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