final round of SCSI updates for the 5.1+ merge window

This is the final round of mostly small fixes in our initial submit. The fix for the read only regressions is the most extensive change and also intrudes outside of SCSI because the partition and read only handling is mostly in block. The specific problem is the inability to distinguish between devices marked read only by the administrator and devices that come up read only but switch to read/write once they are ready to receive data. Without the fix these devices are currently forced to stay read only causing regressions in the enterprise.

The patch is available here:

git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi.git scsi-misc

The short changelog is:

Colin Ian King (1):
scsi: bnx2fc: fix incorrect cast to u64 on shift operation

Erwan Velu (1):
scsi: smartpqi: Reporting unhandled SCSI errors

James Smart (4):
scsi: lpfc: Update lpfc version to 12.2.0.2
scsi: lpfc: add check for loss of ndlp when sending RRQ
scsi: lpfc: correct rcu unlock issue in lpfc_nvme_info_show
scsi: lpfc: resolve lockdep warnings

Martin K. Petersen (1):
scsi: sd: block: Fix regressions in read-only block device handling

Quinn Tran (1):
scsi: qla2xxx: Add cleanup for PCI EEH recovery

YueHaibing (3):
scsi: myrs: Fix uninitialized variable
scsi: qedi: remove set but not used variables 'cdev' and 'udev'
scsi: qedi: remove memset/memcpy to nfunc and use func instead

And the diffstat:

 block/blk-core.c                      |   2 +-
 block/genhd.c                         |  34 ++++--
 block/ioctl.c                         |   4 +
 block/partition-generic.c             |   7 +-
 drivers/scsi/bnx2fc/bnx2fc_hwi.c      |   2 +-
 drivers/scsi/lpfc/lpfc_attr.c         |  37 +++---
 drivers/scsi/lpfc/lpfc_els.c          |   5 +-
 drivers/scsi/lpfc/lpfc_sli.c          |  84 ++++++++-----
 drivers/scsi/lpfc/lpfc_version.h      |   2 +-
 drivers/scsi/myrs.c                   |   2 +-
 drivers/scsi/qedi/qedi_dbg.c          |  32 ++---
 drivers/scsi/qedi/qedi_iscsi.c        |   4 -
 drivers/scsi/qla2xxx/qla_os.c         | 221 +++++++++++++---------------------
 drivers/scsi/sd.c                     |   4 +-
 drivers/scsi/smartpqi/smartpqi_init.c |  23 ++--
 include/linux/genhd.h                 |  11 +-
 16 files changed, 230 insertions(+), 244 deletions(-)

James

No. That code is insane. It looks very fishy indeed to me, and I'm not pulling it this late in the game.

If the partition table gets re-read, the old read-only state should go away - for all ew know, the partition numbers may have changed, for chrissake! So you can't just say "oh, partition number 1 used to be read-only, so now we need to keep it read-only". That partition might be something completely different after the parition table has been invalidated.

So the new model that code uses looks completely bogus to me, and is not anything we've ever done before.

Just revert the oneliner SCSI change that caused the regression.

Linus

Linus,

Yeah, my mess. Sorry.

A couple of people poked me about this issue last week. I merged the patch without much scrutiny since several people had commented and tested when it was originally posted a few months back. In looking over the changes again, however, I agree with your assertion that it is fishy.

My patch wasn't exclusively trying to address the regression wrt. drives that temporarily come up read-only. Device or fabric events can also trigger revalidate and there's a whole can of worms in that department thanks to the intersection between device characteristics changing and the partition table potentially being updated. This was my feeble attempt at fixing several long-standing issues in the read-only device handling which we occasionally hit.

I'll drop the offending patch and revert Jeremy's change for now. And then revisit the gorge of eternal peril that is revalidate...

Martin K. Petersen	Oracle Linux Engineering