Re: [PATCH] mm/damon: validate addr_unit to be power of 2

[SeongJae Park <sj@kernel.org>]

Hi Liew,


Unfortunately I'm bit confused about this patch.  Please answer below
questions.

On Fri, 27 Mar 2026 14:26:27 +0800 Liew Rui Yan <aethernet65535@gmail.com> wrote:

> Problem
> =======
> The 'addr_unit' must be a power of 2 for correct address alignment
> calculations.

It is required to be a power of 2 only if it is smaller than PAGE_SIZE.  Refer
to commit d8fe5d2a3379 ("Docs/mm/damon/design: document the power-of-two
limitation for addr_unit") for more details.

> Previously,

Can we make the timing more explicitly?  I'm confusing when is the time you are
mentioning, since a behavioral change was made recently, by commit c80f46ac228b
("mm/damon/core: disallow non-power of two min_region_sz").

> writing a non-power-of-2 value
> (e.g., addr_unit=3) would be accepted by the sysfs store callback, but
> cause kdamond to exit unexpectedly during parameter application, failing
> silently without returning an error to userspace.

I'm being more confused.  DAMON_SYSFS handles commits synchronously, and return
error to user space when it failed, iirc.  Am I missing something?

> 
> Solution
> ========
> Add an is_power_of_2() check in damon_commit_ctx() to reject invalid
> inputs immediately with -EINVAL. 

If invalid addr_unit already makes damon_commit_ctx() fails and therefore
kdamond exit, why we need more check?

> 
> When damon_commit_ctx() fails, the kdamond thread terminates as
> designed.

Again, I don't get it...


Thanks,
SJ

[...]