Patchwork powerpc: Dont require a dma_ops struct to set dma mask

login
register
mail settings
Submitter Kumar Gala
Date Aug. 2, 2010, 5:21 p.m.
Message ID <1280769682-2839-1-git-send-email-galak@kernel.crashing.org>
Download mbox | patch
Permalink /patch/60558/
State Superseded
Headers show

Comments

Kumar Gala - Aug. 2, 2010, 5:21 p.m.
The only reason to require a dma_ops struct is to see if it has
implemented set_dma_mask.  If not we can fall back to setting the mask
directly.

This resolves an issue with how to sequence the setting of a DMA mask
for platform devices.  Before we had an issue in that we have no way of
setting the DMA mask before the various low level bus notifiers get
called that might check it (swiotlb).

So now we can do:

	pdev = platform_device_alloc("foobar", 0);
	dma_set_mask(&pdev->dev, DMA_BIT_MASK(37));
	platform_device_register(pdev);

And expect the right thing to happen with the bus notifiers get called
via platform_device_register.

Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
---
 arch/powerpc/include/asm/dma-mapping.h |    4 +---
 1 files changed, 1 insertions(+), 3 deletions(-)
Stephen Rothwell - Aug. 3, 2010, 6:17 a.m.
Hi Kumar,

On Mon,  2 Aug 2010 12:21:22 -0500 Kumar Gala <galak@kernel.crashing.org> wrote:
>
> --- a/arch/powerpc/include/asm/dma-mapping.h
> +++ b/arch/powerpc/include/asm/dma-mapping.h
> @@ -131,9 +131,7 @@ static inline int dma_set_mask(struct device *dev, u64 dma_mask)
>  {
>  	struct dma_map_ops *dma_ops = get_dma_ops(dev);
>  
> -	if (unlikely(dma_ops == NULL))
> -		return -EIO;
> -	if (dma_ops->set_dma_mask != NULL)
> +	if (unlikely(dma_ops == NULL) && (dma_ops->set_dma_mask != NULL))

The first part of this condition is backward (should be != (or just
"dma_ops") (and "likely"?)).
roger blofeld - Aug. 3, 2010, 4:28 p.m.
----- Original Message ----

> From: Kumar Gala <galak@kernel.crashing.org>
> To: linuxppc-dev@ozlabs.org
> Sent: Mon, August 2, 2010 12:21:22 PM
> Subject: [PATCH] powerpc: Dont require a dma_ops struct to set dma mask
> 
> The only reason to require a dma_ops struct is to see if it has
> implemented  set_dma_mask.  If not we can fall back to setting the  mask
> directly.
> 
> This resolves an issue with how to sequence the setting  of a DMA mask
> for platform devices.  Before we had an issue in that we  have no way of
> setting the DMA mask before the various low level bus  notifiers get
> called that might check it (swiotlb).
> 
> So now we can  do:
> 
>     pdev = platform_device_alloc("foobar",  0);
>     dma_set_mask(&pdev->dev,  DMA_BIT_MASK(37));
>      platform_device_register(pdev);
> 
> And expect the right thing to happen with  the bus notifiers get called
> via  platform_device_register.
> 
> Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
> ---
>   arch/powerpc/include/asm/dma-mapping.h |    4 +---
>  1 files  changed, 1 insertions(+), 3 deletions(-)
> 
> diff --git  a/arch/powerpc/include/asm/dma-mapping.h  
>b/arch/powerpc/include/asm/dma-mapping.h
> index c85ef23..17d5c17 100644
> ---  a/arch/powerpc/include/asm/dma-mapping.h
> +++  b/arch/powerpc/include/asm/dma-mapping.h
> @@ -131,9 +131,7 @@ static inline  int dma_set_mask(struct device *dev, u64 
>dma_mask)
>  {
>       struct dma_map_ops *dma_ops = get_dma_ops(dev);
> 
> -    if  (unlikely(dma_ops == NULL))
> -        return  -EIO;
> -    if (dma_ops->set_dma_mask !=  NULL)
> +    if (unlikely(dma_ops == NULL) &&  (dma_ops->set_dma_mask != NULL))
>           return dma_ops->set_dma_mask(dev, dma_mask);
>      if  (!dev->dma_mask || !dma_supported(dev, dma_mask))
>           return -EIO;
> -- 
> 1.6.0.6
> 
> _______________________________________________
> Linuxppc-dev  mailing list
> Linuxppc-dev@lists.ozlabs.org
> https://lists.ozlabs.org/listinfo/linuxppc-dev
>

Isn't that test wrong? Perhaps you meant to test for dma_ops non-null before 
dereferencing it?
-roger

Patch

diff --git a/arch/powerpc/include/asm/dma-mapping.h b/arch/powerpc/include/asm/dma-mapping.h
index c85ef23..17d5c17 100644
--- a/arch/powerpc/include/asm/dma-mapping.h
+++ b/arch/powerpc/include/asm/dma-mapping.h
@@ -131,9 +131,7 @@  static inline int dma_set_mask(struct device *dev, u64 dma_mask)
 {
 	struct dma_map_ops *dma_ops = get_dma_ops(dev);
 
-	if (unlikely(dma_ops == NULL))
-		return -EIO;
-	if (dma_ops->set_dma_mask != NULL)
+	if (unlikely(dma_ops == NULL) && (dma_ops->set_dma_mask != NULL))
 		return dma_ops->set_dma_mask(dev, dma_mask);
 	if (!dev->dma_mask || !dma_supported(dev, dma_mask))
 		return -EIO;