xref: /openbmc/linux/drivers/pci/iov.c (revision 7dd65feb)
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/mutex.h>
13 #include <linux/string.h>
14 #include <linux/delay.h>
15 #include "pci.h"
16 
17 #define VIRTFN_ID_LEN	16
18 
19 static inline u8 virtfn_bus(struct pci_dev *dev, int id)
20 {
21 	return dev->bus->number + ((dev->devfn + dev->sriov->offset +
22 				    dev->sriov->stride * id) >> 8);
23 }
24 
25 static inline u8 virtfn_devfn(struct pci_dev *dev, int id)
26 {
27 	return (dev->devfn + dev->sriov->offset +
28 		dev->sriov->stride * id) & 0xff;
29 }
30 
31 static struct pci_bus *virtfn_add_bus(struct pci_bus *bus, int busnr)
32 {
33 	int rc;
34 	struct pci_bus *child;
35 
36 	if (bus->number == busnr)
37 		return bus;
38 
39 	child = pci_find_bus(pci_domain_nr(bus), busnr);
40 	if (child)
41 		return child;
42 
43 	child = pci_add_new_bus(bus, NULL, busnr);
44 	if (!child)
45 		return NULL;
46 
47 	child->subordinate = busnr;
48 	child->dev.parent = bus->bridge;
49 	rc = pci_bus_add_child(child);
50 	if (rc) {
51 		pci_remove_bus(child);
52 		return NULL;
53 	}
54 
55 	return child;
56 }
57 
58 static void virtfn_remove_bus(struct pci_bus *bus, int busnr)
59 {
60 	struct pci_bus *child;
61 
62 	if (bus->number == busnr)
63 		return;
64 
65 	child = pci_find_bus(pci_domain_nr(bus), busnr);
66 	BUG_ON(!child);
67 
68 	if (list_empty(&child->devices))
69 		pci_remove_bus(child);
70 }
71 
72 static int virtfn_add(struct pci_dev *dev, int id, int reset)
73 {
74 	int i;
75 	int rc;
76 	u64 size;
77 	char buf[VIRTFN_ID_LEN];
78 	struct pci_dev *virtfn;
79 	struct resource *res;
80 	struct pci_sriov *iov = dev->sriov;
81 
82 	virtfn = alloc_pci_dev();
83 	if (!virtfn)
84 		return -ENOMEM;
85 
86 	mutex_lock(&iov->dev->sriov->lock);
87 	virtfn->bus = virtfn_add_bus(dev->bus, virtfn_bus(dev, id));
88 	if (!virtfn->bus) {
89 		kfree(virtfn);
90 		mutex_unlock(&iov->dev->sriov->lock);
91 		return -ENOMEM;
92 	}
93 	virtfn->devfn = virtfn_devfn(dev, id);
94 	virtfn->vendor = dev->vendor;
95 	pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_DID, &virtfn->device);
96 	pci_setup_device(virtfn);
97 	virtfn->dev.parent = dev->dev.parent;
98 
99 	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
100 		res = dev->resource + PCI_IOV_RESOURCES + i;
101 		if (!res->parent)
102 			continue;
103 		virtfn->resource[i].name = pci_name(virtfn);
104 		virtfn->resource[i].flags = res->flags;
105 		size = resource_size(res);
106 		do_div(size, iov->total);
107 		virtfn->resource[i].start = res->start + size * id;
108 		virtfn->resource[i].end = virtfn->resource[i].start + size - 1;
109 		rc = request_resource(res, &virtfn->resource[i]);
110 		BUG_ON(rc);
111 	}
112 
113 	if (reset)
114 		__pci_reset_function(virtfn);
115 
116 	pci_device_add(virtfn, virtfn->bus);
117 	mutex_unlock(&iov->dev->sriov->lock);
118 
119 	virtfn->physfn = pci_dev_get(dev);
120 	virtfn->is_virtfn = 1;
121 
122 	rc = pci_bus_add_device(virtfn);
123 	if (rc)
124 		goto failed1;
125 	sprintf(buf, "virtfn%u", id);
126 	rc = sysfs_create_link(&dev->dev.kobj, &virtfn->dev.kobj, buf);
127 	if (rc)
128 		goto failed1;
129 	rc = sysfs_create_link(&virtfn->dev.kobj, &dev->dev.kobj, "physfn");
130 	if (rc)
131 		goto failed2;
132 
133 	kobject_uevent(&virtfn->dev.kobj, KOBJ_CHANGE);
134 
135 	return 0;
136 
137 failed2:
138 	sysfs_remove_link(&dev->dev.kobj, buf);
139 failed1:
140 	pci_dev_put(dev);
141 	mutex_lock(&iov->dev->sriov->lock);
142 	pci_remove_bus_device(virtfn);
143 	virtfn_remove_bus(dev->bus, virtfn_bus(dev, id));
144 	mutex_unlock(&iov->dev->sriov->lock);
145 
146 	return rc;
147 }
148 
149 static void virtfn_remove(struct pci_dev *dev, int id, int reset)
150 {
151 	char buf[VIRTFN_ID_LEN];
152 	struct pci_bus *bus;
153 	struct pci_dev *virtfn;
154 	struct pci_sriov *iov = dev->sriov;
155 
156 	bus = pci_find_bus(pci_domain_nr(dev->bus), virtfn_bus(dev, id));
157 	if (!bus)
158 		return;
159 
160 	virtfn = pci_get_slot(bus, virtfn_devfn(dev, id));
161 	if (!virtfn)
162 		return;
163 
164 	pci_dev_put(virtfn);
165 
166 	if (reset) {
167 		device_release_driver(&virtfn->dev);
168 		__pci_reset_function(virtfn);
169 	}
170 
171 	sprintf(buf, "virtfn%u", id);
172 	sysfs_remove_link(&dev->dev.kobj, buf);
173 	sysfs_remove_link(&virtfn->dev.kobj, "physfn");
174 
175 	mutex_lock(&iov->dev->sriov->lock);
176 	pci_remove_bus_device(virtfn);
177 	virtfn_remove_bus(dev->bus, virtfn_bus(dev, id));
178 	mutex_unlock(&iov->dev->sriov->lock);
179 
180 	pci_dev_put(dev);
181 }
182 
183 static int sriov_migration(struct pci_dev *dev)
184 {
185 	u16 status;
186 	struct pci_sriov *iov = dev->sriov;
187 
188 	if (!iov->nr_virtfn)
189 		return 0;
190 
191 	if (!(iov->cap & PCI_SRIOV_CAP_VFM))
192 		return 0;
193 
194 	pci_read_config_word(dev, iov->pos + PCI_SRIOV_STATUS, &status);
195 	if (!(status & PCI_SRIOV_STATUS_VFM))
196 		return 0;
197 
198 	schedule_work(&iov->mtask);
199 
200 	return 1;
201 }
202 
203 static void sriov_migration_task(struct work_struct *work)
204 {
205 	int i;
206 	u8 state;
207 	u16 status;
208 	struct pci_sriov *iov = container_of(work, struct pci_sriov, mtask);
209 
210 	for (i = iov->initial; i < iov->nr_virtfn; i++) {
211 		state = readb(iov->mstate + i);
212 		if (state == PCI_SRIOV_VFM_MI) {
213 			writeb(PCI_SRIOV_VFM_AV, iov->mstate + i);
214 			state = readb(iov->mstate + i);
215 			if (state == PCI_SRIOV_VFM_AV)
216 				virtfn_add(iov->self, i, 1);
217 		} else if (state == PCI_SRIOV_VFM_MO) {
218 			virtfn_remove(iov->self, i, 1);
219 			writeb(PCI_SRIOV_VFM_UA, iov->mstate + i);
220 			state = readb(iov->mstate + i);
221 			if (state == PCI_SRIOV_VFM_AV)
222 				virtfn_add(iov->self, i, 0);
223 		}
224 	}
225 
226 	pci_read_config_word(iov->self, iov->pos + PCI_SRIOV_STATUS, &status);
227 	status &= ~PCI_SRIOV_STATUS_VFM;
228 	pci_write_config_word(iov->self, iov->pos + PCI_SRIOV_STATUS, status);
229 }
230 
231 static int sriov_enable_migration(struct pci_dev *dev, int nr_virtfn)
232 {
233 	int bir;
234 	u32 table;
235 	resource_size_t pa;
236 	struct pci_sriov *iov = dev->sriov;
237 
238 	if (nr_virtfn <= iov->initial)
239 		return 0;
240 
241 	pci_read_config_dword(dev, iov->pos + PCI_SRIOV_VFM, &table);
242 	bir = PCI_SRIOV_VFM_BIR(table);
243 	if (bir > PCI_STD_RESOURCE_END)
244 		return -EIO;
245 
246 	table = PCI_SRIOV_VFM_OFFSET(table);
247 	if (table + nr_virtfn > pci_resource_len(dev, bir))
248 		return -EIO;
249 
250 	pa = pci_resource_start(dev, bir) + table;
251 	iov->mstate = ioremap(pa, nr_virtfn);
252 	if (!iov->mstate)
253 		return -ENOMEM;
254 
255 	INIT_WORK(&iov->mtask, sriov_migration_task);
256 
257 	iov->ctrl |= PCI_SRIOV_CTRL_VFM | PCI_SRIOV_CTRL_INTR;
258 	pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
259 
260 	return 0;
261 }
262 
263 static void sriov_disable_migration(struct pci_dev *dev)
264 {
265 	struct pci_sriov *iov = dev->sriov;
266 
267 	iov->ctrl &= ~(PCI_SRIOV_CTRL_VFM | PCI_SRIOV_CTRL_INTR);
268 	pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
269 
270 	cancel_work_sync(&iov->mtask);
271 	iounmap(iov->mstate);
272 }
273 
274 static int sriov_enable(struct pci_dev *dev, int nr_virtfn)
275 {
276 	int rc;
277 	int i, j;
278 	int nres;
279 	u16 offset, stride, initial;
280 	struct resource *res;
281 	struct pci_dev *pdev;
282 	struct pci_sriov *iov = dev->sriov;
283 
284 	if (!nr_virtfn)
285 		return 0;
286 
287 	if (iov->nr_virtfn)
288 		return -EINVAL;
289 
290 	pci_read_config_word(dev, iov->pos + PCI_SRIOV_INITIAL_VF, &initial);
291 	if (initial > iov->total ||
292 	    (!(iov->cap & PCI_SRIOV_CAP_VFM) && (initial != iov->total)))
293 		return -EIO;
294 
295 	if (nr_virtfn < 0 || nr_virtfn > iov->total ||
296 	    (!(iov->cap & PCI_SRIOV_CAP_VFM) && (nr_virtfn > initial)))
297 		return -EINVAL;
298 
299 	pci_write_config_word(dev, iov->pos + PCI_SRIOV_NUM_VF, nr_virtfn);
300 	pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_OFFSET, &offset);
301 	pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_STRIDE, &stride);
302 	if (!offset || (nr_virtfn > 1 && !stride))
303 		return -EIO;
304 
305 	nres = 0;
306 	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
307 		res = dev->resource + PCI_IOV_RESOURCES + i;
308 		if (res->parent)
309 			nres++;
310 	}
311 	if (nres != iov->nres) {
312 		dev_err(&dev->dev, "not enough MMIO resources for SR-IOV\n");
313 		return -ENOMEM;
314 	}
315 
316 	iov->offset = offset;
317 	iov->stride = stride;
318 
319 	if (virtfn_bus(dev, nr_virtfn - 1) > dev->bus->subordinate) {
320 		dev_err(&dev->dev, "SR-IOV: bus number out of range\n");
321 		return -ENOMEM;
322 	}
323 
324 	if (iov->link != dev->devfn) {
325 		pdev = pci_get_slot(dev->bus, iov->link);
326 		if (!pdev)
327 			return -ENODEV;
328 
329 		pci_dev_put(pdev);
330 
331 		if (!pdev->is_physfn)
332 			return -ENODEV;
333 
334 		rc = sysfs_create_link(&dev->dev.kobj,
335 					&pdev->dev.kobj, "dep_link");
336 		if (rc)
337 			return rc;
338 	}
339 
340 	iov->ctrl |= PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE;
341 	pci_block_user_cfg_access(dev);
342 	pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
343 	msleep(100);
344 	pci_unblock_user_cfg_access(dev);
345 
346 	iov->initial = initial;
347 	if (nr_virtfn < initial)
348 		initial = nr_virtfn;
349 
350 	for (i = 0; i < initial; i++) {
351 		rc = virtfn_add(dev, i, 0);
352 		if (rc)
353 			goto failed;
354 	}
355 
356 	if (iov->cap & PCI_SRIOV_CAP_VFM) {
357 		rc = sriov_enable_migration(dev, nr_virtfn);
358 		if (rc)
359 			goto failed;
360 	}
361 
362 	kobject_uevent(&dev->dev.kobj, KOBJ_CHANGE);
363 	iov->nr_virtfn = nr_virtfn;
364 
365 	return 0;
366 
367 failed:
368 	for (j = 0; j < i; j++)
369 		virtfn_remove(dev, j, 0);
370 
371 	iov->ctrl &= ~(PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE);
372 	pci_block_user_cfg_access(dev);
373 	pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
374 	ssleep(1);
375 	pci_unblock_user_cfg_access(dev);
376 
377 	if (iov->link != dev->devfn)
378 		sysfs_remove_link(&dev->dev.kobj, "dep_link");
379 
380 	return rc;
381 }
382 
383 static void sriov_disable(struct pci_dev *dev)
384 {
385 	int i;
386 	struct pci_sriov *iov = dev->sriov;
387 
388 	if (!iov->nr_virtfn)
389 		return;
390 
391 	if (iov->cap & PCI_SRIOV_CAP_VFM)
392 		sriov_disable_migration(dev);
393 
394 	for (i = 0; i < iov->nr_virtfn; i++)
395 		virtfn_remove(dev, i, 0);
396 
397 	iov->ctrl &= ~(PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE);
398 	pci_block_user_cfg_access(dev);
399 	pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
400 	ssleep(1);
401 	pci_unblock_user_cfg_access(dev);
402 
403 	if (iov->link != dev->devfn)
404 		sysfs_remove_link(&dev->dev.kobj, "dep_link");
405 
406 	iov->nr_virtfn = 0;
407 }
408 
409 static int sriov_init(struct pci_dev *dev, int pos)
410 {
411 	int i;
412 	int rc;
413 	int nres;
414 	u32 pgsz;
415 	u16 ctrl, total, offset, stride;
416 	struct pci_sriov *iov;
417 	struct resource *res;
418 	struct pci_dev *pdev;
419 
420 	if (dev->pcie_type != PCI_EXP_TYPE_RC_END &&
421 	    dev->pcie_type != PCI_EXP_TYPE_ENDPOINT)
422 		return -ENODEV;
423 
424 	pci_read_config_word(dev, pos + PCI_SRIOV_CTRL, &ctrl);
425 	if (ctrl & PCI_SRIOV_CTRL_VFE) {
426 		pci_write_config_word(dev, pos + PCI_SRIOV_CTRL, 0);
427 		ssleep(1);
428 	}
429 
430 	pci_read_config_word(dev, pos + PCI_SRIOV_TOTAL_VF, &total);
431 	if (!total)
432 		return 0;
433 
434 	ctrl = 0;
435 	list_for_each_entry(pdev, &dev->bus->devices, bus_list)
436 		if (pdev->is_physfn)
437 			goto found;
438 
439 	pdev = NULL;
440 	if (pci_ari_enabled(dev->bus))
441 		ctrl |= PCI_SRIOV_CTRL_ARI;
442 
443 found:
444 	pci_write_config_word(dev, pos + PCI_SRIOV_CTRL, ctrl);
445 	pci_write_config_word(dev, pos + PCI_SRIOV_NUM_VF, total);
446 	pci_read_config_word(dev, pos + PCI_SRIOV_VF_OFFSET, &offset);
447 	pci_read_config_word(dev, pos + PCI_SRIOV_VF_STRIDE, &stride);
448 	if (!offset || (total > 1 && !stride))
449 		return -EIO;
450 
451 	pci_read_config_dword(dev, pos + PCI_SRIOV_SUP_PGSIZE, &pgsz);
452 	i = PAGE_SHIFT > 12 ? PAGE_SHIFT - 12 : 0;
453 	pgsz &= ~((1 << i) - 1);
454 	if (!pgsz)
455 		return -EIO;
456 
457 	pgsz &= ~(pgsz - 1);
458 	pci_write_config_dword(dev, pos + PCI_SRIOV_SYS_PGSIZE, pgsz);
459 
460 	nres = 0;
461 	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
462 		res = dev->resource + PCI_IOV_RESOURCES + i;
463 		i += __pci_read_base(dev, pci_bar_unknown, res,
464 				     pos + PCI_SRIOV_BAR + i * 4);
465 		if (!res->flags)
466 			continue;
467 		if (resource_size(res) & (PAGE_SIZE - 1)) {
468 			rc = -EIO;
469 			goto failed;
470 		}
471 		res->end = res->start + resource_size(res) * total - 1;
472 		nres++;
473 	}
474 
475 	iov = kzalloc(sizeof(*iov), GFP_KERNEL);
476 	if (!iov) {
477 		rc = -ENOMEM;
478 		goto failed;
479 	}
480 
481 	iov->pos = pos;
482 	iov->nres = nres;
483 	iov->ctrl = ctrl;
484 	iov->total = total;
485 	iov->offset = offset;
486 	iov->stride = stride;
487 	iov->pgsz = pgsz;
488 	iov->self = dev;
489 	pci_read_config_dword(dev, pos + PCI_SRIOV_CAP, &iov->cap);
490 	pci_read_config_byte(dev, pos + PCI_SRIOV_FUNC_LINK, &iov->link);
491 	if (dev->pcie_type == PCI_EXP_TYPE_RC_END)
492 		iov->link = PCI_DEVFN(PCI_SLOT(dev->devfn), iov->link);
493 
494 	if (pdev)
495 		iov->dev = pci_dev_get(pdev);
496 	else
497 		iov->dev = dev;
498 
499 	mutex_init(&iov->lock);
500 
501 	dev->sriov = iov;
502 	dev->is_physfn = 1;
503 
504 	return 0;
505 
506 failed:
507 	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
508 		res = dev->resource + PCI_IOV_RESOURCES + i;
509 		res->flags = 0;
510 	}
511 
512 	return rc;
513 }
514 
515 static void sriov_release(struct pci_dev *dev)
516 {
517 	BUG_ON(dev->sriov->nr_virtfn);
518 
519 	if (dev != dev->sriov->dev)
520 		pci_dev_put(dev->sriov->dev);
521 
522 	mutex_destroy(&dev->sriov->lock);
523 
524 	kfree(dev->sriov);
525 	dev->sriov = NULL;
526 }
527 
528 static void sriov_restore_state(struct pci_dev *dev)
529 {
530 	int i;
531 	u16 ctrl;
532 	struct pci_sriov *iov = dev->sriov;
533 
534 	pci_read_config_word(dev, iov->pos + PCI_SRIOV_CTRL, &ctrl);
535 	if (ctrl & PCI_SRIOV_CTRL_VFE)
536 		return;
537 
538 	for (i = PCI_IOV_RESOURCES; i <= PCI_IOV_RESOURCE_END; i++)
539 		pci_update_resource(dev, i);
540 
541 	pci_write_config_dword(dev, iov->pos + PCI_SRIOV_SYS_PGSIZE, iov->pgsz);
542 	pci_write_config_word(dev, iov->pos + PCI_SRIOV_NUM_VF, iov->nr_virtfn);
543 	pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
544 	if (iov->ctrl & PCI_SRIOV_CTRL_VFE)
545 		msleep(100);
546 }
547 
548 /**
549  * pci_iov_init - initialize the IOV capability
550  * @dev: the PCI device
551  *
552  * Returns 0 on success, or negative on failure.
553  */
554 int pci_iov_init(struct pci_dev *dev)
555 {
556 	int pos;
557 
558 	if (!pci_is_pcie(dev))
559 		return -ENODEV;
560 
561 	pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_SRIOV);
562 	if (pos)
563 		return sriov_init(dev, pos);
564 
565 	return -ENODEV;
566 }
567 
568 /**
569  * pci_iov_release - release resources used by the IOV capability
570  * @dev: the PCI device
571  */
572 void pci_iov_release(struct pci_dev *dev)
573 {
574 	if (dev->is_physfn)
575 		sriov_release(dev);
576 }
577 
578 /**
579  * pci_iov_resource_bar - get position of the SR-IOV BAR
580  * @dev: the PCI device
581  * @resno: the resource number
582  * @type: the BAR type to be filled in
583  *
584  * Returns position of the BAR encapsulated in the SR-IOV capability.
585  */
586 int pci_iov_resource_bar(struct pci_dev *dev, int resno,
587 			 enum pci_bar_type *type)
588 {
589 	if (resno < PCI_IOV_RESOURCES || resno > PCI_IOV_RESOURCE_END)
590 		return 0;
591 
592 	BUG_ON(!dev->is_physfn);
593 
594 	*type = pci_bar_unknown;
595 
596 	return dev->sriov->pos + PCI_SRIOV_BAR +
597 		4 * (resno - PCI_IOV_RESOURCES);
598 }
599 
600 /**
601  * pci_sriov_resource_alignment - get resource alignment for VF BAR
602  * @dev: the PCI device
603  * @resno: the resource number
604  *
605  * Returns the alignment of the VF BAR found in the SR-IOV capability.
606  * This is not the same as the resource size which is defined as
607  * the VF BAR size multiplied by the number of VFs.  The alignment
608  * is just the VF BAR size.
609  */
610 int pci_sriov_resource_alignment(struct pci_dev *dev, int resno)
611 {
612 	struct resource tmp;
613 	enum pci_bar_type type;
614 	int reg = pci_iov_resource_bar(dev, resno, &type);
615 
616 	if (!reg)
617 		return 0;
618 
619 	 __pci_read_base(dev, type, &tmp, reg);
620 	return resource_alignment(&tmp);
621 }
622 
623 /**
624  * pci_restore_iov_state - restore the state of the IOV capability
625  * @dev: the PCI device
626  */
627 void pci_restore_iov_state(struct pci_dev *dev)
628 {
629 	if (dev->is_physfn)
630 		sriov_restore_state(dev);
631 }
632 
633 /**
634  * pci_iov_bus_range - find bus range used by Virtual Function
635  * @bus: the PCI bus
636  *
637  * Returns max number of buses (exclude current one) used by Virtual
638  * Functions.
639  */
640 int pci_iov_bus_range(struct pci_bus *bus)
641 {
642 	int max = 0;
643 	u8 busnr;
644 	struct pci_dev *dev;
645 
646 	list_for_each_entry(dev, &bus->devices, bus_list) {
647 		if (!dev->is_physfn)
648 			continue;
649 		busnr = virtfn_bus(dev, dev->sriov->total - 1);
650 		if (busnr > max)
651 			max = busnr;
652 	}
653 
654 	return max ? max - bus->number : 0;
655 }
656 
657 /**
658  * pci_enable_sriov - enable the SR-IOV capability
659  * @dev: the PCI device
660  * @nr_virtfn: number of virtual functions to enable
661  *
662  * Returns 0 on success, or negative on failure.
663  */
664 int pci_enable_sriov(struct pci_dev *dev, int nr_virtfn)
665 {
666 	might_sleep();
667 
668 	if (!dev->is_physfn)
669 		return -ENODEV;
670 
671 	return sriov_enable(dev, nr_virtfn);
672 }
673 EXPORT_SYMBOL_GPL(pci_enable_sriov);
674 
675 /**
676  * pci_disable_sriov - disable the SR-IOV capability
677  * @dev: the PCI device
678  */
679 void pci_disable_sriov(struct pci_dev *dev)
680 {
681 	might_sleep();
682 
683 	if (!dev->is_physfn)
684 		return;
685 
686 	sriov_disable(dev);
687 }
688 EXPORT_SYMBOL_GPL(pci_disable_sriov);
689 
690 /**
691  * pci_sriov_migration - notify SR-IOV core of Virtual Function Migration
692  * @dev: the PCI device
693  *
694  * Returns IRQ_HANDLED if the IRQ is handled, or IRQ_NONE if not.
695  *
696  * Physical Function driver is responsible to register IRQ handler using
697  * VF Migration Interrupt Message Number, and call this function when the
698  * interrupt is generated by the hardware.
699  */
700 irqreturn_t pci_sriov_migration(struct pci_dev *dev)
701 {
702 	if (!dev->is_physfn)
703 		return IRQ_NONE;
704 
705 	return sriov_migration(dev) ? IRQ_HANDLED : IRQ_NONE;
706 }
707 EXPORT_SYMBOL_GPL(pci_sriov_migration);
708 
709 static int ats_alloc_one(struct pci_dev *dev, int ps)
710 {
711 	int pos;
712 	u16 cap;
713 	struct pci_ats *ats;
714 
715 	pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ATS);
716 	if (!pos)
717 		return -ENODEV;
718 
719 	ats = kzalloc(sizeof(*ats), GFP_KERNEL);
720 	if (!ats)
721 		return -ENOMEM;
722 
723 	ats->pos = pos;
724 	ats->stu = ps;
725 	pci_read_config_word(dev, pos + PCI_ATS_CAP, &cap);
726 	ats->qdep = PCI_ATS_CAP_QDEP(cap) ? PCI_ATS_CAP_QDEP(cap) :
727 					    PCI_ATS_MAX_QDEP;
728 	dev->ats = ats;
729 
730 	return 0;
731 }
732 
733 static void ats_free_one(struct pci_dev *dev)
734 {
735 	kfree(dev->ats);
736 	dev->ats = NULL;
737 }
738 
739 /**
740  * pci_enable_ats - enable the ATS capability
741  * @dev: the PCI device
742  * @ps: the IOMMU page shift
743  *
744  * Returns 0 on success, or negative on failure.
745  */
746 int pci_enable_ats(struct pci_dev *dev, int ps)
747 {
748 	int rc;
749 	u16 ctrl;
750 
751 	BUG_ON(dev->ats && dev->ats->is_enabled);
752 
753 	if (ps < PCI_ATS_MIN_STU)
754 		return -EINVAL;
755 
756 	if (dev->is_physfn || dev->is_virtfn) {
757 		struct pci_dev *pdev = dev->is_physfn ? dev : dev->physfn;
758 
759 		mutex_lock(&pdev->sriov->lock);
760 		if (pdev->ats)
761 			rc = pdev->ats->stu == ps ? 0 : -EINVAL;
762 		else
763 			rc = ats_alloc_one(pdev, ps);
764 
765 		if (!rc)
766 			pdev->ats->ref_cnt++;
767 		mutex_unlock(&pdev->sriov->lock);
768 		if (rc)
769 			return rc;
770 	}
771 
772 	if (!dev->is_physfn) {
773 		rc = ats_alloc_one(dev, ps);
774 		if (rc)
775 			return rc;
776 	}
777 
778 	ctrl = PCI_ATS_CTRL_ENABLE;
779 	if (!dev->is_virtfn)
780 		ctrl |= PCI_ATS_CTRL_STU(ps - PCI_ATS_MIN_STU);
781 	pci_write_config_word(dev, dev->ats->pos + PCI_ATS_CTRL, ctrl);
782 
783 	dev->ats->is_enabled = 1;
784 
785 	return 0;
786 }
787 
788 /**
789  * pci_disable_ats - disable the ATS capability
790  * @dev: the PCI device
791  */
792 void pci_disable_ats(struct pci_dev *dev)
793 {
794 	u16 ctrl;
795 
796 	BUG_ON(!dev->ats || !dev->ats->is_enabled);
797 
798 	pci_read_config_word(dev, dev->ats->pos + PCI_ATS_CTRL, &ctrl);
799 	ctrl &= ~PCI_ATS_CTRL_ENABLE;
800 	pci_write_config_word(dev, dev->ats->pos + PCI_ATS_CTRL, ctrl);
801 
802 	dev->ats->is_enabled = 0;
803 
804 	if (dev->is_physfn || dev->is_virtfn) {
805 		struct pci_dev *pdev = dev->is_physfn ? dev : dev->physfn;
806 
807 		mutex_lock(&pdev->sriov->lock);
808 		pdev->ats->ref_cnt--;
809 		if (!pdev->ats->ref_cnt)
810 			ats_free_one(pdev);
811 		mutex_unlock(&pdev->sriov->lock);
812 	}
813 
814 	if (!dev->is_physfn)
815 		ats_free_one(dev);
816 }
817 
818 /**
819  * pci_ats_queue_depth - query the ATS Invalidate Queue Depth
820  * @dev: the PCI device
821  *
822  * Returns the queue depth on success, or negative on failure.
823  *
824  * The ATS spec uses 0 in the Invalidate Queue Depth field to
825  * indicate that the function can accept 32 Invalidate Request.
826  * But here we use the `real' values (i.e. 1~32) for the Queue
827  * Depth; and 0 indicates the function shares the Queue with
828  * other functions (doesn't exclusively own a Queue).
829  */
830 int pci_ats_queue_depth(struct pci_dev *dev)
831 {
832 	int pos;
833 	u16 cap;
834 
835 	if (dev->is_virtfn)
836 		return 0;
837 
838 	if (dev->ats)
839 		return dev->ats->qdep;
840 
841 	pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ATS);
842 	if (!pos)
843 		return -ENODEV;
844 
845 	pci_read_config_word(dev, pos + PCI_ATS_CAP, &cap);
846 
847 	return PCI_ATS_CAP_QDEP(cap) ? PCI_ATS_CAP_QDEP(cap) :
848 				       PCI_ATS_MAX_QDEP;
849 }
850