Bug #15116

X.Org does not start with some NVidia Maxwell and Pascal graphic cards

Added by goupille 2017-12-27 13:48:48 . Updated 2019-11-25 12:14:26 .

Status:
Confirmed
Priority:
Normal
Assignee:
intrigeri
Category:
Hardware support
Target version:
Start date:
2017-12-27
Due date:
% Done:

10%

Feature Branch:
Type of work:
Communicate
Blueprint:

Starter:
Affected tool:
Deliverable for:

Description

Whoever is affected, please try https://nightly.tails.boum.org/build_Tails_ISO_bugfix-15116-newer-nouveau-xorg-driver/lastSuccessful/archive/build-artifacts/ and report back.

The original bug report was about GeForce GTX 970M.

Original report:

> a user reported that that Tails doesn’t start in normal mode on that computer (skylake inside : https://www.zotac.com/gb/product/mini_pcs/nen-steam-machine) with this error :
>
> ACPI Error: [\SB.PCI0.XHC_.RHUB.HS11] Namespace lookup failure, AE_NOT_FOUND (20170531/dswload-210)
> ACPI Exception: AE_NOT_FOUND, During name lookup/catalog (20170531/psobject-252)
> ACPI Exception: AE_NOT_FOUND, (SSDT:xh_rvp10) while loading table (20170531/tbxfload-228)
> ACPI Error: 1 table load failures, 7 successful (20170531/tbxfload-246)
>
> Ubuntu 16.04 and debian-live 9.3 are booting fine on this computer
>
> the USB stick was installed with Tails installer on ubuntu and works fine on other computers.


Subtasks


Related issues

Related to Tails - Bug #11831: NVIDIA Maxwell series graphic card: X.Org doesn't start, or slow graphics operations Resolved 2017-04-10
Related to Tails - Bug #12438: Document workarounds for NVIDIA Maxwell series graphics cards Resolved 2017-04-10
Related to Tails - Bug #14910: NVIDIA Pascal series graphic card: slow graphics operations Resolved 2017-10-31
Related to Tails - Bug #15833: Regression in 3.9~rc1 on some NVIDIA Pascal graphics cards Resolved 2018-08-22
Related to Tails - Bug #16764: Low resolution or no X.Org at all with NVidia NV160 (Turing) Resolved
Related to Tails - Bug #16714: Black screen with Nividia GeForce GTX 1080 (Pascal) if there are multiple monitors Confirmed
Related to Tails - Bug #16755: Call for testing: feature/buster (June 2019 edition) Resolved 2019-06-18
Blocked by Tails - Bug #15132: devel branch FTBFS since aufs-dkms 4.14 is in sid Resolved 2017-12-29
Blocked by Tails - Bug #15270: devel branch FTBFS since torbrowser-launcher 0.2.9 entered sid Resolved 2018-01-30
Blocked by Tails - Bug #15372: devel branch FTBFS since apparmor/2.11.0-3+deb9u2 reached stretch-proposed-updates Resolved 2018-03-03
Blocked by Tails - Bug #15472: Rebase our Tor Browser AppArmor policy on top of torbrowser-launcher 0.2.9-2's Resolved 2018-03-28
Blocked by Tails - Bug #15622: Upgrade to Linux 4.16+ in Tails 3.9 Resolved 2018-05-28

History

#1 Updated by goupille 2017-12-28 14:30:56

The graphics on this computer is a NVidia Maxwell (GeForce GTX 970M).

Tails can completely start with with a low resolution with nouveau.modeset=0 set at boot, but the errors message are still displayed.

Tails can’t start with nouveau.noaccel=1

#2 Updated by intrigeri 2018-01-02 10:33:00

  • Assignee changed from intrigeri to goupille
  • QA Check set to Info Needed

(received incomplete logs, see email)

#3 Updated by goupille 2018-01-02 13:55:55

  • Assignee changed from goupille to intrigeri

#4 Updated by intrigeri 2018-01-02 14:52:57

  • QA Check deleted (Info Needed)

#5 Updated by intrigeri 2018-01-02 14:54:23

  • Subject changed from ACPI errors on a Zotac NEN mini pc to Tails Greeter does not start on a Zotac NEN mini pc
  • Category set to Hardware support

(The ACPI errors also appear when Tails boots successfully, thus retitling.)

#6 Updated by intrigeri 2018-01-02 14:58:45

  • related to Bug #11831: NVIDIA Maxwell series graphic card: X.Org doesn't start, or slow graphics operations added

#7 Updated by intrigeri 2018-01-02 14:58:56

  • related to Bug #12438: Document workarounds for NVIDIA Maxwell series graphics cards added

#8 Updated by intrigeri 2018-01-02 15:00:13

  • Subject changed from Tails Greeter does not start on a Zotac NEN mini pc to X.Org does not start with NVidia Maxwell (GeForce GTX 970M)

The fact nouveau.modeset=0 makes X.Org start suggests this is a problem in the graphics stack (Linux, libdrm, X.Org).

#9 Updated by intrigeri 2018-01-02 15:03:08

  • Description updated

#10 Updated by intrigeri 2018-01-02 15:05:55

  • Description updated

#11 Updated by intrigeri 2018-01-02 15:06:14

  • Status changed from New to Confirmed
  • Feature Branch set to bugfix/15116-newer-nouveau-xorg-driver

#12 Updated by intrigeri 2018-01-02 15:06:26

  • blocked by Bug #15132: devel branch FTBFS since aufs-dkms 4.14 is in sid added

#13 Updated by intrigeri 2018-01-02 15:17:37

  • Status changed from Confirmed to In Progress
  • % Done changed from 0 to 10

Asked the OP to try an experimental ISO.

#14 Updated by intrigeri 2018-01-02 15:18:01

#15 Updated by intrigeri 2018-01-02 16:23:57

  • Target version set to Tails_3.6

If the newer nouveau driver helps, I’ll consider getting it into our next major release.

#16 Updated by goupille 2018-01-11 14:42:46

a user reported that the testing ISO fails to start without adding nouveau.modeset=0 with a NVIDIA GeForce 960M (the report was resent to you)

#17 Updated by intrigeri 2018-01-12 13:12:19

  • Subject changed from X.Org does not start with NVidia Maxwell (GeForce GTX 970M) to X.Org does not start with some NVidia Maxwell and Pascal graphic cards
  • Description updated

#18 Updated by intrigeri 2018-01-12 13:12:27

  • related to Bug #14910: NVIDIA Pascal series graphic card: slow graphics operations added

#19 Updated by intrigeri 2018-01-12 13:29:47

goupille wrote:
> a user reported that the testing ISO fails to start without adding nouveau.modeset=0 with a NVIDIA GeForce 960M (the report was resent to you)

Thanks. Let’s maintain a list of NVIDIA Maxwell and Pascal users who seem to be happy to try new stuff, so I can ask them to test whatever potential solution I find in the future. I see this time we have a contact address. Please add it to our shared repo (Bug #14511) once it’s ready :)

#20 Updated by goupille 2018-01-28 15:53:17

intrigeri wrote:
> Please add it to our shared repo (Bug #14511) once it’s ready :)

done :)

#21 Updated by goupille 2018-01-31 14:04:48

I asked a user (with a Pascal gpu) to try ubuntu 17.10.1 (4.14 kernel) and here are the results :

- in legacy BIOS, the boot process hangs just after the ubuntu boot menu.
- in UEFI, the bott process finishes, Gnome is started but the seesion is unstable and hangs when logging out or shutting down.

with the experimental ISO posted above or tails 3.5, in legacy or UEFI, Tails boot process ends with

          [  OK  ] Started GNOME Display Manager.

#22 Updated by intrigeri 2018-02-08 06:24:31

  • blocked by Bug #15270: devel branch FTBFS since torbrowser-launcher 0.2.9 entered sid added

#23 Updated by intrigeri 2018-02-13 11:59:48

  • Target version changed from Tails_3.6 to Tails_3.9

goupille wrote:
> I asked a user (with a Pascal gpu) to try ubuntu 17.10.1 (4.14 kernel) and here are the results :

Thanks. Even bleeding edge Ubuntu does not support this hardware properly and I don’t see how we could do better in Tails 3.6. Postponing to the following major release.

#24 Updated by intrigeri 2018-02-20 08:45:57

  • blocked by deleted (Feature #13245: Core work 2018Q1: Foundations Team)

#25 Updated by intrigeri 2018-02-20 08:46:24

#26 Updated by intrigeri 2018-03-10 08:08:24

#27 Updated by intrigeri 2018-03-10 08:08:34

#28 Updated by intrigeri 2018-03-10 08:08:47

  • blocked by Bug #15372: devel branch FTBFS since apparmor/2.11.0-3+deb9u2 reached stretch-proposed-updates added

#29 Updated by intrigeri 2018-03-22 07:18:28

  • blocked by deleted (Feature #15139: Core work 2018Q2: Foundations Team)

#30 Updated by intrigeri 2018-03-22 07:18:49

#31 Updated by emmapeel 2018-05-25 09:41:20

The iso referenced on the description is not there anymore.

Does it mean it has been merged into Tails?

What should users with an NVIDIA Maxwell card should test? The user still needs to add
nouveau.modeset=0 to start.

It would be nice to refresh the description, as this ticket is referenced from the t.b.o/gdm page as needing tests: https://tails.boum.org/gdm/#index5h2

#32 Updated by intrigeri 2018-05-25 09:56:51

> The iso referenced on the description is not there anymore.

(There was not activity on the branch since a while so our CI system stopped building it. I’ve refreshed the branch so it builds again; although it might fail to build.)

> Does it mean it has been merged into Tails?

No.

> What should users with an NVIDIA Maxwell card should test?

The ISO from this branch once it builds again, but I’m not hopeful: see the most recent discussion on this ticket.

#33 Updated by intrigeri 2018-06-05 11:21:04

Note to myself: something I could try is installing Mesa from stretch-backports on the topic branch. It might help in some cases.

#34 Updated by intrigeri 2018-06-05 11:24:55

intrigeri wrote:
> Note to myself: something I could try is installing Mesa from stretch-backports on the topic branch. It might help in some cases.

Done, let’s see if it even builds.

#35 Updated by intrigeri 2018-06-06 05:55:53

OK, so the topic branch now has (on top of xserver-xorg-video-nouveau from Buster, which it already had): Mesa + libdrm + friends from stretch-backports; Linux 4.16. There are good chances it improves things on some graphics hardware. It might even improve things a bit on some NVIDIA gear. I’ll quickly test this on the bare metal computers I have around and then will send a call for testing (including the list of “NVIDIA testers” gathered by help desk).

Dear help desk: please keep maintaining your list of NVIDIA testers and add to it whenever you can! It’ll be very useful here.

#36 Updated by intrigeri 2018-06-06 07:19:32

Works smoothly on ThinPad X200 and HP EliteBook 840 G1.

#37 Updated by intrigeri 2018-06-15 08:38:17

Note to myself: once https://release.debian.org/transitions/html/xserver1.20.html is over we won’t be able to install xserver-xorg-video-nouveau from testing anymore and the branch will FTBFS. Once this happens, if we still need to gather feedback about this, I’ll upload a backport to our custom APT repo. Also, then I’ll create a new branch that updates the rest of the graphics stack (mesa, drm, etc.) to stretch-backports but installs the nouveau driver from Stretch: this will help us know where the main improvements come from, and thus decide whether we want to bother backporting the nouveau driver.

#38 Updated by intrigeri 2018-06-15 08:50:02

Sent call for testing.

#39 Updated by intrigeri 2018-06-27 08:44:51

  • blocked by Bug #15472: Rebase our Tor Browser AppArmor policy on top of torbrowser-launcher 0.2.9-2's added

#40 Updated by intrigeri 2018-06-27 08:45:01

  • blocked by Bug #15622: Upgrade to Linux 4.16+ in Tails 3.9 added

#42 Updated by 110100111011 2018-07-10 01:39:26

I’ve just tested this ISO:

tails-amd64-bugfix_15116-newer-nouveau-xorg-driver-3.9-20180709T0442Z-29bb31bfe8+devel@4625f031b1.iso

It works way better than Tails 3.8 or previous versions!

I’m using a GTX 1080, and earlier versions would have very slow graphics, I talked about that here a while ago already: https://labs.riseup.net/code/issues/14910

With this new version, those issues seem to be fixed, and it feels like the system finally uses the GPU to accelerate graphics. Theres also significantly less tearing, though it still feels as if VSync is missing. But way better than before.

I have not tested the modeset and noaccel commands, since as far as I remember those only made things worse. Since it seems to be quite perfect now without any extra commands, it’s surely not needed to test those! If you still need anything else tested, let me know here.

#43 Updated by intrigeri 2018-07-10 08:55:30

> I’ve just tested this ISO:
> tails-amd64-bugfix_15116-newer-nouveau-xorg-driver-3.9-20180709T0442Z-29bb31bfe8+devel@4625f031b1.iso
> It works way better than Tails 3.8 or previous versions!
> I’m using a GTX 1080, and earlier versions would have very slow graphics, […]

Excellent news, thanks for reporting! This might be just the report I needed to convince myself that it was worth applying the changes from bugfix/15116-newer-nouveau-xorg-driver in Tails 3.9. I’ll come back to it early August and will make my final decision for 3.9 then.

#44 Updated by intrigeri 2018-08-01 09:11:50

intrigeri wrote:
> Note to myself: once https://release.debian.org/transitions/html/xserver1.20.html is over we won’t be able to install xserver-xorg-video-nouveau from testing anymore and the branch will FTBFS.

This is now the case.

> Once this happens, if we still need to gather feedback about this, I’ll upload a backport to our custom APT repo.

I’ll first check the feedback I’ve received since my last summary.

> Also, then I’ll create a new branch that updates the rest of the graphics stack (mesa, drm, etc.) to stretch-backports but installs the nouveau driver from Stretch: this will help us know where the main improvements come from, and thus decide whether we want to bother backporting the nouveau driver.

Done: bugfix/15116-update-graphics-stack, which I’ll probably propose for Tails 3.9 if I don’t have time to prepare a backport of the nouveau driver.

#45 Updated by intrigeri 2018-08-01 09:12:14

  • Feature Branch changed from bugfix/15116-newer-nouveau-xorg-driver to bugfix/15116-update-graphics-stack, bugfix/15116-newer-nouveau-xorg-driver

#46 Updated by intrigeri 2018-08-01 10:17:02

Aggregated feedback of the call for testing: https://mailman.boum.org/pipermail/tails-testers/2018-August/001064.html

#47 Updated by intrigeri 2018-08-03 02:21:02

Uploaded a backport so we have now:

  • bugfix/15116-update-graphics-stack: updates libdrm and mesa
  • bugfix/15116-newer-nouveau-xorg-driver: updates libdrm, mesa and the nouveau driver

#48 Updated by intrigeri 2018-08-04 00:48:25

  • Target version changed from Tails_3.9 to Tails_3.11
  • Feature Branch deleted (bugfix/15116-update-graphics-stack, bugfix/15116-newer-nouveau-xorg-driver)

I’ve proposed to merge bugfix/15116-newer-nouveau-xorg-driver on Bug #14910. I’m using Bug #14910 since the only significant improvement that was reported on this branch is about the exact problem that other ticket, and not about “X.Org does not start”. I don’t know what else I can do here right now so I’ll come back to it later.

#49 Updated by intrigeri 2018-08-04 00:48:51

  • blocked by deleted (Feature #15334: Core work 2018Q3: Foundations Team)

#50 Updated by intrigeri 2018-08-29 06:48:32

  • related to Bug #15833: Regression in 3.9~rc1 on some NVIDIA Pascal graphics cards added

#51 Updated by intrigeri 2018-11-05 14:53:28

  • Target version changed from Tails_3.11 to Tails_3.12

FTR I’m leaving this ticket open only because our doc refers to it and it’s a good place to collect info & workarounds before we document them. I see no way for us to fix that ourselves currently, but if time allows I’ll give give the nouveau driver from stretch-backports another try.

#52 Updated by intrigeri 2019-01-02 05:05:25

  • Target version changed from Tails_3.12 to Tails_4.0

> if time allows I’ll give give the nouveau driver from stretch-backports another try.

At this point I’d rather focus on Tails/Buster but I’ll document workarounds that fix such problems for some users (Bug #15774).

#53 Updated by intrigeri 2019-01-05 11:37:54

  • Type of work changed from Research to Communicate

Next step: specifically ask for reports on this front when we publish 4.0~beta1 and call for testing it.

#54 Updated by intrigeri 2019-04-02 15:23:31

  • Target version changed from Tails_4.0 to Tails_3.15

#55 Updated by intrigeri 2019-04-13 08:31:48

  • Status changed from In Progress to Confirmed

#56 Updated by goupille 2019-04-16 16:41:17

Bug report: 8f2dfa36b23e6fef7946731f89f58d22

#57 Updated by intrigeri 2019-05-02 15:09:46

nouveau.modeset=0 is still needed on “NVIDIA Corporation GM107M [GeForce GTX 960M] [10de:139b] (rev a2)” (756051d8ef6e54df5dba53e5abd40c5d).

#58 Updated by intrigeri 2019-05-02 16:05:16

The documented workarounds don’t work on GTX 1080 (049d3e91-978d-290b-a6b8-333eb2a70cbf@riseup.net).

#59 Updated by intrigeri 2019-06-01 10:12:17

  • related to Bug #16764: Low resolution or no X.Org at all with NVidia NV160 (Turing) added

#60 Updated by intrigeri 2019-06-01 10:34:24

  • related to Bug #16714: Black screen with Nividia GeForce GTX 1080 (Pascal) if there are multiple monitors added

#61 Updated by intrigeri 2019-06-01 12:09:46

  • related to Bug #16755: Call for testing: feature/buster (June 2019 edition) added

#62 Updated by intrigeri 2019-06-01 12:10:49

  • Target version deleted (Tails_3.15)

intrigeri wrote:
> Next step: specifically ask for reports on this front when we publish 4.0~beta1 and call for testing it.

I’m tracking this on Bug #16755.

#63 Updated by goupille 2019-11-14 14:30:14

there are still reports about that issue, last one was a user with a GTX 1060, Tails 4.0 shutdown when clicking on the ‘start tails’ button in the Greeter. it seems that it was fixed by adding nouveau.noaccel=1 to the boot parameters.

https://tails.boum.org/gdm still ask users for test results on that ticket, is it still the case ? if yes, could you specify what we’d need to ask the users ?

#64 Updated by intrigeri 2019-11-22 11:39:53

Hi @goupille,

> there are still reports about that issue

This is useful info, thanks!

> last one was a user with a GTX 1060, Tails 4.0 shutdown when clicking on the ‘start tails’ button in the Greeter.

Note to myself: technically, this is a different problem as the Greeter appeared so we’re not in a case when “X.Org does not start”. But I think it makes sense to handle it along with the rest of similar problems, at least for now.

> it seems that it was fixed by adding nouveau.noaccel=1 to the boot parameters.

Good, since this is a documented one for GTX 1060 :)

> https://tails.boum.org/gdm still ask users for test results on that ticket, is it still the case ? if yes, could you specify what we’d need to ask the users ?

At the moment, the most relevant cases when Desk can provide info here are:

  • Whenever a graphics card listed on https://tails.boum.org/support/known_issues/graphics/ starts working out-of-the-box.
    • If we’re lucky enough that a user reports this voluntarily and explicitly (e.g. they’re happy after an upgrade that fixes stuff for them), great: we can clean up that web page a bit. IMO it’s worth reporting.
    • Failing that, somebody (sajolida, or help desk, or myself) could search WhisperBack reports from time to time, looking for successful boots with one of these graphics cards and with no workaround applied. I’m not sure it’s worth the effort. Maybe every 2 years, but not more often?
  • Whenever a workaround is reported to work for a graphics card that’s not listed yet on that web page: then it’s very useful to tell FT about it, so we can document the workaround and the next affected user benefits from it.

But I don’t think it’s particularly useful currently to report that a known workaround is still needed for a graphics card that’s listed already.
This could change once we can do something that’s specifically meant to improve the situation, e.g. upgrading the nouveau X.Org driver (but no such upgrade is available currently).

#65 Updated by goupille 2019-11-25 12:14:26

intrigeri wrote:

> > it seems that it was fixed by adding nouveau.noaccel=1 to the boot parameters.
>
> Good, since this is a documented one for GTX 1060 :)

fun fact -> the user says that once the Persistence was set up, nouveau.noaccel=1 didn’t work anymore to start tails. however the other documented boo parameter (nouveau.modeset=0) does…

>
> > https://tails.boum.org/gdm still ask users for test results on that ticket, is it still the case ? if yes, could you specify what we’d need to ask the users ?
>
> At the moment, the most relevant cases when Desk can provide info here are:
>
> * Whenever a graphics card listed on https://tails.boum.org/support/known_issues/graphics/ starts working out-of-the-box.
> If we’re lucky enough that a user reports this voluntarily and explicitly (e.g. they’re happy after an upgrade that fixes stuff for them), great: we can clean up that web page a bit. IMO it’s worth reporting.
> Failing that, somebody (sajolida, or help desk, or myself) could search WhisperBack reports from time to time, looking for successful boots with one of these graphics cards and with no workaround applied. I’m not sure it’s worth the effort. Maybe every 2 years, but not more often?
> * Whenever a workaround is reported to work for a graphics card that’s not listed yet on that web page: then it’s very useful to tell FT about it, so we can document the workaround and the next affected user benefits from it.
>
> But I don’t think it’s particularly useful currently to report that a known workaround is still needed for a graphics card that’s listed already.
> This could change once we can do something that’s specifically meant to improve the situation, e.g. upgrading the nouveau X.Org driver (but no such upgrade is available currently).

ok, thanks !