backlight device, suspend/resume...

[Werner Almesberger]

Mike (mwester) wrote:
> So what then, when the user reports that "the phone randomly powers
> off"?  Do the hardware/kernel folks just tell the user to go away, and
> bother the application folks?

Ah, but we have that problem anyway. Since it's user space that
triggers all that, users are at the mercy of it, no matter how hard
the kernel tries to know better. And bug reports will often come in
the form of user perception, not in terms of what happens in any of
the underlying mechanisms.

In this context, it would be interesting to have a general panic or
watchdog handler that resets the system to a know to be good state
(which is a lot more than the backlight - e.g., what good is the
backlight if the LCM controller isn't configured ?) and then dumps
its report somewhere.

There is actually a project that aims to do this. It's called kdump,
and it just kexec's a secondary system stowed away in memory. It's a
radical approach, but you can't argue with its power. It might be
interesting to see if something along these lines could also be done
on our platform.

> The secret to customer satisfaction is proactively thinking about what
> might go wrong, and providing the user with information that might be
> helpful.  Sometimes that goes in a different direction from expending
> minimal effort in the kernel.

This time we're lucky - all it takes is to ask Carsten to make his
daemon set the backlight to not-quite-pitch-dark. Everybody wins:
you get the user feedback, Carsten gets a kernel that always does
what he tells it to do, Andy can avoid yet another "your policy
patch is oh so wrong" flame war, and Sean and I get a kernel not
cluttered with policy ;-)

- Werner