KVM: Add irqfd to KVM's list via the vfs_poll() callback

Add the irqfd structure to KVM's list of irqfds in kvm_irqfd_register(),
i.e. via the vfs_poll() callback.  This will allow taking irqfds.lock
across the entire registration sequence (add to waitqueue, add to list),
and more importantly will allow inserting into KVM's list if and only if
adding to the waitqueue succeeds (spoiler alert), without needing to
juggle return codes in weird ways.

Tested-by: K Prateek Nayak <kprateek.nayak@amd.com>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lore.kernel.org/r/20250522235223.3178519-5-seanjc@google.com
Signed-off-by: Sean Christopherson <seanjc@google.com>

diff --git a/virt/kvm/eventfd.c b/virt/kvm/eventfd.c
index 7b7c5269cf18b584d9fb46d2343606c2f7bc0035..170a67aad9834243505c4a1dcb294a58003535cf 100644 (file)
--- a/virt/kvm/eventfd.c
+++ b/virt/kvm/eventfd.c
@@ -245,9 +245,32 @@ irqfd_wakeup(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
        return ret;
 }
 
+static void irqfd_update(struct kvm *kvm, struct kvm_kernel_irqfd *irqfd)
+{
+       struct kvm_kernel_irq_routing_entry *e;
+       struct kvm_kernel_irq_routing_entry entries[KVM_NR_IRQCHIPS];
+       int n_entries;
+
+       lockdep_assert_held(&kvm->irqfds.lock);
+
+       n_entries = kvm_irq_map_gsi(kvm, entries, irqfd->gsi);
+
+       write_seqcount_begin(&irqfd->irq_entry_sc);
+
+       e = entries;
+       if (n_entries == 1)
+               irqfd->irq_entry = *e;
+       else
+               irqfd->irq_entry.type = 0;
+
+       write_seqcount_end(&irqfd->irq_entry_sc);
+}
+
 struct kvm_irqfd_pt {
        struct kvm_kernel_irqfd *irqfd;
+       struct kvm *kvm;
        poll_table pt;
+       int ret;
 };
 
 static void kvm_irqfd_register(struct file *file, wait_queue_head_t *wqh,
@@ -255,6 +278,25 @@ static void kvm_irqfd_register(struct file *file, wait_queue_head_t *wqh,
 {
        struct kvm_irqfd_pt *p = container_of(pt, struct kvm_irqfd_pt, pt);
        struct kvm_kernel_irqfd *irqfd = p->irqfd;
+       struct kvm_kernel_irqfd *tmp;
+       struct kvm *kvm = p->kvm;
+
+       spin_lock_irq(&kvm->irqfds.lock);
+
+       list_for_each_entry(tmp, &kvm->irqfds.items, list) {
+               if (irqfd->eventfd != tmp->eventfd)
+                       continue;
+               /* This fd is used for another irq already. */
+               p->ret = -EBUSY;
+               spin_unlock_irq(&kvm->irqfds.lock);
+               return;
+       }
+
+       irqfd_update(kvm, irqfd);
+
+       list_add_tail(&irqfd->list, &kvm->irqfds.items);
+
+       spin_unlock_irq(&kvm->irqfds.lock);
 
        /*
         * Add the irqfd as a priority waiter on the eventfd, with a custom
@@ -264,26 +306,7 @@ static void kvm_irqfd_register(struct file *file, wait_queue_head_t *wqh,
        init_waitqueue_func_entry(&irqfd->wait, irqfd_wakeup);
 
        add_wait_queue_priority(wqh, &irqfd->wait);
-}
-
-/* Must be called under irqfds.lock */
-static void irqfd_update(struct kvm *kvm, struct kvm_kernel_irqfd *irqfd)
-{
-       struct kvm_kernel_irq_routing_entry *e;
-       struct kvm_kernel_irq_routing_entry entries[KVM_NR_IRQCHIPS];
-       int n_entries;
-
-       n_entries = kvm_irq_map_gsi(kvm, entries, irqfd->gsi);
-
-       write_seqcount_begin(&irqfd->irq_entry_sc);
-
-       e = entries;
-       if (n_entries == 1)
-               irqfd->irq_entry = *e;
-       else
-               irqfd->irq_entry.type = 0;
-
-       write_seqcount_end(&irqfd->irq_entry_sc);
+       p->ret = 0;
 }
 
 #if IS_ENABLED(CONFIG_HAVE_KVM_IRQ_BYPASS)
@@ -308,7 +331,7 @@ void __weak kvm_arch_update_irqfd_routing(struct kvm_kernel_irqfd *irqfd,
 static int
 kvm_irqfd_assign(struct kvm *kvm, struct kvm_irqfd *args)
 {
-       struct kvm_kernel_irqfd *irqfd, *tmp;
+       struct kvm_kernel_irqfd *irqfd;
        struct eventfd_ctx *eventfd = NULL, *resamplefd = NULL;
        struct kvm_irqfd_pt irqfd_pt;
        int ret;
@@ -407,32 +430,22 @@ kvm_irqfd_assign(struct kvm *kvm, struct kvm_irqfd *args)
         */
        idx = srcu_read_lock(&kvm->irq_srcu);
 
-       spin_lock_irq(&kvm->irqfds.lock);
-
-       ret = 0;
-       list_for_each_entry(tmp, &kvm->irqfds.items, list) {
-               if (irqfd->eventfd != tmp->eventfd)
-                       continue;
-               /* This fd is used for another irq already. */
-               ret = -EBUSY;
-               goto fail_duplicate;
-       }
-
-       irqfd_update(kvm, irqfd);
-
-       list_add_tail(&irqfd->list, &kvm->irqfds.items);
-
-       spin_unlock_irq(&kvm->irqfds.lock);
-
        /*
-        * Register the irqfd with the eventfd by polling on the eventfd.  If
-        * there was en event pending on the eventfd prior to registering,
-        * manually trigger IRQ injection.
+        * Register the irqfd with the eventfd by polling on the eventfd, and
+        * simultaneously and the irqfd to KVM's list.  If there was en event
+        * pending on the eventfd prior to registering, manually trigger IRQ
+        * injection.
         */
        irqfd_pt.irqfd = irqfd;
+       irqfd_pt.kvm = kvm;
        init_poll_funcptr(&irqfd_pt.pt, kvm_irqfd_register);
 
        events = vfs_poll(fd_file(f), &irqfd_pt.pt);
+
+       ret = irqfd_pt.ret;
+       if (ret)
+               goto fail_poll;
+
        if (events & EPOLLIN)
                schedule_work(&irqfd->inject);
 
@@ -452,8 +465,7 @@ kvm_irqfd_assign(struct kvm *kvm, struct kvm_irqfd *args)
        srcu_read_unlock(&kvm->irq_srcu, idx);
        return 0;
 
-fail_duplicate:
-       spin_unlock_irq(&kvm->irqfds.lock);
+fail_poll:
        srcu_read_unlock(&kvm->irq_srcu, idx);
 fail:
        if (irqfd->resampler)