arch/x86/pci/xen.c

   1 /*
   2  * Xen PCI - handle PCI (INTx) and MSI infrastructure calls for PV, HVM and
   3  * initial domain support. We also handle the DSDT _PRT callbacks for GSI's
   4  * used in HVM and initial domain mode (PV does not parse ACPI, so it has no
   5  * concept of GSIs). Under PV we hook under the pnbbios API for IRQs and
   6  * 0xcf8 PCI configuration read/write.
   7  *
   8  *   Author: Ryan Wilson <hap9@epoch.ncsc.mil>
   9  *           Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
  10  *           Stefano Stabellini <stefano.stabellini@eu.citrix.com>
  11  */
  12 #include <linux/module.h>
  13 #include <linux/init.h>
  14 #include <linux/pci.h>
  15 #include <linux/acpi.h>
  16
  17 #include <linux/io.h>
  18 #include <asm/io_apic.h>
  19 #include <asm/pci_x86.h>
  20
  21 #include <asm/xen/hypervisor.h>
  22
  23 #include <xen/features.h>
  24 #include <xen/events.h>
  25 #include <asm/xen/pci.h>
  26
  27 static int xen_pcifront_enable_irq(struct pci_dev *dev)
  28 {
  29         int rc;
  30         int share = 1;
  31         int pirq;
  32         u8 gsi;
  33
  34         rc = pci_read_config_byte(dev, PCI_INTERRUPT_LINE, &gsi);
  35         if (rc < 0) {
  36                 dev_warn(&dev->dev, "Xen PCI: failed to read interrupt line: %d\n",
  37                          rc);
  38                 return rc;
  39         }
  40         /* In PV DomU the Xen PCI backend puts the PIRQ in the interrupt line.*/
  41         pirq = gsi;
  42
  43         if (gsi < NR_IRQS_LEGACY)
  44                 share = 0;
  45
  46         rc = xen_bind_pirq_gsi_to_irq(gsi, pirq, share, "pcifront");
  47         if (rc < 0) {
  48                 dev_warn(&dev->dev, "Xen PCI: failed to bind GSI%d (PIRQ%d) to IRQ: %d\n",
  49                          gsi, pirq, rc);
  50                 return rc;
  51         }
  52
  53         dev->irq = rc;
  54         dev_info(&dev->dev, "Xen PCI mapped GSI%d to IRQ%d\n", gsi, dev->irq);
  55         return 0;
  56 }
  57
  58 #ifdef CONFIG_ACPI
  59 static int xen_register_pirq(u32 gsi, int gsi_override, int triggering,
  60                              bool set_pirq)
  61 {
  62         int rc, pirq = -1, irq = -1;
  63         struct physdev_map_pirq map_irq;
  64         int shareable = 0;
  65         char *name;
  66
  67         irq = xen_irq_from_gsi(gsi);
  68         if (irq > 0)
  69                 return irq;
  70
  71         if (set_pirq)
  72                 pirq = gsi;
  73
  74         map_irq.domid = DOMID_SELF;
  75         map_irq.type = MAP_PIRQ_TYPE_GSI;
  76         map_irq.index = gsi;
  77         map_irq.pirq = pirq;
  78
  79         rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
  80         if (rc) {
  81                 printk(KERN_WARNING "xen map irq failed %d\n", rc);
  82                 return -1;
  83         }
  84
  85         if (triggering == ACPI_EDGE_SENSITIVE) {
  86                 shareable = 0;
  87                 name = "ioapic-edge";
  88         } else {
  89                 shareable = 1;
  90                 name = "ioapic-level";
  91         }
  92
  93         if (gsi_override >= 0)
  94                 gsi = gsi_override;
  95
  96         irq = xen_bind_pirq_gsi_to_irq(gsi, map_irq.pirq, shareable, name);
  97         if (irq < 0)
  98                 goto out;
  99
 100         printk(KERN_DEBUG "xen: --> pirq=%d -> irq=%d (gsi=%d)\n", map_irq.pirq, irq, gsi);
 101 out:
 102         return irq;
 103 }
 104
 105 static int acpi_register_gsi_xen_hvm(struct device *dev, u32 gsi,
 106                                      int trigger, int polarity)
 107 {
 108         if (!xen_hvm_domain())
 109                 return -1;
 110
 111         return xen_register_pirq(gsi, -1 /* no GSI override */, trigger,
 112                                  false /* no mapping of GSI to PIRQ */);
 113 }
 114
 115 #ifdef CONFIG_XEN_DOM0
 116 static int xen_register_gsi(u32 gsi, int gsi_override, int triggering, int polarity)
 117 {
 118         int rc, irq;
 119         struct physdev_setup_gsi setup_gsi;
 120
 121         if (!xen_pv_domain())
 122                 return -1;
 123
 124         printk(KERN_DEBUG "xen: registering gsi %u triggering %d polarity %d\n",
 125                         gsi, triggering, polarity);
 126
 127         irq = xen_register_pirq(gsi, gsi_override, triggering, true);
 128
 129         setup_gsi.gsi = gsi;
 130         setup_gsi.triggering = (triggering == ACPI_EDGE_SENSITIVE ? 0 : 1);
 131         setup_gsi.polarity = (polarity == ACPI_ACTIVE_HIGH ? 0 : 1);
 132
 133         rc = HYPERVISOR_physdev_op(PHYSDEVOP_setup_gsi, &setup_gsi);
 134         if (rc == -EEXIST)
 135                 printk(KERN_INFO "Already setup the GSI :%d\n", gsi);
 136         else if (rc) {
 137                 printk(KERN_ERR "Failed to setup GSI :%d, err_code:%d\n",
 138                                 gsi, rc);
 139         }
 140
 141         return irq;
 142 }
 143
 144 static int acpi_register_gsi_xen(struct device *dev, u32 gsi,
 145                                  int trigger, int polarity)
 146 {
 147         return xen_register_gsi(gsi, -1 /* no GSI override */, trigger, polarity);
 148 }
 149 #endif
 150 #endif
 151
 152 #if defined(CONFIG_PCI_MSI)
 153 #include <linux/msi.h>
 154 #include <asm/msidef.h>
 155
 156 struct xen_pci_frontend_ops *xen_pci_frontend;
 157 EXPORT_SYMBOL_GPL(xen_pci_frontend);
 158
 159 static int xen_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
 160 {
 161         int irq, ret, i;
 162         struct msi_desc *msidesc;
 163         int *v;
 164
 165         if (type == PCI_CAP_ID_MSI && nvec > 1)
 166                 return 1;
 167
 168         v = kzalloc(sizeof(int) * max(1, nvec), GFP_KERNEL);
 169         if (!v)
 170                 return -ENOMEM;
 171
 172         if (type == PCI_CAP_ID_MSIX)
 173                 ret = xen_pci_frontend_enable_msix(dev, v, nvec);
 174         else
 175                 ret = xen_pci_frontend_enable_msi(dev, v);
 176         if (ret)
 177                 goto error;
 178         i = 0;
 179         list_for_each_entry(msidesc, &dev->msi_list, list) {
 180                 irq = xen_bind_pirq_msi_to_irq(dev, msidesc, v[i],
 181                                                (type == PCI_CAP_ID_MSIX) ?
 182                                                "pcifront-msi-x" :
 183                                                "pcifront-msi",
 184                                                 DOMID_SELF);
 185                 if (irq < 0) {
 186                         ret = irq;
 187                         goto free;
 188                 }
 189                 i++;
 190         }
 191         kfree(v);
 192         return 0;
 193
 194 error:
 195         dev_err(&dev->dev, "Xen PCI frontend has not registered MSI/MSI-X support!\n");
 196 free:
 197         kfree(v);
 198         return ret;
 199 }
 200
 201 #define XEN_PIRQ_MSI_DATA  (MSI_DATA_TRIGGER_EDGE | \
 202                 MSI_DATA_LEVEL_ASSERT | (3 << 8) | MSI_DATA_VECTOR(0))
 203
 204 static void xen_msi_compose_msg(struct pci_dev *pdev, unsigned int pirq,
 205                 struct msi_msg *msg)
 206 {
 207         /* We set vector == 0 to tell the hypervisor we don't care about it,
 208          * but we want a pirq setup instead.
 209          * We use the dest_id field to pass the pirq that we want. */
 210         msg->address_hi = MSI_ADDR_BASE_HI | MSI_ADDR_EXT_DEST_ID(pirq);
 211         msg->address_lo =
 212                 MSI_ADDR_BASE_LO |
 213                 MSI_ADDR_DEST_MODE_PHYSICAL |
 214                 MSI_ADDR_REDIRECTION_CPU |
 215                 MSI_ADDR_DEST_ID(pirq);
 216
 217         msg->data = XEN_PIRQ_MSI_DATA;
 218 }
 219
 220 static int xen_hvm_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
 221 {
 222         int irq, pirq;
 223         struct msi_desc *msidesc;
 224         struct msi_msg msg;
 225
 226         if (type == PCI_CAP_ID_MSI && nvec > 1)
 227                 return 1;
 228
 229         list_for_each_entry(msidesc, &dev->msi_list, list) {
 230                 __read_msi_msg(msidesc, &msg);
 231                 pirq = MSI_ADDR_EXT_DEST_ID(msg.address_hi) |
 232                         ((msg.address_lo >> MSI_ADDR_DEST_ID_SHIFT) & 0xff);
 233                 if (msg.data != XEN_PIRQ_MSI_DATA ||
 234                     xen_irq_from_pirq(pirq) < 0) {
 235                         pirq = xen_allocate_pirq_msi(dev, msidesc);
 236                         if (pirq < 0) {
 237                                 irq = -ENODEV;
 238                                 goto error;
 239                         }
 240                         xen_msi_compose_msg(dev, pirq, &msg);
 241                         __write_msi_msg(msidesc, &msg);
 242                         dev_dbg(&dev->dev, "xen: msi bound to pirq=%d\n", pirq);
 243                 } else {
 244                         dev_dbg(&dev->dev,
 245                                 "xen: msi already bound to pirq=%d\n", pirq);
 246                 }
 247                 irq = xen_bind_pirq_msi_to_irq(dev, msidesc, pirq,
 248                                                (type == PCI_CAP_ID_MSIX) ?
 249                                                "msi-x" : "msi",
 250                                                DOMID_SELF);
 251                 if (irq < 0)
 252                         goto error;
 253                 dev_dbg(&dev->dev,
 254                         "xen: msi --> pirq=%d --> irq=%d\n", pirq, irq);
 255         }
 256         return 0;
 257
 258 error:
 259         dev_err(&dev->dev,
 260                 "Xen PCI frontend has not registered MSI/MSI-X support!\n");
 261         return irq;
 262 }
 263
 264 #ifdef CONFIG_XEN_DOM0
 265 static bool __read_mostly pci_seg_supported = true;
 266
 267 static int xen_initdom_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
 268 {
 269         int ret = 0;
 270         struct msi_desc *msidesc;
 271
 272         if (type == PCI_CAP_ID_MSI && nvec > 1)
 273                 return 1;
 274
 275         list_for_each_entry(msidesc, &dev->msi_list, list) {
 276                 struct physdev_map_pirq map_irq;
 277                 domid_t domid;
 278
 279                 domid = ret = xen_find_device_domain_owner(dev);
 280                 /* N.B. Casting int's -ENODEV to uint16_t results in 0xFFED,
 281                  * hence check ret value for < 0. */
 282                 if (ret < 0)
 283                         domid = DOMID_SELF;
 284
 285                 memset(&map_irq, 0, sizeof(map_irq));
 286                 map_irq.domid = domid;
 287                 map_irq.type = MAP_PIRQ_TYPE_MSI_SEG;
 288                 map_irq.index = -1;
 289                 map_irq.pirq = -1;
 290                 map_irq.bus = dev->bus->number |
 291                               (pci_domain_nr(dev->bus) << 16);
 292                 map_irq.devfn = dev->devfn;
 293
 294                 if (type == PCI_CAP_ID_MSIX) {
 295                         int pos;
 296                         u32 table_offset, bir;
 297
 298                         pos = dev->msix_cap;
 299                         pci_read_config_dword(dev, pos + PCI_MSIX_TABLE,
 300                                               &table_offset);
 301                         bir = (u8)(table_offset & PCI_MSIX_TABLE_BIR);
 302
 303                         map_irq.table_base = pci_resource_start(dev, bir);
 304                         map_irq.entry_nr = msidesc->msi_attrib.entry_nr;
 305                 }
 306
 307                 ret = -EINVAL;
 308                 if (pci_seg_supported)
 309                         ret = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq,
 310                                                     &map_irq);
 311                 if (ret == -EINVAL && !pci_domain_nr(dev->bus)) {
 312                         map_irq.type = MAP_PIRQ_TYPE_MSI;
 313                         map_irq.index = -1;
 314                         map_irq.pirq = -1;
 315                         map_irq.bus = dev->bus->number;
 316                         ret = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq,
 317                                                     &map_irq);
 318                         if (ret != -EINVAL)
 319                                 pci_seg_supported = false;
 320                 }
 321                 if (ret) {
 322                         dev_warn(&dev->dev, "xen map irq failed %d for %d domain\n",
 323                                  ret, domid);
 324                         goto out;
 325                 }
 326
 327                 ret = xen_bind_pirq_msi_to_irq(dev, msidesc,
 328                                                map_irq.pirq,
 329                                                (type == PCI_CAP_ID_MSIX) ?
 330                                                "msi-x" : "msi",
 331                                                 domid);
 332                 if (ret < 0)
 333                         goto out;
 334         }
 335         ret = 0;
 336 out:
 337         return ret;
 338 }
 339
 340 static void xen_initdom_restore_msi_irqs(struct pci_dev *dev, int irq)
 341 {
 342         int ret = 0;
 343
 344         if (pci_seg_supported) {
 345                 struct physdev_pci_device restore_ext;
 346
 347                 restore_ext.seg = pci_domain_nr(dev->bus);
 348                 restore_ext.bus = dev->bus->number;
 349                 restore_ext.devfn = dev->devfn;
 350                 ret = HYPERVISOR_physdev_op(PHYSDEVOP_restore_msi_ext,
 351                                         &restore_ext);
 352                 if (ret == -ENOSYS)
 353                         pci_seg_supported = false;
 354                 WARN(ret && ret != -ENOSYS, "restore_msi_ext -> %d\n", ret);
 355         }
 356         if (!pci_seg_supported) {
 357                 struct physdev_restore_msi restore;
 358
 359                 restore.bus = dev->bus->number;
 360                 restore.devfn = dev->devfn;
 361                 ret = HYPERVISOR_physdev_op(PHYSDEVOP_restore_msi, &restore);
 362                 WARN(ret && ret != -ENOSYS, "restore_msi -> %d\n", ret);
 363         }
 364 }
 365 #endif
 366
 367 static void xen_teardown_msi_irqs(struct pci_dev *dev)
 368 {
 369         struct msi_desc *msidesc;
 370
 371         msidesc = list_entry(dev->msi_list.next, struct msi_desc, list);
 372         if (msidesc->msi_attrib.is_msix)
 373                 xen_pci_frontend_disable_msix(dev);
 374         else
 375                 xen_pci_frontend_disable_msi(dev);
 376
 377         /* Free the IRQ's and the msidesc using the generic code. */
 378         default_teardown_msi_irqs(dev);
 379 }
 380
 381 static void xen_teardown_msi_irq(unsigned int irq)
 382 {
 383         xen_destroy_irq(irq);
 384 }
 385
 386 #endif
 387
 388 int __init pci_xen_init(void)
 389 {
 390         if (!xen_pv_domain() || xen_initial_domain())
 391                 return -ENODEV;
 392
 393         printk(KERN_INFO "PCI: setting up Xen PCI frontend stub\n");
 394
 395         pcibios_set_cache_line_size();
 396
 397         pcibios_enable_irq = xen_pcifront_enable_irq;
 398         pcibios_disable_irq = NULL;
 399
 400 #ifdef CONFIG_ACPI
 401         /* Keep ACPI out of the picture */
 402         acpi_noirq = 1;
 403 #endif
 404
 405 #ifdef CONFIG_PCI_MSI
 406         x86_msi.setup_msi_irqs = xen_setup_msi_irqs;
 407         x86_msi.teardown_msi_irq = xen_teardown_msi_irq;
 408         x86_msi.teardown_msi_irqs = xen_teardown_msi_irqs;
 409 #endif
 410         return 0;
 411 }
 412
 413 int __init pci_xen_hvm_init(void)
 414 {
 415         if (!xen_have_vector_callback || !xen_feature(XENFEAT_hvm_pirqs))
 416                 return 0;
 417
 418 #ifdef CONFIG_ACPI
 419         /*
 420          * We don't want to change the actual ACPI delivery model,
 421          * just how GSIs get registered.
 422          */
 423         __acpi_register_gsi = acpi_register_gsi_xen_hvm;
 424 #endif
 425
 426 #ifdef CONFIG_PCI_MSI
 427         x86_msi.setup_msi_irqs = xen_hvm_setup_msi_irqs;
 428         x86_msi.teardown_msi_irq = xen_teardown_msi_irq;
 429 #endif
 430         return 0;
 431 }
 432
 433 #ifdef CONFIG_XEN_DOM0
 434 static __init void xen_setup_acpi_sci(void)
 435 {
 436         int rc;
 437         int trigger, polarity;
 438         int gsi = acpi_sci_override_gsi;
 439         int irq = -1;
 440         int gsi_override = -1;
 441
 442         if (!gsi)
 443                 return;
 444
 445         rc = acpi_get_override_irq(gsi, &trigger, &polarity);
 446         if (rc) {
 447                 printk(KERN_WARNING "xen: acpi_get_override_irq failed for acpi"
 448                                 " sci, rc=%d\n", rc);
 449                 return;
 450         }
 451         trigger = trigger ? ACPI_LEVEL_SENSITIVE : ACPI_EDGE_SENSITIVE;
 452         polarity = polarity ? ACPI_ACTIVE_LOW : ACPI_ACTIVE_HIGH;
 453
 454         printk(KERN_INFO "xen: sci override: global_irq=%d trigger=%d "
 455                         "polarity=%d\n", gsi, trigger, polarity);
 456
 457         /* Before we bind the GSI to a Linux IRQ, check whether
 458          * we need to override it with bus_irq (IRQ) value. Usually for
 459          * IRQs below IRQ_LEGACY_IRQ this holds IRQ == GSI, as so:
 460          *  ACPI: INT_SRC_OVR (bus 0 bus_irq 9 global_irq 9 low level)
 461          * but there are oddballs where the IRQ != GSI:
 462          *  ACPI: INT_SRC_OVR (bus 0 bus_irq 9 global_irq 20 low level)
 463          * which ends up being: gsi_to_irq[9] == 20
 464          * (which is what acpi_gsi_to_irq ends up calling when starting the
 465          * the ACPI interpreter and keels over since IRQ 9 has not been
 466          * setup as we had setup IRQ 20 for it).
 467          */
 468         if (acpi_gsi_to_irq(gsi, &irq) == 0) {
 469                 /* Use the provided value if it's valid. */
 470                 if (irq >= 0)
 471                         gsi_override = irq;
 472         }
 473
 474         gsi = xen_register_gsi(gsi, gsi_override, trigger, polarity);
 475         printk(KERN_INFO "xen: acpi sci %d\n", gsi);
 476
 477         return;
 478 }
 479
 480 int __init pci_xen_initial_domain(void)
 481 {
 482         int irq;
 483
 484 #ifdef CONFIG_PCI_MSI
 485         x86_msi.setup_msi_irqs = xen_initdom_setup_msi_irqs;
 486         x86_msi.teardown_msi_irq = xen_teardown_msi_irq;
 487         x86_msi.restore_msi_irqs = xen_initdom_restore_msi_irqs;
 488 #endif
 489         xen_setup_acpi_sci();
 490         __acpi_register_gsi = acpi_register_gsi_xen;
 491         /* Pre-allocate legacy irqs */
 492         for (irq = 0; irq < NR_IRQS_LEGACY; irq++) {
 493                 int trigger, polarity;
 494
 495                 if (acpi_get_override_irq(irq, &trigger, &polarity) == -1)
 496                         continue;
 497
 498                 xen_register_pirq(irq, -1 /* no GSI override */,
 499                         trigger ? ACPI_LEVEL_SENSITIVE : ACPI_EDGE_SENSITIVE,
 500                         true /* Map GSI to PIRQ */);
 501         }
 502         if (0 == nr_ioapics) {
 503                 for (irq = 0; irq < NR_IRQS_LEGACY; irq++)
 504                         xen_bind_pirq_gsi_to_irq(irq, irq, 0, "xt-pic");
 505         }
 506         return 0;
 507 }
 508
 509 struct xen_device_domain_owner {
 510         domid_t domain;
 511         struct pci_dev *dev;
 512         struct list_head list;
 513 };
 514
 515 static DEFINE_SPINLOCK(dev_domain_list_spinlock);
 516 static struct list_head dev_domain_list = LIST_HEAD_INIT(dev_domain_list);
 517
 518 static struct xen_device_domain_owner *find_device(struct pci_dev *dev)
 519 {
 520         struct xen_device_domain_owner *owner;
 521
 522         list_for_each_entry(owner, &dev_domain_list, list) {
 523                 if (owner->dev == dev)
 524                         return owner;
 525         }
 526         return NULL;
 527 }
 528
 529 int xen_find_device_domain_owner(struct pci_dev *dev)
 530 {
 531         struct xen_device_domain_owner *owner;
 532         int domain = -ENODEV;
 533
 534         spin_lock(&dev_domain_list_spinlock);
 535         owner = find_device(dev);
 536         if (owner)
 537                 domain = owner->domain;
 538         spin_unlock(&dev_domain_list_spinlock);
 539         return domain;
 540 }
 541 EXPORT_SYMBOL_GPL(xen_find_device_domain_owner);
 542
 543 int xen_register_device_domain_owner(struct pci_dev *dev, uint16_t domain)
 544 {
 545         struct xen_device_domain_owner *owner;
 546
 547         owner = kzalloc(sizeof(struct xen_device_domain_owner), GFP_KERNEL);
 548         if (!owner)
 549                 return -ENODEV;
 550
 551         spin_lock(&dev_domain_list_spinlock);
 552         if (find_device(dev)) {
 553                 spin_unlock(&dev_domain_list_spinlock);
 554                 kfree(owner);
 555                 return -EEXIST;
 556         }
 557         owner->domain = domain;
 558         owner->dev = dev;
 559         list_add_tail(&owner->list, &dev_domain_list);
 560         spin_unlock(&dev_domain_list_spinlock);
 561         return 0;
 562 }
 563 EXPORT_SYMBOL_GPL(xen_register_device_domain_owner);
 564
 565 int xen_unregister_device_domain_owner(struct pci_dev *dev)
 566 {
 567         struct xen_device_domain_owner *owner;
 568
 569         spin_lock(&dev_domain_list_spinlock);
 570         owner = find_device(dev);
 571         if (!owner) {
 572                 spin_unlock(&dev_domain_list_spinlock);
 573                 return -ENODEV;
 574         }
 575         list_del(&owner->list);
 576         spin_unlock(&dev_domain_list_spinlock);
 577         kfree(owner);
 578         return 0;
 579 }
 580 EXPORT_SYMBOL_GPL(xen_unregister_device_domain_owner);
 581 #endif