linux-kernel - Re: [PATCH v2] clocksource: Defer marking clocksources unstable to kthread

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: <87plir32as.ffs@tglx>
Date: Sat, 08 Mar 2025 17:38:51 +0100
From: Thomas Gleixner <tglx@...utronix.de>
To: paulmck@...nel.org, linux-kernel@...r.kernel.org
Cc: jstultz@...gle.com, sboyd@...nel.org, christian@...sel.eu,
 kernel-team@...a.com, Peter Zijlstra <peterz@...radead.org>, Yafang Shao
 <laoar.shao@...il.com>, Michal Koutný <mkoutny@...e.com>,
 Vincent Guittot
 <vincent.guittot@...aro.org>
Subject: Re: [PATCH v2] clocksource: Defer marking clocksources unstable to
 kthread

On Thu, Mar 06 2025 at 08:06, Paul E. McKenney wrote:
> The clocksource watchdog marks clocksources unstable from within a timer
> handler.  On x86, this marking involves an on_each_cpu_cond_mask(),
> which in turn invokes smp_call_function_many_cond(), which may not be
> invoked from a timer handler.  Doing so results in:
>
> WARNING: CPU: 3 PID: 0 at kernel/smp.c:815 smp_call_function_many_cond+0x46b/0x4c0
>
> Fix this by deferring the marking to the clocksource watchdog kthread.
> Note that marking unstable is already deferred, so deferring it a bit
> more should be just fine.

While this can be done, that's papering over the underlying problem,
which was introduced with:

  8722903cbb8f ("sched: Define sched_clock_irqtime as static key")

That added the static key switch, which is causing the problem. And
"fixing" this in the clocksource watchdog is incomplete because the same
problem exists during CPU hotplug when the TSC synchronization declares
the TSC unstable. It's the exactly same problem as was fixed via:

 6577e42a3e16 ("sched/clock: Fix up clear_sched_clock_stable()")

So as this got introduced in the 6.14 merge window, the proper fix is to
revert commit 8722903cbb8f and send it back to the drawing board. It was
clearly never tested with the various possibilities which invoke
mark_tsc*_unstable().

Thanks,

        tglx