1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * PCI Express I/O Virtualization (IOV) support 4 * Single Root IOV 1.0 5 * Address Translation Service 1.0 6 * 7 * Copyright (C) 2009 Intel Corporation, Yu Zhao <yu.zhao@intel.com> 8 */ 9 10 #include <linux/bitfield.h> 11 #include <linux/bits.h> 12 #include <linux/log2.h> 13 #include <linux/pci.h> 14 #include <linux/sizes.h> 15 #include <linux/slab.h> 16 #include <linux/export.h> 17 #include <linux/string.h> 18 #include <linux/delay.h> 19 #include <asm/div64.h> 20 #include "pci.h" 21 22 #define VIRTFN_ID_LEN 17 /* "virtfn%u\0" for 2^32 - 1 */ 23 24 int pci_iov_virtfn_bus(struct pci_dev *dev, int vf_id) 25 { 26 if (!dev->is_physfn) 27 return -EINVAL; 28 return dev->bus->number + ((dev->devfn + dev->sriov->offset + 29 dev->sriov->stride * vf_id) >> 8); 30 } 31 32 int pci_iov_virtfn_devfn(struct pci_dev *dev, int vf_id) 33 { 34 if (!dev->is_physfn) 35 return -EINVAL; 36 return (dev->devfn + dev->sriov->offset + 37 dev->sriov->stride * vf_id) & 0xff; 38 } 39 EXPORT_SYMBOL_GPL(pci_iov_virtfn_devfn); 40 41 int pci_iov_vf_id(struct pci_dev *dev) 42 { 43 struct pci_dev *pf; 44 45 if (!dev->is_virtfn) 46 return -EINVAL; 47 48 pf = pci_physfn(dev); 49 return (pci_dev_id(dev) - (pci_dev_id(pf) + pf->sriov->offset)) / 50 pf->sriov->stride; 51 } 52 EXPORT_SYMBOL_GPL(pci_iov_vf_id); 53 54 /** 55 * pci_iov_get_pf_drvdata - Return the drvdata of a PF 56 * @dev: VF pci_dev 57 * @pf_driver: Device driver required to own the PF 58 * 59 * This must be called from a context that ensures that a VF driver is attached. 60 * The value returned is invalid once the VF driver completes its remove() 61 * callback. 62 * 63 * Locking is achieved by the driver core. A VF driver cannot be probed until 64 * pci_enable_sriov() is called and pci_disable_sriov() does not return until 65 * all VF drivers have completed their remove(). 66 * 67 * The PF driver must call pci_disable_sriov() before it begins to destroy the 68 * drvdata. 69 */ 70 void *pci_iov_get_pf_drvdata(struct pci_dev *dev, struct pci_driver *pf_driver) 71 { 72 struct pci_dev *pf_dev; 73 74 if (!dev->is_virtfn) 75 return ERR_PTR(-EINVAL); 76 pf_dev = dev->physfn; 77 if (pf_dev->driver != pf_driver) 78 return ERR_PTR(-EINVAL); 79 return pci_get_drvdata(pf_dev); 80 } 81 EXPORT_SYMBOL_GPL(pci_iov_get_pf_drvdata); 82 83 /* 84 * Per SR-IOV spec sec 3.3.10 and 3.3.11, First VF Offset and VF Stride may 85 * change when NumVFs changes. 86 * 87 * Update iov->offset and iov->stride when NumVFs is written. 88 */ 89 static inline void pci_iov_set_numvfs(struct pci_dev *dev, int nr_virtfn) 90 { 91 struct pci_sriov *iov = dev->sriov; 92 93 pci_write_config_word(dev, iov->pos + PCI_SRIOV_NUM_VF, nr_virtfn); 94 pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_OFFSET, &iov->offset); 95 pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_STRIDE, &iov->stride); 96 } 97 98 /* 99 * The PF consumes one bus number. NumVFs, First VF Offset, and VF Stride 100 * determine how many additional bus numbers will be consumed by VFs. 101 * 102 * Iterate over all valid NumVFs, validate offset and stride, and calculate 103 * the maximum number of bus numbers that could ever be required. 104 */ 105 static int compute_max_vf_buses(struct pci_dev *dev) 106 { 107 struct pci_sriov *iov = dev->sriov; 108 int nr_virtfn, busnr, rc = 0; 109 110 for (nr_virtfn = iov->total_VFs; nr_virtfn; nr_virtfn--) { 111 pci_iov_set_numvfs(dev, nr_virtfn); 112 if (!iov->offset || (nr_virtfn > 1 && !iov->stride)) { 113 rc = -EIO; 114 goto out; 115 } 116 117 busnr = pci_iov_virtfn_bus(dev, nr_virtfn - 1); 118 if (busnr > iov->max_VF_buses) 119 iov->max_VF_buses = busnr; 120 } 121 122 out: 123 pci_iov_set_numvfs(dev, 0); 124 return rc; 125 } 126 127 static struct pci_bus *virtfn_add_bus(struct pci_bus *bus, int busnr) 128 { 129 struct pci_bus *child; 130 131 if (bus->number == busnr) 132 return bus; 133 134 child = pci_find_bus(pci_domain_nr(bus), busnr); 135 if (child) 136 return child; 137 138 child = pci_add_new_bus(bus, NULL, busnr); 139 if (!child) 140 return NULL; 141 142 pci_bus_insert_busn_res(child, busnr, busnr); 143 144 return child; 145 } 146 147 static void virtfn_remove_bus(struct pci_bus *physbus, struct pci_bus *virtbus) 148 { 149 if (physbus != virtbus && list_empty(&virtbus->devices)) 150 pci_remove_bus(virtbus); 151 } 152 153 resource_size_t pci_iov_resource_size(struct pci_dev *dev, int resno) 154 { 155 if (!dev->is_physfn) 156 return 0; 157 158 return dev->sriov->barsz[pci_resource_num_to_vf_bar(resno)]; 159 } 160 161 void pci_iov_resource_set_size(struct pci_dev *dev, int resno, int size) 162 { 163 if (!pci_resource_is_iov(resno)) { 164 pci_warn(dev, "%s is not an IOV resource\n", 165 pci_resource_name(dev, resno)); 166 return; 167 } 168 169 resno = pci_resource_num_to_vf_bar(resno); 170 dev->sriov->barsz[resno] = pci_rebar_size_to_bytes(size); 171 } 172 173 bool pci_iov_is_memory_decoding_enabled(struct pci_dev *dev) 174 { 175 u16 cmd; 176 177 pci_read_config_word(dev, dev->sriov->pos + PCI_SRIOV_CTRL, &cmd); 178 179 return cmd & PCI_SRIOV_CTRL_MSE; 180 } 181 182 static void pci_read_vf_config_common(struct pci_dev *virtfn) 183 { 184 struct pci_dev *physfn = virtfn->physfn; 185 186 /* 187 * Some config registers are the same across all associated VFs. 188 * Read them once from VF0 so we can skip reading them from the 189 * other VFs. 190 * 191 * PCIe r4.0, sec 9.3.4.1, technically doesn't require all VFs to 192 * have the same Revision ID and Subsystem ID, but we assume they 193 * do. 194 */ 195 pci_read_config_dword(virtfn, PCI_CLASS_REVISION, 196 &physfn->sriov->class); 197 pci_read_config_byte(virtfn, PCI_HEADER_TYPE, 198 &physfn->sriov->hdr_type); 199 pci_read_config_word(virtfn, PCI_SUBSYSTEM_VENDOR_ID, 200 &physfn->sriov->subsystem_vendor); 201 pci_read_config_word(virtfn, PCI_SUBSYSTEM_ID, 202 &physfn->sriov->subsystem_device); 203 } 204 205 int pci_iov_sysfs_link(struct pci_dev *dev, 206 struct pci_dev *virtfn, int id) 207 { 208 char buf[VIRTFN_ID_LEN]; 209 int rc; 210 211 sprintf(buf, "virtfn%u", id); 212 rc = sysfs_create_link(&dev->dev.kobj, &virtfn->dev.kobj, buf); 213 if (rc) 214 goto failed; 215 rc = sysfs_create_link(&virtfn->dev.kobj, &dev->dev.kobj, "physfn"); 216 if (rc) 217 goto failed1; 218 219 kobject_uevent(&virtfn->dev.kobj, KOBJ_CHANGE); 220 221 return 0; 222 223 failed1: 224 sysfs_remove_link(&dev->dev.kobj, buf); 225 failed: 226 return rc; 227 } 228 229 #ifdef CONFIG_PCI_MSI 230 static ssize_t sriov_vf_total_msix_show(struct device *dev, 231 struct device_attribute *attr, 232 char *buf) 233 { 234 struct pci_dev *pdev = to_pci_dev(dev); 235 u32 vf_total_msix = 0; 236 237 device_lock(dev); 238 if (!pdev->driver || !pdev->driver->sriov_get_vf_total_msix) 239 goto unlock; 240 241 vf_total_msix = pdev->driver->sriov_get_vf_total_msix(pdev); 242 unlock: 243 device_unlock(dev); 244 return sysfs_emit(buf, "%u\n", vf_total_msix); 245 } 246 static DEVICE_ATTR_RO(sriov_vf_total_msix); 247 248 static ssize_t sriov_vf_msix_count_store(struct device *dev, 249 struct device_attribute *attr, 250 const char *buf, size_t count) 251 { 252 struct pci_dev *vf_dev = to_pci_dev(dev); 253 struct pci_dev *pdev = pci_physfn(vf_dev); 254 int val, ret = 0; 255 256 if (kstrtoint(buf, 0, &val) < 0) 257 return -EINVAL; 258 259 if (val < 0) 260 return -EINVAL; 261 262 device_lock(&pdev->dev); 263 if (!pdev->driver || !pdev->driver->sriov_set_msix_vec_count) { 264 ret = -EOPNOTSUPP; 265 goto err_pdev; 266 } 267 268 device_lock(&vf_dev->dev); 269 if (vf_dev->driver) { 270 /* 271 * A driver is already attached to this VF and has configured 272 * itself based on the current MSI-X vector count. Changing 273 * the vector size could mess up the driver, so block it. 274 */ 275 ret = -EBUSY; 276 goto err_dev; 277 } 278 279 ret = pdev->driver->sriov_set_msix_vec_count(vf_dev, val); 280 281 err_dev: 282 device_unlock(&vf_dev->dev); 283 err_pdev: 284 device_unlock(&pdev->dev); 285 return ret ? : count; 286 } 287 static DEVICE_ATTR_WO(sriov_vf_msix_count); 288 #endif 289 290 static struct attribute *sriov_vf_dev_attrs[] = { 291 #ifdef CONFIG_PCI_MSI 292 &dev_attr_sriov_vf_msix_count.attr, 293 #endif 294 NULL, 295 }; 296 297 static umode_t sriov_vf_attrs_are_visible(struct kobject *kobj, 298 struct attribute *a, int n) 299 { 300 struct device *dev = kobj_to_dev(kobj); 301 struct pci_dev *pdev = to_pci_dev(dev); 302 303 if (!pdev->is_virtfn) 304 return 0; 305 306 return a->mode; 307 } 308 309 const struct attribute_group sriov_vf_dev_attr_group = { 310 .attrs = sriov_vf_dev_attrs, 311 .is_visible = sriov_vf_attrs_are_visible, 312 }; 313 314 static struct pci_dev *pci_iov_scan_device(struct pci_dev *dev, int id, 315 struct pci_bus *bus) 316 { 317 struct pci_sriov *iov = dev->sriov; 318 struct pci_dev *virtfn; 319 int rc; 320 321 virtfn = pci_alloc_dev(bus); 322 if (!virtfn) 323 return ERR_PTR(-ENOMEM); 324 325 virtfn->devfn = pci_iov_virtfn_devfn(dev, id); 326 virtfn->vendor = dev->vendor; 327 virtfn->device = iov->vf_device; 328 virtfn->is_virtfn = 1; 329 virtfn->physfn = pci_dev_get(dev); 330 virtfn->no_command_memory = 1; 331 332 if (id == 0) 333 pci_read_vf_config_common(virtfn); 334 335 rc = pci_setup_device(virtfn); 336 if (rc) { 337 pci_dev_put(dev); 338 pci_bus_put(virtfn->bus); 339 kfree(virtfn); 340 return ERR_PTR(rc); 341 } 342 343 return virtfn; 344 } 345 346 int pci_iov_add_virtfn(struct pci_dev *dev, int id) 347 { 348 struct pci_bus *bus; 349 struct pci_dev *virtfn; 350 struct resource *res; 351 int rc, i; 352 u64 size; 353 354 bus = virtfn_add_bus(dev->bus, pci_iov_virtfn_bus(dev, id)); 355 if (!bus) { 356 rc = -ENOMEM; 357 goto failed; 358 } 359 360 virtfn = pci_iov_scan_device(dev, id, bus); 361 if (IS_ERR(virtfn)) { 362 rc = PTR_ERR(virtfn); 363 goto failed0; 364 } 365 366 virtfn->dev.parent = dev->dev.parent; 367 virtfn->multifunction = 0; 368 369 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) { 370 int idx = pci_resource_num_from_vf_bar(i); 371 372 res = &dev->resource[idx]; 373 if (!res->parent) 374 continue; 375 virtfn->resource[i].name = pci_name(virtfn); 376 virtfn->resource[i].flags = res->flags; 377 size = pci_iov_resource_size(dev, idx); 378 resource_set_range(&virtfn->resource[i], 379 res->start + size * id, size); 380 rc = request_resource(res, &virtfn->resource[i]); 381 BUG_ON(rc); 382 } 383 384 pci_device_add(virtfn, virtfn->bus); 385 rc = pci_iov_sysfs_link(dev, virtfn, id); 386 if (rc) 387 goto failed1; 388 389 pci_bus_add_device(virtfn); 390 391 return 0; 392 393 failed1: 394 pci_stop_and_remove_bus_device(virtfn); 395 pci_dev_put(dev); 396 failed0: 397 virtfn_remove_bus(dev->bus, bus); 398 failed: 399 400 return rc; 401 } 402 403 void pci_iov_remove_virtfn(struct pci_dev *dev, int id) 404 { 405 char buf[VIRTFN_ID_LEN]; 406 struct pci_dev *virtfn; 407 408 virtfn = pci_get_domain_bus_and_slot(pci_domain_nr(dev->bus), 409 pci_iov_virtfn_bus(dev, id), 410 pci_iov_virtfn_devfn(dev, id)); 411 if (!virtfn) 412 return; 413 414 sprintf(buf, "virtfn%u", id); 415 sysfs_remove_link(&dev->dev.kobj, buf); 416 /* 417 * pci_stop_dev() could have been called for this virtfn already, 418 * so the directory for the virtfn may have been removed before. 419 * Double check to avoid spurious sysfs warnings. 420 */ 421 if (virtfn->dev.kobj.sd) 422 sysfs_remove_link(&virtfn->dev.kobj, "physfn"); 423 424 pci_stop_and_remove_bus_device(virtfn); 425 virtfn_remove_bus(dev->bus, virtfn->bus); 426 427 /* balance pci_get_domain_bus_and_slot() */ 428 pci_dev_put(virtfn); 429 pci_dev_put(dev); 430 } 431 432 static ssize_t sriov_totalvfs_show(struct device *dev, 433 struct device_attribute *attr, 434 char *buf) 435 { 436 struct pci_dev *pdev = to_pci_dev(dev); 437 438 return sysfs_emit(buf, "%u\n", pci_sriov_get_totalvfs(pdev)); 439 } 440 441 static ssize_t sriov_numvfs_show(struct device *dev, 442 struct device_attribute *attr, 443 char *buf) 444 { 445 struct pci_dev *pdev = to_pci_dev(dev); 446 u16 num_vfs; 447 448 /* Serialize vs sriov_numvfs_store() so readers see valid num_VFs */ 449 device_lock(&pdev->dev); 450 num_vfs = pdev->sriov->num_VFs; 451 device_unlock(&pdev->dev); 452 453 return sysfs_emit(buf, "%u\n", num_vfs); 454 } 455 456 /* 457 * num_vfs > 0; number of VFs to enable 458 * num_vfs = 0; disable all VFs 459 * 460 * Note: SRIOV spec does not allow partial VF 461 * disable, so it's all or none. 462 */ 463 static ssize_t sriov_numvfs_store(struct device *dev, 464 struct device_attribute *attr, 465 const char *buf, size_t count) 466 { 467 struct pci_dev *pdev = to_pci_dev(dev); 468 int ret = 0; 469 u16 num_vfs; 470 471 if (kstrtou16(buf, 0, &num_vfs) < 0) 472 return -EINVAL; 473 474 if (num_vfs > pci_sriov_get_totalvfs(pdev)) 475 return -ERANGE; 476 477 device_lock(&pdev->dev); 478 479 if (num_vfs == pdev->sriov->num_VFs) 480 goto exit; 481 482 /* is PF driver loaded */ 483 if (!pdev->driver) { 484 pci_info(pdev, "no driver bound to device; cannot configure SR-IOV\n"); 485 ret = -ENOENT; 486 goto exit; 487 } 488 489 /* is PF driver loaded w/callback */ 490 if (!pdev->driver->sriov_configure) { 491 pci_info(pdev, "driver does not support SR-IOV configuration via sysfs\n"); 492 ret = -ENOENT; 493 goto exit; 494 } 495 496 if (num_vfs == 0) { 497 /* disable VFs */ 498 pci_lock_rescan_remove(); 499 ret = pdev->driver->sriov_configure(pdev, 0); 500 pci_unlock_rescan_remove(); 501 goto exit; 502 } 503 504 /* enable VFs */ 505 if (pdev->sriov->num_VFs) { 506 pci_warn(pdev, "%d VFs already enabled. Disable before enabling %d VFs\n", 507 pdev->sriov->num_VFs, num_vfs); 508 ret = -EBUSY; 509 goto exit; 510 } 511 512 pci_lock_rescan_remove(); 513 ret = pdev->driver->sriov_configure(pdev, num_vfs); 514 pci_unlock_rescan_remove(); 515 if (ret < 0) 516 goto exit; 517 518 if (ret != num_vfs) 519 pci_warn(pdev, "%d VFs requested; only %d enabled\n", 520 num_vfs, ret); 521 522 exit: 523 device_unlock(&pdev->dev); 524 525 if (ret < 0) 526 return ret; 527 528 return count; 529 } 530 531 static ssize_t sriov_offset_show(struct device *dev, 532 struct device_attribute *attr, 533 char *buf) 534 { 535 struct pci_dev *pdev = to_pci_dev(dev); 536 537 return sysfs_emit(buf, "%u\n", pdev->sriov->offset); 538 } 539 540 static ssize_t sriov_stride_show(struct device *dev, 541 struct device_attribute *attr, 542 char *buf) 543 { 544 struct pci_dev *pdev = to_pci_dev(dev); 545 546 return sysfs_emit(buf, "%u\n", pdev->sriov->stride); 547 } 548 549 static ssize_t sriov_vf_device_show(struct device *dev, 550 struct device_attribute *attr, 551 char *buf) 552 { 553 struct pci_dev *pdev = to_pci_dev(dev); 554 555 return sysfs_emit(buf, "%x\n", pdev->sriov->vf_device); 556 } 557 558 static ssize_t sriov_drivers_autoprobe_show(struct device *dev, 559 struct device_attribute *attr, 560 char *buf) 561 { 562 struct pci_dev *pdev = to_pci_dev(dev); 563 564 return sysfs_emit(buf, "%u\n", pdev->sriov->drivers_autoprobe); 565 } 566 567 static ssize_t sriov_drivers_autoprobe_store(struct device *dev, 568 struct device_attribute *attr, 569 const char *buf, size_t count) 570 { 571 struct pci_dev *pdev = to_pci_dev(dev); 572 bool drivers_autoprobe; 573 574 if (kstrtobool(buf, &drivers_autoprobe) < 0) 575 return -EINVAL; 576 577 pdev->sriov->drivers_autoprobe = drivers_autoprobe; 578 579 return count; 580 } 581 582 static DEVICE_ATTR_RO(sriov_totalvfs); 583 static DEVICE_ATTR_RW(sriov_numvfs); 584 static DEVICE_ATTR_RO(sriov_offset); 585 static DEVICE_ATTR_RO(sriov_stride); 586 static DEVICE_ATTR_RO(sriov_vf_device); 587 static DEVICE_ATTR_RW(sriov_drivers_autoprobe); 588 589 static struct attribute *sriov_pf_dev_attrs[] = { 590 &dev_attr_sriov_totalvfs.attr, 591 &dev_attr_sriov_numvfs.attr, 592 &dev_attr_sriov_offset.attr, 593 &dev_attr_sriov_stride.attr, 594 &dev_attr_sriov_vf_device.attr, 595 &dev_attr_sriov_drivers_autoprobe.attr, 596 #ifdef CONFIG_PCI_MSI 597 &dev_attr_sriov_vf_total_msix.attr, 598 #endif 599 NULL, 600 }; 601 602 static umode_t sriov_pf_attrs_are_visible(struct kobject *kobj, 603 struct attribute *a, int n) 604 { 605 struct device *dev = kobj_to_dev(kobj); 606 607 if (!dev_is_pf(dev)) 608 return 0; 609 610 return a->mode; 611 } 612 613 const struct attribute_group sriov_pf_dev_attr_group = { 614 .attrs = sriov_pf_dev_attrs, 615 .is_visible = sriov_pf_attrs_are_visible, 616 }; 617 618 int __weak pcibios_sriov_enable(struct pci_dev *pdev, u16 num_vfs) 619 { 620 return 0; 621 } 622 623 int __weak pcibios_sriov_disable(struct pci_dev *pdev) 624 { 625 return 0; 626 } 627 628 static int sriov_add_vfs(struct pci_dev *dev, u16 num_vfs) 629 { 630 unsigned int i; 631 int rc; 632 633 if (dev->no_vf_scan) 634 return 0; 635 636 for (i = 0; i < num_vfs; i++) { 637 rc = pci_iov_add_virtfn(dev, i); 638 if (rc) 639 goto failed; 640 } 641 return 0; 642 failed: 643 while (i--) 644 pci_iov_remove_virtfn(dev, i); 645 646 return rc; 647 } 648 649 static int sriov_enable(struct pci_dev *dev, int nr_virtfn) 650 { 651 int rc; 652 int i; 653 int nres; 654 u16 initial; 655 struct resource *res; 656 struct pci_dev *pdev; 657 struct pci_sriov *iov = dev->sriov; 658 int bars = 0; 659 int bus; 660 661 if (!nr_virtfn) 662 return 0; 663 664 if (iov->num_VFs) 665 return -EINVAL; 666 667 pci_read_config_word(dev, iov->pos + PCI_SRIOV_INITIAL_VF, &initial); 668 if (initial > iov->total_VFs || 669 (!(iov->cap & PCI_SRIOV_CAP_VFM) && (initial != iov->total_VFs))) 670 return -EIO; 671 672 if (nr_virtfn < 0 || nr_virtfn > iov->total_VFs || 673 (!(iov->cap & PCI_SRIOV_CAP_VFM) && (nr_virtfn > initial))) 674 return -EINVAL; 675 676 nres = 0; 677 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) { 678 int idx = pci_resource_num_from_vf_bar(i); 679 resource_size_t vf_bar_sz = pci_iov_resource_size(dev, idx); 680 681 bars |= (1 << idx); 682 res = &dev->resource[idx]; 683 if (vf_bar_sz * nr_virtfn > resource_size(res)) 684 continue; 685 if (res->parent) 686 nres++; 687 } 688 if (nres != iov->nres) { 689 pci_err(dev, "not enough MMIO resources for SR-IOV\n"); 690 return -ENOMEM; 691 } 692 693 bus = pci_iov_virtfn_bus(dev, nr_virtfn - 1); 694 if (bus > dev->bus->busn_res.end) { 695 pci_err(dev, "can't enable %d VFs (bus %02x out of range of %pR)\n", 696 nr_virtfn, bus, &dev->bus->busn_res); 697 return -ENOMEM; 698 } 699 700 if (pci_enable_resources(dev, bars)) { 701 pci_err(dev, "SR-IOV: IOV BARS not allocated\n"); 702 return -ENOMEM; 703 } 704 705 if (iov->link != dev->devfn) { 706 pdev = pci_get_slot(dev->bus, iov->link); 707 if (!pdev) 708 return -ENODEV; 709 710 if (!pdev->is_physfn) { 711 pci_dev_put(pdev); 712 return -ENOSYS; 713 } 714 715 rc = sysfs_create_link(&dev->dev.kobj, 716 &pdev->dev.kobj, "dep_link"); 717 pci_dev_put(pdev); 718 if (rc) 719 return rc; 720 } 721 722 iov->initial_VFs = initial; 723 if (nr_virtfn < initial) 724 initial = nr_virtfn; 725 726 rc = pcibios_sriov_enable(dev, initial); 727 if (rc) { 728 pci_err(dev, "failure %d from pcibios_sriov_enable()\n", rc); 729 goto err_pcibios; 730 } 731 732 pci_iov_set_numvfs(dev, nr_virtfn); 733 iov->ctrl |= PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE; 734 pci_cfg_access_lock(dev); 735 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl); 736 msleep(100); 737 pci_cfg_access_unlock(dev); 738 739 rc = sriov_add_vfs(dev, initial); 740 if (rc) 741 goto err_pcibios; 742 743 kobject_uevent(&dev->dev.kobj, KOBJ_CHANGE); 744 iov->num_VFs = nr_virtfn; 745 746 return 0; 747 748 err_pcibios: 749 iov->ctrl &= ~(PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE); 750 pci_cfg_access_lock(dev); 751 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl); 752 ssleep(1); 753 pci_cfg_access_unlock(dev); 754 755 pcibios_sriov_disable(dev); 756 757 if (iov->link != dev->devfn) 758 sysfs_remove_link(&dev->dev.kobj, "dep_link"); 759 760 pci_iov_set_numvfs(dev, 0); 761 return rc; 762 } 763 764 static void sriov_del_vfs(struct pci_dev *dev) 765 { 766 struct pci_sriov *iov = dev->sriov; 767 int i; 768 769 for (i = 0; i < iov->num_VFs; i++) 770 pci_iov_remove_virtfn(dev, i); 771 } 772 773 static void sriov_disable(struct pci_dev *dev) 774 { 775 struct pci_sriov *iov = dev->sriov; 776 777 if (!iov->num_VFs) 778 return; 779 780 sriov_del_vfs(dev); 781 iov->ctrl &= ~(PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE); 782 pci_cfg_access_lock(dev); 783 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl); 784 ssleep(1); 785 pci_cfg_access_unlock(dev); 786 787 pcibios_sriov_disable(dev); 788 789 if (iov->link != dev->devfn) 790 sysfs_remove_link(&dev->dev.kobj, "dep_link"); 791 792 iov->num_VFs = 0; 793 pci_iov_set_numvfs(dev, 0); 794 } 795 796 static int sriov_init(struct pci_dev *dev, int pos) 797 { 798 int i, bar64; 799 int rc; 800 int nres; 801 u32 pgsz; 802 u16 ctrl, total; 803 struct pci_sriov *iov; 804 struct resource *res; 805 const char *res_name; 806 struct pci_dev *pdev; 807 u32 sriovbars[PCI_SRIOV_NUM_BARS]; 808 809 pci_read_config_word(dev, pos + PCI_SRIOV_CTRL, &ctrl); 810 if (ctrl & PCI_SRIOV_CTRL_VFE) { 811 pci_write_config_word(dev, pos + PCI_SRIOV_CTRL, 0); 812 ssleep(1); 813 } 814 815 ctrl = 0; 816 list_for_each_entry(pdev, &dev->bus->devices, bus_list) 817 if (pdev->is_physfn) 818 goto found; 819 820 pdev = NULL; 821 if (pci_ari_enabled(dev->bus)) 822 ctrl |= PCI_SRIOV_CTRL_ARI; 823 824 found: 825 pci_write_config_word(dev, pos + PCI_SRIOV_CTRL, ctrl); 826 827 pci_read_config_word(dev, pos + PCI_SRIOV_TOTAL_VF, &total); 828 if (!total) 829 return 0; 830 831 pci_read_config_dword(dev, pos + PCI_SRIOV_SUP_PGSIZE, &pgsz); 832 i = PAGE_SHIFT > 12 ? PAGE_SHIFT - 12 : 0; 833 pgsz &= ~((1 << i) - 1); 834 if (!pgsz) 835 return -EIO; 836 837 pgsz &= ~(pgsz - 1); 838 pci_write_config_dword(dev, pos + PCI_SRIOV_SYS_PGSIZE, pgsz); 839 840 iov = kzalloc_obj(*iov, GFP_KERNEL); 841 if (!iov) 842 return -ENOMEM; 843 844 /* Sizing SR-IOV BARs with VF Enable cleared - no decode */ 845 __pci_size_stdbars(dev, PCI_SRIOV_NUM_BARS, 846 pos + PCI_SRIOV_BAR, sriovbars); 847 848 nres = 0; 849 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) { 850 int idx = pci_resource_num_from_vf_bar(i); 851 852 res = &dev->resource[idx]; 853 res_name = pci_resource_name(dev, idx); 854 855 /* 856 * If it is already FIXED, don't change it, something 857 * (perhaps EA or header fixups) wants it this way. 858 */ 859 if (res->flags & IORESOURCE_PCI_FIXED) 860 bar64 = (res->flags & IORESOURCE_MEM_64) ? 1 : 0; 861 else 862 bar64 = __pci_read_base(dev, pci_bar_unknown, res, 863 pos + PCI_SRIOV_BAR + i * 4, 864 &sriovbars[i]); 865 if (!res->flags) 866 continue; 867 if (resource_size(res) & (PAGE_SIZE - 1)) { 868 rc = -EIO; 869 goto failed; 870 } 871 iov->barsz[i] = resource_size(res); 872 resource_set_size(res, resource_size(res) * total); 873 pci_info(dev, "%s %pR: contains BAR %d for %d VFs\n", 874 res_name, res, i, total); 875 i += bar64; 876 nres++; 877 } 878 879 iov->pos = pos; 880 iov->nres = nres; 881 iov->ctrl = ctrl; 882 iov->total_VFs = total; 883 iov->driver_max_VFs = total; 884 pci_read_config_word(dev, pos + PCI_SRIOV_VF_DID, &iov->vf_device); 885 iov->pgsz = pgsz; 886 iov->self = dev; 887 iov->drivers_autoprobe = true; 888 pci_read_config_dword(dev, pos + PCI_SRIOV_CAP, &iov->cap); 889 pci_read_config_byte(dev, pos + PCI_SRIOV_FUNC_LINK, &iov->link); 890 if (pci_pcie_type(dev) == PCI_EXP_TYPE_RC_END) 891 iov->link = PCI_DEVFN(PCI_SLOT(dev->devfn), iov->link); 892 iov->vf_rebar_cap = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_VF_REBAR); 893 894 if (pdev) 895 iov->dev = pci_dev_get(pdev); 896 else 897 iov->dev = dev; 898 899 dev->sriov = iov; 900 dev->is_physfn = 1; 901 rc = compute_max_vf_buses(dev); 902 if (rc) 903 goto fail_max_buses; 904 905 return 0; 906 907 fail_max_buses: 908 dev->sriov = NULL; 909 dev->is_physfn = 0; 910 failed: 911 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) { 912 res = &dev->resource[pci_resource_num_from_vf_bar(i)]; 913 res->flags = 0; 914 } 915 916 kfree(iov); 917 return rc; 918 } 919 920 static void sriov_release(struct pci_dev *dev) 921 { 922 BUG_ON(dev->sriov->num_VFs); 923 924 if (dev != dev->sriov->dev) 925 pci_dev_put(dev->sriov->dev); 926 927 kfree(dev->sriov); 928 dev->sriov = NULL; 929 } 930 931 static void sriov_restore_vf_rebar_state(struct pci_dev *dev) 932 { 933 unsigned int pos, nbars, i; 934 u32 ctrl; 935 936 pos = pci_iov_vf_rebar_cap(dev); 937 if (!pos) 938 return; 939 940 pci_read_config_dword(dev, pos + PCI_VF_REBAR_CTRL, &ctrl); 941 nbars = FIELD_GET(PCI_VF_REBAR_CTRL_NBAR_MASK, ctrl); 942 943 for (i = 0; i < nbars; i++, pos += 8) { 944 int bar_idx, size; 945 946 pci_read_config_dword(dev, pos + PCI_VF_REBAR_CTRL, &ctrl); 947 bar_idx = FIELD_GET(PCI_VF_REBAR_CTRL_BAR_IDX, ctrl); 948 size = pci_rebar_bytes_to_size(dev->sriov->barsz[bar_idx]); 949 ctrl &= ~PCI_VF_REBAR_CTRL_BAR_SIZE; 950 ctrl |= FIELD_PREP(PCI_VF_REBAR_CTRL_BAR_SIZE, size); 951 pci_write_config_dword(dev, pos + PCI_VF_REBAR_CTRL, ctrl); 952 } 953 } 954 955 static void sriov_restore_state(struct pci_dev *dev) 956 { 957 int i; 958 u16 ctrl; 959 struct pci_sriov *iov = dev->sriov; 960 961 pci_read_config_word(dev, iov->pos + PCI_SRIOV_CTRL, &ctrl); 962 if (ctrl & PCI_SRIOV_CTRL_VFE) 963 return; 964 965 /* 966 * Restore PCI_SRIOV_CTRL_ARI before pci_iov_set_numvfs() because 967 * it reads offset & stride, which depend on PCI_SRIOV_CTRL_ARI. 968 */ 969 ctrl &= ~PCI_SRIOV_CTRL_ARI; 970 ctrl |= iov->ctrl & PCI_SRIOV_CTRL_ARI; 971 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, ctrl); 972 973 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) 974 pci_update_resource(dev, pci_resource_num_from_vf_bar(i)); 975 976 pci_write_config_dword(dev, iov->pos + PCI_SRIOV_SYS_PGSIZE, iov->pgsz); 977 pci_iov_set_numvfs(dev, iov->num_VFs); 978 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl); 979 if (iov->ctrl & PCI_SRIOV_CTRL_VFE) 980 msleep(100); 981 } 982 983 /** 984 * pci_iov_init - initialize the IOV capability 985 * @dev: the PCI device 986 * 987 * Returns 0 on success, or negative on failure. 988 */ 989 int pci_iov_init(struct pci_dev *dev) 990 { 991 int pos; 992 993 if (!pci_is_pcie(dev)) 994 return -ENODEV; 995 996 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_SRIOV); 997 if (pos) 998 return sriov_init(dev, pos); 999 1000 return -ENODEV; 1001 } 1002 1003 /** 1004 * pci_iov_release - release resources used by the IOV capability 1005 * @dev: the PCI device 1006 */ 1007 void pci_iov_release(struct pci_dev *dev) 1008 { 1009 if (dev->is_physfn) 1010 sriov_release(dev); 1011 } 1012 1013 /** 1014 * pci_iov_remove - clean up SR-IOV state after PF driver is detached 1015 * @dev: the PCI device 1016 */ 1017 void pci_iov_remove(struct pci_dev *dev) 1018 { 1019 struct pci_sriov *iov = dev->sriov; 1020 1021 if (!dev->is_physfn) 1022 return; 1023 1024 iov->driver_max_VFs = iov->total_VFs; 1025 if (iov->num_VFs) 1026 pci_warn(dev, "driver left SR-IOV enabled after remove\n"); 1027 } 1028 1029 /** 1030 * pci_iov_update_resource - update a VF BAR 1031 * @dev: the PCI device 1032 * @resno: the resource number 1033 * 1034 * Update a VF BAR in the SR-IOV capability of a PF. 1035 */ 1036 void pci_iov_update_resource(struct pci_dev *dev, int resno) 1037 { 1038 struct pci_sriov *iov = dev->is_physfn ? dev->sriov : NULL; 1039 struct resource *res = pci_resource_n(dev, resno); 1040 int vf_bar = pci_resource_num_to_vf_bar(resno); 1041 struct pci_bus_region region; 1042 u16 cmd; 1043 u32 new; 1044 int reg; 1045 1046 /* 1047 * The generic pci_restore_bars() path calls this for all devices, 1048 * including VFs and non-SR-IOV devices. If this is not a PF, we 1049 * have nothing to do. 1050 */ 1051 if (!iov) 1052 return; 1053 1054 pci_read_config_word(dev, iov->pos + PCI_SRIOV_CTRL, &cmd); 1055 if ((cmd & PCI_SRIOV_CTRL_VFE) && (cmd & PCI_SRIOV_CTRL_MSE)) { 1056 dev_WARN(&dev->dev, "can't update enabled VF BAR%d %pR\n", 1057 vf_bar, res); 1058 return; 1059 } 1060 1061 /* 1062 * Ignore unimplemented BARs, unused resource slots for 64-bit 1063 * BARs, and non-movable resources, e.g., those described via 1064 * Enhanced Allocation. 1065 */ 1066 if (!res->flags) 1067 return; 1068 1069 if (res->flags & IORESOURCE_UNSET) 1070 return; 1071 1072 if (res->flags & IORESOURCE_PCI_FIXED) 1073 return; 1074 1075 pcibios_resource_to_bus(dev->bus, ®ion, res); 1076 new = region.start; 1077 new |= res->flags & ~PCI_BASE_ADDRESS_MEM_MASK; 1078 1079 reg = iov->pos + PCI_SRIOV_BAR + 4 * vf_bar; 1080 pci_write_config_dword(dev, reg, new); 1081 if (res->flags & IORESOURCE_MEM_64) { 1082 new = region.start >> 16 >> 16; 1083 pci_write_config_dword(dev, reg + 4, new); 1084 } 1085 } 1086 1087 resource_size_t __weak pcibios_iov_resource_alignment(struct pci_dev *dev, 1088 int resno) 1089 { 1090 return pci_iov_resource_size(dev, resno); 1091 } 1092 1093 /** 1094 * pci_sriov_resource_alignment - get resource alignment for VF BAR 1095 * @dev: the PCI device 1096 * @resno: the resource number 1097 * 1098 * Returns the alignment of the VF BAR found in the SR-IOV capability. 1099 * This is not the same as the resource size which is defined as 1100 * the VF BAR size multiplied by the number of VFs. The alignment 1101 * is just the VF BAR size. 1102 */ 1103 resource_size_t pci_sriov_resource_alignment(struct pci_dev *dev, int resno) 1104 { 1105 return pcibios_iov_resource_alignment(dev, resno); 1106 } 1107 1108 /** 1109 * pci_restore_iov_state - restore the state of the IOV capability 1110 * @dev: the PCI device 1111 */ 1112 void pci_restore_iov_state(struct pci_dev *dev) 1113 { 1114 if (dev->is_physfn) { 1115 sriov_restore_vf_rebar_state(dev); 1116 sriov_restore_state(dev); 1117 } 1118 } 1119 1120 /** 1121 * pci_vf_drivers_autoprobe - set PF property drivers_autoprobe for VFs 1122 * @dev: the PCI device 1123 * @auto_probe: set VF drivers auto probe flag 1124 */ 1125 void pci_vf_drivers_autoprobe(struct pci_dev *dev, bool auto_probe) 1126 { 1127 if (dev->is_physfn) 1128 dev->sriov->drivers_autoprobe = auto_probe; 1129 } 1130 1131 /** 1132 * pci_iov_bus_range - find bus range used by Virtual Function 1133 * @bus: the PCI bus 1134 * 1135 * Returns max number of buses (exclude current one) used by Virtual 1136 * Functions. 1137 */ 1138 int pci_iov_bus_range(struct pci_bus *bus) 1139 { 1140 int max = 0; 1141 struct pci_dev *dev; 1142 1143 list_for_each_entry(dev, &bus->devices, bus_list) { 1144 if (!dev->is_physfn) 1145 continue; 1146 if (dev->sriov->max_VF_buses > max) 1147 max = dev->sriov->max_VF_buses; 1148 } 1149 1150 return max ? max - bus->number : 0; 1151 } 1152 1153 /** 1154 * pci_enable_sriov - enable the SR-IOV capability 1155 * @dev: the PCI device 1156 * @nr_virtfn: number of virtual functions to enable 1157 * 1158 * Returns 0 on success, or negative on failure. 1159 */ 1160 int pci_enable_sriov(struct pci_dev *dev, int nr_virtfn) 1161 { 1162 might_sleep(); 1163 1164 if (!dev->is_physfn) 1165 return -ENOSYS; 1166 1167 return sriov_enable(dev, nr_virtfn); 1168 } 1169 EXPORT_SYMBOL_GPL(pci_enable_sriov); 1170 1171 /** 1172 * pci_disable_sriov - disable the SR-IOV capability 1173 * @dev: the PCI device 1174 */ 1175 void pci_disable_sriov(struct pci_dev *dev) 1176 { 1177 might_sleep(); 1178 1179 if (!dev->is_physfn) 1180 return; 1181 1182 sriov_disable(dev); 1183 } 1184 EXPORT_SYMBOL_GPL(pci_disable_sriov); 1185 1186 /** 1187 * pci_num_vf - return number of VFs associated with a PF device_release_driver 1188 * @dev: the PCI device 1189 * 1190 * Returns number of VFs, or 0 if SR-IOV is not enabled. 1191 */ 1192 int pci_num_vf(struct pci_dev *dev) 1193 { 1194 if (!dev->is_physfn) 1195 return 0; 1196 1197 return dev->sriov->num_VFs; 1198 } 1199 EXPORT_SYMBOL_GPL(pci_num_vf); 1200 1201 /** 1202 * pci_vfs_assigned - returns number of VFs are assigned to a guest 1203 * @dev: the PCI device 1204 * 1205 * Returns number of VFs belonging to this device that are assigned to a guest. 1206 * If device is not a physical function returns 0. 1207 */ 1208 int pci_vfs_assigned(struct pci_dev *dev) 1209 { 1210 struct pci_dev *vfdev; 1211 unsigned int vfs_assigned = 0; 1212 unsigned short dev_id; 1213 1214 /* only search if we are a PF */ 1215 if (!dev->is_physfn) 1216 return 0; 1217 1218 /* 1219 * determine the device ID for the VFs, the vendor ID will be the 1220 * same as the PF so there is no need to check for that one 1221 */ 1222 dev_id = dev->sriov->vf_device; 1223 1224 /* loop through all the VFs to see if we own any that are assigned */ 1225 vfdev = pci_get_device(dev->vendor, dev_id, NULL); 1226 while (vfdev) { 1227 /* 1228 * It is considered assigned if it is a virtual function with 1229 * our dev as the physical function and the assigned bit is set 1230 */ 1231 if (vfdev->is_virtfn && (vfdev->physfn == dev) && 1232 pci_is_dev_assigned(vfdev)) 1233 vfs_assigned++; 1234 1235 vfdev = pci_get_device(dev->vendor, dev_id, vfdev); 1236 } 1237 1238 return vfs_assigned; 1239 } 1240 EXPORT_SYMBOL_GPL(pci_vfs_assigned); 1241 1242 /** 1243 * pci_sriov_set_totalvfs -- reduce the TotalVFs available 1244 * @dev: the PCI PF device 1245 * @numvfs: number that should be used for TotalVFs supported 1246 * 1247 * Should be called from PF driver's probe routine with 1248 * device's mutex held. 1249 * 1250 * Returns 0 if PF is an SRIOV-capable device and 1251 * value of numvfs valid. If not a PF return -ENOSYS; 1252 * if numvfs is invalid return -EINVAL; 1253 * if VFs already enabled, return -EBUSY. 1254 */ 1255 int pci_sriov_set_totalvfs(struct pci_dev *dev, u16 numvfs) 1256 { 1257 if (!dev->is_physfn) 1258 return -ENOSYS; 1259 1260 if (numvfs > dev->sriov->total_VFs) 1261 return -EINVAL; 1262 1263 /* Shouldn't change if VFs already enabled */ 1264 if (dev->sriov->ctrl & PCI_SRIOV_CTRL_VFE) 1265 return -EBUSY; 1266 1267 dev->sriov->driver_max_VFs = numvfs; 1268 return 0; 1269 } 1270 EXPORT_SYMBOL_GPL(pci_sriov_set_totalvfs); 1271 1272 /** 1273 * pci_sriov_get_totalvfs -- get total VFs supported on this device 1274 * @dev: the PCI PF device 1275 * 1276 * For a PCIe device with SRIOV support, return the PCIe 1277 * SRIOV capability value of TotalVFs or the value of driver_max_VFs 1278 * if the driver reduced it. Otherwise 0. 1279 */ 1280 int pci_sriov_get_totalvfs(struct pci_dev *dev) 1281 { 1282 if (!dev->is_physfn) 1283 return 0; 1284 1285 return dev->sriov->driver_max_VFs; 1286 } 1287 EXPORT_SYMBOL_GPL(pci_sriov_get_totalvfs); 1288 1289 /** 1290 * pci_sriov_configure_simple - helper to configure SR-IOV 1291 * @dev: the PCI device 1292 * @nr_virtfn: number of virtual functions to enable, 0 to disable 1293 * 1294 * Enable or disable SR-IOV for devices that don't require any PF setup 1295 * before enabling SR-IOV. Return value is negative on error, or number of 1296 * VFs allocated on success. 1297 */ 1298 int pci_sriov_configure_simple(struct pci_dev *dev, int nr_virtfn) 1299 { 1300 int rc; 1301 1302 might_sleep(); 1303 1304 if (!dev->is_physfn) 1305 return -ENODEV; 1306 1307 if (pci_vfs_assigned(dev)) { 1308 pci_warn(dev, "Cannot modify SR-IOV while VFs are assigned\n"); 1309 return -EPERM; 1310 } 1311 1312 if (nr_virtfn == 0) { 1313 sriov_disable(dev); 1314 return 0; 1315 } 1316 1317 rc = sriov_enable(dev, nr_virtfn); 1318 if (rc < 0) 1319 return rc; 1320 1321 return nr_virtfn; 1322 } 1323 EXPORT_SYMBOL_GPL(pci_sriov_configure_simple); 1324 1325 /** 1326 * pci_iov_vf_bar_set_size - set a new size for a VF BAR 1327 * @dev: the PCI device 1328 * @resno: the resource number 1329 * @size: new size as defined in the spec (0=1MB, 31=128TB) 1330 * 1331 * Set the new size of a VF BAR that supports VF resizable BAR capability. 1332 * Unlike pci_resize_resource(), this does not cause the resource that 1333 * reserves the MMIO space (originally up to total_VFs) to be resized, which 1334 * means that following calls to pci_enable_sriov() can fail if the resources 1335 * no longer fit. 1336 * 1337 * Return: 0 on success, or negative on failure. 1338 */ 1339 int pci_iov_vf_bar_set_size(struct pci_dev *dev, int resno, int size) 1340 { 1341 if (!pci_resource_is_iov(resno)) 1342 return -EINVAL; 1343 1344 if (pci_iov_is_memory_decoding_enabled(dev)) 1345 return -EBUSY; 1346 1347 if (!pci_rebar_size_supported(dev, resno, size)) 1348 return -EINVAL; 1349 1350 return pci_rebar_set_size(dev, resno, size); 1351 } 1352 EXPORT_SYMBOL_GPL(pci_iov_vf_bar_set_size); 1353 1354 /** 1355 * pci_iov_vf_bar_get_sizes - get VF BAR sizes allowing to create up to num_vfs 1356 * @dev: the PCI device 1357 * @resno: the resource number 1358 * @num_vfs: number of VFs 1359 * 1360 * Get the sizes of a VF resizable BAR that can accommodate @num_vfs within 1361 * the currently assigned size of the resource @resno. 1362 * 1363 * Return: A bitmask of sizes in format defined in the spec (bit 0=1MB, 1364 * bit 31=128TB). 1365 */ 1366 u32 pci_iov_vf_bar_get_sizes(struct pci_dev *dev, int resno, int num_vfs) 1367 { 1368 u64 vf_len = pci_resource_len(dev, resno); 1369 u64 sizes; 1370 1371 if (!num_vfs) 1372 return 0; 1373 1374 do_div(vf_len, num_vfs); 1375 sizes = (roundup_pow_of_two(vf_len + 1) - 1) >> ilog2(SZ_1M); 1376 1377 return sizes & pci_rebar_get_possible_sizes(dev, resno); 1378 } 1379 EXPORT_SYMBOL_GPL(pci_iov_vf_bar_get_sizes); 1380