On 06/23/2014 07:42 AM, Arnd Bergmann wrote:

On Monday 23 June 2014 07:30:56 Guenter Roeck wrote:

quoted

On 06/23/2014 03:31 AM, Maxime Ripard wrote:

quoted

On Thu, May 22, 2014 at 02:12:07PM -0700, Guenter Roeck wrote:

quoted

On Thu, May 22, 2014 at 10:34:44PM +0200, Maxime Ripard wrote:

quoted

On Mon, May 19, 2014 at 05:04:22PM +0200, Maxime Ripard wrote:

The patches _are_ in my watchdog-next branch and get some coverage from
both my auto-builders and from Fenguang's build robots, so while they are
not in linux-next, they are not completely in the dark either.

So, this patch finally didn't make it into 3.16. Great. Now, we can't
even reboot the boards.

Given how it's just impossible to get something merged reliably
through the watchdog tree, I guess I should just start merging the
patches through mine?

You can not really blame Wim here.

In this case, I suspect the major reason for not accepting the patch
is that I tried to provide a clean method / API for "reset through watchdog
subsystem", which went nowhere, in my understanding because someone objected
that it would be the wrong thing to do [1] and it didn't get approval /
acceptance from the arm maintainers. If it is wrong to reset the board
from the watchdog subsystem in a clean way, it is for sure even more wrong
to do it as you proposed in your patch.

My conclusion therefore is that all board reset code should move back out
of the watchdog subsystem, and that we should not accept such code in the
future. This is not my personal preference, but I do believe that we should
do it in a clean way or not at all.

Moved to where?

I certainly don't want it in the platform directories, and for arm64 we
intentionally don't have a place to put this stuff.

I have no idea, but setting the arm reset function pointer from a watchdog
driver doesn't seem like a good idea either. The arm code _does_ provide
and expect platform code to set the reset function, so having it in the arm
code would at least make more sense than expecting some unrelated driver
code to set it - especially since it is inherently racy [1].

drivers/power/reset would be an option, but then we have to solve the
problem of loading two drivers for one device first.

It sounds crazy and may be huge overkill, but one solution for this is
would be to use mfd to bridge the gap.

Guenter

---
[1] https://lkml.org/lkml/2014/5/15/838