At Fri, 22 Aug 2008 11:41:31 +0200,
Thomas Bogendoerfer wrote:
>
> On Fri, Aug 22, 2008 at 08:07:44AM +0200, Takashi Iwai wrote:
> > I don't think that this must work for *every* platform, too, and it's
> > not expected from the driver. The systems without uncached memory
> > access can simply return an error from dma_mmap_coherent() call, so
> > that the driver can disable the mmap. That'd be enough.
>
> true, I've used snd_pcm_indirect for HAL2 driver, which works even on
> SGI IP28 machines.
>
> > Now, how to handle these exceptions: a question comes into my mind
> > again -- how does the framebuffer handle these as well?
>
> most framebuffers have a dedicated set of video memory and this memory is
> just mmaped uncached either via TLB/MMU (MIPS) or rules inside
> the system (PARISC uses IO space memory, which is always uncached).
> The code which does this mmaping is in drivers/video/fbmem.c plus
> fb_pgprotect out of an include/asm header file.
Thanks. These are the files I already looked at, and pgprot fiddling
is already in my last patches (and apparently they not enough).
> For framebuffers without dedicated video memory the memory is mmaped
> write through or uncached. A driver, which uses this, is
> drivers/video/gbefb.c.
So, adding a code like below to dma_mmap_coherent() for MIPS?
static inline pgprot_t pgprot_mmap(pgprot_t _prot)
{
unsigned long prot = pgprot_val(_prot) & ~_CACHE_MASK;
#ifdef CONFIG_SGI_IP32
#ifdef CONFIG_CPU_R10000
prot = prot | _CACHE_UNCACHED_ACCELERATED;
#else
prot = prot | _CACHE_CACHABLE_NO_WA;
#endif
#else
prot = prot | _CACHE_UNCACHED;
#endif
return __pgprot(prot);
}
dma_mmap_coherent()
{
...
vma->vm_page_prot = pgprot_mmap(vma->vm_page_prot);
...
remap_pfn_range(...);
}
thanks,
Takashi
