linux-kernel - Re: [PATCH 0/2] Fix imbalance issue when balancing fork

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>] [day] [month] [year] [list]

Message-ID: <D25EB7EA-41C7-488A-989D-D7EA9BD6F5EA@os.amperecomputing.com>
Date: Tue, 15 Jul 2025 19:14:46 +0000
From: Shubhang Kaushik Prasanna Kumar <shkaushik@...erecomputing.com>
To: Adam Li OS <adamli@...amperecomputing.com>
CC: "bsegall@...gle.com" <bsegall@...gle.com>, "cl@...ux.com" <cl@...ux.com>,
	"dietmar.eggemann@....com" <dietmar.eggemann@....com>,
	"juri.lelli@...hat.com" <juri.lelli@...hat.com>,
	"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
	"mgorman@...e.de" <mgorman@...e.de>, "mingo@...hat.com" <mingo@...hat.com>,
	Open Source Submission <patches@...erecomputing.com>, "peterz@...radead.org"
	<peterz@...radead.org>, "rostedt@...dmis.org" <rostedt@...dmis.org>,
	"vincent.guittot@...aro.org" <vincent.guittot@...aro.org>,
	"vschneid@...hat.com" <vschneid@...hat.com>
Subject: Re: [PATCH 0/2] Fix imbalance issue when balancing fork

This patch-set tackles a significant task placement imbalance encountered under restricted CPU affinity, particularly on Ampere Altra systems. Our performance tests reveal that the first patch demonstrably improves task placement by 20-30% in scenarios where the number of threads increasingly outnumbers the available cores. However, the second patch unexpectedly introduces regressions when tested against higher thread count scenarios, requiring further investigation.

In light of these findings and the potential regression, the following questions arise:
 Q. Has the regression been investigated to determine if it is specific to two-level scheduling domains or manifests more broadly?
 Q. Were function-level latencies profiled to pinpoint the exact source of the regression within the code path introduced by the second patch?
 Q. How do these patches influence performance and task placement on systems with linear topologies, and do they exhibit more consistent behavior in those configurations?

- Shubhang