Bug 42649 - Regression 3.2 -> 3.3-rc1 commit b4b583d4 inet/media keyboard: stuck "extra" keys
Regression 3.2 -> 3.3-rc1 commit b4b583d4 inet/media keyboard: stuck "extra" ...
Product: Drivers
Classification: Unclassified
Component: Input Devices
All Linux
: P1 normal
Assigned To: drivers_input-devices
Depends on:
Blocks: 42644
  Show dependency treegraph
Reported: 2012-01-24 18:18 UTC by Duncan
Modified: 2012-03-11 21:29 UTC (History)
3 users (show)

See Also:
Kernel Version: 3.3-rc1
Tree: Mainline
Regression: Yes

kernel config (60.21 KB, text/plain)
2012-01-25 09:26 UTC, Duncan

Description Duncan 2012-01-24 18:18:42 UTC
I just bisected a regression down to:

commit b4b583d4e9a5ff28c4a150bb25a4fff5cd4dfbbd
Author: Jiri Kosina <jkosina@suse.cz>
Date:   Mon Oct 31 16:26:22 2011 +0100

    HID: be more strict when ignoring out-of-range fields

I have a logitech USB keyboard with extra inet/media keys.  In kde I have them configured to launch various utilities.  With this commit, these keys will "stick", and instead of a single konsole window or krunner dialog or whatever, it continues launching konsoles or toggling the krunner dialog on and off.  Switching to a text VT and killing X gets me the system back, but of course without X, and launching X/KDE over again results in unresponsive hotkeys.

Apparently, one of these "out of range" fields translates to a key-release for at least some of the extra keys, without which they stay pressed and auto-repeat.

gentoo/amd64 system, logitech wireless usb keyboard, kde 4.7.97 (aka 4.8-rc2), xf86-evdev-2.6.0, xorg-server-1.11.3.

I can attach .config later (headed to work now).
Comment 1 Duncan 2012-01-25 09:26:10 UTC
Created attachment 72189 [details]
kernel config

I just confirmed that reversing that commit against 3.3.0-rc1-167-gf8275f9 eliminates the issue, so it is definitely that commit triggering it.

Meanwhile, here's my current kernel config as promised.
Comment 2 Duncan 2012-02-06 07:37:43 UTC
The problem remains as of 3.3.0-rc2-172-g23783f8.  Reverting still eliminates it.
Comment 3 Duncan 2012-03-01 06:16:09 UTC
Bug seems to have been overlooked in the 3.2 -> 3.3 regression list.  Adding the block on the tracker bug and changing the summary to reflect the regression will fix that, I hope.

AFAIK the bug is still here on rc5.  I've applied the regression patch locally tho, so have to manually verify it with a reset to upstream, build and boot to X/KDE, from time to time.  Last I did, it was still unfixed.
Comment 4 Rafael J. Wysocki 2012-03-04 20:02:04 UTC
First-Bad-Commit : b4b583d4e9a5ff28c4a150bb25a4fff5cd4dfbbd
Comment 5 Rafael J. Wysocki 2012-03-04 21:38:44 UTC
Patch : http://www.spinics.net/lists/linux-input/msg19542.html
Comment 6 Duncan 2012-03-05 00:55:46 UTC
Confirmed, that fixes my problem here, too.
Comment 7 Rafael J. Wysocki 2012-03-11 21:29:16 UTC
Fixed by commit 883e0e366209067e690356e58e19bb2e6693b839 .

Note You need to log in before you can comment on or make changes to this bug.