Hey, Den 2026-02-25 kl. 08:58, skrev Sebastian Andrzej Siewior: > On 2026-02-24 17:59:04 [+0100], To Maarten Lankhorst wrote: > … >> This one I can reproduce while opening a browser and surfing a bit. Let >> me try step by step without lockdep, without RT and maybe just play >> v7.0-rc1 to figure out where this is from… > > It triggers without lockdep but takes longer. It does not trigger > without RT so far. I will try to narrow it down later unless it is > obvious where this is coming from. > > Sebastian Personally I've been running this entirely on xe, and it in stability it is comparable to !PREEMPT_RT.
I didn't entirely fix i915 yet. The changes I've made are the changes from the PREEMPT_RT tree, plus what I found when running the submissions against CI. The most likely problem is that the uncore lock needs to be converted to a raw spinlock. I postponed this, as I have been focused mostly on display and then especially the vblank bits. The selftests failing are very likely related, GPU reset is probably broken on i915 if I look at the CI results. GPU reset failure: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_159035v12/fi-cfl-8700k/igt@i915_selftest@[email protected]#dmesg-warnings533 I'll check what happens if I convert the spinlock to a raw_spinlock. Kind regards, ~Maarten Lankhorst
