* [PATCH v2] selftests: powerpc: Fix CPU affinity for child process
@ 2020-06-09 3:40 Harish
2020-06-09 6:32 ` Satheesh Rajendran
2020-06-09 7:21 ` Kamalesh Babulal
0 siblings, 2 replies; 3+ messages in thread
From: Harish @ 2020-06-09 3:40 UTC (permalink / raw)
To: mpe; +Cc: srikar, kamalesh, shiganta, sandipan, Harish, linuxppc-dev
On systems with large number of cpus, test fails trying to set
affinity for child process by calling sched_setaffinity() with
smaller size for cpuset. This patch fixes it by making sure that
the size of allocated cpu set is dependent on the number of CPUs
as reported by get_nprocs().
Fixes: 00b7ec5c9cf3 ("selftests/powerpc: Import Anton's context_switch2 benchmark")
Reported-by: Shirisha Ganta <shiganta@in.ibm.com>
Signed-off-by: Harish <harish@linux.ibm.com>
Signed-off-by: Sandipan Das <sandipan@linux.ibm.com>
---
.../powerpc/benchmarks/context_switch.c | 18 ++++++++++++------
1 file changed, 12 insertions(+), 6 deletions(-)
diff --git a/tools/testing/selftests/powerpc/benchmarks/context_switch.c b/tools/testing/selftests/powerpc/benchmarks/context_switch.c
index a2e8c9da7fa5..de6c49d6f88f 100644
--- a/tools/testing/selftests/powerpc/benchmarks/context_switch.c
+++ b/tools/testing/selftests/powerpc/benchmarks/context_switch.c
@@ -19,6 +19,7 @@
#include <limits.h>
#include <sys/time.h>
#include <sys/syscall.h>
+#include <sys/sysinfo.h>
#include <sys/types.h>
#include <sys/shm.h>
#include <linux/futex.h>
@@ -104,8 +105,9 @@ static void start_thread_on(void *(*fn)(void *), void *arg, unsigned long cpu)
static void start_process_on(void *(*fn)(void *), void *arg, unsigned long cpu)
{
- int pid;
- cpu_set_t cpuset;
+ int pid, ncpus;
+ cpu_set_t *cpuset;
+ size_t size;
pid = fork();
if (pid == -1) {
@@ -116,12 +118,16 @@ static void start_process_on(void *(*fn)(void *), void *arg, unsigned long cpu)
if (pid)
return;
- CPU_ZERO(&cpuset);
- CPU_SET(cpu, &cpuset);
+ size = CPU_ALLOC_SIZE(ncpus);
+ ncpus = get_nprocs();
+ cpuset = CPU_ALLOC(ncpus);
+ CPU_ZERO_S(size, cpuset);
+ CPU_SET_S(cpu, size, cpuset);
- if (sched_setaffinity(0, sizeof(cpuset), &cpuset)) {
+ if (sched_setaffinity(0, size, cpuset)) {
perror("sched_setaffinity");
- exit(1);
+ CPU_FREE(cpuset);
+ exit(-1);
}
fn(arg);
--
2.24.1
^ permalink raw reply related [flat|nested] 3+ messages in thread
* Re: [PATCH v2] selftests: powerpc: Fix CPU affinity for child process
2020-06-09 3:40 [PATCH v2] selftests: powerpc: Fix CPU affinity for child process Harish
@ 2020-06-09 6:32 ` Satheesh Rajendran
2020-06-09 7:21 ` Kamalesh Babulal
1 sibling, 0 replies; 3+ messages in thread
From: Satheesh Rajendran @ 2020-06-09 6:32 UTC (permalink / raw)
To: Harish; +Cc: srikar, kamalesh, shiganta, sandipan, linuxppc-dev
On Tue, Jun 09, 2020 at 09:10:05AM +0530, Harish wrote:
> On systems with large number of cpus, test fails trying to set
> affinity for child process by calling sched_setaffinity() with
> smaller size for cpuset. This patch fixes it by making sure that
> the size of allocated cpu set is dependent on the number of CPUs
> as reported by get_nprocs().
>
> Fixes: 00b7ec5c9cf3 ("selftests/powerpc: Import Anton's context_switch2 benchmark")
> Reported-by: Shirisha Ganta <shiganta@in.ibm.com>
> Signed-off-by: Harish <harish@linux.ibm.com>
> Signed-off-by: Sandipan Das <sandipan@linux.ibm.com>
> ---
> .../powerpc/benchmarks/context_switch.c | 18 ++++++++++++------
> 1 file changed, 12 insertions(+), 6 deletions(-)
>
> diff --git a/tools/testing/selftests/powerpc/benchmarks/context_switch.c b/tools/testing/selftests/powerpc/benchmarks/context_switch.c
> index a2e8c9da7fa5..de6c49d6f88f 100644
> --- a/tools/testing/selftests/powerpc/benchmarks/context_switch.c
> +++ b/tools/testing/selftests/powerpc/benchmarks/context_switch.c
> @@ -19,6 +19,7 @@
> #include <limits.h>
> #include <sys/time.h>
> #include <sys/syscall.h>
> +#include <sys/sysinfo.h>
> #include <sys/types.h>
> #include <sys/shm.h>
> #include <linux/futex.h>
> @@ -104,8 +105,9 @@ static void start_thread_on(void *(*fn)(void *), void *arg, unsigned long cpu)
>
> static void start_process_on(void *(*fn)(void *), void *arg, unsigned long cpu)
> {
> - int pid;
> - cpu_set_t cpuset;
> + int pid, ncpus;
> + cpu_set_t *cpuset;
> + size_t size;
>
> pid = fork();
> if (pid == -1) {
> @@ -116,12 +118,16 @@ static void start_process_on(void *(*fn)(void *), void *arg, unsigned long cpu)
> if (pid)
> return;
>
> - CPU_ZERO(&cpuset);
> - CPU_SET(cpu, &cpuset);
> + size = CPU_ALLOC_SIZE(ncpus);
> + ncpus = get_nprocs();
above two lines should be interchanged, ncpus not assigned while getting used to get size.
> + cpuset = CPU_ALLOC(ncpus);
> + CPU_ZERO_S(size, cpuset);
> + CPU_SET_S(cpu, size, cpuset);
>
> - if (sched_setaffinity(0, sizeof(cpuset), &cpuset)) {
> + if (sched_setaffinity(0, size, cpuset)) {
> perror("sched_setaffinity");
> - exit(1);
> + CPU_FREE(cpuset);
> + exit(-1);
do we need to change the return value here?
probably other framework might rely on previous value?
Regards,
-Satheesh.
> }
>
> fn(arg);
> --
> 2.24.1
>
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: [PATCH v2] selftests: powerpc: Fix CPU affinity for child process
2020-06-09 3:40 [PATCH v2] selftests: powerpc: Fix CPU affinity for child process Harish
2020-06-09 6:32 ` Satheesh Rajendran
@ 2020-06-09 7:21 ` Kamalesh Babulal
1 sibling, 0 replies; 3+ messages in thread
From: Kamalesh Babulal @ 2020-06-09 7:21 UTC (permalink / raw)
To: Harish; +Cc: srikar, shiganta, sandipan, linuxppc-dev
On 6/9/20 9:10 AM, Harish wrote:
> On systems with large number of cpus, test fails trying to set
> affinity for child process by calling sched_setaffinity() with
> smaller size for cpuset. This patch fixes it by making sure that
> the size of allocated cpu set is dependent on the number of CPUs
> as reported by get_nprocs().
>
> Fixes: 00b7ec5c9cf3 ("selftests/powerpc: Import Anton's context_switch2 benchmark")
> Reported-by: Shirisha Ganta <shiganta@in.ibm.com>
> Signed-off-by: Harish <harish@linux.ibm.com>
> Signed-off-by: Sandipan Das <sandipan@linux.ibm.com>
> ---
> .../powerpc/benchmarks/context_switch.c | 18 ++++++++++++------
> 1 file changed, 12 insertions(+), 6 deletions(-)
>
> diff --git a/tools/testing/selftests/powerpc/benchmarks/context_switch.c b/tools/testing/selftests/powerpc/benchmarks/context_switch.c
> index a2e8c9da7fa5..de6c49d6f88f 100644
> --- a/tools/testing/selftests/powerpc/benchmarks/context_switch.c
> +++ b/tools/testing/selftests/powerpc/benchmarks/context_switch.c
> @@ -19,6 +19,7 @@
> #include <limits.h>
> #include <sys/time.h>
> #include <sys/syscall.h>
> +#include <sys/sysinfo.h>
> #include <sys/types.h>
> #include <sys/shm.h>
> #include <linux/futex.h>
> @@ -104,8 +105,9 @@ static void start_thread_on(void *(*fn)(void *), void *arg, unsigned long cpu)
>
> static void start_process_on(void *(*fn)(void *), void *arg, unsigned long cpu)
> {
> - int pid;
> - cpu_set_t cpuset;
> + int pid, ncpus;
> + cpu_set_t *cpuset;
> + size_t size;
>
> pid = fork();
> if (pid == -1) {
> @@ -116,12 +118,16 @@ static void start_process_on(void *(*fn)(void *), void *arg, unsigned long cpu)
> if (pid)
> return;
>
> - CPU_ZERO(&cpuset);
> - CPU_SET(cpu, &cpuset);
> + size = CPU_ALLOC_SIZE(ncpus);
> + ncpus = get_nprocs();
> + cpuset = CPU_ALLOC(ncpus);
CPU_ALLOC() allocation failure needs to be checked, like malloc() allocations.
> + CPU_ZERO_S(size, cpuset);
> + CPU_SET_S(cpu, size, cpuset);
>
> - if (sched_setaffinity(0, sizeof(cpuset), &cpuset)) {
> + if (sched_setaffinity(0, size, cpuset)) {
> perror("sched_setaffinity");
> - exit(1);
> + CPU_FREE(cpuset);
> + exit(-1);
> }
once the cpu affinity is set, you probably want to free the cpuset mask.
>
> fn(arg);
>
--
Kamalesh
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2020-06-09 7:43 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2020-06-09 3:40 [PATCH v2] selftests: powerpc: Fix CPU affinity for child process Harish
2020-06-09 6:32 ` Satheesh Rajendran
2020-06-09 7:21 ` Kamalesh Babulal
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).