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

[Greg Kroah-Hartman <gregkh@linuxfoundation.org>]

On Sat, Sep 20, 2014 at 08:41:21PM +0000, Paul Zimmerman wrote:
> > From: linux-usb-owner@vger.kernel.org [mailto:linux-usb-owner@vger.kernel.org] On Behalf Of Alan Stern
> > Sent: Friday, September 19, 2014 12:39 PM
> > 
> > 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.
> 
> Hmm. What are the odds that some old kernel monitoring program, in an
> ancient Fedora distribution let's say, will be checking for 'khubd' in
> the kernel process list? Pretty high I would think.

I really doubt it, what would that tell them?

thanks,

greg k-h