Hi in arch/sparc/kernel/ioport.c pci_map_sg() iterates over the scatterlist and assigns: sg->dvma_address = virt_to_phys(page_address(sg->page)). If CONFIG_PCI is set dma_map_sg is just a wrapper around pci_map_sg() as defined in include/asm-generic/dma-mapping.h. But I believe that this makes the behaviour of dma_map_sg() different on sparc compared to i386; In dma_map_sg() on i386 sg->offset is also added to dvma_address. I ran in to this while I was making the DMA API available to non-PCI systems on SPARC. I created the dma_* functions by just copying the pci_-equivalent and changing the BUG_ON(direction ==..) check. But since pci_map_sg() does not include the addition of sg->offset, transfers to buffers which do not begin on a page boundary will not work as expected. In my case this broke the scatterlist mappings in usbcore. Now I am curious as to why sg->offset isn't added to sg->dvma_address in pci_map_sg(). Have I overlooked something or should the offset be added? I am not very familiar with the DMA mapping system, and quite new to kernel programming - so please be gentle :-) Best regards, Jan - To unsubscribe from this list: send the line "unsubscribe sparclinux" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
