RCU branching for the v4.19 merge window

RCU branching for the v4.19 merge window

[Paul E. McKenney]

Hello, Steve!

Another year, another difficult-to-branch set of RCU commits.

In happy contrast to last year, I can make some branches (SRCU, some
of the torture commits, and a few miscellaneous commits), but I will
likely end up with several short branches and one huge one.  My thought
is to keep the long branch, but email the patches out in a few separate
serieses, with each depending on its predecessor.  For example, one series
from the big branch would be folding the ->gpnum and ->completed fields
into a single ->gp_seq, which helps the RCU-flavor consolidation task.
Another series suppresses some rare false-positive splats that have been
plaguing me for more than a year.  Yet another series within this huge
branch applies and optimizes funnel locking for grace-period startup.

The problem is that the conversion to ->gp_seq has a very large footprint,
which of course generates lots of conflicts.  I could of course collapse
these commits into a single commit, but if I did that I would also defer
to the merge window following v4.19 due to the resulting loss of bisection
within that change.

Any advice?

The commits are for-mingo..rcu/dev in my -rcu tree.

							Thanx, Paul

Re: RCU branching for the v4.19 merge window

[Steven Rostedt]

On Thu, 17 May 2018 07:40:44 -0700
"Paul E. McKenney" <paulmck@linux.vnet.ibm.com> wrote:

> Hello, Steve!
> 
> Another year, another difficult-to-branch set of RCU commits.
> 
> In happy contrast to last year, I can make some branches (SRCU, some
> of the torture commits, and a few miscellaneous commits), but I will
> likely end up with several short branches and one huge one.  My thought
> is to keep the long branch, but email the patches out in a few separate
> serieses, with each depending on its predecessor.  For example, one series
> from the big branch would be folding the ->gpnum and ->completed fields
> into a single ->gp_seq, which helps the RCU-flavor consolidation task.
> Another series suppresses some rare false-positive splats that have been
> plaguing me for more than a year.  Yet another series within this huge
> branch applies and optimizes funnel locking for grace-period startup.
> 
> The problem is that the conversion to ->gp_seq has a very large footprint,
> which of course generates lots of conflicts.  I could of course collapse
> these commits into a single commit, but if I did that I would also defer
> to the merge window following v4.19 due to the resulting loss of bisection
> within that change.
> 
> Any advice?
> 
> The commits are for-mingo..rcu/dev in my -rcu tree.

I don't see these branches (and I don't pull tags).

How bad are the conflicts? Or is it too late to respond to help (sorry,
was on vacation :-)

-- Steve

Re: RCU branching for the v4.19 merge window

[Paul E. McKenney]

On Tue, May 22, 2018 at 11:27:36AM -0400, Steven Rostedt wrote:
> On Thu, 17 May 2018 07:40:44 -0700
> "Paul E. McKenney" <paulmck@linux.vnet.ibm.com> wrote:
> 
> > Hello, Steve!
> > 
> > Another year, another difficult-to-branch set of RCU commits.
> > 
> > In happy contrast to last year, I can make some branches (SRCU, some
> > of the torture commits, and a few miscellaneous commits), but I will
> > likely end up with several short branches and one huge one.  My thought
> > is to keep the long branch, but email the patches out in a few separate
> > serieses, with each depending on its predecessor.  For example, one series
> > from the big branch would be folding the ->gpnum and ->completed fields
> > into a single ->gp_seq, which helps the RCU-flavor consolidation task.
> > Another series suppresses some rare false-positive splats that have been
> > plaguing me for more than a year.  Yet another series within this huge
> > branch applies and optimizes funnel locking for grace-period startup.
> > 
> > The problem is that the conversion to ->gp_seq has a very large footprint,
> > which of course generates lots of conflicts.  I could of course collapse
> > these commits into a single commit, but if I did that I would also defer
> > to the merge window following v4.19 due to the resulting loss of bisection
> > within that change.
> > 
> > Any advice?
> > 
> > The commits are for-mingo..rcu/dev in my -rcu tree.
> 
> I don't see these branches (and I don't pull tags).

You don't see them yet because I don't create them until after -rc1 time,
which is a few weeks out.  If you go far enough down from rcu/dev you
will see branches (about 90 commits down from HEAD), but these branches
are already in -tip for v4.18.

> How bad are the conflicts? Or is it too late to respond to help (sorry,
> was on vacation :-)

The conflicts are already causing me substantial hassles when rebasing
bug fixes back to the buggy commits, so the conflicts are non-trivial.
Hence my reaching out to you, given your discomfort with last year's
long-chain RCU submission.

And you do have some time to respond.  ;-)

							Thanx, Paul