From: Ayaz Abdulla <aabdulla@nvidia.com>
To: Jeff Garzik <jgarzik@pobox.com>,
Manfred Spraul <manfred@colorfullife.com>,
Andrew Morton <akpm@osdl.org>,
netdev@vger.kernel.org, Ayaz Abdulla <aabdulla@nvidia.com>,
public@vger.kernel.org
Subject: [PATCH 7/12] forcedeth: tx data path optimization
Date: Sun, 21 Jan 2007 18:10:47 -0500 [thread overview]
Message-ID: <45B3F2F7.9000301@nvidia.com> (raw)
[-- Attachment #1: Type: text/plain, Size: 694 bytes --]
This patch optimizes the tx data paths and cleans up the code (removes
vlan from descr1/2 since only valid for desc3, changes to make code
easier to read, etc).
Signed-Off-By: Ayaz Abdulla <aabdulla@nvidia.com>
-----------------------------------------------------------------------------------
This email message is for the sole use of the intended recipient(s) and may contain
confidential information. Any unauthorized review, use, disclosure or distribution
is prohibited. If you are not the intended recipient, please contact the sender by
reply email and destroy all copies of the original message.
-----------------------------------------------------------------------------------
[-- Attachment #2: patch-tx-opti-cleanup --]
[-- Type: text/plain, Size: 9499 bytes --]
--- orig/drivers/net/forcedeth.c 2007-01-19 11:03:43.000000000 -0500
+++ new/drivers/net/forcedeth.c 2007-01-19 11:07:48.000000000 -0500
@@ -1563,7 +1563,6 @@
u32 size = skb->len-skb->data_len;
u32 entries = (size >> NV_TX2_TSO_MAX_SHIFT) + ((size & (NV_TX2_TSO_MAX_SIZE-1)) ? 1 : 0);
u32 empty_slots;
- u32 tx_flags_vlan = 0;
struct ring_desc* put_tx;
struct ring_desc* start_tx;
struct ring_desc* prev_tx;
@@ -1576,7 +1575,7 @@
}
empty_slots = nv_get_empty_tx_slots(np);
- if (empty_slots <= entries) {
+ if (unlikely(empty_slots <= entries)) {
spin_lock_irq(&np->lock);
netif_stop_queue(dev);
np->tx_stop = 1;
@@ -1596,12 +1595,13 @@
np->put_tx_ctx->dma_len = bcnt;
put_tx->buf = cpu_to_le32(np->put_tx_ctx->dma);
put_tx->flaglen = cpu_to_le32((bcnt-1) | tx_flags);
+
tx_flags = np->tx_flags;
offset += bcnt;
size -= bcnt;
- if (put_tx++ == np->last_tx.orig)
+ if (unlikely(put_tx++ == np->last_tx.orig))
put_tx = np->first_tx.orig;
- if (np->put_tx_ctx++ == np->last_tx_ctx)
+ if (unlikely(np->put_tx_ctx++ == np->last_tx_ctx))
np->put_tx_ctx = np->first_tx_ctx;
} while (size);
@@ -1618,14 +1618,14 @@
np->put_tx_ctx->dma = pci_map_page(np->pci_dev, frag->page, frag->page_offset+offset, bcnt,
PCI_DMA_TODEVICE);
np->put_tx_ctx->dma_len = bcnt;
-
put_tx->buf = cpu_to_le32(np->put_tx_ctx->dma);
put_tx->flaglen = cpu_to_le32((bcnt-1) | tx_flags);
+
offset += bcnt;
size -= bcnt;
- if (put_tx++ == np->last_tx.orig)
+ if (unlikely(put_tx++ == np->last_tx.orig))
put_tx = np->first_tx.orig;
- if (np->put_tx_ctx++ == np->last_tx_ctx)
+ if (unlikely(np->put_tx_ctx++ == np->last_tx_ctx))
np->put_tx_ctx = np->first_tx_ctx;
} while (size);
}
@@ -1642,11 +1642,6 @@
tx_flags_extra = skb->ip_summed == CHECKSUM_PARTIAL ?
NV_TX2_CHECKSUM_L3 | NV_TX2_CHECKSUM_L4 : 0;
- /* vlan tag */
- if (np->vlangrp && vlan_tx_tag_present(skb)) {
- tx_flags_vlan = NV_TX3_VLAN_TAG_PRESENT | vlan_tx_tag_get(skb);
- }
-
spin_lock_irq(&np->lock);
/* set tx flags */
@@ -1669,7 +1664,6 @@
dev->trans_start = jiffies;
writel(NVREG_TXRXCTL_KICK|np->txrxctl_bits, get_hwbase(dev) + NvRegTxRxControl);
- pci_push(get_hwbase(dev));
return NETDEV_TX_OK;
}
@@ -1677,7 +1671,7 @@
{
struct fe_priv *np = netdev_priv(dev);
u32 tx_flags = 0;
- u32 tx_flags_extra = NV_TX2_LASTPACKET;
+ u32 tx_flags_extra;
unsigned int fragments = skb_shinfo(skb)->nr_frags;
unsigned int i;
u32 offset = 0;
@@ -1685,7 +1679,6 @@
u32 size = skb->len-skb->data_len;
u32 entries = (size >> NV_TX2_TSO_MAX_SHIFT) + ((size & (NV_TX2_TSO_MAX_SIZE-1)) ? 1 : 0);
u32 empty_slots;
- u32 tx_flags_vlan = 0;
struct ring_desc_ex* put_tx;
struct ring_desc_ex* start_tx;
struct ring_desc_ex* prev_tx;
@@ -1698,7 +1691,7 @@
}
empty_slots = nv_get_empty_tx_slots(np);
- if (empty_slots <= entries) {
+ if (unlikely(empty_slots <= entries)) {
spin_lock_irq(&np->lock);
netif_stop_queue(dev);
np->tx_stop = 1;
@@ -1719,12 +1712,13 @@
put_tx->bufhigh = cpu_to_le64(np->put_tx_ctx->dma) >> 32;
put_tx->buflow = cpu_to_le64(np->put_tx_ctx->dma) & 0x0FFFFFFFF;
put_tx->flaglen = cpu_to_le32((bcnt-1) | tx_flags);
- tx_flags = np->tx_flags;
+
+ tx_flags = NV_TX2_VALID;
offset += bcnt;
size -= bcnt;
- if (put_tx++ == np->last_tx.ex)
+ if (unlikely(put_tx++ == np->last_tx.ex))
put_tx = np->first_tx.ex;
- if (np->put_tx_ctx++ == np->last_tx_ctx)
+ if (unlikely(np->put_tx_ctx++ == np->last_tx_ctx))
np->put_tx_ctx = np->first_tx_ctx;
} while (size);
@@ -1741,21 +1735,21 @@
np->put_tx_ctx->dma = pci_map_page(np->pci_dev, frag->page, frag->page_offset+offset, bcnt,
PCI_DMA_TODEVICE);
np->put_tx_ctx->dma_len = bcnt;
-
put_tx->bufhigh = cpu_to_le64(np->put_tx_ctx->dma) >> 32;
put_tx->buflow = cpu_to_le64(np->put_tx_ctx->dma) & 0x0FFFFFFFF;
put_tx->flaglen = cpu_to_le32((bcnt-1) | tx_flags);
+
offset += bcnt;
size -= bcnt;
- if (put_tx++ == np->last_tx.ex)
+ if (unlikely(put_tx++ == np->last_tx.ex))
put_tx = np->first_tx.ex;
- if (np->put_tx_ctx++ == np->last_tx_ctx)
+ if (unlikely(np->put_tx_ctx++ == np->last_tx_ctx))
np->put_tx_ctx = np->first_tx_ctx;
} while (size);
}
/* set last fragment flag */
- prev_tx->flaglen |= cpu_to_le32(tx_flags_extra);
+ prev_tx->flaglen |= cpu_to_le32(NV_TX2_LASTPACKET);
/* save skb in this slot's context area */
prev_tx_ctx->skb = skb;
@@ -1767,14 +1761,18 @@
NV_TX2_CHECKSUM_L3 | NV_TX2_CHECKSUM_L4 : 0;
/* vlan tag */
- if (np->vlangrp && vlan_tx_tag_present(skb)) {
- tx_flags_vlan = NV_TX3_VLAN_TAG_PRESENT | vlan_tx_tag_get(skb);
+ if (likely(!np->vlangrp)) {
+ start_tx->txvlan = 0;
+ } else {
+ if (vlan_tx_tag_present(skb))
+ start_tx->txvlan = cpu_to_le32(NV_TX3_VLAN_TAG_PRESENT | vlan_tx_tag_get(skb));
+ else
+ start_tx->txvlan = 0;
}
spin_lock_irq(&np->lock);
/* set tx flags */
- start_tx->txvlan = cpu_to_le32(tx_flags_vlan);
start_tx->flaglen |= cpu_to_le32(tx_flags | tx_flags_extra);
np->put_tx.ex = put_tx;
@@ -1794,7 +1792,6 @@
dev->trans_start = jiffies;
writel(NVREG_TXRXCTL_KICK|np->txrxctl_bits, get_hwbase(dev) + NvRegTxRxControl);
- pci_push(get_hwbase(dev));
return NETDEV_TX_OK;
}
@@ -1807,21 +1804,22 @@
{
struct fe_priv *np = netdev_priv(dev);
u32 flags;
- struct sk_buff *skb;
struct ring_desc* orig_get_tx = np->get_tx.orig;
- while (np->get_tx.orig != np->put_tx.orig) {
- flags = le32_to_cpu(np->get_tx.orig->flaglen);
+ while ((np->get_tx.orig != np->put_tx.orig) &&
+ !((flags = le32_to_cpu(np->get_tx.orig->flaglen)) & NV_TX_VALID)) {
dprintk(KERN_DEBUG "%s: nv_tx_done: flags 0x%x.\n",
dev->name, flags);
- if (flags & NV_TX_VALID)
- break;
+
+ pci_unmap_page(np->pci_dev, np->get_tx_ctx->dma,
+ np->get_tx_ctx->dma_len,
+ PCI_DMA_TODEVICE);
+ np->get_tx_ctx->dma = 0;
+
if (np->desc_ver == DESC_VER_1) {
if (flags & NV_TX_LASTPACKET) {
- skb = np->get_tx_ctx->skb;
- if (flags & (NV_TX_RETRYERROR|NV_TX_CARRIERLOST|NV_TX_LATECOLLISION|
- NV_TX_UNDERFLOW|NV_TX_ERROR)) {
+ if (flags & NV_TX_ERROR) {
if (flags & NV_TX_UNDERFLOW)
np->stats.tx_fifo_errors++;
if (flags & NV_TX_CARRIERLOST)
@@ -1829,14 +1827,14 @@
np->stats.tx_errors++;
} else {
np->stats.tx_packets++;
- np->stats.tx_bytes += skb->len;
+ np->stats.tx_bytes += np->get_tx_ctx->skb->len;
}
+ dev_kfree_skb_any(np->get_tx_ctx->skb);
+ np->get_tx_ctx->skb = NULL;
}
} else {
if (flags & NV_TX2_LASTPACKET) {
- skb = np->get_tx_ctx->skb;
- if (flags & (NV_TX2_RETRYERROR|NV_TX2_CARRIERLOST|NV_TX2_LATECOLLISION|
- NV_TX2_UNDERFLOW|NV_TX2_ERROR)) {
+ if (flags & NV_TX2_ERROR) {
if (flags & NV_TX2_UNDERFLOW)
np->stats.tx_fifo_errors++;
if (flags & NV_TX2_CARRIERLOST)
@@ -1844,17 +1842,18 @@
np->stats.tx_errors++;
} else {
np->stats.tx_packets++;
- np->stats.tx_bytes += skb->len;
+ np->stats.tx_bytes += np->get_tx_ctx->skb->len;
}
+ dev_kfree_skb_any(np->get_tx_ctx->skb);
+ np->get_tx_ctx->skb = NULL;
}
}
- nv_release_txskb(dev, np->get_tx_ctx);
- if (np->get_tx.orig++ == np->last_tx.orig)
+ if (unlikely(np->get_tx.orig++ == np->last_tx.orig))
np->get_tx.orig = np->first_tx.orig;
- if (np->get_tx_ctx++ == np->last_tx_ctx)
+ if (unlikely(np->get_tx_ctx++ == np->last_tx_ctx))
np->get_tx_ctx = np->first_tx_ctx;
}
- if ((np->tx_stop == 1) && (np->get_tx.orig != orig_get_tx)) {
+ if (unlikely((np->tx_stop == 1) && (np->get_tx.orig != orig_get_tx))) {
np->tx_stop = 0;
netif_wake_queue(dev);
}
@@ -1864,20 +1863,21 @@
{
struct fe_priv *np = netdev_priv(dev);
u32 flags;
- struct sk_buff *skb;
struct ring_desc_ex* orig_get_tx = np->get_tx.ex;
- while (np->get_tx.ex == np->put_tx.ex) {
- flags = le32_to_cpu(np->get_tx.ex->flaglen);
+ while ((np->get_tx.ex != np->put_tx.ex) &&
+ !((flags = le32_to_cpu(np->get_tx.ex->flaglen)) & NV_TX_VALID)) {
dprintk(KERN_DEBUG "%s: nv_tx_done_optimized: flags 0x%x.\n",
dev->name, flags);
- if (flags & NV_TX_VALID)
- break;
+
+ pci_unmap_page(np->pci_dev, np->get_tx_ctx->dma,
+ np->get_tx_ctx->dma_len,
+ PCI_DMA_TODEVICE);
+ np->get_tx_ctx->dma = 0;
+
if (flags & NV_TX2_LASTPACKET) {
- skb = np->get_tx_ctx->skb;
- if (flags & (NV_TX2_RETRYERROR|NV_TX2_CARRIERLOST|NV_TX2_LATECOLLISION|
- NV_TX2_UNDERFLOW|NV_TX2_ERROR)) {
+ if (flags & NV_TX2_ERROR) {
if (flags & NV_TX2_UNDERFLOW)
np->stats.tx_fifo_errors++;
if (flags & NV_TX2_CARRIERLOST)
@@ -1885,16 +1885,17 @@
np->stats.tx_errors++;
} else {
np->stats.tx_packets++;
- np->stats.tx_bytes += skb->len;
+ np->stats.tx_bytes += np->get_tx_ctx->skb->len;
}
+ dev_kfree_skb_any(np->get_tx_ctx->skb);
+ np->get_tx_ctx->skb = NULL;
}
- nv_release_txskb(dev, np->get_tx_ctx);
- if (np->get_tx.ex++ == np->last_tx.ex)
+ if (unlikely(np->get_tx.ex++ == np->last_tx.ex))
np->get_tx.ex = np->first_tx.ex;
- if (np->get_tx_ctx++ == np->last_tx_ctx)
+ if (unlikely(np->get_tx_ctx++ == np->last_tx_ctx))
np->get_tx_ctx = np->first_tx_ctx;
}
- if ((np->tx_stop == 1) && (np->get_tx.ex != orig_get_tx)) {
+ if (unlikely((np->tx_stop == 1) && (np->get_tx.ex != orig_get_tx))) {
np->tx_stop = 0;
netif_wake_queue(dev);
}
next reply other threads:[~2007-01-22 1:05 UTC|newest]
Thread overview: 3+ messages / expand[flat|nested] mbox.gz Atom feed top
2007-01-21 23:10 Ayaz Abdulla [this message]
2007-01-23 6:09 ` [PATCH 7/12] forcedeth: tx data path optimization Jeff Garzik
-- strict thread matches above, loose matches on Subject: below --
2007-01-09 18:30 Ayaz Abdulla
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=45B3F2F7.9000301@nvidia.com \
--to=aabdulla@nvidia.com \
--cc=akpm@osdl.org \
--cc=jgarzik@pobox.com \
--cc=manfred@colorfullife.com \
--cc=netdev@vger.kernel.org \
--cc=public@vger.kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).