Re: [PATCH v2 01/22] x86/include/asm/intel-txt.h: constants and accessors for TXT registers and heap

[Jan Beulich <jbeulich@xxxxxxxx>]

On 03.06.2025 10:50, Sergii Dmytruk wrote:
> On Tue, Jun 03, 2025 at 09:06:53AM +0200, Jan Beulich wrote:
>> On 03.06.2025 00:00, Sergii Dmytruk wrote:
>>> On Mon, Jun 02, 2025 at 09:17:37AM +0200, Jan Beulich wrote:
>>>> On 23.05.2025 21:51, Sergii Dmytruk wrote:
>>>>> On Wed, May 21, 2025 at 05:19:57PM +0200, Jan Beulich wrote:
>>>>>>> +static inline uint64_t txt_bios_data_size(void *heap)
>>>>>>
>>>>>> Here, below, and in general: Please try to have code be const-correct, 
>>>>>> i.e.
>>>>>> use pointers-to-const wherever applicable.
>>>>>
>>>>> I assume this doesn't apply to functions returning `void *`.  The
>>>>> approach used in libc is to accept pointers-to-const but then cast the
>>>>> constness away for the return value, but this header isn't a widely-used
>>>>> code.
>>>>
>>>> Which is, from all I know, bad practice not only by my own view.
>>>
>>> I actually ended up doing that to have const-correctness in v3.  In the
>>> absence of function overloads the casts have to be somewhere, can put
>>> them in the calling code instead.
>>
>> Casts of which kind? For context: There shouldn't be any casting away of
>> const-ness (or volatile-ness, for the sake of completeness).
>>
>> Jan
> 
> Casting away const-ness inside of functions like
> 
>     static inline void *txt_bios_data_start(const void *heap)
> 
> If a function accepts a const pointer and returns it, this turns a
> non-const incoming pointer into a const one.  Without duplicating the
> code (either having const and non-const versions or repeating code in
> other ways), nothing can be made const cleanly in here including
> *_size() functions because they call *_start() functions:
> 
>     static inline uint64_t txt_os_mle_data_size(const void *heap)
>     {
>         return *((const uint64_t *)(txt_bios_data_start(heap) +
>                                                      // ^^^^ -- const
>                                     txt_bios_data_size(heap))) -
>                sizeof(uint64_t);
>     }

Yet just to repeat: Besides myself (and maybe others), Misra objects to
the casting away of const-ness.

Jan