Re: [PATCH v5 8/8] unwind: arm64: Use sframe to unwind interrupt frames

Hi Mark,

I'm sending a v6 shortly that should address all/most of your
feedback, but I wanted to circle back on a question you had:

On Fri, May 1, 2026 at 9:46 AM Mark Rutland <[email protected]> wrote:
> > +     /*
> > +      * Consume RA and FP from the stack. The frame record puts FP at a lower
> > +      * address than RA, so we always read FP first.
> > +      */
> > +     if (frame.fp.rule & UNWIND_RULE_DEREF &&
> > +         !get_word(&state->common, &fp))
> > +             return -EINVAL;
>
> Why is this get_word() rather than get_consume_word()?

I use get_word() here because get_consume_word(), in calling
unwind_consume_stack() under the hood, consumes the stack up to the
given address+size such that another unwind step cannot consume it
again. If the subsequent call to get_consume_word() fails, the stack
needs to be in a state such that we can fall back on a frame pointer
unwind. But if we were to use get_consume_word() here, the fallback
call to kunwind_next_frame_record() would not be able to consume the
FP from the stack because it would already have been consumed by the
failed call to unwind_next_frame_sframe().

By only calling get_consume_word() on the RA at the end, we defer
making any changes to the underlying unwind state stack until we are
sure the SFrame unwind step will succeed.

>
> > +
> > +     if (frame.ra.rule & UNWIND_RULE_DEREF &&
> > +         get_consume_word(&state->common, &ra))
> > +             return -EINVAL;
> > +
> > +     state->common.pc = ra;
> > +     state->common.sp = cfa;

Please let me know if this reasoning seems sound.

Thanks,
Dylan