On 08/23/2012 03:45 AM, Sebastian Hesselbarth wrote:

On 8/22/12, Stephen Warren [off-list ref] wrote:

quoted

On 08/22/2012 02:22 AM, Sebastian Hesselbarth wrote:

quoted

+++ b/drivers/pinctrl/pinctrl-mvebu.c

quoted

+static int mvebu_pinctrl_dt_node_to_map(struct pinctrl_dev *pctldev,

quoted

+	of_property_for_each_string(np, "marvell,pins", prop, group) {
+		struct mvebu_pinctrl_group *grp =
+			mvebu_pinctrl_find_group_by_name(pctl, group);
+
+		if (!grp) {
+			dev_err(pctl->dev, "unknown pin %s", group);
+			continue;
+		}
+
+		if (!mvebu_pinctrl_find_setting_by_name(pctl, grp, function)) {
+			dev_err(pctl->dev, "unsupported function %s on pin %s",
+				function, group);
+			continue;
+		}

The error-checking here isn't strictly necessary; the pinctrl core will
error-check all the names if/when the map entries are used. So, you
could probably get away with just assigning the pin/function names
directly into (*map) and hence remove some code here. Still, it's not a
big deal either way.

Hi Stephen,

as these error checks are only performed once, I'd like to have them there
as they give hints when someone uses wrong marvell,pins/function names
in DT.

quoted

quoted

+static int __devinit mvebu_pinctrl_dt_parse_function(struct mvebu_pinctrl
*pctl,

quoted

+static int __devinit mvebu_pinctrl_dt_parse(struct platform_device
*pdev,
+					struct mvebu_pinctrl *pctl)

I don't understand what those two functions do, or a good chunk of
probe(). It seems like they're setting up the ability to get/set pin
configrations in addition to pin muxing, although if that's so, it seems
odd that none of the binding documents specify any way of controlling
pin configuration from device tree. Are you expecting drivers to call
APIs such as pin_config_set() directly, rather than controlling pin
config through DT?

probe() receives some structs from the SoC specific driver stubs that
describe, e.g. how many muxable pins are available, which pingroup they
belong to (marvell,pins), what valid register values are and how they are
named (marvell,function). Although, all mvebu SoC share a good part of
pinmux register layout, there are especially for dove some very different
ways to mux pins. As mvebu pinctrl is a true mux-only driver, i.e. no
pin drive strength or pin direction configuration, we decided not to have
single pins controlable but only pingroups even if they comprise only a
single pin.

dt_parse() and dt_parse_function() build up structs that get used later on
in mvebu_pinmux_ops that require indexed functions. I can join them with
dt_node_to_map() but that would require incremental kzalloc for the
corresponding array. Or I could (functionally) leave dt_parse() to allocate
the array and only join dt_parse_function() with dt_node_to_map().

So everything you said makes sense, in that the core driver is
parameterizable and receives data from a SoC-variant-specific driver
indicating which pins/groups/functions are available. I'm still not sure
though why the translation of the pin/group/function structures passed
to probe into other data structures requires accessing the DT at all;
the set of available pins/groups/functions isn't configured through DT,
and doesn't need to be limited to only those options actually used in
DT, so can't you just process all the data that's passed to probe
without interaction with the DT?