gianfar: Add I/O barriers when touching buffer descriptor ownership.

The hardware must not see that is given ownership of a buffer until it is
completely written, and when the driver receives ownership of a buffer,
it must ensure that any other reads to the buffer reflect its final
state.  Thus, I/O barriers are added where required.

Without this patch, I have observed GCC reordering the setting of
bdp->length and bdp->status in gfar_new_skb.  Hardware reordering
was also theoretically possible.

Signed-off-by: Scott Wood <scottwood@freescale.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

diff --git a/drivers/net/gianfar.c b/drivers/net/gianfar.c
index b666a0c..f5b3cba 100644 (file)
--- a/drivers/net/gianfar.c
+++ b/drivers/net/gianfar.c
@@ -1025,6 +1025,15 @@ static int gfar_start_xmit(struct sk_buff *skb, struct net_device *dev)
 
        dev->trans_start = jiffies;
 
+       /* The powerpc-specific eieio() is used, as wmb() has too strong
+        * semantics (it requires synchronization between cacheable and
+        * uncacheable mappings, which eieio doesn't provide and which we
+        * don't need), thus requiring a more expensive sync instruction.  At
+        * some point, the set of architecture-independent barrier functions
+        * should be expanded to include weaker barriers.
+        */
+
+       eieio();
        txbdp->status = status;
 
        /* If this was the last BD in the ring, the next one */
@@ -1301,6 +1310,7 @@ struct sk_buff * gfar_new_skb(struct net_device *dev, struct rxbd8 *bdp)
        bdp->length = 0;
 
        /* Mark the buffer empty */
+       eieio();
        bdp->status |= (RXBD_EMPTY | RXBD_INTERRUPT);
 
        return skb;
@@ -1484,6 +1494,7 @@ int gfar_clean_rx_ring(struct net_device *dev, int rx_work_limit)
        bdp = priv->cur_rx;
 
        while (!((bdp->status & RXBD_EMPTY) || (--rx_work_limit < 0))) {
+               rmb();
                skb = priv->rx_skbuff[priv->skb_currx];
 
                if (!(bdp->status &