On Sat, 11 Dec 2004, Doug White wrote: > Okay, I'm testing this now ... so far no issues, unless I force things. > > The system is currently running 5.3-STABLE built Dec 5. The machine is an > Althon 3000+ (Barton) on a KT400-based Soyo using either the built-in CMI > audio or a PCI SBLive 5.1 Digital card. SCHED_4BSD, no PREEMPTION, no > apic, no WITNESS/INVARIANTS. > > I played a 160kb VBR MP3 with mpg321 (mpg123 was compiled to use esound on > this system so I decided to try something not so polluted) direct to each > card in turn with a -j3 buildworld running in the background, using the > same disk as the MP3s were located on. I ran the buildworld in a Konsole > window on KDE 3.2.3 and played the mp3's in another. > > I start getting wierdness when the load average hits 5+ with 0% idle. > Otherwise things hold up. > > Interrupt layout: > > zao,ttyp3,~,11>vmstat -i > interrupt total rate > irq0: clk 115654 99 > irq1: atkbd0 3131 2 > irq5: fwohci0 uhci2 2 0 > irq6: fdc0 9 0 > irq8: rtc 148030 127 > irq10: pcm0 ehci0 23453 20 > irq11: vr0 uhci0 23291 20 > irq12: pcm1 uhci1 20841 17 > irq13: npx0 1 0 > irq14: ata0 64869 56 > irq15: ata1 46 0 > Total 399327 344 > > During the test pcm0 is running at 600 interrupts/s and pcm1 is at a > measely 85 interrupts/s. > > If you are getting distortion on an otherwise idle system I'll be more > interested, although that doesn't appear to be the case... > > I'll repeat the runs after world finishes building & installing and see if > its something that changed in the last week. I'd suggest that people > verify that their sound card isn't stuck on the same interrupt as a really > busy device or they aren't having a quiet interrupt storm going off > somewhere else or otherwise running things that chew CPU. > > Of course, make sure WITNESS & INVARIANTS are removed from the running > kernel. WITNESS is guaranteed to cause problems with sound; setting > 'sysctl debug.witness.watch=0' should clear up problems immediately if you > have WITNESS compiled in. > > phk's suggested that usb is a huge hog and sharing interrupts with a USB > controller could cause latency problems. I'll investigate this after the > buildworld finishes and I check if things are different than with last > week's build. Ok, new build, new run. cvsup'd RELENG_5 about 2 hours ago. Things seem better so far; I've turned on some extra hammer tests and everything is quite nice. I added a window that does 'while(1) cat /dev/sndstat; sleep 1' with hw.snd.verbose=2 so the card statistics (including buffer overruns) are printed. Unfortunately now the system hangs if you restart a monitoring tool at the wrong moment. I ^C'd systat to see if the interrupt list would reorder (irq12 was out of order, at the bottom instead of sorted) and the system hung about when the first refresh would have happened. :( I killed it another time ^Cing sysctl so something bad crept in. Serial break for ddb wasn't responding so I'll need to get my PE1750 with an NMI button. > On Thu, 9 Dec 2004, Krzysztof Kowalik wrote: > > > Ivan Voras [ivoras_at_fer.hr] wrote: > > > [...] > > > Not exactly a freeze, but PREEMPTION doesn't help me at all with SB Live > > > (slowdowns, stuttering sound, etc.), especially with large amounts of > > > filesystem operations (this is 5-stable). > > > > Exactly. I tried to use both emu10k1 and emu10kx drivers, with kernel > > with and without PREEMPTION, and both 5.x and 6.0 serie did behave same. > > Intensive I/O made the system quite unusable, and it was not only a > > sound-related issue, but a general one -- slow, lagging mouse in X, etc. > > > > I asked about it on stable_at_ and somehow got no answers but "me too". > > > > Finally, I decided to restore my 4.9 system from the backups, where the > > problem does not exist. > > > > Regards, > > > > -- Doug White | FreeBSD: The Power to Serve dwhite_at_gumbysoft.com | www.FreeBSD.orgReceived on Sun Dec 12 2004 - 00:12:13 UTC
This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:38:24 UTC