On Mon, Mar 6, 2023 at 9:29 AM Linus Torvalds
<torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
>
> The correct thing to do is always that
>
> * Returns >= nr_cpu_ids if no cpus set.
>
> because nr_cpu_ids is always the *smallest* of the access sizes.
>
> Of course, right now Guenter seems to be reporting a problem with that
> optimization, so unless I figure out what is going on I'll just need
> to revert it anyway.
Ahh. And the reason is exactly that people do *not* follow that
"Returns >= nr_cpu_ids" rule.
The drivers/char/random.c code is very wrong, and does
if (cpu == nr_cpumask_bits)
cpu = cpumask_first(&timer_cpus);
which fails miserably exactly because it doesn't use ">=".
Oh well.
I'll have to look for more of this pattern, but basically all those
"xyz_cpumask_bits" things were supposed to always be just internal to
that header file implementation, which is *exactly* why you have to
check the result for ">= nr_cpu_ids".
Linus
