powerpc_flash_init(), wtf!?

[Segher Boessenkool]

>>>   We weren't aware of the of_platform.c work when writing the MTD 
>>> support.
>>>   Note that this function usually probes only the specified set of 
>>> (SoC)
>>> busses, none of which usully contains NOR flash (which is located at 
>>> the
>>> root level).
>
>> The root level?  Um... I don't think so...
>
>     "Trust me". :-)
>     NOR flashes are at the same level as the "memory" node (where else 
> you
> expect them to appear I wonder?).

The "memory" node doesn't describe the RAM devices;
it describes the RAM address space, instead.  You can
have separate nodes for the actual devices.

Now for ROM/flash/NVRAM, nodes _can_ appear directly
under the root, but only if that is where they belong
on your platform (i.e., they sit directly on the "system
bus" (whatever that means on your platform); on most
platforms though, such devices are connected via some
I/O busses, so the nodes should appear under their
respective controllers.

>> I believe the arrangement is similar for most other 4xx systems.  More
>> PC or desktop like systems sometimes have boot flash connected to the
>> south bridge, which I believe puts it on the ISA bus, topologically
>> speaking.

Some have it on the LPC bus as an LPC device, some
have it on the LPC bus but accessed with a separate
protocol, some have it attached to another LPC device
(some "superio" typically), some have it attached
directly to the "south bridge".

>     Not exactly. Boot flash is mapped beyond ISA address space on 386+ 
> -- at
> the top of 4GB (where the "reset vector" is). Although it may be dual 
> mapped
> below 1MB as well (I'm starting to forget x86 :-).

Most "north bridges" have some bits that enable
translation of accesses in the "low bios" area to
the 4GB-minus-a-bit area.  There are many variations
and it all is a big mess :-)


Now, back to the case at hand -- it would be nice to
have a platform-independent way to probe the simple
case -- a single direct-mapped device -- but it isn't
obvious how to make that not clash with the not-so-simple
cases.  A helper function that does the work but is
only called by the platforms that want it would do, I
suppose?


Segher