Project

General

Profile

Bug #16447

Gather information about regression on some Intel GPU (Braswell, Kaby Lake)

Added by mercedes508 8 months ago. Updated 8 days ago.

Status:
In Progress
Priority:
Normal
Assignee:
Category:
Hardware support
Target version:
Start date:
02/08/2019
Due date:
% Done:

0%

Feature Branch:
Type of work:
Research
Blueprint:
Starter:
Affected tool:

Description

Well, since Tails 3.12, many users have been complaining that they get the "Error starting GDM" issue, and finally I got answers confirming that there hardware were working fine with Tails 3.11, so that's a regression.

Here is the list of affected GPU that were working fine with 3.11:

  • Intel Corporation Atom/Celeron/Pentium Precessor x5-E8000/J3xxx/N3xxx Integrated Graphics Controller [8086:22b1] (rev 35) - (2 reports)
  • Intel Corporation 82915G/GV/940GL: integrated Graphics controller [8086:2582] (rev 04)

Here is the list of potential regression (not sure about 3.11 behavior):

  • Integrated Graphics Controller [8086:0a16] (rev0b)
  • Intel Corporation HD Graphics 630 [8086:591b] (rev 04) - (4 reports) → GDM fails to start in troubleshooting mode on 3.13.2; with troubleshooting mode and xorg-driver=modesetting, it fails to start because /dev/dri/card0 does not exist (expected since troubleshooting mode sets nomodeset; note that the nouveau driver is also loaded
  • Intel Corporation UHD Graphics 620 [8086:5917] (rev 07) → GDM fails to start in troubleshooting mode on 3.14.2 (3.14 booted fine); fixed by passing either one of xorg-driver=intel or xorg-driver=modesetting for one user; contradictory report that says xorg-driver=modesetting is not sufficient on the same hardware

And from the few users who answered back, none of the following workaround works:

https://tails.boum.org/support/known_issues/graphics/#index3h2

CancelText.png View - Cancel button text (93.9 KB) dngray, 07/01/2019 05:56 PM


Related issues

Related to Tails - Bug #16224: Black screen after the boot menu with Intel GPU (i915) Resolved 12/13/2018
Related to Tails - Bug #16552: Upgrade to Linux 4.19.28 Resolved 03/13/2019

Associated revisions

Revision 5baca24d (diff)
Added by intrigeri 8 months ago

Document another possible workaround (refs: #16447).

For cards listed in /usr/share/live/config/xserver-xorg/intel.ids,
that is, by default we would force the "intel" X.Org driver,
when things break it might be worth trying to revert to the defaults,
i.e. the "modesetting" X.Org driver.

History

#1 Updated by mercedes508 8 months ago

  • Related to Bug #16224: Black screen after the boot menu with Intel GPU (i915) added

#2 Updated by mercedes508 8 months ago

  • Description updated (diff)

#3 Updated by intrigeri 8 months ago

  • Status changed from Confirmed to In Progress

#4 Updated by intrigeri 8 months ago

  • Assignee changed from intrigeri to mercedes508
  • QA Check set to Info Needed

Hi,

Here is the list of affected GPU that were working fine with 3.11:

  • Intel Corporation Atom/Celeron/Pentium Precessor x5-E8000/J3xxx/N3xxx Integrated Graphics Controller [8086:22b1] (rev 35) - (2 reports)
  • Intel UHD 620 [8086:5917] rev07

Here is the list of potential regression (not sure about 3.11 behavior):

  • Integrated Graphics Controller [8086:0a16] (rev0b)
  • Intel Corporation 82915G/GV/940GL:tegrated Graphics controller [8086:2582] (rev 04)
  • Intel Corporation HD Graphics 630 [8086:591b] (rev 04) - (2 reports)

Thanks a lot for the comprehensive report. This is excellent aggregated data i.e. exactly

All of those, except 8086:0a16 (unclear if that one is a regression at this point), are listed in /usr/share/live/config/xserver-xorg/intel.ids; that is, we force the intel X.Org driver for them. It would be worth testing how they behave with the default driver (modesetting), so I've added this workaround to https://tails.boum.org/support/known_issues/graphics/#intel ⇒ please try affected users to test with xorg-driver=modesetting.

Also, I see that Linux 4.19.16 has a fix for some gen7 GPU issue, so please ask the affected users to try a build from the devel branch (better wait for today's build, which should land there in ca. 1h, as previous builds were using outdated APT snapshots). This could be relevant in particular for the UHD 620 and HD Graphics 630. I need at least the results without any workaround (bonus points if there are results with the 5 documented workarounds too).

Finally, it would be sweet if one of these users was technical enough to add rootpw=sillypassword, switch to a text console, log is as root with "sillypassword", extract the output of journalctl, and send it to me, so I have more info than "X.Org won't start" :) But forget it if they need guidance to do that.

Thanks in advance!

FTR here are the corresponding Intel generations:

  • recent hardware:
    • Braswell (2017-2018): x5-E8000/J3xxx/N3xxx
    • 7th gen Kaby Lake (2017): UHD 620, HD Graphics 630
  • 10+ years old: 82915G/GV/940GL (could be Pentium 4, Celeron D, from 2004, or Core/Core 2 mobile from 2006)

I could not spot any related recent bug filed against xserver-xorg-video-intel. But we did not update it so the problem could be tracked elsewhere, e.g. Mesa, libglvnd, or Linux.

#5 Updated by intrigeri 8 months ago

  • Subject changed from Regression on some Intel GPU to Regression on some Intel GPU (Braswell, Kaby Lake)

#6 Updated by mercedes508 8 months ago

For Intel Corporation Atom/Celeron/Pentium Precessor x5-E8000/J3xxx/N3xxx Integrated Graphics Controller [8086:22b1] (rev 35)

Neither "xorg-driver=modesetting" nor using Tails from devel branch works.

#7 Updated by intrigeri 7 months ago

@mercedes508, @emmapeel , @goupille: we have less than one week left to apply any fix in time for 3.13. What's the status here? Is there any chance we get reports from these users? It's been a month so perhaps it's worth pinging them, in case you did not do it yet?

(I see you've added this to the "Hot topics on our help desk" which made me remember I was waiting for input here :)

#8 Updated by goupille 7 months ago

I asked the users affected if they would be able to get the logs, with no answers yet. one user with a Intel Corporation UHD Graphics 620 [8086:5917] (rev 07) repoted that :

- in normal mode, after the greeter, the Tails desktop appears transiently for a few seconds then the screen goes black.

- "nomodeset", "nomodeset xorg-driver=vesa", "xorg-driver=intel", "xorg-driver=modesetting", or the Troubleshooting mode are giving the "Error starting GDM" screen

#9 Updated by mercedes508 7 months ago

I'm pinging right now the few users who didn't answer already.

#10 Updated by intrigeri 7 months ago

I'm pinging right now the few users who didn't answer already.

Thanks :)

#11 Updated by goupille 7 months ago

  • Assignee changed from mercedes508 to intrigeri

I forwarded you the email of a user that seems able to get those logs.

#12 Updated by intrigeri 7 months ago

  • Intel UHD 620 [8086:5917] rev07

@goupille, @mercedes508: I'm wondering if there's some confusion here. According to the report I've seen from one user of this hardware, the Greeter appears just fine and they don't see "Error starting GDM" (as long as they don't break stuff by adding boot options to debug the problem); instead, the Greeter appears but the screen goes black after login, once the desktop has been briefly displayed.

So could you please check which one(s), on the list of affected GPUs in the ticket description, are actually instances of the regression this ticket is about, and which ones are about the different issue that this user is facing?

#13 Updated by intrigeri 7 months ago

  • Assignee changed from intrigeri to mercedes508

#14 Updated by goupille 7 months ago

the only user who gave a substancial description of their problem (ans answered back) is the one with the Intel UHD 620 [8086:5917] rev07 (same user I forwareded you the email). I assumed that it was an expression of this issue since this exact model is listed above.

#15 Updated by intrigeri 7 months ago

  • Description updated (diff)

goupille wrote:

the only user who gave a substancial description of their problem (ans answered back) is the one with the Intel UHD 620 [8086:5917] rev07 (same user I forwareded you the email). I assumed that it was an expression of this issue since this exact model is listed above.

OK, thanks. I've removed that GPU from the list and my question still stands for the others.

#16 Updated by intrigeri 7 months ago

  • Related to Bug #16552: Upgrade to Linux 4.19.28 added

#17 Updated by intrigeri 7 months ago

@goupille, @mercedes508, @emmapeel, please point users affected by one of the two bugs this ticket is about to: https://lists.autistici.org/message/20190314.121121.001cf4cb.en.html

#18 Updated by CyrilBrulebois 7 months ago

  • Target version changed from Tails_3.13 to Tails_3.14

#19 Updated by fedionso 7 months ago

Hi guys I'm getting this error on 3.13.1 and want to try the latest 3.14 drop from
https://nightly.tails.boum.org/build_Tails_ISO_stable/lastSuccessful/archive/build-artifacts/

Problem: the key used to sign the tails-build-artifacts.shasum (0xD83A438B2F916605) is not listed anywhere on https://tails.boum.org/doc/about/openpgp_keys/index.en.html

I found the following key details at https://pgp.surfnet.nl/pks/lookup?search=0xD83A438B2F916605&fingerprint=on&op=index

pub 4096R/2F916605 2013-09-23 jenkins.tails.boum.org artifact signing key
Fingerprint=AA65 3423 DBE1 0866 0B18 3B17 D83A 438B 2F91 6605

Is this the correct key?

#20 Updated by intrigeri 7 months ago

Is this the correct key?

The one I have is "AA65 3423 DBE1 0866 0B18 3B17 D83A 438B 2F91 6605".
Still, note that these are experimental, unsupported images. Only use them for testing.

#21 Updated by mercedes508 5 months ago

CAFoG=RrV0LSPOF33kJT4=
Intel Corporation HD Graphics 630 [8086:591b] (rev 04)

#22 Updated by CyrilBrulebois 5 months ago

  • Target version changed from Tails_3.14 to Tails_3.15

#23 Updated by mercedes508 5 months ago


Intel Corporation HD Graphics 630 [8086:591b] (rev 04)

#24 Updated by mercedes508 5 months ago

  • Description updated (diff)

#25 Updated by mercedes508 5 months ago

  • Assignee changed from mercedes508 to intrigeri

intri: Do you want me to fwd you the 2 last reports which are not WB reports?

#26 Updated by mercedes508 5 months ago

CALqX=
Intel Corporation HD Graphics 630 [8086:591b] (rev 04)

#27 Updated by goupille 5 months ago

Intel Corporation Atom/Celeron/Pentium Processor x5-E8000/J3xxx/N3xxx
Integrated Graphics Controller [8086:22b1] (rev 35)

I already forwarded the outputs of journalctl

#28 Updated by intrigeri 5 months ago

  • Assignee changed from intrigeri to mercedes508

intri: Do you want me to fwd you the 2 last reports which are not WB reports?

@mercedes508: yes, please. I don't receive those ones.

#29 Updated by intrigeri 5 months ago

It would be sweet to know which workaround (https://tails.boum.org/support/known_issues/graphics/#intel) was tested on which hardware and what the results are. In particular, I'd love to see the Journal for xorg-driver=modesetting on one of the affected cards.

#30 Updated by intrigeri 5 months ago

  • Subject changed from Regression on some Intel GPU (Braswell, Kaby Lake) to Gather information about regression on some Intel GPU (Braswell, Kaby Lake)
  • QA Check deleted (Info Needed)

#31 Updated by mercedes508 5 months ago

intrigeri wrote:

It would be sweet to know which workaround (https://tails.boum.org/support/known_issues/graphics/#intel) was tested on which hardware and what the results are. In particular, I'd love to see the Journal for xorg-driver=modesetting on one of the affected cards.

I forwarded you the reports, will you ask the journal or shall I? Just to avoid duplicate demands.

#32 Updated by mercedes508 5 months ago

  • Assignee changed from mercedes508 to intrigeri

#33 Updated by intrigeri 4 months ago

will you ask the journal or shall I? Just to avoid duplicate demands.

Please gather this info and reassign once you've made it available to me. Thanks! :)

#34 Updated by intrigeri 4 months ago

  • Assignee changed from intrigeri to mercedes508

@mercedes508, see above.

#35 Updated by mercedes508 4 months ago

intrigeri wrote:

@mercedes508, see above.

OK, just asked, let's wait for answers now.

#36 Updated by goupille 4 months ago

Message-ID: <c9b9fa06-e5ce-98ed-0503-941b954f7d46@

#37 Updated by dngray 4 months ago

I'm not sure if this is the right bug, if someone could please advise what to do I will do the research. Hopefully this can be fixed before 3.15 as Tails no longer works for me at all and I cannot stay with vulnerable versions 3.14.0, 3.14.1.

  • I have a Dell XPS 13" 9370 (i7-8550U) with a Kaby Lake processor.
  • I have found that 3.14.2 will not boot for me, whereas 3.14.0 did.
  • I can create an intermediary usb stick with both ie: dd if=tails-amd64-3.14.2.img of=/dev/device and it will work with both versions.

However when booting 3.14.2 after I have used the Tails installer to clone the current 3.14.2 install from the intermediary usb stick I get:

(initramfs) aufs not available

If I boot from the troubleshooting mode I get:

Error starting GDM with your graphics card: Intel Corporation UHD Graphics 620 [8086:5917] (rev 07). Please take note of this error and visit https://tails.boum.org/gdm for troubleshooting.

I tried the Intel startup options however they did not allow the system to boot.

#38 Updated by dngray 4 months ago

Also, something I have observed since 3.13 (worked fine in 3.12.x) was the text on certain buttons gets struck out when highlighting it. Typically it's the kind in this screenshot. I was able to reproduce this in Nautilus and the screenshot app in Gnome.

I haven't had these issues in Archlinux and have run all the kernels from 4.16 to 5.1.15. In Archlinux I do use xf86-video-intel as I got terrible performance when using modesetting

I use i3wm, and compton. I also use TearFree there.

#39 Updated by dngray 4 months ago

  1. (initramfs) aufs not available

    I'm not quite sure what was causing this. I was able to reproduce it with two USB sticks yesterday, but not today.

    Today I booted my machine in BIOS mode and the problem didn't occur, so then I tried UEFI mode with secure boot disabled, no issue. Also no issue with secure boot enabled (as long as I signed BOOTX64.EFI, BOOTIA32.EFI and vmlinuz, ie (sudo sbsign --key db.key --cert db.crt --output {{file to sign}} {{file to sign}}, not doing the signing causes a different very explicit error on my machine anyway.

  2. Error starting GDM with your graphics card: Intel Corporation UHD Graphics 620 [8086:5917] (rev 07). Please take note of this error and visit https://tails.boum.org/gdm for troubleshooting. This was being caused because I was not deleting nomodesetting from the kernel options when booting troubleshooting mode.. Oops.
  3. The issue in #38 seemed to go away as soon as I chose xorg-driver=intel or xorg-driver=modesetting. I can only reproduce it when I don't set either of those options. Both give me 300fps, though I think the intel driver is smoother than the modesetting one.

#40 Updated by dngray 3 months ago

It would seem (initramfs) aufs not available is caused by unclean dismounts.

I solved this by formatting the Tails partition (that's the unencrypted FAT32 partition with the boot, hidden, legacy_boot, esp flags. Ie mkfs.vfat -F32 /dev/disk and then running the Tails installer again.

Sorry to mess up your bug with this unrelated information.

#41 Updated by CyrilBrulebois 3 months ago

  • Target version changed from Tails_3.15 to Tails_3.16

#42 Updated by emmapeel about 2 months ago

intrigeri wrote:

  • Intel UHD 620 [8086:5917] rev07

@goupille, @mercedes508: I'm wondering if there's some confusion here. According to the report I've seen from one user of this hardware, the Greeter appears just fine and they don't see "Error starting GDM" (as long as they don't break stuff by adding boot options to debug the problem); instead, the Greeter appears but the screen goes black after login, once the desktop has been briefly displayed.

ey there! ill forward you now a report from a user with this card, UHD Graphics 620 [8086:5917]

The user could not start Tails after trying several options:

- xorg-driver=intel
- nomodeset
- nomodeset xorg-driver=vesa
- xorg-driver=modesetting

#43 Updated by intrigeri about 2 months ago

  • Assignee changed from mercedes508 to emmapeel

Hi @emmapeel,
@goupille,
and @numbat,

I guess it's not realistic to expect mercedes508 to lead this effort so I'll let you folks organize this in any way you want.

#44 Updated by CyrilBrulebois about 1 month ago

  • Target version changed from Tails_3.16 to Tails_3.17

#45 Updated by intrigeri about 1 month ago

  • Target version changed from Tails_3.17 to Tails_4.0

#46 Updated by goupille 22 days ago

  • Assignee changed from emmapeel to intrigeri

I just resent some logs about that (Message-ID: <> the ticket number is in the subject)

#47 Updated by intrigeri 8 days ago

  • Description updated (diff)
  • Assignee changed from intrigeri to numbat

I just went through all the reports I could find with the info help desk gave me, and aggregated in the ticket whatever useful data I could find there.

(In passing: that did not require particularly technical skills, only reading email and summing things up. If help desk has time to do this work in the future, I think it could be beneficial to the project :)

Unfortunately, this did not result in anything actionable for the Foundations Team: there's no workaround that seems reliable enough to document, and not enough debugging info to forward bug reports upstream.

At this point, our best bet is that 4.0 will fix at least some of these issues. Tails 4.0~rc1 will be released on Friday, so:

  • @emmapeel, it would be awesome if you could ping on Saturday or Sunday the folks who reported such issues; if that's too tight a timeline for you, please ask numbat to do so :)
  • @numbat, it would be awesome if you could collect the replies and give me aggregated feedback

Thanks in advance!

Also available in: Atom PDF