Re: [PATCHv2] thunderbolt: do not double dequeue a request

[Sergey Senozhatsky]

On (25/03/27 16:55), Mika Westerberg wrote:
[..]
> > Yes, if it's scheduled.  If it's already executed then we can schedule
> > again.
> > 
> > 	tb_cfg_request_sync() {
> > 	 tb_cfg_request()
> > 	   schedule_work()
> 
> This point it runs tb_cfg_request_work() which then calls the callback
> (tb_cfg_request_complete()) before it dequeues so "done" is completed.
> 
> > 	                        executes tb_cfg_request_dequeue
> 
> > 	 wait_for_completion_timeout()
> 
> so this will return > 0 as "done" completed..
> 
> > 	   schedule_work()
> > 	                        executes tb_cfg_request_dequeue again
> 
> ..and we don't call this one.

Ah, okay, I see.  Thanks for the explanations.  I'll drop
that one from the commit message then (let me re-spin v3,
just for the history).

[..]
> > Let me see what I can do (we don't normally apply patches that
> > were not in the corresponding subsystem tree).
> > 
> > In the meantime, do you have a subsystem/driver tree that is exposed
> > to linux-next?  If so, would be cool if you can pick up the patch so
> > that it can get some extra testing via linux-next.
> 
> Yes I do, see [1] but it does not work like that. First you should make
> sure you patch works by testing it yourself and then we can pick it up for
> others to test.

Sure, if I had the H/W testing would have done by now.  OK, let me try
to work this out.