Fix synchronize_irq races with IRQ handler

As it is some callers of synchronize_irq rely on memory barriers
to provide synchronisation against the IRQ handlers.  For example,
the tg3 driver does

	tp->irq_sync = 1;
	smp_mb();
	synchronize_irq();

and then in the IRQ handler:

	if (!tp->irq_sync)
		netif_rx_schedule(dev, &tp->napi);

Unfortunately memory barriers only work well when they come in
pairs.  Because we don't actually have memory barriers on the
IRQ path, the memory barrier before the synchronize_irq() doesn't
actually protect us.

In particular, synchronize_irq() may return followed by the
result of netif_rx_schedule being made visible.

This patch (mostly written by Linus) fixes this by using spin
locks instead of memory barries on the synchronize_irq() path.
Signed-off-by: NHerbert Xu <herbert@gondor.apana.org.au>
Acked-by: NBenjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

kernel/irq/manage.c

@@ -29,12 +29,28 @@
 void synchronize_irq(unsigned int irq)
 {
 	struct irq_desc *desc = irq_desc + irq;
+	unsigned int status;
 
 	if (irq >= NR_IRQS)
 		return;
 
-	while (desc->status & IRQ_INPROGRESS)
-		cpu_relax();
+	do {
+		unsigned long flags;
+
+		/*
+		 * Wait until we're out of the critical section.  This might
+		 * give the wrong answer due to the lack of memory barriers.
+		 */
+		while (desc->status & IRQ_INPROGRESS)
+			cpu_relax();
+
+		/* Ok, that indicated we're done: double-check carefully. */
+		spin_lock_irqsave(&desc->lock, flags);
+		status = desc->status;
+		spin_unlock_irqrestore(&desc->lock, flags);
+
+		/* Oops, that failed? */
+	} while (status & IRQ_INPROGRESS);
 }
 EXPORT_SYMBOL(synchronize_irq);