On Mon, 20 Sep 2004 16:10:30 +0200
Harald Welte [off-list ref] wrote:

+static inline void natsemi_irq_enable(struct netdev_private *np)
+{
+	/* Enable interrupts by setting the interrupt mask. */
+	writel(DEFAULT_INTR, np->base_addr + IntrMask);
+	writel(1, np->base_addr + IntrEnable);
+	mb();
+}
+
+static inline void natsemi_irq_disable(struct netdev_private *np)
+{
+	writel(0, np->base_addr + IntrEnable);
+	mb();
+}

Kill the mb(), try using:

	readl(np->base_addr + IntrEnable);

in it's place.

This driver (before the NAPI patch) is a bit buggy, it does
no RX processing SMP locking.  So if one one cpu you're doing
RX interrupt processing, and on another cpu tx_timeout() runs
we're totally screwed.

This is not caused by Harald's patch, but there are implications
for his NAPI patch once it is fixed.