Re: [PATCH v3 0/6] usb: hub: convert khubd into workqueue

From: Alan Stern <stern@rowland.harvard.edu>
To: Petr Mladek <pmladek@suse.cz>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
	Tejun Heo <tj@kernel.org>,
	Joe Lawrence <joe.lawrence@stratus.com>,
	Jiri Kosina <jkosina@suse.cz>, <linux-usb@vger.kernel.org>,
	<linux-kernel@vger.kernel.org>
Subject: Re: [PATCH v3 0/6] usb: hub: convert khubd into workqueue
Date: Fri, 19 Sep 2014 15:38:50 -0400 (EDT)	[thread overview]
Message-ID: <Pine.LNX.4.44L0.1409191536000.1133-100000@iolanthe.rowland.org> (raw)
In-Reply-To: <1411140744-21424-1-git-send-email-pmladek@suse.cz>

On Fri, 19 Sep 2014, Petr Mladek wrote:

> The 3rd version of the patchset is slightly reordered and refactored
> as suggested earlier. See below for more details.
> 
> IMHO, the result is more clean. But feel free to ask me to revert it.
> I do not want to make the review more complicated. Well, I double checked
> the diff of patches files between v2 and v3. It is pretty small and only
> the expected changes are there.
> 
> Changes in:
>     + v3:
> 	+ split the optimization of hub->hdev reference counting
> 	  into separate patch; usable standalone (1st patch)
> 
> 	+ switch order of the two patches that convert the kthread
> 	  and that remove the while cycle in hub_event(); it is
> 	  cleaner (2nd and 3rd patch)
> 
> 	+ use ordered workqueue by default; it makes it compatible
> 	  with the original kthread solution (3rd patch)
> 
>         + added optional patch that allows to process events
> 	  in parallel (6th patch)
> 
> 	+ fixed the suggested stylistic problems (2nd and 3rd patch)
> 
> 
>     + v2:
> 	+ solved potential races:
> 	  	 + get hub->kref in kick_hub_wq()
> 		 + call usb_get_dev(hdev) in hub_probe()
> 		   and 	usb_put_dev(hdev) in hub_release()
> 		 + INIT_WORK only once in hub_probe()
> 
> 	 + do not call cancel_work_sync() in hub_disconnect()
> 	   to keep it fast
> 
> 	 + rename kick_khubd() to kick_hub_wq() already
> 	   in the first patch; IMHO, it is cleaner while
> 	   adding only very few changes
> 
> 
> The workqueue API is well defined and tested. It has many options
> that could be used to tune the scheduling. The code is usually
> easier and thus more safe. It allows to avoid the extra thread
> in most cases. It has has clearly defined behavior vrt. system suspend.
> 
> This patchset converts khubd into the workqueue. It saves one thread,
> lock, and list.
> 
> It  looks huge but the main change is in the first patch. The rest is
> simple renaming of functions, comments and documentation.
> 
> 	   
> Thanks a lot Alan Stern and Tejun Heo for hints and guidance.
> 
> The patches can be applied either against Linus' tree or linux-next.

Very nice work.  I haven't tried it out, but it all looks good.  No
doubt we will be able to sort out unforeseen troubles if any arise,
without much difficulty.

For all six patches,

Acked-by: Alan Stern <stern@rowland.harvard.edu>

Alan Stern