Re: [REGRESSION] x86/debug: After PTRACE_SINGLESTEP DR_STEP is no longer reported in dr6

[Kyle Huey]

On Mon, Oct 26, 2020 at 9:05 AM Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
>
> On Mon, Oct 26, 2020 at 04:55:21PM +0100, Peter Zijlstra wrote:
> > On Mon, Oct 26, 2020 at 07:33:08AM -0700, Kyle Huey wrote:
> > > After resuming a ptracee with PTRACE_SINGLESTEP, in the following
> > > ptrace stop retrieving the dr6 value for the tracee gets a value that
> > > does not include DR_STEP (it is in fact always DR6_RESERVED). I
> > > bisected this to the 13cb73490f475f8e7669f9288be0bcfa85399b1f merge. I
> > > did not bisect further.
> > >
> > > I don't see any handling to ever set DR_STEP in virtual_dr6, so I
> > > think this code is just broken.
> > >
> > > Sorry for not testing this when I was CCd on the original patch series :)
> >
> > Urgh, now I have to try and remember how all that worked again ...
> >
> > I suspect it's either one (or both) of the last two:
> >
> >   f4956cf83ed1 ("x86/debug: Support negative polarity DR6 bits")
> >   d53d9bc0cf78 ("x86/debug: Change thread.debugreg6 to thread.virtual_dr6")
> >
> >
> > Just to clarify, the sequence is something like:
> >
> >  - tracer: ptrace(PTRACE_SINGLESTEP)
> >  - tracee: #DB, DR6 contains DR_STEP
> >  - tracer: ptrace_get_debugreg(6)
> >
> > ?
> >
> > You're right that that would be broken, let me try and figure out what
> > the best way would be 'fix' that.
> >
> > Also, can you confirm that pthread_set_debugreg(6) should not do
> > anything useful?
>
>
> Does something like this make sense?
>
>
> diff --git a/arch/x86/kernel/traps.c b/arch/x86/kernel/traps.c
> index 3c70fb34028b..0e7641ac19a8 100644
> --- a/arch/x86/kernel/traps.c
> +++ b/arch/x86/kernel/traps.c
> @@ -799,6 +799,13 @@ static __always_inline unsigned long debug_read_clear_dr6(void)
>          */
>         current->thread.virtual_dr6 = 0;
>
> +       /*
> +        * If PTRACE requested SINGLE(BLOCK)STEP, make sure to reflect that in
> +        * the ptrace visible DR6 copy.
> +        */
> +       if (test_thread_flag(TIF_BLOCKSTEP) || test_thread_flag(TIF_SINGLESTEP))
> +               current->thread.virtual_dr6 |= dr6 & DR_STEP;
> +
>         /*
>          * The SDM says "The processor clears the BTF flag when it
>          * generates a debug exception."  Clear TIF_BLOCKSTEP to keep

I don't think the `& DR_STEP` should be necessary, that bit should be
set by the hardware, I believe.

Also if a program sets TF on itself in EFLAGS and generates a trap it
should still have DR_STEP set in the ptrace-visible dr6, which this
won't do.

Is there a reason not to always copy dr6 into virtual_dr6 here,
regardless of TIF_SINGLESTEP/etc?

- Kyle