This message is from the T13 list server.

I'm not sure what all the fuss is about redundant transfer-direction
informaiton. SCSI op-codes are partitioned so that a given op-code
corresponds either to a transfer-in, or a transfer-out (bidirectional
and non-transfer commands excepted). There is no SCSI opcode that
sometimes does a transfer-in, and sometimes does a transfer-out,
depending on other information. Nonetheless, all SCSI HBA driver
interfaces provide in a "wrapper" of some sort, and indication of the
direction of the data transfer, even though that information is already
evident in the op-code. Why? Because the HBA doesn't want to be burdened
with parsing the CDB itself.

The very same concept is at play here - with respect to the embedded ATA
command. Only in this case, the SCSI command itself is the "wrapper" for
the embedded ATA command, and so the SCSI CDB should contain an
indication of the transfer direction.

Now - to be fully consistent with existing SCSI models, we recognize
that the SCSI/ATA passthrough CDB has a SCSI op-code. All SCSI op-codes
defined so far indicate a transfer direction as part of the opcode (if
there is a transfer at all). That being the case, it might be prudent to
define distinct SCSI op-codes for transfer-in ATA passthrough commands
and transfer-out ATA passthrough commands. That, apart from being
consistent with the definition of existing SCSI opcodes, frees up
another bit in the CDB to use for some other purpose.

Regards,
Bob

-----Original Message-----
From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Andy
Warner
Sent: Thursday, October 28, 2004 7:39 PM
To: Curtis Stevens
Cc: Andy Warner; [EMAIL PROTECTED]
Subject: Re: [t13] RE: comment on T10 ATA-passthru

This message is from the T13 list server.


Curtis Stevens wrote:
> Andy
> 
>       I am considering removing UDMA entirely since no command
actually
> calls out UDMA as it's protocol. 

You won't hear a peep of complaint from me. I (linux) can't
use it because I'm on the other side of the controller from
any DMA/UDMA distinction. It all looks the same for me.

I don't see how Linux would use Bus Idle for the
same reason. Feel free to chop that if you feel the
urge.

I'd still like to remove the duplicate direction
indications, though. Next week I plan on updating
the linux version, so let me know if you publish
r6 and I'll snap straight to that.

Is the magic number assignment on the agenda for
November's T10 meeting ?
-- 
[EMAIL PROTECTED]

Andy Warner             Voice: (612) 801-8549   Fax: (208) 575-5634

Reply via email to