value); static Eterm add_stacktrace(Process* c_p, Eterm Value, Eterm exc); @@ -993,16 +1037,6 @@ static void save_stacktrace(Process* c_p, BeamInstr* pc, Eterm* reg, BifFunction bf, Eterm args); static struct StackTrace * get_trace_from_exc(Eterm exc); static Eterm make_arglist(Process* c_p, Eterm* reg, int a); -static Eterm call_error_handler(Process* p, BeamInstr* ip, Eterm* reg); -static Eterm call_breakpoint_handler(Process* p, BeamInstr* fi, Eterm* reg); -static BeamInstr* fixed_apply(Process* p, Eterm* reg, Uint arity); -static BeamInstr* apply(Process* p, Eterm module, Eterm function, - Eterm args, Eterm* reg); -static int hibernate(Process* c_p, Eterm module, Eterm function, - Eterm args, Eterm* reg); -static BeamInstr* call_fun(Process* p, int arity, Eterm* reg, Eterm args); -static BeamInstr* apply_fun(Process* p, Eterm fun, Eterm args, Eterm* reg); -static Eterm new_fun(Process* p, Eterm* reg, ErlFunEntry* fe, int num_free); #if defined(VXWORKS) static int init_done; @@ -1146,6 +1180,8 @@ void process_main(void) Uint temp_bits; /* Temporary used by BsSkipBits2 & BsGetInteger2 */ + Eterm pt_arity; /* Used by do_put_tuple */ + ERL_BITS_DECLARE_STATEP; /* Has to be last declaration */ @@ -1246,6 +1282,52 @@ void process_main(void) #define STORE_ARITH_RESULT(res) StoreBifResult(2, (res)); #define ARITH_FUNC(name) erts_gc_##name + { + Eterm increment_reg_val; + Eterm increment_val; + Uint live; + Eterm result; + + OpCase(i_increment_yIId): + increment_reg_val = yb(Arg(0)); + goto do_increment; + + OpCase(i_increment_xIId): + increment_reg_val = xb(Arg(0)); + goto do_increment; + + OpCase(i_increment_rIId): + increment_reg_val = r(0); + I--; + + do_increment: + increment_val = Arg(1); + if (is_small(increment_reg_val)) { + Sint i = signed_val(increment_reg_val) + increment_val; + ASSERT(MY_IS_SSMALL(i) == IS_SSMALL(i)); + if (MY_IS_SSMALL(i)) { + result = make_small(i); + store_result: + StoreBifResult(3, result); + } + } + + live = Arg(2); + SWAPOUT; + reg[0] = r(0); + reg[live] = increment_reg_val; + reg[live+1] = make_small(increment_val); + result = erts_gc_mixed_plus(c_p, reg, live); + r(0) = reg[0]; + SWAPIN; + ERTS_HOLE_CHECK(c_p); + if (is_value(result)) { + goto store_result; + } + ASSERT(c_p->freason != BADMATCH || is_value(c_p->fvalue)); + goto find_func_info; + } + OpCase(i_plus_jId): { Eterm result; @@ -1309,6 +1391,52 @@ void process_main(void) } Next(1); + { + Eterm is_eq_exact_lit_val; + + OpCase(i_is_eq_exact_literal_xfc): + is_eq_exact_lit_val = xb(Arg(0)); + I++; + goto do_is_eq_exact_literal; + + OpCase(i_is_eq_exact_literal_yfc): + is_eq_exact_lit_val = yb(Arg(0)); + I++; + goto do_is_eq_exact_literal; + + OpCase(i_is_eq_exact_literal_rfc): + is_eq_exact_lit_val = r(0); + + do_is_eq_exact_literal: + if (!eq(Arg(1), is_eq_exact_lit_val)) { + ClauseFail(); + } + Next(2); + } + + { + Eterm is_ne_exact_lit_val; + + OpCase(i_is_ne_exact_literal_xfc): + is_ne_exact_lit_val = xb(Arg(0)); + I++; + goto do_is_ne_exact_literal; + + OpCase(i_is_ne_exact_literal_yfc): + is_ne_exact_lit_val = yb(Arg(0)); + I++; + goto do_is_ne_exact_literal; + + OpCase(i_is_ne_exact_literal_rfc): + is_ne_exact_lit_val = r(0); + + do_is_ne_exact_literal: + if (eq(Arg(1), is_ne_exact_lit_val)) { + ClauseFail(); + } + Next(2); + } + OpCase(i_move_call_only_fcr): { r(0) = Arg(1); } @@ -1392,6 +1520,17 @@ void process_main(void) NextPF(1, next); } + OpCase(move_x1_c): { + x(1) = Arg(0); + Next(1); + } + + OpCase(move_x2_c): { + x(2) = Arg(0); + Next(1); + } + + OpCase(return): { SET_I(c_p->cp); /* @@ -1405,16 +1544,6 @@ void process_main(void) Goto(*I); } - OpCase(test_heap_1_put_list_Iy): { - BeamInstr *next; - - PreFetch(2, next); - TestHeap(Arg(0), 1); - PutList(yb(Arg(1)), r(0), r(0), StoreSimpleDest); - CHECK_TERM(r(0)); - NextPF(2, next); - } - /* * Send is almost a standard call-BIF with two arguments, except for: * 1) It cannot be traced. @@ -1447,24 +1576,36 @@ void process_main(void) goto find_func_info; } - OpCase(i_element_jssd): { - Eterm index; - Eterm tuple; - - /* - * Inlined version of element/2 for speed. - */ - GetArg2(1, index, tuple); - if (is_small(index) && is_tuple(tuple)) { - Eterm* tp = tuple_val(tuple); - - if ((signed_val(index) >= 1) && - (signed_val(index) <= arityval(*tp))) { - Eterm result = tp[signed_val(index)]; - StoreBifResult(3, result); - } - } - } + { + Eterm element_index; + Eterm element_tuple; + + OpCase(i_element_xjsd): + element_tuple = xb(Arg(0)); + I++; + goto do_element; + + OpCase(i_element_yjsd): + element_tuple = yb(Arg(0)); + I++; + goto do_element; + + OpCase(i_element_rjsd): + element_tuple = r(0); + /* Fall through */ + + do_element: + GetArg1(1, element_index); + if (is_small(element_index) && is_tuple(element_tuple)) { + Eterm* tp = tuple_val(element_tuple); + + if ((signed_val(element_index) >= 1) && + (signed_val(element_index) <= arityval(*tp))) { + Eterm result = tp[signed_val(element_index)]; + StoreBifResult(2, result); + } + } + } /* Fall through */ OpCase(badarg_j): @@ -1472,24 +1613,32 @@ void process_main(void) c_p->freason = BADARG; goto lb_Cl_error; - OpCase(i_fast_element_jIsd): { - Eterm tuple; - - /* - * Inlined version of element/2 for even more speed. - * The first argument is an untagged integer >= 1. - * The second argument is guaranteed to be a register operand. - */ - GetArg1(2, tuple); - if (is_tuple(tuple)) { - Eterm* tp = tuple_val(tuple); - tmp_arg2 = Arg(1); - if (tmp_arg2 <= arityval(*tp)) { - Eterm result = tp[tmp_arg2]; - StoreBifResult(3, result); - } - } + { + Eterm fast_element_tuple; + + OpCase(i_fast_element_rjId): + fast_element_tuple = r(0); + + do_fast_element: + if (is_tuple(fast_element_tuple)) { + Eterm* tp = tuple_val(fast_element_tuple); + Eterm pos = Arg(1); /* Untagged integer >= 1 */ + if (pos <= arityval(*tp)) { + Eterm result = tp[pos]; + StoreBifResult(2, result); + } + } goto badarg; + + OpCase(i_fast_element_xjId): + fast_element_tuple = xb(Arg(0)); + I++; + goto do_fast_element; + + OpCase(i_fast_element_yjId): + fast_element_tuple = yb(Arg(0)); + I++; + goto do_fast_element; } OpCase(catch_yf): @@ -1842,8 +1991,87 @@ void process_main(void) NextPF(0, next); } - OpCase(i_select_val_sfI): - GetArg1(0, tmp_arg1); + + { + Eterm select_val2; + + OpCase(i_select_tuple_arity2_yfAfAf): + select_val2 = yb(Arg(0)); + goto do_select_tuple_arity2; + + OpCase(i_select_tuple_arity2_xfAfAf): + select_val2 = xb(Arg(0)); + goto do_select_tuple_arity2; + + OpCase(i_select_tuple_arity2_rfAfAf): + select_val2 = r(0); + I--; + + do_select_tuple_arity2: + if (is_not_tuple(select_val2)) { + goto select_val2_fail; + } + select_val2 = *tuple_val(select_val2); + goto do_select_val2; + + OpCase(i_select_val2_yfcfcf): + select_val2 = yb(Arg(0)); + goto do_select_val2; + + OpCase(i_select_val2_xfcfcf): + select_val2 = xb(Arg(0)); + goto do_select_val2; + + OpCase(i_select_val2_rfcfcf): + select_val2 = r(0); + I--; + + do_select_val2: + if (select_val2 == Arg(2)) { + I += 2; + } else if (select_val2 == Arg(4)) { + I += 4; + } + + select_val2_fail: + SET_I((BeamInstr *) Arg(1)); + Goto(*I); + } + + { + Eterm select_val; + + OpCase(i_select_tuple_arity_xfI): + select_val = xb(Arg(0)); + goto do_select_tuple_arity; + + OpCase(i_select_tuple_arity_yfI): + select_val = yb(Arg(0)); + goto do_select_tuple_arity; + + OpCase(i_select_tuple_arity_rfI): + select_val = r(0); + I--; + + do_select_tuple_arity: + if (is_tuple(select_val)) { + select_val = *tuple_val(select_val); + goto do_binary_search; + } + SET_I((BeamInstr *) Arg(1)); + Goto(*I); + + OpCase(i_select_val_xfI): + select_val = xb(Arg(0)); + goto do_binary_search; + + OpCase(i_select_val_yfI): + select_val = yb(Arg(0)); + goto do_binary_search; + + OpCase(i_select_val_rfI): + select_val = r(0); + I--; do_binary_search: { @@ -1880,9 +2108,9 @@ void process_main(void) unsigned int boffset = ((unsigned int)bdiff >> 1) & ~(sizeof(struct Pairs)-1); mid = (struct Pairs*)((char*)low + boffset); - if (tmp_arg1 < mid->val) { + if (select_val < mid->val) { high = mid; - } else if (tmp_arg1 > mid->val) { + } else if (select_val > mid->val) { low = mid + 1; } else { SET_I(mid->addr); @@ -1892,16 +2120,28 @@ void process_main(void) SET_I((BeamInstr *) Arg(1)); Goto(*I); } + } - OpCase(i_jump_on_val_zero_sfI): { - Eterm index; - - GetArg1(0, index); - if (is_small(index)) { - index = signed_val(index); - if (index < Arg(2)) { - SET_I((BeamInstr *) (&Arg(3))[index]); + Eterm jump_on_val_zero_index; + + OpCase(i_jump_on_val_zero_yfI): + jump_on_val_zero_index = yb(Arg(0)); + goto do_jump_on_val_zero_index; + + OpCase(i_jump_on_val_zero_xfI): + jump_on_val_zero_index = xb(Arg(0)); + goto do_jump_on_val_zero_index; + + OpCase(i_jump_on_val_zero_rfI): + jump_on_val_zero_index = r(0); + I--; + + do_jump_on_val_zero_index: + if (is_small(jump_on_val_zero_index)) { + jump_on_val_zero_index = signed_val(jump_on_val_zero_index); + if (jump_on_val_zero_index < Arg(2)) { + SET_I((BeamInstr *) (&Arg(3))[jump_on_val_zero_index]); Goto(*I); } } @@ -1909,15 +2149,27 @@ void process_main(void) Goto(*I); } - OpCase(i_jump_on_val_sfII): { - Eterm index; + Eterm jump_on_val_index; - GetArg1(0, index); - if (is_small(index)) { - index = (Uint) (signed_val(index) - Arg(3)); - if (index < Arg(2)) { - SET_I((BeamInstr *) (&Arg(4))[index]); + + OpCase(i_jump_on_val_yfII): + jump_on_val_index = yb(Arg(0)); + goto do_jump_on_val_index; + + OpCase(i_jump_on_val_xfII): + jump_on_val_index = xb(Arg(0)); + goto do_jump_on_val_index; + + OpCase(i_jump_on_val_rfII): + jump_on_val_index = r(0); + I--; + + do_jump_on_val_index: + if (is_small(jump_on_val_index)) { + jump_on_val_index = (Uint) (signed_val(jump_on_val_index) - Arg(3)); + if (jump_on_val_index < Arg(2)) { + SET_I((BeamInstr *) (&Arg(4))[jump_on_val_index]); Goto(*I); } } @@ -1925,6 +2177,32 @@ void process_main(void) Goto(*I); } + do_put_tuple: { + Eterm* hp = HTOP; + + *hp++ = make_arityval(pt_arity); + + do { + Eterm term = *I++; + switch (term & _TAG_IMMED1_MASK) { + case (R_REG_DEF << _TAG_PRIMARY_SIZE) | TAG_PRIMARY_HEADER: + *hp++ = r(0); + break; + case (X_REG_DEF << _TAG_PRIMARY_SIZE) | TAG_PRIMARY_HEADER: + *hp++ = x(term >> _TAG_IMMED1_SIZE); + break; + case (Y_REG_DEF << _TAG_PRIMARY_SIZE) | TAG_PRIMARY_HEADER: + *hp++ = y(term >> _TAG_IMMED1_SIZE); + break; + default: + *hp++ = term; + break; + } + } while (--pt_arity != 0); + HTOP = hp; + Goto(*I); + } + /* * All guards with zero arguments have special instructions: * self/0 @@ -2562,23 +2840,25 @@ void process_main(void) OpCase(i_int_bnot_jsId): { - GetArg1(1, tmp_arg1); - if (is_small(tmp_arg1)) { - tmp_arg1 = make_small(~signed_val(tmp_arg1)); + Eterm bnot_val; + + GetArg1(1, bnot_val); + if (is_small(bnot_val)) { + bnot_val = make_small(~signed_val(bnot_val)); } else { Uint live = Arg(2); SWAPOUT; reg[0] = r(0); - reg[live] = tmp_arg1; - tmp_arg1 = erts_gc_bnot(c_p, reg, live); + reg[live] = bnot_val; + bnot_val = erts_gc_bnot(c_p, reg, live); r(0) = reg[0]; SWAPIN; ERTS_HOLE_CHECK(c_p); - if (is_nil(tmp_arg1)) { + if (is_nil(bnot_val)) { goto lb_Cl_error; } } - StoreBifResult(3, tmp_arg1); + StoreBifResult(3, bnot_val); } badarith: @@ -2833,121 +3113,6 @@ void process_main(void) goto do_schedule1; } - OpCase(i_select_tuple_arity_sfI): - { - GetArg1(0, tmp_arg1); - - if (is_tuple(tmp_arg1)) { - tmp_arg1 = *tuple_val(tmp_arg1); - goto do_binary_search; - } - SET_I((BeamInstr *) Arg(1)); - Goto(*I); - } - - OpCase(i_select_big_sf): - { - Eterm* bigp; - Uint arity; - Eterm* given; - Uint given_arity; - Uint given_size; - - GetArg1(0, tmp_arg1); - if (is_big(tmp_arg1)) { - - /* - * The loader has sorted the bignumbers in descending order - * on the arity word. Therefore, we know that the search - * has failed as soon as we encounter an arity word less than - * the arity word of the given number. There is a zero word - * (less than any valid arity word) stored after the last bignumber. - */ - - given = big_val(tmp_arg1); - given_arity = given[0]; - given_size = thing_arityval(given_arity); - bigp = (Eterm *) &Arg(2); - while ((arity = bigp[0]) > given_arity) { - bigp += (TermWords(thing_arityval(arity) + 1) + 1) * (sizeof(BeamInstr)/sizeof(Eterm)); - } - while (bigp[0] == given_arity) { - if (memcmp(bigp+1, given+1, sizeof(Eterm)*given_size) == 0) { - BeamInstr *tmp = - ((BeamInstr *) (UWord) bigp) + TermWords(given_size + 1); - SET_I((BeamInstr *) *tmp); - Goto(*I); - } - bigp += (TermWords(thing_arityval(arity) + 1) + 1) * (sizeof(BeamInstr)/sizeof(Eterm)); - } - } - - /* - * Failed. - */ - - SET_I((BeamInstr *) Arg(1)); - Goto(*I); - } - -#if defined(ARCH_64) && !HALFWORD_HEAP - OpCase(i_select_float_sfI): - { - Uint f; - int n; - struct ValLabel { - Uint f; - BeamInstr* addr; - }; - struct ValLabel* ptr; - - GetArg1(0, tmp_arg1); - ASSERT(is_float(tmp_arg1)); - f = float_val(tmp_arg1)[1]; - n = Arg(2); - ptr = (struct ValLabel *) &Arg(3); - while (n-- > 0) { - if (ptr->f == f) { - SET_I(ptr->addr); - Goto(*I); - } - ptr++; - } - SET_I((Eterm *) Arg(1)); - Goto(*I); - } -#else - OpCase(i_select_float_sfI): - { - Uint fpart1; - Uint fpart2; - int n; - struct ValLabel { - Uint fpart1; - Uint fpart2; - BeamInstr* addr; - }; - struct ValLabel* ptr; - - GetArg1(0, tmp_arg1); - ASSERT(is_float(tmp_arg1)); - fpart1 = float_val(tmp_arg1)[1]; - fpart2 = float_val(tmp_arg1)[2]; - - n = Arg(2); - ptr = (struct ValLabel *) &Arg(3); - while (n-- > 0) { - if (ptr->fpart1 == fpart1 && ptr->fpart2 == fpart2) { - SET_I(ptr->addr); - Goto(*I); - } - ptr++; - } - SET_I((BeamInstr *) Arg(1)); - Goto(*I); - } -#endif - OpCase(set_tuple_element_sdP): { Eterm element; Eterm tuple; @@ -2993,15 +3158,17 @@ void process_main(void) the first argument. We also handle atom tags in the first argument for backwards compatibility. */ - GetArg2(0, tmp_arg1, tmp_arg2); - c_p->fvalue = tmp_arg2; + Eterm raise_val1; + Eterm raise_val2; + GetArg2(0, raise_val1, raise_val2); + c_p->fvalue = raise_val2; if (c_p->freason == EXC_NULL) { /* a safety check for the R10-0 case; should not happen */ c_p->ftrace = NIL; c_p->freason = EXC_ERROR; } /* for R10-0 code, keep existing c_p->ftrace and hope it's correct */ - switch (tmp_arg1) { + switch (raise_val1) { case am_throw: c_p->freason = EXC_THROWN & ~EXF_SAVETRACE; break; @@ -3017,8 +3184,8 @@ void process_main(void) passed from a user! Currently only expecting generated calls. */ struct StackTrace *s; - c_p->ftrace = tmp_arg1; - s = get_trace_from_exc(tmp_arg1); + c_p->ftrace = raise_val1; + s = get_trace_from_exc(raise_val1); if (s == NULL) { c_p->freason = EXC_ERROR; } else { @@ -3029,11 +3196,24 @@ void process_main(void) goto find_func_info; } - OpCase(badmatch_s): { - GetArg1(0, tmp_arg1); - c_p->fvalue = tmp_arg1; - c_p->freason = BADMATCH; - } + { + Eterm badmatch_val; + + OpCase(badmatch_y): + badmatch_val = yb(Arg(0)); + goto do_badmatch; + + OpCase(badmatch_x): + badmatch_val = xb(Arg(0)); + goto do_badmatch; + + OpCase(badmatch_r): + badmatch_val = r(0); + + do_badmatch: + c_p->fvalue = badmatch_val; + c_p->freason = BADMATCH; + } /* Fall through here */ find_func_info: { @@ -3056,12 +3236,11 @@ void process_main(void) */ SWAPOUT; reg[0] = r(0); - tmp_arg1 = call_error_handler(c_p, I-3, reg); + I = call_error_handler(c_p, I-3, reg, am_undefined_function); r(0) = reg[0]; SWAPIN; - if (tmp_arg1) { - SET_I(c_p->i); - Dispatch(); + if (I) { + Goto(*I); } /* Fall through */ @@ -3084,128 +3263,142 @@ void process_main(void) } } - OpCase(call_nif): - { - /* - * call_nif is always first instruction in function: - * - * I[-3]: Module - * I[-2]: Function - * I[-1]: Arity - * I[0]: &&call_nif - * I[1]: Function pointer to NIF function - * I[2]: Pointer to erl_module_nif - */ - BifFunction vbf; - - c_p->current = I-3; /* current and vbf set to please handle_error */ - SWAPOUT; - c_p->fcalls = FCALLS - 1; - PROCESS_MAIN_CHK_LOCKS(c_p); - tmp_arg2 = I[-1]; - ERTS_SMP_UNREQ_PROC_MAIN_LOCK(c_p); + { + Eterm nif_bif_result; + Eterm bif_nif_arity; - ASSERT(!ERTS_PROC_IS_EXITING(c_p)); - { - typedef Eterm NifF(struct enif_environment_t*, int argc, Eterm argv[]); - NifF* fp = vbf = (NifF*) I[1]; - struct enif_environment_t env; - erts_pre_nif(&env, c_p, (struct erl_module_nif*)I[2]); - reg[0] = r(0); - tmp_arg1 = (*fp)(&env, tmp_arg2, reg); - erts_post_nif(&env); - } - ASSERT(!ERTS_PROC_IS_EXITING(c_p) || is_non_value(tmp_arg1)); - PROCESS_MAIN_CHK_LOCKS(c_p); - goto apply_bif_or_nif_epilogue; - - OpCase(apply_bif): - /* - * At this point, I points to the code[3] in the export entry for - * the BIF: - * - * code[0]: Module - * code[1]: Function - * code[2]: Arity - * code[3]: &&apply_bif - * code[4]: Function pointer to BIF function - */ + OpCase(call_nif): + { + /* + * call_nif is always first instruction in function: + * + * I[-3]: Module + * I[-2]: Function + * I[-1]: Arity + * I[0]: &&call_nif + * I[1]: Function pointer to NIF function + * I[2]: Pointer to erl_module_nif + */ + BifFunction vbf; - c_p->current = I-3; /* In case we apply process_info/1,2 or load_nif/1 */ - c_p->i = I; /* In case we apply check_process_code/2. */ - c_p->arity = 0; /* To allow garbage collection on ourselves - * (check_process_code/2). - */ - SWAPOUT; - c_p->fcalls = FCALLS - 1; - vbf = (BifFunction) Arg(0); - PROCESS_MAIN_CHK_LOCKS(c_p); - tmp_arg2 = I[-1]; - ASSERT(tmp_arg2 <= 3); - ERTS_SMP_UNREQ_PROC_MAIN_LOCK(c_p); - switch (tmp_arg2) { - case 3: + c_p->current = I-3; /* current and vbf set to please handle_error */ + SWAPOUT; + c_p->fcalls = FCALLS - 1; + PROCESS_MAIN_CHK_LOCKS(c_p); + bif_nif_arity = I[-1]; + ERTS_SMP_UNREQ_PROC_MAIN_LOCK(c_p); + + ASSERT(!ERTS_PROC_IS_EXITING(c_p)); { - Eterm (*bf)(Process*, Eterm, Eterm, Eterm, BeamInstr*) = vbf; - ASSERT(!ERTS_PROC_IS_EXITING(c_p)); - tmp_arg1 = (*bf)(c_p, r(0), x(1), x(2), I); - ASSERT(!ERTS_PROC_IS_EXITING(c_p) || is_non_value(tmp_arg1)); - PROCESS_MAIN_CHK_LOCKS(c_p); + typedef Eterm NifF(struct enif_environment_t*, int argc, Eterm argv[]); + NifF* fp = vbf = (NifF*) I[1]; + struct enif_environment_t env; + erts_pre_nif(&env, c_p, (struct erl_module_nif*)I[2]); + reg[0] = r(0); + nif_bif_result = (*fp)(&env, bif_nif_arity, reg); + erts_post_nif(&env); } - break; - case 2: - { - Eterm (*bf)(Process*, Eterm, Eterm, BeamInstr*) = vbf; - ASSERT(!ERTS_PROC_IS_EXITING(c_p)); - tmp_arg1 = (*bf)(c_p, r(0), x(1), I); - ASSERT(!ERTS_PROC_IS_EXITING(c_p) || is_non_value(tmp_arg1)); - PROCESS_MAIN_CHK_LOCKS(c_p); + ASSERT(!ERTS_PROC_IS_EXITING(c_p) || is_non_value(nif_bif_result)); + PROCESS_MAIN_CHK_LOCKS(c_p); + goto apply_bif_or_nif_epilogue; + + OpCase(apply_bif): + /* + * At this point, I points to the code[3] in the export entry for + * the BIF: + * + * code[0]: Module + * code[1]: Function + * code[2]: Arity + * code[3]: &&apply_bif + * code[4]: Function pointer to BIF function + */ + + c_p->current = I-3; /* In case we apply process_info/1,2 or load_nif/1 */ + c_p->i = I; /* In case we apply check_process_code/2. */ + c_p->arity = 0; /* To allow garbage collection on ourselves + * (check_process_code/2). + */ + SWAPOUT; + c_p->fcalls = FCALLS - 1; + vbf = (BifFunction) Arg(0); + PROCESS_MAIN_CHK_LOCKS(c_p); + bif_nif_arity = I[-1]; + ASSERT(bif_nif_arity <= 3); + ERTS_SMP_UNREQ_PROC_MAIN_LOCK(c_p); + switch (bif_nif_arity) { + case 3: + { + Eterm (*bf)(Process*, Eterm, Eterm, Eterm, BeamInstr*) = vbf; + ASSERT(!ERTS_PROC_IS_EXITING(c_p)); + nif_bif_result = (*bf)(c_p, r(0), x(1), x(2), I); + ASSERT(!ERTS_PROC_IS_EXITING(c_p) || + is_non_value(nif_bif_result)); + PROCESS_MAIN_CHK_LOCKS(c_p); + } + break; + case 2: + { + Eterm (*bf)(Process*, Eterm, Eterm, BeamInstr*) = vbf; + ASSERT(!ERTS_PROC_IS_EXITING(c_p)); + nif_bif_result = (*bf)(c_p, r(0), x(1), I); + ASSERT(!ERTS_PROC_IS_EXITING(c_p) || + is_non_value(nif_bif_result)); + PROCESS_MAIN_CHK_LOCKS(c_p); + } + break; + case 1: + { + Eterm (*bf)(Process*, Eterm, BeamInstr*) = vbf; + ASSERT(!ERTS_PROC_IS_EXITING(c_p)); + nif_bif_result = (*bf)(c_p, r(0), I); + ASSERT(!ERTS_PROC_IS_EXITING(c_p) || + is_non_value(nif_bif_result)); + PROCESS_MAIN_CHK_LOCKS(c_p); + } + break; + case 0: + { + Eterm (*bf)(Process*, BeamInstr*) = vbf; + ASSERT(!ERTS_PROC_IS_EXITING(c_p)); + nif_bif_result = (*bf)(c_p, I); + ASSERT(!ERTS_PROC_IS_EXITING(c_p) || + is_non_value(nif_bif_result)); + PROCESS_MAIN_CHK_LOCKS(c_p); + break; + } + default: + erl_exit(1, "apply_bif: invalid arity: %u\n", + bif_nif_arity); } - break; - case 1: - { - Eterm (*bf)(Process*, Eterm, BeamInstr*) = vbf; - ASSERT(!ERTS_PROC_IS_EXITING(c_p)); - tmp_arg1 = (*bf)(c_p, r(0), I); - ASSERT(!ERTS_PROC_IS_EXITING(c_p) || is_non_value(tmp_arg1)); - PROCESS_MAIN_CHK_LOCKS(c_p); + + apply_bif_or_nif_epilogue: + ERTS_SMP_REQ_PROC_MAIN_LOCK(c_p); + ERTS_HOLE_CHECK(c_p); + if (c_p->mbuf) { + reg[0] = r(0); + nif_bif_result = erts_gc_after_bif_call(c_p, nif_bif_result, + reg, bif_nif_arity); + r(0) = reg[0]; } - break; - case 0: - { - Eterm (*bf)(Process*, BeamInstr*) = vbf; - ASSERT(!ERTS_PROC_IS_EXITING(c_p)); - tmp_arg1 = (*bf)(c_p, I); - ASSERT(!ERTS_PROC_IS_EXITING(c_p) || is_non_value(tmp_arg1)); - PROCESS_MAIN_CHK_LOCKS(c_p); - break; + SWAPIN; /* There might have been a garbage collection. */ + FCALLS = c_p->fcalls; + if (is_value(nif_bif_result)) { + r(0) = nif_bif_result; + CHECK_TERM(r(0)); + SET_I(c_p->cp); + Goto(*I); + } else if (c_p->freason == TRAP) { + SET_I(*((BeamInstr **) (UWord) ((c_p)->def_arg_reg + 3))); + r(0) = c_p->def_arg_reg[0]; + x(1) = c_p->def_arg_reg[1]; + x(2) = c_p->def_arg_reg[2]; + Dispatch(); } - } -apply_bif_or_nif_epilogue: - ERTS_SMP_REQ_PROC_MAIN_LOCK(c_p); - ERTS_HOLE_CHECK(c_p); - if (c_p->mbuf) { reg[0] = r(0); - tmp_arg1 = erts_gc_after_bif_call(c_p, tmp_arg1, reg, tmp_arg2); - r(0) = reg[0]; - } - SWAPIN; /* There might have been a garbage collection. */ - FCALLS = c_p->fcalls; - if (is_value(tmp_arg1)) { - r(0) = tmp_arg1; - CHECK_TERM(r(0)); - SET_I(c_p->cp); - Goto(*I); - } else if (c_p->freason == TRAP) { - SET_I(*((BeamInstr **) (UWord) ((c_p)->def_arg_reg + 3))); - r(0) = c_p->def_arg_reg[0]; - x(1) = c_p->def_arg_reg[1]; - x(2) = c_p->def_arg_reg[2]; - Dispatch(); + I = handle_error(c_p, c_p->cp, reg, vbf); + goto post_error_handling; } - reg[0] = r(0); - I = handle_error(c_p, c_p->cp, reg, vbf); - goto post_error_handling; } OpCase(i_get_sd): @@ -3218,11 +3411,26 @@ apply_bif_or_nif_epilogue: StoreBifResult(1, result); } - OpCase(case_end_s): - GetArg1(0, tmp_arg1); - c_p->fvalue = tmp_arg1; - c_p->freason = EXC_CASE_CLAUSE; - goto find_func_info; + { + Eterm case_end_val; + + OpCase(case_end_x): + case_end_val = xb(Arg(0)); + goto do_case_end; + + OpCase(case_end_y): + case_end_val = yb(Arg(0)); + goto do_case_end; + + OpCase(case_end_r): + case_end_val = r(0); + I--; + + do_case_end: + c_p->fvalue = case_end_val; + c_p->freason = EXC_CASE_CLAUSE; + goto find_func_info; + } OpCase(if_end): c_p->freason = EXC_IF_CLAUSE; @@ -3235,10 +3443,13 @@ apply_bif_or_nif_epilogue: } OpCase(try_case_end_s): - GetArg1(0, tmp_arg1); - c_p->fvalue = tmp_arg1; - c_p->freason = EXC_TRY_CLAUSE; - goto find_func_info; + { + Eterm try_case_end_val; + GetArg1(0, try_case_end_val); + c_p->fvalue = try_case_end_val; + c_p->freason = EXC_TRY_CLAUSE; + goto find_func_info; + } /* * Construction of binaries using new instructions. @@ -3786,19 +3997,20 @@ apply_bif_or_nif_epilogue: Eterm header; BeamInstr *next; Uint slots; + Eterm context; OpCase(i_bs_start_match2_rfIId): { - tmp_arg1 = r(0); + context = r(0); do_start_match: slots = Arg(2); - if (!is_boxed(tmp_arg1)) { + if (!is_boxed(context)) { ClauseFail(); } PreFetch(4, next); - header = *boxed_val(tmp_arg1); + header = *boxed_val(context); if (header_is_bin_matchstate(header)) { - ErlBinMatchState* ms = (ErlBinMatchState *) boxed_val(tmp_arg1); + ErlBinMatchState* ms = (ErlBinMatchState *) boxed_val(context); Uint actual_slots = HEADER_NUM_SLOTS(header); ms->save_offset[0] = ms->mb.offset; if (actual_slots < slots) { @@ -3806,8 +4018,8 @@ apply_bif_or_nif_epilogue: Uint live = Arg(1); Uint wordsneeded = ERL_BIN_MATCHSTATE_SIZE(slots); - TestHeapPreserve(wordsneeded, live, tmp_arg1); - ms = (ErlBinMatchState *) boxed_val(tmp_arg1); + TestHeapPreserve(wordsneeded, live, context); + ms = (ErlBinMatchState *) boxed_val(context); dst = (ErlBinMatchState *) HTOP; *dst = *ms; *HTOP = HEADER_BIN_MATCHSTATE(slots); @@ -3819,12 +4031,12 @@ apply_bif_or_nif_epilogue: Eterm result; Uint live = Arg(1); Uint wordsneeded = ERL_BIN_MATCHSTATE_SIZE(slots); - TestHeapPreserve(wordsneeded, live, tmp_arg1); + TestHeapPreserve(wordsneeded, live, context); HEAP_TOP(c_p) = HTOP; #ifdef DEBUG c_p->stop = E; /* Needed for checking in HeapOnlyAlloc(). */ #endif - result = erts_bs_start_match_2(c_p, tmp_arg1, slots); + result = erts_bs_start_match_2(c_p, context, slots); HTOP = HEAP_TOP(c_p); HEAP_SPACE_VERIFIED(0); if (is_non_value(result)) { @@ -3838,12 +4050,12 @@ apply_bif_or_nif_epilogue: NextPF(4, next); } OpCase(i_bs_start_match2_xfIId): { - tmp_arg1 = xb(Arg(0)); + context = xb(Arg(0)); I++; goto do_start_match; } OpCase(i_bs_start_match2_yfIId): { - tmp_arg1 = yb(Arg(0)); + context = yb(Arg(0)); I++; goto do_start_match; } @@ -3936,93 +4148,105 @@ apply_bif_or_nif_epilogue: NextPF(2, next); } + { + Eterm bs_get_integer8_context; + OpCase(i_bs_get_integer_8_rfd): { - tmp_arg1 = r(0); - goto do_bs_get_integer_8; - } + bs_get_integer8_context = r(0); + goto do_bs_get_integer_8; + } OpCase(i_bs_get_integer_8_xfd): { - tmp_arg1 = xb(Arg(0)); - I++; - } + bs_get_integer8_context = xb(Arg(0)); + I++; + } do_bs_get_integer_8: { - ErlBinMatchBuffer *_mb; - Eterm _result; - _mb = ms_matchbuffer(tmp_arg1); - if (_mb->size - _mb->offset < 8) { - ClauseFail(); - } - if (BIT_OFFSET(_mb->offset) != 0) { - _result = erts_bs_get_integer_2(c_p, 8, 0, _mb); - } else { - _result = make_small(_mb->base[BYTE_OFFSET(_mb->offset)]); - _mb->offset += 8; + ErlBinMatchBuffer *_mb; + Eterm _result; + _mb = ms_matchbuffer(bs_get_integer8_context); + if (_mb->size - _mb->offset < 8) { + ClauseFail(); + } + if (BIT_OFFSET(_mb->offset) != 0) { + _result = erts_bs_get_integer_2(c_p, 8, 0, _mb); + } else { + _result = make_small(_mb->base[BYTE_OFFSET(_mb->offset)]); + _mb->offset += 8; + } + StoreBifResult(1, _result); } - StoreBifResult(1, _result); } - OpCase(i_bs_get_integer_16_rfd): { - tmp_arg1 = r(0); + { + Eterm bs_get_integer_16_context; + + OpCase(i_bs_get_integer_16_rfd): + bs_get_integer_16_context = r(0); goto do_bs_get_integer_16; - } - OpCase(i_bs_get_integer_16_xfd): { - tmp_arg1 = xb(Arg(0)); + OpCase(i_bs_get_integer_16_xfd): + bs_get_integer_16_context = xb(Arg(0)); I++; - } - do_bs_get_integer_16: { - ErlBinMatchBuffer *_mb; - Eterm _result; - _mb = ms_matchbuffer(tmp_arg1); - if (_mb->size - _mb->offset < 16) { - ClauseFail(); - } - if (BIT_OFFSET(_mb->offset) != 0) { - _result = erts_bs_get_integer_2(c_p, 16, 0, _mb); - } else { - _result = make_small(get_int16(_mb->base+BYTE_OFFSET(_mb->offset))); - _mb->offset += 16; + do_bs_get_integer_16: + { + ErlBinMatchBuffer *_mb; + Eterm _result; + _mb = ms_matchbuffer(bs_get_integer_16_context); + if (_mb->size - _mb->offset < 16) { + ClauseFail(); + } + if (BIT_OFFSET(_mb->offset) != 0) { + _result = erts_bs_get_integer_2(c_p, 16, 0, _mb); + } else { + _result = make_small(get_int16(_mb->base+BYTE_OFFSET(_mb->offset))); + _mb->offset += 16; + } + StoreBifResult(1, _result); } - StoreBifResult(1, _result); } - OpCase(i_bs_get_integer_32_rfId): { - tmp_arg1 = r(0); + { + Eterm bs_get_integer_32_context; + + OpCase(i_bs_get_integer_32_rfId): + bs_get_integer_32_context = r(0); goto do_bs_get_integer_32; - } + - OpCase(i_bs_get_integer_32_xfId): { - tmp_arg1 = xb(Arg(0)); + OpCase(i_bs_get_integer_32_xfId): + bs_get_integer_32_context = xb(Arg(0)); I++; - } - do_bs_get_integer_32: { - ErlBinMatchBuffer *_mb; - Uint32 _integer; - Eterm _result; - _mb = ms_matchbuffer(tmp_arg1); - if (_mb->size - _mb->offset < 32) { ClauseFail(); } - if (BIT_OFFSET(_mb->offset) != 0) { - _integer = erts_bs_get_unaligned_uint32(_mb); - } else { - _integer = get_int32(_mb->base + _mb->offset/8); - } - _mb->offset += 32; + + do_bs_get_integer_32: + { + ErlBinMatchBuffer *_mb; + Uint32 _integer; + Eterm _result; + _mb = ms_matchbuffer(bs_get_integer_32_context); + if (_mb->size - _mb->offset < 32) { ClauseFail(); } + if (BIT_OFFSET(_mb->offset) != 0) { + _integer = erts_bs_get_unaligned_uint32(_mb); + } else { + _integer = get_int32(_mb->base + _mb->offset/8); + } + _mb->offset += 32; #if !defined(ARCH_64) || HALFWORD_HEAP - if (IS_USMALL(0, _integer)) { + if (IS_USMALL(0, _integer)) { #endif - _result = make_small(_integer); + _result = make_small(_integer); #if !defined(ARCH_64) || HALFWORD_HEAP - } else { - TestHeap(BIG_UINT_HEAP_SIZE, Arg(1)); - _result = uint_to_big((Uint) _integer, HTOP); - HTOP += BIG_UINT_HEAP_SIZE; - HEAP_SPACE_VERIFIED(0); - } + } else { + TestHeap(BIG_UINT_HEAP_SIZE, Arg(1)); + _result = uint_to_big((Uint) _integer, HTOP); + HTOP += BIG_UINT_HEAP_SIZE; + HEAP_SPACE_VERIFIED(0); + } #endif - StoreBifResult(2, _result); + StoreBifResult(2, _result); + } } /* Operands: Size Live Fail Flags Dst */ @@ -4120,54 +4344,64 @@ apply_bif_or_nif_epilogue: StoreBifResult(3, result); } - /* Operands: MatchContext Fail Dst */ + { + Eterm get_utf8_context; + + /* Operands: MatchContext Fail Dst */ OpCase(i_bs_get_utf8_rfd): { - tmp_arg1 = r(0); - goto do_bs_get_utf8; - } + get_utf8_context = r(0); + goto do_bs_get_utf8; + } OpCase(i_bs_get_utf8_xfd): { - tmp_arg1 = xb(Arg(0)); - I++; - } + get_utf8_context = xb(Arg(0)); + I++; + } - /* - * tmp_arg1 = match_context - * Operands: Fail Dst - */ + /* + * get_utf8_context = match_context + * Operands: Fail Dst + */ - do_bs_get_utf8: { - Eterm result = erts_bs_get_utf8(ms_matchbuffer(tmp_arg1)); - if (is_non_value(result)) { - ClauseFail(); + do_bs_get_utf8: { + Eterm result = erts_bs_get_utf8(ms_matchbuffer(get_utf8_context)); + if (is_non_value(result)) { + ClauseFail(); + } + StoreBifResult(1, result); } - StoreBifResult(1, result); } - /* Operands: MatchContext Fail Flags Dst */ + { + Eterm get_utf16_context; + + /* Operands: MatchContext Fail Flags Dst */ OpCase(i_bs_get_utf16_rfId): { - tmp_arg1 = r(0); - goto do_bs_get_utf16; - } + get_utf16_context = r(0); + goto do_bs_get_utf16; + } OpCase(i_bs_get_utf16_xfId): { - tmp_arg1 = xb(Arg(0)); - I++; - } + get_utf16_context = xb(Arg(0)); + I++; + } - /* - * tmp_arg1 = match_context - * Operands: Fail Flags Dst - */ - do_bs_get_utf16: { - Eterm result = erts_bs_get_utf16(ms_matchbuffer(tmp_arg1), Arg(1)); - if (is_non_value(result)) { - ClauseFail(); + /* + * get_utf16_context = match_context + * Operands: Fail Flags Dst + */ + do_bs_get_utf16: { + Eterm result = erts_bs_get_utf16(ms_matchbuffer(get_utf16_context), + Arg(1)); + if (is_non_value(result)) { + ClauseFail(); + } + StoreBifResult(2, result); } - StoreBifResult(2, result); } { + Eterm context_to_binary_context; ErlBinMatchBuffer* mb; ErlSubBin* sb; Uint size; @@ -4176,27 +4410,29 @@ apply_bif_or_nif_epilogue: Uint hole_size; OpCase(bs_context_to_binary_r): { - tmp_arg1 = x0; + context_to_binary_context = x0; I -= 2; goto do_context_to_binary; } /* Unfortunately, inlining can generate this instruction. */ OpCase(bs_context_to_binary_y): { - tmp_arg1 = yb(Arg(0)); + context_to_binary_context = yb(Arg(0)); goto do_context_to_binary0; } OpCase(bs_context_to_binary_x): { - tmp_arg1 = xb(Arg(0)); + context_to_binary_context = xb(Arg(0)); do_context_to_binary0: I--; } do_context_to_binary: - if (is_boxed(tmp_arg1) && header_is_bin_matchstate(*boxed_val(tmp_arg1))) { - ErlBinMatchState* ms = (ErlBinMatchState *) boxed_val(tmp_arg1); + if (is_boxed(context_to_binary_context) && + header_is_bin_matchstate(*boxed_val(context_to_binary_context))) { + ErlBinMatchState* ms; + ms = (ErlBinMatchState *) boxed_val(context_to_binary_context); mb = &ms->mb; offs = ms->save_offset[0]; size = mb->size - offs; @@ -4205,17 +4441,17 @@ apply_bif_or_nif_epilogue: Next(2); OpCase(i_bs_get_binary_all_reuse_rfI): { - tmp_arg1 = x0; + context_to_binary_context = x0; goto do_bs_get_binary_all_reuse; } OpCase(i_bs_get_binary_all_reuse_xfI): { - tmp_arg1 = xb(Arg(0)); + context_to_binary_context = xb(Arg(0)); I++; } do_bs_get_binary_all_reuse: - mb = ms_matchbuffer(tmp_arg1); + mb = ms_matchbuffer(context_to_binary_context); size = mb->size - mb->offset; if (size % Arg(1) != 0) { ClauseFail(); @@ -4224,7 +4460,7 @@ apply_bif_or_nif_epilogue: do_bs_get_binary_all_reuse_common: orig = mb->orig; - sb = (ErlSubBin *) boxed_val(tmp_arg1); + sb = (ErlSubBin *) boxed_val(context_to_binary_context); hole_size = 1 + header_arity(sb->thing_word) - ERL_SUB_BIN_SIZE; sb->thing_word = HEADER_SUB_BIN; sb->size = BYTE_OFFSET(size); @@ -4240,12 +4476,14 @@ apply_bif_or_nif_epilogue: } { + Eterm match_string_context; + OpCase(i_bs_match_string_rfII): { - tmp_arg1 = r(0); + match_string_context = r(0); goto do_bs_match_string; } OpCase(i_bs_match_string_xfII): { - tmp_arg1 = xb(Arg(0)); + match_string_context = xb(Arg(0)); I++; } @@ -4260,7 +4498,7 @@ apply_bif_or_nif_epilogue: PreFetch(3, next); bits = Arg(1); bytes = (byte *) Arg(2); - mb = ms_matchbuffer(tmp_arg1); + mb = ms_matchbuffer(match_string_context); if (mb->size - mb->offset < bits) { ClauseFail(); } @@ -4723,7 +4961,7 @@ apply_bif_or_nif_epilogue: NextPF(2, next); } - OpCase(fmove_new_ld): { + OpCase(fmove_ld): { Eterm fr = Arg(0); Eterm dest = make_float(HTOP); @@ -4753,11 +4991,6 @@ apply_bif_or_nif_epilogue: NextPF(2, next); } - /* - * Old allocating fmove. - */ - - #ifdef NO_FPE_SIGNALS OpCase(fclearerror): OpCase(i_fcheckerror): @@ -4969,12 +5202,11 @@ apply_bif_or_nif_epilogue: OpCase(i_debug_breakpoint): { SWAPOUT; reg[0] = r(0); - tmp_arg1 = call_breakpoint_handler(c_p, I-3, reg); + I = call_error_handler(c_p, I-3, reg, am_breakpoint); r(0) = reg[0]; SWAPIN; - if (tmp_arg1) { - SET_I(c_p->i); - Dispatch(); + if (I) { + Goto(*I); } goto no_error_handler; } @@ -5724,8 +5956,8 @@ build_stacktrace(Process* c_p, Eterm exc) { } -static Eterm -call_error_handler(Process* p, BeamInstr* fi, Eterm* reg) +static BeamInstr* +call_error_handler(Process* p, BeamInstr* fi, Eterm* reg, Eterm func) { Eterm* hp; Export* ep; @@ -5737,14 +5969,12 @@ call_error_handler(Process* p, BeamInstr* fi, Eterm* reg) /* * Search for the error_handler module. */ - ep = erts_find_function(erts_proc_get_error_handler(p), - am_undefined_function, 3); + ep = erts_find_function(erts_proc_get_error_handler(p), func, 3); if (ep == NULL) { /* No error handler */ p->current = fi; p->freason = EXC_UNDEF; return 0; } - p->i = ep->address; /* * Create a list with all arguments in the x registers. @@ -5764,63 +5994,14 @@ call_error_handler(Process* p, BeamInstr* fi, Eterm* reg) } /* - * Set up registers for call to error_handler:undefined_function/3. + * Set up registers for call to error_handler:<func>/3. */ reg[0] = fi[0]; reg[1] = fi[1]; reg[2] = args; - return 1; -} - -static Eterm -call_breakpoint_handler(Process* p, BeamInstr* fi, Eterm* reg) -{ - Eterm* hp; - Export* ep; - int arity; - Eterm args; - Uint sz; - int i; - - /* - * Search for error handler module. - */ - ep = erts_find_function(erts_proc_get_error_handler(p), - am_breakpoint, 3); - if (ep == NULL) { /* No error handler */ - p->current = fi; - p->freason = EXC_UNDEF; - return 0; - } - p->i = ep->address; - - /* - * Create a list with all arguments in the x registers. - */ - - arity = fi[2]; - sz = 2 * arity; - if (HeapWordsLeft(p) < sz) { - erts_garbage_collect(p, sz, reg, arity); - } - hp = HEAP_TOP(p); - HEAP_TOP(p) += sz; - args = NIL; - for (i = arity-1; i >= 0; i--) { - args = CONS(hp, reg[i], args); - hp += 2; - } - - /* - * Set up registers for call to error_handler:breakpoint/3. - */ - reg[0] = fi[0]; - reg[1] = fi[1]; - reg[2] = args; - return 1; + return ep->address; } - static Export* apply_setup_error_handler(Process* p, Eterm module, Eterm function, Uint arity, Eterm* reg) diff --git a/erts/emulator/beam/beam_load.c b/erts/emulator/beam/beam_load.c index df5602b040..4e0d19dafa 100644 --- a/erts/emulator/beam/beam_load.c +++ b/erts/emulator/beam/beam_load.c @@ -1,7 +1,7 @@ /* * %CopyrightBegin% * - * Copyright Ericsson AB 1996-2010. All Rights Reserved. + * Copyright Ericsson AB 1996-2011. All Rights Reserved. * * The contents of this file are subject to the Erlang Public License, * Version 1.1, (the "License"); you may not use this file except in @@ -89,13 +89,12 @@ typedef struct { } Label; /* - * Type for a operand for a generic instruction. + * Type for an operand for a generic instruction. */ typedef struct { unsigned type; /* Type of operand. */ - BeamInstr val; /* Value of operand. */ - Uint bigarity; /* Arity for bignumbers (only). */ + BeamInstr val; /* Value of operand. */ } GenOpArg; /* @@ -326,11 +325,6 @@ typedef struct { Literal* literals; /* Array of literals. */ LiteralPatch* literal_patches; /* Operands that need to be patched. */ Uint total_literal_size; /* Total heap size for all literals. */ - - /* - * Floating point. - */ - int new_float_instructions; /* New allocation scheme for floating point. */ } LoaderState; typedef struct { @@ -476,12 +470,14 @@ static int read_code_header(LoaderState* stp); static int load_code(LoaderState* stp); static GenOp* gen_element(LoaderState* stp, GenOpArg Fail, GenOpArg Index, GenOpArg Tuple, GenOpArg Dst); -static GenOp* gen_split_values(LoaderState* stp, GenOpArg S, GenOpArg Fail, +static GenOp* gen_split_values(LoaderState* stp, GenOpArg S, + GenOpArg TypeFail, GenOpArg Fail, GenOpArg Size, GenOpArg* Rest); static GenOp* gen_select_val(LoaderState* stp, GenOpArg S, GenOpArg Fail, GenOpArg Size, GenOpArg* Rest); -static GenOp* gen_select_big(LoaderState* stp, GenOpArg S, GenOpArg Fail, - GenOpArg Size, GenOpArg* Rest); +static GenOp* gen_select_literals(LoaderState* stp, GenOpArg S, + GenOpArg Fail, GenOpArg Size, + GenOpArg* Rest); static GenOp* const_select_val(LoaderState* stp, GenOpArg S, GenOpArg Fail, GenOpArg Size, GenOpArg* Rest); static GenOp* gen_func_info(LoaderState* stp, GenOpArg mod, GenOpArg Func, @@ -818,7 +814,6 @@ init_state(LoaderState* stp) stp->total_literal_size = 0; stp->literal_patches = 0; stp->string_patches = 0; - stp->new_float_instructions = 0; stp->may_load_nif = 0; stp->on_load = 0; } @@ -1618,7 +1613,6 @@ load_code(LoaderState* stp) BeamInstr val; BeamInstr words = 0; - stp->new_float_instructions = 1; GetTagAndValue(stp, tag, n); VerifyTag(stp, tag, TAG_u); while (n-- > 0) { @@ -1772,7 +1766,7 @@ load_code(LoaderState* stp) } stp->specific_op = specific; - CodeNeed(opc[stp->specific_op].sz+2); /* Extra margin for packing */ + CodeNeed(opc[stp->specific_op].sz+16); /* Extra margin for packing */ code[ci++] = BeamOpCode(stp->specific_op); } @@ -1936,7 +1930,8 @@ load_code(LoaderState* stp) } code[ci++] = (BeamInstr) stp->import[i].bf; break; - case 'P': /* Byte offset into tuple */ + case 'P': /* Byte offset into tuple or stack */ + case 'Q': /* Like 'P', but packable */ VerifyTag(stp, tag, TAG_u); tmp = tmp_op->a[arg].val; code[ci++] = (BeamInstr) ((tmp_op->a[arg].val+1) * sizeof(Eterm)); @@ -1957,84 +1952,6 @@ load_code(LoaderState* stp) } /* - * Load any list arguments using the primitive tags. - */ - - for ( ; arg < tmp_op->arity; arg++) { - switch (tmp_op->a[arg].type) { - case TAG_i: - CodeNeed(1); - code[ci++] = make_small(tmp_op->a[arg].val); - break; - case TAG_u: - case TAG_a: - case TAG_v: - CodeNeed(1); - code[ci++] = tmp_op->a[arg].val; - break; - case TAG_f: - CodeNeed(1); - code[ci] = stp->labels[tmp_op->a[arg].val].patches; - stp->labels[tmp_op->a[arg].val].patches = ci; - ci++; - break; - case TAG_q: - { - Eterm lit; - - lit = stp->literals[tmp_op->a[arg].val].term; - if (is_big(lit)) { - Eterm* bigp; - Eterm *tmp; - Uint size; - Uint term_size; - - bigp = big_val(lit); - term_size = bignum_header_arity(*bigp); - size = TermWords(term_size + 1); - CodeNeed(size); - tmp = (Eterm *) (code + ci); - *tmp++ = *bigp++; - while (term_size-- > 0) { - *tmp++ = *bigp++; - } - ci +=size; - } else if (is_float(lit)) { -#if defined(ARCH_64) && !HALFWORD_HEAP - CodeNeed(1); - code[ci++] = float_val(stp->literals[tmp_op->a[arg].val].term)[1]; -#elif HALFWORD_HEAP - Eterm* fptr; - Uint size; - Eterm *tmp; - - fptr = float_val(stp->literals[tmp_op->a[arg].val].term)+1; - size = TermWords(2); - CodeNeed(size); - tmp = (Eterm *) (code + ci); - *tmp++ = *fptr++; - *tmp = *fptr; - ci += size; -#else - Eterm* fptr; - - fptr = float_val(stp->literals[tmp_op->a[arg].val].term)+1; - CodeNeed(2); - code[ci++] = *fptr++; - code[ci++] = *fptr; -#endif - } else { - LoadError0(stp, "literal is neither float nor big"); - } - } - break; - default: - LoadError1(stp, "unsupported primitive type '%c'", - tag_to_letter[tmp_op->a[arg].type]); - } - } - - /* * The packing engine. */ if (opc[stp->specific_op].pack[0]) { @@ -2057,6 +1974,11 @@ load_code(LoaderState* stp) case '6': /* Shift 16 steps */ packed = (packed << BEAM_LOOSE_SHIFT) | code[--ci]; break; +#ifdef ARCH_64 + case 'w': /* Shift 32 steps */ + packed = (packed << BEAM_WIDE_SHIFT) | code[--ci]; + break; +#endif case 'p': /* Put instruction (from stack). */ code[ci++] = *--sp; break; @@ -2072,6 +1994,58 @@ load_code(LoaderState* stp) } /* + * Load any list arguments using the primitive tags. + */ + + for ( ; arg < tmp_op->arity; arg++) { + switch (tmp_op->a[arg].type) { + case TAG_i: + CodeNeed(1); + code[ci++] = make_small(tmp_op->a[arg].val); + break; + case TAG_u: + case TAG_a: + case TAG_v: + CodeNeed(1); + code[ci++] = tmp_op->a[arg].val; + break; + case TAG_f: + CodeNeed(1); + code[ci] = stp->labels[tmp_op->a[arg].val].patches; + stp->labels[tmp_op->a[arg].val].patches = ci; + ci++; + break; + case TAG_r: + CodeNeed(1); + code[ci++] = (R_REG_DEF << _TAG_PRIMARY_SIZE) | + TAG_PRIMARY_HEADER; + break; + case TAG_x: + CodeNeed(1); + code[ci++] = (tmp_op->a[arg].val << _TAG_IMMED1_SIZE) | + (X_REG_DEF << _TAG_PRIMARY_SIZE) | TAG_PRIMARY_HEADER; + break; + case TAG_y: + CodeNeed(1); + code[ci++] = (tmp_op->a[arg].val << _TAG_IMMED1_SIZE) | + (Y_REG_DEF << _TAG_PRIMARY_SIZE) | TAG_PRIMARY_HEADER; + break; + case TAG_n: + CodeNeed(1); + code[ci++] = NIL; + break; + case TAG_q: + CodeNeed(1); + new_literal_patch(stp, ci); + code[ci++] = tmp_op->a[arg].val; + break; + default: + LoadError1(stp, "unsupported primitive type '%c'", + tag_to_letter[tmp_op->a[arg].type]); + } + } + + /* * Handle a few special cases. */ switch (stp->specific_op) { @@ -2239,11 +2213,12 @@ use_jump_tab(LoaderState* stp, GenOpArg Size, GenOpArg* Rest) } /* - * Predicate to test whether all values in a table are big numbers. + * Predicate to test whether all values in a table are either + * floats or bignums. */ static int -all_values_are_big(LoaderState* stp, GenOpArg Size, GenOpArg* Rest) +floats_or_bignums(LoaderState* stp, GenOpArg Size, GenOpArg* Rest) { int i; @@ -2255,9 +2230,6 @@ all_values_are_big(LoaderState* stp, GenOpArg Size, GenOpArg* Rest) if (Rest[i].type != TAG_q) { return 0; } - if (is_not_big(stp->literals[Rest[i].val].term)) { - return 0; - } if (Rest[i+1].type != TAG_f) { return 0; } @@ -2317,6 +2289,14 @@ mixed_types(LoaderState* stp, GenOpArg Size, GenOpArg* Rest) return 0; } +static int +same_label(LoaderState* stp, GenOpArg Target, GenOpArg Label) +{ + return Target.type = TAG_f && Label.type == TAG_u && + Target.val == Label.val; +} + + /* * Generate an instruction for element/2. */ @@ -2328,23 +2308,23 @@ gen_element(LoaderState* stp, GenOpArg Fail, GenOpArg Index, GenOp* op; NEW_GENOP(stp, op); - op->op = genop_i_element_4; op->arity = 4; - op->a[0] = Fail; - op->a[1] = Index; - op->a[2] = Tuple; - op->a[3] = Dst; op->next = NULL; - /* - * If safe, generate a faster instruction. - */ - if (Index.type == TAG_i && Index.val > 0 && (Tuple.type == TAG_r || Tuple.type == TAG_x || Tuple.type == TAG_y)) { op->op = genop_i_fast_element_4; - op->a[1].type = TAG_u; - op->a[1].val = Index.val; + op->a[0] = Tuple; + op->a[1] = Fail; + op->a[2].type = TAG_u; + op->a[2].val = Index.val; + op->a[3] = Dst; + } else { + op->op = genop_i_element_4; + op->a[0] = Tuple; + op->a[1] = Fail; + op->a[2] = Index; + op->a[3] = Dst; } return op; @@ -2595,8 +2575,6 @@ binary_too_big_bits(LoaderState* stp, GenOpArg Size) return Size.type == TAG_u && (((Size.val+7)/8) >> (8*sizeof(Uint)-3) != 0); } -#define new_float_allocation(Stp) ((Stp)->new_float_instructions) - static GenOp* gen_put_binary(LoaderState* stp, GenOpArg Fail,GenOpArg Size, GenOpArg Unit, GenOpArg Flags, GenOpArg Src) @@ -2809,6 +2787,52 @@ gen_skip_bits2(LoaderState* stp, GenOpArg Fail, GenOpArg Ms, return op; } +static GenOp* +gen_increment(LoaderState* stp, GenOpArg Reg, GenOpArg Integer, + GenOpArg Live, GenOpArg Dst) +{ + GenOp* op; + + NEW_GENOP(stp, op); + op->op = genop_i_increment_4; + op->arity = 4; + op->next = NULL; + op->a[0] = Reg; + op->a[1].type = TAG_u; + op->a[1].val = Integer.val; + op->a[2] = Live; + op->a[3] = Dst; + return op; +} + +static GenOp* +gen_increment_from_minus(LoaderState* stp, GenOpArg Reg, GenOpArg Integer, + GenOpArg Live, GenOpArg Dst) +{ + GenOp* op; + + NEW_GENOP(stp, op); + op->op = genop_i_increment_4; + op->arity = 4; + op->next = NULL; + op->a[0] = Reg; + op->a[1].type = TAG_u; + op->a[1].val = -Integer.val; + op->a[2] = Live; + op->a[3] = Dst; + return op; +} + +/* + * Test whether the negation of the given number is small. + */ +static int +negation_is_small(LoaderState* stp, GenOpArg Int) +{ + return Int.type == TAG_i && IS_SSMALL(-Int.val); +} + + static int smp(LoaderState* stp) { @@ -3000,6 +3024,21 @@ gen_select_tuple_arity(LoaderState* stp, GenOpArg S, GenOpArg Fail, ASSERT(op->a[i].val < op->a[i+2].val); } #endif + + /* + * Use a special-cased instruction if there are only two values. + */ + if (size == 2) { + op->op = genop_i_select_tuple_arity2_6; + op->arity--; + op->a[2].type = TAG_u; + op->a[2].val = arityval(op->a[3].val); + op->a[3] = op->a[4]; + op->a[4].type = TAG_u; + op->a[4].val = arityval(op->a[5].val); + op->a[5] = op->a[6]; + } + return op; } @@ -3009,18 +3048,24 @@ gen_select_tuple_arity(LoaderState* stp, GenOpArg S, GenOpArg Fail, */ static GenOp* -gen_split_values(LoaderState* stp, GenOpArg S, GenOpArg Fail, - GenOpArg Size, GenOpArg* Rest) +gen_split_values(LoaderState* stp, GenOpArg S, GenOpArg TypeFail, + GenOpArg Fail, GenOpArg Size, GenOpArg* Rest) { GenOp* op1; GenOp* op2; GenOp* label; - Uint type; + GenOp* is_integer; int i; ASSERT(Size.val >= 2 && Size.val % 2 == 0); + NEW_GENOP(stp, is_integer); + is_integer->op = genop_is_integer_2; + is_integer->arity = 2; + is_integer->a[0] = TypeFail; + is_integer->a[1] = S; + NEW_GENOP(stp, label); label->op = genop_label_1; label->arity = 1; @@ -3046,15 +3091,13 @@ gen_split_values(LoaderState* stp, GenOpArg S, GenOpArg Fail, op2->a[2].type = TAG_u; op2->a[2].val = 0; - op1->next = label; - label->next = op2; - op2->next = NULL; - - type = Rest[0].type; + /* + * Split the list. + */ ASSERT(Size.type == TAG_u); for (i = 0; i < Size.val; i += 2) { - GenOp* op = (Rest[i].type == type) ? op1 : op2; + GenOp* op = (Rest[i].type == TAG_q) ? op2 : op1; int dst = 3 + op->a[2].val; ASSERT(Rest[i+1].type == TAG_f); @@ -3063,13 +3106,36 @@ gen_split_values(LoaderState* stp, GenOpArg S, GenOpArg Fail, op->arity += 2; op->a[2].val += 2; } + ASSERT(op1->a[2].val > 0); + ASSERT(op2->a[2].val > 0); /* - * None of the instructions should have zero elements in the list. + * Order the instruction sequence appropriately. */ - ASSERT(op1->a[2].val > 0); - ASSERT(op2->a[2].val > 0); + if (TypeFail.val == Fail.val) { + /* + * select_val L1 S ... (small numbers) + * label L1 + * is_integer Fail S + * select_val Fail S ... (bignums) + */ + op1->next = label; + label->next = is_integer; + is_integer->next = op2; + } else { + /* + * is_integer TypeFail S + * select_val L1 S ... (small numbers) + * label L1 + * select_val Fail S ... (bignums) + */ + is_integer->next = op1; + op1->next = label; + label->next = op2; + op1 = is_integer; + } + op2->next = NULL; return op1; } @@ -3091,6 +3157,29 @@ gen_jump_tab(LoaderState* stp, GenOpArg S, GenOpArg Fail, GenOpArg Size, GenOpAr ASSERT(Size.val >= 2 && Size.val % 2 == 0); /* + * If there is only one choice, don't generate a jump table. + */ + if (Size.val == 2) { + GenOp* jump; + + NEW_GENOP(stp, op); + op->arity = 3; + op->op = genop_is_ne_exact_3; + op->a[0] = Rest[1]; + op->a[1] = S; + op->a[2] = Rest[0]; + + NEW_GENOP(stp, jump); + jump->next = NULL; + jump->arity = 1; + jump->op = genop_jump_1; + jump->a[0] = Fail; + + op->next = jump; + return op; + } + + /* * Calculate the minimum and maximum values and size of jump table. */ @@ -3162,8 +3251,9 @@ genopargcompare(GenOpArg* a, GenOpArg* b) } /* - * Generate a select_val instruction. We know that a jump table is not suitable, - * and that all values are of the same type (integer, atoms, floats; never bignums). + * Generate a select_val instruction. We know that a jump table + * is not suitable, and that all values are of the same type + * (integer or atoms). */ static GenOp* @@ -3177,12 +3267,7 @@ gen_select_val(LoaderState* stp, GenOpArg S, GenOpArg Fail, NEW_GENOP(stp, op); op->next = NULL; - if (Rest[0].type != TAG_q) { - op->op = genop_i_select_val_3; - } else { - ASSERT(is_float(stp->literals[Rest[0].val].term)); - op->op = genop_i_select_float_3; - } + op->op = genop_i_select_val_3; GENOP_ARITY(op, arity); op->a[0] = S; op->a[1] = Fail; @@ -3204,19 +3289,19 @@ gen_select_val(LoaderState* stp, GenOpArg S, GenOpArg Fail, } #endif - return op; -} - -/* - * Compare function for qsort(). - */ + /* + * Use a special-cased instruction if there are only two values. + */ + if (size == 2) { + op->op = genop_i_select_val2_6; + op->arity--; + op->a[2] = op->a[3]; + op->a[3] = op->a[4]; + op->a[4] = op->a[5]; + op->a[5] = op->a[6]; + } -static int -genbigcompare(GenOpArg* a, GenOpArg* b) -{ - int val = (int)(b->bigarity - a->bigarity); - - return val != 0 ? val : ((int) (a->val - b->val)); + return op; } /* @@ -3224,37 +3309,35 @@ genbigcompare(GenOpArg* a, GenOpArg* b) */ static GenOp* -gen_select_big(LoaderState* stp, GenOpArg S, GenOpArg Fail, +gen_select_literals(LoaderState* stp, GenOpArg S, GenOpArg Fail, GenOpArg Size, GenOpArg* Rest) { GenOp* op; - int arity = Size.val + 2 + 1; - int size = Size.val / 2; + GenOp* jump; + GenOp** prev_next = &op; + int i; - NEW_GENOP(stp, op); - op->next = NULL; - op->op = genop_i_select_big_2; - GENOP_ARITY(op, arity); - op->a[0] = S; - op->a[1] = Fail; for (i = 0; i < Size.val; i += 2) { + GenOp* op; ASSERT(Rest[i].type == TAG_q); - op->a[i+2] = Rest[i]; - op->a[i+2].bigarity = *big_val(stp->literals[op->a[i+2].val].term); - op->a[i+3] = Rest[i+1]; - } - ASSERT(i+2 == arity-1); - op->a[arity-1].type = TAG_u; - op->a[arity-1].val = 0; - - /* - * Sort the values in descending arity order. - */ - - qsort(op->a+2, size, 2*sizeof(GenOpArg), - (int (*)(const void *, const void *)) genbigcompare); + NEW_GENOP(stp, op); + op->op = genop_is_ne_exact_3; + op->arity = 3; + op->a[0] = Rest[i+1]; + op->a[1] = S; + op->a[2] = Rest[i]; + *prev_next = op; + prev_next = &op->next; + } + + NEW_GENOP(stp, jump); + jump->next = NULL; + jump->op = genop_jump_1; + jump->arity = 1; + jump->a[0] = Fail; + *prev_next = jump; return op; } @@ -3272,7 +3355,6 @@ const_select_val(LoaderState* stp, GenOpArg S, GenOpArg Fail, int i; ASSERT(Size.type == TAG_u); - ASSERT(S.type == TAG_q); NEW_GENOP(stp, op); op->next = NULL; @@ -3283,18 +3365,32 @@ const_select_val(LoaderState* stp, GenOpArg S, GenOpArg Fail, * Search for a literal matching the controlling expression. */ - if (S.type == TAG_q) { - Eterm expr = stp->literals[S.val].term; - for (i = 0; i < Size.val; i += 2) { - if (Rest[i].type == TAG_q) { - Eterm term = stp->literals[Rest[i].val].term; - if (eq(term, expr)) { - ASSERT(Rest[i+1].type == TAG_f); - op->a[0] = Rest[i+1]; - return op; + switch (S.type) { + case TAG_q: + { + Eterm expr = stp->literals[S.val].term; + for (i = 0; i < Size.val; i += 2) { + if (Rest[i].type == TAG_q) { + Eterm term = stp->literals[Rest[i].val].term; + if (eq(term, expr)) { + ASSERT(Rest[i+1].type == TAG_f); + op->a[0] = Rest[i+1]; + return op; + } } } } + break; + case TAG_i: + case TAG_a: + for (i = 0; i < Size.val; i += 2) { + if (Rest[i].val == S.val && Rest[i].type == S.type) { + ASSERT(Rest[i+1].type == TAG_f); + op->a[0] = Rest[i+1]; + return op; + } + } + break; } /* @@ -3477,6 +3573,56 @@ gen_guard_bif3(LoaderState* stp, GenOpArg Fail, GenOpArg Live, GenOpArg Bif, return op; } +static GenOp* +tuple_append_put5(LoaderState* stp, GenOpArg Arity, GenOpArg Dst, + GenOpArg* Puts, GenOpArg S1, GenOpArg S2, GenOpArg S3, + GenOpArg S4, GenOpArg S5) +{ + GenOp* op; + int arity = Arity.val; /* Arity of tuple, not the instruction */ + int i; + + NEW_GENOP(stp, op); + op->next = NULL; + GENOP_ARITY(op, arity+2+5); + op->op = genop_i_put_tuple_2; + op->a[0] = Dst; + op->a[1].type = TAG_u; + op->a[1].val = arity + 5; + for (i = 0; i < arity; i++) { + op->a[i+2] = Puts[i]; + } + op->a[arity+2] = S1; + op->a[arity+3] = S2; + op->a[arity+4] = S3; + op->a[arity+5] = S4; + op->a[arity+6] = S5; + return op; +} + +static GenOp* +tuple_append_put(LoaderState* stp, GenOpArg Arity, GenOpArg Dst, + GenOpArg* Puts, GenOpArg S) +{ + GenOp* op; + int arity = Arity.val; /* Arity of tuple, not the instruction */ + int i; + + NEW_GENOP(stp, op); + op->next = NULL; + GENOP_ARITY(op, arity+2+1); + op->op = genop_i_put_tuple_2; + op->a[0] = Dst; + op->a[1].type = TAG_u; + op->a[1].val = arity + 1; + for (i = 0; i < arity; i++) { + op->a[i+2] = Puts[i]; + } + op->a[arity+2] = S; + return op; +} + + /* * Freeze the code in memory, move the string table into place, @@ -3876,11 +4022,23 @@ transform_engine(LoaderState* st) if (i == 0) goto restart; break; +#if defined(TOP_is_eq) case TOP_is_eq: ASSERT(ap < instr->arity); if (*pc++ != instr->a[ap].val) goto restart; break; +#endif + case TOP_is_type_eq: + mask = *pc++; + + ASSERT(ap < instr->arity); + ASSERT(instr->a[ap].type < BEAM_NUM_TAGS); + if (((1 << instr->a[ap].type) & mask) == 0) + goto restart; + if (*pc++ != instr->a[ap].val) + goto restart; + break; case TOP_is_same_var: ASSERT(ap < instr->arity); i = *pc++; @@ -4001,14 +4159,17 @@ transform_engine(LoaderState* st) case TOP_rest_args: { int n = *pc++; + int formal_arity = gen_opc[instr->op].arity; + int num_vars = n + (instr->arity - formal_arity); + int j = formal_arity; + var = erts_alloc(ERTS_ALC_T_LOADER_TMP, - instr->arity * sizeof(GenOpArg)); + num_vars * sizeof(GenOpArg)); for (i = 0; i < n; i++) { var[i] = def_vars[i]; } - while (i < instr->arity) { - var[i] = instr->a[i]; - i++; + while (i < num_vars) { + var[i++] = instr->a[j++]; } } break; @@ -5315,6 +5476,9 @@ erts_make_stub_module(Process* p, Eterm Mod, Eterm Beam, Eterm Info) if (state.lambdas != state.def_lambdas) { erts_free(ERTS_ALC_T_LOADER_TMP, (void *) state.lambdas); } + erts_free(ERTS_ALC_T_LOADER_TMP, (void *) state.labels); + erts_free(ERTS_ALC_T_LOADER_TMP, (void *) state.atom); + erts_free(ERTS_ALC_T_LOADER_TMP, (void *) state.export); if (bin != NULL) { driver_free_binary(bin); } @@ -5326,9 +5490,18 @@ erts_make_stub_module(Process* p, Eterm Mod, Eterm Beam, Eterm Info) if (code != NULL) { erts_free(ERTS_ALC_T_CODE, code); } + if (state.labels != NULL) { + erts_free(ERTS_ALC_T_LOADER_TMP, (void *) state.labels); + } if (state.lambdas != state.def_lambdas) { erts_free(ERTS_ALC_T_LOADER_TMP, (void *) state.lambdas); } + if (state.atom != NULL) { + erts_free(ERTS_ALC_T_LOADER_TMP, (void *) state.atom); + } + if (state.export != NULL) { + erts_free(ERTS_ALC_T_LOADER_TMP, (void *) state.export); + } if (bin != NULL) { driver_free_binary(bin); } diff --git a/erts/emulator/beam/bif.c b/erts/emulator/beam/bif.c index 6e9755ad48..bb237e378a 100644 --- a/erts/emulator/beam/bif.c +++ b/erts/emulator/beam/bif.c @@ -813,7 +813,7 @@ BIF_RETTYPE spawn_opt_1(BIF_ALIST_1) so.min_heap_size = H_MIN_SIZE; so.min_vheap_size = BIN_VH_MIN_SIZE; so.priority = PRIORITY_NORMAL; - so.max_gen_gcs = (Uint16) erts_smp_atomic_read(&erts_max_gen_gcs); + so.max_gen_gcs = (Uint16) erts_smp_atomic32_read(&erts_max_gen_gcs); so.scheduler = 0; /* @@ -1351,9 +1351,10 @@ BIF_RETTYPE exit_2(BIF_ALIST_2) #ifdef ERTS_SMP if (rp == BIF_P) rp_locks &= ~ERTS_PROC_LOCK_MAIN; - else + if (rp_locks) + erts_smp_proc_unlock(rp, rp_locks); + if (rp != BIF_P) erts_smp_proc_dec_refc(rp); - erts_smp_proc_unlock(rp, rp_locks); #endif /* * We may have exited ourselves and may have to take action. @@ -3269,12 +3270,13 @@ BIF_RETTYPE ports_0(BIF_ALIST_0) erts_smp_mtx_lock(&ports_snapshot_mtx); /* One snapshot at a time */ - erts_smp_atomic_set(&erts_dead_ports_ptr, (long) (port_buf + erts_max_ports)); + erts_smp_atomic_set(&erts_dead_ports_ptr, + (erts_aint_t) (port_buf + erts_max_ports)); next_ss = erts_smp_atomic_inctest(&erts_ports_snapshot); if (erts_smp_atomic_read(&erts_ports_alive) > 0) { - long i; + erts_aint_t i; for (i = erts_max_ports-1; i >= 0; i--) { Port* prt = &erts_port[i]; erts_smp_port_state_lock(prt); @@ -3289,7 +3291,7 @@ BIF_RETTYPE ports_0(BIF_ALIST_0) } dead_ports = (Eterm*)erts_smp_atomic_xchg(&erts_dead_ports_ptr, - (long)NULL); + (erts_aint_t) NULL); erts_smp_mtx_unlock(&ports_snapshot_mtx); ASSERT(pp <= dead_ports); @@ -3300,7 +3302,7 @@ BIF_RETTYPE ports_0(BIF_ALIST_0) ASSERT((alive+dead) <= erts_max_ports); if (alive+dead > 0) { - long i; + erts_aint_t i; Eterm *hp = HAlloc(BIF_P, (alive+dead)*2); for (i = 0; i < alive; i++) { @@ -3796,7 +3798,8 @@ BIF_RETTYPE system_flag_2(BIF_ALIST_2) goto error; } nval = (n > (Sint) ((Uint16) -1)) ? ((Uint16) -1) : ((Uint16) n); - oval = (Uint) erts_smp_atomic_xchg(&erts_max_gen_gcs, (long) nval); + oval = (Uint) erts_smp_atomic32_xchg(&erts_max_gen_gcs, + (erts_aint32_t) nval); BIF_RET(make_small(oval)); } else if (BIF_ARG_1 == am_min_heap_size) { int oval = H_MIN_SIZE; @@ -4139,7 +4142,7 @@ void erts_init_bif(void) erts_smp_spinlock_init(&make_ref_lock, "make_ref"); erts_smp_mtx_init(&ports_snapshot_mtx, "ports_snapshot"); - erts_smp_atomic_init(&erts_dead_ports_ptr, (long)NULL); + erts_smp_atomic_init(&erts_dead_ports_ptr, (erts_aint_t) NULL); /* * bif_return_trap/1 is a hidden BIF that bifs that need to diff --git a/erts/emulator/beam/bif.tab b/erts/emulator/beam/bif.tab index 60b4b1946b..d9dd80fa8b 100644 --- a/erts/emulator/beam/bif.tab +++ b/erts/emulator/beam/bif.tab @@ -660,6 +660,7 @@ bif erts_debug:display/1 bif 'erl.system.debug':display/1 ebif_erts_debug_display_1 bif erts_debug:dist_ext_to_term/2 bif 'erl.system.debug':dist_ext_to_term/2 ebif_erts_debug_dist_ext_to_term_2 +bif erts_debug:instructions/0 # # Monitor testing bif's... diff --git a/erts/emulator/beam/break.c b/erts/emulator/beam/break.c index f339e19761..d255cf3558 100644 --- a/erts/emulator/beam/break.c +++ b/erts/emulator/beam/break.c @@ -98,7 +98,7 @@ process_killer(void) switch(j) { case 'k': if (rp->status == P_WAITING) { - Uint32 rp_locks = ERTS_PROC_LOCKS_XSIG_SEND; + ErtsProcLocks rp_locks = ERTS_PROC_LOCKS_XSIG_SEND; erts_smp_proc_inc_refc(rp); erts_smp_proc_lock(rp, rp_locks); (void) erts_send_exit_signal(NULL, @@ -558,7 +558,7 @@ do_break(void) #endif #ifdef DEBUG case 't': - p_slpq(); + erts_p_slpq(); return; case 'b': bin_check(); @@ -624,9 +624,9 @@ bin_check(void) erts_printf("Process %T holding binary data \n", rp->id); printed = 1; } - erts_printf("0x%08lx orig_size: %ld, norefs = %ld\n", - (unsigned long)bp->val, - (long)bp->val->orig_size, + erts_printf("%p orig_size: %bpd, norefs = %bpd\n", + bp->val, + bp->val->orig_size, erts_smp_atomic_read(&bp->val->refc)); } } diff --git a/erts/emulator/beam/dist.c b/erts/emulator/beam/dist.c index 694460d702..02910fad90 100644 --- a/erts/emulator/beam/dist.c +++ b/erts/emulator/beam/dist.c @@ -917,6 +917,7 @@ int erts_net_message(Port *prt, Eterm token_size; ErtsMonitor *mon; ErtsLink *lnk; + Uint tuple_arity; int res; #ifdef ERTS_DIST_MSG_DBG int orig_len = len; @@ -1003,29 +1004,23 @@ int erts_net_message(Port *prt, #endif if (is_not_tuple(arg) || - (tuple = tuple_val(arg), arityval(*tuple) < 1) || + (tuple = tuple_val(arg), (tuple_arity = arityval(*tuple)) < 1) || is_not_small(tuple[1])) { - erts_dsprintf_buf_t *dsbufp = erts_create_logger_dsbuf(); - erts_dsprintf(dsbufp, "Invalid distribution message: %.200T", arg); - erts_send_error_to_logger_nogl(dsbufp); - goto data_error; + goto invalid_message; } token_size = 0; switch (type = unsigned_val(tuple[1])) { case DOP_LINK: + if (tuple_arity != 3) { + goto invalid_message; + } from = tuple[2]; to = tuple[3]; /* local proc to link to */ if (is_not_pid(from) || is_not_pid(to)) { - erts_dsprintf_buf_t *dsbufp = erts_create_logger_dsbuf(); - PURIFY_MSG("data error"); - erts_dsprintf(dsbufp, - "Invalid DOP_LINK distribution message: %.200T", - arg); - erts_send_error_to_logger_nogl(dsbufp); - goto data_error; + goto invalid_message; } rp = erts_pid2proc_opt(NULL, 0, @@ -1064,8 +1059,14 @@ int erts_net_message(Port *prt, case DOP_UNLINK: { ErtsDistLinkData dld; + if (tuple_arity != 3) { + goto invalid_message; + } from = tuple[2]; to = tuple[3]; + if (is_not_pid(from) || is_not_pid(to)) { + goto invalid_message; + } rp = erts_pid2proc_opt(NULL, 0, to, ERTS_PROC_LOCK_LINK, @@ -1092,11 +1093,19 @@ int erts_net_message(Port *prt, /* A remote process wants to monitor us, we get: {DOP_MONITOR_P, Remote pid, local pid or name, ref} */ Eterm name; + + if (tuple_arity != 4) { + goto invalid_message; + } watcher = tuple[2]; watched = tuple[3]; /* local proc to monitor */ ref = tuple[4]; + if (is_not_ref(ref)) { + goto invalid_message; + } + if (is_atom(watched)) { name = watched; rp = erts_whereis_process(NULL, 0, @@ -1138,10 +1147,17 @@ int erts_net_message(Port *prt, We get {DOP_DEMONITOR_P, Remote pid, Local pid or name, ref}, We need only the ref of course */ + if (tuple_arity != 4) { + goto invalid_message; + } /* watcher = tuple[2]; */ /* watched = tuple[3]; May be an atom in case of monitor name */ ref = tuple[4]; + if(is_not_ref(ref)) { + goto invalid_message; + } + erts_smp_de_links_lock(dep); mon = erts_remove_monitor(&(dep->monitors),ref); erts_smp_de_links_unlock(dep); @@ -1166,10 +1182,11 @@ int erts_net_message(Port *prt, erts_destroy_monitor(mon); break; - case DOP_NODE_LINK: /* XXX never sent ?? */ - break; - case DOP_REG_SEND_TT: + if (tuple_arity != 5) { + goto invalid_message; + } + token_size = size_object(tuple[5]); /* Fall through ... */ case DOP_REG_SEND: @@ -1180,12 +1197,19 @@ int erts_net_message(Port *prt, * There is intentionally no testing of the cookie (it is always '') * from R9B and onwards. */ + if (type != DOP_REG_SEND_TT && tuple_arity != 4) { + goto invalid_message; + } + #ifdef ERTS_DIST_MSG_DBG dist_msg_dbg(&ede, "MSG", buf, orig_len); #endif from = tuple[2]; to = tuple[4]; + if (is_not_pid(from) || is_not_atom(to)){ + goto invalid_message; + } rp = erts_whereis_process(NULL, 0, to, 0, ERTS_P2P_FLG_SMP_INC_REFC); if (rp) { Uint xsize = (type == DOP_REG_SEND @@ -1217,6 +1241,10 @@ int erts_net_message(Port *prt, break; case DOP_SEND_TT: + if (tuple_arity != 4) { + goto invalid_message; + } + token_size = size_object(tuple[4]); /* Fall through ... */ case DOP_SEND: @@ -1227,8 +1255,13 @@ int erts_net_message(Port *prt, #ifdef ERTS_DIST_MSG_DBG dist_msg_dbg(&ede, "MSG", buf, orig_len); #endif - + if (type != DOP_SEND_TT && tuple_arity != 3) { + goto invalid_message; + } to = tuple[3]; + if (is_not_pid(to)) { + goto invalid_message; + } rp = erts_pid2proc_opt(NULL, 0, to, 0, ERTS_P2P_FLG_SMP_INC_REFC); if (rp) { Uint xsize = type == DOP_SEND ? 0 : ERTS_HEAP_FRAG_SIZE(token_size); @@ -1266,11 +1299,19 @@ int erts_net_message(Port *prt, Eterm sysname; ErtsProcLocks rp_locks = ERTS_PROC_LOCKS_MSG_SEND|ERTS_PROC_LOCK_LINK; + if (tuple_arity != 5) { + goto invalid_message; + } + /* watched = tuple[2]; */ /* remote proc which died */ /* watcher = tuple[3]; */ ref = tuple[4]; reason = tuple[5]; + if(is_not_ref(ref)) { + goto invalid_message; + } + erts_smp_de_links_lock(dep); sysname = dep->sysname; mon = erts_remove_monitor(&(dep->monitors), ref); @@ -1317,24 +1358,25 @@ int erts_net_message(Port *prt, ErtsProcLocks rp_locks = ERTS_PROC_LOCK_LINK|ERTS_PROC_LOCKS_XSIG_SEND; /* 'from', which 'to' is linked to, died */ if (type == DOP_EXIT) { - from = tuple[2]; - to = tuple[3]; - reason = tuple[4]; - token = NIL; + if (tuple_arity != 4) { + goto invalid_message; + } + + from = tuple[2]; + to = tuple[3]; + reason = tuple[4]; + token = NIL; } else { - from = tuple[2]; - to = tuple[3]; - token = tuple[4]; - reason = tuple[5]; + if (tuple_arity != 5) { + goto invalid_message; + } + from = tuple[2]; + to = tuple[3]; + token = tuple[4]; + reason = tuple[5]; } - if (is_not_internal_pid(to)) { - erts_dsprintf_buf_t *dsbufp = erts_create_logger_dsbuf(); - PURIFY_MSG("data error"); - erts_dsprintf(dsbufp, - "Invalid DOP_EXIT distribution message: %.200T", - arg); - erts_send_error_to_logger_nogl(dsbufp); - goto data_error; + if (is_not_pid(from) || is_not_internal_pid(to)) { + goto invalid_message; } rp = erts_pid2proc(NULL, 0, to, rp_locks); @@ -1381,15 +1423,24 @@ int erts_net_message(Port *prt, ErtsProcLocks rp_locks = ERTS_PROC_LOCKS_XSIG_SEND; /* 'from' is send an exit signal to 'to' */ if (type == DOP_EXIT2) { - from = tuple[2]; - to = tuple[3]; - reason = tuple[4]; - token = NIL; + if (tuple_arity != 4) { + goto invalid_message; + } + from = tuple[2]; + to = tuple[3]; + reason = tuple[4]; + token = NIL; } else { - from = tuple[2]; - to = tuple[3]; - token = tuple[4]; - reason = tuple[5]; + if (tuple_arity != 5) { + goto invalid_message; + } + from = tuple[2]; + to = tuple[3]; + token = tuple[4]; + reason = tuple[5]; + } + if (is_not_pid(from) || is_not_internal_pid(to)) { + goto invalid_message; } rp = erts_pid2proc_opt(NULL, 0, to, rp_locks, ERTS_P2P_FLG_SMP_INC_REFC); @@ -1408,10 +1459,14 @@ int erts_net_message(Port *prt, break; } case DOP_GROUP_LEADER: + if (tuple_arity != 3) { + goto invalid_message; + } from = tuple[2]; /* Group leader */ to = tuple[3]; /* new member */ - if (is_not_pid(from)) - break; + if (is_not_pid(from) || is_not_pid(to)) { + goto invalid_message; + } rp = erts_pid2proc(NULL, 0, to, ERTS_PROC_LOCK_MAIN); if (!rp) @@ -1420,16 +1475,8 @@ int erts_net_message(Port *prt, erts_smp_proc_unlock(rp, ERTS_PROC_LOCK_MAIN); break; - default: { - erts_dsprintf_buf_t *dsbufp = erts_create_logger_dsbuf(); - erts_dsprintf(dsbufp, - "Illegal value in distribution dispatch switch: " - "%.200T", - arg); - erts_send_error_to_logger_nogl(dsbufp); - PURIFY_MSG("data error"); - goto data_error; - } + default: + goto invalid_message; } erts_cleanup_offheap(&off_heap); @@ -1441,8 +1488,14 @@ int erts_net_message(Port *prt, UnUseTmpHeapNoproc(DIST_CTL_DEFAULT_SIZE); ERTS_SMP_CHK_NO_PROC_LOCKS; return 0; - + invalid_message: + { + erts_dsprintf_buf_t *dsbufp = erts_create_logger_dsbuf(); + erts_dsprintf(dsbufp, "Invalid distribution message: %.200T", arg); + erts_send_error_to_logger_nogl(dsbufp); + } data_error: + PURIFY_MSG("data error"); erts_cleanup_offheap(&off_heap); #ifndef HYBRID /* FIND ME! */ if (ctl != ctl_default) { diff --git a/erts/emulator/beam/dist.h b/erts/emulator/beam/dist.h index 9ccc3e5ba9..695a4fc3fe 100644 --- a/erts/emulator/beam/dist.h +++ b/erts/emulator/beam/dist.h @@ -52,7 +52,7 @@ #define DOP_SEND 2 #define DOP_EXIT 3 #define DOP_UNLINK 4 -#define DOP_NODE_LINK 5 +/* Ancient DOP_NODE_LINK (5) was here, can be reused */ #define DOP_REG_SEND 6 #define DOP_GROUP_LEADER 7 #define DOP_EXIT2 8 @@ -69,7 +69,6 @@ /* distribution trap functions */ extern Export* dsend2_trap; extern Export* dsend3_trap; -/*extern Export* dsend_nosuspend_trap;*/ extern Export* dlink_trap; extern Export* dunlink_trap; extern Export* dmonitor_node_trap; diff --git a/erts/emulator/beam/erl_alloc.c b/erts/emulator/beam/erl_alloc.c index 7793f60f4f..e85e2d7e3f 100644 --- a/erts/emulator/beam/erl_alloc.c +++ b/erts/emulator/beam/erl_alloc.c @@ -1568,7 +1568,6 @@ erts_memory(int *print_to_p, void *print_to_arg, void *proc, Eterm earg) Eterm atoms[sizeof(size)/sizeof(Uint)]; Uint *uintps[sizeof(size)/sizeof(Uint)]; Eterm euints[sizeof(size)/sizeof(Uint)]; - int need_atom; int want_tot_or_sys; int length; Eterm res = THE_NON_VALUE; @@ -1756,7 +1755,6 @@ erts_memory(int *print_to_p, void *print_to_arg, void *proc, Eterm earg) /* Calculate values needed... */ want_tot_or_sys = want.total || want.system; - need_atom = ERTS_MEM_NEED_ALL_ALCU || want.atom; if (ERTS_MEM_NEED_ALL_ALCU) { size.total = 0; diff --git a/erts/emulator/beam/erl_alloc.types b/erts/emulator/beam/erl_alloc.types index 408ffd12f7..b7b9c6a133 100644 --- a/erts/emulator/beam/erl_alloc.types +++ b/erts/emulator/beam/erl_alloc.types @@ -263,6 +263,8 @@ type XPORTS_LIST SHORT_LIVED SYSTEM extra_port_list type PROC_LCK_WTR LONG_LIVED SYSTEM proc_lock_waiter type PROC_LCK_QS LONG_LIVED SYSTEM proc_lock_queues type RUNQ_BLNS LONG_LIVED SYSTEM run_queue_balancing +type MISC_AUX_WORK_Q LONG_LIVED SYSTEM misc_aux_work_q +type MISC_AUX_WORK SHORT_LIVED SYSTEM misc_aux_work +endif # diff --git a/erts/emulator/beam/erl_bif_ddll.c b/erts/emulator/beam/erl_bif_ddll.c index 2c2e283f65..c9cdcb87a6 100644 --- a/erts/emulator/beam/erl_bif_ddll.c +++ b/erts/emulator/beam/erl_bif_ddll.c @@ -1193,7 +1193,7 @@ int erts_ddll_driver_ok(DE_Handle *dh) static void ddll_no_more_references(void *vdh) { DE_Handle *dh = (DE_Handle *) vdh; - int x; + erts_aint_t x; lock_drv_list(); @@ -1604,7 +1604,7 @@ static int do_load_driver_entry(DE_Handle *dh, char *path, char *name) erts_sys_ddll_close(dh->handle); return ERL_DE_LOAD_ERROR_BAD_NAME; } - erts_smp_atomic_init(&(dh->refc), (long) 0); + erts_smp_atomic_init(&(dh->refc), (erts_aint_t) 0); dh->port_count = 0; dh->full_path = erts_alloc(ERTS_ALC_T_DDLL_HANDLE, sys_strlen(path) + 1); sys_strcpy(dh->full_path, path); @@ -1672,7 +1672,7 @@ static int load_driver_entry(DE_Handle **dhp, char *path, char *name) dh->handle = NULL; dh->procs = NULL; dh->port_count = 0; - erts_refc_init(&(dh->refc), (long) 0); + erts_refc_init(&(dh->refc), (erts_aint_t) 0); dh->status = -1; dh->reload_full_path = NULL; dh->reload_driver_name = NULL; diff --git a/erts/emulator/beam/erl_bif_info.c b/erts/emulator/beam/erl_bif_info.c index 75d8db880c..4a717d7271 100644 --- a/erts/emulator/beam/erl_bif_info.c +++ b/erts/emulator/beam/erl_bif_info.c @@ -2020,7 +2020,7 @@ BIF_RETTYPE system_info_1(BIF_ALIST_1) res = TUPLE2(hp, am_sequential_tracer, val); BIF_RET(res); } else if (BIF_ARG_1 == am_garbage_collection){ - Uint val = (Uint) erts_smp_atomic_read(&erts_max_gen_gcs); + Uint val = (Uint) erts_smp_atomic32_read(&erts_max_gen_gcs); Eterm tup; hp = HAlloc(BIF_P, 3+2 + 3+2 + 3+2); @@ -2035,7 +2035,7 @@ BIF_RETTYPE system_info_1(BIF_ALIST_1) BIF_RET(res); } else if (BIF_ARG_1 == am_fullsweep_after){ - Uint val = (Uint) erts_smp_atomic_read(&erts_max_gen_gcs); + Uint val = (Uint) erts_smp_atomic32_read(&erts_max_gen_gcs); hp = HAlloc(BIF_P, 3); res = TUPLE2(hp, am_fullsweep_after, make_small(val)); BIF_RET(res); @@ -3430,8 +3430,8 @@ BIF_RETTYPE erts_debug_set_internal_state_2(BIF_ALIST_2) */ if (ERTS_IS_ATOM_STR("available_internal_state", BIF_ARG_1) && (BIF_ARG_2 == am_true || BIF_ARG_2 == am_false)) { - long on = (long) (BIF_ARG_2 == am_true); - long prev_on = erts_smp_atomic_xchg(&available_internal_state, on); + erts_aint_t on = (erts_aint_t) (BIF_ARG_2 == am_true); + erts_aint_t prev_on = erts_smp_atomic_xchg(&available_internal_state, on); if (on) { erts_dsprintf_buf_t *dsbufp = erts_create_logger_dsbuf(); erts_dsprintf(dsbufp, "Process %T ", BIF_P->id); @@ -3628,7 +3628,7 @@ BIF_RETTYPE erts_debug_set_internal_state_2(BIF_ALIST_2) } else if (ERTS_IS_ATOM_STR("hipe_test_reschedule_suspend", BIF_ARG_1)) { /* Used by hipe test suites */ - long flag = erts_smp_atomic_read(&hipe_test_reschedule_flag); + erts_aint_t flag = erts_smp_atomic_read(&hipe_test_reschedule_flag); if (!flag && BIF_ARG_2 != am_false) { erts_smp_atomic_set(&hipe_test_reschedule_flag, 1); erts_suspend(BIF_P, ERTS_PROC_LOCK_MAIN, NULL); @@ -3703,7 +3703,7 @@ BIF_RETTYPE erts_debug_set_internal_state_2(BIF_ALIST_2) #ifdef ERTS_ENABLE_LOCK_COUNT static Eterm lcnt_build_lock_stats_term(Eterm **hpp, Uint *szp, erts_lcnt_lock_stats_t *stats, Eterm res) { - unsigned long tries = 0, colls = 0; + Uint tries = 0, colls = 0; unsigned long timer_s = 0, timer_ns = 0, timer_n = 0; unsigned int line = 0; @@ -3716,8 +3716,8 @@ static Eterm lcnt_build_lock_stats_term(Eterm **hpp, Uint *szp, erts_lcnt_lock_s * [{{file, line}, {tries, colls, {seconds, nanoseconds, n_blocks}}}] */ - tries = (unsigned long) ethr_atomic_read(&stats->tries); - colls = (unsigned long) ethr_atomic_read(&stats->colls); + tries = (Uint) ethr_atomic_read(&stats->tries); + colls = (Uint) ethr_atomic_read(&stats->colls); line = stats->line; timer_s = stats->timer.s; diff --git a/erts/emulator/beam/erl_bif_timer.c b/erts/emulator/beam/erl_bif_timer.c index 4ae2f6ebf4..3508e8e0dc 100644 --- a/erts/emulator/beam/erl_bif_timer.c +++ b/erts/emulator/beam/erl_bif_timer.c @@ -478,7 +478,7 @@ setup_bif_timer(Uint32 xflags, tab_insert(btm); ASSERT(btm == tab_find(ref)); btm->tm.active = 0; /* MUST be initalized */ - erl_set_timer(&btm->tm, + erts_set_timer(&btm->tm, (ErlTimeoutProc) bif_timer_timeout, (ErlCancelProc) bif_timer_cleanup, (void *) btm, @@ -550,7 +550,7 @@ BIF_RETTYPE cancel_timer_1(BIF_ALIST_1) res = am_false; } else { - Uint left = time_left(&btm->tm); + Uint left = erts_time_left(&btm->tm); if (!(btm->flags & BTM_FLG_BYNAME)) { erts_smp_proc_lock(btm->receiver.proc.ess, ERTS_PROC_LOCK_MSGQ); unlink_proc(btm); @@ -558,7 +558,7 @@ BIF_RETTYPE cancel_timer_1(BIF_ALIST_1) } tab_remove(btm); ASSERT(!tab_find(BIF_ARG_1)); - erl_cancel_timer(&btm->tm); + erts_cancel_timer(&btm->tm); erts_smp_btm_rwunlock(); res = erts_make_integer(left, BIF_P); } @@ -587,7 +587,7 @@ BIF_RETTYPE read_timer_1(BIF_ALIST_1) res = am_false; } else { - Uint left = time_left(&btm->tm); + Uint left = erts_time_left(&btm->tm); res = erts_make_integer(left, BIF_P); } @@ -613,7 +613,7 @@ erts_print_bif_timer_info(int to, void *to_arg) : btm->receiver.proc.ess->id); erts_print(to, to_arg, "=timer:%T\n", receiver); erts_print(to, to_arg, "Message: %T\n", btm->message); - erts_print(to, to_arg, "Time left: %d ms\n", time_left(&btm->tm)); + erts_print(to, to_arg, "Time left: %d ms\n", erts_time_left(&btm->tm)); } } @@ -640,7 +640,7 @@ erts_cancel_bif_timers(Process *p, ErtsProcLocks plocks) tab_remove(btm); tmp_btm = btm; btm = btm->receiver.proc.next; - erl_cancel_timer(&tmp_btm->tm); + erts_cancel_timer(&tmp_btm->tm); } p->bif_timers = NULL; diff --git a/erts/emulator/beam/erl_bits.c b/erts/emulator/beam/erl_bits.c index 88d2c06246..6f8a7436d5 100644 --- a/erts/emulator/beam/erl_bits.c +++ b/erts/emulator/beam/erl_bits.c @@ -555,10 +555,11 @@ fmt_int(byte *buf, Uint sz, Eterm val, Uint size, Uint flags) { unsigned long offs; - ASSERT(size != 0); offs = BIT_OFFSET(size); if (is_small(val)) { Sint v = signed_val(val); + + ASSERT(size != 0); /* Tested by caller */ if (flags & BSF_LITTLE) { /* Little endian */ sz--; COPY_VAL(buf,1,v,sz); @@ -578,6 +579,9 @@ fmt_int(byte *buf, Uint sz, Eterm val, Uint size, Uint flags) ErtsDigit* dp = big_v(val); int n = MIN(sz,ds); + if (size == 0) { + return 0; + } if (flags & BSF_LITTLE) { sz -= n; /* pad with this amount */ if (sign) { @@ -729,15 +733,13 @@ erts_new_bs_put_integer(ERL_BITS_PROTO_3(Eterm arg, Uint num_bits, unsigned flag Uint b; byte *iptr; - if (num_bits == 0) { - return 1; - } - bit_offset = BIT_OFFSET(bin_offset); if (is_small(arg)) { Uint rbits = 8 - bit_offset; - if (bit_offset + num_bits <= 8) { + if (num_bits == 0) { + return 1; + } else if (bit_offset + num_bits <= 8) { /* * All bits are in the same byte. */ diff --git a/erts/emulator/beam/erl_cpu_topology.c b/erts/emulator/beam/erl_cpu_topology.c index db95c4a5d4..8a6b4d8d6c 100644 --- a/erts/emulator/beam/erl_cpu_topology.c +++ b/erts/emulator/beam/erl_cpu_topology.c @@ -487,7 +487,7 @@ erts_sched_check_cpu_bind_post_suspend(ErtsSchedulerData *esdp) /* Make sure we check if we should bind to a cpu or not... */ if (esdp->run_queue->flags & ERTS_RUNQ_FLG_SHARED_RUNQ) - erts_smp_atomic_set(&esdp->chk_cpu_bind, 1); + erts_smp_atomic32_set(&esdp->chk_cpu_bind, 1); else esdp->run_queue->flags |= ERTS_RUNQ_FLG_CHK_CPU_BIND; } @@ -503,7 +503,7 @@ erts_sched_check_cpu_bind(ErtsSchedulerData *esdp) erts_cpu_groups_callback_call_t *cgcc; #ifdef ERTS_SMP if (erts_common_run_queue) - erts_smp_atomic_set(&esdp->chk_cpu_bind, 0); + erts_smp_atomic32_set(&esdp->chk_cpu_bind, 0); else { esdp->run_queue->flags &= ~ERTS_RUNQ_FLG_CHK_CPU_BIND; } diff --git a/erts/emulator/beam/erl_db.c b/erts/emulator/beam/erl_db.c index 8577354d27..3173d3510e 100644 --- a/erts/emulator/beam/erl_db.c +++ b/erts/emulator/beam/erl_db.c @@ -219,47 +219,68 @@ Export ets_select_continue_exp; * Static traps */ static Export ets_delete_continue_exp; - -static ERTS_INLINE DbTable* db_ref(DbTable* tb, db_lock_kind_t kind) -{ - if (tb != NULL && kind != LCK_READ) { - erts_refc_inc(&tb->common.ref, 2); - } - return tb; -} - -static ERTS_INLINE DbTable* db_unref(DbTable* tb, db_lock_kind_t kind) + +static void +free_dbtable(DbTable* tb) { - if (kind != LCK_READ && !erts_refc_dectest(&tb->common.ref, 0)) { #ifdef HARDDEBUG if (erts_smp_atomic_read(&tb->common.memory_size) != sizeof(DbTable)) { - erts_fprintf(stderr, "ets: db_unref memory remain=%ld fix=%x\n", - erts_smp_atomic_read(&tb->common.memory_size)-sizeof(DbTable), + erts_fprintf(stderr, "ets: free_dbtable memory remain=%ld fix=%x\n", + erts_smp_atomic_read(&tb->common.memory_size)-sizeof(DbTable), tb->common.fixations); } - erts_fprintf(stderr, "ets: db_unref(%T) deleted!!!\r\n", + erts_fprintf(stderr, "ets: free_dbtable(%T) deleted!!!\r\n", tb->common.id); - erts_fprintf(stderr, "ets: db_unref: meta_pid_to_tab common.memory_size = %ld\n", + erts_fprintf(stderr, "ets: free_dbtable: meta_pid_to_tab common.memory_size = %ld\n", erts_smp_atomic_read(&meta_pid_to_tab->common.memory_size)); print_table(ERTS_PRINT_STDOUT, NULL, 1, meta_pid_to_tab); - erts_fprintf(stderr, "ets: db_unref: meta_pid_to_fixed_tab common.memory_size = %ld\n", + erts_fprintf(stderr, "ets: free_dbtable: meta_pid_to_fixed_tab common.memory_size = %ld\n", erts_smp_atomic_read(&meta_pid_to_fixed_tab->common.memory_size)); print_table(ERTS_PRINT_STDOUT, NULL, 1, meta_pid_to_fixed_tab); - #endif #ifdef ERTS_SMP erts_smp_rwmtx_destroy(&tb->common.rwlock); erts_smp_mtx_destroy(&tb->common.fixlock); #endif ASSERT(is_immed(tb->common.heir_data)); - erts_db_free(ERTS_ALC_T_DB_TABLE, tb, (void *) tb, sizeof(DbTable)); + erts_db_free(ERTS_ALC_T_DB_TABLE, tb, (void *) tb, sizeof(DbTable)); ERTS_ETS_MISC_MEM_ADD(-sizeof(DbTable)); - return NULL; - } - return tb; +} + +#ifdef ERTS_SMP +static void +chk_free_dbtable(void *vtb) +{ + DbTable * tb = (DbTable *) vtb; + ERTS_THR_MEMORY_BARRIER; + if (erts_refc_dectest(&tb->common.ref, 0) == 0) + free_dbtable(tb); +} +#endif + +static void schedule_free_dbtable(DbTable* tb) +{ + /* + * NON-SMP case: Caller is *not* allowed to access the *tb + * structure after this function has returned! + * SMP case: Caller is allowed to access the *tb structure + * until the bif has returned (we typically + * need to unlock the table lock after this + * function has returned). + */ +#ifdef ERTS_SMP + int scheds = erts_get_max_no_executing_schedulers(); + ASSERT(scheds >= 1); + ASSERT(erts_refc_read(&tb->common.ref, 0) == 0); + erts_refc_init(&tb->common.ref, scheds); + ERTS_THR_MEMORY_BARRIER; + erts_smp_schedule_misc_aux_work(0, scheds, chk_free_dbtable, tb); +#else + free_dbtable(tb); +#endif } static ERTS_INLINE void db_init_lock(DbTable* tb, int use_frequent_read_lock, @@ -270,8 +291,6 @@ static ERTS_INLINE void db_init_lock(DbTable* tb, int use_frequent_read_lock, if (use_frequent_read_lock) rwmtx_opt.type = ERTS_SMP_RWMTX_TYPE_FREQUENT_READ; #endif - erts_refc_init(&tb->common.ref, 1); - erts_refc_init(&tb->common.fixref, 0); #ifdef ERTS_SMP erts_smp_rwmtx_init_opt_x(&tb->common.rwlock, &rwmtx_opt, rwname, tb->common.the_name); @@ -280,7 +299,7 @@ static ERTS_INLINE void db_init_lock(DbTable* tb, int use_frequent_read_lock, #endif } -static ERTS_INLINE void db_lock_take_over_ref(DbTable* tb, db_lock_kind_t kind) +static ERTS_INLINE void db_lock(DbTable* tb, db_lock_kind_t kind) { #ifdef ERTS_SMP ASSERT(tb != meta_pid_to_tab && tb != meta_pid_to_fixed_tab); @@ -308,16 +327,13 @@ static ERTS_INLINE void db_lock_take_over_ref(DbTable* tb, db_lock_kind_t kind) #endif } -static ERTS_INLINE void db_lock(DbTable* tb, db_lock_kind_t kind) -{ - (void) db_ref(tb, kind); -#ifdef ERTS_SMP - db_lock_take_over_ref(tb, kind); -#endif -} - static ERTS_INLINE void db_unlock(DbTable* tb, db_lock_kind_t kind) { + /* + * In NON-SMP case tb may refer to an already deallocated + * DbTable structure. That is, ONLY the SMP case is allowed + * to follow the tb pointer! + */ #ifdef ERTS_SMP ASSERT(tb != meta_pid_to_tab && tb != meta_pid_to_fixed_tab); @@ -344,7 +360,6 @@ static ERTS_INLINE void db_unlock(DbTable* tb, db_lock_kind_t kind) } } #endif - (void) db_unref(tb, kind); /* May delete table... */ } @@ -371,6 +386,13 @@ DbTable* db_get_table_aux(Process *p, DbTable *tb = NULL; erts_smp_rwmtx_t *mtl = NULL; + /* + * IMPORTANT: Only scheduler threads are allowed + * to access tables. Memory management + * depend on it. + */ + ASSERT(erts_get_scheduler_data()); + if (is_small(id)) { Uint slot = unsigned_val(id) & meta_main_tab_slot_mask; if (!meta_already_locked) { @@ -384,12 +406,8 @@ DbTable* db_get_table_aux(Process *p, || erts_lc_rwmtx_is_rwlocked(test_mtl)); } #endif - if (slot < db_max_tabs && IS_SLOT_ALIVE(slot)) { - /* SMP: inc to prevent race, between unlock of meta_main_tab_lock - * and the table locking outside the meta_main_tab_lock - */ - tb = db_ref(meta_main_tab[slot].u.tb, kind); - } + if (slot < db_max_tabs && IS_SLOT_ALIVE(slot)) + tb = meta_main_tab[slot].u.tb; } else if (is_atom(id)) { struct meta_name_tab_entry* bucket = meta_name_tab_bucket(id,&mtl); @@ -403,16 +421,15 @@ DbTable* db_get_table_aux(Process *p, if (bucket->pu.tb != NULL) { if (is_atom(bucket->u.name_atom)) { /* single */ - if (bucket->u.name_atom == id) { - tb = db_ref(bucket->pu.tb, kind); - } + if (bucket->u.name_atom == id) + tb = bucket->pu.tb; } else { /* multi */ Uint cnt = unsigned_val(bucket->u.mcnt); Uint i; for (i=0; i<cnt; i++) { if (bucket->pu.mvec[i].u.name_atom == id) { - tb = db_ref(bucket->pu.mvec[i].pu.tb, kind); + tb = bucket->pu.mvec[i].pu.tb; break; } } @@ -420,7 +437,7 @@ DbTable* db_get_table_aux(Process *p, } } if (tb) { - db_lock_take_over_ref(tb, kind); + db_lock(tb, kind); if (tb->common.id != id || ((tb->common.status & what) == 0 && p->id != tb->common.owner)) { db_unlock(tb, kind); @@ -594,11 +611,11 @@ done: */ static ERTS_INLINE void local_fix_table(DbTable* tb) { - erts_refc_inc(&tb->common.fixref, 1); + erts_refc_inc(&tb->common.ref, 1); } static ERTS_INLINE void local_unfix_table(DbTable* tb) { - if (erts_refc_dectest(&tb->common.fixref, 0) == 0) { + if (erts_refc_dectest(&tb->common.ref, 0) == 0) { ASSERT(IS_HASH_TABLE(tb->common.status)); db_unfix_table_hash(&(tb->hash)); } @@ -1414,6 +1431,7 @@ BIF_RETTYPE ets_new_2(BIF_ALIST_2) tb->common.type = status & ERTS_ETS_TABLE_TYPES; /* Note, 'type' is *read only* from now on... */ #endif + erts_refc_init(&tb->common.ref, 0); db_init_lock(tb, status & (DB_FINE_LOCKED|DB_FREQ_READ), "db_tab", "db_tab_fix"); tb->common.keypos = keypos; @@ -1436,8 +1454,7 @@ BIF_RETTYPE ets_new_2(BIF_ALIST_2) "** Too many db tables **\n"); free_heir_data(tb); tb->common.meth->db_free_table(tb); - erts_db_free(ERTS_ALC_T_DB_TABLE, tb, (void *) tb, sizeof(DbTable)); - ERTS_ETS_MISC_MEM_ADD(-sizeof(DbTable)); + free_dbtable(tb); BIF_ERROR(BIF_P, SYSTEM_LIMIT); } @@ -1471,9 +1488,10 @@ BIF_RETTYPE ets_new_2(BIF_ALIST_2) free_slot(slot); erts_smp_rwmtx_rwunlock(mmtl); - db_lock_take_over_ref(tb,LCK_WRITE); + db_lock(tb,LCK_WRITE); free_heir_data(tb); tb->common.meth->db_free_table(tb); + schedule_free_dbtable(tb); db_unlock(tb,LCK_WRITE); BIF_ERROR(BIF_P, BADARG); } @@ -2845,8 +2863,7 @@ void init_db(void) meta_pid_to_tab->common.meth = &db_hash; meta_pid_to_tab->common.compress = 0; - erts_refc_init(&meta_pid_to_tab->common.ref, 1); - erts_refc_init(&meta_pid_to_tab->common.fixref, 0); + erts_refc_init(&meta_pid_to_tab->common.ref, 0); /* Neither rwlock or fixlock used db_init_lock(meta_pid_to_tab, "meta_pid_to_tab", "meta_pid_to_tab_FIX");*/ @@ -2878,8 +2895,7 @@ void init_db(void) meta_pid_to_fixed_tab->common.meth = &db_hash; meta_pid_to_fixed_tab->common.compress = 0; - erts_refc_init(&meta_pid_to_fixed_tab->common.ref, 1); - erts_refc_init(&meta_pid_to_fixed_tab->common.fixref, 0); + erts_refc_init(&meta_pid_to_fixed_tab->common.ref, 0); /* Neither rwlock or fixlock used db_init_lock(meta_pid_to_fixed_tab, "meta_pid_to_fixed_tab", "meta_pid_to_fixed_tab_FIX");*/ @@ -3037,12 +3053,10 @@ retry: to_pid, to_locks, ERTS_P2P_FLG_TRY_LOCK); if (to_proc == ERTS_PROC_LOCK_BUSY) { - db_ref(tb, LCK_NONE); /* while unlocked */ db_unlock(tb,LCK_WRITE); to_proc = erts_pid2proc(p, ERTS_PROC_LOCK_MAIN, to_pid, to_locks); db_lock(tb,LCK_WRITE); - tb = db_unref(tb, LCK_NONE); ASSERT(tb != NULL); if (tb->common.owner != p->id) { @@ -3153,13 +3167,13 @@ erts_db_process_exiting(Process *c_p, ErtsProcLocks c_p_locks) erts_smp_rwmtx_t *mmtl = get_meta_main_tab_lock(ix); erts_smp_rwmtx_rlock(mmtl); if (!IS_SLOT_FREE(ix)) { - tb = db_ref(GET_ANY_SLOT_TAB(ix), LCK_WRITE); + tb = GET_ANY_SLOT_TAB(ix); ASSERT(tb); } erts_smp_rwmtx_runlock(mmtl); if (tb) { int do_yield; - db_lock_take_over_ref(tb, LCK_WRITE); + db_lock(tb, LCK_WRITE); /* Ownership may have changed since we looked up the table. */ if (tb->common.owner != pid) { @@ -3241,7 +3255,7 @@ erts_db_process_exiting(Process *c_p, ErtsProcLocks c_p_locks) erts_smp_rwmtx_t *mmtl = get_meta_main_tab_lock(ix); erts_smp_rwmtx_rlock(mmtl); if (IS_SLOT_ALIVE(ix)) { - tb = db_ref(meta_main_tab[ix].u.tb, LCK_WRITE_REC); + tb = meta_main_tab[ix].u.tb; ASSERT(tb); } erts_smp_rwmtx_runlock(mmtl); @@ -3249,7 +3263,7 @@ erts_db_process_exiting(Process *c_p, ErtsProcLocks c_p_locks) int reds; DbFixation** pp; - db_lock_take_over_ref(tb, LCK_WRITE_REC); + db_lock(tb, LCK_WRITE_REC); #ifdef ERTS_SMP erts_smp_mtx_lock(&tb->common.fixlock); #endif @@ -3259,8 +3273,8 @@ erts_db_process_exiting(Process *c_p, ErtsProcLocks c_p_locks) pp = &(*pp)->next) { if ((*pp)->pid == pid) { DbFixation* fix = *pp; - long diff = -(long)fix->counter; - erts_refc_add(&tb->common.fixref,diff,0); + erts_aint_t diff = -((erts_aint_t) fix->counter); + erts_refc_add(&tb->common.ref,diff,0); *pp = fix->next; erts_db_free(ERTS_ALC_T_DB_FIXATION, tb, fix, sizeof(DbFixation)); @@ -3335,7 +3349,7 @@ static void fix_table_locked(Process* p, DbTable* tb) #ifdef ERTS_SMP erts_smp_mtx_lock(&tb->common.fixlock); #endif - erts_refc_inc(&tb->common.fixref,1); + erts_refc_inc(&tb->common.ref,1); fix = tb->common.fixations; if (fix == NULL) { get_now(&(tb->common.megasec), @@ -3389,7 +3403,7 @@ static void unfix_table_locked(Process* p, DbTable* tb, for (pp = &tb->common.fixations; *pp != NULL; pp = &(*pp)->next) { if ((*pp)->pid == p->id) { DbFixation* fix = *pp; - erts_refc_dec(&tb->common.fixref,0); + erts_refc_dec(&tb->common.ref,0); --(fix->counter); ASSERT(fix->counter >= 0); if (fix->counter > 0) { @@ -3415,11 +3429,10 @@ static void unfix_table_locked(Process* p, DbTable* tb, unlocked: if (!IS_FIXED(tb) && IS_HASH_TABLE(tb->common.status) - && erts_smp_atomic_read(&tb->hash.fixdel) != (long)NULL) { + && erts_smp_atomic_read(&tb->hash.fixdel) != (erts_aint_t)NULL) { #ifdef ERTS_SMP if (*kind_p == LCK_READ && tb->common.is_thread_safe) { /* Must have write lock while purging pseudo-deleted (OTP-8166) */ - db_ref(tb, LCK_WRITE); /* LCK_WRITE need it, but not LCK_READ */ erts_smp_rwmtx_runlock(&tb->common.rwlock); erts_smp_rwmtx_rwlock(&tb->common.rwlock); *kind_p = LCK_WRITE; @@ -3438,6 +3451,8 @@ static void free_fixations_locked(DbTable *tb) fix = tb->common.fixations; while (fix != NULL) { + erts_aint_t diff = -((erts_aint_t) fix->counter); + erts_refc_add(&tb->common.ref,diff,0); next_fix = fix->next; db_meta_lock(meta_pid_to_fixed_tab, LCK_WRITE_REC); db_erase_bag_exact2(meta_pid_to_fixed_tab, @@ -3561,10 +3576,6 @@ static int free_table_cont(Process *p, mmtl = get_meta_main_tab_lock(tb->common.slot); #ifdef ERTS_SMP if (erts_smp_rwmtx_tryrwlock(mmtl) == EBUSY) { - /* - * We keep our increased refc over this op in order to - * prevent the table from disapearing. - */ erts_smp_rwmtx_rwunlock(&tb->common.rwlock); erts_smp_rwmtx_rwlock(mmtl); erts_smp_rwmtx_rwlock(&tb->common.rwlock); @@ -3579,7 +3590,7 @@ static int free_table_cont(Process *p, make_small(tb->common.slot)); db_meta_unlock(meta_pid_to_tab, LCK_WRITE_REC); } - db_unref(tb, LCK_NONE); + schedule_free_dbtable(tb); BUMP_REDS(p, 100); return 0; } diff --git a/erts/emulator/beam/erl_db.h b/erts/emulator/beam/erl_db.h index cb2da603f0..e0bdebcb01 100644 --- a/erts/emulator/beam/erl_db.h +++ b/erts/emulator/beam/erl_db.h @@ -83,7 +83,8 @@ Eterm erts_ets_colliding_names(Process*, Eterm name, Uint cnt); #define ERTS_DB_ALC_MEM_UPDATE_(TAB, FREE_SZ, ALLOC_SZ) \ do { \ - long sz__ = ((long) (ALLOC_SZ)) - ((long) (FREE_SZ)); \ + erts_aint_t sz__ = (((erts_aint_t) (ALLOC_SZ)) \ + - ((erts_aint_t) (FREE_SZ))); \ ASSERT((TAB)); \ erts_smp_atomic_add(&(TAB)->common.memory_size, sz__); \ } while (0) diff --git a/erts/emulator/beam/erl_db_hash.c b/erts/emulator/beam/erl_db_hash.c index 14ee63100a..1e50fee554 100644 --- a/erts/emulator/beam/erl_db_hash.c +++ b/erts/emulator/beam/erl_db_hash.c @@ -135,8 +135,8 @@ static ERTS_INLINE Uint hash_to_ix(DbTableHash* tb, HashValue hval) */ static ERTS_INLINE void add_fixed_deletion(DbTableHash* tb, int ix) { - long was_next; - long exp_next; + erts_aint_t was_next; + erts_aint_t exp_next; FixedDeletion* fixd = (FixedDeletion*) erts_db_alloc(ERTS_ALC_T_DB_FIX_DEL, (DbTable *) tb, sizeof(FixedDeletion)); @@ -146,7 +146,9 @@ static ERTS_INLINE void add_fixed_deletion(DbTableHash* tb, int ix) do { /* Lockless atomic insertion in linked list: */ exp_next = was_next; fixd->next = (FixedDeletion*) exp_next; - was_next = erts_smp_atomic_cmpxchg(&tb->fixdel, (long)fixd, exp_next); + was_next = erts_smp_atomic_cmpxchg(&tb->fixdel, + (erts_aint_t) fixd, + exp_next); }while (was_next != exp_next); } @@ -541,12 +543,12 @@ static void restore_fixdel(DbTableHash* tb, FixedDeletion* fixdel) { /*int tries = 0;*/ DEBUG_WAIT(); - if (erts_smp_atomic_cmpxchg(&tb->fixdel, (long)fixdel, - (long)NULL) != (long)NULL) { + if (erts_smp_atomic_cmpxchg(&tb->fixdel, (erts_aint_t)fixdel, + (erts_aint_t)NULL) != (erts_aint_t)NULL) { /* Oboy, must join lists */ FixedDeletion* last = fixdel; - long was_tail; - long exp_tail; + erts_aint_t was_tail; + erts_aint_t exp_tail; while (last->next != NULL) last = last->next; was_tail = erts_smp_atomic_read(&tb->fixdel); @@ -555,7 +557,7 @@ static void restore_fixdel(DbTableHash* tb, FixedDeletion* fixdel) last->next = (FixedDeletion*) exp_tail; /*++tries;*/ DEBUG_WAIT(); - was_tail = erts_smp_atomic_cmpxchg(&tb->fixdel, (long)fixdel, + was_tail = erts_smp_atomic_cmpxchg(&tb->fixdel, (erts_aint_t)fixdel, exp_tail); }while (was_tail != exp_tail); } @@ -573,7 +575,7 @@ void db_unfix_table_hash(DbTableHash *tb) || (erts_smp_lc_rwmtx_is_rlocked(&tb->common.rwlock) && !tb->common.is_thread_safe)); restart: - fixdel = (FixedDeletion*) erts_smp_atomic_xchg(&tb->fixdel, (long)NULL); + fixdel = (FixedDeletion*) erts_smp_atomic_xchg(&tb->fixdel, (erts_aint_t)NULL); while (fixdel != NULL) { FixedDeletion *fx = fixdel; int ix = fx->slot; @@ -642,8 +644,8 @@ int db_create_hash(Process *p, DbTable *tbl) erts_smp_atomic_init(&tb->szm, SEGSZ_MASK); erts_smp_atomic_init(&tb->nactive, SEGSZ); - erts_smp_atomic_init(&tb->fixdel, (long)NULL); - erts_smp_atomic_init(&tb->segtab, (long) alloc_ext_seg(tb,0,NULL)->segtab); + erts_smp_atomic_init(&tb->fixdel, (erts_aint_t)NULL); + erts_smp_atomic_init(&tb->segtab, (erts_aint_t) alloc_ext_seg(tb,0,NULL)->segtab); tb->nsegs = NSEG_1; tb->nslots = SEGSZ; @@ -1715,9 +1717,9 @@ static int db_select_delete_hash(Process *p, Eterm mpb; Eterm egot; #ifdef ERTS_SMP - int fixated_by_me = tb->common.is_thread_safe ? 0 : 1; /* ToDo: something nicer */ + erts_aint_t fixated_by_me = tb->common.is_thread_safe ? 0 : 1; /* ToDo: something nicer */ #else - int fixated_by_me = 0; + erts_aint_t fixated_by_me = 0; #endif erts_smp_rwmtx_t* lck; @@ -2124,11 +2126,11 @@ static int db_free_table_continue_hash(DbTable *tbl) sizeof(FixedDeletion)); ERTS_ETS_MISC_MEM_ADD(-sizeof(FixedDeletion)); if (++done >= 2*DELETE_RECORD_LIMIT) { - erts_smp_atomic_set(&tb->fixdel, (long)fixdel); + erts_smp_atomic_set(&tb->fixdel, (erts_aint_t)fixdel); return 0; /* Not done */ } } - erts_smp_atomic_set(&tb->fixdel, (long)NULL); + erts_smp_atomic_set(&tb->fixdel, (erts_aint_t)NULL); done /= 2; while(tb->nslots != 0) { @@ -2345,7 +2347,7 @@ static int alloc_seg(DbTableHash *tb) struct ext_segment* eseg; eseg = (struct ext_segment*) SEGTAB(tb)[seg_ix-1]; MY_ASSERT(eseg!=NULL && eseg->s.is_ext_segment); - erts_smp_atomic_set(&tb->segtab, (long) eseg->segtab); + erts_smp_atomic_set(&tb->segtab, (erts_aint_t) eseg->segtab); tb->nsegs = eseg->nsegs; } ASSERT(seg_ix < tb->nsegs); @@ -2417,7 +2419,7 @@ static int free_seg(DbTableHash *tb, int free_records) MY_ASSERT(newtop->s.is_ext_segment); if (newtop->prev_segtab != NULL) { /* Time to use a smaller segtab */ - erts_smp_atomic_set(&tb->segtab, (long)newtop->prev_segtab); + erts_smp_atomic_set(&tb->segtab, (erts_aint_t)newtop->prev_segtab); tb->nsegs = seg_ix; ASSERT(tb->nsegs == EXTSEG(SEGTAB(tb))->nsegs); } @@ -2434,7 +2436,7 @@ static int free_seg(DbTableHash *tb, int free_records) if (seg_ix > 0) { if (seg_ix < tb->nsegs) SEGTAB(tb)[seg_ix] = NULL; } else { - erts_smp_atomic_set(&tb->segtab, (long)NULL); + erts_smp_atomic_set(&tb->segtab, (erts_aint_t)NULL); } #endif tb->nslots -= SEGSZ; diff --git a/erts/emulator/beam/erl_db_util.c b/erts/emulator/beam/erl_db_util.c index e773361619..2852fb93fe 100644 --- a/erts/emulator/beam/erl_db_util.c +++ b/erts/emulator/beam/erl_db_util.c @@ -915,7 +915,7 @@ BIF_RETTYPE db_set_trace_control_word_1(Process *p, Eterm new) if (val != ((Uint32)val)) BIF_ERROR(p, BADARG); - old_tcw = (Uint32) erts_smp_atomic_xchg(&trace_control_word, (long) val); + old_tcw = (Uint32) erts_smp_atomic_xchg(&trace_control_word, (erts_aint_t) val); BIF_RET(erts_make_integer((Uint) old_tcw, p)); } diff --git a/erts/emulator/beam/erl_db_util.h b/erts/emulator/beam/erl_db_util.h index 10ba755e80..58ad39d772 100644 --- a/erts/emulator/beam/erl_db_util.h +++ b/erts/emulator/beam/erl_db_util.h @@ -206,8 +206,7 @@ typedef struct db_fixation { */ typedef struct db_table_common { - erts_refc_t ref; - erts_refc_t fixref; /* fixation counter */ + erts_refc_t ref; /* fixation counter and delete counter */ #ifdef ERTS_SMP erts_smp_rwmtx_t rwlock; /* rw lock on table */ erts_smp_mtx_t fixlock; /* Protects fixations,megasec,sec,microsec */ @@ -253,7 +252,7 @@ typedef struct db_table_common { (DB_BAG | DB_SET | DB_DUPLICATE_BAG))) #define IS_TREE_TABLE(Status) (!!((Status) & \ DB_ORDERED_SET)) -#define NFIXED(T) (erts_refc_read(&(T)->common.fixref,0)) +#define NFIXED(T) (erts_refc_read(&(T)->common.ref,0)) #define IS_FIXED(T) (NFIXED(T) != 0) Eterm erts_ets_copy_object(Eterm, Process*); diff --git a/erts/emulator/beam/erl_driver.h b/erts/emulator/beam/erl_driver.h index 9733c0e5b5..13a73e01bb 100644 --- a/erts/emulator/beam/erl_driver.h +++ b/erts/emulator/beam/erl_driver.h @@ -150,6 +150,27 @@ typedef struct { #define ERL_DRV_FLAG_SOFT_BUSY (1 << 1) /* + * Integer types + */ + +typedef unsigned long ErlDrvTermData; +typedef unsigned long ErlDrvUInt; +typedef signed long ErlDrvSInt; + +#if defined(__WIN32__) +typedef unsigned __int64 ErlDrvUInt64; +typedef __int64 ErlDrvSInt64; +#elif SIZEOF_LONG == 8 +typedef unsigned long ErlDrvUInt64; +typedef long ErlDrvSInt64; +#elif SIZEOF_LONG_LONG == 8 +typedef unsigned long long ErlDrvUInt64; +typedef long long ErlDrvSInt64; +#else +#error No 64-bit integer type +#endif + +/* * A binary as seen in a driver. Note that a binary should never be * altered by the driver when it has been sent to Erlang. */ @@ -179,26 +200,6 @@ struct erl_drv_event_data { #endif typedef struct erl_drv_event_data *ErlDrvEventData; /* Event data */ -/* - * Used in monitors... - */ -typedef unsigned long ErlDrvTermData; -typedef unsigned long ErlDrvUInt; -typedef signed long ErlDrvSInt; - -#if defined(__WIN32__) -typedef unsigned __int64 ErlDrvUInt64; -typedef __int64 ErlDrvSInt64; -#elif SIZEOF_LONG == 8 -typedef unsigned long ErlDrvUInt64; -typedef long ErlDrvSInt64; -#elif SIZEOF_LONG_LONG == 8 -typedef unsigned long long ErlDrvUInt64; -typedef long long ErlDrvSInt64; -#else -#error No 64-bit integer type -#endif - /* * A driver monitor */ @@ -394,9 +395,9 @@ EXTERN int driver_exit (ErlDrvPort port, int err); EXTERN ErlDrvPDL driver_pdl_create(ErlDrvPort); EXTERN void driver_pdl_lock(ErlDrvPDL); EXTERN void driver_pdl_unlock(ErlDrvPDL); -EXTERN long driver_pdl_get_refc(ErlDrvPDL); -EXTERN long driver_pdl_inc_refc(ErlDrvPDL); -EXTERN long driver_pdl_dec_refc(ErlDrvPDL); +EXTERN ErlDrvSInt driver_pdl_get_refc(ErlDrvPDL); +EXTERN ErlDrvSInt driver_pdl_inc_refc(ErlDrvPDL); +EXTERN ErlDrvSInt driver_pdl_dec_refc(ErlDrvPDL); /* * Process monitors @@ -432,9 +433,9 @@ EXTERN ErlDrvBinary* driver_realloc_binary(ErlDrvBinary *bin, int size); EXTERN void driver_free_binary(ErlDrvBinary *bin); /* Referenc count on driver binaries */ -EXTERN long driver_binary_get_refc(ErlDrvBinary *dbp); -EXTERN long driver_binary_inc_refc(ErlDrvBinary *dbp); -EXTERN long driver_binary_dec_refc(ErlDrvBinary *dbp); +EXTERN ErlDrvSInt driver_binary_get_refc(ErlDrvBinary *dbp); +EXTERN ErlDrvSInt driver_binary_inc_refc(ErlDrvBinary *dbp); +EXTERN ErlDrvSInt driver_binary_dec_refc(ErlDrvBinary *dbp); /* Allocation interface */ EXTERN void *driver_alloc(size_t size); diff --git a/erts/emulator/beam/erl_fun.c b/erts/emulator/beam/erl_fun.c index 84869f12d6..88947b5536 100644 --- a/erts/emulator/beam/erl_fun.c +++ b/erts/emulator/beam/erl_fun.c @@ -97,7 +97,7 @@ erts_put_fun_entry(Eterm mod, int uniq, int index) { ErlFunEntry template; ErlFunEntry* fe; - long refc; + erts_aint_t refc; ASSERT(is_atom(mod)); template.old_uniq = uniq; template.old_index = index; @@ -119,7 +119,7 @@ erts_put_fun_entry2(Eterm mod, int old_uniq, int old_index, { ErlFunEntry template; ErlFunEntry* fe; - long refc; + erts_aint_t refc; ASSERT(is_atom(mod)); template.old_uniq = old_uniq; @@ -157,7 +157,7 @@ erts_get_fun_entry(Eterm mod, int uniq, int index) erts_fun_read_lock(); ret = (ErlFunEntry *) hash_get(&erts_fun_table, (void*) &template); if (ret) { - long refc = erts_refc_inctest(&ret->refc, 1); + erts_aint_t refc = erts_refc_inctest(&ret->refc, 1); if (refc < 2) /* Pending delete */ erts_refc_inc(&ret->refc, 1); } @@ -257,7 +257,7 @@ erts_dump_fun_entries(int to, void *to_arg) #ifdef HIPE erts_print(to, to_arg, "Native_address: %p\n", fe->native_address); #endif - erts_print(to, to_arg, "Refc: %d\n", erts_refc_read(&fe->refc, 1)); + erts_print(to, to_arg, "Refc: %ld\n", erts_refc_read(&fe->refc, 1)); b = b->next; } } diff --git a/erts/emulator/beam/erl_gc.c b/erts/emulator/beam/erl_gc.c index 0f4d2a2ef9..2aa932e7d1 100644 --- a/erts/emulator/beam/erl_gc.c +++ b/erts/emulator/beam/erl_gc.c @@ -2471,7 +2471,7 @@ erts_check_off_heap2(Process *p, Eterm *htop) old = 0; for (u.hdr = MSO(p).first; u.hdr; u.hdr = u.hdr->next) { - long refc; + erts_aint_t refc; switch (thing_subtag(u.hdr->thing_word)) { case REFC_BINARY_SUBTAG: refc = erts_refc_read(&u.pb->val->refc, 1); diff --git a/erts/emulator/beam/erl_init.c b/erts/emulator/beam/erl_init.c index 464ee750f7..0a57eb6d88 100644 --- a/erts/emulator/beam/erl_init.c +++ b/erts/emulator/beam/erl_init.c @@ -100,7 +100,7 @@ int erts_backtrace_depth; /* How many functions to show in a backtrace int erts_async_max_threads; /* number of threads for async support */ int erts_async_thread_suggested_stack_size; -erts_smp_atomic_t erts_max_gen_gcs; +erts_smp_atomic32_t erts_max_gen_gcs; Eterm erts_error_logger_warnings; /* What to map warning logs to, am_error, am_info or am_warning, am_error is @@ -250,7 +250,7 @@ erl_init(int ncpu) erts_init_monitors(); erts_init_gc(); - init_time(); + erts_init_time(); erts_init_sys_common_misc(); erts_init_process(ncpu); erts_init_scheduling(use_multi_run_queue, @@ -289,7 +289,7 @@ erl_init(int ncpu) erts_delay_trap = erts_export_put(am_erlang, am_delay_trap, 2); erts_late_init_process(); #if HAVE_ERTS_MSEG - erts_mseg_late_init(); /* Must be after timer (init_time()) and thread + erts_mseg_late_init(); /* Must be after timer (erts_init_time()) and thread initializations */ #endif #ifdef HIPE @@ -323,7 +323,7 @@ init_shared_memory(int argc, char **argv) #endif global_gen_gcs = 0; - global_max_gen_gcs = erts_smp_atomic_read(&erts_max_gen_gcs); + global_max_gen_gcs = (Uint16) erts_smp_atomic32_read(&erts_max_gen_gcs); global_gc_flags = erts_default_process_flags; erts_global_offheap.mso = NULL; @@ -651,7 +651,7 @@ early_init(int *argc, char **argv) /* erts_writing_erl_crash_dump = 0; #endif - erts_smp_atomic_init(&erts_max_gen_gcs, (long)((Uint16) -1)); + erts_smp_atomic32_init(&erts_max_gen_gcs, (erts_aint32_t) ((Uint16) -1)); erts_pre_init_process(); #if defined(USE_THREADS) && !defined(ERTS_SMP) @@ -856,7 +856,7 @@ erl_start(int argc, char **argv) envbufsz = sizeof(envbuf); if (erts_sys_getenv("ERL_FULLSWEEP_AFTER", envbuf, &envbufsz) == 0) { Uint16 max_gen_gcs = atoi(envbuf); - erts_smp_atomic_set(&erts_max_gen_gcs, (long) max_gen_gcs); + erts_smp_atomic32_set(&erts_max_gen_gcs, (erts_aint32_t) max_gen_gcs); } envbufsz = sizeof(envbuf); diff --git a/erts/emulator/beam/erl_lock_check.c b/erts/emulator/beam/erl_lock_check.c index 04c7dbd2ec..0185baee6b 100644 --- a/erts/emulator/beam/erl_lock_check.c +++ b/erts/emulator/beam/erl_lock_check.c @@ -177,6 +177,8 @@ static erts_lc_lock_order_t erts_lock_order[] = { { "async_id", NULL }, { "pix_lock", "address" }, { "run_queues_lists", NULL }, + { "misc_aux_work_queue", "index" }, + { "misc_aux_work_pre_alloc_lock", "address" }, { "sched_stat", NULL }, { "run_queue_sleep_list", "address" }, #endif @@ -978,10 +980,10 @@ erts_lc_trylock_force_busy_flg(erts_lc_lock_t *lck, Uint16 op_flags) /* We only force busy if a lock order violation would occur and when on an even millisecond. */ { - erts_thr_timeval_t time; - erts_thr_time_now(&time); + SysTimeval tv; + sys_gettimeofday(&tv); - if ((time.tv_nsec / 1000000) & 1) + if ((tv.tv_usec / 1000) & 1) return 0; } #endif diff --git a/erts/emulator/beam/erl_lock_count.c b/erts/emulator/beam/erl_lock_count.c index 239773f366..a36c53560e 100644 --- a/erts/emulator/beam/erl_lock_count.c +++ b/erts/emulator/beam/erl_lock_count.c @@ -159,7 +159,7 @@ static char* lock_opt(Uint16 flag) { } static void print_lock_x(erts_lcnt_lock_t *lock, Uint16 flag, char *action, char *extra) { - long int colls, tries, w_state, r_state; + erts_aint_t colls, tries, w_state, r_state; erts_lcnt_lock_stats_t *stats = NULL; char *type; @@ -385,7 +385,7 @@ void erts_lcnt_destroy_lock(erts_lcnt_lock_t *lock) { /* lock */ void erts_lcnt_lock_opt(erts_lcnt_lock_t *lock, Uint16 option) { - long r_state = 0, w_state = 0; + erts_aint_t r_state = 0, w_state = 0; erts_lcnt_thread_data_t *eltd; if (erts_lcnt_rt_options & ERTS_LCNT_OPT_SUSPEND) return; @@ -418,7 +418,7 @@ void erts_lcnt_lock_opt(erts_lcnt_lock_t *lock, Uint16 option) { } void erts_lcnt_lock(erts_lcnt_lock_t *lock) { - long w_state; + erts_aint_t w_state; erts_lcnt_thread_data_t *eltd; if (erts_lcnt_rt_options & ERTS_LCNT_OPT_SUSPEND) return; @@ -471,7 +471,7 @@ void erts_lcnt_lock_post_x(erts_lcnt_lock_t *lock, char *file, unsigned int line erts_lcnt_time_t time_wait; erts_lcnt_lock_stats_t *stats; #ifdef DEBUG - long flowstate; + erts_aint_t flowstate; #endif if (erts_lcnt_rt_options & ERTS_LCNT_OPT_SUSPEND) return; @@ -516,8 +516,8 @@ void erts_lcnt_unlock_opt(erts_lcnt_lock_t *lock, Uint16 option) { void erts_lcnt_unlock(erts_lcnt_lock_t *lock) { #ifdef DEBUG - long w_state; - long flowstate; + erts_aint_t w_state; + erts_aint_t flowstate; #endif if (erts_lcnt_rt_options & ERTS_LCNT_OPT_SUSPEND) return; #ifdef DEBUG @@ -552,7 +552,7 @@ void erts_lcnt_trylock_opt(erts_lcnt_lock_t *lock, int res, Uint16 option) { void erts_lcnt_trylock(erts_lcnt_lock_t *lock, int res) { /* Determine lock_state via res instead of state */ #ifdef DEBUG - long flowstate; + erts_aint_t flowstate; #endif if (erts_lcnt_rt_options & ERTS_LCNT_OPT_SUSPEND) return; if (res != EBUSY) { diff --git a/erts/emulator/beam/erl_node_tables.c b/erts/emulator/beam/erl_node_tables.c index 8cdda395df..6daa127d23 100644 --- a/erts/emulator/beam/erl_node_tables.c +++ b/erts/emulator/beam/erl_node_tables.c @@ -235,7 +235,7 @@ erts_sysname_to_connected_dist_entry(Eterm sysname) erts_smp_rwmtx_rlock(&erts_dist_table_rwmtx); res_dep = (DistEntry *) hash_get(&erts_dist_table, (void *) &de); if (res_dep) { - long refc = erts_refc_inctest(&res_dep->refc, 1); + erts_aint_t refc = erts_refc_inctest(&res_dep->refc, 1); if (refc < 2) /* Pending delete */ erts_refc_inc(&res_dep->refc, 1); } @@ -257,7 +257,7 @@ DistEntry *erts_find_or_insert_dist_entry(Eterm sysname) { DistEntry *res; DistEntry de; - long refc; + erts_aint_t refc; res = erts_find_dist_entry(sysname); if (res) return res; @@ -279,7 +279,7 @@ DistEntry *erts_find_dist_entry(Eterm sysname) erts_smp_rwmtx_rlock(&erts_dist_table_rwmtx); res = hash_get(&erts_dist_table, (void *) &de); if (res) { - long refc = erts_refc_inctest(&res->refc, 1); + erts_aint_t refc = erts_refc_inctest(&res->refc, 1); if (refc < 2) /* Pending delete */ erts_refc_inc(&res->refc, 1); } @@ -586,7 +586,7 @@ ErlNode *erts_find_or_insert_node(Eterm sysname, Uint creation) erts_smp_rwmtx_rlock(&erts_node_table_rwmtx); res = hash_get(&erts_node_table, (void *) &ne); if (res && res != erts_this_node) { - long refc = erts_refc_inctest(&res->refc, 0); + erts_aint_t refc = erts_refc_inctest(&res->refc, 0); if (refc < 2) /* New or pending delete */ erts_refc_inc(&res->refc, 1); } @@ -598,7 +598,7 @@ ErlNode *erts_find_or_insert_node(Eterm sysname, Uint creation) res = hash_put(&erts_node_table, (void *) &ne); ASSERT(res); if (res != erts_this_node) { - long refc = erts_refc_inctest(&res->refc, 0); + erts_aint_t refc = erts_refc_inctest(&res->refc, 0); if (refc < 2) /* New or pending delete */ erts_refc_inc(&res->refc, 1); } diff --git a/erts/emulator/beam/erl_port_task.c b/erts/emulator/beam/erl_port_task.c index c10724b951..1b07024ca1 100644 --- a/erts/emulator/beam/erl_port_task.c +++ b/erts/emulator/beam/erl_port_task.c @@ -129,7 +129,7 @@ reset_handle(ErtsPortTask *ptp) { if (ptp->handle) { ASSERT(ptp == handle2task(ptp->handle)); - erts_smp_atomic_set(ptp->handle, (long) NULL); + erts_smp_atomic_set(ptp->handle, (erts_aint_t) NULL); } } @@ -138,7 +138,7 @@ set_handle(ErtsPortTask *ptp, ErtsPortTaskHandle *pthp) { ptp->handle = pthp; if (pthp) { - erts_smp_atomic_set(pthp, (long) ptp); + erts_smp_atomic_set(pthp, (erts_aint_t) ptp); ASSERT(ptp == handle2task(ptp->handle)); } } @@ -568,7 +568,7 @@ erts_port_task_schedule(Eterm id, ErtsRunQueue *xrunq = erts_check_emigration_need(runq, ERTS_PORT_PRIO_LEVEL); if (xrunq) { /* Port emigrated ... */ - erts_smp_atomic_set(&pp->run_queue, (long) xrunq); + erts_smp_atomic_set(&pp->run_queue, (erts_aint_t) xrunq); erts_smp_runq_unlock(runq); runq = xrunq; } @@ -727,7 +727,8 @@ resume_after_block(void *vd) ErtsPortTaskExeBlockData *d = (ErtsPortTaskExeBlockData *) vd; erts_smp_runq_lock(d->runq); if (d->resp) - *d->resp = erts_smp_atomic_read(&erts_port_task_outstanding_io_tasks) != (long) 0; + *d->resp = (erts_smp_atomic_read(&erts_port_task_outstanding_io_tasks) + != (erts_aint_t) 0); } /* @@ -748,7 +749,7 @@ erts_port_task_execute(ErtsRunQueue *runq, Port **curr_port_pp) ErtsPortTask *ptp; int res = 0; int reds = ERTS_PORT_REDS_EXECUTE; - long io_tasks_executed = 0; + erts_aint_t io_tasks_executed = 0; int fpe_was_unmasked; ErtsPortTaskExeBlockData blk_data = {runq, NULL}; @@ -942,7 +943,7 @@ erts_port_task_execute(ErtsRunQueue *runq, Port **curr_port_pp) } else { /* Port emigrated ... */ - erts_smp_atomic_set(&pp->run_queue, (long) xrunq); + erts_smp_atomic_set(&pp->run_queue, (erts_aint_t) xrunq); enqueue_port(xrunq, pp); ASSERT(pp->sched.exe_taskq); pp->sched.exe_taskq = NULL; @@ -953,7 +954,8 @@ erts_port_task_execute(ErtsRunQueue *runq, Port **curr_port_pp) port_was_enqueued = 1; } - res = erts_smp_atomic_read(&erts_port_task_outstanding_io_tasks) != (long) 0; + res = (erts_smp_atomic_read(&erts_port_task_outstanding_io_tasks) + != (erts_aint_t) 0); ERTS_PT_CHK_PRES_PORTQ(runq, pp); @@ -971,7 +973,7 @@ erts_port_task_execute(ErtsRunQueue *runq, Port **curr_port_pp) erts_port_release(pp); #else { - long refc; + erts_aint_t refc; erts_smp_mtx_unlock(pp->lock); refc = erts_smp_atomic_dectest(&pp->refc); ASSERT(refc >= 0); @@ -979,7 +981,8 @@ erts_port_task_execute(ErtsRunQueue *runq, Port **curr_port_pp) erts_smp_runq_unlock(runq); erts_port_cleanup(pp); /* Might aquire runq lock */ erts_smp_runq_lock(runq); - res = erts_smp_atomic_read(&erts_port_task_outstanding_io_tasks) != (long) 0; + res = (erts_smp_atomic_read(&erts_port_task_outstanding_io_tasks) + != (erts_aint_t) 0); } } #endif @@ -1112,7 +1115,7 @@ erts_port_migrate(Port *prt, int *prt_locked, if (!ERTS_PORT_IS_IN_RUNQ(from_rq, prt)) return ERTS_MIGRATE_FAILED_NOT_IN_RUNQ; dequeue_port(from_rq, prt); - erts_smp_atomic_set(&prt->run_queue, (long) to_rq); + erts_smp_atomic_set(&prt->run_queue, (erts_aint_t) to_rq); enqueue_port(to_rq, prt); return ERTS_MIGRATE_SUCCESS; } @@ -1125,7 +1128,7 @@ erts_port_migrate(Port *prt, int *prt_locked, void erts_port_task_init(void) { - erts_smp_atomic_init(&erts_port_task_outstanding_io_tasks, (long) 0); + erts_smp_atomic_init(&erts_port_task_outstanding_io_tasks, (erts_aint_t) 0); init_port_task_alloc(); init_port_taskq_alloc(); } diff --git a/erts/emulator/beam/erl_port_task.h b/erts/emulator/beam/erl_port_task.h index f12d02da0c..714b4ea7dd 100644 --- a/erts/emulator/beam/erl_port_task.h +++ b/erts/emulator/beam/erl_port_task.h @@ -1,7 +1,7 @@ /* * %CopyrightBegin% * - * Copyright Ericsson AB 2006-2009. All Rights Reserved. + * Copyright Ericsson AB 2006-2010. 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 @@ -79,7 +79,7 @@ ERTS_GLB_INLINE int erts_port_task_have_outstanding_io_tasks(void); ERTS_GLB_INLINE void erts_port_task_handle_init(ErtsPortTaskHandle *pthp) { - erts_smp_atomic_init(pthp, (long) NULL); + erts_smp_atomic_init(pthp, (erts_aint_t) NULL); } ERTS_GLB_INLINE int diff --git a/erts/emulator/beam/erl_process.c b/erts/emulator/beam/erl_process.c index fc950af8ce..ddfc27a93f 100644 --- a/erts/emulator/beam/erl_process.c +++ b/erts/emulator/beam/erl_process.c @@ -127,21 +127,22 @@ ErtsLcPSDLocks erts_psd_required_locks[ERTS_PSD_SIZE]; int erts_disable_proc_not_running_opt; -#define ERTS_SCHDLR_SSPND_CHNG_WAITER (((long) 1) << 0) -#define ERTS_SCHDLR_SSPND_CHNG_MSB (((long) 1) << 1) -#define ERTS_SCHDLR_SSPND_CHNG_ONLN (((long) 1) << 2) +#define ERTS_SCHDLR_SSPND_CHNG_WAITER (((erts_aint32_t) 1) << 0) +#define ERTS_SCHDLR_SSPND_CHNG_MSB (((erts_aint32_t) 1) << 1) +#define ERTS_SCHDLR_SSPND_CHNG_ONLN (((erts_aint32_t) 1) << 2) #ifndef DEBUG #define ERTS_SCHDLR_SSPND_CHNG_SET(VAL, OLD_VAL) \ - erts_smp_atomic_set(&schdlr_sspnd.changing, (VAL)) + erts_smp_atomic32_set(&schdlr_sspnd.changing, (VAL)) #else #define ERTS_SCHDLR_SSPND_CHNG_SET(VAL, OLD_VAL) \ do { \ - long old_val__ = erts_smp_atomic_xchg(&schdlr_sspnd.changing, \ - (VAL)); \ + erts_aint32_t old_val__; \ + old_val__ = erts_smp_atomic32_xchg(&schdlr_sspnd.changing, \ + (VAL)); \ ASSERT(old_val__ == (OLD_VAL)); \ } while (0) @@ -154,10 +155,10 @@ static struct { int online; int curr_online; int wait_curr_online; - erts_smp_atomic_t changing; - erts_smp_atomic_t active; + erts_smp_atomic32_t changing; + erts_smp_atomic32_t active; struct { - erts_smp_atomic_t ongoing; + erts_smp_atomic32_t ongoing; long wait_active; ErtsProcList *procs; } msb; /* Multi Scheduling Block */ @@ -165,11 +166,11 @@ static struct { static struct { erts_smp_mtx_t update_mtx; - erts_smp_atomic_t active_runqs; + erts_smp_atomic32_t active_runqs; int last_active_runqs; - erts_smp_atomic_t used_runqs; + erts_smp_atomic32_t used_runqs; int forced_check_balance; - erts_smp_atomic_t checking_balance; + erts_smp_atomic32_t checking_balance; int halftime; int full_reds_history_index; struct { @@ -199,11 +200,11 @@ static erts_tsd_key_t sched_data_key; static erts_smp_mtx_t proc_tab_mtx; -static erts_smp_atomic_t function_calls; +static erts_smp_atomic32_t function_calls; #ifdef ERTS_SMP -static erts_smp_atomic_t doing_sys_schedule; -static erts_smp_atomic_t no_empty_run_queues; +static erts_smp_atomic32_t doing_sys_schedule; +static erts_smp_atomic32_t no_empty_run_queues; #else /* !ERTS_SMP */ ErtsSchedulerData *erts_scheduler_data; #endif @@ -247,7 +248,10 @@ Uint erts_num_active_procs; Process** erts_active_procs; #endif -static erts_smp_atomic_t process_count; +#if ERTS_MAX_PROCESSES > 0x7fffffff +#error "Need to store process_count in another type" +#endif +static erts_smp_atomic32_t process_count; typedef struct ErtsTermProcElement_ ErtsTermProcElement; struct ErtsTermProcElement_ { @@ -407,7 +411,7 @@ erts_init_process(int ncpu) init_proclist_alloc(); - erts_smp_atomic_init(&process_count, 0); + erts_smp_atomic32_init(&process_count, 0); if (erts_use_r9_pids_ports) { proc_bits = ERTS_R9_PROC_BITS; @@ -568,7 +572,7 @@ erts_psd_set_init(Process *p, ErtsProcLocks plocks, int ix, void *data) #ifdef ERTS_SMP void -erts_sched_finish_poke(ErtsSchedulerSleepInfo *ssi, long flags) +erts_sched_finish_poke(ErtsSchedulerSleepInfo *ssi, erts_aint32_t flags) { switch (flags & ERTS_SSI_FLGS_SLEEP_TYPE) { case ERTS_SSI_FLG_POLL_SLEEPING: @@ -586,6 +590,122 @@ erts_sched_finish_poke(ErtsSchedulerSleepInfo *ssi, long flags) } } +typedef struct erts_misc_aux_work_t_ erts_misc_aux_work_t; +struct erts_misc_aux_work_t_ { + erts_misc_aux_work_t *next; + void (*func)(void *); + void *arg; +}; + +typedef struct { + erts_smp_mtx_t mtx; + erts_misc_aux_work_t *first; + erts_misc_aux_work_t *last; +} erts_misc_aux_work_q_t; + +typedef union { + erts_misc_aux_work_q_t data; + char align[ERTS_ALC_CACHE_LINE_ALIGN_SIZE(sizeof(erts_misc_aux_work_q_t))]; +} erts_algnd_misc_aux_work_q_t; + +static erts_algnd_misc_aux_work_q_t *misc_aux_work_queues; + +ERTS_SCHED_PREF_QUICK_ALLOC_IMPL(misc_aux_work, + erts_misc_aux_work_t, + 200, + ERTS_ALC_T_MISC_AUX_WORK) + +static void +init_misc_aux_work(void) +{ + int ix; + + init_misc_aux_work_alloc(); + + misc_aux_work_queues = erts_alloc(ERTS_ALC_T_MISC_AUX_WORK_Q, + (sizeof(erts_algnd_misc_aux_work_q_t) + *(erts_no_schedulers+1))); + if ((((UWord) misc_aux_work_queues) & ERTS_CACHE_LINE_MASK) != 0) + misc_aux_work_queues = ((erts_algnd_misc_aux_work_q_t *) + ((((UWord) misc_aux_work_queues) + & ~ERTS_CACHE_LINE_MASK) + + ERTS_CACHE_LINE_SIZE)); + + for (ix = 0; ix < erts_no_schedulers; ix++) { + erts_smp_mtx_init_x(&misc_aux_work_queues[ix].data.mtx, + "misc_aux_work_queue", + make_small(ix + 1)); + misc_aux_work_queues[ix].data.first = NULL; + misc_aux_work_queues[ix].data.last = NULL; + } +} + +static void +handle_misc_aux_work(ErtsSchedulerData *esdp) +{ + int ix = (int) esdp->no - 1; + erts_misc_aux_work_t *mawp; + + erts_smp_mtx_lock(&misc_aux_work_queues[ix].data.mtx); + mawp = misc_aux_work_queues[ix].data.first; + misc_aux_work_queues[ix].data.first = NULL; + misc_aux_work_queues[ix].data.last = NULL; + erts_smp_mtx_unlock(&misc_aux_work_queues[ix].data.mtx); + + while (mawp) { + erts_misc_aux_work_t *free_mawp; + mawp->func(mawp->arg); + free_mawp = mawp; + mawp = mawp->next; + misc_aux_work_free(free_mawp); + } +} + +void +erts_smp_schedule_misc_aux_work(int ignore_self, + int max_sched, + void (*func)(void *), + void *arg) +{ + int ix, ignore_ix = -1; + + if (ignore_self) { + ErtsSchedulerData *esdp = erts_get_scheduler_data(); + if (esdp) + ignore_ix = (int) esdp->no - 1; + } + + ASSERT(0 <= max_sched && max_sched <= erts_no_schedulers); + + for (ix = 0; ix < max_sched; ix++) { + erts_aint32_t aux_work; + erts_misc_aux_work_t *mawp; + ErtsSchedulerSleepInfo *ssi; + if (ix == ignore_ix) + continue; + + mawp = misc_aux_work_alloc(); + + mawp->func = func; + mawp->arg = arg; + mawp->next = NULL; + + erts_smp_mtx_lock(&misc_aux_work_queues[ix].data.mtx); + if (!misc_aux_work_queues[ix].data.last) + misc_aux_work_queues[ix].data.first = mawp; + else + misc_aux_work_queues[ix].data.last->next = mawp; + misc_aux_work_queues[ix].data.last = mawp; + erts_smp_mtx_unlock(&misc_aux_work_queues[ix].data.mtx); + + ssi = ERTS_SCHED_SLEEP_INFO_IX(ix); + aux_work = erts_smp_atomic32_bor(&ssi->aux_work, + ERTS_SSI_AUX_WORK_MISC); + if ((aux_work & ERTS_SSI_AUX_WORK_MISC) == 0) + erts_sched_poke(ssi); + } +} + #ifdef ERTS_SMP_SCHEDULERS_NEED_TO_CHECK_CHILDREN void erts_smp_notify_check_children_needed(void) @@ -593,11 +713,11 @@ erts_smp_notify_check_children_needed(void) int i; for (i = 0; i < erts_no_schedulers; i++) { - long aux_work; + erts_aint32_t aux_work; ErtsSchedulerSleepInfo *ssi; ssi = ERTS_SCHED_SLEEP_INFO_IX(i); - aux_work = erts_smp_atomic_bor(&ssi->aux_work, - ERTS_SSI_AUX_WORK_CHECK_CHILDREN); + aux_work = erts_smp_atomic32_bor(&ssi->aux_work, + ERTS_SSI_AUX_WORK_CHECK_CHILDREN); if (!(aux_work & ERTS_SSI_AUX_WORK_CHECK_CHILDREN)) erts_sched_poke(ssi); } @@ -605,16 +725,22 @@ erts_smp_notify_check_children_needed(void) #endif #ifdef ERTS_SCHED_NEED_BLOCKABLE_AUX_WORK -static ERTS_INLINE long +static ERTS_INLINE erts_aint32_t blockable_aux_work(ErtsSchedulerData *esdp, ErtsSchedulerSleepInfo *ssi, - long aux_work) + erts_aint32_t aux_work) { if (aux_work & ERTS_SSI_BLOCKABLE_AUX_WORK_MASK) { + if (aux_work & ERTS_SSI_AUX_WORK_MISC) { + aux_work = erts_smp_atomic32_band(&ssi->aux_work, + ~ERTS_SSI_AUX_WORK_MISC); + aux_work &= ~ERTS_SSI_AUX_WORK_MISC; + handle_misc_aux_work(esdp); + } #ifdef ERTS_SMP_SCHEDULERS_NEED_TO_CHECK_CHILDREN if (aux_work & ERTS_SSI_AUX_WORK_CHECK_CHILDREN) { - aux_work = erts_smp_atomic_band(&ssi->aux_work, - ~ERTS_SSI_AUX_WORK_CHECK_CHILDREN); + aux_work = erts_smp_atomic32_band(&ssi->aux_work, + ~ERTS_SSI_AUX_WORK_CHECK_CHILDREN); aux_work &= ~ERTS_SSI_AUX_WORK_CHECK_CHILDREN; erts_check_children(); } @@ -626,10 +752,10 @@ blockable_aux_work(ErtsSchedulerData *esdp, #endif #ifdef ERTS_SCHED_NEED_NONBLOCKABLE_AUX_WORK -static ERTS_INLINE long +static ERTS_INLINE erts_aint32_t nonblockable_aux_work(ErtsSchedulerData *esdp, ErtsSchedulerSleepInfo *ssi, - long aux_work) + erts_aint32_t aux_work) { if (aux_work & ERTS_SSI_NONBLOCKABLE_AUX_WORK_MASK) { @@ -694,10 +820,10 @@ prepare_for_sys_schedule(void) { #ifdef ERTS_SMP while (!erts_port_task_have_outstanding_io_tasks() - && !erts_smp_atomic_xchg(&doing_sys_schedule, 1)) { + && !erts_smp_atomic32_xchg(&doing_sys_schedule, 1)) { if (!erts_port_task_have_outstanding_io_tasks()) return 1; - erts_smp_atomic_set(&doing_sys_schedule, 0); + erts_smp_atomic32_set(&doing_sys_schedule, 0); } return 0; #else @@ -745,53 +871,55 @@ sched_active(Uint no, ErtsRunQueue *rq) static int ERTS_INLINE ongoing_multi_scheduling_block(void) { - return erts_smp_atomic_read(&schdlr_sspnd.msb.ongoing) != 0; + return erts_smp_atomic32_read(&schdlr_sspnd.msb.ongoing) != 0; } static ERTS_INLINE void empty_runq(ErtsRunQueue *rq) { - long oifls = erts_smp_atomic_band(&rq->info_flags, ~ERTS_RUNQ_IFLG_NONEMPTY); + erts_aint32_t oifls = erts_smp_atomic32_band(&rq->info_flags, + ~ERTS_RUNQ_IFLG_NONEMPTY); if (oifls & ERTS_RUNQ_IFLG_NONEMPTY) { #ifdef DEBUG - long empty = erts_smp_atomic_read(&no_empty_run_queues); + erts_aint32_t empty = erts_smp_atomic32_read(&no_empty_run_queues); /* * For a short period of time no_empty_run_queues may have * been increased twice for a specific run queue. */ ASSERT(0 <= empty && empty < 2*erts_no_run_queues); #endif - erts_smp_atomic_inc(&no_empty_run_queues); + erts_smp_atomic32_inc(&no_empty_run_queues); } } static ERTS_INLINE void non_empty_runq(ErtsRunQueue *rq) { - long oifls = erts_smp_atomic_bor(&rq->info_flags, ERTS_RUNQ_IFLG_NONEMPTY); + erts_aint32_t oifls = erts_smp_atomic32_bor(&rq->info_flags, + ERTS_RUNQ_IFLG_NONEMPTY); if (!(oifls & ERTS_RUNQ_IFLG_NONEMPTY)) { #ifdef DEBUG - long empty = erts_smp_atomic_read(&no_empty_run_queues); + erts_aint32_t empty = erts_smp_atomic32_read(&no_empty_run_queues); /* * For a short period of time no_empty_run_queues may have * been increased twice for a specific run queue. */ ASSERT(0 < empty && empty <= 2*erts_no_run_queues); #endif - erts_smp_atomic_dec(&no_empty_run_queues); + erts_smp_atomic32_dec(&no_empty_run_queues); } } -static long +static erts_aint32_t sched_prep_spin_wait(ErtsSchedulerSleepInfo *ssi) { - long oflgs; - long nflgs = (ERTS_SSI_FLG_SLEEPING - | ERTS_SSI_FLG_WAITING); - long xflgs = 0; + erts_aint32_t oflgs; + erts_aint32_t nflgs = (ERTS_SSI_FLG_SLEEPING + | ERTS_SSI_FLG_WAITING); + erts_aint32_t xflgs = 0; do { - oflgs = erts_smp_atomic_cmpxchg(&ssi->flags, nflgs, xflgs); + oflgs = erts_smp_atomic32_cmpxchg(&ssi->flags, nflgs, xflgs); if (oflgs == xflgs) return nflgs; xflgs = oflgs; @@ -799,16 +927,16 @@ sched_prep_spin_wait(ErtsSchedulerSleepInfo *ssi) return oflgs; } -static long +static erts_aint32_t sched_prep_cont_spin_wait(ErtsSchedulerSleepInfo *ssi) { - long oflgs; - long nflgs = (ERTS_SSI_FLG_SLEEPING - | ERTS_SSI_FLG_WAITING); - long xflgs = ERTS_SSI_FLG_WAITING; + erts_aint32_t oflgs; + erts_aint32_t nflgs = (ERTS_SSI_FLG_SLEEPING + | ERTS_SSI_FLG_WAITING); + erts_aint32_t xflgs = ERTS_SSI_FLG_WAITING; do { - oflgs = erts_smp_atomic_cmpxchg(&ssi->flags, nflgs, xflgs); + oflgs = erts_smp_atomic32_cmpxchg(&ssi->flags, nflgs, xflgs); if (oflgs == xflgs) return nflgs; xflgs = oflgs; @@ -817,15 +945,15 @@ sched_prep_cont_spin_wait(ErtsSchedulerSleepInfo *ssi) return oflgs; } -static long +static erts_aint32_t sched_spin_wait(ErtsSchedulerSleepInfo *ssi, int spincount) { - long until_yield = ERTS_SCHED_SPIN_UNTIL_YIELD; + int until_yield = ERTS_SCHED_SPIN_UNTIL_YIELD; int sc = spincount; - long flgs; + erts_aint32_t flgs; do { - flgs = erts_smp_atomic_read(&ssi->flags); + flgs = erts_smp_atomic32_read(&ssi->flags); if ((flgs & (ERTS_SSI_FLG_SLEEPING|ERTS_SSI_FLG_WAITING)) != (ERTS_SSI_FLG_SLEEPING|ERTS_SSI_FLG_WAITING)) { break; @@ -839,18 +967,18 @@ sched_spin_wait(ErtsSchedulerSleepInfo *ssi, int spincount) return flgs; } -static long -sched_set_sleeptype(ErtsSchedulerSleepInfo *ssi, long sleep_type) +static erts_aint32_t +sched_set_sleeptype(ErtsSchedulerSleepInfo *ssi, erts_aint32_t sleep_type) { - long oflgs; - long nflgs = ERTS_SSI_FLG_SLEEPING|ERTS_SSI_FLG_WAITING|sleep_type; - long xflgs = ERTS_SSI_FLG_SLEEPING|ERTS_SSI_FLG_WAITING; + erts_aint32_t oflgs; + erts_aint32_t nflgs = ERTS_SSI_FLG_SLEEPING|ERTS_SSI_FLG_WAITING|sleep_type; + erts_aint32_t xflgs = ERTS_SSI_FLG_SLEEPING|ERTS_SSI_FLG_WAITING; if (sleep_type == ERTS_SSI_FLG_TSE_SLEEPING) erts_tse_reset(ssi->event); while (1) { - oflgs = erts_smp_atomic_cmpxchg(&ssi->flags, nflgs, xflgs); + oflgs = erts_smp_atomic32_cmpxchg(&ssi->flags, nflgs, xflgs); if (oflgs == xflgs) return nflgs; if ((oflgs & (ERTS_SSI_FLG_SLEEPING|ERTS_SSI_FLG_WAITING)) @@ -867,14 +995,14 @@ sched_set_sleeptype(ErtsSchedulerSleepInfo *ssi, long sleep_type) != ERTS_SSI_FLG_WAITING) static void -scheduler_wait(long *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq) +scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq) { ErtsSchedulerSleepInfo *ssi = esdp->ssi; int spincount; - long flgs; + erts_aint32_t flgs; #if defined(ERTS_SCHED_NEED_NONBLOCKABLE_AUX_WORK) \ || defined(ERTS_SCHED_NEED_BLOCKABLE_AUX_WORK) - long aux_work; + erts_aint32_t aux_work; #endif ERTS_SMP_LC_ASSERT(erts_smp_lc_runq_is_locked(rq)); @@ -910,7 +1038,7 @@ scheduler_wait(long *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq) tse_wait: #ifdef ERTS_SCHED_NEED_BLOCKABLE_AUX_WORK - aux_work = erts_smp_atomic_read(&ssi->aux_work); + aux_work = erts_smp_atomic32_read(&ssi->aux_work); tse_blockable_aux_work: aux_work = blockable_aux_work(esdp, ssi, aux_work); #endif @@ -920,7 +1048,7 @@ scheduler_wait(long *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq) #ifdef ERTS_SCHED_NEED_NONBLOCKABLE_AUX_WORK #ifndef ERTS_SCHED_NEED_BLOCKABLE_AUX_WORK - aux_work = erts_smp_atomic_read(&ssi->aux_work); + aux_work = erts_smp_atomic32_read(&ssi->aux_work); #endif nonblockable_aux_work(esdp, ssi, aux_work); #endif @@ -953,7 +1081,7 @@ scheduler_wait(long *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq) } #ifdef ERTS_SCHED_NEED_BLOCKABLE_AUX_WORK - aux_work = erts_smp_atomic_read(&ssi->aux_work); + aux_work = erts_smp_atomic32_read(&ssi->aux_work); if (aux_work & ERTS_SSI_BLOCKABLE_AUX_WORK_MASK) { erts_smp_activity_end(ERTS_ACTIVITY_WAIT, NULL, NULL, NULL); goto tse_blockable_aux_work; @@ -965,16 +1093,16 @@ scheduler_wait(long *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq) erts_smp_activity_end(ERTS_ACTIVITY_WAIT, NULL, NULL, NULL); if (flgs & ~ERTS_SSI_FLG_SUSPENDED) - erts_smp_atomic_band(&ssi->flags, ERTS_SSI_FLG_SUSPENDED); + erts_smp_atomic32_band(&ssi->flags, ERTS_SSI_FLG_SUSPENDED); erts_smp_runq_lock(rq); sched_active(esdp->no, rq); } else { - long dt; + erts_aint_t dt; - erts_smp_atomic_set(&function_calls, 0); + erts_smp_atomic32_set(&function_calls, 0); *fcalls = 0; sched_waiting_sys(esdp->no, rq); @@ -991,23 +1119,23 @@ scheduler_wait(long *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq) erl_sys_schedule(1); /* Might give us something to do */ - dt = do_time_read_and_reset(); - if (dt) bump_timer(dt); + dt = erts_do_time_read_and_reset(); + if (dt) erts_bump_timer(dt); sys_aux_work: #ifdef ERTS_SCHED_NEED_BLOCKABLE_AUX_WORK - aux_work = erts_smp_atomic_read(&ssi->aux_work); + aux_work = erts_smp_atomic32_read(&ssi->aux_work); aux_work = blockable_aux_work(esdp, ssi, aux_work); #endif #ifdef ERTS_SCHED_NEED_NONBLOCKABLE_AUX_WORK #ifndef ERTS_SCHED_NEED_BLOCKABLE_AUX_WORK - aux_work = erts_smp_atomic_read(&ssi->aux_work); + aux_work = erts_smp_atomic32_read(&ssi->aux_work); #endif nonblockable_aux_work(esdp, ssi, aux_work); #endif - flgs = erts_smp_atomic_read(&ssi->flags); + flgs = erts_smp_atomic32_read(&ssi->flags); if (!(flgs & ERTS_SSI_FLG_WAITING)) { ASSERT(!(flgs & ERTS_SSI_FLG_SLEEPING)); goto sys_woken; @@ -1025,7 +1153,7 @@ scheduler_wait(long *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq) * call erl_sys_schedule() until it is handled. */ if (erts_port_task_have_outstanding_io_tasks()) { - erts_smp_atomic_set(&doing_sys_schedule, 0); + erts_smp_atomic32_set(&doing_sys_schedule, 0); /* * Got to check that we still got I/O tasks; otherwise * we have to continue checking for I/O... @@ -1044,7 +1172,7 @@ scheduler_wait(long *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq) * sleep in erl_sys_schedule(). */ if (erts_port_task_have_outstanding_io_tasks()) { - erts_smp_atomic_set(&doing_sys_schedule, 0); + erts_smp_atomic32_set(&doing_sys_schedule, 0); /* * Got to check that we still got I/O tasks; otherwise @@ -1088,8 +1216,8 @@ scheduler_wait(long *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq) erl_sys_schedule(0); - dt = do_time_read_and_reset(); - if (dt) bump_timer(dt); + dt = erts_do_time_read_and_reset(); + if (dt) erts_bump_timer(dt); flgs = sched_prep_cont_spin_wait(ssi); if (flgs & ERTS_SSI_FLG_WAITING) @@ -1098,9 +1226,9 @@ scheduler_wait(long *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq) sys_woken: erts_smp_runq_lock(rq); sys_locked_woken: - erts_smp_atomic_set(&doing_sys_schedule, 0); + erts_smp_atomic32_set(&doing_sys_schedule, 0); if (flgs & ~ERTS_SSI_FLG_SUSPENDED) - erts_smp_atomic_band(&ssi->flags, ERTS_SSI_FLG_SUSPENDED); + erts_smp_atomic32_band(&ssi->flags, ERTS_SSI_FLG_SUSPENDED); sched_active_sys(esdp->no, rq); } } @@ -1108,15 +1236,15 @@ scheduler_wait(long *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq) ERTS_SMP_LC_ASSERT(erts_smp_lc_runq_is_locked(rq)); } -static ERTS_INLINE long +static ERTS_INLINE erts_aint32_t ssi_flags_set_wake(ErtsSchedulerSleepInfo *ssi) { /* reset all flags but suspended */ - long oflgs; - long nflgs = 0; - long xflgs = ERTS_SSI_FLG_SLEEPING|ERTS_SSI_FLG_WAITING; + erts_aint32_t oflgs; + erts_aint32_t nflgs = 0; + erts_aint32_t xflgs = ERTS_SSI_FLG_SLEEPING|ERTS_SSI_FLG_WAITING; while (1) { - oflgs = erts_smp_atomic_cmpxchg(&ssi->flags, nflgs, xflgs); + oflgs = erts_smp_atomic32_cmpxchg(&ssi->flags, nflgs, xflgs); if (oflgs == xflgs) return oflgs; nflgs = oflgs & ERTS_SSI_FLG_SUSPENDED; @@ -1148,7 +1276,7 @@ wake_scheduler(ErtsRunQueue *rq, int incq, int one) if (!ssi) erts_smp_spin_unlock(&sl->lock); else if (one) { - long flgs; + erts_aint32_t flgs; if (ssi->prev) ssi->prev->next = ssi->next; else { @@ -1195,15 +1323,17 @@ wake_all_schedulers(void) static ERTS_INLINE int chk_wake_sched(ErtsRunQueue *crq, int ix, int activate) { - long iflgs; + erts_aint32_t iflgs; ErtsRunQueue *wrq; if (crq->ix == ix) return 0; wrq = ERTS_RUNQ_IX(ix); - iflgs = erts_smp_atomic_read(&wrq->info_flags); + iflgs = erts_smp_atomic32_read(&wrq->info_flags); if (!(iflgs & (ERTS_RUNQ_IFLG_SUSPENDED|ERTS_RUNQ_IFLG_NONEMPTY))) { if (activate) { - if (ix == erts_smp_atomic_cmpxchg(&balance_info.active_runqs, ix+1, ix)) { + if (ix == erts_smp_atomic32_cmpxchg(&balance_info.active_runqs, + ix+1, + ix)) { erts_smp_xrunq_lock(crq, wrq); wrq->flags &= ~ERTS_RUNQ_FLG_INACTIVE; erts_smp_xrunq_unlock(crq, wrq); @@ -1220,8 +1350,8 @@ wake_scheduler_on_empty_runq(ErtsRunQueue *crq) { int ix = crq->ix; int stop_ix = ix; - int active_ix = erts_smp_atomic_read(&balance_info.active_runqs); - int balance_ix = erts_smp_atomic_read(&balance_info.used_runqs); + int active_ix = erts_smp_atomic32_read(&balance_info.active_runqs); + int balance_ix = erts_smp_atomic32_read(&balance_info.used_runqs); if (active_ix > balance_ix) active_ix = balance_ix; @@ -1273,7 +1403,7 @@ erts_sched_notify_check_cpu_bind(void) int ix; if (erts_common_run_queue) { for (ix = 0; ix < erts_no_schedulers; ix++) - erts_smp_atomic_set(&ERTS_SCHEDULER_IX(ix)->chk_cpu_bind, 1); + erts_smp_atomic32_set(&ERTS_SCHEDULER_IX(ix)->chk_cpu_bind, 1); wake_all_schedulers(); } else { @@ -1441,14 +1571,15 @@ evacuate_run_queue(ErtsRunQueue *evac_rq, ErtsRunQueue *rq) erts_smp_runq_lock(evac_rq); - erts_smp_atomic_bor(&evac_rq->scheduler->ssi->flags, ERTS_SSI_FLG_SUSPENDED); + erts_smp_atomic32_bor(&evac_rq->scheduler->ssi->flags, + ERTS_SSI_FLG_SUSPENDED); evac_rq->flags &= ~ERTS_RUNQ_FLGS_IMMIGRATE_QMASK; evac_rq->flags |= (ERTS_RUNQ_FLGS_EMIGRATE_QMASK | ERTS_RUNQ_FLGS_EVACUATE_QMASK | ERTS_RUNQ_FLG_SUSPENDED); - erts_smp_atomic_bor(&evac_rq->info_flags, ERTS_RUNQ_IFLG_SUSPENDED); + erts_smp_atomic32_bor(&evac_rq->info_flags, ERTS_RUNQ_IFLG_SUSPENDED); /* * Need to set up evacuation paths first since we * may release the run queue lock on evac_rq @@ -1697,7 +1828,7 @@ static ERTS_INLINE int check_possible_steal_victim(ErtsRunQueue *rq, int *rq_lockedp, int vix) { ErtsRunQueue *vrq = ERTS_RUNQ_IX(vix); - long iflgs = erts_smp_atomic_read(&vrq->info_flags); + erts_aint32_t iflgs = erts_smp_atomic32_read(&vrq->info_flags); if (iflgs & ERTS_RUNQ_IFLG_NONEMPTY) return try_steal_task_from_victim(rq, rq_lockedp, vrq); else @@ -1727,8 +1858,8 @@ try_steal_task(ErtsRunQueue *rq) ERTS_SMP_LC_CHK_RUNQ_LOCK(rq, rq_locked); - active_rqs = erts_smp_atomic_read(&balance_info.active_runqs); - blnc_rqs = erts_smp_atomic_read(&balance_info.used_runqs); + active_rqs = erts_smp_atomic32_read(&balance_info.active_runqs); + blnc_rqs = erts_smp_atomic32_read(&balance_info.used_runqs); if (active_rqs > blnc_rqs) active_rqs = blnc_rqs; @@ -1739,7 +1870,7 @@ try_steal_task(ErtsRunQueue *rq) if (active_rqs < blnc_rqs) { int no = blnc_rqs - active_rqs; int stop_ix = vix = active_rqs + rq->ix % no; - while (erts_smp_atomic_read(&no_empty_run_queues) < blnc_rqs) { + while (erts_smp_atomic32_read(&no_empty_run_queues) < blnc_rqs) { res = check_possible_steal_victim(rq, &rq_locked, vix); if (res) goto done; @@ -1754,7 +1885,7 @@ try_steal_task(ErtsRunQueue *rq) vix = rq->ix; /* ... then try to steal a job from another active queue... */ - while (erts_smp_atomic_read(&no_empty_run_queues) < blnc_rqs) { + while (erts_smp_atomic32_read(&no_empty_run_queues) < blnc_rqs) { vix++; if (vix >= active_rqs) vix = 0; @@ -1850,15 +1981,15 @@ check_balance(ErtsRunQueue *c_rq) int forced, active, current_active, oowc, half_full_scheds, full_scheds, mmax_len, blnc_no_rqs, qix, pix, freds_hist_ix; - if (erts_smp_atomic_xchg(&balance_info.checking_balance, 1)) { + if (erts_smp_atomic32_xchg(&balance_info.checking_balance, 1)) { c_rq->check_balance_reds = INT_MAX; return; } - blnc_no_rqs = (int) erts_smp_atomic_read(&balance_info.used_runqs); + blnc_no_rqs = (int) erts_smp_atomic32_read(&balance_info.used_runqs); if (blnc_no_rqs == 1) { c_rq->check_balance_reds = INT_MAX; - erts_smp_atomic_set(&balance_info.checking_balance, 0); + erts_smp_atomic32_set(&balance_info.checking_balance, 0); return; } @@ -1866,7 +1997,7 @@ check_balance(ErtsRunQueue *c_rq) if (balance_info.halftime) { balance_info.halftime = 0; - erts_smp_atomic_set(&balance_info.checking_balance, 0); + erts_smp_atomic32_set(&balance_info.checking_balance, 0); ERTS_FOREACH_RUNQ(rq, { if (rq->waiting) @@ -1894,12 +2025,12 @@ check_balance(ErtsRunQueue *c_rq) forced = balance_info.forced_check_balance; balance_info.forced_check_balance = 0; - blnc_no_rqs = (int) erts_smp_atomic_read(&balance_info.used_runqs); + blnc_no_rqs = (int) erts_smp_atomic32_read(&balance_info.used_runqs); if (blnc_no_rqs == 1) { erts_smp_mtx_unlock(&balance_info.update_mtx); erts_smp_runq_lock(c_rq); c_rq->check_balance_reds = INT_MAX; - erts_smp_atomic_set(&balance_info.checking_balance, 0); + erts_smp_atomic32_set(&balance_info.checking_balance, 0); return; } @@ -1908,7 +2039,7 @@ check_balance(ErtsRunQueue *c_rq) if (balance_info.full_reds_history_index >= ERTS_FULL_REDS_HISTORY_SIZE) balance_info.full_reds_history_index = 0; - current_active = erts_smp_atomic_read(&balance_info.active_runqs); + current_active = erts_smp_atomic32_read(&balance_info.active_runqs); /* Read balance information for all run queues */ for (qix = 0; qix < blnc_no_rqs; qix++) { @@ -2243,10 +2374,10 @@ erts_fprintf(stderr, "--------------------------------\n"); } balance_info.last_active_runqs = active; - erts_smp_atomic_set(&balance_info.active_runqs, active); + erts_smp_atomic32_set(&balance_info.active_runqs, active); balance_info.halftime = 1; - erts_smp_atomic_set(&balance_info.checking_balance, 0); + erts_smp_atomic32_set(&balance_info.checking_balance, 0); /* Write migration paths and reset balance statistics in all queues */ for (qix = 0; qix < blnc_no_rqs; qix++) { @@ -2395,7 +2526,7 @@ erts_init_scheduling(int mrq, int no_schedulers, int no_schedulers_online) ASSERT((((UWord) erts_aligned_run_queues) & ERTS_CACHE_LINE_MASK) == 0); #ifdef ERTS_SMP - erts_smp_atomic_init(&no_empty_run_queues, 0); + erts_smp_atomic32_init(&no_empty_run_queues, 0); #endif erts_no_run_queues = n; @@ -2405,7 +2536,7 @@ erts_init_scheduling(int mrq, int no_schedulers, int no_schedulers_online) ErtsRunQueue *rq = ERTS_RUNQ_IX(ix); rq->ix = ix; - erts_smp_atomic_init(&rq->info_flags, ERTS_RUNQ_IFLG_NONEMPTY); + erts_smp_atomic32_init(&rq->info_flags, ERTS_RUNQ_IFLG_NONEMPTY); /* make sure that the "extra" id correponds to the schedulers * id if the esdp->no <-> ix+1 mapping change. @@ -2502,9 +2633,9 @@ erts_init_scheduling(int mrq, int no_schedulers, int no_schedulers_online) ssi->next = NULL; ssi->prev = NULL; #endif - erts_smp_atomic_init(&ssi->flags, 0); + erts_smp_atomic32_init(&ssi->flags, 0); ssi->event = NULL; /* initialized in sched_thread_func */ - erts_smp_atomic_init(&ssi->aux_work, 0); + erts_smp_atomic32_init(&ssi->aux_work, 0); } #endif @@ -2555,7 +2686,7 @@ erts_init_scheduling(int mrq, int no_schedulers, int no_schedulers_online) } #ifdef ERTS_SMP - erts_smp_atomic_init(&esdp->chk_cpu_bind, 0); + erts_smp_atomic32_init(&esdp->chk_cpu_bind, 0); #endif } @@ -2563,21 +2694,21 @@ erts_init_scheduling(int mrq, int no_schedulers, int no_schedulers_online) erts_smp_mtx_init(&schdlr_sspnd.mtx, "schdlr_sspnd"); erts_smp_cnd_init(&schdlr_sspnd.cnd); - erts_smp_atomic_init(&schdlr_sspnd.changing, 0); + erts_smp_atomic32_init(&schdlr_sspnd.changing, 0); schdlr_sspnd.online = no_schedulers_online; schdlr_sspnd.curr_online = no_schedulers; - erts_smp_atomic_init(&schdlr_sspnd.msb.ongoing, 0); - erts_smp_atomic_init(&schdlr_sspnd.active, no_schedulers); + erts_smp_atomic32_init(&schdlr_sspnd.msb.ongoing, 0); + erts_smp_atomic32_init(&schdlr_sspnd.active, no_schedulers); schdlr_sspnd.msb.procs = NULL; - erts_smp_atomic_set(&balance_info.used_runqs, - erts_common_run_queue ? 1 : no_schedulers_online); - erts_smp_atomic_init(&balance_info.active_runqs, no_schedulers); + erts_smp_atomic32_set(&balance_info.used_runqs, + erts_common_run_queue ? 1 : no_schedulers_online); + erts_smp_atomic32_init(&balance_info.active_runqs, no_schedulers); balance_info.last_active_runqs = no_schedulers; erts_smp_mtx_init(&balance_info.update_mtx, "migration_info_update"); balance_info.forced_check_balance = 0; balance_info.halftime = 1; balance_info.full_reds_history_index = 0; - erts_smp_atomic_init(&balance_info.checking_balance, 0); + erts_smp_atomic32_init(&balance_info.checking_balance, 0); balance_info.prev_rise.active_runqs = 0; balance_info.prev_rise.max_len = 0; balance_info.prev_rise.reds = 0; @@ -2586,8 +2717,8 @@ erts_init_scheduling(int mrq, int no_schedulers, int no_schedulers_online) if (no_schedulers_online < no_schedulers) { if (erts_common_run_queue) { for (ix = no_schedulers_online; ix < no_schedulers; ix++) - erts_smp_atomic_bor(&ERTS_SCHED_SLEEP_INFO_IX(ix)->flags, - ERTS_SSI_FLG_SUSPENDED); + erts_smp_atomic32_bor(&ERTS_SCHED_SLEEP_INFO_IX(ix)->flags, + ERTS_SSI_FLG_SUSPENDED); } else { for (ix = no_schedulers_online; ix < erts_no_run_queues; ix++) @@ -2601,7 +2732,9 @@ erts_init_scheduling(int mrq, int no_schedulers, int no_schedulers_online) ERTS_SCHDLR_SSPND_CHNG_SET((ERTS_SCHDLR_SSPND_CHNG_ONLN | ERTS_SCHDLR_SSPND_CHNG_WAITER), 0); - erts_smp_atomic_init(&doing_sys_schedule, 0); + erts_smp_atomic32_init(&doing_sys_schedule, 0); + + init_misc_aux_work(); #else /* !ERTS_SMP */ { @@ -2615,7 +2748,7 @@ erts_init_scheduling(int mrq, int no_schedulers, int no_schedulers_online) erts_no_schedulers = 1; #endif - erts_smp_atomic_init(&function_calls, 0); + erts_smp_atomic32_init(&function_calls, 0); /* init port tasks */ erts_port_task_init(); @@ -2730,6 +2863,19 @@ resume_process(Process *p) p->rstatus = P_FREE; } +int +erts_get_max_no_executing_schedulers(void) +{ +#ifdef ERTS_SMP + if (erts_smp_atomic32_read(&schdlr_sspnd.changing)) + return (int) erts_no_schedulers; + ERTS_THR_MEMORY_BARRIER; + return (int) erts_smp_atomic32_read(&schdlr_sspnd.active); +#else + return 1; +#endif +} + #ifdef ERTS_SMP static void @@ -2748,13 +2894,13 @@ static void scheduler_ix_resume_wake(Uint ix) { ErtsSchedulerSleepInfo *ssi = ERTS_SCHED_SLEEP_INFO_IX(ix); - long xflgs = (ERTS_SSI_FLG_SLEEPING - | ERTS_SSI_FLG_TSE_SLEEPING - | ERTS_SSI_FLG_WAITING - | ERTS_SSI_FLG_SUSPENDED); - long oflgs; + erts_aint32_t xflgs = (ERTS_SSI_FLG_SLEEPING + | ERTS_SSI_FLG_TSE_SLEEPING + | ERTS_SSI_FLG_WAITING + | ERTS_SSI_FLG_SUSPENDED); + erts_aint32_t oflgs; do { - oflgs = erts_smp_atomic_cmpxchg(&ssi->flags, 0, xflgs); + oflgs = erts_smp_atomic32_cmpxchg(&ssi->flags, 0, xflgs); if (oflgs == xflgs) { erts_sched_finish_poke(ssi, oflgs); break; @@ -2763,17 +2909,17 @@ scheduler_ix_resume_wake(Uint ix) } while (oflgs & ERTS_SSI_FLG_SUSPENDED); } -static long -sched_prep_spin_suspended(ErtsSchedulerSleepInfo *ssi, long xpct) +static erts_aint32_t +sched_prep_spin_suspended(ErtsSchedulerSleepInfo *ssi, erts_aint32_t xpct) { - long oflgs; - long nflgs = (ERTS_SSI_FLG_SLEEPING - | ERTS_SSI_FLG_WAITING - | ERTS_SSI_FLG_SUSPENDED); - long xflgs = xpct; + erts_aint32_t oflgs; + erts_aint32_t nflgs = (ERTS_SSI_FLG_SLEEPING + | ERTS_SSI_FLG_WAITING + | ERTS_SSI_FLG_SUSPENDED); + erts_aint32_t xflgs = xpct; do { - oflgs = erts_smp_atomic_cmpxchg(&ssi->flags, nflgs, xflgs); + oflgs = erts_smp_atomic32_cmpxchg(&ssi->flags, nflgs, xflgs); if (oflgs == xflgs) return nflgs; xflgs = oflgs; @@ -2782,15 +2928,15 @@ sched_prep_spin_suspended(ErtsSchedulerSleepInfo *ssi, long xpct) return oflgs; } -static long +static erts_aint32_t sched_spin_suspended(ErtsSchedulerSleepInfo *ssi, int spincount) { int until_yield = ERTS_SCHED_SPIN_UNTIL_YIELD; int sc = spincount; - long flgs; + erts_aint32_t flgs; do { - flgs = erts_smp_atomic_read(&ssi->flags); + flgs = erts_smp_atomic32_read(&ssi->flags); if ((flgs & (ERTS_SSI_FLG_SLEEPING | ERTS_SSI_FLG_WAITING | ERTS_SSI_FLG_SUSPENDED)) @@ -2808,22 +2954,22 @@ sched_spin_suspended(ErtsSchedulerSleepInfo *ssi, int spincount) return flgs; } -static long +static erts_aint32_t sched_set_suspended_sleeptype(ErtsSchedulerSleepInfo *ssi) { - long oflgs; - long nflgs = (ERTS_SSI_FLG_SLEEPING - | ERTS_SSI_FLG_TSE_SLEEPING - | ERTS_SSI_FLG_WAITING - | ERTS_SSI_FLG_SUSPENDED); - long xflgs = (ERTS_SSI_FLG_SLEEPING - | ERTS_SSI_FLG_WAITING - | ERTS_SSI_FLG_SUSPENDED); + erts_aint32_t oflgs; + erts_aint32_t nflgs = (ERTS_SSI_FLG_SLEEPING + | ERTS_SSI_FLG_TSE_SLEEPING + | ERTS_SSI_FLG_WAITING + | ERTS_SSI_FLG_SUSPENDED); + erts_aint32_t xflgs = (ERTS_SSI_FLG_SLEEPING + | ERTS_SSI_FLG_WAITING + | ERTS_SSI_FLG_SUSPENDED); erts_tse_reset(ssi->event); while (1) { - oflgs = erts_smp_atomic_cmpxchg(&ssi->flags, nflgs, xflgs); + oflgs = erts_smp_atomic32_cmpxchg(&ssi->flags, nflgs, xflgs); if (oflgs == xflgs) return nflgs; if ((oflgs & (ERTS_SSI_FLG_SLEEPING @@ -2841,8 +2987,8 @@ sched_set_suspended_sleeptype(ErtsSchedulerSleepInfo *ssi) static void suspend_scheduler(ErtsSchedulerData *esdp) { - long flgs; - int changing; + erts_aint32_t flgs; + erts_aint32_t changing; long no = (long) esdp->no; ErtsSchedulerSleepInfo *ssi = esdp->ssi; long active_schedulers; @@ -2850,7 +2996,7 @@ suspend_scheduler(ErtsSchedulerData *esdp) int wake = 0; #if defined(ERTS_SCHED_NEED_NONBLOCKABLE_AUX_WORK) \ || defined(ERTS_SCHED_NEED_BLOCKABLE_AUX_WORK) - long aux_work; + erts_aint32_t aux_work; #endif /* @@ -2878,15 +3024,15 @@ suspend_scheduler(ErtsSchedulerData *esdp) flgs = sched_prep_spin_suspended(ssi, ERTS_SSI_FLG_SUSPENDED); if (flgs & ERTS_SSI_FLG_SUSPENDED) { - active_schedulers = erts_smp_atomic_dectest(&schdlr_sspnd.active); + active_schedulers = erts_smp_atomic32_dectest(&schdlr_sspnd.active); ASSERT(active_schedulers >= 1); - changing = erts_smp_atomic_read(&schdlr_sspnd.changing); + changing = erts_smp_atomic32_read(&schdlr_sspnd.changing); if (changing & ERTS_SCHDLR_SSPND_CHNG_MSB) { if (active_schedulers == schdlr_sspnd.msb.wait_active) wake = 1; if (active_schedulers == 1) { - changing = erts_smp_atomic_band(&schdlr_sspnd.changing, - ~ERTS_SCHDLR_SSPND_CHNG_MSB); + changing = erts_smp_atomic32_band(&schdlr_sspnd.changing, + ~ERTS_SCHDLR_SSPND_CHNG_MSB); changing &= ~ERTS_SCHDLR_SSPND_CHNG_MSB; } } @@ -2908,8 +3054,8 @@ suspend_scheduler(ErtsSchedulerData *esdp) && schdlr_sspnd.curr_online == schdlr_sspnd.wait_curr_online) wake = 1; if (schdlr_sspnd.online == schdlr_sspnd.curr_online) { - changing = erts_smp_atomic_band(&schdlr_sspnd.changing, - ~ERTS_SCHDLR_SSPND_CHNG_ONLN); + changing = erts_smp_atomic32_band(&schdlr_sspnd.changing, + ~ERTS_SCHDLR_SSPND_CHNG_ONLN); changing &= ~ERTS_SCHDLR_SSPND_CHNG_ONLN; } } @@ -2919,29 +3065,30 @@ suspend_scheduler(ErtsSchedulerData *esdp) wake = 0; } - flgs = erts_smp_atomic_read(&ssi->flags); + flgs = erts_smp_atomic32_read(&ssi->flags); if (!(flgs & ERTS_SSI_FLG_SUSPENDED)) break; erts_smp_mtx_unlock(&schdlr_sspnd.mtx); #ifdef ERTS_SCHED_NEED_BLOCKABLE_AUX_WORK - aux_work = erts_smp_atomic_read(&ssi->aux_work); + aux_work = erts_smp_atomic32_read(&ssi->aux_work); blockable_aux_work: blockable_aux_work(esdp, ssi, aux_work); #endif erts_smp_activity_begin(ERTS_ACTIVITY_WAIT, NULL, NULL, NULL); while (1) { - long flgs; + erts_aint32_t flgs; #ifdef ERTS_SCHED_NEED_NONBLOCKABLE_AUX_WORK #ifndef ERTS_SCHED_NEED_BLOCKABLE_AUX_WORK - aux_work = erts_smp_atomic_read(&ssi->aux_work); + aux_work = erts_smp_atomic32_read(&ssi->aux_work); #endif nonblockable_aux_work(esdp, ssi, aux_work); #endif - flgs = sched_spin_suspended(ssi, ERTS_SCHED_SUSPEND_SLEEP_SPINCOUNT); + flgs = sched_spin_suspended(ssi, + ERTS_SCHED_SUSPEND_SLEEP_SPINCOUNT); if (flgs == (ERTS_SSI_FLG_SLEEPING | ERTS_SSI_FLG_WAITING | ERTS_SSI_FLG_SUSPENDED)) { @@ -2961,13 +3108,13 @@ suspend_scheduler(ErtsSchedulerData *esdp) | ERTS_SSI_FLG_SUSPENDED)); if (!(flgs & ERTS_SSI_FLG_SUSPENDED)) break; - changing = erts_smp_atomic_read(&schdlr_sspnd.changing); + changing = erts_smp_atomic32_read(&schdlr_sspnd.changing); if (changing & ~ERTS_SCHDLR_SSPND_CHNG_WAITER) break; #ifdef ERTS_SCHED_NEED_BLOCKABLE_AUX_WORK - aux_work = erts_smp_atomic_read(&ssi->aux_work); + aux_work = erts_smp_atomic32_read(&ssi->aux_work); if (aux_work & ERTS_SSI_BLOCKABLE_AUX_WORK_MASK) { erts_smp_activity_end(ERTS_ACTIVITY_WAIT, NULL, NULL, NULL); goto blockable_aux_work; @@ -2979,19 +3126,19 @@ suspend_scheduler(ErtsSchedulerData *esdp) erts_smp_activity_end(ERTS_ACTIVITY_WAIT, NULL, NULL, NULL); erts_smp_mtx_lock(&schdlr_sspnd.mtx); - changing = erts_smp_atomic_read(&schdlr_sspnd.changing); + changing = erts_smp_atomic32_read(&schdlr_sspnd.changing); } - active_schedulers = erts_smp_atomic_inctest(&schdlr_sspnd.active); - changing = erts_smp_atomic_read(&schdlr_sspnd.changing); + active_schedulers = erts_smp_atomic32_inctest(&schdlr_sspnd.active); + changing = erts_smp_atomic32_read(&schdlr_sspnd.changing); if ((changing & ERTS_SCHDLR_SSPND_CHNG_MSB) && schdlr_sspnd.online == active_schedulers) { - erts_smp_atomic_band(&schdlr_sspnd.changing, - ~ERTS_SCHDLR_SSPND_CHNG_MSB); + erts_smp_atomic32_band(&schdlr_sspnd.changing, + ~ERTS_SCHDLR_SSPND_CHNG_MSB); } ASSERT(no <= schdlr_sspnd.online); - ASSERT(!erts_smp_atomic_read(&schdlr_sspnd.msb.ongoing)); + ASSERT(!erts_smp_atomic32_read(&schdlr_sspnd.msb.ongoing)); } @@ -3020,7 +3167,7 @@ do { \ (RQ)->flags |= (ERTS_RUNQ_FLG_OUT_OF_WORK \ | ERTS_RUNQ_FLG_HALFTIME_OUT_OF_WORK); \ (RQ)->check_balance_reds = ERTS_RUNQ_CALL_CHECK_BALANCE_REDS; \ - erts_smp_atomic_band(&(RQ)->info_flags, ~ERTS_RUNQ_IFLG_SUSPENDED); \ + erts_smp_atomic32_band(&(RQ)->info_flags, ~ERTS_RUNQ_IFLG_SUSPENDED);\ for (pix__ = 0; pix__ < ERTS_NO_PROC_PRIO_LEVELS; pix__++) { \ (RQ)->procs.prio_info[pix__].max_len = 0; \ (RQ)->procs.prio_info[pix__].reds = 0; \ @@ -3062,9 +3209,9 @@ erts_schedulers_state(Uint *total, int yield_allowed) { int res; - long changing; + erts_aint32_t changing; erts_smp_mtx_lock(&schdlr_sspnd.mtx); - changing = erts_smp_atomic_read(&schdlr_sspnd.changing); + changing = erts_smp_atomic32_read(&schdlr_sspnd.changing); if (yield_allowed && (changing & ~ERTS_SCHDLR_SSPND_CHNG_WAITER)) res = ERTS_SCHDLR_SSPND_YIELD_RESTART; else { @@ -3085,7 +3232,7 @@ erts_set_schedulers_online(Process *p, Sint *old_no) { int ix, res, no, have_unlocked_plocks; - long changing; + erts_aint32_t changing; if (new_no < 1 || erts_no_schedulers < new_no) return ERTS_SCHDLR_SSPND_EINVAL; @@ -3095,7 +3242,7 @@ erts_set_schedulers_online(Process *p, have_unlocked_plocks = 0; no = (int) new_no; - changing = erts_smp_atomic_read(&schdlr_sspnd.changing); + changing = erts_smp_atomic32_read(&schdlr_sspnd.changing); if (changing) { res = ERTS_SCHDLR_SSPND_YIELD_RESTART; } @@ -3142,7 +3289,7 @@ erts_set_schedulers_online(Process *p, ErtsRunQueue *to_rq = ERTS_RUNQ_IX(ix % no); evacuate_run_queue(from_rq, to_rq); } - erts_smp_atomic_set(&balance_info.used_runqs, no); + erts_smp_atomic32_set(&balance_info.used_runqs, no); erts_smp_mtx_unlock(&balance_info.update_mtx); erts_smp_mtx_lock(&schdlr_sspnd.mtx); } @@ -3170,8 +3317,8 @@ erts_set_schedulers_online(Process *p, for (ix = no; ix < online; ix++) { ErtsSchedulerSleepInfo *ssi; ssi = ERTS_SCHED_SLEEP_INFO_IX(ix); - erts_smp_atomic_bor(&ssi->flags, - ERTS_SSI_FLG_SUSPENDED); + erts_smp_atomic32_bor(&ssi->flags, + ERTS_SSI_FLG_SUSPENDED); } wake_all_schedulers(); } @@ -3196,7 +3343,7 @@ erts_set_schedulers_online(Process *p, for (ix = erts_no_run_queues-1; ix >= no; ix--) evacuate_run_queue(ERTS_RUNQ_IX(ix), ERTS_RUNQ_IX(ix % no)); - erts_smp_atomic_set(&balance_info.used_runqs, no); + erts_smp_atomic32_set(&balance_info.used_runqs, no); erts_smp_mtx_unlock(&balance_info.update_mtx); erts_smp_mtx_lock(&schdlr_sspnd.mtx); for (ix = no; ix < online; ix++) { @@ -3218,10 +3365,11 @@ erts_set_schedulers_online(Process *p, NULL); ASSERT(res != ERTS_SCHDLR_SSPND_DONE ? (ERTS_SCHDLR_SSPND_CHNG_WAITER - & erts_smp_atomic_read(&schdlr_sspnd.changing)) + & erts_smp_atomic32_read(&schdlr_sspnd.changing)) : (ERTS_SCHDLR_SSPND_CHNG_WAITER - == erts_smp_atomic_read(&schdlr_sspnd.changing))); - erts_smp_atomic_band(&schdlr_sspnd.changing, ~ERTS_SCHDLR_SSPND_CHNG_WAITER); + == erts_smp_atomic32_read(&schdlr_sspnd.changing))); + erts_smp_atomic32_band(&schdlr_sspnd.changing, + ~ERTS_SCHDLR_SSPND_CHNG_WAITER); } } @@ -3236,11 +3384,11 @@ ErtsSchedSuspendResult erts_block_multi_scheduling(Process *p, ErtsProcLocks plocks, int on, int all) { int ix, res, have_unlocked_plocks = 0; - long changing; + erts_aint32_t changing; ErtsProcList *plp; erts_smp_mtx_lock(&schdlr_sspnd.mtx); - changing = erts_smp_atomic_read(&schdlr_sspnd.changing); + changing = erts_smp_atomic32_read(&schdlr_sspnd.changing); if (changing) { res = ERTS_SCHDLR_SSPND_YIELD_RESTART; /* Yield */ } @@ -3250,7 +3398,7 @@ erts_block_multi_scheduling(Process *p, ErtsProcLocks plocks, int on, int all) plp->next = schdlr_sspnd.msb.procs; schdlr_sspnd.msb.procs = plp; p->flags |= F_HAVE_BLCKD_MSCHED; - ASSERT(erts_smp_atomic_read(&schdlr_sspnd.active) == 1); + ASSERT(erts_smp_atomic32_read(&schdlr_sspnd.active) == 1); ASSERT(p->scheduler_data->no == 1); res = ERTS_SCHDLR_SSPND_DONE_MSCHED_BLOCKED; } @@ -3261,11 +3409,11 @@ erts_block_multi_scheduling(Process *p, ErtsProcLocks plocks, int on, int all) have_unlocked_plocks = 1; erts_smp_proc_unlock(p, plocks); } - ASSERT(0 == erts_smp_atomic_read(&schdlr_sspnd.msb.ongoing)); - erts_smp_atomic_set(&schdlr_sspnd.msb.ongoing, 1); + ASSERT(0 == erts_smp_atomic32_read(&schdlr_sspnd.msb.ongoing)); + erts_smp_atomic32_set(&schdlr_sspnd.msb.ongoing, 1); if (online == 1) { res = ERTS_SCHDLR_SSPND_DONE_MSCHED_BLOCKED; - ASSERT(erts_smp_atomic_read(&schdlr_sspnd.active) == 1); + ASSERT(erts_smp_atomic32_read(&schdlr_sspnd.active) == 1); ASSERT(p->scheduler_data->no == 1); } else { @@ -3285,14 +3433,14 @@ erts_block_multi_scheduling(Process *p, ErtsProcLocks plocks, int on, int all) } if (erts_common_run_queue) { for (ix = 1; ix < online; ix++) - erts_smp_atomic_bor(&ERTS_SCHED_SLEEP_INFO_IX(ix)->flags, - ERTS_SSI_FLG_SUSPENDED); + erts_smp_atomic32_bor(&ERTS_SCHED_SLEEP_INFO_IX(ix)->flags, + ERTS_SSI_FLG_SUSPENDED); wake_all_schedulers(); } else { erts_smp_mtx_unlock(&schdlr_sspnd.mtx); erts_smp_mtx_lock(&balance_info.update_mtx); - erts_smp_atomic_set(&balance_info.used_runqs, 1); + erts_smp_atomic32_set(&balance_info.used_runqs, 1); for (ix = 0; ix < online; ix++) { ErtsRunQueue *rq = ERTS_RUNQ_IX(ix); erts_smp_runq_lock(rq); @@ -3314,7 +3462,7 @@ erts_block_multi_scheduling(Process *p, ErtsProcLocks plocks, int on, int all) susp_sched_prep_block, susp_sched_resume_block, NULL); - while (erts_smp_atomic_read(&schdlr_sspnd.active) + while (erts_smp_atomic32_read(&schdlr_sspnd.active) != schdlr_sspnd.msb.wait_active) erts_smp_cnd_wait(&schdlr_sspnd.cnd, &schdlr_sspnd.mtx); erts_smp_activity_end(ERTS_ACTIVITY_WAIT, @@ -3323,11 +3471,11 @@ erts_block_multi_scheduling(Process *p, ErtsProcLocks plocks, int on, int all) NULL); ASSERT(res != ERTS_SCHDLR_SSPND_DONE_MSCHED_BLOCKED ? (ERTS_SCHDLR_SSPND_CHNG_WAITER - & erts_smp_atomic_read(&schdlr_sspnd.changing)) + & erts_smp_atomic32_read(&schdlr_sspnd.changing)) : (ERTS_SCHDLR_SSPND_CHNG_WAITER - == erts_smp_atomic_read(&schdlr_sspnd.changing))); - erts_smp_atomic_band(&schdlr_sspnd.changing, - ~ERTS_SCHDLR_SSPND_CHNG_WAITER); + == erts_smp_atomic32_read(&schdlr_sspnd.changing))); + erts_smp_atomic32_band(&schdlr_sspnd.changing, + ~ERTS_SCHDLR_SSPND_CHNG_WAITER); } plp = proclist_create(p); plp->next = schdlr_sspnd.msb.procs; @@ -3394,16 +3542,16 @@ erts_block_multi_scheduling(Process *p, ErtsProcLocks plocks, int on, int all) }); #endif p->flags &= ~F_HAVE_BLCKD_MSCHED; - erts_smp_atomic_set(&schdlr_sspnd.msb.ongoing, 0); + erts_smp_atomic32_set(&schdlr_sspnd.msb.ongoing, 0); if (schdlr_sspnd.online == 1) { /* No schedulers to resume */ - ASSERT(erts_smp_atomic_read(&schdlr_sspnd.active) == 1); + ASSERT(erts_smp_atomic32_read(&schdlr_sspnd.active) == 1); ERTS_SCHDLR_SSPND_CHNG_SET(0, ERTS_SCHDLR_SSPND_CHNG_MSB); } else if (erts_common_run_queue) { for (ix = 1; ix < schdlr_sspnd.online; ix++) - erts_smp_atomic_band(&ERTS_SCHED_SLEEP_INFO_IX(ix)->flags, - ~ERTS_SSI_FLG_SUSPENDED); + erts_smp_atomic32_band(&ERTS_SCHED_SLEEP_INFO_IX(ix)->flags, + ~ERTS_SSI_FLG_SUSPENDED); wake_all_schedulers(); } else { @@ -3429,7 +3577,7 @@ erts_block_multi_scheduling(Process *p, ErtsProcLocks plocks, int on, int all) evacuate_run_queue(ERTS_RUNQ_IX(ix), ERTS_RUNQ_IX(ix % online)); - erts_smp_atomic_set(&balance_info.used_runqs, online); + erts_smp_atomic32_set(&balance_info.used_runqs, online); /* Make sure that we balance soon... */ balance_info.forced_check_balance = 1; erts_smp_runq_lock(ERTS_RUNQ_IX(0)); @@ -3453,7 +3601,7 @@ void erts_dbg_multi_scheduling_return_trap(Process *p, Eterm return_value) { if (return_value == am_blocked) { - long active = erts_smp_atomic_read(&schdlr_sspnd.active); + erts_aint32_t active = erts_smp_atomic32_read(&schdlr_sspnd.active); ASSERT(1 <= active && active <= 2); ASSERT(ERTS_PROC_GET_SCHDATA(p)->no == 1); } @@ -3536,12 +3684,12 @@ sched_thread_func(void *vesdp) erts_thread_init_float(); erts_smp_mtx_lock(&schdlr_sspnd.mtx); - ASSERT(erts_smp_atomic_read(&schdlr_sspnd.changing) + ASSERT(erts_smp_atomic32_read(&schdlr_sspnd.changing) & ERTS_SCHDLR_SSPND_CHNG_ONLN); if (--schdlr_sspnd.curr_online == schdlr_sspnd.wait_curr_online) { - erts_smp_atomic_band(&schdlr_sspnd.changing, - ~ERTS_SCHDLR_SSPND_CHNG_ONLN); + erts_smp_atomic32_band(&schdlr_sspnd.changing, + ~ERTS_SCHDLR_SSPND_CHNG_ONLN); if (((ErtsSchedulerData *) vesdp)->no != 1) erts_smp_cnd_signal(&schdlr_sspnd.cnd); } @@ -4914,10 +5062,10 @@ Process *schedule(Process *p, int calls) { ErtsRunQueue *rq; ErtsRunPrioQueue *rpq; - long dt; + erts_aint_t dt; ErtsSchedulerData *esdp; int context_reds; - long fcalls; + int fcalls; int input_reductions; int actual_reds; int reds; @@ -4940,7 +5088,7 @@ Process *schedule(Process *p, int calls) esdp = erts_get_scheduler_data(); rq = erts_get_runq_current(esdp); ASSERT(esdp); - fcalls = erts_smp_atomic_read(&function_calls); + fcalls = (int) erts_smp_atomic32_read(&function_calls); actual_reds = reds = 0; erts_smp_runq_lock(rq); } else { @@ -4958,7 +5106,7 @@ Process *schedule(Process *p, int calls) reds = ERTS_PROC_MIN_CONTEXT_SWITCH_REDS_COST; esdp->virtual_reds = 0; - fcalls = erts_smp_atomic_addtest(&function_calls, reds); + fcalls = (int) erts_smp_atomic32_addtest(&function_calls, reds); ASSERT(esdp && esdp == erts_get_scheduler_data()); rq = erts_get_runq_current(esdp); @@ -5059,10 +5207,10 @@ Process *schedule(Process *p, int calls) ERTS_SMP_CHK_NO_PROC_LOCKS; - dt = do_time_read_and_reset(); + dt = erts_do_time_read_and_reset(); if (dt) { erts_smp_runq_unlock(rq); - bump_timer(dt); + erts_bump_timer(dt); erts_smp_runq_lock(rq); } BM_STOP_TIMER(system); @@ -5091,14 +5239,14 @@ Process *schedule(Process *p, int calls) | ERTS_RUNQ_FLG_CHK_CPU_BIND | ERTS_RUNQ_FLG_SUSPENDED)) { if ((rq->flags & ERTS_RUNQ_FLG_SUSPENDED) - || (erts_smp_atomic_read(&esdp->ssi->flags) + || (erts_smp_atomic32_read(&esdp->ssi->flags) & ERTS_SSI_FLG_SUSPENDED)) { - ASSERT(erts_smp_atomic_read(&esdp->ssi->flags) + ASSERT(erts_smp_atomic32_read(&esdp->ssi->flags) & ERTS_SSI_FLG_SUSPENDED); suspend_scheduler(esdp); } if ((rq->flags & ERTS_RUNQ_FLG_CHK_CPU_BIND) - || erts_smp_atomic_read(&esdp->chk_cpu_bind)) { + || erts_smp_atomic32_read(&esdp->chk_cpu_bind)) { erts_sched_check_cpu_bind(esdp); } } @@ -5107,7 +5255,7 @@ Process *schedule(Process *p, int calls) || defined(ERTS_SCHED_NEED_NONBLOCKABLE_AUX_WORK) { ErtsSchedulerSleepInfo *ssi = esdp->ssi; - long aux_work = erts_smp_atomic_read(&ssi->aux_work); + erts_aint32_t aux_work = erts_smp_atomic32_read(&ssi->aux_work); if (aux_work) { erts_smp_runq_unlock(rq); #ifdef ERTS_SCHED_NEED_BLOCKABLE_AUX_WORK @@ -5149,9 +5297,9 @@ Process *schedule(Process *p, int calls) if (rq->flags & (ERTS_RUNQ_FLG_SHARED_RUNQ | ERTS_RUNQ_FLG_SUSPENDED)) { if ((rq->flags & ERTS_RUNQ_FLG_SUSPENDED) - || (erts_smp_atomic_read(&esdp->ssi->flags) + || (erts_smp_atomic32_read(&esdp->ssi->flags) & ERTS_SSI_FLG_SUSPENDED)) { - ASSERT(erts_smp_atomic_read(&esdp->ssi->flags) + ASSERT(erts_smp_atomic32_read(&esdp->ssi->flags) & ERTS_SSI_FLG_SUSPENDED); non_empty_runq(rq); goto continue_check_activities_to_run; @@ -5193,7 +5341,7 @@ Process *schedule(Process *p, int calls) * Schedule system-level activities. */ - erts_smp_atomic_set(&function_calls, 0); + erts_smp_atomic32_set(&function_calls, 0); fcalls = 0; ASSERT(!erts_port_task_have_outstanding_io_tasks()); @@ -5203,11 +5351,11 @@ Process *schedule(Process *p, int calls) #endif erts_smp_runq_unlock(rq); erl_sys_schedule(runnable); - dt = do_time_read_and_reset(); - if (dt) bump_timer(dt); + dt = erts_do_time_read_and_reset(); + if (dt) erts_bump_timer(dt); #ifdef ERTS_SMP erts_smp_runq_lock(rq); - erts_smp_atomic_set(&doing_sys_schedule, 0); + erts_smp_atomic32_set(&doing_sys_schedule, 0); goto continue_check_activities_to_run; #else if (!runnable) @@ -5235,7 +5383,7 @@ Process *schedule(Process *p, int calls) if (erts_common_run_queue->waiting) wake_scheduler(erts_common_run_queue, 0, 1); } - else if (erts_smp_atomic_read(&no_empty_run_queues) != 0) { + else if (erts_smp_atomic32_read(&no_empty_run_queues) != 0) { wake_scheduler_on_empty_runq(rq); rq->wakeup_other = 0; } @@ -5692,7 +5840,7 @@ erts_test_next_pid(int set, Uint next) Uint erts_process_count(void) { - long res = erts_smp_atomic_read(&process_count); + erts_aint32_t res = erts_smp_atomic32_read(&process_count); ASSERT(res >= 0); return (Uint) res; } @@ -5741,7 +5889,7 @@ alloc_process(void) ASSERT(!process_tab[p_next]); process_tab[p_next] = p; - erts_smp_atomic_inc(&process_count); + erts_smp_atomic32_inc(&process_count); p->id = make_internal_pid(p_serial << p_serial_shift | p_next); if (p->id == ERTS_INVALID_PID) { /* Do not use the invalid pid; change serial */ @@ -5867,7 +6015,7 @@ erl_create_process(Process* parent, /* Parent of process (default group leader). p->min_heap_size = H_MIN_SIZE; p->min_vheap_size = BIN_VH_MIN_SIZE; p->prio = PRIORITY_NORMAL; - p->max_gen_gcs = (Uint16) erts_smp_atomic_read(&erts_max_gen_gcs); + p->max_gen_gcs = (Uint16) erts_smp_atomic32_read(&erts_max_gen_gcs); } p->skipped = 0; ASSERT(p->min_heap_size == erts_next_heap_size(p->min_heap_size, 0)); @@ -7324,8 +7472,8 @@ continue_exit_process(Process *p p->status_flags = 0; #endif process_tab[pix] = NULL; /* Time of death! */ - ASSERT(erts_smp_atomic_read(&process_count) > 0); - erts_smp_atomic_dec(&process_count); + ASSERT(erts_smp_atomic32_read(&process_count) > 0); + erts_smp_atomic32_dec(&process_count); #ifdef ERTS_SMP erts_pix_unlock(pix_lock); @@ -7465,7 +7613,7 @@ cancel_timer(Process* p) #ifdef ERTS_SMP erts_cancel_smp_ptimer(p->u.ptimer); #else - erl_cancel_timer(&p->u.tm); + erts_cancel_timer(&p->u.tm); #endif } @@ -7491,7 +7639,7 @@ set_timer(Process* p, Uint timeout) (ErlTimeoutProc) timeout_proc, timeout); #else - erl_set_timer(&p->u.tm, + erts_set_timer(&p->u.tm, (ErlTimeoutProc) timeout_proc, NULL, (void*) p, diff --git a/erts/emulator/beam/erl_process.h b/erts/emulator/beam/erl_process.h index c038e57b65..d927415f37 100644 --- a/erts/emulator/beam/erl_process.h +++ b/erts/emulator/beam/erl_process.h @@ -174,8 +174,8 @@ extern int erts_sched_thread_suggested_stack_size; #define ERTS_UNSET_RUNQ_FLG_EVACUATE(FLGS, PRIO) \ ((FLGS) &= ~ERTS_RUNQ_FLG_EVACUATE((PRIO))) -#define ERTS_RUNQ_IFLG_SUSPENDED (((long) 1) << 0) -#define ERTS_RUNQ_IFLG_NONEMPTY (((long) 1) << 1) +#define ERTS_RUNQ_IFLG_SUSPENDED (((erts_aint32_t) 1) << 0) +#define ERTS_RUNQ_IFLG_NONEMPTY (((erts_aint32_t) 1) << 1) #ifdef DEBUG @@ -219,11 +219,11 @@ typedef enum { ERTS_MIGRATE_FAILED_RUNQ_SUSPENDED } ErtsMigrateResult; -#define ERTS_SSI_FLG_SLEEPING (((long) 1) << 0) -#define ERTS_SSI_FLG_POLL_SLEEPING (((long) 1) << 1) -#define ERTS_SSI_FLG_TSE_SLEEPING (((long) 1) << 2) -#define ERTS_SSI_FLG_WAITING (((long) 1) << 3) -#define ERTS_SSI_FLG_SUSPENDED (((long) 1) << 4) +#define ERTS_SSI_FLG_SLEEPING (((erts_aint32_t) 1) << 0) +#define ERTS_SSI_FLG_POLL_SLEEPING (((erts_aint32_t) 1) << 1) +#define ERTS_SSI_FLG_TSE_SLEEPING (((erts_aint32_t) 1) << 2) +#define ERTS_SSI_FLG_WAITING (((erts_aint32_t) 1) << 3) +#define ERTS_SSI_FLG_SUSPENDED (((erts_aint32_t) 1) << 4) #define ERTS_SSI_FLGS_SLEEP_TYPE \ (ERTS_SSI_FLG_TSE_SLEEPING|ERTS_SSI_FLG_POLL_SLEEPING) @@ -236,16 +236,14 @@ typedef enum { | ERTS_SSI_FLG_WAITING \ | ERTS_SSI_FLG_SUSPENDED) - -#if !defined(ERTS_SCHED_NEED_BLOCKABLE_AUX_WORK) \ - && defined(ERTS_SMP_SCHEDULERS_NEED_TO_CHECK_CHILDREN) #define ERTS_SCHED_NEED_BLOCKABLE_AUX_WORK -#endif -#define ERTS_SSI_AUX_WORK_CHECK_CHILDREN (((long) 1) << 0) +#define ERTS_SSI_AUX_WORK_CHECK_CHILDREN (((erts_aint32_t) 1) << 0) +#define ERTS_SSI_AUX_WORK_MISC (((erts_aint32_t) 1) << 1) #define ERTS_SSI_BLOCKABLE_AUX_WORK_MASK \ - (ERTS_SSI_AUX_WORK_CHECK_CHILDREN) + (ERTS_SSI_AUX_WORK_CHECK_CHILDREN \ + | ERTS_SSI_AUX_WORK_MISC) #define ERTS_SSI_NONBLOCKABLE_AUX_WORK_MASK \ (0) @@ -259,9 +257,9 @@ typedef struct { struct ErtsSchedulerSleepInfo_ { ErtsSchedulerSleepInfo *next; ErtsSchedulerSleepInfo *prev; - erts_smp_atomic_t flags; + erts_smp_atomic32_t flags; erts_tse_t *event; - erts_smp_atomic_t aux_work; + erts_smp_atomic32_t aux_work; }; /* times to reschedule low prio process before running */ @@ -311,7 +309,7 @@ typedef struct { struct ErtsRunQueue_ { int ix; - erts_smp_atomic_t info_flags; + erts_smp_atomic32_t info_flags; erts_smp_mtx_t mtx; erts_smp_cnd_t cnd; @@ -421,7 +419,7 @@ struct ErtsSchedulerData_ { #ifdef ERTS_SMP /* NOTE: These fields are modified under held mutexes by other threads */ - erts_smp_atomic_t chk_cpu_bind; /* Only used when common run queue */ + erts_smp_atomic32_t chk_cpu_bind; /* Only used when common run queue */ #endif }; @@ -1034,6 +1032,7 @@ int erts_sched_set_wakeup_limit(char *str); #ifdef DEBUG void erts_dbg_multi_scheduling_return_trap(Process *, Eterm); #endif +int erts_get_max_no_executing_schedulers(void); #ifdef ERTS_SMP ErtsSchedSuspendResult erts_schedulers_state(Uint *, Uint *, Uint *, int); @@ -1048,6 +1047,11 @@ int erts_is_multi_scheduling_blocked(void); Eterm erts_multi_scheduling_blockers(Process *); void erts_start_schedulers(void); void erts_smp_notify_check_children_needed(void); +void +erts_smp_schedule_misc_aux_work(int ignore_self, + int max_sched, + void (*func)(void *), + void *arg); #endif void erts_sched_notify_check_cpu_bind(void); Uint erts_active_schedulers(void); @@ -1555,7 +1559,7 @@ extern int erts_disable_proc_not_running_opt; void erts_smp_notify_inc_runq(ErtsRunQueue *runq); #ifdef ERTS_SMP -void erts_sched_finish_poke(ErtsSchedulerSleepInfo *, long); +void erts_sched_finish_poke(ErtsSchedulerSleepInfo *, erts_aint32_t); ERTS_GLB_INLINE void erts_sched_poke(ErtsSchedulerSleepInfo *ssi); #if ERTS_GLB_INLINE_INCL_FUNC_DEF @@ -1563,11 +1567,11 @@ ERTS_GLB_INLINE void erts_sched_poke(ErtsSchedulerSleepInfo *ssi); ERTS_GLB_INLINE void erts_sched_poke(ErtsSchedulerSleepInfo *ssi) { - long flags = erts_smp_atomic_read(&ssi->flags); + erts_aint32_t flags = erts_smp_atomic32_read(&ssi->flags); ASSERT(!(flags & ERTS_SSI_FLG_SLEEPING) || (flags & ERTS_SSI_FLG_WAITING)); if (flags & ERTS_SSI_FLG_SLEEPING) { - flags = erts_smp_atomic_band(&ssi->flags, ~ERTS_SSI_FLGS_SLEEP); + flags = erts_smp_atomic32_band(&ssi->flags, ~ERTS_SSI_FLGS_SLEEP); erts_sched_finish_poke(ssi, flags); } } diff --git a/erts/emulator/beam/erl_process_lock.c b/erts/emulator/beam/erl_process_lock.c index 1bebcdb911..72560aa124 100644 --- a/erts/emulator/beam/erl_process_lock.c +++ b/erts/emulator/beam/erl_process_lock.c @@ -124,7 +124,7 @@ erts_init_proc_lock(int cpus) for (i = 0; i < ERTS_NO_OF_PIX_LOCKS; i++) { #ifdef ERTS_ENABLE_LOCK_COUNT erts_smp_spinlock_init_x(&erts_pix_locks[i].u.spnlck, - "pix_lock", make_small(i)); + "pix_lock", make_small(i)); #else erts_smp_spinlock_init(&erts_pix_locks[i].u.spnlck, "pix_lock"); #endif @@ -413,7 +413,7 @@ transfer_locks(Process *p, do { erts_tse_t *tmp = wake; wake = wake->next; - erts_atomic_set(&tmp->uaflgs, 0); + erts_atomic32_set(&tmp->uaflgs, 0); erts_tse_set(tmp); } while (wake); @@ -509,14 +509,14 @@ wait_for_locks(Process *p, ASSERT((wtr->uflgs & ~ERTS_PROC_LOCKS_ALL) == 0); - erts_atomic_set(&wtr->uaflgs, 1); + erts_atomic32_set(&wtr->uaflgs, 1); erts_pix_unlock(pix_lock); while (1) { int res; erts_tse_reset(wtr); - if (erts_atomic_read(&wtr->uaflgs) == 0) + if (erts_atomic32_read(&wtr->uaflgs) == 0) break; /* @@ -955,7 +955,7 @@ erts_proc_lock_init(Process *p) { /* We always start with all locks locked */ #if ERTS_PROC_LOCK_ATOMIC_IMPL - erts_smp_atomic_init(&p->lock.flags, (long) ERTS_PROC_LOCKS_ALL); + erts_smp_atomic32_init(&p->lock.flags, (erts_aint32_t) ERTS_PROC_LOCKS_ALL); #else p->lock.flags = ERTS_PROC_LOCKS_ALL; #endif @@ -974,7 +974,7 @@ erts_proc_lock_init(Process *p) { int i; for (i = 0; i <= ERTS_PROC_LOCK_MAX_BIT; i++) - erts_smp_atomic_init(&p->lock.locked[i], (long) 1); + erts_smp_atomic32_init(&p->lock.locked[i], (erts_aint32_t) 1); } #endif } diff --git a/erts/emulator/beam/erl_process_lock.h b/erts/emulator/beam/erl_process_lock.h index 4fe30c7209..355179f084 100644 --- a/erts/emulator/beam/erl_process_lock.h +++ b/erts/emulator/beam/erl_process_lock.h @@ -54,20 +54,20 @@ #define ERTS_PROC_LOCK_MAX_BIT 3 -typedef Uint32 ErtsProcLocks; +typedef erts_aint32_t ErtsProcLocks; typedef struct erts_proc_lock_queues_t_ erts_proc_lock_queues_t; typedef struct erts_proc_lock_t_ { #if ERTS_PROC_LOCK_ATOMIC_IMPL - erts_smp_atomic_t flags; + erts_smp_atomic32_t flags; #else ErtsProcLocks flags; #endif erts_proc_lock_queues_t *queues; - long refc; + Sint32 refc; #ifdef ERTS_PROC_LOCK_DEBUG - erts_smp_atomic_t locked[ERTS_PROC_LOCK_MAX_BIT+1]; + erts_smp_atomic32_t locked[ERTS_PROC_LOCK_MAX_BIT+1]; #endif #ifdef ERTS_ENABLE_LOCK_COUNT erts_lcnt_lock_t lcnt_main; @@ -270,17 +270,19 @@ typedef struct { #if ERTS_PROC_LOCK_ATOMIC_IMPL #define ERTS_PROC_LOCK_FLGS_BAND_(L, MSK) \ - ((ErtsProcLocks) erts_smp_atomic_band(&(L)->flags, (long) (MSK))) + ((ErtsProcLocks) erts_smp_atomic32_band(&(L)->flags, (erts_aint32_t) (MSK))) #define ERTS_PROC_LOCK_FLGS_BOR_(L, MSK) \ - ((ErtsProcLocks) erts_smp_atomic_bor(&(L)->flags, (long) (MSK))) + ((ErtsProcLocks) erts_smp_atomic32_bor(&(L)->flags, (erts_aint32_t) (MSK))) #define ERTS_PROC_LOCK_FLGS_CMPXCHG_ACQB_(L, NEW, EXPECTED) \ - ((ErtsProcLocks) erts_smp_atomic_cmpxchg_acqb(&(L)->flags, \ - (long) (NEW), (long) (EXPECTED))) + ((ErtsProcLocks) erts_smp_atomic32_cmpxchg_acqb(&(L)->flags, \ + (erts_aint32_t) (NEW), \ + (erts_aint32_t) (EXPECTED))) #define ERTS_PROC_LOCK_FLGS_CMPXCHG_RELB_(L, NEW, EXPECTED) \ - ((ErtsProcLocks) erts_smp_atomic_cmpxchg_relb(&(L)->flags, \ - (long) (NEW), (long) (EXPECTED))) + ((ErtsProcLocks) erts_smp_atomic32_cmpxchg_relb(&(L)->flags, \ + (erts_aint32_t) (NEW), \ + (erts_aint32_t) (EXPECTED))) #define ERTS_PROC_LOCK_FLGS_READ_(L) \ - ((ErtsProcLocks) erts_smp_atomic_read(&(L)->flags)) + ((ErtsProcLocks) erts_smp_atomic32_read(&(L)->flags)) #else /* no opt atomic ops */ @@ -619,13 +621,13 @@ erts_proc_lock_op_debug(Process *p, ErtsProcLocks locks, int locked) for (i = 0; i <= ERTS_PROC_LOCK_MAX_BIT; i++) { ErtsProcLocks lock = ((ErtsProcLocks) 1) << i; if (locks & lock) { - long lock_count; + erts_aint32_t lock_count; if (locked) { - lock_count = erts_smp_atomic_inctest(&p->lock.locked[i]); + lock_count = erts_smp_atomic32_inctest(&p->lock.locked[i]); ERTS_LC_ASSERT(lock_count == 1); } else { - lock_count = erts_smp_atomic_dectest(&p->lock.locked[i]); + lock_count = erts_smp_atomic32_dectest(&p->lock.locked[i]); ERTS_LC_ASSERT(lock_count == 0); } } diff --git a/erts/emulator/beam/erl_smp.h b/erts/emulator/beam/erl_smp.h index b41fa70476..287327bfe1 100644 --- a/erts/emulator/beam/erl_smp.h +++ b/erts/emulator/beam/erl_smp.h @@ -54,10 +54,10 @@ typedef erts_cnd_t erts_smp_cnd_t; typedef erts_rwmtx_opt_t erts_smp_rwmtx_opt_t; typedef erts_rwmtx_t erts_smp_rwmtx_t; typedef erts_tsd_key_t erts_smp_tsd_key_t; -typedef ethr_atomic_t erts_smp_atomic_t; +typedef erts_atomic_t erts_smp_atomic_t; +typedef erts_atomic32_t erts_smp_atomic32_t; typedef erts_spinlock_t erts_smp_spinlock_t; typedef erts_rwlock_t erts_smp_rwlock_t; -typedef erts_thr_timeval_t erts_smp_thr_timeval_t; void erts_thr_fatal_error(int, char *); /* implemented in erl_init.c */ #else /* #ifdef ERTS_SMP */ @@ -83,7 +83,8 @@ typedef struct { } erts_smp_rwmtx_opt_t; typedef int erts_smp_rwmtx_t; typedef int erts_smp_tsd_key_t; -typedef long erts_smp_atomic_t; +typedef SWord erts_smp_atomic_t; +typedef Uint32 erts_smp_atomic32_t; #if __GNUC__ > 2 typedef struct { } erts_smp_spinlock_t; typedef struct { } erts_smp_rwlock_t; @@ -92,11 +93,6 @@ typedef struct { int gcc_is_buggy; } erts_smp_spinlock_t; typedef struct { int gcc_is_buggy; } erts_smp_rwlock_t; #endif -typedef struct { - long tv_sec; - long tv_nsec; -} erts_smp_thr_timeval_t; - #endif /* #ifdef ERTS_SMP */ ERTS_GLB_INLINE void erts_smp_thr_init(erts_smp_thr_init_data_t *id); @@ -164,33 +160,82 @@ ERTS_GLB_INLINE int erts_smp_rwmtx_tryrwlock(erts_smp_rwmtx_t *rwmtx); ERTS_GLB_INLINE void erts_smp_rwmtx_rwunlock(erts_smp_rwmtx_t *rwmtx); ERTS_GLB_INLINE int erts_smp_lc_rwmtx_is_rlocked(erts_smp_rwmtx_t *mtx); ERTS_GLB_INLINE int erts_smp_lc_rwmtx_is_rwlocked(erts_smp_rwmtx_t *mtx); -ERTS_GLB_INLINE void erts_smp_atomic_init(erts_smp_atomic_t *var, long i); -ERTS_GLB_INLINE void erts_smp_atomic_set(erts_smp_atomic_t *var, long i); -ERTS_GLB_INLINE long erts_smp_atomic_read(erts_smp_atomic_t *var); -ERTS_GLB_INLINE long erts_smp_atomic_inctest(erts_smp_atomic_t *incp); -ERTS_GLB_INLINE long erts_smp_atomic_dectest(erts_smp_atomic_t *decp); +ERTS_GLB_INLINE void erts_smp_atomic_init(erts_smp_atomic_t *var, + erts_aint_t i); +ERTS_GLB_INLINE void erts_smp_atomic_set(erts_smp_atomic_t *var, erts_aint_t i); +ERTS_GLB_INLINE erts_aint_t erts_smp_atomic_read(erts_smp_atomic_t *var); +ERTS_GLB_INLINE erts_aint_t erts_smp_atomic_inctest(erts_smp_atomic_t *incp); +ERTS_GLB_INLINE erts_aint_t erts_smp_atomic_dectest(erts_smp_atomic_t *decp); ERTS_GLB_INLINE void erts_smp_atomic_inc(erts_smp_atomic_t *incp); ERTS_GLB_INLINE void erts_smp_atomic_dec(erts_smp_atomic_t *decp); -ERTS_GLB_INLINE long erts_smp_atomic_addtest(erts_smp_atomic_t *addp, - long i); -ERTS_GLB_INLINE void erts_smp_atomic_add(erts_smp_atomic_t *addp, long i); -ERTS_GLB_INLINE long erts_smp_atomic_xchg(erts_smp_atomic_t *xchgp, - long new); -ERTS_GLB_INLINE long erts_smp_atomic_cmpxchg(erts_smp_atomic_t *xchgp, - long new, - long expected); -ERTS_GLB_INLINE long erts_smp_atomic_bor(erts_smp_atomic_t *var, long mask); -ERTS_GLB_INLINE long erts_smp_atomic_band(erts_smp_atomic_t *var, long mask); -ERTS_GLB_INLINE long erts_smp_atomic_read_acqb(erts_smp_atomic_t *var); -ERTS_GLB_INLINE void erts_smp_atomic_set_relb(erts_smp_atomic_t *var, long i); +ERTS_GLB_INLINE erts_aint_t erts_smp_atomic_addtest(erts_smp_atomic_t *addp, + erts_aint_t i); +ERTS_GLB_INLINE void erts_smp_atomic_add(erts_smp_atomic_t *addp, + erts_aint_t i); +ERTS_GLB_INLINE erts_aint_t erts_smp_atomic_xchg(erts_smp_atomic_t *xchgp, + erts_aint_t new); +ERTS_GLB_INLINE erts_aint_t erts_smp_atomic_cmpxchg(erts_smp_atomic_t *xchgp, + erts_aint_t new, + erts_aint_t expected); +ERTS_GLB_INLINE erts_aint_t erts_smp_atomic_bor(erts_smp_atomic_t *var, + erts_aint_t mask); +ERTS_GLB_INLINE erts_aint_t erts_smp_atomic_band(erts_smp_atomic_t *var, + erts_aint_t mask); +ERTS_GLB_INLINE erts_aint_t erts_smp_atomic_read_acqb(erts_smp_atomic_t *var); +ERTS_GLB_INLINE void erts_smp_atomic_set_relb(erts_smp_atomic_t *var, + erts_aint_t i); ERTS_GLB_INLINE void erts_smp_atomic_dec_relb(erts_smp_atomic_t *decp); -ERTS_GLB_INLINE long erts_smp_atomic_dectest_relb(erts_smp_atomic_t *decp); -ERTS_GLB_INLINE long erts_smp_atomic_cmpxchg_acqb(erts_smp_atomic_t *xchgp, - long new, - long exp); -ERTS_GLB_INLINE long erts_smp_atomic_cmpxchg_relb(erts_smp_atomic_t *xchgp, - long new, - long exp); +ERTS_GLB_INLINE erts_aint_t erts_smp_atomic_dectest_relb(erts_smp_atomic_t *decp); +ERTS_GLB_INLINE erts_aint_t erts_smp_atomic_cmpxchg_acqb(erts_smp_atomic_t *xchgp, + erts_aint_t new, + erts_aint_t exp); +ERTS_GLB_INLINE erts_aint_t erts_smp_atomic_cmpxchg_relb(erts_smp_atomic_t *xchgp, + erts_aint_t new, + erts_aint_t exp); +ERTS_GLB_INLINE void +erts_smp_atomic32_init(erts_smp_atomic32_t *var, erts_aint32_t i); +ERTS_GLB_INLINE void +erts_smp_atomic32_set(erts_smp_atomic32_t *var, erts_aint32_t i); +ERTS_GLB_INLINE erts_aint32_t +erts_smp_atomic32_read(erts_smp_atomic32_t *var); +ERTS_GLB_INLINE erts_aint32_t +erts_smp_atomic32_inctest(erts_smp_atomic32_t *incp); +ERTS_GLB_INLINE erts_aint32_t +erts_smp_atomic32_dectest(erts_smp_atomic32_t *decp); +ERTS_GLB_INLINE void +erts_smp_atomic32_inc(erts_smp_atomic32_t *incp); +ERTS_GLB_INLINE void +erts_smp_atomic32_dec(erts_smp_atomic32_t *decp); +ERTS_GLB_INLINE erts_aint32_t +erts_smp_atomic32_addtest(erts_smp_atomic32_t *addp, erts_aint32_t i); +ERTS_GLB_INLINE void +erts_smp_atomic32_add(erts_smp_atomic32_t *addp, erts_aint32_t i); +ERTS_GLB_INLINE erts_aint32_t +erts_smp_atomic32_xchg(erts_smp_atomic32_t *xchgp, erts_aint32_t new); +ERTS_GLB_INLINE erts_aint32_t +erts_smp_atomic32_cmpxchg(erts_smp_atomic32_t *xchgp, + erts_aint32_t new, + erts_aint32_t expected); +ERTS_GLB_INLINE erts_aint32_t +erts_smp_atomic32_bor(erts_smp_atomic32_t *var, erts_aint32_t mask); +ERTS_GLB_INLINE erts_aint32_t +erts_smp_atomic32_band(erts_smp_atomic32_t *var, erts_aint32_t mask); +ERTS_GLB_INLINE erts_aint32_t +erts_smp_atomic32_read_acqb(erts_smp_atomic32_t *var); +ERTS_GLB_INLINE void +erts_smp_atomic32_set_relb(erts_smp_atomic32_t *var, erts_aint32_t i); +ERTS_GLB_INLINE void +erts_smp_atomic32_dec_relb(erts_smp_atomic32_t *decp); +ERTS_GLB_INLINE erts_aint32_t +erts_smp_atomic32_dectest_relb(erts_smp_atomic32_t *decp); +ERTS_GLB_INLINE erts_aint32_t +erts_smp_atomic32_cmpxchg_acqb(erts_smp_atomic32_t *xchgp, + erts_aint32_t new, + erts_aint32_t exp); +ERTS_GLB_INLINE erts_aint32_t +erts_smp_atomic32_cmpxchg_relb(erts_smp_atomic32_t *xchgp, + erts_aint32_t new, + erts_aint32_t exp); ERTS_GLB_INLINE void erts_smp_spinlock_init_x(erts_smp_spinlock_t *lock, char *name, Eterm extra); @@ -221,7 +266,6 @@ ERTS_GLB_INLINE void erts_smp_write_lock(erts_smp_rwlock_t *lock); ERTS_GLB_INLINE void erts_smp_write_unlock(erts_smp_rwlock_t *lock); ERTS_GLB_INLINE int erts_smp_lc_rwlock_is_rlocked(erts_smp_rwlock_t *lock); ERTS_GLB_INLINE int erts_smp_lc_rwlock_is_rwlocked(erts_smp_rwlock_t *lock); -ERTS_GLB_INLINE void erts_smp_thr_time_now(erts_smp_thr_timeval_t *time); ERTS_GLB_INLINE void erts_smp_tsd_key_create(erts_smp_tsd_key_t *keyp); ERTS_GLB_INLINE void erts_smp_tsd_key_delete(erts_smp_tsd_key_t key); ERTS_GLB_INLINE void erts_smp_tsd_set(erts_smp_tsd_key_t key, void *value); @@ -611,7 +655,7 @@ erts_smp_lc_rwmtx_is_rwlocked(erts_smp_rwmtx_t *mtx) } ERTS_GLB_INLINE void -erts_smp_atomic_init(erts_smp_atomic_t *var, long i) +erts_smp_atomic_init(erts_smp_atomic_t *var, erts_aint_t i) { #ifdef ERTS_SMP erts_atomic_init(var, i); @@ -621,7 +665,7 @@ erts_smp_atomic_init(erts_smp_atomic_t *var, long i) } ERTS_GLB_INLINE void -erts_smp_atomic_set(erts_smp_atomic_t *var, long i) +erts_smp_atomic_set(erts_smp_atomic_t *var, erts_aint_t i) { #ifdef ERTS_SMP erts_atomic_set(var, i); @@ -630,7 +674,7 @@ erts_smp_atomic_set(erts_smp_atomic_t *var, long i) #endif } -ERTS_GLB_INLINE long +ERTS_GLB_INLINE erts_aint_t erts_smp_atomic_read(erts_smp_atomic_t *var) { #ifdef ERTS_SMP @@ -640,7 +684,7 @@ erts_smp_atomic_read(erts_smp_atomic_t *var) #endif } -ERTS_GLB_INLINE long +ERTS_GLB_INLINE erts_aint_t erts_smp_atomic_inctest(erts_smp_atomic_t *incp) { #ifdef ERTS_SMP @@ -650,7 +694,7 @@ erts_smp_atomic_inctest(erts_smp_atomic_t *incp) #endif } -ERTS_GLB_INLINE long +ERTS_GLB_INLINE erts_aint_t erts_smp_atomic_dectest(erts_smp_atomic_t *decp) { #ifdef ERTS_SMP @@ -680,8 +724,8 @@ erts_smp_atomic_dec(erts_smp_atomic_t *decp) #endif } -ERTS_GLB_INLINE long -erts_smp_atomic_addtest(erts_smp_atomic_t *addp, long i) +ERTS_GLB_INLINE erts_aint_t +erts_smp_atomic_addtest(erts_smp_atomic_t *addp, erts_aint_t i) { #ifdef ERTS_SMP return erts_atomic_addtest(addp, i); @@ -691,7 +735,7 @@ erts_smp_atomic_addtest(erts_smp_atomic_t *addp, long i) } ERTS_GLB_INLINE void -erts_smp_atomic_add(erts_smp_atomic_t *addp, long i) +erts_smp_atomic_add(erts_smp_atomic_t *addp, erts_aint_t i) { #ifdef ERTS_SMP erts_atomic_add(addp, i); @@ -700,59 +744,61 @@ erts_smp_atomic_add(erts_smp_atomic_t *addp, long i) #endif } -ERTS_GLB_INLINE long -erts_smp_atomic_xchg(erts_smp_atomic_t *xchgp, long new) +ERTS_GLB_INLINE erts_aint_t +erts_smp_atomic_xchg(erts_smp_atomic_t *xchgp, erts_aint_t new) { #ifdef ERTS_SMP return erts_atomic_xchg(xchgp, new); #else - long old; + erts_aint_t old; old = *xchgp; *xchgp = new; return old; #endif } -ERTS_GLB_INLINE long -erts_smp_atomic_cmpxchg(erts_smp_atomic_t *xchgp, long new, long expected) +ERTS_GLB_INLINE erts_aint_t +erts_smp_atomic_cmpxchg(erts_smp_atomic_t *xchgp, + erts_aint_t new, + erts_aint_t expected) { #ifdef ERTS_SMP return erts_atomic_cmpxchg(xchgp, new, expected); #else - long old = *xchgp; + erts_aint_t old = *xchgp; if (old == expected) *xchgp = new; return old; #endif } -ERTS_GLB_INLINE long -erts_smp_atomic_bor(erts_smp_atomic_t *var, long mask) +ERTS_GLB_INLINE erts_aint_t +erts_smp_atomic_bor(erts_smp_atomic_t *var, erts_aint_t mask) { #ifdef ERTS_SMP return erts_atomic_bor(var, mask); #else - long old; + erts_aint_t old; old = *var; *var |= mask; return old; #endif } -ERTS_GLB_INLINE long -erts_smp_atomic_band(erts_smp_atomic_t *var, long mask) +ERTS_GLB_INLINE erts_aint_t +erts_smp_atomic_band(erts_smp_atomic_t *var, erts_aint_t mask) { #ifdef ERTS_SMP return erts_atomic_band(var, mask); #else - long old; + erts_aint_t old; old = *var; *var &= mask; return old; #endif } -ERTS_GLB_INLINE long +ERTS_GLB_INLINE erts_aint_t erts_smp_atomic_read_acqb(erts_smp_atomic_t *var) { #ifdef ERTS_SMP @@ -763,7 +809,7 @@ erts_smp_atomic_read_acqb(erts_smp_atomic_t *var) } ERTS_GLB_INLINE void -erts_smp_atomic_set_relb(erts_smp_atomic_t *var, long i) +erts_smp_atomic_set_relb(erts_smp_atomic_t *var, erts_aint_t i) { #ifdef ERTS_SMP erts_atomic_set_relb(var, i); @@ -772,7 +818,8 @@ erts_smp_atomic_set_relb(erts_smp_atomic_t *var, long i) #endif } -ERTS_GLB_INLINE void erts_smp_atomic_dec_relb(erts_smp_atomic_t *decp) +ERTS_GLB_INLINE void +erts_smp_atomic_dec_relb(erts_smp_atomic_t *decp) { #ifdef ERTS_SMP erts_atomic_dec_relb(decp); @@ -781,7 +828,7 @@ ERTS_GLB_INLINE void erts_smp_atomic_dec_relb(erts_smp_atomic_t *decp) #endif } -ERTS_GLB_INLINE long +ERTS_GLB_INLINE erts_aint_t erts_smp_atomic_dectest_relb(erts_smp_atomic_t *decp) { #ifdef ERTS_SMP @@ -791,28 +838,244 @@ erts_smp_atomic_dectest_relb(erts_smp_atomic_t *decp) #endif } -ERTS_GLB_INLINE long erts_smp_atomic_cmpxchg_acqb(erts_smp_atomic_t *xchgp, - long new, - long exp) +ERTS_GLB_INLINE erts_aint_t +erts_smp_atomic_cmpxchg_acqb(erts_smp_atomic_t *xchgp, + erts_aint_t new, + erts_aint_t exp) { #ifdef ERTS_SMP return erts_atomic_cmpxchg_acqb(xchgp, new, exp); #else - long old = *xchgp; + erts_aint_t old = *xchgp; if (old == exp) *xchgp = new; return old; #endif } -ERTS_GLB_INLINE long erts_smp_atomic_cmpxchg_relb(erts_smp_atomic_t *xchgp, - long new, - long exp) +ERTS_GLB_INLINE erts_aint_t +erts_smp_atomic_cmpxchg_relb(erts_smp_atomic_t *xchgp, + erts_aint_t new, + erts_aint_t exp) { #ifdef ERTS_SMP return erts_atomic_cmpxchg_relb(xchgp, new, exp); #else - long old = *xchgp; + erts_aint_t old = *xchgp; + if (old == exp) + *xchgp = new; + return old; +#endif +} + +ERTS_GLB_INLINE void +erts_smp_atomic32_init(erts_smp_atomic32_t *var, erts_aint32_t i) +{ +#ifdef ERTS_SMP + erts_atomic32_init(var, i); +#else + *var = i; +#endif +} + +ERTS_GLB_INLINE void +erts_smp_atomic32_set(erts_smp_atomic32_t *var, erts_aint32_t i) +{ +#ifdef ERTS_SMP + erts_atomic32_set(var, i); +#else + *var = i; +#endif +} + +ERTS_GLB_INLINE erts_aint32_t +erts_smp_atomic32_read(erts_smp_atomic32_t *var) +{ +#ifdef ERTS_SMP + return erts_atomic32_read(var); +#else + return *var; +#endif +} + +ERTS_GLB_INLINE erts_aint32_t +erts_smp_atomic32_inctest(erts_smp_atomic32_t *incp) +{ +#ifdef ERTS_SMP + return erts_atomic32_inctest(incp); +#else + return ++(*incp); +#endif +} + +ERTS_GLB_INLINE erts_aint32_t +erts_smp_atomic32_dectest(erts_smp_atomic32_t *decp) +{ +#ifdef ERTS_SMP + return erts_atomic32_dectest(decp); +#else + return --(*decp); +#endif +} + +ERTS_GLB_INLINE void +erts_smp_atomic32_inc(erts_smp_atomic32_t *incp) +{ +#ifdef ERTS_SMP + erts_atomic32_inc(incp); +#else + ++(*incp); +#endif +} + +ERTS_GLB_INLINE void +erts_smp_atomic32_dec(erts_smp_atomic32_t *decp) +{ +#ifdef ERTS_SMP + erts_atomic32_dec(decp); +#else + --(*decp); +#endif +} + +ERTS_GLB_INLINE erts_aint32_t +erts_smp_atomic32_addtest(erts_smp_atomic32_t *addp, erts_aint32_t i) +{ +#ifdef ERTS_SMP + return erts_atomic32_addtest(addp, i); +#else + return *addp += i; +#endif +} + +ERTS_GLB_INLINE void +erts_smp_atomic32_add(erts_smp_atomic32_t *addp, erts_aint32_t i) +{ +#ifdef ERTS_SMP + erts_atomic32_add(addp, i); +#else + *addp += i; +#endif +} + +ERTS_GLB_INLINE erts_aint32_t +erts_smp_atomic32_xchg(erts_smp_atomic32_t *xchgp, erts_aint32_t new) +{ +#ifdef ERTS_SMP + return erts_atomic32_xchg(xchgp, new); +#else + erts_aint32_t old; + old = *xchgp; + *xchgp = new; + return old; +#endif +} + +ERTS_GLB_INLINE erts_aint32_t +erts_smp_atomic32_cmpxchg(erts_smp_atomic32_t *xchgp, + erts_aint32_t new, + erts_aint32_t expected) +{ +#ifdef ERTS_SMP + return erts_atomic32_cmpxchg(xchgp, new, expected); +#else + erts_aint32_t old = *xchgp; + if (old == expected) + *xchgp = new; + return old; +#endif +} + +ERTS_GLB_INLINE erts_aint32_t +erts_smp_atomic32_bor(erts_smp_atomic32_t *var, erts_aint32_t mask) +{ +#ifdef ERTS_SMP + return erts_atomic32_bor(var, mask); +#else + erts_aint32_t old; + old = *var; + *var |= mask; + return old; +#endif +} + +ERTS_GLB_INLINE erts_aint32_t +erts_smp_atomic32_band(erts_smp_atomic32_t *var, erts_aint32_t mask) +{ +#ifdef ERTS_SMP + return erts_atomic32_band(var, mask); +#else + erts_aint32_t old; + old = *var; + *var &= mask; + return old; +#endif +} + +ERTS_GLB_INLINE erts_aint32_t +erts_smp_atomic32_read_acqb(erts_smp_atomic32_t *var) +{ +#ifdef ERTS_SMP + return erts_atomic32_read_acqb(var); +#else + return *var; +#endif +} + +ERTS_GLB_INLINE void +erts_smp_atomic32_set_relb(erts_smp_atomic32_t *var, erts_aint32_t i) +{ +#ifdef ERTS_SMP + erts_atomic32_set_relb(var, i); +#else + *var = i; +#endif +} + +ERTS_GLB_INLINE void +erts_smp_atomic32_dec_relb(erts_smp_atomic32_t *decp) +{ +#ifdef ERTS_SMP + erts_atomic32_dec_relb(decp); +#else + --(*decp); +#endif +} + +ERTS_GLB_INLINE erts_aint32_t +erts_smp_atomic32_dectest_relb(erts_smp_atomic32_t *decp) +{ +#ifdef ERTS_SMP + return erts_atomic32_dectest_relb(decp); +#else + return --(*decp); +#endif +} + +ERTS_GLB_INLINE erts_aint32_t +erts_smp_atomic32_cmpxchg_acqb(erts_smp_atomic32_t *xchgp, + erts_aint32_t new, + erts_aint32_t exp) +{ +#ifdef ERTS_SMP + return erts_atomic32_cmpxchg_acqb(xchgp, new, exp); +#else + erts_aint32_t old = *xchgp; + if (old == exp) + *xchgp = new; + return old; +#endif +} + +ERTS_GLB_INLINE erts_aint32_t +erts_smp_atomic32_cmpxchg_relb(erts_smp_atomic32_t *xchgp, + erts_aint32_t new, + erts_aint32_t exp) +{ +#ifdef ERTS_SMP + return erts_atomic32_cmpxchg_relb(xchgp, new, exp); +#else + erts_aint32_t old = *xchgp; if (old == exp) *xchgp = new; return old; @@ -988,14 +1251,6 @@ erts_smp_lc_rwlock_is_rwlocked(erts_smp_rwlock_t *lock) } ERTS_GLB_INLINE void -erts_smp_thr_time_now(erts_smp_thr_timeval_t *time) -{ -#ifdef ERTS_SMP - erts_thr_time_now(time); -#endif -} - -ERTS_GLB_INLINE void erts_smp_tsd_key_create(erts_smp_tsd_key_t *keyp) { #ifdef ERTS_SMP diff --git a/erts/emulator/beam/erl_term.h b/erts/emulator/beam/erl_term.h index b8e4473141..815cc1beae 100644 --- a/erts/emulator/beam/erl_term.h +++ b/erts/emulator/beam/erl_term.h @@ -193,7 +193,7 @@ struct erl_node_; /* Declared in erl_node_tables.h */ #endif #define _is_aligned(x) (((Uint)(x) & 0x3) == 0) #define _unchecked_make_boxed(x) ((Uint) COMPRESS_POINTER(x) + TAG_PRIMARY_BOXED) -_ET_DECLARE_CHECKED(Eterm,make_boxed,Eterm*); +_ET_DECLARE_CHECKED(Eterm,make_boxed,Eterm*) #define make_boxed(x) _ET_APPLY(make_boxed,(x)) #if 1 #define _is_not_boxed(x) ((x) & (_TAG_PRIMARY_MASK-TAG_PRIMARY_BOXED)) @@ -204,12 +204,12 @@ _ET_DECLARE_CHECKED(int,is_boxed,Eterm) #define is_boxed(x) (((x) & _TAG_PRIMARY_MASK) == TAG_PRIMARY_BOXED) #endif #define _unchecked_boxed_val(x) ((Eterm*) EXPAND_POINTER(((x) - TAG_PRIMARY_BOXED))) -_ET_DECLARE_CHECKED(Eterm*,boxed_val,Eterm); +_ET_DECLARE_CHECKED(Eterm*,boxed_val,Eterm) #define boxed_val(x) _ET_APPLY(boxed_val,(x)) /* cons cell ("list") access methods */ #define _unchecked_make_list(x) ((Uint) COMPRESS_POINTER(x) + TAG_PRIMARY_LIST) -_ET_DECLARE_CHECKED(Eterm,make_list,Eterm*); +_ET_DECLARE_CHECKED(Eterm,make_list,Eterm*) #define make_list(x) _ET_APPLY(make_list,(x)) #if 1 #define _unchecked_is_not_list(x) ((x) & (_TAG_PRIMARY_MASK-TAG_PRIMARY_LIST)) @@ -226,7 +226,7 @@ _ET_DECLARE_CHECKED(int,is_not_list,Eterm) #define _list_precond(x) (is_list(x)) #endif #define _unchecked_list_val(x) ((Eterm*) EXPAND_POINTER((x) - TAG_PRIMARY_LIST)) -_ET_DECLARE_CHECKED(Eterm*,list_val,Eterm); +_ET_DECLARE_CHECKED(Eterm*,list_val,Eterm) #define list_val(x) _ET_APPLY(list_val,(x)) #define CONS(hp, car, cdr) \ @@ -995,14 +995,14 @@ _ET_DECLARE_CHECKED(struct erl_node_*,external_ref_node,Eterm) #endif #define _unchecked_make_cp(x) ((Eterm) COMPRESS_POINTER(x)) -_ET_DECLARE_CHECKED(Eterm,make_cp,BeamInstr*); +_ET_DECLARE_CHECKED(Eterm,make_cp,BeamInstr*) #define make_cp(x) _ET_APPLY(make_cp,(x)) #define is_not_CP(x) ((x) & _CPMASK) #define is_CP(x) (!is_not_CP(x)) #define _unchecked_cp_val(x) ((BeamInstr*) EXPAND_POINTER(x)) -_ET_DECLARE_CHECKED(BeamInstr*,cp_val,Eterm); +_ET_DECLARE_CHECKED(BeamInstr*,cp_val,Eterm) #define cp_val(x) _ET_APPLY(cp_val,(x)) #define make_catch(x) (((x) << _TAG_IMMED2_SIZE) | _TAG_IMMED2_CATCH) diff --git a/erts/emulator/beam/erl_threads.h b/erts/emulator/beam/erl_threads.h index a74cf79b8c..84a20b51f2 100644 --- a/erts/emulator/beam/erl_threads.h +++ b/erts/emulator/beam/erl_threads.h @@ -89,7 +89,10 @@ typedef ethr_rwmutex_opt erts_rwmtx_opt_t; typedef ethr_tsd_key erts_tsd_key_t; typedef ethr_ts_event erts_tse_t; +typedef ethr_sint_t erts_aint_t; typedef ethr_atomic_t erts_atomic_t; +typedef ethr_sint32_t erts_aint32_t; +typedef ethr_atomic32_t erts_atomic32_t; /* spinlock */ typedef struct { @@ -113,7 +116,6 @@ typedef struct { #endif } erts_rwlock_t; -typedef ethr_timeval erts_thr_timeval_t; __decl_noreturn void __noreturn erts_thr_fatal_error(int, char *); /* implemented in erl_init.c */ @@ -152,7 +154,10 @@ typedef struct { typedef int erts_rwmtx_t; typedef int erts_tsd_key_t; typedef int erts_tse_t; -typedef long erts_atomic_t; +typedef SWord erts_aint_t; +typedef SWord erts_atomic_t; +typedef SWord erts_aint32_t; +typedef SWord erts_atomic32_t; #if __GNUC__ > 2 typedef struct { } erts_spinlock_t; typedef struct { } erts_rwlock_t; @@ -160,10 +165,6 @@ typedef struct { } erts_rwlock_t; typedef struct { int gcc_is_buggy; } erts_spinlock_t; typedef struct { int gcc_is_buggy; } erts_rwlock_t; #endif -typedef struct { - long tv_sec; - long tv_nsec; -} erts_thr_timeval_t; #define ERTS_MTX_INITER 0 #define ERTS_CND_INITER 0 @@ -173,6 +174,8 @@ typedef struct { #endif /* #ifdef USE_THREADS */ +#define ERTS_AINT_T_MAX (~(((erts_aint_t) 1) << (sizeof(erts_aint_t)*8-1))) + ERTS_GLB_INLINE void erts_thr_init(erts_thr_init_data_t *id); ERTS_GLB_INLINE void erts_thr_late_init(erts_thr_late_init_data_t *id); ERTS_GLB_INLINE void erts_thr_create(erts_tid_t *tid, void * (*func)(void *), @@ -231,33 +234,65 @@ ERTS_GLB_INLINE int erts_rwmtx_tryrwlock(erts_rwmtx_t *rwmtx); ERTS_GLB_INLINE void erts_rwmtx_rwunlock(erts_rwmtx_t *rwmtx); ERTS_GLB_INLINE int erts_lc_rwmtx_is_rlocked(erts_rwmtx_t *mtx); ERTS_GLB_INLINE int erts_lc_rwmtx_is_rwlocked(erts_rwmtx_t *mtx); -ERTS_GLB_INLINE void erts_atomic_init(erts_atomic_t *var, long i); -ERTS_GLB_INLINE void erts_atomic_set(erts_atomic_t *var, long i); -ERTS_GLB_INLINE long erts_atomic_read(erts_atomic_t *var); -ERTS_GLB_INLINE long erts_atomic_inctest(erts_atomic_t *incp); -ERTS_GLB_INLINE long erts_atomic_dectest(erts_atomic_t *decp); +ERTS_GLB_INLINE void erts_atomic_init(erts_atomic_t *var, erts_aint_t i); +ERTS_GLB_INLINE void erts_atomic_set(erts_atomic_t *var, erts_aint_t i); +ERTS_GLB_INLINE erts_aint_t erts_atomic_read(erts_atomic_t *var); +ERTS_GLB_INLINE erts_aint_t erts_atomic_inctest(erts_atomic_t *incp); +ERTS_GLB_INLINE erts_aint_t erts_atomic_dectest(erts_atomic_t *decp); ERTS_GLB_INLINE void erts_atomic_inc(erts_atomic_t *incp); ERTS_GLB_INLINE void erts_atomic_dec(erts_atomic_t *decp); -ERTS_GLB_INLINE long erts_atomic_addtest(erts_atomic_t *addp, - long i); -ERTS_GLB_INLINE void erts_atomic_add(erts_atomic_t *addp, long i); -ERTS_GLB_INLINE long erts_atomic_xchg(erts_atomic_t *xchgp, - long new); -ERTS_GLB_INLINE long erts_atomic_cmpxchg(erts_atomic_t *xchgp, - long new, - long expected); -ERTS_GLB_INLINE long erts_atomic_bor(erts_atomic_t *var, long mask); -ERTS_GLB_INLINE long erts_atomic_band(erts_atomic_t *var, long mask); -ERTS_GLB_INLINE long erts_atomic_read_acqb(erts_atomic_t *var); -ERTS_GLB_INLINE void erts_atomic_set_relb(erts_atomic_t *var, long i); +ERTS_GLB_INLINE erts_aint_t erts_atomic_addtest(erts_atomic_t *addp, + erts_aint_t i); +ERTS_GLB_INLINE void erts_atomic_add(erts_atomic_t *addp, erts_aint_t i); +ERTS_GLB_INLINE erts_aint_t erts_atomic_xchg(erts_atomic_t *xchgp, + erts_aint_t new); +ERTS_GLB_INLINE erts_aint_t erts_atomic_cmpxchg(erts_atomic_t *xchgp, + erts_aint_t new, + erts_aint_t expected); +ERTS_GLB_INLINE erts_aint_t erts_atomic_bor(erts_atomic_t *var, + erts_aint_t mask); +ERTS_GLB_INLINE erts_aint_t erts_atomic_band(erts_atomic_t *var, + erts_aint_t mask); +ERTS_GLB_INLINE erts_aint_t erts_atomic_read_acqb(erts_atomic_t *var); +ERTS_GLB_INLINE void erts_atomic_set_relb(erts_atomic_t *var, erts_aint_t i); ERTS_GLB_INLINE void erts_atomic_dec_relb(erts_atomic_t *decp); -ERTS_GLB_INLINE long erts_atomic_dectest_relb(erts_atomic_t *decp); -ERTS_GLB_INLINE long erts_atomic_cmpxchg_acqb(erts_atomic_t *xchgp, - long new, - long exp); -ERTS_GLB_INLINE long erts_atomic_cmpxchg_relb(erts_atomic_t *xchgp, - long new, - long exp); +ERTS_GLB_INLINE erts_aint_t erts_atomic_dectest_relb(erts_atomic_t *decp); +ERTS_GLB_INLINE erts_aint_t erts_atomic_cmpxchg_acqb(erts_atomic_t *xchgp, + erts_aint_t new, + erts_aint_t exp); +ERTS_GLB_INLINE erts_aint_t erts_atomic_cmpxchg_relb(erts_atomic_t *xchgp, + erts_aint_t new, + erts_aint_t exp); +ERTS_GLB_INLINE void erts_atomic32_init(erts_atomic32_t *var, erts_aint32_t i); +ERTS_GLB_INLINE void erts_atomic32_set(erts_atomic32_t *var, erts_aint32_t i); +ERTS_GLB_INLINE erts_aint32_t erts_atomic32_read(erts_atomic32_t *var); +ERTS_GLB_INLINE erts_aint32_t erts_atomic32_inctest(erts_atomic32_t *incp); +ERTS_GLB_INLINE erts_aint32_t erts_atomic32_dectest(erts_atomic32_t *decp); +ERTS_GLB_INLINE void erts_atomic32_inc(erts_atomic32_t *incp); +ERTS_GLB_INLINE void erts_atomic32_dec(erts_atomic32_t *decp); +ERTS_GLB_INLINE erts_aint32_t erts_atomic32_addtest(erts_atomic32_t *addp, + erts_aint32_t i); +ERTS_GLB_INLINE void erts_atomic32_add(erts_atomic32_t *addp, erts_aint32_t i); +ERTS_GLB_INLINE erts_aint32_t erts_atomic32_xchg(erts_atomic32_t *xchgp, + erts_aint32_t new); +ERTS_GLB_INLINE erts_aint32_t erts_atomic32_cmpxchg(erts_atomic32_t *xchgp, + erts_aint32_t new, + erts_aint32_t expected); +ERTS_GLB_INLINE erts_aint32_t erts_atomic32_bor(erts_atomic32_t *var, + erts_aint32_t mask); +ERTS_GLB_INLINE erts_aint32_t erts_atomic32_band(erts_atomic32_t *var, + erts_aint32_t mask); +ERTS_GLB_INLINE erts_aint32_t erts_atomic32_read_acqb(erts_atomic32_t *var); +ERTS_GLB_INLINE void erts_atomic32_set_relb(erts_atomic32_t *var, + erts_aint32_t i); +ERTS_GLB_INLINE void erts_atomic32_dec_relb(erts_atomic32_t *decp); +ERTS_GLB_INLINE erts_aint32_t erts_atomic32_dectest_relb(erts_atomic32_t *decp); +ERTS_GLB_INLINE erts_aint32_t erts_atomic32_cmpxchg_acqb(erts_atomic32_t *xchgp, + erts_aint32_t new, + erts_aint32_t exp); +ERTS_GLB_INLINE erts_aint32_t erts_atomic32_cmpxchg_relb(erts_atomic32_t *xchgp, + erts_aint32_t new, + erts_aint32_t exp); ERTS_GLB_INLINE void erts_spinlock_init_x_opt(erts_spinlock_t *lock, char *name, Eterm extra, @@ -292,7 +327,6 @@ ERTS_GLB_INLINE void erts_write_lock(erts_rwlock_t *lock); ERTS_GLB_INLINE void erts_write_unlock(erts_rwlock_t *lock); ERTS_GLB_INLINE int erts_lc_rwlock_is_rlocked(erts_rwlock_t *lock); ERTS_GLB_INLINE int erts_lc_rwlock_is_rwlocked(erts_rwlock_t *lock); -ERTS_GLB_INLINE void erts_thr_time_now(erts_thr_timeval_t *time); ERTS_GLB_INLINE void erts_tsd_key_create(erts_tsd_key_t *keyp); ERTS_GLB_INLINE void erts_tsd_key_delete(erts_tsd_key_t key); ERTS_GLB_INLINE void erts_tsd_set(erts_tsd_key_t key, void *value); @@ -925,7 +959,7 @@ erts_lc_rwmtx_is_rwlocked(erts_rwmtx_t *mtx) } ERTS_GLB_INLINE void -erts_atomic_init(erts_atomic_t *var, long i) +erts_atomic_init(erts_atomic_t *var, erts_aint_t i) { #ifdef USE_THREADS ethr_atomic_init(var, i); @@ -935,7 +969,7 @@ erts_atomic_init(erts_atomic_t *var, long i) } ERTS_GLB_INLINE void -erts_atomic_set(erts_atomic_t *var, long i) +erts_atomic_set(erts_atomic_t *var, erts_aint_t i) { #ifdef USE_THREADS ethr_atomic_set(var, i); @@ -944,7 +978,7 @@ erts_atomic_set(erts_atomic_t *var, long i) #endif } -ERTS_GLB_INLINE long +ERTS_GLB_INLINE erts_aint_t erts_atomic_read(erts_atomic_t *var) { #ifdef USE_THREADS @@ -954,7 +988,7 @@ erts_atomic_read(erts_atomic_t *var) #endif } -ERTS_GLB_INLINE long +ERTS_GLB_INLINE erts_aint_t erts_atomic_inctest(erts_atomic_t *incp) { #ifdef USE_THREADS @@ -964,7 +998,7 @@ erts_atomic_inctest(erts_atomic_t *incp) #endif } -ERTS_GLB_INLINE long +ERTS_GLB_INLINE erts_aint_t erts_atomic_dectest(erts_atomic_t *decp) { #ifdef USE_THREADS @@ -994,8 +1028,8 @@ erts_atomic_dec(erts_atomic_t *decp) #endif } -ERTS_GLB_INLINE long -erts_atomic_addtest(erts_atomic_t *addp, long i) +ERTS_GLB_INLINE erts_aint_t +erts_atomic_addtest(erts_atomic_t *addp, erts_aint_t i) { #ifdef USE_THREADS return ethr_atomic_add_read(addp, i); @@ -1005,7 +1039,7 @@ erts_atomic_addtest(erts_atomic_t *addp, long i) } ERTS_GLB_INLINE void -erts_atomic_add(erts_atomic_t *addp, long i) +erts_atomic_add(erts_atomic_t *addp, erts_aint_t i) { #ifdef USE_THREADS ethr_atomic_add(addp, i); @@ -1014,59 +1048,58 @@ erts_atomic_add(erts_atomic_t *addp, long i) #endif } -ERTS_GLB_INLINE long -erts_atomic_xchg(erts_atomic_t *xchgp, long new) +ERTS_GLB_INLINE erts_aint_t +erts_atomic_xchg(erts_atomic_t *xchgp, erts_aint_t new) { - long old; #ifdef USE_THREADS return ethr_atomic_xchg(xchgp, new); #else - old = *xchgp; + erts_aint_t old = *xchgp; *xchgp = new; -#endif return old; +#endif } -ERTS_GLB_INLINE long -erts_atomic_cmpxchg(erts_atomic_t *xchgp, long new, long expected) +ERTS_GLB_INLINE erts_aint_t +erts_atomic_cmpxchg(erts_atomic_t *xchgp, erts_aint_t new, erts_aint_t expected) { #ifdef USE_THREADS return ethr_atomic_cmpxchg(xchgp, new, expected); #else - long old = *xchgp; + erts_aint_t old = *xchgp; if (old == expected) *xchgp = new; return old; #endif } -ERTS_GLB_INLINE long -erts_atomic_bor(erts_atomic_t *var, long mask) +ERTS_GLB_INLINE erts_aint_t +erts_atomic_bor(erts_atomic_t *var, erts_aint_t mask) { #ifdef USE_THREADS return ethr_atomic_read_bor(var, mask); #else - long old; + erts_aint_t old; old = *var; *var |= mask; return old; #endif } -ERTS_GLB_INLINE long -erts_atomic_band(erts_atomic_t *var, long mask) +ERTS_GLB_INLINE erts_aint_t +erts_atomic_band(erts_atomic_t *var, erts_aint_t mask) { #ifdef USE_THREADS return ethr_atomic_read_band(var, mask); #else - long old; + erts_aint_t old; old = *var; *var &= mask; return old; #endif } -ERTS_GLB_INLINE long +ERTS_GLB_INLINE erts_aint_t erts_atomic_read_acqb(erts_atomic_t *var) { #ifdef USE_THREADS @@ -1077,7 +1110,7 @@ erts_atomic_read_acqb(erts_atomic_t *var) } ERTS_GLB_INLINE void -erts_atomic_set_relb(erts_atomic_t *var, long i) +erts_atomic_set_relb(erts_atomic_t *var, erts_aint_t i) { #ifdef USE_THREADS ethr_atomic_set_relb(var, i); @@ -1096,7 +1129,7 @@ erts_atomic_dec_relb(erts_atomic_t *decp) #endif } -ERTS_GLB_INLINE long +ERTS_GLB_INLINE erts_aint_t erts_atomic_dectest_relb(erts_atomic_t *decp) { #ifdef USE_THREADS @@ -1106,28 +1139,243 @@ erts_atomic_dectest_relb(erts_atomic_t *decp) #endif } -ERTS_GLB_INLINE long erts_atomic_cmpxchg_acqb(erts_atomic_t *xchgp, - long new, - long exp) +ERTS_GLB_INLINE erts_aint_t erts_atomic_cmpxchg_acqb(erts_atomic_t *xchgp, + erts_aint_t new, + erts_aint_t exp) { #ifdef USE_THREADS return ethr_atomic_cmpxchg_acqb(xchgp, new, exp); #else - long old = *xchgp; + erts_aint_t old = *xchgp; if (old == exp) *xchgp = new; return old; #endif } -ERTS_GLB_INLINE long erts_atomic_cmpxchg_relb(erts_atomic_t *xchgp, - long new, - long exp) +ERTS_GLB_INLINE erts_aint_t erts_atomic_cmpxchg_relb(erts_atomic_t *xchgp, + erts_aint_t new, + erts_aint_t exp) { #ifdef USE_THREADS return ethr_atomic_cmpxchg_relb(xchgp, new, exp); #else - long old = *xchgp; + erts_aint_t old = *xchgp; + if (old == exp) + *xchgp = new; + return old; +#endif +} + +/* atomic32 */ + +ERTS_GLB_INLINE void +erts_atomic32_init(erts_atomic32_t *var, erts_aint32_t i) +{ +#ifdef USE_THREADS + ethr_atomic32_init(var, i); +#else + *var = i; +#endif +} + +ERTS_GLB_INLINE void +erts_atomic32_set(erts_atomic32_t *var, erts_aint32_t i) +{ +#ifdef USE_THREADS + ethr_atomic32_set(var, i); +#else + *var = i; +#endif +} + +ERTS_GLB_INLINE erts_aint32_t +erts_atomic32_read(erts_atomic32_t *var) +{ +#ifdef USE_THREADS + return ethr_atomic32_read(var); +#else + return *var; +#endif +} + +ERTS_GLB_INLINE erts_aint32_t +erts_atomic32_inctest(erts_atomic32_t *incp) +{ +#ifdef USE_THREADS + return ethr_atomic32_inc_read(incp); +#else + return ++(*incp); +#endif +} + +ERTS_GLB_INLINE erts_aint32_t +erts_atomic32_dectest(erts_atomic32_t *decp) +{ +#ifdef USE_THREADS + return ethr_atomic32_dec_read(decp); +#else + return --(*decp); +#endif +} + +ERTS_GLB_INLINE void +erts_atomic32_inc(erts_atomic32_t *incp) +{ +#ifdef USE_THREADS + ethr_atomic32_inc(incp); +#else + ++(*incp); +#endif +} + +ERTS_GLB_INLINE void +erts_atomic32_dec(erts_atomic32_t *decp) +{ +#ifdef USE_THREADS + ethr_atomic32_dec(decp); +#else + --(*decp); +#endif +} + +ERTS_GLB_INLINE erts_aint32_t +erts_atomic32_addtest(erts_atomic32_t *addp, erts_aint32_t i) +{ +#ifdef USE_THREADS + return ethr_atomic32_add_read(addp, i); +#else + return *addp += i; +#endif +} + +ERTS_GLB_INLINE void +erts_atomic32_add(erts_atomic32_t *addp, erts_aint32_t i) +{ +#ifdef USE_THREADS + ethr_atomic32_add(addp, i); +#else + *addp += i; +#endif +} + +ERTS_GLB_INLINE erts_aint32_t +erts_atomic32_xchg(erts_atomic32_t *xchgp, erts_aint32_t new) +{ +#ifdef USE_THREADS + return ethr_atomic32_xchg(xchgp, new); +#else + erts_aint32_t old = *xchgp; + *xchgp = new; + return old; +#endif +} + +ERTS_GLB_INLINE erts_aint32_t +erts_atomic32_cmpxchg(erts_atomic32_t *xchgp, + erts_aint32_t new, + erts_aint32_t expected) +{ +#ifdef USE_THREADS + return ethr_atomic32_cmpxchg(xchgp, new, expected); +#else + erts_aint32_t old = *xchgp; + if (old == expected) + *xchgp = new; + return old; +#endif +} + +ERTS_GLB_INLINE erts_aint32_t +erts_atomic32_bor(erts_atomic32_t *var, erts_aint32_t mask) +{ +#ifdef USE_THREADS + return ethr_atomic32_read_bor(var, mask); +#else + erts_aint32_t old; + old = *var; + *var |= mask; + return old; +#endif +} + +ERTS_GLB_INLINE erts_aint32_t +erts_atomic32_band(erts_atomic32_t *var, erts_aint32_t mask) +{ +#ifdef USE_THREADS + return ethr_atomic32_read_band(var, mask); +#else + erts_aint32_t old; + old = *var; + *var &= mask; + return old; +#endif +} + +ERTS_GLB_INLINE erts_aint32_t +erts_atomic32_read_acqb(erts_atomic32_t *var) +{ +#ifdef USE_THREADS + return ethr_atomic32_read_acqb(var); +#else + return *var; +#endif +} + +ERTS_GLB_INLINE void +erts_atomic32_set_relb(erts_atomic32_t *var, erts_aint32_t i) +{ +#ifdef USE_THREADS + ethr_atomic32_set_relb(var, i); +#else + *var = i; +#endif +} + +ERTS_GLB_INLINE void +erts_atomic32_dec_relb(erts_atomic32_t *decp) +{ +#ifdef USE_THREADS + ethr_atomic32_dec_relb(decp); +#else + --(*decp); +#endif +} + +ERTS_GLB_INLINE erts_aint32_t +erts_atomic32_dectest_relb(erts_atomic32_t *decp) +{ +#ifdef USE_THREADS + return ethr_atomic32_dec_read_relb(decp); +#else + return --(*decp); +#endif +} + +ERTS_GLB_INLINE erts_aint32_t +erts_atomic32_cmpxchg_acqb(erts_atomic32_t *xchgp, + erts_aint32_t new, + erts_aint32_t exp) +{ +#ifdef USE_THREADS + return ethr_atomic32_cmpxchg_acqb(xchgp, new, exp); +#else + erts_aint32_t old = *xchgp; + if (old == exp) + *xchgp = new; + return old; +#endif +} + +ERTS_GLB_INLINE erts_aint32_t +erts_atomic32_cmpxchg_relb(erts_atomic32_t *xchgp, + erts_aint32_t new, + erts_aint32_t exp) +{ +#ifdef USE_THREADS + return ethr_atomic32_cmpxchg_relb(xchgp, new, exp); +#else + erts_aint32_t old = *xchgp; if (old == exp) *xchgp = new; return old; @@ -1428,16 +1676,6 @@ erts_lc_rwlock_is_rwlocked(erts_rwlock_t *lock) } ERTS_GLB_INLINE void -erts_thr_time_now(erts_thr_timeval_t *time) -{ -#ifdef USE_THREADS - int res = ethr_time_now(time); - if (res) - erts_thr_fatal_error(res, "get current time"); -#endif -} - -ERTS_GLB_INLINE void erts_tsd_key_create(erts_tsd_key_t *keyp) { #ifdef USE_THREADS diff --git a/erts/emulator/beam/erl_time.h b/erts/emulator/beam/erl_time.h index 6f6b971d34..93d8ea4cb4 100644 --- a/erts/emulator/beam/erl_time.h +++ b/erts/emulator/beam/erl_time.h @@ -20,11 +20,15 @@ #ifndef ERL_TIME_H__ #define ERL_TIME_H__ +extern erts_smp_atomic_t do_time; /* set at clock interrupt */ +extern SysTimeval erts_first_emu_time; + /* ** Timer entry: */ typedef struct erl_timer { struct erl_timer* next; /* next entry tiw slot or chain */ + struct erl_timer* prev; /* prev entry tiw slot or chain */ Uint slot; /* slot in timer wheel */ Uint count; /* number of loops remaining */ int active; /* 1=activated, 0=deactivated */ @@ -39,7 +43,6 @@ typedef void (*ErlTimeoutProc)(void*); typedef void (*ErlCancelProc)(void*); #ifdef ERTS_SMP - /* * Process and port timer */ @@ -61,7 +64,66 @@ void erts_create_smp_ptimer(ErtsSmpPTimer **timer_ref, ErlTimeoutProc timeout_func, Uint timeout); void erts_cancel_smp_ptimer(ErtsSmpPTimer *ptimer); +#endif + +/* timer-wheel api */ +void erts_init_time(void); +void erts_set_timer(ErlTimer*, ErlTimeoutProc, ErlCancelProc, void*, Uint); +void erts_cancel_timer(ErlTimer*); +void erts_bump_timer(erts_aint_t); +Uint erts_timer_wheel_memory_size(void); +Uint erts_time_left(ErlTimer *); +erts_aint_t erts_next_time(void); + +#ifdef DEBUG +void erts_p_slpq(void); #endif +ERTS_GLB_INLINE erts_aint_t erts_do_time_read_and_reset(void); +ERTS_GLB_INLINE void erts_do_time_add(long); + +#if ERTS_GLB_INLINE_INCL_FUNC_DEF + +ERTS_GLB_INLINE erts_aint_t erts_do_time_read_and_reset(void) { return erts_smp_atomic_xchg(&do_time, 0L); } +ERTS_GLB_INLINE void erts_do_time_add(long elapsed) { erts_smp_atomic_add(&do_time, elapsed); } + +#endif /* #if ERTS_GLB_INLINE_INCL_FUNC_DEF */ + + +/* time_sup */ + +#if (defined(HAVE_GETHRVTIME) || defined(HAVE_CLOCK_GETTIME)) +# ifndef HAVE_ERTS_NOW_CPU +# define HAVE_ERTS_NOW_CPU +# ifdef HAVE_GETHRVTIME +# define erts_start_now_cpu() sys_start_hrvtime() +# define erts_stop_now_cpu() sys_stop_hrvtime() +# endif +# endif +void erts_get_now_cpu(Uint* megasec, Uint* sec, Uint* microsec); #endif + +void erts_get_timeval(SysTimeval *tv); +long erts_get_time(void); +void erts_get_emu_time(SysTimeval *); + +ERTS_GLB_INLINE int erts_cmp_timeval(SysTimeval *t1p, SysTimeval *t2p); + +#if ERTS_GLB_INLINE_INCL_FUNC_DEF + +ERTS_GLB_INLINE int +erts_cmp_timeval(SysTimeval *t1p, SysTimeval *t2p) +{ + if (t1p->tv_sec == t2p->tv_sec) { + if (t1p->tv_usec < t2p->tv_usec) + return -1; + else if (t1p->tv_usec > t2p->tv_usec) + return 1; + return 0; + } + return t1p->tv_sec < t2p->tv_sec ? -1 : 1; +} + +#endif /* #if ERTS_GLB_INLINE_INCL_FUNC_DEF */ +#endif /* ERL_TIME_H__ */ diff --git a/erts/emulator/beam/erl_time_sup.c b/erts/emulator/beam/erl_time_sup.c index 7b8706ea13..ca4b54188e 100644 --- a/erts/emulator/beam/erl_time_sup.c +++ b/erts/emulator/beam/erl_time_sup.c @@ -358,10 +358,6 @@ static int clock_resolution; ** instead of something like select. */ -#if defined(ERTS_TIMER_THREAD) -static ERTS_INLINE void init_erts_deliver_time(const SysTimeval *inittv) { } -static ERTS_INLINE void do_erts_deliver_time(const SysTimeval *current) { } -#else static SysTimeval last_delivered; static void init_erts_deliver_time(const SysTimeval *inittv) @@ -389,11 +385,10 @@ static void do_erts_deliver_time(const SysTimeval *current) this by simply pretend as if the time stood still. :) */ if (elapsed > 0) { - do_time_add(elapsed); + erts_do_time_add(elapsed); last_delivered = cur_time; } } -#endif int erts_init_time_sup(void) @@ -786,7 +781,6 @@ get_sys_now(Uint* megasec, Uint* sec, Uint* microsec) to a struct timeval representing current time (to save a gettimeofday() where possible) or NULL */ -#if !defined(ERTS_TIMER_THREAD) void erts_deliver_time(void) { SysTimeval now; @@ -797,7 +791,6 @@ void erts_deliver_time(void) { erts_smp_mtx_unlock(&erts_timeofday_mtx); } -#endif /* get *real* time (not ticks) remaining until next timeout - if there isn't one, give a "long" time, that is guaranteed @@ -806,14 +799,12 @@ void erts_deliver_time(void) { void erts_time_remaining(SysTimeval *rem_time) { int ticks; -#if !defined(ERTS_TIMER_THREAD) SysTimeval cur_time; -#endif long elapsed; - /* next_time() returns no of ticks to next timeout or -1 if none */ + /* erts_next_time() returns no of ticks to next timeout or -1 if none */ - if ((ticks = next_time()) == -1) { + if ((ticks = erts_next_time()) == -1) { /* timer queue empty */ /* this will cause at most 100000000 ticks */ rem_time->tv_sec = 100000; @@ -822,9 +813,6 @@ void erts_time_remaining(SysTimeval *rem_time) /* next timeout after ticks ticks */ ticks *= CLOCK_RESOLUTION; -#if defined(ERTS_TIMER_THREAD) - elapsed = 0; -#else erts_smp_mtx_lock(&erts_timeofday_mtx); get_tolerant_timeofday(&cur_time); @@ -839,7 +827,6 @@ void erts_time_remaining(SysTimeval *rem_time) rem_time->tv_sec = rem_time->tv_usec = 0; return; } -#endif rem_time->tv_sec = (ticks - elapsed) / 1000; rem_time->tv_usec = 1000 * ((ticks - elapsed) % 1000); } diff --git a/erts/emulator/beam/erl_unicode.c b/erts/emulator/beam/erl_unicode.c index 72207df621..545b345a71 100644 --- a/erts/emulator/beam/erl_unicode.c +++ b/erts/emulator/beam/erl_unicode.c @@ -2420,7 +2420,7 @@ void erts_copy_utf8_to_utf16_little(byte *target, byte *bytes, int num_chars) /* * This internal bif converts a filename to whatever format is suitable for the file driver - * It also adds zero termination so that prim_file neednt bother with the character encoding + * It also adds zero termination so that prim_file needn't bother with the character encoding * of the file driver */ BIF_RETTYPE prim_file_internal_name2native_1(BIF_ALIST_1) diff --git a/erts/emulator/beam/erl_vm.h b/erts/emulator/beam/erl_vm.h index 48fa99934e..5b42a2ce11 100644 --- a/erts/emulator/beam/erl_vm.h +++ b/erts/emulator/beam/erl_vm.h @@ -1,7 +1,7 @@ /* * %CopyrightBegin% * - * Copyright Ericsson AB 1996-2010. All Rights Reserved. + * Copyright Ericsson AB 1996-2011. All Rights Reserved. * * The contents of this file are subject to the Erlang Public License, * Version 1.1, (the "License"); you may not use this file except in @@ -47,7 +47,7 @@ #define SEQ_TRACE 1 #define CONTEXT_REDS 2000 /* Swap process out after this number */ -#define MAX_ARG 256 /* Max number of arguments allowed */ +#define MAX_ARG 255 /* Max number of arguments allowed */ #define MAX_REG 1024 /* Max number of x(N) registers used */ /* Scheduler stores data for temporary heaps if diff --git a/erts/emulator/beam/global.h b/erts/emulator/beam/global.h index 524db2a2eb..e8a9d5f32f 100644 --- a/erts/emulator/beam/global.h +++ b/erts/emulator/beam/global.h @@ -544,7 +544,7 @@ ERTS_GLB_INLINE void erts_may_save_closed_port(Port *prt) if (prt->snapshot != erts_smp_atomic_read(&erts_ports_snapshot)) { /* Dead ports are added from the end of the snapshot buffer */ Eterm* tombstone = (Eterm*) erts_smp_atomic_addtest(&erts_dead_ports_ptr, - -(long)sizeof(Eterm)); + -(erts_aint_t)sizeof(Eterm)); ASSERT(tombstone+1 != NULL); ASSERT(prt->snapshot == (Uint32) erts_smp_atomic_read(&erts_ports_snapshot) - 1); *tombstone = prt->id; @@ -563,7 +563,7 @@ extern Uint display_items; /* no of items to display in traces etc */ extern Uint display_loads; /* print info about loaded modules */ extern int erts_backtrace_depth; -extern erts_smp_atomic_t erts_max_gen_gcs; +extern erts_smp_atomic32_t erts_max_gen_gcs; extern int erts_disable_tolerant_timeofday; @@ -1206,7 +1206,7 @@ ERTS_GLB_INLINE void erts_smp_port_unlock(Port *prt) { #ifdef ERTS_SMP - long refc; + erts_aint_t refc; erts_smp_mtx_unlock(prt->lock); refc = erts_smp_atomic_dectest(&prt->refc); ASSERT(refc >= 0); @@ -1425,84 +1425,6 @@ void erl_drv_thr_init(void); /* time.c */ -ERTS_GLB_INLINE long do_time_read_and_reset(void); -#ifdef ERTS_TIMER_THREAD -ERTS_GLB_INLINE int next_time(void); -ERTS_GLB_INLINE void bump_timer(long); -#else -int next_time(void); -void bump_timer(long); -extern erts_smp_atomic_t do_time; /* set at clock interrupt */ -ERTS_GLB_INLINE void do_time_add(long); -#endif - -#if ERTS_GLB_INLINE_INCL_FUNC_DEF - -#ifdef ERTS_TIMER_THREAD -ERTS_GLB_INLINE long do_time_read_and_reset(void) { return 0; } -ERTS_GLB_INLINE int next_time(void) { return -1; } -ERTS_GLB_INLINE void bump_timer(long ignore) { } -#else -ERTS_GLB_INLINE long do_time_read_and_reset(void) -{ - return erts_smp_atomic_xchg(&do_time, 0L); -} -ERTS_GLB_INLINE void do_time_add(long elapsed) -{ - erts_smp_atomic_add(&do_time, elapsed); -} -#endif - -#endif /* #if ERTS_GLB_INLINE_INCL_FUNC_DEF */ - -void init_time(void); -void erl_set_timer(ErlTimer*, ErlTimeoutProc, ErlCancelProc, void*, Uint); -void erl_cancel_timer(ErlTimer*); -Uint time_left(ErlTimer *); - -Uint erts_timer_wheel_memory_size(void); - -#if (defined(HAVE_GETHRVTIME) || defined(HAVE_CLOCK_GETTIME)) -# ifndef HAVE_ERTS_NOW_CPU -# define HAVE_ERTS_NOW_CPU -# ifdef HAVE_GETHRVTIME -# define erts_start_now_cpu() sys_start_hrvtime() -# define erts_stop_now_cpu() sys_stop_hrvtime() -# endif -# endif -void erts_get_now_cpu(Uint* megasec, Uint* sec, Uint* microsec); -#endif - -void erts_get_timeval(SysTimeval *tv); -long erts_get_time(void); - -extern SysTimeval erts_first_emu_time; - -void erts_get_emu_time(SysTimeval *); - -ERTS_GLB_INLINE int erts_cmp_timeval(SysTimeval *t1p, SysTimeval *t2p); - -#if ERTS_GLB_INLINE_INCL_FUNC_DEF - -ERTS_GLB_INLINE int -erts_cmp_timeval(SysTimeval *t1p, SysTimeval *t2p) -{ - if (t1p->tv_sec == t2p->tv_sec) { - if (t1p->tv_usec < t2p->tv_usec) - return -1; - else if (t1p->tv_usec > t2p->tv_usec) - return 1; - return 0; - } - return t1p->tv_sec < t2p->tv_sec ? -1 : 1; -} - -#endif - -#ifdef DEBUG -void p_slpq(void); -#endif - /* utils.c */ /* diff --git a/erts/emulator/beam/io.c b/erts/emulator/beam/io.c index 9ed92bbe03..f21a96c754 100644 --- a/erts/emulator/beam/io.c +++ b/erts/emulator/beam/io.c @@ -428,7 +428,7 @@ setup_port(Port* prt, Eterm pid, erts_driver_t *driver, old_name = prt->name; prt->name = new_name; #ifdef ERTS_SMP - erts_smp_atomic_set(&prt->run_queue, (long) runq); + erts_smp_atomic_set(&prt->run_queue, (erts_aint_t) runq); #endif ASSERT(!prt->drv_ptr); prt->drv_ptr = driver; @@ -670,7 +670,7 @@ erts_open_driver(erts_driver_t* driver, /* Pointer to driver. */ #ifdef ERTS_SMP erts_cancel_smp_ptimer(port->ptimer); #else - erl_cancel_timer(&(port->tm)); + erts_cancel_timer(&(port->tm)); #endif stopq(port); kill_port(port); @@ -1297,7 +1297,7 @@ void init_io(void) erts_port[i].port_data_lock = NULL; } - erts_smp_atomic_init(&erts_ports_snapshot, (long) 0); + erts_smp_atomic_init(&erts_ports_snapshot, (erts_aint_t) 0); last_port_num = 0; erts_smp_spinlock_init(&get_free_port_lck, "get_free_port"); @@ -1839,7 +1839,7 @@ terminate_port(Port *prt) #ifdef ERTS_SMP erts_cancel_smp_ptimer(prt->ptimer); #else - erl_cancel_timer(&prt->tm); + erts_cancel_timer(&prt->tm); #endif drv = prt->drv_ptr; @@ -3252,7 +3252,7 @@ int driver_output_binary(ErlDrvPort ix, char* hbuf, int hlen, return 0; prt->bytes_in += (hlen + len); - erts_smp_atomic_add(&erts_bytes_in, (long) (hlen + len)); + erts_smp_atomic_add(&erts_bytes_in, (erts_aint_t) (hlen + len)); if (prt->status & ERTS_PORT_SFLG_DISTRIBUTION) { return erts_net_message(prt, prt->dist_entry, @@ -3287,7 +3287,7 @@ int driver_output2(ErlDrvPort ix, char* hbuf, int hlen, char* buf, int len) return 0; prt->bytes_in += (hlen + len); - erts_smp_atomic_add(&erts_bytes_in, (long) (hlen + len)); + erts_smp_atomic_add(&erts_bytes_in, (erts_aint_t) (hlen + len)); if (prt->status & ERTS_PORT_SFLG_DISTRIBUTION) { if (len == 0) return erts_net_message(prt, @@ -3364,7 +3364,7 @@ int driver_outputv(ErlDrvPort ix, char* hbuf, int hlen, ErlIOVec* vec, int skip) /* XXX handle distribution !!! */ prt->bytes_in += (hlen + size); - erts_smp_atomic_add(&erts_bytes_in, (long) (hlen + size)); + erts_smp_atomic_add(&erts_bytes_in, (erts_aint_t) (hlen + size)); deliver_vec_message(prt, prt->connected, hbuf, hlen, binv, iov, n, size); return 0; } @@ -3408,25 +3408,25 @@ int len; * reference count on driver binaries... */ -long +ErlDrvSInt driver_binary_get_refc(ErlDrvBinary *dbp) { Binary* bp = ErlDrvBinary2Binary(dbp); - return erts_refc_read(&bp->refc, 1); + return (ErlDrvSInt) erts_refc_read(&bp->refc, 1); } -long +ErlDrvSInt driver_binary_inc_refc(ErlDrvBinary *dbp) { Binary* bp = ErlDrvBinary2Binary(dbp); - return erts_refc_inctest(&bp->refc, 2); + return (ErlDrvSInt) erts_refc_inctest(&bp->refc, 2); } -long +ErlDrvSInt driver_binary_dec_refc(ErlDrvBinary *dbp) { Binary* bp = ErlDrvBinary2Binary(dbp); - return erts_refc_dectest(&bp->refc, 1); + return (ErlDrvSInt) erts_refc_dectest(&bp->refc, 1); } @@ -3541,12 +3541,12 @@ pdl_init_refc(ErlDrvPDL pdl) erts_atomic_init(&pdl->refc, 1); } -static ERTS_INLINE long +static ERTS_INLINE ErlDrvSInt pdl_read_refc(ErlDrvPDL pdl) { - long refc = erts_atomic_read(&pdl->refc); + erts_aint_t refc = erts_atomic_read(&pdl->refc); ERTS_LC_ASSERT(refc >= 0); - return refc; + return (ErlDrvSInt) refc; } static ERTS_INLINE void @@ -3556,12 +3556,12 @@ pdl_inc_refc(ErlDrvPDL pdl) ERTS_LC_ASSERT(driver_pdl_get_refc(pdl) > 1); } -static ERTS_INLINE long +static ERTS_INLINE ErlDrvSInt pdl_inctest_refc(ErlDrvPDL pdl) { - long refc = erts_atomic_inctest(&pdl->refc); + erts_aint_t refc = erts_atomic_inctest(&pdl->refc); ERTS_LC_ASSERT(refc > 1); - return refc; + return (ErlDrvSInt) refc; } #if 0 /* unused */ @@ -3573,12 +3573,12 @@ pdl_dec_refc(ErlDrvPDL pdl) } #endif -static ERTS_INLINE long +static ERTS_INLINE ErlDrvSInt pdl_dectest_refc(ErlDrvPDL pdl) { - long refc = erts_atomic_dectest(&pdl->refc); + erts_aint_t refc = erts_atomic_dectest(&pdl->refc); ERTS_LC_ASSERT(refc >= 0); - return refc; + return (ErlDrvSInt) refc; } static ERTS_INLINE void pdl_destroy(ErlDrvPDL pdl) @@ -3649,7 +3649,7 @@ driver_pdl_lock(ErlDrvPDL pdl) void driver_pdl_unlock(ErlDrvPDL pdl) { - long refc; + ErlDrvSInt refc; #ifdef HARDDEBUG erts_fprintf(stderr, "driver_pdl_unlock(0x%08X)\r\n",(unsigned) pdl); #endif @@ -3659,28 +3659,30 @@ driver_pdl_unlock(ErlDrvPDL pdl) pdl_destroy(pdl); } -long +ErlDrvSInt driver_pdl_get_refc(ErlDrvPDL pdl) { return pdl_read_refc(pdl); } -long +ErlDrvSInt driver_pdl_inc_refc(ErlDrvPDL pdl) { - long refc = pdl_inctest_refc(pdl); + ErlDrvSInt refc = pdl_inctest_refc(pdl); #ifdef HARDDEBUG - erts_fprintf(stderr, "driver_pdl_inc_refc(0x%08X) -> %ld\r\n",(unsigned) pdl, refc); + erts_fprintf(stderr, "driver_pdl_inc_refc(%p) -> %bpd\r\n", + pdl, refc); #endif return refc; } -long +ErlDrvSInt driver_pdl_dec_refc(ErlDrvPDL pdl) { - long refc = pdl_dectest_refc(pdl); + ErlDrvSInt refc = pdl_dectest_refc(pdl); #ifdef HARDDEBUG - erts_fprintf(stderr, "driver_pdl_dec_refc(0x%08X) -> %ld\r\n",(unsigned) pdl, refc); + erts_fprintf(stderr, "driver_pdl_dec_refc(%p) -> %bpd\r\n", + pdl, refc); #endif if (!refc) pdl_destroy(pdl); @@ -4066,7 +4068,7 @@ drv_cancel_timer(Port *prt) #ifdef ERTS_SMP erts_cancel_smp_ptimer(prt->ptimer); #else - erl_cancel_timer(&prt->tm); + erts_cancel_timer(&prt->tm); #endif if (erts_port_task_is_scheduled(&prt->timeout_task)) erts_port_task_abort(prt->id, &prt->timeout_task); @@ -4090,7 +4092,7 @@ int driver_set_timer(ErlDrvPort ix, UWord t) (ErlTimeoutProc) schedule_port_timeout, t); #else - erl_set_timer(&prt->tm, + erts_set_timer(&prt->tm, (ErlTimeoutProc) schedule_port_timeout, NULL, prt, @@ -4121,9 +4123,9 @@ driver_read_timer(ErlDrvPort ix, unsigned long* t) return -1; ERTS_SMP_LC_ASSERT(erts_lc_is_port_locked(prt)); #ifdef ERTS_SMP - *t = prt->ptimer ? time_left(&prt->ptimer->timer.tm) : 0; + *t = prt->ptimer ? erts_time_left(&prt->ptimer->timer.tm) : 0; #else - *t = time_left(&prt->tm); + *t = erts_time_left(&prt->tm); #endif return 0; } diff --git a/erts/emulator/beam/ops.tab b/erts/emulator/beam/ops.tab index a2439d5582..e861f97e7a 100644 --- a/erts/emulator/beam/ops.tab +++ b/erts/emulator/beam/ops.tab @@ -101,16 +101,16 @@ return %macro: test_heap TestHeap -pack allocate t t -allocate_heap I I I +allocate_heap t I t deallocate I init y allocate_zero t t -allocate_heap_zero I I I +allocate_heap_zero t I t trim N Remaining => i_trim N i_trim I -test_heap I I +test_heap I t allocate_heap S u==0 R => allocate S R allocate_heap_zero S u==0 R => allocate_zero S R @@ -124,7 +124,7 @@ init Y1 | init Y2 => init2 Y1 Y2 # Selecting values -select_val S=q Fail=f Size=u Rest=* => const_select_val(S, Fail, Size, Rest) +select_val S=aiq Fail=f Size=u Rest=* => const_select_val(S, Fail, Size, Rest) select_val S=s Fail=f Size=u Rest=* | use_jump_tab(Size, Rest) => \ gen_jump_tab(S, Fail, Size, Rest) @@ -132,34 +132,59 @@ select_val S=s Fail=f Size=u Rest=* | use_jump_tab(Size, Rest) => \ is_integer Fail=f S | select_val S=s Fail=f Size=u Rest=* | use_jump_tab(Size, Rest) => \ gen_jump_tab(S, Fail, Size, Rest) +is_integer TypeFail=f S | select_val S=s Fail=f Size=u Rest=* | \ + mixed_types(Size, Rest) => \ + gen_split_values(S, TypeFail, Fail, Size, Rest) + select_val S=s Fail=f Size=u Rest=* | mixed_types(Size, Rest) => \ - gen_split_values(S, Fail, Size, Rest) + gen_split_values(S, Fail, Fail, Size, Rest) -is_integer Fail=f S | select_val S=s Fail=f Size=u Rest=* | \ +is_integer Fail=f S | select_val S=d Fail=f Size=u Rest=* | \ fixed_size_values(Size, Rest) => gen_select_val(S, Fail, Size, Rest) -is_atom Fail=f S | select_val S=s Fail=f Size=u Rest=* | \ +is_atom Fail=f S | select_val S=d Fail=f Size=u Rest=* | \ fixed_size_values(Size, Rest) => gen_select_val(S, Fail, Size, Rest) -select_val S=s Fail=f Size=u Rest=* | fixed_size_values(Size, Rest) => \ - gen_select_val(S, Fail, Size, Rest) +select_val S=s Fail=f Size=u Rest=* | floats_or_bignums(Size, Rest) => \ + gen_select_literals(S, Fail, Size, Rest) -select_val S=s Fail=f Size=u Rest=* | all_values_are_big(Size, Rest) => \ - gen_select_big(S, Fail, Size, Rest) +select_val S=d Fail=f Size=u Rest=* | fixed_size_values(Size, Rest) => \ + gen_select_val(S, Fail, Size, Rest) -is_tuple Fail=f S | select_tuple_arity S=s Fail=f Size=u Rest=* => \ +is_tuple Fail=f S | select_tuple_arity S=d Fail=f Size=u Rest=* => \ gen_select_tuple_arity(S, Fail, Size, Rest) -select_tuple_arity S=s Fail=f Size=u Rest=* => \ +select_tuple_arity S=d Fail=f Size=u Rest=* => \ gen_select_tuple_arity(S, Fail, Size, Rest) -i_select_val s f I -i_select_tuple_arity s f I -i_select_big s f -i_select_float s f I +i_select_val r f I +i_select_val x f I +i_select_val y f I + +i_select_val2 r f c f c f +i_select_val2 x f c f c f +i_select_val2 y f c f c f + +i_select_tuple_arity2 r f A f A f +i_select_tuple_arity2 x f A f A f +i_select_tuple_arity2 y f A f A f + +i_select_tuple_arity r f I +i_select_tuple_arity x f I +i_select_tuple_arity y f I + +i_jump_on_val_zero r f I +i_jump_on_val_zero x f I +i_jump_on_val_zero y f I + +i_jump_on_val r f I I +i_jump_on_val x f I I +i_jump_on_val y f I I -i_jump_on_val_zero s f I -i_jump_on_val s f I I +jump Target | label Lbl | same_label(Target, Lbl) => label Lbl + +is_ne_exact L1 S1 S2 | jump Fail | label L2 | same_label(L1, L2) => \ + is_eq_exact Fail S1 S2 | label L2 %macro: get_list GetList -pack get_list x x x @@ -234,11 +259,17 @@ is_number Fail Literal=q => move Literal x | is_number Fail x jump f -case_end Literal=q => move Literal x | case_end x -badmatch Literal=q => move Literal x | badmatch x +case_end Literal=cq => move Literal x | case_end x +badmatch Literal=cq => move Literal x | badmatch x + +case_end r +case_end x +case_end y + +badmatch r +badmatch x +badmatch y -case_end s -badmatch s if_end raise s s @@ -248,12 +279,33 @@ system_limit j move R R => +move C=cxy r | jump Lbl => move_jump Lbl C + +%macro: move_jump MoveJump -nonext +move_jump f n +move_jump f c +move_jump f x +move_jump f y + move X1=x Y1=y | move X2=x Y2=y => move2 X1 Y1 X2 Y2 move Y1=y X1=x | move Y2=y X2=x => move2 Y1 X1 Y2 X2 +move X1=x X2=x | move X3=x X4=x => move2 X1 X2 X3 X4 + +move C=aiq X=x==1 => move_x1 C +move C=aiq X=x==2 => move_x2 C + +move_x1 c +move_x2 c %macro: move2 Move2 -pack move2 x y x y move2 y x y x +move2 x x x x + +# The compiler almost never generates a "move Literal y(Y)" instruction, +# so let's cheat if we encounter one. +move S=n D=y => init D +move S=c D=y => move S x | move x D %macro:move Move -pack -gen_dest move x x @@ -265,15 +317,10 @@ move r x move r y move c r move c x -move c y move n x move n r move y y -%cold -move s d -%hot - # Receive operations. loop_rec Fail Src | smp_mark_target_label(Fail) => i_loop_rec Fail Src @@ -306,55 +353,78 @@ i_wait_error_locked send # -# Comparisions. +# Optimized comparisons with one immediate/literal operand. +# + +is_eq_exact Lbl R=rxy C=ian => i_is_eq_exact_immed Lbl R C +is_eq_exact Lbl R=rxy C=q => i_is_eq_exact_literal R Lbl C + +is_ne_exact Lbl R=rxy C=ian => i_is_ne_exact_immed Lbl R C +is_ne_exact Lbl R=rxy C=q => i_is_ne_exact_literal R Lbl C + +%macro: i_is_eq_exact_immed EqualImmed -fail_action +i_is_eq_exact_immed f r c +i_is_eq_exact_immed f x c +i_is_eq_exact_immed f y c + +i_is_eq_exact_literal r f c +i_is_eq_exact_literal x f c +i_is_eq_exact_literal y f c + +%macro: i_is_ne_exact_immed NotEqualImmed -fail_action +i_is_ne_exact_immed f r c +i_is_ne_exact_immed f x c +i_is_ne_exact_immed f y c + +i_is_ne_exact_literal r f c +i_is_ne_exact_literal x f c +i_is_ne_exact_literal y f c + +# +# All other comparisons. # -is_eq_exact Lbl=f R=rxy C=ian => i_is_eq_immed Lbl R C -is_eq Lbl=f R=rxy C=an => i_is_eq_immed Lbl R C +is_eq_exact Lbl S1 S2 => i_fetch S1 S2 | i_is_eq_exact Lbl +is_ne_exact Lbl S1 S2 => i_fetch S1 S2 | i_is_ne_exact Lbl is_ge Lbl S1 S2 => i_fetch S1 S2 | i_is_ge Lbl is_lt Lbl S1 S2 => i_fetch S1 S2 | i_is_lt Lbl is_eq Lbl S1 S2 => i_fetch S1 S2 | i_is_eq Lbl is_ne Lbl S1 S2 => i_fetch S1 S2 | i_is_ne Lbl -is_eq_exact Lbl=f S1 S2 => i_fetch S1 S2 | i_is_eq_exact Lbl -is_ne_exact Lbl S1 S2 => i_fetch S1 S2 | i_is_ne_exact Lbl - +i_is_eq_exact f +i_is_ne_exact f i_is_lt f i_is_ge f i_is_eq f i_is_ne f -i_is_eq_exact f -i_is_ne_exact f - -%macro: i_is_eq_immed EqualImmed -fail_action -i_is_eq_immed f r c -i_is_eq_immed f x c -i_is_eq_immed f y c # # Putting things. # -put_tuple Arity Dst | put V => i_put_tuple Arity V Dst +put_tuple Arity Dst => i_put_tuple Dst u -%macro: i_put_tuple PutTuple -pack -i_put_tuple A x x -i_put_tuple A y x -i_put_tuple A r x -i_put_tuple A n x -i_put_tuple A c x -i_put_tuple A x y -i_put_tuple A x r -i_put_tuple A y r -i_put_tuple A n r -i_put_tuple A c r +i_put_tuple Dst Arity Puts=* | put S1 | put S2 | \ + put S3 | put S4 | put S5 => \ + tuple_append_put5(Arity, Dst, Puts, S1, S2, S3, S4, S5) -%cold -i_put_tuple A r y -i_put_tuple A y y -i_put_tuple A c y -%hot +i_put_tuple Dst Arity Puts=* | put S => \ + tuple_append_put(Arity, Dst, Puts, S) + +i_put_tuple/2 + +%macro:i_put_tuple PutTuple -pack -goto:do_put_tuple +i_put_tuple r I +i_put_tuple x I +i_put_tuple y I + +# +# The instruction "put_list Const [] Dst" will not be generated by +# the current BEAM compiler. But until R15A, play it safe by handling +# that instruction with the following transformation. +# +put_list Const=c n Dst => move Const x | put_list x n Dst %macro:put_list PutList -pack -gen_dest @@ -362,10 +432,8 @@ put_list x n x put_list y n x put_list x x x put_list y x x -put_list c n x put_list x x r put_list y r r -put_list c n r put_list y y x put_list x y x @@ -376,6 +444,13 @@ put_list y y r put_list y r x put_list r n x +put_list x r x +put_list x y r +put_list y x r +put_list y x x + +put_list x r r + # put_list SrcReg Constant Dst put_list r c r put_list r c x @@ -403,17 +478,9 @@ put_list c y x put_list c y y %cold -put_list x r r put_list s s d %hot -%macro: put Put -put x -put r -put y -put c -put n - %macro: i_fetch FetchArgs -pack i_fetch c c i_fetch c r @@ -464,19 +531,20 @@ move_return n r move S r | deallocate D | return => move_deallocate_return S r D -%macro: move_deallocate_return MoveDeallocateReturn -nonext -move_deallocate_return x r P -move_deallocate_return y r P -move_deallocate_return c r P -move_deallocate_return n r P +%macro: move_deallocate_return MoveDeallocateReturn -pack -nonext +move_deallocate_return x r Q +move_deallocate_return y r Q +move_deallocate_return c r Q +move_deallocate_return n r Q deallocate D | return => deallocate_return D %macro: deallocate_return DeallocateReturn -nonext -deallocate_return P +deallocate_return Q test_heap Need u==1 | put_list Y=y r r => test_heap_1_put_list Need Y +%macro: test_heap_1_put_list TestHeapPutList -pack test_heap_1_put_list I y # Test tuple & arity (head) @@ -576,14 +644,14 @@ is_list f y is_nonempty_list Fail=f S=rx | allocate Need Rs => is_nonempty_list_allocate Fail S Need Rs -%macro:is_nonempty_list_allocate IsNonemptyListAllocate -fail_action -is_nonempty_list_allocate f x I I -is_nonempty_list_allocate f r I I +%macro:is_nonempty_list_allocate IsNonemptyListAllocate -fail_action -pack +is_nonempty_list_allocate f x I t +is_nonempty_list_allocate f r I t is_nonempty_list F=f r | test_heap I1 I2 => is_non_empty_list_test_heap F r I1 I2 -%macro: is_non_empty_list_test_heap IsNonemptyListTestHeap -fail_action -is_non_empty_list_test_heap f r I I +%macro: is_non_empty_list_test_heap IsNonemptyListTestHeap -fail_action -pack +is_non_empty_list_test_heap f r I t %macro: is_nonempty_list IsNonemptyList -fail_action is_nonempty_list f x @@ -912,8 +980,13 @@ node x node y %hot -i_fast_element j I s d -i_element j s s d +i_fast_element r j I d +i_fast_element x j I d +i_fast_element y j I d + +i_element r j s d +i_element x j s d +i_element y j s d bif1 f b s d bif1_body b s d @@ -940,11 +1013,11 @@ move S r | call_last Ar P=f D => move_call_last S r P D i_move_call_last f P c r -%macro:move_call_last MoveCallLast -arg_f -nonext +%macro:move_call_last MoveCallLast -arg_f -nonext -pack move_call_last/4 -move_call_last x r f P -move_call_last y r f P +move_call_last x r f Q +move_call_last y r f Q move S=c r | call_only Ar P=f => i_move_call_only P S r move S=x r | call_only Ar P=f => move_call_only S r P @@ -1307,6 +1380,8 @@ fconv Arg=iqan Dst=l => move Arg x | fconv x Dst fmove q l fmove d l +fmove l d + fconv d l i_fadd l l l @@ -1322,12 +1397,6 @@ fcheckerror p => i_fcheckerror i_fcheckerror fclearerror -fmove FR=l Dst=d | new_float_allocation() => fmove_new FR Dst - -# The new instruction for moving a float out of a floating point register. -# (No allocation.) -fmove_new l d - # # New apply instructions in R10B. # @@ -1336,7 +1405,21 @@ apply I apply_last I P # -# New GCing arithmetic instructions. +# Optimize addition and subtraction of small literals using +# the i_increment/4 instruction (in bodies, not in guards). +# + +gc_bif2 p Live u$bif:erlang:splus/2 Int=i Reg=d Dst => \ + gen_increment(Reg, Int, Live, Dst) +gc_bif2 p Live u$bif:erlang:splus/2 Reg=d Int=i Dst => \ + gen_increment(Reg, Int, Live, Dst) + +gc_bif2 p Live u$bif:erlang:sminus/2 Reg=d Int=i Dst | \ + negation_is_small(Int) => \ + gen_increment_from_minus(Reg, Int, Live, Dst) + +# +# GCing arithmetic instructions. # gc_bif2 Fail I u$bif:erlang:splus/2 S1 S2 Dst=d => i_fetch S1 S2 | i_plus Fail I Dst @@ -1359,6 +1442,10 @@ gc_bif1 Fail I u$bif:erlang:bnot/1 Src Dst=d => i_int_bnot Fail Src I Dst gc_bif1 Fail I u$bif:erlang:sminus/1 Src Dst=d => i_fetch i Src | i_minus Fail I Dst gc_bif1 Fail I u$bif:erlang:splus/1 Src Dst=d => i_fetch i Src | i_plus Fail I Dst +i_increment r I I d +i_increment x I I d +i_increment y I I d + i_plus j I d i_minus j I d i_times j I d diff --git a/erts/emulator/beam/sys.h b/erts/emulator/beam/sys.h index 27c5f99320..dff2dc37a2 100644 --- a/erts/emulator/beam/sys.h +++ b/erts/emulator/beam/sys.h @@ -39,13 +39,6 @@ #define ENABLE_CHILD_WAITER_THREAD 1 #endif -/* The ERTS_TIMER_TREAD #define must be visible to the - erl_${OS}_sys.h #include files: it controls whether - certain optional facilities should be defined or not. */ -#if defined(ERTS_SMP) && 0 -#define ERTS_TIMER_THREAD -#endif - #if defined (__WIN32__) # include "erl_win_sys.h" #elif defined (VXWORKS) @@ -562,11 +555,7 @@ extern char *erts_default_arg0; extern char os_type[]; extern int sys_init_time(void); -#if defined(ERTS_TIMER_THREAD) -#define erts_deliver_time() -#else extern void erts_deliver_time(void); -#endif extern void erts_time_remaining(SysTimeval *); extern int erts_init_time_sup(void); extern void erts_sys_init_float(void); @@ -728,11 +717,11 @@ typedef enum { } erts_activity_error_t; typedef struct { - erts_smp_atomic_t do_block; + erts_smp_atomic32_t do_block; struct { - erts_smp_atomic_t wait; - erts_smp_atomic_t gc; - erts_smp_atomic_t io; + erts_smp_atomic32_t wait; + erts_smp_atomic32_t gc; + erts_smp_atomic32_t io; } in_activity; } erts_system_block_state_t; @@ -883,7 +872,7 @@ ERTS_GLB_INLINE int erts_smp_pending_system_block(void) { #ifdef ERTS_SMP - return erts_smp_atomic_read(&erts_system_block_state.do_block); + return (int) erts_smp_atomic32_read(&erts_system_block_state.do_block); #else return 0; #endif @@ -919,7 +908,7 @@ erts_smp_set_activity(erts_activity_t old_activity, case ERTS_ACTIVITY_UNDEFINED: break; case ERTS_ACTIVITY_WAIT: - erts_smp_atomic_dec(&erts_system_block_state.in_activity.wait); + erts_smp_atomic32_dec(&erts_system_block_state.in_activity.wait); if (locked) { /* You are not allowed to leave activity waiting * without supplying the possibility to block @@ -930,10 +919,10 @@ erts_smp_set_activity(erts_activity_t old_activity, } break; case ERTS_ACTIVITY_GC: - erts_smp_atomic_dec(&erts_system_block_state.in_activity.gc); + erts_smp_atomic32_dec(&erts_system_block_state.in_activity.gc); break; case ERTS_ACTIVITY_IO: - erts_smp_atomic_dec(&erts_system_block_state.in_activity.io); + erts_smp_atomic32_dec(&erts_system_block_state.in_activity.io); break; default: erts_set_activity_error(ERTS_ACT_ERR_LEAVE_UNKNOWN_ACTIVITY, @@ -949,13 +938,13 @@ erts_smp_set_activity(erts_activity_t old_activity, case ERTS_ACTIVITY_UNDEFINED: break; case ERTS_ACTIVITY_WAIT: - erts_smp_atomic_inc(&erts_system_block_state.in_activity.wait); + erts_smp_atomic32_inc(&erts_system_block_state.in_activity.wait); break; case ERTS_ACTIVITY_GC: - erts_smp_atomic_inc(&erts_system_block_state.in_activity.gc); + erts_smp_atomic32_inc(&erts_system_block_state.in_activity.gc); break; case ERTS_ACTIVITY_IO: - erts_smp_atomic_inc(&erts_system_block_state.in_activity.io); + erts_smp_atomic32_inc(&erts_system_block_state.in_activity.io); break; default: erts_set_activity_error(ERTS_ACT_ERR_ENTER_UNKNOWN_ACTIVITY, @@ -990,27 +979,31 @@ erts_smp_set_activity(erts_activity_t old_activity, typedef erts_smp_atomic_t erts_refc_t; -ERTS_GLB_INLINE void erts_refc_init(erts_refc_t *refcp, long val); -ERTS_GLB_INLINE void erts_refc_inc(erts_refc_t *refcp, long min_val); -ERTS_GLB_INLINE long erts_refc_inctest(erts_refc_t *refcp, long min_val); -ERTS_GLB_INLINE void erts_refc_dec(erts_refc_t *refcp, long min_val); -ERTS_GLB_INLINE long erts_refc_dectest(erts_refc_t *refcp, long min_val); -ERTS_GLB_INLINE void erts_refc_add(erts_refc_t *refcp, long diff, long min_val); -ERTS_GLB_INLINE long erts_refc_read(erts_refc_t *refcp, long min_val); +ERTS_GLB_INLINE void erts_refc_init(erts_refc_t *refcp, erts_aint_t val); +ERTS_GLB_INLINE void erts_refc_inc(erts_refc_t *refcp, erts_aint_t min_val); +ERTS_GLB_INLINE erts_aint_t erts_refc_inctest(erts_refc_t *refcp, + erts_aint_t min_val); +ERTS_GLB_INLINE void erts_refc_dec(erts_refc_t *refcp, erts_aint_t min_val); +ERTS_GLB_INLINE erts_aint_t erts_refc_dectest(erts_refc_t *refcp, + erts_aint_t min_val); +ERTS_GLB_INLINE void erts_refc_add(erts_refc_t *refcp, erts_aint_t diff, + erts_aint_t min_val); +ERTS_GLB_INLINE erts_aint_t erts_refc_read(erts_refc_t *refcp, + erts_aint_t min_val); #if ERTS_GLB_INLINE_INCL_FUNC_DEF ERTS_GLB_INLINE void -erts_refc_init(erts_refc_t *refcp, long val) +erts_refc_init(erts_refc_t *refcp, erts_aint_t val) { erts_smp_atomic_init((erts_smp_atomic_t *) refcp, val); } ERTS_GLB_INLINE void -erts_refc_inc(erts_refc_t *refcp, long min_val) +erts_refc_inc(erts_refc_t *refcp, erts_aint_t min_val) { #ifdef ERTS_REFC_DEBUG - long val = erts_smp_atomic_inctest((erts_smp_atomic_t *) refcp); + erts_aint_t val = erts_smp_atomic_inctest((erts_smp_atomic_t *) refcp); if (val < min_val) erl_exit(ERTS_ABORT_EXIT, "erts_refc_inc(): Bad refc found (refc=%ld < %ld)!\n", @@ -1020,10 +1013,10 @@ erts_refc_inc(erts_refc_t *refcp, long min_val) #endif } -ERTS_GLB_INLINE long -erts_refc_inctest(erts_refc_t *refcp, long min_val) +ERTS_GLB_INLINE erts_aint_t +erts_refc_inctest(erts_refc_t *refcp, erts_aint_t min_val) { - long val = erts_smp_atomic_inctest((erts_smp_atomic_t *) refcp); + erts_aint_t val = erts_smp_atomic_inctest((erts_smp_atomic_t *) refcp); #ifdef ERTS_REFC_DEBUG if (val < min_val) erl_exit(ERTS_ABORT_EXIT, @@ -1034,10 +1027,10 @@ erts_refc_inctest(erts_refc_t *refcp, long min_val) } ERTS_GLB_INLINE void -erts_refc_dec(erts_refc_t *refcp, long min_val) +erts_refc_dec(erts_refc_t *refcp, erts_aint_t min_val) { #ifdef ERTS_REFC_DEBUG - long val = erts_smp_atomic_dectest((erts_smp_atomic_t *) refcp); + erts_aint_t val = erts_smp_atomic_dectest((erts_smp_atomic_t *) refcp); if (val < min_val) erl_exit(ERTS_ABORT_EXIT, "erts_refc_dec(): Bad refc found (refc=%ld < %ld)!\n", @@ -1047,10 +1040,10 @@ erts_refc_dec(erts_refc_t *refcp, long min_val) #endif } -ERTS_GLB_INLINE long -erts_refc_dectest(erts_refc_t *refcp, long min_val) +ERTS_GLB_INLINE erts_aint_t +erts_refc_dectest(erts_refc_t *refcp, erts_aint_t min_val) { - long val = erts_smp_atomic_dectest((erts_smp_atomic_t *) refcp); + erts_aint_t val = erts_smp_atomic_dectest((erts_smp_atomic_t *) refcp); #ifdef ERTS_REFC_DEBUG if (val < min_val) erl_exit(ERTS_ABORT_EXIT, @@ -1061,10 +1054,10 @@ erts_refc_dectest(erts_refc_t *refcp, long min_val) } ERTS_GLB_INLINE void -erts_refc_add(erts_refc_t *refcp, long diff, long min_val) +erts_refc_add(erts_refc_t *refcp, erts_aint_t diff, erts_aint_t min_val) { #ifdef ERTS_REFC_DEBUG - long val = erts_smp_atomic_addtest((erts_smp_atomic_t *) refcp, diff); + erts_aint_t val = erts_smp_atomic_addtest((erts_smp_atomic_t *) refcp, diff); if (val < min_val) erl_exit(ERTS_ABORT_EXIT, "erts_refc_add(%ld): Bad refc found (refc=%ld < %ld)!\n", @@ -1074,10 +1067,10 @@ erts_refc_add(erts_refc_t *refcp, long diff, long min_val) #endif } -ERTS_GLB_INLINE long -erts_refc_read(erts_refc_t *refcp, long min_val) +ERTS_GLB_INLINE erts_aint_t +erts_refc_read(erts_refc_t *refcp, erts_aint_t min_val) { - long val = erts_smp_atomic_read((erts_smp_atomic_t *) refcp); + erts_aint_t val = erts_smp_atomic_read((erts_smp_atomic_t *) refcp); #ifdef ERTS_REFC_DEBUG if (val < min_val) erl_exit(ERTS_ABORT_EXIT, diff --git a/erts/emulator/beam/time.c b/erts/emulator/beam/time.c index 53d39aef0e..c65cc37fc6 100644 --- a/erts/emulator/beam/time.c +++ b/erts/emulator/beam/time.c @@ -99,80 +99,37 @@ static erts_smp_mtx_t tiw_lock; static ErlTimer** tiw; /* the timing wheel, allocated in init_time() */ static Uint tiw_pos; /* current position in wheel */ static Uint tiw_nto; /* number of timeouts in wheel */ +static Uint tiw_min; +static ErlTimer *tiw_min_ptr; /* END tiw_lock protected variables */ /* Actual interval time chosen by sys_init_time() */ static int itime; /* Constant after init */ -#if defined(ERTS_TIMER_THREAD) -static SysTimeval time_start; /* start of current time interval */ -static long ticks_end; /* time_start+ticks_end == time_wakeup */ -static long ticks_latest; /* delta from time_start at latest time update*/ - -static ERTS_INLINE long time_gettimeofday(SysTimeval *now) -{ - long elapsed; - - erts_get_timeval(now); - now->tv_usec = 1000 * (now->tv_usec / 1000); /* ms resolution */ - elapsed = (1000 * (now->tv_sec - time_start.tv_sec) + - (now->tv_usec - time_start.tv_usec) / 1000); - // elapsed /= CLOCK_RESOLUTION; - return elapsed; -} - -static long do_time_update(void) -{ - SysTimeval now; - long elapsed; - - elapsed = time_gettimeofday(&now); - ticks_latest = elapsed; - return elapsed; -} - -static ERTS_INLINE long do_time_read(void) -{ - return ticks_latest; -} - -static long do_time_reset(void) -{ - SysTimeval now; - long elapsed; - - elapsed = time_gettimeofday(&now); - time_start = now; - ticks_end = LONG_MAX; - ticks_latest = 0; - return elapsed; -} - -static ERTS_INLINE void do_time_init(void) -{ - (void)do_time_reset(); -} - -#else erts_smp_atomic_t do_time; /* set at clock interrupt */ -static ERTS_INLINE long do_time_read(void) { return erts_smp_atomic_read(&do_time); } -static ERTS_INLINE long do_time_update(void) { return do_time_read(); } +static ERTS_INLINE erts_aint_t do_time_read(void) { return erts_smp_atomic_read(&do_time); } +static ERTS_INLINE erts_aint_t do_time_update(void) { return do_time_read(); } static ERTS_INLINE void do_time_init(void) { erts_smp_atomic_init(&do_time, 0L); } -#endif /* get the time (in units of itime) to the next timeout, or -1 if there are no timeouts */ -static int next_time_internal(void) /* PRE: tiw_lock taken by caller */ +static erts_aint_t next_time_internal(void) /* PRE: tiw_lock taken by caller */ { int i, tm, nto; unsigned int min; ErlTimer* p; - long dt; + erts_aint_t dt; if (tiw_nto == 0) return -1; /* no timeouts in wheel */ + + if (tiw_min_ptr) { + min = tiw_min; + dt = do_time_read(); + return ((min >= dt) ? (min - dt) : 0); + } /* start going through wheel to find next timeout */ tm = nto = 0; @@ -185,11 +142,17 @@ static int next_time_internal(void) /* PRE: tiw_lock taken by caller */ if (p->count == 0) { /* found next timeout */ dt = do_time_read(); + /* p->count is zero */ + tiw_min_ptr = p; + tiw_min = tm; return ((tm >= dt) ? (tm - dt) : 0); } else { /* keep shortest time in 'min' */ - if (tm + p->count*TIW_SIZE < min) + if (tm + p->count*TIW_SIZE < min) { min = tm + p->count*TIW_SIZE; + tiw_min_ptr = p; + tiw_min = min; + } } p = p->next; } @@ -202,11 +165,35 @@ static int next_time_internal(void) /* PRE: tiw_lock taken by caller */ return ((min >= dt) ? (min - dt) : 0); } -#if !defined(ERTS_TIMER_THREAD) +static void remove_timer(ErlTimer *p) { + /* first */ + if (!p->prev) { + tiw[p->slot] = p->next; + if(p->next) + p->next->prev = NULL; + } else { + p->prev->next = p->next; + } + + /* last */ + if (!p->next) { + if (p->prev) + p->prev->next = NULL; + } else { + p->next->prev = p->prev; + } + + p->next = NULL; + p->prev = NULL; + /* Make sure cancel callback isn't called */ + p->active = 0; + tiw_nto--; +} + /* Private export to erl_time_sup.c */ -int next_time(void) +erts_aint_t erts_next_time(void) { - int ret; + erts_aint_t ret; erts_smp_mtx_lock(&tiw_lock); (void)do_time_update(); @@ -214,14 +201,13 @@ int next_time(void) erts_smp_mtx_unlock(&tiw_lock); return ret; } -#endif -static ERTS_INLINE void bump_timer_internal(long dt) /* PRE: tiw_lock is write-locked */ +static ERTS_INLINE void bump_timer_internal(erts_aint_t dt) /* PRE: tiw_lock is write-locked */ { Uint keep_pos; Uint count; ErlTimer *p, **prev, *timeout_head, **timeout_tail; - Uint dtime = (unsigned long)dt; + Uint dtime = (Uint) dt; /* no need to bump the position if there aren't any timeouts */ if (tiw_nto == 0) { @@ -242,12 +228,16 @@ static ERTS_INLINE void bump_timer_internal(long dt) /* PRE: tiw_lock is write-l if (tiw_pos == keep_pos) count--; prev = &tiw[tiw_pos]; while ((p = *prev) != NULL) { + ASSERT( p != p->next); if (p->count < count) { /* we have a timeout */ - *prev = p->next; /* Remove from list */ - tiw_nto--; - p->next = NULL; - p->active = 0; /* Make sure cancel callback - isn't called */ + /* remove min time */ + if (tiw_min_ptr == p) { + tiw_min_ptr = NULL; + tiw_min = 0; + } + + /* Remove from list */ + remove_timer(p); *timeout_tail = p; /* Insert in timeout queue */ timeout_tail = &p->next; } @@ -261,6 +251,8 @@ static ERTS_INLINE void bump_timer_internal(long dt) /* PRE: tiw_lock is write-l dtime--; } tiw_pos = keep_pos; + if (tiw_min_ptr) + tiw_min -= dt; erts_smp_mtx_unlock(&tiw_lock); @@ -275,24 +267,17 @@ static ERTS_INLINE void bump_timer_internal(long dt) /* PRE: tiw_lock is write-l * callback is called. */ p->next = NULL; + p->prev = NULL; p->slot = 0; (*p->timeout)(p->arg); } } -#if defined(ERTS_TIMER_THREAD) -static void timer_thread_bump_timer(void) -{ - erts_smp_mtx_lock(&tiw_lock); - bump_timer_internal(do_time_reset()); -} -#else -void bump_timer(long dt) /* dt is value from do_time */ +void erts_bump_timer(erts_aint_t dt) /* dt is value from do_time */ { erts_smp_mtx_lock(&tiw_lock); bump_timer_internal(dt); } -#endif Uint erts_timer_wheel_memory_size(void) @@ -300,82 +285,10 @@ erts_timer_wheel_memory_size(void) return (Uint) TIW_SIZE * sizeof(ErlTimer*); } -#if defined(ERTS_TIMER_THREAD) -static struct erts_iwait *timer_thread_iwait; - -static int timer_thread_setup_delay(SysTimeval *rem_time) -{ - long elapsed; - int ticks; - - erts_smp_mtx_lock(&tiw_lock); - elapsed = do_time_update(); - ticks = next_time_internal(); - if (ticks == -1) /* timer queue empty */ - ticks = 100*1000*1000; - if (elapsed > ticks) - elapsed = ticks; - ticks -= elapsed; - //ticks *= CLOCK_RESOLUTION; - rem_time->tv_sec = ticks / 1000; - rem_time->tv_usec = 1000 * (ticks % 1000); - ticks_end = ticks; - erts_smp_mtx_unlock(&tiw_lock); - return ticks; -} - -static void *timer_thread_start(void *ignore) -{ - SysTimeval delay; - -#ifdef ERTS_ENABLE_LOCK_CHECK - erts_lc_set_thread_name("timer"); -#endif - erts_register_blockable_thread(); - - for(;;) { - if (timer_thread_setup_delay(&delay)) { - erts_smp_activity_begin(ERTS_ACTIVITY_WAIT, NULL, NULL, NULL); - ASSERT_NO_LOCKED_LOCKS; - erts_iwait_wait(timer_thread_iwait, &delay); - ASSERT_NO_LOCKED_LOCKS; - erts_smp_activity_end(ERTS_ACTIVITY_WAIT, NULL, NULL, NULL); - } - else - erts_smp_chk_system_block(NULL, NULL, NULL); - timer_thread_bump_timer(); - ASSERT_NO_LOCKED_LOCKS; - } - /*NOTREACHED*/ - return NULL; -} - -static ERTS_INLINE void timer_thread_post_insert(Uint ticks) -{ - if ((Sint)ticks < ticks_end) - erts_iwait_interrupt(timer_thread_iwait); -} - -static void timer_thread_init(void) -{ - erts_thr_opts_t opts = ERTS_THR_OPTS_DEFAULT_INITER; - erts_tid_t tid; - - opts->detached = 1; - - timer_thread_iwait = erts_iwait_init(); - erts_thr_create(&tid, timer_thread_start, NULL, &opts); -} - -#else -static ERTS_INLINE void timer_thread_post_insert(Uint ticks) { } -static ERTS_INLINE void timer_thread_init(void) { } -#endif - /* this routine links the time cells into a free list at the start and sets the time queue as empty */ void -init_time(void) +erts_init_time(void) { int i; @@ -391,10 +304,13 @@ init_time(void) tiw[i] = NULL; do_time_init(); tiw_pos = tiw_nto = 0; - - timer_thread_init(); + tiw_min_ptr = NULL; + tiw_min = 0; } + + + /* ** Insert a process into the time queue, with a timeout 't' */ @@ -424,16 +340,31 @@ insert_timer(ErlTimer* p, Uint t) /* insert at head of list at slot */ p->next = tiw[tm]; + p->prev = NULL; + if (p->next != NULL) + p->next->prev = p; tiw[tm] = p; - tiw_nto++; - timer_thread_post_insert(ticks); + + /* insert min time */ + if ((tiw_nto == 0) || ((tiw_min_ptr != NULL) && (ticks < tiw_min))) { + tiw_min = ticks; + tiw_min_ptr = p; + } + if ((tiw_min_ptr == p) && (ticks > tiw_min)) { + /* some other timer might be 'min' now */ + tiw_min = 0; + tiw_min_ptr = NULL; + } + + tiw_nto++; } void -erl_set_timer(ErlTimer* p, ErlTimeoutProc timeout, ErlCancelProc cancel, +erts_set_timer(ErlTimer* p, ErlTimeoutProc timeout, ErlCancelProc cancel, void* arg, Uint t) { + erts_deliver_time(); erts_smp_mtx_lock(&tiw_lock); if (p->active) { /* XXX assert ? */ @@ -446,42 +377,34 @@ erl_set_timer(ErlTimer* p, ErlTimeoutProc timeout, ErlCancelProc cancel, p->active = 1; insert_timer(p, t); erts_smp_mtx_unlock(&tiw_lock); -#if defined(ERTS_SMP) && !defined(ERTS_TIMER_THREAD) +#if defined(ERTS_SMP) if (t <= (Uint) LONG_MAX) erts_sys_schedule_interrupt_timed(1, (long) t); #endif } void -erl_cancel_timer(ErlTimer* p) +erts_cancel_timer(ErlTimer* p) { - ErlTimer *tp; - ErlTimer **prev; - erts_smp_mtx_lock(&tiw_lock); if (!p->active) { /* allow repeated cancel (drivers) */ erts_smp_mtx_unlock(&tiw_lock); return; } - /* find p in linked list at slot p->slot and remove it */ - prev = &tiw[p->slot]; - while ((tp = *prev) != NULL) { - if (tp == p) { - *prev = p->next; /* Remove from list */ - tiw_nto--; - p->next = NULL; - p->slot = p->count = 0; - p->active = 0; - if (p->cancel != NULL) { - erts_smp_mtx_unlock(&tiw_lock); - (*p->cancel)(p->arg); - } else { - erts_smp_mtx_unlock(&tiw_lock); - } - return; - } else { - prev = &tp->next; - } + + /* is it the 'min' timer, remove min */ + if (p == tiw_min_ptr) { + tiw_min_ptr = NULL; + tiw_min = 0; + } + + remove_timer(p); + p->slot = p->count = 0; + + if (p->cancel != NULL) { + erts_smp_mtx_unlock(&tiw_lock); + (*p->cancel)(p->arg); + return; } erts_smp_mtx_unlock(&tiw_lock); } @@ -493,10 +416,10 @@ erl_cancel_timer(ErlTimer* p) immediately if it hadn't been cancelled). */ Uint -time_left(ErlTimer *p) +erts_time_left(ErlTimer *p) { Uint left; - long dt; + erts_aint_t dt; erts_smp_mtx_lock(&tiw_lock); @@ -517,12 +440,11 @@ time_left(ErlTimer *p) erts_smp_mtx_unlock(&tiw_lock); - return left * itime; + return (Uint) left * itime; } #ifdef DEBUG - -void p_slpq() +void erts_p_slpq() { int i; ErlTimer* p; @@ -551,5 +473,4 @@ void p_slpq() erts_smp_mtx_unlock(&tiw_lock); } - #endif /* DEBUG */ diff --git a/erts/emulator/beam/utils.c b/erts/emulator/beam/utils.c index ab5e8b5d4a..1d60b54d21 100644 --- a/erts/emulator/beam/utils.c +++ b/erts/emulator/beam/utils.c @@ -59,13 +59,6 @@ /* profile_scheduler mini message queue */ -#ifdef ERTS_TIMER_THREAD -/* A timer thread is not welcomed with this lock violation work around. - * - Bj�rn-Egil - */ -#error Timer thread may not be enabled due to lock violation. -#endif - typedef struct { Uint scheduler_id; Uint no_schedulers; @@ -3183,7 +3176,7 @@ erts_create_smp_ptimer(ErtsSmpPTimer **timer_ref, *timer_ref = res; - erl_set_timer(&res->timer.tm, + erts_set_timer(&res->timer.tm, (ErlTimeoutProc) ptimer_timeout, (ErlCancelProc) ptimer_cancelled, (void*) res, @@ -3197,7 +3190,7 @@ erts_cancel_smp_ptimer(ErtsSmpPTimer *ptimer) ASSERT(*ptimer->timer.timer_ref == ptimer); *ptimer->timer.timer_ref = NULL; ptimer->timer.flags |= ERTS_PTMR_FLG_CANCELLED; - erl_cancel_timer(&ptimer->timer.tm); + erts_cancel_timer(&ptimer->timer.tm); } } @@ -3637,19 +3630,19 @@ erts_set_activity_error(erts_activity_error_t error, char *file, int line) } -static ERTS_INLINE int +static ERTS_INLINE erts_aint32_t threads_not_under_control(void) { - int res = system_block_state.threads_to_block; + erts_aint32_t res = system_block_state.threads_to_block; /* Waiting is always an allowed activity... */ - res -= erts_smp_atomic_read(&erts_system_block_state.in_activity.wait); + res -= erts_smp_atomic32_read(&erts_system_block_state.in_activity.wait); if (system_block_state.allowed_activities & ERTS_BS_FLG_ALLOW_GC) - res -= erts_smp_atomic_read(&erts_system_block_state.in_activity.gc); + res -= erts_smp_atomic32_read(&erts_system_block_state.in_activity.gc); if (system_block_state.allowed_activities & ERTS_BS_FLG_ALLOW_IO) - res -= erts_smp_atomic_read(&erts_system_block_state.in_activity.io); + res -= erts_smp_atomic32_read(&erts_system_block_state.in_activity.io); if (res < 0) { ASSERT(0); @@ -3709,7 +3702,7 @@ erts_block_system(Uint32 allowed_activities) } else { - erts_smp_atomic_inc(&erts_system_block_state.do_block); + erts_smp_atomic32_inc(&erts_system_block_state.do_block); /* Someone else might be waiting for us to block... */ if (do_block) { @@ -3761,11 +3754,11 @@ erts_emergency_block_system(long timeout, Uint32 allowed_activities) another_blocker = erts_smp_pending_system_block(); system_block_state.emergency = 1; - erts_smp_atomic_inc(&erts_system_block_state.do_block); + erts_smp_atomic32_inc(&erts_system_block_state.do_block); if (another_blocker) { if (is_blocker()) { - erts_smp_atomic_dec(&erts_system_block_state.do_block); + erts_smp_atomic32_dec(&erts_system_block_state.do_block); res = 0; goto done; } @@ -3822,7 +3815,7 @@ erts_release_system(void) if (system_block_state.recursive_block) system_block_state.recursive_block--; else { - do_block = erts_smp_atomic_dectest(&erts_system_block_state.do_block); + do_block = erts_smp_atomic32_dectest(&erts_system_block_state.do_block); system_block_state.have_blocker = 0; if (is_blockable_thread()) system_block_state.threads_to_block++; @@ -3957,10 +3950,10 @@ erts_system_block_init(void) /* Global state... */ - erts_smp_atomic_init(&erts_system_block_state.do_block, 0L); - erts_smp_atomic_init(&erts_system_block_state.in_activity.wait, 0L); - erts_smp_atomic_init(&erts_system_block_state.in_activity.gc, 0L); - erts_smp_atomic_init(&erts_system_block_state.in_activity.io, 0L); + erts_smp_atomic32_init(&erts_system_block_state.do_block, 0); + erts_smp_atomic32_init(&erts_system_block_state.in_activity.wait, 0); + erts_smp_atomic32_init(&erts_system_block_state.in_activity.gc, 0); + erts_smp_atomic32_init(&erts_system_block_state.in_activity.io, 0); /* Make sure blockable threads unregister when exiting... */ erts_smp_install_exit_handler(erts_unregister_blockable_thread); diff --git a/erts/emulator/drivers/common/efile_drv.c b/erts/emulator/drivers/common/efile_drv.c index 786fa7da77..6449c6f506 100644 --- a/erts/emulator/drivers/common/efile_drv.c +++ b/erts/emulator/drivers/common/efile_drv.c @@ -385,7 +385,6 @@ struct t_data ErlDrvBinary *binp; int size; int offset; - char name[1]; } read_file; struct { struct t_readdir_buf *first_buf; @@ -1117,7 +1116,7 @@ static void invoke_read_file(void *data) Sint64 size; if (! (d->result_ok = - efile_openfile(&d->errInfo, d->c.read_file.name, + efile_openfile(&d->errInfo, d->b, EFILE_MODE_READ, &fd, &size))) { goto done; } @@ -3071,7 +3070,7 @@ file_outputv(ErlDrvData e, ErlIOVec *ev) { d->command = command; d->reply = !0; /* Copy name */ - FILENAME_COPY(d->c.read_file.name, filename); + FILENAME_COPY(d->b, filename); d->c.read_file.binp = NULL; d->invoke = invoke_read_file; d->free = free_read_file; diff --git a/erts/emulator/drivers/common/inet_drv.c b/erts/emulator/drivers/common/inet_drv.c index 1382d1dfe4..818bc6334e 100644 --- a/erts/emulator/drivers/common/inet_drv.c +++ b/erts/emulator/drivers/common/inet_drv.c @@ -4945,39 +4945,45 @@ static int inet_ctl_getifaddrs(inet_descriptor* desc_p, *buf_p++ = '\0'; *buf_p++ = INET_IFOPT_FLAGS; put_int32(IFGET_FLAGS(ifa_p->ifa_flags), buf_p); buf_p += 4; - if (ifa_p->ifa_addr->sa_family == AF_INET + if (ifa_p->ifa_addr) { + if (ifa_p->ifa_addr->sa_family == AF_INET #if defined(AF_INET6) - || ifa_p->ifa_addr->sa_family == AF_INET6 + || ifa_p->ifa_addr->sa_family == AF_INET6 #endif - ) { - SOCKADDR_TO_BUF(INET_IFOPT_ADDR, ifa_p->ifa_addr); - BUF_ENSURE(1); - SOCKADDR_TO_BUF(INET_IFOPT_NETMASK, ifa_p->ifa_netmask); - if (ifa_p->ifa_flags & IFF_POINTOPOINT) { - BUF_ENSURE(1); - SOCKADDR_TO_BUF(INET_IFOPT_DSTADDR, ifa_p->ifa_dstaddr); - } else if (ifa_p->ifa_flags & IFF_BROADCAST) { - BUF_ENSURE(1); - SOCKADDR_TO_BUF(INET_IFOPT_BROADADDR, ifa_p->ifa_broadaddr); + ) { + SOCKADDR_TO_BUF(INET_IFOPT_ADDR, ifa_p->ifa_addr); + if (ifa_p->ifa_netmask) { + BUF_ENSURE(1); + SOCKADDR_TO_BUF(INET_IFOPT_NETMASK, ifa_p->ifa_netmask); + } + if (ifa_p->ifa_dstaddr && + (ifa_p->ifa_flags & IFF_POINTOPOINT)) { + BUF_ENSURE(1); + SOCKADDR_TO_BUF(INET_IFOPT_DSTADDR, ifa_p->ifa_dstaddr); + } else if (ifa_p->ifa_broadaddr && + (ifa_p->ifa_flags & IFF_BROADCAST)) { + BUF_ENSURE(1); + SOCKADDR_TO_BUF(INET_IFOPT_BROADADDR, ifa_p->ifa_broadaddr); + } } - } #if defined(AF_LINK) || defined(AF_PACKET) - else if ( + else if ( #if defined(AF_LINK) - ifa_p->ifa_addr->sa_family == AF_LINK + ifa_p->ifa_addr->sa_family == AF_LINK #else - 0 + 0 #endif #if defined(AF_PACKET) - || ifa_p->ifa_addr->sa_family == AF_PACKET + || ifa_p->ifa_addr->sa_family == AF_PACKET #endif - ) { - char *bp = buf_p; - BUF_ENSURE(1); - SOCKADDR_TO_BUF(INET_IFOPT_HWADDR, ifa_p->ifa_addr); - if (buf_p - bp < 4) buf_p = bp; /* Empty hwaddr */ - } + ) { + char *bp = buf_p; + BUF_ENSURE(1); + SOCKADDR_TO_BUF(INET_IFOPT_HWADDR, ifa_p->ifa_addr); + if (buf_p - bp < 4) buf_p = bp; /* Empty hwaddr */ + } #endif + } BUF_ENSURE(1); *buf_p++ = '\0'; } diff --git a/erts/emulator/drivers/unix/unix_efile.c b/erts/emulator/drivers/unix/unix_efile.c index 6297ccb8bc..4b3934657c 100644 --- a/erts/emulator/drivers/unix/unix_efile.c +++ b/erts/emulator/drivers/unix/unix_efile.c @@ -620,7 +620,7 @@ efile_readdir(Efile_error* errInfo, /* Where to return error codes. */ if (IS_DOT_OR_DOTDOT(dirp->d_name)) continue; buffer[0] = '\0'; - strncat(buffer, dirp->d_name, size-1); + strncat(buffer, dirp->d_name, (*size)-1); *size = strlen(dirp->d_name); return 1; } diff --git a/erts/emulator/drivers/win32/win_con.c b/erts/emulator/drivers/win32/win_con.c index 2202ca655f..14f7941643 100644 --- a/erts/emulator/drivers/win32/win_con.c +++ b/erts/emulator/drivers/win32/win_con.c @@ -704,6 +704,18 @@ FrameWndProc(HWND hwnd, UINT iMsg, WPARAM wParam, LPARAM lParam) } write_inbuf(&c, 1); return 0; + case WM_MOUSEWHEEL: + { + int delta = GET_WHEEL_DELTA_WPARAM(wParam); + if (delta < 0) { + PostMessage(hClientWnd, WM_VSCROLL, MAKELONG(SB_THUMBTRACK, + (iVscrollPos + 5)),0); + } else { + WORD pos = ((iVscrollPos - 5) < 0) ? 0 : (iVscrollPos - 5); + PostMessage(hClientWnd, WM_VSCROLL, MAKELONG(SB_THUMBTRACK,pos),0); + } + return 0; + } case WM_CHAR: c = (TCHAR)wParam; write_inbuf(&c,1); diff --git a/erts/emulator/drivers/win32/win_efile.c b/erts/emulator/drivers/win32/win_efile.c index 4ec9579529..101853736a 100755 --- a/erts/emulator/drivers/win32/win_efile.c +++ b/erts/emulator/drivers/win32/win_efile.c @@ -1016,8 +1016,8 @@ efile_write_info(Efile_error* errInfo, } \ } - MKTIME(ModifyFileTime, pInfo->accessTime, mtime); - MKTIME(AccessFileTime, pInfo->modifyTime, atime); + MKTIME(ModifyFileTime, pInfo->modifyTime, mtime); + MKTIME(AccessFileTime, pInfo->accessTime, atime); MKTIME(CreationFileTime, pInfo->cTime, ctime); #undef MKTIME diff --git a/erts/emulator/sys/common/erl_mseg.c b/erts/emulator/sys/common/erl_mseg.c index b1ee165489..010d60eb63 100644 --- a/erts/emulator/sys/common/erl_mseg.c +++ b/erts/emulator/sys/common/erl_mseg.c @@ -35,6 +35,7 @@ #include "global.h" #include "erl_threads.h" #include "erl_mtrace.h" +#include "erl_time.h" #include "big.h" #if HAVE_ERTS_MSEG @@ -271,7 +272,7 @@ schedule_cache_check(void) #endif { cache_check_timer.active = 0; - erl_set_timer(&cache_check_timer, + erts_set_timer(&cache_check_timer, check_cache, NULL, NULL, diff --git a/erts/emulator/sys/common/erl_poll.c b/erts/emulator/sys/common/erl_poll.c index c17806d96c..4d0ca97889 100644 --- a/erts/emulator/sys/common/erl_poll.c +++ b/erts/emulator/sys/common/erl_poll.c @@ -124,9 +124,9 @@ erts_smp_mtx_unlock(&(PS)->mtx) #define ERTS_POLLSET_SET_POLLED_CHK(PS) \ - ((int) erts_smp_atomic_xchg(&(PS)->polled, (long) 1)) + ((int) erts_smp_atomic_xchg(&(PS)->polled, (erts_aint_t) 1)) #define ERTS_POLLSET_UNSET_POLLED(PS) \ - erts_smp_atomic_set(&(PS)->polled, (long) 0) + erts_smp_atomic_set(&(PS)->polled, (erts_aint_t) 0) #define ERTS_POLLSET_IS_POLLED(PS) \ ((int) erts_smp_atomic_read(&(PS)->polled)) @@ -134,11 +134,11 @@ #define ERTS_POLLSET_SET_POLLER_WOKEN(PS) \ do { \ ERTS_THR_MEMORY_BARRIER; \ - erts_smp_atomic_set(&(PS)->woken, (long) 1); \ + erts_smp_atomic_set(&(PS)->woken, (erts_aint_t) 1); \ } while (0) #define ERTS_POLLSET_UNSET_POLLER_WOKEN(PS) \ do { \ - erts_smp_atomic_set(&(PS)->woken, (long) 0); \ + erts_smp_atomic_set(&(PS)->woken, (erts_aint_t) 0); \ ERTS_THR_MEMORY_BARRIER; \ } while (0) #define ERTS_POLLSET_IS_POLLER_WOKEN(PS) \ @@ -179,9 +179,9 @@ do { \ #if ERTS_POLL_USE_UPDATE_REQUESTS_QUEUE #define ERTS_POLLSET_SET_HAVE_UPDATE_REQUESTS(PS) \ - erts_smp_atomic_set(&(PS)->have_update_requests, (long) 1) + erts_smp_atomic_set(&(PS)->have_update_requests, (erts_aint_t) 1) #define ERTS_POLLSET_UNSET_HAVE_UPDATE_REQUESTS(PS) \ - erts_smp_atomic_set(&(PS)->have_update_requests, (long) 0) + erts_smp_atomic_set(&(PS)->have_update_requests, (erts_aint_t) 0) #define ERTS_POLLSET_HAVE_UPDATE_REQUESTS(PS) \ ((int) erts_smp_atomic_read(&(PS)->have_update_requests)) #else @@ -202,13 +202,13 @@ do { \ #define ERTS_POLLSET_UNSET_INTERRUPTED_CHK(PS) unset_interrupted_chk((PS)) #define ERTS_POLLSET_UNSET_INTERRUPTED(PS) \ do { \ - erts_smp_atomic_set(&(PS)->interrupt, (long) 0); \ + erts_smp_atomic_set(&(PS)->interrupt, (erts_aint_t) 0); \ ERTS_THR_MEMORY_BARRIER; \ } while (0) #define ERTS_POLLSET_SET_INTERRUPTED(PS) \ do { \ ERTS_THR_MEMORY_BARRIER; \ - erts_smp_atomic_set(&(PS)->interrupt, (long) 1); \ + erts_smp_atomic_set(&(PS)->interrupt, (erts_aint_t) 1); \ } while (0) #define ERTS_POLLSET_IS_INTERRUPTED(PS) \ ((int) erts_smp_atomic_read(&(PS)->interrupt)) @@ -356,7 +356,7 @@ unset_interrupted_chk(ErtsPollSet ps) res = ps->interrupt; ps->interrupt = 0; #else - res = (int) erts_smp_atomic_xchg(&ps->interrupt, (long) 0); + res = (int) erts_smp_atomic_xchg(&ps->interrupt, (erts_aint_t) 0); ERTS_THR_MEMORY_BARRIER; #endif return res; @@ -369,7 +369,7 @@ static ERTS_INLINE int set_poller_woken_chk(ErtsPollSet ps) { ERTS_THR_MEMORY_BARRIER; - return (int) erts_smp_atomic_xchg(&ps->woken, (long) 1); + return (int) erts_smp_atomic_xchg(&ps->woken, (erts_aint_t) 1); } #endif @@ -1918,7 +1918,7 @@ check_fd_events(ErtsPollSet ps, SysTimeval *tv, int max_res, int *ps_locked) return 0; } else { - long timeout = tv->tv_sec*1000 + tv->tv_usec/1000; + erts_aint_t timeout = tv->tv_sec*1000 + tv->tv_usec/1000; ASSERT(timeout >= 0); erts_smp_atomic_set(&ps->timeout, timeout); #if ERTS_POLL_USE_FALLBACK @@ -2112,7 +2112,7 @@ ERTS_POLL_EXPORT(erts_poll_wait)(ErtsPollSet ps, #endif done: - erts_smp_atomic_set(&ps->timeout, LONG_MAX); + erts_smp_atomic_set(&ps->timeout, ERTS_AINT_T_MAX); #ifdef ERTS_POLL_DEBUG_PRINT erts_printf("Leaving %s = erts_poll_wait()\n", res == 0 ? "0" : erl_errno_id(res)); @@ -2150,10 +2150,12 @@ ERTS_POLL_EXPORT(erts_poll_interrupt)(ErtsPollSet ps, int set) * is not guaranteed that it will timeout before 'msec' milli seconds. */ void -ERTS_POLL_EXPORT(erts_poll_interrupt_timed)(ErtsPollSet ps, int set, long msec) +ERTS_POLL_EXPORT(erts_poll_interrupt_timed)(ErtsPollSet ps, + int set, + long msec) { if (set) { - if (erts_smp_atomic_read(&ps->timeout) > msec) { + if (erts_smp_atomic_read(&ps->timeout) > (erts_aint_t) msec) { ERTS_POLLSET_SET_INTERRUPTED(ps); #if ERTS_POLL_ASYNC_INTERRUPT_SUPPORT || defined(ERTS_SMP) wake_poller(ps); @@ -2315,7 +2317,7 @@ ERTS_POLL_EXPORT(erts_poll_create_pollset)(void) #else erts_smp_atomic_init(&ps->interrupt, 0); #endif - erts_smp_atomic_init(&ps->timeout, LONG_MAX); + erts_smp_atomic_init(&ps->timeout, ERTS_AINT_T_MAX); #ifdef ERTS_POLL_COUNT_AVOIDED_WAKEUPS erts_smp_atomic_init(&ps->no_avoided_wakeups, 0); erts_smp_atomic_init(&ps->no_avoided_interrupts, 0); diff --git a/erts/emulator/sys/unix/erl_unix_sys.h b/erts/emulator/sys/unix/erl_unix_sys.h index 2d5ef882f6..824678a0bb 100644 --- a/erts/emulator/sys/unix/erl_unix_sys.h +++ b/erts/emulator/sys/unix/erl_unix_sys.h @@ -329,11 +329,4 @@ extern int exit_async(void); #define ERTS_EXIT_AFTER_DUMP _exit -#ifdef ERTS_TIMER_THREAD -struct erts_iwait; /* opaque for clients */ -extern struct erts_iwait *erts_iwait_init(void); -extern void erts_iwait_wait(struct erts_iwait *iwait, struct timeval *delay); -extern void erts_iwait_interrupt(struct erts_iwait *iwait); -#endif /* ERTS_TIMER_THREAD */ - #endif /* #ifndef _ERL_UNIX_SYS_H */ diff --git a/erts/emulator/sys/unix/sys.c b/erts/emulator/sys/unix/sys.c index 01ba773688..bfc04faa45 100644 --- a/erts/emulator/sys/unix/sys.c +++ b/erts/emulator/sys/unix/sys.c @@ -237,9 +237,9 @@ static int max_files = -1; #ifdef ERTS_SMP erts_smp_atomic_t erts_break_requested; #define ERTS_SET_BREAK_REQUESTED \ - erts_smp_atomic_set(&erts_break_requested, (long) 1) + erts_smp_atomic_set(&erts_break_requested, (erts_aint_t) 1) #define ERTS_UNSET_BREAK_REQUESTED \ - erts_smp_atomic_set(&erts_break_requested, (long) 0) + erts_smp_atomic_set(&erts_break_requested, (erts_aint_t) 0) #else volatile int erts_break_requested = 0; #define ERTS_SET_BREAK_REQUESTED (erts_break_requested = 1) @@ -3109,226 +3109,3 @@ erl_sys_args(int* argc, char** argv) } *argc = j; } - -#ifdef ERTS_TIMER_THREAD - -/* - * Interruptible-wait facility: low-level synchronisation state - * and methods that are implementation dependent. - * - * Constraint: Every implementation must define 'struct erts_iwait' - * with a field 'erts_smp_atomic_t state;'. - */ - -/* values for struct erts_iwait's state field */ -#define IWAIT_WAITING 0 -#define IWAIT_AWAKE 1 -#define IWAIT_INTERRUPT 2 - -#if 0 /* XXX: needs feature test in erts/configure.in */ - -/* - * This is an implementation of the interruptible wait facility on - * top of Linux-specific futexes. - */ -#include <asm/unistd.h> -#define FUTEX_WAIT 0 -#define FUTEX_WAKE 1 -static int sys_futex(void *futex, int op, int val, const struct timespec *timeout) -{ - return syscall(__NR_futex, futex, op, val, timeout); -} - -struct erts_iwait { - erts_smp_atomic_t state; /* &state.counter is our futex */ -}; - -static void iwait_lowlevel_init(struct erts_iwait *iwait) { /* empty */ } - -static void iwait_lowlevel_wait(struct erts_iwait *iwait, struct timeval *delay) -{ - struct timespec timeout; - int res; - - timeout.tv_sec = delay->tv_sec; - timeout.tv_nsec = delay->tv_usec * 1000; - res = sys_futex((void*)&iwait->state.counter, FUTEX_WAIT, IWAIT_WAITING, &timeout); - if (res < 0 && errno != ETIMEDOUT && errno != EWOULDBLOCK && errno != EINTR) - perror("FUTEX_WAIT"); -} - -static void iwait_lowlevel_interrupt(struct erts_iwait *iwait) -{ - int res = sys_futex((void*)&iwait->state.counter, FUTEX_WAKE, 1, NULL); - if (res < 0) - perror("FUTEX_WAKE"); -} - -#else /* using poll() or select() */ - -/* - * This is an implementation of the interruptible wait facility on - * top of pipe(), poll() or select(), read(), and write(). - */ -struct erts_iwait { - erts_smp_atomic_t state; - int read_fd; /* wait polls and reads this fd */ - int write_fd; /* interrupt writes this fd */ -}; - -static void iwait_lowlevel_init(struct erts_iwait *iwait) -{ - int fds[2]; - - if (pipe(fds) < 0) { - perror("pipe()"); - exit(1); - } - iwait->read_fd = fds[0]; - iwait->write_fd = fds[1]; -} - -#if defined(ERTS_USE_POLL) - -#include <sys/poll.h> -#define PERROR_POLL "poll()" - -static int iwait_lowlevel_poll(int read_fd, struct timeval *delay) -{ - struct pollfd pollfd; - int timeout; - - pollfd.fd = read_fd; - pollfd.events = POLLIN; - pollfd.revents = 0; - timeout = delay->tv_sec * 1000 + delay->tv_usec / 1000; - return poll(&pollfd, 1, timeout); -} - -#else /* !ERTS_USE_POLL */ - -#include <sys/select.h> -#define PERROR_POLL "select()" - -static int iwait_lowlevel_poll(int read_fd, struct timeval *delay) -{ - fd_set readfds; - - FD_ZERO(&readfds); - FD_SET(read_fd, &readfds); - return select(read_fd + 1, &readfds, NULL, NULL, delay); -} - -#endif /* !ERTS_USE_POLL */ - -static void iwait_lowlevel_wait(struct erts_iwait *iwait, struct timeval *delay) -{ - int res; - char buf[64]; - - res = iwait_lowlevel_poll(iwait->read_fd, delay); - if (res > 0) - (void)read(iwait->read_fd, buf, sizeof buf); - else if (res < 0 && errno != EINTR) - perror(PERROR_POLL); -} - -static void iwait_lowlevel_interrupt(struct erts_iwait *iwait) -{ - int res = write(iwait->write_fd, "!", 1); - if (res < 0) - perror("write()"); -} - -#endif /* using poll() or select() */ - -#if 0 /* not using poll() or select() */ -/* - * This is an implementation of the interruptible wait facility on - * top of pthread_cond_timedwait(). This has two problems: - * 1. pthread_cond_timedwait() requires an absolute time point, - * so the relative delay must be converted to absolute time. - * Worse, this breaks if the machine's time is adjusted while - * we're preparing to wait. - * 2. Each cond operation requires additional mutex lock/unlock operations. - * - * Problem 2 is probably not too bad on Linux (they'll just become - * relatively cheap futex operations), but problem 1 is the real killer. - * Only use this implementation if no better alternatives are available! - */ -struct erts_iwait { - erts_smp_atomic_t state; - pthread_cond_t cond; - pthread_mutex_t mutex; -}; - -static void iwait_lowlevel_init(struct erts_iwait *iwait) -{ - iwait->cond = (pthread_cond_t) PTHREAD_COND_INITIALIZER; - iwait->mutex = (pthread_mutex_t) PTHREAD_MUTEX_INITIALIZER; -} - -static void iwait_lowlevel_wait(struct erts_iwait *iwait, struct timeval *delay) -{ - struct timeval tmp; - struct timespec timeout; - - /* Due to pthread_cond_timedwait()'s use of absolute - time, this must be the real gettimeofday(), _not_ - the "smoothed" one beam/erl_time_sup.c implements. */ - gettimeofday(&tmp, NULL); - - tmp.tv_sec += delay->tv_sec; - tmp.tv_usec += delay->tv_usec; - if (tmp.tv_usec >= 1000*1000) { - tmp.tv_usec -= 1000*1000; - tmp.tv_sec += 1; - } - timeout.tv_sec = tmp.tv_sec; - timeout.tv_nsec = tmp.tv_usec * 1000; - pthread_mutex_lock(&iwait->mutex); - pthread_cond_timedwait(&iwait->cond, &iwait->mutex, &timeout); - pthread_mutex_unlock(&iwait->mutex); -} - -static void iwait_lowlevel_interrupt(struct erts_iwait *iwait) -{ - pthread_mutex_lock(&iwait->mutex); - pthread_cond_signal(&iwait->cond); - pthread_mutex_unlock(&iwait->mutex); -} - -#endif /* not using POLL */ - -/* - * Interruptible-wait facility. This is just a wrapper around the - * low-level synchronisation code, where we maintain our logical - * state in order to suppress some state transitions. - */ - -struct erts_iwait *erts_iwait_init(void) -{ - struct erts_iwait *iwait = malloc(sizeof *iwait); - if (!iwait) { - perror("malloc"); - exit(1); - } - iwait_lowlevel_init(iwait); - erts_smp_atomic_init(&iwait->state, IWAIT_AWAKE); - return iwait; -} - -void erts_iwait_wait(struct erts_iwait *iwait, struct timeval *delay) -{ - if (erts_smp_atomic_xchg(&iwait->state, IWAIT_WAITING) != IWAIT_INTERRUPT) - iwait_lowlevel_wait(iwait, delay); - erts_smp_atomic_set(&iwait->state, IWAIT_AWAKE); -} - -void erts_iwait_interrupt(struct erts_iwait *iwait) -{ - if (erts_smp_atomic_xchg(&iwait->state, IWAIT_INTERRUPT) == IWAIT_WAITING) - iwait_lowlevel_interrupt(iwait); -} - -#endif /* ERTS_TIMER_THREAD */ diff --git a/erts/emulator/sys/vxworks/sys.c b/erts/emulator/sys/vxworks/sys.c index 411b4b37cf..c6e7b65f32 100644 --- a/erts/emulator/sys/vxworks/sys.c +++ b/erts/emulator/sys/vxworks/sys.c @@ -85,7 +85,7 @@ EXTERN_FUNCTION(void, erl_exit, (int n, char*, _DOTS_)); EXTERN_FUNCTION(void, erl_error, (char*, va_list)); EXTERN_FUNCTION(int, driver_interrupt, (int, int)); EXTERN_FUNCTION(void, increment_time, (int)); -EXTERN_FUNCTION(int, next_time, (_VOID_)); +EXTERN_FUNCTION(int, erts_next_time, (_VOID_)); EXTERN_FUNCTION(void, set_reclaim_free_function, (FreeFunction)); EXTERN_FUNCTION(int, erl_mem_info_get, (MEM_PART_STATS *)); EXTERN_FUNCTION(void, erl_crash_dump, (char* file, int line, char* fmt, ...)); diff --git a/erts/emulator/sys/win32/erl_poll.c b/erts/emulator/sys/win32/erl_poll.c index a766fe9575..d84ae2ede2 100644 --- a/erts/emulator/sys/win32/erl_poll.c +++ b/erts/emulator/sys/win32/erl_poll.c @@ -297,11 +297,11 @@ struct ErtsPollSet_ { #define ERTS_POLLSET_UNLOCK(PS) \ erts_smp_mtx_unlock(&(PS)->mtx) #define ERTS_POLLSET_SET_POLLED_CHK(PS) \ - ((int) erts_smp_atomic_xchg(&(PS)->polled, (long) 1)) + ((int) erts_smp_atomic_xchg(&(PS)->polled, (erts_aint_t) 1)) #define ERTS_POLLSET_SET_POLLED(PS) \ - erts_smp_atomic_set(&(PS)->polled, (long) 1) + erts_smp_atomic_set(&(PS)->polled, (erts_aint_t) 1) #define ERTS_POLLSET_UNSET_POLLED(PS) \ - erts_smp_atomic_set(&(PS)->polled, (long) 0) + erts_smp_atomic_set(&(PS)->polled, (erts_aint_t) 0) #define ERTS_POLLSET_IS_POLLED(PS) \ ((int) erts_smp_atomic_read(&(PS)->polled)) @@ -309,11 +309,11 @@ struct ErtsPollSet_ { #define ERTS_POLLSET_SET_POLLER_WOKEN(PS) \ do { \ ERTS_THR_MEMORY_BARRIER; \ - erts_smp_atomic_set(&(PS)->woken, (long) 1); \ + erts_smp_atomic_set(&(PS)->woken, (erts_aint_t) 1); \ } while (0) #define ERTS_POLLSET_UNSET_POLLER_WOKEN(PS) \ do { \ - erts_smp_atomic_set(&(PS)->woken, (long) 0); \ + erts_smp_atomic_set(&(PS)->woken, (erts_aint_t) 0); \ ERTS_THR_MEMORY_BARRIER; \ } while (0) #define ERTS_POLLSET_IS_POLLER_WOKEN(PS) \ @@ -322,13 +322,13 @@ do { \ #define ERTS_POLLSET_UNSET_INTERRUPTED_CHK(PS) unset_interrupted_chk((PS)) #define ERTS_POLLSET_UNSET_INTERRUPTED(PS) \ do { \ - erts_smp_atomic_set(&(PS)->interrupt, (long) 0); \ + erts_smp_atomic_set(&(PS)->interrupt, (erts_aint_t) 0); \ ERTS_THR_MEMORY_BARRIER; \ } while (0) #define ERTS_POLLSET_SET_INTERRUPTED(PS) \ do { \ ERTS_THR_MEMORY_BARRIER; \ - erts_smp_atomic_set(&(PS)->interrupt, (long) 1); \ + erts_smp_atomic_set(&(PS)->interrupt, (erts_aint_t) 1); \ } while (0) #define ERTS_POLLSET_IS_INTERRUPTED(PS) \ ((int) erts_smp_atomic_read(&(PS)->interrupt)) @@ -336,7 +336,7 @@ do { \ static ERTS_INLINE int unset_interrupted_chk(ErtsPollSet ps) { - int res = (int) erts_smp_atomic_xchg(&ps->interrupt, (long) 0); + int res = (int) erts_smp_atomic_xchg(&ps->interrupt, (erts_aint_t) 0); ERTS_THR_MEMORY_BARRIER; return res; @@ -346,7 +346,7 @@ static ERTS_INLINE int set_poller_woken_chk(ErtsPollSet ps) { ERTS_THR_MEMORY_BARRIER; - return (int) erts_smp_atomic_xchg(&ps->woken, (long) 1); + return (int) erts_smp_atomic_xchg(&ps->woken, (erts_aint_t) 1); } #else @@ -413,9 +413,9 @@ set_poller_woken_chk(ErtsPollSet ps) #ifdef ERTS_SMP extern erts_smp_atomic_t erts_break_requested; #define ERTS_SET_BREAK_REQUESTED \ - erts_smp_atomic_set(&erts_break_requested, (long) 1) + erts_smp_atomic_set(&erts_break_requested, (erts_aint_t) 1) #define ERTS_UNSET_BREAK_REQUESTED \ - erts_smp_atomic_set(&erts_break_requested, (long) 0) + erts_smp_atomic_set(&erts_break_requested, (erts_aint_t) 0) #else extern volatile int erts_break_requested; #define ERTS_SET_BREAK_REQUESTED (erts_break_requested = 1) @@ -986,7 +986,7 @@ void erts_poll_interrupt_timed(ErtsPollSet ps, HARDTRACEF(("In erts_poll_interrupt_timed(%d,%ld)",set,msec)); #ifdef ERTS_SMP if (set) { - if (erts_smp_atomic_read(&ps->timeout) > msec) { + if (erts_smp_atomic_read(&ps->timeout) > (erts_aint_t) msec) { ERTS_POLLSET_SET_INTERRUPTED(ps); wake_poller(ps); } @@ -1228,7 +1228,7 @@ int erts_poll_wait(ErtsPollSet ps, erts_mtx_unlock(&w->mtx); } done: - erts_smp_atomic_set(&ps->timeout, LONG_MAX); + erts_smp_atomic_set(&ps->timeout, ERTS_AINT_T_MAX); *len = num; ERTS_POLLSET_UNLOCK(ps); HARDTRACEF(("Out erts_poll_wait")); @@ -1314,7 +1314,7 @@ ErtsPollSet erts_poll_create_pollset(void) erts_smp_mtx_init(&ps->mtx, "pollset"); erts_smp_atomic_init(&ps->interrupt, 0); #endif - erts_smp_atomic_init(&ps->timeout, LONG_MAX); + erts_smp_atomic_init(&ps->timeout, ERTS_AINT_T_MAX); HARDTRACEF(("Out erts_poll_create_pollset")); return ps; diff --git a/erts/emulator/sys/win32/erl_win_dyn_driver.h b/erts/emulator/sys/win32/erl_win_dyn_driver.h index 4949998abc..1347eead91 100644 --- a/erts/emulator/sys/win32/erl_win_dyn_driver.h +++ b/erts/emulator/sys/win32/erl_win_dyn_driver.h @@ -87,15 +87,15 @@ WDD_TYPEDEF(unsigned long, erts_alc_test, (unsigned long, unsigned long, unsigned long, unsigned long)); -WDD_TYPEDEF(long, driver_binary_get_refc, (ErlDrvBinary *dbp)); -WDD_TYPEDEF(long, driver_binary_inc_refc, (ErlDrvBinary *dbp)); -WDD_TYPEDEF(long, driver_binary_dec_refc, (ErlDrvBinary *dbp)); +WDD_TYPEDEF(ErlDrvSInt, driver_binary_get_refc, (ErlDrvBinary *dbp)); +WDD_TYPEDEF(ErlDrvSInt, driver_binary_inc_refc, (ErlDrvBinary *dbp)); +WDD_TYPEDEF(ErlDrvSInt, driver_binary_dec_refc, (ErlDrvBinary *dbp)); WDD_TYPEDEF(ErlDrvPDL, driver_pdl_create, (ErlDrvPort)); WDD_TYPEDEF(void, driver_pdl_lock, (ErlDrvPDL)); WDD_TYPEDEF(void, driver_pdl_unlock, (ErlDrvPDL)); -WDD_TYPEDEF(long, driver_pdl_get_refc, (ErlDrvPDL)); -WDD_TYPEDEF(long, driver_pdl_inc_refc, (ErlDrvPDL)); -WDD_TYPEDEF(long, driver_pdl_dec_refc, (ErlDrvPDL)); +WDD_TYPEDEF(ErlDrvSInt, driver_pdl_get_refc, (ErlDrvPDL)); +WDD_TYPEDEF(ErlDrvSInt, driver_pdl_inc_refc, (ErlDrvPDL)); +WDD_TYPEDEF(ErlDrvSInt, driver_pdl_dec_refc, (ErlDrvPDL)); WDD_TYPEDEF(void, driver_system_info, (ErlDrvSysInfo *, size_t)); WDD_TYPEDEF(int, driver_get_now, (ErlDrvNowData *)); WDD_TYPEDEF(int, driver_monitor_process, (ErlDrvPort port, diff --git a/erts/emulator/sys/win32/sys_interrupt.c b/erts/emulator/sys/win32/sys_interrupt.c index d2449a1bdb..262f84babc 100644 --- a/erts/emulator/sys/win32/sys_interrupt.c +++ b/erts/emulator/sys/win32/sys_interrupt.c @@ -1,7 +1,7 @@ /* * %CopyrightBegin% * - * Copyright Ericsson AB 1997-2009. All Rights Reserved. + * Copyright Ericsson AB 1997-2010. 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 @@ -33,9 +33,9 @@ #ifdef ERTS_SMP erts_smp_atomic_t erts_break_requested; #define ERTS_SET_BREAK_REQUESTED \ - erts_smp_atomic_set(&erts_break_requested, (long) 1) + erts_smp_atomic_set(&erts_break_requested, (erts_aint_t) 1) #define ERTS_UNSET_BREAK_REQUESTED \ - erts_smp_atomic_set(&erts_break_requested, (long) 0) + erts_smp_atomic_set(&erts_break_requested, (erts_aint_t) 0) #else volatile int erts_break_requested = 0; #define ERTS_SET_BREAK_REQUESTED (erts_break_requested = 1) diff --git a/erts/emulator/test/beam_SUITE.erl b/erts/emulator/test/beam_SUITE.erl index 228ff15341..32ac07cb2d 100644 --- a/erts/emulator/test/beam_SUITE.erl +++ b/erts/emulator/test/beam_SUITE.erl @@ -20,7 +20,8 @@ -module(beam_SUITE). -export([all/1, packed_registers/1, apply_last/1, apply_last_bif/1, - buildo_mucho/1, heap_sizes/1, big_lists/1, fconv/1]). + buildo_mucho/1, heap_sizes/1, big_lists/1, fconv/1, + select_val/1]). -export([applied/2]). @@ -28,7 +29,7 @@ all(suite) -> [packed_registers, apply_last, apply_last_bif, buildo_mucho, - heap_sizes, big_lists]. + heap_sizes, big_lists, select_val]. %% Verify that apply(M, F, A) is really tail recursive. @@ -302,3 +303,19 @@ do_fconv(nil, Float) when is_float(Float) -> Float + []; do_fconv(tuple_literal, Float) when is_float(Float) -> Float + {a,b}. + +select_val(Config) when is_list(Config) -> + ?line zero = do_select_val(0), + ?line big = do_select_val(1 bsl 64), + ?line integer = do_select_val(42), + ok. + +do_select_val(X) -> + case X of + 0 -> + zero; + 1 bsl 64 -> + big; + Int when is_integer(Int) -> + integer + end. diff --git a/erts/emulator/test/beam_literals_SUITE.erl b/erts/emulator/test/beam_literals_SUITE.erl index 75841adbfc..1eda939cf8 100644 --- a/erts/emulator/test/beam_literals_SUITE.erl +++ b/erts/emulator/test/beam_literals_SUITE.erl @@ -23,7 +23,8 @@ matching_bigs/1, matching_more_bigs/1, matching_bigs_and_smalls/1, badmatch/1, case_clause/1, receiving/1, literal_type_tests/1, - put_list/1, fconv/1, literal_case_expression/1]). + put_list/1, fconv/1, literal_case_expression/1, + increment/1]). -include("test_server.hrl"). @@ -32,7 +33,7 @@ all(suite) -> matching_bigs, matching_more_bigs, matching_bigs_and_smalls, badmatch, case_clause, receiving, literal_type_tests, - put_list, fconv, literal_case_expression]. + put_list, fconv, literal_case_expression, increment]. putting(doc) -> "Test creating lists and tuples containing big number literals."; putting(Config) when is_list(Config) -> @@ -48,6 +49,7 @@ matching_bigs(doc) -> "Test matching of a few big number literals (in Beam," matching_bigs(Config) when is_list(Config) -> a = matching1(3972907842873739), b = matching1(-389789298378939783333333333333333333784), + other = matching1(3141699999999999999999999999999999999), other = matching1(42). matching_smalls(doc) -> "Test matching small numbers (both positive and negative)."; @@ -405,14 +407,51 @@ fconv_2(F) when is_float(F) -> literal_case_expression(Config) when is_list(Config) -> ?line DataDir = ?config(data_dir, Config), ?line Src = filename:join(DataDir, "literal_case_expression"), - ?line {ok,literal_case_expression=Mod,Code} = compile:file(Src, [from_asm,binary]), + ?line {ok,literal_case_expression=Mod,Code} = + compile:file(Src, [from_asm,binary]), ?line {module,Mod} = code:load_binary(Mod, Src, Code), ?line ok = Mod:x(), ?line ok = Mod:y(), + ?line ok = Mod:zi1(), + ?line ok = Mod:zi2(), + ?line ok = Mod:za1(), + ?line ok = Mod:za2(), ?line true = code:delete(Mod), ?line code:purge(Mod), ok. +%% Test the i_increment instruction. +increment(Config) when is_list(Config) -> + %% In the 32-bit emulator, Neg32 can be represented as a small, + %% but -Neg32 cannot. Therefore the i_increment instruction must + %% not be used in the subtraction that follows (since i_increment + %% cannot handle a bignum literal). + Neg32 = -(1 bsl 27), + Big32 = id(1 bsl 32), + Result32 = (1 bsl 32) + (1 bsl 27), + ?line Result32 = Big32 + (1 bsl 27), + ?line Result32 = Big32 - Neg32, + + %% Same thing, but for the 64-bit emulator. + Neg64 = -(1 bsl 59), + Big64 = id(1 bsl 64), + Result64 = (1 bsl 64) + (1 bsl 59), + ?line Result64 = Big64 + (1 bsl 59), + ?line Result64 = Big64 - Neg64, + + %% Test error handling for the i_increment instruction. + Bad = id(bad), + ?line {'EXIT',{badarith,_}} = (catch Bad + 42), + + %% Small operands, but a big result. + Res32 = 1 bsl 27, + Small32 = id(Res32-1), + ?line Res32 = Small32 + 1, + Res64 = 1 bsl 59, + Small64 = id(Res64-1), + ?line Res64 = Small64 + 1, + ok. + %% Help functions. chksum(Term) -> diff --git a/erts/emulator/test/beam_literals_SUITE_data/literal_case_expression.S b/erts/emulator/test/beam_literals_SUITE_data/literal_case_expression.S index c0ffe9ab53..bfdfc079dc 100644 --- a/erts/emulator/test/beam_literals_SUITE_data/literal_case_expression.S +++ b/erts/emulator/test/beam_literals_SUITE_data/literal_case_expression.S @@ -1,10 +1,11 @@ {module, literal_case_expression}. %% version = 0 -{exports, [{module_info,0},{module_info,1},{x,0},{y,0}]}. +{exports, [{module_info,0},{module_info,1},{x,0},{y,0}, + {zi1,0},{zi2,0},{za1,0},{za2,0}]}. {attributes, []}. -{labels, 15}. +{labels, 32}. {function, x, 0, 2}. @@ -52,6 +53,81 @@ {label,10}. {case_end,{float,34.0000}}. +{function, zi1, 0, 16}. + {label,15}. + {func_info,{atom,literal_case_expression},{atom,zi1},0}. + {label,16}. + {test,is_integer,{f,19},[{integer,42}]}. + {select_val,{integer,42}, + {f,18}, + {list,[{integer,42}, + {f,17}, + {integer,1000}, + {f,18}]}}. + {label,17}. + {move,{atom,ok},{x,0}}. + return. + {label,18}. + {move,{atom,error},{x,0}}. + return. + {label,19}. + {case_end,{integer,42}}. + +{function, zi2, 0, 16}. + {label,20}. + {func_info,{atom,literal_case_expression},{atom,zi2},0}. + {label,21}. + {test,is_integer,{f,23},[{integer,42}]}. + {select_val,{integer,42}, + {f,23}, + {list,[{integer,42}, + {f,22}, + {integer,1000}, + {f,23}]}}. + {label,22}. + {move,{atom,ok},{x,0}}. + return. + {label,23}. + {move,{atom,error},{x,0}}. + return. + +{function, za1, 0, 25}. + {label,24}. + {func_info,{atom,literal_case_expression},{atom,za1},0}. + {label,25}. + {test,is_atom,{f,28},[{atom,x}]}. + {select_val,{atom,x}, + {f,27}, + {list,[{atom,a}, + {f,27}, + {atom,x}, + {f,26}]}}. + {label,26}. + {move,{atom,ok},{x,0}}. + return. + {label,27}. + {move,{atom,error},{x,0}}. + return. + {label,28}. + {case_end,{atom,x}}. + +{function, za2, 0, 30}. + {label,29}. + {func_info,{atom,literal_case_expression},{atom,za2},0}. + {label,30}. + {test,is_atom,{f,32},[{atom,x}]}. + {select_val,{atom,x}, + {f,32}, + {list,[{atom,a}, + {f,32}, + {atom,x}, + {f,31}]}}. + {label,31}. + {move,{atom,ok},{x,0}}. + return. + {label,32}. + {move,{atom,error},{x,0}}. + return. {function, module_info, 0, 12}. {label,11}. diff --git a/erts/emulator/test/bs_construct_SUITE.erl b/erts/emulator/test/bs_construct_SUITE.erl index 3d9b51d278..138a19a626 100644 --- a/erts/emulator/test/bs_construct_SUITE.erl +++ b/erts/emulator/test/bs_construct_SUITE.erl @@ -1,7 +1,7 @@ %% %% %CopyrightBegin% %% -%% Copyright Ericsson AB 1999-2009. All Rights Reserved. +%% Copyright Ericsson AB 1999-2010. 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 @@ -27,7 +27,7 @@ mem_leak/1, coerce_to_float/1, bjorn/1, huge_float_field/1, huge_binary/1, system_limit/1, badarg/1, copy_writable_binary/1, kostis/1, dynamic/1, bs_add/1, - otp_7422/1]). + otp_7422/1, zero_width/1]). -include("test_server.hrl"). @@ -36,7 +36,7 @@ all(suite) -> not_used, in_guard, mem_leak, coerce_to_float, bjorn, huge_float_field, huge_binary, system_limit, badarg, copy_writable_binary, kostis, dynamic, bs_add, - otp_7422]. + otp_7422, zero_width]. big(1) -> 57285702734876389752897683. @@ -786,5 +786,20 @@ otp_7422_bin(N) when N < 512 -> end), otp_7422_bin(N+1); otp_7422_bin(_) -> ok. + +zero_width(Config) when is_list(Config) -> + ?line Z = id(0), + Small = id(42), + Big = id(1 bsl 128), + ?line <<>> = <<Small:Z>>, + ?line <<>> = <<Small:0>>, + ?line <<>> = <<Big:Z>>, + ?line <<>> = <<Big:0>>, + + ?line {'EXIT',{badarg,_}} = (catch <<not_a_number:0>>), + ?line {'EXIT',{badarg,_}} = (catch <<(id(not_a_number)):Z>>), + ?line {'EXIT',{badarg,_}} = (catch <<(id(not_a_number)):0>>), + + ok. id(I) -> I. diff --git a/erts/emulator/test/code_SUITE.erl b/erts/emulator/test/code_SUITE.erl index 33351a3cc9..5707de30d3 100644 --- a/erts/emulator/test/code_SUITE.erl +++ b/erts/emulator/test/code_SUITE.erl @@ -1,7 +1,7 @@ %% %% %CopyrightBegin% %% -%% Copyright Ericsson AB 1999-2009. All Rights Reserved. +%% Copyright Ericsson AB 1999-2011. All Rights Reserved. %% %% The contents of this file are subject to the Erlang Public License, %% Version 1.1, (the "License"); you may not use this file except in @@ -320,6 +320,9 @@ make_stub(Config) when is_list(Config) -> (catch code:make_stub_module(my_code_test, bit_sized_binary(Code), {[],[]})), + ?line {'EXIT',{badarg,_}} = + (catch code:make_stub_module(my_code_test_with_wrong_name, + Code, {[],[]})), ok. make_stub_many_funs(Config) when is_list(Config) -> diff --git a/erts/emulator/test/distribution_SUITE.erl b/erts/emulator/test/distribution_SUITE.erl index 79252d0593..f26455e6da 100644 --- a/erts/emulator/test/distribution_SUITE.erl +++ b/erts/emulator/test/distribution_SUITE.erl @@ -39,6 +39,7 @@ atom_roundtrip/1, atom_roundtrip_r12b/1, contended_atom_cache_entry/1, + bad_dist_structure/1, bad_dist_ext/1, bad_dist_ext_receive/1, bad_dist_ext_process_info/1, @@ -61,6 +62,7 @@ all(suite) -> [ stop_dist, trap_bif, dist_auto_connect, dist_parallel_send, atom_roundtrip, atom_roundtrip_r12b, contended_atom_cache_entry, + bad_dist_structure, bad_dist_ext ]. @@ -174,7 +176,7 @@ bulk_sendsend2(Terms, BinSize, BusyBufSize) -> ?line {ok, NodeRecv} = start_node(bulk_receiver), ?line Recv = spawn(NodeRecv, erlang, apply, [fun receiver/2, [0, 0]]), ?line Bin = list_to_binary(lists:duplicate(BinSize*1024, 253)), - ?line Size = Terms*size(Bin), + %%?line Size = Terms*size(Bin), %% SLF LEFT OFF HERE. %% When the caller uses small hunks, like 4k via @@ -187,7 +189,7 @@ bulk_sendsend2(Terms, BinSize, BusyBufSize) -> ?line {ok, NodeSend} = start_node(bulk_sender, "+zdbbl " ++ integer_to_list(BusyBufSize)), ?line _Send = spawn(NodeSend, erlang, apply, [fun sendersender/4, [self(), Recv, Bin, Terms]]), - ?line {Elapsed, {TermsN, SizeN}, MonitorCount} = + ?line {Elapsed, {_TermsN, SizeN}, MonitorCount} = receive {sendersender, BigRes} -> BigRes end, @@ -227,7 +229,7 @@ sendersender3(To, _Bin, 0, SendDone, MonitorCount) -> ok end, receive - {monitor, _Pid, _Type, _Info} = M -> + {monitor, _Pid, _Type, _Info} -> sendersender3(To, _Bin, 0, SendDone, MonitorCount + 1) after 0 -> if SendDone -> @@ -522,7 +524,7 @@ sink1() -> lost_exit(doc) -> "Test that EXIT and DOWN messages send to another node are not lost if " - "if the distribution port is busy."; + "the distribution port is busy."; lost_exit(Config) when is_list(Config) -> ?line {ok, Node} = start_node(lost_exit), @@ -1143,8 +1145,7 @@ contended_atom_cache_entry(Config) when is_list(Config) -> ?line {ok, SNode} = start_node(Config), ?line {ok, RNode} = start_node(Config), ?line Success = make_ref(), - ?line Mstr - = spawn_link( + ?line spawn_link( SNode, fun () -> erts_debug:set_internal_state(available_internal_state, @@ -1201,13 +1202,13 @@ contended_atom_cache_entry(Config) when is_list(Config) -> ?line stop_node(RNode), ?line ok. -send_ref_atom(To, Ref, Atom, 0) -> +send_ref_atom(_To, _Ref, _Atom, 0) -> ok; send_ref_atom(To, Ref, Atom, N) -> To ! {Ref, Atom}, send_ref_atom(To, Ref, Atom, N-1). -receive_ref_atom(Ref, Atom, 0) -> +receive_ref_atom(_Ref, _Atom, 0) -> ok; receive_ref_atom(Ref, Atom, N) -> receive @@ -1242,7 +1243,7 @@ unwanted_cixs() -> nodes()). -get_conflicting_atoms(CIX, 0) -> +get_conflicting_atoms(_CIX, 0) -> []; get_conflicting_atoms(CIX, N) -> {A, B, C} = now(), @@ -1256,6 +1257,186 @@ get_conflicting_atoms(CIX, N) -> get_conflicting_atoms(CIX, N) end. +-define(COOKIE, ''). +-define(DOP_LINK, 1). +-define(DOP_SEND, 2). +-define(DOP_EXIT, 3). +-define(DOP_UNLINK, 4). +-define(DOP_REG_SEND, 6). +-define(DOP_GROUP_LEADER, 7). +-define(DOP_EXIT2, 8). + +-define(DOP_SEND_TT, 12). +-define(DOP_EXIT_TT, 13). +-define(DOP_REG_SEND_TT, 16). +-define(DOP_EXIT2_TT, 18). + +-define(DOP_MONITOR_P, 19). +-define(DOP_DEMONITOR_P, 20). +-define(DOP_MONITOR_P_EXIT, 21). + +start_monitor(Offender,P) -> + ?line Parent = self(), + ?line Q = spawn(Offender, + fun () -> + Ref = erlang:monitor(process,P), + Parent ! {self(),ref,Ref}, + receive + just_stay_alive -> ok + end + end), + ?line Ref = receive + {Q,ref,R} -> + R + after 5000 -> + error + end, + io:format("Ref is ~p~n",[Ref]), + ok. +start_link(Offender,P) -> + ?line Parent = self(), + ?line Q = spawn(Offender, + fun () -> + process_flag(trap_exit,true), + link(P), + Parent ! {self(),ref,P}, + receive + just_stay_alive -> ok + end + end), + ?line Ref = receive + {Q,ref,R} -> + R + after 5000 -> + error + end, + io:format("Ref is ~p~n",[Ref]), + ok. + +bad_dist_structure(suite) -> + []; +bad_dist_structure(doc) -> + ["Test dist messages with valid structure (binary to term ok) but malformed" + "control content"]; +bad_dist_structure(Config) when is_list(Config) -> + %process_flag(trap_exit,true), + ODog = ?config(watchdog, Config), + ?t:timetrap_cancel(ODog), + Dog = ?t:timetrap(?t:seconds(15)), + + ?line {ok, Offender} = start_node(bad_dist_structure_offender), + ?line {ok, Victim} = start_node(bad_dist_structure_victim), + ?line start_node_monitors([Offender,Victim]), + ?line Parent = self(), + ?line P = spawn(Victim, + fun () -> + process_flag(trap_exit,true), + Parent ! {self(), started}, + receive check_msgs -> ok end, + bad_dist_struct_check_msgs([one, + two]), + Parent ! {self(), messages_checked}, + receive done -> ok end + end), + ?line receive {P, started} -> ok end, + ?line pong = rpc:call(Victim, net_adm, ping, [Offender]), + ?line verify_up(Offender, Victim), + ?line true = lists:member(Offender, rpc:call(Victim, erlang, nodes, [])), + ?line start_monitor(Offender,P), + ?line P ! one, + ?line send_bad_structure(Offender, P,{?DOP_MONITOR_P_EXIT,'replace',P,normal},2), + ?line pong = rpc:call(Victim, net_adm, ping, [Offender]), + ?line start_monitor(Offender,P), + ?line send_bad_structure(Offender, P,{?DOP_MONITOR_P_EXIT,'replace',P,normal,normal},2), + ?line pong = rpc:call(Victim, net_adm, ping, [Offender]), + ?line start_link(Offender,P), + ?line send_bad_structure(Offender, P,{?DOP_LINK},0), + ?line pong = rpc:call(Victim, net_adm, ping, [Offender]), + ?line start_link(Offender,P), + ?line send_bad_structure(Offender, P,{?DOP_UNLINK,'replace'},2), + ?line pong = rpc:call(Victim, net_adm, ping, [Offender]), + ?line start_link(Offender,P), + ?line send_bad_structure(Offender, P,{?DOP_UNLINK,'replace',make_ref()},2), + ?line pong = rpc:call(Victim, net_adm, ping, [Offender]), + ?line start_link(Offender,P), + ?line send_bad_structure(Offender, P,{?DOP_UNLINK,make_ref(),P},0), + ?line pong = rpc:call(Victim, net_adm, ping, [Offender]), + ?line start_link(Offender,P), + ?line send_bad_structure(Offender, P,{?DOP_UNLINK,normal,normal},0), + ?line pong = rpc:call(Victim, net_adm, ping, [Offender]), + ?line start_monitor(Offender,P), + ?line send_bad_structure(Offender, P,{?DOP_MONITOR_P,'replace',P},2), + ?line pong = rpc:call(Victim, net_adm, ping, [Offender]), + ?line start_monitor(Offender,P), + ?line send_bad_structure(Offender, P,{?DOP_MONITOR_P,'replace',P,normal},2), + ?line pong = rpc:call(Victim, net_adm, ping, [Offender]), + ?line start_monitor(Offender,P), + ?line send_bad_structure(Offender, P,{?DOP_DEMONITOR_P,'replace',P},2), + ?line pong = rpc:call(Victim, net_adm, ping, [Offender]), + ?line start_monitor(Offender,P), + ?line send_bad_structure(Offender, P,{?DOP_DEMONITOR_P,'replace',P,normal},2), + ?line pong = rpc:call(Victim, net_adm, ping, [Offender]), + ?line send_bad_structure(Offender, P,{?DOP_EXIT,'replace',P},2), + ?line pong = rpc:call(Victim, net_adm, ping, [Offender]), + ?line send_bad_structure(Offender, P,{?DOP_EXIT,make_ref(),normal,normal},0), + ?line pong = rpc:call(Victim, net_adm, ping, [Offender]), + ?line send_bad_structure(Offender, P,{?DOP_EXIT_TT,'replace',token,P},2), + ?line pong = rpc:call(Victim, net_adm, ping, [Offender]), + ?line send_bad_structure(Offender, P,{?DOP_EXIT_TT,make_ref(),token,normal,normal},0), + ?line pong = rpc:call(Victim, net_adm, ping, [Offender]), + ?line send_bad_structure(Offender, P,{?DOP_EXIT2,'replace',P},2), + ?line pong = rpc:call(Victim, net_adm, ping, [Offender]), + ?line send_bad_structure(Offender, P,{?DOP_EXIT2,make_ref(),normal,normal},0), + ?line pong = rpc:call(Victim, net_adm, ping, [Offender]), + ?line send_bad_structure(Offender, P,{?DOP_EXIT2_TT,'replace',token,P},2), + ?line pong = rpc:call(Victim, net_adm, ping, [Offender]), + ?line send_bad_structure(Offender, P,{?DOP_EXIT2_TT,make_ref(),token,normal,normal},0), + ?line pong = rpc:call(Victim, net_adm, ping, [Offender]), + ?line send_bad_structure(Offender, P,{?DOP_GROUP_LEADER,'replace'},2), + ?line pong = rpc:call(Victim, net_adm, ping, [Offender]), + ?line send_bad_structure(Offender, P,{?DOP_GROUP_LEADER,'replace','atomic'},2), + ?line pong = rpc:call(Victim, net_adm, ping, [Offender]), + ?line send_bad_structure(Offender, P,{?DOP_GROUP_LEADER,'replace',P},0), + ?line pong = rpc:call(Victim, net_adm, ping, [Offender]), + ?line send_bad_structure(Offender, P,{?DOP_REG_SEND_TT,'replace','',name},2,{message}), + ?line pong = rpc:call(Victim, net_adm, ping, [Offender]), + ?line send_bad_structure(Offender, P,{?DOP_REG_SEND_TT,'replace','',name,token},0,{message}), + ?line pong = rpc:call(Victim, net_adm, ping, [Offender]), + ?line send_bad_structure(Offender, P,{?DOP_REG_SEND,'replace',''},2,{message}), + ?line pong = rpc:call(Victim, net_adm, ping, [Offender]), + ?line send_bad_structure(Offender, P,{?DOP_REG_SEND,'replace','',P},0,{message}), + ?line pong = rpc:call(Victim, net_adm, ping, [Offender]), + ?line send_bad_structure(Offender, P,{?DOP_REG_SEND,'replace','',name},0,{message}), + ?line pong = rpc:call(Victim, net_adm, ping, [Offender]), + ?line send_bad_structure(Offender, P,{?DOP_REG_SEND,'replace','',name,{token}},2,{message}), + ?line pong = rpc:call(Victim, net_adm, ping, [Offender]), + ?line send_bad_structure(Offender, P,{?DOP_SEND_TT,'',P},0,{message}), + ?line pong = rpc:call(Victim, net_adm, ping, [Offender]), + ?line send_bad_structure(Offender, P,{?DOP_SEND_TT,'',name,token},0,{message}), + ?line pong = rpc:call(Victim, net_adm, ping, [Offender]), + ?line send_bad_structure(Offender, P,{?DOP_SEND,''},0,{message}), + ?line pong = rpc:call(Victim, net_adm, ping, [Offender]), + ?line send_bad_structure(Offender, P,{?DOP_SEND,'',name},0,{message}), + ?line pong = rpc:call(Victim, net_adm, ping, [Offender]), + ?line send_bad_structure(Offender, P,{?DOP_SEND,'',P,{token}},0,{message}), + ?line pong = rpc:call(Victim, net_adm, ping, [Offender]), + ?line P ! two, + ?line P ! check_msgs, + ?line receive + {P, messages_checked} -> ok + after 5000 -> + exit(victim_is_dead) + end, + + ?line {message_queue_len, 0} + = rpc:call(Victim, erlang, process_info, [P, message_queue_len]), + + ?line unlink(P), + ?line P ! done, + ?line stop_node(Offender), + ?line stop_node(Victim), + ?t:timetrap_cancel(Dog), + ok. bad_dist_ext(doc) -> []; bad_dist_ext(suite) -> @@ -1483,6 +1664,22 @@ bad_dist_ext_connection_id(Config) when is_list(Config) -> ?line stop_node(Victim). +bad_dist_struct_check_msgs([]) -> + receive + Msg -> + exit({unexpected_message, Msg}) + after 0 -> + ok + end; +bad_dist_struct_check_msgs([M|Ms]) -> + receive + {'EXIT',_,_} = EM -> + io:format("Ignoring exit message: ~p~n",[EM]), + bad_dist_struct_check_msgs([M|Ms]); + Msg -> + M = Msg, + bad_dist_struct_check_msgs(Ms) + end. bad_dist_ext_check_msgs([]) -> receive Msg -> @@ -1497,24 +1694,6 @@ bad_dist_ext_check_msgs([M|Ms]) -> bad_dist_ext_check_msgs(Ms) end. --define(COOKIE, ''). --define(DOP_LINK, 1). --define(DOP_SEND, 2). --define(DOP_EXIT, 3). --define(DOP_UNLINK, 4). --define(DOP_NODE_LINK, 5). --define(DOP_REG_SEND, 6). --define(DOP_GROUP_LEADER, 7). --define(DOP_EXIT2, 8). - --define(DOP_SEND_TT, 12). --define(DOP_EXIT_TT, 13). --define(DOP_REG_SEND_TT, 16). --define(DOP_EXIT2_TT, 18). - --define(DOP_MONITOR_P, 19). --define(DOP_DEMONITOR_P, 20). --define(DOP_MONITOR_P_EXIT, 21). dport_reg_send(Node, Name, Msg) -> DPrt = case dport(Node) of @@ -1546,6 +1725,39 @@ dport_send(To, Msg) -> ?COOKIE, To}), dmsg_ext(Msg)]). +send_bad_structure(Offender,Victim,Bad,WhereToPutSelf) -> + send_bad_structure(Offender,Victim,Bad,WhereToPutSelf,[]). +send_bad_structure(Offender,Victim,Bad,WhereToPutSelf,PayLoad) -> + Parent = self(), + Done = make_ref(), + spawn(Offender, + fun () -> + Node = node(Victim), + pong = net_adm:ping(Node), + DPrt = dport(Node), + Bad1 = case WhereToPutSelf of + 0 -> + Bad; + N when N > 0 -> + setelement(N,Bad,self()) + end, + DData = [dmsg_hdr(), + dmsg_ext(Bad1)] ++ + case PayLoad of + [] -> []; + _Other -> [dmsg_ext(PayLoad)] + end, + port_command(DPrt, DData), + Parent ! {DData,Done} + end), + receive + {WhatSent,Done} -> + io:format("Offender sent ~p~n",[WhatSent]), + ok + after 5000 -> + exit(unable_to_send) + end. + %% send_bad_msgs(): %% Send a valid distribution header and control message @@ -1629,10 +1841,10 @@ dmsg_bad_hdr() -> 255]. % 255 atom references -dmsg_fake_hdr1() -> - A = <<"fake header atom 1">>, - [131, % Version Magic - $D, 1, 16#8, 0, size(A), A]. % Fake header +%% dmsg_fake_hdr1() -> +%% A = <<"fake header atom 1">>, +%% [131, % Version Magic +%% $D, 1, 16#8, 0, size(A), A]. % Fake header dmsg_fake_hdr2() -> A1 = <<"fake header atom 1">>, @@ -1817,7 +2029,7 @@ flush_node_changes() -> node_monitor_loop(Master) -> receive - {nodeup, Node, InfoList} = Msg -> + {nodeup, Node, _InfoList} = Msg -> Master ! {nodeup, node(), Node}, ?t:format("~p ~p: ~p~n", [node(), erlang:now(), Msg]), node_monitor_loop(Master); @@ -1854,9 +2066,9 @@ verify_no_down(A, B) -> ok end. -verify_down(A, B) -> - receive {nodedown, A, B, _} -> ok end, - receive {nodedown, B, A, _} -> ok end. +%% verify_down(A, B) -> +%% receive {nodedown, A, B, _} -> ok end, +%% receive {nodedown, B, A, _} -> ok end. verify_down(A, ReasonA, B, ReasonB) -> receive @@ -1876,11 +2088,11 @@ from(H, [H | T]) -> T; from(H, [_ | T]) -> from(H, T); from(_, []) -> []. -fun_spawn(Fun) -> - fun_spawn(Fun, []). +%% fun_spawn(Fun) -> +%% fun_spawn(Fun, []). -fun_spawn(Fun, Args) -> - spawn_link(erlang, apply, [Fun, Args]). +%% fun_spawn(Fun, Args) -> +%% spawn_link(erlang, apply, [Fun, Args]). long_or_short() -> diff --git a/erts/emulator/test/erl_link_SUITE.erl b/erts/emulator/test/erl_link_SUITE.erl index 542c8dffbe..11a7a61586 100644 --- a/erts/emulator/test/erl_link_SUITE.erl +++ b/erts/emulator/test/erl_link_SUITE.erl @@ -1,7 +1,7 @@ %% %% %CopyrightBegin% %% -%% Copyright Ericsson AB 2001-2009. All Rights Reserved. +%% Copyright Ericsson AB 2001-2010. 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 @@ -1050,7 +1050,6 @@ stop_node(Node) -> -define(DOP_SEND, 2). -define(DOP_EXIT, 3). -define(DOP_UNLINK, 4). --define(DOP_NODE_LINK, 5). -define(DOP_REG_SEND, 6). -define(DOP_GROUP_LEADER, 7). -define(DOP_EXIT2, 8). diff --git a/erts/emulator/test/erts_debug_SUITE.erl b/erts/emulator/test/erts_debug_SUITE.erl index e60a999df1..934a1b10a4 100644 --- a/erts/emulator/test/erts_debug_SUITE.erl +++ b/erts/emulator/test/erts_debug_SUITE.erl @@ -21,10 +21,10 @@ -include("test_server.hrl"). -export([all/1,init_per_testcase/2,fin_per_testcase/2, - flat_size/1,flat_size_big/1,df/1]). + flat_size/1,flat_size_big/1,df/1,instructions/1]). all(suite) -> - [flat_size,flat_size_big,df]. + [flat_size,flat_size_big,df,instructions]. init_per_testcase(Func, Config) when is_atom(Func), is_list(Config) -> Dog=?t:timetrap(?t:minutes(2)), @@ -70,3 +70,8 @@ df(Config) when is_list(Config) -> pps() -> {erlang:ports()}. + +instructions(Config) when is_list(Config) -> + ?line Is = erts_debug:instructions(), + ?line _ = [list_to_atom(I) || I <- Is], + ok. diff --git a/erts/emulator/utils/beam_makeops b/erts/emulator/utils/beam_makeops index de19a2e35b..e7c57142c0 100755 --- a/erts/emulator/utils/beam_makeops +++ b/erts/emulator/utils/beam_makeops @@ -27,6 +27,7 @@ my $outdir = "."; # Directory for output files. my $verbose = 0; my $hot = 1; my $num_file_opcodes = 0; +my $wordsize = 32; # This is shift counts and mask for the packer. my $WHOLE_WORD = ''; @@ -36,12 +37,20 @@ my @pack_mask; $pack_instr[2] = ['6', 'i']; $pack_instr[3] = ['0', '0', 'i']; +$pack_instr[4] = ['6', '6', '6', 'i']; # Only for 64 bit wordsize $pack_shift[2] = ['0', 'BEAM_LOOSE_SHIFT']; $pack_shift[3] = ['0', 'BEAM_TIGHT_SHIFT', '(2*BEAM_TIGHT_SHIFT)']; +$pack_shift[4] = ['0', 'BEAM_LOOSE_SHIFT', # Only for 64 bit wordsize + '(2*BEAM_LOOSE_SHIFT)', + '(3*BEAM_LOOSE_SHIFT)']; $pack_mask[2] = ['BEAM_LOOSE_MASK', $WHOLE_WORD]; $pack_mask[3] = ['BEAM_TIGHT_MASK', 'BEAM_TIGHT_MASK', 'BEAM_TIGHT_MASK']; +$pack_mask[4] = ['BEAM_LOOSE_MASK', # Only for 64 bit wordsize + 'BEAM_LOOSE_MASK', + 'BEAM_LOOSE_MASK', + $WHOLE_WORD]; # There are two types of instructions: generic and specific. # The generic instructions are those generated by the Beam compiler. @@ -80,6 +89,8 @@ my %cold_code; my @unnumbered_generic; my %unnumbered; +my %is_transformed; + # # Code transformations. # @@ -118,7 +129,8 @@ my %arg_size = ('r' => 0, # x(0) - x register zero 't' => 1, # untagged integer -- can be packed 'b' => 1, # pointer to bif 'A' => 1, # arity value - 'P' => 1, # byte offset into tuple + 'P' => 1, # byte offset into tuple or stack + 'Q' => 1, # like 'P', but packable 'h' => 1, # character 'l' => 1, # float reg 'q' => 1, # literal term @@ -157,6 +169,7 @@ my @tag_type; $type_bit{'U'} = $type_bit{'u'}; $type_bit{'e'} = $type_bit{'u'}; $type_bit{'P'} = $type_bit{'u'}; + $type_bit{'Q'} = $type_bit{'u'}; } # @@ -169,6 +182,7 @@ while (@ARGV && $ARGV[0] =~ /^-(.*)/) { ($target = \&emulator_output), next if /^emulator/; ($target = \&compiler_output), next if /^compiler/; ($outdir = shift), next if /^outdir/; + ($wordsize = shift), next if /^wordsize/; ($verbose = 1), next if /^v/; die "$0: Bad option: -$_\n"; } @@ -474,8 +488,9 @@ sub emulator_output { $gen_transform_offset{$key} : -1; my($spec_op) = $gen_to_spec{$key}; my($num_specific) = $num_specific{$key}; - defined $spec_op or $tr != -1 or + defined $spec_op or $obsolete[$gen_opnum{$name,$arity}] or + $is_transformed{$name,$arity} or error("instruction $key has no specific instruction"); $spec_op = -1 unless defined $spec_op; &init_item($name, $arity, $spec_op, $num_specific, $tr, $min_window{$key}); @@ -498,12 +513,14 @@ sub emulator_output { print "#define NUM_SPECIFIC_OPS ", scalar(@op_to_name), "\n"; print "\n"; print "#ifdef ARCH_64\n"; + print "# define BEAM_WIDE_MASK 0xFFFFUL\n"; print "# define BEAM_LOOSE_MASK 0x1FFFUL\n"; print "#if HALFWORD_HEAP\n"; print "# define BEAM_TIGHT_MASK 0x1FFCUL\n"; print "#else\n"; print "# define BEAM_TIGHT_MASK 0x1FF8UL\n"; print "#endif\n"; + print "# define BEAM_WIDE_SHIFT 32\n"; print "# define BEAM_LOOSE_SHIFT 16\n"; print "# define BEAM_TIGHT_SHIFT 16\n"; print "#else\n"; @@ -796,6 +813,7 @@ sub basic_generator { 'I' => 1, 't' => 1, 'P' => 1, + 'Q' => 1, ); # Pick up the macro to use and its flags (if any). @@ -916,7 +934,18 @@ sub basic_generator { $var_decls .= "BeamInstr tmp_packed2;" if $macro_code =~ /tmp_packed2/; if ($flags =~ /-nonext/) { - $code = "$macro_code\n"; + $code = join("\n", + "{ $var_decls", + $macro_code, + "}"); + } elsif ($flags =~ /-goto:(\S*)/) { + my $goto = $1; + $code = join("\n", + "{ $var_decls", + $macro_code, + "I += $size + 1;", + "goto $goto;", + "}"); } else { $code = join("\n", "{ $var_decls", @@ -935,18 +964,31 @@ sub basic_generator { sub do_pack { my(@args) = @_; - my($i); my($packable_args) = 0; + my @is_packable; # Packability (boolean) for each argument. + my $wide_packing = 0; # # Count the number of packable arguments. If we encounter any 's' or 'd' # arguments, packing is not possible. # - for ($i = 0; $i < @args; $i++) { - if ($args[$i] =~ /[xyt]/) { + my $packable_types = "xytQ"; + foreach my $arg (@args) { + if ($arg =~ /^[$packable_types]/) { $packable_args++; - } elsif ($args[$i] =~ /[sd]/) { + push @is_packable, 1; + } elsif ($arg =~ /^I/ and $wordsize == 64 and $packable_args < 2) { + $wide_packing = 1; + push @is_packable, 1; + if (++$packable_args == 2) { + # We can only pack two arguments. Turn off packing + # for the rest of the arguments. + $packable_types = "\xFF"; + } + } elsif ($arg =~ /^[sd]/) { return ('', '', @args); + } else { + push @is_packable, 0; } } @@ -962,10 +1004,27 @@ sub do_pack { # beginning). my($up) = ''; # Pack commands (storing back while # moving forward). - my($args_per_word) = $packable_args < 4 ? $packable_args : 2; - my(@shift) = @{$pack_shift[$args_per_word]}; - my(@mask) = @{$pack_mask[$args_per_word]}; - my(@pack_instr) = @{$pack_instr[$args_per_word]}; + my $args_per_word; + if ($packable_args < 4 or $wordsize == 64) { + $args_per_word = $packable_args; + } else { + # 4 packable argument, 32 bit wordsize. Need 2 words. + $args_per_word = 2; + } + + my @shift; + my @mask; + my @instr; + + if ($wide_packing) { + @shift = ('0', 'BEAM_WIDE_SHIFT'); + @mask = ('BEAM_WIDE_MASK', $WHOLE_WORD); + @instr = ('w', 'i'); + } else { + @shift = @{$pack_shift[$args_per_word]}; + @mask = @{$pack_mask[$args_per_word]}; + @instr = @{$pack_instr[$args_per_word]}; + } # # Now generate the packing instructions. One complication is that @@ -979,10 +1038,10 @@ sub do_pack { my($ap) = 0; # Argument number within word. my($tmpnum) = 1; # Number of temporary variable. my($expr) = ''; - for ($i = 0; $i < @args; $i++) { + for (my $i = 0; $i < @args; $i++) { my($reg) = $args[$i]; my($this_size) = $arg_size{$reg}; - if ($reg =~ /[xyt]/) { + if ($is_packable[$i]) { $this_size = 0; $did_some_packing = 1; @@ -993,7 +1052,7 @@ sub do_pack { $this_size = 1; } - $down = "$pack_instr[$ap]$down"; + $down = "$instr[$ap]$down"; my($unpack) = &make_unpack($tmpnum, $shift[$ap], $mask[$ap]); $args[$i] = "pack:$this_size:$reg" . "b($unpack)"; @@ -1103,6 +1162,10 @@ sub compile_transform { if ($obsolete[$gen_opnum{$name,$arity}]) { error("obsolete function must not be used in transformations"); } + + if ($src) { + $is_transformed{$name,$arity} = 1; + } [$name,$arity,@ops]; } @@ -1291,13 +1354,28 @@ sub tr_gen_from { my($var, $type, $type_val, $cond, $val) = @$op; if ($type ne '' && $type ne '*') { - my($types) = ''; - my($type_mask) = 0; - foreach (split('', $type)) { - $types .= "$_ "; - $type_mask |= $type_bit{$_}; + # + # The is_bif, is_not_bif, and is_func instructions have + # their own built-in type test and don't need to + # be guarded with a type test instruction. + # + unless ($cond eq 'is_bif' or + $cond eq 'is_not_bif' or + $cond eq 'is_func') { + my($types) = ''; + my($type_mask) = 0; + foreach (split('', $type)) { + $types .= "$_ "; + $type_mask |= $type_bit{$_}; + } + if ($cond ne 'is_eq') { + push(@code, &make_op($types, 'is_type', $type_mask)); + } else { + $cond = ''; + push(@code, &make_op($types, 'is_type_eq', + $type_mask, $val)); + } } - push(@code, &make_op($types, 'is_type', $type_mask)); } if ($cond eq 'is_func') { diff --git a/erts/emulator/utils/count b/erts/emulator/utils/count new file mode 100755 index 0000000000..617f5c25e8 --- /dev/null +++ b/erts/emulator/utils/count @@ -0,0 +1,127 @@ +%% -*- erlang -*- +%% +%% %CopyrightBegin% +%% +%% Copyright Ericsson AB 1998-2010. 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% +%% + +-mode(compile). + +main(_) -> + DisDir = "./dis", + ok = filelib:ensure_dir(filename:join(DisDir, "dummy")), + io:format("Dissambling to ~s\n", [DisDir]), + ok = file:set_cwd(DisDir), + Path = code:get_path() -- ["."], + Beams0 = [filelib:wildcard(filename:join(Dir, "*.beam")) || + Dir <- Path], + Beams = lists:append(Beams0), + Mods0 = [list_to_atom(filename:rootname(filename:basename(F))) || + F <- Beams], + Mods = lists:usort(Mods0), + start_sem(), + Ps = [begin + {_,Ref} = spawn_monitor(fun() -> count(M) end), + Ref + end || M <- Mods], + [put(list_to_atom(I), 0) || I <- erts_debug:instructions()], + Res = wait_for_all(Ps, 1), + OutFile = "count", + {ok,Out} = file:open(OutFile, [write]), + [io:format(Out, "~s ~p\n", [I,C]) || {I,C} <- Res], + ok = file:close(Out), + io:format("\nResult written to ~s\n", + [filename:join(DisDir, OutFile)]), + ok. + +wait_for_all([], _) -> + lists:reverse(lists:keysort(2, get())); +wait_for_all([_|_]=Ps, I) -> + receive + {'DOWN',Ref,process,_,Result} -> + io:format("\r~p", [I]), + [increment(Key, Count) || {Key,Count} <- Result], + wait_for_all(Ps -- [Ref], I+1) + end. + +count(M) -> + down(), + erts_debug:df(M), + {ok,Fd} = file:open(atom_to_list(M) ++ ".dis", [read,raw]), + count_is(Fd), + ok = file:close(Fd), + exit(get()). + +count_is(Fd) -> + case file:read_line(Fd) of + {ok,Line} -> + count_instr(Line), + count_is(Fd); + eof -> + ok + end. + +count_instr([$\s|T]) -> + count_instr_1(T, []); +count_instr([_|T]) -> + count_instr(T); +count_instr([]) -> + %% Empty line. + ok. + +count_instr_1([$\s|_], Acc) -> + Instr = list_to_atom(lists:reverse(Acc)), + increment(Instr, 1); +count_instr_1([H|T], Acc) -> + count_instr_1(T, [H|Acc]). + +increment(Key, Inc) -> + case get(Key) of + undefined -> + put(Key, Inc); + Count -> + put(Key, Count+Inc) + end. + +%%% +%%% Counting sempahore to limit the number of processes that +%%% can run concurrently. +%%% + +down() -> + sem ! {down,self()}, + receive + sem_taken -> ok + end. + +start_sem() -> + spawn(fun() -> + register(sem, self()), + process_flag(trap_exit, true), + do_sem(erlang:system_info(schedulers)+1) end). + +do_sem(0) -> + receive + {'EXIT',_,_} -> + do_sem(1) + end; +do_sem(C) -> + receive + {down,Pid} -> + link(Pid), + Pid ! sem_taken, + do_sem(C-1) + end. diff --git a/erts/epmd/src/epmd_srv.c b/erts/epmd/src/epmd_srv.c index ef471a473a..3499ab2934 100644 --- a/erts/epmd/src/epmd_srv.c +++ b/erts/epmd/src/epmd_srv.c @@ -2,7 +2,7 @@ /* * %CopyrightBegin% * - * Copyright Ericsson AB 1998-2010. All Rights Reserved. + * Copyright Ericsson AB 1998-2011. All Rights Reserved. * * The contents of this file are subject to the Erlang Public License, * Version 1.1, (the "License"); you may not use this file except in @@ -157,8 +157,10 @@ void run(EpmdVars *g) dbg_printf(g,2,"starting"); - listen(listensock, SOMAXCONN); - + if(listen(listensock, SOMAXCONN) < 0) { + dbg_perror(g,"failed to listen on socket"); + epmd_cleanup_exit(g,1); + } FD_ZERO(&g->orig_read_mask); FD_SET(listensock,&g->orig_read_mask); diff --git a/erts/include/internal/ethr_atomics.h b/erts/include/internal/ethr_atomics.h new file mode 100644 index 0000000000..1caf4d0567 --- /dev/null +++ b/erts/include/internal/ethr_atomics.h @@ -0,0 +1,726 @@ +/* + * %CopyrightBegin% + * + * Copyright Ericsson AB 2010. 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: The ethread atomic API + * Author: Rickard Green + */ + +#ifndef ETHR_ATOMIC_H__ +#define ETHR_ATOMIC_H__ + +#if !defined(ETHR_TRY_INLINE_FUNCS) || defined(ETHR_ATOMIC_IMPL__) +# define ETHR_NEED_ATOMIC_PROTOTYPES__ +#endif + +#ifndef ETHR_HAVE_NATIVE_ATOMICS +/* + * No native atomic implementation available. :( + * Use fallback... + */ +typedef ethr_sint32_t ethr_atomic32_t; +typedef ethr_sint_t ethr_atomic_t; +#else +/* + * Map ethread native atomics to ethread API atomics. + * + * We do at least have a native atomic implementation that + * can handle integers of a size larger than or equal to + * the size of pointers. + */ + +/* -- Pointer size atomics -- */ + +#undef ETHR_NAINT_T__ +#undef ETHR_NATMC_FUNC__ +#undef ETHR_NATMC_ADDR_FUNC__ +#if ETHR_SIZEOF_PTR == 8 +# if defined(ETHR_HAVE_NATIVE_ATOMIC64) +# define ETHR_NATMC_ADDR_FUNC__ ethr_native_atomic64_addr +typedef ethr_native_atomic64_t ethr_atomic_t; +# define ETHR_NAINT_T__ ethr_sint64_t +# define ETHR_NATMC_FUNC__(X) ethr_native_atomic64_ ## X +# else +# error "Missing native atomic implementation" +# endif +#elif ETHR_SIZEOF_PTR == 4 +# define ETHR_NATMC_ADDR_FUNC__ ethr_native_atomic32_addr +# ifdef ETHR_HAVE_NATIVE_ATOMIC32 +typedef ethr_native_atomic32_t ethr_atomic_t; +# define ETHR_NAINT_T__ ethr_sint32_t +# define ETHR_NATMC_FUNC__(X) ethr_native_atomic32_ ## X +# elif defined(ETHR_HAVE_NATIVE_ATOMIC64) +typedef ethr_native_atomic64_t ethr_atomic_t; +# define ETHR_NATMC_T__ ethr_native_atomic64_t +# define ETHR_NAINT_T__ ethr_sint64_t +# define ETHR_NATMC_FUNC__(X) ethr_native_atomic64_ ## X +# else +# error "Missing native atomic implementation" +# endif +#endif + +/* -- 32-bit atomics -- */ + +#undef ETHR_NAINT32_T__ +#undef ETHR_NATMC32_FUNC__ +#if defined(ETHR_HAVE_NATIVE_ATOMIC32) +typedef ethr_native_atomic32_t ethr_atomic32_t; +# define ETHR_NAINT32_T__ ethr_sint32_t +# define ETHR_NATMC32_FUNC__(X) ethr_native_atomic32_ ## X +#elif defined(ETHR_HAVE_NATIVE_ATOMIC64) +typedef ethr_native_atomic64_t ethr_atomic32_t; +# define ETHR_NAINT32_T__ ethr_sint64_t +# define ETHR_NATMC32_FUNC__(X) ethr_native_atomic64_ ## X +#else +# error "Missing native atomic implementation" +#endif + +#endif + +#ifdef ETHR_NEED_ATOMIC_PROTOTYPES__ +ethr_sint_t *ethr_atomic_addr(ethr_atomic_t *); +void ethr_atomic_init(ethr_atomic_t *, ethr_sint_t); +void ethr_atomic_set(ethr_atomic_t *, ethr_sint_t); +ethr_sint_t ethr_atomic_read(ethr_atomic_t *); +ethr_sint_t ethr_atomic_inc_read(ethr_atomic_t *); +ethr_sint_t ethr_atomic_dec_read(ethr_atomic_t *); +void ethr_atomic_inc(ethr_atomic_t *); +void ethr_atomic_dec(ethr_atomic_t *); +ethr_sint_t ethr_atomic_add_read(ethr_atomic_t *, ethr_sint_t); +void ethr_atomic_add(ethr_atomic_t *, ethr_sint_t); +ethr_sint_t ethr_atomic_read_band(ethr_atomic_t *, ethr_sint_t); +ethr_sint_t ethr_atomic_read_bor(ethr_atomic_t *, ethr_sint_t); +ethr_sint_t ethr_atomic_xchg(ethr_atomic_t *, ethr_sint_t); +ethr_sint_t ethr_atomic_cmpxchg(ethr_atomic_t *, ethr_sint_t, ethr_sint_t); +ethr_sint_t ethr_atomic_read_acqb(ethr_atomic_t *); +ethr_sint_t ethr_atomic_inc_read_acqb(ethr_atomic_t *); +void ethr_atomic_set_relb(ethr_atomic_t *, ethr_sint_t); +void ethr_atomic_dec_relb(ethr_atomic_t *); +ethr_sint_t ethr_atomic_dec_read_relb(ethr_atomic_t *); +ethr_sint_t ethr_atomic_cmpxchg_acqb(ethr_atomic_t *, ethr_sint_t, ethr_sint_t); +ethr_sint_t ethr_atomic_cmpxchg_relb(ethr_atomic_t *, ethr_sint_t, ethr_sint_t); + +ethr_sint32_t *ethr_atomic32_addr(ethr_atomic32_t *); +void ethr_atomic32_init(ethr_atomic32_t *, ethr_sint32_t); +void ethr_atomic32_set(ethr_atomic32_t *, ethr_sint32_t); +ethr_sint32_t ethr_atomic32_read(ethr_atomic32_t *); +ethr_sint32_t ethr_atomic32_inc_read(ethr_atomic32_t *); +ethr_sint32_t ethr_atomic32_dec_read(ethr_atomic32_t *); +void ethr_atomic32_inc(ethr_atomic32_t *); +void ethr_atomic32_dec(ethr_atomic32_t *); +ethr_sint32_t ethr_atomic32_add_read(ethr_atomic32_t *, ethr_sint32_t); +void ethr_atomic32_add(ethr_atomic32_t *, ethr_sint32_t); +ethr_sint32_t ethr_atomic32_read_band(ethr_atomic32_t *, ethr_sint32_t); +ethr_sint32_t ethr_atomic32_read_bor(ethr_atomic32_t *, ethr_sint32_t); +ethr_sint32_t ethr_atomic32_xchg(ethr_atomic32_t *, ethr_sint32_t); +ethr_sint32_t ethr_atomic32_cmpxchg(ethr_atomic32_t *, + ethr_sint32_t, + ethr_sint32_t); +ethr_sint32_t ethr_atomic32_read_acqb(ethr_atomic32_t *); +ethr_sint32_t ethr_atomic32_inc_read_acqb(ethr_atomic32_t *); +void ethr_atomic32_set_relb(ethr_atomic32_t *, ethr_sint32_t); +void ethr_atomic32_dec_relb(ethr_atomic32_t *); +ethr_sint32_t ethr_atomic32_dec_read_relb(ethr_atomic32_t *); +ethr_sint32_t ethr_atomic32_cmpxchg_acqb(ethr_atomic32_t *, + ethr_sint32_t, + ethr_sint32_t); +ethr_sint32_t ethr_atomic32_cmpxchg_relb(ethr_atomic32_t *, + ethr_sint32_t, + ethr_sint32_t); +#endif + +int ethr_init_atomics(void); + +#if defined(ETHR_TRY_INLINE_FUNCS) || defined(ETHR_ATOMIC_IMPL__) + +#ifndef ETHR_HAVE_NATIVE_ATOMICS +/* + * Fallbacks for atomics used in absence of a native implementation. + */ + +#define ETHR_ATOMIC_ADDR_BITS 10 +#define ETHR_ATOMIC_ADDR_SHIFT 6 + +typedef struct { + union { + ethr_spinlock_t lck; + char buf[ETHR_CACHE_LINE_SIZE]; + } u; +} ethr_atomic_protection_t; + +extern ethr_atomic_protection_t ethr_atomic_protection__[1 << ETHR_ATOMIC_ADDR_BITS]; + +#define ETHR_ATOMIC_PTR2LCK__(PTR) \ +(ðr_atomic_protection__[((((ethr_uint_t) (PTR)) >> ETHR_ATOMIC_ADDR_SHIFT) \ + & ((1 << ETHR_ATOMIC_ADDR_BITS) - 1))].u.lck) + + +#define ETHR_ATOMIC_OP_FALLBACK_IMPL__(AP, EXPS) \ +do { \ + ethr_spinlock_t *slp__ = ETHR_ATOMIC_PTR2LCK__((AP)); \ + ethr_spin_lock(slp__); \ + { EXPS; } \ + ethr_spin_unlock(slp__); \ +} while (0) + +#endif + +/* + * --- Pointer size atomics --------------------------------------------------- + */ + +static ETHR_INLINE ethr_sint_t * +ETHR_INLINE_FUNC_NAME_(ethr_atomic_addr)(ethr_atomic_t *var) +{ +#ifdef ETHR_HAVE_NATIVE_ATOMICS + return (ethr_sint_t *) ETHR_NATMC_ADDR_FUNC__(var); +#else + return (ethr_sint_t *) var; +#endif +} + +static ETHR_INLINE void +ETHR_INLINE_FUNC_NAME_(ethr_atomic_init)(ethr_atomic_t *var, ethr_sint_t i) +{ +#ifdef ETHR_HAVE_NATIVE_ATOMICS + ETHR_NATMC_FUNC__(init)(var, (ETHR_NAINT_T__) i); +#else + ETHR_ATOMIC_OP_FALLBACK_IMPL__(var, *var = i); +#endif +} + +static ETHR_INLINE void +ETHR_INLINE_FUNC_NAME_(ethr_atomic_set)(ethr_atomic_t *var, ethr_sint_t i) +{ +#ifdef ETHR_HAVE_NATIVE_ATOMICS + ETHR_NATMC_FUNC__(set)(var, (ETHR_NAINT_T__) i); +#else + ETHR_ATOMIC_OP_FALLBACK_IMPL__(var, *var = i); +#endif +} + +static ETHR_INLINE ethr_sint_t +ETHR_INLINE_FUNC_NAME_(ethr_atomic_read)(ethr_atomic_t *var) +{ +#ifdef ETHR_HAVE_NATIVE_ATOMICS + return (ethr_sint_t) ETHR_NATMC_FUNC__(read)(var); +#else + ethr_sint_t res; + ETHR_ATOMIC_OP_FALLBACK_IMPL__(var, res = *var); + return res; +#endif +} + +static ETHR_INLINE void +ETHR_INLINE_FUNC_NAME_(ethr_atomic_add)(ethr_atomic_t *var, ethr_sint_t incr) +{ +#ifdef ETHR_HAVE_NATIVE_ATOMICS + ETHR_NATMC_FUNC__(add)(var, (ETHR_NAINT_T__) incr); +#else + ETHR_ATOMIC_OP_FALLBACK_IMPL__(var, *var += incr); +#endif +} + +static ETHR_INLINE ethr_sint_t +ETHR_INLINE_FUNC_NAME_(ethr_atomic_add_read)(ethr_atomic_t *var, ethr_sint_t i) +{ +#ifdef ETHR_HAVE_NATIVE_ATOMICS + return (ethr_sint_t) ETHR_NATMC_FUNC__(add_return)(var, (ETHR_NAINT_T__) i); +#else + ethr_sint_t res; + ETHR_ATOMIC_OP_FALLBACK_IMPL__(var, *var += i; res = *var); + return res; +#endif +} + +static ETHR_INLINE void +ETHR_INLINE_FUNC_NAME_(ethr_atomic_inc)(ethr_atomic_t *var) +{ +#ifdef ETHR_HAVE_NATIVE_ATOMICS + ETHR_NATMC_FUNC__(inc)(var); +#else + ETHR_ATOMIC_OP_FALLBACK_IMPL__(var, ++(*var)); +#endif +} + +static ETHR_INLINE void +ETHR_INLINE_FUNC_NAME_(ethr_atomic_dec)(ethr_atomic_t *var) +{ +#ifdef ETHR_HAVE_NATIVE_ATOMICS + ETHR_NATMC_FUNC__(dec)(var); +#else + ETHR_ATOMIC_OP_FALLBACK_IMPL__(var, --(*var)); +#endif +} + +static ETHR_INLINE ethr_sint_t +ETHR_INLINE_FUNC_NAME_(ethr_atomic_inc_read)(ethr_atomic_t *var) +{ +#ifdef ETHR_HAVE_NATIVE_ATOMICS + return (ethr_sint_t) ETHR_NATMC_FUNC__(inc_return)(var); +#else + ethr_sint_t res; + ETHR_ATOMIC_OP_FALLBACK_IMPL__(var, res = ++(*var)); + return res; +#endif +} + +static ETHR_INLINE ethr_sint_t +ETHR_INLINE_FUNC_NAME_(ethr_atomic_dec_read)(ethr_atomic_t *var) +{ +#ifdef ETHR_HAVE_NATIVE_ATOMICS + return (ethr_sint_t) ETHR_NATMC_FUNC__(dec_return)(var); +#else + ethr_sint_t res; + ETHR_ATOMIC_OP_FALLBACK_IMPL__(var, res = --(*var)); + return res; +#endif +} + +static ETHR_INLINE ethr_sint_t +ETHR_INLINE_FUNC_NAME_(ethr_atomic_read_band)(ethr_atomic_t *var, + ethr_sint_t mask) +{ +#ifdef ETHR_HAVE_NATIVE_ATOMICS + return (ethr_sint_t) ETHR_NATMC_FUNC__(and_retold)(var, + (ETHR_NAINT_T__) mask); +#else + ethr_sint_t res; + ETHR_ATOMIC_OP_FALLBACK_IMPL__(var, res = *var; *var &= mask); + return res; +#endif +} + +static ETHR_INLINE ethr_sint_t +ETHR_INLINE_FUNC_NAME_(ethr_atomic_read_bor)(ethr_atomic_t *var, + ethr_sint_t mask) +{ +#ifdef ETHR_HAVE_NATIVE_ATOMICS + return (ethr_sint_t) ETHR_NATMC_FUNC__(or_retold)(var, + (ETHR_NAINT_T__) mask); +#else + ethr_sint_t res; + ETHR_ATOMIC_OP_FALLBACK_IMPL__(var, res = *var; *var |= mask); + return res; +#endif +} + +static ETHR_INLINE ethr_sint_t +ETHR_INLINE_FUNC_NAME_(ethr_atomic_xchg)(ethr_atomic_t *var, ethr_sint_t new) +{ +#ifdef ETHR_HAVE_NATIVE_ATOMICS + return (ethr_sint_t) ETHR_NATMC_FUNC__(xchg)(var, + (ETHR_NAINT_T__) new); +#else + ethr_sint_t res; + ETHR_ATOMIC_OP_FALLBACK_IMPL__(var, res = *var; *var = new); + return res; +#endif +} + +static ETHR_INLINE ethr_sint_t +ETHR_INLINE_FUNC_NAME_(ethr_atomic_cmpxchg)(ethr_atomic_t *var, + ethr_sint_t new, + ethr_sint_t exp) +{ +#ifdef ETHR_HAVE_NATIVE_ATOMICS + return (ethr_sint_t) ETHR_NATMC_FUNC__(cmpxchg)(var, + (ETHR_NAINT_T__) new, + (ETHR_NAINT_T__) exp); +#else + ethr_sint_t res; + ETHR_ATOMIC_OP_FALLBACK_IMPL__(var, + { + res = *var; + if (__builtin_expect(res == exp, 1)) + *var = new; + }); + return res; +#endif +} + +/* + * Important memory barrier requirements. + * + * The following atomic operations *must* supply a memory barrier of + * at least the type specified by its suffix: + * _acqb = acquire barrier + * _relb = release barrier + */ + +static ETHR_INLINE ethr_sint_t +ETHR_INLINE_FUNC_NAME_(ethr_atomic_read_acqb)(ethr_atomic_t *var) +{ +#ifdef ETHR_HAVE_NATIVE_ATOMICS + return (ethr_sint_t) ETHR_NATMC_FUNC__(read_acqb)(var); +#else + return ETHR_INLINE_FUNC_NAME_(ethr_atomic_read)(var); +#endif +} + +static ETHR_INLINE ethr_sint_t +ETHR_INLINE_FUNC_NAME_(ethr_atomic_inc_read_acqb)(ethr_atomic_t *var) +{ +#ifdef ETHR_HAVE_NATIVE_ATOMICS + return (ethr_sint_t) ETHR_NATMC_FUNC__(inc_return_acqb)(var); +#else + return ETHR_INLINE_FUNC_NAME_(ethr_atomic_inc_read)(var); +#endif +} + +static ETHR_INLINE void +ETHR_INLINE_FUNC_NAME_(ethr_atomic_set_relb)(ethr_atomic_t *var, + ethr_sint_t val) +{ +#ifdef ETHR_HAVE_NATIVE_ATOMICS + ETHR_NATMC_FUNC__(set_relb)(var, (ETHR_NAINT_T__) val); +#else + ETHR_INLINE_FUNC_NAME_(ethr_atomic_set)(var, val); +#endif +} + +static ETHR_INLINE void +ETHR_INLINE_FUNC_NAME_(ethr_atomic_dec_relb)(ethr_atomic_t *var) +{ +#ifdef ETHR_HAVE_NATIVE_ATOMICS + ETHR_NATMC_FUNC__(dec_relb)(var); +#else + ETHR_INLINE_FUNC_NAME_(ethr_atomic_dec)(var); +#endif +} + +static ETHR_INLINE ethr_sint_t +ETHR_INLINE_FUNC_NAME_(ethr_atomic_dec_read_relb)(ethr_atomic_t *var) +{ +#ifdef ETHR_HAVE_NATIVE_ATOMICS + return (ethr_sint_t) ETHR_NATMC_FUNC__(dec_return_relb)(var); +#else + return ETHR_INLINE_FUNC_NAME_(ethr_atomic_dec_read)(var); +#endif +} + +static ETHR_INLINE ethr_sint_t +ETHR_INLINE_FUNC_NAME_(ethr_atomic_cmpxchg_acqb)(ethr_atomic_t *var, + ethr_sint_t new, + ethr_sint_t exp) +{ +#ifdef ETHR_HAVE_NATIVE_ATOMICS + return (ethr_sint_t) ETHR_NATMC_FUNC__(cmpxchg_acqb)(var, + (ETHR_NAINT_T__) new, + (ETHR_NAINT_T__) exp); +#else + return ETHR_INLINE_FUNC_NAME_(ethr_atomic_cmpxchg)(var, new, exp); +#endif +} + +static ETHR_INLINE ethr_sint_t +ETHR_INLINE_FUNC_NAME_(ethr_atomic_cmpxchg_relb)(ethr_atomic_t *var, + ethr_sint_t new, + ethr_sint_t exp) +{ +#ifdef ETHR_HAVE_NATIVE_ATOMICS + return (ethr_sint_t) ETHR_NATMC_FUNC__(cmpxchg_relb)(var, + (ETHR_NAINT_T__) new, + (ETHR_NAINT_T__) exp); +#else + return ETHR_INLINE_FUNC_NAME_(ethr_atomic_cmpxchg)(var, new, exp); +#endif +} + +/* + * --- 32-bit atomics --------------------------------------------------------- + */ + +static ETHR_INLINE ethr_sint32_t * +ETHR_INLINE_FUNC_NAME_(ethr_atomic32_addr)(ethr_atomic32_t *var) +{ +#ifdef ETHR_HAVE_NATIVE_ATOMICS + return ethr_native_atomic32_addr(var); +#else + return (ethr_sint32_t *) var; +#endif +} + +static ETHR_INLINE void +ETHR_INLINE_FUNC_NAME_(ethr_atomic32_init)(ethr_atomic32_t *var, + ethr_sint32_t i) +{ +#ifdef ETHR_HAVE_NATIVE_ATOMICS + ETHR_NATMC32_FUNC__(init)(var, (ETHR_NAINT32_T__) i); +#else + ETHR_ATOMIC_OP_FALLBACK_IMPL__(var, *var = i); +#endif +} + +static ETHR_INLINE void +ETHR_INLINE_FUNC_NAME_(ethr_atomic32_set)(ethr_atomic32_t *var, ethr_sint32_t i) +{ +#ifdef ETHR_HAVE_NATIVE_ATOMICS + ETHR_NATMC32_FUNC__(set)(var, (ETHR_NAINT32_T__) i); +#else + ETHR_ATOMIC_OP_FALLBACK_IMPL__(var, *var = i); +#endif +} + +static ETHR_INLINE ethr_sint32_t +ETHR_INLINE_FUNC_NAME_(ethr_atomic32_read)(ethr_atomic32_t *var) +{ +#ifdef ETHR_HAVE_NATIVE_ATOMICS + return (ethr_sint32_t) ETHR_NATMC32_FUNC__(read)(var); +#else + ethr_sint32_t res; + ETHR_ATOMIC_OP_FALLBACK_IMPL__(var, res = *var); + return res; +#endif +} + +static ETHR_INLINE void +ETHR_INLINE_FUNC_NAME_(ethr_atomic32_add)(ethr_atomic32_t *var, + ethr_sint32_t incr) +{ +#ifdef ETHR_HAVE_NATIVE_ATOMICS + ETHR_NATMC32_FUNC__(add)(var, (ETHR_NAINT32_T__) incr); +#else + ETHR_ATOMIC_OP_FALLBACK_IMPL__(var, *var += incr); +#endif +} + +static ETHR_INLINE ethr_sint32_t +ETHR_INLINE_FUNC_NAME_(ethr_atomic32_add_read)(ethr_atomic32_t *var, + ethr_sint32_t i) +{ +#ifdef ETHR_HAVE_NATIVE_ATOMICS + return (ethr_sint32_t) + ETHR_NATMC32_FUNC__(add_return)(var, (ETHR_NAINT32_T__) i); +#else + ethr_sint32_t res; + ETHR_ATOMIC_OP_FALLBACK_IMPL__(var, *var += i; res = *var); + return res; +#endif +} + +static ETHR_INLINE void +ETHR_INLINE_FUNC_NAME_(ethr_atomic32_inc)(ethr_atomic32_t *var) +{ +#ifdef ETHR_HAVE_NATIVE_ATOMICS + ETHR_NATMC32_FUNC__(inc)(var); +#else + ETHR_ATOMIC_OP_FALLBACK_IMPL__(var, ++(*var)); +#endif +} + +static ETHR_INLINE void +ETHR_INLINE_FUNC_NAME_(ethr_atomic32_dec)(ethr_atomic32_t *var) +{ +#ifdef ETHR_HAVE_NATIVE_ATOMICS + ETHR_NATMC32_FUNC__(dec)(var); +#else + ETHR_ATOMIC_OP_FALLBACK_IMPL__(var, --(*var)); +#endif +} + +static ETHR_INLINE ethr_sint32_t +ETHR_INLINE_FUNC_NAME_(ethr_atomic32_inc_read)(ethr_atomic32_t *var) +{ +#ifdef ETHR_HAVE_NATIVE_ATOMICS + return (ethr_sint32_t) ETHR_NATMC32_FUNC__(inc_return)(var); +#else + ethr_sint32_t res; + ETHR_ATOMIC_OP_FALLBACK_IMPL__(var, res = ++(*var)); + return res; +#endif +} + +static ETHR_INLINE ethr_sint32_t +ETHR_INLINE_FUNC_NAME_(ethr_atomic32_dec_read)(ethr_atomic32_t *var) +{ +#ifdef ETHR_HAVE_NATIVE_ATOMICS + return (ethr_sint32_t) ETHR_NATMC32_FUNC__(dec_return)(var); +#else + ethr_sint32_t res; + ETHR_ATOMIC_OP_FALLBACK_IMPL__(var, res = --(*var)); + return res; +#endif +} + +static ETHR_INLINE ethr_sint32_t +ETHR_INLINE_FUNC_NAME_(ethr_atomic32_read_band)(ethr_atomic32_t *var, + ethr_sint32_t mask) +{ +#ifdef ETHR_HAVE_NATIVE_ATOMICS + return (ethr_sint32_t) + ETHR_NATMC32_FUNC__(and_retold)(var, (ETHR_NAINT32_T__) mask); +#else + ethr_sint32_t res; + ETHR_ATOMIC_OP_FALLBACK_IMPL__(var, res = *var; *var &= mask); + return res; +#endif +} + +static ETHR_INLINE ethr_sint32_t +ETHR_INLINE_FUNC_NAME_(ethr_atomic32_read_bor)(ethr_atomic32_t *var, + ethr_sint32_t mask) +{ +#ifdef ETHR_HAVE_NATIVE_ATOMICS + return + (ethr_sint32_t) ETHR_NATMC32_FUNC__(or_retold)(var, + (ETHR_NAINT32_T__) mask); +#else + ethr_sint32_t res; + ETHR_ATOMIC_OP_FALLBACK_IMPL__(var, res = *var; *var |= mask); + return res; +#endif +} + +static ETHR_INLINE ethr_sint32_t +ETHR_INLINE_FUNC_NAME_(ethr_atomic32_xchg)(ethr_atomic32_t *var, + ethr_sint32_t new) +{ +#ifdef ETHR_HAVE_NATIVE_ATOMICS + return (ethr_sint32_t) ETHR_NATMC32_FUNC__(xchg)(var, + (ETHR_NAINT32_T__) new); +#else + ethr_sint32_t res; + ETHR_ATOMIC_OP_FALLBACK_IMPL__(var, res = *var; *var = new); + return res; +#endif +} + +static ETHR_INLINE ethr_sint32_t +ETHR_INLINE_FUNC_NAME_(ethr_atomic32_cmpxchg)(ethr_atomic32_t *var, + ethr_sint32_t new, + ethr_sint32_t exp) +{ +#ifdef ETHR_HAVE_NATIVE_ATOMICS + return (ethr_sint32_t) ETHR_NATMC32_FUNC__(cmpxchg)(var, + (ETHR_NAINT32_T__) new, + (ETHR_NAINT32_T__) exp); +#else + ethr_sint32_t res; + ETHR_ATOMIC_OP_FALLBACK_IMPL__(var, + { + res = *var; + if (__builtin_expect(res == exp, 1)) + *var = new; + }); + return res; +#endif +} + +/* + * Important memory barrier requirements. + * + * The following atomic operations *must* supply a memory barrier of + * at least the type specified by its suffix: + * _acqb = acquire barrier + * _relb = release barrier + */ + +static ETHR_INLINE ethr_sint32_t +ETHR_INLINE_FUNC_NAME_(ethr_atomic32_read_acqb)(ethr_atomic32_t *var) +{ +#ifdef ETHR_HAVE_NATIVE_ATOMICS + return (ethr_sint32_t) ETHR_NATMC32_FUNC__(read_acqb)(var); +#else + return ETHR_INLINE_FUNC_NAME_(ethr_atomic32_read)(var); +#endif +} + +static ETHR_INLINE ethr_sint32_t +ETHR_INLINE_FUNC_NAME_(ethr_atomic32_inc_read_acqb)(ethr_atomic32_t *var) +{ +#ifdef ETHR_HAVE_NATIVE_ATOMICS + return (ethr_sint32_t) ETHR_NATMC32_FUNC__(inc_return_acqb)(var); +#else + return ETHR_INLINE_FUNC_NAME_(ethr_atomic32_inc_read)(var); +#endif +} + +static ETHR_INLINE void +ETHR_INLINE_FUNC_NAME_(ethr_atomic32_set_relb)(ethr_atomic32_t *var, + ethr_sint32_t val) +{ +#ifdef ETHR_HAVE_NATIVE_ATOMICS + ETHR_NATMC32_FUNC__(set_relb)(var, (ETHR_NAINT32_T__) val); +#else + ETHR_INLINE_FUNC_NAME_(ethr_atomic32_set)(var, val); +#endif +} + +static ETHR_INLINE void +ETHR_INLINE_FUNC_NAME_(ethr_atomic32_dec_relb)(ethr_atomic32_t *var) +{ +#ifdef ETHR_HAVE_NATIVE_ATOMICS + ETHR_NATMC32_FUNC__(dec_relb)(var); +#else + ETHR_INLINE_FUNC_NAME_(ethr_atomic32_dec)(var); +#endif +} + +static ETHR_INLINE ethr_sint32_t +ETHR_INLINE_FUNC_NAME_(ethr_atomic32_dec_read_relb)(ethr_atomic32_t *var) +{ +#ifdef ETHR_HAVE_NATIVE_ATOMICS + return (ethr_sint32_t) ETHR_NATMC32_FUNC__(dec_return_relb)(var); +#else + return ETHR_INLINE_FUNC_NAME_(ethr_atomic32_dec_read)(var); +#endif +} + +static ETHR_INLINE ethr_sint32_t +ETHR_INLINE_FUNC_NAME_(ethr_atomic32_cmpxchg_acqb)(ethr_atomic32_t *var, + ethr_sint32_t new, + ethr_sint32_t exp) +{ +#ifdef ETHR_HAVE_NATIVE_ATOMICS + return (ethr_sint32_t) + ETHR_NATMC32_FUNC__(cmpxchg_acqb)(var, + (ETHR_NAINT32_T__) new, + (ETHR_NAINT32_T__) exp); +#else + return ETHR_INLINE_FUNC_NAME_(ethr_atomic32_cmpxchg)(var, new, exp); +#endif +} + +static ETHR_INLINE ethr_sint32_t +ETHR_INLINE_FUNC_NAME_(ethr_atomic32_cmpxchg_relb)(ethr_atomic32_t *var, + ethr_sint32_t new, + ethr_sint32_t exp) +{ +#ifdef ETHR_HAVE_NATIVE_ATOMICS + return (ethr_sint32_t) + ETHR_NATMC32_FUNC__(cmpxchg_relb)(var, + (ETHR_NAINT32_T__) new, + (ETHR_NAINT32_T__) exp); +#else + return ETHR_INLINE_FUNC_NAME_(ethr_atomic32_cmpxchg)(var, new, exp); +#endif +} + + +#endif /* ETHR_TRY_INLINE_FUNCS */ + +#undef ETHR_NAINT_T__ +#undef ETHR_NATMC_FUNC__ +#undef ETHR_NATMC_ADDR_FUNC__ + +#undef ETHR_NAINT32_T__ +#undef ETHR_NATMC32_FUNC__ + +#endif diff --git a/erts/include/internal/ethr_mutex.h b/erts/include/internal/ethr_mutex.h index 01855864e3..fadaf1e2a4 100644 --- a/erts/include/internal/ethr_mutex.h +++ b/erts/include/internal/ethr_mutex.h @@ -78,13 +78,13 @@ # error Need a qlock implementation #endif -#define ETHR_RWMTX_W_FLG__ (((long) 1) << 31) -#define ETHR_RWMTX_W_WAIT_FLG__ (((long) 1) << 30) -#define ETHR_RWMTX_R_WAIT_FLG__ (((long) 1) << 29) +#define ETHR_RWMTX_W_FLG__ (((ethr_sint32_t) 1) << 31) +#define ETHR_RWMTX_W_WAIT_FLG__ (((ethr_sint32_t) 1) << 30) +#define ETHR_RWMTX_R_WAIT_FLG__ (((ethr_sint32_t) 1) << 29) /* frequent read kind */ -#define ETHR_RWMTX_R_FLG__ (((long) 1) << 28) -#define ETHR_RWMTX_R_ABRT_UNLCK_FLG__ (((long) 1) << 27) +#define ETHR_RWMTX_R_FLG__ (((ethr_sint32_t) 1) << 28) +#define ETHR_RWMTX_R_ABRT_UNLCK_FLG__ (((ethr_sint32_t) 1) << 27) #define ETHR_RWMTX_R_PEND_UNLCK_MASK__ (ETHR_RWMTX_R_ABRT_UNLCK_FLG__ - 1) /* normal kind */ @@ -106,28 +106,28 @@ #endif #define ETHR_MTX_DBG_CHK_UNUSED_FLG_BITS(MTX) \ - ETHR_DBG_CHK_UNUSED_FLG_BITS(ethr_atomic_read(&(MTX)->mtxb.flgs)) + ETHR_DBG_CHK_UNUSED_FLG_BITS(ethr_atomic32_read(&(MTX)->mtxb.flgs)) struct ethr_mutex_base_ { #ifdef ETHR_MTX_HARD_DEBUG_FENCE long pre_fence; #endif - ethr_atomic_t flgs; - ETHR_MTX_QLOCK_TYPE__ qlck; - ethr_ts_event *q; + ethr_atomic32_t flgs; short aux_scnt; short main_scnt; + ETHR_MTX_QLOCK_TYPE__ qlck; + ethr_ts_event *q; #ifdef ETHR_MTX_HARD_DEBUG_WSQ int ws; #endif #ifdef ETHR_MTX_CHK_EXCL - ethr_atomic_t exclusive; + ethr_atomic32_t exclusive; #endif #ifdef ETHR_MTX_CHK_NON_EXCL - ethr_atomic_t non_exclusive; + ethr_atomic32_t non_exclusive; #endif #ifdef ETHR_MTX_HARD_DEBUG_LFS - ethr_atomic_t hdbg_lfs; + ethr_atomic32_t hdbg_lfs; #endif }; @@ -236,7 +236,7 @@ typedef struct { typedef union { struct { - ethr_atomic_t readers; + ethr_atomic32_t readers; int waiting_readers; int byte_offset; ethr_rwmutex_lived lived; @@ -298,13 +298,13 @@ void ethr_rwmutex_rwunlock(ethr_rwmutex *); #ifdef ETHR_MTX_HARD_DEBUG_LFS # define ETHR_MTX_HARD_DEBUG_LFS_INIT(MTXB) \ do { \ - ethr_atomic_init(&(MTXB)->hdbg_lfs, 0); \ + ethr_atomic32_init(&(MTXB)->hdbg_lfs, 0); \ } while (0) # define ETHR_MTX_HARD_DEBUG_LFS_RLOCK(MTXB) \ do { \ - long val__; \ + ethr_sint32_t val__; \ ETHR_COMPILER_BARRIER; \ - val__ = ethr_atomic_inc_read(&(MTXB)->hdbg_lfs); \ + val__ = ethr_atomic32_inc_read(&(MTXB)->hdbg_lfs); \ ETHR_MTX_HARD_ASSERT(val__ > 0); \ } while (0) # define ETHR_MTX_HARD_DEBUG_LFS_TRYRLOCK(MTXB, RES) \ @@ -317,15 +317,15 @@ do { \ } while (0) # define ETHR_MTX_HARD_DEBUG_LFS_RUNLOCK(MTXB) \ do { \ - long val__ = ethr_atomic_dec_read(&(MTXB)->hdbg_lfs); \ + ethr_sint32_t val__ = ethr_atomic32_dec_read(&(MTXB)->hdbg_lfs); \ ETHR_MTX_HARD_ASSERT(val__ >= 0); \ ETHR_COMPILER_BARRIER; \ } while (0) # define ETHR_MTX_HARD_DEBUG_LFS_RWLOCK(MTXB) \ do { \ - long val__; \ + ethr_sint32_t val__; \ ETHR_COMPILER_BARRIER; \ - val__ = ethr_atomic_dec_read(&(MTXB)->hdbg_lfs); \ + val__ = ethr_atomic32_dec_read(&(MTXB)->hdbg_lfs); \ ETHR_MTX_HARD_ASSERT(val__ == -1); \ } while (0) # define ETHR_MTX_HARD_DEBUG_LFS_TRYRWLOCK(MTXB, RES) \ @@ -338,7 +338,7 @@ do { \ } while (0) # define ETHR_MTX_HARD_DEBUG_LFS_RWUNLOCK(MTXB) \ do { \ - long val__ = ethr_atomic_inctest(&(MTXB)->hdbg_lfs); \ + ethr_sint32_t val__ = ethr_atomic32_inctest(&(MTXB)->hdbg_lfs); \ ETHR_MTX_HARD_ASSERT(val__ == 0); \ ETHR_COMPILER_BARRIER; \ } while (0) @@ -386,12 +386,12 @@ do { \ #endif # define ETHR_MTX_CHK_EXCL_INIT__(MTXB) \ - ethr_atomic_init(&(MTXB)->exclusive, 0) + ethr_atomic32_init(&(MTXB)->exclusive, 0) # define ETHR_MTX_CHK_EXCL_IS_EXCL(MTXB) \ do { \ ETHR_COMPILER_BARRIER; \ - if (!ethr_atomic_read(&(MTXB)->exclusive)) \ + if (!ethr_atomic32_read(&(MTXB)->exclusive)) \ ethr_assert_failed(__FILE__, __LINE__, __func__,\ "is exclusive"); \ ETHR_COMPILER_BARRIER; \ @@ -399,7 +399,7 @@ do { \ # define ETHR_MTX_CHK_EXCL_IS_NOT_EXCL(MTXB) \ do { \ ETHR_COMPILER_BARRIER; \ - if (ethr_atomic_read(&(MTXB)->exclusive)) \ + if (ethr_atomic32_read(&(MTXB)->exclusive)) \ ethr_assert_failed(__FILE__, __LINE__, __func__,\ "is not exclusive"); \ ETHR_COMPILER_BARRIER; \ @@ -407,13 +407,13 @@ do { \ # define ETHR_MTX_CHK_EXCL_SET_EXCL(MTXB) \ do { \ ETHR_MTX_CHK_EXCL_IS_NOT_EXCL((MTXB)); \ - ethr_atomic_set(&(MTXB)->exclusive, 1); \ + ethr_atomic32_set(&(MTXB)->exclusive, 1); \ ETHR_COMPILER_BARRIER; \ } while (0) # define ETHR_MTX_CHK_EXCL_UNSET_EXCL(MTXB) \ do { \ ETHR_MTX_CHK_EXCL_IS_EXCL((MTXB)); \ - ethr_atomic_set(&(MTXB)->exclusive, 0); \ + ethr_atomic32_set(&(MTXB)->exclusive, 0); \ ETHR_COMPILER_BARRIER; \ } while (0) @@ -424,11 +424,11 @@ do { \ #endif # define ETHR_MTX_CHK_NON_EXCL_INIT__(MTXB) \ - ethr_atomic_init(&(MTXB)->non_exclusive, 0) + ethr_atomic32_init(&(MTXB)->non_exclusive, 0) # define ETHR_MTX_CHK_EXCL_IS_NON_EXCL(MTXB) \ do { \ ETHR_COMPILER_BARRIER; \ - if (!ethr_atomic_read(&(MTXB)->non_exclusive)) \ + if (!ethr_atomic32_read(&(MTXB)->non_exclusive)) \ ethr_assert_failed(__FILE__, __LINE__, __func__,\ "is non-exclusive"); \ ETHR_COMPILER_BARRIER; \ @@ -436,7 +436,7 @@ do { \ # define ETHR_MTX_CHK_EXCL_IS_NOT_NON_EXCL(MTXB) \ do { \ ETHR_COMPILER_BARRIER; \ - if (ethr_atomic_read(&(MTXB)->non_exclusive)) \ + if (ethr_atomic32_read(&(MTXB)->non_exclusive)) \ ethr_assert_failed(__FILE__, __LINE__, __func__,\ "is not non-exclusive"); \ ETHR_COMPILER_BARRIER; \ @@ -444,19 +444,19 @@ do { \ # define ETHR_MTX_CHK_EXCL_SET_NON_EXCL(MTXB) \ do { \ ETHR_COMPILER_BARRIER; \ - ethr_atomic_inc(&(MTXB)->non_exclusive); \ + ethr_atomic32_inc(&(MTXB)->non_exclusive); \ ETHR_COMPILER_BARRIER; \ } while (0) # define ETHR_MTX_CHK_EXCL_SET_NON_EXCL_NO(MTXB, NO) \ do { \ ETHR_COMPILER_BARRIER; \ - ethr_atomic_add(&(MTXB)->non_exclusive, (NO)); \ + ethr_atomic32_add(&(MTXB)->non_exclusive, (NO)); \ ETHR_COMPILER_BARRIER; \ } while (0) # define ETHR_MTX_CHK_EXCL_UNSET_NON_EXCL(MTXB) \ do { \ ETHR_COMPILER_BARRIER; \ - ethr_atomic_dec(&(MTXB)->non_exclusive); \ + ethr_atomic32_dec(&(MTXB)->non_exclusive); \ ETHR_COMPILER_BARRIER; \ } while (0) #else @@ -501,18 +501,18 @@ do { \ #if defined(ETHR_TRY_INLINE_FUNCS) || defined(ETHR_MUTEX_IMPL__) -void ethr_mutex_lock_wait__(ethr_mutex *, long); -void ethr_mutex_unlock_wake__(ethr_mutex *, long); +void ethr_mutex_lock_wait__(ethr_mutex *, ethr_sint32_t); +void ethr_mutex_unlock_wake__(ethr_mutex *, ethr_sint32_t); static ETHR_INLINE int ETHR_INLINE_FUNC_NAME_(ethr_mutex_trylock)(ethr_mutex *mtx) { - long act; + ethr_sint32_t act; int res; ETHR_MTX_HARD_DEBUG_FENCE_CHK(mtx); ETHR_MTX_DBG_CHK_UNUSED_FLG_BITS(mtx); - act = ethr_atomic_cmpxchg_acqb(&mtx->mtxb.flgs, ETHR_RWMTX_W_FLG__, 0); + act = ethr_atomic32_cmpxchg_acqb(&mtx->mtxb.flgs, ETHR_RWMTX_W_FLG__, 0); res = (act == 0) ? 0 : EBUSY; #ifdef ETHR_MTX_CHK_EXCL @@ -531,11 +531,11 @@ ETHR_INLINE_FUNC_NAME_(ethr_mutex_trylock)(ethr_mutex *mtx) static ETHR_INLINE void ETHR_INLINE_FUNC_NAME_(ethr_mutex_lock)(ethr_mutex *mtx) { - long act; + ethr_sint32_t act; ETHR_MTX_HARD_DEBUG_FENCE_CHK(mtx); ETHR_MTX_DBG_CHK_UNUSED_FLG_BITS(mtx); - act = ethr_atomic_cmpxchg_acqb(&mtx->mtxb.flgs, ETHR_RWMTX_W_FLG__, 0); + act = ethr_atomic32_cmpxchg_acqb(&mtx->mtxb.flgs, ETHR_RWMTX_W_FLG__, 0); if (act != 0) ethr_mutex_lock_wait__(mtx, act); @@ -551,7 +551,7 @@ ETHR_INLINE_FUNC_NAME_(ethr_mutex_lock)(ethr_mutex *mtx) static ETHR_INLINE void ETHR_INLINE_FUNC_NAME_(ethr_mutex_unlock)(ethr_mutex *mtx) { - long act; + ethr_sint32_t act; ETHR_COMPILER_BARRIER; ETHR_MTX_HARD_DEBUG_FENCE_CHK(mtx); ETHR_MTX_HARD_DEBUG_LFS_RWUNLOCK(&mtx->mtxb); @@ -559,7 +559,7 @@ ETHR_INLINE_FUNC_NAME_(ethr_mutex_unlock)(ethr_mutex *mtx) ETHR_MTX_CHK_EXCL_UNSET_EXCL(&mtx->mtxb); - act = ethr_atomic_cmpxchg_relb(&mtx->mtxb.flgs, 0, ETHR_RWMTX_W_FLG__); + act = ethr_atomic32_cmpxchg_relb(&mtx->mtxb.flgs, 0, ETHR_RWMTX_W_FLG__); if (act != ETHR_RWMTX_W_FLG__) ethr_mutex_unlock_wake__(mtx, act); diff --git a/erts/include/internal/ethr_optimized_fallbacks.h b/erts/include/internal/ethr_optimized_fallbacks.h index 2f9f987d0b..8e04692856 100644 --- a/erts/include/internal/ethr_optimized_fallbacks.h +++ b/erts/include/internal/ethr_optimized_fallbacks.h @@ -71,36 +71,46 @@ ethr_opt_spin_lock(ethr_opt_spinlock_t *lock) #define ETHR_HAVE_NATIVE_SPINLOCKS 1 #define ETHR_HAVE_OPTIMIZED_SPINLOCKS 1 -typedef ethr_native_atomic_t ethr_native_spinlock_t; +#if defined(ETHR_HAVE_NATIVE_ATOMIC32) +typedef ethr_native_atomic32_t ethr_native_spinlock_t; +# define ETHR_NATMC_FUNC__(X) ethr_native_atomic32_ ## X +#elif defined(ETHR_HAVE_NATIVE_ATOMIC64) +typedef ethr_native_atomic64_t ethr_native_spinlock_t; +# define ETHR_NATMC_FUNC__(X) ethr_native_atomic64_ ## X +#else +# error "Missing native atomic implementation" +#endif #if defined(ETHR_TRY_INLINE_FUNCS) || defined(ETHR_AUX_IMPL__) static ETHR_INLINE void ethr_native_spinlock_init(ethr_native_spinlock_t *lock) { - ethr_native_atomic_init((ethr_native_atomic_t *) lock, 0); + ETHR_NATMC_FUNC__(init)(lock, 0); } static ETHR_INLINE void ethr_native_spin_unlock(ethr_native_spinlock_t *lock) { ETHR_COMPILER_BARRIER; - ETHR_ASSERT(ethr_native_atomic_read((ethr_native_atomic_t *) lock) == 1); - ethr_native_atomic_set_relb((ethr_native_atomic_t *) lock, 0); + ETHR_ASSERT(ETHR_NATMC_FUNC__(read)(lock) == 1); + ETHR_NATMC_FUNC__(set_relb)(lock, 0); } static ETHR_INLINE void ethr_native_spin_lock(ethr_native_spinlock_t *lock) { - while (ethr_native_atomic_cmpxchg_acqb((ethr_native_atomic_t *) lock, - (long) 1, (long) 0) != 0) { - ETHR_SPIN_BODY; + while (ETHR_NATMC_FUNC__(cmpxchg_acqb)(lock, 1, 0) != 0) { + while (ETHR_NATMC_FUNC__(read)(lock) != 0) + ETHR_SPIN_BODY; } ETHR_COMPILER_BARRIER; } #endif +#undef ETHR_NATMC_FUNC__ + #endif @@ -111,16 +121,26 @@ ethr_native_spin_lock(ethr_native_spinlock_t *lock) #define ETHR_HAVE_NATIVE_RWSPINLOCKS 1 #define ETHR_HAVE_OPTIMIZED_RWSPINLOCKS 1 -typedef ethr_native_atomic_t ethr_native_rwlock_t; +#if defined(ETHR_HAVE_NATIVE_ATOMIC32) +typedef ethr_native_atomic32_t ethr_native_rwlock_t; +# define ETHR_NAINT_T__ ethr_sint32_t +# define ETHR_WLOCK_FLAG__ (((ethr_sint32_t) 1) << 30) +# define ETHR_NATMC_FUNC__(X) ethr_native_atomic32_ ## X +#elif defined(ETHR_HAVE_NATIVE_ATOMIC64) +typedef ethr_native_atomic64_t ethr_native_rwlock_t; +# define ETHR_NAINT_T__ ethr_sint64_t +# define ETHR_WLOCK_FLAG__ (((ethr_sint64_t) 1) << 62) +# define ETHR_NATMC_FUNC__(X) ethr_native_atomic64_ ## X +#else +# error "Missing native atomic implementation" +#endif #if defined(ETHR_TRY_INLINE_FUNCS) || defined(ETHR_AUX_IMPL__) -#define ETHR_WLOCK_FLAG__ (((long) 1) << 30) - static ETHR_INLINE void ethr_native_rwlock_init(ethr_native_rwlock_t *lock) { - ethr_native_atomic_init((ethr_native_atomic_t *) lock, 0); + ETHR_NATMC_FUNC__(init)(lock, 0); } static ETHR_INLINE void @@ -128,22 +148,24 @@ ethr_native_read_unlock(ethr_native_rwlock_t *lock) { ETHR_COMPILER_BARRIER; #ifdef DEBUG - ETHR_ASSERT(ethr_native_atomic_read((ethr_native_atomic_t *) lock) >= 0); + ETHR_ASSERT(ETHR_NATMC_FUNC__(read)(lock) >= 0); #endif - ethr_native_atomic_dec_relb((ethr_native_atomic_t *) lock); + ETHR_NATMC_FUNC__(dec_relb)(lock); } static ETHR_INLINE void ethr_native_read_lock(ethr_native_rwlock_t *lock) { - long act, exp = 0; + ETHR_NAINT_T__ act, exp = 0; while (1) { - act = ethr_native_atomic_cmpxchg_acqb((ethr_native_atomic_t *) lock, - exp+1, exp); + act = ETHR_NATMC_FUNC__(cmpxchg_acqb)(lock, exp+1, exp); if (act == exp) break; - ETHR_SPIN_BODY; - exp = (act & ETHR_WLOCK_FLAG__) ? 0 : act; + while (act & ETHR_WLOCK_FLAG__) { + ETHR_SPIN_BODY; + act = ETHR_NATMC_FUNC__(read)(lock); + } + exp = act; } ETHR_COMPILER_BARRIER; } @@ -152,18 +174,16 @@ static ETHR_INLINE void ethr_native_write_unlock(ethr_native_rwlock_t *lock) { ETHR_COMPILER_BARRIER; - ETHR_ASSERT(ethr_native_atomic_read((ethr_native_atomic_t *) lock) - == ETHR_WLOCK_FLAG__); - ethr_native_atomic_set_relb((ethr_native_atomic_t *) lock, 0); + ETHR_ASSERT(ETHR_NATMC_FUNC__(read)(lock) == ETHR_WLOCK_FLAG__); + ETHR_NATMC_FUNC__(set_relb)(lock, 0); } static ETHR_INLINE void ethr_native_write_lock(ethr_native_rwlock_t *lock) { - long act, exp = 0; + ETHR_NAINT_T__ act, exp = 0; while (1) { - act = ethr_native_atomic_cmpxchg_acqb((ethr_native_atomic_t *) lock, - exp|ETHR_WLOCK_FLAG__, exp); + act = ETHR_NATMC_FUNC__(cmpxchg_acqb)(lock, exp|ETHR_WLOCK_FLAG__, exp); if (act == exp) break; ETHR_SPIN_BODY; @@ -173,13 +193,17 @@ ethr_native_write_lock(ethr_native_rwlock_t *lock) /* Wait for readers to leave */ while (act != ETHR_WLOCK_FLAG__) { ETHR_SPIN_BODY; - act = ethr_native_atomic_read_acqb((ethr_native_atomic_t *) lock); + act = ETHR_NATMC_FUNC__(read_acqb)(lock); } ETHR_COMPILER_BARRIER; } #endif +#undef ETHR_NAINT_T__ +#undef ETHR_NATMC_FUNC__ +#undef ETHR_WLOCK_FLAG__ + #endif #endif diff --git a/erts/include/internal/ethread.h b/erts/include/internal/ethread.h index 53fa1acdc2..4cd95faf6a 100644 --- a/erts/include/internal/ethread.h +++ b/erts/include/internal/ethread.h @@ -37,11 +37,6 @@ #undef ETHR_HAVE_OPTIMIZED_SPINLOCK #undef ETHR_HAVE_OPTIMIZED_RWSPINLOCK -typedef struct { - long tv_sec; - long tv_nsec; -} ethr_timeval; - #if defined(DEBUG) # define ETHR_DEBUG #endif @@ -73,7 +68,7 @@ typedef struct { #endif /* Assume 64-byte cache line size */ -#define ETHR_CACHE_LINE_SIZE 64L +#define ETHR_CACHE_LINE_SIZE ((ethr_uint_t) 64) #define ETHR_CACHE_LINE_MASK (ETHR_CACHE_LINE_SIZE - 1) #define ETHR_CACHE_LINE_ALIGN_SIZE(SZ) \ @@ -171,6 +166,22 @@ typedef pthread_key_t ethr_tsd_key; # undef WIN32_LEAN_AND_MEAN #endif +#if defined(_MSC_VER) + +#if ETHR_SIZEOF_LONG == 4 +#define ETHR_HAVE_INT32_T 1 +typedef long ethr_sint32_t; +typedef unsigned long ethr_uint32_t; +#endif + +#if ETHR_SIZEOF___INT64 == 8 +#define ETHR_HAVE_INT64_T 1 +typedef __int64 ethr_sint64_t; +typedef unsigned __int64 ethr_uint64_t; +#endif + +#endif + struct ethr_join_data_; /* Types */ @@ -198,12 +209,48 @@ typedef DWORD ethr_tsd_key; #endif -#ifdef SIZEOF_LONG -#if SIZEOF_LONG < ETHR_SIZEOF_PTR -#error size of long currently needs to be at least the same as size of void * +#ifndef ETHR_HAVE_INT32_T +#if ETHR_SIZEOF_INT == 4 +#define ETHR_HAVE_INT32_T 1 +typedef int ethr_sint32_t; +typedef unsigned int ethr_uint32_t; +#elif ETHR_SIZEOF_LONG == 4 +#define ETHR_HAVE_INT32_T 1 +typedef long ethr_sint32_t; +typedef unsigned long ethr_uint32_t; #endif #endif +#ifndef ETHR_HAVE_INT64_T +#if ETHR_SIZEOF_INT == 8 +#define ETHR_HAVE_INT64_T 1 +typedef int ethr_sint64_t; +typedef unsigned int ethr_uint64_t; +#elif ETHR_SIZEOF_LONG == 8 +#define ETHR_HAVE_INT64_T 1 +typedef long ethr_sint64_t; +typedef unsigned long ethr_uint64_t; +#elif ETHR_SIZEOF_LONG_LONG == 8 +#define ETHR_HAVE_INT64_T 1 +typedef long long ethr_sint64_t; +typedef unsigned long long ethr_uint64_t; +#endif +#endif + +#if ETHR_SIZEOF_PTR == 4 +#ifndef ETHR_HAVE_INT32_T +#error "No 32-bit integer type found" +#endif +typedef ethr_sint32_t ethr_sint_t; +typedef ethr_uint32_t ethr_uint_t; +#elif ETHR_SIZEOF_PTR == 8 +#ifndef ETHR_HAVE_INT64_T +#error "No 64-bit integer type found" +#endif +typedef ethr_sint64_t ethr_sint_t; +typedef ethr_uint64_t ethr_uint_t; +#endif + /* __builtin_expect() is needed by both native atomics code * and the fallback code */ #if !defined(__GNUC__) || (__GNUC__ < 2) || (__GNUC__ == 2 && __GNUC_MINOR__ < 96) @@ -386,7 +433,6 @@ typedef struct { #if !defined(ETHR_TRY_INLINE_FUNCS) || defined(ETHR_AUX_IMPL__) # define ETHR_NEED_SPINLOCK_PROTOTYPES__ # define ETHR_NEED_RWSPINLOCK_PROTOTYPES__ -# define ETHR_NEED_ATOMIC_PROTOTYPES__ #endif int ethr_init(ethr_init_data *); @@ -399,7 +445,6 @@ void ethr_thr_exit(void *); ethr_tid ethr_self(void); int ethr_equal_tids(ethr_tid, ethr_tid); -int ethr_time_now(ethr_timeval *); int ethr_tsd_key_create(ethr_tsd_key *); int ethr_tsd_key_delete(ethr_tsd_key); int ethr_tsd_set(ethr_tsd_key, void *); @@ -502,312 +547,7 @@ ETHR_INLINE_FUNC_NAME_(ethr_spin_lock)(ethr_spinlock_t *lock) #endif /* ETHR_TRY_INLINE_FUNCS */ -#ifdef ETHR_HAVE_NATIVE_ATOMICS -/* - * Map ethread native atomics to ethread API atomics. - */ -typedef ethr_native_atomic_t ethr_atomic_t; -#else -typedef long ethr_atomic_t; -#endif - -#ifdef ETHR_NEED_ATOMIC_PROTOTYPES__ -void ethr_atomic_init(ethr_atomic_t *, long); -void ethr_atomic_set(ethr_atomic_t *, long); -long ethr_atomic_read(ethr_atomic_t *); -long ethr_atomic_inc_read(ethr_atomic_t *); -long ethr_atomic_dec_read(ethr_atomic_t *); -void ethr_atomic_inc(ethr_atomic_t *); -void ethr_atomic_dec(ethr_atomic_t *); -long ethr_atomic_add_read(ethr_atomic_t *, long); -void ethr_atomic_add(ethr_atomic_t *, long); -long ethr_atomic_read_band(ethr_atomic_t *, long); -long ethr_atomic_read_bor(ethr_atomic_t *, long); -long ethr_atomic_xchg(ethr_atomic_t *, long); -long ethr_atomic_cmpxchg(ethr_atomic_t *, long, long); -long ethr_atomic_read_acqb(ethr_atomic_t *); -long ethr_atomic_inc_read_acqb(ethr_atomic_t *); -void ethr_atomic_set_relb(ethr_atomic_t *, long); -void ethr_atomic_dec_relb(ethr_atomic_t *); -long ethr_atomic_dec_read_relb(ethr_atomic_t *); -long ethr_atomic_cmpxchg_acqb(ethr_atomic_t *, long, long); -long ethr_atomic_cmpxchg_relb(ethr_atomic_t *, long, long); -#endif - -#if defined(ETHR_TRY_INLINE_FUNCS) || defined(ETHR_AUX_IMPL__) - -#ifndef ETHR_HAVE_NATIVE_ATOMICS -/* - * Fallbacks for atomics used in absence of a native implementation. - */ - -#define ETHR_ATOMIC_ADDR_BITS 10 -#define ETHR_ATOMIC_ADDR_SHIFT 6 - -typedef struct { - union { - ethr_spinlock_t lck; - char buf[ETHR_CACHE_LINE_SIZE]; - } u; -} ethr_atomic_protection_t; - -extern ethr_atomic_protection_t ethr_atomic_protection__[1 << ETHR_ATOMIC_ADDR_BITS]; - -#define ETHR_ATOMIC_PTR2LCK__(PTR) \ -(ðr_atomic_protection__[((((unsigned long) (PTR)) >> ETHR_ATOMIC_ADDR_SHIFT) \ - & ((1 << ETHR_ATOMIC_ADDR_BITS) - 1))].u.lck) - - -#define ETHR_ATOMIC_OP_FALLBACK_IMPL__(AP, EXPS) \ -do { \ - ethr_spinlock_t *slp__ = ETHR_ATOMIC_PTR2LCK__((AP)); \ - ethr_spin_lock(slp__); \ - { EXPS; } \ - ethr_spin_unlock(slp__); \ -} while (0) - -#endif - -static ETHR_INLINE void -ETHR_INLINE_FUNC_NAME_(ethr_atomic_init)(ethr_atomic_t *var, long i) -{ -#ifdef ETHR_HAVE_NATIVE_ATOMICS - ethr_native_atomic_init(var, i); -#else - ETHR_ATOMIC_OP_FALLBACK_IMPL__(var, *var = i); -#endif -} - -static ETHR_INLINE void -ETHR_INLINE_FUNC_NAME_(ethr_atomic_set)(ethr_atomic_t *var, long i) -{ -#ifdef ETHR_HAVE_NATIVE_ATOMICS - ethr_native_atomic_set(var, i); -#else - ETHR_ATOMIC_OP_FALLBACK_IMPL__(var, *var = i); -#endif -} - -static ETHR_INLINE long -ETHR_INLINE_FUNC_NAME_(ethr_atomic_read)(ethr_atomic_t *var) -{ -#ifdef ETHR_HAVE_NATIVE_ATOMICS - return ethr_native_atomic_read(var); -#else - long res; - ETHR_ATOMIC_OP_FALLBACK_IMPL__(var, res = (long) *var); - return res; -#endif -} - -static ETHR_INLINE void -ETHR_INLINE_FUNC_NAME_(ethr_atomic_add)(ethr_atomic_t *var, long incr) -{ -#ifdef ETHR_HAVE_NATIVE_ATOMICS - ethr_native_atomic_add(var, incr); -#else - ETHR_ATOMIC_OP_FALLBACK_IMPL__(var, *var += incr); -#endif -} - -static ETHR_INLINE long -ETHR_INLINE_FUNC_NAME_(ethr_atomic_add_read)(ethr_atomic_t *var, long i) -{ -#ifdef ETHR_HAVE_NATIVE_ATOMICS - return ethr_native_atomic_add_return(var, i); -#else - long res; - ETHR_ATOMIC_OP_FALLBACK_IMPL__(var, *var += i; res = *var); - return res; -#endif -} - -static ETHR_INLINE void -ETHR_INLINE_FUNC_NAME_(ethr_atomic_inc)(ethr_atomic_t *var) -{ -#ifdef ETHR_HAVE_NATIVE_ATOMICS - ethr_native_atomic_inc(var); -#else - ETHR_ATOMIC_OP_FALLBACK_IMPL__(var, ++(*var)); -#endif -} - -static ETHR_INLINE void -ETHR_INLINE_FUNC_NAME_(ethr_atomic_dec)(ethr_atomic_t *var) -{ -#ifdef ETHR_HAVE_NATIVE_ATOMICS - ethr_native_atomic_dec(var); -#else - ETHR_ATOMIC_OP_FALLBACK_IMPL__(var, --(*var)); -#endif -} - -static ETHR_INLINE long -ETHR_INLINE_FUNC_NAME_(ethr_atomic_inc_read)(ethr_atomic_t *var) -{ -#ifdef ETHR_HAVE_NATIVE_ATOMICS - return ethr_native_atomic_inc_return(var); -#else - long res; - ETHR_ATOMIC_OP_FALLBACK_IMPL__(var, res = (long) ++(*var)); - return res; -#endif -} - -static ETHR_INLINE long -ETHR_INLINE_FUNC_NAME_(ethr_atomic_dec_read)(ethr_atomic_t *var) -{ -#ifdef ETHR_HAVE_NATIVE_ATOMICS - return ethr_native_atomic_dec_return(var); -#else - long res; - ETHR_ATOMIC_OP_FALLBACK_IMPL__(var, res = (long) --(*var)); - return res; -#endif -} - -static ETHR_INLINE long -ETHR_INLINE_FUNC_NAME_(ethr_atomic_read_band)(ethr_atomic_t *var, - long mask) -{ -#ifdef ETHR_HAVE_NATIVE_ATOMICS - return ethr_native_atomic_and_retold(var, mask); -#else - long res; - ETHR_ATOMIC_OP_FALLBACK_IMPL__(var, res = *var; *var &= mask); - return res; -#endif -} - -static ETHR_INLINE long -ETHR_INLINE_FUNC_NAME_(ethr_atomic_read_bor)(ethr_atomic_t *var, - long mask) -{ -#ifdef ETHR_HAVE_NATIVE_ATOMICS - return ethr_native_atomic_or_retold(var, mask); -#else - long res; - ETHR_ATOMIC_OP_FALLBACK_IMPL__(var, res = *var; *var |= mask); - return res; -#endif -} - -static ETHR_INLINE long -ETHR_INLINE_FUNC_NAME_(ethr_atomic_xchg)(ethr_atomic_t *var, - long new) -{ -#ifdef ETHR_HAVE_NATIVE_ATOMICS - return ethr_native_atomic_xchg(var, new); -#else - long res; - ETHR_ATOMIC_OP_FALLBACK_IMPL__(var, res = *var; *var = new); - return res; -#endif -} - -static ETHR_INLINE long -ETHR_INLINE_FUNC_NAME_(ethr_atomic_cmpxchg)(ethr_atomic_t *var, - long new, - long exp) -{ -#ifdef ETHR_HAVE_NATIVE_ATOMICS - return ethr_native_atomic_cmpxchg(var, new, exp); -#else - long res; - ETHR_ATOMIC_OP_FALLBACK_IMPL__(var, - { - res = *var; - if (__builtin_expect(res == exp, 1)) - *var = new; - }); - return res; -#endif -} - -/* - * Important memory barrier requirements. - * - * The following atomic operations *must* supply a memory barrier of - * at least the type specified by its suffix: - * _acqb = acquire barrier - * _relb = release barrier - */ - -static ETHR_INLINE long -ETHR_INLINE_FUNC_NAME_(ethr_atomic_read_acqb)(ethr_atomic_t *var) -{ -#ifdef ETHR_HAVE_NATIVE_ATOMICS - return ethr_native_atomic_read_acqb(var); -#else - return ETHR_INLINE_FUNC_NAME_(ethr_atomic_read)(var); -#endif -} - -static ETHR_INLINE long -ETHR_INLINE_FUNC_NAME_(ethr_atomic_inc_read_acqb)(ethr_atomic_t *var) -{ -#ifdef ETHR_HAVE_NATIVE_ATOMICS - return ethr_native_atomic_inc_return_acqb(var); -#else - return ETHR_INLINE_FUNC_NAME_(ethr_atomic_inc_read)(var); -#endif -} - -static ETHR_INLINE void -ETHR_INLINE_FUNC_NAME_(ethr_atomic_set_relb)(ethr_atomic_t *var, long val) -{ -#ifdef ETHR_HAVE_NATIVE_ATOMICS - ethr_native_atomic_set_relb(var, val); -#else - ETHR_INLINE_FUNC_NAME_(ethr_atomic_set)(var, val); -#endif -} - -static ETHR_INLINE void -ETHR_INLINE_FUNC_NAME_(ethr_atomic_dec_relb)(ethr_atomic_t *var) -{ -#ifdef ETHR_HAVE_NATIVE_ATOMICS - ethr_native_atomic_dec_relb(var); -#else - ETHR_INLINE_FUNC_NAME_(ethr_atomic_dec)(var); -#endif -} - -static ETHR_INLINE long -ETHR_INLINE_FUNC_NAME_(ethr_atomic_dec_read_relb)(ethr_atomic_t *var) -{ -#ifdef ETHR_HAVE_NATIVE_ATOMICS - return ethr_native_atomic_dec_return_relb(var); -#else - return ETHR_INLINE_FUNC_NAME_(ethr_atomic_dec_read)(var); -#endif -} - -static ETHR_INLINE long -ETHR_INLINE_FUNC_NAME_(ethr_atomic_cmpxchg_acqb)(ethr_atomic_t *var, - long new, - long exp) -{ -#ifdef ETHR_HAVE_NATIVE_ATOMICS - return ethr_native_atomic_cmpxchg_acqb(var, new, exp); -#else - return ETHR_INLINE_FUNC_NAME_(ethr_atomic_cmpxchg)(var, new, exp); -#endif -} - -static ETHR_INLINE long -ETHR_INLINE_FUNC_NAME_(ethr_atomic_cmpxchg_relb)(ethr_atomic_t *var, - long new, - long exp) -{ -#ifdef ETHR_HAVE_NATIVE_ATOMICS - return ethr_native_atomic_cmpxchg_relb(var, new, exp); -#else - return ETHR_INLINE_FUNC_NAME_(ethr_atomic_cmpxchg)(var, new, exp); -#endif -} - -#endif /* ETHR_TRY_INLINE_FUNCS */ +#include "ethr_atomics.h" typedef struct ethr_ts_event_ ethr_ts_event; /* Needed by ethr_mutex.h */ @@ -825,7 +565,7 @@ struct ethr_ts_event_ { ethr_ts_event *prev; ethr_event event; void *udata; - ethr_atomic_t uaflgs; + ethr_atomic32_t uaflgs; unsigned uflgs; unsigned iflgs; /* for ethr lib only */ short rgix; /* for ethr lib only */ diff --git a/erts/include/internal/ethread_header_config.h.in b/erts/include/internal/ethread_header_config.h.in index 5debb44756..f394d790d2 100644 --- a/erts/include/internal/ethread_header_config.h.in +++ b/erts/include/internal/ethread_header_config.h.in @@ -20,6 +20,21 @@ /* Define to the size of pointers */ #undef ETHR_SIZEOF_PTR +/* Define to the size of int */ +#undef ETHR_SIZEOF_INT + +/* Define to the size of long */ +#undef ETHR_SIZEOF_LONG + +/* Define to the size of long long */ +#undef ETHR_SIZEOF_LONG_LONG + +/* Define to the size of __int64 */ +#undef ETHR_SIZEOF___INT64 + +/* Define if bigendian */ +#undef ETHR_BIGENDIAN + /* Define if you want to disable native ethread implementations */ #undef ETHR_DISABLE_NATIVE_IMPLS @@ -100,6 +115,27 @@ /* Define to the size of AO_t if libatomic_ops is used */ #undef ETHR_SIZEOF_AO_T +/* Define if you have _InterlockedCompareExchange64() */ +#undef ETHR_HAVE__INTERLOCKEDCOMPAREEXCHANGE64 + +/* Define if you have _InterlockedDecrement64() */ +#undef ETHR_HAVE__INTERLOCKEDDECREMENT64 + +/* Define if you have _InterlockedIncrement64() */ +#undef ETHR_HAVE__INTERLOCKEDINCREMENT64 + +/* Define if you have _InterlockedExchangeAdd64() */ +#undef ETHR_HAVE__INTERLOCKEDEXCHANGEADD64 + +/* Define if you have _InterlockedExchange64() */ +#undef ETHR_HAVE__INTERLOCKEDEXCHANGE64 + +/* Define if you have _InterlockedAnd64() */ +#undef ETHR_HAVE__INTERLOCKEDAND64 + +/* Define if you have _InterlockedOr64() */ +#undef ETHR_HAVE__INTERLOCKEDOR64 + /* Define if you want to turn on extra sanity checking in the ethread library */ #undef ETHR_XCHK diff --git a/erts/include/internal/gcc/ethr_atomic.h b/erts/include/internal/gcc/ethr_atomic.h index e8e529dd48..16935084b1 100644 --- a/erts/include/internal/gcc/ethr_atomic.h +++ b/erts/include/internal/gcc/ethr_atomic.h @@ -22,24 +22,35 @@ * Author: Rickard Green */ -#ifndef ETHR_GCC_ATOMIC_H__ -#define ETHR_GCC_ATOMIC_H__ +#undef ETHR_INCLUDE_ATOMIC_IMPL__ +#if !defined(ETHR_GCC_ATOMIC32_H__) && defined(ETHR_ATOMIC_WANT_32BIT_IMPL__) +#define ETHR_GCC_ATOMIC32_H__ +#define ETHR_INCLUDE_ATOMIC_IMPL__ 4 +#undef ETHR_ATOMIC_WANT_32BIT_IMPL__ +#elif !defined(ETHR_GCC_ATOMIC64_H__) && defined(ETHR_ATOMIC_WANT_64BIT_IMPL__) +#define ETHR_GCC_ATOMIC64_H__ +#define ETHR_INCLUDE_ATOMIC_IMPL__ 8 +#undef ETHR_ATOMIC_WANT_64BIT_IMPL__ +#endif + +#ifdef ETHR_INCLUDE_ATOMIC_IMPL__ -#if !defined(ETHR_HAVE_NATIVE_ATOMICS) && defined(ETHR_HAVE_GCC_ATOMIC_OPS) -#define ETHR_HAVE_NATIVE_ATOMICS 1 +#ifndef ETHR_GCC_ATOMIC_COMMON__ +#define ETHR_GCC_ATOMIC_COMMON__ -#define ETHR_IMMED_ATOMIC_SET_GET_SAFE__ 0 -/* Enable immediate read/write on platforms where we know it is safe */ +#define ETHR_READ_AND_SET_WITHOUT_SYNC_OP__ 0 #if defined(__i386__) || defined(__x86_64__) || defined(__sparc__) \ || defined(__powerpc__) || defined(__ppc__) || defined(__mips__) -# undef ETHR_IMMED_ATOMIC_SET_GET_SAFE__ -# define ETHR_IMMED_ATOMIC_SET_GET_SAFE__ 1 +# undef ETHR_READ_AND_SET_WITHOUT_SYNC_OP__ +# define ETHR_READ_AND_SET_WITHOUT_SYNC_OP__ 1 #endif -typedef struct { - volatile long counter; -} ethr_native_atomic_t; - +#if defined(__x86_64__) || (defined(__i386__) \ + && !defined(ETHR_PRE_PENTIUM4_COMPAT)) +# define ETHR_READ_ACQB_AND_SET_RELB_COMPILER_BARRIER_ONLY__ 1 +#else +# define ETHR_READ_ACQB_AND_SET_RELB_COMPILER_BARRIER_ONLY__ 0 +#endif /* * According to the documentation this is what we want: @@ -47,34 +58,73 @@ typedef struct { * However, __sync_synchronize() is known to erroneously be * a noop on at least some platforms with some gcc versions. * This has suposedly been fixed in some gcc version, but we - * don't know from which version. Therefore, we use the - * workaround implemented below on all gcc versions except - * for gcc 4.2 or above for MIPS, where it's been verified. + * don't know from which version. Therefore, we only use + * it when it has been verified to work. Otherwise + * we use a workaround. */ #if defined(__mips__) && (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 2)) +/* __sync_synchronize() has been verified to work here */ #define ETHR_MEMORY_BARRIER __sync_synchronize() +#define ETHR_READ_DEPEND_MEMORY_BARRIER __sync_synchronize() +#elif defined(__x86_64__) || (defined(__i386__) \ + && !defined(ETHR_PRE_PENTIUM4_COMPAT)) +/* Use fence instructions directly instead of workaround */ +#define ETHR_MEMORY_BARRIER __asm__ __volatile__("mfence" : : : "memory") +#define ETHR_WRITE_MEMORY_BARRIER __asm__ __volatile__("sfence" : : : "memory") +#define ETHR_READ_MEMORY_BARRIER __asm__ __volatile__("lfence" : : : "memory") +#define ETHR_READ_DEPEND_MEMORY_BARRIER __asm__ __volatile__("" : : : "memory") #else +/* Workaround */ #define ETHR_MEMORY_BARRIER \ do { \ - volatile long x___ = 0; \ - (void) __sync_val_compare_and_swap(&x___, (long) 0, (long) 1); \ + volatile ethr_sint32_t x___ = 0; \ + (void) __sync_val_compare_and_swap(&x___, (ethr_sint32_t) 0, (ethr_sint32_t) 1); \ } while (0) -#endif #define ETHR_READ_DEPEND_MEMORY_BARRIER ETHR_MEMORY_BARRIER +#endif + +#define ETHR_COMPILER_BARRIER __asm__ __volatile__("" : : : "memory") -#if defined(ETHR_TRY_INLINE_FUNCS) || defined(ETHR_AUX_IMPL__) +#endif /* ETHR_GCC_ATOMIC_COMMON__ */ + +#if ETHR_INCLUDE_ATOMIC_IMPL__ == 4 +#define ETHR_HAVE_NATIVE_ATOMIC32 1 +#define ETHR_NATMC_FUNC__(X) ethr_native_atomic32_ ## X +#define ETHR_ATMC_T__ ethr_native_atomic32_t +#define ETHR_AINT_T__ ethr_sint32_t +#elif ETHR_INCLUDE_ATOMIC_IMPL__ == 8 +#define ETHR_HAVE_NATIVE_ATOMIC64 1 +#define ETHR_NATMC_FUNC__(X) ethr_native_atomic64_ ## X +#define ETHR_ATMC_T__ ethr_native_atomic64_t +#define ETHR_AINT_T__ ethr_sint64_t +#else +#error "Unsupported integer size" +#endif + +typedef struct { + volatile ETHR_AINT_T__ counter; +} ETHR_ATMC_T__; + + +#if defined(ETHR_TRY_INLINE_FUNCS) || defined(ETHR_ATOMIC_IMPL__) + +static ETHR_INLINE ETHR_AINT_T__ * +ETHR_NATMC_FUNC__(addr)(ETHR_ATMC_T__ *var) +{ + return (ETHR_AINT_T__ *) &var->counter; +} static ETHR_INLINE void -ethr_native_atomic_set(ethr_native_atomic_t *var, long value) +ETHR_NATMC_FUNC__(set)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ value) { -#if ETHR_IMMED_ATOMIC_SET_GET_SAFE__ +#if ETHR_READ_AND_SET_WITHOUT_SYNC_OP__ var->counter = value; #else /* * Unfortunately no __sync_store() or similar exist in the gcc atomic * op interface. We therefore have to simulate it this way... */ - long act = 0, exp; + ETHR_AINT_T__ act = 0, exp; do { exp = act; act = __sync_val_compare_and_swap(&var->counter, exp, value); @@ -82,80 +132,86 @@ ethr_native_atomic_set(ethr_native_atomic_t *var, long value) #endif } -#define ethr_native_atomic_init ethr_native_atomic_set +static ETHR_INLINE void +ETHR_NATMC_FUNC__(init)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ value) +{ + ETHR_NATMC_FUNC__(set)(var, value); +} -static ETHR_INLINE long -ethr_native_atomic_read(ethr_native_atomic_t *var) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(read)(ETHR_ATMC_T__ *var) { -#if ETHR_IMMED_ATOMIC_SET_GET_SAFE__ +#if ETHR_READ_AND_SET_WITHOUT_SYNC_OP__ return var->counter; #else /* * Unfortunately no __sync_fetch() or similar exist in the gcc atomic * op interface. We therefore have to simulate it this way... */ - return __sync_add_and_fetch(&var->counter, (long) 0); + return __sync_add_and_fetch(&var->counter, (ETHR_AINT_T__) 0); #endif } static ETHR_INLINE void -ethr_native_atomic_add(ethr_native_atomic_t *var, long incr) +ETHR_NATMC_FUNC__(add)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ incr) { (void) __sync_add_and_fetch(&var->counter, incr); } -static ETHR_INLINE long -ethr_native_atomic_add_return(ethr_native_atomic_t *var, long incr) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(add_return)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ incr) { return __sync_add_and_fetch(&var->counter, incr); } static ETHR_INLINE void -ethr_native_atomic_inc(ethr_native_atomic_t *var) +ETHR_NATMC_FUNC__(inc)(ETHR_ATMC_T__ *var) { - (void) __sync_add_and_fetch(&var->counter, (long) 1); + (void) __sync_add_and_fetch(&var->counter, (ETHR_AINT_T__) 1); } static ETHR_INLINE void -ethr_native_atomic_dec(ethr_native_atomic_t *var) +ETHR_NATMC_FUNC__(dec)(ETHR_ATMC_T__ *var) { - (void) __sync_sub_and_fetch(&var->counter, (long) 1); + (void) __sync_sub_and_fetch(&var->counter, (ETHR_AINT_T__) 1); } -static ETHR_INLINE long -ethr_native_atomic_inc_return(ethr_native_atomic_t *var) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(inc_return)(ETHR_ATMC_T__ *var) { - return __sync_add_and_fetch(&var->counter, (long) 1); + return __sync_add_and_fetch(&var->counter, (ETHR_AINT_T__) 1); } -static ETHR_INLINE long -ethr_native_atomic_dec_return(ethr_native_atomic_t *var) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(dec_return)(ETHR_ATMC_T__ *var) { - return __sync_sub_and_fetch(&var->counter, (long) 1); + return __sync_sub_and_fetch(&var->counter, (ETHR_AINT_T__) 1); } -static ETHR_INLINE long -ethr_native_atomic_and_retold(ethr_native_atomic_t *var, long mask) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(and_retold)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ mask) { return __sync_fetch_and_and(&var->counter, mask); } -static ETHR_INLINE long -ethr_native_atomic_or_retold(ethr_native_atomic_t *var, long mask) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(or_retold)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ mask) { - return (long) __sync_fetch_and_or(&var->counter, mask); + return (ETHR_AINT_T__) __sync_fetch_and_or(&var->counter, mask); } -static ETHR_INLINE long -ethr_native_atomic_cmpxchg(ethr_native_atomic_t *var, long new, long old) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(cmpxchg)(ETHR_ATMC_T__ *var, + ETHR_AINT_T__ new, + ETHR_AINT_T__ old) { return __sync_val_compare_and_swap(&var->counter, old, new); } -static ETHR_INLINE long -ethr_native_atomic_xchg(ethr_native_atomic_t *var, long new) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(xchg)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ new) { - long exp, act = 0; + ETHR_AINT_T__ exp, act = 0; do { exp = act; act = __sync_val_compare_and_swap(&var->counter, exp, new); @@ -167,22 +223,68 @@ ethr_native_atomic_xchg(ethr_native_atomic_t *var, long new) * Atomic ops with at least specified barriers. */ -static ETHR_INLINE long -ethr_native_atomic_read_acqb(ethr_native_atomic_t *var) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(read_acqb)(ETHR_ATMC_T__ *var) { - return __sync_add_and_fetch(&var->counter, (long) 0); +#if ETHR_READ_ACQB_AND_SET_RELB_COMPILER_BARRIER_ONLY__ + ETHR_AINT_T__ val = var->counter; + ETHR_COMPILER_BARRIER; + return val; +#else + return __sync_add_and_fetch(&var->counter, (ETHR_AINT_T__) 0); +#endif } -#define ethr_native_atomic_inc_return_acqb ethr_native_atomic_inc_return -#define ethr_native_atomic_set_relb ethr_native_atomic_xchg -#define ethr_native_atomic_dec_relb ethr_native_atomic_dec_return -#define ethr_native_atomic_dec_return_relb ethr_native_atomic_dec_return +static ETHR_INLINE void +ETHR_NATMC_FUNC__(set_relb)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ i) +{ +#if ETHR_READ_ACQB_AND_SET_RELB_COMPILER_BARRIER_ONLY__ + ETHR_COMPILER_BARRIER; + var->counter = i; +#else + (void) ETHR_NATMC_FUNC__(xchg)(var, i); +#endif +} -#define ethr_native_atomic_cmpxchg_acqb ethr_native_atomic_cmpxchg -#define ethr_native_atomic_cmpxchg_relb ethr_native_atomic_cmpxchg +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(inc_return_acqb)(ETHR_ATMC_T__ *var) +{ + return ETHR_NATMC_FUNC__(inc_return)(var); +} -#endif +static ETHR_INLINE void +ETHR_NATMC_FUNC__(dec_relb)(ETHR_ATMC_T__ *var) +{ + ETHR_NATMC_FUNC__(dec)(var); +} + +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(dec_return_relb)(ETHR_ATMC_T__ *var) +{ + return ETHR_NATMC_FUNC__(dec_return)(var); +} + +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(cmpxchg_acqb)(ETHR_ATMC_T__ *var, + ETHR_AINT_T__ new, + ETHR_AINT_T__ old) +{ + return ETHR_NATMC_FUNC__(cmpxchg)(var, new, old); +} + +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(cmpxchg_relb)(ETHR_ATMC_T__ *var, + ETHR_AINT_T__ new, + ETHR_AINT_T__ old) +{ + return ETHR_NATMC_FUNC__(cmpxchg)(var, new, old); +} #endif +#undef ETHR_NATMC_FUNC__ +#undef ETHR_ATMC_T__ +#undef ETHR_AINT_T__ +#undef ETHR_AINT_SUFFIX__ + #endif diff --git a/erts/include/internal/gcc/ethread.h b/erts/include/internal/gcc/ethread.h index bb378e31e0..392a1aa2b2 100644 --- a/erts/include/internal/gcc/ethread.h +++ b/erts/include/internal/gcc/ethread.h @@ -25,6 +25,16 @@ #ifndef ETHREAD_GCC_H__ #define ETHREAD_GCC_H__ +#if !defined(ETHR_HAVE_NATIVE_ATOMICS) && defined(ETHR_HAVE_GCC_ATOMIC_OPS) +#define ETHR_HAVE_NATIVE_ATOMICS 1 + +#define ETHR_ATOMIC_WANT_32BIT_IMPL__ #include "ethr_atomic.h" +#if ETHR_SIZEOF_PTR == 8 +# define ETHR_ATOMIC_WANT_64BIT_IMPL__ +# include "ethr_atomic.h" +#endif + +#endif #endif diff --git a/erts/include/internal/i386/atomic.h b/erts/include/internal/i386/atomic.h index 52d01aab32..4e402f261a 100644 --- a/erts/include/internal/i386/atomic.h +++ b/erts/include/internal/i386/atomic.h @@ -23,14 +23,24 @@ * * This code requires a 486 or newer processor. */ -#ifndef ETHREAD_I386_ATOMIC_H -#define ETHREAD_I386_ATOMIC_H -/* An atomic is an aligned long accessed via locked operations. - */ -typedef struct { - volatile long counter; -} ethr_native_atomic_t; +#undef ETHR_INCLUDE_ATOMIC_IMPL__ +#if !defined(ETHR_X86_ATOMIC32_H__) && defined(ETHR_ATOMIC_WANT_32BIT_IMPL__) +#define ETHR_X86_ATOMIC32_H__ +#define ETHR_INCLUDE_ATOMIC_IMPL__ 4 +#undef ETHR_ATOMIC_WANT_32BIT_IMPL__ +#elif !defined(ETHR_X86_ATOMIC64_H__) && defined(ETHR_ATOMIC_WANT_64BIT_IMPL__) +#define ETHR_X86_ATOMIC64_H__ +#define ETHR_INCLUDE_ATOMIC_IMPL__ 8 +#undef ETHR_ATOMIC_WANT_64BIT_IMPL__ +#endif + +#ifdef ETHR_INCLUDE_ATOMIC_IMPL__ + +#ifndef ETHR_X86_ATOMIC_COMMON__ +#define ETHR_X86_ATOMIC_COMMON__ + +#define ETHR_ATOMIC_HAVE_INC_DEC_INSTRUCTIONS 1 #if defined(__x86_64__) || !defined(ETHR_PRE_PENTIUM4_COMPAT) #define ETHR_MEMORY_BARRIER __asm__ __volatile__("mfence" : : : "memory") @@ -40,123 +50,161 @@ typedef struct { #else #define ETHR_MEMORY_BARRIER \ do { \ - volatile long x___ = 0; \ + volatile ethr_sint32_t x___ = 0; \ __asm__ __volatile__("lock; incl %0" : "=m"(x___) : "m"(x___) : "memory"); \ } while (0) #endif -#define ETHR_ATOMIC_HAVE_INC_DEC_INSTRUCTIONS 1 - -#if defined(ETHR_TRY_INLINE_FUNCS) || defined(ETHR_AUX_IMPL__) +#endif /* ETHR_X86_ATOMIC_COMMON__ */ -#ifdef __x86_64__ -#define LONG_SUFFIX "q" +#if ETHR_INCLUDE_ATOMIC_IMPL__ == 4 +#define ETHR_HAVE_NATIVE_ATOMIC32 1 +#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_INCLUDE_ATOMIC_IMPL__ == 8 +#define ETHR_HAVE_NATIVE_ATOMIC64 1 +#define ETHR_NATMC_FUNC__(X) ethr_native_atomic64_ ## X +#define ETHR_ATMC_T__ ethr_native_atomic64_t +#define ETHR_AINT_T__ ethr_sint64_t +#define ETHR_AINT_SUFFIX__ "q" #else -#define LONG_SUFFIX "l" +#error "Unsupported integer size" #endif +/* An atomic is an aligned ETHR_AINT_T__ accessed via locked operations. + */ +typedef struct { + volatile ETHR_AINT_T__ counter; +} ETHR_ATMC_T__; + +#if defined(ETHR_TRY_INLINE_FUNCS) || defined(ETHR_ATOMIC_IMPL__) + +static ETHR_INLINE ETHR_AINT_T__ * +ETHR_NATMC_FUNC__(addr)(ETHR_ATMC_T__ *var) +{ + return (ETHR_AINT_T__ *) &var->counter; +} + static ETHR_INLINE void -ethr_native_atomic_init(ethr_native_atomic_t *var, long i) +ETHR_NATMC_FUNC__(init)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ i) { var->counter = i; } -#define ethr_native_atomic_set(v, i) ethr_native_atomic_init((v), (i)) -static ETHR_INLINE long -ethr_native_atomic_read(ethr_native_atomic_t *var) +static ETHR_INLINE void +ETHR_NATMC_FUNC__(set)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ i) +{ + var->counter = i; +} + +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(read)(ETHR_ATMC_T__ *var) { return var->counter; } static ETHR_INLINE void -ethr_native_atomic_add(ethr_native_atomic_t *var, long incr) +ETHR_NATMC_FUNC__(add)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ incr) { __asm__ __volatile__( - "lock; add" LONG_SUFFIX " %1, %0" + "lock; add" ETHR_AINT_SUFFIX__ " %1, %0" : "=m"(var->counter) : "ir"(incr), "m"(var->counter)); } static ETHR_INLINE void -ethr_native_atomic_inc(ethr_native_atomic_t *var) +ETHR_NATMC_FUNC__(inc)(ETHR_ATMC_T__ *var) { __asm__ __volatile__( - "lock; inc" LONG_SUFFIX " %0" + "lock; inc" ETHR_AINT_SUFFIX__ " %0" : "=m"(var->counter) : "m"(var->counter)); } static ETHR_INLINE void -ethr_native_atomic_dec(ethr_native_atomic_t *var) +ETHR_NATMC_FUNC__(dec)(ETHR_ATMC_T__ *var) { __asm__ __volatile__( - "lock; dec" LONG_SUFFIX " %0" + "lock; dec" ETHR_AINT_SUFFIX__ " %0" : "=m"(var->counter) : "m"(var->counter)); } -static ETHR_INLINE long -ethr_native_atomic_add_return(ethr_native_atomic_t *var, long incr) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(add_return)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ incr) { - long tmp; + ETHR_AINT_T__ tmp; tmp = incr; __asm__ __volatile__( - "lock; xadd" LONG_SUFFIX " %0, %1" /* xadd didn't exist prior to the 486 */ + "lock; xadd" ETHR_AINT_SUFFIX__ " %0, %1" /* xadd didn't exist prior to the 486 */ : "=r"(tmp) : "m"(var->counter), "0"(tmp)); /* now tmp is the atomic's previous value */ return tmp + incr; } -#define ethr_native_atomic_inc_return(var) ethr_native_atomic_add_return((var), 1) -#define ethr_native_atomic_dec_return(var) ethr_native_atomic_add_return((var), -1) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(inc_return)(ETHR_ATMC_T__ *var) +{ + return ETHR_NATMC_FUNC__(add_return)(var, (ETHR_AINT_T__) 1); +} -static ETHR_INLINE long -ethr_native_atomic_cmpxchg(ethr_native_atomic_t *var, long new, long old) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(dec_return)(ETHR_ATMC_T__ *var) +{ + return ETHR_NATMC_FUNC__(add_return)(var, (ETHR_AINT_T__) -1); +} + +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(cmpxchg)(ETHR_ATMC_T__ *var, + ETHR_AINT_T__ new, + ETHR_AINT_T__ old) { __asm__ __volatile__( - "lock; cmpxchg" LONG_SUFFIX " %2, %3" + "lock; cmpxchg" ETHR_AINT_SUFFIX__ " %2, %3" : "=a"(old), "=m"(var->counter) : "r"(new), "m"(var->counter), "0"(old) : "cc", "memory"); /* full memory clobber to make this a compiler barrier */ return old; } -static ETHR_INLINE long -ethr_native_atomic_and_retold(ethr_native_atomic_t *var, long mask) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(and_retold)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ mask) { - long tmp, old; + ETHR_AINT_T__ tmp, old; tmp = var->counter; do { old = tmp; - tmp = ethr_native_atomic_cmpxchg(var, tmp & mask, tmp); + tmp = ETHR_NATMC_FUNC__(cmpxchg)(var, tmp & mask, tmp); } while (__builtin_expect(tmp != old, 0)); /* now tmp is the atomic's previous value */ return tmp; } -static ETHR_INLINE long -ethr_native_atomic_or_retold(ethr_native_atomic_t *var, long mask) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(or_retold)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ mask) { - long tmp, old; + ETHR_AINT_T__ tmp, old; tmp = var->counter; do { old = tmp; - tmp = ethr_native_atomic_cmpxchg(var, tmp | mask, tmp); + tmp = ETHR_NATMC_FUNC__(cmpxchg)(var, tmp | mask, tmp); } while (__builtin_expect(tmp != old, 0)); /* now tmp is the atomic's previous value */ return tmp; } -static ETHR_INLINE long -ethr_native_atomic_xchg(ethr_native_atomic_t *var, long val) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(xchg)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ val) { - long tmp = val; + ETHR_AINT_T__ tmp = val; __asm__ __volatile__( - "xchg" LONG_SUFFIX " %0, %1" + "xchg" ETHR_AINT_SUFFIX__ " %0, %1" : "=r"(tmp) : "m"(var->counter), "0"(tmp)); /* now tmp is the atomic's previous value */ @@ -167,57 +215,73 @@ ethr_native_atomic_xchg(ethr_native_atomic_t *var, long val) * Atomic ops with at least specified barriers. */ -static ETHR_INLINE long -ethr_native_atomic_read_acqb(ethr_native_atomic_t *var) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(read_acqb)(ETHR_ATMC_T__ *var) { - long val; + ETHR_AINT_T__ val; #if defined(__x86_64__) || !defined(ETHR_PRE_PENTIUM4_COMPAT) val = var->counter; #else - val = ethr_native_atomic_add_return(var, 0); + val = ETHR_NATMC_FUNC__(add_return)(var, 0); #endif __asm__ __volatile__("" : : : "memory"); return val; } static ETHR_INLINE void -ethr_native_atomic_set_relb(ethr_native_atomic_t *var, long i) +ETHR_NATMC_FUNC__(set_relb)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ i) { __asm__ __volatile__("" : : : "memory"); #if defined(__x86_64__) || !defined(ETHR_PRE_PENTIUM4_COMPAT) var->counter = i; #else - (void) ethr_native_atomic_xchg(var, i); + (void) ETHR_NATMC_FUNC__(xchg)(var, i); #endif } -static ETHR_INLINE long -ethr_native_atomic_inc_return_acqb(ethr_native_atomic_t *var) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(inc_return_acqb)(ETHR_ATMC_T__ *var) { - long res = ethr_native_atomic_inc_return(var); + ETHR_AINT_T__ res = ETHR_NATMC_FUNC__(inc_return)(var); __asm__ __volatile__("" : : : "memory"); return res; } static ETHR_INLINE void -ethr_native_atomic_dec_relb(ethr_native_atomic_t *var) +ETHR_NATMC_FUNC__(dec_relb)(ETHR_ATMC_T__ *var) { __asm__ __volatile__("" : : : "memory"); - ethr_native_atomic_dec(var); + ETHR_NATMC_FUNC__(dec)(var); } -static ETHR_INLINE long -ethr_native_atomic_dec_return_relb(ethr_native_atomic_t *var) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(dec_return_relb)(ETHR_ATMC_T__ *var) { __asm__ __volatile__("" : : : "memory"); - return ethr_native_atomic_dec_return(var); + return ETHR_NATMC_FUNC__(dec_return)(var); } -#define ethr_native_atomic_cmpxchg_acqb ethr_native_atomic_cmpxchg -#define ethr_native_atomic_cmpxchg_relb ethr_native_atomic_cmpxchg +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(cmpxchg_acqb)(ETHR_ATMC_T__ *var, + ETHR_AINT_T__ new, + ETHR_AINT_T__ old) +{ + return ETHR_NATMC_FUNC__(cmpxchg)(var, new, old); +} -#undef LONG_SUFFIX +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(cmpxchg_relb)(ETHR_ATMC_T__ *var, + ETHR_AINT_T__ new, + ETHR_AINT_T__ old) +{ + return ETHR_NATMC_FUNC__(cmpxchg)(var, new, old); +} #endif /* ETHR_TRY_INLINE_FUNCS */ -#endif /* ETHREAD_I386_ATOMIC_H */ +#undef ETHR_NATMC_FUNC__ +#undef ETHR_ATMC_T__ +#undef ETHR_AINT_T__ +#undef ETHR_AINT_SUFFIX__ + +#endif /* ETHR_INCLUDE_ATOMIC_IMPL__ */ diff --git a/erts/include/internal/i386/ethread.h b/erts/include/internal/i386/ethread.h index ed43e77279..b5a17caefb 100644 --- a/erts/include/internal/i386/ethread.h +++ b/erts/include/internal/i386/ethread.h @@ -1,7 +1,7 @@ /* * %CopyrightBegin% * - * Copyright Ericsson AB 2005-2009. All Rights Reserved. + * Copyright Ericsson AB 2005-2010. 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 @@ -24,7 +24,12 @@ #ifndef ETHREAD_I386_ETHREAD_H #define ETHREAD_I386_ETHREAD_H +#define ETHR_ATOMIC_WANT_32BIT_IMPL__ #include "atomic.h" +#if ETHR_SIZEOF_PTR == 8 +# define ETHR_ATOMIC_WANT_64BIT_IMPL__ +# include "atomic.h" +#endif #include "spinlock.h" #include "rwlock.h" diff --git a/erts/include/internal/libatomic_ops/ethr_atomic.h b/erts/include/internal/libatomic_ops/ethr_atomic.h index a6eb43a0bd..d56693dbf8 100644 --- a/erts/include/internal/libatomic_ops/ethr_atomic.h +++ b/erts/include/internal/libatomic_ops/ethr_atomic.h @@ -46,17 +46,39 @@ * - AO_store() * - AO_compare_and_swap() * - * The `AO_t' type also have to be at least as large as - * `void *' and `long' types. + * The `AO_t' type also have to be at least as large as the `void *' type. */ #if ETHR_SIZEOF_AO_T < ETHR_SIZEOF_PTR #error The AO_t type is too small #endif +#if ETHR_SIZEOF_AO_T == 4 +#define ETHR_HAVE_NATIVE_ATOMIC32 1 +#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_NATMC_FUNC__(X) ethr_native_atomic64_ ## X +#define ETHR_ATMC_T__ ethr_native_atomic64_t +#define ETHR_AINT_T__ ethr_sint64_t +#define ETHR_AINT_SUFFIX__ "q" +#else +#error "Unsupported integer size" +#endif + +#if ETHR_SIZEOF_AO_T == 8 +typedef union { + volatile AO_t counter; + ethr_sint32_t sint32[2]; +} ETHR_ATMC_T__; +#else typedef struct { volatile AO_t counter; -} ethr_native_atomic_t; +} ETHR_ATMC_T__; +#endif #define ETHR_MEMORY_BARRIER AO_nop_full() #ifdef AO_HAVE_nop_write @@ -72,123 +94,151 @@ typedef struct { #ifdef AO_NO_DD_ORDERING # define ETHR_READ_DEPEND_MEMORY_BARRIER ETHR_READ_MEMORY_BARRIER #else -# define ETHR_READ_DEPEND_MEMORY_BARRIER __asm__ __volatile__("":::"memory") +# define ETHR_READ_DEPEND_MEMORY_BARRIER AO_compiler_barrier() +#endif + +#if defined(ETHR_TRY_INLINE_FUNCS) || defined(ETHR_ATOMIC_IMPL__) + +static ETHR_INLINE ETHR_AINT_T__ * +ETHR_NATMC_FUNC__(addr)(ETHR_ATMC_T__ *var) +{ + return (ETHR_AINT_T__ *) &var->counter; +} + +#if ETHR_SIZEOF_AO_T == 8 +/* + * We also need to provide an ethr_native_atomic32_addr(), since + * this 64-bit implementation will be used implementing 32-bit + * native atomics. + */ + +static ETHR_INLINE ethr_sint32_t * +ethr_native_atomic32_addr(ETHR_ATMC_T__ *var) +{ + ETHR_ASSERT(((void *) &var->sint32[0]) == ((void *) &var->counter)); +#ifdef ETHR_BIGENDIAN + return &var->sint32[1]; +#else + return &var->sint32[0]; #endif +} -#if defined(ETHR_TRY_INLINE_FUNCS) || defined(ETHR_AUX_IMPL__) +#endif /* ETHR_SIZEOF_AO_T == 8 */ static ETHR_INLINE void -ethr_native_atomic_set(ethr_native_atomic_t *var, long value) +ETHR_NATMC_FUNC__(set)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ value) { AO_store(&var->counter, (AO_t) value); } static ETHR_INLINE void -ethr_native_atomic_init(ethr_native_atomic_t *var, long value) +ETHR_NATMC_FUNC__(init)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ value) { - ethr_native_atomic_set(var, value); + ETHR_NATMC_FUNC__(set)(var, value); } -static ETHR_INLINE long -ethr_native_atomic_read(ethr_native_atomic_t *var) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(read)(ETHR_ATMC_T__ *var) { - return (long) AO_load(&var->counter); + return (ETHR_AINT_T__) AO_load(&var->counter); } -static ETHR_INLINE long -ethr_native_atomic_add_return(ethr_native_atomic_t *var, long incr) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(add_return)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ incr) { #ifdef AO_HAVE_fetch_and_add - return ((long) AO_fetch_and_add(&var->counter, (AO_t) incr)) + incr; + return ((ETHR_AINT_T__) AO_fetch_and_add(&var->counter, (AO_t) incr)) + incr; #else while (1) { AO_t exp = AO_load(&var->counter); AO_t new = exp + (AO_t) incr; if (AO_compare_and_swap(&var->counter, exp, new)) - return (long) new; + return (ETHR_AINT_T__) new; } #endif } static ETHR_INLINE void -ethr_native_atomic_add(ethr_native_atomic_t *var, long incr) +ETHR_NATMC_FUNC__(add)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ incr) { - (void) ethr_native_atomic_add_return(var, incr); + (void) ETHR_NATMC_FUNC__(add_return)(var, incr); } -static ETHR_INLINE long -ethr_native_atomic_inc_return(ethr_native_atomic_t *var) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(inc_return)(ETHR_ATMC_T__ *var) { #ifdef AO_HAVE_fetch_and_add1 - return ((long) AO_fetch_and_add1(&var->counter)) + 1; + return ((ETHR_AINT_T__) AO_fetch_and_add1(&var->counter)) + 1; #else - return ethr_native_atomic_add_return(var, 1); + return ETHR_NATMC_FUNC__(add_return)(var, 1); #endif } static ETHR_INLINE void -ethr_native_atomic_inc(ethr_native_atomic_t *var) +ETHR_NATMC_FUNC__(inc)(ETHR_ATMC_T__ *var) { - (void) ethr_native_atomic_inc_return(var); + (void) ETHR_NATMC_FUNC__(inc_return)(var); } -static ETHR_INLINE long -ethr_native_atomic_dec_return(ethr_native_atomic_t *var) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(dec_return)(ETHR_ATMC_T__ *var) { #ifdef AO_HAVE_fetch_and_sub1 - return ((long) AO_fetch_and_sub1(&var->counter)) - 1; + return ((ETHR_AINT_T__) AO_fetch_and_sub1(&var->counter)) - 1; #else - return ethr_native_atomic_add_return(var, -1); + return ETHR_NATMC_FUNC__(add_return)(var, -1); #endif } static ETHR_INLINE void -ethr_native_atomic_dec(ethr_native_atomic_t *var) +ETHR_NATMC_FUNC__(dec)(ETHR_ATMC_T__ *var) { - (void) ethr_native_atomic_dec_return(var); + (void) ETHR_NATMC_FUNC__(dec_return)(var); } -static ETHR_INLINE long -ethr_native_atomic_and_retold(ethr_native_atomic_t *var, long mask) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(and_retold)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ mask) { while (1) { AO_t exp = AO_load(&var->counter); AO_t new = exp & ((AO_t) mask); if (AO_compare_and_swap(&var->counter, exp, new)) - return (long) exp; + return (ETHR_AINT_T__) exp; } } -static ETHR_INLINE long -ethr_native_atomic_or_retold(ethr_native_atomic_t *var, long mask) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(or_retold)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ mask) { while (1) { AO_t exp = AO_load(&var->counter); AO_t new = exp | ((AO_t) mask); if (AO_compare_and_swap(&var->counter, exp, new)) - return (long) exp; + return (ETHR_AINT_T__) exp; } } -static ETHR_INLINE long -ethr_native_atomic_cmpxchg(ethr_native_atomic_t *var, long new, long exp) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(cmpxchg)(ETHR_ATMC_T__ *var, + ETHR_AINT_T__ new, + ETHR_AINT_T__ exp) { - long act; + ETHR_AINT_T__ act; do { if (AO_compare_and_swap(&var->counter, (AO_t) exp, (AO_t) new)) return exp; - act = (long) AO_load(&var->counter); + act = (ETHR_AINT_T__) AO_load(&var->counter); } while (act == exp); return act; } -static ETHR_INLINE long -ethr_native_atomic_xchg(ethr_native_atomic_t *var, long new) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(xchg)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ new) { while (1) { AO_t exp = AO_load(&var->counter); if (AO_compare_and_swap(&var->counter, exp, (AO_t) new)) - return (long) exp; + return (ETHR_AINT_T__) exp; } } @@ -196,97 +246,105 @@ ethr_native_atomic_xchg(ethr_native_atomic_t *var, long new) * Atomic ops with at least specified barriers. */ -static ETHR_INLINE long -ethr_native_atomic_read_acqb(ethr_native_atomic_t *var) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(read_acqb)(ETHR_ATMC_T__ *var) { #ifdef AO_HAVE_load_acquire - return (long) AO_load_acquire(&var->counter); + return (ETHR_AINT_T__) AO_load_acquire(&var->counter); #else - long res = ethr_native_atomic_read(var); + ETHR_AINT_T__ res = ETHR_NATMC_FUNC__(read)(var); ETHR_MEMORY_BARRIER; return res; #endif } -static ETHR_INLINE long -ethr_native_atomic_inc_return_acqb(ethr_native_atomic_t *var) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(inc_return_acqb)(ETHR_ATMC_T__ *var) { #ifdef AO_HAVE_fetch_and_add1_acquire - return ((long) AO_fetch_and_add1_acquire(&var->counter)) + 1; + return ((ETHR_AINT_T__) AO_fetch_and_add1_acquire(&var->counter)) + 1; #else - long res = ethr_native_atomic_add_return(var, 1); + ETHR_AINT_T__ res = ETHR_NATMC_FUNC__(add_return)(var, 1); ETHR_MEMORY_BARRIER; return res; #endif } static ETHR_INLINE void -ethr_native_atomic_set_relb(ethr_native_atomic_t *var, long value) +ETHR_NATMC_FUNC__(set_relb)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ value) { #ifdef AO_HAVE_store_release AO_store_release(&var->counter, (AO_t) value); #else ETHR_MEMORY_BARRIER; - ethr_native_atomic_set(var, value); + ETHR_NATMC_FUNC__(set)(var, value); #endif } -static ETHR_INLINE long -ethr_native_atomic_dec_return_relb(ethr_native_atomic_t *var) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(dec_return_relb)(ETHR_ATMC_T__ *var) { #ifdef AO_HAVE_fetch_and_sub1_release - return ((long) AO_fetch_and_sub1_release(&var->counter)) - 1; + return ((ETHR_AINT_T__) AO_fetch_and_sub1_release(&var->counter)) - 1; #else ETHR_MEMORY_BARRIER; - return ethr_native_atomic_dec_return(var); + return ETHR_NATMC_FUNC__(dec_return)(var); #endif } static ETHR_INLINE void -ethr_native_atomic_dec_relb(ethr_native_atomic_t *var) +ETHR_NATMC_FUNC__(dec_relb)(ETHR_ATMC_T__ *var) { - (void) ethr_native_atomic_dec_return_relb(var); + (void) ETHR_NATMC_FUNC__(dec_return_relb)(var); } -static ETHR_INLINE long -ethr_native_atomic_cmpxchg_acqb(ethr_native_atomic_t *var, long new, long exp) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(cmpxchg_acqb)(ETHR_ATMC_T__ *var, + ETHR_AINT_T__ new, + ETHR_AINT_T__ exp) { #ifdef AO_HAVE_compare_and_swap_acquire - long act; + ETHR_AINT_T__ act; do { if (AO_compare_and_swap_acquire(&var->counter, (AO_t) exp, (AO_t) new)) return exp; - act = (long) AO_load(&var->counter); + act = (ETHR_AINT_T__) AO_load(&var->counter); } while (act == exp); AO_nop_full(); return act; #else - long act = ethr_native_atomic_cmpxchg(var, new, exp); + ETHR_AINT_T__ act = ETHR_NATMC_FUNC__(cmpxchg)(var, new, exp); ETHR_MEMORY_BARRIER; return act; #endif } -static ETHR_INLINE long -ethr_native_atomic_cmpxchg_relb(ethr_native_atomic_t *var, long new, long exp) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(cmpxchg_relb)(ETHR_ATMC_T__ *var, + ETHR_AINT_T__ new, + ETHR_AINT_T__ exp) { #ifdef AO_HAVE_compare_and_swap_release - long act; + ETHR_AINT_T__ act; do { if (AO_compare_and_swap_release(&var->counter, (AO_t) exp, (AO_t) new)) return exp; - act = (long) AO_load(&var->counter); + act = (ETHR_AINT_T__) AO_load(&var->counter); } while (act == exp); return act; #else ETHR_MEMORY_BARRIER; - return ethr_native_atomic_cmpxchg(var, new, exp); + return ETHR_NATMC_FUNC__(cmpxchg)(var, new, exp); #endif } -#endif +#endif /* ETHR_TRY_INLINE_FUNCS */ -#endif +#undef ETHR_NATMC_FUNC__ +#undef ETHR_ATMC_T__ +#undef ETHR_AINT_T__ -#endif +#endif /* !defined(ETHR_HAVE_NATIVE_ATOMICS) && defined(ETHR_HAVE_LIBATOMIC_OPS) */ + +#endif /* ETHR_LIBATOMIC_OPS_ATOMIC_H__ */ diff --git a/erts/include/internal/ppc32/atomic.h b/erts/include/internal/ppc32/atomic.h index f21f7c9588..522f433649 100644 --- a/erts/include/internal/ppc32/atomic.h +++ b/erts/include/internal/ppc32/atomic.h @@ -28,31 +28,39 @@ #ifndef ETHREAD_PPC_ATOMIC_H #define ETHREAD_PPC_ATOMIC_H +#define ETHR_HAVE_NATIVE_ATOMIC32 1 + typedef struct { - volatile int counter; -} ethr_native_atomic_t; + volatile ethr_sint32_t counter; +} ethr_native_atomic32_t; #define ETHR_MEMORY_BARRIER __asm__ __volatile__("sync" : : : "memory") -#if defined(ETHR_TRY_INLINE_FUNCS) || defined(ETHR_AUX_IMPL__) +#if defined(ETHR_TRY_INLINE_FUNCS) || defined(ETHR_ATOMIC_IMPL__) + +static ETHR_INLINE ethr_sint32_t * +ethr_native_atomic32_addr(ethr_native_atomic32_t *var) +{ + return (ethr_sint32_t *) &var->counter; +} static ETHR_INLINE void -ethr_native_atomic_init(ethr_native_atomic_t *var, int i) +ethr_native_atomic32_init(ethr_native_atomic32_t *var, ethr_sint32_t i) { var->counter = i; } -#define ethr_native_atomic_set(v, i) ethr_native_atomic_init((v), (i)) +#define ethr_native_atomic32_set(v, i) ethr_native_atomic32_init((v), (i)) -static ETHR_INLINE int -ethr_native_atomic_read(ethr_native_atomic_t *var) +static ETHR_INLINE ethr_sint32_t +ethr_native_atomic32_read(ethr_native_atomic32_t *var) { return var->counter; } -static ETHR_INLINE int -ethr_native_atomic_add_return(ethr_native_atomic_t *var, int incr) +static ETHR_INLINE ethr_sint32_t +ethr_native_atomic32_add_return(ethr_native_atomic32_t *var, ethr_sint32_t incr) { - int tmp; + ethr_sint32_t tmp; __asm__ __volatile__( "eieio\n\t" @@ -69,16 +77,16 @@ ethr_native_atomic_add_return(ethr_native_atomic_t *var, int incr) } static ETHR_INLINE void -ethr_native_atomic_add(ethr_native_atomic_t *var, int incr) +ethr_native_atomic32_add(ethr_native_atomic32_t *var, ethr_sint32_t incr) { /* XXX: could use weaker version here w/o eieio+isync */ - (void)ethr_native_atomic_add_return(var, incr); + (void)ethr_native_atomic32_add_return(var, incr); } -static ETHR_INLINE int -ethr_native_atomic_inc_return(ethr_native_atomic_t *var) +static ETHR_INLINE ethr_sint32_t +ethr_native_atomic32_inc_return(ethr_native_atomic32_t *var) { - int tmp; + ethr_sint32_t tmp; __asm__ __volatile__( "eieio\n\t" @@ -95,16 +103,16 @@ ethr_native_atomic_inc_return(ethr_native_atomic_t *var) } static ETHR_INLINE void -ethr_native_atomic_inc(ethr_native_atomic_t *var) +ethr_native_atomic32_inc(ethr_native_atomic32_t *var) { /* XXX: could use weaker version here w/o eieio+isync */ - (void)ethr_native_atomic_inc_return(var); + (void)ethr_native_atomic32_inc_return(var); } -static ETHR_INLINE int -ethr_native_atomic_dec_return(ethr_native_atomic_t *var) +static ETHR_INLINE ethr_sint32_t +ethr_native_atomic32_dec_return(ethr_native_atomic32_t *var) { - int tmp; + ethr_sint32_t tmp; __asm__ __volatile__( "eieio\n\t" @@ -121,16 +129,16 @@ ethr_native_atomic_dec_return(ethr_native_atomic_t *var) } static ETHR_INLINE void -ethr_native_atomic_dec(ethr_native_atomic_t *var) +ethr_native_atomic32_dec(ethr_native_atomic32_t *var) { /* XXX: could use weaker version here w/o eieio+isync */ - (void)ethr_native_atomic_dec_return(var); + (void)ethr_native_atomic32_dec_return(var); } -static ETHR_INLINE int -ethr_native_atomic_and_retold(ethr_native_atomic_t *var, int mask) +static ETHR_INLINE ethr_sint32_t +ethr_native_atomic32_and_retold(ethr_native_atomic32_t *var, ethr_sint32_t mask) { - int old, new; + ethr_sint32_t old, new; __asm__ __volatile__( "eieio\n\t" @@ -146,10 +154,10 @@ ethr_native_atomic_and_retold(ethr_native_atomic_t *var, int mask) return old; } -static ETHR_INLINE int -ethr_native_atomic_or_retold(ethr_native_atomic_t *var, int mask) +static ETHR_INLINE ethr_sint32_t +ethr_native_atomic32_or_retold(ethr_native_atomic32_t *var, ethr_sint32_t mask) { - int old, new; + ethr_sint32_t old, new; __asm__ __volatile__( "eieio\n\t" @@ -165,10 +173,10 @@ ethr_native_atomic_or_retold(ethr_native_atomic_t *var, int mask) return old; } -static ETHR_INLINE int -ethr_native_atomic_xchg(ethr_native_atomic_t *var, int val) +static ETHR_INLINE ethr_sint32_t +ethr_native_atomic32_xchg(ethr_native_atomic32_t *var, ethr_sint32_t val) { - int tmp; + ethr_sint32_t tmp; __asm__ __volatile__( "eieio\n\t" @@ -183,10 +191,12 @@ ethr_native_atomic_xchg(ethr_native_atomic_t *var, int val) return tmp; } -static ETHR_INLINE int -ethr_native_atomic_cmpxchg(ethr_native_atomic_t *var, int new, int expected) +static ETHR_INLINE ethr_sint32_t +ethr_native_atomic32_cmpxchg(ethr_native_atomic32_t *var, + ethr_sint32_t new, + ethr_sint32_t expected) { - int old; + ethr_sint32_t old; __asm__ __volatile__( "eieio\n\t" @@ -210,20 +220,20 @@ ethr_native_atomic_cmpxchg(ethr_native_atomic_t *var, int new, int expected) */ static ETHR_INLINE long -ethr_native_atomic_read_acqb(ethr_native_atomic_t *var) +ethr_native_atomic32_read_acqb(ethr_native_atomic32_t *var) { - long res = ethr_native_atomic_read(var); + long res = ethr_native_atomic32_read(var); ETHR_MEMORY_BARRIER; return res; } -#define ethr_native_atomic_set_relb ethr_native_atomic_xchg -#define ethr_native_atomic_inc_return_acqb ethr_native_atomic_inc_return -#define ethr_native_atomic_dec_relb ethr_native_atomic_dec_return -#define ethr_native_atomic_dec_return_relb ethr_native_atomic_dec_return +#define ethr_native_atomic32_set_relb ethr_native_atomic32_xchg +#define ethr_native_atomic32_inc_return_acqb ethr_native_atomic32_inc_return +#define ethr_native_atomic32_dec_relb ethr_native_atomic32_dec_return +#define ethr_native_atomic32_dec_return_relb ethr_native_atomic32_dec_return -#define ethr_native_atomic_cmpxchg_acqb ethr_native_atomic_cmpxchg -#define ethr_native_atomic_cmpxchg_relb ethr_native_atomic_cmpxchg +#define ethr_native_atomic32_cmpxchg_acqb ethr_native_atomic32_cmpxchg +#define ethr_native_atomic32_cmpxchg_relb ethr_native_atomic32_cmpxchg #endif /* ETHR_TRY_INLINE_FUNCS */ diff --git a/erts/include/internal/pthread/ethr_event.h b/erts/include/internal/pthread/ethr_event.h index 104ec287e0..93da8a0429 100644 --- a/erts/include/internal/pthread/ethr_event.h +++ b/erts/include/internal/pthread/ethr_event.h @@ -30,31 +30,9 @@ #include <linux/futex.h> #include <sys/time.h> -/* - * Note: Linux futexes operate on 32-bit integers, but - * ethr_native_atomic_t are 64-bits on 64-bit - * platforms. This has to be taken into account. - * Therefore, in each individual value used each - * byte look the same. - */ - -#if ETHR_SIZEOF_PTR == 8 - -#define ETHR_EVENT_OFF_WAITER__ 0xffffffffffffffffL -#define ETHR_EVENT_OFF__ 0x7777777777777777L -#define ETHR_EVENT_ON__ 0L - -#elif ETHR_SIZEOF_PTR == 4 - -#define ETHR_EVENT_OFF_WAITER__ 0xffffffffL -#define ETHR_EVENT_OFF__ 0x77777777L -#define ETHR_EVENT_ON__ 0L - -#else - -#error ehrm... - -#endif +#define ETHR_EVENT_OFF_WAITER__ ((ethr_sint32_t) -1) +#define ETHR_EVENT_OFF__ ((ethr_sint32_t) 1) +#define ETHR_EVENT_ON__ ((ethr_sint32_t) 0) #if defined(FUTEX_WAIT_PRIVATE) && defined(FUTEX_WAKE_PRIVATE) # define ETHR_FUTEX_WAIT__ FUTEX_WAIT_PRIVATE @@ -65,11 +43,17 @@ #endif typedef struct { - ethr_atomic_t futex; + ethr_atomic32_t futex; } ethr_event; -#define ETHR_FUTEX__(FTX, OP, VAL) \ - (-1 == syscall(__NR_futex, (void *) (FTX), (OP), (int) (VAL), NULL, NULL, 0)\ +#define ETHR_FUTEX__(FTX, OP, VAL) \ + (-1 == syscall(__NR_futex, \ + (void *) ethr_atomic32_addr((FTX)), \ + (OP), \ + (int) (VAL), \ + NULL, \ + NULL, \ + 0) \ ? errno : 0) #if defined(ETHR_TRY_INLINE_FUNCS) || defined(ETHR_EVENT_IMPL__) @@ -77,9 +61,9 @@ typedef struct { static void ETHR_INLINE ETHR_INLINE_FUNC_NAME_(ethr_event_set)(ethr_event *e) { - long val; + ethr_sint32_t val; ETHR_WRITE_MEMORY_BARRIER; - val = ethr_atomic_xchg(&e->futex, ETHR_EVENT_ON__); + val = ethr_atomic32_xchg(&e->futex, ETHR_EVENT_ON__); if (val == ETHR_EVENT_OFF_WAITER__) { int res = ETHR_FUTEX__(&e->futex, ETHR_FUTEX_WAKE__, 1); if (res != 0) @@ -90,7 +74,7 @@ ETHR_INLINE_FUNC_NAME_(ethr_event_set)(ethr_event *e) static void ETHR_INLINE ETHR_INLINE_FUNC_NAME_(ethr_event_reset)(ethr_event *e) { - ethr_atomic_set(&e->futex, ETHR_EVENT_OFF__); + ethr_atomic32_set(&e->futex, ETHR_EVENT_OFF__); ETHR_MEMORY_BARRIER; } @@ -100,7 +84,7 @@ ETHR_INLINE_FUNC_NAME_(ethr_event_reset)(ethr_event *e) /* --- Posix mutex/cond implementation of events ---------------------------- */ typedef struct { - ethr_atomic_t state; + ethr_atomic32_t state; pthread_mutex_t mtx; pthread_cond_t cnd; } ethr_event; @@ -114,9 +98,9 @@ typedef struct { static void ETHR_INLINE ETHR_INLINE_FUNC_NAME_(ethr_event_set)(ethr_event *e) { - long val; + ethr_sint32_t val; ETHR_WRITE_MEMORY_BARRIER; - val = ethr_atomic_xchg(&e->state, ETHR_EVENT_ON__); + val = ethr_atomic32_xchg(&e->state, ETHR_EVENT_ON__); if (val == ETHR_EVENT_OFF_WAITER__) { int res = pthread_mutex_lock(&e->mtx); if (res != 0) @@ -133,7 +117,7 @@ ETHR_INLINE_FUNC_NAME_(ethr_event_set)(ethr_event *e) static void ETHR_INLINE ETHR_INLINE_FUNC_NAME_(ethr_event_reset)(ethr_event *e) { - ethr_atomic_set(&e->state, ETHR_EVENT_OFF__); + ethr_atomic32_set(&e->state, ETHR_EVENT_OFF__); ETHR_MEMORY_BARRIER; } diff --git a/erts/include/internal/sparc32/atomic.h b/erts/include/internal/sparc32/atomic.h index 2da6472393..00380dbf07 100644 --- a/erts/include/internal/sparc32/atomic.h +++ b/erts/include/internal/sparc32/atomic.h @@ -21,49 +21,86 @@ * Native ethread atomics on SPARC V9. * Author: Mikael Pettersson. */ -#ifndef ETHR_SPARC32_ATOMIC_H -#define ETHR_SPARC32_ATOMIC_H -typedef struct { - volatile long counter; -} ethr_native_atomic_t; +#undef ETHR_INCLUDE_ATOMIC_IMPL__ +#if !defined(ETHR_SPARC_V9_ATOMIC32_H__) && defined(ETHR_ATOMIC_WANT_32BIT_IMPL__) +#define ETHR_SPARC_V9_ATOMIC32_H__ +#define ETHR_INCLUDE_ATOMIC_IMPL__ 4 +#undef ETHR_ATOMIC_WANT_32BIT_IMPL__ +#elif !defined(ETHR_SPARC_V9_ATOMIC64_H__) && defined(ETHR_ATOMIC_WANT_64BIT_IMPL__) +#define ETHR_SPARC_V9_ATOMIC64_H__ +#define ETHR_INCLUDE_ATOMIC_IMPL__ 8 +#undef ETHR_ATOMIC_WANT_64BIT_IMPL__ +#endif + +#ifdef ETHR_INCLUDE_ATOMIC_IMPL__ + +#ifndef ETHR_SPARC_V9_ATOMIC_COMMON__ +#define ETHR_SPARC_V9_ATOMIC_COMMON__ #define ETHR_MEMORY_BARRIER \ __asm__ __volatile__("membar #LoadLoad|#LoadStore|#StoreLoad|#StoreStore\n" \ : : : "memory") -#if defined(ETHR_TRY_INLINE_FUNCS) || defined(ETHR_AUX_IMPL__) - -#if defined(__arch64__) -#define CASX "casx" +#endif /* ETHR_SPARC_V9_ATOMIC_COMMON__ */ + +#if ETHR_INCLUDE_ATOMIC_IMPL__ == 4 +#define ETHR_HAVE_NATIVE_ATOMIC32 1 +#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_CAS__ "cas" +#elif ETHR_INCLUDE_ATOMIC_IMPL__ == 8 +#define ETHR_HAVE_NATIVE_ATOMIC64 1 +#define ETHR_NATMC_FUNC__(X) ethr_native_atomic64_ ## X +#define ETHR_ATMC_T__ ethr_native_atomic64_t +#define ETHR_AINT_T__ ethr_sint64_t +#define ETHR_CAS__ "casx" #else -#define CASX "cas" +#error "Unsupported integer size" #endif +typedef struct { + volatile ETHR_AINT_T__ counter; +} ETHR_ATMC_T__; + +#if defined(ETHR_TRY_INLINE_FUNCS) || defined(ETHR_ATOMIC_IMPL__) + +static ETHR_INLINE ETHR_AINT_T__ * +ETHR_NATMC_FUNC__(addr)(ETHR_ATMC_T__ *var) +{ + return (ETHR_AINT_T__ *) &var->counter; +} + +static ETHR_INLINE void +ETHR_NATMC_FUNC__(init)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ i) +{ + var->counter = i; +} + static ETHR_INLINE void -ethr_native_atomic_init(ethr_native_atomic_t *var, long i) +ETHR_NATMC_FUNC__(set)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ i) { var->counter = i; } -#define ethr_native_atomic_set(v, i) ethr_native_atomic_init((v), (i)) -static ETHR_INLINE long -ethr_native_atomic_read(ethr_native_atomic_t *var) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(read)(ETHR_ATMC_T__ *var) { return var->counter; } -static ETHR_INLINE long -ethr_native_atomic_add_return(ethr_native_atomic_t *var, long incr) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(add_return)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ incr) { - long old, tmp; + ETHR_AINT_T__ old, tmp; __asm__ __volatile__("membar #LoadLoad|#StoreLoad\n"); do { old = var->counter; tmp = old+incr; __asm__ __volatile__( - CASX " [%2], %1, %0" + ETHR_CAS__ " [%2], %1, %0" : "=&r"(tmp) : "r"(old), "r"(&var->counter), "0"(tmp) : "memory"); @@ -73,46 +110,46 @@ ethr_native_atomic_add_return(ethr_native_atomic_t *var, long incr) } static ETHR_INLINE void -ethr_native_atomic_add(ethr_native_atomic_t *var, long incr) +ETHR_NATMC_FUNC__(add)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ incr) { - (void)ethr_native_atomic_add_return(var, incr); + (void)ETHR_NATMC_FUNC__(add_return)(var, incr); } -static ETHR_INLINE long -ethr_native_atomic_inc_return(ethr_native_atomic_t *var) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(inc_return)(ETHR_ATMC_T__ *var) { - return ethr_native_atomic_add_return(var, 1); + return ETHR_NATMC_FUNC__(add_return)(var, 1); } static ETHR_INLINE void -ethr_native_atomic_inc(ethr_native_atomic_t *var) +ETHR_NATMC_FUNC__(inc)(ETHR_ATMC_T__ *var) { - (void)ethr_native_atomic_add_return(var, 1); + (void)ETHR_NATMC_FUNC__(add_return)(var, 1); } -static ETHR_INLINE long -ethr_native_atomic_dec_return(ethr_native_atomic_t *var) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(dec_return)(ETHR_ATMC_T__ *var) { - return ethr_native_atomic_add_return(var, -1); + return ETHR_NATMC_FUNC__(add_return)(var, -1); } static ETHR_INLINE void -ethr_native_atomic_dec(ethr_native_atomic_t *var) +ETHR_NATMC_FUNC__(dec)(ETHR_ATMC_T__ *var) { - (void)ethr_native_atomic_add_return(var, -1); + (void)ETHR_NATMC_FUNC__(add_return)(var, -1); } -static ETHR_INLINE long -ethr_native_atomic_and_retold(ethr_native_atomic_t *var, long mask) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(and_retold)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ mask) { - long old, tmp; + ETHR_AINT_T__ old, tmp; __asm__ __volatile__("membar #LoadLoad|#StoreLoad\n"); do { old = var->counter; tmp = old & mask; __asm__ __volatile__( - CASX " [%2], %1, %0" + ETHR_CAS__ " [%2], %1, %0" : "=&r"(tmp) : "r"(old), "r"(&var->counter), "0"(tmp) : "memory"); @@ -121,17 +158,17 @@ ethr_native_atomic_and_retold(ethr_native_atomic_t *var, long mask) return old; } -static ETHR_INLINE long -ethr_native_atomic_or_retold(ethr_native_atomic_t *var, long mask) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(or_retold)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ mask) { - long old, tmp; + ETHR_AINT_T__ old, tmp; __asm__ __volatile__("membar #LoadLoad|#StoreLoad\n"); do { old = var->counter; tmp = old | mask; __asm__ __volatile__( - CASX " [%2], %1, %0" + ETHR_CAS__ " [%2], %1, %0" : "=&r"(tmp) : "r"(old), "r"(&var->counter), "0"(tmp) : "memory"); @@ -140,17 +177,17 @@ ethr_native_atomic_or_retold(ethr_native_atomic_t *var, long mask) return old; } -static ETHR_INLINE long -ethr_native_atomic_xchg(ethr_native_atomic_t *var, long val) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(xchg)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ val) { - long old, new; + ETHR_AINT_T__ old, new; __asm__ __volatile__("membar #LoadLoad|#StoreLoad"); do { old = var->counter; new = val; __asm__ __volatile__( - CASX " [%2], %1, %0" + ETHR_CAS__ " [%2], %1, %0" : "=&r"(new) : "r"(old), "r"(&var->counter), "0"(new) : "memory"); @@ -159,12 +196,12 @@ ethr_native_atomic_xchg(ethr_native_atomic_t *var, long val) return old; } -static ETHR_INLINE long -ethr_native_atomic_cmpxchg(ethr_native_atomic_t *var, long new, long old) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(cmpxchg)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ new, ETHR_AINT_T__ old) { __asm__ __volatile__("membar #LoadLoad|#StoreLoad\n"); __asm__ __volatile__( - CASX " [%2], %1, %0" + ETHR_CAS__ " [%2], %1, %0" : "=&r"(new) : "r"(old), "r"(&var->counter), "0"(new) : "memory"); @@ -178,58 +215,63 @@ ethr_native_atomic_cmpxchg(ethr_native_atomic_t *var, long new, long old) /* TODO: relax acquire barriers */ -static ETHR_INLINE long -ethr_native_atomic_read_acqb(ethr_native_atomic_t *var) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(read_acqb)(ETHR_ATMC_T__ *var) { - long res = ethr_native_atomic_read(var); + ETHR_AINT_T__ res = ETHR_NATMC_FUNC__(read)(var); __asm__ __volatile__("membar #LoadLoad|#LoadStore|#StoreLoad|#StoreStore" : : : "memory"); return res; } static ETHR_INLINE void -ethr_native_atomic_set_relb(ethr_native_atomic_t *var, long i) +ETHR_NATMC_FUNC__(set_relb)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ i) { __asm__ __volatile__("membar #LoadStore|#StoreStore" : : : "memory"); - ethr_native_atomic_set(var, i); + ETHR_NATMC_FUNC__(set)(var, i); } -static ETHR_INLINE long -ethr_native_atomic_inc_return_acqb(ethr_native_atomic_t *var) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(inc_return_acqb)(ETHR_ATMC_T__ *var) { - long res = ethr_native_atomic_inc_return(var); + ETHR_AINT_T__ res = ETHR_NATMC_FUNC__(inc_return)(var); __asm__ __volatile__("membar #LoadLoad|#LoadStore" : : : "memory"); return res; } static ETHR_INLINE void -ethr_native_atomic_dec_relb(ethr_native_atomic_t *var) +ETHR_NATMC_FUNC__(dec_relb)(ETHR_ATMC_T__ *var) { __asm__ __volatile__("membar #LoadStore|#StoreStore" : : : "memory"); - ethr_native_atomic_dec(var); + ETHR_NATMC_FUNC__(dec)(var); } -static ETHR_INLINE long -ethr_native_atomic_dec_return_relb(ethr_native_atomic_t *var) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(dec_return_relb)(ETHR_ATMC_T__ *var) { __asm__ __volatile__("membar #LoadStore|#StoreStore" : : : "memory"); - return ethr_native_atomic_dec_return(var); + return ETHR_NATMC_FUNC__(dec_return)(var); } -static ETHR_INLINE long -ethr_native_atomic_cmpxchg_acqb(ethr_native_atomic_t *var, long new, long old) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(cmpxchg_acqb)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ new, ETHR_AINT_T__ old) { - long res = ethr_native_atomic_cmpxchg(var, new, old); + ETHR_AINT_T__ res = ETHR_NATMC_FUNC__(cmpxchg)(var, new, old); __asm__ __volatile__("membar #LoadLoad|#LoadStore" : : : "memory"); return res; } -static ETHR_INLINE long -ethr_native_atomic_cmpxchg_relb(ethr_native_atomic_t *var, long new, long old) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(cmpxchg_relb)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ new, ETHR_AINT_T__ old) { __asm__ __volatile__("membar #LoadStore|#StoreStore" : : : "memory"); - return ethr_native_atomic_cmpxchg(var, new, old); + return ETHR_NATMC_FUNC__(cmpxchg)(var, new, old); } #endif /* ETHR_TRY_INLINE_FUNCS */ -#endif /* ETHR_SPARC32_ATOMIC_H */ +#undef ETHR_NATMC_FUNC__ +#undef ETHR_ATMC_T__ +#undef ETHR_AINT_T__ +#undef ETHR_CAS__ + +#endif /* ETHR_INCLUDE_ATOMIC_IMPL__ */ diff --git a/erts/include/internal/sparc32/ethread.h b/erts/include/internal/sparc32/ethread.h index dca113b4d6..aea9794390 100644 --- a/erts/include/internal/sparc32/ethread.h +++ b/erts/include/internal/sparc32/ethread.h @@ -1,7 +1,7 @@ /* * %CopyrightBegin% * - * Copyright Ericsson AB 2005-2009. All Rights Reserved. + * Copyright Ericsson AB 2005-2010. 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 @@ -24,7 +24,12 @@ #ifndef ETHREAD_SPARC32_ETHREAD_H #define ETHREAD_SPARC32_ETHREAD_H +#define ETHR_ATOMIC_WANT_32BIT_IMPL__ #include "atomic.h" +#if ETHR_SIZEOF_PTR == 8 +# define ETHR_ATOMIC_WANT_64BIT_IMPL__ +# include "atomic.h" +#endif #include "spinlock.h" #include "rwlock.h" diff --git a/erts/include/internal/tile/atomic.h b/erts/include/internal/tile/atomic.h index 69569d82d1..48e4c0c6c8 100644 --- a/erts/include/internal/tile/atomic.h +++ b/erts/include/internal/tile/atomic.h @@ -24,92 +24,102 @@ #ifndef ETHREAD_TILE_ATOMIC_H #define ETHREAD_TILE_ATOMIC_H +#define ETHR_HAVE_NATIVE_ATOMIC32 1 + #include <atomic.h> /* An atomic is an aligned int accessed via locked operations. */ typedef struct { - volatile long counter; -} ethr_native_atomic_t; + volatile ethr_sint32_t counter; +} ethr_native_atomic32_t; #define ETHR_MEMORY_BARRIER __insn_mf() -#if defined(ETHR_TRY_INLINE_FUNCS) || defined(ETHR_AUX_IMPL__) +#if defined(ETHR_TRY_INLINE_FUNCS) || defined(ETHR_ATOMIC_IMPL__) + +static ETHR_INLINE ethr_sint32_t * +ethr_native_atomic32_addr(ethr_native_atomic32_t *var) +{ + return (ethr_sint32_t *) &var->counter; +} static ETHR_INLINE void -ethr_native_atomic_init(ethr_native_atomic_t *var, long i) +ethr_native_atomic32_init(ethr_native_atomic32_t *var, ethr_sint32_t i) { var->counter = i; } static ETHR_INLINE void -ethr_native_atomic_set(ethr_native_atomic_t *var, long i) +ethr_native_atomic32_set(ethr_native_atomic32_t *var, ethr_sint32_t i) { atomic_exchange_acq(&var->counter, i); } -static ETHR_INLINE long -ethr_native_atomic_read(ethr_native_atomic_t *var) +static ETHR_INLINE ethr_sint32_t +ethr_native_atomic32_read(ethr_native_atomic32_t *var) { return var->counter; } static ETHR_INLINE void -ethr_native_atomic_add(ethr_native_atomic_t *var, long incr) +ethr_native_atomic32_add(ethr_native_atomic32_t *var, ethr_sint32_t incr) { atomic_add(&var->counter, incr); } static ETHR_INLINE void -ethr_native_atomic_inc(ethr_native_atomic_t *var) +ethr_native_atomic32_inc(ethr_native_atomic32_t *var) { atomic_increment(&var->counter); } static ETHR_INLINE void -ethr_native_atomic_dec(ethr_native_atomic_t *var) +ethr_native_atomic32_dec(ethr_native_atomic32_t *var) { atomic_decrement(&var->counter); } -static ETHR_INLINE long -ethr_native_atomic_add_return(ethr_native_atomic_t *var, long incr) +static ETHR_INLINE ethr_sint32_t +ethr_native_atomic32_add_return(ethr_native_atomic32_t *var, ethr_sint32_t incr) { return atomic_exchange_and_add(&var->counter, incr) + incr; } -static ETHR_INLINE long -ethr_native_atomic_inc_return(ethr_native_atomic_t *var) +static ETHR_INLINE ethr_sint32_t +ethr_native_atomic32_inc_return(ethr_native_atomic32_t *var) { - return ethr_native_atomic_add_return(var, 1); + return ethr_native_atomic32_add_return(var, 1); } -static ETHR_INLINE long -ethr_native_atomic_dec_return(ethr_native_atomic_t *var) +static ETHR_INLINE ethr_sint32_t +ethr_native_atomic32_dec_return(ethr_native_atomic32_t *var) { - return ethr_native_atomic_add_return(var, -1); + return ethr_native_atomic32_add_return(var, -1); } -static ETHR_INLINE long -ethr_native_atomic_and_retold(ethr_native_atomic_t *var, long mask) +static ETHR_INLINE ethr_sint32_t +ethr_native_atomic32_and_retold(ethr_native_atomic32_t *var, ethr_sint32_t mask) { return atomic_and_val(&var->counter, mask); } -static ETHR_INLINE long -ethr_native_atomic_or_retold(ethr_native_atomic_t *var, long mask) +static ETHR_INLINE ethr_sint32_t +ethr_native_atomic32_or_retold(ethr_native_atomic32_t *var, ethr_sint32_t mask) { return atomic_or_val(&var->counter, mask); } -static ETHR_INLINE long -ethr_native_atomic_xchg(ethr_native_atomic_t *var, long val) +static ETHR_INLINE ethr_sint32_t +ethr_native_atomic32_xchg(ethr_native_atomic32_t *var, ethr_sint32_t val) { return atomic_exchange_acq(&var->counter, val); } -static ETHR_INLINE long -ethr_native_atomic_cmpxchg(ethr_native_atomic_t *var, long new, long expected) +static ETHR_INLINE ethr_sint32_t +ethr_native_atomic32_cmpxchg(ethr_native_atomic32_t *var, + ethr_sint32_t new, + ethr_sint32_t expected) { return atomic_compare_and_exchange_val_acq(&var->counter, new, expected); } @@ -118,54 +128,58 @@ ethr_native_atomic_cmpxchg(ethr_native_atomic_t *var, long new, long expected) * Atomic ops with at least specified barriers. */ -static ETHR_INLINE long -ethr_native_atomic_read_acqb(ethr_native_atomic_t *var) +static ETHR_INLINE ethr_sint32_t +ethr_native_atomic32_read_acqb(ethr_native_atomic32_t *var) { - long res = ethr_native_atomic_read(var); + ethr_sint32_t res = ethr_native_atomic32_read(var); ETHR_MEMORY_BARRIER; return res; } -static ETHR_INLINE long -ethr_native_atomic_inc_return_acqb(ethr_native_atomic_t *var) +static ETHR_INLINE ethr_sint32_t +ethr_native_atomic32_inc_return_acqb(ethr_native_atomic32_t *var) { - long res = ethr_native_atomic_inc_return(var); + ethr_sint32_t res = ethr_native_atomic32_inc_return(var); ETHR_MEMORY_BARRIER; return res; } static ETHR_INLINE void -ethr_native_atomic_set_relb(ethr_native_atomic_t *var, long val) +ethr_native_atomic32_set_relb(ethr_native_atomic32_t *var, ethr_sint32_t val) { ETHR_MEMORY_BARRIER; - ethr_native_atomic_set(var, val); + ethr_native_atomic32_set(var, val); } static ETHR_INLINE void -ethr_native_atomic_dec_relb(ethr_native_atomic_t *var) +ethr_native_atomic32_dec_relb(ethr_native_atomic32_t *var) { ETHR_MEMORY_BARRIER; - ethr_native_atomic_dec(var); + ethr_native_atomic32_dec(var); } -static ETHR_INLINE long -ethr_native_atomic_dec_return_relb(ethr_native_atomic_t *var) +static ETHR_INLINE ethr_sint32_t +ethr_native_atomic32_dec_return_relb(ethr_native_atomic32_t *var) { ETHR_MEMORY_BARRIER; - return ethr_native_atomic_dec_return(var); + return ethr_native_atomic32_dec_return(var); } -static ETHR_INLINE long -ethr_native_atomic_cmpxchg_acqb(ethr_native_atomic_t *var, long new, long exp) +static ETHR_INLINE ethr_sint32_t +ethr_native_atomic32_cmpxchg_acqb(ethr_native_atomic32_t *var, + ethr_sint32_t new, + ethr_sint32_t exp) { - return ethr_native_atomic_cmpxchg(var, new, exp); + return ethr_native_atomic32_cmpxchg(var, new, exp); } -static ETHR_INLINE long -ethr_native_atomic_cmpxchg_relb(ethr_native_atomic_t *var, long new, long exp) +static ETHR_INLINE ethr_sint32_t +ethr_native_atomic32_cmpxchg_relb(ethr_native_atomic32_t *var, + ethr_sint32_t new, + ethr_sint32_t exp) { ETHR_MEMORY_BARRIER; - return ethr_native_atomic_cmpxchg(var, new, exp); + return ethr_native_atomic32_cmpxchg(var, new, exp); } #endif /* ETHR_TRY_INLINE_FUNCS */ diff --git a/erts/include/internal/win/ethr_atomic.h b/erts/include/internal/win/ethr_atomic.h index 500459dd6c..60def01a7e 100644 --- a/erts/include/internal/win/ethr_atomic.h +++ b/erts/include/internal/win/ethr_atomic.h @@ -22,223 +22,394 @@ * Author: Rickard Green */ -#ifndef ETHR_WIN_ATOMIC_H__ -#define ETHR_WIN_ATOMIC_H__ - -#ifdef _MSC_VER -# if _MSC_VER < 1300 -# define ETHR_IMMED_ATOMIC_SET_GET_SAFE__ 0 /* Dont trust really old compilers */ -# else -# if defined(_M_IX86) -# define ETHR_IMMED_ATOMIC_SET_GET_SAFE__ 1 -# else /* I.e. IA64 */ -# if _MSC_VER >= 1400 -# define ETHR_IMMED_ATOMIC_SET_GET_SAFE__ 1 -# else -# define ETHR_IMMED_ATOMIC_SET_GET_SAFE__ 0 -# endif -# endif -# endif -# if _MSC_VER >= 1400 -# include <intrin.h> -# undef ETHR_COMPILER_BARRIER -# define ETHR_COMPILER_BARRIER _ReadWriteBarrier() -# endif -#pragma intrinsic(_ReadWriteBarrier) -#pragma intrinsic(_InterlockedAnd) -#pragma intrinsic(_InterlockedOr) +#undef ETHR_INCLUDE_ATOMIC_IMPL__ +#if !defined(ETHR_WIN_ATOMIC32_H__) && defined(ETHR_ATOMIC_WANT_32BIT_IMPL__) +#define ETHR_WIN_ATOMIC32_H__ +#define ETHR_INCLUDE_ATOMIC_IMPL__ 4 +#undef ETHR_ATOMIC_WANT_32BIT_IMPL__ +#elif !defined(ETHR_WIN_ATOMIC64_H__) && defined(ETHR_ATOMIC_WANT_64BIT_IMPL__) +#define ETHR_WIN_ATOMIC64_H__ +#ifdef ETHR_HAVE__INTERLOCKEDCOMPAREEXCHANGE64 +/* _InterlockedCompareExchange64() required... */ +#define ETHR_INCLUDE_ATOMIC_IMPL__ 8 +#endif +#undef ETHR_ATOMIC_WANT_64BIT_IMPL__ +#endif + +#ifdef ETHR_INCLUDE_ATOMIC_IMPL__ + +#if defined(_MSC_VER) && _MSC_VER >= 1400 + +#ifndef ETHR_WIN_ATOMIC_COMMON__ +#define ETHR_WIN_ATOMIC_COMMON__ + +#define ETHR_HAVE_NATIVE_ATOMICS 1 + +#if defined(_M_IX86) || defined(_M_AMD64) || defined(_M_IA64) +# define ETHR_READ_AND_SET_WITHOUT_INTERLOCKED_OP__ 1 #else -# define ETHR_IMMED_ATOMIC_SET_GET_SAFE__ 0 +# define ETHR_READ_AND_SET_WITHOUT_INTERLOCKED_OP__ 0 #endif +#if defined(_M_AMD64) || (defined(_M_IX86) \ + && !defined(ETHR_PRE_PENTIUM4_COMPAT)) +# define ETHR_READ_ACQB_AND_SET_RELB_COMPILER_BARRIER_ONLY__ 1 +#else +# define ETHR_READ_ACQB_AND_SET_RELB_COMPILER_BARRIER_ONLY__ 0 +#endif /* - * No configure test checking for _Interlocked*_{acq,rel} and - * Interlocked*{Acquire,Release} have been written yet... + * No configure test checking for interlocked acquire/release + * versions have been written, yet. It should define + * ETHR_HAVE_INTERLOCKED_ACQUIRE_RELEASE_BARRIERS if, and + * only if, all used interlocked operations with barriers + * exists. * * Note, that these are pure optimizations for the itanium * processor. */ -#ifdef ETHR_HAVE_INTERLOCKEDCOMPAREEXCHANGE_ACQ -#pragma intrinsic(_InterlockedCompareExchange_acq) +#include <intrin.h> +#undef ETHR_COMPILER_BARRIER +#define ETHR_COMPILER_BARRIER _ReadWriteBarrier() +#pragma intrinsic(_ReadWriteBarrier) +#pragma intrinsic(_InterlockedCompareExchange) + +#if defined(_M_AMD64) || (defined(_M_IX86) \ + && !defined(ETHR_PRE_PENTIUM4_COMPAT)) +#include <emmintrin.h> +#include <mmintrin.h> +#pragma intrinsic(_mm_mfence) +#define ETHR_MEMORY_BARRIER _mm_mfence() +#pragma intrinsic(_mm_sfence) +#define ETHR_WRITE_MEMORY_BARRIER _mm_sfence() +#pragma intrinsic(_mm_lfence) +#define ETHR_READ_MEMORY_BARRIER _mm_lfence() +#define ETHR_READ_DEPEND_MEMORY_BARRIER ETHR_COMPILER_BARRIER + +#else + +#define ETHR_MEMORY_BARRIER \ +do { \ + volatile long x___ = 0; \ + _InterlockedCompareExchange(&x___, (long) 1, (long) 0); \ +} while (0) + #endif -#ifdef ETHR_HAVE_INTERLOCKEDCOMPAREEXCHANGE_REL + +#endif /* ETHR_WIN_ATOMIC_COMMON__ */ + +#if ETHR_INCLUDE_ATOMIC_IMPL__ == 4 + +#define ETHR_HAVE_NATIVE_ATOMIC32 1 + +/* + * All used operations available as 32-bit intrinsics + */ + +#pragma intrinsic(_InterlockedDecrement) +#pragma intrinsic(_InterlockedIncrement) +#pragma intrinsic(_InterlockedExchangeAdd) +#pragma intrinsic(_InterlockedExchange) +#pragma intrinsic(_InterlockedAnd) +#pragma intrinsic(_InterlockedOr) +#ifdef ETHR_HAVE_INTERLOCKED_ACQUIRE_RELEASE_BARRIERS +#pragma intrinsic(_InterlockedExchangeAdd_acq) +#pragma intrinsic(_InterlockedIncrement_acq) +#pragma intrinsic(_InterlockedDecrement_rel) +#pragma intrinsic(_InterlockedCompareExchange_acq) #pragma intrinsic(_InterlockedCompareExchange_rel) #endif +#define ETHR_ILCKD__(X) _Interlocked ## X +#ifdef ETHR_HAVE_INTERLOCKED_ACQUIRE_RELEASE_BARRIERS +#define ETHR_ILCKD_ACQ__(X) _Interlocked ## X ## _acq +#define ETHR_ILCKD_REL__(X) _Interlocked ## X ## _rel +#else +#define ETHR_ILCKD_ACQ__(X) _Interlocked ## X +#define ETHR_ILCKD_REL__(X) _Interlocked ## X +#endif + +#define ETHR_NATMC_FUNC__(X) ethr_native_atomic32_ ## X +#define ETHR_ATMC_T__ ethr_native_atomic32_t +#define ETHR_AINT_T__ ethr_sint32_t + +#elif ETHR_INCLUDE_ATOMIC_IMPL__ == 8 + +#define ETHR_HAVE_NATIVE_ATOMIC64 1 + +/* + * _InterlockedCompareExchange64() is required. The other may not + * be available, but if so, we can generate them. + */ +#pragma intrinsic(_InterlockedCompareExchange64) + +#if ETHR_READ_AND_SET_WITHOUT_INTERLOCKED_OP__ +#define ETHR_OWN_ILCKD_INIT_VAL__(PTR) *(PTR) +#else +#define ETHR_OWN_ILCKD_INIT_VAL__(PTR) (__int64) 0 +#endif + +#define ETHR_OWN_ILCKD_BODY_IMPL__(FUNC, PTR, NEW, ACT, EXP, OPS, RET) \ +{ \ + __int64 NEW, ACT, EXP; \ + ACT = ETHR_OWN_ILCKD_INIT_VAL__(PTR); \ + do { \ + EXP = ACT; \ + { OPS; } \ + ACT = _InterlockedCompareExchange64(PTR, NEW, EXP); \ + } while (ACT != EXP); \ + return RET; \ +} + +#define ETHR_OWN_ILCKD_1_IMPL__(FUNC, NEW, ACT, EXP, OPS, RET) \ +static __forceinline __int64 \ +FUNC(__int64 volatile *ptr) \ +ETHR_OWN_ILCKD_BODY_IMPL__(FUNC, ptr, NEW, ACT, EXP, OPS, RET) + +#define ETHR_OWN_ILCKD_2_IMPL__(FUNC, NEW, ACT, EXP, OPS, ARG, RET) \ +static __forceinline __int64 \ +FUNC(__int64 volatile *ptr, __int64 ARG) \ +ETHR_OWN_ILCKD_BODY_IMPL__(FUNC, ptr, NEW, ACT, EXP, OPS, RET) + + +#ifdef ETHR_HAVE__INTERLOCKEDDECREMENT64 +#pragma intrinsic(_InterlockedDecrement64) +#else +ETHR_OWN_ILCKD_1_IMPL__(_InterlockedDecrement64, new, act, exp, + new = act - 1, new) +#endif +#ifdef ETHR_HAVE__INTERLOCKEDINCREMENT64 +#pragma intrinsic(_InterlockedIncrement64) +#else +ETHR_OWN_ILCKD_1_IMPL__(_InterlockedIncrement64, new, act, exp, + new = act + 1, new) +#endif +#ifdef ETHR_HAVE__INTERLOCKEDEXCHANGEADD64 +#pragma intrinsic(_InterlockedExchangeAdd64) +#else +ETHR_OWN_ILCKD_2_IMPL__(_InterlockedExchangeAdd64, new, act, exp, + new = act + arg, arg, act) +#endif +#ifdef ETHR_HAVE__INTERLOCKEDEXCHANGE64 +#pragma intrinsic(_InterlockedExchange64) +#else +ETHR_OWN_ILCKD_2_IMPL__(_InterlockedExchange64, new, act, exp, + new = arg, arg, act) +#endif +#ifdef ETHR_HAVE__INTERLOCKEDAND64 +#pragma intrinsic(_InterlockedAnd64) +#else +ETHR_OWN_ILCKD_2_IMPL__(_InterlockedAnd64, new, act, exp, + new = act & arg, arg, act) +#endif +#ifdef ETHR_HAVE__INTERLOCKEDOR64 +#pragma intrinsic(_InterlockedOr64) +#else +ETHR_OWN_ILCKD_2_IMPL__(_InterlockedOr64, new, act, exp, + new = act | arg, arg, act) +#endif +#ifdef ETHR_HAVE_INTERLOCKED_ACQUIRE_RELEASE_BARRIERS +#pragma intrinsic(_InterlockedExchangeAdd64_acq) +#pragma intrinsic(_InterlockedIncrement64_acq) +#pragma intrinsic(_InterlockedDecrement64_rel) +#pragma intrinsic(_InterlockedCompareExchange64_acq) +#pragma intrinsic(_InterlockedCompareExchange64_rel) +#endif + +#define ETHR_ILCKD__(X) _Interlocked ## X ## 64 +#ifdef ETHR_HAVE_INTERLOCKED_ACQUIRE_RELEASE_BARRIERS +#define ETHR_ILCKD_ACQ__(X) _Interlocked ## X ## 64_acq +#define ETHR_ILCKD_REL__(X) _Interlocked ## X ## 64_rel +#else +#define ETHR_ILCKD_ACQ__(X) _Interlocked ## X ## 64 +#define ETHR_ILCKD_REL__(X) _Interlocked ## X ## 64 +#endif + +#define ETHR_NATMC_FUNC__(X) ethr_native_atomic64_ ## X +#define ETHR_ATMC_T__ ethr_native_atomic64_t +#define ETHR_AINT_T__ ethr_sint64_t + +#else +#error "Unsupported integer size" +#endif typedef struct { - volatile LONG value; -} ethr_native_atomic_t; + volatile ETHR_AINT_T__ value; +} ETHR_ATMC_T__; -#define ETHR_MEMORY_BARRIER \ -do { \ - volatile LONG x___ = 0; \ - _InterlockedCompareExchange(&x___, (LONG) 1, (LONG) 0); \ -} while (0) +#if defined(ETHR_TRY_INLINE_FUNCS) || defined(ETHR_ATOMIC_IMPL__) -#if defined(ETHR_TRY_INLINE_FUNCS) || defined(ETHR_AUX_IMPL__) +static ETHR_INLINE ETHR_AINT_T__ * +ETHR_NATMC_FUNC__(addr)(ETHR_ATMC_T__ *var) +{ + return (ETHR_AINT_T__ *) &var->value; +} static ETHR_INLINE void -ethr_native_atomic_init(ethr_native_atomic_t *var, long i) +ETHR_NATMC_FUNC__(init)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ i) { - var->value = (LONG) i; +#if ETHR_READ_AND_SET_WITHOUT_INTERLOCKED_OP__ + var->value = i; +#else + (void) ETHR_ILCKD__(Exchange)(&var->value, i); +#endif } static ETHR_INLINE void -ethr_native_atomic_set(ethr_native_atomic_t *var, long i) +ETHR_NATMC_FUNC__(set)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ i) { -#if ETHR_IMMED_ATOMIC_SET_GET_SAFE__ - var->value = (LONG) i; +#if ETHR_READ_AND_SET_WITHOUT_INTERLOCKED_OP__ + var->value = i; #else - (void) InterlockedExchange(&var->value, (LONG) i); + (void) ETHR_ILCKD__(Exchange)(&var->value, i); #endif } -static ETHR_INLINE long -ethr_native_atomic_read(ethr_native_atomic_t *var) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(read)(ETHR_ATMC_T__ *var) { -#if ETHR_IMMED_ATOMIC_SET_GET_SAFE__ +#if ETHR_READ_AND_SET_WITHOUT_INTERLOCKED_OP__ return var->value; #else - return InterlockedExchangeAdd(&var->value, (LONG) 0); + return ETHR_ILCKD__(ExchangeAdd)(&var->value, (ETHR_AINT_T__) 0); #endif } static ETHR_INLINE void -ethr_native_atomic_add(ethr_native_atomic_t *var, long incr) +ETHR_NATMC_FUNC__(add)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ incr) { - (void) InterlockedExchangeAdd(&var->value, (LONG) incr); + (void) ETHR_ILCKD__(ExchangeAdd)(&var->value, incr); } -static ETHR_INLINE long -ethr_native_atomic_add_return(ethr_native_atomic_t *var, long i) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(add_return)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ i) { - LONG tmp = InterlockedExchangeAdd(&var->value, (LONG) i); - return tmp + i; + return ETHR_ILCKD__(ExchangeAdd)(&var->value, i) + i; } static ETHR_INLINE void -ethr_native_atomic_inc(ethr_native_atomic_t *var) +ETHR_NATMC_FUNC__(inc)(ETHR_ATMC_T__ *var) { - (void) InterlockedIncrement(&var->value); + (void) ETHR_ILCKD__(Increment)(&var->value); } static ETHR_INLINE void -ethr_native_atomic_dec(ethr_native_atomic_t *var) +ETHR_NATMC_FUNC__(dec)(ETHR_ATMC_T__ *var) { - (void) InterlockedDecrement(&var->value); + (void) ETHR_ILCKD__(Decrement)(&var->value); } -static ETHR_INLINE long -ethr_native_atomic_inc_return(ethr_native_atomic_t *var) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(inc_return)(ETHR_ATMC_T__ *var) { - return (long) InterlockedIncrement(&var->value); + return ETHR_ILCKD__(Increment)(&var->value); } -static ETHR_INLINE long -ethr_native_atomic_dec_return(ethr_native_atomic_t *var) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(dec_return)(ETHR_ATMC_T__ *var) { - return (long) InterlockedDecrement(&var->value); + return ETHR_ILCKD__(Decrement)(&var->value); } -static ETHR_INLINE long -ethr_native_atomic_and_retold(ethr_native_atomic_t *var, long mask) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(and_retold)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ mask) { - return (long) _InterlockedAnd(&var->value, mask); + return ETHR_ILCKD__(And)(&var->value, mask); } -static ETHR_INLINE long -ethr_native_atomic_or_retold(ethr_native_atomic_t *var, long mask) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(or_retold)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ mask) { - return (long) _InterlockedOr(&var->value, mask); + return ETHR_ILCKD__(Or)(&var->value, mask); } - -static ETHR_INLINE long -ethr_native_atomic_cmpxchg(ethr_native_atomic_t *var, long new, long old) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(cmpxchg)(ETHR_ATMC_T__ *var, + ETHR_AINT_T__ new, + ETHR_AINT_T__ old) { - return (long) _InterlockedCompareExchange(&var->value, (LONG) new, (LONG) old); + return ETHR_ILCKD__(CompareExchange)(&var->value, new, old); } -static ETHR_INLINE long -ethr_native_atomic_xchg(ethr_native_atomic_t *var, long new) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(xchg)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ new) { - return (long) InterlockedExchange(&var->value, (LONG) new); + return ETHR_ILCKD__(Exchange)(&var->value, new); } /* * Atomic ops with at least specified barriers. */ -static ETHR_INLINE long -ethr_native_atomic_read_acqb(ethr_native_atomic_t *var) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(read_acqb)(ETHR_ATMC_T__ *var) { -#ifdef ETHR_HAVE_INTERLOCKEDEXCHANGEADDACQUIRE - return (long) InterlockedExchangeAddAcquire(&var->value, (LONG) 0); +#if ETHR_READ_ACQB_AND_SET_RELB_COMPILER_BARRIER_ONLY__ + ETHR_AINT_T__ val = var->value; + ETHR_COMPILER_BARRIER; + return val; #else - return (long) InterlockedExchangeAdd(&var->value, (LONG) 0); + return ETHR_ILCKD_ACQ__(ExchangeAdd)(&var->value, (ETHR_AINT_T__) 0); #endif } -static ETHR_INLINE long -ethr_native_atomic_inc_return_acqb(ethr_native_atomic_t *var) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(inc_return_acqb)(ETHR_ATMC_T__ *var) { -#ifdef ETHR_HAVE_INTERLOCKEDINCREMENTACQUIRE - return (long) InterlockedIncrementAcquire(&var->value); -#else - return (long) InterlockedIncrement(&var->value); -#endif + return ETHR_ILCKD_ACQ__(Increment)(&var->value); } static ETHR_INLINE void -ethr_native_atomic_set_relb(ethr_native_atomic_t *var, long i) +ETHR_NATMC_FUNC__(set_relb)(ETHR_ATMC_T__ *var, ETHR_AINT_T__ i) { - (void) InterlockedExchange(&var->value, (LONG) i); +#if ETHR_READ_ACQB_AND_SET_RELB_COMPILER_BARRIER_ONLY__ + ETHR_COMPILER_BARRIER; + var->value = i; +#else + (void) ETHR_ILCKD_REL__(Exchange)(&var->value, i); +#endif } static ETHR_INLINE void -ethr_native_atomic_dec_relb(ethr_native_atomic_t *var) +ETHR_NATMC_FUNC__(dec_relb)(ETHR_ATMC_T__ *var) { -#ifdef ETHR_HAVE_INTERLOCKEDDECREMENTRELEASE - (void) InterlockedDecrementRelease(&var->value); -#else - (void) InterlockedDecrement(&var->value); -#endif + (void) ETHR_ILCKD_REL__(Decrement)(&var->value); } -static ETHR_INLINE long -ethr_native_atomic_dec_return_relb(ethr_native_atomic_t *var) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(dec_return_relb)(ETHR_ATMC_T__ *var) { -#ifdef ETHR_HAVE_INTERLOCKEDDECREMENTRELEASE - return (long) InterlockedDecrementRelease(&var->value); -#else - return (long) InterlockedDecrement(&var->value); -#endif + return ETHR_ILCKD_REL__(Decrement)(&var->value); } -static ETHR_INLINE long -ethr_native_atomic_cmpxchg_acqb(ethr_native_atomic_t *var, long new, long old) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(cmpxchg_acqb)(ETHR_ATMC_T__ *var, + ETHR_AINT_T__ new, + ETHR_AINT_T__ old) { -#ifdef ETHR_HAVE_INTERLOCKEDCOMPAREEXCHANGE_ACQ - return (long) _InterlockedCompareExchange_acq(&var->value, (LONG) new, (LONG) old); -#else - return (long) _InterlockedCompareExchange(&var->value, (LONG) new, (LONG) old); -#endif + return ETHR_ILCKD_ACQ__(CompareExchange)(&var->value, new, old); } -static ETHR_INLINE long -ethr_native_atomic_cmpxchg_relb(ethr_native_atomic_t *var, long new, long old) +static ETHR_INLINE ETHR_AINT_T__ +ETHR_NATMC_FUNC__(cmpxchg_relb)(ETHR_ATMC_T__ *var, + ETHR_AINT_T__ new, + ETHR_AINT_T__ old) { - -#ifdef ETHR_HAVE_INTERLOCKEDCOMPAREEXCHANGE_REL - return (long) _InterlockedCompareExchange_rel(&var->value, (LONG) new, (LONG) old); -#else - return (long) _InterlockedCompareExchange(&var->value, (LONG) new, (LONG) old); -#endif + return ETHR_ILCKD_REL__(CompareExchange)(&var->value, new, old); } -#endif +#endif /* ETHR_TRY_INLINE_FUNCS */ -#endif +#undef ETHR_ILCKD__ +#undef ETHR_ILCKD_ACQ__ +#undef ETHR_ILCKD_REL__ +#undef ETHR_NATMC_FUNC__ +#undef ETHR_ATMC_T__ +#undef ETHR_AINT_T__ +#undef ETHR_READ_AND_SET_WITHOUT_INTERLOCKED_OP__ +#undef ETHR_READ_ACQB_AND_SET_RELB_COMPILER_BARRIER_ONLY__ + +#endif /* _MSC_VER */ + +#endif /* ETHR_INCLUDE_ATOMIC_IMPL__ */ diff --git a/erts/include/internal/win/ethr_event.h b/erts/include/internal/win/ethr_event.h index af57c20f91..598816b2c6 100644 --- a/erts/include/internal/win/ethr_event.h +++ b/erts/include/internal/win/ethr_event.h @@ -21,22 +21,24 @@ * Author: Rickard Green */ -#define ETHR_EVENT_OFF_WAITER__ ((LONG) -1) -#define ETHR_EVENT_OFF__ ((LONG) 1) -#define ETHR_EVENT_ON__ ((LONG) 0) +#define ETHR_EVENT_OFF_WAITER__ ((long) -1) +#define ETHR_EVENT_OFF__ ((long) 1) +#define ETHR_EVENT_ON__ ((long) 0) typedef struct { - volatile LONG state; + volatile long state; HANDLE handle; } ethr_event; #if defined(ETHR_TRY_INLINE_FUNCS) || defined(ETHR_EVENT_IMPL__) +#pragma intrinsic(_InterlockedExchange) + static ETHR_INLINE void ETHR_INLINE_FUNC_NAME_(ethr_event_set)(ethr_event *e) { - /* InterlockedExchange() imply a full memory barrier which is important */ - LONG state = InterlockedExchange(&e->state, ETHR_EVENT_ON__); + /* _InterlockedExchange() imply a full memory barrier which is important */ + long state = _InterlockedExchange(&e->state, ETHR_EVENT_ON__); if (state == ETHR_EVENT_OFF_WAITER__) { if (!SetEvent(e->handle)) ETHR_FATAL_ERROR__(ethr_win_get_errno__()); @@ -46,7 +48,7 @@ ETHR_INLINE_FUNC_NAME_(ethr_event_set)(ethr_event *e) static ETHR_INLINE void ETHR_INLINE_FUNC_NAME_(ethr_event_reset)(ethr_event *e) { - /* InterlockedExchange() imply a full memory barrier which is important */ + /* _InterlockedExchange() imply a full memory barrier which is important */ InterlockedExchange(&e->state, ETHR_EVENT_OFF__); } diff --git a/erts/include/internal/win/ethread.h b/erts/include/internal/win/ethread.h index b52710f6a3..c01b17cf14 100644 --- a/erts/include/internal/win/ethread.h +++ b/erts/include/internal/win/ethread.h @@ -25,7 +25,11 @@ #ifndef ETHREAD_WIN_H__ #define ETHREAD_WIN_H__ +#define ETHR_ATOMIC_WANT_32BIT_IMPL__ #include "ethr_atomic.h" -#define ETHR_HAVE_NATIVE_ATOMICS 1 +#if ETHR_SIZEOF_PTR == 8 +# define ETHR_ATOMIC_WANT_64BIT_IMPL__ +# include "ethr_atomic.h" +#endif #endif diff --git a/erts/lib_src/Makefile.in b/erts/lib_src/Makefile.in index 0d3181cace..757b3b24e2 100644 --- a/erts/lib_src/Makefile.in +++ b/erts/lib_src/Makefile.in @@ -283,6 +283,7 @@ endif ETHR_THR_LIB_BASE_DIR=@ETHR_THR_LIB_BASE_DIR@ ifneq ($(strip $(ETHR_LIB_NAME)),) ETHREAD_LIB_SRC=common/ethr_aux.c \ + common/ethr_atomics.c \ common/ethr_mutex.c \ common/ethr_cbf.c \ $(ETHR_THR_LIB_BASE_DIR)/ethread.c \ @@ -381,6 +382,11 @@ $(ERTS_LIB): $(ERTS_LIB_OBJS) # Object files # +ifeq ($(TYPE)-@GCC@,debug-yes) +$(r_OBJ_DIR)/ethr_aux.o: common/ethr_aux.c + $(CC) $(THR_DEFS) $(CFLAGS) -Wno-unused-function $(INCLUDES) -c $< -o $@ +endif + $(r_OBJ_DIR)/%.o: common/%.c $(CC) $(THR_DEFS) $(CFLAGS) $(INCLUDES) -c $< -o $@ @@ -445,6 +451,7 @@ INTERNAL_RELEASE_INCLUDES= \ $(ERTS_INCL_INT)/ethread.h \ $(ERTS_INCL_INT)/ethr_mutex.h \ $(ERTS_INCL_INT)/ethr_optimized_fallbacks.h \ + $(ERTS_INCL_INT)/ethr_atomics.h \ $(ERTS_INCL_INT)/$(TARGET)/ethread.mk \ $(ERTS_INCL_INT)/$(TARGET)/erts_internal.mk \ $(ERTS_INCL_INT)/$(TARGET)/ethread_header_config.h \ diff --git a/erts/lib_src/common/ethr_atomics.c b/erts/lib_src/common/ethr_atomics.c new file mode 100644 index 0000000000..94557d904a --- /dev/null +++ b/erts/lib_src/common/ethr_atomics.c @@ -0,0 +1,402 @@ +/* + * %CopyrightBegin% + * + * Copyright Ericsson AB 2010. 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: The ethread atomic API + * Author: Rickard Green + */ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#define ETHR_INLINE_FUNC_NAME_(X) X ## __ +#define ETHR_ATOMIC_IMPL__ + +#include "ethread.h" +#include "ethr_internal.h" + +#ifndef ETHR_HAVE_NATIVE_ATOMICS +ethr_atomic_protection_t ethr_atomic_protection__[1 << ETHR_ATOMIC_ADDR_BITS]; +#endif + +int +ethr_init_atomics(void) +{ +#ifndef ETHR_HAVE_NATIVE_ATOMICS + { + int i; + for (i = 0; i < (1 << ETHR_ATOMIC_ADDR_BITS); i++) { + int res = ethr_spinlock_init(ðr_atomic_protection__[i].u.lck); + if (res != 0) + return res; + } + } +#endif + return 0; +} + +/* + * --- Pointer size atomics --------------------------------------------------- + */ + +ethr_sint_t * +ethr_atomic_addr(ethr_atomic_t *var) +{ + ETHR_ASSERT(var); + return ethr_atomic_addr__(var); +} + +void +ethr_atomic_init(ethr_atomic_t *var, ethr_sint_t i) +{ + ETHR_ASSERT(var); + ethr_atomic_init__(var, i); +} + +void +ethr_atomic_set(ethr_atomic_t *var, ethr_sint_t i) +{ + ETHR_ASSERT(!ethr_not_inited__); + ETHR_ASSERT(var); + ethr_atomic_set__(var, i); +} + +ethr_sint_t +ethr_atomic_read(ethr_atomic_t *var) +{ + ETHR_ASSERT(!ethr_not_inited__); + ETHR_ASSERT(var); + return ethr_atomic_read__(var); +} + +ethr_sint_t +ethr_atomic_add_read(ethr_atomic_t *var, ethr_sint_t incr) +{ + ETHR_ASSERT(!ethr_not_inited__); + ETHR_ASSERT(var); + return ethr_atomic_add_read__(var, incr); +} + +ethr_sint_t +ethr_atomic_inc_read(ethr_atomic_t *var) +{ + ETHR_ASSERT(!ethr_not_inited__); + ETHR_ASSERT(var); + return ethr_atomic_inc_read__(var); +} + +ethr_sint_t +ethr_atomic_dec_read(ethr_atomic_t *var) +{ + ETHR_ASSERT(!ethr_not_inited__); + ETHR_ASSERT(var); + return ethr_atomic_dec_read__(var); +} + +void +ethr_atomic_add(ethr_atomic_t *var, ethr_sint_t incr) +{ + ETHR_ASSERT(!ethr_not_inited__); + ETHR_ASSERT(var); + ethr_atomic_add__(var, incr); +} + +void +ethr_atomic_inc(ethr_atomic_t *var) +{ + ETHR_ASSERT(!ethr_not_inited__); + ETHR_ASSERT(var); + ethr_atomic_inc__(var); +} + +void +ethr_atomic_dec(ethr_atomic_t *var) +{ + ETHR_ASSERT(!ethr_not_inited__); + ETHR_ASSERT(var); + ethr_atomic_dec__(var); +} + +ethr_sint_t +ethr_atomic_read_band(ethr_atomic_t *var, ethr_sint_t mask) +{ + ETHR_ASSERT(!ethr_not_inited__); + ETHR_ASSERT(var); + return ethr_atomic_read_band__(var, mask); +} + +ethr_sint_t +ethr_atomic_read_bor(ethr_atomic_t *var, ethr_sint_t mask) +{ + ETHR_ASSERT(!ethr_not_inited__); + ETHR_ASSERT(var); + return ethr_atomic_read_bor__(var, mask); +} + +ethr_sint_t +ethr_atomic_xchg(ethr_atomic_t *var, ethr_sint_t new) +{ + ETHR_ASSERT(!ethr_not_inited__); + ETHR_ASSERT(var); + return ethr_atomic_xchg__(var, new); +} + +ethr_sint_t +ethr_atomic_cmpxchg(ethr_atomic_t *var, ethr_sint_t new, ethr_sint_t expected) +{ + ETHR_ASSERT(!ethr_not_inited__); + ETHR_ASSERT(var); + return ethr_atomic_cmpxchg__(var, new, expected); +} + +ethr_sint_t +ethr_atomic_read_acqb(ethr_atomic_t *var) +{ + ETHR_ASSERT(!ethr_not_inited__); + ETHR_ASSERT(var); + return ethr_atomic_read_acqb__(var); +} + +ethr_sint_t +ethr_atomic_inc_read_acqb(ethr_atomic_t *var) +{ + ETHR_ASSERT(!ethr_not_inited__); + ETHR_ASSERT(var); + return ethr_atomic_inc_read_acqb__(var); +} + +void +ethr_atomic_set_relb(ethr_atomic_t *var, ethr_sint_t i) +{ + ETHR_ASSERT(!ethr_not_inited__); + ETHR_ASSERT(var); + ethr_atomic_set_relb__(var, i); +} + +void +ethr_atomic_dec_relb(ethr_atomic_t *var) +{ + ETHR_ASSERT(!ethr_not_inited__); + ETHR_ASSERT(var); + ethr_atomic_dec_relb__(var); +} + +ethr_sint_t +ethr_atomic_dec_read_relb(ethr_atomic_t *var) +{ + ETHR_ASSERT(!ethr_not_inited__); + ETHR_ASSERT(var); + return ethr_atomic_dec_read_relb__(var); +} + +ethr_sint_t +ethr_atomic_cmpxchg_acqb(ethr_atomic_t *var, ethr_sint_t new, ethr_sint_t exp) +{ + ETHR_ASSERT(!ethr_not_inited__); + ETHR_ASSERT(var); + return ethr_atomic_cmpxchg_acqb__(var, new, exp); +} + +ethr_sint_t +ethr_atomic_cmpxchg_relb(ethr_atomic_t *var, ethr_sint_t new, ethr_sint_t exp) +{ + ETHR_ASSERT(!ethr_not_inited__); + ETHR_ASSERT(var); + return ethr_atomic_cmpxchg_relb__(var, new, exp); +} + + +/* + * --- 32-bit atomics --------------------------------------------------------- + */ + +ethr_sint32_t * +ethr_atomic32_addr(ethr_atomic32_t *var) +{ + ETHR_ASSERT(var); + return ethr_atomic32_addr__(var); +} + +void +ethr_atomic32_init(ethr_atomic32_t *var, ethr_sint32_t i) +{ + ETHR_ASSERT(var); + ethr_atomic32_init__(var, i); +} + +void +ethr_atomic32_set(ethr_atomic32_t *var, ethr_sint32_t i) +{ + ETHR_ASSERT(!ethr_not_inited__); + ETHR_ASSERT(var); + ethr_atomic32_set__(var, i); +} + +ethr_sint32_t +ethr_atomic32_read(ethr_atomic32_t *var) +{ + ETHR_ASSERT(!ethr_not_inited__); + ETHR_ASSERT(var); + return ethr_atomic32_read__(var); +} + + +ethr_sint32_t +ethr_atomic32_add_read(ethr_atomic32_t *var, ethr_sint32_t incr) +{ + ETHR_ASSERT(!ethr_not_inited__); + ETHR_ASSERT(var); + return ethr_atomic32_add_read__(var, incr); +} + +ethr_sint32_t +ethr_atomic32_inc_read(ethr_atomic32_t *var) +{ + ETHR_ASSERT(!ethr_not_inited__); + ETHR_ASSERT(var); + return ethr_atomic32_inc_read__(var); +} + +ethr_sint32_t +ethr_atomic32_dec_read(ethr_atomic32_t *var) +{ + ETHR_ASSERT(!ethr_not_inited__); + ETHR_ASSERT(var); + return ethr_atomic32_dec_read__(var); +} + +void +ethr_atomic32_add(ethr_atomic32_t *var, ethr_sint32_t incr) +{ + ETHR_ASSERT(!ethr_not_inited__); + ETHR_ASSERT(var); + ethr_atomic32_add__(var, incr); +} + +void +ethr_atomic32_inc(ethr_atomic32_t *var) +{ + ETHR_ASSERT(!ethr_not_inited__); + ETHR_ASSERT(var); + ethr_atomic32_inc__(var); +} + +void +ethr_atomic32_dec(ethr_atomic32_t *var) +{ + ETHR_ASSERT(!ethr_not_inited__); + ETHR_ASSERT(var); + ethr_atomic32_dec__(var); +} + +ethr_sint32_t +ethr_atomic32_read_band(ethr_atomic32_t *var, ethr_sint32_t mask) +{ + ETHR_ASSERT(!ethr_not_inited__); + ETHR_ASSERT(var); + return ethr_atomic32_read_band__(var, mask); +} + +ethr_sint32_t +ethr_atomic32_read_bor(ethr_atomic32_t *var, ethr_sint32_t mask) +{ + ETHR_ASSERT(!ethr_not_inited__); + ETHR_ASSERT(var); + return ethr_atomic32_read_bor__(var, mask); +} + +ethr_sint32_t +ethr_atomic32_xchg(ethr_atomic32_t *var, ethr_sint32_t new) +{ + ETHR_ASSERT(!ethr_not_inited__); + ETHR_ASSERT(var); + return ethr_atomic32_xchg__(var, new); +} + +ethr_sint32_t +ethr_atomic32_cmpxchg(ethr_atomic32_t *var, + ethr_sint32_t new, + ethr_sint32_t expected) +{ + ETHR_ASSERT(!ethr_not_inited__); + ETHR_ASSERT(var); + return ethr_atomic32_cmpxchg__(var, new, expected); +} + +ethr_sint32_t +ethr_atomic32_read_acqb(ethr_atomic32_t *var) +{ + ETHR_ASSERT(!ethr_not_inited__); + ETHR_ASSERT(var); + return ethr_atomic32_read_acqb__(var); +} + +ethr_sint32_t +ethr_atomic32_inc_read_acqb(ethr_atomic32_t *var) +{ + ETHR_ASSERT(!ethr_not_inited__); + ETHR_ASSERT(var); + return ethr_atomic32_inc_read_acqb__(var); +} + +void +ethr_atomic32_set_relb(ethr_atomic32_t *var, ethr_sint32_t i) +{ + ETHR_ASSERT(!ethr_not_inited__); + ETHR_ASSERT(var); + ethr_atomic32_set_relb__(var, i); +} + +void +ethr_atomic32_dec_relb(ethr_atomic32_t *var) +{ + ETHR_ASSERT(!ethr_not_inited__); + ETHR_ASSERT(var); + ethr_atomic32_dec_relb__(var); +} + +ethr_sint32_t +ethr_atomic32_dec_read_relb(ethr_atomic32_t *var) +{ + ETHR_ASSERT(!ethr_not_inited__); + ETHR_ASSERT(var); + return ethr_atomic32_dec_read_relb__(var); +} + +ethr_sint32_t +ethr_atomic32_cmpxchg_acqb(ethr_atomic32_t *var, + ethr_sint32_t new, + ethr_sint32_t exp) +{ + ETHR_ASSERT(!ethr_not_inited__); + ETHR_ASSERT(var); + return ethr_atomic32_cmpxchg_acqb__(var, new, exp); +} + +ethr_sint32_t +ethr_atomic32_cmpxchg_relb(ethr_atomic32_t *var, + ethr_sint32_t new, + ethr_sint32_t exp) +{ + ETHR_ASSERT(!ethr_not_inited__); + ETHR_ASSERT(var); + return ethr_atomic32_cmpxchg_relb__(var, new, exp); +} + diff --git a/erts/lib_src/common/ethr_aux.c b/erts/lib_src/common/ethr_aux.c index 4db4cffd3a..2c3e25a805 100644 --- a/erts/lib_src/common/ethr_aux.c +++ b/erts/lib_src/common/ethr_aux.c @@ -31,7 +31,10 @@ #define ETHR_INLINE_FUNC_NAME_(X) X ## __ #define ETHR_AUX_IMPL__ - +#define ETHR_ATOMIC_IMPL__ /* Needed in order to pull in + native atomic implementations + for optimized fallbacks of + spinlocks and rwspinlocks */ #include "ethread.h" #include "ethr_internal.h" #include <string.h> @@ -51,10 +54,6 @@ int ethr_not_inited__ = 1; ethr_memory_allocators ethr_mem__ = ETHR_MEM_ALLOCS_DEF_INITER__; -#ifndef ETHR_HAVE_OPTIMIZED_ATOMIC_OPS -ethr_atomic_protection_t ethr_atomic_protection__[1 << ETHR_ATOMIC_ADDR_BITS]; -#endif - void *(*ethr_thr_prepare_func__)(void) = NULL; void (*ethr_thr_parent_func__)(void *) = NULL; void (*ethr_thr_child_func__)(void *) = NULL; @@ -138,16 +137,9 @@ ethr_init_common__(ethr_init_data *id) #endif ethr_max_stack_size__ = ETHR_B2KW(ethr_max_stack_size__); -#ifndef ETHR_HAVE_OPTIMIZED_ATOMIC_OPS - { - int i; - for (i = 0; i < (1 << ETHR_ATOMIC_ADDR_BITS); i++) { - res = ethr_spinlock_init(ðr_atomic_protection__[i].u.lck); - if (res != 0) - return res; - } - } -#endif + res = ethr_init_atomics(); + if (res != 0) + return res; res = ethr_mutex_lib_init(erts_get_cpu_configured(ethr_cpu_info__)); if (res != 0) @@ -279,14 +271,6 @@ typedef union { static ethr_spinlock_t ts_ev_alloc_lock; static ethr_ts_event *free_ts_ev; -#if SIZEOF_VOID_P == SIZEOF_INT -typedef unsigned int EthrPtrSzUInt; -#elif SIZEOF_VOID_P == SIZEOF_LONG -typedef unsigned long EthrPtrSzUInt; -#else -#error No pointer sized integer type -#endif - static ethr_ts_event *ts_event_pool(int size, ethr_ts_event **endpp) { int i; @@ -295,16 +279,16 @@ static ethr_ts_event *ts_event_pool(int size, ethr_ts_event **endpp) + ETHR_CACHE_LINE_SIZE); if (!atsev) return NULL; - if ((((EthrPtrSzUInt) atsev) & ETHR_CACHE_LINE_MASK) == 0) + if ((((ethr_uint_t) atsev) & ETHR_CACHE_LINE_MASK) == 0) atsev = ((ethr_aligned_ts_event *) - ((((EthrPtrSzUInt) atsev) & ~ETHR_CACHE_LINE_MASK) + ((((ethr_uint_t) atsev) & ~ETHR_CACHE_LINE_MASK) + ETHR_CACHE_LINE_SIZE)); for (i = 1; i < size; i++) { atsev[i-1].ts_ev.next = &atsev[i].ts_ev; - ethr_atomic_init(&atsev[i-1].ts_ev.uaflgs, 0); + ethr_atomic32_init(&atsev[i-1].ts_ev.uaflgs, 0); atsev[i-1].ts_ev.iflgs = 0; } - ethr_atomic_init(&atsev[size-1].ts_ev.uaflgs, 0); + ethr_atomic32_init(&atsev[size-1].ts_ev.uaflgs, 0); atsev[size-1].ts_ev.iflgs = 0; atsev[size-1].ts_ev.next = NULL; if (endpp) @@ -466,170 +450,6 @@ int ethr_get_main_thr_status(int *on) return 0; } - -/* Atomics */ - -void -ethr_atomic_init(ethr_atomic_t *var, long i) -{ - ETHR_ASSERT(var); - ethr_atomic_init__(var, i); -} - -void -ethr_atomic_set(ethr_atomic_t *var, long i) -{ - ETHR_ASSERT(!ethr_not_inited__); - ETHR_ASSERT(var); - ethr_atomic_set__(var, i); -} - -long -ethr_atomic_read(ethr_atomic_t *var) -{ - ETHR_ASSERT(!ethr_not_inited__); - ETHR_ASSERT(var); - return ethr_atomic_read__(var); -} - - -long -ethr_atomic_add_read(ethr_atomic_t *var, long incr) -{ - ETHR_ASSERT(!ethr_not_inited__); - ETHR_ASSERT(var); - return ethr_atomic_add_read__(var, incr); -} - -long -ethr_atomic_inc_read(ethr_atomic_t *var) -{ - ETHR_ASSERT(!ethr_not_inited__); - ETHR_ASSERT(var); - return ethr_atomic_inc_read__(var); -} - -long -ethr_atomic_dec_read(ethr_atomic_t *var) -{ - ETHR_ASSERT(!ethr_not_inited__); - ETHR_ASSERT(var); - return ethr_atomic_dec_read__(var); -} - -void -ethr_atomic_add(ethr_atomic_t *var, long incr) -{ - ETHR_ASSERT(!ethr_not_inited__); - ETHR_ASSERT(var); - ethr_atomic_add__(var, incr); -} - -void -ethr_atomic_inc(ethr_atomic_t *var) -{ - ETHR_ASSERT(!ethr_not_inited__); - ETHR_ASSERT(var); - ethr_atomic_inc__(var); -} - -void -ethr_atomic_dec(ethr_atomic_t *var) -{ - ETHR_ASSERT(!ethr_not_inited__); - ETHR_ASSERT(var); - ethr_atomic_dec__(var); -} - -long -ethr_atomic_read_band(ethr_atomic_t *var, long mask) -{ - ETHR_ASSERT(!ethr_not_inited__); - ETHR_ASSERT(var); - return ethr_atomic_read_band__(var, mask); -} - -long -ethr_atomic_read_bor(ethr_atomic_t *var, long mask) -{ - ETHR_ASSERT(!ethr_not_inited__); - ETHR_ASSERT(var); - return ethr_atomic_read_bor__(var, mask); -} - -long -ethr_atomic_xchg(ethr_atomic_t *var, long new) -{ - ETHR_ASSERT(!ethr_not_inited__); - ETHR_ASSERT(var); - return ethr_atomic_xchg__(var, new); -} - -long -ethr_atomic_cmpxchg(ethr_atomic_t *var, long new, long expected) -{ - ETHR_ASSERT(!ethr_not_inited__); - ETHR_ASSERT(var); - return ethr_atomic_cmpxchg__(var, new, expected); -} - -long -ethr_atomic_read_acqb(ethr_atomic_t *var) -{ - ETHR_ASSERT(!ethr_not_inited__); - ETHR_ASSERT(var); - return ethr_atomic_read_acqb__(var); -} - -long -ethr_atomic_inc_read_acqb(ethr_atomic_t *var) -{ - ETHR_ASSERT(!ethr_not_inited__); - ETHR_ASSERT(var); - return ethr_atomic_inc_read_acqb__(var); -} - -void -ethr_atomic_set_relb(ethr_atomic_t *var, long i) -{ - ETHR_ASSERT(!ethr_not_inited__); - ETHR_ASSERT(var); - ethr_atomic_set_relb__(var, i); -} - -void -ethr_atomic_dec_relb(ethr_atomic_t *var) -{ - ETHR_ASSERT(!ethr_not_inited__); - ETHR_ASSERT(var); - ethr_atomic_dec_relb__(var); -} - -long -ethr_atomic_dec_read_relb(ethr_atomic_t *var) -{ - ETHR_ASSERT(!ethr_not_inited__); - ETHR_ASSERT(var); - return ethr_atomic_dec_read_relb__(var); -} - -long -ethr_atomic_cmpxchg_acqb(ethr_atomic_t *var, long new, long exp) -{ - ETHR_ASSERT(!ethr_not_inited__); - ETHR_ASSERT(var); - return ethr_atomic_cmpxchg_acqb__(var, new, exp); -} - -long -ethr_atomic_cmpxchg_relb(ethr_atomic_t *var, long new, long exp) -{ - ETHR_ASSERT(!ethr_not_inited__); - ETHR_ASSERT(var); - return ethr_atomic_cmpxchg_relb__(var, new, exp); -} - - /* Spinlocks and rwspinlocks */ int diff --git a/erts/lib_src/common/ethr_mutex.c b/erts/lib_src/common/ethr_mutex.c index a2fbf3a454..2ddef32dfc 100644 --- a/erts/lib_src/common/ethr_mutex.c +++ b/erts/lib_src/common/ethr_mutex.c @@ -206,16 +206,16 @@ static void hard_debug_chk_q__(struct ethr_mutex_base_ *, int); #ifdef ETHR_USE_OWN_RWMTX_IMPL__ static void rwmutex_transfer_read_lock(ethr_rwmutex *rwmtx, - long initial, + ethr_sint32_t initial, int q_locked); static void rwmutex_unlock_wake(ethr_rwmutex *rwmtx, int have_w, - long initial, + ethr_sint32_t initial, int transfer_read_lock); static int rwmutex_try_complete_runlock(ethr_rwmutex *rwmtx, - long initial, + ethr_sint32_t initial, ethr_ts_event *tse, int start_next_ix, int check_before_try, @@ -242,12 +242,12 @@ rwmutex_freqread_rdrs_add(ethr_rwmutex *rwmtx, int inc) { if (type == ETHR_RWMUTEX_TYPE_FREQUENT_READ || ix == 0) - ethr_atomic_add(&rwmtx->tdata.ra[ix].data.readers, inc); + ethr_atomic32_add(&rwmtx->tdata.ra[ix].data.readers, inc); else { ETHR_ASSERT(type == ETHR_RWMUTEX_TYPE_EXTREMELY_FREQUENT_READ); - ETHR_ASSERT(ethr_atomic_read(&rwmtx->tdata.ra[ix].data.readers) == 0); + ETHR_ASSERT(ethr_atomic32_read(&rwmtx->tdata.ra[ix].data.readers) == 0); ETHR_ASSERT(inc == 1); - ethr_atomic_set(&rwmtx->tdata.ra[ix].data.readers, (long) 1); + ethr_atomic32_set(&rwmtx->tdata.ra[ix].data.readers, (ethr_sint32_t) 1); } } @@ -258,15 +258,15 @@ rwmutex_freqread_rdrs_inc(ethr_rwmutex *rwmtx, ethr_ts_event *tse) if (rwmtx->type == ETHR_RWMUTEX_TYPE_FREQUENT_READ) { ix = tse->rgix; atomic_inc: - ethr_atomic_inc(&rwmtx->tdata.ra[ix].data.readers); + ethr_atomic32_inc(&rwmtx->tdata.ra[ix].data.readers); } else { ix = tse->mtix; if (ix == 0) goto atomic_inc; ETHR_ASSERT(rwmtx->type == ETHR_RWMUTEX_TYPE_EXTREMELY_FREQUENT_READ); - ETHR_ASSERT(ethr_atomic_read(&rwmtx->tdata.ra[ix].data.readers) == 0); - ethr_atomic_set(&rwmtx->tdata.ra[ix].data.readers, (long) 1); + ETHR_ASSERT(ethr_atomic32_read(&rwmtx->tdata.ra[ix].data.readers) == 0); + ethr_atomic32_set(&rwmtx->tdata.ra[ix].data.readers, (ethr_sint32_t) 1); } } @@ -279,64 +279,65 @@ rwmutex_freqread_rdrs_dec(ethr_rwmutex *rwmtx, ethr_ts_event *tse) if (rwmtx->type == ETHR_RWMUTEX_TYPE_FREQUENT_READ) { ix = tse->rgix; atomic_dec: - ethr_atomic_dec(&rwmtx->tdata.ra[ix].data.readers); + ethr_atomic32_dec(&rwmtx->tdata.ra[ix].data.readers); } else { ix = tse->mtix; if (ix == 0) goto atomic_dec; ETHR_ASSERT(rwmtx->type == ETHR_RWMUTEX_TYPE_EXTREMELY_FREQUENT_READ); - ETHR_ASSERT(ethr_atomic_read(&rwmtx->tdata.ra[ix].data.readers) == 1); - ethr_atomic_set(&rwmtx->tdata.ra[ix].data.readers, (long) 0); + ETHR_ASSERT(ethr_atomic32_read(&rwmtx->tdata.ra[ix].data.readers) == 1); + ethr_atomic32_set(&rwmtx->tdata.ra[ix].data.readers, (ethr_sint32_t) 0); } } #endif -static ETHR_INLINE long +static ETHR_INLINE ethr_sint32_t rwmutex_freqread_rdrs_dec_read(ethr_rwmutex *rwmtx, ethr_ts_event *tse) { int ix; if (rwmtx->type == ETHR_RWMUTEX_TYPE_FREQUENT_READ) { ix = tse->rgix; atomic_dec_read: - return ethr_atomic_dec_read(&rwmtx->tdata.ra[ix].data.readers); + return ethr_atomic32_dec_read(&rwmtx->tdata.ra[ix].data.readers); } else { ix = tse->mtix; if (ix == 0) goto atomic_dec_read; ETHR_ASSERT(rwmtx->type == ETHR_RWMUTEX_TYPE_EXTREMELY_FREQUENT_READ); - ETHR_ASSERT(ethr_atomic_read(&rwmtx->tdata.ra[ix].data.readers) == 1); - ethr_atomic_set(&rwmtx->tdata.ra[ix].data.readers, (long) 0); - return (long) 0; + ETHR_ASSERT(ethr_atomic32_read(&rwmtx->tdata.ra[ix].data.readers) == 1); + ethr_atomic32_set(&rwmtx->tdata.ra[ix].data.readers, (ethr_sint32_t) 0); + return (ethr_sint32_t) 0; } } -static ETHR_INLINE long +static ETHR_INLINE ethr_sint32_t rwmutex_freqread_rdrs_dec_read_relb(ethr_rwmutex *rwmtx, ethr_ts_event *tse) { int ix; if (rwmtx->type == ETHR_RWMUTEX_TYPE_FREQUENT_READ) { ix = tse->rgix; atomic_dec_read: - return ethr_atomic_dec_read_relb(&rwmtx->tdata.ra[ix].data.readers); + return ethr_atomic32_dec_read_relb(&rwmtx->tdata.ra[ix].data.readers); } else { ix = tse->mtix; if (ix == 0) goto atomic_dec_read; ETHR_ASSERT(rwmtx->type == ETHR_RWMUTEX_TYPE_EXTREMELY_FREQUENT_READ); - ETHR_ASSERT(ethr_atomic_read(&rwmtx->tdata.ra[ix].data.readers) == 1); - ethr_atomic_set_relb(&rwmtx->tdata.ra[ix].data.readers, (long) 0); - return (long) 0; + ETHR_ASSERT(ethr_atomic32_read(&rwmtx->tdata.ra[ix].data.readers) == 1); + ethr_atomic32_set_relb(&rwmtx->tdata.ra[ix].data.readers, + (ethr_sint32_t) 0); + return (ethr_sint32_t) 0; } } -static ETHR_INLINE long +static ETHR_INLINE ethr_sint32_t rwmutex_freqread_rdrs_read(ethr_rwmutex *rwmtx, int ix) { - long res = ethr_atomic_read(&rwmtx->tdata.ra[ix].data.readers); + ethr_sint32_t res = ethr_atomic32_read(&rwmtx->tdata.ra[ix].data.readers); #ifdef ETHR_DEBUG switch (rwmtx->type) { case ETHR_RWMUTEX_TYPE_FREQUENT_READ: @@ -402,19 +403,19 @@ static void event_wait(struct ethr_mutex_base_ *mtxb, ethr_ts_event *tse, int spincount, - long type, + ethr_sint32_t type, int is_rwmtx, int is_freq_read) { int locked = 0; - long act; + ethr_sint32_t act; int need_try_complete_runlock = 0; int transfer_read_lock = 0; /* Need to enqueue and wait... */ tse->uflgs = type; - ethr_atomic_set(&tse->uaflgs, type); + ethr_atomic32_set(&tse->uaflgs, type); ETHR_MTX_Q_LOCK(&mtxb->qlck); locked = 1; @@ -423,7 +424,7 @@ event_wait(struct ethr_mutex_base_ *mtxb, hard_debug_chk_q__(mtxb, is_rwmtx); #endif - act = ethr_atomic_read(&mtxb->flgs); + act = ethr_atomic32_read(&mtxb->flgs); if (act & type) { @@ -453,7 +454,7 @@ event_wait(struct ethr_mutex_base_ *mtxb, /* Set wait bit */ while (1) { - long new, exp = act; + ethr_sint32_t new, exp = act; need_try_complete_runlock = 0; transfer_read_lock = 0; @@ -484,7 +485,7 @@ event_wait(struct ethr_mutex_base_ *mtxb, } } - act = ethr_atomic_cmpxchg_acqb(&mtxb->flgs, new, exp); + act = ethr_atomic32_cmpxchg_acqb(&mtxb->flgs, new, exp); if (exp == act) { if (new & type) { act = new; @@ -559,7 +560,7 @@ event_wait(struct ethr_mutex_base_ *mtxb, while (1) { ethr_event_reset(&tse->event); - act = ethr_atomic_read_acqb(&tse->uaflgs); + act = ethr_atomic32_read_acqb(&tse->uaflgs); if (!act) goto done; /* Got it */ @@ -567,7 +568,7 @@ event_wait(struct ethr_mutex_base_ *mtxb, ethr_event_swait(&tse->event, spincount); /* swait result: 0 || EINTR */ - act = ethr_atomic_read_acqb(&tse->uaflgs); + act = ethr_atomic32_read_acqb(&tse->uaflgs); if (!act) goto done; /* Got it */ } @@ -587,7 +588,7 @@ wake_writer(struct ethr_mutex_base_ *mtxb, int is_rwmtx) dequeue(&mtxb->q, tse, tse); ETHR_ASSERT(tse->uflgs == ETHR_RWMTX_W_WAIT_FLG__); - ETHR_ASSERT(ethr_atomic_read(&tse->uaflgs) == ETHR_RWMTX_W_WAIT_FLG__); + ETHR_ASSERT(ethr_atomic32_read(&tse->uaflgs) == ETHR_RWMTX_W_WAIT_FLG__); #ifdef ETHR_MTX_HARD_DEBUG_WSQ mtxb->ws--; #endif @@ -597,7 +598,7 @@ wake_writer(struct ethr_mutex_base_ *mtxb, int is_rwmtx) ETHR_MTX_Q_UNLOCK(&mtxb->qlck); - ethr_atomic_set(&tse->uaflgs, 0); + ethr_atomic32_set(&tse->uaflgs, 0); ethr_event_set(&tse->event); } @@ -649,11 +650,11 @@ int check_readers_array(ethr_rwmutex *rwmtx, static ETHR_INLINE void write_lock_wait(struct ethr_mutex_base_ *mtxb, - long initial, + ethr_sint32_t initial, int is_rwmtx, int is_freq_read) { - long act = initial; + ethr_sint32_t act = initial; int scnt, start_scnt; ethr_ts_event *tse = NULL; int until_yield = ETHR_YIELD_AFTER_BUSY_LOOPS; @@ -706,13 +707,13 @@ write_lock_wait(struct ethr_mutex_base_ *mtxb, until_yield = ETHR_YIELD_AFTER_BUSY_LOOPS; ETHR_YIELD(); } - act = ethr_atomic_read(&mtxb->flgs); + act = ethr_atomic32_read(&mtxb->flgs); scnt--; } - act = ethr_atomic_cmpxchg_acqb(&mtxb->flgs, - ETHR_RWMTX_W_FLG__, - 0); + act = ethr_atomic32_cmpxchg_acqb(&mtxb->flgs, + ETHR_RWMTX_W_FLG__, + 0); if (act == 0) goto done; /* Got it */ } @@ -756,16 +757,16 @@ mtxb_init(struct ethr_mutex_base_ *mtxb, } mtxb->q = NULL; - ethr_atomic_init(&mtxb->flgs, 0); + ethr_atomic32_init(&mtxb->flgs, 0); return ETHR_MTX_QLOCK_INIT(&mtxb->qlck); } static int mtxb_destroy(struct ethr_mutex_base_ *mtxb) { - long act; + ethr_sint32_t act; ETHR_MTX_Q_LOCK(&mtxb->qlck); - act = ethr_atomic_read(&mtxb->flgs); + act = ethr_atomic32_read(&mtxb->flgs); ETHR_MTX_Q_UNLOCK(&mtxb->qlck); if (act != 0) return EINVAL; @@ -831,13 +832,13 @@ ethr_mutex_destroy(ethr_mutex *mtx) } void -ethr_mutex_lock_wait__(ethr_mutex *mtx, long initial) +ethr_mutex_lock_wait__(ethr_mutex *mtx, ethr_sint32_t initial) { write_lock_wait(&mtx->mtxb, initial, 0, 0); } void -ethr_mutex_unlock_wake__(ethr_mutex *mtx, long initial) +ethr_mutex_unlock_wake__(ethr_mutex *mtx, ethr_sint32_t initial) { ethr_ts_event *tse; @@ -845,7 +846,7 @@ ethr_mutex_unlock_wake__(ethr_mutex *mtx, long initial) tse = mtx->mtxb.q; ETHR_ASSERT(tse); - ETHR_ASSERT(ethr_atomic_read(&mtx->mtxb.flgs) + ETHR_ASSERT(ethr_atomic32_read(&mtx->mtxb.flgs) == (ETHR_RWMTX_W_FLG__|ETHR_RWMTX_W_WAIT_FLG__)); ETHR_ASSERT(initial & ETHR_RWMTX_W_WAIT_FLG__); ETHR_MTX_HARD_DEBUG_CHK_Q(mtx); @@ -855,7 +856,7 @@ ethr_mutex_unlock_wake__(ethr_mutex *mtx, long initial) * mtxb->flgs; otherwise, we need to clear the write wait bit... */ if (tse->next == mtx->mtxb.q) - ethr_atomic_set(&mtx->mtxb.flgs, ETHR_RWMTX_W_FLG__); + ethr_atomic32_set(&mtx->mtxb.flgs, ETHR_RWMTX_W_FLG__); wake_writer(&mtx->mtxb, 0); } @@ -865,7 +866,7 @@ ethr_mutex_unlock_wake__(ethr_mutex *mtx, long initial) static void enqueue_mtx(ethr_mutex *mtx, ethr_ts_event *tse_start, ethr_ts_event *tse_end) { - long act; + ethr_sint32_t act; /* * `ethr_cond_signal()' and `ethr_cond_broadcast()' end up here. If `mtx' @@ -894,7 +895,7 @@ enqueue_mtx(ethr_mutex *mtx, ethr_ts_event *tse_start, ethr_ts_event *tse_end) } #endif - act = ethr_atomic_read(&mtx->mtxb.flgs); + act = ethr_atomic32_read(&mtx->mtxb.flgs); ETHR_ASSERT(act == 0 || act == ETHR_RWMTX_W_FLG__ || act == (ETHR_RWMTX_W_FLG__|ETHR_RWMTX_W_WAIT_FLG__)); @@ -902,10 +903,10 @@ enqueue_mtx(ethr_mutex *mtx, ethr_ts_event *tse_start, ethr_ts_event *tse_end) /* The normal sane case */ if (!(act & ETHR_RWMTX_W_WAIT_FLG__)) { ETHR_ASSERT(!mtx->mtxb.q); - act = ethr_atomic_cmpxchg(&mtx->mtxb.flgs, - (ETHR_RWMTX_W_FLG__ - | ETHR_RWMTX_W_WAIT_FLG__), - ETHR_RWMTX_W_FLG__); + act = ethr_atomic32_cmpxchg(&mtx->mtxb.flgs, + (ETHR_RWMTX_W_FLG__ + | ETHR_RWMTX_W_WAIT_FLG__), + ETHR_RWMTX_W_FLG__); if (act != ETHR_RWMTX_W_FLG__) { /* * Sigh... this wasn't so sane after all since, the mutex was @@ -937,14 +938,14 @@ enqueue_mtx(ethr_mutex *mtx, ethr_ts_event *tse_start, ethr_ts_event *tse_end) multi = tse_start != tse_end; while (1) { - long new, exp = act; + ethr_sint32_t new, exp = act; if (multi || (act & ETHR_RWMTX_W_FLG__)) new = ETHR_RWMTX_W_FLG__|ETHR_RWMTX_W_WAIT_FLG__; else new = ETHR_RWMTX_W_FLG__; - act = ethr_atomic_cmpxchg(&mtx->mtxb.flgs, new, exp); + act = ethr_atomic32_cmpxchg(&mtx->mtxb.flgs, new, exp); if (exp == act) { ETHR_ASSERT(!mtx->mtxb.q); if (act & ETHR_RWMTX_W_FLG__) { @@ -972,7 +973,7 @@ enqueue_mtx(ethr_mutex *mtx, ethr_ts_event *tse_start, ethr_ts_event *tse_end) ETHR_MTX_HARD_DEBUG_CHK_Q(mtx); ETHR_MTX_Q_UNLOCK(&mtx->mtxb.qlck); - ethr_atomic_set(&tse_start->uaflgs, 0); + ethr_atomic32_set(&tse_start->uaflgs, 0); ethr_event_set(&tse_start->event); } break; @@ -1063,9 +1064,9 @@ ethr_cond_signal(ethr_cond *cnd) ETHR_MTX_HARD_DEBUG_FENCE_CHK(mtx); ETHR_ASSERT(tse->uflgs == ETHR_RWMTX_W_WAIT_FLG__); - ETHR_ASSERT(ethr_atomic_read(&tse->uaflgs) == ETHR_CND_WAIT_FLG__); + ETHR_ASSERT(ethr_atomic32_read(&tse->uaflgs) == ETHR_CND_WAIT_FLG__); - ethr_atomic_set(&tse->uaflgs, ETHR_RWMTX_W_WAIT_FLG__); + ethr_atomic32_set(&tse->uaflgs, ETHR_RWMTX_W_WAIT_FLG__); dequeue(&cnd->q, tse, tse); @@ -1116,10 +1117,11 @@ ethr_cond_broadcast(ethr_cond *cnd) /* The normal case */ ETHR_ASSERT(tse_tmp->uflgs == ETHR_RWMTX_W_WAIT_FLG__); - ETHR_ASSERT(ethr_atomic_read(&tse_tmp->uaflgs) + ETHR_ASSERT(ethr_atomic32_read(&tse_tmp->uaflgs) == ETHR_CND_WAIT_FLG__); - ethr_atomic_set(&tse_tmp->uaflgs, ETHR_RWMTX_W_WAIT_FLG__); + ethr_atomic32_set(&tse_tmp->uaflgs, + ETHR_RWMTX_W_WAIT_FLG__); } else { /* Should be very unusual */ @@ -1172,7 +1174,7 @@ ethr_cond_wait(ethr_cond *cnd, ethr_mutex *mtx) tse->udata = (void *) mtx; tse->uflgs = ETHR_RWMTX_W_WAIT_FLG__; /* Prep for mutex lock op */ - ethr_atomic_set(&tse->uaflgs, ETHR_CND_WAIT_FLG__); + ethr_atomic32_set(&tse->uaflgs, ETHR_CND_WAIT_FLG__); ETHR_MTX_Q_LOCK(&cnd->qlck); @@ -1185,11 +1187,11 @@ ethr_cond_wait(ethr_cond *cnd, ethr_mutex *mtx) /* Wait */ woken = 0; while (1) { - long act; + ethr_sint32_t act; ethr_event_reset(&tse->event); - act = ethr_atomic_read_acqb(&tse->uaflgs); + act = ethr_atomic32_read_acqb(&tse->uaflgs); if (!act) break; /* Mtx locked */ @@ -1205,7 +1207,7 @@ ethr_cond_wait(ethr_cond *cnd, ethr_mutex *mtx) */ if (act == ETHR_CND_WAIT_FLG__) { ETHR_MTX_Q_LOCK(&cnd->qlck); - act = ethr_atomic_read(&tse->uaflgs); + act = ethr_atomic32_read(&tse->uaflgs); ETHR_ASSERT(act == ETHR_CND_WAIT_FLG__ || act == ETHR_RWMTX_W_WAIT_FLG__); /* @@ -1407,7 +1409,7 @@ wake_readers(ethr_rwmutex *rwmtx, int rs) rwmtx->rq_end = NULL; ETHR_ASSERT(!rwmtx->mtxb.q - || (ethr_atomic_read(&rwmtx->mtxb.q->uaflgs) + || (ethr_atomic32_read(&rwmtx->mtxb.q->uaflgs) == ETHR_RWMTX_W_WAIT_FLG__)); ETHR_RWMTX_HARD_DEBUG_CHK_Q(rwmtx); @@ -1418,7 +1420,7 @@ wake_readers(ethr_rwmutex *rwmtx, int rs) #ifdef ETHR_DEBUG ETHR_ASSERT(tse->uflgs == ETHR_RWMTX_R_WAIT_FLG__); - ETHR_ASSERT(ethr_atomic_read(&tse->uaflgs) + ETHR_ASSERT(ethr_atomic32_read(&tse->uaflgs) == ETHR_RWMTX_R_WAIT_FLG__); drs++; #endif @@ -1426,7 +1428,7 @@ wake_readers(ethr_rwmutex *rwmtx, int rs) tse_next = tse->next; /* we aren't allowed to read tse->next after we have reset uaflgs */ - ethr_atomic_set(&tse->uaflgs, 0); + ethr_atomic32_set(&tse->uaflgs, 0); ethr_event_set(&tse->event); tse = tse_next; } @@ -1469,7 +1471,7 @@ int check_readers_array(ethr_rwmutex *rwmtx, ETHR_MEMORY_BARRIER; do { - long act = rwmutex_freqread_rdrs_read(rwmtx, ix); + ethr_sint32_t act = rwmutex_freqread_rdrs_read(rwmtx, ix); if (act != 0) return EBUSY; ix++; @@ -1483,9 +1485,9 @@ int check_readers_array(ethr_rwmutex *rwmtx, static void rwmutex_freqread_rdrs_dec_chk_wakeup(ethr_rwmutex *rwmtx, ethr_ts_event *tse, - long initial) + ethr_sint32_t initial) { - long act = initial; + ethr_sint32_t act = initial; if ((act & (ETHR_RWMTX_W_FLG__| ETHR_RWMTX_R_ABRT_UNLCK_FLG__)) == 0) { @@ -1515,7 +1517,7 @@ rwmutex_freqread_rdrs_dec_chk_wakeup(ethr_rwmutex *rwmtx, if (!rwmtx->mtxb.q) ETHR_MTX_Q_UNLOCK(&rwmtx->mtxb.qlck); else if (is_w_waiter(rwmtx->mtxb.q)) { - act = ethr_atomic_read(&rwmtx->mtxb.flgs); + act = ethr_atomic32_read(&rwmtx->mtxb.flgs); ETHR_MTX_Q_UNLOCK(&rwmtx->mtxb.qlck); if ((act & ETHR_RWMTX_W_FLG__) == 0) rwmutex_try_complete_runlock(rwmtx, act, tse, 1, 0, 0); @@ -1525,7 +1527,7 @@ rwmutex_freqread_rdrs_dec_chk_wakeup(ethr_rwmutex *rwmtx, * rwmutex_transfer_read_lock() will * unlock Q lock. */ - act = ethr_atomic_read(&rwmtx->mtxb.flgs); + act = ethr_atomic32_read(&rwmtx->mtxb.flgs); if (act & ETHR_RWMTX_W_FLG__) ETHR_MTX_Q_UNLOCK(&rwmtx->mtxb.qlck); else @@ -1539,7 +1541,7 @@ static void rwmutex_freqread_restore_failed_tryrlock(ethr_rwmutex *rwmtx, ethr_ts_event *tse) { - long act; + ethr_sint32_t act; /* * Restore failed increment */ @@ -1548,21 +1550,21 @@ rwmutex_freqread_restore_failed_tryrlock(ethr_rwmutex *rwmtx, ETHR_MEMORY_BARRIER; if (act == 0) { - act = ethr_atomic_read(&rwmtx->mtxb.flgs); + act = ethr_atomic32_read(&rwmtx->mtxb.flgs); rwmutex_freqread_rdrs_dec_chk_wakeup(rwmtx, tse, act); } } static int rwmutex_try_complete_runlock(ethr_rwmutex *rwmtx, - long initial, + ethr_sint32_t initial, ethr_ts_event *tse, int start_next_ix, int check_before_try, int try_write_lock) { ethr_ts_event *tse_tmp; - long act = initial; + ethr_sint32_t act = initial; int six, res, length; ETHR_ASSERT((act & ETHR_RWMTX_W_FLG__) == 0); @@ -1606,15 +1608,15 @@ rwmutex_try_complete_runlock(ethr_rwmutex *rwmtx, restart: while (1) { - long exp = act; - long new = act+1; + ethr_sint32_t exp = act; + ethr_sint32_t new = act+1; ETHR_ASSERT((act & ETHR_RWMTX_R_ABRT_UNLCK_FLG__) == 0); ETHR_ASSERT((act & ETHR_RWMTX_R_PEND_UNLCK_MASK__) < ETHR_RWMTX_R_PEND_UNLCK_MASK__); - act = ethr_atomic_cmpxchg(&rwmtx->mtxb.flgs, new, exp); + act = ethr_atomic32_cmpxchg(&rwmtx->mtxb.flgs, new, exp); if (exp == act) { act = new; break; @@ -1651,8 +1653,8 @@ rwmutex_try_complete_runlock(ethr_rwmutex *rwmtx, while (1) { int finished_abort = 0; - long exp = act; - long new = act; + ethr_sint32_t exp = act; + ethr_sint32_t new = act; new--; if (act & ETHR_RWMTX_R_ABRT_UNLCK_FLG__) { @@ -1668,7 +1670,7 @@ rwmutex_try_complete_runlock(ethr_rwmutex *rwmtx, ETHR_ASSERT(act & ETHR_RWMTX_R_PEND_UNLCK_MASK__); - act = ethr_atomic_cmpxchg(&rwmtx->mtxb.flgs, new, exp); + act = ethr_atomic32_cmpxchg(&rwmtx->mtxb.flgs, new, exp); if (exp == act) { act = new; if (act & ETHR_RWMTX_W_FLG__) @@ -1702,9 +1704,9 @@ rwmutex_try_complete_runlock(ethr_rwmutex *rwmtx, tryrwlock: /* Try to write lock it */ - act = ethr_atomic_cmpxchg_acqb(&rwmtx->mtxb.flgs, - ETHR_RWMTX_W_FLG__, - 0); + act = ethr_atomic32_cmpxchg_acqb(&rwmtx->mtxb.flgs, + ETHR_RWMTX_W_FLG__, + 0); return act == 0 ? 0 : EBUSY; } @@ -1713,11 +1715,11 @@ rwmutex_try_complete_runlock(ethr_rwmutex *rwmtx, static ETHR_INLINE void rwmutex_incdec_restore_failed_tryrlock(ethr_rwmutex *rwmtx) { - long act; + ethr_sint32_t act; /* * Restore failed increment */ - act = ethr_atomic_dec_read(&rwmtx->mtxb.flgs); + act = ethr_atomic32_dec_read(&rwmtx->mtxb.flgs); if ((act & ETHR_RWMTX_WAIT_FLGS__) && (act & ~ETHR_RWMTX_WAIT_FLGS__) == 0) { rwmutex_unlock_wake(rwmtx, 0, act, 0); @@ -1727,10 +1729,9 @@ rwmutex_incdec_restore_failed_tryrlock(ethr_rwmutex *rwmtx) #endif static void -rwmutex_normal_rlock_wait(ethr_rwmutex *rwmtx, - long initial) +rwmutex_normal_rlock_wait(ethr_rwmutex *rwmtx, ethr_sint32_t initial) { - long act = initial, exp; + ethr_sint32_t act = initial, exp; int scnt, start_scnt; ethr_ts_event *tse = NULL; int until_yield = ETHR_YIELD_AFTER_BUSY_LOOPS; @@ -1746,7 +1747,7 @@ rwmutex_normal_rlock_wait(ethr_rwmutex *rwmtx, #ifdef ETHR_RLOCK_WITH_INC_DEC rwmutex_incdec_restore_failed_tryrlock(rwmtx); - act = ethr_atomic_read(&rwmtx->mtxb.flgs); + act = ethr_atomic32_read(&rwmtx->mtxb.flgs); #endif while (act & (ETHR_RWMTX_W_FLG__|ETHR_RWMTX_W_WAIT_FLG__)) { @@ -1763,17 +1764,17 @@ rwmutex_normal_rlock_wait(ethr_rwmutex *rwmtx, until_yield = ETHR_YIELD_AFTER_BUSY_LOOPS; ETHR_YIELD(); } - act = ethr_atomic_read(&rwmtx->mtxb.flgs); + act = ethr_atomic32_read(&rwmtx->mtxb.flgs); scnt--; } exp = act; #ifdef ETHR_RLOCK_WITH_INC_DEC - act = ethr_atomic_inc_read(&rwmtx->mtxb.flgs); + act = ethr_atomic32_inc_read(&rwmtx->mtxb.flgs); if ((act & (ETHR_RWMTX_W_FLG__|ETHR_RWMTX_W_WAIT_FLG__)) == 0) goto done; /* Got it */ #else - act = ethr_atomic_cmpxchg_acqb(&rwmtx->mtxb.flgs, exp+1, exp); + act = ethr_atomic32_cmpxchg_acqb(&rwmtx->mtxb.flgs, exp+1, exp); if (act == exp) goto done; /* Got it */ #endif @@ -1792,19 +1793,19 @@ static int rwmutex_freqread_rlock(ethr_rwmutex *rwmtx, ethr_ts_event *tse, int trylock) { int res = 0; - long act; + ethr_sint32_t act; rwmutex_freqread_rdrs_inc(rwmtx, tse); ETHR_MEMORY_BARRIER; - act = ethr_atomic_read_acqb(&rwmtx->mtxb.flgs); + act = ethr_atomic32_read_acqb(&rwmtx->mtxb.flgs); if (act != ETHR_RWMTX_R_FLG__) { int wake_other_readers; while (1) { - long exp, new; + ethr_sint32_t exp, new; wake_other_readers = 0; @@ -1846,7 +1847,7 @@ rwmutex_freqread_rlock(ethr_rwmutex *rwmtx, ethr_ts_event *tse, int trylock) } exp = act; - act = ethr_atomic_cmpxchg_acqb(&rwmtx->mtxb.flgs, new, exp); + act = ethr_atomic32_cmpxchg_acqb(&rwmtx->mtxb.flgs, new, exp); if (act == exp) break; } @@ -1862,7 +1863,7 @@ static void rwmutex_freqread_rlock_wait(ethr_rwmutex *rwmtx, ethr_ts_event *tse) { - long act; + ethr_sint32_t act; int scnt, start_scnt; int until_yield = ETHR_YIELD_AFTER_BUSY_LOOPS; @@ -1875,7 +1876,7 @@ rwmutex_freqread_rlock_wait(ethr_rwmutex *rwmtx, while (1) { - act = ethr_atomic_read(&rwmtx->mtxb.flgs); + act = ethr_atomic32_read(&rwmtx->mtxb.flgs); while (act & (ETHR_RWMTX_W_FLG__|ETHR_RWMTX_W_WAIT_FLG__)) { if (scnt <= 0) { @@ -1890,7 +1891,7 @@ rwmutex_freqread_rlock_wait(ethr_rwmutex *rwmtx, until_yield = ETHR_YIELD_AFTER_BUSY_LOOPS; ETHR_YIELD(); } - act = ethr_atomic_read(&rwmtx->mtxb.flgs); + act = ethr_atomic32_read(&rwmtx->mtxb.flgs); scnt--; } @@ -1900,21 +1901,23 @@ rwmutex_freqread_rlock_wait(ethr_rwmutex *rwmtx, } static void -rwmutex_normal_rwlock_wait(ethr_rwmutex *rwmtx, long initial) +rwmutex_normal_rwlock_wait(ethr_rwmutex *rwmtx, ethr_sint32_t initial) { write_lock_wait(&rwmtx->mtxb, initial, 1, 0); } static void -rwmutex_freqread_rwlock_wait(ethr_rwmutex *rwmtx, long initial) +rwmutex_freqread_rwlock_wait(ethr_rwmutex *rwmtx, ethr_sint32_t initial) { write_lock_wait(&rwmtx->mtxb, initial, 1, 1); } static ETHR_INLINE void -rwlock_wake_set_flags(ethr_rwmutex *rwmtx, long new_initial, long act_initial) +rwlock_wake_set_flags(ethr_rwmutex *rwmtx, + ethr_sint32_t new_initial, + ethr_sint32_t act_initial) { - long act, act_mask; + ethr_sint32_t act, act_mask; int chk_abrt_flg; ETHR_MEMORY_BARRIER; @@ -1935,18 +1938,18 @@ rwlock_wake_set_flags(ethr_rwmutex *rwmtx, long new_initial, long act_initial) #else /* rs mask always zero */ ETHR_ASSERT((act_initial & ETHR_RWMTX_RS_MASK__) == 0); - ethr_atomic_set(&rwmtx->mtxb.flgs, new_initial); + ethr_atomic32_set(&rwmtx->mtxb.flgs, new_initial); return; #endif } act = act_initial; while (1) { - long exp = act; - long new = new_initial + (act & act_mask); + ethr_sint32_t exp = act; + ethr_sint32_t new = new_initial + (act & act_mask); if (chk_abrt_flg && (act & act_mask)) new |= ETHR_RWMTX_R_ABRT_UNLCK_FLG__; - act = ethr_atomic_cmpxchg(&rwmtx->mtxb.flgs, new, exp); + act = ethr_atomic32_cmpxchg(&rwmtx->mtxb.flgs, new, exp); if (act == exp) break; exp = act; @@ -1960,7 +1963,7 @@ dbg_unlock_wake(ethr_rwmutex *rwmtx, int have_w, ethr_ts_event *tse) { - long exp, act, imask; + ethr_sint32_t exp, act, imask; exp = have_w ? ETHR_RWMTX_W_FLG__ : 0; @@ -1982,7 +1985,7 @@ dbg_unlock_wake(ethr_rwmutex *rwmtx, if (rwmtx->rq_end) { exp |= ETHR_RWMTX_R_WAIT_FLG__; } - act = ethr_atomic_read(&rwmtx->mtxb.flgs); + act = ethr_atomic32_read(&rwmtx->mtxb.flgs); ETHR_ASSERT((exp & ~imask) == (act & ~imask)); ETHR_RWMTX_HARD_DEBUG_CHK_Q(rwmtx); @@ -2001,7 +2004,7 @@ dbg_unlock_wake(ethr_rwmutex *rwmtx, imask |= ETHR_RWMTX_RS_MASK__; } } - act = ethr_atomic_read(&rwmtx->mtxb.flgs); + act = ethr_atomic32_read(&rwmtx->mtxb.flgs); ETHR_ASSERT((exp & ~imask) == (act & ~imask)); ETHR_RWMTX_HARD_DEBUG_CHK_Q(rwmtx); @@ -2012,9 +2015,11 @@ dbg_unlock_wake(ethr_rwmutex *rwmtx, #endif static void -rwmutex_transfer_read_lock(ethr_rwmutex *rwmtx, long initial, int q_locked) +rwmutex_transfer_read_lock(ethr_rwmutex *rwmtx, + ethr_sint32_t initial, + int q_locked) { - long act = initial; + ethr_sint32_t act = initial; if (!q_locked) { ethr_ts_event *tse; @@ -2022,7 +2027,7 @@ rwmutex_transfer_read_lock(ethr_rwmutex *rwmtx, long initial, int q_locked) ETHR_ASSERT((initial & ETHR_RWMTX_W_FLG__) == 0); ETHR_MTX_Q_LOCK(&rwmtx->mtxb.qlck); - act = ethr_atomic_read(&rwmtx->mtxb.flgs); + act = ethr_atomic32_read(&rwmtx->mtxb.flgs); tse = rwmtx->mtxb.q; if ((act & ETHR_RWMTX_W_FLG__) || !tse || is_w_waiter(tse)) { /* Someone else woke the readers up... */ @@ -2035,10 +2040,10 @@ rwmutex_transfer_read_lock(ethr_rwmutex *rwmtx, long initial, int q_locked) } static void -rwmutex_unlock_wake(ethr_rwmutex *rwmtx, int have_w, long initial, +rwmutex_unlock_wake(ethr_rwmutex *rwmtx, int have_w, ethr_sint32_t initial, int transfer_read_lock) { - long new, act = initial; + ethr_sint32_t new, act = initial; ethr_ts_event *tse; if (transfer_read_lock) { @@ -2060,9 +2065,9 @@ rwmutex_unlock_wake(ethr_rwmutex *rwmtx, int have_w, long initial, return; else { while ((act & ETHR_RWMTX_WAIT_FLGS__) == 0) { - long exp = act; + ethr_sint32_t exp = act; new = exp & ~ETHR_RWMTX_W_FLG__; - act = ethr_atomic_cmpxchg(&rwmtx->mtxb.flgs, new, exp); + act = ethr_atomic32_cmpxchg(&rwmtx->mtxb.flgs, new, exp); if (act == exp) return; } @@ -2075,12 +2080,12 @@ rwmutex_unlock_wake(ethr_rwmutex *rwmtx, int have_w, long initial, if (!have_w) { if (!tse) { #ifdef ETHR_DEBUG - act = ethr_atomic_read(&rwmtx->mtxb.flgs); + act = ethr_atomic32_read(&rwmtx->mtxb.flgs); ETHR_ASSERT((act & ETHR_RWMTX_WAIT_FLGS__) == 0); #endif goto already_served; } - act = ethr_atomic_read(&rwmtx->mtxb.flgs); + act = ethr_atomic32_read(&rwmtx->mtxb.flgs); if (act == (ETHR_RWMTX_R_WAIT_FLG__|ETHR_RWMTX_R_FLG__)) { ETHR_ASSERT(tse && !is_w_waiter(tse)); } @@ -2099,7 +2104,7 @@ rwmutex_unlock_wake(ethr_rwmutex *rwmtx, int have_w, long initial, if (is_w_waiter(tse)) { if (!have_w) { - act = ethr_atomic_read_bor(&rwmtx->mtxb.flgs, + act = ethr_atomic32_read_bor(&rwmtx->mtxb.flgs, ETHR_RWMTX_W_FLG__); ETHR_ASSERT((act & ~(ETHR_RWMTX_WAIT_FLGS__ | (rwmtx->type == ETHR_RWMUTEX_TYPE_NORMAL @@ -2131,7 +2136,7 @@ rwmutex_unlock_wake(ethr_rwmutex *rwmtx, int have_w, long initial, if (rwmtx->type == ETHR_RWMUTEX_TYPE_NORMAL) { rs = rwmtx->tdata.rs; - new = (long) rs; + new = (ethr_sint32_t) rs; rwmtx->tdata.rs = 0; } else { @@ -2187,16 +2192,16 @@ alloc_readers_array(int length, ethr_rwmutex_lived lived) if (!mem) return NULL; - if ((((unsigned long) mem) & ETHR_CACHE_LINE_MASK) == 0) { + if ((((ethr_uint_t) mem) & ETHR_CACHE_LINE_MASK) == 0) { ra = (ethr_rwmtx_readers_array__ *) mem; ra->data.byte_offset = 0; } else { ra = ((ethr_rwmtx_readers_array__ *) - ((((unsigned long) mem) & ~ETHR_CACHE_LINE_MASK) + ((((ethr_uint_t) mem) & ~ETHR_CACHE_LINE_MASK) + ETHR_CACHE_LINE_SIZE)); - ra->data.byte_offset = (int) ((unsigned long) ra - - (unsigned long) mem); + ra->data.byte_offset = (int) ((ethr_uint_t) ra + - (ethr_uint_t) mem); } ra->data.lived = lived; return ra; @@ -2270,7 +2275,7 @@ ethr_rwmutex_init_opt(ethr_rwmutex *rwmtx, ethr_rwmutex_opt *opt) rwmtx->tdata.ra = ra; for (ix = 0; ix < length; ix++) { - ethr_atomic_init(&rwmtx->tdata.ra[ix].data.readers, 0); + ethr_atomic32_init(&rwmtx->tdata.ra[ix].data.readers, 0); rwmtx->tdata.ra[ix].data.waiting_readers = 0; } break; @@ -2324,7 +2329,7 @@ ethr_rwmutex_destroy(ethr_rwmutex *rwmtx) #endif ETHR_MTX_DBG_CHK_UNUSED_FLG_BITS(rwmtx); if (rwmtx->type != ETHR_RWMUTEX_TYPE_NORMAL) { - long act = ethr_atomic_read(&rwmtx->mtxb.flgs); + ethr_sint32_t act = ethr_atomic32_read(&rwmtx->mtxb.flgs); if (act == ETHR_RWMTX_R_FLG__) rwmutex_try_complete_runlock(rwmtx, act, NULL, 0, 0, 0); } @@ -2345,7 +2350,7 @@ int ethr_rwmutex_tryrlock(ethr_rwmutex *rwmtx) { int res = 0; - long act; + ethr_sint32_t act; ETHR_ASSERT(!ethr_not_inited__); ETHR_ASSERT(rwmtx); @@ -2358,22 +2363,22 @@ ethr_rwmutex_tryrlock(ethr_rwmutex *rwmtx) switch (rwmtx->type) { case ETHR_RWMUTEX_TYPE_NORMAL: { #ifdef ETHR_RLOCK_WITH_INC_DEC - act = ethr_atomic_read(&rwmtx->mtxb.flgs); + act = ethr_atomic32_read(&rwmtx->mtxb.flgs); if (act & (ETHR_RWMTX_W_FLG__|ETHR_RWMTX_W_WAIT_FLG__)) res = EBUSY; else { - act = ethr_atomic_inc_read_acqb(&rwmtx->mtxb.flgs); + act = ethr_atomic32_inc_read_acqb(&rwmtx->mtxb.flgs); if (act & (ETHR_RWMTX_W_FLG__|ETHR_RWMTX_W_WAIT_FLG__)) { rwmutex_incdec_restore_failed_tryrlock(rwmtx); res = EBUSY; } } #else - long exp = 0; + ethr_sint32_t exp = 0; int tries = 0; while (1) { - act = ethr_atomic_cmpxchg_acqb(&rwmtx->mtxb.flgs, exp+1, exp); + act = ethr_atomic32_cmpxchg_acqb(&rwmtx->mtxb.flgs, exp+1, exp); if (act == exp) { res = 0; break; @@ -2416,7 +2421,7 @@ ethr_rwmutex_tryrlock(ethr_rwmutex *rwmtx) void ethr_rwmutex_rlock(ethr_rwmutex *rwmtx) { - long act; + ethr_sint32_t act; ETHR_ASSERT(!ethr_not_inited__); ETHR_ASSERT(rwmtx); @@ -2429,14 +2434,14 @@ ethr_rwmutex_rlock(ethr_rwmutex *rwmtx) switch (rwmtx->type) { case ETHR_RWMUTEX_TYPE_NORMAL: { #ifdef ETHR_RLOCK_WITH_INC_DEC - act = ethr_atomic_inc_read_acqb(&rwmtx->mtxb.flgs); + act = ethr_atomic32_inc_read_acqb(&rwmtx->mtxb.flgs); if (act & (ETHR_RWMTX_W_FLG__|ETHR_RWMTX_W_WAIT_FLG__)) rwmutex_normal_rlock_wait(rwmtx, act); #else - long exp = 0; + ethr_sint32_t exp = 0; while (1) { - act = ethr_atomic_cmpxchg_acqb(&rwmtx->mtxb.flgs, exp+1, exp); + act = ethr_atomic32_cmpxchg_acqb(&rwmtx->mtxb.flgs, exp+1, exp); if (act == exp) break; @@ -2469,7 +2474,7 @@ ethr_rwmutex_rlock(ethr_rwmutex *rwmtx) void ethr_rwmutex_runlock(ethr_rwmutex *rwmtx) { - long act; + ethr_sint32_t act; ETHR_MTX_CHK_EXCL_IS_NOT_EXCL(&rwmtx->mtxb); ETHR_MTX_CHK_EXCL_UNSET_NON_EXCL(&rwmtx->mtxb); @@ -2484,7 +2489,7 @@ ethr_rwmutex_runlock(ethr_rwmutex *rwmtx) switch (rwmtx->type) { case ETHR_RWMUTEX_TYPE_NORMAL: - act = ethr_atomic_dec_read_relb(&rwmtx->mtxb.flgs); + act = ethr_atomic32_dec_read_relb(&rwmtx->mtxb.flgs); if ((act & ETHR_RWMTX_WAIT_FLGS__) && (act & ~ETHR_RWMTX_WAIT_FLGS__) == 0) { ETHR_ASSERT((act & ETHR_RWMTX_W_FLG__) == 0); @@ -2503,7 +2508,7 @@ ethr_rwmutex_runlock(ethr_rwmutex *rwmtx) ETHR_MEMORY_BARRIER; if (act == 0) { - act = ethr_atomic_read(&rwmtx->mtxb.flgs); + act = ethr_atomic32_read(&rwmtx->mtxb.flgs); if (act != ETHR_RWMTX_R_FLG__) rwmutex_freqread_rdrs_dec_chk_wakeup(rwmtx, tse, act); } @@ -2521,7 +2526,7 @@ int ethr_rwmutex_tryrwlock(ethr_rwmutex *rwmtx) { int res = 0; - long act; + ethr_sint32_t act; ETHR_ASSERT(!ethr_not_inited__); ETHR_ASSERT(rwmtx); @@ -2533,8 +2538,8 @@ ethr_rwmutex_tryrwlock(ethr_rwmutex *rwmtx) switch (rwmtx->type) { case ETHR_RWMUTEX_TYPE_NORMAL: - act = ethr_atomic_cmpxchg_acqb(&rwmtx->mtxb.flgs, - ETHR_RWMTX_W_FLG__, 0); + act = ethr_atomic32_cmpxchg_acqb(&rwmtx->mtxb.flgs, + ETHR_RWMTX_W_FLG__, 0); if (act != 0) res = EBUSY; break; @@ -2543,13 +2548,13 @@ ethr_rwmutex_tryrwlock(ethr_rwmutex *rwmtx) case ETHR_RWMUTEX_TYPE_EXTREMELY_FREQUENT_READ: res = 0; - act = ethr_atomic_read(&rwmtx->mtxb.flgs); + act = ethr_atomic32_read(&rwmtx->mtxb.flgs); do { if (act == 0) - act = ethr_atomic_cmpxchg_acqb(&rwmtx->mtxb.flgs, - ETHR_RWMTX_W_FLG__, 0); + act = ethr_atomic32_cmpxchg_acqb(&rwmtx->mtxb.flgs, + ETHR_RWMTX_W_FLG__, 0); else if (act == ETHR_RWMTX_R_FLG__) { res = rwmutex_try_complete_runlock(rwmtx, act, NULL, 0, 1, 1); @@ -2582,7 +2587,7 @@ ethr_rwmutex_tryrwlock(ethr_rwmutex *rwmtx) void ethr_rwmutex_rwlock(ethr_rwmutex *rwmtx) { - long act; + ethr_sint32_t act; ETHR_ASSERT(!ethr_not_inited__); ETHR_ASSERT(rwmtx); ETHR_ASSERT(rwmtx->initialized == ETHR_RWMUTEX_INITIALIZED); @@ -2593,8 +2598,8 @@ ethr_rwmutex_rwlock(ethr_rwmutex *rwmtx) switch (rwmtx->type) { case ETHR_RWMUTEX_TYPE_NORMAL: - act = ethr_atomic_cmpxchg_acqb(&rwmtx->mtxb.flgs, - ETHR_RWMTX_W_FLG__, 0); + act = ethr_atomic32_cmpxchg_acqb(&rwmtx->mtxb.flgs, + ETHR_RWMTX_W_FLG__, 0); if (act != 0) rwmutex_normal_rwlock_wait(rwmtx, act); break; @@ -2602,7 +2607,7 @@ ethr_rwmutex_rwlock(ethr_rwmutex *rwmtx) case ETHR_RWMUTEX_TYPE_FREQUENT_READ: case ETHR_RWMUTEX_TYPE_EXTREMELY_FREQUENT_READ: - act = ethr_atomic_read(&rwmtx->mtxb.flgs); + act = ethr_atomic32_read(&rwmtx->mtxb.flgs); do { @@ -2611,8 +2616,8 @@ ethr_rwmutex_rwlock(ethr_rwmutex *rwmtx) break; } - act = ethr_atomic_cmpxchg_acqb(&rwmtx->mtxb.flgs, - ETHR_RWMTX_W_FLG__, 0); + act = ethr_atomic32_cmpxchg_acqb(&rwmtx->mtxb.flgs, + ETHR_RWMTX_W_FLG__, 0); } while (act != 0); @@ -2630,7 +2635,7 @@ ethr_rwmutex_rwlock(ethr_rwmutex *rwmtx) void ethr_rwmutex_rwunlock(ethr_rwmutex *rwmtx) { - long act; + ethr_sint32_t act; ETHR_ASSERT(!ethr_not_inited__); ETHR_ASSERT(rwmtx); ETHR_ASSERT(rwmtx->initialized == ETHR_RWMUTEX_INITIALIZED); @@ -2645,16 +2650,16 @@ ethr_rwmutex_rwunlock(ethr_rwmutex *rwmtx) switch (rwmtx->type) { case ETHR_RWMUTEX_TYPE_NORMAL: - act = ethr_atomic_cmpxchg_relb(&rwmtx->mtxb.flgs, - 0, ETHR_RWMTX_W_FLG__); + act = ethr_atomic32_cmpxchg_relb(&rwmtx->mtxb.flgs, + 0, ETHR_RWMTX_W_FLG__); if (act != ETHR_RWMTX_W_FLG__) rwmutex_unlock_wake(rwmtx, 1, act, 0); break; case ETHR_RWMUTEX_TYPE_FREQUENT_READ: case ETHR_RWMUTEX_TYPE_EXTREMELY_FREQUENT_READ: - act = ethr_atomic_cmpxchg_relb(&rwmtx->mtxb.flgs, 0, - ETHR_RWMTX_W_FLG__); + act = ethr_atomic32_cmpxchg_relb(&rwmtx->mtxb.flgs, 0, + ETHR_RWMTX_W_FLG__); if (act != ETHR_RWMTX_W_FLG__) rwmutex_unlock_wake(rwmtx, 1, act, 0); break; @@ -2779,7 +2784,7 @@ static void hard_debug_chk_q__(struct ethr_mutex_base_ *mtxb, int is_rwmtx) { int res; - long flgs = ethr_atomic_read(&mtxb->flgs); + ethr_sint32_t flgs = ethr_atomic32_read(&mtxb->flgs); ETHR_MTX_HARD_ASSERT(res == 0); @@ -2802,12 +2807,12 @@ hard_debug_chk_q__(struct ethr_mutex_base_ *mtxb, int is_rwmtx) tse = mtxb->q; do { - long type; + ethr_sint32_t type; ETHR_MTX_HARD_ASSERT(tse->next->prev == tse); ETHR_MTX_HARD_ASSERT(tse->prev->next == tse); - type = ethr_atomic_read(&tse->uaflgs); + type = ethr_atomic32_read(&tse->uaflgs); ETHR_MTX_HARD_ASSERT(type == tse->uflgs); switch (type) { case ETHR_RWMTX_W_WAIT_FLG__: diff --git a/erts/lib_src/pthread/ethr_event.c b/erts/lib_src/pthread/ethr_event.c index 6731c0eb46..9434d60d0a 100644 --- a/erts/lib_src/pthread/ethr_event.c +++ b/erts/lib_src/pthread/ethr_event.c @@ -24,6 +24,10 @@ #define ETHR_INLINE_FUNC_NAME_(X) X ## __ #define ETHR_EVENT_IMPL__ +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + #include "ethread.h" #if defined(ETHR_LINUX_FUTEX_IMPL__) @@ -37,7 +41,7 @@ int ethr_event_init(ethr_event *e) { - ethr_atomic_init(&e->futex, ETHR_EVENT_OFF__); + ethr_atomic32_init(&e->futex, ETHR_EVENT_OFF__); return 0; } @@ -52,7 +56,7 @@ wait__(ethr_event *e, int spincount) { unsigned sc = spincount; int res; - long val; + ethr_sint32_t val; int until_yield = ETHR_YIELD_AFTER_BUSY_LOOPS; if (spincount < 0) @@ -60,7 +64,7 @@ wait__(ethr_event *e, int spincount) while (1) { while (1) { - val = ethr_atomic_read(&e->futex); + val = ethr_atomic32_read(&e->futex); if (val == ETHR_EVENT_ON__) return 0; if (sc == 0) @@ -76,16 +80,18 @@ wait__(ethr_event *e, int spincount) } if (val != ETHR_EVENT_OFF_WAITER__) { - val = ethr_atomic_cmpxchg(&e->futex, - ETHR_EVENT_OFF_WAITER__, - ETHR_EVENT_OFF__); + val = ethr_atomic32_cmpxchg(&e->futex, + ETHR_EVENT_OFF_WAITER__, + ETHR_EVENT_OFF__); if (val == ETHR_EVENT_ON__) return 0; ETHR_ASSERT(val == ETHR_EVENT_OFF__); } - res = ETHR_FUTEX__(&e->futex, ETHR_FUTEX_WAIT__, ETHR_EVENT_OFF_WAITER__); + res = ETHR_FUTEX__(&e->futex, + ETHR_FUTEX_WAIT__, + ETHR_EVENT_OFF_WAITER__); if (res == EINTR) break; if (res != 0 && res != EWOULDBLOCK) @@ -102,7 +108,7 @@ int ethr_event_init(ethr_event *e) { int res; - ethr_atomic_init(&e->state, ETHR_EVENT_OFF__); + ethr_atomic32_init(&e->state, ETHR_EVENT_OFF__); res = pthread_mutex_init(&e->mtx, NULL); if (res != 0) return res; @@ -131,7 +137,7 @@ static ETHR_INLINE int wait__(ethr_event *e, int spincount) { int sc = spincount; - long val; + ethr_sint32_t val; int res, ulres; int until_yield = ETHR_YIELD_AFTER_BUSY_LOOPS; @@ -139,7 +145,7 @@ wait__(ethr_event *e, int spincount) ETHR_FATAL_ERROR__(EINVAL); while (1) { - val = ethr_atomic_read(&e->state); + val = ethr_atomic32_read(&e->state); if (val == ETHR_EVENT_ON__) return 0; if (sc == 0) @@ -155,9 +161,9 @@ wait__(ethr_event *e, int spincount) } if (val != ETHR_EVENT_OFF_WAITER__) { - val = ethr_atomic_cmpxchg(&e->state, - ETHR_EVENT_OFF_WAITER__, - ETHR_EVENT_OFF__); + val = ethr_atomic32_cmpxchg(&e->state, + ETHR_EVENT_OFF_WAITER__, + ETHR_EVENT_OFF__); if (val == ETHR_EVENT_ON__) return 0; ETHR_ASSERT(val == ETHR_EVENT_OFF__); @@ -172,7 +178,7 @@ wait__(ethr_event *e, int spincount) while (1) { - val = ethr_atomic_read(&e->state); + val = ethr_atomic32_read(&e->state); if (val == ETHR_EVENT_ON__) break; diff --git a/erts/lib_src/pthread/ethread.c b/erts/lib_src/pthread/ethread.c index ea1d9d43f0..f047104103 100644 --- a/erts/lib_src/pthread/ethread.c +++ b/erts/lib_src/pthread/ethread.c @@ -72,7 +72,7 @@ static void thr_exit_cleanup(void) /* Argument passed to thr_wrapper() */ typedef struct { - ethr_atomic_t result; + ethr_atomic32_t result; ethr_ts_event *tse; void *(*thr_func)(void *); void *arg; @@ -81,14 +81,14 @@ typedef struct { static void *thr_wrapper(void *vtwd) { - long result; + ethr_sint32_t result; void *res; ethr_thr_wrap_data__ *twd = (ethr_thr_wrap_data__ *) vtwd; void *(*thr_func)(void *) = twd->thr_func; void *arg = twd->arg; ethr_ts_event *tsep = NULL; - result = (long) ethr_make_ts_event__(&tsep); + result = (ethr_sint32_t) ethr_make_ts_event__(&tsep); if (result == 0) { tsep->iflgs |= ETHR_TS_EV_ETHREAD; @@ -99,7 +99,7 @@ static void *thr_wrapper(void *vtwd) tsep = twd->tse; /* We aren't allowed to follow twd after result has been set! */ - ethr_atomic_set(&twd->result, result); + ethr_atomic32_set(&twd->result, result); ethr_event_set(&tsep->event); @@ -191,7 +191,7 @@ ethr_thr_create(ethr_tid *tid, void * (*func)(void *), void *arg, } #endif - ethr_atomic_init(&twd.result, -1); + ethr_atomic32_init(&twd.result, (ethr_sint32_t) -1); twd.tse = ethr_get_ts_event(); twd.thr_func = func; twd.arg = arg; @@ -252,10 +252,10 @@ ethr_thr_create(ethr_tid *tid, void * (*func)(void *), void *arg, /* Wait for child to initialize... */ while (1) { - long result; + ethr_sint32_t result; ethr_event_reset(&twd.tse->event); - result = ethr_atomic_read(&twd.result); + result = ethr_atomic32_read(&twd.result); if (result == 0) break; @@ -349,32 +349,6 @@ ethr_leave_ts_event(ethr_ts_event *tsep) } /* - * Current time - */ - -int -ethr_time_now(ethr_timeval *time) -{ - int res; - struct timeval tv; -#if ETHR_XCHK - if (ethr_not_inited__) { - ETHR_ASSERT(0); - return EACCES; - } - if (!time) { - ETHR_ASSERT(0); - return EINVAL; - } -#endif - - res = gettimeofday(&tv, NULL); - time->tv_sec = (long) tv.tv_sec; - time->tv_nsec = ((long) tv.tv_usec)*1000; - return res; -} - -/* * Thread specific data */ diff --git a/erts/lib_src/win/ethr_event.c b/erts/lib_src/win/ethr_event.c index ddb4780ff1..68f093f49c 100644 --- a/erts/lib_src/win/ethr_event.c +++ b/erts/lib_src/win/ethr_event.c @@ -28,6 +28,9 @@ /* --- Windows implementation of thread events ------------------------------ */ +#pragma intrinsic(_InterlockedExchangeAdd) +#pragma intrinsic(_InterlockedCompareExchange) + int ethr_event_init(ethr_event *e) { @@ -72,10 +75,10 @@ wait(ethr_event *e, int spincount) while (1) { long on; while (1) { -#if ETHR_IMMED_ATOMIC_SET_GET_SAFE__ +#if ETHR_READ_AND_SET_WITHOUT_INTERLOCKED_OP__ state = e->state; #else - state = InterlockedExchangeAdd(&e->state, (LONG) 0); + state = _InterlockedExchangeAdd(&e->state, (LONG) 0); #endif if (state == ETHR_EVENT_ON__) return 0; diff --git a/erts/lib_src/win/ethread.c b/erts/lib_src/win/ethread.c index 69523edf94..789a360b11 100644 --- a/erts/lib_src/win/ethread.c +++ b/erts/lib_src/win/ethread.c @@ -49,7 +49,7 @@ /* Argument passed to thr_wrapper() */ typedef struct { ethr_tid *tid; - ethr_atomic_t result; + ethr_atomic32_t result; ethr_ts_event *tse; void *(*thr_func)(void *); void *arg; @@ -93,20 +93,20 @@ static void thr_exit_cleanup(ethr_tid *tid, void *res) static unsigned __stdcall thr_wrapper(LPVOID vtwd) { ethr_tid my_tid; - long result; + ethr_sint32_t result; void *res; ethr_thr_wrap_data__ *twd = (ethr_thr_wrap_data__ *) vtwd; void *(*thr_func)(void *) = twd->thr_func; void *arg = twd->arg; ethr_ts_event *tsep = NULL; - result = (long) ethr_make_ts_event__(&tsep); + result = (ethr_sint32_t) ethr_make_ts_event__(&tsep); if (result == 0) { tsep->iflgs |= ETHR_TS_EV_ETHREAD; my_tid = *twd->tid; if (!TlsSetValue(own_tid_key, (LPVOID) &my_tid)) { - result = (long) ethr_win_get_errno__(); + result = (ethr_sint32_t) ethr_win_get_errno__(); ethr_free_ts_event__(tsep); } else { @@ -118,7 +118,7 @@ static unsigned __stdcall thr_wrapper(LPVOID vtwd) tsep = twd->tse; /* We aren't allowed to follow twd after result has been set! */ - ethr_atomic_set(&twd->result, result); + ethr_atomic32_set(&twd->result, result); ethr_event_set(&tsep->event); @@ -128,28 +128,6 @@ static unsigned __stdcall thr_wrapper(LPVOID vtwd) return 0; } -#ifdef __GNUC__ -#define LL_LITERAL(X) X##LL -#else -#define LL_LITERAL(X) X##i64 -#endif - -#define EPOCH_JULIAN_DIFF LL_LITERAL(11644473600) - -static ETHR_INLINE void -get_curr_time(long *sec, long *nsec) -{ - SYSTEMTIME t; - FILETIME ft; - LONGLONG lft; - - GetSystemTime(&t); - SystemTimeToFileTime(&t, &ft); - memcpy(&lft, &ft, sizeof(lft)); - *nsec = ((long) (lft % LL_LITERAL(10000000)))*100; - *sec = (long) ((lft / LL_LITERAL(10000000)) - EPOCH_JULIAN_DIFF); -} - /* internal exports */ int @@ -320,7 +298,7 @@ ethr_thr_create(ethr_tid *tid, void * (*func)(void *), void *arg, ETHR_PAGE_ALIGN(ETHR_KW2B(suggested_stack_size)); } - ethr_atomic_init(&twd.result, -1); + ethr_atomic32_init(&twd.result, -1); twd.tid = tid; twd.thr_func = func; @@ -352,11 +330,11 @@ ethr_thr_create(ethr_tid *tid, void * (*func)(void *), void *arg, /* Wait for child to initialize... */ while (1) { - long result; + ethr_sint32_t result; int err; ethr_event_reset(&twd.tse->event); - result = ethr_atomic_read(&twd.result); + result = ethr_atomic32_read(&twd.result); if (result == 0) break; @@ -517,23 +495,6 @@ ethr_equal_tids(ethr_tid tid1, ethr_tid tid2) return tid1.id == tid2.id && tid1.id != ETHR_INVALID_TID_ID; } -int -ethr_time_now(ethr_timeval *time) -{ -#if ETHR_XCHK - if (ethr_not_inited__) { - ETHR_ASSERT(0); - return EACCES; - } - if (!time) { - ETHR_ASSERT(0); - return EINVAL; - } -#endif - get_curr_time(&time->tv_sec, &time->tv_nsec); - return 0; -} - /* * Thread specific data */ diff --git a/erts/test/ethread_SUITE.erl b/erts/test/ethread_SUITE.erl index 93e27fa8d3..69e5af802f 100644 --- a/erts/test/ethread_SUITE.erl +++ b/erts/test/ethread_SUITE.erl @@ -37,7 +37,6 @@ equal_tids/1, mutex/1, try_lock_mutex/1, - time_now/1, cond_wait/1, broadcast/1, detached_thread/1, @@ -55,7 +54,6 @@ tests() -> equal_tids, mutex, try_lock_mutex, - time_now, cond_wait, broadcast, detached_thread, @@ -104,17 +102,6 @@ try_lock_mutex(suite) -> try_lock_mutex(Config) -> run_case(Config, "try_lock_mutex", ""). -time_now(doc) -> - ["Tests ethr_time_now by comparing time values with Erlang."]; -time_now(suite) -> - []; -time_now(Config) -> - run_case(Config, "time_now", "", fun (P) -> - spawn_link(fun () -> - watchdog(P) - end) - end). - wd_dispatch(P) -> receive bye -> diff --git a/erts/test/ethread_SUITE_data/ethread_tests.c b/erts/test/ethread_SUITE_data/ethread_tests.c index 7fc71d8047..0b59ff5aa6 100644 --- a/erts/test/ethread_SUITE_data/ethread_tests.c +++ b/erts/test/ethread_SUITE_data/ethread_tests.c @@ -514,69 +514,6 @@ try_lock_mutex_test(void) } /* - * The time now test. - * - * Tests ethr_time_now by comparing time values with Erlang. - */ -#define TNT_MAX_TIME_DIFF 200000 -#define TNT_MAX_TIME_VALUES 52 - -static void -time_now_test(void) -{ - int scanf_res, time_now_res, i, no_values, max_abs_diff; - static ethr_timeval tv[TNT_MAX_TIME_VALUES]; - static int ms[TNT_MAX_TIME_VALUES]; - - i = 0; - do { - ASSERT(i < TNT_MAX_TIME_VALUES); - scanf_res = scanf("%d", &ms[i]); - time_now_res = ethr_time_now(&tv[i]); - ASSERT(scanf_res == 1); - ASSERT(time_now_res == 0); -#if 0 - print_line("Got %d; %ld:%ld", ms[i], tv[i].tv_sec, tv[i].tv_nsec); -#endif - i++; - } while (ms[i-1] >= 0); - - no_values = i-1; - - ASSERT(ms[0] == 0); - - print_line("TNT_MAX_TIME_DIFF = %d (us)", TNT_MAX_TIME_DIFF); - - max_abs_diff = 0; - - for (i = 1; i < no_values; i++) { - long diff; - long tn_us; - long e_us; - - tn_us = (tv[i].tv_sec - tv[0].tv_sec) * 1000000; - tn_us += (tv[i].tv_nsec - tv[0].tv_nsec)/1000; - - e_us = ms[i]*1000; - - diff = e_us - tn_us; - - print_line("Erlang time = %ld us; ethr_time_now = %ld us; diff %ld us", - e_us, tn_us, diff); - - if (max_abs_diff < abs((int) diff)) { - max_abs_diff = abs((int) diff); - } - - ASSERT(e_us - TNT_MAX_TIME_DIFF <= tn_us); - ASSERT(tn_us <= e_us + TNT_MAX_TIME_DIFF); - } - - print_line("Max absolute diff = %d us", max_abs_diff); - succeed("Max absolute diff = %d us", max_abs_diff); -} - -/* * The cond wait test case. * * Tests ethr_cond_wait with ethr_cond_signal and ethr_cond_broadcast. @@ -1538,8 +1475,6 @@ main(int argc, char *argv[]) mutex_test(); else if (strcmp(testcase, "try_lock_mutex") == 0) try_lock_mutex_test(); - else if (strcmp(testcase, "time_now") == 0) - time_now_test(); else if (strcmp(testcase, "cond_wait") == 0) cond_wait_test(); else if (strcmp(testcase, "broadcast") == 0) |