11 years ago[SCSI] iscsi class: fix endpoint id handling
Mike Christie [Wed, 24 Sep 2008 16:46:11 +0000]
[SCSI] iscsi class: fix endpoint id handling

Some endpoint code was using unsigned int and some
was using uint64_t. This converts it all to uint64_t.

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

11 years ago[SCSI] libiscsi: Support drivers initiating session removal
Mike Christie [Wed, 24 Sep 2008 16:46:10 +0000]
[SCSI] libiscsi: Support drivers initiating session removal

If the driver knows when hardware is removed like with cxgb3i,
bnx2i, qla4xxx and iser then we will want to remove the sessions/devices
that are bound to that device before removing the host.

cxgb3i and in the future bnx2i will remove the host and that will
remove all the sessions on the hba. iser can call iscsi_kill_session
when it gets an event that indicates that a hca is removed.
And when qla4xxx is hooked in to the lib (it is only hooked into
the class right now) it can call iscsi remove host like the
partial offload card drivers.

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

11 years ago[SCSI] libiscsi: fix data corruption when target has to resend data-in packets
Mike Christie [Wed, 24 Sep 2008 16:46:09 +0000]
[SCSI] libiscsi: fix data corruption when target has to resend data-in packets

iscsi_tcp was updating the exp_statsn (exp_statsn acknowledges
status and tells the target it is ok to let the resources for
a iscsi pdu to be reused) before it got all the data for pdu read
into OS buffers. Data corruption was occuring if something happens
to a packet and the network layer requests a retransmit, and the
initiator has told the target about the udpated exp_statsn ack,
then the target may be sending data from a buffer it has reused
for a new iscsi pdu. This fixes the problem by having the LLD
(iscsi_tcp in this case) just handle the transferring of data, and
has libiscsi handle the processing of status (libiscsi completion
processing is done after LLD data transfers are complete).

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

11 years ago[SCSI] sd: Switch kernel printing level for DIF messages
Martin K. Petersen [Wed, 1 Oct 2008 05:37:21 +0000]
[SCSI] sd: Switch kernel printing level for DIF messages

For some reason these messages ended up being printed with KERN_INFO
rendering them invisible to pretty much everyone.  Switch to
KERN_NOTICE.

Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

11 years ago[SCSI] sd: Correctly handle all combinations of DIF and DIX
Martin K. Petersen [Fri, 19 Sep 2008 22:47:21 +0000]
[SCSI] sd: Correctly handle all combinations of DIF and DIX

The old detection code couldn't handle all possible combinations of
DIX and DIF.  This version does, giving priority to DIX if the
controller is capable.

Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

11 years ago[SCSI] sd: Always print actual protection_type
Martin K. Petersen [Fri, 19 Sep 2008 22:47:20 +0000]
[SCSI] sd: Always print actual protection_type

Now that we no longer use protection_type as trigger for preparing
protected CDBs we can remove the places that set it to zero.  This
allows userland to see which protection type the device is formatted
with regardless of whether the HBA supports DIF or not.

Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

11 years ago[SCSI] sd: Issue correct protection operation
Martin K. Petersen [Fri, 19 Sep 2008 22:47:19 +0000]
[SCSI] sd: Issue correct protection operation

Use the same logic to prepare RD/WRPROTECT and the protection
operation.  Fixes a corner case where we could issue an unprotected
CDB and yet tell the HBA to do DIF to the drive.

Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

11 years ago[SCSI] scsi_error: fix target reset handling
James Bottomley [Fri, 12 Sep 2008 21:46:51 +0000]
[SCSI] scsi_error: fix target reset handling

There's a target reset bug.

This loop:

for (id = 0; id <= shost->max_id; id++) {

Never terminates if shost->max_id is set to ~0, like aic94xx does.

It's also pretty inefficient since you mostly have compact target
numbers, but the max_id can be very high.  The best way would be to
sort the recovery list by target id and skip them if they're equal,
but even a worst case O(N^2) traversal is probably OK here, so fix it
by finding the next highest target number (assuming n+1) and
terminating when there isn't one.

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

11 years ago[SCSI] lpfc 8.2.8 v2 : Add statistical reporting control and additional fc vendor...
James Smart [Sun, 7 Sep 2008 15:52:10 +0000]
[SCSI] lpfc 8.2.8 v2 : Add statistical reporting control and additional fc vendor events

Added support for new sysfs attributes: lpfc_stat_data_ctrl and
lpfc_max_scsicmpl_time. The attributes control statistical reporting
of io load.

Added support for new fc vendor events for error reporting.

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

11 years ago[SCSI] lpfc 8.2.8 v2 : Add sysfs control of target queue depth handling
James Smart [Sun, 7 Sep 2008 15:52:04 +0000]
[SCSI] lpfc 8.2.8 v2 : Add sysfs control of target queue depth handling

Added new sysfs attribute lpfc_max_scsicmpl_time. Attribute, when enabled,
will control target queue depth based on I/O completion time.

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

11 years ago[SCSI] lpfc 8.2.8 v2 : Revert target busy in favor of transport disrupted
James Smart [Sun, 7 Sep 2008 15:51:56 +0000]
[SCSI] lpfc 8.2.8 v2 : Revert target busy in favor of transport disrupted

Revert the target busy response in favor of the transport disrupted
response for node state transitions.

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

11 years ago[SCSI] scsi_dh_alua: remove REQ_NOMERGE
Mike Christie [Sat, 6 Sep 2008 13:39:16 +0000]
[SCSI] scsi_dh_alua: remove REQ_NOMERGE

We do not need to set REQ_NOMERGE because when the module calls
blk_execute_rq -> blk_execute_rq_nowait, blk_execute_rq_nowait sets
it for us. This brings all the modules in sync for those bits.

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

11 years ago[SCSI] lpfc 8.2.8 : update driver version to 8.2.8
James Smart [Mon, 25 Aug 2008 01:50:45 +0000]
[SCSI] lpfc 8.2.8 : update driver version to 8.2.8

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

11 years ago[SCSI] lpfc 8.2.8 : Add MSI-X support
James Smart [Mon, 25 Aug 2008 01:50:30 +0000]
[SCSI] lpfc 8.2.8 : Add MSI-X support

Add support for MSI-X Multi-Message interrupts. We use different vectors
for fast-path interrupts (i/o) and slow-patch interrupts (discovery, etc).

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

11 years ago[SCSI] lpfc 8.2.8 : Update driver to use new Host byte error code DID_TRANSPORT_DISRUPTED
James Smart [Mon, 25 Aug 2008 01:50:18 +0000]
[SCSI] lpfc 8.2.8 : Update driver to use new Host byte error code DID_TRANSPORT_DISRUPTED

[jejb: drop rejecting hunk altered by target busy patches]
Signed-off-by: James Smart <james.smart@emulex.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

11 years ago[SCSI] lpfc 8.2.8 : Add support for PCI-EEH permanent disabling
James Smart [Mon, 25 Aug 2008 01:50:11 +0000]
[SCSI] lpfc 8.2.8 : Add support for PCI-EEH permanent disabling

Add support for PCI-EEH permanent-disabling a device via lpfc_pci_remove_one()

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

11 years ago[SCSI] lpfc 8.2.8 : Add new FCOE hardware support
James Smart [Mon, 25 Aug 2008 01:50:06 +0000]
[SCSI] lpfc 8.2.8 : Add new FCOE hardware support

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

11 years ago[SCSI] lpfc 8.2.8 : Miscellaneous Bug Fixes
James Smart [Mon, 25 Aug 2008 01:50:00 +0000]
[SCSI] lpfc 8.2.8 : Miscellaneous Bug Fixes

Miscellaneous Fixes:
- Fix the wrong variable name used for checking node active usage status
- Fix numerous duplicate log message numbers
- Fix change KERN_WARNING messages to KERN_INFO.
- Stop sending erroneous LOGO to fabric after vport is already terminated
- Fix HBQ allocates that were kalloc'ing w/ GFP_KERNEL while holding a lock.
- Fix gcc 4.3.2 compiler warnings and a sparse warning
- Fix bugs in handling unsolicited ct event queue
- Reorder some of the initial link up checks, to remove odd VPI states.
- Correct poor VPI handling
- Add debug messages
- Expand Update_CFG mailbox definition
- Fix handling of VPD data offsets
- Reorder loopback flags
- convert to use offsetof()

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

11 years ago[SCSI] lpfc 8.2.8 : Update driver for new SLI-3 features
James Smart [Mon, 25 Aug 2008 01:49:55 +0000]
[SCSI] lpfc 8.2.8 : Update driver for new SLI-3 features

Update driver for new SLI-3 features:
- interrupt enhancements
- lose adapter doorbell writes
- inlining support for FCP_Ixx cmds

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

11 years ago[SCSI] lpfc 8.2.8 : Miscellaneous Discovery Fixes
James Smart [Mon, 25 Aug 2008 01:49:45 +0000]
[SCSI] lpfc 8.2.8 : Miscellaneous Discovery Fixes

Miscellaneous Discovery fixes:
- Fix rejection followed by acceptance in handling RPL and RPS
  unsolicited events
- Fix for vport delete crash
- Fix PLOGI vs ADISC race condition

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

11 years ago[SCSI] lpfc 8.2.8 : Add kernel-doc function headers
James Smart [Mon, 25 Aug 2008 01:49:00 +0000]
[SCSI] lpfc 8.2.8 : Add kernel-doc function headers

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

11 years ago[SCSI] modify scsi to handle new fail fast flags.
Mike Christie [Tue, 19 Aug 2008 23:45:31 +0000]
[SCSI] modify scsi to handle new fail fast flags.

This checks the errors the scsi-ml determined were retryable
and returns if we should fast fail it based on the request
fail fast flags.

Without the patch, drivers like lpfc, qla2xxx and fcoe would return
DID_ERROR for what it determines is a temporary communication problem.
There is no loss of connectivity at that time and the driver thinks
that it would be fast to retry at the driver level. SCSI-ml will however
sees fast fail on the request and DID_ERROR and will fast fail the io.
This will then cause dm-multipath to fail the path and possibley switch
target controllers when we should be retrying at the scsi layer.

We also were fast failing device errors to dm multiapth when
unless the scsi_dh modules think otherwis we want to retry at
the scsi layer because multipath can only retry the IO like scsi
should have done. multipath is a little dumber though because it
does not what the error was for and assumes that it should fail
the paths.

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

11 years ago[SCSI] block: separate failfast into multiple bits.
Mike Christie [Tue, 19 Aug 2008 23:45:30 +0000]
[SCSI] block: separate failfast into multiple bits.

Multipath is best at handling transport errors. If it gets a device
error then there is not much the multipath layer can do. It will just
access the same device but from a different path.

This patch breaks up failfast into device, transport and driver errors.
The multipath layers (md and dm mutlipath) only ask the lower levels to
fast fail transport errors. The user of failfast, read ahead, will ask
to fast fail on all errors.

Note that blk_noretry_request will return true if any failfast bit
is set. This allows drivers that do not support the multipath failfast
bits to continue to fail on any failfast error like before. Drivers
like scsi that are able to fail fast specific errors can check
for the specific fail fast type. In the next patch I will convert
scsi.

Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Cc: Jens Axboe <jens.axboe@oracle.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

11 years ago[SCSI] qla2xxx: use new host byte transport errors.
Mike Christie [Tue, 19 Aug 2008 23:45:29 +0000]
[SCSI] qla2xxx: use new host byte transport errors.

This has qla2xxx use the new transport error values instead of
DID_BUS_BUSY. I am not sure if all the errors
in qla_isr.c I changed are transport related. We end up blocking/deleting
the rport for all of them so it is better to use the new transport error since
the fc classs will decide when to fail the IO.

With this patch if I pull a cable then IO that had reached
the driver, will be failed with DID_TRANSPORT_DISRUPTED (not including
tape). The fc class will then fail the IO when the fast io fail tmo
has fired, and the driver will flush any other commands running.

Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Cc: Andrew Vasquez <andrew.vasquez@qlogic.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

11 years ago[SCSI] fc class: Add support for new transport errors
Mike Christie [Tue, 19 Aug 2008 23:45:27 +0000]
[SCSI] fc class: Add support for new transport errors

If the target is blocked and fast io fail tmo has not fired
then we requeue with DID_TRANSPORT_DISRUPTED. Once that
tmo fires we fail with DID_TRANSPORT_FAILFAST.

v2
- seperate from
"fc class: unblock target after calling terminate callback"
to make it easier to review.
- Add JamesS's ack from list.
v2
- initial patch

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

11 years ago[SCSI] iscsi class, libiscsi and qla4xxx: convert to new transport host byte values
Mike Christie [Tue, 19 Aug 2008 23:45:26 +0000]
[SCSI] iscsi class, libiscsi and qla4xxx: convert to new transport host byte values

This patch converts the iscsi drivers to the new host byte values.

v2
Drop some conversions. Want to avoid conflicts with other patches.
v1
initial patch.

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

11 years ago[SCSI] scsi: add transport host byte errors (v3)
Mike Christie [Tue, 19 Aug 2008 23:45:25 +0000]
[SCSI] scsi: add transport host byte errors (v3)

Currently, if there is a transport problem the iscsi drivers will return
outstanding commands (commands being exeucted by the driver/fw/hw) with
DID_BUS_BUSY and block the session so no new commands can be queued.
Commands that are caught between the failure handling and blocking are
failed with DID_IMM_RETRY or one of the scsi ml queuecommand return values.
When the recovery_timeout fires, the iscsi drivers then fail IO with
DID_NO_CONNECT.

For fcp, some drivers will fail some outstanding IO (disk but possibly not
tape) with DID_BUS_BUSY or DID_ERROR or some other value that causes a retry
and hits the scsi_error.c failfast check, block the rport, and commands
caught in the race are failed with DID_IMM_RETRY. Other drivers, may
hold onto all IO and wait for the terminate_rport_io or dev_loss_tmo_callbk
to be called.

The following patches attempt to unify what upper layers will see drivers
like multipath can make a good guess. This relies on drivers being
hooked into their transport class.

This first patch just defines two new host byte errors so drivers can
return the same value for when a rport/session is blocked and for
when the fast_io_fail_tmo fires.

The idea is that if the LLD/class detects a problem and is going to block
a rport/session, then if the LLD wants or must return the command to scsi-ml,
then it can return it with DID_TRANSPORT_DISRUPTED. This will requeue
the IO into the same scsi queue it came from, until the fast io fail timer
fires and the class decides what to do.

When using multipath and the fast_io_fail_tmo fires then the class
can fail commands with DID_TRANSPORT_FAILFAST or drivers can use
DID_TRANSPORT_FAILFAST in their terminate_rport_io callbacks or
the equivlent in iscsi if we ever implement more advanced recovery methods.
A LLD, like lpfc, could continue to return DID_ERROR and then it will hit
the normal failfast path, so drivers do not have fully be ported to
work better. The point of the patches is that upper layers will
not see a failure that could be recovered from while the rport/session is
blocked until fast_io_fail_tmo/recovery_timeout fires.

V3
Remove some comments.
V2
Fixed patch/diff errors and renamed DID_TRANSPORT_BLOCKED to
DID_TRANSPORT_DISRUPTED.
V1
initial patch.

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

11 years ago[SCSI] ibmvfc, qla2xxx, lpfc: remove scsi_target_unblock calls in terminate callbacks
Mike Christie [Tue, 19 Aug 2008 23:45:24 +0000]
[SCSI] ibmvfc, qla2xxx, lpfc: remove scsi_target_unblock calls in terminate callbacks

The fc class now calls scsi_target_unblock after calling the
terminate callback, so this patch removes the calls from the
drivers.

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

11 years ago[SCSI] fc class: unblock target after calling terminate callback (take 2)
Mike Christie [Tue, 19 Aug 2008 23:45:23 +0000]
[SCSI] fc class: unblock target after calling terminate callback (take 2)

When we block a rport and the driver implements the terminate
callback we will fail IO that was running quickly. However
IO that was in the scsi_device/block queue sits there until
the dev_loss_tmo fires, and this can make it look like IO is
lost because new IO will get executed but that IO stuck in
the blocked queue sits there for some time longer.

With this patch when the fast io fail tmo fires, we will
fail the blocked IO and any new IO. This patch also allows
all drivers to partially support the fast io fail tmo. If the
terminate io callback is not implemented, we will still fail blocked
IO and any new IO, so multipath can handle that.

This patch also allows the fc and iscsi classes to implement the
same behavior. The timers are just unfornately named differently.

This patch also fixes the problem where drivers were unblocking
the target in their terminate callback, which was needed for
rport removal, but for fast io fail timeout it would cause
IO to bounce arround the scsi/block layer and the LLD queuecommand.
And it for drivers that could have IO stuck but did not have
a terminate callback the unblock calls in the class will fix
them.

v2.
- fix up bit setting style to meet JamesS's pref.
- Broke out new host byte error changes to make it easier to read.
- added JamesS's ack from list.
v1
- initial patch

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

11 years ago[SCSI] lpfc: use SCSI_MLQUEUE_TARGET_BUSY when catching the rport transition race
Mike Christie [Sun, 17 Aug 2008 20:24:41 +0000]
[SCSI] lpfc: use SCSI_MLQUEUE_TARGET_BUSY when catching the rport transition race

We do want to call right back into the queuecommand during the race,
so we can just use SCSI_MLQUEUE_TARGET_BUSY.

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

11 years ago[SCSI] libiscsi: Use SCSI_MLQUEUE_TARGET_BUSY
Mike Christie [Sun, 17 Aug 2008 20:24:43 +0000]
[SCSI] libiscsi: Use SCSI_MLQUEUE_TARGET_BUSY

For the conditions below we do not want the queuecommand
function to call us right back, so return SCSI_MLQUEUE_TARGET_BUSY.

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

11 years ago[SCSI] qla2xxx: return SCSI_MLQUEUE_TARGET_BUSY when driver has detected rport error...
Mike Christie [Sun, 17 Aug 2008 20:24:40 +0000]
[SCSI] qla2xxx: return SCSI_MLQUEUE_TARGET_BUSY when driver has detected rport error or race

If the fcport is not online then we do not want to block IO to all ports on
the host. We just want to stop IO on port not online, so we should be using
the SCSI_MLQUEUE_TARGET_BUSY return value.

For the case where we race with the rport memset initialization
we do not want the queuecommand to be called again so we can just use
SCSI_MLQUEUE_TARGET_BUSY for this.

Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Acked-by: Andrew Vasquez <andrew.vasquez@qlogic.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

11 years ago[SCSI] qla4xxx: return SCSI_MLQUEUE_TARGET_BUSY when driver has detected session...
Mike Christie [Sun, 17 Aug 2008 20:24:39 +0000]
[SCSI] qla4xxx: return SCSI_MLQUEUE_TARGET_BUSY when driver has detected session error

When qla4xxx begins recovery and the iscsi class is firing up to handle
it, we need to retrn SCSI_MLQUEUE_TARGET_BUSY from the driver instead
of host busy, because the session recovery only affects the one target.

Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Acked-by: David C Somayajulu <david.somayajulu@qlogic.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

11 years ago[SCSI] Add helper code so transport classes/driver can control queueing (v3)
Mike Christie [Sun, 17 Aug 2008 20:24:38 +0000]
[SCSI] Add helper code so transport classes/driver can control queueing (v3)

SCSI-ml manages the queueing limits for the device and host, but
does not do so at the target level. However something something similar
can come in userful when a driver is transitioning a transport object to
the the blocked state, becuase at that time we do not want to queue
io and we do not want the queuecommand to be called again.

The patch adds code similar to the exisiting SCSI_ML_*BUSY handlers.
You can now return SCSI_MLQUEUE_TARGET_BUSY when we hit
a transport level queueing issue like the hw cannot allocate some
resource at the iscsi session/connection level, or the target has temporarily
closed or shrunk the queueing window, or if we are transitioning
to the blocked state.

bnx2i, when they rework their firmware according to netdev
developers requests, will also need to be able to limit queueing at this
level. bnx2i will hook into libiscsi, but will allocate a scsi host per
netdevice/hba, so unlike pure software iscsi/iser which is allocating
a host per session, it cannot set the scsi_host->can_queue and return
SCSI_MLQUEUE_HOST_BUSY to reflect queueing limits on the transport.

The iscsi class/driver can also set a scsi_target->can_queue value which
reflects the max commands the driver/class can support. For iscsi this
reflects the number of commands we can support for each session due to
session/connection hw limits, driver limits, and to also reflect the
session/targets's queueing window.

Changes:
v1 - initial patch.
v2 - Fix scsi_run_queue handling of multiple blocked targets.
Previously we would break from the main loop if a device was added back on
the starved list. We now run over the list and check if any target is
blocked.
v3 - Rediff for scsi-misc.

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

11 years agoMAINTAINERS: remove colon from headings
Randy Dunlap [Sun, 12 Oct 2008 23:11:58 +0000]
MAINTAINERS: remove colon from headings

Remove ending ':' from some of the Topic lines for consistency.

Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

11 years agoMAINTAINERS: add F: and acronyms
Randy Dunlap [Sun, 12 Oct 2008 23:11:45 +0000]
MAINTAINERS: add F: and acronyms

Add F: definition since it is being used.

Expand acronyms or add them so that users of MAINTAINERS file can find
entries more easily.

Correct driver spelling.

Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

11 years agoMAINTAINERS: alpha sort
Randy Dunlap [Sun, 12 Oct 2008 23:11:31 +0000]
MAINTAINERS: alpha sort

Sorted case-insensitive (sort -f).
Groups that begin with ARM, INTEL, etc. not sorted.

Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

11 years agoMerge branch 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4
Linus Torvalds [Sun, 12 Oct 2008 23:10:29 +0000]
Merge branch 'for_linus' of git://git./linux/kernel/git/tytso/ext4

* 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4:
  ext4: fix kconfig typo and extra whitespace
  ext4: fix build failure without procfs
  ext4: add an option to control error handling on file data
  jbd2: don't dirty original metadata buffer on abort
  ext4: add checks for errors from jbd2
  jbd2: fix error handling for checkpoint io
  jbd2: abort when failed to log metadata buffers

11 years agoext4: fix kconfig typo and extra whitespace
Jan Engelhardt [Sun, 12 Oct 2008 19:53:01 +0000]
ext4: fix kconfig typo and extra whitespace

Signed-off-by: Jan Engelhardt <jengelh@medozas.de>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>

11 years agoext4: fix build failure without procfs
Alexander Beregalov [Sun, 12 Oct 2008 21:27:49 +0000]
ext4: fix build failure without procfs

fs/ext4/super.c: In function 'ext4_fill_super':
fs/ext4/super.c:2226: error: 'ext4_ui_proc_fops' undeclared (first use
in this function)
fs/ext4/super.c:2226: error: (Each undeclared identifier is reported
only once
fs/ext4/super.c:2226: error: for each function it appears in.)

Signed-off-by: Alexander Beregalov <a.beregalov@gmail.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>

11 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/sfrench/cifs-2.6
Linus Torvalds [Sun, 12 Oct 2008 19:42:29 +0000]
Merge git://git./linux/kernel/git/sfrench/cifs-2.6

* git://git.kernel.org/pub/scm/linux/kernel/git/sfrench/cifs-2.6:
  [CIFS] cifs: remove pointless lock and unlock of GlobalMid_Lock in header_assemble

11 years agoFix Kconfig dependency for WM97xx battery driver
Linus Torvalds [Sun, 12 Oct 2008 19:37:16 +0000]
Fix Kconfig dependency for WM97xx battery driver

It needs the WM97xx touchscreen driver to be compiled in, not as a
module.  And it cannot itself be a module, due to some unfortunate
interactions with platform data.

Reported-by: Herbert Xu <herbert@gondor.apana.org.au>
Acked-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

11 years agoMerge branch 'x86-core-v2-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 12 Oct 2008 19:04:59 +0000]
Merge branch 'x86-core-v2-for-linus' of git://git./linux/kernel/git/tip/linux-2.6-tip

This merges in:

  x86/build, x86/microcode, x86/spinlocks, x86/memory-corruption-check,
  x86/early-printk, x86/xsave, x86/quirks, x86/setup, x86/signal,
  core/signal, x86/urgent, x86/xen

* 'x86-core-v2-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip: (142 commits)
  x86: make processor type select depend on CONFIG_EMBEDDED
  x86: extend processor type select help text
  x86, amd-iommu: propagate PCI device enabling error
  warnings: fix arch/x86/kernel/io_apic_64.c
  warnings: fix arch/x86/kernel/early_printk.c
  x86, fpu: check __clear_user() return value
  x86: memory corruption check - cleanup
  x86: ioperm user_regset
  xen: do not reserve 2 pages of padding between hypervisor and fixmap.
  xen: use spin_lock_nest_lock when pinning a pagetable
  x86: xsave: set FP, SSE bits in the xsave header in the user sigcontext
  x86: xsave: fix error condition in save_i387_xstate()
  x86: SB450: deprioritize DMI quirks
  x86: SB450: skip IRQ0 override if it is not routed to INT2 of IOAPIC
  x86: replace a magic number with a named constant in the VESA boot code
  x86 setup: remove IMAGE_OFFSET
  x86 setup: remove DEF_INITSEG and DEF_SETUPSEG
  Revert "x86: fix ghost EDD devices in /sys again"
  x86 setup: fix ghost entries under /sys/firmware/edd take 3
  x86: signal: remove indent in restore_sigcontext()
  ...

11 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/hskinnemoen/avr32-2.6
Linus Torvalds [Sun, 12 Oct 2008 19:00:23 +0000]
Merge git://git./linux/kernel/git/hskinnemoen/avr32-2.6

* git://git.kernel.org/pub/scm/linux/kernel/git/hskinnemoen/avr32-2.6:
  avr32: Fix build failures in board code
  avr32: Allow selecting multiple pins at once
  avr32: Minor pm_power_off cleanup
  avr32: Implement {read,write}[bwl]_be
  avr32: Replace static clock list with dynamic linked list
  avr32: Use platform_driver_probe for pdc platform driver
  avr32: Use platform_driver_probe for pio platform driver
  avr32: Provide a way to deselect pins in the portmux
  ngw100: export J15 through sysfs
  avr32: Allow fine-grained control over LCDC pins
  avr32: added mem kernel command line option support
  Add kernel support for oprofile callgraphs on AVR32
  avr32: use the new byteorder headers

11 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/drzeus/mmc
Linus Torvalds [Sun, 12 Oct 2008 18:51:57 +0000]
Merge branch 'for-linus' of git://git./linux/kernel/git/drzeus/mmc

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/drzeus/mmc: (24 commits)
  MMC: Use timeout values from CSR
  MMC: CSD and CID timeout values
  sdhci: 'scratch' may be used uninitialized
  mmc: explicitly mention SDIO support in Kconfig
  mmc: remove redundant "depends on"
  Fix comment in include/linux/mmc/host.h
  sdio: high-speed support
  mmc_block: hard code 512 byte block size
  sdhci: force high speed capability on some controllers
  mmc_block: filter out PC requests
  mmc_block: indicate strict ordering
  mmc_block: inform block layer about sector count restriction
  sdio: give sdio irq thread a host specific name
  sdio: make sleep on error interruptable
  sdhci: reduce card detection delay
  sdhci: let the controller wait for busy state to end
  atmel-mci: Add missing flush_dcache_page() in PIO transfer code
  atmel-mci: Don't overwrite error bits when NOTBUSY is set
  atmel-mci: Add experimental DMA support
  atmel-mci: support multiple mmc slots
  ...

11 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/wim/linux-2.6-watchdog
Linus Torvalds [Sun, 12 Oct 2008 18:51:32 +0000]
Merge git://git./linux/kernel/git/wim/linux-2.6-watchdog

* git://git.kernel.org/pub/scm/linux/kernel/git/wim/linux-2.6-watchdog:
  [WATCHDOG] orion5x_wdt.c: add spinlocking
  [WATCHDOG] Orion: add hardware watchdog support
  [WATCHDOG] omap_wdt.c: cleanup a bit omap_wdt.c
  [WATCHDOG] omap_wdt.c: another ioremap() fix
  [WATCHDOG] omap_wdt.c: sync linux-omap changes
  [WATCHDOG] Add AT91SAM9X watchdog
  [WATCHDOG] Add driver for winbond w83697ug/uf watchdog feature
  [WATCHDOG] add watchdog driver IT8716 IT8726 IT8712J/K

11 years agoMerge branch 'x86-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sun, 12 Oct 2008 18:50:37 +0000]
Merge branch 'x86-fixes-for-linus' of git://git./linux/kernel/git/tip/linux-2.6-tip

* 'x86-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
  x86: fix early panic on amd64 due to typo in supported CPU section
  x86, early_ioremap: fix fencepost error
  x86: avoid dereferencing beyond stack + THREAD_SIZE

11 years agoprovide generic_block_fiemap() only with BLOCK=y
Adrian Bunk [Sun, 12 Oct 2008 04:15:19 +0000]
provide generic_block_fiemap() only with BLOCK=y

This fixes the following compile error with CONFIG_BLOCK=n caused by
commit 68c9d702bb72f367f3b148963ec6cf5e07ff7f65 ("generic block based
fiemap implementation"):

CC fs/ioctl.o
  fs/ioctl.c: In function 'generic_block_fiemap':
  fs/ioctl.c:249: error: storage size of 'tmp' isn't known
  fs/ioctl.c:272: error: invalid application of 'sizeof' to incomplete type 'struct buffer_head'
  fs/ioctl.c:280: error: implicit declaration of function 'buffer_mapped'
  fs/ioctl.c:249: warning: unused variable 'tmp'
  make[2]: *** [fs/ioctl.o] Error 1

Signed-off-by: Adrian Bunk <bunk@kernel.org>
Acked-by: Josef Bacik <jbacik@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

11 years agoadd key_revoke() dummy for KEYS=n
Adrian Bunk [Sun, 12 Oct 2008 04:10:50 +0000]
add key_revoke() dummy for KEYS=n

This fixes the following build error with CONFIG_KEYS=n, caused by
commit dfd15c46a6c2cafb006183c0c14f07e59eee4ac0 ("cifs: explicitly
revoke SPNEGO key after session setup"):

CC [M] fs/cifs/sess.o
  fs/cifs/sess.c: In function 'CIFS_SessSetup':
  fs/cifs/sess.c:628: error: implicit declaration of function 'key_revoke'
  make[3]: *** [fs/cifs/sess.o] Error 1

Signed-off-by: Adrian Bunk <bunk@kernel.org>
Acked-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

11 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/sparc-next-2.6
Linus Torvalds [Sun, 12 Oct 2008 18:40:55 +0000]
Merge git://git./linux/kernel/git/davem/sparc-next-2.6

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/sparc-next-2.6: (180 commits)
  leo: disable cursor when leaving graphics mode
  cg6: disable cursor when leaving graphics mode
  sparc32: sun4m interrupt mask cleanup
  drivers/rtc/Kconfig: don't build rtc-cmos.o on sparc32
  sparc: arch/sparc/kernel/pmc.c -- extra #include?
  sparc32: Add more extensive documentation of sun4m interrupts.
  sparc32: Kill irq_rcvreg from sun4m_irq.c
  sparc32: Delete master_l10_limit.
  sparc32: Use PROM device probing for sun4c timers.
  sparc32: Use PROM device probing for sun4c interrupt register.
  sparc32: Delete claim_ticker14().
  sparc32: Stop calling claim_ticker14() from sun4c_irq.c
  sparc32: Kill clear_profile_irq btfixup entry.
  sparc32: Call sun4m_clear_profile_irq() directly from sun4m_smp.c
  sparc32: Remove #if 0'd code from sun4c_irq.c
  sparc32: Remove some SMP ifdefs in sun4d_irq.c
  sparc32: Use PROM infrastructure for probing and mapping sun4d timers.
  sparc32: Use PROM device probing for sun4m irq registers.
  sparc32: Use PROM device probing for sun4m timer registers.
  sparc: Fix user_regset 'n' field values.
  ...

11 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6
Linus Torvalds [Sun, 12 Oct 2008 18:40:34 +0000]
Merge git://git./linux/kernel/git/davem/net-2.6

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6:
  smc911x: Fix external PHY detection
  e1000: allow VLAN devices to use TSO and CSUM offload
  gre: Initialise rtnl_link tunnel parameters properly
  ipvs: Add proper dependencies on IP_VS, and fix description header line.

11 years agoFix RTC wakealarm sysfs interface breakage.
Linus Torvalds [Sun, 12 Oct 2008 18:30:08 +0000]
Fix RTC wakealarm sysfs interface breakage.

Commit ed458df4d2470adc02762a87a9ad665d0b1a2bd4 ("PnP: move
pnpacpi/pnpbios_init to after PCI init") moved the PnP RTC discovery
later, and now the ACPI RTC glue code doesn't find it any more, breaking
the RTC wakealarm sysfs interfaces, as reported by Rafael.

This really is fairly messy, and we have several annoying ordering
constraints here - the PnP code that sets up the RTC resources wants to
run after the PCI resources have to be registered, which in turn needs
to run after ACPI has at least enumerated the root PCI buses etc.  Our
initcall ordering is not fine-grained enough to make this all painless.

So this moves the ACPI RTC glue ("acpi_rtc_init()") down to a regular
module call, which fixes the problem Rafael has.  The reason this isn't
wonderful is that we really should do acpi_rtc_init before we do the
rtc_cmos init, and now those two are in the same module_init() section.

Which happens to work, but only because drivers/rtc is linked after
drivers/acpi.  In other words, we still have a very subtle ordering
issue here. Grr.

Reported-and-tested-by: Rafael J. Wysocki <rjw@sisk.pl>
Acked-by: David Brownell <david-b@pacbell.net>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

11 years agoavr32: Fix build failures in board code
Haavard Skinnemoen [Sun, 12 Oct 2008 14:02:46 +0000]
avr32: Fix build failures in board code

Fix a few instances of board code breakage introduced by the atmel-mci
platform interface changes.

Signed-off-by: Haavard Skinnemoen <haavard.skinnemoen@atmel.com>

11 years agoavr32: Allow selecting multiple pins at once
Julien May [Wed, 24 Sep 2008 08:30:47 +0000]
avr32: Allow selecting multiple pins at once

at32_select_periph() now takes an u32 bitmask rather than a single pin.
This allows to set multiple pins at once.

Signed-off-by: Alex Raimondi <mailinglist@miromico.ch>
Signed-off-by: Haavard Skinnemoen <haavard.skinnemoen@atmel.com>

11 years agoMerge branches 'x86/xen', 'x86/build', 'x86/microcode', 'x86/mm-debug-v2', 'x86/memor...
Ingo Molnar [Sun, 12 Oct 2008 13:50:02 +0000]
Merge branches 'x86/xen', 'x86/build', 'x86/microcode', 'x86/mm-debug-v2', 'x86/memory-corruption-check', 'x86/early-printk', 'x86/xsave', 'x86/ptrace-v2', 'x86/quirks', 'x86/setup', 'x86/spinlocks' and 'x86/signal' into x86/core-v2

11 years agoMerge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/hskinnemoen...
Haavard Skinnemoen [Sun, 12 Oct 2008 13:44:33 +0000]
Merge branch 'master' of git://git./linux/kernel/git/hskinnemoen/atmel-mci-2.6.28

11 years agox86: make processor type select depend on CONFIG_EMBEDDED
Ingo Molnar [Sun, 12 Oct 2008 13:40:45 +0000]
x86: make processor type select depend on CONFIG_EMBEDDED

deselecting one of the CPU type CONFIG_CPU_SUP_* config options
can render a kernel unbootable. Make sure this option is only
available if CONFIG_EMBEDDED is enabled.

Signed-off-by: Ingo Molnar <mingo@elte.hu>

11 years agox86: extend processor type select help text
Ingo Molnar [Sun, 12 Oct 2008 13:36:24 +0000]
x86: extend processor type select help text

extend the help text of the CONFIG_CPU_SUP_* config options to
express what it does and what effects it has.

Signed-off-by: Ingo Molnar <mingo@elte.hu>

11 years ago[CIFS] cifs: remove pointless lock and unlock of GlobalMid_Lock in header_assemble
Jeff Layton [Sun, 12 Oct 2008 13:34:11 +0000]
[CIFS] cifs: remove pointless lock and unlock of GlobalMid_Lock in header_assemble

We lock GlobalMid_Lock in header_assemble and then immediately unlock it
again without doing anything. Not sure what this was intended to do, but
remove it.

Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Steve French <sfrench@us.ibm.com>

11 years agox86, amd-iommu: propagate PCI device enabling error
Ingo Molnar [Sun, 12 Oct 2008 13:24:53 +0000]
x86, amd-iommu: propagate PCI device enabling error

propagate an error in enabling the PCI device.

Also eliminates this warning:

 arch/x86/kernel/amd_iommu_init.c: In function ‘init_iommu_one’:
 arch/x86/kernel/amd_iommu_init.c:726: warning: ignoring return value of ‘pci_enable_device’, declared with attribute warn_unused_result

Signed-off-by: Ingo Molnar <mingo@elte.hu>

11 years agowarnings: fix arch/x86/kernel/io_apic_64.c
Ingo Molnar [Sun, 12 Oct 2008 13:22:22 +0000]
warnings: fix arch/x86/kernel/io_apic_64.c

fix:

 arch/x86/kernel/io_apic_64.c: In function ‘print_local_APIC’:
 arch/x86/kernel/io_apic_64.c:1284: warning: format ‘%08x’ expects type ‘unsigned int’, but argument 2 has type ‘long unsigned int’
 arch/x86/kernel/io_apic_64.c:1285: warning: format ‘%08x’ expects type ‘unsigned int’, but argument 2 has type ‘long unsigned int’

We want to print the two halves of 'icr' at 32 bit width.

Signed-off-by: Ingo Molnar <mingo@elte.hu>

11 years agowarnings: fix arch/x86/kernel/early_printk.c
Ingo Molnar [Wed, 27 Aug 2008 08:37:14 +0000]
warnings: fix arch/x86/kernel/early_printk.c

fix warning:

  arch/x86/kernel/early_printk.c:993: warning: ‘enable_debug_console’ defined but not used

Eliminate dead code.

Signed-off-by: Ingo Molnar <mingo@elte.hu>

11 years agox86, fpu: check __clear_user() return value
Ingo Molnar [Mon, 18 Aug 2008 10:59:32 +0000]
x86, fpu: check __clear_user() return value

fix warning:

  arch/x86/kernel/xsave.c: In function ‘save_i387_xstate’:
  arch/x86/kernel/xsave.c:98: warning: ignoring return value of ‘__clear_user’, declared with attribute warn_unused_result

check the return value and act on it. We should not be ignoring faults
at this point.

Signed-off-by: Ingo Molnar <mingo@elte.hu>

11 years agoMerge branch 'linus' into x86/xsave
Ingo Molnar [Sun, 12 Oct 2008 13:17:14 +0000]
Merge branch 'linus' into x86/xsave

11 years agox86: memory corruption check - cleanup
Ingo Molnar [Sun, 12 Oct 2008 13:06:29 +0000]
x86: memory corruption check - cleanup

Move the prototypes from the generic kernel.h header to the more
appropriate include/asm-x86/bios_ebda.h header file.

Also, remove the check from the power management code - this is a
pure x86 matter for now.

Signed-off-by: Ingo Molnar <mingo@elte.hu>

11 years agoMerge branch 'linus' into x86/memory-corruption-check
Ingo Molnar [Sun, 12 Oct 2008 13:05:39 +0000]
Merge branch 'linus' into x86/memory-corruption-check

11 years agoMerge branches 'core/signal' and 'x86/spinlocks' into x86/xen
Ingo Molnar [Sun, 12 Oct 2008 11:20:25 +0000]
Merge branches 'core/signal' and 'x86/spinlocks' into x86/xen

Conflicts:
include/asm-x86/spinlock.h

11 years agoMerge branch 'x86/signal' into core/signal
Ingo Molnar [Sun, 12 Oct 2008 11:17:07 +0000]
Merge branch 'x86/signal' into core/signal

11 years agoMerge branch 'linus' into x86/signal
Ingo Molnar [Sun, 12 Oct 2008 10:49:27 +0000]
Merge branch 'linus' into x86/signal

Conflicts:
arch/x86/kernel/signal_64.c

11 years agoMerge branch 'linus' into x86/quirks
Ingo Molnar [Sun, 12 Oct 2008 10:43:21 +0000]
Merge branch 'linus' into x86/quirks

Conflicts:
arch/x86/kernel/early-quirks.c

11 years agoMerge branch 'linus' into x86/spinlocks
Ingo Molnar [Sun, 12 Oct 2008 10:39:30 +0000]
Merge branch 'linus' into x86/spinlocks

Done to prevent this failure of an Octopus merge:

 Added arch/arm/include/asm/byteorder.h in both, but differently.
 ERROR: Merge conflict in arch/arm/include/asm/byteorder.h
 Auto-merging include/asm-x86/spinlock.h
 ERROR: Merge conflict in include/asm-x86/spinlock.h
 fatal: merge program failed

11 years agoMerge branch 'linus' into x86/xen
Ingo Molnar [Sun, 12 Oct 2008 10:35:23 +0000]
Merge branch 'linus' into x86/xen

Conflicts:
arch/x86/kernel/cpu/common.c
arch/x86/kernel/process_64.c
arch/x86/xen/enlighten.c

11 years agox86: ioperm user_regset
Roland McGrath [Fri, 8 Aug 2008 22:58:39 +0000]
x86: ioperm user_regset

This adds a user_regset type for the x86 io permissions bitmap.
This makes it appear in core dumps (when ioperm has been used).
It will also make it visible to debuggers in the future.

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
[conflict resolutions: Signed-off-by: Ingo Molnar <mingo@elte.hu> ]

11 years agoMerge branch 'x86/urgent' into core/signal
Ingo Molnar [Sun, 12 Oct 2008 09:32:17 +0000]
Merge branch 'x86/urgent' into core/signal

Conflicts:
arch/x86/kernel/signal_64.c

11 years agox86: fix early panic on amd64 due to typo in supported CPU section
Petr Vandrovec [Sun, 12 Oct 2008 08:51:03 +0000]
x86: fix early panic on amd64 due to typo in supported CPU section

Do not crash when enumerating supported CPU architectures

SECURITY_INIT somehow ended up in x86_cpu_dev.init section.  That caused printk
in code which prints supported architectures to hit #GP due to non-canonical
address being used.

Signed-off-by: Petr Vandrovec <petr@vandrovec.name>
Cc: thomas.petazzoni@free-electrons.com
Signed-off-by: Ingo Molnar <mingo@elte.hu>

11 years agox86, early_ioremap: fix fencepost error
Alan Cox [Fri, 10 Oct 2008 09:46:45 +0000]
x86, early_ioremap: fix fencepost error

The x86 implementation of early_ioremap has an off by one error. If we get
an object which ends on the first byte of a page we undermap by one page and
this causes a crash on boot with the ASUS P5QL whose DMI table happens to fit
this alignment.

The size computation is currently

last_addr = phys_addr + size - 1;
npages = (PAGE_ALIGN(last_addr) - phys_addr)

(Consider a request for 1 byte at alignment 0...)

Closes #11693

Debugging work by Ian Campbell/Felix Geyer

Signed-off-by: Alan Cox <alan@rehat.com>
Cc: <stable@kernel.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

11 years agox86: avoid dereferencing beyond stack + THREAD_SIZE
David Rientjes [Tue, 7 Oct 2008 21:15:11 +0000]
x86: avoid dereferencing beyond stack + THREAD_SIZE

It's possible for get_wchan() to dereference past task->stack + THREAD_SIZE
while iterating through instruction pointers if fp equals the upper boundary,
causing a kernel panic.

Signed-off-by: David Rientjes <rientjes@google.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

11 years agoMerge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/hskinnemoen...
Pierre Ossman [Sun, 12 Oct 2008 09:08:46 +0000]
Merge branch 'master' of git://git./linux/kernel/git/hskinnemoen/atmel-mci-2.6.28

Conflicts:
drivers/mmc/host/atmel-mci.c

11 years agoMMC: Use timeout values from CSR
Matthew Fleming [Thu, 2 Oct 2008 11:21:42 +0000]
MMC: Use timeout values from CSR

Hard-coded timeout values of 250ms for writes and 100ms for reads are
currently used for MMC transactions over SPI. The spec states that the
timeout values from the card should be used.

Signed-off-by: Matthew Fleming <matthew.fleming@imgtec.com>
Signed-off-by: Pierre Ossman <drzeus@drzeus.cx>

11 years agoMMC: CSD and CID timeout values
Matthew Fleming [Thu, 2 Oct 2008 11:24:05 +0000]
MMC: CSD and CID timeout values

The MMC spec states that the timeout for accessing the CSD and CID
registers is 64 clock cycles.

Signed-off-by: Matthew Fleming <matthew.fleming@imgtec.com>
Signed-off-by: Pierre Ossman <drzeus@drzeus.cx>

11 years agosdhci: 'scratch' may be used uninitialized
Steven Noonan [Wed, 1 Oct 2008 08:50:25 +0000]
sdhci: 'scratch' may be used uninitialized

The variable 'scratch' is always initialized before it's used. The
conditional which is responsible for initialization of 'scratch' will
always evaluate 'true' when the first loop iteration occurs, and thus,
it's properly initialized. GCC doesn't see this, of course, so using
the uninitialized_var() macro seems to work for silencing this case.

Signed-off-by: Steven Noonan <steven@uplinklabs.net>
Signed-off-by: Pierre Ossman <drzeus@drzeus.cx>

11 years agommc: explicitly mention SDIO support in Kconfig
Pierre Ossman [Sun, 7 Sep 2008 11:16:58 +0000]
mmc: explicitly mention SDIO support in Kconfig

Signed-off-by: Pierre Ossman <drzeus@drzeus.cx>

11 years agommc: remove redundant "depends on"
Pierre Ossman [Sun, 7 Sep 2008 11:07:57 +0000]
mmc: remove redundant "depends on"

Signed-off-by: Pierre Ossman <drzeus@drzeus.cx>

11 years agoFix comment in include/linux/mmc/host.h
Thomas Petazzoni [Tue, 2 Sep 2008 08:14:13 +0000]
Fix comment in include/linux/mmc/host.h

In include/linux/mmc/host.h, it is mentionned that the callback to
know if a card is present or not is get_ro(). But it's get_cd().

Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Signed-off-by: Pierre Ossman <drzeus@drzeus.cx>

11 years agosdio: high-speed support
Pierre Ossman [Sun, 31 Aug 2008 15:22:46 +0000]
sdio: high-speed support

Signed-off-by: Pierre Ossman <drzeus@drzeus.cx>

11 years agommc_block: hard code 512 byte block size
Pierre Ossman [Sun, 31 Aug 2008 12:10:08 +0000]
mmc_block: hard code 512 byte block size

We use 512 byte blocks on all cards, and newer cards support nothing
else, so hard code it and make the code less complex.

Signed-off-by: Pierre Ossman <drzeus@drzeus.cx>

11 years agosdhci: force high speed capability on some controllers
Pierre Ossman [Sat, 16 Aug 2008 18:43:04 +0000]
sdhci: force high speed capability on some controllers

Some high speed capable controllers forget to set the high speed
capability bit. Make sure we enable the functionality anyway.

Signed-off-by: Pierre Ossman <drzeus@drzeus.cx>

11 years agommc_block: filter out PC requests
Pierre Ossman [Sat, 16 Aug 2008 18:43:48 +0000]
mmc_block: filter out PC requests

We do not support PC (SCSI) commands, so don't pretend we do by
letting them through.

Signed-off-by: Pierre Ossman <drzeus@drzeus.cx>

11 years agommc_block: indicate strict ordering
Pierre Ossman [Sat, 16 Aug 2008 19:15:50 +0000]
mmc_block: indicate strict ordering

The MMC block driver services requests one at a time and in strict
order. Indicate this to the block layer so that it can handle barriers
in an efficient manner.

Signed-off-by: Pierre Ossman <drzeus@drzeus.cx>

11 years agommc_block: inform block layer about sector count restriction
Pierre Ossman [Sat, 16 Aug 2008 19:34:02 +0000]
mmc_block: inform block layer about sector count restriction

Make sure we consider the maximum block count when we tell the block
layer about the maximum sector count. That way we don't have to chop
up the request ourselves.

Signed-off-by: Pierre Ossman <drzeus@drzeus.cx>

11 years agosdio: give sdio irq thread a host specific name
Pierre Ossman [Sun, 31 Aug 2008 11:42:00 +0000]
sdio: give sdio irq thread a host specific name

There is one thread per host controller so make sure they all get
unique names.

Signed-off-by: Pierre Ossman <drzeus@drzeus.cx>

11 years agosdio: make sleep on error interruptable
Pierre Ossman [Sun, 31 Aug 2008 11:38:54 +0000]
sdio: make sleep on error interruptable

Make sure we can be woken from the forced sleep that is done on errors.
Removing a card often results in -ENOMEDIUM or -EILSEQ so we previously
locked up the removal process for a second.

We could completely exit on -ENOMEDIUM, but it might be a transient
glitch so treat it like any other error.

Signed-off-by: Pierre Ossman <drzeus@drzeus.cx>

11 years agosdhci: reduce card detection delay
Pierre Ossman [Mon, 18 Aug 2008 20:18:14 +0000]
sdhci: reduce card detection delay

The card detection delay was added early when the behaviour of the
card interrupt was still very much unknown (i.e. before there was a
public specification). As it is now known that it is a debounced signal,
reduce the delay to something more sensible.

Signed-off-by: Pierre Ossman <drzeus@drzeus.cx>

11 years agosdhci: let the controller wait for busy state to end
Pierre Ossman [Thu, 24 Jul 2008 23:09:08 +0000]
sdhci: let the controller wait for busy state to end

The sdhci controllers can interrupt us when the busy state from the
card has ended, saving CPU cycles and power.

Signed-off-by: Pierre Ossman <drzeus@drzeus.cx>

11 years agosmc911x: Fix external PHY detection
Guennadi Liakhovetski [Sat, 11 Oct 2008 22:07:16 +0000]
smc911x: Fix external PHY detection

If an external PHY is found the driver falls through to the default
case in the switch and overwrites the PHY ID. Add the missing break.

Signed-off-by: Guennadi Liakhovetski <lg@denx.de>
Signed-off-by: David S. Miller <davem@davemloft.net>

11 years agoMerge branch 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4
Linus Torvalds [Sat, 11 Oct 2008 20:23:48 +0000]
Merge branch 'for_linus' of git://git./linux/kernel/git/tytso/ext4

* 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4: (43 commits)
  ext4: Rename ext4dev to ext4
  ext4: Avoid double dirtying of super block in ext4_put_super()
  Update ext4 MAINTAINERS file
  Hook ext4 to the vfs fiemap interface.
  generic block based fiemap implementation
  ocfs2: fiemap support
  vfs: vfs-level fiemap interface
  ext4: fix xattr deadlock
  jbd2: Fix buffer head leak when writing the commit block
  ext4: Add debugging markers that can be used by systemtap
  jbd2: abort instead of waiting for nonexistent transaction
  ext4: fix initialization of UNINIT bitmap blocks
  ext4: Remove old legacy block allocator
  ext4: Use readahead when reading an inode from the inode table
  ext4: Improve the documentation for ext4's /proc tunables
  ext4: Combine proc file handling into a single set of functions
  ext4: move /proc setup and teardown out of mballoc.c
  ext4: Don't use 'struct dentry' for internal lookups
  ext4/jbd2: Avoid WARN() messages when failing to write to the superblock
  ext4: use percpu data structures for lg_prealloc_list
  ...

11 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/bart/ide-2.6
Linus Torvalds [Sat, 11 Oct 2008 20:22:33 +0000]
Merge git://git./linux/kernel/git/bart/ide-2.6

* git://git.kernel.org/pub/scm/linux/kernel/git/bart/ide-2.6: (71 commits)
  ide: Remove ide_spin_wait_hwgroup() and use special requests instead
  ide: move IDE{FLOPPY,TAPE}_WAIT_CMD defines to <linux/ide.h>
  ide: add ide_do_test_unit_ready() helper
  ide: add ide_do_start_stop() helper
  ide: add ide_set_media_lock() helper
  ide-floppy: move floppy ioctls handling to ide-floppy_ioctl.c
  ide-floppy: ->{srfp,wp} -> IDE_AFLAG_{SRFP,WP}
  ide: add ide_queue_pc_tail() helper
  ide: add ide_queue_pc_head() helper
  ide: add ide_init_pc() helper
  ide-tape: add ide_tape_set_media_lock() helper
  ide-floppy: add ide_floppy_set_media_lock() helper
  ide: add ide_io_buffers() helper
  ide-scsi: cleanup ide_scsi_io_buffers()
  ide-floppy: remove MODE_SENSE_* defines
  ide-{floppy,tape}: remove packet command stack
  ide-{floppy,tape}: remove request stack
  ide-generic: handle probing of legacy io-ports v5
  ide-floppy: use scatterlists for pio transfers
  ide-tape: remove idetape_init_rq()
  ...

11 years agoMerge branch 'master' of master.kernel.org:/pub/scm/linux/kernel/git/torvalds/linux-2.6
David S. Miller [Sat, 11 Oct 2008 19:39:35 +0000]
Merge branch 'master' of /linux/kernel/git/torvalds/linux-2.6

Conflicts:

sound/core/memalloc.c

11 years agoe1000: allow VLAN devices to use TSO and CSUM offload
Patrick McHardy [Sat, 11 Oct 2008 19:25:59 +0000]
e1000: allow VLAN devices to use TSO and CSUM offload

This patch changes e1000 to set vlan_features so TSO and CSUM
offload can be used by VLAN devices, similar as with the other
Intel drivers.

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

11 years agogre: Initialise rtnl_link tunnel parameters properly
Herbert Xu [Sat, 11 Oct 2008 19:20:15 +0000]
gre: Initialise rtnl_link tunnel parameters properly

Brown paper bag error of calling memset with sizeof(p) instead
of sizeof(*p).

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>