Re: [PATCH] flush_cpu_workqueue: don't flush an empty ->worklist

[Andrew Morton <akpm@osdl.org>]

On Tue, 9 Jan 2007 10:34:17 +0530
Srivatsa Vaddagiri <vatsa@in.ibm.com> wrote:

> On Mon, Jan 08, 2007 at 03:54:28PM -0800, Andrew Morton wrote:
> > Furthermore I don't know which of these need to be tossed overboard if/when
> > we get around to using the task freezer for CPU hotplug synchronisation.
> > Hopefully, a lot of them.  I don't really understand why we're continuing
> > to struggle with the existing approach before that question is settled.
> 
> Good point!
> 
> Fundamentally, I think we need to answer this question:
> 
> "Do we provide *some* mechanism to block concurrent hotplug operations
> from happening? By hotplug operations I mean both changes to the bitmap
> and execution of all baclbacks in CPU_DEAD/ONLINE etc"
> 
> If NO, then IMHO we will be forever fixing races
> 
> If YES, then what is that mechanism? freeze_processes()? or a magical
> lock?
> 
> freeze_processes() cant be that mechanism, if my understanding of it is
> correct - see http://lkml.org/lkml/2007/1/8/149

That's not correct.  freeze_processes() will freeze *all* processes.  All
of them are forced to enter refrigerator().  With the mysterious exception
of some I/O-related kernel threads, which might need some thought.

> and 
> http://marc.theaimsgroup.com/?l=linux-kernel&m=116817460726058.

Am not sure how that's related.

> I would be happy to be corrected if the above impression of
> freeze_processes() is corrected ..

It could be that the freezer needs a bit of work for this application. 
Obviously we're not interested in the handling of disk I/O, so we'd really
like to do a simple
try_to_freeze_tasks(FREEZER_USER_SPACE|FREEZER_KERNEL_THREADS), but the
code isn't set up to do that (it should be).  The other non-swsusp callers
probably want this change as well.  But that's all a minor matter.