ncr5380: Drop legacy scsi.h include

Convert Scsi_Cmnd to struct scsi_cmnd and drop the #include "scsi.h".
The sun3_NCR5380.c core driver already uses struct scsi_cmnd so converting
the other core drivers reduces the diff which makes them easier to unify.

Signed-off-by: Finn Thain <fthain@telegraphics.com.au>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Tested-by: Michael Schmitz <schmitzmic@gmail.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

ncr5380: Remove *_RELEASE macros

The *_RELEASE macros don't tell me anything. In some cases the version in
the macro contradicts the version in the comments. Anyway, the Linux kernel
version is sufficient information. Remove these macros to improve readability.

Signed-off-by: Finn Thain <fthain@telegraphics.com.au>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Tested-by: Michael Schmitz <schmitzmic@gmail.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

ncr5380: Remove pointless compiler command line override macros

Compile-time override of scsi host defaults is pointless for drivers that
provide module parameters and __setup options for that. Too many macros make
the code hard to read so remove them.

Signed-off-by: Finn Thain <fthain@telegraphics.com.au>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Tested-by: Michael Schmitz <schmitzmic@gmail.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

ncr5380: Move static PDMA spin counters to host data

Static variables from dtc.c and pas16.c should not appear in the core
NCR5380.c driver. Aside from being a layering issue this worsens the
divergence between the three core driver variants (atari_NCR5380.c and
sun3_NCR5380.c don't support PSEUDO_DMA) and it can mean multiple hosts
share the same counters.

Fix this by making the pseudo DMA spin counters in the core more generic.
This also avoids the abuse of the {DTC,PAS16}_PUBLIC_RELEASE macros, so
they can be removed.

oak.c doesn't use PDMA and hence it doesn't use the counters and hence it
needs no write_info() method. Remove it.

Signed-off-by: Finn Thain <fthain@telegraphics.com.au>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Tested-by: Michael Schmitz <schmitzmic@gmail.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

ncr5380: Cleanup host info() methods

If the host->info() method is not set, then host->name is used by default.
For atari_scsi, that is exactly the same text. So remove the redundant
info() method. Keep sun3_scsi.c in line with atari_scsi.

Some NCR5380 drivers return an empty string from the info() method
(arm/cumana_1.c arm/oak.c mac_scsi.c) while other drivers use the default
(dmx3191d dtc.c g_NCR5380.c pas16.c t128.c).

Implement a common info() method to replace a lot of duplicated code which
the various drivers use to announce the same information.

This replaces most of the (deprecated) show_info() output and all of the
NCR5380_print_info() output. This also eliminates a bunch of code in
g_NCR5380 which just duplicates functionality in the core driver.

Signed-off-by: Finn Thain <fthain@telegraphics.com.au>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Tested-by: Michael Schmitz <schmitzmic@gmail.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

ncr5380: Remove NCR5380_STATS

The NCR5380_STATS option is only enabled by g_NCR5380 yet it adds
clutter to all three core drivers. The atari_NCR5380.c and sun3_NCR5380.c
core drivers have a slightly different implementation of the
NCR5380_STATS option.

Out of all ten NCR5380 drivers, only one of them (g_NCR5380) actually
has the code to report on the collected stats. Aside from being unreadable,
that code seems to be broken because there's no initialization of timebase.
sun3_NCR5380.c and atari_NCR5380.c have the timebase initialization but
lack the code to report the stats.

Remove all of this code to improve readability and reduce divergence
between the three core drivers.

This patch and the next one completely eliminate the PRINTP and ANDP
pre-processor abuse.

Signed-off-by: Finn Thain <fthain@telegraphics.com.au>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Tested-by: Michael Schmitz <schmitzmic@gmail.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

ncr5380: Fix SCSI_IRQ_NONE bugs

Oak scsi doesn't use any IRQ, but it sets irq = IRQ_NONE rather than
SCSI_IRQ_NONE. Problem is, the core NCR5380 driver expects SCSI_IRQ_NONE
if it is to issue IDENTIFY commands that prevent target disconnection.
And, as Geert points out, IRQ_NONE is part of enum irqreturn.

Other drivers, when they can't get an IRQ or can't use one, will set
host->irq = SCSI_IRQ_NONE (that is, 255). But when they exit they will
attempt to free IRQ 255 which was never requested.

Fix these bugs by using NO_IRQ in place of SCSI_IRQ_NONE and IRQ_NONE.
That means IRQ 0 is no longer probed by ISA drivers but I don't think
this matters.

Setting IRQ = 255 for these ISA drivers is understood to mean no IRQ.
This remains supported so as to avoid breaking existing ISA setups (which
can be difficult to get working) and because existing documentation
(SANE, TLDP etc) describes this usage for the ISA NCR5380 driver options.

Signed-off-by: Finn Thain <fthain@telegraphics.com.au>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Tested-by: Michael Schmitz <schmitzmic@gmail.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

ncr5380: Remove duplicate comments

The LIMIT_TRANSFERSIZE, PSEUDO_DMA, PARITY and UNSAFE options are all
documented in the core drivers where they are used. The same goes for the
chip databook reference. Remove the duplicate comments.

Signed-off-by: Finn Thain <fthain@telegraphics.com.au>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Christoph Hellwig <hch@lst.de>

ncr5380: Remove redundant AUTOSENSE macro

Every NCR5380 driver sets AUTOSENSE so it need not be optional (and the
mid-layer expects it). Remove this redundant macro to improve readability.

Signed-off-by: Finn Thain <fthain@telegraphics.com.au>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Tested-by: Michael Schmitz <schmitzmic@gmail.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

ncr5380: Cleanup TAG_NEXT and TAG_NONE macros

Both atari_NCR5380.c and sun3_NCR5380.c core drivers #undef TAG_NONE and
then redefine it. But the original definition is unused because NCR5380.c
lacks support for tagged queueing. So just define it once.

The TAG_NEXT macro only appears in the arguments to NCR5380_select() calls.
But that routine doesn't use its tag argument as the tag was already
assigned in NCR5380_main(). So remove the unused argument and the macro.

Signed-off-by: Finn Thain <fthain@telegraphics.com.au>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Tested-by: Michael Schmitz <schmitzmic@gmail.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

ncr5380: Remove more useless prototypes

Make use of the host template static initializer instead of assigning
handlers at run-time. Move __maybe_unused qualifiers from declarations
to definitions. Move the atari_scsi_bus_reset() wrapper after the
definition of NCR5380_bus_reset(). All of the host template handler
prototypes are now redundant so remove them.

The write_info() handler is only relevant to drivers using PSEUDO_DMA so
this patch fixes the compiler warning in atari_NCR5380.c and sun3_NCR5380.c:

  CC      drivers/scsi/atari_scsi.o
  drivers/scsi/NCR5380.h:329: warning: 'NCR5380_write_info' declared 'static' but never defined

Signed-off-by: Finn Thain <fthain@telegraphics.com.au>
Tested-by: Michael Schmitz <schmitzmic@gmail.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Christoph Hellwig <hch@lst.de>

ncr5380: Remove useless prototypes

Add missing static qualifiers and remove the now pointless prototypes. The
NCR5380_* prototypes are all declared in NCR5380.h and renamed using macros.
Further declarations are redundant (some are completely unused). Remove
them.

Signed-off-by: Finn Thain <fthain@telegraphics.com.au>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Tested-by: Michael Schmitz <schmitzmic@gmail.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

ncr5380: Remove unused macros

Some macros are never evaluated (i.e. FOO, USLEEP, SCSI2 and USE_WRAPPER;
and in some drivers, NCR5380_intr and NCR5380_proc_info). DRIVER_SETUP
serves no purpose anymore. Remove these macro definitions.

Signed-off-by: Finn Thain <fthain@telegraphics.com.au>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Tested-by: Michael Schmitz <schmitzmic@gmail.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

ncr5380: Fix compiler warnings and __setup options

Some __setup() options mentioned in Documentation/scsi don't work because
a few lines of code went missing sometime since Linux 2.4. Fix the options
and thus fix some compiler warnings for both the non-modular case,

    CC      drivers/scsi/dtc.o
drivers/scsi/dtc.c:176:20: warning: 'dtc_setup' defined but not used [-Wunused-function]

and the modular case,

    CC [M]  drivers/scsi/pas16.o
drivers/scsi/pas16.c:335:20: warning: 'pas16_setup' defined but not used [-Wunused-function]
    CC [M]  drivers/scsi/t128.o
drivers/scsi/t128.c:147:20: warning: 't128_setup' defined but not used [-Wunused-function]

Signed-off-by: Finn Thain <fthain@telegraphics.com.au>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Christoph Hellwig <hch@lst.de>

ncr5380: Remove unused hostdata fields

Remove unused fields from hostdata structs declared with the
NCR5380_implementation_fields macro.

Signed-off-by: Finn Thain <fthain@telegraphics.com.au>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Christoph Hellwig <hch@lst.de>

ncr5380: Use printk() not pr_debug()

Having defined NDEBUG, and having set the console log level, I'd like to see
some output. Don't use pr_debug(), it's annoying to have to define DEBUG as
well.

Signed-off-by: Finn Thain <fthain@telegraphics.com.au>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Tested-by: Michael Schmitz <schmitzmic@gmail.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

Fnic: Fnic Driver crashed with NULL pointer reference

When issuing I/O request, if the I/O completes before returning from
fnic_queuecommand(), we may be referencing scsi_cmnd structure that may
be freed by interrupt handler. Acquring IO lock would synchronize
fnic_queuecommand and interrupt handler.

- Increment fnic version from 1.6.0.15 to 1.6.0.16

Signed-off-by: Hiral Shah <hishah@cisco.com>
Signed-off-by: Sesidhar Baddela <sebaddel@cisco.com>
Signed-off-by: Anil Chintalapati <achintal@cisco.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

Fnic: For Standalone C series, "sending VLAN request" message seen even if the link is down

When physical link between standalone C series and switch is down,
the fip timer is not turned off and timer expiration will keep sending
vlan request.

It can be fixed by stopping the fip_timer and
it will be restarted automatically when Link is up.

- Increment fnic version from 1.6.0.14 to 1.6.0.15

Signed-off-by: Hiral Shah <hishah@cisco.com>
Signed-off-by: Sesidhar Baddela <sebaddel@cisco.com>
Signed-off-by: Anil Chintalapati <achintal@cisco.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

Fnic: Improper resue of exchange Ids

IOs belonging to an rport are aborted with Internal terminate option
when rport goes offline. Any new IO issued to the rport during this
time can reuse the terminated exchange which will cause inconsistent
state of the exchange between local port and remote port.

fc_rport_priv is set to RPORT_ST_DELETE before exchanges are aborted by
libfc. Not issuing amy more I/O requests when RPORT_ST_DELETE is set,
will avoid inconsistent state of the exchange between local port and
remote port.

- Increment fnic version from 1.6.0.13 to 1.6.0.14

Signed-off-by: Hiral Shah <hishah@cisco.com>
Signed-off-by: Sesidhar Baddela <sebaddel@cisco.com>
Signed-off-by: Anil Chintalapati <achintal@cisco.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

Fnic: Memcopy only mimumum of data or trace buffer

In case of receive path, we do not have eth header or fcoe header available
when we take a trace so we fill the fc trace buffer with 0xff for both
values. We copy only mimimum of received data or trace buffer size -
fc header - eth and fcoe header

- Increment fnic version from 1.6.0.12 to 1.6.0.13

Signed-off-by: Hiral Shah <hishah@cisco.com>
Signed-off-by: Sesidhar Baddela <sebaddel@cisco.com>
Signed-off-by: Anil Chintalapati <achintal@cisco.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

Fnic: Not probing all the vNICS via fnic_probe on boot

In fnic_dev_wait, Wait for finish to complete at least three times in two
seconds while loop before returning -ETIMEDOUT as sometime
schedule_timeout_uninterruptible takes more than two seconds to wake up.

- Increment fnic version from 1.6.0.11 to 1.6.0.12

Signed-off-by: Hiral Shah <hishah@cisco.com>
Signed-off-by: Sesidhar Baddela <sebaddel@cisco.com>
Signed-off-by: Anil Chintalapati <achintal@cisco.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

bnx2fc: fix an error code in _bnx2fc_create()

We should be returning an error code here instead of success.  Either
-ENODEV or -ENOMEM would work.  There is also a failure message in
printk().

Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Acked-by: Chad Dupuis <chad.dupuis@qlogic.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

bnx2fc: check IS_ERR() instead of NULL

The bnx2fc_if_create() function returns NULL on failure, it never
returns an error pointer.

Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Acked-by: Chad Dupuis <chad.dupuis@qlogic.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

IB/srp: Fix a race condition triggered by destroying a queue pair

At least LID reassignment can trigger a race condition in the SRP
initiator driver, namely the receive completion handler trying to
post a request on a QP during or after QP destruction and before
the CQ's have been destroyed. Avoid this race by modifying a QP
into the error state and by waiting until all receive completions
have been processed before destroying a QP.

Reported-by: Max Gurtuvoy <maxg@mellanox.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

IB/srp: Add multichannel support

Improve performance by using multiple RDMA/RC channels per SCSI
host for communication with an SRP target. About the
implementation:
- Introduce a loop over all channels in the code that uses
  target->ch.
- Set the SRP_MULTICHAN_MULTI flag during login for the creation
  of the second and subsequent channels.
- RDMA completion vectors are chosen such that RDMA completion
  interrupts are handled by the CPU socket that submitted the I/O
  request. As one can see in this patch it has been assumed if a
  system contains n CPU sockets and m RDMA completion vectors
  have been assigned to an RDMA HCA that IRQ affinity has been
  configured such that completion vectors [i*m/n..(i+1)*m/n) are
  bound to CPU socket i with 0 <= i < n.
- Modify srp_free_ch_ib() and srp_free_req_data() such that it
  becomes safe to invoke these functions after the corresponding
  allocation function failed.
- Add a ch_count sysfs attribute per target port.

Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

IB/srp: Use block layer tags

Since the block layer already contains functionality to assign
a tag to each request, use that functionality instead of
reimplementing that functionality in the SRP initiator driver.
This change makes the free_reqs list superfluous. Hence remove
that list.

[hch: updated to use .use_blk_tags instead scsi_activate_tcq]
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

IB/srp: Separate target and channel variables

Changes in this patch:
- Move channel variables into a new structure (struct srp_rdma_ch).
- Add an srp_target_port pointer, 'lock' and 'comp_vector' members
  in struct srp_rdma_ch.
- Add code to initialize these three new member variables.
- Many boring "target->" into "ch->" changes.
- The cm_id and completion handler context pointers are now of type
  srp_rdma_ch * instead of srp_target_port *.
- Three kzalloc(a * b, f) calls have been changed into kcalloc(a, b, f)
  to avoid that this patch would trigger a checkpatch warning.
- Two casts from u64 into unsigned long long have been left out
  because these are superfluous. Since considerable time u64 is
  defined as unsigned long long for all architectures supported by
  the Linux kernel.

Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Acked-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

IB/srp: Introduce two new srp_target_port member variables

Introduce the srp_target_port member variables 'sgid' and 'pkey'.
Change the type of 'orig_dgid' from __be16[8] into union ib_gid.
This patch does not change any functionality but makes the
"Separate target and channel variables" patch easier to verify.

Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

IB/srp: Avoid that I/O hangs due to a cable pull during LUN scanning

If a cable is pulled during LUN scanning it can happen that the
SRP rport and the SCSI host have been created but no LUNs have been
added to the SCSI host. Since multipathd only sends SCSI commands
to a SCSI target if one or more SCSI devices are present and since
there is no keepalive mechanism for IB queue pairs this means that
after a LUN scan failed and after a reconnect has succeeded no
data will be sent over the QP and hence that a subsequent cable
pull will not be detected. Avoid this by not creating an rport or
SCSI host if a cable is pulled during a SCSI LUN scan.

Note: so far the above behavior has only been observed with the
kernel module parameter ch_count set to a value >= 2.

Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

IB/srp: Remove stale connection retry mechanism

Attempting to connect three times may be insufficient after an
initiator system tries to relogin, especially if the relogin
attempt occurs before the SRP target service ID has been
registered. Since the srp_daemon retries a failed login attempt
anyway, remove the stale connection retry mechanism.

Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

IB/srp: Move ib_destroy_cm_id() call into srp_free_ch_ib()

The patch that adds multichannel support into the SRP initiator
driver introduces an additional call to srp_free_ch_ib(). This
patch helps to keep that later patch simple.

Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Sagi Grimberg <sagig@mellanox.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

qla2xxx: remove redundant declaration in 'qla_gbl.h'

Remove 2 redundant extern inline functions: qla8044_set_qsnt_ready() and
qla8044_need_reset_handler(). At present, within upstream next kernel
source code, they are only used within "drivers/scsi/qla2xxx/qla_nx2.c".

The related error and warnings (with allmodconfig under tile):

    CC [M]  drivers/scsi/qla2xxx/qla_nx2.o
  drivers/scsi/qla2xxx/qla_nx2.c:1633:1: error: static declaration of 'qla8044_need_reset_handler' follows non-static declaration
   qla8044_need_reset_handler(struct scsi_qla_host *vha)
   ^
  In file included from drivers/scsi/qla2xxx/qla_def.h:3706:0,
                   from drivers/scsi/qla2xxx/qla_nx2.c:11:
  drivers/scsi/qla2xxx/qla_gbl.h:756:20: note: previous declaration of 'qla8044_need_reset_handler' was here
   extern inline void qla8044_need_reset_handler(struct scsi_qla_host *vha);
                      ^
  drivers/scsi/qla2xxx/qla_gbl.h:756:20: warning: inline function 'qla8044_need_reset_handler' declared but never defined
  make[3]: *** [drivers/scsi/qla2xxx/qla_nx2.o] Error 1
  make[2]: *** [drivers/scsi/qla2xxx] Error 2
  make[1]: *** [drivers/scsi] Error 2
  make: *** [drivers] Error 2

    CC [M]  drivers/scsi/qla2xxx/qla_tmpl.o
  In file included from drivers/scsi/qla2xxx/qla_def.h:3706:0,
                   from drivers/scsi/qla2xxx/qla_tmpl.c:7:
  drivers/scsi/qla2xxx/qla_gbl.h:755:20: warning: inline function 'qla8044_set_qsnt_ready' declared but never defined
   extern inline void qla8044_set_qsnt_ready(struct scsi_qla_host *vha);
                    ^

Signed-off-by: Chen Gang <gang.chen.5i5j@gmail.com>
Acked-by: Saurav Kashyap <saurav.kashyap@qlogic.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

ipr: set coherent DMA mask

Use dma_set_mask_and_coherent() to set both the DMA and coherent
DMA mask.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

ipr: convert to generic DMA API

Even though the ipr driver is only used on PCI, convert it
to use the generic DMA API.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

vmw_pvscsi: fixup tagging

The request (and SCSI command) tag is the tag number assigned
by the generic block-tagging code, not the SCSI-II tag messages.
Those are represented by the device flags 'tagged_supported',
'simple_tags', and 'ordered_tags'.
(The SCSI midlayer doesn't use HEAD_OF_QUEUE tags).
So fixup vmw_pvscsi to assign the correct tag type.

[hch: fixed up to never set MSG_ORDERED_TAG]
Signed-off-by: Hannes Reinecke <hare@suse.de>
Acked-by: Arvind Kumar <arvindkumar@vmware.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

scsi_debug: error message should say scsi_host_alloc not scsi_register

Signed-off-by: Finn Thain <fthain@telegraphics.com.au>
Acked-by: Douglas Gilbert <dgilbert@interlog.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

tmscim: remove unused SCSI_IRQ_NONE macro definition

This macro is only used in the NCR5380 drivers and they don't include
this header.

Signed-off-by: Finn Thain <fthain@telegraphics.com.au>
Acked-by: Guennadi Liakhovetski <g.liakhovetski@gmx.de>
Signed-off-by: Christoph Hellwig <hch@lst.de>

ufs: remove spurious scsi_set_tag_type call

ufs never looks at the tag type, so there is no need to set it either.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Mike Christie <michaelc@cs.wisc.edu>
Reviewed-by: Hannes Reinecke <hare@suse.de>

scsi: don't force tagged_supported in drivers

Now that we also get proper values in cmd->request->tag for untagged
commands, there is no need to force tagged_supported to on in drivers
that need host-wide tags.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Mike Christie <michaelc@cs.wisc.edu>
Reviewed-by: Hannes Reinecke <hare@suse.de>

scsi: don't set tagging state from scsi_adjust_queue_depth

Remove the tagged argument from scsi_adjust_queue_depth, and just let it
handle the queue depth.  For most drivers those two are fairly separate,
given that most modern drivers don't care about the SCSI "tagged" status
of a command at all, and many old drivers allow queuing of multiple
untagged commands in the driver.

Instead we start out with the ->simple_tags flag set before calling
->slave_configure, which is how all drivers actually looking at
->simple_tags except for one worke anyway.  The one other case looks
broken, but I've kept the behavior as-is for now.

Except for that we only change ->simple_tags from the ->change_queue_type,
and when rejecting a tag message in a single driver, so keeping this
churn out of scsi_adjust_queue_depth is a clear win.

Now that the usage of scsi_adjust_queue_depth is more obvious we can
also remove all the trivial instances in ->slave_alloc or ->slave_configure
that just set it to the cmd_per_lun default.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Mike Christie <michaelc@cs.wisc.edu>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: always assign block layer tags if enabled

Allow a driver to ask for block layer tags by setting .use_blk_tags in the
host template, in which case it will always see a valid value in
request->tag, similar to the behavior when using blk-mq.  This means even
SCSI "untagged" commands will now have a tag, which is especially useful
when using a host-wide tag map.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Mike Christie <michaelc@cs.wisc.edu>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>

scsi: remove use_blk_tcq Scsi_Host field

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Mike Christie <michaelc@cs.wisc.edu>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>

mptfusion: don't change queue type in ->change_queue_depth

This function shouldn't change the queue type, just the depth.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Mike Christie <michaelc@cs.wisc.edu>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>

scsi: remove abuses of scsi_populate_tag

Unless we want to build a SPI tag message we should just check SCMD_TAGGED
instead of reverse engineering a tag type through the use of
scsi_populate_tag_msg.

Also rename the function to spi_populate_tag_msg, make it behave like the
other spi message helpers, and move it to the spi transport class.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Mike Christie <michaelc@cs.wisc.edu>
Reviewed-by: Hannes Reinecke <hare@suse.de>

scsi: remove ordered_tag host template field

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Mike Christie <michaelc@cs.wisc.edu>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>

scsi: remove ordered_tags scsi_device field

Remove the ordered_tags field, we haven't been issuing ordered tags based
on it since the big barrier rework in 2010.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Mike Christie <michaelc@cs.wisc.edu>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>

scsi: add new scsi-command flag for tagged commands

Currently scsi piggy backs on the block layer to define the concept
of a tagged command.  But we want to be able to have block-level host-wide
tags assigned even for untagged commands like the initial INQUIRY, so add
a new SCSI-level flag for commands that are tagged at the scsi level, so
that even commands without that set can have tags assigned to them.  Note
that this alredy is the case for the blk-mq code path, and this just lets
the old path catch up with it.

We also set this flag based upon sdev->simple_tags instead of the block
queue flag, so that it is entirely independent of the block layer tagging,
and thus always correct even if a driver doesn't use block level tagging
yet.

Also remove the old blk_rq_tagged; it was only used by SCSI drivers, and
removing it forces them to look for the proper replacement.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Mike Christie <michaelc@cs.wisc.edu>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>

scsi: provide a generic change_queue_type method

Most drivers use exactly the same implementation, so provide it as a
library function.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Mike Christie <michaelc@cs.wisc.edu>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>

scsi: handle more device handler setup/teardown in common code

Move all code to set up and tear down sdev->scsi_dh_data to common code.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Mike Christie <michaelc@cs.wisc.edu>
Reviewed-by: Hannes Reinecke <hare@suse.de>

scsi: device handlers must have attach and detach methods

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Mike Christie <michaelc@cs.wisc.edu>
Reviewed-by: Hannes Reinecke <hare@suse.de>

scsi: remove struct scsi_dh_devlist

All drivers now do their own matching, so there is no more need to expose
a device list as part of the interface.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Mike Christie <michaelc@cs.wisc.edu>
Reviewed-by: Hannes Reinecke <hare@suse.de>

scsi: use container_of to get at device handler private data

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Mike Christie <michaelc@cs.wisc.edu>
Reviewed-by: Hannes Reinecke <hare@suse.de>

scsi_dh: get module reference outside of device handler

We need to grab a reference to the module before calling the attach
routines to avoid a small race vs module removal.  It also cleans up
the code significantly as a side effect.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Mike Christie <michaelc@cs.wisc.edu>
Reviewed-by: Hannes Reinecke <hare@suse.de>

scsi_dh_hp_sw: fix return value on failed allocation

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Mike Christie <michaelc@cs.wisc.edu>
Reviewed-by: Hannes Reinecke <hare@suse.de>

sd: disable discard_zeroes_data for UNMAP

The T10 SBC UNMAP command does not provide any hard guarantees that
blocks will return zeroes on a subsequent READ. This is due to the fact
that the device server is free to silently ignore all or parts of the
request.

The only way to ensure that a block consistently returns zeroes after
being unmapped is to use WRITE SAME with the UNMAP bit set. Should the
device be unable to unmap one or more blocks described by the command it
is required to manually write zeroes to them.

Until now we have preferred UNMAP over the WRITE SAME variants to
accommodate thinly provisioned devices that predated the final SBC-3
spec. This patch changes the heuristic so that we favor WRITE SAME(16)
or (10) over UNMAP if these commands are marked as supported in the
Logical Block Provisioning VPD page.

The patch also disables discard_zeroes_data for devices operating in
UNMAP mode.

Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

scsi: return EAGAIN when resetting a device under EH

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>

osst: call scsi_set_medium_removal directly

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>

st: call scsi_set_medium_removal directly

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>

sd: fix up ->compat_ioctl

No need to verify the passthrough ioctls, the real handler will
take care of that.  Also make sure not to block for resets on
O_NONBLOCK fds.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>

scsi: split scsi_nonblockable_ioctl

The calling conventions for this function are bad as it could return
-ENODEV both for a device not currently online and a not recognized ioctl.

Add a new scsi_ioctl_block_when_processing_errors function that wraps
scsi_block_when_processing_errors with the a special case for the
SG_SCSI_RESET ioctl command, and handle the SG_SCSI_RESET case itself
in scsi_ioctl.  All callers of scsi_ioctl now must call the above helper
to check for the EH state, so that the ioctl handler itself doesn't
have to.

Reported-by: Robert Elliott <Elliott@hp.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>

scsi: refactor scsi_reset_provider handling

Pull the common code from the two callers into the function,
and rename it to scsi_ioctl_reset.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>

scsi: add support for multiple hardware queues in scsi_(host_)find_tag

Modify scsi_find_tag() and scsi_host_find_tag() such that these
functions can translate a tag generated by blk_mq_unique_tag().

Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

scsi: add support for multiple hardware queues

Allow a SCSI LLD to declare how many hardware queues it supports
by setting Scsi_Host.nr_hw_queues before calling scsi_add_host().

Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Sagi Grimberg <sagig@mellanox.com>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

blk-mq: add blk_mq_unique_tag()

The queuecommand() callback functions in SCSI low-level drivers
need to know which hardware context has been selected by the
block layer. Since this information is not available in the
request structure, and since passing the hctx pointer directly to
the queuecommand callback function would require modification of
all SCSI LLDs, add a function to the block layer that allows to
query the hardware context index.

Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Acked-by: Jens Axboe <axboe@kernel.dk>
Reviewed-by: Sagi Grimberg <sagig@mellanox.com>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

scsi: ratelimit I/O error messages

There can be quite a lot of I/O error messages, even on smaller
machines. So we need to ratelimit them to not overwhelm logging.

Signed-off-by: Hannes Reinecke <hare@suse.de>
Tested-by: Robert Elliott <elliott@hp.com>
Reviewed-by: Robert Elliott <elliott@hp.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

scsi: correct return values for .eh_abort_handler implementations

The .eh_abort_handler needs to return SUCCESS, FAILED, or
FAST_IO_FAIL. So fixup all callers to adhere to this requirement.

Reviewed-by: Robert Elliott <elliott@hp.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Christoph Hellwig <hch@lst.de>

scsi: document scsi_try_to_abort_cmd

scsi_try_to_abort_cmd() should only return SUCCESS, FAILED, or
FAST_IO_FAIL. So document that in the function description and simplify
the logging message.

Signed-off-by: Hannes Reinecke <hare@suse.de>
Suggested-by: Christoph Hellwig <hch@infradead.org>
Reviewed-by: Robert Elliott <elliott@hp.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

scsi: use shost argument in scsi_eh_prt_fail_stats

The EH statistics are per host, so we should be using
shost_printk() here.

Signed-off-by: Hannes Reinecke <hare@suse.de>
Suggested-by: Robert Elliott <elliott@hp.com>
Reviewed-by: Robert Elliott <elliott@hp.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

scsi: fixup logging messages in scsi_error.c

Use the matching scope for logging messages to allow for
better command tracing.

Signed-off-by: Hannes Reinecke <hare@suse.de>
Suggested-by: Robert Elliott <elliott@hp.com>
Reviewed-by: Robert Elliott <elliott@hp.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

scsi: simplify scsi_log_(send|completion)

Simplify scsi_log_(send|completion) by externalizing
scsi_mlreturn_string() and always print the command address.

Signed-off-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Robert Elliott <elliott@hp.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

scsi: remove scsi_show_result()

Open-code scsi_print_result in sd.c, and cleanup logging to
not print duplicate informations.
Also remove the call to scsi_show_result() in ufshcd.c
to be consistent with other callers of scsi_execute().

With that we can remove scsi_show_result in constants.c

Signed-off-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Robert Elliott <elliott@hp.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

scsi: separate out scsi_(host|driver)byte_string()

Export functions for later use.

Signed-off-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Robert Elliott <elliott@hp.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

scsi: Remove scsi_print_command when calling abort

Calling scsi_print_command should not be necessary during abort;
if the information is required one should enable scsi logging.

Signed-off-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Robert Elliott <elliott@hp.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

scsi: repurpose the last argument from print_opcode_name()

print_opcode_name() was only ever called with a '0' argument
from LLDDs and ULDs which were _not_ supporting variable length
CDBs, so the 'if' clause was never triggered.
Instead we should be using the last argument to specify
the cdb length to avoid accidental overflow when reading
the cdb buffer.

Signed-off-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Robert Elliott <elliott@hp.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

scsi: consolidate opcode lookup in scsi_opcode_sa_name()

Consolidate the CDB opcode lookup in scsi_opcode_sa_name(),
so that we don't have to call several functions to figure
out the CDB opcode string.

Signed-off-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Robert Elliott <elliott@hp.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

scsi: merge print_opcode_name()

Instead of having two versions of print_opcode_name() we
should be consolidating them into one version.

Signed-off-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Robert Elliott <elliott@hp.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

scsi: implement scsi_opcode_sa_name

Implement a lookup array for SERVICE ACTION commands instead
of hardcoding it in a large switch statement.

Signed-off-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Robert Elliott <elliott@hp.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

scsi: remove scsi_print_status()

Last caller is gone, so we can remove it.

Signed-off-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Robert Elliott <elliott@hp.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

scsi: use 'bool' as return value for scsi_normalize_sense()

Convert scsi_normalize_sense() and friends to return 'bool'
instead of an integer.

Signed-off-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Robert Elliott <elliott@hp.com>
Reviewed-by: Yoshihiro Yunomae <yoshihiro.yunomae.ez@hitachi.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

scsi: do not decode sense extras

Currently we're only decoding sense extras for tape devices.
And even there only for fixed format sense formats.
As this is of rather limited use in the general case we should
be stop trying to decode sense extras; the tape driver does
its own decoding anyway.

Signed-off-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Robert Elliott <elliott@hp.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

scsi: stop decoding if scsi_normalize_sense() fails

If scsi_normalize_sense() fails we couldn't decode the sense
buffer, and the scsi_sense_hdr fields are invalid.
For those cases we should rather dump the sense buffer
and not try to decode invalid fields.

Signed-off-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Robert Elliott <elliott@hp.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

53c700: remove scsi_print_sense() usage

The 53c700 driver would be using scsi_print_sense() in a debug
statement, which was never compiled in. Plus the same information
can get retrieved with logging. So remove it.

Signed-off-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Robert Elliott <elliott@hp.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

fas216: update logging messages

Update logging messages to use dev_printk() variants for correct
device annotations.

Signed-off-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Robert Elliott <elliott@hp.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

fas216: return DID_ERROR for incomplete data transfer

fas216 returns DID_BAD_TARGET for an incomplete data
transfer. The midlayer uses DID_BAD_TARGET to signal
a non-existing or not reachable target. So we should
rather be using DID_ERROR here.

Signed-off-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Robert Elliott <elliott@hp.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

acornscsi: use scsi_print_command()

Update acornscsi to use scsi_print_command() instead of the
underscore version and use scmd_printk() in acornscsi_done().
This will add correct device annotations in the resulting message.
And we should be using set_host_byte() for setting the
final result.

Signed-off-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Robert Elliott <elliott@hp.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

scsi: use sdev as argument for sense code printing

We should be using the standard dev_printk() variants for
sense code printing.

[hch: remove __scsi_print_sense call in xen-scsiback, Acked by Juergen]
[hch: folded bracing fix from Dan Carpenter]
Signed-off-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Robert Elliott <elliott@hp.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

scsi: introduce sdev_prefix_printk()

Like scmd_printk(), but the device name is passed in as
a string. Can be used by eg ULDs which do not have access
to the scsi_cmnd structure.

Signed-off-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Robert Elliott <elliott@hp.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

aha152x: debug output update and whitespace cleanup

Remove all uncommented debugging code and move all
printk() statements over to dev_printk().
And while we're at it we should be doing a whitespace
cleanup, too.

Signed-off-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Robert Elliott <elliott@hp.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

sd: remove scsi_print_sense() in sd_done()

sd_done() was calling scsi_print_sense() for a sense code
of 'NO_SENSE'.

Signed-off-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Robert Elliott <elliott@hp.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

scsi: remove scsi_cmd_print_sense_hdr()

Unused.

Signed-off-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Robert Elliott <elliott@hp.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

enclosure: handle non-unique element descriptors

Some SES devices give non-unique Element Descriptors as part of the
Element Descriptor diag page. Since we use these for creating sysfs
entries, they need to be unique. The specification doesn't require
these to be unique.

Eg:
$ sg_ses -p 7 /dev/sg0
  FTS CORP  TXS6_SAS20BPX12   0500
    enclosure services device
Element descriptor In diagnostic page:
  generation code: 0x0
  element descriptor by type list
    Element type: Array device, subenclosure id: 0
      Overall descriptor: ArrayDevicesInSubEnclsr0
      Element 1 descriptor: ArrayDevice00
      Element 2 descriptor: ArrayDevice01
      Element 3 descriptor: ArrayDevice02
      Element 4 descriptor: ArrayDevice03
      Element 5 descriptor: ArrayDevice03
      Element 6 descriptor: ArrayDevice03
      Element 7 descriptor: ArrayDevice03
      Element 8 descriptor: ArrayDevice03
      Element 9 descriptor: ArrayDevice03
      Element 10 descriptor: ArrayDevice03
      Element 11 descriptor: ArrayDevice03
      Element 12 descriptor: ArrayDevice03

Based on http://thread.gmane.org/gmane.linux.scsi/69289. This
version implements James' ideas about the naming convention

Signed-off-by: Markus Stockhausen <stockhausen@collogia.de>
Acked-by: James Bottomley <James.Bottomley@HansenPartnership.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

st: add a debug_flag module parameter request

This patch adds a debug_flag parameter that can be set on module load, and allows the DEBUG facility without a module recompile.
Note that now DEBUG 1 is the default with this patch.

Usage: modprobe st debug_flag=1

Signed-off-by: Laurence Oberman <loberman@redhat.com>
Acked-by: Kai M??kisara <kai.makisara@kolumbus.fi>
Signed-off-by: Christoph Hellwig <hch@lst.de>

scsi: add SG_SCSI_RESET_NO_ESCALATE flag to SG_SCSI_RESET ioctl

Further to a January 2013 thread titled: "[PATCH] SG_SCSI_RESET ioctl
should only perform requested operation" by Jeremy Linton a patch (v3)
is presented that expands the existing ioctl to include "no_escalate"
versions to the existing resets. This requires no changes to SCSI low
level drivers (LLDs); it adds several more finely tuned reset options
to the user space. For example:

   /* This call remains the same, with the same escalating semantics
    * if the device (LU) reset fail. That is: on failure to try a
    * target reset and if that fails, try a bus reset, and if that fails
    * try a host (i.e. LLD) reset. */
   val = SG_SCSI_RESET_DEVICE;
   res = ioctl(<sg_or_block_fd>, SG_SCSI_RESET, &val);

   /* What follows is a new option introduced by this patch series. Only
    * a device reset is attempted. If that fails then an appropriate
    * error code is provided. N.B. There is no reset escalation. */
   val = SG_SCSI_RESET_DEVICE | SG_SCSI_RESET_NO_ESCALATE;
   res = ioctl(<sg_or_block_fd>, SG_SCSI_RESET, &val);

Signed-off-by: Douglas Gilbert <dgilbert@interlog.com>
Reviewed-by: Jeremy Linton <jlinton@tributary.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Christoph Hellwig <hch@lst.de>

scsi: PC partition tables are little endian

As sparse correctly pointed out, scsi_partsize should use get_unaligned_le32
to read PC partition tables from disk, as they are little endian.

The result of this bug is that we returned incorrect geometries on big
endian systems when using the scsicam variant.  Which probably doesn't
matter as only old x86 systems every cared about the geometry.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>

scsi: resolve some missing-field-initializers warnings

Resolve some missing-field-initializers warnings by using
designated initialization.

[hch: W=2 with modern gcc warns about this.  Pretty pointless to me, but
 I'd prefer to keep us warning free]

Signed-off-by: Mark Rustad <mark.d.rustad@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

scsi: fix off-by-one LUN check in scsi_scan_host_selected()

The Scsi_Host structure max_lun field is the maximum allowed LUN plus 1. So
a LUN value is invalid if >= max_lun.

Signed-off-by: Mark Knibbs <markk@clara.co.uk>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

scsi: fix trivial typos in scsi_scan.c comment

Signed-off-by: Mark Knibbs <markk@clara.co.uk>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

bnx2fc: fix tgt spinlock locking

bnx2fc_queuecommand(): when allocating a new io_req, the tgt_lock
spinlock must be locked before calling bnx2fc_cmd_alloc().

The spinlock should also be kept locked until bnx2fc_post_io_req() has
been completed.
If not, a kernel thread may call bnx2fc_process_cq_compl() that extracts
the newly allocated io_req from hba->cmd_mgr->cmds and destroys it while
it is still being used by bnx2fc_post_io_req().

BUG: unable to handle kernel NULL pointer dereference at 000000000000004c
IP: [<ffffffffa03130da>] bnx2fc_init_task+0x6a/0x230 [bnx2fc]
PGD 0
Oops: 0000 [#1] SMP
last sysfs file: /sys/devices/pci0000:00/0000:00:02.0/0000:04:00.3/net/eth3/type
CPU 33
Modules linked in: autofs4 target_core_iblock target_core_file target_core_pscsi target_core_mod configfs bnx2fc cnic uio fcoe libfcoe libfc scsi_transport_fc 8021q garp scsi_tgt stp llc cpufreq_ondemand freq_table pcc_cpufreq ipt_REJECT nf_conntrack_ipv4 nf_defrag_ipv4 iptable_filter ip_tables ip6t_REJECT nf_conntrack_ipv6 nf_defrag_ipv6 xt_state nf_conntrack ip6table_filter ip6_tables ipv6 power_meter microcode iTCO_wdt iTCO_vendor_support hpilo hpwdt sg bnx2x libcrc32c mdio serio_raw lpc_ich mfd_core shpchp ext4 jbd2 mbcache sd_mod crc_t10dif hpsa video output dm_mirror dm_region_hash dm_log dm_mod [last unloaded: scsi_wait_scan]

Pid: 7355, comm: bnx2fc_thread/3 Not tainted 2.6.32-431.el6.x86_64 #1 HP ProLiant BL460c Gen8
RIP: 0010:[<ffffffffa03130da>]  [<ffffffffa03130da>] bnx2fc_init_task+0x6a/0x230 [bnx2fc]
RSP: 0018:ffff8820b0da3b68  EFLAGS: 00010246
RAX: 0000000000000000 RBX: ffff882003801080 RCX: 0000000000000000
RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffff882003801100
RBP: ffff8820b0da3bc8 R08: ffffffff8160d4e8 R09: 0000000000000040
R10: 0000000000000000 R11: 0000000000000000 R12: ffff88400e600e00
R13: ffff8840108fbe40 R14: ffff88200ffe5400 R15: 0000000000000000
FS:  0000000000000000(0000) GS:ffff8820b0da0000(0000) knlGS:0000000000000000
CS:  0010 DS: 0018 ES: 0018 CR0: 000000008005003b
CR2: 000000000000004c CR3: 0000002010b67000 CR4: 00000000001407e0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400
Process bnx2fc_thread/3 (pid: 7355, threadinfo ffff88401f940000, task ffff884012f5f540)
Stack:
 ffff8820b0da3bc8 ffffffff81527303 ffff884000000020 ffff8820b0da3bd8
<d> ffff8820b0da3b98 000000028138931a ffff88400f506078 ffff88400e600e00
<d> ffff88200ffe5400 ffff88200ffe5590 0000000000000021 0000000000000002
Call Trace:
 <IRQ>
 [<ffffffff81527303>] ? printk+0x41/0x46
 [<ffffffffa03169bc>] bnx2fc_post_io_req+0x11c/0x440 [bnx2fc]
 [<ffffffff812825b9>] ? cpumask_next_and+0x29/0x50
 [<ffffffff8137ffd0>] ? scsi_done+0x0/0x60
 [<ffffffffa0316df7>] bnx2fc_queuecommand+0x117/0x140 [bnx2fc]
 [<ffffffff81380245>] scsi_dispatch_cmd+0xe5/0x310
 [<ffffffff81388b9e>] scsi_request_fn+0x5ee/0x7a0
 [<ffffffff812658f1>] __blk_run_queue+0x31/0x40
 [<ffffffff81265a40>] blk_run_queue+0x30/0x50
 [<ffffffff81387da6>] scsi_run_queue+0xc6/0x270
 [<ffffffff81260f92>] ? elv_requeue_request+0x52/0xa0
 [<ffffffff813897a0>] scsi_requeue_command+0x90/0xb0
 [<ffffffff81389b84>] scsi_io_completion+0x154/0x6c0
 [<ffffffff8137ff62>] scsi_finish_command+0xc2/0x130
 [<ffffffff8138a255>] scsi_softirq_done+0x145/0x170
 [<ffffffff8126e865>] blk_done_softirq+0x85/0xa0
 [<ffffffff8107a8e1>] __do_softirq+0xc1/0x1e0
 [<ffffffff8100c30c>] ? call_softirq+0x1c/0x30
 [<ffffffff8100c30c>] call_softirq+0x1c/0x30
 <EOI>
 [<ffffffff8100fa75>] ? do_softirq+0x65/0xa0
 [<ffffffff8107a40a>] local_bh_enable_ip+0x9a/0xb0
 [<ffffffff8152a4eb>] _spin_unlock_bh+0x1b/0x20
 [<ffffffffa0313937>] bnx2fc_process_cq_compl+0x257/0x2b0 [bnx2fc]
 [<ffffffffa03114ea>] bnx2fc_percpu_io_thread+0xea/0x160 [bnx2fc]
 [<ffffffffa0311400>] ? bnx2fc_percpu_io_thread+0x0/0x160 [bnx2fc]
 [<ffffffff8109aef6>] kthread+0x96/0xa0
 [<ffffffff8100c20a>] child_rip+0xa/0x20
 [<ffffffff8109ae60>] ? kthread+0x0/0xa0
 [<ffffffff8100c200>] ? child_rip+0x0/0x20
Code: 89 df 45 8b 7e 30 0f 85 75 01 00 00 89 d1 31 c0 c1 e9 03 83 e2 04 89 c9 f3 48 ab 74 06 c7 07 00 00 00 00 49 89 9c 24 88 01 00 00 <83> 7e 4c 01 b8 01 00 00 00 0f 84 e7 00 00 00 89 c2 0a 53 38 41
RIP  [<ffffffffa03130da>] bnx2fc_init_task+0x6a/0x230 [bnx2fc]
 RSP <ffff8820b0da3b68>
CR2: 000000000000004c

Signed-off-by: Maurizio Lombardi <mlombard@redhat.com>
Acked-by: Chad Dupuis <chad.dupuis@qlogic.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

megaraid_sas: fix bug in handling return value of pci_enable_msix_range()

Function pci_enable_msix_range() may return negative values for error
conditions. So it's a bug by checking (pci_enable_msix_range() != 0)
for success and causes failure to megaraid driver when MSI is disabled.
[   16.487267] megaraid_sas 0000:02:00.0: Controller type: iMR
[   16.487275] genirq: Flags mismatch irq 0. 00000000 (megasas) vs. 00015a00 (tii
mer)
[   16.487347] megasas: Failed to register IRQ for vector 0.

Fixes: 8ae80ed1734b "megaraid: Use pci_enable_msix_range() instead of pci_enable_msix()"
Signed-off-by: Jiang Liu <jiang.liu@linux.intel.com>
Acked-by: Sumit Saxena <sumit.saxena@avagotech.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Cc: <stable@vger.kernel.org> # 3.17

cxgb4i: send abort_rpl correctly

Connection retries were not being cleaned up correctly if they failed as a
result of link down. Applies on top of drivers-for-3.18.

Signed-off-by: Anish Bhatt <anish@chelsio.com>
Signed-off-by: Karen Xie <kxie@chelsio.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>