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