[GTALUG] for any hardware gurus

[Russell]

On July 18, 2017 10:59:57 PM EDT, o1bigtenor via talk <talk at gtalug.org> wrote:
>*working with xrandr 1.5 not finding how to connect second or third
>gpu*
>------------------------------
>Greetings
>
>Running debian buster (testing) using lxqt. Have 3 nvidia 570 gpus and
>4
>1920x1080 monitors.
>Have had this system now for almost 5.5 years and have had it running
>the
>four monitors all that time.

Just to be clear, you were using the vendors drivers on Debian stable and all was well?

>
>Updated to buster (new testing) and found that there were some issues
>with

Older hardware and testing releases don't necessarily play well together.

>proprietary drivers nvidia 375.66 - - - was having one gpu halt and
>upon
>reboot things seemed to work again.

Was it overheating and shutting down or just dropping off? The difference would be helpful in determining which logic is not working.

>
>The time between reboots was getting down to less than 12 hrs when I

If the the mean time between failures is decreasing, this might indicate hardware EOL, but not necessarily, could just as easily be bottlenecking.

>landed
>up with a system with corrupted networking on a reboot.
>So in the process of re-install I decided to try the Nouveau drivers
>instead. So now I have a one gpu 2 monitor system available. Got to
>very
>much like the four monitors for my work layout so I want the other two
>monitors working.

Nouveau development requires someone to harvest the firmware blob from the vendors device and reverse engineer the software, so it is usually not completely functional.

Nvidia cooperates just a little bit with open source but would prefer that people use the drivers they write. Makes sense from a support perspective, it frees them from having to support every spin out there. 

>
>Using xrandr I can see Provider 0,1 and 2.
>DVI-I-1 and DVI-I-2 are connected.
>DVI-I-2-5 and DVI-I-2-6 are listed as disconnected.
>
>I have been able to create a virtual screen of 3840 x 3000 which is how
>I
>had things before this reinstall
>using $xrandr --fb 3840x3000 .
>Tried $xrandr --setprovideroutputsource 1 0
>no changes
>looked for pci addresses
>found 01:00.0. 02:00.0 and 03:00.0
>
>Looked in the kernel log and all I can see is a error code of -19 (for
>both
>Provider 1 and 2) which, after a lot of digging, seems to be saying "no
>such device"

Obviously something broke the hardware abstraction. Was something your bus needed dropped from the testing kernel?

>Tried $ xrandr --addmode DVI-I-2-5 1920x1080
>then DVI-I-2-5 still shows as disconnected but with a resolution of
>1920x1080 60.00
>Next I tried $ xrandr --output VGA-2 --mode 1920x1080 --pos 1280x1280
>--rotate inverted
>result - - warning: output VGA-2 not found; ignoring
>
>It used to be that this setup happened in xorg.conf (and was a
>monolithic
>file), then things were changed to files placed in /etc/xorg.conf.d (as
>individual descriptions - - - ie. devices, monitors and AIUI screens)
>and CV
>in my system these kind of files are now in /etc/X11/Xsession.d but
>there
>aren't any to do with graphics issues.

This change in descriptors seems to be a function of migrating to systemd while preserving LSB init functions.

>
>Any information I haven't provided that would help others help me?

The output of lshw gives an overview of your system topology.

>
>I can't find any guides on how to setup xrandr to use multiple gpus AND
>more than 2 or 3 monitors (usually from only 1 gpu).
>Most of the randr information I can find is for at newest randr 1.3 and
>I
>have 1.5. Randr 1.4 is where the multiple gpu option was introduced so
>it
>should be possible but . . . .
>
>I've found
>https://cgit.freedesktop.org/xorg/pr...oto-1.5.0#n160

Can't resolve this link, but at this time it is likely your current issue is with using Nouveau on the testing kernel.

Nvidia has a new beta driver which has dealt with some Dbus issues, among other things.

>likely everything I need is hidden in this file but for the life of me
>I
>can't figure out what commands to use. If I were an expert I wouldn't
>be
>asking but I'm not and there are no examples to follow (or try).
>
>Suggestions please?
>
>(Maybe do I have to go for newer proprietary drivers? (rather not I
>sort of
>like the no issues right now with Nouveau!)

Last month I installed Fedora alpha on a five year old system with GT440 Nvidia hardware for a friend. Nouveau works, but yesterday we had to install the driver from the vendors install script to connect kaffene to the TV tuner.

The following link to step by step instructions for Fedora might be helpfull if you are tentative. This site was pretty convoluted earlier but now changes to the runlevels, in order to blacklist nouveau, are given using systemctl.

If systemd is inevatible, the migration process is going to be painfull to older hardware.

https://www.if-not-true-then-false.com/2015/fedora-nvidia-guide/

Lxqt is a quicktime rewrite of LDXE. Maybe LDXE or another lightweight will handle what looks like framebuffering issues to me; at least in the first instance after switching kernels. Currently your issues seem to be with Nouveau.

>
>TIA

Hope this helps a bit.

>
>Dee

I'm more blunt force hacker and not guru, but the following info may help. 
-- 
Russell
Sent by K-9 Mail