diff options
author | Ayaz Abdulla <aabdulla@nvidia.com> | 2007-01-21 18:10:37 -0500 |
---|---|---|
committer | Jeff Garzik <jeff@garzik.org> | 2007-02-05 16:58:48 -0500 |
commit | 86b22b0dfbf462e6ed75e54fc83575dae01e3c69 (patch) | |
tree | 8a6efa296e3184acc0cdf8ce9bc72a9fac1e4913 /drivers/net/forcedeth.c | |
parent | 658f648ad1c2876e0ce5401e087d2d21d0262441 (diff) |
forcedeth: optimized routines
This patch breaks up the routines into two versions, one for legacy
descriptor versions (ver 1 and ver 2) and one for desc ver 3. This will
make the new desc functions more leaner and further reductions will be
made in next few patches.
Signed-Off-By: Ayaz Abdulla <aabdulla@nvidia.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
Diffstat (limited to 'drivers/net/forcedeth.c')
-rw-r--r-- | drivers/net/forcedeth.c | 634 |
1 files changed, 501 insertions, 133 deletions
diff --git a/drivers/net/forcedeth.c b/drivers/net/forcedeth.c index 0fc078625090..f28ae12d8569 100644 --- a/drivers/net/forcedeth.c +++ b/drivers/net/forcedeth.c | |||
@@ -1307,50 +1307,57 @@ static struct net_device_stats *nv_get_stats(struct net_device *dev) | |||
1307 | static int nv_alloc_rx(struct net_device *dev) | 1307 | static int nv_alloc_rx(struct net_device *dev) |
1308 | { | 1308 | { |
1309 | struct fe_priv *np = netdev_priv(dev); | 1309 | struct fe_priv *np = netdev_priv(dev); |
1310 | union ring_type less_rx; | 1310 | struct ring_desc* less_rx; |
1311 | 1311 | ||
1312 | if (np->desc_ver == DESC_VER_1 || np->desc_ver == DESC_VER_2) { | 1312 | less_rx = np->get_rx.orig; |
1313 | less_rx.orig = np->get_rx.orig; | 1313 | if (less_rx-- == np->first_rx.orig) |
1314 | if (less_rx.orig-- == np->first_rx.orig) | 1314 | less_rx = np->last_rx.orig; |
1315 | less_rx.orig = np->last_rx.orig; | ||
1316 | } else { | ||
1317 | less_rx.ex = np->get_rx.ex; | ||
1318 | if (less_rx.ex-- == np->first_rx.ex) | ||
1319 | less_rx.ex = np->last_rx.ex; | ||
1320 | } | ||
1321 | 1315 | ||
1322 | while (1) { | 1316 | while (np->put_rx.orig != less_rx) { |
1323 | struct sk_buff *skb; | 1317 | struct sk_buff *skb = dev_alloc_skb(np->rx_buf_sz + NV_RX_ALLOC_PAD); |
1324 | 1318 | if (skb) { | |
1325 | if (np->desc_ver == DESC_VER_1 || np->desc_ver == DESC_VER_2) { | 1319 | skb->dev = dev; |
1326 | if (np->put_rx.orig == less_rx.orig) | 1320 | np->put_rx_ctx->skb = skb; |
1327 | break; | 1321 | np->put_rx_ctx->dma = pci_map_single(np->pci_dev, skb->data, |
1322 | skb->end-skb->data, PCI_DMA_FROMDEVICE); | ||
1323 | np->put_rx_ctx->dma_len = skb->end-skb->data; | ||
1324 | np->put_rx.orig->buf = cpu_to_le32(np->put_rx_ctx->dma); | ||
1325 | wmb(); | ||
1326 | np->put_rx.orig->flaglen = cpu_to_le32(np->rx_buf_sz | NV_RX_AVAIL); | ||
1327 | if (np->put_rx.orig++ == np->last_rx.orig) | ||
1328 | np->put_rx.orig = np->first_rx.orig; | ||
1329 | if (np->put_rx_ctx++ == np->last_rx_ctx) | ||
1330 | np->put_rx_ctx = np->first_rx_ctx; | ||
1328 | } else { | 1331 | } else { |
1329 | if (np->put_rx.ex == less_rx.ex) | 1332 | return 1; |
1330 | break; | ||
1331 | } | 1333 | } |
1334 | } | ||
1335 | return 0; | ||
1336 | } | ||
1337 | |||
1338 | static int nv_alloc_rx_optimized(struct net_device *dev) | ||
1339 | { | ||
1340 | struct fe_priv *np = netdev_priv(dev); | ||
1341 | struct ring_desc_ex* less_rx; | ||
1342 | |||
1343 | less_rx = np->get_rx.ex; | ||
1344 | if (less_rx-- == np->first_rx.ex) | ||
1345 | less_rx = np->last_rx.ex; | ||
1332 | 1346 | ||
1333 | skb = dev_alloc_skb(np->rx_buf_sz + NV_RX_ALLOC_PAD); | 1347 | while (np->put_rx.ex != less_rx) { |
1348 | struct sk_buff *skb = dev_alloc_skb(np->rx_buf_sz + NV_RX_ALLOC_PAD); | ||
1334 | if (skb) { | 1349 | if (skb) { |
1335 | skb->dev = dev; | 1350 | skb->dev = dev; |
1336 | np->put_rx_ctx->skb = skb; | 1351 | np->put_rx_ctx->skb = skb; |
1337 | np->put_rx_ctx->dma = pci_map_single(np->pci_dev, skb->data, | 1352 | np->put_rx_ctx->dma = pci_map_single(np->pci_dev, skb->data, |
1338 | skb->end-skb->data, PCI_DMA_FROMDEVICE); | 1353 | skb->end-skb->data, PCI_DMA_FROMDEVICE); |
1339 | np->put_rx_ctx->dma_len = skb->end-skb->data; | 1354 | np->put_rx_ctx->dma_len = skb->end-skb->data; |
1340 | if (np->desc_ver == DESC_VER_1 || np->desc_ver == DESC_VER_2) { | 1355 | np->put_rx.ex->bufhigh = cpu_to_le64(np->put_rx_ctx->dma) >> 32; |
1341 | np->put_rx.orig->buf = cpu_to_le32(np->put_rx_ctx->dma); | 1356 | np->put_rx.ex->buflow = cpu_to_le64(np->put_rx_ctx->dma) & 0x0FFFFFFFF; |
1342 | wmb(); | 1357 | wmb(); |
1343 | np->put_rx.orig->flaglen = cpu_to_le32(np->rx_buf_sz | NV_RX_AVAIL); | 1358 | np->put_rx.ex->flaglen = cpu_to_le32(np->rx_buf_sz | NV_RX2_AVAIL); |
1344 | if (np->put_rx.orig++ == np->last_rx.orig) | 1359 | if (np->put_rx.ex++ == np->last_rx.ex) |
1345 | np->put_rx.orig = np->first_rx.orig; | 1360 | np->put_rx.ex = np->first_rx.ex; |
1346 | } else { | ||
1347 | np->put_rx.ex->bufhigh = cpu_to_le64(np->put_rx_ctx->dma) >> 32; | ||
1348 | np->put_rx.ex->buflow = cpu_to_le64(np->put_rx_ctx->dma) & 0x0FFFFFFFF; | ||
1349 | wmb(); | ||
1350 | np->put_rx.ex->flaglen = cpu_to_le32(np->rx_buf_sz | NV_RX2_AVAIL); | ||
1351 | if (np->put_rx.ex++ == np->last_rx.ex) | ||
1352 | np->put_rx.ex = np->first_rx.ex; | ||
1353 | } | ||
1354 | if (np->put_rx_ctx++ == np->last_rx_ctx) | 1361 | if (np->put_rx_ctx++ == np->last_rx_ctx) |
1355 | np->put_rx_ctx = np->first_rx_ctx; | 1362 | np->put_rx_ctx = np->first_rx_ctx; |
1356 | } else { | 1363 | } else { |
@@ -1374,6 +1381,7 @@ static void nv_do_rx_refill(unsigned long data) | |||
1374 | { | 1381 | { |
1375 | struct net_device *dev = (struct net_device *) data; | 1382 | struct net_device *dev = (struct net_device *) data; |
1376 | struct fe_priv *np = netdev_priv(dev); | 1383 | struct fe_priv *np = netdev_priv(dev); |
1384 | int retcode; | ||
1377 | 1385 | ||
1378 | if (!using_multi_irqs(dev)) { | 1386 | if (!using_multi_irqs(dev)) { |
1379 | if (np->msi_flags & NV_MSI_X_ENABLED) | 1387 | if (np->msi_flags & NV_MSI_X_ENABLED) |
@@ -1383,7 +1391,11 @@ static void nv_do_rx_refill(unsigned long data) | |||
1383 | } else { | 1391 | } else { |
1384 | disable_irq(np->msi_x_entry[NV_MSI_X_VECTOR_RX].vector); | 1392 | disable_irq(np->msi_x_entry[NV_MSI_X_VECTOR_RX].vector); |
1385 | } | 1393 | } |
1386 | if (nv_alloc_rx(dev)) { | 1394 | if (np->desc_ver == DESC_VER_1 || np->desc_ver == DESC_VER_2) |
1395 | retcode = nv_alloc_rx(dev); | ||
1396 | else | ||
1397 | retcode = nv_alloc_rx_optimized(dev); | ||
1398 | if (retcode) { | ||
1387 | spin_lock_irq(&np->lock); | 1399 | spin_lock_irq(&np->lock); |
1388 | if (!np->in_shutdown) | 1400 | if (!np->in_shutdown) |
1389 | mod_timer(&np->oom_kick, jiffies + OOM_REFILL); | 1401 | mod_timer(&np->oom_kick, jiffies + OOM_REFILL); |
@@ -1456,9 +1468,14 @@ static void nv_init_tx(struct net_device *dev) | |||
1456 | 1468 | ||
1457 | static int nv_init_ring(struct net_device *dev) | 1469 | static int nv_init_ring(struct net_device *dev) |
1458 | { | 1470 | { |
1471 | struct fe_priv *np = netdev_priv(dev); | ||
1472 | |||
1459 | nv_init_tx(dev); | 1473 | nv_init_tx(dev); |
1460 | nv_init_rx(dev); | 1474 | nv_init_rx(dev); |
1461 | return nv_alloc_rx(dev); | 1475 | if (np->desc_ver == DESC_VER_1 || np->desc_ver == DESC_VER_2) |
1476 | return nv_alloc_rx(dev); | ||
1477 | else | ||
1478 | return nv_alloc_rx_optimized(dev); | ||
1462 | } | 1479 | } |
1463 | 1480 | ||
1464 | static int nv_release_txskb(struct net_device *dev, struct nv_skb_map* tx_skb) | 1481 | static int nv_release_txskb(struct net_device *dev, struct nv_skb_map* tx_skb) |
@@ -1554,9 +1571,9 @@ static int nv_start_xmit(struct sk_buff *skb, struct net_device *dev) | |||
1554 | u32 entries = (size >> NV_TX2_TSO_MAX_SHIFT) + ((size & (NV_TX2_TSO_MAX_SIZE-1)) ? 1 : 0); | 1571 | u32 entries = (size >> NV_TX2_TSO_MAX_SHIFT) + ((size & (NV_TX2_TSO_MAX_SIZE-1)) ? 1 : 0); |
1555 | u32 empty_slots; | 1572 | u32 empty_slots; |
1556 | u32 tx_flags_vlan = 0; | 1573 | u32 tx_flags_vlan = 0; |
1557 | union ring_type put_tx; | 1574 | struct ring_desc* put_tx; |
1558 | union ring_type start_tx; | 1575 | struct ring_desc* start_tx; |
1559 | union ring_type prev_tx; | 1576 | struct ring_desc* prev_tx; |
1560 | struct nv_skb_map* prev_tx_ctx; | 1577 | struct nv_skb_map* prev_tx_ctx; |
1561 | 1578 | ||
1562 | /* add fragments to entries count */ | 1579 | /* add fragments to entries count */ |
@@ -1573,10 +1590,7 @@ static int nv_start_xmit(struct sk_buff *skb, struct net_device *dev) | |||
1573 | return NETDEV_TX_BUSY; | 1590 | return NETDEV_TX_BUSY; |
1574 | } | 1591 | } |
1575 | 1592 | ||
1576 | if (np->desc_ver == DESC_VER_1 || np->desc_ver == DESC_VER_2) | 1593 | start_tx = put_tx = np->put_tx.orig; |
1577 | start_tx.orig = put_tx.orig = np->put_tx.orig; | ||
1578 | else | ||
1579 | start_tx.ex = put_tx.ex = np->put_tx.ex; | ||
1580 | 1594 | ||
1581 | /* setup the header buffer */ | 1595 | /* setup the header buffer */ |
1582 | do { | 1596 | do { |
@@ -1586,24 +1600,13 @@ static int nv_start_xmit(struct sk_buff *skb, struct net_device *dev) | |||
1586 | np->put_tx_ctx->dma = pci_map_single(np->pci_dev, skb->data + offset, bcnt, | 1600 | np->put_tx_ctx->dma = pci_map_single(np->pci_dev, skb->data + offset, bcnt, |
1587 | PCI_DMA_TODEVICE); | 1601 | PCI_DMA_TODEVICE); |
1588 | np->put_tx_ctx->dma_len = bcnt; | 1602 | np->put_tx_ctx->dma_len = bcnt; |
1589 | if (np->desc_ver == DESC_VER_1 || np->desc_ver == DESC_VER_2) { | 1603 | put_tx->buf = cpu_to_le32(np->put_tx_ctx->dma); |
1590 | put_tx.orig->buf = cpu_to_le32(np->put_tx_ctx->dma); | 1604 | put_tx->flaglen = cpu_to_le32((bcnt-1) | tx_flags); |
1591 | put_tx.orig->flaglen = cpu_to_le32((bcnt-1) | tx_flags); | ||
1592 | } else { | ||
1593 | put_tx.ex->bufhigh = cpu_to_le64(np->put_tx_ctx->dma) >> 32; | ||
1594 | put_tx.ex->buflow = cpu_to_le64(np->put_tx_ctx->dma) & 0x0FFFFFFFF; | ||
1595 | put_tx.ex->flaglen = cpu_to_le32((bcnt-1) | tx_flags); | ||
1596 | } | ||
1597 | tx_flags = np->tx_flags; | 1605 | tx_flags = np->tx_flags; |
1598 | offset += bcnt; | 1606 | offset += bcnt; |
1599 | size -= bcnt; | 1607 | size -= bcnt; |
1600 | if (np->desc_ver == DESC_VER_1 || np->desc_ver == DESC_VER_2) { | 1608 | if (put_tx++ == np->last_tx.orig) |
1601 | if (put_tx.orig++ == np->last_tx.orig) | 1609 | put_tx = np->first_tx.orig; |
1602 | put_tx.orig = np->first_tx.orig; | ||
1603 | } else { | ||
1604 | if (put_tx.ex++ == np->last_tx.ex) | ||
1605 | put_tx.ex = np->first_tx.ex; | ||
1606 | } | ||
1607 | if (np->put_tx_ctx++ == np->last_tx_ctx) | 1610 | if (np->put_tx_ctx++ == np->last_tx_ctx) |
1608 | np->put_tx_ctx = np->first_tx_ctx; | 1611 | np->put_tx_ctx = np->first_tx_ctx; |
1609 | } while (size); | 1612 | } while (size); |
@@ -1622,33 +1625,19 @@ static int nv_start_xmit(struct sk_buff *skb, struct net_device *dev) | |||
1622 | PCI_DMA_TODEVICE); | 1625 | PCI_DMA_TODEVICE); |
1623 | np->put_tx_ctx->dma_len = bcnt; | 1626 | np->put_tx_ctx->dma_len = bcnt; |
1624 | 1627 | ||
1625 | if (np->desc_ver == DESC_VER_1 || np->desc_ver == DESC_VER_2) { | 1628 | put_tx->buf = cpu_to_le32(np->put_tx_ctx->dma); |
1626 | put_tx.orig->buf = cpu_to_le32(np->put_tx_ctx->dma); | 1629 | put_tx->flaglen = cpu_to_le32((bcnt-1) | tx_flags); |
1627 | put_tx.orig->flaglen = cpu_to_le32((bcnt-1) | tx_flags); | ||
1628 | } else { | ||
1629 | put_tx.ex->bufhigh = cpu_to_le64(np->put_tx_ctx->dma) >> 32; | ||
1630 | put_tx.ex->buflow = cpu_to_le64(np->put_tx_ctx->dma) & 0x0FFFFFFFF; | ||
1631 | put_tx.ex->flaglen = cpu_to_le32((bcnt-1) | tx_flags); | ||
1632 | } | ||
1633 | offset += bcnt; | 1630 | offset += bcnt; |
1634 | size -= bcnt; | 1631 | size -= bcnt; |
1635 | if (np->desc_ver == DESC_VER_1 || np->desc_ver == DESC_VER_2) { | 1632 | if (put_tx++ == np->last_tx.orig) |
1636 | if (put_tx.orig++ == np->last_tx.orig) | 1633 | put_tx = np->first_tx.orig; |
1637 | put_tx.orig = np->first_tx.orig; | ||
1638 | } else { | ||
1639 | if (put_tx.ex++ == np->last_tx.ex) | ||
1640 | put_tx.ex = np->first_tx.ex; | ||
1641 | } | ||
1642 | if (np->put_tx_ctx++ == np->last_tx_ctx) | 1634 | if (np->put_tx_ctx++ == np->last_tx_ctx) |
1643 | np->put_tx_ctx = np->first_tx_ctx; | 1635 | np->put_tx_ctx = np->first_tx_ctx; |
1644 | } while (size); | 1636 | } while (size); |
1645 | } | 1637 | } |
1646 | 1638 | ||
1647 | /* set last fragment flag */ | 1639 | /* set last fragment flag */ |
1648 | if (np->desc_ver == DESC_VER_1 || np->desc_ver == DESC_VER_2) | 1640 | prev_tx->flaglen |= cpu_to_le32(tx_flags_extra); |
1649 | prev_tx.orig->flaglen |= cpu_to_le32(tx_flags_extra); | ||
1650 | else | ||
1651 | prev_tx.ex->flaglen |= cpu_to_le32(tx_flags_extra); | ||
1652 | 1641 | ||
1653 | /* save skb in this slot's context area */ | 1642 | /* save skb in this slot's context area */ |
1654 | prev_tx_ctx->skb = skb; | 1643 | prev_tx_ctx->skb = skb; |
@@ -1667,14 +1656,8 @@ static int nv_start_xmit(struct sk_buff *skb, struct net_device *dev) | |||
1667 | spin_lock_irq(&np->lock); | 1656 | spin_lock_irq(&np->lock); |
1668 | 1657 | ||
1669 | /* set tx flags */ | 1658 | /* set tx flags */ |
1670 | if (np->desc_ver == DESC_VER_1 || np->desc_ver == DESC_VER_2) { | 1659 | start_tx->flaglen |= cpu_to_le32(tx_flags | tx_flags_extra); |
1671 | start_tx.orig->flaglen |= cpu_to_le32(tx_flags | tx_flags_extra); | 1660 | np->put_tx.orig = put_tx; |
1672 | np->put_tx.orig = put_tx.orig; | ||
1673 | } else { | ||
1674 | start_tx.ex->txvlan = cpu_to_le32(tx_flags_vlan); | ||
1675 | start_tx.ex->flaglen |= cpu_to_le32(tx_flags | tx_flags_extra); | ||
1676 | np->put_tx.ex = put_tx.ex; | ||
1677 | } | ||
1678 | 1661 | ||
1679 | spin_unlock_irq(&np->lock); | 1662 | spin_unlock_irq(&np->lock); |
1680 | 1663 | ||
@@ -1696,6 +1679,130 @@ static int nv_start_xmit(struct sk_buff *skb, struct net_device *dev) | |||
1696 | return NETDEV_TX_OK; | 1679 | return NETDEV_TX_OK; |
1697 | } | 1680 | } |
1698 | 1681 | ||
1682 | static int nv_start_xmit_optimized(struct sk_buff *skb, struct net_device *dev) | ||
1683 | { | ||
1684 | struct fe_priv *np = netdev_priv(dev); | ||
1685 | u32 tx_flags = 0; | ||
1686 | u32 tx_flags_extra = NV_TX2_LASTPACKET; | ||
1687 | unsigned int fragments = skb_shinfo(skb)->nr_frags; | ||
1688 | unsigned int i; | ||
1689 | u32 offset = 0; | ||
1690 | u32 bcnt; | ||
1691 | u32 size = skb->len-skb->data_len; | ||
1692 | u32 entries = (size >> NV_TX2_TSO_MAX_SHIFT) + ((size & (NV_TX2_TSO_MAX_SIZE-1)) ? 1 : 0); | ||
1693 | u32 empty_slots; | ||
1694 | u32 tx_flags_vlan = 0; | ||
1695 | struct ring_desc_ex* put_tx; | ||
1696 | struct ring_desc_ex* start_tx; | ||
1697 | struct ring_desc_ex* prev_tx; | ||
1698 | struct nv_skb_map* prev_tx_ctx; | ||
1699 | |||
1700 | /* add fragments to entries count */ | ||
1701 | for (i = 0; i < fragments; i++) { | ||
1702 | entries += (skb_shinfo(skb)->frags[i].size >> NV_TX2_TSO_MAX_SHIFT) + | ||
1703 | ((skb_shinfo(skb)->frags[i].size & (NV_TX2_TSO_MAX_SIZE-1)) ? 1 : 0); | ||
1704 | } | ||
1705 | |||
1706 | empty_slots = nv_get_empty_tx_slots(np); | ||
1707 | if ((empty_slots - np->tx_limit_stop) <= entries) { | ||
1708 | spin_lock_irq(&np->lock); | ||
1709 | netif_stop_queue(dev); | ||
1710 | spin_unlock_irq(&np->lock); | ||
1711 | return NETDEV_TX_BUSY; | ||
1712 | } | ||
1713 | |||
1714 | start_tx = put_tx = np->put_tx.ex; | ||
1715 | |||
1716 | /* setup the header buffer */ | ||
1717 | do { | ||
1718 | prev_tx = put_tx; | ||
1719 | prev_tx_ctx = np->put_tx_ctx; | ||
1720 | bcnt = (size > NV_TX2_TSO_MAX_SIZE) ? NV_TX2_TSO_MAX_SIZE : size; | ||
1721 | np->put_tx_ctx->dma = pci_map_single(np->pci_dev, skb->data + offset, bcnt, | ||
1722 | PCI_DMA_TODEVICE); | ||
1723 | np->put_tx_ctx->dma_len = bcnt; | ||
1724 | put_tx->bufhigh = cpu_to_le64(np->put_tx_ctx->dma) >> 32; | ||
1725 | put_tx->buflow = cpu_to_le64(np->put_tx_ctx->dma) & 0x0FFFFFFFF; | ||
1726 | put_tx->flaglen = cpu_to_le32((bcnt-1) | tx_flags); | ||
1727 | tx_flags = np->tx_flags; | ||
1728 | offset += bcnt; | ||
1729 | size -= bcnt; | ||
1730 | if (put_tx++ == np->last_tx.ex) | ||
1731 | put_tx = np->first_tx.ex; | ||
1732 | if (np->put_tx_ctx++ == np->last_tx_ctx) | ||
1733 | np->put_tx_ctx = np->first_tx_ctx; | ||
1734 | } while (size); | ||
1735 | |||
1736 | /* setup the fragments */ | ||
1737 | for (i = 0; i < fragments; i++) { | ||
1738 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | ||
1739 | u32 size = frag->size; | ||
1740 | offset = 0; | ||
1741 | |||
1742 | do { | ||
1743 | prev_tx = put_tx; | ||
1744 | prev_tx_ctx = np->put_tx_ctx; | ||
1745 | bcnt = (size > NV_TX2_TSO_MAX_SIZE) ? NV_TX2_TSO_MAX_SIZE : size; | ||
1746 | np->put_tx_ctx->dma = pci_map_page(np->pci_dev, frag->page, frag->page_offset+offset, bcnt, | ||
1747 | PCI_DMA_TODEVICE); | ||
1748 | np->put_tx_ctx->dma_len = bcnt; | ||
1749 | |||
1750 | put_tx->bufhigh = cpu_to_le64(np->put_tx_ctx->dma) >> 32; | ||
1751 | put_tx->buflow = cpu_to_le64(np->put_tx_ctx->dma) & 0x0FFFFFFFF; | ||
1752 | put_tx->flaglen = cpu_to_le32((bcnt-1) | tx_flags); | ||
1753 | offset += bcnt; | ||
1754 | size -= bcnt; | ||
1755 | if (put_tx++ == np->last_tx.ex) | ||
1756 | put_tx = np->first_tx.ex; | ||
1757 | if (np->put_tx_ctx++ == np->last_tx_ctx) | ||
1758 | np->put_tx_ctx = np->first_tx_ctx; | ||
1759 | } while (size); | ||
1760 | } | ||
1761 | |||
1762 | /* set last fragment flag */ | ||
1763 | prev_tx->flaglen |= cpu_to_le32(tx_flags_extra); | ||
1764 | |||
1765 | /* save skb in this slot's context area */ | ||
1766 | prev_tx_ctx->skb = skb; | ||
1767 | |||
1768 | if (skb_is_gso(skb)) | ||
1769 | tx_flags_extra = NV_TX2_TSO | (skb_shinfo(skb)->gso_size << NV_TX2_TSO_SHIFT); | ||
1770 | else | ||
1771 | tx_flags_extra = skb->ip_summed == CHECKSUM_PARTIAL ? | ||
1772 | NV_TX2_CHECKSUM_L3 | NV_TX2_CHECKSUM_L4 : 0; | ||
1773 | |||
1774 | /* vlan tag */ | ||
1775 | if (np->vlangrp && vlan_tx_tag_present(skb)) { | ||
1776 | tx_flags_vlan = NV_TX3_VLAN_TAG_PRESENT | vlan_tx_tag_get(skb); | ||
1777 | } | ||
1778 | |||
1779 | spin_lock_irq(&np->lock); | ||
1780 | |||
1781 | /* set tx flags */ | ||
1782 | start_tx->txvlan = cpu_to_le32(tx_flags_vlan); | ||
1783 | start_tx->flaglen |= cpu_to_le32(tx_flags | tx_flags_extra); | ||
1784 | np->put_tx.ex = put_tx; | ||
1785 | |||
1786 | spin_unlock_irq(&np->lock); | ||
1787 | |||
1788 | dprintk(KERN_DEBUG "%s: nv_start_xmit_optimized: entries %d queued for transmission. tx_flags_extra: %x\n", | ||
1789 | dev->name, entries, tx_flags_extra); | ||
1790 | { | ||
1791 | int j; | ||
1792 | for (j=0; j<64; j++) { | ||
1793 | if ((j%16) == 0) | ||
1794 | dprintk("\n%03x:", j); | ||
1795 | dprintk(" %02x", ((unsigned char*)skb->data)[j]); | ||
1796 | } | ||
1797 | dprintk("\n"); | ||
1798 | } | ||
1799 | |||
1800 | dev->trans_start = jiffies; | ||
1801 | writel(NVREG_TXRXCTL_KICK|np->txrxctl_bits, get_hwbase(dev) + NvRegTxRxControl); | ||
1802 | pci_push(get_hwbase(dev)); | ||
1803 | return NETDEV_TX_OK; | ||
1804 | } | ||
1805 | |||
1699 | /* | 1806 | /* |
1700 | * nv_tx_done: check for completed packets, release the skbs. | 1807 | * nv_tx_done: check for completed packets, release the skbs. |
1701 | * | 1808 | * |
@@ -1707,16 +1814,8 @@ static void nv_tx_done(struct net_device *dev) | |||
1707 | u32 flags; | 1814 | u32 flags; |
1708 | struct sk_buff *skb; | 1815 | struct sk_buff *skb; |
1709 | 1816 | ||
1710 | while (1) { | 1817 | while (np->get_tx.orig != np->put_tx.orig) { |
1711 | if (np->desc_ver == DESC_VER_1 || np->desc_ver == DESC_VER_2) { | 1818 | flags = le32_to_cpu(np->get_tx.orig->flaglen); |
1712 | if (np->get_tx.orig == np->put_tx.orig) | ||
1713 | break; | ||
1714 | flags = le32_to_cpu(np->get_tx.orig->flaglen); | ||
1715 | } else { | ||
1716 | if (np->get_tx.ex == np->put_tx.ex) | ||
1717 | break; | ||
1718 | flags = le32_to_cpu(np->get_tx.ex->flaglen); | ||
1719 | } | ||
1720 | 1819 | ||
1721 | dprintk(KERN_DEBUG "%s: nv_tx_done: flags 0x%x.\n", | 1820 | dprintk(KERN_DEBUG "%s: nv_tx_done: flags 0x%x.\n", |
1722 | dev->name, flags); | 1821 | dev->name, flags); |
@@ -1754,13 +1853,45 @@ static void nv_tx_done(struct net_device *dev) | |||
1754 | } | 1853 | } |
1755 | } | 1854 | } |
1756 | nv_release_txskb(dev, np->get_tx_ctx); | 1855 | nv_release_txskb(dev, np->get_tx_ctx); |
1757 | if (np->desc_ver == DESC_VER_1 || np->desc_ver == DESC_VER_2) { | 1856 | if (np->get_tx.orig++ == np->last_tx.orig) |
1758 | if (np->get_tx.orig++ == np->last_tx.orig) | 1857 | np->get_tx.orig = np->first_tx.orig; |
1759 | np->get_tx.orig = np->first_tx.orig; | 1858 | if (np->get_tx_ctx++ == np->last_tx_ctx) |
1760 | } else { | 1859 | np->get_tx_ctx = np->first_tx_ctx; |
1761 | if (np->get_tx.ex++ == np->last_tx.ex) | 1860 | } |
1762 | np->get_tx.ex = np->first_tx.ex; | 1861 | if (nv_get_empty_tx_slots(np) > np->tx_limit_start) |
1862 | netif_wake_queue(dev); | ||
1863 | } | ||
1864 | |||
1865 | static void nv_tx_done_optimized(struct net_device *dev) | ||
1866 | { | ||
1867 | struct fe_priv *np = netdev_priv(dev); | ||
1868 | u32 flags; | ||
1869 | struct sk_buff *skb; | ||
1870 | |||
1871 | while (np->get_tx.ex == np->put_tx.ex) { | ||
1872 | flags = le32_to_cpu(np->get_tx.ex->flaglen); | ||
1873 | |||
1874 | dprintk(KERN_DEBUG "%s: nv_tx_done_optimized: flags 0x%x.\n", | ||
1875 | dev->name, flags); | ||
1876 | if (flags & NV_TX_VALID) | ||
1877 | break; | ||
1878 | if (flags & NV_TX2_LASTPACKET) { | ||
1879 | skb = np->get_tx_ctx->skb; | ||
1880 | if (flags & (NV_TX2_RETRYERROR|NV_TX2_CARRIERLOST|NV_TX2_LATECOLLISION| | ||
1881 | NV_TX2_UNDERFLOW|NV_TX2_ERROR)) { | ||
1882 | if (flags & NV_TX2_UNDERFLOW) | ||
1883 | np->stats.tx_fifo_errors++; | ||
1884 | if (flags & NV_TX2_CARRIERLOST) | ||
1885 | np->stats.tx_carrier_errors++; | ||
1886 | np->stats.tx_errors++; | ||
1887 | } else { | ||
1888 | np->stats.tx_packets++; | ||
1889 | np->stats.tx_bytes += skb->len; | ||
1890 | } | ||
1763 | } | 1891 | } |
1892 | nv_release_txskb(dev, np->get_tx_ctx); | ||
1893 | if (np->get_tx.ex++ == np->last_tx.ex) | ||
1894 | np->get_tx.ex = np->first_tx.ex; | ||
1764 | if (np->get_tx_ctx++ == np->last_tx_ctx) | 1895 | if (np->get_tx_ctx++ == np->last_tx_ctx) |
1765 | np->get_tx_ctx = np->first_tx_ctx; | 1896 | np->get_tx_ctx = np->first_tx_ctx; |
1766 | } | 1897 | } |
@@ -1837,7 +1968,10 @@ static void nv_tx_timeout(struct net_device *dev) | |||
1837 | nv_stop_tx(dev); | 1968 | nv_stop_tx(dev); |
1838 | 1969 | ||
1839 | /* 2) check that the packets were not sent already: */ | 1970 | /* 2) check that the packets were not sent already: */ |
1840 | nv_tx_done(dev); | 1971 | if (np->desc_ver == DESC_VER_1 || np->desc_ver == DESC_VER_2) |
1972 | nv_tx_done(dev); | ||
1973 | else | ||
1974 | nv_tx_done_optimized(dev); | ||
1841 | 1975 | ||
1842 | /* 3) if there are dead entries: clear everything */ | 1976 | /* 3) if there are dead entries: clear everything */ |
1843 | if (np->get_tx_ctx != np->put_tx_ctx) { | 1977 | if (np->get_tx_ctx != np->put_tx_ctx) { |
@@ -1913,22 +2047,14 @@ static int nv_rx_process(struct net_device *dev, int limit) | |||
1913 | u32 vlanflags = 0; | 2047 | u32 vlanflags = 0; |
1914 | int count; | 2048 | int count; |
1915 | 2049 | ||
1916 | for (count = 0; count < limit; ++count) { | 2050 | for (count = 0; count < limit; ++count) { |
1917 | struct sk_buff *skb; | 2051 | struct sk_buff *skb; |
1918 | int len; | 2052 | int len; |
1919 | 2053 | ||
1920 | if (np->desc_ver == DESC_VER_1 || np->desc_ver == DESC_VER_2) { | 2054 | if (np->get_rx.orig == np->put_rx.orig) |
1921 | if (np->get_rx.orig == np->put_rx.orig) | 2055 | break; /* we scanned the whole ring - do not continue */ |
1922 | break; /* we scanned the whole ring - do not continue */ | 2056 | flags = le32_to_cpu(np->get_rx.orig->flaglen); |
1923 | flags = le32_to_cpu(np->get_rx.orig->flaglen); | 2057 | len = nv_descr_getlength(np->get_rx.orig, np->desc_ver); |
1924 | len = nv_descr_getlength(np->get_rx.orig, np->desc_ver); | ||
1925 | } else { | ||
1926 | if (np->get_rx.ex == np->put_rx.ex) | ||
1927 | break; /* we scanned the whole ring - do not continue */ | ||
1928 | flags = le32_to_cpu(np->get_rx.ex->flaglen); | ||
1929 | len = nv_descr_getlength_ex(np->get_rx.ex, np->desc_ver); | ||
1930 | vlanflags = le32_to_cpu(np->get_rx.ex->buflow); | ||
1931 | } | ||
1932 | 2058 | ||
1933 | dprintk(KERN_DEBUG "%s: nv_rx_process: flags 0x%x.\n", | 2059 | dprintk(KERN_DEBUG "%s: nv_rx_process: flags 0x%x.\n", |
1934 | dev->name, flags); | 2060 | dev->name, flags); |
@@ -2076,13 +2202,133 @@ static int nv_rx_process(struct net_device *dev, int limit) | |||
2076 | np->stats.rx_packets++; | 2202 | np->stats.rx_packets++; |
2077 | np->stats.rx_bytes += len; | 2203 | np->stats.rx_bytes += len; |
2078 | next_pkt: | 2204 | next_pkt: |
2079 | if (np->desc_ver == DESC_VER_1 || np->desc_ver == DESC_VER_2) { | 2205 | if (np->get_rx.orig++ == np->last_rx.orig) |
2080 | if (np->get_rx.orig++ == np->last_rx.orig) | 2206 | np->get_rx.orig = np->first_rx.orig; |
2081 | np->get_rx.orig = np->first_rx.orig; | 2207 | if (np->get_rx_ctx++ == np->last_rx_ctx) |
2082 | } else { | 2208 | np->get_rx_ctx = np->first_rx_ctx; |
2083 | if (np->get_rx.ex++ == np->last_rx.ex) | 2209 | } |
2084 | np->get_rx.ex = np->first_rx.ex; | 2210 | |
2211 | return count; | ||
2212 | } | ||
2213 | |||
2214 | static int nv_rx_process_optimized(struct net_device *dev, int limit) | ||
2215 | { | ||
2216 | struct fe_priv *np = netdev_priv(dev); | ||
2217 | u32 flags; | ||
2218 | u32 vlanflags = 0; | ||
2219 | int count; | ||
2220 | |||
2221 | for (count = 0; count < limit; ++count) { | ||
2222 | struct sk_buff *skb; | ||
2223 | int len; | ||
2224 | |||
2225 | if (np->get_rx.ex == np->put_rx.ex) | ||
2226 | break; /* we scanned the whole ring - do not continue */ | ||
2227 | flags = le32_to_cpu(np->get_rx.ex->flaglen); | ||
2228 | len = nv_descr_getlength_ex(np->get_rx.ex, np->desc_ver); | ||
2229 | vlanflags = le32_to_cpu(np->get_rx.ex->buflow); | ||
2230 | |||
2231 | dprintk(KERN_DEBUG "%s: nv_rx_process_optimized: flags 0x%x.\n", | ||
2232 | dev->name, flags); | ||
2233 | |||
2234 | if (flags & NV_RX_AVAIL) | ||
2235 | break; /* still owned by hardware, */ | ||
2236 | |||
2237 | /* | ||
2238 | * the packet is for us - immediately tear down the pci mapping. | ||
2239 | * TODO: check if a prefetch of the first cacheline improves | ||
2240 | * the performance. | ||
2241 | */ | ||
2242 | pci_unmap_single(np->pci_dev, np->get_rx_ctx->dma, | ||
2243 | np->get_rx_ctx->dma_len, | ||
2244 | PCI_DMA_FROMDEVICE); | ||
2245 | skb = np->get_rx_ctx->skb; | ||
2246 | np->get_rx_ctx->skb = NULL; | ||
2247 | |||
2248 | { | ||
2249 | int j; | ||
2250 | dprintk(KERN_DEBUG "Dumping packet (flags 0x%x).",flags); | ||
2251 | for (j=0; j<64; j++) { | ||
2252 | if ((j%16) == 0) | ||
2253 | dprintk("\n%03x:", j); | ||
2254 | dprintk(" %02x", ((unsigned char*)skb->data)[j]); | ||
2255 | } | ||
2256 | dprintk("\n"); | ||
2085 | } | 2257 | } |
2258 | /* look at what we actually got: */ | ||
2259 | if (!(flags & NV_RX2_DESCRIPTORVALID)) { | ||
2260 | dev_kfree_skb(skb); | ||
2261 | goto next_pkt; | ||
2262 | } | ||
2263 | |||
2264 | if (flags & NV_RX2_ERROR) { | ||
2265 | if (flags & (NV_RX2_ERROR1|NV_RX2_ERROR2|NV_RX2_ERROR3)) { | ||
2266 | np->stats.rx_errors++; | ||
2267 | dev_kfree_skb(skb); | ||
2268 | goto next_pkt; | ||
2269 | } | ||
2270 | if (flags & NV_RX2_CRCERR) { | ||
2271 | np->stats.rx_crc_errors++; | ||
2272 | np->stats.rx_errors++; | ||
2273 | dev_kfree_skb(skb); | ||
2274 | goto next_pkt; | ||
2275 | } | ||
2276 | if (flags & NV_RX2_OVERFLOW) { | ||
2277 | np->stats.rx_over_errors++; | ||
2278 | np->stats.rx_errors++; | ||
2279 | dev_kfree_skb(skb); | ||
2280 | goto next_pkt; | ||
2281 | } | ||
2282 | if (flags & NV_RX2_ERROR4) { | ||
2283 | len = nv_getlen(dev, skb->data, len); | ||
2284 | if (len < 0) { | ||
2285 | np->stats.rx_errors++; | ||
2286 | dev_kfree_skb(skb); | ||
2287 | goto next_pkt; | ||
2288 | } | ||
2289 | } | ||
2290 | /* framing errors are soft errors */ | ||
2291 | if (flags & NV_RX2_FRAMINGERR) { | ||
2292 | if (flags & NV_RX2_SUBSTRACT1) { | ||
2293 | len--; | ||
2294 | } | ||
2295 | } | ||
2296 | } | ||
2297 | if (np->rx_csum) { | ||
2298 | flags &= NV_RX2_CHECKSUMMASK; | ||
2299 | if (flags == NV_RX2_CHECKSUMOK1 || | ||
2300 | flags == NV_RX2_CHECKSUMOK2 || | ||
2301 | flags == NV_RX2_CHECKSUMOK3) { | ||
2302 | dprintk(KERN_DEBUG "%s: hw checksum hit!.\n", dev->name); | ||
2303 | skb->ip_summed = CHECKSUM_UNNECESSARY; | ||
2304 | } else { | ||
2305 | dprintk(KERN_DEBUG "%s: hwchecksum miss!.\n", dev->name); | ||
2306 | } | ||
2307 | } | ||
2308 | /* got a valid packet - forward it to the network core */ | ||
2309 | skb_put(skb, len); | ||
2310 | skb->protocol = eth_type_trans(skb, dev); | ||
2311 | dprintk(KERN_DEBUG "%s: nv_rx_process: %d bytes, proto %d accepted.\n", | ||
2312 | dev->name, len, skb->protocol); | ||
2313 | #ifdef CONFIG_FORCEDETH_NAPI | ||
2314 | if (np->vlangrp && (vlanflags & NV_RX3_VLAN_TAG_PRESENT)) | ||
2315 | vlan_hwaccel_receive_skb(skb, np->vlangrp, | ||
2316 | vlanflags & NV_RX3_VLAN_TAG_MASK); | ||
2317 | else | ||
2318 | netif_receive_skb(skb); | ||
2319 | #else | ||
2320 | if (np->vlangrp && (vlanflags & NV_RX3_VLAN_TAG_PRESENT)) | ||
2321 | vlan_hwaccel_rx(skb, np->vlangrp, | ||
2322 | vlanflags & NV_RX3_VLAN_TAG_MASK); | ||
2323 | else | ||
2324 | netif_rx(skb); | ||
2325 | #endif | ||
2326 | dev->last_rx = jiffies; | ||
2327 | np->stats.rx_packets++; | ||
2328 | np->stats.rx_bytes += len; | ||
2329 | next_pkt: | ||
2330 | if (np->get_rx.ex++ == np->last_rx.ex) | ||
2331 | np->get_rx.ex = np->first_rx.ex; | ||
2086 | if (np->get_rx_ctx++ == np->last_rx_ctx) | 2332 | if (np->get_rx_ctx++ == np->last_rx_ctx) |
2087 | np->get_rx_ctx = np->first_rx_ctx; | 2333 | np->get_rx_ctx = np->first_rx_ctx; |
2088 | } | 2334 | } |
@@ -2655,6 +2901,117 @@ static irqreturn_t nv_nic_irq(int foo, void *data) | |||
2655 | return IRQ_RETVAL(i); | 2901 | return IRQ_RETVAL(i); |
2656 | } | 2902 | } |
2657 | 2903 | ||
2904 | static irqreturn_t nv_nic_irq_optimized(int foo, void *data) | ||
2905 | { | ||
2906 | struct net_device *dev = (struct net_device *) data; | ||
2907 | struct fe_priv *np = netdev_priv(dev); | ||
2908 | u8 __iomem *base = get_hwbase(dev); | ||
2909 | u32 events; | ||
2910 | int i; | ||
2911 | |||
2912 | dprintk(KERN_DEBUG "%s: nv_nic_irq_optimized\n", dev->name); | ||
2913 | |||
2914 | for (i=0; ; i++) { | ||
2915 | if (!(np->msi_flags & NV_MSI_X_ENABLED)) { | ||
2916 | events = readl(base + NvRegIrqStatus) & NVREG_IRQSTAT_MASK; | ||
2917 | writel(NVREG_IRQSTAT_MASK, base + NvRegIrqStatus); | ||
2918 | } else { | ||
2919 | events = readl(base + NvRegMSIXIrqStatus) & NVREG_IRQSTAT_MASK; | ||
2920 | writel(NVREG_IRQSTAT_MASK, base + NvRegMSIXIrqStatus); | ||
2921 | } | ||
2922 | pci_push(base); | ||
2923 | dprintk(KERN_DEBUG "%s: irq: %08x\n", dev->name, events); | ||
2924 | if (!(events & np->irqmask)) | ||
2925 | break; | ||
2926 | |||
2927 | spin_lock(&np->lock); | ||
2928 | nv_tx_done_optimized(dev); | ||
2929 | spin_unlock(&np->lock); | ||
2930 | |||
2931 | if (events & NVREG_IRQ_LINK) { | ||
2932 | spin_lock(&np->lock); | ||
2933 | nv_link_irq(dev); | ||
2934 | spin_unlock(&np->lock); | ||
2935 | } | ||
2936 | if (np->need_linktimer && time_after(jiffies, np->link_timeout)) { | ||
2937 | spin_lock(&np->lock); | ||
2938 | nv_linkchange(dev); | ||
2939 | spin_unlock(&np->lock); | ||
2940 | np->link_timeout = jiffies + LINK_TIMEOUT; | ||
2941 | } | ||
2942 | if (events & (NVREG_IRQ_TX_ERR)) { | ||
2943 | dprintk(KERN_DEBUG "%s: received irq with events 0x%x. Probably TX fail.\n", | ||
2944 | dev->name, events); | ||
2945 | } | ||
2946 | if (events & (NVREG_IRQ_UNKNOWN)) { | ||
2947 | printk(KERN_DEBUG "%s: received irq with unknown events 0x%x. Please report\n", | ||
2948 | dev->name, events); | ||
2949 | } | ||
2950 | if (unlikely(events & NVREG_IRQ_RECOVER_ERROR)) { | ||
2951 | spin_lock(&np->lock); | ||
2952 | /* disable interrupts on the nic */ | ||
2953 | if (!(np->msi_flags & NV_MSI_X_ENABLED)) | ||
2954 | writel(0, base + NvRegIrqMask); | ||
2955 | else | ||
2956 | writel(np->irqmask, base + NvRegIrqMask); | ||
2957 | pci_push(base); | ||
2958 | |||
2959 | if (!np->in_shutdown) { | ||
2960 | np->nic_poll_irq = np->irqmask; | ||
2961 | np->recover_error = 1; | ||
2962 | mod_timer(&np->nic_poll, jiffies + POLL_WAIT); | ||
2963 | } | ||
2964 | spin_unlock(&np->lock); | ||
2965 | break; | ||
2966 | } | ||
2967 | |||
2968 | #ifdef CONFIG_FORCEDETH_NAPI | ||
2969 | if (events & NVREG_IRQ_RX_ALL) { | ||
2970 | netif_rx_schedule(dev); | ||
2971 | |||
2972 | /* Disable furthur receive irq's */ | ||
2973 | spin_lock(&np->lock); | ||
2974 | np->irqmask &= ~NVREG_IRQ_RX_ALL; | ||
2975 | |||
2976 | if (np->msi_flags & NV_MSI_X_ENABLED) | ||
2977 | writel(NVREG_IRQ_RX_ALL, base + NvRegIrqMask); | ||
2978 | else | ||
2979 | writel(np->irqmask, base + NvRegIrqMask); | ||
2980 | spin_unlock(&np->lock); | ||
2981 | } | ||
2982 | #else | ||
2983 | nv_rx_process_optimized(dev, dev->weight); | ||
2984 | if (nv_alloc_rx_optimized(dev)) { | ||
2985 | spin_lock(&np->lock); | ||
2986 | if (!np->in_shutdown) | ||
2987 | mod_timer(&np->oom_kick, jiffies + OOM_REFILL); | ||
2988 | spin_unlock(&np->lock); | ||
2989 | } | ||
2990 | #endif | ||
2991 | if (i > max_interrupt_work) { | ||
2992 | spin_lock(&np->lock); | ||
2993 | /* disable interrupts on the nic */ | ||
2994 | if (!(np->msi_flags & NV_MSI_X_ENABLED)) | ||
2995 | writel(0, base + NvRegIrqMask); | ||
2996 | else | ||
2997 | writel(np->irqmask, base + NvRegIrqMask); | ||
2998 | pci_push(base); | ||
2999 | |||
3000 | if (!np->in_shutdown) { | ||
3001 | np->nic_poll_irq = np->irqmask; | ||
3002 | mod_timer(&np->nic_poll, jiffies + POLL_WAIT); | ||
3003 | } | ||
3004 | printk(KERN_DEBUG "%s: too many iterations (%d) in nv_nic_irq.\n", dev->name, i); | ||
3005 | spin_unlock(&np->lock); | ||
3006 | break; | ||
3007 | } | ||
3008 | |||
3009 | } | ||
3010 | dprintk(KERN_DEBUG "%s: nv_nic_irq_optimized completed\n", dev->name); | ||
3011 | |||
3012 | return IRQ_RETVAL(i); | ||
3013 | } | ||
3014 | |||
2658 | static irqreturn_t nv_nic_irq_tx(int foo, void *data) | 3015 | static irqreturn_t nv_nic_irq_tx(int foo, void *data) |
2659 | { | 3016 | { |
2660 | struct net_device *dev = (struct net_device *) data; | 3017 | struct net_device *dev = (struct net_device *) data; |
@@ -2675,7 +3032,7 @@ static irqreturn_t nv_nic_irq_tx(int foo, void *data) | |||
2675 | break; | 3032 | break; |
2676 | 3033 | ||
2677 | spin_lock_irqsave(&np->lock, flags); | 3034 | spin_lock_irqsave(&np->lock, flags); |
2678 | nv_tx_done(dev); | 3035 | nv_tx_done_optimized(dev); |
2679 | spin_unlock_irqrestore(&np->lock, flags); | 3036 | spin_unlock_irqrestore(&np->lock, flags); |
2680 | 3037 | ||
2681 | if (events & (NVREG_IRQ_TX_ERR)) { | 3038 | if (events & (NVREG_IRQ_TX_ERR)) { |
@@ -2711,7 +3068,10 @@ static int nv_napi_poll(struct net_device *dev, int *budget) | |||
2711 | u8 __iomem *base = get_hwbase(dev); | 3068 | u8 __iomem *base = get_hwbase(dev); |
2712 | unsigned long flags; | 3069 | unsigned long flags; |
2713 | 3070 | ||
2714 | pkts = nv_rx_process(dev, limit); | 3071 | if (np->desc_ver == DESC_VER_1 || np->desc_ver == DESC_VER_2) |
3072 | pkts = nv_rx_process(dev, limit); | ||
3073 | else | ||
3074 | pkts = nv_rx_process_optimized(dev, limit); | ||
2715 | 3075 | ||
2716 | if (nv_alloc_rx(dev)) { | 3076 | if (nv_alloc_rx(dev)) { |
2717 | spin_lock_irqsave(&np->lock, flags); | 3077 | spin_lock_irqsave(&np->lock, flags); |
@@ -2782,8 +3142,8 @@ static irqreturn_t nv_nic_irq_rx(int foo, void *data) | |||
2782 | if (!(events & np->irqmask)) | 3142 | if (!(events & np->irqmask)) |
2783 | break; | 3143 | break; |
2784 | 3144 | ||
2785 | nv_rx_process(dev, dev->weight); | 3145 | nv_rx_process_optimized(dev, dev->weight); |
2786 | if (nv_alloc_rx(dev)) { | 3146 | if (nv_alloc_rx_optimized(dev)) { |
2787 | spin_lock_irqsave(&np->lock, flags); | 3147 | spin_lock_irqsave(&np->lock, flags); |
2788 | if (!np->in_shutdown) | 3148 | if (!np->in_shutdown) |
2789 | mod_timer(&np->oom_kick, jiffies + OOM_REFILL); | 3149 | mod_timer(&np->oom_kick, jiffies + OOM_REFILL); |
@@ -2942,6 +3302,16 @@ static int nv_request_irq(struct net_device *dev, int intr_test) | |||
2942 | u8 __iomem *base = get_hwbase(dev); | 3302 | u8 __iomem *base = get_hwbase(dev); |
2943 | int ret = 1; | 3303 | int ret = 1; |
2944 | int i; | 3304 | int i; |
3305 | irqreturn_t (*handler)(int foo, void *data); | ||
3306 | |||
3307 | if (intr_test) { | ||
3308 | handler = nv_nic_irq_test; | ||
3309 | } else { | ||
3310 | if (np->desc_ver == DESC_VER_3) | ||
3311 | handler = nv_nic_irq_optimized; | ||
3312 | else | ||
3313 | handler = nv_nic_irq; | ||
3314 | } | ||
2945 | 3315 | ||
2946 | if (np->msi_flags & NV_MSI_X_CAPABLE) { | 3316 | if (np->msi_flags & NV_MSI_X_CAPABLE) { |
2947 | for (i = 0; i < (np->msi_flags & NV_MSI_X_VECTORS_MASK); i++) { | 3317 | for (i = 0; i < (np->msi_flags & NV_MSI_X_VECTORS_MASK); i++) { |
@@ -2979,10 +3349,7 @@ static int nv_request_irq(struct net_device *dev, int intr_test) | |||
2979 | set_msix_vector_map(dev, NV_MSI_X_VECTOR_OTHER, NVREG_IRQ_OTHER); | 3349 | set_msix_vector_map(dev, NV_MSI_X_VECTOR_OTHER, NVREG_IRQ_OTHER); |
2980 | } else { | 3350 | } else { |
2981 | /* Request irq for all interrupts */ | 3351 | /* Request irq for all interrupts */ |
2982 | if ((!intr_test && | 3352 | if (request_irq(np->msi_x_entry[NV_MSI_X_VECTOR_ALL].vector, handler, IRQF_SHARED, dev->name, dev) != 0) { |
2983 | request_irq(np->msi_x_entry[NV_MSI_X_VECTOR_ALL].vector, &nv_nic_irq, IRQF_SHARED, dev->name, dev) != 0) || | ||
2984 | (intr_test && | ||
2985 | request_irq(np->msi_x_entry[NV_MSI_X_VECTOR_ALL].vector, &nv_nic_irq_test, IRQF_SHARED, dev->name, dev) != 0)) { | ||
2986 | printk(KERN_INFO "forcedeth: request_irq failed %d\n", ret); | 3353 | printk(KERN_INFO "forcedeth: request_irq failed %d\n", ret); |
2987 | pci_disable_msix(np->pci_dev); | 3354 | pci_disable_msix(np->pci_dev); |
2988 | np->msi_flags &= ~NV_MSI_X_ENABLED; | 3355 | np->msi_flags &= ~NV_MSI_X_ENABLED; |
@@ -2998,8 +3365,7 @@ static int nv_request_irq(struct net_device *dev, int intr_test) | |||
2998 | if (ret != 0 && np->msi_flags & NV_MSI_CAPABLE) { | 3365 | if (ret != 0 && np->msi_flags & NV_MSI_CAPABLE) { |
2999 | if ((ret = pci_enable_msi(np->pci_dev)) == 0) { | 3366 | if ((ret = pci_enable_msi(np->pci_dev)) == 0) { |
3000 | np->msi_flags |= NV_MSI_ENABLED; | 3367 | np->msi_flags |= NV_MSI_ENABLED; |
3001 | if ((!intr_test && request_irq(np->pci_dev->irq, &nv_nic_irq, IRQF_SHARED, dev->name, dev) != 0) || | 3368 | if (request_irq(np->pci_dev->irq, handler, IRQF_SHARED, dev->name, dev) != 0) { |
3002 | (intr_test && request_irq(np->pci_dev->irq, &nv_nic_irq_test, IRQF_SHARED, dev->name, dev) != 0)) { | ||
3003 | printk(KERN_INFO "forcedeth: request_irq failed %d\n", ret); | 3369 | printk(KERN_INFO "forcedeth: request_irq failed %d\n", ret); |
3004 | pci_disable_msi(np->pci_dev); | 3370 | pci_disable_msi(np->pci_dev); |
3005 | np->msi_flags &= ~NV_MSI_ENABLED; | 3371 | np->msi_flags &= ~NV_MSI_ENABLED; |
@@ -3014,8 +3380,7 @@ static int nv_request_irq(struct net_device *dev, int intr_test) | |||
3014 | } | 3380 | } |
3015 | } | 3381 | } |
3016 | if (ret != 0) { | 3382 | if (ret != 0) { |
3017 | if ((!intr_test && request_irq(np->pci_dev->irq, &nv_nic_irq, IRQF_SHARED, dev->name, dev) != 0) || | 3383 | if (request_irq(np->pci_dev->irq, handler, IRQF_SHARED, dev->name, dev) != 0) |
3018 | (intr_test && request_irq(np->pci_dev->irq, &nv_nic_irq_test, IRQF_SHARED, dev->name, dev) != 0)) | ||
3019 | goto out_err; | 3384 | goto out_err; |
3020 | 3385 | ||
3021 | } | 3386 | } |
@@ -4629,7 +4994,10 @@ static int __devinit nv_probe(struct pci_dev *pci_dev, const struct pci_device_i | |||
4629 | 4994 | ||
4630 | dev->open = nv_open; | 4995 | dev->open = nv_open; |
4631 | dev->stop = nv_close; | 4996 | dev->stop = nv_close; |
4632 | dev->hard_start_xmit = nv_start_xmit; | 4997 | if (np->desc_ver == DESC_VER_1 || np->desc_ver == DESC_VER_2) |
4998 | dev->hard_start_xmit = nv_start_xmit; | ||
4999 | else | ||
5000 | dev->hard_start_xmit = nv_start_xmit_optimized; | ||
4633 | dev->get_stats = nv_get_stats; | 5001 | dev->get_stats = nv_get_stats; |
4634 | dev->change_mtu = nv_change_mtu; | 5002 | dev->change_mtu = nv_change_mtu; |
4635 | dev->set_mac_address = nv_set_mac_address; | 5003 | dev->set_mac_address = nv_set_mac_address; |