Hello Alexandre,


On Sat, Jan 24, 2015 at 12:17:38AM +0100, Alexandre Belloni wrote:

This is a rework, what is part of linux-3.10-at91 and not yet present in
mainline should be part of a following series. I would prefer not mixing
reworks and "new" functionalities (they have been present in the atmel
tree for a while but never mainlined).

I agree, I just didn't know a new series will follow, maybe I missed 
this point.

Maybe I am a bit too picky (or boring: if I am, please told me), but 
this series by itself adds regression to all users of >= 9x5 boards 
(sama5, ?) because it merges MEM target and MEM+SLOW_CLOCK target, which 
used to be too different target states, not selectable at runtime indeed 
but this is still in practice two different target states. Note that I 
am not saying that MEM target and MEM+SLOW_CLOCK target should not be 
merged, they should, absolutely ;-). For >= 9x5 boards (sama5, ?), MEM 
target works and MEM+SLOW_CLOCK target does not work, MEM and 
MEM+SLOW_CLOCK merge breaks MEM target for those boards.

There is however a good news !, at91_pm_verify_clocks() used to be 
called for MEM target without considering if it was MEM (~STANDBY) or 
MEM+SLOW_CLOCK. It means that all MEM target users can with very good 
chance go to a deeper sleep state without issue because 
at91_pm_verify_clocks() successfully checked on those boards and is why 
we can merge MEM and MEM+SLOW_CLOCK without adding a regression.

Care should be taken to pull-request at the same time both the rework 
and the above cited following series about slow clock support for all 
known boards so we don't break MEM target for a release cycle.

I would say that PM on 9x5, n12 and sama5 in mainline is clearly not 
well tested and is lagging behind the atmel tree.

Well, at the current mainline state, everything works fine for me except 
slow clock mode, I thoroughly checked everything else.


Sylvain