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