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#include <linux/ktime.h> 24#include <xen/platform_pci.h> 25 26#include <asm/xen/swiotlb-xen.h> 27#define INVALID_GRANT_REF (0) 28#define INVALID_EVTCHN (-1) 29 30struct pci_bus_entry { 31 struct list_head list; 32 struct pci_bus *bus; 33}; 34 35#define _PDEVB_op_active (0) 36#define PDEVB_op_active (1 << (_PDEVB_op_active)) 37 38struct pcifront_device { 39 struct xenbus_device *xdev; 40 struct list_head root_buses; 41 42 int evtchn; 43 int gnt_ref; 44 45 int irq; 46 47 /* Lock this when doing any operations in sh_info */ 48 spinlock_t sh_info_lock; 49 struct xen_pci_sharedinfo *sh_info; 50 struct work_struct op_work; 51 unsigned long flags; 52 53}; 54 55struct pcifront_sd { 56 struct pci_sysdata sd; 57 struct pcifront_device *pdev; 58}; 59 60static inline struct pcifront_device * 61pcifront_get_pdev(struct pcifront_sd *sd) 62{ 63 return sd->pdev; 64} 65 66static inline void pcifront_init_sd(struct pcifront_sd *sd, 67 unsigned int domain, unsigned int bus, 68 struct pcifront_device *pdev) 69{ 70 /* Because we do not expose that information via XenBus. */ 71 sd->sd.node = first_online_node; 72 sd->sd.domain = domain; 73 sd->pdev = pdev; 74} 75 76static DEFINE_SPINLOCK(pcifront_dev_lock); 77static struct pcifront_device *pcifront_dev; 78 79static int verbose_request; 80module_param(verbose_request, int, 0644); 81 82static int errno_to_pcibios_err(int errno) 83{ 84 switch (errno) { 85 case XEN_PCI_ERR_success: 86 return PCIBIOS_SUCCESSFUL; 87 88 case XEN_PCI_ERR_dev_not_found: 89 return PCIBIOS_DEVICE_NOT_FOUND; 90 91 case XEN_PCI_ERR_invalid_offset: 92 case XEN_PCI_ERR_op_failed: 93 return PCIBIOS_BAD_REGISTER_NUMBER; 94 95 case XEN_PCI_ERR_not_implemented: 96 return PCIBIOS_FUNC_NOT_SUPPORTED; 97 98 case XEN_PCI_ERR_access_denied: 99 return PCIBIOS_SET_FAILED; 100 } 101 return errno; 102} 103 104static inline void schedule_pcifront_aer_op(struct pcifront_device *pdev) 105{ 106 if (test_bit(_XEN_PCIB_active, (unsigned long *)&pdev->sh_info->flags) 107 && !test_and_set_bit(_PDEVB_op_active, &pdev->flags)) { 108 dev_dbg(&pdev->xdev->dev, "schedule aer frontend job\n"); 109 schedule_work(&pdev->op_work); 110 } 111} 112 113static int do_pci_op(struct pcifront_device *pdev, struct xen_pci_op *op) 114{ 115 int err = 0; 116 struct xen_pci_op *active_op = &pdev->sh_info->op; 117 unsigned long irq_flags; 118 evtchn_port_t port = pdev->evtchn; 119 unsigned irq = pdev->irq; 120 s64 ns, ns_timeout; 121 122 spin_lock_irqsave(&pdev->sh_info_lock, irq_flags); 123 124 memcpy(active_op, op, sizeof(struct xen_pci_op)); 125 126 /* Go */ 127 wmb(); 128 set_bit(_XEN_PCIF_active, (unsigned long *)&pdev->sh_info->flags); 129 notify_remote_via_evtchn(port); 130 131 /* 132 * We set a poll timeout of 3 seconds but give up on return after 133 * 2 seconds. It is better to time out too late rather than too early 134 * (in the latter case we end up continually re-executing poll() with a 135 * timeout in the past). 1s difference gives plenty of slack for error. 136 */ 137 ns_timeout = ktime_get_ns() + 2 * (s64)NSEC_PER_SEC; 138 139 xen_clear_irq_pending(irq); 140 141 while (test_bit(_XEN_PCIF_active, 142 (unsigned long *)&pdev->sh_info->flags)) { 143 xen_poll_irq_timeout(irq, jiffies + 3*HZ); 144 xen_clear_irq_pending(irq); 145 ns = ktime_get_ns(); 146 if (ns > ns_timeout) { 147 dev_err(&pdev->xdev->dev, 148 "pciback not responding!!!\n"); 149 clear_bit(_XEN_PCIF_active, 150 (unsigned long *)&pdev->sh_info->flags); 151 err = XEN_PCI_ERR_dev_not_found; 152 goto out; 153 } 154 } 155 156 /* 157 * We might lose backend service request since we 158 * reuse same evtchn with pci_conf backend response. So re-schedule 159 * aer pcifront service. 160 */ 161 if (test_bit(_XEN_PCIB_active, 162 (unsigned long *)&pdev->sh_info->flags)) { 163 dev_err(&pdev->xdev->dev, 164 "schedule aer pcifront service\n"); 165 schedule_pcifront_aer_op(pdev); 166 } 167 168 memcpy(op, active_op, sizeof(struct xen_pci_op)); 169 170 err = op->err; 171out: 172 spin_unlock_irqrestore(&pdev->sh_info_lock, irq_flags); 173 return err; 174} 175 176/* Access to this function is spinlocked in drivers/pci/access.c */ 177static int pcifront_bus_read(struct pci_bus *bus, unsigned int devfn, 178 int where, int size, u32 *val) 179{ 180 int err = 0; 181 struct xen_pci_op op = { 182 .cmd = XEN_PCI_OP_conf_read, 183 .domain = pci_domain_nr(bus), 184 .bus = bus->number, 185 .devfn = devfn, 186 .offset = where, 187 .size = size, 188 }; 189 struct pcifront_sd *sd = bus->sysdata; 190 struct pcifront_device *pdev = pcifront_get_pdev(sd); 191 192 if (verbose_request) 193 dev_info(&pdev->xdev->dev, 194 "read dev=%04x:%02x:%02x.%d - offset %x size %d\n", 195 pci_domain_nr(bus), bus->number, PCI_SLOT(devfn), 196 PCI_FUNC(devfn), where, size); 197 198 err = do_pci_op(pdev, &op); 199 200 if (likely(!err)) { 201 if (verbose_request) 202 dev_info(&pdev->xdev->dev, "read got back value %x\n", 203 op.value); 204 205 *val = op.value; 206 } else if (err == -ENODEV) { 207 /* No device here, pretend that it just returned 0 */ 208 err = 0; 209 *val = 0; 210 } 211 212 return errno_to_pcibios_err(err); 213} 214 215/* Access to this function is spinlocked in drivers/pci/access.c */ 216static int pcifront_bus_write(struct pci_bus *bus, unsigned int devfn, 217 int where, int size, u32 val) 218{ 219 struct xen_pci_op op = { 220 .cmd = XEN_PCI_OP_conf_write, 221 .domain = pci_domain_nr(bus), 222 .bus = bus->number, 223 .devfn = devfn, 224 .offset = where, 225 .size = size, 226 .value = val, 227 }; 228 struct pcifront_sd *sd = bus->sysdata; 229 struct pcifront_device *pdev = pcifront_get_pdev(sd); 230 231 if (verbose_request) 232 dev_info(&pdev->xdev->dev, 233 "write dev=%04x:%02x:%02x.%d - " 234 "offset %x size %d val %x\n", 235 pci_domain_nr(bus), bus->number, 236 PCI_SLOT(devfn), PCI_FUNC(devfn), where, size, val); 237 238 return errno_to_pcibios_err(do_pci_op(pdev, &op)); 239} 240 241static struct pci_ops pcifront_bus_ops = { 242 .read = pcifront_bus_read, 243 .write = pcifront_bus_write, 244}; 245 246#ifdef CONFIG_PCI_MSI 247static int pci_frontend_enable_msix(struct pci_dev *dev, 248 int vector[], int nvec) 249{ 250 int err; 251 int i; 252 struct xen_pci_op op = { 253 .cmd = XEN_PCI_OP_enable_msix, 254 .domain = pci_domain_nr(dev->bus), 255 .bus = dev->bus->number, 256 .devfn = dev->devfn, 257 .value = nvec, 258 }; 259 struct pcifront_sd *sd = dev->bus->sysdata; 260 struct pcifront_device *pdev = pcifront_get_pdev(sd); 261 struct msi_desc *entry; 262 263 if (nvec > SH_INFO_MAX_VEC) { 264 dev_err(&dev->dev, "too much vector for pci frontend: %x." 265 " Increase SH_INFO_MAX_VEC.\n", nvec); 266 return -EINVAL; 267 } 268 269 i = 0; 270 for_each_pci_msi_entry(entry, dev) { 271 op.msix_entries[i].entry = entry->msi_attrib.entry_nr; 272 /* Vector is useless at this point. */ 273 op.msix_entries[i].vector = -1; 274 i++; 275 } 276 277 err = do_pci_op(pdev, &op); 278 279 if (likely(!err)) { 280 if (likely(!op.value)) { 281 /* we get the result */ 282 for (i = 0; i < nvec; i++) { 283 if (op.msix_entries[i].vector <= 0) { 284 dev_warn(&dev->dev, "MSI-X entry %d is invalid: %d!\n", 285 i, op.msix_entries[i].vector); 286 err = -EINVAL; 287 vector[i] = -1; 288 continue; 289 } 290 vector[i] = op.msix_entries[i].vector; 291 } 292 } else { 293 printk(KERN_DEBUG "enable msix get value %x\n", 294 op.value); 295 err = op.value; 296 } 297 } else { 298 dev_err(&dev->dev, "enable msix get err %x\n", err); 299 } 300 return err; 301} 302 303static void pci_frontend_disable_msix(struct pci_dev *dev) 304{ 305 int err; 306 struct xen_pci_op op = { 307 .cmd = XEN_PCI_OP_disable_msix, 308 .domain = pci_domain_nr(dev->bus), 309 .bus = dev->bus->number, 310 .devfn = dev->devfn, 311 }; 312 struct pcifront_sd *sd = dev->bus->sysdata; 313 struct pcifront_device *pdev = pcifront_get_pdev(sd); 314 315 err = do_pci_op(pdev, &op); 316 317 /* What should do for error ? */ 318 if (err) 319 dev_err(&dev->dev, "pci_disable_msix get err %x\n", err); 320} 321 322static int pci_frontend_enable_msi(struct pci_dev *dev, int vector[]) 323{ 324 int err; 325 struct xen_pci_op op = { 326 .cmd = XEN_PCI_OP_enable_msi, 327 .domain = pci_domain_nr(dev->bus), 328 .bus = dev->bus->number, 329 .devfn = dev->devfn, 330 }; 331 struct pcifront_sd *sd = dev->bus->sysdata; 332 struct pcifront_device *pdev = pcifront_get_pdev(sd); 333 334 err = do_pci_op(pdev, &op); 335 if (likely(!err)) { 336 vector[0] = op.value; 337 if (op.value <= 0) { 338 dev_warn(&dev->dev, "MSI entry is invalid: %d!\n", 339 op.value); 340 err = -EINVAL; 341 vector[0] = -1; 342 } 343 } else { 344 dev_err(&dev->dev, "pci frontend enable msi failed for dev " 345 "%x:%x\n", op.bus, op.devfn); 346 err = -EINVAL; 347 } 348 return err; 349} 350 351static void pci_frontend_disable_msi(struct pci_dev *dev) 352{ 353 int err; 354 struct xen_pci_op op = { 355 .cmd = XEN_PCI_OP_disable_msi, 356 .domain = pci_domain_nr(dev->bus), 357 .bus = dev->bus->number, 358 .devfn = dev->devfn, 359 }; 360 struct pcifront_sd *sd = dev->bus->sysdata; 361 struct pcifront_device *pdev = pcifront_get_pdev(sd); 362 363 err = do_pci_op(pdev, &op); 364 if (err == XEN_PCI_ERR_dev_not_found) { 365 /* XXX No response from backend, what shall we do? */ 366 printk(KERN_DEBUG "get no response from backend for disable MSI\n"); 367 return; 368 } 369 if (err) 370 /* how can pciback notify us fail? */ 371 printk(KERN_DEBUG "get fake response frombackend\n"); 372} 373 374static struct xen_pci_frontend_ops pci_frontend_ops = { 375 .enable_msi = pci_frontend_enable_msi, 376 .disable_msi = pci_frontend_disable_msi, 377 .enable_msix = pci_frontend_enable_msix, 378 .disable_msix = pci_frontend_disable_msix, 379}; 380 381static void pci_frontend_registrar(int enable) 382{ 383 if (enable) 384 xen_pci_frontend = &pci_frontend_ops; 385 else 386 xen_pci_frontend = NULL; 387}; 388#else 389static inline void pci_frontend_registrar(int enable) { }; 390#endif /* CONFIG_PCI_MSI */ 391 392/* Claim resources for the PCI frontend as-is, backend won't allow changes */ 393static int pcifront_claim_resource(struct pci_dev *dev, void *data) 394{ 395 struct pcifront_device *pdev = data; 396 int i; 397 struct resource *r; 398 399 for (i = 0; i < PCI_NUM_RESOURCES; i++) { 400 r = &dev->resource[i]; 401 402 if (!r->parent && r->start && r->flags) { 403 dev_info(&pdev->xdev->dev, "claiming resource %s/%d\n", 404 pci_name(dev), i); 405 if (pci_claim_resource(dev, i)) { 406 dev_err(&pdev->xdev->dev, "Could not claim resource %s/%d! " 407 "Device offline. Try using e820_host=1 in the guest config.\n", 408 pci_name(dev), i); 409 } 410 } 411 } 412 413 return 0; 414} 415 416static int pcifront_scan_bus(struct pcifront_device *pdev, 417 unsigned int domain, unsigned int bus, 418 struct pci_bus *b) 419{ 420 struct pci_dev *d; 421 unsigned int devfn; 422 423 /* Scan the bus for functions and add. 424 * We omit handling of PCI bridge attachment because pciback prevents 425 * bridges from being exported. 426 */ 427 for (devfn = 0; devfn < 0x100; devfn++) { 428 d = pci_get_slot(b, devfn); 429 if (d) { 430 /* Device is already known. */ 431 pci_dev_put(d); 432 continue; 433 } 434 435 d = pci_scan_single_device(b, devfn); 436 if (d) 437 dev_info(&pdev->xdev->dev, "New device on " 438 "%04x:%02x:%02x.%d found.\n", domain, bus, 439 PCI_SLOT(devfn), PCI_FUNC(devfn)); 440 } 441 442 return 0; 443} 444 445static int pcifront_scan_root(struct pcifront_device *pdev, 446 unsigned int domain, unsigned int bus) 447{ 448 struct pci_bus *b; 449 LIST_HEAD(resources); 450 struct pcifront_sd *sd = NULL; 451 struct pci_bus_entry *bus_entry = NULL; 452 int err = 0; 453 static struct resource busn_res = { 454 .start = 0, 455 .end = 255, 456 .flags = IORESOURCE_BUS, 457 }; 458 459#ifndef CONFIG_PCI_DOMAINS 460 if (domain != 0) { 461 dev_err(&pdev->xdev->dev, 462 "PCI Root in non-zero PCI Domain! domain=%d\n", domain); 463 dev_err(&pdev->xdev->dev, 464 "Please compile with CONFIG_PCI_DOMAINS\n"); 465 err = -EINVAL; 466 goto err_out; 467 } 468#endif 469 470 dev_info(&pdev->xdev->dev, "Creating PCI Frontend Bus %04x:%02x\n", 471 domain, bus); 472 473 bus_entry = kzalloc(sizeof(*bus_entry), GFP_KERNEL); 474 sd = kzalloc(sizeof(*sd), GFP_KERNEL); 475 if (!bus_entry || !sd) { 476 err = -ENOMEM; 477 goto err_out; 478 } 479 pci_add_resource(&resources, &ioport_resource); 480 pci_add_resource(&resources, &iomem_resource); 481 pci_add_resource(&resources, &busn_res); 482 pcifront_init_sd(sd, domain, bus, pdev); 483 484 pci_lock_rescan_remove(); 485 486 b = pci_scan_root_bus(&pdev->xdev->dev, bus, 487 &pcifront_bus_ops, sd, &resources); 488 if (!b) { 489 dev_err(&pdev->xdev->dev, 490 "Error creating PCI Frontend Bus!\n"); 491 err = -ENOMEM; 492 pci_unlock_rescan_remove(); 493 pci_free_resource_list(&resources); 494 goto err_out; 495 } 496 497 bus_entry->bus = b; 498 499 list_add(&bus_entry->list, &pdev->root_buses); 500 501 /* pci_scan_root_bus skips devices which do not have a 502 * devfn==0. The pcifront_scan_bus enumerates all devfn. */ 503 err = pcifront_scan_bus(pdev, domain, bus, b); 504 505 /* Claim resources before going "live" with our devices */ 506 pci_walk_bus(b, pcifront_claim_resource, pdev); 507 508 /* Create SysFS and notify udev of the devices. Aka: "going live" */ 509 pci_bus_add_devices(b); 510 511 pci_unlock_rescan_remove(); 512 return err; 513 514err_out: 515 kfree(bus_entry); 516 kfree(sd); 517 518 return err; 519} 520 521static int pcifront_rescan_root(struct pcifront_device *pdev, 522 unsigned int domain, unsigned int bus) 523{ 524 int err; 525 struct pci_bus *b; 526 527#ifndef CONFIG_PCI_DOMAINS 528 if (domain != 0) { 529 dev_err(&pdev->xdev->dev, 530 "PCI Root in non-zero PCI Domain! domain=%d\n", domain); 531 dev_err(&pdev->xdev->dev, 532 "Please compile with CONFIG_PCI_DOMAINS\n"); 533 return -EINVAL; 534 } 535#endif 536 537 dev_info(&pdev->xdev->dev, "Rescanning PCI Frontend Bus %04x:%02x\n", 538 domain, bus); 539 540 b = pci_find_bus(domain, bus); 541 if (!b) 542 /* If the bus is unknown, create it. */ 543 return pcifront_scan_root(pdev, domain, bus); 544 545 err = pcifront_scan_bus(pdev, domain, bus, b); 546 547 /* Claim resources before going "live" with our devices */ 548 pci_walk_bus(b, pcifront_claim_resource, pdev); 549 550 /* Create SysFS and notify udev of the devices. Aka: "going live" */ 551 pci_bus_add_devices(b); 552 553 return err; 554} 555 556static void free_root_bus_devs(struct pci_bus *bus) 557{ 558 struct pci_dev *dev; 559 560 while (!list_empty(&bus->devices)) { 561 dev = container_of(bus->devices.next, struct pci_dev, 562 bus_list); 563 dev_dbg(&dev->dev, "removing device\n"); 564 pci_stop_and_remove_bus_device(dev); 565 } 566} 567 568static void pcifront_free_roots(struct pcifront_device *pdev) 569{ 570 struct pci_bus_entry *bus_entry, *t; 571 572 dev_dbg(&pdev->xdev->dev, "cleaning up root buses\n"); 573 574 pci_lock_rescan_remove(); 575 list_for_each_entry_safe(bus_entry, t, &pdev->root_buses, list) { 576 list_del(&bus_entry->list); 577 578 free_root_bus_devs(bus_entry->bus); 579 580 kfree(bus_entry->bus->sysdata); 581 582 device_unregister(bus_entry->bus->bridge); 583 pci_remove_bus(bus_entry->bus); 584 585 kfree(bus_entry); 586 } 587 pci_unlock_rescan_remove(); 588} 589 590static pci_ers_result_t pcifront_common_process(int cmd, 591 struct pcifront_device *pdev, 592 pci_channel_state_t state) 593{ 594 pci_ers_result_t result; 595 struct pci_driver *pdrv; 596 int bus = pdev->sh_info->aer_op.bus; 597 int devfn = pdev->sh_info->aer_op.devfn; 598 struct pci_dev *pcidev; 599 int flag = 0; 600 601 dev_dbg(&pdev->xdev->dev, 602 "pcifront AER process: cmd %x (bus:%x, devfn%x)", 603 cmd, bus, devfn); 604 result = PCI_ERS_RESULT_NONE; 605 606 pcidev = pci_get_bus_and_slot(bus, devfn); 607 if (!pcidev || !pcidev->driver) { 608 dev_err(&pdev->xdev->dev, "device or AER driver is NULL\n"); 609 pci_dev_put(pcidev); 610 return result; 611 } 612 pdrv = pcidev->driver; 613 614 if (pdrv) { 615 if (pdrv->err_handler && pdrv->err_handler->error_detected) { 616 dev_dbg(&pcidev->dev, 617 "trying to call AER service\n"); 618 if (pcidev) { 619 flag = 1; 620 switch (cmd) { 621 case XEN_PCI_OP_aer_detected: 622 result = pdrv->err_handler-> 623 error_detected(pcidev, state); 624 break; 625 case XEN_PCI_OP_aer_mmio: 626 result = pdrv->err_handler-> 627 mmio_enabled(pcidev); 628 break; 629 case XEN_PCI_OP_aer_slotreset: 630 result = pdrv->err_handler-> 631 slot_reset(pcidev); 632 break; 633 case XEN_PCI_OP_aer_resume: 634 pdrv->err_handler->resume(pcidev); 635 break; 636 default: 637 dev_err(&pdev->xdev->dev, 638 "bad request in aer recovery " 639 "operation!\n"); 640 641 } 642 } 643 } 644 } 645 if (!flag) 646 result = PCI_ERS_RESULT_NONE; 647 648 return result; 649} 650 651 652static void pcifront_do_aer(struct work_struct *data) 653{ 654 struct pcifront_device *pdev = 655 container_of(data, struct pcifront_device, op_work); 656 int cmd = pdev->sh_info->aer_op.cmd; 657 pci_channel_state_t state = 658 (pci_channel_state_t)pdev->sh_info->aer_op.err; 659 660 /*If a pci_conf op is in progress, 661 we have to wait until it is done before service aer op*/ 662 dev_dbg(&pdev->xdev->dev, 663 "pcifront service aer bus %x devfn %x\n", 664 pdev->sh_info->aer_op.bus, pdev->sh_info->aer_op.devfn); 665 666 pdev->sh_info->aer_op.err = pcifront_common_process(cmd, pdev, state); 667 668 /* Post the operation to the guest. */ 669 wmb(); 670 clear_bit(_XEN_PCIB_active, (unsigned long *)&pdev->sh_info->flags); 671 notify_remote_via_evtchn(pdev->evtchn); 672 673 /*in case of we lost an aer request in four lines time_window*/ 674 smp_mb__before_atomic(); 675 clear_bit(_PDEVB_op_active, &pdev->flags); 676 smp_mb__after_atomic(); 677 678 schedule_pcifront_aer_op(pdev); 679 680} 681 682static irqreturn_t pcifront_handler_aer(int irq, void *dev) 683{ 684 struct pcifront_device *pdev = dev; 685 schedule_pcifront_aer_op(pdev); 686 return IRQ_HANDLED; 687} 688static int pcifront_connect_and_init_dma(struct pcifront_device *pdev) 689{ 690 int err = 0; 691 692 spin_lock(&pcifront_dev_lock); 693 694 if (!pcifront_dev) { 695 dev_info(&pdev->xdev->dev, "Installing PCI frontend\n"); 696 pcifront_dev = pdev; 697 } else 698 err = -EEXIST; 699 700 spin_unlock(&pcifront_dev_lock); 701 702 if (!err && !swiotlb_nr_tbl()) { 703 err = pci_xen_swiotlb_init_late(); 704 if (err) 705 dev_err(&pdev->xdev->dev, "Could not setup SWIOTLB!\n"); 706 } 707 return err; 708} 709 710static void pcifront_disconnect(struct pcifront_device *pdev) 711{ 712 spin_lock(&pcifront_dev_lock); 713 714 if (pdev == pcifront_dev) { 715 dev_info(&pdev->xdev->dev, 716 "Disconnecting PCI Frontend Buses\n"); 717 pcifront_dev = NULL; 718 } 719 720 spin_unlock(&pcifront_dev_lock); 721} 722static struct pcifront_device *alloc_pdev(struct xenbus_device *xdev) 723{ 724 struct pcifront_device *pdev; 725 726 pdev = kzalloc(sizeof(struct pcifront_device), GFP_KERNEL); 727 if (pdev == NULL) 728 goto out; 729 730 pdev->sh_info = 731 (struct xen_pci_sharedinfo *)__get_free_page(GFP_KERNEL); 732 if (pdev->sh_info == NULL) { 733 kfree(pdev); 734 pdev = NULL; 735 goto out; 736 } 737 pdev->sh_info->flags = 0; 738 739 /*Flag for registering PV AER handler*/ 740 set_bit(_XEN_PCIB_AERHANDLER, (void *)&pdev->sh_info->flags); 741 742 dev_set_drvdata(&xdev->dev, pdev); 743 pdev->xdev = xdev; 744 745 INIT_LIST_HEAD(&pdev->root_buses); 746 747 spin_lock_init(&pdev->sh_info_lock); 748 749 pdev->evtchn = INVALID_EVTCHN; 750 pdev->gnt_ref = INVALID_GRANT_REF; 751 pdev->irq = -1; 752 753 INIT_WORK(&pdev->op_work, pcifront_do_aer); 754 755 dev_dbg(&xdev->dev, "Allocated pdev @ 0x%p pdev->sh_info @ 0x%p\n", 756 pdev, pdev->sh_info); 757out: 758 return pdev; 759} 760 761static void free_pdev(struct pcifront_device *pdev) 762{ 763 dev_dbg(&pdev->xdev->dev, "freeing pdev @ 0x%p\n", pdev); 764 765 pcifront_free_roots(pdev); 766 767 cancel_work_sync(&pdev->op_work); 768 769 if (pdev->irq >= 0) 770 unbind_from_irqhandler(pdev->irq, pdev); 771 772 if (pdev->evtchn != INVALID_EVTCHN) 773 xenbus_free_evtchn(pdev->xdev, pdev->evtchn); 774 775 if (pdev->gnt_ref != INVALID_GRANT_REF) 776 gnttab_end_foreign_access(pdev->gnt_ref, 0 /* r/w page */, 777 (unsigned long)pdev->sh_info); 778 else 779 free_page((unsigned long)pdev->sh_info); 780 781 dev_set_drvdata(&pdev->xdev->dev, NULL); 782 783 kfree(pdev); 784} 785 786static int pcifront_publish_info(struct pcifront_device *pdev) 787{ 788 int err = 0; 789 struct xenbus_transaction trans; 790 grant_ref_t gref; 791 792 err = xenbus_grant_ring(pdev->xdev, pdev->sh_info, 1, &gref); 793 if (err < 0) 794 goto out; 795 796 pdev->gnt_ref = gref; 797 798 err = xenbus_alloc_evtchn(pdev->xdev, &pdev->evtchn); 799 if (err) 800 goto out; 801 802 err = bind_evtchn_to_irqhandler(pdev->evtchn, pcifront_handler_aer, 803 0, "pcifront", pdev); 804 805 if (err < 0) 806 return err; 807 808 pdev->irq = err; 809 810do_publish: 811 err = xenbus_transaction_start(&trans); 812 if (err) { 813 xenbus_dev_fatal(pdev->xdev, err, 814 "Error writing configuration for backend " 815 "(start transaction)"); 816 goto out; 817 } 818 819 err = xenbus_printf(trans, pdev->xdev->nodename, 820 "pci-op-ref", "%u", pdev->gnt_ref); 821 if (!err) 822 err = xenbus_printf(trans, pdev->xdev->nodename, 823 "event-channel", "%u", pdev->evtchn); 824 if (!err) 825 err = xenbus_printf(trans, pdev->xdev->nodename, 826 "magic", XEN_PCI_MAGIC); 827 828 if (err) { 829 xenbus_transaction_end(trans, 1); 830 xenbus_dev_fatal(pdev->xdev, err, 831 "Error writing configuration for backend"); 832 goto out; 833 } else { 834 err = xenbus_transaction_end(trans, 0); 835 if (err == -EAGAIN) 836 goto do_publish; 837 else if (err) { 838 xenbus_dev_fatal(pdev->xdev, err, 839 "Error completing transaction " 840 "for backend"); 841 goto out; 842 } 843 } 844 845 xenbus_switch_state(pdev->xdev, XenbusStateInitialised); 846 847 dev_dbg(&pdev->xdev->dev, "publishing successful!\n"); 848 849out: 850 return err; 851} 852 853static int pcifront_try_connect(struct pcifront_device *pdev) 854{ 855 int err = -EFAULT; 856 int i, num_roots, len; 857 char str[64]; 858 unsigned int domain, bus; 859 860 861 /* Only connect once */ 862 if (xenbus_read_driver_state(pdev->xdev->nodename) != 863 XenbusStateInitialised) 864 goto out; 865 866 err = pcifront_connect_and_init_dma(pdev); 867 if (err && err != -EEXIST) { 868 xenbus_dev_fatal(pdev->xdev, err, 869 "Error setting up PCI Frontend"); 870 goto out; 871 } 872 873 err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, 874 "root_num", "%d", &num_roots); 875 if (err == -ENOENT) { 876 xenbus_dev_error(pdev->xdev, err, 877 "No PCI Roots found, trying 0000:00"); 878 err = pcifront_scan_root(pdev, 0, 0); 879 if (err) { 880 xenbus_dev_fatal(pdev->xdev, err, 881 "Error scanning PCI root 0000:00"); 882 goto out; 883 } 884 num_roots = 0; 885 } else if (err != 1) { 886 if (err == 0) 887 err = -EINVAL; 888 xenbus_dev_fatal(pdev->xdev, err, 889 "Error reading number of PCI roots"); 890 goto out; 891 } 892 893 for (i = 0; i < num_roots; i++) { 894 len = snprintf(str, sizeof(str), "root-%d", i); 895 if (unlikely(len >= (sizeof(str) - 1))) { 896 err = -ENOMEM; 897 goto out; 898 } 899 900 err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str, 901 "%x:%x", &domain, &bus); 902 if (err != 2) { 903 if (err >= 0) 904 err = -EINVAL; 905 xenbus_dev_fatal(pdev->xdev, err, 906 "Error reading PCI root %d", i); 907 goto out; 908 } 909 910 err = pcifront_scan_root(pdev, domain, bus); 911 if (err) { 912 xenbus_dev_fatal(pdev->xdev, err, 913 "Error scanning PCI root %04x:%02x", 914 domain, bus); 915 goto out; 916 } 917 } 918 919 err = xenbus_switch_state(pdev->xdev, XenbusStateConnected); 920 921out: 922 return err; 923} 924 925static int pcifront_try_disconnect(struct pcifront_device *pdev) 926{ 927 int err = 0; 928 enum xenbus_state prev_state; 929 930 931 prev_state = xenbus_read_driver_state(pdev->xdev->nodename); 932 933 if (prev_state >= XenbusStateClosing) 934 goto out; 935 936 if (prev_state == XenbusStateConnected) { 937 pcifront_free_roots(pdev); 938 pcifront_disconnect(pdev); 939 } 940 941 err = xenbus_switch_state(pdev->xdev, XenbusStateClosed); 942 943out: 944 945 return err; 946} 947 948static int pcifront_attach_devices(struct pcifront_device *pdev) 949{ 950 int err = -EFAULT; 951 int i, num_roots, len; 952 unsigned int domain, bus; 953 char str[64]; 954 955 if (xenbus_read_driver_state(pdev->xdev->nodename) != 956 XenbusStateReconfiguring) 957 goto out; 958 959 err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, 960 "root_num", "%d", &num_roots); 961 if (err == -ENOENT) { 962 xenbus_dev_error(pdev->xdev, err, 963 "No PCI Roots found, trying 0000:00"); 964 err = pcifront_rescan_root(pdev, 0, 0); 965 if (err) { 966 xenbus_dev_fatal(pdev->xdev, err, 967 "Error scanning PCI root 0000:00"); 968 goto out; 969 } 970 num_roots = 0; 971 } else if (err != 1) { 972 if (err == 0) 973 err = -EINVAL; 974 xenbus_dev_fatal(pdev->xdev, err, 975 "Error reading number of PCI roots"); 976 goto out; 977 } 978 979 for (i = 0; i < num_roots; i++) { 980 len = snprintf(str, sizeof(str), "root-%d", i); 981 if (unlikely(len >= (sizeof(str) - 1))) { 982 err = -ENOMEM; 983 goto out; 984 } 985 986 err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str, 987 "%x:%x", &domain, &bus); 988 if (err != 2) { 989 if (err >= 0) 990 err = -EINVAL; 991 xenbus_dev_fatal(pdev->xdev, err, 992 "Error reading PCI root %d", i); 993 goto out; 994 } 995 996 err = pcifront_rescan_root(pdev, domain, bus); 997 if (err) { 998 xenbus_dev_fatal(pdev->xdev, err, 999 "Error scanning PCI root %04x:%02x", 1000 domain, bus); 1001 goto out; 1002 } 1003 } 1004 1005 xenbus_switch_state(pdev->xdev, XenbusStateConnected); 1006 1007out: 1008 return err; 1009} 1010 1011static int pcifront_detach_devices(struct pcifront_device *pdev) 1012{ 1013 int err = 0; 1014 int i, num_devs; 1015 unsigned int domain, bus, slot, func; 1016 struct pci_dev *pci_dev; 1017 char str[64]; 1018 1019 if (xenbus_read_driver_state(pdev->xdev->nodename) != 1020 XenbusStateConnected) 1021 goto out; 1022 1023 err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, "num_devs", "%d", 1024 &num_devs); 1025 if (err != 1) { 1026 if (err >= 0) 1027 err = -EINVAL; 1028 xenbus_dev_fatal(pdev->xdev, err, 1029 "Error reading number of PCI devices"); 1030 goto out; 1031 } 1032 1033 /* Find devices being detached and remove them. */ 1034 for (i = 0; i < num_devs; i++) { 1035 int l, state; 1036 l = snprintf(str, sizeof(str), "state-%d", i); 1037 if (unlikely(l >= (sizeof(str) - 1))) { 1038 err = -ENOMEM; 1039 goto out; 1040 } 1041 err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str, "%d", 1042 &state); 1043 if (err != 1) 1044 state = XenbusStateUnknown; 1045 1046 if (state != XenbusStateClosing) 1047 continue; 1048 1049 /* Remove device. */ 1050 l = snprintf(str, sizeof(str), "vdev-%d", i); 1051 if (unlikely(l >= (sizeof(str) - 1))) { 1052 err = -ENOMEM; 1053 goto out; 1054 } 1055 err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str, 1056 "%x:%x:%x.%x", &domain, &bus, &slot, &func); 1057 if (err != 4) { 1058 if (err >= 0) 1059 err = -EINVAL; 1060 xenbus_dev_fatal(pdev->xdev, err, 1061 "Error reading PCI device %d", i); 1062 goto out; 1063 } 1064 1065 pci_dev = pci_get_domain_bus_and_slot(domain, bus, 1066 PCI_DEVFN(slot, func)); 1067 if (!pci_dev) { 1068 dev_dbg(&pdev->xdev->dev, 1069 "Cannot get PCI device %04x:%02x:%02x.%d\n", 1070 domain, bus, slot, func); 1071 continue; 1072 } 1073 pci_lock_rescan_remove(); 1074 pci_stop_and_remove_bus_device(pci_dev); 1075 pci_dev_put(pci_dev); 1076 pci_unlock_rescan_remove(); 1077 1078 dev_dbg(&pdev->xdev->dev, 1079 "PCI device %04x:%02x:%02x.%d removed.\n", 1080 domain, bus, slot, func); 1081 } 1082 1083 err = xenbus_switch_state(pdev->xdev, XenbusStateReconfiguring); 1084 1085out: 1086 return err; 1087} 1088 1089static void __init_refok pcifront_backend_changed(struct xenbus_device *xdev, 1090 enum xenbus_state be_state) 1091{ 1092 struct pcifront_device *pdev = dev_get_drvdata(&xdev->dev); 1093 1094 switch (be_state) { 1095 case XenbusStateUnknown: 1096 case XenbusStateInitialising: 1097 case XenbusStateInitWait: 1098 case XenbusStateInitialised: 1099 break; 1100 1101 case XenbusStateConnected: 1102 pcifront_try_connect(pdev); 1103 break; 1104 1105 case XenbusStateClosed: 1106 if (xdev->state == XenbusStateClosed) 1107 break; 1108 /* Missed the backend's CLOSING state -- fallthrough */ 1109 case XenbusStateClosing: 1110 dev_warn(&xdev->dev, "backend going away!\n"); 1111 pcifront_try_disconnect(pdev); 1112 break; 1113 1114 case XenbusStateReconfiguring: 1115 pcifront_detach_devices(pdev); 1116 break; 1117 1118 case XenbusStateReconfigured: 1119 pcifront_attach_devices(pdev); 1120 break; 1121 } 1122} 1123 1124static int pcifront_xenbus_probe(struct xenbus_device *xdev, 1125 const struct xenbus_device_id *id) 1126{ 1127 int err = 0; 1128 struct pcifront_device *pdev = alloc_pdev(xdev); 1129 1130 if (pdev == NULL) { 1131 err = -ENOMEM; 1132 xenbus_dev_fatal(xdev, err, 1133 "Error allocating pcifront_device struct"); 1134 goto out; 1135 } 1136 1137 err = pcifront_publish_info(pdev); 1138 if (err) 1139 free_pdev(pdev); 1140 1141out: 1142 return err; 1143} 1144 1145static int pcifront_xenbus_remove(struct xenbus_device *xdev) 1146{ 1147 struct pcifront_device *pdev = dev_get_drvdata(&xdev->dev); 1148 if (pdev) 1149 free_pdev(pdev); 1150 1151 return 0; 1152} 1153 1154static const struct xenbus_device_id xenpci_ids[] = { 1155 {"pci"}, 1156 {""}, 1157}; 1158 1159static struct xenbus_driver xenpci_driver = { 1160 .name = "pcifront", 1161 .ids = xenpci_ids, 1162 .probe = pcifront_xenbus_probe, 1163 .remove = pcifront_xenbus_remove, 1164 .otherend_changed = pcifront_backend_changed, 1165}; 1166 1167static int __init pcifront_init(void) 1168{ 1169 if (!xen_pv_domain() || xen_initial_domain()) 1170 return -ENODEV; 1171 1172 if (!xen_has_pv_devices()) 1173 return -ENODEV; 1174 1175 pci_frontend_registrar(1 /* enable */); 1176 1177 return xenbus_register_frontend(&xenpci_driver); 1178} 1179 1180static void __exit pcifront_cleanup(void) 1181{ 1182 xenbus_unregister_driver(&xenpci_driver); 1183 pci_frontend_registrar(0 /* disable */); 1184} 1185module_init(pcifront_init); 1186module_exit(pcifront_cleanup); 1187 1188MODULE_DESCRIPTION("Xen PCI passthrough frontend."); 1189MODULE_LICENSE("GPL"); 1190MODULE_ALIAS("xen:pci"); 1191