Re: [RFC PATCH 1/5] mm/damon/core: introduce damos_quota_goal_tuner

[SeongJae Park]

On Wed, 11 Feb 2026 22:23:08 -0800 SeongJae Park <sj@xxxxxxxxxx> wrote:

> Currently DAMOS quota goal feature utilizes a single feedback loop based
> algorithm for automatic tuning of the effective quota.  It is proven to
> be useful on dynamic environments that operate systems with only
> kernels.  But, no one fits all, and we got multiple reports [1,2] of
> cases that the algorithm is not optimum.
> 
> Introduce a new field of 'struct damos_quotas', namely 'goal_tuner'.  It
> specifies what tuning algorithm the given scheme should use, and allows
> DAMON API callers to set it as they want.  Nonetheless, this commit does
> not introduce a new tuning algorithm but only the interface.  This
> commit hence makes no behavioral change.  A new algorithm will be added
> by the following commit.
> 
> [1] https://lore.kernel.org/CALa+Y17__d=ZsM1yX+MXx0ozVdsXnFqF4p0g+kATEitrWyZFfg@xxxxxxxxxxxxxx
> [2] https://lore.kernel.org/20260204022537.814-1-yunjeong.mun@xxxxxx
> 
> Signed-off-by: SeongJae Park <sj@xxxxxxxxxx>
> ---
>  include/linux/damon.h | 9 +++++++++
>  mm/damon/core.c       | 4 +++-
>  2 files changed, 12 insertions(+), 1 deletion(-)
> 
> diff --git a/include/linux/damon.h b/include/linux/damon.h
> index a4fea23da8576..25345b5f821b9 100644
> --- a/include/linux/damon.h
> +++ b/include/linux/damon.h
> @@ -213,6 +213,14 @@ struct damos_quota_goal {
>  	struct list_head list;
>  };
>  
> +/**
> + * enum damos_quota_goal_tuner - Goal-based quota tuning logic.
> + * @DAMOS_QUOTA_GOAL_TUNER_CONSIST:	Aim long term consistent quota.
> + */
> +enum damos_quota_goal_tuner {
> +	DAMOS_QUOTA_GOAL_TUNER_CONSIST,
> +};
> +
>  /**
>   * struct damos_quota - Controls the aggressiveness of the given scheme.
>   * @reset_interval:	Charge reset interval in milliseconds.
> @@ -260,6 +268,7 @@ struct damos_quota {
>  	unsigned long ms;
>  	unsigned long sz;
>  	struct list_head goals;
> +	enum damos_quota_goal_tuner goal_tuner;
>  	unsigned long esz;
>  
>  	unsigned int weight_sz;
> diff --git a/mm/damon/core.c b/mm/damon/core.c
> index 0ff190ed8a599..1c126d910fe62 100644
> --- a/mm/damon/core.c
> +++ b/mm/damon/core.c
> @@ -405,8 +405,9 @@ struct damos *damon_new_scheme(struct damos_access_pattern *pattern,
>  	INIT_LIST_HEAD(&scheme->list);
>  
>  	scheme->quota = *(damos_quota_init(quota));
> -	/* quota.goals should be separately set by caller */
> +	/* quota.goals and .goal_tuner should be separately set by caller */
>  	INIT_LIST_HEAD(&scheme->quota.goals);
> +	scheme->quota.goal_tuner = DAMOS_QUOTA_GOAL_TUNER_CONSIST;

Ravi found this is resetting tuner setup in damon_sysfs_add_quota_score() in
sysfs-schemes.c.  As a result, users were unable to use goal_tuner feature.
Ravi further suggested a fix that sets the goal_tuner after damon_new_scheme(),
as suggested here.

I was thinking that's the right fix assuming the comment is right.  I was also
thinking I entirely forgot setting the goal_tuner field in sysfs-schemes.c.
But I now find I didn't entirely forget it, as the third patch of this series
does set it on the quota struct that is passed to damon_new_scheme().  However
I was stupidly forgetting the fact that damon_new_scheme() overwrites the
field, and I asked callers to set it after the call.

Now I think the above initialization and the instruction on the comment is just
wrong.  There is no reason to refuse the caller-set value in the parameter
struct.  Rather, it only confuses callers.

I will fix the issue by simply dropping the above diff.

[1] https://lore.kernel.org/20260223123232.12851-2-ravis.opensrc@xxxxxxxxx


Thanks,
SJ

[...]