[igt-dev] [i-g-t, v2, 2/3] benchmarks/gem_syslatency: don't die with SIGSEGV

[Kamil Konieczny]

On 2022-05-31 at 08:52:10 +0200, Mauro Carvalho Chehab wrote:
> From: Mauro Carvalho Chehab <mchehab at kernel.org>
> 
> Fix the pthread_create() logic for it to properly handle temporary
> issues returned with EAGAIN and ensure that thread creation
> won't fail, as, if it fails, it will cause a Segmentation fault:
> 
> 	Received signal SIGSEGV.
> 	Stack trace:
> 	 #0 [fatal_sig_handler+0x163]
> 	 #1 [__sigaction+0x50]
> 	 #2 [__pthread_clockjoin_ex+0x22]
> 	 #3 [main+0x346]
> 	 #4 [__libc_start_call_main+0x80]
> 	 #5 [__libc_start_main+0x89]
> 	 #6 [_start+0x25]
> 	Segmentation fault
> 
> Signed-off-by: Mauro Carvalho Chehab <mchehab at kernel.org>
> ---
>  benchmarks/gem_syslatency.c | 14 ++++++++++----
>  1 file changed, 10 insertions(+), 4 deletions(-)
> 
> diff --git a/benchmarks/gem_syslatency.c b/benchmarks/gem_syslatency.c
> index 035ee9346251..4b097e1c9316 100644
> --- a/benchmarks/gem_syslatency.c
> +++ b/benchmarks/gem_syslatency.c
> @@ -351,7 +351,7 @@ int main(int argc, char **argv)
>  	bool leak = false;
>  	bool interrupts = false;
>  	long batch = 0;
> -	int n, c;
> +	int ret, n, c;
>  
>  	while ((c = getopt(argc, argv, "r:t:f:bmni1")) != -1) {
>  		switch (c) {
> @@ -409,8 +409,11 @@ int main(int argc, char **argv)
>  			busy[n].sz = batch;
>  			busy[n].leak = leak;
>  			busy[n].interrupts = interrupts;
> -			pthread_create(&busy[n].thread, &attr,
> -				       gem_busyspin, &busy[n]);
> +			do {
> +				ret = pthread_create(&busy[n].thread, &attr,
> +						     gem_busyspin, &busy[n]);
> +			} while (ret == EAGAIN);

Should we make this loop and wait for other error indefinitly ?
imho it is better to make some counter after which it fails ?

> +			igt_assert_f(!ret, "Can't create task on CPU#%d\n", n);
>  		}
>  	}
>  
> @@ -420,7 +423,10 @@ int main(int argc, char **argv)
>  	for (n = 0; n < ncpus; n++) {
>  		igt_mean_init(&wait[n].mean);
>  		bind_cpu(&attr, n);
> -		pthread_create(&wait[n].thread, &attr, sys_fn, &wait[n]);
> +		do {
> +			ret = pthread_create(&wait[n].thread, &attr, sys_fn, &wait[n]);
> +		} while (ret == EAGAIN);
> +		igt_assert_f(!ret, "Can't create task on CPU#%d\n", n);

This is the same code, maybe it should be placed in igt lib as
a macro or function ?

--
Kamil

>  	}
>  
>  	sleep(time);