[vfio-users] amd gpu not being seen

Fri Oct 2 17:14:59 UTC 2015

My win10 VM goes black on boot from time to time. Just sayin'
Also, there is a neighbor thread about some kind of regression in recent
kernels.
On Oct 2, 2015 8:08 PM, "Ryan Clawfish" <chimmychainsaw at hotmail.com> wrote:

> I tried the solutions recommended to no avail. Switching pci slots and
> blacklisting the drivers seemed to have no effect on the horrible boot
> performance. I asked my friend if i could borrow my old 770 back to test it
> out in my arch machine and i had the same issues as before with the radeon
> card. I can't help but think the issue is either with my install or a bug
> with the latest 4.2.2.1 kernel as everything was working somewhat fine just
> a few days ago.
>
>  I decided to install fedora on a separate partition thinking if any os
> should work it's one made by redhat. And it did work! Sorta. Following
> alex's awesome vfio blog i got everything setup pretty much properly.
> Windows 10 boots normally with my 280x attached. My only issue now is
> trying to get windows 10 to stop crashing.Using seabios it crashes with the
> atikmdag error and with uefi it simply goes black.
> ------------------------------
> Date: Wed, 30 Sep 2015 10:55:39 +0300
> Subject: Re: [vfio-users] amd gpu not being seen
> From: ihatethisfield at gmail.com
> To: chimmychainsaw at hotmail.com
> CC: vfio-users at redhat.com
>
> Alright, i've studied your dmesg(fell asleep in one place), so:
> You should blacklist the radeon driver. It seems like it's interfering not
> only with vfio, but somehow manages to screw up nvidia host driver.
> And there is a high possibility that you should remove that wireless
> adapter and plug your third GPU there, because iommu grouping may be messed
> up.
> Your host GPU is plugged in the first slot and should be the only one
> doing VGA. But for some reason it is not.
>
> After blacklisting the radeon driver, you should reinstall the guest
> system.
> On Sep 29, 2015 6:23 PM, "Ryan Clawfish" <chimmychainsaw at hotmail.com>
> wrote:
>
> The output was huge and I thought only the vfio entries were relevant. My
> bad. I'm using vfio-pci because it's recommended on the arch wiki. I have
> not yet tried pci-stub.
>
> I've added: *options vfio-pci ids=1002:6798,1002:aaa0* to
> /etc/modprobe.d/vfio.conf
>
> everything else is on http://pastebin.com/3DA3wUBd
>
>
> ------------------------------
> Date: Tue, 29 Sep 2015 17:43:02 +0300
> Subject: Re: [vfio-users] amd gpu not being seen
> From: ihatethisfield at gmail.com
> To: chimmychainsaw at hotmail.com
> CC: vfio-users at redhat.com
>
> Why did you grep...
> Post the full one on pastebin or search for whatever using VGA or that
> GPUs. Do you use any kind of pre-binding(vfio-pci.ids=...) or pci-stub?
> On Sep 29, 2015 5:36 PM, "Ryan Clawfish" <chimmychainsaw at hotmail.com>
> wrote:
>
> Hotmail really needs to start accepting code tags, it would make things so
> much neater.
>
> I'm using an fx 8350 with an asus m5a99 pro R2.0 motherboard.
>
> * lspci -nn*
> 04:00.0 VGA compatible controller [0300]: Advanced Micro Devices, Inc.
> [AMD/ATI] Tahiti XT [Radeon HD 7970/8970 OEM / R9 280X] [1002:6798]
> 04:00.1 Audio device [0403]: Advanced Micro Devices, Inc. [AMD/ATI] Tahiti
> XT HDMI Audio [Radeon HD 7970 Series] [1002:aaa0]
>
> *[ryan at earth-archbox ~]$ dmesg | grep -i vfio*
> [    1.066842] VFIO - User Level meta-driver version: 0.3
> [    1.079992] vfio_pci: add [1002:6798[ffff:ffff]] class 0x000000/00000000
> [    1.093349] vfio_pci: add [1002:aaa0[ffff:ffff]] class 0x000000/00000000
> [  108.858636] vfio-pci 0000:04:00.0: enabling device (0000 -> 0003)
> [  108.858820] vfio_ecap_init: 0000:04:00.0 hiding ecap 0x19 at 0x270
> [  108.858829] vfio_ecap_init: 0000:04:00.0 hiding ecap 0x1b at 0x2d0
> [  108.875285] vfio-pci 0000:04:00.1: enabling device (0000 -> 0002)
> [ 1284.893125] vfio_ecap_init: 0000:04:00.0 hiding ecap 0x19 at 0x270
> [ 1284.893133] vfio_ecap_init: 0000:04:00.0 hiding ecap 0x1b at 0x2d0
> [ 1284.909667] vfio-pci 0000:04:00.1: enabling device (0400 -> 0402)
> [ 1519.790793] vfio_ecap_init: 0000:04:00.0 hiding ecap 0x19 at 0x270
> [ 1519.790805] vfio_ecap_init: 0000:04:00.0 hiding ecap 0x1b at 0x2d0
> [ 1519.810519] vfio-pci 0000:04:00.1: enabling device (0400 -> 0402)
>
> * Hostdev*
> </video>
>     <hostdev mode='subsystem' type='pci' managed='yes'>
>       <source>
>         <address domain='0x0000' bus='0x04' slot='0x00' function='0x0'/>
>       </source>
>       <address type='pci' domain='0x0000' bus='0x00' slot='0x08'
> function='0x0'/>
>     </hostdev>
>     <hostdev mode='subsystem' type='pci' managed='yes'>
>       <source>
>         <address domain='0x0000' bus='0x04' slot='0x00' function='0x1'/>
>       </source>
>       <address type='pci' domain='0x0000' bus='0x00' slot='0x09'
> function='0x0'/>
>     </hostdev>
>
> So you managed to get crossfire inside the emulation? I didn't think it
> was possible but that would be beyond cool. Pain in the ass sure but still
> cool as hell.
>
> also: In my pre coffee daze i accidentally replied to ihatethisfield and
> not the mailing list. Sorry about that.
>
>
> ------------------------------
> Date: Tue, 29 Sep 2015 07:46:23 +0300
> Subject: Re: [vfio-users] amd gpu not being seen
> From: ihatethisfield at gmail.com
> To: chimmychainsaw at hotmail.com
> CC: vfio-users at redhat.com
>
> Your attached XML file lacks the most important part: <hostdev> entries
> with your GPUs.
> Tell us about your CPU-MB combo and do lspci -nn and find all your needed
> devices in their respective iommu groups. And it would be good to see your
> dmesg, maybe you've forgot to blacklist some driver or something else
> tampers with VGA somehow.
>
> Also, on my 2xHD7750 crossfire works flawlessly. They don't need a bridge
> to interconnect, but your cards might need it.
> On Sep 29, 2015 4:26 AM, "Ryan Clawfish" <chimmychainsaw at hotmail.com>
> wrote:
>
> I'm rather happy with 2x 280x's. Hopefully i can use linux virtualization
> for convinence gaming and reboot into a windows partition when i want
> insane crossfire graphics.
>
> To be honest i'm a noob at this and i don't really know what the hell i'm
> doing. I'm mostly following the arch wiki guide
> https://wiki.archlinux.org/index.php/PCI_passthrough_via_OVMF  Because
> /dev/vfios/17 did not exist I decided to reinstall arch and restart the
> process from scratch, roughly documenting my steps. Here's what i did.
>
> installed qemu and rpmextract
> extracted the gerd hoffmons eufi and moved it to /usr/share
> downloaded libvirt
> enabled libvirt and start it
> installed ebtables dnsmasq  bridge-utils and bsd-netcat
> added amd_iommu=on to /etc/default/grub, regenerated grub and rebuilt
> mkinitpcio
> added modprobe vfio-pci
> ran mkinitpcio
> /etc/libvirt/qemu.conf added root permissions + /dev/vfio/17
> created raw file for emulated hdd
> installed virt-manager
> sudo cp /usr/share/edk2.git/ovmf-x64/OVMF-pure-efi.fd
> /usr/share/ovmf/OVMF-win10.fd #pointless?
> sudo ln -s /usr/share/ovmf/OVMF-win10.fd
> /usr/share/ovmf/OVMF-win10.fd                      #pointless
> added uefi into in the os section of my xml file:
>        <loader readonly='yes'
> type='pflash'>/usr/share/edk2.git/ovmf-x64/OVMF_CODE-pure-efi.fd</loader>
>        <nvram
> template='/usr/share/edk2.git/ovmf-x64/OVMF_VARS-pure-efi.fd'/>
> installed windows 10
> installed radeon drivers
> added target pci (gpu) from virt-manager gui
>
> When i add the my gpu  the uefi is delayed and booting windows 10 takes
> forever. There are lots of artifacts and it lags to the point of being
> completely unusable.  I've tried adding my card via virt-manager gui and
> via the (depreciated?)<qemu:commandline>. Both have the same effect.
>
> Here's my xml virt config file http://pastebin.com/sNxvXiQ6
>
>
>
> ------------------------------
> Date: Mon, 28 Sep 2015 09:40:22 -0600
> Subject: Re: [vfio-users] amd gpu not being seen
> From: alex.l.williamson at gmail.com
> To: chimmychainsaw at hotmail.com
> CC: vfio-users at redhat.com
>
> On Mon, Sep 28, 2015 at 9:12 AM, Ryan Clawfish <chimmychainsaw at hotmail.com
> > wrote:
>
> My main issue is i keep getting an error saying "failed to get group 17"
> and "/dev/vfio/17 does not exist."  If i ls -a my /dev , i see /dev/vfio
> does not exist. I created it using mkdir, chmod 666 it and added it in
> /etc/libvirt/qemu.conf but the error persists.
>
>
> So all along, you've probably been trying to use legacy KVM device
> assignment, which led too the Code 43 with the GeForce card.  Too bad,
> IMHO, the 770 was a better card.  vfio has been the default for PCI hostdev
> devices since libvirt 1.1.3 and /dev/vfio/ should be present even before
> vfio is load since kernel v3.14.  Are you even building vfio support into
> your kernel?
>
> _______________________________________________
> vfio-users mailing list
> vfio-users at redhat.com
> https://www.redhat.com/mailman/listinfo/vfio-users
>
> http://pastebin.com/HrCWcd6Shttp://pastebin.com/HrCWcd6Shttp://pastebin.com/HrCWcd6S
>
>
> _______________________________________________
> vfio-users mailing list
> vfio-users at redhat.com
> https://www.redhat.com/mailman/listinfo/vfio-users
>
>
> _______________________________________________
> vfio-users mailing list
> vfio-users at redhat.com
> https://www.redhat.com/mailman/listinfo/vfio-users
>
>
> _______________________________________________
> 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/20151002/4beeddb7/attachment.htm>