On 12 August 2016 at 13:18, Tom Yan [off-list ref] wrote:

On 12 August 2016 at 10:01, Martin K. Petersen
[off-list ref] wrote:

quoted

Again, the point of max_hw_sectors and max_dev_sectors is to enforce the
hard limits of controller and device respectively. Nothing else.

Sounds like libata-scsi is doing something wrong then. It should not
set max_hw_sectors to dev->max_sectors that is set by libata-core:

quoted

blk_queue_max_hw_sectors(q, dev->max_sectors);

Though we'll still have to "abuse" max_hw_sectors for ATAPI class
devices, since neither the SATL or sr cares about VPD. The only
devices that need ATA_HORKAGE_MAX_SEC_LBA48 are ATAPI devices as well.

but instead it should report it as Maximum Transfer Length and let sd
set it as max_dev_sectors:

quoted

put_unaligned_be32(dev->max_sectors, &rbuf[8]);

While max_hw_sectors will be left untouched (in the case of AHCI, for
example, since its SCSI host template does not have max_sectors set;
so max_hw_sectors will be SCSI_DEFAULT_MAX_SECTORS).

Although that means setting dev->max_sectors to a value larger than
1024 will probably be a no-op, if that's really an issue, we should
have the host templates in libata updated.

Make sense?