aux button interrupt to pmu?

Werner Almesberger werner at openmoko.org
Fri Jun 6 12:29:13 CEST 2008


Andy Green wrote:
> In fact there should be limited trouble it can cause since we jump to a
> wake address in a fixed CPU register that survives suspend,

Yes, but there's still quite a lot of code that gets executed before
this. What if there are differences in what this one does ? We already
have a small one, namely the wakeup reason. (The "jump to GSTATUS2" is
as old as time, so NOR has that one as well.)

And it gets worse: even if there's no incompatibility between u-boot in
NAND and NOR, AUX may very well get deasserted between the time the CPU
decides whether to execute SteppingStone or not, and the time before we
jump back into the kernel. In this case, the content of the BootSRAM is
unpredictable, and we switch right into it for execution.

I can't predict how often this race condition would happen, and how
often one would find valid data in the BootSRAM, but the fact that the
slightest touch of AUX would make the system execute this code path,
and there's no debouncing, latching, or similar, makes it plausible
that this would lead to spurious crashes from time to time.

Also, I don't know if SteppingStone takes it kindly if there is a
transition of the NOR/NAND selection while it's doing its work.

So I think the bug is in that new spec trying to lead us somewhere we
really don't want to go :)

- Werner




More information about the openmoko-kernel mailing list