linux-kernel - Re: [PATCH 07/10] sched/fair: Provide can_migrate_task

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]

Date:   Thu, 1 Nov 2018 11:16:37 +0000
From:   Valentin Schneider <valentin.schneider@....com>
To:     Peter Zijlstra <peterz@...radead.org>
Cc:     Steven Sistare <steven.sistare@...cle.com>, mingo@...hat.com,
        subhra.mazumdar@...cle.com, dhaval.giani@...cle.com,
        rohit.k.jain@...cle.com, daniel.m.jordan@...cle.com,
        pavel.tatashin@...rosoft.com, matt@...eblueprint.co.uk,
        umgwanakikbuti@...il.com, riel@...hat.com, jbacik@...com,
        juri.lelli@...hat.com, linux-kernel@...r.kernel.org
Subject: Re: [PATCH 07/10] sched/fair: Provide can_migrate_task_llc

On 31/10/2018 19:14, Peter Zijlstra wrote:
> On Mon, Oct 29, 2018 at 07:34:50PM +0000, Valentin Schneider wrote:
>> On a sidenote, I find it a bit odd that the exec_start threshold depends on
>> sysctl_sched_migration_cost, which to me is more about idle_balance() cost
>> than "how long does it take for a previously run task to go cache cold".
> 
> A long long (think 2.6.20 long ago) there was code that did boot-time
> measurement of the various cache topology costs and that threshold was
> related to that.
> 
> That code got killed because of boot to boot variance and dubious
> benefits.
> 
> The migration cost is what it was replaced with as a single measure.
> 
> migration cost was then later abused in the newidle balance because it
> was over eager.


Thanks! I tried following the blame rabbit hole but got a bit lost along
the way.

> Ideally we'd get rid of it there, because we've now got
> that much more elaborate accounting, but Rohit tried and found some
> regression because of that.
> 
> Maybe we should remove it from newidle anyway.
> 

Well, if we ditch idle_balance() in favor of stealing, that would "solve"
it...