Re: [PATCH -v2 4/6] locking, arch: Update spin_unlock_wait()

From: Martin Schwidefsky
Date: Fri May 27 2016 - 04:37:20 EST

On Thu, 26 May 2016 16:19:26 +0200
Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:

> This patch updates/fixes all spin_unlock_wait() implementations.
> The update is in semantics; where it previously was only a control
> dependency, we now upgrade to a full load-acquire to match the
> store-release from the spin_unlock() we waited on. This ensures that
> when spin_unlock_wait() returns, we're guaranteed to observe the full
> critical section we waited on.
> This fixes a number of spin_unlock_wait() users that (not
> unreasonably) rely on this.

All that is missing is an smp_rmb(), no?

> --- a/arch/s390/include/asm/spinlock.h
> +++ b/arch/s390/include/asm/spinlock.h
> @@ -95,8 +95,7 @@ static inline void arch_spin_unlock(arch
> static inline void arch_spin_unlock_wait(arch_spinlock_t *lock)
> {
> - while (arch_spin_is_locked(lock))
> - arch_spin_relax(lock);
> + smp_cond_load_acquire(&lock->lock, !VAL);
> }
> /*

This change adds the smp_rmb() at the end of the waiting loop, but
it also replaces arch_spin_relax() alias arch_lock_relax() with a
cpu_relax(). This is not good, these two functions do *very* different
things. cpu_relax() does an undirected yield with diagnose 0x44 but
only if the system is non-SMT. arch_lock_relax() does an additional
cpu_is_preempted() to test if the target cpu is running and does a
directed yield with diagnose 0x9c.

Why can't we just add the smp_rmb() to the arch_spin_unlock_wait()?

blue skies,

"Reality continues to ruin my life." - Calvin.