Re: [PATCH 0/2] pinctrl: fix group/function table leaks on consumer re-probe

[Linus Walleij]

On Thu, Oct 1, 2026 at 10:52 AM A. Sverdlin
<alexander.sverdlin@xxxxxxxxxxx> wrote:

> From: Alexander Sverdlin <alexander.sverdlin@xxxxxxxxxxx>
>
> Pinctrl drivers that create their groups/functions lazily from the
> consumer's device-tree node in .dt_node_to_map() leak the per-mapping
> tables when the same node is mapped again. The core deduplicates
> groups/functions by name and keeps the first instance for the controller's
> lifetime, so it silently drops the copies built on every subsequent
> mapping - e.g. on each re-probe of the consumer device. kmemleak does not
> flag them because they stay reachable from the controller's devres list.
>
> Reproducer (patch 1, pinctrl-single) - any board where a device referencing
> a pinctrl-single state is repeatedly re-probed; here an I2C TPM whose node
> has a "default" pinctrl state:
>
>   while modprobe tpm_tis_i2c; do rmmod tpm_tis_i2c; done
>
> The tables land in a generic kmalloc-N cache (kmalloc-192 on arm64, where
> ARCH_KMALLOC_MINALIGN pads every devres allocation; the bucket differs by
> config), so watch it cache-independently by diffing /proc/slabinfo after
> dropping the reclaimable slab:
>
>   sync; echo 3 > /proc/sys/vm/drop_caches
>   cat /proc/slabinfo | awk 'NR>2{print $1,$2}' | sort > /tmp/a
>   <run N cycles>
>   sync; echo 3 > /proc/sys/vm/drop_caches
>   cat /proc/slabinfo | awk 'NR>2{print $1,$2}' | sort > /tmp/b
>   join /tmp/a /tmp/b | awk '$3>$2{print $1,$3-$2}'
>
> Patch 1 (pinctrl-single) was found and fixed with the above test on real
> hardware. Patch 2 (ti-iodelay) has the same bug pattern - plus a missing
> .dt_free_map that leaks the mapping table as well - but was found by code
> review only and is compile-tested only; I have no such hardware to
> reproduce it.
>
> Alexander Sverdlin (2):
>   pinctrl: single: don't leak group/function tables on consumer re-probe
>   pinctrl: ti-iodelay: fix memory leaks on consumer re-probe

I got exactly zero reviews on this patch set from the pinctrl-single
maintainers, so I have now just applied it.

Maybe the old maintainers have other things to do, have you considered
signing up as co-maintainer? You seem to know what you are doing.

Yours,
Linus Walleij