kvm: add support for irqfd

(Applies to kvm.git:833367b57)

KVM provides a complete virtual system environment for guests, including support for injecting interrupts modeled after the real exception/interrupt facilities present on the native platform (such as the IDT on x86). Virtual interrupts can come from a variety of sources (emulated devices, pass-through devices, etc) but all must be injected to the guest via the KVM infrastructure. This patch adds a new mechanism to inject a specific interrupt to a guest using a decoupled eventfd mechnanism: Any legal signal on the irqfd (using eventfd semantics from either userspace or kernel) will translate into an injected interrupt in the guest at the next available interrupt window.

I am reasonably satisfied with this series, so Avi please consider for inclusion.

[ Changelog:

v6:
*) Moved eventfd creation back to userspace, per Avi's request *) Dropped no longer necessary supporting patches from series *) Rebased to kvm.git:833367b57

v5:
*) Added padding to the ioctl structure
*) Added proper ref-count increment to the file before returning
success. (Needs review by Al Viro, Davide Libenzi)
*) Cleaned up error-handling path to make sure we remove ourself
from the waitq if necessary.
*) Make sure we only add ourselves to kvm->irqfds if successful
creating the irqfd in the first place.
*) Rebased to kvm.git:66b0aed4

v4:
*) Changed allocation model to create the new fd last, after
we get past the last potential error point by using Davide's new eventfd_file_create interface (Al Viro, Davide Libenzi)
*) We no longer export sys_eventfd2() since it is replaced
functionally with eventfd_file_create();
*) Rebased to kvm.git:7da2e3ba

v3:
*) The kernel now allocates the eventfd (need to export sys_eventfd2) *) Added a flags field for future expansion to kvm_irqfd() *) We properly toggle the irq level 1+0.
*) We re-use the USERSPACE_SRC_ID instead of creating our own *) Properly check for failures establishing a poll-table with eventfd *) Fixed fd/file leaks on failure
*) Rebased to lateste kvm.git::41b76d8d04

v2:
*) Dropped notifier_chain based callbacks in favor of
wait_queue_t::func and file::poll based callbacks (Thanks to Davide for the suggestion)

v1:
*) Initial release

---

 arch/x86/kvm/Makefile    |    2 -
 arch/x86/kvm/x86.c       |    1 
 include/linux/kvm.h      |    9 +++
 include/linux/kvm_host.h |    4 +
 virt/kvm/irqfd.c         |  150 ++++++++++++++++++++++++++++++++++++++++++++++
 virt/kvm/kvm_main.c      |   11 +++
 6 files changed, 176 insertions(+), 1 deletions(-)
 create mode 100644 virt/kvm/irqfd.c

index b43c4ef..d5fff51 100644
@@ -3,7 +3,7 @@
 #
 
 common-objs = $(addprefix ../../../virt/kvm/, kvm_main.o ioapic.o \
-                coalesced_mmio.o irq_comm.o)
+                coalesced_mmio.o irq_comm.o irqfd.o)
 ifeq ($(CONFIG_KVM_TRACE),y)
 common-objs += $(addprefix ../../../virt/kvm/, kvm_trace.o)
 endif

index c6d3ff3..1d062eb 100644
@@ -1026,6 +1026,7 @@ int kvm_dev_ioctl_check_extension(long ext)
 	case KVM_CAP_REINJECT_CONTROL:
 	case KVM_CAP_IRQ_INJECT_STATUS:
 	case KVM_CAP_ASSIGN_DEV_IRQ:
+	case KVM_CAP_IRQFD:
 		r = 1;
 		break;
 	case KVM_CAP_COALESCED_MMIO:

index 3db5d8d..bb4f497 100644
@@ -415,6 +415,7 @@ struct kvm_trace_rec {
 #define KVM_CAP_ASSIGN_DEV_IRQ 29
 /* Another bug in KVM_SET_USER_MEMORY_REGION fixed: */
 #define KVM_CAP_JOIN_MEMORY_REGIONS_WORKS 30
+#define KVM_CAP_IRQFD 31
 
 #ifdef KVM_CAP_IRQ_ROUTING
 

@@ -454,6 +455,13 @@ struct kvm_irq_routing {
 
 #endif
 
+struct kvm_irqfd {
+	__u32 fd;
+	__u32 gsi;
+	__u32 flags;
+	__u8  pad[20];
+};
+
 /*
  * ioctls for VM fds
  */

@@ -498,6 +506,7 @@ struct kvm_irq_routing {
 #define KVM_ASSIGN_SET_MSIX_ENTRY \
 			_IOW(KVMIO, 0x74, struct kvm_assigned_msix_entry)
 #define KVM_DEASSIGN_DEV_IRQ       _IOW(KVMIO, 0x75, struct kvm_assigned_irq)
+#define KVM_IRQFD               _IOW(KVMIO, 0x76, struct kvm_irqfd)
 
 /*
  * ioctls for vcpu fds

index 2b8df0c..dc91610 100644
@@ -134,6 +134,7 @@ struct kvm {
 	struct list_head vm_list;
 	struct kvm_io_bus mmio_bus;
 	struct kvm_io_bus pio_bus;
+	struct list_head irqfds;
 	struct kvm_vm_stat stat;
 	struct kvm_arch arch;
 	atomic_t users_count;

@@ -525,4 +526,7 @@ static inline void kvm_free_irq_routing(struct kvm *kvm) {}
 
 #endif
 
+int kvm_irqfd(struct kvm *kvm, int fd, int gsi, int flags);
+void kvm_irqfd_release(struct kvm *kvm);
+
 #endif

index 4d00942..b2db766 100644
@@ -983,6 +983,7 @@ static struct kvm *kvm_create_vm(void)
 	atomic_inc(&kvm->mm->mm_count);
 	spin_lock_init(&kvm->mmu_lock);
 	kvm_io_bus_init(&kvm->pio_bus);
+	INIT_LIST_HEAD(&kvm->irqfds);
 	mutex_init(&kvm->lock);
 	kvm_io_bus_init(&kvm->mmio_bus);
 	init_rwsem(&kvm->slots_lock);

@@ -1034,6 +1035,7 @@ static void kvm_destroy_vm(struct kvm *kvm)
 	spin_lock(&kvm_lock);
 	list_del(&kvm->vm_list);
 	spin_unlock(&kvm_lock);
+	kvm_irqfd_release(kvm);
 	kvm_free_irq_routing(kvm);
 	kvm_io_bus_destroy(&kvm->pio_bus);
 	kvm_io_bus_destroy(&kvm->mmio_bus);

@@ -2208,6 +2210,15 @@ static long kvm_vm_ioctl(struct file *filp,
 	}
 #endif
 #endif /* KVM_CAP_IRQ_ROUTING */
+	case KVM_IRQFD: {
+		struct kvm_irqfd data;
+
+		r = -EFAULT;
+		if (copy_from_user(&data, argp, sizeof data))
+			goto out;
+		r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
+		break;
+	}
 	default:
 		r = kvm_arch_vm_ioctl(filp, ioctl, arg);
 	}

Looks good. Please add a way to disassociate an irqfd (for hotunplug) and I'll merge.

I have a truly marvellous patch that fixes the bug which this
signature is too narrow to contain.

Done, and v7 sent out. Since I completed the testing on the "hypercall-fd" (I call them "iofd" for reasons described in the patch headers) I included those as well for review. Since the irqfd stuff is heavily reviewed and the iofd patches haven't seen the light of day until today, feel free to merge them independently. I included them together because I wanted to avoid forking the CAP bits, etc, for two closely related subsystems. But they technically don't need to go in together if you prefer.

Thanks Avi,
-Greg