Re: [PATCH 1/2] perf: Add munmap callback

From: Peter Zijlstra <peterz@infradead.org>
To: Andi Kleen <ak@linux.intel.com>
Cc: "Liang, Kan" <kan.liang@linux.intel.com>,
	Arnaldo Carvalho de Melo <acme@kernel.org>,
	Thomas Gleixner <tglx@linutronix.de>,
	Ingo Molnar <mingo@kernel.org>,
	linux-kernel@vger.kernel.org, bp@alien8.de,
	Stephane Eranian <eranian@google.com>
Subject: Re: [PATCH 1/2] perf: Add munmap callback
Date: Thu, 25 Oct 2018 02:31:30 +0200	[thread overview]
Message-ID: <20181025003130.GU3109@worktop.c.hoisthospitality.com> (raw)
In-Reply-To: <20181024182854.GH6218@tassilo.jf.intel.com>

On Wed, Oct 24, 2018 at 11:28:54AM -0700, Andi Kleen wrote:
> > > void perf_event_mmap(struct vm_area_struct *vma)
> > > {
> > >          struct perf_mmap_event mmap_event;
> > > 
> > >          if (!atomic_read(&nr_mmap_events))
> > >                  return;
> > > <SNIP>
> > > }
> > > 
> > 
> > Thanks. I'll add the nr_mmap_events check in V2.
> 
> No, that's the wrong check here. The PEBS flush is independent of mmap
> events being requested.
> 
> If anything would need to check for any PEBS events active, which
> would need a new counter.  I think the easiest is to just check if 
> this_cpu_ptr(&sched_cb_list)
> is empty, which should be good enough.

It is just the CLI+STI, not PUSHF;CLI+POPF that are required and that is
a lot cheaper already. Also, you need to have preemption disabled in
order to check the per-cpu cb list.

So I don't think it really makes much sense to try and frob a fast path
in there.