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 #include <asm/i8259.h>
  27
  28 static int xen_pcifront_enable_irq(struct pci_dev *dev)
  29 {
  30         int rc;
  31         int share = 1;
  32         int pirq;
  33         u8 gsi;
  34
  35         rc = pci_read_config_byte(dev, PCI_INTERRUPT_LINE, &gsi);
  36         if (rc < 0) {
  37                 dev_warn(&dev->dev, "Xen PCI: failed to read interrupt line: %d\n",
  38                          rc);
  39                 return rc;
  40         }
  41         /* In PV DomU the Xen PCI backend puts the PIRQ in the interrupt line.*/
  42         pirq = gsi;
  43
  44         if (gsi < nr_legacy_irqs())
  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 xen_register_pirq(u32 gsi, int gsi_override, int triggering,
  61                              bool set_pirq)
  62 {
  63         int rc, pirq = -1, irq = -1;
  64         struct physdev_map_pirq map_irq;
  65         int shareable = 0;
  66         char *name;
  67
  68         irq = xen_irq_from_gsi(gsi);
  69         if (irq > 0)
  70                 return irq;
  71
  72         if (set_pirq)
  73                 pirq = gsi;
  74
  75         map_irq.domid = DOMID_SELF;
  76         map_irq.type = MAP_PIRQ_TYPE_GSI;
  77         map_irq.index = gsi;
  78         map_irq.pirq = pirq;
  79
  80         rc = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, &map_irq);
  81         if (rc) {
  82                 printk(KERN_WARNING "xen map irq failed %d\n", rc);
  83                 return -1;
  84         }
  85
  86         if (triggering == ACPI_EDGE_SENSITIVE) {
  87                 shareable = 0;
  88                 name = "ioapic-edge";
  89         } else {
  90                 shareable = 1;
  91                 name = "ioapic-level";
  92         }
  93
  94         if (gsi_override >= 0)
  95                 gsi = gsi_override;
  96
  97         irq = xen_bind_pirq_gsi_to_irq(gsi, map_irq.pirq, shareable, name);
  98         if (irq < 0)
  99                 goto out;
 100
 101         printk(KERN_DEBUG "xen: --> pirq=%d -> irq=%d (gsi=%d)\n", map_irq.pirq, irq, gsi);
 102 out:
 103         return irq;
 104 }
 105
 106 static int acpi_register_gsi_xen_hvm(struct device *dev, u32 gsi,
 107                                      int trigger, int polarity)
 108 {
 109         if (!xen_hvm_domain())
 110                 return -1;
 111
 112         return xen_register_pirq(gsi, -1 /* no GSI override */, trigger,
 113                                  false /* no mapping of GSI to PIRQ */);
 114 }
 115
 116 #ifdef CONFIG_XEN_DOM0
 117 static int xen_register_gsi(u32 gsi, int gsi_override, int triggering, int polarity)
 118 {
 119         int rc, irq;
 120         struct physdev_setup_gsi setup_gsi;
 121
 122         if (!xen_pv_domain())
 123                 return -1;
 124
 125         printk(KERN_DEBUG "xen: registering gsi %u triggering %d polarity %d\n",
 126                         gsi, triggering, polarity);
 127
 128         irq = xen_register_pirq(gsi, gsi_override, triggering, true);
 129
 130         setup_gsi.gsi = gsi;
 131         setup_gsi.triggering = (triggering == ACPI_EDGE_SENSITIVE ? 0 : 1);
 132         setup_gsi.polarity = (polarity == ACPI_ACTIVE_HIGH ? 0 : 1);
 133
 134         rc = HYPERVISOR_physdev_op(PHYSDEVOP_setup_gsi, &setup_gsi);
 135         if (rc == -EEXIST)
 136                 printk(KERN_INFO "Already setup the GSI :%d\n", gsi);
 137         else if (rc) {
 138                 printk(KERN_ERR "Failed to setup GSI :%d, err_code:%d\n",
 139                                 gsi, rc);
 140         }
 141
 142         return irq;
 143 }
 144
 145 static int acpi_register_gsi_xen(struct device *dev, u32 gsi,
 146                                  int trigger, int polarity)
 147 {
 148         return xen_register_gsi(gsi, -1 /* no GSI override */, trigger, polarity);
 149 }
 150 #endif
 151 #endif
 152
 153 #if defined(CONFIG_PCI_MSI)
 154 #include <linux/msi.h>
 155 #include <asm/msidef.h>
 156
 157 struct xen_pci_frontend_ops *xen_pci_frontend;
 158 EXPORT_SYMBOL_GPL(xen_pci_frontend);
 159
 160 static int xen_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
 161 {
 162         int irq, ret, i;
 163         struct msi_desc *msidesc;
 164         int *v;
 165
 166         if (type == PCI_CAP_ID_MSI && nvec > 1)
 167                 return 1;
 168
 169         v = kzalloc(sizeof(int) * max(1, nvec), GFP_KERNEL);
 170         if (!v)
 171                 return -ENOMEM;
 172
 173         if (type == PCI_CAP_ID_MSIX)
 174                 ret = xen_pci_frontend_enable_msix(dev, v, nvec);
 175         else
 176                 ret = xen_pci_frontend_enable_msi(dev, v);
 177         if (ret)
 178                 goto error;
 179         i = 0;
 180         list_for_each_entry(msidesc, &dev->msi_list, list) {
 181                 irq = xen_bind_pirq_msi_to_irq(dev, msidesc, v[i],
 182                                                (type == PCI_CAP_ID_MSI) ? nvec : 1,
 183                                                (type == PCI_CAP_ID_MSIX) ?
 184                                                "pcifront-msi-x" :
 185                                                "pcifront-msi",
 186                                                 DOMID_SELF);
 187                 if (irq < 0) {
 188                         ret = irq;
 189                         goto free;
 190                 }
 191                 i++;
 192         }
 193         kfree(v);
 194         return 0;
 195
 196 error:
 197         dev_err(&dev->dev, "Xen PCI frontend has not registered MSI/MSI-X support!\n");
 198 free:
 199         kfree(v);
 200         return ret;
 201 }
 202
 203 #define XEN_PIRQ_MSI_DATA  (MSI_DATA_TRIGGER_EDGE | \
 204                 MSI_DATA_LEVEL_ASSERT | (3 << 8) | MSI_DATA_VECTOR(0))
 205
 206 static void xen_msi_compose_msg(struct pci_dev *pdev, unsigned int pirq,
 207                 struct msi_msg *msg)
 208 {
 209         /* We set vector == 0 to tell the hypervisor we don't care about it,
 210          * but we want a pirq setup instead.
 211          * We use the dest_id field to pass the pirq that we want. */
 212         msg->address_hi = MSI_ADDR_BASE_HI | MSI_ADDR_EXT_DEST_ID(pirq);
 213         msg->address_lo =
 214                 MSI_ADDR_BASE_LO |
 215                 MSI_ADDR_DEST_MODE_PHYSICAL |
 216                 MSI_ADDR_REDIRECTION_CPU |
 217                 MSI_ADDR_DEST_ID(pirq);
 218
 219         msg->data = XEN_PIRQ_MSI_DATA;
 220 }
 221
 222 static int xen_hvm_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
 223 {
 224         int irq, pirq;
 225         struct msi_desc *msidesc;
 226         struct msi_msg msg;
 227
 228         if (type == PCI_CAP_ID_MSI && nvec > 1)
 229                 return 1;
 230
 231         list_for_each_entry(msidesc, &dev->msi_list, list) {
 232                 __pci_read_msi_msg(msidesc, &msg);
 233                 pirq = MSI_ADDR_EXT_DEST_ID(msg.address_hi) |
 234                         ((msg.address_lo >> MSI_ADDR_DEST_ID_SHIFT) & 0xff);
 235                 if (msg.data != XEN_PIRQ_MSI_DATA ||
 236                     xen_irq_from_pirq(pirq) < 0) {
 237                         pirq = xen_allocate_pirq_msi(dev, msidesc);
 238                         if (pirq < 0) {
 239                                 irq = -ENODEV;
 240                                 goto error;
 241                         }
 242                         xen_msi_compose_msg(dev, pirq, &msg);
 243                         __write_msi_msg(msidesc, &msg);
 244                         dev_dbg(&dev->dev, "xen: msi bound to pirq=%d\n", pirq);
 245                 } else {
 246                         dev_dbg(&dev->dev,
 247                                 "xen: msi already bound to pirq=%d\n", pirq);
 248                 }
 249                 irq = xen_bind_pirq_msi_to_irq(dev, msidesc, pirq,
 250                                                (type == PCI_CAP_ID_MSI) ? nvec : 1,
 251                                                (type == PCI_CAP_ID_MSIX) ?
 252                                                "msi-x" : "msi",
 253                                                DOMID_SELF);
 254                 if (irq < 0)
 255                         goto error;
 256                 dev_dbg(&dev->dev,
 257                         "xen: msi --> pirq=%d --> irq=%d\n", pirq, irq);
 258         }
 259         return 0;
 260
 261 error:
 262         dev_err(&dev->dev,
 263                 "Xen PCI frontend has not registered MSI/MSI-X support!\n");
 264         return irq;
 265 }
 266
 267 #ifdef CONFIG_XEN_DOM0
 268 static bool __read_mostly pci_seg_supported = true;
 269
 270 static int xen_initdom_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
 271 {
 272         int ret = 0;
 273         struct msi_desc *msidesc;
 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_MSI && nvec > 1) {
 295                         map_irq.type = MAP_PIRQ_TYPE_MULTI_MSI;
 296                         map_irq.entry_nr = nvec;
 297                 } else if (type == PCI_CAP_ID_MSIX) {
 298                         int pos;
 299                         u32 table_offset, bir;
 300
 301                         pos = dev->msix_cap;
 302                         pci_read_config_dword(dev, pos + PCI_MSIX_TABLE,
 303                                               &table_offset);
 304                         bir = (u8)(table_offset & PCI_MSIX_TABLE_BIR);
 305
 306                         map_irq.table_base = pci_resource_start(dev, bir);
 307                         map_irq.entry_nr = msidesc->msi_attrib.entry_nr;
 308                 }
 309
 310                 ret = -EINVAL;
 311                 if (pci_seg_supported)
 312                         ret = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq,
 313                                                     &map_irq);
 314                 if (type == PCI_CAP_ID_MSI && nvec > 1 && ret) {
 315                         /*
 316                          * If MAP_PIRQ_TYPE_MULTI_MSI is not available
 317                          * there's nothing else we can do in this case.
 318                          * Just set ret > 0 so driver can retry with
 319                          * single MSI.
 320                          */
 321                         ret = 1;
 322                         goto out;
 323                 }
 324                 if (ret == -EINVAL && !pci_domain_nr(dev->bus)) {
 325                         map_irq.type = MAP_PIRQ_TYPE_MSI;
 326                         map_irq.index = -1;
 327                         map_irq.pirq = -1;
 328                         map_irq.bus = dev->bus->number;
 329                         ret = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq,
 330                                                     &map_irq);
 331                         if (ret != -EINVAL)
 332                                 pci_seg_supported = false;
 333                 }
 334                 if (ret) {
 335                         dev_warn(&dev->dev, "xen map irq failed %d for %d domain\n",
 336                                  ret, domid);
 337                         goto out;
 338                 }
 339
 340                 ret = xen_bind_pirq_msi_to_irq(dev, msidesc, map_irq.pirq,
 341                                                (type == PCI_CAP_ID_MSI) ? nvec : 1,
 342                                                (type == PCI_CAP_ID_MSIX) ? "msi-x" : "msi",
 343                                                domid);
 344                 if (ret < 0)
 345                         goto out;
 346         }
 347         ret = 0;
 348 out:
 349         return ret;
 350 }
 351
 352 static void xen_initdom_restore_msi_irqs(struct pci_dev *dev)
 353 {
 354         int ret = 0;
 355
 356         if (pci_seg_supported) {
 357                 struct physdev_pci_device restore_ext;
 358
 359                 restore_ext.seg = pci_domain_nr(dev->bus);
 360                 restore_ext.bus = dev->bus->number;
 361                 restore_ext.devfn = dev->devfn;
 362                 ret = HYPERVISOR_physdev_op(PHYSDEVOP_restore_msi_ext,
 363                                         &restore_ext);
 364                 if (ret == -ENOSYS)
 365                         pci_seg_supported = false;
 366                 WARN(ret && ret != -ENOSYS, "restore_msi_ext -> %d\n", ret);
 367         }
 368         if (!pci_seg_supported) {
 369                 struct physdev_restore_msi restore;
 370
 371                 restore.bus = dev->bus->number;
 372                 restore.devfn = dev->devfn;
 373                 ret = HYPERVISOR_physdev_op(PHYSDEVOP_restore_msi, &restore);
 374                 WARN(ret && ret != -ENOSYS, "restore_msi -> %d\n", ret);
 375         }
 376 }
 377 #endif
 378
 379 static void xen_teardown_msi_irqs(struct pci_dev *dev)
 380 {
 381         struct msi_desc *msidesc;
 382
 383         msidesc = list_entry(dev->msi_list.next, struct msi_desc, list);
 384         if (msidesc->msi_attrib.is_msix)
 385                 xen_pci_frontend_disable_msix(dev);
 386         else
 387                 xen_pci_frontend_disable_msi(dev);
 388
 389         /* Free the IRQ's and the msidesc using the generic code. */
 390         default_teardown_msi_irqs(dev);
 391 }
 392
 393 static void xen_teardown_msi_irq(unsigned int irq)
 394 {
 395         xen_destroy_irq(irq);
 396 }
 397
 398 #endif
 399
 400 int __init pci_xen_init(void)
 401 {
 402         if (!xen_pv_domain() || xen_initial_domain())
 403                 return -ENODEV;
 404
 405         printk(KERN_INFO "PCI: setting up Xen PCI frontend stub\n");
 406
 407         pcibios_set_cache_line_size();
 408
 409         pcibios_enable_irq = xen_pcifront_enable_irq;
 410         pcibios_disable_irq = NULL;
 411
 412 #ifdef CONFIG_ACPI
 413         /* Keep ACPI out of the picture */
 414         acpi_noirq = 1;
 415 #endif
 416
 417 #ifdef CONFIG_PCI_MSI
 418         x86_msi.setup_msi_irqs = xen_setup_msi_irqs;
 419         x86_msi.teardown_msi_irq = xen_teardown_msi_irq;
 420         x86_msi.teardown_msi_irqs = xen_teardown_msi_irqs;
 421         pci_msi_ignore_mask = 1;
 422 #endif
 423         return 0;
 424 }
 425
 426 int __init pci_xen_hvm_init(void)
 427 {
 428         if (!xen_have_vector_callback || !xen_feature(XENFEAT_hvm_pirqs))
 429                 return 0;
 430
 431 #ifdef CONFIG_ACPI
 432         /*
 433          * We don't want to change the actual ACPI delivery model,
 434          * just how GSIs get registered.
 435          */
 436         __acpi_register_gsi = acpi_register_gsi_xen_hvm;
 437 #endif
 438
 439 #ifdef CONFIG_PCI_MSI
 440         x86_msi.setup_msi_irqs = xen_hvm_setup_msi_irqs;
 441         x86_msi.teardown_msi_irq = xen_teardown_msi_irq;
 442 #endif
 443         return 0;
 444 }
 445
 446 #ifdef CONFIG_XEN_DOM0
 447 static __init void xen_setup_acpi_sci(void)
 448 {
 449         int rc;
 450         int trigger, polarity;
 451         int gsi = acpi_sci_override_gsi;
 452         int irq = -1;
 453         int gsi_override = -1;
 454
 455         if (!gsi)
 456                 return;
 457
 458         rc = acpi_get_override_irq(gsi, &trigger, &polarity);
 459         if (rc) {
 460                 printk(KERN_WARNING "xen: acpi_get_override_irq failed for acpi"
 461                                 " sci, rc=%d\n", rc);
 462                 return;
 463         }
 464         trigger = trigger ? ACPI_LEVEL_SENSITIVE : ACPI_EDGE_SENSITIVE;
 465         polarity = polarity ? ACPI_ACTIVE_LOW : ACPI_ACTIVE_HIGH;
 466
 467         printk(KERN_INFO "xen: sci override: global_irq=%d trigger=%d "
 468                         "polarity=%d\n", gsi, trigger, polarity);
 469
 470         /* Before we bind the GSI to a Linux IRQ, check whether
 471          * we need to override it with bus_irq (IRQ) value. Usually for
 472          * IRQs below IRQ_LEGACY_IRQ this holds IRQ == GSI, as so:
 473          *  ACPI: INT_SRC_OVR (bus 0 bus_irq 9 global_irq 9 low level)
 474          * but there are oddballs where the IRQ != GSI:
 475          *  ACPI: INT_SRC_OVR (bus 0 bus_irq 9 global_irq 20 low level)
 476          * which ends up being: gsi_to_irq[9] == 20
 477          * (which is what acpi_gsi_to_irq ends up calling when starting the
 478          * the ACPI interpreter and keels over since IRQ 9 has not been
 479          * setup as we had setup IRQ 20 for it).
 480          */
 481         if (acpi_gsi_to_irq(gsi, &irq) == 0) {
 482                 /* Use the provided value if it's valid. */
 483                 if (irq >= 0)
 484                         gsi_override = irq;
 485         }
 486
 487         gsi = xen_register_gsi(gsi, gsi_override, trigger, polarity);
 488         printk(KERN_INFO "xen: acpi sci %d\n", gsi);
 489
 490         return;
 491 }
 492
 493 int __init pci_xen_initial_domain(void)
 494 {
 495         int irq;
 496
 497 #ifdef CONFIG_PCI_MSI
 498         x86_msi.setup_msi_irqs = xen_initdom_setup_msi_irqs;
 499         x86_msi.teardown_msi_irq = xen_teardown_msi_irq;
 500         x86_msi.restore_msi_irqs = xen_initdom_restore_msi_irqs;
 501         pci_msi_ignore_mask = 1;
 502 #endif
 503         xen_setup_acpi_sci();
 504         __acpi_register_gsi = acpi_register_gsi_xen;
 505         /* Pre-allocate legacy irqs */
 506         for (irq = 0; irq < nr_legacy_irqs(); irq++) {
 507                 int trigger, polarity;
 508
 509                 if (acpi_get_override_irq(irq, &trigger, &polarity) == -1)
 510                         continue;
 511
 512                 xen_register_pirq(irq, -1 /* no GSI override */,
 513                         trigger ? ACPI_LEVEL_SENSITIVE : ACPI_EDGE_SENSITIVE,
 514                         true /* Map GSI to PIRQ */);
 515         }
 516         if (0 == nr_ioapics) {
 517                 for (irq = 0; irq < nr_legacy_irqs(); irq++)
 518                         xen_bind_pirq_gsi_to_irq(irq, irq, 0, "xt-pic");
 519         }
 520         return 0;
 521 }
 522
 523 struct xen_device_domain_owner {
 524         domid_t domain;
 525         struct pci_dev *dev;
 526         struct list_head list;
 527 };
 528
 529 static DEFINE_SPINLOCK(dev_domain_list_spinlock);
 530 static struct list_head dev_domain_list = LIST_HEAD_INIT(dev_domain_list);
 531
 532 static struct xen_device_domain_owner *find_device(struct pci_dev *dev)
 533 {
 534         struct xen_device_domain_owner *owner;
 535
 536         list_for_each_entry(owner, &dev_domain_list, list) {
 537                 if (owner->dev == dev)
 538                         return owner;
 539         }
 540         return NULL;
 541 }
 542
 543 int xen_find_device_domain_owner(struct pci_dev *dev)
 544 {
 545         struct xen_device_domain_owner *owner;
 546         int domain = -ENODEV;
 547
 548         spin_lock(&dev_domain_list_spinlock);
 549         owner = find_device(dev);
 550         if (owner)
 551                 domain = owner->domain;
 552         spin_unlock(&dev_domain_list_spinlock);
 553         return domain;
 554 }
 555 EXPORT_SYMBOL_GPL(xen_find_device_domain_owner);
 556
 557 int xen_register_device_domain_owner(struct pci_dev *dev, uint16_t domain)
 558 {
 559         struct xen_device_domain_owner *owner;
 560
 561         owner = kzalloc(sizeof(struct xen_device_domain_owner), GFP_KERNEL);
 562         if (!owner)
 563                 return -ENODEV;
 564
 565         spin_lock(&dev_domain_list_spinlock);
 566         if (find_device(dev)) {
 567                 spin_unlock(&dev_domain_list_spinlock);
 568                 kfree(owner);
 569                 return -EEXIST;
 570         }
 571         owner->domain = domain;
 572         owner->dev = dev;
 573         list_add_tail(&owner->list, &dev_domain_list);
 574         spin_unlock(&dev_domain_list_spinlock);
 575         return 0;
 576 }
 577 EXPORT_SYMBOL_GPL(xen_register_device_domain_owner);
 578
 579 int xen_unregister_device_domain_owner(struct pci_dev *dev)
 580 {
 581         struct xen_device_domain_owner *owner;
 582
 583         spin_lock(&dev_domain_list_spinlock);
 584         owner = find_device(dev);
 585         if (!owner) {
 586                 spin_unlock(&dev_domain_list_spinlock);
 587                 return -ENODEV;
 588         }
 589         list_del(&owner->list);
 590         spin_unlock(&dev_domain_list_spinlock);
 591         kfree(owner);
 592         return 0;
 593 }
 594 EXPORT_SYMBOL_GPL(xen_unregister_device_domain_owner);
 595 #endif