Hi Rob,

On 08/16/2016 09:54 AM, Rob Herring wrote:

On Fri, Aug 12, 2016 at 11:43 AM, Bjorn Andersson
[off-list ref] wrote:

quoted

On Fri 12 Aug 09:00 PDT 2016, Suman Anna wrote:

quoted

On 08/12/2016 06:02 AM, Lee Jones wrote:

quoted

On Thu, 11 Aug 2016, Suman Anna wrote:

quoted

The remoteproc framework has been enhanced recently to provide new
OF API to retrieve a remoteproc handle by client drivers through a
standard 'rprocs' property in client nodes. The wkup_m3_ipc driver
has been using a custom 'ti,rproc' property until now, switch this
to use this new OF infrastructure. The wkup_m3_ipc driver has been
fixed up to provide backward compatibility for older DTBs by
replacing the older property with the standard newer property.

The wkup_m3_ipc binding has also been updated accordingly.

Signed-off-by: Suman Anna <redacted>
---
This patch is based on the discussion [1] for introducing new standard
OF API in remoteproc core series [2] and the exporting of couple of
functions in OF base code [3].

With this in place, the remoteproc core need not be looking for the
TI specific property anymore. I will submit the DTS changes once this
makes it to mainline.

regards
Suman

[1] https://patchwork.kernel.org/patch/9237767/
[2] http://marc.info/?l=linux-kernel&m=146894341701010&w=2
[3] https://patchwork.kernel.org/patch/9276181/

 .../devicetree/bindings/soc/ti/wkup_m3_ipc.txt     |  9 ++++++--
 drivers/soc/ti/wkup_m3_ipc.c                       | 26 +++++++++++++++++++++-
 2 files changed, 32 insertions(+), 3 deletions(-)

diff --git a/Documentation/devicetree/bindings/soc/ti/wkup_m3_ipc.txt b/Documentation/devicetree/bindings/soc/ti/wkup_m3_ipc.txt
index 401550487ed6..2ea7dd91acff 100644
--- a/Documentation/devicetree/bindings/soc/ti/wkup_m3_ipc.txt
+++ b/Documentation/devicetree/bindings/soc/ti/wkup_m3_ipc.txt

@@ -23,12 +23,17 @@ Required properties:
                   with the Wakeup M3 processor
 - interrupts:             Contains the interrupt information for the wkup_m3
                   interrupt that signals the MPU.
-- ti,rproc:               phandle to the wkup_m3 rproc node so the IPC driver
+- rprocs:         phandle to the wkup_m3 rproc node so the IPC driver
                   can boot it.
 - mboxes:         phandles used by IPC framework to get correct mbox
                   channel for communication. Must point to appropriate
                   mbox_wkupm3 child node.

+Deprecated properties:
+----------------------
+- ti,rproc:               This property has been replaced with the "rprocs"
+                  property.
+
 Example:
 --------
 /* AM33xx */

@@ -48,7 +53,7 @@ Example:
                           compatible = "ti,am3352-wkup-m3-ipc";
                           reg = <0x1324 0x24>;
                           interrupts = <78>;
-                          ti,rproc = <&wkup_m3>;
+                          rprocs = <&wkup_m3>;
                           mboxes = <&mailbox &mbox_wkupm3>;
                   };

diff --git a/drivers/soc/ti/wkup_m3_ipc.c b/drivers/soc/ti/wkup_m3_ipc.c
index 8823cc81ae45..86085f9bf6a8 100644
--- a/drivers/soc/ti/wkup_m3_ipc.c
+++ b/drivers/soc/ti/wkup_m3_ipc.c

@@ -1,7 +1,7 @@
 /*
  * AMx3 Wkup M3 IPC driver
  *
- * Copyright (C) 2015 Texas Instruments, Inc.
+ * Copyright (C) 2015-2016 Texas Instruments, Inc.
  *
  * Dave Gerlach <d-gerlach@ti.com>
  *

@@ -390,6 +390,8 @@ static int wkup_m3_ipc_probe(struct platform_device *pdev)
   struct resource *res;
   struct task_struct *task;
   struct wkup_m3_ipc *m3_ipc;
+  struct property *nprop, *oprop;
+  const char nprop_name[] = "rprocs";

   m3_ipc = devm_kzalloc(dev, sizeof(*m3_ipc), GFP_KERNEL);
   if (!m3_ipc)

@@ -415,6 +417,28 @@ static int wkup_m3_ipc_probe(struct platform_device *pdev)
           return ret;
   }

+  /* provide compatibility for older DTBs using ti,rproc property */
+  nprop = of_find_property(dev->of_node, "rprocs", NULL);
+  if (!nprop) {
+          oprop = of_find_property(dev->of_node, "ti,rproc", NULL);
+          if (!oprop) {
+                  dev_err(&pdev->dev, "node is missing ti,rproc property\n");
+                  return -ENODEV;
+          }
+
+          nprop = kzalloc(sizeof(*nprop) + sizeof(nprop_name),
+                          GFP_KERNEL);
+          if (!nprop)
+                  return -ENOMEM;
+
+          nprop->name = (char *)nprop + sizeof(*nprop);
+          snprintf(nprop->name, sizeof(nprop_name), nprop_name);
+          nprop->value = oprop->value;
+          nprop->length = oprop->length;
+          of_update_property(dev->of_node, nprop);
+          of_remove_property(dev->of_node, oprop);
+  }
+

+1 for getting the functionality out of core code.

-100 for having to jump though all these hoops.

If you are going to keep all of this, I would at least tuck it away in
a header file or something.

Hiding it somewhere else won't make much of a difference though. If this
is such an eyesore, we still can do this cleanly without jumping through
these hoops. I would say this is a direct result of the removal of the
rproc_get_by_phandle API in Patch 2 of your OF remoteproc series [2].
That API is still an agnostic API. We can just add the of_rproc_get()
convenience helper in your Patch 1 and drop patch 2 completely with no
references to ti,rproc in the core code in the first place. The other
option, if we do want to remove that API is to add an additional
property name argument (NULL to mean "rprocs") to
of_rproc_get_by_index(), and it can be hidden away under of_rproc_get()
API - which is what I am expecting most of the new client users would
use anyway.

Once "rprocs" hits mainline, I will definitely switch over the
wkup_m3_ipc nodes to use that standard property and can fix this driver
independently.

We're stuck with this problem all over the place, as the world continues
to evolve we will have issues with DT being static. This has been
discussed many times before and the suggested solution is always what
you implemented here - make the code deal with both versions, preferably
by patching.

The fact that you had to export the of_ operations indicates that no-one
has tried this before and I'm happy you did. I'm however not happy about
the size of the chunk of code it takes to do this dance.

I think for this to be practical we need to provide higher level
operations for DT modification - in this case a of_rename_property().

@Rob, any comments on this?

I agree. Pantelis submitted some helpers in this area a while back
(for the changeset API IIRC). I believe they were mostly fine, but
needed some users.

Is this the series you are talking about?
http://marc.info/?l=devicetree&m=146341689512653&w=2

It looks like that series is effective only when OF_DYNAMIC is enabled.
Probably a dumb question, but is this limited to DT Overlays? This
particular usage is a one-time change (first-time module is insmod'd)
mainly to provide compatibility for older DTBs, thereafter we wouldn't
be required to make any changes. It is definitely not a bulk update and
we don't want to unroll the changes even if we removed the module.

regards
Suman