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 int 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 *kvm, struct kvm_memory_slot *slot,
  46                            gfn_t gfn, unsigned long size)
  47 {
  48         gfn_t end_gfn;
  49         pfn_t pfn;
  50
  51         pfn     = gfn_to_pfn_memslot(kvm, slot, gfn);
  52         end_gfn = gfn + (size >> PAGE_SHIFT);
  53         gfn    += 1;
  54
  55         if (is_error_pfn(pfn))
  56                 return pfn;
  57
  58         while (gfn < end_gfn)
  59                 gfn_to_pfn_memslot(kvm, 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 | IOMMU_WRITE;
  80         if (kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY)
  81                 flags |= IOMMU_CACHE;
  82
  83
  84         while (gfn < end_gfn) {
  85                 unsigned long page_size;
  86
  87                 /* Check if already mapped */
  88                 if (iommu_iova_to_phys(domain, gfn_to_gpa(gfn))) {
  89                         gfn += 1;
  90                         continue;
  91                 }
  92
  93                 /* Get the page size we could use to map */
  94                 page_size = kvm_host_page_size(kvm, gfn);
  95
  96                 /* Make sure the page_size does not exceed the memslot */
  97                 while ((gfn + (page_size >> PAGE_SHIFT)) > end_gfn)
  98                         page_size >>= 1;
  99
 100                 /* Make sure gfn is aligned to the page size we want to map */
 101                 while ((gfn << PAGE_SHIFT) & (page_size - 1))
 102                         page_size >>= 1;
 103
 104                 /*
 105                  * Pin all pages we are about to map in memory. This is
 106                  * important because we unmap and unpin in 4kb steps later.
 107                  */
 108                 pfn = kvm_pin_pages(kvm, slot, gfn, page_size);
 109                 if (is_error_pfn(pfn)) {
 110                         gfn += 1;
 111                         continue;
 112                 }
 113
 114                 /* Map into IO address space */
 115                 r = iommu_map(domain, gfn_to_gpa(gfn), pfn_to_hpa(pfn),
 116                               get_order(page_size), flags);
 117                 if (r) {
 118                         printk(KERN_ERR "kvm_iommu_map_address:"
 119                                "iommu failed to map pfn=%llx\n", pfn);
 120                         goto unmap_pages;
 121                 }
 122
 123                 gfn += page_size >> PAGE_SHIFT;
 124
 125
 126         }
 127
 128         return 0;
 129
 130 unmap_pages:
 131         kvm_iommu_put_pages(kvm, slot->base_gfn, gfn);
 132         return r;
 133 }
 134
 135 static int kvm_iommu_map_memslots(struct kvm *kvm)
 136 {
 137         int i, idx, r = 0;
 138         struct kvm_memslots *slots;
 139
 140         idx = srcu_read_lock(&kvm->srcu);
 141         slots = kvm_memslots(kvm);
 142
 143         for (i = 0; i < slots->nmemslots; i++) {
 144                 r = kvm_iommu_map_pages(kvm, &slots->memslots[i]);
 145                 if (r)
 146                         break;
 147         }
 148         srcu_read_unlock(&kvm->srcu, idx);
 149
 150         return r;
 151 }
 152
 153 int kvm_assign_device(struct kvm *kvm,
 154                       struct kvm_assigned_dev_kernel *assigned_dev)
 155 {
 156         struct pci_dev *pdev = NULL;
 157         struct iommu_domain *domain = kvm->arch.iommu_domain;
 158         int r, last_flags;
 159
 160         /* check if iommu exists and in use */
 161         if (!domain)
 162                 return 0;
 163
 164         pdev = assigned_dev->dev;
 165         if (pdev == NULL)
 166                 return -ENODEV;
 167
 168         r = iommu_attach_device(domain, &pdev->dev);
 169         if (r) {
 170                 printk(KERN_ERR "assign device %x:%x:%x.%x failed",
 171                         pci_domain_nr(pdev->bus),
 172                         pdev->bus->number,
 173                         PCI_SLOT(pdev->devfn),
 174                         PCI_FUNC(pdev->devfn));
 175                 return r;
 176         }
 177
 178         last_flags = kvm->arch.iommu_flags;
 179         if (iommu_domain_has_cap(kvm->arch.iommu_domain,
 180                                  IOMMU_CAP_CACHE_COHERENCY))
 181                 kvm->arch.iommu_flags |= KVM_IOMMU_CACHE_COHERENCY;
 182
 183         /* Check if need to update IOMMU page table for guest memory */
 184         if ((last_flags ^ kvm->arch.iommu_flags) ==
 185                         KVM_IOMMU_CACHE_COHERENCY) {
 186                 kvm_iommu_unmap_memslots(kvm);
 187                 r = kvm_iommu_map_memslots(kvm);
 188                 if (r)
 189                         goto out_unmap;
 190         }
 191
 192         pdev->dev_flags |= PCI_DEV_FLAGS_ASSIGNED;
 193
 194         printk(KERN_DEBUG "assign device %x:%x:%x.%x\n",
 195                 assigned_dev->host_segnr,
 196                 assigned_dev->host_busnr,
 197                 PCI_SLOT(assigned_dev->host_devfn),
 198                 PCI_FUNC(assigned_dev->host_devfn));
 199
 200         return 0;
 201 out_unmap:
 202         kvm_iommu_unmap_memslots(kvm);
 203         return r;
 204 }
 205
 206 int kvm_deassign_device(struct kvm *kvm,
 207                         struct kvm_assigned_dev_kernel *assigned_dev)
 208 {
 209         struct iommu_domain *domain = kvm->arch.iommu_domain;
 210         struct pci_dev *pdev = NULL;
 211
 212         /* check if iommu exists and in use */
 213         if (!domain)
 214                 return 0;
 215
 216         pdev = assigned_dev->dev;
 217         if (pdev == NULL)
 218                 return -ENODEV;
 219
 220         iommu_detach_device(domain, &pdev->dev);
 221
 222         pdev->dev_flags &= ~PCI_DEV_FLAGS_ASSIGNED;
 223
 224         printk(KERN_DEBUG "deassign device %x:%x:%x.%x\n",
 225                 assigned_dev->host_segnr,
 226                 assigned_dev->host_busnr,
 227                 PCI_SLOT(assigned_dev->host_devfn),
 228                 PCI_FUNC(assigned_dev->host_devfn));
 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         kvm->arch.iommu_domain = iommu_domain_alloc(&pci_bus_type);
 243         if (!kvm->arch.iommu_domain)
 244                 return -ENOMEM;
 245
 246         if (!allow_unsafe_assigned_interrupts &&
 247             !iommu_domain_has_cap(kvm->arch.iommu_domain,
 248                                   IOMMU_CAP_INTR_REMAP)) {
 249                 printk(KERN_WARNING "%s: No interrupt remapping support,"
 250                        " disallowing device assignment."
 251                        " Re-enble with \"allow_unsafe_assigned_interrupts=1\""
 252                        " module option.\n", __func__);
 253                 iommu_domain_free(kvm->arch.iommu_domain);
 254                 kvm->arch.iommu_domain = NULL;
 255                 return -EPERM;
 256         }
 257
 258         r = kvm_iommu_map_memslots(kvm);
 259         if (r)
 260                 goto out_unmap;
 261
 262         return 0;
 263
 264 out_unmap:
 265         kvm_iommu_unmap_memslots(kvm);
 266         return r;
 267 }
 268
 269 static void kvm_unpin_pages(struct kvm *kvm, pfn_t pfn, unsigned long npages)
 270 {
 271         unsigned long i;
 272
 273         for (i = 0; i < npages; ++i)
 274                 kvm_release_pfn_clean(pfn + i);
 275 }
 276
 277 static void kvm_iommu_put_pages(struct kvm *kvm,
 278                                 gfn_t base_gfn, unsigned long npages)
 279 {
 280         struct iommu_domain *domain;
 281         gfn_t end_gfn, gfn;
 282         pfn_t pfn;
 283         u64 phys;
 284
 285         domain  = kvm->arch.iommu_domain;
 286         end_gfn = base_gfn + npages;
 287         gfn     = base_gfn;
 288
 289         /* check if iommu exists and in use */
 290         if (!domain)
 291                 return;
 292
 293         while (gfn < end_gfn) {
 294                 unsigned long unmap_pages;
 295                 int order;
 296
 297                 /* Get physical address */
 298                 phys = iommu_iova_to_phys(domain, gfn_to_gpa(gfn));
 299                 pfn  = phys >> PAGE_SHIFT;
 300
 301                 /* Unmap address from IO address space */
 302                 order       = iommu_unmap(domain, gfn_to_gpa(gfn), 0);
 303                 unmap_pages = 1ULL << order;
 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 }
 311
 312 static int kvm_iommu_unmap_memslots(struct kvm *kvm)
 313 {
 314         int i, idx;
 315         struct kvm_memslots *slots;
 316
 317         idx = srcu_read_lock(&kvm->srcu);
 318         slots = kvm_memslots(kvm);
 319
 320         for (i = 0; i < slots->nmemslots; i++) {
 321                 kvm_iommu_put_pages(kvm, slots->memslots[i].base_gfn,
 322                                     slots->memslots[i].npages);
 323         }
 324         srcu_read_unlock(&kvm->srcu, idx);
 325
 326         return 0;
 327 }
 328
 329 int kvm_iommu_unmap_guest(struct kvm *kvm)
 330 {
 331         struct iommu_domain *domain = kvm->arch.iommu_domain;
 332
 333         /* check if iommu exists and in use */
 334         if (!domain)
 335                 return 0;
 336
 337         kvm_iommu_unmap_memslots(kvm);
 338         iommu_domain_free(domain);
 339         return 0;
 340 }