[linux-audio-user] Re: Linux IRQ datapoint

[Lee Revell]

On Mon, 2004-10-11 at 17:40, Mark Knecht wrote:
> Lee Revell wrote:
> > On Mon, 2004-10-11 at 16:59, Mark Knecht wrote:
> > 
> >>I was building a new kernel under FC2 this morning trying to make FC2
> >>equal to my Gentoo kernel which is working very well. I made a mistake
> >>on this new kernel and forgot to turn on 'APIC contoller for UMP
> >>machines'. When the kernel booted I got all sorts of xruns. The IRQ
> >>was threaded, so I set it for non-threaded but still got a lot of
> >>xruns. (>100 in 10 minutes)
> >>
> >>I rebuilt the kernel where the only change I made was to enable APIC
> >>interrupts and IO-APIC controllers for UMP. The kernel booted and has
> >>been running for 30 minutes with no xruns...
> >>
> >>Just another IRQ datapoint, but that was the only change I made... 
> >>
> > 
> > 
> > That is really weird, most people report better results _disabling_ APIC
> > on UP.
> > 
> > Lee
> > 
> > 
> Who knows?!? I just take data. I don't interpret it! ;-)
> 
> Could be specific to this ATI chipset's interrupt controller for all I 
> know. I built a similar kernel yesterday, also under FC2, but on my 
> older Via KT266 MB with an Athlon XP (I think...) 1600+. That MB only 
> supports legacy style interrupts and the kernel worked fine, at least at 
> -p64/-n2 using an HDSP 9652 sound card.
> 
> I did find two strange things about that setup, both sound card related 
> I think:
> 
> 1) The HDSP driver would not allow me to even try and set -p <64. Just 
> told me it wasn't allowed. This may be by design. Possibly the way it 
> moves DMA data it just doesn't support less than that.
> 
> 2) There are two 'regions' of sample rates with that card
> 
> - 32K, 44.1K & 48K (i.e. - 1x)
> - 64K, 88.2K & 96K (i.e. - 2x)
> 
> If I try to change from any 1x sample rate to any 2 x sample rate using 
> Jack I get a message that it cannot do it. Parameters not defined or not 
> available IIRC. Same going from 2x to 1x.
> 
> However, if I use hdspconf and manually shift from 44.1K (1x) to say 96K 
> (2x) hdspconf reports that it changed. I can then use Jack to change 
> between all 2x rates, but I cannot go back to 1x. To go back to 1x I 
> have to once again use hdspconf. Sounds like a driver issue to me, but 
> who knows. (I'm not supposed to interpret!) ;-)

Yeah, it sounds like that behavior could just reflect the way the
hardware works.  You should post that to alsa-user if you suspect a bug,
the RME stuff is very well supported as you probably know.

If you are being limited by the size of a DMA transfer or your cards
constraint on DMA buffer size, on some cards (emu10k1) you can get lower
latency by recording more channels.

Lee

Lee