From: Kristoffer Glembo <kristoffer@xxxxxxxxxxx> Date: Mon, 16 Nov 2009 10:45:40 +0100 > I finally had the time to look at this again and I'm not sure > how I would do this cleanly using only the SBUS stuff. They > use the IOMMU routines (mm/iommu.c or mm/io-unit.c) and since > we don't currently have an IOMMU I don't want to mess with that. > > On LEON system that are not snooping DMA accesses I want to flush > the dcache when unmapping or syncing the buffer. That's it. > > I have a suggestion for a patch. It uses sbus_alloc_coherent and > sbus_free_coherent with some added ifdefs and then the PCI > routines for the rest (I've added unmap_page and patched map_page). > At least the code duplication is quite minimal ... > > In the somewhat limited testing I have done it works fine. It looks good from a code sharing perspective, but the virt_to_phys() change has to be undone as per feedback I gave to the patch #2 posting. > If you explain what you think is very messy with the PCI routines > maybe I can clean them up further. On LEON we use both the generic > DMA-API and the PCI-API. Bad coding style, "#if 0" sections, it does things like: n = (n + PAGE_SIZE-1) & PAGE_MASK; when a simple PAGE_ALIGN() would do, etc. Also, tests like this: BUG_ON(page_address(sg_page(sg)) == NULL); Don't make any sense, because highmem pages are legal to DMA from/to, and such pages would return NULL here. -- 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
