Re: [PATCH] perf: Fix topology_sibling_cpumask check warning on ARM

From: Steven Price
Date: Thu Sep 12 2024 - 11:53:34 EST


On 12/09/2024 15:50, kan.liang@xxxxxxxxxxxxxxx wrote:
> From: Kan Liang <kan.liang@xxxxxxxxxxxxxxx>
>
> The below warning is triggered when building with arm
> multi_v7_defconfig.
>
> kernel/events/core.c: In function 'perf_event_setup_cpumask':
> kernel/events/core.c:14012:13: warning: the comparison will always
> evaluate as 'true' for the address of 'thread_sibling' will never be
> NULL [-Waddress]
> 14012 | if (!topology_sibling_cpumask(cpu)) {
>
> The perf_event_init_cpu() may be invoked at the early boot stage, while
> the topology_*_cpumask hasn't been initialized yet. The check is to
> specially handle the case, and initialize the perf_online_<domain>_masks
> on the boot CPU.
> X86 uses a per-cpu cpumask pointer, which could be NULL at the early
> boot stage. However, ARM uses a global variable, which never be NULL.
>
> Use perf_online_mask as an indicator instead. Only initialize the
> perf_online_<domain>_masks when perf_online_mask is empty.
>
> Fix a typo as well.
>
> Fixes: 4ba4f1afb6a9 ("perf: Generic hotplug support for a PMU with a scope")
> Reported-by: Stephen Rothwell <sfr@xxxxxxxxxxxxxxxx>
> Closes: https://lore.kernel.org/lkml/20240911153854.240bbc1f@xxxxxxxxxxxxxxxx/
> Reported-by: Steven Price <steven.price@xxxxxxx>
> Closes: https://lore.kernel.org/lkml/1835eb6d-3e05-47f3-9eae-507ce165c3bf@xxxxxxx/
> Signed-off-by: Kan Liang <kan.liang@xxxxxxxxxxxxxxx>

Tested-by: Steven Price <steven.price@xxxxxxx>

Thanks,
Steve

> ---
> kernel/events/core.c | 10 +++++-----
> 1 file changed, 5 insertions(+), 5 deletions(-)
>
> diff --git a/kernel/events/core.c b/kernel/events/core.c
> index 7a028474caef..20e97c1aa4d6 100644
> --- a/kernel/events/core.c
> +++ b/kernel/events/core.c
> @@ -13954,21 +13954,19 @@ static void perf_event_setup_cpumask(unsigned int cpu)
> struct cpumask *pmu_cpumask;
> unsigned int scope;
>
> - cpumask_set_cpu(cpu, perf_online_mask);
> -
> /*
> * Early boot stage, the cpumask hasn't been set yet.
> * The perf_online_<domain>_masks includes the first CPU of each domain.
> - * Always uncondifionally set the boot CPU for the perf_online_<domain>_masks.
> + * Always unconditionally set the boot CPU for the perf_online_<domain>_masks.
> */
> - if (!topology_sibling_cpumask(cpu)) {
> + if (cpumask_empty(perf_online_mask)) {
> for (scope = PERF_PMU_SCOPE_NONE + 1; scope < PERF_PMU_MAX_SCOPE; scope++) {
> pmu_cpumask = perf_scope_cpumask(scope);
> if (WARN_ON_ONCE(!pmu_cpumask))
> continue;
> cpumask_set_cpu(cpu, pmu_cpumask);
> }
> - return;
> + goto end;
> }
>
> for (scope = PERF_PMU_SCOPE_NONE + 1; scope < PERF_PMU_MAX_SCOPE; scope++) {
> @@ -13983,6 +13981,8 @@ static void perf_event_setup_cpumask(unsigned int cpu)
> cpumask_any_and(pmu_cpumask, cpumask) >= nr_cpu_ids)
> cpumask_set_cpu(cpu, pmu_cpumask);
> }
> +end:
> + cpumask_set_cpu(cpu, perf_online_mask);
> }
>
> int perf_event_init_cpu(unsigned int cpu)