On 07/25/17 10:48, Hristo Iliev wrote: > This sounds really a lot like that old kernel MTRR problem
Yes, it does. http://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/?id=879ae1880449c88db11c1ebdaedc2da79b2fe73f (part of v4.4) The symptoms described in this thread (matching the commit message above) finger the phase of the firmware where all processors are started up initially. Moving back to ancient OVMF builds might paper over the symptom (it's possible that those OVMF builds don't even boot up the APs at all). Either way, the time when all APs are executing in parallel in the firmware shouldn't show as even a blip in anyone's favorite load monitor. This sounds like a really messy problem and I don't think I can give any advice until I can log into a machine that reproduces this problem -- even then, KVM tracing looks necessary, interpreting which is usually horribly difficult (to me anyway). So, I'm specifically not asking for kernel versions, OVMF logs and the like. Any chance you guys can use the most recent OVMF build from Gerd's repo, and bisect the host kernel instead? Check out the most recent releases of a few stable / longterm kernel streams first, I guess: 4.12.3, 4.11.12, 4.9.39, 4.4.78 ... Thanks Laszlo _______________________________________________ vfio-users mailing list vfio-users@redhat.com https://www.redhat.com/mailman/listinfo/vfio-users
