On Wed, Nov 29, 2023 at 2:31 PM Yosry Ahmed <yosryahmed@xxxxxxxxxx> wrote:
>
> On Wed, Nov 29, 2023 at 2:29 PM Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> wrote:
> >
> > On Wed, 29 Nov 2023 14:18:15 -0800 Yosry Ahmed <yosryahmed@xxxxxxxxxx> wrote:
> >
> > > On Wed, Nov 29, 2023 at 1:53 PM Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> wrote:
> > > >
> > > > On Wed, 29 Nov 2023 13:43:13 -0800 Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> wrote:
> > > >
> > > > > On Wed, 29 Nov 2023 23:42:11 +0800 kernel test robot <lkp@xxxxxxxxx> wrote:
> > > > >
> > > > > > >> mm/zswap.c:825:17: error: dereferencing pointer to incomplete type 'struct mem_cgroup'
> > > > > > css_get(&memcg->css);
> > > > > > ^~
> > > > >
> > > > > OK, thanks, patchset needs work for CONFIG_MEMCG=n. I'll drop this version.
> > > >
> > > > Well that's annoying - the "mm: memcg: subtree stats flushing and
> > > > thresholds" series had lots of dependencies on this series.
> > >
> > > FWIW, the "mm: memcg: subtree stats flushing and thresholds" series
> > > has no actual dependency on the zswap series. The conflicts come from
> > > patch 2, which moves some code in mm/memcontrol.c, which happens to be
> > > touched by the zswap series. The first 2 patches of the stats series
> > > are just refactoring with no functional changes, so if those two can
> > > remain in mm-unstable, independent of the remaining of the series or
> > > the zswap series, then either Nhat or I could easily rebased our
> > > patches on top of them, and you can easily move the series in & out of
> > > mm-unstable without conflicts.
> > >
> > > Another thing is, the only difference between v3 & v4 of the "mm:
> > > memcg: subtree stats flushing and thresholds" series is the rebase on
> > > top of the zswap series. So if you want, you can take both series out,
> > > and add in v3 of the stats series instead of v4. If you need to remove
> > > the stast series again in the future, you can leave the first two
> > > patches to avoid conflicts with the zswap series.
> > >
> >
> > Ok, thanks. I prefer not to make what is in mm.git too different from
> > what was sent.
> >
> > I've left everything in place:
> >
> > list_lru-allows-explicit-memcg-and-numa-node-selection.patch
> > memcontrol-add-a-new-function-to-traverse-online-only-memcg-hierarchy.patch
> > zswap-make-shrinking-memcg-aware.patch
> > zswap-make-shrinking-memcg-aware-fix.patch
> > mm-memcg-add-per-memcg-zswap-writeback-stat.patch
> > selftests-cgroup-update-per-memcg-zswap-writeback-selftest.patch
> > zswap-shrinks-zswap-pool-based-on-memory-pressure.patch
> > #
> > ...
> > #
> > mm-memcg-change-flush_next_time-to-flush_last_time.patch
> > mm-memcg-move-vmstats-structs-definition-above-flushing-code.patch
> > mm-memcg-make-stats-flushing-threshold-per-memcg.patch
> > mm-workingset-move-the-stats-flush-into-workingset_test_recent.patch
> > mm-memcg-restore-subtree-stats-flushing.patch
> >
> > in the hope that a new version of the first series ("workload-specific
> > and memory pressure-driven zswap writeback") can be dropped in place of
> > the old version.
> >
>
> I see. So Nhat needs to *not* rebase his patches on top of the current
> mm-unstable, right?
FWIW, I'll have to re-send a v8 to fix the kernel test robots/build
issues that I've missed and incorporate the comments/suggestions from
Michal and Johannes. IIUC, your series is ready right? Then we can
have your v3 in mm-unstable first, then I can rebase my patch's v8 on
top of that (fixing any merge conflicts along the way), then send out
the whole thing again.
Does this sound good?
