On 01/27/14 15:45, Thomas Petazzoni wrote:

On Sat, 25 Jan 2014 19:34:10 +0100, Sebastian Hesselbarth wrote:

quoted

Allocating the pinctrl resource in common pinctrl-mvebu was a misdesign,
as it does not allow SoC specific parts to access the allocated resource.
This moves resource allocation from mvebu_pinctrl_probe to SoC specific
_probe functions and passes the base address to common pinctrl driver
instead.

Signed-off-by: Sebastian Hesselbarth <sebastian.hesselbarth@gmail.com>

I definitely agree with that: I had the same problem several months ago
when I started doing the pinctrl driver for Orion5x, which has a
non-linear MPP register set.

However, I'd like this to go a little bit further if possible. See
below.

Agreed.

quoted

-	return mvebu_pinctrl_probe(pdev);
+	return mvebu_pinctrl_probe(pdev, base);

I think there is no need to pass "base" to mvebu_pinctrl_probe(). The
only reason we have this is because the base gets stored in the
mvebu_pinctrl structure so that the mvebu_common_mpp_get() and
mvebu_common_mpp_set() functions that are the default behavior
for mvebu_pinconf_group_get() and mvebu_pinconf_group_set() work
properly.

Shouldn't we turn these functions mvebu_common_mpp_get() and
mvebu_common_mpp_set() into helper functions, accessible from the
per-SoC pinctrl drivers, so that they can easily implement their
->mpp_get() and ->mpp_set() callbacks?

Sounds reasonable to do so. I have a look at removing the base address
from common.c completely.

Sebastian

This way, the "base" thing is completely owned by the per-SoC driver,
which would be more logical I believe.