embedded-trainings-2020/advanced
Jorge Aparicio fcf0e310ab - rework RTIC resources section in the adv. material
- change the USB PID so it matches the date of the workshop
2020-06-16 10:21:25 +02:00
..
common - rework RTIC resources section in the adv. material 2020-06-16 10:21:25 +02:00
firmware - rework RTIC resources section in the adv. material 2020-06-16 10:21:25 +02:00
host/print-descs - rework RTIC resources section in the adv. material 2020-06-16 10:21:25 +02:00
README.md - rework RTIC resources section in the adv. material 2020-06-16 10:21:25 +02:00

advanced

Advanced workshop

In this workshop we'll build a toy USB device application that gets enumerated by the host.

The embedded application will run in a fully event driven fashion: only doing work when the host asks for it.

The nRF52840

Some details about the nRF52840 microcontroller that are relevant to this workshop.

  • single core ARM Cortex-M4 processor clocked at 64 MHz
  • 1 MB of Flash (at address 0x0000_0000)
  • 256 KB of RAM (at address 0x2000_0000)
  • IEEE 802.15.4 and BLE (Bluetooth Low Energy) compatible radio
  • USB controller (device function)

The nRF52840 Development Kit

The development board we'll use has two USB ports: J2 and J3 -- you can find a description of the board in the top-level README of this repository -- and an on-board J-Link programmer / debugger. USB port J2 is the J-Link's USB port. USB port J3 is the nRF52840's USB port.

Code organization

The advanced folder contains both "host" code, code that will run on the host, and "firmware" code, code that will run on the nRF52840 SoC. "host" and "firmware" source code has been placed in different Cargo workspaces so that each can be compiled with different compilation profiles. The host workspace will be natively compiled, whereas the firmware workspace will be cross-compiled for the ARM Cortex-M architecture.

$ cd advanced

$ tree -L 1 .
.
├── common
├── firmware
├── host
└── README.md

In addition to these two workspaces there's a third folder called "common". This folder contains no_std code that can be depended on by either "host" code or "firmware" code.

Listing USB devices

In the tools folder you'll find usb-list: a minimal cross-platform version of the lsusb tool. Run it (cargo run from tools/usb-list) to list all USB devices.

$ cargo run
Bus 002 Device 001: ID 1d6b:0003
Bus 001 Device 002: ID 0cf3:e300
Bus 001 Device 003: ID 0c45:6713
Bus 001 Device 001: ID 1d6b:0002

The goal is to get the nRF52840 SoC to show in this list. The embedded application will use the vendor ID (VID) and product ID (PID) defined in advanced/common/consts; the usb-list tool will highlight the USB device that matches that VID PID pair.

$ # expected output
$ cargo run
Bus 002 Device 001: ID 1d6b:0003
Bus 001 Device 002: ID 0cf3:e300
Bus 001 Device 003: ID 0c45:6713
Bus 001 Device 001: ID 1d6b:0002
Bus 001 Device 059: ID 2020:0717 <- nRF52840 on the nRF52840 Development Kit

Hello, world!

First, open the tools/dk-run folder and run cargo install --path . -f to install the dk-run tool.

Next open the advanced/firmware folder in VS Code. If have already opened the root of the repository (embedded-trainings-2020) then please also open the advanced/firmware folder: right click the left side panel, select "Add folder to workspace" and add the advanced/apps folder.

Now, on the left side panel, open the src/bin/hello.rs file from under the advanced/apps folder.

Give Rust Analyzer some time to analyze the file and its dependency graph. When it's done, a "Run" button will appear over the main function -- you may need to edit the file contents to make the "Run" button appear.

Click the "Run" button to run the application on the microcontroller.

If you are not using VS code run the cargo run --bin hello command from the advanced/firmware folder.

NOTE if you run into an error along the lines of "Debug power request failed" retry the operation and the error should disappear

The firmware workspace has been configured to cross-compile applications to the ARM Cortex-M architecture and then run them through the dk-run custom Cargo runner. The dk-run tool will load and run the embedded application on the microcontroller and collect logs from the microcontroller.

The dk-run process will terminate when the microcontroller enters the "halted" state. From the embedded application, one can enter the "halted" state using the asm::bkpt function. For convenience, an exit function is provided in the dk Hardware Abstraction Layer (HAL). This function is divergent like std::process::exit (fn() -> !) and can be used to halt the device and terminate the dk-run process.

Note that when the dk-run tool sees the device enter the halted state it will proceed to reset-halt the device. This is particularly important when writing USB applications because simply leaving the device in a halted state will make it appear as an unresponsive USB device to the host; some OSes (e.g. Linux) will try to make an unresponsive device respond by power cycling the entire USB bus -- this will cause all other USB devices on the bus to be re-enumerated. Reset-halting the device will cause it to be electrically disconnected from the host USB bus and avoid the "power cycle the USB bus" scenario.

RTIC hello

RTIC, Real Time on Integrated Circuits, is a framework for building evented, time sensitive applications.

Open the src/bin/rtic-hello.rs file.

RTIC applications are written in RTIC's Domain Specific Language (DSL). The DSL extends Rust syntax with custom attributes like #[init] and #[idle].

RTIC makes a clearer distinction between the application's initialization phase, the #[init] function, and the application's main loop or main logic, the #[idle] function. The initialization phase runs with interrupts disabled and interrupts are re-enabled before the idle function is executed.

rtic::app is a procedural macro that generates extra Rust code, in addition to the user's functions. The fully expanded version of the macro can be found in the file target/rtic-expansion.rs. This file will contain the expansion of the procedural macro for the last compiled RTIC application.

If you look at the rtic-expansion.rs file generated for the build of the rtic-hello example you can confirm that interrupts are disabled during the execution of the init function.

fn main() -> ! {
    rtic::export::interrupt::disable();
    let late = init(init::Context::new(/* .. */));
    rtic::export::interrupt::enable();
    idle(idle::Context::new(/* .. */))
}

Dealing with registers

Open the src/bin/rtic-events.rs file.

In this and the next section we'll look into the RTIC's event handling features. To explore these features we'll use the action of connecting a USB cable to the DK's port J2 as the event we'd like to handle.

The example application enables the signaling of this "USB power" event in the init function. This is done using the low level register API generated by the svd2rust tool. The register API was generated from a SVD (System View Description) file, a file that describes all the peripherals and registers, and their memory layout, on a Cortex-M microcontroller.

In the svd2rust API, peripherals are represented as structs. The fields of each peripheral struct are the registers associated to that peripheral. Each register field exposes methods to read and write to the register in a single memory operation.

The read and write methods take closure arguments. These closures in turn grant access to a "constructor" value, usually named r or w, which provides methods to modify the bitfields of a register. At the same time the API of these "constructors" prevent you from modifying the reserved parts of the register: you cannot write arbitrary values into registers; you can only write valid values into registers.

In Cortex-M devices interrupt handling needs to be enabled on two sides: on the peripheral side and on the core side. The register operations done in init take care of the peripheral side. The core side of the operation involves writing to the registers of the Nested Vector Interrupt Controller (NVIC) peripheral. This second part doesn't need to be in RTIC application because the framework takes care of it.

Event handling

Below the idle function you'll see a #[task] handler (function). This task is bound to the POWER_CLOCK interrupt signal and will be executed, function call style, every time the interrupt signal is raised by the hardware.

"Run" the rtic-events application. Then connect a micro-USB cable to your PC/laptop then connect the other end to the DK (port J2). You'll see the "POWER event occurred" message after the cable is connected.

Note that all tasks will be prioritized over the idle function so the execution of idle will be interrupted (paused) by the on_power_event task. When the on_power_event task finishes (returns) the execution of the idle will be resumed. This will become more obvious in the next section.

Try this: add an infinite loop to init so that it never returns. Now run the program and connect the USB cable. What behavior do you observe?

Task state

Now let's say we want to change the previous program to count how many times the USB cable (port J3) has been connected and disconnected.

Tasks run from start to finish, like functions, in response to events. To preserve some state between the different executions of a task we can add a resource to the task. In RTIC, resources are the mechanism used to share data between different tasks in a memory safe manner but they can also be used to hold task state.

To get the desired behavior we'll want to store some counter in the state of the on_power_event task.

Open the src/bin/rtic-resources.rs file. The starter code shows the syntax to declare a resource, the Resources struct, and the syntax to associate a resource to a task, the resources list in the #[task] attribute.

In the starter code a resource is used to move the POWER peripheral from init to the on_power_event task. The POWER peripheral then becomes part of the state of the on_power_event task. The resources of a task are available via the Context argument of the task.

We have moved the POWER peripheral into the task because we want to clear the USBDETECTED interrupt flag after it has been set by the hardware. If we miss this step the on_power_event task (function) will be called again once it returns and then again and again and again (ad infinitum).

Also note that in the starter code the idle function has been modified. Pay attention to the logs when you run the starter code.

Your task in this section will be to modify the program so that it prints the number of times the USB cable has been connected to the DK each time the cable is connected.

USB basics

Some basics about the USB protocol. The protocol is complex so we'll leave out many details and focus on the concepts required to get enumeration working.

A USB device, the nRF52840 in our case, can be one of these three states: the Default state, the Address state or the Configured state.

After being powered the device will start in the Default state. The enumeration process will take the device from the Default state to the Address state. As a result of the enumeration process the device will be assigned an address, in the range 1..=127, by the host.

Each OS may perform the enumeration process slightly differently but the process will always involve these host actions:

  • USB reset. This will put the device in the Default state, regardless of what state it was in.
  • GET_DESCRIPTOR request to get the device descriptor.
  • SET_ADDRESS request to assign an address to the device.

The device descriptor is a binary encoded data structure sent by the device to the host. It contains information about the device, like its product and vendor identifiers and how many configurations it has.

A configuration is akin to an operation mode. USB devices usually have a single configuration that will be the only mode in which they'll operate, for example a USB mouse will always act as a USB mouse. Some devices, though, may provide a second configuration for the purpose of firmware upgrades. For example a printer may enter DFU (Device Firmware Upgrade) mode, a second configuration, so that a user can update its firmware; while in DFU mode the printer will not provide printing functionality.

Like the device descriptor, the configuration descriptor is also a binary encoded data structure sent by the device to the host. This descriptor contains information about the interfaces and endpoints the configuration exposes.

An interface is closest to a USB device's function. For example, a USB mouse may expose a single HID (Human Interface Device) interface to report user input to the host. USB devices can expose multiple interfaces. For example, the nRF52840 Dongle could expose both a CDC ACM interface (AKA virtual serial port) and a HID interface; the first interface could be used for (log::info!-style) logs; and the second one could provide a RPC (Remote Procedure Call) interface to the host for controlling the nRF52840's radio.

An interface is made up of one or more endpoints. An endpoint is similar to a UDP or TCP port on a PC in that they allow logical multiplexing of data over a single physical USB bus. Endpoints have directions: a endpoint can either be an IN endpoint or an OUT endpoint. The direction is always from the perspective of the host so in an IN endpoint data travels from the device to the host and in an OUT endpoint data travels from the host to the device. To give an example, a HID interface can use two (interrupt) endpoints, one IN and one OUT, for bidirectional communication with the host. A single endpoint cannot be used by more than one interface (with the exception of the special "endpoint 0").

Endpoints are identified by their address, a zero-based index, and direction. There are three types of non-zero endpoints ("endpoint 0" is special): bulk endpoints, interrupt endpoints and isochronous endpoints. Each endpoint type has different reliability and latency data transfer properties but it's not important to discuss them for this workshop.

"Endpoint 0", also known as the control pipe, actually refers to two endpoints: endpoint 0 IN and endpoint 0 OUT so the control pipe supports data transfers in both directions. The control pipe is mandatory: it must always be present and must always be active.

In this workshop we'll implement the minimal amount of functionality to make enumeration work. To that end you need to consider the following requirements:

  • a USB device must support at least one configuration
  • each configuration must expose at least one interface
  • the control pipe (endpoint 0) must be implemented
  • endpoint 0 is implicitly associated to all interfaces
  • the number of endpoints bound to an interface can be zero -- endpoint 0 is never included in the endpoint count of an interface

Although the control pipe should be bidirectional, in practice to complete the enumeration data only needs to be transferred from the device to the host (IN direction).

Dealing with USB events

Open the src/bin/rtic-usb-1.rs file.

The USB peripheral contains a series of registers, called EVENTS registers, that indicate the reason for entering the USBD event handler. These events must be handled by the application to complete the enumeration process.

In this stage you will need to deal with the following USB events until you reach the EP0STATUS event.

  • USBRESET. This event indicates that the host issued a USB reset signal. According to the USB specification this will move the device from any state to the Default state. Where are not dealing with any other state so doing nothing in response to this event is OK for now.

  • EP0SETUP. The USBD peripheral has detected the SETUP stage of a control transfer. If you get to this point move to the next section.

  • EP0SETUP. The USBD peripheral is signaling the end of the DATA stage of a control transfer. You won't encounter this event just yet.

SETUP stage

Control transfers over endpoint 0 consists of three stages: a SETUP stage, an optional DATA stage and a STATUS stage.

The SETUP stage consists of the host sending a 8 byte header to the device. This header describes the host request: is the host expecting a response (data) from the device, is the host about to send some data to the device or is this a request with no data payload? The SETUP stage conveys this information.

The nRF52840 USBD peripheral will parse this header and store its contents in the following registers: BMREQUESTTYPE, BREQUEST, WVALUE{L,H}, WINDEX{L,H} and WLENGTH{L,H}. These registers match the logical division of the setup packet, in fields, as specified in the USB 2.0 specification. The fields that start with the letter 'b' are 1-byte large; the ones that start with the letter 'w' are 2-bytes large.

Open the src/bin/rtic-usb-2.rs file.

The task here is to write a SETUP packet parser in the common/usb crate and reach the GOAL comment when executing the program. The starter code has already read the relevant registers using helper functions.

To implement the parser, refer to table 9-3 under section 9.4 of the USB 2.0 specification (page 250). Note that at this stage you only need to be able to parse the GetDescriptor variant of the Request enum and within that variant you only need to handle the Device variant of the Descriptor enum.

You should develop this part completely on the host. Switch the workspace to the common/usb directory and run the unit tests on the host using Rust Analyzer's "Test" button.

Device descriptor

After receiving a GET_DESCRIPTOR request during the SETUP stage the device needs to respond with a descriptor during the DATA stage.

Open the src/bin/rtic-usb-3.rs file.

This starter code parses the GET_DESCRIPTOR request and sends back a zero-byte response to the host using the Ep0In abstraction, which we'll cover in the next section. The starter code is wrong because a zero-byte response is not a valid descriptor.

The task in this section is to build a device descriptor using the usb2::device::Descriptor API, turn that Descriptor instance into a sequence of bytes and respond with that. Note that the device must send back at most length bytes to the host; if the byte representation of Descriptor is longer than length then the slice must be truncated to length bytes.

As for the contents of the descriptor you should use these values:

  • Vendor ID: consts::VID
  • Product ID: consts::PID
  • Max packet size for endpoint 0: 64 bytes
  • Number of configurations: 1
  • bcdDevice: 0x01_00, which means version "1.00"
  • everything else can be set to 0 or None

We suggest you check the API docs of the usb2 crate by running the command cargo doc -p usb2 --open. This should open the API docs in your browser.

More information about the fields of the device descriptor can be found in section 9.6.1 of the USB 2.0 spec.

After that has been fixed you'll reach the EP0DATADONE event. It indicates that the data transfer has completed. You must call the Ep0In.end method at that point. After that you'll likely get a new request from the host.

DMA

Let's zoom into the Ep0In abstraction next. You can use the "go to definition" to see the implementation of the Ep0In.start method. What this method does is start a DMA transfer to send bytes to the host. The data (bytes) is first copied into an internal buffer and then the DMA is configured to move the data from that internal buffer to the USBD peripheral.

The signature of the start method does not ensure that (a) bytes won't be deallocated before the DMA transfer is over (e.g. bytes could be pointing into the stack) or that (b) bytes won't be modified right after the DMA transfer starts (this would be a data race in the general case). For these two safety reasons the API is implemented using an internal buffer. The internal buffer has a 'static lifetime so it's guaranteed to never be deallocated -- this prevent issue (a). The busy flag prevents any further modification to the buffer -- from the public API -- while the DMA transfer is in progress.

Apart from thinking about lifetimes and explicit data races in the surface API one must internally use memory fences to prevent reordering of memory operations (e.g. by the compiler), which can also cause data races. DMA transfers run in parallel to the instructions performed by the processor and are "invisible" to the compiler.

In the implementation of the start method, data is copied from bytes to the internal buffer, memcpy operation, and then the DMA transfer is started. The compiler sees the start of the DMA transfer as an unrelated memory operation so it may move the memcpy to after the DMA transfer has started. This reordering results in a data race: the processor modifies the internal buffer while the DMA is reading data out from it. To avoid this reordering a memory fence, dma_start, is used. The fence pairs with the store operation that starts the DMA transfer and prevents the previous memcpy, and any other memory operation, from being move to after the store operation.

Another memory fence, dma_end, is need at the end of the DMA transfer. In the general case, this prevents instruction reordering that would result in the processor accessing the internal buffer before the DMA transfer has finished. This is particularly problematic with DMA transfers that modify a region of memory which the processor intends to read after the transfer.

Not relevant to the DMA operation but relevant to the USB specification, the start method sets a shortcut in the USBD peripheral to issue a STATUS stage right after the DATA stage is finished. Thanks to this it is not necessary to manually start a STATUS stage after calling the end method.

Error handling: stalling the endpoint

The USB specification defines a device-side procedure for "stalling a endpoint", which amounts to the device telling the host that it doesn't support some request. This procedure should be used to deal with invalid requests, requests whose SETUP stage doesn't match any USB 2.0 standard request, and requests not supported by the device, for instance the SET_DESCRIPTOR request is not mandatory.

Open the src/bin/rtic-usb-4.rs file.

In this starter code the code that handles the SETUP stage of endpoint 0 has been refactored into a separate function, ep0setup, that uses Rust's built-in error handling features. This function will return the Err variant when it encounters an invalid request or a request that is not supported.

You can use the usbd:ep0stall helper function or write 1 to the TASKS_EP0STALL register to stall endpoint 0.

State

The starter code in src/bin/rtic-usb-4.rs also passes a State variable to the ep0setup function. This variable tracks the state of the USB device, which can be one of Default, Address or Configured.

From this point you'll need to track the state of the device in software to be able to correctly respond to the host requests.

The handling of the USB reset condition (Event::UsbReset) will need to be updated. According to the USB specification this needs to change the device state, from any state, to the Default state.

SET_ADDRESS

This request changes the device's state as follows:

  • If the device is in the Default state, then

    • if the requested address was 0 (None in the usb2 API) then the device should stay in the Default state
    • otherwise the device should move to the Address state
  • If the device is in the Address state, then

    • if the requested address was 0 (None in the usb2 API) then the device should return to the Default state
    • otherwise the device should remain in the Address state but start using the new address
  • If the device is in the Configured state this request results in "unspecified" behavior according to the USB specication. You should stall the endpoint in this case.

According to the USB specification the device needs to respond to this request with a STATUS stage -- the DATA stage is omitted. The nRF52840 USBD peripheral will automatically issue the STATUS stage and switch to listening only to the requested address (see the USBADDR register) so no further interaction with the USBD peripheral is required for this request.

For more details, read the introduction of section 6.35.9 of the nRF52840 Product Specification 1.0 (pages 486 and 487).

Configuration descriptors

When the host issues a GET_DESCRIPTOR request to request a configuration descriptor the device needs to respond with the requested configuration descriptor plus all the interface and endpoint descriptors associated to that configuration descriptor during the DATA stage.

For simplicity we'll report that the device has zero interfaces and zero endpoints, other than the control endpoint 0 which doesn't need to be reported. So in response to a GET_DESCRIPTOR Configuration 0 request the device should respond with the binary representation of a usb2::configuration::Descriptor instance. The instance should contain these values:

  • wTotalLength: usb2::configuration::Descriptor::SIZE, as just the configuration descriptor is being reported back
  • bNumInterfaces: 0
  • bConfigurationValue: 1
  • bmAttributes: { self_powered: true, remote_wakeup: false }
  • bMaxPower: 250, equivalent to 500 mA

Note that the index of the GET_DESCRIPTOR Configuration request must be 0 because the device reported a single configuration in its device descriptor. Any other index is an "out of bounds" attempt and should be handled by stalling the endpoint.

(Linux) SET_CONFIGURATION

On Linux, the host will sent a SET_CONFIGURATION request, after enumeration, to put the device in the Configured state. You'll need to handle the request according to the following logic:

  • If the device is in the Default state, you should stall the endpoint.

  • If the device is in the Address state, then

    • if the requested configuration value is 0 (None in the usb2 API) then stay in the Address state
    • if the requested configuration value is non-zero and valid (was previously reported in a configuration descriptor) then move to the Configured state
    • if the requested configuration value is not valid then stall the endpoint
  • If the device is in the Configured state, then

    • if the requested configuration value is 0 (None in the usb2 API) then return to the Address state
    • if the requested configuration value is non-zero and valid (was previously reported in a configuration descriptor) then move to the Configured state with the new configuration value
    • if the requested configuration value is not valid then stall the endpoint

In all the cases where you did not stall the endpoint (returned Err) you'll need to acknowledge the request by starting a STATUS stage. This can be done by writing 1 to the TASKS_EP0STATUS register.

For more details, read the section 9.4.7 'SET_CONFIGURATION' of the USB 2.0 specification.

(homework) String descriptors

NOTE(japaric) I hardly think the workshop can fit any more material within the allocated time frame but this is one, of many things, people could continue working on

References