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

[Yevgeny Petrilin]

Roland Dreier wrote:
>  > +#include <linux/random.h>
> 
> this isn't needed, is it?
Correct, I should have removed it

> 
>  > +		queue_delayed_work(sense->sense_wq , &sense->sense_poll,
>  > +				   round_jiffies(MLX4_SENSE_RANGE));
> 
> should be round_jiffies_relative, right?
I did not find critical difference between the two:
unsigned long round_jiffies(unsigned long j)
{
	return round_jiffies_common(j, raw_smp_processor_id(), false);
}

unsigned long __round_jiffies_relative(unsigned long j, int cpu)
{
	unsigned long j0 = jiffies;

	/* Use j0 because jiffies might change while we run */
	return round_jiffies_common(j + j0, cpu, false) - j0;
}

>
>  > +	if (sense->resched)
>
> Do we need resched?  Can't we just use cancel_delayed_work_sync() when
> we want to stop this from running?
> 
>  > +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");
> 
You are right about this, and I am using cancel_delayed_work_sync() in mlx4_sense_cleanup().
I was just using the resched flag as an indicator to rearm or not rearm the task, I guess it is redundant
and just should use cancel_delayed_work_sync() in mlx4_stop_sense();

> Do we really another work queue, or can we share one queue for the
> catastrophic error and port sensing work?
There is no special reason for having another queue other then convenience.
The catastrophic work queue is defined in catas.c


Thanks,
Yevgeny