GL in user namespace

scott.c.arnold · April 6, 2020, 4:52pm

Gl graphics display blank screen when ran in user namespace:

/bin/glxgears

scott.c.arnold · April 6, 2020, 4:59pm

Finishing post:

#/bin/glxgears // works as expected
#unshare -U /bin/glxgears // blank window

RH 7.6
Kernel 5.3.6
DMI: HP ProLiant DL580 Gen9/ProLiant DL580 Gen9, BIOS U17 01/22/2018

±----------------------------------------------------------------------------+
| Processes: GPU Memory |
| GPU PID Type Process name Usage |
|=============================================================================|
| 0 25483 G /usr/bin/Xorg 30MiB |
±----------------------------------------------------------------------------+
Any insight greatly appreciated
Thanks
Scott

scott.c.arnold · April 7, 2020, 4:15pm

More info:
No errors reported.
Reported frame rate is the same.
It seems GL thinks it is rendering the image, but screen is blank.
Machine has 72 cores (Intel Xenon E7-8867) and 512G ram.
This works correctly on a Ubuntu machine w/NVS315 card and 390 series drivers.

Thanks
Scott

scott.c.arnold · April 7, 2020, 10:08pm

Maybe these are part of problem:

[Tue Apr 7 17:07:25 2020] NVRM: Xid (PCI:0000:04:00): 13, pid=44855, Graphics Exception on GPC 0: 3D-C MEMLAYOUT Violation. Coordinates: (0x8, 0xc)
[Tue Apr 7 17:07:25 2020] NVRM: Xid (PCI:0000:04:00): 13, pid=44855, Graphics Exception: ESR 0x500420=0x80000400 0x500434=0xc0008 0x500438=0x2a00 0x50043c=0x10000
[Tue Apr 7 17:07:25 2020] NVRM: Xid (PCI:0000:04:00): 13, pid=44855, Graphics Exception on GPC 1: 3D-C MEMLAYOUT Violation. Coordinates: (0x3, 0xc)
[Tue Apr 7 17:07:25 2020] NVRM: Xid (PCI:0000:04:00): 13, pid=44855, Graphics Exception: ESR 0x508420=0x80000400 0x508434=0xc0003 0x508438=0x2a00 0x50843c=0x10000
[Tue Apr 7 17:07:25 2020] NVRM: Xid (PCI:0000:04:00): 13, pid=44855, Graphics Exception: ChID 0010, Class 0000c197, Offset 000023c8, Data bf800000

Only occur when unshared

scott.c.arnold · April 23, 2020, 9:55pm

Same problem with 390.132 drivers. Going to try nouveau, at least I have the source and can fix it if it doesn’t work.

amrits · April 24, 2020, 5:54am

Hi Scott,

I tried reproducing issue on my test setup where I observed that running command “unshare -U /bin/glxgears” leaves a window blank but fps decreases drastically.

My config - Precision T7610 + Ubuntu 18.04.4 + 5.5.6 + GeForce RTX 2080 + Driver 440.82

root@oemqa-Precision-T7610:~# unshare -U glxgears &
[3] 14823
root@oemqa-Precision-T7610:~# Running synchronized to the vertical refresh. The framerate should be
approximately the same as the monitor refresh rate.
11 frames in 5.2 seconds = 2.128 FPS
10 frames in 5.2 seconds = 1.925 FPS
10 frames in 5.2 seconds = 1.926 FPS
10 frames in 5.2 seconds = 1.925 FPS
10 frames in 5.2 seconds = 1.927 FPS
10 frames in 5.2 seconds = 1.924 FPS
10 frames in 5.2 seconds = 1.926 FPS

However in your case, fps is same in both cases, correct ?
Also can you please confirm , every-time when you run “unshare -U /bin/glxgears”, you hit with Xid 13 errors because i have not observed on my setup so far althiugh screen is blank.

amrits · April 24, 2020, 7:48am

I tried with below configuration setup again and looks like I am able to replicate issue.

HP Z8 G4 Workstation + Ubuntu 18.04.4 + 5.3.0-45-generic + Driver 440.59 + Quadro P6000

Repro Steps Attempted -

Started Bare X server
Run glxgears which executed successfully and observed application running on screen.
root@nghodake-HP-Z840-Workstation:~# Running synchronized to the vertical refresh. The framerate should be
approximately the same as the monitor refresh rate.
24/04/2020 12:51:20 copy_tiles: allocating first_line at size 121
24/04/2020 12:51:20 client 1 network rate 124.0 KB/sec (1996.4 eff KB/sec)
24/04/2020 12:51:20 client 1 latency: 39.8 ms
24/04/2020 12:51:20 dt1: 0.0128, dt2: 0.0700 dt3: 0.0398 bytes: 10259
24/04/2020 12:51:20 link_rate: LR_BROADBAND - 39 ms, 123 KB/s
296 frames in 5.0 seconds = 59.111 FPS
300 frames in 5.0 seconds = 59.990 FPS
300 frames in 5.0 seconds = 59.997 FPS
300 frames in 5.0 seconds = 59.998 FPS
300 frames in 5.0 seconds = 59.997 FPS
300 frames in 5.0 seconds = 59.996 FPS
Later I ran “unshare -U /bin/glxgears” which also executed successfully but screes is blank as reported by you.
root@nghodake-HP-Z840-Workstation:~# unshare -U glxgears
Running synchronized to the vertical refresh. The framerate should be
approximately the same as the monitor refresh rate.
2 frames in 5.1 seconds = 0.391 FPS
300 frames in 5.0 seconds = 59.967 FPS
300 frames in 5.0 seconds = 59.996 FPS
300 frames in 5.0 seconds = 59.998 FPS
300 frames in 5.0 seconds = 59.997 FPS
300 frames in 5.0 seconds = 59.996 FPS
300 frames in 5.0 seconds = 59.999 FPS
300 frames in 5.0 seconds = 59.997 FPS
300 frames in 5.0 seconds = 59.997 FPS
^C

However, I observed Xid 31 after quitting glxgears after step 3. Can you please confirm if you always get Xid 13 error or any other errors.

root@nghodake-HP-Z840-Workstation:~# dmesg |grep -i xid
[761903.782958] NVRM: Xid (PCI:0000:2d:00): 31, pid=89943, Ch 00000010, intr 10000000. MMU Fault: ENGINE GRAPHICS GPCCLIENT_PROP_0 faulted @ 0x1_04800000. Fault is of type FAULT_PDE ACCESS_TYPE_WRITE
[761955.674472] NVRM: Xid (PCI:0000:2d:00): 31, pid=90699, Ch 00000010, intr 10000000. MMU Fault: ENGINE GRAPHICS GPCCLIENT_PROP_0 faulted @ 0x1_04800000. Fault is of type FAULT_PDE ACCESS_TYPE_WRITE

scott.c.arnold · April 24, 2020, 3:34pm

Hello,

Errors do not seem to be logged every time but usually have some kind of error.

The following occurred with the 390.132 driver running unshare –U glxgears :

Apr 23 15:31:05 jstsdxmpcv01 kernel: NVRM: Xid (PCI:0000:04:00): 31, Ch 00000010, engmask 00000105, intr 10000000

Apr 23 15:31:16 jstsdxmpcv01 kernel: NVRM: Xid (PCI:0000:04:00): 13, Graphics Exception on GPC 0: 3D-C MEMLAYOUT Violation. Coordinates: (0xfb, 0x5c)

Apr 23 15:31:16 jstsdxmpcv01 kernel: NVRM: Xid (PCI:0000:04:00): 13, Graphics Exception: ESR 0x500420=0x80000400 0x500434=0x5c00fb 0x500438=0x2a00 0x50043c=0x310000

Apr 23 15:31:16 jstsdxmpcv01 kernel: NVRM: Xid (PCI:0000:04:00): 13, Graphics Exception on GPC 1: 3D-C MEMLAYOUT Violation. Coordinates: (0x100, 0x5c)

Apr 23 15:31:16 jstsdxmpcv01 kernel: NVRM: Xid (PCI:0000:04:00): 13, Graphics Exception: ESR 0x508420=0x80000400 0x508434=0x5c0100 0x508438=0x2a00 0x50843c=0x310000

Also sometimes the frame rate will be very low, maybe when logging errors ?

Thanks

Scott

scott.c.arnold · April 24, 2020, 4:12pm

Hello,

I also tested a P2000 card on Ubuntu 18.04.4 LTS + 4.15.0-96-generic + Driver 390.116 and it worked correctly, no errors logged.

I think the problem may be 5.x kernel related.

Thanks

Scott

scott.c.arnold · April 24, 2020, 11:29pm

Unshare -u glxgears works as expected with nouveau driver.

Thanks
Scott

amrits · April 26, 2020, 9:12am

Hi Scott,

Just to confirm issue reported is blank screen when we run GL in user namespace.
We will work on it and keep you update.
We have filed a bug 200611077 internally for tracking purpose.

ahuillet · May 25, 2020, 12:53pm

Hello, we have fixed the bug in our development branch. The fix will appear in a future release, and there should be a changelog entry covering it.
What trips the driver currently is that the /dev/nvidia-modeset is seen to have nobody:nobody ownership, and it expects root:root. Maybe that is enough for you to find a workaround until the fix lands.

scott.c.arnold · May 26, 2020, 3:49pm

Thanks we will test it out.

How do I get the development driver ?

Thanks

Scott

scott.c.arnold · May 26, 2020, 3:54pm

Hello,

On our machines the owner when unshared is nfsnobody:nfsnobody.

Will that break the fix ?

Thanks

Scott

ahuillet · May 26, 2020, 4:17pm

This is our internal branch, you do not get a build from it directly. If your company has an existing relationship with us and a NDA, we can give you pre-release build from a release branch.
No, the owner of /dev/nvidia* will not matter after the fix. The driver will try to open it no matter what, and also handle failure gracefully.

scott.c.arnold · May 26, 2020, 4:38pm

Hello,

Any idea when the updated driver will be available ? Need for scheduling.

Could I or a NASA official sign a NDA ?

Thanks

Scott

archenroot · October 3, 2021, 3:53pm

Hello @ahuillet, I am on Gentoo Linux where the file mentioned by you has following permissions (not root:root, but video as group):
[17:47:47] zangetsu@andromeda  $  ~  ls -la /dev/nvidia-modeset
crw-rw---- 1 root video 195, 254 Oct 3 11:02 /dev/nvidia-modeset

I actually didn’t experience this behavior with gtx 1030, but obtained quadro p4000 for triple monitor work and this started happening, the behavior is different:

sometimes its freezes completely my desktop
sometimes it restarts my desktop to complete fresh instanstance

I use KDE Plasma. I will report it in new fresh thread again for better tracking, but this thread found via google direct search, so will bit duplicate content.

ahuillet · October 4, 2021, 10:26am

I do not understand what you are referring to when you say “this behavior”.

Topic		Replies	Views
Refresh rates seemed to be capped at 40fps on laptop display, even after setting it to 165hz in settings Linux ubuntu	13	1738	May 5, 2021
Recovered GPU Errors in nvidia-settings Linux	10	19984	October 10, 2014
Ubuntu 19.04 Driver Installed but not Used Linux	102	16120	October 12, 2021
Xid 61 (black screen on startup) Ubuntu 18.04 GTX 1060 mobile Linux	12	3364	August 11, 2020
Laptop - GTX 1050 [driver 470.63.01] Manjaro linux-5.13.19-2 - X server black screen Linux driver	7	2268	November 14, 2021
Non-existent shared VRAM on NVIDIA Linux drivers Linux	54	9880	December 12, 2024
Reproducible: NVRM: GPU at 0000:01:00.0 has fallen off the bus. -- Both screens black, Xorg at 100% Linux	24	50921	December 16, 2015
[390.25] Random flicker in mid screen when deskop is mostly idle/empty. Linux	47	13108	August 20, 2019
Driver allocating memory over pci slot size Linux kernel	12	2578	February 16, 2021
Ubuntu 19.04 Drivers 440.26 Multiple Issues Linux	13	1267	January 29, 2020

GL in user namespace

/bin/glxgears

Related topics