Hi Arnd,





Thanks,
with regards,
Ram


On Tue, Jan 13, 2015 at 12:25 AM, Arnd Bergmann [off-list ref] wrote:

On Monday 12 January 2015 14:40:09 Rameshwar Sahu wrote:

quoted

quoted

+
+       dma_set_mask_and_coherent(&pdev->dev,
+                                 (sizeof(dma_addr_t) == sizeof(u64)) ?
+                                 DMA_BIT_MASK(64) : DMA_BIT_MASK(32));
+

I commented on this on v1 already: always set the device specific mask here,
it's independent of the CPU architecture, but do check the return value.

Got it, I will set first 64 bit dma mask here  due to device support
64 bit address dma operations, and fall down to 32 - bit in case
failure with 64 bit.

Also add the correct dma-ranges property in the parent, or else this will
fail in the future once we add proper checks to dma_set_mask.

I see arch/arm64 kernel doesn't use dma range value parsed from
dma-range property currently, I will check it more.
It doesn't use dma_pfn_offset value calculated by the dma-ranges in
dma mapping API.
dma-ranges value should test properly before adding.
Is there any use to add this dma-ranges properly if arch code doesn't support ??

        Arnd