Linux Audio

Check our new training course

Loading...
v5.4
   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/mutex.h>
  13#include <linux/export.h>
  14#include <linux/string.h>
  15#include <linux/delay.h>
  16#include "pci.h"
  17
  18#define VIRTFN_ID_LEN	16
  19
  20int pci_iov_virtfn_bus(struct pci_dev *dev, int vf_id)
  21{
  22	if (!dev->is_physfn)
  23		return -EINVAL;
  24	return dev->bus->number + ((dev->devfn + dev->sriov->offset +
  25				    dev->sriov->stride * vf_id) >> 8);
  26}
  27
  28int pci_iov_virtfn_devfn(struct pci_dev *dev, int vf_id)
  29{
  30	if (!dev->is_physfn)
  31		return -EINVAL;
  32	return (dev->devfn + dev->sriov->offset +
  33		dev->sriov->stride * vf_id) & 0xff;
  34}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  35
  36/*
  37 * Per SR-IOV spec sec 3.3.10 and 3.3.11, First VF Offset and VF Stride may
  38 * change when NumVFs changes.
  39 *
  40 * Update iov->offset and iov->stride when NumVFs is written.
  41 */
  42static inline void pci_iov_set_numvfs(struct pci_dev *dev, int nr_virtfn)
  43{
  44	struct pci_sriov *iov = dev->sriov;
  45
  46	pci_write_config_word(dev, iov->pos + PCI_SRIOV_NUM_VF, nr_virtfn);
  47	pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_OFFSET, &iov->offset);
  48	pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_STRIDE, &iov->stride);
  49}
  50
  51/*
  52 * The PF consumes one bus number.  NumVFs, First VF Offset, and VF Stride
  53 * determine how many additional bus numbers will be consumed by VFs.
  54 *
  55 * Iterate over all valid NumVFs, validate offset and stride, and calculate
  56 * the maximum number of bus numbers that could ever be required.
  57 */
  58static int compute_max_vf_buses(struct pci_dev *dev)
  59{
  60	struct pci_sriov *iov = dev->sriov;
  61	int nr_virtfn, busnr, rc = 0;
  62
  63	for (nr_virtfn = iov->total_VFs; nr_virtfn; nr_virtfn--) {
  64		pci_iov_set_numvfs(dev, nr_virtfn);
  65		if (!iov->offset || (nr_virtfn > 1 && !iov->stride)) {
  66			rc = -EIO;
  67			goto out;
  68		}
  69
  70		busnr = pci_iov_virtfn_bus(dev, nr_virtfn - 1);
  71		if (busnr > iov->max_VF_buses)
  72			iov->max_VF_buses = busnr;
  73	}
  74
  75out:
  76	pci_iov_set_numvfs(dev, 0);
  77	return rc;
  78}
  79
  80static struct pci_bus *virtfn_add_bus(struct pci_bus *bus, int busnr)
  81{
  82	struct pci_bus *child;
  83
  84	if (bus->number == busnr)
  85		return bus;
  86
  87	child = pci_find_bus(pci_domain_nr(bus), busnr);
  88	if (child)
  89		return child;
  90
  91	child = pci_add_new_bus(bus, NULL, busnr);
  92	if (!child)
  93		return NULL;
  94
  95	pci_bus_insert_busn_res(child, busnr, busnr);
  96
  97	return child;
  98}
  99
 100static void virtfn_remove_bus(struct pci_bus *physbus, struct pci_bus *virtbus)
 101{
 102	if (physbus != virtbus && list_empty(&virtbus->devices))
 103		pci_remove_bus(virtbus);
 104}
 105
 106resource_size_t pci_iov_resource_size(struct pci_dev *dev, int resno)
 107{
 108	if (!dev->is_physfn)
 109		return 0;
 110
 111	return dev->sriov->barsz[resno - PCI_IOV_RESOURCES];
 112}
 113
 114static void pci_read_vf_config_common(struct pci_dev *virtfn)
 115{
 116	struct pci_dev *physfn = virtfn->physfn;
 117
 118	/*
 119	 * Some config registers are the same across all associated VFs.
 120	 * Read them once from VF0 so we can skip reading them from the
 121	 * other VFs.
 122	 *
 123	 * PCIe r4.0, sec 9.3.4.1, technically doesn't require all VFs to
 124	 * have the same Revision ID and Subsystem ID, but we assume they
 125	 * do.
 126	 */
 127	pci_read_config_dword(virtfn, PCI_CLASS_REVISION,
 128			      &physfn->sriov->class);
 129	pci_read_config_byte(virtfn, PCI_HEADER_TYPE,
 130			     &physfn->sriov->hdr_type);
 131	pci_read_config_word(virtfn, PCI_SUBSYSTEM_VENDOR_ID,
 132			     &physfn->sriov->subsystem_vendor);
 133	pci_read_config_word(virtfn, PCI_SUBSYSTEM_ID,
 134			     &physfn->sriov->subsystem_device);
 135}
 136
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 137int pci_iov_add_virtfn(struct pci_dev *dev, int id)
 138{
 139	int i;
 140	int rc = -ENOMEM;
 141	u64 size;
 142	char buf[VIRTFN_ID_LEN];
 143	struct pci_dev *virtfn;
 144	struct resource *res;
 145	struct pci_sriov *iov = dev->sriov;
 146	struct pci_bus *bus;
 147
 148	bus = virtfn_add_bus(dev->bus, pci_iov_virtfn_bus(dev, id));
 149	if (!bus)
 150		goto failed;
 151
 152	virtfn = pci_alloc_dev(bus);
 153	if (!virtfn)
 154		goto failed0;
 155
 156	virtfn->devfn = pci_iov_virtfn_devfn(dev, id);
 157	virtfn->vendor = dev->vendor;
 158	virtfn->device = iov->vf_device;
 159	virtfn->is_virtfn = 1;
 160	virtfn->physfn = pci_dev_get(dev);
 
 161
 162	if (id == 0)
 163		pci_read_vf_config_common(virtfn);
 164
 165	rc = pci_setup_device(virtfn);
 166	if (rc)
 167		goto failed1;
 168
 169	virtfn->dev.parent = dev->dev.parent;
 170	virtfn->multifunction = 0;
 171
 172	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
 173		res = &dev->resource[i + PCI_IOV_RESOURCES];
 174		if (!res->parent)
 175			continue;
 176		virtfn->resource[i].name = pci_name(virtfn);
 177		virtfn->resource[i].flags = res->flags;
 178		size = pci_iov_resource_size(dev, i + PCI_IOV_RESOURCES);
 179		virtfn->resource[i].start = res->start + size * id;
 180		virtfn->resource[i].end = virtfn->resource[i].start + size - 1;
 181		rc = request_resource(res, &virtfn->resource[i]);
 182		BUG_ON(rc);
 183	}
 184
 185	pci_device_add(virtfn, virtfn->bus);
 186
 187	sprintf(buf, "virtfn%u", id);
 188	rc = sysfs_create_link(&dev->dev.kobj, &virtfn->dev.kobj, buf);
 189	if (rc)
 190		goto failed2;
 191	rc = sysfs_create_link(&virtfn->dev.kobj, &dev->dev.kobj, "physfn");
 192	if (rc)
 193		goto failed3;
 194
 195	kobject_uevent(&virtfn->dev.kobj, KOBJ_CHANGE);
 196
 197	pci_bus_add_device(virtfn);
 198
 199	return 0;
 200
 201failed3:
 202	sysfs_remove_link(&dev->dev.kobj, buf);
 203failed2:
 204	pci_stop_and_remove_bus_device(virtfn);
 205failed1:
 
 206	pci_dev_put(dev);
 207failed0:
 208	virtfn_remove_bus(dev->bus, bus);
 209failed:
 210
 211	return rc;
 212}
 213
 214void pci_iov_remove_virtfn(struct pci_dev *dev, int id)
 215{
 216	char buf[VIRTFN_ID_LEN];
 217	struct pci_dev *virtfn;
 218
 219	virtfn = pci_get_domain_bus_and_slot(pci_domain_nr(dev->bus),
 220					     pci_iov_virtfn_bus(dev, id),
 221					     pci_iov_virtfn_devfn(dev, id));
 222	if (!virtfn)
 223		return;
 224
 225	sprintf(buf, "virtfn%u", id);
 226	sysfs_remove_link(&dev->dev.kobj, buf);
 227	/*
 228	 * pci_stop_dev() could have been called for this virtfn already,
 229	 * so the directory for the virtfn may have been removed before.
 230	 * Double check to avoid spurious sysfs warnings.
 231	 */
 232	if (virtfn->dev.kobj.sd)
 233		sysfs_remove_link(&virtfn->dev.kobj, "physfn");
 234
 235	pci_stop_and_remove_bus_device(virtfn);
 236	virtfn_remove_bus(dev->bus, virtfn->bus);
 237
 238	/* balance pci_get_domain_bus_and_slot() */
 239	pci_dev_put(virtfn);
 240	pci_dev_put(dev);
 241}
 242
 243static ssize_t sriov_totalvfs_show(struct device *dev,
 244				   struct device_attribute *attr,
 245				   char *buf)
 246{
 247	struct pci_dev *pdev = to_pci_dev(dev);
 248
 249	return sprintf(buf, "%u\n", pci_sriov_get_totalvfs(pdev));
 250}
 251
 252static ssize_t sriov_numvfs_show(struct device *dev,
 253				 struct device_attribute *attr,
 254				 char *buf)
 255{
 256	struct pci_dev *pdev = to_pci_dev(dev);
 
 
 
 
 
 
 257
 258	return sprintf(buf, "%u\n", pdev->sriov->num_VFs);
 259}
 260
 261/*
 262 * num_vfs > 0; number of VFs to enable
 263 * num_vfs = 0; disable all VFs
 264 *
 265 * Note: SRIOV spec does not allow partial VF
 266 *	 disable, so it's all or none.
 267 */
 268static ssize_t sriov_numvfs_store(struct device *dev,
 269				  struct device_attribute *attr,
 270				  const char *buf, size_t count)
 271{
 272	struct pci_dev *pdev = to_pci_dev(dev);
 273	int ret;
 274	u16 num_vfs;
 275
 276	ret = kstrtou16(buf, 0, &num_vfs);
 277	if (ret < 0)
 278		return ret;
 279
 280	if (num_vfs > pci_sriov_get_totalvfs(pdev))
 281		return -ERANGE;
 282
 283	device_lock(&pdev->dev);
 284
 285	if (num_vfs == pdev->sriov->num_VFs)
 286		goto exit;
 287
 
 
 
 
 
 
 
 288	/* is PF driver loaded w/callback */
 289	if (!pdev->driver || !pdev->driver->sriov_configure) {
 290		pci_info(pdev, "Driver does not support SRIOV configuration via sysfs\n");
 291		ret = -ENOENT;
 292		goto exit;
 293	}
 294
 295	if (num_vfs == 0) {
 296		/* disable VFs */
 297		ret = pdev->driver->sriov_configure(pdev, 0);
 298		goto exit;
 299	}
 300
 301	/* enable VFs */
 302	if (pdev->sriov->num_VFs) {
 303		pci_warn(pdev, "%d VFs already enabled. Disable before enabling %d VFs\n",
 304			 pdev->sriov->num_VFs, num_vfs);
 305		ret = -EBUSY;
 306		goto exit;
 307	}
 308
 309	ret = pdev->driver->sriov_configure(pdev, num_vfs);
 310	if (ret < 0)
 311		goto exit;
 312
 313	if (ret != num_vfs)
 314		pci_warn(pdev, "%d VFs requested; only %d enabled\n",
 315			 num_vfs, ret);
 316
 317exit:
 318	device_unlock(&pdev->dev);
 319
 320	if (ret < 0)
 321		return ret;
 322
 323	return count;
 324}
 325
 326static ssize_t sriov_offset_show(struct device *dev,
 327				 struct device_attribute *attr,
 328				 char *buf)
 329{
 330	struct pci_dev *pdev = to_pci_dev(dev);
 331
 332	return sprintf(buf, "%u\n", pdev->sriov->offset);
 333}
 334
 335static ssize_t sriov_stride_show(struct device *dev,
 336				 struct device_attribute *attr,
 337				 char *buf)
 338{
 339	struct pci_dev *pdev = to_pci_dev(dev);
 340
 341	return sprintf(buf, "%u\n", pdev->sriov->stride);
 342}
 343
 344static ssize_t sriov_vf_device_show(struct device *dev,
 345				    struct device_attribute *attr,
 346				    char *buf)
 347{
 348	struct pci_dev *pdev = to_pci_dev(dev);
 349
 350	return sprintf(buf, "%x\n", pdev->sriov->vf_device);
 351}
 352
 353static ssize_t sriov_drivers_autoprobe_show(struct device *dev,
 354					    struct device_attribute *attr,
 355					    char *buf)
 356{
 357	struct pci_dev *pdev = to_pci_dev(dev);
 358
 359	return sprintf(buf, "%u\n", pdev->sriov->drivers_autoprobe);
 360}
 361
 362static ssize_t sriov_drivers_autoprobe_store(struct device *dev,
 363					     struct device_attribute *attr,
 364					     const char *buf, size_t count)
 365{
 366	struct pci_dev *pdev = to_pci_dev(dev);
 367	bool drivers_autoprobe;
 368
 369	if (kstrtobool(buf, &drivers_autoprobe) < 0)
 370		return -EINVAL;
 371
 372	pdev->sriov->drivers_autoprobe = drivers_autoprobe;
 373
 374	return count;
 375}
 376
 377static DEVICE_ATTR_RO(sriov_totalvfs);
 378static DEVICE_ATTR_RW(sriov_numvfs);
 379static DEVICE_ATTR_RO(sriov_offset);
 380static DEVICE_ATTR_RO(sriov_stride);
 381static DEVICE_ATTR_RO(sriov_vf_device);
 382static DEVICE_ATTR_RW(sriov_drivers_autoprobe);
 383
 384static struct attribute *sriov_dev_attrs[] = {
 385	&dev_attr_sriov_totalvfs.attr,
 386	&dev_attr_sriov_numvfs.attr,
 387	&dev_attr_sriov_offset.attr,
 388	&dev_attr_sriov_stride.attr,
 389	&dev_attr_sriov_vf_device.attr,
 390	&dev_attr_sriov_drivers_autoprobe.attr,
 
 
 
 391	NULL,
 392};
 393
 394static umode_t sriov_attrs_are_visible(struct kobject *kobj,
 395				       struct attribute *a, int n)
 396{
 397	struct device *dev = kobj_to_dev(kobj);
 398
 399	if (!dev_is_pf(dev))
 400		return 0;
 401
 402	return a->mode;
 403}
 404
 405const struct attribute_group sriov_dev_attr_group = {
 406	.attrs = sriov_dev_attrs,
 407	.is_visible = sriov_attrs_are_visible,
 408};
 409
 410int __weak pcibios_sriov_enable(struct pci_dev *pdev, u16 num_vfs)
 411{
 412	return 0;
 413}
 414
 415int __weak pcibios_sriov_disable(struct pci_dev *pdev)
 416{
 417	return 0;
 418}
 419
 420static int sriov_add_vfs(struct pci_dev *dev, u16 num_vfs)
 421{
 422	unsigned int i;
 423	int rc;
 424
 425	if (dev->no_vf_scan)
 426		return 0;
 427
 428	for (i = 0; i < num_vfs; i++) {
 429		rc = pci_iov_add_virtfn(dev, i);
 430		if (rc)
 431			goto failed;
 432	}
 433	return 0;
 434failed:
 435	while (i--)
 436		pci_iov_remove_virtfn(dev, i);
 437
 438	return rc;
 439}
 440
 441static int sriov_enable(struct pci_dev *dev, int nr_virtfn)
 442{
 443	int rc;
 444	int i;
 445	int nres;
 446	u16 initial;
 447	struct resource *res;
 448	struct pci_dev *pdev;
 449	struct pci_sriov *iov = dev->sriov;
 450	int bars = 0;
 451	int bus;
 452
 453	if (!nr_virtfn)
 454		return 0;
 455
 456	if (iov->num_VFs)
 457		return -EINVAL;
 458
 459	pci_read_config_word(dev, iov->pos + PCI_SRIOV_INITIAL_VF, &initial);
 460	if (initial > iov->total_VFs ||
 461	    (!(iov->cap & PCI_SRIOV_CAP_VFM) && (initial != iov->total_VFs)))
 462		return -EIO;
 463
 464	if (nr_virtfn < 0 || nr_virtfn > iov->total_VFs ||
 465	    (!(iov->cap & PCI_SRIOV_CAP_VFM) && (nr_virtfn > initial)))
 466		return -EINVAL;
 467
 468	nres = 0;
 469	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
 470		bars |= (1 << (i + PCI_IOV_RESOURCES));
 471		res = &dev->resource[i + PCI_IOV_RESOURCES];
 472		if (res->parent)
 473			nres++;
 474	}
 475	if (nres != iov->nres) {
 476		pci_err(dev, "not enough MMIO resources for SR-IOV\n");
 477		return -ENOMEM;
 478	}
 479
 480	bus = pci_iov_virtfn_bus(dev, nr_virtfn - 1);
 481	if (bus > dev->bus->busn_res.end) {
 482		pci_err(dev, "can't enable %d VFs (bus %02x out of range of %pR)\n",
 483			nr_virtfn, bus, &dev->bus->busn_res);
 484		return -ENOMEM;
 485	}
 486
 487	if (pci_enable_resources(dev, bars)) {
 488		pci_err(dev, "SR-IOV: IOV BARS not allocated\n");
 489		return -ENOMEM;
 490	}
 491
 492	if (iov->link != dev->devfn) {
 493		pdev = pci_get_slot(dev->bus, iov->link);
 494		if (!pdev)
 495			return -ENODEV;
 496
 497		if (!pdev->is_physfn) {
 498			pci_dev_put(pdev);
 499			return -ENOSYS;
 500		}
 501
 502		rc = sysfs_create_link(&dev->dev.kobj,
 503					&pdev->dev.kobj, "dep_link");
 504		pci_dev_put(pdev);
 505		if (rc)
 506			return rc;
 507	}
 508
 509	iov->initial_VFs = initial;
 510	if (nr_virtfn < initial)
 511		initial = nr_virtfn;
 512
 513	rc = pcibios_sriov_enable(dev, initial);
 514	if (rc) {
 515		pci_err(dev, "failure %d from pcibios_sriov_enable()\n", rc);
 516		goto err_pcibios;
 517	}
 518
 519	pci_iov_set_numvfs(dev, nr_virtfn);
 520	iov->ctrl |= PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE;
 521	pci_cfg_access_lock(dev);
 522	pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
 523	msleep(100);
 524	pci_cfg_access_unlock(dev);
 525
 526	rc = sriov_add_vfs(dev, initial);
 527	if (rc)
 528		goto err_pcibios;
 529
 530	kobject_uevent(&dev->dev.kobj, KOBJ_CHANGE);
 531	iov->num_VFs = nr_virtfn;
 532
 533	return 0;
 534
 535err_pcibios:
 536	iov->ctrl &= ~(PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE);
 537	pci_cfg_access_lock(dev);
 538	pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
 539	ssleep(1);
 540	pci_cfg_access_unlock(dev);
 541
 542	pcibios_sriov_disable(dev);
 543
 544	if (iov->link != dev->devfn)
 545		sysfs_remove_link(&dev->dev.kobj, "dep_link");
 546
 547	pci_iov_set_numvfs(dev, 0);
 548	return rc;
 549}
 550
 551static void sriov_del_vfs(struct pci_dev *dev)
 552{
 553	struct pci_sriov *iov = dev->sriov;
 554	int i;
 555
 556	if (dev->no_vf_scan)
 557		return;
 558
 559	for (i = 0; i < iov->num_VFs; i++)
 560		pci_iov_remove_virtfn(dev, i);
 561}
 562
 563static void sriov_disable(struct pci_dev *dev)
 564{
 565	struct pci_sriov *iov = dev->sriov;
 566
 567	if (!iov->num_VFs)
 568		return;
 569
 570	sriov_del_vfs(dev);
 571	iov->ctrl &= ~(PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE);
 572	pci_cfg_access_lock(dev);
 573	pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
 574	ssleep(1);
 575	pci_cfg_access_unlock(dev);
 576
 577	pcibios_sriov_disable(dev);
 578
 579	if (iov->link != dev->devfn)
 580		sysfs_remove_link(&dev->dev.kobj, "dep_link");
 581
 582	iov->num_VFs = 0;
 583	pci_iov_set_numvfs(dev, 0);
 584}
 585
 586static int sriov_init(struct pci_dev *dev, int pos)
 587{
 588	int i, bar64;
 589	int rc;
 590	int nres;
 591	u32 pgsz;
 592	u16 ctrl, total;
 593	struct pci_sriov *iov;
 594	struct resource *res;
 595	struct pci_dev *pdev;
 596
 597	pci_read_config_word(dev, pos + PCI_SRIOV_CTRL, &ctrl);
 598	if (ctrl & PCI_SRIOV_CTRL_VFE) {
 599		pci_write_config_word(dev, pos + PCI_SRIOV_CTRL, 0);
 600		ssleep(1);
 601	}
 602
 603	ctrl = 0;
 604	list_for_each_entry(pdev, &dev->bus->devices, bus_list)
 605		if (pdev->is_physfn)
 606			goto found;
 607
 608	pdev = NULL;
 609	if (pci_ari_enabled(dev->bus))
 610		ctrl |= PCI_SRIOV_CTRL_ARI;
 611
 612found:
 613	pci_write_config_word(dev, pos + PCI_SRIOV_CTRL, ctrl);
 614
 615	pci_read_config_word(dev, pos + PCI_SRIOV_TOTAL_VF, &total);
 616	if (!total)
 617		return 0;
 618
 619	pci_read_config_dword(dev, pos + PCI_SRIOV_SUP_PGSIZE, &pgsz);
 620	i = PAGE_SHIFT > 12 ? PAGE_SHIFT - 12 : 0;
 621	pgsz &= ~((1 << i) - 1);
 622	if (!pgsz)
 623		return -EIO;
 624
 625	pgsz &= ~(pgsz - 1);
 626	pci_write_config_dword(dev, pos + PCI_SRIOV_SYS_PGSIZE, pgsz);
 627
 628	iov = kzalloc(sizeof(*iov), GFP_KERNEL);
 629	if (!iov)
 630		return -ENOMEM;
 631
 632	nres = 0;
 633	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
 634		res = &dev->resource[i + PCI_IOV_RESOURCES];
 635		/*
 636		 * If it is already FIXED, don't change it, something
 637		 * (perhaps EA or header fixups) wants it this way.
 638		 */
 639		if (res->flags & IORESOURCE_PCI_FIXED)
 640			bar64 = (res->flags & IORESOURCE_MEM_64) ? 1 : 0;
 641		else
 642			bar64 = __pci_read_base(dev, pci_bar_unknown, res,
 643						pos + PCI_SRIOV_BAR + i * 4);
 644		if (!res->flags)
 645			continue;
 646		if (resource_size(res) & (PAGE_SIZE - 1)) {
 647			rc = -EIO;
 648			goto failed;
 649		}
 650		iov->barsz[i] = resource_size(res);
 651		res->end = res->start + resource_size(res) * total - 1;
 652		pci_info(dev, "VF(n) BAR%d space: %pR (contains BAR%d for %d VFs)\n",
 653			 i, res, i, total);
 654		i += bar64;
 655		nres++;
 656	}
 657
 658	iov->pos = pos;
 659	iov->nres = nres;
 660	iov->ctrl = ctrl;
 661	iov->total_VFs = total;
 662	iov->driver_max_VFs = total;
 663	pci_read_config_word(dev, pos + PCI_SRIOV_VF_DID, &iov->vf_device);
 664	iov->pgsz = pgsz;
 665	iov->self = dev;
 666	iov->drivers_autoprobe = true;
 667	pci_read_config_dword(dev, pos + PCI_SRIOV_CAP, &iov->cap);
 668	pci_read_config_byte(dev, pos + PCI_SRIOV_FUNC_LINK, &iov->link);
 669	if (pci_pcie_type(dev) == PCI_EXP_TYPE_RC_END)
 670		iov->link = PCI_DEVFN(PCI_SLOT(dev->devfn), iov->link);
 671
 672	if (pdev)
 673		iov->dev = pci_dev_get(pdev);
 674	else
 675		iov->dev = dev;
 676
 677	dev->sriov = iov;
 678	dev->is_physfn = 1;
 679	rc = compute_max_vf_buses(dev);
 680	if (rc)
 681		goto fail_max_buses;
 682
 683	return 0;
 684
 685fail_max_buses:
 686	dev->sriov = NULL;
 687	dev->is_physfn = 0;
 688failed:
 689	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
 690		res = &dev->resource[i + PCI_IOV_RESOURCES];
 691		res->flags = 0;
 692	}
 693
 694	kfree(iov);
 695	return rc;
 696}
 697
 698static void sriov_release(struct pci_dev *dev)
 699{
 700	BUG_ON(dev->sriov->num_VFs);
 701
 702	if (dev != dev->sriov->dev)
 703		pci_dev_put(dev->sriov->dev);
 704
 705	kfree(dev->sriov);
 706	dev->sriov = NULL;
 707}
 708
 709static void sriov_restore_state(struct pci_dev *dev)
 710{
 711	int i;
 712	u16 ctrl;
 713	struct pci_sriov *iov = dev->sriov;
 714
 715	pci_read_config_word(dev, iov->pos + PCI_SRIOV_CTRL, &ctrl);
 716	if (ctrl & PCI_SRIOV_CTRL_VFE)
 717		return;
 718
 719	/*
 720	 * Restore PCI_SRIOV_CTRL_ARI before pci_iov_set_numvfs() because
 721	 * it reads offset & stride, which depend on PCI_SRIOV_CTRL_ARI.
 722	 */
 723	ctrl &= ~PCI_SRIOV_CTRL_ARI;
 724	ctrl |= iov->ctrl & PCI_SRIOV_CTRL_ARI;
 725	pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, ctrl);
 726
 727	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++)
 728		pci_update_resource(dev, i + PCI_IOV_RESOURCES);
 729
 730	pci_write_config_dword(dev, iov->pos + PCI_SRIOV_SYS_PGSIZE, iov->pgsz);
 731	pci_iov_set_numvfs(dev, iov->num_VFs);
 732	pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
 733	if (iov->ctrl & PCI_SRIOV_CTRL_VFE)
 734		msleep(100);
 735}
 736
 737/**
 738 * pci_iov_init - initialize the IOV capability
 739 * @dev: the PCI device
 740 *
 741 * Returns 0 on success, or negative on failure.
 742 */
 743int pci_iov_init(struct pci_dev *dev)
 744{
 745	int pos;
 746
 747	if (!pci_is_pcie(dev))
 748		return -ENODEV;
 749
 750	pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_SRIOV);
 751	if (pos)
 752		return sriov_init(dev, pos);
 753
 754	return -ENODEV;
 755}
 756
 757/**
 758 * pci_iov_release - release resources used by the IOV capability
 759 * @dev: the PCI device
 760 */
 761void pci_iov_release(struct pci_dev *dev)
 762{
 763	if (dev->is_physfn)
 764		sriov_release(dev);
 765}
 766
 767/**
 768 * pci_iov_remove - clean up SR-IOV state after PF driver is detached
 769 * @dev: the PCI device
 770 */
 771void pci_iov_remove(struct pci_dev *dev)
 772{
 773	struct pci_sriov *iov = dev->sriov;
 774
 775	if (!dev->is_physfn)
 776		return;
 777
 778	iov->driver_max_VFs = iov->total_VFs;
 779	if (iov->num_VFs)
 780		pci_warn(dev, "driver left SR-IOV enabled after remove\n");
 781}
 782
 783/**
 784 * pci_iov_update_resource - update a VF BAR
 785 * @dev: the PCI device
 786 * @resno: the resource number
 787 *
 788 * Update a VF BAR in the SR-IOV capability of a PF.
 789 */
 790void pci_iov_update_resource(struct pci_dev *dev, int resno)
 791{
 792	struct pci_sriov *iov = dev->is_physfn ? dev->sriov : NULL;
 793	struct resource *res = dev->resource + resno;
 794	int vf_bar = resno - PCI_IOV_RESOURCES;
 795	struct pci_bus_region region;
 796	u16 cmd;
 797	u32 new;
 798	int reg;
 799
 800	/*
 801	 * The generic pci_restore_bars() path calls this for all devices,
 802	 * including VFs and non-SR-IOV devices.  If this is not a PF, we
 803	 * have nothing to do.
 804	 */
 805	if (!iov)
 806		return;
 807
 808	pci_read_config_word(dev, iov->pos + PCI_SRIOV_CTRL, &cmd);
 809	if ((cmd & PCI_SRIOV_CTRL_VFE) && (cmd & PCI_SRIOV_CTRL_MSE)) {
 810		dev_WARN(&dev->dev, "can't update enabled VF BAR%d %pR\n",
 811			 vf_bar, res);
 812		return;
 813	}
 814
 815	/*
 816	 * Ignore unimplemented BARs, unused resource slots for 64-bit
 817	 * BARs, and non-movable resources, e.g., those described via
 818	 * Enhanced Allocation.
 819	 */
 820	if (!res->flags)
 821		return;
 822
 823	if (res->flags & IORESOURCE_UNSET)
 824		return;
 825
 826	if (res->flags & IORESOURCE_PCI_FIXED)
 827		return;
 828
 829	pcibios_resource_to_bus(dev->bus, &region, res);
 830	new = region.start;
 831	new |= res->flags & ~PCI_BASE_ADDRESS_MEM_MASK;
 832
 833	reg = iov->pos + PCI_SRIOV_BAR + 4 * vf_bar;
 834	pci_write_config_dword(dev, reg, new);
 835	if (res->flags & IORESOURCE_MEM_64) {
 836		new = region.start >> 16 >> 16;
 837		pci_write_config_dword(dev, reg + 4, new);
 838	}
 839}
 840
 841resource_size_t __weak pcibios_iov_resource_alignment(struct pci_dev *dev,
 842						      int resno)
 843{
 844	return pci_iov_resource_size(dev, resno);
 845}
 846
 847/**
 848 * pci_sriov_resource_alignment - get resource alignment for VF BAR
 849 * @dev: the PCI device
 850 * @resno: the resource number
 851 *
 852 * Returns the alignment of the VF BAR found in the SR-IOV capability.
 853 * This is not the same as the resource size which is defined as
 854 * the VF BAR size multiplied by the number of VFs.  The alignment
 855 * is just the VF BAR size.
 856 */
 857resource_size_t pci_sriov_resource_alignment(struct pci_dev *dev, int resno)
 858{
 859	return pcibios_iov_resource_alignment(dev, resno);
 860}
 861
 862/**
 863 * pci_restore_iov_state - restore the state of the IOV capability
 864 * @dev: the PCI device
 865 */
 866void pci_restore_iov_state(struct pci_dev *dev)
 867{
 868	if (dev->is_physfn)
 869		sriov_restore_state(dev);
 870}
 871
 872/**
 873 * pci_vf_drivers_autoprobe - set PF property drivers_autoprobe for VFs
 874 * @dev: the PCI device
 875 * @auto_probe: set VF drivers auto probe flag
 876 */
 877void pci_vf_drivers_autoprobe(struct pci_dev *dev, bool auto_probe)
 878{
 879	if (dev->is_physfn)
 880		dev->sriov->drivers_autoprobe = auto_probe;
 881}
 882
 883/**
 884 * pci_iov_bus_range - find bus range used by Virtual Function
 885 * @bus: the PCI bus
 886 *
 887 * Returns max number of buses (exclude current one) used by Virtual
 888 * Functions.
 889 */
 890int pci_iov_bus_range(struct pci_bus *bus)
 891{
 892	int max = 0;
 893	struct pci_dev *dev;
 894
 895	list_for_each_entry(dev, &bus->devices, bus_list) {
 896		if (!dev->is_physfn)
 897			continue;
 898		if (dev->sriov->max_VF_buses > max)
 899			max = dev->sriov->max_VF_buses;
 900	}
 901
 902	return max ? max - bus->number : 0;
 903}
 904
 905/**
 906 * pci_enable_sriov - enable the SR-IOV capability
 907 * @dev: the PCI device
 908 * @nr_virtfn: number of virtual functions to enable
 909 *
 910 * Returns 0 on success, or negative on failure.
 911 */
 912int pci_enable_sriov(struct pci_dev *dev, int nr_virtfn)
 913{
 914	might_sleep();
 915
 916	if (!dev->is_physfn)
 917		return -ENOSYS;
 918
 919	return sriov_enable(dev, nr_virtfn);
 920}
 921EXPORT_SYMBOL_GPL(pci_enable_sriov);
 922
 923/**
 924 * pci_disable_sriov - disable the SR-IOV capability
 925 * @dev: the PCI device
 926 */
 927void pci_disable_sriov(struct pci_dev *dev)
 928{
 929	might_sleep();
 930
 931	if (!dev->is_physfn)
 932		return;
 933
 934	sriov_disable(dev);
 935}
 936EXPORT_SYMBOL_GPL(pci_disable_sriov);
 937
 938/**
 939 * pci_num_vf - return number of VFs associated with a PF device_release_driver
 940 * @dev: the PCI device
 941 *
 942 * Returns number of VFs, or 0 if SR-IOV is not enabled.
 943 */
 944int pci_num_vf(struct pci_dev *dev)
 945{
 946	if (!dev->is_physfn)
 947		return 0;
 948
 949	return dev->sriov->num_VFs;
 950}
 951EXPORT_SYMBOL_GPL(pci_num_vf);
 952
 953/**
 954 * pci_vfs_assigned - returns number of VFs are assigned to a guest
 955 * @dev: the PCI device
 956 *
 957 * Returns number of VFs belonging to this device that are assigned to a guest.
 958 * If device is not a physical function returns 0.
 959 */
 960int pci_vfs_assigned(struct pci_dev *dev)
 961{
 962	struct pci_dev *vfdev;
 963	unsigned int vfs_assigned = 0;
 964	unsigned short dev_id;
 965
 966	/* only search if we are a PF */
 967	if (!dev->is_physfn)
 968		return 0;
 969
 970	/*
 971	 * determine the device ID for the VFs, the vendor ID will be the
 972	 * same as the PF so there is no need to check for that one
 973	 */
 974	dev_id = dev->sriov->vf_device;
 975
 976	/* loop through all the VFs to see if we own any that are assigned */
 977	vfdev = pci_get_device(dev->vendor, dev_id, NULL);
 978	while (vfdev) {
 979		/*
 980		 * It is considered assigned if it is a virtual function with
 981		 * our dev as the physical function and the assigned bit is set
 982		 */
 983		if (vfdev->is_virtfn && (vfdev->physfn == dev) &&
 984			pci_is_dev_assigned(vfdev))
 985			vfs_assigned++;
 986
 987		vfdev = pci_get_device(dev->vendor, dev_id, vfdev);
 988	}
 989
 990	return vfs_assigned;
 991}
 992EXPORT_SYMBOL_GPL(pci_vfs_assigned);
 993
 994/**
 995 * pci_sriov_set_totalvfs -- reduce the TotalVFs available
 996 * @dev: the PCI PF device
 997 * @numvfs: number that should be used for TotalVFs supported
 998 *
 999 * Should be called from PF driver's probe routine with
1000 * device's mutex held.
1001 *
1002 * Returns 0 if PF is an SRIOV-capable device and
1003 * value of numvfs valid. If not a PF return -ENOSYS;
1004 * if numvfs is invalid return -EINVAL;
1005 * if VFs already enabled, return -EBUSY.
1006 */
1007int pci_sriov_set_totalvfs(struct pci_dev *dev, u16 numvfs)
1008{
1009	if (!dev->is_physfn)
1010		return -ENOSYS;
1011
1012	if (numvfs > dev->sriov->total_VFs)
1013		return -EINVAL;
1014
1015	/* Shouldn't change if VFs already enabled */
1016	if (dev->sriov->ctrl & PCI_SRIOV_CTRL_VFE)
1017		return -EBUSY;
1018
1019	dev->sriov->driver_max_VFs = numvfs;
1020	return 0;
1021}
1022EXPORT_SYMBOL_GPL(pci_sriov_set_totalvfs);
1023
1024/**
1025 * pci_sriov_get_totalvfs -- get total VFs supported on this device
1026 * @dev: the PCI PF device
1027 *
1028 * For a PCIe device with SRIOV support, return the PCIe
1029 * SRIOV capability value of TotalVFs or the value of driver_max_VFs
1030 * if the driver reduced it.  Otherwise 0.
1031 */
1032int pci_sriov_get_totalvfs(struct pci_dev *dev)
1033{
1034	if (!dev->is_physfn)
1035		return 0;
1036
1037	return dev->sriov->driver_max_VFs;
1038}
1039EXPORT_SYMBOL_GPL(pci_sriov_get_totalvfs);
1040
1041/**
1042 * pci_sriov_configure_simple - helper to configure SR-IOV
1043 * @dev: the PCI device
1044 * @nr_virtfn: number of virtual functions to enable, 0 to disable
1045 *
1046 * Enable or disable SR-IOV for devices that don't require any PF setup
1047 * before enabling SR-IOV.  Return value is negative on error, or number of
1048 * VFs allocated on success.
1049 */
1050int pci_sriov_configure_simple(struct pci_dev *dev, int nr_virtfn)
1051{
1052	int rc;
1053
1054	might_sleep();
1055
1056	if (!dev->is_physfn)
1057		return -ENODEV;
1058
1059	if (pci_vfs_assigned(dev)) {
1060		pci_warn(dev, "Cannot modify SR-IOV while VFs are assigned\n");
1061		return -EPERM;
1062	}
1063
1064	if (nr_virtfn == 0) {
1065		sriov_disable(dev);
1066		return 0;
1067	}
1068
1069	rc = sriov_enable(dev, nr_virtfn);
1070	if (rc < 0)
1071		return rc;
1072
1073	return nr_virtfn;
1074}
1075EXPORT_SYMBOL_GPL(pci_sriov_configure_simple);
v6.2
   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
  19int 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
  27int 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}
  34EXPORT_SYMBOL_GPL(pci_iov_virtfn_devfn);
  35
  36int pci_iov_vf_id(struct pci_dev *dev)
  37{
  38	struct pci_dev *pf;
  39
  40	if (!dev->is_virtfn)
  41		return -EINVAL;
  42
  43	pf = pci_physfn(dev);
  44	return (((dev->bus->number << 8) + dev->devfn) -
  45		((pf->bus->number << 8) + pf->devfn + pf->sriov->offset)) /
  46	       pf->sriov->stride;
  47}
  48EXPORT_SYMBOL_GPL(pci_iov_vf_id);
  49
  50/**
  51 * pci_iov_get_pf_drvdata - Return the drvdata of a PF
  52 * @dev: VF pci_dev
  53 * @pf_driver: Device driver required to own the PF
  54 *
  55 * This must be called from a context that ensures that a VF driver is attached.
  56 * The value returned is invalid once the VF driver completes its remove()
  57 * callback.
  58 *
  59 * Locking is achieved by the driver core. A VF driver cannot be probed until
  60 * pci_enable_sriov() is called and pci_disable_sriov() does not return until
  61 * all VF drivers have completed their remove().
  62 *
  63 * The PF driver must call pci_disable_sriov() before it begins to destroy the
  64 * drvdata.
  65 */
  66void *pci_iov_get_pf_drvdata(struct pci_dev *dev, struct pci_driver *pf_driver)
  67{
  68	struct pci_dev *pf_dev;
  69
  70	if (!dev->is_virtfn)
  71		return ERR_PTR(-EINVAL);
  72	pf_dev = dev->physfn;
  73	if (pf_dev->driver != pf_driver)
  74		return ERR_PTR(-EINVAL);
  75	return pci_get_drvdata(pf_dev);
  76}
  77EXPORT_SYMBOL_GPL(pci_iov_get_pf_drvdata);
  78
  79/*
  80 * Per SR-IOV spec sec 3.3.10 and 3.3.11, First VF Offset and VF Stride may
  81 * change when NumVFs changes.
  82 *
  83 * Update iov->offset and iov->stride when NumVFs is written.
  84 */
  85static inline void pci_iov_set_numvfs(struct pci_dev *dev, int nr_virtfn)
  86{
  87	struct pci_sriov *iov = dev->sriov;
  88
  89	pci_write_config_word(dev, iov->pos + PCI_SRIOV_NUM_VF, nr_virtfn);
  90	pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_OFFSET, &iov->offset);
  91	pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_STRIDE, &iov->stride);
  92}
  93
  94/*
  95 * The PF consumes one bus number.  NumVFs, First VF Offset, and VF Stride
  96 * determine how many additional bus numbers will be consumed by VFs.
  97 *
  98 * Iterate over all valid NumVFs, validate offset and stride, and calculate
  99 * the maximum number of bus numbers that could ever be required.
 100 */
 101static int compute_max_vf_buses(struct pci_dev *dev)
 102{
 103	struct pci_sriov *iov = dev->sriov;
 104	int nr_virtfn, busnr, rc = 0;
 105
 106	for (nr_virtfn = iov->total_VFs; nr_virtfn; nr_virtfn--) {
 107		pci_iov_set_numvfs(dev, nr_virtfn);
 108		if (!iov->offset || (nr_virtfn > 1 && !iov->stride)) {
 109			rc = -EIO;
 110			goto out;
 111		}
 112
 113		busnr = pci_iov_virtfn_bus(dev, nr_virtfn - 1);
 114		if (busnr > iov->max_VF_buses)
 115			iov->max_VF_buses = busnr;
 116	}
 117
 118out:
 119	pci_iov_set_numvfs(dev, 0);
 120	return rc;
 121}
 122
 123static struct pci_bus *virtfn_add_bus(struct pci_bus *bus, int busnr)
 124{
 125	struct pci_bus *child;
 126
 127	if (bus->number == busnr)
 128		return bus;
 129
 130	child = pci_find_bus(pci_domain_nr(bus), busnr);
 131	if (child)
 132		return child;
 133
 134	child = pci_add_new_bus(bus, NULL, busnr);
 135	if (!child)
 136		return NULL;
 137
 138	pci_bus_insert_busn_res(child, busnr, busnr);
 139
 140	return child;
 141}
 142
 143static void virtfn_remove_bus(struct pci_bus *physbus, struct pci_bus *virtbus)
 144{
 145	if (physbus != virtbus && list_empty(&virtbus->devices))
 146		pci_remove_bus(virtbus);
 147}
 148
 149resource_size_t pci_iov_resource_size(struct pci_dev *dev, int resno)
 150{
 151	if (!dev->is_physfn)
 152		return 0;
 153
 154	return dev->sriov->barsz[resno - PCI_IOV_RESOURCES];
 155}
 156
 157static void pci_read_vf_config_common(struct pci_dev *virtfn)
 158{
 159	struct pci_dev *physfn = virtfn->physfn;
 160
 161	/*
 162	 * Some config registers are the same across all associated VFs.
 163	 * Read them once from VF0 so we can skip reading them from the
 164	 * other VFs.
 165	 *
 166	 * PCIe r4.0, sec 9.3.4.1, technically doesn't require all VFs to
 167	 * have the same Revision ID and Subsystem ID, but we assume they
 168	 * do.
 169	 */
 170	pci_read_config_dword(virtfn, PCI_CLASS_REVISION,
 171			      &physfn->sriov->class);
 172	pci_read_config_byte(virtfn, PCI_HEADER_TYPE,
 173			     &physfn->sriov->hdr_type);
 174	pci_read_config_word(virtfn, PCI_SUBSYSTEM_VENDOR_ID,
 175			     &physfn->sriov->subsystem_vendor);
 176	pci_read_config_word(virtfn, PCI_SUBSYSTEM_ID,
 177			     &physfn->sriov->subsystem_device);
 178}
 179
 180int pci_iov_sysfs_link(struct pci_dev *dev,
 181		struct pci_dev *virtfn, int id)
 182{
 183	char buf[VIRTFN_ID_LEN];
 184	int rc;
 185
 186	sprintf(buf, "virtfn%u", id);
 187	rc = sysfs_create_link(&dev->dev.kobj, &virtfn->dev.kobj, buf);
 188	if (rc)
 189		goto failed;
 190	rc = sysfs_create_link(&virtfn->dev.kobj, &dev->dev.kobj, "physfn");
 191	if (rc)
 192		goto failed1;
 193
 194	kobject_uevent(&virtfn->dev.kobj, KOBJ_CHANGE);
 195
 196	return 0;
 197
 198failed1:
 199	sysfs_remove_link(&dev->dev.kobj, buf);
 200failed:
 201	return rc;
 202}
 203
 204#ifdef CONFIG_PCI_MSI
 205static ssize_t sriov_vf_total_msix_show(struct device *dev,
 206					struct device_attribute *attr,
 207					char *buf)
 208{
 209	struct pci_dev *pdev = to_pci_dev(dev);
 210	u32 vf_total_msix = 0;
 211
 212	device_lock(dev);
 213	if (!pdev->driver || !pdev->driver->sriov_get_vf_total_msix)
 214		goto unlock;
 215
 216	vf_total_msix = pdev->driver->sriov_get_vf_total_msix(pdev);
 217unlock:
 218	device_unlock(dev);
 219	return sysfs_emit(buf, "%u\n", vf_total_msix);
 220}
 221static DEVICE_ATTR_RO(sriov_vf_total_msix);
 222
 223static ssize_t sriov_vf_msix_count_store(struct device *dev,
 224					 struct device_attribute *attr,
 225					 const char *buf, size_t count)
 226{
 227	struct pci_dev *vf_dev = to_pci_dev(dev);
 228	struct pci_dev *pdev = pci_physfn(vf_dev);
 229	int val, ret = 0;
 230
 231	if (kstrtoint(buf, 0, &val) < 0)
 232		return -EINVAL;
 233
 234	if (val < 0)
 235		return -EINVAL;
 236
 237	device_lock(&pdev->dev);
 238	if (!pdev->driver || !pdev->driver->sriov_set_msix_vec_count) {
 239		ret = -EOPNOTSUPP;
 240		goto err_pdev;
 241	}
 242
 243	device_lock(&vf_dev->dev);
 244	if (vf_dev->driver) {
 245		/*
 246		 * A driver is already attached to this VF and has configured
 247		 * itself based on the current MSI-X vector count. Changing
 248		 * the vector size could mess up the driver, so block it.
 249		 */
 250		ret = -EBUSY;
 251		goto err_dev;
 252	}
 253
 254	ret = pdev->driver->sriov_set_msix_vec_count(vf_dev, val);
 255
 256err_dev:
 257	device_unlock(&vf_dev->dev);
 258err_pdev:
 259	device_unlock(&pdev->dev);
 260	return ret ? : count;
 261}
 262static DEVICE_ATTR_WO(sriov_vf_msix_count);
 263#endif
 264
 265static struct attribute *sriov_vf_dev_attrs[] = {
 266#ifdef CONFIG_PCI_MSI
 267	&dev_attr_sriov_vf_msix_count.attr,
 268#endif
 269	NULL,
 270};
 271
 272static umode_t sriov_vf_attrs_are_visible(struct kobject *kobj,
 273					  struct attribute *a, int n)
 274{
 275	struct device *dev = kobj_to_dev(kobj);
 276	struct pci_dev *pdev = to_pci_dev(dev);
 277
 278	if (!pdev->is_virtfn)
 279		return 0;
 280
 281	return a->mode;
 282}
 283
 284const struct attribute_group sriov_vf_dev_attr_group = {
 285	.attrs = sriov_vf_dev_attrs,
 286	.is_visible = sriov_vf_attrs_are_visible,
 287};
 288
 289int pci_iov_add_virtfn(struct pci_dev *dev, int id)
 290{
 291	int i;
 292	int rc = -ENOMEM;
 293	u64 size;
 
 294	struct pci_dev *virtfn;
 295	struct resource *res;
 296	struct pci_sriov *iov = dev->sriov;
 297	struct pci_bus *bus;
 298
 299	bus = virtfn_add_bus(dev->bus, pci_iov_virtfn_bus(dev, id));
 300	if (!bus)
 301		goto failed;
 302
 303	virtfn = pci_alloc_dev(bus);
 304	if (!virtfn)
 305		goto failed0;
 306
 307	virtfn->devfn = pci_iov_virtfn_devfn(dev, id);
 308	virtfn->vendor = dev->vendor;
 309	virtfn->device = iov->vf_device;
 310	virtfn->is_virtfn = 1;
 311	virtfn->physfn = pci_dev_get(dev);
 312	virtfn->no_command_memory = 1;
 313
 314	if (id == 0)
 315		pci_read_vf_config_common(virtfn);
 316
 317	rc = pci_setup_device(virtfn);
 318	if (rc)
 319		goto failed1;
 320
 321	virtfn->dev.parent = dev->dev.parent;
 322	virtfn->multifunction = 0;
 323
 324	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
 325		res = &dev->resource[i + PCI_IOV_RESOURCES];
 326		if (!res->parent)
 327			continue;
 328		virtfn->resource[i].name = pci_name(virtfn);
 329		virtfn->resource[i].flags = res->flags;
 330		size = pci_iov_resource_size(dev, i + PCI_IOV_RESOURCES);
 331		virtfn->resource[i].start = res->start + size * id;
 332		virtfn->resource[i].end = virtfn->resource[i].start + size - 1;
 333		rc = request_resource(res, &virtfn->resource[i]);
 334		BUG_ON(rc);
 335	}
 336
 337	pci_device_add(virtfn, virtfn->bus);
 338	rc = pci_iov_sysfs_link(dev, virtfn, id);
 
 
 
 
 
 339	if (rc)
 340		goto failed1;
 
 
 341
 342	pci_bus_add_device(virtfn);
 343
 344	return 0;
 345
 
 
 
 
 346failed1:
 347	pci_stop_and_remove_bus_device(virtfn);
 348	pci_dev_put(dev);
 349failed0:
 350	virtfn_remove_bus(dev->bus, bus);
 351failed:
 352
 353	return rc;
 354}
 355
 356void pci_iov_remove_virtfn(struct pci_dev *dev, int id)
 357{
 358	char buf[VIRTFN_ID_LEN];
 359	struct pci_dev *virtfn;
 360
 361	virtfn = pci_get_domain_bus_and_slot(pci_domain_nr(dev->bus),
 362					     pci_iov_virtfn_bus(dev, id),
 363					     pci_iov_virtfn_devfn(dev, id));
 364	if (!virtfn)
 365		return;
 366
 367	sprintf(buf, "virtfn%u", id);
 368	sysfs_remove_link(&dev->dev.kobj, buf);
 369	/*
 370	 * pci_stop_dev() could have been called for this virtfn already,
 371	 * so the directory for the virtfn may have been removed before.
 372	 * Double check to avoid spurious sysfs warnings.
 373	 */
 374	if (virtfn->dev.kobj.sd)
 375		sysfs_remove_link(&virtfn->dev.kobj, "physfn");
 376
 377	pci_stop_and_remove_bus_device(virtfn);
 378	virtfn_remove_bus(dev->bus, virtfn->bus);
 379
 380	/* balance pci_get_domain_bus_and_slot() */
 381	pci_dev_put(virtfn);
 382	pci_dev_put(dev);
 383}
 384
 385static ssize_t sriov_totalvfs_show(struct device *dev,
 386				   struct device_attribute *attr,
 387				   char *buf)
 388{
 389	struct pci_dev *pdev = to_pci_dev(dev);
 390
 391	return sysfs_emit(buf, "%u\n", pci_sriov_get_totalvfs(pdev));
 392}
 393
 394static ssize_t sriov_numvfs_show(struct device *dev,
 395				 struct device_attribute *attr,
 396				 char *buf)
 397{
 398	struct pci_dev *pdev = to_pci_dev(dev);
 399	u16 num_vfs;
 400
 401	/* Serialize vs sriov_numvfs_store() so readers see valid num_VFs */
 402	device_lock(&pdev->dev);
 403	num_vfs = pdev->sriov->num_VFs;
 404	device_unlock(&pdev->dev);
 405
 406	return sysfs_emit(buf, "%u\n", num_vfs);
 407}
 408
 409/*
 410 * num_vfs > 0; number of VFs to enable
 411 * num_vfs = 0; disable all VFs
 412 *
 413 * Note: SRIOV spec does not allow partial VF
 414 *	 disable, so it's all or none.
 415 */
 416static ssize_t sriov_numvfs_store(struct device *dev,
 417				  struct device_attribute *attr,
 418				  const char *buf, size_t count)
 419{
 420	struct pci_dev *pdev = to_pci_dev(dev);
 421	int ret = 0;
 422	u16 num_vfs;
 423
 424	if (kstrtou16(buf, 0, &num_vfs) < 0)
 425		return -EINVAL;
 
 426
 427	if (num_vfs > pci_sriov_get_totalvfs(pdev))
 428		return -ERANGE;
 429
 430	device_lock(&pdev->dev);
 431
 432	if (num_vfs == pdev->sriov->num_VFs)
 433		goto exit;
 434
 435	/* is PF driver loaded */
 436	if (!pdev->driver) {
 437		pci_info(pdev, "no driver bound to device; cannot configure SR-IOV\n");
 438		ret = -ENOENT;
 439		goto exit;
 440	}
 441
 442	/* is PF driver loaded w/callback */
 443	if (!pdev->driver->sriov_configure) {
 444		pci_info(pdev, "driver does not support SR-IOV configuration via sysfs\n");
 445		ret = -ENOENT;
 446		goto exit;
 447	}
 448
 449	if (num_vfs == 0) {
 450		/* disable VFs */
 451		ret = pdev->driver->sriov_configure(pdev, 0);
 452		goto exit;
 453	}
 454
 455	/* enable VFs */
 456	if (pdev->sriov->num_VFs) {
 457		pci_warn(pdev, "%d VFs already enabled. Disable before enabling %d VFs\n",
 458			 pdev->sriov->num_VFs, num_vfs);
 459		ret = -EBUSY;
 460		goto exit;
 461	}
 462
 463	ret = pdev->driver->sriov_configure(pdev, num_vfs);
 464	if (ret < 0)
 465		goto exit;
 466
 467	if (ret != num_vfs)
 468		pci_warn(pdev, "%d VFs requested; only %d enabled\n",
 469			 num_vfs, ret);
 470
 471exit:
 472	device_unlock(&pdev->dev);
 473
 474	if (ret < 0)
 475		return ret;
 476
 477	return count;
 478}
 479
 480static ssize_t sriov_offset_show(struct device *dev,
 481				 struct device_attribute *attr,
 482				 char *buf)
 483{
 484	struct pci_dev *pdev = to_pci_dev(dev);
 485
 486	return sysfs_emit(buf, "%u\n", pdev->sriov->offset);
 487}
 488
 489static ssize_t sriov_stride_show(struct device *dev,
 490				 struct device_attribute *attr,
 491				 char *buf)
 492{
 493	struct pci_dev *pdev = to_pci_dev(dev);
 494
 495	return sysfs_emit(buf, "%u\n", pdev->sriov->stride);
 496}
 497
 498static ssize_t sriov_vf_device_show(struct device *dev,
 499				    struct device_attribute *attr,
 500				    char *buf)
 501{
 502	struct pci_dev *pdev = to_pci_dev(dev);
 503
 504	return sysfs_emit(buf, "%x\n", pdev->sriov->vf_device);
 505}
 506
 507static ssize_t sriov_drivers_autoprobe_show(struct device *dev,
 508					    struct device_attribute *attr,
 509					    char *buf)
 510{
 511	struct pci_dev *pdev = to_pci_dev(dev);
 512
 513	return sysfs_emit(buf, "%u\n", pdev->sriov->drivers_autoprobe);
 514}
 515
 516static ssize_t sriov_drivers_autoprobe_store(struct device *dev,
 517					     struct device_attribute *attr,
 518					     const char *buf, size_t count)
 519{
 520	struct pci_dev *pdev = to_pci_dev(dev);
 521	bool drivers_autoprobe;
 522
 523	if (kstrtobool(buf, &drivers_autoprobe) < 0)
 524		return -EINVAL;
 525
 526	pdev->sriov->drivers_autoprobe = drivers_autoprobe;
 527
 528	return count;
 529}
 530
 531static DEVICE_ATTR_RO(sriov_totalvfs);
 532static DEVICE_ATTR_RW(sriov_numvfs);
 533static DEVICE_ATTR_RO(sriov_offset);
 534static DEVICE_ATTR_RO(sriov_stride);
 535static DEVICE_ATTR_RO(sriov_vf_device);
 536static DEVICE_ATTR_RW(sriov_drivers_autoprobe);
 537
 538static struct attribute *sriov_pf_dev_attrs[] = {
 539	&dev_attr_sriov_totalvfs.attr,
 540	&dev_attr_sriov_numvfs.attr,
 541	&dev_attr_sriov_offset.attr,
 542	&dev_attr_sriov_stride.attr,
 543	&dev_attr_sriov_vf_device.attr,
 544	&dev_attr_sriov_drivers_autoprobe.attr,
 545#ifdef CONFIG_PCI_MSI
 546	&dev_attr_sriov_vf_total_msix.attr,
 547#endif
 548	NULL,
 549};
 550
 551static umode_t sriov_pf_attrs_are_visible(struct kobject *kobj,
 552					  struct attribute *a, int n)
 553{
 554	struct device *dev = kobj_to_dev(kobj);
 555
 556	if (!dev_is_pf(dev))
 557		return 0;
 558
 559	return a->mode;
 560}
 561
 562const struct attribute_group sriov_pf_dev_attr_group = {
 563	.attrs = sriov_pf_dev_attrs,
 564	.is_visible = sriov_pf_attrs_are_visible,
 565};
 566
 567int __weak pcibios_sriov_enable(struct pci_dev *pdev, u16 num_vfs)
 568{
 569	return 0;
 570}
 571
 572int __weak pcibios_sriov_disable(struct pci_dev *pdev)
 573{
 574	return 0;
 575}
 576
 577static int sriov_add_vfs(struct pci_dev *dev, u16 num_vfs)
 578{
 579	unsigned int i;
 580	int rc;
 581
 582	if (dev->no_vf_scan)
 583		return 0;
 584
 585	for (i = 0; i < num_vfs; i++) {
 586		rc = pci_iov_add_virtfn(dev, i);
 587		if (rc)
 588			goto failed;
 589	}
 590	return 0;
 591failed:
 592	while (i--)
 593		pci_iov_remove_virtfn(dev, i);
 594
 595	return rc;
 596}
 597
 598static int sriov_enable(struct pci_dev *dev, int nr_virtfn)
 599{
 600	int rc;
 601	int i;
 602	int nres;
 603	u16 initial;
 604	struct resource *res;
 605	struct pci_dev *pdev;
 606	struct pci_sriov *iov = dev->sriov;
 607	int bars = 0;
 608	int bus;
 609
 610	if (!nr_virtfn)
 611		return 0;
 612
 613	if (iov->num_VFs)
 614		return -EINVAL;
 615
 616	pci_read_config_word(dev, iov->pos + PCI_SRIOV_INITIAL_VF, &initial);
 617	if (initial > iov->total_VFs ||
 618	    (!(iov->cap & PCI_SRIOV_CAP_VFM) && (initial != iov->total_VFs)))
 619		return -EIO;
 620
 621	if (nr_virtfn < 0 || nr_virtfn > iov->total_VFs ||
 622	    (!(iov->cap & PCI_SRIOV_CAP_VFM) && (nr_virtfn > initial)))
 623		return -EINVAL;
 624
 625	nres = 0;
 626	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
 627		bars |= (1 << (i + PCI_IOV_RESOURCES));
 628		res = &dev->resource[i + PCI_IOV_RESOURCES];
 629		if (res->parent)
 630			nres++;
 631	}
 632	if (nres != iov->nres) {
 633		pci_err(dev, "not enough MMIO resources for SR-IOV\n");
 634		return -ENOMEM;
 635	}
 636
 637	bus = pci_iov_virtfn_bus(dev, nr_virtfn - 1);
 638	if (bus > dev->bus->busn_res.end) {
 639		pci_err(dev, "can't enable %d VFs (bus %02x out of range of %pR)\n",
 640			nr_virtfn, bus, &dev->bus->busn_res);
 641		return -ENOMEM;
 642	}
 643
 644	if (pci_enable_resources(dev, bars)) {
 645		pci_err(dev, "SR-IOV: IOV BARS not allocated\n");
 646		return -ENOMEM;
 647	}
 648
 649	if (iov->link != dev->devfn) {
 650		pdev = pci_get_slot(dev->bus, iov->link);
 651		if (!pdev)
 652			return -ENODEV;
 653
 654		if (!pdev->is_physfn) {
 655			pci_dev_put(pdev);
 656			return -ENOSYS;
 657		}
 658
 659		rc = sysfs_create_link(&dev->dev.kobj,
 660					&pdev->dev.kobj, "dep_link");
 661		pci_dev_put(pdev);
 662		if (rc)
 663			return rc;
 664	}
 665
 666	iov->initial_VFs = initial;
 667	if (nr_virtfn < initial)
 668		initial = nr_virtfn;
 669
 670	rc = pcibios_sriov_enable(dev, initial);
 671	if (rc) {
 672		pci_err(dev, "failure %d from pcibios_sriov_enable()\n", rc);
 673		goto err_pcibios;
 674	}
 675
 676	pci_iov_set_numvfs(dev, nr_virtfn);
 677	iov->ctrl |= PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE;
 678	pci_cfg_access_lock(dev);
 679	pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
 680	msleep(100);
 681	pci_cfg_access_unlock(dev);
 682
 683	rc = sriov_add_vfs(dev, initial);
 684	if (rc)
 685		goto err_pcibios;
 686
 687	kobject_uevent(&dev->dev.kobj, KOBJ_CHANGE);
 688	iov->num_VFs = nr_virtfn;
 689
 690	return 0;
 691
 692err_pcibios:
 693	iov->ctrl &= ~(PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE);
 694	pci_cfg_access_lock(dev);
 695	pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
 696	ssleep(1);
 697	pci_cfg_access_unlock(dev);
 698
 699	pcibios_sriov_disable(dev);
 700
 701	if (iov->link != dev->devfn)
 702		sysfs_remove_link(&dev->dev.kobj, "dep_link");
 703
 704	pci_iov_set_numvfs(dev, 0);
 705	return rc;
 706}
 707
 708static void sriov_del_vfs(struct pci_dev *dev)
 709{
 710	struct pci_sriov *iov = dev->sriov;
 711	int i;
 
 
 
 712
 713	for (i = 0; i < iov->num_VFs; i++)
 714		pci_iov_remove_virtfn(dev, i);
 715}
 716
 717static void sriov_disable(struct pci_dev *dev)
 718{
 719	struct pci_sriov *iov = dev->sriov;
 720
 721	if (!iov->num_VFs)
 722		return;
 723
 724	sriov_del_vfs(dev);
 725	iov->ctrl &= ~(PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE);
 726	pci_cfg_access_lock(dev);
 727	pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
 728	ssleep(1);
 729	pci_cfg_access_unlock(dev);
 730
 731	pcibios_sriov_disable(dev);
 732
 733	if (iov->link != dev->devfn)
 734		sysfs_remove_link(&dev->dev.kobj, "dep_link");
 735
 736	iov->num_VFs = 0;
 737	pci_iov_set_numvfs(dev, 0);
 738}
 739
 740static int sriov_init(struct pci_dev *dev, int pos)
 741{
 742	int i, bar64;
 743	int rc;
 744	int nres;
 745	u32 pgsz;
 746	u16 ctrl, total;
 747	struct pci_sriov *iov;
 748	struct resource *res;
 749	struct pci_dev *pdev;
 750
 751	pci_read_config_word(dev, pos + PCI_SRIOV_CTRL, &ctrl);
 752	if (ctrl & PCI_SRIOV_CTRL_VFE) {
 753		pci_write_config_word(dev, pos + PCI_SRIOV_CTRL, 0);
 754		ssleep(1);
 755	}
 756
 757	ctrl = 0;
 758	list_for_each_entry(pdev, &dev->bus->devices, bus_list)
 759		if (pdev->is_physfn)
 760			goto found;
 761
 762	pdev = NULL;
 763	if (pci_ari_enabled(dev->bus))
 764		ctrl |= PCI_SRIOV_CTRL_ARI;
 765
 766found:
 767	pci_write_config_word(dev, pos + PCI_SRIOV_CTRL, ctrl);
 768
 769	pci_read_config_word(dev, pos + PCI_SRIOV_TOTAL_VF, &total);
 770	if (!total)
 771		return 0;
 772
 773	pci_read_config_dword(dev, pos + PCI_SRIOV_SUP_PGSIZE, &pgsz);
 774	i = PAGE_SHIFT > 12 ? PAGE_SHIFT - 12 : 0;
 775	pgsz &= ~((1 << i) - 1);
 776	if (!pgsz)
 777		return -EIO;
 778
 779	pgsz &= ~(pgsz - 1);
 780	pci_write_config_dword(dev, pos + PCI_SRIOV_SYS_PGSIZE, pgsz);
 781
 782	iov = kzalloc(sizeof(*iov), GFP_KERNEL);
 783	if (!iov)
 784		return -ENOMEM;
 785
 786	nres = 0;
 787	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
 788		res = &dev->resource[i + PCI_IOV_RESOURCES];
 789		/*
 790		 * If it is already FIXED, don't change it, something
 791		 * (perhaps EA or header fixups) wants it this way.
 792		 */
 793		if (res->flags & IORESOURCE_PCI_FIXED)
 794			bar64 = (res->flags & IORESOURCE_MEM_64) ? 1 : 0;
 795		else
 796			bar64 = __pci_read_base(dev, pci_bar_unknown, res,
 797						pos + PCI_SRIOV_BAR + i * 4);
 798		if (!res->flags)
 799			continue;
 800		if (resource_size(res) & (PAGE_SIZE - 1)) {
 801			rc = -EIO;
 802			goto failed;
 803		}
 804		iov->barsz[i] = resource_size(res);
 805		res->end = res->start + resource_size(res) * total - 1;
 806		pci_info(dev, "VF(n) BAR%d space: %pR (contains BAR%d for %d VFs)\n",
 807			 i, res, i, total);
 808		i += bar64;
 809		nres++;
 810	}
 811
 812	iov->pos = pos;
 813	iov->nres = nres;
 814	iov->ctrl = ctrl;
 815	iov->total_VFs = total;
 816	iov->driver_max_VFs = total;
 817	pci_read_config_word(dev, pos + PCI_SRIOV_VF_DID, &iov->vf_device);
 818	iov->pgsz = pgsz;
 819	iov->self = dev;
 820	iov->drivers_autoprobe = true;
 821	pci_read_config_dword(dev, pos + PCI_SRIOV_CAP, &iov->cap);
 822	pci_read_config_byte(dev, pos + PCI_SRIOV_FUNC_LINK, &iov->link);
 823	if (pci_pcie_type(dev) == PCI_EXP_TYPE_RC_END)
 824		iov->link = PCI_DEVFN(PCI_SLOT(dev->devfn), iov->link);
 825
 826	if (pdev)
 827		iov->dev = pci_dev_get(pdev);
 828	else
 829		iov->dev = dev;
 830
 831	dev->sriov = iov;
 832	dev->is_physfn = 1;
 833	rc = compute_max_vf_buses(dev);
 834	if (rc)
 835		goto fail_max_buses;
 836
 837	return 0;
 838
 839fail_max_buses:
 840	dev->sriov = NULL;
 841	dev->is_physfn = 0;
 842failed:
 843	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
 844		res = &dev->resource[i + PCI_IOV_RESOURCES];
 845		res->flags = 0;
 846	}
 847
 848	kfree(iov);
 849	return rc;
 850}
 851
 852static void sriov_release(struct pci_dev *dev)
 853{
 854	BUG_ON(dev->sriov->num_VFs);
 855
 856	if (dev != dev->sriov->dev)
 857		pci_dev_put(dev->sriov->dev);
 858
 859	kfree(dev->sriov);
 860	dev->sriov = NULL;
 861}
 862
 863static void sriov_restore_state(struct pci_dev *dev)
 864{
 865	int i;
 866	u16 ctrl;
 867	struct pci_sriov *iov = dev->sriov;
 868
 869	pci_read_config_word(dev, iov->pos + PCI_SRIOV_CTRL, &ctrl);
 870	if (ctrl & PCI_SRIOV_CTRL_VFE)
 871		return;
 872
 873	/*
 874	 * Restore PCI_SRIOV_CTRL_ARI before pci_iov_set_numvfs() because
 875	 * it reads offset & stride, which depend on PCI_SRIOV_CTRL_ARI.
 876	 */
 877	ctrl &= ~PCI_SRIOV_CTRL_ARI;
 878	ctrl |= iov->ctrl & PCI_SRIOV_CTRL_ARI;
 879	pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, ctrl);
 880
 881	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++)
 882		pci_update_resource(dev, i + PCI_IOV_RESOURCES);
 883
 884	pci_write_config_dword(dev, iov->pos + PCI_SRIOV_SYS_PGSIZE, iov->pgsz);
 885	pci_iov_set_numvfs(dev, iov->num_VFs);
 886	pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
 887	if (iov->ctrl & PCI_SRIOV_CTRL_VFE)
 888		msleep(100);
 889}
 890
 891/**
 892 * pci_iov_init - initialize the IOV capability
 893 * @dev: the PCI device
 894 *
 895 * Returns 0 on success, or negative on failure.
 896 */
 897int pci_iov_init(struct pci_dev *dev)
 898{
 899	int pos;
 900
 901	if (!pci_is_pcie(dev))
 902		return -ENODEV;
 903
 904	pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_SRIOV);
 905	if (pos)
 906		return sriov_init(dev, pos);
 907
 908	return -ENODEV;
 909}
 910
 911/**
 912 * pci_iov_release - release resources used by the IOV capability
 913 * @dev: the PCI device
 914 */
 915void pci_iov_release(struct pci_dev *dev)
 916{
 917	if (dev->is_physfn)
 918		sriov_release(dev);
 919}
 920
 921/**
 922 * pci_iov_remove - clean up SR-IOV state after PF driver is detached
 923 * @dev: the PCI device
 924 */
 925void pci_iov_remove(struct pci_dev *dev)
 926{
 927	struct pci_sriov *iov = dev->sriov;
 928
 929	if (!dev->is_physfn)
 930		return;
 931
 932	iov->driver_max_VFs = iov->total_VFs;
 933	if (iov->num_VFs)
 934		pci_warn(dev, "driver left SR-IOV enabled after remove\n");
 935}
 936
 937/**
 938 * pci_iov_update_resource - update a VF BAR
 939 * @dev: the PCI device
 940 * @resno: the resource number
 941 *
 942 * Update a VF BAR in the SR-IOV capability of a PF.
 943 */
 944void pci_iov_update_resource(struct pci_dev *dev, int resno)
 945{
 946	struct pci_sriov *iov = dev->is_physfn ? dev->sriov : NULL;
 947	struct resource *res = dev->resource + resno;
 948	int vf_bar = resno - PCI_IOV_RESOURCES;
 949	struct pci_bus_region region;
 950	u16 cmd;
 951	u32 new;
 952	int reg;
 953
 954	/*
 955	 * The generic pci_restore_bars() path calls this for all devices,
 956	 * including VFs and non-SR-IOV devices.  If this is not a PF, we
 957	 * have nothing to do.
 958	 */
 959	if (!iov)
 960		return;
 961
 962	pci_read_config_word(dev, iov->pos + PCI_SRIOV_CTRL, &cmd);
 963	if ((cmd & PCI_SRIOV_CTRL_VFE) && (cmd & PCI_SRIOV_CTRL_MSE)) {
 964		dev_WARN(&dev->dev, "can't update enabled VF BAR%d %pR\n",
 965			 vf_bar, res);
 966		return;
 967	}
 968
 969	/*
 970	 * Ignore unimplemented BARs, unused resource slots for 64-bit
 971	 * BARs, and non-movable resources, e.g., those described via
 972	 * Enhanced Allocation.
 973	 */
 974	if (!res->flags)
 975		return;
 976
 977	if (res->flags & IORESOURCE_UNSET)
 978		return;
 979
 980	if (res->flags & IORESOURCE_PCI_FIXED)
 981		return;
 982
 983	pcibios_resource_to_bus(dev->bus, &region, res);
 984	new = region.start;
 985	new |= res->flags & ~PCI_BASE_ADDRESS_MEM_MASK;
 986
 987	reg = iov->pos + PCI_SRIOV_BAR + 4 * vf_bar;
 988	pci_write_config_dword(dev, reg, new);
 989	if (res->flags & IORESOURCE_MEM_64) {
 990		new = region.start >> 16 >> 16;
 991		pci_write_config_dword(dev, reg + 4, new);
 992	}
 993}
 994
 995resource_size_t __weak pcibios_iov_resource_alignment(struct pci_dev *dev,
 996						      int resno)
 997{
 998	return pci_iov_resource_size(dev, resno);
 999}
1000
1001/**
1002 * pci_sriov_resource_alignment - get resource alignment for VF BAR
1003 * @dev: the PCI device
1004 * @resno: the resource number
1005 *
1006 * Returns the alignment of the VF BAR found in the SR-IOV capability.
1007 * This is not the same as the resource size which is defined as
1008 * the VF BAR size multiplied by the number of VFs.  The alignment
1009 * is just the VF BAR size.
1010 */
1011resource_size_t pci_sriov_resource_alignment(struct pci_dev *dev, int resno)
1012{
1013	return pcibios_iov_resource_alignment(dev, resno);
1014}
1015
1016/**
1017 * pci_restore_iov_state - restore the state of the IOV capability
1018 * @dev: the PCI device
1019 */
1020void pci_restore_iov_state(struct pci_dev *dev)
1021{
1022	if (dev->is_physfn)
1023		sriov_restore_state(dev);
1024}
1025
1026/**
1027 * pci_vf_drivers_autoprobe - set PF property drivers_autoprobe for VFs
1028 * @dev: the PCI device
1029 * @auto_probe: set VF drivers auto probe flag
1030 */
1031void pci_vf_drivers_autoprobe(struct pci_dev *dev, bool auto_probe)
1032{
1033	if (dev->is_physfn)
1034		dev->sriov->drivers_autoprobe = auto_probe;
1035}
1036
1037/**
1038 * pci_iov_bus_range - find bus range used by Virtual Function
1039 * @bus: the PCI bus
1040 *
1041 * Returns max number of buses (exclude current one) used by Virtual
1042 * Functions.
1043 */
1044int pci_iov_bus_range(struct pci_bus *bus)
1045{
1046	int max = 0;
1047	struct pci_dev *dev;
1048
1049	list_for_each_entry(dev, &bus->devices, bus_list) {
1050		if (!dev->is_physfn)
1051			continue;
1052		if (dev->sriov->max_VF_buses > max)
1053			max = dev->sriov->max_VF_buses;
1054	}
1055
1056	return max ? max - bus->number : 0;
1057}
1058
1059/**
1060 * pci_enable_sriov - enable the SR-IOV capability
1061 * @dev: the PCI device
1062 * @nr_virtfn: number of virtual functions to enable
1063 *
1064 * Returns 0 on success, or negative on failure.
1065 */
1066int pci_enable_sriov(struct pci_dev *dev, int nr_virtfn)
1067{
1068	might_sleep();
1069
1070	if (!dev->is_physfn)
1071		return -ENOSYS;
1072
1073	return sriov_enable(dev, nr_virtfn);
1074}
1075EXPORT_SYMBOL_GPL(pci_enable_sriov);
1076
1077/**
1078 * pci_disable_sriov - disable the SR-IOV capability
1079 * @dev: the PCI device
1080 */
1081void pci_disable_sriov(struct pci_dev *dev)
1082{
1083	might_sleep();
1084
1085	if (!dev->is_physfn)
1086		return;
1087
1088	sriov_disable(dev);
1089}
1090EXPORT_SYMBOL_GPL(pci_disable_sriov);
1091
1092/**
1093 * pci_num_vf - return number of VFs associated with a PF device_release_driver
1094 * @dev: the PCI device
1095 *
1096 * Returns number of VFs, or 0 if SR-IOV is not enabled.
1097 */
1098int pci_num_vf(struct pci_dev *dev)
1099{
1100	if (!dev->is_physfn)
1101		return 0;
1102
1103	return dev->sriov->num_VFs;
1104}
1105EXPORT_SYMBOL_GPL(pci_num_vf);
1106
1107/**
1108 * pci_vfs_assigned - returns number of VFs are assigned to a guest
1109 * @dev: the PCI device
1110 *
1111 * Returns number of VFs belonging to this device that are assigned to a guest.
1112 * If device is not a physical function returns 0.
1113 */
1114int pci_vfs_assigned(struct pci_dev *dev)
1115{
1116	struct pci_dev *vfdev;
1117	unsigned int vfs_assigned = 0;
1118	unsigned short dev_id;
1119
1120	/* only search if we are a PF */
1121	if (!dev->is_physfn)
1122		return 0;
1123
1124	/*
1125	 * determine the device ID for the VFs, the vendor ID will be the
1126	 * same as the PF so there is no need to check for that one
1127	 */
1128	dev_id = dev->sriov->vf_device;
1129
1130	/* loop through all the VFs to see if we own any that are assigned */
1131	vfdev = pci_get_device(dev->vendor, dev_id, NULL);
1132	while (vfdev) {
1133		/*
1134		 * It is considered assigned if it is a virtual function with
1135		 * our dev as the physical function and the assigned bit is set
1136		 */
1137		if (vfdev->is_virtfn && (vfdev->physfn == dev) &&
1138			pci_is_dev_assigned(vfdev))
1139			vfs_assigned++;
1140
1141		vfdev = pci_get_device(dev->vendor, dev_id, vfdev);
1142	}
1143
1144	return vfs_assigned;
1145}
1146EXPORT_SYMBOL_GPL(pci_vfs_assigned);
1147
1148/**
1149 * pci_sriov_set_totalvfs -- reduce the TotalVFs available
1150 * @dev: the PCI PF device
1151 * @numvfs: number that should be used for TotalVFs supported
1152 *
1153 * Should be called from PF driver's probe routine with
1154 * device's mutex held.
1155 *
1156 * Returns 0 if PF is an SRIOV-capable device and
1157 * value of numvfs valid. If not a PF return -ENOSYS;
1158 * if numvfs is invalid return -EINVAL;
1159 * if VFs already enabled, return -EBUSY.
1160 */
1161int pci_sriov_set_totalvfs(struct pci_dev *dev, u16 numvfs)
1162{
1163	if (!dev->is_physfn)
1164		return -ENOSYS;
1165
1166	if (numvfs > dev->sriov->total_VFs)
1167		return -EINVAL;
1168
1169	/* Shouldn't change if VFs already enabled */
1170	if (dev->sriov->ctrl & PCI_SRIOV_CTRL_VFE)
1171		return -EBUSY;
1172
1173	dev->sriov->driver_max_VFs = numvfs;
1174	return 0;
1175}
1176EXPORT_SYMBOL_GPL(pci_sriov_set_totalvfs);
1177
1178/**
1179 * pci_sriov_get_totalvfs -- get total VFs supported on this device
1180 * @dev: the PCI PF device
1181 *
1182 * For a PCIe device with SRIOV support, return the PCIe
1183 * SRIOV capability value of TotalVFs or the value of driver_max_VFs
1184 * if the driver reduced it.  Otherwise 0.
1185 */
1186int pci_sriov_get_totalvfs(struct pci_dev *dev)
1187{
1188	if (!dev->is_physfn)
1189		return 0;
1190
1191	return dev->sriov->driver_max_VFs;
1192}
1193EXPORT_SYMBOL_GPL(pci_sriov_get_totalvfs);
1194
1195/**
1196 * pci_sriov_configure_simple - helper to configure SR-IOV
1197 * @dev: the PCI device
1198 * @nr_virtfn: number of virtual functions to enable, 0 to disable
1199 *
1200 * Enable or disable SR-IOV for devices that don't require any PF setup
1201 * before enabling SR-IOV.  Return value is negative on error, or number of
1202 * VFs allocated on success.
1203 */
1204int pci_sriov_configure_simple(struct pci_dev *dev, int nr_virtfn)
1205{
1206	int rc;
1207
1208	might_sleep();
1209
1210	if (!dev->is_physfn)
1211		return -ENODEV;
1212
1213	if (pci_vfs_assigned(dev)) {
1214		pci_warn(dev, "Cannot modify SR-IOV while VFs are assigned\n");
1215		return -EPERM;
1216	}
1217
1218	if (nr_virtfn == 0) {
1219		sriov_disable(dev);
1220		return 0;
1221	}
1222
1223	rc = sriov_enable(dev, nr_virtfn);
1224	if (rc < 0)
1225		return rc;
1226
1227	return nr_virtfn;
1228}
1229EXPORT_SYMBOL_GPL(pci_sriov_configure_simple);