Re: 3.17-rc1: leds blink workqueue causes sleeping BUGs

Hello,

Sabrina Dubroca, le Mon 25 Aug 2014 23:13:40 +0200, a écrit :

2014-08-19, 13:06:07 -0400, Valdis.Kletnieks@vt.edu wrote:

quoted

On Sat, 16 Aug 2014 20:27:01 -0700, Hugh Dickins said:

quoted

Can we safely revert your 8b37e1bef5a6 ("leds: convert blink timer to
workqueue"), or have there been other changes which now depend upon it?

I suspect there's something else busted.  I hand-reverted that patch, and I *still*
see the following lockdep whine that looks related (as it talks about
leddev_list_lock).  next-0811 was OK, looks like next-0815 and -0818 had this....

I had a look at the code, led_trigger_event calls vt_led_set, which
calls led_trigger_event again.

Yes, that is expected: the vt::* leds actually generate the
corresponding vt-* trigger events, which are used by the various
input*::* leds. We could indeed have a loop if the user was making the
VT::* leds use the vt-* trigger, but otherwise it's safe since it's a
different trigger. We can add code to prevent the user from building
loops, but otherwise it's a false positive.

Samuel