Re: [RFC PATCH v3 0/4] mm/damon: Introduce node_eligible_mem_bp and node_ineligible_mem_bp Quota Goal Metrics

[SeongJae Park]

On Wed, 25 Feb 2026 10:19:11 -0800 Ravi Jonnalagadda <ravis.opensrc@xxxxxxxxx> wrote:

> On Mon, Feb 23, 2026 at 9:36 PM SeongJae Park <sj@xxxxxxxxxx> wrote:
> >
> > Hello Ravi,
> >
> > On Mon, 23 Feb 2026 12:32:28 +0000 Ravi Jonnalagadda <ravis.opensrc@xxxxxxxxx> wrote:
> >
> > > MIME-Version: 1.0
> > > Content-Type: text/plain; charset=UTF-8
> > > Content-Transfer-Encoding: 8bit
> > >
> > > This series introduces two new DAMON quota goal metrics for controlling
> > > memory migration in heterogeneous memory systems (e.g., DRAM and CXL
> > > memory tiering) using physical address (PA) mode monitoring.
> >
> > Thank you for keep working on and sharing this :)
> 
> Thank you for the detailed review!

My pleasure!

[...]
> > > - Added PA-mode detection lag compensation cache (see dedicated section
> > >   below for design details).
> >
> > I'm not very sure if this is really needed, though.  I'll leave comment on the
> > dedicated section below.
> 
> Understood. I consciously separated the cache implementation (patch 4)
> from the core metrics (patch 3) because the cache is ONE possible approach to
> handle detection lag - not necessarily THE approach. My goal was to share
> what was needed to achieve equilibrium with my synthetic benchmark
> workload (multiload),
> while making it clear that the cache mechanism could be dropped or
> replaced with alternatives.

That was indeed helpful for reviewing, thank you!

> 
> >
> > >
> > > - Added fix for esz=0 quota bypass that allowed unlimited migration when
> > >   goal was achieved.
> > >
> > > - Added fix for goal_tuner sysfs setting being ignored due to
> > >   damon_new_scheme() always defaulting to CONSIST.
> >
> > Thank you for finding and fixing these issues in my previously shared RFC patch
> > series!  I left a few comments to the patches.  In short, the second fix looks
> > good and I will add that to the next revision of my RFC patch series, if you
> > don't mind.  For the first fix, I'd like to take more time on thinking more
> > cleaner solution.
> 
> Sounds good. Please go ahead and incorporate the goal_tuner fix into
> your series.
> Happy to test whatever approach you come up with for the esz=0 issue.

Thank you, I will do!

[...]
> > > In PA-mode, when pages are migrated:
> > > 1. Source node detection drops immediately (pages are gone)
> > > 2. Target node detection increases slowly (new addresses need sampling)
> >
> > I agree.  And this is not what I clearly expected during the previous
> > discussion.  Thank you for sharing this issue.
> 
> I'm glad this observation is useful. It was something I discovered during
> testing that wasn't obvious until I looked at the trace data closely.

Thank you for sharing the pain point.  I recently added a few more DAMOS
tracepoints motivated by our offline discussion.  I'm planning to add better
supports of those in DAMON user-space tool.  Knowing this kind of pain points
is essential and useful at improving DAMON, thank you!

[...]
> > I will leave more comments to the patch implementing this.  But this seems too
> > much at the current stage, unless there are clear test results showing its
> > needs.  I'd recommend proceeding without this, and later revisit if the problem
> > becomes clearly significant.
> 
> I agree. Let's drop patch 4 for now and focus on getting the core
> metrics merged.
> The cache mechanism can be revisited later if real-world usage shows
> it's needed.

Thank you for flexibly accepting my suggestion!

[...]
> > I'm yet to further reply to the fourth patch, but I hope my comments be worthy
> > :)
> >
> 
> Very much so! Your feedback has been invaluable in shaping this work. :-)

More than exciting to hear that :D

> 
> I'm currently on a break and will be back after March 10th. Once I return,
> I'll send the updated patch 3 and share test results with CONSIST
> tuner.

Sounds perfect, I hope you to have great break!


Thanks,
SJ

[...]