Re: [ltt-dev] Re: [PATCH] LTT for 2.5.36 9/9: ARM trace support

From: Frank Rowand (frowand@mvista.com)
Date: Thu Sep 19 2002 - 18:27:06 EST


Russell King wrote:
>
> You should probably explicitly CC the architecture maintainers with
> these patches. Generally, there is no guarantee that they'll read
> lkml.
>
> On Thu, Sep 19, 2002 at 06:35:06PM -0400, Karim Yaghmour wrote:
> > diff -urpN linux-2.5.36/arch/arm/kernel/entry-common.S linux-2.5.36-ltt/arch/arm/kernel/entry-common.S
> > --- linux-2.5.36/arch/arm/kernel/entry-common.S Tue Sep 17 20:58:42 2002
> > +++ linux-2.5.36-ltt/arch/arm/kernel/entry-common.S Thu Sep 19 16:29:55 2002
> > @@ -35,6 +35,11 @@ ENTRY(__do_softirq)
> > * stack.
> > */
> > ret_fast_syscall:
> > +#if (CONFIG_TRACE || CONFIG_TRACE_MODULE)
> > + mov r7, r0 @ save returned r0
> > + bl SYMBOL_NAME(trace_real_syscall_exit)
> > + mov r0, r7
> > +#endif
>
> This misses the slow syscall exit path.

Thanks, I'll fix that.

>
> > +#if (CONFIG_TRACE || CONFIG_TRACE_MODULE)
> > +asmlinkage void trace_real_syscall_entry(int scno,struct pt_regs * regs)
> > +{
> > + int depth = 0;
> > + unsigned long end_code;
> > + unsigned long *fp; /* frame pointer */
> > + unsigned long lower_bound;
> > + unsigned long lr; /* link register */
> > + unsigned long *prev_fp;
> > + int seek_depth;
> > + unsigned long start_code;
> > + unsigned long *start_stack;
> > + trace_syscall_entry trace_syscall_event;
> > + unsigned long upper_bound;
> > + int use_bounds;
> > + int use_depth;
> > +
> > + trace_syscall_event.syscall_id = (uint8_t)scno;
> > + trace_syscall_event.address = instruction_pointer(regs);
> > +
> > + if (! (user_mode(regs) ))
> > + goto trace_syscall_end;
> > +
> > + if (trace_get_config(&use_depth,
> > + &use_bounds,
> > + &seek_depth,
> > + (void*)&lower_bound,
> > + (void*)&upper_bound) < 0)
> > + goto trace_syscall_end;
> > +
> > + if ((use_depth == 1) || (use_bounds == 1)) {
> > + fp = (unsigned long *)regs->ARM_fp;
>
> You can't rely on FP being set to anything real. Although the "APCS"
> ABI defines that FP will be either zero or a pointer to a valid frame,
> this isn't always true; a binary built with -fomit-frame-pointer will
> use FP for its own purposes. This means that there exists the possibility
> for a program without any frames on the stack (although we could be
> many functions deep within the program.)
>
> Do you care about this?

Yes, but there isn't much we can to about it. If a program doesn't have
valid frame pointers then we just won't be able to capture a valid
address of where the program made the syscall from. There are
plenty of paranoia checks to limit the search through the frames (even
with non-existent frame pointers), so this won't be catastrophic.

It looks like I could add one more paranoia check for the initial
value of regs->ARM_fp.

>
> --
> Russell King (rmk@arm.linux.org.uk) The developer of ARM Linux
> http://www.arm.linux.org.uk/personal/aboutme.html
>
> _______________________________________________
> ltt-dev mailing list
> ltt-dev@listserv.shafik.org
> http://www.listserv.shafik.org/listserv/listinfo/ltt-dev

Thanks for the comments!

-Frank

-- 
Frank Rowand <frank_rowand@mvista.com>
MontaVista Software, Inc
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/



This archive was generated by hypermail 2b29 : Mon Sep 23 2002 - 22:00:28 EST