Re: [PATCH] lglock: Use spinlock_t instead of arch_spinlock_t

From: Peter Zijlstra <peterz@infradead.org>
To: Daniel Wagner <daniel.wagner@bmw-carit.de>
Cc: Jeff Layton <jlayton@poochiereds.net>,
	Ingo Molnar <mingo@redhat.com>,
	Alexander Viro <viro@zeniv.linux.org.uk>,
	"J. Bruce Fields" <bfields@fieldses.org>,
	Thomas Gleixner <tglx@linutronix.de>,
	Andi Kleen <andi@firstfloor.org>,
	linux-fsdevel@vger.kernel.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH] lglock: Use spinlock_t instead of arch_spinlock_t
Date: Thu, 26 Mar 2015 17:03:25 +0100	[thread overview]
Message-ID: <20150326160325.GA21418@twins.programming.kicks-ass.net> (raw)
In-Reply-To: <1427382128-12541-1-git-send-email-daniel.wagner@bmw-carit.de>

On Thu, Mar 26, 2015 at 04:02:08PM +0100, Daniel Wagner wrote:
> @@ -67,9 +67,9 @@ void lg_global_lock(struct lglock *lg)
>  	preempt_disable();
>  	lock_acquire_exclusive(&lg->lock_dep_map, 0, 0, NULL, _RET_IP_);
>  	for_each_possible_cpu(i) {
> -		arch_spinlock_t *lock;
> +		spinlock_t *lock;
>  		lock = per_cpu_ptr(lg->lock, i);
> -		arch_spin_lock(lock);
> +		spin_lock(lock);
>  	}
>  }

Nope, that'll blow up in two separate places.

One: lockdep, it can only track a limited number of held locks, and it
will further report a recursion warning on the 2nd cpu.

Second: preempt_count_add(), spin_lock() does preempt_disable(), with
enough CPUs you'll overflow the preempt counter (255).