Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Tue, 23 Aug 2022 09:10:02 +0200
From:   Peter Zijlstra <peterz@infradead.org>
To:     Ravi Bangoria <ravi.bangoria@amd.com>
Cc:     acme@kernel.org, alexander.shishkin@linux.intel.com,
        jolsa@redhat.com, namhyung@kernel.org, songliubraving@fb.com,
        eranian@google.com, alexey.budankov@linux.intel.com,
        ak@linux.intel.com, mark.rutland@arm.com, megha.dey@intel.com,
        frederic@kernel.org, maddy@linux.ibm.com, irogers@google.com,
        kim.phillips@amd.com, linux-kernel@vger.kernel.org,
        santosh.shukla@amd.com
Subject: Re: [RFC v2] perf: Rewrite core context handling
Message-ID: <YwR9ShCHDBgrvT9s@worktop.programming.kicks-ass.net>
References: <20220113134743.1292-1-ravi.bangoria@amd.com>
 <2acacbcb-2bdb-30d6-3300-8d125b06a0f3@amd.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <2acacbcb-2bdb-30d6-3300-8d125b06a0f3@amd.com>
Precedence: bulk

On Tue, Aug 02, 2022 at 11:43:03AM +0530, Ravi Bangoria wrote:
> [...]
> 
> >  /*
> > @@ -2718,7 +2706,6 @@ static void ctx_resched(struct perf_cpu_context *cpuctx,
> >  			struct perf_event_context *task_ctx,
> >  			enum event_type_t event_type)
> >  {
> > -	enum event_type_t ctx_event_type;
> >  	bool cpu_event = !!(event_type & EVENT_CPU);
> >  
> >  	/*
> > @@ -2728,11 +2715,13 @@ static void ctx_resched(struct perf_cpu_context *cpuctx,
> >  	if (event_type & EVENT_PINNED)
> >  		event_type |= EVENT_FLEXIBLE;
> >  
> > -	ctx_event_type = event_type & EVENT_ALL;
> > +	event_type &= EVENT_ALL;
> >  
> > -	perf_pmu_disable(cpuctx->ctx.pmu);
> > -	if (task_ctx)
> > -		task_ctx_sched_out(cpuctx, task_ctx, event_type);
> > +	perf_ctx_disable(&cpuctx->ctx);
> > +	if (task_ctx) {
> > +		perf_ctx_disable(task_ctx);
> > +		task_ctx_sched_out(task_ctx, event_type);
> > +	}
> >  
> >  	/*
> >  	 * Decide which cpu ctx groups to schedule out based on the types
> > @@ -2742,17 +2731,20 @@ static void ctx_resched(struct perf_cpu_context *cpuctx,
> >  	 *  - otherwise, do nothing more.
> >  	 */
> >  	if (cpu_event)
> > -		cpu_ctx_sched_out(cpuctx, ctx_event_type);
> > -	else if (ctx_event_type & EVENT_PINNED)
> > -		cpu_ctx_sched_out(cpuctx, EVENT_FLEXIBLE);
> > +		ctx_sched_out(&cpuctx->ctx, event_type);
> > +	else if (event_type & EVENT_PINNED)
> > +		ctx_sched_out(&cpuctx->ctx, EVENT_FLEXIBLE);
> >  
> >  	perf_event_sched_in(cpuctx, task_ctx, current);
> > -	perf_pmu_enable(cpuctx->ctx.pmu);
> > +
> > +	perf_ctx_enable(&cpuctx->ctx);
> > +	if (task_ctx)
> > +		perf_ctx_enable(task_ctx);
> >  }
> 
> ctx_resched() reschedule entire perf_event_context while adding new event
> to the context or enabling existing event in the context. We can probably
> optimize it by rescheduling only affected pmu_ctx.

Yes, it would probably make sense to add a pmu argument there and limit
the rescheduling where possible.