[SCSI] scsi_scan.c: Release mutex in error handling code

The mutex is released on a successful return, so it would seem that it
should be released on an error return as well.

The semantic patch that makes this change is as follows:
(http://www.emn.fr/x-info/coccinelle/)

// <smpl>
@@
expression l;
@@

mutex_lock(l);
... when != mutex_unlock(l)
    when any
    when strict
(
if (...) { ... when != mutex_unlock(l)
+   mutex_unlock(l);
    return ...;
}
|
mutex_unlock(l);
)
// </smpl>

Signed-off-by: Julia Lawall <julia@diku.dk>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] scsi_eh_prep_cmnd should save scmd->underflow

This patch (as1116) fixes a bug in scsi_eh_prep_cmnd() and
scsi_eh_restore_cmnd().  These routines are supposed to save any
values they change and restore them later, but someone forgot to
save & restore scmd->underflow.

This fixes part of the problem reported in Bugzilla #9638.

[jejb: fix up rejections around DIF/DIX]
Signed-off-by: Alan Stern <stern@rowland.harvard.edu>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] sd: Support for SCSI disk (SBC) Data Integrity Field

Support for controllers and disks that implement DIF protection
information:

 - During command preparation the RDPROTECT/WRPROTECT must be set
   correctly if the target has DIF enabled.

 - READ(6) and WRITE(6) are not supported when DIF is on.

 - The controller must be told how to handle the I/O via the
   protection operation field in scsi_cmnd.

 - Refactor the I/O completion code that extracts failed LBA from the
   returned sense data and handle DIF failures correctly.

 - sd_dif.c implements the functions required to prepare and complete
   requests with protection information attached.

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

[SCSI] sd: Identify DIF protection type and application tag ownership

If a disk is formatted with protection information (Inquiry bit
PROTECT=1) it is required to support Read Capacity(16).  Force use of
the 16-bit command in this case and extract the P_TYPE field which
indicates whether the disk is formatted using DIF Type 1, 2 or 3.

The ATO (App Tag Own) bit in the Control Mode Page indicates whether
the storage device or the initiator own the contents of the
DIF application tag.

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

[SCSI] Do not retry a request whose data integrity check failed

If initiator or target reject the I/O due to DIF errors there is no
point in retrying.

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

[SCSI] Support devices with protection information

Implement support for DMA of protection information for devices that
are data integrity capable.

 - Add support for mapping an extra scatter-gather list containing
   the protection information.

 - Allocate protection scsi_data_buffer if host is DIX (integrity DMA)
   capable.

 - Accessor function for checking whether a device has protection
   enabled.

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

[SCSI] Command protection operation

Controllers that support DMA of protection information must be told
explicitly how to handle the I/O.  The controller has no knowledge of
the protection capabilities of the target device so this information
must be passed in the scsi_cmnd.

 - The protection operation tells the HBA whether to generate, strip or
   verify protection information.

 - The protection type tells the HBA which layout the target is
   formatted with.  This is necessary because the controller must be
   able to correctly interpret the included protection information in
   order to verify it.

 - When a scsi_cmnd is reused for error handling the protection
   operation must be cleared and saved while error handling is in
   progress.

 - prot_op and prot_type are placed in an existing hole in scsi_cmnd
   and don't cause the structure to grow.

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

[SCSI] Host protection capabilities

Controllers that support protection information must indicate this to
the SCSI midlayer so that the ULD can prepare scsi_cmnds accordingly.

This patch implements a host mask and various types of protection:

 - DIF Type 1-3 (between HBA and disk)
 - DIX Type 0-3 (between OS and HBA)

The patch also allows the HBA to set the guard type to something
different than the T10-mandated CRC.

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

[SCSI] scsi_dh: create lookup cache

Create a cache of devices that are seen in a system. This will avoid
the unnecessary traversal of the device list in the scsi_dh when there
are multiple luns of a same type.

Signed-off-by: Chandra Seetharaman <sekharan@us.ibm.com>
Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] scsi_dh: attach to hardware handler from dm-mpath

multipath keeps a separate device table which may be
more current than the built-in one.
So we should make sure to always call ->attach whenever
a multipath map with hardware handler is instantiated.
And we should call ->detach on removal, too.

[sekharan: update as per comments from agk]
Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Chandra Seetharaman <sekharan@us.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] scsi_dh: add generic SPC-3 alua handler

Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Chandra Seetharaman <sekharan@us.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] scsi_dh: Update RDAC device handler

This patch updates the RDAC device handler to
refuse to attach to devices not supporting the
RDAC vpd pages.

Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Chandra Seetharaman <sekharan@us.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] scsi_dh: Update hp_sw hardware handler

This patch updates the hp_sw device handler to properly
check the return codes etc.
And adds the 'correct' machine definitions.

Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Chandra Seetharaman <sekharan@us.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] scsi_dh: Update EMC handler

This patch converts the EMC device handler to use a proper
state machine. We now also parse the extended INQUIRY
information to determine if long trespass commands are
supported. And we're now using the long trespass command
correctly. And finally there's now an check at init time
to refuse to attach to devices not supporting EMC-specific
VPD pages.

Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Chandra Seetharaman <sekharan@us.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] scsi_dh: Add 'dh_state' sysfs attribute

Implement a 'dh_state' sdev attribute for dynamic device handler
manipulation. A read on the attribute will return the name of
the currently attached device handler or 'detached' if no handler
is attached.
The attribute allows the following strings to be written:
- The name of the device handler to be attached if the state is
  'detached'.
- 'activate' to trigger path activation if a device handler
  is attached.
- 'detach' to detach the currently attached device handler.

Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Chandra Seetharaman <sekharan@us.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] scsi_dh: Implement common device table handling

Instead of having each and every driver implement its own
device table scanning code we should rather implement a common
routine and scan the device tables there.
This allows us also to implement a general notifier chain
callback for all device handler instead for one per handler.

[sekharan: Fix rejections caused by conflicting bug fix]
Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Chandra Seetharaman <sekharan@us.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] Make host_no an unsigned int

Daniel Debonzi reports that he has managed to wrap host_no.  Increasing
the number of host numbers available to 32-bit from 16-bit allows the
problem to be evaded for another hundred years.

Signed-off-by: Matthew Wilcox <willy@linux.intel.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] sd: update index allocation and use ida instead of idr

Update index allocation as follows.

* sd_index_idr is used only for ID allocation and mapping
  functionality is not used.  Use more memory efficient ida instead.

* idr and ida have their own locks inside them and don't need them for
  operation.  Drop it.

* index wasn't freed if probing failed after index allocation.  fix
  it.

* ida allocation should be repeated if it fails with -EAGAIN.

Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] fix shared tag map tag allocation

When drivers use a shared tag map we can end up with more requests
than tags, because the tag map is shost->can_queue tags and there
can be sdevs * sdev->queue_depth requests. In scsi_request_fn
if tag allocation fails we just drop down to just dequeueing the
tag without a tag. The problem is that drivers using the shared tag
map rely on a valid tag always being set, because it will use the
tag number to lookup commands later.

This patch has us check if we got a valid tag when the host lock
is held right before we check if the host queue is ready. We do the
check here because to allocate the tag we need the q lock, but
if the tag is bad we want to add the device/q onto the starved list
which requires the host lock.

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

[SCSI] stex: fix queue depth setting

We want to set the queue depth to something reasonable - not
the can_queue.

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

[SCSI] qla4xxx: fix queue depth setting

We want to set the queue depth to something reasonable - not
the can_queue.

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

[SCSI] fix shared tag map setup

Currently qla4xxx and stex pass in their can_queue values into
scsi_activate_tcq because they wanted the tag map that large.
The problem with this is that it ends up also setting the queue
depth to that large value. All we want to do this in this case
is set the device queue depth and the other device settings.
We do not need to touch the tag map sizing because the drivers
had setup that map according to their can_queue limits when the
shared map was created.

The scsi mid layer in request_fn will then handle the case where we
have more requests than available tags when it checks the host
queue ready function.

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

[SCSI] st: Remove bogus memset

Mike Christie noticed a bogus memset. It can be removed as dead code
since the number of bytes in the driver buffer in fixed block mode is
always a multiple of the tape block size.

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

[SCSI] st: Move buffer pointer back when data could not be written.

Move buffer pointer back when data could not be written. Bug found by
Mike Christie.

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

[SCSI] qla2xxx: Update version number to 8.02.01-k5.

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

[SCSI] qla2xxx: Don't hardcode fw_transfer_size for ISP2[45]xx parts.

Use the full buffer size available, as there's no reason to limit
the firwmare-image load-segment size for these parts.

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

[SCSI] qla2xxx: Verify the RISC is not in ROM code if firmware-load is disabled.

Add an additional check to verify that the current executing
firmware is in fact non-ROM code.  The non-ROM Get-ID mailbox
command is used for verification.

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

[SCSI] qla2xxx: Correct overflow during dump-processing on large-memory ISP23xx parts.

Total ram words can exceed a 16bit value on large-memory boards.
Safely extend to a 32bit width.

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

[SCSI] qla2xxx: Skip FDMI registration on ISP21xx/22xx parts.

Firmware does not have the facilities to issue management server
IOCBs.

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

[SCSI] qla2xxx: Correct rport/fcport visibility-state handling during loop-resync.

There were several issues here, one, during RSCN handling if a
follow-on RSCN occurred (within interrupt context) the DPC thread
could inadvertantly leave the fcport in a stale lost state.
Secondly, scheduled rport removal is handled exclusively by the
'parent' DPC thread, so wake up the proper thread.  Finally,
process vport loop-resync's only when the vport has in an
"active" state (ID acquired).

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>

[SCSI] qla2xxx: Correct vport management of MBA_PORT_UPDATE.

By allowing the qla2x00_alert_all_vps() to manage per-vport
recognition of the MBA.

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>

[SCSI] qla2xxx: Correct fcport state-management during loss.

All fcport->state management should be done within
qla2x00_mark_device_lost(), the assignment of state within
qla2x00_mark_vp_devices_dead() caused associated rports to not be
removed.

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>

[SCSI] qla2xxx: Always aquire the parent's hardware_lock.

While issuing a marker, manipulating the request/response queues
and modifying the outstanding command array.

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>

[SCSI] qla2xxx: Swap enablement order of EFT and FCE.

The firmware group has suggested that FCE (Fibre Channel Event)
tracing be enabled prior to EFT (Extended Firmware Tracing) to
maximize the capturing of data on the wire.  This change has no
real semantic effect on driver operation, as it's mostly a
shuffling of code.

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

[SCSI] qla2xxx: Retrieve board serial-number and description from VPD.

Recent ISPs have this information written at manufacturing time,
so use the information.  This also reduces future churn of the
qla_devtbl.h file contents, as the driver can now depend on the
information to be present in VPD.

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

[SCSI] qla2xxx: Allow the user the option of disabling iIDMA.

iIDMA support requires the driver issue several additional
fabric-managegment (FM) commands per port discovered during SNS
scanning -- GFPN (Get Fabric Port Name) and GPSC (Get Port Speed
Capabilities).  It has been found during testing that some
switches do not respond as *well* as expected to these commands
(silence -- no ACC nor BS_RJT).  So, to handle such conditions,
allow the user the ability to indirectly disable the FM commands
by disabling iIDMA with the ql2xiidmaenable module-parameter.

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

[SCSI] qla2xxx: Cleanup NPIV related functions

Removed repeated or unnecessary operations during vport
creation/deletion.

Signed-off-by: Shyam Sundar <shyam.sundar@qlogic.com>
Signed-off-by: Seokmann Ju <seokmann.ju@qlogic.com>
Signed-off-by: Ravi Anand <ravi.anand@qlogic.com>
Signed-off-by: Andrew Vasquez <andrew.vasquez@qlogic.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

[SCSI] qla2xxx: Add LIP count to FC-transport statistics.

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

[SCSI] qla2xxx: Track total number of ISP aborts.

This parameter counts the total number of ISP aborts during
driver execution.  The value is exported through a DEVICE_ATTR()
off the scsi_host.

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

[SCSI] qla2xxx: Set an rport's dev_loss_tmo value in a consistent manner.

As there's no point in adding a fixed-fudge value (originally 5
seconds), honor the user settings only.  We also remove the
driver's dead-callback get_rport_dev_loss_tmo function
(qla2x00_get_rport_loss_tmo()).

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

[SCSI] qla2xxx: Add dev_loss_tmo_callbk/terminate_rport_io callback support.

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>

[SCSI] qla2xxx: Correct locking during NVRAM manipulation.

Commit 2c96d8d0c17978bbf5eb82314d488f46d4a51280 pushed the
acquisition of hardware_lock to too fine a level, which in turn
will cause problems with cond_resched()s added with
40a2e34a94c336b716f631b2952d233e1ba76e3c.

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

x86_64: fix ia32 AMD syscall audit fast-path

The new code in commit 5cbf1565f29eb57a86a305b08836613508e294d7
has a bug in the version supporting the AMD 'syscall' instruction.
It clobbers the user's %ecx register value (with the %ebp value).

This change fixes it.

Signed-off-by: Roland McGrath <roland@redhat.com>

Merge git://git./linux/kernel/git/davem/net-2.6

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6:
  ipsec: ipcomp - Decompress into frags if necessary
  ipsec: ipcomp - Merge IPComp implementations
  pkt_sched: Fix locking in shutdown_scheduler_queue()

MFD_TC6393XB is ARM-only

Compile error on other architectures:

    CC      drivers/mfd/tc6393xb.o
  /home/bunk/linux/kernel-2.6/git/linux-2.6/drivers/mfd/tc6393xb.c: In function ‘tc6393xb_attach_irq’:
  /home/bunk/linux/kernel-2.6/git/linux-2.6/drivers/mfd/tc6393xb.c:324: error: implicit declaration of function ‘set_irq_flags’
  ...

Reported-by: Adrian Bunk <bunk@kernel.org>
Signed-off-by: Adrian Bunk <bunk@kernel.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Merge branch 'merge' of git://git./linux/kernel/git/benh/powerpc

* 'merge' of git://git.kernel.org/pub/scm/linux/kernel/git/benh/powerpc:
  powerpc: Fix boot problem due to AT_BASE_PLATFORM change

Merge git://git./linux/kernel/git/davem/sparc-2.6

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/sparc-2.6:
  sparc: Wire up new system calls.

Merge branch 'release' of git://git./linux/kernel/git/aegl/linux-2.6

* 'release' of git://git.kernel.org/pub/scm/linux/kernel/git/aegl/linux-2.6:
  [IA64] Wire up new system calls

sys_paccept definition missing __user annotation

Introduced by commit aaca0bdca573f3f51ea03139f9c7289541e7bca3 ("flag
parameters: paccept"):

  net/socket.c:1515:17: error: symbol 'sys_paccept' redeclared with different type (originally declared at include/linux/syscalls.h:413) - incompatible argument 4 (different address spaces)

Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

powerpc: Fix boot problem due to AT_BASE_PLATFORM change

Commit 9115d13453dee22473a1e8cacc90a8d64a9c4bc9 ("powerpc: Enable
AT_BASE_PLATFORM aux vector") broke boot on 32-bit powerpc systems; we
have to use PTRRELOC to initialize powerpc_base_platform this early in
boot.

Bug reported by Jon Smirl.

Signed-off-by: Nathan Lynch <ntl@pobox.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>

sparc: Wire up new system calls.

This wires up the recently added Wire up signalfd4, eventfd2,
epoll_create1, dup3, pipe2, and inotify_init1 system calls.

Signed-off-by: David S. Miller <davem@davemloft.net>

Merge git://git.infradead.org/embedded-2.6

* git://git.infradead.org/embedded-2.6:
  Make console charset translation optional

Merge git://git.infradead.org/~dwmw2/random-2.6

* git://git.infradead.org/~dwmw2/random-2.6:
  remove dummy asm/kvm.h files
  firmware: create firmware binaries during 'make modules'.

bootmem: Move node allocation macros back to !HAVE_ARCH_BOOTMEM_NODE

These got unintentionally moved, put them back as x86 provides its own
versions.

Signed-off-by: Johannes Weiner <hannes@saeurebad.de>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

remove dummy asm/kvm.h files

This patch removes the dummy asm/kvm.h files on architectures not (yet)
supporting KVM and uses the same conditional headers installation as
already used for a.out.h .

Also removed are superfluous install rules in the s390 and x86 Kbuild
files (they are already in Kbuild.asm).

Signed-off-by: Adrian Bunk <bunk@kernel.org>
Acked-by: Sam Ravnborg <sam@ravnborg.org>
Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

signalfd: fix undefined reference to `compat_sys_signalfd4' when !CONFIG_SIGNALFD

fix:

arch/x86/ia32/built-in.o: In function `ia32_sys_call_table':
(.rodata+0xa38): undefined reference to `compat_sys_signalfd4'

on !CONFIG_SIGNALFD.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/hugetlb.c: fix build failure with !CONFIG_SYSCTL

on !CONFIG_SYSCTL on x86 with latest -git i get:

     mm/hugetlb.c: In function 'decrement_hugepage_resv_vma':
     mm/hugetlb.c:83: error: 'reserve' undeclared (first use in this function)
     mm/hugetlb.c:83: error: (Each undeclared identifier is reported only once
     mm/hugetlb.c:83: error: for each function it appears in.)

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

firmware: create firmware binaries during 'make modules'.

This means that we no longer need write access to the source tree while
doing 'make modules_install'.

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>

Merge branch 'merge' of git://git./linux/kernel/git/benh/powerpc

* 'merge' of git://git.kernel.org/pub/scm/linux/kernel/git/benh/powerpc: (34 commits)
  powerpc: Wireup new syscalls
  Move update_mmu_cache() declaration from tlbflush.h to pgtable.h
  powerpc/pseries: Remove kmalloc call in handling writes to lparcfg
  powerpc/pseries: Update arch vector to indicate support for CMO
  ibmvfc: Add support for collaborative memory overcommit
  ibmvscsi: driver enablement for CMO
  ibmveth: enable driver for CMO
  ibmveth: Automatically enable larger rx buffer pools for larger mtu
  powerpc/pseries: Verify CMO memory entitlement updates with virtual I/O
  powerpc/pseries: vio bus support for CMO
  powerpc/pseries: iommu enablement for CMO
  powerpc/pseries: Add CMO paging statistics
  powerpc/pseries: Add collaborative memory manager
  powerpc/pseries: Utilities to set firmware page state
  powerpc/pseries: Enable CMO feature during platform setup
  powerpc/pseries: Split retrieval of processor entitlement data into a helper routine
  powerpc/pseries: Add memory entitlement capabilities to /proc/ppc64/lparcfg
  powerpc/pseries: Split processor entitlement retrieval and gathering to helper routines
  powerpc/pseries: Remove extraneous error reporting for hcall failures in lparcfg
  powerpc: Fix compile error with binutils 2.15
  ...

Fixed up conflict in arch/powerpc/platforms/52xx/Kconfig manually.

Merge branch 'linux-next' of git://git.infradead.org/~dedekind/ubi-2.6

* 'linux-next' of git://git.infradead.org/~dedekind/ubi-2.6: (22 commits)
  UBI: always start the background thread
  UBI: fix gcc warning
  UBI: remove pre-sqnum images support
  UBI: fix kernel-doc errors and warnings
  UBI: fix checkpatch.pl errors and warnings
  UBI: bugfix - do not torture PEB needlessly
  UBI: rework scrubbing messages
  UBI: implement multiple volumes rename
  UBI: fix and re-work debugging stuff
  UBI: amend commentaries
  UBI: fix error message
  UBI: improve mkvol request validation
  UBI: add ubi_sync() interface
  UBI: fix 64-bit calculations
  UBI: fix LEB locking
  UBI: fix memory leak on error path
  UBI: do not forget to free internal volumes
  UBI: fix memory leak
  UBI: avoid unnecessary division operations
  UBI: fix buffer padding
  ...

Fix ahci driver 'flags' type

The new type checking of the flags arguments to irqsave and friends
(commit 3f307891ce0e7b0438c432af1aacd656a092ff45) pointed out this thing
with a big nice warning.

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

edac: mpc85xx fix pci ofdev 2nd pass

Convert PCI err device from platform to open firmware of_dev to comply
with powerpc schemes.

[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Dave Jiang <djiang@mvista.com>
Signed-off-by: Doug Thompson <dougthompson@xmission.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

edac: mv64x60 add pci fixup

Fixup of missing bit 0 on 64360 PCIx_ERR_MASK and errata FEr-#11 and
FEr-#16 for the 64460.  Bit 0 must remain 0.

Signed-off-by: Dave Jiang <djiang@mvista.com>
Signed-off-by: Doug Thompson <dougthompson.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

edac: mv64x60 fix get_property

Update get_property() call to use of_get_property() in order to fix compile

Signed-off-by: Dave Jiang <djiang@mvista.com>
Signed-off-by: Doug Thompson <dougthompson.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

edac: e752x fix too loud on nonmemory errors

This module harvests more than just memory errors, it also harvests
various bus and dma errors that the Chipset detects.  Previously, it would
report all such errors, which would cause output to be TOO loud.

This patches therefore adds a parameter which is used to turn off
NON-MEMORY error reports by default.  Or the reporting can be enabled via
the parameter

Also did code style cleanup: less than 80 characters per line rule

Signed-off-by: Doug Thompson <dougthompson@xmission.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

edac: core fix added newline to sysfs dimm labels

The channel DIMM label does not seem to be used much in the edac code.
However, where it is used (in the core code), it is assumed to not have a
newline embedded.  This leaves the sysfs file newline free which looks
funny when cat'ing it.  Here we just add the trailing newline to the sysfs
chX_dimm_label output...

[Doug Thompson note: the DIMM label is one of the primary uses of EDAC.
User space daemon scripts, edac-utils@sourceforge, populate the DIMM label
fields, via /sys/devices/system/edac attributes, with the silk screen
labels of the motherboard in use.  dmidecode access BIOS tables, but BIOS
tables are well known to be incorrect and useless in these respects.
edac-utils will strip off any newlines before its use of the output, when
displaying DIMM slot silk screen labels.

Signed-off-by: Arthur Jones <ajones@riverbed.com>
Signed-off-by: Doug Thompson <dougthompson@xmission.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

edac: core fix static to dynamic kset

Static kobjects and ksets are not supported in Linux kernel.  Convert the
mc_kset from static to dynamic.  This patch depends on my previous patch
to remove the module parameter attributes from mc...

Signed-off-by: Arthur Jones <ajones@riverbed.com>
Signed-off-by: Doug Thompson <dougthompson@xmission.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

edac: core fix redundant sysfs controls to parameters

/sys/devices/system/edac/mc has a few files which are duplicated in
/sys/module/edac_core/parameters.  Now that all the functionality is
duplicated between these two locations, we remove the former kobject
attributes and update the documentation.

Signed-off-by: Arthur Jones <ajones@riverbed.com>
Signed-off-by: Doug Thompson <dougthompson@xmission.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

edac: core fix workq timer

When updating the edac_mc_poll_msec module parameter from the sysfs
/sys/module/edac_core/parameters/edac_mc_poll_msec file, we don't update
the workq timers.  So that, if we move from a big poll time to a small
one, the small one won't take effect until the big one has timed out.

Here we provide a new module parameter set method to call out to the
update routine.  This brings the /sys/module/edac_core/parameters
functionality up to that provided by the /sys/drivers/system/edac/mc sysfs
module parameter files so that we can remove them or at least link to the
/sys/module files...

Signed-off-by: Arthur Jones <ajones@riverbed.com>
Signed-off-by: Doug Thompson <dougthompson@xmission.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

edac: core fix to use dynamic kobject

Static kobjects are not supported in linux kernel.  Convert the
edac_pci_top_main_kobj from static to dynamic.  This avoids the double
free of the edac_pci_top_main_kobj.name that we see on module reload of
the e752x edac driver (and probably others as well).

In addition Greg KH <greg@kroah.com> has pointed out that this code may be
cleaned up significantly.  I will look at that as a follow-on patch, for
now, I just want the minimum fix to get this double-free oops bug
squashed...

Many thanks to Greg KH for his patience in showing me what the
Documentation/kobject.txt already said (oops)...

Signed-off-by: Arthur Jones <ajones@riverbed.com>
Signed-off-by: Doug Thompson <dougthompson@xmission.com>
Acked-by: Greg Kroah-Hartman <gregkh@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

edac: i5100: cleanup

Some code cleanliness issues found by Andrew Morton (thanks!) which should
not affect functionality, but which should help make the code more
maintainable.

In particular, we now:

* convert all #define's w/ a parameter to static inlines
* use 1UL rather than 1ULL when calculating an unsigned long
* use pci_disable_device

The resulting code is tested and seems to work fine...

Signed-off-by: Arthur Jones <ajones@riverbed.com>
Cc: Doug Thompson <dougthompson@xmission.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

edac: i5100 fix unmask ecc bits

Explicitly unmask ECC errors we are interested in reporting.

Signed-off-by: Arthur Jones <ajones@riverbed.com>
Signed-off-by: Doug Thompson <dougthompson@xmission.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

edac: i5100 fix enable ecc hardware

It is possible that the BIOS did not enable ECC at boot time.  We check
for that case and fail to load if it is true.

Signed-off-by: Arthur Jones <ajones@riverbed.com>
Signed-off-by: Doug Thompson <dougthompson@xmission.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

edac: i5100 fix missing bits

The error mask we use to trigger ECC notifications is missing many bits of
interest.  We add these bits here so that all possible ECC errors can be
reported.

Signed-off-by: Arthur Jones <ajones@riverbed.com>
Signed-off-by: Doug Thompson <dougthompson@xmission.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

edac: i5100 new intel chipset driver

Preliminary support for the Intel 5100 MCH.  CE and UE errors are reported
along with the current DIMM label information and other memory parameters.

Reasons why this is preliminary:

1) This chip has 2 independent memory controllers which, for best
   perforance, use interleaved accesses to the DDR2 memory.  This
   architecture does not map very well to the current edac data structures
   which depend on symmetric channel access to the interleaved data.
   Without core changes, the best I could do for now is to map both memory
   controllers to different csrows (first all ranks of controller 0, then
   all ranks of controller 1).  Someone much more familiar with the edac
   core than I will probably need to come up with a more general data
   structure to handle the interleaving and de-interleaving of the two
   memory controllers.

2) I have not yet tackled the de-interleaving of the rank/controller
   address space into the physical address space of the CPU.  There is
   nothing fundamentally missing, it is just ending up to be a lot of
   code, and I'd rather keep it separate for now, esp since it doesn't
   work yet...

3) The code depends on a particular i5100 chip select to DIMM mainboard
   chip select mapping.  This mapping seems obvious to me in order to
   support dual and single ranked memory, but it is not unique and DIMM
   labels could be wrong on other mainboards.  There is no way to query
   this mapping that I know of.

4) The code requires that the i5100 is in 32GB mode.  Only 4 ranks per
   controller, 2 ranks per DIMM are supported.  I do not have hardware
   (nor do I expect to have hardware anytime soon) for the 48GB (6 ranks
   per controller) mode.

5) The serial presence detect code should be broken out into a "real"
   i2c driver so that decode-dimms.pl can work.

Signed-off-by: Arthur Jones <ajones@riverbed.com>
Signed-off-by: Doug Thompson <dougthompson@xmission.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fuse: lockd support

If fuse filesystem doesn't define it's own lock operations, then allow the
lock manager to work with fuse.

Adding lockd support for remote locking is also possible, but more rarely
used, so leave it till later.

Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Cc: "J. Bruce Fields" <bfields@fieldses.org>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: Matthew Wilcox <matthew@wil.cx>
Cc: David Teigland <teigland@redhat.com>
Cc: Christoph Hellwig <hch@lst.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fuse: nfs export special lookups

Implement the get_parent export operation by sending a LOOKUP request with
".." as the name.

Implement looking up an inode by node ID after it has been evicted from
the cache.  This is done by seding a LOOKUP request with "." as the name
(for all file types, not just directories).

The filesystem can set the FUSE_EXPORT_SUPPORT flag in the INIT reply, to
indicate that it supports these special lookups.

Thanks to John Muir for the original implementation of this feature.

Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Cc: "J. Bruce Fields" <bfields@fieldses.org>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: Matthew Wilcox <matthew@wil.cx>
Cc: David Teigland <teigland@redhat.com>
Cc: Christoph Hellwig <hch@lst.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fuse: add fuse_lookup_name() helper

Add a new helper function which sends a LOOKUP request with the supplied
name.  This will be used by the next patch to send special LOOKUP requests
with "." and ".." as the name.

Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Cc: Christoph Hellwig <hch@lst.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fuse: add export operations

Implement export_operations, to allow fuse filesystems to be exported to
NFS.  This feature has been in the out-of-tree fuse module, and is widely
used and tested.

It has not been originally merged into mainline, because doing the NFS
export in userspace was thought to be a cleaner and more efficient way of
doing it, than through the kernel.

While that is true, it would also have involved a lot of duplicated effort
at reimplementing NFS exporting (all the different versions of the
protocol).  This effort was unfortunately not undertaken by anyone, so we
are left with doing it the easy but less efficient way.

If this feature goes in, the out-of-tree fuse module can go away,
which would have several advantages:

  - not having to maintain two versions
  - less confusion for users
  - no bugs due to kernel API changes

Comment from hch:
 - Use the same fh_type values as XFS, since we use the same fh encoding.

Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Cc: Christoph Hellwig <hch@lst.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fuse: prepare lookup for nfs export

Use d_splice_alias() instead of d_add() in fuse lookup code, to allow NFS
exporting.

Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Cc: Christoph Hellwig <hch@lst.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

locks: allow ->lock() to return FILE_LOCK_DEFERRED

Allow filesystem's ->lock() method to call posix_lock_file() instead of
posix_lock_file_wait(), and return FILE_LOCK_DEFERRED.  This makes it
possible to implement a such a ->lock() function, that works with the lock
manager, which needs the call to be asynchronous.

Now the vfs_lock_file() helper can be used, so this is a cleanup as well.

Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Cc: "J. Bruce Fields" <bfields@fieldses.org>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: Matthew Wilcox <matthew@wil.cx>
Cc: David Teigland <teigland@redhat.com>
Cc: Christoph Hellwig <hch@lst.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

locks: cleanup code duplication

Extract common code into a function.

Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Cc: "J. Bruce Fields" <bfields@fieldses.org>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: Matthew Wilcox <matthew@wil.cx>
Cc: David Teigland <teigland@redhat.com>
Cc: Christoph Hellwig <hch@lst.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

locks: add special return value for asynchronous locks

Use a special error value FILE_LOCK_DEFERRED to mean that a locking
operation returned asynchronously.  This is returned by

  posix_lock_file() for sleeping locks to mean that the lock has been
  queued on the block list, and will be woken up when it might become
  available and needs to be retried (either fl_lmops->fl_notify() is
  called or fl_wait is woken up).

  f_op->lock() to mean either the above, or that the filesystem will
  call back with fl_lmops->fl_grant() when the result of the locking
  operation is known.  The filesystem can do this for sleeping as well
  as non-sleeping locks.

This is to make sure, that return values of -EAGAIN and -EINPROGRESS by
filesystems are not mistaken to mean an asynchronous locking.

This also makes error handling in fs/locks.c and lockd/svclock.c slightly
cleaner.

Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: "J. Bruce Fields" <bfields@fieldses.org>
Cc: Matthew Wilcox <matthew@wil.cx>
Cc: David Teigland <teigland@redhat.com>
Cc: Christoph Hellwig <hch@lst.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

lockd: dont return EAGAIN for a permanent error

Fix nlm_fopen() to return NLM_FAILED (or NLM_LCK_DENIED_NOLOCKS) instead
of NLM_LCK_DENIED.  The latter means the lock request failed because of a
conflicting lock (i.e.  a temporary error), which is wrong in this case.

Also fix the client to return ENOLCK instead of EAGAIN if a blocking lock
request returns with NLM_LOCK_DENIED.

Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: "J. Bruce Fields" <bfields@fieldses.org>
Cc: Matthew Wilcox <matthew@wil.cx>
Cc: David Teigland <teigland@redhat.com>
Cc: Christoph Hellwig <hch@lst.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

taskstats: remove initialization of static per-cpu variable

Cc: Shailabh Nagar <nagar@watson.ibm.com>
Signed-off-by: Vegard Nossum <vegard.nossum@gmail.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

per-task-delay-accounting: update document and getdelays.c for memory reclaim

Update document and make getdelays.c show delay accounting for memory reclaim.

For making a distinction between "swapping in pages" and "memory reclaim"
in getdelays.c, MEM is changed to SWAP.

Signed-off-by: Keika Kobayashi <kobayashi.kk@ncos.nec.co.jp>
Acked-by: Balbir Singh <balbir@linux.vnet.ibm.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

per-task-delay-accounting: update taskstats for memory reclaim delay

Add members for memory reclaim delay to taskstats, and accumulate them in
__delayacct_add_tsk() .

Signed-off-by: Keika Kobayashi <kobayashi.kk@ncos.nec.co.jp>
Cc: Hiroshi Shimamoto <h-shimamoto@ct.jp.nec.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

per-task-delay-accounting: add memory reclaim delay

Sometimes, application responses become bad under heavy memory load.
Applications take a bit time to reclaim memory.  The statistics, how long
memory reclaim takes, will be useful to measure memory usage.

This patch adds accounting memory reclaim to per-task-delay-accounting for
accounting the time of do_try_to_free_pages().

<i.e>

- When System is under low memory load,
  memory reclaim may not occur.

$ free
             total       used       free     shared    buffers     cached
Mem:       8197800    1577300    6620500          0       4808    1516724
-/+ buffers/cache:      55768    8142032
Swap:     16386292          0   16386292

$ vmstat 1
procs -----------memory---------- ---swap-- -----io---- -system-- ----cpu----
 r  b   swpd   free   buff  cache   si   so    bi    bo   in   cs us sy id wa
 0  0      0 5069748  10612 3014060    0    0     0     0    3   26  0  0 100  0
 0  0      0 5069748  10612 3014060    0    0     0     0    4   22  0  0 100  0
 0  0      0 5069748  10612 3014060    0    0     0     0    3   18  0  0 100  0

Measure the time of tar command.

$ ls -s test.dat
1501472 test.dat

$ time tar cvf test.tar test.dat
real    0m13.388s
user    0m0.116s
sys     0m5.304s

$ ./delayget -d -p <pid>
CPU             count     real total  virtual total    delay total
                  428     5528345500     5477116080       62749891
IO              count    delay total
                  338     8078977189
SWAP            count    delay total
                    0              0
RECLAIM         count    delay total
                    0              0

- When system is under heavy memory load
  memory reclaim may occur.

$ vmstat 1
procs -----------memory---------- ---swap-- -----io---- -system-- ----cpu----
 r  b   swpd   free   buff  cache   si   so    bi    bo   in   cs us sy id wa
 0  0 7159032  49724   1812   3012    0    0     0     0    3   24  0  0 100  0
 0  0 7159032  49724   1812   3012    0    0     0     0    4   24  0  0 100  0
 0  0 7159032  49848   1812   3012    0    0     0     0    3   22  0  0 100  0

In this case, one process uses more 8G memory
by execution of malloc() and memset().

$ time tar cvf test.tar test.dat
real    1m38.563s        <-  increased by 85 sec
user    0m0.140s
sys     0m7.060s

$ ./delayget -d -p <pid>
CPU             count     real total  virtual total    delay total
                 9021     7140446250     7315277975      923201824
IO              count    delay total
                 8965    90466349669
SWAP            count    delay total
                    3       21036367
RECLAIM         count    delay total
                  740    61011951153

In the later case, the value of RECLAIM is increasing.
So, taskstats can show how much memory reclaim influences TAT.

Signed-off-by: Keika Kobayashi <kobayashi.kk@ncos.nec.co.jp>
Acked-by: Balbir Singh <balbir@linux.vnet.ibm.com>
Acked-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujistu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

tsacct: fix bacct_add_tsk()'s use of do_div()

Fix bacct_add_tsk()'s use of do_div() on an s64 by making ac_etime a u64
instead and dividing that.

Possibly this should be guarded lest the interval calculation turn up
negative, but the possible negativity of the result of the division is
cast away, and it shouldn't end up negative anyway.

This was introduced by patch f3cef7a99469afc159fec3a61b42dc7ca5b6824f.

Signed-off-by: David Howells <dhowells@redhat.com>
Cc: Jay Lan <jlan@engr.sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

task IO accounting: provide distinct tgid/tid I/O statistics

Report per-thread I/O statistics in /proc/pid/task/tid/io and aggregate
parent I/O statistics in /proc/pid/io.  This approach follows the same
model used to account per-process and per-thread CPU times.

As a practial application, this allows for example to quickly find the top
I/O consumer when a process spawns many child threads that perform the
actual I/O work, because the aggregated I/O statistics can always be found
in /proc/pid/io.

[ Oleg Nesterov points out that we should check that the task is still
  alive before we iterate over the threads, but also says that we can do
  that fixup on top of this later.  - Linus ]

Acked-by: Balbir Singh <balbir@linux.vnet.ibm.com>
Signed-off-by: Andrea Righi <righi.andrea@gmail.com>
Cc: Matt Heaton <matt@hostmonster.com>
Cc: Shailabh Nagar <nagar@watson.ibm.com>
Acked-by-with-comments: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

bsdacct: fix and add comments around acct_process()

Fix the one describing what this function is and add one more - about
locking absence around pid namespaces loop.

Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Cc: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

bsdacct: account dying tasks in all relevant namespaces

This just makes the acct_proces walk the pid namespaces from current up to
the top and account a task in each with the accounting turned on.

ns->parent access if safe lockless, since current it still alive and holds
its namespace, which in turn holds its parent.

Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

bsdacct: turn acct off for all pidns-s on umount time

All the bsd_acct_strcts with opened accounting are linked into a global
list.  So, the acct_auto_close(_mnt) walks one and drops the accounting
for each.

Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

bsdacct: switch from global bsd_acct_struct instance to per-pidns one

Allocate the structure on the first call to sys_acct().  After this each
namespace, that ordered the accounting, will live with this structure till
its own death.

Two notes
- routines, that close the accounting on fs umount time use
  the init_pid_ns's acct by now;
- accounting routine accounts to dying task's namespace
  (also by now).

Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

bsdacct: make internal code work with passed bsd_acct_struct, not global

This adds the appropriate pointer to all the internal (i.e.  static)
functions that work with global acct instance.  API calls pass a global
instance to them (while we still have such).

Mostly this is a s/acct_globals./acct->/ over the file.

Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

bsdacct: turn the acct_lock from on-the-struct to global

Don't use per-bsd-acct-struct lock, but work with a global one.

This lock is taken for short periods, so it doesn't seem it'll become a
bottleneck, but it will allow us to easily avoid many locking difficulties
in the future.

So this is a mostly s/acct_globals.lock/acct_lock/ over the file.

Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

bsdacct: make check timer accept a bsd_acct_struct argument

We're going to have many bsd_acct_struct instances, not just one, so the
timer (currently working with a global one) has to know which one to work
with.

Use a handy setup_timer macro for it (thanks to Oleg for one).

Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

bsdacct: "truthify" a comment near acct_process

The acct_process does not accept any arguments actually.

Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

pidns: add the struct bsd_acct_struct pointer on pid_namespace struct

All the bsdacct-related info will be stored in the area, pointer by this
one.

It will be NULL automatically for all new namespaces.

Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

pidns: use kzalloc when allocating new pid_namespace struct

It makes many fields initialization implicit helping in auto-setting
#ifdef-ed fields (bsd-acct related pointer will be such).

Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>