virt/kvm/iommu.c

   1 /*
   2  * Copyright (c) 2006, Intel Corporation.
   3  *
   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.
   7  *
   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
  11  * more details.
  12  *
  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.
  16  *
  17  * Copyright (C) 2006-2008 Intel Corporation
  18  * Copyright IBM Corporation, 2008
  19  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
  20  *
  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>
  24  */
  25
  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>
  34
  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.");
  40
  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);
  44
  45 static pfn_t kvm_pin_pages(struct kvm_memory_slot *slot, gfn_t gfn,
  46                            unsigned long npages)
  47 {
  48         gfn_t end_gfn;
  49         pfn_t pfn;
  50
  51         pfn     = gfn_to_pfn_memslot(slot, gfn);
  52         end_gfn = gfn + npages;
  53         gfn    += 1;
  54
  55         if (is_error_noslot_pfn(pfn))
  56                 return pfn;
  57
  58         while (gfn < end_gfn)
  59                 gfn_to_pfn_memslot(slot, gfn++);
  60
  61         return pfn;
  62 }
  63
  64 static void kvm_unpin_pages(struct kvm *kvm, pfn_t pfn, unsigned long npages)
  65 {
  66         unsigned long i;
  67
  68         for (i = 0; i < npages; ++i)
  69                 kvm_release_pfn_clean(pfn + i);
  70 }
  71
  72 int kvm_iommu_map_pages(struct kvm *kvm, struct kvm_memory_slot *slot)
  73 {
  74         gfn_t gfn, end_gfn;
  75         pfn_t pfn;
  76         int r = 0;
  77         struct iommu_domain *domain = kvm->arch.iommu_domain;
  78         int flags;
  79
  80         /* check if iommu exists and in use */
  81         if (!domain)
  82                 return 0;
  83
  84         gfn     = slot->base_gfn;
  85         end_gfn = gfn + slot->npages;
  86
  87         flags = IOMMU_READ;
  88         if (!(slot->flags & KVM_MEM_READONLY))
  89                 flags |= IOMMU_WRITE;
  90         if (kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY)
  91                 flags |= IOMMU_CACHE;
  92
  93
  94         while (gfn < end_gfn) {
  95                 unsigned long page_size;
  96
  97                 /* Check if already mapped */
  98                 if (iommu_iova_to_phys(domain, gfn_to_gpa(gfn))) {
  99                         gfn += 1;
 100                         continue;
 101                 }
 102
 103                 /* Get the page size we could use to map */
 104                 page_size = kvm_host_page_size(kvm, gfn);
 105
 106                 /* Make sure the page_size does not exceed the memslot */
 107                 while ((gfn + (page_size >> PAGE_SHIFT)) > end_gfn)
 108                         page_size >>= 1;
 109
 110                 /* Make sure gfn is aligned to the page size we want to map */
 111                 while ((gfn << PAGE_SHIFT) & (page_size - 1))
 112                         page_size >>= 1;
 113
 114                 /* Make sure hva is aligned to the page size we want to map */
 115                 while (__gfn_to_hva_memslot(slot, gfn) & (page_size - 1))
 116                         page_size >>= 1;
 117
 118                 /*
 119                  * Pin all pages we are about to map in memory. This is
 120                  * important because we unmap and unpin in 4kb steps later.
 121                  */
 122                 pfn = kvm_pin_pages(slot, gfn, page_size >> PAGE_SHIFT);
 123                 if (is_error_noslot_pfn(pfn)) {
 124                         gfn += 1;
 125                         continue;
 126                 }
 127
 128                 /* Map into IO address space */
 129                 r = iommu_map(domain, gfn_to_gpa(gfn), pfn_to_hpa(pfn),
 130                               page_size, flags);
 131                 if (r) {
 132                         printk(KERN_ERR "kvm_iommu_map_address:"
 133                                "iommu failed to map pfn=%llx\n", pfn);
 134                         kvm_unpin_pages(kvm, pfn, page_size >> PAGE_SHIFT);
 135                         goto unmap_pages;
 136                 }
 137
 138                 gfn += page_size >> PAGE_SHIFT;
 139
 140
 141         }
 142
 143         return 0;
 144
 145 unmap_pages:
 146         kvm_iommu_put_pages(kvm, slot->base_gfn, gfn - slot->base_gfn);
 147         return r;
 148 }
 149
 150 static int kvm_iommu_map_memslots(struct kvm *kvm)
 151 {
 152         int idx, r = 0;
 153         struct kvm_memslots *slots;
 154         struct kvm_memory_slot *memslot;
 155
 156         idx = srcu_read_lock(&kvm->srcu);
 157         slots = kvm_memslots(kvm);
 158
 159         kvm_for_each_memslot(memslot, slots) {
 160                 r = kvm_iommu_map_pages(kvm, memslot);
 161                 if (r)
 162                         break;
 163         }
 164         srcu_read_unlock(&kvm->srcu, idx);
 165
 166         return r;
 167 }
 168
 169 int kvm_assign_device(struct kvm *kvm,
 170                       struct kvm_assigned_dev_kernel *assigned_dev)
 171 {
 172         struct pci_dev *pdev = NULL;
 173         struct iommu_domain *domain = kvm->arch.iommu_domain;
 174         int r, last_flags;
 175
 176         /* check if iommu exists and in use */
 177         if (!domain)
 178                 return 0;
 179
 180         pdev = assigned_dev->dev;
 181         if (pdev == NULL)
 182                 return -ENODEV;
 183
 184         r = iommu_attach_device(domain, &pdev->dev);
 185         if (r) {
 186                 dev_err(&pdev->dev, "kvm assign device failed ret %d", r);
 187                 return r;
 188         }
 189
 190         last_flags = kvm->arch.iommu_flags;
 191         if (iommu_domain_has_cap(kvm->arch.iommu_domain,
 192                                  IOMMU_CAP_CACHE_COHERENCY))
 193                 kvm->arch.iommu_flags |= KVM_IOMMU_CACHE_COHERENCY;
 194
 195         /* Check if need to update IOMMU page table for guest memory */
 196         if ((last_flags ^ kvm->arch.iommu_flags) ==
 197                         KVM_IOMMU_CACHE_COHERENCY) {
 198                 kvm_iommu_unmap_memslots(kvm);
 199                 r = kvm_iommu_map_memslots(kvm);
 200                 if (r)
 201                         goto out_unmap;
 202         }
 203
 204         pdev->dev_flags |= PCI_DEV_FLAGS_ASSIGNED;
 205
 206         printk(KERN_DEBUG "assign device %x:%x:%x.%x\n",
 207                 assigned_dev->host_segnr,
 208                 assigned_dev->host_busnr,
 209                 PCI_SLOT(assigned_dev->host_devfn),
 210                 PCI_FUNC(assigned_dev->host_devfn));
 211
 212         return 0;
 213 out_unmap:
 214         kvm_iommu_unmap_memslots(kvm);
 215         return r;
 216 }
 217
 218 int kvm_deassign_device(struct kvm *kvm,
 219                         struct kvm_assigned_dev_kernel *assigned_dev)
 220 {
 221         struct iommu_domain *domain = kvm->arch.iommu_domain;
 222         struct pci_dev *pdev = NULL;
 223
 224         /* check if iommu exists and in use */
 225         if (!domain)
 226                 return 0;
 227
 228         pdev = assigned_dev->dev;
 229         if (pdev == NULL)
 230                 return -ENODEV;
 231
 232         iommu_detach_device(domain, &pdev->dev);
 233
 234         pdev->dev_flags &= ~PCI_DEV_FLAGS_ASSIGNED;
 235
 236         printk(KERN_DEBUG "deassign device %x:%x:%x.%x\n",
 237                 assigned_dev->host_segnr,
 238                 assigned_dev->host_busnr,
 239                 PCI_SLOT(assigned_dev->host_devfn),
 240                 PCI_FUNC(assigned_dev->host_devfn));
 241
 242         return 0;
 243 }
 244
 245 int kvm_iommu_map_guest(struct kvm *kvm)
 246 {
 247         int r;
 248
 249         if (!iommu_present(&pci_bus_type)) {
 250                 printk(KERN_ERR "%s: iommu not found\n", __func__);
 251                 return -ENODEV;
 252         }
 253
 254         mutex_lock(&kvm->slots_lock);
 255
 256         kvm->arch.iommu_domain = iommu_domain_alloc(&pci_bus_type);
 257         if (!kvm->arch.iommu_domain) {
 258                 r = -ENOMEM;
 259                 goto out_unlock;
 260         }
 261
 262         if (!allow_unsafe_assigned_interrupts &&
 263             !iommu_domain_has_cap(kvm->arch.iommu_domain,
 264                                   IOMMU_CAP_INTR_REMAP)) {
 265                 printk(KERN_WARNING "%s: No interrupt remapping support,"
 266                        " disallowing device assignment."
 267                        " Re-enble with \"allow_unsafe_assigned_interrupts=1\""
 268                        " module option.\n", __func__);
 269                 iommu_domain_free(kvm->arch.iommu_domain);
 270                 kvm->arch.iommu_domain = NULL;
 271                 r = -EPERM;
 272                 goto out_unlock;
 273         }
 274
 275         r = kvm_iommu_map_memslots(kvm);
 276         if (r)
 277                 kvm_iommu_unmap_memslots(kvm);
 278
 279 out_unlock:
 280         mutex_unlock(&kvm->slots_lock);
 281         return r;
 282 }
 283
 284 static void kvm_iommu_put_pages(struct kvm *kvm,
 285                                 gfn_t base_gfn, unsigned long npages)
 286 {
 287         struct iommu_domain *domain;
 288         gfn_t end_gfn, gfn;
 289         pfn_t pfn;
 290         u64 phys;
 291
 292         domain  = kvm->arch.iommu_domain;
 293         end_gfn = base_gfn + npages;
 294         gfn     = base_gfn;
 295
 296         /* check if iommu exists and in use */
 297         if (!domain)
 298                 return;
 299
 300         while (gfn < end_gfn) {
 301                 unsigned long unmap_pages;
 302                 size_t size;
 303
 304                 /* Get physical address */
 305                 phys = iommu_iova_to_phys(domain, gfn_to_gpa(gfn));
 306
 307                 if (!phys) {
 308                         gfn++;
 309                         continue;
 310                 }
 311
 312                 pfn  = phys >> PAGE_SHIFT;
 313
 314                 /* Unmap address from IO address space */
 315                 size       = iommu_unmap(domain, gfn_to_gpa(gfn), PAGE_SIZE);
 316                 unmap_pages = 1ULL << get_order(size);
 317
 318                 /* Unpin all pages we just unmapped to not leak any memory */
 319                 kvm_unpin_pages(kvm, pfn, unmap_pages);
 320
 321                 gfn += unmap_pages;
 322         }
 323 }
 324
 325 void kvm_iommu_unmap_pages(struct kvm *kvm, struct kvm_memory_slot *slot)
 326 {
 327         kvm_iommu_put_pages(kvm, slot->base_gfn, slot->npages);
 328 }
 329
 330 static int kvm_iommu_unmap_memslots(struct kvm *kvm)
 331 {
 332         int idx;
 333         struct kvm_memslots *slots;
 334         struct kvm_memory_slot *memslot;
 335
 336         idx = srcu_read_lock(&kvm->srcu);
 337         slots = kvm_memslots(kvm);
 338
 339         kvm_for_each_memslot(memslot, slots)
 340                 kvm_iommu_unmap_pages(kvm, memslot);
 341
 342         srcu_read_unlock(&kvm->srcu, idx);
 343
 344         return 0;
 345 }
 346
 347 int kvm_iommu_unmap_guest(struct kvm *kvm)
 348 {
 349         struct iommu_domain *domain = kvm->arch.iommu_domain;
 350
 351         /* check if iommu exists and in use */
 352         if (!domain)
 353                 return 0;
 354
 355         mutex_lock(&kvm->slots_lock);
 356         kvm_iommu_unmap_memslots(kvm);
 357         kvm->arch.iommu_domain = NULL;
 358         mutex_unlock(&kvm->slots_lock);
 359
 360         iommu_domain_free(domain);
 361         return 0;
 362 }