Re: [PATCH 2/6] locking: Introduce range reader/writer lock

From: Laurent Dufour
Date: Thu Apr 13 2017 - 04:58:18 EST


On 13/04/2017 10:38, Jan Kara wrote:
> On Thu 13-04-17 10:07:48, Laurent Dufour wrote:
>> On 06/04/2017 18:50, Davidlohr Bueso wrote:
>>> On Thu, 06 Apr 2017, Laurent Dufour wrote:
>>>
>>>> How is 'seqnum' wrapping handled here ?
>>>> I'd rather see something like time_before() here, isn't it ?
>>>
>>> Its a 64bit counter, no overflows.
>>
>> I should have miss something, what prevents this 64bit counter to not
>> overflow ?
>> At some point of time, this counter could reach ~0UL and then 0UL, which
>> is breaking this check.
>
> Count for yourself how long would it take for the counter to overflow if we
> incremented it say every nanosecond?

I do agree that this will take time, but systems become larger and
having 200 threads hitting a lock like the mmap_sem every nanoseconds,
it will increment faster (still need about 3 years)... if there are 1000
threads, less than 1 year to overflow this counter...
This is not for today, I agree.