Re: [PATCH 2/3] Provide and use an always inline version of finish_task_switch

From: Segher Boessenkool <segher@kernel.crashing.org>
To: Rik van Riel <riel@surriel.com>
Cc: Xie Yuanbin <qq570070308@gmail.com>,
	linux@armlinux.org.uk, mathieu.desnoyers@efficios.com,
	paulmck@kernel.org, pjw@kernel.org, palmer@dabbelt.com,
	aou@eecs.berkeley.edu, alex@ghiti.fr, hca@linux.ibm.com,
	gor@linux.ibm.com, agordeev@linux.ibm.com,
	borntraeger@linux.ibm.com, svens@linux.ibm.com,
	davem@davemloft.net, andreas@gaisler.com, tglx@linutronix.de,
	mingo@redhat.com, bp@alien8.de, dave.hansen@linux.intel.com,
	hpa@zytor.com, luto@kernel.org, peterz@infradead.org,
	acme@kernel.org, namhyung@kernel.org, mark.rutland@arm.com,
	alexander.shishkin@linux.intel.com, jolsa@kernel.org,
	irogers@google.com, adrian.hunter@intel.com,
	anna-maria@linutronix.de, frederic@kernel.org,
	juri.lelli@redhat.com, vincent.guittot@linaro.org,
	dietmar.eggemann@arm.com, rostedt@goodmis.org,
	bsegall@google.com, mgorman@suse.de, vschneid@redhat.com,
	thuth@redhat.com, akpm@linux-foundation.org, david@redhat.com,
	lorenzo.stoakes@oracle.com, ryan.roberts@arm.com,
	max.kellermann@ionos.com, urezki@gmail.com, nysal@linux.ibm.com,
	x86@kernel.org, linux-arm-kernel@lists.infradead.org,
	linux-kernel@vger.kernel.org, linux-riscv@lists.infradead.org,
	linux-s390@vger.kernel.org, sparclinux@vger.kernel.org,
	linux-perf-users@vger.kernel.org, will@kernel.org
Subject: Re: [PATCH 2/3] Provide and use an always inline version of finish_task_switch
Date: Sat, 25 Oct 2025 09:36:12 -0500	[thread overview]
Message-ID: <aPzgXBd34r2wVy8I@gate> (raw)
In-Reply-To: <18734a4944e47952b7ad3e10a36c902392bdaa91.camel@surriel.com>

On Fri, Oct 24, 2025 at 05:36:06PM -0400, Rik van Riel wrote:
> On Sat, 2025-10-25 at 02:35 +0800, Xie Yuanbin wrote:
> > finish_task_switch is called during context switching,
> > inlining it can bring some performance benefits.
> > 
> > Add an always inline version `finish_task_switch_ainline` to be
> > called
> > during context switching, and keep the original version for being
> > called
> > elsewhere, so as to take into account the size impact.
> 
> Does that actually work, or does the compiler
> still inline some of those "non-inlined" versions,
> anyway?

Of course the compiler does!  That is part of the compiler's job after
all, to generate fast, efficient code!

The compiler will inline stuff when a) it *can*, mostly it has to have
the function body available; and b) it estimates it to be a win to
inline it.  There is a whole bunch of heuristics for this.  One of those
is that the always_inline attribute will do the utmost to get inlining
to happen.

(All that is assuming you have -finline-functions turned on, like you do
have at -O2).

Segher