Skip to contentSkip to conversation

: powerpc/cell spidernet bugfixes, etc.

PATCH In review Linas Vepstas 7 participants 34 messages Oct 10, 2006 – Oct 14, 2006

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

cover : powerpc/cell spidernet bugfixes, etc.

Andrew, please apply/forward upstream.

The following set of 21 patches (!) are all aimed at the the spidernet ethernet device driver. The spidernet is an etherenet controller built into the Toshiba southbridge for the PowerPC Cell processor. (This is the only device in existance that with this ethernet hardware in it).

These patches re-package/re-order/re-cleanup a previous set of patches I've previously mailed. Thus, some have been previously Acked-by lines, most do not. Most of these patches are tiny, and handle problems that cropped up during testing. Sorry about there being so many of them.

The first set of 12 patches fix a large variety of mostly minor bugs.

The important patches are 13 through 17: these overcome a debilitating performance problem on transmit (6 megabits per second !!) on transmit of patches 500 bytes or larger. After applying these, I am able to get the following:

pkt sz   speed (100K buffs)       speed (4M buffs)
------   -----------------        ----------------
1500     700 Mbits/sec            951 Mbits/sec
1000     658 Mbits/sec            770
800      600                      648
500      500                      500
300      372                      372

60 70 70

Above buf size refers to /proc/sys/net/core/wmem_default

----

I'm not planning on any further patches for a long while. I tried to do som RX work, but gave up. RX performance could be improved.

FYI, Christoph Hellwig's node-aware patches seem to make no difference at all any more.

I tried to base these on linux-2.6.19-rc1-mm1 but hit a kernel BUG in copy_fdtable at fs/file.c:138! (reported earlire today by Olof)

--linas

#1 lore raw

PATCH 1/21 : powerpc/cell spidernet ethtool -i version number info.

This patch adds version information as reported by ethtool -i to the Spidernet driver.

From: James K Lewis <[email protected]>

----

 drivers/net/spider_net.c         |    3 +++
 drivers/net/spider_net.h         |    2 ++
 drivers/net/spider_net_ethtool.c |    2 +-
 3 files changed, 6 insertions(+), 1 deletion(-)
drivers/net/spider_net.c +3 -0
Index: linux-2.6.18-mm2/drivers/net/spider_net.c
===================================================================
@@ -55,6 +55,7 @@ MODULE_AUTHOR("Utz Bacher <utz.bacher@de
 	      "<[email protected]>");
 MODULE_DESCRIPTION("Spider Southbridge Gigabit Ethernet driver");
 MODULE_LICENSE("GPL");
+MODULE_VERSION(VERSION);
 
 static int rx_descriptors = SPIDER_NET_RX_DESCRIPTORS_DEFAULT;
 static int tx_descriptors = SPIDER_NET_TX_DESCRIPTORS_DEFAULT;
@@ -2252,6 +2253,8 @@ static struct pci_driver spider_net_driv
  */
 static int __init spider_net_init(void)
 {
+	printk(KERN_INFO "Spidernet version %s.\n", VERSION);
+
 	if (rx_descriptors < SPIDER_NET_RX_DESCRIPTORS_MIN) {
 		rx_descriptors = SPIDER_NET_RX_DESCRIPTORS_MIN;
 		pr_info("adjusting rx descriptors to %i.\n", rx_descriptors);
drivers/net/spider_net.h +2 -0
Index: linux-2.6.18-mm2/drivers/net/spider_net.h
===================================================================
@@ -24,6 +24,8 @@
 #ifndef _SPIDER_NET_H
 #define _SPIDER_NET_H
 
+#define VERSION "1.1 A"
+
 #include "sungem_phy.h"
 
 extern int spider_net_stop(struct net_device *netdev);
drivers/net/spider_net_ethtool.c +1 -1
Index: linux-2.6.18-mm2/drivers/net/spider_net_ethtool.c
===================================================================
@@ -76,7 +76,7 @@ spider_net_ethtool_get_drvinfo(struct ne
 	/* clear and fill out info */
 	memset(drvinfo, 0, sizeof(struct ethtool_drvinfo));
 	strncpy(drvinfo->driver, spider_net_driver_name, 32);
-	strncpy(drvinfo->version, "0.1", 32);
+	strncpy(drvinfo->version, VERSION, 32);
 	strcpy(drvinfo->fw_version, "no information");
 	strncpy(drvinfo->bus_info, pci_name(card->pdev), 32);
 }
#2 lore raw

PATCH 2/21 : powerpc/cell spidernet burst alignment patch.

In reply to Linas Vepstas

This patch increases the Burst Address alignment from 64 to 1024 in the Spidernet driver. This improves transmit performance for large packets.

From: James K Lewis <[email protected]>

----

 drivers/net/spider_net.h |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/net/spider_net.h +1 -1
Index: linux-2.6.18-mm2/drivers/net/spider_net.h
===================================================================
@@ -211,7 +211,7 @@ extern char spider_net_driver_name[];
 #define SPIDER_NET_DMA_RX_FEND_VALUE	0x00030003
 /* to set TX_DMA_EN */
 #define SPIDER_NET_TX_DMA_EN		0x80000000
-#define SPIDER_NET_GDTDCEIDIS		0x00000002
+#define SPIDER_NET_GDTDCEIDIS		0x00000302
 #define SPIDER_NET_DMA_TX_VALUE		SPIDER_NET_TX_DMA_EN | \
 					SPIDER_NET_GDTDCEIDIS
 #define SPIDER_NET_DMA_TX_FEND_VALUE	0x00030003
#3 lore raw

PATCH 3/21 : Spidernet module parm permissions

In reply to Linas Vepstas

The module param permsissions should bw read-only, not writable.

From: James K Lewis <[email protected]>

----

 drivers/net/spider_net.c |    4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/net/spider_net.c +2 -2
Index: linux-2.6.18-mm2/drivers/net/spider_net.c
===================================================================
@@ -60,8 +60,8 @@ MODULE_VERSION(VERSION);
 static int rx_descriptors = SPIDER_NET_RX_DESCRIPTORS_DEFAULT;
 static int tx_descriptors = SPIDER_NET_TX_DESCRIPTORS_DEFAULT;
 
-module_param(rx_descriptors, int, 0644);
-module_param(tx_descriptors, int, 0644);
+module_param(rx_descriptors, int, 0444);
+module_param(tx_descriptors, int, 0444);
 
 MODULE_PARM_DESC(rx_descriptors, "number of descriptors used " \
 		 "in rx chains");
#4 lore raw

PATCH 4/21 : powerpc/cell spidernet force-end fix

In reply to Linas Vepstas

Bugfix: when cleaning up the transmit queue upon device close, be sure to walk the entire queue.

Cc James K Lewis <[email protected]>
Cc Arnd Bergmann <[email protected]>

----

 drivers/net/spider_net.c |    5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
drivers/net/spider_net.c +4 -1
Index: linux-2.6.18-mm2/drivers/net/spider_net.c
===================================================================
@@ -699,6 +699,8 @@ spider_net_release_tx_descr(struct spide
 
 	/* unmap the skb */
 	skb = descr->skb;
+	if (!skb)
+		return;
 	pci_unmap_single(card->pdev, descr->buf_addr, skb->len,
 			PCI_DMA_TODEVICE);
 	dev_kfree_skb_any(skb);
@@ -751,7 +753,8 @@ spider_net_release_tx_chain(struct spide
 
 		default:
 			card->netdev_stats.tx_dropped++;
-			return 1;
+			if (!brutal)
+				return 1;
 		}
 		spider_net_release_tx_descr(card);
 	}
#5 lore raw

PATCH 5/21 : powerpc/cell spidernet zlen min packet length

In reply to Linas Vepstas

Polite device drivers pad short packets to 60 bytes, so that mean-spirited users don't accidentally DOS some other OS that can't handle short packets.

Cc James K Lewis <[email protected]>
Cc Arnd Bergmann <[email protected]>

----

 drivers/net/spider_net.c |   18 ++++++++++++++----
 1 file changed, 14 insertions(+), 4 deletions(-)
drivers/net/spider_net.c +14 -4
Index: linux-2.6.18-mm2/drivers/net/spider_net.c
===================================================================
@@ -648,18 +648,26 @@ spider_net_prepare_tx_descr(struct spide
 {
 	struct spider_net_descr *descr = card->tx_chain.head;
 	dma_addr_t buf;
+	int length;
 
-	buf = pci_map_single(card->pdev, skb->data, skb->len, PCI_DMA_TODEVICE);
+	length = skb->len;
+	if (length < ETH_ZLEN) {
+		if (skb_pad(skb, ETH_ZLEN-length))
+			return 0;
+		length = ETH_ZLEN;
+	}
+
+	buf = pci_map_single(card->pdev, skb->data, length, PCI_DMA_TODEVICE);
 	if (pci_dma_mapping_error(buf)) {
 		if (netif_msg_tx_err(card) && net_ratelimit())
 			pr_err("could not iommu-map packet (%p, %i). "
-				  "Dropping packet\n", skb->data, skb->len);
+				  "Dropping packet\n", skb->data, length);
 		card->spider_stats.tx_iommu_map_error++;
 		return -ENOMEM;
 	}
 
 	descr->buf_addr = buf;
-	descr->buf_size = skb->len;
+	descr->buf_size = length;
 	descr->next_descr_addr = 0;
 	descr->skb = skb;
 	descr->data_status = 0;
@@ -693,6 +701,7 @@ spider_net_release_tx_descr(struct spide
 {
 	struct spider_net_descr *descr = card->tx_chain.tail;
 	struct sk_buff *skb;
+	unsigned int len;
 
 	card->tx_chain.tail = card->tx_chain.tail->next;
 	descr->dmac_cmd_status |= SPIDER_NET_DESCR_NOT_IN_USE;
@@ -701,7 +710,8 @@ spider_net_release_tx_descr(struct spide
 	skb = descr->skb;
 	if (!skb)
 		return;
-	pci_unmap_single(card->pdev, descr->buf_addr, skb->len,
+	len = skb->len < ETH_ZLEN ? ETH_ZLEN : skb->len;
+	pci_unmap_single(card->pdev, descr->buf_addr, len,
 			PCI_DMA_TODEVICE);
 	dev_kfree_skb_any(skb);
 }
#6 lore raw
In reply to Linas Vepstas
Quoted 2 lines · On Tue, 10 Oct 2006 15:49:47 -0500 [email protected] (Linas Vepstas) wrote:

Andrew, please apply/forward upstream.

s/Andrew/Jeff/;)

Quoted 3 lines

I tried to base these on linux-2.6.19-rc1-mm1 but hit a kernel BUG in copy_fdtable at fs/file.c:138! (reported earlire today by Olof)

yup, ftp://ftp.kernel.org/pub/linux/kernel/people/akpm/patches/2.6/2.6.19-rc1/2.6.19-rc1-mm1/hot-fixes/ holds what should be a fix.

#7 lore raw

PATCH 6/21 : powerpc/cell spidernet add missing netdev watchdog

In reply to Linas Vepstas

Set the netdev watchdog timer.

Cc James K Lewis <[email protected]>
Cc Arnd Bergmann <[email protected]>

----

 drivers/net/spider_net.c |    1 +
 1 file changed, 1 insertion(+)
drivers/net/spider_net.c +1 -0
Index: linux-2.6.18-mm2/drivers/net/spider_net.c
===================================================================
@@ -686,6 +686,7 @@ spider_net_prepare_tx_descr(struct spide
 
 	descr->prev->next_descr_addr = descr->bus_addr;
 
+	card->netdev->trans_start = jiffies; /* set netdev watchdog timer */
 	return 0;
 }
 
#8 lore raw

PATCH 7/21 : Spidernet fix register field definitions

In reply to Linas Vepstas

This patch fixes the names of a few fields in the DMA control register. There is no functional change.

Cc James K Lewis <[email protected]>
Cc Arnd Bergmann <[email protected]>

----

 drivers/net/spider_net.c |    2 +-
 drivers/net/spider_net.h |   16 +++++++++++-----
 2 files changed, 12 insertions(+), 6 deletions(-)
drivers/net/spider_net.c +1 -1
Index: linux-2.6.18-mm2/drivers/net/spider_net.c
===================================================================
@@ -1614,7 +1614,7 @@ spider_net_enable_card(struct spider_net
 			     SPIDER_NET_INT2_MASK_VALUE);
 
 	spider_net_write_reg(card, SPIDER_NET_GDTDMACCNTR,
-			     SPIDER_NET_GDTDCEIDIS);
+			     SPIDER_NET_GDTBSTA | SPIDER_NET_GDTDCEIDIS);
 }
 
 /**
drivers/net/spider_net.h +11 -5
Index: linux-2.6.18-mm2/drivers/net/spider_net.h
===================================================================
@@ -191,7 +191,9 @@ extern char spider_net_driver_name[];
 #define SPIDER_NET_MACMODE_VALUE	0x00000001
 #define SPIDER_NET_BURSTLMT_VALUE	0x00000200 /* about 16 us */
 
-/* 1(0)					enable r/tx dma
+/* DMAC control register GDMACCNTR
+ *
+ * 1(0)				enable r/tx dma
  *  0000000				fixed to 0
  *
  *         000000			fixed to 0
@@ -200,6 +202,7 @@ extern char spider_net_driver_name[];
  *
  *                 000000		fixed to 0
  *                       00		burst alignment: 128 bytes
+ *                       11		burst alignment: 1024 bytes
  *
  *                         00000	fixed to 0
  *                              0	descr writeback size 32 bytes
@@ -210,10 +213,13 @@ extern char spider_net_driver_name[];
 #define SPIDER_NET_DMA_RX_VALUE		0x80000000
 #define SPIDER_NET_DMA_RX_FEND_VALUE	0x00030003
 /* to set TX_DMA_EN */
-#define SPIDER_NET_TX_DMA_EN		0x80000000
-#define SPIDER_NET_GDTDCEIDIS		0x00000302
-#define SPIDER_NET_DMA_TX_VALUE		SPIDER_NET_TX_DMA_EN | \
-					SPIDER_NET_GDTDCEIDIS
+#define SPIDER_NET_TX_DMA_EN           0x80000000
+#define SPIDER_NET_GDTBSTA             0x00000300
+#define SPIDER_NET_GDTDCEIDIS          0x00000002
+#define SPIDER_NET_DMA_TX_VALUE        SPIDER_NET_TX_DMA_EN | \
+                                       SPIDER_NET_GDTBSTA | \
+                                       SPIDER_NET_GDTDCEIDIS
+
 #define SPIDER_NET_DMA_TX_FEND_VALUE	0x00030003
 
 /* SPIDER_NET_UA_DESCR_VALUE is OR'ed with the unicast address */
#9 lore raw

PATCH 8/21 : Spidernet stop queue when queue is full.

In reply to Linas Vepstas

This patch adds a call to netif_stop_queue() when there is no more room for more packets on the transmit queue.

Cc James K Lewis <[email protected]>
Cc Arnd Bergmann <[email protected]>

----

 drivers/net/spider_net.c |   33 ++++++++++-----------------------
 1 file changed, 10 insertions(+), 23 deletions(-)
drivers/net/spider_net.c +10 -23
Index: linux-2.6.18-mm2/drivers/net/spider_net.c
===================================================================
@@ -823,39 +823,25 @@ spider_net_xmit(struct sk_buff *skb, str
 	struct spider_net_descr_chain *chain = &card->tx_chain;
 	struct spider_net_descr *descr = chain->head;
 	unsigned long flags;
-	int result;
 
 	spin_lock_irqsave(&chain->lock, flags);
 
 	spider_net_release_tx_chain(card, 0);
 
-	if (chain->head->next == chain->tail->prev) {
-		card->netdev_stats.tx_dropped++;
-		result = NETDEV_TX_LOCKED;
-		goto out;
-	}
+	if ((chain->head->next == chain->tail->prev) ||
+	   (spider_net_get_descr_status(descr) != SPIDER_NET_DESCR_NOT_IN_USE) ||
+	   (spider_net_prepare_tx_descr(card, skb) != 0)) {
 
-	if (spider_net_get_descr_status(descr) != SPIDER_NET_DESCR_NOT_IN_USE) {
 		card->netdev_stats.tx_dropped++;
-		result = NETDEV_TX_LOCKED;
-		goto out;
+		spin_unlock_irqrestore(&chain->lock, flags);
+		netif_stop_queue(netdev);
+		return NETDEV_TX_BUSY;
 	}
 
-	if (spider_net_prepare_tx_descr(card, skb) != 0) {
-		card->netdev_stats.tx_dropped++;
-		result = NETDEV_TX_BUSY;
-		goto out;
-	}
-
-	result = NETDEV_TX_OK;
-
 	spider_net_kick_tx_dma(card);
 	card->tx_chain.head = card->tx_chain.head->next;
-
-out:
 	spin_unlock_irqrestore(&chain->lock, flags);
-	netif_wake_queue(netdev);
-	return result;
+	return NETDEV_TX_OK;
 }
 
 /**
@@ -874,9 +860,10 @@ spider_net_cleanup_tx_ring(struct spider
 	spin_lock_irqsave(&card->tx_chain.lock, flags);
 
 	if ((spider_net_release_tx_chain(card, 0) != 0) &&
-	    (card->netdev->flags & IFF_UP))
+	    (card->netdev->flags & IFF_UP)) {
 		spider_net_kick_tx_dma(card);
-
+		netif_wake_queue(card->netdev);
+	}
 	spin_unlock_irqrestore(&card->tx_chain.lock, flags);
 }
 
#10 lore raw

PATCH 9/21 : powerpc/cell spidernet bogus rx interrupt bit

In reply to Linas Vepstas

The current receive interrupt mask sets a bogus bit that doesn't even belong to the definition of this register. Remove it.

Cc James K Lewis <[email protected]>
Cc Arnd Bergmann <[email protected]>

----

 drivers/net/spider_net.h |    5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)
drivers/net/spider_net.h +2 -3
Index: linux-2.6.18-mm2/drivers/net/spider_net.h
===================================================================
@@ -332,9 +332,8 @@ enum spider_net_int2_status {
 				  (1 << SPIDER_NET_GDTDCEINT) | \
 				  (1 << SPIDER_NET_GDTFDCINT) )
 
-/* we rely on flagged descriptor interrupts*/
-#define SPIDER_NET_RXINT	( (1 << SPIDER_NET_GDAFDCINT) | \
-				  (1 << SPIDER_NET_GRMFLLINT) )
+/* We rely on flagged descriptor interrupts */
+#define SPIDER_NET_RXINT	( (1 << SPIDER_NET_GDAFDCINT) )
 
 #define SPIDER_NET_ERRINT	( 0xffffffff & \
 				  (~SPIDER_NET_TXINT) & \
#11 lore raw

PATCH 10/21 : powerpc/cell spidernet fix error interrupt print

In reply to Linas Vepstas

The print message associated with the descriptor chain end interrupt prints a bogs value. Fix that.

Cc James K Lewis <[email protected]>
Cc Arnd Bergmann <[email protected]>

----

 drivers/net/spider_net.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/net/spider_net.c +1 -1
Index: linux-2.6.18-mm2/drivers/net/spider_net.c
===================================================================
@@ -1356,7 +1356,7 @@ spider_net_handle_error_irq(struct spide
 		if (netif_msg_intr(card))
 			pr_err("got descriptor chain end interrupt, "
 			       "restarting DMAC %c.\n",
-			       'D'+i-SPIDER_NET_GDDDCEINT);
+			       'D'-(i-SPIDER_NET_GDDDCEINT)/3);
 		spider_net_refill_rx_chain(card);
 		spider_net_enable_rxdmac(card);
 		show_error = 0;
#12 lore raw

PATCH 11/21 : powerpc/cell spidernet stop error printing patch.

In reply to Linas Vepstas

Turn off mis-interpretation of the queue-empty interrupt status bit as an error.

----

 drivers/net/spider_net.c |   13 +++++++++----
 1 file changed, 9 insertions(+), 4 deletions(-)
drivers/net/spider_net.c +9 -4
Index: linux-2.6.18-mm2/drivers/net/spider_net.c
===================================================================
@@ -1245,12 +1245,15 @@ spider_net_handle_error_irq(struct spide
 	case SPIDER_NET_PHYINT:
 	case SPIDER_NET_GMAC2INT:
 	case SPIDER_NET_GMAC1INT:
-	case SPIDER_NET_GIPSINT:
 	case SPIDER_NET_GFIFOINT:
 	case SPIDER_NET_DMACINT:
 	case SPIDER_NET_GSYSINT:
 		break; */
 
+	case SPIDER_NET_GIPSINT:
+		show_error = 0;
+		break;
+
 	case SPIDER_NET_GPWOPCMPINT:
 		/* PHY write operation completed */
 		show_error = 0;
@@ -1309,9 +1312,10 @@ spider_net_handle_error_irq(struct spide
 	case SPIDER_NET_GDTDCEINT:
 		/* chain end. If a descriptor should be sent, kick off
 		 * tx dma
-		if (card->tx_chain.tail == card->tx_chain.head)
+		if (card->tx_chain.tail != card->tx_chain.head)
 			spider_net_kick_tx_dma(card);
-		show_error = 0; */
+		*/
+		show_error = 0;
 		break;
 
 	/* case SPIDER_NET_G1TMCNTINT: not used. print a message */
@@ -1425,8 +1429,9 @@ spider_net_handle_error_irq(struct spide
 	}
 
 	if ((show_error) && (netif_msg_intr(card)))
-		pr_err("Got error interrupt, GHIINT0STS = 0x%08x, "
+		pr_err("Got error interrupt on %s, GHIINT0STS = 0x%08x, "
 		       "GHIINT1STS = 0x%08x, GHIINT2STS = 0x%08x\n",
+		       card->netdev->name,
 		       status_reg, error_reg1, error_reg2);
 
 	/* clear interrupt sources */
#13 lore raw

PATCH 12/21 : powerpc/cell spidernet incorrect offset

In reply to Linas Vepstas

Bugfix -- the rx chain is in memory after the tx chain -- the offset being used was wrong, resulting in memory corruption when the size of the rx and tx rings weren't exactly the same.

Cc James K Lewis <[email protected]>
Cc Arnd Bergmann <[email protected]>

----

 drivers/net/spider_net.c |    4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
drivers/net/spider_net.c +3 -1
Index: linux-2.6.18-mm2/drivers/net/spider_net.c
===================================================================
@@ -1628,8 +1628,10 @@ spider_net_open(struct net_device *netde
 	if (spider_net_init_chain(card, &card->tx_chain, card->descr,
 			PCI_DMA_TODEVICE, card->tx_desc))
 		goto alloc_tx_failed;
+
+	/* rx_chain is after tx_chain, so offset is descr + tx_count */
 	if (spider_net_init_chain(card, &card->rx_chain,
-			card->descr + card->rx_desc,
+			card->descr + card->tx_desc,
 			PCI_DMA_FROMDEVICE, card->rx_desc))
 		goto alloc_rx_failed;
 
#14 lore raw

PATCH 13/21 : powerpc/cell spidernet low watermark patch.

In reply to Linas Vepstas

Implement basic low-watermark support for the transmit queue. Hardware low-watermarks allow a properly configured kernel to continously stream data to a device and not have to handle any interrupts at all in doing so. Correct zero-interrupt operation can be actually observed for this driver, when the socket buffer is made large enough.

The basic idea of a low-watermark interrupt is as follows. The device driver queues up a bunch of packets for the hardware to transmit, and then kicks the hardware to get it started. As the hardware drains the queue of pending, untransmitted packets, the device driver will want to know when the queue is almost empty, so that it can queue some more packets.

If the queue drains down to the low waterark, then an interrupt will be generated. However, if the kernel/driver continues to add enough packets to keep the queue partially filled, no interrupt will actually be generated, and the hardware can continue streaming packets indefinitely in this mode.

The impelmentation is done by setting the DESCR_TXDESFLG flag in one of the packets. When the hardware sees this flag, it will interrupt the device driver. Because this flag is on a fixed packet, rather than at fixed location in the queue, the code below needs to move the flag as more packets are queued up. This implementation attempts to keep the flag at about 1/4 from "empty".

----

 drivers/net/spider_net.c |   43 +++++++++++++++++++++++++++++++++++++++++++
 drivers/net/spider_net.h |    8 ++++----
 2 files changed, 47 insertions(+), 4 deletions(-)
drivers/net/spider_net.c +43 -0
Index: linux-2.6.18-mm2/drivers/net/spider_net.c
===================================================================
@@ -684,6 +684,7 @@ spider_net_prepare_tx_descr(struct spide
 			break;
 		}
 
+	/* Chain the bus address, so that the DMA engine finds this descr. */
 	descr->prev->next_descr_addr = descr->bus_addr;
 
 	card->netdev->trans_start = jiffies; /* set netdev watchdog timer */
@@ -717,6 +718,41 @@ spider_net_release_tx_descr(struct spide
 	dev_kfree_skb_any(skb);
 }
 
+static void
+spider_net_set_low_watermark(struct spider_net_card *card)
+{
+	int status;
+	int cnt=0;
+	int i;
+	struct spider_net_descr *descr = card->tx_chain.tail;
+
+	/* Measure the length of the queue. */
+	while (descr != card->tx_chain.head) {
+		status = descr->dmac_cmd_status & SPIDER_NET_DESCR_NOT_IN_USE;
+		if (status == SPIDER_NET_DESCR_NOT_IN_USE)
+			break;
+		descr = descr->next;
+		cnt++;
+	}
+
+	/* If TX queue is short, don't even bother with interrupts */
+	if (cnt < card->tx_desc/4)
+		return;
+
+	/* Set low-watermark 3/4th's of the way into the queue. */
+	descr = card->tx_chain.tail;
+	cnt = (cnt*3)/4;
+	for (i=0;i<cnt; i++)
+		descr = descr->next;
+
+	/* Set the new watermark, clear the old watermark */
+	descr->dmac_cmd_status |= SPIDER_NET_DESCR_TXDESFLG;
+	if (card->low_watermark && card->low_watermark != descr)
+		card->low_watermark->dmac_cmd_status =
+		     card->low_watermark->dmac_cmd_status & ~SPIDER_NET_DESCR_TXDESFLG;
+	card->low_watermark = descr;
+}
+
 /**
  * spider_net_release_tx_chain - processes sent tx descriptors
  * @card: adapter structure
@@ -838,6 +874,7 @@ spider_net_xmit(struct sk_buff *skb, str
 		return NETDEV_TX_BUSY;
 	}
 
+	spider_net_set_low_watermark(card);
 	spider_net_kick_tx_dma(card);
 	card->tx_chain.head = card->tx_chain.head->next;
 	spin_unlock_irqrestore(&chain->lock, flags);
@@ -1467,6 +1504,10 @@ spider_net_interrupt(int irq, void *ptr,
 		spider_net_rx_irq_off(card);
 		netif_rx_schedule(netdev);
 	}
+	if (status_reg & SPIDER_NET_TXINT ) {
+		spider_net_cleanup_tx_ring(card);
+		netif_wake_queue(netdev);
+	}
 
 	if (status_reg & SPIDER_NET_ERRINT )
 		spider_net_handle_error_irq(card, status_reg);
@@ -1629,6 +1670,8 @@ spider_net_open(struct net_device *netde
 			PCI_DMA_TODEVICE, card->tx_desc))
 		goto alloc_tx_failed;
 
+	card->low_watermark = NULL;
+
 	/* rx_chain is after tx_chain, so offset is descr + tx_count */
 	if (spider_net_init_chain(card, &card->rx_chain,
 			card->descr + card->tx_desc,
drivers/net/spider_net.h +4 -4
Index: linux-2.6.18-mm2/drivers/net/spider_net.h
===================================================================
@@ -49,7 +49,7 @@ extern char spider_net_driver_name[];
 #define SPIDER_NET_TX_DESCRIPTORS_MIN		16
 #define SPIDER_NET_TX_DESCRIPTORS_MAX		512
 
-#define SPIDER_NET_TX_TIMER			20
+#define SPIDER_NET_TX_TIMER			(HZ/5)
 
 #define SPIDER_NET_RX_CSUM_DEFAULT		1
 
@@ -328,9 +328,7 @@ enum spider_net_int2_status {
 	SPIDER_NET_GRISPDNGINT
 };
 
-#define SPIDER_NET_TXINT	( (1 << SPIDER_NET_GTTEDINT) | \
-				  (1 << SPIDER_NET_GDTDCEINT) | \
-				  (1 << SPIDER_NET_GDTFDCINT) )
+#define SPIDER_NET_TXINT	( (1 << SPIDER_NET_GDTFDCINT) )
 
 /* We rely on flagged descriptor interrupts */
 #define SPIDER_NET_RXINT	( (1 << SPIDER_NET_GDAFDCINT) )
@@ -356,6 +354,7 @@ enum spider_net_int2_status {
 #define SPIDER_NET_DESCR_FORCE_END		0x50000000 /* used in rx and tx */
 #define SPIDER_NET_DESCR_CARDOWNED		0xA0000000 /* used in rx and tx */
 #define SPIDER_NET_DESCR_NOT_IN_USE		0xF0000000
+#define SPIDER_NET_DESCR_TXDESFLG		0x00800000
 
 struct spider_net_descr {
 	/* as defined by the hardware */
@@ -440,6 +439,7 @@ struct spider_net_card {
 
 	struct spider_net_descr_chain tx_chain;
 	struct spider_net_descr_chain rx_chain;
+	struct spider_net_descr *low_watermark;
 
 	struct net_device_stats netdev_stats;
 
#15 lore raw

PATCH 14/21 : powerpc/cell spidernet NAPI polling info.

In reply to Linas Vepstas

This patch moves transmit queue cleanup code out of the interrupt context, and into the NAPI polling routine.

----

 drivers/net/spider_net.c |   16 ++++++++--------
 1 file changed, 8 insertions(+), 8 deletions(-)
drivers/net/spider_net.c +8 -8
Index: linux-2.6.18-mm2/drivers/net/spider_net.c
===================================================================
@@ -715,7 +715,7 @@ spider_net_release_tx_descr(struct spide
 	len = skb->len < ETH_ZLEN ? ETH_ZLEN : skb->len;
 	pci_unmap_single(card->pdev, descr->buf_addr, len,
 			PCI_DMA_TODEVICE);
-	dev_kfree_skb_any(skb);
+	dev_kfree_skb(skb);
 }
 
 static void
@@ -885,9 +885,10 @@ spider_net_xmit(struct sk_buff *skb, str
  * spider_net_cleanup_tx_ring - cleans up the TX ring
  * @card: card structure
  *
- * spider_net_cleanup_tx_ring is called by the tx_timer (as we don't use
- * interrupts to cleanup our TX ring) and returns sent packets to the stack
- * by freeing them
+ * spider_net_cleanup_tx_ring is called by either the tx_timer
+ * or from the NAPI polling routine.
+ * This routine releases resources associted with transmitted
+ * packets, including updating the queue tail pointer.
  */
 static void
 spider_net_cleanup_tx_ring(struct spider_net_card *card)
@@ -1092,6 +1093,7 @@ spider_net_poll(struct net_device *netde
 	int packets_to_do, packets_done = 0;
 	int no_more_packets = 0;
 
+	spider_net_cleanup_tx_ring(card);
 	packets_to_do = min(*budget, netdev->quota);
 
 	while (packets_to_do) {
@@ -1504,10 +1506,8 @@ spider_net_interrupt(int irq, void *ptr,
 		spider_net_rx_irq_off(card);
 		netif_rx_schedule(netdev);
 	}
-	if (status_reg & SPIDER_NET_TXINT ) {
-		spider_net_cleanup_tx_ring(card);
-		netif_wake_queue(netdev);
-	}
+	if (status_reg & SPIDER_NET_TXINT)
+		netif_rx_schedule(netdev);
 
 	if (status_reg & SPIDER_NET_ERRINT )
 		spider_net_handle_error_irq(card, status_reg);
#16 lore raw

PATCH 15/21 : powerpc/cell spidernet refine locking

In reply to Linas Vepstas

The transmit side of the spider ethernet driver currently places locks around some very large chunks of code. This results in a fair amount of lock contention is some cases. This patch makes the locks much more fine-grained, protecting only the cirtical sections. One lock is used to protect three locations: the queue head and tail pointers, and the queue low-watermark location.

Cc Arnd Bergmann <[email protected]>
Cc James K Lewis <[email protected]>

----

 drivers/net/spider_net.c |   95 +++++++++++++++++++++--------------------------
 1 file changed, 43 insertions(+), 52 deletions(-)
drivers/net/spider_net.c +43 -52
Index: linux-2.6.18-mm2/drivers/net/spider_net.c
===================================================================
@@ -646,8 +646,9 @@ static int
 spider_net_prepare_tx_descr(struct spider_net_card *card,
 			    struct sk_buff *skb)
 {
-	struct spider_net_descr *descr = card->tx_chain.head;
+	struct spider_net_descr *descr;
 	dma_addr_t buf;
+	unsigned long flags;
 	int length;
 
 	length = skb->len;
@@ -666,6 +667,10 @@ spider_net_prepare_tx_descr(struct spide
 		return -ENOMEM;
 	}
 
+	spin_lock_irqsave(&card->tx_chain.lock, flags);
+	descr = card->tx_chain.head;
+	card->tx_chain.head = descr->next;
+
 	descr->buf_addr = buf;
 	descr->buf_size = length;
 	descr->next_descr_addr = 0;
@@ -674,6 +679,8 @@ spider_net_prepare_tx_descr(struct spide
 
 	descr->dmac_cmd_status =
 			SPIDER_NET_DESCR_CARDOWNED | SPIDER_NET_DMAC_NOCS;
+	spin_unlock_irqrestore(&card->tx_chain.lock, flags);
+
 	if (skb->protocol == htons(ETH_P_IP))
 		switch (skb->nh.iph->protocol) {
 		case IPPROTO_TCP:
@@ -691,42 +698,17 @@ spider_net_prepare_tx_descr(struct spide
 	return 0;
 }
 
-/**
- * spider_net_release_tx_descr - processes a used tx descriptor
- * @card: card structure
- * @descr: descriptor to release
- *
- * releases a used tx descriptor (unmapping, freeing of skb)
- */
-static inline void
-spider_net_release_tx_descr(struct spider_net_card *card)
-{
-	struct spider_net_descr *descr = card->tx_chain.tail;
-	struct sk_buff *skb;
-	unsigned int len;
-
-	card->tx_chain.tail = card->tx_chain.tail->next;
-	descr->dmac_cmd_status |= SPIDER_NET_DESCR_NOT_IN_USE;
-
-	/* unmap the skb */
-	skb = descr->skb;
-	if (!skb)
-		return;
-	len = skb->len < ETH_ZLEN ? ETH_ZLEN : skb->len;
-	pci_unmap_single(card->pdev, descr->buf_addr, len,
-			PCI_DMA_TODEVICE);
-	dev_kfree_skb(skb);
-}
-
 static void
 spider_net_set_low_watermark(struct spider_net_card *card)
 {
+	unsigned long flags;
 	int status;
 	int cnt=0;
 	int i;
 	struct spider_net_descr *descr = card->tx_chain.tail;
 
-	/* Measure the length of the queue. */
+	/* Measure the length of the queue. Measurement does not
+	 * need to be precise -- does not need a lock. */
 	while (descr != card->tx_chain.head) {
 		status = descr->dmac_cmd_status & SPIDER_NET_DESCR_NOT_IN_USE;
 		if (status == SPIDER_NET_DESCR_NOT_IN_USE)
@@ -746,11 +728,13 @@ spider_net_set_low_watermark(struct spid
 		descr = descr->next;
 
 	/* Set the new watermark, clear the old watermark */
+	spin_lock_irqsave(&card->tx_chain.lock, flags);
 	descr->dmac_cmd_status |= SPIDER_NET_DESCR_TXDESFLG;
 	if (card->low_watermark && card->low_watermark != descr)
 		card->low_watermark->dmac_cmd_status =
 		     card->low_watermark->dmac_cmd_status & ~SPIDER_NET_DESCR_TXDESFLG;
 	card->low_watermark = descr;
+	spin_unlock_irqrestore(&card->tx_chain.lock, flags);
 }
 
 /**
@@ -769,21 +753,31 @@ static int
 spider_net_release_tx_chain(struct spider_net_card *card, int brutal)
 {
 	struct spider_net_descr_chain *chain = &card->tx_chain;
+	struct spider_net_descr *descr;
+	struct sk_buff *skb;
+	u32 buf_addr;
+	unsigned long flags;
 	int status;
 
 	spider_net_read_reg(card, SPIDER_NET_GDTDMACCNTR);
 
 	while (chain->tail != chain->head) {
-		status = spider_net_get_descr_status(chain->tail);
+		spin_lock_irqsave(&chain->lock, flags);
+		descr = chain->tail;
+
+		status = spider_net_get_descr_status(descr);
 		switch (status) {
 		case SPIDER_NET_DESCR_COMPLETE:
 			card->netdev_stats.tx_packets++;
-			card->netdev_stats.tx_bytes += chain->tail->skb->len;
+			card->netdev_stats.tx_bytes += descr->skb->len;
 			break;
 
 		case SPIDER_NET_DESCR_CARDOWNED:
-			if (!brutal)
+			if (!brutal) {
+				spin_unlock_irqrestore(&chain->lock, flags);
 				return 1;
+			}
+
 			/* fallthrough, if we release the descriptors
 			 * brutally (then we don't care about
 			 * SPIDER_NET_DESCR_CARDOWNED) */
@@ -800,12 +794,25 @@ spider_net_release_tx_chain(struct spide
 
 		default:
 			card->netdev_stats.tx_dropped++;
-			if (!brutal)
+			if (!brutal) {
+				spin_unlock_irqrestore(&chain->lock, flags);
 				return 1;
+			}
 		}
-		spider_net_release_tx_descr(card);
-	}
 
+		chain->tail = descr->next;
+		descr->dmac_cmd_status |= SPIDER_NET_DESCR_NOT_IN_USE;
+		skb = descr->skb;
+		buf_addr = descr->buf_addr;
+		spin_unlock_irqrestore(&chain->lock, flags);
+
+		/* unmap the skb */
+		if (skb) {
+			int len = skb->len < ETH_ZLEN ? ETH_ZLEN : skb->len;
+			pci_unmap_single(card->pdev, buf_addr, len, PCI_DMA_TODEVICE);
+			dev_kfree_skb(skb);
+		}
+	}
 	return 0;
 }
 
@@ -857,27 +864,19 @@ spider_net_xmit(struct sk_buff *skb, str
 {
 	struct spider_net_card *card = netdev_priv(netdev);
 	struct spider_net_descr_chain *chain = &card->tx_chain;
-	struct spider_net_descr *descr = chain->head;
-	unsigned long flags;
-
-	spin_lock_irqsave(&chain->lock, flags);
 
 	spider_net_release_tx_chain(card, 0);
 
 	if ((chain->head->next == chain->tail->prev) ||
-	   (spider_net_get_descr_status(descr) != SPIDER_NET_DESCR_NOT_IN_USE) ||
 	   (spider_net_prepare_tx_descr(card, skb) != 0)) {
 
 		card->netdev_stats.tx_dropped++;
-		spin_unlock_irqrestore(&chain->lock, flags);
 		netif_stop_queue(netdev);
 		return NETDEV_TX_BUSY;
 	}
 
 	spider_net_set_low_watermark(card);
 	spider_net_kick_tx_dma(card);
-	card->tx_chain.head = card->tx_chain.head->next;
-	spin_unlock_irqrestore(&chain->lock, flags);
 	return NETDEV_TX_OK;
 }
 
@@ -893,16 +892,11 @@ spider_net_xmit(struct sk_buff *skb, str
 static void
 spider_net_cleanup_tx_ring(struct spider_net_card *card)
 {
-	unsigned long flags;
-
-	spin_lock_irqsave(&card->tx_chain.lock, flags);
-
 	if ((spider_net_release_tx_chain(card, 0) != 0) &&
 	    (card->netdev->flags & IFF_UP)) {
 		spider_net_kick_tx_dma(card);
 		netif_wake_queue(card->netdev);
 	}
-	spin_unlock_irqrestore(&card->tx_chain.lock, flags);
 }
 
 /**
@@ -1930,10 +1924,7 @@ spider_net_stop(struct net_device *netde
 	spider_net_disable_rxdmac(card);
 
 	/* release chains */
-	if (spin_trylock(&card->tx_chain.lock)) {
-		spider_net_release_tx_chain(card, 1);
-		spin_unlock(&card->tx_chain.lock);
-	}
+	spider_net_release_tx_chain(card, 1);
 
 	spider_net_free_chain(card, &card->tx_chain);
 	spider_net_free_chain(card, &card->rx_chain);
#17 lore raw

PATCH 16/21 : powerpc/cell spidernet

In reply to Linas Vepstas

Remove a dummy register read that is not needed. This reduces CPU usage notably during transmit.

Cc Arnd Bergmann <[email protected]>
Cc James K Lewis <[email protected]>

----

 drivers/net/spider_net.c |    2 --
 1 file changed, 2 deletions(-)
drivers/net/spider_net.c +0 -2
Index: linux-2.6.18-mm2/drivers/net/spider_net.c
===================================================================
@@ -759,8 +759,6 @@ spider_net_release_tx_chain(struct spide
 	unsigned long flags;
 	int status;
 
-	spider_net_read_reg(card, SPIDER_NET_GDTDMACCNTR);
-
 	while (chain->tail != chain->head) {
 		spin_lock_irqsave(&chain->lock, flags);
 		descr = chain->tail;
#18 lore raw

PATCH 17/21 : powerpc/cell spidernet reduce DMA kicking

In reply to Linas Vepstas

The current code attempts to start the TX dma every time a packet is queued. This is too conservative, and wastes CPU time. This patch changes behaviour to call the kick-dma function less often, only when the tx queue is at risk of emptying.

This reduces cpu usage, improves performance.

Cc James K Lewis <[email protected]>
Cc Arnd Bergmann <[email protected]>

----

 drivers/net/spider_net.c |   19 +++++++++++++------
 1 file changed, 13 insertions(+), 6 deletions(-)
drivers/net/spider_net.c +13 -6
Index: linux-2.6.18-mm2/drivers/net/spider_net.c
===================================================================
@@ -698,7 +698,7 @@ spider_net_prepare_tx_descr(struct spide
 	return 0;
 }
 
-static void
+static int
 spider_net_set_low_watermark(struct spider_net_card *card)
 {
 	unsigned long flags;
@@ -719,7 +719,7 @@ spider_net_set_low_watermark(struct spid
 
 	/* If TX queue is short, don't even bother with interrupts */
 	if (cnt < card->tx_desc/4)
-		return;
+		return cnt;
 
 	/* Set low-watermark 3/4th's of the way into the queue. */
 	descr = card->tx_chain.tail;
@@ -735,6 +735,7 @@ spider_net_set_low_watermark(struct spid
 		     card->low_watermark->dmac_cmd_status & ~SPIDER_NET_DESCR_TXDESFLG;
 	card->low_watermark = descr;
 	spin_unlock_irqrestore(&card->tx_chain.lock, flags);
+	return cnt;
 }
 
 /**
@@ -819,8 +820,12 @@ spider_net_release_tx_chain(struct spide
  * @card: card structure
  * @descr: descriptor address to enable TX processing at
  *
- * spider_net_kick_tx_dma writes the current tx chain head as start address
- * of the tx descriptor chain and enables the transmission DMA engine
+ * This routine will start the transmit DMA running if
+ * it is not already running. This routine ned only be
+ * called when queueing a new packet to an empty tx queue.
+ * Writes the current tx chain head as start address
+ * of the tx descriptor chain and enables the transmission
+ * DMA engine.
  */
 static inline void
 spider_net_kick_tx_dma(struct spider_net_card *card)
@@ -860,6 +865,7 @@ out:
 static int
 spider_net_xmit(struct sk_buff *skb, struct net_device *netdev)
 {
+	int cnt;
 	struct spider_net_card *card = netdev_priv(netdev);
 	struct spider_net_descr_chain *chain = &card->tx_chain;
 
@@ -873,8 +879,9 @@ spider_net_xmit(struct sk_buff *skb, str
 		return NETDEV_TX_BUSY;
 	}
 
-	spider_net_set_low_watermark(card);
-	spider_net_kick_tx_dma(card);
+	cnt = spider_net_set_low_watermark(card);
+	if (cnt < 5)
+		spider_net_kick_tx_dma(card);
 	return NETDEV_TX_OK;
 }
 
#19 lore raw

PATCH 18/21 : powerpc/cell spidernet variable name change

In reply to Linas Vepstas

Cosmetic patch: give the variable holding the numer of descriptors a more descriptive name, so to avoid confusion.

Cc James K Lewis <[email protected]>
Cc Arnd Bergmann <[email protected]>

----

 drivers/net/spider_net.c         |   12 ++++++------
 drivers/net/spider_net.h         |    4 ++--
 drivers/net/spider_net_ethtool.c |    4 ++--
 3 files changed, 10 insertions(+), 10 deletions(-)
drivers/net/spider_net.c +6 -6
Index: linux-2.6.18-mm2/drivers/net/spider_net.c
===================================================================
@@ -718,7 +718,7 @@ spider_net_set_low_watermark(struct spid
 	}
 
 	/* If TX queue is short, don't even bother with interrupts */
-	if (cnt < card->tx_desc/4)
+	if (cnt < card->num_tx_desc/4)
 		return cnt;
 
 	/* Set low-watermark 3/4th's of the way into the queue. */
@@ -1666,15 +1666,15 @@ spider_net_open(struct net_device *netde
 
 	result = -ENOMEM;
 	if (spider_net_init_chain(card, &card->tx_chain, card->descr,
-			PCI_DMA_TODEVICE, card->tx_desc))
+			PCI_DMA_TODEVICE, card->num_tx_desc))
 		goto alloc_tx_failed;
 
 	card->low_watermark = NULL;
 
 	/* rx_chain is after tx_chain, so offset is descr + tx_count */
 	if (spider_net_init_chain(card, &card->rx_chain,
-			card->descr + card->tx_desc,
-			PCI_DMA_FROMDEVICE, card->rx_desc))
+			card->descr + card->num_tx_desc,
+			PCI_DMA_FROMDEVICE, card->num_rx_desc))
 		goto alloc_rx_failed;
 
 	/* allocate rx skbs */
@@ -2060,8 +2060,8 @@ spider_net_setup_netdev(struct spider_ne
 
 	card->options.rx_csum = SPIDER_NET_RX_CSUM_DEFAULT;
 
-	card->tx_desc = tx_descriptors;
-	card->rx_desc = rx_descriptors;
+	card->num_tx_desc = tx_descriptors;
+	card->num_rx_desc = rx_descriptors;
 
 	spider_net_setup_netdev_ops(netdev);
 
drivers/net/spider_net.h +2 -2
Index: linux-2.6.18-mm2/drivers/net/spider_net.h
===================================================================
@@ -455,8 +455,8 @@ struct spider_net_card {
 
 	/* for ethtool */
 	int msg_enable;
-	int rx_desc;
-	int tx_desc;
+	int num_rx_desc;
+	int num_tx_desc;
 	struct spider_net_extra_stats spider_stats;
 
 	struct spider_net_descr descr[0];
drivers/net/spider_net_ethtool.c +2 -2
Index: linux-2.6.18-mm2/drivers/net/spider_net_ethtool.c
===================================================================
@@ -158,9 +158,9 @@ spider_net_ethtool_get_ringparam(struct 
 	struct spider_net_card *card = netdev->priv;
 
 	ering->tx_max_pending = SPIDER_NET_TX_DESCRIPTORS_MAX;
-	ering->tx_pending = card->tx_desc;
+	ering->tx_pending = card->num_tx_desc;
 	ering->rx_max_pending = SPIDER_NET_RX_DESCRIPTORS_MAX;
-	ering->rx_pending = card->rx_desc;
+	ering->rx_pending = card->num_rx_desc;
 }
 
 static int spider_net_get_stats_count(struct net_device *netdev)
#20 lore raw

PATCH 19/21 : powerpc/cell spidernet DMA direction fix

In reply to Linas Vepstas

The ring buffer descriptors are DMA-accessed bidirectionally, but are not declared in this way. Fix this.

Cc James K Lewis <[email protected]>
Cc Arnd Bergmann <[email protected]>

----

 drivers/net/spider_net.c |   26 +++++++++++++-------------
 1 file changed, 13 insertions(+), 13 deletions(-)
drivers/net/spider_net.c +13 -13
Index: linux-2.6.18-mm2/drivers/net/spider_net.c
===================================================================
@@ -301,7 +301,7 @@ static int
 spider_net_init_chain(struct spider_net_card *card,
 		       struct spider_net_descr_chain *chain,
 		       struct spider_net_descr *start_descr,
-		       int direction, int no)
+		       int no)
 {
 	int i;
 	struct spider_net_descr *descr;
@@ -316,7 +316,7 @@ spider_net_init_chain(struct spider_net_
 
 		buf = pci_map_single(card->pdev, descr,
 				     SPIDER_NET_DESCR_SIZE,
-				     direction);
+				     PCI_DMA_BIDIRECTIONAL);
 
 		if (pci_dma_mapping_error(buf))
 			goto iommu_error;
@@ -330,11 +330,6 @@ spider_net_init_chain(struct spider_net_
 	(descr-1)->next = start_descr;
 	start_descr->prev = descr-1;
 
-	descr = start_descr;
-	if (direction == PCI_DMA_FROMDEVICE)
-		for (i=0; i < no; i++, descr++)
-			descr->next_descr_addr = descr->next->bus_addr;
-
 	spin_lock_init(&chain->lock);
 	chain->head = start_descr;
 	chain->tail = start_descr;
@@ -347,7 +342,7 @@ iommu_error:
 		if (descr->bus_addr)
 			pci_unmap_single(card->pdev, descr->bus_addr,
 					 SPIDER_NET_DESCR_SIZE,
-					 direction);
+					 PCI_DMA_BIDIRECTIONAL);
 	return -ENOMEM;
 }
 
@@ -368,7 +363,7 @@ spider_net_free_rx_chain_contents(struct
 			dev_kfree_skb(descr->skb);
 			pci_unmap_single(card->pdev, descr->buf_addr,
 					 SPIDER_NET_MAX_FRAME,
-					 PCI_DMA_FROMDEVICE);
+					 PCI_DMA_BIDIRECTIONAL);
 		}
 		descr = descr->next;
 	}
@@ -1662,21 +1657,26 @@ int
 spider_net_open(struct net_device *netdev)
 {
 	struct spider_net_card *card = netdev_priv(netdev);
-	int result;
+	struct spider_net_descr *descr;
+	int i, result;
 
 	result = -ENOMEM;
 	if (spider_net_init_chain(card, &card->tx_chain, card->descr,
-			PCI_DMA_TODEVICE, card->num_tx_desc))
+	                          card->num_tx_desc))
 		goto alloc_tx_failed;
 
 	card->low_watermark = NULL;
 
 	/* rx_chain is after tx_chain, so offset is descr + tx_count */
 	if (spider_net_init_chain(card, &card->rx_chain,
-			card->descr + card->num_tx_desc,
-			PCI_DMA_FROMDEVICE, card->num_rx_desc))
+	                          card->descr + card->num_tx_desc,
+	                          card->num_rx_desc))
 		goto alloc_rx_failed;
 
+	descr = card->rx_chain.head;
+	for (i=0; i < card->num_rx_desc; i++, descr++)
+		descr->next_descr_addr = descr->next->bus_addr;
+
 	/* allocate rx skbs */
 	if (spider_net_alloc_rx_skbs(card))
 		goto alloc_skbs_failed;
#21 lore raw

PATCH 20/21 : powerpc/cell spidernet release all descrs

In reply to Linas Vepstas

Bugfix: rx descriptor release function fails to visit the last entry while walking receive descriptor ring.

Cc James K Lewis <[email protected]>
Cc Arnd Bergmann <[email protected]>

----

 drivers/net/spider_net.c |    4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/net/spider_net.c +2 -2
Index: linux-2.6.18-mm2/drivers/net/spider_net.c
===================================================================
@@ -358,7 +358,7 @@ spider_net_free_rx_chain_contents(struct
 	struct spider_net_descr *descr;
 
 	descr = card->rx_chain.head;
-	while (descr->next != card->rx_chain.head) {
+	do {
 		if (descr->skb) {
 			dev_kfree_skb(descr->skb);
 			pci_unmap_single(card->pdev, descr->buf_addr,
@@ -366,7 +366,7 @@ spider_net_free_rx_chain_contents(struct
 					 PCI_DMA_BIDIRECTIONAL);
 		}
 		descr = descr->next;
-	}
+	} while (descr != card->rx_chain.head);
 }
 
 /**
#22 lore raw

PATCH 21/21 : powerpc/cell spidernet DMA coalescing

In reply to Linas Vepstas

The current driver code performs 512 DMA mappns of a bunch of 32-byte structures. This is silly, as they are all in contiguous memory. Ths patch changes the code to DMA map the entie area with just one call.

Cc James K Lewis <[email protected]>
Cc Arnd Bergmann <[email protected]>

----

 drivers/net/spider_net.c |  107 +++++++++++++++++++++++------------------------
 drivers/net/spider_net.h |   16 ++-----
 2 files changed, 59 insertions(+), 64 deletions(-)
drivers/net/spider_net.c +53 -54
Index: linux-2.6.18-mm2/drivers/net/spider_net.c
===================================================================
@@ -267,25 +267,6 @@ spider_net_get_descr_status(struct spide
 }
 
 /**
- * spider_net_free_chain - free descriptor chain
- * @card: card structure
- * @chain: address of chain
- *
- */
-static void
-spider_net_free_chain(struct spider_net_card *card,
-		      struct spider_net_descr_chain *chain)
-{
-	struct spider_net_descr *descr;
-
-	for (descr = chain->tail; !descr->bus_addr; descr = descr->next) {
-		pci_unmap_single(card->pdev, descr->bus_addr,
-				 SPIDER_NET_DESCR_SIZE, PCI_DMA_BIDIRECTIONAL);
-		descr->bus_addr = 0;
-	}
-}
-
-/**
  * spider_net_init_chain - links descriptor chain
  * @card: card structure
  * @chain: address of chain
@@ -297,15 +278,15 @@ spider_net_free_chain(struct spider_net_
  *
  * returns 0 on success, <0 on failure
  */
-static int
+static void
 spider_net_init_chain(struct spider_net_card *card,
 		       struct spider_net_descr_chain *chain,
 		       struct spider_net_descr *start_descr,
+		       dma_addr_t buf,
 		       int no)
 {
 	int i;
 	struct spider_net_descr *descr;
-	dma_addr_t buf;
 
 	descr = start_descr;
 	memset(descr, 0, sizeof(*descr) * no);
@@ -314,17 +295,12 @@ spider_net_init_chain(struct spider_net_
 	for (i=0; i<no; i++, descr++) {
 		descr->dmac_cmd_status = SPIDER_NET_DESCR_NOT_IN_USE;
 
-		buf = pci_map_single(card->pdev, descr,
-				     SPIDER_NET_DESCR_SIZE,
-				     PCI_DMA_BIDIRECTIONAL);
-
-		if (pci_dma_mapping_error(buf))
-			goto iommu_error;
-
 		descr->bus_addr = buf;
+		descr->next_descr_addr = 0;
 		descr->next = descr + 1;
 		descr->prev = descr - 1;
 
+		buf += sizeof(struct spider_net_descr);
 	}
 	/* do actual circular list */
 	(descr-1)->next = start_descr;
@@ -333,17 +309,6 @@ spider_net_init_chain(struct spider_net_
 	spin_lock_init(&chain->lock);
 	chain->head = start_descr;
 	chain->tail = start_descr;
-
-	return 0;
-
-iommu_error:
-	descr = start_descr;
-	for (i=0; i < no; i++, descr++)
-		if (descr->bus_addr)
-			pci_unmap_single(card->pdev, descr->bus_addr,
-					 SPIDER_NET_DESCR_SIZE,
-					 PCI_DMA_BIDIRECTIONAL);
-	return -ENOMEM;
 }
 
 /**
@@ -1658,24 +1623,32 @@ spider_net_open(struct net_device *netde
 {
 	struct spider_net_card *card = netdev_priv(netdev);
 	struct spider_net_descr *descr;
-	int i, result;
+	int result = -ENOMEM;
 
-	result = -ENOMEM;
-	if (spider_net_init_chain(card, &card->tx_chain, card->descr,
-	                          card->num_tx_desc))
-		goto alloc_tx_failed;
+	card->descr_dma_addr = pci_map_single(card->pdev, card->descr,
+	        (card->num_tx_desc+card->num_rx_desc)*sizeof(struct spider_net_descr),
+				     PCI_DMA_BIDIRECTIONAL);
+	if (pci_dma_mapping_error(card->descr_dma_addr))
+		return -ENOMEM;
+
+	spider_net_init_chain(card, &card->tx_chain, card->descr,
+	                          card->descr_dma_addr,
+	                          card->num_tx_desc);
 
 	card->low_watermark = NULL;
 
 	/* rx_chain is after tx_chain, so offset is descr + tx_count */
-	if (spider_net_init_chain(card, &card->rx_chain,
+	spider_net_init_chain(card, &card->rx_chain,
 	                          card->descr + card->num_tx_desc,
-	                          card->num_rx_desc))
-		goto alloc_rx_failed;
+	                          card->descr_dma_addr
+					+ card->num_tx_desc * sizeof(struct spider_net_descr),
+	                          card->num_rx_desc);
 
 	descr = card->rx_chain.head;
-	for (i=0; i < card->num_rx_desc; i++, descr++)
+	do {
 		descr->next_descr_addr = descr->next->bus_addr;
+		descr = descr->next;
+	} while (descr != card->rx_chain.head);
 
 	/* allocate rx skbs */
 	if (spider_net_alloc_rx_skbs(card))
@@ -1701,10 +1674,21 @@ spider_net_open(struct net_device *netde
 register_int_failed:
 	spider_net_free_rx_chain_contents(card);
 alloc_skbs_failed:
-	spider_net_free_chain(card, &card->rx_chain);
-alloc_rx_failed:
-	spider_net_free_chain(card, &card->tx_chain);
-alloc_tx_failed:
+	descr = card->rx_chain.head;
+	do {
+		descr->bus_addr = 0;
+		descr = descr->next;
+	} while (descr != card->rx_chain.head);
+
+	descr = card->tx_chain.head;
+	do {
+		descr->bus_addr = 0;
+		descr = descr->next;
+	} while (descr != card->tx_chain.head);
+
+	pci_unmap_single(card->pdev, card->descr_dma_addr,
+	   (card->num_tx_desc+card->num_rx_desc)*sizeof(struct spider_net_descr),
+	                 PCI_DMA_BIDIRECTIONAL);
 	return result;
 }
 
@@ -1907,6 +1891,7 @@ int
 spider_net_stop(struct net_device *netdev)
 {
 	struct spider_net_card *card = netdev_priv(netdev);
+	struct spider_net_descr *descr;
 
 	tasklet_kill(&card->rxram_full_tl);
 	netif_poll_disable(netdev);
@@ -1930,9 +1915,23 @@ spider_net_stop(struct net_device *netde
 
 	/* release chains */
 	spider_net_release_tx_chain(card, 1);
+	spider_net_free_rx_chain_contents(card);
+
+	descr = card->rx_chain.head;
+	do {
+		descr->bus_addr = 0;
+		descr = descr->next;
+	} while (descr != card->rx_chain.head);
+
+	descr = card->tx_chain.head;
+	do {
+		descr->bus_addr = 0;
+		descr = descr->next;
+	} while (descr != card->tx_chain.head);
 
-	spider_net_free_chain(card, &card->tx_chain);
-	spider_net_free_chain(card, &card->rx_chain);
+	pci_unmap_single(card->pdev, card->descr_dma_addr,
+	      (card->num_tx_desc+card->num_rx_desc)*sizeof(struct spider_net_descr),
+	                 PCI_DMA_BIDIRECTIONAL);
 
 	return 0;
 }
drivers/net/spider_net.h +6 -10
Index: linux-2.6.18-mm2/drivers/net/spider_net.h
===================================================================
@@ -397,8 +397,6 @@ struct spider_net_descr_chain {
  * 701b8000 would be correct, but every packets gets that flag */
 #define SPIDER_NET_DESTROY_RX_FLAGS	0x700b8000
 
-#define SPIDER_NET_DESCR_SIZE		32
-
 /* this will be bigger some time */
 struct spider_net_options {
 	int rx_csum; /* for rx: if 0 ip_summed=NONE,
@@ -437,28 +435,26 @@ struct spider_net_card {
 
 	void __iomem *regs;
 
+	int num_rx_desc;
+	int num_tx_desc;
 	struct spider_net_descr_chain tx_chain;
 	struct spider_net_descr_chain rx_chain;
 	struct spider_net_descr *low_watermark;
+	dma_addr_t descr_dma_addr;
 
-	struct net_device_stats netdev_stats;
-
-	struct spider_net_options options;
-
-	spinlock_t intmask_lock;
 	struct tasklet_struct rxram_full_tl;
 	struct timer_list tx_timer;
-
 	struct work_struct tx_timeout_task;
 	atomic_t tx_timeout_task_counter;
 	wait_queue_head_t waitq;
 
 	/* for ethtool */
 	int msg_enable;
-	int num_rx_desc;
-	int num_tx_desc;
+	struct net_device_stats netdev_stats;
 	struct spider_net_extra_stats spider_stats;
+	struct spider_net_options options;
 
+	/* Must be last element in the structure */
 	struct spider_net_descr descr[0];
 };
 
#23 lore raw

Re: [PATCH 21/21]: powerpc/cell spidernet DMA coalescing

Quoted 6 lines · Linas Vepstas

The current driver code performs 512 DMA mappns of a bunch of 32-byte structures. This is silly, as they are all in contiguous memory. Ths patch changes the code to DMA map the entie area with just one call.

Cc James K Lewis <[email protected]>
Cc Arnd Bergmann <[email protected]>

The others look good, but this one complicates the code and doesn't have any benefit. 20 for 21 isn't bad.

#24 lore raw

Re: [PATCH 21/21]: powerpc/cell spidernet DMA coalescing

Quoted 6 lines · jschopp
Quoted 5 lines · Linas Vepstas

The current driver code performs 512 DMA mappns of a bunch of 32-byte structures. This is silly, as they are all in contiguous memory. Ths patch changes the code to DMA map the entie area with just one call.

Quoted 6 lines
Quoted 3 lines
Cc James K Lewis <[email protected]>
Cc Arnd Bergmann <[email protected]>

The others look good, but this one complicates the code and doesn't have any benefit. 20 for 21 isn't bad.

Linas,

Is the motivation for this change to improve performance by reducing the overhead of the mapping calls? If so, there may be some benefit for some systems. Could you please elaborate?

-Geoff

#25 lore raw

Re: [PATCH 21/21]: powerpc/cell spidernet DMA coalescing

Quoted 8 lines · Geoff Levand
Quoted 3 lines

The others look good, but this one complicates the code and doesn't have any benefit.  20 for 21 isn't bad.

Is the motivation for this change to improve performance by reducing the overhead of the mapping calls?  If so, there may be some benefit for some systems.  Could you please elaborate?

From what I understand, this patch drastically reduces the number of I/O PTEs that are needed in the iommu. With the current static IOMMU mapping, it should only make a difference during initialization, but any platform that uses a dynamic mapping of iommu entries will benefit a lot from it, because:

- the card can do better prefetching of consecutive memory
- there are more I/O ptes available for other drivers.

	Arnd <><
#26 lore raw

Re: [PATCH 21/21]: powerpc/cell spidernet DMA coalescing

In reply to Geoff Levand
Quoted 10 lines · Geoff Levand
Quoted 5 lines
Quoted 4 lines · Linas Vepstas

The current driver code performs 512 DMA mappns of a bunch of 32-byte structures. This is silly, as they are all in contiguous memory. Ths patch changes the code to DMA map the entie area with just one call.

Linas,

Is the motivation for this change to improve performance by reducing the overhead of the mapping calls?

Yes.

Quoted 2 lines

If so, there may be some benefit for some systems. Could you please elaborate?

I started writingthe patch thinking it will have some huge effect on performance, based on a false assumption on how i/o was done on this machine

*If* this were another pSeries system, then each call to pci_map_single() chews up an actual hardware "translation control entry" (TCE) that maps pci bus addresses into system RAM addresses. These are somewhat limited resources, and so one shouldn't squander them. Furthermore, I thouhght TCE's have TLB's associated with them (similar to how virtual memory page tables are backed by hardware page TLB's), of which there are even less of. I was thinking that TLB thrashing would have a big hit on performance.

Turns out that there was no difference to performance at all, and a quick look at "cell_map_single()" in arch/powerpc/platforms/cell made it clear why: there's no fancy i/o address mapping.

Thus, the patch has only mrginal benefit; I submit it only in the name of "its the right thing to do anyway".

--linas

#27 lore raw

Re: [PATCH 21/21]: powerpc/cell spidernet DMA coalescing

Quoted 31 lines · Linas Vepstas
Quoted 10 lines · Geoff Levand
Quoted 5 lines
Quoted 4 lines · Linas Vepstas

The current driver code performs 512 DMA mappns of a bunch of 32-byte structures. This is silly, as they are all in contiguous memory. Ths patch changes the code to DMA map the entie area with just one call.

Linas,

Is the motivation for this change to improve performance by reducing the overhead of the mapping calls?

Yes.

Quoted 2 lines

If so, there may be some benefit for some systems. Could you please elaborate?

I started writingthe patch thinking it will have some huge effect on performance, based on a false assumption on how i/o was done on this machine

*If* this were another pSeries system, then each call to pci_map_single() chews up an actual hardware "translation control entry" (TCE) that maps pci bus addresses into system RAM addresses. These are somewhat limited resources, and so one shouldn't squander them. Furthermore, I thouhght TCE's have TLB's associated with them (similar to how virtual memory page tables are backed by hardware page TLB's), of which there are even less of. I was thinking that TLB thrashing would have a big hit on performance.

Quoted 3 lines

Turns out that there was no difference to performance at all, and a quick look at "cell_map_single()" in arch/powerpc/platforms/cell made it clear why: there's no fancy i/o address mapping.

OK, thanks for the explanation. Actually, the current cell DMA mapping implementation uses a simple 'linear' mapping, in that, all of RAM is mapped into the bus DMA address space at once, and in fact, it is all just done at system startup.

There is ongoing work to implement 'dynamic' mapping, where DMA pages are mapped into the bus DMA address space on demand. I think a key point to understand the benefit to this is that the cell processor's I/O controller maps pages per device, so you can map one DMA page to one device. I currently have this working for my platform, but have not released that work. There is some overhead to managing the mapped buffers and to request pages be mapped by the hypervisor, etc., so I was thinking that is this work of yours to consolidate the memory buffers prior to requesting the mapping could be of benefit if it was in an often executed code path.

-Geoff

#28 lore raw

Re: [PATCH 21/21]: powerpc/cell spidernet DMA coalescing

In reply to Linas Vepstas
Quoted 15 lines

I started writingthe patch thinking it will have some huge effect on performance, based on a false assumption on how i/o was done on this machine

*If* this were another pSeries system, then each call to pci_map_single() chews up an actual hardware "translation control entry" (TCE) that maps pci bus addresses into system RAM addresses. These are somewhat limited resources, and so one shouldn't squander them. Furthermore, I thouhght TCE's have TLB's associated with them (similar to how virtual memory page tables are backed by hardware page TLB's), of which there are even less of. I was thinking that TLB thrashing would have a big hit on performance.

Quoted 6 lines

Turns out that there was no difference to performance at all, and a quick look at "cell_map_single()" in arch/powerpc/platforms/cell made it clear why: there's no fancy i/o address mapping.

Thus, the patch has only mrginal benefit; I submit it only in the name of "its the right thing to do anyway".

Well, there is no fancy iommu mapping ... yet.

It's been implemented and is coming after we put together some workarounds for various other spider hardware issues that trigger when using it (bogus prefetches and bogus pci ordering).

I think the hypervisor based platforms will be happy with that patch too.

Ben.

#29 lore raw

Re: [PATCH 21/21]: powerpc/cell spidernet DMA coalescing

In reply to jschopp
Quoted 6 lines · jschopp
Quoted 5 lines · Linas Vepstas

The current driver code performs 512 DMA mappns of a bunch of 32-byte structures. This is silly, as they are all in contiguous memory. Ths patch changes the code to DMA map the entie area with just one call.

Quoted 6 lines
Quoted 3 lines
Cc James K Lewis <[email protected]>
Cc Arnd Bergmann <[email protected]>

The others look good, but this one complicates the code and doesn't have any benefit. 20 for 21 isn't bad.

Hi Joel !

I'm not sure what you mean here.... (especially your 20 to 21 comment).

The patch looks perfectly fine to me, and in fact removes more lines of code than it adds :)

Cheers,
Ben.

Acked-by on v1 21/21
#30 lore raw

Re: [PATCH 21/21]: powerpc/cell spidernet DMA coalescing

In reply to Linas Vepstas
Quoted 8 lines · Linas Vepstas

The current driver code performs 512 DMA mappns of a bunch of 32-byte structures. This is silly, as they are all in contiguous memory. Ths patch changes the code to DMA map the entie area with just one call.

Cc James K Lewis <[email protected]>
Cc Arnd Bergmann <[email protected]>

While this patch wasn't useful in the current cell implementation of pci_map_single it sounds like people are going to be making changes to that sometime. In light of that new information (new to me anyway) this should probably go in after all. Sorry for causing trouble.

#31 lore raw
In reply to Linas Vepstas
Quoted 31 lines · Linas Vepstas

Andrew, please apply/forward upstream.

The following set of 21 patches (!) are all aimed at the the spidernet ethernet device driver. The spidernet is an etherenet controller built into the Toshiba southbridge for the PowerPC Cell processor. (This is the only device in existance that with this ethernet hardware in it).

These patches re-package/re-order/re-cleanup a previous set of patches I've previously mailed. Thus, some have been previously Acked-by lines, most do not. Most of these patches are tiny, and handle problems that cropped up during testing. Sorry about there being so many of them.

The first set of 12 patches fix a large variety of mostly minor bugs.

The important patches are 13 through 17: these overcome a debilitating performance problem on transmit (6 megabits per second !!) on transmit of patches 500 bytes or larger. After applying these, I am able to get the following:

pkt sz   speed (100K buffs)       speed (4M buffs) ------   -----------------        ---------------- 1500     700 Mbits/sec            951 Mbits/sec 1000     658 Mbits/sec            770
800      600                      648
500      500                      500
300      372                      372
60        70                       70

Quoted 1 line

Above buf size refers to /proc/sys/net/core/wmem_default

Excellent work! I guess this the best tx performance we've seen so far on this hardware.

Consider this as an Acked-by: for all the patches, I'll save the effort of replying to each one of them separately.

Jeff, do you plan on merging these fixes for 2.6.19?

	Arnd <><
#32 lore raw

Please don't be confused by the numbers in the 4M column, they don't
mean anything to the end user. We had a bit better performance (approx. 720 Mbps) at one time but at 100% CPU usage. These new patches lower the max. to about 700 Mbps but decrease CPU usage down to about 30% which I think will help all around. I have been testing this driver all night and have not found a problem yet. Consider this my Acked-by: for all these patches.

Jim Lewis
Advisory Software Engineer
IBM Linux Technology Center
512-838-7754

Arnd Bergmann <[email protected]>
10/11/2006 11:02 AM

To
Linas Vepstas <[email protected]>
cc
[email protected], [email protected], James K Lewis/Austin/IBM@ibmus, [email protected], [email protected], [email protected]
Subject
Re: [PATCH 0/21]: powerpc/cell spidernet bugfixes, etc.

Quoted 29 lines · Linas Vepstas

Andrew, please apply/forward upstream.

The following set of 21 patches (!) are all aimed at the the spidernet ethernet device driver. The spidernet is an etherenet controller built into the Toshiba southbridge for the PowerPC Cell processor. (This is the only device in existance that with this ethernet hardware in it).

These patches re-package/re-order/re-cleanup a previous set of patches I've previously mailed. Thus, some have been previously Acked-by lines, most do not. Most of these patches are tiny, and handle problems that cropped up during testing. Sorry about there being so many of them.

The first set of 12 patches fix a large variety of mostly minor bugs.

The important patches are 13 through 17: these overcome a debilitating performance problem on transmit (6 megabits per second !!) on transmit of patches 500 bytes or larger. After applying these, I am able to get the following:

pkt sz   speed (100K buffs)       speed (4M buffs)
------   -----------------        ----------------
1500     700 Mbits/sec            951 Mbits/sec
1000     658 Mbits/sec            770
800      600                      648
500      500                      500
300      372                      372
Quoted 3 lines

60 70 70

Above buf size refers to /proc/sys/net/core/wmem_default

Excellent work! I guess this the best tx performance we've seen so far on this hardware.

Consider this as an Acked-by: for all the patches, I'll save the effort of replying to each one of them separately.

Jeff, do you plan on merging these fixes for 2.6.19?

                 Arnd <><

_______________________________________________
Linuxppc-dev mailing list
[email protected]
https://ozlabs.org/mailman/listinfo/linuxppc-dev

#33 lore raw
In reply to Arnd Bergmann

Jeff,

Please apply and forward these Spidernet patches. Thanks!

Jim Lewis
Advisory Software Engineer
IBM Linux Technology Center
512-838-7754

Arnd Bergmann <[email protected]>
10/11/2006 11:02 AM

To
Linas Vepstas <[email protected]>
cc
[email protected], [email protected], James K Lewis/Austin/IBM@ibmus, [email protected], [email protected], [email protected]
Subject
Re: [PATCH 0/21]: powerpc/cell spidernet bugfixes, etc.

Quoted 29 lines · Linas Vepstas

Andrew, please apply/forward upstream.

The following set of 21 patches (!) are all aimed at the the spidernet ethernet device driver. The spidernet is an etherenet controller built into the Toshiba southbridge for the PowerPC Cell processor. (This is the only device in existance that with this ethernet hardware in it).

These patches re-package/re-order/re-cleanup a previous set of patches I've previously mailed. Thus, some have been previously Acked-by lines, most do not. Most of these patches are tiny, and handle problems that cropped up during testing. Sorry about there being so many of them.

The first set of 12 patches fix a large variety of mostly minor bugs.

The important patches are 13 through 17: these overcome a debilitating performance problem on transmit (6 megabits per second !!) on transmit of patches 500 bytes or larger. After applying these, I am able to get the following:

pkt sz   speed (100K buffs)       speed (4M buffs)
------   -----------------        ----------------
1500     700 Mbits/sec            951 Mbits/sec
1000     658 Mbits/sec            770
800      600                      648
500      500                      500
300      372                      372
Quoted 3 lines

60 70 70

Above buf size refers to /proc/sys/net/core/wmem_default

Excellent work! I guess this the best tx performance we've seen so far on this hardware.

Consider this as an Acked-by: for all the patches, I'll save the effort of replying to each one of them separately.

Jeff, do you plan on merging these fixes for 2.6.19?

                 Arnd <><

_______________________________________________
Linuxppc-dev mailing list
[email protected]
https://ozlabs.org/mailman/listinfo/linuxppc-dev