Skip to contentSkip to conversation

at91_mci: don't use coherent dma buffers

PATCH In review David Brownell 4 participants 6 messages Aug 18, 2008 – Aug 27, 2008

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

PATCH 1/1 at91_mci: don't use coherent dma buffers

From: David Brownell <[email protected]>

At91_mci is abusing dma_free_coherent(), which may not be called with IRQs disabled. I saw "mkfs.ext3" on an MMC card objecting voluminously as each write completed:

 WARNING: at arch/arm/mm/consistent.c:368 dma_free_coherent+0x2c/0x224()
 [<c002726c>] (dump_stack+0x0/0x14) from [<c00387d4>] (warn_on_slowpath+0x4c/0x68)
 [<c0038788>] (warn_on_slowpath+0x0/0x68) from [<c0028768>] (dma_free_coherent+0x2c/0x224)
  r6:00008008 r5:ffc06000 r4:00000000
 [<c002873c>] (dma_free_coherent+0x0/0x224) from [<c01918ac>] (at91_mci_irq+0x374/0x420)
 [<c0191538>] (at91_mci_irq+0x0/0x420) from [<c0065d9c>] (handle_IRQ_event+0x2c/0x6c)
 ...

This bug has been around for a LONG time. The MM warning is from late 2005, but the driver merged a year later ... so I'm puzzled why nobody noticed this before now.

The fix involves noting that this buffer shouldn't be DMA-coherent; it's just used for normal DMA writes. So replace it with standard kmalloc() buffering and DMA mapping calls.

This is the quickie fix. A better one would not rely on allocating large bounce buffers. (Note that dma_alloc_coherent could have failed too, but that case was ignored... kmalloc is a bit more likely to fail though.)

---
 drivers/mmc/host/at91_mci.c |   20 ++++++++++++++++----
 1 file changed, 16 insertions(+), 4 deletions(-)
drivers/mmc/host/at91_mci.c +16 -4
@@ -621,12 +621,21 @@ static void at91_mci_send_command(struct
 				if (cpu_is_at91sam9260 () || cpu_is_at91sam9263())
 					if (host->total_length < 12)
 						host->total_length = 12;
-				host->buffer = dma_alloc_coherent(NULL,
-						host->total_length,
-						&host->physical_address, GFP_KERNEL);
+
+				host->buffer = kmalloc(host->total_length, GFP_KERNEL);
+				if (!host->buffer) {
+					pr_debug("Can't alloc tx buffer\n");
+					cmd->error = -ENOMEM;
+					mmc_request_done(host->mmc, host->request);
+					return;
+				}
 
 				at91_mci_sg_to_dma(host, data);
 
+				host->physical_address = dma_map_single(NULL,
+						host->buffer, host->total_length,
+						DMA_TO_DEVICE);
+
 				pr_debug("Transmitting %d bytes\n", host->total_length);
 
 				at91_mci_write(host, ATMEL_PDC_TPR, host->physical_address);
@@ -694,7 +703,10 @@ static void at91_mci_completed_command(s
 	cmd->resp[3] = at91_mci_read(host, AT91_MCI_RSPR(3));
 
 	if (host->buffer) {
-		dma_free_coherent(NULL, host->total_length, host->buffer, host->physical_address);
+		dma_unmap_single(NULL,
+				host->physical_address, host->total_length,
+				DMA_TO_DEVICE);
+		kfree(host->buffer);
 		host->buffer = NULL;
 	}
 
#1 lore raw

Ok with you nicolas?

Quoted 72 lines · On Mon, 18 Aug 2008 16:07:21 -0700 David Brownell <[email protected]> wrote:

From: David Brownell <[email protected]>

At91_mci is abusing dma_free_coherent(), which may not be called with IRQs disabled. I saw "mkfs.ext3" on an MMC card objecting voluminously as each write completed:

 WARNING: at arch/arm/mm/consistent.c:368 dma_free_coherent+0x2c/0x224()
 [<c002726c>] (dump_stack+0x0/0x14) from [<c00387d4>] (warn_on_slowpath+0x4c/0x68)
 [<c0038788>] (warn_on_slowpath+0x0/0x68) from [<c0028768>] (dma_free_coherent+0x2c/0x224)
  r6:00008008 r5:ffc06000 r4:00000000
 [<c002873c>] (dma_free_coherent+0x0/0x224) from [<c01918ac>] (at91_mci_irq+0x374/0x420)
 [<c0191538>] (at91_mci_irq+0x0/0x420) from [<c0065d9c>] (handle_IRQ_event+0x2c/0x6c)
 ...

This bug has been around for a LONG time. The MM warning is from late 2005, but the driver merged a year later ... so I'm puzzled why nobody noticed this before now.

The fix involves noting that this buffer shouldn't be DMA-coherent; it's just used for normal DMA writes. So replace it with standard kmalloc() buffering and DMA mapping calls.

This is the quickie fix. A better one would not rely on allocating large bounce buffers. (Note that dma_alloc_coherent could have failed too, but that case was ignored... kmalloc is a bit more likely to fail though.)

---
 drivers/mmc/host/at91_mci.c |   20 ++++++++++++++++----
 1 file changed, 16 insertions(+), 4 deletions(-)
drivers/mmc/host/at91_mci.c +16 -4
@@ -621,12 +621,21 @@ static void at91_mci_send_command(struct
 				if (cpu_is_at91sam9260 () || cpu_is_at91sam9263())
 					if (host->total_length < 12)
 						host->total_length = 12;
-				host->buffer = dma_alloc_coherent(NULL,
-						host->total_length,
-						&host->physical_address, GFP_KERNEL);
+
+				host->buffer = kmalloc(host->total_length, GFP_KERNEL);
+				if (!host->buffer) {
+					pr_debug("Can't alloc tx buffer\n");
+					cmd->error = -ENOMEM;
+					mmc_request_done(host->mmc, host->request);
+					return;
+				}
 
 				at91_mci_sg_to_dma(host, data);
 
+				host->physical_address = dma_map_single(NULL,
+						host->buffer, host->total_length,
+						DMA_TO_DEVICE);
+
 				pr_debug("Transmitting %d bytes\n", host->total_length);
 
 				at91_mci_write(host, ATMEL_PDC_TPR, host->physical_address);
@@ -694,7 +703,10 @@ static void at91_mci_completed_command(s
 	cmd->resp[3] = at91_mci_read(host, AT91_MCI_RSPR(3));
 
 	if (host->buffer) {
-		dma_free_coherent(NULL, host->total_length, host->buffer, host->physical_address);
+		dma_unmap_single(NULL,
+				host->physical_address, host->total_length,
+				DMA_TO_DEVICE);
+		kfree(host->buffer);
 		host->buffer = NULL;
 	}
 
Signature · 9 lines
     -- Pierre Ossman

  Linux kernel, MMC maintainer        http://www.kernel.org
  rdesktop, core developer          http://www.rdesktop.org

  WARNING: This correspondence is being monitored by the
  Swedish government. Make sure your server uses encryption
  for SMTP traffic and consider using PGP for end-to-end
  encryption.
#2 lore raw
In reply to David Brownell

Pierre, Andrew,

You can add my...

David Brownell :

Quoted 28 lines

From: David Brownell <[email protected]>

At91_mci is abusing dma_free_coherent(), which may not be called with IRQs disabled. I saw "mkfs.ext3" on an MMC card objecting voluminously as each write completed:

 WARNING: at arch/arm/mm/consistent.c:368 dma_free_coherent+0x2c/0x224()
 [<c002726c>] (dump_stack+0x0/0x14) from [<c00387d4>] (warn_on_slowpath+0x4c/0x68)
 [<c0038788>] (warn_on_slowpath+0x0/0x68) from [<c0028768>] (dma_free_coherent+0x2c/0x224)
  r6:00008008 r5:ffc06000 r4:00000000
 [<c002873c>] (dma_free_coherent+0x0/0x224) from [<c01918ac>] (at91_mci_irq+0x374/0x420)
 [<c0191538>] (at91_mci_irq+0x0/0x420) from [<c0065d9c>] (handle_IRQ_event+0x2c/0x6c)
 ...

This bug has been around for a LONG time. The MM warning is from late 2005, but the driver merged a year later ... so I'm puzzled why nobody noticed this before now.

The fix involves noting that this buffer shouldn't be DMA-coherent; it's just used for normal DMA writes. So replace it with standard kmalloc() buffering and DMA mapping calls.

This is the quickie fix. A better one would not rely on allocating large bounce buffers. (Note that dma_alloc_coherent could have failed too, but that case was ignored... kmalloc is a bit more likely to fail though.)

...

Signature · 1 line
Nicolas Ferre
Acked-by
#3 lore raw

Pierre, Andrew,

Based on Andrew comments, I rewrite this email..

You can add my...

Quoted 27 lines

David Brownell :

Quoted 26 lines

From: David Brownell <[email protected]>

At91_mci is abusing dma_free_coherent(), which may not be called with IRQs disabled. I saw "mkfs.ext3" on an MMC card objecting voluminously as each write completed:

 WARNING: at arch/arm/mm/consistent.c:368 dma_free_coherent+0x2c/0x224()
 [<c002726c>] (dump_stack+0x0/0x14) from [<c00387d4>] 
(warn_on_slowpath+0x4c/0x68)
 [<c0038788>] (warn_on_slowpath+0x0/0x68) from [<c0028768>] 
(dma_free_coherent+0x2c/0x224)
  r6:00008008 r5:ffc06000 r4:00000000
 [<c002873c>] (dma_free_coherent+0x0/0x224) from [<c01918ac>] 
(at91_mci_irq+0x374/0x420)
 [<c0191538>] (at91_mci_irq+0x0/0x420) from [<c0065d9c>] 
(handle_IRQ_event+0x2c/0x6c)
 ...

This bug has been around for a LONG time. The MM warning is from late 2005, but the driver merged a year later ... so I'm puzzled why nobody noticed this before now.

The fix involves noting that this buffer shouldn't be DMA-coherent; it's just used for normal DMA writes. So replace it with standard kmalloc() buffering and DMA mapping calls.

Quoted 8 lines
Quoted 6 lines

This is the quickie fix. A better one would not rely on allocating large bounce buffers. (Note that dma_alloc_coherent could have failed too, but that case was ignored... kmalloc is a bit more likely to fail though.)

...

Kind regards,

Signature · 1 line
Nicolas Ferre
Acked-by
#4 lore raw
Quoted 7 lines · On Fri, 22 Aug 2008 10:40:30 +0200 Nicolas Ferre <[email protected]>…
Quoted 4 lines
Quoted 2 lines

...

I have nothing else in my queue, so Andrew, as I suspect you have other things going to Linus, if you could please include this one as well. :)

Rgds

Signature · 9 lines
     -- Pierre Ossman

  Linux kernel, MMC maintainer        http://www.kernel.org
  rdesktop, core developer          http://www.rdesktop.org

  WARNING: This correspondence is being monitored by the
  Swedish government. Make sure your server uses encryption
  for SMTP traffic and consider using PGP for end-to-end
  encryption.
#5 lore raw
Quoted 6 lines · On Wed, 27 Aug 2008 21:01:05 +0200 Pierre Ossman <[email protected]> wrote:
Quoted 3 lines · On Fri, 22 Aug 2008 10:40:30 +0200 Nicolas Ferre <[email protected]>…
Quoted 3 lines
Quoted 2 lines
Quoted 7 lines
Quoted 4 lines
Quoted 1 line

...

I have nothing else in my queue, so Andrew, as I suspect you have other things going to Linus, if you could please include this one as well. :)

No probs, thanks.