nvidia 384.59 issues [SOLVED!]

thelongdivider · 2017-08-05 12:45:24

I've been needing to revert back to the 381 series of drivers because on the 384 series my card pins at 100% usage and decode usage while watching video. Has anyone else had this issue?

Edit: I have a gtx 1070.

Last edited by thelongdivider (2017-09-14 14:39:25)

seth · 2017-08-05 14:00:30

Yup. MPEG2 so far only, though. To the degree of a kernel halt.
https://bugs.archlinux.org/task/55038

thelongdivider · 2017-08-05 14:45:39

seth wrote:

Yup. MPEG2 so far only, though. To the degree of a kernel halt.
https://bugs.archlinux.org/task/55038

Indeed. The entire system locks up when opening any video in mpv/chrome.

thelongdivider · 2017-08-07 11:54:52

The issue persists in nvidia-384.59-3.

seth · 2017-08-07 12:28:05

A rebuild is never gonna fix this.

thelongdivider · 2017-08-12 21:13:25

I don't know if this is actually the solution, but I can at least manage to not lock up the system by using nvidia-modprobe on every boot. I do it both as user and as root every time I boot.

Last edited by thelongdivider (2017-08-12 21:13:46)

seth · 2017-08-12 21:39:17

Edit: Worked once
Might have been some über-lucky conincidence?
Every subsequent time didn't prevent the kernel halt.

I stracegrepped it for "open" calls, but there's nothing interesing. Going to check whether the UID0 call is required and also dump full straces with both driver versions.
Edit: the straces barely differ, nvidia-modprobe is suid, so running it as root should not be required?

Last edited by seth (2017-08-12 22:10:58)

thelongdivider · 2017-08-18 14:47:01

seth wrote:

Edit: Worked once
Might have been some über-lucky conincidence?
Every subsequent time didn't prevent the kernel halt.
I stracegrepped it for "open" calls, but there's nothing interesing. Going to check whether the UID0 call is required and also dump full straces with both driver versions.
Edit: the straces barely differ, nvidia-modprobe is suid, so running it as root should not be required?

For me it seemed to work sporadically as well. I unfortunately need to be on 384 for now because I don't have the lib32-nvidia-utils-381 package. Hopefully this issue is fixed by nvidia soon.

TheSgtBilko · 2017-08-18 16:03:46

https://wiki.archlinux.org/index.php/Arch_Linux_Archive
https://archive.archlinux.org/packages/ … dia-utils/

Last edited by TheSgtBilko (2017-08-18 16:09:43)

thelongdivider · 2017-08-18 16:32:21

TheSgtBilko wrote:

https://wiki.archlinux.org/index.php/Arch_Linux_Archive
https://archive.archlinux.org/packages/ … dia-utils/

Awesome. Thank you for the link. I will return to the 381 series until a further version of nvidia-384 gets released.

thelongdivider · 2017-08-23 12:51:51

The issue persists in 384.69.

The error being reported is this:

NVRM: Xid (PCI:0000:09:00): 31, Ch 00000020, engmask 00008100, intr 100000000

Last edited by thelongdivider (2017-08-26 20:01:01)

thelongdivider · 2017-09-14 14:39:02

AT LAST!

I found a solution to my problem. Nvidia lists this as an mmu problem. I was aware that ryzen had some iommu issues to begin with, but it was working fine on earlier nvidia versions so I never had any problems.

In my Gigabyte-x370 gaming k7 UEFI, there is an option called IOMMU. I switched this option from "auto" to "enabled". This fixes this issue and I can now enable hardware decoding on Nvidia-384.69!

I hope this helps someone else in the future.

seth · 2017-09-30 20:14:12

Afaics this problem is gone entirely with 384.90

thelongdivider · 2017-10-05 12:43:55

seth wrote:

Afaics this problem is gone entirely with 384.90

This issue actually returned for me on 384.90. I'm on x370 and I am getting this xid 31 error on my gtx 1070. What is your hardware like?

It also occurs in ffmpeg decoding directly, so it's not only mpv.

seth · 2017-10-05 13:04:39

gt 520 - and still i686 (I'm waiting till the last day because otherwise, where's the thrill? ;-)

thelongdivider · 2017-10-05 13:13:39

seth wrote:

gt 520 - and still i686 (I'm waiting till the last day because otherwise, where's the thrill? ;-)

So it effects a lot of platforms and generations of cards. That's good to know.

I installed 381.22 again, but the dkms variant to avoid conflicts. I'm going to hold here for awhile before returning to the current branch.

Arch Linux

#1 2017-08-05 12:45:24

nvidia 384.59 issues [SOLVED!]

#2 2017-08-05 14:00:30

Re: nvidia 384.59 issues [SOLVED!]

#3 2017-08-05 14:45:39

Re: nvidia 384.59 issues [SOLVED!]

#4 2017-08-07 11:54:52

Re: nvidia 384.59 issues [SOLVED!]

#5 2017-08-07 12:28:05

Re: nvidia 384.59 issues [SOLVED!]

#6 2017-08-12 21:13:25

Re: nvidia 384.59 issues [SOLVED!]

#7 2017-08-12 21:39:17

Re: nvidia 384.59 issues [SOLVED!]

#8 2017-08-18 14:47:01

Re: nvidia 384.59 issues [SOLVED!]

#9 2017-08-18 16:03:46

Re: nvidia 384.59 issues [SOLVED!]

#10 2017-08-18 16:32:21

Re: nvidia 384.59 issues [SOLVED!]

#11 2017-08-23 12:51:51

Re: nvidia 384.59 issues [SOLVED!]

#12 2017-09-14 14:39:02

Re: nvidia 384.59 issues [SOLVED!]

#13 2017-09-30 20:14:12

Re: nvidia 384.59 issues [SOLVED!]

#14 2017-10-05 12:43:55

Re: nvidia 384.59 issues [SOLVED!]

#15 2017-10-05 13:04:39

Re: nvidia 384.59 issues [SOLVED!]

#16 2017-10-05 13:13:39

Re: nvidia 384.59 issues [SOLVED!]

Board footer