[PATCH v3 7/7] ARM: brcmstb: dts: add a reference DTS for Broadcom 7445

[Michal Simek]

On 01/16/2014 12:19 PM, Arnd Bergmann wrote:
> On Thursday 16 January 2014, Marc C wrote:
>>> And then you can add a regular device driver to drivers/reset that provides a
>>> device_reset() API to other drivers, or a system-reset function to be registered as
>>> arm_pm_restart. This driver would use syscon_regmap_lookup_by_phandle() to get access
>>> to a regmap pointer, and then use either hardcoded offsets into the regmap, or get
>>> those offsets from numbers in the devicetree, as provided in the example above.
>>
>> I was able to port a standalone "reboot" driver using syscon + regmap. However, for the
>> SMP initialization case, it turns out that syscon is configured after SMP init. Do you
>> have any advice for this type of situation?
>>
>> I'd hate to go around in circles, but without resorting to hard-coded offsets, perhaps I
>> can just add the remaining "non-regmap" register offsets in the DT?
> 
> You are not the first one to stumble over this problem. There are
> two ways to get out of it, and we should probably implement both in
> the long run:
> 
> 1. Other platforms also require the syscon driver to be active before
> the regular device driver probing starts. Michal Simek has the same
> issue in the zynq clock driver that you have for SMP initialization.
> We have talked about this with Mark Brown already, and I think we will
> find a solution for this in the end, but it's not as straightforward
> as I first hoped. We can probably use help in this area.

I wouldn't say issue because we have created workaround which should be acceptable.
For more info.
https://lkml.org/lkml/2014/1/6/349  patch 2/7 is that what we have done
to get it work.
Also I hope that this can go to mainline before any nice solution will be find
not to stop work on others drivers which use regmap.

But there is definitely a space for doing this better.


> 2. There is actually no reason for the SMP code to be called this early,
> and multiple platforms would like to move SMP init to a later stage.
> Stephen Boyd has recently started reworking the way we do SMP init,
> and he may have some more insight.
> 
> In the meantime, I'd suggest you do the binding under the assumption
> that it will work eventually, and then work around the current limitations
> in the SMP code by looking for the device nodes you need and map them
> manually, as you did in the previous versions of your patch set.

Yep, that's exactly what we have done just because of other driver developing
till this is resolved.

Thanks,
Michal