On 6/25/09 11:20 PM, Andrew Morton wrote:

On Tue, 23 Jun 2009 13:54:48 +0200
Richard R__jfors[off-list ref]  wrote:

quoted

This patch removes the HR timer, since it's bad to do synchronous I2C
in the HR timer callback context. The new implementation makes use
of the global workqueue. The work is scheduled every 5ms when polling
rather than 5 us.

"it's bad" isn't a very good description of the problem which the patch
fixes.

The problem is that the synchronous I2C calls in HR-timer callback 
context might end up in a deadlock.

This matters.  People wish to make decisions about whether this patch
is needed in 2.6.29.x, 2.6.30.x, 2.6.31, 2.6.32, etc.  Without knowing
the effects of the problem which the patch fixes, we cannot make that
decision!

quoted

+	/* cancel any work */
+	cancel_delayed_work(&ts->work);
+

Should this have been cancel_delayed_work_sync()?

It should, I will post a new patch shortly which should be applied after 
these two which fixes this.

--Richard