On Feb 17, 2010, at 12:23 PM, Oleg Nesterov [off-list ref] wrote:

On 02/17, Dmitry Torokhov wrote:

quoted

It apppears that it is allowed to try to cancel work that has never  
been
queued

Yes,

quoted

and I believe that canceled or completed work should be exactly
the same as never been queued work (which is apparently not the case
currently).

And yes, currently this is not the case.

As I said, I agree that cancel() could clear ->data. Will this change
help? (in any case this change is not for 2.6.33)

But I don't see how "completed" can do this, please see my previous
email. Note that flush() can't do this too.

quoted

Yes, it is certainly possible to work around the issue in every  
driver
that may happen to shut down and re-create workqueue as needed. The
question is whether it is the right thing to do.

I'd say, the question is whether we can improve this ;) Well, see  
above.

I understand. I think if we just change cancel to do the right thing  
that should be enough to avoid surprises.

-- 
Dmitry