netlink: specs: finish up operation enum-models

I had a (bright?) idea of introducing the concept of enum-models
to account for all the weird ways families enumerate their messages.
I've never finished it because generating C code for each of them
is pretty daunting. But for languages which can use ID values directly
the support is simple enough, so clean this up a bit.

"unified" model is what I recommend going forward.
"directional" model is what ethtool uses.
"notify-split" is used by the proposed DPLL code, but we can just
make them use "unified", it hasn't been merged :)

Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Documentation/netlink/genetlink-c.yaml

@@ -218,9 +218,7 @@ properties:
           to a single enum.
           "directional" has the messages sent to the kernel and from the kernel
           enumerated separately.
-          "notify-split" has the notifications and request-response types in
+        enum: [ unified ]
-          different enums.
-        enum: [ unified, directional, notify-split ]
       name-prefix:
         description: |
           Prefix for the C enum name of the command. The name is formed by concatenating

Documentation/netlink/genetlink-legacy.yaml

@@ -241,9 +241,7 @@ properties:
           to a single enum.
           "directional" has the messages sent to the kernel and from the kernel
           enumerated separately.
-          "notify-split" has the notifications and request-response types in
+        enum: [ unified, directional ] # Trim
-          different enums.
-        enum: [ unified, directional, notify-split ]
       name-prefix:
         description: |
           Prefix for the C enum name of the command. The name is formed by concatenating
@@ -307,6 +305,13 @@ properties:
                       type: array
                       items:
                         type: string
+                    # Start genetlink-legacy
+                    value:
+                      description: |
+                        ID of this message if value for request and response differ,
+                        i.e. requests and responses have different message enums.
+                      $ref: '#/$defs/uint'
+                    # End genetlink-legacy
                 reply: *subop-attr-list
                 pre:
                   description: Hook for a function to run before the main callback (pre_doit or start).

Documentation/netlink/genetlink.yaml

@@ -188,9 +188,7 @@ properties:
           to a single enum.
           "directional" has the messages sent to the kernel and from the kernel
           enumerated separately.
-          "notify-split" has the notifications and request-response types in
+        enum: [ unified ]
-          different enums.
-        enum: [ unified, directional, notify-split ]
       name-prefix:
         description: |
           Prefix for the C enum name of the command. The name is formed by concatenating

Documentation/userspace-api/netlink/genetlink-legacy.rst

@@ -74,6 +74,88 @@ type. Inside the attr-index nest are the policy attributes. Modern
 Netlink families should have instead defined this as a flat structure,
 the nesting serves no good purpose here.
 
+Operations
+==========
+
+Enum (message ID) model
+-----------------------
+
+unified
+~~~~~~~
+
+Modern families use the ``unified`` message ID model, which uses
+a single enumeration for all messages within family. Requests and
+responses share the same message ID. Notifications have separate
+IDs from the same space. For example given the following list
+of operations:
+
+.. code-block:: yaml
+
+  -
+    name: a
+    value: 1
+    do: ...
+  -
+    name: b
+    do: ...
+  -
+    name: c
+    value: 4
+    notify: a
+  -
+    name: d
+    do: ...
+
+Requests and responses for operation ``a`` will have the ID of 1,
+the requests and responses of ``b`` - 2 (since there is no explicit
+``value`` it's previous operation ``+ 1``). Notification ``c`` will
+use the ID of 4, operation ``d`` 5 etc.
+
+directional
+~~~~~~~~~~~
+
+The ``directional`` model splits the ID assignment by the direction of
+the message. Messages from and to the kernel can't be confused with
+each other so this conserves the ID space (at the cost of making
+the programming more cumbersome).
+
+In this case ``value`` attribute should be specified in the ``request``
+``reply`` sections of the operations (if an operation has both ``do``
+and ``dump`` the IDs are shared, ``value`` should be set in ``do``).
+For notifications the ``value`` is provided at the op level but it
+only allocates a ``reply`` (i.e. a "from-kernel" ID). Let's look
+at an example:
+
+.. code-block:: yaml
+
+  -
+    name: a
+    do:
+      request:
+        value: 2
+        attributes: ...
+      reply:
+        value: 1
+        attributes: ...
+  -
+    name: b
+    notify: a
+  -
+    name: c
+    notify: a
+    value: 7
+  -
+    name: d
+    do: ...
+
+In this case ``a`` will use 2 when sending the message to the kernel
+and expects message with ID 1 in response. Notification ``b`` allocates
+a "from-kernel" ID which is 2. ``c`` allocates "from-kernel" ID of 7.
+If operation ``d`` does not set ``values`` explicitly in the spec
+it will be allocated 3 for the request (``a`` is the previous operation
+with a request section and the value of 2) and 8 for response (``c`` is
+the previous operation in the "from-kernel" direction).
+
 Other quirks (todo)
 ===================