Description of problem: This has similarities to the following bug, but is much rarer: https://bugzilla.redhat.com/show_bug.cgi?id=1194366 I have a script which boots a guest 1000 times on the Fedora/aarch64 host. Guest and host kernels are identical (4.2.0-1.fc24.aarch64). About 5 in every 1000 boots fail. The host kernel prints: kvm [3683]: load/store instruction decoding not implemented The corresponding qemu process hangs after printing: error: kvm run failed Function not implemented PC=000000006bbfd238 SP=00000000700000b0 X00=aa1903e1aa0303e2 X01=0000000068e67b40 X02=000000006baa7cec X03=0000000000e68200 X04=000000006bac1398 X05=00000000009ffaf8 X06=0000000000000000 X07=000000006f04c85c X08=000000006f04cb78 X09=0000000000000000 X10=0000000000000004 X11=0000000000000000 X12=00000000700fe0fa X13=0000000000000000 X14=0000000000000000 X15=0000000000000000 X16=000000006f04cdf0 X17=0000000000000000 X18=0000000000000000 X19=000000006bff0018 X20=0000000000000000 X21=0000000000000000 X22=0000000000000000 X23=0000000000000000 X24=0000000000000000 X25=0000000000000000 X26=0000000000000000 X27=0000000000000000 X28=0000000000000000 X29=0000000000000000 X30=0000000000000000 PSTATE=60000305 (flags -ZC-) The PC address does not correspond to any kernel address. Version-Release number of selected component (if applicable): kernel 4.2.0-1.fc24.aarch64 How reproducible: Rare, approximately 1 in 200 boots. Steps to Reproduce: 1. In the libguestfs test suite, run: ./tests/qemu/qemu-boot -n 1000 Additional info: The error message usually indicates that the guest has jumped into random code. I'm still investigating this bug, will update this bug with further details as I collect it.
Created attachment 1081033 [details] serial port log from guest Even though this is Fedora, I'm using AAVMF (AAVMF-20141113-5.git77d5dac.aa7a.aarch64). It appears to be crashing in AAVMF. Attached is the guest serial port log up to the point where the guest stops.
Created attachment 1081034 [details] libvirt/qemu log from guest This is the libvirt/qemu log from the same guest as in comment 1.
As this works with Gerd's edk2.git/aarch64 package, I'm going to say this was because I was using a too old version of AAVMF, so it's likely not a bug.