On Thu, Jun 6, 2013 at 5:21 PM, Xenia Ragiadakou <burzalodowa@xxxxxxxxx> wrote: > This patch adds a check on whether the host machine > supports the xHC DMA address mask and sets the DMA > mask for coherent DMA address allocation via an > explicit call to dma_set_coherent_mask(). > > According to DMA-API-HOWTO, if coherent DMA address > mask has not been set explicitly via dma_set_coherent_mask(), > and the driver calls dma_alloc_coherent() or > dma_pool_create() to allocate consistent DMA memory > blocks, the consistent DMA mapping interface will > return by default DMA addresses which are 32-bit > addressable. > > Hence, if 64-bit DMA mapping is supported, it > is appropriate to call dma_set_coherent_mask() > with DMA_BIT_MASK(64) to take advantage of it. > > Also, according to DMA-API-HOWTO, dma_set_coherent_mask() > is guaranteed to set successfully the same or a smaller > mask as dma_set_mask(). It looks for me overcomplicated. We have *dma_mask and dma_coherent mask in the struct device. First question, who is allocating memory for dma_mask? Second, in case of dma_mask == NULL, dma_set_mask fails with -EIO. It doesn't mean we have no support of this one. How do you handle this case? I think it's pretty simple to set dma_coherent_mask and then apply its address to the dma_mask, because you set the same values anyway. -- With Best Regards, Andy Shevchenko -- To unsubscribe from this list: send the line "unsubscribe linux-usb" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
