Re: [RFC][PATCH] x86: optimization to avoid CAL+RES IPIs

From: Josh Don <joshdon@google.com>
To: Andy Lutomirski <luto@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>,
	Ingo Molnar <mingo@redhat.com>, Borislav Petkov <bp@alien8.de>,
	X86 ML <x86@kernel.org>, "H . Peter Anvin" <hpa@zytor.com>,
	Linux PM <linux-pm@vger.kernel.org>,
	LKML <linux-kernel@vger.kernel.org>,
	"Rafael J . Wysocki" <rjw@rjwysocki.net>,
	Daniel Lezcano <daniel.lezcano@linaro.org>,
	Peter Zijlstra <peterz@infradead.org>,
	Juri Lelli <juri.lelli@redhat.com>,
	Vincent Guittot <vincent.guittot@linaro.org>,
	Dietmar Eggemann <dietmar.eggemann@arm.com>,
	Steven Rostedt <rostedt@goodmis.org>,
	Ben Segall <bsegall@google.com>, Mel Gorman <mgorman@suse.de>,
	Paul Turner <pjt@google.com>
Subject: Re: [RFC][PATCH] x86: optimization to avoid CAL+RES IPIs
Date: Fri, 17 Jul 2020 22:53:50 -0700	[thread overview]
Message-ID: <CABk29NsN_3kH4xyY1v0FixMDuGb_rG_iUYDFtwd3U5n10LsjaA@mail.gmail.com> (raw)
In-Reply-To: <CALCETrXHEhwBS72_Fiv9LbPjsBGp_rMmU16oYMTZJFa=wTki8A@mail.gmail.com>

Hey Andy, thanks for taking a look.

On Fri, Jul 17, 2020 at 8:14 PM Andy Lutomirski <luto@kernel.org> wrote:
>
> PeterZ and I fixed a whole series of bugs a few years ago, and remote
> wakeups *should* already do this. Did we miss something?  Did it
> regress?  Even the call_function_single path ought to go through this:
>
> void send_call_function_single_ipi(int cpu)
> {
>         struct rq *rq = cpu_rq(cpu);
>
>         if (!set_nr_if_polling(rq->idle))
>                 arch_send_call_function_single_ipi(cpu);
>         else
>                 trace_sched_wake_idle_without_ipi(cpu);
> }
>

Yep, I was sitting on this for a bit and raced with b2a02fc43a there.
90b5363ac also got rid of the last smp_send_reschedule() that was
triggering the ipiless handling.

One of the nice parts of the patch was that it could blanket apply to
all of the smp_call/reschedule.  However, with the above patches that
isn't a concern; it makes more sense to keep the existing
TIF_POLLING_NRFLAG logic.