Re: can device drivers return non-ram via vm_ops->nopage?

[Russell King <rmk+lkml@arm.linux.org.uk>]

On Sat, Mar 20, 2004 at 04:57:39PM +0100, Andrea Arcangeli wrote:
> > One of my current projects is fixing this crap in ALSA.
> 
> Do you agree it should be fixed by returning a PFN from ->nopage?

No.  How would you return the PFN from a remapped page?  It's far
easier to provide an interface which returns the struct page* for
the underlying pages, thusly:

static struct page *
dma_coherent_to_page(struct device *dev, void *cpu_addr,
		     dma_addr_t handle, unsigned int offset)

And this is precisely what I would be working on if I weren't writing
this mail. 8)

Take a moment to think about the problem.  We've allocated some memory
for coherent DMA via the dma_alloc_coherent() interface.  At some point,
we've had to get a struct page* in this allocator.  However, the
allocator has had to do some architecture defined operations to provide
coherent memory.

Only the architecture can translate the results from dma_alloc_coherent()
back to a struct page* - which it needs to be able to do if
dma_free_coherent() is going to work.

Therefore, what we need to do to solve the ALSA problem is require all
architectures to provide dma_coherent_to_page() and make ALSA use that.

(A related problem is that some architectures need pgprot_dmacoherent()
to modify the page protections so that the user space mapping is also
DMA-coherent.  However, that discussion should be the subject of a
new thread.)

-- 
Russell King
 Linux kernel    2.6 ARM Linux   - http://www.arm.linux.org.uk/
 maintainer of:  2.6 PCMCIA      - http://pcmcia.arm.linux.org.uk/
                 2.6 Serial core