lkml.org 
[lkml]   [2004]   [Sep]   [10]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRe: voluntary-preemption: understanding latency trace
From
Date
On Fri, 2004-09-10 at 02:37, Ingo Molnar wrote:
> * Lee Revell <rlrevell@joe-job.com> wrote:
>
> > > I've got a SCHED_FIFO kernel thread at the highest priority
> > > (MAX_USER_RT_PRIO-1) and it's sleeping on a wait queue. The wake is
> > > called from an ISR. Since this thread is the highest priority in the
> > > system, I expect it to run before the ISR threads and softIRQ threads
> > > etc.
> > >
> > > In the ISR I sample sched_clock() just before the call to wake_up()
> > > and in the thread I sample sched_clock() again just after the call to
> > > sleep. I'm seeing an almost 4ms latency between the call to wake_up
> > > and the actual wakeup. However, in /proc/latency_trace, the worst
> > > latency I see during the running of this test is <500us.
>
> > Ingo, any ideas here? Looks like maybe the use of sched_clock is the
> > problem.
>
> sched_clock() is not 100% accurate (it takes a few shortcuts to avoid a
> division) but it should be better than 90% so 4 msec measured means
> there's likely some big delay.
>
> if the priority setup is indeed as described above then the RT task
> should have run much faster. First i'd suggest to check whether it's not
> console printing (printing of a stacktrace or a latency trace) that
> slows things down.
>

Ah, this is probably it, this is pretty close to the latency I get when
/proc/latency_trace is updated, and this is also the one latency that
doesn't show in the traces by design.

rlrevell@mindpipe:~$ ./amlat-rlr/amlat
599.895 MHz
secondsPerTick=0.000000
ticksPerSecond=599894954.372806
599.895 MHz
Using rtc interval of 1024
u=0
latency = 53 microseconds
latency = 60 microseconds
latency = 60 microseconds
latency = 62 microseconds
latency = 66 microseconds
latency = 76 microseconds
latency = 78 microseconds
latency = 2548 microseconds

The last line is the latency trace being updated. If I turn off tracing
or set preempt_max_latency to a high value then amlat doesn't show
these.

Lee


-
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/

\
 
 \ /
  Last update: 2005-03-22 14:06    [W:0.219 / U:0.032 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site