> > diff --git a/drivers/nvdimm/region_devs.c b/drivers/nvdimm/region_devs.c
> > index 63cb01ef4ef0..0eecc8670f80 100644
> > --- a/drivers/nvdimm/region_devs.c
> > +++ b/drivers/nvdimm/region_devs.c
> > @@ -926,18 +926,15 @@ int nd_blk_region_init(struct nd_region *nd_region)
> > unsigned int nd_region_acquire_lane(struct nd_region *nd_region) {
> > unsigned int cpu, lane;
> > + struct nd_percpu_lane *ndl_lock, *ndl_count;
> >
> > - cpu = get_cpu();
> > - if (nd_region->num_lanes < nr_cpu_ids) {
> > - struct nd_percpu_lane *ndl_lock, *ndl_count;
> > + cpu = get_cpu_light();
>
> what is wrong with using raw_smp_processor_id() instead of get_cpu()? In
> case the schedulers pushes the task to another CPU the it will access
> cross-CPU data which is locked.
>
> >
> > - lane = cpu % nd_region->num_lanes;
> > - ndl_count = per_cpu_ptr(nd_region->lane, cpu);
> > - ndl_lock = per_cpu_ptr(nd_region->lane, lane);
> > - if (ndl_count->count++ == 0)
Wouldn't there is chance of race with both get_cpu_light() & raw_smp_processor_id()?
Maybe I am wrong.
Thanks,
Pankaj
> > - spin_lock(&ndl_lock->lock);
> > - } else
> > - lane = cpu;
> > + lane = cpu % nd_region->num_lanes;
> > + ndl_count = per_cpu_ptr(nd_region->lane, cpu);
> > + ndl_lock = per_cpu_ptr(nd_region->lane, lane);
> > + if (ndl_count->count++ == 0)
> > + spin_lock(&ndl_lock->lock);
> >
> > return lane;
> > }
>
> Sebastian
>
