[T10] ATA PASS-THROUGH(32) straw poll

Knight, Frederick Frederick.Knight at netapp.com
Fri Nov 20 15:04:06 PST 2015


What I heard at the meeting was that the original proposed approach was SATA FIS friendly.  It had been modeled after the existing PASS THROUGH command which was also supposedly FIS friendly.

Did anyone verify that?

                Fred

From: t10-bounces at t10.org [mailto:t10-bounces at t10.org] On Behalf Of Gerry Houlder
Sent: Tuesday, November 17, 2015 11:43 AM
Cc: t10 at t10.org; terry.denney at hpe.com; nathan seiler
Subject: Re: [T10] ATA PASS-THROUGH(32) straw poll

Just my opinion, but I don't see anything that is "SCSI friendly" about the second choice.

  *   The LBA bytes don't land within bytes 14-19, where the lower 6 bytes of LBA are for 32 byte commands.
  *   The Count field doesn't land in bytes 28-31, where the 4 byte transfer length is for 32 byte commands.
  *   Most  of the rest of the ATA fields don't have a SCSI equivalent.
Any "SCSI friendly" arrangement would seem to be completely arbitrary and therefore not very helpful. This leads me to think that the third option (ATA friendly) makes the most sense.

On Mon, Nov 16, 2015 at 10:21 PM, Neil Wanamaker <ntw20 at earthlink.net<mailto:ntw20 at earthlink.net>> wrote:
John,

I must agree that your presentation makes it much easier to follow the differences.

I should likely have renamed Control(7:0) to be Device Control(7:0) to disambiguate it..

Thanks,
Neil


-----Original Message-----
>From: "John Geldman (jgeldman)" <jgeldman at micron.com<mailto:jgeldman at micron.com>>
>Sent: Nov 16, 2015 8:01 PM
>To: Neil Wanamaker <ntw20 at earthlink.net<mailto:ntw20 at earthlink.net>>, "t10 at t10.org<mailto:t10 at t10.org>" <t10 at t10.org<mailto:t10 at t10.org>>
>Cc: nathan seiler <Nathan.Seiler at InsightGlobal.net<mailto:Nathan.Seiler at InsightGlobal.net>>, "terry.denney at hpe.com<mailto:terry.denney at hpe.com>" <terry.denney at hpe.com<mailto:terry.denney at hpe.com>>
>Subject: RE: [T10] ATA PASS-THROUGH(32) straw poll
>
>Neil,
>
>It may be late and I may be intellectually lazy tonight, but that is the most convoluted presentation of material that I've seen in a while. It takes more effort than I can muster this evening to visualize each two page table and mentally find the differences.
>
>The attached spreadsheet is an easier way to look. (Now I see the SCSI-friendly version has the fields in contiguous order, while the SATA-friendly version has the fields in all master/all slave order).
>
>The attached does also reveal that your 2nd and 3rd table have a Control (7:0) field that doesn't exist in the first table.
>
>While I fixed inconsistent names, explaining the 'odd' Control field is up to you (I think it is a duplicate of the SCSI field).
>
>Grumpily yours,
>John
>
>
>-----Original Message-----
>From: t10-bounces at t10.org<mailto:t10-bounces at t10.org> [mailto:t10-bounces at t10.org<mailto:t10-bounces at t10.org>] On Behalf Of Neil Wanamaker
>Sent: Monday, November 16, 2015 6:57 PM
>To: t10 at t10.org<mailto:t10 at t10.org>
>Cc: nathan seiler; terry.denney at hpe.com<mailto:terry.denney at hpe.com>
>Subject: [T10] ATA PASS-THROUGH(32) straw poll
>
>All:
>
>At the recent T10 CAP meeting, Gerry Houlder presented 15-269r0, which included a new ATA PASS-THROUGH(32) command. In his proposal, he made the CDB structure consistent with the form of the ATA PASS-THROUGH(16), adding AUX and ICC at the end.
>
>In general, I applaud consistency, but... the ATA PASS-THROUGH(16) command seems to have few redeeming features - it is neither convenient for the SCSI initiator or for the SATL, what with the LBAs being scrambled, etc.
>
>To make the new command convenient for either the initiator or the SATL, I would like feedback on which of the three formats in the attachment would be preferable.
>
>The first is Gerry's proposal unchanged apart from using a service action of the form xx00, as the SPC editor suggests reserving a block of service actions (such as 3000h-37ffh)for SAT, as shown in Table E.8 of SPC-5.
>
>The second is a SCSI-friendly structure, putting LBA and AUX in big-endian form aligned to appropriate boundaries. I have added the CONTROL(7:0) field (which could be added to the first proposal, if people wanted to go that route.
>
>The third is a SAT (or ATA if you prefer) - friendly structure, which may be familiar to some of you.
>
>I would appreciate feedback from the audience as to which of these best meets your needs.
>
>Best regards,
>Neil Wanamaker
_______________________________________________
T10 mailing list
T10 at t10.org<mailto:T10 at t10.org>
https://urldefense.proofpoint.com/v2/url?u=http-3A__www.t10.org_mailman_listinfo_t10&d=CwICAg&c=IGDlg0lD0b-nebmJJ0Kp8A&r=TxI1DC4HavpWBdSmUqvdNvSwgOklhaW328zLt5AOpPM&m=kDiQzQG_DdEmmFm5EOXjyHs0hVuNkK_MWKoCQhfj7Kc&s=3zkQdT50a01kbTZT0dpwSSKBCg3GMS8_MO99TRN3Cn4&e=

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.t10.org/pipermail/t10/attachments/20151120/07150126/attachment.html>


More information about the T10 mailing list