Debugging information lost for some aborts

[Jakub Jermář]

Starting with changeset:mainline,845, an aborting task will not print a stack trace itself, but will delegate this responsibility to taskdump transitively via the SYS_TASK_EXIT syscall.

This is an elegant solution, but does not work in two cases:

• when taskdump is not (yet) running
• when the aborting task is required for successful invocation of taskdump, such as the root file system server, VFS and perhaps the root block device driver.

The result is that for some kinds of aborts, the user will not get any useful debugging information. There appear to be several possible solutions:

• run taskdump as an init task so that it does not need the file system to load
• return to printing the stacktrace before calling the syscall
• have the kernel print the stack trace when processing the syscall, especially when it can tell that taskdump will not manage to do it itself

[Jakub Jermář]

Kernel prints the stack trace for any crashing task.