Re: [Qemu-devel] [PATCH 00/16] SCSI sense and target request overhaul

[Anthony Liguori <anthony@codemonkey.ws>]

On 08/03/2011 03:49 AM, Paolo Bonzini wrote:
> This is a pretty important step in the modernization and improvement
> of the SCSI backends, and a prerequisite for pretty much everything
> that is on the table: migration, addressing, improved CD-ROM support,
> hotplug.

Applied all.  Thanks.

It fuzzed a bit, it seems pretty trivial to me but wouldn't hurt to 
double check.  I tested SCSI quite a bit afterwards too.

Regards,

Anthony Liguori

>
> The series touches two main parts:
>
> 1) sense data.  Autosense is made a first-class citizen of the
> subsystem.  Sense data is communicated to the target primarily
> via autosense, and it is then the target that uses this information
> to reply to REQUEST SENSE commands.  In addition, each LUN or
> target can be in a unit attention condition, which will be
> communicated via either autosense or REQUEST SENSE, or cleared
> automatically.
>
> 2) target requests.  The common example of this is REPORT LUNS
> and commands sent to invalid LUNs.  Handling of these commands
> previously had to be done for every SCSIDevice.  This series splits
> request-related SCSIDevice callbacks out into a new SCSIReqOps
> structure, thus allowing the target (SCSIBus) to hijack those requests
> and reply to them without involving the SCSIDevice at all.  The
> system is quite flexible, and is also used for invalid commands,
> REQUEST SENSE and commands sent while a unit attention
> condition is pending.
>
> Finally, other parts of the SCSIDevice code are moved to common
> code including general parsing of requests.
>
> The two parts are slightly intertwined.  Sense data is handled by patches
> 2/3/4/12/13/14/15, and target requests are handled by patches 5 to 11.
> (patches 1 and 16 are somewhat unrelated).  The reason for this is that
> the current device-specific handling of sense data conflicts heavily
> with the goal of handling some commands in a device-independent way.
> So, the series starts by moving sense handling to generic code, and
> comes back to generic handling of REQUEST SENSE and unit attention after
> implementing a few others device-independent requests.
>
> It conflicts somewhat with Markus's tray series, but not in a
> fatal way; the conflicts are trivial.
>
> Paolo Bonzini (16):
>    scsi-disk: no need to call scsi_req_data on a short read
>    vscsi: always use get_sense
>    scsi: pass status when completing
>    scsi: move sense handling to generic code
>    scsi: introduce SCSIReqOps
>    scsi: move request-related callbacks from SCSIDeviceInfo to SCSIReqOps
>    scsi: pass cdb already to scsi_req_new
>    scsi: introduce SCSICommand
>    scsi: push lun field to SCSIDevice
>    scsi: move request parsing to common code
>    scsi: move handling of REPORT LUNS and invalid LUNs to common code
>    scsi: move handling of REQUEST SENSE to common code
>    scsi: add a bunch more common sense codes
>    scsi: add support for unit attention conditions
>    scsi: report unit attention on reset
>    scsi: add special traces for common commands
>
>   hw/esp.c          |    4 +-
>   hw/lsi53c895a.c   |    4 +-
>   hw/scsi-bus.c     |  578 +++++++++++++++++++++++++++++++++++++++++++++--------
>   hw/scsi-defs.h    |    4 +
>   hw/scsi-disk.c    |  164 +++++-----------
>   hw/scsi-generic.c |  156 ++++-----------
>   hw/scsi.h         |   72 +++++--
>   hw/spapr_vscsi.c  |   95 ++-------
>   hw/usb-msd.c      |    4 +-
>   trace-events      |    5 +
>   10 files changed, 668 insertions(+), 418 deletions(-)
>