Messages in this thread | | | Date | Wed, 7 Oct 2009 00:30:37 -0700 | Subject | Re: [BISECTED] "conservative" cpufreq governor broken | From | Steven Noonan <> |
| |
On Tue, Oct 6, 2009 at 10:27 PM, Eero Nurkkala <ext-eero.nurkkala@nokia.com> wrote: > On Wed, 2009-10-07 at 03:22 +0200, ext Steven Noonan wrote: >> >> >> >> Tracers may by themselves add some load into the system. >> >> >> >> If I were you, I'd add something like: (I have only one CPU BTW) >> >> >> >> static ktime_t time_prior_idle; >> >> static int64_t idle_total; >> >> >> >> time_prior_idle = ktime_get(); >> >> <idle stuff> >> >> idle_total += ktime_to_ns(ktime_sub(ktime_get(), time_prior_idle)); >> >> >> >> and have a sysfs hook (something already present, so you can just cat >> >> it) with a trace: >> >> >> >> printk("Times: %lld, %lld \n", idle_total, ktime_to_ns(ktime_get())); >> >> >> >> Sample output: >> >> 374758812519, 386768249832 >> >> >> >> So I have 386768249832 / 386768249832 = 96.9 % time spent in idle in >> >> this case. >> >> >> >> (Right, this should provide somewhat descent info, hopefully ;) ) >> >> >> > >> > Well, I tried adding the code to cpu_idle() as suggested, but it never >> > printed anything. Apparently cpu_idle() isn't ever being called here. >> > Even added a 'BUG();' at the beginning of the function and it never >> > hit it. Of course, I'm probably missing something obvious. Is there a >> > separate cpu_idle()-esque function for SMP? >> > >> >> Oh crap. Perhaps it's more insidious. I reverted the bisected commit >> and it _DID_ hit the line I added. So cpu_idle is never entered with >> the bisected commit. Bizarre. >> >> - Steven > > So you're saying that cpu_idle() is not ever called.. The commit in > question - whether you got it or had it reverted - should, and must not > be a trigger for your system to hit cpu_idle(). What if you took a look > at init/main.c, and added traces before the first call to cpu_idle(): > > start_kernel() -> > rest_init() -> > cpu_idle()
Okay, wow, I'm a moron. I misread what cpu_idle() was intended to be for. I thought that cpu_idle() was a function that was periodically called whenever the CPU had nothing to do, but now I see that it's actually the main loop. I should really read the code next time.
I've moved the statistics printout code to the _inside_ of that infinite loop and retested. I had it print every several hundred iterations. Here's the results (note the machine was idle the whole time, except for about the first 10-20 seconds while the machine booted):
[ 3.627716] timings[0]: 2250511125 / 3627716116 [ 6.946216] timings[0]: 4780901366 / 6946213531 [ 13.355182] timings[0]: 9385417604 / 13355183525 [ 18.551304] timings[1]: 16300853077 / 18551301189 [ 21.589039] timings[0]: 15984495433 / 21589037480 [ 47.152733] timings[1]: 44386121538 / 47152731476 [ 51.682630] timings[0]: 45713834076 / 51682628295 [ 79.587359] timings[0]: 73524821916 / 79587356820 [ 88.630110] timings[1]: 85324277596 / 88630109605 [ 96.082386] timings[0]: 89691306072 / 96082384539
> > With the patch not reverted? > > So I guess, that an irq may actually start your "system up" (patch > reverted) somehow: > > irq_exit() -> > tick_nohz_stop_sched_tick() -> > tick_nohz_start_idle() -> > sched_clock_idle_sleep_event() ? > > So it looks that maybe the conservative governor isn't broken after > all ;)
Symptomatically, it is broken. But it's likely just that: a symptom of a larger issue.
> > > (BTW I guess, with SMP, you may wish to have fix things to work in your > setup > static ktime_t time_prior_idle[2]; > static int64_t idle_total[2]; > and > time_prior_idle[smp_processor_id()] = ktime_get(); > etc, and print out times separately). > > - Eero > >
| |