Messages in this thread | | | Date | Sat, 3 Oct 2015 10:52:32 +0200 | From | Ingo Molnar <> | Subject | Re: 4.2.2: NR_CPUS effectively being 1 bug |
| |
* Mike Galbraith <umgwanakikbuti@gmail.com> wrote:
> On Sat, 2015-10-03 at 10:18 +0200, Ingo Molnar wrote: > > * Mike Galbraith <umgwanakikbuti@gmail.com> wrote: > > > > > On Sat, 2015-10-03 at 09:38 +0200, Ingo Molnar wrote: > > > > * Peter Zijlstra <peterz@infradead.org> wrote: > > > > > > > > > On Fri, Oct 02, 2015 at 02:00:33PM -0700, Andrew Morton wrote: > > > > > > On Fri, 2 Oct 2015 23:46:59 +0300 Alexey Dobriyan <adobriyan@gmail.com> wrote: > > > > > > > > > > > > > FYI, I've upgraded from 4.1.7 to 4.2.1 (and retested with 4.2.2) and > > > > > > > everything is scheduled on 1 CPU out of 4 (i5 760). > > > > > > > > > > > > > > $ sudo cat /proc/1/status | grep cpu -i > > > > > > > Cpus_allowed: 1 > > > > > > > Cpus_allowed_list: 0 > > > > > > > > > > > > > > Every process inherits this tiny cpumask. > > > > > > > > > > > > Sell the other CPUs on ebay? > > > > > > > > > > > > I haven't seen such a report before - maybe it rings a bell with Peter > > > > > > & Ingo? > > > > > > > > > > I think this is related to some NO_HZ_FULL quackery. People seem to have > > > > > enabled stuff they've really no sane reason for. > > > > > > > > So the question is, is CONFIG_NO_HZ_FULL_ALL=y enabled? If yes then please disable > > > > it. > > > > > > > > Frederic, is there a fix for that? The Kconfig help text for CONFIG_NO_HZ_FULL_ALL > > > > says:: > > > > > > > > CONFIG_NO_HZ_FULL_ALL: > > > > > > > > If the user doesn't pass the nohz_full boot option to > > > > define the range of full dynticks CPUs, consider that all > > > > CPUs in the system are full dynticks by default. > > > > Note the boot CPU will still be kept outside the range to > > > > handle the timekeeping duty. > > > > > > > > I can see people enabling that. Why are all CPUs lost if it's done? > > > > > > Simple. Rik made it such that cpu_isolated_map is immune to cpusets, > > > and Chris made tick_nohz_full_mask automatically set cpu_isolated_map, > > > so now if nohz_full is ever turned on, that CPU is gone from the generic > > > pool forever, with obvious consequences for CONFIG_NO_HZ_FULL_ALL. > > > > > > Lovely eh? > > > > Which commit is that? Needs to be reverted... > > 8cb9764f nohz: Set isolcpus when nohz_full is set > > That's the one that makes them all automatically go away.
Ok, looks like a patch with good intentions but bad effects. Mind sending a revert patch, changelogged, signed off?
Thanks,
Ingo
| |