drivers/pci/iov.c

   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/pci.h>
  11 #include <linux/slab.h>
  12 #include <linux/export.h>
  13 #include <linux/string.h>
  14 #include <linux/delay.h>
  15 #include "pci.h"
  16
  17 #define VIRTFN_ID_LEN   16
  18
  19 int pci_iov_virtfn_bus(struct pci_dev *dev, int vf_id)
  20 {
  21         if (!dev->is_physfn)
  22                 return -EINVAL;
  23         return dev->bus->number + ((dev->devfn + dev->sriov->offset +
  24                                     dev->sriov->stride * vf_id) >> 8);
  25 }
  26
  27 int pci_iov_virtfn_devfn(struct pci_dev *dev, int vf_id)
  28 {
  29         if (!dev->is_physfn)
  30                 return -EINVAL;
  31         return (dev->devfn + dev->sriov->offset +
  32                 dev->sriov->stride * vf_id) & 0xff;
  33 }
  34
  35 /*
  36  * Per SR-IOV spec sec 3.3.10 and 3.3.11, First VF Offset and VF Stride may
  37  * change when NumVFs changes.
  38  *
  39  * Update iov->offset and iov->stride when NumVFs is written.
  40  */
  41 static inline void pci_iov_set_numvfs(struct pci_dev *dev, int nr_virtfn)
  42 {
  43         struct pci_sriov *iov = dev->sriov;
  44
  45         pci_write_config_word(dev, iov->pos + PCI_SRIOV_NUM_VF, nr_virtfn);
  46         pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_OFFSET, &iov->offset);
  47         pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_STRIDE, &iov->stride);
  48 }
  49
  50 /*
  51  * The PF consumes one bus number.  NumVFs, First VF Offset, and VF Stride
  52  * determine how many additional bus numbers will be consumed by VFs.
  53  *
  54  * Iterate over all valid NumVFs, validate offset and stride, and calculate
  55  * the maximum number of bus numbers that could ever be required.
  56  */
  57 static int compute_max_vf_buses(struct pci_dev *dev)
  58 {
  59         struct pci_sriov *iov = dev->sriov;
  60         int nr_virtfn, busnr, rc = 0;
  61
  62         for (nr_virtfn = iov->total_VFs; nr_virtfn; nr_virtfn--) {
  63                 pci_iov_set_numvfs(dev, nr_virtfn);
  64                 if (!iov->offset || (nr_virtfn > 1 && !iov->stride)) {
  65                         rc = -EIO;
  66                         goto out;
  67                 }
  68
  69                 busnr = pci_iov_virtfn_bus(dev, nr_virtfn - 1);
  70                 if (busnr > iov->max_VF_buses)
  71                         iov->max_VF_buses = busnr;
  72         }
  73
  74 out:
  75         pci_iov_set_numvfs(dev, 0);
  76         return rc;
  77 }
  78
  79 static struct pci_bus *virtfn_add_bus(struct pci_bus *bus, int busnr)
  80 {
  81         struct pci_bus *child;
  82
  83         if (bus->number == busnr)
  84                 return bus;
  85
  86         child = pci_find_bus(pci_domain_nr(bus), busnr);
  87         if (child)
  88                 return child;
  89
  90         child = pci_add_new_bus(bus, NULL, busnr);
  91         if (!child)
  92                 return NULL;
  93
  94         pci_bus_insert_busn_res(child, busnr, busnr);
  95
  96         return child;
  97 }
  98
  99 static void virtfn_remove_bus(struct pci_bus *physbus, struct pci_bus *virtbus)
 100 {
 101         if (physbus != virtbus && list_empty(&virtbus->devices))
 102                 pci_remove_bus(virtbus);
 103 }
 104
 105 resource_size_t pci_iov_resource_size(struct pci_dev *dev, int resno)
 106 {
 107         if (!dev->is_physfn)
 108                 return 0;
 109
 110         return dev->sriov->barsz[resno - PCI_IOV_RESOURCES];
 111 }
 112
 113 static void pci_read_vf_config_common(struct pci_dev *virtfn)
 114 {
 115         struct pci_dev *physfn = virtfn->physfn;
 116
 117         /*
 118          * Some config registers are the same across all associated VFs.
 119          * Read them once from VF0 so we can skip reading them from the
 120          * other VFs.
 121          *
 122          * PCIe r4.0, sec 9.3.4.1, technically doesn't require all VFs to
 123          * have the same Revision ID and Subsystem ID, but we assume they
 124          * do.
 125          */
 126         pci_read_config_dword(virtfn, PCI_CLASS_REVISION,
 127                               &physfn->sriov->class);
 128         pci_read_config_byte(virtfn, PCI_HEADER_TYPE,
 129                              &physfn->sriov->hdr_type);
 130         pci_read_config_word(virtfn, PCI_SUBSYSTEM_VENDOR_ID,
 131                              &physfn->sriov->subsystem_vendor);
 132         pci_read_config_word(virtfn, PCI_SUBSYSTEM_ID,
 133                              &physfn->sriov->subsystem_device);
 134 }
 135
 136 int pci_iov_add_virtfn(struct pci_dev *dev, int id)
 137 {
 138         int i;
 139         int rc = -ENOMEM;
 140         u64 size;
 141         char buf[VIRTFN_ID_LEN];
 142         struct pci_dev *virtfn;
 143         struct resource *res;
 144         struct pci_sriov *iov = dev->sriov;
 145         struct pci_bus *bus;
 146
 147         bus = virtfn_add_bus(dev->bus, pci_iov_virtfn_bus(dev, id));
 148         if (!bus)
 149                 goto failed;
 150
 151         virtfn = pci_alloc_dev(bus);
 152         if (!virtfn)
 153                 goto failed0;
 154
 155         virtfn->devfn = pci_iov_virtfn_devfn(dev, id);
 156         virtfn->vendor = dev->vendor;
 157         virtfn->device = iov->vf_device;
 158         virtfn->is_virtfn = 1;
 159         virtfn->physfn = pci_dev_get(dev);
 160
 161         if (id == 0)
 162                 pci_read_vf_config_common(virtfn);
 163
 164         rc = pci_setup_device(virtfn);
 165         if (rc)
 166                 goto failed1;
 167
 168         virtfn->dev.parent = dev->dev.parent;
 169         virtfn->multifunction = 0;
 170
 171         for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
 172                 res = &dev->resource[i + PCI_IOV_RESOURCES];
 173                 if (!res->parent)
 174                         continue;
 175                 virtfn->resource[i].name = pci_name(virtfn);
 176                 virtfn->resource[i].flags = res->flags;
 177                 size = pci_iov_resource_size(dev, i + PCI_IOV_RESOURCES);
 178                 virtfn->resource[i].start = res->start + size * id;
 179                 virtfn->resource[i].end = virtfn->resource[i].start + size - 1;
 180                 rc = request_resource(res, &virtfn->resource[i]);
 181                 BUG_ON(rc);
 182         }
 183
 184         pci_device_add(virtfn, virtfn->bus);
 185
 186         sprintf(buf, "virtfn%u", id);
 187         rc = sysfs_create_link(&dev->dev.kobj, &virtfn->dev.kobj, buf);
 188         if (rc)
 189                 goto failed2;
 190         rc = sysfs_create_link(&virtfn->dev.kobj, &dev->dev.kobj, "physfn");
 191         if (rc)
 192                 goto failed3;
 193
 194         kobject_uevent(&virtfn->dev.kobj, KOBJ_CHANGE);
 195
 196         pci_bus_add_device(virtfn);
 197
 198         return 0;
 199
 200 failed3:
 201         sysfs_remove_link(&dev->dev.kobj, buf);
 202 failed2:
 203         pci_stop_and_remove_bus_device(virtfn);
 204 failed1:
 205         pci_dev_put(dev);
 206 failed0:
 207         virtfn_remove_bus(dev->bus, bus);
 208 failed:
 209
 210         return rc;
 211 }
 212
 213 void pci_iov_remove_virtfn(struct pci_dev *dev, int id)
 214 {
 215         char buf[VIRTFN_ID_LEN];
 216         struct pci_dev *virtfn;
 217
 218         virtfn = pci_get_domain_bus_and_slot(pci_domain_nr(dev->bus),
 219                                              pci_iov_virtfn_bus(dev, id),
 220                                              pci_iov_virtfn_devfn(dev, id));
 221         if (!virtfn)
 222                 return;
 223
 224         sprintf(buf, "virtfn%u", id);
 225         sysfs_remove_link(&dev->dev.kobj, buf);
 226         /*
 227          * pci_stop_dev() could have been called for this virtfn already,
 228          * so the directory for the virtfn may have been removed before.
 229          * Double check to avoid spurious sysfs warnings.
 230          */
 231         if (virtfn->dev.kobj.sd)
 232                 sysfs_remove_link(&virtfn->dev.kobj, "physfn");
 233
 234         pci_stop_and_remove_bus_device(virtfn);
 235         virtfn_remove_bus(dev->bus, virtfn->bus);
 236
 237         /* balance pci_get_domain_bus_and_slot() */
 238         pci_dev_put(virtfn);
 239         pci_dev_put(dev);
 240 }
 241
 242 static ssize_t sriov_totalvfs_show(struct device *dev,
 243                                    struct device_attribute *attr,
 244                                    char *buf)
 245 {
 246         struct pci_dev *pdev = to_pci_dev(dev);
 247
 248         return sprintf(buf, "%u\n", pci_sriov_get_totalvfs(pdev));
 249 }
 250
 251 static ssize_t sriov_numvfs_show(struct device *dev,
 252                                  struct device_attribute *attr,
 253                                  char *buf)
 254 {
 255         struct pci_dev *pdev = to_pci_dev(dev);
 256         u16 num_vfs;
 257
 258         /* Serialize vs sriov_numvfs_store() so readers see valid num_VFs */
 259         device_lock(&pdev->dev);
 260         num_vfs = pdev->sriov->num_VFs;
 261         device_unlock(&pdev->dev);
 262
 263         return sprintf(buf, "%u\n", num_vfs);
 264 }
 265
 266 /*
 267  * num_vfs > 0; number of VFs to enable
 268  * num_vfs = 0; disable all VFs
 269  *
 270  * Note: SRIOV spec does not allow partial VF
 271  *       disable, so it's all or none.
 272  */
 273 static ssize_t sriov_numvfs_store(struct device *dev,
 274                                   struct device_attribute *attr,
 275                                   const char *buf, size_t count)
 276 {
 277         struct pci_dev *pdev = to_pci_dev(dev);
 278         int ret;
 279         u16 num_vfs;
 280
 281         ret = kstrtou16(buf, 0, &num_vfs);
 282         if (ret < 0)
 283                 return ret;
 284
 285         if (num_vfs > pci_sriov_get_totalvfs(pdev))
 286                 return -ERANGE;
 287
 288         device_lock(&pdev->dev);
 289
 290         if (num_vfs == pdev->sriov->num_VFs)
 291                 goto exit;
 292
 293         /* is PF driver loaded w/callback */
 294         if (!pdev->driver || !pdev->driver->sriov_configure) {
 295                 pci_info(pdev, "Driver does not support SRIOV configuration via sysfs\n");
 296                 ret = -ENOENT;
 297                 goto exit;
 298         }
 299
 300         if (num_vfs == 0) {
 301                 /* disable VFs */
 302                 ret = pdev->driver->sriov_configure(pdev, 0);
 303                 goto exit;
 304         }
 305
 306         /* enable VFs */
 307         if (pdev->sriov->num_VFs) {
 308                 pci_warn(pdev, "%d VFs already enabled. Disable before enabling %d VFs\n",
 309                          pdev->sriov->num_VFs, num_vfs);
 310                 ret = -EBUSY;
 311                 goto exit;
 312         }
 313
 314         ret = pdev->driver->sriov_configure(pdev, num_vfs);
 315         if (ret < 0)
 316                 goto exit;
 317
 318         if (ret != num_vfs)
 319                 pci_warn(pdev, "%d VFs requested; only %d enabled\n",
 320                          num_vfs, ret);
 321
 322 exit:
 323         device_unlock(&pdev->dev);
 324
 325         if (ret < 0)
 326                 return ret;
 327
 328         return count;
 329 }
 330
 331 static ssize_t sriov_offset_show(struct device *dev,
 332                                  struct device_attribute *attr,
 333                                  char *buf)
 334 {
 335         struct pci_dev *pdev = to_pci_dev(dev);
 336
 337         return sprintf(buf, "%u\n", pdev->sriov->offset);
 338 }
 339
 340 static ssize_t sriov_stride_show(struct device *dev,
 341                                  struct device_attribute *attr,
 342                                  char *buf)
 343 {
 344         struct pci_dev *pdev = to_pci_dev(dev);
 345
 346         return sprintf(buf, "%u\n", pdev->sriov->stride);
 347 }
 348
 349 static ssize_t sriov_vf_device_show(struct device *dev,
 350                                     struct device_attribute *attr,
 351                                     char *buf)
 352 {
 353         struct pci_dev *pdev = to_pci_dev(dev);
 354
 355         return sprintf(buf, "%x\n", pdev->sriov->vf_device);
 356 }
 357
 358 static ssize_t sriov_drivers_autoprobe_show(struct device *dev,
 359                                             struct device_attribute *attr,
 360                                             char *buf)
 361 {
 362         struct pci_dev *pdev = to_pci_dev(dev);
 363
 364         return sprintf(buf, "%u\n", pdev->sriov->drivers_autoprobe);
 365 }
 366
 367 static ssize_t sriov_drivers_autoprobe_store(struct device *dev,
 368                                              struct device_attribute *attr,
 369                                              const char *buf, size_t count)
 370 {
 371         struct pci_dev *pdev = to_pci_dev(dev);
 372         bool drivers_autoprobe;
 373
 374         if (kstrtobool(buf, &drivers_autoprobe) < 0)
 375                 return -EINVAL;
 376
 377         pdev->sriov->drivers_autoprobe = drivers_autoprobe;
 378
 379         return count;
 380 }
 381
 382 static DEVICE_ATTR_RO(sriov_totalvfs);
 383 static DEVICE_ATTR_RW(sriov_numvfs);
 384 static DEVICE_ATTR_RO(sriov_offset);
 385 static DEVICE_ATTR_RO(sriov_stride);
 386 static DEVICE_ATTR_RO(sriov_vf_device);
 387 static DEVICE_ATTR_RW(sriov_drivers_autoprobe);
 388
 389 static struct attribute *sriov_dev_attrs[] = {
 390         &dev_attr_sriov_totalvfs.attr,
 391         &dev_attr_sriov_numvfs.attr,
 392         &dev_attr_sriov_offset.attr,
 393         &dev_attr_sriov_stride.attr,
 394         &dev_attr_sriov_vf_device.attr,
 395         &dev_attr_sriov_drivers_autoprobe.attr,
 396         NULL,
 397 };
 398
 399 static umode_t sriov_attrs_are_visible(struct kobject *kobj,
 400                                        struct attribute *a, int n)
 401 {
 402         struct device *dev = kobj_to_dev(kobj);
 403
 404         if (!dev_is_pf(dev))
 405                 return 0;
 406
 407         return a->mode;
 408 }
 409
 410 const struct attribute_group sriov_dev_attr_group = {
 411         .attrs = sriov_dev_attrs,
 412         .is_visible = sriov_attrs_are_visible,
 413 };
 414
 415 int __weak pcibios_sriov_enable(struct pci_dev *pdev, u16 num_vfs)
 416 {
 417         return 0;
 418 }
 419
 420 int __weak pcibios_sriov_disable(struct pci_dev *pdev)
 421 {
 422         return 0;
 423 }
 424
 425 static int sriov_add_vfs(struct pci_dev *dev, u16 num_vfs)
 426 {
 427         unsigned int i;
 428         int rc;
 429
 430         if (dev->no_vf_scan)
 431                 return 0;
 432
 433         for (i = 0; i < num_vfs; i++) {
 434                 rc = pci_iov_add_virtfn(dev, i);
 435                 if (rc)
 436                         goto failed;
 437         }
 438         return 0;
 439 failed:
 440         while (i--)
 441                 pci_iov_remove_virtfn(dev, i);
 442
 443         return rc;
 444 }
 445
 446 static int sriov_enable(struct pci_dev *dev, int nr_virtfn)
 447 {
 448         int rc;
 449         int i;
 450         int nres;
 451         u16 initial;
 452         struct resource *res;
 453         struct pci_dev *pdev;
 454         struct pci_sriov *iov = dev->sriov;
 455         int bars = 0;
 456         int bus;
 457
 458         if (!nr_virtfn)
 459                 return 0;
 460
 461         if (iov->num_VFs)
 462                 return -EINVAL;
 463
 464         pci_read_config_word(dev, iov->pos + PCI_SRIOV_INITIAL_VF, &initial);
 465         if (initial > iov->total_VFs ||
 466             (!(iov->cap & PCI_SRIOV_CAP_VFM) && (initial != iov->total_VFs)))
 467                 return -EIO;
 468
 469         if (nr_virtfn < 0 || nr_virtfn > iov->total_VFs ||
 470             (!(iov->cap & PCI_SRIOV_CAP_VFM) && (nr_virtfn > initial)))
 471                 return -EINVAL;
 472
 473         nres = 0;
 474         for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
 475                 bars |= (1 << (i + PCI_IOV_RESOURCES));
 476                 res = &dev->resource[i + PCI_IOV_RESOURCES];
 477                 if (res->parent)
 478                         nres++;
 479         }
 480         if (nres != iov->nres) {
 481                 pci_err(dev, "not enough MMIO resources for SR-IOV\n");
 482                 return -ENOMEM;
 483         }
 484
 485         bus = pci_iov_virtfn_bus(dev, nr_virtfn - 1);
 486         if (bus > dev->bus->busn_res.end) {
 487                 pci_err(dev, "can't enable %d VFs (bus %02x out of range of %pR)\n",
 488                         nr_virtfn, bus, &dev->bus->busn_res);
 489                 return -ENOMEM;
 490         }
 491
 492         if (pci_enable_resources(dev, bars)) {
 493                 pci_err(dev, "SR-IOV: IOV BARS not allocated\n");
 494                 return -ENOMEM;
 495         }
 496
 497         if (iov->link != dev->devfn) {
 498                 pdev = pci_get_slot(dev->bus, iov->link);
 499                 if (!pdev)
 500                         return -ENODEV;
 501
 502                 if (!pdev->is_physfn) {
 503                         pci_dev_put(pdev);
 504                         return -ENOSYS;
 505                 }
 506
 507                 rc = sysfs_create_link(&dev->dev.kobj,
 508                                         &pdev->dev.kobj, "dep_link");
 509                 pci_dev_put(pdev);
 510                 if (rc)
 511                         return rc;
 512         }
 513
 514         iov->initial_VFs = initial;
 515         if (nr_virtfn < initial)
 516                 initial = nr_virtfn;
 517
 518         rc = pcibios_sriov_enable(dev, initial);
 519         if (rc) {
 520                 pci_err(dev, "failure %d from pcibios_sriov_enable()\n", rc);
 521                 goto err_pcibios;
 522         }
 523
 524         pci_iov_set_numvfs(dev, nr_virtfn);
 525         iov->ctrl |= PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE;
 526         pci_cfg_access_lock(dev);
 527         pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
 528         msleep(100);
 529         pci_cfg_access_unlock(dev);
 530
 531         rc = sriov_add_vfs(dev, initial);
 532         if (rc)
 533                 goto err_pcibios;
 534
 535         kobject_uevent(&dev->dev.kobj, KOBJ_CHANGE);
 536         iov->num_VFs = nr_virtfn;
 537
 538         return 0;
 539
 540 err_pcibios:
 541         iov->ctrl &= ~(PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE);
 542         pci_cfg_access_lock(dev);
 543         pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
 544         ssleep(1);
 545         pci_cfg_access_unlock(dev);
 546
 547         pcibios_sriov_disable(dev);
 548
 549         if (iov->link != dev->devfn)
 550                 sysfs_remove_link(&dev->dev.kobj, "dep_link");
 551
 552         pci_iov_set_numvfs(dev, 0);
 553         return rc;
 554 }
 555
 556 static void sriov_del_vfs(struct pci_dev *dev)
 557 {
 558         struct pci_sriov *iov = dev->sriov;
 559         int i;
 560
 561         if (dev->no_vf_scan)
 562                 return;
 563
 564         for (i = 0; i < iov->num_VFs; i++)
 565                 pci_iov_remove_virtfn(dev, i);
 566 }
 567
 568 static void sriov_disable(struct pci_dev *dev)
 569 {
 570         struct pci_sriov *iov = dev->sriov;
 571
 572         if (!iov->num_VFs)
 573                 return;
 574
 575         sriov_del_vfs(dev);
 576         iov->ctrl &= ~(PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE);
 577         pci_cfg_access_lock(dev);
 578         pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
 579         ssleep(1);
 580         pci_cfg_access_unlock(dev);
 581
 582         pcibios_sriov_disable(dev);
 583
 584         if (iov->link != dev->devfn)
 585                 sysfs_remove_link(&dev->dev.kobj, "dep_link");
 586
 587         iov->num_VFs = 0;
 588         pci_iov_set_numvfs(dev, 0);
 589 }
 590
 591 static int sriov_init(struct pci_dev *dev, int pos)
 592 {
 593         int i, bar64;
 594         int rc;
 595         int nres;
 596         u32 pgsz;
 597         u16 ctrl, total;
 598         struct pci_sriov *iov;
 599         struct resource *res;
 600         struct pci_dev *pdev;
 601
 602         pci_read_config_word(dev, pos + PCI_SRIOV_CTRL, &ctrl);
 603         if (ctrl & PCI_SRIOV_CTRL_VFE) {
 604                 pci_write_config_word(dev, pos + PCI_SRIOV_CTRL, 0);
 605                 ssleep(1);
 606         }
 607
 608         ctrl = 0;
 609         list_for_each_entry(pdev, &dev->bus->devices, bus_list)
 610                 if (pdev->is_physfn)
 611                         goto found;
 612
 613         pdev = NULL;
 614         if (pci_ari_enabled(dev->bus))
 615                 ctrl |= PCI_SRIOV_CTRL_ARI;
 616
 617 found:
 618         pci_write_config_word(dev, pos + PCI_SRIOV_CTRL, ctrl);
 619
 620         pci_read_config_word(dev, pos + PCI_SRIOV_TOTAL_VF, &total);
 621         if (!total)
 622                 return 0;
 623
 624         pci_read_config_dword(dev, pos + PCI_SRIOV_SUP_PGSIZE, &pgsz);
 625         i = PAGE_SHIFT > 12 ? PAGE_SHIFT - 12 : 0;
 626         pgsz &= ~((1 << i) - 1);
 627         if (!pgsz)
 628                 return -EIO;
 629
 630         pgsz &= ~(pgsz - 1);
 631         pci_write_config_dword(dev, pos + PCI_SRIOV_SYS_PGSIZE, pgsz);
 632
 633         iov = kzalloc(sizeof(*iov), GFP_KERNEL);
 634         if (!iov)
 635                 return -ENOMEM;
 636
 637         nres = 0;
 638         for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
 639                 res = &dev->resource[i + PCI_IOV_RESOURCES];
 640                 /*
 641                  * If it is already FIXED, don't change it, something
 642                  * (perhaps EA or header fixups) wants it this way.
 643                  */
 644                 if (res->flags & IORESOURCE_PCI_FIXED)
 645                         bar64 = (res->flags & IORESOURCE_MEM_64) ? 1 : 0;
 646                 else
 647                         bar64 = __pci_read_base(dev, pci_bar_unknown, res,
 648                                                 pos + PCI_SRIOV_BAR + i * 4);
 649                 if (!res->flags)
 650                         continue;
 651                 if (resource_size(res) & (PAGE_SIZE - 1)) {
 652                         rc = -EIO;
 653                         goto failed;
 654                 }
 655                 iov->barsz[i] = resource_size(res);
 656                 res->end = res->start + resource_size(res) * total - 1;
 657                 pci_info(dev, "VF(n) BAR%d space: %pR (contains BAR%d for %d VFs)\n",
 658                          i, res, i, total);
 659                 i += bar64;
 660                 nres++;
 661         }
 662
 663         iov->pos = pos;
 664         iov->nres = nres;
 665         iov->ctrl = ctrl;
 666         iov->total_VFs = total;
 667         iov->driver_max_VFs = total;
 668         pci_read_config_word(dev, pos + PCI_SRIOV_VF_DID, &iov->vf_device);
 669         iov->pgsz = pgsz;
 670         iov->self = dev;
 671         iov->drivers_autoprobe = true;
 672         pci_read_config_dword(dev, pos + PCI_SRIOV_CAP, &iov->cap);
 673         pci_read_config_byte(dev, pos + PCI_SRIOV_FUNC_LINK, &iov->link);
 674         if (pci_pcie_type(dev) == PCI_EXP_TYPE_RC_END)
 675                 iov->link = PCI_DEVFN(PCI_SLOT(dev->devfn), iov->link);
 676
 677         if (pdev)
 678                 iov->dev = pci_dev_get(pdev);
 679         else
 680                 iov->dev = dev;
 681
 682         dev->sriov = iov;
 683         dev->is_physfn = 1;
 684         rc = compute_max_vf_buses(dev);
 685         if (rc)
 686                 goto fail_max_buses;
 687
 688         return 0;
 689
 690 fail_max_buses:
 691         dev->sriov = NULL;
 692         dev->is_physfn = 0;
 693 failed:
 694         for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
 695                 res = &dev->resource[i + PCI_IOV_RESOURCES];
 696                 res->flags = 0;
 697         }
 698
 699         kfree(iov);
 700         return rc;
 701 }
 702
 703 static void sriov_release(struct pci_dev *dev)
 704 {
 705         BUG_ON(dev->sriov->num_VFs);
 706
 707         if (dev != dev->sriov->dev)
 708                 pci_dev_put(dev->sriov->dev);
 709
 710         kfree(dev->sriov);
 711         dev->sriov = NULL;
 712 }
 713
 714 static void sriov_restore_state(struct pci_dev *dev)
 715 {
 716         int i;
 717         u16 ctrl;
 718         struct pci_sriov *iov = dev->sriov;
 719
 720         pci_read_config_word(dev, iov->pos + PCI_SRIOV_CTRL, &ctrl);
 721         if (ctrl & PCI_SRIOV_CTRL_VFE)
 722                 return;
 723
 724         /*
 725          * Restore PCI_SRIOV_CTRL_ARI before pci_iov_set_numvfs() because
 726          * it reads offset & stride, which depend on PCI_SRIOV_CTRL_ARI.
 727          */
 728         ctrl &= ~PCI_SRIOV_CTRL_ARI;
 729         ctrl |= iov->ctrl & PCI_SRIOV_CTRL_ARI;
 730         pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, ctrl);
 731
 732         for (i = 0; i < PCI_SRIOV_NUM_BARS; i++)
 733                 pci_update_resource(dev, i + PCI_IOV_RESOURCES);
 734
 735         pci_write_config_dword(dev, iov->pos + PCI_SRIOV_SYS_PGSIZE, iov->pgsz);
 736         pci_iov_set_numvfs(dev, iov->num_VFs);
 737         pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
 738         if (iov->ctrl & PCI_SRIOV_CTRL_VFE)
 739                 msleep(100);
 740 }
 741
 742 /**
 743  * pci_iov_init - initialize the IOV capability
 744  * @dev: the PCI device
 745  *
 746  * Returns 0 on success, or negative on failure.
 747  */
 748 int pci_iov_init(struct pci_dev *dev)
 749 {
 750         int pos;
 751
 752         if (!pci_is_pcie(dev))
 753                 return -ENODEV;
 754
 755         pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_SRIOV);
 756         if (pos)
 757                 return sriov_init(dev, pos);
 758
 759         return -ENODEV;
 760 }
 761
 762 /**
 763  * pci_iov_release - release resources used by the IOV capability
 764  * @dev: the PCI device
 765  */
 766 void pci_iov_release(struct pci_dev *dev)
 767 {
 768         if (dev->is_physfn)
 769                 sriov_release(dev);
 770 }
 771
 772 /**
 773  * pci_iov_remove - clean up SR-IOV state after PF driver is detached
 774  * @dev: the PCI device
 775  */
 776 void pci_iov_remove(struct pci_dev *dev)
 777 {
 778         struct pci_sriov *iov = dev->sriov;
 779
 780         if (!dev->is_physfn)
 781                 return;
 782
 783         iov->driver_max_VFs = iov->total_VFs;
 784         if (iov->num_VFs)
 785                 pci_warn(dev, "driver left SR-IOV enabled after remove\n");
 786 }
 787
 788 /**
 789  * pci_iov_update_resource - update a VF BAR
 790  * @dev: the PCI device
 791  * @resno: the resource number
 792  *
 793  * Update a VF BAR in the SR-IOV capability of a PF.
 794  */
 795 void pci_iov_update_resource(struct pci_dev *dev, int resno)
 796 {
 797         struct pci_sriov *iov = dev->is_physfn ? dev->sriov : NULL;
 798         struct resource *res = dev->resource + resno;
 799         int vf_bar = resno - PCI_IOV_RESOURCES;
 800         struct pci_bus_region region;
 801         u16 cmd;
 802         u32 new;
 803         int reg;
 804
 805         /*
 806          * The generic pci_restore_bars() path calls this for all devices,
 807          * including VFs and non-SR-IOV devices.  If this is not a PF, we
 808          * have nothing to do.
 809          */
 810         if (!iov)
 811                 return;
 812
 813         pci_read_config_word(dev, iov->pos + PCI_SRIOV_CTRL, &cmd);
 814         if ((cmd & PCI_SRIOV_CTRL_VFE) && (cmd & PCI_SRIOV_CTRL_MSE)) {
 815                 dev_WARN(&dev->dev, "can't update enabled VF BAR%d %pR\n",
 816                          vf_bar, res);
 817                 return;
 818         }
 819
 820         /*
 821          * Ignore unimplemented BARs, unused resource slots for 64-bit
 822          * BARs, and non-movable resources, e.g., those described via
 823          * Enhanced Allocation.
 824          */
 825         if (!res->flags)
 826                 return;
 827
 828         if (res->flags & IORESOURCE_UNSET)
 829                 return;
 830
 831         if (res->flags & IORESOURCE_PCI_FIXED)
 832                 return;
 833
 834         pcibios_resource_to_bus(dev->bus, &region, res);
 835         new = region.start;
 836         new |= res->flags & ~PCI_BASE_ADDRESS_MEM_MASK;
 837
 838         reg = iov->pos + PCI_SRIOV_BAR + 4 * vf_bar;
 839         pci_write_config_dword(dev, reg, new);
 840         if (res->flags & IORESOURCE_MEM_64) {
 841                 new = region.start >> 16 >> 16;
 842                 pci_write_config_dword(dev, reg + 4, new);
 843         }
 844 }
 845
 846 resource_size_t __weak pcibios_iov_resource_alignment(struct pci_dev *dev,
 847                                                       int resno)
 848 {
 849         return pci_iov_resource_size(dev, resno);
 850 }
 851
 852 /**
 853  * pci_sriov_resource_alignment - get resource alignment for VF BAR
 854  * @dev: the PCI device
 855  * @resno: the resource number
 856  *
 857  * Returns the alignment of the VF BAR found in the SR-IOV capability.
 858  * This is not the same as the resource size which is defined as
 859  * the VF BAR size multiplied by the number of VFs.  The alignment
 860  * is just the VF BAR size.
 861  */
 862 resource_size_t pci_sriov_resource_alignment(struct pci_dev *dev, int resno)
 863 {
 864         return pcibios_iov_resource_alignment(dev, resno);
 865 }
 866
 867 /**
 868  * pci_restore_iov_state - restore the state of the IOV capability
 869  * @dev: the PCI device
 870  */
 871 void pci_restore_iov_state(struct pci_dev *dev)
 872 {
 873         if (dev->is_physfn)
 874                 sriov_restore_state(dev);
 875 }
 876
 877 /**
 878  * pci_vf_drivers_autoprobe - set PF property drivers_autoprobe for VFs
 879  * @dev: the PCI device
 880  * @auto_probe: set VF drivers auto probe flag
 881  */
 882 void pci_vf_drivers_autoprobe(struct pci_dev *dev, bool auto_probe)
 883 {
 884         if (dev->is_physfn)
 885                 dev->sriov->drivers_autoprobe = auto_probe;
 886 }
 887
 888 /**
 889  * pci_iov_bus_range - find bus range used by Virtual Function
 890  * @bus: the PCI bus
 891  *
 892  * Returns max number of buses (exclude current one) used by Virtual
 893  * Functions.
 894  */
 895 int pci_iov_bus_range(struct pci_bus *bus)
 896 {
 897         int max = 0;
 898         struct pci_dev *dev;
 899
 900         list_for_each_entry(dev, &bus->devices, bus_list) {
 901                 if (!dev->is_physfn)
 902                         continue;
 903                 if (dev->sriov->max_VF_buses > max)
 904                         max = dev->sriov->max_VF_buses;
 905         }
 906
 907         return max ? max - bus->number : 0;
 908 }
 909
 910 /**
 911  * pci_enable_sriov - enable the SR-IOV capability
 912  * @dev: the PCI device
 913  * @nr_virtfn: number of virtual functions to enable
 914  *
 915  * Returns 0 on success, or negative on failure.
 916  */
 917 int pci_enable_sriov(struct pci_dev *dev, int nr_virtfn)
 918 {
 919         might_sleep();
 920
 921         if (!dev->is_physfn)
 922                 return -ENOSYS;
 923
 924         return sriov_enable(dev, nr_virtfn);
 925 }
 926 EXPORT_SYMBOL_GPL(pci_enable_sriov);
 927
 928 /**
 929  * pci_disable_sriov - disable the SR-IOV capability
 930  * @dev: the PCI device
 931  */
 932 void pci_disable_sriov(struct pci_dev *dev)
 933 {
 934         might_sleep();
 935
 936         if (!dev->is_physfn)
 937                 return;
 938
 939         sriov_disable(dev);
 940 }
 941 EXPORT_SYMBOL_GPL(pci_disable_sriov);
 942
 943 /**
 944  * pci_num_vf - return number of VFs associated with a PF device_release_driver
 945  * @dev: the PCI device
 946  *
 947  * Returns number of VFs, or 0 if SR-IOV is not enabled.
 948  */
 949 int pci_num_vf(struct pci_dev *dev)
 950 {
 951         if (!dev->is_physfn)
 952                 return 0;
 953
 954         return dev->sriov->num_VFs;
 955 }
 956 EXPORT_SYMBOL_GPL(pci_num_vf);
 957
 958 /**
 959  * pci_vfs_assigned - returns number of VFs are assigned to a guest
 960  * @dev: the PCI device
 961  *
 962  * Returns number of VFs belonging to this device that are assigned to a guest.
 963  * If device is not a physical function returns 0.
 964  */
 965 int pci_vfs_assigned(struct pci_dev *dev)
 966 {
 967         struct pci_dev *vfdev;
 968         unsigned int vfs_assigned = 0;
 969         unsigned short dev_id;
 970
 971         /* only search if we are a PF */
 972         if (!dev->is_physfn)
 973                 return 0;
 974
 975         /*
 976          * determine the device ID for the VFs, the vendor ID will be the
 977          * same as the PF so there is no need to check for that one
 978          */
 979         dev_id = dev->sriov->vf_device;
 980
 981         /* loop through all the VFs to see if we own any that are assigned */
 982         vfdev = pci_get_device(dev->vendor, dev_id, NULL);
 983         while (vfdev) {
 984                 /*
 985                  * It is considered assigned if it is a virtual function with
 986                  * our dev as the physical function and the assigned bit is set
 987                  */
 988                 if (vfdev->is_virtfn && (vfdev->physfn == dev) &&
 989                         pci_is_dev_assigned(vfdev))
 990                         vfs_assigned++;
 991
 992                 vfdev = pci_get_device(dev->vendor, dev_id, vfdev);
 993         }
 994
 995         return vfs_assigned;
 996 }
 997 EXPORT_SYMBOL_GPL(pci_vfs_assigned);
 998
 999 /**
1000  * pci_sriov_set_totalvfs -- reduce the TotalVFs available
1001  * @dev: the PCI PF device
1002  * @numvfs: number that should be used for TotalVFs supported
1003  *
1004  * Should be called from PF driver's probe routine with
1005  * device's mutex held.
1006  *
1007  * Returns 0 if PF is an SRIOV-capable device and
1008  * value of numvfs valid. If not a PF return -ENOSYS;
1009  * if numvfs is invalid return -EINVAL;
1010  * if VFs already enabled, return -EBUSY.
1011  */
1012 int pci_sriov_set_totalvfs(struct pci_dev *dev, u16 numvfs)
1013 {
1014         if (!dev->is_physfn)
1015                 return -ENOSYS;
1016
1017         if (numvfs > dev->sriov->total_VFs)
1018                 return -EINVAL;
1019
1020         /* Shouldn't change if VFs already enabled */
1021         if (dev->sriov->ctrl & PCI_SRIOV_CTRL_VFE)
1022                 return -EBUSY;
1023
1024         dev->sriov->driver_max_VFs = numvfs;
1025         return 0;
1026 }
1027 EXPORT_SYMBOL_GPL(pci_sriov_set_totalvfs);
1028
1029 /**
1030  * pci_sriov_get_totalvfs -- get total VFs supported on this device
1031  * @dev: the PCI PF device
1032  *
1033  * For a PCIe device with SRIOV support, return the PCIe
1034  * SRIOV capability value of TotalVFs or the value of driver_max_VFs
1035  * if the driver reduced it.  Otherwise 0.
1036  */
1037 int pci_sriov_get_totalvfs(struct pci_dev *dev)
1038 {
1039         if (!dev->is_physfn)
1040                 return 0;
1041
1042         return dev->sriov->driver_max_VFs;
1043 }
1044 EXPORT_SYMBOL_GPL(pci_sriov_get_totalvfs);
1045
1046 /**
1047  * pci_sriov_configure_simple - helper to configure SR-IOV
1048  * @dev: the PCI device
1049  * @nr_virtfn: number of virtual functions to enable, 0 to disable
1050  *
1051  * Enable or disable SR-IOV for devices that don't require any PF setup
1052  * before enabling SR-IOV.  Return value is negative on error, or number of
1053  * VFs allocated on success.
1054  */
1055 int pci_sriov_configure_simple(struct pci_dev *dev, int nr_virtfn)
1056 {
1057         int rc;
1058
1059         might_sleep();
1060
1061         if (!dev->is_physfn)
1062                 return -ENODEV;
1063
1064         if (pci_vfs_assigned(dev)) {
1065                 pci_warn(dev, "Cannot modify SR-IOV while VFs are assigned\n");
1066                 return -EPERM;
1067         }
1068
1069         if (nr_virtfn == 0) {
1070                 sriov_disable(dev);
1071                 return 0;
1072         }
1073
1074         rc = sriov_enable(dev, nr_virtfn);
1075         if (rc < 0)
1076                 return rc;
1077
1078         return nr_virtfn;
1079 }
1080 EXPORT_SYMBOL_GPL(pci_sriov_configure_simple);