On Sun, Apr 15, 2007 at 11:26:05PM +0800, tiansm@xxxxxxxxxx wrote:
(Adding a few more people to the cc'list)
> sound/core/pcm_native.c | 10 ++++++++++
> sound/core/sgbuf.c | 9 +++++++++
> 2 files changed, 19 insertions(+), 0 deletions(-)
>
> diff --git a/sound/core/pcm_native.c b/sound/core/pcm_native.c
> index 3e276fc..9005bac 100644
> --- a/sound/core/pcm_native.c
> +++ b/sound/core/pcm_native.c
> @@ -3145,7 +3145,11 @@ static struct page *snd_pcm_mmap_data_nopage(struct vm_area_struct *area,
> return NOPAGE_OOM; /* XXX: is this really due to OOM? */
> } else {
> vaddr = runtime->dma_area + offset;
> +#if defined(__mips__) && defined(CONFIG_DMA_NONCOHERENT)
Please use CONFIG_MIPS instead of __mips__ in #if / #ifdefs.
The question if #ifdefing is the right approach to solve this problem is
something else but I think no, ....
> + page = virt_to_page(CAC_ADDR(vaddr));
> +#else
> page = virt_to_page(vaddr);
> +#endif
So this is needed because the MIPS virt_to_page is returning a unsuitable
value if vaddress is not a KSEG0 (64-bit: cached XKPHYS) address which is
what GFP allocations and the slab will return. So now we have to deciede
if
a) the MIPS __pa() should be changed to handle uncached addresses.
b) the sound code here is simply broken.
Some drivers seem to allocate runtime->dma_area from vmalloc, so this
whole area in the sound code is looking like built on quicksand ...
> }
> get_page(page);
> if (type)
> @@ -3261,6 +3265,12 @@ static int snd_pcm_mmap(struct file *file, struct vm_area_struct *area)
> substream = pcm_file->substream;
> snd_assert(substream != NULL, return -ENXIO);
>
> +#if defined(__mips__) && defined(CONFIG_DMA_NONCOHERENT)
> + /* all mmap using uncached mode */
> + area->vm_page_prot = pgprot_noncached(area->vm_page_prot);
> + area->vm_flags |= ( VM_RESERVED | VM_IO);
VM_RESERVED will prevent the buffer from being freed. I assume that is
another workaround for some kernel subsystem blowing up when being fed a
pointer to an uncached RAM address? This smells like a memory leak.
> +#endif
> +
> offset = area->vm_pgoff << PAGE_SHIFT;
> switch (offset) {
> case SNDRV_PCM_MMAP_OFFSET_STATUS:
> diff --git a/sound/core/sgbuf.c b/sound/core/sgbuf.c
> index cefd228..535f0bc 100644
> --- a/sound/core/sgbuf.c
> +++ b/sound/core/sgbuf.c
> @@ -91,12 +91,21 @@ void *snd_malloc_sgbuf_pages(struct device *device,
> }
> sgbuf->table[i].buf = tmpb.area;
> sgbuf->table[i].addr = tmpb.addr;
> +#if defined(__mips__) && defined(CONFIG_DMA_NONCOHERENT)
> + sgbuf->page_table[i] = virt_to_page(CAC_ADDR(tmpb.area));
VM_RESERVED will prevent the buffer from being freed. I assume that is
another workaround for some kernel subsystem blowing up when being fed a
pointer to an uncached RAM address? This smells like a memory leak.
> +#else
> sgbuf->page_table[i] = virt_to_page(tmpb.area);
> +#endif
> sgbuf->pages++;
> }
>
> sgbuf->size = size;
> +#if defined(__mips__) && defined(CONFIG_DMA_NONCOHERENT)
> + /* maybe we should use uncached accelerated mode */
> + dmab->area = vmap(sgbuf->page_table, sgbuf->pages, VM_MAP | VM_IO, pgprot_noncached(PAGE_KERNEL));
> +#else
> dmab->area = vmap(sgbuf->page_table, sgbuf->pages, VM_MAP, PAGE_KERNEL);
> +#endif
I would suggest to get rid of this ifdef with a new arch-specific function
like vmap_io_buffer which will do whatever a platform seems fit for this
case?
> if (! dmab->area)
> goto _failed;
> return dmab->area;
Ralf
