[SCSI] fcoe: reduces lock cost when adding a new skb to fcoe_pending_queue

Currently fcoe_pending_queue.lock held twice for every new skb
adding to this queue when already least one pkt is pending in this
queue and that is not uncommon once skb pkts starts getting queued
here upon fcoe_start_io => dev_queue_xmit failure.

This patch moves most fcoe_pending_queue logic to fcoe_check_wait_queue
function, this new logic grabs fcoe_pending_queue.lock only once to
add a new skb instead twice as used to be.

I think after this patch call flow around fcoe_check_wait_queue
calling in fcoe_xmit is bit simplified with modified
fcoe_check_wait_queue function taking care of adding and
removing pending skb in one function.

Signed-off-by: Vasu Dev <vasu.dev@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] libfc: Check if exchange is completed when receiving a sequence

When a sequence is received in response to an exchange we issued previously,
we should check to see if the exchange has completed. If yes, the sequence
should be discarded. Since the exchange might be still in the completion
process, it should be untouched.

Signed-off-by: Steve Ma <steve.ma@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] libfc: use DID_ERROR when we have internall aborted command

If we aborted a command, because it timed out we should not use
DID_ABORT. It will fail the command right away back to the upper
layer. We want to use something that indicated that the problem
did not complete normally, but it was not a fatal problem.

Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] fcoe: use ETH_P_FIP for skb->protocol of FIP frames

FIP frames should leave the fcoe layer with skb->protocol set to
ETH_P_FIP, not ETH_P_802_3.

Signed-off-by: Chris Leech <christopher.leech@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] libfcoe: fip: fix non-FIP-mode FLOGI state after reset.

When a reset is sent using fcoeadm on a non-FIP mode NIC,
there's no link flap, so the fcoe_ctlr stays in non-FIP mode.

In that case, FIP wasn't setting the flogi_oxid or map_dest flag,
causing the FLOGI to be sent with the both wrong source MAC and
the wrong destination MAC address, causing it to fail.

This leads to a non-functioning HBA until a link flap or
instance delete/create.

Signed-off-by: Joe Eykholt <jeykholt@cisco.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] cxgb3i: Include net/dst.h for struct dst_cache

This driver needs dst_cache->dev so it should include net/dst.h
to ensure that it builds.  While net/tcp.h probably includes it
already, we shouldn't rely on that since there is no guarantee
that this won't change in future.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Acked-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] ibmvfc: Driver version 1.0.6

Bump driver version

Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] ibmvfc: Improve LOGO/PRLO ELS handling

There are several scenarios where the ibmvfc driver needs to
try to log back into a target on the fabric. Today when these events
occur, we simply go through re-discovery for all attached targets,
assuming that either the query of the name server or an ADISC will
indicate we might need to log back into the target, which doesn't
work for all scenarios. Fix this by taking note of the affected target(s)
in these conditions and ensuring we try to PLOGI back into the target.

Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] ibmvfc: Improve device rediscovery

For certain scenarios during device rediscovery, we detect we need
to log back into a target. Currently we do just that - PLOGI/PRLI
back into the target. Change the code to delete and add the target
from the FC transport layer as well, to ensure we handle any cases
where the target may have changed.

Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] ibmvfc: Add flush on halt support

The virtual I/O server controlling the NPIV adapter associated with
a virtual fibre channel adapter can send a HALT event to the client.
When this occurs, the client can no longer send commands until a RESUME
is received. By adding support for flush on halt, we will get all of
our outstanding commands flushed back before the Virtual I/O server
enters the halt state, eliminating potential command timeouts for
outstanding commands which might occur if we did not support this feature.

Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] ibmvfc: Add support for NPIV Logout

This patch adds support for a new command supported by the Virtual I/O
Server, NPIV Logout. The command will abort all outstanding commands
and log out of the fabric. Currently, the only way to do this is
by breaking the CRQ, which can take a fairly long time when lots of
commands are outstanding. The NPIV Logout commands provides a mechanism
to accomplish virtually the same function, but is much faster.

Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] ibmvfc: Fix deadlock in EH

Fixes the following deadlock scenario shown below. We currently allow
queuecommand to send commands when the ibmvfc workqueue is scanning for
new rports, so we should also allow EH to function at this time as well.

scsi_eh_3     D 0000000000000000 12304  1279      2
Call Trace:
[c0000002f7257730] [c0000002f72577e0] 0xc0000002f72577e0 (unreliable)
[c0000002f7257900] [c0000000000118f4] .__switch_to+0x158/0x1a0
[c0000002f72579a0] [c0000000004f8b40] .schedule+0x8d4/0x9dc
[c0000002f7257b60] [c0000000004f8f08] .schedule_timeout+0xa8/0xe8
[c0000002f7257c50] [d0000000001d23e0] .ibmvfc_wait_while_resetting+0xe4/0x140 [ibmvfc]
[c0000002f7257d20] [d0000000001d3984] .ibmvfc_eh_abort_handler+0x60/0xe4 [ibmvfc]
[c0000002f7257dc0] [d000000000366714] .scsi_error_handler+0x38c/0x674 [scsi_mod]
[c0000002f7257f00] [c0000000000a7470] .kthread+0x78/0xc4
[c0000002f7257f90] [c000000000029b8c] .kernel_thread+0x4c/0x68
ibmvfc_3      D 0000000000000000 12432  1280      2
Call Trace:
[c0000002f7253540] [c0000002f72535f0] 0xc0000002f72535f0 (unreliable)
[c0000002f7253710] [c0000000000118f4] .__switch_to+0x158/0x1a0
[c0000002f72537b0] [c0000000004f8b40] .schedule+0x8d4/0x9dc
[c0000002f7253970] [c0000000004f8e98] .schedule_timeout+0x38/0xe8
[c0000002f7253a60] [c0000000004f80cc] .wait_for_common+0x138/0x220
[c0000002f7253b40] [c0000000000a2784] .flush_cpu_workqueue+0xac/0xcc
[c0000002f7253c10] [c0000000000a2960] .flush_workqueue+0x58/0xa0
[c0000002f7253ca0] [d0000000000827fc] .fc_flush_work+0x4c/0x64 [scsi_transport_fc]
[c0000002f7253d20] [d000000000082db4] .fc_remote_port_add+0x48/0x6c4 [scsi_transport_fc]
[c0000002f7253dd0] [d0000000001d7d04] .ibmvfc_work+0x820/0xa7c [ibmvfc]
[c0000002f7253f00] [c0000000000a7470] .kthread+0x78/0xc4
[c0000002f7253f90] [c000000000029b8c] .kernel_thread+0x4c/0x68
fc_wq_3       D 0000000000000000 10720  1283      2
Call Trace:
[c0000002f559ac30] [c0000002f559ace0] 0xc0000002f559ace0 (unreliable)
[c0000002f559ae00] [c0000000000118f4] .__switch_to+0x158/0x1a0
[c0000002f559aea0] [c0000000004f8b40] .schedule+0x8d4/0x9dc
[c0000002f559b060] [c0000000004f8e98] .schedule_timeout+0x38/0xe8
[c0000002f559b150] [c0000000004f80cc] .wait_for_common+0x138/0x220
[c0000002f559b230] [c0000000002721c4] .blk_execute_rq+0xb4/0x100
[c0000002f559b360] [d00000000036a1f8] .scsi_execute+0x118/0x194 [scsi_mod]
[c0000002f559b420] [d00000000036a32c] .scsi_execute_req+0xb8/0x124 [scsi_mod]
[c0000002f559b500] [d0000000000c1330] .sd_sync_cache+0x8c/0x108 [sd_mod]
[c0000002f559b5e0] [d0000000000c15b4] .sd_shutdown+0x9c/0x158 [sd_mod]
[c0000002f559b660] [d0000000000c16d0] .sd_remove+0x60/0xb4 [sd_mod]
[c0000002f559b700] [c000000000392ecc] .__device_release_driver+0xd0/0x118
[c0000002f559b7a0] [c000000000393080] .device_release_driver+0x30/0x54
[c0000002f559b830] [c000000000392108] .bus_remove_device+0x128/0x16c
[c0000002f559b8d0] [c00000000038f94c] .device_del+0x158/0x234
[c0000002f559b960] [d00000000036f078] .__scsi_remove_device+0x5c/0xd4 [scsi_mod]
[c0000002f559b9f0] [d00000000036f124] .scsi_remove_device+0x34/0x58 [scsi_mod]
[c0000002f559ba80] [d00000000036f204] .__scsi_remove_target+0xb4/0x120 [scsi_mod]
[c0000002f559bb10] [d00000000036f338] .__remove_child+0x2c/0x44 [scsi_mod]
[c0000002f559bb90] [c00000000038f11c] .device_for_each_child+0x54/0xb4
[c0000002f559bc50] [d00000000036f2e0] .scsi_remove_target+0x70/0x9c [scsi_mod]
[c0000002f559bce0] [d000000000083454] .fc_starget_delete+0x24/0x3c [scsi_transport_fc]
[c0000002f559bd70] [c0000000000a2368] .run_workqueue+0x118/0x208
[c0000002f559be30] [c0000000000a2580] .worker_thread+0x128/0x154
[c0000002f559bf00] [c0000000000a7470] .kthread+0x78/0xc4
[c0000002f559bf90] [c000000000029b8c] .kernel_thread+0x4c/0x68

Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] ibmvfc: Reduce error logging noise

The ibmvfc driver currently logs errors during discovery for several
transient fabric errors, which generally get retried. If retries
do not work, we see multiple errors in the log. If retries do work,
we see errors in the log which may be confusing since the retry worked.
This patch enhances the discovery time error logging to only log errors
for command failures during discovery if all allowed retries have been
used up. The existing behavior of logging all failures can be restored
by setting the hosts log_level to a value of 3 or greater.

Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] ibmvfc: Use DEVICE_ATTR macro

Use DEVICE_ATTR macro for defining device sysfs attributes.

Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] ibmvfc: Fixup GFP flags for target allocations

Since target allocations can occur while resetting the virtual adapter,
we shouldn't be using GFP_KERNEL for them as it could hang. Switch to
use GFP_NOIO.

Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] ibmvfc: Fix invalid error response handling

Fix an obvious bug in processing error responses for SCSI commands
which can result in successful responses being incorrectly returned
with DID_ERROR.

Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] sd: fix bug in SCSI async probing

The async split up of probing in sd.c created a potential failure case where
something goes wrong with device_add(), but which we don't recover properly.
Since, in general, asynchronous error handling is hard, move the device_add()
into the asynchronous path (it should be fast) and make sure all the deferred
processing cannot fail.

Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] fix up scsi_eh_lock_door()

The Documentation is incorrect (we removed some functions referred to), and
none of the bug warnings now apply.  Additionally remove the spurious check on
the return from blk_get_request() which can't fail if __GFP_WAIT is passed in.

Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] fix documentation for two functions

Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] lpfc 8.3.2 : Update the lpfc driver version to 8.3.2

Signed-off-by: James Smart <james.smart@emulex.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] lpfc 8.3.2 : Persistent Vport Support

Add support for persistent vport definitions at creation at boot time

Also includes a few misc fixes for:
- conversion to vpi name from vport slang name
- couple of small mailbox references
- some additional discovery mods

Signed-off-by: James Smart <james.smart@emulex.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] lpfc 8.3.2 : Miscellaneous Changes

Miscellaneous Changes:
- Convert from SLI2_ACTIVE flag to more correct SLI_ACTIVE (generic) flag
- Reposition log verbose messaging definitions
- Update naming for vpi object name from vport slang name
- Handle deferred error attention condition
- Add 10G link support
- Small bug fixup

Signed-off-by: James Smart <james.smart@emulex.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] lpfc 8.3.2 : Update of copyrights

Update of copyrights on modified files

Signed-off-by: James Smart <james.smart@emulex.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] lpfc 8.3.2 : Addition of SLI4 Interface - FCOE Discovery support

SLI4 supports both FC and FCOE, with some extended topology objects.
This patch adss support for the objects, and updates the disovery
engines for their use.

Signed-off-by: James Smart <james.smart@emulex.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] lpfc 8.3.2 : Addition of SLI4 Interface - Mailbox handling

The mailbox commands themselves are the same, or very similar to
their SLI3 counterparts. This patch genericizes mailbox command
handling and adds support for the new SLI4 mailbox queue.

Signed-off-by: James Smart <james.smart@emulex.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] lpfc 8.3.2 : Addition of SLI4 Interface - Queues

Adds support for the new queues in the SLI-4 interface.  There are :
- Work Queues - host-to-adapter for fast-path traffic
- Mailbox Queues - host-to-adapter for control (slow-path)
- Buffer Queues - host-to-adapter for posting buffers for async receive
- Completion Queues - adapter-to-host for posting async events,
       completions for fast or slow patch work, receipt of async
       receive traffic
- Event Queues - tied to MSI-X vectors, binds completion queues with
       interrupts

These patches add the all the support code to tie into command submission
and response paths, updates the interrupt handling, etc.

Signed-off-by: James Smart <james.smart@emulex.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] lpfc 8.3.2 : Addition of SLI4 Interface - Base Support

Adds new hardware and interface definitions.

Adds new interface routines - utilizing the reorganized layout of the
driver. Adds SLI-4 specific functions for attachment, initialization,
teardown, etc.

Signed-off-by: James Smart <james.smart@emulex.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] lpfc 8.3.2 : Reorganization for SLI4

Preps the organization of the driver so that the bottom half, which
interacts with the hardware, can share common code sequences for
attachment, detachment, initialization, teardown, etc with new hardware.

For very common code sections, which become specific to the interface
type, the driver uses an indirect function call. The function is set at
initialization. For less common sections, such as initialization, the
driver looks at the interface type and calls the routines relative to
the interface.

Signed-off-by: James Smart <james.smart@emulex.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] libfc,fcoe,fnic: Separate rport and lport max retry counts

This allows fnic to configure number of retries for lport and rport
separately.

Signed-off-by: Abhijeet Joglekar <abjoglek@cisco.com>
Acked-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] mpt2sas: bump driver version to 01.100.03.00

Bump driver version.

Signed-off-by: Eric Moore <eric.moore@lsi.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] mpt2sas: using the same naming convention for all static function

This fix's is for all local function so their name has the "_" preceeding
the module name, then function name.  Most the code is already is using this
naming convention.

Signed-off-by: Eric Moore <eric.moore@lsi.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] mpt2sas: add query task support for MPT2COMMAND ioctl

This patch will find an active mid for a query_task request via the ioctl path.

This code is already there for task_abort, so this patch combining code using
the same fuction _ctl_set_task_mid(), previously _ctl_do_task_abort().

Signed-off-by: Eric Moore <eric.moore@lsi.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] mpt2sas: LUN Reset Support

Adding new eh_target_reset_handler for target reset. Change the
eh_device_reset_handler so its sending
MPI2_SCSITASKMGMT_TASKTYPE_LOGICAL_UNIT_RESET, instead of
MPI2_SCSITASKMGMT_TASKTYPE_TARGET_RESET.  Add new function
_scsih_scsi_lookup_find_by_lun as a sanity check to insure I_T_L commands are
completed upon completing lun reset.

Signed-off-by: Eric Moore <eric.moore@lsi.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] mpt2sas: T10 DIF Support

This add support for type 1 and 3 DIF support per the Oracle API.

Signed-off-by: Eric Moore <eric.moore@lsi.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] zfcp: Keep ccw device and model id in zfcp_ccw.c

Keep the information about the device and model id in zfcp_ccw. This
requires an additional helper function to check for the privileged
cfdc subchannel, but it allows the removal of the redundant defines
from the zfcp_def header file.

Reviewed-by: Swen Schillig <swen@vnet.ibm.com>
Signed-off-by: Christof Schmitt <christof.schmitt@de.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] zfcp: Changed D_ID left port disabled

If the destination ID (D_ID) of a remote storage port changed, e.g.
re-plugged cable on the switch in a different switch port, the port
was never (re-)attached within Linux. This patch fixes the broken
mapping between the WWPN and the D_ID.

Signed-off-by: Swen Schillig <swen@vnet.ibm.com>
Signed-off-by: Christof Schmitt <christof.schmitt@de.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] zfcp: Increase ref counter for port open requests

In rare cases, open port request might timeout, erp calls
zfcp_port_put, port gets dequeued. Now, the late returning (or
dismissed) fsf-port-open calls the fsf_port_open_handler that tries to
reference the port data structure leading to a kernel oops.

Signed-off-by: Martin Petermann <martin.petermann@de.ibm.com>
Signed-off-by: Christof Schmitt <christof.schmitt@de.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] zfcp: Add comments to switch/case fallthroughs

Add comments where there is a deliberate fall through in switch/case
statements. This makes some code checkers happy and makes it clear
that there is no missing break statement.

Reviewed-by: Swen Schillig <swen@vnet.ibm.com>
Signed-off-by: Christof Schmitt <christof.schmitt@de.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] zfcp: Remove unnecessary default case and assignments

enum dma_data_direction only has the 4 values DMA_BIDIRECTIONAL,
DMA_TO_DEVICE, DMA_FROM_DEVICE and DMA_NONE. No need to have the
default case. While changing this, setup sbtype in one place to make
sparse happy.

The default value of retval is already -EIO, so remove the
additional assignment for these two cases.

Reviewed-by: Swen Schillig <swen@vnet.ibm.com>
Signed-off-by: Christof Schmitt <christof.schmitt@de.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] zfcp: Make queue_depth adjustable

zfcp did always set the queue_depth for SCSI devices to 32, not
allowing to change this. Introduce a kernel parameter zfcp.queue_depth
and the change_queue_depth callback to allow changing the queue_depth
when it is required.

Reviewed-by: Swen Schillig <swen@vnet.ibm.com>
Signed-off-by: Christof Schmitt <christof.schmitt@de.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] zfcp: Update message and add description

Update the newly introduced message for the boxed status to conform to
match the style of s390 and zfcp messages.

Reviewed-by: Swen Schillig <swen@vnet.ibm.com>
Signed-off-by: Christof Schmitt <christof.schmitt@de.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] zfcp: Use correct req_id for traces

The zfcp traces used the fsf_req address in place of the req_id.
Change this to save the correct req_id.

Reviewed-by: Swen Schillig <swen@vnet.ibm.com>
Signed-off-by: Christof Schmitt <christof.schmitt@de.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] mvsas: remove all the casts from void * or to void *

Signed-off-by: Ying Chu <jasonchu@marvell.com>
Signed-off-by: Andy Yan <ayan@marvell.com>
Signed-off-by: Ke Wei <kewei@marvell.com
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] libiscsi: add debug printks for iscsi command completion path

This patch just adds some debug statements for the abort
and completion paths.

Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] libiscsi: add task aborted state

If a task did not complete normally due to a TMF, libiscsi will
now complete the task with the state ISCSI_TASK_ABRT_TMF. Drivers
like bnx2i that need to free resources if a command did not complete normally
can then check the task state. If a driver does not need to send
a special command if we have dropped the session then they can check
for ISCSI_TASK_ABRT_SESS_RECOV.

Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] libiscsi: check if iscsi host has work queue before queueing work

Instead of having libiscsi check if the offload bit is set, have
it check if the lld created a work queue. I think this is more
clear.

Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] libiscsi: don't let io sit in queue when session has failed

If the session is failed, but we have not yet fully transitioned
to the recovery stage we were still queueuing IO. The idea is
that for some failures we can recvover at the command level
and still continue to execute other IO. Well, we never have
added the recovery within a command code, so queueing up IO here
just creates the possibility that it might time time out so
this just has us requeue the IO the scsi layer for now.

Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] libiscsi: handle cleanup task races

bnx2i needs to send a hardware specific cleanup command if
a command has not completed normally (iscsi/scsi response from
target), and the session is still ok (this is the case when we
send a TMF to stop the command).

At this time it will need to drop the session lock. The problem
with the current code is that fail_all_commands assumes we
will hold the lock the entire time, so it uses list_for_each_entry_safe.
If while bnx2i drops the session lock multiple cmds complete then
list_for_each_entry_safe will not handle this correctly.

This patch removes the running lists and just has us loop over
the cmds array (in later patches we will then replace that
array with a block tag map at the session level). It also fixes
up the completion path so that if the TMF code and the normal recv
path were completing the same command then they both do not try
to do release the refcount taken when the task is queued.

Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] libiscsi: fix iscsi transport checks to account for slower links

If we have not got any pdus for recv_timeout seconds, then we will
send a iscsi ping/nop to make sure the target is still around. The
problem is if this is a slow link, and the ping got queued after
the data for a data_out (read), then the transport code could think
the ping has failed when it is just slowly making its way through
the network. This patch has us check if we are making progress while
the nop is outstanding. If we are still reading in data, then we
do not fail the session at that time.

Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] libiscsi_tcp: update recv tracking for each skb instead of iscsi pdu

Everytime we read in a pdu libiscsi will update a tracking field.
It uses this to decide when to check if the transport might be bad.
If we have not got data in recv_timeout seconds then we will
send a iscsi ping/nop.

If we are on a slow link then it could take a while to read in all
the data for a data_in. In that case we might send a ping/nop when
we do not need to or we might drop a session thinking it is bad
when the lower layer is making forward progress on it.

This patch has libiscsi_tcp update the recv tracking for each skb
(basically network packet from our point of view) instead of the
entire iscsi pdu+data, so we account for these cases where data is
coming in slowly.

Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] libiscsi: fix nop response/reply and session cleanup race

If we are responding to a nop from the target by sending our nop,
and the session is getting torn down, then iscsi_start_session_recovery
could set the conn stop bits while the recv path is sending the nop
response and we will hit the bug ons in __iscsi_conn_send_pdu.

This has us check the state in __iscsi_conn_send_pdu and fail all
incoming mgmt IO if we are  not logged in and if the pdu is not login
related. It also changes the ordering of the setting of conn stop state
bits so they are set after the session state is set (both are set under
the session lock).

Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] libiscsi: have iscsi_data_in_rsp call iscsi_update_cmdsn

This has iscsi_data_in_rsp call iscsi_update_cmdsn when a pdu is
completed like is done for other pdu's that are don.

For libiscsi_tcp, this means that it calls iscsi_update_cmdsn when
it is handling the pdu internally to only transfer data, but if there is
status then it does not need to call it since the completion handling
will do it.

Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] libiscsi: export iscsi_itt_to_task for bnx2i

bnx2i needs to be able to look up mgmt task like login and nop, because
it does some processing of them on the completion path. This exports
iscsi_itt_to_task so it can look up the task.

Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] libiscsi: handle param allocation failures

If we could not allocate the initiator name or some other id like
the hwaddress or netdev, then userspace could deal with the failure
by just running in a dregraded mode.

Now we want to be able to switch values for the params and we
want some feedback, so this patch will check if a string like
the initiatorname could not be allocated and return an error.

Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] libiscsi: check of LLD has a alloc pdu callout.

bnx2i does not have one. It currently preallocates the bdt
when the session is setup.

We probably want to change that to a dma pool, then allocate from
the pool in the alloc pdu. Until then check if there is a alloc
pdu callout.

Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] iscsi: pass ep connect shost

When we create the tcp/ip connection by calling ep_connect, we currently
just go by the routing table info.

I think there are two problems with this.

1. Some drivers do not have access to a routing table. Some drivers like
qla4xxx do not even know about other ports.

2. If you have two initiator ports on the same subnet, the user may have
set things up so that session1 was supposed to be run through port1. and
session2 was supposed to be run through port2. It looks like we could
end with both sessions going through one of the ports.

Fixes for cxgb3i from Karen Xie.

Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] NCR_D700: fix IRQ handler return type

drivers/scsi/NCR_D700.c: In function `NCR_D700_probe':
drivers/scsi/NCR_D700.c:322: warning: passing argument 2 of `request_irq' from incompatible pointer type
drivers/scsi/NCR_D700.c:322: warning: passing argument 2 of `request_irq' from incompatible pointer type
drivers/scsi/NCR_D700.c:322: warning: passing argument 2 of `request_irq' from incompatible pointer type

Signed-off-by: Zhenwen Xu <helight.xu@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] mvsas: performance improvement using domain_device->lldd_dev

Using sticky field to improve retrieve performance by eliminating some
lookups in . Remove some spurious casts.

Signed-off-by: Ying Chu <jasonchu@marvell.com>
Signed-off-by: Andy Yan <ayan@marvell.com>
Signed-off-by: Ke Wei <kewei@marvell.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] mvsas: correct bit map usage

Utilize DECLARE_BITMAP to define the tags array.

Signed-off-by: Ying Chu <jasonchu@marvell.com>
Signed-off-by: Andy Yan <ayan@marvell.com>
Signed-off-by: Ke Wei <kewei@marvell.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] mvsas: bug fix, null pointer may be used

Null pointer check to avoid corruption.

Signed-off-by: Ying Chu <jasonchu@marvell.com>
Signed-off-by: Andy Yan <ayan@marvell.com>
Signed-off-by: Ke Wei <kewei@marvell.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] mvsas: bug fix of dead lock

TMF task should be issued with Interrupt Disabled, or Deadlock may take place.
Clean-up unused parameters and conditonal lock.

Signed-off-by: Ying Chu <jasonchu@marvell.com>
Signed-off-by: Andy Yan <ayan@marvell.com>
Signed-off-by: Ke Wei <kewei@marvell.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] mvsas: bug fix with setting task management frame type

Correct frame type setting according to parameter.

Signed-off-by: Ying Chu <jasonchu@marvell.com>
Signed-off-by: Andy Yan <ayan@marvell.com>
Signed-off-by: Ke Wei <kewei@marvell.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] ipr: fix PCI permanent error handler

The ipr driver can hang if it encounters enough PCI errors
to trigger the permanent error handler. The driver will attempt
to initiate a "bringdown" of the adapter and fail all pending
ops back. However, this bringdown is unlike any other bringdown
of the adapter in the code as the driver. In this code path we
end up failing back ops with allow_cmds still set to 1. This results
in some commands, the HCAM commands in particular, getting immediately
re-issued to the adapter on the done call, which results in
an infinite loop in ipr_fail_all_ops. Fix this by setting allow_cmds
to zero in this path.

Signed-off-by: Kleber S. Souza <klebers@linux.vnet.ibm.com>
[brking@linux.vnet.ibm.com: alternate patch substituted]
Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] Update wording of CONFIG_SCSI_MULTI_LUN help

I had to set CONFIG_SCSI_MULTI_LUN to y in order to get my SE W595
working when plugging it as a mass storage. Looking at SCSI option to
get a phone behaving correctly was convoluted to say the least. There
are quite a few other reports about USB card readers needing this option
as well. This patch improves the help text to make the use of the option
more obvious.

Signed-off-by: Eric Piel <eric.piel@tremplin-utc.net>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] ibmvscsi: Remove redundant test on unsigned.

Signed-off-by: Roel Kluin <roel.kluin@gmail.com>
Acked by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] st: fix gcc 4.4 warning

This patch fixes the GCC 4.4 warning reported by David Binderman and Sergey
Senozhatsky. The old version was working correctly but was not easy to read.

Signed-off-by: Kai Makisara <kai.makisara@kolumbus.fi>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] limit state transitions in scsi_internal_device_unblock

scsi timeout on two or more devices may cause extremely long execution
time for user applications because SDEV_OFFLINE state is changed to
SDEV_RUNNING state during scsi error recovery procedures triggered by
a bus reset or a host reset of scsi LLD, and scsi timeout can happens
on the same devices many times.

This happens because scsi_internal_device_unblock() changes device's
state to SDEV_RUNNING even if a device in other states than SDEV_BLOCK,
while the following two transitions are required in this function.

  SDEV_BLOCK -> SDEV_RUNNING
  SDEV_CREATED_BLOCK -> SDEV_CREATED

Otherwise, it returns -EINVAL.

Signed-off-by: Takahiro Yasui <tyasui@redhat.com>
[matthew@wil.cx: supplied rewritten base for patch]
Signed-off-by: Matthew Wilcox <matthew@wil.cx>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] fcoe, libfc: fix function declarations to be ANSI-compliant

Fix function declarations:

drivers/scsi/fcoe/fcoe.c:1356:28: warning: non-ANSI function declaration of function 'fcoe_dev_setup'
drivers/scsi/libfc/fc_rport.c:1293:20: warning: non-ANSI function declaration of function 'fc_setup_rport'
drivers/scsi/libfc/fc_rport.c:1302:23: warning: non-ANSI function declaration of function 'fc_destroy_rport'

[jejb: fixed wrong doc in comment noticed during inspection]
Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] scsi_debug: fix virtual disk larger than 1TB

Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Acked-by: Douglas Gilbert <dgilbert@interlog.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] ipr: ipr_remove() marked __devexit

Marking the ipr clean up function ipr_remove() as __devexit and using
__devexit_p() macro in its address reference.

Signed-off-by: Kleber Sacilotto de Souza <kleber@linux.vnet.ibm.com>
Reported-by: Breno Leitao <leitao@linux.vnet.ibm.com>
Acked-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] scsi_dh_rdac: Retry for NOT_READY(02/04/01) in rdac device handler

During device discovery read capacity fails with 0x020401 and sets the
device size to 0. As a reason any I/O submitted to this path gets
killed at sd_prep_fn with BLKPREP_KILL. This patch is to retry for
0x020401. NEED_RETRY in scsi_decide_disposition does not give
sufficient time for the device to become ready.

Signed-off-by: Vijay Chauhan <vijay.chauhan@lsi.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] sym53c8xx_2: slave_alloc/destroy safety (2.6.27.5)

Make the sym53c8xx_2 driver slave_alloc/destroy less unsafe. References
to the destroyed LCB are cleared from the target structure (instead of
leaving a dangling pointer), and when the last LCB for the target is
destroyed the reference to the upper layer target data is cleared. The
host lock is used to prevent a race with the interrupt handler. Also
user commands are prevented for targets with all LCBs destroyed.

Signed-off-by: Aaro Koskinen <Aaro.Koskinen@nokia.com>
Tested-by: Tony Battersby <tonyb@cybernetics.com>
Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] sym53c8xx_2: lun to_clear flag not re-initialized (2.6.27.5)

(Resent with proper formatting)

Fix for the sym53c8xx_2 driver to initialize lun's to_clear flag after
a bus reset (a failed clear can trigger a bus reset and it should not
be attemped again after that).

Signed-off-by: Aaro Koskinen <Aaro.Koskinen@nokia.com>
Tested-by: Tony Battersby <tonyb@cybernetics.com>
Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] qla1280: error recovery rewrite

The driver now waits for the scsi commands associated with a
particular error recovery step to be returned to the mid-layer,
and returns the appropriate SUCCESS or FAILED status.  Removes
unneeded polling of chip for interrupts.

This patch also bumps the driver version number.

Signed-off-by: Michael Reed <mdr@sgi.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] qla1280: driver clean up

Remove some unneeded, inactive and unused code, make some trivial
corrections to comments and a printk, and return a proper status
in qla1280_queuecommand.  No fundamental logic changes are made.

Signed-off-by: Michael Reed <mdr@sgi.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] Increase default timeout for INQUIRY

This patch (as1224) changes the default timeout for INQUIRY commands
from 3 seconds to 20 seconds, which is the value used by Windows for
USB Mass-Storage devices.  Some of these devices, like the Corsair
Flash Voyager (see Bugzilla #12188) really do need a long timeout.

Signed-off-by: Alan Stern <stern@rowland.harvard.edu>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] mvsas: add support for 94xx; layout change; bug fixes

This version contains following main changes
  - Switch to new layout to support more types of ASIC.
  - SSP TMF supported and related Error Handing enhanced.
  - Support flash feature with delay 2*HZ when PHY changed.
  - Support Marvell 94xx series ASIC for 6G SAS/SATA, which has 2
88SE64xx chips but any different register description.
  - Support SPI flash for HBA-related configuration info.
  - Other patch enhanced from kernel side such as increasing PHY type

[jejb: fold back in DMA_BIT_MASK changes]
Signed-off-by: Ying Chu <jasonchu@marvell.com>
Signed-off-by: Andy Yan <ayan@marvell.com>
Signed-off-by: Ke Wei <kewei@marvell.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] mvsas: split driver into multiple files

Split mvsas driver into multiple source codes, based on the split
and function distribution found in Marvell's mvsas update.

Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] mvsas: move into new directory drivers/scsi/mvsas/

Zero functional changes, just file movement.

This commit prepares for the upcoming integration of the
Marvell-provided driver update that splits the driver into support
for both 64xx and 94xx chip families.

Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] qla2xxx: Update version number to 8.03.01-k2.

Signed-off-by: Andrew Vasquez <andrew.vasquez@qlogic.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] qla2xxx: Use port number to compute nvram/vpd parameter offsets.

Read adapter's physical port number from interrupt pin register
and use it instead of pci function number to offset into the
nvram to obtain the port's configuration parameters.

Signed-off-by: Anirban Chakraborty <anirban.chakraborty@qlogic.com>
Signed-off-by: Andrew Vasquez <andrew.vasquez@qlogic.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] qla2xxx: Add an override option to specify ISP firmware load semantics.

As it may be useful during debugging to use a specific firmware
image.

Signed-off-by: Andrew Vasquez <andrew.vasquez@qlogic.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] qla2xxx: Don't try to 'stop' firmware if already in ROM code.

Signed-off-by: Andrew Vasquez <andrew.vasquez@qlogic.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] qla2xxx: Conditionally disable automatic queue full tracking.

Changing a lun's queue depth (/sys/block/sdX/device/queue_depth)
isn't sticky when the device is connected via a QLogic fibre
channel adapter.

The QLogic qla2xxx fibre channel driver dynamically adjusts a
lun's queue depth.  If a user has a specific need to limit the
number of commands issued to a lun (say a tape drive, or a shared
raid where the total commands issued to all luns is limited at
the controller level, for example) and writes a limiting value to
/sys/block/sdXX/device/queue_depth, the qla2xxx driver will
silently and gradually increase the queue depth back to the
driver limit of ql2xmaxqdepth.  While reducing this value (module
parameter) or increasing the interval between ramp ups
(ql2xqfullrampup) offers the potential for a work around it would
be better to have the option of just disabling the dynamic
adjustment of queue depth.

This patch implements an "off switch" as a module parameter.

Signed-off-by: Michael Reed <mdr@sgi.com>
Signed-off-by: Andrew Vasquez <andrew.vasquez@qlogic.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] qla2xxx: Perform an implicit login to the Management Server.

Set the conditional plogi option bit whenever logging in the
fabric management server (if it is already logged in, it does not
need an explicit login; an implicit login suffices).

Signed-off-by: Andrew Vasquez <andrew.vasquez@qlogic.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] qla2xxx: Restrict model-name/description device-table usage.

Information present in static table is only valid for pre-ISP25xx
adapters.

Signed-off-by: Andrew Vasquez <andrew.vasquez@qlogic.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] qla2xxx: Correct hard-coded address of a second-port's NVRAM.

Signed-off-by: Harish Zunjarrao <harish.zunjarrao@qlogic.com>
Signed-off-by: Andrew Vasquez <andrew.vasquez@qlogic.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] qla2xxx: Correct typo in read_nvram() callback.

Signed-off-by: Andrew Vasquez <andrew.vasquez@qlogic.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] qla2xxx: Remove reference to request queue from scsi request block.

srbs used to maintain a reference to the request queue on which
it was enqueued. This is no longer required as the request queue
pointer is now maintained in the scsi host that issues the srb.

Signed-off-by: Anirban Chakraborty <anirban.chakraborty@qlogic.com>
Signed-off-by: Andrew Vasquez <andrew.vasquez@qlogic.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] qla2xxx: Add CPU affinity support.

Set the module parameter ql2xmultique_tag to 1 to enable this
feature. In this mode, the total number of response queues
created is equal to the number of online cpus. Turning the block
layer's rq_affinity mode on enables requests to be routed to the
proper cpu and at the same time it enables completion of the IO
in a response queue that is affined to the cpu in the request
path.

Signed-off-by: Anirban Chakraborty <anirban.chakraborty@qlogic.com>
Signed-off-by: Andrew Vasquez <andrew.vasquez@qlogic.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] qla2xxx: Add QoS support.

Set the number of request queues to the module paramater
ql2xmaxqueues.  Each vport gets a request queue. The QoS value
set to the request queues determines priority control for queued
IOs. If QoS value is not specified, the vports use the default
queue 0.

Signed-off-by: Anirban Chakraborty <anirban.chakraborty@qlogic.com>
Signed-off-by: Andrew Vasquez <andrew.vasquez@qlogic.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] qla2xxx: Correct compilation failures when DEBUG'n' options are enabled.

Signed-off-by: Andrew Vasquez <andrew.vasquez@qlogic.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] qla2xxx: Export additional FCoE attributes for application support.

Cull and export VN_Port MAC address and VLAN_ID information on
supported FCoE ISPs.

Signed-off-by: Andrew Vasquez <andrew.vasquez@qlogic.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] qla2xxx: Correct bus-reset behaviour with recent ISPs.

The short-circuit to skip the non-applicable 'full-login-lip'
process on 81xx ISPs was nested too deeply in the 'bus-reset'
routine, as the code in qla2x00_loop_reset() should skip the
whole enable_lip_full_login process.  The original code could
cause device tear-down due to the qla2x00_wait_for_loop_ready()
call taking a large amount of time.

Signed-off-by: Seokmann Ju <seokmann.ju@qlogic.com>
Signed-off-by: Andrew Vasquez <andrew.vasquez@qlogic.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

Merge branch 'scsi-fixes' into merge-base

Merge branch 'fixes-for-linus' of git://git.monstr.eu/linux-2.6-microblaze

* 'fixes-for-linus' of git://git.monstr.eu/linux-2.6-microblaze:
  microblaze: Fix kind-of-intr checking against number of interrupts
  microblaze: Update Microblaze defconfig

Merge branch 'for-linus' of git://git./linux/kernel/git/lrg/voltage-2.6

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/lrg/voltage-2.6:
  regulator: da903x: add missing __devexit_p()

Avoid ICE in get_random_int() with gcc-3.4.5

Martin Knoblauch reports that trying to build 2.6.30-rc6-git3 with
RHEL4.3 userspace (gcc (GCC) 3.4.5 20051201 (Red Hat 3.4.5-2)) causes an
internal compiler error (ICE):

    drivers/char/random.c: In function `get_random_int':
    drivers/char/random.c:1672: error: unrecognizable insn:
    (insn 202 148 150 0 /scratch/build/linux-2.6.30-rc6-git3/arch/x86/include/asm/tsc.h:23 (set (reg:SI 0 ax [91])
            (subreg:SI (plus:DI (plus:DI (reg:DI 0 ax [88])
                        (subreg:DI (reg:SI 6 bp) 0))
                    (const_int -4 [0xfffffffffffffffc])) 0)) -1 (nil)
        (nil))
    drivers/char/random.c:1672: internal compiler error: in extract_insn, at recog.c:2083

and after some debugging it turns out that it's due to the code trying
to figure out the rough value of the current stack pointer by taking an
address of an uninitialized variable and casting that to an integer.

This is clearly a compiler bug, but it's not worth fighting - while the
current stack kernel pointer might be somewhat hard to predict in user
space, it's also not generally going to change for a lot of the call
chains for a particular process.

So just drop it, and mumble some incoherent curses at the compiler.

Tested-by: Martin Knoblauch <spamtrap@knobisoft.de>
Cc: Matt Mackall <mpm@selenic.com>
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

nfs: Fix NFS v4 client handling of MAY_EXEC in nfs_permission.

The problem is that permission checking is skipped if atomic open is
possible, but when exec opens a file, it just opens it O_READONLY which
means EXEC permission will not be checked at that time.

This problem is observed by the following sequence (executed as root):

  mount -t nfs4 server:/ /mnt4
  echo "ls" >/mnt4/foo
  chmod 744 /mnt4/foo
  su guest -c "mnt4/foo"

Signed-off-by: Frank Filz <ffilzlnx@us.ibm.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Cc: stable@kernel.org
Tested-by: Eugene Teo <eugeneteo@kernel.sg>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

[SCSI] mpt2sas: fix driver version inconsistency

In Commit

commit 3b8b5c9b1f08660583e5dfe095c24170df62f1d2
Author: Eric Moore <eric.moore@lsi.com>
Date:   Tue Apr 21 15:44:27 2009 -0600

    [SCSI] mpt2sas : bump driver version to 01.100.02.00

The MPT2SAS_MAJOR_VERSION didn't get bumped from 00 to 01 so
applications will see it incorrectly as 00.100.02.00 driver instead of
01.100.02.00.  Fix by making MPT2SAS_MAJOR_VERSION match the major
number in MPT2SAS_DRIVER_VERSION

Signed-off-by: Eric Moore <eric.moore@lsi.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>