Re: [PATCH 1/2] sched/core: switch struct rq->nr_iowait to a normal int

[Thomas Gleixner]

On Thu, Feb 29 2024 at 15:30, Jens Axboe wrote:
> On 2/29/24 12:52 PM, Thomas Gleixner wrote:
>>         return atomic_read(&cpu_rq(cpu)->nr_iowait) & ((1 << 16) - 1);
>> 
>> Obviously written with proper inline wrappers and defines, but you get
>> the idea.
>
> I'll play with this a bit, but do we want to switch to an atomic_long_t
> for this? 2^16 in iowait seems extreme, but it definitely seems possible
> to overflow it.

Indeed. 32bit has PID_MAX_LIMIT == 0x8000 which obviously fits into 16
bits, while 64bit lifts that limit and relies on memory exhaustion to
limit the number of concurrent threads on the machine, but that
obviously can exceed 16bits.

Whether more than 2^16 sit in iowait concurrently on a single CPU that's
a different question and probably more academic. :)

Though as this will touch all nr_iowait places anyway changing it to
atomic_long_t in a preparatory patch first makes a lot of sense.

Thanks,

        tglx