| How to create vmcores for tests |
| =============================== |
| |
| 1. Boot a FreeBSD VM with as little memory as possible and create a core dump |
| per `FreeBSD Handbook Kernel Debugging Chapter`_. Note that you may need to |
| reboot with more memory after the kernel panic as otherwise savecore(8) may |
| fail. |
| |
| For instance, I was able to boot FreeBSD and qemu-system-x86_64 with 128 MiB |
| RAM but had to increase it to 256 MiB for the boot after kernel panic. |
| |
| 2. Transfer the kernel image (``/boot/kernel/kernel``) and vmcore |
| (``/var/crash/vmcore.latest``) from the VM. |
| |
| 3. Patch libfbsdvmcore using ``libfbsdvmcore-hacks.patch`` and build LLDB |
| against the patched library. |
| |
| 4. Patch LLDB using ``lldb-minimize-processes.patch`` and build it. |
| |
| WARNING: LLDB will now modify core files in order to make the resulting |
| test vmcores smaller. Make a backup if necessary. |
| |
| 5. Do a test run of ``test.script`` in LLDB against the kernel + vmcore:: |
| |
| lldb -b -s test.script --core /path/to/core /path/to/kernel |
| |
| If everything works fine, the LLDB output should be interspersed with |
| ``%RD`` lines. The vmcore will also be modified to shorten the process |
| list in ``allproc``. |
| |
| 6. Open the vmcore in the patched LLDB again and choose interesting threads |
| for testing. Update thread numbers in ``test.script`` if necessary. |
| |
| 7. Use the ``copy-sparse.py`` tool to create a sparse version of the vmcore:: |
| |
| lldb -b -s test.script --core /path/to/core /path/to/kernel | |
| grep '^% RD' | python copy-sparse.py /path/to/core vmcore.sparse |
| |
| 8. Compress the sparse vmcore file using ``bzip2``:: |
| |
| bzip2 -9 vmcore.sparse |
| |
| |
| .. _FreeBSD Handbook Kernel Debugging Chapter: |
| https://docs.freebsd.org/en/books/developers-handbook/kerneldebug/ |
