linux-kernel - Re: [PATCH V2] arch_topology: Clear LLC sibling when cacheinfo teardown

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: <20230316092910.doolw3xiuwwakile@bogus>
Date:   Thu, 16 Mar 2023 09:29:10 +0000
From:   Sudeep Holla <sudeep.holla@....com>
To:     Song Shuai <suagrfillet@...il.com>
Cc:     gregkh@...uxfoundation.org, rafael@...nel.org,
        conor.dooley@...rochip.com, ionela.voinescu@....com,
        Pierre.Gondois@....com, Sudeep Holla <sudeep.holla@....com>,
        linux-kernel@...r.kernel.org
Subject: Re: [PATCH V2] arch_topology: Clear LLC sibling when cacheinfo
 teardown

On Tue, Mar 14, 2023 at 03:53:45PM +0800, Song Shuai wrote:
> The teardown of CPUHP_AP_BASE_CACHEINFO_ONLINE now only invokes
> free_cache_attributes() to clear share_cpu_map of cacheinfo list.
> At the same time, clearing cpu_topology[].llc_sibling is
> called quite late at the teardown code in hotplug STARTING section.
>
> To avoid the incorrect LLC sibling masks generated, move its clearing
> right after free_cache_attributes().
>

Technically in terms of flow/timing this is correct. However I would like
to know if you are seeing any issues without this change ?

Technically, if a cpu is hotplugged out, the cacheinfo is reset first
and then the topology. Until the cpu is removes, the LLC info in the
topology is still valid. Also I am not sure if anything gets scheduled
and this LLC info is utilised once the teardown of CPUHP_AP_BASE_CACHEINFO_ONLINE
has started.

So I am trying to understand if we really need this change. Please let me
know if I am missing anything here.

-- 
Regards,
Sudeep