Re: [PATCH V3 1/1] mm: add swapiness= arg to memory.reclaim

[Dan Schatzberg]

On Tue, Dec 12, 2023 at 01:31:46PM -0800, Yosry Ahmed wrote:
> On Tue, Dec 12, 2023 at 1:27 PM Dan Schatzberg <schatzberg.dan@xxxxxxxxx> wrote:
> >
> > On Mon, Dec 11, 2023 at 11:41:24AM -0800, Yosry Ahmed wrote:
> > > On Mon, Dec 11, 2023 at 6:04 AM Dan Schatzberg <schatzberg.dan@xxxxxxxxx> wrote:
> > >
> > > contains* the*
> > >
> > > I think this statement was only important because no keys were
> > > supported, so I think we can remove it completely and rely on
> > > documenting the supported keys below like other interfaces, see my
> > > next comment.
> > >
> > > > +       to reclaim.
> > > >
> > > >         Example::
> > > >
> > > > @@ -1304,6 +1304,17 @@ PAGE_SIZE multiple when read back.
> > > >         This means that the networking layer will not adapt based on
> > > >         reclaim induced by memory.reclaim.
> > > >
> > > > +       This file also allows the user to specify the swappiness value
> > > > +       to be used for the reclaim. For example:
> > > > +
> > > > +         echo "1G swappiness=60" > memory.reclaim
> > > > +
> > > > +       The above instructs the kernel to perform the reclaim with
> > > > +       a swappiness value of 60. Note that this has the same semantics
> > > > +       as the vm.swappiness sysctl - it sets the relative IO cost of
> > > > +       reclaiming anon vs file memory but does not allow for reclaiming
> > > > +       specific amounts of anon or file memory.
> > > > +
> > >
> > > Can we instead follow the same format used by other nested-keyed files
> > > (e.g. io.max)? This usually involves a table of supported keys and
> > > such.
> >
> > Thanks, both are good suggestions. Will address these.
> >
> > > > +       while ((start = strsep(&buf, " ")) != NULL) {
> > > > +               if (!strlen(start))
> > > > +                       continue;
> > > > +               switch (match_token(start, if_tokens, args)) {
> > > > +               case MEMORY_RECLAIM_SWAPPINESS:
> > > > +                       if (match_int(&args[0], &swappiness))
> > > > +                               return -EINVAL;
> > > > +                       if (swappiness < 0 || swappiness > 200)
> > >
> > > I am not a fan of extending the hardcoded 0 and 200 values, and now
> > > the new -1 value. Maybe it's time to create constants for the min and
> > > max swappiness values instead of hardcoding them everywhere? This can
> > > be a separate preparatory patch. Then, -1 (or any invalid value) can
> > > also be added as a constant with a useful name, instead of passing -1
> > > to all other callers.
> > >
> > > This should make the code a little bit more readable and easier to extend.
> >
> > I'm not sure I understand the concern. This check just validates that
> > the swappiness value inputted is between 0 and 200 (inclusive)
> > otherwise the interface returns -EINVAL. Are you just concerned that
> > these constants are not named explicitly so they can be reused
> > elsewhere in the code?
> 
> Yes. The 0 and 200 values are already hardcoded in multiple places,
> and we are adding more places now and more hardcoded values (i.e. -1).

Understood, I'll add a preparatory patch which adds DEFINEs for
MIN_SWAPPINESS and MAX_SWAPPINESS and change the usages of 0 and 200
to those. I'll also eliminate the use of -1 as Chris suggested.