Installing nvidia drivers in dom0

[ver...@riseup.net]

Due to graphical issues (i.e., I can't run X at all), I need to install
the nvidia drivers. However, this must be done in dom0, and I can't figure
out how to get it to install there.

I have tried following the instructions at
https://www.qubes-os.org/doc/InstallNvidiaDriver/, but they're outdated
now and the command

"yum install kernel-devel rpm-build kmodtool rpmbuild --nodeps -D "kernels
`uname -r`" --rebuild nvidia-kmod-260.19.36-1.fc13.3.src.rpm"

in particular didn't work (I ran that one in Fedora 21, btw), so I've been
hoping that I would be able to make do with the RPMs that I WAS able to
fetch.

I transferred the RPM files that I was able to fetch on Fedora over to a
USB, which I have made available locally on my Qubes installation. I have
tried "sudo yum install /mnt/usb/kmod-nvidia...rpm," "sudo
qubes-dom0-update /mnt/usb/kmod-nvidia...rpm" and so on. They don't work.
Qubes tries to fetch them from a server and can't because, I presume, dom0
isn't networked.

Anybody had success installing nvidia drivers (though I'm willing to
settle for ANY drivers at this point) in dom0?

[raf...@elitemail.org]

Yes, I have. You will want to use the 3.19.8-100 kernel in dom0, and
installing the drivers requires one to trust RpmFusion in dom0, as
well. No compiling or rebuilding is necessary.

First, download the signing key from RpmFusion here:

http://rpmfusion.org/keys?action=AttachFile&do=get&target=RPM-GPG-KEY-rpmfusion-nonfree-fedora-20

Then move that to dom0 following these instructions (safer than using a
usb stick):

https://www.qubes-os.org/doc/CopyToDomZero/

Then import the key in dom0:

@dom0 ~]$ sudo rpmkeys --import /path/to/signing/key

Then download at least the kmod-nvidia package and the
xorg-x11-drv-nvidia package (probably need the xorg-x11-drv-nvidia-libs
package, too) from here:

http://download1.rpmfusion.org/nonfree/fedora/updates/20/x86_64/repoview/index.html

...and here, respectively:

http://download1.rpmfusion.org/nonfree/fedora/updates/20/x86_64/repoview/letter_x.group.html

Then move the packages to dom0 using the previous instructions.

Then verify the packages in dom0:

@dom0 ~]$ rpm -K /path/to/packages/*.rpm

(if you see anything saying 'NOT ok', then signature verification failed)

Then, simply install the packages:

@dom0 ~]$ sudo yum install /path/to/packages/*.rpm

Reboot.

---

These steps will get the packages, verify, and install them in dom0.
From this point, it becomes a matter of getting X to use the driver in
conjunction with your hardware. I ran into issues at that point, and
didn't take it farther as I don't need my Nvidia gpu in Qubes, but YMMV
depending on your hardware, etc.

Another option would be to simply blacklist the nouveau driver at boot
(you might want to try this first, as it should be easier to follow the
above instructions after blacklisting):

@dom0 ~]$ sudo vi /etc/default/grub

add:

nouveau.modeset=0 rd.driver.blacklist=nouveau

to the end of the line beginning with GRUB_CMDLINE_LINUX=... (you could
also do this one-time-only by adding those options at the grub boot promt)

then apply:

@dom0 ~]$ sudo grub2-mkconfig -o /boot/grub2/grub.cfg

and reboot.


Blacklisting nouveau forces use of my integrated Intel gpu, and X works
fine at that point. Again, YMMV depending on your hardware.

[ver...@riseup.net]

> Then, simply install the packages:
>
> @dom0 ~]$ sudo yum install /path/to/packages/*.rpm
>
> Reboot.

It's the "simply install the packages" step that trips me up. I copied all
the files from USB (I'm a hobbyist who's just trying to figure out if I
can get this to work before I think about a serious installation, so USB
safety isn't a concern at this point), checked them with 'rpm -K
/path/to/file/*.rpm, and everything came up OK. But when I try to install
the packages, I get a screenful of info—only half of which I can see
because the screen prints it so fast–telling me to "disable the
qubes-dom0-cached repo" or "configure the failing repo to be skipped" (I'm
assuming yum is trying to get on the internet but can't because dom0 isn't
connected to it). The only info I can see in full is at the bottom of the
screen:

"failure: repodata/repomd.xml from qubes-dom0-cached: [Errno 256] No more
mirrors to try.
file:///var/lib/qubes/updates/repodata/repomd.xml"

Anyway, I just want to say thank you SO MUCH for offering a helping hand
here! I've been trying to get it working for 3 weeks or so now, and if I
can just figure out how to get it installed in dom0, I think I'll be good
to go (or at least good to both a bunch of different people on an nvidia
board! ;)).

> Another option would be to simply blacklist the nouveau driver at boot
> (you might want to try this first, as it should be easier to follow the
> above instructions after blacklisting):
>
> @dom0 ~]$ sudo vi /etc/default/grub
>
> add:
>
> nouveau.modeset=0 rd.driver.blacklist=nouveau
>
> to the end of the line beginning with GRUB_CMDLINE_LINUX=... (you could
> also do this one-time-only by adding those options at the grub boot promt)
>
> then apply:
>
> @dom0 ~]$ sudo grub2-mkconfig -o /boot/grub2/grub.cfg
>
> and reboot.
>
>
> Blacklisting nouveau forces use of my integrated Intel gpu, and X works
> fine at that point. Again, YMMV depending on your hardware.
>

I tried this step first per your instructions, but no dice. I still have
to wait for it to boot into its usual blank screen and then hit
fn+ctrl+alt+F2 to switch to tty2 and get all my work done from there.

[Marek Marczykowski-Górecki]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA256

On Sun, Sep 27, 2015 at 07:39:40PM -0700, ver...@riseup.net wrote:
> > Then, simply install the packages:
> >
> > @dom0 ~]$ sudo yum install /path/to/packages/*.rpm
> >
> > Reboot.
>
> It's the "simply install the packages" step that trips me up. I copied all
> the files from USB (I'm a hobbyist who's just trying to figure out if I
> can get this to work before I think about a serious installation, so USB
> safety isn't a concern at this point), checked them with 'rpm -K
> /path/to/file/*.rpm, and everything came up OK. But when I try to install
> the packages, I get a screenful of info—only half of which I can see
> because the screen prints it so fast–telling me to "disable the
> qubes-dom0-cached repo" or "configure the failing repo to be skipped" (I'm
> assuming yum is trying to get on the internet but can't because dom0 isn't
> connected to it). The only info I can see in full is at the bottom of the
> screen:
>
> "failure: repodata/repomd.xml from qubes-dom0-cached: [Errno 256] No more
> mirrors to try.
> file:///var/lib/qubes/updates/repodata/repomd.xml"

This repository is used to install packages downloaded by
qubes-dom0-update tool. You can disable it by
- --disablerepo=qubes-dom0-cached yum option.

- --
Best Regards,
Marek Marczykowski-Górecki
Invisible Things Lab
A: Because it messes up the order in which people normally read text.
Q: Why is top-posting such a bad thing?
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1

iQEcBAEBCAAGBQJWCPFlAAoJENuP0xzK19cs6ZYH/2YWZ0vpgZViADUUfU85XTtu
Y9cu6lgsoJaG1xN8em0gfn4Yfmf4BQ9yRZpodVgzw8EykVvBRXgjexLmotEhHK+w
d8yrHYc3T/hknshY9pEsX4BssJ7sjnpZjRdXAh0Utyfkg0YBFln+jbsE2h/KI2Ho
zu+zeRJdVKgzIPBva4OUFOv0owyRJxvVOgTu+BZ3GezKKc1T09LYYm6YIKNmghwq
P2jAv+lZn8EhYtCrulFxEYfSBDJK22XKDpirr4hYRhCVwpfNIMd8nW39K6NKMd4d
0/xARW9GQzAC4wsESaUxy2ghxj6WgmF2C81gMOWPRkJBAtcWqXoq9ZwE9C8DFwo=
=OJ+H
-----END PGP SIGNATURE-----

[ver...@riseup.net]

> This repository is used to install packages downloaded by
> qubes-dom0-update tool. You can disable it by
> - --disablerepo=qubes-dom0-cached yum option.
>

Ah! That seems to work...or rather, it would work if I weren't getting
prompted to install dependencies (specifically, libvdpau >= 0.5).

I'll check whether this is a one-off or indicative of dependency hell and
update my findings.

[ver...@riseup.net]

The good news: libvdpau was the only dependency I needed to install. But
interestingly, I had to install the nvidia RPMs in a batch (i.e., "sudo
yum install *.rpm"), because when I tried to do it one at a time, the
nvidia package would say "we need the nvidia-libs package," and the
nvidia-libs package would say, "we need the nvidia package," throwing me
in a loop. Just a note in case anyone runs into that problem.

The bad news: installing the graphics drivers seems to have made things
worse. After successfully installing the 331xx versions of all the nvidia
drivers, I rebooted, chose kernel 3.19, and bam........right into a blank
screen. And I mean blank: I couldn't even switch into tty2 using the
typical shortcut fn+ctrl+option+F2. The only thing the computer seemed
capable of doing was getting hot (and boy, did it get HOT).

Though I'm inexplicably reinstalling Qubes on my Macbook now (just to see
if I haven't done LITERALLY everything I can think of to make it work), I
think we can safely chalk late 2008 MBPs up as "aggressively unsupported"
for the time being.

Thanks for all your help, though!

["Marek Marczykowski-Górecki"]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA256

Do you know anything about support for this particular hardware in
baremetal Linux distributions (like Ubuntu, Fedora)? I guess this
version of binary nvidia drivers wont work (based on your experience),
but maybe there is some way to get open source driver working...
Watch this ticket: https://github.com/QubesOS/qubes-issues/issues/794
I'll be posting there next test images soon - besides EFI support, it
also will contain updated kernel (4.1.x) and X drivers.

- --
Best Regards,
Marek Marczykowski-Górecki
Invisible Things Lab
A: Because it messes up the order in which people normally read text.
Q: Why is top-posting such a bad thing?
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1

iQEcBAEBCAAGBQJWCfBNAAoJENuP0xzK19csCGgH/3A19/5yJB8N1KK7z/5E0qoC
7XhvBYRU1uNKVcJuVgCREh8pYzTKFrsDkDEq/GelcvCDMep27WuasLN1Nw3O4RRK
lV6Pv6U6G1kNQ9JkmkKWidKks9BEybAcrdtMg5pLPG3mgXcmC3Vb4gE/2f0m44Mj
vvEFiljMpZmhvvrC5rVJv7myd8rt2hI10EsnLxqGAS8DF+hcXnOmgyNPc6rnxK87
AVdL2YP7b3A6PltatEPFH6p+O4Nh1SQpL5TjyWB1gpUNC9xJdcp0SSBTdUCjUj5F
ZnbygntlvJpkipcZvC9NOunWWDbmRLpi0IqpN69039l12nWQYrNMMyjvFpJt/Xk=
=OquF
-----END PGP SIGNATURE-----

[ver...@riseup.net]

> Do you know anything about support for this particular hardware in
> baremetal Linux distributions (like Ubuntu, Fedora)?

I've successfully installed Ubuntu, Xubuntu, Fedora 22 (in basic graphics
mode), PC-BSD, OpenBSD, and elementary OS on this Macbook. Ubuntu,
Xubuntu, and elementary OS were all extremely choppy with Nouveau so I had
to use the closed-source drivers, but OpenBSD and Fedora 22 ran fine
without any driver modification.

> I guess this
> version of binary nvidia drivers wont work (based on your experience),
> but maybe there is some way to get open source driver working...
> Watch this ticket: https://github.com/QubesOS/qubes-issues/issues/794
> I'll be posting there next test images soon - besides EFI support, it
> also will contain updated kernel (4.1.x) and X drivers.

Thanks for the tip! I'll definitely try the next image and write an update
if it ends up working.

[daltong defourne]

Sorry to necro a year-old discussion, but I'd like to ask if any of the participants (OP?) have had success with this operation.

I need to get GTX 1070 box sorta kinda working with Qubes 3.2, and permanently installing a different GPU or using integrated graphics is, sadly, not an option (can use diff GPU as temporary solution in order to get through install process tho) ...