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.
21 * Author: Allen M. Kay <allen.m.kay@intel.com>
22 * Author: Weidong Han <weidong.han@intel.com>
23 * Author: Ben-Ami Yassour <benami@il.ibm.com>
26 #include <linux/list.h>
27 #include <linux/kvm_host.h>
28 #include <linux/module.h>
29 #include <linux/pci.h>
30 #include <linux/stat.h>
31 #include <linux/dmar.h>
32 #include <linux/iommu.h>
33 #include <linux/intel-iommu.h>
35 static bool allow_unsafe_assigned_interrupts;
36 module_param_named(allow_unsafe_assigned_interrupts,
37 allow_unsafe_assigned_interrupts, bool, S_IRUGO | S_IWUSR);
38 MODULE_PARM_DESC(allow_unsafe_assigned_interrupts,
39 "Enable device assignment on platforms without interrupt remapping support.");
41 static int kvm_iommu_unmap_memslots(struct kvm *kvm);
42 static void kvm_iommu_put_pages(struct kvm *kvm,
43 gfn_t base_gfn, unsigned long npages);
45 static pfn_t kvm_pin_pages(struct kvm_memory_slot *slot, gfn_t gfn,
51 pfn = gfn_to_pfn_memslot(slot, gfn);
52 end_gfn = gfn + (size >> PAGE_SHIFT);
55 if (is_error_noslot_pfn(pfn))
59 gfn_to_pfn_memslot(slot, gfn++);
64 int kvm_iommu_map_pages(struct kvm *kvm, struct kvm_memory_slot *slot)
69 struct iommu_domain *domain = kvm->arch.iommu_domain;
72 /* check if iommu exists and in use */
77 end_gfn = gfn + slot->npages;
80 if (!(slot->flags & KVM_MEM_READONLY))
82 if (kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY)
86 while (gfn < end_gfn) {
87 unsigned long page_size;
89 /* Check if already mapped */
90 if (iommu_iova_to_phys(domain, gfn_to_gpa(gfn))) {
95 /* Get the page size we could use to map */
96 page_size = kvm_host_page_size(kvm, gfn);
98 /* Make sure the page_size does not exceed the memslot */
99 while ((gfn + (page_size >> PAGE_SHIFT)) > end_gfn)
102 /* Make sure gfn is aligned to the page size we want to map */
103 while ((gfn << PAGE_SHIFT) & (page_size - 1))
107 * Pin all pages we are about to map in memory. This is
108 * important because we unmap and unpin in 4kb steps later.
110 pfn = kvm_pin_pages(slot, gfn, page_size);
111 if (is_error_noslot_pfn(pfn)) {
116 /* Map into IO address space */
117 r = iommu_map(domain, gfn_to_gpa(gfn), pfn_to_hpa(pfn),
120 printk(KERN_ERR "kvm_iommu_map_address:"
121 "iommu failed to map pfn=%llx\n", pfn);
125 gfn += page_size >> PAGE_SHIFT;
133 kvm_iommu_put_pages(kvm, slot->base_gfn, gfn);
137 static int kvm_iommu_map_memslots(struct kvm *kvm)
140 struct kvm_memslots *slots;
141 struct kvm_memory_slot *memslot;
143 idx = srcu_read_lock(&kvm->srcu);
144 slots = kvm_memslots(kvm);
146 kvm_for_each_memslot(memslot, slots) {
147 r = kvm_iommu_map_pages(kvm, memslot);
151 srcu_read_unlock(&kvm->srcu, idx);
156 int kvm_assign_device(struct kvm *kvm,
157 struct kvm_assigned_dev_kernel *assigned_dev)
159 struct pci_dev *pdev = NULL;
160 struct iommu_domain *domain = kvm->arch.iommu_domain;
163 /* check if iommu exists and in use */
167 pdev = assigned_dev->dev;
171 r = iommu_attach_device(domain, &pdev->dev);
173 dev_err(&pdev->dev, "kvm assign device failed ret %d", r);
177 last_flags = kvm->arch.iommu_flags;
178 if (iommu_domain_has_cap(kvm->arch.iommu_domain,
179 IOMMU_CAP_CACHE_COHERENCY))
180 kvm->arch.iommu_flags |= KVM_IOMMU_CACHE_COHERENCY;
182 /* Check if need to update IOMMU page table for guest memory */
183 if ((last_flags ^ kvm->arch.iommu_flags) ==
184 KVM_IOMMU_CACHE_COHERENCY) {
185 kvm_iommu_unmap_memslots(kvm);
186 r = kvm_iommu_map_memslots(kvm);
191 pdev->dev_flags |= PCI_DEV_FLAGS_ASSIGNED;
193 printk(KERN_DEBUG "assign device %x:%x:%x.%x\n",
194 assigned_dev->host_segnr,
195 assigned_dev->host_busnr,
196 PCI_SLOT(assigned_dev->host_devfn),
197 PCI_FUNC(assigned_dev->host_devfn));
201 kvm_iommu_unmap_memslots(kvm);
205 int kvm_deassign_device(struct kvm *kvm,
206 struct kvm_assigned_dev_kernel *assigned_dev)
208 struct iommu_domain *domain = kvm->arch.iommu_domain;
209 struct pci_dev *pdev = NULL;
211 /* check if iommu exists and in use */
215 pdev = assigned_dev->dev;
219 iommu_detach_device(domain, &pdev->dev);
221 pdev->dev_flags &= ~PCI_DEV_FLAGS_ASSIGNED;
223 printk(KERN_DEBUG "deassign device %x:%x:%x.%x\n",
224 assigned_dev->host_segnr,
225 assigned_dev->host_busnr,
226 PCI_SLOT(assigned_dev->host_devfn),
227 PCI_FUNC(assigned_dev->host_devfn));
232 int kvm_iommu_map_guest(struct kvm *kvm)
236 if (!iommu_present(&pci_bus_type)) {
237 printk(KERN_ERR "%s: iommu not found\n", __func__);
241 mutex_lock(&kvm->slots_lock);
243 kvm->arch.iommu_domain = iommu_domain_alloc(&pci_bus_type);
244 if (!kvm->arch.iommu_domain) {
249 if (!allow_unsafe_assigned_interrupts &&
250 !iommu_domain_has_cap(kvm->arch.iommu_domain,
251 IOMMU_CAP_INTR_REMAP)) {
252 printk(KERN_WARNING "%s: No interrupt remapping support,"
253 " disallowing device assignment."
254 " Re-enble with \"allow_unsafe_assigned_interrupts=1\""
255 " module option.\n", __func__);
256 iommu_domain_free(kvm->arch.iommu_domain);
257 kvm->arch.iommu_domain = NULL;
262 r = kvm_iommu_map_memslots(kvm);
264 kvm_iommu_unmap_memslots(kvm);
267 mutex_unlock(&kvm->slots_lock);
271 static void kvm_unpin_pages(struct kvm *kvm, pfn_t pfn, unsigned long npages)
275 for (i = 0; i < npages; ++i)
276 kvm_release_pfn_clean(pfn + i);
279 static void kvm_iommu_put_pages(struct kvm *kvm,
280 gfn_t base_gfn, unsigned long npages)
282 struct iommu_domain *domain;
287 domain = kvm->arch.iommu_domain;
288 end_gfn = base_gfn + npages;
291 /* check if iommu exists and in use */
295 while (gfn < end_gfn) {
296 unsigned long unmap_pages;
299 /* Get physical address */
300 phys = iommu_iova_to_phys(domain, gfn_to_gpa(gfn));
307 pfn = phys >> PAGE_SHIFT;
309 /* Unmap address from IO address space */
310 size = iommu_unmap(domain, gfn_to_gpa(gfn), PAGE_SIZE);
311 unmap_pages = 1ULL << get_order(size);
313 /* Unpin all pages we just unmapped to not leak any memory */
314 kvm_unpin_pages(kvm, pfn, unmap_pages);
320 void kvm_iommu_unmap_pages(struct kvm *kvm, struct kvm_memory_slot *slot)
322 kvm_iommu_put_pages(kvm, slot->base_gfn, slot->npages);
325 static int kvm_iommu_unmap_memslots(struct kvm *kvm)
328 struct kvm_memslots *slots;
329 struct kvm_memory_slot *memslot;
331 idx = srcu_read_lock(&kvm->srcu);
332 slots = kvm_memslots(kvm);
334 kvm_for_each_memslot(memslot, slots)
335 kvm_iommu_unmap_pages(kvm, memslot);
337 srcu_read_unlock(&kvm->srcu, idx);
342 int kvm_iommu_unmap_guest(struct kvm *kvm)
344 struct iommu_domain *domain = kvm->arch.iommu_domain;
346 /* check if iommu exists and in use */
350 mutex_lock(&kvm->slots_lock);
351 kvm_iommu_unmap_memslots(kvm);
352 kvm->arch.iommu_domain = NULL;
353 mutex_unlock(&kvm->slots_lock);
355 iommu_domain_free(domain);