Re: skype on current/5.x and maestro-2E sound

On Mon, May 23, 2005 at 02:47:06PM +0200, cpghost wrote:
 > Pyun YongHyeon wrote:
 > 
 > >On Sun, May 15, 2005 at 02:41:24AM -0700, Julian Elischer wrote:
 > >> cpghost_at_cordula.ws wrote:
 > >> >On Sat, May 14, 2005 at 06:51:48PM -0700, Julian Elischer wrote:
 > >> >
 > >> >>Has anyone run skype successfully on these versions (5 or 6) of 
 > >freeBSD?
 > >> >>I can run it successfully on 4.x but on my 5.x machine the audio is 
 > >> >>completely
 > >> >>broken up. like someone is chopping the audio stream.
 > >> >
 > >> >
 > >> >I'm running Skype on 5.4 (via82c686). On an AMD Duron 1200 MHz, the
 > >> >sound quality is all right; on an EPIA 5000 Eden 500 MHz (also 
 > >via82c686),
 > >> >the sound is totally chopped and it is impossible to follow.
 > >> 
 > >> hmm so maybe its the fact that my machine is too slow.. it's also 500MHz
 > >> my 1GHz 4.11 machine seems to run it fine.
 > >> 
 > >
 > >I don't think 500MHz is too slow. Check if kernel converters
 > >are active when you play audio samples.(cat /dev/sndstat after
 > >setting hw.snd.verbose=2).
 > > 
 > >
 > This is what I get while skype plays the chopped sound of the echo service:
 > 
 > epia2# cat /dev/sndstat
 > FreeBSD Audio Driver (newpcm)
 > Installed devices:
 > pcm0: <VIA VT82C686A> at io 0xdc00 irq 10 kld snd_via82c686 (1p/1r/0v 
 > channels duplex default)
 >        [pcm0:play:0]: spd 48000, fmt 0x00000010, flags 0x00003030, 
 > 0x00000000, pid 78793
 >        interrupts 18062, underruns 4824, ready 3580
 >        {userland} -> feeder_root(0x00000010) -> {hardware}
 >        [pcm0:record:0]: spd 48000, fmt 0x00000010, flags 0x00003030, 
 > 0x00000000, pid 78793
 >        interrupts 16968, overruns 4808, hfree 512, sfree 3584
 >        {hardware} -> feeder_root(0x00000010) -> {userland}
 > 
 > No idea how to interpret this.
 > 

Format : sampling rate 48KHz, signed 16bit little endian format
channel flags : opened, block size set, DMA active, triggered
feeder flags : no flags(no kernel converter involed)

It seems that no kernel converters are active. But you have large
feeder underrun counters. It means sound quality would be poor.
Quick looking the driver source showed that via82c686 has fixed
sampling rate(48KHz) for playback. So userland application
may need to resampling operations which requires additional CPU cycles.

I vaguely guess the driver should work better than other cheap audio
cards as it supports scatter/gatter DMA operations. If 500MHz CPU is
not sufficient for the driver to work correctly I think something is
wrong in the driver. Since I don't have a copy of 82C686 data sheet
and hardware I don't know where the problem is. :-(

 > >Since the driver also needs Giant lock it may suffer from interrupt
 > >latencies with other devices. In addition if it share IRQ with
 > >other devices(e.g. USB) the issue would be noticable.
 > > 
 > >
 > None that I know of:
 > 
 > epia2# vmstat -i
 > interrupt                          total       rate
 > irq0: clk                       30624081         99
 > irq1: atkbd0                      416592          1
 > irq8: rtc                       39199914        128
 > irq10: pcm0                      1210380          3
 > irq11: vr0                        531082          1
 > irq12: psm0                      1354640          4
 > irq13: npx0                            1          0
 > irq14: ata0                      3681643         12
 > Total                           77018333        251
 > 
 > irc10 is not shared by any other likely interrupt source, or so it seems.
 > 

Do you use VCHAN? How about increasing DMA buffer size to 8 or 16KB?
Default DMA buffer size is 4KB(via82c686.c, line 44) which I guess
too small to support scatter/gatter DMA operations.
How many pcm interrupts do you see when you use skype?
(Use systat -vmstat 1 to to see the number of interrupts/second.)

-- 
Regards,
Pyun YongHyeon
http://www.kr.freebsd.org/~yongari	|	yongari_at_freebsd.org