linux-kernel - Re: [PATCH V7 9/9] sched/rt: Remove default bandwidth control

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <Z0nTZXHDggHD8raj@jlelli-thinkpadt14gen4.remote.csb>
Date: Fri, 29 Nov 2024 15:44:53 +0100
From: Juri Lelli <juri.lelli@...hat.com>
To: Michal Koutný <mkoutny@...e.com>
Cc: Ingo Molnar <mingo@...hat.com>, Peter Zijlstra <peterz@...radead.org>,
	Vincent Guittot <vincent.guittot@...aro.org>,
	Dietmar Eggemann <dietmar.eggemann@....com>,
	Steven Rostedt <rostedt@...dmis.org>,
	Ben Segall <bsegall@...gle.com>, Mel Gorman <mgorman@...e.de>,
	Valentin Schneider <vschneid@...hat.com>,
	linux-kernel@...r.kernel.org,
	Luca Abeni <luca.abeni@...tannapisa.it>,
	Tommaso Cucinotta <tommaso.cucinotta@...tannapisa.it>,
	Thomas Gleixner <tglx@...utronix.de>,
	Joel Fernandes <joel@...lfernandes.org>,
	Vineeth Pillai <vineeth@...byteword.org>,
	Shuah Khan <skhan@...uxfoundation.org>,
	Phil Auld <pauld@...hat.com>,
	Suleiman Souhlal <suleiman@...gle.com>
Subject: Re: [PATCH V7 9/9] sched/rt: Remove default bandwidth control

On 29/11/24 11:02, Michal Koutný wrote:
> On Wed, Nov 27, 2024 at 04:35:39PM GMT, Juri Lelli <juri.lelli@...hat.com> wrote:
> > > The default DL bandwidth is thus is 95%. The fair server is given 5%.
> > > Is that 5% of those 95%?
> > 
> > Yes, it is indeed.
> 
> Thanks for navigating me. I have followup questions about
> /proc/sys/kernel/sched_rt_runtime_us / /proc/sys/kernel/sched_rt_period_us
> (a ratio, without CONGIG_RT_GROUP_SCHED)
> 
> - 0
>   - disables DL, (not RT, so they can monopolize a CPU)
> - 1
>   - DL tasks can monopolize CPU, SCHED_NORMAL have 5% thanks to
>     fair_server
> - 1-Δ
>   - SCHED_NORMAL tasks have
>     - (1-Δ)*5% on behalf of DL (above RT)
>     - Δ regularly (below RT)
> 
> Is this breakdown correct?

So, sched_rt_runtime_us/sched_rt_period_us only applies to admission
control (for DEADLINE tasks, including dl_servers). The actual
parameters controlling the dl_server for SCHED_NORMAL are under sched/
debug as per-cpu values, e.g.:

# grep . /sys/kernel/debug/sched/fair_server/cpu*/*
/sys/kernel/debug/sched/fair_server/cpu0/period:1000000000
/sys/kernel/debug/sched/fair_server/cpu0/runtime:50000000
/sys/kernel/debug/sched/fair_server/cpu1/period:1000000000
/sys/kernel/debug/sched/fair_server/cpu1/runtime:50000000
/sys/kernel/debug/sched/fair_server/cpu2/period:1000000000
/sys/kernel/debug/sched/fair_server/cpu2/runtime:50000000
/sys/kernel/debug/sched/fair_server/cpu3/period:1000000000
/sys/kernel/debug/sched/fair_server/cpu3/runtime:50000000
/sys/kernel/debug/sched/fair_server/cpu4/period:1000000000
/sys/kernel/debug/sched/fair_server/cpu4/runtime:50000000
/sys/kernel/debug/sched/fair_server/cpu5/period:1000000000
/sys/kernel/debug/sched/fair_server/cpu5/runtime:50000000
/sys/kernel/debug/sched/fair_server/cpu6/period:1000000000
/sys/kernel/debug/sched/fair_server/cpu6/runtime:50000000
/sys/kernel/debug/sched/fair_server/cpu7/period:1000000000
/sys/kernel/debug/sched/fair_server/cpu7/runtime:50000000

You can disable admission control by echoing -1 in sched_rt_runtime_us,
but still have the dl_server working for SCHED_NORMAL tasks. By
disabling admission control SCHED_DEADLINE can indeed monopolize CPU
(over subscription).

Best,
Juri