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