Re: objtool warning "uses BP as a scratch register" with clang-9

[Linus Torvalds]

On Thu, Aug 29, 2019 at 10:35 AM Josh Poimboeuf <jpoimboe@xxxxxxxxxx> wrote:
>
> Peter suggested to try WRITE_ONCE for the two zero writes to see if that
> "fixes" it.

I'm sure it "fixes" it.

.. and then where else will we hit this?

It's one thing to turn a structure zeroing into "memset()", but some
places really can't do it.

We use "-ffreestanding" in some places to make sure that gcc doesn't
start calling random libc routines. I wonder if we need to make it a
general rule that it's done unconditionally.

Sadly, I think that ends up also disabling things like
"__builtin_memcpy()" and friends. Which we _do_ want to have access
to, because then gcc can inline the memcpy() when we _do_ use
memcpy().

We used to do all of those heuristics by hand, but wanted to let the
compiler do them for us.

So:

 - we do want "memcpy()" to become "__builtin_memcpy()" which can then
be optimized to either individual inlined assignments _or_ to an
out-of-line call to memcpy().

 - we do *not* want individual assignments to be randomly turned into
memset/memcpy(), because of various different reasons (including
function tracing, but also store tearing, yadda yadda)

Conceptually, "-ffreestanding" is definitely what a kernel needs, but
it has been *too* big of a hammer and disables real code generation,
iirc.

             Linus