Summary: | [skl] Kwin crashing | ||
---|---|---|---|
Product: | Mesa | Reporter: | Mike Lothian <mike> |
Component: | Drivers/DRI/i965 | Assignee: | Intel 3D Bugs Mailing List <intel-3d-bugs> |
Status: | RESOLVED FIXED | QA Contact: | Intel 3D Bugs Mailing List <intel-3d-bugs> |
Severity: | normal | ||
Priority: | medium | CC: | mike, topi.pohjolainen |
Version: | git | ||
Hardware: | Other | ||
OS: | All | ||
Whiteboard: | |||
i915 platform: | i915 features: |
Description
Mike Lothian
2017-07-23 13:16:25 UTC
Looking at the backtrace it looks like there is some sort of of clean-up going on, destructor for KWin::SceneOpenGL::Texture getting called. What surprises me a bit is that it seems to kick off rendering with GPU as isl_emit_depth_stencil_hiz_s() gets called to setup the hardware. For normal rendering there is its own path for setting depth+stencil and ending up here means that the GPU is being setup for blorp. Anyway, is ~Texture() really suppose to render something and more importantly is the driver context still suppose to be valid at this point. Looking at: 6 0x00007f294cfdb534 in isl_gen5_emit_depth_stencil_hiz_s (dev=0x0, batch=0x1344830, info=0x1) at /var/tmp/portage/media-libs/mesa-9999/work/mesa-9999/src/intel/isl/isl_emit_depth_stencil.c:67 All three arguments are pointers and must be valid, first is simply NULL and third clearly invalid "0x1". And setting up depth stencil for gen5 means you would need to be running ILK. I'm guessing this is newer hardware, right? If so, it would also indicate that the driver context isn't sound anymore as it contains wrong hardware generation specifier. I just tried kwin 5.10.4 on my Kabylake system (should be basically identical to Skylake), and it seems to be working fine... :/ I noticed the trace said KWin::AbstractEglTexture::~AbstractEglTexture() I thought kwin had removed EGL support, turns out they just removed the setting in Control Panels and I've been using EGL rather than GLX since the option was removed The option in in ~/.config/kwinrc under compositing GLPlatformInterface=egl I'll try removing it and see if I still have issues, it could be a sign that something is broken in Mesa's EGL code Would the 0x1 be related to https://cgit.freedesktop.org/mesa/mesa/commit/?id=63c251e38f9b4ac3606094a72c8c5b658402f3d0 No, the last argument "info=0x1" is a pointer to a structure (struct isl_depth_stencil_hiz_emit_info) that specifies the depth and stencil buffers against which the depth-stencil state setup for the hardware is based on. unless someone can reproduce this, it shouldn't block the release. |
Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.