[vfio-users] [FEEDBACK NEEDED] Additionnal tips to include in the Arch wiki article

Alex Williamson alex.l.williamson at gmail.com
Fri May 20 20:27:17 UTC 2016


On Fri, May 20, 2016 at 11:38 AM, Nicolas Roy-Renaud <
nicolas.roy-renaud.1 at ens.etsmtl.ca> wrote:

> Did you have any issues with efifb or some other driver grabbing the gard
> before vfio-pci? What are your IOMMU groups like, which chard is your
> boot_vga, what script or method do you use to get vfio-pci to bind with
> your card, what's your kernel command line and what modules are included in
> your initramfs?
>
> I've struggled with getting my primary GPU to work correctly for a
> passthrough like this, and I ended up switching it up with the secondary
> because I simply couldn't get it to work properly (I'd keep getting
> something about "Invalid ROM contents") and I couldn't see my boot logs
> until the host driver finished loading. It's really inconvenient now
> becuase my motherboard is designed so that having a full-size GPU in the
> second PCIe slot blocks 4 SATA ports out of 6, so I need to have L-shaped
> cables in a bunch of places to make everything work properly.
>

The boot VGA ROM is handled differently from other PCI ROMs in Linux, when
you read it you're actually reading a shadow copy of it placed at 0xc0000,
which has its roots back in the original IBM PC history.  Execution of the
VGA ROM is allowed to modify this copy of the ROM image in memory.  Thus
you're really not getting a pristine copy of the device ROM and it can
often lead to these invalid ROM contents messages.  I wish we provided raw
access to the PCI option ROM in these cases as well.  For now, if you do
have these sorts of invalid ROM messages, I would suggest that you at least
boot with the card as secondary, dump the ROM, then use that collected
image with the romfile= option rather than use the shadow copy.  I had to
add some code to vfio to fix device IDs and checksums for IGD assignment
because of this, I don't know if there's any standard fixup we can do for
other vendors.  Thanks,

Alex
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listman.redhat.com/archives/vfio-users/attachments/20160520/c4f92139/attachment.htm>


More information about the vfio-users mailing list