On Thu, 2021-12-09 at 01:37 +0000, Sean Christopherson wrote: > On Thu, Dec 09, 2021, Maxim Levitsky wrote: > > On Thu, 2021-12-09 at 01:00 +0200, Maxim Levitsky wrote: > > > Probably just luck (can't reproduce this anymore) but > > > while running some kvm unit tests with this patch series (and few my patches > > > for AVIC co-existance which shouldn't affect this) I got this > > > > > > (warning about is_running already set) > > ... > > > Also got this while trying a VM with passed through device: > > A tangentially related question: have you seen any mysterious crashes on your AMD > system? I've been bisecting (well, attempting to bisect) bizarre crashes that > AFAICT showed up between v5.15 and v5.16-rc2. Things like runqueues being NULL > deep in the scheduler when a CPU is coming out of idle. I _think_ the issues have > been fixed as of v5.16-rc4, but I don't have a good reproducer so bisecting in > either direction has been a complete mess. I've reproduced on multiple AMD hosts, > but never on an Intel system. I have a sinking feeling that the issue is > relatively unique to our systems :-/ I did had my 3970X lockup hard once on 5.16-rc2. It locked up completely without even sending anything over a a pcie serial port card I. I don't remember what I was doing during the crash but probably had some VMs running. Since then it didn't happen again, and I am running 5.16-rc3 for some time with Paolo's kvm/queue merged and my own patches. > > And a request: any testing and bug fixes you can throw at the AVIC changes would be > greatly appreciated. I've been partially blocked on testing the AVIC stuff for the > better part of the week. If the crashes I'm seeing have been resolved, then I should > be able to help hunt down the issues, but if not... > This is what I started doing last evening, and I'll go through all of the usual testing I do soon. Best regards, Maxim Levitsky