Re: [PATCH 1/8] kcsan: Add Kernel Concurrency Sanitizer infrastructure

[Marco Elver]

On Wed, 16 Oct 2019 at 20:44, Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
>
> On Wed, Oct 16, 2019 at 10:39:52AM +0200, Marco Elver wrote:
>
> > +bool __kcsan_check_watchpoint(const volatile void *ptr, size_t size,
> > +                           bool is_write)
> > +{
> > +     atomic_long_t *watchpoint;
> > +     long encoded_watchpoint;
> > +     unsigned long flags;
> > +     enum kcsan_report_type report_type;
> > +
> > +     if (unlikely(!is_enabled()))
> > +             return false;
> > +
> > +     watchpoint = find_watchpoint((unsigned long)ptr, size, !is_write,
> > +                                  &encoded_watchpoint);
> > +     if (watchpoint == NULL)
> > +             return true;
> > +
> > +     flags = user_access_save();
>
> Could use a comment on why find_watchpoint() is save to call without
> user_access_save() on.

Thanks, will add a comment for v2.

> > +     if (!try_consume_watchpoint(watchpoint, encoded_watchpoint)) {
> > +             /*
> > +              * The other thread may not print any diagnostics, as it has
> > +              * already removed the watchpoint, or another thread consumed
> > +              * the watchpoint before this thread.
> > +              */
> > +             kcsan_counter_inc(kcsan_counter_report_races);
> > +             report_type = kcsan_report_race_check_race;
> > +     } else {
> > +             report_type = kcsan_report_race_check;
> > +     }
> > +
> > +     /* Encountered a data-race. */
> > +     kcsan_counter_inc(kcsan_counter_data_races);
> > +     kcsan_report(ptr, size, is_write, raw_smp_processor_id(), report_type);
> > +
> > +     user_access_restore(flags);
> > +     return false;
> > +}
> > +EXPORT_SYMBOL(__kcsan_check_watchpoint);
> > +
> > +void __kcsan_setup_watchpoint(const volatile void *ptr, size_t size,
> > +                           bool is_write)
> > +{
> > +     atomic_long_t *watchpoint;
> > +     union {
> > +             u8 _1;
> > +             u16 _2;
> > +             u32 _4;
> > +             u64 _8;
> > +     } expect_value;
> > +     bool is_expected = true;
> > +     unsigned long ua_flags = user_access_save();
> > +     unsigned long irq_flags;
> > +
> > +     if (!should_watch(ptr))
> > +             goto out;
> > +
> > +     if (!check_encodable((unsigned long)ptr, size)) {
> > +             kcsan_counter_inc(kcsan_counter_unencodable_accesses);
> > +             goto out;
> > +     }
> > +
> > +     /*
> > +      * Disable interrupts & preemptions, to ignore races due to accesses in
> > +      * threads running on the same CPU.
> > +      */
> > +     local_irq_save(irq_flags);
> > +     preempt_disable();
>
> Is there a point to that preempt_disable() here?

We want to avoid being preempted while the watchpoint is set up;
otherwise, we would report data-races for CPU-local data, which is
incorrect. An alternative would be adding the source CPU to the
watchpoint, and checking that the CPU != this_cpu. There are several
problems with that alternative:
1. We do not want to steal more bits from the watchpoint encoding for
things other than read/write, size, and address, as not only does it
affect accuracy, it would also increase performance overhead in the
fast-path.
2. As a consequence, if we get a preemption and run a task on the same
CPU, and there *is* a genuine data-race, we would *not* report it; and
since this is the common case (and not accesses to CPU-local data), it
makes more sense (from a data-race detection PoV) to simply disable
preemptions and ensure that all tasks are run on other CPUs as well as
avoid the problem of point (1).

I can add a comment to that effect here for v2.

Thanks,
-- Marco