[PATCH v2 1/1] mm/mmu_notifier: Add async OOM cleanup via call_srcu()

[PATCH v2 1/1] mm/mmu_notifier: Add async OOM cleanup via call_srcu()

[shaikh.kamal]

When an mm undergoes OOM kill, the OOM reaper unmaps memory while
holding the mmap_lock. MMU notifier subscribers (notably KVM) need
to be informed so they can tear down their secondary mappings. The
current synchronous unregister path can deadlock on PREEMPT_RT
because synchronize_srcu() is called from contexts that cannot
safely sleep.

This patch implements the asynchronous cleanup design proposed by
Paolo Bonzini in v1 review: a new optional after_oom_unregister
callback in struct mmu_notifier_ops, invoked after the SRCU grace
period via call_srcu() so that no readers can still reference the
subscription when cleanup runs.

The flow is:

  1. The OOM reaper calls mmu_notifier_oom_enter() from
     __oom_reap_task_mm().
  2. mmu_notifier_oom_enter() walks the subscription list and, for
     each subscriber that provides after_oom_unregister, detaches
     the subscription from the active list and schedules a
     call_srcu() callback.
  3. The deferred callback invokes after_oom_unregister once the
     grace period has elapsed and all in-flight readers have
     finished.
  4. Subsystems waiting to free structures referenced by the
     callback can call the new mmu_notifier_barrier() helper, which
     wraps srcu_barrier() to wait for all outstanding callbacks
     scheduled this way.

after_oom_unregister is mutually exclusive with alloc_notifier
because allocated notifiers can have additional outstanding
references that the OOM path cannot safely drop.

KVM is updated to provide after_oom_unregister, which clears
mn_active_invalidate_count, and to detect via hlist_unhashed() in
kvm_destroy_vm() when its subscription was already detached by the
OOM path; in that case it calls mmu_notifier_barrier() and drops
the mm reference rather than calling mmu_notifier_unregister().

Reported-by: syzbot+c3178b6b512446632bac@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=c3178b6b512446632bac
Suggested-by: Paolo Bonzini <pbonzini@redhat.com>
Link: https://lore.kernel.org/all/20260209161527.31978-1-shaikhkamal2012@gmail.com/

Signed-off-by: shaikh.kamal <shaikhkamal2012@gmail.com>
---
 include/linux/mmu_notifier.h |  10 +++
 mm/mmu_notifier.c            | 123 +++++++++++++++++++++++++++++++++++
 mm/oom_kill.c                |   3 +
 virt/kvm/kvm_main.c          |  27 +++++++-
 4 files changed, 162 insertions(+), 1 deletion(-)

diff --git a/include/linux/mmu_notifier.h b/include/linux/mmu_notifier.h
index 07a2bbaf86e9..0ccd590f55d3 100644
--- a/include/linux/mmu_notifier.h
+++ b/include/linux/mmu_notifier.h
@@ -88,6 +88,14 @@ struct mmu_notifier_ops {
 	void (*release)(struct mmu_notifier *subscription,
 			struct mm_struct *mm);

+	/*
+	 * Any mmu notifier that defines this is automatically unregistered
+	 * when its mm is the subject of an OOM kill.  after_oom_unregister()
+	 * is invoked after all other outstanding callbacks have terminated.
+	 */
+	void (*after_oom_unregister)(struct mmu_notifier *subscription,
+				     struct mm_struct *mm);
+
 	/*
 	 * clear_flush_young is called after the VM is
 	 * test-and-clearing the young/accessed bitflag in the
@@ -375,6 +383,8 @@ mmu_interval_check_retry(struct mmu_interval_notifier *interval_sub,

 extern void __mmu_notifier_subscriptions_destroy(struct mm_struct *mm);
 extern void __mmu_notifier_release(struct mm_struct *mm);
+void mmu_notifier_oom_enter(struct mm_struct *mm);
+extern void mmu_notifier_barrier(void);
 extern int __mmu_notifier_clear_flush_young(struct mm_struct *mm,
 					  unsigned long start,
 					  unsigned long end);
diff --git a/mm/mmu_notifier.c b/mm/mmu_notifier.c
index a6cdf3674bdc..b8fa58fe6b7d 100644
--- a/mm/mmu_notifier.c
+++ b/mm/mmu_notifier.c
@@ -49,6 +49,37 @@ struct mmu_notifier_subscriptions {
 	struct hlist_head deferred_list;
 };

+/*
+ * Callback structure for asynchronous OOM cleanup.
+ * Used with call_srcu() to defer after_oom_unregister callbacks
+ * until after SRCU grace period completes.
+ */
+struct mmu_notifier_oom_callback {
+	struct rcu_head rcu;
+	struct mmu_notifier *subscription;
+	struct mm_struct *mm;
+};
+
+/*
+ * Callback function invoked after SRCU grace period.
+ * Safely calls after_oom_unregister once all readers have finished.
+ */
+static void mmu_notifier_oom_callback_fn(struct rcu_head *rcu)
+{
+	struct mmu_notifier_oom_callback *cb =
+		container_of(rcu, struct mmu_notifier_oom_callback, rcu);
+
+	/* Safe - all SRCU readers have finished */
+	cb->subscription->ops->after_oom_unregister(cb->subscription, cb->mm);
+
+	/* Release mm reference taken when callback was scheduled */
+	WARN_ON_ONCE(atomic_read(&cb->mm->mm_count) <= 0);
+	mmdrop(cb->mm);
+
+	/* Free callback structure */
+	kfree(cb);
+}
+
 /*
  * This is a collision-retry read-side/write-side 'lock', a lot like a
  * seqcount, however this allows multiple write-sides to hold it at
@@ -359,6 +390,85 @@ void __mmu_notifier_release(struct mm_struct *mm)
 		mn_hlist_release(subscriptions, mm);
 }

+void mmu_notifier_oom_enter(struct mm_struct *mm)
+{
+	struct mmu_notifier_subscriptions *subscriptions =
+						mm->notifier_subscriptions;
+	struct mmu_notifier *subscription;
+	struct hlist_node *tmp;
+	HLIST_HEAD(oom_list);
+	int id;
+
+	if (!subscriptions)
+		return;
+
+	id = srcu_read_lock(&srcu);
+
+	/*
+	 * Prevent further calls to the MMU notifier, except for
+	 * release and after_oom_unregister.
+	 */
+	spin_lock(&subscriptions->lock);
+	hlist_for_each_entry_safe(subscription, tmp,
+				  &subscriptions->list, hlist) {
+		if (!subscription->ops->after_oom_unregister)
+			continue;
+
+		/*
+		 * after_oom_unregister and alloc_notifier are incompatible,
+		 * because there could be other references to allocated
+		 * notifiers.
+		 */
+		if (WARN_ON(subscription->ops->alloc_notifier))
+			continue;
+
+		hlist_del_init_rcu(&subscription->hlist);
+		hlist_add_head(&subscription->hlist, &oom_list);
+	}
+	spin_unlock(&subscriptions->lock);
+	hlist_for_each_entry(subscription, &oom_list, hlist)
+		if (subscription->ops->release)
+			subscription->ops->release(subscription, mm);
+
+	srcu_read_unlock(&srcu, id);
+
+	if (hlist_empty(&oom_list))
+		return;
+
+	hlist_for_each_entry_safe(subscription, tmp,
+				  &oom_list, hlist) {
+		struct mmu_notifier_oom_callback *cb;
+		/*
+		 * Remove from stack-based oom_list and reset hlist to unhashed state.
+		 * This sets subscription->hlist.pprev = NULL, so future callers of
+		 * mmu_notifier_unregister() (e.g. kvm_destroy_vm) will see
+		 * hlist_unhashed() == true and take the safe path, avoiding
+		 * use-after-free on the stack-allocated oom_list head.
+		 */
+		hlist_del_init(&subscription->hlist);
+
+		/*
+		 * GFP_ATOMIC failure is exceedingly rare. We cannot sleep
+		 * here (would reintroduce the deadlock this patch fixes)
+		 * and cannot call after_oom_unregister synchronously
+		 * without first waiting for SRCU readers. The subscriber
+		 * will not receive after_oom_unregister but cleanup will
+		 * eventually happen via the unregister path.
+		 */
+		cb = kmalloc(sizeof(*cb), GFP_ATOMIC);
+		if (!cb)
+			continue;
+
+		cb->subscription = subscription;
+		cb->mm = mm;
+		mmgrab(mm);
+
+		/* Schedule callback - returns immediately */
+		call_srcu(&srcu, &cb->rcu, mmu_notifier_oom_callback_fn);
+	}
+
+}
+
 /*
  * If no young bitflag is supported by the hardware, ->clear_flush_young can
  * unmap the address and return 1 or 0 depending if the mapping previously
@@ -1096,3 +1206,16 @@ void mmu_notifier_synchronize(void)
 	synchronize_srcu(&srcu);
 }
 EXPORT_SYMBOL_GPL(mmu_notifier_synchronize);
+
+/**
+ * mmu_notifier_barrier - Wait for all pending MMU notifier callbacks
+ *
+ * Waits for all call_srcu() callbacks scheduled by mmu_notifier_oom_enter()
+ * to complete. Used by subsystems during cleanup to prevent use-after-free
+ * when destroying structures accessed by the callbacks.
+ */
+void mmu_notifier_barrier(void)
+{
+	srcu_barrier(&srcu);
+}
+EXPORT_SYMBOL_GPL(mmu_notifier_barrier);
diff --git a/mm/oom_kill.c b/mm/oom_kill.c
index 5c6c95c169ee..029e041afc57 100644
--- a/mm/oom_kill.c
+++ b/mm/oom_kill.c
@@ -519,6 +519,9 @@ static bool __oom_reap_task_mm(struct mm_struct *mm)
 	bool ret = true;
 	MA_STATE(mas, &mm->mm_mt, ULONG_MAX, ULONG_MAX);

+	/* Notify MMU notifiers about the OOM event */
+	mmu_notifier_oom_enter(mm);
+
 	/*
 	 * Tell all users of get_user/copy_from_user etc... that the content
 	 * is no longer stable. No barriers really needed because unmapping
diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
index 1bc1da66b4b0..a2df83d3b413 100644
--- a/virt/kvm/kvm_main.c
+++ b/virt/kvm/kvm_main.c
@@ -885,6 +885,24 @@ static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 	srcu_read_unlock(&kvm->srcu, idx);
 }

+static void kvm_mmu_notifier_after_oom_unregister(struct mmu_notifier *mn,
+					struct mm_struct *mm)
+{
+	struct kvm *kvm;
+
+	kvm = mmu_notifier_to_kvm(mn);
+
+	/*
+	 * At this point the unregister has completed and all other callbacks
+	 * have terminated. Clean up any unbalanced invalidation counts.
+	 */
+	WARN_ON(rcuwait_active(&kvm->mn_memslots_update_rcuwait));
+	if (kvm->mn_active_invalidate_count)
+		kvm->mn_active_invalidate_count = 0;
+	else
+		WARN_ON(kvm->mmu_invalidate_in_progress);
+}
+
 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 	.invalidate_range_start	= kvm_mmu_notifier_invalidate_range_start,
 	.invalidate_range_end	= kvm_mmu_notifier_invalidate_range_end,
@@ -892,6 +910,7 @@ static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 	.clear_young		= kvm_mmu_notifier_clear_young,
 	.test_young		= kvm_mmu_notifier_test_young,
 	.release		= kvm_mmu_notifier_release,
+	.after_oom_unregister	= kvm_mmu_notifier_after_oom_unregister,
 };

 static int kvm_init_mmu_notifier(struct kvm *kvm)
@@ -1280,7 +1299,13 @@ static void kvm_destroy_vm(struct kvm *kvm)
 		kvm->buses[i] = NULL;
 	}
 	kvm_coalesced_mmio_free(kvm);
-	mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
+	if (hlist_unhashed(&kvm->mmu_notifier.hlist)) {
+		/* Subscription removed by OOM. Wait for async callback. */
+		mmu_notifier_barrier();
+		mmdrop(kvm->mm);
+	} else {
+		mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
+	}
 	/*
 	 * At this point, pending calls to invalidate_range_start()
 	 * have completed but no more MMU notifiers will run, so
--
2.43.0

Re: [PATCH] KVM: x86/xen: Fix sleeping lock in hard IRQ context on PREEMPT_RT

[Sean Christopherson]

On Mon, Mar 30, 2026, David Woodhouse wrote:
> On Mon, 2026-03-30 at 10:18 -0400, Steven Rostedt wrote:
> > 
> > > +static void xen_timer_inject_irqwork(struct irq_work *work)
> > > +{
> > > +     struct kvm_vcpu_xen *xen = container_of(work, struct kvm_vcpu_xen,
> > > +                                             timer_inject_irqwork);
> > > +     struct kvm_vcpu *vcpu = container_of(xen, struct kvm_vcpu, arch.xen);
> > > +     struct kvm_xen_evtchn e;
> > > +     int rc;
> > > +
> > > +     e.vcpu_id = vcpu->vcpu_id;
> > > +     e.vcpu_idx = vcpu->vcpu_idx;
> > > +     e.port = vcpu->arch.xen.timer_virq;
> > > +     e.priority = KVM_IRQ_ROUTING_XEN_EVTCHN_PRIO_2LEVEL;
> > > +
> > > +     rc = kvm_xen_set_evtchn_fast(&e, vcpu->kvm);
> > > +     if (rc != -EWOULDBLOCK)
> > > +             vcpu->arch.xen.timer_expires = 0;
> > > +}
> > 
> > Why duplicate this code and not simply make a static inline helper
> > function that is used in both places?
> 
> It's already duplicating the functionality; the original
> xen_timer_callback() will already fall back to injecting the IRQ in
> process context when it needs to (by setting vcpu-
> >arch.xen.timer_pending and then setting KVM_REQ_UNBLOCK).
> 
> All you had to do was make kvm_xen_set_evtchn_fast() return 
> -EWOULDBLOCK in the in_hardirq() case in order to use the existing
> fallback, surely? 
> 
> Better still, can't kvm_xen_set_evtchn_fast() just use read_trylock()
> instead?

Re-reading through the thread where you proposed using trylock, and through
commit bbe17c625d68 ("KVM: x86/xen: Fix potential deadlock in kvm_xen_update_runstate_guest()"),
I think I agree with using trylock for "fast" paths.

Though I would prefer to not make it unconditional for the "fast" helper instead
of conditional based on in_interrupt().  And before we start doing surgery to
"fix" a setup no one uses, and also before we use gpcs more broadly, I think we
should try to up-level the gpc APIs to reduce the amount of duplicate, boilerplate
code.  kvm_xen_update_runstate_guest() and maybe kvm_xen_set_evtchn() will likely
need to open code some amount of logic, but 

Side topic, looks like kvm_xen_shared_info_init() is buggy in that it fails to
mark the slot as dirty.

E.g. sans the API implementations, I think we can and should end up with code
like this:

---
 arch/x86/kvm/x86.c |  14 ++---
 arch/x86/kvm/xen.c | 127 ++++++++++++---------------------------------
 2 files changed, 37 insertions(+), 104 deletions(-)

diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index 0b5d48e75b65..65bad25fd9d4 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -3274,15 +3274,8 @@ static void kvm_setup_guest_pvclock(struct pvclock_vcpu_time_info *ref_hv_clock,
 
 	memcpy(&hv_clock, ref_hv_clock, sizeof(hv_clock));
 
-	read_lock_irqsave(&gpc->lock, flags);
-	while (!kvm_gpc_check(gpc, offset + sizeof(*guest_hv_clock))) {
-		read_unlock_irqrestore(&gpc->lock, flags);
-
-		if (kvm_gpc_refresh(gpc, offset + sizeof(*guest_hv_clock)))
-			return;
-
-		read_lock_irqsave(&gpc->lock, flags);
-	}
+	if (kvm_gpc_acquire(gpc))
+		return;
 
 	guest_hv_clock = (void *)(gpc->khva + offset);
 
@@ -3305,8 +3298,7 @@ static void kvm_setup_guest_pvclock(struct pvclock_vcpu_time_info *ref_hv_clock,
 
 	guest_hv_clock->version = ++hv_clock.version;
 
-	kvm_gpc_mark_dirty_in_slot(gpc);
-	read_unlock_irqrestore(&gpc->lock, flags);
+	kvm_gpc_release_dirty(gpc);
 
 	trace_kvm_pvclock_update(vcpu->vcpu_id, &hv_clock);
 }
diff --git a/arch/x86/kvm/xen.c b/arch/x86/kvm/xen.c
index 91fd3673c09a..a97fd88ee99c 100644
--- a/arch/x86/kvm/xen.c
+++ b/arch/x86/kvm/xen.c
@@ -42,19 +42,12 @@ static int kvm_xen_shared_info_init(struct kvm *kvm)
 	u32 *wc_sec_hi;
 	u32 wc_version;
 	u64 wall_nsec;
-	int ret = 0;
 	int idx = srcu_read_lock(&kvm->srcu);
+	int ret;
 
-	read_lock_irq(&gpc->lock);
-	while (!kvm_gpc_check(gpc, PAGE_SIZE)) {
-		read_unlock_irq(&gpc->lock);
-
-		ret = kvm_gpc_refresh(gpc, PAGE_SIZE);
-		if (ret)
-			goto out;
-
-		read_lock_irq(&gpc->lock);
-	}
+	ret = kvm_gpc_acquire(gpc);
+	if (ret)
+		goto out;
 
 	/*
 	 * This code mirrors kvm_write_wall_clock() except that it writes
@@ -96,7 +89,7 @@ static int kvm_xen_shared_info_init(struct kvm *kvm)
 	smp_wmb();
 
 	wc->version = wc_version + 1;
-	read_unlock_irq(&gpc->lock);
+	kvm_gpc_release_dirty(gpc);
 
 	kvm_make_all_cpus_request(kvm, KVM_REQ_MASTERCLOCK_UPDATE);
 
@@ -155,22 +148,14 @@ static int xen_get_guest_pvclock(struct kvm_vcpu *vcpu,
 				 struct gfn_to_pfn_cache *gpc,
 				 unsigned int offset)
 {
-	unsigned long flags;
 	int r;
 
-	read_lock_irqsave(&gpc->lock, flags);
-	while (!kvm_gpc_check(gpc, offset + sizeof(*hv_clock))) {
-		read_unlock_irqrestore(&gpc->lock, flags);
-
-		r = kvm_gpc_refresh(gpc, offset + sizeof(*hv_clock));
-		if (r)
-			return r;
-
-		read_lock_irqsave(&gpc->lock, flags);
-	}
+	r = kvm_gpc_acquire(gpc);
+	if (r)
+		return r;
 
 	memcpy(hv_clock, gpc->khva + offset, sizeof(*hv_clock));
-	read_unlock_irqrestore(&gpc->lock, flags);
+	kvm_gpc_release_clean(gpc);
 
 	/*
 	 * Sanity check TSC shift+multiplier to verify the guest's view of time
@@ -420,27 +405,8 @@ static void kvm_xen_update_runstate_guest(struct kvm_vcpu *v, bool atomic)
 	 * Attempt to obtain the GPC lock on *both* (if there are two)
 	 * gfn_to_pfn caches that cover the region.
 	 */
-	if (atomic) {
-		local_irq_save(flags);
-		if (!read_trylock(&gpc1->lock)) {
-			local_irq_restore(flags);
-			return;
-		}
-	} else {
-		read_lock_irqsave(&gpc1->lock, flags);
-	}
-	while (!kvm_gpc_check(gpc1, user_len1)) {
-		read_unlock_irqrestore(&gpc1->lock, flags);
-
-		/* When invoked from kvm_sched_out() we cannot sleep */
-		if (atomic)
-			return;
-
-		if (kvm_gpc_refresh(gpc1, user_len1))
-			return;
-
-		read_lock_irqsave(&gpc1->lock, flags);
-	}
+	if (__kvm_gpc_acquire(gpc, atomic))
+		return;
 
 	if (likely(!user_len2)) {
 		/*
@@ -465,6 +431,7 @@ static void kvm_xen_update_runstate_guest(struct kvm_vcpu *v, bool atomic)
 		 * gpc1 lock to make lockdep shut up about it.
 		 */
 		lock_set_subclass(&gpc1->lock.dep_map, 1, _THIS_IP_);
+
 		if (atomic) {
 			if (!read_trylock(&gpc2->lock)) {
 				read_unlock_irqrestore(&gpc1->lock, flags);
@@ -575,13 +542,10 @@ static void kvm_xen_update_runstate_guest(struct kvm_vcpu *v, bool atomic)
 		smp_wmb();
 	}
 
-	if (user_len2) {
-		kvm_gpc_mark_dirty_in_slot(gpc2);
-		read_unlock(&gpc2->lock);
-	}
+	if (user_len2)
+		kvm_gpc_release_dirty(gpc2);
 
-	kvm_gpc_mark_dirty_in_slot(gpc1);
-	read_unlock_irqrestore(&gpc1->lock, flags);
+	kvm_gpc_release_dirty(gpc1);
 }
 
 void kvm_xen_update_runstate(struct kvm_vcpu *v, int state)
@@ -645,20 +609,8 @@ void kvm_xen_inject_pending_events(struct kvm_vcpu *v)
 	if (!evtchn_pending_sel)
 		return;
 
-	/*
-	 * Yes, this is an open-coded loop. But that's just what put_user()
-	 * does anyway. Page it in and retry the instruction. We're just a
-	 * little more honest about it.
-	 */
-	read_lock_irqsave(&gpc->lock, flags);
-	while (!kvm_gpc_check(gpc, sizeof(struct vcpu_info))) {
-		read_unlock_irqrestore(&gpc->lock, flags);
-
-		if (kvm_gpc_refresh(gpc, sizeof(struct vcpu_info)))
-			return;
-
-		read_lock_irqsave(&gpc->lock, flags);
-	}
+	if (kvm_gpc_acquire(gpc))
+		return;
 
 	/* Now gpc->khva is a valid kernel address for the vcpu_info */
 	if (IS_ENABLED(CONFIG_64BIT) && v->kvm->arch.xen.long_mode) {
@@ -686,8 +638,7 @@ void kvm_xen_inject_pending_events(struct kvm_vcpu *v)
 		WRITE_ONCE(vi->evtchn_upcall_pending, 1);
 	}
 
-	kvm_gpc_mark_dirty_in_slot(gpc);
-	read_unlock_irqrestore(&gpc->lock, flags);
+	kvm_gpc_release_dirty(gpc);
 
 	/* For the per-vCPU lapic vector, deliver it as MSI. */
 	if (v->arch.xen.upcall_vector)
@@ -697,8 +648,8 @@ void kvm_xen_inject_pending_events(struct kvm_vcpu *v)
 int __kvm_xen_has_interrupt(struct kvm_vcpu *v)
 {
 	struct gfn_to_pfn_cache *gpc = &v->arch.xen.vcpu_info_cache;
-	unsigned long flags;
 	u8 rc = 0;
+	int r;
 
 	/*
 	 * If the global upcall vector (HVMIRQ_callback_vector) is set and
@@ -713,33 +664,23 @@ int __kvm_xen_has_interrupt(struct kvm_vcpu *v)
 	BUILD_BUG_ON(sizeof(rc) !=
 		     sizeof_field(struct compat_vcpu_info, evtchn_upcall_pending));
 
-	read_lock_irqsave(&gpc->lock, flags);
-	while (!kvm_gpc_check(gpc, sizeof(struct vcpu_info))) {
-		read_unlock_irqrestore(&gpc->lock, flags);
-
-		/*
-		 * This function gets called from kvm_vcpu_block() after setting the
-		 * task to TASK_INTERRUPTIBLE, to see if it needs to wake immediately
-		 * from a HLT. So we really mustn't sleep. If the page ended up absent
-		 * at that point, just return 1 in order to trigger an immediate wake,
-		 * and we'll end up getting called again from a context where we *can*
-		 * fault in the page and wait for it.
-		 */
-		if (in_atomic() || !task_is_running(current))
-			return 1;
-
-		if (kvm_gpc_refresh(gpc, sizeof(struct vcpu_info))) {
-			/*
-			 * If this failed, userspace has screwed up the
-			 * vcpu_info mapping. No interrupts for you.
-			 */
-			return 0;
-		}
-		read_lock_irqsave(&gpc->lock, flags);
-	}
+	/*
+	 * This function gets called from kvm_vcpu_block() after setting the
+	 * task to TASK_INTERRUPTIBLE, to see if it needs to wake immediately
+	 * from a HLT. So we really mustn't sleep. If the page ended up absent
+	 * at that point, just return 1 in order to trigger an immediate wake,
+	 * and we'll end up getting called again from a context where we *can*
+	 * fault in the page and wait for it.
+	 *
+	 * If acquiring the cache fails completely, then userspace has screwed
+	 * up the vcpu_info mapping. No interrupts for you.
+	 */
+	r = __kvm_gpc_acquire(gpc, in_atomic() || !task_is_running(current));
+	if (r)
+		return r == -EWOULDBLOCK ? 1 : 0;
 
 	rc = ((struct vcpu_info *)gpc->khva)->evtchn_upcall_pending;
-	read_unlock_irqrestore(&gpc->lock, flags);
+	kvm_gpc_release_clean(gpc);
 	return rc;
 }
 

base-commit: 3d6cdcc8883b5726513d245eef0e91cabfc397f7
-- 

[*] https://lore.kernel.org/all/76c61e1cb86e04df892d74c10976597700fe4cb5.camel@infradead.org

[PATCH v2 1/1] mm/mmu_notifier: Add async OOM cleanup via call_srcu()

[shaikh.kamal]

When an mm undergoes OOM kill, the OOM reaper unmaps memory while
holding the mmap_lock. MMU notifier subscribers (notably KVM) need
to be informed so they can tear down their secondary mappings. The
current synchronous unregister path can deadlock on PREEMPT_RT
because synchronize_srcu() is called from contexts that cannot
safely sleep.

This patch implements the asynchronous cleanup design proposed by
Paolo Bonzini in v1 review: a new optional after_oom_unregister
callback in struct mmu_notifier_ops, invoked after the SRCU grace
period via call_srcu() so that no readers can still reference the
subscription when cleanup runs.

The flow is:

  1. The OOM reaper calls mmu_notifier_oom_enter() from
     __oom_reap_task_mm().
  2. mmu_notifier_oom_enter() walks the subscription list and, for
     each subscriber that provides after_oom_unregister, detaches
     the subscription from the active list and schedules a
     call_srcu() callback.
  3. The deferred callback invokes after_oom_unregister once the
     grace period has elapsed and all in-flight readers have
     finished.
  4. Subsystems waiting to free structures referenced by the
     callback can call the new mmu_notifier_barrier() helper, which
     wraps srcu_barrier() to wait for all outstanding callbacks
     scheduled this way.

after_oom_unregister is mutually exclusive with alloc_notifier
because allocated notifiers can have additional outstanding
references that the OOM path cannot safely drop.

KVM is updated to provide after_oom_unregister, which clears
mn_active_invalidate_count, and to detect via hlist_unhashed() in
kvm_destroy_vm() when its subscription was already detached by the
OOM path; in that case it calls mmu_notifier_barrier() and drops
the mm reference rather than calling mmu_notifier_unregister().

Reported-by: syzbot+c3178b6b512446632bac@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=c3178b6b512446632bac
Suggested-by: Paolo Bonzini <pbonzini@redhat.com>
Link: https://lore.kernel.org/all/20260209161527.31978-1-shaikhkamal2012@gmail.com/

Signed-off-by: shaikh.kamal <shaikhkamal2012@gmail.com>
---
 include/linux/mmu_notifier.h |  10 +++
 mm/mmu_notifier.c            | 123 +++++++++++++++++++++++++++++++++++
 mm/oom_kill.c                |   3 +
 virt/kvm/kvm_main.c          |  27 +++++++-
 4 files changed, 162 insertions(+), 1 deletion(-)

diff --git a/include/linux/mmu_notifier.h b/include/linux/mmu_notifier.h
index 07a2bbaf86e9..0ccd590f55d3 100644
--- a/include/linux/mmu_notifier.h
+++ b/include/linux/mmu_notifier.h
@@ -88,6 +88,14 @@ struct mmu_notifier_ops {
 	void (*release)(struct mmu_notifier *subscription,
 			struct mm_struct *mm);

+	/*
+	 * Any mmu notifier that defines this is automatically unregistered
+	 * when its mm is the subject of an OOM kill.  after_oom_unregister()
+	 * is invoked after all other outstanding callbacks have terminated.
+	 */
+	void (*after_oom_unregister)(struct mmu_notifier *subscription,
+				     struct mm_struct *mm);
+
 	/*
 	 * clear_flush_young is called after the VM is
 	 * test-and-clearing the young/accessed bitflag in the
@@ -375,6 +383,8 @@ mmu_interval_check_retry(struct mmu_interval_notifier *interval_sub,

 extern void __mmu_notifier_subscriptions_destroy(struct mm_struct *mm);
 extern void __mmu_notifier_release(struct mm_struct *mm);
+void mmu_notifier_oom_enter(struct mm_struct *mm);
+extern void mmu_notifier_barrier(void);
 extern int __mmu_notifier_clear_flush_young(struct mm_struct *mm,
 					  unsigned long start,
 					  unsigned long end);
diff --git a/mm/mmu_notifier.c b/mm/mmu_notifier.c
index a6cdf3674bdc..b8fa58fe6b7d 100644
--- a/mm/mmu_notifier.c
+++ b/mm/mmu_notifier.c
@@ -49,6 +49,37 @@ struct mmu_notifier_subscriptions {
 	struct hlist_head deferred_list;
 };

+/*
+ * Callback structure for asynchronous OOM cleanup.
+ * Used with call_srcu() to defer after_oom_unregister callbacks
+ * until after SRCU grace period completes.
+ */
+struct mmu_notifier_oom_callback {
+	struct rcu_head rcu;
+	struct mmu_notifier *subscription;
+	struct mm_struct *mm;
+};
+
+/*
+ * Callback function invoked after SRCU grace period.
+ * Safely calls after_oom_unregister once all readers have finished.
+ */
+static void mmu_notifier_oom_callback_fn(struct rcu_head *rcu)
+{
+	struct mmu_notifier_oom_callback *cb =
+		container_of(rcu, struct mmu_notifier_oom_callback, rcu);
+
+	/* Safe - all SRCU readers have finished */
+	cb->subscription->ops->after_oom_unregister(cb->subscription, cb->mm);
+
+	/* Release mm reference taken when callback was scheduled */
+	WARN_ON_ONCE(atomic_read(&cb->mm->mm_count) <= 0);
+	mmdrop(cb->mm);
+
+	/* Free callback structure */
+	kfree(cb);
+}
+
 /*
  * This is a collision-retry read-side/write-side 'lock', a lot like a
  * seqcount, however this allows multiple write-sides to hold it at
@@ -359,6 +390,85 @@ void __mmu_notifier_release(struct mm_struct *mm)
 		mn_hlist_release(subscriptions, mm);
 }

+void mmu_notifier_oom_enter(struct mm_struct *mm)
+{
+	struct mmu_notifier_subscriptions *subscriptions =
+						mm->notifier_subscriptions;
+	struct mmu_notifier *subscription;
+	struct hlist_node *tmp;
+	HLIST_HEAD(oom_list);
+	int id;
+
+	if (!subscriptions)
+		return;
+
+	id = srcu_read_lock(&srcu);
+
+	/*
+	 * Prevent further calls to the MMU notifier, except for
+	 * release and after_oom_unregister.
+	 */
+	spin_lock(&subscriptions->lock);
+	hlist_for_each_entry_safe(subscription, tmp,
+				  &subscriptions->list, hlist) {
+		if (!subscription->ops->after_oom_unregister)
+			continue;
+
+		/*
+		 * after_oom_unregister and alloc_notifier are incompatible,
+		 * because there could be other references to allocated
+		 * notifiers.
+		 */
+		if (WARN_ON(subscription->ops->alloc_notifier))
+			continue;
+
+		hlist_del_init_rcu(&subscription->hlist);
+		hlist_add_head(&subscription->hlist, &oom_list);
+	}
+	spin_unlock(&subscriptions->lock);
+	hlist_for_each_entry(subscription, &oom_list, hlist)
+		if (subscription->ops->release)
+			subscription->ops->release(subscription, mm);
+
+	srcu_read_unlock(&srcu, id);
+
+	if (hlist_empty(&oom_list))
+		return;
+
+	hlist_for_each_entry_safe(subscription, tmp,
+				  &oom_list, hlist) {
+		struct mmu_notifier_oom_callback *cb;
+		/*
+		 * Remove from stack-based oom_list and reset hlist to unhashed state.
+		 * This sets subscription->hlist.pprev = NULL, so future callers of
+		 * mmu_notifier_unregister() (e.g. kvm_destroy_vm) will see
+		 * hlist_unhashed() == true and take the safe path, avoiding
+		 * use-after-free on the stack-allocated oom_list head.
+		 */
+		hlist_del_init(&subscription->hlist);
+
+		/*
+		 * GFP_ATOMIC failure is exceedingly rare. We cannot sleep
+		 * here (would reintroduce the deadlock this patch fixes)
+		 * and cannot call after_oom_unregister synchronously
+		 * without first waiting for SRCU readers. The subscriber
+		 * will not receive after_oom_unregister but cleanup will
+		 * eventually happen via the unregister path.
+		 */
+		cb = kmalloc(sizeof(*cb), GFP_ATOMIC);
+		if (!cb)
+			continue;
+
+		cb->subscription = subscription;
+		cb->mm = mm;
+		mmgrab(mm);
+
+		/* Schedule callback - returns immediately */
+		call_srcu(&srcu, &cb->rcu, mmu_notifier_oom_callback_fn);
+	}
+
+}
+
 /*
  * If no young bitflag is supported by the hardware, ->clear_flush_young can
  * unmap the address and return 1 or 0 depending if the mapping previously
@@ -1096,3 +1206,16 @@ void mmu_notifier_synchronize(void)
 	synchronize_srcu(&srcu);
 }
 EXPORT_SYMBOL_GPL(mmu_notifier_synchronize);
+
+/**
+ * mmu_notifier_barrier - Wait for all pending MMU notifier callbacks
+ *
+ * Waits for all call_srcu() callbacks scheduled by mmu_notifier_oom_enter()
+ * to complete. Used by subsystems during cleanup to prevent use-after-free
+ * when destroying structures accessed by the callbacks.
+ */
+void mmu_notifier_barrier(void)
+{
+	srcu_barrier(&srcu);
+}
+EXPORT_SYMBOL_GPL(mmu_notifier_barrier);
diff --git a/mm/oom_kill.c b/mm/oom_kill.c
index 5c6c95c169ee..029e041afc57 100644
--- a/mm/oom_kill.c
+++ b/mm/oom_kill.c
@@ -519,6 +519,9 @@ static bool __oom_reap_task_mm(struct mm_struct *mm)
 	bool ret = true;
 	MA_STATE(mas, &mm->mm_mt, ULONG_MAX, ULONG_MAX);

+	/* Notify MMU notifiers about the OOM event */
+	mmu_notifier_oom_enter(mm);
+
 	/*
 	 * Tell all users of get_user/copy_from_user etc... that the content
 	 * is no longer stable. No barriers really needed because unmapping
diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
index 1bc1da66b4b0..a2df83d3b413 100644
--- a/virt/kvm/kvm_main.c
+++ b/virt/kvm/kvm_main.c
@@ -885,6 +885,24 @@ static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 	srcu_read_unlock(&kvm->srcu, idx);
 }

+static void kvm_mmu_notifier_after_oom_unregister(struct mmu_notifier *mn,
+					struct mm_struct *mm)
+{
+	struct kvm *kvm;
+
+	kvm = mmu_notifier_to_kvm(mn);
+
+	/*
+	 * At this point the unregister has completed and all other callbacks
+	 * have terminated. Clean up any unbalanced invalidation counts.
+	 */
+	WARN_ON(rcuwait_active(&kvm->mn_memslots_update_rcuwait));
+	if (kvm->mn_active_invalidate_count)
+		kvm->mn_active_invalidate_count = 0;
+	else
+		WARN_ON(kvm->mmu_invalidate_in_progress);
+}
+
 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 	.invalidate_range_start	= kvm_mmu_notifier_invalidate_range_start,
 	.invalidate_range_end	= kvm_mmu_notifier_invalidate_range_end,
@@ -892,6 +910,7 @@ static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 	.clear_young		= kvm_mmu_notifier_clear_young,
 	.test_young		= kvm_mmu_notifier_test_young,
 	.release		= kvm_mmu_notifier_release,
+	.after_oom_unregister	= kvm_mmu_notifier_after_oom_unregister,
 };

 static int kvm_init_mmu_notifier(struct kvm *kvm)
@@ -1280,7 +1299,13 @@ static void kvm_destroy_vm(struct kvm *kvm)
 		kvm->buses[i] = NULL;
 	}
 	kvm_coalesced_mmio_free(kvm);
-	mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
+	if (hlist_unhashed(&kvm->mmu_notifier.hlist)) {
+		/* Subscription removed by OOM. Wait for async callback. */
+		mmu_notifier_barrier();
+		mmdrop(kvm->mm);
+	} else {
+		mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
+	}
 	/*
 	 * At this point, pending calls to invalidate_range_start()
 	 * have completed but no more MMU notifiers will run, so
--
2.43.0

Re: [PATCH v2 1/1] mm/mmu_notifier: Add async OOM cleanup via call_srcu()

[kernel test robot]

Hi shaikh.kamal,

kernel test robot noticed the following build errors:

[auto build test ERROR on v7.0]
[cannot apply to akpm-mm/mm-everything kvm/queue kvm/next kvm/linux-next v7.1-rc1 linus/master next-20260430]
[If your patch is applied to the wrong git tree, kindly drop us a note.
And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch#_base_tree_information]

url:    https://github.com/intel-lab-lkp/linux/commits/shaikh-kamal/mm-mmu_notifier-Add-async-OOM-cleanup-via-call_srcu/20260430-202943
base:   v7.0
patch link:    https://lore.kernel.org/r/20260429222548.25475-1-shaikhkamal2012%40gmail.com
patch subject: [PATCH v2 1/1] mm/mmu_notifier: Add async OOM cleanup via call_srcu()
config: hexagon-allnoconfig (https://download.01.org/0day-ci/archive/20260503/202605031115.qmkkOLQc-lkp@intel.com/config)
compiler: clang version 23.0.0git (https://github.com/llvm/llvm-project 5bac06718f502014fade905512f1d26d578a18f3)
reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20260503/202605031115.qmkkOLQc-lkp@intel.com/reproduce)

If you fix the issue in a separate patch/commit (i.e. not just a new version of
the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <lkp@intel.com>
| Closes: https://lore.kernel.org/oe-kbuild-all/202605031115.qmkkOLQc-lkp@intel.com/

All errors (new ones prefixed by >>):

>> mm/oom_kill.c:523:2: error: call to undeclared function 'mmu_notifier_oom_enter'; ISO C99 and later do not support implicit function declarations [-Wimplicit-function-declaration]
     523 |         mmu_notifier_oom_enter(mm);
         |         ^
   mm/oom_kill.c:523:2: note: did you mean 'mmu_notifier_release'?
   include/linux/mmu_notifier.h:610:20: note: 'mmu_notifier_release' declared here
     610 | static inline void mmu_notifier_release(struct mm_struct *mm)
         |                    ^
   mm/oom_kill.c:511:28: warning: variable 'oom_reaper_th' set but not used [-Wunused-but-set-global]
     511 | static struct task_struct *oom_reaper_th;
         |                            ^
   1 warning and 1 error generated.


vim +/mmu_notifier_oom_enter +523 mm/oom_kill.c

   515	
   516	static bool __oom_reap_task_mm(struct mm_struct *mm)
   517	{
   518		struct vm_area_struct *vma;
   519		bool ret = true;
   520		MA_STATE(mas, &mm->mm_mt, ULONG_MAX, ULONG_MAX);
   521	
   522		/* Notify MMU notifiers about the OOM event */
 > 523		mmu_notifier_oom_enter(mm);
   524	
   525		/*
   526		 * Tell all users of get_user/copy_from_user etc... that the content
   527		 * is no longer stable. No barriers really needed because unmapping
   528		 * should imply barriers already and the reader would hit a page fault
   529		 * if it stumbled over a reaped memory.
   530		 */
   531		mm_flags_set(MMF_UNSTABLE, mm);
   532	
   533		/*
   534		 * It might start racing with the dying task and compete for shared
   535		 * resources - e.g. page table lock contention has been observed.
   536		 * Reduce those races by reaping the oom victim from the other end
   537		 * of the address space.
   538		 */
   539		mas_for_each_rev(&mas, vma, 0) {
   540			if (vma->vm_flags & (VM_HUGETLB|VM_PFNMAP))
   541				continue;
   542	
   543			/*
   544			 * Only anonymous pages have a good chance to be dropped
   545			 * without additional steps which we cannot afford as we
   546			 * are OOM already.
   547			 *
   548			 * We do not even care about fs backed pages because all
   549			 * which are reclaimable have already been reclaimed and
   550			 * we do not want to block exit_mmap by keeping mm ref
   551			 * count elevated without a good reason.
   552			 */
   553			if (vma_is_anonymous(vma) || !(vma->vm_flags & VM_SHARED)) {
   554				struct mmu_notifier_range range;
   555				struct mmu_gather tlb;
   556	
   557				mmu_notifier_range_init(&range, MMU_NOTIFY_UNMAP, 0,
   558							mm, vma->vm_start,
   559							vma->vm_end);
   560				tlb_gather_mmu(&tlb, mm);
   561				if (mmu_notifier_invalidate_range_start_nonblock(&range)) {
   562					tlb_finish_mmu(&tlb);
   563					ret = false;
   564					continue;
   565				}
   566				unmap_page_range(&tlb, vma, range.start, range.end, NULL);
   567				mmu_notifier_invalidate_range_end(&range);
   568				tlb_finish_mmu(&tlb);
   569			}
   570		}
   571	
   572		return ret;
   573	}
   574	

-- 
0-DAY CI Kernel Test Service
https://github.com/intel/lkp-tests/wiki

Re: [PATCH v2 1/1] mm/mmu_notifier: Add async OOM cleanup via call_srcu()

[kernel test robot]

Hi shaikh.kamal,

kernel test robot noticed the following build errors:

[auto build test ERROR on v7.0]
[cannot apply to akpm-mm/mm-everything kvm/queue kvm/next kvm/linux-next v7.1-rc1 linus/master next-20260430]
[If your patch is applied to the wrong git tree, kindly drop us a note.
And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch#_base_tree_information]

url:    https://github.com/intel-lab-lkp/linux/commits/shaikh-kamal/mm-mmu_notifier-Add-async-OOM-cleanup-via-call_srcu/20260430-202943
base:   v7.0
patch link:    https://lore.kernel.org/r/20260429222548.25475-1-shaikhkamal2012%40gmail.com
patch subject: [PATCH v2 1/1] mm/mmu_notifier: Add async OOM cleanup via call_srcu()
config: arc-allnoconfig (https://download.01.org/0day-ci/archive/20260503/202605031109.uxckW5L3-lkp@intel.com/config)
compiler: arc-linux-gcc (GCC) 15.2.0
reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20260503/202605031109.uxckW5L3-lkp@intel.com/reproduce)

If you fix the issue in a separate patch/commit (i.e. not just a new version of
the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <lkp@intel.com>
| Closes: https://lore.kernel.org/oe-kbuild-all/202605031109.uxckW5L3-lkp@intel.com/

All errors (new ones prefixed by >>):

   mm/oom_kill.c: In function '__oom_reap_task_mm':
>> mm/oom_kill.c:523:9: error: implicit declaration of function 'mmu_notifier_oom_enter'; did you mean 'mmu_notifier_release'? [-Wimplicit-function-declaration]
     523 |         mmu_notifier_oom_enter(mm);
         |         ^~~~~~~~~~~~~~~~~~~~~~
         |         mmu_notifier_release


vim +523 mm/oom_kill.c

   515	
   516	static bool __oom_reap_task_mm(struct mm_struct *mm)
   517	{
   518		struct vm_area_struct *vma;
   519		bool ret = true;
   520		MA_STATE(mas, &mm->mm_mt, ULONG_MAX, ULONG_MAX);
   521	
   522		/* Notify MMU notifiers about the OOM event */
 > 523		mmu_notifier_oom_enter(mm);
   524	
   525		/*
   526		 * Tell all users of get_user/copy_from_user etc... that the content
   527		 * is no longer stable. No barriers really needed because unmapping
   528		 * should imply barriers already and the reader would hit a page fault
   529		 * if it stumbled over a reaped memory.
   530		 */
   531		mm_flags_set(MMF_UNSTABLE, mm);
   532	
   533		/*
   534		 * It might start racing with the dying task and compete for shared
   535		 * resources - e.g. page table lock contention has been observed.
   536		 * Reduce those races by reaping the oom victim from the other end
   537		 * of the address space.
   538		 */
   539		mas_for_each_rev(&mas, vma, 0) {
   540			if (vma->vm_flags & (VM_HUGETLB|VM_PFNMAP))
   541				continue;
   542	
   543			/*
   544			 * Only anonymous pages have a good chance to be dropped
   545			 * without additional steps which we cannot afford as we
   546			 * are OOM already.
   547			 *
   548			 * We do not even care about fs backed pages because all
   549			 * which are reclaimable have already been reclaimed and
   550			 * we do not want to block exit_mmap by keeping mm ref
   551			 * count elevated without a good reason.
   552			 */
   553			if (vma_is_anonymous(vma) || !(vma->vm_flags & VM_SHARED)) {
   554				struct mmu_notifier_range range;
   555				struct mmu_gather tlb;
   556	
   557				mmu_notifier_range_init(&range, MMU_NOTIFY_UNMAP, 0,
   558							mm, vma->vm_start,
   559							vma->vm_end);
   560				tlb_gather_mmu(&tlb, mm);
   561				if (mmu_notifier_invalidate_range_start_nonblock(&range)) {
   562					tlb_finish_mmu(&tlb);
   563					ret = false;
   564					continue;
   565				}
   566				unmap_page_range(&tlb, vma, range.start, range.end, NULL);
   567				mmu_notifier_invalidate_range_end(&range);
   568				tlb_finish_mmu(&tlb);
   569			}
   570		}
   571	
   572		return ret;
   573	}
   574	

-- 
0-DAY CI Kernel Test Service
https://github.com/intel/lkp-tests/wiki

Re: [PATCH] KVM: mmu_notifier: make mn_invalidate_lock non-sleeping for non-blocking invalidations

[Paolo Bonzini]

[-- Attachment #1: Type: text/plain, Size: 1406 bytes --]

On Sat, Mar 28, 2026 at 3:50 PM shaikh kamaluddin
<shaikhkamal2012@gmail.com> wrote:
> +void __mmu_notifier_oom_enter(struct mm_struct *mm)
> +{
> +       struct mmu_notifier *subscription;
> +       int id;
> +       pr_info("Entering :func:%s\n", __func__);
> +       if (!mm->notifier_subscriptions)
> +               return;
> +
> +       id = srcu_read_lock(&srcu);
> +       hlist_for_each_entry_rcu(subscription,
> +                                &mm->notifier_subscriptions->list, hlist,
> +                                rcu_read_lock_held(&srcu)) {
> +               if(subscription->ops->oom_enter)
> +                       subscription->ops->oom_enter(subscription, mm);
> +
> +       }
> +       srcu_read_unlock(&srcu, id);
> +       pr_info("Done:%s\n", __func__);

Yeah, calling mmu_notifier_unregister() won't work from within this function.

One possibility is for the new method to be something like this:

       void (*after_oom_unregister)(struct mmu_notifier *subscription);

So it only has to do

kvm->mn_registered = false; /* or xchg, it's the same */
WARN_ON(rcuwait_active(&kvm->mn_memslots_update_rcuwait));
if (kvm->mn_active_invalidate_count)
    kvm->mn_active_invalidate_count = 0;
else
    WARN_ON(kvm->mmu_invalidate_in_progress);

or something like that. See the attached sketch, feel free to reuse it
as you see fit.

Paolo

[-- Attachment #2: mm.patch --]
[-- Type: text/x-patch, Size: 4393 bytes --]

diff --git a/arch/x86/kvm/svm/nested.c b/arch/x86/kvm/svm/nested.c
index 3b670ee4eb26..7b14d8099cc1 100644
--- a/arch/x86/kvm/svm/nested.c
+++ b/arch/x86/kvm/svm/nested.c
@@ -1973,12 +1973,15 @@ static gpa_t svm_translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa,
 				      struct x86_exception *exception,
 				      u64 pte_access)
 {
+	struct vcpu_svm *svm = to_svm(vcpu);
 	struct kvm_mmu *mmu = vcpu->arch.mmu;
 
 	BUG_ON(!mmu_is_nested(vcpu));
 
-	/* NPT walks are always user-walks */
-	access |= PFERR_USER_MASK;
+	/* Non-GMET walks are always user-walks */
+	if (!(svm->nested.ctl.nested_ctl & SVM_NESTED_CTL_GMET_ENABLE))
+		access |= PFERR_USER_MASK;
+
 	return mmu->gva_to_gpa(vcpu, mmu, gpa, access, exception);
 }
 
diff --git a/arch/x86/kvm/vmx/nested.c b/arch/x86/kvm/vmx/nested.c
index e4cb317807ab..4a1c1f5297c4 100644
--- a/arch/x86/kvm/vmx/nested.c
+++ b/arch/x86/kvm/vmx/nested.c
@@ -7444,6 +7444,15 @@ static gpa_t vmx_translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa,
 	struct kvm_mmu *mmu = vcpu->arch.mmu;
 
 	BUG_ON(!mmu_is_nested(vcpu));
+
+	/*
+	 * MBEC differentiates based on the effective U/S bit of
+	 * the guest page tables; not the processor CPL.
+	 */
+	access &= ~PFERR_USER_MASK;
+	if ((pte_access & ACC_USER_MASK) && (access & PFERR_GUEST_FINAL_MASK))
+		access |= PFERR_USER_MASK;
+
 	return mmu->gva_to_gpa(vcpu, mmu, gpa, access, exception);
 }
 
diff --git a/include/linux/mmu_notifier.h b/include/linux/mmu_notifier.h
index 8450e18a87c2..3c67ec15c09c 100644
--- a/include/linux/mmu_notifier.h
+++ b/include/linux/mmu_notifier.h
@@ -212,6 +212,14 @@ struct mmu_notifier_ops {
 	 */
 	struct mmu_notifier *(*alloc_notifier)(struct mm_struct *mm);
 	void (*free_notifier)(struct mmu_notifier *subscription);
+
+	/*
+	 * Any mmu notifier that defines this is automatically unregistered
+	 * when its mm is the subject of an OOM kill.  after_oom_unregister()
+	 * is invoked after all other outstanding callbacks have terminated.
+	 */
+	void (*after_oom_unregister)(struct mmu_notifier *subscription,
+				     struct mm_struct *mm);
 };
 
 /*
@@ -287,6 +295,7 @@ mmu_notifier_get(const struct mmu_notifier_ops *ops, struct mm_struct *mm)
 }
 void mmu_notifier_put(struct mmu_notifier *subscription);
 void mmu_notifier_synchronize(void);
+void mmu_notifier_oom_enter(struct mm_struct *mm);
 
 extern int mmu_notifier_register(struct mmu_notifier *subscription,
 				 struct mm_struct *mm);
@@ -661,6 +670,10 @@ static inline void mmu_notifier_synchronize(void)
 {
 }
 
+static inline void mmu_notifier_oom_enter(struct mm_struct *mm)
+{
+}
+
 #endif /* CONFIG_MMU_NOTIFIER */
 
 #endif /* _LINUX_MMU_NOTIFIER_H */
diff --git a/mm/mmu_notifier.c b/mm/mmu_notifier.c
index a6cdf3674bdc..deba056468b1 100644
--- a/mm/mmu_notifier.c
+++ b/mm/mmu_notifier.c
@@ -835,6 +835,56 @@ void mmu_notifier_unregister(struct mmu_notifier *subscription,
 }
 EXPORT_SYMBOL_GPL(mmu_notifier_unregister);
 
+void mmu_notifier_oom_enter(struct mm_struct *mm)
+{
+	struct mmu_notifier_subscriptions *subscriptions = mm->notifier_subscriptions;
+	struct mmu_notifier *subscription;
+	struct hlist_node *tmp;
+	HLIST_HEAD(oom_list);
+	int id;
+
+	id = srcu_read_lock(&srcu);
+
+	/*
+	 * Prevent further calls to the MMU notifier, except for
+	 * release and after_oom_unregister.
+	 */
+	spin_lock(&subscriptions->lock);
+	hlist_for_each_entry_safe(subscription, tmp, &subscriptions->list, hlist) {
+		if (!subscription->ops->after_oom_unregister)
+			continue;
+
+		/*
+		 * after_oom_unregister and alloc_notifier are incompatible,
+		 * because there could be other references to allocated
+		 * notifiers.
+		 */
+		if (WARN_ON(subscription->ops->alloc_notifier))
+			continue;
+
+		hlist_del_init_rcu(&subscription->hlist);
+		hlist_add_head(&subscription->hlist, &oom_list);
+	}
+	spin_unlock(&subscriptions->lock);
+
+	hlist_for_each_entry(subscription, &oom_list, hlist)
+		if (subscription->ops->release)
+			subscription->ops->release(subscription, mm);
+	srcu_read_unlock(&srcu, id);
+
+	if (hlist_empty(&oom_list))
+		return;
+
+	synchronize_srcu(&srcu);
+
+	hlist_for_each_entry_safe(subscription, tmp, &oom_list, hlist) {
+		subscription->ops->after_oom_unregister(subscription, mm);
+
+		BUG_ON(atomic_read(&mm->mm_count) <= 0);
+		mmdrop(mm);
+	}
+}
+
 static void mmu_notifier_free_rcu(struct rcu_head *rcu)
 {
 	struct mmu_notifier *subscription =

[PATCH v2 1/1] mm/mmu_notifier: Add async OOM cleanup via call_srcu()

[shaikh.kamal]

When an mm undergoes OOM kill, the OOM reaper unmaps memory while
holding the mmap_lock. MMU notifier subscribers (notably KVM) need
to be informed so they can tear down their secondary mappings. The
current synchronous unregister path can deadlock on PREEMPT_RT
because synchronize_srcu() is called from contexts that cannot
safely sleep.

This patch implements the asynchronous cleanup design proposed by
Paolo Bonzini in v1 review: a new optional after_oom_unregister
callback in struct mmu_notifier_ops, invoked after the SRCU grace
period via call_srcu() so that no readers can still reference the
subscription when cleanup runs.

The flow is:

  1. The OOM reaper calls mmu_notifier_oom_enter() from
     __oom_reap_task_mm().
  2. mmu_notifier_oom_enter() walks the subscription list and, for
     each subscriber that provides after_oom_unregister, detaches
     the subscription from the active list and schedules a
     call_srcu() callback.
  3. The deferred callback invokes after_oom_unregister once the
     grace period has elapsed and all in-flight readers have
     finished.
  4. Subsystems waiting to free structures referenced by the
     callback can call the new mmu_notifier_barrier() helper, which
     wraps srcu_barrier() to wait for all outstanding callbacks
     scheduled this way.

after_oom_unregister is mutually exclusive with alloc_notifier
because allocated notifiers can have additional outstanding
references that the OOM path cannot safely drop.

KVM is updated to provide after_oom_unregister, which clears
mn_active_invalidate_count, and to detect via hlist_unhashed() in
kvm_destroy_vm() when its subscription was already detached by the
OOM path; in that case it calls mmu_notifier_barrier() and drops
the mm reference rather than calling mmu_notifier_unregister().

Reported-by: syzbot+c3178b6b512446632bac@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=c3178b6b512446632bac
Suggested-by: Paolo Bonzini <pbonzini@redhat.com>
Link: https://lore.kernel.org/all/20260209161527.31978-1-shaikhkamal2012@gmail.com/

Signed-off-by: shaikh.kamal <shaikhkamal2012@gmail.com>
---
 include/linux/mmu_notifier.h |  10 +++
 mm/mmu_notifier.c            | 123 +++++++++++++++++++++++++++++++++++
 mm/oom_kill.c                |   3 +
 virt/kvm/kvm_main.c          |  27 +++++++-
 4 files changed, 162 insertions(+), 1 deletion(-)

diff --git a/include/linux/mmu_notifier.h b/include/linux/mmu_notifier.h
index 07a2bbaf86e9..0ccd590f55d3 100644
--- a/include/linux/mmu_notifier.h
+++ b/include/linux/mmu_notifier.h
@@ -88,6 +88,14 @@ struct mmu_notifier_ops {
 	void (*release)(struct mmu_notifier *subscription,
 			struct mm_struct *mm);

+	/*
+	 * Any mmu notifier that defines this is automatically unregistered
+	 * when its mm is the subject of an OOM kill.  after_oom_unregister()
+	 * is invoked after all other outstanding callbacks have terminated.
+	 */
+	void (*after_oom_unregister)(struct mmu_notifier *subscription,
+				     struct mm_struct *mm);
+
 	/*
 	 * clear_flush_young is called after the VM is
 	 * test-and-clearing the young/accessed bitflag in the
@@ -375,6 +383,8 @@ mmu_interval_check_retry(struct mmu_interval_notifier *interval_sub,

 extern void __mmu_notifier_subscriptions_destroy(struct mm_struct *mm);
 extern void __mmu_notifier_release(struct mm_struct *mm);
+void mmu_notifier_oom_enter(struct mm_struct *mm);
+extern void mmu_notifier_barrier(void);
 extern int __mmu_notifier_clear_flush_young(struct mm_struct *mm,
 					  unsigned long start,
 					  unsigned long end);
diff --git a/mm/mmu_notifier.c b/mm/mmu_notifier.c
index a6cdf3674bdc..b8fa58fe6b7d 100644
--- a/mm/mmu_notifier.c
+++ b/mm/mmu_notifier.c
@@ -49,6 +49,37 @@ struct mmu_notifier_subscriptions {
 	struct hlist_head deferred_list;
 };

+/*
+ * Callback structure for asynchronous OOM cleanup.
+ * Used with call_srcu() to defer after_oom_unregister callbacks
+ * until after SRCU grace period completes.
+ */
+struct mmu_notifier_oom_callback {
+	struct rcu_head rcu;
+	struct mmu_notifier *subscription;
+	struct mm_struct *mm;
+};
+
+/*
+ * Callback function invoked after SRCU grace period.
+ * Safely calls after_oom_unregister once all readers have finished.
+ */
+static void mmu_notifier_oom_callback_fn(struct rcu_head *rcu)
+{
+	struct mmu_notifier_oom_callback *cb =
+		container_of(rcu, struct mmu_notifier_oom_callback, rcu);
+
+	/* Safe - all SRCU readers have finished */
+	cb->subscription->ops->after_oom_unregister(cb->subscription, cb->mm);
+
+	/* Release mm reference taken when callback was scheduled */
+	WARN_ON_ONCE(atomic_read(&cb->mm->mm_count) <= 0);
+	mmdrop(cb->mm);
+
+	/* Free callback structure */
+	kfree(cb);
+}
+
 /*
  * This is a collision-retry read-side/write-side 'lock', a lot like a
  * seqcount, however this allows multiple write-sides to hold it at
@@ -359,6 +390,85 @@ void __mmu_notifier_release(struct mm_struct *mm)
 		mn_hlist_release(subscriptions, mm);
 }

+void mmu_notifier_oom_enter(struct mm_struct *mm)
+{
+	struct mmu_notifier_subscriptions *subscriptions =
+						mm->notifier_subscriptions;
+	struct mmu_notifier *subscription;
+	struct hlist_node *tmp;
+	HLIST_HEAD(oom_list);
+	int id;
+
+	if (!subscriptions)
+		return;
+
+	id = srcu_read_lock(&srcu);
+
+	/*
+	 * Prevent further calls to the MMU notifier, except for
+	 * release and after_oom_unregister.
+	 */
+	spin_lock(&subscriptions->lock);
+	hlist_for_each_entry_safe(subscription, tmp,
+				  &subscriptions->list, hlist) {
+		if (!subscription->ops->after_oom_unregister)
+			continue;
+
+		/*
+		 * after_oom_unregister and alloc_notifier are incompatible,
+		 * because there could be other references to allocated
+		 * notifiers.
+		 */
+		if (WARN_ON(subscription->ops->alloc_notifier))
+			continue;
+
+		hlist_del_init_rcu(&subscription->hlist);
+		hlist_add_head(&subscription->hlist, &oom_list);
+	}
+	spin_unlock(&subscriptions->lock);
+	hlist_for_each_entry(subscription, &oom_list, hlist)
+		if (subscription->ops->release)
+			subscription->ops->release(subscription, mm);
+
+	srcu_read_unlock(&srcu, id);
+
+	if (hlist_empty(&oom_list))
+		return;
+
+	hlist_for_each_entry_safe(subscription, tmp,
+				  &oom_list, hlist) {
+		struct mmu_notifier_oom_callback *cb;
+		/*
+		 * Remove from stack-based oom_list and reset hlist to unhashed state.
+		 * This sets subscription->hlist.pprev = NULL, so future callers of
+		 * mmu_notifier_unregister() (e.g. kvm_destroy_vm) will see
+		 * hlist_unhashed() == true and take the safe path, avoiding
+		 * use-after-free on the stack-allocated oom_list head.
+		 */
+		hlist_del_init(&subscription->hlist);
+
+		/*
+		 * GFP_ATOMIC failure is exceedingly rare. We cannot sleep
+		 * here (would reintroduce the deadlock this patch fixes)
+		 * and cannot call after_oom_unregister synchronously
+		 * without first waiting for SRCU readers. The subscriber
+		 * will not receive after_oom_unregister but cleanup will
+		 * eventually happen via the unregister path.
+		 */
+		cb = kmalloc(sizeof(*cb), GFP_ATOMIC);
+		if (!cb)
+			continue;
+
+		cb->subscription = subscription;
+		cb->mm = mm;
+		mmgrab(mm);
+
+		/* Schedule callback - returns immediately */
+		call_srcu(&srcu, &cb->rcu, mmu_notifier_oom_callback_fn);
+	}
+
+}
+
 /*
  * If no young bitflag is supported by the hardware, ->clear_flush_young can
  * unmap the address and return 1 or 0 depending if the mapping previously
@@ -1096,3 +1206,16 @@ void mmu_notifier_synchronize(void)
 	synchronize_srcu(&srcu);
 }
 EXPORT_SYMBOL_GPL(mmu_notifier_synchronize);
+
+/**
+ * mmu_notifier_barrier - Wait for all pending MMU notifier callbacks
+ *
+ * Waits for all call_srcu() callbacks scheduled by mmu_notifier_oom_enter()
+ * to complete. Used by subsystems during cleanup to prevent use-after-free
+ * when destroying structures accessed by the callbacks.
+ */
+void mmu_notifier_barrier(void)
+{
+	srcu_barrier(&srcu);
+}
+EXPORT_SYMBOL_GPL(mmu_notifier_barrier);
diff --git a/mm/oom_kill.c b/mm/oom_kill.c
index 5c6c95c169ee..029e041afc57 100644
--- a/mm/oom_kill.c
+++ b/mm/oom_kill.c
@@ -519,6 +519,9 @@ static bool __oom_reap_task_mm(struct mm_struct *mm)
 	bool ret = true;
 	MA_STATE(mas, &mm->mm_mt, ULONG_MAX, ULONG_MAX);

+	/* Notify MMU notifiers about the OOM event */
+	mmu_notifier_oom_enter(mm);
+
 	/*
 	 * Tell all users of get_user/copy_from_user etc... that the content
 	 * is no longer stable. No barriers really needed because unmapping
diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
index 1bc1da66b4b0..a2df83d3b413 100644
--- a/virt/kvm/kvm_main.c
+++ b/virt/kvm/kvm_main.c
@@ -885,6 +885,24 @@ static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 	srcu_read_unlock(&kvm->srcu, idx);
 }

+static void kvm_mmu_notifier_after_oom_unregister(struct mmu_notifier *mn,
+					struct mm_struct *mm)
+{
+	struct kvm *kvm;
+
+	kvm = mmu_notifier_to_kvm(mn);
+
+	/*
+	 * At this point the unregister has completed and all other callbacks
+	 * have terminated. Clean up any unbalanced invalidation counts.
+	 */
+	WARN_ON(rcuwait_active(&kvm->mn_memslots_update_rcuwait));
+	if (kvm->mn_active_invalidate_count)
+		kvm->mn_active_invalidate_count = 0;
+	else
+		WARN_ON(kvm->mmu_invalidate_in_progress);
+}
+
 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 	.invalidate_range_start	= kvm_mmu_notifier_invalidate_range_start,
 	.invalidate_range_end	= kvm_mmu_notifier_invalidate_range_end,
@@ -892,6 +910,7 @@ static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 	.clear_young		= kvm_mmu_notifier_clear_young,
 	.test_young		= kvm_mmu_notifier_test_young,
 	.release		= kvm_mmu_notifier_release,
+	.after_oom_unregister	= kvm_mmu_notifier_after_oom_unregister,
 };

 static int kvm_init_mmu_notifier(struct kvm *kvm)
@@ -1280,7 +1299,13 @@ static void kvm_destroy_vm(struct kvm *kvm)
 		kvm->buses[i] = NULL;
 	}
 	kvm_coalesced_mmio_free(kvm);
-	mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
+	if (hlist_unhashed(&kvm->mmu_notifier.hlist)) {
+		/* Subscription removed by OOM. Wait for async callback. */
+		mmu_notifier_barrier();
+		mmdrop(kvm->mm);
+	} else {
+		mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
+	}
 	/*
 	 * At this point, pending calls to invalidate_range_start()
 	 * have completed but no more MMU notifiers will run, so
--
2.43.0