linux-2.6-block.git
3 years agoMerge branch 'clocksource.2021.11.30c' into HEAD
Paul E. McKenney [Thu, 9 Dec 2021 19:40:42 +0000 (11:40 -0800)]
Merge branch 'clocksource.2021.11.30c' into HEAD

clocksource.2021.11.30c: Clocksource watchdog updates.

3 years agoMerge branch 'lkmm-dev.2021.11.30c' into HEAD
Paul E. McKenney [Thu, 9 Dec 2021 19:39:39 +0000 (11:39 -0800)]
Merge branch 'lkmm-dev.2021.11.30c' into HEAD

lkmm-dev.2021.11.30c: LKMM development branch.

3 years agoMerge branch 'lkmm.2021.11.30c' into HEAD
Paul E. McKenney [Thu, 9 Dec 2021 19:38:57 +0000 (11:38 -0800)]
Merge branch 'lkmm.2021.11.30c' into HEAD

lkmm.2021.11.30c: Linux-kernel memory-model updates.

3 years agoMerge branches 'doc.2021.11.30c', 'exp.2021.12.07a', 'fastnohz.2021.11.30c', 'fixes...
Paul E. McKenney [Thu, 9 Dec 2021 19:38:09 +0000 (11:38 -0800)]
Merge branches 'doc.2021.11.30c', 'exp.2021.12.07a', 'fastnohz.2021.11.30c', 'fixes.2021.11.30c', 'nocb.2021.12.09a', 'nolibc.2021.11.30c', 'tasks.2021.12.09a', 'torture.2021.12.07a' and 'torturescript.2021.11.30c' into HEAD

doc.2021.11.30c: Documentation updates.
exp.2021.12.07a: Expedited-grace-period fixes.
fastnohz.2021.11.30c: Remove CONFIG_RCU_FAST_NO_HZ.
fixes.2021.11.30c: Miscellaneous fixes.
nocb.2021.12.09a: No-CB CPU updates.
nolibc.2021.11.30c: Tiny in-kernel library updates.
tasks.2021.12.09a: RCU-tasks updates, including update-side scalability.
torture.2021.12.07a: Torture-test in-kernel module updates.
torturescript.2021.11.30c: Torture-test scripting updates.

3 years agorcu/nocb: Merge rcu_spawn_cpu_nocb_kthread() and rcu_spawn_one_nocb_kthread()
Frederic Weisbecker [Tue, 23 Nov 2021 00:37:08 +0000 (01:37 +0100)]
rcu/nocb: Merge rcu_spawn_cpu_nocb_kthread() and rcu_spawn_one_nocb_kthread()

The rcu_spawn_one_nocb_kthread() function is called only from
rcu_spawn_cpu_nocb_kthread().  Therefore, inline the former into
the latter, saving a few lines of code.

Reviewed-by: Neeraj Upadhyay <quic_neeraju@quicinc.com>
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Cc: Boqun Feng <boqun.feng@gmail.com>
Cc: Uladzislau Rezki <urezki@gmail.com>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Joel Fernandes <joel@joelfernandes.org>
Tested-by: Juri Lelli <juri.lelli@redhat.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorcu/nocb: Allow empty "rcu_nocbs" kernel parameter
Frederic Weisbecker [Tue, 23 Nov 2021 00:37:07 +0000 (01:37 +0100)]
rcu/nocb: Allow empty "rcu_nocbs" kernel parameter

Allow the rcu_nocbs kernel parameter to be specified just by itself,
without specifying any CPUs.  This allows systems administrators to use
"rcu_nocbs" to specify that none of the CPUs are to be offloaded at boot
time, but than any of them may be offloaded at runtime via cpusets.

In contrast, if the "rcu_nocbs" or "nohz_full" kernel parameters are not
specified at all, then not only are none of the CPUs offloaded at boot,
none of them can be offloaded at runtime, either.

While in the area, modernize the description of the "rcuo" kthreads'
naming scheme.

Reviewed-by: Neeraj Upadhyay <quic_neeraju@quicinc.com>
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Cc: Boqun Feng <boqun.feng@gmail.com>
Cc: Uladzislau Rezki <urezki@gmail.com>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Joel Fernandes <joel@joelfernandes.org>
Tested-by: Juri Lelli <juri.lelli@redhat.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorcu/nocb: Create kthreads on all CPUs if "rcu_nocbs=" or "nohz_full=" are passed
Frederic Weisbecker [Tue, 23 Nov 2021 00:37:06 +0000 (01:37 +0100)]
rcu/nocb: Create kthreads on all CPUs if "rcu_nocbs=" or "nohz_full=" are passed

In order to be able to (de-)offload any CPU using cpusets in the future,
create the NOCB data structures for all possible CPUs.  For now this is
done only as long as the "rcu_nocbs=" or "nohz_full=" kernel parameters
are passed to avoid the unnecessary overhead for most users.

Note that the rcuog and rcuoc kthreads are not created until at least
one of the corresponding CPUs comes online.  This approach avoids the
creation of excess kthreads when firmware lies about the number of CPUs
present on the system.

Reviewed-by: Neeraj Upadhyay <quic_neeraju@quicinc.com>
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Cc: Boqun Feng <boqun.feng@gmail.com>
Cc: Uladzislau Rezki <urezki@gmail.com>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Joel Fernandes <joel@joelfernandes.org>
Tested-by: Juri Lelli <juri.lelli@redhat.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorcu/nocb: Optimize kthreads and rdp initialization
Frederic Weisbecker [Tue, 23 Nov 2021 00:37:05 +0000 (01:37 +0100)]
rcu/nocb: Optimize kthreads and rdp initialization

Currently cpumask_available() is used to prevent from unwanted NOCB
initialization.  However if neither "rcu_nocbs=" nor "nohz_full="
parameters are passed to a kernel built with CONFIG_CPUMASK_OFFSTACK=n,
the initialization path is still taken, running through all sorts of
needless operations and iterations on an empty cpumask.

Fix this by relying on a real initialization state instead.  This also
optimizes kthread creation, preventing needless iteration over all online
CPUs when the kernel is booted without any offloaded CPUs.

Reviewed-by: Neeraj Upadhyay <quic_neeraju@quicinc.com>
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Cc: Boqun Feng <boqun.feng@gmail.com>
Cc: Uladzislau Rezki <urezki@gmail.com>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Joel Fernandes <joel@joelfernandes.org>
Tested-by: Juri Lelli <juri.lelli@redhat.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorcu/nocb: Prepare nocb_cb_wait() to start with a non-offloaded rdp
Frederic Weisbecker [Tue, 23 Nov 2021 00:37:04 +0000 (01:37 +0100)]
rcu/nocb: Prepare nocb_cb_wait() to start with a non-offloaded rdp

In order to be able to toggle the offloaded state from cpusets, a nocb
kthread will need to be created for all possible CPUs whenever either
of the "rcu_nocbs=" or "nohz_full=" parameters are specified.

Therefore, the nocb_cb_wait() kthread must be prepared to start running
on a de-offloaded rdp.  To accomplish this, simply move the sleeping
condition to the beginning of the nocb_cb_wait() function, which prevents
this kthread from attempting to invoke callbacks before the corresponding
CPU is offloaded.

Reviewed-by: Neeraj Upadhyay <quic_neeraju@quicinc.com>
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Cc: Boqun Feng <boqun.feng@gmail.com>
Cc: Uladzislau Rezki <urezki@gmail.com>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Joel Fernandes <joel@joelfernandes.org>
Tested-by: Juri Lelli <juri.lelli@redhat.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorcu/nocb: Remove rcu_node structure from nocb list when de-offloaded
Frederic Weisbecker [Tue, 23 Nov 2021 00:37:03 +0000 (01:37 +0100)]
rcu/nocb: Remove rcu_node structure from nocb list when de-offloaded

The nocb_gp_wait() function iterates over all CPUs in its group,
including even those CPUs that have been de-offloaded.  This is of
course suboptimal, especially if none of the CPUs within the group are
currently offloaded.  This will become even more of a problem once a
nocb kthread is created for all possible CPUs.

Therefore use a standard double linked list to link all the offloaded
rcu_data structures and safely add or delete these structure as we
offload or de-offload them, respectively.

Reviewed-by: Neeraj Upadhyay <quic_neeraju@quicinc.com>
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Cc: Boqun Feng <boqun.feng@gmail.com>
Cc: Uladzislau Rezki <urezki@gmail.com>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Joel Fernandes <joel@joelfernandes.org>
Tested-by: Juri Lelli <juri.lelli@redhat.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorcu-tasks: Use fewer callbacks queues if callback flood ends
Paul E. McKenney [Tue, 30 Nov 2021 00:52:31 +0000 (16:52 -0800)]
rcu-tasks: Use fewer callbacks queues if callback flood ends

By default, when lock contention is encountered, the RCU Tasks flavors
of RCU switch to using per-CPU queueing.  However, if the callback
flood ends, per-CPU queueing continues to be used, which introduces
significant additional overhead, especially for callback invocation,
which fans out a series of workqueue handlers.

This commit therefore switches back to single-queue operation if at the
beginning of a grace period there are very few callbacks.  The definition
of "very few" is set by the rcupdate.rcu_task_collapse_lim module
parameter, which defaults to 10.  This switch happens in two phases,
with the first phase causing future callbacks to be enqueued on CPU 0's
queue, but with all queues continuing to be checked for grace periods
and callback invocation.  The second phase checks to see if an RCU grace
period has elapsed and if all remaining RCU-Tasks callbacks are queued
on CPU 0.  If so, only CPU 0 is checked for future grace periods and
callback operation.

Of course, the return of contention anywhere during this process will
result in returning to per-CPU callback queueing.

Reported-by: Martin Lau <kafai@fb.com>
Cc: Neeraj Upadhyay <neeraj.iitr10@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorcu-tasks: Use separate ->percpu_dequeue_lim for callback dequeueing
Paul E. McKenney [Mon, 29 Nov 2021 19:46:33 +0000 (11:46 -0800)]
rcu-tasks: Use separate ->percpu_dequeue_lim for callback dequeueing

Decreasing the number of callback queues is a bit tricky because it
is necessary to handle callbacks that were queued before the number of
queues decreased, but which were not ready to invoke until afterwards.
This commit takes a first step in this direction by maintaining a separate
->percpu_dequeue_lim to control callback dequeueing, in addition to the
existing ->percpu_enqueue_lim which now controls only enqueueing.

Reported-by: Martin Lau <kafai@fb.com>
Cc: Neeraj Upadhyay <neeraj.iitr10@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorcu-tasks: Use more callback queues if contention encountered
Paul E. McKenney [Wed, 24 Nov 2021 23:12:15 +0000 (15:12 -0800)]
rcu-tasks: Use more callback queues if contention encountered

The rcupdate.rcu_task_enqueue_lim module parameter allows system
administrators to tune the number of callback queues used by the RCU
Tasks flavors.  However if callback storms are infrequent, it would
be better to operate with a single queue on a given system unless and
until that system actually needed more queues.  Systems not needing
more queues can then avoid the overhead of checking the extra queues
and especially avoid the overhead of fanning workqueue handlers out to
all CPUs to invoke callbacks.

This commit therefore switches to using all the CPUs' callback queues if
call_rcu_tasks_generic() encounters too much lock contention.  The amount
of lock contention to tolerate defaults to 100 contended lock acquisitions
per jiffy, and can be adjusted using the new rcupdate.rcu_task_contend_lim
module parameter.

Such switching is undertaken only if the rcupdate.rcu_task_enqueue_lim
module parameter is negative, which is its default value (-1).
This allows savvy systems administrators to set the number of queues
to some known good value and to not have to worry about the kernel doing
any second guessing.

[ paulmck: Apply feedback from Guillaume Tucker and kernelci. ]

Reported-by: Martin Lau <kafai@fb.com>
Cc: Neeraj Upadhyay <neeraj.iitr10@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorcu-tasks: Avoid raw-spinlocked wakeups from call_rcu_tasks_generic()
Paul E. McKenney [Wed, 24 Nov 2021 00:16:50 +0000 (16:16 -0800)]
rcu-tasks: Avoid raw-spinlocked wakeups from call_rcu_tasks_generic()

If the caller of of call_rcu_tasks(), call_rcu_tasks_rude(),
or call_rcu_tasks_trace() holds a raw spinlock, and then if
call_rcu_tasks_generic() determines that the grace-period kthread must
be awakened, then the wakeup might acquire a normal spinlock while a
raw spinlock is held.  This results in lockdep splats when the
kernel is built with CONFIG_PROVE_RAW_LOCK_NESTING=y.

This commit therefore defers the wakeup using irq_work_queue().

It would be nice to directly invoke wakeup when a raw spinlock is not
held, but there is currently no way to check for this in all kernels.

Reported-by: Martin Lau <kafai@fb.com>
Cc: Neeraj Upadhyay <neeraj.iitr10@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorcu-tasks: Count trylocks to estimate call_rcu_tasks() contention
Paul E. McKenney [Mon, 22 Nov 2021 21:38:42 +0000 (13:38 -0800)]
rcu-tasks: Count trylocks to estimate call_rcu_tasks() contention

This commit converts the unconditional raw_spin_lock_rcu_node() lock
acquisition in call_rcu_tasks_generic() to a trylock followed by an
unconditional acquisition if the trylock fails.  If the trylock fails,
the failure is counted, but the count is reset to zero on each new jiffy.

This statistic will be used to determine when to move from a single
callback queue to per-CPU callback queues.

Reported-by: Martin Lau <kafai@fb.com>
Cc: Neeraj Upadhyay <neeraj.iitr10@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorcu-tasks: Add rcupdate.rcu_task_enqueue_lim to set initial queueing
Paul E. McKenney [Fri, 12 Nov 2021 15:33:40 +0000 (07:33 -0800)]
rcu-tasks: Add rcupdate.rcu_task_enqueue_lim to set initial queueing

This commit adds a rcupdate.rcu_task_enqueue_lim module parameter that
sets the initial number of callback queues to use for the RCU Tasks
family of RCU implementations.  This parameter allows testing of various
fanout values.

Reported-by: Martin Lau <kafai@fb.com>
Cc: Neeraj Upadhyay <neeraj.iitr10@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorcu-tasks: Make rcu_barrier_tasks*() handle multiple callback queues
Paul E. McKenney [Thu, 11 Nov 2021 22:53:43 +0000 (14:53 -0800)]
rcu-tasks: Make rcu_barrier_tasks*() handle multiple callback queues

Currently, rcu_barrier_tasks(), rcu_barrier_tasks_rude(),
and rcu_barrier_tasks_trace() simply invoke the corresponding
synchronize_rcu_tasks*() function.  This works because there is only
one callback queue.

However, there will soon be multiple callback queues.  This commit
therefore scans the queues currently in use, entraining a callback on
each non-empty queue.  Sequence numbers and reference counts are used
to synchronize this process in a manner similar to the approach taken
by rcu_barrier().

Reported-by: Martin Lau <kafai@fb.com>
Cc: Neeraj Upadhyay <neeraj.iitr10@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorcu-tasks: Use workqueues for multiple rcu_tasks_invoke_cbs() invocations
Paul E. McKenney [Wed, 10 Nov 2021 23:56:40 +0000 (15:56 -0800)]
rcu-tasks: Use workqueues for multiple rcu_tasks_invoke_cbs() invocations

If there is a flood of callbacks, it is necessary to put multiple
CPUs to work invoking those callbacks.  This commit therefore uses a
workqueue-flooding approach to parallelize RCU Tasks callback execution.

Reported-by: Martin Lau <kafai@fb.com>
Cc: Neeraj Upadhyay <neeraj.iitr10@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorcu-tasks: Abstract invocations of callbacks
Paul E. McKenney [Tue, 9 Nov 2021 23:34:56 +0000 (15:34 -0800)]
rcu-tasks: Abstract invocations of callbacks

This commit adds a rcu_tasks_invoke_cbs() function that invokes all
ready callbacks on all of the per-CPU lists that are currently in use.

Reported-by: Martin Lau <kafai@fb.com>
Cc: Neeraj Upadhyay <neeraj.iitr10@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorcu-tasks: Abstract checking of callback lists
Paul E. McKenney [Tue, 9 Nov 2021 21:37:34 +0000 (13:37 -0800)]
rcu-tasks: Abstract checking of callback lists

This commit adds a rcu_tasks_need_gpcb() function that returns an
indication of whether another grace period is required, and if no grace
period is required, whether there are callbacks that need to be invoked.
The function scans all per-CPU lists currently in use.

Reported-by: Martin Lau <kafai@fb.com>
Cc: Neeraj Upadhyay <neeraj.iitr10@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorcu-tasks: Add a ->percpu_enqueue_lim to the rcu_tasks structure
Paul E. McKenney [Tue, 9 Nov 2021 19:11:32 +0000 (11:11 -0800)]
rcu-tasks: Add a ->percpu_enqueue_lim to the rcu_tasks structure

This commit adds a ->percpu_enqueue_lim field to the rcu_tasks structure.
This field contains two to the power of the ->percpu_enqueue_shift
field, easing construction of iterators over the per-CPU queues that
might contain RCU Tasks callbacks.  Such iterators will be introduced
in later commits.

Reported-by: Martin Lau <kafai@fb.com>
Cc: Neeraj Upadhyay <neeraj.iitr10@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorcu-tasks: Inspect stalled task's trc state in locked state
Neeraj Upadhyay [Tue, 9 Nov 2021 11:22:14 +0000 (16:52 +0530)]
rcu-tasks: Inspect stalled task's trc state in locked state

On RCU tasks trace stall, inspect the RCU-tasks-trace specific
states of stalled task in locked down state, using try_invoke_
on_locked_down_task(), to get reliable trc state of a non-running
stalled task.

This was tested using the following command:

tools/testing/selftests/rcutorture/bin/kvm.sh --cpus 8 --configs TRACE01 \
--bootargs "rcutorture.torture_type=tasks-tracing rcutorture.stall_cpu=10 \
rcutorture.stall_cpu_block=1 rcupdate.rcu_task_stall_timeout=100" --trust-make

As expected, this produced the following console output for running and
sleeping tasks.

[   21.520291] INFO: rcu_tasks_trace detected stalls on tasks:
[   21.521292] P85: ... nesting: 1N cpu: 2
[   21.521966] task:rcu_torture_sta state:D stack:15080 pid:   85 ppid:     2
flags:0x00004000
[   21.523384] Call Trace:
[   21.523808]  __schedule+0x273/0x6e0
[   21.524428]  schedule+0x35/0xa0
[   21.524971]  schedule_timeout+0x1ed/0x270
[   21.525690]  ? del_timer_sync+0x30/0x30
[   21.526371]  ? rcu_torture_writer+0x720/0x720
[   21.527106]  rcu_torture_stall+0x24a/0x270
[   21.527816]  kthread+0x115/0x140
[   21.528401]  ? set_kthread_struct+0x40/0x40
[   21.529136]  ret_from_fork+0x22/0x30
[   21.529766]  1 holdouts
[   21.632300] INFO: rcu_tasks_trace detected stalls on tasks:
[   21.632345] rcu_torture_stall end.
[   21.633293] P85: .
[   21.633294] task:rcu_torture_sta state:R  running task stack:15080 pid:
85 ppid:     2 flags:0x00004000
[   21.633299] Call Trace:
[   21.633301]  ? vprintk_emit+0xab/0x180
[   21.633306]  ? vprintk_emit+0x11a/0x180
[   21.633308]  ? _printk+0x4d/0x69
[   21.633311]  ? __default_send_IPI_shortcut+0x1f/0x40

[ paulmck: Update to new v5.16 task_call_func() name. ]

Signed-off-by: Neeraj Upadhyay <quic_neeraju@quicinc.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorcu-tasks: Use spin_lock_rcu_node() and friends
Paul E. McKenney [Tue, 9 Nov 2021 00:52:02 +0000 (16:52 -0800)]
rcu-tasks: Use spin_lock_rcu_node() and friends

This commit renames the rcu_tasks_percpu structure's ->cbs_pcpu_lock
to ->lock and then uses spin_lock_rcu_node() and friends to acquire and
release this lock, preparing for upcoming commits that will spread the
grace-period process across multiple CPUs and kthreads.

[ paulmck: Apply feedback from kernel test robot. ]

Reported-by: Martin Lau <kafai@fb.com>
Cc: Neeraj Upadhyay <neeraj.iitr10@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorcutorture: Combine n_max_cbs from all kthreads in a callback flood
Paul E. McKenney [Tue, 23 Nov 2021 21:51:11 +0000 (13:51 -0800)]
rcutorture: Combine n_max_cbs from all kthreads in a callback flood

With the addition of multiple callback-flood kthreads, the maximum number
of callbacks from any one of those kthreads is reported in the rcutorture
run summary.  This commit changes this to report the sum of each kthread's
maximum number of callbacks in a given callback-flooding episode.

Cc: Neeraj Upadhyay <neeraj.iitr10@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorcutorture: Add ability to limit callback-flood intensity
Paul E. McKenney [Tue, 23 Nov 2021 19:53:52 +0000 (11:53 -0800)]
rcutorture: Add ability to limit callback-flood intensity

The RCU tasks flavors of RCU now need concurrent callback flooding to
test their ability to switch between single-queue mode and per-CPU queue
mode, but their lack of heavy-duty forward-progress features rules out
the use of rcutorture's current callback-flooding code.  This commit
therefore provides the ability to limit the intensity of the callback
floods using a new ->cbflood_max field in the rcu_operations structure.
When this field is zero, there is no limit, otherwise, each callback-flood
kthread allocates at most ->cbflood_max callbacks.

Cc: Neeraj Upadhyay <neeraj.iitr10@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorcutorture: Enable multiple concurrent callback-flood kthreads
Paul E. McKenney [Tue, 23 Nov 2021 04:55:18 +0000 (20:55 -0800)]
rcutorture: Enable multiple concurrent callback-flood kthreads

This commit converts the rcutorture.fwd_progress module parameter from
bool to int, so that it specifies the number of callback-flood kthreads.
Values less than zero specify one kthread per CPU, however, the number of
kthreads executing concurrently is limited to the number of online CPUs.
This commit also reverse the order of the need-resched and callback-flood
operations to cause the callback flooding to happen more nearly at the
same time.

Cc: Neeraj Upadhyay <neeraj.iitr10@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorcutorture: Test RCU-tasks multiqueue callback queueing
Paul E. McKenney [Fri, 12 Nov 2021 16:16:42 +0000 (08:16 -0800)]
rcutorture: Test RCU-tasks multiqueue callback queueing

This commit modifies the TASKS01 scenario to use four callback queues
and the TRACE01 scenario to use two queues, thus providing testing of
multiple queues by default.

Cc: Neeraj Upadhyay <neeraj.iitr10@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorcutorture: Avoid soft lockup during cpu stall
Wander Lairson Costa [Wed, 10 Nov 2021 14:37:45 +0000 (11:37 -0300)]
rcutorture: Avoid soft lockup during cpu stall

If we use the module stall_cpu option, we may get a soft lockup warning
in case we also don't pass the stall_cpu_block option.

Introduce the stall_no_softlockup option to avoid a soft lockup on
cpu stall even if we don't use the stall_cpu_block option.

Signed-off-by: Wander Lairson Costa <wander@redhat.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agolocktorture,rcutorture,torture: Always log error message
Li Zhijian [Wed, 3 Nov 2021 08:30:28 +0000 (16:30 +0800)]
locktorture,rcutorture,torture: Always log error message

Unconditionally log messages corresponding to errors.

Acked-by: Davidlohr Bueso <dbueso@suse.de>
Signed-off-by: Li Zhijian <zhijianx.li@intel.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agoscftorture: Always log error message
Li Zhijian [Wed, 3 Nov 2021 08:30:27 +0000 (16:30 +0800)]
scftorture: Always log error message

Unconditionally log messages corresponding to errors.

Acked-by: Davidlohr Bueso <dbueso@suse.de>
Signed-off-by: Li Zhijian <zhijianx.li@intel.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorcuscale: Always log error message
Li Zhijian [Fri, 29 Oct 2021 09:40:28 +0000 (17:40 +0800)]
rcuscale: Always log error message

Unconditionally log messages corresponding to errors.

Acked-by: Davidlohr Bueso <dbueso@suse.de>
Signed-off-by: Li Zhijian <zhijianx.li@intel.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agoscftorture: Remove unused SCFTORTOUT
Li Zhijian [Fri, 29 Oct 2021 09:40:26 +0000 (17:40 +0800)]
scftorture: Remove unused SCFTORTOUT

There are no longer any users of SCFTORTOUT(), so this commit removes it.

Acked-by: Davidlohr Bueso <dbueso@suse.de>
Signed-off-by: Li Zhijian <zhijianx.li@intel.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agoscftorture: Add missing '\n' to flush message
Li Zhijian [Fri, 29 Oct 2021 09:40:25 +0000 (17:40 +0800)]
scftorture: Add missing '\n' to flush message

Add '\n' to macros to flush message for each call.

Acked-by: Davidlohr Bueso <dbueso@suse.de>
Signed-off-by: Li Zhijian <zhijianx.li@intel.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorefscale: Add missing '\n' to flush message
Li Zhijian [Fri, 29 Oct 2021 09:40:24 +0000 (17:40 +0800)]
refscale: Add missing '\n' to flush message

Add '\n' to macros to flush message for each call.

Acked-by: Davidlohr Bueso <dbueso@suse.de>
Signed-off-by: Li Zhijian <zhijianx.li@intel.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorefscale: Always log the error message
Li Zhijian [Mon, 25 Oct 2021 03:26:58 +0000 (11:26 +0800)]
refscale: Always log the error message

An OOM is a serious error that should be logged even in non-verbose runs.
This commit therefore adds an unconditional SCALEOUT_ERRSTRING() macro
and uses it instead of VERBOSE_SCALEOUT_ERRSTRING() when reporting an OOM.

[ paulmck: Drop do-while from SCALEOUT_ERRSTRING() due to only single statement. ]

Signed-off-by: Li Zhijian <lizhijian@cn.fujitsu.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorcu_tasks: Convert bespoke callback list to rcu_segcblist structure
Paul E. McKenney [Tue, 9 Nov 2021 00:18:57 +0000 (16:18 -0800)]
rcu_tasks: Convert bespoke callback list to rcu_segcblist structure

This commit moves from a bespoke head and tail pointer in the
rcu_tasks_percpu structure to an rcu_segcblist structure, thus allowing
associating the grace-period sequence number with groups of callbacks.
This in turn will allow callbacks to be invoked independently on
different CPUs.

Reported-by: Martin Lau <kafai@fb.com>
Cc: Neeraj Upadhyay <neeraj.iitr10@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorcu-tasks: Convert grace-period counter to grace-period sequence number
Paul E. McKenney [Mon, 8 Nov 2021 22:14:43 +0000 (14:14 -0800)]
rcu-tasks: Convert grace-period counter to grace-period sequence number

This commit moves the rcu_tasks structure's ->n_gps grace-period-counter
field to a ->task_gp_seq grce-period sequence number in order to enable
use of the rcu_segcblist structure for the callback lists.  This in turn
permits CPUs to lag behind the RCU Tasks grace-period sequence number
without suffering long-term slowdowns in callback invocation.

Reported-by: Martin Lau <kafai@fb.com>
Cc: Neeraj Upadhyay <neeraj.iitr10@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorcu-tasks: Introduce ->percpu_enqueue_shift for dynamic queue selection
Paul E. McKenney [Mon, 8 Nov 2021 18:51:13 +0000 (10:51 -0800)]
rcu-tasks: Introduce ->percpu_enqueue_shift for dynamic queue selection

This commit introduces a ->percpu_enqueue_shift field to the rcu_tasks
structure, and uses it to shift down the CPU number in order to
select a rcu_tasks_percpu structure.  This field is currently set to a
sufficiently large shift count to always select the CPU-0 instance of
the rcu_tasks_percpu structure, and later commits will adjust this.

Reported-by: Martin Lau <kafai@fb.com>
Cc: Neeraj Upadhyay <neeraj.iitr10@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorcu-tasks: Create per-CPU callback lists
Paul E. McKenney [Sat, 6 Nov 2021 04:52:00 +0000 (21:52 -0700)]
rcu-tasks: Create per-CPU callback lists

Currently, RCU Tasks Trace (as well as the other two flavors of RCU Tasks)
use a single global callback list.  This works well and is simple, but
expected changes in workload will cause this list to become a bottleneck.
This commit therefore creates per-CPU callback lists for the various
flavors of RCU Tasks, but continues queueing on a single list, namely
that of CPU 0.  Later commits will dynamically vary the number of lists
in use to accommodate dynamic changes in workload.

Reported-by: Martin Lau <kafai@fb.com>
Cc: Neeraj Upadhyay <neeraj.iitr10@gmail.com>
Tested-by: kernel test robot <beibei.si@intel.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorcu/nocb: Don't invoke local rcu core on callback overload from nocb kthread
Frederic Weisbecker [Tue, 19 Oct 2021 00:08:16 +0000 (02:08 +0200)]
rcu/nocb: Don't invoke local rcu core on callback overload from nocb kthread

rcu_core() tries to ensure that its self-invocation in case of callbacks
overload only happen in softirq/rcuc mode. Indeed it doesn't make sense
to trigger local RCU core from nocb_cb kthread since it can execute
on a CPU different from the target rdp. Also in case of overload, the
nocb_cb kthread simply iterates a new loop of callbacks processing.

However the "offloaded" check that aims at preventing misplaced
rcu_core() invocations is wrong. First of all that state is volatile
and second: softirq/rcuc can execute while the target rdp is offloaded.
As a result rcu_core() can be invoked on the wrong CPU while in the
process of (de-)offloading.

Fix that with moving the rcu_core() self-invocation to rcu_core() itself,
irrespective of the rdp offloaded state.

Tested-by: Valentin Schneider <valentin.schneider@arm.com>
Tested-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Cc: Valentin Schneider <valentin.schneider@arm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Joel Fernandes <joel@joelfernandes.org>
Cc: Boqun Feng <boqun.feng@gmail.com>
Cc: Neeraj Upadhyay <neeraju@codeaurora.org>
Cc: Uladzislau Rezki <urezki@gmail.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorcu: Apply callbacks processing time limit only on softirq
Frederic Weisbecker [Tue, 19 Oct 2021 00:08:15 +0000 (02:08 +0200)]
rcu: Apply callbacks processing time limit only on softirq

Time limit only makes sense when callbacks are serviced in softirq mode
because:

_ In case we need to get back to the scheduler,
  cond_resched_tasks_rcu_qs() is called after each callback.

_ In case some other softirq vector needs the CPU, the call to
  local_bh_enable() before cond_resched_tasks_rcu_qs() takes care about
  them via a call to do_softirq().

Therefore, make sure the time limit only applies to softirq mode.

Reviewed-by: Valentin Schneider <valentin.schneider@arm.com>
Tested-by: Valentin Schneider <valentin.schneider@arm.com>
Tested-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Cc: Valentin Schneider <valentin.schneider@arm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Joel Fernandes <joel@joelfernandes.org>
Cc: Boqun Feng <boqun.feng@gmail.com>
Cc: Neeraj Upadhyay <neeraju@codeaurora.org>
Cc: Uladzislau Rezki <urezki@gmail.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorcu: Fix callbacks processing time limit retaining cond_resched()
Frederic Weisbecker [Tue, 19 Oct 2021 00:08:14 +0000 (02:08 +0200)]
rcu: Fix callbacks processing time limit retaining cond_resched()

The callbacks processing time limit makes sure we are not exceeding a
given amount of time executing the queue.

However its "continue" clause bypasses the cond_resched() call on
rcuc and NOCB kthreads, delaying it until we reach the limit, which can
be very long...

Make sure the scheduler has a higher priority than the time limit.

Reviewed-by: Valentin Schneider <valentin.schneider@arm.com>
Tested-by: Valentin Schneider <valentin.schneider@arm.com>
Tested-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Cc: Valentin Schneider <valentin.schneider@arm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Joel Fernandes <joel@joelfernandes.org>
Cc: Boqun Feng <boqun.feng@gmail.com>
Cc: Neeraj Upadhyay <neeraju@codeaurora.org>
Cc: Uladzislau Rezki <urezki@gmail.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorcu/nocb: Limit number of softirq callbacks only on softirq
Frederic Weisbecker [Tue, 19 Oct 2021 00:08:13 +0000 (02:08 +0200)]
rcu/nocb: Limit number of softirq callbacks only on softirq

The current condition to limit the number of callbacks executed in a
row checks the offloaded state of the rdp. Not only is it volatile
but it is also misleading: the rcu_core() may well be executing
callbacks concurrently with NOCB kthreads, and the offloaded state
would then be verified on both cases. As a result the limit would
spuriously not apply anymore on softirq while in the middle of
(de-)offloading process.

Fix and clarify the condition with those constraints in mind:

_ If callbacks are processed either by rcuc or NOCB kthread, the call
  to cond_resched_tasks_rcu_qs() is enough to take care of the overload.

_ If instead callbacks are processed by softirqs:
  * If need_resched(), exit the callbacks processing
  * Otherwise if CPU is idle we can continue
  * Otherwise exit because a softirq shouldn't interrupt a task for too
    long nor deprive other pending softirq vectors of the CPU.

Tested-by: Valentin Schneider <valentin.schneider@arm.com>
Tested-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Cc: Valentin Schneider <valentin.schneider@arm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Joel Fernandes <joel@joelfernandes.org>
Cc: Boqun Feng <boqun.feng@gmail.com>
Cc: Neeraj Upadhyay <neeraju@codeaurora.org>
Cc: Uladzislau Rezki <urezki@gmail.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorcu/nocb: Use appropriate rcu_nocb_lock_irqsave()
Frederic Weisbecker [Tue, 19 Oct 2021 00:08:12 +0000 (02:08 +0200)]
rcu/nocb: Use appropriate rcu_nocb_lock_irqsave()

Instead of hardcoding IRQ save and nocb lock, use the consolidated
API (and fix a comment as per Valentin Schneider's suggestion).

Reviewed-by: Valentin Schneider <valentin.schneider@arm.com>
Tested-by: Valentin Schneider <valentin.schneider@arm.com>
Tested-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Cc: Valentin Schneider <valentin.schneider@arm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Joel Fernandes <joel@joelfernandes.org>
Cc: Boqun Feng <boqun.feng@gmail.com>
Cc: Neeraj Upadhyay <neeraju@codeaurora.org>
Cc: Uladzislau Rezki <urezki@gmail.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorcu/nocb: Check a stable offloaded state to manipulate qlen_last_fqs_check
Frederic Weisbecker [Tue, 19 Oct 2021 00:08:11 +0000 (02:08 +0200)]
rcu/nocb: Check a stable offloaded state to manipulate qlen_last_fqs_check

It's not entirely obvious why rdp->qlen_last_fqs_check is updated before
processing the queue only on offloaded rdp. There can be different
effect to that, either in favour of triggering the force quiescent state
path or not. For example:

1) If the number of callbacks has decreased since the last
   rdp->qlen_last_fqs_check update (because we recently called
   rcu_do_batch() and we executed below qhimark callbacks) and the number
   of processed callbacks on a subsequent do_batch() arranges for
   exceeding qhimark on non-offloaded but not on offloaded setup, then we
   may spare a later run to the force quiescent state
   slow path on __call_rcu_nocb_wake(), as compared to the non-offloaded
   counterpart scenario.

   Here is such an offloaded scenario instance:

    qhimark = 1000
    rdp->last_qlen_last_fqs_check = 3000
    rcu_segcblist_n_cbs(rdp) = 2000

    rcu_do_batch() {
        if (offloaded)
            rdp->last_qlen_fqs_check = rcu_segcblist_n_cbs(rdp) // 2000
        // run 1000 callback
        rcu_segcblist_n_cbs(rdp) = 1000
        // Not updating rdp->qlen_last_fqs_check
        if (count < rdp->qlen_last_fqs_check - qhimark)
            rdp->qlen_last_fqs_check = count;
    }

    call_rcu() * 1001 {
        __call_rcu_nocb_wake() {
            // not taking the fqs slowpath:
            // rcu_segcblist_n_cbs(rdp) == 2001
            // rdp->qlen_last_fqs_check == 2000
            // qhimark == 1000
            if (len > rdp->qlen_last_fqs_check + qhimark)
                ...
    }

    In the case of a non-offloaded scenario, rdp->qlen_last_fqs_check
    would be 1000 and the fqs slowpath would have executed.

2) If the number of callbacks has increased since the last
   rdp->qlen_last_fqs_check update (because we recently queued below
   qhimark callbacks) and the number of callbacks executed in rcu_do_batch()
   doesn't exceed qhimark for either offloaded or non-offloaded setup,
   then it's possible that the offloaded scenario later run the force
   quiescent state slow path on __call_rcu_nocb_wake() while the
   non-offloaded doesn't.

    qhimark = 1000
    rdp->last_qlen_last_fqs_check = 3000
    rcu_segcblist_n_cbs(rdp) = 2000

    rcu_do_batch() {
        if (offloaded)
            rdp->last_qlen_last_fqs_check = rcu_segcblist_n_cbs(rdp) // 2000
        // run 100 callbacks
        // concurrent queued 100
        rcu_segcblist_n_cbs(rdp) = 2000
        // Not updating rdp->qlen_last_fqs_check
        if (count < rdp->qlen_last_fqs_check - qhimark)
            rdp->qlen_last_fqs_check = count;
    }

    call_rcu() * 1001 {
        __call_rcu_nocb_wake() {
            // Taking the fqs slowpath:
            // rcu_segcblist_n_cbs(rdp) == 3001
            // rdp->qlen_last_fqs_check == 2000
            // qhimark == 1000
            if (len > rdp->qlen_last_fqs_check + qhimark)
                ...
    }

    In the case of a non-offloaded scenario, rdp->qlen_last_fqs_check
    would be 3000 and the fqs slowpath would have executed.

The reason for updating rdp->qlen_last_fqs_check when invoking callbacks
for offloaded CPUs is that there is usually no point in waking up either
the rcuog or rcuoc kthreads while in this state.  After all, both threads
are prohibited from indefinite sleeps.

The exception is when some huge number of callbacks are enqueued while
rcu_do_batch() is in the midst of invoking, in which case interrupting
the rcuog kthread's timed sleep might get more callbacks set up for the
next grace period.

Reported-and-tested-by: Valentin Schneider <valentin.schneider@arm.com>
Tested-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Original-patch-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Cc: Valentin Schneider <valentin.schneider@arm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Joel Fernandes <joel@joelfernandes.org>
Cc: Boqun Feng <boqun.feng@gmail.com>
Cc: Neeraj Upadhyay <neeraju@codeaurora.org>
Cc: Uladzislau Rezki <urezki@gmail.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorcu/nocb: Make rcu_core() callbacks acceleration (de-)offloading safe
Frederic Weisbecker [Tue, 19 Oct 2021 00:08:10 +0000 (02:08 +0200)]
rcu/nocb: Make rcu_core() callbacks acceleration (de-)offloading safe

When callbacks are offloaded, the NOCB kthreads handle the callbacks
progression on behalf of rcu_core().

However during the (de-)offloading process, the kthread may not be
entirely up to the task. As a result some callbacks grace period
sequence number may remain stale for a while because rcu_core() won't
take care of them either.

Fix this with forcing callbacks acceleration from rcu_core() as long
as the offloading process isn't complete.

Reported-and-tested-by: Valentin Schneider <valentin.schneider@arm.com>
Tested-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Cc: Valentin Schneider <valentin.schneider@arm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Joel Fernandes <joel@joelfernandes.org>
Cc: Boqun Feng <boqun.feng@gmail.com>
Cc: Neeraj Upadhyay <neeraju@codeaurora.org>
Cc: Uladzislau Rezki <urezki@gmail.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorcu/nocb: Make rcu_core() callbacks acceleration preempt-safe
Thomas Gleixner [Tue, 19 Oct 2021 00:08:09 +0000 (02:08 +0200)]
rcu/nocb: Make rcu_core() callbacks acceleration preempt-safe

While reporting a quiescent state for a given CPU, rcu_core() takes
advantage of the freshly loaded grace period sequence number and the
locked rnp to accelerate the callbacks whose sequence number have been
assigned a stale value.

This action is only necessary when the rdp isn't offloaded, otherwise
the NOCB kthreads already take care of the callbacks progression.

However the check for the offloaded state is volatile because it is
performed outside the IRQs disabled section. It's possible for the
offloading process to preempt rcu_core() at that point on PREEMPT_RT.

This is dangerous because rcu_core() may end up accelerating callbacks
concurrently with NOCB kthreads without appropriate locking.

Fix this with moving the offloaded check inside the rnp locking section.

Reported-and-tested-by: Valentin Schneider <valentin.schneider@arm.com>
Reviewed-by: Valentin Schneider <valentin.schneider@arm.com>
Tested-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Joel Fernandes <joel@joelfernandes.org>
Cc: Boqun Feng <boqun.feng@gmail.com>
Cc: Neeraj Upadhyay <neeraju@codeaurora.org>
Cc: Uladzislau Rezki <urezki@gmail.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorcu/nocb: Invoke rcu_core() at the start of deoffloading
Frederic Weisbecker [Tue, 19 Oct 2021 00:08:08 +0000 (02:08 +0200)]
rcu/nocb: Invoke rcu_core() at the start of deoffloading

On PREEMPT_RT, if rcu_core() is preempted by the de-offloading process,
some work, such as callbacks acceleration and invocation, may be left
unattended due to the volatile checks on the offloaded state.

In the worst case this work is postponed until the next rcu_pending()
check that can take a jiffy to reach, which can be a problem in case
of callbacks flooding.

Solve that with invoking rcu_core() early in the de-offloading process.
This way any work dismissed by an ongoing rcu_core() call fooled by
a preempting deoffloading process will be caught up by a nearby future
recall to rcu_core(), this time fully aware of the de-offloading state.

Tested-by: Valentin Schneider <valentin.schneider@arm.com>
Tested-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Cc: Valentin Schneider <valentin.schneider@arm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Joel Fernandes <joel@joelfernandes.org>
Cc: Boqun Feng <boqun.feng@gmail.com>
Cc: Neeraj Upadhyay <neeraju@codeaurora.org>
Cc: Uladzislau Rezki <urezki@gmail.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorcu/nocb: Prepare state machine for a new step
Frederic Weisbecker [Tue, 19 Oct 2021 00:08:07 +0000 (02:08 +0200)]
rcu/nocb: Prepare state machine for a new step

Currently SEGCBLIST_SOFTIRQ_ONLY is a bit of an exception among the
segcblist flags because it is an exclusive state that doesn't mix up
with the other flags. Remove it in favour of:

_ A flag specifying that rcu_core() needs to perform callbacks execution
  and acceleration

and

_ A flag specifying we want the nocb lock to be held in any needed
  circumstances

This clarifies the code and is more flexible: It allows to have a state
where rcu_core() runs with locking while offloading hasn't started yet.
This is a necessary step to prepare for triggering rcu_core() at the
very beginning of the de-offloading process so that rcu_core() won't
dismiss work while being preempted by the de-offloading process, at
least not without a pending subsequent rcu_core() that will quickly
catch up.

Reviewed-by: Valentin Schneider <Valentin.Schneider@arm.com>
Tested-by: Valentin Schneider <valentin.schneider@arm.com>
Tested-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Joel Fernandes <joel@joelfernandes.org>
Cc: Boqun Feng <boqun.feng@gmail.com>
Cc: Neeraj Upadhyay <neeraju@codeaurora.org>
Cc: Uladzislau Rezki <urezki@gmail.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorcu/nocb: Make local rcu_nocb_lock_irqsave() safe against concurrent deoffloading
Frederic Weisbecker [Mon, 11 Oct 2021 14:51:30 +0000 (16:51 +0200)]
rcu/nocb: Make local rcu_nocb_lock_irqsave() safe against concurrent deoffloading

rcu_nocb_lock_irqsave() can be preempted between the call to
rcu_segcblist_is_offloaded() and the actual locking. This matters now
that rcu_core() is preemptible on PREEMPT_RT and the (de-)offloading
process can interrupt the softirq or the rcuc kthread.

As a result we may locklessly call into code that requires nocb locking.
In practice this is a problem while we accelerate callbacks on rcu_core().

Simply disabling interrupts before (instead of after) checking the NOCB
offload state fixes the issue.

Reported-and-tested-by: Valentin Schneider <valentin.schneider@arm.com>
Tested-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Cc: Valentin Schneider <valentin.schneider@arm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Joel Fernandes <joel@joelfernandes.org>
Cc: Boqun Feng <boqun.feng@gmail.com>
Cc: Neeraj Upadhyay <neeraju@codeaurora.org>
Cc: Uladzislau Rezki <urezki@gmail.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorcu: Tighten rcu_advance_cbs_nowake() checks
Paul E. McKenney [Fri, 17 Sep 2021 22:04:48 +0000 (15:04 -0700)]
rcu: Tighten rcu_advance_cbs_nowake() checks

Currently, rcu_advance_cbs_nowake() checks that a grace period is in
progress, however, that grace period could end just after the check.
This commit rechecks that a grace period is still in progress while
holding the rcu_node structure's lock.  The grace period cannot end while
the current CPU's rcu_node structure's ->lock is held, thus avoiding
false positives from the WARN_ON_ONCE().

As Daniel Vacek noted, it is not necessary for the rcu_node structure
to have a CPU that has not yet passed through its quiescent state.

Tested-by: Guillaume Morin <guillaume@morinfr.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorcu/exp: Mark current CPU as exp-QS in IPI loop second pass
Frederic Weisbecker [Tue, 30 Nov 2021 16:21:08 +0000 (17:21 +0100)]
rcu/exp: Mark current CPU as exp-QS in IPI loop second pass

Expedited RCU grace periods invoke sync_rcu_exp_select_node_cpus(), which
takes two passes over the leaf rcu_node structure's CPUs.  The first
pass gathers up the current CPU and CPUs that are in dynticks idle mode.
The workqueue will report a quiescent state on their behalf later.
The second pass sends IPIs to the rest of the CPUs, but excludes the
current CPU, incorrectly assuming it has been included in the first
pass's list of CPUs.

Unfortunately the current CPU may have changed between the first and
second pass, due to the fact that the various rcu_node structures'
->lock fields have been dropped, thus momentarily enabling preemption.
This means that if the second pass's CPU was not on the first pass's
list, it will be ignored completely.  There will be no IPI sent to
it, and there will be no reporting of quiescent states on its behalf.
Unfortunately, the expedited grace period will nevertheless be waiting
for that CPU to report a quiescent state, but with that CPU having no
reason to believe that such a report is needed.

The result will be an expedited grace period stall.

Fix this by no longer excluding the current CPU from consideration during
the second pass.

Fixes: b9ad4d6ed18e ("rcu: Avoid self-IPI in sync_rcu_exp_select_node_cpus()")
Reviewed-by: Neeraj Upadhyay <quic_neeraju@quicinc.com>
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Cc: Uladzislau Rezki <urezki@gmail.com>
Cc: Neeraj Upadhyay <quic_neeraju@quicinc.com>
Cc: Boqun Feng <boqun.feng@gmail.com>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Joel Fernandes <joel@joelfernandes.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorcu: Make idle entry report expedited quiescent states
Paul E. McKenney [Wed, 29 Sep 2021 18:09:34 +0000 (11:09 -0700)]
rcu: Make idle entry report expedited quiescent states

In non-preemptible kernels, an unfortunately timed expedited grace period
can result in the rcu_exp_handler() IPI handler setting the rcu_data
structure's cpu_no_qs.b.exp field just as the target CPU enters idle.
There are situations in which this field will not be checked until after
that CPU exits idle.  The resulting grace-period latency does not qualify
as "expedited".

This commit therefore checks this field upon non-preemptible idle entry in
the rcu_preempt_deferred_qs() function.  It also qualifies the rcu_core()
preempt_count() check with IS_ENABLED(CONFIG_PREEMPT_COUNT) to prevent
false-positive quiescent states from count-free kernels.

Reported-by: Neeraj Upadhyay <neeraju@codeaurora.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorcu: Prevent expedited GP from enabling tick on offline CPU
Paul E. McKenney [Wed, 29 Sep 2021 16:21:34 +0000 (09:21 -0700)]
rcu: Prevent expedited GP from enabling tick on offline CPU

If an RCU expedited grace period starts just when a CPU is in the process
of going offline, so that the outgoing CPU has completed its pass through
stop-machine but has not yet completed its final dive into the idle loop,
RCU will attempt to enable that CPU's scheduling-clock tick via a call
to tick_dep_set_cpu().  For this to happen, that CPU has to have been
online when the expedited grace period completed its CPU-selection phase.

This is pointless:  The outgoing CPU has interrupts disabled, so it cannot
take a scheduling-clock tick anyway.  In addition, the tick_dep_set_cpu()
function's eventual call to irq_work_queue_on() will splat as follows:

smpboot: CPU 1 is now offline
WARNING: CPU: 6 PID: 124 at kernel/irq_work.c:95
+irq_work_queue_on+0x57/0x60
Modules linked in:
CPU: 6 PID: 124 Comm: kworker/6:2 Not tainted 5.15.0-rc1+ #3
Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS
+rel-1.14.0-0-g155821a-rebuilt.opensuse.org 04/01/2014
Workqueue: rcu_gp wait_rcu_exp_gp
RIP: 0010:irq_work_queue_on+0x57/0x60
Code: 8b 05 1d c7 ea 62 a9 00 00 f0 00 75 21 4c 89 ce 44 89 c7 e8
+9b 37 fa ff ba 01 00 00 00 89 d0 c3 4c 89 cf e8 3b ff ff ff eb ee <0f> 0b eb b7
+0f 0b eb db 90 48 c7 c0 98 2a 02 00 65 48 03 05 91
 6f
RSP: 0000:ffffb12cc038fe48 EFLAGS: 00010282
RAX: 0000000000000001 RBX: 0000000000005208 RCX: 0000000000000020
RDX: 0000000000000001 RSI: 0000000000000001 RDI: ffff9ad01f45a680
RBP: 000000000004c990 R08: 0000000000000001 R09: ffff9ad01f45a680
R10: ffffb12cc0317db0 R11: 0000000000000001 R12: 00000000fffecee8
R13: 0000000000000001 R14: 0000000000026980 R15: ffffffff9e53ae00
FS:  0000000000000000(0000) GS:ffff9ad01f580000(0000)
+knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 0000000000000000 CR3: 000000000de0c000 CR4: 00000000000006e0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
Call Trace:
 tick_nohz_dep_set_cpu+0x59/0x70
 rcu_exp_wait_wake+0x54e/0x870
 ? sync_rcu_exp_select_cpus+0x1fc/0x390
 process_one_work+0x1ef/0x3c0
 ? process_one_work+0x3c0/0x3c0
 worker_thread+0x28/0x3c0
 ? process_one_work+0x3c0/0x3c0
 kthread+0x115/0x140
 ? set_kthread_struct+0x40/0x40
 ret_from_fork+0x22/0x30
---[ end trace c5bf75eb6aa80bc6 ]---

This commit therefore avoids invoking tick_dep_set_cpu() on offlined
CPUs to limit both futility and false-positive splats.

Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorcu: Mark sync_sched_exp_online_cleanup() ->cpu_no_qs.b.exp load
Paul E. McKenney [Tue, 28 Sep 2021 18:06:35 +0000 (11:06 -0700)]
rcu: Mark sync_sched_exp_online_cleanup() ->cpu_no_qs.b.exp load

The sync_sched_exp_online_cleanup() is called from rcutree_online_cpu(),
which can be invoked with interrupts enabled.  This means that
the ->cpu_no_qs.b.exp field is subject to data races from the
rcu_exp_handler() IPI handler, so this commit marks the load from
that field.

Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorcu: Remove rcu_data.exp_deferred_qs and convert to rcu_data.cpu no_qs.b.exp
Frederic Weisbecker [Thu, 16 Sep 2021 12:10:48 +0000 (14:10 +0200)]
rcu: Remove rcu_data.exp_deferred_qs and convert to rcu_data.cpu no_qs.b.exp

Having two fields for the same purpose with subtle differences on
different RCU flavours is confusing, especially when both fields always
exist on both RCU flavours.

Fortunately, it is now safe for preemptible RCU to rely on the rcu_data
structure's ->cpu_no_qs.b.exp field, just like non-preemptible RCU.
This commit therefore removes the ad-hoc ->exp_deferred_qs field.

Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorcu: Move rcu_data.cpu_no_qs.b.exp reset to rcu_export_exp_rdp()
Frederic Weisbecker [Thu, 16 Sep 2021 12:10:47 +0000 (14:10 +0200)]
rcu: Move rcu_data.cpu_no_qs.b.exp reset to rcu_export_exp_rdp()

On non-preemptible RCU, move clearing of the rcu_data structure's
->cpu_no_qs.b.exp filed to the actual expedited quiescent state report
function, matching hw preemptible RCU handles the ->exp_deferred_qs field.

This prepares for removing ->exp_deferred_qs in favor of ->cpu_no_qs.b.exp
for both preemptible and non-preemptible RCU.

Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorcu: Ignore rdp.cpu_no_qs.b.exp on preemptible RCU's rcu_qs()
Frederic Weisbecker [Thu, 16 Sep 2021 12:10:45 +0000 (14:10 +0200)]
rcu: Ignore rdp.cpu_no_qs.b.exp on preemptible RCU's rcu_qs()

Preemptible RCU does not use the rcu_data structure's ->cpu_no_qs.b.exp,
instead using a separate ->exp_deferred_qs field to record the need for
an expedited quiescent state.

In fact ->cpu_no_qs.b.exp should never be set in preemptible RCU because
preemptible RCU's expedited grace periods use other mechanisms to record
quiescent states.

This commit therefore removes the implicit rcu_qs() reference to
->cpu_no_qs.b.exp in favor of a direct reference to ->cpu_no_qs.b.norm.

Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agotools/memory-model: Use "-unroll 0" to keep --hw runs finite
Paul E. McKenney [Tue, 25 Jun 2019 05:30:32 +0000 (22:30 -0700)]
tools/memory-model: Use "-unroll 0" to keep --hw runs finite

Litmus tests involving atomic operations produce LL/SC loops on a number
of architectures, and unrolling these loops can result in excessive
verification times or even stack overflows.  This commit therefore uses
the "-unroll 0" herd7 argument to avoid unrolling, on the grounds that
additional passes through an LL/SC loop should not change the verification.

Note however, that certain bugs in the mapping of the LL/SC loop to
machine instructions may go undetected.  On the other hand, herd7 might
not be the best vehicle for finding such bugs in any case.  (You do
stress-test your architecture-specific code, don't you?)

Suggested-by: Luc Maranget <luc.maranget@inria.fr>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agotools/memory-model: Make judgelitmus.sh handle scripted Result: tag
Paul E. McKenney [Thu, 6 Jun 2019 09:13:27 +0000 (02:13 -0700)]
tools/memory-model: Make judgelitmus.sh handle scripted Result: tag

The scripts that generate the litmus tests in the "auto" directory of
the https://github.com/paulmckrcu/litmus archive place the "Result:"
tag into a single-line ocaml comment, which judgelitmus.sh currently
does not recognize.  This commit therefore makes judgelitmus.sh
recognize both the multiline comment format that it currently does
and the automatically generated single-line format.

Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agotools/memory-model: Add data-race capabilities to judgelitmus.sh
Paul E. McKenney [Fri, 3 May 2019 14:34:20 +0000 (07:34 -0700)]
tools/memory-model: Add data-race capabilities to judgelitmus.sh

This commit adds functionality to judgelitmus.sh to allow it to handle
both the "DATARACE" markers in the "Result:" comments in litmus tests
and the "Flag data-race" markers in LKMM output.  For C-language tests,
if either marker is present, the other must also be as well, at least for
litmus tests having a "Result:" comment.  If the LKMM output indicates
a data race, then failures of the Always/Sometimes/Never portion of the
"Result:" prediction are forgiven.

The reason for forgiving "Result:" mispredictions is that data races can
result in "interesting" compiler optimizations, so that all bets are off
in the data-race case.

[ paulmck: Apply Akira Yokosawa feedback. ]
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agotools/memory-model: Add checktheselitmus.sh to run specified litmus tests
Paul E. McKenney [Thu, 2 May 2019 17:05:14 +0000 (10:05 -0700)]
tools/memory-model: Add checktheselitmus.sh to run specified litmus tests

This commit adds a checktheselitmus.sh script that runs the litmus tests
specified on the command line.  This is useful for verifying fixes to
specific litmus tests.

Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agotools/memory-model: Repair parseargs.sh header comment
Paul E. McKenney [Thu, 2 May 2019 17:03:29 +0000 (10:03 -0700)]
tools/memory-model: Repair parseargs.sh header comment

Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agotools/memory-model: Add "--" to parseargs.sh for additional arguments
Paul E. McKenney [Thu, 2 May 2019 16:51:57 +0000 (09:51 -0700)]
tools/memory-model:  Add "--" to parseargs.sh for additional arguments

Currently, parseargs.sh expects to consume all the command-line arguments,
which prevents the calling script from having any of its own arguments.
This commit therefore causes parseargs.sh to stop consuming arguments
when it encounters a "--" argument, leaving any remaining arguments for
the calling script.

Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agotools/memory-model: Make history-check scripts use mselect7
Paul E. McKenney [Mon, 8 Apr 2019 17:02:23 +0000 (10:02 -0700)]
tools/memory-model: Make history-check scripts use mselect7

The history-check scripts currently use grep to ignore non-C-language
litmus tests, which is a bit fragile.  This commit therefore enlists the
aid of "mselect7 -arch C", given Luc Maraget's recent modifications that
allow mselect7 to operate in filter mode.

This change requires herdtools 7.52-32-g1da3e0e50977 or later.

Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agotools/memory-model: Make checkghlitmus.sh use mselect7
Paul E. McKenney [Mon, 8 Apr 2019 16:27:28 +0000 (09:27 -0700)]
tools/memory-model: Make checkghlitmus.sh use mselect7

The checkghlitmus.sh script currently uses grep to ignore non-C-language
litmus tests, which is a bit fragile.  This commit therefore enlists the
aid of "mselect7 -arch C", given Luc Maraget's recent modifications that
allow mselect7 to operate in filter mode.

This change requires herdtools 7.52-32-g1da3e0e50977 or later.

Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agotools/memory-model: Fix scripting --jobs argument
Paul E. McKenney [Wed, 27 Mar 2019 18:47:14 +0000 (11:47 -0700)]
tools/memory-model: Fix scripting --jobs argument

The parseargs.sh regular expression for the --jobs argument incorrectly
requires that the number of jobs be at least 10, that is, have at least
two digits.  This commit therefore adjusts this regular expression to
allow single-digit numbers of jobs to be specified.

Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agotools/memory-model: Implement --hw support for checkghlitmus.sh
Paul E. McKenney [Sat, 23 Mar 2019 00:18:43 +0000 (17:18 -0700)]
tools/memory-model: Implement --hw support for checkghlitmus.sh

This commits enables the "--hw" argument for the checkghlitmus.sh script,
causing it to convert any applicable C-language litmus tests to the
specified flavor of assembly language, to verify these assembly-language
litmus tests, and checking compatibility of the outcomes.

Note that the conversion does not yet handle locking, RCU, SRCU, plain
C-language memory accesses, or casts.

Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agotools/memory-model: Add -v flag to jingle7 runs
Paul E. McKenney [Fri, 5 Apr 2019 19:34:56 +0000 (12:34 -0700)]
tools/memory-model: Add -v flag to jingle7 runs

Adding the -v flag to jingle7 invocations gives much useful information
on why jingle7 didn't like a given litmus test.  This commit therefore
adds this flag and saves off any such information into a .err file.

Suggested-by: Luc Maranget <luc.maranget@inria.fr>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agotools/memory-model: Make runlitmus.sh check for jingle errors
Paul E. McKenney [Tue, 26 Mar 2019 00:20:51 +0000 (17:20 -0700)]
tools/memory-model: Make runlitmus.sh check for jingle errors

It turns out that the jingle7 tool is currently a bit picky about
the litmus tests it is willing to process.  This commit therefore
ensures that jingle7 failures are reported.

Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agotools/memory-model: Allow herd to deduce CPU type
Paul E. McKenney [Fri, 22 Mar 2019 15:57:20 +0000 (08:57 -0700)]
tools/memory-model: Allow herd to deduce CPU type

Currently, the scripts specify the CPU's .cat file to herd.  But this is
pointless because herd will select a good and sufficient .cat file from
the assembly-language litmus test itself.  This commit therefore removes
the -model argument to herd, allowing herd to figure the CPU family out
itself.

Note that the user can override herd's choice using the "--herdopts"
argument to the scripts.

Suggested-by: Luc Maranget <luc.maranget@inria.fr>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agotools/memory-model: Keep assembly-language litmus tests
Paul E. McKenney [Thu, 21 Mar 2019 21:44:09 +0000 (14:44 -0700)]
tools/memory-model: Keep assembly-language litmus tests

This commit retains the assembly-language litmus tests generated from
the C-language litmus tests, appending the hardware tag to the original
C-language litmus test's filename.  Thus, S+poonceonces.litmus.AArch64
contains the Armv8 assembly language corresponding to the C-language
S+poonceonces.litmus test.

This commit also updates the .gitignore to avoid committing these
automatically generated assembly-language litmus tests.

Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agotools/memory-model: Move from .AArch64.litmus.out to .litmus.AArch.out
Paul E. McKenney [Thu, 21 Mar 2019 21:06:27 +0000 (14:06 -0700)]
tools/memory-model: Move from .AArch64.litmus.out to .litmus.AArch.out

When the github scripts see ".litmus.out", they assume that there must be
a corresponding C-language ".litmus" file.  Won't they be disappointed
when they instead see nothing, or, worse yet, the corresponding
assembly-language litmus test?  This commit therefore swaps the hardware
tag with the "litmus" to avoid this sort of disappointment.

This commit also adjusts the .gitignore file so as to avoid adding these
new ".out" files to git.

[ paulmck: Apply Akira Yokosawa feedback. ]
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agotools/memory-model: Make runlitmus.sh generate .litmus.out for --hw
Paul E. McKenney [Wed, 20 Mar 2019 23:41:41 +0000 (16:41 -0700)]
tools/memory-model: Make runlitmus.sh generate .litmus.out for --hw

In the absence of "Result:" comments, the runlitmus.sh script relies on
litmus.out files from prior LKMM runs.  This can be a bit user-hostile,
so this commit makes runlitmus.sh generate any needed .litmus.out files
that don't already exist.

Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agotools/memory-model: Split runlitmus.sh out of checklitmus.sh
Paul E. McKenney [Wed, 20 Mar 2019 21:57:56 +0000 (14:57 -0700)]
tools/memory-model: Split runlitmus.sh out of checklitmus.sh

This commit prepares for adding --hw capability to github litmus-test
scripts by splitting runlitmus.sh (which simply runs the verification)
out of checklitmus.sh (which also judges the results).

Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agotools/memory-model: Make judgelitmus.sh ransack .litmus.out files
Paul E. McKenney [Wed, 20 Mar 2019 21:37:46 +0000 (14:37 -0700)]
tools/memory-model: Make judgelitmus.sh ransack .litmus.out files

The judgelitmus.sh script currently relies solely on the "Result:"
comment in the .litmus file.  This is problematic when using the --hw
argument, because it is necessary to check the hardware model against
LKMM even in the absence of "Result:" comments.

This commit therefore modifies judgelitmus.sh to check the observation
in a .litmus.out file, in case one was generated by a previous LKMM run.

Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agotools/memory-model: Hardware checking for check{,all}litmus.sh
Paul E. McKenney [Wed, 20 Mar 2019 19:39:27 +0000 (12:39 -0700)]
tools/memory-model: Hardware checking for check{,all}litmus.sh

This commit makes checklitmus.sh and checkalllitmus.sh check to see
if a hardware verification was specified (via the --hw command-line
argument, which sets the LKMM_HW_MAP_FILE environment variable).
If so, the C-language litmus test is converted to the specified type
of assembly-language litmus test and herd is run on it.  Hardware is
permitted to be stronger than LKMM requires, so "Always" and "Never"
verifications of "Sometimes" C-language litmus tests are forgiven.

Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agotools/memory-model: Fix checkalllitmus.sh comment
Paul E. McKenney [Tue, 19 Mar 2019 23:37:01 +0000 (16:37 -0700)]
tools/memory-model: Fix checkalllitmus.sh comment

The checkalllitmus.sh runs litmus tests in the litmus-tests directory,
not those in the github archive, so this commit updates the comment to
reflect this reality.

Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agotools/memory-model: Add simpletest.sh to check locking, RCU, and SRCU
Paul E. McKenney [Tue, 19 Mar 2019 23:21:09 +0000 (16:21 -0700)]
tools/memory-model: Add simpletest.sh to check locking, RCU, and SRCU

This commit abstracts out common function to check a given litmus test
for locking, RCU, and SRCU in order to avoid duplicating code.

Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agotools/memory-model: Make judgelitmus.sh handle hardware verifications
Paul E. McKenney [Tue, 19 Mar 2019 21:39:10 +0000 (14:39 -0700)]
tools/memory-model: Make judgelitmus.sh handle hardware verifications

This commit makes the judgelitmus.sh script check the --hw argument
(AKA the LKMM_HW_MAP_FILE environment variable) and to adjust its
judgment for a run where a C-language litmus test has been translated to
assembly and the assembly version verified.  In this case, the assembly
verification output is checked against the C-language script's "Result:"
comment.  However, because hardware can be stronger than LKMM requires,
the judgelitmus.sh script forgives verification mismatches featuring
a "Sometimes" in the C-language script and an "Always" or "Never"
assembly-language verification.

Note that deadlock is not forgiven, however, this should not normally be
an issue given that C-language tests containing locking, RCU, or SRCU
cannot be translated to assembly.  However, this issue can crop up in
litmus tests that mimic deadlock by using the "filter" clause to ignore
all executions.  It can also crop up when certain herd arguments are
used to autofilter everything that does not match the "exists" clause
in cases where the "exists" clause cannot be satisfied.

Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agotools/memory-model: Update parseargs.sh for hardware verification
Paul E. McKenney [Tue, 19 Mar 2019 22:59:26 +0000 (15:59 -0700)]
tools/memory-model: Update parseargs.sh for hardware verification

This commit adds a --hw argument to parseargs.sh to specify the CPU
family for a hardware verification.  For example, "--hw AArch64" will
specify that a C-language litmus test is to be translated to ARMv8 and
the result verified.  This will set the LKMM_HW_MAP_FILE environment
variable accordingly.  If there is no --hw argument, this environment
variable will be set to the empty string.

Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agotools/memory-model: Fix paulmck email address on pre-existing scripts
Paul E. McKenney [Thu, 11 Apr 2019 14:33:18 +0000 (07:33 -0700)]
tools/memory-model: Fix paulmck email address on pre-existing scripts

Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agotools/memory-model: Make judgelitmus.sh detect hard deadlocks
Paul E. McKenney [Tue, 19 Mar 2019 21:27:06 +0000 (14:27 -0700)]
tools/memory-model: Make judgelitmus.sh detect hard deadlocks

If a litmus test specifies "Result: Never" and if it contains an
unconditional ("hard") deadlock, then running checklitmus.sh on it will
not flag any errors, despite the fact that there are no executions.
This commit therefore updates judgelitmus.sh to complain about tests
with no executions that are marked, but not as "Result: DEADLOCK".

Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agotools/memory-model: Make judgelitmus.sh identify bad macros
Paul E. McKenney [Mon, 18 Mar 2019 20:40:57 +0000 (13:40 -0700)]
tools/memory-model: Make judgelitmus.sh identify bad macros

Currently, judgelitmus.sh treats use of unknown primitives (such as
srcu_read_lock() prior to SRCU support) as "!!! Verification error".
This can be misleading because it fails to call out typos and running
a version LKMM on a litmus test requiring a feature not provided by
that version.  This commit therefore changes judgelitmus.sh to check
for unknown primitives and to report them, for example, with:

'!!! Current LKMM version does not know "rcu_write_lock"'.

Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agotools/memory-model: Make cmplitmushist.sh note timeouts
Paul E. McKenney [Mon, 18 Mar 2019 20:07:46 +0000 (13:07 -0700)]
tools/memory-model: Make cmplitmushist.sh note timeouts

Currently, cmplitmushist.sh treats timeouts (as in the "--timeout"
argument) as "Missing Observation line".  This can be misleading because
it is quite possible that running the test longer would have produced
a verification.  This commit therefore changes cmplitmushist.sh to check
for timeouts and to report them with "Timed out".

Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agotools/memory-model: Make judgelitmus.sh note timeouts
Paul E. McKenney [Mon, 18 Mar 2019 18:53:50 +0000 (11:53 -0700)]
tools/memory-model: Make judgelitmus.sh note timeouts

Currently, judgelitmus.sh treats timeouts (as in the "--timeout" argument)
as "!!! Verification error".  This can be misleading because it is quite
possible that running the test longer would have produced a verification.
This commit therefore changes judgelitmus.sh to check for timeouts and
to report them with "!!! Timeout".

Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agotools/memory-model: Document locking corner cases
Paul E. McKenney [Fri, 14 Aug 2020 23:14:34 +0000 (16:14 -0700)]
tools/memory-model:  Document locking corner cases

Most Linux-kernel uses of locking are straightforward, but there are
corner-case uses that rely on less well-known aspects of the lock and
unlock primitives.  This commit therefore adds a locking.txt and litmus
tests in Documentation/litmus-tests/locking to explain these corner-case
uses.

Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agotools/memory-model: litmus: Add two tests for unlock(A)+lock(B) ordering
Boqun Feng [Mon, 25 Oct 2021 14:54:16 +0000 (22:54 +0800)]
tools/memory-model: litmus: Add two tests for unlock(A)+lock(B) ordering

The memory model has been updated to provide a stronger ordering
guarantee for unlock(A)+lock(B) on the same CPU/thread. Therefore add
two litmus tests describing this new guarantee, these tests are simple
yet can clearly show the usage of the new guarantee, also they can serve
as the self tests for the modification in the model.

Co-developed-by: Alan Stern <stern@rowland.harvard.edu>
Signed-off-by: Alan Stern <stern@rowland.harvard.edu>
Signed-off-by: Boqun Feng <boqun.feng@gmail.com>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agotools/memory-model: doc: Describe the requirement of the litmus-tests directory
Boqun Feng [Mon, 25 Oct 2021 14:54:15 +0000 (22:54 +0800)]
tools/memory-model: doc: Describe the requirement of the litmus-tests directory

It's better that we have some "standard" about which test should be put
in the litmus-tests directory because it helps future contributors
understand whether they should work on litmus-tests in kernel or Paul's
GitHub repo. Therefore explain a little bit on what a "representative"
litmus test is.

Signed-off-by: Boqun Feng <boqun.feng@gmail.com>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agotools/memory-model: Provide extra ordering for unlock+lock pair on the same CPU
Boqun Feng [Mon, 25 Oct 2021 14:54:14 +0000 (22:54 +0800)]
tools/memory-model: Provide extra ordering for unlock+lock pair on the same CPU

A recent discussion[1] shows that we are in favor of strengthening the
ordering of unlock + lock on the same CPU: a unlock and a po-after lock
should provide the so-called RCtso ordering, that is a memory access S
po-before the unlock should be ordered against a memory access R
po-after the lock, unless S is a store and R is a load.

The strengthening meets programmers' expection that "sequence of two
locked regions to be ordered wrt each other" (from Linus), and can
reduce the mental burden when using locks. Therefore add it in LKMM.

[1]: https://lore.kernel.org/lkml/20210909185937.GA12379@rowland.harvard.edu/

Co-developed-by: Alan Stern <stern@rowland.harvard.edu>
Signed-off-by: Alan Stern <stern@rowland.harvard.edu>
Signed-off-by: Boqun Feng <boqun.feng@gmail.com>
Reviewed-by: Michael Ellerman <mpe@ellerman.id.au> (powerpc)
Acked-by: Palmer Dabbelt <palmerdabbelt@google.com> (RISC-V)
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agotorture: Properly redirect kvm-remote.sh "echo" commands
Paul E. McKenney [Tue, 30 Nov 2021 19:47:37 +0000 (11:47 -0800)]
torture: Properly redirect kvm-remote.sh "echo" commands

The echo commands following initialization of the "oldrun" variable need
to be "tee"d to $oldrun/remote-log.  This commit fixes several stragglers.

Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agotorture: Fix incorrectly redirected "exit" in kvm-remote.sh
Paul E. McKenney [Tue, 30 Nov 2021 19:33:01 +0000 (11:33 -0800)]
torture: Fix incorrectly redirected "exit" in kvm-remote.sh

The "exit 4" in kvm-remote.sh is pointlessly redirected, so this commit
removes the redirection.

Fixes: 0092eae4cb4e ("torture: Add kvm-remote.sh script for distributed rcutorture test runs")
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorcutorture: Test RCU Tasks lock-contention detection
Paul E. McKenney [Thu, 25 Nov 2021 00:04:51 +0000 (16:04 -0800)]
rcutorture: Test RCU Tasks lock-contention detection

This commit adjusts the TRACE02 scenario to use a pair of callback-flood
kthreads.  This in turn forces lock contention on the single RCU Tasks
Trace callback queue, which forces use of all CPUs' queues, thus testing
this transition.  (No, there is not yet any way to transition back.

Cc: Neeraj Upadhyay <neeraj.iitr10@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorcutorture: Cause TREE02 and TREE10 scenarios to do more callback flooding
Paul E. McKenney [Tue, 23 Nov 2021 05:02:43 +0000 (21:02 -0800)]
rcutorture: Cause TREE02 and TREE10 scenarios to do more callback flooding

This commit enables two callback-flood kthreads for the TREE02 scenario
and 28 for the TREE10 scenario.

Cc: Neeraj Upadhyay <neeraj.iitr10@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agotorture: Retry download once before giving up
Paul E. McKenney [Mon, 22 Nov 2021 20:47:50 +0000 (12:47 -0800)]
torture: Retry download once before giving up

Currently, a transient network error can kill a run if it happens while
downloading the tarball to one of the target systems.  This commit
therefore does a 60-second wait and then a retry.  If further experience
indicates, a more elaborate mechanism might be used later.

Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agotorture: Make kvm-find-errors.sh report link-time undefined symbols
Paul E. McKenney [Wed, 10 Nov 2021 15:47:41 +0000 (07:47 -0800)]
torture: Make kvm-find-errors.sh report link-time undefined symbols

This commit makes kvm-find-errors.sh check for and report undefined
symbols that are detected at link time.

Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agotorture: Catch kvm.sh help text up with actual options
Paul E. McKenney [Tue, 21 Sep 2021 19:50:05 +0000 (12:50 -0700)]
torture: Catch kvm.sh help text up with actual options

This commit brings the kvm.sh script's help text up to date with recently
(and some not-so-recently) added parameters.

Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorefscale: Prevent buffer to pr_alert() being too long
Li Zhijian [Mon, 25 Oct 2021 03:26:57 +0000 (11:26 +0800)]
refscale: Prevent buffer to pr_alert() being too long

0Day/LKP observed that the refscale results fail to complete when larger
values of nrun (such as 300) are specified.  The problem is that printk()
can accept at most a 1024-byte buffer.  This commit therefore prints
the buffer whenever its length exceeds 800 bytes.

CC: Philip Li <philip.li@intel.com>
Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: Li Zhijian <lizhijian@cn.fujitsu.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorefscale: Simplify the errexit checkpoint
Li Zhijian [Mon, 25 Oct 2021 03:26:56 +0000 (11:26 +0800)]
refscale: Simplify the errexit checkpoint

There is only the one OOM error case in main_func(), so this commit
eliminates the errexit local variable in favor of a branch to cleanup
code.

Signed-off-by: Li Zhijian <lizhijian@cn.fujitsu.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
3 years agorcutorture: Suppress pi-lock-across read-unlock testing for Tiny SRCU
Paul E. McKenney [Sat, 25 Sep 2021 04:30:26 +0000 (21:30 -0700)]
rcutorture: Suppress pi-lock-across read-unlock testing for Tiny SRCU

Because Tiny srcu_read_unlock() directly calls swake_up_one(), lockdep
complains when a pi lock is held across that srcu_read_unlock().
Although this is a lockdep false positive (there is no other CPU to
complete the deadlock cycle), lockdep is what it is at the moment.
This commit therefore prevents rcutorture from holding pi lock across
a Tiny srcu_read_unlock().

Signed-off-by: Paul E. McKenney <paulmck@kernel.org>