1 /* 2 * Xen PCI Frontend. 3 * 4 * Author: Ryan Wilson <hap9@epoch.ncsc.mil> 5 */ 6 #include <linux/module.h> 7 #include <linux/init.h> 8 #include <linux/mm.h> 9 #include <xen/xenbus.h> 10 #include <xen/events.h> 11 #include <xen/grant_table.h> 12 #include <xen/page.h> 13 #include <linux/spinlock.h> 14 #include <linux/pci.h> 15 #include <linux/msi.h> 16 #include <xen/interface/io/pciif.h> 17 #include <asm/xen/pci.h> 18 #include <linux/interrupt.h> 19 #include <linux/atomic.h> 20 #include <linux/workqueue.h> 21 #include <linux/bitops.h> 22 #include <linux/time.h> 23 24 #define INVALID_GRANT_REF (0) 25 #define INVALID_EVTCHN (-1) 26 27 struct pci_bus_entry { 28 struct list_head list; 29 struct pci_bus *bus; 30 }; 31 32 #define _PDEVB_op_active (0) 33 #define PDEVB_op_active (1 << (_PDEVB_op_active)) 34 35 struct pcifront_device { 36 struct xenbus_device *xdev; 37 struct list_head root_buses; 38 39 int evtchn; 40 int gnt_ref; 41 42 int irq; 43 44 /* Lock this when doing any operations in sh_info */ 45 spinlock_t sh_info_lock; 46 struct xen_pci_sharedinfo *sh_info; 47 struct work_struct op_work; 48 unsigned long flags; 49 50 }; 51 52 struct pcifront_sd { 53 int domain; 54 struct pcifront_device *pdev; 55 }; 56 57 static inline struct pcifront_device * 58 pcifront_get_pdev(struct pcifront_sd *sd) 59 { 60 return sd->pdev; 61 } 62 63 static inline void pcifront_init_sd(struct pcifront_sd *sd, 64 unsigned int domain, unsigned int bus, 65 struct pcifront_device *pdev) 66 { 67 sd->domain = domain; 68 sd->pdev = pdev; 69 } 70 71 static DEFINE_SPINLOCK(pcifront_dev_lock); 72 static struct pcifront_device *pcifront_dev; 73 74 static int verbose_request; 75 module_param(verbose_request, int, 0644); 76 77 static int errno_to_pcibios_err(int errno) 78 { 79 switch (errno) { 80 case XEN_PCI_ERR_success: 81 return PCIBIOS_SUCCESSFUL; 82 83 case XEN_PCI_ERR_dev_not_found: 84 return PCIBIOS_DEVICE_NOT_FOUND; 85 86 case XEN_PCI_ERR_invalid_offset: 87 case XEN_PCI_ERR_op_failed: 88 return PCIBIOS_BAD_REGISTER_NUMBER; 89 90 case XEN_PCI_ERR_not_implemented: 91 return PCIBIOS_FUNC_NOT_SUPPORTED; 92 93 case XEN_PCI_ERR_access_denied: 94 return PCIBIOS_SET_FAILED; 95 } 96 return errno; 97 } 98 99 static inline void schedule_pcifront_aer_op(struct pcifront_device *pdev) 100 { 101 if (test_bit(_XEN_PCIB_active, (unsigned long *)&pdev->sh_info->flags) 102 && !test_and_set_bit(_PDEVB_op_active, &pdev->flags)) { 103 dev_dbg(&pdev->xdev->dev, "schedule aer frontend job\n"); 104 schedule_work(&pdev->op_work); 105 } 106 } 107 108 static int do_pci_op(struct pcifront_device *pdev, struct xen_pci_op *op) 109 { 110 int err = 0; 111 struct xen_pci_op *active_op = &pdev->sh_info->op; 112 unsigned long irq_flags; 113 evtchn_port_t port = pdev->evtchn; 114 unsigned irq = pdev->irq; 115 s64 ns, ns_timeout; 116 struct timeval tv; 117 118 spin_lock_irqsave(&pdev->sh_info_lock, irq_flags); 119 120 memcpy(active_op, op, sizeof(struct xen_pci_op)); 121 122 /* Go */ 123 wmb(); 124 set_bit(_XEN_PCIF_active, (unsigned long *)&pdev->sh_info->flags); 125 notify_remote_via_evtchn(port); 126 127 /* 128 * We set a poll timeout of 3 seconds but give up on return after 129 * 2 seconds. It is better to time out too late rather than too early 130 * (in the latter case we end up continually re-executing poll() with a 131 * timeout in the past). 1s difference gives plenty of slack for error. 132 */ 133 do_gettimeofday(&tv); 134 ns_timeout = timeval_to_ns(&tv) + 2 * (s64)NSEC_PER_SEC; 135 136 xen_clear_irq_pending(irq); 137 138 while (test_bit(_XEN_PCIF_active, 139 (unsigned long *)&pdev->sh_info->flags)) { 140 xen_poll_irq_timeout(irq, jiffies + 3*HZ); 141 xen_clear_irq_pending(irq); 142 do_gettimeofday(&tv); 143 ns = timeval_to_ns(&tv); 144 if (ns > ns_timeout) { 145 dev_err(&pdev->xdev->dev, 146 "pciback not responding!!!\n"); 147 clear_bit(_XEN_PCIF_active, 148 (unsigned long *)&pdev->sh_info->flags); 149 err = XEN_PCI_ERR_dev_not_found; 150 goto out; 151 } 152 } 153 154 /* 155 * We might lose backend service request since we 156 * reuse same evtchn with pci_conf backend response. So re-schedule 157 * aer pcifront service. 158 */ 159 if (test_bit(_XEN_PCIB_active, 160 (unsigned long *)&pdev->sh_info->flags)) { 161 dev_err(&pdev->xdev->dev, 162 "schedule aer pcifront service\n"); 163 schedule_pcifront_aer_op(pdev); 164 } 165 166 memcpy(op, active_op, sizeof(struct xen_pci_op)); 167 168 err = op->err; 169 out: 170 spin_unlock_irqrestore(&pdev->sh_info_lock, irq_flags); 171 return err; 172 } 173 174 /* Access to this function is spinlocked in drivers/pci/access.c */ 175 static int pcifront_bus_read(struct pci_bus *bus, unsigned int devfn, 176 int where, int size, u32 *val) 177 { 178 int err = 0; 179 struct xen_pci_op op = { 180 .cmd = XEN_PCI_OP_conf_read, 181 .domain = pci_domain_nr(bus), 182 .bus = bus->number, 183 .devfn = devfn, 184 .offset = where, 185 .size = size, 186 }; 187 struct pcifront_sd *sd = bus->sysdata; 188 struct pcifront_device *pdev = pcifront_get_pdev(sd); 189 190 if (verbose_request) 191 dev_info(&pdev->xdev->dev, 192 "read dev=%04x:%02x:%02x.%01x - offset %x size %d\n", 193 pci_domain_nr(bus), bus->number, PCI_SLOT(devfn), 194 PCI_FUNC(devfn), where, size); 195 196 err = do_pci_op(pdev, &op); 197 198 if (likely(!err)) { 199 if (verbose_request) 200 dev_info(&pdev->xdev->dev, "read got back value %x\n", 201 op.value); 202 203 *val = op.value; 204 } else if (err == -ENODEV) { 205 /* No device here, pretend that it just returned 0 */ 206 err = 0; 207 *val = 0; 208 } 209 210 return errno_to_pcibios_err(err); 211 } 212 213 /* Access to this function is spinlocked in drivers/pci/access.c */ 214 static int pcifront_bus_write(struct pci_bus *bus, unsigned int devfn, 215 int where, int size, u32 val) 216 { 217 struct xen_pci_op op = { 218 .cmd = XEN_PCI_OP_conf_write, 219 .domain = pci_domain_nr(bus), 220 .bus = bus->number, 221 .devfn = devfn, 222 .offset = where, 223 .size = size, 224 .value = val, 225 }; 226 struct pcifront_sd *sd = bus->sysdata; 227 struct pcifront_device *pdev = pcifront_get_pdev(sd); 228 229 if (verbose_request) 230 dev_info(&pdev->xdev->dev, 231 "write dev=%04x:%02x:%02x.%01x - " 232 "offset %x size %d val %x\n", 233 pci_domain_nr(bus), bus->number, 234 PCI_SLOT(devfn), PCI_FUNC(devfn), where, size, val); 235 236 return errno_to_pcibios_err(do_pci_op(pdev, &op)); 237 } 238 239 struct pci_ops pcifront_bus_ops = { 240 .read = pcifront_bus_read, 241 .write = pcifront_bus_write, 242 }; 243 244 #ifdef CONFIG_PCI_MSI 245 static int pci_frontend_enable_msix(struct pci_dev *dev, 246 int vector[], int nvec) 247 { 248 int err; 249 int i; 250 struct xen_pci_op op = { 251 .cmd = XEN_PCI_OP_enable_msix, 252 .domain = pci_domain_nr(dev->bus), 253 .bus = dev->bus->number, 254 .devfn = dev->devfn, 255 .value = nvec, 256 }; 257 struct pcifront_sd *sd = dev->bus->sysdata; 258 struct pcifront_device *pdev = pcifront_get_pdev(sd); 259 struct msi_desc *entry; 260 261 if (nvec > SH_INFO_MAX_VEC) { 262 dev_err(&dev->dev, "too much vector for pci frontend: %x." 263 " Increase SH_INFO_MAX_VEC.\n", nvec); 264 return -EINVAL; 265 } 266 267 i = 0; 268 list_for_each_entry(entry, &dev->msi_list, list) { 269 op.msix_entries[i].entry = entry->msi_attrib.entry_nr; 270 /* Vector is useless at this point. */ 271 op.msix_entries[i].vector = -1; 272 i++; 273 } 274 275 err = do_pci_op(pdev, &op); 276 277 if (likely(!err)) { 278 if (likely(!op.value)) { 279 /* we get the result */ 280 for (i = 0; i < nvec; i++) { 281 if (op.msix_entries[i].vector <= 0) { 282 dev_warn(&dev->dev, "MSI-X entry %d is invalid: %d!\n", 283 i, op.msix_entries[i].vector); 284 err = -EINVAL; 285 vector[i] = -1; 286 continue; 287 } 288 vector[i] = op.msix_entries[i].vector; 289 } 290 } else { 291 printk(KERN_DEBUG "enable msix get value %x\n", 292 op.value); 293 } 294 } else { 295 dev_err(&dev->dev, "enable msix get err %x\n", err); 296 } 297 return err; 298 } 299 300 static void pci_frontend_disable_msix(struct pci_dev *dev) 301 { 302 int err; 303 struct xen_pci_op op = { 304 .cmd = XEN_PCI_OP_disable_msix, 305 .domain = pci_domain_nr(dev->bus), 306 .bus = dev->bus->number, 307 .devfn = dev->devfn, 308 }; 309 struct pcifront_sd *sd = dev->bus->sysdata; 310 struct pcifront_device *pdev = pcifront_get_pdev(sd); 311 312 err = do_pci_op(pdev, &op); 313 314 /* What should do for error ? */ 315 if (err) 316 dev_err(&dev->dev, "pci_disable_msix get err %x\n", err); 317 } 318 319 static int pci_frontend_enable_msi(struct pci_dev *dev, int vector[]) 320 { 321 int err; 322 struct xen_pci_op op = { 323 .cmd = XEN_PCI_OP_enable_msi, 324 .domain = pci_domain_nr(dev->bus), 325 .bus = dev->bus->number, 326 .devfn = dev->devfn, 327 }; 328 struct pcifront_sd *sd = dev->bus->sysdata; 329 struct pcifront_device *pdev = pcifront_get_pdev(sd); 330 331 err = do_pci_op(pdev, &op); 332 if (likely(!err)) { 333 vector[0] = op.value; 334 if (op.value <= 0) { 335 dev_warn(&dev->dev, "MSI entry is invalid: %d!\n", 336 op.value); 337 err = -EINVAL; 338 vector[0] = -1; 339 } 340 } else { 341 dev_err(&dev->dev, "pci frontend enable msi failed for dev " 342 "%x:%x\n", op.bus, op.devfn); 343 err = -EINVAL; 344 } 345 return err; 346 } 347 348 static void pci_frontend_disable_msi(struct pci_dev *dev) 349 { 350 int err; 351 struct xen_pci_op op = { 352 .cmd = XEN_PCI_OP_disable_msi, 353 .domain = pci_domain_nr(dev->bus), 354 .bus = dev->bus->number, 355 .devfn = dev->devfn, 356 }; 357 struct pcifront_sd *sd = dev->bus->sysdata; 358 struct pcifront_device *pdev = pcifront_get_pdev(sd); 359 360 err = do_pci_op(pdev, &op); 361 if (err == XEN_PCI_ERR_dev_not_found) { 362 /* XXX No response from backend, what shall we do? */ 363 printk(KERN_DEBUG "get no response from backend for disable MSI\n"); 364 return; 365 } 366 if (err) 367 /* how can pciback notify us fail? */ 368 printk(KERN_DEBUG "get fake response frombackend\n"); 369 } 370 371 static struct xen_pci_frontend_ops pci_frontend_ops = { 372 .enable_msi = pci_frontend_enable_msi, 373 .disable_msi = pci_frontend_disable_msi, 374 .enable_msix = pci_frontend_enable_msix, 375 .disable_msix = pci_frontend_disable_msix, 376 }; 377 378 static void pci_frontend_registrar(int enable) 379 { 380 if (enable) 381 xen_pci_frontend = &pci_frontend_ops; 382 else 383 xen_pci_frontend = NULL; 384 }; 385 #else 386 static inline void pci_frontend_registrar(int enable) { }; 387 #endif /* CONFIG_PCI_MSI */ 388 389 /* Claim resources for the PCI frontend as-is, backend won't allow changes */ 390 static int pcifront_claim_resource(struct pci_dev *dev, void *data) 391 { 392 struct pcifront_device *pdev = data; 393 int i; 394 struct resource *r; 395 396 for (i = 0; i < PCI_NUM_RESOURCES; i++) { 397 r = &dev->resource[i]; 398 399 if (!r->parent && r->start && r->flags) { 400 dev_info(&pdev->xdev->dev, "claiming resource %s/%d\n", 401 pci_name(dev), i); 402 if (pci_claim_resource(dev, i)) { 403 dev_err(&pdev->xdev->dev, "Could not claim " 404 "resource %s/%d! Device offline. Try " 405 "giving less than 4GB to domain.\n", 406 pci_name(dev), i); 407 } 408 } 409 } 410 411 return 0; 412 } 413 414 static int __devinit pcifront_scan_bus(struct pcifront_device *pdev, 415 unsigned int domain, unsigned int bus, 416 struct pci_bus *b) 417 { 418 struct pci_dev *d; 419 unsigned int devfn; 420 421 /* Scan the bus for functions and add. 422 * We omit handling of PCI bridge attachment because pciback prevents 423 * bridges from being exported. 424 */ 425 for (devfn = 0; devfn < 0x100; devfn++) { 426 d = pci_get_slot(b, devfn); 427 if (d) { 428 /* Device is already known. */ 429 pci_dev_put(d); 430 continue; 431 } 432 433 d = pci_scan_single_device(b, devfn); 434 if (d) 435 dev_info(&pdev->xdev->dev, "New device on " 436 "%04x:%02x:%02x.%02x found.\n", domain, bus, 437 PCI_SLOT(devfn), PCI_FUNC(devfn)); 438 } 439 440 return 0; 441 } 442 443 static int __devinit pcifront_scan_root(struct pcifront_device *pdev, 444 unsigned int domain, unsigned int bus) 445 { 446 struct pci_bus *b; 447 struct pcifront_sd *sd = NULL; 448 struct pci_bus_entry *bus_entry = NULL; 449 int err = 0; 450 451 #ifndef CONFIG_PCI_DOMAINS 452 if (domain != 0) { 453 dev_err(&pdev->xdev->dev, 454 "PCI Root in non-zero PCI Domain! domain=%d\n", domain); 455 dev_err(&pdev->xdev->dev, 456 "Please compile with CONFIG_PCI_DOMAINS\n"); 457 err = -EINVAL; 458 goto err_out; 459 } 460 #endif 461 462 dev_info(&pdev->xdev->dev, "Creating PCI Frontend Bus %04x:%02x\n", 463 domain, bus); 464 465 bus_entry = kmalloc(sizeof(*bus_entry), GFP_KERNEL); 466 sd = kmalloc(sizeof(*sd), GFP_KERNEL); 467 if (!bus_entry || !sd) { 468 err = -ENOMEM; 469 goto err_out; 470 } 471 pcifront_init_sd(sd, domain, bus, pdev); 472 473 b = pci_scan_bus_parented(&pdev->xdev->dev, bus, 474 &pcifront_bus_ops, sd); 475 if (!b) { 476 dev_err(&pdev->xdev->dev, 477 "Error creating PCI Frontend Bus!\n"); 478 err = -ENOMEM; 479 goto err_out; 480 } 481 482 bus_entry->bus = b; 483 484 list_add(&bus_entry->list, &pdev->root_buses); 485 486 /* pci_scan_bus_parented skips devices which do not have a have 487 * devfn==0. The pcifront_scan_bus enumerates all devfn. */ 488 err = pcifront_scan_bus(pdev, domain, bus, b); 489 490 /* Claim resources before going "live" with our devices */ 491 pci_walk_bus(b, pcifront_claim_resource, pdev); 492 493 /* Create SysFS and notify udev of the devices. Aka: "going live" */ 494 pci_bus_add_devices(b); 495 496 return err; 497 498 err_out: 499 kfree(bus_entry); 500 kfree(sd); 501 502 return err; 503 } 504 505 static int __devinit pcifront_rescan_root(struct pcifront_device *pdev, 506 unsigned int domain, unsigned int bus) 507 { 508 int err; 509 struct pci_bus *b; 510 511 #ifndef CONFIG_PCI_DOMAINS 512 if (domain != 0) { 513 dev_err(&pdev->xdev->dev, 514 "PCI Root in non-zero PCI Domain! domain=%d\n", domain); 515 dev_err(&pdev->xdev->dev, 516 "Please compile with CONFIG_PCI_DOMAINS\n"); 517 return -EINVAL; 518 } 519 #endif 520 521 dev_info(&pdev->xdev->dev, "Rescanning PCI Frontend Bus %04x:%02x\n", 522 domain, bus); 523 524 b = pci_find_bus(domain, bus); 525 if (!b) 526 /* If the bus is unknown, create it. */ 527 return pcifront_scan_root(pdev, domain, bus); 528 529 err = pcifront_scan_bus(pdev, domain, bus, b); 530 531 /* Claim resources before going "live" with our devices */ 532 pci_walk_bus(b, pcifront_claim_resource, pdev); 533 534 /* Create SysFS and notify udev of the devices. Aka: "going live" */ 535 pci_bus_add_devices(b); 536 537 return err; 538 } 539 540 static void free_root_bus_devs(struct pci_bus *bus) 541 { 542 struct pci_dev *dev; 543 544 while (!list_empty(&bus->devices)) { 545 dev = container_of(bus->devices.next, struct pci_dev, 546 bus_list); 547 dev_dbg(&dev->dev, "removing device\n"); 548 pci_remove_bus_device(dev); 549 } 550 } 551 552 static void pcifront_free_roots(struct pcifront_device *pdev) 553 { 554 struct pci_bus_entry *bus_entry, *t; 555 556 dev_dbg(&pdev->xdev->dev, "cleaning up root buses\n"); 557 558 list_for_each_entry_safe(bus_entry, t, &pdev->root_buses, list) { 559 list_del(&bus_entry->list); 560 561 free_root_bus_devs(bus_entry->bus); 562 563 kfree(bus_entry->bus->sysdata); 564 565 device_unregister(bus_entry->bus->bridge); 566 pci_remove_bus(bus_entry->bus); 567 568 kfree(bus_entry); 569 } 570 } 571 572 static pci_ers_result_t pcifront_common_process(int cmd, 573 struct pcifront_device *pdev, 574 pci_channel_state_t state) 575 { 576 pci_ers_result_t result; 577 struct pci_driver *pdrv; 578 int bus = pdev->sh_info->aer_op.bus; 579 int devfn = pdev->sh_info->aer_op.devfn; 580 struct pci_dev *pcidev; 581 int flag = 0; 582 583 dev_dbg(&pdev->xdev->dev, 584 "pcifront AER process: cmd %x (bus:%x, devfn%x)", 585 cmd, bus, devfn); 586 result = PCI_ERS_RESULT_NONE; 587 588 pcidev = pci_get_bus_and_slot(bus, devfn); 589 if (!pcidev || !pcidev->driver) { 590 dev_err(&pdev->xdev->dev, "device or AER driver is NULL\n"); 591 if (pcidev) 592 pci_dev_put(pcidev); 593 return result; 594 } 595 pdrv = pcidev->driver; 596 597 if (get_driver(&pdrv->driver)) { 598 if (pdrv->err_handler && pdrv->err_handler->error_detected) { 599 dev_dbg(&pcidev->dev, 600 "trying to call AER service\n"); 601 if (pcidev) { 602 flag = 1; 603 switch (cmd) { 604 case XEN_PCI_OP_aer_detected: 605 result = pdrv->err_handler-> 606 error_detected(pcidev, state); 607 break; 608 case XEN_PCI_OP_aer_mmio: 609 result = pdrv->err_handler-> 610 mmio_enabled(pcidev); 611 break; 612 case XEN_PCI_OP_aer_slotreset: 613 result = pdrv->err_handler-> 614 slot_reset(pcidev); 615 break; 616 case XEN_PCI_OP_aer_resume: 617 pdrv->err_handler->resume(pcidev); 618 break; 619 default: 620 dev_err(&pdev->xdev->dev, 621 "bad request in aer recovery " 622 "operation!\n"); 623 624 } 625 } 626 } 627 put_driver(&pdrv->driver); 628 } 629 if (!flag) 630 result = PCI_ERS_RESULT_NONE; 631 632 return result; 633 } 634 635 636 static void pcifront_do_aer(struct work_struct *data) 637 { 638 struct pcifront_device *pdev = 639 container_of(data, struct pcifront_device, op_work); 640 int cmd = pdev->sh_info->aer_op.cmd; 641 pci_channel_state_t state = 642 (pci_channel_state_t)pdev->sh_info->aer_op.err; 643 644 /*If a pci_conf op is in progress, 645 we have to wait until it is done before service aer op*/ 646 dev_dbg(&pdev->xdev->dev, 647 "pcifront service aer bus %x devfn %x\n", 648 pdev->sh_info->aer_op.bus, pdev->sh_info->aer_op.devfn); 649 650 pdev->sh_info->aer_op.err = pcifront_common_process(cmd, pdev, state); 651 652 /* Post the operation to the guest. */ 653 wmb(); 654 clear_bit(_XEN_PCIB_active, (unsigned long *)&pdev->sh_info->flags); 655 notify_remote_via_evtchn(pdev->evtchn); 656 657 /*in case of we lost an aer request in four lines time_window*/ 658 smp_mb__before_clear_bit(); 659 clear_bit(_PDEVB_op_active, &pdev->flags); 660 smp_mb__after_clear_bit(); 661 662 schedule_pcifront_aer_op(pdev); 663 664 } 665 666 static irqreturn_t pcifront_handler_aer(int irq, void *dev) 667 { 668 struct pcifront_device *pdev = dev; 669 schedule_pcifront_aer_op(pdev); 670 return IRQ_HANDLED; 671 } 672 static int pcifront_connect(struct pcifront_device *pdev) 673 { 674 int err = 0; 675 676 spin_lock(&pcifront_dev_lock); 677 678 if (!pcifront_dev) { 679 dev_info(&pdev->xdev->dev, "Installing PCI frontend\n"); 680 pcifront_dev = pdev; 681 } else { 682 dev_err(&pdev->xdev->dev, "PCI frontend already installed!\n"); 683 err = -EEXIST; 684 } 685 686 spin_unlock(&pcifront_dev_lock); 687 688 return err; 689 } 690 691 static void pcifront_disconnect(struct pcifront_device *pdev) 692 { 693 spin_lock(&pcifront_dev_lock); 694 695 if (pdev == pcifront_dev) { 696 dev_info(&pdev->xdev->dev, 697 "Disconnecting PCI Frontend Buses\n"); 698 pcifront_dev = NULL; 699 } 700 701 spin_unlock(&pcifront_dev_lock); 702 } 703 static struct pcifront_device *alloc_pdev(struct xenbus_device *xdev) 704 { 705 struct pcifront_device *pdev; 706 707 pdev = kzalloc(sizeof(struct pcifront_device), GFP_KERNEL); 708 if (pdev == NULL) 709 goto out; 710 711 pdev->sh_info = 712 (struct xen_pci_sharedinfo *)__get_free_page(GFP_KERNEL); 713 if (pdev->sh_info == NULL) { 714 kfree(pdev); 715 pdev = NULL; 716 goto out; 717 } 718 pdev->sh_info->flags = 0; 719 720 /*Flag for registering PV AER handler*/ 721 set_bit(_XEN_PCIB_AERHANDLER, (void *)&pdev->sh_info->flags); 722 723 dev_set_drvdata(&xdev->dev, pdev); 724 pdev->xdev = xdev; 725 726 INIT_LIST_HEAD(&pdev->root_buses); 727 728 spin_lock_init(&pdev->sh_info_lock); 729 730 pdev->evtchn = INVALID_EVTCHN; 731 pdev->gnt_ref = INVALID_GRANT_REF; 732 pdev->irq = -1; 733 734 INIT_WORK(&pdev->op_work, pcifront_do_aer); 735 736 dev_dbg(&xdev->dev, "Allocated pdev @ 0x%p pdev->sh_info @ 0x%p\n", 737 pdev, pdev->sh_info); 738 out: 739 return pdev; 740 } 741 742 static void free_pdev(struct pcifront_device *pdev) 743 { 744 dev_dbg(&pdev->xdev->dev, "freeing pdev @ 0x%p\n", pdev); 745 746 pcifront_free_roots(pdev); 747 748 cancel_work_sync(&pdev->op_work); 749 750 if (pdev->irq >= 0) 751 unbind_from_irqhandler(pdev->irq, pdev); 752 753 if (pdev->evtchn != INVALID_EVTCHN) 754 xenbus_free_evtchn(pdev->xdev, pdev->evtchn); 755 756 if (pdev->gnt_ref != INVALID_GRANT_REF) 757 gnttab_end_foreign_access(pdev->gnt_ref, 0 /* r/w page */, 758 (unsigned long)pdev->sh_info); 759 else 760 free_page((unsigned long)pdev->sh_info); 761 762 dev_set_drvdata(&pdev->xdev->dev, NULL); 763 764 kfree(pdev); 765 } 766 767 static int pcifront_publish_info(struct pcifront_device *pdev) 768 { 769 int err = 0; 770 struct xenbus_transaction trans; 771 772 err = xenbus_grant_ring(pdev->xdev, virt_to_mfn(pdev->sh_info)); 773 if (err < 0) 774 goto out; 775 776 pdev->gnt_ref = err; 777 778 err = xenbus_alloc_evtchn(pdev->xdev, &pdev->evtchn); 779 if (err) 780 goto out; 781 782 err = bind_evtchn_to_irqhandler(pdev->evtchn, pcifront_handler_aer, 783 0, "pcifront", pdev); 784 785 if (err < 0) 786 return err; 787 788 pdev->irq = err; 789 790 do_publish: 791 err = xenbus_transaction_start(&trans); 792 if (err) { 793 xenbus_dev_fatal(pdev->xdev, err, 794 "Error writing configuration for backend " 795 "(start transaction)"); 796 goto out; 797 } 798 799 err = xenbus_printf(trans, pdev->xdev->nodename, 800 "pci-op-ref", "%u", pdev->gnt_ref); 801 if (!err) 802 err = xenbus_printf(trans, pdev->xdev->nodename, 803 "event-channel", "%u", pdev->evtchn); 804 if (!err) 805 err = xenbus_printf(trans, pdev->xdev->nodename, 806 "magic", XEN_PCI_MAGIC); 807 808 if (err) { 809 xenbus_transaction_end(trans, 1); 810 xenbus_dev_fatal(pdev->xdev, err, 811 "Error writing configuration for backend"); 812 goto out; 813 } else { 814 err = xenbus_transaction_end(trans, 0); 815 if (err == -EAGAIN) 816 goto do_publish; 817 else if (err) { 818 xenbus_dev_fatal(pdev->xdev, err, 819 "Error completing transaction " 820 "for backend"); 821 goto out; 822 } 823 } 824 825 xenbus_switch_state(pdev->xdev, XenbusStateInitialised); 826 827 dev_dbg(&pdev->xdev->dev, "publishing successful!\n"); 828 829 out: 830 return err; 831 } 832 833 static int __devinit pcifront_try_connect(struct pcifront_device *pdev) 834 { 835 int err = -EFAULT; 836 int i, num_roots, len; 837 char str[64]; 838 unsigned int domain, bus; 839 840 841 /* Only connect once */ 842 if (xenbus_read_driver_state(pdev->xdev->nodename) != 843 XenbusStateInitialised) 844 goto out; 845 846 err = pcifront_connect(pdev); 847 if (err) { 848 xenbus_dev_fatal(pdev->xdev, err, 849 "Error connecting PCI Frontend"); 850 goto out; 851 } 852 853 err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, 854 "root_num", "%d", &num_roots); 855 if (err == -ENOENT) { 856 xenbus_dev_error(pdev->xdev, err, 857 "No PCI Roots found, trying 0000:00"); 858 err = pcifront_scan_root(pdev, 0, 0); 859 num_roots = 0; 860 } else if (err != 1) { 861 if (err == 0) 862 err = -EINVAL; 863 xenbus_dev_fatal(pdev->xdev, err, 864 "Error reading number of PCI roots"); 865 goto out; 866 } 867 868 for (i = 0; i < num_roots; i++) { 869 len = snprintf(str, sizeof(str), "root-%d", i); 870 if (unlikely(len >= (sizeof(str) - 1))) { 871 err = -ENOMEM; 872 goto out; 873 } 874 875 err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str, 876 "%x:%x", &domain, &bus); 877 if (err != 2) { 878 if (err >= 0) 879 err = -EINVAL; 880 xenbus_dev_fatal(pdev->xdev, err, 881 "Error reading PCI root %d", i); 882 goto out; 883 } 884 885 err = pcifront_scan_root(pdev, domain, bus); 886 if (err) { 887 xenbus_dev_fatal(pdev->xdev, err, 888 "Error scanning PCI root %04x:%02x", 889 domain, bus); 890 goto out; 891 } 892 } 893 894 err = xenbus_switch_state(pdev->xdev, XenbusStateConnected); 895 896 out: 897 return err; 898 } 899 900 static int pcifront_try_disconnect(struct pcifront_device *pdev) 901 { 902 int err = 0; 903 enum xenbus_state prev_state; 904 905 906 prev_state = xenbus_read_driver_state(pdev->xdev->nodename); 907 908 if (prev_state >= XenbusStateClosing) 909 goto out; 910 911 if (prev_state == XenbusStateConnected) { 912 pcifront_free_roots(pdev); 913 pcifront_disconnect(pdev); 914 } 915 916 err = xenbus_switch_state(pdev->xdev, XenbusStateClosed); 917 918 out: 919 920 return err; 921 } 922 923 static int __devinit pcifront_attach_devices(struct pcifront_device *pdev) 924 { 925 int err = -EFAULT; 926 int i, num_roots, len; 927 unsigned int domain, bus; 928 char str[64]; 929 930 if (xenbus_read_driver_state(pdev->xdev->nodename) != 931 XenbusStateReconfiguring) 932 goto out; 933 934 err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, 935 "root_num", "%d", &num_roots); 936 if (err == -ENOENT) { 937 xenbus_dev_error(pdev->xdev, err, 938 "No PCI Roots found, trying 0000:00"); 939 err = pcifront_rescan_root(pdev, 0, 0); 940 num_roots = 0; 941 } else if (err != 1) { 942 if (err == 0) 943 err = -EINVAL; 944 xenbus_dev_fatal(pdev->xdev, err, 945 "Error reading number of PCI roots"); 946 goto out; 947 } 948 949 for (i = 0; i < num_roots; i++) { 950 len = snprintf(str, sizeof(str), "root-%d", i); 951 if (unlikely(len >= (sizeof(str) - 1))) { 952 err = -ENOMEM; 953 goto out; 954 } 955 956 err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str, 957 "%x:%x", &domain, &bus); 958 if (err != 2) { 959 if (err >= 0) 960 err = -EINVAL; 961 xenbus_dev_fatal(pdev->xdev, err, 962 "Error reading PCI root %d", i); 963 goto out; 964 } 965 966 err = pcifront_rescan_root(pdev, domain, bus); 967 if (err) { 968 xenbus_dev_fatal(pdev->xdev, err, 969 "Error scanning PCI root %04x:%02x", 970 domain, bus); 971 goto out; 972 } 973 } 974 975 xenbus_switch_state(pdev->xdev, XenbusStateConnected); 976 977 out: 978 return err; 979 } 980 981 static int pcifront_detach_devices(struct pcifront_device *pdev) 982 { 983 int err = 0; 984 int i, num_devs; 985 unsigned int domain, bus, slot, func; 986 struct pci_bus *pci_bus; 987 struct pci_dev *pci_dev; 988 char str[64]; 989 990 if (xenbus_read_driver_state(pdev->xdev->nodename) != 991 XenbusStateConnected) 992 goto out; 993 994 err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, "num_devs", "%d", 995 &num_devs); 996 if (err != 1) { 997 if (err >= 0) 998 err = -EINVAL; 999 xenbus_dev_fatal(pdev->xdev, err, 1000 "Error reading number of PCI devices"); 1001 goto out; 1002 } 1003 1004 /* Find devices being detached and remove them. */ 1005 for (i = 0; i < num_devs; i++) { 1006 int l, state; 1007 l = snprintf(str, sizeof(str), "state-%d", i); 1008 if (unlikely(l >= (sizeof(str) - 1))) { 1009 err = -ENOMEM; 1010 goto out; 1011 } 1012 err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str, "%d", 1013 &state); 1014 if (err != 1) 1015 state = XenbusStateUnknown; 1016 1017 if (state != XenbusStateClosing) 1018 continue; 1019 1020 /* Remove device. */ 1021 l = snprintf(str, sizeof(str), "vdev-%d", i); 1022 if (unlikely(l >= (sizeof(str) - 1))) { 1023 err = -ENOMEM; 1024 goto out; 1025 } 1026 err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str, 1027 "%x:%x:%x.%x", &domain, &bus, &slot, &func); 1028 if (err != 4) { 1029 if (err >= 0) 1030 err = -EINVAL; 1031 xenbus_dev_fatal(pdev->xdev, err, 1032 "Error reading PCI device %d", i); 1033 goto out; 1034 } 1035 1036 pci_bus = pci_find_bus(domain, bus); 1037 if (!pci_bus) { 1038 dev_dbg(&pdev->xdev->dev, "Cannot get bus %04x:%02x\n", 1039 domain, bus); 1040 continue; 1041 } 1042 pci_dev = pci_get_slot(pci_bus, PCI_DEVFN(slot, func)); 1043 if (!pci_dev) { 1044 dev_dbg(&pdev->xdev->dev, 1045 "Cannot get PCI device %04x:%02x:%02x.%02x\n", 1046 domain, bus, slot, func); 1047 continue; 1048 } 1049 pci_remove_bus_device(pci_dev); 1050 pci_dev_put(pci_dev); 1051 1052 dev_dbg(&pdev->xdev->dev, 1053 "PCI device %04x:%02x:%02x.%02x removed.\n", 1054 domain, bus, slot, func); 1055 } 1056 1057 err = xenbus_switch_state(pdev->xdev, XenbusStateReconfiguring); 1058 1059 out: 1060 return err; 1061 } 1062 1063 static void __init_refok pcifront_backend_changed(struct xenbus_device *xdev, 1064 enum xenbus_state be_state) 1065 { 1066 struct pcifront_device *pdev = dev_get_drvdata(&xdev->dev); 1067 1068 switch (be_state) { 1069 case XenbusStateUnknown: 1070 case XenbusStateInitialising: 1071 case XenbusStateInitWait: 1072 case XenbusStateInitialised: 1073 case XenbusStateClosed: 1074 break; 1075 1076 case XenbusStateConnected: 1077 pcifront_try_connect(pdev); 1078 break; 1079 1080 case XenbusStateClosing: 1081 dev_warn(&xdev->dev, "backend going away!\n"); 1082 pcifront_try_disconnect(pdev); 1083 break; 1084 1085 case XenbusStateReconfiguring: 1086 pcifront_detach_devices(pdev); 1087 break; 1088 1089 case XenbusStateReconfigured: 1090 pcifront_attach_devices(pdev); 1091 break; 1092 } 1093 } 1094 1095 static int pcifront_xenbus_probe(struct xenbus_device *xdev, 1096 const struct xenbus_device_id *id) 1097 { 1098 int err = 0; 1099 struct pcifront_device *pdev = alloc_pdev(xdev); 1100 1101 if (pdev == NULL) { 1102 err = -ENOMEM; 1103 xenbus_dev_fatal(xdev, err, 1104 "Error allocating pcifront_device struct"); 1105 goto out; 1106 } 1107 1108 err = pcifront_publish_info(pdev); 1109 if (err) 1110 free_pdev(pdev); 1111 1112 out: 1113 return err; 1114 } 1115 1116 static int pcifront_xenbus_remove(struct xenbus_device *xdev) 1117 { 1118 struct pcifront_device *pdev = dev_get_drvdata(&xdev->dev); 1119 if (pdev) 1120 free_pdev(pdev); 1121 1122 return 0; 1123 } 1124 1125 static const struct xenbus_device_id xenpci_ids[] = { 1126 {"pci"}, 1127 {""}, 1128 }; 1129 1130 static struct xenbus_driver xenbus_pcifront_driver = { 1131 .name = "pcifront", 1132 .owner = THIS_MODULE, 1133 .ids = xenpci_ids, 1134 .probe = pcifront_xenbus_probe, 1135 .remove = pcifront_xenbus_remove, 1136 .otherend_changed = pcifront_backend_changed, 1137 }; 1138 1139 static int __init pcifront_init(void) 1140 { 1141 if (!xen_pv_domain() || xen_initial_domain()) 1142 return -ENODEV; 1143 1144 pci_frontend_registrar(1 /* enable */); 1145 1146 return xenbus_register_frontend(&xenbus_pcifront_driver); 1147 } 1148 1149 static void __exit pcifront_cleanup(void) 1150 { 1151 xenbus_unregister_driver(&xenbus_pcifront_driver); 1152 pci_frontend_registrar(0 /* disable */); 1153 } 1154 module_init(pcifront_init); 1155 module_exit(pcifront_cleanup); 1156 1157 MODULE_DESCRIPTION("Xen PCI passthrough frontend."); 1158 MODULE_LICENSE("GPL"); 1159 MODULE_ALIAS("xen:pci"); 1160