xref: /openbmc/linux/drivers/pci/iov.c (revision c4ee0af3)
1 /*
2  * drivers/pci/iov.c
3  *
4  * Copyright (C) 2009 Intel Corporation, Yu Zhao <yu.zhao@intel.com>
5  *
6  * PCI Express I/O Virtualization (IOV) support.
7  *   Single Root IOV 1.0
8  *   Address Translation Service 1.0
9  */
10 
11 #include <linux/pci.h>
12 #include <linux/slab.h>
13 #include <linux/mutex.h>
14 #include <linux/export.h>
15 #include <linux/string.h>
16 #include <linux/delay.h>
17 #include <linux/pci-ats.h>
18 #include "pci.h"
19 
20 #define VIRTFN_ID_LEN	16
21 
22 static inline u8 virtfn_bus(struct pci_dev *dev, int id)
23 {
24 	return dev->bus->number + ((dev->devfn + dev->sriov->offset +
25 				    dev->sriov->stride * id) >> 8);
26 }
27 
28 static inline u8 virtfn_devfn(struct pci_dev *dev, int id)
29 {
30 	return (dev->devfn + dev->sriov->offset +
31 		dev->sriov->stride * id) & 0xff;
32 }
33 
34 static struct pci_bus *virtfn_add_bus(struct pci_bus *bus, int busnr)
35 {
36 	struct pci_bus *child;
37 
38 	if (bus->number == busnr)
39 		return bus;
40 
41 	child = pci_find_bus(pci_domain_nr(bus), busnr);
42 	if (child)
43 		return child;
44 
45 	child = pci_add_new_bus(bus, NULL, busnr);
46 	if (!child)
47 		return NULL;
48 
49 	pci_bus_insert_busn_res(child, busnr, busnr);
50 
51 	return child;
52 }
53 
54 static void virtfn_remove_bus(struct pci_bus *physbus, struct pci_bus *virtbus)
55 {
56 	if (physbus != virtbus && list_empty(&virtbus->devices))
57 		pci_remove_bus(virtbus);
58 }
59 
60 static int virtfn_add(struct pci_dev *dev, int id, int reset)
61 {
62 	int i;
63 	int rc = -ENOMEM;
64 	u64 size;
65 	char buf[VIRTFN_ID_LEN];
66 	struct pci_dev *virtfn;
67 	struct resource *res;
68 	struct pci_sriov *iov = dev->sriov;
69 	struct pci_bus *bus;
70 
71 	mutex_lock(&iov->dev->sriov->lock);
72 	bus = virtfn_add_bus(dev->bus, virtfn_bus(dev, id));
73 	if (!bus)
74 		goto failed;
75 
76 	virtfn = pci_alloc_dev(bus);
77 	if (!virtfn)
78 		goto failed0;
79 
80 	virtfn->devfn = virtfn_devfn(dev, id);
81 	virtfn->vendor = dev->vendor;
82 	pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_DID, &virtfn->device);
83 	pci_setup_device(virtfn);
84 	virtfn->dev.parent = dev->dev.parent;
85 	virtfn->physfn = pci_dev_get(dev);
86 	virtfn->is_virtfn = 1;
87 
88 	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
89 		res = dev->resource + PCI_IOV_RESOURCES + i;
90 		if (!res->parent)
91 			continue;
92 		virtfn->resource[i].name = pci_name(virtfn);
93 		virtfn->resource[i].flags = res->flags;
94 		size = resource_size(res);
95 		do_div(size, iov->total_VFs);
96 		virtfn->resource[i].start = res->start + size * id;
97 		virtfn->resource[i].end = virtfn->resource[i].start + size - 1;
98 		rc = request_resource(res, &virtfn->resource[i]);
99 		BUG_ON(rc);
100 	}
101 
102 	if (reset)
103 		__pci_reset_function(virtfn);
104 
105 	pci_device_add(virtfn, virtfn->bus);
106 	mutex_unlock(&iov->dev->sriov->lock);
107 
108 	rc = pci_bus_add_device(virtfn);
109 	sprintf(buf, "virtfn%u", id);
110 	rc = sysfs_create_link(&dev->dev.kobj, &virtfn->dev.kobj, buf);
111 	if (rc)
112 		goto failed1;
113 	rc = sysfs_create_link(&virtfn->dev.kobj, &dev->dev.kobj, "physfn");
114 	if (rc)
115 		goto failed2;
116 
117 	kobject_uevent(&virtfn->dev.kobj, KOBJ_CHANGE);
118 
119 	return 0;
120 
121 failed2:
122 	sysfs_remove_link(&dev->dev.kobj, buf);
123 failed1:
124 	pci_dev_put(dev);
125 	mutex_lock(&iov->dev->sriov->lock);
126 	pci_stop_and_remove_bus_device(virtfn);
127 failed0:
128 	virtfn_remove_bus(dev->bus, bus);
129 failed:
130 	mutex_unlock(&iov->dev->sriov->lock);
131 
132 	return rc;
133 }
134 
135 static void virtfn_remove(struct pci_dev *dev, int id, int reset)
136 {
137 	char buf[VIRTFN_ID_LEN];
138 	struct pci_dev *virtfn;
139 	struct pci_sriov *iov = dev->sriov;
140 
141 	virtfn = pci_get_domain_bus_and_slot(pci_domain_nr(dev->bus),
142 					     virtfn_bus(dev, id),
143 					     virtfn_devfn(dev, id));
144 	if (!virtfn)
145 		return;
146 
147 	if (reset) {
148 		device_release_driver(&virtfn->dev);
149 		__pci_reset_function(virtfn);
150 	}
151 
152 	sprintf(buf, "virtfn%u", id);
153 	sysfs_remove_link(&dev->dev.kobj, buf);
154 	/*
155 	 * pci_stop_dev() could have been called for this virtfn already,
156 	 * so the directory for the virtfn may have been removed before.
157 	 * Double check to avoid spurious sysfs warnings.
158 	 */
159 	if (virtfn->dev.kobj.sd)
160 		sysfs_remove_link(&virtfn->dev.kobj, "physfn");
161 
162 	mutex_lock(&iov->dev->sriov->lock);
163 	pci_stop_and_remove_bus_device(virtfn);
164 	virtfn_remove_bus(dev->bus, virtfn->bus);
165 	mutex_unlock(&iov->dev->sriov->lock);
166 
167 	/* balance pci_get_domain_bus_and_slot() */
168 	pci_dev_put(virtfn);
169 	pci_dev_put(dev);
170 }
171 
172 static int sriov_migration(struct pci_dev *dev)
173 {
174 	u16 status;
175 	struct pci_sriov *iov = dev->sriov;
176 
177 	if (!iov->num_VFs)
178 		return 0;
179 
180 	if (!(iov->cap & PCI_SRIOV_CAP_VFM))
181 		return 0;
182 
183 	pci_read_config_word(dev, iov->pos + PCI_SRIOV_STATUS, &status);
184 	if (!(status & PCI_SRIOV_STATUS_VFM))
185 		return 0;
186 
187 	schedule_work(&iov->mtask);
188 
189 	return 1;
190 }
191 
192 static void sriov_migration_task(struct work_struct *work)
193 {
194 	int i;
195 	u8 state;
196 	u16 status;
197 	struct pci_sriov *iov = container_of(work, struct pci_sriov, mtask);
198 
199 	for (i = iov->initial_VFs; i < iov->num_VFs; i++) {
200 		state = readb(iov->mstate + i);
201 		if (state == PCI_SRIOV_VFM_MI) {
202 			writeb(PCI_SRIOV_VFM_AV, iov->mstate + i);
203 			state = readb(iov->mstate + i);
204 			if (state == PCI_SRIOV_VFM_AV)
205 				virtfn_add(iov->self, i, 1);
206 		} else if (state == PCI_SRIOV_VFM_MO) {
207 			virtfn_remove(iov->self, i, 1);
208 			writeb(PCI_SRIOV_VFM_UA, iov->mstate + i);
209 			state = readb(iov->mstate + i);
210 			if (state == PCI_SRIOV_VFM_AV)
211 				virtfn_add(iov->self, i, 0);
212 		}
213 	}
214 
215 	pci_read_config_word(iov->self, iov->pos + PCI_SRIOV_STATUS, &status);
216 	status &= ~PCI_SRIOV_STATUS_VFM;
217 	pci_write_config_word(iov->self, iov->pos + PCI_SRIOV_STATUS, status);
218 }
219 
220 static int sriov_enable_migration(struct pci_dev *dev, int nr_virtfn)
221 {
222 	int bir;
223 	u32 table;
224 	resource_size_t pa;
225 	struct pci_sriov *iov = dev->sriov;
226 
227 	if (nr_virtfn <= iov->initial_VFs)
228 		return 0;
229 
230 	pci_read_config_dword(dev, iov->pos + PCI_SRIOV_VFM, &table);
231 	bir = PCI_SRIOV_VFM_BIR(table);
232 	if (bir > PCI_STD_RESOURCE_END)
233 		return -EIO;
234 
235 	table = PCI_SRIOV_VFM_OFFSET(table);
236 	if (table + nr_virtfn > pci_resource_len(dev, bir))
237 		return -EIO;
238 
239 	pa = pci_resource_start(dev, bir) + table;
240 	iov->mstate = ioremap(pa, nr_virtfn);
241 	if (!iov->mstate)
242 		return -ENOMEM;
243 
244 	INIT_WORK(&iov->mtask, sriov_migration_task);
245 
246 	iov->ctrl |= PCI_SRIOV_CTRL_VFM | PCI_SRIOV_CTRL_INTR;
247 	pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
248 
249 	return 0;
250 }
251 
252 static void sriov_disable_migration(struct pci_dev *dev)
253 {
254 	struct pci_sriov *iov = dev->sriov;
255 
256 	iov->ctrl &= ~(PCI_SRIOV_CTRL_VFM | PCI_SRIOV_CTRL_INTR);
257 	pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
258 
259 	cancel_work_sync(&iov->mtask);
260 	iounmap(iov->mstate);
261 }
262 
263 static int sriov_enable(struct pci_dev *dev, int nr_virtfn)
264 {
265 	int rc;
266 	int i, j;
267 	int nres;
268 	u16 offset, stride, initial;
269 	struct resource *res;
270 	struct pci_dev *pdev;
271 	struct pci_sriov *iov = dev->sriov;
272 	int bars = 0;
273 
274 	if (!nr_virtfn)
275 		return 0;
276 
277 	if (iov->num_VFs)
278 		return -EINVAL;
279 
280 	pci_read_config_word(dev, iov->pos + PCI_SRIOV_INITIAL_VF, &initial);
281 	if (initial > iov->total_VFs ||
282 	    (!(iov->cap & PCI_SRIOV_CAP_VFM) && (initial != iov->total_VFs)))
283 		return -EIO;
284 
285 	if (nr_virtfn < 0 || nr_virtfn > iov->total_VFs ||
286 	    (!(iov->cap & PCI_SRIOV_CAP_VFM) && (nr_virtfn > initial)))
287 		return -EINVAL;
288 
289 	pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_OFFSET, &offset);
290 	pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_STRIDE, &stride);
291 	if (!offset || (nr_virtfn > 1 && !stride))
292 		return -EIO;
293 
294 	nres = 0;
295 	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
296 		bars |= (1 << (i + PCI_IOV_RESOURCES));
297 		res = dev->resource + PCI_IOV_RESOURCES + i;
298 		if (res->parent)
299 			nres++;
300 	}
301 	if (nres != iov->nres) {
302 		dev_err(&dev->dev, "not enough MMIO resources for SR-IOV\n");
303 		return -ENOMEM;
304 	}
305 
306 	iov->offset = offset;
307 	iov->stride = stride;
308 
309 	if (virtfn_bus(dev, nr_virtfn - 1) > dev->bus->busn_res.end) {
310 		dev_err(&dev->dev, "SR-IOV: bus number out of range\n");
311 		return -ENOMEM;
312 	}
313 
314 	if (pci_enable_resources(dev, bars)) {
315 		dev_err(&dev->dev, "SR-IOV: IOV BARS not allocated\n");
316 		return -ENOMEM;
317 	}
318 
319 	if (iov->link != dev->devfn) {
320 		pdev = pci_get_slot(dev->bus, iov->link);
321 		if (!pdev)
322 			return -ENODEV;
323 
324 		if (!pdev->is_physfn) {
325 			pci_dev_put(pdev);
326 			return -ENOSYS;
327 		}
328 
329 		rc = sysfs_create_link(&dev->dev.kobj,
330 					&pdev->dev.kobj, "dep_link");
331 		pci_dev_put(pdev);
332 		if (rc)
333 			return rc;
334 	}
335 
336 	pci_write_config_word(dev, iov->pos + PCI_SRIOV_NUM_VF, nr_virtfn);
337 	iov->ctrl |= PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE;
338 	pci_cfg_access_lock(dev);
339 	pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
340 	msleep(100);
341 	pci_cfg_access_unlock(dev);
342 
343 	iov->initial_VFs = initial;
344 	if (nr_virtfn < initial)
345 		initial = nr_virtfn;
346 
347 	for (i = 0; i < initial; i++) {
348 		rc = virtfn_add(dev, i, 0);
349 		if (rc)
350 			goto failed;
351 	}
352 
353 	if (iov->cap & PCI_SRIOV_CAP_VFM) {
354 		rc = sriov_enable_migration(dev, nr_virtfn);
355 		if (rc)
356 			goto failed;
357 	}
358 
359 	kobject_uevent(&dev->dev.kobj, KOBJ_CHANGE);
360 	iov->num_VFs = nr_virtfn;
361 
362 	return 0;
363 
364 failed:
365 	for (j = 0; j < i; j++)
366 		virtfn_remove(dev, j, 0);
367 
368 	iov->ctrl &= ~(PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE);
369 	pci_cfg_access_lock(dev);
370 	pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
371 	pci_write_config_word(dev, iov->pos + PCI_SRIOV_NUM_VF, 0);
372 	ssleep(1);
373 	pci_cfg_access_unlock(dev);
374 
375 	if (iov->link != dev->devfn)
376 		sysfs_remove_link(&dev->dev.kobj, "dep_link");
377 
378 	return rc;
379 }
380 
381 static void sriov_disable(struct pci_dev *dev)
382 {
383 	int i;
384 	struct pci_sriov *iov = dev->sriov;
385 
386 	if (!iov->num_VFs)
387 		return;
388 
389 	if (iov->cap & PCI_SRIOV_CAP_VFM)
390 		sriov_disable_migration(dev);
391 
392 	for (i = 0; i < iov->num_VFs; i++)
393 		virtfn_remove(dev, i, 0);
394 
395 	iov->ctrl &= ~(PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE);
396 	pci_cfg_access_lock(dev);
397 	pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
398 	ssleep(1);
399 	pci_cfg_access_unlock(dev);
400 
401 	if (iov->link != dev->devfn)
402 		sysfs_remove_link(&dev->dev.kobj, "dep_link");
403 
404 	iov->num_VFs = 0;
405 	pci_write_config_word(dev, iov->pos + PCI_SRIOV_NUM_VF, 0);
406 }
407 
408 static int sriov_init(struct pci_dev *dev, int pos)
409 {
410 	int i;
411 	int rc;
412 	int nres;
413 	u32 pgsz;
414 	u16 ctrl, total, offset, stride;
415 	struct pci_sriov *iov;
416 	struct resource *res;
417 	struct pci_dev *pdev;
418 
419 	if (pci_pcie_type(dev) != PCI_EXP_TYPE_RC_END &&
420 	    pci_pcie_type(dev) != PCI_EXP_TYPE_ENDPOINT)
421 		return -ENODEV;
422 
423 	pci_read_config_word(dev, pos + PCI_SRIOV_CTRL, &ctrl);
424 	if (ctrl & PCI_SRIOV_CTRL_VFE) {
425 		pci_write_config_word(dev, pos + PCI_SRIOV_CTRL, 0);
426 		ssleep(1);
427 	}
428 
429 	pci_read_config_word(dev, pos + PCI_SRIOV_TOTAL_VF, &total);
430 	if (!total)
431 		return 0;
432 
433 	ctrl = 0;
434 	list_for_each_entry(pdev, &dev->bus->devices, bus_list)
435 		if (pdev->is_physfn)
436 			goto found;
437 
438 	pdev = NULL;
439 	if (pci_ari_enabled(dev->bus))
440 		ctrl |= PCI_SRIOV_CTRL_ARI;
441 
442 found:
443 	pci_write_config_word(dev, pos + PCI_SRIOV_CTRL, ctrl);
444 	pci_read_config_word(dev, pos + PCI_SRIOV_VF_OFFSET, &offset);
445 	pci_read_config_word(dev, pos + PCI_SRIOV_VF_STRIDE, &stride);
446 	if (!offset || (total > 1 && !stride))
447 		return -EIO;
448 
449 	pci_read_config_dword(dev, pos + PCI_SRIOV_SUP_PGSIZE, &pgsz);
450 	i = PAGE_SHIFT > 12 ? PAGE_SHIFT - 12 : 0;
451 	pgsz &= ~((1 << i) - 1);
452 	if (!pgsz)
453 		return -EIO;
454 
455 	pgsz &= ~(pgsz - 1);
456 	pci_write_config_dword(dev, pos + PCI_SRIOV_SYS_PGSIZE, pgsz);
457 
458 	nres = 0;
459 	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
460 		res = dev->resource + PCI_IOV_RESOURCES + i;
461 		i += __pci_read_base(dev, pci_bar_unknown, res,
462 				     pos + PCI_SRIOV_BAR + i * 4);
463 		if (!res->flags)
464 			continue;
465 		if (resource_size(res) & (PAGE_SIZE - 1)) {
466 			rc = -EIO;
467 			goto failed;
468 		}
469 		res->end = res->start + resource_size(res) * total - 1;
470 		nres++;
471 	}
472 
473 	iov = kzalloc(sizeof(*iov), GFP_KERNEL);
474 	if (!iov) {
475 		rc = -ENOMEM;
476 		goto failed;
477 	}
478 
479 	iov->pos = pos;
480 	iov->nres = nres;
481 	iov->ctrl = ctrl;
482 	iov->total_VFs = total;
483 	iov->offset = offset;
484 	iov->stride = stride;
485 	iov->pgsz = pgsz;
486 	iov->self = dev;
487 	pci_read_config_dword(dev, pos + PCI_SRIOV_CAP, &iov->cap);
488 	pci_read_config_byte(dev, pos + PCI_SRIOV_FUNC_LINK, &iov->link);
489 	if (pci_pcie_type(dev) == PCI_EXP_TYPE_RC_END)
490 		iov->link = PCI_DEVFN(PCI_SLOT(dev->devfn), iov->link);
491 
492 	if (pdev)
493 		iov->dev = pci_dev_get(pdev);
494 	else
495 		iov->dev = dev;
496 
497 	mutex_init(&iov->lock);
498 
499 	dev->sriov = iov;
500 	dev->is_physfn = 1;
501 
502 	return 0;
503 
504 failed:
505 	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
506 		res = dev->resource + PCI_IOV_RESOURCES + i;
507 		res->flags = 0;
508 	}
509 
510 	return rc;
511 }
512 
513 static void sriov_release(struct pci_dev *dev)
514 {
515 	BUG_ON(dev->sriov->num_VFs);
516 
517 	if (dev != dev->sriov->dev)
518 		pci_dev_put(dev->sriov->dev);
519 
520 	mutex_destroy(&dev->sriov->lock);
521 
522 	kfree(dev->sriov);
523 	dev->sriov = NULL;
524 }
525 
526 static void sriov_restore_state(struct pci_dev *dev)
527 {
528 	int i;
529 	u16 ctrl;
530 	struct pci_sriov *iov = dev->sriov;
531 
532 	pci_read_config_word(dev, iov->pos + PCI_SRIOV_CTRL, &ctrl);
533 	if (ctrl & PCI_SRIOV_CTRL_VFE)
534 		return;
535 
536 	for (i = PCI_IOV_RESOURCES; i <= PCI_IOV_RESOURCE_END; i++)
537 		pci_update_resource(dev, i);
538 
539 	pci_write_config_dword(dev, iov->pos + PCI_SRIOV_SYS_PGSIZE, iov->pgsz);
540 	pci_write_config_word(dev, iov->pos + PCI_SRIOV_NUM_VF, iov->num_VFs);
541 	pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
542 	if (iov->ctrl & PCI_SRIOV_CTRL_VFE)
543 		msleep(100);
544 }
545 
546 /**
547  * pci_iov_init - initialize the IOV capability
548  * @dev: the PCI device
549  *
550  * Returns 0 on success, or negative on failure.
551  */
552 int pci_iov_init(struct pci_dev *dev)
553 {
554 	int pos;
555 
556 	if (!pci_is_pcie(dev))
557 		return -ENODEV;
558 
559 	pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_SRIOV);
560 	if (pos)
561 		return sriov_init(dev, pos);
562 
563 	return -ENODEV;
564 }
565 
566 /**
567  * pci_iov_release - release resources used by the IOV capability
568  * @dev: the PCI device
569  */
570 void pci_iov_release(struct pci_dev *dev)
571 {
572 	if (dev->is_physfn)
573 		sriov_release(dev);
574 }
575 
576 /**
577  * pci_iov_resource_bar - get position of the SR-IOV BAR
578  * @dev: the PCI device
579  * @resno: the resource number
580  * @type: the BAR type to be filled in
581  *
582  * Returns position of the BAR encapsulated in the SR-IOV capability.
583  */
584 int pci_iov_resource_bar(struct pci_dev *dev, int resno,
585 			 enum pci_bar_type *type)
586 {
587 	if (resno < PCI_IOV_RESOURCES || resno > PCI_IOV_RESOURCE_END)
588 		return 0;
589 
590 	BUG_ON(!dev->is_physfn);
591 
592 	*type = pci_bar_unknown;
593 
594 	return dev->sriov->pos + PCI_SRIOV_BAR +
595 		4 * (resno - PCI_IOV_RESOURCES);
596 }
597 
598 /**
599  * pci_sriov_resource_alignment - get resource alignment for VF BAR
600  * @dev: the PCI device
601  * @resno: the resource number
602  *
603  * Returns the alignment of the VF BAR found in the SR-IOV capability.
604  * This is not the same as the resource size which is defined as
605  * the VF BAR size multiplied by the number of VFs.  The alignment
606  * is just the VF BAR size.
607  */
608 resource_size_t pci_sriov_resource_alignment(struct pci_dev *dev, int resno)
609 {
610 	struct resource tmp;
611 	enum pci_bar_type type;
612 	int reg = pci_iov_resource_bar(dev, resno, &type);
613 
614 	if (!reg)
615 		return 0;
616 
617 	 __pci_read_base(dev, type, &tmp, reg);
618 	return resource_alignment(&tmp);
619 }
620 
621 /**
622  * pci_restore_iov_state - restore the state of the IOV capability
623  * @dev: the PCI device
624  */
625 void pci_restore_iov_state(struct pci_dev *dev)
626 {
627 	if (dev->is_physfn)
628 		sriov_restore_state(dev);
629 }
630 
631 /**
632  * pci_iov_bus_range - find bus range used by Virtual Function
633  * @bus: the PCI bus
634  *
635  * Returns max number of buses (exclude current one) used by Virtual
636  * Functions.
637  */
638 int pci_iov_bus_range(struct pci_bus *bus)
639 {
640 	int max = 0;
641 	u8 busnr;
642 	struct pci_dev *dev;
643 
644 	list_for_each_entry(dev, &bus->devices, bus_list) {
645 		if (!dev->is_physfn)
646 			continue;
647 		busnr = virtfn_bus(dev, dev->sriov->total_VFs - 1);
648 		if (busnr > max)
649 			max = busnr;
650 	}
651 
652 	return max ? max - bus->number : 0;
653 }
654 
655 /**
656  * pci_enable_sriov - enable the SR-IOV capability
657  * @dev: the PCI device
658  * @nr_virtfn: number of virtual functions to enable
659  *
660  * Returns 0 on success, or negative on failure.
661  */
662 int pci_enable_sriov(struct pci_dev *dev, int nr_virtfn)
663 {
664 	might_sleep();
665 
666 	if (!dev->is_physfn)
667 		return -ENOSYS;
668 
669 	return sriov_enable(dev, nr_virtfn);
670 }
671 EXPORT_SYMBOL_GPL(pci_enable_sriov);
672 
673 /**
674  * pci_disable_sriov - disable the SR-IOV capability
675  * @dev: the PCI device
676  */
677 void pci_disable_sriov(struct pci_dev *dev)
678 {
679 	might_sleep();
680 
681 	if (!dev->is_physfn)
682 		return;
683 
684 	sriov_disable(dev);
685 }
686 EXPORT_SYMBOL_GPL(pci_disable_sriov);
687 
688 /**
689  * pci_sriov_migration - notify SR-IOV core of Virtual Function Migration
690  * @dev: the PCI device
691  *
692  * Returns IRQ_HANDLED if the IRQ is handled, or IRQ_NONE if not.
693  *
694  * Physical Function driver is responsible to register IRQ handler using
695  * VF Migration Interrupt Message Number, and call this function when the
696  * interrupt is generated by the hardware.
697  */
698 irqreturn_t pci_sriov_migration(struct pci_dev *dev)
699 {
700 	if (!dev->is_physfn)
701 		return IRQ_NONE;
702 
703 	return sriov_migration(dev) ? IRQ_HANDLED : IRQ_NONE;
704 }
705 EXPORT_SYMBOL_GPL(pci_sriov_migration);
706 
707 /**
708  * pci_num_vf - return number of VFs associated with a PF device_release_driver
709  * @dev: the PCI device
710  *
711  * Returns number of VFs, or 0 if SR-IOV is not enabled.
712  */
713 int pci_num_vf(struct pci_dev *dev)
714 {
715 	if (!dev->is_physfn)
716 		return 0;
717 
718 	return dev->sriov->num_VFs;
719 }
720 EXPORT_SYMBOL_GPL(pci_num_vf);
721 
722 /**
723  * pci_vfs_assigned - returns number of VFs are assigned to a guest
724  * @dev: the PCI device
725  *
726  * Returns number of VFs belonging to this device that are assigned to a guest.
727  * If device is not a physical function returns 0.
728  */
729 int pci_vfs_assigned(struct pci_dev *dev)
730 {
731 	struct pci_dev *vfdev;
732 	unsigned int vfs_assigned = 0;
733 	unsigned short dev_id;
734 
735 	/* only search if we are a PF */
736 	if (!dev->is_physfn)
737 		return 0;
738 
739 	/*
740 	 * determine the device ID for the VFs, the vendor ID will be the
741 	 * same as the PF so there is no need to check for that one
742 	 */
743 	pci_read_config_word(dev, dev->sriov->pos + PCI_SRIOV_VF_DID, &dev_id);
744 
745 	/* loop through all the VFs to see if we own any that are assigned */
746 	vfdev = pci_get_device(dev->vendor, dev_id, NULL);
747 	while (vfdev) {
748 		/*
749 		 * It is considered assigned if it is a virtual function with
750 		 * our dev as the physical function and the assigned bit is set
751 		 */
752 		if (vfdev->is_virtfn && (vfdev->physfn == dev) &&
753 		    (vfdev->dev_flags & PCI_DEV_FLAGS_ASSIGNED))
754 			vfs_assigned++;
755 
756 		vfdev = pci_get_device(dev->vendor, dev_id, vfdev);
757 	}
758 
759 	return vfs_assigned;
760 }
761 EXPORT_SYMBOL_GPL(pci_vfs_assigned);
762 
763 /**
764  * pci_sriov_set_totalvfs -- reduce the TotalVFs available
765  * @dev: the PCI PF device
766  * @numvfs: number that should be used for TotalVFs supported
767  *
768  * Should be called from PF driver's probe routine with
769  * device's mutex held.
770  *
771  * Returns 0 if PF is an SRIOV-capable device and
772  * value of numvfs valid. If not a PF return -ENOSYS;
773  * if numvfs is invalid return -EINVAL;
774  * if VFs already enabled, return -EBUSY.
775  */
776 int pci_sriov_set_totalvfs(struct pci_dev *dev, u16 numvfs)
777 {
778 	if (!dev->is_physfn)
779 		return -ENOSYS;
780 	if (numvfs > dev->sriov->total_VFs)
781 		return -EINVAL;
782 
783 	/* Shouldn't change if VFs already enabled */
784 	if (dev->sriov->ctrl & PCI_SRIOV_CTRL_VFE)
785 		return -EBUSY;
786 	else
787 		dev->sriov->driver_max_VFs = numvfs;
788 
789 	return 0;
790 }
791 EXPORT_SYMBOL_GPL(pci_sriov_set_totalvfs);
792 
793 /**
794  * pci_sriov_get_totalvfs -- get total VFs supported on this device
795  * @dev: the PCI PF device
796  *
797  * For a PCIe device with SRIOV support, return the PCIe
798  * SRIOV capability value of TotalVFs or the value of driver_max_VFs
799  * if the driver reduced it.  Otherwise 0.
800  */
801 int pci_sriov_get_totalvfs(struct pci_dev *dev)
802 {
803 	if (!dev->is_physfn)
804 		return 0;
805 
806 	if (dev->sriov->driver_max_VFs)
807 		return dev->sriov->driver_max_VFs;
808 
809 	return dev->sriov->total_VFs;
810 }
811 EXPORT_SYMBOL_GPL(pci_sriov_get_totalvfs);
812