Felix Radensky wrote:
>
> Stuart Hughes wrote:
>
> > Felix Radensky wrote:
> > >
> > > [EMAIL PROTECTED] wrote:
> > >
> > > > On Thu, Oct 28, 1999 at 01:21:07PM +0200, Felix Radensky wrote:
> > > > > Hi,
> > > > >
> > > > > I've applied rtl v1.3 to linux-2.0.37 and changed
> > > > > CFLAGS to -Wall -Wstrict-prototypes -O2 -fno-omit-frame-pointer
> > > > > -fno-strength-reduce
> > > > > The difference is -fno-omit-frame-pointer flag, that is needed for
> > > > > proper stack trace
> > > > > debugging. The kernel just cannot boot. The normal kernel boots fine.
> > > > > Any ideas would
> > > > > be greatly appreciated.
> > > >
> > > > Do you mean that the normal kernel boots fine when compiled with those
> > >
> > > Exactly. Some kernel debuggers require kernel to be compiled with -g
> > > -fno-omit-frame-pointers
> > > in order for stack backtracing to work properly. This is true for kdb
> > > (http://reality.sgi.com/slurn_engr/)
> > > and ikd (ftp://e-mind.com/pub/andrea/ikd/). GCC-HOWTO also says that
> > > -fno-omit-frame-pointers
> > > should be used for debugging.
> > > Normal kernels (I've tried 2.0.35, 2.0.36 & 2.0.37 ) when compiled with
> > > this flag (gcc-2.7.2.3) boot
> > > ok. With kernels patched with rtl (v1.2 & v1.3) the system prints
> > > "Loading linux ..." and reboots immediately.
> > >
> > > By the way, 2.2.13 kernel patched with rtl v2.0 and compiled with
> > > -fno-omit-frame-pointers has no problems booting. But since we use
> > > 2.0.x series kernel, it would be nice to have rtl working as expected
> > > with these kernels also.
> > >
> >
> > Hi Felix,
> >
> > I don't know if this helps you, but I use a modified gsbstubs patch to
> > do Linux kernel debugging. The combination I use is Linux 2.0.36 + RTL
> > 0.9J (which are close to what you have). If I look in the Linux
> > Makefile, the CFLAGS are:
> >
> > CFLAGS = -Wall -Wstrict-prototypes -O2 -g -fomit-frame-pointer
> > -fno-strength-reduce
> >
> > This work without any problem, using host/target debug over a serial
> > line (the host has the unstripped vmlinux, the target the zImage).
> >
> > The gsbstubs package can be found at:
> >
> > ftp://ftp.gcom.com/pub/linux/src/gdbstub-2.0.36/
> >
> > Note: this works for 2.2 as well despite the name.
> >
> > Regards
> >
> > Stuart.
>
> Hi, Stuart
>
> We are using almost the same configuration: Linux 2.0.35 + RTL 0.9F +
> gdbstub-2.0.36.
> But we have problems viewing the values of local variables and tracing the
> function calls.
> I thought that compiling the kernel with -fno-omit-frame-pointers should solve
> the problem.
> Are you saying that in your configuration all our problems is not an issue ?
Hi Felix,
I've been working mostly on 2.2.10+RTAI0.7, although I briefly looked
at the 2.0.36 + 0.9J combination. On the 2.2 kernel (and possibly the
2.0), there was a stack pointer problem that caused using the 'next'
command to fail, also auto variables were invalid. I have attached the
arch/i386/kernel/i386-stubs.c file that contains my fixes (see line
740). I'm *not* sure this applies to 2.0 (you need to look carefully
through entry.S I think).
Regards
Stuart.
i386-stub.c
