Re: [PATCH 3/4] Eclair: relax "noreturn" function-pointer conversion deviation

[Nicola Vetrini <nicola.vetrini@xxxxxxxxxxx>]

On 2026-09-03 13:44, Jan Beulich wrote:
> Like misra/rules.rst says, function arguments other than "void *" are 
> okay
> as well.
>
> Signed-off-by: Jan Beulich <jbeulich@xxxxxxxx>
> ---
> I can't explain why this covers the violation in mce.c:mce_callbacks'es
> initializer, but not the one in mce.c:default_handler's.
Possibly differing attributes (e.g. cf_check vs section attributes)? 
Just a guess that would need to be tested, though.
> As a result of 6852334f8416 ("Arm/GIC: add noreturn in a few more
> places"), vgic_v2_lpi_to_pending() and vgic_v2_lpi_get_priority() (both
> returning non-void) would also need covering. (As said in a remark 
> there,
> non-void together with noreturn is somewhat odd.)
Indeed

> Really before and after this change there's no checking that parameter 
> and
> return types actually match. I have no clue how one would express such
> checks.
The presence of a bitcast indicates that the two types do not match 
exactly. Typically function attributes are not relevant towards 
determining a type difference, but different compilers may model 
non-standard features differently (rightly so), in such a way that some 
make a difference in the AST, and others do not.
To check for compatibility of function pointers I would try activating 
service STD.funptrcv, which essentially mirrors 
-Wincompatible-pointer-types:
caution for rule STD.funptrcv: (rule) A pointer is used to call a 
function whose type is not compatible with the pointed-to type. 
(untagged)
p.c:8.8-8.8: Loc #1 [culprit: implicit cast converts from `void(*)(int)' 
to `__typeof__(@EXPR@)*' (that is `void(*)(void)')]
  qq = m;
       ^
p.c: In function ‘h’:
p.c:8:6: error: assignment to ‘void (*)(void)’ from incompatible pointer 
type ‘void (*)(int)’ [-Wincompatible-pointer-types]
    8 |   qq = m;
      |      ^
p.c:3:6: note: ‘m’ declared here
    3 | void m(int x);
      |      ^


> --- a/automation/eclair_analysis/ECLAIR/deviations.ecl
> +++ b/automation/eclair_analysis/ECLAIR/deviations.ecl
> @@ -391,11 +391,11 @@ constant expressions are required.\""
>  }
>  -doc_end
>
> --doc_begin="The conversion from 'void noreturn (*)(void *)' to 'void 
> (*)(void *)' is safe
> +-doc_begin="The conversion from 'void noreturn (*)(...)' to 'void 
> (*)(...)' is safe
>  because the semantics of the 'noreturn' attribute do not alter the 
> calling convention or behavior of the resulting code."
>  -config=MC3A2.R11.1,casts+={safe,
> -  
> "kind(bitcast)&&to(type(pointer(inner(return(builtin(void))&&all_param(1, 
> pointer(builtin(void)))))))&&from(expr(skip(!syntactic(),
> -   ref(property(noreturn)))))"}
> +  
> "kind(bitcast)&&to(type(pointer(inner(return(builtin(void))))))&&from(expr(skip(!syntactic(),ref(property(noreturn)))))"
> +}
>  -doc_end
>
>  -doc_begin="The conversion from a pointer to an incomplete type to 
> unsigned long does not lose any information, provided that the target 
> type has enough bits to store it."
--
Nicola Vetrini, B.Sc.
Software Engineer
BUGSENG (https://bugseng.com)
LinkedIn: https://www.linkedin.com/in/nicola-vetrini-a42471253