Re: [PATCH 1/2] block: add max_open_zones to blk-sysfs

[Niklas Cassel]

On Tue, Jun 30, 2020 at 01:49:41AM +0000, Damien Le Moal wrote:
> On 2020/06/16 19:28, Niklas Cassel wrote:
> > diff --git a/drivers/nvme/host/zns.c b/drivers/nvme/host/zns.c
> > index c08f6281b614..af156529f3b6 100644
> > --- a/drivers/nvme/host/zns.c
> > +++ b/drivers/nvme/host/zns.c
> > @@ -82,6 +82,7 @@ int nvme_update_zone_info(struct gendisk *disk, struct nvme_ns *ns,
> >  
> >  	q->limits.zoned = BLK_ZONED_HM;
> >  	blk_queue_flag_set(QUEUE_FLAG_ZONE_RESETALL, q);
> > +	blk_queue_max_open_zones(q, le32_to_cpu(id->mor) + 1);
> >  free_data:
> >  	kfree(id);
> >  	return status;
> > diff --git a/drivers/scsi/sd_zbc.c b/drivers/scsi/sd_zbc.c
> > index 183a20720da9..aa3564139b40 100644
> > --- a/drivers/scsi/sd_zbc.c
> > +++ b/drivers/scsi/sd_zbc.c
> > @@ -717,6 +717,10 @@ int sd_zbc_read_zones(struct scsi_disk *sdkp, unsigned char *buf)
> >  	/* The drive satisfies the kernel restrictions: set it up */
> >  	blk_queue_flag_set(QUEUE_FLAG_ZONE_RESETALL, q);
> >  	blk_queue_required_elevator_features(q, ELEVATOR_F_ZBD_SEQ_WRITE);
> > +	if (sdkp->zones_max_open == U32_MAX)
> > +		blk_queue_max_open_zones(q, 0);
> > +	else
> > +		blk_queue_max_open_zones(q, sdkp->zones_max_open);
> 
> This is correct only for host-managed drives. Host-aware models define the
> "OPTIMAL NUMBER OF OPEN SEQUENTIAL WRITE PREFERRED ZONES" instead of a maximum
> number of open sequential write required zones.
> 
> Since the standard does not actually explicitly define what the value of the
> maximum number of open sequential write required zones should be for a
> host-aware drive, I would suggest to always have the max_open_zones value set to
> 0 for host-aware disks.

Isn't this already the case?

At least according to the comments:

https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/scsi/sd_zbc.c?h=v5.8-rc3#n555

https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/scsi/sd_zbc.c?h=v5.8-rc3#n561

We seem to set

sdkp->zones_max_open = 0;

for host-aware, and

sdkp->zones_max_open = get_unaligned_be32(&buf[16]);

for host-managed.

So the blk_queue_max_open_zones(q, sdkp->zones_max_open) call in
sd_zbc_read_zones() should already export this new sysfs property
as 0 for host-aware disks.


Kind regards,
Niklas

> 
> >  	nr_zones = round_up(sdkp->capacity, zone_blocks) >> ilog2(zone_blocks);
> >  
> >  	/* READ16/WRITE16 is mandatory for ZBC disks */
> > diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
> > index 8fd900998b4e..2f332f00501d 100644
> > --- a/include/linux/blkdev.h
> > +++ b/include/linux/blkdev.h
> > @@ -520,6 +520,7 @@ struct request_queue {
> >  	unsigned int		nr_zones;
> >  	unsigned long		*conv_zones_bitmap;
> >  	unsigned long		*seq_zones_wlock;
> > +	unsigned int		max_open_zones;
> >  #endif /* CONFIG_BLK_DEV_ZONED */
> >  
> >  	/*
> > @@ -729,6 +730,17 @@ static inline bool blk_queue_zone_is_seq(struct request_queue *q,
> >  		return true;
> >  	return !test_bit(blk_queue_zone_no(q, sector), q->conv_zones_bitmap);
> >  }
> > +
> > +static inline void blk_queue_max_open_zones(struct request_queue *q,
> > +		unsigned int max_open_zones)
> > +{
> > +	q->max_open_zones = max_open_zones;
> > +}
> > +
> > +static inline unsigned int queue_max_open_zones(const struct request_queue *q)
> > +{
> > +	return q->max_open_zones;
> > +}
> >  #else /* CONFIG_BLK_DEV_ZONED */
> >  static inline unsigned int blk_queue_nr_zones(struct request_queue *q)
> >  {
> > @@ -744,6 +756,14 @@ static inline unsigned int blk_queue_zone_no(struct request_queue *q,
> >  {
> >  	return 0;
> >  }
> > +static inline void blk_queue_max_open_zones(struct request_queue *q,
> > +		unsigned int max_open_zones)
> > +{
> > +}
> 
> Why is this one necessary ? For the !CONFIG_BLK_DEV_ZONED case, no driver should
> ever call this function.

Will remove in v2.