Jan, have patience my friend. I will be reporting bugs regarding LTT tonight. ;) I have a day job that unfortunately requires my expertise.
I have local APIC disabled in the kernel because we don't seem to have one. I see context switches like these whether or not I use idle=poll, and thus the TSC. Here's my /proc/xenomai/timer: status=on+watchdog:setup=3:clock=40035659193:timerdev=pit:clockdev=tsc It seems that idle=poll didn't prevent the tsc from being marked as unstable despite what the mailing lists say. Steven > -----Original Message----- > From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] > Sent: Monday, February 11, 2008 1:20 PM > To: Steven Seeger > Cc: [email protected] > Subject: Re: [Xenomai-help] irq0 > > Steven Seeger wrote: > > I used the idle=poll option so I'm running the tsc now, and irq0 handler > > still has a lot of overhead. > > > > > > > > CPU PID MSW CSW PF STAT %CPU NAME > > > > 0 0 0 2 0 00500080 94.2 ROOT > > > > 0 0 0 1 0 00100082 0.0 sound task > > > > 0 0 0 1 0 00100082 0.0 tone task > > > > 0 0 0 775585 0 00000000 5.6 IRQ0: [timer] > > > > > > > > I'm seeing about 17,000 context switches a second for IRQ0. Sound task > > and tone task are doing nothing. They're just blocked. > > > > Ah! That could mean the timer gets misprogrammed, fires too early and > thus too often. I had this effect on an standard PC when we used a > slightly wrong APIC frequency. Do you use the APIC? -> /proc/xenomai/timer > > Also this effect can nicely be confirmed via LTTng traces. > > Jan _______________________________________________ Xenomai-help mailing list [email protected] https://mail.gna.org/listinfo/xenomai-help
