Re: [PATCH v3 4/5] locking/rwsem: Enable direct rwsem lock handoff

[Waiman Long]

On 10/18/22 22:29, Hillf Danton wrote:
> On 18 Oct 2022 20:39:59 -0400 Waiman Long <longman@xxxxxxxxxx>
> > On 10/18/22 19:51, Hillf Danton wrote:
> > > On 18 Oct 2022 13:37:20 -0400 Waiman Long <longman@xxxxxxxxxx>
> > > > On 10/18/22 10:13, Mukesh Ojha wrote:
> > > > > On 10/18/2022 4:44 PM, Hillf Danton wrote:
> > > > > > On 17 Oct 2022 17:13:55 -0400 Waiman Long <longman@xxxxxxxxxx>
> > > > > > > @@ -1067,13 +1119,33 @@ rwsem_down_read_slowpath(struct rw_semaphore
> > > > > > >                 return sem;
> > > > > > >             }
> > > > > > >             adjustment += RWSEM_FLAG_WAITERS;
> > > > > > > +    } else if ((count & RWSEM_FLAG_HANDOFF) &&
> > > > > > > +          ((count & RWSEM_LOCK_MASK) == RWSEM_READER_BIAS)) {
> > > > > > Could a couple of CPUs go read slow path in parallel?
> > > > This is under wait_lock protection. So no parallel execution is possible.
> > > They individually add RWSEM_READER_BIAS to count before taking wait_lock,
> > > and the check for BIAS here does not cover the case of readers in parallel.
> > > Is this intended?
> > >
> > > Hillf
> > As I said in the patch description, the lock handoff can only be done if
> > we can be sure that there is no other active locks outstanding with the
> > handoff bit set. If at the time of the check, another reader come in and
> > adds its RWSEM_READER_BIAS, the check fail and the cpu will proceed to
> > put its waiter in the queue and begin sleeping. Hopefully, the last one
> > left will find that count has only its RWSEM_READER_BIAS and it can
> > start the handoff process.
> If handoff grants rwsem to a read waiter then the read fast path may revive.
I don't quite understand what you mean by "read fast path may revive".
> And at the time of the check, multiple readers do not break handoff IMO.

I am not saying that multiple readers will break handoff. They will just 
delay it until all their temporary RWSEM_READ_BIAS are taken off.

Cheers,
Longman