dma16, PAS16 and static noise

[Rasmus Tamstorf <rt brandx net>]

Hi there,

After digging through web-pages, archives, and source code I give in and
hope that someone here can help me. 

I'm trying to get an old Pro Audio Specturm 16 sound card to work with the
2.2.14 kernel. It sort of works in the sense, that I get sound ... but
regardless of whether I compile the sound system into the kernel or as a
module I get a lot of static noise. Looking at deja-news I can see that
other people have had that problem and that it's most likely a dma or irq
conflict. Unfortunately the best advice I found for resolving it is trial 
and error and I've tried some of that to no avail. 

Now, before I continue with the trial and error I'd like to understand
enough to *not* choose settings which are certain not to work. This leads
to the following DMA related questions :

1) Why can you specify both a dma and a dma16 channel for the PAS16
   driver? I assume that 'dma' is an 8 bit channel, while 'dma16' is a
   16 bit channel, but I thought the native (non-SB-compatible) mode for
   PAS16 was 16 bit, so why does it need an 8 bit channel ? And why can
   you set a 16 bit dma channel for the SB part of the card (sb_dma16)
   when that part is only capable of doing mono (8 bit, right ?) ?

2) Given that you can specify both 'dma' and 'dma16' can they be equal, or
   should they never be ? What are the implications either way ?

3) Are the lower DMA numbers (like 1-3) reserved for 8 bit DMA channels
   while, say, DMA 4-7 are 16 bit or can any DMA channel be used for
   either 8 bit or 16 bit ? 

4) I've seen the abbreviation "drq" used. Is that short for "dma and irq",
   is it a synonym for "dma" or what does it mean ?

An (almost) non-dma related question concerns whether to compile the
soundsystem into the kernel or as a module. In one posting on
comp.linux.setup someone claimed to be unable to make the PAS16 card work
correctly if the driver weren't compiled into the kernel. Aside from the
problem there might be allocating dma buffers after a long uptime (if a
persistent dma buffer is not used), is there any reason this should be
true ? In other words is the module solution inferior to the compiled-in
solution ?

I hope someone has actually read this far, and look forward to hearing
about any insights. 

Rasmus