[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <3mc7l6otsn4ufmyaiuqgpf64rfcukilgpjainslniwid6ajqm7@ltxbi5qennh7>
Date: Tue, 1 Apr 2025 17:00:00 +0200
From: Michal Koutný <mkoutny@...e.com>
To: Mateusz Guzik <mjguzik@...il.com>
Cc: Yosry Ahmed <yosry.ahmed@...ux.dev>, Greg Thelen <gthelen@...gle.com>,
Tejun Heo <tj@...nel.org>, Johannes Weiner <hannes@...xchg.org>,
Andrew Morton <akpm@...ux-foundation.org>, Eric Dumazet <edumzaet@...gle.com>, cgroups@...r.kernel.org,
linux-mm@...ck.org, linux-kernel@...r.kernel.org, Eric Dumazet <edumazet@...gle.com>
Subject: Re: [PATCH] cgroup/rstat: avoid disabling irqs for O(num_cpu)
Hello Mateusz.
On Thu, Mar 27, 2025 at 06:47:56PM +0100, Mateusz Guzik <mjguzik@...il.com> wrote:
> I feel compelled to note atomics on x86-64 were expensive for as long
> as the architecture was around so I'm confused what's up with the
> resistance to the notion that they remain costly even with modern
> uarchs. If anything, imo claims that they are cheap require strong
> evidence.
I don't there's strong resistance, your measurements show that it's not
negligible under given conditions.
The question is -- how much benefit would flushers have in practice with
coalesced unlock-locks. There is the approach now with releasing for
each CPU that is simple and benefits latency of irq dependants.
If you see practical issues with the limited throughputs of stat readers
(or flushers in general) because of this, please send a patch for
discussion that resolves it while preserving (some of) the irq freedom.
Also there is ongoing work of splitting up flushing per controller --
I'd like to see whether the given locks become "small" enough to require
no _irq exclusion at all during flushing.
Michal
Download attachment "signature.asc" of type "application/pgp-signature" (229 bytes)
Powered by blists - more mailing lists