Created attachment 171781 [details] dmesg output under kernel 4.0-rc5. Under kernel 4.0-rc5 get a warning and trace in syslog on boot to console only at (I presume) i915 module load, and just before console switch from vga to i915 driver. So far no other adverse effects are apparent, and can carry on to start X, etc ok. Was last OK under kernel 4.0-rc4. See also BUG ID 94251. ------------------------------------ Hardware: Asus P7H55-M SI with Intel Core i5 processor and Integrated Graphics [i915 driver]. The board has 3 physical graphics connectors :- VGA = VGA1 DVI-D = HDMI1 in use. HDMI = HDMI2 Distro: Slackware-14.1, but with newer kernel. ------------------------------------ See trace in attached dmesg output produced on boot up to console only under kernel 4.0-rc5.
Created attachment 172381 [details] dmesg output +1 Can't tell if it's the same issue, bisecting now...
Got it: 319c1d420a0b62d9dbb88104afebaabc968cdbfa is the first bad commit commit 319c1d420a0b62d9dbb88104afebaabc968cdbfa Author: Xi Ruoyao <xry111@outlook.com> Date: Thu Mar 12 20:16:32 2015 +0800 drm/i915: Ensure plane->state->fb stays in sync with plane->fb plane->state->fb and plane->fb should always reference the same FB so that atomic and legacy codepaths have the same view of display state. However, there are some places in kernel code that directly set plane->fb and neglect to update plane->state->fb. If we never do a successful update through the atomic pipeline, the RmFB cleanup code will look at the plane->state->fb pointer, which has never actually been set to a legitimate value, and try to clean it up, leading to BUG's. Add a quick helper function to synchronize plane->state->fb with plane->fb and call it everywhere the driver tries to manually set plane->fb outside of the atomic pipeline. In this function, use drm_atomic_set_fb_for_plane instead of writing plane->state->fb directly to keep the reference count right. This is modified from Matt Roper's patch to drm-intel-nightly with commit id commit afd65eb4cc0578a9c07d621acdb8a570e2782bf7 Author: Matt Roper <matthew.d.roper@intel.com> Date: Tue Feb 3 13:10:04 2015 -0800 drm/i915: Ensure plane->state->fb stays in sync with plane->fb However this bug exists in mainline kernel too, so I created this to fix it in mainline kernel. A minor change is to use drm_atomic_set_fb_for_plane instead of update reference count manually. Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=88909 Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=93711 Signed-off-by: Xi Ruoyao <xry111@outlook.com> Reviewed-by: Matt Roper <matthew.d.roper@intel.com> [Jani: included the patch notes in the commit message] Signed-off-by: Jani Nikula <jani.nikula@intel.com> :040000 040000 baa4beae6b27f959570b3507f3c1237c6d9181bf d451c8e6ea1e02fc7779518c90c9d87a14786a39 M drivers
Please try drm-intel-fixes branch of http://cgit.freedesktop.org/drm-intel.
Yep, drm-intel-fixes-2015-03-26 is clean, no warnings... I didn't have to bisect last night, only two relevant commits, :facepalm :)
Thanks for the report and testing; hopefully this will make it to v4.0-rc6.
Applied patches from drm-intel-fixes-2015-03-26 to kernel 4.0-rc5 patch-8218c3f4df3bb1c637c17552405039a6dd3c1ee1 patch-59a58cb34d3fe73e6c899cc5d9a87428ca662925 patch-32a3aad1e2927b1684e7369c9f36a370e0b95da patch-3164a803416832d61268b758112e8dfd7e35cdf7 patch-5f407751b0ca9bd876fe8f15ff28153661c6ba0a Now does the console switch cleanly for me too - thanks.
(In reply to William from comment #6) > Now does the console switch cleanly for me too - thanks. Thanks for the additional testing. The commits are now in Linus' master.