sera/0ad
1
0
Fork 0
forked from 0ad/0ad
Code Pull Requests Activity

proper fix for brain-dead Linux affinity API (that code was disabled in #547)

Browse Source 
refs #985

This was SVN commit r10373.
This commit is contained in:
janwas 2011-10-08 16:42:30 +00:00
parent 90cac1b7b0
commit 39e5cb5dc9
1 changed files with 45 additions and 20 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

65
source/lib/sysdep/os/linux/lcpu.cpp
View File

@ -1,4 +1,4 @@
/* Copyright (c) 2010 Wildfire Games
/* Copyright (c) 2011 Wildfire Games
*
* Permission is hereby granted, free of charge, to any person obtaining
* a copy of this software and associated documentation files (the
@ -25,6 +25,7 @@
#include "lib/sysdep/os_cpu.h"
#include "lib/alignment.h"
#include "lib/bits.h"
#include "lib/module_init.h"
#if OS_LINUX
#include "valgrind.h"
@ -98,47 +99,71 @@ size_t os_cpu_MemoryAvailable()
}
// glibc __CPU_SETSIZE=1024 is smaller than required on some Linux (4096),
// but the CONFIG_NR_CPUS in a header may not reflect the actual kernel,
// so we have to detect the limit at runtime.
// (see http://trac.wildfiregames.com/ticket/547 for additional information)
static size_t maxCpus;
static bool IsMaxCpusSufficient()
{
const size_t setSize = CPU_ALLOC_SIZE(maxCpus);
cpu_set_t* set = CPU_ALLOC(maxCpus);
const int ret = sched_getaffinity(0, setSize, set);
CPU_FREE(set);
if(ret == 0)
return true;
ENSURE(errno == EINVAL);
return false;
}
static Status DetectMaxCpus()
{
// the most I have ever heard of is CONFIG_NR_CPUS=4096,
// and even that limit should be enough for years and years.
for(maxCpus = 64; maxCpus <= 65536; maxCpus *= 2)
{
if(IsMaxCpusSufficient())
return INFO::OK;
}
return ERR::FAIL;
}
uintptr_t os_cpu_SetThreadAffinityMask(uintptr_t processorMask)
{
// This code is broken on kernels with CONFIG_NR_CPUS >= 1024, since cpu_set_t
// is too small by default. See <http://trac.wildfiregames.com/ticket/547>.
// It seems the most reliable solution is to use dynamically-allocated sets
// (CPU_SET_S etc), and do it in a loop until we've allocated a set large
// enough to fit all the CPUs.
// That's a pain and we currently don't actually need this code at all, and on OS X
// we don't implement this function anyway, so just disable it here.
#if 0
int ret;
cpu_set_t set;
static ModuleInitState maxCpusInitState;
(void)ModuleInit(&maxCpusInitState, DetectMaxCpus);
const size_t setSize = CPU_ALLOC_SIZE(maxCpus);
cpu_set_t* set = CPU_ALLOC(maxCpus);
ENSURE(set);
uintptr_t previousProcessorMask = 0;
{
ret = sched_getaffinity(0, sizeof(set), &set);
int ret = sched_getaffinity(0, setSize, set);
ENSURE(ret == 0);
for(size_t processor = 0; processor < os_cpu_NumProcessors(); processor++)
{
if(CPU_ISSET(processor, &set))
if(CPU_ISSET_S(processor, setSize, set))
previousProcessorMask |= uintptr_t(1) << processor;
}
}
CPU_ZERO(&set);
CPU_ZERO_S(setSize, set);
for(size_t processor = 0; processor < os_cpu_NumProcessors(); processor++)
{
if(IsBitSet(processorMask, processor))
CPU_SET(processor, &set);
CPU_SET_S(processor, setSize, set);
}
ret = sched_setaffinity(0, sizeof(set), &set);
int ret = sched_setaffinity(0, setSize, set);
ENSURE(ret == 0);
// (The process gets migrated immediately by the setaffinity call)
CPU_FREE(set);
return previousProcessorMask;
#endif
UNUSED2(processorMask);
return os_cpu_ProcessorMask();
}