[Testing] [OLPC New Zealand] Testing Summary: 3 December 2011 - Auckland, New Zealand

[James Cameron]

On Mon, Dec 05, 2011 at 10:22:32PM +1300, Tom Parker wrote:
> On Mon, 2011-12-05 at 13:57 +1100, James Cameron wrote:
> 
> > > Speak - starts but no speaking and no lips moving. There is a high
> > > pitched whine (Tom estimates 10KHz) on clicky keys C1 and then it
> > > locked up and stopped updating the display. The other C1 also had no
> > > speaking and no lips moving and then locked in the same way. Both are
> > > successfully killed using the force quit option if you stop from the
> > > frame.
> > 
> > The estimated 10kHz tone is likely to be aliasing.  See #11334 for
> > technical background.  The lockup is unfamiliar, but if automatic power
> > management was enabled we know it happens.
> 
> Really, automatic power management can lock up just one application? It
> appeared like the thread that responds to X events had stopped -- areas
> damaged from the frame was not redrawn. Having said that, I'm not sure
> it was totally locked -- it might have just been very very very slow.

The audio driver lacks suspend and resume support, so the behaviour of
an application that uses audio is difficult to predict when automatic
power management is enabled.  Please try to reproduce the problem with
it disabled.

I'm not familiar with the event loop of the Speak activity, or how it
reacts to the situation.

I'd test it myself, but my units are all on a critical test task.

> My experience so far diagnosing locks and poor performance with python
> has been poor. With a significant amount of effort, bandwidth,
> reconfiguration, and most importantly, time I have been able to do back
> traces with gdb. Since yum is barely functional I haven't tried on the
> 1.75. I'm used to java where you can send a signal and get a thread dump
> (at work we even built a profile viewer which takes a series of thread
> dumps for offline profiling in production). Is there anything like this
> available in python?

Python has profiling infrastructure, but I don't recall how to get a
backtrace.

If the problem occurs again, I suggest checking the CPU load, checking
the processes consuming CPU, trying to record and play sound using ALSA
utilities, and figure out if the lack of response to expose events
extends to other applications.

-- 
James Cameron
http://quozl.linux.org.au/