Open karlowich opened 7 months ago
The memcpy()
of the mmap'ed register causes an HVF data exit (a vcpu trap/VMEXIT) with len=1
if copying more than 8 bytes. That's in itself a little weird. QEMU's trap handler ends up asserting on isv
(which I also have no idea what is).
I'm 100% out of my comfort zone here, but I do not believe the bug to be in hw/nvme
. It actually kinda feels like the bug would be in the guest kernel doing something that would also be illegal on real hardware (which justifies the QEMU assert).
The
memcpy
inxnvme_be_spdk_sync_cmd_pseudo
causeshvf
in QEMU to fail an assert, resulting in a crash. It is unclear exactly where the bug is - for now it will remain a known issue.