https://bugzilla.kernel.org/show_bug.cgi?id=209333 Bug ID: 209333 Summary: VM not starting anymore with 5.8.8 - lots of page faults Product: Virtualization Version: unspecified Kernel Version: 5.8.8 Hardware: All OS: Linux Tree: Mainline Status: NEW Severity: normal Priority: P1 Component: kvm Assignee: virtualization_kvm@xxxxxxxxxxxxxxxxxxxx Reporter: kernel@xxxxxxxxxxxxxxxxxx Regression: No Host system is a Threadripper 1920x on a X399 Motherboard. I have a VM here, that I pass through a Samsung 960 EVO SSD. > 08:00.0 Non-Volatile memory controller: Samsung Electronics Co Ltd NVMe SSD > Controller SM961/PM961 The VM starts up and works fine with Kernel 5.8.7, and when I start the VM with a 5.8.8 kernel, libvirt/QEMU say the VM is started, but it doesn't come up, and I see lots and lots of these in dmesg: > AMD-Vi: Event logged [IO_PAGE_FAULT device=08:00.0 domain=0x002c > address=0xfffffffdf8000000 flags=0x0008] Now, one thing to mention is that the VM uses AMD AVIC to directly deliver interrupts to the VM. There are 2 IOMMU related changes in the 5.8.8 changelog that seem likely for the layman that I am, both by Suravee Suthikulpanit from AMD: iommu/amd: Use cmpxchg_double() when updating 128-bit IRTE iommu/amd: Restore IRTE.RemapEn bit after programming IRTE I observe the same behaviour with 5.8.10. Happy to provide more info if needed! -- You are receiving this mail because: You are watching the assignee of the bug.
