Thank you for the patch. Overall - LGTM, however I got a request from Amazon, to wait with landing after their release 2.6.2 patches (a matter of day or two).
If you wish, I can take care of their patches and this one - is that ok for you?

LGTM, thanks!

The default config Armada 38x should run with is sys/arm/conf/GENERIC. Can you please add a commit with that?

In D36793#837252, @bz wrote:

In D36793#837249, @mw wrote:

That looks really good, thanks a lot!

One idea:
OF_device_from_xref(OF_xref_from_node(sc->sfp))
I think it would be useful to have a wrapper: OF_device_from_node

I was thinking long about that and I had actually done an OF_dev_from_xref_node() to not lose the xref but I decided it's a few one-time calls ... If we do that we should do it seperately and fix the entire tree (as this is a more-than-once pattern).

That looks really good, thanks a lot!

In D36793#837060, @dsl wrote:

@mw, this diff is supposed to substitute some pieces from sys/dev/dpaa2/dpaa2_mc_fdt.c and sys/dev/dpaa2/dpaa2_mc_acpi.c (https://reviews.freebsd.org/D36638):

static int
dpaa2_mac_dev_attach(device_t dev)
{
...
hobj = NULL;
	s = device_get_property(dev, "phy-handle", &hobj, sizeof(hobj),
	    DEVICE_PROP_ANY);
	if (s > 0) {
		if (hobj->Type != ACPI_TYPE_PACKAGE) {
			device_printf(dev, "Cannot get 'phy-handle' "
			    "package obj, not ACPI_TYPE_PACKAGE\n");
			goto out;
		}
		if (hobj->Package.Count != 1) {
			device_printf(dev, "Cannot get 'phy-handle' "
			    "package obj, count %d\n", hobj->Package.Count);
			goto out;
		}
		pobj = &hobj->Package.Elements[0];
		if (pobj == NULL) {
			device_printf(dev, "Cannot get 'phy-handle' "
			    "package pobj is NULL\n");
			goto out;
		}
		if (pobj->Type != ACPI_TYPE_LOCAL_REFERENCE) {
			device_printf(dev, "Cannot get 'phy-handle' package "
			    "pobj, no ACPI_TYPE_LOCAL_REFERENCE\n");
			goto out;
		}
		sc->phy_channel = acpi_GetReference(NULL, pobj);
		if (sc->phy_channel == NULL) {
			device_printf(dev, "Cannot get 'phy-handle' reference "
			    "handle ph is NULL\n");
			goto out;
		}
	}
...
}

You've requested such change in https://reviews.freebsd.org/D36677#833135, haven't you?

Overall, LGTM - thank you for this patch! One nit/suggestion inline. Do you have an example code locally that shows real-life usage - I would like to see a code snippet/diff.

On the occasion, it would be great to extend device_get_property or add a new routine (e.g. device_get_reference) for obtaining phandles in a unified way. That would allow minimizing the code related to HW description in https://reviews.freebsd.org/D36638 (and some future code I'm working on :) ).

I agree the wrapper routines should not be suffixed with _locked. However, because we have a different case than e6000sw_readreg/e6000sw_readreg_wrapper, I propose a following change:
new functions should take the original name (the DEVMETHODs declarations will remain intact). i.e.:

• e6000sw_writephy_wrapper -> e6000sw_writephy
• e6000sw_readphy_wrapper -> e6000sw_readphy

LGTM, thanks!

Nice catch!

Thanks for splitting all the patches!