e1000_clean_tx_ring

e1000_clean_tx_ring

[Anton Blanchard]

Hi,

I was looking over the e1000 driver and noticed what I think is a bug
in e1000_clean_tx_ring. We wouldnt call pci_unmap_page on tx ring
entries that didnt have ->skb filled, eg zero copy packets.

This is on latest 2.6 BK.

Anton

===== e1000_main.c 1.120 vs edited =====
--- 1.120/drivers/net/e1000/e1000_main.c	Sat Jun 19 10:00:00 2004
+++ edited/e1000_main.c	Thu Jun 24 02:16:42 2004
@@ -1070,14 +1070,19 @@
 
 	for(i = 0; i < tx_ring->count; i++) {
 		buffer_info = &tx_ring->buffer_info[i];
-		if(buffer_info->skb) {
+		if (buffer_info->dma) {
 
 			pci_unmap_page(pdev,
 			               buffer_info->dma,
 			               buffer_info->length,
 			               PCI_DMA_TODEVICE);
 
-			dev_kfree_skb(buffer_info->skb);
+			buffer_info->dma = NULL;
+		}
+
+		if (buffer_info->skb) {
+
+			dev_kfree_skb_any(buffer_info->skb);
 
 			buffer_info->skb = NULL;
 		}

RE: e1000_clean_tx_ring

[Venkatesan, Ganesh]

Anton:

We will integrate this patch into the next release of our driver.

Thanks,
ganesh 
 
-------------------------------------------------
Ganesh Venkatesan
Network/Storage Division, Hillsboro, OR

-----Original Message-----
From: Anton Blanchard [mailto:anton@samba.org] 
Sent: Thursday, June 24, 2004 4:23 PM
To: netdev@oss.sgi.com
Cc: cramerj; Ronciak, John; Venkatesan, Ganesh
Subject: e1000_clean_tx_ring


Hi,

I was looking over the e1000 driver and noticed what I think is a bug
in e1000_clean_tx_ring. We wouldnt call pci_unmap_page on tx ring
entries that didnt have ->skb filled, eg zero copy packets.

This is on latest 2.6 BK.

Anton

===== e1000_main.c 1.120 vs edited =====
--- 1.120/drivers/net/e1000/e1000_main.c	Sat Jun 19 10:00:00 2004
+++ edited/e1000_main.c	Thu Jun 24 02:16:42 2004
@@ -1070,14 +1070,19 @@
 
 	for(i = 0; i < tx_ring->count; i++) {
 		buffer_info = &tx_ring->buffer_info[i];
-		if(buffer_info->skb) {
+		if (buffer_info->dma) {
 
 			pci_unmap_page(pdev,
 			               buffer_info->dma,
 			               buffer_info->length,
 			               PCI_DMA_TODEVICE);
 
-			dev_kfree_skb(buffer_info->skb);
+			buffer_info->dma = NULL;
+		}
+
+		if (buffer_info->skb) {
+
+			dev_kfree_skb_any(buffer_info->skb);
 
 			buffer_info->skb = NULL;
 		}

Re: e1000_clean_tx_ring

[David Greaves]

FYI
It hasn't helped with either of my problems; watchdog timeout or mtu 
9000 malloc

David

Venkatesan, Ganesh wrote:

>Anton:
>
>We will integrate this patch into the next release of our driver.
>
>Thanks,
>ganesh 
> 
>-------------------------------------------------
>Ganesh Venkatesan
>Network/Storage Division, Hillsboro, OR
>
>-----Original Message-----
>From: Anton Blanchard [mailto:anton@samba.org] 
>Sent: Thursday, June 24, 2004 4:23 PM
>To: netdev@oss.sgi.com
>Cc: cramerj; Ronciak, John; Venkatesan, Ganesh
>Subject: e1000_clean_tx_ring
>
>
>Hi,
>
>I was looking over the e1000 driver and noticed what I think is a bug
>in e1000_clean_tx_ring. We wouldnt call pci_unmap_page on tx ring
>entries that didnt have ->skb filled, eg zero copy packets.
>
>This is on latest 2.6 BK.
>
>Anton
>
>===== e1000_main.c 1.120 vs edited =====
>--- 1.120/drivers/net/e1000/e1000_main.c	Sat Jun 19 10:00:00 2004
>+++ edited/e1000_main.c	Thu Jun 24 02:16:42 2004
>@@ -1070,14 +1070,19 @@
> 
> 	for(i = 0; i < tx_ring->count; i++) {
> 		buffer_info = &tx_ring->buffer_info[i];
>-		if(buffer_info->skb) {
>+		if (buffer_info->dma) {
> 
> 			pci_unmap_page(pdev,
> 			               buffer_info->dma,
> 			               buffer_info->length,
> 			               PCI_DMA_TODEVICE);
> 
>-			dev_kfree_skb(buffer_info->skb);
>+			buffer_info->dma = NULL;
>+		}
>+
>+		if (buffer_info->skb) {
>+
>+			dev_kfree_skb_any(buffer_info->skb);
> 
> 			buffer_info->skb = NULL;
> 		}
>
>
>
>
>
>  
>