On Wed, Jul 06, 2005 at 12:29:12PM +0900, Atsushi Nemoto wrote:
> >>>>> On Tue, 5 Jul 2005 21:03:09 +0100, Ralf Baechle <ralf@linux-mips.org>
> >>>>> said:
> ralf> If the WCHAN column of ps axl is supposed to be any useful we
> ralf> need to unwind the stack until we find the caller of the
> ralf> sleeping or scheduling function. Very useful for debugging.
>
> Yes, but many sleeping/scheduling (such as schedule_timeout(),
> __down(), etc.) are compiled without -fno-omit-frame-pointer, so
> you can not find the caller of such functions anyway.
Without additional information a framepointer isn't terribly useful on
MIPS. Unfortunately it's stored at a non-constant offset in a function's
stackframe.
> And some sleeping/scheduling functions which are compiled with
> -fno-omit-frame-pointer are static or deprecated (sleep_on(), etc.)
>
> You can find the caller of "schedule()" even with simple
> thread_saved_pc(). I think it is enough so I do not think it is worth
> to fix (and maintain) current minfo[].
The alternative would be to finally bite the bullet and add a wchan field
to thread_struct and initialize it in all the sleeping functions.
The IA-64 people have something like a DWARF-based frame unwinder but
that just seems to heavy.
Ralf
|