[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <umy4kicwvlsaszajo62pcpgptcdqi2yp7pponvlch5h6tirabb@mhhq2en3piel>
Date: Tue, 27 May 2025 17:33:55 +0200
From: Michal Koutný <mkoutny@...e.com>
To: Yafang Shao <laoar.shao@...il.com>
Cc: mingo@...hat.com, peterz@...radead.org, hannes@...xchg.org,
juri.lelli@...hat.com, vincent.guittot@...aro.org, dietmar.eggemann@....com,
rostedt@...dmis.org, bsegall@...gle.com, mgorman@...e.de, vschneid@...hat.com,
surenb@...gle.com, linux-kernel@...r.kernel.org, cgroups@...r.kernel.org,
lkp@...el.com, Eric Dumazet <edumazet@...gle.com>
Subject: Re: [PATCH v9 2/2] sched: Annotate sched_clock_irqtime with
__read_mostly
On Sun, May 11, 2025 at 11:08:00AM +0800, Yafang Shao <laoar.shao@...il.com> wrote:
> Eric reported an issue [0] as follows,
> : rebalance_domains() can attempt to change sched_balance_running
> : more than 350,000 times per second on our servers.
>
> : If sched_clock_irqtime and sched_balance_running share the
> : same cache line, we see a very high cost on hosts with 480 threads
> : dealing with many interrupts.
I'd say this patch could be independent from the "series".
> While the rebalance_domains() issue has been resolved [1], we should
> proactively annotate sched_clock_irqtime with __read_mostly to prevent
> potential cacheline false sharing. This optimization is particularly
> justified since sched_clock_irqtime is only modified during TSC instability
> events.
>
> Link: https://lore.kernel.org/all/20250423174634.3009657-1-edumazet@google.com/ [0]
> Link: https://lore.kernel.org/all/20250416035823.1846307-1-tim.c.chen@linux.intel.com/ [1]
>
> Reported-by: Eric Dumazet <edumazet@...gle.com>
> Debugged-by: Eric Dumazet <edumazet@...gle.com>
> Signed-off-by: Yafang Shao <laoar.shao@...il.com>
> Cc: Eric Dumazet <edumazet@...gle.com>
I can say
Reviewed-by: Michal Koutný <mkoutny@...e.com>
but it'd be good to have also Tested-by: wrt the cache traffic
reduction.
0.02€,
Michal
Download attachment "signature.asc" of type "application/pgp-signature" (229 bytes)
Powered by blists - more mailing lists