Re: [PATCH mm-new v2 0/5] mm: remove unused functions

From: Garg, Shivank

Date: Wed Nov 18 2026 - 13:55:19 EST


On Wed, 2026-11-18 at 22:44 +0530, Dev Jain wrote:
>
> On 18/11/26 10:35 pm, Dev Jain wrote:
> >
> >
> > On 17/11/26 12:46 pm, Shivank Garg wrote:
> > > This series removes five functions and one variable in mm that have no
> > > users left anywhere in the tree. I found them with the kunused scripts
> > > [1], which I wrote with LLM's help. The v1 cover letter [2] has more
> > > details on the scripts and how LLM was used.
> > >
> > > Changes in v2:
> > > - Split v1 [2] up by MM tree, as David suggested. These five patches
> > > are based on mm-new.
> > > - No changes to the patches themselves.
> > >
> > > [1] https://github.com/AMDESE/linux-mm/tree/shivank/kunused
> > > [2] https://lore.kernel.org/r/20261026-unused-sym-v1-0-82ca2dceabe1@xxxxxxx
> > >
> > > Signed-off-by: Shivank Garg <shivankg@xxxxxxx>
> > > ---
> >
> > Okay I just recalled that there have been numerous cases where people ran
> > scripts / compiler with flags etc and sent patches to remove unused functions,
> > but maintainers rejected them citing they are there for a future use. Here
> > are some of those instances:
> >
> > https://lore.kernel.org/all/20240907021506.4018676-1-Liam.Howlett@xxxxxxxxxx/
> > https://lore.kernel.org/linux-mm/20100211052247.GA15392@localhost/
> >
> > So I think the commit messages need to do more, they need to prove that the
> > unused function was a genuine oversight.
> >
>
> One way to do that would be to mention the commit and its date, if its too old,
> or, if it was part of a patch series but then the author changed their
> strategy in the later patches making the function redundant, etc etc.
>
> In patch 1, you remove mapping_clear_release_always() but d73065e60dcc 's description
> implies it is going to get used, which is confusing. Maybe we are missing
> a genuine bug, or the author forgot to drop the stale description across
> revisions.
>

Hi Dev,

Thanks for the review. I agree that some unused symbols may be kept
intentionally for a planned future user, which I checked for before
sending.

For most patches I had included some history, like the commit that
removed the last caller or the commit that added them without one. I
missed that for mapping_clear_release_always() in this patch and for
make_guard_swp_entry() in patch 2, so here is the history for both:

- mapping_clear_release_always() was added in b4fa966f03b7 ("mm, netfs,
fscache: stop read optimisation when folio removed from pagecache") and
never had a caller. Earlier version of had called it from ->evict_inode()
but V4 dropped those calls as unnecessary [1] leaving the helper.

- make_guard_swp_entry() was added in 7c53dfbdb024 ("mm: add PTE_MARKER_GUARD
PTE marker"), has never had a caller, and none has been posted in 2 years.

I left some symbols out because a posted series adds callers, for
example folio_dup_file_rmap_pmd() and folio_is_pfmemalloc(), or because
they are one half of a pair whose other half is still used, like
kmsan_poison_memory().

[1] https://lore.kernel.org/all/166924370539.1772793.13730698360771821317.stgit@xxxxxxxxxxxxxxxxxxxxxx

Thanks,
Shivank