GitHub/LineageOS/android_kernel_motorola_exynos9610.git
9 years agoMerge branches 'doc.2015.10.06a', 'percpu-rwsem.2015.10.06a' and 'torture.2015.10...
Paul E. McKenney [Wed, 7 Oct 2015 23:06:25 +0000 (16:06 -0700)]
Merge branches 'doc.2015.10.06a', 'percpu-rwsem.2015.10.06a' and 'torture.2015.10.06a' into HEAD

doc.2015.10.06a:  Documentation updates.
percpu-rwsem.2015.10.06a:  Optimization of per-CPU reader-writer semaphores.
torture.2015.10.06a:  Torture-test updates.

9 years agoMerge branches 'fixes.2015.10.06a' and 'exp.2015.10.07a' into HEAD
Paul E. McKenney [Wed, 7 Oct 2015 23:05:21 +0000 (16:05 -0700)]
Merge branches 'fixes.2015.10.06a' and 'exp.2015.10.07a' into HEAD

exp.2015.10.07a:  Reduce OS jitter of RCU-sched expedited grace periods.
fixes.2015.10.06a:  Miscellaneous fixes.

9 years agorcu: Better hotplug handling for synchronize_sched_expedited()
Paul E. McKenney [Thu, 3 Sep 2015 07:45:02 +0000 (00:45 -0700)]
rcu: Better hotplug handling for synchronize_sched_expedited()

Earlier versions of synchronize_sched_expedited() can prematurely end
grace periods due to the fact that a CPU marked as cpu_is_offline()
can still be using RCU read-side critical sections during the time that
CPU makes its last pass through the scheduler and into the idle loop
and during the time that a given CPU is in the process of coming online.
This commit therefore eliminates this window by adding additional
interaction with the CPU-hotplug operations.

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
9 years agorcu: Enable stall warnings for synchronize_rcu_expedited()
Paul E. McKenney [Tue, 18 Aug 2015 19:17:29 +0000 (12:17 -0700)]
rcu: Enable stall warnings for synchronize_rcu_expedited()

This commit redirects synchronize_rcu_expedited()'s wait to
synchronize_sched_expedited_wait(), thus enabling RCU CPU
stall warnings.

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
9 years agorcu: Add tasks to expedited stall-warning messages
Paul E. McKenney [Tue, 18 Aug 2015 18:25:48 +0000 (11:25 -0700)]
rcu: Add tasks to expedited stall-warning messages

This commit adds task-print ability to the expedited RCU CPU stall
warning messages in preparation for adding stall warnings to
synchornize_rcu_expedited().

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
9 years agorcu: Add online/offline info to expedited stall warning message
Paul E. McKenney [Tue, 18 Aug 2015 17:20:43 +0000 (10:20 -0700)]
rcu: Add online/offline info to expedited stall warning message

This commit makes the RCU CPU stall warning message print online/offline
indications immediately after the CPU number.  A "O" indicates global
offline, a "." global online, and a "o" indicates RCU believes that the
CPU is offline for the current grace period and "." otherwise, and an
"N" indicates that RCU believes that the CPU will be offline for the
next grace period, and "." otherwise, all right after the CPU number.
So for CPU 10, you would normally see "10-...:" indicating that everything
believes that the CPU is online.

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
9 years agorcu: Consolidate expedited CPU selection
Paul E. McKenney [Sun, 16 Aug 2015 02:00:31 +0000 (19:00 -0700)]
rcu: Consolidate expedited CPU selection

Now that sync_sched_exp_select_cpus() and sync_rcu_exp_select_cpus()
are identical aside from the the argument to smp_call_function_single(),
this commit consolidates them with a functional argument.

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
9 years agorcu: Prepare for consolidating expedited CPU selection
Paul E. McKenney [Sun, 16 Aug 2015 00:20:58 +0000 (17:20 -0700)]
rcu: Prepare for consolidating expedited CPU selection

This commit brings sync_sched_exp_select_cpus() into alignment with
sync_rcu_exp_select_cpus(), as a first step towards consolidating them
into one function.

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
9 years agocpu: Remove try_get_online_cpus()
Paul E. McKenney [Fri, 7 Aug 2015 22:14:30 +0000 (15:14 -0700)]
cpu: Remove try_get_online_cpus()

Now that synchronize_sched_expedited() no longer uses it, there are
no users of try_get_online_cpus() in mainline.  This commit therefore
removes it.

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
9 years agorcu: Stop excluding CPU hotplug in synchronize_sched_expedited()
Paul E. McKenney [Fri, 7 Aug 2015 19:03:45 +0000 (12:03 -0700)]
rcu: Stop excluding CPU hotplug in synchronize_sched_expedited()

Now that synchronize_sched_expedited() uses IPIs, a hook in
rcu_sched_qs(), and the ->expmask field in the rcu_node combining
tree, it is no longer necessary to exclude CPU hotplug.  Any
races with CPU hotplug will be detected when attempting to send
the IPI.  This commit therefore removes the code excluding
CPU hotplug operations.

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
9 years agorcu: Stop silencing lockdep false positive for expedited grace periods
Paul E. McKenney [Fri, 7 Aug 2015 03:43:02 +0000 (20:43 -0700)]
rcu: Stop silencing lockdep false positive for expedited grace periods

This reverts commit af859beaaba4 (rcu: Silence lockdep false positive
for expedited grace periods).  Because synchronize_rcu_expedited()
no longer invokes synchronize_sched_expedited(), ->exp_funnel_mutex
acquisition is no longer nested, so the false positive no longer happens.
This commit therefore removes the extra lockdep data structures, as they
are no longer needed.

9 years agorcu: Switch synchronize_sched_expedited() to IPI
Paul E. McKenney [Thu, 6 Aug 2015 23:50:39 +0000 (16:50 -0700)]
rcu: Switch synchronize_sched_expedited() to IPI

This commit switches synchronize_sched_expedited() from stop_one_cpu_nowait()
to smp_call_function_single(), thus moving from an IPI and a pair of
context switches to an IPI and a single pass through the scheduler.
Of course, if the scheduler actually does decide to switch to a different
task, there will still be a pair of context switches, but there would
likely have been a pair of context switches anyway, just a bit later.

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
9 years agolocktorture: Fix module unwind when bad torture_type specified
Paul E. McKenney [Mon, 31 Aug 2015 03:01:48 +0000 (20:01 -0700)]
locktorture: Fix module unwind when bad torture_type specified

The locktorture module has a list of torture types, and specifying
a type not on this list is supposed to cleanly fail the module load.
Unfortunately, the "fail" happens without the "cleanly".  This commit
therefore adds the needed clean-up after an incorrect torture_type.

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
9 years agotorture: Forgive non-plural arguments
Paul E. McKenney [Wed, 26 Aug 2015 00:12:09 +0000 (17:12 -0700)]
torture: Forgive non-plural arguments

This commit allows --bootarg instead of --bootargs, --config instead of
--configs, and --qemu-arg instead of --qemu-args.  For those cases where
a native English speaker might auto-correct the argument to be incorrect.

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
9 years agorcutorture: Fix unused-function warning for torturing_tasks()
Paul E. McKenney [Mon, 24 Aug 2015 20:13:51 +0000 (13:13 -0700)]
rcutorture: Fix unused-function warning for torturing_tasks()

The torturing_tasks() function is used only in kernels built with
CONFIG_PROVE_RCU=y, so the second definition can result in unused-function
compiler warnings.  This commit adds __maybe_unused to suppress these
warnings.

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
9 years agorcutorture: Fix module unwind when bad torture_type specified
Paul E. McKenney [Mon, 24 Aug 2015 18:37:58 +0000 (11:37 -0700)]
rcutorture: Fix module unwind when bad torture_type specified

The rcutorture module has a list of torture types, and specifying a
type not on this list is supposed to cleanly fail the module load.
Unfortunately, the "fail" happens without the "cleanly".  This commit
therefore adds the needed clean-up after an incorrect torture_type.

Reported-by: David Miller <davem@davemloft.net>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Acked-by: David Miller <davem@davemloft.net>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
9 years agorcu_sync: Cleanup the CONFIG_PROVE_RCU checks
Oleg Nesterov [Fri, 11 Sep 2015 15:59:18 +0000 (17:59 +0200)]
rcu_sync: Cleanup the CONFIG_PROVE_RCU checks

1. Rename __rcu_sync_is_idle() to rcu_sync_lockdep_assert() and
   change it to use rcu_lockdep_assert().

2. Change rcu_sync_is_idle() to return rsp->gp_state == GP_IDLE
   unconditonally, this way we can remove the same check from
   rcu_sync_lockdep_assert() and clearly isolate the debugging
   code.

Note: rcu_sync_enter()->wait_event(gp_state == GP_PASSED) needs
another CONFIG_PROVE_RCU check, the same as is done in ->sync(); but
this needs some simple preparations in the core RCU code to avoid the
code duplication.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
9 years agolocking/percpu-rwsem: Clean up the lockdep annotations in percpu_down_read()
Oleg Nesterov [Fri, 21 Aug 2015 17:43:03 +0000 (19:43 +0200)]
locking/percpu-rwsem: Clean up the lockdep annotations in percpu_down_read()

Based on Peter Zijlstra's earlier patch.

Change percpu_down_read() to use __down_read(), this way we can
do rwsem_acquire_read() unconditionally at the start to make this
code more symmetric and clean.

Originally-From: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
9 years agolocking/percpu-rwsem: Fix the comments outdated by rcu_sync
Oleg Nesterov [Fri, 21 Aug 2015 17:43:00 +0000 (19:43 +0200)]
locking/percpu-rwsem: Fix the comments outdated by rcu_sync

Update the comments broken by the previous change.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
9 years agolocking/percpu-rwsem: Make use of the rcu_sync infrastructure
Oleg Nesterov [Fri, 21 Aug 2015 17:42:57 +0000 (19:42 +0200)]
locking/percpu-rwsem: Make use of the rcu_sync infrastructure

Currently down_write/up_write calls synchronize_sched_expedited()
twice, which is evil.  Change this code to rely on rcu-sync primitives.
This avoids the _expedited "big hammer", and this can be faster in
the contended case or even in the case when a single thread does
down_write/up_write in a loop.

Of course, a single down_write() will take more time, but otoh it
will be much more friendly to the whole system.

To simplify the review this patch doesn't update the comments, fixed
by the next change.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
9 years agolocking/percpu-rwsem: Make percpu_free_rwsem() after kzalloc() safe
Oleg Nesterov [Fri, 21 Aug 2015 17:42:55 +0000 (19:42 +0200)]
locking/percpu-rwsem: Make percpu_free_rwsem() after kzalloc() safe

This is the temporary ugly hack which will be reverted later. We only
need it to ensure that the next patch will not break "change sb_writers
to use percpu_rw_semaphore" patches routed via the VFS tree.

The alloc_super()->destroy_super() error path assumes that it is safe
to call percpu_free_rwsem() after kzalloc() without percpu_init_rwsem(),
so let's not disappoint it.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
9 years agorcu_sync: Introduce rcu_sync_dtor()
Oleg Nesterov [Fri, 21 Aug 2015 17:42:52 +0000 (19:42 +0200)]
rcu_sync: Introduce rcu_sync_dtor()

This commit allows rcu_sync structures to be safely deallocated,
The trick is to add a new ->wait field to the gp_ops array.
This field is a pointer to the rcu_barrier() function corresponding
to the flavor of RCU in question.  This allows a new rcu_sync_dtor()
to wait for any outstanding callbacks before freeing the rcu_sync
structure.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
9 years agorcu_sync: Add CONFIG_PROVE_RCU checks
Oleg Nesterov [Fri, 21 Aug 2015 17:42:50 +0000 (19:42 +0200)]
rcu_sync: Add CONFIG_PROVE_RCU checks

This commit validates that the caller of rcu_sync_is_idle() holds the
corresponding type of RCU read-side lock, but only in kernels built
with CONFIG_PROVE_RCU=y.  This validation is carried out via a new
rcu_sync_ops->held() method that is checked within rcu_sync_is_idle().

Note that although this does add code to the fast path, it only does so
in kernels built with CONFIG_PROVE_RCU=y.

Suggested-by: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
9 years agorcu_sync: Simplify rcu_sync using new rcu_sync_ops structure
Oleg Nesterov [Fri, 21 Aug 2015 17:42:47 +0000 (19:42 +0200)]
rcu_sync: Simplify rcu_sync using new rcu_sync_ops structure

This commit adds the new struct rcu_sync_ops which holds sync/call
methods, and turns the function pointers in rcu_sync_struct into an array
of struct rcu_sync_ops.  This simplifies the "init" helpers by collapsing
a switch statement and explicit multiple definitions into a simple
assignment and a helper macro, respectively.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
9 years agorcu: Create rcu_sync infrastructure
Oleg Nesterov [Fri, 21 Aug 2015 17:42:44 +0000 (19:42 +0200)]
rcu: Create rcu_sync infrastructure

The rcu_sync infrastructure can be thought of as infrastructure to be
used to implement reader-writer primitives having extremely lightweight
readers during times when there are no writers.  The first use is in
the percpu_rwsem used by the VFS subsystem.

This infrastructure is functionally equivalent to

        struct rcu_sync_struct {
                atomic_t counter;
        };

/* Check possibility of fast-path read-side operations. */
        static inline bool rcu_sync_is_idle(struct rcu_sync_struct *rss)
        {
                return atomic_read(&rss->counter) == 0;
        }

/* Tell readers to use slowpaths. */
        static inline void rcu_sync_enter(struct rcu_sync_struct *rss)
        {
                atomic_inc(&rss->counter);
                synchronize_sched();
        }

/* Allow readers to once again use fastpaths. */
        static inline void rcu_sync_exit(struct rcu_sync_struct *rss)
        {
                synchronize_sched();
                atomic_dec(&rss->counter);
        }

The main difference is that it records the state and only calls
synchronize_sched() if required.  At least some of the calls to
synchronize_sched() will be optimized away when rcu_sync_enter() and
rcu_sync_exit() are invoked repeatedly in quick succession.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
9 years agotorture: Consolidate cond_resched_rcu_qs() into stutter_wait()
Paul E. McKenney [Sun, 30 Aug 2015 10:29:58 +0000 (03:29 -0700)]
torture: Consolidate cond_resched_rcu_qs() into stutter_wait()

This commit moves cond_resched_rcu_qs() into stutter_wait(), saving
a line and also avoiding RCU CPU stall warnings from all torture
loops containing a stutter_wait().

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
9 years agolocktorture: Add torture tests for percpu_rwsem
Paul E. McKenney [Sat, 29 Aug 2015 21:46:29 +0000 (14:46 -0700)]
locktorture: Add torture tests for percpu_rwsem

This commit adds percpu_rwsem tests based on the earlier rwsem tests.

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Davidlohr Bueso <dave@stgolabs.net>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
9 years agolocking/percpu-rwsem: Export symbols for locktorture
Paul E. McKenney [Tue, 1 Sep 2015 03:21:59 +0000 (20:21 -0700)]
locking/percpu-rwsem: Export symbols for locktorture

This commit exports percpu_down_read(), percpu_down_write(),
__percpu_init_rwsem(), percpu_up_read(), and percpu_up_write() to allow
locktorture to test them when built as a module.

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
9 years agolocktorture: Support rtmutex torturing
Davidlohr Bueso [Wed, 22 Jul 2015 21:07:27 +0000 (14:07 -0700)]
locktorture: Support rtmutex torturing

Real time mutexes is one of the few general primitives
that we do not have in locktorture. Address this -- a few
considerations:

o To spice things up, enable competing thread(s) to become
rt, such that we can stress different prio boosting paths
in the rtmutex code. Introduce a ->task_boost callback,
only used by rtmutex-torturer. Tasks will boost/deboost
around every 50k (arbitrarily) lock/unlock operations.

o Hold times are similar to what we have for other locks:
only occasionally having longer hold times (per ~200k ops).
So we roughly do two full rt boost+deboosting ops with
short hold times.

Signed-off-by: Davidlohr Bueso <dbueso@suse.de>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
9 years agodocumentation: Add lockless_dereference()
Paul E. McKenney [Thu, 17 Sep 2015 15:18:32 +0000 (08:18 -0700)]
documentation: Add lockless_dereference()

The recently added lockless_dereference() macro is not present in the
Documentation/ directory, so this commit fixes that.

Reported-by: Dmitry Vyukov <dvyukov@google.com>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
9 years agodocumentation: Catch up list of torture_type options
Paul E. McKenney [Mon, 24 Aug 2015 18:46:00 +0000 (11:46 -0700)]
documentation: Catch up list of torture_type options

This commit rids the documentation of long-obsolete torture_type options
such as rcu_sync and adds new ones such as tasks.  Also add verbiage
noting the fact that rcutorture now concurrently tests the asynchrounous,
synchronous, expedited synchronous, and polling grace-period primitives.

Reported-by: David Miller <davem@davemloft.net>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Acked-by: David Miller <davem@davemloft.net>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
9 years agodocumentation: Correct doc to use rcu_dereference_protected
Jason A. Donenfeld [Tue, 11 Aug 2015 12:26:33 +0000 (14:26 +0200)]
documentation: Correct doc to use rcu_dereference_protected

As there is lots of misinformation and outdated information on the
Internet about nearly all topics related to the kernel, I thought it
would be best if I based my RCU code on the guidelines of the examples
in the Documentation/ tree of the latest kernel. One thing that stuck
out when reading the whatisRCU.txt document was, "interesting how we
don't need any function to dereference rcu protected pointers when doing
updates if a lock is held. I wonder how static analyzers will work with
that." Then, a few weeks later, upon discovering sparse's __rcu support,
I ran it over my code, and lo and behold, things weren't done right.
Examining other RCU usages in the kernel reveal consistent usage of
rcu_dereference_protected, passing in lockdep_is_held as the
conditional. So, this patch adds that idiom to the documentation, so
that others ahead of me won't endure the same exercise.

Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
9 years agodocumentation: Call out slow consoles as cause of stall warnings
Paul E. McKenney [Tue, 4 Aug 2015 18:54:04 +0000 (11:54 -0700)]
documentation: Call out slow consoles as cause of stall warnings

The Linux kernel outputs copious text during boot, and a slow serial
console can result in stall warnings, particularly when messages are
printed with interrupts disabled.  This commit adds this to the list
of causes of RCU CPU stall warning messages.

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
9 years agodocumentation: No acquire/release for RCU readers
Paul E. McKenney [Sat, 1 Aug 2015 17:45:26 +0000 (10:45 -0700)]
documentation: No acquire/release for RCU readers

Documentation/memory-barriers.txt calls out RCU as one of the sets
of primitives associated with ACQUIRE and RELEASE.  There really
is an association in that rcu_assign_pointer() includes a RELEASE
operation, but a quick read can convince people that rcu_read_lock() and
rcu_read_unlock() have ACQUIRE and RELEASE semantics, which they do not.

This commit therefore removes RCU from this list in order to avoid
this confusion.

Reported-by: Boqun Feng <boqun.feng@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
9 years agorculist: Use WRITE_ONCE() when deleting from reader-visible list
Paul E. McKenney [Fri, 18 Sep 2015 15:45:22 +0000 (08:45 -0700)]
rculist: Use WRITE_ONCE() when deleting from reader-visible list

The various RCU list-deletion macros (list_del_rcu(),
hlist_del_init_rcu(), hlist_del_rcu(), hlist_bl_del_init_rcu(),
hlist_bl_del_rcu(), hlist_nulls_del_init_rcu(), and hlist_nulls_del_rcu())
do plain stores into the ->next pointer of the preceding list elemment.
Unfortunately, the compiler is within its rights to (for example) use
byte-at-a-time writes to update the pointer, which would fatally confuse
concurrent readers.  This patch therefore adds the needed WRITE_ONCE()
macros.

KernelThreadSanitizer (KTSAN) reported the __hlist_del() issue, which
is a problem when __hlist_del() is invoked by hlist_del_rcu().

Reported-by: Dmitry Vyukov <dvyukov@google.com>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
9 years agorcu: Remove deprecated rcu_lockdep_assert()
Paul E. McKenney [Mon, 14 Sep 2015 19:23:01 +0000 (12:23 -0700)]
rcu: Remove deprecated rcu_lockdep_assert()

The old rcu_lockdep_assert() was retained to ease handling of incoming
patches, but any use will result in deprecated warnings.  However, its
replacement, RCU_LOCKDEP_WARN(), is now upstream.  It is therefore
time to remove rcu_lockdep_assert(), which this commit does.

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
9 years agorculist: Make list_entry_rcu() use lockless_dereference()
Patrick Marlier [Fri, 11 Sep 2015 22:50:35 +0000 (15:50 -0700)]
rculist: Make list_entry_rcu() use lockless_dereference()

The current list_entry_rcu() implementation copies the pointer to a stack
variable, then invokes rcu_dereference_raw() on it.  This results in an
additional store-load pair.  Now, most compilers will emit normal store
and load instructions, which might seem to be of negligible overhead,
but this results in a load-hit-store situation that can cause surprisingly
long pipeline stalls, even on modern microprocessors.  The problem is
that it takes time for the store to get the store buffer updated, which
can delay the subsequent load, which immediately follows.

This commit therefore switches to the lockless_dereference() primitive,
which does not expect the __rcu annotations (that are anyway not present
in the list_head structure) and which, like rcu_dereference_raw(),
does not check for an enclosing RCU read-side critical section.
Most importantly, it does not copy the pointer, thus avoiding the
load-hit-store overhead.

Signed-off-by: Patrick Marlier <patrick.marlier@gmail.com>
[ paulmck: Switched to lockless_dereference() to suppress sparse warnings. ]
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
9 years agorcu: Add rcu_pointer_handoff()
Paul E. McKenney [Thu, 10 Sep 2015 23:29:02 +0000 (16:29 -0700)]
rcu: Add rcu_pointer_handoff()

This commit adds an rcu_pointer_handoff() that is intended to mark
situations where a structure's protection transitions from RCU to some
other mechanism (locking, reference counting, whatever).  These markings
should allow external tools to more easily spot bugs involving leaking
pointers out of RCU read-side critical sections.

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
9 years agorcu: Correct comment for values of ->gp_state field
Paul E. McKenney [Thu, 10 Sep 2015 18:21:28 +0000 (11:21 -0700)]
rcu: Correct comment for values of ->gp_state field

This commit corrects the comment for the values of the ->gp_state field,
which previously incorrectly said that these were for the ->gp_flags
field.

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
9 years agorcu: Finish folding ->fqs_state into ->gp_state
Petr Mladek [Wed, 9 Sep 2015 19:09:49 +0000 (12:09 -0700)]
rcu: Finish folding ->fqs_state into ->gp_state

Commit commit 4cdfc175c25c89ee ("rcu: Move quiescent-state forcing
into kthread") started the process of folding the old ->fqs_state into
->gp_state, but did not complete it.  This situation does not cause
any malfunction, but can result in extremely confusing trace output.
This commit completes this task of eliminating ->fqs_state in favor
of ->gp_state.

The old ->fqs_state was also used to decide when to collect dyntick-idle
snapshots.  For this purpose, we add a boolean variable into the kthread,
which is set on the first call to rcu_gp_fqs() for a given grace period
and clear otherwise.

Signed-off-by: Petr Mladek <pmladek@suse.com>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
9 years agorcu: Move preemption disabling out of __srcu_read_lock()
Paul E. McKenney [Tue, 1 Sep 2015 07:42:57 +0000 (00:42 -0700)]
rcu: Move preemption disabling out of __srcu_read_lock()

Currently, __srcu_read_lock() cannot be invoked from restricted
environments because it contains calls to preempt_disable() and
preempt_enable(), both of which can invoke lockdep, which is a bad
idea in some restricted execution modes.  This commit therefore moves
the preempt_disable() and preempt_enable() from __srcu_read_lock()
to srcu_read_lock().  It also inserts the preempt_disable() and
preempt_enable() around the call to __srcu_read_lock() in do_exit().

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
9 years agorcu: Add online/offline info to stall warning message
Paul E. McKenney [Sat, 15 Aug 2015 21:49:50 +0000 (14:49 -0700)]
rcu: Add online/offline info to stall warning message

This commit makes the RCU CPU stall warning message print online/offline
indications immediately after a hyphen following the CPU number.  A "O"
indicates that the global CPU-hotplug system believes that the CPU is
online, a "o" that RCU perceived the CPU to be online at the beginning
of the current expedited grace period, and an "N" that RCU currently
believes that it will perceive the CPU as being online at the beginning
of the next expedited grace period, with "." otherwise for all three
indications.  So for CPU 10, you would normally see "10-OoN:" indicating
that everything believes that the CPU is online.

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
9 years agorcu: Eliminate panic when silly boot-time fanout specified
Paul E. McKenney [Fri, 31 Jul 2015 15:28:35 +0000 (08:28 -0700)]
rcu: Eliminate panic when silly boot-time fanout specified

This commit loosens rcutree.rcu_fanout_leaf range checks
and replaces a panic() with a fallback to compile-time values.
This fallback is accompanied by a WARN_ON(), and both occur when the
rcutree.rcu_fanout_leaf value is too small to accommodate the number of
CPUs.  For example, given the current four-level limit for the rcu_node
tree, a system with more than 16 CPUs built with CONFIG_FANOUT=2 must
have rcutree.rcu_fanout_leaf larger than 2.

Reported-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
9 years agorcu: Don't disable preemption for Tiny and Tree RCU readers
Boqun Feng [Thu, 30 Jul 2015 23:55:38 +0000 (16:55 -0700)]
rcu: Don't disable preemption for Tiny and Tree RCU readers

Because preempt_disable() maps to barrier() for non-debug builds,
it forces the compiler to spill and reload registers.  Because Tree
RCU and Tiny RCU now only appear in CONFIG_PREEMPT=n builds, these
barrier() instances generate needless extra code for each instance of
rcu_read_lock() and rcu_read_unlock().  This extra code slows down Tree
RCU and bloats Tiny RCU.

This commit therefore removes the preempt_disable() and preempt_enable()
from the non-preemptible implementations of __rcu_read_lock() and
__rcu_read_unlock(), respectively.  However, for debug purposes,
preempt_disable() and preempt_enable() are still invoked if
CONFIG_PREEMPT_COUNT=y, because this allows detection of sleeping inside
atomic sections in non-preemptible kernels.

However, Tiny and Tree RCU operates by coalescing all RCU read-side
critical sections on a given CPU that lie between successive quiescent
states.  It is therefore necessary to compensate for removing barriers
from __rcu_read_lock() and __rcu_read_unlock() by adding them to a
couple of the RCU functions invoked during quiescent states, namely to
rcu_all_qs() and rcu_note_context_switch().  However, note that the latter
is more paranoia than necessity, at least until link-time optimizations
become more aggressive.

This is based on an earlier patch by Paul E. McKenney, fixing
a bug encountered in kernels built with CONFIG_PREEMPT=n and
CONFIG_PREEMPT_COUNT=y.

Signed-off-by: Boqun Feng <boqun.feng@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
9 years agorcu: Use call_rcu_func_t to replace explicit type equivalents
Boqun Feng [Wed, 29 Jul 2015 05:29:39 +0000 (13:29 +0800)]
rcu: Use call_rcu_func_t to replace explicit type equivalents

We have had the call_rcu_func_t typedef for a quite awhile, but we still
use explicit function pointer types in some places.  These types can
confuse cscope and can be hard to read.  This patch therefore replaces
these types with the call_rcu_func_t typedef.

Signed-off-by: Boqun Feng <boqun.feng@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
9 years agorcu: Use rcu_callback_t in call_rcu*() and friends
Boqun Feng [Wed, 29 Jul 2015 05:29:38 +0000 (13:29 +0800)]
rcu: Use rcu_callback_t in call_rcu*() and friends

As we now have rcu_callback_t typedefs as the type of rcu callbacks, we
should use it in call_rcu*() and friends as the type of parameters. This
could save us a few lines of code and make it clear which function
requires an rcu callbacks rather than other callbacks as its argument.

Besides, this can also help cscope to generate a better database for
code reading.

Signed-off-by: Boqun Feng <boqun.feng@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
9 years agosched: Export sched_setscheduler_nocheck
Davidlohr Bueso [Wed, 2 Sep 2015 08:28:44 +0000 (01:28 -0700)]
sched: Export sched_setscheduler_nocheck

The new locktorture rtmutex_lock tests exercise priority boosting, which
means that they need to set some tasks to real-time priority.  To do this,
they use sched_setscheduler_nocheck().  However, this is not exported to
modules, which results in the following error when building locktorture
as a module:

ERROR: "sched_setscheduler_nocheck" [kernel/locking/locktorture.ko] undefined!

This commit therefore adds an EXPORT_SYMBOL_GPL() to allow this function
to be invoked from locktorture when built as a module.

Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Davidlohr Bueso <dave@stgolabs.net>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Acked-by: Ingo Molnar <mingo@kernel.org>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
9 years agorcu: Make ->cpu_no_qs be a union for aggregate OR
Paul E. McKenney [Thu, 6 Aug 2015 22:16:57 +0000 (15:16 -0700)]
rcu: Make ->cpu_no_qs be a union for aggregate OR

This commit converts the rcu_data structure's ->cpu_no_qs field
to a union.  The bytewise side of this union allows individual access
to indications as to whether this CPU needs to find a quiescent state
for a normal (.norm) and/or expedited (.exp) grace period.  The setwise
side of the union allows testing whether or not a quiescent state is
needed at all, for either type of grace period.

For now, only .norm is used.  A later commit will introduce the expedited
usage.

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
9 years agorcu: Invert passed_quiesce and rename to cpu_no_qs
Paul E. McKenney [Thu, 6 Aug 2015 21:17:29 +0000 (14:17 -0700)]
rcu: Invert passed_quiesce and rename to cpu_no_qs

This commit inverts the sense of the rcu_data structure's ->passed_quiesce
field and renames it to ->cpu_no_qs.  This will allow a later commit to
use an "aggregate OR" operation to test expedited as well as normal grace
periods without added overhead.

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
9 years agorcu: Rename qs_pending to core_needs_qs
Paul E. McKenney [Thu, 6 Aug 2015 18:31:51 +0000 (11:31 -0700)]
rcu: Rename qs_pending to core_needs_qs

An upcoming commit needs to invert the sense of the ->passed_quiesce
rcu_data structure field, so this commit is taking this opportunity
to clarify things a bit by renaming ->qs_pending to ->core_needs_qs.

So if !rdp->core_needs_qs, then this CPU need not concern itself with
quiescent states, in particular, it need not acquire its leaf rcu_node
structure's ->lock to check.  Otherwise, it needs to report the next
quiescent state.

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
9 years agorcu: Move synchronize_sched_expedited() to combining tree
Paul E. McKenney [Wed, 5 Aug 2015 23:03:54 +0000 (16:03 -0700)]
rcu: Move synchronize_sched_expedited() to combining tree

Currently, synchronize_sched_expedited() uses a single global counter
to track the number of remaining context switches that the current
expedited grace period must wait on.  This is problematic on large
systems, where the resulting memory contention can be pathological.
This commit therefore makes synchronize_sched_expedited() instead use
the combining tree in the same manner as synchronize_rcu_expedited(),
keeping memory contention down to a dull roar.

This commit creates a temporary function sync_sched_exp_select_cpus()
that is very similar to sync_rcu_exp_select_cpus().  A later commit
will consolidate these two functions, which becomes possible when
synchronize_sched_expedited() switches from stop_one_cpu_nowait() to
smp_call_function_single().

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
9 years agorcu: Use single-stage IPI algorithm for RCU expedited grace period
Paul E. McKenney [Sun, 2 Aug 2015 20:53:17 +0000 (13:53 -0700)]
rcu: Use single-stage IPI algorithm for RCU expedited grace period

The current preemptible-RCU expedited grace-period algorithm invokes
synchronize_sched_expedited() to enqueue all tasks currently running
in a preemptible-RCU read-side critical section, then waits for all the
->blkd_tasks lists to drain.  This works, but results in both an IPI and
a double context switch even on CPUs that do not happen to be running
in a preemptible RCU read-side critical section.

This commit implements a new algorithm that causes less OS jitter.
This new algorithm IPIs all online CPUs that are not idle (from an
RCU perspective), but refrains from self-IPIs.  If a CPU receiving
this IPI is not in a preemptible RCU read-side critical section (or
is just now exiting one), it pushes quiescence up the rcu_node tree,
otherwise, it sets a flag that will be handled by the upcoming outermost
rcu_read_unlock(), which will then push quiescence up the tree.

The expedited grace period must of course wait on any pre-existing blocked
readers, and newly blocked readers must be queued carefully based on
the state of both the normal and the expedited grace periods.  This
new queueing approach also avoids the need to update boost state,
courtesy of the fact that blocked tasks are no longer ever migrated to
the root rcu_node structure.

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
9 years agorcu: Consolidate tree setup for synchronize_rcu_expedited()
Paul E. McKenney [Fri, 31 Jul 2015 23:04:45 +0000 (16:04 -0700)]
rcu: Consolidate tree setup for synchronize_rcu_expedited()

This commit replaces sync_rcu_preempt_exp_init1(() and
sync_rcu_preempt_exp_init2() with sync_exp_reset_tree_hotplug()
and sync_exp_reset_tree(), which will also be used by
synchronize_sched_expedited(), and sync_rcu_exp_select_nodes(), which
contains code specific to synchronize_rcu_expedited().

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
9 years agorcu: Move rcu_report_exp_rnp() to allow consolidation
Paul E. McKenney [Fri, 31 Jul 2015 20:34:32 +0000 (13:34 -0700)]
rcu: Move rcu_report_exp_rnp() to allow consolidation

This is a nearly pure code-movement commit, moving rcu_report_exp_rnp(),
sync_rcu_preempt_exp_done(), and rcu_preempted_readers_exp() so
that later commits can make synchronize_sched_expedited() use them.
The non-code-movement portion of this commit tags rcu_report_exp_rnp()
as __maybe_unused to avoid build errors when CONFIG_PREEMPT=n.

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
9 years agorcu: Use rsp->expedited_wq instead of sync_rcu_preempt_exp_wq
Paul E. McKenney [Thu, 30 Jul 2015 00:28:11 +0000 (17:28 -0700)]
rcu: Use rsp->expedited_wq instead of sync_rcu_preempt_exp_wq

Now that there is an ->expedited_wq waitqueue in each rcu_state structure,
there is no need for the sync_rcu_preempt_exp_wq global variable.  This
commit therefore substitutes ->expedited_wq for sync_rcu_preempt_exp_wq.
It also initializes ->expedited_wq only once at boot instead of at the
start of each expedited grace period.

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
9 years agorcu: Suppress lockdep false positive for rcp->exp_funnel_mutex
Paul E. McKenney [Mon, 21 Sep 2015 04:01:22 +0000 (21:01 -0700)]
rcu: Suppress lockdep false positive for rcp->exp_funnel_mutex

In kernels built with CONFIG_PREEMPT=y, synchronize_rcu_expedited()
invokes synchronize_sched_expedited() while holding RCU-preempt's
root rcu_node structure's ->exp_funnel_mutex, which is acquired after
the rcu_data structure's ->exp_funnel_mutex.  The first thing that
synchronize_sched_expedited() will do is acquire RCU-sched's rcu_data
structure's ->exp_funnel_mutex.   There is no danger of an actual deadlock
because the locking order is always from RCU-preempt's expedited mutexes
to those of RCU-sched.  Unfortunately, lockdep considers both rcu_data
structures' ->exp_funnel_mutex to be in the same lock class and therefore
reports a deadlock cycle.

This commit silences this false positive by placing RCU-sched's rcu_data
structures' ->exp_funnel_mutex locks into their own lock class.

Reported-by: Sasha Levin <sasha.levin@oracle.com>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
9 years agorcu: Change _wait_rcu_gp() to work around GCC bug 67055
Oleg Nesterov [Tue, 25 Aug 2015 18:45:18 +0000 (20:45 +0200)]
rcu: Change _wait_rcu_gp() to work around GCC bug 67055

Code like this in inline functions confuses some recent versions of gcc:

const int n = const-expr;
whatever_t array[n];

For more details, see:

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=67055#c13

This compiler bug results in the following failure after 114b7fd4b (rcu:
Create rcu_sync infrastructure):

In file included from include/linux/rcupdate.h:429:0,
  from include/linux/rcu_sync.h:5,
  from kernel/rcu/sync.c:1:
include/linux/rcutiny.h: In function 'rcu_barrier_sched':
include/linux/rcutiny.h:55:20: internal compiler error: Segmentation fault
  static inline void rcu_barrier_sched(void)

This commit therefore eliminates the constant local variable in favor of
direct use of the expression.

Reported-and-tested-by: Mark Salter <msalter@redhat.com>
Reported-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
9 years agoLinux 4.3-rc2
Linus Torvalds [Sun, 20 Sep 2015 21:32:34 +0000 (14:32 -0700)]
Linux 4.3-rc2

9 years agoMerge branch 'fixes' of git://ftp.arm.linux.org.uk/~rmk/linux-arm
Linus Torvalds [Sun, 20 Sep 2015 04:05:02 +0000 (21:05 -0700)]
Merge branch 'fixes' of git://ftp.arm.linux.org.uk/~rmk/linux-arm

Pull ARM fixes from Russell King:
 "Three fixes and a resulting cleanup for -rc2:

   - Andre Przywara reported that he was seeing a warning with the new
     cast inside DMA_ERROR_CODE's definition, and fixed the incorrect
     use.

   - Doug Anderson noticed that kgdb causes a "scheduling while atomic"
     bug.

   - OMAP5 folk noticed that their Thumb-2 compiled X servers crashed
     when enabling support to cover ARMv6 CPUs due to a kernel bug
     leaking some conditional context into the signal handler"

* 'fixes' of git://ftp.arm.linux.org.uk/~rmk/linux-arm:
  ARM: 8425/1: kgdb: Don't try to stop the machine when setting breakpoints
  ARM: 8437/1: dma-mapping: fix build warning with new DMA_ERROR_CODE definition
  ARM: get rid of needless #if in signal handling code
  ARM: fix Thumb2 signal handling when ARMv6 is enabled

9 years agoMerge tag 'linux-kselftest-4.3-rc2' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 20 Sep 2015 03:57:45 +0000 (20:57 -0700)]
Merge tag 'linux-kselftest-4.3-rc2' of git://git./linux/kernel/git/shuah/linux-kselftest

Pull kselftest fixes from Shuah Khan:
 "This update contains 7 fixes for problems ranging from build failurs
  to incorrect error reporting"

* tag 'linux-kselftest-4.3-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/shuah/linux-kselftest:
  selftests: exec: revert to default emit rule
  selftests: change install command to rsync
  selftests: mqueue: simplify the Makefile
  selftests: mqueue: allow extra cflags
  selftests: rename jump label to static_keys
  selftests/seccomp: add support for s390
  seltests/zram: fix syntax error

9 years agoMerge tag 'pm+acpi-4.3-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael...
Linus Torvalds [Sun, 20 Sep 2015 03:41:31 +0000 (20:41 -0700)]
Merge tag 'pm+acpi-4.3-rc2' of git://git./linux/kernel/git/rafael/linux-pm

Pull power management and ACPI updates from Rafael Wysocki:
 "Included are: a somewhat late devfreq update which however is mostly
  fixes and cleanups with one new thing only (the PPMUv2 support on
  Exynos5433), an ACPI cpufreq driver fixup and two ACPI core cleanups
  related to preprocessor directives.

  Specifics:

   - Fix a memory allocation size in the devfreq core (Xiaolong Ye).

   - Fix a mistake in the exynos-ppmu DT binding (Javier Martinez
     Canillas).

   - Add support for PPMUv2 ((Platform Performance Monitoring Unit
     version 2.0) on the Exynos5433 SoCs (Chanwoo Choi).

   - Fix a type casting bug in the Exynos PPMU code (MyungJoo Ham).

   - Assorted devfreq code cleanups and optimizations (Javi Merino,
     MyungJoo Ham, Viresh Kumar).

   - Fix up the ACPI cpufreq driver to use a more lightweight way to get
     to its private data in the ->get() callback (Rafael J Wysocki).

   - Fix a CONFIG_ prefix bug in one of the ACPI drivers and make the
     ACPI subsystem use IS_ENABLED() instead of #ifdefs in function
     bodies (Sudeep Holla)"

* tag 'pm+acpi-4.3-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm:
  cpufreq: acpi-cpufreq: Use cpufreq_cpu_get_raw() in ->get()
  ACPI: Eliminate CONFIG_.*{, _MODULE} #ifdef in favor of IS_ENABLED()
  ACPI: int340x_thermal: add missing CONFIG_ prefix
  PM / devfreq: Fix incorrect type issue.
  PM / devfreq: tegra: Update governor to use devfreq_update_stats()
  PM / devfreq: comments for get_dev_status usage updated
  PM / devfreq: drop comment about thermal setting max_freq
  PM / devfreq: cache the last call to get_dev_status()
  PM / devfreq: Drop unlikely before IS_ERR(_OR_NULL)
  PM / devfreq: exynos-ppmu: bit-wise operation bugfix.
  PM / devfreq: exynos-ppmu: Update documentation to support PPMUv2
  PM / devfreq: exynos-ppmu: Add the support of PPMUv2 for Exynos5433
  PM / devfreq: event: Remove incorrect property in exynos-ppmu DT binding

9 years agoMerge tag 'clk-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sun, 20 Sep 2015 03:17:40 +0000 (20:17 -0700)]
Merge tag 'clk-fixes-for-linus' of git://git./linux/kernel/git/clk/linux

Pull clk fixes from Stephen Boyd:
 "A few driver fixes for tegra, rockchip, and st SoCs and a two-liner in
  the framework to avoid oops when get_parent ops return out of range
  values on tegra platforms"

* tag 'clk-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/clk/linux:
  drivers: clk: st: Rename st_pll3200c32_407_c0_x into st_pll3200c32_cx_x
  clk: check for invalid parent index of orphans in __clk_init()
  clk: tegra: dfll: Properly protect OPP list
  clk: rockchip: add critical clock for rk3368

9 years agoMerge tag 'led-fixes-for-v4.3-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sun, 20 Sep 2015 03:10:30 +0000 (20:10 -0700)]
Merge tag 'led-fixes-for-v4.3-rc2' of git://git./linux/kernel/git/j.anaszewski/linux-leds

Pull LED fixes from Jacek Anaszewski:
 - fix module autoload for six OF platform drivers (aat1290, bcm6328,
   bcm6358, ktd2692, max77693, ns2)
 - aat1290: add missing static modifier
 - ipaq-micro: add missing LEDS_CLASS dependency
 - lp55xx: correct Kconfig dependecy for f/w user helper

* tag 'led-fixes-for-v4.3-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/j.anaszewski/linux-leds:
  leds:lp55xx: Correct Kconfig dependency for f/w user helper
  leds: leds-ipaq-micro: Add LEDS_CLASS dependency
  leds: aat1290: add 'static' modifier to init_mm_current_scale
  leds: leds-ns2: Fix module autoload for OF platform driver
  leds: max77693: Fix module autoload for OF platform driver
  leds: ktd2692: Fix module autoload for OF platform driver
  leds: bcm6358: Fix module autoload for OF platform driver
  leds: bcm6328: Fix module autoload for OF platform driver
  leds: aat1290: Fix module autoload for OF platform driver

9 years agoMerge tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dledford/rdma
Linus Torvalds [Sun, 20 Sep 2015 03:04:11 +0000 (20:04 -0700)]
Merge tag 'for-linus' of git://git./linux/kernel/git/dledford/rdma

Pull rdma fixes from Doug Ledford:
 "The new hfi1 driver in staging/rdma has had a number of fixup patches
  since being added to the tree.  This is the first batch of those fixes"

* tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dledford/rdma:
  IB/hfi: Properly set permissions for user device files
  IB/hfi1: mask vs shift confusion
  IB/hfi1: clean up some defines
  IB/hfi1: info leak in get_ctxt_info()
  IB/hfi1: fix a locking bug
  IB/hfi1: checking for NULL instead of IS_ERR
  IB/hfi1: fix sdma_descq_cnt parameter parsing
  IB/hfi1: fix copy_to/from_user() error handling
  IB/hfi1: fix pstateinfo from returning improperly byteswapped value

9 years agoMerge branch 'libnvdimm-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/nvdim...
Linus Torvalds [Sun, 20 Sep 2015 02:13:03 +0000 (19:13 -0700)]
Merge branch 'libnvdimm-fixes' of git://git./linux/kernel/git/nvdimm/nvdimm

Pull libnvdimm fixes from Dan Williams:

 - a boot regression (since v4.2) fix for some ARM configurations from
   Tyler

 - regression (since v4.1) fixes for mkfs.xfs on a DAX enabled device
   from Jeff.  These are tagged for -stable.

 - a pair of locking fixes from Axel that are hidden from lockdep since
   they involve device_lock().  The "btt" one is tagged for -stable, the
   other only applies to the new "pfn" mechanism in v4.3.

 - a fix for the pmem ->rw_page() path to use wmb_pmem() from Ross.

* 'libnvdimm-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/nvdimm/nvdimm:
  mm: fix type cast in __pfn_to_phys()
  pmem: add proper fencing to pmem_rw_page()
  libnvdimm: pfn_devs: Fix locking in namespace_store
  libnvdimm: btt_devs: Fix locking in namespace_store
  blockdev: don't set S_DAX for misaligned partitions
  dax: fix O_DIRECT I/O to the last block of a blockdev

9 years agoMerge branch 'for-linus' of git://git.kernel.dk/linux-block
Linus Torvalds [Sun, 20 Sep 2015 01:57:09 +0000 (18:57 -0700)]
Merge branch 'for-linus' of git://git.kernel.dk/linux-block

Pull block updates from Jens Axboe:
 "This is a bit bigger than it should be, but I could (did) not want to
  send it off last week due to both wanting extra testing, and expecting
  a fix for the bounce regression as well.  In any case, this contains:

   - Fix for the blk-merge.c compilation warning on gcc 5.x from me.

   - A set of back/front SG gap merge fixes, from me and from Sagi.
     This ensures that we honor SG gapping for integrity payloads as
     well.

   - Two small fixes for null_blk from Matias, fixing a leak and a
     capacity propagation issue.

   - A blkcg fix from Tejun, fixing a NULL dereference.

   - A fast clone optimization from Ming, fixing a performance
     regression since the arbitrarily sized bio's were introduced.

   - Also from Ming, a regression fix for bouncing IOs"

* 'for-linus' of git://git.kernel.dk/linux-block:
  block: fix bounce_end_io
  block: blk-merge: fast-clone bio when splitting rw bios
  block: blkg_destroy_all() should clear q->root_blkg and ->root_rl.blkg
  block: Copy a user iovec if it includes gaps
  block: Refuse adding appending a gapped integrity page to a bio
  block: Refuse request/bio merges with gaps in the integrity payload
  block: Check for gaps on front and back merges
  null_blk: fix wrong capacity when bs is not 512 bytes
  null_blk: fix memory leak on cleanup
  block: fix bogus compiler warnings in blk-merge.c

9 years agofs-writeback: unplug before cond_resched in writeback_sb_inodes
Chris Mason [Fri, 18 Sep 2015 17:35:08 +0000 (13:35 -0400)]
fs-writeback: unplug before cond_resched in writeback_sb_inodes

Commit 505a666ee3fc ("writeback: plug writeback in wb_writeback() and
writeback_inodes_wb()") has us holding a plug during writeback_sb_inodes,
which increases the merge rate when relatively contiguous small files
are written by the filesystem.  It helps both on flash and spindles.

For an fs_mark workload creating 4K files in parallel across 8 drives,
this commit improves performance ~9% more by unplugging before calling
cond_resched().  cond_resched() doesn't trigger an implicit unplug, so
explicitly getting the IO down to the device before scheduling reduces
latencies for anyone waiting on clean pages.

It also cuts down on how often we use kblockd to unplug, which means
less work bouncing from one workqueue to another.

Many more details about how we got here:

  https://lkml.org/lkml/2015/9/11/570

Signed-off-by: Chris Mason <clm@fb.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agomm: fix type cast in __pfn_to_phys()
Tyler Baker [Sat, 19 Sep 2015 07:58:10 +0000 (03:58 -0400)]
mm: fix type cast in __pfn_to_phys()

The various definitions of __pfn_to_phys() have been consolidated to
use a generic macro in include/asm-generic/memory_model.h. This hit
mainline in the form of 012dcef3f058 "mm: move __phys_to_pfn and
__pfn_to_phys to asm/generic/memory_model.h". When the generic macro
was implemented the type cast to phys_addr_t was dropped which caused
boot regressions on ARM platforms with more than 4GB of memory and
LPAE enabled.

It was suggested to use PFN_PHYS() defined in include/linux/pfn.h
as provides the correct logic and avoids further duplication.

Reported-by: kernelci.org bot <bot@kernelci.org>
Suggested-by: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: Tyler Baker <tyler.baker@linaro.org>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
9 years agoMerge branch 'acpi-bus'
Rafael J. Wysocki [Fri, 18 Sep 2015 21:07:46 +0000 (23:07 +0200)]
Merge branch 'acpi-bus'

* acpi-bus:
  ACPI: Eliminate CONFIG_.*{, _MODULE} #ifdef in favor of IS_ENABLED()
  ACPI: int340x_thermal: add missing CONFIG_ prefix

9 years agoMerge branches 'pm-cpufreq' and 'pm-devfreq'
Rafael J. Wysocki [Fri, 18 Sep 2015 21:05:28 +0000 (23:05 +0200)]
Merge branches 'pm-cpufreq' and 'pm-devfreq'

* pm-cpufreq:
  cpufreq: acpi-cpufreq: Use cpufreq_cpu_get_raw() in ->get()

* pm-devfreq:
  PM / devfreq: Fix incorrect type issue.
  PM / devfreq: tegra: Update governor to use devfreq_update_stats()
  PM / devfreq: comments for get_dev_status usage updated
  PM / devfreq: drop comment about thermal setting max_freq
  PM / devfreq: cache the last call to get_dev_status()
  PM / devfreq: Drop unlikely before IS_ERR(_OR_NULL)
  PM / devfreq: exynos-ppmu: bit-wise operation bugfix.
  PM / devfreq: exynos-ppmu: Update documentation to support PPMUv2
  PM / devfreq: exynos-ppmu: Add the support of PPMUv2 for Exynos5433
  PM / devfreq: event: Remove incorrect property in exynos-ppmu DT binding

9 years agoMerge tag 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mst/vhost
Linus Torvalds [Fri, 18 Sep 2015 16:28:20 +0000 (09:28 -0700)]
Merge tag 'for_linus' of git://git./linux/kernel/git/mst/vhost

Pull virtio fixes and cleanups from Michael Tsirkin:
 "This fixes the virtio-test tool, and improves the error handling for
  virtio-ccw"

* tag 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mst/vhost:
  virtio/s390: handle failures of READ_VQ_CONF ccw
  tools/virtio: propagate V=X to kernel build
  vhost: move features to core
  tools/virtio: fix build after 4.2 changes

9 years agoMerge tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm
Linus Torvalds [Fri, 18 Sep 2015 16:23:08 +0000 (09:23 -0700)]
Merge tag 'for-linus' of git://git./virt/kvm/kvm

Pull KVM fixes from Paolo Bonzini:
 "Mostly stable material, a lot of ARM fixes"

* tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm: (22 commits)
  sched: access local runqueue directly in single_task_running
  arm/arm64: KVM: Remove 'config KVM_ARM_MAX_VCPUS'
  arm64: KVM: Remove all traces of the ThumbEE registers
  arm: KVM: Disable virtual timer even if the guest is not using it
  arm64: KVM: Disable virtual timer even if the guest is not using it
  arm/arm64: KVM: vgic: Check for !irqchip_in_kernel() when mapping resources
  KVM: s390: Replace incorrect atomic_or with atomic_andnot
  arm: KVM: Fix incorrect device to IPA mapping
  arm64: KVM: Fix user access for debug registers
  KVM: vmx: fix VPID is 0000H in non-root operation
  KVM: add halt_attempted_poll to VCPU stats
  kvm: fix zero length mmio searching
  kvm: fix double free for fast mmio eventfd
  kvm: factor out core eventfd assign/deassign logic
  kvm: don't try to register to KVM_FAST_MMIO_BUS for non mmio eventfd
  KVM: make the declaration of functions within 80 characters
  KVM: arm64: add workaround for Cortex-A57 erratum #852523
  KVM: fix polling for guest halt continued even if disable it
  arm/arm64: KVM: Fix PSCI affinity info return value for non valid cores
  arm64: KVM: set {v,}TCR_EL2 RES1 bits
  ...

9 years agoIB/hfi: Properly set permissions for user device files
Ira Weiny [Thu, 17 Sep 2015 17:47:49 +0000 (13:47 -0400)]
IB/hfi: Properly set permissions for user device files

Some of the device files are required to be user accessible for PSM while
most should remain accessible only by root.

Add a parameter to hfi1_cdev_init which controls if the user should have access
to this device which places it in a different class with the appropriate
devnode callback.

In addition set the devnode call back for the existing class to be a bit more
explicit for those permissions.

Finally remove the unnecessary null check before class_destroy

Tested-by: Donald Dutile <ddutile@redhat.com>
Signed-off-by: Haralanov, Mitko (mitko.haralanov@intel.com)
Signed-off-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
9 years agoIB/hfi1: mask vs shift confusion
Dan Carpenter [Wed, 16 Sep 2015 16:03:45 +0000 (19:03 +0300)]
IB/hfi1: mask vs shift confusion

We are shifting by the _MASK macros instead of the _SHIFT ones.

Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
9 years agoIB/hfi1: clean up some defines
Dan Carpenter [Wed, 16 Sep 2015 16:02:54 +0000 (19:02 +0300)]
IB/hfi1: clean up some defines

I added spaces around operators so it matches kernel style because
normally "-1ULL" is a number and " - 1" is a subtract operation.  Also
removed some superflous "ULL" types so "1ULL" becomes "1".

Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
9 years agoIB/hfi1: info leak in get_ctxt_info()
Dan Carpenter [Wed, 16 Sep 2015 06:42:25 +0000 (09:42 +0300)]
IB/hfi1: info leak in get_ctxt_info()

The cinfo struct has a hole after the last struct member so we need to
zero it out.  Otherwise we disclose some uninitialized stack data.

Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
9 years agoIB/hfi1: fix a locking bug
Dan Carpenter [Wed, 16 Sep 2015 06:22:51 +0000 (09:22 +0300)]
IB/hfi1: fix a locking bug

mutex_trylock() returns zero on failure, not EBUSY.

Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
9 years agoIB/hfi1: checking for NULL instead of IS_ERR
Dan Carpenter [Wed, 16 Sep 2015 06:22:20 +0000 (09:22 +0300)]
IB/hfi1: checking for NULL instead of IS_ERR

__get_txreq() returns an ERR_PTR() but this checks for NULL so it would
oops on failure.

Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
9 years agoIB/hfi1: fix sdma_descq_cnt parameter parsing
Mike Marciniszyn [Tue, 15 Sep 2015 14:19:27 +0000 (10:19 -0400)]
IB/hfi1: fix sdma_descq_cnt parameter parsing

The boolean tests should have been or-ed.

Reported-by: David Binderman <dcb314@hotmail.com>
Reviewed-by: Jubin John <jubin.john@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
9 years agoIB/hfi1: fix copy_to/from_user() error handling
Dan Carpenter [Tue, 15 Sep 2015 10:35:25 +0000 (13:35 +0300)]
IB/hfi1: fix copy_to/from_user() error handling

copy_to/from_user() returns the number of bytes which we were not able
to copy.  It doesn't return an error code.

Also a couple places had a printk() on error and I removed that because
people can take advantage of it to fill /var/log/messages with spam.

Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Acked-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
9 years agoIB/hfi1: fix pstateinfo from returning improperly byteswapped value
Ira Weiny [Wed, 9 Sep 2015 05:28:21 +0000 (01:28 -0400)]
IB/hfi1: fix pstateinfo from returning improperly byteswapped value

Byteswap link_width_downgrade_*_active values before sending on the wire.  In
addition properly define the Port State Info structure.

Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Reviewed-by: Christian Gomez <christian.gomez@intel.com>
Signed-off-by: Rimmer, Todd <todd.rimmer@intel.com>
Signed-off-by: Ira Weiny <ira.weiny@intel.com>
Acked-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
9 years agoMerge branch 'irq-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Fri, 18 Sep 2015 15:11:42 +0000 (08:11 -0700)]
Merge branch 'irq-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull irq updates from Thomas Gleixner:
 "This is a rather large update post rc1 due to the final steps of
  cleanups and API changes which had to wait for the preparatory patches
  to hit your tree.

   - Regression fixes for ARM GIC irqchips

   - Regression fixes and lockdep anotations for renesas irq chips

   - The leftovers of the cleanup and preparatory patches which have
     been ignored by maintainers

   - Final conversions of the newly merged users of obsolete APIs

   - Final removal of obsolete APIs

   - Final removal of ARM artifacts which had been introduced during the
     conversion of ARM to the generic interrupt code.

   - Final split of the irq_data into chip specific and common data to
     reflect the needs of hierarchical irq domains.

   - Treewide removal of the first argument of interrupt flow handlers,
     i.e. the irq number, which is not used by the majority of handlers
     and simple to retrieve from the other argument the irq descriptor.

   - A few comment updates and build warning fixes"

* 'irq-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (40 commits)
  arm64: Remove ununsed set_irq_flags
  ARM: Remove ununsed set_irq_flags
  sh: Kill off set_irq_flags usage
  irqchip: Kill off set_irq_flags usage
  gpu/drm: Kill off set_irq_flags usage
  genirq: Remove irq argument from irq flow handlers
  genirq: Move field 'msi_desc' from irq_data into irq_common_data
  genirq: Move field 'affinity' from irq_data into irq_common_data
  genirq: Move field 'handler_data' from irq_data into irq_common_data
  genirq: Move field 'node' from irq_data into irq_common_data
  irqchip/gic-v3: Use IRQD_FORWARDED_TO_VCPU flag
  irqchip/gic: Use IRQD_FORWARDED_TO_VCPU flag
  genirq: Provide IRQD_FORWARDED_TO_VCPU status flag
  genirq: Simplify irq_data_to_desc()
  genirq: Remove __irq_set_handler_locked()
  pinctrl/pistachio: Use irq_set_handler_locked
  gpio: vf610: Use irq_set_handler_locked
  powerpc/mpc8xx: Use irq_set_handler_locked()
  powerpc/ipic: Use irq_set_handler_locked()
  powerpc/cpm2: Use irq_set_handler_locked()
  ...

9 years agoMerge branch 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Fri, 18 Sep 2015 15:06:28 +0000 (08:06 -0700)]
Merge branch 'x86-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull x86 fix from Thomas Gleixner:
 "A single regression fix for the x86 dma allocator which got wreckaged
  in the merge window"

* 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  x86/pci/dma: Fix gfp flags for coherent DMA memory allocation

9 years agoMerge tag 'powerpc-4.3-2' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc...
Linus Torvalds [Fri, 18 Sep 2015 15:01:06 +0000 (08:01 -0700)]
Merge tag 'powerpc-4.3-2' of git://git./linux/kernel/git/powerpc/linux

Pull powerpc fixes from Michael Ellerman:

 - Fix 32-bit TCE table init in kdump kernel from Nish

 - Fix kdump with non-power-of-2 crashkernel= from Nish

 - Abort cxl_pci_enable_device_hook() if PCI channel is offline from
   Andrew

 - Fix to release DRC when configure_connector() fails from Bharata

 - Wire up sys_userfaultfd()

 - Fix race condition in tearing down MSI interrupts from Paul

 - Fix unbalanced pci_dev_get() in cxl_probe() from Daniel

 - Fix cxl build failure due to -Wunused-variable gcc behaviour change
   from Ian

 - Tell the toolchain to use ABI v2 when building an LE boot wrapper
   from Benh

 - Fix THP to recompute hash value after a failed update from Aneesh

 - 32-bit memcpy/memset: only use dcbz once cache is enabled from
   Christophe

* tag 'powerpc-4.3-2' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc/linux:
  powerpc32: memset: only use dcbz once cache is enabled
  powerpc32: memcpy: only use dcbz once cache is enabled
  powerpc/mm: Recompute hash value after a failed update
  powerpc/boot: Specify ABI v2 when building an LE boot wrapper
  cxl: Fix build failure due to -Wunused-variable behaviour change
  cxl: Fix unbalanced pci_dev_get in cxl_probe
  powerpc/MSI: Fix race condition in tearing down MSI interrupts
  powerpc: Wire up sys_userfaultfd()
  powerpc/pseries: Release DRC when configure_connector fails
  cxl: abort cxl_pci_enable_device_hook() if PCI channel is offline
  powerpc/powernv/pci-ioda: fix kdump with non-power-of-2 crashkernel=
  powerpc/powernv/pci-ioda: fix 32-bit TCE table init in kdump kernel

9 years agosched: access local runqueue directly in single_task_running
Dominik Dingel [Fri, 18 Sep 2015 09:27:45 +0000 (11:27 +0200)]
sched: access local runqueue directly in single_task_running

Commit 2ee507c47293 ("sched: Add function single_task_running to let a task
check if it is the only task running on a cpu") referenced the current
runqueue with the smp_processor_id.  When CONFIG_DEBUG_PREEMPT is enabled,
that is only allowed if preemption is disabled or the currrent task is
bound to the local cpu (e.g. kernel worker).

With commit f78195129963 ("kvm: add halt_poll_ns module parameter") KVM
calls single_task_running. If CONFIG_DEBUG_PREEMPT is enabled that
generates a lot of kernel messages.

To avoid adding preemption in that cases, as it would limit the usefulness,
we change single_task_running to access directly the cpu local runqueue.

Cc: Tim Chen <tim.c.chen@linux.intel.com>
Suggested-by: Peter Zijlstra <peterz@infradead.org>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: <stable@vger.kernel.org>
Fixes: 2ee507c472939db4b146d545352b8a7c79ef47f8
Signed-off-by: Dominik Dingel <dingel@linux.vnet.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
9 years agoMerge tag 'platform-drivers-x86-v4.3-2' of git://git.infradead.org/users/dvhart/linux...
Linus Torvalds [Fri, 18 Sep 2015 04:41:02 +0000 (21:41 -0700)]
Merge tag 'platform-drivers-x86-v4.3-2' of git://git.infradead.org/users/dvhart/linux-platform-drivers-x86

Pull x86 platform driver fixes from Darren Hart:
 "Fix an issue introduced by the previous major toshiba rework.  Add a
  quirk.  Workaround a few platform specific firmware items.  One
  cleanup to wmi I inadvertently dropped from a previous pull request.

  Details:

  hp-wmi:
   - limit hotkey enable

  toshiba_acpi:
   - Fix hotkeys registration on some toshiba models
   - Fix USB Sleep and Music always disabled

  wmi:
   - Remove private %pUL implementation

  asus-nb-wmi:
   - Add wapf=4 quirk for X456UA/X456UF"

* tag 'platform-drivers-x86-v4.3-2' of git://git.infradead.org/users/dvhart/linux-platform-drivers-x86:
  hp-wmi: limit hotkey enable
  toshiba_acpi: Fix hotkeys registration on some toshiba models
  toshiba_acpi: Fix USB Sleep and Music always disabled
  wmi: Remove private %pUL implementation
  asus-nb-wmi: Add wapf=4 quirk for X456UA/X456UF

9 years agoMerge branch 'akpm' (patches from Andrew)
Linus Torvalds [Fri, 18 Sep 2015 04:16:47 +0000 (21:16 -0700)]
Merge branch 'akpm' (patches from Andrew)

Merge misc fixes from ANdrew Morton:
 "8 fixes"

* emailed patches from Andrew Morton <akpm@linux-foundation.org>:
  revert "mm: make sure all file VMAs have ->vm_ops set"
  MAINTAINERS: update LTP mailing list
  userfaultfd: add missing mmput() in error path
  lib/string_helpers.c: fix infinite loop in string_get_size()
  alpha: lib: export __delay
  alpha: io: define ioremap_uc
  kasan: fix last shadow judgement in memory_is_poisoned_16()
  zram: fix possible use after free in zcomp_create()

9 years agorevert "mm: make sure all file VMAs have ->vm_ops set"
Andrew Morton [Thu, 17 Sep 2015 23:02:00 +0000 (16:02 -0700)]
revert "mm: make sure all file VMAs have ->vm_ops set"

Revert commit 6dc296e7df4c "mm: make sure all file VMAs have ->vm_ops
set".

Will Deacon reports that it "causes some mmap regressions in LTP, which
appears to use a MAP_PRIVATE mmap of /dev/zero as a way to get anonymous
pages in some of its tests (specifically mmap10 [1])".

William Shuman reports Oracle crashes.

So revert the patch while we work out what to do.

Reported-by: William Shuman <wshuman3@gmail.com>
Reported-by: Will Deacon <will.deacon@arm.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agoMAINTAINERS: update LTP mailing list
Cyril Hrubis [Thu, 17 Sep 2015 23:01:57 +0000 (16:01 -0700)]
MAINTAINERS: update LTP mailing list

[akpm@linux-foundation.org: Wanlong Gao has moved]
Signed-off-by: Cyril Hrubis <chrubis@suse.cz>
Cc: Jan Stancek <jstancek@redhat.com>
Cc: Stanislav Kholmanskikh <stanislav.kholmanskikh@oracle.com>
Cc: Alexey Kodanev <alexey.kodanev@oracle.com>
Cc: Wanlong Gao <wanlong.gao@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agouserfaultfd: add missing mmput() in error path
Eric Biggers [Thu, 17 Sep 2015 23:01:54 +0000 (16:01 -0700)]
userfaultfd: add missing mmput() in error path

This fixes a memleak if anon_inode_getfile() fails in userfaultfd().

Signed-off-by: Eric Biggers <ebiggers3@gmail.com>
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agolib/string_helpers.c: fix infinite loop in string_get_size()
Vitaly Kuznetsov [Thu, 17 Sep 2015 23:01:51 +0000 (16:01 -0700)]
lib/string_helpers.c: fix infinite loop in string_get_size()

Some string_get_size() calls (e.g.:
 string_get_size(1, 512, STRING_UNITS_10, ..., ...)
 string_get_size(15, 64, STRING_UNITS_10, ..., ...)
) result in an infinite loop. The problem is that if size is equal to
divisor[units]/blk_size and is smaller than divisor[units] we'll end
up with size == 0 when we start doing sf_cap calculations:

For string_get_size(1, 512, STRING_UNITS_10, ..., ...) case:
   ...
   remainder = do_div(size, divisor[units]); -> size is 0, remainder is 1
   remainder *= blk_size; -> remainder is 512
   ...
   size *= blk_size; -> size is still 0
   size += remainder / divisor[units]; -> size is still 0

The caller causing the issue is sd_read_capacity(), the problem was
noticed on Hyper-V, such weird size was reported by host when scanning
collides with device removal.  This is probably a separate issue worth
fixing, this patch is intended to prevent the library routine from
infinite looping.

Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Acked-by: James Bottomley <JBottomley@Odin.com>
Cc: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Cc: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Cc: "K. Y. Srinivasan" <kys@microsoft.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agoalpha: lib: export __delay
Sudip Mukherjee [Thu, 17 Sep 2015 23:01:49 +0000 (16:01 -0700)]
alpha: lib: export __delay

__delay was not exported as a result while building with allmodconfig we
were getting build error of undefined symbol.  __delay is being used by:
drivers/net/phy/mdio-octeon.c

Signed-off-by: Sudip Mukherjee <sudip@vectorindia.org>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: Matt Turner <mattst88@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agoalpha: io: define ioremap_uc
Sudip Mukherjee [Thu, 17 Sep 2015 23:01:46 +0000 (16:01 -0700)]
alpha: io: define ioremap_uc

ioremap_uc was not defined and as a result while building with
allmodconfig were getting build error of: implicit declaration of
function 'ioremap_uc'.

Signed-off-by: Sudip Mukherjee <sudip@vectorindia.org>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: Matt Turner <mattst88@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agokasan: fix last shadow judgement in memory_is_poisoned_16()
Xishi Qiu [Thu, 17 Sep 2015 23:01:43 +0000 (16:01 -0700)]
kasan: fix last shadow judgement in memory_is_poisoned_16()

The shadow which correspond 16 bytes memory may span 2 or 3 bytes.  If
the memory is aligned on 8, then the shadow takes only 2 bytes.  So we
check "shadow_first_bytes" is enough, and need not to call
"memory_is_poisoned_1(addr + 15);".  But the code "if
(likely(!last_byte))" is wrong judgement.

e.g.  addr=0, so last_byte = 15 & KASAN_SHADOW_MASK = 7, then the code
will continue to call "memory_is_poisoned_1(addr + 15);"

Signed-off-by: Xishi Qiu <qiuxishi@huawei.com>
Acked-by: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Andrey Konovalov <adech.fo@gmail.com>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Michal Marek <mmarek@suse.cz>
Cc: <zhongjiang@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agozram: fix possible use after free in zcomp_create()
Luis Henriques [Thu, 17 Sep 2015 23:01:40 +0000 (16:01 -0700)]
zram: fix possible use after free in zcomp_create()

zcomp_create() verifies the success of zcomp_strm_{multi,single}_create()
through comp->stream, which can potentially be pointing to memory that
was freed if these functions returned an error.

While at it, replace a 'ERR_PTR(-ENOMEM)' by a more generic
'ERR_PTR(error)' as in the future zcomp_strm_{multi,siggle}_create()
could return other error codes.  Function documentation updated
accordingly.

Fixes: beca3ec71fe5 ("zram: add multi stream functionality")
Signed-off-by: Luis Henriques <luis.henriques@canonical.com>
Acked-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Acked-by: Minchan Kim <minchan@kernel.org>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agohp-wmi: limit hotkey enable
Kyle Evans [Fri, 11 Sep 2015 15:40:17 +0000 (10:40 -0500)]
hp-wmi: limit hotkey enable

Do not write initialize magic on systems that do not have
feature query 0xb. Fixes Bug #82451.

Redefine FEATURE_QUERY to align with 0xb and FEATURE2 with 0xd
for code clearity.

Add a new test function, hp_wmi_bios_2008_later() & simplify
hp_wmi_bios_2009_later(), which fixes a bug in cases where
an improper value is returned. Probably also fixes Bug #69131.

Add missing __init tag.

Signed-off-by: Kyle Evans <kvans32@gmail.com>
Cc: stable@vger.kernel.org
Signed-off-by: Darren Hart <dvhart@linux.intel.com>
9 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/sage/ceph...
Linus Torvalds [Thu, 17 Sep 2015 19:32:40 +0000 (12:32 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/sage/ceph-client

Pull Ceph fixes from Sage Weil:
 "These are both fixes to the new and improved keepalive2 behavior"

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/sage/ceph-client:
  libceph: advertise support for keepalive2
  libceph: don't access invalid memory in keepalive2 path

9 years agoMerge tag 'for-v4.3-rc' of git://git.kernel.org/pub/scm/linux/kernel/git/sre/linux...
Linus Torvalds [Thu, 17 Sep 2015 19:25:42 +0000 (12:25 -0700)]
Merge tag 'for-v4.3-rc' of git://git./linux/kernel/git/sre/linux-power-supply

Pull power supply fixes from Sebastian Reichel:
 "twl4030-charger fixes"

* tag 'for-v4.3-rc' of git://git.kernel.org/pub/scm/linux/kernel/git/sre/linux-power-supply:
  twl4030_charger: fix another compile error
  Revert "twl4030_charger: correctly handle -EPROBE_DEFER from devm_usb_get_phy_by_node"

9 years agodrivers: clk: st: Rename st_pll3200c32_407_c0_x into st_pll3200c32_cx_x
Gabriel Fernandez [Wed, 16 Sep 2015 07:42:59 +0000 (09:42 +0200)]
drivers: clk: st: Rename st_pll3200c32_407_c0_x into st_pll3200c32_cx_x

Use a generic name for this kind of PLL

Correction in dts files are already done here:
commit 5eb26c605909 ("ARM: STi: DT: Rename st_pll3200c32_407_c0_x into st_pll3200c32_cx_x")

Signed-off-by: Gabriel Fernandez <gabriel.fernandez@linaro.org>
Signed-off-by: Stephen Boyd <sboyd@codeaurora.org>
9 years agoMerge tag 'arm64-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux
Linus Torvalds [Thu, 17 Sep 2015 18:28:17 +0000 (11:28 -0700)]
Merge tag 'arm64-fixes' of git://git./linux/kernel/git/arm64/linux

Pull arm64 fixes from Will Deacon:
 "This addresses some problems with filesystem writeback due to the
  recently merged hardware DBM patches, which caused us to treat some
  read-only pages as dirty.

  There are also some other, less significant fixes that are described
  in the summary below:

  A mixture of fixes for regressions introduced during the merge window,
  some longer standing problems that we spotted and a couple of hardware
  errata.  The main changes are:

   - Fix fallout from the h/w DBM patches, causing filesystem writeback
     issues on both v8 and v8.1 CPUs

   - Workaround for Cortex-A53 erratum #843419 in the module loader

   - Fix for long-standing issue with compat big-endian signal handlers
     using the saved floating point state"

* tag 'arm64-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux:
  arm64: errata: add module build workaround for erratum #843419
  arm64: compat: fix vfp save/restore across signal handlers in big-endian
  arm64: cpu hotplug: ensure we mask out CPU_TASKS_FROZEN in notifiers
  arm64: head.S: initialise mdcr_el2 in el2_setup
  arm64: enable generic idle loop
  arm64: pgtable: use a single bit for PTE_WRITE regardless of DBM
  arm64: Fix pte_modify() to preserve the hardware dirty information
  arm64: Fix the pte_hw_dirty() check when AF/DBM is enabled
  arm64: dma-mapping: check whether cma area is initialized or not