sdhci_add_host() calls regulator_get(..., "vmmc"), which under device-
tree calls of_get_regulator(), which contains:
regnode = of_parse_phandle(dev->of_node, prop_name, 0);
if (!regnode) {
dev_warn(dev, "%s property in node %s references invalid phandle",
prop_name, dev->of_node->full_name);
return NULL;
}
Thus if the vmmc-supply property is missing, because a board has no
regulator for a particular SD controller (power is always on), then
the warning is printed.
Can we eliminate this warning in of_get_regulator(), and let clients
Control whether they warn when a regulator isn't found, if they think
one is mandatory? I think I'd prefer this option; it's consistent with
the non-DT path in regulator_dev_lookup().
Or, should I set up dummy regulators in device tree to cover this case,
such that an SD controller's vmmc-supply always points at a valid phandle.
In which case, I'd have to add DT support to the dummy regulator.
Thanks.
--
nvpublic
On Fri, Feb 24, 2012 at 01:35:08PM -0800, Stephen Warren wrote:
> Can we eliminate this warning in of_get_regulator(), and let clients
> Control whether they warn when a regulator isn't found, if they think
> one is mandatory? I think I'd prefer this option; it's consistent with
> the non-DT path in regulator_dev_lookup().
This would be sensible but...
> Or, should I set up dummy regulators in device tree to cover this case,
> such that an SD controller's vmmc-supply always points at a valid phandle.
> In which case, I'd have to add DT support to the dummy regulator.
...you really ought to be doing this anyway (except with the fixed
voltage regulator, the dummy regulator is for stubbing out regulator
support entirely on systems with just one or two software controlled
regulators or where you don't have any information on the board design).
The MMC regulator usage is a bit of a mess for historical reasons, it
shouldn't really be conditional.