2002-09-28 16:37:01

by David Mosberger

[permalink] [raw]
Subject: show_stack()/show_trace() prototypes

Ingo, the ksymoops patch adds these to linux/sched.h:

extern void show_trace(unsigned long *stack);
extern void show_stack(unsigned long *stack);

This is not good. In general, it is not possible to do a reliable
backtrace with just a stack pointer as a starting point (it is
necessary to have access to the entire "preserved" machine state
instead). I'd suggest to either change the argument to a task_struct
pointer (and update half a dozen platforms or so accordingly), or to
leave the declarations platform-specific like they were before.

Thanks,

--david

PS: force_sig_info() still drops the siginfo pointer passed to it.
Will you fix this soon as promised?


2002-09-28 18:21:15

by Maciej W. Rozycki

[permalink] [raw]
Subject: Re: show_stack()/show_trace() prototypes

On Sat, 28 Sep 2002, David Mosberger wrote:

> Ingo, the ksymoops patch adds these to linux/sched.h:
>
> extern void show_trace(unsigned long *stack);
> extern void show_stack(unsigned long *stack);
>
> This is not good. In general, it is not possible to do a reliable
> backtrace with just a stack pointer as a starting point (it is
> necessary to have access to the entire "preserved" machine state
> instead). I'd suggest to either change the argument to a task_struct
> pointer (and update half a dozen platforms or so accordingly), or to
> leave the declarations platform-specific like they were before.

Also the prototypes may differ, e.g. for MIPS/MIPS64 we have:

extern void show_trace(long *stack);

since MIPS addresses are signed and then, for consistency, we have:

extern void show_stack(long *stack);

as well.

--
+ Maciej W. Rozycki, Technical University of Gdansk, Poland +
+--------------------------------------------------------------+
+ e-mail: [email protected], PGP key available +

2002-09-29 00:16:09

by Paul Mackerras

[permalink] [raw]
Subject: Re: show_stack()/show_trace() prototypes

David Mosberger writes:

> Ingo, the ksymoops patch adds these to linux/sched.h:
>
> extern void show_trace(unsigned long *stack);
> extern void show_stack(unsigned long *stack);
>
> This is not good.

I agree. It is unnecessary to have these declarations in
linux/sched.h since these functions are only called from arch code.
So I would also vote for having the declarations arch-specific.

Paul.