Messages in this thread | | | From | Bryan Wu <> | Date | Mon, 22 Oct 2012 16:12:13 -0700 | Subject | Re: [PATCH] ledtrig-cpu: use spin_lock to replace mutex lock |
| |
On Mon, Oct 22, 2012 at 3:28 PM, Nathan Lynch <ntl@pobox.com> wrote: > Hi Bryan, > > On Thu, 2012-10-18 at 11:18 -0700, Bryan Wu wrote: >> @@ -117,14 +117,14 @@ static int __init ledtrig_cpu_init(void) >> for_each_possible_cpu(cpu) { >> struct led_trigger_cpu *trig = &per_cpu(cpu_trig, cpu); >> >> - mutex_init(&trig->lock); >> + spin_lock_init(&trig->lock); >> >> snprintf(trig->name, MAX_NAME_LEN, "cpu%d", cpu); >> >> - mutex_lock(&trig->lock); >> + spin_lock(&trig->lock); >> led_trigger_register_simple(trig->name, &trig->_trig); >> trig->lock_is_inited = 1; >> - mutex_unlock(&trig->lock); >> + spin_unlock(&trig->lock); > > I wouldn't know how to fix the original problem, but I don't think this > patch is okay -- led_trigger_register_simple() does things that > potentially sleep (GFP_KERNEL allocation, down_write), so it's not safe > to call while holding a spinlock. >
Looks like we got a hard issue, since led_trigger_register() families and even led_trigger_event() might use rwsem for locking, these are all potentially sleep, which are not easy to modified for atomic context.
Any hints are welcome.
Thanks, -Bryan
| |