On 2014-11-28 23:22, Arnd Bergmann wrote:

On Friday 28 November 2014 23:09:09 Stefan Agner wrote:

quoted

On 2014-11-28 22:24, Arnd Bergmann wrote:

quoted

On Friday 28 November 2014 22:02:01 Stefan Agner wrote:

quoted

quoted

If the SRC is also capable of resetting individual blocks instead of just
the entire machine, it would be a reset driver in drivers/reset instead.

Beside the system reset, there is only a mask functionality for the
watchdogs (there are two watchdogs, one for Cortex-A5 and one for the
M4). This makes the SRC module in the Vybrid a bit different then what
is available on other i.MX SoC's...

If you already have the watchdog registers in there and want to have
a watchdog driver too, the easiest way would be to register the reboot
handler from the watchdog driver.

Hm, not sure we speak about the same here. The SRC module has two
(multi-)bit fields to mask the watchdog reset event for each watchdog.
Beside that, there are two full watchdog register maps, which are in
different areas. There is already a driver for this watchdogs. I'm not
sure what the idea behind this is exactly, I guess it would easily allow
to (temporary) mask the other CPU's watchdog. However, I don't think we
need that functionality, so I don't care about that right now.

Ok, I see, thanks for the clarification!

quoted

There is also a restart handler in the watchdog driver, but I prefer to
use the reset capabilities of the SRC since it has immediate effect.

Lets get to the big picture again: I could register the whole SRC
register map as a syscon device and then access the registers from my
suspend/resume implementation later on. And similar in the restart
driver, I would use syscon_regmap_lookup_by_compatible to check if it
contains the vf610-src compatible string and register the restart
driver/handler if available.

Correct, and also in the watchdog driver, I guess.

Instead of syscon_regmap_lookup_by_compatible, please use
syscon_regmap_lookup_by_phandle and put the link to the syscon
device into the device accessing it.

Also, see if you can use or extend drivers/power/reset/syscon-reboot.c
for your use case.

Nice, this allows to do the reset with almost no code. The only thing
which might be a problem is the priority: I used 192 since 128 is
already used by the watchdog driver. Maybe we can change that
syscon-reboot is a bit above watchdogs by default, or maybe even a dt
property? I will try to use that driver for v2. Thx! 

--
Stefan