Re: [PATCH v4 1/3] driver core: platform: remove software node on release()

["Danilo Krummrich" <dakr@kernel.org>]

On Mon May 11, 2026 at 4:50 PM CEST, Bartosz Golaszewski wrote:
> On Mon, May 11, 2026 at 4:31 PM Danilo Krummrich <dakr@kernel.org> wrote:
>>
>> On Mon May 11, 2026 at 4:17 PM CEST, Bartosz Golaszewski wrote:
>> > On Mon, 11 May 2026 16:00:23 +0200, Danilo Krummrich <dakr@kernel.org> said:
>> >> On Mon May 11, 2026 at 9:33 AM CEST, Bartosz Golaszewski wrote:
>> >>> Danilo: if there are no further comments, can you pick it up for v7.1?
>> >>
>> >> It seems that sashiko has a valid concern in [1]; can you confirm?
>> >>
>> >> [1] https://sashiko.dev/#/patchset/20260430-swnode-remove-on-dev-unreg-v4-0-01574da0aed3%40oss.qualcomm.com
>> >>
>> >
>> > Yes, I explained it here[1]. Basically it's similar to how we need to call
>> > platform_device_add_data() for devices created with platform_device_alloc().
>> >
>> > We can consider adding platform_device_add_software_node() once there's
>> > a potential user but for now I'd just leave it like this.
>>
>> But there are users that already need this, no? There is Xe [1] and Surface GPE
>> [2], or am I missing something?
>>
>> [1] https://elixir.bootlin.com/linux/v7.1-rc2/source/drivers/gpu/drm/xe/xe_i2c.c#L99
>> [2] https://elixir.bootlin.com/linux/v7.1-rc2/source/drivers/platform/surface/surface_gpe.c#L308
>
> Right, I was not aware of these. That could indeed cause a regression.
> I'd like to fix the problem in v7.1 but also keep it minimal so adding
> platform_device_add_software_node() and updating drivers to using it
> may be the next step but for now: how about adding
> platform_device_release_full() which would call
> device_remove_software_node() and then the existing
> platform_device_release()? We'd replace the .release callback of
> struct device in platform_device_register_full() but if the user just
> uses platform_device_alloc(), they would keep the regular .release()
> that doesn't remove the software node?
>
> That would go into v7.1 and then I'd provide
> platform_device_add_software_node(), use it in all drivers that need
> it, and then we'd remove platform_device_release_full() and go back to
> a single, unified release callback?
>
> Does it make sense?

If it is just those two (and at least I did not find any other drivers) it might
be easier to just add platform_device_add_software_node(), use it within those
two drivers and still land it for v7.1.

I feel like this change turns out simpler and less error prone than the above
workaround to keep changes local to the driver core.