Re: [PATCH] mm/damon: modified damon_call_control from static to kmalloc

[JaeJoon Jung <rgbi3307@gmail.com>]

On Thu, 11 Dec 2025 at 15:04, SeongJae Park <sj@kernel.org> wrote:
>
> On Thu, 11 Dec 2025 12:29:59 +0900 JaeJoon Jung <rgbi3307@gmail.com> wrote:
>
> > On Wed, 10 Dec 2025 at 11:54, SeongJae Park <sj@kernel.org> wrote:
> > >
> > > On Tue, 9 Dec 2025 21:20:42 +0900 JaeJoon Jung <rgbi3307@gmail.com> wrote:
> [...]
> > I suggested removing the dealloc_on_cancel condition from the patch.
> > When using the damon_call_control structure, if you only use kmalloc()/kfree()
> > when canceled without dealloc_on_cancel.  However, currently, in other modules,
> > since it is static struct damon_call_control { ... }, kfree() is not allowed,
> > so dealloc_on_cancel condition is additionally required.
> > Because kmalloc()/kfree() and static are mixed together,
> > dealloc_on_cancel condition is required.
> > As you pointed out in the previous email, kmalloc()/kfree() has the burden of
> > memory allocation and deallocation, and static has the disadvantage of
> > increasing
> > the code size.  Both have their pros and cons.
> > Among these, I proposed unifying them into kmalloc()/kfree() methods and
> > removing the dealloc_on_cancel
>
> Thank you for kindly clarifying this, JeaJoon!
>
> >
> > >
> > > Meanwhile I find the feature might look complicated, or not well documented.
> > > Specifically, dealloc_on_cancel should be set on only dynamic-allocated
> > > damon_call_control object, but that is not well documented on the kernel-doc
> > > comments.  Are you saying removing it is an advantage because it makes reading
> > > code easier?  If that's the case, how about improving the documentation?
> > >
> >
> > dealloc_on_cancel is only used in mm/damon/sysfs.c.  Here, since kmalloc
> > the damon_call_control structure, the dealloc_on_cancel=true condition is
> > required.  If you want to keep the current code as is and only change the
> > comment, it would be clearer to comment it out like this:
> >
> > De-allocate when canceled.
> > --> To perform kfree() if allocated with kmalloc() when canceled.
>
> Sounds good.  Would you mind sending a patch for that?

Would you like me to send you the patch with only the comments corrected?

Thanks,
JaeJoon

>
>
> Thanks,
> SJ
>
> [...]