[PATCH 1/7] DMAENGINE: add generic PrimeCell interface to COH 901 318 v6

[Dan Williams]

On Wed, May 12, 2010 at 12:39 AM, Linus WALLEIJ
<linus.walleij at stericsson.com> wrote:
> [Dan]
>
>> >> +void dma_set_ambaconfig(struct dma_chan *chan,
>> >> +                       struct amba_dma_channel_config *config)
>> >
>> > What happens when two primecell dma devices are present on one soc?
>> > Can that happen?  In that case we'll have conflicting providers of
>> > dma_set_ambaconfig().
>> >
>>
>> ...the ->device_control() method is already approaching an ioctl()
>> let's just add the argument parameter to complete the capability,
>> i.e.:
>>
>> int (*device_control)(struct dma_chan *chan, enum dma_ctrl_cmd cmd,
>> unsigned long arg);
>
> So the unsigned long arg is to match an ioctl? Okidoki...
>
> If it was only for in-kernel use a void* would be preferred
> I believe since assuming unsigned long to be == bus width
> is ... well, disturbing, since we'll be casting this into
> a struct pointer.

It's safe to assume that Linux does not run on any architecture where
sizeof(unsigned long) < sizeof(void *).

> So this should also be used for the AMBA configuration stuff
> then, so I have to make first a patch to convert all the
> device_control functions and then a patch on top of that to
> implement the AMBA call as a device_control() thingie?
>
> (Should work out...)
>
> Just a bit of work, so will take some more hours :-)

Apologies for not discovering this earlier in the review process, but
it gets rid of the last bit of 'specialness' of the Primecell api and
I can see it allowing other weird commands to be implemented
generically down the road.  Just let me know if you want to run any
compile tests for you.

> I will just have to make a proof-of-concept for the AMBA config
> interface for Russell first, then I'll be on the ioctl-like API.

Ok, I really appreciate your patience, I'll see what I can do to push
this post-rc1 if things turn out that way.

Regards,
Dan