Re: [PATCH 1/2] drm/panthor: Replace sleep locks with spinlocks in fdinfo path

[Tvrtko Ursulin]

On 12/02/2025 16:33, Boris Brezillon wrote:
> On Tue, 11 Feb 2025 11:39:49 +0000
> Tvrtko Ursulin <tursulin@xxxxxxxxxxx> wrote:
>
> > On 10/02/2025 16:08, Adrián Larumbe wrote:
> > > Hi Tvrtko,
> >
> > Thanks!
> >
> > > [18153.770244] BUG: sleeping function called from invalid context at kernel/locking/mutex.c:562
> > > [18153.771059] in_atomic(): 1, irqs_disabled(): 0, non_block: 0, pid: 203412, name: cat
> > > [18153.771757] preempt_count: 1, expected: 0
> > > [18153.772164] RCU nest depth: 0, expected: 0
> > > [18153.772538] INFO: lockdep is turned off.
> > > [18153.772898] CPU: 4 UID: 0 PID: 203412 Comm: cat Tainted: G        W          6.14.0-rc1-panthor-next-rk3588-fdinfo+ #1
> > > [18153.772906] Tainted: [W]=WARN
> > > [18153.772908] Hardware name: Radxa ROCK 5B (DT)
> > > [18153.772911] Call trace:
> > > [18153.772913]  show_stack+0x24/0x38 (C)
> > > [18153.772927]  dump_stack_lvl+0x3c/0x98
> > > [18153.772935]  dump_stack+0x18/0x24
> > > [18153.772941]  __might_resched+0x298/0x2b0
> > > [18153.772948]  __might_sleep+0x6c/0xb0
> > > [18153.772953]  __mutex_lock_common+0x7c/0x1950
> > > [18153.772962]  mutex_lock_nested+0x38/0x50
> > > [18153.772969]  panthor_fdinfo_gather_group_samples+0x80/0x138 [panthor]
> > > [18153.773042]  panthor_show_fdinfo+0x80/0x228 [panthor]
> > > [18153.773109]  drm_show_fdinfo+0x1a4/0x1e0 [drm]
> > > [18153.773397]  seq_show+0x274/0x358
> > > [18153.773404]  seq_read_iter+0x1d4/0x630
> >
> > There is a mutex_lock literally in seq_read_iter.
> >
> > So colour me confused. What created the atomic context between then and
> > Panthor code?! I just don't see it.
>
> It's actually the xa_lock() we take when gathering fdinfo data that
> puts us in this atomic context. In other words, the fix is correct, but
> the explanation is wrong :-).

Huh I did not spot that when glancing over it yesterday or so, but 
anyway, excellent you figured it out! Now the commit message can be 
fixed and not mislead someone in the future to think mutexes cannot be 
used in fdinfo. Thanks for taking time to check it!

Regards,

Tvrtko