Op 15-08-10 15:36, Jassi Brar schreef:

On Sun, Aug 15, 2010 at 10:28 PM, Jassi Brar [off-list ref] wrote:

quoted

On Mon, May 31, 2010 at 8:34 PM, Maurus Cuelenaere
[off-list ref] wrote:

quoted

When a DMA channel is freed, its pending requests should be flushed and the
channel should be halted. This patch ensures that happens.

Signed-off-by: Maurus Cuelenaere <redacted>
---
 arch/arm/mach-s3c64xx/dma.c |    3 ++-
 1 files changed, 2 insertions(+), 1 deletions(-)

diff --git a/arch/arm/mach-s3c64xx/dma.c b/arch/arm/mach-s3c64xx/dma.c
index 1fd9d0c..e1f22af 100644
--- a/arch/arm/mach-s3c64xx/dma.c
+++ b/arch/arm/mach-s3c64xx/dma.c

@@ -560,7 +560,8 @@ int s3c2410_dma_free(unsigned int channel, struct s3c2410_dma_client *client)
                      channel, chan->client, client);
       }

-       /* sort out stopping and freeing the channel */
+       s3c64xx_dma_flush(chan);
+       s3c64xx_dma_stop(chan);

I think, the order should be reverted.

You're probably right.

Though it's a matter of S3C DMA API spec.
Shouldn't the client be made to explicitly do S3C2410_DMAOP_FLUSH
a pending req before freeing the channel? If so, the patch may not be needed.

I can't find anything about the client needing to flush the request itself.

Documentation/arm/Samsung-S3C24XX/DMA.txt doesn't mention anything regarding it
and the comment above s3c2410_dma_free() says "release the given channel back to
the system, will stop and flush any outstanding transfers, and ensure the
channel is ready for the next claimant."


-- 
Maurus Cuelenaere