Re: ddb(4) spoils kernel stack in CURRENT?

On Wednesday 20 December 2006 15:41, Kip Macy wrote:
> I worried that gdb probably had workaround for the large stack argument.
> I'll have to dig it up, thanks for the heads up.

Both kgdb and ddb have explicit knowledge of trapframes and interrupt frames
and know how to step across them, so both of those will need to be updated.
For ddb, you'll have to update the 'db_nextframe()' function in
sys/{amd64,i386}/{amd64,i386}/db_trace.c.  Specifically this section:

        /*
         * Point to base of trapframe which is just above the
         * current frame.
         */
        if (frame_type == INTERRUPT)
                tf = (struct trapframe *)((int)*fp + 12);
        else
                tf = (struct trapframe *)((int)*fp + 8);

The easy change is to just bump those up by 4 (by 8 on amd64).  Similarly,
for kgdb you'll have to update src/gnu/usr.bin/gdb/kgdb/trgt_{amd64,i386}.c.

NB: kgdb doesn't properly distinguish interrupt frames from other trap
frames.  It treats nearly all trap frames as interrupt frames (though its
handling of interrupt frames is more like a trap frame I think).

>   -Kip
> 
> On 12/20/06, Dmitry Pryanishnikov <dmitry_at_atlantis.dp.ua> wrote:
> >
> >
> > Hello!
> >
> > On Wed, 20 Dec 2006, Kostik Belousov wrote:
> > >>> So it looks like a regression in CURRENT vs RELENG_6 (either ddb
> > 'spoils'
> > >>> the stack somehow, or kgdb fails to unwind it).
> > >
> > > Could you further localize the problem, i.e. try to backtrace CURRENT
> > dump
> >
> >    Good news: I've managed to localize the bug! I'm Feeling Lucky (TM) ;)
> > just because CURRENT on my notebook was updated approx. at 17-Dec 00:00,
> > and it didn't manifest such a behaviour! So it was easy to identify the
> > regression - it comes with the following commit:
> >
> > -----------------------------------------------------------------------
> >
> > Date: Sun, 17 Dec 2006 05:07:01 +0000 (UTC)
> > From: Kip Macy <kmacy_at_freebsd.org>
> > To: src-committers_at_freebsd.org, cvs-src_at_freebsd.org, cvs-all_at_freebsd.org
> > Subject: cvs commit: src/sys/i386/i386 apic_vector.s exception.slocal_apic.c
> >           trap.c vm86.c vm86bios.s src/sys/i386/include apicvar.h
> >           src/sys/i386/isa atpic.c atpic_vector.s icu.h
> >
> > kmacy       2006-12-17 05:07:01 UTC
> >
> >    FreeBSD src repository
> >
> >    Modified files:
> >      sys/i386/i386        apic_vector.s exception.s local_apic.c
> >                           trap.c vm86.c vm86bios.s
> >      sys/i386/include     apicvar.h
> >      sys/i386/isa         atpic.c atpic_vector.s icu.h
> >    Log:
> >    Evidently FreeBSD has long relied on the compiler to treat structures
> >    passed by value (trap frames) as if they were in fact being passed by
> >    reference. For better or worse, this incorrect behaviour is no longer
> >    present in gcc 4.1. In this patch I convert all trapframe arguments to
> >    be explicitly pass by reference. I also remove vm86_initflags, pushing
> >    the very little work that it actually does up into vm86_prepcall.
> >
> > -----------------------------------------------------------------------
> >
> > So kernel built from sources as of date=2006.12.17.05.00.00 gives dump
> > with analyzable backtrace, and kernel built from sources as of
> > date=2006.12.17.05.10.00 (which include this commit) gives dump
> > which confuses kgdb. I believe that commit itself is correct,
> > but kgdb contains some workaround against the old (incorrect) behaviour
> > of the kernel, so it's the kgdb that should be fixed.
> >
> > Sincerely, Dmitry
> > --
> > Atlantis ISP, System Administrator
> > e-mail:  dmitry_at_atlantis.dp.ua
> > nic-hdl: LYNX-RIPE
> > _______________________________________________
> > freebsd-current_at_freebsd.org mailing list
> > http://lists.freebsd.org/mailman/listinfo/freebsd-current
> > To unsubscribe, send any mail to "freebsd-current-unsubscribe_at_freebsd.org"
> >
> _______________________________________________
> freebsd-current_at_freebsd.org mailing list
> http://lists.freebsd.org/mailman/listinfo/freebsd-current
> To unsubscribe, send any mail to "freebsd-current-unsubscribe_at_freebsd.org"
> 

-- 
John Baldwin