On Tue, 2019-01-15 at 16:20 +0100, hch@xxxxxx wrote: > On Tue, Jan 15, 2019 at 03:24:55PM +0100, Christian König wrote: > > Yeah, indeed. Bounce buffers are an absolute no-go for GPUs. > > > > If the DMA API finds that a piece of memory is not directly > > accessible by > > the GPU we need to return an error and not try to use bounce > > buffers behind > > the surface. > > > > That is something which always annoyed me with the DMA API, which > > is > > otherwise rather cleanly defined. > > That is exactly what I want to fix with my series to make > DMA_ATTR_NON_CONSISTENT more useful and always available: > > https://na01.safelinks.protection.outlook.com/?url=https%3A%2F%2Flists.linuxfoundation.org%2Fpipermail%2Fiommu%2F2018-December%2F031985.html&data=02%7C01%7Cthellstrom%40vmware.com%7Cb1799c4073024a824f9408d67afcfcea%7Cb39138ca3cee4b4aa4d6cd83d9dd62f0%7C0%7C0%7C636831624340834140&sdata=JiRBfzZMvN3joJ4vKiErbzXAHaNzuBcLapRJDL%2Bt6Hc%3D&reserved=0 > > With that you allocate the memory using dma_alloc_attrs with > DMA_ATTR_NON_CONSISTENT, and use dma_sync_single_* to transfer > ownership to the cpu and back to the device, with a gurantee that > there won't be any bouncing. So far the interest by the parties that > requested the feature has been rather lacklustre, though. In the graphics case, it's probably because it doesn't fit the graphics use-cases: 1) Memory typically needs to be mappable by another device. (the "dma- buf" interface) 2) DMA buffers are exported to user-space and is sub-allocated by it. Mostly there are no GPU user-space kernel interfaces to sync / flush subregions and these syncs may happen on a smaller-than-cache-line granularity. So to help the graphics driver, that coherent flag would be much more useful. /Thomas _______________________________________________ dri-devel mailing list dri-devel@xxxxxxxxxxxxxxxxxxxxx https://lists.freedesktop.org/mailman/listinfo/dri-devel
