1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Xen PCI Frontend 4 * 5 * Author: Ryan Wilson <hap9@epoch.ncsc.mil> 6 */ 7 #include <linux/module.h> 8 #include <linux/init.h> 9 #include <linux/mm.h> 10 #include <xen/xenbus.h> 11 #include <xen/events.h> 12 #include <xen/grant_table.h> 13 #include <xen/page.h> 14 #include <linux/spinlock.h> 15 #include <linux/pci.h> 16 #include <linux/msi.h> 17 #include <xen/interface/io/pciif.h> 18 #include <asm/xen/pci.h> 19 #include <linux/interrupt.h> 20 #include <linux/atomic.h> 21 #include <linux/workqueue.h> 22 #include <linux/bitops.h> 23 #include <linux/time.h> 24 #include <linux/ktime.h> 25 #include <xen/platform_pci.h> 26 27 #include <asm/xen/swiotlb-xen.h> 28 #define INVALID_GRANT_REF (0) 29 #define INVALID_EVTCHN (-1) 30 31 struct pci_bus_entry { 32 struct list_head list; 33 struct pci_bus *bus; 34 }; 35 36 #define _PDEVB_op_active (0) 37 #define PDEVB_op_active (1 << (_PDEVB_op_active)) 38 39 struct pcifront_device { 40 struct xenbus_device *xdev; 41 struct list_head root_buses; 42 43 int evtchn; 44 int gnt_ref; 45 46 int irq; 47 48 /* Lock this when doing any operations in sh_info */ 49 spinlock_t sh_info_lock; 50 struct xen_pci_sharedinfo *sh_info; 51 struct work_struct op_work; 52 unsigned long flags; 53 54 }; 55 56 struct pcifront_sd { 57 struct pci_sysdata sd; 58 struct pcifront_device *pdev; 59 }; 60 61 static inline struct pcifront_device * 62 pcifront_get_pdev(struct pcifront_sd *sd) 63 { 64 return sd->pdev; 65 } 66 67 static inline void pcifront_init_sd(struct pcifront_sd *sd, 68 unsigned int domain, unsigned int bus, 69 struct pcifront_device *pdev) 70 { 71 /* Because we do not expose that information via XenBus. */ 72 sd->sd.node = first_online_node; 73 sd->sd.domain = domain; 74 sd->pdev = pdev; 75 } 76 77 static DEFINE_SPINLOCK(pcifront_dev_lock); 78 static struct pcifront_device *pcifront_dev; 79 80 static int verbose_request; 81 module_param(verbose_request, int, 0644); 82 83 static int errno_to_pcibios_err(int errno) 84 { 85 switch (errno) { 86 case XEN_PCI_ERR_success: 87 return PCIBIOS_SUCCESSFUL; 88 89 case XEN_PCI_ERR_dev_not_found: 90 return PCIBIOS_DEVICE_NOT_FOUND; 91 92 case XEN_PCI_ERR_invalid_offset: 93 case XEN_PCI_ERR_op_failed: 94 return PCIBIOS_BAD_REGISTER_NUMBER; 95 96 case XEN_PCI_ERR_not_implemented: 97 return PCIBIOS_FUNC_NOT_SUPPORTED; 98 99 case XEN_PCI_ERR_access_denied: 100 return PCIBIOS_SET_FAILED; 101 } 102 return errno; 103 } 104 105 static inline void schedule_pcifront_aer_op(struct pcifront_device *pdev) 106 { 107 if (test_bit(_XEN_PCIB_active, (unsigned long *)&pdev->sh_info->flags) 108 && !test_and_set_bit(_PDEVB_op_active, &pdev->flags)) { 109 dev_dbg(&pdev->xdev->dev, "schedule aer frontend job\n"); 110 schedule_work(&pdev->op_work); 111 } 112 } 113 114 static int do_pci_op(struct pcifront_device *pdev, struct xen_pci_op *op) 115 { 116 int err = 0; 117 struct xen_pci_op *active_op = &pdev->sh_info->op; 118 unsigned long irq_flags; 119 evtchn_port_t port = pdev->evtchn; 120 unsigned irq = pdev->irq; 121 s64 ns, ns_timeout; 122 123 spin_lock_irqsave(&pdev->sh_info_lock, irq_flags); 124 125 memcpy(active_op, op, sizeof(struct xen_pci_op)); 126 127 /* Go */ 128 wmb(); 129 set_bit(_XEN_PCIF_active, (unsigned long *)&pdev->sh_info->flags); 130 notify_remote_via_evtchn(port); 131 132 /* 133 * We set a poll timeout of 3 seconds but give up on return after 134 * 2 seconds. It is better to time out too late rather than too early 135 * (in the latter case we end up continually re-executing poll() with a 136 * timeout in the past). 1s difference gives plenty of slack for error. 137 */ 138 ns_timeout = ktime_get_ns() + 2 * (s64)NSEC_PER_SEC; 139 140 xen_clear_irq_pending(irq); 141 142 while (test_bit(_XEN_PCIF_active, 143 (unsigned long *)&pdev->sh_info->flags)) { 144 xen_poll_irq_timeout(irq, jiffies + 3*HZ); 145 xen_clear_irq_pending(irq); 146 ns = ktime_get_ns(); 147 if (ns > ns_timeout) { 148 dev_err(&pdev->xdev->dev, 149 "pciback not responding!!!\n"); 150 clear_bit(_XEN_PCIF_active, 151 (unsigned long *)&pdev->sh_info->flags); 152 err = XEN_PCI_ERR_dev_not_found; 153 goto out; 154 } 155 } 156 157 /* 158 * We might lose backend service request since we 159 * reuse same evtchn with pci_conf backend response. So re-schedule 160 * aer pcifront service. 161 */ 162 if (test_bit(_XEN_PCIB_active, 163 (unsigned long *)&pdev->sh_info->flags)) { 164 dev_err(&pdev->xdev->dev, 165 "schedule aer pcifront service\n"); 166 schedule_pcifront_aer_op(pdev); 167 } 168 169 memcpy(op, active_op, sizeof(struct xen_pci_op)); 170 171 err = op->err; 172 out: 173 spin_unlock_irqrestore(&pdev->sh_info_lock, irq_flags); 174 return err; 175 } 176 177 /* Access to this function is spinlocked in drivers/pci/access.c */ 178 static int pcifront_bus_read(struct pci_bus *bus, unsigned int devfn, 179 int where, int size, u32 *val) 180 { 181 int err = 0; 182 struct xen_pci_op op = { 183 .cmd = XEN_PCI_OP_conf_read, 184 .domain = pci_domain_nr(bus), 185 .bus = bus->number, 186 .devfn = devfn, 187 .offset = where, 188 .size = size, 189 }; 190 struct pcifront_sd *sd = bus->sysdata; 191 struct pcifront_device *pdev = pcifront_get_pdev(sd); 192 193 if (verbose_request) 194 dev_info(&pdev->xdev->dev, 195 "read dev=%04x:%02x:%02x.%d - offset %x size %d\n", 196 pci_domain_nr(bus), bus->number, PCI_SLOT(devfn), 197 PCI_FUNC(devfn), where, size); 198 199 err = do_pci_op(pdev, &op); 200 201 if (likely(!err)) { 202 if (verbose_request) 203 dev_info(&pdev->xdev->dev, "read got back value %x\n", 204 op.value); 205 206 *val = op.value; 207 } else if (err == -ENODEV) { 208 /* No device here, pretend that it just returned 0 */ 209 err = 0; 210 *val = 0; 211 } 212 213 return errno_to_pcibios_err(err); 214 } 215 216 /* Access to this function is spinlocked in drivers/pci/access.c */ 217 static int pcifront_bus_write(struct pci_bus *bus, unsigned int devfn, 218 int where, int size, u32 val) 219 { 220 struct xen_pci_op op = { 221 .cmd = XEN_PCI_OP_conf_write, 222 .domain = pci_domain_nr(bus), 223 .bus = bus->number, 224 .devfn = devfn, 225 .offset = where, 226 .size = size, 227 .value = val, 228 }; 229 struct pcifront_sd *sd = bus->sysdata; 230 struct pcifront_device *pdev = pcifront_get_pdev(sd); 231 232 if (verbose_request) 233 dev_info(&pdev->xdev->dev, 234 "write dev=%04x:%02x:%02x.%d - " 235 "offset %x size %d val %x\n", 236 pci_domain_nr(bus), bus->number, 237 PCI_SLOT(devfn), PCI_FUNC(devfn), where, size, val); 238 239 return errno_to_pcibios_err(do_pci_op(pdev, &op)); 240 } 241 242 static struct pci_ops pcifront_bus_ops = { 243 .read = pcifront_bus_read, 244 .write = pcifront_bus_write, 245 }; 246 247 #ifdef CONFIG_PCI_MSI 248 static int pci_frontend_enable_msix(struct pci_dev *dev, 249 int vector[], int nvec) 250 { 251 int err; 252 int i; 253 struct xen_pci_op op = { 254 .cmd = XEN_PCI_OP_enable_msix, 255 .domain = pci_domain_nr(dev->bus), 256 .bus = dev->bus->number, 257 .devfn = dev->devfn, 258 .value = nvec, 259 }; 260 struct pcifront_sd *sd = dev->bus->sysdata; 261 struct pcifront_device *pdev = pcifront_get_pdev(sd); 262 struct msi_desc *entry; 263 264 if (nvec > SH_INFO_MAX_VEC) { 265 pci_err(dev, "too many vectors (0x%x) for PCI frontend:" 266 " Increase SH_INFO_MAX_VEC\n", nvec); 267 return -EINVAL; 268 } 269 270 i = 0; 271 for_each_pci_msi_entry(entry, dev) { 272 op.msix_entries[i].entry = entry->msi_attrib.entry_nr; 273 /* Vector is useless at this point. */ 274 op.msix_entries[i].vector = -1; 275 i++; 276 } 277 278 err = do_pci_op(pdev, &op); 279 280 if (likely(!err)) { 281 if (likely(!op.value)) { 282 /* we get the result */ 283 for (i = 0; i < nvec; i++) { 284 if (op.msix_entries[i].vector <= 0) { 285 pci_warn(dev, "MSI-X entry %d is invalid: %d!\n", 286 i, op.msix_entries[i].vector); 287 err = -EINVAL; 288 vector[i] = -1; 289 continue; 290 } 291 vector[i] = op.msix_entries[i].vector; 292 } 293 } else { 294 printk(KERN_DEBUG "enable msix get value %x\n", 295 op.value); 296 err = op.value; 297 } 298 } else { 299 pci_err(dev, "enable msix get err %x\n", err); 300 } 301 return err; 302 } 303 304 static void pci_frontend_disable_msix(struct pci_dev *dev) 305 { 306 int err; 307 struct xen_pci_op op = { 308 .cmd = XEN_PCI_OP_disable_msix, 309 .domain = pci_domain_nr(dev->bus), 310 .bus = dev->bus->number, 311 .devfn = dev->devfn, 312 }; 313 struct pcifront_sd *sd = dev->bus->sysdata; 314 struct pcifront_device *pdev = pcifront_get_pdev(sd); 315 316 err = do_pci_op(pdev, &op); 317 318 /* What should do for error ? */ 319 if (err) 320 pci_err(dev, "pci_disable_msix get err %x\n", err); 321 } 322 323 static int pci_frontend_enable_msi(struct pci_dev *dev, int vector[]) 324 { 325 int err; 326 struct xen_pci_op op = { 327 .cmd = XEN_PCI_OP_enable_msi, 328 .domain = pci_domain_nr(dev->bus), 329 .bus = dev->bus->number, 330 .devfn = dev->devfn, 331 }; 332 struct pcifront_sd *sd = dev->bus->sysdata; 333 struct pcifront_device *pdev = pcifront_get_pdev(sd); 334 335 err = do_pci_op(pdev, &op); 336 if (likely(!err)) { 337 vector[0] = op.value; 338 if (op.value <= 0) { 339 pci_warn(dev, "MSI entry is invalid: %d!\n", 340 op.value); 341 err = -EINVAL; 342 vector[0] = -1; 343 } 344 } else { 345 pci_err(dev, "pci frontend enable msi failed for dev " 346 "%x:%x\n", op.bus, op.devfn); 347 err = -EINVAL; 348 } 349 return err; 350 } 351 352 static void pci_frontend_disable_msi(struct pci_dev *dev) 353 { 354 int err; 355 struct xen_pci_op op = { 356 .cmd = XEN_PCI_OP_disable_msi, 357 .domain = pci_domain_nr(dev->bus), 358 .bus = dev->bus->number, 359 .devfn = dev->devfn, 360 }; 361 struct pcifront_sd *sd = dev->bus->sysdata; 362 struct pcifront_device *pdev = pcifront_get_pdev(sd); 363 364 err = do_pci_op(pdev, &op); 365 if (err == XEN_PCI_ERR_dev_not_found) { 366 /* XXX No response from backend, what shall we do? */ 367 printk(KERN_DEBUG "get no response from backend for disable MSI\n"); 368 return; 369 } 370 if (err) 371 /* how can pciback notify us fail? */ 372 printk(KERN_DEBUG "get fake response frombackend\n"); 373 } 374 375 static struct xen_pci_frontend_ops pci_frontend_ops = { 376 .enable_msi = pci_frontend_enable_msi, 377 .disable_msi = pci_frontend_disable_msi, 378 .enable_msix = pci_frontend_enable_msix, 379 .disable_msix = pci_frontend_disable_msix, 380 }; 381 382 static void pci_frontend_registrar(int enable) 383 { 384 if (enable) 385 xen_pci_frontend = &pci_frontend_ops; 386 else 387 xen_pci_frontend = NULL; 388 }; 389 #else 390 static inline void pci_frontend_registrar(int enable) { }; 391 #endif /* CONFIG_PCI_MSI */ 392 393 /* Claim resources for the PCI frontend as-is, backend won't allow changes */ 394 static int pcifront_claim_resource(struct pci_dev *dev, void *data) 395 { 396 struct pcifront_device *pdev = data; 397 int i; 398 struct resource *r; 399 400 for (i = 0; i < PCI_NUM_RESOURCES; i++) { 401 r = &dev->resource[i]; 402 403 if (!r->parent && r->start && r->flags) { 404 dev_info(&pdev->xdev->dev, "claiming resource %s/%d\n", 405 pci_name(dev), i); 406 if (pci_claim_resource(dev, i)) { 407 dev_err(&pdev->xdev->dev, "Could not claim resource %s/%d! " 408 "Device offline. Try using e820_host=1 in the guest config.\n", 409 pci_name(dev), i); 410 } 411 } 412 } 413 414 return 0; 415 } 416 417 static int pcifront_scan_bus(struct pcifront_device *pdev, 418 unsigned int domain, unsigned int bus, 419 struct pci_bus *b) 420 { 421 struct pci_dev *d; 422 unsigned int devfn; 423 424 /* Scan the bus for functions and add. 425 * We omit handling of PCI bridge attachment because pciback prevents 426 * bridges from being exported. 427 */ 428 for (devfn = 0; devfn < 0x100; devfn++) { 429 d = pci_get_slot(b, devfn); 430 if (d) { 431 /* Device is already known. */ 432 pci_dev_put(d); 433 continue; 434 } 435 436 d = pci_scan_single_device(b, devfn); 437 if (d) 438 dev_info(&pdev->xdev->dev, "New device on " 439 "%04x:%02x:%02x.%d found.\n", domain, bus, 440 PCI_SLOT(devfn), PCI_FUNC(devfn)); 441 } 442 443 return 0; 444 } 445 446 static int pcifront_scan_root(struct pcifront_device *pdev, 447 unsigned int domain, unsigned int bus) 448 { 449 struct pci_bus *b; 450 LIST_HEAD(resources); 451 struct pcifront_sd *sd = NULL; 452 struct pci_bus_entry *bus_entry = NULL; 453 int err = 0; 454 static struct resource busn_res = { 455 .start = 0, 456 .end = 255, 457 .flags = IORESOURCE_BUS, 458 }; 459 460 #ifndef CONFIG_PCI_DOMAINS 461 if (domain != 0) { 462 dev_err(&pdev->xdev->dev, 463 "PCI Root in non-zero PCI Domain! domain=%d\n", domain); 464 dev_err(&pdev->xdev->dev, 465 "Please compile with CONFIG_PCI_DOMAINS\n"); 466 err = -EINVAL; 467 goto err_out; 468 } 469 #endif 470 471 dev_info(&pdev->xdev->dev, "Creating PCI Frontend Bus %04x:%02x\n", 472 domain, bus); 473 474 bus_entry = kzalloc(sizeof(*bus_entry), GFP_KERNEL); 475 sd = kzalloc(sizeof(*sd), GFP_KERNEL); 476 if (!bus_entry || !sd) { 477 err = -ENOMEM; 478 goto err_out; 479 } 480 pci_add_resource(&resources, &ioport_resource); 481 pci_add_resource(&resources, &iomem_resource); 482 pci_add_resource(&resources, &busn_res); 483 pcifront_init_sd(sd, domain, bus, pdev); 484 485 pci_lock_rescan_remove(); 486 487 b = pci_scan_root_bus(&pdev->xdev->dev, bus, 488 &pcifront_bus_ops, sd, &resources); 489 if (!b) { 490 dev_err(&pdev->xdev->dev, 491 "Error creating PCI Frontend Bus!\n"); 492 err = -ENOMEM; 493 pci_unlock_rescan_remove(); 494 pci_free_resource_list(&resources); 495 goto err_out; 496 } 497 498 bus_entry->bus = b; 499 500 list_add(&bus_entry->list, &pdev->root_buses); 501 502 /* pci_scan_root_bus skips devices which do not have a 503 * devfn==0. The pcifront_scan_bus enumerates all devfn. */ 504 err = pcifront_scan_bus(pdev, domain, bus, b); 505 506 /* Claim resources before going "live" with our devices */ 507 pci_walk_bus(b, pcifront_claim_resource, pdev); 508 509 /* Create SysFS and notify udev of the devices. Aka: "going live" */ 510 pci_bus_add_devices(b); 511 512 pci_unlock_rescan_remove(); 513 return err; 514 515 err_out: 516 kfree(bus_entry); 517 kfree(sd); 518 519 return err; 520 } 521 522 static int pcifront_rescan_root(struct pcifront_device *pdev, 523 unsigned int domain, unsigned int bus) 524 { 525 int err; 526 struct pci_bus *b; 527 528 #ifndef CONFIG_PCI_DOMAINS 529 if (domain != 0) { 530 dev_err(&pdev->xdev->dev, 531 "PCI Root in non-zero PCI Domain! domain=%d\n", domain); 532 dev_err(&pdev->xdev->dev, 533 "Please compile with CONFIG_PCI_DOMAINS\n"); 534 return -EINVAL; 535 } 536 #endif 537 538 dev_info(&pdev->xdev->dev, "Rescanning PCI Frontend Bus %04x:%02x\n", 539 domain, bus); 540 541 b = pci_find_bus(domain, bus); 542 if (!b) 543 /* If the bus is unknown, create it. */ 544 return pcifront_scan_root(pdev, domain, bus); 545 546 err = pcifront_scan_bus(pdev, domain, bus, b); 547 548 /* Claim resources before going "live" with our devices */ 549 pci_walk_bus(b, pcifront_claim_resource, pdev); 550 551 /* Create SysFS and notify udev of the devices. Aka: "going live" */ 552 pci_bus_add_devices(b); 553 554 return err; 555 } 556 557 static void free_root_bus_devs(struct pci_bus *bus) 558 { 559 struct pci_dev *dev; 560 561 while (!list_empty(&bus->devices)) { 562 dev = container_of(bus->devices.next, struct pci_dev, 563 bus_list); 564 pci_dbg(dev, "removing device\n"); 565 pci_stop_and_remove_bus_device(dev); 566 } 567 } 568 569 static void pcifront_free_roots(struct pcifront_device *pdev) 570 { 571 struct pci_bus_entry *bus_entry, *t; 572 573 dev_dbg(&pdev->xdev->dev, "cleaning up root buses\n"); 574 575 pci_lock_rescan_remove(); 576 list_for_each_entry_safe(bus_entry, t, &pdev->root_buses, list) { 577 list_del(&bus_entry->list); 578 579 free_root_bus_devs(bus_entry->bus); 580 581 kfree(bus_entry->bus->sysdata); 582 583 device_unregister(bus_entry->bus->bridge); 584 pci_remove_bus(bus_entry->bus); 585 586 kfree(bus_entry); 587 } 588 pci_unlock_rescan_remove(); 589 } 590 591 static pci_ers_result_t pcifront_common_process(int cmd, 592 struct pcifront_device *pdev, 593 pci_channel_state_t state) 594 { 595 pci_ers_result_t result; 596 struct pci_driver *pdrv; 597 int bus = pdev->sh_info->aer_op.bus; 598 int devfn = pdev->sh_info->aer_op.devfn; 599 int domain = pdev->sh_info->aer_op.domain; 600 struct pci_dev *pcidev; 601 int flag = 0; 602 603 dev_dbg(&pdev->xdev->dev, 604 "pcifront AER process: cmd %x (bus:%x, devfn%x)", 605 cmd, bus, devfn); 606 result = PCI_ERS_RESULT_NONE; 607 608 pcidev = pci_get_domain_bus_and_slot(domain, bus, devfn); 609 if (!pcidev || !pcidev->driver) { 610 dev_err(&pdev->xdev->dev, "device or AER driver is NULL\n"); 611 pci_dev_put(pcidev); 612 return result; 613 } 614 pdrv = pcidev->driver; 615 616 if (pdrv) { 617 if (pdrv->err_handler && pdrv->err_handler->error_detected) { 618 pci_dbg(pcidev, "trying to call AER service\n"); 619 if (pcidev) { 620 flag = 1; 621 switch (cmd) { 622 case XEN_PCI_OP_aer_detected: 623 result = pdrv->err_handler-> 624 error_detected(pcidev, state); 625 break; 626 case XEN_PCI_OP_aer_mmio: 627 result = pdrv->err_handler-> 628 mmio_enabled(pcidev); 629 break; 630 case XEN_PCI_OP_aer_slotreset: 631 result = pdrv->err_handler-> 632 slot_reset(pcidev); 633 break; 634 case XEN_PCI_OP_aer_resume: 635 pdrv->err_handler->resume(pcidev); 636 break; 637 default: 638 dev_err(&pdev->xdev->dev, 639 "bad request in aer recovery " 640 "operation!\n"); 641 642 } 643 } 644 } 645 } 646 if (!flag) 647 result = PCI_ERS_RESULT_NONE; 648 649 return result; 650 } 651 652 653 static void pcifront_do_aer(struct work_struct *data) 654 { 655 struct pcifront_device *pdev = 656 container_of(data, struct pcifront_device, op_work); 657 int cmd = pdev->sh_info->aer_op.cmd; 658 pci_channel_state_t state = 659 (pci_channel_state_t)pdev->sh_info->aer_op.err; 660 661 /*If a pci_conf op is in progress, 662 we have to wait until it is done before service aer op*/ 663 dev_dbg(&pdev->xdev->dev, 664 "pcifront service aer bus %x devfn %x\n", 665 pdev->sh_info->aer_op.bus, pdev->sh_info->aer_op.devfn); 666 667 pdev->sh_info->aer_op.err = pcifront_common_process(cmd, pdev, state); 668 669 /* Post the operation to the guest. */ 670 wmb(); 671 clear_bit(_XEN_PCIB_active, (unsigned long *)&pdev->sh_info->flags); 672 notify_remote_via_evtchn(pdev->evtchn); 673 674 /*in case of we lost an aer request in four lines time_window*/ 675 smp_mb__before_atomic(); 676 clear_bit(_PDEVB_op_active, &pdev->flags); 677 smp_mb__after_atomic(); 678 679 schedule_pcifront_aer_op(pdev); 680 681 } 682 683 static irqreturn_t pcifront_handler_aer(int irq, void *dev) 684 { 685 struct pcifront_device *pdev = dev; 686 schedule_pcifront_aer_op(pdev); 687 return IRQ_HANDLED; 688 } 689 static int pcifront_connect_and_init_dma(struct pcifront_device *pdev) 690 { 691 int err = 0; 692 693 spin_lock(&pcifront_dev_lock); 694 695 if (!pcifront_dev) { 696 dev_info(&pdev->xdev->dev, "Installing PCI frontend\n"); 697 pcifront_dev = pdev; 698 } else 699 err = -EEXIST; 700 701 spin_unlock(&pcifront_dev_lock); 702 703 if (!err && !swiotlb_nr_tbl()) { 704 err = pci_xen_swiotlb_init_late(); 705 if (err) 706 dev_err(&pdev->xdev->dev, "Could not setup SWIOTLB!\n"); 707 } 708 return err; 709 } 710 711 static void pcifront_disconnect(struct pcifront_device *pdev) 712 { 713 spin_lock(&pcifront_dev_lock); 714 715 if (pdev == pcifront_dev) { 716 dev_info(&pdev->xdev->dev, 717 "Disconnecting PCI Frontend Buses\n"); 718 pcifront_dev = NULL; 719 } 720 721 spin_unlock(&pcifront_dev_lock); 722 } 723 static struct pcifront_device *alloc_pdev(struct xenbus_device *xdev) 724 { 725 struct pcifront_device *pdev; 726 727 pdev = kzalloc(sizeof(struct pcifront_device), GFP_KERNEL); 728 if (pdev == NULL) 729 goto out; 730 731 pdev->sh_info = 732 (struct xen_pci_sharedinfo *)__get_free_page(GFP_KERNEL); 733 if (pdev->sh_info == NULL) { 734 kfree(pdev); 735 pdev = NULL; 736 goto out; 737 } 738 pdev->sh_info->flags = 0; 739 740 /*Flag for registering PV AER handler*/ 741 set_bit(_XEN_PCIB_AERHANDLER, (void *)&pdev->sh_info->flags); 742 743 dev_set_drvdata(&xdev->dev, pdev); 744 pdev->xdev = xdev; 745 746 INIT_LIST_HEAD(&pdev->root_buses); 747 748 spin_lock_init(&pdev->sh_info_lock); 749 750 pdev->evtchn = INVALID_EVTCHN; 751 pdev->gnt_ref = INVALID_GRANT_REF; 752 pdev->irq = -1; 753 754 INIT_WORK(&pdev->op_work, pcifront_do_aer); 755 756 dev_dbg(&xdev->dev, "Allocated pdev @ 0x%p pdev->sh_info @ 0x%p\n", 757 pdev, pdev->sh_info); 758 out: 759 return pdev; 760 } 761 762 static void free_pdev(struct pcifront_device *pdev) 763 { 764 dev_dbg(&pdev->xdev->dev, "freeing pdev @ 0x%p\n", pdev); 765 766 pcifront_free_roots(pdev); 767 768 cancel_work_sync(&pdev->op_work); 769 770 if (pdev->irq >= 0) 771 unbind_from_irqhandler(pdev->irq, pdev); 772 773 if (pdev->evtchn != INVALID_EVTCHN) 774 xenbus_free_evtchn(pdev->xdev, pdev->evtchn); 775 776 if (pdev->gnt_ref != INVALID_GRANT_REF) 777 gnttab_end_foreign_access(pdev->gnt_ref, 0 /* r/w page */, 778 (unsigned long)pdev->sh_info); 779 else 780 free_page((unsigned long)pdev->sh_info); 781 782 dev_set_drvdata(&pdev->xdev->dev, NULL); 783 784 kfree(pdev); 785 } 786 787 static int pcifront_publish_info(struct pcifront_device *pdev) 788 { 789 int err = 0; 790 struct xenbus_transaction trans; 791 grant_ref_t gref; 792 793 err = xenbus_grant_ring(pdev->xdev, pdev->sh_info, 1, &gref); 794 if (err < 0) 795 goto out; 796 797 pdev->gnt_ref = gref; 798 799 err = xenbus_alloc_evtchn(pdev->xdev, &pdev->evtchn); 800 if (err) 801 goto out; 802 803 err = bind_evtchn_to_irqhandler(pdev->evtchn, pcifront_handler_aer, 804 0, "pcifront", pdev); 805 806 if (err < 0) 807 return err; 808 809 pdev->irq = err; 810 811 do_publish: 812 err = xenbus_transaction_start(&trans); 813 if (err) { 814 xenbus_dev_fatal(pdev->xdev, err, 815 "Error writing configuration for backend " 816 "(start transaction)"); 817 goto out; 818 } 819 820 err = xenbus_printf(trans, pdev->xdev->nodename, 821 "pci-op-ref", "%u", pdev->gnt_ref); 822 if (!err) 823 err = xenbus_printf(trans, pdev->xdev->nodename, 824 "event-channel", "%u", pdev->evtchn); 825 if (!err) 826 err = xenbus_printf(trans, pdev->xdev->nodename, 827 "magic", XEN_PCI_MAGIC); 828 829 if (err) { 830 xenbus_transaction_end(trans, 1); 831 xenbus_dev_fatal(pdev->xdev, err, 832 "Error writing configuration for backend"); 833 goto out; 834 } else { 835 err = xenbus_transaction_end(trans, 0); 836 if (err == -EAGAIN) 837 goto do_publish; 838 else if (err) { 839 xenbus_dev_fatal(pdev->xdev, err, 840 "Error completing transaction " 841 "for backend"); 842 goto out; 843 } 844 } 845 846 xenbus_switch_state(pdev->xdev, XenbusStateInitialised); 847 848 dev_dbg(&pdev->xdev->dev, "publishing successful!\n"); 849 850 out: 851 return err; 852 } 853 854 static int pcifront_try_connect(struct pcifront_device *pdev) 855 { 856 int err = -EFAULT; 857 int i, num_roots, len; 858 char str[64]; 859 unsigned int domain, bus; 860 861 862 /* Only connect once */ 863 if (xenbus_read_driver_state(pdev->xdev->nodename) != 864 XenbusStateInitialised) 865 goto out; 866 867 err = pcifront_connect_and_init_dma(pdev); 868 if (err && err != -EEXIST) { 869 xenbus_dev_fatal(pdev->xdev, err, 870 "Error setting up PCI Frontend"); 871 goto out; 872 } 873 874 err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, 875 "root_num", "%d", &num_roots); 876 if (err == -ENOENT) { 877 xenbus_dev_error(pdev->xdev, err, 878 "No PCI Roots found, trying 0000:00"); 879 err = pcifront_scan_root(pdev, 0, 0); 880 if (err) { 881 xenbus_dev_fatal(pdev->xdev, err, 882 "Error scanning PCI root 0000:00"); 883 goto out; 884 } 885 num_roots = 0; 886 } else if (err != 1) { 887 if (err == 0) 888 err = -EINVAL; 889 xenbus_dev_fatal(pdev->xdev, err, 890 "Error reading number of PCI roots"); 891 goto out; 892 } 893 894 for (i = 0; i < num_roots; i++) { 895 len = snprintf(str, sizeof(str), "root-%d", i); 896 if (unlikely(len >= (sizeof(str) - 1))) { 897 err = -ENOMEM; 898 goto out; 899 } 900 901 err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str, 902 "%x:%x", &domain, &bus); 903 if (err != 2) { 904 if (err >= 0) 905 err = -EINVAL; 906 xenbus_dev_fatal(pdev->xdev, err, 907 "Error reading PCI root %d", i); 908 goto out; 909 } 910 911 err = pcifront_scan_root(pdev, domain, bus); 912 if (err) { 913 xenbus_dev_fatal(pdev->xdev, err, 914 "Error scanning PCI root %04x:%02x", 915 domain, bus); 916 goto out; 917 } 918 } 919 920 err = xenbus_switch_state(pdev->xdev, XenbusStateConnected); 921 922 out: 923 return err; 924 } 925 926 static int pcifront_try_disconnect(struct pcifront_device *pdev) 927 { 928 int err = 0; 929 enum xenbus_state prev_state; 930 931 932 prev_state = xenbus_read_driver_state(pdev->xdev->nodename); 933 934 if (prev_state >= XenbusStateClosing) 935 goto out; 936 937 if (prev_state == XenbusStateConnected) { 938 pcifront_free_roots(pdev); 939 pcifront_disconnect(pdev); 940 } 941 942 err = xenbus_switch_state(pdev->xdev, XenbusStateClosed); 943 944 out: 945 946 return err; 947 } 948 949 static int pcifront_attach_devices(struct pcifront_device *pdev) 950 { 951 int err = -EFAULT; 952 int i, num_roots, len; 953 unsigned int domain, bus; 954 char str[64]; 955 956 if (xenbus_read_driver_state(pdev->xdev->nodename) != 957 XenbusStateReconfiguring) 958 goto out; 959 960 err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, 961 "root_num", "%d", &num_roots); 962 if (err == -ENOENT) { 963 xenbus_dev_error(pdev->xdev, err, 964 "No PCI Roots found, trying 0000:00"); 965 err = pcifront_rescan_root(pdev, 0, 0); 966 if (err) { 967 xenbus_dev_fatal(pdev->xdev, err, 968 "Error scanning PCI root 0000:00"); 969 goto out; 970 } 971 num_roots = 0; 972 } else if (err != 1) { 973 if (err == 0) 974 err = -EINVAL; 975 xenbus_dev_fatal(pdev->xdev, err, 976 "Error reading number of PCI roots"); 977 goto out; 978 } 979 980 for (i = 0; i < num_roots; i++) { 981 len = snprintf(str, sizeof(str), "root-%d", i); 982 if (unlikely(len >= (sizeof(str) - 1))) { 983 err = -ENOMEM; 984 goto out; 985 } 986 987 err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str, 988 "%x:%x", &domain, &bus); 989 if (err != 2) { 990 if (err >= 0) 991 err = -EINVAL; 992 xenbus_dev_fatal(pdev->xdev, err, 993 "Error reading PCI root %d", i); 994 goto out; 995 } 996 997 err = pcifront_rescan_root(pdev, domain, bus); 998 if (err) { 999 xenbus_dev_fatal(pdev->xdev, err, 1000 "Error scanning PCI root %04x:%02x", 1001 domain, bus); 1002 goto out; 1003 } 1004 } 1005 1006 xenbus_switch_state(pdev->xdev, XenbusStateConnected); 1007 1008 out: 1009 return err; 1010 } 1011 1012 static int pcifront_detach_devices(struct pcifront_device *pdev) 1013 { 1014 int err = 0; 1015 int i, num_devs; 1016 unsigned int domain, bus, slot, func; 1017 struct pci_dev *pci_dev; 1018 char str[64]; 1019 1020 if (xenbus_read_driver_state(pdev->xdev->nodename) != 1021 XenbusStateConnected) 1022 goto out; 1023 1024 err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, "num_devs", "%d", 1025 &num_devs); 1026 if (err != 1) { 1027 if (err >= 0) 1028 err = -EINVAL; 1029 xenbus_dev_fatal(pdev->xdev, err, 1030 "Error reading number of PCI devices"); 1031 goto out; 1032 } 1033 1034 /* Find devices being detached and remove them. */ 1035 for (i = 0; i < num_devs; i++) { 1036 int l, state; 1037 l = snprintf(str, sizeof(str), "state-%d", i); 1038 if (unlikely(l >= (sizeof(str) - 1))) { 1039 err = -ENOMEM; 1040 goto out; 1041 } 1042 state = xenbus_read_unsigned(pdev->xdev->otherend, str, 1043 XenbusStateUnknown); 1044 1045 if (state != XenbusStateClosing) 1046 continue; 1047 1048 /* Remove device. */ 1049 l = snprintf(str, sizeof(str), "vdev-%d", i); 1050 if (unlikely(l >= (sizeof(str) - 1))) { 1051 err = -ENOMEM; 1052 goto out; 1053 } 1054 err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str, 1055 "%x:%x:%x.%x", &domain, &bus, &slot, &func); 1056 if (err != 4) { 1057 if (err >= 0) 1058 err = -EINVAL; 1059 xenbus_dev_fatal(pdev->xdev, err, 1060 "Error reading PCI device %d", i); 1061 goto out; 1062 } 1063 1064 pci_dev = pci_get_domain_bus_and_slot(domain, bus, 1065 PCI_DEVFN(slot, func)); 1066 if (!pci_dev) { 1067 dev_dbg(&pdev->xdev->dev, 1068 "Cannot get PCI device %04x:%02x:%02x.%d\n", 1069 domain, bus, slot, func); 1070 continue; 1071 } 1072 pci_lock_rescan_remove(); 1073 pci_stop_and_remove_bus_device(pci_dev); 1074 pci_dev_put(pci_dev); 1075 pci_unlock_rescan_remove(); 1076 1077 dev_dbg(&pdev->xdev->dev, 1078 "PCI device %04x:%02x:%02x.%d removed.\n", 1079 domain, bus, slot, func); 1080 } 1081 1082 err = xenbus_switch_state(pdev->xdev, XenbusStateReconfiguring); 1083 1084 out: 1085 return err; 1086 } 1087 1088 static void __ref pcifront_backend_changed(struct xenbus_device *xdev, 1089 enum xenbus_state be_state) 1090 { 1091 struct pcifront_device *pdev = dev_get_drvdata(&xdev->dev); 1092 1093 switch (be_state) { 1094 case XenbusStateUnknown: 1095 case XenbusStateInitialising: 1096 case XenbusStateInitWait: 1097 case XenbusStateInitialised: 1098 break; 1099 1100 case XenbusStateConnected: 1101 pcifront_try_connect(pdev); 1102 break; 1103 1104 case XenbusStateClosed: 1105 if (xdev->state == XenbusStateClosed) 1106 break; 1107 /* Missed the backend's CLOSING state -- fallthrough */ 1108 case XenbusStateClosing: 1109 dev_warn(&xdev->dev, "backend going away!\n"); 1110 pcifront_try_disconnect(pdev); 1111 break; 1112 1113 case XenbusStateReconfiguring: 1114 pcifront_detach_devices(pdev); 1115 break; 1116 1117 case XenbusStateReconfigured: 1118 pcifront_attach_devices(pdev); 1119 break; 1120 } 1121 } 1122 1123 static int pcifront_xenbus_probe(struct xenbus_device *xdev, 1124 const struct xenbus_device_id *id) 1125 { 1126 int err = 0; 1127 struct pcifront_device *pdev = alloc_pdev(xdev); 1128 1129 if (pdev == NULL) { 1130 err = -ENOMEM; 1131 xenbus_dev_fatal(xdev, err, 1132 "Error allocating pcifront_device struct"); 1133 goto out; 1134 } 1135 1136 err = pcifront_publish_info(pdev); 1137 if (err) 1138 free_pdev(pdev); 1139 1140 out: 1141 return err; 1142 } 1143 1144 static int pcifront_xenbus_remove(struct xenbus_device *xdev) 1145 { 1146 struct pcifront_device *pdev = dev_get_drvdata(&xdev->dev); 1147 if (pdev) 1148 free_pdev(pdev); 1149 1150 return 0; 1151 } 1152 1153 static const struct xenbus_device_id xenpci_ids[] = { 1154 {"pci"}, 1155 {""}, 1156 }; 1157 1158 static struct xenbus_driver xenpci_driver = { 1159 .name = "pcifront", 1160 .ids = xenpci_ids, 1161 .probe = pcifront_xenbus_probe, 1162 .remove = pcifront_xenbus_remove, 1163 .otherend_changed = pcifront_backend_changed, 1164 }; 1165 1166 static int __init pcifront_init(void) 1167 { 1168 if (!xen_pv_domain() || xen_initial_domain()) 1169 return -ENODEV; 1170 1171 if (!xen_has_pv_devices()) 1172 return -ENODEV; 1173 1174 pci_frontend_registrar(1 /* enable */); 1175 1176 return xenbus_register_frontend(&xenpci_driver); 1177 } 1178 1179 static void __exit pcifront_cleanup(void) 1180 { 1181 xenbus_unregister_driver(&xenpci_driver); 1182 pci_frontend_registrar(0 /* disable */); 1183 } 1184 module_init(pcifront_init); 1185 module_exit(pcifront_cleanup); 1186 1187 MODULE_DESCRIPTION("Xen PCI passthrough frontend."); 1188 MODULE_LICENSE("GPL"); 1189 MODULE_ALIAS("xen:pci"); 1190