A patchset worked out together with Peter Zijlstra. Ingo is OK with taking
it through the DRM tree:
This is a small fallout from a work to allow batching WW mutex locks and
unlocks.
Our Wound-Wait mutexes actually don't use the Wound-Wait algorithm but
the Wait-Die algorithm. One could perhaps rename those mutexes tree-wide to
"Wait-Die mutexes" or "Deadlock Avoidance mutexes". Another approach suggested
here is to implement also the "Wound-Wait" algorithm as a per-WW-class
choice, as it has advantages in some cases. See for example
http://www.mathcs.emory.edu/~cheung/Courses/554/Syllabus/8-recv+serial/deadlock-compare.html
Now Wound-Wait is a preemptive algorithm, and the preemption is implemented
using a lazy scheme: If a wounded transaction is about to go to sleep on
a contended WW mutex, we return -EDEADLK. That is sufficient for deadlock
prevention. Since with WW mutexes we also require the aborted transaction to
sleep waiting to lock the WW mutex it was aborted on, this choice also provides
a suitable WW mutex to sleep on. If we were to return -EDEADLK on the first
WW mutex lock after the transaction was wounded whether the WW mutex was
contended or not, the transaction might frequently be restarted without a wait,
which is far from optimal. Note also that with the lazy preemption scheme,
contrary to Wait-Die there will be no rollbacks on lock contention of locks
held by a transaction that has completed its locking sequence.
The modeset locks are then changed from Wait-Die to Wound-Wait since the
typical locking pattern of those locks very well matches the criterion for
a substantial reduction in the number of rollbacks. For reservation objects,
the benefit is more unclear at this point and they remain using Wait-Die.
Signed-off-by: Dave Airlie <airlied@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180703105339.4461-1-thellstrom@vmware.com
The current Wound-Wait mutex algorithm is actually not Wound-Wait but
Wait-Die. Implement also Wound-Wait as a per-ww-class choice. Wound-Wait
is, contrary to Wait-Die a preemptive algorithm and is known to generate
fewer backoffs. Testing reveals that this is true if the
number of simultaneous contending transactions is small.
As the number of simultaneous contending threads increases, Wait-Wound
becomes inferior to Wait-Die in terms of elapsed time.
Possibly due to the larger number of held locks of sleeping transactions.
Update documentation and callers.
Timings using git://people.freedesktop.org/~thomash/ww_mutex_test
tag patch-18-06-15
Each thread runs 100000 batches of lock / unlock 800 ww mutexes randomly
chosen out of 100000. Four core Intel x86_64:
Algorithm #threads Rollbacks time
Wound-Wait 4 ~100 ~17s.
Wait-Die 4 ~150000 ~19s.
Wound-Wait 16 ~360000 ~109s.
Wait-Die 16 ~450000 ~82s.
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Gustavo Padovan <gustavo@padovan.org>
Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Cc: Sean Paul <seanpaul@chromium.org>
Cc: David Airlie <airlied@linux.ie>
Cc: Davidlohr Bueso <dave@stgolabs.net>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Kate Stewart <kstewart@linuxfoundation.org>
Cc: Philippe Ombredanne <pombredanne@nexb.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: linux-doc@vger.kernel.org
Cc: linux-media@vger.kernel.org
Cc: linaro-mm-sig@lists.linaro.org
Co-authored-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Ingo Molnar <mingo@kernel.org>
This commit adds "#define pr_fmt(fmt) fmt" to the torture-test files
in order to keep the current dmesg format. Once Joe's commits have
hit mainline, these definitions will be changed in order to automatically
generate the dmesg line prefix that the scripts expect. This will have
the beneficial side-effect of allowing printk() formats to be used more
widely and of shortening some pr_*() lines.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Joe Perches <joe@perches.com>
Some bugs reproduce quickly only at high CPU-hotplug rates, so the
rcutorture TREE03 scenario now has only 200 milliseconds spacing between
CPU-hotplug operations. At this rate, the torture-test pair of console
messages per operation becomes a bit voluminous. This commit therefore
converts the torture-test set of "verbose" kernel-boot arguments from
bool to int, and prints the extra console messages only when verbose=2.
The default is still verbose=1.
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Pull locking fixes from Thomas Gleixner:
"A set of fixes and updates for the locking code:
- Prevent lockdep from updating irq state within its own code and
thereby confusing itself.
- Buid fix for older GCCs which mistreat anonymous unions
- Add a missing lockdep annotation in down_read_non_onwer() which
causes up_read_non_owner() to emit a lockdep splat
- Remove the custom alpha dec_and_lock() implementation which is
incorrect in terms of ordering and use the generic one.
The remaining two commits are not strictly fixes. They provide irqsave
variants of atomic_dec_and_lock() and refcount_dec_and_lock(). These
are required to merge the relevant updates and cleanups into different
maintainer trees for 4.19, so routing them into mainline without
actual users is the sanest approach.
They should have been in -rc1, but last weekend I took the liberty to
just avoid computers in order to regain some mental sanity"
* 'locking-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
locking/qspinlock: Fix build for anonymous union in older GCC compilers
locking/lockdep: Do not record IRQ state within lockdep code
locking/rwsem: Fix up_read_non_owner() warning with DEBUG_RWSEMS
locking/refcounts: Implement refcount_dec_and_lock_irqsave()
atomic: Add irqsave variant of atomic_dec_and_lock()
alpha: Remove custom dec_and_lock() implementation
Pull locking updates from Ingo Molnar:
- Lots of tidying up changes all across the map for Linux's formal
memory/locking-model tooling, by Alan Stern, Akira Yokosawa, Andrea
Parri, Paul E. McKenney and SeongJae Park.
Notable changes beyond an overall update in the tooling itself is the
tidying up of spin_is_locked() semantics, which spills over into the
kernel proper as well.
- qspinlock improvements: the locking algorithm now guarantees forward
progress whereas the previous implementation in mainline could starve
threads indefinitely in cmpxchg() loops. Also other related cleanups
to the qspinlock code (Will Deacon)
- misc smaller improvements, cleanups and fixes all across the locking
subsystem
* 'locking-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (51 commits)
locking/rwsem: Simplify the is-owner-spinnable checks
tools/memory-model: Add reference for 'Simplifying ARM concurrency'
tools/memory-model: Update ASPLOS information
MAINTAINERS, tools/memory-model: Update e-mail address for Andrea Parri
tools/memory-model: Fix coding style in 'lock.cat'
tools/memory-model: Remove out-of-date comments and code from lock.cat
tools/memory-model: Improve mixed-access checking in lock.cat
tools/memory-model: Improve comments in lock.cat
tools/memory-model: Remove duplicated code from lock.cat
tools/memory-model: Flag "cumulativity" and "propagation" tests
tools/memory-model: Add model support for spin_is_locked()
tools/memory-model: Add scripts to test memory model
tools/memory-model: Fix coding style in 'linux-kernel.def'
tools/memory-model: Model 'smp_store_mb()'
tools/memory-order: Update the cheat-sheet to show that smp_mb__after_atomic() orders later RMW operations
tools/memory-order: Improve key for SELF and SV
tools/memory-model: Fix cheat sheet typo
tools/memory-model: Update required version of herdtools7
tools/memory-model: Redefine rb in terms of rcu-fence
tools/memory-model: Rename link and rcu-path to rcu-link and rb
...
The filesystem freezing code needs to transfer ownership of a rwsem
embedded in a percpu-rwsem from the task that does the freezing to
another one that does the thawing by calling percpu_rwsem_release()
after freezing and percpu_rwsem_acquire() before thawing.
However, the new rwsem debug code runs afoul with this scheme by warning
that the task that releases the rwsem isn't the one that acquires it,
as reported by Amir Goldstein:
DEBUG_LOCKS_WARN_ON(sem->owner != get_current())
WARNING: CPU: 1 PID: 1401 at /home/amir/build/src/linux/kernel/locking/rwsem.c:133 up_write+0x59/0x79
Call Trace:
percpu_up_write+0x1f/0x28
thaw_super_locked+0xdf/0x120
do_vfs_ioctl+0x270/0x5f1
ksys_ioctl+0x52/0x71
__x64_sys_ioctl+0x16/0x19
do_syscall_64+0x5d/0x167
entry_SYSCALL_64_after_hwframe+0x49/0xbe
To work properly with the rwsem debug code, we need to annotate that the
rwsem ownership is unknown during the tranfer period until a brave soul
comes forward to acquire the ownership. During that period, optimistic
spinning will be disabled.
Reported-by: Amir Goldstein <amir73il@gmail.com>
Tested-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Waiman Long <longman@redhat.com>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Davidlohr Bueso <dave@stgolabs.net>
Cc: Jan Kara <jack@suse.cz>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Theodore Y. Ts'o <tytso@mit.edu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Will Deacon <will.deacon@arm.com>
Cc: linux-fsdevel@vger.kernel.org
Link: http://lkml.kernel.org/r/1526420991-21213-3-git-send-email-longman@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
There are use cases where a rwsem can be acquired by one task, but
released by another task. In thess cases, optimistic spinning may need
to be disabled. One example will be the filesystem freeze/thaw code
where the task that freezes the filesystem will acquire a write lock
on a rwsem and then un-owns it before returning to userspace. Later on,
another task will come along, acquire the ownership, thaw the filesystem
and release the rwsem.
Bit 0 of the owner field was used to designate that it is a reader
owned rwsem. It is now repurposed to mean that the owner of the rwsem
is not known. If only bit 0 is set, the rwsem is reader owned. If bit
0 and other bits are set, it is writer owned with an unknown owner.
One such value for the latter case is (-1L). So we can set owner to 1 for
reader-owned, -1 for writer-owned. The owner is unknown in both cases.
To handle transfer of rwsem ownership, the higher level code should
set the owner field to -1 to indicate a write-locked rwsem with unknown
owner. Optimistic spinning will be disabled in this case.
Once the higher level code figures who the new owner is, it can then
set the owner field accordingly.
Tested-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Waiman Long <longman@redhat.com>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Davidlohr Bueso <dave@stgolabs.net>
Cc: Jan Kara <jack@suse.cz>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Theodore Y. Ts'o <tytso@mit.edu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Will Deacon <will.deacon@arm.com>
Cc: linux-fsdevel@vger.kernel.org
Link: http://lkml.kernel.org/r/1526420991-21213-2-git-send-email-longman@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Variants of proc_create{,_data} that directly take a seq_file show
callback and drastically reduces the boilerplate code in the callers.
All trivial callers converted over.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Variants of proc_create{,_data} that directly take a struct seq_operations
argument and drastically reduces the boilerplate code in the callers.
All trivial callers converted over.
Signed-off-by: Christoph Hellwig <hch@lst.de>
The qspinlock slowpath must ensure that the MCS node is fully initialised
before it can be reached by another other CPU. This is currently enforced
by using a RELEASE operation when updating the tail and also when linking
the node into the waitqueue, since the control dependency off xchg_tail
is insufficient to enforce sufficient ordering, see:
95bcade33a ("locking/qspinlock: Ensure node is initialised before updating prev->next")
Back-to-back RELEASE operations may be expensive on some architectures,
particularly those that implement them using fences under the hood. We
can replace the two RELEASE operations with a single smp_wmb() fence and
use RELAXED operations for the subsequent publishing of the node.
Signed-off-by: Will Deacon <will.deacon@arm.com>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Waiman Long <longman@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: boqun.feng@gmail.com
Cc: linux-arm-kernel@lists.infradead.org
Cc: paulmck@linux.vnet.ibm.com
Link: http://lkml.kernel.org/r/1524738868-31318-12-git-send-email-will.deacon@arm.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
The qspinlock locking slowpath utilises a "pending" bit as a simple form
of an embedded test-and-set lock that can avoid the overhead of explicit
queuing in cases where the lock is held but uncontended. This bit is
managed using a cmpxchg() loop which tries to transition the uncontended
lock word from (0,0,0) -> (0,0,1) or (0,0,1) -> (0,1,1).
Unfortunately, the cmpxchg() loop is unbounded and lockers can be starved
indefinitely if the lock word is seen to oscillate between unlocked
(0,0,0) and locked (0,0,1). This could happen if concurrent lockers are
able to take the lock in the cmpxchg() loop without queuing and pass it
around amongst themselves.
This patch fixes the problem by unconditionally setting _Q_PENDING_VAL
using atomic_fetch_or, and then inspecting the old value to see whether
we need to spin on the current lock owner, or whether we now effectively
hold the lock. The tricky scenario is when concurrent lockers end up
queuing on the lock and the lock becomes available, causing us to see
a lockword of (n,0,0). With pending now set, simply queuing could lead
to deadlock as the head of the queue may not have observed the pending
flag being cleared. Conversely, if the head of the queue did observe
pending being cleared, then it could transition the lock from (n,0,0) ->
(0,0,1) meaning that any attempt to "undo" our setting of the pending
bit could race with a concurrent locker trying to set it.
We handle this race by preserving the pending bit when taking the lock
after reaching the head of the queue and leaving the tail entry intact
if we saw pending set, because we know that the tail is going to be
updated shortly.
Signed-off-by: Will Deacon <will.deacon@arm.com>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Waiman Long <longman@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: boqun.feng@gmail.com
Cc: linux-arm-kernel@lists.infradead.org
Cc: paulmck@linux.vnet.ibm.com
Link: http://lkml.kernel.org/r/1524738868-31318-6-git-send-email-will.deacon@arm.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
If a locker taking the qspinlock slowpath reads a lock value indicating
that only the pending bit is set, then it will spin whilst the
concurrent pending->locked transition takes effect.
Unfortunately, there is no guarantee that such a transition will ever be
observed since concurrent lockers could continuously set pending and
hand over the lock amongst themselves, leading to starvation. Whilst
this would probably resolve in practice, it means that it is not
possible to prove liveness properties about the lock and means that lock
acquisition time is unbounded.
Rather than removing the pending->locked spinning from the slowpath
altogether (which has been shown to heavily penalise a 2-threaded
locking stress test on x86), this patch replaces the explicit spinning
with a call to atomic_cond_read_relaxed and allows the architecture to
provide a bound on the number of spins. For architectures that can
respond to changes in cacheline state in their smp_cond_load implementation,
it should be sufficient to use the default bound of 1.
Suggested-by: Waiman Long <longman@redhat.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Waiman Long <longman@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: boqun.feng@gmail.com
Cc: linux-arm-kernel@lists.infradead.org
Cc: paulmck@linux.vnet.ibm.com
Link: http://lkml.kernel.org/r/1524738868-31318-4-git-send-email-will.deacon@arm.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
The lock debug output in print_lock() has a few shortcomings:
- It prints the hlock->acquire_ip field in %px and %pS format. That's
redundant information.
- It lacks information about the lock object itself. The lock class is
not helpful to identify a particular instance of a lock.
Change the output so it prints:
- hlock->instance to allow identification of a particular lock instance.
- only the %pS format of hlock->ip_acquire which is sufficient to decode
the actual code line with faddr2line.
The resulting output is:
3 locks held by a.out/31106:
#0: 00000000b0f753ba (&mm->mmap_sem){++++}, at: copy_process.part.41+0x10d5/0x1fe0
#1: 00000000ef64d539 (&mm->mmap_sem/1){+.+.}, at: copy_process.part.41+0x10fe/0x1fe0
#2: 00000000b41a282e (&mapping->i_mmap_rwsem){++++}, at: copy_process.part.41+0x12f2/0x1fe0
[ tglx: Massaged changelog ]
Signed-off-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: David Rientjes <rientjes@google.com>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Cc: linux-mm@kvack.org
Cc: Borislav Petkov <bp@suse.de>
Link: https://lkml.kernel.org/r/201803271941.GBE57310.tVSOJLQOFFOHFM@I-love.SAKURA.ne.jp
In -RT task_blocks_on_rt_mutex() may return with -EAGAIN due to
(->pi_blocked_on == PI_WAKEUP_INPROGRESS) before it added itself as a
waiter. In such a case remove_waiter() must not be called because without a
waiter it will trigger the BUG_ON() statement.
This was initially reported by Yimin Deng. Thomas Gleixner fixed it then
with an explicit check for waiters before calling remove_waiter().
Instead of an explicit NULL check before calling rt_mutex_top_waiter() make
the function return NULL if there are no waiters. With that fixed the now
pointless NULL check is removed from rt_mutex_slowlock().
Reported-and-debugged-by: Yimin Deng <yimin11.deng@gmail.com>
Suggested-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lkml.kernel.org/r/CAAh1qt=DCL9aUXNxanP5BKtiPp3m+qj4yB+gDohhXPVFCxWwzg@mail.gmail.com
Link: https://lkml.kernel.org/r/20180327121438.sss7hxg3crqy4ecd@linutronix.de
When running rcutorture with TREE03 config, CONFIG_PROVE_LOCKING=y, and
kernel cmdline argument "rcutorture.gp_exp=1", lockdep reports a
HARDIRQ-safe->HARDIRQ-unsafe deadlock:
================================
WARNING: inconsistent lock state
4.16.0-rc4+ #1 Not tainted
--------------------------------
inconsistent {IN-HARDIRQ-W} -> {HARDIRQ-ON-W} usage.
takes:
__schedule+0xbe/0xaf0
{IN-HARDIRQ-W} state was registered at:
_raw_spin_lock+0x2a/0x40
scheduler_tick+0x47/0xf0
...
other info that might help us debug this:
Possible unsafe locking scenario:
CPU0
----
lock(&rq->lock);
<Interrupt>
lock(&rq->lock);
*** DEADLOCK ***
1 lock held by rcu_torture_rea/724:
rcu_torture_read_lock+0x0/0x70
stack backtrace:
CPU: 2 PID: 724 Comm: rcu_torture_rea Not tainted 4.16.0-rc4+ #1
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.11.0-20171110_100015-anatol 04/01/2014
Call Trace:
lock_acquire+0x90/0x200
? __schedule+0xbe/0xaf0
_raw_spin_lock+0x2a/0x40
? __schedule+0xbe/0xaf0
__schedule+0xbe/0xaf0
preempt_schedule_irq+0x2f/0x60
retint_kernel+0x1b/0x2d
RIP: 0010:rcu_read_unlock_special+0x0/0x680
? rcu_torture_read_unlock+0x60/0x60
__rcu_read_unlock+0x64/0x70
rcu_torture_read_unlock+0x17/0x60
rcu_torture_reader+0x275/0x450
? rcutorture_booster_init+0x110/0x110
? rcu_torture_stall+0x230/0x230
? kthread+0x10e/0x130
kthread+0x10e/0x130
? kthread_create_worker_on_cpu+0x70/0x70
? call_usermodehelper_exec_async+0x11a/0x150
ret_from_fork+0x3a/0x50
This happens with the following even sequence:
preempt_schedule_irq();
local_irq_enable();
__schedule():
local_irq_disable(); // irq off
...
rcu_note_context_switch():
rcu_note_preempt_context_switch():
rcu_read_unlock_special():
local_irq_save(flags);
...
raw_spin_unlock_irqrestore(...,flags); // irq remains off
rt_mutex_futex_unlock():
raw_spin_lock_irq();
...
raw_spin_unlock_irq(); // accidentally set irq on
<return to __schedule()>
rq_lock():
raw_spin_lock(); // acquiring rq->lock with irq on
which means rq->lock becomes a HARDIRQ-unsafe lock, which can cause
deadlocks in scheduler code.
This problem was introduced by commit 02a7c234e5 ("rcu: Suppress
lockdep false-positive ->boost_mtx complaints"). That brought the user
of rt_mutex_futex_unlock() with irq off.
To fix this, replace the *lock_irq() in rt_mutex_futex_unlock() with
*lock_irq{save,restore}() to make it safe to call rt_mutex_futex_unlock()
with irq off.
Fixes: 02a7c234e5 ("rcu: Suppress lockdep false-positive ->boost_mtx complaints")
Signed-off-by: Boqun Feng <boqun.feng@gmail.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Lai Jiangshan <jiangshanlai@gmail.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Cc: "Paul E . McKenney" <paulmck@linux.vnet.ibm.com>
Link: https://lkml.kernel.org/r/20180309065630.8283-1-boqun.feng@gmail.com
When queuing on the qspinlock, the count field for the current CPU's head
node is incremented. This needn't be atomic because locking in e.g. IRQ
context is balanced and so an IRQ will return with node->count as it
found it.
However, the compiler could in theory reorder the initialisation of
node[idx] before the increment of the head node->count, causing an
IRQ to overwrite the initialised node and potentially corrupt the lock
state.
Avoid the potential for this harmful compiler reordering by placing a
barrier() between the increment of the head node->count and the subsequent
node initialisation.
Signed-off-by: Will Deacon <will.deacon@arm.com>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/1518528177-19169-3-git-send-email-will.deacon@arm.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
When a locker ends up queuing on the qspinlock locking slowpath, we
initialise the relevant mcs node and publish it indirectly by updating
the tail portion of the lock word using xchg_tail. If we find that there
was a pre-existing locker in the queue, we subsequently update their
->next field to point at our node so that we are notified when it's our
turn to take the lock.
This can be roughly illustrated as follows:
/* Initialise the fields in node and encode a pointer to node in tail */
tail = initialise_node(node);
/*
* Exchange tail into the lockword using an atomic read-modify-write
* operation with release semantics
*/
old = xchg_tail(lock, tail);
/* If there was a pre-existing waiter ... */
if (old & _Q_TAIL_MASK) {
prev = decode_tail(old);
smp_read_barrier_depends();
/* ... then update their ->next field to point to node.
WRITE_ONCE(prev->next, node);
}
The conditional update of prev->next therefore relies on the address
dependency from the result of xchg_tail ensuring order against the
prior initialisation of node. However, since the release semantics of
the xchg_tail operation apply only to the write portion of the RmW,
then this ordering is not guaranteed and it is possible for the CPU
to return old before the writes to node have been published, consequently
allowing us to point prev->next to an uninitialised node.
This patch fixes the problem by making the update of prev->next a RELEASE
operation, which also removes the reliance on dependency ordering.
Signed-off-by: Will Deacon <will.deacon@arm.com>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/1518528177-19169-2-git-send-email-will.deacon@arm.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Pull locking updates from Ingo Molnar:
"The main changes relate to making lock_is_held() et al (and external
wrappers of them) work on const data types - this requires const
propagation through the depths of lockdep.
This removes a number of ugly type hacks the external helpers used"
* 'locking-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
lockdep: Convert some users to const
lockdep: Make lockdep checking constant
lockdep: Assign lock keys on registration
Pull RCU updates from Ingo Molnar:
"The main RCU changes in this cycle were:
- Updates to use cond_resched() instead of cond_resched_rcu_qs()
where feasible (currently everywhere except in kernel/rcu and in
kernel/torture.c). Also a couple of fixes to avoid sending IPIs to
offline CPUs.
- Updates to simplify RCU's dyntick-idle handling.
- Updates to remove almost all uses of smp_read_barrier_depends() and
read_barrier_depends().
- Torture-test updates.
- Miscellaneous fixes"
* 'core-rcu-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (72 commits)
torture: Save a line in stutter_wait(): while -> for
torture: Eliminate torture_runnable and perf_runnable
torture: Make stutter less vulnerable to compilers and races
locking/locktorture: Fix num reader/writer corner cases
locking/locktorture: Fix rwsem reader_delay
torture: Place all torture-test modules in one MAINTAINERS group
rcutorture/kvm-build.sh: Skip build directory check
rcutorture: Simplify functions.sh include path
rcutorture: Simplify logging
rcutorture/kvm-recheck-*: Improve result directory readability check
rcutorture/kvm.sh: Support execution from any directory
rcutorture/kvm.sh: Use consistent help text for --qemu-args
rcutorture/kvm.sh: Remove unused variable, `alldone`
rcutorture: Remove unused script, config2frag.sh
rcutorture/configinit: Fix build directory error message
rcutorture: Preempt RCU-preempt readers more vigorously
torture: Reduce #ifdefs for preempt_schedule()
rcu: Remove have_rcu_nocb_mask from tree_plugin.h
rcu: Add comment giving debug strategy for double call_rcu()
tracing, rcu: Hide trace event rcu_nocb_wake when not used
...
Julia reported futex state corruption in the following scenario:
waiter waker stealer (prio > waiter)
futex(WAIT_REQUEUE_PI, uaddr, uaddr2,
timeout=[N ms])
futex_wait_requeue_pi()
futex_wait_queue_me()
freezable_schedule()
<scheduled out>
futex(LOCK_PI, uaddr2)
futex(CMP_REQUEUE_PI, uaddr,
uaddr2, 1, 0)
/* requeues waiter to uaddr2 */
futex(UNLOCK_PI, uaddr2)
wake_futex_pi()
cmp_futex_value_locked(uaddr2, waiter)
wake_up_q()
<woken by waker>
<hrtimer_wakeup() fires,
clears sleeper->task>
futex(LOCK_PI, uaddr2)
__rt_mutex_start_proxy_lock()
try_to_take_rt_mutex() /* steals lock */
rt_mutex_set_owner(lock, stealer)
<preempted>
<scheduled in>
rt_mutex_wait_proxy_lock()
__rt_mutex_slowlock()
try_to_take_rt_mutex() /* fails, lock held by stealer */
if (timeout && !timeout->task)
return -ETIMEDOUT;
fixup_owner()
/* lock wasn't acquired, so,
fixup_pi_state_owner skipped */
return -ETIMEDOUT;
/* At this point, we've returned -ETIMEDOUT to userspace, but the
* futex word shows waiter to be the owner, and the pi_mutex has
* stealer as the owner */
futex_lock(LOCK_PI, uaddr2)
-> bails with EDEADLK, futex word says we're owner.
And suggested that what commit:
73d786bd04 ("futex: Rework inconsistent rt_mutex/futex_q state")
removes from fixup_owner() looks to be just what is needed. And indeed
it is -- I completely missed that requeue_pi could also result in this
case. So we need to restore that, except that subsequent patches, like
commit:
16ffa12d74 ("futex: Pull rt_mutex_futex_unlock() out from under hb->lock")
changed all the locking rules. Even without that, the sequence:
- if (rt_mutex_futex_trylock(&q->pi_state->pi_mutex)) {
- locked = 1;
- goto out;
- }
- raw_spin_lock_irq(&q->pi_state->pi_mutex.wait_lock);
- owner = rt_mutex_owner(&q->pi_state->pi_mutex);
- if (!owner)
- owner = rt_mutex_next_owner(&q->pi_state->pi_mutex);
- raw_spin_unlock_irq(&q->pi_state->pi_mutex.wait_lock);
- ret = fixup_pi_state_owner(uaddr, q, owner);
already suggests there were races; otherwise we'd never have to look
at next_owner.
So instead of doing 3 consecutive wait_lock sections with who knows
what races, we do it all in a single section. Additionally, the usage
of pi_state->owner in fixup_owner() was only safe because only the
rt_mutex owner would modify it, which this additional case wrecks.
Luckily the values can only change away and not to the value we're
testing, this means we can do a speculative test and double check once
we have the wait_lock.
Fixes: 73d786bd04 ("futex: Rework inconsistent rt_mutex/futex_q state")
Reported-by: Julia Cartwright <julia@ni.com>
Reported-by: Gratian Crisan <gratian.crisan@ni.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Julia Cartwright <julia@ni.com>
Tested-by: Gratian Crisan <gratian.crisan@ni.com>
Cc: Darren Hart <dvhart@infradead.org>
Cc: stable@vger.kernel.org
Link: https://lkml.kernel.org/r/20171208124939.7livp7no2ov65rrc@hirez.programming.kicks-ass.net
Pull RCU updates from Paul E. McKenney:
- Updates to use cond_resched() instead of cond_resched_rcu_qs()
where feasible (currently everywhere except in kernel/rcu and
in kernel/torture.c). Also a couple of fixes to avoid sending
IPIs to offline CPUs.
- Updates to simplify RCU's dyntick-idle handling.
- Updates to remove almost all uses of smp_read_barrier_depends()
and read_barrier_depends().
- Miscellaneous fixes.
- Torture-test updates.
Signed-off-by: Ingo Molnar <mingo@kernel.org>
This code (CONFIG_LOCKDEP_CROSSRELEASE=y and CONFIG_LOCKDEP_COMPLETIONS=y),
while it found a number of old bugs initially, was also causing too many
false positives that caused people to disable lockdep - which is arguably
a worse overall outcome.
If we disable cross-release by default but keep the code upstream then
in practice the most likely outcome is that we'll allow the situation
to degrade gradually, by allowing entropy to introduce more and more
false positives, until it overwhelms maintenance capacity.
Another bad side effect was that people were trying to work around
the false positives by uglifying/complicating unrelated code. There's
a marked difference between annotating locking operations and
uglifying good code just due to bad lock debugging code ...
This gradual decrease in quality happened to a number of debugging
facilities in the kernel, and lockdep is pretty complex already,
so we cannot risk this outcome.
Either cross-release checking can be done right with no false positives,
or it should not be included in the upstream kernel.
( Note that it might make sense to maintain it out of tree and go through
the false positives every now and then and see whether new bugs were
introduced. )
Cc: Byungchul Park <byungchul.park@lge.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
