Sorry for the late reply, I was at Linux Plumbers, and had a bunch of stuff to catch up on when I returned.
On Sat, 21 Sep 2013 00:07:36 +0200 Mario Kleiner mario.kleiner@tuebingen.mpg.de wrote:
Steven, would it then be acceptable to convert that "faster" lock into a raw_spinlock_t or is this unacceptable? If so, the preempt_disable() could stay, right?
If a spinlock is tight (not held for more than 2us on todays processors), and has little contention, than I would be fine with converting it to raw. And if that's the only lock held you could do the preempt_disable() call.
In fact, if you want, you can leave the preempt_disable() out of mainline, and send a patch to us that uses "preempt_disable_rt()" and add a comment to it. In the -rt patch, preempt_disable_rt() is a nop when PREEMPT_RT is not set, and is preempt_disable() when it is.
-- Steve