On Sat, 8 Dec 2018 at 17:17, Christoph Hellwig <hch@xxxxxx> wrote: > > On Sun, Dec 02, 2018 at 05:11:02PM +1100, Benjamin Herrenschmidt wrote: > > Talking of which ... Christoph, not sure if we can do something about > > this at the DMA API level or keep hacks but some adapters such as the > > nVidia GPUs have a HW hack we can use to work around their limitations > > in that case. > > > > They have a register that can program a fixed value for the top bits > > that they don't support. > > > > This works fine for any linear mapping with an offset, provided they > > can program the offset in that register and they have enough DMA range > > to cover all memory from that offset. > > > > I can probably get the info about this from them so we can exploit it > > in nouveau. > > I think we can expose the direct mapping offset if people care enough, > we just have to be very careful designing the API. I'll happily leave > that to those that actually want to use it, but I'll gladly help > reviewing it. Hi, Christoph and Ben, It just came to my mind (and this is most likely a stupid question, but still)… Is there any possibility of these changes having an (positive) effect on the long-standing problem of Power Mac machines with AGP graphics cards (which have to be limited to PCI transfers, otherwise they'll hang, due to coherence issues)? If so, I have a G4 machine where I'd gladly test them. Thanks, Rui
