Re: rtmutex, pi_blocked_on, and blk_flush_plug()

[Sebastian Andrzej Siewior <bigeasy@linutronix.de>]

On 2023-03-04 23:39:57 [-0600], Crystal Wood wrote:
> > This still leaves the problem vs. io_wq_worker_sleeping() and it's
> > running() counterpart after schedule().
> 
> The closest thing I can see to a problem there is io_wqe_dec_running()->
> io_queue_worker_create()->io_wq_cancel_tw_create()->kfree(), but that only
> happens with func == create_worker_cont(), and io_wqe_dec_running() uses 
> create_worker_cb(). 

So we may good then. The while loop in io_wq_cancel_tw_create() worries
me a little. I am not sure if only the submitted work gets cancel or
maybe other as well, including the one leading the kfree. 

> Are there any workloads I could run to stress out that path (with my
> asserts in place)?

None that I can think of. Maybe something from the io-ring test suite.
But then you may need to bend to code to get the task_add() to fail.
Maybe Jens knows something.

> -Scott

Sebastian