Patchwork [PULL,v5,00/25] SCSI subsystem improvements

login
register
mail settings
Submitter Paolo Bonzini
Date May 26, 2011, 10:56 a.m.
Message ID <1306407411-4290-1-git-send-email-pbonzini@redhat.com>
Download mbox
Permalink /patch/97535/
State New
Headers show

Pull-request

git://github.com/bonzini/qemu.git scsi.2

Comments

Paolo Bonzini - May 26, 2011, 10:56 a.m.
The following changes since commit aa29141d84d58171c2d219f0a4b599bd76fb2e37:

  Merge remote-tracking branch 'kraxel/CVE-2011-1751' into staging (2011-05-25 07:04:13 -0500)

are available in the git repository at:

  git://github.com/bonzini/qemu.git scsi.2

This series includes the following improvements to the SCSI subsystem:

1) introduction of SCSIBusOps that generalize the existing
command_complete callback;

2) widespread use of the SCSIRequest abstraction, with simpler memory
management (refcounting) and with various common idioms converted into
simple C functions instead of duplicating them all over the place;

3) support for autosense.

Some patches are from Hannes Reinecke's megasas patchset posted last
November, forward ported and applied to the new vSCSI controller as
well.  Most have been acked by Christoph; I fixed all issues he
pointed out in the others.

I already planned the following two series too:

1) adding support for zerocopy.  Previous attempts were rejected
because they were applied to real devices (thus making for example an
IOMMU hard to impossible).  However, for PV devices zerocopy should be
uncontroversial---and it is a must to get competitive performance WRT
virtio-blk.  I'll use vmw-pvscsi for the first implementation and for
benchmarking.

2) adding support for multiple LUNs; see recently posted RFC for this.

After this I'll work on the virtio-scsi device model.

Testing:
- RHEL6.1 install complete to scsi-disk with lsi, from scsi-generic CD
- iozone run with lsi on scsi-disk target
- RHEL6.1 install to usb-msd from IDE CD is too slow, but it manages to
  format /boot in ~10 minutes with or without the patches
- RHEL6.1 install with vscsi with scsi-disk CD and installation disk
  hung at the same place with or without the series, perhaps it was
  just me being impatient.

esp is only compile tested.

Please review and merge.

v4->v5:
    SCSIBusOps now const-ified.  Added braces here and there.
    Added patch 5.

v3->v4:
    renamed scsi_req_kick to scsi_req_continue.  Changed SCSIBusOps to
    pointer.  Removed scsi_req_free, merged it into scsi_req_unref.
    Added head comments for some functions.  Renamed arguments to
    HBA callbacks.  Added patches 23 and 24

v2->v3:
    included fixes for Jonathan Nieder's recently reported bug

v1->v2:
    rebased, added patch 21

Hannes Reinecke (4):
  scsi: Use 'SCSIRequest' directly
  scsi: Update sense code handling
  scsi: Implement 'get_sense' callback
  scsi-disk: add data direction checking

Paolo Bonzini (21):
  scsi: add tracing of scsi requests
  scsi-generic: Remove bogus double complete
  scsi: introduce scsi_req_data
  scsi: introduce SCSIBusOps
  scsi-generic: do not use a stale aiocb
  scsi: reference-count requests
  lsi: extract lsi_find_by_tag
  scsi: commonize purging requests
  scsi: introduce scsi_req_abort
  scsi: introduce scsi_req_cancel
  scsi: use scsi_req_complete
  scsi: do not call send_command directly
  scsi: introduce scsi_req_new
  scsi: introduce scsi_req_continue
  scsi: introduce scsi_req_get_buf
  scsi: make write_data return void
  scsi-generic: Handle queue full
  esp: rename sense to status
  scsi: split command_complete callback in two
  scsi: rename arguments to the new callbacks
  scsi: ignore LUN field in the CDB

 hw/esp.c          |  119 ++++++++++++++----------
 hw/lsi53c895a.c   |  200 ++++++++++++++++++++++++---------------
 hw/scsi-bus.c     |  220 +++++++++++++++++++++++++++++++++++++++----
 hw/scsi-disk.c    |  271 ++++++++++++++++++++++-------------------------------
 hw/scsi-generic.c |  223 +++++++++++++++++++++-----------------------
 hw/scsi.h         |   91 ++++++++++++++----
 hw/spapr_vscsi.c  |  187 ++++++++++++++++++++----------------
 hw/usb-msd.c      |  120 ++++++++++++++----------
 trace-events      |    8 ++
 9 files changed, 866 insertions(+), 573 deletions(-)
Anthony Liguori - May 31, 2011, 1:38 p.m.
On 05/26/2011 05:56 AM, Paolo Bonzini wrote:
> The following changes since commit aa29141d84d58171c2d219f0a4b599bd76fb2e37:
>
>    Merge remote-tracking branch 'kraxel/CVE-2011-1751' into staging (2011-05-25 07:04:13 -0500)
>
> are available in the git repository at:
>
>    git://github.com/bonzini/qemu.git scsi.2
>
> This series includes the following improvements to the SCSI subsystem:

Pulled.  Thanks.

Regards,

Anthony Liguori

>
> 1) introduction of SCSIBusOps that generalize the existing
> command_complete callback;
>
> 2) widespread use of the SCSIRequest abstraction, with simpler memory
> management (refcounting) and with various common idioms converted into
> simple C functions instead of duplicating them all over the place;
>
> 3) support for autosense.
>
> Some patches are from Hannes Reinecke's megasas patchset posted last
> November, forward ported and applied to the new vSCSI controller as
> well.  Most have been acked by Christoph; I fixed all issues he
> pointed out in the others.
>
> I already planned the following two series too:
>
> 1) adding support for zerocopy.  Previous attempts were rejected
> because they were applied to real devices (thus making for example an
> IOMMU hard to impossible).  However, for PV devices zerocopy should be
> uncontroversial---and it is a must to get competitive performance WRT
> virtio-blk.  I'll use vmw-pvscsi for the first implementation and for
> benchmarking.
>
> 2) adding support for multiple LUNs; see recently posted RFC for this.
>
> After this I'll work on the virtio-scsi device model.
>
> Testing:
> - RHEL6.1 install complete to scsi-disk with lsi, from scsi-generic CD
> - iozone run with lsi on scsi-disk target
> - RHEL6.1 install to usb-msd from IDE CD is too slow, but it manages to
>    format /boot in ~10 minutes with or without the patches
> - RHEL6.1 install with vscsi with scsi-disk CD and installation disk
>    hung at the same place with or without the series, perhaps it was
>    just me being impatient.
>
> esp is only compile tested.
>
> Please review and merge.
>
> v4->v5:
>      SCSIBusOps now const-ified.  Added braces here and there.
>      Added patch 5.
>
> v3->v4:
>      renamed scsi_req_kick to scsi_req_continue.  Changed SCSIBusOps to
>      pointer.  Removed scsi_req_free, merged it into scsi_req_unref.
>      Added head comments for some functions.  Renamed arguments to
>      HBA callbacks.  Added patches 23 and 24
>
> v2->v3:
>      included fixes for Jonathan Nieder's recently reported bug
>
> v1->v2:
>      rebased, added patch 21
>
> Hannes Reinecke (4):
>    scsi: Use 'SCSIRequest' directly
>    scsi: Update sense code handling
>    scsi: Implement 'get_sense' callback
>    scsi-disk: add data direction checking
>
> Paolo Bonzini (21):
>    scsi: add tracing of scsi requests
>    scsi-generic: Remove bogus double complete
>    scsi: introduce scsi_req_data
>    scsi: introduce SCSIBusOps
>    scsi-generic: do not use a stale aiocb
>    scsi: reference-count requests
>    lsi: extract lsi_find_by_tag
>    scsi: commonize purging requests
>    scsi: introduce scsi_req_abort
>    scsi: introduce scsi_req_cancel
>    scsi: use scsi_req_complete
>    scsi: do not call send_command directly
>    scsi: introduce scsi_req_new
>    scsi: introduce scsi_req_continue
>    scsi: introduce scsi_req_get_buf
>    scsi: make write_data return void
>    scsi-generic: Handle queue full
>    esp: rename sense to status
>    scsi: split command_complete callback in two
>    scsi: rename arguments to the new callbacks
>    scsi: ignore LUN field in the CDB
>
>   hw/esp.c          |  119 ++++++++++++++----------
>   hw/lsi53c895a.c   |  200 ++++++++++++++++++++++++---------------
>   hw/scsi-bus.c     |  220 +++++++++++++++++++++++++++++++++++++++----
>   hw/scsi-disk.c    |  271 ++++++++++++++++++++++-------------------------------
>   hw/scsi-generic.c |  223 +++++++++++++++++++++-----------------------
>   hw/scsi.h         |   91 ++++++++++++++----
>   hw/spapr_vscsi.c  |  187 ++++++++++++++++++++----------------
>   hw/usb-msd.c      |  120 ++++++++++++++----------
>   trace-events      |    8 ++
>   9 files changed, 866 insertions(+), 573 deletions(-)
>
Andreas Färber - June 2, 2011, 2:54 p.m.
Am 31.05.2011 um 15:38 schrieb Anthony Liguori:

> On 05/26/2011 05:56 AM, Paolo Bonzini wrote:
>> The following changes since commit  
>> aa29141d84d58171c2d219f0a4b599bd76fb2e37:
>>
>>   Merge remote-tracking branch 'kraxel/CVE-2011-1751' into staging  
>> (2011-05-25 07:04:13 -0500)
>>
>> are available in the git repository at:
>>
>>   git://github.com/bonzini/qemu.git scsi.2
>>
>> This series includes the following improvements to the SCSI  
>> subsystem:
>
> Pulled.  Thanks.

Unfortunately that pulled in the v5 version, breaking simple trace  
build.

Paolo, do you have differential patches for your v6 already?

Andreas
Paolo Bonzini - June 3, 2011, 6:58 a.m.
On 06/02/2011 04:54 PM, Andreas Färber wrote:
>
> Unfortunately that pulled in the v5 version, breaking simple trace build.
>
> Paolo, do you have differential patches for your v6 already?

Yes, sending them today.

Paolo