trying to load the nvidia drivers gets a gpf as it tries to read/write MTRRs. ron ------------------------------------------------------- This SF.Net email sponsored by Black Hat Briefings & Training. Attend Black Hat Briefings & Training, Las Vegas July 24-29 - digital self defense, top technical experts, no vendor pitches, unmatched networking opportunities. Visit www.blackhat.com _______________________________________________ Xen-devel mailing list Xen-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/xen-devel
> trying to load the nvidia drivers gets a gpf as it tries to read/write > MTRRs.Which nvidia driver? One in the Linux kernel, or can it work just by using a suitable driver in the Xserver? What mechanism does it use to write the mtrr''s ? (these could be virtualised without too much trouble -- there''s already an rd/wrmsr.) We''ve had X working with a bunch of different graphics cards, but I don''t think I''ve seen an nvidia. Ian ------------------------------------------------------- This SF.Net email sponsored by Black Hat Briefings & Training. Attend Black Hat Briefings & Training, Las Vegas July 24-29 - digital self defense, top technical experts, no vendor pitches, unmatched networking opportunities. Visit www.blackhat.com _______________________________________________ Xen-devel mailing list Xen-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/xen-devel
On Tue, 13 Jul 2004, Ian Pratt wrote:> Which nvidia driver? One in the Linux kernel, or can it work just > by using a suitable driver in the Xserver? What mechanism does > it use to write the mtrr''s ? (these could be virtualised without > too much trouble -- there''s already an rd/wrmsr.)latest driver, which is of course mostly a binary module. I''ll look for what source I can. I am guessing it stupidly does a rdmsr/wrmsr since it got the GPF.> We''ve had X working with a bunch of different graphics cards, but > I don''t think I''ve seen an nvidia.Good cards, hard company to work with. ron ------------------------------------------------------- This SF.Net email sponsored by Black Hat Briefings & Training. Attend Black Hat Briefings & Training, Las Vegas July 24-29 - digital self defense, top technical experts, no vendor pitches, unmatched networking opportunities. Visit www.blackhat.com _______________________________________________ Xen-devel mailing list Xen-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/xen-devel
> latest driver, which is of course mostly a binary module. I''ll look for > what source I can.IIRC, for decent 3D acceleration you need a binary-only kernel module (no source due to licensing issues) but there is open-source support for 2D (I guess in an X server module). I would have thought binary kernel modules from i386 Linux wouldn''t work with Linux on Xen at present. Mark ------------------------------------------------------- This SF.Net email sponsored by Black Hat Briefings & Training. Attend Black Hat Briefings & Training, Las Vegas July 24-29 - digital self defense, top technical experts, no vendor pitches, unmatched networking opportunities. Visit www.blackhat.com _______________________________________________ Xen-devel mailing list Xen-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/xen-devel
> On Tue, 13 Jul 2004, Ian Pratt wrote: > > > Which nvidia driver? One in the Linux kernel, or can it work just > > by using a suitable driver in the Xserver? What mechanism does > > it use to write the mtrr''s ? (these could be virtualised without > > too much trouble -- there''s already an rd/wrmsr.) > > latest driver, which is of course mostly a binary module. I''ll look for > what source I can. > > I am guessing it stupidly does a rdmsr/wrmsr since it got the GPF.Ouch, if its a binary driver its probably pretty doomed. All sorts of things could be causing problems e.g. cli/sti. Doing an objdump -d and doing a bit of grep''ing might reveal the scale of the problem. For a laugh, why not fire-up your favourite binary editor and turn the rd/wrmsr''s into nops. (You could do do them in the src wrapper anyhow.) If you actually get somewhere with this approach, we could consider having a special linux GPF handler that decodes the instruction and patches these up. Barf. Ian ------------------------------------------------------- This SF.Net email sponsored by Black Hat Briefings & Training. Attend Black Hat Briefings & Training, Las Vegas July 24-29 - digital self defense, top technical experts, no vendor pitches, unmatched networking opportunities. Visit www.blackhat.com _______________________________________________ Xen-devel mailing list Xen-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/xen-devel
On Tue, 13 Jul 2004, Mark Williamson wrote:> I would have thought binary kernel modules from i386 Linux wouldn''t work with > Linux on Xen at present.works very well until it hits the attempt to get an MTRR :-) ron ------------------------------------------------------- This SF.Net email sponsored by Black Hat Briefings & Training. Attend Black Hat Briefings & Training, Las Vegas July 24-29 - digital self defense, top technical experts, no vendor pitches, unmatched networking opportunities. Visit www.blackhat.com _______________________________________________ Xen-devel mailing list Xen-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/xen-devel
On Tue, 13 Jul 2004, Ian Pratt wrote:> If you actually get somewhere with this approach, we could > consider having a special linux GPF handler that decodes the > instruction and patches these up. Barf.you guys have nicely avoided putting in insn decode so far, best avoided on these horrible CPUs :-0 ron ------------------------------------------------------- This SF.Net email sponsored by Black Hat Briefings & Training. Attend Black Hat Briefings & Training, Las Vegas July 24-29 - digital self defense, top technical experts, no vendor pitches, unmatched networking opportunities. Visit www.blackhat.com _______________________________________________ Xen-devel mailing list Xen-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/xen-devel
> On Tue, 13 Jul 2004, Ian Pratt wrote: > > > If you actually get somewhere with this approach, we could > > consider having a special linux GPF handler that decodes the > > instruction and patches these up. Barf. > > you guys have nicely avoided putting in insn decode so far, best avoided > on these horrible CPUs :-0If Mark''s right about there being a 2D driver in the Xserver that works without the nvidia binary-only kernel driver, then that''s definitely the way to go. I seriously think it''s worth a doing a disassembly of the binary module and seeing the scale of the problem. If we''re lucky and it''s just the rd/wrmsr that are the problem, they''ll be easy to deal with by nop''ing them. If there are cli/sti then that''s more of a pain, but it would actually be pretty easy to do in a Linux (as opposed to Xen) GPF handler that spots them and does the appropriate Xen call. They''re both 1 byte instructions (0xfa/b) with no operands, so easy to spot. As an alternative, you could use static binary rewriting. I''m not too familiar with the various x86 binary re-writing tools that exist, but perhaps one is up to the job? It''s almost do-able with objdump -d | awk | gas ;-) Can anyone recommend a good static binary rewriting tool for x86? Ian ------------------------------------------------------- This SF.Net email sponsored by Black Hat Briefings & Training. Attend Black Hat Briefings & Training, Las Vegas July 24-29 - digital self defense, top technical experts, no vendor pitches, unmatched networking opportunities. Visit www.blackhat.com _______________________________________________ Xen-devel mailing list Xen-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/xen-devel
> If Mark''s right about there being a 2D driver in the Xserver that > works without the nvidia binary-only kernel driver, then that''s > definitely the way to go. > > I seriously think it''s worth a doing a disassembly of the binary > module and seeing the scale of the problem. If we''re lucky and > it''s just the rd/wrmsr that are the problem, they''ll be easy to > deal with by nop''ing them. If there are cli/sti then that''s more > of a pain, but it would actually be pretty easy to do in a Linux > (as opposed to Xen) GPF handler that spots them and does the > appropriate Xen call. They''re both 1 byte instructions (0xfa/b) > with no operands, so easy to spot.Retaining the possibility of doing this kind of thing is another good reason for keeping PAGE_OFFSET the same as native in our Linux 2.6 port as well as the 2.4 port. If __va() and __pa() change then that''s likely to be a much bigger rewriting problem. -- Keir ------------------------------------------------------- This SF.Net email sponsored by Black Hat Briefings & Training. Attend Black Hat Briefings & Training, Las Vegas July 24-29 - digital self defense, top technical experts, no vendor pitches, unmatched networking opportunities. Visit www.blackhat.com _______________________________________________ Xen-devel mailing list Xen-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/xen-devel