On Mon, Sep 19, 2016 at 01:52:05PM +0200, Michal Hocko wrote:
> On Sun 18-09-16 13:29:43, Piotr Kwapulinski wrote:
> > The MPOL_F_STATIC_NODES and MPOL_F_RELATIVE_NODES flags are irrelevant
> > when setting them for MPOL_LOCAL NUMA memory policy via set_mempolicy.
> > Return the "invalid argument" from set_mempolicy whenever
> > any of these flags is passed along with MPOL_LOCAL.
> 
> man 2 set_mempolicy doesn't list this as invalid option. Maybe this is a
> documentation bug but is it possible that somebody will see this as an
> unexpected error?
> 
The MPOL_LOCAL is currently not documented in "man set_mempolicy(2)".
In case the nodemask is empty for MPOL_LOCAL it is transformed into 
MPOL_PREFERRED.
The motivation for disabling MPOL_F_STATIC_NODES and MPOL_F_RELATIVE_NODES
flags for MPOL_PREFERRED with empty nodemask is described at this commit
3e1f064562fcff7. Currently I call set_mempolicy(MPOL_LOCAL, ...) via the 
syscall()
but despite of that it is inconsistent with MPOL_PREFERRED.

> > It is consistent with MPOL_PREFERRED passed with empty nodemask.
> > It also slightly shortens the execution time in paths where these flags
> > are used e.g. when trying to rebind the NUMA nodes for changes in
> > cgroups cpuset mems (mpol_rebind_preferred()) or when just printing
> > the mempolicy structure (/proc/PID/numa_maps).
> 
> I am not sure I understand this argument. What does this patch actually
> fix? If this is about the execution time then why not just bail out
> early when MPOL_LOCAL && (MPOL_F_STATIC_NODES || MPOL_F_RELATIVE_NODES)
>
The mpol_new() performs additional checks on nodemask.

> > Isolated tests done.
> > 
> > Signed-off-by: Piotr Kwapulinski <kwapulinski.pi...@gmail.com>
> > ---
> >  mm/mempolicy.c | 4 +++-
> >  1 file changed, 3 insertions(+), 1 deletion(-)
> > 
> > diff --git a/mm/mempolicy.c b/mm/mempolicy.c
> > index 2da72a5..27b07d1 100644
> > --- a/mm/mempolicy.c
> > +++ b/mm/mempolicy.c
> > @@ -276,7 +276,9 @@ static struct mempolicy *mpol_new(unsigned short mode, 
> > unsigned short flags,
> >                             return ERR_PTR(-EINVAL);
> >             }
> >     } else if (mode == MPOL_LOCAL) {
> > -           if (!nodes_empty(*nodes))
> > +           if (!nodes_empty(*nodes) ||
> > +               (flags & MPOL_F_STATIC_NODES) ||
> > +               (flags & MPOL_F_RELATIVE_NODES))
> >                     return ERR_PTR(-EINVAL);
> >             mode = MPOL_PREFERRED;
> >     } else if (nodes_empty(*nodes))
> > -- 
> > 2.9.2
> 
> -- 
> Michal Hocko
> SUSE Labs

--
Piotr Kwapulinski

Reply via email to