[vfio-users] BIOS memory mapping or chipset issue with MMIO prevents IGD pass-through in legacy mode?

[globalgorrilla at fastmail.fm]

Hi Alex,

The first time QEMU is run the display attached via DP comes out of 
standby and then goes into power saving.

This is used to capture the SeaBIOS debug outputs:

   <qemu:commandline>
     <qemu:arg value='-global'/>
     <qemu:arg value='isa-debugcon.iobase=0x402'/>
     <qemu:arg value='-debugcon'/>
     <qemu:arg value='file:/tmp/archlinux.log'/>
   </qemu:commandline>

No 00:02.0 (boots to grub from CD-ROM ISO):
https://pastebin.com/y0xtURW5

With 00:02.0 (hangs):
https://pastebin.com/GPJ3nRq2

/proc/iomem
https://pastebin.com/b4WtKy89

/proc/ioports
https://pastebin.com/8RAbkGy8

Thanks

On 27 Jul 2016, at 19:52, Alex Williamson wrote:

> On Wed, 27 Jul 2016 18:42:29 -0700
> globalgorrilla at fastmail.fm wrote:
>
>> On Wed, Jul 27, 2016, at 06:04 PM, Alex Williamson wrote:
>>
>>> Hmm, the kernel not exposing the VGA regions is really the only way 
>>> we
>>> can fail here and the only two ways that the kernel won't expose 
>>> VGA,
>>> given that enabled in the config, is if the device isn't VGA (which 
>>> we
>>> can see by your class code, it is) or if the disable_vga option to
>>> vfio_pci is used.  You don't happen to have that left over from
>>> assigning discrete devices, do you?
>>>
>>> # cat /sys/module/vfio_pci/parameters/disable_vga
>>
>> Indeed I did, thanks! Now with it gone I still get the DMAR messages 
>> ...
>>
>> [ 4683.644755] DMAR: DRHD: handling fault status reg 3
>> [ 4683.703123] DMAR: [DMA Read] Request device [00:02.0] fault addr
>> 8f440000 [fault reason 06] PTE Read access is not set
>> [ 4683.831145] DMAR: DRHD: handling fault status reg 2
>> [ 4683.889487] DMAR: [DMA Read] Request device [00:02.0] fault addr
>> 8f494000 [fault reason 06] PTE Read access is not set
>
> As noted in the blog post, these aren't necessarily bad so long as 
> they
> are not continuous.  They're probably just in the gap between 
> resetting
> the device and reprogramming the stolen memory range.
>
>> ... but I also get pass-through:
>>
>> [ 4870.366423] kvm: zapping shadow pages for mmio generation 
>> wraparound
>> [ 4887.800200] vgaarb: device changed decodes:
>> PCI:0000:04:00.0,olddecodes=io+mem,decodes=io+mem:owns=none
>> [ 4887.800202] vgaarb: device changed decodes:
>> PCI:0000:00:02.0,olddecodes=io+mem,decodes=io+mem:owns=io+mem
>>
>> No output yet.   Is vgaarb looking correct above? 0000:04:00.0 is an 
>> R9
>> 290.
>
> Matches what I see on my system.  I don't spot anything else wrong.
> Make sure your bios.bin is from your QEMU 2.7-rc0 build tree.  Make
> sure that IGD is set to primary graphics on the host (I assume it is 
> or
> you wouldn't have a ROM).  Try adding '-chardev stdio,id=seabios
> -device isa-debugcon,iobase=0x402,chardev=seabios' to your command 
> line
> and post the log from it.