Re: [PATCH RFT RESEND linux-next] c6x: dma-mapping: support debug_dma_mapping_error

[Shuah Khan <shuah.khan@hp.com>]

On Fri, 2012-11-02 at 16:59 -0400, Mark Salter wrote:
> On Fri, 2012-11-02 at 14:26 -0600, Shuah Khan wrote:
> > On Fri, 2012-11-02 at 16:15 -0400, Mark Salter wrote:
> > > On Fri, 2012-11-02 at 13:53 -0600, Shuah Khan wrote:
> > > > On Fri, 2012-11-02 at 15:10 -0400, Mark Salter wrote:
> > > > > On Fri, 2012-11-02 at 10:44 -0600, Shuah Khan wrote:
> > > > > > On Fri, 2012-10-26 at 09:40 -0600, Shuah Khan wrote:
> > > > > > > Add support for debug_dma_mapping_error() call to avoid warning from
> > > > > > > debug_dma_unmap() interface when it checks for mapping error checked
> > > > > > > status. Without this patch, device driver failed to check map error
> > > > > > > warning is generated.
> > > > > > > 
> > > > > > > Signed-off-by: Shuah Khan <shuah.khan@hp.com>
> > > > > > > ---
> > > > > > >  arch/c6x/include/asm/dma-mapping.h |    1 +
> > > > > > >  1 file changed, 1 insertion(+)
> > > > > 
> > > > > > Would you like to this patch go through c6x arch tree or linux-next?
> > > > > > Please let me know your preference.
> > > > > 
> > > > > I tried to test this but I get a build error with CONFIG_DMA_API_DEBUG:
> > > > > 
> > > > > /linux-next/lib/dma-debug.c: In function 'has_mapping_error':
> > > > > /linux-next/lib/dma-debug.c:863:15: error: implicit declaration of function 'get_dma_ops' [-Werror=implicit-function-declaration]
> > > > > /linux-next/lib/dma-debug.c:863:34: warning: initialization makes pointer from integer without a cast [enabled by default]
> > > > > 
> > > > > C6X (along with some other architectures) doesn't have a get_dma_ops()
> > > > > function defined.
> > > > 
> > > > That is a problem I didn't think about. I did a check and looks like c6x
> > > > and frv are the only ones that don't have get_dma_ops() defined. frv is
> > > 
> > > By my count, there are 14 architectures with get_dma_ops() and 14
> > > without.
> > Right. I should have explained more. The following archs
> > 
> > arch/avr32/include/asm/dma-mapping.h
> > arch/blackfin/include/asm/dma-mapping.h
> > arch/cris/include/asm/dma-mapping.h
> > arch/mn10300/include/asm/dma-mapping.h
> > arch/parisc/include/asm/dma-mapping.h
> > arch/xtensa/include/asm/dma-mapping.h
> > 
> > define dma_map_page() and dma_map_single() and not call
> > debug_dma_map_page() interface. There is no risk of mis-matched debug
> > and non-debug mapping and mapping error checks like in the case of other
> > archs and c6x.
> 
> Ah, okay. Not all architectures support HAVE_DMA_API_DEBUG. So, of those
> that do, c6x seems to be the only one with dma_ops.
> 
> > 
> > > > in a different category as it doesn't use dma_debug interfaces. IN the
> > > > case c6x, now with my change to add debug_dma_mapping_error(), we will
> > > > start seeing warnings since dma_map_page() and dma_map_single() are
> > > > debugged with a call to debug_dma_map_page() and the corresponding
> > > > dma_mapping_error() interface doesn't call debug_dma_mapping_error()
> > > > interface
> > > > 
> > > > - Does adding get_dma_ops() make sense? Doesn't look like c6x exports
> > > > dma_ops?
> > > > 
> > > > Any other ideas?
> > > 
> > > I'm not sure. I don't know what get_dma_ops() does and it doesn't seem
> > > to be documented anywhere.
> > 
> > It returns pointer to dma_ops like the one on alpha:
> > 
> > static inline struct dma_map_ops *get_dma_ops(struct device *dev)
> > {
> >         return dma_ops;
> > }
> 
> Okay, so what is dma_ops used for? Looks like maybe supporting different
> dma features/functionality on different busses/devices.
> 
> > 
> > c6x doesn't define dma_ops looks like. Is that correct? Returning null
> > from get_dma_ops() is not an option as get_dma_ops() return is assumed
> > to be not null.
> 
> As things stand, c6x DMA hw doesn't really need dma_ops and until this
> patch, I could build in DMA debug support without dma_ops. So do we
> really want to require dma_ops for dma debug support even for those
> architectures which don't otherwise need it? I could add dma_ops, but
> it seems silly to do so only for dma debug.

I agree it doesn't make sense to add dma_ops when there is no need to.
Until now dma-debug didn't depend on dma_ops. This dependency has been
introduced with the debug_dma_mapping_error() interface I added :(

Let me go back and see if I can come up with a way to not require
dma_ops for debug-dma.

Thanks,
-- Shuah