xref: /linux/drivers/pci/iov.c (revision e9f0878c4b2004ac19581274c1ae4c61ae3ca70e)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * PCI Express I/O Virtualization (IOV) support
4  *   Single Root IOV 1.0
5  *   Address Translation Service 1.0
6  *
7  * Copyright (C) 2009 Intel Corporation, Yu Zhao <yu.zhao@intel.com>
8  */
9 
10 #include <linux/pci.h>
11 #include <linux/slab.h>
12 #include <linux/mutex.h>
13 #include <linux/export.h>
14 #include <linux/string.h>
15 #include <linux/delay.h>
16 #include <linux/pci-ats.h>
17 #include "pci.h"
18 
19 #define VIRTFN_ID_LEN	16
20 
21 int pci_iov_virtfn_bus(struct pci_dev *dev, int vf_id)
22 {
23 	if (!dev->is_physfn)
24 		return -EINVAL;
25 	return dev->bus->number + ((dev->devfn + dev->sriov->offset +
26 				    dev->sriov->stride * vf_id) >> 8);
27 }
28 
29 int pci_iov_virtfn_devfn(struct pci_dev *dev, int vf_id)
30 {
31 	if (!dev->is_physfn)
32 		return -EINVAL;
33 	return (dev->devfn + dev->sriov->offset +
34 		dev->sriov->stride * vf_id) & 0xff;
35 }
36 
37 /*
38  * Per SR-IOV spec sec 3.3.10 and 3.3.11, First VF Offset and VF Stride may
39  * change when NumVFs changes.
40  *
41  * Update iov->offset and iov->stride when NumVFs is written.
42  */
43 static inline void pci_iov_set_numvfs(struct pci_dev *dev, int nr_virtfn)
44 {
45 	struct pci_sriov *iov = dev->sriov;
46 
47 	pci_write_config_word(dev, iov->pos + PCI_SRIOV_NUM_VF, nr_virtfn);
48 	pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_OFFSET, &iov->offset);
49 	pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_STRIDE, &iov->stride);
50 }
51 
52 /*
53  * The PF consumes one bus number.  NumVFs, First VF Offset, and VF Stride
54  * determine how many additional bus numbers will be consumed by VFs.
55  *
56  * Iterate over all valid NumVFs, validate offset and stride, and calculate
57  * the maximum number of bus numbers that could ever be required.
58  */
59 static int compute_max_vf_buses(struct pci_dev *dev)
60 {
61 	struct pci_sriov *iov = dev->sriov;
62 	int nr_virtfn, busnr, rc = 0;
63 
64 	for (nr_virtfn = iov->total_VFs; nr_virtfn; nr_virtfn--) {
65 		pci_iov_set_numvfs(dev, nr_virtfn);
66 		if (!iov->offset || (nr_virtfn > 1 && !iov->stride)) {
67 			rc = -EIO;
68 			goto out;
69 		}
70 
71 		busnr = pci_iov_virtfn_bus(dev, nr_virtfn - 1);
72 		if (busnr > iov->max_VF_buses)
73 			iov->max_VF_buses = busnr;
74 	}
75 
76 out:
77 	pci_iov_set_numvfs(dev, 0);
78 	return rc;
79 }
80 
81 static struct pci_bus *virtfn_add_bus(struct pci_bus *bus, int busnr)
82 {
83 	struct pci_bus *child;
84 
85 	if (bus->number == busnr)
86 		return bus;
87 
88 	child = pci_find_bus(pci_domain_nr(bus), busnr);
89 	if (child)
90 		return child;
91 
92 	child = pci_add_new_bus(bus, NULL, busnr);
93 	if (!child)
94 		return NULL;
95 
96 	pci_bus_insert_busn_res(child, busnr, busnr);
97 
98 	return child;
99 }
100 
101 static void virtfn_remove_bus(struct pci_bus *physbus, struct pci_bus *virtbus)
102 {
103 	if (physbus != virtbus && list_empty(&virtbus->devices))
104 		pci_remove_bus(virtbus);
105 }
106 
107 resource_size_t pci_iov_resource_size(struct pci_dev *dev, int resno)
108 {
109 	if (!dev->is_physfn)
110 		return 0;
111 
112 	return dev->sriov->barsz[resno - PCI_IOV_RESOURCES];
113 }
114 
115 static void pci_read_vf_config_common(struct pci_dev *virtfn)
116 {
117 	struct pci_dev *physfn = virtfn->physfn;
118 
119 	/*
120 	 * Some config registers are the same across all associated VFs.
121 	 * Read them once from VF0 so we can skip reading them from the
122 	 * other VFs.
123 	 *
124 	 * PCIe r4.0, sec 9.3.4.1, technically doesn't require all VFs to
125 	 * have the same Revision ID and Subsystem ID, but we assume they
126 	 * do.
127 	 */
128 	pci_read_config_dword(virtfn, PCI_CLASS_REVISION,
129 			      &physfn->sriov->class);
130 	pci_read_config_byte(virtfn, PCI_HEADER_TYPE,
131 			     &physfn->sriov->hdr_type);
132 	pci_read_config_word(virtfn, PCI_SUBSYSTEM_VENDOR_ID,
133 			     &physfn->sriov->subsystem_vendor);
134 	pci_read_config_word(virtfn, PCI_SUBSYSTEM_ID,
135 			     &physfn->sriov->subsystem_device);
136 }
137 
138 int pci_iov_add_virtfn(struct pci_dev *dev, int id)
139 {
140 	int i;
141 	int rc = -ENOMEM;
142 	u64 size;
143 	char buf[VIRTFN_ID_LEN];
144 	struct pci_dev *virtfn;
145 	struct resource *res;
146 	struct pci_sriov *iov = dev->sriov;
147 	struct pci_bus *bus;
148 
149 	bus = virtfn_add_bus(dev->bus, pci_iov_virtfn_bus(dev, id));
150 	if (!bus)
151 		goto failed;
152 
153 	virtfn = pci_alloc_dev(bus);
154 	if (!virtfn)
155 		goto failed0;
156 
157 	virtfn->devfn = pci_iov_virtfn_devfn(dev, id);
158 	virtfn->vendor = dev->vendor;
159 	virtfn->device = iov->vf_device;
160 	virtfn->is_virtfn = 1;
161 	virtfn->physfn = pci_dev_get(dev);
162 
163 	if (id == 0)
164 		pci_read_vf_config_common(virtfn);
165 
166 	rc = pci_setup_device(virtfn);
167 	if (rc)
168 		goto failed1;
169 
170 	virtfn->dev.parent = dev->dev.parent;
171 	virtfn->multifunction = 0;
172 
173 	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
174 		res = &dev->resource[i + PCI_IOV_RESOURCES];
175 		if (!res->parent)
176 			continue;
177 		virtfn->resource[i].name = pci_name(virtfn);
178 		virtfn->resource[i].flags = res->flags;
179 		size = pci_iov_resource_size(dev, i + PCI_IOV_RESOURCES);
180 		virtfn->resource[i].start = res->start + size * id;
181 		virtfn->resource[i].end = virtfn->resource[i].start + size - 1;
182 		rc = request_resource(res, &virtfn->resource[i]);
183 		BUG_ON(rc);
184 	}
185 
186 	pci_device_add(virtfn, virtfn->bus);
187 
188 	sprintf(buf, "virtfn%u", id);
189 	rc = sysfs_create_link(&dev->dev.kobj, &virtfn->dev.kobj, buf);
190 	if (rc)
191 		goto failed2;
192 	rc = sysfs_create_link(&virtfn->dev.kobj, &dev->dev.kobj, "physfn");
193 	if (rc)
194 		goto failed3;
195 
196 	kobject_uevent(&virtfn->dev.kobj, KOBJ_CHANGE);
197 
198 	pci_bus_add_device(virtfn);
199 
200 	return 0;
201 
202 failed3:
203 	sysfs_remove_link(&dev->dev.kobj, buf);
204 failed2:
205 	pci_stop_and_remove_bus_device(virtfn);
206 failed1:
207 	pci_dev_put(dev);
208 failed0:
209 	virtfn_remove_bus(dev->bus, bus);
210 failed:
211 
212 	return rc;
213 }
214 
215 void pci_iov_remove_virtfn(struct pci_dev *dev, int id)
216 {
217 	char buf[VIRTFN_ID_LEN];
218 	struct pci_dev *virtfn;
219 
220 	virtfn = pci_get_domain_bus_and_slot(pci_domain_nr(dev->bus),
221 					     pci_iov_virtfn_bus(dev, id),
222 					     pci_iov_virtfn_devfn(dev, id));
223 	if (!virtfn)
224 		return;
225 
226 	sprintf(buf, "virtfn%u", id);
227 	sysfs_remove_link(&dev->dev.kobj, buf);
228 	/*
229 	 * pci_stop_dev() could have been called for this virtfn already,
230 	 * so the directory for the virtfn may have been removed before.
231 	 * Double check to avoid spurious sysfs warnings.
232 	 */
233 	if (virtfn->dev.kobj.sd)
234 		sysfs_remove_link(&virtfn->dev.kobj, "physfn");
235 
236 	pci_stop_and_remove_bus_device(virtfn);
237 	virtfn_remove_bus(dev->bus, virtfn->bus);
238 
239 	/* balance pci_get_domain_bus_and_slot() */
240 	pci_dev_put(virtfn);
241 	pci_dev_put(dev);
242 }
243 
244 int __weak pcibios_sriov_enable(struct pci_dev *pdev, u16 num_vfs)
245 {
246 	return 0;
247 }
248 
249 int __weak pcibios_sriov_disable(struct pci_dev *pdev)
250 {
251 	return 0;
252 }
253 
254 static int sriov_enable(struct pci_dev *dev, int nr_virtfn)
255 {
256 	int rc;
257 	int i;
258 	int nres;
259 	u16 initial;
260 	struct resource *res;
261 	struct pci_dev *pdev;
262 	struct pci_sriov *iov = dev->sriov;
263 	int bars = 0;
264 	int bus;
265 
266 	if (!nr_virtfn)
267 		return 0;
268 
269 	if (iov->num_VFs)
270 		return -EINVAL;
271 
272 	pci_read_config_word(dev, iov->pos + PCI_SRIOV_INITIAL_VF, &initial);
273 	if (initial > iov->total_VFs ||
274 	    (!(iov->cap & PCI_SRIOV_CAP_VFM) && (initial != iov->total_VFs)))
275 		return -EIO;
276 
277 	if (nr_virtfn < 0 || nr_virtfn > iov->total_VFs ||
278 	    (!(iov->cap & PCI_SRIOV_CAP_VFM) && (nr_virtfn > initial)))
279 		return -EINVAL;
280 
281 	nres = 0;
282 	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
283 		bars |= (1 << (i + PCI_IOV_RESOURCES));
284 		res = &dev->resource[i + PCI_IOV_RESOURCES];
285 		if (res->parent)
286 			nres++;
287 	}
288 	if (nres != iov->nres) {
289 		pci_err(dev, "not enough MMIO resources for SR-IOV\n");
290 		return -ENOMEM;
291 	}
292 
293 	bus = pci_iov_virtfn_bus(dev, nr_virtfn - 1);
294 	if (bus > dev->bus->busn_res.end) {
295 		pci_err(dev, "can't enable %d VFs (bus %02x out of range of %pR)\n",
296 			nr_virtfn, bus, &dev->bus->busn_res);
297 		return -ENOMEM;
298 	}
299 
300 	if (pci_enable_resources(dev, bars)) {
301 		pci_err(dev, "SR-IOV: IOV BARS not allocated\n");
302 		return -ENOMEM;
303 	}
304 
305 	if (iov->link != dev->devfn) {
306 		pdev = pci_get_slot(dev->bus, iov->link);
307 		if (!pdev)
308 			return -ENODEV;
309 
310 		if (!pdev->is_physfn) {
311 			pci_dev_put(pdev);
312 			return -ENOSYS;
313 		}
314 
315 		rc = sysfs_create_link(&dev->dev.kobj,
316 					&pdev->dev.kobj, "dep_link");
317 		pci_dev_put(pdev);
318 		if (rc)
319 			return rc;
320 	}
321 
322 	iov->initial_VFs = initial;
323 	if (nr_virtfn < initial)
324 		initial = nr_virtfn;
325 
326 	rc = pcibios_sriov_enable(dev, initial);
327 	if (rc) {
328 		pci_err(dev, "failure %d from pcibios_sriov_enable()\n", rc);
329 		goto err_pcibios;
330 	}
331 
332 	pci_iov_set_numvfs(dev, nr_virtfn);
333 	iov->ctrl |= PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE;
334 	pci_cfg_access_lock(dev);
335 	pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
336 	msleep(100);
337 	pci_cfg_access_unlock(dev);
338 
339 	for (i = 0; i < initial; i++) {
340 		rc = pci_iov_add_virtfn(dev, i);
341 		if (rc)
342 			goto failed;
343 	}
344 
345 	kobject_uevent(&dev->dev.kobj, KOBJ_CHANGE);
346 	iov->num_VFs = nr_virtfn;
347 
348 	return 0;
349 
350 failed:
351 	while (i--)
352 		pci_iov_remove_virtfn(dev, i);
353 
354 err_pcibios:
355 	iov->ctrl &= ~(PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE);
356 	pci_cfg_access_lock(dev);
357 	pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
358 	ssleep(1);
359 	pci_cfg_access_unlock(dev);
360 
361 	pcibios_sriov_disable(dev);
362 
363 	if (iov->link != dev->devfn)
364 		sysfs_remove_link(&dev->dev.kobj, "dep_link");
365 
366 	pci_iov_set_numvfs(dev, 0);
367 	return rc;
368 }
369 
370 static void sriov_disable(struct pci_dev *dev)
371 {
372 	int i;
373 	struct pci_sriov *iov = dev->sriov;
374 
375 	if (!iov->num_VFs)
376 		return;
377 
378 	for (i = 0; i < iov->num_VFs; i++)
379 		pci_iov_remove_virtfn(dev, i);
380 
381 	iov->ctrl &= ~(PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE);
382 	pci_cfg_access_lock(dev);
383 	pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
384 	ssleep(1);
385 	pci_cfg_access_unlock(dev);
386 
387 	pcibios_sriov_disable(dev);
388 
389 	if (iov->link != dev->devfn)
390 		sysfs_remove_link(&dev->dev.kobj, "dep_link");
391 
392 	iov->num_VFs = 0;
393 	pci_iov_set_numvfs(dev, 0);
394 }
395 
396 static int sriov_init(struct pci_dev *dev, int pos)
397 {
398 	int i, bar64;
399 	int rc;
400 	int nres;
401 	u32 pgsz;
402 	u16 ctrl, total;
403 	struct pci_sriov *iov;
404 	struct resource *res;
405 	struct pci_dev *pdev;
406 
407 	pci_read_config_word(dev, pos + PCI_SRIOV_CTRL, &ctrl);
408 	if (ctrl & PCI_SRIOV_CTRL_VFE) {
409 		pci_write_config_word(dev, pos + PCI_SRIOV_CTRL, 0);
410 		ssleep(1);
411 	}
412 
413 	ctrl = 0;
414 	list_for_each_entry(pdev, &dev->bus->devices, bus_list)
415 		if (pdev->is_physfn)
416 			goto found;
417 
418 	pdev = NULL;
419 	if (pci_ari_enabled(dev->bus))
420 		ctrl |= PCI_SRIOV_CTRL_ARI;
421 
422 found:
423 	pci_write_config_word(dev, pos + PCI_SRIOV_CTRL, ctrl);
424 
425 	pci_read_config_word(dev, pos + PCI_SRIOV_TOTAL_VF, &total);
426 	if (!total)
427 		return 0;
428 
429 	pci_read_config_dword(dev, pos + PCI_SRIOV_SUP_PGSIZE, &pgsz);
430 	i = PAGE_SHIFT > 12 ? PAGE_SHIFT - 12 : 0;
431 	pgsz &= ~((1 << i) - 1);
432 	if (!pgsz)
433 		return -EIO;
434 
435 	pgsz &= ~(pgsz - 1);
436 	pci_write_config_dword(dev, pos + PCI_SRIOV_SYS_PGSIZE, pgsz);
437 
438 	iov = kzalloc(sizeof(*iov), GFP_KERNEL);
439 	if (!iov)
440 		return -ENOMEM;
441 
442 	nres = 0;
443 	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
444 		res = &dev->resource[i + PCI_IOV_RESOURCES];
445 		/*
446 		 * If it is already FIXED, don't change it, something
447 		 * (perhaps EA or header fixups) wants it this way.
448 		 */
449 		if (res->flags & IORESOURCE_PCI_FIXED)
450 			bar64 = (res->flags & IORESOURCE_MEM_64) ? 1 : 0;
451 		else
452 			bar64 = __pci_read_base(dev, pci_bar_unknown, res,
453 						pos + PCI_SRIOV_BAR + i * 4);
454 		if (!res->flags)
455 			continue;
456 		if (resource_size(res) & (PAGE_SIZE - 1)) {
457 			rc = -EIO;
458 			goto failed;
459 		}
460 		iov->barsz[i] = resource_size(res);
461 		res->end = res->start + resource_size(res) * total - 1;
462 		pci_info(dev, "VF(n) BAR%d space: %pR (contains BAR%d for %d VFs)\n",
463 			 i, res, i, total);
464 		i += bar64;
465 		nres++;
466 	}
467 
468 	iov->pos = pos;
469 	iov->nres = nres;
470 	iov->ctrl = ctrl;
471 	iov->total_VFs = total;
472 	iov->driver_max_VFs = total;
473 	pci_read_config_word(dev, pos + PCI_SRIOV_VF_DID, &iov->vf_device);
474 	iov->pgsz = pgsz;
475 	iov->self = dev;
476 	iov->drivers_autoprobe = true;
477 	pci_read_config_dword(dev, pos + PCI_SRIOV_CAP, &iov->cap);
478 	pci_read_config_byte(dev, pos + PCI_SRIOV_FUNC_LINK, &iov->link);
479 	if (pci_pcie_type(dev) == PCI_EXP_TYPE_RC_END)
480 		iov->link = PCI_DEVFN(PCI_SLOT(dev->devfn), iov->link);
481 
482 	if (pdev)
483 		iov->dev = pci_dev_get(pdev);
484 	else
485 		iov->dev = dev;
486 
487 	dev->sriov = iov;
488 	dev->is_physfn = 1;
489 	rc = compute_max_vf_buses(dev);
490 	if (rc)
491 		goto fail_max_buses;
492 
493 	return 0;
494 
495 fail_max_buses:
496 	dev->sriov = NULL;
497 	dev->is_physfn = 0;
498 failed:
499 	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
500 		res = &dev->resource[i + PCI_IOV_RESOURCES];
501 		res->flags = 0;
502 	}
503 
504 	kfree(iov);
505 	return rc;
506 }
507 
508 static void sriov_release(struct pci_dev *dev)
509 {
510 	BUG_ON(dev->sriov->num_VFs);
511 
512 	if (dev != dev->sriov->dev)
513 		pci_dev_put(dev->sriov->dev);
514 
515 	kfree(dev->sriov);
516 	dev->sriov = NULL;
517 }
518 
519 static void sriov_restore_state(struct pci_dev *dev)
520 {
521 	int i;
522 	u16 ctrl;
523 	struct pci_sriov *iov = dev->sriov;
524 
525 	pci_read_config_word(dev, iov->pos + PCI_SRIOV_CTRL, &ctrl);
526 	if (ctrl & PCI_SRIOV_CTRL_VFE)
527 		return;
528 
529 	/*
530 	 * Restore PCI_SRIOV_CTRL_ARI before pci_iov_set_numvfs() because
531 	 * it reads offset & stride, which depend on PCI_SRIOV_CTRL_ARI.
532 	 */
533 	ctrl &= ~PCI_SRIOV_CTRL_ARI;
534 	ctrl |= iov->ctrl & PCI_SRIOV_CTRL_ARI;
535 	pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, ctrl);
536 
537 	for (i = PCI_IOV_RESOURCES; i <= PCI_IOV_RESOURCE_END; i++)
538 		pci_update_resource(dev, i);
539 
540 	pci_write_config_dword(dev, iov->pos + PCI_SRIOV_SYS_PGSIZE, iov->pgsz);
541 	pci_iov_set_numvfs(dev, iov->num_VFs);
542 	pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
543 	if (iov->ctrl & PCI_SRIOV_CTRL_VFE)
544 		msleep(100);
545 }
546 
547 /**
548  * pci_iov_init - initialize the IOV capability
549  * @dev: the PCI device
550  *
551  * Returns 0 on success, or negative on failure.
552  */
553 int pci_iov_init(struct pci_dev *dev)
554 {
555 	int pos;
556 
557 	if (!pci_is_pcie(dev))
558 		return -ENODEV;
559 
560 	pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_SRIOV);
561 	if (pos)
562 		return sriov_init(dev, pos);
563 
564 	return -ENODEV;
565 }
566 
567 /**
568  * pci_iov_release - release resources used by the IOV capability
569  * @dev: the PCI device
570  */
571 void pci_iov_release(struct pci_dev *dev)
572 {
573 	if (dev->is_physfn)
574 		sriov_release(dev);
575 }
576 
577 /**
578  * pci_iov_remove - clean up SR-IOV state after PF driver is detached
579  * @dev: the PCI device
580  */
581 void pci_iov_remove(struct pci_dev *dev)
582 {
583 	struct pci_sriov *iov = dev->sriov;
584 
585 	if (!dev->is_physfn)
586 		return;
587 
588 	iov->driver_max_VFs = iov->total_VFs;
589 	if (iov->num_VFs)
590 		pci_warn(dev, "driver left SR-IOV enabled after remove\n");
591 }
592 
593 /**
594  * pci_iov_update_resource - update a VF BAR
595  * @dev: the PCI device
596  * @resno: the resource number
597  *
598  * Update a VF BAR in the SR-IOV capability of a PF.
599  */
600 void pci_iov_update_resource(struct pci_dev *dev, int resno)
601 {
602 	struct pci_sriov *iov = dev->is_physfn ? dev->sriov : NULL;
603 	struct resource *res = dev->resource + resno;
604 	int vf_bar = resno - PCI_IOV_RESOURCES;
605 	struct pci_bus_region region;
606 	u16 cmd;
607 	u32 new;
608 	int reg;
609 
610 	/*
611 	 * The generic pci_restore_bars() path calls this for all devices,
612 	 * including VFs and non-SR-IOV devices.  If this is not a PF, we
613 	 * have nothing to do.
614 	 */
615 	if (!iov)
616 		return;
617 
618 	pci_read_config_word(dev, iov->pos + PCI_SRIOV_CTRL, &cmd);
619 	if ((cmd & PCI_SRIOV_CTRL_VFE) && (cmd & PCI_SRIOV_CTRL_MSE)) {
620 		dev_WARN(&dev->dev, "can't update enabled VF BAR%d %pR\n",
621 			 vf_bar, res);
622 		return;
623 	}
624 
625 	/*
626 	 * Ignore unimplemented BARs, unused resource slots for 64-bit
627 	 * BARs, and non-movable resources, e.g., those described via
628 	 * Enhanced Allocation.
629 	 */
630 	if (!res->flags)
631 		return;
632 
633 	if (res->flags & IORESOURCE_UNSET)
634 		return;
635 
636 	if (res->flags & IORESOURCE_PCI_FIXED)
637 		return;
638 
639 	pcibios_resource_to_bus(dev->bus, &region, res);
640 	new = region.start;
641 	new |= res->flags & ~PCI_BASE_ADDRESS_MEM_MASK;
642 
643 	reg = iov->pos + PCI_SRIOV_BAR + 4 * vf_bar;
644 	pci_write_config_dword(dev, reg, new);
645 	if (res->flags & IORESOURCE_MEM_64) {
646 		new = region.start >> 16 >> 16;
647 		pci_write_config_dword(dev, reg + 4, new);
648 	}
649 }
650 
651 resource_size_t __weak pcibios_iov_resource_alignment(struct pci_dev *dev,
652 						      int resno)
653 {
654 	return pci_iov_resource_size(dev, resno);
655 }
656 
657 /**
658  * pci_sriov_resource_alignment - get resource alignment for VF BAR
659  * @dev: the PCI device
660  * @resno: the resource number
661  *
662  * Returns the alignment of the VF BAR found in the SR-IOV capability.
663  * This is not the same as the resource size which is defined as
664  * the VF BAR size multiplied by the number of VFs.  The alignment
665  * is just the VF BAR size.
666  */
667 resource_size_t pci_sriov_resource_alignment(struct pci_dev *dev, int resno)
668 {
669 	return pcibios_iov_resource_alignment(dev, resno);
670 }
671 
672 /**
673  * pci_restore_iov_state - restore the state of the IOV capability
674  * @dev: the PCI device
675  */
676 void pci_restore_iov_state(struct pci_dev *dev)
677 {
678 	if (dev->is_physfn)
679 		sriov_restore_state(dev);
680 }
681 
682 /**
683  * pci_vf_drivers_autoprobe - set PF property drivers_autoprobe for VFs
684  * @dev: the PCI device
685  * @auto_probe: set VF drivers auto probe flag
686  */
687 void pci_vf_drivers_autoprobe(struct pci_dev *dev, bool auto_probe)
688 {
689 	if (dev->is_physfn)
690 		dev->sriov->drivers_autoprobe = auto_probe;
691 }
692 
693 /**
694  * pci_iov_bus_range - find bus range used by Virtual Function
695  * @bus: the PCI bus
696  *
697  * Returns max number of buses (exclude current one) used by Virtual
698  * Functions.
699  */
700 int pci_iov_bus_range(struct pci_bus *bus)
701 {
702 	int max = 0;
703 	struct pci_dev *dev;
704 
705 	list_for_each_entry(dev, &bus->devices, bus_list) {
706 		if (!dev->is_physfn)
707 			continue;
708 		if (dev->sriov->max_VF_buses > max)
709 			max = dev->sriov->max_VF_buses;
710 	}
711 
712 	return max ? max - bus->number : 0;
713 }
714 
715 /**
716  * pci_enable_sriov - enable the SR-IOV capability
717  * @dev: the PCI device
718  * @nr_virtfn: number of virtual functions to enable
719  *
720  * Returns 0 on success, or negative on failure.
721  */
722 int pci_enable_sriov(struct pci_dev *dev, int nr_virtfn)
723 {
724 	might_sleep();
725 
726 	if (!dev->is_physfn)
727 		return -ENOSYS;
728 
729 	return sriov_enable(dev, nr_virtfn);
730 }
731 EXPORT_SYMBOL_GPL(pci_enable_sriov);
732 
733 /**
734  * pci_disable_sriov - disable the SR-IOV capability
735  * @dev: the PCI device
736  */
737 void pci_disable_sriov(struct pci_dev *dev)
738 {
739 	might_sleep();
740 
741 	if (!dev->is_physfn)
742 		return;
743 
744 	sriov_disable(dev);
745 }
746 EXPORT_SYMBOL_GPL(pci_disable_sriov);
747 
748 /**
749  * pci_num_vf - return number of VFs associated with a PF device_release_driver
750  * @dev: the PCI device
751  *
752  * Returns number of VFs, or 0 if SR-IOV is not enabled.
753  */
754 int pci_num_vf(struct pci_dev *dev)
755 {
756 	if (!dev->is_physfn)
757 		return 0;
758 
759 	return dev->sriov->num_VFs;
760 }
761 EXPORT_SYMBOL_GPL(pci_num_vf);
762 
763 /**
764  * pci_vfs_assigned - returns number of VFs are assigned to a guest
765  * @dev: the PCI device
766  *
767  * Returns number of VFs belonging to this device that are assigned to a guest.
768  * If device is not a physical function returns 0.
769  */
770 int pci_vfs_assigned(struct pci_dev *dev)
771 {
772 	struct pci_dev *vfdev;
773 	unsigned int vfs_assigned = 0;
774 	unsigned short dev_id;
775 
776 	/* only search if we are a PF */
777 	if (!dev->is_physfn)
778 		return 0;
779 
780 	/*
781 	 * determine the device ID for the VFs, the vendor ID will be the
782 	 * same as the PF so there is no need to check for that one
783 	 */
784 	dev_id = dev->sriov->vf_device;
785 
786 	/* loop through all the VFs to see if we own any that are assigned */
787 	vfdev = pci_get_device(dev->vendor, dev_id, NULL);
788 	while (vfdev) {
789 		/*
790 		 * It is considered assigned if it is a virtual function with
791 		 * our dev as the physical function and the assigned bit is set
792 		 */
793 		if (vfdev->is_virtfn && (vfdev->physfn == dev) &&
794 			pci_is_dev_assigned(vfdev))
795 			vfs_assigned++;
796 
797 		vfdev = pci_get_device(dev->vendor, dev_id, vfdev);
798 	}
799 
800 	return vfs_assigned;
801 }
802 EXPORT_SYMBOL_GPL(pci_vfs_assigned);
803 
804 /**
805  * pci_sriov_set_totalvfs -- reduce the TotalVFs available
806  * @dev: the PCI PF device
807  * @numvfs: number that should be used for TotalVFs supported
808  *
809  * Should be called from PF driver's probe routine with
810  * device's mutex held.
811  *
812  * Returns 0 if PF is an SRIOV-capable device and
813  * value of numvfs valid. If not a PF return -ENOSYS;
814  * if numvfs is invalid return -EINVAL;
815  * if VFs already enabled, return -EBUSY.
816  */
817 int pci_sriov_set_totalvfs(struct pci_dev *dev, u16 numvfs)
818 {
819 	if (!dev->is_physfn)
820 		return -ENOSYS;
821 
822 	if (numvfs > dev->sriov->total_VFs)
823 		return -EINVAL;
824 
825 	/* Shouldn't change if VFs already enabled */
826 	if (dev->sriov->ctrl & PCI_SRIOV_CTRL_VFE)
827 		return -EBUSY;
828 
829 	dev->sriov->driver_max_VFs = numvfs;
830 	return 0;
831 }
832 EXPORT_SYMBOL_GPL(pci_sriov_set_totalvfs);
833 
834 /**
835  * pci_sriov_get_totalvfs -- get total VFs supported on this device
836  * @dev: the PCI PF device
837  *
838  * For a PCIe device with SRIOV support, return the PCIe
839  * SRIOV capability value of TotalVFs or the value of driver_max_VFs
840  * if the driver reduced it.  Otherwise 0.
841  */
842 int pci_sriov_get_totalvfs(struct pci_dev *dev)
843 {
844 	if (!dev->is_physfn)
845 		return 0;
846 
847 	return dev->sriov->driver_max_VFs;
848 }
849 EXPORT_SYMBOL_GPL(pci_sriov_get_totalvfs);
850 
851 /**
852  * pci_sriov_configure_simple - helper to configure SR-IOV
853  * @dev: the PCI device
854  * @nr_virtfn: number of virtual functions to enable, 0 to disable
855  *
856  * Enable or disable SR-IOV for devices that don't require any PF setup
857  * before enabling SR-IOV.  Return value is negative on error, or number of
858  * VFs allocated on success.
859  */
860 int pci_sriov_configure_simple(struct pci_dev *dev, int nr_virtfn)
861 {
862 	int rc;
863 
864 	might_sleep();
865 
866 	if (!dev->is_physfn)
867 		return -ENODEV;
868 
869 	if (pci_vfs_assigned(dev)) {
870 		pci_warn(dev, "Cannot modify SR-IOV while VFs are assigned\n");
871 		return -EPERM;
872 	}
873 
874 	if (nr_virtfn == 0) {
875 		sriov_disable(dev);
876 		return 0;
877 	}
878 
879 	rc = sriov_enable(dev, nr_virtfn);
880 	if (rc < 0)
881 		return rc;
882 
883 	return nr_virtfn;
884 }
885 EXPORT_SYMBOL_GPL(pci_sriov_configure_simple);
886