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