[vfio-users] [help] 2 identical GPUs in Arch

[Garland Key]

Thanks Alex,

I consider myself an intermediate power user. I know enough to understand
what the line that you provided does, but not enough to write a solution of
my own. I inserted the line that you provided above in
/modprobe.d/vfio.conf and it caused X to fail. I had to alt+f2 to comment
it out, rebuilt initramfs and then reboot.

It was suggested that I could do this through a systemd service in the arch
irc channel but nobody was able to provide details other than "man
systemd.service".  Any further pointers is greatly appreciated.

Best,
Garland

On Fri, Feb 19, 2016 at 6:06 PM Alex Williamson <alex.williamson at redhat.com>
wrote:

> On Fri, 19 Feb 2016 16:43:47 -0500
> Ben J <btpprograms at gmail.com> wrote:
>
> > For your latter question, it doesn't seem like you should need to modify
> it
> > as his original script already generically finds all GPUs. Did it not
> work
> > when you tried it? Also, what type of cards are you using? If they're
> > NVIDIA with nvidia drivers then I doubt that script will work for you.
> > I'm using the following guide:
> >
> https://vfio.blogspot.com/2015/05/vfio-gpu-how-to-series-part-3-host.html.
> > I'm having trouble adapting some of the Fedora focused instructions for
> use
> > in Arch.
> >
> > *1. Dracut*
> > Dracut isn't used in Arch but the mkinitcpio is. You're asked to
> > insert *add_drivers+="vfio
> > vfio_iommu_type1 vfio_pci vfio_virqfd" *in:
> */etc/dracut.conf.d/local.conf*.
> > Farther down, I'm supposed to insert
> > *install_items+="/sbin/vfio-pci-override-vga.sh
> > /usr/bin/find /usr/bin/dirname" *in: */etc/dracut.conf.d/local.conf*.
> >
> > Should I instead add these lines to a modprobe.d conf file or should I
> add
> > it to the modules line in */etc/mkinitcpio.conf*?
> >
> > *2. driver_override*
> > My bash-foo isn't very good and I need to adapt the following script (see
> > below) so that it points to the correct directories. In Arch, the
> directory
> > for the GPU (02:00.0) and GPU audio (02:00.1) are here:
> > */sys/devices/pci0000:00/0000:00:02.0/0000:02:00.0/* &
> > */sys/devices/pci0000:00/0000:00:02.0/0000:02:00.1/*
> >
> >
> >
> >
> >
> >
> >
> > *    #!/bin/sh    DEVS="0000:02:00.0 0000:02:00.1"    for DEV in $DEVS;
> > do echo "vfio-pci" > /sys/bus/pci/devices/$DEV/driver_override    done
> > modprobe -i vfio-pci*
>
> There are really just so many ways to solve this problem, don't feel
> like you need to stick with the approaches I've given.  If you want
> something that only depends on modprobe.d, then the "install" directive
> is really powerful.  For instance, let's say I have a pair of identical
> nvidia cards, one at 01:00.0 and the other at 02:00.0 and I want to use
> 02: for my guest.  I could do something like:
>
> install nvidia "echo vfio-pci >
> /sys/bus/pci/devices/0000:02:00.0/driver_override; modprobe
> --ignore-install nvidia"
>
> This tells modprobe that before loading the nvidia driver, set the
> driver override so that only vfio-pci can claim 02:00.0, then load the
> nvidia driver.  You can do the same with the audio function and
> snd_hda_intel driver.  You can wait til later to actually load the
> vfio-pci module, it's not really necessary to include them in your
> initramfs.
>
> As for sysfs paths, the actual devices do live in /sys/devices/, but
> there are links to them in /sys/bus/pci/devices/, this is a kernel
> property, your choice of distribution isn't going to change that.
> Thanks,
>
> Alex
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listman.redhat.com/archives/vfio-users/attachments/20160221/a9f66cdb/attachment.htm>