On Monday 12 February 2001 11:30 pm, Reinoud Zandijk wrote:
> 'evening Chris (last time this night! :))

I know that feeling ;)  Perhaps we'd be better of using ICQ or talkd :)

> On Mon, 12 Feb 2001, Chris Gilbert wrote:
> > I think other things use it, IOMD possibly.  As I said it locked up
> > (normally doing a make build)  I suspect that we need to work out what
> > buffers are DMA and make the malloc's use a dma version.  Preference
> > would be to not use the simms but for DMA.  However it needs pmap work I
> > believe.  I've not got time at the mo to investigate, I'm moving in a few
> > days time.
>
> Well yeah that would be good to investigate. Sound is definataly one ...
> it seems to use uvm_km_alloc(kernel_map, ....), so doing a grep yields :
> mainly sound !!! vidc/lmaudio.c, vidc/vidcaudio.c and
> vidc/console/vidcconsole.c (cursor).... hmmm... they could use a MD
> allocation routine that gets mapped to the UVM routines when you don't
> have a Kinetic or all memory is DMA-able or get the memory from the pool
> of DMA DRAM.  Pity I cant test this out... I'll try to work out the
> routines etc. but real testing is a bit more difficult. Oh well... we'll
> see..
>
> There are also two SCSI podule drivers that seem to use it ... but those
> might be less critical, but never the less need patching....
>
> They all seem to be claiming DMA buffers one time ... so even that doesn't
> have to be a problem but needs investigating.

Perhaps as Richard suggests we need to look into doing bus_space DMA stuff, 
IE do it properly, not some strange way and then have to do the bus_space 
stuff.

> When do you think you'll be back agian ? or rather online :) ... well i
> completely understand you wont have the time ... take some time off :)

I'll be online, however I'm spending next week moving so probably a fortnight 
for anything serious, test booting a kernel should be fine.

Cheers,
Chris