[PATCH v5 15/70] kernel/fork: Use maple tree for dup_mmap() during forking

[Liam Howlett]

* Mark Hemment <markhemm at googlemail.com> [220203 07:00]:
> On Wed, 2 Feb 2022 at 03:23, Liam Howlett <liam.howlett at oracle.com> wrote:
> >
> > From: "Liam R. Howlett" <Liam.Howlett at Oracle.com>
> >
> > The maple tree was already tracking VMAs in this function by an earlier
> > commit, but the rbtree iterator was being used to iterate the list.
> > Change the iterator to use a maple tree native iterator and switch to
> > the maple tree advanced API to avoid multiple walks of the tree during
> > insert operations.  Unexport the now-unused vma_store() function.
> >
> > We track whether we need to free the VMAs and tree nodes through RCU
> > (ie whether there have been multiple threads that can see the mm_struct
> > simultaneously; by pthread(), ptrace() or looking at /proc/$pid/maps).
> > This setting is sticky because it's too tricky to decide when it's safe
> > to exit RCU mode.
> >
> > For performance reasons we bulk allocate the maple tree nodes.  The node
> > calculations are done internally to the tree and use the VMA count and
> > assume the worst-case node requirements.  The VM_DONT_COPY flag does
> > not allow for the most efficient copy method of the tree and so a bulk
> > loading algorithm is used.
> >
> > Signed-off-by: Matthew Wilcox (Oracle) <willy at infradead.org>
> > Signed-off-by: Liam R. Howlett <Liam.Howlett at Oracle.com>
> > Acked-by: Vlastimil Babka <vbabka at suse.cz>
> > ---
> >  include/linux/mm.h       |  2 --
> >  include/linux/sched/mm.h | 13 +++++++++++++
> >  kernel/fork.c            | 35 +++++++++++++++++++++++++++++------
> >  3 files changed, 42 insertions(+), 8 deletions(-)
> ....
> > diff --git a/kernel/fork.c b/kernel/fork.c
> > index 51a7971651ef..8ea683fcefcd 100644
> > --- a/kernel/fork.c
> > +++ b/kernel/fork.c
> > @@ -377,6 +377,16 @@ void vm_area_free(struct vm_area_struct *vma)
> >         kmem_cache_free(vm_area_cachep, vma);
> >  }
> >
> > +void mm_set_in_rcu(struct mm_struct *mm)
> > +{
> > +       if (!mt_in_rcu(&mm->mm_mt))
> > +               return;
> > +       mmap_write_lock(mm);
> > +       mm->mm_mt.ma_flags |= MT_FLAGS_USE_RCU;
> > +       mmap_write_unlock(mm);
> > +}
> > +EXPORT_SYMBOL(mm_set_in_rcu);
> 
> The mt_in_rcu() test looks wrong (inverted).
> mt_in_rcu() returns true only when MT_FLAGS_USE_RCU is set, so this
> flag will never set here.
> All callers of mm_set_in_rcu() check mt_in_rcu() so the test here
> could be removed.

Yes, you are correct.  Thanks.  I will rerun my benchmarks with this
change.

It's not a critical thing at this point since all the vma work is still
under the mmap_lock.  But it could affect performance.