__blk_mq_alloc_rq_maps() can be invoked multiple times, if we scale
back the queue depth if we are low on memory. So don't clear
set->tags when we fail, this is handled directly in
the parent function, blk_mq_alloc_tag_set().
Reported-by: Robert Elliott  <Elliott@hp.com>
Signed-off-by: Jens Axboe <axboe@fb.com>

We should not insert requests into the flush state machine from
blk_mq_insert_request.  All incoming flush requests come through
blk_{m,s}q_make_request and are handled there, while blk_execute_rq_nowait
should only be called for BLOCK_PC requests.  All other callers
deal with requests that already went through the flush statemchine
and shouldn't be reinserted into it.
Reported-by: Robert Elliott  <Elliott@hp.com>
Debugged-by: Ming Lei <ming.lei@canonical.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>

This patch should fix the bug reported in
https://lkml.org/lkml/2014/9/11/249.

We have to initialize at least the atomic_flags and the cmd_flags when
allocating storage for the requests.

Otherwise blk_mq_timeout_check() might dereference uninitialized
pointers when racing with the creation of a request.

Also move the reset of cmd_flags for the initializing code to the point
where a request is freed. So we will never end up with pending flush
request indicators that might trigger dereferences of invalid pointers
in blk_mq_timeout_check().

Cc: stable@vger.kernel.org
Signed-off-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Reported-by: Paulo De Rezende Pinatti <ppinatti@linux.vnet.ibm.com>
Tested-by: Paulo De Rezende Pinatti <ppinatti@linux.vnet.ibm.com>
Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Jens Axboe <axboe@fb.com>

When we start the request, we set the deadline and flip the bits
marking the request as started and non-complete. However, it's
important that the deadline store is ordered before flipping the
bits, otherwise we could have a small window where the request is
marked started but with an invalid deadline. This can confuse the
timeout handling.
Suggested-by: Ming Lei <tom.leiming@gmail.com>
Signed-off-by: Jens Axboe <axboe@fb.com>

Pull gfs2 fixes from Steven Whitehouse:
 "Here are a number of small fixes for GFS2.

  There is a fix for FIEMAP on large sparse files, a negative dentry
  hashing fix, a fix for flock, and a bug fix relating to d_splice_alias
  usage.

  There are also (patches 1 and 5) a couple of updates which are less
  critical, but small and low risk"

* tag 'gfs2-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/steve/gfs2-3.0-fixes:
  GFS2: fix d_splice_alias() misuses
  GFS2: Don't use MAXQUOTAS value
  GFS2: Hash the negative dentry during inode lookup
  GFS2: Request demote when a "try" flock fails
  GFS2: Change maxlen variables to size_t
  GFS2: fs/gfs2/super.c: replace seq_printf by seq_puts

Commit d6bb3e90 ("vfs: simplify and shrink stack frame of
link_path_walk()") introduced build problems with GCC versions older
than 4.6 due to the initialisation of a member of an anonymous union in
struct qstr without enclosing braces.

This hits GCC bug 10676 [1] (which was fixed in GCC 4.6 by [2]), and
causes the following build error:

  fs/namei.c: In function 'link_path_walk':
  fs/namei.c:1778: error: unknown field 'hash_len' specified in initializer

This is worked around by adding explicit braces.

[1] https://gcc.gnu.org/bugzilla/show_bug.cgi?id=10676
[2] https://gcc.gnu.org/viewcvs/gcc?view=revision&revision=159206

Fixes: d6bb3e90 (vfs: simplify and shrink stack frame of link_path_walk())
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: linux-fsdevel@vger.kernel.org
Cc: linux-metag@vger.kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Pull virtio fixes from Rusty Russell:
 "virtio-rng corner case fixes, with cc:stable.

  Survived a few days in linux-next"

* tag 'fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rusty/linux:
  virtio-rng: skip reading when we start to remove the device
  virtio-rng: fix stuck of hot-unplugging busy device

Pull regmap fix from Mark Brown:
 "Fix registers file in debugfs

  Ensure that the mode reported for the registers file in debugfs is
  accurate by marking it as read only when the define to enable writes
  has not been set.  This is on the edge of being a bug fix but it's
  debugfs and it makes it much easier for users to spot what's going
  wrong when they forget to enable writeability"

* tag 'regmap-v3.17-rc5' of git://git.kernel.org/pub/scm/linux/kernel/git/broonie/regmap:
  regmap: Fix debugfs-file 'registers' mode

Pull input updates from Dmitry Torokhov:
 "A few quirks for i8042/AT keyboards and a small device tree doc fix
  for Atmel Touchscreens"

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input:
  Input: atmel_mxt_ts - fix merge in DT documentation
  Input: i8042 - also set the firmware id for MUXed ports
  Input: i8042 - add nomux quirk for Avatar AVIU-145A6
  Input: i8042 - add Fujitsu U574 to no_timeout dmi table
  Input: atkbd - do not try 'deactivate' keyboard on any LG laptops

Cut & paste typo from the line above.
Reported-by: Ben Hutchings <ben@decadent.org.uk>
Signed-off-by: Tony Luck <tony.luck@intel.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Commit 9226b5b4 ("vfs: avoid non-forwarding large load after small
store in path lookup") made link_path_walk() always access the
"hash_len" field as a single 64-bit entity, in order to avoid mixed size
accesses to the members.

However, what I didn't notice was that that effectively means that the
whole "struct qstr this" is now basically redundant.  We already
explicitly track the "const char *name", and if we just use "u64
hash_len" instead of "long len", there is nothing else left of the
"struct qstr".

We do end up wanting the "struct qstr" if we have a filesystem with a
"d_hash()" function, but that's a rare case, and we might as well then
just squirrell away the name and hash_len at that point.

End result: fewer live variables in the loop, a smaller stack frame, and
better code generation.  And we don't need to pass in pointers variables
to helper functions any more, because the return value contains all the
relevant information.  So this removes more lines than it adds, and the
source code is clearer too.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Pull crypto fixes from Herbert Xu:
 "This fixes the newly added drbg generator so that it actually works on
  32-bit machines.  Previously the code was only tested on 64-bit and on
  32-bit it overflowed and simply doesn't work"

* git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6:
  crypto: drbg - remove check for uninitialized DRBG handle
  crypto: drbg - backport "fix maximum value checks on 32 bit systems"

Pull vfs fixes from Al Viro:
 "double iput() on failure exit in lustre, racy removal of spliced
  dentries from ->s_anon in __d_materialise_dentry() plus a bunch of
  assorted RCU pathwalk fixes"

The RCU pathwalk fixes end up fixing a couple of cases where we
incorrectly dropped out of RCU walking, due to incorrect initialization
and testing of the sequence locks in some corner cases.  Since dropping
out of RCU walk mode forces the slow locked accesses, those corner cases
slowed down quite dramatically.

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs:
  be careful with nd->inode in path_init() and follow_dotdot_rcu()
  don't bugger nd->seq on set_root_rcu() from follow_dotdot_rcu()
  fix bogus read_seqretry() checks introduced in b37199e6
  move the call of __d_drop(anon) into __d_materialise_unique(dentry, anon)
  [fix] lustre: d_make_root() does iput() on dentry allocation failure

The performance regression that Josef Bacik reported in the pathname
lookup (see commit 99d263d4 "vfs: fix bad hashing of dentries") made
me look at performance stability of the dcache code, just to verify that
the problem was actually fixed.  That turned up a few other problems in
this area.

There are a few cases where we exit RCU lookup mode and go to the slow
serializing case when we shouldn't, Al has fixed those and they'll come
in with the next VFS pull.

But my performance verification also shows that link_path_walk() turns
out to have a very unfortunate 32-bit store of the length and hash of
the name we look up, followed by a 64-bit read of the combined hash_len
field.  That screws up the processor store to load forwarding, causing
an unnecessary hickup in this critical routine.

It's caused by the ugly calling convention for the "hash_name()"
function, and easily fixed by just making hash_name() fill in the whole
'struct qstr' rather than passing it a pointer to just the hash value.

With that, the profile for this function looks much smoother.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Pull parisc updates from Helge Deller:
 "The most important patch is a new Light Weigth Syscall (LWS) for 8,
  16, 32 and 64 bit atomic CAS operations which is required in order to
  be able to implement the atomic gcc builtins on our platform.

  Other than that, we wire up the seccomp, getrandom and memfd_create
  syscalls, fixes a minor off-by-one bug and a wrong printk string"

* 'parisc-3.17-1' of git://git.kernel.org/pub/scm/linux/kernel/git/deller/parisc-linux:
  parisc: Implement new LWS CAS supporting 64 bit operations.
  parisc: Wire up seccomp, getrandom and memfd_create syscalls
  parisc: dino: fix %d confusingly prefixed with 0x in format string
  parisc: sys_hpux: NUL terminator is one past the end

in the former we simply check if dentry is still valid after picking
its ->d_inode; in the latter we fetch ->d_inode in the same places
where we fetch dentry and its ->d_seq, under the same checks.

Cc: stable@vger.kernel.org # 2.6.38+
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

return the value instead, and have path_init() do the assignment.  Broken by
"vfs: Fix absolute RCU path walk failures due to uninitialized seq number",
which was Cc-stable with 2.6.38+ as destination.  This one should go where
it went.

To avoid dummy value returned in case when root is already set (it would do
no harm, actually, since the only caller that doesn't ignore the return value
is guaranteed to have nd->root *not* set, but it's more obvious that way),
lift the check into callers.  And do the same to set_root(), to keep them
in sync.

Cc: stable@vger.kernel.org # 2.6.38+
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

Pull ntb driver bugfixes from Jon Mason:
 "NTB driver fixes for queue spread and buffer alignment.  Also, update
  to MAINTAINERS to reflect new e-mail address"

* tag 'ntb-3.17' of git://github.com/jonmason/ntb:
  ntb: Add alignment check to meet hardware requirement
  MAINTAINERS: update NTB info
  NTB: correct the spread of queues over mw's

Pull ARM irq chip fixes from Thomas Gleixner:
 "Another pile of ARM specific irq chip fixlets:

   - off by one bugs in the crossbar driver
   - missing annotations
   - a bunch of "make it compile" updates

  I pulled the lot today from Jason, but it has been in -next for at
  least a week"

* 'irq-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  irqchip: gic-v3: Declare rdist as __percpu pointer to __iomem pointer
  irqchip: gic: Make gic_default_routable_irq_domain_ops static
  irqchip: exynos-combiner: Fix compilation error on ARM64
  irqchip: crossbar: Off by one bugs in init
  irqchip: gic-v3: Tag all low level accessors __maybe_unused
  irqchip: gic-v3: Only define gic_peek_irq() when building SMP

irqchip fixes for v3.17 from Jason Cooper

 - GIC/GICV3: Various fixlets
 - crossbar: Fix off-by-one bug
 - exynos-combiner: Fix arm64 build error

The NTB translate register must have the value to be BAR size aligned.
This alignment check make sure that the DMA memory allocated has the
proper alignment. Another requirement for NTB to function properly with
memory window BAR size greater or equal to 4M is to use the CMA feature
in 3.16 kernel with the appropriate CONFIG_CMA_ALIGNMENT and
CONFIG_CMA_SIZE_MBYTES set.
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Signed-off-by: Jon Mason <jdmason@kudzu.us>

Update my contact info to my personal email address and add Dave Jiang.
Signed-off-by: Jon Mason <jon.mason@intel.com>
Signed-off-by: Dave Jiang <dave.jiang@intel.com>

The detection of an uneven number of queues on the given memory windows
was not correct.  The mw_num is zero based and the mod should be
division to spread them evenly over the mw's.
Signed-off-by: Jon Mason <jon.mason@intel.com>

read_seqretry() returns true on mismatch, not on match...

Cc: stable@vger.kernel.org # 3.15+
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

and lock the right list there

Cc: stable@vger.kernel.org
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

double-free is a bad thing
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

Merge branches 'locking-urgent-for-linus' and 'timers-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip

Pull futex and timer fixes from Thomas Gleixner:
 "A oneliner bugfix for the jinxed futex code:

   - Drop hash bucket lock in the error exit path.  I really could slap
     myself for intruducing that bug while fixing all the other horror
     in that code three month ago ...

  and the timer department is not too proud about the following fixes:

   - Deal with a long standing rounding bug in the timeval to jiffies
     conversion.  It's a real issue and this fix fell through the cracks
     for quite some time.

   - Another round of alarmtimer fixes.  Finally this code gets used
     more widely and the subtle issues hidden for quite some time are
     noticed and fixed.  Nothing really exciting, just the itty bitty
     details which bite the serious users here and there"

* 'locking-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  futex: Unlock hb->lock in futex_wait_requeue_pi() error path

* 'timers-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  alarmtimer: Lock k_itimer during timer callback
  alarmtimer: Do not signal SIGEV_NONE timers
  alarmtimer: Return relative times in timer_gettime
  jiffies: Fix timeval conversion to jiffies

The current LWS cas only works correctly for 32bit. The new LWS allows
for CAS operations of variable size.
Signed-off-by: Guy Martin <gmsoft@tuxicoman.be>
Cc: <stable@vger.kernel.org> # 3.13+
Signed-off-by: Helge Deller <deller@gmx.de>

Josef Bacik found a performance regression between 3.2 and 3.10 and
narrowed it down to commit bfcfaa77 ("vfs: use 'unsigned long'
accesses for dcache name comparison and hashing"). He reports:

 "The test case is essentially

      for (i = 0; i < 1000000; i++)
              mkdir("a$i");

  On xfs on a fio card this goes at about 20k dir/sec with 3.2, and 12k
  dir/sec with 3.10.  This is because we spend waaaaay more time in
  __d_lookup on 3.10 than in 3.2.

  The new hashing function for strings is suboptimal for <
  sizeof(unsigned long) string names (and hell even > sizeof(unsigned
  long) string names that I've tested).  I broke out the old hashing
  function and the new one into a userspace helper to get real numbers
  and this is what I'm getting:

      Old hash table had 1000000 entries, 0 dupes, 0 max dupes
      New hash table had 12628 entries, 987372 dupes, 900 max dupes
      We had 11400 buckets with a p50 of 30 dupes, p90 of 240 dupes, p99 of 567 dupes for the new hash

  My test does the hash, and then does the d_hash into a integer pointer
  array the same size as the dentry hash table on my system, and then
  just increments the value at the address we got to see how many
  entries we overlap with.

  As you can see the old hash function ended up with all 1 million
  entries in their own bucket, whereas the new one they are only
  distributed among ~12.5k buckets, which is why we're using so much
  more CPU in __d_lookup".

The reason for this hash regression is two-fold:

 - On 64-bit architectures the down-mixing of the original 64-bit
   word-at-a-time hash into the final 32-bit hash value is very
   simplistic and suboptimal, and just adds the two 32-bit parts
   together.

   In particular, because there is no bit shuffling and the mixing
   boundary is also a byte boundary, similar character patterns in the
   low and high word easily end up just canceling each other out.

 - the old byte-at-a-time hash mixed each byte into the final hash as it
   hashed the path component name, resulting in the low bits of the hash
   generally being a good source of hash data.  That is not true for the
   word-at-a-time case, and the hash data is distributed among all the
   bits.

The fix is the same in both cases: do a better job of mixing the bits up
and using as much of the hash data as possible.  We already have the
"hash_32|64()" functions to do that.
Reported-by: Josef Bacik <jbacik@fb.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Chris Mason <clm@fb.com>
Cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

The hash_64() function historically does the multiply by the
GOLDEN_RATIO_PRIME_64 number with explicit shifts and adds, because
unlike the 32-bit case, gcc seems unable to turn the constant multiply
into the more appropriate shift and adds when required.

However, that means that we generate those shifts and adds even when the
architecture has a fast multiplier, and could just do it better in
hardware.

Use the now-cleaned-up CONFIG_ARCH_HAS_FAST_MULTIPLIER (together with
"is it a 64-bit architecture") to decide whether to use an integer
multiply or the explicit sequence of shift/add instructions.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

It used to be an ad-hoc hack defined by the x86 version of
<asm/bitops.h> that enabled a couple of library routines to know whether
an integer multiply is faster than repeated shifts and additions.

This just makes it use the real Kconfig system instead, and makes x86
(which was the only architecture that did this) select the option.

NOTE! Even for x86, this really is kind of wrong.  If we cared, we would
probably not enable this for builds optimized for netburst (P4), where
shifts-and-adds are generally faster than multiplies.  This patch does
*not* change that kind of logic, though, it is purely a syntactic change
with no code changes.

This was triggered by the fact that we have other places that really
want to know "do I want to expand multiples by constants by hand or
not", particularly the hash generation code.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Pull device mapper fix from Mike Snitzer:
 "Fix a race in the DM cache target that caused dirty blocks to be
  marked as clean.  This could cause no writeback to occur or spurious
  dirty block counts"

* tag 'dm-3.17-fix2' of git://git.kernel.org/pub/scm/linux/kernel/git/device-mapper/linux-dm:
  dm cache: fix race causing dirty blocks to be marked as clean

Pull block fixes from Jens Axboe:
 "A small collection of fixes for the current rc series.  This contains:

   - Two small blk-mq patches from Rob Elliott, cleaning up error case
     at init time.

   - A fix from Ming Lei, fixing SG merging for blk-mq where
     QUEUE_FLAG_SG_NO_MERGE is the default.

   - A dev_t minor lifetime fix from Keith, fixing an issue where a
     minor might be reused before all references to it were gone.

   - Fix from Alan Stern where an unbalanced queue bypass caused SCSI
     some headaches when it does a series of add/del on devices without
     fully registrering the queue.

   - A fix from me for improving the scaling of tag depth in blk-mq if
     we are short on memory"

* 'for-linus' of git://git.kernel.dk/linux-block:
  blk-mq: scale depth and rq map appropriate if low on memory
  Block: fix unbalanced bypass-disable in blk_register_queue
  block: Fix dev_t minor allocation lifetime
  blk-mq: cleanup after blk_mq_init_rq_map failures
  blk-mq: pass along blk_mq_alloc_tag_set return values
  blk-merge: fix blk_recount_segments

Merge tag 'stable/for-linus-3.17-b-rc4-arm-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/xen/tip

Pull Xen ARM bugfix from Stefano Stabellini:
 "The patches fix the "xen_add_mach_to_phys_entry: cannot add" bug that
  has been affecting xen on arm and arm64 guests since 3.16.  They
  require a few hypervisor side changes that just went in xen-unstable.

  A couple of days ago David sent out a pull request with a few other
  Xen fixes (it is already in master).  Sorry we didn't synchronized
  better among us"

* tag 'stable/for-linus-3.17-b-rc4-arm-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/xen/tip:
  xen/arm: remove mach_to_phys rbtree
  xen/arm: reimplement xen_dma_unmap_page & friends
  xen/arm: introduce XENFEAT_grant_map_identity

Locks the k_itimer's it_lock member when handling the alarm timer's
expiry callback.

The regular posix timers defined in posix-timers.c have this lock held
during timout processing because their callbacks are routed through
posix_timer_fn().  The alarm timers follow a different path, so they
ought to grab the lock somewhere else.

Cc: stable@vger.kernel.org
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Richard Cochran <richardcochran@gmail.com>
Cc: Prarit Bhargava <prarit@redhat.com>
Cc: Sharvil Nanavati <sharvil@google.com>
Signed-off-by: Richard Larocque <rlarocque@google.com>
Signed-off-by: John Stultz <john.stultz@linaro.org>

Avoids sending a signal to alarm timers created with sigev_notify set to
SIGEV_NONE by checking for that special case in the timeout callback.

The regular posix timers avoid sending signals to SIGEV_NONE timers by
not scheduling any callbacks for them in the first place.  Although it
would be possible to do something similar for alarm timers, it's simpler
to handle this as a special case in the timeout.

Prior to this patch, the alarm timer would ignore the sigev_notify value
and try to deliver signals to the process anyway.  Even worse, the
sanity check for the value of sigev_signo is skipped when SIGEV_NONE was
specified, so the signal number could be bogus.  If sigev_signo was an
unitialized value (as it often would be if SIGEV_NONE is used), then
it's hard to predict which signal will be sent.

Cc: stable@vger.kernel.org
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Richard Cochran <richardcochran@gmail.com>
Cc: Prarit Bhargava <prarit@redhat.com>
Cc: Sharvil Nanavati <sharvil@google.com>
Signed-off-by: Richard Larocque <rlarocque@google.com>
Signed-off-by: John Stultz <john.stultz@linaro.org>

Returns the time remaining for an alarm timer, rather than the time at
which it is scheduled to expire.  If the timer has already expired or it
is not currently scheduled, the it_value's members are set to zero.

This new behavior matches that of the other posix-timers and the POSIX
specifications.

This is a change in user-visible behavior, and may break existing
applications.  Hopefully, few users rely on the old incorrect behavior.

Cc: stable@vger.kernel.org
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Richard Cochran <richardcochran@gmail.com>
Cc: Prarit Bhargava <prarit@redhat.com>
Cc: Sharvil Nanavati <sharvil@google.com>
Signed-off-by: Richard Larocque <rlarocque@google.com>
[jstultz: minor style tweak]
Signed-off-by: John Stultz <john.stultz@linaro.org>

timeval_to_jiffies tried to round a timeval up to an integral number
of jiffies, but the logic for doing so was incorrect: intervals
corresponding to exactly N jiffies would become N+1. This manifested
itself particularly repeatedly stopping/starting an itimer:

setitimer(ITIMER_PROF, &val, NULL);
setitimer(ITIMER_PROF, NULL, &val);

would add a full tick to val, _even if it was exactly representable in
terms of jiffies_ (say, the result of a previous rounding.)  Doing
this repeatedly would cause unbounded growth in val.  So fix the math.

Here's what was wrong with the conversion: we essentially computed
(eliding seconds)

jiffies = usec  * (NSEC_PER_USEC/TICK_NSEC)

by using scaling arithmetic, which took the best approximation of
NSEC_PER_USEC/TICK_NSEC with denominator of 2^USEC_JIFFIE_SC =
x/(2^USEC_JIFFIE_SC), and computed:

jiffies = (usec * x) >> USEC_JIFFIE_SC

and rounded this calculation up in the intermediate form (since we
can't necessarily exactly represent TICK_NSEC in usec.) But the
scaling arithmetic is a (very slight) *over*approximation of the true
value; that is, instead of dividing by (1 usec/ 1 jiffie), we
effectively divided by (1 usec/1 jiffie)-epsilon (rounding
down). This would normally be fine, but we want to round timeouts up,
and we did so by adding 2^USEC_JIFFIE_SC - 1 before the shift; this
would be fine if our division was exact, but dividing this by the
slightly smaller factor was equivalent to adding just _over_ 1 to the
final result (instead of just _under_ 1, as desired.)

In particular, with HZ=1000, we consistently computed that 10000 usec
was 11 jiffies; the same was true for any exact multiple of
TICK_NSEC.

We could possibly still round in the intermediate form, adding
something less than 2^USEC_JIFFIE_SC - 1, but easier still is to
convert usec->nsec, round in nanoseconds, and then convert using
time*spec*_to_jiffies.  This adds one constant multiplication, and is
not observably slower in microbenchmarks on recent x86 hardware.

Tested: the following program:

int main() {
  struct itimerval zero = {{0, 0}, {0, 0}};
  /* Initially set to 10 ms. */
  struct itimerval initial = zero;
  initial.it_interval.tv_usec = 10000;
  setitimer(ITIMER_PROF, &initial, NULL);
  /* Save and restore several times. */
  for (size_t i = 0; i < 10; ++i) {
    struct itimerval prev;
    setitimer(ITIMER_PROF, &zero, &prev);
    /* on old kernels, this goes up by TICK_USEC every iteration */
    printf("previous value: %ld %ld %ld %ld\n",
           prev.it_interval.tv_sec, prev.it_interval.tv_usec,
           prev.it_value.tv_sec, prev.it_value.tv_usec);
    setitimer(ITIMER_PROF, &prev, NULL);
  }
    return 0;
}

Cc: stable@vger.kernel.org
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Paul Turner <pjt@google.com>
Cc: Richard Cochran <richardcochran@gmail.com>
Cc: Prarit Bhargava <prarit@redhat.com>
Reviewed-by: Paul Turner <pjt@google.com>
Reported-by: Aaron Jacobs <jacobsa@google.com>
Signed-off-by: Andrew Hunter <ahh@google.com>
[jstultz: Tweaked to apply to 3.17-rc]
Signed-off-by: John Stultz <john.stultz@linaro.org>

futex_wait_requeue_pi() calls futex_wait_setup(). If
futex_wait_setup() succeeds it returns with hb->lock held and
preemption disabled. Now the sanity check after this does:

        if (match_futex(&q.key, &key2)) {
	   	ret = -EINVAL;
		goto out_put_keys;
	}

which releases the keys but does not release hb->lock.

So we happily return to user space with hb->lock held and therefor
preemption disabled.

Unlock hb->lock before taking the exit route.
Reported-by: Dave "Trinity" Jones <davej@redhat.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Darren Hart <dvhart@linux.intel.com>
Reviewed-by: Davidlohr Bueso <dave@stgolabs.net>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: stable@vger.kernel.org
Link: http://lkml.kernel.org/r/alpine.DEB.2.10.1409112318500.4178@nanosSigned-off-by: Thomas Gleixner <tglx@linutronix.de>