Re: [PATCH v3 2/4] dt-bindings: Add TI SCI PM Domains

From: Dave Gerlach
Date: Fri Jan 13 2017 - 15:28:58 EST


On 01/13/2017 01:25 PM, Rob Herring wrote:
On Thu, Jan 12, 2017 at 9:27 AM, Dave Gerlach <d-gerlach@xxxxxx> wrote:
Rob,

On 01/11/2017 03:34 PM, Rob Herring wrote:

On Mon, Jan 9, 2017 at 11:57 AM, Dave Gerlach <d-gerlach@xxxxxx> wrote:

Rob,

On 01/09/2017 11:50 AM, Rob Herring wrote:


On Wed, Jan 04, 2017 at 02:55:34PM -0600, Dave Gerlach wrote:


Add a generic power domain implementation, TI SCI PM Domains, that
will hook into the genpd framework and allow the TI SCI protocol to
control device power states.

Also, provide macros representing each device index as understood
by TI SCI to be used in the device node power-domain references.
These are identifiers for the K2G devices managed by the PMMC.

Signed-off-by: Nishanth Menon <nm@xxxxxx>
Signed-off-by: Dave Gerlach <d-gerlach@xxxxxx>
---
v2->v3:
Update k2g_pds node docs to show it should be a child of pmmc
node.
In early versions a phandle was used to point to pmmc and docs
still
incorrectly showed this.

.../devicetree/bindings/soc/ti/sci-pm-domain.txt | 59 ++++++++++++++
MAINTAINERS | 2 +
include/dt-bindings/genpd/k2g.h | 90
++++++++++++++++++++++
3 files changed, 151 insertions(+)
create mode 100644
Documentation/devicetree/bindings/soc/ti/sci-pm-domain.txt
create mode 100644 include/dt-bindings/genpd/k2g.h

diff --git a/Documentation/devicetree/bindings/soc/ti/sci-pm-domain.txt
b/Documentation/devicetree/bindings/soc/ti/sci-pm-domain.txt
new file mode 100644
index 000000000000..4c9064e512cb
--- /dev/null
+++ b/Documentation/devicetree/bindings/soc/ti/sci-pm-domain.txt
@@ -0,0 +1,59 @@
+Texas Instruments TI-SCI Generic Power Domain
+---------------------------------------------
+
+Some TI SoCs contain a system controller (like the PMMC, etc...) that
is
+responsible for controlling the state of the IPs that are present.
+Communication between the host processor running an OS and the system
+controller happens through a protocol known as TI-SCI [1]. This pm
domain
+implementation plugs into the generic pm domain framework and makes
use
of
+the TI SCI protocol power on and off each device when needed.
+
+[1] Documentation/devicetree/bindings/arm/keystone/ti,sci.txt
+
+PM Domain Node
+==============
+The PM domain node represents the global PM domain managed by the
PMMC,
+which in this case is the single implementation as documented by the
generic
+PM domain bindings in
Documentation/devicetree/bindings/power/power_domain.txt.
+Because this relies on the TI SCI protocol to communicate with the
PMMC
it
+must be a child of the pmmc node.
+
+Required Properties:
+--------------------
+- compatible: should be "ti,sci-pm-domain"
+- #power-domain-cells: Must be 0.
+
+Example (K2G):
+-------------
+ pmmc: pmmc {
+ compatible = "ti,k2g-sci";
+ ...
+
+ k2g_pds: k2g_pds {
+ compatible = "ti,sci-pm-domain";
+ #power-domain-cells = <0>;
+ };
+ };
+
+PM Domain Consumers
+===================
+Hardware blocks that require SCI control over their state must provide
+a reference to the sci-pm-domain they are part of and a unique device
+specific ID that identifies the device.
+
+Required Properties:
+--------------------
+- power-domains: phandle pointing to the corresponding PM domain node.
+- ti,sci-id: index representing the device id to be passed oevr SCI to
+ be used for device control.



As I've already stated before, this goes in power-domain cells. When you
have a single thing (i.e. node) that controls multiple things, then you
you need to specify the ID for each of them in phandle args. This is how
irqs, gpio, clocks, *everything* in DT works.



You think the reasoning for doing it this way provided by both Ulf and
myself on v2 [1] is not valid then?

From Ulf:

To me, the TI SCI ID, is similar to a "conid" for any another "device
resource" (like clock, pinctrl, regulator etc) which we can describe
in DT and assign to a device node. The only difference here, is that
we don't have common API to fetch the resource (like clk_get(),
regulator_get()), but instead we fetches the device's resource from
SoC specific code, via genpd's device ->attach() callback.


Sorry, but that sounds like a kernel problem to me and has nothing to
do with DT bindings.

From me:

Yes, you've pretty much hit it on the head. It is not an index into a
list
of genpds but rather identifies the device *within* a single genpd. It is
a
property specific to each device that resides in a ti-sci-genpd, not a
mapping describing which genpd the device belongs to. The generic power
domain binding is concerned with mapping the device to a specific genpd,
which is does fine for us, but we have a sub mapping for devices that
exist
inside a genpd which, we must describe as well, hence the ti,sci-id.


So to summarize, the genpd framework does interpret the phandle arg as an
index into multiple genpds, just as you've said other frameworks do, but
this is not what I am trying to do, we have multiple devices within this
*single* genpd, hence the need for the ti,sci-id property.


Fix the genpd framework rather than work around it in DT.


I still disagree that this has nothing to do with DT bindings, as the
current DT binding represents something different already. I am trying to
extend it to give me additional information needed for our platforms. Are
you saying that we should break what the current DT binding already
represents to mean something else?

No idea because what's the current binding? From the patch, looks like
a new binding to me.

Yes, ti,sci-id is a new binding. I am referring to the current meaning of the "power-domains" binding, which is where you are asking this property to be added, in "power-domains" cells. This is documented here [1] in the kernel, although looking at it I must admit it is not very clear.

The power-domains cell represents an offset into an array of power domains, if you choose to use it. That's what the genpd framework is hard coded to interpret it as. This is correct, as it is an index into a static list of power domains, used to identify which power domain a device belongs to, which is exactly what the genpd framework itself is concerned with. This is already how it is used in the kernel today.

My ti,sci-id is not an index into a list of power domains, so it should not go in the power-domains cells and go against what the power-domains binding says that the cell expects. We have one single power domain, and the new ti,sci-id binding is not something the genpd framework itself is concerned with as it's our property to identify a device inside a power domain, not to identify which power domain it is associated with.

Regards,
Dave

[1] Documentation/devicetree/bindings/power/power_domain.txt


Rob