net: bcmgenet: fix TX reclaim accounting for fragments

The GENET driver supports SKB fragments, and succeeds in transmitting
them properly, but when reclaiming these transmitted fragments, we will
only update the count of free buffer descriptors by 1, even for SKBs
with fragments. This leads to the networking stack thinking it has more
room than the hardware has when pushing new SKBs, and backing off
consequently because we return NETDEV_TX_BUSY.

Fix this by accounting for the SKB nr_frags plus one (itself) and update
ring->free_bds accordingly with that value for each iteration loop in
__bcmgenet_tx_reclaim().

Fixes: 1c1008c793 ("net: bcmgenet: add main driver file")
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This commit is contained in:
Florian Fainelli 2014-09-22 11:54:42 -07:00 committed by David S. Miller
parent a35165ca10
commit 478a010c92

View file

@ -875,6 +875,7 @@ static void __bcmgenet_tx_reclaim(struct net_device *dev,
int last_tx_cn, last_c_index, num_tx_bds;
struct enet_cb *tx_cb_ptr;
struct netdev_queue *txq;
unsigned int bds_compl;
unsigned int c_index;
/* Compute how many buffers are transmitted since last xmit call */
@ -899,7 +900,9 @@ static void __bcmgenet_tx_reclaim(struct net_device *dev,
/* Reclaim transmitted buffers */
while (last_tx_cn-- > 0) {
tx_cb_ptr = ring->cbs + last_c_index;
bds_compl = 0;
if (tx_cb_ptr->skb) {
bds_compl = skb_shinfo(tx_cb_ptr->skb)->nr_frags + 1;
dev->stats.tx_bytes += tx_cb_ptr->skb->len;
dma_unmap_single(&dev->dev,
dma_unmap_addr(tx_cb_ptr, dma_addr),
@ -916,7 +919,7 @@ static void __bcmgenet_tx_reclaim(struct net_device *dev,
dma_unmap_addr_set(tx_cb_ptr, dma_addr, 0);
}
dev->stats.tx_packets++;
ring->free_bds += 1;
ring->free_bds += bds_compl;
last_c_index++;
last_c_index &= (num_tx_bds - 1);