Messages in this thread | | | Date | Mon, 29 Nov 2004 15:33:16 +0100 | From | Ingo Molnar <> | Subject | Re: Real-Time Preemption, -RT-2.6.10-rc2-mm3-V0.7.31-7 |
| |
* Rui Nuno Capela <rncbc@rncbc.org> wrote:
> > the trace buffer is too small to capture a significant portion of the > > xrun - i'd suggest for you to increase it from 4096-1 to 4096*16-1, to > > be able to capture a couple of hundreds of millisecs worth of traces. > > > > and how I do that? Is it some /proc magic or its in the kernel configuration?
it's the MAX_TRACE define in kernel/latency.c.
but please try to the -31-10 kernel that i've just uploaded, it has a number of tracer enhancements:
- process name tracking. The new format is:
bash- 3633 80000003 0.264ms (+0.000ms): idle_cpu (wake_idle) bash- 3633 80000003 0.264ms (+0.000ms): idle_cpu (wake_idle) bash- 3633 80000003 0.264ms (+0.000ms): find_next_bit (wake_idle)
this makes it easier to identify which process does what. This feature has no significant overhead in the tracer itself, all the hard work is done when /proc/latency_trace is read by the user.
- /proc/sys/kernel/mcount_enabled flag: if disabled then /proc/latency_trace will only contain 'custom' events, but no per-function entries. This can be useful to trace really long latencies, to get a rough idea of what is going on.
- /proc/latency_trace atomicity. It was fundamentally non-atomic, due to it being a 4K-granular file interface. Now the kernel has a second layer of saved-trace logic, which makes sure that the only time the trace is switched is when the header of the trace is read. I.e. when cp-ing or cat-ing /proc/latency_trace, no new trace info will be used. This change could solve some of the 'truncated traces' weirdnesses reported.
Ingo - 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/
| |