Skip to content

drm: defer hotplug rescans behind a settle timer - #493

Closed
YF36 wants to merge 1 commit into
hyprwm:mainfrom
YF36:drm/hotplug-settle-timer
Closed

YF36 wants to merge 1 commit into
hyprwm:mainfrom
YF36:drm/hotplug-settle-timer

Conversation

@YF36

@YF36 YF36 commented Nov 12, 2026

Copy link
Copy Markdown

DPMS-off on nvidia GB10 emits udev connector-change events and transiently reports the connector disconnected to forced probes while the output is still live. The immediate recheckOutputs() on every hotplug event then destroys a live output on a transient reading; the status settles back to connected shortly after, the output reconnects, and the DPMS-off state machine re-disables it. The resulting modeset war burns a full core until the CPU watchdog kills the compositor.

Coalesce rescans instead: every hotplug event re-arms a 500ms one-shot timerfd, and the rescan runs once the event stream has been quiet for the settle window. The timer rides the backend poll loop; if the timerfd cannot be created we fall back to the old immediate rescan.

Twelve disable/enable rounds against a live session on the affected hardware survive with the connector status stable and no watchdog kills, where an unpatched build dies on the first disable.

DPMS-off on nvidia GB10 emits udev connector-change events and transiently
reports the connector disconnected to forced probes while the output is
still live. The immediate recheckOutputs() on every hotplug event then
destroys a live output on a transient reading; the status settles back to
connected shortly after, the output reconnects, and the DPMS-off state
machine re-disables it. The resulting modeset war burns a full core until
the CPU watchdog kills the compositor.

Coalesce rescans instead: every hotplug event re-arms a 500ms one-shot
timerfd, and the rescan runs once the event stream has been quiet for the
settle window. The timer rides the backend poll loop; if the timerfd
cannot be created we fall back to the old immediate rescan.

Twelve disable/enable rounds against a live session on the affected
hardware survive with the connector status stable and no watchdog kills,
where an unpatched build dies on the first disable.

Co-Authored-By: Claude Code <noreply@anthropic.com>
@github-actions

Copy link
Copy Markdown

Hi @YF36,

Thanks for your contribution! Unfortunately, Hyprland and hypr* no longer
accept contributions from unvouched contributors.

To get vouched, please see our contributing guide at wiki.hypr.land/Contributing-and-Debugging

This PR has been closed automatically and will not be re-opened.

@github-actions github-actions Bot closed this Nov 12, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant