xref: /linux/drivers/pci/iov.c (revision ff5599816711d2e67da2d7561fd36ac48debd433)
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_write_config_word(dev, iov->pos + PCI_SRIOV_NUM_VF, nr_virtfn);
290 	pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_OFFSET, &offset);
291 	pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_STRIDE, &stride);
292 	if (!offset || (nr_virtfn > 1 && !stride))
293 		return -EIO;
294 
295 	nres = 0;
296 	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
297 		bars |= (1 << (i + PCI_IOV_RESOURCES));
298 		res = dev->resource + PCI_IOV_RESOURCES + i;
299 		if (res->parent)
300 			nres++;
301 	}
302 	if (nres != iov->nres) {
303 		dev_err(&dev->dev, "not enough MMIO resources for SR-IOV\n");
304 		return -ENOMEM;
305 	}
306 
307 	iov->offset = offset;
308 	iov->stride = stride;
309 
310 	if (virtfn_bus(dev, nr_virtfn - 1) > dev->bus->busn_res.end) {
311 		dev_err(&dev->dev, "SR-IOV: bus number out of range\n");
312 		return -ENOMEM;
313 	}
314 
315 	if (pci_enable_resources(dev, bars)) {
316 		dev_err(&dev->dev, "SR-IOV: IOV BARS not allocated\n");
317 		return -ENOMEM;
318 	}
319 
320 	if (iov->link != dev->devfn) {
321 		pdev = pci_get_slot(dev->bus, iov->link);
322 		if (!pdev)
323 			return -ENODEV;
324 
325 		if (!pdev->is_physfn) {
326 			pci_dev_put(pdev);
327 			return -ENODEV;
328 		}
329 
330 		rc = sysfs_create_link(&dev->dev.kobj,
331 					&pdev->dev.kobj, "dep_link");
332 		pci_dev_put(pdev);
333 		if (rc)
334 			return rc;
335 	}
336 
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 	ssleep(1);
372 	pci_cfg_access_unlock(dev);
373 
374 	if (iov->link != dev->devfn)
375 		sysfs_remove_link(&dev->dev.kobj, "dep_link");
376 
377 	return rc;
378 }
379 
380 static void sriov_disable(struct pci_dev *dev)
381 {
382 	int i;
383 	struct pci_sriov *iov = dev->sriov;
384 
385 	if (!iov->num_VFs)
386 		return;
387 
388 	if (iov->cap & PCI_SRIOV_CAP_VFM)
389 		sriov_disable_migration(dev);
390 
391 	for (i = 0; i < iov->num_VFs; i++)
392 		virtfn_remove(dev, i, 0);
393 
394 	iov->ctrl &= ~(PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE);
395 	pci_cfg_access_lock(dev);
396 	pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
397 	ssleep(1);
398 	pci_cfg_access_unlock(dev);
399 
400 	if (iov->link != dev->devfn)
401 		sysfs_remove_link(&dev->dev.kobj, "dep_link");
402 
403 	iov->num_VFs = 0;
404 }
405 
406 static int sriov_init(struct pci_dev *dev, int pos)
407 {
408 	int i;
409 	int rc;
410 	int nres;
411 	u32 pgsz;
412 	u16 ctrl, total, offset, stride;
413 	struct pci_sriov *iov;
414 	struct resource *res;
415 	struct pci_dev *pdev;
416 
417 	if (pci_pcie_type(dev) != PCI_EXP_TYPE_RC_END &&
418 	    pci_pcie_type(dev) != PCI_EXP_TYPE_ENDPOINT)
419 		return -ENODEV;
420 
421 	pci_read_config_word(dev, pos + PCI_SRIOV_CTRL, &ctrl);
422 	if (ctrl & PCI_SRIOV_CTRL_VFE) {
423 		pci_write_config_word(dev, pos + PCI_SRIOV_CTRL, 0);
424 		ssleep(1);
425 	}
426 
427 	pci_read_config_word(dev, pos + PCI_SRIOV_TOTAL_VF, &total);
428 	if (!total)
429 		return 0;
430 
431 	ctrl = 0;
432 	list_for_each_entry(pdev, &dev->bus->devices, bus_list)
433 		if (pdev->is_physfn)
434 			goto found;
435 
436 	pdev = NULL;
437 	if (pci_ari_enabled(dev->bus))
438 		ctrl |= PCI_SRIOV_CTRL_ARI;
439 
440 found:
441 	pci_write_config_word(dev, pos + PCI_SRIOV_CTRL, ctrl);
442 	pci_read_config_word(dev, pos + PCI_SRIOV_VF_OFFSET, &offset);
443 	pci_read_config_word(dev, pos + PCI_SRIOV_VF_STRIDE, &stride);
444 	if (!offset || (total > 1 && !stride))
445 		return -EIO;
446 
447 	pci_read_config_dword(dev, pos + PCI_SRIOV_SUP_PGSIZE, &pgsz);
448 	i = PAGE_SHIFT > 12 ? PAGE_SHIFT - 12 : 0;
449 	pgsz &= ~((1 << i) - 1);
450 	if (!pgsz)
451 		return -EIO;
452 
453 	pgsz &= ~(pgsz - 1);
454 	pci_write_config_dword(dev, pos + PCI_SRIOV_SYS_PGSIZE, pgsz);
455 
456 	nres = 0;
457 	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
458 		res = dev->resource + PCI_IOV_RESOURCES + i;
459 		i += __pci_read_base(dev, pci_bar_unknown, res,
460 				     pos + PCI_SRIOV_BAR + i * 4);
461 		if (!res->flags)
462 			continue;
463 		if (resource_size(res) & (PAGE_SIZE - 1)) {
464 			rc = -EIO;
465 			goto failed;
466 		}
467 		res->end = res->start + resource_size(res) * total - 1;
468 		nres++;
469 	}
470 
471 	iov = kzalloc(sizeof(*iov), GFP_KERNEL);
472 	if (!iov) {
473 		rc = -ENOMEM;
474 		goto failed;
475 	}
476 
477 	iov->pos = pos;
478 	iov->nres = nres;
479 	iov->ctrl = ctrl;
480 	iov->total_VFs = total;
481 	iov->offset = offset;
482 	iov->stride = stride;
483 	iov->pgsz = pgsz;
484 	iov->self = dev;
485 	pci_read_config_dword(dev, pos + PCI_SRIOV_CAP, &iov->cap);
486 	pci_read_config_byte(dev, pos + PCI_SRIOV_FUNC_LINK, &iov->link);
487 	if (pci_pcie_type(dev) == PCI_EXP_TYPE_RC_END)
488 		iov->link = PCI_DEVFN(PCI_SLOT(dev->devfn), iov->link);
489 
490 	if (pdev)
491 		iov->dev = pci_dev_get(pdev);
492 	else
493 		iov->dev = dev;
494 
495 	mutex_init(&iov->lock);
496 
497 	dev->sriov = iov;
498 	dev->is_physfn = 1;
499 
500 	return 0;
501 
502 failed:
503 	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
504 		res = dev->resource + PCI_IOV_RESOURCES + i;
505 		res->flags = 0;
506 	}
507 
508 	return rc;
509 }
510 
511 static void sriov_release(struct pci_dev *dev)
512 {
513 	BUG_ON(dev->sriov->num_VFs);
514 
515 	if (dev != dev->sriov->dev)
516 		pci_dev_put(dev->sriov->dev);
517 
518 	mutex_destroy(&dev->sriov->lock);
519 
520 	kfree(dev->sriov);
521 	dev->sriov = NULL;
522 }
523 
524 static void sriov_restore_state(struct pci_dev *dev)
525 {
526 	int i;
527 	u16 ctrl;
528 	struct pci_sriov *iov = dev->sriov;
529 
530 	pci_read_config_word(dev, iov->pos + PCI_SRIOV_CTRL, &ctrl);
531 	if (ctrl & PCI_SRIOV_CTRL_VFE)
532 		return;
533 
534 	for (i = PCI_IOV_RESOURCES; i <= PCI_IOV_RESOURCE_END; i++)
535 		pci_update_resource(dev, i);
536 
537 	pci_write_config_dword(dev, iov->pos + PCI_SRIOV_SYS_PGSIZE, iov->pgsz);
538 	pci_write_config_word(dev, iov->pos + PCI_SRIOV_NUM_VF, iov->num_VFs);
539 	pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
540 	if (iov->ctrl & PCI_SRIOV_CTRL_VFE)
541 		msleep(100);
542 }
543 
544 /**
545  * pci_iov_init - initialize the IOV capability
546  * @dev: the PCI device
547  *
548  * Returns 0 on success, or negative on failure.
549  */
550 int pci_iov_init(struct pci_dev *dev)
551 {
552 	int pos;
553 
554 	if (!pci_is_pcie(dev))
555 		return -ENODEV;
556 
557 	pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_SRIOV);
558 	if (pos)
559 		return sriov_init(dev, pos);
560 
561 	return -ENODEV;
562 }
563 
564 /**
565  * pci_iov_release - release resources used by the IOV capability
566  * @dev: the PCI device
567  */
568 void pci_iov_release(struct pci_dev *dev)
569 {
570 	if (dev->is_physfn)
571 		sriov_release(dev);
572 }
573 
574 /**
575  * pci_iov_resource_bar - get position of the SR-IOV BAR
576  * @dev: the PCI device
577  * @resno: the resource number
578  * @type: the BAR type to be filled in
579  *
580  * Returns position of the BAR encapsulated in the SR-IOV capability.
581  */
582 int pci_iov_resource_bar(struct pci_dev *dev, int resno,
583 			 enum pci_bar_type *type)
584 {
585 	if (resno < PCI_IOV_RESOURCES || resno > PCI_IOV_RESOURCE_END)
586 		return 0;
587 
588 	BUG_ON(!dev->is_physfn);
589 
590 	*type = pci_bar_unknown;
591 
592 	return dev->sriov->pos + PCI_SRIOV_BAR +
593 		4 * (resno - PCI_IOV_RESOURCES);
594 }
595 
596 /**
597  * pci_sriov_resource_alignment - get resource alignment for VF BAR
598  * @dev: the PCI device
599  * @resno: the resource number
600  *
601  * Returns the alignment of the VF BAR found in the SR-IOV capability.
602  * This is not the same as the resource size which is defined as
603  * the VF BAR size multiplied by the number of VFs.  The alignment
604  * is just the VF BAR size.
605  */
606 resource_size_t pci_sriov_resource_alignment(struct pci_dev *dev, int resno)
607 {
608 	struct resource tmp;
609 	enum pci_bar_type type;
610 	int reg = pci_iov_resource_bar(dev, resno, &type);
611 
612 	if (!reg)
613 		return 0;
614 
615 	 __pci_read_base(dev, type, &tmp, reg);
616 	return resource_alignment(&tmp);
617 }
618 
619 /**
620  * pci_restore_iov_state - restore the state of the IOV capability
621  * @dev: the PCI device
622  */
623 void pci_restore_iov_state(struct pci_dev *dev)
624 {
625 	if (dev->is_physfn)
626 		sriov_restore_state(dev);
627 }
628 
629 /**
630  * pci_iov_bus_range - find bus range used by Virtual Function
631  * @bus: the PCI bus
632  *
633  * Returns max number of buses (exclude current one) used by Virtual
634  * Functions.
635  */
636 int pci_iov_bus_range(struct pci_bus *bus)
637 {
638 	int max = 0;
639 	u8 busnr;
640 	struct pci_dev *dev;
641 
642 	list_for_each_entry(dev, &bus->devices, bus_list) {
643 		if (!dev->is_physfn)
644 			continue;
645 		busnr = virtfn_bus(dev, dev->sriov->total_VFs - 1);
646 		if (busnr > max)
647 			max = busnr;
648 	}
649 
650 	return max ? max - bus->number : 0;
651 }
652 
653 /**
654  * pci_enable_sriov - enable the SR-IOV capability
655  * @dev: the PCI device
656  * @nr_virtfn: number of virtual functions to enable
657  *
658  * Returns 0 on success, or negative on failure.
659  */
660 int pci_enable_sriov(struct pci_dev *dev, int nr_virtfn)
661 {
662 	might_sleep();
663 
664 	if (!dev->is_physfn)
665 		return -ENODEV;
666 
667 	return sriov_enable(dev, nr_virtfn);
668 }
669 EXPORT_SYMBOL_GPL(pci_enable_sriov);
670 
671 /**
672  * pci_disable_sriov - disable the SR-IOV capability
673  * @dev: the PCI device
674  */
675 void pci_disable_sriov(struct pci_dev *dev)
676 {
677 	might_sleep();
678 
679 	if (!dev->is_physfn)
680 		return;
681 
682 	sriov_disable(dev);
683 }
684 EXPORT_SYMBOL_GPL(pci_disable_sriov);
685 
686 /**
687  * pci_sriov_migration - notify SR-IOV core of Virtual Function Migration
688  * @dev: the PCI device
689  *
690  * Returns IRQ_HANDLED if the IRQ is handled, or IRQ_NONE if not.
691  *
692  * Physical Function driver is responsible to register IRQ handler using
693  * VF Migration Interrupt Message Number, and call this function when the
694  * interrupt is generated by the hardware.
695  */
696 irqreturn_t pci_sriov_migration(struct pci_dev *dev)
697 {
698 	if (!dev->is_physfn)
699 		return IRQ_NONE;
700 
701 	return sriov_migration(dev) ? IRQ_HANDLED : IRQ_NONE;
702 }
703 EXPORT_SYMBOL_GPL(pci_sriov_migration);
704 
705 /**
706  * pci_num_vf - return number of VFs associated with a PF device_release_driver
707  * @dev: the PCI device
708  *
709  * Returns number of VFs, or 0 if SR-IOV is not enabled.
710  */
711 int pci_num_vf(struct pci_dev *dev)
712 {
713 	if (!dev->is_physfn)
714 		return 0;
715 
716 	return dev->sriov->num_VFs;
717 }
718 EXPORT_SYMBOL_GPL(pci_num_vf);
719 
720 /**
721  * pci_vfs_assigned - returns number of VFs are assigned to a guest
722  * @dev: the PCI device
723  *
724  * Returns number of VFs belonging to this device that are assigned to a guest.
725  * If device is not a physical function returns -ENODEV.
726  */
727 int pci_vfs_assigned(struct pci_dev *dev)
728 {
729 	struct pci_dev *vfdev;
730 	unsigned int vfs_assigned = 0;
731 	unsigned short dev_id;
732 
733 	/* only search if we are a PF */
734 	if (!dev->is_physfn)
735 		return 0;
736 
737 	/*
738 	 * determine the device ID for the VFs, the vendor ID will be the
739 	 * same as the PF so there is no need to check for that one
740 	 */
741 	pci_read_config_word(dev, dev->sriov->pos + PCI_SRIOV_VF_DID, &dev_id);
742 
743 	/* loop through all the VFs to see if we own any that are assigned */
744 	vfdev = pci_get_device(dev->vendor, dev_id, NULL);
745 	while (vfdev) {
746 		/*
747 		 * It is considered assigned if it is a virtual function with
748 		 * our dev as the physical function and the assigned bit is set
749 		 */
750 		if (vfdev->is_virtfn && (vfdev->physfn == dev) &&
751 		    (vfdev->dev_flags & PCI_DEV_FLAGS_ASSIGNED))
752 			vfs_assigned++;
753 
754 		vfdev = pci_get_device(dev->vendor, dev_id, vfdev);
755 	}
756 
757 	return vfs_assigned;
758 }
759 EXPORT_SYMBOL_GPL(pci_vfs_assigned);
760 
761 /**
762  * pci_sriov_set_totalvfs -- reduce the TotalVFs available
763  * @dev: the PCI PF device
764  * @numvfs: number that should be used for TotalVFs supported
765  *
766  * Should be called from PF driver's probe routine with
767  * device's mutex held.
768  *
769  * Returns 0 if PF is an SRIOV-capable device and
770  * value of numvfs valid. If not a PF with VFS, return -EINVAL;
771  * if VFs already enabled, return -EBUSY.
772  */
773 int pci_sriov_set_totalvfs(struct pci_dev *dev, u16 numvfs)
774 {
775 	if (!dev->is_physfn || (numvfs > dev->sriov->total_VFs))
776 		return -EINVAL;
777 
778 	/* Shouldn't change if VFs already enabled */
779 	if (dev->sriov->ctrl & PCI_SRIOV_CTRL_VFE)
780 		return -EBUSY;
781 	else
782 		dev->sriov->driver_max_VFs = numvfs;
783 
784 	return 0;
785 }
786 EXPORT_SYMBOL_GPL(pci_sriov_set_totalvfs);
787 
788 /**
789  * pci_sriov_get_totalvfs -- get total VFs supported on this devic3
790  * @dev: the PCI PF device
791  *
792  * For a PCIe device with SRIOV support, return the PCIe
793  * SRIOV capability value of TotalVFs or the value of driver_max_VFs
794  * if the driver reduced it.  Otherwise, -EINVAL.
795  */
796 int pci_sriov_get_totalvfs(struct pci_dev *dev)
797 {
798 	if (!dev->is_physfn)
799 		return -EINVAL;
800 
801 	if (dev->sriov->driver_max_VFs)
802 		return dev->sriov->driver_max_VFs;
803 
804 	return dev->sriov->total_VFs;
805 }
806 EXPORT_SYMBOL_GPL(pci_sriov_get_totalvfs);
807