Overview of embedded programming and debugging

This section intends to provide an overview of how to go about programming and debugging microcontrollers on embedded hardware. Later sections of this book will dive into numerous topics from simply flashing a binary onto a chip to single stepping through instructions, inspecting memory and registers, instructing the CPU to halt execution under certain conditions and wait for input, look for changes in specific memory locations, etc.

Why programmers and debuggers

What are debuggers and programmers?

Generally, an additional hardware device is required to allow the host PC you are developing on to connect to the target microcontroller. These devices have several names such as debugger, debug probe or programmer. Sometimes they are referred to by a specific brand or protocol they speak, e.g. ST-Link, JLink. These devices are typically distinct pieces of hardware, but in case you are using development boards, they are often integrated into the board.

To add to the confusion, the software used to control the hardware device, in our case gdb is also typically referred to as 'debugger'.

In contrast to creating software for a PC, the compiled program will be running on a different device and needs to be transferred to the target which needs to be instructed to execute the program. This is accomplished by the debugger device. The debugger is therefor our main tool to understand and debug the inner workings of the system being developed. These tools are very powerful and give the user complete access and jurisdiction over a microcontroller, even to the point that the user can rewrite arbitrary memory manually if desired.

SHOULD WE ADD IMAGES OF COMMON HARDWARE TOGETHER WITH SOME DESCRIPTION?

GDB

Our main option for debugging embedded targets is a version of gdb specifically targetted to the processor architecture being used, for example, arm-none-eabi-gdb for bare metal arm targets. Unlike debugging on a local machine, the gdb instance and application code typically run on different devices. gdb supports this through the use of servers and remote targets, whereby the debugger device implements the gdb server, and the gdb instance can connect to this to interact with the running application.

While the invocation of GDB is typically independent of the debugger and thus gdb server you are using, you may need to change the remote server address when connecting and provide specific options (such as semihosting) which are not universally available on all debugger devices.

Connecting

  • Run gdb ELF where ELF is your output file to start gdb (ie. gdb target/thumbv7em-none-eabihf/release/hello_world)
  • Enter target extended-remote :PORT to connect to the gdb server (ie. target extended-remote :4242), note that this may be a TCP or serial port depending on the programmer
  • Get debugging!

gdb will automatically execute commands from a .gdbinit file in the directory from which it is executed, allowing you to automate the connection to the gdb server and setting any breakpoints you might need. A useful minimal example is:

	# Connect to the remote target
	target extended-remote :4242
	# Load the application
	load
	# Set a breakpoint on our entry to main
	break main
	# Start execution
	continue

Useful Commands

In general gdb will interpret the shortest series of characters required to uniquely identify a command as that command, for example tar instead of target.

  • target extended-remote ADDR connects to a remote target.
  • break LOCATION or b LOCATION sets a breakpoint, for example b main sets a breakpoint on the entry to main.
  • print VARIABLE or p VARIABLE prints the value of a variable at the point you are inspecting, p\x VARIABLE prints it in hex.
  • delete N or d N deletes a breakpoint by index.
  • continue or c continues to the next breakpoint.
  • step or s steps through execution line by line.
  • backtrace or bt prints a backtrace from the point you are currently inspecting.
  • load flashes the current binary to the device.
  • run re-starts an application.
  • layout LAYOUT switches to different views, useful options are SRC for source, REGS for registers, ASM for assembly.
  • quit or q gets you out, though you may have to interrupt with ctrl+c if the application is currently running.
  • source FILENAME runs the commands specified in FILENAME.
  • set substitute-path SRC DST substitutes paths that start with SRC with DST. This is useful for stepping into the rust sources as explained later.

You can invoke gdb with the source layout loaded by passing the --tui argument in the command line. Note that tui mode starts with the source view selected so normal control keys will scroll the source view instead of the terminal, you can move through previous and next commands with ctrl+P and ctrl+N respectively, or use ctrl+x o to move focus between the source and terminal views and use your arrow keys and page-up/page-down as normal.

You may encounter errors stepping through code concerning sources that cannot be found. In case the source paths contains hashes like: /rustc/e305df1846a6d985315917ae0c81b74af8b4e641/... the cause may be rust sources which were compiled on a build server in the directory /rustc/{commit_hash}. To find this commit hash you can use the command rustc -Vv. The set substitute-path command can then be used to substitute this with your {RUST_SRC_PATH}/lib/rustlib/src/rust (RUST_SRC_PATH can be found from the output of rustc --print=sysroot). For example:

set substitute-path /rustc/e305df1846a6d985315917ae0c81b74af8b4e641 "C:/Users/username/.rustup/toolchains/nightly-x86_64-pc-windows-msvc/lib/rustlib/src/rust"

VSCode Integration

The Native Debug extension can be used to debug Rust code directly in the editor. To use it you will need to add a launch configuration to your .vscode/launch.json file. Below is an example that starts gdb and executes the commands specified in debug.gdb:

"configurations": [
    {
        "name": "Remote debug",
        "type": "gdb",
        "request": "launch",
        "cwd": "${workspaceRoot}",
        "target": "${workspaceRoot}/target/thumbv7em-none-eabihf/debug/hello", 
        "gdbpath" : "arm-none-eabi-gdb",
        "autorun": [
            "source -v debug.gdb",
        ]
    }
]

Make sure to change the "target" to match the destination of your output file.

Before launching the debugger in VSCode you must ensure that JLinkGDBServer, OpenOCD or other debugger of choice is running on the correct port. Once running you can then set breakpoints in the code margin and hover over local or global variables to see their current value

VSCode screenshot showing breakpoints and hovering over variables to see their value

Cargo run integration

The cargo run command can be configured to start the debugger. To use this, simply add the following to the appropriate target in your .cargo/config:

runner = "arm-none-eabi-gdb -q -x debug.gdb"

Interfaces / Protocols

Two common protocols exist to communicate and introspect microcontrollers: JTAG (Joint Test Action Group) and SWD (Serial Wire Debug).

JTAG

JTAG has a long history, and was originally intended to daisy-chain multiple micro processors, FPGAs, memory chips, simply anything which supports JTAG, and access them over the same JTAG bus. To enable this, it requires more signals than SWD:

  • TDI (Test Data In)
  • TDO (Test Data Out)
  • TCK (Test Clock)
  • TMS (Test Mode Select)
  • TRST (Test Reset) which is optional

(a variant of JTAG called cJTAG, c for compact exists that uses 2 signals)

The following image illustrates a common communication example with a JTAG bus with multiple devices:

Alt text

SWD

SWD is quite similar to JTAG, as it uses the same protocol but it is limited to two electrical signals and cannot address multiple targets on a single connection. SWD is also a proprietary ARM technology. While JTAG is vastly more versatile, SWD's strength is in its simplicity as many systems only have one microcontroller to program. In these cases the reduction of signals while still having full debugability is a major win. In SWD the following signals are available:

  • SWDIO (Serial Wire Debug Input / Output)
  • SWCLK (Serial Wire Clock)

The following image illustrates a common communication example with SWD communicating with a single microcontroller:

Alt text

Which one to use?

When designing a system, we should select the interface which is most convenient for the system, and there are simple criteria to consider:

  • Does the target support SWD?
  • Does the system have more than a single component which needs to be programmed or debugged?
  • How valuable is board space and is the larger area required for JTAG acceptable?
  • What does the component manufacturer recommend?

Debugging Applications

Program Files

Depending on the debugger you use, you may need to convert your compiled output to a different format. By default the compiler outputs a rich ELF file containing section information (and debug symbols if you have them turned on). This information is useful to debug software, i.e. gbd, but most microcontrollers require a very minimal binary (BIN) or hexadecimal (HEX) file, consisting of only the program instructions.

cargo-binutils provides an objcopy command to convert between formats. Since objcopy can only operate on a single file, as with all cargo commands, the concrete target needs to be specified. For embedded projects the target can either be:

  • an application binary: specify this using --bin NAME where NAME is either the name of the crate (in this case the implicit target with code in src/main.rs) or the name of the binary as defined in Cargo.toml

  • an example: use --example NAME where NAME is the name of the example source code in the examples folder

To create a bin file execute:

	$ cargo objcopy TARGET --release -- -O binary OUTPUT 

or for a hex file:

	cargo objcopy TARGET --release -- -O ihex OUTPUT

where TARGET is the target specification as explained above and OUTPUT the desired output filename.

This is equivalent to using arm-none-eabi-objcopy for the same purpose, but cargo objcopy is aware of the context can determine the target output directory.

TODO: do we always have to build with release to get reasonable binary sizes? TODO: could cargo objcopy just do these things by default?

Common debugger platforms

OpenOCD

OpenOCD (Open On-Chip Debugger) is a very popular piece of software which provides an interface for a debugger software, such as GDB, to connect to in order to provide in-system programming and boundary-scan testing for embedded devices.

However, OpenOCD only provides the gbdserver for the debug software to connect to and requires a debug adapter to connect to the target microcontroller.

OpenOCD supports both JTAG and SWD signaling schemes throughsupported debug adapters. These are typically connected to the PC via USB. These devices range from very cheap eBay reseller devices, are built into evaluation boards from microcontroller manufacturers, to very expensive, special purpose, adapters.

TODO: list workable openocd hardware

Segger JLink

Segger provide a JLink family of programmers and debuggers commonly used in industry. JLink devices require closed source binaries to use, however these are available for common platforms here. JLink probes are typically compatible with both SWD and JTAG based debugging.

Flashing is managed using JLinkExe and a custom gdb server is provided in JLinkGdbServer, the JLinkExe commands are documented here, note that on windows this is called JLink.exe.

Programming

  1. Run the JLinkExe --speed 4000 --if SWD command to connect to the debugger in SWD mode
  2. Select your device with device DEVICE eg. device EFM32G210
  3. Halt the processor with halt or h
  4. Load your binary with loadbin BINARY, FLASH_ADDRESS eg. loadbin test.bin 0x0000
  5. Verify your binary flashed correctly with verifybin BINARY, FLASH_ADDRESS eg. loadbin test.bin 0x0000
  6. Reset the processor with reset or r
  7. Quit the JLinkExe gui with q

It is also possible to pass scripts to JLinkExe with the --CommanderScript option, allowing automation of commands.

Debugging

To debug with the JLink device you run the JLinkGDBServer command with the specified device, speed, and interface. For example, JLinkGDBServer -device DEVICE -speed 4000 -if SWD. You can then launch a GDB instance with the appropriate command for your target (eg. arm-none-eabi-gdb BINARY.elf) and connect to the GDB server using target remote localhost:2331 (specifying the default JLinkGDBServer port).

A common gotcha with JLinkGDBServer is interrupts not firing. If you experience this issue then add monitor reset to the end of your GDB initialization commands. You can also enable semihosting and print to stdout with monitor semihosting enable and monitor semihosting IOClient 2 respectively.

ARM DAPLink

DAPLink is a project by ARM to develop an open source cortex debug probe, this provides a set of interfaces to simplify programming and debugging and can be implemented on nearly any USB-capable hardware. DAPLink provides a set of endpoints including a CMSIS-DAP interface for debugging, a USB disk for drag-and-drop programming, and an optional serial port for communication with the target system. The USB mass storage approach, in which the target device appears as a thumbdrive in the host operating system is useful for programming of devices in-field as it requires no additonal software, however is not always reliable for development use.

Programming

After connecting a DAPLink device to your system, a USB drive should appear. To flash, copy your binary file (*.bin) to the drive and wait for completion.

Debugging

TODO: try it out

STLink

STLink debuggers are integrated on most ST development boards, and are thus one of the most common programmers you will come across.

ST provides vendor utilities for programming on Windows, and these devices are typically compatible with OpenOCD. You can also use the texane/stlink package which provides a set of utilities to interact with STLink debuggers.

Note that while a STLink programmer might technically work for programming / debugging any SWD compatible device, you are legally prohibited from using it with devices not manufactured by ST.

Programming

Flash with st-flash --reset flash BINARY FLASH_ADDRESS where BINARY is your bin file, and FLASH_ADDRESS is the flash address to write the file. You can also erase the memory wth st-flash erase which due to a bug may sometimes be required prior to flashing.

Debugging

Launch a gdb server on the default port (:4242) with st-util , check the help with st-util --help for other options.

Black Magic Probe

The BMP (Black Magic Probe) is different from the other devices listed here in that it does not require OpenOCD or other intermediary host software to provide a gbd-server. Instead the device itself implements a GDB server over serial port connection. In a sense, it is OpenOCD combined with a debug adapter in the same dongle. The BMP is open hardware and runs open source software inside it as well, plus using GDB with the BMP very similar to OpenOCD. BMP is based on a common STM chip and the firmware can easily be ported to other boards.

Debugger used in this book

This book uses the STM32F3Discovery platform for all examples, just like the Discovery Book and The Embedded Rust Book. This eval board integrates a STLink v2 debug adapter with which we will communicate using OpenOCD.

Debugging Tools

There are currently two main debugging softwares on the PC side, the Gnu Debugger (GDB) and LLVM's Debugger (LLDB). At the time of writing this book, LLDB does not yet have the required support to debug the targets we are using, and we will be using GDB in this book.