]> www.pilppa.org Git - linux-2.6-omap-h63xx.git/commitdiff
mv643xx_eth: enforce multiple-of-8-bytes receive buffer size restriction
authorLennert Buytenhek <buytenh@wantstofly.org>
Sun, 24 Aug 2008 01:00:20 +0000 (03:00 +0200)
committerLennert Buytenhek <buytenh@marvell.com>
Sun, 24 Aug 2008 01:33:44 +0000 (03:33 +0200)
The mv643xx_eth hardware ignores the lower three bits of the buffer
size field in receive descriptors, causing the reception of full-sized
packets to fail at some MTUs.  Fix this by rounding the size of
allocated receive buffers up to a multiple of eight bytes.

While we are at it, add a bit of extra space to each receive buffer so
that we can handle multiple vlan tags on ingress.

Signed-off-by: Lennert Buytenhek <buytenh@marvell.com>
drivers/net/mv643xx_eth.c

index e33dfc0165f610cb940fd9d792ae2e5cb041ce91..a02a5f4a02949c3f8dbf38849391ce30b3715097 100644 (file)
@@ -474,11 +474,19 @@ static void rxq_refill(struct rx_queue *rxq)
                /*
                 * Reserve 2+14 bytes for an ethernet header (the
                 * hardware automatically prepends 2 bytes of dummy
-                * data to each received packet), 4 bytes for a VLAN
-                * header, and 4 bytes for the trailing FCS -- 24
-                * bytes total.
+                * data to each received packet), 16 bytes for up to
+                * four VLAN tags, and 4 bytes for the trailing FCS
+                * -- 36 bytes total.
                 */
-               skb_size = mp->dev->mtu + 24;
+               skb_size = mp->dev->mtu + 36;
+
+               /*
+                * Make sure that the skb size is a multiple of 8
+                * bytes, as the lower three bits of the receive
+                * descriptor's buffer size field are ignored by
+                * the hardware.
+                */
+               skb_size = (skb_size + 7) & ~7;
 
                skb = dev_alloc_skb(skb_size + dma_get_cache_alignment() - 1);
                if (skb == NULL)
@@ -552,7 +560,7 @@ static int rxq_process(struct rx_queue *rxq, int budget)
                spin_unlock_irqrestore(&mp->lock, flags);
 
                dma_unmap_single(NULL, rx_desc->buf_ptr + 2,
-                                mp->dev->mtu + 24, DMA_FROM_DEVICE);
+                                rx_desc->buf_size, DMA_FROM_DEVICE);
                rxq->rx_desc_count--;
                rx++;