Re: [PATCH v2 11/11] perf/x86: Simplify put_exclusive_constraints

From: Peter Zijlstra <peterz@infradead.org>
To: mingo@kernel.org
Cc: vincent.weaver@maine.edu, eranian@google.com, jolsa@redhat.com,
	linux-kernel@vger.kernel.org
Subject: Re: [PATCH v2 11/11] perf/x86: Simplify put_exclusive_constraints
Date: Fri, 22 May 2015 15:38:43 +0200	[thread overview]
Message-ID: <20150522133843.GF3644@twins.programming.kicks-ass.net> (raw)
In-Reply-To: <20150522133136.355195312@infradead.org>

On Fri, May 22, 2015 at 03:29:16PM +0200, Peter Zijlstra wrote:
> @@ -2149,22 +2147,25 @@ static void intel_put_excl_constraints(s
>  	}
>  
>  	/*
> +	 * If event was actually assigned, then mark the counter state as
> +	 * unused now.
>  	 */
> +	if (hwc->idx >= 0) {
> +		xl = &excl_cntrs->states[tid];
> +
> +		/*
> +		 * put_constraint may be called from x86_schedule_events()
> +		 * which already has the lock held so here make locking
> +		 * conditional.
> +		 */
> +		if (!xl->sched_started)
> +			raw_spin_lock(&excl_cntrs->lock);
>  
>  		xl->state[hwc->idx] = INTEL_EXCL_UNUSED;
>  
> +		if (!xl->sched_started)
> +			raw_spin_unlock(&excl_cntrs->lock);
> +	}
>  }

Hmm, we could probably avoid the lock and use WRITE_ONCE() to clear this
state, but I'm too tired to really think about it.