[PATCH 2/2] [POWERPC] MPC8349E-mITX: use platform IDE driver for CF interface

[Sergei Shtylyov]

Scott Wood wrote:

>> Scott Wood wrote:

>>>>    Also, what mmio-ide in the compat properly means in the context 
>>>> of ide_platform which is able to handle both port and memory mapped 
>>>> IDE.

>>> I/O-space is only valid in the context of PCI, ISA, or similar buses, 
>>> and
>>> the bus-specific reg format indicates whether it's mmio-space or
>>> io-space.

>>    You could save time on lecturing me (and use it to look on the 
>> driver ;-).

> Sorry, I misread the question as being a mismatch between the 
> capabilities of the device binding and the driver, not about the 
> specific compatible name.

    That too. :-)

> Something like "generic-ide" would probably be better.

    I strongly disagree with "generic" part. The generic IDE could only be 
said of 1:1 I/O mapped IDE ports, not about this fancy mapping.

>>> What is board specific about a set of standard IDE registers at a given

>>    The regisrer mapping used is highly non-standard.

> The gap between registers is nonstandard, but that's a fairly common 
> type of noncompliance in embedded-land, and probably merits being 

    That is only a common variation of embedded non-compliancy (which doesn't 
make it a compliancy. ;-)
    There are worse cases in the bi-endian land, even with the standard 8-bit 
regs and 1-byte stride. *Hopefully*, this driver could also support those...

> supported in a generic way.  I wouldn't call it "highly" nonstandard.

    Yeah, there are also 8250 "compatible" UARTs that use 32-bit memory 
accesses, and even worse -- with some registers mapped differently than on 
8250 (those can't be called compatible by any means), yet 8250.c drives all of 
them.  I'm not really sure it was such a good idea to merge, say Alchemy UART 
support into 8250.c.

> Is there some other non-standardness that I'm missing?

    *Hopefully*, none.  The original Kumar's driver pretended to handle 
byte-lane swapping too (but that was ugly :-).

>>    We're already in board specific code, so why the heck not? :-)

>>> various ns16550-compatibles out there as well?

>>    I never suggested that -- what I did suggest was make of_serial.c 
>> recognize certain chip types and register them with 8250 driver.

> What would be the advantage of maintaining a list of chips whose only 

    Nobody's talking about the advantages, just about the device tree accepted 
practices (which we've already tried to bypass with MTD node -- causing a lot 
of bashing until David Woodhouse came to help :-).

> difference is register spacing, rather than just using reg-shift and 
> being done with it?

    Please read the linuxppc-dev archive's threads following form David's 
patches.  Or maybe Segher could repeat this for you. ;-)

> -Scott

MBR, Sergei