Re: Please can I have a stable KCSAN branch for 5.8?

From: Paul E. McKenney
Date: Thu May 07 2020 - 11:32:15 EST


On Thu, May 07, 2020 at 11:34:55AM +0200, Marco Elver wrote:
> On Wed, 6 May 2020 at 23:47, Paul E. McKenney <paulmck@xxxxxxxxxx> wrote:
> >
> > On Wed, May 06, 2020 at 12:37:13PM -0700, Paul E. McKenney wrote:
> > > On Wed, May 06, 2020 at 09:11:23PM +0200, Marco Elver wrote:
> > > > On Wed, 6 May 2020 at 20:02, Paul E. McKenney <paulmck@xxxxxxxxxx> wrote:
> > > > >
> > > > > On Wed, May 06, 2020 at 05:26:56PM +0200, Marco Elver wrote:
> > > > > > On Wed, 6 May 2020 at 16:41, Will Deacon <will@xxxxxxxxxx> wrote:
> > > > > > >
> > > > > > > Hi Paul,
> > > > > > >
> > > > > > > Cheers for the quick reply!
> > > > > > >
> > > > > > > On Wed, May 06, 2020 at 07:36:16AM -0700, Paul E. McKenney wrote:
> > > > > > > > On Wed, May 06, 2020 at 02:28:17PM +0100, Will Deacon wrote:
> > > > > > > > > I'm looking to rebase my READ_ONCE() series [1] on top of the KCSAN patches
> > > > > > > > > so that we can get them in for 5.8. However, tip/locking/kcsan seems to be
> > > > > > > > > missing some bits:
> > > > > > > > >
> > > > > > > > > * An update to checkpatch.pl to warn about missing comments for
> > > > > > > > > data_race():
> > > > > > > > >
> > > > > > > > > https://lore.kernel.org/r/20200401101714.44781-1-elver@xxxxxxxxxx
> > > > > > > >
> > > > > > > > For some reason, I thought this was going up some other tree, but I do
> > > > > > > > not see it in -next. So unless I hear otherwise, I will pull it into
> > > > > > > > the v5.8 kcsan branch.
> > > > > > >
> > > > > > > Brill, thanks.
> > > > > > >
> > > > > > > > > * I'm unable to apply these two patches from Marco that are needed for
> > > > > > > > > my READ_ONCE() work:
> > > > > > > > >
> > > > > > > > > https://lore.kernel.org/lkml/20200424154730.190041-1-elver@xxxxxxxxxx/
> > > > > > > > >
> > > > > > > > > I think these depend on stuff that has been queued by Paul, and appears
> > > > > > > > > in linux-next, but to be honest with you I'm quite confused about what
> > > > > > > > > is queued for 5.8 and what isn't.
> > > > > > > >
> > > > > > > > This one is queued, but I currently have it in the v5.9 pile (but
> > > > > > > > tentatively for v5.8). Unless Marco tells me otherwise, I will move it
> > > > > > > > to the v5.8 branch, which will be part of my pull request next week.
> > > > > > >
> > > > > > > Great, then this would all show up on tip/locking/kscan, right?
> > > > > > >
> > > > > > > > > What's the best base for me to use?
> > > > > > > >
> > > > > > > > The -next tree has the latter, but not yet the former.
> > > > > > >
> > > > > > > That probably means -next is good enough for me to cook a new version of my
> > > > > > > series, and then I can make a proper branch next week.
> > > > > > >
> > > > > > > > Hopefully we can get this straightened out, and please accept my apologies
> > > > > > > > for the hassle!
> > > > > > >
> > > > > > > No need to apologise, I just couldn't figure out what was what and decided
> > > > > > > it was easier to ask the experts ;)
> > > > > >
> > > > > > Just confirming that I don't see any issues with the plan -- the
> > > > > > patches that Will needs are good to go into the v5.8 branch.
> > > > >
> > > > > OK, I have updated -rcu's kcsan and kcsan-dev branches. Could you please
> > > > > double-check, given that pull-request time is quite soon?
> > > >
> > > > I believe "objtool, kcsan: Add kcsan_disable_current() and
> > > > kcsan_enable_current_nowarn()" is missing, which should go after
> > > > "kcsan: Add __kcsan_{enable,disable}_current() variants".
> > >
> > > Thank you for checking! I will move that one also.
> >
> > And it is moved. FYI, I will likely be sending my KCSAN pull request
> > late tomorrow (Thursday) Pacific Time, a few days earlier than normal.
>
> Looks good, thank you!

And it did pass overnight testing, so looking good! Next step is a
quick test without KCSAN enabled.

Thanx, Paul