[v2] selftests: powerpc: Fix CPU affinity for child process

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(-)

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
>

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);
>