platform/kernel/linux-exynos.git
7 years agoscsi: mpt: Move scsi_remove_host() out of mptscsih_remove_host()
Hannes Reinecke [Mon, 24 Apr 2017 09:22:32 +0000 (11:22 +0200)]
scsi: mpt: Move scsi_remove_host() out of mptscsih_remove_host()

Commit c5ce0abeb628 ("scsi: sas: move scsi_remove_host call...")  moved
the call to scsi_remove_host() into sas_remove_host(), but forgot to
modify the mpt drivers.

Fixes: c5ce0abeb628 ("scsi: sas: move scsi_remove_host call into sas_remove_host")
Signed-off-by: Hannes Reinecke <hare@suse.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: sg: reset 'res_in_use' after unlinking reserved array
Hannes Reinecke [Mon, 24 Apr 2017 08:26:36 +0000 (10:26 +0200)]
scsi: sg: reset 'res_in_use' after unlinking reserved array

Once the reserved page array is unused we can reset the 'res_in_use'
state; here we can do a lazy update without holding the mutex as we only
need to check against concurrent access, not concurrent release.

[mkp: checkpatch]

Fixes: 1bc0eb044615 ("scsi: sg: protect accesses to 'reserved' page array")
Signed-off-by: Hannes Reinecke <hare@suse.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: mvumi: remove code handling zero scsi_sg_count(scmd) case
Alexey Khoroshilov [Sun, 23 Apr 2017 23:01:00 +0000 (02:01 +0300)]
scsi: mvumi: remove code handling zero scsi_sg_count(scmd) case

As Christoph Hellwig noted, SCSI commands that transfer data always have
a SG entry. The patch removes dead code in mvumi_make_sgl(),
mvumi_complete_cmd() and mvumi_timed_out() that handle zero
scsi_sg_count(scmd) case.

Also the patch adds pci_unmap_sg() on failure path in mvumi_make_sgl().

Signed-off-by: Alexey Khoroshilov <khoroshilov@ispras.ru>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: fusion: fix spelling mistake: "Persistancy" -> "Persistency"
Colin Ian King [Sat, 22 Apr 2017 13:02:02 +0000 (14:02 +0100)]
scsi: fusion: fix spelling mistake: "Persistancy" -> "Persistency"

trivial fix to spelling mistake

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: pmcraid: use normal copy_from_user
Arnd Bergmann [Fri, 21 Apr 2017 22:02:31 +0000 (00:02 +0200)]
scsi: pmcraid: use normal copy_from_user

As pointed out by Al Viro for my previous series, the driver has no need
to call access_ok() and __copy_from_user()/__copy_to_user(). Changing
it to regular copy_from_user()/copy_to_user() simplifies the code without
any real downsides, making it less error-prone at best.

This patch by itself also addresses the warning about the access_ok()
macro on MIPS, but both fixes improve the code, so ideally we apply
them both.

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: pmcraid: fix minor sparse warnings
Arnd Bergmann [Thu, 20 Apr 2017 17:54:48 +0000 (19:54 +0200)]
scsi: pmcraid: fix minor sparse warnings

pmcraid_minor is only used in this one file and should be 'static' as suggested
by sparse:

drivers/scsi/pmcraid.c:80:1: warning: symbol 'pmcraid_minor' was not declared. Should it be static?

In Linux coding style, a literal '0' integer should not be used to represent
a NULL pointer:

drivers/scsi/pmcraid.c:348:29: warning: Using plain integer as NULL pointer
drivers/scsi/pmcraid.c:4824:49: warning: Using plain integer as NULL pointer

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: pmcraid: fix endianess sparse annotations
Arnd Bergmann [Thu, 20 Apr 2017 17:54:47 +0000 (19:54 +0200)]
scsi: pmcraid: fix endianess sparse annotations

The use of le32_to_cpu() etc in this driver looks completely arbitrary.
It may have made sense at some point, but it is not applied consistently,
so this driver presumably won't work on big-endian kernel builds.

Unfortunately it's unclear whether the type names or the calls to
le32_to_cpu() are the correct ones. I'm taking educated guesses here
and assume that most of the __le32 and __le16 annotations are correct,
adding the conversion helpers whereever we access those fields.

The exceptions are the 'fw_version' field that is always accessed as
big-endian, so I'm changing the type here, and the 'hrrq' values that
are accessed as little-endian, so I'm changing those the other way.

None of these changes should have any effect on little-endian
architectures like x86, but it addresses the sparse warnings.

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: pmcraid: use __iomem pointers for ioctl argument
Arnd Bergmann [Thu, 20 Apr 2017 17:54:45 +0000 (19:54 +0200)]
scsi: pmcraid: use __iomem pointers for ioctl argument

kernelci.org reports a new compile warning for old code in the pmcraid
driver:

arch/mips/include/asm/uaccess.h:138:21: warning: passing argument 1 of '__access_ok' makes pointer from integer without a cast [-Wint-conversion]

The warning got introduced by a cleanup to the access_ok() helper that
requires the argument to be a pointer, where the old version silently
accepts 'unsigned long' arguments as it still does on most other
architectures.

The new behavior in MIPS however seems absolutely sensible, and so far I
could only find one other file with the same issue, so the best solution
seems to be to clean up the pmcraid driver.

This makes the driver consistently use 'void __iomem *' pointers for
passing around the address of the user space ioctl arguments, which gets
rid of the kernelci warning as well as several sparse warnings.

Fixes: f0a955f4eeec ("mips: sanitize __access_ok()")
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: pmcraid: fix lock imbalance in pmcraid_reset_reload()
Christoph Hellwig [Sun, 23 Apr 2017 08:33:23 +0000 (10:33 +0200)]
scsi: pmcraid: fix lock imbalance in pmcraid_reset_reload()

sparse found a bug that has always been present since the driver was
merged:

drivers/scsi/pmcraid.c:2353:12: warning: context imbalance in 'pmcraid_reset_reload' - different lock contexts for basic block

Fix this by using a common unlock goto label, and also reduce the
indentation level in the function.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reported-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: sas: move scsi_remove_host call into sas_remove_host
Johannes Thumshirn [Fri, 21 Apr 2017 12:11:41 +0000 (14:11 +0200)]
scsi: sas: move scsi_remove_host call into sas_remove_host

Move scsi_remove_host call into sas_remove_host and remove it from SAS
HBA drivers, so we don't mess up the ordering. This solves an issue with
double deleting sysfs entries that was introduced by the change of sysfs
behaviour from commit bcdde7e221a8 ("sysfs: make __sysfs_remove_dir()
recursive").

[mkp: addressed checkpatch complaints]

Signed-off-by: Johannes Thumshirn <jthumshirn@suse.de>
Suggested-by: Christoph Hellwig <hch@lst.de>
Cc: Hannes Reinecke <hare@suse.de>
Cc: James Bottomley <jejb@linux.vnet.ibm.com>
Cc: Jinpu Wang <jinpu.wang@profitbricks.com>
Cc: John Garry <john.garry@huawei.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jinpu Wang <jinpu.wang@profitbricks.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: BusLogic: fix incorrect spelling of adatper_reset_req
Colin Ian King [Tue, 21 Feb 2017 16:27:11 +0000 (16:27 +0000)]
scsi: BusLogic: fix incorrect spelling of adatper_reset_req

Trivial fix to spelling mistake, adatper_reset_req should be
adapter_reset_req.  Also break up very long seq_printf statement into
multiple lines.

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Acked-by: Khalid Aziz <khalid@gonehiking.org>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: bfa: use designated initializers
Kees Cook [Thu, 20 Apr 2017 22:00:02 +0000 (15:00 -0700)]
scsi: bfa: use designated initializers

Prepare to mark sensitive kernel structures for randomization by making
sure they're using designated initializers. This also initializes the
array members using the enum used to look up __port_action entries.

Signed-off-by: Kees Cook <keescook@chromium.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: aic7xxx: fix order of arguments in function prototype
Colin Ian King [Sat, 8 Apr 2017 17:28:42 +0000 (18:28 +0100)]
scsi: aic7xxx: fix order of arguments in function prototype

The vendor/device and subvendor/subdevice arguments to the function
prototype ahc_9005_subdevinfo_valid are in the wrong order and need to
be swapped to fix this.  Detected with PVS-Studio studio.

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: bfa: remove bfa_module_s madness
Christoph Hellwig [Thu, 13 Apr 2017 08:02:56 +0000 (10:02 +0200)]
scsi: bfa: remove bfa_module_s madness

Just call the functions directly and remove a giant pile of boilerplate
code.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: qedi: qedf: Use designated initializers
Kees Cook [Wed, 29 Mar 2017 20:59:23 +0000 (13:59 -0700)]
scsi: qedi: qedf: Use designated initializers

Prepare to mark sensitive kernel structures for randomization by making
sure they're using designated initializers. These were identified during
allyesconfig builds of x86, arm, and arm64, with most initializer fixes
extracted from grsecurity.

For these cases, terminate the list with { }, which will be zero-filled,
instead of undesignated NULLs.

Signed-off-by: Kees Cook <keescook@chromium.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: storvsc: Add support for FC rport.
Cathy Avery [Mon, 17 Apr 2017 18:37:46 +0000 (14:37 -0400)]
scsi: storvsc: Add support for FC rport.

Included in the current storvsc driver for Hyper-V is the ability to
access luns on an FC fabric via a virtualized fiber channel adapter
exposed by the Hyper-V host. The driver also attaches to the FC
transport to allow host and port names to be published under
/sys/class/fc_host/hostX. Current customer tools running on the VM
require that these names be available in the well known standard
location under fc_host/hostX.

This patch stubs in an rport per fc_host and sets its rport role as
FC_PORT_ROLE_FCP_DUMMY_INITIATOR to indicate to the fc_transport that it
is a pseudo rport in order to scan the scsi stack via echo "- - -" >
/sys/class/scsi_host/hostX/scan.

Signed-off-by: Cathy Avery <cavery@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: scsi_transport_fc: Add dummy initiator role to rport
Cathy Avery [Mon, 17 Apr 2017 18:37:45 +0000 (14:37 -0400)]
scsi: scsi_transport_fc: Add dummy initiator role to rport

This patch allows scsi drivers that expose virturalized fibre channel
devices but that do not expose rports to successfully rescan the scsi
bus via echo "- - -" > /sys/class/scsi_host/hostX/scan.  Drivers can
create a pseudo rport and indicate FC_PORT_ROLE_FCP_DUMMY_INITIATOR as
the rport's role in fc_rport_identifiers. This insures that a valid
scsi_target_id is assigned to the newly created rport and it can meet
the requirements of fc_user_scan_tgt calling scsi_scan_target.

Signed-off-by: Cathy Avery <cavery@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: virtio_scsi: Always try to read VPD pages
David Gibson [Thu, 13 Apr 2017 02:13:00 +0000 (12:13 +1000)]
scsi: virtio_scsi: Always try to read VPD pages

Passed through SCSI targets may have transfer limits which come from the
host SCSI controller or something on the host side other than the target
itself.

To make this work properly, the hypervisor can adjust the target's VPD
information to advertise these limits.  But for that to work, the guest
has to look at the VPD pages, which we won't do by default if it is an
SPC-2 device, even if it does actually support it.

This adds a workaround to address this, forcing devices attached to a
virtio-scsi controller to always check the VPD pages.  This is modelled
on a similar workaround for the storvsc (Hyper-V) SCSI controller,
although that exists for slightly different reasons.

A specific case which causes this is a volume from IBM's IPR RAID
controller (which presents as an SPC-2 device, although it does support
VPD) passed through with qemu's 'scsi-block' device.

[mkp: fixed typo]

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: lpfc: fix potential buffer overflow.
Maurizio Lombardi [Tue, 18 Apr 2017 09:55:51 +0000 (11:55 +0200)]
scsi: lpfc: fix potential buffer overflow.

This patch fixes a potential buffer overflow in lpfc_nvme_info_show().

Signed-off-by: Maurizio Lombardi <mlombard@redhat.com>
Reviewed-by: Ewan D. Milne <emilne@redhat.com>
Acked-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: cxgb4i: update module description
Varun Prakash [Mon, 17 Apr 2017 16:02:41 +0000 (21:32 +0530)]
scsi: cxgb4i: update module description

Signed-off-by: Varun Prakash <varun@chelsio.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: fc: remove redundant check of an unsigned long being less than zero
Colin Ian King [Fri, 14 Apr 2017 13:58:02 +0000 (14:58 +0100)]
scsi: fc: remove redundant check of an unsigned long being less than zero

The check for an unsigned long being less than zero is always false so
it is a redundant check and can be removed.

Detected by static analysis with by PVS-Studio

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Reviewed-by: Tyrel Datwyler <tyreld@linux.vnet.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: ibmvfc: don't check for failure from mempool_alloc()
NeilBrown [Mon, 10 Apr 2017 02:15:13 +0000 (12:15 +1000)]
scsi: ibmvfc: don't check for failure from mempool_alloc()

mempool_alloc() cannot fail when passed GFP_NOIO or any other gfp
setting that is permitted to sleep.  So remove this pointless code.

Signed-off-by: NeilBrown <neilb@suse.com>
Acked-by: Tyrel Datwyler <tyreld@linux.vnet.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: storvsc: remove return at end of void function
Miguel Bernal Marin [Thu, 16 Mar 2017 06:59:57 +0000 (00:59 -0600)]
scsi: storvsc: remove return at end of void function

storvsc_on_channel_callback is a void function and the return
statement at the end is not useful.

Found with checkpatch.

Signed-off-by: Miguel Bernal Marin <miguel.bernal.marin@linux.intel.com>
Reviewed-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: storvsc: Prefer kcalloc over kzalloc with multiply
Miguel Bernal Marin [Thu, 16 Mar 2017 06:58:23 +0000 (00:58 -0600)]
scsi: storvsc: Prefer kcalloc over kzalloc with multiply

Use kcalloc for allocating an array instead of kzalloc with multiply,
kcalloc is the preferred API.

Found with checkpatch.

Signed-off-by: Miguel Bernal Marin <miguel.bernal.marin@linux.intel.com>
Reviewed-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: cxlflash: Introduce hardware queue steering
Matthew R. Ochs [Wed, 12 Apr 2017 19:16:02 +0000 (14:16 -0500)]
scsi: cxlflash: Introduce hardware queue steering

As an enhancement to distribute requests to multiple hardware queues, add the
infrastructure to hash a SCSI command into a particular hardware queue.
Support the following scenarios when deriving which queue to use: single
queue, tagging when SCSI-MQ enabled, and simple hash via CPU ID when SCSI-MQ
is disabled. Rather than altering the existing send API, the derived hardware
queue is stored in the AFU command where it can be used for sending a command
to the chosen hardware queue.

Signed-off-by: Matthew R. Ochs <mrochs@linux.vnet.ibm.com>
Signed-off-by: Uma Krishnan <ukrishn@linux.vnet.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: cxlflash: Add hardware queues attribute
Matthew R. Ochs [Wed, 12 Apr 2017 19:15:53 +0000 (14:15 -0500)]
scsi: cxlflash: Add hardware queues attribute

As staging for supporting multiple hardware queues, add an attribute to show
and set the current number of hardware queues for the host. Support specifying
a hard limit or a CPU affinitized value. This will allow the number of
hardware queues to be tuned by a system administrator.

Signed-off-by: Matthew R. Ochs <mrochs@linux.vnet.ibm.com>
Signed-off-by: Uma Krishnan <ukrishn@linux.vnet.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: cxlflash: Support multiple hardware queues
Uma Krishnan [Wed, 12 Apr 2017 19:15:42 +0000 (14:15 -0500)]
scsi: cxlflash: Support multiple hardware queues

Introduce multiple hardware queues to improve legacy I/O path performance.
Each hardware queue is comprised of a master context and associated I/O
resources. The hardware queues are initially implemented as a static array
embedded in the AFU. This will be transitioned to a dynamic allocation in a
later series to improve the memory footprint of the driver.

Signed-off-by: Uma Krishnan <ukrishn@linux.vnet.ibm.com>
Acked-by: Matthew R. Ochs <mrochs@linux.vnet.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: cxlflash: Improve asynchronous interrupt processing
Matthew R. Ochs [Wed, 12 Apr 2017 19:15:29 +0000 (14:15 -0500)]
scsi: cxlflash: Improve asynchronous interrupt processing

The method used to decode asynchronous interrupts involves unnecessary loops
to match up bits that are set with corresponding entries in the asynchronous
interrupt information table. This algorithm is wasteful and does not scale
well as new status bits are supported.

As an improvement, use the for_each_set_bit() service to iterate over the
asynchronous status bits and refactor the information table such that it can
be indexed by bit position.

Signed-off-by: Matthew R. Ochs <mrochs@linux.vnet.ibm.com>
Signed-off-by: Uma Krishnan <ukrishn@linux.vnet.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: cxlflash: Fix warnings/errors
Matthew R. Ochs [Wed, 12 Apr 2017 19:15:20 +0000 (14:15 -0500)]
scsi: cxlflash: Fix warnings/errors

As a general cleanup, address all reasonable checkpatch warnings and
errors. These include enforcement of comment styles and including named
identifiers in function prototypes.

Signed-off-by: Matthew R. Ochs <mrochs@linux.vnet.ibm.com>
Signed-off-by: Uma Krishnan <ukrishn@linux.vnet.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: cxlflash: Fix power-of-two validations
Matthew R. Ochs [Wed, 12 Apr 2017 19:15:11 +0000 (14:15 -0500)]
scsi: cxlflash: Fix power-of-two validations

Validation statements to enforce assumptions about specific defines are not
being evaluated by the compiler due to the fact that they reside in a routine
that is not used. To activate them, call the routine as part of module
initialization. As an additional, related cleanup, remove the now-defunct
CXLFLASH_NUM_CMDS.

Signed-off-by: Matthew R. Ochs <mrochs@linux.vnet.ibm.com>
Signed-off-by: Uma Krishnan <ukrishn@linux.vnet.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: cxlflash: Remove unnecessary DMA mapping
Matthew R. Ochs [Wed, 12 Apr 2017 19:15:02 +0000 (14:15 -0500)]
scsi: cxlflash: Remove unnecessary DMA mapping

Devices supported by the cxlflash driver are fully coherent and do not require
a bus address mapping. Avoid unnecessary path length by using the virtual
address and length already present in the scatter-gather entry.

Signed-off-by: Matthew R. Ochs <mrochs@linux.vnet.ibm.com>
Signed-off-by: Uma Krishnan <ukrishn@linux.vnet.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: cxlflash: Fence EEH during probe
Matthew R. Ochs [Wed, 12 Apr 2017 19:14:51 +0000 (14:14 -0500)]
scsi: cxlflash: Fence EEH during probe

An EEH during probe can lead to a crash as the recovery thread races with the
probe thread. To avoid this issue, introduce new states to fence out EEH
recovery until probe has completed. Also ensure the reset wait queue is
flushed during device removal to avoid orphaned threads.

Signed-off-by: Matthew R. Ochs <mrochs@linux.vnet.ibm.com>
Signed-off-by: Uma Krishnan <ukrishn@linux.vnet.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: cxlflash: Support up to 4 ports
Matthew R. Ochs [Wed, 12 Apr 2017 19:14:41 +0000 (14:14 -0500)]
scsi: cxlflash: Support up to 4 ports

Update the driver to allow for future cards with 4 ports.

Signed-off-by: Matthew R. Ochs <mrochs@linux.vnet.ibm.com>
Signed-off-by: Uma Krishnan <ukrishn@linux.vnet.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: cxlflash: SISlite updates to support 4 ports
Matthew R. Ochs [Wed, 12 Apr 2017 19:14:28 +0000 (14:14 -0500)]
scsi: cxlflash: SISlite updates to support 4 ports

Update the SISlite header to support 4 ports as outlined in the SISlite
specification. Address fallout from structure renames and refreshed
organization throughout the driver. Determine the number of ports supported by
a card from the global port selection mask register reset value.

Signed-off-by: Matthew R. Ochs <mrochs@linux.vnet.ibm.com>
Signed-off-by: Uma Krishnan <ukrishn@linux.vnet.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: cxlflash: Hide FC internals behind common access routine
Matthew R. Ochs [Wed, 12 Apr 2017 19:14:17 +0000 (14:14 -0500)]
scsi: cxlflash: Hide FC internals behind common access routine

As staging to support FC-related updates to the SISlite specification,
introduce helper routines to obtain references to FC resources that exist
within the global map. This will allow changes to the underlying global map
structure without impacting existing code paths.

Signed-off-by: Matthew R. Ochs <mrochs@linux.vnet.ibm.com>
Signed-off-by: Uma Krishnan <ukrishn@linux.vnet.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: cxlflash: Remove port configuration assumptions
Matthew R. Ochs [Wed, 12 Apr 2017 19:14:05 +0000 (14:14 -0500)]
scsi: cxlflash: Remove port configuration assumptions

At present, the cxlflash driver only supports hardware with two FC ports. The
code was initially designed with this assumption and is dependent on having
two FC ports - adding more ports will break logic within the driver.

To mitigate this issue, remove the existing port assumptions and transition
the code to support more than two ports. As a side effect, clarify the
interpretation of the DK_CXLFLASH_ALL_PORTS_ACTIVE flag.

Signed-off-by: Matthew R. Ochs <mrochs@linux.vnet.ibm.com>
Signed-off-by: Uma Krishnan <ukrishn@linux.vnet.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: cxlflash: Support dynamic number of FC ports
Matthew R. Ochs [Wed, 12 Apr 2017 19:13:50 +0000 (14:13 -0500)]
scsi: cxlflash: Support dynamic number of FC ports

Transition from a static number of FC ports to a value that is derived during
probe. For now, a static value is used but this will later be based on the
type of card being configured.

Signed-off-by: Matthew R. Ochs <mrochs@linux.vnet.ibm.com>
Signed-off-by: Uma Krishnan <ukrishn@linux.vnet.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: cxlflash: Update sysfs helper routines to pass config structure
Matthew R. Ochs [Wed, 12 Apr 2017 19:13:34 +0000 (14:13 -0500)]
scsi: cxlflash: Update sysfs helper routines to pass config structure

As staging for future function, pass the config pointer instead of the AFU
pointer for port-related sysfs helper routines.

Signed-off-by: Matthew R. Ochs <mrochs@linux.vnet.ibm.com>
Signed-off-by: Uma Krishnan <ukrishn@linux.vnet.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: cxlflash: Implement IRQ polling for RRQ processing
Matthew R. Ochs [Wed, 12 Apr 2017 19:13:20 +0000 (14:13 -0500)]
scsi: cxlflash: Implement IRQ polling for RRQ processing

Currently, RRQ processing takes place on hardware interrupt context. This can
be a heavy burden in some environments due to the overhead encountered while
completing RRQ entries. In an effort to improve system performance, use the
IRQ polling API to schedule this processing on softirq context.

This function will be disabled by default until starting values can be
established for the hardware supported by this driver.

Signed-off-by: Matthew R. Ochs <mrochs@linux.vnet.ibm.com>
Signed-off-by: Uma Krishnan <ukrishn@linux.vnet.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: cxlflash: Serialize RRQ access and support offlevel processing
Matthew R. Ochs [Wed, 12 Apr 2017 19:12:55 +0000 (14:12 -0500)]
scsi: cxlflash: Serialize RRQ access and support offlevel processing

As further staging to support processing the HRRQ by other means, access to
the HRRQ needs to be serialized by a disabled lock. This will allow safe
access in other non-hardware interrupt contexts. In an effort to minimize the
period where interrupts are disabled, support is added to queue up commands
harvested from the RRQ such that they can be processed with hardware
interrupts enabled. While this doesn't offer any improvement with processing
on a hardware interrupt it will help when IRQ polling is supported and the
command completions can execute on softirq context.

Signed-off-by: Matthew R. Ochs <mrochs@linux.vnet.ibm.com>
Signed-off-by: Uma Krishnan <ukrishn@linux.vnet.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: cxlflash: Separate RRQ processing from the RRQ interrupt handler
Matthew R. Ochs [Wed, 12 Apr 2017 19:11:44 +0000 (14:11 -0500)]
scsi: cxlflash: Separate RRQ processing from the RRQ interrupt handler

In order to support processing the HRRQ by other means (e.g. polling), the
processing portion of the current RRQ interrupt handler needs to be broken out
into a separate routine. This will allow RRQ processing from places other than
the RRQ hardware interrupt handler.

Signed-off-by: Matthew R. Ochs <mrochs@linux.vnet.ibm.com>
Signed-off-by: Uma Krishnan <ukrishn@linux.vnet.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: snic: fix spelling mistake: "Cann't" -> "Cannot"
Colin Ian King [Wed, 12 Apr 2017 14:12:35 +0000 (15:12 +0100)]
scsi: snic: fix spelling mistake: "Cann't" -> "Cannot"

Trivial fix to spelling mistake in SNIC_ERR error message text, one
cannot have "Cann't".

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: hisi_sas: controller reset for multi-bits ECC and AXI fatal errors
Xiang Chen [Mon, 10 Apr 2017 13:22:01 +0000 (21:22 +0800)]
scsi: hisi_sas: controller reset for multi-bits ECC and AXI fatal errors

For 1 bit ECC errors, those errors can be recovered by hw. But for
multi-bits ECC and AXI errors, there are something wrong with whole
module or system, so try reset the controller to recover those errors
instead of calling panic().

Signed-off-by: Xiang Chen <chenxiang66@hisilicon.com>
Signed-off-by: John Garry <john.garry@huawei.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: hisi_sas: fix NULL deference when TMF timeouts
John Garry [Mon, 10 Apr 2017 13:22:00 +0000 (21:22 +0800)]
scsi: hisi_sas: fix NULL deference when TMF timeouts

If a TMF timeouts (maybe due to unlikely scenario of an expander being
unplugged when TMF for remote device is active), when we eventually try
to free the slot, we crash as we dereference the slot's task, which has
already been released.

As a fix, add checks in the slot release code for a NULL task.

Signed-off-by: John Garry <john.garry@huawei.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: hisi_sas: add v2 hw internal abort timeout workaround
John Garry [Mon, 10 Apr 2017 13:21:59 +0000 (21:21 +0800)]
scsi: hisi_sas: add v2 hw internal abort timeout workaround

This patch is a workaround for a SoC bug where an internal abort command
may timeout. In v2 hw, the channel should become idle in order to finish
abort process. If the target side has been sending HOLD, host side
channel failed to complete the frame to send, and can not enter the idle
state. Then internal abort command will timeout.

As this issue is only in v2 hw, we deal with it in the hw layer.  Our
workaround solution is: If abort is not finished within a certain period
of time, we will check HOLD status. If HOLD has been sending, we will
send break command.

Signed-off-by: John Garry <john.garry@huawei.com>
Signed-off-by: Xiaofei Tan <tanxiaofei@huawei.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: hisi_sas: workaround SoC about abort timeout bug
Xiaofei Tan [Mon, 10 Apr 2017 13:21:58 +0000 (21:21 +0800)]
scsi: hisi_sas: workaround SoC about abort timeout bug

This patch adds a workaround solution for a SoC bug which may cause SoC
logic fatal error when disabling a PHY.  Then we find internal abort IO
timeout may occur, and the controller IO breakpoint may be corrupted.

We work around this SoC bug by optimizing the flow of disabling a PHY.

Signed-off-by: Xiaofei Tan <tanxiaofei@huawei.com>
Signed-off-by: John Garry <john.garry@huawei.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: hisi_sas: workaround a SoC SATA IO processing bug
Xiaofei Tan [Mon, 10 Apr 2017 13:21:57 +0000 (21:21 +0800)]
scsi: hisi_sas: workaround a SoC SATA IO processing bug

This patch provides a workaround a SoC bug where SATA IPTTs for
different devices may conflict.

The workaround solution requests the following:
1. SATA device id must be even and not equal to SAS IPTT.
2. SATA device can not share the same IPTT with other SAS or
   SATA device.

Besides we shall consider IPTT value 0 is reserved for another SoC bug
(STP device open link at firstly after SAS controller reset).

To sum up, the solution is: Each SATA device uses independent and
continuous 32 even IPTT from 64 to 4094, then v2 hw can only support 63
SATA devices.  All SAS device(SSP/SMP devices) share odd IPTT value from
1 to 4095.

Signed-off-by: Xiaofei Tan <tanxiaofei@huawei.com>
Signed-off-by: John Garry <john.garry@huawei.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: hisi_sas: workaround STP link SoC bug
Xiaofei Tan [Mon, 10 Apr 2017 13:21:56 +0000 (21:21 +0800)]
scsi: hisi_sas: workaround STP link SoC bug

After resetting the controller, the process of scanning SATA disks
attached to an expander may fail occasionally. The issue is that the
controller can't close the STP link created by target if the max link
time is 0.

To workaround this issue, we reject STP link after resetting the
controller, and change the corresponding PHY to accept STP link only
after receiving data.

We do this check in cq interrupt handler. In order not to reduce
efficiency, we use an variable to control whether we should check and
change PHY to accept STP link.

The function phys_reject_stp_links_v2_hw() should be called after
resetting the controller.

The solution of another SoC bug "SATA IO timeout", that also uses the
same register to control STP link, is not effective before the PHY
accepts STP link.

Signed-off-by: Xiaofei Tan <tanxiaofei@huawei.com>
Signed-off-by: John Garry <john.garry@huawei.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: libfc: directly call ELS request handlers
Johannes Thumshirn [Fri, 7 Apr 2017 13:42:24 +0000 (15:42 +0200)]
scsi: libfc: directly call ELS request handlers

Directly call ELS request handler functions in fc_lport_recv_els_req
instead of saving the pointer to the handler's receive function and then
later dereferencing this pointer.

This makes the code a bit more obvious.

Signed-off-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: Chad Dupuis <chad.dupuis@cavium.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: sg: close race condition in sg_remove_sfp_usercontext()
Hannes Reinecke [Fri, 7 Apr 2017 07:34:17 +0000 (09:34 +0200)]
scsi: sg: close race condition in sg_remove_sfp_usercontext()

sg_remove_sfp_usercontext() is clearing any sg requests, but needs to
take 'rq_list_lock' when modifying the list.

Reported-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Hannes Reinecke <hare@suse.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Tested-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: sg: use standard lists for sg_requests
Hannes Reinecke [Fri, 7 Apr 2017 07:34:16 +0000 (09:34 +0200)]
scsi: sg: use standard lists for sg_requests

'Sg_request' is using a private list implementation; convert it to
standard lists.

Signed-off-by: Hannes Reinecke <hare@suse.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Tested-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: sg: check for valid direction before starting the request
Johannes Thumshirn [Fri, 7 Apr 2017 07:34:15 +0000 (09:34 +0200)]
scsi: sg: check for valid direction before starting the request

Check for a valid direction before starting the request, otherwise we
risk running into an assertion in the scsi midlayer checking for valid
requests.

[mkp: fixed typo]

Signed-off-by: Johannes Thumshirn <jthumshirn@suse.de>
Link: http://www.spinics.net/lists/linux-scsi/msg104400.html
Reported-by: Dmitry Vyukov <dvyukov@google.com>
Signed-off-by: Hannes Reinecke <hare@suse.com>
Tested-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: sg: protect accesses to 'reserved' page array
Hannes Reinecke [Fri, 7 Apr 2017 07:34:14 +0000 (09:34 +0200)]
scsi: sg: protect accesses to 'reserved' page array

The 'reserved' page array is used as a short-cut for mapping data,
saving us to allocate pages per request. However, the 'reserved' array
is only capable of holding one request, so this patch introduces a mutex
for protect 'sg_fd' against concurrent accesses.

Signed-off-by: Hannes Reinecke <hare@suse.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Tested-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: sg: remove 'save_scat_len'
Hannes Reinecke [Fri, 7 Apr 2017 07:34:13 +0000 (09:34 +0200)]
scsi: sg: remove 'save_scat_len'

Unused.

Signed-off-by: Hannes Reinecke <hare@suse.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Tested-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: sg: disable SET_FORCE_LOW_DMA
Hannes Reinecke [Fri, 7 Apr 2017 07:34:12 +0000 (09:34 +0200)]
scsi: sg: disable SET_FORCE_LOW_DMA

The ioctl SET_FORCE_LOW_DMA has never worked since the initial git
check-in, and the respective setting is nowadays handled correctly. So
disable it entirely.

Signed-off-by: Hannes Reinecke <hare@suse.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Tested-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: qla2xxx: remove some redundant pointer assignments
Colin Ian King [Thu, 6 Apr 2017 11:19:57 +0000 (12:19 +0100)]
scsi: qla2xxx: remove some redundant pointer assignments

There are several local or function parameter pointers that are being
assigned NULL after a kfree where and these have no effect and hence can
be removed.

Fixes various cppcheck warnings:

"Assignment of function parameter has no effect outside the
function. Did you forget dereferencing it"

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Acked-by: Himanshu Madhani <himanshu.madhani@cavium.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: qla4xxx: drop redundant init_completion
Nicholas Mc Guire [Wed, 22 Mar 2017 10:44:22 +0000 (11:44 +0100)]
scsi: qla4xxx: drop redundant init_completion

The redundant init_completion() here seems to be a cut&past error as
struct scsi_qla_host only has 4 completion elements to initialize, thus
the duplicate init_completion(disable_acb_comp) is simply removed.

Signed-off-by: Nicholas Mc Guire <der.herr@hofr.at>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: make asynchronous aborts mandatory
Hannes Reinecke [Thu, 6 Apr 2017 13:36:35 +0000 (15:36 +0200)]
scsi: make asynchronous aborts mandatory

There hasn't been any reports for HBAs where asynchronous abort
would not work, so we should make it mandatory and remove
the fallback.

Signed-off-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: Bart Van Assche <bart.vanassche@sandisk.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: make scsi_eh_scmd_add() always succeed
Hannes Reinecke [Thu, 6 Apr 2017 13:36:34 +0000 (15:36 +0200)]
scsi: make scsi_eh_scmd_add() always succeed

scsi_eh_scmd_add() currently only will fail if no
error handler thread is started (which will never be the
case) or if the state machine encounters an illegal transition.

But if we're encountering an invalid state transition
chances is we cannot fixup things with the error handler.
So better add a WARN_ON for illegal host states and
make scsi_dh_scmd_add() a void function.

Signed-off-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: Bart Van Assche <bart.vanassche@sandisk.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: make eh_eflags persistent
Hannes Reinecke [Thu, 6 Apr 2017 13:36:33 +0000 (15:36 +0200)]
scsi: make eh_eflags persistent

If a failed command is retried and fails again we need
to enter SCSI EH, otherwise we will never be able to
recover the command.
To detect this situation we must not clear scmd->eh_eflags
when EH finishes but rather make it persistent throughout
the lifetime of the command.

Signed-off-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Benjamin Block <bblock@linux.vnet.ibm.com>
Reviewed-by: Bart Van Assche <bart.vanassche@sandisk.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: libsas: allow async aborts
Christoph Hellwig [Thu, 6 Apr 2017 13:36:32 +0000 (15:36 +0200)]
scsi: libsas: allow async aborts

We now first try to call ->eh_abort_handler from a work queue, but libsas
was always failing that for no good reason.  Allow async aborts.

Reviewed-by: Johannes Thumshirn <jth@kernel.org>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: always send command aborts
Hannes Reinecke [Thu, 6 Apr 2017 13:36:31 +0000 (15:36 +0200)]
scsi: always send command aborts

When a command has timed out we always should be sending an
abort; with the previous code a failed abort might signal
SCSI EH to start, and all other timed out commands will
never be aborted, even though they might belong to a
different ITL nexus.

Cc: Benjamin Block <bblock@linux.vnet.ibm.com>
Signed-off-by: Hannes Reinecke <hare@suse.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: sd: Return SUCCESS in sd_eh_action() after device offline
Hannes Reinecke [Thu, 6 Apr 2017 13:36:30 +0000 (15:36 +0200)]
scsi: sd: Return SUCCESS in sd_eh_action() after device offline

If sd_eh_action() decides to take the device offline there is
no point in returning FAILED, as taking the device offline
is the ultimate step in SCSI EH anyway.
So further escalation via SCSI EH is not likely to make a
difference and we can as well return SUCCESS.

Cc: Benjamin Block <bblock@linux.vnet.ibm.com>
Signed-off-by: Hannes Reinecke <hare@suse.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: scsi_error: count medium access timeout only once per EH run
Hannes Reinecke [Thu, 6 Apr 2017 13:36:29 +0000 (15:36 +0200)]
scsi: scsi_error: count medium access timeout only once per EH run

The current medium access timeout counter will be increased for
each command, so if there are enough failed commands we'll hit
the medium access timeout for even a single device failure and
the following kernel message is displayed:

sd H:C:T:L: [sdXY] Medium access timeout failure. Offlining disk!

Fix this by making the timeout per EH run, ie the counter will
only be increased once per device and EH run.

Fixes: 18a4d0a ("[SCSI] Handle disk devices which can not process medium access commands")
Cc: Ewan Milne <emilne@redhat.com>
Cc: Lawrence Obermann <loberman@redhat.com>
Cc: Benjamin Block <bblock@linux.vnet.ibm.com>
Cc: Steffen Maier <maier@linux.vnet.ibm.com>
Signed-off-by: Hannes Reinecke <hare@suse.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: csiostor: switch to pci_alloc_irq_vectors
Christoph Hellwig [Thu, 12 Jan 2017 10:17:29 +0000 (11:17 +0100)]
scsi: csiostor: switch to pci_alloc_irq_vectors

And get automatic MSI-X affinity for free.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Acked-by: Varun Prakash <varun@chelsio.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: ses: don't get power status of SES device slot on probe
Mauricio Faria de Oliveira [Wed, 5 Apr 2017 15:18:19 +0000 (12:18 -0300)]
scsi: ses: don't get power status of SES device slot on probe

The commit 08024885a2a3 ("ses: Add power_status to SES device slot")
introduced the 'power_status' attribute to enclosure components and
the associated callbacks.

There are 2 callbacks available to get the power status of a device:
1) ses_get_power_status() for 'struct enclosure_component_callbacks'
2) get_component_power_status() for the sysfs device attribute
(these are available for kernel-space and user-space, respectively.)

However, despite both methods being available to get power status
on demand, that commit also introduced a call to get power status
in ses_enclosure_data_process().

This dramatically increased the total probe time for SCSI devices
on larger configurations, because ses_enclosure_data_process() is
called several times during the SCSI devices probe and loops over
the component devices (but that is another problem, another patch).

That results in a tremendous continuous hammering of SCSI Receive
Diagnostics commands to the enclosure-services device, which does
delay the total probe time for the SCSI devices __significantly__:

  Originally, ~34 minutes on a system attached to ~170 disks:

    [ 9214.490703] mpt3sas version 13.100.00.00 loaded
    ...
    [11256.580231] scsi 17:0:177:0: qdepth(16), tagged(1), simple(0),
                   ordered(0), scsi_level(6), cmd_que(1)

  With this patch, it decreased to ~2.5 minutes -- a 13.6x faster

    [ 1002.992533] mpt3sas version 13.100.00.00 loaded
    ...
    [ 1151.978831] scsi 11:0:177:0: qdepth(16), tagged(1), simple(0),
                   ordered(0), scsi_level(6), cmd_que(1)

Back to the commit discussion.. on the ses_get_power_status() call
introduced in ses_enclosure_data_process(): impact of removing it.

That may possibly be in place to initialize the power status value
on device probe.  However, those 2 functions available to retrieve
that value _do_ automatically refresh/update it.  So the potential
benefit would be a direct access of the 'power_status' field which
does not use the callbacks...

But the only reader of 'struct enclosure_component::power_status'
is the get_component_power_status() callback for sysfs attribute,
and it _does_ check for and call the .get_power_status callback,
(which indeed is defined and implemented by that commit), so the
power status value is, again, automatically updated.

So, the remaining potential for a direct/non-callback access to
the power_status attribute would be out-of-tree modules -- well,
for those, if they are for whatever reason interested in values
that are set during device probe and not up-to-date by the time
they need it.. well, that would be curious.

Well, to handle that more properly, set the initial power state
value to '-1' (i.e., uninitialized) instead of '1' (power 'on'),
and check for it in that callback which may do an direct access
to the field value _if_ a callback function is not defined.

Signed-off-by: Mauricio Faria de Oliveira <mauricfo@linux.vnet.ibm.com>
Fixes: 08024885a2a3 ("ses: Add power_status to SES device slot")
Reviewed-by: Dan Williams <dan.j.williams@intel.com>
Reviewed-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: Make checking the scsi_device_get() return value mandatory
Bart Van Assche [Wed, 5 Apr 2017 16:52:50 +0000 (09:52 -0700)]
scsi: Make checking the scsi_device_get() return value mandatory

Now that all scsi_device_get() callers check the return value of this
function, make checking that return value mandatory.

Signed-off-by: Bart Van Assche <bart.vanassche@sandisk.com>
Cc: Hannes Reinecke <hare@suse.de>
Cc: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: osd_uld: Check scsi_device_get() return value
Bart Van Assche [Thu, 30 Mar 2017 17:17:17 +0000 (17:17 +0000)]
scsi: osd_uld: Check scsi_device_get() return value

scsi_device_get() can fail. Hence check its return value.

Signed-off-by: Bart Van Assche <bart.vanassche@sandisk.com>
Cc: Boaz Harrosh <bharrosh@panasas.com>
Acked-by: Boaz Harrosh <ooo@electrozaur.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: sas: remove sas_domain_release_transport
Johannes Thumshirn [Mon, 3 Apr 2017 14:32:50 +0000 (16:32 +0200)]
scsi: sas: remove sas_domain_release_transport

sas_domain_release_transport is unused since at least v3.13, remove it.

Signed-off-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: qla2xxx: Fix typo in driver
Milan P Gandhi [Fri, 31 Mar 2017 21:37:04 +0000 (14:37 -0700)]
scsi: qla2xxx: Fix typo in driver

Signed-off-by: Milan P Gandhi <mgandhi@redhat.com>
Signed-off-by: Himanshu Madhani <himanshu.madhani@cavium.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: advansys: fix uninitialized data access
Arnd Bergmann [Thu, 23 Mar 2017 15:02:18 +0000 (16:02 +0100)]
scsi: advansys: fix uninitialized data access

gcc-7.0.1 now warns about a previously unnoticed access of uninitialized
struct members:

drivers/scsi/advansys.c: In function 'AscMsgOutSDTR':
drivers/scsi/advansys.c:3860:26: error: '*((void *)&sdtr_buf+5)' may be used uninitialized in this function [-Werror=maybe-uninitialized]
         ((ushort)s_buffer[i + 1] << 8) | s_buffer[i]);
                          ^
drivers/scsi/advansys.c:3860:26: error: '*((void *)&sdtr_buf+7)' may be used uninitialized in this function [-Werror=maybe-uninitialized]
drivers/scsi/advansys.c:3860:26: error: '*((void *)&sdtr_buf+5)' may be used uninitialized in this function [-Werror=maybe-uninitialized]
drivers/scsi/advansys.c:3860:26: error: '*((void *)&sdtr_buf+7)' may be used uninitialized in this function [-Werror=maybe-uninitialized]

The code has existed in this exact form at least since v2.6.12, and the
warning seems correct. This uses named initializers to ensure we
initialize all members of the structure.

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: be2iscsi: switch to pci_alloc_irq_vectors
Christoph Hellwig [Fri, 13 Jan 2017 16:29:48 +0000 (17:29 +0100)]
scsi: be2iscsi: switch to pci_alloc_irq_vectors

And get automatic MSI-X affinity for free.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Acked-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoRevert "scsi: ufs: add queries retry mechanism"
Szymon Mielczarek [Wed, 29 Mar 2017 06:19:21 +0000 (08:19 +0200)]
Revert "scsi: ufs: add queries retry mechanism"

This reverts commit 61e073590b82a539654626ecae91b8fab11db3f3.

The patch introduced redundant query retries as we already had such
mechanism provided with _retry functions.  Both ufshcd_read_desc and
ufshcd_read_unit_desc_param functions call ufshcd_query_descriptor_retry
wrapper.

Signed-off-by: Szymon Mielczarek <szymonx.mielczarek@intel.com>
Reviewed-by: Subhash Jadavani <subhashj@codeaurora.org>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: hpsa: change driver version
Don Brace [Tue, 28 Mar 2017 21:40:20 +0000 (16:40 -0500)]
scsi: hpsa: change driver version

Reviewed-by: Gerry Morong <gerry.morong@microsemi.com>
Reviewed-by: Scott Teel <scott.teel@microsemi.com>
Signed-off-by: Don Brace <don.brace@microsemi.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: hpsa: update pci ids
Don Brace [Tue, 28 Mar 2017 21:40:13 +0000 (16:40 -0500)]
scsi: hpsa: update pci ids

Reviewed-by: Gerry Morong <gerry.morong@microsemi.com>
Reviewed-by: Scott Teel <scott.teel@microsemi.com>
Signed-off-by: Don Brace <don.brace@microsemi.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: hisi_sas: fix SATA dependency
Arnd Bergmann [Tue, 28 Mar 2017 14:22:03 +0000 (16:22 +0200)]
scsi: hisi_sas: fix SATA dependency

Removing the 'select SCSI_SAS_LIBSAS' statement in Kconfig resulted in a
link failure in configurations that have hisi_sas built-in but libsas as
a loadable module:

drivers/scsi/built-in.o: In function `hisi_sas_scan_finished':
hisi_sas_main.c:(.text+0x37ce9): undefined reference to `sas_drain_work'
drivers/scsi/built-in.o: In function `hisi_sas_slave_configure':
hisi_sas_main.c:(.text+0x37d17): undefined reference to `sas_slave_configure'
hisi_sas_main.c:(.text+0x37d40): undefined reference to `sas_change_queue_depth'
drivers/scsi/built-in.o: In function `hisi_sas_remove':

All other libsas users have the 'select' statement, so we should do the
same here for consistency. For all I can tell, the patch that added the
sata softreset does not actually introduce a dependency on SCSI_SAS_ATA
but instead adds calls into libata itself, so we can express that with a
more specific dependency.

We cannot have 'select SCSI_SAS_LIBSAS; depends on SCSI_SAS_ATA' as that
would cause a dependency loop.

Fixes: 7c594f0407de ("scsi: hisi_sas: add softreset function for SATA disk")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Acked-by: John Garry <john.garry@huawei.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: ufs: just use sizeof() for snprintf()
Tomohiro Kusumi [Tue, 28 Mar 2017 13:49:29 +0000 (16:49 +0300)]
scsi: ufs: just use sizeof() for snprintf()

Not much reason to use ARRAY_SIZE() when we know it's for a C string.

Signed-off-by: Tomohiro Kusumi <tkusumi@tuxera.com>
Reviewed-by: Subhash Jadavani <subhashj@codeaurora.org>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: ufs: remove deprecated enum for hw interrupt
Tomohiro Kusumi [Tue, 28 Mar 2017 13:49:28 +0000 (16:49 +0300)]
scsi: ufs: remove deprecated enum for hw interrupt

These flags are no longer needed after 2fbd009b in 2013.

Signed-off-by: Tomohiro Kusumi <tkusumi@tuxera.com>
Reviewed-by: Subhash Jadavani <subhashj@codeaurora.org>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: ufs: add missing macros for register bits from UFSHCI spec
Tomohiro Kusumi [Tue, 28 Mar 2017 13:49:27 +0000 (16:49 +0300)]
scsi: ufs: add missing macros for register bits from UFSHCI spec

Add macros for register bits that can be found in JESD223C (v2.1).

Not all registers are defined in ufshci.h (i.e. some are unused whether
macros are defined or undefined), but all the bits for those registers
that are already defined should appear here.

Signed-off-by: Tomohiro Kusumi <tkusumi@tuxera.com>
Reviewed-by: Subhash Jadavani <subhashj@codeaurora.org>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: ufs: non functional macro fix
Tomohiro Kusumi [Tue, 28 Mar 2017 13:49:26 +0000 (16:49 +0300)]
scsi: ufs: non functional macro fix

Not having () isn't likely to do any harm in this case, but all the
other macros below do have it. Also add "are" in a comment.

Signed-off-by: Tomohiro Kusumi <tkusumi@tuxera.com>
Reviewed-by: Subhash Jadavani <subhashj@codeaurora.org>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: ufs: use existing macro CONTROLLER_ENABLE to test register bit
Tomohiro Kusumi [Tue, 28 Mar 2017 13:49:25 +0000 (16:49 +0300)]
scsi: ufs: use existing macro CONTROLLER_ENABLE to test register bit

Signed-off-by: Tomohiro Kusumi <tkusumi@tuxera.com>
Reviewed-by: Subhash Jadavani <subhashj@codeaurora.org>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: ufs: make ufshcd_is_{device_present, hba_active}() return bool
Tomohiro Kusumi [Tue, 28 Mar 2017 13:49:24 +0000 (16:49 +0300)]
scsi: ufs: make ufshcd_is_{device_present, hba_active}() return bool

ufshcd driver generally uses bool for is_xxx type things instead of int,
so conform to its style.

Signed-off-by: Tomohiro Kusumi <tkusumi@tuxera.com>
Reviewed-by: Subhash Jadavani <subhashj@codeaurora.org>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: hisi_sas: add missing break in switch statement
Colin Ian King [Tue, 28 Mar 2017 11:12:22 +0000 (12:12 +0100)]
scsi: hisi_sas: add missing break in switch statement

It appears that a break in the TRANS_TX_OPEN_CNX_ERR_NO_DESTINATION case
got accidentally removed in an earlier commit, as it stands, the
ts->stat and ts->open_rej_reason are being updated twice for this case
which looks incorrect.  Fix this by adding in the missing break
statement.

Detected by CoverityScan, CID#1422110 ("Missing break in switch")

Fixes: 634a9585f49c7 ("scsi: hisi_sas: process error codes according to their priority")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Acked-by: John Garry <john.garry@huawei.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoMAINTAINERS: remove pmchba list for PM8001
Jack Wang [Fri, 24 Mar 2017 09:49:52 +0000 (10:49 +0100)]
MAINTAINERS: remove pmchba list for PM8001

The email address is undeliverable for some time now, so just remove it.

Signed-off-by: Jack Wang <jinpu.wang@profitbricks.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: be2iscsi: Update driver version
Jitendra Bhivare [Fri, 24 Mar 2017 08:41:49 +0000 (14:11 +0530)]
scsi: be2iscsi: Update driver version

Version 11.4.0.0

Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Reviewed-by: Tomas Henzl <thenzl@redhat.com>
Reviewed-by: Chris Leech <cleech@redhat.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: be2iscsi: Update Copyright
Jitendra Bhivare [Fri, 24 Mar 2017 08:41:48 +0000 (14:11 +0530)]
scsi: be2iscsi: Update Copyright

Update Broadcom Copyright markings in all files.

Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: be2iscsi: Check size before copying ASYNC handle
Jitendra Bhivare [Fri, 24 Mar 2017 08:41:47 +0000 (14:11 +0530)]
scsi: be2iscsi: Check size before copying ASYNC handle

Data in buffers are gathered into a single buffer before giving to iSCSI
layer. Though less likely to have payload more than 8K in ASYNC PDU, the
data length is provide by FW and check is missing for overrun.

Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Reviewed-by: Tomas Henzl <thenzl@redhat.com>
Reviewed-by: Chris Leech <cleech@redhat.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: be2iscsi: Remove free_list for ASYNC handles
Jitendra Bhivare [Fri, 24 Mar 2017 08:41:46 +0000 (14:11 +0530)]
scsi: be2iscsi: Remove free_list for ASYNC handles

With previous patch adding ASYNC Rx buffers to free_list is not
required.  Remove all free_list related operations.

Add in_use to track if buffer posted is being processed by driver and
purge all buffers received for connection if found so.

Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Reviewed-by: Tomas Henzl <thenzl@redhat.com>
Reviewed-by: Chris Leech <cleech@redhat.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: be2iscsi: Use num_cons field in Rx CQE
Jitendra Bhivare [Fri, 24 Mar 2017 08:41:45 +0000 (14:11 +0530)]
scsi: be2iscsi: Use num_cons field in Rx CQE

FW runs out of buffer if buffers are not posted back soon.  ASYNC Rx CQE
indicates that FW has consumed 8 RQEs.  Use it to post back buffers
instead of waiting for buffers to be processed and freed by driver.

Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Reviewed-by: Tomas Henzl <thenzl@redhat.com>
Reviewed-by: Chris Leech <cleech@redhat.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: be2iscsi: Increase HDQ default queue size
Jitendra Bhivare [Fri, 24 Mar 2017 08:41:44 +0000 (14:11 +0530)]
scsi: be2iscsi: Increase HDQ default queue size

Currently, ASYNC PDU default queue size is set to max connections.  This
leaves only one buffer per connection for any ASYNC PDUs from targets.

Double the size of the default queue.

Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Reviewed-by: Tomas Henzl <thenzl@redhat.com>
Reviewed-by: Chris Leech <cleech@redhat.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: scsi_transport_iscsi: Use flush_work in iscsi_remove_session
Jitendra Bhivare [Fri, 24 Mar 2017 08:41:43 +0000 (14:11 +0530)]
scsi: scsi_transport_iscsi: Use flush_work in iscsi_remove_session

scsi_flush_work flushes workqueue for the Scsi_Host.  In iSCSI offload
enabled host, this would wait for all other sessions under the host.

Use flush_work for the session being removed instead.

Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Reviewed-by: Tomas Henzl <thenzl@redhat.com>
Reviewed-by: Chris Leech <cleech@redhat.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: be2iscsi: Replace spin_unlock_bh with spin_lock
Jitendra Bhivare [Fri, 24 Mar 2017 08:41:42 +0000 (14:11 +0530)]
scsi: be2iscsi: Replace spin_unlock_bh with spin_lock

spin_unlock_bh back_lock is used in beiscsi_eh_device_reset instead of
spin_lock.

Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Reviewed-by: Tomas Henzl <thenzl@redhat.com>
Reviewed-by: Chris Leech <cleech@redhat.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: be2iscsi: Fix closing of connection
Jitendra Bhivare [Fri, 24 Mar 2017 08:41:41 +0000 (14:11 +0530)]
scsi: be2iscsi: Fix closing of connection

CID needs to be freed even when invalidate or upload connection fails.
Attempt to close connection 3 times before freeing CID.

Set cleanup_type to INVALIDATE instead of force TCP_RST.  This
unnecessarily is terminating connection with reset instead of gracefully
closing it.

Set save_cfg to 0 - session not to be saved on flash.

Add delay and process CQ before uploading connection.

Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Reviewed-by: Tomas Henzl <thenzl@redhat.com>
Reviewed-by: Chris Leech <cleech@redhat.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: be2iscsi: Check tag in beiscsi_mccq_compl_wait
Jitendra Bhivare [Fri, 24 Mar 2017 08:41:40 +0000 (14:11 +0530)]
scsi: be2iscsi: Check tag in beiscsi_mccq_compl_wait

scsi host12: BS_1377 : mgmt_invalidate_connection Failed for cid=256
BUG: unable to handle kernel NULL pointer dereference at 0000000000000008
IP: [<ffffffff81332ebf>] __list_add+0xf/0xc0
PGD 0
Oops: 0000 [#1] SMP
Modules linked in:
...
CPU: 9 PID: 1542 Comm: iscsid Tainted: G               ------------ T 3.10.0-514.el7.x86_64 #1
Hardware name: HP ProLiant DL360 Gen9/ProLiant DL360 Gen9, BIOS P89 09/12/2016
task: ffff88076f310fb0 ti: ffff88076bba8000 task.ti: ffff88076bba8000
RIP: 0010:[<ffffffff81332ebf>]  [<ffffffff81332ebf>] __list_add+0xf/0xc0
RSP: 0018:ffff88076bbab8e8  EFLAGS: 00010046
RAX: 0000000000000246 RBX: ffff88076bbab990 RCX: 0000000000000000
RDX: 0000000000000000 RSI: ffff880468badf58 RDI: ffff88076bbab990
RBP: ffff88076bbab900 R08: 0000000000000246 R09: 00000000000020de
R10: 0000000000000000 R11: ffff88076bbab5be R12: 0000000000000000
R13: ffff880468badf58 R14: 000000000001adb0 R15: ffff88076f310fb0
FS:  00007f377124a880(0000) GS:ffff88046fa40000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 0000000000000008 CR3: 0000000771318000 CR4: 00000000001407e0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400
Stack:
ffff88076bbab990 ffff880468badf50 0000000000000001 ffff88076bbab938
ffffffff810b128b 0000000000000246 00000000cf9b7040 ffff880468bac7a0
0000000000000000 ffff880468bac7a0 ffff88076bbab9d0 ffffffffa05a6ea3

Call Trace:
[<ffffffff810b128b>] prepare_to_wait+0x7b/0x90
[<ffffffffa05a6ea3>] beiscsi_mccq_compl_wait+0x153/0x330 [be2iscsi]
[<ffffffff810b1600>] ? wake_up_atomic_t+0x30/0x30
[<ffffffffa05981b1>] beiscsi_ep_disconnect+0x91/0x2d0 [be2iscsi]
[<ffffffffa0202ffa>] iscsi_if_ep_disconnect.isra.14+0x5a/0x70 [scsi_transport_iscsi]
[<ffffffffa02042fb>] iscsi_if_recv_msg+0x113b/0x14a0 [scsi_transport_iscsi]
[<ffffffff811dffd8>] ? __kmalloc_node_track_caller+0x58/0x290
[<ffffffffa02046ee>] iscsi_if_rx+0x8e/0x1f0 [scsi_transport_iscsi]
[<ffffffff815a351d>] netlink_unicast+0xed/0x1b0
[<ffffffff815a38fe>] netlink_sendmsg+0x31e/0x690
[<ffffffff815a03e4>] ? netlink_rcv_wake+0x44/0x60
[<ffffffff815a19e3>] ? netlink_recvmsg+0x1e3/0x450

beiscsi_mccq_compl_wait gets called even when MCC tag allocation failed
for mgmt_invalidate_connection.  mcc_wait is not initialized for tag 0
so causes crash in prepare_to_wait.

Signed-off-by: Jitendra Bhivare <jitendra.bhivare@broadcom.com>
Reviewed-by: Tomas Henzl <thenzl@redhat.com>
Reviewed-by: Chris Leech <cleech@redhat.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: ufs: fix wrong/ambiguous fall through comments
Tomohiro Kusumi [Thu, 23 Mar 2017 10:49:04 +0000 (12:49 +0200)]
scsi: ufs: fix wrong/ambiguous fall through comments

These aren't really falling through to anywhere meaningful.

Signed-off-by: Tomohiro Kusumi <tkusumi@tuxera.com>
Reviewed-by: Subhash Jadavani <subhashj@codeaurora.org>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: osd_uld: remove an unneeded NULL check
Dan Carpenter [Thu, 23 Mar 2017 10:41:42 +0000 (13:41 +0300)]
scsi: osd_uld: remove an unneeded NULL check

We don't call the remove() function unless probe() succeeds so "oud"
can't be NULL here.  Plus, if it were NULL, we dereference it on the
next line so it would crash anyway.

[mkp: applied by hand]

Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Acked-by: Boaz Harrosh <ooo@electrozaur.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: ipr: Driver version 2.6.4
Brian King [Wed, 15 Mar 2017 21:58:42 +0000 (16:58 -0500)]
scsi: ipr: Driver version 2.6.4

Bump driver version

Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Reviewed-by: Wendy Xiong <wenxiong@linux.vnet.ibm.com>
Tested-by: Wendy Xiong <wenxiong@linux.vnet.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: ipr: Fix SATA EH hang
Brian King [Wed, 15 Mar 2017 21:58:41 +0000 (16:58 -0500)]
scsi: ipr: Fix SATA EH hang

This patch fixes a hang that can occur in ATA EH with ipr. With ipr's
usage of libata, commands should never end up on ap->eh_done_q. The
timeout function we use for ipr, even for SATA devices, is
scsi_times_out, so ATA_QCFLAG_EH_SCHEDULED never gets set for ipr and EH
is driven completely by ipr and SCSI. The SCSI EH thread ends up calling
ipr's eh_device_reset_handler, which then calls
ata_std_error_handler. This ends up calling ipr_sata_reset, which issues
a reset to the device. This should result in all pending commands
getting failed back and having ata_qc_complete called for them, which
should end up clearing ATA_QCFLAG_FAILED as qc->flags gets zeroed in
ata_qc_free.  This ensures that when we end up in ata_eh_finish, we
don't do anything more with the command.

On adapters that only support a single interrupt and when running with
two MSI-X vectors or less, the adapter firmware guarantees that
responses to all outstanding commands are sent back prior to sending the
response to the SATA reset command.  On newer adapters supporting
multiple HRRQs, however, this can no longer be guaranteed, since the
command responses and reset response may be processed on different
HRRQs.

If ipr returns from ipr_sata_reset before the outstanding command was
returned, this sends us down the path of __ata_eh_qc_complete which then
moves the associated scsi_cmd from the work_q in
scsi_eh_bus_device_reset to ap->eh_done_q, which then will sit there
forever and we will be wedged.

This patch fixes this up by ensuring that any outstanding commands are
flushed before returning from eh_device_reset_handler for a SATA device.

Reported-by: David Jeffery <djeffery@redhat.com>
Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Reviewed-by: Wendy Xiong <wenxiong@linux.vnet.ibm.com>
Tested-by: Wendy Xiong <wenxiong@linux.vnet.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: ipr: Error path locking fixes
Brian King [Wed, 15 Mar 2017 21:58:39 +0000 (16:58 -0500)]
scsi: ipr: Error path locking fixes

This patch closes up some potential race conditions observed in the
error handling paths in ipr while debugging an issue resulting in a hang
with SATA error handling. These patches ensure we are holding the
correct lock when adding and removing commands from the free and pending
queues in some error scenarios.

Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Reviewed-by: Wendy Xiong <wenxiong@linux.vnet.ibm.com>
Tested-by: Wendy Xiong <wenxiong@linux.vnet.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
7 years agoscsi: ipr: Fix abort path race condition
Brian King [Wed, 15 Mar 2017 21:58:39 +0000 (16:58 -0500)]
scsi: ipr: Fix abort path race condition

This fixes a race condition in the error handlomg paths of ipr. While a
command is outstanding to the adapter, it is placed on a pending queue
for the hrrq it is associated with, while holding the HRRQ lock. When a
command is completed, it is removed from the pending queue, under HRRQ
lock, and placed on a local list.  This list is then iterated through
without any locks and each command's done function is invoked, inside of
which, the command gets returned to the free list while grabbing the
HRRQ lock. This fixes two race conditions when commands have been
removed from the pending list but have not yet been added to the free
list. Both of these changes fix race conditions that could result in
returning success from eh_abort_handler and then later calling scsi_done
for the same request.

The first race condition is in ipr_cancel_op. It looks through each
pending queue to see if the command to be aborted is still outstanding
or not. Rather than looking on the pending queue, reverse the logic to
check to look for commands that are NOT on the free queue.  The second
race condition can occur when in ipr_wait_for_ops where we are waiting
for responses for commands we've aborted.

Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Reviewed-by: Wendy Xiong <wenxiong@linux.vnet.ibm.com>
Tested-by: Wendy Xiong <wenxiong@linux.vnet.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>