Question on NVMe SCSI Translation Ref 1.5

Question on NVMe SCSI Translation Ref 1.5

[Jeffrey Lien]

In the SCSI Translation Ref 1.5 spec, section 4.5 Reports LUNs, the actions for the different Select Report fields are defined.   For value 11h, it describes what should be done when the LUN addressed in the command is for LUN 0 vs not.  My question is how do we know which LUN is being addressed?   I didn't see anything in the CDB or the sg_io_hdr that specified  a LUN.   Does anyone know how to clarify this?   

Jeff Lien

Jeff.Lien at hgst.com
507 322-2416/23-2416
3605 Hwy 52 N.  
Rochester, MN 55901

Question on NVMe SCSI Translation Ref 1.5

[Christoph Hellwig]

Hi Jeffrey,

SG_IO is sent to a device node.  For actualy SCSI devices you'll
get the LUN from sdev->lun.  For NVMe there obviously aren't any LUNs,
so you'll have to come up with a tranlation.  The current Linux driver
translated namespaces 1 to LUN 0 and so on.  In general sending a
REPORT LUNS to a NVMe device node does not seem very useful, and I have
no idea why it even was specified.  Then again the whole SCSI
translation layer concept seems incredbile braindead and I think we
should at least optionally remove it from the NVMe driver.

Question on NVMe SCSI Translation Ref 1.5

[Robles, Raymond C]

Hi Chris/Jeff,

The SCSI to NVMe translation spec was written as part of the NVMe working group initiative to enable broad adoption of NVMe in the ecosystem. There are other OS environments besides Linux (i.e. VMware, Windows, etc.) where the NVMe driver resides within a SCSI storage stack and therefore receives native OS SCSI commands. For Linux, agreed it does not have much usage considering where the NVMe driver sites in the Linux kernel. However, the NVMe spec or translation spec was not written just for Linux. 

Thanks,
Ray

-----Original Message-----
From: Linux-nvme [mailto:linux-nvme-bounces@lists.infradead.org] On Behalf Of Christoph Hellwig
Sent: Friday, November 20, 2015 11:32 AM
To: Jeffrey Lien
Cc: linux-nvme at lists.infradead.org
Subject: Re: Question on NVMe SCSI Translation Ref 1.5

Hi Jeffrey,

SG_IO is sent to a device node.  For actualy SCSI devices you'll get the LUN from sdev->lun.  For NVMe there obviously aren't any LUNs, so you'll have to come up with a tranlation.  The current Linux driver translated namespaces 1 to LUN 0 and so on.  In general sending a REPORT LUNS to a NVMe device node does not seem very useful, and I have no idea why it even was specified.  Then again the whole SCSI translation layer concept seems incredbile braindead and I think we should at least optionally remove it from the NVMe driver.

_______________________________________________
Linux-nvme mailing list
Linux-nvme at lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-nvme

Question on NVMe SCSI Translation Ref 1.5

['Christoph Hellwig']

Hi Raymond,

I'm not complaining about the spec - I think it's clearly useful
as a reference.  For example it allows my pNFS SCSI layout spec to be
used with NVMe by providing a reference mapping, even if we never even
send SCSI commands in that case.  I just think the implementing it in
the Linux kernel is useless as we provide a rich block I/O abstraction.