2 * Copyright (c) 2006, Intel Corporation.
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms and conditions of the GNU General Public License,
6 * version 2, as published by the Free Software Foundation.
8 * This program is distributed in the hope it will be useful, but WITHOUT
9 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * You should have received a copy of the GNU General Public License along with
14 * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
15 * Place - Suite 330, Boston, MA 02111-1307 USA.
17 * Copyright (C) 2006-2008 Intel Corporation
18 * Copyright IBM Corporation, 2008
19 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
26 #include <linux/list.h>
27 #include <linux/kvm_host.h>
28 #include <linux/pci.h>
29 #include <linux/dmar.h>
30 #include <linux/iommu.h>
31 #include <linux/intel-iommu.h>
33 static int allow_unsafe_assigned_interrupts;
34 module_param_named(allow_unsafe_assigned_interrupts,
35 allow_unsafe_assigned_interrupts, bool, S_IRUGO | S_IWUSR);
36 MODULE_PARM_DESC(allow_unsafe_assigned_interrupts,
37 "Enable device assignment on platforms without interrupt remapping support.");
39 static int kvm_iommu_unmap_memslots(struct kvm *kvm);
40 static void kvm_iommu_put_pages(struct kvm *kvm,
41 gfn_t base_gfn, unsigned long npages);
43 static pfn_t kvm_pin_pages(struct kvm *kvm, struct kvm_memory_slot *slot,
44 gfn_t gfn, unsigned long size)
49 pfn = gfn_to_pfn_memslot(kvm, slot, gfn);
50 end_gfn = gfn + (size >> PAGE_SHIFT);
53 if (is_error_pfn(pfn))
57 gfn_to_pfn_memslot(kvm, slot, gfn++);
62 int kvm_iommu_map_pages(struct kvm *kvm, struct kvm_memory_slot *slot)
67 struct iommu_domain *domain = kvm->arch.iommu_domain;
70 /* check if iommu exists and in use */
75 end_gfn = gfn + slot->npages;
77 flags = IOMMU_READ | IOMMU_WRITE;
78 if (kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY)
82 while (gfn < end_gfn) {
83 unsigned long page_size;
85 /* Check if already mapped */
86 if (iommu_iova_to_phys(domain, gfn_to_gpa(gfn))) {
91 /* Get the page size we could use to map */
92 page_size = kvm_host_page_size(kvm, gfn);
94 /* Make sure the page_size does not exceed the memslot */
95 while ((gfn + (page_size >> PAGE_SHIFT)) > end_gfn)
98 /* Make sure gfn is aligned to the page size we want to map */
99 while ((gfn << PAGE_SHIFT) & (page_size - 1))
103 * Pin all pages we are about to map in memory. This is
104 * important because we unmap and unpin in 4kb steps later.
106 pfn = kvm_pin_pages(kvm, slot, gfn, page_size);
107 if (is_error_pfn(pfn)) {
112 /* Map into IO address space */
113 r = iommu_map(domain, gfn_to_gpa(gfn), pfn_to_hpa(pfn),
114 get_order(page_size), flags);
116 printk(KERN_ERR "kvm_iommu_map_address:"
117 "iommu failed to map pfn=%llx\n", pfn);
121 gfn += page_size >> PAGE_SHIFT;
129 kvm_iommu_put_pages(kvm, slot->base_gfn, gfn);
133 static int kvm_iommu_map_memslots(struct kvm *kvm)
136 struct kvm_memslots *slots;
138 idx = srcu_read_lock(&kvm->srcu);
139 slots = kvm_memslots(kvm);
141 for (i = 0; i < slots->nmemslots; i++) {
142 r = kvm_iommu_map_pages(kvm, &slots->memslots[i]);
146 srcu_read_unlock(&kvm->srcu, idx);
151 int kvm_assign_device(struct kvm *kvm,
152 struct kvm_assigned_dev_kernel *assigned_dev)
154 struct pci_dev *pdev = NULL;
155 struct iommu_domain *domain = kvm->arch.iommu_domain;
158 /* check if iommu exists and in use */
162 pdev = assigned_dev->dev;
166 r = iommu_attach_device(domain, &pdev->dev);
168 printk(KERN_ERR "assign device %x:%x:%x.%x failed",
169 pci_domain_nr(pdev->bus),
171 PCI_SLOT(pdev->devfn),
172 PCI_FUNC(pdev->devfn));
176 last_flags = kvm->arch.iommu_flags;
177 if (iommu_domain_has_cap(kvm->arch.iommu_domain,
178 IOMMU_CAP_CACHE_COHERENCY))
179 kvm->arch.iommu_flags |= KVM_IOMMU_CACHE_COHERENCY;
181 /* Check if need to update IOMMU page table for guest memory */
182 if ((last_flags ^ kvm->arch.iommu_flags) ==
183 KVM_IOMMU_CACHE_COHERENCY) {
184 kvm_iommu_unmap_memslots(kvm);
185 r = kvm_iommu_map_memslots(kvm);
190 pdev->dev_flags |= PCI_DEV_FLAGS_ASSIGNED;
192 printk(KERN_DEBUG "assign device %x:%x:%x.%x\n",
193 assigned_dev->host_segnr,
194 assigned_dev->host_busnr,
195 PCI_SLOT(assigned_dev->host_devfn),
196 PCI_FUNC(assigned_dev->host_devfn));
200 kvm_iommu_unmap_memslots(kvm);
204 int kvm_deassign_device(struct kvm *kvm,
205 struct kvm_assigned_dev_kernel *assigned_dev)
207 struct iommu_domain *domain = kvm->arch.iommu_domain;
208 struct pci_dev *pdev = NULL;
210 /* check if iommu exists and in use */
214 pdev = assigned_dev->dev;
218 iommu_detach_device(domain, &pdev->dev);
220 pdev->dev_flags &= ~PCI_DEV_FLAGS_ASSIGNED;
222 printk(KERN_DEBUG "deassign device %x:%x:%x.%x\n",
223 assigned_dev->host_segnr,
224 assigned_dev->host_busnr,
225 PCI_SLOT(assigned_dev->host_devfn),
226 PCI_FUNC(assigned_dev->host_devfn));
231 int kvm_iommu_map_guest(struct kvm *kvm)
235 if (!iommu_present(&pci_bus_type)) {
236 printk(KERN_ERR "%s: iommu not found\n", __func__);
240 kvm->arch.iommu_domain = iommu_domain_alloc(&pci_bus_type);
241 if (!kvm->arch.iommu_domain)
244 if (!allow_unsafe_assigned_interrupts &&
245 !iommu_domain_has_cap(kvm->arch.iommu_domain,
246 IOMMU_CAP_INTR_REMAP)) {
247 printk(KERN_WARNING "%s: No interrupt remapping support,"
248 " disallowing device assignment."
249 " Re-enble with \"allow_unsafe_assigned_interrupts=1\""
250 " module option.\n", __func__);
251 iommu_domain_free(kvm->arch.iommu_domain);
252 kvm->arch.iommu_domain = NULL;
256 r = kvm_iommu_map_memslots(kvm);
263 kvm_iommu_unmap_memslots(kvm);
267 static void kvm_unpin_pages(struct kvm *kvm, pfn_t pfn, unsigned long npages)
271 for (i = 0; i < npages; ++i)
272 kvm_release_pfn_clean(pfn + i);
275 static void kvm_iommu_put_pages(struct kvm *kvm,
276 gfn_t base_gfn, unsigned long npages)
278 struct iommu_domain *domain;
283 domain = kvm->arch.iommu_domain;
284 end_gfn = base_gfn + npages;
287 /* check if iommu exists and in use */
291 while (gfn < end_gfn) {
292 unsigned long unmap_pages;
295 /* Get physical address */
296 phys = iommu_iova_to_phys(domain, gfn_to_gpa(gfn));
297 pfn = phys >> PAGE_SHIFT;
299 /* Unmap address from IO address space */
300 order = iommu_unmap(domain, gfn_to_gpa(gfn), 0);
301 unmap_pages = 1ULL << order;
303 /* Unpin all pages we just unmapped to not leak any memory */
304 kvm_unpin_pages(kvm, pfn, unmap_pages);
310 static int kvm_iommu_unmap_memslots(struct kvm *kvm)
313 struct kvm_memslots *slots;
315 idx = srcu_read_lock(&kvm->srcu);
316 slots = kvm_memslots(kvm);
318 for (i = 0; i < slots->nmemslots; i++) {
319 kvm_iommu_put_pages(kvm, slots->memslots[i].base_gfn,
320 slots->memslots[i].npages);
322 srcu_read_unlock(&kvm->srcu, idx);
327 int kvm_iommu_unmap_guest(struct kvm *kvm)
329 struct iommu_domain *domain = kvm->arch.iommu_domain;
331 /* check if iommu exists and in use */
335 kvm_iommu_unmap_memslots(kvm);
336 iommu_domain_free(domain);