Re: [PATCH] Hugetlb pages rss accounting is incorrect in /proc/<pid>/smaps

[Naoya Horiguchi]

On Wed, Oct 25, 2017 at 08:55:27AM +0200, Michal Hocko wrote:
> [CCing Naoya]
> 
> On Tue 24-10-17 16:56:08, Prakash Sangappa wrote:
> > Resident set size(Rss) accounting of hugetlb pages is not done
> > currently in /proc/<pid>/smaps. The pmap command reads rss from
> > this file and so it shows Rss to be 0 in pmap -x output for
> > hugetlb mapped vmas. This patch fixes it.
> 
> We do not account in rss because we do have a dedicated counters
> depending on whether the hugetlb page is mapped privately or it is
> shared. The reason this is not in RSS IIRC is that a large unexpected
> RSS from hugetlb pages might confuse system monitors.

Yes, that was the intention of separate counters for hugetlb.

> This is one of
> those backward compatibility issues when you start accounting something
> too late.

So new monitoring applications are supposed to check the new counters
to track hugetlb usages.

Thanks,
Naoya Horiguchi


> 
> > Signed-off-by: Prakash Sangappa <prakash.sangappa@xxxxxxxxxx>
> > ---
> >  fs/proc/task_mmu.c | 1 +
> >  1 file changed, 1 insertion(+)
> > 
> > diff --git a/fs/proc/task_mmu.c b/fs/proc/task_mmu.c
> > index 5589b4b..c7e1048 100644
> > --- a/fs/proc/task_mmu.c
> > +++ b/fs/proc/task_mmu.c
> > @@ -724,6 +724,7 @@ static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
> >  			mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
> >  		else
> >  			mss->private_hugetlb += huge_page_size(hstate_vma(vma));
> > +		mss->resident += huge_page_size(hstate_vma(vma));
> >  	}
> >  	return 0;
> >  }
> > -- 
> > 2.7.4
> > 
> 
> -- 
> Michal Hocko
> SUSE Labs
> 
> --
> To unsubscribe, send a message with 'unsubscribe linux-mm' in
> the body to majordomo@xxxxxxxxxx  For more info on Linux MM,
> see: http://www.linux-mm.org/ .
> Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>