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/pci.h>
  29 #include <linux/dmar.h>
  30 #include <linux/iommu.h>
  31 #include <linux/intel-iommu.h>
  32
  33 static int allow_unsafe_assigned_interrupts;
  34 module_param_named(allow_unsafe_assigned_interrupts,
  35                    allow_unsafe_assigned_interrupts, bool, S_IRUGO | S_IWUSR);
  36 MODULE_PARM_DESC(allow_unsafe_assigned_interrupts,
  37  "Enable device assignment on platforms without interrupt remapping support.");
  38
  39 static int kvm_iommu_unmap_memslots(struct kvm *kvm);
  40 static void kvm_iommu_put_pages(struct kvm *kvm,
  41                                 gfn_t base_gfn, unsigned long npages);
  42
  43 static pfn_t kvm_pin_pages(struct kvm *kvm, struct kvm_memory_slot *slot,
  44                            gfn_t gfn, unsigned long size)
  45 {
  46         gfn_t end_gfn;
  47         pfn_t pfn;
  48
  49         pfn     = gfn_to_pfn_memslot(kvm, slot, gfn);
  50         end_gfn = gfn + (size >> PAGE_SHIFT);
  51         gfn    += 1;
  52
  53         if (is_error_pfn(pfn))
  54                 return pfn;
  55
  56         while (gfn < end_gfn)
  57                 gfn_to_pfn_memslot(kvm, slot, gfn++);
  58
  59         return pfn;
  60 }
  61
  62 int kvm_iommu_map_pages(struct kvm *kvm, struct kvm_memory_slot *slot)
  63 {
  64         gfn_t gfn, end_gfn;
  65         pfn_t pfn;
  66         int r = 0;
  67         struct iommu_domain *domain = kvm->arch.iommu_domain;
  68         int flags;
  69
  70         /* check if iommu exists and in use */
  71         if (!domain)
  72                 return 0;
  73
  74         gfn     = slot->base_gfn;
  75         end_gfn = gfn + slot->npages;
  76
  77         flags = IOMMU_READ | IOMMU_WRITE;
  78         if (kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY)
  79                 flags |= IOMMU_CACHE;
  80
  81
  82         while (gfn < end_gfn) {
  83                 unsigned long page_size;
  84
  85                 /* Check if already mapped */
  86                 if (iommu_iova_to_phys(domain, gfn_to_gpa(gfn))) {
  87                         gfn += 1;
  88                         continue;
  89                 }
  90
  91                 /* Get the page size we could use to map */
  92                 page_size = kvm_host_page_size(kvm, gfn);
  93
  94                 /* Make sure the page_size does not exceed the memslot */
  95                 while ((gfn + (page_size >> PAGE_SHIFT)) > end_gfn)
  96                         page_size >>= 1;
  97
  98                 /* Make sure gfn is aligned to the page size we want to map */
  99                 while ((gfn << PAGE_SHIFT) & (page_size - 1))
 100                         page_size >>= 1;
 101
 102                 /*
 103                  * Pin all pages we are about to map in memory. This is
 104                  * important because we unmap and unpin in 4kb steps later.
 105                  */
 106                 pfn = kvm_pin_pages(kvm, slot, gfn, page_size);
 107                 if (is_error_pfn(pfn)) {
 108                         gfn += 1;
 109                         continue;
 110                 }
 111
 112                 /* Map into IO address space */
 113                 r = iommu_map(domain, gfn_to_gpa(gfn), pfn_to_hpa(pfn),
 114                               get_order(page_size), flags);
 115                 if (r) {
 116                         printk(KERN_ERR "kvm_iommu_map_address:"
 117                                "iommu failed to map pfn=%llx\n", pfn);
 118                         goto unmap_pages;
 119                 }
 120
 121                 gfn += page_size >> PAGE_SHIFT;
 122
 123
 124         }
 125
 126         return 0;
 127
 128 unmap_pages:
 129         kvm_iommu_put_pages(kvm, slot->base_gfn, gfn);
 130         return r;
 131 }
 132
 133 static int kvm_iommu_map_memslots(struct kvm *kvm)
 134 {
 135         int i, idx, r = 0;
 136         struct kvm_memslots *slots;
 137
 138         idx = srcu_read_lock(&kvm->srcu);
 139         slots = kvm_memslots(kvm);
 140
 141         for (i = 0; i < slots->nmemslots; i++) {
 142                 r = kvm_iommu_map_pages(kvm, &slots->memslots[i]);
 143                 if (r)
 144                         break;
 145         }
 146         srcu_read_unlock(&kvm->srcu, idx);
 147
 148         return r;
 149 }
 150
 151 int kvm_assign_device(struct kvm *kvm,
 152                       struct kvm_assigned_dev_kernel *assigned_dev)
 153 {
 154         struct pci_dev *pdev = NULL;
 155         struct iommu_domain *domain = kvm->arch.iommu_domain;
 156         int r, last_flags;
 157
 158         /* check if iommu exists and in use */
 159         if (!domain)
 160                 return 0;
 161
 162         pdev = assigned_dev->dev;
 163         if (pdev == NULL)
 164                 return -ENODEV;
 165
 166         r = iommu_attach_device(domain, &pdev->dev);
 167         if (r) {
 168                 printk(KERN_ERR "assign device %x:%x:%x.%x failed",
 169                         pci_domain_nr(pdev->bus),
 170                         pdev->bus->number,
 171                         PCI_SLOT(pdev->devfn),
 172                         PCI_FUNC(pdev->devfn));
 173                 return r;
 174         }
 175
 176         last_flags = kvm->arch.iommu_flags;
 177         if (iommu_domain_has_cap(kvm->arch.iommu_domain,
 178                                  IOMMU_CAP_CACHE_COHERENCY))
 179                 kvm->arch.iommu_flags |= KVM_IOMMU_CACHE_COHERENCY;
 180
 181         /* Check if need to update IOMMU page table for guest memory */
 182         if ((last_flags ^ kvm->arch.iommu_flags) ==
 183                         KVM_IOMMU_CACHE_COHERENCY) {
 184                 kvm_iommu_unmap_memslots(kvm);
 185                 r = kvm_iommu_map_memslots(kvm);
 186                 if (r)
 187                         goto out_unmap;
 188         }
 189
 190         pdev->dev_flags |= PCI_DEV_FLAGS_ASSIGNED;
 191
 192         printk(KERN_DEBUG "assign device %x:%x:%x.%x\n",
 193                 assigned_dev->host_segnr,
 194                 assigned_dev->host_busnr,
 195                 PCI_SLOT(assigned_dev->host_devfn),
 196                 PCI_FUNC(assigned_dev->host_devfn));
 197
 198         return 0;
 199 out_unmap:
 200         kvm_iommu_unmap_memslots(kvm);
 201         return r;
 202 }
 203
 204 int kvm_deassign_device(struct kvm *kvm,
 205                         struct kvm_assigned_dev_kernel *assigned_dev)
 206 {
 207         struct iommu_domain *domain = kvm->arch.iommu_domain;
 208         struct pci_dev *pdev = NULL;
 209
 210         /* check if iommu exists and in use */
 211         if (!domain)
 212                 return 0;
 213
 214         pdev = assigned_dev->dev;
 215         if (pdev == NULL)
 216                 return -ENODEV;
 217
 218         iommu_detach_device(domain, &pdev->dev);
 219
 220         pdev->dev_flags &= ~PCI_DEV_FLAGS_ASSIGNED;
 221
 222         printk(KERN_DEBUG "deassign device %x:%x:%x.%x\n",
 223                 assigned_dev->host_segnr,
 224                 assigned_dev->host_busnr,
 225                 PCI_SLOT(assigned_dev->host_devfn),
 226                 PCI_FUNC(assigned_dev->host_devfn));
 227
 228         return 0;
 229 }
 230
 231 int kvm_iommu_map_guest(struct kvm *kvm)
 232 {
 233         int r;
 234
 235         if (!iommu_present(&pci_bus_type)) {
 236                 printk(KERN_ERR "%s: iommu not found\n", __func__);
 237                 return -ENODEV;
 238         }
 239
 240         kvm->arch.iommu_domain = iommu_domain_alloc(&pci_bus_type);
 241         if (!kvm->arch.iommu_domain)
 242                 return -ENOMEM;
 243
 244         if (!allow_unsafe_assigned_interrupts &&
 245             !iommu_domain_has_cap(kvm->arch.iommu_domain,
 246                                   IOMMU_CAP_INTR_REMAP)) {
 247                 printk(KERN_WARNING "%s: No interrupt remapping support,"
 248                        " disallowing device assignment."
 249                        " Re-enble with \"allow_unsafe_assigned_interrupts=1\""
 250                        " module option.\n", __func__);
 251                 iommu_domain_free(kvm->arch.iommu_domain);
 252                 kvm->arch.iommu_domain = NULL;
 253                 return -EPERM;
 254         }
 255
 256         r = kvm_iommu_map_memslots(kvm);
 257         if (r)
 258                 goto out_unmap;
 259
 260         return 0;
 261
 262 out_unmap:
 263         kvm_iommu_unmap_memslots(kvm);
 264         return r;
 265 }
 266
 267 static void kvm_unpin_pages(struct kvm *kvm, pfn_t pfn, unsigned long npages)
 268 {
 269         unsigned long i;
 270
 271         for (i = 0; i < npages; ++i)
 272                 kvm_release_pfn_clean(pfn + i);
 273 }
 274
 275 static void kvm_iommu_put_pages(struct kvm *kvm,
 276                                 gfn_t base_gfn, unsigned long npages)
 277 {
 278         struct iommu_domain *domain;
 279         gfn_t end_gfn, gfn;
 280         pfn_t pfn;
 281         u64 phys;
 282
 283         domain  = kvm->arch.iommu_domain;
 284         end_gfn = base_gfn + npages;
 285         gfn     = base_gfn;
 286
 287         /* check if iommu exists and in use */
 288         if (!domain)
 289                 return;
 290
 291         while (gfn < end_gfn) {
 292                 unsigned long unmap_pages;
 293                 int order;
 294
 295                 /* Get physical address */
 296                 phys = iommu_iova_to_phys(domain, gfn_to_gpa(gfn));
 297                 pfn  = phys >> PAGE_SHIFT;
 298
 299                 /* Unmap address from IO address space */
 300                 order       = iommu_unmap(domain, gfn_to_gpa(gfn), 0);
 301                 unmap_pages = 1ULL << order;
 302
 303                 /* Unpin all pages we just unmapped to not leak any memory */
 304                 kvm_unpin_pages(kvm, pfn, unmap_pages);
 305
 306                 gfn += unmap_pages;
 307         }
 308 }
 309
 310 static int kvm_iommu_unmap_memslots(struct kvm *kvm)
 311 {
 312         int i, idx;
 313         struct kvm_memslots *slots;
 314
 315         idx = srcu_read_lock(&kvm->srcu);
 316         slots = kvm_memslots(kvm);
 317
 318         for (i = 0; i < slots->nmemslots; i++) {
 319                 kvm_iommu_put_pages(kvm, slots->memslots[i].base_gfn,
 320                                     slots->memslots[i].npages);
 321         }
 322         srcu_read_unlock(&kvm->srcu, idx);
 323
 324         return 0;
 325 }
 326
 327 int kvm_iommu_unmap_guest(struct kvm *kvm)
 328 {
 329         struct iommu_domain *domain = kvm->arch.iommu_domain;
 330
 331         /* check if iommu exists and in use */
 332         if (!domain)
 333                 return 0;
 334
 335         kvm_iommu_unmap_memslots(kvm);
 336         iommu_domain_free(domain);
 337         return 0;
 338 }