Merge tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm

Pull KVM changes from Paolo Bonzini:
 "Here are the 3.13 KVM changes.  There was a lot of work on the PPC
  side: the HV and emulation flavors can now coexist in a single kernel
  is probably the most interesting change from a user point of view.

  On the x86 side there are nested virtualization improvements and a few
  bugfixes.

  ARM got transparent huge page support, improved overcommit, and
  support for big endian guests.

  Finally, there is a new interface to connect KVM with VFIO.  This
  helps with devices that use NoSnoop PCI transactions, letting the
  driver in the guest execute WBINVD instructions.  This includes some
  nVidia cards on Windows, that fail to start without these patches and
  the corresponding userspace changes"

* tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm: (146 commits)
  kvm, vmx: Fix lazy FPU on nested guest
  arm/arm64: KVM: PSCI: propagate caller endianness to the incoming vcpu
  arm/arm64: KVM: MMIO support for BE guest
  kvm, cpuid: Fix sparse warning
  kvm: Delete prototype for non-existent function kvm_check_iopl
  kvm: Delete prototype for non-existent function complete_pio
  hung_task: add method to reset detector
  pvclock: detect watchdog reset at pvclock read
  kvm: optimize out smp_mb after srcu_read_unlock
  srcu: API for barrier after srcu read unlock
  KVM: remove vm mmap method
  KVM: IOMMU: hva align mapping page size
  KVM: x86: trace cpuid emulation when called from emulator
  KVM: emulator: cleanup decode_register_operand() a bit
  KVM: emulator: check rex prefix inside decode_register()
  KVM: x86: fix emulation of "movzbl %bpl, %eax"
  kvm_host: typo fix
  KVM: x86: emulate SAHF instruction
  MAINTAINERS: add tree for kvm.git
  Documentation/kvm: add a 00-INDEX file
  ...

5 files changed, 348 insertions, 113 deletions

diff --git a/virt/kvm/Kconfig b/virt/kvm/Kconfig
index 779262f..fbe1a48 100644
--- a/virt/kvm/Kconfig
+++ b/virt/kvm/Kconfig
@@ -27,3 +27,6 @@ config HAVE_KVM_MSI
 
 config HAVE_KVM_CPU_RELAX_INTERCEPT
        bool
+
+config KVM_VFIO
+       bool
diff --git a/virt/kvm/async_pf.c b/virt/kvm/async_pf.c
index 8a39dda..8631d9c 100644
--- a/virt/kvm/async_pf.c
+++ b/virt/kvm/async_pf.c
@@ -56,7 +56,6 @@ void kvm_async_pf_vcpu_init(struct kvm_vcpu *vcpu)
 
 static void async_pf_execute(struct work_struct *work)
 {
-	struct page *page = NULL;
 	struct kvm_async_pf *apf =
 		container_of(work, struct kvm_async_pf, work);
 	struct mm_struct *mm = apf->mm;
@@ -68,14 +67,12 @@ static void async_pf_execute(struct work_struct *work)
 
 	use_mm(mm);
 	down_read(&mm->mmap_sem);
-	get_user_pages(current, mm, addr, 1, 1, 0, &page, NULL);
+	get_user_pages(current, mm, addr, 1, 1, 0, NULL, NULL);
 	up_read(&mm->mmap_sem);
 	unuse_mm(mm);
 
 	spin_lock(&vcpu->async_pf.lock);
 	list_add_tail(&apf->link, &vcpu->async_pf.done);
-	apf->page = page;
-	apf->done = true;
 	spin_unlock(&vcpu->async_pf.lock);
 
 	/*
@@ -83,7 +80,7 @@ static void async_pf_execute(struct work_struct *work)
 	 * this point
 	 */
 
-	trace_kvm_async_pf_completed(addr, page, gva);
+	trace_kvm_async_pf_completed(addr, gva);
 
 	if (waitqueue_active(&vcpu->wq))
 		wake_up_interruptible(&vcpu->wq);
@@ -99,9 +96,8 @@ void kvm_clear_async_pf_completion_queue(struct kvm_vcpu *vcpu)
 		struct kvm_async_pf *work =
 			list_entry(vcpu->async_pf.queue.next,
 				   typeof(*work), queue);
-		cancel_work_sync(&work->work);
 		list_del(&work->queue);
-		if (!work->done) { /* work was canceled */
+		if (cancel_work_sync(&work->work)) {
 			mmdrop(work->mm);
 			kvm_put_kvm(vcpu->kvm); /* == work->vcpu->kvm */
 			kmem_cache_free(async_pf_cache, work);
@@ -114,8 +110,6 @@ void kvm_clear_async_pf_completion_queue(struct kvm_vcpu *vcpu)
 			list_entry(vcpu->async_pf.done.next,
 				   typeof(*work), link);
 		list_del(&work->link);
-		if (!is_error_page(work->page))
-			kvm_release_page_clean(work->page);
 		kmem_cache_free(async_pf_cache, work);
 	}
 	spin_unlock(&vcpu->async_pf.lock);
@@ -135,14 +129,11 @@ void kvm_check_async_pf_completion(struct kvm_vcpu *vcpu)
 		list_del(&work->link);
 		spin_unlock(&vcpu->async_pf.lock);
 
-		if (work->page)
-			kvm_arch_async_page_ready(vcpu, work);
+		kvm_arch_async_page_ready(vcpu, work);
 		kvm_arch_async_page_present(vcpu, work);
 
 		list_del(&work->queue);
 		vcpu->async_pf.queued--;
-		if (!is_error_page(work->page))
-			kvm_release_page_clean(work->page);
 		kmem_cache_free(async_pf_cache, work);
 	}
 }
@@ -165,8 +156,7 @@ int kvm_setup_async_pf(struct kvm_vcpu *vcpu, gva_t gva, gfn_t gfn,
 	if (!work)
 		return 0;
 
-	work->page = NULL;
-	work->done = false;
+	work->wakeup_all = false;
 	work->vcpu = vcpu;
 	work->gva = gva;
 	work->addr = gfn_to_hva(vcpu->kvm, gfn);
@@ -206,7 +196,7 @@ int kvm_async_pf_wakeup_all(struct kvm_vcpu *vcpu)
 	if (!work)
 		return -ENOMEM;
 
-	work->page = KVM_ERR_PTR_BAD_PAGE;
+	work->wakeup_all = true;
 	INIT_LIST_HEAD(&work->queue); /* for list_del to work */
 
 	spin_lock(&vcpu->async_pf.lock);
diff --git a/virt/kvm/iommu.c b/virt/kvm/iommu.c
index 72a130b..0df7d4b 100644
--- a/virt/kvm/iommu.c
+++ b/virt/kvm/iommu.c
@@ -79,7 +79,7 @@ int kvm_iommu_map_pages(struct kvm *kvm, struct kvm_memory_slot *slot)
 	flags = IOMMU_READ;
 	if (!(slot->flags & KVM_MEM_READONLY))
 		flags |= IOMMU_WRITE;
-	if (kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY)
+	if (!kvm->arch.iommu_noncoherent)
 		flags |= IOMMU_CACHE;
 
 
@@ -103,6 +103,10 @@ int kvm_iommu_map_pages(struct kvm *kvm, struct kvm_memory_slot *slot)
 		while ((gfn << PAGE_SHIFT) & (page_size - 1))
 			page_size >>= 1;
 
+		/* Make sure hva is aligned to the page size we want to map */
+		while (__gfn_to_hva_memslot(slot, gfn) & (page_size - 1))
+			page_size >>= 1;
+
 		/*
 		 * Pin all pages we are about to map in memory. This is
 		 * important because we unmap and unpin in 4kb steps later.
@@ -140,6 +144,9 @@ static int kvm_iommu_map_memslots(struct kvm *kvm)
 	struct kvm_memslots *slots;
 	struct kvm_memory_slot *memslot;
 
+	if (kvm->arch.iommu_noncoherent)
+		kvm_arch_register_noncoherent_dma(kvm);
+
 	idx = srcu_read_lock(&kvm->srcu);
 	slots = kvm_memslots(kvm);
 
@@ -158,7 +165,8 @@ int kvm_assign_device(struct kvm *kvm,
 {
 	struct pci_dev *pdev = NULL;
 	struct iommu_domain *domain = kvm->arch.iommu_domain;
-	int r, last_flags;
+	int r;
+	bool noncoherent;
 
 	/* check if iommu exists and in use */
 	if (!domain)
@@ -174,15 +182,13 @@ int kvm_assign_device(struct kvm *kvm,
 		return r;
 	}
 
-	last_flags = kvm->arch.iommu_flags;
-	if (iommu_domain_has_cap(kvm->arch.iommu_domain,
-				 IOMMU_CAP_CACHE_COHERENCY))
-		kvm->arch.iommu_flags |= KVM_IOMMU_CACHE_COHERENCY;
+	noncoherent = !iommu_domain_has_cap(kvm->arch.iommu_domain,
+					    IOMMU_CAP_CACHE_COHERENCY);
 
 	/* Check if need to update IOMMU page table for guest memory */
-	if ((last_flags ^ kvm->arch.iommu_flags) ==
-			KVM_IOMMU_CACHE_COHERENCY) {
+	if (noncoherent != kvm->arch.iommu_noncoherent) {
 		kvm_iommu_unmap_memslots(kvm);
+		kvm->arch.iommu_noncoherent = noncoherent;
 		r = kvm_iommu_map_memslots(kvm);
 		if (r)
 			goto out_unmap;
@@ -190,11 +196,7 @@ int kvm_assign_device(struct kvm *kvm,
 
 	pdev->dev_flags |= PCI_DEV_FLAGS_ASSIGNED;
 
-	printk(KERN_DEBUG "assign device %x:%x:%x.%x\n",
-		assigned_dev->host_segnr,
-		assigned_dev->host_busnr,
-		PCI_SLOT(assigned_dev->host_devfn),
-		PCI_FUNC(assigned_dev->host_devfn));
+	dev_info(&pdev->dev, "kvm assign device\n");
 
 	return 0;
 out_unmap:
@@ -220,11 +222,7 @@ int kvm_deassign_device(struct kvm *kvm,
 
 	pdev->dev_flags &= ~PCI_DEV_FLAGS_ASSIGNED;
 
-	printk(KERN_DEBUG "deassign device %x:%x:%x.%x\n",
-		assigned_dev->host_segnr,
-		assigned_dev->host_busnr,
-		PCI_SLOT(assigned_dev->host_devfn),
-		PCI_FUNC(assigned_dev->host_devfn));
+	dev_info(&pdev->dev, "kvm deassign device\n");
 
 	return 0;
 }
@@ -336,6 +334,9 @@ static int kvm_iommu_unmap_memslots(struct kvm *kvm)
 
 	srcu_read_unlock(&kvm->srcu, idx);
 
+	if (kvm->arch.iommu_noncoherent)
+		kvm_arch_unregister_noncoherent_dma(kvm);
+
 	return 0;
 }
 
@@ -350,6 +351,7 @@ int kvm_iommu_unmap_guest(struct kvm *kvm)
 	mutex_lock(&kvm->slots_lock);
 	kvm_iommu_unmap_memslots(kvm);
 	kvm->arch.iommu_domain = NULL;
+	kvm->arch.iommu_noncoherent = false;
 	mutex_unlock(&kvm->slots_lock);
 
 	iommu_domain_free(domain);
diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
index 1cf9ccb..662f34c 100644
--- a/virt/kvm/kvm_main.c
+++ b/virt/kvm/kvm_main.c
@@ -70,7 +70,8 @@ MODULE_LICENSE("GPL");
  * 		kvm->lock --> kvm->slots_lock --> kvm->irq_lock
  */
 
-DEFINE_RAW_SPINLOCK(kvm_lock);
+DEFINE_SPINLOCK(kvm_lock);
+static DEFINE_RAW_SPINLOCK(kvm_count_lock);
 LIST_HEAD(vm_list);
 
 static cpumask_var_t cpus_hardware_enabled;
@@ -186,6 +187,7 @@ void kvm_flush_remote_tlbs(struct kvm *kvm)
 		++kvm->stat.remote_tlb_flush;
 	cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
 }
+EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
 
 void kvm_reload_remote_mmus(struct kvm *kvm)
 {
@@ -490,9 +492,9 @@ static struct kvm *kvm_create_vm(unsigned long type)
 	if (r)
 		goto out_err;
 
-	raw_spin_lock(&kvm_lock);
+	spin_lock(&kvm_lock);
 	list_add(&kvm->vm_list, &vm_list);
-	raw_spin_unlock(&kvm_lock);
+	spin_unlock(&kvm_lock);
 
 	return kvm;
 
@@ -540,13 +542,13 @@ static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
 /*
  * Free any memory in @free but not in @dont.
  */
-static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
+static void kvm_free_physmem_slot(struct kvm *kvm, struct kvm_memory_slot *free,
 				  struct kvm_memory_slot *dont)
 {
 	if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 		kvm_destroy_dirty_bitmap(free);
 
-	kvm_arch_free_memslot(free, dont);
+	kvm_arch_free_memslot(kvm, free, dont);
 
 	free->npages = 0;
 }
@@ -557,7 +559,7 @@ void kvm_free_physmem(struct kvm *kvm)
 	struct kvm_memory_slot *memslot;
 
 	kvm_for_each_memslot(memslot, slots)
-		kvm_free_physmem_slot(memslot, NULL);
+		kvm_free_physmem_slot(kvm, memslot, NULL);
 
 	kfree(kvm->memslots);
 }
@@ -581,9 +583,9 @@ static void kvm_destroy_vm(struct kvm *kvm)
 	struct mm_struct *mm = kvm->mm;
 
 	kvm_arch_sync_events(kvm);
-	raw_spin_lock(&kvm_lock);
+	spin_lock(&kvm_lock);
 	list_del(&kvm->vm_list);
-	raw_spin_unlock(&kvm_lock);
+	spin_unlock(&kvm_lock);
 	kvm_free_irq_routing(kvm);
 	for (i = 0; i < KVM_NR_BUSES; i++)
 		kvm_io_bus_destroy(kvm->buses[i]);
@@ -821,7 +823,7 @@ int __kvm_set_memory_region(struct kvm *kvm,
 	if (change == KVM_MR_CREATE) {
 		new.userspace_addr = mem->userspace_addr;
 
-		if (kvm_arch_create_memslot(&new, npages))
+		if (kvm_arch_create_memslot(kvm, &new, npages))
 			goto out_free;
 	}
 
@@ -872,6 +874,19 @@ int __kvm_set_memory_region(struct kvm *kvm,
 			goto out_free;
 	}
 
+	/* actual memory is freed via old in kvm_free_physmem_slot below */
+	if (change == KVM_MR_DELETE) {
+		new.dirty_bitmap = NULL;
+		memset(&new.arch, 0, sizeof(new.arch));
+	}
+
+	old_memslots = install_new_memslots(kvm, slots, &new);
+
+	kvm_arch_commit_memory_region(kvm, mem, &old, change);
+
+	kvm_free_physmem_slot(kvm, &old, &new);
+	kfree(old_memslots);
+
 	/*
 	 * IOMMU mapping:  New slots need to be mapped.  Old slots need to be
 	 * un-mapped and re-mapped if their base changes.  Since base change
@@ -883,29 +898,15 @@ int __kvm_set_memory_region(struct kvm *kvm,
 	 */
 	if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
 		r = kvm_iommu_map_pages(kvm, &new);
-		if (r)
-			goto out_slots;
-	}
-
-	/* actual memory is freed via old in kvm_free_physmem_slot below */
-	if (change == KVM_MR_DELETE) {
-		new.dirty_bitmap = NULL;
-		memset(&new.arch, 0, sizeof(new.arch));
+		return r;
 	}
 
-	old_memslots = install_new_memslots(kvm, slots, &new);
-
-	kvm_arch_commit_memory_region(kvm, mem, &old, change);
-
-	kvm_free_physmem_slot(&old, &new);
-	kfree(old_memslots);
-
 	return 0;
 
 out_slots:
 	kfree(slots);
 out_free:
-	kvm_free_physmem_slot(&new, &old);
+	kvm_free_physmem_slot(kvm, &new, &old);
 out:
 	return r;
 }
@@ -964,6 +965,7 @@ int kvm_get_dirty_log(struct kvm *kvm,
 out:
 	return r;
 }
+EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
 
 bool kvm_largepages_enabled(void)
 {
@@ -1654,6 +1656,7 @@ void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
 	memslot = gfn_to_memslot(kvm, gfn);
 	mark_page_dirty_in_slot(kvm, memslot, gfn);
 }
+EXPORT_SYMBOL_GPL(mark_page_dirty);
 
 /*
  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
@@ -1679,6 +1682,7 @@ void kvm_vcpu_block(struct kvm_vcpu *vcpu)
 
 	finish_wait(&vcpu->wq, &wait);
 }
+EXPORT_SYMBOL_GPL(kvm_vcpu_block);
 
 #ifndef CONFIG_S390
 /*
@@ -2271,6 +2275,11 @@ static int kvm_ioctl_create_device(struct kvm *kvm,
 		ops = &kvm_xics_ops;
 		break;
 #endif
+#ifdef CONFIG_KVM_VFIO
+	case KVM_DEV_TYPE_VFIO:
+		ops = &kvm_vfio_ops;
+		break;
+#endif
 	default:
 		return -ENODEV;
 	}
@@ -2519,44 +2528,12 @@ out:
 }
 #endif
 
-static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
-{
-	struct page *page[1];
-	unsigned long addr;
-	int npages;
-	gfn_t gfn = vmf->pgoff;
-	struct kvm *kvm = vma->vm_file->private_data;
-
-	addr = gfn_to_hva(kvm, gfn);
-	if (kvm_is_error_hva(addr))
-		return VM_FAULT_SIGBUS;
-
-	npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
-				NULL);
-	if (unlikely(npages != 1))
-		return VM_FAULT_SIGBUS;
-
-	vmf->page = page[0];
-	return 0;
-}
-
-static const struct vm_operations_struct kvm_vm_vm_ops = {
-	.fault = kvm_vm_fault,
-};
-
-static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
-{
-	vma->vm_ops = &kvm_vm_vm_ops;
-	return 0;
-}
-
 static struct file_operations kvm_vm_fops = {
 	.release        = kvm_vm_release,
 	.unlocked_ioctl = kvm_vm_ioctl,
 #ifdef CONFIG_COMPAT
 	.compat_ioctl   = kvm_vm_compat_ioctl,
 #endif
-	.mmap           = kvm_vm_mmap,
 	.llseek		= noop_llseek,
 };
 
@@ -2683,11 +2660,12 @@ static void hardware_enable_nolock(void *junk)
 	}
 }
 
-static void hardware_enable(void *junk)
+static void hardware_enable(void)
 {
-	raw_spin_lock(&kvm_lock);
-	hardware_enable_nolock(junk);
-	raw_spin_unlock(&kvm_lock);
+	raw_spin_lock(&kvm_count_lock);
+	if (kvm_usage_count)
+		hardware_enable_nolock(NULL);
+	raw_spin_unlock(&kvm_count_lock);
 }
 
 static void hardware_disable_nolock(void *junk)
@@ -2700,11 +2678,12 @@ static void hardware_disable_nolock(void *junk)
 	kvm_arch_hardware_disable(NULL);
 }
 
-static void hardware_disable(void *junk)
+static void hardware_disable(void)
 {
-	raw_spin_lock(&kvm_lock);
-	hardware_disable_nolock(junk);
-	raw_spin_unlock(&kvm_lock);
+	raw_spin_lock(&kvm_count_lock);
+	if (kvm_usage_count)
+		hardware_disable_nolock(NULL);
+	raw_spin_unlock(&kvm_count_lock);
 }
 
 static void hardware_disable_all_nolock(void)
@@ -2718,16 +2697,16 @@ static void hardware_disable_all_nolock(void)
 
 static void hardware_disable_all(void)
 {
-	raw_spin_lock(&kvm_lock);
+	raw_spin_lock(&kvm_count_lock);
 	hardware_disable_all_nolock();
-	raw_spin_unlock(&kvm_lock);
+	raw_spin_unlock(&kvm_count_lock);
 }
 
 static int hardware_enable_all(void)
 {
 	int r = 0;
 
-	raw_spin_lock(&kvm_lock);
+	raw_spin_lock(&kvm_count_lock);
 
 	kvm_usage_count++;
 	if (kvm_usage_count == 1) {
@@ -2740,7 +2719,7 @@ static int hardware_enable_all(void)
 		}
 	}
 
-	raw_spin_unlock(&kvm_lock);
+	raw_spin_unlock(&kvm_count_lock);
 
 	return r;
 }
@@ -2750,20 +2729,17 @@ static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
 {
 	int cpu = (long)v;
 
-	if (!kvm_usage_count)
-		return NOTIFY_OK;
-
 	val &= ~CPU_TASKS_FROZEN;
 	switch (val) {
 	case CPU_DYING:
 		printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
 		       cpu);
-		hardware_disable(NULL);
+		hardware_disable();
 		break;
 	case CPU_STARTING:
 		printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
 		       cpu);
-		hardware_enable(NULL);
+		hardware_enable();
 		break;
 	}
 	return NOTIFY_OK;
@@ -3056,10 +3032,10 @@ static int vm_stat_get(void *_offset, u64 *val)
 	struct kvm *kvm;
 
 	*val = 0;
-	raw_spin_lock(&kvm_lock);
+	spin_lock(&kvm_lock);
 	list_for_each_entry(kvm, &vm_list, vm_list)
 		*val += *(u32 *)((void *)kvm + offset);
-	raw_spin_unlock(&kvm_lock);
+	spin_unlock(&kvm_lock);
 	return 0;
 }
 
@@ -3073,12 +3049,12 @@ static int vcpu_stat_get(void *_offset, u64 *val)
 	int i;
 
 	*val = 0;
-	raw_spin_lock(&kvm_lock);
+	spin_lock(&kvm_lock);
 	list_for_each_entry(kvm, &vm_list, vm_list)
 		kvm_for_each_vcpu(i, vcpu, kvm)
 			*val += *(u32 *)((void *)vcpu + offset);
 
-	raw_spin_unlock(&kvm_lock);
+	spin_unlock(&kvm_lock);
 	return 0;
 }
 
@@ -3133,7 +3109,7 @@ static int kvm_suspend(void)
 static void kvm_resume(void)
 {
 	if (kvm_usage_count) {
-		WARN_ON(raw_spin_is_locked(&kvm_lock));
+		WARN_ON(raw_spin_is_locked(&kvm_count_lock));
 		hardware_enable_nolock(NULL);
 	}
 }
diff --git a/virt/kvm/vfio.c b/virt/kvm/vfio.c
new file mode 100644
index 0000000..ca4260e
--- /dev/null
+++ b/virt/kvm/vfio.c
@@ -0,0 +1,264 @@
+/*
+ * VFIO-KVM bridge pseudo device
+ *
+ * Copyright (C) 2013 Red Hat, Inc.  All rights reserved.
+ *     Author: Alex Williamson <alex.williamson@redhat.com>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include <linux/errno.h>
+#include <linux/file.h>
+#include <linux/kvm_host.h>
+#include <linux/list.h>
+#include <linux/module.h>
+#include <linux/mutex.h>
+#include <linux/slab.h>
+#include <linux/uaccess.h>
+#include <linux/vfio.h>
+
+struct kvm_vfio_group {
+	struct list_head node;
+	struct vfio_group *vfio_group;
+};
+
+struct kvm_vfio {
+	struct list_head group_list;
+	struct mutex lock;
+	bool noncoherent;
+};
+
+static struct vfio_group *kvm_vfio_group_get_external_user(struct file *filep)
+{
+	struct vfio_group *vfio_group;
+	struct vfio_group *(*fn)(struct file *);
+
+	fn = symbol_get(vfio_group_get_external_user);
+	if (!fn)
+		return ERR_PTR(-EINVAL);
+
+	vfio_group = fn(filep);
+
+	symbol_put(vfio_group_get_external_user);
+
+	return vfio_group;
+}
+
+static void kvm_vfio_group_put_external_user(struct vfio_group *vfio_group)
+{
+	void (*fn)(struct vfio_group *);
+
+	fn = symbol_get(vfio_group_put_external_user);
+	if (!fn)
+		return;
+
+	fn(vfio_group);
+
+	symbol_put(vfio_group_put_external_user);
+}
+
+/*
+ * Groups can use the same or different IOMMU domains.  If the same then
+ * adding a new group may change the coherency of groups we've previously
+ * been told about.  We don't want to care about any of that so we retest
+ * each group and bail as soon as we find one that's noncoherent.  This
+ * means we only ever [un]register_noncoherent_dma once for the whole device.
+ */
+static void kvm_vfio_update_coherency(struct kvm_device *dev)
+{
+	struct kvm_vfio *kv = dev->private;
+	bool noncoherent = false;
+	struct kvm_vfio_group *kvg;
+
+	mutex_lock(&kv->lock);
+
+	list_for_each_entry(kvg, &kv->group_list, node) {
+		/*
+		 * TODO: We need an interface to check the coherency of
+		 * the IOMMU domain this group is using.  For now, assume
+		 * it's always noncoherent.
+		 */
+		noncoherent = true;
+		break;
+	}
+
+	if (noncoherent != kv->noncoherent) {
+		kv->noncoherent = noncoherent;
+
+		if (kv->noncoherent)
+			kvm_arch_register_noncoherent_dma(dev->kvm);
+		else
+			kvm_arch_unregister_noncoherent_dma(dev->kvm);
+	}
+
+	mutex_unlock(&kv->lock);
+}
+
+static int kvm_vfio_set_group(struct kvm_device *dev, long attr, u64 arg)
+{
+	struct kvm_vfio *kv = dev->private;
+	struct vfio_group *vfio_group;
+	struct kvm_vfio_group *kvg;
+	void __user *argp = (void __user *)arg;
+	struct fd f;
+	int32_t fd;
+	int ret;
+
+	switch (attr) {
+	case KVM_DEV_VFIO_GROUP_ADD:
+		if (get_user(fd, (int32_t __user *)argp))
+			return -EFAULT;
+
+		f = fdget(fd);
+		if (!f.file)
+			return -EBADF;
+
+		vfio_group = kvm_vfio_group_get_external_user(f.file);
+		fdput(f);
+
+		if (IS_ERR(vfio_group))
+			return PTR_ERR(vfio_group);
+
+		mutex_lock(&kv->lock);
+
+		list_for_each_entry(kvg, &kv->group_list, node) {
+			if (kvg->vfio_group == vfio_group) {
+				mutex_unlock(&kv->lock);
+				kvm_vfio_group_put_external_user(vfio_group);
+				return -EEXIST;
+			}
+		}
+
+		kvg = kzalloc(sizeof(*kvg), GFP_KERNEL);
+		if (!kvg) {
+			mutex_unlock(&kv->lock);
+			kvm_vfio_group_put_external_user(vfio_group);
+			return -ENOMEM;
+		}
+
+		list_add_tail(&kvg->node, &kv->group_list);
+		kvg->vfio_group = vfio_group;
+
+		mutex_unlock(&kv->lock);
+
+		kvm_vfio_update_coherency(dev);
+
+		return 0;
+
+	case KVM_DEV_VFIO_GROUP_DEL:
+		if (get_user(fd, (int32_t __user *)argp))
+			return -EFAULT;
+
+		f = fdget(fd);
+		if (!f.file)
+			return -EBADF;
+
+		vfio_group = kvm_vfio_group_get_external_user(f.file);
+		fdput(f);
+
+		if (IS_ERR(vfio_group))
+			return PTR_ERR(vfio_group);
+
+		ret = -ENOENT;
+
+		mutex_lock(&kv->lock);
+
+		list_for_each_entry(kvg, &kv->group_list, node) {
+			if (kvg->vfio_group != vfio_group)
+				continue;
+
+			list_del(&kvg->node);
+			kvm_vfio_group_put_external_user(kvg->vfio_group);
+			kfree(kvg);
+			ret = 0;
+			break;
+		}
+
+		mutex_unlock(&kv->lock);
+
+		kvm_vfio_group_put_external_user(vfio_group);
+
+		kvm_vfio_update_coherency(dev);
+
+		return ret;
+	}
+
+	return -ENXIO;
+}
+
+static int kvm_vfio_set_attr(struct kvm_device *dev,
+			     struct kvm_device_attr *attr)
+{
+	switch (attr->group) {
+	case KVM_DEV_VFIO_GROUP:
+		return kvm_vfio_set_group(dev, attr->attr, attr->addr);
+	}
+
+	return -ENXIO;
+}
+
+static int kvm_vfio_has_attr(struct kvm_device *dev,
+			     struct kvm_device_attr *attr)
+{
+	switch (attr->group) {
+	case KVM_DEV_VFIO_GROUP:
+		switch (attr->attr) {
+		case KVM_DEV_VFIO_GROUP_ADD:
+		case KVM_DEV_VFIO_GROUP_DEL:
+			return 0;
+		}
+
+		break;
+	}
+
+	return -ENXIO;
+}
+
+static void kvm_vfio_destroy(struct kvm_device *dev)
+{
+	struct kvm_vfio *kv = dev->private;
+	struct kvm_vfio_group *kvg, *tmp;
+
+	list_for_each_entry_safe(kvg, tmp, &kv->group_list, node) {
+		kvm_vfio_group_put_external_user(kvg->vfio_group);
+		list_del(&kvg->node);
+		kfree(kvg);
+	}
+
+	kvm_vfio_update_coherency(dev);
+
+	kfree(kv);
+	kfree(dev); /* alloc by kvm_ioctl_create_device, free by .destroy */
+}
+
+static int kvm_vfio_create(struct kvm_device *dev, u32 type)
+{
+	struct kvm_device *tmp;
+	struct kvm_vfio *kv;
+
+	/* Only one VFIO "device" per VM */
+	list_for_each_entry(tmp, &dev->kvm->devices, vm_node)
+		if (tmp->ops == &kvm_vfio_ops)
+			return -EBUSY;
+
+	kv = kzalloc(sizeof(*kv), GFP_KERNEL);
+	if (!kv)
+		return -ENOMEM;
+
+	INIT_LIST_HEAD(&kv->group_list);
+	mutex_init(&kv->lock);
+
+	dev->private = kv;
+
+	return 0;
+}
+
+struct kvm_device_ops kvm_vfio_ops = {
+	.name = "kvm-vfio",
+	.create = kvm_vfio_create,
+	.destroy = kvm_vfio_destroy,
+	.set_attr = kvm_vfio_set_attr,
+	.has_attr = kvm_vfio_has_attr,
+};