[PATCH] [MTD] ofpart: Partitions at same address cannot have thesame name v3

[Segher Boessenkool]

>> Your special "partition" isn't really a partition then, is it.
>> Because of that, device nodes to represent your partitions doesn't
>> work very well.
>
> I think that they work pretty well.

With "real" devices you don't normally have two devices sitting at
the same address.  Since you _do_ have that with your partitions,
it doesn't fit the device node abstraction very well.

> Unfortunately, since
> 4b08e149c0e02e97ec49c2a31d14a0d3a02f8074 all the partiton must be
> named "partition"

Ah!  After writing my original reply, I read the documentation,
and this had me confused.

4b08e14 looks wrong to me.  It also didn't update the documentation.

>> You really want to use something else, a partition table on the
>> flash itself for example.  Or maybe the (platform? MTD?) code
>> should create a Linux device for the "full" device.
>
> This wont be very flexible. With the device tree aproach we can define
> partitions not only for the full flash, but also for two partitions
> merged, a partition inside a partition....

Well, except you have problems doing just that :-P

>>> Because two nodes of a device tree cannot have the same name,
>>
>> This isn't true.
>
> Are you sure?

Yes.

> This is what I get if I try to compile a device tree
> with two partitions with the same name starting at the same address.
>
> $ arch/powerpc/boot/dtc -O dtb -b 0 -p 1024
> arch/powerpc/boot/dts/q5-avnet.dts -o /tmp/kk
> DTC: dts->dtb  on file "arch/powerpc/boot/dts/q5-avnet.dts"
> ERROR (duplicate_node_names): Duplicate node name
> /plb at 0/flash at ff000000/partition at ff000000
> ERROR: Input tree has errors, aborting (use -f to force output)

I think this error is trying to say both the names and the unit
addresses are identical (and the latter exist), which isn't
technically incorrect, but not very useful either.

>>> but all the
>>> partitions must be named "partition",
>>
>> Bad binding, no cookie for you!
>
> Sorry, I dont understand want you want to say here.

I was trying to say that if the "partition" binding says the "name"
of every partition node should be "partition", then it is a very
bad binding.  The binding does _not_ say that though, the code is
wrong.

>> You cannot claim a name as generic as "partition" for this.  Pick
>> something else if you really must do things this way.
>
> You choose :)

Picking good names is hard work (more than 50% of all programming, heh).
You choose :-)  [Hint: something with a comma, maybe  
"linux,mtd,partition"
or similar -- I don't really care.  OTOH, it seems you won't need this
at all].


Segher