Here's how it works. There are two machines:
The test machine contains a development build of the kernel
and system applications.
The debugging machine contains the symbols for the kernel
and for the system applications. This allows the debugging
machine to perform both user-mode and kernel-mode debugging.
When a crash occurs the NT equivalent of the OOPS massage is sent
via a serial cable to the debugging machine. The debugger
application catches the kernel dump message stream, loads the
operating system symbolic information and enters command mode.
Then someone sitting at the debugging machine can step through the
system memory, alter memory states and step through code. It's
incredibly powerful.
But, here's the really cool bit: if the debugging machine is on a
LAN or has a dialup server, a remote developer (in Israel, say) can
connect up and debug the system.
This seems to me to be functionality that would be super-useful for
the Linux development community. The are so many sharp minds out
there that have domain knowledge that has been hard-won and is
difficult to find/develop. I think a system should be created where
development systems' crashes can be registered on a website and
kernel developers can then dial in to the crashed machine's
debugging machine to trace and identify the fault. This would
increase our productivity by a few orders of magnitude.
We could also set up a method whereby the debugging machine
automatically posts the full OOPS message to the web for analysis.
What do you all think?
Miles Lane
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/