problem with sound driver in iBook2

[Heinz Kirchmann <kirchman@dfki.uni-kl.de>]

Hi,

I tried to collect some info, why the dmasound driver does not work 
correctly on my iBook2. Besides the fact, that recording is not possible 
with the standard dmasound_pmac and dmasound_core drivers, I noticed 
problems with audacity, too. After calling audacity (version 1.0.1) 
neither recording nor playing were possible any more. I made some tests 
and it turned out, that opening /dev/dsp with flag O_RDWR caused the 
driver not to release the device cleanly.

System:
  Hardware: iBook2 G3 800 12"
  OS: Debian 3.0
  Kernel: 2.4.21-ben2 (from rsync, no configuration changes)

I took a closer look at the kernel sources of dmasound_core.c and 
dmasound_awacs.c in subdirectory 'drivers/sound/dmasound'.

I noticed the following strange things in dmasound_core.c and 
dmasound_awacs.c:

a) in sq_open (dmasound_core.c) I found two (conflicting?) concepts to 
decide, if the soundcard offers recording possibility or not: macro 
'HAS_RECORD' was set and dmasound.mach.record was _not_ set. I decided 
to change the corresponding line in function dmasound_awacs_init (file 
dmasound_awacs.c) around line 3109:

        if (awacs_revision != AWACS_TUMBLER &&
            awacs_revision != AWACS_SNAPPER &&
            awacs_revision != AWACS_DACA)
		dmasound.mach.capabilities = DSP_CAP_DUPLEX ;
		dmasound.mach.record = PMacRecord ;
...

I commented out the line with AWACS_SNAPPER in it, thus causing 
dmasound.mach.record to be set to a value != NULL.

b) the structure sq_fops of type 'struct file_operations' (in 
dmasound_core.c) does not define a function for 'read'. Even though 
'HAS_RECORD' is defined, the corresponding value is set to NULL. What am 
I missing here?
I decided to set the value for 'read' to 'sq_read'.

These two changes changed the behaviour of the driver as follows:
a) open for reading now possible, but a call to read will cause the read 
function to return immediately with successful return value but returned 
data does not contain any senseful samples.
b) driver does not lock up sound device after opening with O_RDWR.

That's all: not too much, but I did not investigate further, because I 
could not understand the driver's read function. If someone can give me 
some hints, how to continue, I will try.

Heinz