Re: [SCSI] Add support for braindead Cypress USB ATA passthrough CDBs

[James Bottomley <James.Bottomley@SteelEye.com>]

On Fri, 2005-12-23 at 12:41 -0500, Jeff Garzik wrote:
> James Bottomley wrote:
> > 1) Simply don't mangle the LUN for SCSI_UNKNOWN and then have all the
> > subsystems lying about SCSI_2 compliance instead set their mangled level
> > to SCSI_UNKNOWN (which seems to be more truthful)
> 
> It is specified in MMC that ATAPI or USB may be SCSI level to zero. 
> Thus "lying" is not really accurate at all.  We need to update SCSI to 
> handle these devices as specified.

Isn't that what I proposed?

But it only fixes the devices whose SCSI level is correctly passed on to
the mid-layer.  There's still the problem of USB and other subsystems
mangling the level back to SCSI_2 which causes the LUN to be placed in
cdb[1] as per spec.  

> This is why I snoop INQUIRY for ATAPI devices in libata-scsi, and force 
> the SCSI version to 0x05, if SCSI version is returned from the device as 
> zero.  It is conditional because -- due to the wonderful world of 
> hardware -- some ATAPI devices (such as SCSI devices plugged into an 
> SPI<->ATA bridge) and some USB devices correctly report SCSI version.

Well, that's SCSI3 (SPC-3), which is fine, but does mean we try
REPORT_LUNS scanning.  The problem in other subsystems is that some
devices hang when they see a REPORT_LUNS command.

What problems do you see returning zero (apart from cdb[1] changes)?

James