[vfio-users] Passthrough VGA on first slot

Sat Oct 3 20:37:16 UTC 2015

I didn't gave up hope yet, If it was a pci-e 3.0 to pci-e 2.0 I would be
using the second slot without problems, but, in this case, it's from a
pci-e 2.0 x16 to a x4. And yes, MSI was also reducing my performance, but
if there is a way for me to guarantee 100% of the available GPU power
available, why would I give up...

I decided to change the GPUs again, log everything I could as well log when
it doesn't work.
I've resumed them all here:
https://gist.github.com/PedroHLC/5c54785a00efa3f08154
As well as I added relevant settings and some hardware specs.

I've observed some things, while on PCIEX4 bus (secondary), I do not need a
ROM, and if I pass one, even a downloaded one or one I ripped myself, It
wouldn't work.
QEMU only throws ROM errors when on PCIEX16 bus (primary), as well it seems
to be missing this on dmesg: "vfio-pci 0000:0?:00.0: enabling device (????
-> ????)"

So as, VFIO isn't enabling it, there is still something holding it up? If
it was, shouldn't it appear on /proc/iomem? Is there a way to vfio verbose
more intel?

2015-10-02 18:26 GMT-03:00 Blank Field <ihatethisfield at gmail.com>:

> To answer your last question you must examine your dmesg.
>
> Don't worry, on a system level it is possible to eject any device in
> windows. I've seen a hardware platform that could eject one of cpu's cores
> and system boot drive too. As in Linux, when you do that either nothing
> happens or nothing good happens.
>
> I personally would test if your GPU really needs the full pci-e 3.0
> bandwidth. I mean, when you got it running, maybe it lags due to msi, cpu
> latency issues or something else?
> Because HDMI audio stuttering problem points me directly to alex's blog
> post about message signalled interrupts.
> And yeah, if you could switch the primary GPU the gigabyte way - that
> would solve your initial problem. So consider updating your firmware...
> Both Seabios and OVMF.
> Turn on/off the machine? I did tried restarting it sometimes...
> Attaching an QXL video I can see NVIDIA driver with the classic 43 error,
> but I don't thing of it being qemu fault as when vga is in slot2 this
> doesn't happen.
> If I disable nvidia device and reenable it, it goes back to error 43.
> Weird observation, I can eject the GTX780, and the SCSI drivers in Windows
> eject tray icon....
> For one last test, I uninstalled the drive in devmgmt, turned off the
> machine, turned it on without QXL, waited a long time to IDK, maybe Windows
> to reinstall the GPU, called "system_powerdown", ran it again in hope, but
> none.
>
> Also, should my GPU still be the "target" of vga_arbiter?
> If I'm using 'options vfio-pci ids=' should I also try pci-stub?
>
> 2015-10-02 13:47 GMT-03:00 Blank Field <ihatethisfield at gmail.com>:
>
>> Seabios? OVMF? Have you tried turining it on and off?^W^W^W^W^W
>> reinstalling windows guest drivers? Attaching an emulated QXL device to see
>> devmgmt.msc?
>> On Oct 2, 2015 7:43 PM, "Pedro Henrique Lara Campos" <
>> pedro.laracampos at gmail.com> wrote:
>>
>>> Yep, you both we're right, uvesafb was using GPU, now nothing besides
>>> vfio is using it, yet I have no video, but there are no more mmap
>>> unsupported errors, with a romcard there is no ROM error, there are also no
>>> errors in AMD-Vi, qemu, IOMMU. Windows still loads fine, I can turn it off
>>> through monitor, every qemu setting is identical it was when GTX was in
>>> slot2 (in this slot I could see from OVMF bios until machine goes off) with
>>> exception of host pci addr but in slot1 it doesn't seems to want to start
>>> at all...
>>>
>>> 2015-10-02 10:27 GMT-03:00 Blank Field <ihatethisfield at gmail.com>:
>>>
>>>> Do cat /proc/iomem to find out who is using your GPU. I predict uvesafb
>>>> or something like that. Disable or uninstall this.
>>>> Also, don't you have pci-e 3.0 onboard?
>>>> On Oct 2, 2015 4:22 PM, "Pedro Henrique Lara Campos" <
>>>> pedro.laracampos at gmail.com> wrote:
>>>>
>>>>> Hello masters of vfio. I've been playing with vfio for 3 weeks now,
>>>>> and here is a resume of what i got: When I put my loved GTX780 on the
>>>>> second pci-e slot of my GA-970A-DS3P everything works on Windows (guest)
>>>>> while my Linux (host) stays comfortable with a Radeon HD 6670.
>>>>> But, as I wouldn't be writing this email if nothing was wrong, my
>>>>> second pci-e slot is "PCIEX4" only, I believe that's because my FPS goes
>>>>> down and HDMI audio keeps crackin. I had to change the slots, my bios
>>>>> doesn't have an option to choose which GPU to use, so it uses the GTX, GRUB
>>>>> does it too, then gpu gets binded by modprobe and I have no video until
>>>>> XOrg starts on Radeon, if I start qemu, I've get:
>>>>>   "vfio_ecap_init: 0000:01:00.0 hiding ecap 0x19 at 0x900"
>>>>>   "BAR 3 mmap unsupported. Performance may be slow"
>>>>>   "vfio-pci 0000:01:00.0: Invalid ROM contents"
>>>>> No video on GTX, Windows boots normally, I can even hear notifications
>>>>> sounds, I can turn it off through "system_powerdown".
>>>>> Tried using a romfile, it only makes the third message goes away, no
>>>>> other thing changes...
>>>>>
>>>>> _______________________________________________
>>>>> vfio-users mailing list
>>>>> vfio-users at redhat.com
>>>>> https://www.redhat.com/mailman/listinfo/vfio-users
>>>>>
>>>>>
>>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listman.redhat.com/archives/vfio-users/attachments/20151003/5cc79df8/attachment.htm>