Re: [RFC][PATCH 4/7] x86/debug: Move historical SYSENTER junk into exc_debug_kernel()

[peterz]

On Thu, Aug 20, 2020 at 10:16:59AM -0500, Josh Poimboeuf wrote:
> On Thu, Aug 20, 2020 at 05:08:41PM +0200, peterz@xxxxxxxxxxxxx wrote:
> > On Thu, Aug 20, 2020 at 10:45:12AM -0400, Brian Gerst wrote:
> > > On Thu, Aug 20, 2020 at 6:53 AM Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
> > > >
> > > >
> > > > Signed-off-by: Peter Zijlstra (Intel) <peterz@xxxxxxxxxxxxx>
> > > > ---
> > > >  arch/x86/kernel/traps.c |   24 ++++++++++++------------
> > > >  1 file changed, 12 insertions(+), 12 deletions(-)
> > > >
> > > > --- a/arch/x86/kernel/traps.c
> > > > +++ b/arch/x86/kernel/traps.c
> > > > @@ -820,18 +820,6 @@ static void handle_debug(struct pt_regs
> > > >                 goto out;
> > > >         }
> > > >
> > > > -       if (WARN_ON_ONCE((dr6 & DR_STEP) && !user_mode(regs))) {
> > > > -               /*
> > > > -                * Historical junk that used to handle SYSENTER single-stepping.
> > > > -                * This should be unreachable now.  If we survive for a while
> > > > -                * without anyone hitting this warning, we'll turn this into
> > > > -                * an oops.
> > > > -                */
> > > > -               tsk->thread.debugreg6 &= ~DR_STEP;
> > > > -               set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
> > > > -               regs->flags &= ~X86_EFLAGS_TF;
> > > > -       }
> > > > -
> > > >         si_code = get_si_code(tsk->thread.debugreg6);
> > > >         if (tsk->thread.debugreg6 & (DR_STEP | DR_TRAP_BITS) || user_icebp)
> > > >                 send_sigtrap(regs, 0, si_code);
> > > > @@ -874,6 +862,18 @@ static __always_inline void exc_debug_ke
> > > >         if (kprobe_debug_handler(regs))
> > > >                 goto out;
> > > >
> > > > +       if (WARN_ON_ONCE(dr6 & DR_STEP)) {
> > > > +               /*
> > > > +                * Historical junk that used to handle SYSENTER single-stepping.
> > > > +                * This should be unreachable now.  If we survive for a while
> > > > +                * without anyone hitting this warning, we'll turn this into
> > > > +                * an oops.
> > > > +                */
> > > > +               dr6 &= ~DR_STEP;
> > > > +               set_thread_flag(TIF_SINGLESTEP);
> > > > +               regs->flags &= ~X86_EFLAGS_TF;
> > > > +       }
> > > > +
> > > >         handle_debug(regs, dr6, false);
> > > >
> > > >  out:
> > > 
> > > Can this be removed or changed to a BUG()?  The warning has been there
> > > since 2016 and nobody has apparently complained about it.
> > 
> > Something like:
> > 
> > 	/*
> > 	 * The kernel doesn't do TF outside of Kprobes, so if we get
> > 	 * here, something's fishy.
> > 	 */
> > 	BUG_ON(dr6 & DR_STEP);
> > 
> > ?
> 
> What about kernel debuggers like kgdb and qemu gdb?

qemu-gdb stub should eat the event before it lands in the guest, kgdb
should eat it through the notifier.

Thank you!

Because that means I got the order wrong :/