On 25/07/16 16:29, Andrew Jones wrote: > On Mon, Jul 25, 2016 at 04:07:20PM +0100, Marc Zyngier wrote: >> On 25/07/16 15:28, Andrew Jones wrote: >>> On Mon, Jul 25, 2016 at 03:05:46PM +0100, Marc Zyngier wrote: >>>> News flash: Mark Rutland just saw it crashing on his Seattle using my >>>> kernel that doesn't crash on mine. So we're looking at external factors >>>> now. DT, firmware, whatever. >>> >>> Interesting, especially with the information that the bug is also >>> on v7 with uboot (where it's present with 4k pages), and on v8 >>> with UEFI (where it's not present with 4k pages, only 64k) >> >> Here's what I get with Mark's DT: >> >> [ 2.199618] kvm [1]: 8-bit VMID >> [ 2.202756] kvm [1]: Hyp mode initialized successfully >> [ 2.207907] kvm [1]: GICV size 0x2000 not a multiple of page size 0x10000 >> [ 2.214736] kvm [1]: virtual timer IRQ4 >> >> Somehow, we fail to detect that we cannot run with 64k pages, >> because we cannot safely pass the 8kB region to the guest. >> KVM should fail to initialize, or at least report that there >> is no VGIC. Drew, is that what you have on your system as well? > > Yup, precisely. I wish I would have grepped and reported that before... > > I also now see that when booting with old vgic and 64k pages I get an > additional message after the GICV... line > > [ 1.889241] kvm [1]: error: KVM vGIC probing failed > > So we should port that error message over. Not just the message. Somehow, the rest of KVM still thinks we have things up and running, while we won't be able to deliver an interrupt. Pretty shit. Now, coming back to Stefan's issue, I see that his DT lacks a maintenance interrupt in the GIC node, which would be another failure mode. Except that he says that it works with the old vgic. That's weird. Basically, our error handling is bonkers and needs fixing. Thanks, M. -- Jazz is not dead. It just smells funny... _______________________________________________ kvmarm mailing list kvmarm@xxxxxxxxxxxxxxxxxxxxx https://lists.cs.columbia.edu/mailman/listinfo/kvmarm