cvs commit: src/sys/alpha/alpha support.s src/sys/i386/i386 swtch.s src/sys/kern kern_shutdown.c src/sys/sys systm.h

Nate Lawson nate at root.org
Tue Jan 20 14:09:50 PST 2004


On Tue, 20 Jan 2004, Poul-Henning Kamp wrote:
> In message <200401201445.24897.jhb at FreeBSD.org>, John Baldwin writes:
> >On Tuesday 20 January 2004 01:23 pm, Poul-Henning Kamp wrote:
> >> In message <200401201234.45472.jhb at FreeBSD.org>, John Baldwin writes:
> >> >On Monday 19 January 2004 04:27 pm, Poul-Henning Kamp wrote:
> >> >>   Log:
> >> >>   Add linenumber and source filename to panic(9) output.
>
> This thread already represents a prime example of the worst of the
> FreeBSD projects social dimension:  The Gratuituos Bike Shed.
>
> On a generic kernel, this change adds 32088 non-executed bytes of
> text-segment to i386/GENERIC.
>
> For reference this 0.5% of the kernel and 12% of the size of the
> ACPI module.

I'm not against this because of its size, I think it's the wrong way to
go.

> And yes, we can do much better, tracebacks would be a great start,
> better panics also a great thing.

We already have tracebacks if options DDB is enabled.  (It probably should
be for the GENERIC installed by sysinstall).

> And no, this does not solve the mid-east crisis, but it is still
> an improvement, even for me:  If I save 1 minute because I do not
> have to hunt for the panic in the first place, then that is one
> minute more I can spend on the code.
>
> So if this makes your eyes water, I suggest you comment it out in
> your local source tree and pop in on the next meeting in your
> local user-group for some much needed perspective.

The 10 seconds of grep time (guess I'm faster on this) is far outweighed
by the one to INF email roundtrips necessary to get a user to enable
options DDB, recompile, set up a serial console, etc.  Most panics are
page faults so you have to go to the traceback anyway as knowing the
file/line of the page fault handler is not useful.

I appreciate your desire to improve debugging and hope you'll be
interested in tackling some of the outstanding issues (like the ones I
listed).

-Nate


More information about the cvs-src mailing list