Re: [PATCH v6 2/3] treewide: Switch memcpy() users of 'task->comm' to a more safer implementation

[Linus Torvalds <torvalds@linux-foundation.org>]

On Sat, 26 Jul 2025 at 16:19, Kees Cook <kees@kernel.org> wrote:
>
> That works for me! I just get twitchy around seeing memcpy used for strings. :) if we're gonna NUL after the memcpy, just use strscpy_pad().

I do worry a tiny bit about performance.

Because 'memcpy+set last byte to NUL' really is just a couple of
instructions when we're talking small constant-sized arrays.

strscpy_pad() isn't horrible, but it's still at another level. And
most of the cost is that "return the length" which people often don't
care about.

Dang, I wish we had some compiler trick to say "if the value isn't
used, do X, if it _is_ used do Y".

It's such a trivial thing in the compiler itself, and the information
is there, but I don't think it is exposed in any useful way.

In fact, it *is* exposed in one way I can think of:

   __attribute__((__warn_unused_result__))

but not in a useful form for actually generating different code.

Some kind of "__builtin_if_used(x,y)" where it picks 'x' if the value
is used, and 'y' if it isn't would be lovely for this.

Then you could do things like

    #define my_helper(x) \
        __builtin_if_used( \
                full_semantics(x), \
                simpler_version(x))

when having a return value means extra work and most people don't care.

Maybe it exists in some form that I haven't thought of?

Any compiler people around?

                 Linus