Re: [patch] libata: add ioctls to support SMART

[Jeff Garzik <jgarzik@pobox.com>]

John W. Linville wrote:
> Support for HDIO_DRIVE_CMD and HDIO_DRIVE_TASK in libata.  Useful for
> supporting SMART w/ unmodified smartctl and smartd userland binaries.

First let me say that it's a damn fine first attempt, and people should 
be able to use this for SMART until support is merged officially.


> Not happy w/ loop after failed ata_qc_new_init(), but needed because smartctl
> and smartd did not retry after failure.  Likely need an option to wait for
> available qc?  Also not sure all the error return codes are correct...

I'd like to implement it a bit differently, and I think this different 
method will solve some of the open questions you have.

Take a look at http://www.t10.org/ftp/t10/document.04/04-260r2.pdf

I would like to implement HDIO_DRIVE_CMD and HDIO_DRIVE_TASK completely 
inside libata-scsi.c.  These ioctls should translate the ioctl arguments 
into an ATA-passthru SCSI command, and use the standard "issue a scsi 
command" kernel API to submit the command and wait for a result.

That implies, then, that you would add code to libata-scsi.c that 
translates the ATA-passthru SCSI command into an ATA command using the 
ata_scsi_translate() infrastructure.

Note that you'll need to make up a SCSI opcode, inside the SCSI 
vendor-specific opcode space, since the ATA-passthru hasn't yet been 
assigned an official SCSI opcode.  SPC-3 
(http://www.t10.org/ftp/t10/drafts/spc3/spc3r20a.pdf) lists the 
available opcodes in section C.3, denoted with a 'V' across all columns.

Once libata-scsi.c can handle the ATA-passthru SCSI command, 
implementing HDIO_DRIVE_{TASK,CMD} should be quite trivial.

And SMART support will be complete :)

	Jeff