diff mbox

MMC: OMAP MSDI: fix broken PIO mode

Message ID alpine.DEB.2.00.1208240555210.25436@utopia.booyaka.com (mailing list archive)
State New, archived
Headers show

Commit Message

Paul Walmsley Aug. 24, 2012, 6 a.m. UTC
After commit 26b88520b80695a6fa5fd95b5d97c03f4daf87e0 ("mmc:
omap_hsmmc: remove private DMA API implementation"), the Nokia N800
here stopped booting:

[    2.086181] Waiting for root device /dev/mmcblk0p1...
[    2.324066] Unhandled fault: imprecise external abort (0x406) at 0x00000000
[    2.331451] Internal error: : 406 [#1] ARM
[    2.335784] Modules linked in:
[    2.339050] CPU: 0    Not tainted  (3.6.0-rc3 #60)
[    2.344146] PC is at default_idle+0x28/0x30
[    2.348602] LR is at trace_hardirqs_on_caller+0x15c/0x1b0

...

This turned out to be due to memory corruption caused by long-broken
PIO code in drivers/mmc/host/omap.c.  (Previously, this driver had
been using DMA; but the above commit caused the MMC driver to fall
back to PIO mode with an unmodified Kconfig.)

The PIO code, added with the rest of the driver in commit
730c9b7e6630f786fcec026fb11d2e6f2c90fdcb ("[MMC] Add OMAP MMC host
driver"), confused bytes with 16-bit words.  This bug caused memory
located after the PIO transfer buffer to be corrupted with transfers
larger than 32 bytes.  The driver also did not increment the buffer
pointer after the transfer occurred.  This bug resulted in data
corruption during any transfer larger than 64 bytes.

Signed-off-by: Paul Walmsley <paul@pwsan.com>
Cc: Juha Yrjölä <juha.yrjola@nokia.com>
Cc: Tony Lindgren <tony@atomide.com>
Cc: Carlos Aguiar <carlos.aguiar@indt.org.br>
Cc: Russell King <rmk+kernel@arm.linux.org.uk>
---
 drivers/mmc/host/omap.c |   14 +++++++++++---
 1 file changed, 11 insertions(+), 3 deletions(-)

Comments

Felipe Balbi Aug. 24, 2012, 7:25 a.m. UTC | #1
Hi,

On Fri, Aug 24, 2012 at 06:00:18AM +0000, Paul Walmsley wrote:
> 
> After commit 26b88520b80695a6fa5fd95b5d97c03f4daf87e0 ("mmc:
> omap_hsmmc: remove private DMA API implementation"), the Nokia N800
> here stopped booting:
> 
> [    2.086181] Waiting for root device /dev/mmcblk0p1...
> [    2.324066] Unhandled fault: imprecise external abort (0x406) at 0x00000000
> [    2.331451] Internal error: : 406 [#1] ARM
> [    2.335784] Modules linked in:
> [    2.339050] CPU: 0    Not tainted  (3.6.0-rc3 #60)
> [    2.344146] PC is at default_idle+0x28/0x30
> [    2.348602] LR is at trace_hardirqs_on_caller+0x15c/0x1b0
> 
> ...
> 
> This turned out to be due to memory corruption caused by long-broken
> PIO code in drivers/mmc/host/omap.c.  (Previously, this driver had
> been using DMA; but the above commit caused the MMC driver to fall
> back to PIO mode with an unmodified Kconfig.)
> 
> The PIO code, added with the rest of the driver in commit
> 730c9b7e6630f786fcec026fb11d2e6f2c90fdcb ("[MMC] Add OMAP MMC host
> driver"), confused bytes with 16-bit words.  This bug caused memory
> located after the PIO transfer buffer to be corrupted with transfers
> larger than 32 bytes.  The driver also did not increment the buffer
> pointer after the transfer occurred.  This bug resulted in data
> corruption during any transfer larger than 64 bytes.
> 
> Signed-off-by: Paul Walmsley <paul@pwsan.com>
> Cc: Juha Yrjölä <juha.yrjola@nokia.com>
> Cc: Tony Lindgren <tony@atomide.com>
> Cc: Carlos Aguiar <carlos.aguiar@indt.org.br>
> Cc: Russell King <rmk+kernel@arm.linux.org.uk>

this looks good to me, though I don't have how to test. FWIW:

Reviewed-by: Felipe Balbi <balbi@ti.com>

> ---
>  drivers/mmc/host/omap.c |   14 +++++++++++---
>  1 file changed, 11 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/mmc/host/omap.c b/drivers/mmc/host/omap.c
> index 50e08f0..a5999a7 100644
> --- a/drivers/mmc/host/omap.c
> +++ b/drivers/mmc/host/omap.c
> @@ -668,7 +668,7 @@ mmc_omap_clk_timer(unsigned long data)
>  static void
>  mmc_omap_xfer_data(struct mmc_omap_host *host, int write)
>  {
> -	int n;
> +	int n, nwords;
>  
>  	if (host->buffer_bytes_left == 0) {
>  		host->sg_idx++;
> @@ -678,15 +678,23 @@ mmc_omap_xfer_data(struct mmc_omap_host *host, int write)
>  	n = 64;
>  	if (n > host->buffer_bytes_left)
>  		n = host->buffer_bytes_left;
> +
> +	nwords = n / 2;
> +	nwords += n & 1; /* handle odd number of bytes to transfer */
> +
>  	host->buffer_bytes_left -= n;
>  	host->total_bytes_left -= n;
>  	host->data->bytes_xfered += n;
>  
>  	if (write) {
> -		__raw_writesw(host->virt_base + OMAP_MMC_REG(host, DATA), host->buffer, n);
> +		__raw_writesw(host->virt_base + OMAP_MMC_REG(host, DATA),
> +			      host->buffer, nwords);
>  	} else {
> -		__raw_readsw(host->virt_base + OMAP_MMC_REG(host, DATA), host->buffer, n);
> +		__raw_readsw(host->virt_base + OMAP_MMC_REG(host, DATA),
> +			     host->buffer, nwords);
>  	}
> +
> +	host->buffer += nwords;
>  }
>  
>  static inline void mmc_omap_report_irq(u16 status)
> -- 
> 1.7.10.4
Tony Lindgren Aug. 24, 2012, 3:53 p.m. UTC | #2
* Felipe Balbi <balbi@ti.com> [120824 00:29]:
> 
> this looks good to me, though I don't have how to test. FWIW:
> 
> Reviewed-by: Felipe Balbi <balbi@ti.com>

Works for me:

Tested-by: Tony Lindgren <tony@atomide.com>
--
To unsubscribe from this list: send the line "unsubscribe linux-omap" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Chris Ball Aug. 27, 2012, 10:42 p.m. UTC | #3
Hi,

On Fri, Aug 24 2012, Paul Walmsley wrote:
> After commit 26b88520b80695a6fa5fd95b5d97c03f4daf87e0 ("mmc:
> omap_hsmmc: remove private DMA API implementation"), the Nokia N800
> here stopped booting:
>
> [    2.086181] Waiting for root device /dev/mmcblk0p1...
> [    2.324066] Unhandled fault: imprecise external abort (0x406) at 0x00000000
> [    2.331451] Internal error: : 406 [#1] ARM
> [    2.335784] Modules linked in:
> [    2.339050] CPU: 0    Not tainted  (3.6.0-rc3 #60)
> [    2.344146] PC is at default_idle+0x28/0x30
> [    2.348602] LR is at trace_hardirqs_on_caller+0x15c/0x1b0
>
> ...
>
> This turned out to be due to memory corruption caused by long-broken
> PIO code in drivers/mmc/host/omap.c.  (Previously, this driver had
> been using DMA; but the above commit caused the MMC driver to fall
> back to PIO mode with an unmodified Kconfig.)
>
> The PIO code, added with the rest of the driver in commit
> 730c9b7e6630f786fcec026fb11d2e6f2c90fdcb ("[MMC] Add OMAP MMC host
> driver"), confused bytes with 16-bit words.  This bug caused memory
> located after the PIO transfer buffer to be corrupted with transfers
> larger than 32 bytes.  The driver also did not increment the buffer
> pointer after the transfer occurred.  This bug resulted in data
> corruption during any transfer larger than 64 bytes.
>
> Signed-off-by: Paul Walmsley <paul@pwsan.com>

Yuck.  Thanks for tracking that down, Paul -- pushed to mmc-next for 3.6.

- Chris.
diff mbox

Patch

diff --git a/drivers/mmc/host/omap.c b/drivers/mmc/host/omap.c
index 50e08f0..a5999a7 100644
--- a/drivers/mmc/host/omap.c
+++ b/drivers/mmc/host/omap.c
@@ -668,7 +668,7 @@  mmc_omap_clk_timer(unsigned long data)
 static void
 mmc_omap_xfer_data(struct mmc_omap_host *host, int write)
 {
-	int n;
+	int n, nwords;
 
 	if (host->buffer_bytes_left == 0) {
 		host->sg_idx++;
@@ -678,15 +678,23 @@  mmc_omap_xfer_data(struct mmc_omap_host *host, int write)
 	n = 64;
 	if (n > host->buffer_bytes_left)
 		n = host->buffer_bytes_left;
+
+	nwords = n / 2;
+	nwords += n & 1; /* handle odd number of bytes to transfer */
+
 	host->buffer_bytes_left -= n;
 	host->total_bytes_left -= n;
 	host->data->bytes_xfered += n;
 
 	if (write) {
-		__raw_writesw(host->virt_base + OMAP_MMC_REG(host, DATA), host->buffer, n);
+		__raw_writesw(host->virt_base + OMAP_MMC_REG(host, DATA),
+			      host->buffer, nwords);
 	} else {
-		__raw_readsw(host->virt_base + OMAP_MMC_REG(host, DATA), host->buffer, n);
+		__raw_readsw(host->virt_base + OMAP_MMC_REG(host, DATA),
+			     host->buffer, nwords);
 	}
+
+	host->buffer += nwords;
 }
 
 static inline void mmc_omap_report_irq(u16 status)