[PATCH] e100: Do pci_dma_sync after skb_alloc for proper operation on ixp4xx

[PATCH] e100: Do pci_dma_sync after skb_alloc for proper operation on ixp4xx

[Kevin Hao]

Hi,

The E100 device can't work on current kernel (2.6.26-rc6) and will cause
kernel corruption on intel ixdp4xx. This patch fix this regression.

---
commit c1f24812b9de182a4272724a2e84ea1ae5dd4659
Author: Kevin Hao <kexin.hao@windriver.com>
Date:   Sat Jun 21 16:50:50 2008 +0800

    e100: Do pci_dma_sync after skb_alloc for proper operation on ixp4xx
    
    Do a DMA sync after we've allocated a new skb and link it with the previous
    RFD.  Otherwise the previous RFD may reference the previously freed skb
    and cause kernel corruption.
    
    Previously there was such a DMA sync at this place, but it was removed
    by commit 7734f6e6bcd7ba78b00e93e74a4ddafd9886cdea.
    Restoring this sync operation eliminates the regression.
    
    Signed-off-by: Kevin Hao <kexin.hao@windriver.com>

diff --git a/drivers/net/e100.c b/drivers/net/e100.c
index f3cba5e..1037b13 100644
--- a/drivers/net/e100.c
+++ b/drivers/net/e100.c
@@ -1803,6 +1803,8 @@ static int e100_rx_alloc_skb(struct nic *nic, struct rx *rx)
 	if (rx->prev->skb) {
 		struct rfd *prev_rfd = (struct rfd *)rx->prev->skb->data;
 		put_unaligned_le32(rx->dma_addr, &prev_rfd->link);
+		pci_dma_sync_single_for_device(nic->pdev, rx->prev->dma_addr,
+			sizeof(struct rfd), PCI_DMA_TODEVICE);
 	}
 
 	return 0;
---

Best Regards,
Kevin

Re: [PATCH] e100: Do pci_dma_sync after skb_alloc for proper operation on ixp4xx

[David Acker]

Kevin Hao wrote:
> Hi,
> 
> The E100 device can't work on current kernel (2.6.26-rc6) and will cause
> kernel corruption on intel ixdp4xx. This patch fix this regression.
> 
Hello, this is pretty much the same patch that Anders Grafström sent on 
6/18.  I think it looks good; I am sorry I missed this.  I think this 
can be ok'ed if folks can test it on a few different ARM platforms and 
of course x86.

I am getting sinus surgery this Monday but I should be back in the 
office within a week.  I will give it a try then.  I will test it on an 
ixp425 board and an PXA255 board.  Just so you know, I found the 
in-kernel packet generator to be a good test.  I also found using the 
board as a router between wired and wireless devices to catch issues 
that led me to posting patches for this driver in the first place.

-ack

> ---
> commit c1f24812b9de182a4272724a2e84ea1ae5dd4659
> Author: Kevin Hao <kexin.hao@windriver.com>
> Date:   Sat Jun 21 16:50:50 2008 +0800
> 
>     e100: Do pci_dma_sync after skb_alloc for proper operation on ixp4xx
>     
>     Do a DMA sync after we've allocated a new skb and link it with the previous
>     RFD.  Otherwise the previous RFD may reference the previously freed skb
>     and cause kernel corruption.
>     
>     Previously there was such a DMA sync at this place, but it was removed
>     by commit 7734f6e6bcd7ba78b00e93e74a4ddafd9886cdea.
>     Restoring this sync operation eliminates the regression.
>     
>     Signed-off-by: Kevin Hao <kexin.hao@windriver.com>
> 
> diff --git a/drivers/net/e100.c b/drivers/net/e100.c
> index f3cba5e..1037b13 100644
> --- a/drivers/net/e100.c
> +++ b/drivers/net/e100.c
> @@ -1803,6 +1803,8 @@ static int e100_rx_alloc_skb(struct nic *nic, struct rx *rx)
>  	if (rx->prev->skb) {
>  		struct rfd *prev_rfd = (struct rfd *)rx->prev->skb->data;
>  		put_unaligned_le32(rx->dma_addr, &prev_rfd->link);
> +		pci_dma_sync_single_for_device(nic->pdev, rx->prev->dma_addr,
> +			sizeof(struct rfd), PCI_DMA_TODEVICE);
>  	}
>  
>  	return 0;
> ---
> 
> Best Regards,
> Kevin
> 

Re: [PATCH] e100: Do pci_dma_sync after skb_alloc for proper operation on ixp4xx

[Jeff Garzik]

Kevin Hao wrote:
> Hi,
> 
> The E100 device can't work on current kernel (2.6.26-rc6) and will cause
> kernel corruption on intel ixdp4xx. This patch fix this regression.
> 
> ---
> commit c1f24812b9de182a4272724a2e84ea1ae5dd4659
> Author: Kevin Hao <kexin.hao@windriver.com>
> Date:   Sat Jun 21 16:50:50 2008 +0800
> 
>     e100: Do pci_dma_sync after skb_alloc for proper operation on ixp4xx
>     
>     Do a DMA sync after we've allocated a new skb and link it with the previous
>     RFD.  Otherwise the previous RFD may reference the previously freed skb
>     and cause kernel corruption.
>     
>     Previously there was such a DMA sync at this place, but it was removed
>     by commit 7734f6e6bcd7ba78b00e93e74a4ddafd9886cdea.
>     Restoring this sync operation eliminates the regression.
>     
>     Signed-off-by: Kevin Hao <kexin.hao@windriver.com>
> 
> diff --git a/drivers/net/e100.c b/drivers/net/e100.c
> index f3cba5e..1037b13 100644
> --- a/drivers/net/e100.c
> +++ b/drivers/net/e100.c
> @@ -1803,6 +1803,8 @@ static int e100_rx_alloc_skb(struct nic *nic, struct rx *rx)
>  	if (rx->prev->skb) {
>  		struct rfd *prev_rfd = (struct rfd *)rx->prev->skb->data;
>  		put_unaligned_le32(rx->dma_addr, &prev_rfd->link);
> +		pci_dma_sync_single_for_device(nic->pdev, rx->prev->dma_addr,
> +			sizeof(struct rfd), PCI_DMA_TODEVICE);

applied