Re: [PATCH RFC v6 00/21] DEPT(Dependency Tracker)

From: Byungchul Park
Date: Sun May 08 2022 - 21:30:33 EST


On Wed, May 04, 2022 at 11:17:02AM -0700, Linus Torvalds wrote:
> On Wed, May 4, 2022 at 1:19 AM Byungchul Park <byungchul.park@xxxxxxx> wrote:
> >
> > Hi Linus and folks,
> >
> > I've been developing a tool for detecting deadlock possibilities by
> > tracking wait/event rather than lock(?) acquisition order to try to
> > cover all synchonization machanisms.
>
> So what is the actual status of reports these days?

I'd like to mention one important thing here. Reportability would get
stronger if the more wait-event pairs get tagged everywhere DEPT can
work.

Everything e.g. HW-SW interface, any retry logic and so on can be a
wait-event pair if they work wait or event anyway. For example, polling
on an IO mapped read register and initiating the HW to go for the event
also can be a pair. Definitely those make DEPT more useful.

---

The way to use the APIs:

1. Define SDT(Simple Dependency Tracker)

DEFINE_DEPT_SDT(my_hw_event); <- add this

2. Tag on the waits

sdt_wait(&my_hw_event); <- add this
... retry logic until my hw work done ... <- the original code

3. Tag on the events

sdt_event(&my_hw_event); <- add this
run_my_hw(); <- the original code

---

These are all we should do. I believe DEPT would be a very useful tool
once all wait-event pairs get tagged by the developers in all subsystems
and device drivers.

Byungchul

> Last time I looked at some reports, it gave a lot of false positives
> due to mis-understanding prepare_to_sleep().
>
> For this all to make sense, it would need to not have false positives
> (or at least a very small number of them together with a way to sanely
> get rid of them), and also have a track record of finding things that
> lockdep doesn't.
>
> Maybe such reports have been sent out with the current situation, and
> I haven't seen them.
>
> Linus