Generate core dump from crashed pods

[anwu]

Background

We should be able to save a core dump of a halted POD so that we can continue debugging when the device is no longer attached to the debugger.

Goal

Be able to run a dump tool like so... iotile hw --port jlink connect_direct device=nrf52 debug dump_core <core.file> Where you can then do arm-none-eabi-gdb --core <core.file> And then debug from there.

Implementation

A core dump file is pretty much an .elf file of the program with certain sections modified with memory contents of the device at the time of crash.

A more in depth analysis can be read here.

The following implementation steps will be referencing to this article.

• Use an existing compiled .elf file of a controller and modify its properties
• The e_type of the ELF header should be set to 0x04 to indicate that the .elf file is a core file
• ELF files are not expected to have section headers
• SHT_NOBITS are not present in core files
• SHT_PROGBITS section are present in core file
• SHT_NOTE section header mapd to the PT_NOTE program header... this is important, this will point us to the saved register information that the core file will use
• Use the jlink adapter's flash_forensic tool to dump_memory of the device's register information
• Fill the PT_NOTE section and format it accordingly with dumped register information

[anwu]

Capability will probably need to have some inherent knowledge of where certain FreeRTOS stacks live.

[anwu]

Adding New Program Header

• Increment e_phnum in the ELF Header
• Insert bytes wherever Program Header starts by size of e_phentsize
• Update the e_shoff by incrementing it by e_phentsize
• For the section .shstrtab, increment offset by e_phentsize