Re: Clarifying confusion of our variable placement rules caused by cleanup.h

[Linus Torvalds]

On Tue, 18 Nov 2025 at 12:21, James Bottomley
<James.Bottomley@xxxxxxxxxxxxxxxxxxxxx> wrote:
>
> On Tue, 2025-11-18 at 14:17 -0500, Steven Rostedt wrote:
> > I think the code could also be better optimized? I haven't run an
> > objcopy to confirm but now early exits do not require calling the
> > __free() function on NULL pointers.
>
> Yes, I can confirm that (at least from reading the docs not actually
> from disassembling the code).

Actually, I _have_ been disassembling some of that code, and most of
the time the compiler is actually good at eliding those things and not
calling kfree() with a NULL pointer.

Now, the reason for that is actually that we spent some effort on this
in <linux/cleanup.h> (and by "we" I mean mostly PeterZ & co with me
being involved in the discussions).

So you'll see those destructor functions being inline functions with
things like that

   DEFINE_FREE(kfree, void *, if (_T) kfree(_T))

where that "if (_T)" being integral to having the compiler able to see
inline that "oh, it's statically NULL at this stage, I don't need to
call any external function".

But yes, sometimes having the declaration later can simplify this all
for the compiler too. But the *primary* thing should be about making
the code itself legible and maintainable to humans.

          Linus