mv643xx_eth: rework receive skb cache alignment
On the platforms that mv643xx_eth is used on, the manual skb->data alignment logic in mv643xx_eth can be simplified, as the only case we need to handle is where NET_SKB_PAD is not a multiple of the cache line size. If this is the case, the extra padding we need can be computed at compile time, while if NET_SKB_PAD _is_ a multiple of the cache line size, the code can be optimised out entirely. Signed-off-by: Lennert Buytenhek <buytenh@marvell.com> Signed-off-by: David S. Miller <davem@davemloft.net>
This commit is contained in:
parent
becfad979d
commit
7fd96ce47f
|
@ -174,6 +174,7 @@ static char mv643xx_eth_driver_version[] = "1.4";
|
|||
*/
|
||||
#define DEFAULT_RX_QUEUE_SIZE 128
|
||||
#define DEFAULT_TX_QUEUE_SIZE 256
|
||||
#define SKB_DMA_REALIGN ((PAGE_SIZE - NET_SKB_PAD) % SMP_CACHE_BYTES)
|
||||
|
||||
|
||||
/*
|
||||
|
@ -649,23 +650,20 @@ static int rxq_refill(struct rx_queue *rxq, int budget)
|
|||
refilled = 0;
|
||||
while (refilled < budget && rxq->rx_desc_count < rxq->rx_ring_size) {
|
||||
struct sk_buff *skb;
|
||||
int unaligned;
|
||||
int rx;
|
||||
struct rx_desc *rx_desc;
|
||||
|
||||
skb = __skb_dequeue(&mp->rx_recycle);
|
||||
if (skb == NULL)
|
||||
skb = dev_alloc_skb(mp->skb_size +
|
||||
dma_get_cache_alignment() - 1);
|
||||
skb = dev_alloc_skb(mp->skb_size);
|
||||
|
||||
if (skb == NULL) {
|
||||
mp->oom = 1;
|
||||
goto oom;
|
||||
}
|
||||
|
||||
unaligned = (u32)skb->data & (dma_get_cache_alignment() - 1);
|
||||
if (unaligned)
|
||||
skb_reserve(skb, dma_get_cache_alignment() - unaligned);
|
||||
if (SKB_DMA_REALIGN)
|
||||
skb_reserve(skb, SKB_DMA_REALIGN);
|
||||
|
||||
refilled++;
|
||||
rxq->rx_desc_count++;
|
||||
|
@ -964,8 +962,7 @@ static int txq_reclaim(struct tx_queue *txq, int budget, int force)
|
|||
if (skb != NULL) {
|
||||
if (skb_queue_len(&mp->rx_recycle) <
|
||||
mp->rx_ring_size &&
|
||||
skb_recycle_check(skb, mp->skb_size +
|
||||
dma_get_cache_alignment() - 1))
|
||||
skb_recycle_check(skb, mp->skb_size))
|
||||
__skb_queue_head(&mp->rx_recycle, skb);
|
||||
else
|
||||
dev_kfree_skb(skb);
|
||||
|
@ -2336,6 +2333,14 @@ static void mv643xx_eth_recalc_skb_size(struct mv643xx_eth_private *mp)
|
|||
* size field are ignored by the hardware.
|
||||
*/
|
||||
mp->skb_size = (skb_size + 7) & ~7;
|
||||
|
||||
/*
|
||||
* If NET_SKB_PAD is smaller than a cache line,
|
||||
* netdev_alloc_skb() will cause skb->data to be misaligned
|
||||
* to a cache line boundary. If this is the case, include
|
||||
* some extra space to allow re-aligning the data area.
|
||||
*/
|
||||
mp->skb_size += SKB_DMA_REALIGN;
|
||||
}
|
||||
|
||||
static int mv643xx_eth_open(struct net_device *dev)
|
||||
|
|
Loading…
Reference in a new issue