Re: [PATCH 15/16] dma-debug: x86 architecture bindings

[Joerg Roedel <joro@8bytes.org>]

On Sun, Jan 11, 2009 at 03:25:47PM +0900, FUJITA Tomonori wrote:
> On Fri, 9 Jan 2009 17:19:29 +0100
> Joerg Roedel <joerg.roedel@amd.com> wrote:
> 
> > Impact: make use of DMA-API debugging code in x86
> > 
> > Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
> > ---
> >  arch/x86/Kconfig                   |    1 +
> >  arch/x86/include/asm/dma-mapping.h |   30 ++++++++++++++++++++++++++----
> >  arch/x86/kernel/pci-dma.c          |    5 +++++
> >  3 files changed, 32 insertions(+), 4 deletions(-)
> > 
> > diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
> > index 862adb9..68a806c 100644
> > --- a/arch/x86/Kconfig
> > +++ b/arch/x86/Kconfig
> > @@ -39,6 +39,7 @@ config X86
> >  	select HAVE_GENERIC_DMA_COHERENT if X86_32
> >  	select HAVE_EFFICIENT_UNALIGNED_ACCESS
> >  	select USER_STACKTRACE_SUPPORT
> > +	select HAVE_DMA_API_DEBUG
> >  
> >  config ARCH_DEFCONFIG
> >  	string
> > diff --git a/arch/x86/include/asm/dma-mapping.h b/arch/x86/include/asm/dma-mapping.h
> > index 4035357..939d5b3 100644
> > --- a/arch/x86/include/asm/dma-mapping.h
> > +++ b/arch/x86/include/asm/dma-mapping.h
> > @@ -7,6 +7,7 @@
> >   */
> >  
> >  #include <linux/scatterlist.h>
> > +#include <linux/dma-debug.h>
> >  #include <asm/io.h>
> >  #include <asm/swiotlb.h>
> >  #include <asm-generic/dma-coherent.h>
> > @@ -93,9 +94,12 @@ dma_map_single(struct device *hwdev, void *ptr, size_t size,
> >  	       int direction)
> >  {
> >  	struct dma_mapping_ops *ops = get_dma_ops(hwdev);
> > +	dma_addr_t addr;
> >  
> >  	BUG_ON(!valid_dma_direction(direction));
> > -	return ops->map_single(hwdev, virt_to_phys(ptr), size, direction);
> > +	addr = ops->map_single(hwdev, virt_to_phys(ptr), size, direction);
> > +	debug_map_single(hwdev, ptr, size, direction, addr);
> > +	return addr;
> 
> What happens if ops->map_single fails?
> 
> Seems that debug_map_single doesn't check the dma mapping fails. So it
> allocates a new entries and nobody frees it?

Ah true. The debug_map_single function has to check for
dma_mapping_error. Same is true for debug_map_sg.

> Another problem is that what happens if ops->map_single succeeds but
> debug_map_single fails? Seems that it gives a false warning.

No, this can not happen. If the code fails it disables itself so no more
warnings will be printed.

Joerg