Re: [iPAQ] supported hardware?

never mind ... I found it. :
http://cvs.handhelds.org/cgi-bin/viewcvs.cgi/linux/kernel/Documentation/

Nicolas Pitre wrote:

> On Fri, 11 Aug 2000, Charlie Flynn wrote:
>
> > Thanks Nicholas,
> >
> > I saw your code and I also studied the code in the audio driver you mentioned.
> > I have been able to get DMA working using a combination of the existing code and
> > your code ( specifically the function "sa1100_start_dma" ). My code isn't
> > generic stuff but at least it's working and I can tidy it up later. I was also
> > in a steep DMA learning curve.
> >
> > My understanding is that the "generic" API as presented in
> > /arch/arm/kernel/dma.c (kernel 2.4.0-test4) needs the architecture specific
> > functions (arch_xxx() ). These functions are still under development (and are
> > presently commented out in 2.4.0-test4). If this is true can you give some
> > indication as to when they will be finished.?
>
> Actually I don't know since I'm not settled on the best API to use. The
> generic one isn't appropriate for SA1100 DMA and probably a separate one
> might be best.
>
> My idea is to use a function to commit some buffers in DMA queues and have
> a callback function be called to notify the interested code when the
> buffer has been processed. Also buffers size shouldn't be limited by the
> physical DMA buffer size, etc.
>
> As soon as I can resurface from everything that stacked while I was on
> vacation I'll try to complete the code.
>
> > Your help now and in the future would be much appreciated. One question comes to
> > mind since you offer ;-)
> > The physical and virtual addresses returned from the call to
> > "consistent_alloc()" were different (as expected) but when I tried to convert
> > physical to virtual using the macros "virt_to_phys" and "phys_to_virt".the
> > addresses were the same. I only got DMA working when I ditched these macros and
> > stayed with the physical address returned from "consistent_alloc()". Any
> > comments?
>
> The consist_alloc() creates a separate virtual mapping for the allocated
> memory which is marked as non cacheable. Since the returned virtual
> address is the new mapping and not the actually allocated memory, you
> can't use virt_to_phys() on it.
>
> If you wanted to use the physical address of a buffer like a skbuf for
> example, you should use pci_map_single() as described in
> linux/Documentation/DMA-mapping.txt (here you should pass NULL as the PCI
> dev argument). This is especially suited for streaming DMA like
> networking data where you can avoid the extra memcpy to/from the static
> coherent buffer.
>
> Nicolas
>
> _______________________________________________
> iPAQ mailing list
> iPAQ@handhelds.org
> http://handhelds.org/mailman/listinfo/ipaq
Received on Fri Aug 11 11:11:20 2000