Overview
How to Clone
How to Build
How to Generate Crash Dumps
How to Run
CrashDebug is a post-mortem debugging tool for Cortex-M microcontrollers. It allows developers to investigate a crash with GDB at a later point in time and without requiring direct access to the failing device. GDB supports core dump generation and debugging on other targets (Linux for example) but not for Cortex-M targets. CrashDebug fills that gap.
A developer just needs these things to conduct post-mortem debugging:
- This CrashDebug remote debugging stub for GDB.
- arm-none-eabi-gdb which can be found and installed from this website.
- The .elf which corresponds to the firmware running on the Cortex-M processor at the time of the crash.
- Contains the symbols required by GDB.
- Provides CrashDebug with the read-only contents of FLASH. The crash dump therefore only needs to provide the contents of RAM and the processor registers at the time of the crash.
- A crash dump to debug.
- The crash dump provides the contents of RAM and the processor registers (r0-r12, sp, lr, pc, xpsr) at the time of the crash.
- See this section to learn how to obtain these crash dumps.
This project uses submodules (ie. MRI and CrashCatcher). Cloning therefore requires a few more steps to get all of the necessary code.
git clone --recursive [email protected]:adamgreen/CrashDebug.git
- or -
git clone [email protected]:adamgreen/CrashDebug.git cd CrashDebug git submodule init git submodule update cd mri git submodule init git submodule update cd ../CrashCatcher git submodule init git submodule update
If you don't want to build the CrashDebug utility yourself, you can use the pre-built binaries (macOS, Windows, and Ubuntu) for CrashDebug that can be found at https://github.com/adamgreen/CrashDebug/tree/master/bins.
You will need a gcc or clang toolchain appropriate for your platform:
- macOS: Xcode and it's command line tools.
- Windows: MinGW
- Linux: You will need gcc and g++.
CrashDebug uses a single non-recursive makefile at the root of the project to build everything. It supports these top level targets:
- all: This builds the CrashDebug code, builds the unit tests, executes the unit tests, and reports the test results. This is the default target if no other is provided to make.
- clean: Cleans up all ouptut files from any previous builds. This forces everything to be rebuilt.
- gcov: Like the all target, this builds all of the CrashDebug code and runs the unit tests but it also instruments the binaries with code coverage tracking and then reports the code coverage obtained from executing these unit tests.
Example:
make all - Build CrashDebug by just rebuilding what has changed since last build.
make clean all - Build CrashDebug by rebuilding everything.
The makefile is constructed in such a way that all dependencies for any single target within the build system are known by GNU Make, including header file dependencies. This means that the user can specify a specific binary target on the make command line and it will build it and all of its dependencies.
Examples:
make lib/libCrashDebug.a - Build the main CrashDebug library.
make LIBCRASHDEBUG_tests - Build the main CrashDebug library, unit tests, and all required dependencies.
make RUN_LIBCRASHDEBUG_TESTS - This does what the previous example does but also runs the unit tests.
make GCOV_LIBCRASHDEBUG - This is similar to the previous example except that it builds binaries which include code coverage instrumentation and then reports the resulting code coverage results after running the unit tests.
make CrashDebug - Builds the CrashDebug application and all of the libraries on which it depends while skipping the unit tests. Windows users would run make CrashDebug.exe instead.
The build has been tested on the following operating systems:
- macOS Big Sur
- Windows 10
- Ubuntu 20.04
All intermediate binaries (*.o and *.a) for code coverage runs are found in the gcov/obj and gcov/lib directories. If you want to examine the line by line coverage results for a library like libCrashDebug, you would find it in the gcov/LIBCRASHDEBUG_tests/ directory.
If a user has GDB attached to a halted Cortex-M processor then they can manually generate a dump of the processor state at the time of the crash. This crash dump can be used for further debugging at a later point in time after the device has been freed from the debugger or it could be sent along with the corresponding .elf file to another developer for further investigation.
We want to have GDB dump all of the RAM on the device. The info mem command will ask GDB to display its knowledge of the memory map to the user. The following shows an example of what you might get on a LPC1768 device:
(gdb) info mem Using memory regions provided by the target. Num Enb Low Addr High Addr Attrs 0 y 0x00000000 0x00010000 flash blocksize 0x1000 nocache 1 y 0x00010000 0x00080000 flash blocksize 0x8000 nocache 2 y 0x10000000 0x10008000 rw nocache 3 y 0x2007c000 0x20084000 rw nocache
In the above example, the first two regions are for read-only FLASH so we can ignore those as their contents can already be found in the .elf. The next two regions, 0x10000000 to 0x10008000 and 0x2007c000 to 0x20084000, are the RAM regions we are interested in dumping.
The following GDB commands will generate a gdb.txt file which contains the crash dump state required for post-mortem debugging:
set pagination off set logging on set var $ptr=0x10000000 while $ptr < 0x10008000 x/4wx $ptr set var $ptr+=16 end set var $ptr= 0x2007c000 while $ptr < 0x20084000 x/4wx $ptr set var $ptr+=16 end info all-registers set logging off set pagination on
What is all of that doing?
- Turn pagination off so that you don't have to keep pressing Enter to continue scrolling the text.
- Ask GDB to save the output from the rest of these commands to a file named gdb.txt in the current directory at the time GDB was launched.
- A debugger variable, $ptr, is used to execute a loop, dumping RAM four words at a time. This example contains two such loops, one for each RAM region shown in the earlier info mem output.
- Dump the CPU registers.
- Stop the logging of information to gdb.txt now that we have collected all of the information we need.
- Turn pagination back on which is more friendly for interactive debugging.
The following is an example of what the user will see when the CrashCatcher HexModule generates a crash dump.
CRASH ENCOUNTERED Enable logging and then press any key to start dump. 63430200 00000000 01000000587D0010647F0010FFFFFFFF 3C010010CDAB56347856341200000000 00C002407F0000005FE19BFBFF3FF7AC DC000010 687F0010 1D0200003C02000000000081 03000080 0000001000800010 00BE0AE00D782D0668400824400000D3 ... 28ED00E03CED00E0 008200000000004008000000FFFFFFFF FFFFFFFF End of dump CRASH ENCOUNTERED Enable logging and then press any key to start dump.
If the hex dump is sent over a serial connection, the user can record the contents of the dump by having their terminal program log the results. Before the text from such a log can be used, the extra text from before and after the actual dump should be deleted. The above example should therefore be edited to look like the following before using it with the CrashDebug utility:
63430200 00000000 01000000587D0010647F0010FFFFFFFF 3C010010CDAB56347856341200000000 00C002407F0000005FE19BFBFF3FF7AC DC000010 687F0010 1D0200003C02000000000081 03000080 0000001000800010 00BE0AE00D782D0668400824400000D3 ... 28ED00E03CED00E0 008200000000004008000000FFFFFFFF FFFFFFFF
The first 4 characters in the hex dump must contain the "6343" signature and the dump should contain no extra text at the end which CrashDebug might mistake as an extra (and corrupt) memory region.
A binary CrashCatcher dump file such as the CRASH.DMP generated by the LocalFileSystem sample should be ready to use by CrashDebug.
CrashDebug is launched from within GDB as part of the target remote command. The following example shows how to start arm-none-eabi-gdb such that it automatically launches CrashDebug as a remote debugging stub.
arm-none-eabi-gdb main.elf -ex "set target-charset ASCII" -ex "target remote | CrashDebug --elf main.elf --dump MainCrashDump.txt"
CrashDebug (--elf elfFilename | --bin imageFilename baseAddress)
--dump dumpFilenameNOTE: The --elf and --bin options are mutually exclusive. Use one or the other but not both.
--elf is used to provide the filename of the .elf image containing the device's FLASH contents at the time of the crash.
--bin is used to provide the filename of the binary image loaded into the device's FLASH when the crash occurred. These binary images are typically generated by running arm-none-eabi-objcopy -O binary input.elf output.bin The baseAddress parameter indicates where the contents of the .bin file was loaded into FLASH. This address will typically be 0x00000000 unless a boot loader was in use.
--dump is used to provide the filename of the crash dump which contains the contents of RAM and the CPU registers at the time of the crash. This dump can be a gdb.txt manually created by a user from within GDB, a hex dump generated by the CrashCatcher module or a binary dump generated by the CrashCatcher module. See this section to learn more about generating crash dumps.
Windows Users: Don't use backslashes (\) when specifying the path for CrashDebug, the elf file, or the dump file. Instead use forward slashes (/). GDB deletes backslashes that it encounters in -ex command line parameters.