Messages in this thread |  | | Date | Mon, 3 Oct 2016 11:36:24 -0400 | From | Steven Rostedt <> | Subject | Re: [RFC][PATCH 4/4] futex: Rewrite FUTEX_UNLOCK_PI |
| |
On Mon, 03 Oct 2016 11:12:38 +0200 Peter Zijlstra <peterz@infradead.org> wrote:
> There's a number of 'interesting' problems with FUTEX_UNLOCK_PI, all > caused by holding hb->lock while doing the rt_mutex_unlock() > equivalient. > > This patch doesn't attempt to fix any of the actual problems, but > instead reworks the code to not hold hb->lock across the unlock, > paving the way to actually fix the problems later. > > The current reason we hold hb->lock over unlock is that it serializes > against FUTEX_LOCK_PI and avoids new waiters from coming in, this then > ensures the rt_mutex_next_owner() value is stable and can be written > into the user-space futex value before doing the unlock. Such that the > unlock will indeed end up at new_owner. > > This patch recognises that holding rt_mutex::wait_lock results in the > very same guarantee, no new waiters can come in while we hold that > lock -- after all, waiters would need this lock to queue themselves. > > It therefore restructures the code to keep rt_mutex::wait_lock held. > > This (of course) is not entirely straight forward either, see the > comment in rt_mutex_slowunlock(), doing the unlock itself might drop > wait_lock, letting new waiters in. To cure this > rt_mutex_futex_unlock() becomes a variant of rt_mutex_slowunlock() > that return -EAGAIN instead. This ensures the FUTEX_UNLOCK_PI code > aborts and restarts the entire operation. > > Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> > --- > kernel/futex.c | 63 +++++++++++++++++-------------- > kernel/locking/rtmutex.c | 81 ++++++++++++++++++++++++++++++++++++---- > kernel/locking/rtmutex_common.h | 4 - > 3 files changed, 110 insertions(+), 38 deletions(-) > > --- a/kernel/futex.c > +++ b/kernel/futex.c > @@ -1294,24 +1294,21 @@ static void mark_wake_futex(struct wake_ > static int wake_futex_pi(u32 __user *uaddr, u32 uval, struct futex_q *top_waiter, > struct futex_hash_bucket *hb) > { > - struct task_struct *new_owner; > struct futex_pi_state *pi_state = top_waiter->pi_state; > u32 uninitialized_var(curval), newval; > + struct task_struct *new_owner; > WAKE_Q(wake_q); > - bool deboost; > int ret = 0; > > - if (!pi_state) > - return -EINVAL; > + raw_spin_lock_irq(&pi_state->pi_mutex.wait_lock); > > + WARN_ON_ONCE(!atomic_inc_not_zero(&pi_state->refcount));
Don't we have a rule where WARN_ON() and BUG_ON() should never have "side effects"? That is, they should only check values, but their contents should not update values.
hence have:
ret = atomic_inc_not_zero(&pi_state->refcount); WARN_ON_ONCE(!ret);
> /* > - * If current does not own the pi_state then the futex is > - * inconsistent and user space fiddled with the futex value. > + * Now that we hold wait_lock, no new waiters can happen on the > + * rt_mutex and new owner is stable. Drop hb->lock. > */ > - if (pi_state->owner != current) > - return -EINVAL; > + spin_unlock(&hb->lock); >
Also, as Sebastian has said before, I believe this breaks rt's migrate disable code. As migrate disable and migrate_enable are a nop if preemption is disabled, thus if you hold a raw_spin_lock across a spin_unlock() when the migrate enable will be a nop, and the migrate_disable() will never stop.
-- Steve
> - raw_spin_lock_irq(&pi_state->pi_mutex.wait_lock); > new_owner = rt_mutex_next_owner(&pi_state->pi_mutex); > > /*
|  |