Re: [PATCH 3/3] kcov: introduce extended PC coverage collection mode

[Jann Horn]

On Fri, Mar 13, 2026 at 8:58 AM Dmitry Vyukov <dvyukov@xxxxxxxxxx> wrote:
> On Wed, 11 Mar 2026 at 22:06, Jann Horn <jannh@xxxxxxxxxx> wrote:
> > This is the second half of CONFIG_KCOV_EXT_RECORDS.
> >
> > Introduce a new KCOV mode KCOV_TRACE_PC_EXT which replaces the upper 8 bits
> > of recorded instruction pointers with metadata. For now, userspace can use
> > this metadata to distinguish three types of records:
[...]
> > @@ -1519,8 +1519,10 @@ struct task_struct {
> >         int                             kcov_sequence;
> >
> >         /* Collect coverage from softirq context: */
> > -       unsigned int                    kcov_softirq;
> > -#endif
> > +       unsigned int                    kcov_softirq : 1;
> > +       /* Emit KCOV records in extended format: */
> > +       unsigned int                    kcov_ext_format : 1;
>
> Setting/saving/restoring this flag is fragile. I afraid some of future
> patches can break it in some corner cases.
> Can we have a new kcov_mode and use some mask check on tracing fast
> path, so that it's as cheap as the current == kcov_mode comparison?

Yeah, I also thought that what I'm doing here didn't look particularly
pretty... I'll try to implement something like what you suggested for
v2.

> >  void notrace __sanitizer_cov_trace_pc_exit(void)
> >  {
> > +       unsigned long record;
> > +
> > +       /*
> > +        * Unlike __sanitizer_cov_trace_pc_entry(), this PC should only be
> > +        * reported in extended mode.
>
> It would help to explain _why_. The fact that it's not traced is
> already in the code.

Right, I'll change the comment to explain that this callback isn't at
the start of a basic block, and that the basic block is already
covered by a preceding hook call.