Re: [PATCH] static_call: fix function type mismatch

[Peter Zijlstra]

On Wed, Mar 24, 2021 at 06:33:39PM +0100, Peter Zijlstra wrote:
> On Wed, Mar 24, 2021 at 05:45:52PM +0100, Rasmus Villemoes wrote:
> > Sorry, I think I misread the code. The static calls are indeed
> > initialized with a function with the right prototype. Try adding
> > "preempt=full" on the command line so that we exercise these lines
> > 
> >                static_call_update(cond_resched,
> > (typeof(&__cond_resched)) __static_call_return0);
> >                 static_call_update(might_resched,
> > (typeof(&__cond_resched)) __static_call_return0);
> > 
> > I would expect that to blow up, since we end up calling a long (*)(void)
> > function using a function pointer of type int (*)(void).
> 
> Note that on x86 there won't actually be any calling of function
> pointers. See what arch/x86/kernel/static_call.c does :-)
> 
> But I think some of this code might need some __va_function() love when
> combined with CFI.
> 
> But yes, this is why I think something like -fcdecl might be a good
> idea, that ought to tell the compiler about the calling convention,
> which ought to be enough for the compiler to figure out that this magic
> really is ok.
> 
> Notable things we rely on:
> 
>  - caller cleanup of stack; the function caller sets up any stack
>    arguments and is also responsible for cleanin up the stack once the
>    function returns.

  - the arguments are pushed on stack right to left;

>  - the return value is in a register.
> 
> Per the first we can call a function that has a partial (empty per
> extremum) argument list. 

That extra constraint is required to make partial args work; as it
happens we only use empty args, and as such don't really care about this
atm.

> Per the second we can call a function with a
> different return type as long as they all fit in the same register.
> 
> The calling of a 'long (*)()' function for a 'int (*)()' type then
> becomes idential to something like: 'int x = (long)y', and that is
> something C is perfectly fine with.
> 
> We then slightly push things with the other __static_call_return0()
> usage in the kernel, where we basically end up with: 'void *x =
> (long)y', which is something C really rather would have a cast on.