linux-kernel - Re: [RFC PATCH 3/3] cgroup: make css_rstat

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: <6u7ccequ5ye3e4iqblcdeqsigindo3xjpsvkdb6hyaw7cpjddc@u2ujv7ymlxc6>
Date: Thu, 1 May 2025 15:10:20 -0700
From: Shakeel Butt <shakeel.butt@...ux.dev>
To: Yosry Ahmed <yosry.ahmed@...ux.dev>
Cc: Tejun Heo <tj@...nel.org>, Andrew Morton <akpm@...ux-foundation.org>, 
	Alexei Starovoitov <ast@...nel.org>, Johannes Weiner <hannes@...xchg.org>, 
	Michal Hocko <mhocko@...nel.org>, Roman Gushchin <roman.gushchin@...ux.dev>, 
	Muchun Song <muchun.song@...ux.dev>, Michal Koutný <mkoutny@...e.com>, 
	Vlastimil Babka <vbabka@...e.cz>, Sebastian Andrzej Siewior <bigeasy@...utronix.de>, 
	JP Kobryn <inwardvessel@...il.com>, bpf@...r.kernel.org, linux-mm@...ck.org, cgroups@...r.kernel.org, 
	linux-kernel@...r.kernel.org, Meta kernel team <kernel-team@...a.com>
Subject: Re: [RFC PATCH 3/3] cgroup: make css_rstat_updated nmi safe

On Wed, Apr 30, 2025 at 06:14:28AM -0700, Yosry Ahmed wrote:
[...]
> > +
> > +	if (!_css_rstat_cpu_trylock(css, cpu, &flags)) {
> 
> 
> IIUC this trylock will only fail if a BPF program runs in NMI context
> and tries to update cgroup stats, interrupting a context that is already
> holding the lock (i.e. updating or flushing stats).
> 

Correct (though note that flushing side can be on a different CPU).

> How often does this happen in practice tho? Is it worth the complexity?

This is about correctness, so even a chance of occurance need the
solution.

> 
> I wonder if it's better if we make css_rstat_updated() inherently
> lockless instead.
> 
> What if css_rstat_updated() always just adds to a lockless tree,

Here I assume you meant lockless list instead of tree.

> and we
> defer constructing the proper tree to the flushing side? This should
> make updates generally faster and avoids locking or disabling interrupts
> in the fast path. We essentially push more work to the flushing side.
> 
> We may be able to consolidate some of the code too if all the logic
> manipulating the tree is on the flushing side.
> 
> WDYT? Am I missing something here?
> 

Yes this can be done but I don't think we need to tie that to current
series. I think we can start with lockless in the nmi context and then
iteratively make css_rstat_updated() lockless for all contexts.