Re: Validation of DMA params breaks e100 driver (2.6.28-rc2)

Re: Validation of DMA params breaks e100 driver (2.6.28-rc2)

[Russell King - ARM Linux]

On Thu, Oct 30, 2008 at 05:27:59PM +0100, Anders Grafström wrote:
> The e100 driver triggers BUG_ON(buf->direction != dir)
> by doing pci_map_single(..., PCI_DMA_BIDIRECTIONAL)
> and pci_dma_sync_single_for_device(..., PCI_DMA_TODEVICE).
> 
> I'm guessing it's allowed to do that and that something like
> the patch below is called for?

No, it is not allowed to do that - that's why it's called "BUG_ON".
Changing the DMA direction, especially with dmabounce will result
in unexpected behaviour.

> -----------------
> 
> [PATCH] [ARM] dma: Fix DMA params validation
> 
> This patch makes the DMA params validation less strict
> and more like the generic implementation.
> 
> Signed-off-by: Anders Grafström <grfstrm@users.sourceforge.net>
> ---
>  arch/arm/common/dmabounce.c |   13 ++++++-------
>  1 files changed, 6 insertions(+), 7 deletions(-)
> 
> diff --git a/arch/arm/common/dmabounce.c b/arch/arm/common/dmabounce.c
> index f030f07..a0c9c7b 100644
> --- a/arch/arm/common/dmabounce.c
> +++ b/arch/arm/common/dmabounce.c
> @@ -289,7 +289,6 @@ static inline void unmap_single(struct device *dev, dma_addr_t dma_addr,
> 
>  	if (buf) {
>  		BUG_ON(buf->size != size);
> -		BUG_ON(buf->direction != dir);
> 
>  		dev_dbg(dev,
>  			"%s: unsafe buffer %p (dma=%#x) mapped to %p (dma=%#x)\n",
> @@ -382,8 +381,6 @@ int dmabounce_sync_for_cpu(struct device *dev, dma_addr_t addr,
>  	if (!buf)
>  		return 1;
> 
> -	BUG_ON(buf->direction != dir);
> -
>  	dev_dbg(dev, "%s: unsafe buffer %p (dma=%#x) mapped to %p (dma=%#x)\n",
>  		__func__, buf->ptr, virt_to_dma(dev, buf->ptr),
>  		buf->safe, buf->safe_dma_addr);
> @@ -394,7 +391,9 @@ int dmabounce_sync_for_cpu(struct device *dev, dma_addr_t addr,
>  		dev_dbg(dev, "%s: copy back safe %p to unsafe %p size %d\n",
>  			__func__, buf->safe + off, buf->ptr + off, sz);
>  		memcpy(buf->ptr + off, buf->safe + off, sz);
> -	}
> +	} else
> +		BUG_ON(dir != DMA_TO_DEVICE);
> +
>  	return 0;
>  }
>  EXPORT_SYMBOL(dmabounce_sync_for_cpu);
> @@ -411,8 +410,6 @@ int dmabounce_sync_for_device(struct device *dev, dma_addr_t addr,
>  	if (!buf)
>  		return 1;
> 
> -	BUG_ON(buf->direction != dir);
> -
>  	dev_dbg(dev, "%s: unsafe buffer %p (dma=%#x) mapped to %p (dma=%#x)\n",
>  		__func__, buf->ptr, virt_to_dma(dev, buf->ptr),
>  		buf->safe, buf->safe_dma_addr);
> @@ -423,7 +420,9 @@ int dmabounce_sync_for_device(struct device *dev, dma_addr_t addr,
>  		dev_dbg(dev, "%s: copy out unsafe %p to safe %p, size %d\n",
>  			__func__,buf->ptr + off, buf->safe + off, sz);
>  		memcpy(buf->safe + off, buf->ptr + off, sz);
> -	}
> +	} else
> +		BUG_ON(dir != DMA_FROM_DEVICE);
> +
>  	return 0;
>  }
>  EXPORT_SYMBOL(dmabounce_sync_for_device);
> -- 
> 1.5.6.5
> 
> 

Re: Validation of DMA params breaks e100 driver (2.6.28-rc2)

[Brandeburg, Jesse]

[-- Attachment #1: Type: TEXT/PLAIN, Size: 5403 bytes --]

On Fri, 31 Oct 2008, Russell King - ARM Linux wrote:

> On Thu, Oct 30, 2008 at 05:27:59PM +0100, Anders Grafström wrote:
> > The e100 driver triggers BUG_ON(buf->direction != dir)
> > by doing pci_map_single(..., PCI_DMA_BIDIRECTIONAL)
> > and pci_dma_sync_single_for_device(..., PCI_DMA_TODEVICE).
> > 
> > I'm guessing it's allowed to do that and that something like
> > the patch below is called for?
> 
> No, it is not allowed to do that - that's why it's called "BUG_ON".
> Changing the DMA direction, especially with dmabounce will result
> in unexpected behaviour.

okay, how about this patch...  only compile tested as I couldn't get 
net-next-2.6 to boot on my test machine.  I'll get some testing done on 
this, but in the meantime....

e100: fix dma error in direction for mapping

From: Jesse Brandeburg <jesse.brandeburg@intel.com>

Russell King - ARM Linux wrote:
> On Thu, Oct 30, 2008 at 05:27:59PM +0100, Anders Grafström wrote:
>> The e100 driver triggers BUG_ON(buf->direction != dir)
>> by doing pci_map_single(..., PCI_DMA_BIDIRECTIONAL)
>> and pci_dma_sync_single_for_device(..., PCI_DMA_TODEVICE).
>> 
>> I'm guessing it's allowed to do that and that something like
>> the patch below is called for?
> 
> No, it is not allowed to do that - that's why it's called "BUG_ON".
> Changing the DMA direction, especially with dmabounce will result
> in unexpected behaviour.

This is my attempt to fix this issue:

Reported by: Anders Grafstrom <grfstrm@users.sourceforge.net>
CC: Russell King <linux@arm.linux.org.uk>
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
---

 drivers/net/e100.c |   20 ++++++++++----------
 1 files changed, 10 insertions(+), 10 deletions(-)

diff --git a/drivers/net/e100.c b/drivers/net/e100.c
index 62cdefa..c9c7079 100644
--- a/drivers/net/e100.c
+++ b/drivers/net/e100.c
@@ -166,7 +166,7 @@
 
 #define DRV_NAME		"e100"
 #define DRV_EXT			"-NAPI"
-#define DRV_VERSION		"3.5.23-k4"DRV_EXT
+#define DRV_VERSION		"3.5.23-k6"DRV_EXT
 #define DRV_DESCRIPTION		"Intel(R) PRO/100 Network Driver"
 #define DRV_COPYRIGHT		"Copyright(c) 1999-2006 Intel Corporation"
 #define PFX			DRV_NAME ": "
@@ -1804,7 +1804,7 @@ static int e100_rx_alloc_skb(struct nic *nic, struct rx *rx)
 		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);
+			sizeof(struct rfd), PCI_DMA_BIDIRECTIONAL);
 	}
 
 	return 0;
@@ -1823,7 +1823,7 @@ static int e100_rx_indicate(struct nic *nic, struct rx *rx,
 
 	/* Need to sync before taking a peek at cb_complete bit */
 	pci_dma_sync_single_for_cpu(nic->pdev, rx->dma_addr,
-		sizeof(struct rfd), PCI_DMA_FROMDEVICE);
+		sizeof(struct rfd), PCI_DMA_BIDIRECTIONAL);
 	rfd_status = le16_to_cpu(rfd->status);
 
 	DPRINTK(RX_STATUS, DEBUG, "status=0x%04X\n", rfd_status);
@@ -1850,7 +1850,7 @@ static int e100_rx_indicate(struct nic *nic, struct rx *rx,
 
 	/* Get data */
 	pci_unmap_single(nic->pdev, rx->dma_addr,
-		RFD_BUF_LEN, PCI_DMA_FROMDEVICE);
+		RFD_BUF_LEN, PCI_DMA_BIDIRECTIONAL);
 
 	/* If this buffer has the el bit, but we think the receiver
 	 * is still running, check to see if it really stopped while
@@ -1942,7 +1942,7 @@ static void e100_rx_clean(struct nic *nic, unsigned int *work_done,
 		new_before_last_rfd->command |= cpu_to_le16(cb_el);
 		pci_dma_sync_single_for_device(nic->pdev,
 			new_before_last_rx->dma_addr, sizeof(struct rfd),
-			PCI_DMA_TODEVICE);
+			PCI_DMA_BIDIRECTIONAL);
 
 		/* Now that we have a new stopping point, we can clear the old
 		 * stopping point.  We must sync twice to get the proper
@@ -1950,11 +1950,11 @@ static void e100_rx_clean(struct nic *nic, unsigned int *work_done,
 		old_before_last_rfd->command &= ~cpu_to_le16(cb_el);
 		pci_dma_sync_single_for_device(nic->pdev,
 			old_before_last_rx->dma_addr, sizeof(struct rfd),
-			PCI_DMA_TODEVICE);
+			PCI_DMA_BIDIRECTIONAL);
 		old_before_last_rfd->size = cpu_to_le16(VLAN_ETH_FRAME_LEN);
 		pci_dma_sync_single_for_device(nic->pdev,
 			old_before_last_rx->dma_addr, sizeof(struct rfd),
-			PCI_DMA_TODEVICE);
+			PCI_DMA_BIDIRECTIONAL);
 	}
 
 	if(restart_required) {
@@ -1977,7 +1977,7 @@ static void e100_rx_clean_list(struct nic *nic)
 		for(rx = nic->rxs, i = 0; i < count; rx++, i++) {
 			if(rx->skb) {
 				pci_unmap_single(nic->pdev, rx->dma_addr,
-					RFD_BUF_LEN, PCI_DMA_FROMDEVICE);
+					RFD_BUF_LEN, PCI_DMA_BIDIRECTIONAL);
 				dev_kfree_skb(rx->skb);
 			}
 		}
@@ -2020,7 +2020,7 @@ static int e100_rx_alloc_list(struct nic *nic)
 	before_last->command |= cpu_to_le16(cb_el);
 	before_last->size = 0;
 	pci_dma_sync_single_for_device(nic->pdev, rx->dma_addr,
-		sizeof(struct rfd), PCI_DMA_TODEVICE);
+		sizeof(struct rfd), PCI_DMA_BIDIRECTIONAL);
 
 	nic->rx_to_use = nic->rx_to_clean = nic->rxs;
 	nic->ru_running = RU_SUSPENDED;
@@ -2221,7 +2221,7 @@ static int e100_loopback_test(struct nic *nic, enum loopback loopback_mode)
 	msleep(10);
 
 	pci_dma_sync_single_for_cpu(nic->pdev, nic->rx_to_clean->dma_addr,
-			RFD_BUF_LEN, PCI_DMA_FROMDEVICE);
+			RFD_BUF_LEN, PCI_DMA_BIDIRECTIONAL);
 
 	if(memcmp(nic->rx_to_clean->skb->data + sizeof(struct rfd),
 	   skb->data, ETH_DATA_LEN))

[-- Attachment #2: Type: text/plain, Size: 363 bytes --]

-------------------------------------------------------------------------
This SF.Net email is sponsored by the Moblin Your Move Developer's challenge
Build the coolest Linux based applications with Moblin SDK & win great prizes
Grand prize is a trip for two to an Open Source event anywhere in the world
http://moblin-contest.org/redirect.php?banner_id=100&url=/

[-- Attachment #3: Type: text/plain, Size: 164 bytes --]

_______________________________________________
E1000-devel mailing list
E1000-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/e1000-devel

Re: Validation of DMA params breaks e100 driver (2.6.28-rc2)

[David Miller]

From: "Brandeburg, Jesse" <jesse.brandeburg@intel.com>
Date: Thu, 13 Nov 2008 17:21:50 -0800 (Pacific Standard Time)

> On Fri, 31 Oct 2008, Russell King - ARM Linux wrote:
> 
> > On Thu, Oct 30, 2008 at 05:27:59PM +0100, Anders Grafström wrote:
> > > The e100 driver triggers BUG_ON(buf->direction != dir)
> > > by doing pci_map_single(..., PCI_DMA_BIDIRECTIONAL)
> > > and pci_dma_sync_single_for_device(..., PCI_DMA_TODEVICE).
> > > 
> > > I'm guessing it's allowed to do that and that something like
> > > the patch below is called for?
> > 
> > No, it is not allowed to do that - that's why it's called "BUG_ON".
> > Changing the DMA direction, especially with dmabounce will result
> > in unexpected behaviour.
> 
> okay, how about this patch...  only compile tested as I couldn't get 
> net-next-2.6 to boot on my test machine.  I'll get some testing done on 
> this, but in the meantime....
> 
> e100: fix dma error in direction for mapping
> 
> From: Jesse Brandeburg <jesse.brandeburg@intel.com>

FWIW this looks good to me.

Acked-by: David S. Miller <davem@davemloft.net>

Re: Validation of DMA params breaks e100 driver (2.6.28-rc2)

[Anders Grafström]

Brandeburg, Jesse wrote:
> On Fri, 31 Oct 2008, Russell King - ARM Linux wrote:
> 
>> On Thu, Oct 30, 2008 at 05:27:59PM +0100, Anders Grafström wrote:
>>> The e100 driver triggers BUG_ON(buf->direction != dir)
>>> by doing pci_map_single(..., PCI_DMA_BIDIRECTIONAL)
>>> and pci_dma_sync_single_for_device(..., PCI_DMA_TODEVICE).
>>>
>>> I'm guessing it's allowed to do that and that something like
>>> the patch below is called for?
>> No, it is not allowed to do that - that's why it's called "BUG_ON".
>> Changing the DMA direction, especially with dmabounce will result
>> in unexpected behaviour.
> 
> okay, how about this patch...  only compile tested as I couldn't get 
> net-next-2.6 to boot on my test machine.  I'll get some testing done on 
> this, but in the meantime....
> 
> e100: fix dma error in direction for mapping
> 
> From: Jesse Brandeburg <jesse.brandeburg@intel.com>

I failed to save this patch (flowed format?) but I applied it manually
to 2.6.28-rc4 and tried it. Appears to work.


-------------------------------------------------------------------
List admin: http://lists.arm.linux.org.uk/mailman/listinfo/linux-arm-kernel
FAQ:        http://www.arm.linux.org.uk/mailinglists/faq.php
Etiquette:  http://www.arm.linux.org.uk/mailinglists/etiquette.php

Re: Validation of DMA params breaks e100 driver (2.6.28-rc2)

[Brandeburg, Jesse]

Anders Grafström wrote:
> Brandeburg, Jesse wrote:
>> On Fri, 31 Oct 2008, Russell King - ARM Linux wrote:
>>
>>> On Thu, Oct 30, 2008 at 05:27:59PM +0100, Anders Grafström wrote:
>>>> The e100 driver triggers BUG_ON(buf->direction != dir)
>>>> by doing pci_map_single(..., PCI_DMA_BIDIRECTIONAL)
>>>> and pci_dma_sync_single_for_device(..., PCI_DMA_TODEVICE).
>>>>
>>>> I'm guessing it's allowed to do that and that something like
>>>> the patch below is called for?
>>> No, it is not allowed to do that - that's why it's called "BUG_ON".
>>> Changing the DMA direction, especially with dmabounce will result
>>> in unexpected behaviour.
>>
>> okay, how about this patch...  only compile tested as I couldn't get
>> net-next-2.6 to boot on my test machine.  I'll get some testing done
>> on this, but in the meantime....
>>
>> e100: fix dma error in direction for mapping
>>
>> From: Jesse Brandeburg <jesse.brandeburg@intel.com>
>
> I failed to save this patch (flowed format?) but I applied it manually
> to 2.6.28-rc4 and tried it. Appears to work.

Strange... sorry for the thrash on the patch, and thanks for testing!


-------------------------------------------------------------------------
This SF.Net email is sponsored by the Moblin Your Move Developer's challenge
Build the coolest Linux based applications with Moblin SDK & win great prizes
Grand prize is a trip for two to an Open Source event anywhere in the world
http://moblin-contest.org/redirect.php?banner_id=100&url=/