From 1b273b618002d65159453fdfb9520a9476e4423a Mon Sep 17 00:00:00 2001 From: Rickard Green Date: Mon, 16 Aug 2010 19:38:33 +0200 Subject: Make it possible to reread and update detected CPU information Calling erlang:system_info/1 with the new argument 'update_cpu_info' will make the runtime system reread and update the internally stored CPU information. For more information see the documentation of erlang:system_info(update_cpu_info). --- erts/doc/src/erlang.xml | 55 ++++++++- erts/emulator/beam/erl_bif_info.c | 24 +++- erts/emulator/beam/erl_process.c | 35 +++++- erts/emulator/beam/erl_process.h | 1 + erts/include/internal/erl_misc_utils.h | 4 +- erts/lib_src/common/erl_misc_utils.c | 212 +++++++++++++++++++++++---------- 6 files changed, 258 insertions(+), 73 deletions(-) (limited to 'erts') diff --git a/erts/doc/src/erlang.xml b/erts/doc/src/erlang.xml index 54a0a80536..cb3112acfe 100644 --- a/erts/doc/src/erlang.xml +++ b/erts/doc/src/erlang.xml @@ -5533,7 +5533,7 @@ true CpuTopology type to change.

- {cpu_topology, defined} + {cpu_topology, defined}

Returns the user defined CpuTopology. For more information see the documentation of @@ -5543,7 +5543,7 @@ true argument.

- {cpu_topology, detected} + {cpu_topology, detected}

Returns the automatically detected CpuTopology. The emulator currently only detects the CPU topology on some newer @@ -5727,11 +5727,34 @@ true information see the "How to interpret the Erlang crash dumps" chapter in the ERTS User's Guide.

 - logical_processors + logical_processors -

Returns the number of logical processors detected on the - system as an integer or the atom unknown if the - emulator wasn't able to detect any. +

Returns the detected number of logical processors configured + on the system. The return value is either an integer, or + the atom unknown if the emulator wasn't able to + detect logical processors configured. +

+ + logical_processors_available + +

Returns the detected number of logical processors available to + the Erlang runtime system. The return value is either an + integer, or the atom unknown if the emulator wasn't + able to detect logical processors available. The number + of logical processors available is less than or equal to + the number of logical + processors online. +

+ + logical_processors_online + +

Returns the detected number of logical processors online on + the system. The return value is either an integer, + or the atom unknown if the emulator wasn't able to + detect logical processors online. The number of logical + processors online is less than or equal to the number of + logical processors + configured.

machine @@ -5936,6 +5959,26 @@ true get_tcw in "Match Specifications in Erlang", ERTS User's Guide.

+ update_cpu_info + +

The runtime system rereads the CPU information available and + updates its internally stored information about the + detected CPU + topology and the amount of logical processors + configured, + online, and + available. + If the CPU information has changed since the last time it was read, + the atom changed is returned; otherwise, the atom + unchanged is returned. If the CPU information has changed + you probably want to + adjust the amount + of schedulers online. You typically want to have as + many schedulers online as + logical processors + available. +

+ version

Returns a string containing the version number of the diff --git a/erts/emulator/beam/erl_bif_info.c b/erts/emulator/beam/erl_bif_info.c index 5128776c47..d5ae3d2174 100644 --- a/erts/emulator/beam/erl_bif_info.c +++ b/erts/emulator/beam/erl_bif_info.c @@ -2316,6 +2316,15 @@ BIF_RETTYPE system_info_1(BIF_ALIST_1) } else if (ERTS_IS_ATOM_STR("cpu_topology", BIF_ARG_1)) { res = erts_get_cpu_topology_term(BIF_P, am_used); BIF_TRAP1(erts_format_cpu_topology_trap, BIF_P, res); + } else if (ERTS_IS_ATOM_STR("update_cpu_info", BIF_ARG_1)) { + if (erts_update_cpu_info()) { + ERTS_DECL_AM(changed); + BIF_RET(AM_changed); + } + else { + ERTS_DECL_AM(unchanged); + BIF_RET(AM_unchanged); + } #if defined(__GNUC__) && defined(HAVE_SOLARIS_SPARC_PERFMON) } else if (ERTS_IS_ATOM_STR("ultrasparc_read_tick1", BIF_ARG_1)) { register unsigned high asm("%l0"); @@ -2387,7 +2396,10 @@ BIF_RETTYPE system_info_1(BIF_ALIST_1) } /* Arguments that are unusual follow ... */ else if (ERTS_IS_ATOM_STR("logical_processors", BIF_ARG_1)) { - int no = erts_get_cpu_configured(erts_cpuinfo); + int no; + erts_smp_rwmtx_rlock(&erts_cpu_bind_rwmtx); + no = erts_get_cpu_configured(erts_cpuinfo); + erts_smp_rwmtx_runlock(&erts_cpu_bind_rwmtx); if (no > 0) BIF_RET(make_small((Uint) no)); else { @@ -2396,7 +2408,10 @@ BIF_RETTYPE system_info_1(BIF_ALIST_1) } } else if (ERTS_IS_ATOM_STR("logical_processors_online", BIF_ARG_1)) { - int no = erts_get_cpu_online(erts_cpuinfo); + int no; + erts_smp_rwmtx_rlock(&erts_cpu_bind_rwmtx); + no = erts_get_cpu_online(erts_cpuinfo); + erts_smp_rwmtx_runlock(&erts_cpu_bind_rwmtx); if (no > 0) BIF_RET(make_small((Uint) no)); else { @@ -2405,7 +2420,10 @@ BIF_RETTYPE system_info_1(BIF_ALIST_1) } } else if (ERTS_IS_ATOM_STR("logical_processors_available", BIF_ARG_1)) { - int no = erts_get_cpu_available(erts_cpuinfo); + int no; + erts_smp_rwmtx_rlock(&erts_cpu_bind_rwmtx); + no = erts_get_cpu_available(erts_cpuinfo); + erts_smp_rwmtx_runlock(&erts_cpu_bind_rwmtx); if (no > 0) BIF_RET(make_small((Uint) no)); else { diff --git a/erts/emulator/beam/erl_process.c b/erts/emulator/beam/erl_process.c index a644520442..457e283034 100644 --- a/erts/emulator/beam/erl_process.c +++ b/erts/emulator/beam/erl_process.c @@ -3962,7 +3962,7 @@ check_cpu_bind(ErtsSchedulerData *esdp) goto unbind; } } - else if (cpu_id < 0 && scheduler2cpu_map[esdp->no].bound_id >= 0) { + else if (cpu_id < 0) /* && scheduler2cpu_map[esdp->no].bound_id >= 0) */ { unbind: /* Get rid of old binding */ res = erts_unbind_from_cpu(erts_cpuinfo); @@ -5521,6 +5521,39 @@ late_cpu_bind_init(void) } } +int +erts_update_cpu_info(void) +{ + int changed; + erts_smp_rwmtx_rwlock(&erts_cpu_bind_rwmtx); + changed = erts_cpu_info_update(erts_cpuinfo); + if (changed) { + erts_cpu_topology_t *cpudata; + int cpudata_size; + erts_free(ERTS_ALC_T_CPUDATA, system_cpudata); + + system_cpudata_size = erts_get_cpu_topology_size(erts_cpuinfo); + system_cpudata = erts_alloc(ERTS_ALC_T_CPUDATA, + (sizeof(erts_cpu_topology_t) + * system_cpudata_size)); + + if (!erts_get_cpu_topology(erts_cpuinfo, system_cpudata) + || ERTS_INIT_CPU_TOPOLOGY_OK != verify_topology(system_cpudata, + system_cpudata_size)) { + erts_free(ERTS_ALC_T_CPUDATA, system_cpudata); + system_cpudata = NULL; + system_cpudata_size = 0; + } + + create_tmp_cpu_topology_copy(&cpudata, &cpudata_size); + ASSERT(cpudata); + signal_schedulers_bind_change(cpudata, cpudata_size); + destroy_tmp_cpu_topology_copy(cpudata); + } + erts_smp_rwmtx_rwunlock(&erts_cpu_bind_rwmtx); + return changed; +} + #ifdef ERTS_SMP static void diff --git a/erts/emulator/beam/erl_process.h b/erts/emulator/beam/erl_process.h index c3fef6d38e..0cc4a715ad 100644 --- a/erts/emulator/beam/erl_process.h +++ b/erts/emulator/beam/erl_process.h @@ -1025,6 +1025,7 @@ int erts_init_scheduler_bind_type(char *how); #define ERTS_INIT_CPU_TOPOLOGY_MISSING 9 int erts_init_cpu_topology(char *topology_str); +int erts_update_cpu_info(void); void erts_pre_init_process(void); void erts_late_init_process(void); diff --git a/erts/include/internal/erl_misc_utils.h b/erts/include/internal/erl_misc_utils.h index 6b875ff824..dda2cab5c6 100644 --- a/erts/include/internal/erl_misc_utils.h +++ b/erts/include/internal/erl_misc_utils.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 @@ -34,7 +34,7 @@ typedef struct { erts_cpu_info_t *erts_cpu_info_create(void); void erts_cpu_info_destroy(erts_cpu_info_t *cpuinfo); -void erts_cpu_info_update(erts_cpu_info_t *cpuinfo); +int erts_cpu_info_update(erts_cpu_info_t *cpuinfo); int erts_get_cpu_configured(erts_cpu_info_t *cpuinfo); int erts_get_cpu_online(erts_cpu_info_t *cpuinfo); int erts_get_cpu_available(erts_cpu_info_t *cpuinfo); diff --git a/erts/lib_src/common/erl_misc_utils.c b/erts/lib_src/common/erl_misc_utils.c index d2ef7140a5..6346fe361d 100644 --- a/erts/lib_src/common/erl_misc_utils.c +++ b/erts/lib_src/common/erl_misc_utils.c @@ -62,10 +62,10 @@ #if defined(HAVE_SCHED_xETAFFINITY) # include # define ERTS_HAVE_MISC_UTIL_AFFINITY_MASK__ -#define ERTS_MU_GET_PROC_AFFINITY__(CPUINFOP) \ +#define ERTS_MU_GET_PROC_AFFINITY__(CPUINFOP, CPUSET) \ (sched_getaffinity((CPUINFOP)->pid, \ sizeof(cpu_set_t), \ - &(CPUINFOP)->cpuset) != 0 ? -errno : 0) + (CPUSET)) != 0 ? -errno : 0) #define ERTS_MU_SET_THR_AFFINITY__(SETP) \ (sched_setaffinity(0, sizeof(cpu_set_t), (SETP)) != 0 ? -errno : 0) #elif defined(__WIN32__) @@ -99,6 +99,26 @@ static int read_topology(erts_cpu_info_t *cpuinfo); +#if defined(ERTS_HAVE_MISC_UTIL_AFFINITY_MASK__) +static int +cpu_sets_are_eq(cpu_set_t *x, cpu_set_t *y) +{ + int i; + for (i = 0; i < CPU_SETSIZE; i++) { + if (CPU_ISSET(i, x)) { + if (!CPU_ISSET(i, y)) + return 0; + } + else { + if (CPU_ISSET(i, y)) + return 0; + } + } + return 1; +} + +#endif + int erts_milli_sleep(long ms) { @@ -137,15 +157,15 @@ struct erts_cpu_info_t_ { #if defined(__WIN32__) static __forceinline int -get_proc_affinity(erts_cpu_info_t *cpuinfo) +get_proc_affinity(erts_cpu_info_t *cpuinfo, cpu_set_t *cpuset) { DWORD pamask, samask; if (GetProcessAffinityMask(GetCurrentProcess(), &pamask, &samask)) { - cpuinfo->cpuset = (cpu_set_t) pamask; + *cpuset = (cpu_set_t) pamask; return 0; } else { - cpuinfo->cpuset = (cpu_set_t) 0; + *cpuset = (cpu_set_t) 0; return -erts_get_last_win_errno(); } } @@ -179,6 +199,9 @@ erts_cpu_info_create(void) #endif cpuinfo->topology_size = 0; cpuinfo->topology = NULL; + cpuinfo->configured = -1; + cpuinfo->online = -1; + cpuinfo->available = -1; erts_cpu_info_update(cpuinfo); return cpuinfo; } @@ -203,34 +226,40 @@ erts_cpu_info_destroy(erts_cpu_info_t *cpuinfo) } } -void +int erts_cpu_info_update(erts_cpu_info_t *cpuinfo) { - cpuinfo->configured = 0; - cpuinfo->online = 0; - cpuinfo->available = 0; + int changed = 0; + int configured = 0; + int online = 0; + int available = 0; + erts_cpu_topology_t *old_topology; + int old_topology_size; +#if defined(ERTS_HAVE_MISC_UTIL_AFFINITY_MASK__) + cpu_set_t cpuset; +#endif #ifdef __WIN32__ { int i; SYSTEM_INFO sys_info; GetSystemInfo(&sys_info); - cpuinfo->configured = (int) sys_info.dwNumberOfProcessors; + configured = (int) sys_info.dwNumberOfProcessors; for (i = 0; i < sizeof(DWORD)*8; i++) if (sys_info.dwActiveProcessorMask & (((DWORD) 1) << i)) - cpuinfo->online++; + online++; } #elif !defined(NO_SYSCONF) && (defined(_SC_NPROCESSORS_CONF) \ || defined(_SC_NPROCESSORS_ONLN)) #ifdef _SC_NPROCESSORS_CONF - cpuinfo->configured = (int) sysconf(_SC_NPROCESSORS_CONF); - if (cpuinfo->configured < 0) - cpuinfo->configured = 0; + configured = (int) sysconf(_SC_NPROCESSORS_CONF); + if (configured < 0) + configured = 0; #endif #ifdef _SC_NPROCESSORS_ONLN - cpuinfo->online = (int) sysconf(_SC_NPROCESSORS_ONLN); - if (cpuinfo->online < 0) - cpuinfo->online = 0; + online = (int) sysconf(_SC_NPROCESSORS_ONLN); + if (online < 0) + online = 0; #endif #elif defined(HAVE_SYS_SYSCTL_H) && defined(CTL_HW) && (defined(HW_NCPU) \ || defined(HW_AVAILCPU)) @@ -242,71 +271,138 @@ erts_cpu_info_update(erts_cpu_info_t *cpuinfo) len = sizeof(int); mib[0] = CTL_HW; mib[1] = HW_NCPU; - if (sysctl(&mib[0], 2, &cpuinfo->configured, &len, NULL, 0) < 0) - cpuinfo->configured = 0; + if (sysctl(&mib[0], 2, &configured, &len, NULL, 0) < 0) + configured = 0; #endif #ifdef HW_AVAILCPU len = sizeof(int); mib[0] = CTL_HW; mib[1] = HW_AVAILCPU; - if (sysctl(&mib[0], 2, &cpuinfo->online, &len, NULL, 0) < 0) - cpuinfo->online = 0; + if (sysctl(&mib[0], 2, &online, &len, NULL, 0) < 0) + online = 0; #endif } #endif - if (cpuinfo->online > cpuinfo->configured) - cpuinfo->online = cpuinfo->configured; + if (online > configured) + online = configured; + + if (cpuinfo->configured != configured) + changed = 1; + if (cpuinfo->online != online) + changed = 1; #if defined(ERTS_HAVE_MISC_UTIL_AFFINITY_MASK__) - if (ERTS_MU_GET_PROC_AFFINITY__(cpuinfo) == 0) { - int i, c, cn, si; - c = cn = 0; - si = sizeof(cpuinfo->affinity_str_buf) - 1; - cpuinfo->affinity_str_buf[si] = '\0'; - for (i = 0; i < CPU_SETSIZE; i++) { - if (CPU_ISSET(i, &cpuinfo->cpuset)) { - c |= 1 << cn; - cpuinfo->available++; + if (ERTS_MU_GET_PROC_AFFINITY__(cpuinfo, &cpuset) == 0) { + if (!changed && !cpu_sets_are_eq(&cpuset, &cpuinfo->cpuset)) + changed = 1; + + if (!changed) + available = cpuinfo->available; + else { + int i, c, cn, si; + + memcpy((void *) &cpuinfo->cpuset, + (void *) &cpuset, + sizeof(cpu_set_t)); + + c = cn = 0; + si = sizeof(cpuinfo->affinity_str_buf) - 1; + cpuinfo->affinity_str_buf[si] = '\0'; + for (i = 0; i < CPU_SETSIZE; i++) { + if (CPU_ISSET(i, &cpuinfo->cpuset)) { + c |= 1 << cn; + available++; + } + cn++; + if (cn == 4) { + cpuinfo->affinity_str_buf[--si] = (c < 10 + ? '0' + c + : 'A' + c - 10); + c = cn = 0; + } } - cn++; - if (cn == 4) { + if (c) cpuinfo->affinity_str_buf[--si] = (c < 10 ? '0' + c : 'A' + c - 10); - c = cn = 0; - } + while (cpuinfo->affinity_str_buf[si] == '0') + si++; + cpuinfo->affinity_str = &cpuinfo->affinity_str_buf[si]; } - if (c) - cpuinfo->affinity_str_buf[--si] = (c < 10 - ? '0' + c - : 'A' + c - 10); - while (cpuinfo->affinity_str_buf[si] == '0') - si++; - cpuinfo->affinity_str = &cpuinfo->affinity_str_buf[si]; } #elif defined(HAVE_PSET_INFO) { - uint_t numcpus = cpuinfo->configured; - if (cpuinfo->cpuids) - free(cpuinfo->cpuids); - cpuinfo->cpuids = malloc(sizeof(processorid_t)*numcpus); - if (cpuinfo->cpuids) { - if (pset_info(PS_MYID, NULL, &numcpus, &cpuinfo->cpuids) == 0) - cpuinfo->available = (int) numcpus; - if (cpuinfo->available < 0) { - free(cpuinfo->cpuid); - cpuinfo->available = 0; + processorid_t *cpuids; + uint_t numcpus = configured; + cpuids = malloc(sizeof(processorid_t)*numcpus); + if (cpuids) { + if (pset_info(PS_MYID, NULL, &numcpus, &cpuids) == 0) + available = (int) numcpus; + if (available < 0) { + free(cpuids); + cpuids = NULL; + available = 0; } } + if (!cpuids) { + if (cpuinfo->cpuids) + changed = 1; + } + else { + if (cpuinfo->cpuids) + changed = 1; + if (memcmp((void *) cpuinfo->cpuids, + (void *) cpuids, + sizeof(processorid_t)*numcpus) != 0) + changed = 1; + + } + if (!changed) { + if (cpuids) + free(cpuids); + } + else { + if (cpuinfo->cpuids) + free(cpuinfo->cpuids); + cpuinfo->cpuids = cpuids; + } } #endif - if (cpuinfo->available > cpuinfo->online) - cpuinfo->available = cpuinfo->online; + if (available > online) + available = online; + + if (cpuinfo->available != available) + changed = 1; + + cpuinfo->configured = configured; + cpuinfo->online = online; + cpuinfo->available = available; + + old_topology = cpuinfo->topology; + old_topology_size = cpuinfo->topology_size; + cpuinfo->topology = NULL; read_topology(cpuinfo); + if (cpuinfo->topology_size != old_topology_size + || (old_topology_size != 0 + && memcmp((void *) cpuinfo->topology, + (void *) old_topology, + (sizeof(erts_cpu_topology_t) + * old_topology_size)) != 0)) { + changed = 1; + if (old_topology) + free(old_topology); + } + else { + if (cpuinfo->topology) + free(cpuinfo->topology); + cpuinfo->topology = old_topology; + } + + return changed; } int @@ -641,9 +737,6 @@ read_topology(erts_cpu_info_t *cpuinfo) errno = 0; - if (cpuinfo->topology) - free(cpuinfo->topology); - if (cpuinfo->configured < 1) goto error; @@ -896,9 +989,6 @@ read_topology(erts_cpu_info_t *cpuinfo) errno = 0; - if (cpuinfo->topology) - free(cpuinfo->topology); - if (cpuinfo->configured < 1) goto error; -- cgit v1.2.3 From f45560001d8128193ed4933ab9dca1b105ae752a Mon Sep 17 00:00:00 2001 From: Rickard Green Date: Tue, 17 Aug 2010 19:43:08 +0200 Subject: Implement automatic detection of CPU topology on Windows The CPU topology is now automatically detected on Windows systems with less than 33 logical processors. The runtime system will now, also on Windows, by default bind schedulers to logical processors using the 'default_bind' bind type if the amount of schedulers is at least equal to the amount of logical processors configured, binding of schedulers is supported, and a CPU topology is available at startup. --- erts/doc/src/erl.xml | 4 +- erts/doc/src/erlang.xml | 12 +- erts/include/internal/erl_misc_utils.h | 1 + erts/lib_src/common/erl_misc_utils.c | 381 +++++++++++++++++++++++++++++---- 4 files changed, 345 insertions(+), 53 deletions(-) (limited to 'erts') diff --git a/erts/doc/src/erl.xml b/erts/doc/src/erl.xml index 0e26d62548..f477280a6f 100644 --- a/erts/doc/src/erl.xml +++ b/erts/doc/src/erl.xml @@ -685,8 +685,8 @@ erlang:system_flag(scheduler_bind_type, default_bind).

-

Binding of schedulers are currently only supported on newer - Linux and Solaris systems.

+

Binding of schedulers is currently only supported on newer + Linux, Solaris, and Windows systems.

If no CPU topology is available when the +sbt flag is processed and BindType is any other type than u, the runtime system will fail to start. CPU diff --git a/erts/doc/src/erlang.xml b/erts/doc/src/erlang.xml index cb3112acfe..a17a1d0ead 100644 --- a/erts/doc/src/erlang.xml +++ b/erts/doc/src/erlang.xml @@ -5175,9 +5175,9 @@ true schedulers actually have bound as requested, call erlang:system_info(scheduler_bindings).

-

Schedulers can currently only be bound on newer Linux - and Solaris systems, but more systems will be supported - in the future. +

Schedulers can currently only be bound on newer Linux, + Solaris, and Windows systems, but more systems will be + supported in the future.

In order for the runtime system to be able to bind schedulers, the CPU topology needs to be known. If the runtime system fails @@ -5547,8 +5547,10 @@ true

Returns the automatically detected CpuTopology. The emulator currently only detects the CPU topology on some newer - linux and solaris systems. For more information see the - documentation of the + Linux, Solaris, and Windows systems. On Windows system with + more than 32 logical processors the CPU topology is not detected. +

+

For more information see the documentation of the cpu_topology argument.

diff --git a/erts/include/internal/erl_misc_utils.h b/erts/include/internal/erl_misc_utils.h index dda2cab5c6..507e1726f4 100644 --- a/erts/include/internal/erl_misc_utils.h +++ b/erts/include/internal/erl_misc_utils.h @@ -51,6 +51,7 @@ int erts_unbind_from_cpu_str(char *str); int erts_milli_sleep(long); #ifdef __WIN32__ +int erts_map_win_error_to_errno(DWORD win_error); int erts_get_last_win_errno(void); #endif diff --git a/erts/lib_src/common/erl_misc_utils.c b/erts/lib_src/common/erl_misc_utils.c index 6346fe361d..116c9886d8 100644 --- a/erts/lib_src/common/erl_misc_utils.c +++ b/erts/lib_src/common/erl_misc_utils.c @@ -21,10 +21,13 @@ #include "config.h" #endif +#if defined(__WIN32__) +# include +#endif + #include "erl_misc_utils.h" #if defined(__WIN32__) -# include #elif defined(VXWORKS) # include #else /* UNIX */ @@ -684,6 +687,56 @@ cpu_cmp(const void *vx, const void *vy) return 0; } +static void +adjust_processor_nodes(erts_cpu_info_t *cpuinfo, int no_nodes) +{ + erts_cpu_topology_t *prev, *this, *last; + if (no_nodes > 1) { + int processor = -1; + int processor_node = 0; + int node = -1; + + qsort(cpuinfo->topology, + cpuinfo->topology_size, + sizeof(erts_cpu_topology_t), + pn_cmp); + + prev = NULL; + this = &cpuinfo->topology[0]; + last = &cpuinfo->topology[cpuinfo->configured-1]; + while (1) { + if (processor == this->processor) { + if (node != this->node) + processor_node = 1; + } + else { + if (processor_node) { + make_processor_node: + while (prev->processor == processor) { + prev->processor_node = prev->node; + prev->node = -1; + if (prev == &cpuinfo->topology[0]) + break; + prev--; + } + processor_node = 0; + } + processor = this->processor; + node = this->node; + } + if (this == last) { + if (processor_node) { + prev = this; + goto make_processor_node; + } + break; + } + prev = this++; + } + } +} + + #ifdef __linux__ static int @@ -850,49 +903,7 @@ read_topology(erts_cpu_info_t *cpuinfo) cpuinfo->topology = t; } - if (no_nodes > 1) { - int processor = -1; - int processor_node = 0; - int node = -1; - - qsort(cpuinfo->topology, - cpuinfo->topology_size, - sizeof(erts_cpu_topology_t), - pn_cmp); - - prev = NULL; - this = &cpuinfo->topology[0]; - last = &cpuinfo->topology[cpuinfo->configured-1]; - while (1) { - if (processor == this->processor) { - if (node != this->node) - processor_node = 1; - } - else { - if (processor_node) { - make_processor_node: - while (prev->processor == processor) { - prev->processor_node = prev->node; - prev->node = -1; - if (prev == &cpuinfo->topology[0]) - break; - prev--; - } - processor_node = 0; - } - processor = this->processor; - node = this->node; - } - if (this == last) { - if (processor_node) { - prev = this; - goto make_processor_node; - } - break; - } - prev = this++; - } - } + adjust_processor_nodes(cpuinfo, no_nodes); qsort(cpuinfo->topology, cpuinfo->topology_size, @@ -1075,6 +1086,8 @@ read_topology(erts_cpu_info_t *cpuinfo) } } + adjust_processor_nodes(cpuinfo, 1); + error: if (res == 0) { @@ -1093,6 +1106,275 @@ read_topology(erts_cpu_info_t *cpuinfo) } +#elif defined(__WIN32__) + +/* + * We cannot use Relation* out of the box since all of them are not + * always part of the LOGICAL_PROCESSOR_RELATIONSHIP enum. They are + * however documented as follows... + */ +#define ERTS_MU_RELATION_PROCESSOR_CORE 0 /* RelationProcessorCore */ +#define ERTS_MU_RELATION_NUMA_NODE 1 /* RelationNumaNode */ +#define ERTS_MU_RELATION_CACHE 2 /* RelationCache */ +#define ERTS_MU_RELATION_PROCESSOR_PACKAGE 3 /* RelationProcessorPackage */ + +static __forceinline int +rel_cmp_val(int r) +{ + switch (r) { + case ERTS_MU_RELATION_NUMA_NODE: return 0; + case ERTS_MU_RELATION_PROCESSOR_PACKAGE: return 1; + case ERTS_MU_RELATION_PROCESSOR_CORE: return 2; + default: /* currently not used */ return 3; + } +} + +static int +slpi_cmp(const void *vx, const void *vy) +{ + PSYSTEM_LOGICAL_PROCESSOR_INFORMATION x, y; + x = (PSYSTEM_LOGICAL_PROCESSOR_INFORMATION) vx; + y = (PSYSTEM_LOGICAL_PROCESSOR_INFORMATION) vy; + + if ((int) x->Relationship != (int) y->Relationship) + return (rel_cmp_val((int) x->Relationship) + - rel_cmp_val((int) y->Relationship)); + + switch ((int) x->Relationship) { + case ERTS_MU_RELATION_NUMA_NODE: + if (x->NumaNode.NodeNumber == y->NumaNode.NodeNumber) + break; + return ((int) x->NumaNode.NodeNumber) - ((int) y->NumaNode.NodeNumber); + case ERTS_MU_RELATION_PROCESSOR_CORE: + case ERTS_MU_RELATION_PROCESSOR_PACKAGE: + default: + break; + } + + if (x->ProcessorMask == y->ProcessorMask) + return 0; + return x->ProcessorMask < y->ProcessorMask ? -1 : 1; +} + +typedef BOOL (WINAPI *glpi_t)(PSYSTEM_LOGICAL_PROCESSOR_INFORMATION, PDWORD); + +static int +read_topology(erts_cpu_info_t *cpuinfo) +{ + int res = 0; + glpi_t glpi; + int *core_id = NULL; + PSYSTEM_LOGICAL_PROCESSOR_INFORMATION slpip = NULL; + int wix, rix, max_l, l, packages, nodes, no_slpi; + DWORD slpi_size = 0; + + + glpi = (glpi_t) GetProcAddress(GetModuleHandle("kernel32"), + "GetLogicalProcessorInformation"); + if (!glpi) + return -ENOTSUP; + + cpuinfo->topology = NULL; + + if (cpuinfo->configured < 1 || sizeof(ULONG_PTR)*8 < cpuinfo->configured) + goto error; + + while (1) { + DWORD werr; + if (TRUE == glpi(slpip, &slpi_size)) + break; + werr = GetLastError(); + if (werr != ERROR_INSUFFICIENT_BUFFER) { + res = -erts_map_win_error_to_errno(werr); + goto error; + } + if (slpip) + free(slpip); + slpip = malloc(slpi_size); + if (!slpip) { + res = -ENOMEM; + goto error; + } + } + + no_slpi = (int) slpi_size/sizeof(SYSTEM_LOGICAL_PROCESSOR_INFORMATION); + + qsort(slpip, + no_slpi, + sizeof(SYSTEM_LOGICAL_PROCESSOR_INFORMATION), + slpi_cmp); + + /* + * Now numa node relations appear before package relations which + * appear before core relations which appear before relations + * we aren't interested in... + */ + + max_l = 0; + packages = 0; + nodes = 0; + for (rix = 0; rix < no_slpi; rix++) { + PSYSTEM_LOGICAL_PROCESSOR_INFORMATION this = &slpip[rix]; + for (l = sizeof(ULONG_PTR)*8 - 1; l > 0; l--) { + if (slpip[rix].ProcessorMask & (((ULONG_PTR) 1) << l)) { + if (max_l < l) + max_l = l; + break; + } + } + if ((int) slpip[rix].Relationship == ERTS_MU_RELATION_PROCESSOR_PACKAGE) + packages++; + if ((int) slpip[rix].Relationship == ERTS_MU_RELATION_NUMA_NODE) + nodes++; + } + + core_id = malloc(sizeof(int)*(packages ? packages : 1)); + if (!core_id) { + res = -ENOMEM; + goto error; + } + + for (rix = 0; rix < packages; rix++) + core_id[rix] = 0; + + cpuinfo->topology_size = max_l + 1; + cpuinfo->topology = malloc(sizeof(erts_cpu_topology_t) + * cpuinfo->topology_size); + if (!cpuinfo->topology) { + res = -ENOMEM; + goto error; + } + + for (wix = 0; wix < cpuinfo->topology_size; wix++) { + cpuinfo->topology[wix].node = -1; + cpuinfo->topology[wix].processor = -1; + cpuinfo->topology[wix].processor_node = -1; + cpuinfo->topology[wix].core = -1; + cpuinfo->topology[wix].thread = -1; + cpuinfo->topology[wix].logical = -1; + } + + nodes = 0; + packages = 0; + + for (rix = 0; rix < no_slpi; rix++) { + + switch ((int) slpip[rix].Relationship) { + case ERTS_MU_RELATION_NUMA_NODE: + for (l = 0; l < sizeof(ULONG_PTR)*8; l++) { + if (slpip[rix].ProcessorMask & (((ULONG_PTR) 1) << l)) { + cpuinfo->topology[l].logical = l; + cpuinfo->topology[l].node = slpip[rix].NumaNode.NodeNumber; + } + } + nodes++; + break; + case ERTS_MU_RELATION_PROCESSOR_PACKAGE: + for (l = 0; l < sizeof(ULONG_PTR)*8; l++) { + if (slpip[rix].ProcessorMask & (((ULONG_PTR) 1) << l)) { + cpuinfo->topology[l].logical = l; + cpuinfo->topology[l].processor = packages; + } + } + packages++; + break; + case ERTS_MU_RELATION_PROCESSOR_CORE: { + int thread = 0; + int processor = -1; + for (l = 0; l < sizeof(ULONG_PTR)*8; l++) { + /* + * Nodes and packages may not be supported; pretend + * that there are one if this is the case... + */ + if (!nodes) + cpuinfo->topology[l].node = 0; + if (!packages) + cpuinfo->topology[l].processor = 0; + if (slpip[rix].ProcessorMask & (((ULONG_PTR) 1) << l)) { + if (processor < 0) { + processor = cpuinfo->topology[l].processor; + if (processor < 0) { + res = -EINVAL; + goto error; + } + } + else if (processor != cpuinfo->topology[l].processor) { + res = -EINVAL; + goto error; + } + cpuinfo->topology[l].logical = l; + cpuinfo->topology[l].thread = thread; + cpuinfo->topology[l].core = core_id[processor]; + thread++; + } + } + core_id[processor]++; + break; + } + default: + /* + * We have reached the end of the relationships + * that we (currently) are interested in... + */ + goto relationships_done; + } + } + + relationships_done: + + /* + * There may be unused entries; remove them... + */ + for (rix = wix = 0; rix < cpuinfo->topology_size; rix++) { + if (cpuinfo->topology[rix].logical >= 0) { + if (wix != rix) + cpuinfo->topology[wix] = cpuinfo->topology[rix]; + wix++; + } + } + + if (cpuinfo->topology_size != wix) { + erts_cpu_topology_t *new = cpuinfo->topology; + new = realloc(cpuinfo->topology, + sizeof(erts_cpu_topology_t)*wix); + if (!new) { + res = -ENOMEM; + goto error; + } + cpuinfo->topology = new; + cpuinfo->topology_size = wix; + } + + res = wix; + + adjust_processor_nodes(cpuinfo, nodes); + + qsort(cpuinfo->topology, + cpuinfo->topology_size, + sizeof(erts_cpu_topology_t), + cpu_cmp); + + if (res < cpuinfo->online) + res = -EINVAL; + + error: + + if (res <= 0) { + cpuinfo->topology_size = 0; + if (cpuinfo->topology) { + free(cpuinfo->topology); + cpuinfo->topology = NULL; + } + } + + if (slpip) + free(slpip); + if (core_id) + free(core_id); + + return res; +} + #else static int @@ -1106,9 +1388,9 @@ read_topology(erts_cpu_info_t *cpuinfo) #if defined(__WIN32__) int -erts_get_last_win_errno(void) +erts_map_win_error_to_errno(DWORD win_error) { - switch (GetLastError()) { + switch (win_error) { case ERROR_INVALID_FUNCTION: return EINVAL; /* 1 */ case ERROR_FILE_NOT_FOUND: return ENOENT; /* 2 */ case ERROR_PATH_NOT_FOUND: return ENOENT; /* 3 */ @@ -1189,4 +1471,11 @@ erts_get_last_win_errno(void) } } +int +erts_get_last_win_errno(void) +{ + return erts_map_win_error_to_errno(GetLastError()); +} + + #endif -- cgit v1.2.3 From 026940a3cf528a115dcc5e2bfb89f0ce78907031 Mon Sep 17 00:00:00 2001 From: Rickard Green Date: Tue, 17 Aug 2010 23:05:05 +0200 Subject: Fix faulty assertions --- erts/lib_src/common/ethr_mutex.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) (limited to 'erts') diff --git a/erts/lib_src/common/ethr_mutex.c b/erts/lib_src/common/ethr_mutex.c index f918bba81d..aac0d44a32 100644 --- a/erts/lib_src/common/ethr_mutex.c +++ b/erts/lib_src/common/ethr_mutex.c @@ -1048,7 +1048,7 @@ ethr_cond_signal(ethr_cond *cnd) ethr_ts_event *tse; ETHR_ASSERT(!ethr_not_inited__); - ETHR_ASSERT(!cnd); + ETHR_ASSERT(cnd); ETHR_ASSERT(cnd->initialized == ETHR_COND_INITIALIZED); ETHR_MTX_HARD_DEBUG_FENCE_CHK(cnd); @@ -1089,7 +1089,7 @@ ethr_cond_broadcast(ethr_cond *cnd) int got_all; ethr_ts_event *tse; ETHR_ASSERT(!ethr_not_inited__); - ETHR_ASSERT(!cnd); + ETHR_ASSERT(cnd); ETHR_ASSERT(cnd->initialized == ETHR_COND_INITIALIZED); ETHR_MTX_HARD_DEBUG_FENCE_CHK(cnd); @@ -1158,9 +1158,9 @@ ethr_cond_wait(ethr_cond *cnd, ethr_mutex *mtx) ethr_ts_event *tse; ETHR_ASSERT(!ethr_not_inited__); - ETHR_ASSERT(!cnd); + ETHR_ASSERT(cnd); ETHR_ASSERT(cnd->initialized == ETHR_COND_INITIALIZED); - ETHR_ASSERT(!mtx); + ETHR_ASSERT(mtx); ETHR_ASSERT(mtx->initialized == ETHR_MUTEX_INITIALIZED); tse = ethr_get_ts_event(); -- cgit v1.2.3 From d389ff6f96cbd855e95a7455407b6140f1601d0f Mon Sep 17 00:00:00 2001 From: Rickard Green Date: Tue, 17 Aug 2010 23:05:43 +0200 Subject: Initialize environment functionality after thread lib An assertion failed due to the thread library not being initialized when initializing an rwmutex. This was however harmless. --- erts/emulator/sys/win32/sys.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) (limited to 'erts') diff --git a/erts/emulator/sys/win32/sys.c b/erts/emulator/sys/win32/sys.c index 54f71b202d..f37b7bbbe5 100644 --- a/erts/emulator/sys/win32/sys.c +++ b/erts/emulator/sys/win32/sys.c @@ -3002,7 +3002,6 @@ erts_sys_pre_init(void) } #endif erts_smp_atomic_init(&sys_misc_mem_sz, 0); - erts_sys_env_init(); } void noinherit_std_handle(DWORD type) @@ -3018,6 +3017,8 @@ void erl_sys_init(void) { HANDLE handle; + erts_sys_env_init(); + noinherit_std_handle(STD_OUTPUT_HANDLE); noinherit_std_handle(STD_INPUT_HANDLE); noinherit_std_handle(STD_ERROR_HANDLE); -- cgit v1.2.3