linux-kernel - Re: [PATCH] perf/core: Fix warning warning due to unordred pmu_ctx

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <fac5bb75-8bce-4181-8278-c971d35b8b3a@linux.intel.com>
Date: Mon, 20 Jan 2025 15:49:09 -0500
From: "Liang, Kan" <kan.liang@...ux.intel.com>
To: Luo Gengkun <luogengkun@...weicloud.com>, peterz@...radead.org
Cc: mingo@...hat.com, acme@...nel.org, namhyung@...nel.org,
 mark.rutland@....com, alexander.shishkin@...ux.intel.com, jolsa@...nel.org,
 irogers@...gle.com, adrian.hunter@...el.com, ravi.bangoria@....com,
 linux-perf-users@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH] perf/core: Fix warning warning due to unordred
 pmu_ctx_list

A redundant "warning" is in the title.

On 2025-01-20 6:43 a.m., Luo Gengkun wrote:
> Syskaller triggers a warning due to prev_epc->pmu != next_epc->pmu in
> perf_event_swap_task_ctx_data. vmcore shows that two lists have the same
> perf_event_pmu_context, but not in the same order.
> 
> The problem is that when inheritance is performed, it traverses the ordered
> groups of events, and inserts the new perf_event_pmu_context into
> child_ctx->pmu_ctx_list which is unordered. So the order of pmu_ctx_list in
> the parent and child may be different.

I think the order of pmu_ctx_list for the parent should be impacted by
the time when an event/pmu is added.
While the order for a child should be impacted by the event order in the
pinned_groups and flexible_groups.

> 
> The follow testcase can trigger above warning:
> 
>  # perf record -e cycles --call-graph lbr -- taskset -c 3 ./a.out &
>  # perf stat -e cpu-clock,cs -p xxx // xxx is the pid of a.out
> 
> test.c
> 
> void main() {
>         int count = 0;
>         pid_t pid;
> 
>         printf("%d running\n", getpid());
>         sleep(30);
>         printf("running\n");
> 
>         pid = fork();
>         if (pid == -1) {
>                 printf("fork error\n");
>                 return;
>         }
>         if (pid == 0) {
>                 while (1) {
>                         count++;
>                 }
>         } else {
>                 while (1) {
>                         count++;
>                 }
>         }
> }
> 
> The testcase first open a lbr event, so it will alloc task_ctx_data, and
> then open tracepoint and software events, so the parent ctx will have 3
> different perf_event_pmu_contexts. When doing inherit, child ctx will
> insert the perf_event_pmu_context in another order then the warning will
> trigger.
> 
> To fix this problem, add pmu_ctx_insertion_sort to make sure the
> pmu_ctx_list is ordered.
> 
> Fixes: bd2756811766 ("perf: Rewrite core context handling")
> Signed-off-by: Luo Gengkun <luogengkun@...weicloud.com>
> ---
>  kernel/events/core.c | 22 ++++++++++++++++++++--
>  1 file changed, 20 insertions(+), 2 deletions(-)
> 
> diff --git a/kernel/events/core.c b/kernel/events/core.c
> index 95b01a51139d..1bdff3ef0ce2 100644
> --- a/kernel/events/core.c
> +++ b/kernel/events/core.c
> @@ -4953,6 +4953,24 @@ find_get_context(struct task_struct *task, struct perf_event *event)
>  	return ERR_PTR(err);
>  }
>  
> +/*
> + * This function ensures that ctx->pmu_ctx_list is ordered, so that no warning
> + * is triggered due to prev_epc->pmu != next_epc->pmu.
> + */
> +static void pmu_ctx_insertion_sort(struct perf_event_pmu_context *new,
> +				   struct perf_event_context *ctx)
> +{
> +	struct perf_event_pmu_context *epc;
> +
> +	lockdep_assert_held(&ctx->lock);
> +
> +	list_for_each_entry(epc, &ctx->pmu_ctx_list, pmu_ctx_entry) {
> +		if (epc->pmu > new->pmu)
> +			break;
> +	}
> +	list_add(&new->pmu_ctx_entry, epc->pmu_ctx_entry.prev);
> +}
> +
>  static struct perf_event_pmu_context *
>  find_get_pmu_context(struct pmu *pmu, struct perf_event_context *ctx,
>  		     struct perf_event *event)
> @@ -4974,7 +4992,7 @@ find_get_pmu_context(struct pmu *pmu, struct perf_event_context *ctx,
>  		if (!epc->ctx) {
>  			atomic_set(&epc->refcount, 1);
>  			epc->embedded = 1;
> -			list_add(&epc->pmu_ctx_entry, &ctx->pmu_ctx_list);
> +			pmu_ctx_insertion_sort(epc, ctx);

The CPU event and per-task event should have a different ctx.
The warning should only be triggered for the per-task event, right?
If so, I don't think a sort is required here.

>  			epc->ctx = ctx;
>  		} else {
>  			WARN_ON_ONCE(epc->ctx != ctx);
> @@ -5021,7 +5039,7 @@ find_get_pmu_context(struct pmu *pmu, struct perf_event_context *ctx,
>  	printk(KERN_INFO
>  		"lgk: ctx %p insert pmu ctx %p, pmu is %p!\n", ctx, epc, epc->pmu);

Seems your debug code. Please send a clean patch.

>  
> -	list_add(&epc->pmu_ctx_entry, &ctx->pmu_ctx_list);
> +	pmu_ctx_insertion_sort(epc, ctx);

I think the pmu_ctx_list has already traversed to find a matched pmu
right before. The traverse in the pmu_ctx_insertion_sort() can be avoided.

Thanks,
Kan
>  	epc->ctx = ctx;
>  
>  found_epc: