On Sun, Apr 26, 2026 at 6:24 AM Dmitry Torokhov <[email protected]> wrote: > > Still please check sashiko review - I think some of the concerns about > error unwinding are correct. >
Sigh... I hate the walls of text it generates. Can we set it to "caveman mode"? :) The first part of the review - while valid - already has a precedence with this function. If someone allocates the platform device with platform_device_alloc() and assigns the address of a static structure as its platform data, on release() we'll unconditionally call kfree() on it. The user must not assign platform data directly but use the dedicated platform_device_add_data() helper. We could consider providing platform_device_add_software_node() but since there are no users, let's cross that bridge when we get there? For the second point: sashiko's right, we need to bump the refcount earlier. For the third: I need to look into it deeper. I thought my second test case from patch 3/3 checks that use-case but the review makes sense. Bart
