Re: Should queue_max_hw_sectors() always be <= max_dev_sectors?

[Jens Axboe <axboe@kernel.dk>]

On 08/30/2016 10:54 AM, Andy Grover wrote:
> Hi Martin, (you're the last person to touch this? :)
>
> If I create a local SCSI device using LIO's loopback fabric, and backed
> by a file, /sys/block/sda/queue/max_hw_sectors_kb reports 32767. This
> reflects the loopback scsi_host_template reporting 65536 max_sectors,
> but that's the "controller" -- the actual device reports via the Block
> Limits VPD of supporting a max size of 16384 512-byte blocks.
>
> Therefore, using max_hw_sectors_kb will give a wrong impression for what
> the max data size that can be issued to the device is -- 32767KiB
> instead of 8192KiB. Should max_hw_sectors_kb be reported as the lesser
> of max_hw_sectors and max_dev_sectors? Something like:
>
> --- a/include/linux/blkdev.h
> +++ b/include/linux/blkdev.h
> @@ -1196,7 +1196,7 @@ static inline unsigned int
> queue_max_sectors(struct request_queue *q)
>
>  static inline unsigned int queue_max_hw_sectors(struct request_queue *q)
>  {
> -       return q->limits.max_hw_sectors;
> +       return min_not_zero(q->limits.max_hw_sectors,
> q->limits.max_dev_sectors);
>  }
>
> ...because limits.max_hw_sectors internally seems to be just the
> controller limit, but what users of queue_max_hw_sectors (including
> sysfs) really want is the lesser of limits.max_hw_sectors and
> limits.max_dev_sectors.

It should be the maximum supported to the device, which is the minimum
of any limitations in that path. So I'd say your change is correct.


-- 
Jens Axboe