[SNIP]

^^ the correct solution is to implement arch_initcall function
which will create fixed PHYs, and then leave only
snprintf(fpi->bus_id, 16, PHY_ID_FMT, 0, *data); part in the
fs_enet's find_phy().

Try add something like this to the fsl_soc.c (compile untested):

- - - -
static int __init of_add_fixed_phys(void)
{
	struct device_node *np;
	const u32 *prop;
	struct fixed_phy_status status = {};

	while ((np = of_find_node_by_name(NULL, "ethernet"))) {
		data  = of_get_property(np, "fixed-link", NULL);
		if (!data)
			continue;

		status.link = 1;
		status.duplex = data[1];
		status.speed  = data[2];

What about Pause and Asym_Pause? Dunno why so few, if any, eth drivers
impl. it, but the PHY lib supports it.
Even if fixed PHYs doesn't support it directly I think the OF interface
should have it.

    - fixed-link : <a b c d e> where a is emulated phy id - choose any,
      but unique to the all specified fixed-links, b is duplex - 0 half,
      1 full, c is link speed - d#10/d#100/d#1000, d is pause - 0 no pause,
      1 pause, d asym_pause - 0 no asym_pause, 1 asym_pause.

Jocke

		ret = fixed_phy_add(PHY_POLL, data[0], &status);
		if (ret)
			return ret;
	}

	return 0;
}
arch_initcall(of_add_fixed_phys);
- - - -

And remove fixed_phy_add() from the fs_enet. This should work
nicely and also should be ideologically correct. ;-)

quoted

How is this supposed to work for modules or for the
PPC_CPM_NEW_BINDING mode where the device tree is no longer scanned
during fs_soc initialization but during device initialization?

We should mark fixed.c as bool. Fake/virtual/fixed/platform PHYs
creation is architecture code anyway, can't be =m.

-- 
Anton Vorontsov
email: cbou@mail.ru
backup email: ya-cbou@yandex.ru
irc://irc.freenode.net/bd2