Re: [PATCH] perf: Rewrite core context handling

From: Ravi Bangoria <ravi.bangoria@amd.com>
To: Peter Zijlstra <peterz@infradead.org>
Cc: acme@kernel.org, alexander.shishkin@linux.intel.com,
	jolsa@redhat.com, namhyung@kernel.org, songliubraving@fb.com,
	eranian@google.com, ak@linux.intel.com, mark.rutland@arm.com,
	frederic@kernel.org, maddy@linux.ibm.com, irogers@google.com,
	will@kernel.org, robh@kernel.org, mingo@redhat.com,
	catalin.marinas@arm.com, linux-arm-kernel@lists.infradead.org,
	linux-perf-users@vger.kernel.org, linuxppc-dev@lists.ozlabs.org,
	linux-s390@vger.kernel.org, linux-kernel@vger.kernel.org,
	sandipan.das@amd.com, ananth.narayan@amd.com,
	kim.phillips@amd.com, santosh.shukla@amd.com,
	ravi.bangoria@amd.com
Subject: Re: [PATCH] perf: Rewrite core context handling
Date: Mon, 5 Sep 2022 10:10:14 +0530	[thread overview]
Message-ID: <8f5b618e-e0a1-c5f2-3d4d-a41ea96fd59c@amd.com> (raw)
In-Reply-To: <YxCXqjfiJca3izQd@hirez.programming.kicks-ass.net>

> So the basic issue I mentioned is that:
> 
> 
> /*
>  *           ,------------------------[1:n]---------------------.
>  *           V                                                  V
>  * perf_event_context <-[1:n]-> perf_event_pmu_context <--- perf_event
>  *           ^                      ^     |                     |
>  *           `--------[1:n]---------'     `-[n:1]-> pmu <-[1:n]-'
>  *
>  *
>  * XXX destroy epc when empty
>  *   refcount, !rcu
>  *
>  * XXX epc locking
>  *
>  *   event->pmu_ctx            ctx->mutex && inactive
>  *   ctx->pmu_ctx_list         ctx->mutex && ctx->lock
>  *
>  */
> struct perf_event_pmu_context {
> 	...
>         atomic_t                        refcount; /* event <-> epc */
> 	...
> }
> 
> But that then also suggests that:
> 
> struct perf_event_context {
> 	...
>         refcount_t                      refcount;
> 	...
> }
> 
> should be: ctx <-> epc, and that is not so, the ctx::refcount still
> counts the number of events.
> 
> Now this might not be bad, but it is confusing.

I don't have strong opinion, but we store events in ctx, not in pmu_ctx.
So, I think it makes sense to keep refcount as ctx <-> event?

[...]

> +void perf_pmu_migrate_context(struct pmu *pmu, int src_cpu, int dst_cpu)
> +{
> +	struct perf_event_context *src_ctx, *dst_ctx;
> +	LIST_HEAD(events);
> +
> +	src_ctx = &per_cpu_ptr(&cpu_context, src_cpu)->ctx;
> +	dst_ctx = &per_cpu_ptr(&cpu_context, dst_cpu)->ctx;
> +
> +	/*
> +	 * See perf_event_ctx_lock() for comments on the details
> +	 * of swizzling perf_event::ctx.
> +	 */
> +	mutex_lock_double(&src_ctx->mutex, &dst_ctx->mutex);
> +
> +	__perf_pmu_remove(src_ctx, src_cpu, pmu, &src_src->pinned_groups, &events);
> +	__perf_pmu_remove(src_ctx, src_cpu, pmu, &src_src->flexible_groups, &events);

Rbtrees does not contain sibling events. Shouldn't we continue using
event_list here?

Thanks,
Ravi

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel