arch/x86/pci/xen.c

   1 /*
   2  * Xen PCI Frontend Stub - puts some "dummy" functions in to the Linux
   3  *                         x86 PCI core to support the Xen PCI Frontend
   4  *
   5  *   Author: Ryan Wilson <hap9@epoch.ncsc.mil>
   6  */
   7 #include <linux/module.h>
   8 #include <linux/init.h>
   9 #include <linux/pci.h>
  10 #include <linux/acpi.h>
  11
  12 #include <linux/io.h>
  13 #include <asm/io_apic.h>
  14 #include <asm/pci_x86.h>
  15
  16 #include <asm/xen/hypervisor.h>
  17
  18 #include <xen/features.h>
  19 #include <xen/events.h>
  20 #include <asm/xen/pci.h>
  21
  22 static int xen_pcifront_enable_irq(struct pci_dev *dev)
  23 {
  24         int rc;
  25         int share = 1;
  26         int pirq;
  27         u8 gsi;
  28
  29         rc = pci_read_config_byte(dev, PCI_INTERRUPT_LINE, &gsi);
  30         if (rc < 0) {
  31                 dev_warn(&dev->dev, "Xen PCI: failed to read interrupt line: %d\n",
  32                          rc);
  33                 return rc;
  34         }
  35
  36         rc = xen_allocate_pirq_gsi(gsi);
  37         if (rc < 0) {
  38                 dev_warn(&dev->dev, "Xen PCI: failed to allocate a PIRQ for GSI%d: %d\n",
  39                          gsi, rc);
  40                 return rc;
  41         }
  42         pirq = rc;
  43
  44         if (gsi < NR_IRQS_LEGACY)
  45                 share = 0;
  46
  47         rc = xen_bind_pirq_gsi_to_irq(gsi, pirq, share, "pcifront");
  48         if (rc < 0) {
  49                 dev_warn(&dev->dev, "Xen PCI: failed to bind GSI%d (PIRQ%d) to IRQ: %d\n",
  50                          gsi, pirq, rc);
  51                 return rc;
  52         }
  53
  54         dev->irq = rc;
  55         dev_info(&dev->dev, "Xen PCI mapped GSI%d to IRQ%d\n", gsi, dev->irq);
  56         return 0;
  57 }
  58
  59 #ifdef CONFIG_ACPI
  60 static int acpi_register_gsi_xen_hvm(struct device *dev, u32 gsi,
  61                                  int trigger, int polarity)
  62 {
  63         int rc, irq;
  64         struct physdev_map_pirq map_irq;
  65         int shareable = 0;
  66         char *name;
  67
  68         if (!xen_hvm_domain())
  69                 return -1;
  70
  71         map_irq.domid = DOMID_SELF;
  72         map_irq.type = MAP_PIRQ_TYPE_GSI;
  73         map_irq.index = gsi;
  74         map_irq.pirq = -1;
  75
  76         rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
  77         if (rc) {
  78                 printk(KERN_WARNING "xen map irq failed %d\n", rc);
  79                 return -1;
  80         }
  81
  82         if (trigger == ACPI_EDGE_SENSITIVE) {
  83                 shareable = 0;
  84                 name = "ioapic-edge";
  85         } else {
  86                 shareable = 1;
  87                 name = "ioapic-level";
  88         }
  89
  90         irq = xen_bind_pirq_gsi_to_irq(gsi, map_irq.pirq, shareable, name);
  91
  92         printk(KERN_DEBUG "xen: --> irq=%d, pirq=%d\n", irq, map_irq.pirq);
  93
  94         return irq;
  95 }
  96 #endif
  97
  98 #ifdef CONFIG_XEN_DOM0
  99 static int xen_register_pirq(u32 gsi, int gsi_override, int triggering)
 100 {
 101         int rc, pirq, irq = -1;
 102         struct physdev_map_pirq map_irq;
 103         int shareable = 0;
 104         char *name;
 105
 106         if (!xen_pv_domain())
 107                 return -1;
 108
 109         if (triggering == ACPI_EDGE_SENSITIVE) {
 110                 shareable = 0;
 111                 name = "ioapic-edge";
 112         } else {
 113                 shareable = 1;
 114                 name = "ioapic-level";
 115         }
 116         pirq = xen_allocate_pirq_gsi(gsi);
 117         if (pirq < 0)
 118                 goto out;
 119
 120         if (gsi_override >= 0)
 121                 irq = xen_bind_pirq_gsi_to_irq(gsi_override, pirq, shareable, name);
 122         else
 123                 irq = xen_bind_pirq_gsi_to_irq(gsi, pirq, shareable, name);
 124         if (irq < 0)
 125                 goto out;
 126
 127         printk(KERN_DEBUG "xen: --> pirq=%d -> irq=%d (gsi=%d)\n", pirq, irq, gsi);
 128
 129         map_irq.domid = DOMID_SELF;
 130         map_irq.type = MAP_PIRQ_TYPE_GSI;
 131         map_irq.index = gsi;
 132         map_irq.pirq = pirq;
 133
 134         rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
 135         if (rc) {
 136                 printk(KERN_WARNING "xen map irq failed %d\n", rc);
 137                 return -1;
 138         }
 139
 140 out:
 141         return irq;
 142 }
 143
 144 static int xen_register_gsi(u32 gsi, int gsi_override, int triggering, int polarity)
 145 {
 146         int rc, irq;
 147         struct physdev_setup_gsi setup_gsi;
 148
 149         if (!xen_pv_domain())
 150                 return -1;
 151
 152         printk(KERN_DEBUG "xen: registering gsi %u triggering %d polarity %d\n",
 153                         gsi, triggering, polarity);
 154
 155         irq = xen_register_pirq(gsi, gsi_override, triggering);
 156
 157         setup_gsi.gsi = gsi;
 158         setup_gsi.triggering = (triggering == ACPI_EDGE_SENSITIVE ? 0 : 1);
 159         setup_gsi.polarity = (polarity == ACPI_ACTIVE_HIGH ? 0 : 1);
 160
 161         rc = HYPERVISOR_physdev_op(PHYSDEVOP_setup_gsi, &setup_gsi);
 162         if (rc == -EEXIST)
 163                 printk(KERN_INFO "Already setup the GSI :%d\n", gsi);
 164         else if (rc) {
 165                 printk(KERN_ERR "Failed to setup GSI :%d, err_code:%d\n",
 166                                 gsi, rc);
 167         }
 168
 169         return irq;
 170 }
 171
 172 static int acpi_register_gsi_xen(struct device *dev, u32 gsi,
 173                                  int trigger, int polarity)
 174 {
 175         return xen_register_gsi(gsi, -1 /* no GSI override */, trigger, polarity);
 176 }
 177 #endif
 178
 179 #if defined(CONFIG_PCI_MSI)
 180 #include <linux/msi.h>
 181 #include <asm/msidef.h>
 182
 183 struct xen_pci_frontend_ops *xen_pci_frontend;
 184 EXPORT_SYMBOL_GPL(xen_pci_frontend);
 185
 186 /*
 187  * For MSI interrupts we have to use drivers/xen/event.s functions to
 188  * allocate an irq_desc and setup the right */
 189 static int xen_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
 190 {
 191         int irq, ret, i;
 192         struct msi_desc *msidesc;
 193         int *v;
 194
 195         v = kzalloc(sizeof(int) * max(1, nvec), GFP_KERNEL);
 196         if (!v)
 197                 return -ENOMEM;
 198
 199         if (type == PCI_CAP_ID_MSIX)
 200                 ret = xen_pci_frontend_enable_msix(dev, v, nvec);
 201         else
 202                 ret = xen_pci_frontend_enable_msi(dev, v);
 203         if (ret)
 204                 goto error;
 205         i = 0;
 206         list_for_each_entry(msidesc, &dev->msi_list, list) {
 207                 irq = xen_bind_pirq_msi_to_irq(dev, msidesc, v[i], 0,
 208                                                (type == PCI_CAP_ID_MSIX) ?
 209                                                "pcifront-msi-x" :
 210                                                "pcifront-msi",
 211                                                 DOMID_SELF);
 212                 if (irq < 0)
 213                         goto free;
 214                 i++;
 215         }
 216         kfree(v);
 217         return 0;
 218
 219 error:
 220         dev_err(&dev->dev, "Xen PCI frontend has not registered MSI/MSI-X support!\n");
 221 free:
 222         kfree(v);
 223         return ret;
 224 }
 225
 226 #define XEN_PIRQ_MSI_DATA  (MSI_DATA_TRIGGER_EDGE | \
 227                 MSI_DATA_LEVEL_ASSERT | (3 << 8) | MSI_DATA_VECTOR(0))
 228
 229 static void xen_msi_compose_msg(struct pci_dev *pdev, unsigned int pirq,
 230                 struct msi_msg *msg)
 231 {
 232         /* We set vector == 0 to tell the hypervisor we don't care about it,
 233          * but we want a pirq setup instead.
 234          * We use the dest_id field to pass the pirq that we want. */
 235         msg->address_hi = MSI_ADDR_BASE_HI | MSI_ADDR_EXT_DEST_ID(pirq);
 236         msg->address_lo =
 237                 MSI_ADDR_BASE_LO |
 238                 MSI_ADDR_DEST_MODE_PHYSICAL |
 239                 MSI_ADDR_REDIRECTION_CPU |
 240                 MSI_ADDR_DEST_ID(pirq);
 241
 242         msg->data = XEN_PIRQ_MSI_DATA;
 243 }
 244
 245 static int xen_hvm_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
 246 {
 247         int irq, pirq;
 248         struct msi_desc *msidesc;
 249         struct msi_msg msg;
 250
 251         list_for_each_entry(msidesc, &dev->msi_list, list) {
 252                 __read_msi_msg(msidesc, &msg);
 253                 pirq = MSI_ADDR_EXT_DEST_ID(msg.address_hi) |
 254                         ((msg.address_lo >> MSI_ADDR_DEST_ID_SHIFT) & 0xff);
 255                 if (msg.data != XEN_PIRQ_MSI_DATA ||
 256                     xen_irq_from_pirq(pirq) < 0) {
 257                         pirq = xen_allocate_pirq_msi(dev, msidesc);
 258                         if (pirq < 0)
 259                                 goto error;
 260                         xen_msi_compose_msg(dev, pirq, &msg);
 261                         __write_msi_msg(msidesc, &msg);
 262                         dev_dbg(&dev->dev, "xen: msi bound to pirq=%d\n", pirq);
 263                 } else {
 264                         dev_dbg(&dev->dev,
 265                                 "xen: msi already bound to pirq=%d\n", pirq);
 266                 }
 267                 irq = xen_bind_pirq_msi_to_irq(dev, msidesc, pirq, 0,
 268                                                (type == PCI_CAP_ID_MSIX) ?
 269                                                "msi-x" : "msi",
 270                                                DOMID_SELF);
 271                 if (irq < 0)
 272                         goto error;
 273                 dev_dbg(&dev->dev,
 274                         "xen: msi --> pirq=%d --> irq=%d\n", pirq, irq);
 275         }
 276         return 0;
 277
 278 error:
 279         dev_err(&dev->dev,
 280                 "Xen PCI frontend has not registered MSI/MSI-X support!\n");
 281         return -ENODEV;
 282 }
 283
 284 #ifdef CONFIG_XEN_DOM0
 285 static int xen_initdom_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
 286 {
 287         int ret = 0;
 288         struct msi_desc *msidesc;
 289
 290         list_for_each_entry(msidesc, &dev->msi_list, list) {
 291                 struct physdev_map_pirq map_irq;
 292                 domid_t domid;
 293
 294                 domid = ret = xen_find_device_domain_owner(dev);
 295                 /* N.B. Casting int's -ENODEV to uint16_t results in 0xFFED,
 296                  * hence check ret value for < 0. */
 297                 if (ret < 0)
 298                         domid = DOMID_SELF;
 299
 300                 memset(&map_irq, 0, sizeof(map_irq));
 301                 map_irq.domid = domid;
 302                 map_irq.type = MAP_PIRQ_TYPE_MSI;
 303                 map_irq.index = -1;
 304                 map_irq.pirq = -1;
 305                 map_irq.bus = dev->bus->number;
 306                 map_irq.devfn = dev->devfn;
 307
 308                 if (type == PCI_CAP_ID_MSIX) {
 309                         int pos;
 310                         u32 table_offset, bir;
 311
 312                         pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
 313
 314                         pci_read_config_dword(dev, pos + PCI_MSIX_TABLE,
 315                                               &table_offset);
 316                         bir = (u8)(table_offset & PCI_MSIX_FLAGS_BIRMASK);
 317
 318                         map_irq.table_base = pci_resource_start(dev, bir);
 319                         map_irq.entry_nr = msidesc->msi_attrib.entry_nr;
 320                 }
 321
 322                 ret = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
 323                 if (ret) {
 324                         dev_warn(&dev->dev, "xen map irq failed %d for %d domain\n",
 325                                  ret, domid);
 326                         goto out;
 327                 }
 328
 329                 ret = xen_bind_pirq_msi_to_irq(dev, msidesc,
 330                                                map_irq.pirq, map_irq.index,
 331                                                (type == PCI_CAP_ID_MSIX) ?
 332                                                "msi-x" : "msi",
 333                                                 domid);
 334                 if (ret < 0)
 335                         goto out;
 336         }
 337         ret = 0;
 338 out:
 339         return ret;
 340 }
 341 #endif
 342
 343 static void xen_teardown_msi_irqs(struct pci_dev *dev)
 344 {
 345         struct msi_desc *msidesc;
 346
 347         msidesc = list_entry(dev->msi_list.next, struct msi_desc, list);
 348         if (msidesc->msi_attrib.is_msix)
 349                 xen_pci_frontend_disable_msix(dev);
 350         else
 351                 xen_pci_frontend_disable_msi(dev);
 352
 353         /* Free the IRQ's and the msidesc using the generic code. */
 354         default_teardown_msi_irqs(dev);
 355 }
 356
 357 static void xen_teardown_msi_irq(unsigned int irq)
 358 {
 359         xen_destroy_irq(irq);
 360 }
 361
 362 #endif
 363
 364
 365 int __init pci_xen_init(void)
 366 {
 367         if (!xen_pv_domain() || xen_initial_domain())
 368                 return -ENODEV;
 369
 370         printk(KERN_INFO "PCI: setting up Xen PCI frontend stub\n");
 371
 372         pcibios_set_cache_line_size();
 373
 374         pcibios_enable_irq = xen_pcifront_enable_irq;
 375         pcibios_disable_irq = NULL;
 376
 377 #ifdef CONFIG_ACPI
 378         /* Keep ACPI out of the picture */
 379         acpi_noirq = 1;
 380 #endif
 381
 382 #ifdef CONFIG_PCI_MSI
 383         x86_msi.setup_msi_irqs = xen_setup_msi_irqs;
 384         x86_msi.teardown_msi_irq = xen_teardown_msi_irq;
 385         x86_msi.teardown_msi_irqs = xen_teardown_msi_irqs;
 386 #endif
 387         return 0;
 388 }
 389
 390 int __init pci_xen_hvm_init(void)
 391 {
 392         if (!xen_feature(XENFEAT_hvm_pirqs))
 393                 return 0;
 394
 395 #ifdef CONFIG_ACPI
 396         /*
 397          * We don't want to change the actual ACPI delivery model,
 398          * just how GSIs get registered.
 399          */
 400         __acpi_register_gsi = acpi_register_gsi_xen_hvm;
 401 #endif
 402
 403 #ifdef CONFIG_PCI_MSI
 404         x86_msi.setup_msi_irqs = xen_hvm_setup_msi_irqs;
 405         x86_msi.teardown_msi_irq = xen_teardown_msi_irq;
 406 #endif
 407         return 0;
 408 }
 409
 410 #ifdef CONFIG_XEN_DOM0
 411 static __init void xen_setup_acpi_sci(void)
 412 {
 413         int rc;
 414         int trigger, polarity;
 415         int gsi = acpi_sci_override_gsi;
 416         int irq = -1;
 417         int gsi_override = -1;
 418
 419         if (!gsi)
 420                 return;
 421
 422         rc = acpi_get_override_irq(gsi, &trigger, &polarity);
 423         if (rc) {
 424                 printk(KERN_WARNING "xen: acpi_get_override_irq failed for acpi"
 425                                 " sci, rc=%d\n", rc);
 426                 return;
 427         }
 428         trigger = trigger ? ACPI_LEVEL_SENSITIVE : ACPI_EDGE_SENSITIVE;
 429         polarity = polarity ? ACPI_ACTIVE_LOW : ACPI_ACTIVE_HIGH;
 430
 431         printk(KERN_INFO "xen: sci override: global_irq=%d trigger=%d "
 432                         "polarity=%d\n", gsi, trigger, polarity);
 433
 434         /* Before we bind the GSI to a Linux IRQ, check whether
 435          * we need to override it with bus_irq (IRQ) value. Usually for
 436          * IRQs below IRQ_LEGACY_IRQ this holds IRQ == GSI, as so:
 437          *  ACPI: INT_SRC_OVR (bus 0 bus_irq 9 global_irq 9 low level)
 438          * but there are oddballs where the IRQ != GSI:
 439          *  ACPI: INT_SRC_OVR (bus 0 bus_irq 9 global_irq 20 low level)
 440          * which ends up being: gsi_to_irq[9] == 20
 441          * (which is what acpi_gsi_to_irq ends up calling when starting the
 442          * the ACPI interpreter and keels over since IRQ 9 has not been
 443          * setup as we had setup IRQ 20 for it).
 444          */
 445         /* Check whether the GSI != IRQ */
 446         if (acpi_gsi_to_irq(gsi, &irq) == 0) {
 447                 if (irq >= 0 && irq != gsi)
 448                         /* Bugger, we MUST have that IRQ. */
 449                         gsi_override = irq;
 450         }
 451
 452         gsi = xen_register_gsi(gsi, gsi_override, trigger, polarity);
 453         printk(KERN_INFO "xen: acpi sci %d\n", gsi);
 454
 455         return;
 456 }
 457
 458 static int __init pci_xen_initial_domain(void)
 459 {
 460 #ifdef CONFIG_PCI_MSI
 461         x86_msi.setup_msi_irqs = xen_initdom_setup_msi_irqs;
 462         x86_msi.teardown_msi_irq = xen_teardown_msi_irq;
 463 #endif
 464         xen_setup_acpi_sci();
 465         __acpi_register_gsi = acpi_register_gsi_xen;
 466
 467         return 0;
 468 }
 469
 470 void __init xen_setup_pirqs(void)
 471 {
 472         int pirq, irq;
 473
 474         pci_xen_initial_domain();
 475
 476         if (0 == nr_ioapics) {
 477                 for (irq = 0; irq < NR_IRQS_LEGACY; irq++) {
 478                         pirq = xen_allocate_pirq_gsi(irq);
 479                         if (WARN(pirq < 0,
 480                                  "Could not allocate PIRQ for legacy interrupt\n"))
 481                                 break;
 482                         irq = xen_bind_pirq_gsi_to_irq(irq, pirq, 0, "xt-pic");
 483                 }
 484                 return;
 485         }
 486
 487         /* Pre-allocate legacy irqs */
 488         for (irq = 0; irq < NR_IRQS_LEGACY; irq++) {
 489                 int trigger, polarity;
 490
 491                 if (acpi_get_override_irq(irq, &trigger, &polarity) == -1)
 492                         continue;
 493
 494                 xen_register_pirq(irq, -1 /* no GSI override */,
 495                         trigger ? ACPI_LEVEL_SENSITIVE : ACPI_EDGE_SENSITIVE);
 496         }
 497 }
 498
 499 struct xen_device_domain_owner {
 500         domid_t domain;
 501         struct pci_dev *dev;
 502         struct list_head list;
 503 };
 504
 505 static DEFINE_SPINLOCK(dev_domain_list_spinlock);
 506 static struct list_head dev_domain_list = LIST_HEAD_INIT(dev_domain_list);
 507
 508 static struct xen_device_domain_owner *find_device(struct pci_dev *dev)
 509 {
 510         struct xen_device_domain_owner *owner;
 511
 512         list_for_each_entry(owner, &dev_domain_list, list) {
 513                 if (owner->dev == dev)
 514                         return owner;
 515         }
 516         return NULL;
 517 }
 518
 519 int xen_find_device_domain_owner(struct pci_dev *dev)
 520 {
 521         struct xen_device_domain_owner *owner;
 522         int domain = -ENODEV;
 523
 524         spin_lock(&dev_domain_list_spinlock);
 525         owner = find_device(dev);
 526         if (owner)
 527                 domain = owner->domain;
 528         spin_unlock(&dev_domain_list_spinlock);
 529         return domain;
 530 }
 531 EXPORT_SYMBOL_GPL(xen_find_device_domain_owner);
 532
 533 int xen_register_device_domain_owner(struct pci_dev *dev, uint16_t domain)
 534 {
 535         struct xen_device_domain_owner *owner;
 536
 537         owner = kzalloc(sizeof(struct xen_device_domain_owner), GFP_KERNEL);
 538         if (!owner)
 539                 return -ENODEV;
 540
 541         spin_lock(&dev_domain_list_spinlock);
 542         if (find_device(dev)) {
 543                 spin_unlock(&dev_domain_list_spinlock);
 544                 kfree(owner);
 545                 return -EEXIST;
 546         }
 547         owner->domain = domain;
 548         owner->dev = dev;
 549         list_add_tail(&owner->list, &dev_domain_list);
 550         spin_unlock(&dev_domain_list_spinlock);
 551         return 0;
 552 }
 553 EXPORT_SYMBOL_GPL(xen_register_device_domain_owner);
 554
 555 int xen_unregister_device_domain_owner(struct pci_dev *dev)
 556 {
 557         struct xen_device_domain_owner *owner;
 558
 559         spin_lock(&dev_domain_list_spinlock);
 560         owner = find_device(dev);
 561         if (!owner) {
 562                 spin_unlock(&dev_domain_list_spinlock);
 563                 return -ENODEV;
 564         }
 565         list_del(&owner->list);
 566         spin_unlock(&dev_domain_list_spinlock);
 567         kfree(owner);
 568         return 0;
 569 }
 570 EXPORT_SYMBOL_GPL(xen_unregister_device_domain_owner);
 571 #endif