On Fri, May 24, 2013 at 12:40:26AM +0200, Sebastian Hesselbarth wrote:

On 05/23/2013 08:40 PM, Jason Cooper wrote:

quoted

On Thu, May 23, 2013 at 11:53:57AM -0600, Jason Gunthorpe wrote:

quoted

On Thu, May 23, 2013 at 01:23:39PM -0400, Jason Cooper wrote:

quoted

Shouldn't it rather be

compatible = "marvell,kirkwood-eth", "marvell,orion-eth";

Not sure about orion-eth?

Sorry, yep, one or the other.

Jason, Jason,

For a second, I read this as "tsk tsk tsk..." ;-)

sorry I didn't came back to this conversation earlier. I already
reworked the patch to rely on
of_device_is_compatible(.."marvell,kirkwood-eth"..). This is a
kirkwood only thing as other Orions cannot do clock gating or
retain critcal register content (Dove). I will stick with orion-eth
for all other and maybe introduce new compatible strings (and new
fixes) as soon as issues surface.

Okay, as I mentioned to Jason, I would like to test 6282 before we
settle on this path.  Other than that, I'm fine with it.

quoted

quoted

quoted

I'm inclined to go with of_machine_is_compatible() since the only
concrete difference we know is that the tweak is needed on kirkwood and
nowhere else.

But there is a larger problem here then just this one bit.

The PSC1 register must be set properly for the board layout, and today
we rely on the bootloader to set it. In fact, even with Sebastian's
change the ethernet port won't work without bootloader
intervention. The PortReset bit should also be cleared by the driver
(and it is only present on some variants of this IP block,
apparently).

Actually, fixing modular scenarios is only for the sake of multiarch
someday. I don't see the point in running current kernel without eth
compiled in _on a NAS SoC_ ;)

Good point, but if the eth can be gated to save power, we shouldn't
limit the user's ability just because the SoC is primarily in NAS's.

thx,

Jason.