On Tue, Aug 01, 2017 at 04:25:56AM +0000, Mathieu Desnoyers wrote:
> ----- On Aug 1, 2017, at 12:03 AM, Paul E. McKenney
> paul...@linux.vnet.ibm.com wrote:
>
> > On Tue, Aug 01, 2017 at 12:04:05AM +0000, Mathieu Desnoyers wrote:
> >> ----- On Jul 31, 2017, at 12:13 PM, Paul E. McKenney
> >> paul...@linux.vnet.ibm.com
> >> wrote:
> >>
> >> > On Mon, Jul 31, 2017 at 01:50:29PM +1000, Stephen Rothwell wrote:
> >> >> Hi Paul,
> >> >>
> >> >> Today's linux-next merge of the rcu tree got a conflict in:
> >> >>
> >> >> arch/x86/mm/tlb.c
> >> >>
> >> >> between commit:
> >> >>
> >> >> 94b1b03b519b ("x86/mm: Rework lazy TLB mode and TLB freshness
> >> >> tracking")
> >> >>
> >> >> from the tip tree and commit:
> >> >>
> >> >> d7713e8f8b23 ("membarrier: Expedited private command")
> >> >>
> >> >> from the rcu tree.
> >> >>
> >> >> I fixed it up (the former removed the comment and the load_cr3(), so I
> >> >> just dropped the commend change in the latter) and can carry the fix as
> >> >> necessary. This is now fixed as far as linux-next is concerned, but any
> >> >> non trivial conflicts should be mentioned to your upstream maintainer
> >> >> when your tree is submitted for merging. You may also want to consider
> >> >> cooperating with the maintainer of the conflicting tree to minimise any
> >> >> particularly complex conflicts.
> >> >
> >> > Thank you, Stephen!
> >> >
> >> > Mathieu, Peter, our commit log reads as if removal of load_cr3() would
> >> > simply result in relying on the ordering provided by the atomic ops
> >> > in switch_mm() for mm_cpumask(), so that only the commit log and the
> >> > comment need changing.
> >> >
> >> > Please let me know if I am missing something here.
> >>
> >> I think you are right. Both load_cr3() and mm_cpumask update operations
> >> (LOCK prefixed) provide the appropriate barriers on x86. So it's just a
> >> matter of adapting the comment to the new reality.
> >
> > Like this?
>
> The updated comment in the commit message looks good, but I would be
> tempted to add a comment in x86 switch_mm_irqs_off() stating the
> following requirement just before the line invoking cpumask_set_cpu():
>
> /*
> * The full memory barrier implied by mm_cpumask update operations
> * is required by the membarrier system call.
> */
This looks good to me, but I will give the discussion another day or
so to settle out. ;-)
Thanx, Paul
