latency logger for realtime-preempt-2.6.12-final-V0.7.51-30
Quoted 9 lines
Quoted 5 lines
Hi, Ingo
This patch can record interrupt-off latency, preemption-off latency and wakeup latency in a big history array, in the meanwhile, it dummies up printks produced by these latency timing.
looks pretty good! I'll look at merging your patch after KS/OLS.
Ingo
Hi, ingo
Do you have any trouble while you merge that latency logger patch?
Quoted 3 lines · Yang Yi
Quoted 1 line
looks pretty good! I'll look at merging your patch after KS/OLS.
Do you have any trouble while you merge that latency logger patch?
i've merged it to the -52-11 patch that, and i've uploaded it a couple of minutes ago.
i have done a number of cleanups on it - e.g. instead of latency logging it's now called latency histogram, and i've also fixed a number of coding style issues. Please double-check that it's still OK, seems to work here.
would be nice to clean up the impact of the latency-histogram code some more though: e.g. the #ifdef jungle check_critical_timing() is disgusting. Could be cleaned up by always recording the latency_type being currently traced into a new tr->latency_type field, and then using that in check_critical_timing().
Ingo
Quoted 5 lines · Ingo Molnar
would be nice to clean up the impact of the latency-histogram code some more though: e.g. the #ifdef jungle
check_critical_timing()is disgusting. Could be cleaned up by always recording thelatency_typebeing currently traced into a newtr->latency_typefield, and then using that incheck_critical_timing().
the code appears to be adding ifdefs to make preempt_max_latency work like preempt_threshold , I think all the nasty ifdefs could go away if we just made it use preempt_threshold .. Plus it logs latency during boot up cause preempt_max_latency equals zero .
Daniel
Quoted 8 lines · Daniel Walker
Quoted 5 lines · Ingo Molnar
would be nice to clean up the impact of the latency-histogram code some more though: e.g. the #ifdef jungle
check_critical_timing()is disgusting. Could be cleaned up by always recording thelatency_typebeing currently traced into a newtr->latency_typefield, and then using that incheck_critical_timing().
Quoted 4 lines
the code appears to be adding ifdefs to make
preempt_max_latencywork likepreempt_threshold, I think all the nasty ifdefs could go away if we just made it usepreempt_threshold.. Plus it logs latency during boot up causepreempt_max_latencyequals zero .
yes. Would be nice to have these cleanups too. The #ifdef impact of the LATENCY_HIST code is quite high at the moment, and it needs to be decreased.
Ingo