Public bug reported:

SUMMARY

After resuming from S3 suspend, both external monitors connected through a
USB-C hub stop working. The DRM connectors report status=connected but with
ZERO modes, so Mutter cannot create the outputs and the screens stay dark.
A physical unplug/replug of the hub always restores them.

This is a regression in the 7.0 kernel series. The same hardware, hub, NVIDIA
driver and userspace worked correctly on every 6.8 kernel.

ENVIRONMENT

  Distro    : Ubuntu 26.04.1 LTS (Resolute Raccoon)
  Kernel    : 7.0.0-44-generic (also reproduced on 7.0.0-38, -42, -43)
  Machine   : Lenovo ThinkPad P53 (20QQS20R03), BIOS N2NET61W 1.46, 2024-07-23
  GPUs      : Intel UHD 630 8086:3e9b (i915) + NVIDIA Quadro T1000 Mobile
              10de:1fb9 (nvidia)
  Driver    : nvidia-driver-580 580.178.04-0ubuntu0.26.04.1, 
nvidia-drm.modeset=1
  Session   : GNOME 50.1 / Mutter 50.1, Wayland
  Sleep     : deep (S3); s2idle also available
  Displays  : 2 x Samsung S24C450 1920x1200 behind a USB-C hub (0a05:7211)

The monitors are NOT on the HDMI port. They arrive as DisplayPort Alt-Mode over
USB-C and terminate on the NVIDIA GPU:

  card2-DP-2     connected     <- monitor 1
  card2-DP-3     connected     <- monitor 2
  card2-HDMI-A-1 disconnected
  card1-eDP-1    connected     <- internal panel (i915)

UCSI is broken on this machine on ALL kernels (PPM init failed, con2: failed to
register alt modes), so the kernel never manages DP Alt-Mode here - the EC
firmware does. This is pre-existing and NOT the regression (verified present on
6.8.0-88, -90, -107, -136 as well).

STEPS TO REPRODUCE

  1. Connect two monitors through the USB-C hub, confirm they work.
  2. systemctl suspend
  3. Wake the machine (open lid / keypress)

EXPECTED RESULT

Both external monitors come back.

ACTUAL RESULT

Both monitors stay dark. Per resume, journalctl shows once per monitor:

  gnome-shell: Failed to create KMS output: No modes available

and the connectors are in this state:

  card2-DP-2: status=connected  modes=0  enabled=disabled  dpms=Off
  card2-DP-3: status=connected  modes=0  enabled=disabled  dpms=Off

Mutter then reports only eDP-1. Physically replugging the hub restores
everything immediately.

EVIDENCE: 6.8 -> 7.0 REGRESSION

Counted across every boot retained in the journal (~1 year), using
"PM: suspend exit" for resumes and "journalctl _COMM=gnome-shell" for the
"No modes available" messages:

  Kernel series                                   Resumes   No-modes
  6.8 series (6.8.0-87 .. 6.8.0-147, 12 versions)     346          0
  7.0 series (7.0.0-38, -42, -43, -44)                 59        140

Per version:

  6.8.0-87   resumes=23  nomodes=0
  6.8.0-88   resumes=45  nomodes=0
  6.8.0-90   resumes=57  nomodes=0
  6.8.0-94   resumes=13  nomodes=0
  6.8.0-100  resumes=90  nomodes=0
  6.8.0-107  resumes=26  nomodes=0
  6.8.0-111  resumes=21  nomodes=0
  6.8.0-117  resumes=6   nomodes=0
  6.8.0-124  resumes=8   nomodes=0
  6.8.0-136  resumes=19  nomodes=0
  6.8.0-139  resumes=33  nomodes=0
  6.8.0-146  resumes=5   nomodes=0
  6.8.0-147  resumes=0   nomodes=0
  7.0.0-38   resumes=17  nomodes=38
  7.0.0-42   resumes=7   nomodes=17
  7.0.0-43   resumes=30  nomodes=74
  7.0.0-44   resumes=5   nomodes=11

346 resumes on the 6.8 series produced ZERO occurrences of this message; every
7.0 resume reproduces it (~2.4 messages/resume = one per monitor plus retries).
The system was upgraded 24.04 -> 26.04 on 2026-10-15
(/var/log/dist-upgrade/20261015-1023), which is exactly the 6.8 -> 7.0 switch.

Last known good: 6.8.0-147. First bad: 7.0.0-38.
Still bad as of 7.0.0-44 (2026-11-20), newest in resolute-updates.

ADDITIONAL OBSERVATIONS

  - The USB-C hub frequently re-enumerates ~2-11 s AFTER wake
    (usb 1-4: USB disconnect -> new full-speed USB device). This happens on 6.8
    too (rarely) but on 6.8 it never broke the display.
  - The post-resume connector probe appears to run ~1 s after wake, before the
    DP Alt-Mode link / AUX channel is usable, so the EDID read fails and the
    connector is left with an empty mode list. My inference is that 6.8
    retained the connector's cached mode list across S3, so userspace could
    re-apply a mode without a fresh EDID read - that part is not proven.
  - ~/.config/monitors.xml records the panels as DP-1-0.8 / DP-1-0.9 (DP-MST)
    while the kernel presents them as DP-2 / DP-3 (SST) in the failed state, so
    the saved layout does not auto-apply either.

WORKAROUNDS

Does NOT work (reproduced):
  - udevadm trigger --action=change --subsystem-match=drm
    (synthetic uevent without HOTPLUG=1; Mutter ignores it)
  - echo detect > /sys/class/drm/card2-DP-2/status on its own
    This DOES re-read the EDID and restores modes=18, but the outputs remain
    enabled=disabled / dpms=Off and Mutter still shows only eDP-1, so the
    screens stay dark. It notifies nobody.
  - reloading ucsi_acpi / typec_ucsi (also reported as crashing in LP #2166167)

DOES work (verified):
  1) restore the mode list:
       for c in /sys/class/drm/card2-DP-*/status; do echo detect > "$c"; done
  2) force a REAL hotplug - the kernel only emits
     drm_sysfs_connector_hotplug_event() when a connector's status CHANGES:
       for c in /sys/class/drm/card2-DP-2 /sys/class/drm/card2-DP-3; do echo 
off > "$c/status"; done
       sleep 2
       for c in /sys/class/drm/card2-DP-2 /sys/class/drm/card2-DP-3; do echo 
detect > "$c/status"; done

     Afterwards: enabled=enabled, dpms=On, Mutter reports DP-2, DP-3, eDP-1 and
     the screens come back without touching the cable.

I wrapped both steps in a systemd-sleep post hook that runs a few seconds after
resume, retries until the hub has re-enumerated, then does the off->detect flip.
It now recovers automatically on every wake. Happy to attach it; it is only a
workaround, the kernel side still needs fixing.

Side note for anyone copying this: a systemd-sleep hook must launch its work via
systemd-run (separate cgroup). Simply backgrounding with "setsid ... &" gets
killed the moment systemd-suspend.service tears its cgroup down.

REQUEST

Please consider this a resume-ordering / connector-invalidation regression in
the 7.0 DRM core (or in how nvidia-drm interacts with it). Useful directions:

  - should connectors be invalidated (mode list cleared) on resume at all, or
    should the previous modes be retained as on 6.8?
  - should the post-resume probe be deferred/retried rather than performed once,
    immediately after wake?
  - should the nvidia-drm resume path re-read EDID after the USB-C Alt-Mode/hub
    has settled?

I can test patches, run drm.debug=0x1e, or bisect if that helps.

ATTACHMENTS

** Affects: linux (Ubuntu)
     Importance: Undecided
         Status: New

** Attachment added: "attachment-evidence.txt"
   
https://bugs.launchpad.net/bugs/2175414/+attachment/6019721/+files/attachment-evidence.txt

-- 
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/2175414

Title:
  After resuming from S3 suspend, both external monitors connected
  through a USB-C hub stop working.

To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/linux/+bug/2175414/+subscriptions


-- 
ubuntu-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs

Reply via email to