Re: [PATCH 04/65] x86/hypercall: Annotate fnptr targets

[Jan Beulich <jbeulich@xxxxxxxx>]

On 29.11.2021 10:38, Jan Beulich wrote:
> On 26.11.2021 15:28, Andrew Cooper wrote:
>> On 26/11/2021 14:21, Jan Beulich wrote:
>>> On 26.11.2021 13:33, Andrew Cooper wrote:
>>>> --- a/xen/include/xen/hypercall.h
>>>> +++ b/xen/include/xen/hypercall.h
>>>> @@ -18,12 +18,12 @@
>>>>  #include <asm/hypercall.h>
>>>>  #include <xsm/xsm.h>
>>>>  
>>>> -extern long
>>>> +extern long cf_check
>>>>  do_sched_op(
>>>>      int cmd,
>>>>      XEN_GUEST_HANDLE_PARAM(void) arg);
>>> What purpose does the attribute serve on a declaration? On the surface
>>> I would consider it meaningful only on definitions, like e.g. __init.
>>
>> Because GCC treats cf_check (and nocf_check) as part of the function
>> type.  Simply getting it wrong will yield a "definition doesn't match
>> prototype" error.
>>
>> Furthermore, it needs to be visible across translation units so one TU
>> can spot (and complain at) creating a function pointer to a non-local
>> non-endbr'd function.
> 
> Hmm, it might well be that way, but that's not what the doc (for 11.2)
> says. While suggesting things are along the lines of what you say for
> "nocf_check", for "cf_check" it really only talks about code generation:
> "The cf_check attribute on a function is used to inform the compiler
> that ENDBR instruction should be placed at the function entry when
> ‘-fcf-protection=branch’ is enabled." And even for "nocf_check" it talks
> about extra compile time checks only when the attribute is applied to a
> function pointer variable/field, not when applied to a function.

The two attributes looks to indeed be as different as the doc suggests,
after some experimenting. But I now understand that's what patch 56 is
about.

Jan