Re: [PATCH] lockdep: Speed up lockdep_unregister_key() with expedited RCU synchronization

[Boqun Feng <boqun.feng@gmail.com>]

On Wed, Jul 09, 2025 at 09:57:44AM -0400, Waiman Long wrote:
> On 7/9/25 6:00 AM, Breno Leitao wrote:
> > Hello Waiman, Boqun,
> > 
> > On Fri, Mar 21, 2025 at 02:30:49AM -0700, Breno Leitao wrote:
> > > lockdep_unregister_key() is called from critical code paths, including
> > > sections where rtnl_lock() is held. For example, when replacing a qdisc
> > > in a network device, network egress traffic is disabled while
> > > __qdisc_destroy() is called for every network queue.
> > > 
> > > If lockdep is enabled, __qdisc_destroy() calls lockdep_unregister_key(),
> > > which gets blocked waiting for synchronize_rcu() to complete.
> > > 
> > > For example, a simple tc command to replace a qdisc could take 13
> > > seconds:
> > > 
> > >    # time /usr/sbin/tc qdisc replace dev eth0 root handle 0x1: mq
> > >      real    0m13.195s
> > >      user    0m0.001s
> > >      sys     0m2.746s
> > > 
> > > During this time, network egress is completely frozen while waiting for
> > > RCU synchronization.
> > > 
> > > Use synchronize_rcu_expedited() instead to minimize the impact on
> > > critical operations like network connectivity changes.
> > > 
> > > This improves 10x the function call to tc, when replacing the qdisc for
> > > a network card.
> > > 
> > >     # time /usr/sbin/tc qdisc replace dev eth0 root handle 0x1: mq
> > >       real     0m1.789s
> > >       user     0m0.000s
> > >       sys      0m1.613s
> > Can I have this landed as a workaround for the problem above, while
> > hazard pointers doesn't get merged?
> > 
> > This is affecting some systems that runs the Linus' upstream kernel with
> > some debug flags enabled, and I would like to have they unblocked.
> > 
> > Once hazard pointer lands, this will be reverted. Is this a fair
> > approach?
> > 
> > Thanks for your help,
> > --breno
> 
> I am fine with this patch going in as a temporary workaround. Boqun, what do
> you think?
> 

Seems good to me. We can add a "// TODO" comment to call out it's a
temporary workaround and should be replaced. Although, I believe Peter
had some some concern about IPI, I would like to get his opinion on
using synchronize_rcu_expedited() as a temporary solution. Peter?

Regards,
Boqun

> Cheers,
> Longman
>