lkml.org 
[lkml]   [2017]   [Mar]   [7]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH -v5 10/14] futex: Pull rt_mutex_futex_unlock() out from under hb->lock
On Tue, Mar 07, 2017 at 03:08:17PM +0100, Thomas Gleixner wrote:
> On Sat, 4 Mar 2017, Peter Zijlstra wrote:
> > @@ -1035,6 +1037,9 @@ static int attach_to_pi_state(u32 __user
> > * has dropped the hb->lock in between queue_me() and unqueue_me_pi(),
> > * which in turn means that futex_lock_pi() still has a reference on
> > * our pi_state.
> > + *
> > + * IOW, we cannot race against the unlocked put_pi_state() in
> > + * futex_unlock_pi().
>
> That 'IOW' made my head spin for a while. I rather prefer to spell it out
> more explicitely:
>
> * The waiter holding a reference on @pi_state protects also
> * against the unlocked put_pi_state() in futex_unlock_pi(),
> * futex_lock_pi() and futex_wait_requeue_pi() as it cannot go to 0
> * and consequentely free pi state before we can take a reference
> * ourself.

Right you are. After staring at this for too damn long one tends to
forget what 'obvious' means.

>
> > */
> > WARN_ON(!atomic_read(&pi_state->refcount));
> >
> > @@ -1378,47 +1383,33 @@ static void mark_wake_futex(struct wake_
> > smp_store_release(&q->lock_ptr, NULL);
> > }
> >
> > -static int wake_futex_pi(u32 __user *uaddr, u32 uval, struct futex_q *top_waiter,
> > - struct futex_hash_bucket *hb)
>
> Please add a comment, that the caller must hold a reference on @pi_state

Will do.

> > +static int wake_futex_pi(u32 __user *uaddr, u32 uval, struct futex_pi_state *pi_state)
> > {
> > u32 uninitialized_var(curval), newval;
> > + struct task_struct *new_owner;
> > + bool deboost = false;
> > DEFINE_WAKE_Q(wake_q);
> > int ret = 0;
> >
> > raw_spin_lock_irq(&pi_state->pi_mutex.wait_lock);
> > new_owner = rt_mutex_next_owner(&pi_state->pi_mutex);
> > if (!new_owner) {
> > + /*
> > + * Since we held neither hb->lock nor wait_lock when coming
> > + * into this function, we could have raced with futex_lock_pi()
> > + * such that it will have removed the waiter that brought us
> > + * here.
>
> Hmm. That's not entirely correct. There are two cases:
>
> lock_pi()
> queue_me() <- Makes it visible as waiter in the hash bucket
> unlock(hb->lock)
>
> [1]
>
> rtmutex_futex_lock()
>
> [2]
>
> lock(hb->lock)
>
> Both [1] and [2] are valid reasons why the top waiter is not a rtmutex
> waiter.

Correct, I've even drawn similar state pictures elsewhere in this
series. I'll update.

\
 
 \ /
  Last update: 2017-03-07 19:45    [W:0.180 / U:0.104 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site