Re: [RFC] Track mlock()ed pages

[Andrew Morton]

On Fri, 26 Jan 2007 10:23:44 -0800 (PST)
Christoph Lameter <clameter@xxxxxxx> wrote:

> On Fri, 26 Jan 2007, Andrew Morton wrote:
> 
> > > Large amounts of mlocked pages may be a problem for 
> > > 
> > > 1. Reclaim behavior.
> > > 
> > > 2. Defragmentation
> > > 
> > 
> > We know that.  What has that to do with this patch?
> 
> Knowing how much mlocked pages are where is necessary to solve these 
> issues.

If we continue this dialogue for long enough, we'll actually have a changlog.

> > > > You could perhaps go for a walk across all the other vmas which presently
> > > > map this page.  If any of them have VM_LOCKED, don't increment the counter.
> > > > Similar on removal: only decrement the counter when the final mlocked VMA
> > > > is dropping the pte.
> > > 
> > > For that we would need an additional refcount for vmlocked maps in the 
> > > page struct.
> > 
> > No you don't.  The refcount is already there.  It is "the sum of the VM_LOCKED
> > VMAs which map this page".
> > 
> > It might be impractical or expensive to calculate it, but it's there.
> 
> Correct. Its so expensive that it cannot be used to build vm stats for 
> mlocked pages. F.e. Determination of the final mlocked VMA dropping the 
> page would require a scan over all vmas mapping the page.

Of course it would.  But how do you know it is "too expensive"?  We "scan
all the vmas mapping a page" as a matter of course in the page scanner -
millions of times a minute.  If that's "too expensive" then ouch.

That, plus if we have so many vmas mapping a page for this effect to
matter, then your change as proposed will be so inaccurate as to be
useless, no?
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/