snd-pcsp driver takes chip->substream_lock together with PCM substream
lock.  These are even mixed up with hrtimer's lock, resulting in messy
lock depencies.  Right now, snd-pcsp driver resolves the deadlock by
using HRTIMER_CB_SOFTIRQ.  However, this isn't nice for a really fast
path like bit-flipping.

This patch introduces a tasklet for PCM period handling so that the
hrtimer callback can be handled fast.  This also reduce the use of
chip->substream_lock to avoid deadlocks.  It's still used in pointer
callback, but even this could be removed with a proper barrier.

Another good solution is to introduce async trigger callback.  But,
this will involve with a major rewrite of the PCM core code, so I
take first this easy fix.
Signed-off-by: NTakashi Iwai <tiwai@suse.de>