generalize async notification and add queued operation manager

CODEOWNERS

@@ -118,7 +118,7 @@
 /doc/scripts/                             @carlescufi
 /doc/guides/bluetooth/                    @joerchan @jhedberg @Vudentz
 /doc/reference/bluetooth/                 @joerchan @jhedberg @Vudentz
-/doc/reference/kernel/other/resource_mgmt.rst @pabigot
+/doc/reference/resource_management/       @pabigot
 /doc/reference/networking/can*            @alexanderwachter
 /drivers/debug/                           @nashif
 /drivers/*/*cc13xx_cc26xx*                @bwitherspoon

doc/reference/index.rst

@@ -6,6 +6,7 @@ API Reference
 .. toctree::
    :maxdepth: 1
 
+   misc/async_notify.rst
    bluetooth/index.rst
    kconfig/index.rst
    drivers/index.rst
@@ -18,6 +19,7 @@ API Reference
    peripherals/index.rst
    power_management/index.rst
    random/index.rst
+   resource_management/index.rst
    shell/index.rst
    storage/index.rst
    usb/index.rst

doc/reference/kernel/index.rst

@@ -116,7 +116,6 @@ These pages cover other kernel services.
    other/atomic.rst
    other/float.rst
    other/ring_buffers.rst
-   other/resource_mgmt.rst
    other/cxx_support.rst
    other/version.rst
    other/fatal.rst

doc/reference/misc/async_notify.rst

@@ -0,0 +1,25 @@
+.. _async_notification:
+
+Asynchronous Notification APIs
+##############################
+
+Zephyr APIs often include :ref:`api_term_async` functions where an
+operation is initiated and the application needs to be informed when it
+completes, and whether it succeeded.  Using :cpp:func:`k_poll()` is
+often a good method, but some application architectures may be more
+suited to a callback notification, and operations like enabling clocks
+and power rails may need to be invoked before kernel functions are
+available so a busy-wait for completion may be needed.
+
+This API is intended to be embedded within specific subsystems such as
+:ref:`resource_mgmt_onoff` and other APIs that support async
+transactions.  The subsystem wrappers are responsible for extracting
+operation-specific data from requests that include a notification
+element, and for invoking callbacks with the parameters required by the
+API.
+
+API Reference
+*************
+
+.. doxygengroup:: async_notify_apis
+   :project: Zephyr

doc/reference/resource_management/index.rst

@@ -0,0 +1,146 @@
+.. _resource_mgmt:
+
+Resource Management
+###################
+
+There are various situations where it's necessary to coordinate resource
+use at runtime among multiple clients.  These include power rails,
+clocks, other peripherals, and binary device power management. The
+complexity of properly managing multiple consumers of a device in a
+multithreaded system, especially when transitions may be asynchronous,
+suggests that a shared implementation is desirable.
+
+Zephyr provides managers for several coordination policies.  These
+managers are embedded into services that use them for specific
+functions.
+
+.. contents::
+    :local:
+    :depth: 2
+
+.. _resource_mgmt_onoff:
+
+On-Off Manager
+**************
+
+An on-off manager supports an arbitrary number of clients of a service
+which has a binary state.  Example applications are power rails, clocks,
+and binary device power management.
+
+The manager has the following properties:
+
+* The stable states are off, on, and error.  The service always begins
+  in the off state.  The service may also be in a transition to a given
+  state.
+* The core operations are request (add a dependency) and release (remove
+  a dependency). The service manages the state based on calls to
+  functions that initiate these operations.
+* The service transitions from off to on when first client request is
+  received.
+* The service transitions from on to off when last client release is
+  received.
+* Each service configuration provides functions that implement the
+  transition from off to on, from on to off, and optionally from an
+  error state to off.  Transitions that may put a calling thread to
+  sleep must be flagged in the configuration to support safe invocation
+  from non-thread context.
+* All operations are asynchronous, and are initiated by a function call
+  that references a specific service and is given client notification
+  data. The function call will succeed or fail. On success, the
+  operation is guaranteed to be initiated, but whether the operation
+  itself succeeds or fails is indicated through client notification.
+  The initiation functions can be invoked from pre-kernel, thread, or
+  ISR context.  In contexts and states where the operation cannot
+  be started the function will result in an error.
+* Requests to turn on may be queued while a transition to off is in
+  progress: when the service has turned off successfully it will be
+  immediately turned on again (where context allows) and waiting clients
+  notified when the start completes.
+
+Requests are reference counted, but not tracked. That means clients are
+responsible for recording whether their requests were accepted, and for
+initiating a release only if they have previously successfully completed
+a request.  Improper use of the API can cause an active client to be
+shut out, and the manager does not maintain a record of specific clients
+that have been granted a request.
+
+Failures in executing a transition are recorded and inhibit further
+requests or releases until the manager is reset. Pending requests are
+notified (and cancelled) when errors are discovered.
+
+Transition operation completion notifications are provided through the
+standard :ref:`async_notification`, supporting these methods:
+
+* Signal: A pointer to a :c:type:`struct k_poll_signal` is provided, and
+  the signal is raised when the transition completes. The operation
+  completion code is stored as the signal value.
+* Callback: a function pointer is provided by the client along with an
+  opaque pointer, and on completion of the operation the function is
+  invoked with the pointer and the operation completion code.
+* Spin-wait: the client is required to check for operation completion
+  using the :cpp:func:`onoff_client_fetch_result()` function.
+
+Synchronous transition may be implemented by a caller based on its
+context, for example by using :cpp:func:`k_poll()` to wait until the
+completion is signalled.
+
+.. doxygengroup:: resource_mgmt_onoff_apis
+   :project: Zephyr
+
+.. _resource_mgmt_queued_operation:
+
+Queued Operation Manager
+************************
+
+While :ref:`resource_mgmt_onoff` supports a shared resource that must be
+available as long as any user still depends on it, the queued operation
+manager provides serialized exclusive access to a resource that executes
+operations asynchronously.  This can be used to support (for example)
+ADC sampling for different sensors, or groups of bus transactions.
+Clients submit a operation request that is processed when the device
+becomes available, with clients being notified of the completion of the
+operation though the standard :ref:`async_notification`.
+
+As with the on-off manager, the queued resource manager is a generic
+infrastructure tool that should be used by a extending service, such as
+an I2C bus controller or an ADC.  The manager has the following
+characteristics:
+
+* The stable states are idle and processing.  The manager always begins
+  in the idle state.
+* The core client operations are submit (add an operation) and cancel
+  (remove an operation before it starts).
+* Ownership of the operation object transitions from the client to the
+  manager when a queue request is accepted, and is returned to the
+  client when the manager notifies the client of operation completion.
+* The core client event is completion.  Manager state changes only as a
+  side effect from submitting or completing an operation.
+* The service transitions from idle to processing when an operation is
+  submitted.
+* The service transitions from processing to idle when notification of
+  the last operation has completed and there are no queued operations.
+* The manager selects the next operation to process when notification of
+  completion has itself completed.  In particular, changes to the set of
+  pending operations that are made during a completion callback affect
+  the next operation to execute.
+* Each submitted operation includes a priority that orders execution by
+  first-come-first-served within priority.
+* Operations are asynchronous, with completion notification through the
+  :ref:`async_notification`.  The operations and notifications are run
+  in a context that is service-specific.  This may be one or more
+  dedicated threads, or work queues.  Notifications may come from
+  interrupt handlers.  Note that for some services certain operations
+  may complete before the submit request has returned to its caller.
+
+The generic infrastructure holds the active operation and a queue of
+pending operations.  A service extension shall provide functions that:
+
+* check that a request is well-formed, i.e. can be added to the queue;
+* receive notification that a new operation is to be processed, or that
+  no operations are available (allowing the service to enter a
+  power-down mode);
+* translate a generic completion callback into a service-specific
+  callback.
+
+.. doxygengroup:: resource_mgmt_queued_operation_apis
+   :project: Zephyr