Messages in this thread | | | Date | Tue, 16 Nov 2004 22:32:43 +0100 | From | Florian Schmidt <> | Subject | Re: [patch] Real-Time Preemption, -RT-2.6.10-rc2-mm1-V0.7.27-3 |
| |
On Tue, 16 Nov 2004 22:20:39 +0100 Florian Schmidt <mista.tapas@gmx.net> wrote:
> /proc/latency_trace doesn't show that high latencies either on console > switch: >
correction: now i do see large wakeup latencies in /proc/latency_trace, but only when rtc_wakeup is not running [??? :)]. something's fishy..
(IRQ 1/18/CPU#0): new 2 us maximum-latency wakeup. (IRQ 1/18/CPU#0): new 3 us maximum-latency wakeup. (syslogd/302/CPU#0): new 4 us maximum-latency wakeup. (ksoftirqd/0/3/CPU#0): new 4 us maximum-latency wakeup. (ksoftirqd/0/3/CPU#0): new 5 us maximum-latency wakeup. (IRQ 0/2/CPU#0): new 857 us maximum-latency wakeup. (IRQ 0/2/CPU#0): new 891 us maximum-latency wakeup. (IRQ 0/2/CPU#0): new 976 us maximum-latency wakeup. (IRQ 0/2/CPU#0): new 1117 us maximum-latency wakeup.
It seems this excerpt from below trace is characteristic for all the long traces:
5 80000002 0.001ms (+1.114ms): __do_softirq (do_softirq) 5 00000000 1.115ms (+0.000ms): preempt_schedule (_mmx_memcpy)
preemption latency trace v1.0.7 on 2.6.10-rc2-mm1-RT-V0.7.27-10 ------------------------------------------------------- latency: 1117 us, entries: 22 (22) | [VP:0 KP:1 SP:1 HP:1 #CPUS:1] ----------------- | task: IRQ 0/2, uid:0 nice:0 policy:1 rt_prio:50 ----------------- => started at: try_to_wake_up+0x51/0x170 <c010f3a1> => ended at: finish_task_switch+0x51/0xb0 <c010f911> =======> 5 80010004 0.000ms (+0.000ms): trace_start_sched_wakeup (try_to_wake_up) 5 80010003 0.000ms (+0.000ms): (49) ((98)) 5 80010003 0.000ms (+0.000ms): (2) ((5)) 5 80010003 0.000ms (+0.000ms): try_to_wake_up (wake_up_process) 5 80010003 0.000ms (+0.000ms): (0) ((1)) 5 80010002 0.000ms (+0.000ms): preempt_schedule (try_to_wake_up) 5 80010002 0.000ms (+0.000ms): wake_up_process (redirect_hardirq) 5 80010001 0.000ms (+0.000ms): preempt_schedule (__do_IRQ) 5 80010001 0.000ms (+0.000ms): irq_exit (do_IRQ) 5 80000002 0.000ms (+0.000ms): do_softirq (irq_exit) 5 80000002 0.001ms (+1.114ms): __do_softirq (do_softirq) 5 00000000 1.115ms (+0.000ms): preempt_schedule (_mmx_memcpy) 5 90000000 1.115ms (+0.000ms): __schedule (preempt_schedule) 5 90000000 1.115ms (+0.000ms): profile_hit (__schedule) 5 90000001 1.116ms (+0.000ms): sched_clock (__schedule) 2 80000002 1.116ms (+0.000ms): __switch_to (__schedule) 2 80000002 1.116ms (+0.000ms): (5) ((2)) 2 80000002 1.116ms (+0.000ms): (98) ((49)) 2 80000002 1.116ms (+0.000ms): finish_task_switch (__schedule) 2 80000001 1.116ms (+0.000ms): trace_stop_sched_switched (finish_task_switch) 2 80000001 1.116ms (+0.003ms): (2) ((49)) 2 80000001 1.120ms (+0.000ms): trace_stop_sched_switched (finish_task_switch)
Ah, and one more thing: When i boot up the computer my init script sets irq 3 to prio 98. But it seems the irq handler's priority changes actually when the soundcard is used the first time. So i need to re-set the irq prio _after_ i have used the soundcard for the first time..
flo - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
| |