On Fri, 28 Feb 2025 at 17:55, Brendan Jackman <jackmanb@xxxxxxxxxx> wrote:
>
> This calculation divides a fixed parameter by an environment-dependent
> parameter i.e. the number of CPUs.
>
> The simple way to avoid machine-specific failures here is to just put a
> cap on the max value of the latter.
>
> Suggested-by: Mateusz Guzik <mjguzik@xxxxxxxxx>
> Signed-off-by: Brendan Jackman <jackmanb@xxxxxxxxxx>
> ---
> tools/testing/selftests/mm/uffd-stress.c | 11 ++++++++++-
> 1 file changed, 10 insertions(+), 1 deletion(-)
>
> diff --git a/tools/testing/selftests/mm/uffd-stress.c b/tools/testing/selftests/mm/uffd-stress.c
> index efe2051c393096e237d942c04a264b6611a6e127..5656128590373ed376b3b5d9259e5ca3867a4099 100644
> --- a/tools/testing/selftests/mm/uffd-stress.c
> +++ b/tools/testing/selftests/mm/uffd-stress.c
> @@ -434,6 +434,7 @@ static void sigalrm(int sig)
>
> int main(int argc, char **argv)
> {
> + unsigned long nr_cpus;
> size_t bytes;
>
> if (argc < 4)
> @@ -452,7 +453,15 @@ int main(int argc, char **argv)
> return KSFT_SKIP;
> }
>
> - nr_threads = sysconf(_SC_NPROCESSORS_ONLN);
> + nr_cpus = sysconf(_SC_NPROCESSORS_ONLN);
> + if (nr_cpus > 32) {
> + /* Don't let calculation below go to zero. */
> + ksft_print_msg("_SC_NPROCESSORS_ONLN (%lu) too large, capping nr_threads to 32\n",
> + nr_cpus);
> + nr_threads = 32;
> + } else {
> + nr_cpus = nr_threads;
Won't have time to send v4 for a few days so I'll just note here: this
shoudl be nr_thread = nr_cpus. This causes a division by zero on
machines with less than 30 CPUs.
