Messages in this thread | | | Date | Thu, 7 Nov 2019 10:38:22 +0100 | From | Peter Zijlstra <> | Subject | Re: [patch 08/12] futex: Sanitize exit state handling |
| |
On Wed, Nov 06, 2019 at 10:55:42PM +0100, Thomas Gleixner wrote: > Instead of having a smp_mb() and an empty lock/unlock of task::pi_lock move > the state setting into to the lock section. > > Signed-off-by: Thomas Gleixner <tglx@linutronix.de> > --- > kernel/futex.c | 17 ++++++++++------- > 1 file changed, 10 insertions(+), 7 deletions(-) > > --- a/kernel/futex.c > +++ b/kernel/futex.c > @@ -3703,16 +3703,19 @@ void futex_exit_done(struct task_struct > > void futex_exit_release(struct task_struct *tsk) > { > - tsk->futex_state = FUTEX_STATE_EXITING; > - /* > - * Ensure that all new tsk->pi_lock acquisitions must observe > - * FUTEX_STATE_EXITING. Serializes against attach_to_pi_owner(). > - */ > - smp_mb(); > /* > - * Ensure that we must observe the pi_state in exit_pi_state_list(). > + * Switch the state to FUTEX_STATE_EXITING under tsk->pi_lock. > + * > + * This ensures that all subsequent checks of tsk->futex_state in > + * attach_to_pi_owner() must observe FUTEX_STATE_EXITING with > + * tsk->pi_lock held. > + * > + * It guarantees also that a pi_state which was queued right before > + * the state change under tsk->pi_lock by a concurrent waiter must > + * be observed in exit_pi_state_list(). > */ > raw_spin_lock_irq(&tsk->pi_lock); > + tsk->futex_state = FUTEX_STATE_EXITING; > raw_spin_unlock_irq(&tsk->pi_lock);
Right, much saner.
So this used to be:
exit_signals(tsk) /* sets PF_EXITING */ smp_mb(); raw_spin_unlock_wait(&tsk->pi_lock);
Which is in fact (possibly) faster than the new sane code, since unlock_wait() only has to wait for any current lock holder to complete.
However due to terrible semantics and implementation issues we got rid of *spin_unlock_wait() and well.. lets all forget this :-)
| |