Write a minimal DFv2 driver

This guide walks through the steps involved in creating a minimal DFv2 driver.

The instructions in this guide are based on the minimal skeleton driver, which provides the minimum implementation necessary to build, load, and register a new DFv2 driver in a Fuchsia system.

The steps are:

  1. Create a driver header file.
  2. Create a driver source file.
  3. Add the driver export macro.
  4. Create a build file.
  5. Write bind rules.
  6. Create a driver component.

For more DFv2-related features, see Additional tasks.

Create a driver header file

To create a header file for your DFv2 driver, do the following:

  1. Create a new header file (.h) for the driver (for example, skeleton_driver.h).

  2. Include the following interface to the header file:

    #include <lib/driver/component/cpp/driver_base.h>
    
  3. Add an interface for the DriverBase class, for example:

    #include <lib/driver/component/cpp/driver_base.h>
    
    namespace skeleton {
    
    class SkeletonDriver : public fdf::DriverBase {
     public:
      SkeletonDriver(fdf::DriverStartArgs start_args, fdf::UnownedSynchronizedDispatcher driver_dispatcher);
    
      // Called by the driver framework to initialize the driver instance.
      zx::result<> SkeletonDriver::Start() override;
    };
    
    }  // namespace skeleton
    

    (Source: skeleton_driver.h)

Create a driver source file

To implement the basic methods for the DriverBase class, do the following:

  1. Create a new source file (.cc) for the driver (for example, skeleton_driver.cc).

  2. Include the header file created for the driver, for example:

    #include "skeleton_driver.h"
    
  3. Implement the basic methods for the class, for example:

    #include "skeleton_driver.h"
    
    namespace skeleton {
    
    SkeletonDriver::SkeletonDriver(fdf::DriverStartArgs start_args,
                               fdf::UnownedSynchronizedDispatcher driver_dispatcher)
        : DriverBase("skeleton_driver", std::move(start_args),
              std::move(driver_dispatcher)) {
    }
    
    zx::result<> SkeletonDriver::Start() {
      return zx::ok();
    }
    
    }  // namespace skeleton
    

    (Source: skeleton_driver.cc)

    This driver constructor needs to pass the driver name (for example, "skeleton_driver"), start_args, and driver_dispatcher to the DriverBase class.

Add the driver export macro

To add the driver export macro, do the following:

  1. In the driver source file, include the following header file:

    #include <lib/driver/component/cpp/driver_export.h>
    
  2. Add the following macro (which exports the driver class) at the bottom of the driver source file:

    FUCHSIA_DRIVER_EXPORT(skeleton::SkeletonDriver);
    

    For example:

    #include <lib/driver/component/cpp/driver_base.h>
    #include <lib/driver/component/cpp/driver_export.h>
    
    #include "skeleton_driver.h"
    
    namespace skeleton {
    
    SkeletonDriver::SkeletonDriver(fdf::DriverStartArgs start_args,
                               fdf::UnownedSynchronizedDispatcher driver_dispatcher)
        : DriverBase("skeleton_driver", std::move(start_args),
              std::move(driver_dispatcher)) {
    }
    
    zx::result<> SkeletonDriver::Start() {
      return zx::ok();
    }
    
    }  // namespace skeleton
    
    FUCHSIA_DRIVER_EXPORT(skeleton::SkeletonDriver);
    

    (Source: skeleton_driver.cc)

Create a build file

To create a build file for the driver, do the following:

  1. Create a new BUILD.gn file.
  2. Include the following line to import the driver build rules:

    import("//build/drivers.gni")
    
  3. Add a target for the driver, for example:

    fuchsia_cc_driver("driver") {
      output_name = "skeleton_driver"
      sources = [ "skeleton_driver.cc" ]
      deps = [
        "//sdk/lib/driver/component/cpp",
        "//src/devices/lib/driver:driver_runtime",
      ]
    }
    

    (Source: BUILD.gn)

    The output_name field must be unique among all drivers.

Write bind rules

To write bind rules for your driver, do the following:

  1. Create a new bind rule file (.bind) for the driver (for example, skeleton_driver.bind) in the meta directory.

  2. Add basic bind rules, for example:

    using gizmo.example;
    gizmo.example.TEST_NODE_ID == "skeleton_driver";
    

    (Source: skeleton_driver.bind)

  3. In the BUILD.gn file, include the following line to import the bind build rules:

    import("//build/bind/bind.gni")
    
  4. In the BUILD.gn file, add a target for the driver's bind rules, for example:

    driver_bind_rules("bind") {
      rules = "meta/skeleton.bind"
      bind_output = "skeleton_driver.bindbc"
      deps = [ "//examples/drivers/bind_library:gizmo.example" ]
    }
    

    (Source: BUILD.gn)

    The bind_output field must be unique among all drivers.

Create a driver component

To create a Fuchsia component for the driver, do the following:

  1. Create a new component manifest file (.cml) in the meta directory (for example, skeleton_driver.cml).

  2. Include the following component shards:

    {
        include: [
            "inspect/client.shard.cml",
            "syslog/client.shard.cml",
        ],
    }
    
  3. Add the driver's program information using the following format:

    {
        program: {
            runner: "driver",
            binary: "driver/<OUTPUT_NAME>.so",
            bind: "meta/bind/<BIND_OUTPUT>",
        },
    }
    

    The binary field must match the output_name field in the fuchsia_driver target of the BUILD.gn file, and the bind field must match bind_output in the driver_bind_rules target, for example:

    {
        include: [
            "inspect/client.shard.cml",
            "syslog/client.shard.cml",
        ],
        program: {
            runner: "driver",
            binary: "driver/skeleton_driver.so",
            bind: "meta/bind/skeleton.bindbc",
        },
    }
    

    (Source: skeleton_driver.cml)

  4. Create a new JSON file to provide the component's information (for example, component-info.json) in the meta directory.

  5. Add the driver component's information in JSON format, for example:

    {
        "short_description": "Driver Framework example for a skeleton DFv2 driver",
        "manufacturer": "",
        "families": [],
        "models": [],
        "areas": [
            "DriverFramework"
        ]
    }
    

    (Source: component-info.json)

  6. In the BUILD.gn file, include the following line to import the component build rules:

    import("//build/components.gni")
    
  7. In the BUILD.gn file, add a target for the driver component, for example:

    fuchsia_driver_component("component") {
      component_name = "skeleton"
      manifest = "meta/skeleton.cml"
      deps = [
         ":bind",
         ":driver"
      ]
      info = "component-info.json"
    }
    

    (Source: BUILD.gn)

    See the rules for these fields below:

    • Set the manifest field to the location of the driver's .cml file.
    • Set the info field to the location of the driver component information JSON file.
    • Set the deps array to include the fuchsia_driver and driver_bind_rules targets from the BUILD.gn file.

You can now build, load, and register this DFv2 driver in a Fuchsia system

Additional tasks

This section provides additional features you can add to your minimal DFv2 driver:

Add logs

By default, to print logs from a DFv2 driver, use the FDF_LOG macro, for example:

FDF_LOG(INFO, "Starting SimpleDriver")

In addition to using the FDF_LOG macro, you can also print logs using Fuchsia's structured logger library (structured_logger.h), which uses the FDF_SLOG macro.

To use structured logs from your DFv2 driver, do the following:

  1. Include the following header:

    #include <lib/driver/logging/cpp/structured_logger.h>
    
  2. Use the FDF_SLOG macro to print logs, for example:

    FDF_SLOG(ERROR, "Failed to add child", KV("status", result.status_string()));
    

Add a child node

A DFv2 driver can add child nodes using the following Node protocol in the fuchsia.driver.framework FIDL library:

open protocol Node {
    flexible AddChild(resource struct {
        args NodeAddArgs;
        controller server_end:NodeController;
        node server_end:<Node, optional>;
    }) -> () error NodeError;
};

To facilitate this, during startup the driver framework provides a client of the bound node's Node protocol to the DFv2 driver, through the DriverBase. The driver can access its node client at any time to create child nodes on it. However directly using this FIDL library requires a setup that includes creating FIDL channel pairs and constructing the NodeAddArgs table. Therefore the DriverBase class provides a set of helper functions to make adding child nodes easier. (To see these helpers, check out the driver_base.h file.)

There are two types of nodes a DFv2 driver can add: unowned and owned. The main difference between an unowned node and an owned node is whether they participate in the driver match process or not.

The driver framework tries to find a driver that matches the properties of unowned nodes so it can bind a driver to the node. Once a driver is matched and bound to a node, the bound driver becomes the owner of the node. On the other hand, owned nodes do not participate in matching since the driver that created the node is already the owner.

DriverBase helper functions

The client to the node that your driver is currently bound to is stored in the DriverBase object. This allows the driver to use the DriverBase class's AddChild() and AddOwnedChild() functions to add a child node to this node.

However, to use these DriverBase helper functions, the node must not have been moved out of the driver. If the node is moved out or your target node is not the node that the driver is currently bound to (ie. for a grand-child node), you need to use the namespace methods available in the add_child.h file instead. These methods are the same as the DriverBase helper functions except they can be used to add a child to a node beyond the reach of the DriverBase object, by providing the correct parent node client as a target.

Lastly, these helper functions take care of logging errors if they happen, so no logging is needed by the driver.

Create an unowned node

To create an unowned node, a driver can use the DriverBase::AddChild() helper functions. There are two types of these functions: one that allows providing DevfsAddArgs and the other that does not. These functions allow setting the properties on an unowned node, which the driver framework uses to find a matching driver. The return result of both is a client end to the NodeController protocol, which can either be kept by the driver or discarded safely.

The example code below creates an unowned node under the driver's bound node:

// Add a child node.
auto properties = std::vector{fdf::MakeProperty2(bind_fuchsia_test::TEST_CHILD, "simple")};
zx::result child_result = AddChild(child_name, properties, compat_server_.CreateOffers2());
if (child_result.is_error()) {
  return child_result.take_error();
}

child_controller_.Bind(std::move(child_result.value()));

(Source: simple_driver.cc)

Create an owned node

To create an owned node, a driver can use the DriverBase::AddOwnedChild() helper functions. There are again two types: one that allows DevfsAddArgs and the other that does not. These functions do not provide a properties argument since an owned node does not participate in driver matching. The return result of both is an OwnedChildNode object that contains a client end to the NodeController (which is safe to discard) and a client end to the Node protocol, which is not safe to discard. The driver must hold on to the Node client for as long as it wants the owned node to stay around. Dropping this client will cause the driver framework to remove the node.

The example code below creates an owned node with devfs arguments:

fuchsia_driver_framework::DevfsAddArgs devfs_args{{.connector = std::move(connector.value())}};
zx::result owned_child = AddOwnedChild("retriever", devfs_args);
if (owned_child.is_error()) {
  return owned_child.error_value();
}

child_node_.emplace(std::move(owned_child.value()));

(Source: retriever-driver.cc)

Clean up the driver

If a DFv2 driver needs to perform teardowns before it is stopped (for example, stopping threads), then you need to override and implement additional DriverBase methods: PrepareStop() and Stop()

The PrepareStop() function is called before the driver's fdf dispatchers are shut down and the driver is deallocated. Therefore, the driver needs to implement PrepareStop()if it needs to perform certain operations before the driver's dispatchers shut down, for example:

void SimpleDriver::PrepareStop(fdf::PrepareStopCompleter completer) {
 // Teardown threads
  FDF_LOG(INFO, "Preparing to stop SimpleDriver");
  completer(zx::ok());
}

The Stop() function is called after all dispatchers belonging to this driver are shut down, for example:

void SimpleDriver::Stop() {
  FDF_LOG(INFO, "Stopping SimpleDriver");
}

Add a compat device server

If your DFv2 driver has descendant DFv1 drivers that haven't yet migrated to DFv2, you need to use the compatibility shim to enable your DFv2 driver to talk to other DFv1 drivers in the system. For more details, see the Set up the compat device server in a DFv2 driver guide.