Re: [PATCH 12/13] block: introduce blk-iolatency io controller

[Tejun Heo]

Hello,

Just interface nits.

On Tue, May 29, 2018 at 05:17:23PM -0400, Josef Bacik wrote:
...
> +static size_t iolatency_pd_stat(struct blkg_policy_data *pd, char *buf,
> +				size_t size)
> +{
> +	struct iolatency_grp *iolat = pd_to_lat(pd);
> +	struct blkcg_gq *blkg = pd_to_blkg(pd);
> +	unsigned use_delay = atomic_read(&blkg->use_delay);
> +
> +	if (!iolat->min_lat_nsec)
> +		return 0;
> +
> +	return snprintf(buf, size,
> +			" depth=%u delay=%llu use_delay=%u total_lat_avg=%llu",
> +			iolat->rq_depth.max_depth,

Can we please use "max" as depth value when there is no restriction?

> +			(unsigned long long)(use_delay ?
> +				atomic64_read(&blkg->delay_nsec) /
> +				NSEC_PER_USEC : 0),
> +			use_delay,
> +			(unsigned long long)iolat->total_lat_avg /
> +			NSEC_PER_USEC);

and "avg_lat".

I'm a bit worried about exposing anything other than avg_lat given
that they're inherently implementation details.  I think it might be a
good idea to gate them behind a kernel boot param (which hopefully can
be turned on/off while the system is running).

Thanks.

-- 
tejun