This fixes the problem of getting extra bytes inserted at the
beginning of a recording when using the Apple i2s interface and DBDMA
controller.  It turns out that we can't just abort the DMA; we have to
let it stop at the end of a command, and then wait for the S7 bit to
be set before turning off the DBDMA controller.  Doing that for
playback doesn't seem to be necessary, but doesn't hurt either.
We use the technique used by the Darwin driver: make each transfer
command branch to a stop command if the S0 status bit is set.  Thus we
can ask the DMA controller to stop at the end of the current command
by setting S0.
The interrupt routine now looks at and clears the status word of the
DBDMA command ring.  This is necessary so it can know when the DBDMA
controller has seen that S0 is set, and so when it should look for the
DBDMA controller being stopped and S7 being set.  This also ended up
simplifying the calculation in i2sbus_pcm_pointer.
Tested on a 15 inch albook.
[Addition by Johannes]
I modified this patch and added the suspend/resume bits to it to get my
powermac into a decent state when playing sound across suspend to disk
that has a different bitrate from what the firmware programs the
hardware to.
I also added the SNDRV_PCM_INFO_JOINT_DUPLEX flag because it seemed the
right thing to do and I was looking at the info stuff.

Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Jaroslav Kysela <perex@suse.cz>