On Fri, 28 Nov 2008 08:52:58 +0100, Jan Nikitenko <jan.nikitenko@xxxxxxxxx> wrote:
> dma_cache_wback_inv() expects virtual address, but physical was provided
> due to translation via plat_dma_addr_to_phys().
> If replaced with dma_addr_to_virt(), page fault oops from dma_unmap_page()
> is gone on au1550 platform.
>
> Signed-off-by: Jan Nikitenko <jan.nikitenko@xxxxxxxxx>
> ---
> arch/mips/mm/dma-default.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/arch/mips/mm/dma-default.c b/arch/mips/mm/dma-default.c
> index 5b98d0e..5f336c1 100644
> --- a/arch/mips/mm/dma-default.c
> +++ b/arch/mips/mm/dma-default.c
> @@ -222,7 +222,7 @@ void dma_unmap_page(struct device *dev, dma_addr_t dma_address, size_t size,
> if (!plat_device_is_coherent(dev) && direction != DMA_TO_DEVICE) {
> unsigned long addr;
>
> - addr = plat_dma_addr_to_phys(dma_address);
> + addr = dma_addr_to_virt(dma_address);
> dma_cache_wback_inv(addr, size);
> }
Acked-by: Atsushi Nemoto <anemo@xxxxxxxxxxxxx>
I'm also wondering why dma_map_page and dma_unmap_page are using
dma_cache_wback_inv instead of __dma_sync. They also lack special
r10000 handling. Hmm...
---
Atsushi Nemoto
