[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20191023152920.GG19358@hirez.programming.kicks-ass.net>
Date: Wed, 23 Oct 2019 17:29:20 +0200
From: Peter Zijlstra <peterz@...radead.org>
To: Stephane Eranian <eranian@...gle.com>
Cc: LKML <linux-kernel@...r.kernel.org>, mingo@...e.hu,
Arnaldo Carvalho de Melo <acme@...hat.com>,
Jiri Olsa <jolsa@...hat.com>,
"Liang, Kan" <kan.liang@...el.com>,
Song Liu <songliubraving@...com>,
Ian Rogers <irogers@...gle.com>
Subject: Re: [PATCH] perf/core: fix multiplexing event scheduling issue
On Wed, Oct 23, 2019 at 11:37:57AM +0200, Peter Zijlstra wrote:
> Further, since we set it on reschedule, I propose you change the above
> like:
>
> if (ctx->rotate_necessary) {
> int type = get_event_type(event);
> /*
> * comment..
> */
> if (type & EVENT_PINNED)
> type |= EVENT_FLEXIBLE;
> + /*
> + * Will be reset by ctx_resched()'s flexible_sched_in().
> + */
> + ctx->rotate_necessary = 0;
> ctx_resched(cpuctx, cpuctx->task_ctx, type);
> }
n/m, that is already done through ctx_sched_out().
Powered by blists - more mailing lists