Re: [PATCH RESEND] soundwire: fix usages of device_get_named_child_node()

From: Pierre-Louis Bossart
Date: Tue May 28 2024 - 11:41:10 EST




On 5/28/24 10:19, Markus Elfring wrote:
>> Add fwnode_handle_put() to avoid leaked references.
>
> Are you going to respond also to my previous patch review
> in more constructive ways?
> https://lore.kernel.org/lkml/eb15ab0a-e416-4ae9-98bb-610fdc04492c@xxxxxx/
> https://lkml.org/lkml/2024/4/29/493

Sorry about that, both Bard and I missed your comments.

On the Fixes tag: I made a deliberate choice to add all the fixes in one
patch, to show that the usage was copy-pasted and done 'wrong' in
multiple places. That makes it really hard to add a Fixes tag since the
different uses were added in a time interval of about 5 years.
We could split and have multiple patches if that was desired, but I
would still not include a Fixes tag since the leaked references are not
that bad, we read the Manager properties on probe, and the peripheral
properties are generally not used by codec drivers, so it's unlikely
that any user will ever see a problem that requires a backport in
linux-stable. This problem was found by reading the documentation while
adding new things, not by any user report or test failure.

The error flow was revisited and hardened in a follow-up patchset which
also adds new properties for MIPI DisCo 2.1 spec [1], we'll share the
patches in this kernel cycle.

[1] https://github.com/thesofproject/linux/pull/4857