On Sat, Jan 11, 2014 at 07:55:57AM +0800, Thomas Petazzoni wrote:

Dear Jason Cooper,

On Fri, 10 Jan 2014 12:44:12 -0500, Jason Cooper wrote:

quoted

Lior, thanks for the clarification.  Simon, care to respin this with a
check for "marvell,armada-370-xp" root compatible string?  It should be
safe to say that if there is no DT, don't write the register.

Why check a root compatible string? If we do this, then we will have to
change the SATA driver for each and every new Marvell SoC that has this
PHY speed control register (and these new SOCs will not use the
"marvell,armada-370-xp" root compatible string, since they are clearly
not Armada 370/XP).

Instead, we should introduce an additional compatible string for the
SATA driver itself.

Agreed.  

quoted

Alternatively, we could do as Lior suggests, and create a new sata
compatible string.  But I think that is overkill.

No, this is the right thing to do, IMO.

quoted

Also, I'm growing more leery creating compatible strings for IP blocks
which are tied to the SoC revision.  If the IP block doesn't get issued
it's own version number/codename, we should just use the root compatible
strings to determine which SoC we are on.  I'll expand on this though as
I get caught up with Gregory's series's. 

I really disagree. It means that whenever a new root compatible string
is created for a new SOC, we will have to edit gazillions of drivers.
It's not because two SOCs have the same SATA IP that they are globally
compatible, and can therefore share the same root compatible strings.

Yes, you're right.  I really went off the deep end on that one.  Simon,
please do as Lior and Thomas suggested.

thx,

Jason.