On Wed, Jan 27, 2010 at 9:06 PM, John Kacur <jkacur@xxxxxxxxxx> wrote:
> On Wed, Jan 27, 2010 at 12:18 PM, yi li <liyi.dev@xxxxxxxxx> wrote:
>> Hi,
>>
>> In cyclictest, by default "tracing_thresh" will be set to "tracelimit"
>> (given to cyclictest with option "-b tracelimit").
>> If the maximum latency of the tracer currently used is less than
>> "tracelimit", no trace will be recorded.
>
> So, can't you just set the tracing_thresh to your desired value with the
> -b tracelimit option?
>
> Explain to me how cyclictest will behave if tracelimit and tracing_thresh
> are set to different values.
>
> If you showed us two sets of outputs (with and without your option)
> to show what setting tracing_thresh separately would achieve that
> would be helpful.
>
Here is an example:
1. run cyclictest with "wakeup-rt" tracer (In my case, max latency of
cyclictest is caused by "wakeup"):
root:/> ./cyclictest -p 80 -n -t 1 -l 10000 -i 20000 -b 5000 -W
could not set latency_hist/wakeup_latency/reset to 1
policy: fifo: loadavg: 4.98 3.32 1.47 6/30 363
T: 0 ( 359) P:80 I:20000 C: 373 Min: 74 Act: 5194 Avg: 158 Max: 5194
# Thread Ids: 00359
# Break thread: 359
root:/> cat /sys/kernel/debug/tracing/trace
# tracer: wakeup_rt
#
# wakeup_rt latency trace v1.1.5 on 2.6.32.6-ADI-2010R1-pre-svn8240
# --------------------------------------------------------------------
# latency: 5138 us, #5/5, CPU#0 | (M:desktop VP:0, KP:0, SP:0 HP:0)
# -----------------
# | task: -359 (uid:0 nice:0 policy:1 rt_prio:80)
# -----------------
#
# _------=> CPU#
# / _-----=> irqs-off
# | / _----=> need-resched
# || / _---=> hardirq/softirq
# ||| / _--=> preempt-depth
# |||| /_--=> lock-depth
# |||||/ delay
# cmd pid |||||| time | caller
# \ / |||||| \ | /
ls-364 0dNh.. 7us+: 364:120:R + [000] 359: 19:S cyclictest
ls-364 0dNh.. 20us!: _wake_up_process <-_hrtimer_wakeup
ls-364 0d.... 5127us+: _schedule <-_system_call
ls-364 0d.... 5130us : 364:120:R ==> [000] 359: 19:R cyclictest
2. Now we want to run cyclictest with "irqsoff" tracer:
Without my patch, "irqsoff" tracer cannot recored valid trace,
because "tracing_thresh" will be set to "5000", and max latency of
"irqsoff" in this case is less than "5000".
root:/> ./cyclictest -p 80 -n -t 1 -l 10000 -i 20000 -b 5000 -I
could not set latency_hist/wakeup_latency/reset to 1
policy: fifo: loadavg: 5.08 3.78 1.80 6/30 408
T: 0 ( 402) P:80 I:20000 C: 1217 Min: 84 Act: 179 Avg: 167 Max: 3933
# Thread Ids: 00402
# Break thread: 402
root:/> cat /sys/kernel/debug/tracing/trace
# tracer: irqsoff
#
# irqsoff latency trace v1.1.5 on 2.6.32.6-ADI-2010R1-pre-svn8240
# --------------------------------------------------------------------
# latency: 5138 us, #5/5, CPU#0 | (M:desktop VP:0, KP:0, SP:0 HP:0)
# -----------------
# | task: -359 (uid:0 nice:0 policy:1 rt_prio:80)
# -----------------
#
# _------=> CPU#
# / _-----=> irqs-off
# | / _----=> need-resched
# || / _---=> hardirq/softirq
# ||| / _--=> preempt-depth
# |||| /_--=> lock-depth
# |||||/ delay
# cmd pid |||||| time | caller
# \ / |||||| \ | /
ls-364 0dNh.. 7us+: 364:120:R + [000] 359: 19:S cyclictest
ls-364 0dNh.. 20us!: _wake_up_process <-_hrtimer_wakeup
ls-364 0d.... 5127us+: _schedule <-_system_call
ls-364 0d.... 5130us : 364:120:R ==> [000] 359: 19:R cyclictest
3. With my patch, run cyclic test with "irqsoff" tracer can get expected trace
(because "tracing_thresh" is reset to "0" in this case).
root:/> ./cyclictest -p 80 -n -t 1 -l 10000 -i 20000 -b 5000 -I
could not set latency_hist/wakeup_latency/reset to 1
policy: fifo: loadavg: 5.00 4.79 3.15 6/30 701
T: 0 ( 700) P:80 I:20000 C: 147 Min: 92 Act: 8706 Avg: 226 Max: 8706
# Thread Ids: 00700
# Break thread: 700
root:/> cat /sys/kernel/debug/tracing/trace
# tracer: irqsoff
#
# irqsoff latency trace v1.1.5 on 2.6.32.6-ADI-2010R1-pre-svn8240
# --------------------------------------------------------------------
# latency: 116 us, #3/3, CPU#0 | (M:desktop VP:0, KP:0, SP:0 HP:0)
# -----------------
# | task: -694 (uid:0 nice:0 policy:0 rt_prio:0)
# -----------------
# => started at: _user_path_at
# => ended at: _irq_exit
#
#
# _------=> CPU#
# / _-----=> irqs-off
# | / _----=> need-resched
# || / _---=> hardirq/softirq
# ||| / _--=> preempt-depth
# |||| /_--=> lock-depth
# |||||/ delay
# cmd pid |||||| time | caller
# \ / |||||| \ | /
ls-694 0d.... 1us!: __common_int_entry <-_user_path_at
ls-694 0dNs.. 112us+: ___do_softirq <-_irq_exit
ls-694 0dNs.. 119us : _trace_hardirqs_on <-_irq_exit
4. Or we can set a smaller "tracing_thresh" (e.g 100).
root:/> ./cyclictest -p 80 -n -t 1 -l 10000 -i 20000 -b 5000 -I --tracing_thresh
100
could not set latency_hist/wakeup_latency/reset to 1
policy: fifo: loadavg: 5.04 5.01 4.92 6/30 4235
T: 0 ( 4234) P:80 I:20000 C: 262 Min: 83 Act: 100 Avg: 165 Max: 1366
# Thread Ids: 04234
# Break thread: 4234
root:/> cat /sys/kernel/debug/tracing/tracing_thresh
100
root:/> cat /sys/kernel/debug/tracing/trace
# tracer: irqsoff
#
# irqsoff latency trace v1.1.5 on 2.6.32.6-ADI-2010R1-pre-svn8240
# --------------------------------------------------------------------
# latency: 102 us, #3/3, CPU#0 | (M:desktop VP:0, KP:0, SP:0 HP:0)
# -----------------
# | task: -4231 (uid:0 nice:0 policy:0 rt_prio:0)
# -----------------
# => started at: _dput
# => ended at: _irq_exit
#
#
# _------=> CPU#
# / _-----=> irqs-off
# | / _----=> need-resched
# || / _---=> hardirq/softirq
# ||| / _--=> preempt-depth
# |||| /_--=> lock-depth
# |||||/ delay
# cmd pid |||||| time | caller
# \ / |||||| \ | /
ls-4231 0d.... 1us+: __common_int_entry <-_dput
ls-4231 0d.s.. 99us+: ___do_softirq <-_irq_exit
ls-4231 0d.s.. 105us : _trace_hardirqs_on <-_irq_exit
>> +
>> + if (tracing_thresh > 0 && tracing_thresh <= tracelimit) {
>> + sprintf(buffer, "%d", tracing_thresh);
>> + setkernvar("tracing_thresh", buffer);
>> + }
>
> Do you need an else part that sets tracing_thresh to tracelimit?
>
I think by default set "tracing_thresh" to "0" makes more sense, i.e,
disable tracing thresh.
So max latency of current tracer can be recorded.
If people want the old behavior, cyclictest can be called with option
like " cyclictest -b 1000 --tracing_thresh 1000".
I will change this code and resend the patch:
+ if (tracing_thresh >= 0 && tracing_thresh <= tracelimit) {
+ sprintf(buffer, "%d", tracing_thresh);
+ setkernvar("tracing_thresh", buffer);
+ }
>
>>
>> /* ftrace_enabled is a sysctl variable */
>> fileprefix = procfileprefix;
>> @@ -766,6 +769,7 @@ static void display_help(int error)
>> " without -t default = 1\n"
>> "-T TRACE --tracer=TRACER set tracing function\n"
>> " configured tracers: %s\n"
>> + "--tracing_thresh=THRESH set tracing_thresh of current
>> tracer to THRESH\n"
>
> If you run your patch through checkpatch you'll get
>
> ERROR: patch seems to be corrupt (line wrapped?)
> #64: FILE: src/cyclictest/cyclictest.c:772:
> tracer to THRESH\n"
>
> And because of the above, I am unable to apply your patch
> with git am
>
> We don't follow all the kernel style recommendations in cyclictest
> but we try to follow most of them. So please run your patch
> through checkpatch and fix-up the warnings.
>
Sorry about that. I will resend the patch soon.
-Yi
--
To unsubscribe from this list: send the line "unsubscribe linux-rt-users" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
