linux-kernel - RE: [RFC] RAS/CEC: Should cec_notifier() set MCE_HANDLED

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: <SJ1PR11MB60833BFA53B5E617C526828EFC772@SJ1PR11MB6083.namprd11.prod.outlook.com>
Date: Tue, 1 Oct 2024 18:24:19 +0000
From: "Luck, Tony" <tony.luck@...el.com>
To: "Meyer, Kyle" <kyle.meyer@....com>, "bp@...en8.de" <bp@...en8.de>
CC: "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>
Subject: RE: [RFC] RAS/CEC: Should cec_notifier() set MCE_HANDLED_CEC after a
 soft-offline?

> I noticed CEC should indicate whether it took action to log or handle an error
> by setting MCE_HANDLED_CEC (commit 1de08dc) and that EDAC and dev-mcelog should
> skip errors that have been processed by CEC (commit 23ba710).
>
> cec_notifier() does not set MCE_HANDLED_CEC when the offlining threshold
> is reached in cec_add_elem() because the return code is not zero. Is that
> intentional?

Kyle,

It seems a bit murky. You are right that cec_add_elem() appears to expect three
different actions from its caller based on the return value being <0, 0, >0. But
cec_notifier() only has two actions (0 and !0).

But I think this may be OK. The main purpose of CEC is to avoid over-reacting
to simple corrected memory errors. Many (most?) are due to particle bit flips and
no action is needed. So setting MCE_HANDLED_CEC for the case where CEC
counted the error, but took no action feels like the right thing to do.

Conversely, if action was taken (because this was an error that repeated
enough to hit the threshold) the we do want mcelog/EDAC to give additional
reporting.

-Tony