Re: [PATCH] x86/sev: Mark snp_abort() noreturn

[Segher Boessenkool]

Hi!

On Wed, Aug 24, 2022 at 10:45:12PM +0200, Peter Zijlstra wrote:
> On Wed, Aug 24, 2022 at 12:29:29PM -0500, Segher Boessenkool wrote:
> > Would -Wmissing-noreturn have caught this?  It sounds like you need this
> > (and then fix all resulting warnings) to not upset objtool?
> > 
> > It is nice to have this anyway (if there aren't a zillion false
> > positives), but it seems objtool is very fragile.
> 
> Well, just like gcc has noreturn heuristics so has objtool, it just
> turns into pain when they don't agree with one another.
> 
> Ideally noreturn would be reflected in the object file so we don't have
> to guess at it.

It is!  A noreturn function (that doesn't warn like "warning: 'noreturn'
function does return") does not have whatever your architecture uses for
function returns in it.  Just like most non-noreturn functions that do
not return btw: the attribute affects code generation of the *caller* of
such functions.

> STT_FUNC_NORETURN would do I suppose, except then all
> the tools will need to be taught how to deal with that, which is also
> very painful.

What is that?  Even Google has no idea.  Hrm.

> Another options is something like .symtab.noreturn which is another
> symbol table explicitly listing the noreturn functions. Since it's an
> extra section tools that don't know about it can freely ignore it and
> carry on as usual.

The noreturn attribute, like any attribute, is informing the compiler
about some attributes of the code it is compiling.  Trying to use it
at a completely different level, for a completely different purpose,
will only end in tears.  Say no to tears.

The only thing a compiler does with it (not counting diagnostics) is it
says that calls to such a function never return, are really just a jump,
not a call.  This allows the compiler to generate slightly faster code
(smaller, and it won't realistically be in any hot path).

   - - -

What fundamental problem does objtool have in dealing with any normal
compiled code itself?  Does it try to understand the semantics of the
machine code (not very tractable), does it expect some magic markup to
be generated together with the machine code, does it want to have
compilers hamstrung wrt what kind of code they can generate?

There is some serious disconnect here, and I'm not even completely sure
what it is :-(


Segher