Hello.

Vitaly Bordug wrote:

quoted

quoted

  I acn undertand your complaint in the context of an OF driver
(which we don't have yet) but "mmio-ide" just means nothing to the
current driver, and it doesn't convery enough info on the
programming interface for the conceivable OF driver, it also does
need to know at least "reg-stride" (and maybe "reg-size" in case
only 16/32-bit accesses can be used).  Well, if such driver will be
written, I/O mapping support will probably be dropped from it, so
indeed, calling it mmio-ide.c would make sense.  But that can be
added when this driver is done, and for now 

quoted

I don't think the details of what Linux code currently exists should 
drive the device tree binding.  That the current patches use 
platform_device glue code is an implementation detail (and one I'd 
rather see go away, in favor of a driver that supports both 
platform_device and of_device).

quoted

quoted

I'd really prefer the board name to appear in the "compatible" prop
(to which "mmio-ide" can be appended)...

quoted

Sure, that's always good...  it was the "instead" that I objected to.

Hmmm. So what is finally suggested devicetree node for this beast - can somebody refine?

I am a little bit confused about decided device_type

    My understanding is that "ata" has been already used, so there's no sense 
in introducing "ide". Anyway, Segher will just say that "device_type" shoudn't 
matter and even be present. ;-)

 > and compatible fields...

    In my understanding, as "mmio-ide" currectly makes no sense, it shouldn't 
even appear there.  And since "mpc8349emitx-cf" (or whatever would be most 
generic name for those boards with the same type of CF IDE mapping) should be 
imply the shift value, this property should also be optional, i.e. passing 
hard coded value with platform_device would do.

MBR, Sergei