arch/x86/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 #include "assigned-dev.h"
  35
  36 static bool allow_unsafe_assigned_interrupts;
  37 module_param_named(allow_unsafe_assigned_interrupts,
  38                    allow_unsafe_assigned_interrupts, bool, S_IRUGO | S_IWUSR);
  39 MODULE_PARM_DESC(allow_unsafe_assigned_interrupts,
  40  "Enable device assignment on platforms without interrupt remapping support.");
  41
  42 static int kvm_iommu_unmap_memslots(struct kvm *kvm);
  43 static void kvm_iommu_put_pages(struct kvm *kvm,
  44                                 gfn_t base_gfn, unsigned long npages);
  45
  46 static kvm_pfn_t kvm_pin_pages(struct kvm_memory_slot *slot, gfn_t gfn,
  47                            unsigned long npages)
  48 {
  49         gfn_t end_gfn;
  50         kvm_pfn_t pfn;
  51
  52         pfn     = gfn_to_pfn_memslot(slot, gfn);
  53         end_gfn = gfn + npages;
  54         gfn    += 1;
  55
  56         if (is_error_noslot_pfn(pfn))
  57                 return pfn;
  58
  59         while (gfn < end_gfn)
  60                 gfn_to_pfn_memslot(slot, gfn++);
  61
  62         return pfn;
  63 }
  64
  65 static void kvm_unpin_pages(struct kvm *kvm, kvm_pfn_t pfn,
  66                 unsigned long npages)
  67 {
  68         unsigned long i;
  69
  70         for (i = 0; i < npages; ++i)
  71                 kvm_release_pfn_clean(pfn + i);
  72 }
  73
  74 int kvm_iommu_map_pages(struct kvm *kvm, struct kvm_memory_slot *slot)
  75 {
  76         gfn_t gfn, end_gfn;
  77         kvm_pfn_t pfn;
  78         int r = 0;
  79         struct iommu_domain *domain = kvm->arch.iommu_domain;
  80         int flags;
  81
  82         /* check if iommu exists and in use */
  83         if (!domain)
  84                 return 0;
  85
  86         gfn     = slot->base_gfn;
  87         end_gfn = gfn + slot->npages;
  88
  89         flags = IOMMU_READ;
  90         if (!(slot->flags & KVM_MEM_READONLY))
  91                 flags |= IOMMU_WRITE;
  92         if (!kvm->arch.iommu_noncoherent)
  93                 flags |= IOMMU_CACHE;
  94
  95
  96         while (gfn < end_gfn) {
  97                 unsigned long page_size;
  98
  99                 /* Check if already mapped */
 100                 if (iommu_iova_to_phys(domain, gfn_to_gpa(gfn))) {
 101                         gfn += 1;
 102                         continue;
 103                 }
 104
 105                 /* Get the page size we could use to map */
 106                 page_size = kvm_host_page_size(kvm, gfn);
 107
 108                 /* Make sure the page_size does not exceed the memslot */
 109                 while ((gfn + (page_size >> PAGE_SHIFT)) > end_gfn)
 110                         page_size >>= 1;
 111
 112                 /* Make sure gfn is aligned to the page size we want to map */
 113                 while ((gfn << PAGE_SHIFT) & (page_size - 1))
 114                         page_size >>= 1;
 115
 116                 /* Make sure hva is aligned to the page size we want to map */
 117                 while (__gfn_to_hva_memslot(slot, gfn) & (page_size - 1))
 118                         page_size >>= 1;
 119
 120                 /*
 121                  * Pin all pages we are about to map in memory. This is
 122                  * important because we unmap and unpin in 4kb steps later.
 123                  */
 124                 pfn = kvm_pin_pages(slot, gfn, page_size >> PAGE_SHIFT);
 125                 if (is_error_noslot_pfn(pfn)) {
 126                         gfn += 1;
 127                         continue;
 128                 }
 129
 130                 /* Map into IO address space */
 131                 r = iommu_map(domain, gfn_to_gpa(gfn), pfn_to_hpa(pfn),
 132                               page_size, flags);
 133                 if (r) {
 134                         printk(KERN_ERR "kvm_iommu_map_address:"
 135                                "iommu failed to map pfn=%llx\n", pfn);
 136                         kvm_unpin_pages(kvm, pfn, page_size >> PAGE_SHIFT);
 137                         goto unmap_pages;
 138                 }
 139
 140                 gfn += page_size >> PAGE_SHIFT;
 141
 142                 cond_resched();
 143         }
 144
 145         return 0;
 146
 147 unmap_pages:
 148         kvm_iommu_put_pages(kvm, slot->base_gfn, gfn - slot->base_gfn);
 149         return r;
 150 }
 151
 152 static int kvm_iommu_map_memslots(struct kvm *kvm)
 153 {
 154         int idx, r = 0;
 155         struct kvm_memslots *slots;
 156         struct kvm_memory_slot *memslot;
 157
 158         if (kvm->arch.iommu_noncoherent)
 159                 kvm_arch_register_noncoherent_dma(kvm);
 160
 161         idx = srcu_read_lock(&kvm->srcu);
 162         slots = kvm_memslots(kvm);
 163
 164         kvm_for_each_memslot(memslot, slots) {
 165                 r = kvm_iommu_map_pages(kvm, memslot);
 166                 if (r)
 167                         break;
 168         }
 169         srcu_read_unlock(&kvm->srcu, idx);
 170
 171         return r;
 172 }
 173
 174 int kvm_assign_device(struct kvm *kvm, struct pci_dev *pdev)
 175 {
 176         struct iommu_domain *domain = kvm->arch.iommu_domain;
 177         int r;
 178         bool noncoherent;
 179
 180         /* check if iommu exists and in use */
 181         if (!domain)
 182                 return 0;
 183
 184         if (pdev == NULL)
 185                 return -ENODEV;
 186
 187         r = iommu_attach_device(domain, &pdev->dev);
 188         if (r) {
 189                 dev_err(&pdev->dev, "kvm assign device failed ret %d", r);
 190                 return r;
 191         }
 192
 193         noncoherent = !iommu_capable(&pci_bus_type, IOMMU_CAP_CACHE_COHERENCY);
 194
 195         /* Check if need to update IOMMU page table for guest memory */
 196         if (noncoherent != kvm->arch.iommu_noncoherent) {
 197                 kvm_iommu_unmap_memslots(kvm);
 198                 kvm->arch.iommu_noncoherent = noncoherent;
 199                 r = kvm_iommu_map_memslots(kvm);
 200                 if (r)
 201                         goto out_unmap;
 202         }
 203
 204         kvm_arch_start_assignment(kvm);
 205         pci_set_dev_assigned(pdev);
 206
 207         dev_info(&pdev->dev, "kvm assign device\n");
 208
 209         return 0;
 210 out_unmap:
 211         kvm_iommu_unmap_memslots(kvm);
 212         return r;
 213 }
 214
 215 int kvm_deassign_device(struct kvm *kvm, struct pci_dev *pdev)
 216 {
 217         struct iommu_domain *domain = kvm->arch.iommu_domain;
 218
 219         /* check if iommu exists and in use */
 220         if (!domain)
 221                 return 0;
 222
 223         if (pdev == NULL)
 224                 return -ENODEV;
 225
 226         iommu_detach_device(domain, &pdev->dev);
 227
 228         pci_clear_dev_assigned(pdev);
 229         kvm_arch_end_assignment(kvm);
 230
 231         dev_info(&pdev->dev, "kvm deassign device\n");
 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_capable(&pci_bus_type, IOMMU_CAP_INTR_REMAP)) {
 255                 printk(KERN_WARNING "%s: No interrupt remapping support,"
 256                        " disallowing device assignment."
 257                        " Re-enable with \"allow_unsafe_assigned_interrupts=1\""
 258                        " module option.\n", __func__);
 259                 iommu_domain_free(kvm->arch.iommu_domain);
 260                 kvm->arch.iommu_domain = NULL;
 261                 r = -EPERM;
 262                 goto out_unlock;
 263         }
 264
 265         r = kvm_iommu_map_memslots(kvm);
 266         if (r)
 267                 kvm_iommu_unmap_memslots(kvm);
 268
 269 out_unlock:
 270         mutex_unlock(&kvm->slots_lock);
 271         return r;
 272 }
 273
 274 static void kvm_iommu_put_pages(struct kvm *kvm,
 275                                 gfn_t base_gfn, unsigned long npages)
 276 {
 277         struct iommu_domain *domain;
 278         gfn_t end_gfn, gfn;
 279         kvm_pfn_t pfn;
 280         u64 phys;
 281
 282         domain  = kvm->arch.iommu_domain;
 283         end_gfn = base_gfn + npages;
 284         gfn     = base_gfn;
 285
 286         /* check if iommu exists and in use */
 287         if (!domain)
 288                 return;
 289
 290         while (gfn < end_gfn) {
 291                 unsigned long unmap_pages;
 292                 size_t size;
 293
 294                 /* Get physical address */
 295                 phys = iommu_iova_to_phys(domain, gfn_to_gpa(gfn));
 296
 297                 if (!phys) {
 298                         gfn++;
 299                         continue;
 300                 }
 301
 302                 pfn  = phys >> PAGE_SHIFT;
 303
 304                 /* Unmap address from IO address space */
 305                 size       = iommu_unmap(domain, gfn_to_gpa(gfn), PAGE_SIZE);
 306                 unmap_pages = 1ULL << get_order(size);
 307
 308                 /* Unpin all pages we just unmapped to not leak any memory */
 309                 kvm_unpin_pages(kvm, pfn, unmap_pages);
 310
 311                 gfn += unmap_pages;
 312
 313                 cond_resched();
 314         }
 315 }
 316
 317 void kvm_iommu_unmap_pages(struct kvm *kvm, struct kvm_memory_slot *slot)
 318 {
 319         kvm_iommu_put_pages(kvm, slot->base_gfn, slot->npages);
 320 }
 321
 322 static int kvm_iommu_unmap_memslots(struct kvm *kvm)
 323 {
 324         int idx;
 325         struct kvm_memslots *slots;
 326         struct kvm_memory_slot *memslot;
 327
 328         idx = srcu_read_lock(&kvm->srcu);
 329         slots = kvm_memslots(kvm);
 330
 331         kvm_for_each_memslot(memslot, slots)
 332                 kvm_iommu_unmap_pages(kvm, memslot);
 333
 334         srcu_read_unlock(&kvm->srcu, idx);
 335
 336         if (kvm->arch.iommu_noncoherent)
 337                 kvm_arch_unregister_noncoherent_dma(kvm);
 338
 339         return 0;
 340 }
 341
 342 int kvm_iommu_unmap_guest(struct kvm *kvm)
 343 {
 344         struct iommu_domain *domain = kvm->arch.iommu_domain;
 345
 346         /* check if iommu exists and in use */
 347         if (!domain)
 348                 return 0;
 349
 350         mutex_lock(&kvm->slots_lock);
 351         kvm_iommu_unmap_memslots(kvm);
 352         kvm->arch.iommu_domain = NULL;
 353         kvm->arch.iommu_noncoherent = false;
 354         mutex_unlock(&kvm->slots_lock);
 355
 356         iommu_domain_free(domain);
 357         return 0;
 358 }