[U-Boot] [PATCH 03/11] DM: add block controller core

[Marek Vasut]

Dear Pavel Herrmann,

> On Friday 21 of September 2012 17:39:21 Marek Vasut wrote:
> > Dear Pavel Herrmann,
> > 
> > [...]
> > 
> > > > Can't the old driver just have a compat section in them? Like I did
> > > > with serial stuff:
> > > > 
> > > > 1) rename the internal functions to ${driver}_${function_name} from
> > > > pure ${function_name} and introduce section which behaves as a
> > > > wrapper (implement ${function_name} calling
> > > > ${driver}_${function_name} ). 2) Add your DM goo, implement #ifdef
> > > > around it so either the compat section or DM section is enabled.
> > > 
> > > I actually did something of this sort, see [4/11], with less touching.
> > > 
> > > the problem is that while SATA drivers are easy to convert, IDE ones
> > > are not. I would actually propose to do a ide_legacy driver (mostly
> > > out of the code currently in common/cmd_ide.c), and keep it as the
> > > only option until IDE dies completely.
> > 
> > IDE will be around for a LONG time.
> > 
> > You introduce that CONFIG_SYS_SATA_LEGACY for no reason, if you did it as
> > said above, simple CONFIG_DM would suffice as the drivers would be
> > intacts with DM disabled. Note the compiler will opt-out these proxy
> > calls.
> > 
> > Besides, with this approach of yours, you need to enable SATA_LEGACY for
> > every single board now, introducing a lot of churn into the patches and
> > if it's not defined, every board using SATA is broken, right?
> 
> No, if you dont define CONFIG_DM, you get the old way of interacting with
> disks. only if you define CONFIG_DM you only need CONFIG_SATA_LEGACY to
> plug old SATA drivers into DM codepaths

So if you enable DM for a board, you also need to enable this compat layer? My 
opinion is, that you either enable DM on the board and the board is ready for it 
or don't enable it.

Would you need the compat layer at all were you to follow my advice?

The whole idea goes deeper, see if you prepended this patchset with such a 
conversion as above, you'd already have a readily defined structure of blockdev 
operations in each driver to use in this patchset. This patchset would then 
really only be the DM stuff. The DM-part addition to the drivers would be 
mechanical.

> > > > How does that work? It's much cleaner.
> > > > 
> > > > > Pavel Herrmann
> > > > 
> > > > Best regards,
> > > > Marek Vasut
> > 
> > Best regards,
> > Marek Vasut

Best regards,
Marek Vasut