Re: Race between io_wqe_worker() and io_wqe_wake_worker()

[Jens Axboe <axboe@kernel.dk>]

On 8/3/21 1:20 PM, Nadav Amit wrote:
> 
> 
>> On Aug 3, 2021, at 11:14 AM, Jens Axboe <axboe@kernel.dk> wrote:
>>
>> On 8/3/21 12:04 PM, Nadav Amit wrote:
>>>
>>>
>>> Thanks for the quick response.
>>>
>>> I tried you version. It works better, but my workload still gets stuck
>>> occasionally (less frequently though). It is pretty obvious that the
>>> version you sent still has a race, so I didn’t put the effort into
>>> debugging it.
>>
>> All good, thanks for testing! Is it a test case you can share? Would
>> help with confidence in the final solution.
> 
> Unfortunately no, since it is an entire WIP project that I am working
> on (with undetermined license at this point). But I will be happy to
> test any solution that you provide.

OK no worries, I'll see if I can tighten this up. I don't particularly
hate your solution, it would just be nice to avoid creating a new worker
if we can just keep running the current one.

I'll toss something your way in a bit...

-- 
Jens Axboe