On Wed, Aug 05, 2026 at 05:23:38AM -0700, Puranjay Mohan wrote:
> RCU's per-CPU callback list is only touched with interrupts disabled:
> the enqueue runs under local_irq_save() (and the nocb locks when
> offloaded), as do callback invocation and grace-period work. A
> call_rcu() that arrives with interrupts already disabled, whether from an
> NMI or from instrumentation that re-enters RCU, can interrupt one of
> those and corrupt the list or deadlock.
>
> Handle it by deferring: stage the callback on a per-CPU llist and raise
> an irq_work that re-issues it once interrupts are on. The re-issue goes
> straight to the enqueue so it cannot defer again. Callers that only hold
> interrupts off are deferred too, which is harmless. Skip the gate while
> the scheduler is down (RCU_SCHEDULER_INACTIVE), since irq_work is not
> usable that early and rcu_init() already calls call_rcu().
>
> rcu_barrier() flushes deferred callbacks before it scans the lists by
> draining every CPU's ->defer_head itself, and rcutree_migrate_callbacks()
> drains an outgoing CPU's. ->defer_lock is held across llist_del_all() and
> the re-issue, so these drainers serialize: a drain that finds the list
> empty knows any re-issue in flight has already reached a callback list.
>
> A deferred callback is re-issued without the lazy hint: it has already
> waited for the drain, so batching it further would only add latency.
>
> The re-issue runs with interrupts disabled, so instrumentation on the
> enqueue path can re-enter call_rcu(), stage another callback, and
> re-raise the irq_work, livelocking the drain. A per-CPU flag guards it:
> a call_rcu() that tries to defer while the drain is running, and is not
> from an NMI, is dropped with WARN_ONCE() instead of re-queued. Dropping
> leaks the callback, but the alternative is an unbounded loop.
>
> The irq_work is IRQ_WORK_INIT_HARD so the re-issue stays prompt on
> PREEMPT_RT, where a non-HARD irq_work runs in a kthread that can be
> delayed under load. A hidden CONFIG_RCU_DEFER gates the code and its
> IRQ_WORK dependency; without it call_rcu() enqueues directly as before.
> Under CONFIG_PROVE_RCU, warn if the direct path is reached from an NMI.
>
> Suggested-by: Paul E. McKenney <[email protected]>
> Signed-off-by: Puranjay Mohan <[email protected]>
I have replaced the previous version with this one for further review
and testing, thank you!
Thanx, Paul
> ---
> kernel/rcu/Kconfig | 6 ++
> kernel/rcu/rcu.h | 15 +++++
> kernel/rcu/tree.c | 139 +++++++++++++++++++++++++++++++++++++++++----
> kernel/rcu/tree.h | 6 ++
> 4 files changed, 156 insertions(+), 10 deletions(-)
>
> diff --git a/kernel/rcu/Kconfig b/kernel/rcu/Kconfig
> index f15da8038d0ba..1a5fb3156c062 100644
> --- a/kernel/rcu/Kconfig
> +++ b/kernel/rcu/Kconfig
> @@ -175,6 +175,12 @@ config RCU_STALL_COMMON
> config RCU_NEED_SEGCBLIST
> def_bool ( TREE_RCU || TREE_SRCU || TASKS_RCU_GENERIC )
>
> +# The deferral (and the IRQ_WORK it uses) is only needed where call_rcu() /
> +# call_srcu() can be invoked while a callback-list operation is in flight.
> +config RCU_DEFER
> + def_bool HAVE_NMI || KPROBES || FUNCTION_TRACER || TRACEPOINTS
> + select IRQ_WORK
> +
> config RCU_FANOUT
> int "Tree-based hierarchical RCU fanout value"
> range 2 64 if 64BIT
> diff --git a/kernel/rcu/rcu.h b/kernel/rcu/rcu.h
> index 39a9f6fa9a7b2..fd075d91b80cf 100644
> --- a/kernel/rcu/rcu.h
> +++ b/kernel/rcu/rcu.h
> @@ -572,6 +572,21 @@ static inline void tasks_cblist_init_generic(void) { }
> #define RCU_SCHEDULER_INIT 1
> #define RCU_SCHEDULER_RUNNING 2
>
> +/*
> + * Defer a call_rcu()/call_srcu() callback rather than enqueue it now? Defer
> + * whenever interrupts are disabled, since a callback-list operation may be
> in
> + * flight on this CPU. Not before the scheduler is up, though: that covers
> + * early boot, where irq_work is not yet usable and rcu_init() itself already
> + * calls call_rcu().
> + */
> +static inline bool should_rcu_defer(void)
> +{
> + if (!IS_ENABLED(CONFIG_RCU_DEFER))
> + return false;
> +
> + return irqs_disabled() && rcu_scheduler_active !=
> RCU_SCHEDULER_INACTIVE;
> +}
> +
> enum rcutorture_type {
> RCU_FLAVOR,
> RCU_TASKS_FLAVOR,
> diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c
> index 21b6ce1dffb63..6a83408974c2b 100644
> --- a/kernel/rcu/tree.c
> +++ b/kernel/rcu/tree.c
> @@ -24,6 +24,7 @@
> #include <linux/smp.h>
> #include <linux/rcupdate_wait.h>
> #include <linux/interrupt.h>
> +#include <linux/llist.h>
> #include <linux/sched.h>
> #include <linux/sched/debug.h>
> #include <linux/nmi.h>
> @@ -3148,21 +3149,25 @@ static void check_cb_ovld(struct rcu_data *rdp)
> raw_spin_unlock_rcu_node(rnp);
> }
>
> -static void
> -__call_rcu_common(struct rcu_head *head, rcu_callback_t func, bool lazy_in)
> +/*
> + * The callback list is only accessed with interrupts disabled, so a
> call_rcu()
> + * that arrives with interrupts off (see should_rcu_defer()) stages the
> callback
> + * on a per-CPU llist that an irq_work re-issues once interrupts are on.
> + */
> +static void rcu_defer_drain(struct irq_work *iw);
> +
> +/*
> + * Enqueue @head on this CPU's rcu_segcblist. Also called by
> rcu_defer_drain()
> + * to re-issue a deferred callback, so it must not re-check the deferral
> + * condition. Either caller may have interrupts already disabled.
> + */
> +static void rcu_do_enqueue(struct rcu_head *head, rcu_callback_t func, bool
> lazy_in)
> {
> static atomic_t doublefrees;
> unsigned long flags;
> bool lazy;
> struct rcu_data *rdp;
>
> - /* Misaligned rcu_head! */
> - WARN_ON_ONCE((unsigned long)head & (sizeof(void *) - 1));
> -
> - /* Avoid NULL dereference if callback is NULL. */
> - if (WARN_ON_ONCE(!func))
> - return;
> -
> if (debug_rcu_head_queue(head)) {
> /*
> * Probable double call_rcu(), so leak the callback.
> @@ -3206,6 +3211,104 @@ __call_rcu_common(struct rcu_head *head,
> rcu_callback_t func, bool lazy_in)
> local_irq_restore(flags);
> }
>
> +/*
> + * Re-issue deferred callbacks straight to the enqueue so they cannot defer
> + * again. ->defer_lock serializes the drainers: this CPU's irq_work,
> + * rcu_defer_flush() and rcutree_migrate_callbacks().
> + */
> +static void __rcu_defer_drain(struct rcu_data *rdp, bool guard)
> +{
> + struct llist_node *node, *next;
> + unsigned long flags;
> +
> + if (!IS_ENABLED(CONFIG_RCU_DEFER))
> + return;
> +
> + raw_spin_lock_irqsave(&rdp->defer_lock, flags);
> + if (guard)
> + WRITE_ONCE(rdp->defer_draining, true);
> + llist_for_each_safe(node, next, llist_del_all(&rdp->defer_head)) {
> + struct rcu_head *head = (struct rcu_head *)node;
> +
> + head->next = NULL;
> + rcu_do_enqueue(head, head->func, false);
> + }
> + if (guard)
> + WRITE_ONCE(rdp->defer_draining, false);
> + raw_spin_unlock_irqrestore(&rdp->defer_lock, flags);
> +}
> +
> +/*
> + * Only the irq_work drain can be re-fed by its own re-issue, so only it sets
> + * ->defer_draining. A direct drain re-issues onto this CPU, and anything
> + * staged during it is picked up by that CPU's own irq_work.
> + */
> +static void rcu_defer_drain(struct irq_work *iw)
> +{
> + __rcu_defer_drain(container_of(iw, struct rcu_data, defer_work), true);
> +}
> +
> +/* Stage @head for this CPU's irq_work when call_rcu() cannot enqueue now. */
> +static void call_rcu_defer(struct rcu_head *head, rcu_callback_t func)
> +{
> + struct rcu_data *rdp = this_cpu_ptr(&rcu_data);
> +
> + /*
> + * Instrumentation on the enqueue path can re-enter here from inside
> + * rcu_defer_drain(). Re-queuing would livelock the drain, so drop the
> + * callback; an NMI is one-shot and cannot loop, so let it through.
> + */
> + if (READ_ONCE(rdp->defer_draining) && !in_nmi()) {
> + WARN_ONCE(IS_ENABLED(CONFIG_PROVE_RCU),
> + "call_rcu() re-entered during callback drain; leaking
> callback\n");
> + return;
> + }
> + head->func = func;
> + if (llist_add((struct llist_node *)head, &rdp->defer_head))
> + irq_work_queue(&rdp->defer_work);
> +}
> +
> +/*
> + * Register pending deferred callbacks into the callback lists so a following
> + * rcu_barrier() waits for them. This runs before rcu_barrier() scans the
> + * lists.
> + */
> +static void rcu_defer_flush(void)
> +{
> + int cpu;
> +
> + if (!IS_ENABLED(CONFIG_RCU_DEFER))
> + return;
> +
> + for_each_possible_cpu(cpu) {
> + struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu);
> +
> + if (!llist_empty(&rdp->defer_head))
> + __rcu_defer_drain(rdp, false);
> + }
> +}
> +
> +static void
> +__call_rcu_common(struct rcu_head *head, rcu_callback_t func, bool lazy_in)
> +{
> + /* Misaligned rcu_head! */
> + WARN_ON_ONCE((unsigned long)head & (sizeof(void *) - 1));
> +
> + /* Avoid NULL dereference if callback is NULL. */
> + if (WARN_ON_ONCE(!func))
> + return;
> +
> + if (should_rcu_defer()) {
> + call_rcu_defer(head, func);
> + return;
> + }
> +
> + /* An NMI reaching here entered with irqs enabled, so the enqueue can
> race. */
> + WARN_ON_ONCE(IS_ENABLED(CONFIG_PROVE_RCU) && in_nmi());
> +
> + rcu_do_enqueue(head, func, lazy_in);
> +}
> +
> #ifdef CONFIG_RCU_LAZY
> static bool enable_rcu_lazy __read_mostly =
> !IS_ENABLED(CONFIG_RCU_LAZY_DEFAULT_OFF);
> module_param(enable_rcu_lazy, bool, 0444);
> @@ -3896,8 +3999,12 @@ void rcu_barrier(void)
> unsigned long flags;
> unsigned long gseq;
> struct rcu_data *rdp;
> - unsigned long s = rcu_seq_snap(&rcu_state.barrier_sequence);
> + unsigned long s;
>
> + /* Register any deferred callbacks before snapshotting the sequence. */
> + rcu_defer_flush();
> +
> + s = rcu_seq_snap(&rcu_state.barrier_sequence);
> rcu_barrier_trace(TPS("Begin"), -1, s);
>
> /* Take mutex to serialize concurrent rcu_barrier() requests. */
> @@ -4231,6 +4338,10 @@ rcu_boot_init_percpu_data(int cpu)
> rdp->rcu_onl_gp_state = RCU_GP_CLEANED;
> rdp->last_sched_clock = jiffies;
> rdp->cpu = cpu;
> + init_llist_head(&rdp->defer_head);
> + raw_spin_lock_init(&rdp->defer_lock);
> + /* Hard irq_work so the re-issue runs promptly. */
> + rdp->defer_work = IRQ_WORK_INIT_HARD(rcu_defer_drain);
> rcu_boot_init_nocb_percpu_data(rdp);
> }
>
> @@ -4528,6 +4639,14 @@ void rcutree_migrate_callbacks(int cpu)
> struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu);
> bool needwake;
>
> + /*
> + * Callbacks the outgoing CPU deferred late in the offline path (past
> the
> + * point its irq_work can run) sit on ->defer_head, which the ->cblist
> + * migration below does not cover. Drain them here, before the early
> + * returns; the re-issue lands on this CPU.
> + */
> + __rcu_defer_drain(rdp, false);
> +
> if (rcu_rdp_is_offloaded(rdp))
> return;
>
> diff --git a/kernel/rcu/tree.h b/kernel/rcu/tree.h
> index eedfa43059e80..b7cac7a13b4f2 100644
> --- a/kernel/rcu/tree.h
> +++ b/kernel/rcu/tree.h
> @@ -229,6 +229,12 @@ struct rcu_data {
> struct rcu_head barrier_head;
> int exp_watching_snap; /* Double-check need for IPI. */
>
> + /* Deferral of an NMI/reentrant call_rcu(); see __call_rcu_common(). */
> + struct llist_head defer_head;
> + struct irq_work defer_work;
> + raw_spinlock_t defer_lock;
> + bool defer_draining;
> +
> /* 5) Callback offloading. */
> #ifdef CONFIG_RCU_NOCB_CPU
> struct swait_queue_head nocb_cb_wq; /* For nocb kthreads to sleep on. */
> --
> 2.53.0-Meta
>
