platform/kernel/linux-rpi.git
3 years agofutex: Clarify comment in futex_requeue()
Thomas Gleixner [Sun, 15 Aug 2021 21:29:14 +0000 (23:29 +0200)]
futex: Clarify comment in futex_requeue()

The comment about the restriction of the number of waiters to wake for the
REQUEUE_PI case is confusing at best. Rewrite it.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211305.524990421@linutronix.de
3 years agofutex: Restructure futex_requeue()
Thomas Gleixner [Sun, 15 Aug 2021 21:29:12 +0000 (23:29 +0200)]
futex: Restructure futex_requeue()

No point in taking two more 'requeue_pi' conditionals just to get to the
requeue. Same for the requeue_pi case just the other way round.

No functional change.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211305.468835790@linutronix.de
3 years agofutex: Correct the number of requeued waiters for PI
Thomas Gleixner [Sun, 15 Aug 2021 21:29:10 +0000 (23:29 +0200)]
futex: Correct the number of requeued waiters for PI

The accounting is wrong when either the PI sanity check or the
requeue PI operation fails. Adjust it in the failure path.

Will be simplified in the next step.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211305.416427548@linutronix.de
3 years agofutex: Remove bogus condition for requeue PI
Thomas Gleixner [Sun, 15 Aug 2021 21:29:09 +0000 (23:29 +0200)]
futex: Remove bogus condition for requeue PI

For requeue PI it's required to establish PI state for the PI futex to
which waiters are requeued. This either acquires the user space futex on
behalf of the top most waiter on the inner 'waitqueue' futex, or attaches to
the PI state of an existing waiter, or creates on attached to the owner of
the futex.

This code can retry in case of failure, but retry can never happen when the
pi state was successfully created. The condition to run this code is:

  (task_count - nr_wake) < nr_requeue

which is always true because:

   task_count = 0
   nr_wake = 1
   nr_requeue >= 0

Remove it completely.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211305.362730187@linutronix.de
3 years agofutex: Clarify futex_requeue() PI handling
Thomas Gleixner [Sun, 15 Aug 2021 21:29:07 +0000 (23:29 +0200)]
futex: Clarify futex_requeue() PI handling

When requeuing to a PI futex, then the requeue code tries to trylock the PI
futex on behalf of the topmost waiter on the inner 'waitqueue' futex. If
that succeeds, then PI state has to be allocated in order to requeue further
waiters to the PI futex.

The comment and the code are confusing, as the PI state allocation uses
lookup_pi_state(), which either attaches to an existing waiter or to the
owner. As the PI futex was just acquired, there cannot be a waiter on the
PI futex because the hash bucket lock is held.

Clarify the comment and use attach_to_pi_owner() directly. As the task on
which behalf the PI futex has been acquired is guaranteed to be alive and
not exiting, this call must succeed. Add a WARN_ON() in case that fails.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211305.305142462@linutronix.de
3 years agofutex: Clean up stale comments
Thomas Gleixner [Sun, 15 Aug 2021 21:29:06 +0000 (23:29 +0200)]
futex: Clean up stale comments

The futex key reference mechanism is long gone. Clean up the stale comments
which still mention it.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211305.249178312@linutronix.de
3 years agofutex: Validate waiter correctly in futex_proxy_trylock_atomic()
Thomas Gleixner [Sun, 15 Aug 2021 21:29:04 +0000 (23:29 +0200)]
futex: Validate waiter correctly in futex_proxy_trylock_atomic()

The loop in futex_requeue() has a sanity check for the waiter, which is
missing in futex_proxy_trylock_atomic(). In theory the key2 check is
sufficient, but futexes are cursed so add it for completeness and paranoia
sake.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211305.193767519@linutronix.de
3 years agolib/test_lockup: Adapt to changed variables
Sebastian Andrzej Siewior [Sun, 15 Aug 2021 21:29:03 +0000 (23:29 +0200)]
lib/test_lockup: Adapt to changed variables

The inner parts of certain locks (mutex, rwlocks) changed due to a rework for
RT and non RT code. Most users remain unaffected, but those who fiddle around
in the inner parts need to be updated.

Match the struct names to the new layout.

Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211305.137982730@linutronix.de
3 years agolocking/rtmutex: Add mutex variant for RT
Thomas Gleixner [Sun, 15 Aug 2021 21:29:01 +0000 (23:29 +0200)]
locking/rtmutex: Add mutex variant for RT

Add the necessary defines, helpers and API functions for replacing struct mutex on
a PREEMPT_RT enabled kernel with an rtmutex based variant.

No functional change when CONFIG_PREEMPT_RT=n

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211305.081517417@linutronix.de
3 years agolocking/ww_mutex: Implement rtmutex based ww_mutex API functions
Peter Zijlstra [Sun, 15 Aug 2021 21:29:00 +0000 (23:29 +0200)]
locking/ww_mutex: Implement rtmutex based ww_mutex API functions

Add the actual ww_mutex API functions which replace the mutex based variant
on RT enabled kernels.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211305.024057938@linutronix.de
3 years agolocking/rtmutex: Extend the rtmutex core to support ww_mutex
Peter Zijlstra [Sun, 15 Aug 2021 21:28:58 +0000 (23:28 +0200)]
locking/rtmutex: Extend the rtmutex core to support ww_mutex

Add a ww acquire context pointer to the waiter and various functions and
add the ww_mutex related invocations to the proper spots in the locking
code, similar to the mutex based variant.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211304.966139174@linutronix.de
3 years agolocking/ww_mutex: Add rt_mutex based lock type and accessors
Peter Zijlstra [Sun, 15 Aug 2021 21:28:56 +0000 (23:28 +0200)]
locking/ww_mutex: Add rt_mutex based lock type and accessors

Provide the defines for RT mutex based ww_mutexes and fix up the debug logic
so it's either enabled by DEBUG_MUTEXES or DEBUG_RT_MUTEXES on RT kernels.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211304.908012566@linutronix.de
3 years agolocking/ww_mutex: Add RT priority to W/W order
Peter Zijlstra [Sun, 15 Aug 2021 21:28:55 +0000 (23:28 +0200)]
locking/ww_mutex: Add RT priority to W/W order

RT mutex based ww_mutexes cannot order based on timestamps. They have to
order based on priority. Add the necessary decision logic.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211304.847536630@linutronix.de
3 years agolocking/ww_mutex: Implement rt_mutex accessors
Peter Zijlstra [Sun, 15 Aug 2021 21:28:53 +0000 (23:28 +0200)]
locking/ww_mutex: Implement rt_mutex accessors

Provide the type defines and the helper inlines for rtmutex based ww_mutexes.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211304.790760545@linutronix.de
3 years agolocking/ww_mutex: Abstract out internal lock accesses
Thomas Gleixner [Sun, 15 Aug 2021 21:28:52 +0000 (23:28 +0200)]
locking/ww_mutex: Abstract out internal lock accesses

Accessing the internal wait_lock of mutex and rtmutex is slightly
different. Provide helper functions for that.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211304.734635961@linutronix.de
3 years agolocking/ww_mutex: Abstract out mutex types
Peter Zijlstra [Sun, 15 Aug 2021 21:28:50 +0000 (23:28 +0200)]
locking/ww_mutex: Abstract out mutex types

Some ww_mutex helper functions use pointers for the underlying mutex and
mutex_waiter. The upcoming rtmutex based implementation needs to share
these functions. Add and use defines for the types and replace the direct
types in the affected functions.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211304.678720245@linutronix.de
3 years agolocking/ww_mutex: Abstract out mutex accessors
Peter Zijlstra [Sun, 15 Aug 2021 21:28:49 +0000 (23:28 +0200)]
locking/ww_mutex: Abstract out mutex accessors

Move the mutex related access from various ww_mutex functions into helper
functions so they can be substituted for rtmutex based ww_mutex later.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211304.622477030@linutronix.de
3 years agolocking/ww_mutex: Abstract out waiter enqueueing
Peter Zijlstra [Sun, 15 Aug 2021 21:28:47 +0000 (23:28 +0200)]
locking/ww_mutex: Abstract out waiter enqueueing

The upcoming rtmutex based ww_mutex needs a different handling for
enqueueing a waiter. Split it out into a helper function.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211304.566318143@linutronix.de
3 years agolocking/ww_mutex: Abstract out the waiter iteration
Peter Zijlstra [Sun, 15 Aug 2021 21:28:45 +0000 (23:28 +0200)]
locking/ww_mutex: Abstract out the waiter iteration

Split out the waiter iteration functions so they can be substituted for a
rtmutex based ww_mutex later.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211304.509186185@linutronix.de
3 years agolocking/ww_mutex: Remove the __sched annotation from ww_mutex APIs
Peter Zijlstra [Sun, 15 Aug 2021 21:28:44 +0000 (23:28 +0200)]
locking/ww_mutex: Remove the __sched annotation from ww_mutex APIs

None of these functions will be on the stack when blocking in
schedule(), hence __sched is not needed.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211304.453235952@linutronix.de
3 years agolocking/ww_mutex: Split out the W/W implementation logic into kernel/locking/ww_mutex.h
Peter Zijlstra (Intel) [Tue, 17 Aug 2021 14:31:54 +0000 (16:31 +0200)]
locking/ww_mutex: Split out the W/W implementation logic into kernel/locking/ww_mutex.h

Split the W/W mutex helper functions out into a separate header file, so
they can be shared with a rtmutex based variant later.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211304.396893399@linutronix.de
3 years agolocking/ww_mutex: Split up ww_mutex_unlock()
Peter Zijlstra (Intel) [Tue, 17 Aug 2021 14:19:04 +0000 (16:19 +0200)]
locking/ww_mutex: Split up ww_mutex_unlock()

Split the ww related part out into a helper function so it can be reused
for a rtmutex based ww_mutex implementation.

[ mingo: Fixed bisection failure. ]

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211304.340166556@linutronix.de
3 years agolocking/ww_mutex: Gather mutex_waiter initialization
Peter Zijlstra [Sun, 15 Aug 2021 21:28:39 +0000 (23:28 +0200)]
locking/ww_mutex: Gather mutex_waiter initialization

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211304.281927514@linutronix.de
3 years agolocking/ww_mutex: Simplify lockdep annotations
Peter Zijlstra [Sun, 15 Aug 2021 21:28:38 +0000 (23:28 +0200)]
locking/ww_mutex: Simplify lockdep annotations

No functional change.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211304.222921634@linutronix.de
3 years agolocking/mutex: Make mutex::wait_lock raw
Thomas Gleixner [Sun, 15 Aug 2021 21:28:36 +0000 (23:28 +0200)]
locking/mutex: Make mutex::wait_lock raw

The wait_lock of mutex is really a low level lock. Convert it to a
raw_spinlock like the wait_lock of rtmutex.

[ mingo: backmerged the test_lockup.c build fix by bigeasy. ]

Co-developed-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211304.166863404@linutronix.de
3 years agolocking/ww_mutex: Move the ww_mutex definitions from <linux/mutex.h> into <linux...
Thomas Gleixner [Sun, 15 Aug 2021 21:28:34 +0000 (23:28 +0200)]
locking/ww_mutex: Move the ww_mutex definitions from <linux/mutex.h> into <linux/ww_mutex.h>

Move the ww_mutex definitions into the ww_mutex specific header where they
belong.

Preparatory change to allow compiling ww_mutexes standalone.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211304.110216293@linutronix.de
3 years agolocking/mutex: Move the 'struct mutex_waiter' definition from <linux/mutex.h> to...
Thomas Gleixner [Sun, 15 Aug 2021 21:28:33 +0000 (23:28 +0200)]
locking/mutex: Move the 'struct mutex_waiter' definition from <linux/mutex.h> to the internal header

Move the mutex waiter declaration from the public <linux/mutex.h> header
to the internal kernel/locking/mutex.h header.

There is no reason to expose it outside of the core code.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211304.054325923@linutronix.de
3 years agolocking/mutex: Consolidate core headers, remove kernel/locking/mutex-debug.h
Thomas Gleixner [Tue, 17 Aug 2021 14:17:38 +0000 (16:17 +0200)]
locking/mutex: Consolidate core headers, remove kernel/locking/mutex-debug.h

Having two header files which contain just the non-debug and debug variants
is mostly waste of disc space and has no real value. Stick the debug
variants into the common mutex.h file as counterpart to the stubs for the
non-debug case.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211303.995350521@linutronix.de
3 years agolocking/rtmutex: Squash !RT tasks to DEFAULT_PRIO
Peter Zijlstra [Sun, 15 Aug 2021 21:28:30 +0000 (23:28 +0200)]
locking/rtmutex: Squash !RT tasks to DEFAULT_PRIO

Ensure all !RT tasks have the same prio such that they end up in FIFO
order and aren't split up according to nice level.

The reason why nice levels were taken into account so far is historical. In
the early days of the rtmutex code it was done to give the PI boosting and
deboosting a larger coverage.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211303.938676930@linutronix.de
3 years agolocking/rwlock: Provide RT variant
Thomas Gleixner [Sun, 15 Aug 2021 21:28:28 +0000 (23:28 +0200)]
locking/rwlock: Provide RT variant

Similar to rw_semaphores, on RT the rwlock substitution is not writer fair,
because it's not feasible to have a writer inherit its priority to
multiple readers. Readers blocked on a writer follow the normal rules of
priority inheritance. Like RT spinlocks, RT rwlocks are state preserving
across the slow lock operations (contended case).

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211303.882793524@linutronix.de
3 years agolocking/spinlock: Provide RT variant
Thomas Gleixner [Sun, 15 Aug 2021 21:28:27 +0000 (23:28 +0200)]
locking/spinlock: Provide RT variant

Provide the actual locking functions which make use of the general and
spinlock specific rtmutex code.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211303.826621464@linutronix.de
3 years agolocking/rtmutex: Provide the spin/rwlock core lock function
Thomas Gleixner [Sun, 15 Aug 2021 21:28:25 +0000 (23:28 +0200)]
locking/rtmutex: Provide the spin/rwlock core lock function

A simplified version of the rtmutex slowlock function, which neither handles
signals nor timeouts, and is careful about preserving the state of the
blocked task across the lock operation.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211303.770228446@linutronix.de
3 years agolocking/spinlock: Provide RT variant header: <linux/spinlock_rt.h>
Thomas Gleixner [Sun, 15 Aug 2021 21:28:23 +0000 (23:28 +0200)]
locking/spinlock: Provide RT variant header: <linux/spinlock_rt.h>

Provide the necessary wrappers around the actual rtmutex based spinlock
implementation.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211303.712897671@linutronix.de
3 years agolocking/spinlock: Provide RT specific spinlock_t
Thomas Gleixner [Sun, 15 Aug 2021 21:28:22 +0000 (23:28 +0200)]
locking/spinlock: Provide RT specific spinlock_t

RT replaces spinlocks with a simple wrapper around an rtmutex, which turns
spinlocks on RT into 'sleeping' spinlocks. The actual implementation of the
spinlock API differs from a regular rtmutex, as it does neither handle
timeouts nor signals and it is state preserving across the lock operation.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211303.654230709@linutronix.de
3 years agolocking/rtmutex: Reduce <linux/rtmutex.h> header dependencies, only include <linux...
Sebastian Andrzej Siewior [Sun, 15 Aug 2021 21:28:20 +0000 (23:28 +0200)]
locking/rtmutex: Reduce <linux/rtmutex.h> header dependencies, only include <linux/rbtree_types.h>

We have the following header dependency problem on RT:

 - <linux/rtmutex.h> needs the definition of 'struct rb_root_cached'.
 - <linux/rbtree.h> includes <linux/kernel.h>, which includes <linux/spinlock.h>

That works nicely for non-RT enabled kernels, but on RT enabled kernels
spinlocks are based on rtmutexes, which creates another circular header
dependency as <linux/spinlocks.h> will require <linux/rtmutex.h>.

Include <linux/rbtree_types.h> instead.

Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211303.598003167@linutronix.de
3 years agorbtree: Split out the rbtree type definitions into <linux/rbtree_types.h>
Sebastian Andrzej Siewior [Sun, 15 Aug 2021 21:28:19 +0000 (23:28 +0200)]
rbtree: Split out the rbtree type definitions into <linux/rbtree_types.h>

So we have this header dependency problem on RT:

 - <linux/rtmutex.h> needs the definition of 'struct rb_root_cached'.
 - <linux/rbtree.h> includes <linux/kernel.h>, which includes <linux/spinlock.h>.

That works nicely for non-RT enabled kernels, but on RT enabled kernels
spinlocks are based on rtmutexes, which creates another circular header
dependency, as <linux/spinlocks.h> will require <linux/rtmutex.h>.

Split out the type definitions and move them into their own header file so
the rtmutex header can include just those.

Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211303.542123501@linutronix.de
3 years agolocking/lockdep: Reduce header dependencies in <linux/debug_locks.h>
Sebastian Andrzej Siewior [Sun, 15 Aug 2021 21:28:17 +0000 (23:28 +0200)]
locking/lockdep: Reduce header dependencies in <linux/debug_locks.h>

The inclusion of printk.h leads to a circular dependency if spinlock_t is
based on rtmutexes on RT enabled kernels.

Include only atomic.h (xchg()) and cache.h (__read_mostly) which is all
what debug_locks.h requires.

Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211303.484161136@linutronix.de
3 years agolocking/rtmutex: Prevent future include recursion hell
Sebastian Andrzej Siewior [Sun, 15 Aug 2021 21:28:16 +0000 (23:28 +0200)]
locking/rtmutex: Prevent future include recursion hell

rtmutex only needs raw_spinlock_t, but it includes spinlock_types.h, which
is not a problem on an non RT enabled kernel.

RT kernels substitute regular spinlocks with 'sleeping' spinlocks, which
are based on rtmutexes, and therefore must be able to include rtmutex.h.

Include <linux/spinlock_types_raw.h> instead.

Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211303.428224188@linutronix.de
3 years agolocking/spinlock: Split the lock types header, and move the raw types into <linux...
Thomas Gleixner [Sun, 15 Aug 2021 21:28:14 +0000 (23:28 +0200)]
locking/spinlock: Split the lock types header, and move the raw types into <linux/spinlock_types_raw.h>

Move raw_spinlock into its own file. Prepare for RT 'sleeping spinlocks', to
avoid header recursion, as RT locks require rtmutex.h, which in turn requires
the raw spinlock types.

No functional change.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211303.371269088@linutronix.de
3 years agolocking/rtmutex: Guard regular sleeping locks specific functions
Thomas Gleixner [Sun, 15 Aug 2021 21:28:12 +0000 (23:28 +0200)]
locking/rtmutex: Guard regular sleeping locks specific functions

Guard the regular sleeping lock specific functionality, which is used for
rtmutex on non-RT enabled kernels and for mutex, rtmutex and semaphores on
RT enabled kernels so the code can be reused for the RT specific
implementation of spinlocks and rwlocks in a different compilation unit.

No functional change.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211303.311535693@linutronix.de
3 years agolocking/rtmutex: Prepare RT rt_mutex_wake_q for RT locks
Thomas Gleixner [Sun, 15 Aug 2021 21:28:11 +0000 (23:28 +0200)]
locking/rtmutex: Prepare RT rt_mutex_wake_q for RT locks

Add an rtlock_task pointer to rt_mutex_wake_q, which allows to handle the RT
specific wakeup for spin/rwlock waiters. The pointer is just consuming 4/8
bytes on the stack so it is provided unconditionaly to avoid #ifdeffery all
over the place.

This cannot use a regular wake_q, because a task can have concurrent wakeups which
would make it miss either lock or the regular wakeups, depending on what gets
queued first, unless task struct gains a separate wake_q_node for this, which
would be overkill, because there can only be a single task which gets woken
up in the spin/rw_lock unlock path.

No functional change for non-RT enabled kernels.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211303.253614678@linutronix.de
3 years agolocking/rtmutex: Use rt_mutex_wake_q_head
Thomas Gleixner [Sun, 15 Aug 2021 21:28:09 +0000 (23:28 +0200)]
locking/rtmutex: Use rt_mutex_wake_q_head

Prepare for the required state aware handling of waiter wakeups via wake_q
and switch the rtmutex code over to the rtmutex specific wrapper.

No functional change.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211303.197113263@linutronix.de
3 years agolocking/rtmutex: Provide rt_wake_q_head and helpers
Thomas Gleixner [Sun, 15 Aug 2021 21:28:08 +0000 (23:28 +0200)]
locking/rtmutex: Provide rt_wake_q_head and helpers

To handle the difference between wakeups for regular sleeping locks (mutex,
rtmutex, rw_semaphore) and the wakeups for 'sleeping' spin/rwlocks on
PREEMPT_RT enabled kernels correctly, it is required to provide a
wake_q_head construct which allows to keep them separate.

Provide a wrapper around wake_q_head and the required helpers, which will be
extended with the state handling later.

No functional change.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211303.139337655@linutronix.de
3 years agolocking/rtmutex: Add wake_state to rt_mutex_waiter
Thomas Gleixner [Sun, 15 Aug 2021 21:28:06 +0000 (23:28 +0200)]
locking/rtmutex: Add wake_state to rt_mutex_waiter

Regular sleeping locks like mutexes, rtmutexes and rw_semaphores are always
entering and leaving a blocking section with task state == TASK_RUNNING.

On a non-RT kernel spinlocks and rwlocks never affect the task state, but
on RT kernels these locks are converted to rtmutex based 'sleeping' locks.

So in case of contention the task goes to block, which requires to carefully
preserve the task state, and restore it after acquiring the lock taking
regular wakeups for the task into account, which happened while the task was
blocked. This state preserving is achieved by having a separate task state
for blocking on a RT spin/rwlock and a saved_state field in task_struct
along with careful handling of these wakeup scenarios in try_to_wake_up().

To avoid conditionals in the rtmutex code, store the wake state which has
to be used for waking a lock waiter in rt_mutex_waiter which allows to
handle the regular and RT spin/rwlocks by handing it to wake_up_state().

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211303.079800739@linutronix.de
3 years agolocking/rwsem: Add rtmutex based R/W semaphore implementation
Thomas Gleixner [Sun, 15 Aug 2021 21:28:05 +0000 (23:28 +0200)]
locking/rwsem: Add rtmutex based R/W semaphore implementation

The RT specific R/W semaphore implementation used to restrict the number of
readers to one, because a writer cannot block on multiple readers and
inherit its priority or budget.

The single reader restricting was painful in various ways:

 - Performance bottleneck for multi-threaded applications in the page fault
   path (mmap sem)

 - Progress blocker for drivers which are carefully crafted to avoid the
   potential reader/writer deadlock in mainline.

The analysis of the writer code paths shows that properly written RT tasks
should not take them. Syscalls like mmap(), file access which take mmap sem
write locked have unbound latencies, which are completely unrelated to mmap
sem. Other R/W sem users like graphics drivers are not suitable for RT tasks
either.

So there is little risk to hurt RT tasks when the RT rwsem implementation is
done in the following way:

 - Allow concurrent readers

 - Make writers block until the last reader left the critical section. This
   blocking is not subject to priority/budget inheritance.

 - Readers blocked on a writer inherit their priority/budget in the normal
   way.

There is a drawback with this scheme: R/W semaphores become writer unfair
though the applications which have triggered writer starvation (mostly on
mmap_sem) in the past are not really the typical workloads running on a RT
system. So while it's unlikely to hit writer starvation, it's possible. If
there are unexpected workloads on RT systems triggering it, the problem
has to be revisited.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211303.016885947@linutronix.de
3 years agolocking/rt: Add base code for RT rw_semaphore and rwlock
Thomas Gleixner [Sun, 15 Aug 2021 21:28:03 +0000 (23:28 +0200)]
locking/rt: Add base code for RT rw_semaphore and rwlock

On PREEMPT_RT, rw_semaphores and rwlocks are substituted with an rtmutex and
a reader count. The implementation is writer unfair, as it is not feasible
to do priority inheritance on multiple readers, but experience has shown
that real-time workloads are not the typical workloads which are sensitive
to writer starvation.

The inner workings of rw_semaphores and rwlocks on RT are almost identical
except for the task state and signal handling. rw_semaphores are not state
preserving over a contention, they are expected to enter and leave with state
== TASK_RUNNING. rwlocks have a mechanism to preserve the state of the task
at entry and restore it after unblocking taking potential non-lock related
wakeups into account. rw_semaphores can also be subject to signal handling
interrupting a blocked state, while rwlocks ignore signals.

To avoid code duplication, provide a shared implementation which takes the
small difference vs. state and signals into account. The code is included
into the relevant rw_semaphore/rwlock base code and compiled for each use
case separately.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211302.957920571@linutronix.de
3 years agolocking/rtmutex: Provide rt_mutex_base_is_locked()
Thomas Gleixner [Sun, 15 Aug 2021 21:28:02 +0000 (23:28 +0200)]
locking/rtmutex: Provide rt_mutex_base_is_locked()

Provide rt_mutex_base_is_locked(), which will be used for various wrapped
locking primitives for RT.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211302.899572818@linutronix.de
3 years agolocking/rtmutex: Provide rt_mutex_slowlock_locked()
Thomas Gleixner [Sun, 15 Aug 2021 21:28:00 +0000 (23:28 +0200)]
locking/rtmutex: Provide rt_mutex_slowlock_locked()

Split the inner workings of rt_mutex_slowlock() out into a separate
function, which can be reused by the upcoming RT lock substitutions,
e.g. for rw_semaphores.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211302.841971086@linutronix.de
3 years agolocking/rtmutex: Split out the inner parts of 'struct rtmutex'
Peter Zijlstra [Sun, 15 Aug 2021 21:27:58 +0000 (23:27 +0200)]
locking/rtmutex: Split out the inner parts of 'struct rtmutex'

RT builds substitutions for rwsem, mutex, spinlock and rwlock around
rtmutexes. Split the inner working out so each lock substitution can use
them with the appropriate lockdep annotations. This avoids having an extra
unused lockdep map in the wrapped rtmutex.

No functional change.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211302.784739994@linutronix.de
3 years agolocking/rtmutex: Split API from implementation
Thomas Gleixner [Sun, 15 Aug 2021 21:27:57 +0000 (23:27 +0200)]
locking/rtmutex: Split API from implementation

Prepare for reusing the inner functions of rtmutex for RT lock
substitutions: introduce kernel/locking/rtmutex_api.c and move
them there.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211302.726560996@linutronix.de
3 years agolocking/rtmutex: Switch to from cmpxchg_*() to try_cmpxchg_*()
Thomas Gleixner [Sun, 15 Aug 2021 21:27:55 +0000 (23:27 +0200)]
locking/rtmutex: Switch to from cmpxchg_*() to try_cmpxchg_*()

Allows the compiler to generate better code depending on the architecture.

Suggested-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211302.668958502@linutronix.de
3 years agolocking/rtmutex: Convert macros to inlines
Sebastian Andrzej Siewior [Sun, 15 Aug 2021 21:27:54 +0000 (23:27 +0200)]
locking/rtmutex: Convert macros to inlines

Inlines are type-safe...

Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211302.610830960@linutronix.de
3 years agolocking/rtmutex: Remove rt_mutex_is_locked()
Peter Zijlstra [Sun, 15 Aug 2021 21:27:52 +0000 (23:27 +0200)]
locking/rtmutex: Remove rt_mutex_is_locked()

There are no more users left.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211302.552218335@linutronix.de
3 years agomedia/atomisp: Use lockdep instead of *mutex_is_locked()
Peter Zijlstra [Sun, 15 Aug 2021 21:27:51 +0000 (23:27 +0200)]
media/atomisp: Use lockdep instead of *mutex_is_locked()

The only user of rt_mutex_is_locked() is an anti-pattern, remove it.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211302.491442626@linutronix.de
3 years agosched/wake_q: Provide WAKE_Q_HEAD_INITIALIZER()
Thomas Gleixner [Sun, 15 Aug 2021 21:27:49 +0000 (23:27 +0200)]
sched/wake_q: Provide WAKE_Q_HEAD_INITIALIZER()

The RT specific spin/rwlock implementation requires special handling of the
to be woken waiters. Provide a WAKE_Q_HEAD_INITIALIZER(), which can be used by
the rtmutex code to implement an RT aware wake_q derivative.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211302.429918071@linutronix.de
3 years agosched/core: Provide a scheduling point for RT locks
Thomas Gleixner [Sun, 15 Aug 2021 21:27:48 +0000 (23:27 +0200)]
sched/core: Provide a scheduling point for RT locks

RT enabled kernels substitute spin/rwlocks with 'sleeping' variants based
on rtmutexes. Blocking on such a lock is similar to preemption versus:

 - I/O scheduling and worker handling, because these functions might block
   on another substituted lock, or come from a lock contention within these
   functions.

 - RCU considers this like a preemption, because the task might be in a read
   side critical section.

Add a separate scheduling point for this, and hand a new scheduling mode
argument to __schedule() which allows, along with separate mode masks, to
handle this gracefully from within the scheduler, without proliferating that
to other subsystems like RCU.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211302.372319055@linutronix.de
3 years agosched/core: Rework the __schedule() preempt argument
Thomas Gleixner [Sun, 15 Aug 2021 21:27:46 +0000 (23:27 +0200)]
sched/core: Rework the __schedule() preempt argument

PREEMPT_RT needs to hand a special state into __schedule() when a task
blocks on a 'sleeping' spin/rwlock. This is required to handle
rcu_note_context_switch() correctly without having special casing in the
RCU code. From an RCU point of view the blocking on the sleeping spinlock
is equivalent to preemption, because the task might be in a read side
critical section.

schedule_debug() also has a check which would trigger with the !preempt
case, but that could be handled differently.

To avoid adding another argument and extra checks which cannot be optimized
out by the compiler, the following solution has been chosen:

 - Replace the boolean 'preempt' argument with an unsigned integer
   'sched_mode' argument and define constants to hand in:
   (0 == no preemption, 1 = preemption).

 - Add two masks to apply on that mode: one for the debug/rcu invocations,
   and one for the actual scheduling decision.

   For a non RT kernel these masks are UINT_MAX, i.e. all bits are set,
   which allows the compiler to optimize the AND operation out, because it is
   not masking out anything. IOW, it's not different from the boolean.

   RT enabled kernels will define these masks separately.

No functional change.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211302.315473019@linutronix.de
3 years agosched/wakeup: Prepare for RT sleeping spin/rwlocks
Thomas Gleixner [Sun, 15 Aug 2021 21:27:44 +0000 (23:27 +0200)]
sched/wakeup: Prepare for RT sleeping spin/rwlocks

Waiting for spinlocks and rwlocks on non RT enabled kernels is task::state
preserving. Any wakeup which matches the state is valid.

RT enabled kernels substitutes them with 'sleeping' spinlocks. This creates
an issue vs. task::__state.

In order to block on the lock, the task has to overwrite task::__state and a
consecutive wakeup issued by the unlocker sets the state back to
TASK_RUNNING. As a consequence the task loses the state which was set
before the lock acquire and also any regular wakeup targeted at the task
while it is blocked on the lock.

To handle this gracefully, add a 'saved_state' member to task_struct which
is used in the following way:

 1) When a task blocks on a 'sleeping' spinlock, the current state is saved
    in task::saved_state before it is set to TASK_RTLOCK_WAIT.

 2) When the task unblocks and after acquiring the lock, it restores the saved
    state.

 3) When a regular wakeup happens for a task while it is blocked then the
    state change of that wakeup is redirected to operate on task::saved_state.

    This is also required when the task state is running because the task
    might have been woken up from the lock wait and has not yet restored
    the saved state.

To make it complete, provide the necessary helpers to save and restore the
saved state along with the necessary documentation how the RT lock blocking
is supposed to work.

For non-RT kernels there is no functional change.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211302.258751046@linutronix.de
3 years agosched/wakeup: Reorganize the current::__state helpers
Thomas Gleixner [Sun, 15 Aug 2021 21:27:43 +0000 (23:27 +0200)]
sched/wakeup: Reorganize the current::__state helpers

In order to avoid more duplicate implementations for the debug and
non-debug variants of the state change macros, split the debug portion out
and make that conditional on CONFIG_DEBUG_ATOMIC_SLEEP=y.

Suggested-by: Waiman Long <longman@redhat.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211302.200898048@linutronix.de
Signed-off-by: Ingo Molnar <mingo@kernel.org>
3 years agosched/wakeup: Introduce the TASK_RTLOCK_WAIT state bit
Thomas Gleixner [Sun, 15 Aug 2021 21:27:41 +0000 (23:27 +0200)]
sched/wakeup: Introduce the TASK_RTLOCK_WAIT state bit

RT kernels have an extra quirk for try_to_wake_up() to handle task state
preservation across periods of blocking on a 'sleeping' spin/rwlock.

For this to function correctly and under all circumstances try_to_wake_up()
must be able to identify whether the wakeup is lock related or not and
whether the task is waiting for a lock or not.

The original approach was to use a special wake_flag argument for
try_to_wake_up() and just use TASK_UNINTERRUPTIBLE for the tasks wait state
and the try_to_wake_up() state argument.

This works in principle, but due to the fact that try_to_wake_up() cannot
determine whether the task is waiting for an RT lock wakeup or for a regular
wakeup it's suboptimal.

RT kernels save the original task state when blocking on an RT lock and
restore it when the lock has been acquired. Any non lock related wakeup is
checked against the saved state and if it matches the saved state is set to
running so that the wakeup is not lost when the state is restored.

While the necessary logic for the wake_flag based solution is trivial, the
downside is that any regular wakeup with TASK_UNINTERRUPTIBLE in the state
argument set will wake the task despite the fact that it is still blocked
on the lock. That's not a fatal problem as the lock wait has do deal with
spurious wakeups anyway, but it introduces unnecessary latencies.

Introduce the TASK_RTLOCK_WAIT state bit which will be set when a task
blocks on an RT lock.

The lock wakeup will use wake_up_state(TASK_RTLOCK_WAIT), so both the
waiting state and the wakeup state are distinguishable, which avoids
spurious wakeups and allows better analysis.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211302.144989915@linutronix.de
Signed-off-by: Ingo Molnar <mingo@kernel.org>
3 years agosched/wakeup: Split out the wakeup ->__state check
Thomas Gleixner [Sun, 15 Aug 2021 21:27:40 +0000 (23:27 +0200)]
sched/wakeup: Split out the wakeup ->__state check

RT kernels have a slightly more complicated handling of wakeups due to
'sleeping' spin/rwlocks. If a task is blocked on such a lock then the
original state of the task is preserved over the blocking period, and
any regular (non lock related) wakeup has to be targeted at the
saved state to ensure that these wakeups are not lost.

Once the task acquires the lock it restores the task state from the saved state.

To avoid cluttering try_to_wake_up() with that logic, split the wakeup
state check out into an inline helper and use it at both places where
task::__state is checked against the state argument of try_to_wake_up().

No functional change.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211302.088945085@linutronix.de
3 years agolocking/rtmutex: Set proper wait context for lockdep
Thomas Gleixner [Sun, 15 Aug 2021 21:27:38 +0000 (23:27 +0200)]
locking/rtmutex: Set proper wait context for lockdep

RT mutexes belong to the LD_WAIT_SLEEP class. Make them so.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211302.031014562@linutronix.de
3 years agolocking/local_lock: Add missing owner initialization
Thomas Gleixner [Sun, 15 Aug 2021 21:27:37 +0000 (23:27 +0200)]
locking/local_lock: Add missing owner initialization

If CONFIG_DEBUG_LOCK_ALLOC=y is enabled then local_lock_t has an 'owner'
member which is checked for consistency, but nothing initialized it to
zero explicitly.

The static initializer does so implicit, and the run time allocated per CPU
storage is usually zero initialized as well, but relying on that is not
really good practice.

Fixes: 91710728d172 ("locking: Introduce local_lock()")
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20210815211301.969975279@linutronix.de
3 years agoMerge tag 'v5.14-rc6' into locking/core, to pick up fixes
Ingo Molnar [Tue, 17 Aug 2021 14:16:29 +0000 (16:16 +0200)]
Merge tag 'v5.14-rc6' into locking/core, to pick up fixes

Signed-off-by: Ingo Molnar <mingo@kernel.org>
3 years agoLinux 5.14-rc6
Linus Torvalds [Sun, 15 Aug 2021 23:40:53 +0000 (13:40 -1000)]
Linux 5.14-rc6

3 years agoMerge tag 'powerpc-5.14-5' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc...
Linus Torvalds [Sun, 15 Aug 2021 16:57:43 +0000 (06:57 -1000)]
Merge tag 'powerpc-5.14-5' of git://git./linux/kernel/git/powerpc/linux

Pull powerpc fixes from Michael Ellerman:

 - Fix crashes coming out of nap on 32-bit Book3s (eg. powerbooks).

 - Fix critical and debug interrupts on BookE, seen as crashes when
   using ptrace.

 - Fix an oops when running an SMP kernel on a UP system.

 - Update pseries LPAR security flavor after partition migration.

 - Fix an oops when using kprobes on BookE.

 - Fix oops on 32-bit pmac by not calling do_IRQ() from
   timer_interrupt().

 - Fix softlockups on CPU hotplug into a CPU-less node with xive (P9).

Thanks to Cédric Le Goater, Christophe Leroy, Finn Thain, Geetika
Moolchandani, Laurent Dufour, Laurent Vivier, Nicholas Piggin, Pu Lehui,
Radu Rendec, Srikar Dronamraju, and Stan Johnson.

* tag 'powerpc-5.14-5' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc/linux:
  powerpc/xive: Do not skip CPU-less nodes when creating the IPIs
  powerpc/interrupt: Do not call single_step_exception() from other exceptions
  powerpc/interrupt: Fix OOPS by not calling do_IRQ() from timer_interrupt()
  powerpc/kprobes: Fix kprobe Oops happens in booke
  powerpc/pseries: Fix update of LPAR security flavor after LPM
  powerpc/smp: Fix OOPS in topology_init()
  powerpc/32: Fix critical and debug interrupts on BOOKE
  powerpc/32s: Fix napping restore in data storage interrupt (DSI)

3 years agoMerge tag 'irq-urgent-2021-08-15' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sun, 15 Aug 2021 16:49:40 +0000 (06:49 -1000)]
Merge tag 'irq-urgent-2021-08-15' of git://git./linux/kernel/git/tip/tip

Pull irq fixes from Thomas Gleixner:
 "A set of fixes for PCI/MSI and x86 interrupt startup:

   - Mask all MSI-X entries when enabling MSI-X otherwise stale unmasked
     entries stay around e.g. when a crashkernel is booted.

   - Enforce masking of a MSI-X table entry when updating it, which
     mandatory according to speification

   - Ensure that writes to MSI[-X} tables are flushed.

   - Prevent invalid bits being set in the MSI mask register

   - Properly serialize modifications to the mask cache and the mask
     register for multi-MSI.

   - Cure the violation of the affinity setting rules on X86 during
     interrupt startup which can cause lost and stale interrupts. Move
     the initial affinity setting ahead of actualy enabling the
     interrupt.

   - Ensure that MSI interrupts are completely torn down before freeing
     them in the error handling case.

   - Prevent an array out of bounds access in the irq timings code"

* tag 'irq-urgent-2021-08-15' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  driver core: Add missing kernel doc for device::msi_lock
  genirq/msi: Ensure deactivation on teardown
  genirq/timings: Prevent potential array overflow in __irq_timings_store()
  x86/msi: Force affinity setup before startup
  x86/ioapic: Force affinity setup before startup
  genirq: Provide IRQCHIP_AFFINITY_PRE_STARTUP
  PCI/MSI: Protect msi_desc::masked for multi-MSI
  PCI/MSI: Use msi_mask_irq() in pci_msi_shutdown()
  PCI/MSI: Correct misleading comments
  PCI/MSI: Do not set invalid bits in MSI mask
  PCI/MSI: Enforce MSI[X] entry updates to be visible
  PCI/MSI: Enforce that MSI-X table entry is masked for update
  PCI/MSI: Mask all unused MSI-X entries
  PCI/MSI: Enable and mask MSI-X early

3 years agoMerge tag 'locking_urgent_for_v5.14_rc6' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 15 Aug 2021 16:46:04 +0000 (06:46 -1000)]
Merge tag 'locking_urgent_for_v5.14_rc6' of git://git./linux/kernel/git/tip/tip

Pull locking fix from Borislav Petkov:

 - Fix a CONFIG symbol's spelling

* tag 'locking_urgent_for_v5.14_rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  locking/rtmutex: Use the correct rtmutex debugging config option

3 years agoMerge tag 'efi_urgent_for_v5.14_rc6' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 15 Aug 2021 16:38:26 +0000 (06:38 -1000)]
Merge tag 'efi_urgent_for_v5.14_rc6' of git://git./linux/kernel/git/tip/tip

Pull EFI fixes from Borislav Petkov:
 "A batch of fixes for the arm64 stub image loader:

   - fix a logic bug that can make the random page allocator fail
     spuriously

   - force reallocation of the Image when it overlaps with firmware
     reserved memory regions

   - fix an oversight that defeated on optimization introduced earlier
     where images loaded at a suitable offset are never moved if booting
     without randomization

   - complain about images that were not loaded at the right offset by
     the firmware image loader"

* tag 'efi_urgent_for_v5.14_rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  efi/libstub: arm64: Double check image alignment at entry
  efi/libstub: arm64: Warn when efi_random_alloc() fails
  efi/libstub: arm64: Relax 2M alignment again for relocatable kernels
  efi/libstub: arm64: Force Image reallocation if BSS was not reserved
  arm64: efi: kaslr: Fix occasional random alloc (and boot) failure

3 years agoMerge tag 'x86_urgent_for_v5.14_rc6' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 15 Aug 2021 16:30:24 +0000 (06:30 -1000)]
Merge tag 'x86_urgent_for_v5.14_rc6' of git://git./linux/kernel/git/tip/tip

Pull x86 fixes from Borislav Petkov:
 "Two fixes:

   - An objdump checker fix to ignore parenthesized strings in the
     objdump version

   - Fix resctrl default monitoring groups reporting when new subgroups
     get created"

* tag 'x86_urgent_for_v5.14_rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  x86/resctrl: Fix default monitoring groups reporting
  x86/tools: Fix objdump version check again

3 years agoMerge tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm
Linus Torvalds [Sun, 15 Aug 2021 16:21:30 +0000 (06:21 -1000)]
Merge tag 'for-linus' of git://git./virt/kvm/kvm

Pull KVM fixes from Paolo Bonzini:
 "ARM:

   - Plug race between enabling MTE and creating vcpus

   - Fix off-by-one bug when checking whether an address range is RAM

  x86:

   - Fixes for the new MMU, especially a memory leak on hosts with <39
     physical address bits

   - Remove bogus EFER.NX checks on 32-bit non-PAE hosts

   - WAITPKG fix"

* tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm:
  KVM: x86/mmu: Protect marking SPs unsync when using TDP MMU with spinlock
  KVM: x86/mmu: Don't step down in the TDP iterator when zapping all SPTEs
  KVM: x86/mmu: Don't leak non-leaf SPTEs when zapping all SPTEs
  KVM: nVMX: Use vmx_need_pf_intercept() when deciding if L0 wants a #PF
  kvm: vmx: Sync all matching EPTPs when injecting nested EPT fault
  KVM: x86: remove dead initialization
  KVM: x86: Allow guest to set EFER.NX=1 on non-PAE 32-bit kernels
  KVM: VMX: Use current VMCS to query WAITPKG support for MSR emulation
  KVM: arm64: Fix race when enabling KVM_ARM_CAP_MTE
  KVM: arm64: Fix off-by-one in range_is_memory

3 years agoMerge tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi
Linus Torvalds [Sun, 15 Aug 2021 05:51:58 +0000 (19:51 -1000)]
Merge tag 'scsi-fixes' of git://git./linux/kernel/git/jejb/scsi

Pull SCSI fixes from James Bottomley:
 "Three minor fixes, all in drivers"

* tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi:
  scsi: mpt3sas: Fix incorrectly assigned error return and check
  scsi: storvsc: Log TEST_UNIT_READY errors as warnings
  scsi: lpfc: Move initialization of phba->poll_list earlier to avoid crash

3 years agoMerge tag 'libnvdimm-fixes-5.14-rc6' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 15 Aug 2021 05:46:39 +0000 (19:46 -1000)]
Merge tag 'libnvdimm-fixes-5.14-rc6' of git://git./linux/kernel/git/nvdimm/nvdimm

Pull libnvdimm fixes from Dan Williams:
 "A couple of fixes for long standing bugs, a warning fixup, and some
  miscellaneous dax cleanups.

  The bugs were recently found due to new platforms looking to use the
  ACPI NFIT "virtual" device definition, and new error injection
  capabilities to trigger error responses to label area requests. Ira's
  cleanups have been long pending, I neglected to send them earlier, and
  see no harm in including them now. This has all appeared in -next with
  no reported issues.

  Summary:

   - Fix support for NFIT "virtual" ranges (BIOS-defined memory disks)

   - Fix recovery from failed label storage areas on NVDIMM devices

   - Miscellaneous cleanups from Ira's investigation of
     dax_direct_access paths preparing for stray-write protection"

* tag 'libnvdimm-fixes-5.14-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/nvdimm/nvdimm:
  tools/testing/nvdimm: Fix missing 'fallthrough' warning
  libnvdimm/region: Fix label activation vs errors
  ACPI: NFIT: Fix support for virtual SPA ranges
  dax: Ensure errno is returned from dax_direct_access
  fs/dax: Clarify nr_pages to dax_direct_access()
  fs/fuse: Remove unneeded kaddr parameter

3 years agoMerge tag 'usb-5.14-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/usb
Linus Torvalds [Sun, 15 Aug 2021 05:22:33 +0000 (19:22 -1000)]
Merge tag 'usb-5.14-rc6' of git://git./linux/kernel/git/gregkh/usb

Pull USB fix from Greg KH:
 "A single revert of a commit that caused problems in 5.14-rc5 for
  5.14-rc6. It has been in linux-next almost all week, and has resolved
  the issues that were reported on lots of different systems that were
  not the platform that the change was originally tested on (gotta love
  SoC cores used in multiple devices from multiple vendors...)"

* tag 'usb-5.14-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/usb:
  Revert "usb: dwc3: gadget: Use list_replace_init() before traversing lists"

3 years agoMerge tag 'staging-5.14-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh...
Linus Torvalds [Sun, 15 Aug 2021 05:16:30 +0000 (19:16 -1000)]
Merge tag 'staging-5.14-rc6' of git://git./linux/kernel/git/gregkh/staging

Pull IIO driver fixes from Greg KH:
 "Here are some small IIO driver fixes for reported problems for
  5.14-rc6 (no staging driver fixes at the moment).

  All of them resolve reported issues and have been in linux-next all
  week with no reported problems. Full details are in the shortlog"

* tag 'staging-5.14-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/staging:
  iio: adc: Fix incorrect exit of for-loop
  iio: humidity: hdc100x: Add margin to the conversion time
  dt-bindings: iio: st: Remove wrong items length check
  iio: accel: fxls8962af: fix i2c dependency
  iio: adis: set GPIO reset pin direction
  iio: adc: ti-ads7950: Ensure CS is deasserted after reading channels
  iio: accel: fxls8962af: fix potential use of uninitialized symbol

3 years agoMerge branch 'i2c/for-current' of git://git.kernel.org/pub/scm/linux/kernel/git/wsa...
Linus Torvalds [Sun, 15 Aug 2021 04:59:53 +0000 (18:59 -1000)]
Merge branch 'i2c/for-current' of git://git./linux/kernel/git/wsa/linux

Pull i2c fixes from Wolfram Sang:
 "One driver bugfix, a documentation bugfix, and an "uninitialized data"
  leak fix for the core"

* 'i2c/for-current' of git://git.kernel.org/pub/scm/linux/kernel/git/wsa/linux:
  Documentation: i2c: add i2c-sysfs into index
  i2c: dev: zero out array used for i2c reads from userspace
  i2c: iproc: fix race between client unreg and tasklet

3 years agoMerge tag 'for-linus-5.14-rc6-tag' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sat, 14 Aug 2021 16:31:22 +0000 (06:31 -1000)]
Merge tag 'for-linus-5.14-rc6-tag' of git://git./linux/kernel/git/xen/tip

Pull xen fixes from Juergen Gross:
 "A small cleanup patch and a fix of a rare race in the Xen evtchn
  driver"

* tag 'for-linus-5.14-rc6-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/xen/tip:
  xen/events: Fix race in set_evtchn_to_irq
  xen/events: remove redundant initialization of variable irq

3 years agoMerge tag 'riscv-for-linus-5.14-rc6' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sat, 14 Aug 2021 16:28:19 +0000 (06:28 -1000)]
Merge tag 'riscv-for-linus-5.14-rc6' of git://git./linux/kernel/git/riscv/linux

Pull RISC-V fixes from Palmer Dabbelt:

 - avoid passing -mno-relax to compilers that don't support it

 - a comment fix

* tag 'riscv-for-linus-5.14-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/riscv/linux:
  riscv: Fix comment regarding kernel mapping overlapping with IS_ERR_VALUE
  riscv: kexec: do not add '-mno-relax' flag if compiler doesn't support it

3 years agoMerge tag 'configfs-5.14' of git://git.infradead.org/users/hch/configfs
Linus Torvalds [Sat, 14 Aug 2021 16:22:42 +0000 (06:22 -1000)]
Merge tag 'configfs-5.14' of git://git.infradead.org/users/hch/configfs

Pull configfs fix from Christoph Hellwig:

 - fix to revert to the historic write behavior (Bart Van Assche)

* tag 'configfs-5.14' of git://git.infradead.org/users/hch/configfs:
  configfs: restore the kernel v5.13 text attribute write behavior

3 years agoMerge branch 'akpm' (patches from Andrew)
Linus Torvalds [Sat, 14 Aug 2021 01:05:23 +0000 (15:05 -1000)]
Merge branch 'akpm' (patches from Andrew)

Merge misc fixes from Andrew Morton:
 "7 patches.

  Subsystems affected by this patch series: mm (kasan, mm/slub,
  mm/madvise, and memcg), and lib"

* emailed patches from Andrew Morton <akpm@linux-foundation.org>:
  lib: use PFN_PHYS() in devmem_is_allowed()
  mm/memcg: fix incorrect flushing of lruvec data in obj_stock
  mm/madvise: report SIGBUS as -EFAULT for MADV_POPULATE_(READ|WRITE)
  mm: slub: fix slub_debug disabling for list of slabs
  slub: fix kmalloc_pagealloc_invalid_free unit test
  kasan, slub: reset tag when printing address
  kasan, kmemleak: reset tags when scanning block

3 years agoMerge tag '5.14-rc5-smb3-fixes' of git://git.samba.org/sfrench/cifs-2.6
Linus Torvalds [Sat, 14 Aug 2021 00:44:32 +0000 (14:44 -1000)]
Merge tag '5.14-rc5-smb3-fixes' of git://git.samba.org/sfrench/cifs-2.6

Pull cifs fixes from Steve French:
 "Four CIFS/SMB3 Fixes, all for stable, two relating to deferred close,
  and one for the 'modefromsid' mount option (when 'idsfromsid' not
  specified)"

* tag '5.14-rc5-smb3-fixes' of git://git.samba.org/sfrench/cifs-2.6:
  cifs: Call close synchronously during unlink/rename/lease break.
  cifs: Handle race conditions during rename
  cifs: use the correct max-length for dentry_path_raw()
  cifs: create sd context must be a multiple of 8

3 years agoMerge tag 'linux-kselftest-fixes-5.14-rc6' of git://git.kernel.org/pub/scm/linux...
Linus Torvalds [Sat, 14 Aug 2021 00:32:38 +0000 (14:32 -1000)]
Merge tag 'linux-kselftest-fixes-5.14-rc6' of git://git./linux/kernel/git/shuah/linux-kselftest

Pull Kselftest fix from Shuah Khan:
 "A single patch to sgx test to fix Q1 and Q2 calculation"

* tag 'linux-kselftest-fixes-5.14-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/shuah/linux-kselftest:
  selftests/sgx: Fix Q1 and Q2 calculation in sigstruct.c

3 years agolib: use PFN_PHYS() in devmem_is_allowed()
Liang Wang [Fri, 13 Aug 2021 23:54:45 +0000 (16:54 -0700)]
lib: use PFN_PHYS() in devmem_is_allowed()

The physical address may exceed 32 bits on 32-bit systems with more than
32 bits of physcial address.  Use PFN_PHYS() in devmem_is_allowed(), or
the physical address may overflow and be truncated.

We found this bug when mapping a high addresses through devmem tool,
when CONFIG_STRICT_DEVMEM is enabled on the ARM with ARM_LPAE and devmem
is used to map a high address that is not in the iomem address range, an
unexpected error indicating no permission is returned.

This bug was initially introduced from v2.6.37, and the function was
moved to lib in v5.11.

Link: https://lkml.kernel.org/r/20210731025057.78825-1-wangliang101@huawei.com
Fixes: 087aaffcdf9c ("ARM: implement CONFIG_STRICT_DEVMEM by disabling access to RAM via /dev/mem")
Fixes: 527701eda5f1 ("lib: Add a generic version of devmem_is_allowed()")
Signed-off-by: Liang Wang <wangliang101@huawei.com>
Reviewed-by: Luis Chamberlain <mcgrof@kernel.org>
Cc: Palmer Dabbelt <palmerdabbelt@google.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Russell King <linux@armlinux.org.uk>
Cc: Liang Wang <wangliang101@huawei.com>
Cc: Xiaoming Ni <nixiaoming@huawei.com>
Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: <stable@vger.kernel.org> [2.6.37+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
3 years agomm/memcg: fix incorrect flushing of lruvec data in obj_stock
Waiman Long [Fri, 13 Aug 2021 23:54:41 +0000 (16:54 -0700)]
mm/memcg: fix incorrect flushing of lruvec data in obj_stock

When mod_objcg_state() is called with a pgdat that is different from
that in the obj_stock, the old lruvec data cached in obj_stock are
flushed out.  Unfortunately, they were flushed to the new pgdat and so
the data go to the wrong node.  This will screw up the slab data
reported in /sys/devices/system/node/node*/meminfo.

Fix that by flushing the data to the cached pgdat instead.

Link: https://lkml.kernel.org/r/20210802143834.30578-1-longman@redhat.com
Fixes: 68ac5b3c8db2 ("mm/memcg: cache vmstat data in percpu memcg_stock_pcp")
Signed-off-by: Waiman Long <longman@redhat.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Acked-by: Roman Gushchin <guro@fb.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Christoph Lameter <cl@linux.com>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Muchun Song <songmuchun@bytedance.com>
Cc: Alex Shi <alex.shi@linux.alibaba.com>
Cc: Chris Down <chris@chrisdown.name>
Cc: Yafang Shao <laoar.shao@gmail.com>
Cc: Wei Yang <richard.weiyang@gmail.com>
Cc: Masayoshi Mizuma <msys.mizuma@gmail.com>
Cc: Xing Zhengjun <zhengjun.xing@linux.intel.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Waiman Long <longman@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
3 years agomm/madvise: report SIGBUS as -EFAULT for MADV_POPULATE_(READ|WRITE)
David Hildenbrand [Fri, 13 Aug 2021 23:54:37 +0000 (16:54 -0700)]
mm/madvise: report SIGBUS as -EFAULT for MADV_POPULATE_(READ|WRITE)

Doing some extended tests and polishing the man page update for
MADV_POPULATE_(READ|WRITE), I realized that we end up converting also
SIGBUS (via -EFAULT) to -EINVAL, making it look like yet another
madvise() user error.

We want to report only problematic mappings and permission problems that
the user could have know as -EINVAL.

Let's not convert -EFAULT arising due to SIGBUS (or SIGSEGV) to -EINVAL,
but instead indicate -EFAULT to user space.  While we could also convert
it to -ENOMEM, using -EFAULT looks more helpful when user space might
want to troubleshoot what's going wrong: MADV_POPULATE_(READ|WRITE) is
not part of an final Linux release and we can still adjust the behavior.

Link: https://lkml.kernel.org/r/20210726154932.102880-1-david@redhat.com
Fixes: 4ca9b3859dac ("mm/madvise: introduce MADV_POPULATE_(READ|WRITE) to prefault page tables")
Signed-off-by: David Hildenbrand <david@redhat.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Jann Horn <jannh@google.com>
Cc: Jason Gunthorpe <jgg@ziepe.ca>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Rik van Riel <riel@surriel.com>
Cc: Michael S. Tsirkin <mst@redhat.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: Matt Turner <mattst88@gmail.com>
Cc: Thomas Bogendoerfer <tsbogend@alpha.franken.de>
Cc: "James E.J. Bottomley" <James.Bottomley@HansenPartnership.com>
Cc: Helge Deller <deller@gmx.de>
Cc: Chris Zankel <chris@zankel.net>
Cc: Max Filippov <jcmvbkbc@gmail.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Peter Xu <peterx@redhat.com>
Cc: Rolf Eike Beer <eike-kernel@sf-tec.de>
Cc: Ram Pai <linuxram@us.ibm.com>
Cc: Shuah Khan <shuah@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
3 years agomm: slub: fix slub_debug disabling for list of slabs
Vlastimil Babka [Fri, 13 Aug 2021 23:54:34 +0000 (16:54 -0700)]
mm: slub: fix slub_debug disabling for list of slabs

Vijayanand Jitta reports:

  Consider the scenario where CONFIG_SLUB_DEBUG_ON is set and we would
  want to disable slub_debug for few slabs. Using boot parameter with
  slub_debug=-,slab_name syntax doesn't work as expected i.e; only
  disabling debugging for the specified list of slabs. Instead it
  disables debugging for all slabs, which is wrong.

This patch fixes it by delaying the moment when the global slub_debug
flags variable is updated.  In case a "slub_debug=-,slab_name" has been
passed, the global flags remain as initialized (depending on
CONFIG_SLUB_DEBUG_ON enabled or disabled) and are not simply reset to 0.

Link: https://lkml.kernel.org/r/8a3d992a-473a-467b-28a0-4ad2ff60ab82@suse.cz
Signed-off-by: Vlastimil Babka <vbabka@suse.cz>
Reported-by: Vijayanand Jitta <vjitta@codeaurora.org>
Reviewed-by: Vijayanand Jitta <vjitta@codeaurora.org>
Acked-by: David Rientjes <rientjes@google.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Vinayak Menon <vinmenon@codeaurora.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
3 years agoslub: fix kmalloc_pagealloc_invalid_free unit test
Shakeel Butt [Fri, 13 Aug 2021 23:54:31 +0000 (16:54 -0700)]
slub: fix kmalloc_pagealloc_invalid_free unit test

The unit test kmalloc_pagealloc_invalid_free makes sure that for the
higher order slub allocation which goes to page allocator, the free is
called with the correct address i.e.  the virtual address of the head
page.

Commit f227f0faf63b ("slub: fix unreclaimable slab stat for bulk free")
unified the free code paths for page allocator based slub allocations
but instead of using the address passed by the caller, it extracted the
address from the page.  Thus making the unit test
kmalloc_pagealloc_invalid_free moot.  So, fix this by using the address
passed by the caller.

Should we fix this? I think yes because dev expect kasan to catch these
type of programming bugs.

Link: https://lkml.kernel.org/r/20210802180819.1110165-1-shakeelb@google.com
Fixes: f227f0faf63b ("slub: fix unreclaimable slab stat for bulk free")
Signed-off-by: Shakeel Butt <shakeelb@google.com>
Reported-by: Nathan Chancellor <nathan@kernel.org>
Tested-by: Nathan Chancellor <nathan@kernel.org>
Acked-by: Roman Gushchin <guro@fb.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Muchun Song <songmuchun@bytedance.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
3 years agokasan, slub: reset tag when printing address
Kuan-Ying Lee [Fri, 13 Aug 2021 23:54:27 +0000 (16:54 -0700)]
kasan, slub: reset tag when printing address

The address still includes the tags when it is printed.  With hardware
tag-based kasan enabled, we will get a false positive KASAN issue when
we access metadata.

Reset the tag before we access the metadata.

Link: https://lkml.kernel.org/r/20210804090957.12393-3-Kuan-Ying.Lee@mediatek.com
Fixes: aa1ef4d7b3f6 ("kasan, mm: reset tags when accessing metadata")
Signed-off-by: Kuan-Ying Lee <Kuan-Ying.Lee@mediatek.com>
Reviewed-by: Marco Elver <elver@google.com>
Reviewed-by: Andrey Konovalov <andreyknvl@gmail.com>
Cc: Alexander Potapenko <glider@google.com>
Cc: Andrey Ryabinin <ryabinin.a.a@gmail.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Chinwen Chang <chinwen.chang@mediatek.com>
Cc: Nicholas Tang <nicholas.tang@mediatek.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
3 years agokasan, kmemleak: reset tags when scanning block
Kuan-Ying Lee [Fri, 13 Aug 2021 23:54:24 +0000 (16:54 -0700)]
kasan, kmemleak: reset tags when scanning block

Patch series "kasan, slub: reset tag when printing address", v3.

With hardware tag-based kasan enabled, we reset the tag when we access
metadata to avoid from false alarm.

This patch (of 2):

Kmemleak needs to scan kernel memory to check memory leak.  With hardware
tag-based kasan enabled, when it scans on the invalid slab and
dereference, the issue will occur as below.

Hardware tag-based KASAN doesn't use compiler instrumentation, we can not
use kasan_disable_current() to ignore tag check.

Based on the below report, there are 11 0xf7 granules, which amounts to
176 bytes, and the object is allocated from the kmalloc-256 cache.  So
when kmemleak accesses the last 256-176 bytes, it causes faults, as those
are marked with KASAN_KMALLOC_REDZONE == KASAN_TAG_INVALID == 0xfe.

Thus, we reset tags before accessing metadata to avoid from false positives.

  BUG: KASAN: out-of-bounds in scan_block+0x58/0x170
  Read at addr f7ff0000c0074eb0 by task kmemleak/138
  Pointer tag: [f7], memory tag: [fe]

  CPU: 7 PID: 138 Comm: kmemleak Not tainted 5.14.0-rc2-00001-g8cae8cd89f05-dirty #134
  Hardware name: linux,dummy-virt (DT)
  Call trace:
   dump_backtrace+0x0/0x1b0
   show_stack+0x1c/0x30
   dump_stack_lvl+0x68/0x84
   print_address_description+0x7c/0x2b4
   kasan_report+0x138/0x38c
   __do_kernel_fault+0x190/0x1c4
   do_tag_check_fault+0x78/0x90
   do_mem_abort+0x44/0xb4
   el1_abort+0x40/0x60
   el1h_64_sync_handler+0xb4/0xd0
   el1h_64_sync+0x78/0x7c
   scan_block+0x58/0x170
   scan_gray_list+0xdc/0x1a0
   kmemleak_scan+0x2ac/0x560
   kmemleak_scan_thread+0xb0/0xe0
   kthread+0x154/0x160
   ret_from_fork+0x10/0x18

  Allocated by task 0:
   kasan_save_stack+0x2c/0x60
   __kasan_kmalloc+0xec/0x104
   __kmalloc+0x224/0x3c4
   __register_sysctl_paths+0x200/0x290
   register_sysctl_table+0x2c/0x40
   sysctl_init+0x20/0x34
   proc_sys_init+0x3c/0x48
   proc_root_init+0x80/0x9c
   start_kernel+0x648/0x6a4
   __primary_switched+0xc0/0xc8

  Freed by task 0:
   kasan_save_stack+0x2c/0x60
   kasan_set_track+0x2c/0x40
   kasan_set_free_info+0x44/0x54
   ____kasan_slab_free.constprop.0+0x150/0x1b0
   __kasan_slab_free+0x14/0x20
   slab_free_freelist_hook+0xa4/0x1fc
   kfree+0x1e8/0x30c
   put_fs_context+0x124/0x220
   vfs_kern_mount.part.0+0x60/0xd4
   kern_mount+0x24/0x4c
   bdev_cache_init+0x70/0x9c
   vfs_caches_init+0xdc/0xf4
   start_kernel+0x638/0x6a4
   __primary_switched+0xc0/0xc8

  The buggy address belongs to the object at ffff0000c0074e00
   which belongs to the cache kmalloc-256 of size 256
  The buggy address is located 176 bytes inside of
   256-byte region [ffff0000c0074e00ffff0000c0074f00)
  The buggy address belongs to the page:
  page:(____ptrval____) refcount:1 mapcount:0 mapping:0000000000000000 index:0x0 pfn:0x100074
  head:(____ptrval____) order:2 compound_mapcount:0 compound_pincount:0
  flags: 0xbfffc0000010200(slab|head|node=0|zone=2|lastcpupid=0xffff|kasantag=0x0)
  raw: 0bfffc0000010200 0000000000000000 dead000000000122 f5ff0000c0002300
  raw: 0000000000000000 0000000000200020 00000001ffffffff 0000000000000000
  page dumped because: kasan: bad access detected

  Memory state around the buggy address:
   ffff0000c0074c00: f0 f0 f0 f0 f0 f0 f0 f0 f0 fe fe fe fe fe fe fe
   ffff0000c0074d00: fe fe fe fe fe fe fe fe fe fe fe fe fe fe fe fe
  >ffff0000c0074e00: f7 f7 f7 f7 f7 f7 f7 f7 f7 f7 f7 fe fe fe fe fe
                                                      ^
   ffff0000c0074f00: fe fe fe fe fe fe fe fe fe fe fe fe fe fe fe fe
   ffff0000c0075000: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
  ==================================================================
  Disabling lock debugging due to kernel taint
  kmemleak: 181 new suspected memory leaks (see /sys/kernel/debug/kmemleak)

Link: https://lkml.kernel.org/r/20210804090957.12393-1-Kuan-Ying.Lee@mediatek.com
Link: https://lkml.kernel.org/r/20210804090957.12393-2-Kuan-Ying.Lee@mediatek.com
Signed-off-by: Kuan-Ying Lee <Kuan-Ying.Lee@mediatek.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Reviewed-by: Andrey Konovalov <andreyknvl@gmail.com>
Cc: Marco Elver <elver@google.com>
Cc: Nicholas Tang <nicholas.tang@mediatek.com>
Cc: Andrey Ryabinin <ryabinin.a.a@gmail.com>
Cc: Alexander Potapenko <glider@google.com>
Cc: Chinwen Chang <chinwen.chang@mediatek.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
3 years agoMerge tag 'block-5.14-2021-08-13' of git://git.kernel.dk/linux-block
Linus Torvalds [Fri, 13 Aug 2021 23:36:42 +0000 (13:36 -1000)]
Merge tag 'block-5.14-2021-08-13' of git://git.kernel.dk/linux-block

Pull block fixes from Jens Axboe:
 "A few fixes for block that should go into 5.14:

   - Revert the mq-deadline cgroup addition. More work is needed on this
     front, let's revert it for now and get it right before having it in
     a released kernel (Tejun)

   - blk-iocost lockdep fix (Ming)

   - nbd double completion fix (Xie)

   - Fix for non-idling when clearing the shared tag flag (Yu)"

* tag 'block-5.14-2021-08-13' of git://git.kernel.dk/linux-block:
  nbd: Aovid double completion of a request
  blk-mq: clear active_queues before clearing BLK_MQ_F_TAG_QUEUE_SHARED
  Revert "block/mq-deadline: Add cgroup support"
  blk-iocost: fix lockdep warning on blkcg->lock

3 years agoMerge tag 'io_uring-5.14-2021-08-13' of git://git.kernel.dk/linux-block
Linus Torvalds [Fri, 13 Aug 2021 23:25:08 +0000 (13:25 -1000)]
Merge tag 'io_uring-5.14-2021-08-13' of git://git.kernel.dk/linux-block

Pull io_uring fixes from Jens Axboe:
 "A bit bigger than the previous weeks, but mostly just a few stable
  bound fixes. In detail:

   - Followup fixes to patches from last week for io-wq, turns out they
     weren't complete (Hao)

   - Two lockdep reported fixes out of the RT camp (me)

   - Sync the io_uring-cp example with liburing, as a few bug fixes
     never made it to the kernel carried version (me)

   - SQPOLL related TIF_NOTIFY_SIGNAL fix (Nadav)

   - Use WRITE_ONCE() when writing sq flags (Nadav)

   - io_rsrc_put_work() deadlock fix (Pavel)"

* tag 'io_uring-5.14-2021-08-13' of git://git.kernel.dk/linux-block:
  tools/io_uring/io_uring-cp: sync with liburing example
  io_uring: fix ctx-exit io_rsrc_put_work() deadlock
  io_uring: drop ctx->uring_lock before flushing work item
  io-wq: fix IO_WORKER_F_FIXED issue in create_io_worker()
  io-wq: fix bug of creating io-wokers unconditionally
  io_uring: rsrc ref lock needs to be IRQ safe
  io_uring: Use WRITE_ONCE() when writing to sq_flags
  io_uring: clear TIF_NOTIFY_SIGNAL when running task work

3 years agoMerge tag 'pinctrl-v5.14-2' of git://git.kernel.org/pub/scm/linux/kernel/git/linusw...
Linus Torvalds [Fri, 13 Aug 2021 22:41:45 +0000 (12:41 -1000)]
Merge tag 'pinctrl-v5.14-2' of git://git./linux/kernel/git/linusw/linux-pinctrl

Pull pin control fixes from Linus Walleij:
 "An assortment of pin control fixes of varying importance, the most
  important ones affecting Intel and AMD laptops turned up the recent
  few days so it's time to push this to your tree.

   - Fix the Kconfig dependency for Qualcomm SM8350 pin controller

   - Fix pin biasing fallback behaviour on the Mediatek pin controller

   - Fix the GPIO numbering scheme for Intel Tiger Lake-H to correspond
     to the products that are now actually out on the market

   - Fix a pin control function itemization in the Sunxi driver
     out-of-bounds access bug

   - Fix disable clocking for the RISC-V K210 pin controller on the
     errorpath

   - Fix a system shutdown bug affecting AMD Ryzen-based laptops, the
     system would not suspend but just bounce back up"

* tag 'pinctrl-v5.14-2' of git://git.kernel.org/pub/scm/linux/kernel/git/linusw/linux-pinctrl:
  pinctrl: amd: Fix an issue with shutdown when system set to s0ix
  pinctrl: k210: Fix k210_fpioa_probe()
  pinctrl: sunxi: Don't underestimate number of functions
  pinctrl: tigerlake: Fix GPIO mapping for newer version of software
  pinctrl: mediatek: Fix fallback behavior for bias_set_combo
  pinctrl: qcom: fix GPIOLIB dependencies

3 years agonbd: Aovid double completion of a request
Xie Yongji [Fri, 13 Aug 2021 15:13:30 +0000 (23:13 +0800)]
nbd: Aovid double completion of a request

There is a race between iterating over requests in
nbd_clear_que() and completing requests in recv_work(),
which can lead to double completion of a request.

To fix it, flush the recv worker before iterating over
the requests and don't abort the completed request
while iterating.

Fixes: 96d97e17828f ("nbd: clear_sock on netlink disconnect")
Reported-by: Jiang Yadong <jiangyadong@bytedance.com>
Signed-off-by: Xie Yongji <xieyongji@bytedance.com>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Link: https://lore.kernel.org/r/20210813151330.96-1-xieyongji@bytedance.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agotools/io_uring/io_uring-cp: sync with liburing example
Jens Axboe [Fri, 13 Aug 2021 14:57:07 +0000 (08:57 -0600)]
tools/io_uring/io_uring-cp: sync with liburing example

This example is missing a few fixes that are in the liburing version,
synchronize with the upstream version.

Reported-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agoblk-mq: clear active_queues before clearing BLK_MQ_F_TAG_QUEUE_SHARED
Yu Kuai [Sat, 31 Jul 2021 06:21:30 +0000 (14:21 +0800)]
blk-mq: clear active_queues before clearing BLK_MQ_F_TAG_QUEUE_SHARED

We run a test that delete and recover devcies frequently(two devices on
the same host), and we found that 'active_queues' is super big after a
period of time.

If device a and device b share a tag set, and a is deleted, then
blk_mq_exit_queue() will clear BLK_MQ_F_TAG_QUEUE_SHARED because there
is only one queue that are using the tag set. However, if b is still
active, the active_queues of b might never be cleared even if b is
deleted.

Thus clear active_queues before BLK_MQ_F_TAG_QUEUE_SHARED is cleared.

Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20210731062130.1533893-1-yukuai3@huawei.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agodriver core: Add missing kernel doc for device::msi_lock
Thomas Gleixner [Fri, 13 Aug 2021 10:36:14 +0000 (12:36 +0200)]
driver core: Add missing kernel doc for device::msi_lock

Fixes: 77e89afc25f3 ("PCI/MSI: Protect msi_desc::masked for multi-MSI")
Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
3 years agoMerge branch 'kvm-tdpmmu-fixes' into kvm-master
Paolo Bonzini [Fri, 13 Aug 2021 07:33:13 +0000 (03:33 -0400)]
Merge branch 'kvm-tdpmmu-fixes' into kvm-master

Merge topic branch with fixes for both 5.14-rc6 and 5.15.

3 years agoKVM: x86/mmu: Protect marking SPs unsync when using TDP MMU with spinlock
Sean Christopherson [Thu, 12 Aug 2021 18:18:15 +0000 (11:18 -0700)]
KVM: x86/mmu: Protect marking SPs unsync when using TDP MMU with spinlock

Add yet another spinlock for the TDP MMU and take it when marking indirect
shadow pages unsync.  When using the TDP MMU and L1 is running L2(s) with
nested TDP, KVM may encounter shadow pages for the TDP entries managed by
L1 (controlling L2) when handling a TDP MMU page fault.  The unsync logic
is not thread safe, e.g. the kvm_mmu_page fields are not atomic, and
misbehaves when a shadow page is marked unsync via a TDP MMU page fault,
which runs with mmu_lock held for read, not write.

Lack of a critical section manifests most visibly as an underflow of
unsync_children in clear_unsync_child_bit() due to unsync_children being
corrupted when multiple CPUs write it without a critical section and
without atomic operations.  But underflow is the best case scenario.  The
worst case scenario is that unsync_children prematurely hits '0' and
leads to guest memory corruption due to KVM neglecting to properly sync
shadow pages.

Use an entirely new spinlock even though piggybacking tdp_mmu_pages_lock
would functionally be ok.  Usurping the lock could degrade performance when
building upper level page tables on different vCPUs, especially since the
unsync flow could hold the lock for a comparatively long time depending on
the number of indirect shadow pages and the depth of the paging tree.

For simplicity, take the lock for all MMUs, even though KVM could fairly
easily know that mmu_lock is held for write.  If mmu_lock is held for
write, there cannot be contention for the inner spinlock, and marking
shadow pages unsync across multiple vCPUs will be slow enough that
bouncing the kvm_arch cacheline should be in the noise.

Note, even though L2 could theoretically be given access to its own EPT
entries, a nested MMU must hold mmu_lock for write and thus cannot race
against a TDP MMU page fault.  I.e. the additional spinlock only _needs_ to
be taken by the TDP MMU, as opposed to being taken by any MMU for a VM
that is running with the TDP MMU enabled.  Holding mmu_lock for read also
prevents the indirect shadow page from being freed.  But as above, keep
it simple and always take the lock.

Alternative #1, the TDP MMU could simply pass "false" for can_unsync and
effectively disable unsync behavior for nested TDP.  Write protecting leaf
shadow pages is unlikely to noticeably impact traditional L1 VMMs, as such
VMMs typically don't modify TDP entries, but the same may not hold true for
non-standard use cases and/or VMMs that are migrating physical pages (from
L1's perspective).

Alternative #2, the unsync logic could be made thread safe.  In theory,
simply converting all relevant kvm_mmu_page fields to atomics and using
atomic bitops for the bitmap would suffice.  However, (a) an in-depth audit
would be required, (b) the code churn would be substantial, and (c) legacy
shadow paging would incur additional atomic operations in performance
sensitive paths for no benefit (to legacy shadow paging).

Fixes: a2855afc7ee8 ("KVM: x86/mmu: Allow parallel page faults for the TDP MMU")
Cc: stable@vger.kernel.org
Cc: Ben Gardon <bgardon@google.com>
Signed-off-by: Sean Christopherson <seanjc@google.com>
Message-Id: <20210812181815.3378104-1-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86/mmu: Don't step down in the TDP iterator when zapping all SPTEs
Sean Christopherson [Thu, 12 Aug 2021 18:14:14 +0000 (11:14 -0700)]
KVM: x86/mmu: Don't step down in the TDP iterator when zapping all SPTEs

Set the min_level for the TDP iterator at the root level when zapping all
SPTEs to optimize the iterator's try_step_down().  Zapping a non-leaf
SPTE will recursively zap all its children, thus there is no need for the
iterator to attempt to step down.  This avoids rereading the top-level
SPTEs after they are zapped by causing try_step_down() to short-circuit.

In most cases, optimizing try_step_down() will be in the noise as the cost
of zapping SPTEs completely dominates the overall time.  The optimization
is however helpful if the zap occurs with relatively few SPTEs, e.g. if KVM
is zapping in response to multiple memslot updates when userspace is adding
and removing read-only memslots for option ROMs.  In that case, the task
doing the zapping likely isn't a vCPU thread, but it still holds mmu_lock
for read and thus can be a noisy neighbor of sorts.

Reviewed-by: Ben Gardon <bgardon@google.com>
Signed-off-by: Sean Christopherson <seanjc@google.com>
Message-Id: <20210812181414.3376143-3-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
3 years agoKVM: x86/mmu: Don't leak non-leaf SPTEs when zapping all SPTEs
Sean Christopherson [Thu, 12 Aug 2021 18:14:13 +0000 (11:14 -0700)]
KVM: x86/mmu: Don't leak non-leaf SPTEs when zapping all SPTEs

Pass "all ones" as the end GFN to signal "zap all" for the TDP MMU and
really zap all SPTEs in this case.  As is, zap_gfn_range() skips non-leaf
SPTEs whose range exceeds the range to be zapped.  If shadow_phys_bits is
not aligned to the range size of top-level SPTEs, e.g. 512gb with 4-level
paging, the "zap all" flows will skip top-level SPTEs whose range extends
beyond shadow_phys_bits and leak their SPs when the VM is destroyed.

Use the current upper bound (based on host.MAXPHYADDR) to detect that the
caller wants to zap all SPTEs, e.g. instead of using the max theoretical
gfn, 1 << (52 - 12).  The more precise upper bound allows the TDP iterator
to terminate its walk earlier when running on hosts with MAXPHYADDR < 52.

Add a WARN on kmv->arch.tdp_mmu_pages when the TDP MMU is destroyed to
help future debuggers should KVM decide to leak SPTEs again.

The bug is most easily reproduced by running (and unloading!) KVM in a
VM whose host.MAXPHYADDR < 39, as the SPTE for gfn=0 will be skipped.

  =============================================================================
  BUG kvm_mmu_page_header (Not tainted): Objects remaining in kvm_mmu_page_header on __kmem_cache_shutdown()
  -----------------------------------------------------------------------------
  Slab 0x000000004d8f7af1 objects=22 used=2 fp=0x00000000624d29ac flags=0x4000000000000200(slab|zone=1)
  CPU: 0 PID: 1582 Comm: rmmod Not tainted 5.14.0-rc2+ #420
  Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 0.0.0 02/06/2015
  Call Trace:
   dump_stack_lvl+0x45/0x59
   slab_err+0x95/0xc9
   __kmem_cache_shutdown.cold+0x3c/0x158
   kmem_cache_destroy+0x3d/0xf0
   kvm_mmu_module_exit+0xa/0x30 [kvm]
   kvm_arch_exit+0x5d/0x90 [kvm]
   kvm_exit+0x78/0x90 [kvm]
   vmx_exit+0x1a/0x50 [kvm_intel]
   __x64_sys_delete_module+0x13f/0x220
   do_syscall_64+0x3b/0xc0
   entry_SYSCALL_64_after_hwframe+0x44/0xae

Fixes: faaf05b00aec ("kvm: x86/mmu: Support zapping SPTEs in the TDP MMU")
Cc: stable@vger.kernel.org
Cc: Ben Gardon <bgardon@google.com>
Signed-off-by: Sean Christopherson <seanjc@google.com>
Message-Id: <20210812181414.3376143-2-seanjc@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>