On 25/08/22 14:14, Yury Norov wrote:
> On Thu, Aug 25, 2022 at 07:12:05PM +0100, Valentin Schneider wrote:
>> +#define for_each_cpu_andnot(cpu, mask1, mask2) \
>> + for ((cpu) = -1; \
>> + (cpu) = cpumask_next_andnot((cpu), (mask1), (mask2)), \
>> + (cpu) < nr_cpu_ids;)
>
> The standard doesn't guarantee the order of execution of last 2 lines,
> so you might end up with unreliable code. Can you do it in a more
> conventional style:
> #define for_each_cpu_andnot(cpu, mask1, mask2) \
> for ((cpu) = cpumask_next_andnot(-1, (mask1), (mask2)); \
> (cpu) < nr_cpu_ids; \
> (cpu) = cpumask_next_andnot((cpu), (mask1), (mask2)))
>
IIUC the order of execution *is* guaranteed as this is a comma operator,
not argument passing:
6.5.17 Comma operator
The left operand of a comma operator is evaluated as a void expression;
there is a sequence point after its evaluation. Then the right operand is
evaluated; the result has its type and value.
for_each_cpu{_and}() uses the same pattern (which I simply copied here).
Still, I'd be up for making this a bit more readable. I did a bit of
digging to figure out how we ended up with that pattern, and found
7baac8b91f98 ("cpumask: make for_each_cpu_mask a bit smaller")
so this appears to have been done to save up on generated instructions.
*if* it is actually OK standard-wise, I'd vote to leave it as-is.
>> +
>> /**
>> * cpumask_any_but - return a "random" in a cpumask, but not this one.
>> * @mask: the cpumask to search
>> --
>> 2.31.1
