Re: [PATCH V5 6/6] proc: show MADV_FREE pages info in smaps

[Minchan Kim]

Hi Andrew,

On Tue, Mar 07, 2017 at 02:43:38PM -0800, Andrew Morton wrote:
> On Tue, 7 Mar 2017 11:05:45 +0100 Michal Hocko <mhocko@xxxxxxxxxx> wrote:
> 
> > On Fri 03-03-17 16:10:27, Andrew Morton wrote:
> > > On Thu, 2 Mar 2017 17:30:54 +0100 Michal Hocko <mhocko@xxxxxxxxxx> wrote:
> > > 
> > > > > It's not that I think you're wrong: it *is* an implementation detail.
> > > > > But we take a bit of incoherency from batching all over the place, so
> > > > > it's a little odd to take a stand over this particular instance of it
> > > > > - whether demanding that it'd be fixed, or be documented, which would
> > > > > only suggest to users that this is special when it really isn't etc.
> > > > 
> > > > I am not aware of other counter printed in smaps that would suffer from
> > > > the same problem, but I haven't checked too deeply so I might be wrong. 
> > > > 
> > > > Anyway it seems that I am alone in my position so I will not insist.
> > > > If we have any bug report then we can still fix it.
> > > 
> > > A single lru_add_drain_all() right at the top level (in smaps_show()?)
> > > won't kill us
> > 
> > I do not think we want to put lru_add_drain_all cost to a random
> > process reading /proc/<pid>/smaps.
> 
> Why not?  It's that process which is calling for the work to be done.
> 
> > If anything the one which does the
> > madvise should be doing this.
> 
> But it would be silly to do extra work in madvise() if nobody will be
> reading smaps for the next two months.
> 
> How much work is it anyway?  What would be the relative impact upon a
> smaps read?

I agree only if the draining guarantees all of mapped pages in the range
could be marked to lazyfree. However, it's not true because there are a
few of logics to skip the page marking in madvise_free_pte_range.

So, my conclusion is drainning helps a bit but not gaurantees.
In such case, IMHO, let's not do the effort to make better.

Thanks.