Re: violating function pointer signature

[Steven Rostedt <rostedt@goodmis.org>]

On Thu, 19 Nov 2020 09:04:57 -0800
Alexei Starovoitov <alexei.starovoitov@gmail.com> wrote:

> On Thu, Nov 19, 2020 at 6:59 AM Steven Rostedt <rostedt@goodmis.org> wrote:
> > Linux obviously
> > supports multiple architectures (more than any other OS), but it is pretty
> > stuck to gcc as a compiler (with LLVM just starting to work too).
> >
> > We are fine with being stuck to a compiler if it gives us what we want.  
> 
> I beg to disagree.

I think you misunderstood.

> android, chrome and others changed their kernel builds to
> "make LLVM=1" some time ago.
> It's absolutely vital for the health of the kernel to be built with
> both gcc and llvm.

That's what I meant with "LLVM just starting to work too".

LLVM has been working hard to make sure that it can do the same tricks that
the kernel depends on gcc for. And LLVM appears to be working fine with the
nop stub logic (it's already in 5.10-rc with with the static callers).

We can easily create a boot up test (config option) that will test it, and
if a compiler breaks it, this test would trigger a failure.

Again, both static calls and tracepoint callbacks are limited in what they
can do. Both return void, and both do are not variadic functions. Although
passing in a struct as a parameter is possible, we could add testing to
detect this, as that's rather slow to begin with.

-- Steve