The per-CPU rcu_data structure's ->defer_qs_iw field is initialized
by IRQ_WORK_INIT_HARD(), which means that the subsequent invocation of
rcu_preempt_deferred_qs_handler() will always be executed with interrupts
disabled. This commit therefore removes the local_irq_save/restore()
operations from rcu_preempt_deferred_qs_handler() and adds a call to
lockdep_assert_irqs_disabled() in order to enable lockdep to diagnose
mistaken invocations of this function from interrupts-enabled code.
Signed-off-by: Zqiang <qiang.zhang@linux.dev>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
RCU re-initializes the deferred QS irq work everytime before attempting
to queue it. However there are situations where the irq work is
attempted to be queued even though it is already queued. In that case
re-initializing messes-up with the irq work queue that is about to be
handled.
The chances for that to happen are higher when the architecture doesn't
support self-IPIs and irq work are then all lazy, such as with the
following sequence:
1) rcu_read_unlock() is called when IRQs are disabled and there is a
grace period involving blocked tasks on the node. The irq work
is then initialized and queued.
2) The related tasks are unblocked and the CPU quiescent state
is reported. rdp->defer_qs_iw_pending is reset to DEFER_QS_IDLE,
allowing the irq work to be requeued in the future (note the previous
one hasn't fired yet).
3) A new grace period starts and the node has blocked tasks.
4) rcu_read_unlock() is called when IRQs are disabled again. The irq work
is re-initialized (but it's queued! and its node is cleared) and
requeued. Which means it's requeued to itself.
5) The irq work finally fires with the tick. But since it was requeued
to itself, it loops and hangs.
Fix this with initializing the irq work only once before the CPU boots.
Fixes: b41642c877 ("rcu: Fix rcu_read_unlock() deadloop due to IRQ work")
Reported-by: kernel test robot <oliver.sang@intel.com>
Closes: https://lore.kernel.org/oe-lkp/202508071303.c1134cce-lkp@intel.com
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Reviewed-by: Joel Fernandes <joelagnelf@nvidia.com>
Signed-off-by: Neeraj Upadhyay (AMD) <neeraj.upadhyay@kernel.org>
Extract the complex expedited handling condition in rcu_read_unlock_special()
into a separate function rcu_unlock_needs_exp_handling() with detailed
comments explaining each condition.
This improves code readability. No functional change intended.
Reviewed-by: "Paul E. McKenney" <paulmck@kernel.org>
Signed-off-by: Joel Fernandes <joelagnelf@nvidia.com>
Signed-off-by: Neeraj Upadhyay (AMD) <neeraj.upadhyay@kernel.org>
During rcu_read_unlock_special(), if this happens during irq_exit(), we
can lockup if an IPI is issued. This is because the IPI itself triggers
the irq_exit() path causing a recursive lock up.
This is precisely what Xiongfeng found when invoking a BPF program on
the trace_tick_stop() tracepoint As shown in the trace below. Fix by
managing the irq_work state correctly.
irq_exit()
__irq_exit_rcu()
/* in_hardirq() returns false after this */
preempt_count_sub(HARDIRQ_OFFSET)
tick_irq_exit()
tick_nohz_irq_exit()
tick_nohz_stop_sched_tick()
trace_tick_stop() /* a bpf prog is hooked on this trace point */
__bpf_trace_tick_stop()
bpf_trace_run2()
rcu_read_unlock_special()
/* will send a IPI to itself */
irq_work_queue_on(&rdp->defer_qs_iw, rdp->cpu);
A simple reproducer can also be obtained by doing the following in
tick_irq_exit(). It will hang on boot without the patch:
static inline void tick_irq_exit(void)
{
+ rcu_read_lock();
+ WRITE_ONCE(current->rcu_read_unlock_special.b.need_qs, true);
+ rcu_read_unlock();
+
Reported-by: Xiongfeng Wang <wangxiongfeng2@huawei.com>
Closes: https://lore.kernel.org/all/9acd5f9f-6732-7701-6880-4b51190aa070@huawei.com/
Tested-by: Qi Xi <xiqi2@huawei.com>
Signed-off-by: Joel Fernandes <joelagnelf@nvidia.com>
Reviewed-by: "Paul E. McKenney" <paulmck@kernel.org>
Reported-by: Linux Kernel Functional Testing <lkft@linaro.org>
[neeraj: Apply Frederic's suggested fix for PREEMPT_RT]
Signed-off-by: Neeraj Upadhyay (AMD) <neeraj.upadhyay@kernel.org>
On kernels built with CONFIG_IRQ_WORK=y, when rcu_read_unlock() is
invoked within an interrupts-disabled region of code [1], it will invoke
rcu_read_unlock_special(), which uses an irq-work handler to force the
system to notice when the RCU read-side critical section actually ends.
That end won't happen until interrupts are enabled at the soonest.
In some kernels, such as those booted with rcutree.use_softirq=y, the
irq-work handler is used unconditionally.
The per-CPU rcu_data structure's ->defer_qs_iw_pending field is
updated by the irq-work handler and is both read and updated by
rcu_read_unlock_special(). This resulted in the following KCSAN splat:
------------------------------------------------------------------------
BUG: KCSAN: data-race in rcu_preempt_deferred_qs_handler / rcu_read_unlock_special
read to 0xffff96b95f42d8d8 of 1 bytes by task 90 on cpu 8:
rcu_read_unlock_special+0x175/0x260
__rcu_read_unlock+0x92/0xa0
rt_spin_unlock+0x9b/0xc0
__local_bh_enable+0x10d/0x170
__local_bh_enable_ip+0xfb/0x150
rcu_do_batch+0x595/0xc40
rcu_cpu_kthread+0x4e9/0x830
smpboot_thread_fn+0x24d/0x3b0
kthread+0x3bd/0x410
ret_from_fork+0x35/0x40
ret_from_fork_asm+0x1a/0x30
write to 0xffff96b95f42d8d8 of 1 bytes by task 88 on cpu 8:
rcu_preempt_deferred_qs_handler+0x1e/0x30
irq_work_single+0xaf/0x160
run_irq_workd+0x91/0xc0
smpboot_thread_fn+0x24d/0x3b0
kthread+0x3bd/0x410
ret_from_fork+0x35/0x40
ret_from_fork_asm+0x1a/0x30
no locks held by irq_work/8/88.
irq event stamp: 200272
hardirqs last enabled at (200272): [<ffffffffb0f56121>] finish_task_switch+0x131/0x320
hardirqs last disabled at (200271): [<ffffffffb25c7859>] __schedule+0x129/0xd70
softirqs last enabled at (0): [<ffffffffb0ee093f>] copy_process+0x4df/0x1cc0
softirqs last disabled at (0): [<0000000000000000>] 0x0
------------------------------------------------------------------------
The problem is that irq-work handlers run with interrupts enabled, which
means that rcu_preempt_deferred_qs_handler() could be interrupted,
and that interrupt handler might contain an RCU read-side critical
section, which might invoke rcu_read_unlock_special(). In the strict
KCSAN mode of operation used by RCU, this constitutes a data race on
the ->defer_qs_iw_pending field.
This commit therefore disables interrupts across the portion of the
rcu_preempt_deferred_qs_handler() that updates the ->defer_qs_iw_pending
field. This suffices because this handler is not a fast path.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Neeraj Upadhyay (AMD) <neeraj.upadhyay@kernel.org>
A full memory barrier in the RCU-PREEMPT task unblock path advertizes
to order the context switch (or rather the accesses prior to
rcu_read_unlock()) with the expedited grace period fastpath.
However the grace period can not complete without the rnp calling into
rcu_report_exp_rnp() with the node locked. This reports the quiescent
state in a fully ordered fashion against updater's accesses thanks to:
1) The READ-SIDE smp_mb__after_unlock_lock() barrier across nodes
locking while propagating QS up to the root.
2) The UPDATE-SIDE smp_mb__after_unlock_lock() barrier while holding the
the root rnp to wait/check for the GP completion.
3) The (perhaps redundant given step 1) and 2)) smp_mb() in rcu_seq_end()
before the grace period completes.
This makes the explicit barrier in this place superfluous. Therefore
remove it as it is confusing.
Acked-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Joel Fernandes <joelagnelf@nvidia.com>
Signed-off-by: Neeraj Upadhyay (AMD) <neeraj.upadhyay@kernel.org>
For built with CONFIG_PROVE_RCU=y and CONFIG_PREEMPT_RT=y kernels,
Disable BH does not change the SOFTIRQ corresponding bits in
preempt_count(), but change current->softirq_disable_cnt, this
resulted in the following splat:
WARNING: suspicious RCU usage
kernel/rcu/tree_plugin.h:36 Unsafe read of RCU_NOCB offloaded state!
stack backtrace:
CPU: 0 UID: 0 PID: 22 Comm: rcuc/0
Call Trace:
[ 0.407907] <TASK>
[ 0.407910] dump_stack_lvl+0xbb/0xd0
[ 0.407917] dump_stack+0x14/0x20
[ 0.407920] lockdep_rcu_suspicious+0x133/0x210
[ 0.407932] rcu_rdp_is_offloaded+0x1c3/0x270
[ 0.407939] rcu_core+0x471/0x900
[ 0.407942] ? lockdep_hardirqs_on+0xd5/0x160
[ 0.407954] rcu_cpu_kthread+0x25f/0x870
[ 0.407959] ? __pfx_rcu_cpu_kthread+0x10/0x10
[ 0.407966] smpboot_thread_fn+0x34c/0xa50
[ 0.407970] ? trace_preempt_on+0x54/0x120
[ 0.407977] ? __pfx_smpboot_thread_fn+0x10/0x10
[ 0.407982] kthread+0x40e/0x840
[ 0.407990] ? __pfx_kthread+0x10/0x10
[ 0.407994] ? rt_spin_unlock+0x4e/0xb0
[ 0.407997] ? rt_spin_unlock+0x4e/0xb0
[ 0.408000] ? __pfx_kthread+0x10/0x10
[ 0.408006] ? __pfx_kthread+0x10/0x10
[ 0.408011] ret_from_fork+0x40/0x70
[ 0.408013] ? __pfx_kthread+0x10/0x10
[ 0.408018] ret_from_fork_asm+0x1a/0x30
[ 0.408042] </TASK>
Currently, triggering an rdp offloaded state change need the
corresponding rdp's CPU goes offline, and at this time the rcuc
kthreads has already in parking state. this means the corresponding
rcuc kthreads can safely read offloaded state of rdp while it's
corresponding cpu is online.
This commit therefore add softirq_count() check for
Preempt-RT kernels.
Suggested-by: Joel Fernandes <joelagnelf@nvidia.com>
Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Zqiang <qiang.zhang1211@gmail.com>
Signed-off-by: Joel Fernandes <joelagnelf@nvidia.com>
With PREEMPT_RCU=n, cond_resched() provides urgently needed quiescent
states for read-side critical sections via rcu_all_qs().
One reason why this was needed: lacking preempt-count, the tick
handler has no way of knowing whether it is executing in a
read-side critical section or not.
With (PREEMPT_LAZY=y, PREEMPT_DYNAMIC=n), we get (PREEMPT_COUNT=y,
PREEMPT_RCU=n). In this configuration cond_resched() is a stub and
does not provide quiescent states via rcu_all_qs().
(PREEMPT_RCU=y provides this information via rcu_read_unlock() and
its nesting counter.)
So, use the availability of preempt_count() to report quiescent states
in rcu_flavor_sched_clock_irq().
Suggested-by: Paul E. McKenney <paulmck@kernel.org>
Reviewed-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Ankur Arora <ankur.a.arora@oracle.com>
Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Boqun Feng <boqun.feng@gmail.com>
rcu_read_unlock_strict() can be called with preemption enabled
which can make for an unstable rdp and a racy norm value.
Fix this by dropping the preempt-count in __rcu_read_unlock()
after the call to rcu_read_unlock_strict(), adjusting the
preempt-count check appropriately.
Suggested-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Ankur Arora <ankur.a.arora@oracle.com>
Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Boqun Feng <boqun.feng@gmail.com>
Pull kthread updates from Frederic Weisbecker:
"Kthreads affinity follow either of 4 existing different patterns:
1) Per-CPU kthreads must stay affine to a single CPU and never
execute relevant code on any other CPU. This is currently handled
by smpboot code which takes care of CPU-hotplug operations.
Affinity here is a correctness constraint.
2) Some kthreads _have_ to be affine to a specific set of CPUs and
can't run anywhere else. The affinity is set through
kthread_bind_mask() and the subsystem takes care by itself to
handle CPU-hotplug operations. Affinity here is assumed to be a
correctness constraint.
3) Per-node kthreads _prefer_ to be affine to a specific NUMA node.
This is not a correctness constraint but merely a preference in
terms of memory locality. kswapd and kcompactd both fall into this
category. The affinity is set manually like for any other task and
CPU-hotplug is supposed to be handled by the relevant subsystem so
that the task is properly reaffined whenever a given CPU from the
node comes up. Also care should be taken so that the node affinity
doesn't cross isolated (nohz_full) cpumask boundaries.
4) Similar to the previous point except kthreads have a _preferred_
affinity different than a node. Both RCU boost kthreads and RCU
exp kworkers fall into this category as they refer to "RCU nodes"
from a distinctly distributed tree.
Currently the preferred affinity patterns (3 and 4) have at least 4
identified users, with more or less success when it comes to handle
CPU-hotplug operations and CPU isolation. Each of which do it in its
own ad-hoc way.
This is an infrastructure proposal to handle this with the following
API changes:
- kthread_create_on_node() automatically affines the created kthread
to its target node unless it has been set as per-cpu or bound with
kthread_bind[_mask]() before the first wake-up.
- kthread_affine_preferred() is a new function that can be called
right after kthread_create_on_node() to specify a preferred
affinity different than the specified node.
When the preferred affinity can't be applied because the possible
targets are offline or isolated (nohz_full), the kthread is affine to
the housekeeping CPUs (which means to all online CPUs most of the time
or only the non-nohz_full CPUs when nohz_full= is set).
kswapd, kcompactd, RCU boost kthreads and RCU exp kworkers have been
converted, along with a few old drivers.
Summary of the changes:
- Consolidate a bunch of ad-hoc implementations of
kthread_run_on_cpu()
- Introduce task_cpu_fallback_mask() that defines the default last
resort affinity of a task to become nohz_full aware
- Add some correctness check to ensure kthread_bind() is always
called before the first kthread wake up.
- Default affine kthread to its preferred node.
- Convert kswapd / kcompactd and remove their halfway working ad-hoc
affinity implementation
- Implement kthreads preferred affinity
- Unify kthread worker and kthread API's style
- Convert RCU kthreads to the new API and remove the ad-hoc affinity
implementation"
* tag 'kthread-for-6.14-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/frederic/linux-dynticks:
kthread: modify kernel-doc function name to match code
rcu: Use kthread preferred affinity for RCU exp kworkers
treewide: Introduce kthread_run_worker[_on_cpu]()
kthread: Unify kthread_create_on_cpu() and kthread_create_worker_on_cpu() automatic format
rcu: Use kthread preferred affinity for RCU boost
kthread: Implement preferred affinity
mm: Create/affine kswapd to its preferred node
mm: Create/affine kcompactd to its preferred node
kthread: Default affine kthread to its preferred NUMA node
kthread: Make sure kthread hasn't started while binding it
sched,arm64: Handle CPU isolation on last resort fallback rq selection
arm64: Exclude nohz_full CPUs from 32bits el0 support
lib: test_objpool: Use kthread_run_on_cpu()
kallsyms: Use kthread_run_on_cpu()
soc/qman: test: Use kthread_run_on_cpu()
arm/bL_switcher: Use kthread_run_on_cpu()
Now that kthreads have an infrastructure to handle preferred affinity
against CPU hotplug and housekeeping cpumask, convert RCU boost to use
it instead of handling all the constraints by itself.
Acked-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
The value of rdp->cpu_no_qs.b.exp may be changed only by the corresponding
CPU, and that CPU is not even allowed to race with itself, for example,
via interrupt handlers. This commit therefore adds KCSAN exclusive-writer
assertions to check this constraint.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Cc: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Uladzislau Rezki (Sony) <urezki@gmail.com>
The context_tracking.state RCU_DYNTICKS subvariable has been renamed to
RCU_WATCHING, replace "dyntick_idle" into "eqs" to drop the dyntick
reference.
Signed-off-by: Valentin Schneider <vschneid@redhat.com>
Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Neeraj Upadhyay <neeraj.upadhyay@kernel.org>
The barrier_mutex is used currently to protect (de-)offloading
operations and prevent from nocb_lock locking imbalance in rcu_barrier()
and shrinker, and also from misordered RCU barrier invocation.
Now since RCU (de-)offloading is going to happen on offline CPUs, an RCU
barrier will have to be executed while transitionning from offloaded to
de-offloaded state. And this can't happen while holding the
barrier_mutex.
Introduce a NOCB mutex to protect (de-)offloading transitions. The
barrier_mutex is still held for now when necessary to avoid barrier
callbacks reordering and nocb_lock imbalance.
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Reviewed-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Neeraj Upadhyay <neeraj.upadhyay@kernel.org>
Checking for races against concurrent (de-)offloading implies the
creation of !CONFIG_RCU_NOCB_CPU stubs to check if each relevant lock
is held. For now this only implies the nocb_lock but more are to be
expected.
Create instead a NOCB specific version of RCU_LOCKDEP_WARN() to avoid
the proliferation of stubs.
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Reviewed-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Neeraj Upadhyay <neeraj.upadhyay@kernel.org>
If a CPU is running either a userspace application or a guest OS in
nohz_full mode, it is possible for a system call to occur just as an
RCU grace period is starting. If that CPU also has the scheduling-clock
tick enabled for any reason (such as a second runnable task), and if the
system was booted with rcutree.use_softirq=0, then RCU can add insult to
injury by awakening that CPU's rcuc kthread, resulting in yet another
task and yet more OS jitter due to switching to that task, running it,
and switching back.
In addition, in the common case where that system call is not of
excessively long duration, awakening the rcuc task is pointless.
This pointlessness is due to the fact that the CPU will enter an extended
quiescent state upon returning to the userspace application or guest OS.
In this case, the rcuc kthread cannot do anything that the main RCU
grace-period kthread cannot do on its behalf, at least if it is given
a few additional milliseconds (for example, given the time duration
specified by rcutree.jiffies_till_first_fqs, give or take scheduling
delays).
This commit therefore adds a rcutree.nohz_full_patience_delay kernel
boot parameter that specifies the grace period age (in milliseconds,
rounded to jiffies) before which RCU will refrain from awakening the
rcuc kthread. Preliminary experimentation suggests a value of 1000,
that is, one second. Increasing rcutree.nohz_full_patience_delay will
increase grace-period latency and in turn increase memory footprint,
so systems with constrained memory might choose a smaller value.
Systems with less-aggressive OS-jitter requirements might choose the
default value of zero, which keeps the traditional immediate-wakeup
behavior, thus avoiding increases in grace-period latency.
[ paulmck: Apply Leonardo Bras feedback. ]
Link: https://lore.kernel.org/all/20240328171949.743211-1-leobras@redhat.com/
Reported-by: Leonardo Bras <leobras@redhat.com>
Suggested-by: Leonardo Bras <leobras@redhat.com>
Suggested-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Reviewed-by: Leonardo Bras <leobras@redhat.com>
Upon NOCB deoffloading, the rcuo kthread must be forced to sleep
until the corresponding rdp is ever offloaded again. The deoffloader
clears the SEGCBLIST_OFFLOADED flag, wakes up the rcuo kthread which
then notices that change and clears in turn its SEGCBLIST_KTHREAD_CB
flag before going to sleep, until it ever sees the SEGCBLIST_OFFLOADED
flag again, should a re-offloading happen.
Upon NOCB offloading, the rcuo kthread must be forced to wake up and
handle callbacks until the corresponding rdp is ever deoffloaded again.
The offloader sets the SEGCBLIST_OFFLOADED flag, wakes up the rcuo
kthread which then notices that change and sets in turn its
SEGCBLIST_KTHREAD_CB flag before going to check callbacks, until it
ever sees the SEGCBLIST_OFFLOADED flag cleared again, should a
de-offloading happen again.
This is all a crude ad-hoc and error-prone kthread (un-)parking
re-implementation.
Consolidate the behaviour with the appropriate API instead.
[ paulmck: Apply Qiang Zhang feedback provided in Link: below. ]
Link: https://lore.kernel.org/all/20240509074046.15629-1-qiang.zhang1211@gmail.com/
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Currently, there are rcu_data structure fields named ->rcu_onl_gp_seq
and ->rcu_ofl_gp_seq that track the rcu_state.gp_flags field at the
time of the corresponding CPU's last online or offline operation,
respectively. However, this information is not particularly useful.
It would be better to instead track the grace period state kept
in rcu_state.gp_state. This would also be consistent with the
initialization in rcu_boot_init_percpu_data(), which is to RCU_GP_CLEANED
(an rcu_state.gp_state value), and also with the diagnostics in
rcu_implicit_dynticks_qs(), whose format is consistent with an integer,
not a bitmask.
This commit therefore makes this change and changes the names to
->rcu_onl_gp_flags and ->rcu_ofl_gp_flags, respectively.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Uladzislau Rezki (Sony) <urezki@gmail.com>
Affine the parallel expedited gp kworkers to their respective RCU node
in order to make them close to the cache their are playing with.
This reuses the boost kthreads machinery that probe into CPU hotplug
operations such that the kthreads become/stay affine to their respective
node as soon/long as they contain online CPUs. Otherwise and if the
current CPU going down was the last online on the leaf node, the related
kthread is affine to the housekeeping CPUs.
In the long run, this affinity VS CPU hotplug operation game should
probably be implemented at the generic kthread level.
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Reviewed-by: Paul E. McKenney <paulmck@kernel.org>
[boqun: s/* rcu_boost_task/*rcu_boost_task as reported by checkpatch]
Signed-off-by: Boqun Feng <boqun.feng@gmail.com>
