Re: [PATCH V3 0/2] Fix perf bench numa to work with machines having #CPUs > 1K
Em Tue, Apr 12, 2022 at 10:10:57PM +0530, Athira Rajeev escreveu: > The perf benchmark for collections: numa hits failure in system > configuration with CPU's more than 1024. These benchmarks uses > "sched_getaffinity" and "sched_setaffinity" in the code to > work with affinity. Thanks, applied. - Arnaldo > Example snippet from numa benchmark: > <<>> > perf: bench/numa.c:302: bind_to_node: Assertion `!(ret)' failed. > Aborted (core dumped) > <<>> > > bind_to_node function uses "sched_getaffinity" to save the cpumask. > This fails with EINVAL because the default mask size in glibc is 1024 > > To overcome this 1024 CPUs mask size limitation of cpu_set_t, > change the mask size using the CPU_*_S macros ie, use CPU_ALLOC to > allocate cpumask, CPU_ALLOC_SIZE for size, CPU_SET_S to set mask bit. > > Fix all the relevant places in the code to use mask size which is large > enough to represent number of possible CPU's in the system. > > This patchset also address a fix for parse_setup_cpu_list function in > numa bench to check if input CPU is online before binding task to > that CPU. This is to fix failures where, though CPU number is within > max CPU, it could happen that CPU is offline. Here, sched_setaffinity > will result in failure when using cpumask having that cpu bit set > in the mask. > > Patch 1 address fix for parse_setup_cpu_list to check if CPU used to bind > task is online. Patch 2 has fix for bench numa to work with machines > having #CPUs > 1K > > Athira Rajeev (2): > tools/perf: Fix perf bench numa testcase to check if CPU used to bind > task is online > perf bench: Fix numa bench to fix usage of affinity for machines with > #CPUs > 1K > > Changelog: > v2 -> v3 > Link to the v2 version : > https://lore.kernel.org/all/20220406175113.87881-1-atraj...@linux.vnet.ibm.com/ > - From the v2 version, patch 1 and patch 2 are now part of upstream. > - This v3 version separates patch 3 and patch 4 to address review >comments from arnaldo which includes using sysfs__read_str for reading >sysfs file and fixing the compilation issues observed in debian > > tools/perf/bench/numa.c | 136 +-- > tools/perf/util/header.c | 51 +++ > tools/perf/util/header.h | 1 + > 3 files changed, 153 insertions(+), 35 deletions(-) > > -- > 2.35.1 -- - Arnaldo
Re: [PATCH V3 0/2] Fix perf bench numa to work with machines having #CPUs > 1K
-Original Message- From: Athira Rajeev To: a...@kernel.org, jo...@kernel.org, disg...@linux.vnet.ibm.com Cc: m...@ellerman.id.au, linux-perf-us...@vger.kernel.org, linuxppc-dev@lists.ozlabs.org, ma...@linux.vnet.ibm.com, rnsas...@linux.ibm.com, kj...@linux.ibm.com, linux-ker...@vger.kernel.org, sri...@linux.vnet.ibm.com, irog...@google.com Subject: [PATCH V3 0/2] Fix perf bench numa to work with machines having #CPUs > 1K Date: Tue, 12 Apr 2022 22:10:57 +0530 The perf benchmark for collections: numa hits failure in systemconfiguration with CPU's more than 1024. These benchmarks uses"sched_getaffinity" and "sched_setaffinity" in the code towork with affinity. Example snippet from numa benchmark:<<>>perf: bench/numa.c:302: bind_to_node: Assertion `!(ret)' failed.Aborted (core dumped)<<>> bind_to_node function uses "sched_getaffinity" to save the cpumask.This fails with EINVAL because the default mask size in glibc is 1024 To overcome this 1024 CPUs mask size limitation of cpu_set_t,change the mask size using the CPU_*_S macros ie, use CPU_ALLOC toallocate cpumask, CPU_ALLOC_SIZE for size, CPU_SET_S to set mask bit. Fix all the relevant places in the code to use mask size which is largeenough to represent number of possible CPU's in the system. This patchset also address a fix for parse_setup_cpu_list function innuma bench to check if input CPU is online before binding task tothat CPU. This is to fix failures where, though CPU number is withinmax CPU, it could happen that CPU is offline. Here, sched_setaffinitywill result in failure when using cpumask having that cpu bit setin the mask. Patch 1 address fix for parse_setup_cpu_list to check if CPU used to bindtask is online. Patch 2 has fix for bench numa to work with machineshaving #CPUs > 1K Athira Rajeev (2): tools/perf: Fix perf bench numa testcase to check if CPU used to bindtask is online perf bench: Fix numa bench to fix usage of affinity for machines with#CPUs > 1K Changelog:v2 -> v3Link to the v2 version : https://lore.kernel.org/all/20220406175113.87881-1-atraj...@linux.vnet.ibm.com/ - From the v2 version, patch 1 and patch 2 are now part of upstream. - This v3 version separates patch 3 and patch 4 to address review comments from arnaldo which includes using sysfs__read_str for reading sysfs file and fixing the compilation issues observed in debian Tesed the patches on powerpc with CPU > 1K and other configurations as well, verified the perf bench numa with the patch set.Tested-by: Disha Goel tools/perf/bench/numa.c | 136 +-- tools/perf/util/header.c | 51 +++ tools/perf/util/header.h | 1 + 3 files changed, 153 insertions(+), 35 deletions(-)
[PATCH V3 0/2] Fix perf bench numa to work with machines having #CPUs > 1K
The perf benchmark for collections: numa hits failure in system configuration with CPU's more than 1024. These benchmarks uses "sched_getaffinity" and "sched_setaffinity" in the code to work with affinity. Example snippet from numa benchmark: <<>> perf: bench/numa.c:302: bind_to_node: Assertion `!(ret)' failed. Aborted (core dumped) <<>> bind_to_node function uses "sched_getaffinity" to save the cpumask. This fails with EINVAL because the default mask size in glibc is 1024 To overcome this 1024 CPUs mask size limitation of cpu_set_t, change the mask size using the CPU_*_S macros ie, use CPU_ALLOC to allocate cpumask, CPU_ALLOC_SIZE for size, CPU_SET_S to set mask bit. Fix all the relevant places in the code to use mask size which is large enough to represent number of possible CPU's in the system. This patchset also address a fix for parse_setup_cpu_list function in numa bench to check if input CPU is online before binding task to that CPU. This is to fix failures where, though CPU number is within max CPU, it could happen that CPU is offline. Here, sched_setaffinity will result in failure when using cpumask having that cpu bit set in the mask. Patch 1 address fix for parse_setup_cpu_list to check if CPU used to bind task is online. Patch 2 has fix for bench numa to work with machines having #CPUs > 1K Athira Rajeev (2): tools/perf: Fix perf bench numa testcase to check if CPU used to bind task is online perf bench: Fix numa bench to fix usage of affinity for machines with #CPUs > 1K Changelog: v2 -> v3 Link to the v2 version : https://lore.kernel.org/all/20220406175113.87881-1-atraj...@linux.vnet.ibm.com/ - From the v2 version, patch 1 and patch 2 are now part of upstream. - This v3 version separates patch 3 and patch 4 to address review comments from arnaldo which includes using sysfs__read_str for reading sysfs file and fixing the compilation issues observed in debian tools/perf/bench/numa.c | 136 +-- tools/perf/util/header.c | 51 +++ tools/perf/util/header.h | 1 + 3 files changed, 153 insertions(+), 35 deletions(-) -- 2.35.1
