[ofa-general][PATCH 1/2] mlx4_core: Link sensing support

[Roland Dreier]

 >  mlx4_core-y :=	alloc.o catas.o cmd.o cq.o eq.o fw.o icm.o intf.o main.o mcg.o \
 > -		mr.o pd.o port.o profile.o qp.o reset.o srq.o
 > +		mr.o pd.o port.o profile.o qp.o reset.o srq.o sense.o

trivial but if this is going to be respun, please keep the filenames in
alphabetical order -- ie sense.o goes before srq.o

 > +#include <linux/random.h>

this isn't needed, is it?

 > +	if (sense->resched)

Do we need resched?  Can't we just use cancel_delayed_work_sync() when
we want to stop this from running?

 > +		queue_delayed_work(sense->sense_wq , &sense->sense_poll,
 > +				   round_jiffies(MLX4_SENSE_RANGE));

should be round_jiffies_relative, right?

 > +void mlx4_stop_sense(struct mlx4_dev *dev)
 > +{
 > +	mlx4_priv(dev)->sense.resched = 0;
 > +}

This doesn't stop anything... we need cancel_delayed_work_sync().

 > +	sense->sense_wq = create_singlethread_workqueue("mlx4_sense");

Do we really another work queue, or can we share one queue for the
catastrophic error and port sensing work?

 - R.