| Message ID | 1590714370-67182-3-git-send-email-feng.tang@intel.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | make vm_committed_as_batch aware of vm overcommit policy | expand |
On Fri 29-05-20 09:06:08, Feng Tang wrote: > percpu_counter_sum_positive() will provide more accurate info. > > As with percpu_counter_read_positive(), in worst case the deviation > could be 'batch * nr_cpus', which is totalram_pages/256 for now, > and will be more when the batch gets enlarged. > > Its time cost is about 800 nanoseconds on a 2C/4T platform and > 2~3 microseconds on a 2S/36C/72T server in normal case, and in > worst case where vm_committed_as's spinlock is under severe > contention, it costs 30~40 microseconds for the 2S/36C/72T sever, > which should be fine for its only two users: /proc/meminfo and > HyperV balloon driver's status trace per second. > > Signed-off-by: Feng Tang <feng.tang@intel.com> I cannot speak for HyperV part. Cc maintainers but this shouldn't be a problem for meminfo. Acked-by: Michal Hocko <mhocko@suse.com> > --- > mm/util.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/mm/util.c b/mm/util.c > index 9b3be03..3c7a08c 100644 > --- a/mm/util.c > +++ b/mm/util.c > @@ -790,7 +790,7 @@ struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp; > */ > unsigned long vm_memory_committed(void) > { > - return percpu_counter_read_positive(&vm_committed_as); > + return percpu_counter_sum_positive(&vm_committed_as); > } > EXPORT_SYMBOL_GPL(vm_memory_committed); > > -- > 2.7.4 >
> Its time cost is about 800 nanoseconds on a 2C/4T platform and > 2~3 microseconds on a 2S/36C/72T server in normal case, and in > worst case where vm_committed_as's spinlock is under severe > contention, it costs 30~40 microseconds for the 2S/36C/72T sever, This will be likely 40-80us on larger systems, although the overhead is often non linear so it might get worse. > which should be fine for its only two users: /proc/meminfo and > HyperV balloon driver's status trace per second. There are some setups who do frequent sampling of /proc/meminfo in the background. Increased overhead could be a problem for them. But not proposing a change now. If someone complains have to revisit I guess, perhaps adding a rate limit of some sort. -Andi
On Wed, Jun 03, 2020 at 07:28:53AM -0700, Andi Kleen wrote: > > Its time cost is about 800 nanoseconds on a 2C/4T platform and > > 2~3 microseconds on a 2S/36C/72T server in normal case, and in > > worst case where vm_committed_as's spinlock is under severe > > contention, it costs 30~40 microseconds for the 2S/36C/72T sever, > > This will be likely 40-80us on larger systems, although the overhead > is often non linear so it might get worse. > > > which should be fine for its only two users: /proc/meminfo and > > HyperV balloon driver's status trace per second. > > There are some setups who do frequent sampling of /proc/meminfo > in the background. Increased overhead could be a problem for them. > But not proposing a change now. If someone complains have to > revisit I guess, perhaps adding a rate limit of some sort. Agree. Maybe I should also put the time cost info into the code comments in case someone noticed the slowdown. Thanks, Feng > > -Andi
diff --git a/mm/util.c b/mm/util.c index 9b3be03..3c7a08c 100644 --- a/mm/util.c +++ b/mm/util.c @@ -790,7 +790,7 @@ struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp; */ unsigned long vm_memory_committed(void) { - return percpu_counter_read_positive(&vm_committed_as); + return percpu_counter_sum_positive(&vm_committed_as); } EXPORT_SYMBOL_GPL(vm_memory_committed);
percpu_counter_sum_positive() will provide more accurate info. As with percpu_counter_read_positive(), in worst case the deviation could be 'batch * nr_cpus', which is totalram_pages/256 for now, and will be more when the batch gets enlarged. Its time cost is about 800 nanoseconds on a 2C/4T platform and 2~3 microseconds on a 2S/36C/72T server in normal case, and in worst case where vm_committed_as's spinlock is under severe contention, it costs 30~40 microseconds for the 2S/36C/72T sever, which should be fine for its only two users: /proc/meminfo and HyperV balloon driver's status trace per second. Signed-off-by: Feng Tang <feng.tang@intel.com> --- mm/util.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)