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 27static 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 59static 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); 101out: 102 return irq; 103} 104 105static 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 116static 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 144static 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 156struct xen_pci_frontend_ops *xen_pci_frontend; 157EXPORT_SYMBOL_GPL(xen_pci_frontend); 158 159static 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 v = kzalloc(sizeof(int) * max(1, nvec), GFP_KERNEL); 166 if (!v) 167 return -ENOMEM; 168 169 if (type == PCI_CAP_ID_MSIX) 170 ret = xen_pci_frontend_enable_msix(dev, v, nvec); 171 else 172 ret = xen_pci_frontend_enable_msi(dev, v); 173 if (ret) 174 goto error; 175 i = 0; 176 list_for_each_entry(msidesc, &dev->msi_list, list) { 177 irq = xen_bind_pirq_msi_to_irq(dev, msidesc, v[i], 0, 178 (type == PCI_CAP_ID_MSIX) ? 179 "pcifront-msi-x" : 180 "pcifront-msi", 181 DOMID_SELF); 182 if (irq < 0) { 183 ret = irq; 184 goto free; 185 } 186 i++; 187 } 188 kfree(v); 189 return 0; 190 191error: 192 dev_err(&dev->dev, "Xen PCI frontend has not registered MSI/MSI-X support!\n"); 193free: 194 kfree(v); 195 return ret; 196} 197 198#define XEN_PIRQ_MSI_DATA (MSI_DATA_TRIGGER_EDGE | \ 199 MSI_DATA_LEVEL_ASSERT | (3 << 8) | MSI_DATA_VECTOR(0)) 200 201static void xen_msi_compose_msg(struct pci_dev *pdev, unsigned int pirq, 202 struct msi_msg *msg) 203{ 204 /* We set vector == 0 to tell the hypervisor we don't care about it, 205 * but we want a pirq setup instead. 206 * We use the dest_id field to pass the pirq that we want. */ 207 msg->address_hi = MSI_ADDR_BASE_HI | MSI_ADDR_EXT_DEST_ID(pirq); 208 msg->address_lo = 209 MSI_ADDR_BASE_LO | 210 MSI_ADDR_DEST_MODE_PHYSICAL | 211 MSI_ADDR_REDIRECTION_CPU | 212 MSI_ADDR_DEST_ID(pirq); 213 214 msg->data = XEN_PIRQ_MSI_DATA; 215} 216 217static int xen_hvm_setup_msi_irqs(struct pci_dev *dev, int nvec, int type) 218{ 219 int irq, pirq; 220 struct msi_desc *msidesc; 221 struct msi_msg msg; 222 223 list_for_each_entry(msidesc, &dev->msi_list, list) { 224 __read_msi_msg(msidesc, &msg); 225 pirq = MSI_ADDR_EXT_DEST_ID(msg.address_hi) | 226 ((msg.address_lo >> MSI_ADDR_DEST_ID_SHIFT) & 0xff); 227 if (msg.data != XEN_PIRQ_MSI_DATA || 228 xen_irq_from_pirq(pirq) < 0) { 229 pirq = xen_allocate_pirq_msi(dev, msidesc); 230 if (pirq < 0) { 231 irq = -ENODEV; 232 goto error; 233 } 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 252error: 253 dev_err(&dev->dev, 254 "Xen PCI frontend has not registered MSI/MSI-X support!\n"); 255 return irq; 256} 257 258#ifdef CONFIG_XEN_DOM0 259static bool __read_mostly pci_seg_supported = true; 260 261static int xen_initdom_setup_msi_irqs(struct pci_dev *dev, int nvec, int type) 262{ 263 int ret = 0; 264 struct msi_desc *msidesc; 265 266 list_for_each_entry(msidesc, &dev->msi_list, list) { 267 struct physdev_map_pirq map_irq; 268 domid_t domid; 269 270 domid = ret = xen_find_device_domain_owner(dev); 271 /* N.B. Casting int's -ENODEV to uint16_t results in 0xFFED, 272 * hence check ret value for < 0. */ 273 if (ret < 0) 274 domid = DOMID_SELF; 275 276 memset(&map_irq, 0, sizeof(map_irq)); 277 map_irq.domid = domid; 278 map_irq.type = MAP_PIRQ_TYPE_MSI_SEG; 279 map_irq.index = -1; 280 map_irq.pirq = -1; 281 map_irq.bus = dev->bus->number | 282 (pci_domain_nr(dev->bus) << 16); 283 map_irq.devfn = dev->devfn; 284 285 if (type == PCI_CAP_ID_MSIX) { 286 int pos; 287 u32 table_offset, bir; 288 289 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX); 290 291 pci_read_config_dword(dev, pos + PCI_MSIX_TABLE, 292 &table_offset); 293 bir = (u8)(table_offset & PCI_MSIX_FLAGS_BIRMASK); 294 295 map_irq.table_base = pci_resource_start(dev, bir); 296 map_irq.entry_nr = msidesc->msi_attrib.entry_nr; 297 } 298 299 ret = -EINVAL; 300 if (pci_seg_supported) 301 ret = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, 302 &map_irq); 303 if (ret == -EINVAL && !pci_domain_nr(dev->bus)) { 304 map_irq.type = MAP_PIRQ_TYPE_MSI; 305 map_irq.index = -1; 306 map_irq.pirq = -1; 307 map_irq.bus = dev->bus->number; 308 ret = HYPERVISOR_physdev_op(PHYSDEVOP_map_pirq, 309 &map_irq); 310 if (ret != -EINVAL) 311 pci_seg_supported = false; 312 } 313 if (ret) { 314 dev_warn(&dev->dev, "xen map irq failed %d for %d domain\n", 315 ret, domid); 316 goto out; 317 } 318 319 ret = xen_bind_pirq_msi_to_irq(dev, msidesc, 320 map_irq.pirq, map_irq.index, 321 (type == PCI_CAP_ID_MSIX) ? 322 "msi-x" : "msi", 323 domid); 324 if (ret < 0) 325 goto out; 326 } 327 ret = 0; 328out: 329 return ret; 330} 331 332static void xen_initdom_restore_msi_irqs(struct pci_dev *dev, int irq) 333{ 334 int ret = 0; 335 336 if (pci_seg_supported) { 337 struct physdev_pci_device restore_ext; 338 339 restore_ext.seg = pci_domain_nr(dev->bus); 340 restore_ext.bus = dev->bus->number; 341 restore_ext.devfn = dev->devfn; 342 ret = HYPERVISOR_physdev_op(PHYSDEVOP_restore_msi_ext, 343 &restore_ext); 344 if (ret == -ENOSYS) 345 pci_seg_supported = false; 346 WARN(ret && ret != -ENOSYS, "restore_msi_ext -> %d\n", ret); 347 } 348 if (!pci_seg_supported) { 349 struct physdev_restore_msi restore; 350 351 restore.bus = dev->bus->number; 352 restore.devfn = dev->devfn; 353 ret = HYPERVISOR_physdev_op(PHYSDEVOP_restore_msi, &restore); 354 WARN(ret && ret != -ENOSYS, "restore_msi -> %d\n", ret); 355 } 356} 357#endif 358 359static void xen_teardown_msi_irqs(struct pci_dev *dev) 360{ 361 struct msi_desc *msidesc; 362 363 msidesc = list_entry(dev->msi_list.next, struct msi_desc, list); 364 if (msidesc->msi_attrib.is_msix) 365 xen_pci_frontend_disable_msix(dev); 366 else 367 xen_pci_frontend_disable_msi(dev); 368 369 /* Free the IRQ's and the msidesc using the generic code. */ 370 default_teardown_msi_irqs(dev); 371} 372 373static void xen_teardown_msi_irq(unsigned int irq) 374{ 375 xen_destroy_irq(irq); 376} 377 378#endif 379 380int __init pci_xen_init(void) 381{ 382 if (!xen_pv_domain() || xen_initial_domain()) 383 return -ENODEV; 384 385 printk(KERN_INFO "PCI: setting up Xen PCI frontend stub\n"); 386 387 pcibios_set_cache_line_size(); 388 389 pcibios_enable_irq = xen_pcifront_enable_irq; 390 pcibios_disable_irq = NULL; 391 392#ifdef CONFIG_ACPI 393 /* Keep ACPI out of the picture */ 394 acpi_noirq = 1; 395#endif 396 397#ifdef CONFIG_PCI_MSI 398 x86_msi.setup_msi_irqs = xen_setup_msi_irqs; 399 x86_msi.teardown_msi_irq = xen_teardown_msi_irq; 400 x86_msi.teardown_msi_irqs = xen_teardown_msi_irqs; 401#endif 402 return 0; 403} 404 405int __init pci_xen_hvm_init(void) 406{ 407 if (!xen_have_vector_callback || !xen_feature(XENFEAT_hvm_pirqs)) 408 return 0; 409 410#ifdef CONFIG_ACPI 411 /* 412 * We don't want to change the actual ACPI delivery model, 413 * just how GSIs get registered. 414 */ 415 __acpi_register_gsi = acpi_register_gsi_xen_hvm; 416#endif 417 418#ifdef CONFIG_PCI_MSI 419 x86_msi.setup_msi_irqs = xen_hvm_setup_msi_irqs; 420 x86_msi.teardown_msi_irq = xen_teardown_msi_irq; 421#endif 422 return 0; 423} 424 425#ifdef CONFIG_XEN_DOM0 426static __init void xen_setup_acpi_sci(void) 427{ 428 int rc; 429 int trigger, polarity; 430 int gsi = acpi_sci_override_gsi; 431 int irq = -1; 432 int gsi_override = -1; 433 434 if (!gsi) 435 return; 436 437 rc = acpi_get_override_irq(gsi, &trigger, &polarity); 438 if (rc) { 439 printk(KERN_WARNING "xen: acpi_get_override_irq failed for acpi" 440 " sci, rc=%d\n", rc); 441 return; 442 } 443 trigger = trigger ? ACPI_LEVEL_SENSITIVE : ACPI_EDGE_SENSITIVE; 444 polarity = polarity ? ACPI_ACTIVE_LOW : ACPI_ACTIVE_HIGH; 445 446 printk(KERN_INFO "xen: sci override: global_irq=%d trigger=%d " 447 "polarity=%d\n", gsi, trigger, polarity); 448 449 /* Before we bind the GSI to a Linux IRQ, check whether 450 * we need to override it with bus_irq (IRQ) value. Usually for 451 * IRQs below IRQ_LEGACY_IRQ this holds IRQ == GSI, as so: 452 * ACPI: INT_SRC_OVR (bus 0 bus_irq 9 global_irq 9 low level) 453 * but there are oddballs where the IRQ != GSI: 454 * ACPI: INT_SRC_OVR (bus 0 bus_irq 9 global_irq 20 low level) 455 * which ends up being: gsi_to_irq[9] == 20 456 * (which is what acpi_gsi_to_irq ends up calling when starting the 457 * the ACPI interpreter and keels over since IRQ 9 has not been 458 * setup as we had setup IRQ 20 for it). 459 */ 460 if (acpi_gsi_to_irq(gsi, &irq) == 0) { 461 /* Use the provided value if it's valid. */ 462 if (irq >= 0) 463 gsi_override = irq; 464 } 465 466 gsi = xen_register_gsi(gsi, gsi_override, trigger, polarity); 467 printk(KERN_INFO "xen: acpi sci %d\n", gsi); 468 469 return; 470} 471 472int __init pci_xen_initial_domain(void) 473{ 474 int irq; 475 476#ifdef CONFIG_PCI_MSI 477 x86_msi.setup_msi_irqs = xen_initdom_setup_msi_irqs; 478 x86_msi.teardown_msi_irq = xen_teardown_msi_irq; 479 x86_msi.restore_msi_irqs = xen_initdom_restore_msi_irqs; 480#endif 481 xen_setup_acpi_sci(); 482 __acpi_register_gsi = acpi_register_gsi_xen; 483 /* Pre-allocate legacy irqs */ 484 for (irq = 0; irq < NR_IRQS_LEGACY; irq++) { 485 int trigger, polarity; 486 487 if (acpi_get_override_irq(irq, &trigger, &polarity) == -1) 488 continue; 489 490 xen_register_pirq(irq, -1 /* no GSI override */, 491 trigger ? ACPI_LEVEL_SENSITIVE : ACPI_EDGE_SENSITIVE, 492 true /* Map GSI to PIRQ */); 493 } 494 if (0 == nr_ioapics) { 495 for (irq = 0; irq < NR_IRQS_LEGACY; irq++) 496 xen_bind_pirq_gsi_to_irq(irq, irq, 0, "xt-pic"); 497 } 498 return 0; 499} 500 501struct xen_device_domain_owner { 502 domid_t domain; 503 struct pci_dev *dev; 504 struct list_head list; 505}; 506 507static DEFINE_SPINLOCK(dev_domain_list_spinlock); 508static struct list_head dev_domain_list = LIST_HEAD_INIT(dev_domain_list); 509 510static struct xen_device_domain_owner *find_device(struct pci_dev *dev) 511{ 512 struct xen_device_domain_owner *owner; 513 514 list_for_each_entry(owner, &dev_domain_list, list) { 515 if (owner->dev == dev) 516 return owner; 517 } 518 return NULL; 519} 520 521int xen_find_device_domain_owner(struct pci_dev *dev) 522{ 523 struct xen_device_domain_owner *owner; 524 int domain = -ENODEV; 525 526 spin_lock(&dev_domain_list_spinlock); 527 owner = find_device(dev); 528 if (owner) 529 domain = owner->domain; 530 spin_unlock(&dev_domain_list_spinlock); 531 return domain; 532} 533EXPORT_SYMBOL_GPL(xen_find_device_domain_owner); 534 535int xen_register_device_domain_owner(struct pci_dev *dev, uint16_t domain) 536{ 537 struct xen_device_domain_owner *owner; 538 539 owner = kzalloc(sizeof(struct xen_device_domain_owner), GFP_KERNEL); 540 if (!owner) 541 return -ENODEV; 542 543 spin_lock(&dev_domain_list_spinlock); 544 if (find_device(dev)) { 545 spin_unlock(&dev_domain_list_spinlock); 546 kfree(owner); 547 return -EEXIST; 548 } 549 owner->domain = domain; 550 owner->dev = dev; 551 list_add_tail(&owner->list, &dev_domain_list); 552 spin_unlock(&dev_domain_list_spinlock); 553 return 0; 554} 555EXPORT_SYMBOL_GPL(xen_register_device_domain_owner); 556 557int xen_unregister_device_domain_owner(struct pci_dev *dev) 558{ 559 struct xen_device_domain_owner *owner; 560 561 spin_lock(&dev_domain_list_spinlock); 562 owner = find_device(dev); 563 if (!owner) { 564 spin_unlock(&dev_domain_list_spinlock); 565 return -ENODEV; 566 } 567 list_del(&owner->list); 568 spin_unlock(&dev_domain_list_spinlock); 569 kfree(owner); 570 return 0; 571} 572EXPORT_SYMBOL_GPL(xen_unregister_device_domain_owner); 573#endif 574