On Tue, Jul 17, 2018 at 5:28 PM Andrew Jeffery [off-list ref] wrote:

On Tue, 17 Jul 2018, at 14:26, Benjamin Herrenschmidt wrote:

quoted

On Mon, 2018-07-16 at 07:55 -0600, Rob Herring wrote:

quoted

If that data is one set per SoC, then i'm not that concerned having
platform-specific data in the driver. That doesn't mean the driver is
not "generic". It's still not clear to me in this thread, how much of
this is board specific, but given that you've placed all the data in
an SoC dtsi file it seems to be all per SoC.

So Rob, I think that's precisely where the disconnect is.

I think we all (well hopefully) agree that those few tunables don't fit
in any existing subystem and aren't likely to ever do (famous last
words...).

Where we disagree is we want to make this parametrized via the DT, and
you want us to hard wire the list in some kind of SoC driver for a
given SoC family/version.

The reason I think hard wiring the list in the driver is not a great
solution is that that list in itself is prone to variations, possibly
fairly often, between boards, vendors, versions of boards, etc...

We can't know for sure every SoC tunable (out of the gazillions in
those chips) are going to be needed for a given system. We know which
ones we do use for ours, and that's a couple of handfuls, but it could
be that Dell need a slightly different set, and so might Yadro, or so
might our next board revision for that matter.

Now, updating the device-tree in the board flash with whatever vendor
specific information is needed is a LOT easier than getting the kernel
driver constantly updated. The device-tree after all is there to
reflect among other things system specific ways in which the SoC is
wired and configured. This is rather close...

Not sure this helps, but I feel that the proposal pretty closely matches what's described in Documentation/devicetree/bindings/mfd/mfd.txt. It's intended that nodes using the bindings I'm proposing are children of a 'compatible = "syscon", "simple-mfd"' node (this is the case with the features we're hoping to describe for our SoC). I should explicitly call that out.

IMO, any binding that has only those compatibles is not correct and a
specific compatible is needed. We should be able identify a specific
h/w block.

But to go on, "simple-mfd" is effectively an alias of "simple-bus", which means its intended to match child node compatibles to drivers provided by the kernel. If we shouldn't be describing minor features of a SoC in the devicetree, doesn't this invalidate the case for simple-mfd? What is the *correct* use of simple-mfd? When is it not used to expose minor features in set of "miscellaneous system registers"? Why doesn't this proposed case fit?

I'm no fan of simple-mfd either. I think it is abused and often a sign
of bad binding design. The general problem with MFD bindings is people
define child nodes based on what drivers they happen to have for some
OS. DT is not the only way to instantiate drivers. Child nodes are
really only needed when you have resources per child that need to be
defined. For example, if the MFD has an interrupt controller and
interrupts to sub-blocks or sub-blocks have their own clocks.
"simple-mfd" was for when the parent node has no driver or the child
nodes have no dependency on the parent.

Rob