fuse: update MAINTAINERS entry

Signed-off-by: Miklos Szeredi <[email protected]>

watchdog: omap_wdt: early_enable module parameter

Add a early_enable module parameter to the omap_wdt that starts the
watchdog on module insertion. The default value is 0 which does not
start the watchdog - which also does not change the behavior if the
parameter is not given.

Signed-off-by: Lars Poeschel <[email protected]>
Reviewed-by: Guenter Roeck <[email protected]>
Signed-off-by: Wim Van Sebroeck <[email protected]>

fuse: separate pqueue for clones

Make each fuse device clone refer to a separate processing queue.  The only
constraint on userspace code is that the request answer must be written to
the same device clone as it was read off.

Signed-off-by: Miklos Szeredi <[email protected]>

fuse: introduce per-instance fuse_dev structure

Allow fuse device clones to refer to be distinguished.  This patch just
adds the infrastructure by associating a separate "struct fuse_dev" with
each clone.

Signed-off-by: Miklos Szeredi <[email protected]>
Reviewed-by: Ashish Samant <[email protected]>

fuse: device fd clone

Allow an open fuse device to be "cloned".  Userspace can create a clone by:

      newfd = open("/dev/fuse", O_RDWR)
      ioctl(newfd, FUSE_DEV_IOC_CLONE, &oldfd);

At this point newfd will refer to the same fuse connection as oldfd.

Signed-off-by: Miklos Szeredi <[email protected]>
Reviewed-by: Ashish Samant <[email protected]>

fuse: abort: no fc->lock needed for request ending

In fuse_abort_conn() when all requests are on private lists we no longer
need fc->lock protection.

Signed-off-by: Miklos Szeredi <[email protected]>
Reviewed-by: Ashish Samant <[email protected]>

fuse: no fc->lock for pqueue parts

Remove fc->lock protection from processing queue members, now protected by
fpq->lock.

Signed-off-by: Miklos Szeredi <[email protected]>
Reviewed-by: Ashish Samant <[email protected]>

fuse: no fc->lock in request_end()

No longer need to call request_end() with the connection lock held.  We
still protect the background counters and queue with fc->lock, so acquire
it if necessary.

Signed-off-by: Miklos Szeredi <[email protected]>
Reviewed-by: Ashish Samant <[email protected]>

fuse: cleanup request_end()

Now that we atomically test having already done everything we no longer
need other protection.

Signed-off-by: Miklos Szeredi <[email protected]>
Reviewed-by: Ashish Samant <[email protected]>

fuse: request_end(): do once

When the connection is aborted it is possible that request_end() will be
called twice.  Use atomic test and set to do the actual ending only once.

test_and_set_bit() also provides the necessary barrier semantics so no
explicit smp_wmb() is necessary.

Signed-off-by: Miklos Szeredi <[email protected]>
Reviewed-by: Ashish Samant <[email protected]>

fuse: add req flag for private list

When an unlocked request is aborted, it is moved from fpq->io to a private
list.  Then, after unlocking fpq->lock, the private list is processed and
the requests are finished off.

To protect the private list, we need to mark the request with a flag, so if
in the meantime the request is unlocked the list is not corrupted.

Signed-off-by: Miklos Szeredi <[email protected]>
Reviewed-by: Ashish Samant <[email protected]>

fuse: pqueue locking

Add a fpq->lock for protecting members of struct fuse_pqueue and FR_LOCKED
request flag.

Signed-off-by: Miklos Szeredi <[email protected]>
Reviewed-by: Ashish Samant <[email protected]>

fuse: abort: group pqueue accesses

Rearrange fuse_abort_conn() so that processing queue accesses are grouped
together.

Signed-off-by: Miklos Szeredi <[email protected]>
Reviewed-by: Ashish Samant <[email protected]>

fuse: cleanup fuse_dev_do_read()

 - locked list_add() + list_del_init() cancel out

 - common handling of case when request is ended here in the read phase

Signed-off-by: Miklos Szeredi <[email protected]>
Reviewed-by: Ashish Samant <[email protected]>

fuse: move list_del_init() from request_end() into callers

Signed-off-by: Miklos Szeredi <[email protected]>

fuse: duplicate ->connected in pqueue

This will allow checking ->connected just with the processing queue lock.

Signed-off-by: Miklos Szeredi <[email protected]>
Reviewed-by: Ashish Samant <[email protected]>

fuse: separate out processing queue

This is just two fields: fc->io and fc->processing.

This patch just rearranges the fields, no functional change.

Signed-off-by: Miklos Szeredi <[email protected]>
Reviewed-by: Ashish Samant <[email protected]>

fuse: simplify request_wait()

wait_event_interruptible_exclusive_locked() will do everything
request_wait() does, so replace it.

Signed-off-by: Miklos Szeredi <[email protected]>
Reviewed-by: Ashish Samant <[email protected]>

fuse: no fc->lock for iqueue parts

Remove fc->lock protection from input queue members, now protected by
fiq->waitq.lock.

Signed-off-by: Miklos Szeredi <[email protected]>
Reviewed-by: Ashish Samant <[email protected]>

fuse: allow interrupt queuing without fc->lock

Interrupt is only queued after the request has been sent to userspace.
This is either done in request_wait_answer() or fuse_dev_do_read()
depending on which state the request is in at the time of the interrupt.
If it's not yet sent, then queuing the interrupt is postponed until the
request is read.  Otherwise (the request has already been read and is
waiting for an answer) the interrupt is queued immedidately.

We want to call queue_interrupt() without fc->lock protection, in which
case there can be a race between the two functions:

 - neither of them queue the interrupt (thinking the other one has already
   done it).

 - both of them queue the interrupt

The first one is prevented by adding memory barriers, the second is
prevented by checking (under fiq->waitq.lock) if the interrupt has already
been queued.

Signed-off-by: Miklos Szeredi <[email protected]>

fuse: iqueue locking

Use fiq->waitq.lock for protecting members of struct fuse_iqueue and
FR_PENDING request flag, previously protected by fc->lock.

Following patches will remove fc->lock protection from these members.

Signed-off-by: Miklos Szeredi <[email protected]>
Reviewed-by: Ashish Samant <[email protected]>

fuse: dev read: split list_move

Different lists will need different locks.

Signed-off-by: Miklos Szeredi <[email protected]>
Reviewed-by: Ashish Samant <[email protected]>

fuse: abort: group iqueue accesses

Rearrange fuse_abort_conn() so that input queue accesses are grouped
together.

Signed-off-by: Miklos Szeredi <[email protected]>
Reviewed-by: Ashish Samant <[email protected]>

fuse: duplicate ->connected in iqueue

This will allow checking ->connected just with the input queue lock.

Signed-off-by: Miklos Szeredi <[email protected]>
Reviewed-by: Ashish Samant <[email protected]>

fuse: separate out input queue

The input queue contains normal requests (fc->pending), forgets
(fc->forget_*) and interrupts (fc->interrupts).  There's also fc->waitq and
fc->fasync for waking up the readers of the fuse device when a request is
available.

The fc->reqctr is also moved to the input queue (assigned to the request
when the request is added to the input queue.

This patch just rearranges the fields, no functional change.

Signed-off-by: Miklos Szeredi <[email protected]>
Reviewed-by: Ashish Samant <[email protected]>

fuse: req state use flags

Use flags for representing the state in fuse_req.  This is needed since
req->list will be protected by different locks in different states, hence
we'll want the state itself to be split into distinct bits, each protected
with the relevant lock in that state.

Signed-off-by: Miklos Szeredi <[email protected]>

fuse: simplify req states

FUSE_REQ_INIT is actually the same state as FUSE_REQ_PENDING and
FUSE_REQ_READING and FUSE_REQ_WRITING can be merged into a common
FUSE_REQ_IO state.

Signed-off-by: Miklos Szeredi <[email protected]>
Reviewed-by: Ashish Samant <[email protected]>

fuse: don't hold lock over request_wait_answer()

Only hold fc->lock over sections of request_wait_answer() that actually
need it.  If wait_event_interruptible() returns zero, it means that the
request finished.  Need to add memory barriers, though, to make sure that
all relevant data in the request is synchronized.

Signed-off-by: Miklos Szeredi <[email protected]>

fuse: simplify unique ctr

Since it's a 64bit counter, it's never gonna wrap around.  Remove code
dealing with that possibility.

Signed-off-by: Miklos Szeredi <[email protected]>
Reviewed-by: Ashish Samant <[email protected]>

fuse: rework abort

Splice fc->pending and fc->processing lists into a common kill list while
holding fc->lock.

By the time we release fc->lock, pending and processing lists are empty and
the io list contains only locked requests.

Signed-off-by: Miklos Szeredi <[email protected]>
Reviewed-by: Ashish Samant <[email protected]>

fuse: fold helpers into abort

Fold end_io_requests() and end_queued_requests() into fuse_abort_conn().

Signed-off-by: Miklos Szeredi <[email protected]>
Reviewed-by: Ashish Samant <[email protected]>

fuse: use per req lock for lock/unlock_request()

Reuse req->waitq.lock for protecting FR_ABORTED and FR_LOCKED flags.

Signed-off-by: Miklos Szeredi <[email protected]>
Reviewed-by: Ashish Samant <[email protected]>

fuse: req use bitops

Finer grained locking will mean there's no single lock to protect
modification of bitfileds in fuse_req.

So move to using bitops.  Can use the non-atomic variants for those which
happen while the request definitely has only one reference.

Signed-off-by: Miklos Szeredi <[email protected]>
Reviewed-by: Ashish Samant <[email protected]>

fuse: simplify request abort

 - don't end the request while req->locked is true

 - make unlock_request() return an error if the connection was aborted

Signed-off-by: Miklos Szeredi <[email protected]>
Reviewed-by: Ashish Samant <[email protected]>

fuse: call fuse_abort_conn() in dev release

fuse_abort_conn() does all the work done by fuse_dev_release() and more.
"More" consists of:

end_io_requests(fc);
wake_up_all(&fc->waitq);
kill_fasync(&fc->fasync, SIGIO, POLL_IN);

All of which should be no-op (WARN_ON's added).

Signed-off-by: Miklos Szeredi <[email protected]>
Reviewed-by: Ashish Samant <[email protected]>

fuse: fold fuse_request_send_nowait() into single caller

And the same with fuse_request_send_nowait_locked().

Signed-off-by: Miklos Szeredi <[email protected]>
Reviewed-by: Ashish Samant <[email protected]>

fuse: check conn_error earlier

fc->conn_error is set once in FUSE_INIT reply and never cleared.  Check it
in request allocation, there's no sense in doing all the preparation if
sending will surely fail.

Signed-off-by: Miklos Szeredi <[email protected]>
Reviewed-by: Ashish Samant <[email protected]>

fuse: account as waiting before queuing for background

Move accounting of fc->num_waiting to the point where the request actually
starts waiting.  This is earlier than the current queue_request() for
background requests, since they might be waiting on the fc->bg_queue before
being queued on fc->pending.

Signed-off-by: Miklos Szeredi <[email protected]>
Reviewed-by: Ashish Samant <[email protected]>

fuse: reset waiting

Reset req->waiting in fuse_put_request().  This is needed for correct
accounting in fc->num_waiting for reserved requests.

Signed-off-by: Miklos Szeredi <[email protected]>

fuse: fix background request if not connected

request_end() expects fc->num_background and fc->active_background to have
been incremented, which is not the case in fuse_request_send_nowait()
failure path.  So instead just call the ->end() callback (which is actually
set by all callers).

Signed-off-by: Miklos Szeredi <[email protected]>
Reviewed-by: Ashish Samant <[email protected]>

fuse: initialize fc->release before calling it

fc->release is called from fuse_conn_put() which was used in the error
cleanup before fc->release was initialized.

[Jeremiah Mahler <[email protected]>: assign fc->release after calling
fuse_conn_init(fc) instead of before.]

Signed-off-by: Miklos Szeredi <[email protected]>
Fixes: a325f9b92273 ("fuse: update fuse_conn_init() and separate out fuse_conn_kill()")
Cc: <[email protected]> #v2.6.31+

arm64: Don't report clear pmds and puds as huge

The current pmd_huge() and pud_huge() functions simply check if the table
bit is not set and reports the entries as huge in that case.  This is
counter-intuitive as a clear pmd/pud cannot also be a huge pmd/pud, and
it is inconsistent with at least arm and x86.

To prevent others from making the same mistake as me in looking at code
that calls these functions and to fix an issue with KVM on arm64 that
causes memory corruption due to incorrect page reference counting
resulting from this mistake, let's change the behavior.

Signed-off-by: Christoffer Dall <[email protected]>
Reviewed-by: Steve Capper <[email protected]>
Acked-by: Marc Zyngier <[email protected]>
Fixes: 084bd29810a5 ("ARM64: mm: HugeTLB support.")
Cc: <[email protected]> # 3.11+
Signed-off-by: Catalin Marinas <[email protected]>

hwspinlock: qcom: Correct msb in regmap_field

msb of the regmap_field was mistakenly given the value 32, to set all bits
in the regmap update mask; although incorrect this worked until 921cc294,
where the mask calculation was corrected.

Signed-off-by: Bjorn Andersson <[email protected]>
Signed-off-by: Ohad Ben-Cohen <[email protected]>

printk: Increase maximum CONFIG_LOG_BUF_SHIFT from 21 to 25

So I tried to some kernel debugging that produced a ton of kernel messages
on a big box, and wanted to save them all: but CONFIG_LOG_BUF_SHIFT maxes
out at 21 (2 MB).

Increase it to 25 (32 MB).

This does not affect any existing config or defaults.

Cc: Linus Torvalds <[email protected]>
Cc: Andrew Morton <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: [email protected]
Signed-off-by: Ingo Molnar <[email protected]>

time: Remove development rules from Kbuild/Makefile

time.o gets rebuilt unconditionally due to a leftover Makefile rule
which was placed there for development purposes.

Remove it along with the commented out always rule in the toplevel
Kbuild file.

Fixes: 0a227985d4a9 'time: Move timeconst.h into include/generated'
Reported-by; Stephen Boyd <[email protected]>
Signed-off-by: Thomas Gleixner <[email protected]>
Cc: Nicholas Mc Guire <[email protected]>

iommu/amd: Introduce protection_domain_init() function

This function contains the common parts between the
initialization of dma_ops_domains and usual protection
domains. This also fixes a long-standing bug which was
uncovered by recent changes, in which the api_lock was not
initialized for dma_ops_domains.

Reported-by: George Wang <[email protected]>
Tested-by: George Wang <[email protected]>
Signed-off-by: Joerg Roedel <[email protected]>

fs/file.c: __fget() and dup2() atomicity rules

__fget() does lockless fetch of pointer from the descriptor
table, attempts to grab a reference and treats "it was already
zero" as "it's already gone from the table, we just hadn't
seen the store, let's fail".  Unfortunately, that breaks the
atomicity of dup2() - __fget() might see the old pointer,
notice that it's been already dropped and treat that as
"it's closed".  What we should be getting is either the
old file or new one, depending whether we come before or after
dup2().

Dmitry had following test failing sometimes :

int fd;
void *Thread(void *x) {
  char buf;
  int n = read(fd, &buf, 1);
  if (n != 1)
    exit(printf("read failed: n=%d errno=%d\n", n, errno));
  return 0;
}

int main()
{
  fd = open("/dev/urandom", O_RDONLY);
  int fd2 = open("/dev/urandom", O_RDONLY);
  if (fd == -1 || fd2 == -1)
    exit(printf("open failed\n"));
  pthread_t th;
  pthread_create(&th, 0, Thread, 0);
  if (dup2(fd2, fd) == -1)
    exit(printf("dup2 failed\n"));
  pthread_join(th, 0);
  if (close(fd) == -1)
    exit(printf("close failed\n"));
  if (close(fd2) == -1)
    exit(printf("close failed\n"));
  printf("DONE\n");
  return 0;
}

Signed-off-by: Eric Dumazet <[email protected]>
Reported-by: Dmitry Vyukov <[email protected]>
Signed-off-by: Al Viro <[email protected]>

fs/file.c: don't acquire files->file_lock in fd_install()

Mateusz Guzik reported :

 Currently obtaining a new file descriptor results in locking fdtable
 twice - once in order to reserve a slot and second time to fill it.

Holding the spinlock in __fd_install() is needed in case a resize is
done, or to prevent a resize.

Mateusz provided an RFC patch and a micro benchmark :
  http://people.redhat.com/~mguzik/pipebench.c

A resize is an unlikely operation in a process lifetime,
as table size is at least doubled at every resize.

We can use RCU instead of the spinlock.

__fd_install() must wait if a resize is in progress.

The resize must block new __fd_install() callers from starting,
and wait that ongoing install are finished (synchronize_sched())

resize should be attempted by a single thread to not waste resources.

rcu_sched variant is used, as __fd_install() and expand_fdtable() run
from process context.

It gives us a ~30% speedup using pipebench on a dual Intel(R) Xeon(R)
CPU E5-2696 v2 @ 2.50GHz

Signed-off-by: Eric Dumazet <[email protected]>
Reported-by: Mateusz Guzik <[email protected]>
Acked-by: Mateusz Guzik <[email protected]>
Tested-by: Mateusz Guzik <[email protected]>
Signed-off-by: Al Viro <[email protected]>

fs:super:get_anon_bdev: fix race condition could cause dev exceed its upper limitation

Execution of get_anon_bdev concurrently and preemptive kernel all
could bring race condition, it isn't enough to check dev against
its upper limitation with equality operator only.

This patch fix it.

Signed-off-by: Wang YanQing <[email protected]>
Signed-off-by: Al Viro <[email protected]>

Merge git://git.kernel.org/pub/scm/linux/kernel/git/cmetcalf/linux-tile

Pull arch/tile updates from Chris Metcalf:
 "These are a grab bag of changes to improve debugging and respond to a
  variety of issues raised on LKML over the last couple of months"

* git://git.kernel.org/pub/scm/linux/kernel/git/cmetcalf/linux-tile:
  tile: avoid a "label not used" warning in do_page_fault()
  tile: vdso: use raw_read_seqcount_begin() in vdso
  tile: force CONFIG_TILEGX if ARCH != tilepro
  tile: improve stack backtrace
  tile: fix "odd fault" warning for stack backtraces
  tile: set up initial stack top to honor STACK_TOP_DELTA
  tile: support delivering NMIs for multicore backtrace
  drivers/tty/hvc/hvc_tile.c: properly return -EAGAIN
  tile: add <asm/word-at-a-time.h> and enable support functions
  tile: use READ_ONCE() in arch_spin_is_locked()
  tile: modify arch_spin_unlock_wait() semantics

Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/s390/linux

Pull more s390 updates from Martin Schwidefsky:
 "There is one larger patch for the AP bus code to make it work with the
  longer reset periods of the latest crypto cards.

  A new default configuration, a naming cleanup for SMP and a few fixes"

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/s390/linux:
  s390/kdump: fix compile for !SMP
  s390/kdump: fix nosmt kernel parameter
  s390: new default configuration
  s390/smp: cleanup core vs. cpu in the SCLP interface
  s390/smp: fix sigp cpu detection loop
  s390/zcrypt: Fixed reset and interrupt handling of AP queues
  s390/kdump: fix REGSET_VX_LOW vector register ELF notes
  s390/bpf: Fix backward jumps

Merge branch 'for-next' of git://git.samba.org/sfrench/cifs-2.6

Pull CIFS/SMB3 updates from Steve French:
 "Includes two bug fixes, as well as (minimal) support for the new
  protocol dialect (SMB3.1.1), and support for two ioctls including
  reflink (duplicate extents) file copy and set integrity"

* 'for-next' of git://git.samba.org/sfrench/cifs-2.6:
  cifs: Unset CIFS_MOUNT_POSIX_PATHS flag when following dfs mounts
  Update negotiate protocol for SMB3.11 dialect
  Add ioctl to set integrity
  Add Get/Set Integrity Information structure definitions
  Add reflink copy over SMB3.11 with new FSCTL_DUPLICATE_EXTENTS
  Add SMB3.11 mount option synonym for new dialect
  add struct FILE_STANDARD_INFO
  Make dialect negotiation warning message easier to read
  Add defines and structs for smb3.1 dialect
  Allow parsing vers=3.11 on cifs mount
  client MUST ignore EncryptionKeyLength if CAP_EXTENDED_SECURITY is set

vfs: avoid creation of inode number 0 in get_next_ino

currently, get_next_ino() is able to create inodes with inode number = 0.
This have a bad impact in the filesystems relying in this function to generate
inode numbers.

While there is no problem at all in having inodes with number 0, userspace tools
which handle file management tasks can have problems handling these files, like
for example, the impossiblity of users to delete these files, since glibc will
ignore them. So, I believe the best way is kernel to avoid creating them.

This problem has been raised previously, but the old thread didn't have any
other update for a year+, and I've seen too many users hitting the same issue
regarding the impossibility to delete files while using filesystems relying on
this function. So, I'm starting the thread again, with the same patch
that I believe is enough to address this problem.

Signed-off-by: Carlos Maiolino <[email protected]>
Signed-off-by: Al Viro <[email protected]>

Merge tag 'xfs-for-linus-4.2-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/dgc/linux-xfs

Pul xfs updates from Dave Chinner:
 "There's a couple of small API changes to the core DAX code which
  required small changes to the ext2 and ext4 code bases, but otherwise
  everything is within the XFS codebase.

  This update contains:

   - A new sparse on-disk inode record format to allow small extents to
     be used for inode allocation when free space is fragmented.

   - DAX support.  This includes minor changes to the DAX core code to
     fix problems with lock ordering and bufferhead mapping abuse.

   - transaction commit interface cleanup

   - removal of various unnecessary XFS specific type definitions

   - cleanup and optimisation of freelist preparation before allocation

   - various minor cleanups

   - bug fixes for
- transaction reservation leaks
- incorrect inode logging in unwritten extent conversion
- mmap lock vs freeze ordering
- remote symlink mishandling
- attribute fork removal issues"

* tag 'xfs-for-linus-4.2-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/dgc/linux-xfs: (49 commits)
  xfs: don't truncate attribute extents if no extents exist
  xfs: clean up XFS_MIN_FREELIST macros
  xfs: sanitise error handling in xfs_alloc_fix_freelist
  xfs: factor out free space extent length check
  xfs: xfs_alloc_fix_freelist() can use incore perag structures
  xfs: remove xfs_caddr_t
  xfs: use void pointers in log validation helpers
  xfs: return a void pointer from xfs_buf_offset
  xfs: remove inst_t
  xfs: remove __psint_t and __psunsigned_t
  xfs: fix remote symlinks on V5/CRC filesystems
  xfs: fix xfs_log_done interface
  xfs: saner xfs_trans_commit interface
  xfs: remove the flags argument to xfs_trans_cancel
  xfs: pass a boolean flag to xfs_trans_free_items
  xfs: switch remaining xfs_trans_dup users to xfs_trans_roll
  xfs: check min blks for random debug mode sparse allocations
  xfs: fix sparse inodes 32-bit compile failure
  xfs: add initial DAX support
  xfs: add DAX IO path support
  ...

Merge branch 'for-linus-4.2' of git://git.kernel.org/pub/scm/linux/kernel/git/mason/linux-btrfs

Pull btrfs updates from Chris Mason:
 "Outside of our usual batch of fixes, this integrates the subvolume
  quota updates that Qu Wenruo from Fujitsu has been working on for a
  few releases now.  He gets an extra gold star for making btrfs smaller
  this time, and fixing a number of quota corners in the process.

  Dave Sterba tested and integrated Anand Jain's sysfs improvements.
  Outside of exporting a symbol (ack'd by Greg) these are all internal
  to btrfs and it's mostly cleanups and fixes.  Anand also attached some
  of our sysfs objects to our internal device management structs instead
  of an object off the super block.  It will make device management
  easier overall and it's a better fit for how the sysfs files are used.
  None of the existing sysfs files are moved around.

  Thanks for all the fixes everyone"

* 'for-linus-4.2' of git://git.kernel.org/pub/scm/linux/kernel/git/mason/linux-btrfs: (87 commits)
  btrfs: delayed-ref: double free in btrfs_add_delayed_tree_ref()
  Btrfs: Check if kobject is initialized before put
  lib: export symbol kobject_move()
  Btrfs: sysfs: add support to show replacing target in the sysfs
  Btrfs: free the stale device
  Btrfs: use received_uuid of parent during send
  Btrfs: fix use-after-free in btrfs_replay_log
  btrfs: wait for delayed iputs on no space
  btrfs: qgroup: Make snapshot accounting work with new extent-oriented qgroup.
  btrfs: qgroup: Add the ability to skip given qgroup for old/new_roots.
  btrfs: ulist: Add ulist_del() function.
  btrfs: qgroup: Cleanup the old ref_node-oriented mechanism.
  btrfs: qgroup: Switch self test to extent-oriented qgroup mechanism.
  btrfs: qgroup: Switch to new extent-oriented qgroup mechanism.
  btrfs: qgroup: Switch rescan to new mechanism.
  btrfs: qgroup: Add new qgroup calculation function btrfs_qgroup_account_extents().
  btrfs: backref: Add special time_seq == (u64)-1 case for btrfs_find_all_roots().
  btrfs: qgroup: Add new function to record old_roots.
  btrfs: qgroup: Record possible quota-related extent for qgroup.
  btrfs: qgroup: Add function qgroup_update_counters().
  ...

Merge branch 'for-linus' of git://git.kernel.dk/linux-block

Pull more block layer patches from Jens Axboe:
 "A few later arrivers that I didn't fold into the first pull request,
  so we had a chance to run some testing.  This contains:

   - NVMe:
        - Set of fixes from Keith
        - 4.4 and earlier gcc build fix from Andrew

   - small set of xen-blk{back,front} fixes from Bob Liu.

   - warnings fix for bogus inline statement in I_BDEV() from Geert.

   - error code fixup for SG_IO ioctl from Paolo Bonzini"

* 'for-linus' of git://git.kernel.dk/linux-block:
  drivers/block/nvme-core.c: fix build with gcc-4.4.4
  bdi: Remove "inline" keyword from exported I_BDEV() implementation
  block: fix bogus EFAULT error from SG_IO ioctl
  NVMe: Fix filesystem deadlock on removal
  NVMe: Failed controller initialization fixes
  NVMe: Unify controller probe and resume
  NVMe: Don't use fake status on cancelled command
  NVMe: Fix device cleanup on initialization failure
  drivers: xen-blkfront: only talk_to_blkback() when in XenbusStateInitialising
  xen/block: add multi-page ring support
  driver: xen-blkfront: move talk_to_blkback to a more suitable place
  drivers: xen-blkback: delay pending_req allocation to connect_ring

genalloc: rename of_get_named_gen_pool() to of_gen_pool_get()

To be consistent with other kernel interface namings, rename
of_get_named_gen_pool() to of_gen_pool_get().  In the original function
name "_named" suffix references to a device tree property, which contains
a phandle to a device and the corresponding device driver is assumed to
register a gen_pool object.

Due to a weak relation and to avoid any confusion (e.g.  in future
possible scenario if gen_pool objects are named) the suffix is removed.

[[email protected]: crypto/marvell/cesa - fix up for of_get_named_gen_pool() rename]
Signed-off-by: Vladimir Zapolskiy <[email protected]>
Cc: Nicolas Ferre <[email protected]>
Cc: Philipp Zabel <[email protected]>
Cc: Shawn Guo <[email protected]>
Cc: Sascha Hauer <[email protected]>
Cc: Alexandre Belloni <[email protected]>
Cc: Russell King <[email protected]>
Cc: Mauro Carvalho Chehab <[email protected]>
Cc: Vinod Koul <[email protected]>
Cc: Takashi Iwai <[email protected]>
Cc: Jaroslav Kysela <[email protected]>
Signed-off-by: Stephen Rothwell <[email protected]>
Cc: Herbert Xu <[email protected]>
Cc: Boris BREZILLON <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

genalloc: rename dev_get_gen_pool() to gen_pool_get()

To be consistent with other genalloc interface namings, rename
dev_get_gen_pool() to gen_pool_get().  The original omitted "dev_" prefix
is removed, since it points to argument type of the function, and so it
does not bring any useful information.

[[email protected]: update arch/arm/mach-socfpga/pm.c]
Signed-off-by: Vladimir Zapolskiy <[email protected]>
Acked-by: Nicolas Ferre <[email protected]>
Cc: Philipp Zabel <[email protected]>
Cc: Shawn Guo <[email protected]>
Cc: Sascha Hauer <[email protected]>
Cc: Alexandre Belloni <[email protected]>
Cc: Russell King <[email protected]>
Cc: Mauro Carvalho Chehab <[email protected]>
Cc: Vinod Koul <[email protected]>
Cc: Takashi Iwai <[email protected]>
Cc: Jaroslav Kysela <[email protected]>
Cc: Mark Brown <[email protected]>
Cc: Nicolas Ferre <[email protected]>
Cc: Alan Tull <[email protected]>
Cc: Dinh Nguyen <[email protected]>
Cc: Kevin Hilman <[email protected]>
Cc: Stephen Rothwell <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

x86: opt into HAVE_COPY_THREAD_TLS, for both 32-bit and 64-bit

For 32-bit userspace on a 64-bit kernel, this requires modifying
stub32_clone to actually swap the appropriate arguments to match
CONFIG_CLONE_BACKWARDS, rather than just leaving the C argument for tls
broken.

Patch co-authored by Josh Triplett and Thiago Macieira.

Signed-off-by: Josh Triplett <[email protected]>
Acked-by: Andy Lutomirski <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: "H. Peter Anvin" <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Thiago Macieira <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

MAINTAINERS: add zpool

Add entry for zpool to MAINTAINERS file.

Signed-off-by: Dan Streetman <[email protected]>
Cc: Seth Jennings <[email protected]>
Cc: Minchan Kim <[email protected]>
Cc: Nitin Gupta <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

MAINTAINERS: BCACHE: Kent Overstreet has changed email address

Kent's email address in MAINTAINERS seems to be invalid.
This was his last sign-off address, so use that if appropriate.

Fix the S: status entry while there.

Signed-off-by: Joe Perches <[email protected]>
Acked-by: Kent Overstreet <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

MAINTAINERS: move Jens Osterkamp to CREDITS

Jens' email address bounces, so move his name and entry to CREDITS.

Signed-off-by: Joe Perches <[email protected]>
Cc: Ishizaki Kou <[email protected]>
Cc: Jens Osterkamp <[email protected]>
Cc: Benjamin Herrenschmidt <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

MAINTAINERS: remove unused nbd.h pattern

Commit 13e71d69cc74 ("nbd: Remove kernel internal header") deleted the
file, remove the pattern.

Signed-off-by: Joe Perches <[email protected]>
Acked-by: Markus Pargmann <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

MAINTAINERS: update brcm gpio filename pattern

Commit 23a71fd616bf ("dt-bindings: brcm: rationalize Broadcom
documentation naming") renamed the file, update the pattern.

Signed-off-by: Joe Perches <[email protected]>
Acked-by: Scott Branden <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

MAINTAINERS: update brcm dts pattern

Commit 8c0b9ee8665c ("MIPS: Move device-trees into vendor
sub-directories") moved the files, update the pattern.

Signed-off-by: Joe Perches <[email protected]>
Cc: Andrew Bresticker <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

MAINTAINERS: update sound soc intel patterns

Commit 2106241a6803 ("ASoC: Intel: create common folder and move common
files in") moved the files around.  Update the patterns.

Signed-off-by: Joe Perches <[email protected]>
Cc: Jie Yang <[email protected]>
Acked-by: Jie Yang <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

MAINTAINERS: remove website for paride

The webpage mentioned is not working,

Signed-off-by: Sudip Mukherjee <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

MAINTAINERS: update Emulex ocrdma email addresses

@emulex.com addresses respond to use @avagotech.com.

Signed-off-by: Laurent Navet <[email protected]>
Acked-By: Devesh Sharma <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

bcache: use kvfree() in various places

Use kvfree() instead of open-coding it.

Signed-off-by: Pekka Enberg <[email protected]>
Cc: Kent Overstreet <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

libcxgbi: use kvfree() in cxgbi_free_big_mem()

Use kvfree() instead of open-coding it.

Signed-off-by: Pekka Enberg <[email protected]>
Cc: "James E.J. Bottomley" <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

target: use kvfree() in session alloc and free

Use kvfree() instead of open-coding it.

Signed-off-by: Pekka Enberg <[email protected]>
Cc: "Nicholas A. Bellinger" <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

IB/ehca: use kvfree() in ipz_queue_{cd}tor()

Use kvfree() instead of open-coding it.

Signed-off-by: Pekka Enberg <[email protected]>
Cc: Hoang-Nam Nguyen <[email protected]>
Cc: Christoph Raisch <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

drm/nouveau/gem: use kvfree() in u_free()

Use kvfree() instead of open-coding it.

Signed-off-by: Pekka Enberg <[email protected]>
Cc: David Airlie <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

drm: use kvfree() in drm_free_large()

Use kvfree() instead of open-coding it.

Signed-off-by: Pekka Enberg <[email protected]>
Cc: David Airlie <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

cxgb4: use kvfree() in t4_free_mem()

Use kvfree() instead of open-coding it.

Signed-off-by: Pekka Enberg <[email protected]>
Cc: Hariprasad S <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

cxgb3: use kvfree() in cxgb_free_mem()

Use kvfree() instead of open-coding it.

Signed-off-by: Pekka Enberg <[email protected]>
Cc: Santosh Raspatur <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

kernel/relay.c: use kvfree() in relay_free_page_array()

Use kvfree() instead of open-coding it.

Signed-off-by: Pekka Enberg <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

printk: improve the description of /dev/kmsg line format

The comment about /dev/kmsg does not mention the additional values which
may actually be exported, fix that.

Also move up the part of the comment instructing the users to ignore these
additional values, this way the reading is more fluent and logically
compact.

Signed-off-by: Antonio Ospite <[email protected]>
Cc: Joe Perches <[email protected]>
Cc: Jonathan Corbet <[email protected]>
Cc: Tejun Heo <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

arch/unicore32/kernel/fpu-ucf64.c: remove unnecessary KERN_ERR

Signed-off-by: Masanari Iida <[email protected]>
Cc: Guan Xuetao <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

drivers/scsi/scsi_debug.c: resolve sg buffer const-ness issue

do_device_access() takes a separate parameter to indicate the direction of
data transfer, which it used to use to select the appropriate function out
of sg_pcopy_{to,from}_buffer().  However these two functions now have

So this patch makes it bypass these wrappers and call the underlying
function sg_copy_buffer() directly; this has the same calling style as
do_device_access() i.e.  a separate direction-of-transfer parameter and no
pointers-to-const, so skipping the wrappers not only eliminates the
warning, it also make the code simpler :)

[[email protected]: fix very broken build]
Signed-off-by: Dave Gordon <[email protected]>
Acked-by: Arnd Bergmann <[email protected]>
Cc: James Bottomley <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

lib/scatterlist: mark input buffer parameters as 'const'

The 'buf' parameter of sg(p)copy_from_buffer() can and should be
const-qualified, although because of the shared implementation of
_to_buffer() and _from_buffer(), we have to cast this away internally.

This means that callers who have a 'const' buffer containing the data to
be copied to the sg-list no longer have to cast away the const-ness
themselves.  It also enables improved coverage by code analysis tools.

Signed-off-by: Dave Gordon <[email protected]>
Cc: Akinobu Mita <[email protected]>
Cc: "Martin K. Petersen" <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

lib/scatterlist.c: fix kerneldoc for sg_pcopy_{to,from}_buffer()

The kerneldoc for the functions doesn't match the code; the last two
parameters (buflen, skip) have been transposed, which is confusing,
especially as they're both integral types and the compiler won't warn
about swapping them.

These functions and the kerneldoc were introduced in commit:
    df642cea lib/scatterlist: introduce sg_pcopy_from_buffer() ...
    Author: Akinobu Mita <[email protected]>
    Date:   Mon Jul 8 16:01:54 2013 -0700

    The only difference between sg_pcopy_{from,to}_buffer() and
    sg_copy_{from,to}_buffer() is an additional argument that
    specifies the number of bytes to skip the SG list before
    copying.

The functions have the extra argument at the end, but the kerneldoc
lists it in penultimate position.

Signed-off-by: Dave Gordon <[email protected]>
Reviewed-by: Akinobu Mita <[email protected]>
Cc: "Martin K. Petersen" <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

ipc,sysv: return -EINVAL upon incorrect id/seqnum

In ipc_obtain_object_check we return -EIDRM when a bogus sequence number
is detected via ipc_checkid, while the ipc manpages state the following
return codes for such errors:

   EIDRM  <ID> points to a removed identifier.
   EINVAL Invalid <ID> value, or unaligned, etc.

EIDRM should only be returned upon a RMID call (->deleted check), and thus
return EINVAL for wrong seq.  This difference in semantics has also caused
real bugs, ie: https://bugzilla.redhat.com/show_bug.cgi?id=246509

Signed-off-by: Davidlohr Bueso <[email protected]>
Cc: Manfred Spraul <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

ipc,sysv: make return -EIDRM when racing with RMID consistent

The ipc_lock helper is used by all forms of sysv ipc to acquire the ipc
object's spinlock.  Upon error (bogus identifier), we always return
-EINVAL, whether the problem be in the idr path or because we raced with a
task performing RMID.  For the later, however, all ipc related manpages,
state the that for:

       EIDRM  <ID> points to a removed identifier.

And return:

       EINVAL Invalid <ID> value, or unaligned, etc.

Which (EINVAL) should only return once the ipc resource is deleted.  For
all types of ipc this is done immediately upon a RMID command.  However,
shared memory behaves slightly different as it can merely mark a segment
for deletion, and delay the actual freeing until there are no more active
consumers.  Per shmctl(IPC_RMID) manpage:

""
Mark  the  segment to be destroyed.  The segment will only actually
be destroyed after the last process detaches it (i.e., when the
shm_nattch member of the associated structure shmid_ds is zero).
""

Unlike ipc_lock, paths that behave "correctly", at least per the manpage,
involve controlling the ipc resource via *ctl(), doing the exact same
validity check as ipc_lock after right acquiring the spinlock:

if (!ipc_valid_object()) {
err = -EIDRM;
goto out_unlock;
}

Thus make ipc_lock consistent with the rest of ipc code and return -EIDRM
in ipc_lock when !ipc_valid_object().

Signed-off-by: Davidlohr Bueso <[email protected]>
Cc: Manfred Spraul <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

ipc: rename ipc_obtain_object

...  to ipc_obtain_object_idr, which is more meaningful and makes the code
slightly easier to follow.

Signed-off-by: Davidlohr Bueso <[email protected]>
Cc: Manfred Spraul <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

ipc,msg: provide barrier pairings for lockless receive

We currently use a full barrier on the sender side to to avoid receiver
tasks disappearing on us while still performing on the sender side wakeup.
 We lack however, the proper CPU-CPU interactions pairing on the receiver
side which busy-waits for the message.  Similarly, we do not need a full
smp_mb, and can relax the semantics for the writer and reader sides of the
message.  This is safe as we are only ordering loads and stores to r_msg.
And in both smp_wmb and smp_rmb, there are no stores after the calls
_anyway_.

This obviously applies for pipelined_send and expunge_all, for EIRDM when
destroying a queue.

Signed-off-by: Davidlohr Bueso <[email protected]>
Cc: Manfred Spraul <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

ipc,shm: move BUG_ON check into shm_lock

Upon every shm_lock call, we BUG_ON if an error was returned, indicating
racing either in idr or in shm_destroy.  Move this logic into the locking.

[[email protected]: simplify code]
Signed-off-by: Davidlohr Bueso <[email protected]>
Cc: Manfred Spraul <[email protected]>
Cc: Davidlohr Bueso <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

ipc/util.c: use kvfree() in ipc_rcu_free()

Use kvfree() instead of open-coding it.

Signed-off-by: Pekka Enberg <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

arc: use for_each_sg()

This replaces the plain loop over the sglist array with for_each_sg()
macro which consists of sg_next() function calls.  Since arc doesn't
select ARCH_HAS_SG_CHAIN, it is not necessary to use for_each_sg() in
order to loop over each sg element.  But this can help find problems with
drivers that do not properly initialize their sg tables when
CONFIG_DEBUG_SG is enabled.

Signed-off-by: Akinobu Mita <[email protected]>
Acked-by: Vineet Gupta <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

devpts: if initialization failed, don't crash when opening /dev/ptmx

If devpts failed to initialize, it would store an ERR_PTR in the global
devpts_mnt.  A subsequent open of /dev/ptmx would call devpts_new_index,
which would dereference devpts_mnt and crash.

Avoid storing invalid values in devpts_mnt; leave it NULL instead.  Make
both devpts_new_index and devpts_pty_new fail gracefully with ENODEV in
that case, which then becomes the return value to the userspace open call
on /dev/ptmx.

[[email protected]: remove unneeded static]
Signed-off-by: Josh Triplett <[email protected]>
Reported-by: Fengguang Wu <[email protected]>
Reviewed-by: Peter Hurley <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

scripts/gdb: remove useless global instruction

Signed-off-by: Thiébaud Weksteen <[email protected]>
Signed-off-by: Jan Kiszka <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

scripts/gdb: add ps command

Signed-off-by: Thiébaud Weksteen <[email protected]>
Signed-off-by: Jan Kiszka <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

scripts/gdb: fix PEP8 compliance

Signed-off-by: Thiébaud Weksteen <[email protected]>
Signed-off-by: Jan Kiszka <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

scripts/gdb: fix typo in exception name

Signed-off-by: Thiébaud Weksteen <[email protected]>
Signed-off-by: Jan Kiszka <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

scripts/gdb: enable completion for lx-list-check parameter

Signed-off-by: Jan Kiszka <[email protected]>
Cc: Thiébaud Weksteen <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

scripts/gdb: also allow list_head pointer as lx-list-check paramter

This makes the usage more flexible.

Signed-off-by: Jan Kiszka <[email protected]>
Cc: Thiébaud Weksteen <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

scripts/gdb: add command to check list consistency

Add a gdb script to verify the consistency of lists.

Signed-off-by: Thiébaud Weksteen <[email protected]>
Signed-off-by: Jan Kiszka <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

memstick: remove deprecated use of pci api

Replace occurences of the pci api by appropriate call to the dma api.

A simplified version of the semantic patch that finds this problem is as
follows: (http://coccinelle.lip6.fr)

@deprecated@
idexpression id;
position p;
@@

(
  pci_dma_supported@p ( id, ...)
|
  pci_alloc_consistent@p ( id, ...)
)

@bad1@
idexpression id;
position deprecated.p;
@@
...when != &id->dev
   when != pci_get_drvdata ( id )
   when != pci_enable_device ( id )
(
  pci_dma_supported@p ( id, ...)
|
  pci_alloc_consistent@p ( id, ...)
)

@depends on !bad1@
idexpression id;
expression direction;
position deprecated.p;
@@

(
- pci_dma_supported@p ( id,
+ dma_supported ( &id->dev,
...
+ , GFP_ATOMIC
  )
|
- pci_alloc_consistent@p ( id,
+ dma_alloc_coherent ( &id->dev,
...
+ , GFP_ATOMIC
  )
)

Signed-off-by: Quentin Lambert <[email protected]>
Cc: Maxim Levitsky <[email protected]>
Cc: Greg KH <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

fs/affs/symlink.c: remove unneeded err variable

err is only assigned to -EIO.  Return that value at the end of fail
context.

Signed-off-by: Fabian Frederick <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

fs/affs/amigaffs.c: remove unneeded initialization

bh is initialized unconditionally in affs_remove_link()

Signed-off-by: Fabian Frederick <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>