Michel Dänzer <[EMAIL PROTECTED]> writes: > > And one major feature on x86 machines are the iommu facilities of the > > AGP chipsets. Implementing a fake "AGP" driver for architectures which > > already have iommu facilities for their PCI subsystems seems to be > > fairly easy. The user-land drivers of the X-server can (perhaps?) > > remain unchanged. The drm kernel modules need to be adjusted, of course. > > Sounds plausible to me, but I can't tell for sure. > > Alternatively, maybe you could leverage the pcigart code of the ATI > drivers.
Mmmh. Actually, it doesn't work yet. I think there is still something wrong with the DMA stuff, though I don't know what (at the moment). I'm using vmalloc to allocate a large chunk of memory, and then walk page tables to find the underlying physical nodes; then I program the iommu to map the pages to a virtually contiguous region on the pci bus. After three or four frames the PRIMADDRESS ptr of the G400 is set to zero and remains zero; and the X server is caught in an infinite loop where it tries to flush the DMA buffers. Oh well. I'll have a look at the chip specifications. Actutally, the PCI version of the G450 comes as a PCI-PCI bridge where the G450 sits on the the secondary bus as a single device. Thx for your answer. Greetings Claus -- Claus-Justus Heine [EMAIL PROTECTED] http://www.instmath.rwth-aachen.de/~heine/ Ftape - the Linux Floppy Tape Project Home Page : http://www.instmath.rwth-aachen.de/~heine/ftape/ _______________________________________________ Dri-devel mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/dri-devel
