Re: [XEN PATCH][for-4.19 v4 1/8] xen/include: add macro ISOLATE_LOW_BIT

[Nicola Vetrini <nicola.vetrini@bugseng.com>]

On 2023-10-31 11:20, Jan Beulich wrote:
> On 31.10.2023 11:03, Nicola Vetrini wrote:
>> On 2023-10-31 09:28, Nicola Vetrini wrote:
>>> On 2023-10-31 08:43, Jan Beulich wrote:
>>>> On 30.10.2023 23:44, Stefano Stabellini wrote:
>>>>> On Mon, 30 Oct 2023, Jan Beulich wrote:
>>>>>> On 27.10.2023 15:34, Nicola Vetrini wrote:
>>>>>>> --- a/xen/include/xen/macros.h
>>>>>>> +++ b/xen/include/xen/macros.h
>>>>>>> @@ -8,8 +8,14 @@
>>>>>>>  #define DIV_ROUND(n, d) (((n) + (d) / 2) / (d))
>>>>>>>  #define DIV_ROUND_UP(n, d) (((n) + (d) - 1) / (d))
>>>>>>> 
>>>>>>> -#define MASK_EXTR(v, m) (((v) & (m)) / ((m) & -(m)))
>>>>>>> -#define MASK_INSR(v, m) (((v) * ((m) & -(m))) & (m))
>>>>>>> +/*
>>>>>>> + * Given an unsigned integer argument, expands to a mask where
>>>>>>> just the least
>>>>>>> + * significant nonzero bit of the argument is set, or 0 if no 
>>>>>>> bits
>>>>>>> are set.
>>>>>>> + */
>>>>>>> +#define ISOLATE_LOW_BIT(x) ((x) & -(x))
>>>>>> 
>>>>>> Not even considering future Misra changes (which aiui may require
>>>>>> that
>>>>>> anyway), this generalization of the macro imo demands that its
>>>>>> argument
>>>>>> now be evaluated only once.
>>>>> 
>>>>> Fur sure that would be an improvement, but I don't see a trivial 
>>>>> way
>>>>> to
>>>>> do it and this issue is also present today before the patch.
>>>> 
>>>> This was an issue here for MASK_EXTR() and MASK_INSR(), yes, but the
>>>> new
>>>> macro has wider use, and there was no issue elsewhere so far.
>>>> 
>>>>> I think it
>>>>> would be better to avoid scope-creeping this patch as we are 
>>>>> already
>>>>> at
>>>>> v4 for something that was expected to be a trivial mechanical 
>>>>> change.
>>>>> I
>>>>> would rather review the fix as a separate patch, maybe sent by you 
>>>>> as
>>>>> you probably have a specific implementation in mind?
>>>> 
>>>> #define ISOLATE_LOW_BIT(x) ({ \
>>>>     typeof(x) x_ = (x); \
>>>>     x_ & -x_; \
>>>> })
>>>> 
>>>> Hard to see the scope creep here. What I would consider scope creep 
>>>> I
>>>> specifically didn't even ask for: I'd like this macro to be
>>>> overridable
>>>> by an arch. Specifically (see my earlier naming hint) I'd like to 
>>>> use
>>>> x86's BMI insn BLSI in the context of "x86: allow Kconfig control 
>>>> over
>>>> psABI level", when ABI v2 or higher is in use.
>>> 
>>> I appreciate you suggesting an implementation; I'll send a v5
>>> incorporating it.
>> 
>> There's an issue with this approach, though: since the macro is used
>> indirectly
>> in expressions that are e.g. case labels or array sizes, the build 
>> fails
>> (see [1] for instance).
>> Perhaps it's best to leave it as is?
> 
> Hmm. I'm afraid it's not an option to "leave as is", not the least 
> because
> - as said - I'm under the impression that another Misra rule requires
> macro arguments to be evaluated exactly once. Best I can think of right
> away is to have a macro for limited use (to address such build issues)
> plus an inline function (for general use). But yes, maybe that then 
> indeed
> needs to be a 2nd step.
> 
> Jan
> 
>> [1] 
>> https://gitlab.com/xen-project/people/bugseng/xen/-/jobs/5423693947
>> 

There is no such MISRA Rule afaik: R23.7 is similar, but only for C11 
generic selections.

-- 
Nicola Vetrini, BSc
Software Engineer, BUGSENG srl (https://bugseng.com)