Re: [PATCH net 09/10] mptcp: pm: avoid sleeping while holding rcu_read_lock

[Matthieu Baerts <matttbe@kernel.org>]

Hi Sebastian,

Thank you for the review!

On 01/06/2026 16:58, Sebastian Andrzej Siewior wrote:
> On 2026-06-01 13:10:05 [+1000], Matthieu Baerts (NGI0) wrote:
>> sk_stop_timer_sync() calls del_timer_sync(), which spin-waits for the
> There is no del_timer_sync()

Indeed, not sure why I put that one. I wanted to say timer_delete_sync()

>> timer callback to complete on non-RT kernels. But on PREEMPT_RT, it can
>> sleep. Sleeping inside an RCU read-side critical section might trigger a
>> lockdep splat.
> 
> It can not sleep. timer_delete_sync() does not sleep. It can block on
> spinlock_t and schedule but this is okay within a RCU read section.

Sorry, I got confused by the comment in __timer_delete_sync().

> Do you have a report for this?
No, sorry. I was exploring the code in timer.c after a related comment
from Sashiko [1]. I noticed the comment in __timer_delete_sync() about
being able to sleep on PREEMPT_RT, then saw that timer_delete_sync() was
called inside a rcu_read_lock(), and I thought it was wrong. I should
have tried to reproduce the warning on PREEMPT_RT, but I switched to the
next bug instead :-/

Sorry for the noise, let's drop this patch if it is not needed then.

[1] https://github.com/multipath-tcp/mptcp_net-next/issues/623

Cheers,
Matt
-- 
Sponsored by the NGI0 Core fund.