| Message ID | 1588922717-63697-3-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 08-05-20 15:25:16, Feng Tang wrote: > percpu_counter_sum_positive() will provide more accurate info. Why do we need that? > 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 two users: /proc/meminfo and HV ballon > driver's status trace per second. OK, this explains that the additional overhead is not terrible but there is no actual information on why the imprecision is a problem. > Signed-off-by: Feng Tang <feng.tang@intel.com> > --- > mm/util.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/mm/util.c b/mm/util.c > index 988d11e..3de78e9 100644 > --- a/mm/util.c > +++ b/mm/util.c > @@ -774,7 +774,7 @@ 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
On Fri, May 15, 2020 at 03:23:07PM +0800, Michal Hocko wrote: > On Fri 08-05-20 15:25:16, Feng Tang wrote: > > percpu_counter_sum_positive() will provide more accurate info. > > Why do we need that? This is a preparation for patch 3/3, which will enlarge the batch size of percpu-counter 'vm_committed_as'. Andi Kleen concerned that may hurt the accuracy for those readers (currently only /proc/meminfo and hyperV balloon drivers for status check) As in worst case, the deviation could be 'batch * nr_cpus', and with patch 3/3, that could be totalram_pages/16. Thanks, Feng > > 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 two users: /proc/meminfo and HV ballon > > driver's status trace per second. > > OK, this explains that the additional overhead is not terrible but there > is no actual information on why the imprecision is a problem. > > > Signed-off-by: Feng Tang <feng.tang@intel.com> > > --- > > mm/util.c | 2 +- > > 1 file changed, 1 insertion(+), 1 deletion(-) > > > > diff --git a/mm/util.c b/mm/util.c > > index 988d11e..3de78e9 100644 > > --- a/mm/util.c > > +++ b/mm/util.c > > @@ -774,7 +774,7 @@ 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 > > -- > Michal Hocko > SUSE Labs
On Fri 15-05-20 16:11:09, Feng Tang wrote: > On Fri, May 15, 2020 at 03:23:07PM +0800, Michal Hocko wrote: > > On Fri 08-05-20 15:25:16, Feng Tang wrote: > > > percpu_counter_sum_positive() will provide more accurate info. > > > > Why do we need that? > > This is a preparation for patch 3/3, which will enlarge the batch size > of percpu-counter 'vm_committed_as'. Andi Kleen concerned that may > hurt the accuracy for those readers (currently only /proc/meminfo > and hyperV balloon drivers for status check) > > As in worst case, the deviation could be 'batch * nr_cpus', and > with patch 3/3, that could be totalram_pages/16. This information belongs to the changelog.
On Fri, May 15, 2020 at 05:04:40PM +0800, Michal Hocko wrote: > On Fri 15-05-20 16:11:09, Feng Tang wrote: > > On Fri, May 15, 2020 at 03:23:07PM +0800, Michal Hocko wrote: > > > On Fri 08-05-20 15:25:16, Feng Tang wrote: > > > > percpu_counter_sum_positive() will provide more accurate info. > > > > > > Why do we need that? > > > > This is a preparation for patch 3/3, which will enlarge the batch size > > of percpu-counter 'vm_committed_as'. Andi Kleen concerned that may > > hurt the accuracy for those readers (currently only /proc/meminfo > > and hyperV balloon drivers for status check) > > > > As in worst case, the deviation could be 'batch * nr_cpus', and > > with patch 3/3, that could be totalram_pages/16. > > This information belongs to the changelog. Will add it in v3, thanks - Feng
diff --git a/mm/util.c b/mm/util.c index 988d11e..3de78e9 100644 --- a/mm/util.c +++ b/mm/util.c @@ -774,7 +774,7 @@ 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. 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 two users: /proc/meminfo and HV ballon 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(-)