ixgbe: performance tweaks
drop variables that had cache lines modified in simultaneous hot paths.
keep some variables modified on hot paths but make their storage per queue.
cache align DMA data buffer start addresses.
cache align (padding) some structures that end within a cacheline.
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Acked-by: Peter P Waskiewicz Jr <peter.p.waskiewicz.jr@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
diff --git a/drivers/net/ixgbe/ixgbe_main.c b/drivers/net/ixgbe/ixgbe_main.c
index e3dc68b..db05030 100644
--- a/drivers/net/ixgbe/ixgbe_main.c
+++ b/drivers/net/ixgbe/ixgbe_main.c
@@ -413,7 +413,7 @@
if (__netif_subqueue_stopped(netdev, tx_ring->queue_index) &&
!test_bit(__IXGBE_DOWN, &adapter->state)) {
netif_wake_subqueue(netdev, tx_ring->queue_index);
- ++adapter->restart_queue;
+ ++tx_ring->restart_queue;
}
}
@@ -624,7 +624,6 @@
/* It must be a TCP or UDP packet with a valid checksum */
skb->ip_summed = CHECKSUM_UNNECESSARY;
- adapter->hw_csum_rx_good++;
}
static inline void ixgbe_release_rx_desc(struct ixgbe_hw *hw,
@@ -681,14 +680,19 @@
if (!bi->skb) {
struct sk_buff *skb;
- skb = netdev_alloc_skb_ip_align(adapter->netdev,
- rx_ring->rx_buf_len);
+ /* netdev_alloc_skb reserves 32 bytes up front!! */
+ uint bufsz = rx_ring->rx_buf_len + SMP_CACHE_BYTES;
+ skb = netdev_alloc_skb(adapter->netdev, bufsz);
if (!skb) {
adapter->alloc_rx_buff_failed++;
goto no_buffers;
}
+ /* advance the data pointer to the next cache line */
+ skb_reserve(skb, (PTR_ALIGN(skb->data, SMP_CACHE_BYTES)
+ - skb->data));
+
bi->skb = skb;
bi->dma = pci_map_single(pdev, skb->data,
rx_ring->rx_buf_len,
@@ -801,8 +805,6 @@
hdr_info = le16_to_cpu(ixgbe_get_hdr_info(rx_desc));
len = (hdr_info & IXGBE_RXDADV_HDRBUFLEN_MASK) >>
IXGBE_RXDADV_HDRBUFLEN_SHIFT;
- if (hdr_info & IXGBE_RXDADV_SPH)
- adapter->rx_hdr_split++;
if (len > IXGBE_RX_HDR_SIZE)
len = IXGBE_RX_HDR_SIZE;
upper_len = le16_to_cpu(rx_desc->wb.upper.length);
@@ -812,7 +814,7 @@
cleaned = true;
skb = rx_buffer_info->skb;
- prefetch(skb->data - NET_IP_ALIGN);
+ prefetch(skb->data);
rx_buffer_info->skb = NULL;
if (rx_buffer_info->dma) {
@@ -884,7 +886,7 @@
skb->next = next_buffer->skb;
skb->next->prev = skb;
}
- adapter->non_eop_descs++;
+ rx_ring->non_eop_descs++;
goto next_desc;
}
@@ -4511,6 +4513,13 @@
adapter->rsc_total_flush = rsc_flush;
}
+ /* gather some stats to the adapter struct that are per queue */
+ for (i = 0; i < adapter->num_tx_queues; i++)
+ adapter->restart_queue += adapter->tx_ring[i].restart_queue;
+
+ for (i = 0; i < adapter->num_rx_queues; i++)
+ adapter->non_eop_descs += adapter->tx_ring[i].non_eop_descs;
+
adapter->stats.crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
for (i = 0; i < 8; i++) {
/* for packet buffers not used, the register should read 0 */
@@ -4893,14 +4902,12 @@
iph->daddr, 0,
IPPROTO_TCP,
0);
- adapter->hw_tso_ctxt++;
} else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
ipv6_hdr(skb)->payload_len = 0;
tcp_hdr(skb)->check =
~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
&ipv6_hdr(skb)->daddr,
0, IPPROTO_TCP, 0);
- adapter->hw_tso6_ctxt++;
}
i = tx_ring->next_to_use;
@@ -5019,7 +5026,6 @@
tx_buffer_info->time_stamp = jiffies;
tx_buffer_info->next_to_watch = i;
- adapter->hw_csum_tx_good++;
i++;
if (i == tx_ring->count)
i = 0;
@@ -5256,8 +5262,6 @@
static int __ixgbe_maybe_stop_tx(struct net_device *netdev,
struct ixgbe_ring *tx_ring, int size)
{
- struct ixgbe_adapter *adapter = netdev_priv(netdev);
-
netif_stop_subqueue(netdev, tx_ring->queue_index);
/* Herbert's original patch had:
* smp_mb__after_netif_stop_queue();
@@ -5271,7 +5275,7 @@
/* A reprieve! - use start_queue because it doesn't call schedule */
netif_start_subqueue(netdev, tx_ring->queue_index);
- ++adapter->restart_queue;
+ ++tx_ring->restart_queue;
return 0;
}