On 13.03.20 09:55, Jan Beulich wrote:
On 13.03.2020 09:05, Juergen Gross wrote:@@ -199,10 +199,10 @@ unsigned long _spin_lock_irqsave(spinlock_t *lock) void _spin_unlock(spinlock_t *lock) { arch_lock_release_barrier(); - preempt_enable(); LOCK_PROFILE_REL; rel_lock(&lock->debug); add_sized(&lock->tickets.head, 1); + preempt_enable(); arch_lock_signal(); }arch_lock_signal() is a barrier on Arm, hence just like for patch 1 I wonder whether the insertion wouldn't better come after it.
Either way is fine for me. It should be noted that preemption is only relevant on the local cpu. So this is about trading lock state visibility against preemption disabled time, and I agree the visible time of the lock held should be minimized at higher priority than the preemption disabled time. I'll modify my patches accordingly, adding a note in this regard to the commit message. Juergen _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
