When __pa is applied to virtual address in uncached KSEG region the result is incorrect. Fix it by checking if the original address is in the uncached KSEG and adjusting the result. It looks better than masking off bits because pfn_valid would correctly work with new __pa results and it may be made working in noMMU case, once we get definition for uncached memory view. This is required for the dma_common_mmap and other functions calling __pa on coherent DMA memory addresses. Cc: stable@xxxxxxxxxxxxxxx Tested-by: Boris Brezillon <boris.brezillon@xxxxxxxxxxxxxxxxxx> Reviewed-by: Boris Brezillon <boris.brezillon@xxxxxxxxxxxxxxxxxx> Signed-off-by: Max Filippov <jcmvbkbc@xxxxxxxxx> --- arch/xtensa/include/asm/page.h | 13 +++++++++++++ 1 file changed, 13 insertions(+) diff --git a/arch/xtensa/include/asm/page.h b/arch/xtensa/include/asm/page.h index 976b1d7..4ddbfd5 100644 --- a/arch/xtensa/include/asm/page.h +++ b/arch/xtensa/include/asm/page.h @@ -164,8 +164,21 @@ void copy_user_highpage(struct page *to, struct page *from, #define ARCH_PFN_OFFSET (PHYS_OFFSET >> PAGE_SHIFT) +#ifdef CONFIG_MMU +static inline unsigned long ___pa(unsigned long va) +{ + unsigned long off = va - PAGE_OFFSET; + + if (off >= XCHAL_KSEG_SIZE) + off -= XCHAL_KSEG_SIZE; + + return off + PHYS_OFFSET; +} +#define __pa(x) ___pa((unsigned long)(x)) +#else #define __pa(x) \ ((unsigned long) (x) - PAGE_OFFSET + PHYS_OFFSET) +#endif #define __va(x) \ ((void *)((unsigned long) (x) - PHYS_OFFSET + PAGE_OFFSET)) #define pfn_valid(pfn) \ -- 2.1.4