Paul Janzen <[EMAIL PROTECTED]> writes:

> In mv643xx_eth_start_xmit:
> [...]
>       spin_lock_irqsave(&mp->lock, flags);
> [...]
>               /* Since hardware can't handle unaligned fragments smaller
>                * than 9 bytes, if we find any, we linearize the skb
>                * and start again. */
> [...]
>                               skb_linearize(skb, GFP_ATOMIC);
> [...]
>
> which ends up calling kunmap_skb_frag(vaddr), which, when
> CONFIG_HIGHMEM=y, calls local_bh_enable with interrupts off.

A patch for this problem is enclosed.  I believe it also solves a
potential deadlock if skb_linearize() returns -ENOMEM. 

Signed-off-by: Paul Janzen <[EMAIL PROTECTED]>

--- a/drivers/net/mv643xx_eth.c 2005-12-24 15:47:48.000000000 -0800
+++ b/drivers/net/mv643xx_eth.c 2006-01-08 20:30:25.000000000 -0800
@@ -1093,6 +1093,27 @@ static int mv643xx_poll(struct net_devic
 }
 #endif
 
+/* Hardware can't handle unaligned fragments smaller than 9 bytes.
+ * This helper function detects that case.  When I've seen it, it's
+ * always been the first frag (probably near the end of the page), but
+ * we check all frags to be safe.
+ */
+
+static inline unsigned int has_tiny_unaligned_frags(struct sk_buff *skb)
+{
+        unsigned int frag;
+        skb_frag_t *fragp;
+
+        for (frag = 0; frag < skb_shinfo(skb)->nr_frags; frag++) {
+                fragp = &skb_shinfo(skb)->frags[frag];
+                if (fragp->size <= 8 && fragp->page_offset & 0x7) 
+                        return 1;
+                
+        }
+        return 0;
+}
+
+
 /*
  * mv643xx_eth_start_xmit
  *
@@ -1136,12 +1157,22 @@ static int mv643xx_eth_start_xmit(struct
                return 1;
        }
 
+        if (has_tiny_unaligned_frags(skb)) {
+                if ((skb_linearize(skb, GFP_ATOMIC) != 0) 
+                    || has_tiny_unaligned_frags(skb)) { 
+                        stats->tx_dropped++;
+                        printk(KERN_ERR 
+                               "%s: failed to linearize tiny unaligned 
fragment\n",
+                                dev->name);
+                        return 1;
+                }
+        }
+
        spin_lock_irqsave(&mp->lock, flags);
 
        /* Update packet info data structure -- DMA owned, first last */
 #ifdef MV643XX_CHECKSUM_OFFLOAD_TX
        if (!skb_shinfo(skb)->nr_frags) {
-linear:
                if (skb->ip_summed != CHECKSUM_HW) {
                        /* Errata BTS #50, IHL must be 5 if no HW checksum */
                        pkt_info.cmd_sts = ETH_TX_ENABLE_INTERRUPT |
@@ -1183,26 +1214,6 @@ linear:
        } else {
                unsigned int frag;
 
-               /* Since hardware can't handle unaligned fragments smaller
-                * than 9 bytes, if we find any, we linearize the skb
-                * and start again.  When I've seen it, it's always been
-                * the first frag (probably near the end of the page),
-                * but we check all frags to be safe.
-                */
-               for (frag = 0; frag < skb_shinfo(skb)->nr_frags; frag++) {
-                       skb_frag_t *fragp;
-
-                       fragp = &skb_shinfo(skb)->frags[frag];
-                       if (fragp->size <= 8 && fragp->page_offset & 0x7) {
-                               skb_linearize(skb, GFP_ATOMIC);
-                               printk(KERN_DEBUG "%s: unaligned tiny fragment"
-                                               "%d of %d, fixed\n",
-                                               dev->name, frag,
-                                               skb_shinfo(skb)->nr_frags);
-                               goto linear;
-                       }
-               }
-
                /* first frag which is skb header */
                pkt_info.byte_cnt = skb_headlen(skb);
                pkt_info.buf_ptr = dma_map_single(NULL, skb->data,


-- Paul
-
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to [EMAIL PROTECTED]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Reply via email to