Re: [PATCH 4.4 50/92] mm: filemap: avoid unnecessary calls to lock_page when waiting for IO to complete during a read

[Hugh Dickins]

On Thu, May 24, 2018 at 12:06 PM Greg KH <gregkh@xxxxxxxxxxxxxxxxxxx> wrote:
> On Thu, May 24, 2018 at 10:27:59AM -0700, Hugh Dickins wrote:
> > On Thu, May 24, 2018 at 4:28 AM Greg KH <gregkh@xxxxxxxxxxxxxxxxxxx>
wrote:
> > > On Thu, May 24, 2018 at 04:17:12AM -0700, Hugh Dickins wrote:
> > > >   Thu, May 24, 2018 at 4:06 AM Greg Kroah-Hartman
> > > > <gregkh@xxxxxxxxxxxxxxxxxxx>
> > > > wrote:
> > > >
> > > > > On Thu, May 24, 2018 at 12:50:11PM +0200, Jan Kara wrote:
> > > > > > On Thu 24-05-18 11:38:27, Greg Kroah-Hartman wrote:
> > > > > > > 4.4-stable review patch.  If anyone has any objections, please
> > let me
> > > > know.
> > > > > >
> > > > > > Just one objection: Why does stable care about this (and the
> > previous
> > > > > > patch)? I've checked the stable queue and I don't see anything
that
> > > > would
> > > > > > have these patches as a prerequisite. And on their own, they are
> > only
> > > > > > cleanups without substantial gains.
> > > >
> > > > > There's a small gain here:
> > > >
> > > > > > > paralleldd
> > > > > > >                                     4.4.0
4.4.0
> > > > > > >                                   vanilla
avoidlock
> > > > > > > Amean    Elapsd-1          5.28 (  0.00%)        5.15 (
  2.50%)
> > > > > > > Amean    Elapsd-4          5.29 (  0.00%)        5.17 (
  2.12%)
> > > > > > > Amean    Elapsd-7          5.28 (  0.00%)        5.18 (
  1.78%)
> > > > > > > Amean    Elapsd-12         5.20 (  0.00%)        5.33 (
-2.50%)
> > > > > > > Amean    Elapsd-21         5.14 (  0.00%)        5.21 (
-1.41%)
> > > > > > > Amean    Elapsd-30         5.30 (  0.00%)        5.12 (
  3.38%)
> > > > > > > Amean    Elapsd-48         5.78 (  0.00%)        5.42 (
  6.21%)
> > > > > > > Amean    Elapsd-79         6.78 (  0.00%)        6.62 (
  2.46%)
> > > > > > > Amean    Elapsd-110        9.09 (  0.00%)        8.99 (
  1.15%)
> > > > > > > Amean    Elapsd-128       10.60 (  0.00%)       10.43 (
  1.66%)
> > > > > > >
> > > > > > > The impact is small but intuitively, it makes sense to avoid
> > > > unnecessary
> > > > > > > calls to lock_page.
> > > >
> > > > > Yes, it's small, but it's marked in the SLES kernel as "needs to
be
> > > > > merged into stable", so obviously it matters to someone :)
> > > >
> > > > Hmm. I had the same reaction to these two as Jan, but assumed that
they
> > > > made applying later patches easier, and didn't take the trouble he
did
> > to
> > > > find that's not so.
> > > >
> > > > I've no wish to be disputatious, but it does seem that the
definition of
> > > > "stable" has changed, and not necessarily for the better, if it's
now a
> > > > home for small gains: I thought we left those to upstream.
> >
> > > This is in the SLES kernel for a reason, and again, it's in the
section
> > > that says "this should be pushed to stable".  So if it's good enough
for
> > > the SLES kernel, why isn't it good enough for all users of this kernel
> > > tree?
> >
> > > If you all think it should be dropped in both places, that's fine with
> > > me :)
> >
> > I think they are perfectly fine in SLES: folding in good work is a part
of
> > what distros are about.

> And it's also what stable is for.  We have had backports of performance
> improvements in the past, along with lots of other things over the
> years.  This is a performance improvement.  A tiny one, yes, but getting
> rid of a lock is a good thing, and I picked it up as part of my review
> of what a distro decided was worth adding for their users, as that's a
> huge signal that might be of value to others.

> > But I cannot find anything in stable-kernel-rules.rst that would admit
them
> > - perhaps that's just out of date?

> Nope, that's the list I use to say "no" to.  You can't describe
> everything in that file, it's a judgement call.

> > If -stable is to be a compendium of "this looks nice, you might like to
> > include it", so be it: but the rules should then be updated.

> This is a "a bunch of people I trust took it in their kernel, and it's
> been running on zillion of machines for a while and causes no harm and a
> slight benefit, so let's add it!" type of thing.  It's not the only
> patch in this series that was like that, but for some reason this one is
> the one the triggered the debate, which is funny to me as this does have
> numbers in it showing that it is an improvement :)

Thank you for looking after the -stable trees: please let me not waste your
time any further. I have no specific objection to the two patches, which
are certainly not egregious offenders. But I do still find the disconnect
between stable-kernel-rules.rst and reality confusing - or perhaps I just
find reality confusing :)

Hugh