Re: [PATCH v4 00/10] Add Kernel Concurrency Sanitizer (KCSAN)

From: Paul E. McKenney
Date: Sat Nov 16 2019 - 10:40:57 EST


On Sat, Nov 16, 2019 at 09:20:54AM +0100, Marco Elver wrote:
> On Fri, 15 Nov 2019 at 21:43, Paul E. McKenney <paulmck@xxxxxxxxxx> wrote:
> >
> > On Fri, Nov 15, 2019 at 06:14:46PM +0100, Marco Elver wrote:
> > > On Fri, 15 Nov 2019 at 17:42, Paul E. McKenney <paulmck@xxxxxxxxxx> wrote:
> > > >
> > > > On Fri, Nov 15, 2019 at 01:02:08PM +0100, Marco Elver wrote:
> > > > > On Thu, 14 Nov 2019 at 23:16, Paul E. McKenney <paulmck@xxxxxxxxxx> wrote:
> > > > > >
> > > > > > On Thu, Nov 14, 2019 at 10:33:03PM +0100, Marco Elver wrote:
> > > > > > > On Thu, 14 Nov 2019, Paul E. McKenney wrote:
> > > > > > >
> > > > > > > > On Thu, Nov 14, 2019 at 07:02:53PM +0100, Marco Elver wrote:
> > > > > > > > > This is the patch-series for the Kernel Concurrency Sanitizer (KCSAN).
> > > > > > > > > KCSAN is a sampling watchpoint-based *data race detector*. More details
> > > > > > > > > are included in **Documentation/dev-tools/kcsan.rst**. This patch-series
> > > > > > > > > only enables KCSAN for x86, but we expect adding support for other
> > > > > > > > > architectures is relatively straightforward (we are aware of
> > > > > > > > > experimental ARM64 and POWER support).
> > > > > > > > >
> > > > > > > > > To gather early feedback, we announced KCSAN back in September, and have
> > > > > > > > > integrated the feedback where possible:
> > > > > > > > > http://lkml.kernel.org/r/CANpmjNPJ_bHjfLZCAPV23AXFfiPiyXXqqu72n6TgWzb2Gnu1eA@xxxxxxxxxxxxxx
> > > > > > > > >
> > > > > > > > > The current list of known upstream fixes for data races found by KCSAN
> > > > > > > > > can be found here:
> > > > > > > > > https://github.com/google/ktsan/wiki/KCSAN#upstream-fixes-of-data-races-found-by-kcsan
> > > > > > > > >
> > > > > > > > > We want to point out and acknowledge the work surrounding the LKMM,
> > > > > > > > > including several articles that motivate why data races are dangerous
> > > > > > > > > [1, 2], justifying a data race detector such as KCSAN.
> > > > > > > > >
> > > > > > > > > [1] https://lwn.net/Articles/793253/
> > > > > > > > > [2] https://lwn.net/Articles/799218/
> > > > > > > >
> > > > > > > > I queued this and ran a quick rcutorture on it, which completed
> > > > > > > > successfully with quite a few reports.
> > > > > > >
> > > > > > > Great. Many thanks for queuing this in -rcu. And regarding merge window
> > > > > > > you mentioned, we're fine with your assumption to targeting the next
> > > > > > > (v5.6) merge window.
> > > > > > >
> > > > > > > I've just had a look at linux-next to check what a future rebase
> > > > > > > requires:
> > > > > > >
> > > > > > > - There is a change in lib/Kconfig.debug and moving KCSAN to the
> > > > > > > "Generic Kernel Debugging Instruments" section seems appropriate.
> > > > > > > - bitops-instrumented.h was removed and split into 3 files, and needs
> > > > > > > re-inserting the instrumentation into the right places.
> > > > > > >
> > > > > > > Otherwise there are no issues. Let me know what you recommend.
> > > > > >
> > > > > > Sounds good!
> > > > > >
> > > > > > I will be rebasing onto v5.5-rc1 shortly after it comes out. My usual
> > > > > > approach is to fix any conflicts during that rebasing operation.
> > > > > > Does that make sense, or would you prefer to send me a rebased stack at
> > > > > > that point? Either way is fine for me.
> > > > >
> > > > > That's fine with me, thanks! To avoid too much additional churn on
> > > > > your end, I just replied to the bitops patch with a version that will
> > > > > apply with the change to bitops-instrumented infrastructure.
> > > >
> > > > My first thought was to replace 8/10 of the previous version of your
> > > > patch in -rcu (047ca266cfab "asm-generic, kcsan: Add KCSAN instrumentation
> > > > for bitops"), but this does not apply. So I am guessing that I instead
> > > > do this substitution when a rebase onto -rc1..
> > > >
> > > > Except...
> > > >
> > > > > Also considering the merge window, we had a discussion and there are
> > > > > some arguments for targeting the v5.5 merge window:
> > > > > - we'd unblock ARM and POWER ports;
> > > > > - we'd unblock people wanting to use the data_race macro;
> > > > > - we'd unblock syzbot just tracking upstream;
> > > > > Unless there are strong reasons to not target v5.5, I leave it to you
> > > > > if you think it's appropriate.
> > > >
> > > > My normal process is to send the pull request shortly after -rc5 comes
> > > > out, but you do call out some benefits of getting it in sooner, so...
> > > >
> > > > What I will do is to rebase your series onto (say) -rc7, test it, and
> > > > see about an RFC pull request.
> > > >
> > > > One possible complication is the new 8/10 patch. But maybe it will
> > > > apply against -rc7?
> > > >
> > > > Another possible complication is this:
> > > >
> > > > scripts/kconfig/conf --syncconfig Kconfig
> > > > *
> > > > * Restart config...
> > > > *
> > > > *
> > > > * KCSAN: watchpoint-based dynamic data race detector
> > > > *
> > > > KCSAN: watchpoint-based dynamic data race detector (KCSAN) [N/y/?] (NEW)
> > > >
> > > > Might be OK in this case because it is quite obvious what it is doing.
> > > > (Avoiding pain from this is the reason that CONFIG_RCU_EXPERT exists.)
> > > >
> > > > But I will just mention this in the pull request.
> > > >
> > > > If there is a -rc8, there is of course a higher probability of making it
> > > > into the next merge window.
> > > >
> > > > Fair enough?
> > >
> > > Totally fine with that, sounds like a good plan, thanks!
> > >
> > > If it helps, in theory we can also drop and delay the bitops
> > > instrumentation patch until the new bitops instrumentation
> > > infrastructure is in 5.5-rc1. There won't be any false positives if
> > > this is missing, we might just miss a few data races until we have it.
> >
> > That sounds advisable for an attempt to hit this coming merge window.
> >
> > So just to make sure I understand, I drop 8/10 and keep the rest during
> > a rebase to 5.4-rc7, correct?
>
> Yes, that's right.

Very good, I just now pushed a "kcsan" branch on -rcu, and am running
rcutorture, first without KCSAN enabled and then with it turned on.
If all that works out, I set my -next branch to that point and see what
-next testing and kbuild test robot think about it. If all goes well,
an RFC pull request.

Look OK?

Thanx, Paul