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