Re: [PATCH] of: provide of_platform_unpopulate()

[Sebastian Andrzej Siewior <bigeasy@linutronix.de>]

* Grant Likely | 2013-07-24 15:19:58 [+0100]:

>> Was there more breakage than imx6 and amba devices? Your first version
>> had a fallback case for powerpc. Couldn't we do just allow that for more
>> than just powerpc? I'd much rather see some work-around within the core
>> DT code with a warning to prevent more proliferation than putting this
>> into drivers.
>
>It's tricky stuff. I've not figured out a solution I'm happy with.
>Trying to figure out when to apply a work around is hard because the
>resource reservation makes assumptions about the memory range layout
>that doesn't match the assumptions made by device tree code.

I can't really follow. Do you have a simple at hand?

>One /possible/ option is to not add the resources to the devices at all
>when the device is registered and instead resolve them right at bind
>time. Jean Christophe proposed doing this already to solve a different
>problem; obtaining resources that require other drivers to be probed
>first. If the resources are resolved at .probe() time, then the resource
>registration problem should also go away.
>
>The downside to that approach is that it makes each deferred probe more
>expensive; potentially a *lot* more expensive depending on how much work
>the xlate functions have to do. It would be worth prototyping though to
>see how well it works.

So you say defer the io ressources until the device-tree device is
actually probed. I don't really understand why that defer part should
solve the problem but I would try and see how it goes.
Jean-Christophe proposed that only, that means no patches yet, right?

>g.

Sebastian