[vfio-users] Physical disk passthrough and OVMF

[Dan Ziemba]

On Fri, 2016-03-25 at 01:10 +0300, Blank Field wrote:
> Hello again, fellow subscribers.
> 
> Since 4.3.X kernels I am experiencing some strange problems with disk
> IO.
> 
> I am passing through the whole /dev/sda(since i also dualboot into
> that system for some weird software or hardware) to my VM, and since
> recent versions of kernel the VM gets stuck in an endless loop,
> consuming 100% one core.
> 
> The ISA debug console says stuff like:
> PartitionValidMbr: Bad MBR partition size EndingLBA(D99299D3) >
> LastLBA(31FFF)
> 
> The thing is that the disk in question has nothing to do with MBR, it
> is formatted into GPT.
> I can't pinpoint the source of that weird D99299D3 number.
> 
> That kind of error happens for every partition detected.
> 
> Device          Start        End    Sectors   Size Type
> /dev/sda1        2048 1171875000 1171872953 558.8G Microsoft basic
> data
> /dev/sda2  1171875840 1172080639     204800   100M EFI System
> /dev/sda3  1172080640 1172342783     262144   128M Microsoft reserved
> /dev/sda4  1172342784 1465145343  292802560 139.6G Microsoft basic
> data
> /dev/sda5  1465145344 1953525134  488379791 232.9G Microsoft basic
> data
> That looks like a valid partition table for me. Moreover, the fact
> that i am able to boot an UEFI-capable system bare-metal confirms
> that
> the partition table is correct.
> 
> I remember i've seen some related messages on the list in february or
> something, but they only stated the existence of this error.
> 
> Have any of you seen stuff like that? If so, how do you live with it?
> 
> Maybe i should write to edk2-devel mailing list for this, because it
> isn't much related to VFIO, but normal qemu folks rarely sacrifice
> the
> whole disk to a VM.
> 
> Versions:
> QEMU emulator version 2.4.1 (qemu-2.4.1-7.fc23)
> kernel 4.4.5-300.fc23.x86_64(doesn't matter much from 4.3.X)
> edk2.git-0-20160311.b1594.gf6326d1.x86_64
> 

A lot of people had problems booting with OVMF with kernels 4.2 and
4.3.  It's fixed for myself and I think most others with 4.4, but I
think you may need fairly recent qemu and ovmf versions.  Looks like
your OVMF is new enough if that date means what I think it does, but
you should try upgrading to qemu 2.5.