Re: [PATCH v4 6/6] scsi: core: Improve IOPS in case of host-wide tags

[Bart Van Assche <bart.vanassche@gmail.com>]

On 12/19/25 4:13 PM, Damien Le Moal wrote:
> E.g.: "host->can_queue is identical to the maximum queue depth per
> logical unit" -> As I mentioned, SCSI does not define/advertize a maximum queue
> depth per LU (beside the transport defined maximum of course). So Is this
> something that UFS defines outside of SCSI/SBC ?

No, this is something that is supported since a long time by the Linux
kernel. scsi_alloc_sdev() uses host->cmd_per_lun when allocating the
SCSI device budget map. Hence, host->cmd_per_lun is the maximum queue
depth for a SCSI device. This limit is enforced since a very long time.
Before the budget map was introduced, the number of commands per SCSI
device was set as follows:

        scsi_change_queue_depth(sdev, sdev->host->cmd_per_lun ?: 1);

> Also, for UFS, is it always one
> host per LU ? (that would be odd, the "device" here should be the host and you
> say it can have multiple LUs).

No. There is one SCSI host per UFS device and there can be multiple
logical units per UFS device.

> But if I understand this correctly, you are saying that a UFS device is like
> SATA and can_queue == device max queue depth, so we are always guaranteed that
> if you can allocate a tag, you will be able to issue the command, right ?

That's correct.

Thanks,

Bart.