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 pfn_t kvm_pin_pages(struct kvm_memory_slot *slot, gfn_t gfn,
  47                            unsigned long npages)
  48 {
  49         gfn_t end_gfn;
  50         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, pfn_t pfn, unsigned long npages)
  66 {
  67         unsigned long i;
  68
  69         for (i = 0; i < npages; ++i)
  70                 kvm_release_pfn_clean(pfn + i);
  71 }
  72
  73 int kvm_iommu_map_pages(struct kvm *kvm, struct kvm_memory_slot *slot)
  74 {
  75         gfn_t gfn, end_gfn;
  76         pfn_t pfn;
  77         int r = 0;
  78         struct iommu_domain *domain = kvm->arch.iommu_domain;
  79         int flags;
  80
  81         /* check if iommu exists and in use */
  82         if (!domain)
  83                 return 0;
  84
  85         gfn     = slot->base_gfn;
  86         end_gfn = gfn + slot->npages;
  87
  88         flags = IOMMU_READ;
  89         if (!(slot->flags & KVM_MEM_READONLY))
  90                 flags |= IOMMU_WRITE;
  91         if (!kvm->arch.iommu_noncoherent)
  92                 flags |= IOMMU_CACHE;
  93
  94
  95         while (gfn < end_gfn) {
  96                 unsigned long page_size;
  97
  98                 /* Check if already mapped */
  99                 if (iommu_iova_to_phys(domain, gfn_to_gpa(gfn))) {
 100                         gfn += 1;
 101                         continue;
 102                 }
 103
 104                 /* Get the page size we could use to map */
 105                 page_size = kvm_host_page_size(kvm, gfn);
 106
 107                 /* Make sure the page_size does not exceed the memslot */
 108                 while ((gfn + (page_size >> PAGE_SHIFT)) > end_gfn)
 109                         page_size >>= 1;
 110
 111                 /* Make sure gfn is aligned to the page size we want to map */
 112                 while ((gfn << PAGE_SHIFT) & (page_size - 1))
 113                         page_size >>= 1;
 114
 115                 /* Make sure hva is aligned to the page size we want to map */
 116                 while (__gfn_to_hva_memslot(slot, gfn) & (page_size - 1))
 117                         page_size >>= 1;
 118
 119                 /*
 120                  * Pin all pages we are about to map in memory. This is
 121                  * important because we unmap and unpin in 4kb steps later.
 122                  */
 123                 pfn = kvm_pin_pages(slot, gfn, page_size >> PAGE_SHIFT);
 124                 if (is_error_noslot_pfn(pfn)) {
 125                         gfn += 1;
 126                         continue;
 127                 }
 128
 129                 /* Map into IO address space */
 130                 r = iommu_map(domain, gfn_to_gpa(gfn), pfn_to_hpa(pfn),
 131                               page_size, flags);
 132                 if (r) {
 133                         printk(KERN_ERR "kvm_iommu_map_address:"
 134                                "iommu failed to map pfn=%llx\n", pfn);
 135                         kvm_unpin_pages(kvm, pfn, page_size >> PAGE_SHIFT);
 136                         goto unmap_pages;
 137                 }
 138
 139                 gfn += page_size >> PAGE_SHIFT;
 140
 141                 cond_resched();
 142         }
 143
 144         return 0;
 145
 146 unmap_pages:
 147         kvm_iommu_put_pages(kvm, slot->base_gfn, gfn - slot->base_gfn);
 148         return r;
 149 }
 150
 151 static int kvm_iommu_map_memslots(struct kvm *kvm)
 152 {
 153         int idx, r = 0;
 154         struct kvm_memslots *slots;
 155         struct kvm_memory_slot *memslot;
 156
 157         if (kvm->arch.iommu_noncoherent)
 158                 kvm_arch_register_noncoherent_dma(kvm);
 159
 160         idx = srcu_read_lock(&kvm->srcu);
 161         slots = kvm_memslots(kvm);
 162
 163         kvm_for_each_memslot(memslot, slots) {
 164                 r = kvm_iommu_map_pages(kvm, memslot);
 165                 if (r)
 166                         break;
 167         }
 168         srcu_read_unlock(&kvm->srcu, idx);
 169
 170         return r;
 171 }
 172
 173 int kvm_assign_device(struct kvm *kvm, struct pci_dev *pdev)
 174 {
 175         struct iommu_domain *domain = kvm->arch.iommu_domain;
 176         int r;
 177         bool noncoherent;
 178
 179         /* check if iommu exists and in use */
 180         if (!domain)
 181                 return 0;
 182
 183         if (pdev == NULL)
 184                 return -ENODEV;
 185
 186         r = iommu_attach_device(domain, &pdev->dev);
 187         if (r) {
 188                 dev_err(&pdev->dev, "kvm assign device failed ret %d", r);
 189                 return r;
 190         }
 191
 192         noncoherent = !iommu_capable(&pci_bus_type, IOMMU_CAP_CACHE_COHERENCY);
 193
 194         /* Check if need to update IOMMU page table for guest memory */
 195         if (noncoherent != kvm->arch.iommu_noncoherent) {
 196                 kvm_iommu_unmap_memslots(kvm);
 197                 kvm->arch.iommu_noncoherent = noncoherent;
 198                 r = kvm_iommu_map_memslots(kvm);
 199                 if (r)
 200                         goto out_unmap;
 201         }
 202
 203         pci_set_dev_assigned(pdev);
 204
 205         dev_info(&pdev->dev, "kvm assign device\n");
 206
 207         return 0;
 208 out_unmap:
 209         kvm_iommu_unmap_memslots(kvm);
 210         return r;
 211 }
 212
 213 int kvm_deassign_device(struct kvm *kvm, struct pci_dev *pdev)
 214 {
 215         struct iommu_domain *domain = kvm->arch.iommu_domain;
 216
 217         /* check if iommu exists and in use */
 218         if (!domain)
 219                 return 0;
 220
 221         if (pdev == NULL)
 222                 return -ENODEV;
 223
 224         iommu_detach_device(domain, &pdev->dev);
 225
 226         pci_clear_dev_assigned(pdev);
 227
 228         dev_info(&pdev->dev, "kvm deassign device\n");
 229
 230         return 0;
 231 }
 232
 233 int kvm_iommu_map_guest(struct kvm *kvm)
 234 {
 235         int r;
 236
 237         if (!iommu_present(&pci_bus_type)) {
 238                 printk(KERN_ERR "%s: iommu not found\n", __func__);
 239                 return -ENODEV;
 240         }
 241
 242         mutex_lock(&kvm->slots_lock);
 243
 244         kvm->arch.iommu_domain = iommu_domain_alloc(&pci_bus_type);
 245         if (!kvm->arch.iommu_domain) {
 246                 r = -ENOMEM;
 247                 goto out_unlock;
 248         }
 249
 250         if (!allow_unsafe_assigned_interrupts &&
 251             !iommu_capable(&pci_bus_type, 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;
 258                 r = -EPERM;
 259                 goto out_unlock;
 260         }
 261
 262         r = kvm_iommu_map_memslots(kvm);
 263         if (r)
 264                 kvm_iommu_unmap_memslots(kvm);
 265
 266 out_unlock:
 267         mutex_unlock(&kvm->slots_lock);
 268         return r;
 269 }
 270
 271 static void kvm_iommu_put_pages(struct kvm *kvm,
 272                                 gfn_t base_gfn, unsigned long npages)
 273 {
 274         struct iommu_domain *domain;
 275         gfn_t end_gfn, gfn;
 276         pfn_t pfn;
 277         u64 phys;
 278
 279         domain  = kvm->arch.iommu_domain;
 280         end_gfn = base_gfn + npages;
 281         gfn     = base_gfn;
 282
 283         /* check if iommu exists and in use */
 284         if (!domain)
 285                 return;
 286
 287         while (gfn < end_gfn) {
 288                 unsigned long unmap_pages;
 289                 size_t size;
 290
 291                 /* Get physical address */
 292                 phys = iommu_iova_to_phys(domain, gfn_to_gpa(gfn));
 293
 294                 if (!phys) {
 295                         gfn++;
 296                         continue;
 297                 }
 298
 299                 pfn  = phys >> PAGE_SHIFT;
 300
 301                 /* Unmap address from IO address space */
 302                 size       = iommu_unmap(domain, gfn_to_gpa(gfn), PAGE_SIZE);
 303                 unmap_pages = 1ULL << get_order(size);
 304
 305                 /* Unpin all pages we just unmapped to not leak any memory */
 306                 kvm_unpin_pages(kvm, pfn, unmap_pages);
 307
 308                 gfn += unmap_pages;
 309
 310                 cond_resched();
 311         }
 312 }
 313
 314 void kvm_iommu_unmap_pages(struct kvm *kvm, struct kvm_memory_slot *slot)
 315 {
 316         kvm_iommu_put_pages(kvm, slot->base_gfn, slot->npages);
 317 }
 318
 319 static int kvm_iommu_unmap_memslots(struct kvm *kvm)
 320 {
 321         int idx;
 322         struct kvm_memslots *slots;
 323         struct kvm_memory_slot *memslot;
 324
 325         idx = srcu_read_lock(&kvm->srcu);
 326         slots = kvm_memslots(kvm);
 327
 328         kvm_for_each_memslot(memslot, slots)
 329                 kvm_iommu_unmap_pages(kvm, memslot);
 330
 331         srcu_read_unlock(&kvm->srcu, idx);
 332
 333         if (kvm->arch.iommu_noncoherent)
 334                 kvm_arch_unregister_noncoherent_dma(kvm);
 335
 336         return 0;
 337 }
 338
 339 int kvm_iommu_unmap_guest(struct kvm *kvm)
 340 {
 341         struct iommu_domain *domain = kvm->arch.iommu_domain;
 342
 343         /* check if iommu exists and in use */
 344         if (!domain)
 345                 return 0;
 346
 347         mutex_lock(&kvm->slots_lock);
 348         kvm_iommu_unmap_memslots(kvm);
 349         kvm->arch.iommu_domain = NULL;
 350         kvm->arch.iommu_noncoherent = false;
 351         mutex_unlock(&kvm->slots_lock);
 352
 353         iommu_domain_free(domain);
 354         return 0;
 355 }