xref: /linux/drivers/pci/iov.c (revision 6beeaf48db6c548fcfc2ad32739d33af2fef3a5b)
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 EXPORT_SYMBOL_GPL(pci_iov_virtfn_devfn);
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  */
42 static 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  */
58 static 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 
75 out:
76 	pci_iov_set_numvfs(dev, 0);
77 	return rc;
78 }
79 
80 static 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 
100 static 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 
106 resource_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 
114 static 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 
137 int pci_iov_sysfs_link(struct pci_dev *dev,
138 		struct pci_dev *virtfn, int id)
139 {
140 	char buf[VIRTFN_ID_LEN];
141 	int rc;
142 
143 	sprintf(buf, "virtfn%u", id);
144 	rc = sysfs_create_link(&dev->dev.kobj, &virtfn->dev.kobj, buf);
145 	if (rc)
146 		goto failed;
147 	rc = sysfs_create_link(&virtfn->dev.kobj, &dev->dev.kobj, "physfn");
148 	if (rc)
149 		goto failed1;
150 
151 	kobject_uevent(&virtfn->dev.kobj, KOBJ_CHANGE);
152 
153 	return 0;
154 
155 failed1:
156 	sysfs_remove_link(&dev->dev.kobj, buf);
157 failed:
158 	return rc;
159 }
160 
161 #ifdef CONFIG_PCI_MSI
162 static ssize_t sriov_vf_total_msix_show(struct device *dev,
163 					struct device_attribute *attr,
164 					char *buf)
165 {
166 	struct pci_dev *pdev = to_pci_dev(dev);
167 	u32 vf_total_msix = 0;
168 
169 	device_lock(dev);
170 	if (!pdev->driver || !pdev->driver->sriov_get_vf_total_msix)
171 		goto unlock;
172 
173 	vf_total_msix = pdev->driver->sriov_get_vf_total_msix(pdev);
174 unlock:
175 	device_unlock(dev);
176 	return sysfs_emit(buf, "%u\n", vf_total_msix);
177 }
178 static DEVICE_ATTR_RO(sriov_vf_total_msix);
179 
180 static ssize_t sriov_vf_msix_count_store(struct device *dev,
181 					 struct device_attribute *attr,
182 					 const char *buf, size_t count)
183 {
184 	struct pci_dev *vf_dev = to_pci_dev(dev);
185 	struct pci_dev *pdev = pci_physfn(vf_dev);
186 	int val, ret;
187 
188 	ret = kstrtoint(buf, 0, &val);
189 	if (ret)
190 		return ret;
191 
192 	if (val < 0)
193 		return -EINVAL;
194 
195 	device_lock(&pdev->dev);
196 	if (!pdev->driver || !pdev->driver->sriov_set_msix_vec_count) {
197 		ret = -EOPNOTSUPP;
198 		goto err_pdev;
199 	}
200 
201 	device_lock(&vf_dev->dev);
202 	if (vf_dev->driver) {
203 		/*
204 		 * A driver is already attached to this VF and has configured
205 		 * itself based on the current MSI-X vector count. Changing
206 		 * the vector size could mess up the driver, so block it.
207 		 */
208 		ret = -EBUSY;
209 		goto err_dev;
210 	}
211 
212 	ret = pdev->driver->sriov_set_msix_vec_count(vf_dev, val);
213 
214 err_dev:
215 	device_unlock(&vf_dev->dev);
216 err_pdev:
217 	device_unlock(&pdev->dev);
218 	return ret ? : count;
219 }
220 static DEVICE_ATTR_WO(sriov_vf_msix_count);
221 #endif
222 
223 static struct attribute *sriov_vf_dev_attrs[] = {
224 #ifdef CONFIG_PCI_MSI
225 	&dev_attr_sriov_vf_msix_count.attr,
226 #endif
227 	NULL,
228 };
229 
230 static umode_t sriov_vf_attrs_are_visible(struct kobject *kobj,
231 					  struct attribute *a, int n)
232 {
233 	struct device *dev = kobj_to_dev(kobj);
234 	struct pci_dev *pdev = to_pci_dev(dev);
235 
236 	if (!pdev->is_virtfn)
237 		return 0;
238 
239 	return a->mode;
240 }
241 
242 const struct attribute_group sriov_vf_dev_attr_group = {
243 	.attrs = sriov_vf_dev_attrs,
244 	.is_visible = sriov_vf_attrs_are_visible,
245 };
246 
247 int pci_iov_add_virtfn(struct pci_dev *dev, int id)
248 {
249 	int i;
250 	int rc = -ENOMEM;
251 	u64 size;
252 	struct pci_dev *virtfn;
253 	struct resource *res;
254 	struct pci_sriov *iov = dev->sriov;
255 	struct pci_bus *bus;
256 
257 	bus = virtfn_add_bus(dev->bus, pci_iov_virtfn_bus(dev, id));
258 	if (!bus)
259 		goto failed;
260 
261 	virtfn = pci_alloc_dev(bus);
262 	if (!virtfn)
263 		goto failed0;
264 
265 	virtfn->devfn = pci_iov_virtfn_devfn(dev, id);
266 	virtfn->vendor = dev->vendor;
267 	virtfn->device = iov->vf_device;
268 	virtfn->is_virtfn = 1;
269 	virtfn->physfn = pci_dev_get(dev);
270 	virtfn->no_command_memory = 1;
271 
272 	if (id == 0)
273 		pci_read_vf_config_common(virtfn);
274 
275 	rc = pci_setup_device(virtfn);
276 	if (rc)
277 		goto failed1;
278 
279 	virtfn->dev.parent = dev->dev.parent;
280 	virtfn->multifunction = 0;
281 
282 	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
283 		res = &dev->resource[i + PCI_IOV_RESOURCES];
284 		if (!res->parent)
285 			continue;
286 		virtfn->resource[i].name = pci_name(virtfn);
287 		virtfn->resource[i].flags = res->flags;
288 		size = pci_iov_resource_size(dev, i + PCI_IOV_RESOURCES);
289 		virtfn->resource[i].start = res->start + size * id;
290 		virtfn->resource[i].end = virtfn->resource[i].start + size - 1;
291 		rc = request_resource(res, &virtfn->resource[i]);
292 		BUG_ON(rc);
293 	}
294 
295 	pci_device_add(virtfn, virtfn->bus);
296 	rc = pci_iov_sysfs_link(dev, virtfn, id);
297 	if (rc)
298 		goto failed1;
299 
300 	pci_bus_add_device(virtfn);
301 
302 	return 0;
303 
304 failed1:
305 	pci_stop_and_remove_bus_device(virtfn);
306 	pci_dev_put(dev);
307 failed0:
308 	virtfn_remove_bus(dev->bus, bus);
309 failed:
310 
311 	return rc;
312 }
313 
314 void pci_iov_remove_virtfn(struct pci_dev *dev, int id)
315 {
316 	char buf[VIRTFN_ID_LEN];
317 	struct pci_dev *virtfn;
318 
319 	virtfn = pci_get_domain_bus_and_slot(pci_domain_nr(dev->bus),
320 					     pci_iov_virtfn_bus(dev, id),
321 					     pci_iov_virtfn_devfn(dev, id));
322 	if (!virtfn)
323 		return;
324 
325 	sprintf(buf, "virtfn%u", id);
326 	sysfs_remove_link(&dev->dev.kobj, buf);
327 	/*
328 	 * pci_stop_dev() could have been called for this virtfn already,
329 	 * so the directory for the virtfn may have been removed before.
330 	 * Double check to avoid spurious sysfs warnings.
331 	 */
332 	if (virtfn->dev.kobj.sd)
333 		sysfs_remove_link(&virtfn->dev.kobj, "physfn");
334 
335 	pci_stop_and_remove_bus_device(virtfn);
336 	virtfn_remove_bus(dev->bus, virtfn->bus);
337 
338 	/* balance pci_get_domain_bus_and_slot() */
339 	pci_dev_put(virtfn);
340 	pci_dev_put(dev);
341 }
342 
343 static ssize_t sriov_totalvfs_show(struct device *dev,
344 				   struct device_attribute *attr,
345 				   char *buf)
346 {
347 	struct pci_dev *pdev = to_pci_dev(dev);
348 
349 	return sysfs_emit(buf, "%u\n", pci_sriov_get_totalvfs(pdev));
350 }
351 
352 static ssize_t sriov_numvfs_show(struct device *dev,
353 				 struct device_attribute *attr,
354 				 char *buf)
355 {
356 	struct pci_dev *pdev = to_pci_dev(dev);
357 	u16 num_vfs;
358 
359 	/* Serialize vs sriov_numvfs_store() so readers see valid num_VFs */
360 	device_lock(&pdev->dev);
361 	num_vfs = pdev->sriov->num_VFs;
362 	device_unlock(&pdev->dev);
363 
364 	return sysfs_emit(buf, "%u\n", num_vfs);
365 }
366 
367 /*
368  * num_vfs > 0; number of VFs to enable
369  * num_vfs = 0; disable all VFs
370  *
371  * Note: SRIOV spec does not allow partial VF
372  *	 disable, so it's all or none.
373  */
374 static ssize_t sriov_numvfs_store(struct device *dev,
375 				  struct device_attribute *attr,
376 				  const char *buf, size_t count)
377 {
378 	struct pci_dev *pdev = to_pci_dev(dev);
379 	int ret;
380 	u16 num_vfs;
381 
382 	ret = kstrtou16(buf, 0, &num_vfs);
383 	if (ret < 0)
384 		return ret;
385 
386 	if (num_vfs > pci_sriov_get_totalvfs(pdev))
387 		return -ERANGE;
388 
389 	device_lock(&pdev->dev);
390 
391 	if (num_vfs == pdev->sriov->num_VFs)
392 		goto exit;
393 
394 	/* is PF driver loaded */
395 	if (!pdev->driver) {
396 		pci_info(pdev, "no driver bound to device; cannot configure SR-IOV\n");
397 		ret = -ENOENT;
398 		goto exit;
399 	}
400 
401 	/* is PF driver loaded w/callback */
402 	if (!pdev->driver->sriov_configure) {
403 		pci_info(pdev, "driver does not support SR-IOV configuration via sysfs\n");
404 		ret = -ENOENT;
405 		goto exit;
406 	}
407 
408 	if (num_vfs == 0) {
409 		/* disable VFs */
410 		ret = pdev->driver->sriov_configure(pdev, 0);
411 		goto exit;
412 	}
413 
414 	/* enable VFs */
415 	if (pdev->sriov->num_VFs) {
416 		pci_warn(pdev, "%d VFs already enabled. Disable before enabling %d VFs\n",
417 			 pdev->sriov->num_VFs, num_vfs);
418 		ret = -EBUSY;
419 		goto exit;
420 	}
421 
422 	ret = pdev->driver->sriov_configure(pdev, num_vfs);
423 	if (ret < 0)
424 		goto exit;
425 
426 	if (ret != num_vfs)
427 		pci_warn(pdev, "%d VFs requested; only %d enabled\n",
428 			 num_vfs, ret);
429 
430 exit:
431 	device_unlock(&pdev->dev);
432 
433 	if (ret < 0)
434 		return ret;
435 
436 	return count;
437 }
438 
439 static ssize_t sriov_offset_show(struct device *dev,
440 				 struct device_attribute *attr,
441 				 char *buf)
442 {
443 	struct pci_dev *pdev = to_pci_dev(dev);
444 
445 	return sysfs_emit(buf, "%u\n", pdev->sriov->offset);
446 }
447 
448 static ssize_t sriov_stride_show(struct device *dev,
449 				 struct device_attribute *attr,
450 				 char *buf)
451 {
452 	struct pci_dev *pdev = to_pci_dev(dev);
453 
454 	return sysfs_emit(buf, "%u\n", pdev->sriov->stride);
455 }
456 
457 static ssize_t sriov_vf_device_show(struct device *dev,
458 				    struct device_attribute *attr,
459 				    char *buf)
460 {
461 	struct pci_dev *pdev = to_pci_dev(dev);
462 
463 	return sysfs_emit(buf, "%x\n", pdev->sriov->vf_device);
464 }
465 
466 static ssize_t sriov_drivers_autoprobe_show(struct device *dev,
467 					    struct device_attribute *attr,
468 					    char *buf)
469 {
470 	struct pci_dev *pdev = to_pci_dev(dev);
471 
472 	return sysfs_emit(buf, "%u\n", pdev->sriov->drivers_autoprobe);
473 }
474 
475 static ssize_t sriov_drivers_autoprobe_store(struct device *dev,
476 					     struct device_attribute *attr,
477 					     const char *buf, size_t count)
478 {
479 	struct pci_dev *pdev = to_pci_dev(dev);
480 	bool drivers_autoprobe;
481 
482 	if (kstrtobool(buf, &drivers_autoprobe) < 0)
483 		return -EINVAL;
484 
485 	pdev->sriov->drivers_autoprobe = drivers_autoprobe;
486 
487 	return count;
488 }
489 
490 static DEVICE_ATTR_RO(sriov_totalvfs);
491 static DEVICE_ATTR_RW(sriov_numvfs);
492 static DEVICE_ATTR_RO(sriov_offset);
493 static DEVICE_ATTR_RO(sriov_stride);
494 static DEVICE_ATTR_RO(sriov_vf_device);
495 static DEVICE_ATTR_RW(sriov_drivers_autoprobe);
496 
497 static struct attribute *sriov_pf_dev_attrs[] = {
498 	&dev_attr_sriov_totalvfs.attr,
499 	&dev_attr_sriov_numvfs.attr,
500 	&dev_attr_sriov_offset.attr,
501 	&dev_attr_sriov_stride.attr,
502 	&dev_attr_sriov_vf_device.attr,
503 	&dev_attr_sriov_drivers_autoprobe.attr,
504 #ifdef CONFIG_PCI_MSI
505 	&dev_attr_sriov_vf_total_msix.attr,
506 #endif
507 	NULL,
508 };
509 
510 static umode_t sriov_pf_attrs_are_visible(struct kobject *kobj,
511 					  struct attribute *a, int n)
512 {
513 	struct device *dev = kobj_to_dev(kobj);
514 
515 	if (!dev_is_pf(dev))
516 		return 0;
517 
518 	return a->mode;
519 }
520 
521 const struct attribute_group sriov_pf_dev_attr_group = {
522 	.attrs = sriov_pf_dev_attrs,
523 	.is_visible = sriov_pf_attrs_are_visible,
524 };
525 
526 int __weak pcibios_sriov_enable(struct pci_dev *pdev, u16 num_vfs)
527 {
528 	return 0;
529 }
530 
531 int __weak pcibios_sriov_disable(struct pci_dev *pdev)
532 {
533 	return 0;
534 }
535 
536 static int sriov_add_vfs(struct pci_dev *dev, u16 num_vfs)
537 {
538 	unsigned int i;
539 	int rc;
540 
541 	if (dev->no_vf_scan)
542 		return 0;
543 
544 	for (i = 0; i < num_vfs; i++) {
545 		rc = pci_iov_add_virtfn(dev, i);
546 		if (rc)
547 			goto failed;
548 	}
549 	return 0;
550 failed:
551 	while (i--)
552 		pci_iov_remove_virtfn(dev, i);
553 
554 	return rc;
555 }
556 
557 static int sriov_enable(struct pci_dev *dev, int nr_virtfn)
558 {
559 	int rc;
560 	int i;
561 	int nres;
562 	u16 initial;
563 	struct resource *res;
564 	struct pci_dev *pdev;
565 	struct pci_sriov *iov = dev->sriov;
566 	int bars = 0;
567 	int bus;
568 
569 	if (!nr_virtfn)
570 		return 0;
571 
572 	if (iov->num_VFs)
573 		return -EINVAL;
574 
575 	pci_read_config_word(dev, iov->pos + PCI_SRIOV_INITIAL_VF, &initial);
576 	if (initial > iov->total_VFs ||
577 	    (!(iov->cap & PCI_SRIOV_CAP_VFM) && (initial != iov->total_VFs)))
578 		return -EIO;
579 
580 	if (nr_virtfn < 0 || nr_virtfn > iov->total_VFs ||
581 	    (!(iov->cap & PCI_SRIOV_CAP_VFM) && (nr_virtfn > initial)))
582 		return -EINVAL;
583 
584 	nres = 0;
585 	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
586 		bars |= (1 << (i + PCI_IOV_RESOURCES));
587 		res = &dev->resource[i + PCI_IOV_RESOURCES];
588 		if (res->parent)
589 			nres++;
590 	}
591 	if (nres != iov->nres) {
592 		pci_err(dev, "not enough MMIO resources for SR-IOV\n");
593 		return -ENOMEM;
594 	}
595 
596 	bus = pci_iov_virtfn_bus(dev, nr_virtfn - 1);
597 	if (bus > dev->bus->busn_res.end) {
598 		pci_err(dev, "can't enable %d VFs (bus %02x out of range of %pR)\n",
599 			nr_virtfn, bus, &dev->bus->busn_res);
600 		return -ENOMEM;
601 	}
602 
603 	if (pci_enable_resources(dev, bars)) {
604 		pci_err(dev, "SR-IOV: IOV BARS not allocated\n");
605 		return -ENOMEM;
606 	}
607 
608 	if (iov->link != dev->devfn) {
609 		pdev = pci_get_slot(dev->bus, iov->link);
610 		if (!pdev)
611 			return -ENODEV;
612 
613 		if (!pdev->is_physfn) {
614 			pci_dev_put(pdev);
615 			return -ENOSYS;
616 		}
617 
618 		rc = sysfs_create_link(&dev->dev.kobj,
619 					&pdev->dev.kobj, "dep_link");
620 		pci_dev_put(pdev);
621 		if (rc)
622 			return rc;
623 	}
624 
625 	iov->initial_VFs = initial;
626 	if (nr_virtfn < initial)
627 		initial = nr_virtfn;
628 
629 	rc = pcibios_sriov_enable(dev, initial);
630 	if (rc) {
631 		pci_err(dev, "failure %d from pcibios_sriov_enable()\n", rc);
632 		goto err_pcibios;
633 	}
634 
635 	pci_iov_set_numvfs(dev, nr_virtfn);
636 	iov->ctrl |= PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE;
637 	pci_cfg_access_lock(dev);
638 	pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
639 	msleep(100);
640 	pci_cfg_access_unlock(dev);
641 
642 	rc = sriov_add_vfs(dev, initial);
643 	if (rc)
644 		goto err_pcibios;
645 
646 	kobject_uevent(&dev->dev.kobj, KOBJ_CHANGE);
647 	iov->num_VFs = nr_virtfn;
648 
649 	return 0;
650 
651 err_pcibios:
652 	iov->ctrl &= ~(PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE);
653 	pci_cfg_access_lock(dev);
654 	pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
655 	ssleep(1);
656 	pci_cfg_access_unlock(dev);
657 
658 	pcibios_sriov_disable(dev);
659 
660 	if (iov->link != dev->devfn)
661 		sysfs_remove_link(&dev->dev.kobj, "dep_link");
662 
663 	pci_iov_set_numvfs(dev, 0);
664 	return rc;
665 }
666 
667 static void sriov_del_vfs(struct pci_dev *dev)
668 {
669 	struct pci_sriov *iov = dev->sriov;
670 	int i;
671 
672 	for (i = 0; i < iov->num_VFs; i++)
673 		pci_iov_remove_virtfn(dev, i);
674 }
675 
676 static void sriov_disable(struct pci_dev *dev)
677 {
678 	struct pci_sriov *iov = dev->sriov;
679 
680 	if (!iov->num_VFs)
681 		return;
682 
683 	sriov_del_vfs(dev);
684 	iov->ctrl &= ~(PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE);
685 	pci_cfg_access_lock(dev);
686 	pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
687 	ssleep(1);
688 	pci_cfg_access_unlock(dev);
689 
690 	pcibios_sriov_disable(dev);
691 
692 	if (iov->link != dev->devfn)
693 		sysfs_remove_link(&dev->dev.kobj, "dep_link");
694 
695 	iov->num_VFs = 0;
696 	pci_iov_set_numvfs(dev, 0);
697 }
698 
699 static int sriov_init(struct pci_dev *dev, int pos)
700 {
701 	int i, bar64;
702 	int rc;
703 	int nres;
704 	u32 pgsz;
705 	u16 ctrl, total;
706 	struct pci_sriov *iov;
707 	struct resource *res;
708 	struct pci_dev *pdev;
709 
710 	pci_read_config_word(dev, pos + PCI_SRIOV_CTRL, &ctrl);
711 	if (ctrl & PCI_SRIOV_CTRL_VFE) {
712 		pci_write_config_word(dev, pos + PCI_SRIOV_CTRL, 0);
713 		ssleep(1);
714 	}
715 
716 	ctrl = 0;
717 	list_for_each_entry(pdev, &dev->bus->devices, bus_list)
718 		if (pdev->is_physfn)
719 			goto found;
720 
721 	pdev = NULL;
722 	if (pci_ari_enabled(dev->bus))
723 		ctrl |= PCI_SRIOV_CTRL_ARI;
724 
725 found:
726 	pci_write_config_word(dev, pos + PCI_SRIOV_CTRL, ctrl);
727 
728 	pci_read_config_word(dev, pos + PCI_SRIOV_TOTAL_VF, &total);
729 	if (!total)
730 		return 0;
731 
732 	pci_read_config_dword(dev, pos + PCI_SRIOV_SUP_PGSIZE, &pgsz);
733 	i = PAGE_SHIFT > 12 ? PAGE_SHIFT - 12 : 0;
734 	pgsz &= ~((1 << i) - 1);
735 	if (!pgsz)
736 		return -EIO;
737 
738 	pgsz &= ~(pgsz - 1);
739 	pci_write_config_dword(dev, pos + PCI_SRIOV_SYS_PGSIZE, pgsz);
740 
741 	iov = kzalloc(sizeof(*iov), GFP_KERNEL);
742 	if (!iov)
743 		return -ENOMEM;
744 
745 	nres = 0;
746 	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
747 		res = &dev->resource[i + PCI_IOV_RESOURCES];
748 		/*
749 		 * If it is already FIXED, don't change it, something
750 		 * (perhaps EA or header fixups) wants it this way.
751 		 */
752 		if (res->flags & IORESOURCE_PCI_FIXED)
753 			bar64 = (res->flags & IORESOURCE_MEM_64) ? 1 : 0;
754 		else
755 			bar64 = __pci_read_base(dev, pci_bar_unknown, res,
756 						pos + PCI_SRIOV_BAR + i * 4);
757 		if (!res->flags)
758 			continue;
759 		if (resource_size(res) & (PAGE_SIZE - 1)) {
760 			rc = -EIO;
761 			goto failed;
762 		}
763 		iov->barsz[i] = resource_size(res);
764 		res->end = res->start + resource_size(res) * total - 1;
765 		pci_info(dev, "VF(n) BAR%d space: %pR (contains BAR%d for %d VFs)\n",
766 			 i, res, i, total);
767 		i += bar64;
768 		nres++;
769 	}
770 
771 	iov->pos = pos;
772 	iov->nres = nres;
773 	iov->ctrl = ctrl;
774 	iov->total_VFs = total;
775 	iov->driver_max_VFs = total;
776 	pci_read_config_word(dev, pos + PCI_SRIOV_VF_DID, &iov->vf_device);
777 	iov->pgsz = pgsz;
778 	iov->self = dev;
779 	iov->drivers_autoprobe = true;
780 	pci_read_config_dword(dev, pos + PCI_SRIOV_CAP, &iov->cap);
781 	pci_read_config_byte(dev, pos + PCI_SRIOV_FUNC_LINK, &iov->link);
782 	if (pci_pcie_type(dev) == PCI_EXP_TYPE_RC_END)
783 		iov->link = PCI_DEVFN(PCI_SLOT(dev->devfn), iov->link);
784 
785 	if (pdev)
786 		iov->dev = pci_dev_get(pdev);
787 	else
788 		iov->dev = dev;
789 
790 	dev->sriov = iov;
791 	dev->is_physfn = 1;
792 	rc = compute_max_vf_buses(dev);
793 	if (rc)
794 		goto fail_max_buses;
795 
796 	return 0;
797 
798 fail_max_buses:
799 	dev->sriov = NULL;
800 	dev->is_physfn = 0;
801 failed:
802 	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
803 		res = &dev->resource[i + PCI_IOV_RESOURCES];
804 		res->flags = 0;
805 	}
806 
807 	kfree(iov);
808 	return rc;
809 }
810 
811 static void sriov_release(struct pci_dev *dev)
812 {
813 	BUG_ON(dev->sriov->num_VFs);
814 
815 	if (dev != dev->sriov->dev)
816 		pci_dev_put(dev->sriov->dev);
817 
818 	kfree(dev->sriov);
819 	dev->sriov = NULL;
820 }
821 
822 static void sriov_restore_state(struct pci_dev *dev)
823 {
824 	int i;
825 	u16 ctrl;
826 	struct pci_sriov *iov = dev->sriov;
827 
828 	pci_read_config_word(dev, iov->pos + PCI_SRIOV_CTRL, &ctrl);
829 	if (ctrl & PCI_SRIOV_CTRL_VFE)
830 		return;
831 
832 	/*
833 	 * Restore PCI_SRIOV_CTRL_ARI before pci_iov_set_numvfs() because
834 	 * it reads offset & stride, which depend on PCI_SRIOV_CTRL_ARI.
835 	 */
836 	ctrl &= ~PCI_SRIOV_CTRL_ARI;
837 	ctrl |= iov->ctrl & PCI_SRIOV_CTRL_ARI;
838 	pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, ctrl);
839 
840 	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++)
841 		pci_update_resource(dev, i + PCI_IOV_RESOURCES);
842 
843 	pci_write_config_dword(dev, iov->pos + PCI_SRIOV_SYS_PGSIZE, iov->pgsz);
844 	pci_iov_set_numvfs(dev, iov->num_VFs);
845 	pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
846 	if (iov->ctrl & PCI_SRIOV_CTRL_VFE)
847 		msleep(100);
848 }
849 
850 /**
851  * pci_iov_init - initialize the IOV capability
852  * @dev: the PCI device
853  *
854  * Returns 0 on success, or negative on failure.
855  */
856 int pci_iov_init(struct pci_dev *dev)
857 {
858 	int pos;
859 
860 	if (!pci_is_pcie(dev))
861 		return -ENODEV;
862 
863 	pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_SRIOV);
864 	if (pos)
865 		return sriov_init(dev, pos);
866 
867 	return -ENODEV;
868 }
869 
870 /**
871  * pci_iov_release - release resources used by the IOV capability
872  * @dev: the PCI device
873  */
874 void pci_iov_release(struct pci_dev *dev)
875 {
876 	if (dev->is_physfn)
877 		sriov_release(dev);
878 }
879 
880 /**
881  * pci_iov_remove - clean up SR-IOV state after PF driver is detached
882  * @dev: the PCI device
883  */
884 void pci_iov_remove(struct pci_dev *dev)
885 {
886 	struct pci_sriov *iov = dev->sriov;
887 
888 	if (!dev->is_physfn)
889 		return;
890 
891 	iov->driver_max_VFs = iov->total_VFs;
892 	if (iov->num_VFs)
893 		pci_warn(dev, "driver left SR-IOV enabled after remove\n");
894 }
895 
896 /**
897  * pci_iov_update_resource - update a VF BAR
898  * @dev: the PCI device
899  * @resno: the resource number
900  *
901  * Update a VF BAR in the SR-IOV capability of a PF.
902  */
903 void pci_iov_update_resource(struct pci_dev *dev, int resno)
904 {
905 	struct pci_sriov *iov = dev->is_physfn ? dev->sriov : NULL;
906 	struct resource *res = dev->resource + resno;
907 	int vf_bar = resno - PCI_IOV_RESOURCES;
908 	struct pci_bus_region region;
909 	u16 cmd;
910 	u32 new;
911 	int reg;
912 
913 	/*
914 	 * The generic pci_restore_bars() path calls this for all devices,
915 	 * including VFs and non-SR-IOV devices.  If this is not a PF, we
916 	 * have nothing to do.
917 	 */
918 	if (!iov)
919 		return;
920 
921 	pci_read_config_word(dev, iov->pos + PCI_SRIOV_CTRL, &cmd);
922 	if ((cmd & PCI_SRIOV_CTRL_VFE) && (cmd & PCI_SRIOV_CTRL_MSE)) {
923 		dev_WARN(&dev->dev, "can't update enabled VF BAR%d %pR\n",
924 			 vf_bar, res);
925 		return;
926 	}
927 
928 	/*
929 	 * Ignore unimplemented BARs, unused resource slots for 64-bit
930 	 * BARs, and non-movable resources, e.g., those described via
931 	 * Enhanced Allocation.
932 	 */
933 	if (!res->flags)
934 		return;
935 
936 	if (res->flags & IORESOURCE_UNSET)
937 		return;
938 
939 	if (res->flags & IORESOURCE_PCI_FIXED)
940 		return;
941 
942 	pcibios_resource_to_bus(dev->bus, &region, res);
943 	new = region.start;
944 	new |= res->flags & ~PCI_BASE_ADDRESS_MEM_MASK;
945 
946 	reg = iov->pos + PCI_SRIOV_BAR + 4 * vf_bar;
947 	pci_write_config_dword(dev, reg, new);
948 	if (res->flags & IORESOURCE_MEM_64) {
949 		new = region.start >> 16 >> 16;
950 		pci_write_config_dword(dev, reg + 4, new);
951 	}
952 }
953 
954 resource_size_t __weak pcibios_iov_resource_alignment(struct pci_dev *dev,
955 						      int resno)
956 {
957 	return pci_iov_resource_size(dev, resno);
958 }
959 
960 /**
961  * pci_sriov_resource_alignment - get resource alignment for VF BAR
962  * @dev: the PCI device
963  * @resno: the resource number
964  *
965  * Returns the alignment of the VF BAR found in the SR-IOV capability.
966  * This is not the same as the resource size which is defined as
967  * the VF BAR size multiplied by the number of VFs.  The alignment
968  * is just the VF BAR size.
969  */
970 resource_size_t pci_sriov_resource_alignment(struct pci_dev *dev, int resno)
971 {
972 	return pcibios_iov_resource_alignment(dev, resno);
973 }
974 
975 /**
976  * pci_restore_iov_state - restore the state of the IOV capability
977  * @dev: the PCI device
978  */
979 void pci_restore_iov_state(struct pci_dev *dev)
980 {
981 	if (dev->is_physfn)
982 		sriov_restore_state(dev);
983 }
984 
985 /**
986  * pci_vf_drivers_autoprobe - set PF property drivers_autoprobe for VFs
987  * @dev: the PCI device
988  * @auto_probe: set VF drivers auto probe flag
989  */
990 void pci_vf_drivers_autoprobe(struct pci_dev *dev, bool auto_probe)
991 {
992 	if (dev->is_physfn)
993 		dev->sriov->drivers_autoprobe = auto_probe;
994 }
995 
996 /**
997  * pci_iov_bus_range - find bus range used by Virtual Function
998  * @bus: the PCI bus
999  *
1000  * Returns max number of buses (exclude current one) used by Virtual
1001  * Functions.
1002  */
1003 int pci_iov_bus_range(struct pci_bus *bus)
1004 {
1005 	int max = 0;
1006 	struct pci_dev *dev;
1007 
1008 	list_for_each_entry(dev, &bus->devices, bus_list) {
1009 		if (!dev->is_physfn)
1010 			continue;
1011 		if (dev->sriov->max_VF_buses > max)
1012 			max = dev->sriov->max_VF_buses;
1013 	}
1014 
1015 	return max ? max - bus->number : 0;
1016 }
1017 
1018 /**
1019  * pci_enable_sriov - enable the SR-IOV capability
1020  * @dev: the PCI device
1021  * @nr_virtfn: number of virtual functions to enable
1022  *
1023  * Returns 0 on success, or negative on failure.
1024  */
1025 int pci_enable_sriov(struct pci_dev *dev, int nr_virtfn)
1026 {
1027 	might_sleep();
1028 
1029 	if (!dev->is_physfn)
1030 		return -ENOSYS;
1031 
1032 	return sriov_enable(dev, nr_virtfn);
1033 }
1034 EXPORT_SYMBOL_GPL(pci_enable_sriov);
1035 
1036 /**
1037  * pci_disable_sriov - disable the SR-IOV capability
1038  * @dev: the PCI device
1039  */
1040 void pci_disable_sriov(struct pci_dev *dev)
1041 {
1042 	might_sleep();
1043 
1044 	if (!dev->is_physfn)
1045 		return;
1046 
1047 	sriov_disable(dev);
1048 }
1049 EXPORT_SYMBOL_GPL(pci_disable_sriov);
1050 
1051 /**
1052  * pci_num_vf - return number of VFs associated with a PF device_release_driver
1053  * @dev: the PCI device
1054  *
1055  * Returns number of VFs, or 0 if SR-IOV is not enabled.
1056  */
1057 int pci_num_vf(struct pci_dev *dev)
1058 {
1059 	if (!dev->is_physfn)
1060 		return 0;
1061 
1062 	return dev->sriov->num_VFs;
1063 }
1064 EXPORT_SYMBOL_GPL(pci_num_vf);
1065 
1066 /**
1067  * pci_vfs_assigned - returns number of VFs are assigned to a guest
1068  * @dev: the PCI device
1069  *
1070  * Returns number of VFs belonging to this device that are assigned to a guest.
1071  * If device is not a physical function returns 0.
1072  */
1073 int pci_vfs_assigned(struct pci_dev *dev)
1074 {
1075 	struct pci_dev *vfdev;
1076 	unsigned int vfs_assigned = 0;
1077 	unsigned short dev_id;
1078 
1079 	/* only search if we are a PF */
1080 	if (!dev->is_physfn)
1081 		return 0;
1082 
1083 	/*
1084 	 * determine the device ID for the VFs, the vendor ID will be the
1085 	 * same as the PF so there is no need to check for that one
1086 	 */
1087 	dev_id = dev->sriov->vf_device;
1088 
1089 	/* loop through all the VFs to see if we own any that are assigned */
1090 	vfdev = pci_get_device(dev->vendor, dev_id, NULL);
1091 	while (vfdev) {
1092 		/*
1093 		 * It is considered assigned if it is a virtual function with
1094 		 * our dev as the physical function and the assigned bit is set
1095 		 */
1096 		if (vfdev->is_virtfn && (vfdev->physfn == dev) &&
1097 			pci_is_dev_assigned(vfdev))
1098 			vfs_assigned++;
1099 
1100 		vfdev = pci_get_device(dev->vendor, dev_id, vfdev);
1101 	}
1102 
1103 	return vfs_assigned;
1104 }
1105 EXPORT_SYMBOL_GPL(pci_vfs_assigned);
1106 
1107 /**
1108  * pci_sriov_set_totalvfs -- reduce the TotalVFs available
1109  * @dev: the PCI PF device
1110  * @numvfs: number that should be used for TotalVFs supported
1111  *
1112  * Should be called from PF driver's probe routine with
1113  * device's mutex held.
1114  *
1115  * Returns 0 if PF is an SRIOV-capable device and
1116  * value of numvfs valid. If not a PF return -ENOSYS;
1117  * if numvfs is invalid return -EINVAL;
1118  * if VFs already enabled, return -EBUSY.
1119  */
1120 int pci_sriov_set_totalvfs(struct pci_dev *dev, u16 numvfs)
1121 {
1122 	if (!dev->is_physfn)
1123 		return -ENOSYS;
1124 
1125 	if (numvfs > dev->sriov->total_VFs)
1126 		return -EINVAL;
1127 
1128 	/* Shouldn't change if VFs already enabled */
1129 	if (dev->sriov->ctrl & PCI_SRIOV_CTRL_VFE)
1130 		return -EBUSY;
1131 
1132 	dev->sriov->driver_max_VFs = numvfs;
1133 	return 0;
1134 }
1135 EXPORT_SYMBOL_GPL(pci_sriov_set_totalvfs);
1136 
1137 /**
1138  * pci_sriov_get_totalvfs -- get total VFs supported on this device
1139  * @dev: the PCI PF device
1140  *
1141  * For a PCIe device with SRIOV support, return the PCIe
1142  * SRIOV capability value of TotalVFs or the value of driver_max_VFs
1143  * if the driver reduced it.  Otherwise 0.
1144  */
1145 int pci_sriov_get_totalvfs(struct pci_dev *dev)
1146 {
1147 	if (!dev->is_physfn)
1148 		return 0;
1149 
1150 	return dev->sriov->driver_max_VFs;
1151 }
1152 EXPORT_SYMBOL_GPL(pci_sriov_get_totalvfs);
1153 
1154 /**
1155  * pci_sriov_configure_simple - helper to configure SR-IOV
1156  * @dev: the PCI device
1157  * @nr_virtfn: number of virtual functions to enable, 0 to disable
1158  *
1159  * Enable or disable SR-IOV for devices that don't require any PF setup
1160  * before enabling SR-IOV.  Return value is negative on error, or number of
1161  * VFs allocated on success.
1162  */
1163 int pci_sriov_configure_simple(struct pci_dev *dev, int nr_virtfn)
1164 {
1165 	int rc;
1166 
1167 	might_sleep();
1168 
1169 	if (!dev->is_physfn)
1170 		return -ENODEV;
1171 
1172 	if (pci_vfs_assigned(dev)) {
1173 		pci_warn(dev, "Cannot modify SR-IOV while VFs are assigned\n");
1174 		return -EPERM;
1175 	}
1176 
1177 	if (nr_virtfn == 0) {
1178 		sriov_disable(dev);
1179 		return 0;
1180 	}
1181 
1182 	rc = sriov_enable(dev, nr_virtfn);
1183 	if (rc < 0)
1184 		return rc;
1185 
1186 	return nr_virtfn;
1187 }
1188 EXPORT_SYMBOL_GPL(pci_sriov_configure_simple);
1189