Re: (sashiko review) [RFC PATCH v5 03/11] mm/damon/core: introduce failed region quota charge ratio

[SeongJae Park <sj@kernel.org>]

TL; DR: sashiko asks same question to previous one, which I decided to ignore.
My opinion is not changed.

Forwarding sashiko review in a reply format with my inline
comments below, for details of my view and doing discussions
via mails if needed.

> # review url: https://sashiko.dev/#/patchset/20260410142034.83798-4-sj@kernel.org
> # start of sashiko.dev inline review
> commit f672ea44b246c3c6bb3fac52d693ca50e48cbbac
> Author: SeongJae Park <sj@kernel.org>
> 
> mm/damon/core: introduce failed region quota charge ratio
> 
> This commit introduces a DAMON core API for setting the charge ratio for
> action-failed memory. It allows API callers to specify the ratio flexibly by
> setting a numerator and denominator.
> 
> > diff --git a/mm/damon/core.c b/mm/damon/core.c
> > index 97b375c720251..cb8f65db0f4b9 100644
> > --- a/mm/damon/core.c
> > +++ b/mm/damon/core.c
> [ ... ]
> > @@ -2041,6 +2043,23 @@ static void damos_walk_cancel(struct damon_ctx *ctx)
> >  	mutex_unlock(&ctx->walk_control_lock);
> >  }
> >  
> > +static void damos_charge_quota(struct damos_quota *quota,
> > +		unsigned long sz_region, unsigned long sz_applied)
> > +{
> > +	/*
> > +	 * sz_applied could be bigger than sz_region, depending on ops
> > +	 * implementation of the action, e.g., damos_pa_pageout().  Charge only
> > +	 * the region size in the case.
> > +	 */
> > +	if (!quota->fail_charge_denom || sz_applied > sz_region)
> > +		quota->charged_sz += sz_region;
> > +	else
> > +		quota->charged_sz += sz_applied + mult_frac(
> > +				(sz_region - sz_applied),
> > +				quota->fail_charge_num,
> > +				quota->fail_charge_denom);
> 
> Can the mult_frac() calculation silently overflow on 32-bit architectures?
> 
> The sz_region parameter is an unsigned long. On 32-bit architectures, this
> means intermediate calculations in mult_frac() will be bounded to 32 bits.
> 
> If a user supplies large values (e.g., fail_charge_num = 1,000,000 and
> fail_charge_denom = 1,000,000), could the remainder multiplication exceed the
> 32-bit ULONG_MAX limit? This could under-charge the quota for failed regions,
> causing the DAMOS scheme to exceed its configured quota limits.
> 
> Additionally, can setting the failed region charge ratio to 0 (e.g.,
> fail_charge_num = 0) neutralize the DAMON time quota (ms) and cause kdamond to
> hog the CPU?
> 
> If actions consistently fail (e.g., sz_applied == 0), the total_charged_sz
> will remain 0, leading to an estimated throughput of 0. This results in the
> effective size quota (esz) being bounded to min_region_sz.
> 
> Since charged_sz remains 0, the check in damos_quota_is_full():
> 
>     return quota->esz - quota->charged_sz < min_region_sz;
> 
> would evaluate to min_region_sz - 0 < min_region_sz, which is never met.
> Does this cause DAMON to bypass the time limit and process the entire target
> address space without stopping?

Same question was asked to the previous version, and my opinion is also still
same.  I will keep this as is.

> 
> > +}
> > +
> 
> 
> # end of sashiko.dev inline review
> # review url: https://sashiko.dev/#/patchset/20260410142034.83798-4-sj@kernel.org


Thanks,
SJ

# hkml [1] generated a draft of this mail.  You can regenerate
# this using below command:
#
#     hkml patch sashiko_dev --for_forwarding \
#             20260410142034.83798-4-sj@kernel.org
#
# [1] https://github.com/sjp38/hackermail