Re: [Xen-devel] spinlock wakeup

[Julien Grall]

Hi Boris,

On 24/02/17 16:15, Boris Ostrovsky wrote:
Looking at
void _spin_lock(spinlock_t *lock)
{
    spinlock_tickets_t tickets = SPINLOCK_TICKET_INC;
    LOCK_PROFILE_VAR;

    check_lock(&lock->debug);
    tickets.head_tail = arch_fetch_and_add(&lock->tickets.head_tail,
                                           tickets.head_tail);
    while ( tickets.tail != observe_head(&lock->tickets) )
    {
        LOCK_PROFILE_BLOCK;
        arch_lock_relax();
    }
    LOCK_PROFILE_GOT;
    preempt_disable();
    arch_lock_acquire_barrier();
}

If the lock is dropped during LOCK_PROFILE_BLOCK, isn't there a chance
that we are never woken up from arch_lock_relax()? (This would only be
an issue on ARM which IIUIC sleeps there).

The function arch_lock_signal will broadcast the event to all the 
processors and each of them will record the event. This event will be 
only cleared by a WFE (the implementation of arch_lock_acquire_barrier).

This means that if the event was happening during LOCK_PROFILE_BLOCK, 
the processor will not go to sleep and try again.

Cheers,

--
Julien Grall

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
https://lists.xen.org/xen-devel