Re: [PATCH V5] x86 spinlock: Fix memory corruption on completing completions
From: Oleg Nesterov <oleg@redhat.com>
Date: 2015-02-15 16:19:53
Also in:
kvm, lkml
Well, I regret I mentioned the lack of barrier after enter_slowpath ;) On 02/15, Raghavendra K T wrote:
quoted hunk ↗ jump to hunk
@@ -46,7 +46,8 @@ static __always_inline bool static_key_false(struct static_key *key); static inline void __ticket_enter_slowpath(arch_spinlock_t *lock) { - set_bit(0, (volatile unsigned long *)&lock->tickets.tail); + set_bit(0, (volatile unsigned long *)&lock->tickets.head); + barrier(); }
Because this barrier() looks really confusing. Firsty, it is equally unneeded on x86. At the same time, it can not help. We need a memory barrier() between set_bit(SLOWPATH) and READ_ONCE(head) to avoid the race with spin_unlock(). So I think you should replace it with smp_mb__after_atomic() or remove it. Other than that I believe this version is correct. So I won't insist, this is cosmetic after all. Oleg.