Wolfgang Grandegger wrote:
> roland Tollenaar wrote:
>> Hi
>>
>>> Two conclusions:
>>>
>>>  - You are running your kernel as i586 without TSC support - suboptimal,
>>>    costs you a few micros.
>> I am aware of this. For the current machine I will stick to this.
>>
>>
>>>  - The reported latency perfectly matches the trace, nothing
>>>    pathological there. The trace looks like this: Timer fired,
>>>    measurement task woken up, two interrupts squeeze themselves between
>>>    wakeup and time stamp acquisition. All sane.
>> Is it not a bit strange that a machine as fast as this one is supposed
>> to be give worse latencies than much slower machines. Are the 2
>> interrupts causing the latency?
> 
> Yes, what puzzels me is actually the high min latency of more than 24us:
> 
>  == Sampling period: 100 us
> == Test mode: periodic user-mode task
> == All results in microseconds
> warming up...
> RTT|  00:00:01  (periodic user-mode task, 100 us period, priority 99)
> RTH|--lat min|---lat avg|-----lat max|-overrun|----lat best|---lat worst
> RTD|   24.304|    35.199|      65.371|       0|      24.304|      65.371
> 
> Can the TSC cause such high delays?
>

You mean TSC emulation? Definitely yes for the worst-case. Almost every
rthal_get_8254_tsc() costs about 3 us, and I counted >15 in Roland's
critical trace path.

> And what is the output of "/proc/xenomai/latency"?

Indeed interesting. Maybe the system is not well tuned. But the overhead
of rthal_get_8254_tsc() will remain for the non-minimal case (e.g.
host-tick hitting before or after task-wakeup tick).

Jan

Attachment: signature.asc
Description: OpenPGP digital signature

_______________________________________________
Xenomai-help mailing list
[email protected]
https://mail.gna.org/listinfo/xenomai-help

Reply via email to