xref: /linux/drivers/pci/iov.c (revision 87c9c16317882dd6dbbc07e349bc3223e14f3244)
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 sprintf(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 sprintf(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 w/callback */
395 	if (!pdev->driver || !pdev->driver->sriov_configure) {
396 		pci_info(pdev, "Driver does not support SRIOV configuration via sysfs\n");
397 		ret = -ENOENT;
398 		goto exit;
399 	}
400 
401 	if (num_vfs == 0) {
402 		/* disable VFs */
403 		ret = pdev->driver->sriov_configure(pdev, 0);
404 		goto exit;
405 	}
406 
407 	/* enable VFs */
408 	if (pdev->sriov->num_VFs) {
409 		pci_warn(pdev, "%d VFs already enabled. Disable before enabling %d VFs\n",
410 			 pdev->sriov->num_VFs, num_vfs);
411 		ret = -EBUSY;
412 		goto exit;
413 	}
414 
415 	ret = pdev->driver->sriov_configure(pdev, num_vfs);
416 	if (ret < 0)
417 		goto exit;
418 
419 	if (ret != num_vfs)
420 		pci_warn(pdev, "%d VFs requested; only %d enabled\n",
421 			 num_vfs, ret);
422 
423 exit:
424 	device_unlock(&pdev->dev);
425 
426 	if (ret < 0)
427 		return ret;
428 
429 	return count;
430 }
431 
432 static ssize_t sriov_offset_show(struct device *dev,
433 				 struct device_attribute *attr,
434 				 char *buf)
435 {
436 	struct pci_dev *pdev = to_pci_dev(dev);
437 
438 	return sprintf(buf, "%u\n", pdev->sriov->offset);
439 }
440 
441 static ssize_t sriov_stride_show(struct device *dev,
442 				 struct device_attribute *attr,
443 				 char *buf)
444 {
445 	struct pci_dev *pdev = to_pci_dev(dev);
446 
447 	return sprintf(buf, "%u\n", pdev->sriov->stride);
448 }
449 
450 static ssize_t sriov_vf_device_show(struct device *dev,
451 				    struct device_attribute *attr,
452 				    char *buf)
453 {
454 	struct pci_dev *pdev = to_pci_dev(dev);
455 
456 	return sprintf(buf, "%x\n", pdev->sriov->vf_device);
457 }
458 
459 static ssize_t sriov_drivers_autoprobe_show(struct device *dev,
460 					    struct device_attribute *attr,
461 					    char *buf)
462 {
463 	struct pci_dev *pdev = to_pci_dev(dev);
464 
465 	return sprintf(buf, "%u\n", pdev->sriov->drivers_autoprobe);
466 }
467 
468 static ssize_t sriov_drivers_autoprobe_store(struct device *dev,
469 					     struct device_attribute *attr,
470 					     const char *buf, size_t count)
471 {
472 	struct pci_dev *pdev = to_pci_dev(dev);
473 	bool drivers_autoprobe;
474 
475 	if (kstrtobool(buf, &drivers_autoprobe) < 0)
476 		return -EINVAL;
477 
478 	pdev->sriov->drivers_autoprobe = drivers_autoprobe;
479 
480 	return count;
481 }
482 
483 static DEVICE_ATTR_RO(sriov_totalvfs);
484 static DEVICE_ATTR_RW(sriov_numvfs);
485 static DEVICE_ATTR_RO(sriov_offset);
486 static DEVICE_ATTR_RO(sriov_stride);
487 static DEVICE_ATTR_RO(sriov_vf_device);
488 static DEVICE_ATTR_RW(sriov_drivers_autoprobe);
489 
490 static struct attribute *sriov_pf_dev_attrs[] = {
491 	&dev_attr_sriov_totalvfs.attr,
492 	&dev_attr_sriov_numvfs.attr,
493 	&dev_attr_sriov_offset.attr,
494 	&dev_attr_sriov_stride.attr,
495 	&dev_attr_sriov_vf_device.attr,
496 	&dev_attr_sriov_drivers_autoprobe.attr,
497 #ifdef CONFIG_PCI_MSI
498 	&dev_attr_sriov_vf_total_msix.attr,
499 #endif
500 	NULL,
501 };
502 
503 static umode_t sriov_pf_attrs_are_visible(struct kobject *kobj,
504 					  struct attribute *a, int n)
505 {
506 	struct device *dev = kobj_to_dev(kobj);
507 
508 	if (!dev_is_pf(dev))
509 		return 0;
510 
511 	return a->mode;
512 }
513 
514 const struct attribute_group sriov_pf_dev_attr_group = {
515 	.attrs = sriov_pf_dev_attrs,
516 	.is_visible = sriov_pf_attrs_are_visible,
517 };
518 
519 int __weak pcibios_sriov_enable(struct pci_dev *pdev, u16 num_vfs)
520 {
521 	return 0;
522 }
523 
524 int __weak pcibios_sriov_disable(struct pci_dev *pdev)
525 {
526 	return 0;
527 }
528 
529 static int sriov_add_vfs(struct pci_dev *dev, u16 num_vfs)
530 {
531 	unsigned int i;
532 	int rc;
533 
534 	if (dev->no_vf_scan)
535 		return 0;
536 
537 	for (i = 0; i < num_vfs; i++) {
538 		rc = pci_iov_add_virtfn(dev, i);
539 		if (rc)
540 			goto failed;
541 	}
542 	return 0;
543 failed:
544 	while (i--)
545 		pci_iov_remove_virtfn(dev, i);
546 
547 	return rc;
548 }
549 
550 static int sriov_enable(struct pci_dev *dev, int nr_virtfn)
551 {
552 	int rc;
553 	int i;
554 	int nres;
555 	u16 initial;
556 	struct resource *res;
557 	struct pci_dev *pdev;
558 	struct pci_sriov *iov = dev->sriov;
559 	int bars = 0;
560 	int bus;
561 
562 	if (!nr_virtfn)
563 		return 0;
564 
565 	if (iov->num_VFs)
566 		return -EINVAL;
567 
568 	pci_read_config_word(dev, iov->pos + PCI_SRIOV_INITIAL_VF, &initial);
569 	if (initial > iov->total_VFs ||
570 	    (!(iov->cap & PCI_SRIOV_CAP_VFM) && (initial != iov->total_VFs)))
571 		return -EIO;
572 
573 	if (nr_virtfn < 0 || nr_virtfn > iov->total_VFs ||
574 	    (!(iov->cap & PCI_SRIOV_CAP_VFM) && (nr_virtfn > initial)))
575 		return -EINVAL;
576 
577 	nres = 0;
578 	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
579 		bars |= (1 << (i + PCI_IOV_RESOURCES));
580 		res = &dev->resource[i + PCI_IOV_RESOURCES];
581 		if (res->parent)
582 			nres++;
583 	}
584 	if (nres != iov->nres) {
585 		pci_err(dev, "not enough MMIO resources for SR-IOV\n");
586 		return -ENOMEM;
587 	}
588 
589 	bus = pci_iov_virtfn_bus(dev, nr_virtfn - 1);
590 	if (bus > dev->bus->busn_res.end) {
591 		pci_err(dev, "can't enable %d VFs (bus %02x out of range of %pR)\n",
592 			nr_virtfn, bus, &dev->bus->busn_res);
593 		return -ENOMEM;
594 	}
595 
596 	if (pci_enable_resources(dev, bars)) {
597 		pci_err(dev, "SR-IOV: IOV BARS not allocated\n");
598 		return -ENOMEM;
599 	}
600 
601 	if (iov->link != dev->devfn) {
602 		pdev = pci_get_slot(dev->bus, iov->link);
603 		if (!pdev)
604 			return -ENODEV;
605 
606 		if (!pdev->is_physfn) {
607 			pci_dev_put(pdev);
608 			return -ENOSYS;
609 		}
610 
611 		rc = sysfs_create_link(&dev->dev.kobj,
612 					&pdev->dev.kobj, "dep_link");
613 		pci_dev_put(pdev);
614 		if (rc)
615 			return rc;
616 	}
617 
618 	iov->initial_VFs = initial;
619 	if (nr_virtfn < initial)
620 		initial = nr_virtfn;
621 
622 	rc = pcibios_sriov_enable(dev, initial);
623 	if (rc) {
624 		pci_err(dev, "failure %d from pcibios_sriov_enable()\n", rc);
625 		goto err_pcibios;
626 	}
627 
628 	pci_iov_set_numvfs(dev, nr_virtfn);
629 	iov->ctrl |= PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE;
630 	pci_cfg_access_lock(dev);
631 	pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
632 	msleep(100);
633 	pci_cfg_access_unlock(dev);
634 
635 	rc = sriov_add_vfs(dev, initial);
636 	if (rc)
637 		goto err_pcibios;
638 
639 	kobject_uevent(&dev->dev.kobj, KOBJ_CHANGE);
640 	iov->num_VFs = nr_virtfn;
641 
642 	return 0;
643 
644 err_pcibios:
645 	iov->ctrl &= ~(PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE);
646 	pci_cfg_access_lock(dev);
647 	pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
648 	ssleep(1);
649 	pci_cfg_access_unlock(dev);
650 
651 	pcibios_sriov_disable(dev);
652 
653 	if (iov->link != dev->devfn)
654 		sysfs_remove_link(&dev->dev.kobj, "dep_link");
655 
656 	pci_iov_set_numvfs(dev, 0);
657 	return rc;
658 }
659 
660 static void sriov_del_vfs(struct pci_dev *dev)
661 {
662 	struct pci_sriov *iov = dev->sriov;
663 	int i;
664 
665 	for (i = 0; i < iov->num_VFs; i++)
666 		pci_iov_remove_virtfn(dev, i);
667 }
668 
669 static void sriov_disable(struct pci_dev *dev)
670 {
671 	struct pci_sriov *iov = dev->sriov;
672 
673 	if (!iov->num_VFs)
674 		return;
675 
676 	sriov_del_vfs(dev);
677 	iov->ctrl &= ~(PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE);
678 	pci_cfg_access_lock(dev);
679 	pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
680 	ssleep(1);
681 	pci_cfg_access_unlock(dev);
682 
683 	pcibios_sriov_disable(dev);
684 
685 	if (iov->link != dev->devfn)
686 		sysfs_remove_link(&dev->dev.kobj, "dep_link");
687 
688 	iov->num_VFs = 0;
689 	pci_iov_set_numvfs(dev, 0);
690 }
691 
692 static int sriov_init(struct pci_dev *dev, int pos)
693 {
694 	int i, bar64;
695 	int rc;
696 	int nres;
697 	u32 pgsz;
698 	u16 ctrl, total;
699 	struct pci_sriov *iov;
700 	struct resource *res;
701 	struct pci_dev *pdev;
702 
703 	pci_read_config_word(dev, pos + PCI_SRIOV_CTRL, &ctrl);
704 	if (ctrl & PCI_SRIOV_CTRL_VFE) {
705 		pci_write_config_word(dev, pos + PCI_SRIOV_CTRL, 0);
706 		ssleep(1);
707 	}
708 
709 	ctrl = 0;
710 	list_for_each_entry(pdev, &dev->bus->devices, bus_list)
711 		if (pdev->is_physfn)
712 			goto found;
713 
714 	pdev = NULL;
715 	if (pci_ari_enabled(dev->bus))
716 		ctrl |= PCI_SRIOV_CTRL_ARI;
717 
718 found:
719 	pci_write_config_word(dev, pos + PCI_SRIOV_CTRL, ctrl);
720 
721 	pci_read_config_word(dev, pos + PCI_SRIOV_TOTAL_VF, &total);
722 	if (!total)
723 		return 0;
724 
725 	pci_read_config_dword(dev, pos + PCI_SRIOV_SUP_PGSIZE, &pgsz);
726 	i = PAGE_SHIFT > 12 ? PAGE_SHIFT - 12 : 0;
727 	pgsz &= ~((1 << i) - 1);
728 	if (!pgsz)
729 		return -EIO;
730 
731 	pgsz &= ~(pgsz - 1);
732 	pci_write_config_dword(dev, pos + PCI_SRIOV_SYS_PGSIZE, pgsz);
733 
734 	iov = kzalloc(sizeof(*iov), GFP_KERNEL);
735 	if (!iov)
736 		return -ENOMEM;
737 
738 	nres = 0;
739 	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
740 		res = &dev->resource[i + PCI_IOV_RESOURCES];
741 		/*
742 		 * If it is already FIXED, don't change it, something
743 		 * (perhaps EA or header fixups) wants it this way.
744 		 */
745 		if (res->flags & IORESOURCE_PCI_FIXED)
746 			bar64 = (res->flags & IORESOURCE_MEM_64) ? 1 : 0;
747 		else
748 			bar64 = __pci_read_base(dev, pci_bar_unknown, res,
749 						pos + PCI_SRIOV_BAR + i * 4);
750 		if (!res->flags)
751 			continue;
752 		if (resource_size(res) & (PAGE_SIZE - 1)) {
753 			rc = -EIO;
754 			goto failed;
755 		}
756 		iov->barsz[i] = resource_size(res);
757 		res->end = res->start + resource_size(res) * total - 1;
758 		pci_info(dev, "VF(n) BAR%d space: %pR (contains BAR%d for %d VFs)\n",
759 			 i, res, i, total);
760 		i += bar64;
761 		nres++;
762 	}
763 
764 	iov->pos = pos;
765 	iov->nres = nres;
766 	iov->ctrl = ctrl;
767 	iov->total_VFs = total;
768 	iov->driver_max_VFs = total;
769 	pci_read_config_word(dev, pos + PCI_SRIOV_VF_DID, &iov->vf_device);
770 	iov->pgsz = pgsz;
771 	iov->self = dev;
772 	iov->drivers_autoprobe = true;
773 	pci_read_config_dword(dev, pos + PCI_SRIOV_CAP, &iov->cap);
774 	pci_read_config_byte(dev, pos + PCI_SRIOV_FUNC_LINK, &iov->link);
775 	if (pci_pcie_type(dev) == PCI_EXP_TYPE_RC_END)
776 		iov->link = PCI_DEVFN(PCI_SLOT(dev->devfn), iov->link);
777 
778 	if (pdev)
779 		iov->dev = pci_dev_get(pdev);
780 	else
781 		iov->dev = dev;
782 
783 	dev->sriov = iov;
784 	dev->is_physfn = 1;
785 	rc = compute_max_vf_buses(dev);
786 	if (rc)
787 		goto fail_max_buses;
788 
789 	return 0;
790 
791 fail_max_buses:
792 	dev->sriov = NULL;
793 	dev->is_physfn = 0;
794 failed:
795 	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
796 		res = &dev->resource[i + PCI_IOV_RESOURCES];
797 		res->flags = 0;
798 	}
799 
800 	kfree(iov);
801 	return rc;
802 }
803 
804 static void sriov_release(struct pci_dev *dev)
805 {
806 	BUG_ON(dev->sriov->num_VFs);
807 
808 	if (dev != dev->sriov->dev)
809 		pci_dev_put(dev->sriov->dev);
810 
811 	kfree(dev->sriov);
812 	dev->sriov = NULL;
813 }
814 
815 static void sriov_restore_state(struct pci_dev *dev)
816 {
817 	int i;
818 	u16 ctrl;
819 	struct pci_sriov *iov = dev->sriov;
820 
821 	pci_read_config_word(dev, iov->pos + PCI_SRIOV_CTRL, &ctrl);
822 	if (ctrl & PCI_SRIOV_CTRL_VFE)
823 		return;
824 
825 	/*
826 	 * Restore PCI_SRIOV_CTRL_ARI before pci_iov_set_numvfs() because
827 	 * it reads offset & stride, which depend on PCI_SRIOV_CTRL_ARI.
828 	 */
829 	ctrl &= ~PCI_SRIOV_CTRL_ARI;
830 	ctrl |= iov->ctrl & PCI_SRIOV_CTRL_ARI;
831 	pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, ctrl);
832 
833 	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++)
834 		pci_update_resource(dev, i + PCI_IOV_RESOURCES);
835 
836 	pci_write_config_dword(dev, iov->pos + PCI_SRIOV_SYS_PGSIZE, iov->pgsz);
837 	pci_iov_set_numvfs(dev, iov->num_VFs);
838 	pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
839 	if (iov->ctrl & PCI_SRIOV_CTRL_VFE)
840 		msleep(100);
841 }
842 
843 /**
844  * pci_iov_init - initialize the IOV capability
845  * @dev: the PCI device
846  *
847  * Returns 0 on success, or negative on failure.
848  */
849 int pci_iov_init(struct pci_dev *dev)
850 {
851 	int pos;
852 
853 	if (!pci_is_pcie(dev))
854 		return -ENODEV;
855 
856 	pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_SRIOV);
857 	if (pos)
858 		return sriov_init(dev, pos);
859 
860 	return -ENODEV;
861 }
862 
863 /**
864  * pci_iov_release - release resources used by the IOV capability
865  * @dev: the PCI device
866  */
867 void pci_iov_release(struct pci_dev *dev)
868 {
869 	if (dev->is_physfn)
870 		sriov_release(dev);
871 }
872 
873 /**
874  * pci_iov_remove - clean up SR-IOV state after PF driver is detached
875  * @dev: the PCI device
876  */
877 void pci_iov_remove(struct pci_dev *dev)
878 {
879 	struct pci_sriov *iov = dev->sriov;
880 
881 	if (!dev->is_physfn)
882 		return;
883 
884 	iov->driver_max_VFs = iov->total_VFs;
885 	if (iov->num_VFs)
886 		pci_warn(dev, "driver left SR-IOV enabled after remove\n");
887 }
888 
889 /**
890  * pci_iov_update_resource - update a VF BAR
891  * @dev: the PCI device
892  * @resno: the resource number
893  *
894  * Update a VF BAR in the SR-IOV capability of a PF.
895  */
896 void pci_iov_update_resource(struct pci_dev *dev, int resno)
897 {
898 	struct pci_sriov *iov = dev->is_physfn ? dev->sriov : NULL;
899 	struct resource *res = dev->resource + resno;
900 	int vf_bar = resno - PCI_IOV_RESOURCES;
901 	struct pci_bus_region region;
902 	u16 cmd;
903 	u32 new;
904 	int reg;
905 
906 	/*
907 	 * The generic pci_restore_bars() path calls this for all devices,
908 	 * including VFs and non-SR-IOV devices.  If this is not a PF, we
909 	 * have nothing to do.
910 	 */
911 	if (!iov)
912 		return;
913 
914 	pci_read_config_word(dev, iov->pos + PCI_SRIOV_CTRL, &cmd);
915 	if ((cmd & PCI_SRIOV_CTRL_VFE) && (cmd & PCI_SRIOV_CTRL_MSE)) {
916 		dev_WARN(&dev->dev, "can't update enabled VF BAR%d %pR\n",
917 			 vf_bar, res);
918 		return;
919 	}
920 
921 	/*
922 	 * Ignore unimplemented BARs, unused resource slots for 64-bit
923 	 * BARs, and non-movable resources, e.g., those described via
924 	 * Enhanced Allocation.
925 	 */
926 	if (!res->flags)
927 		return;
928 
929 	if (res->flags & IORESOURCE_UNSET)
930 		return;
931 
932 	if (res->flags & IORESOURCE_PCI_FIXED)
933 		return;
934 
935 	pcibios_resource_to_bus(dev->bus, &region, res);
936 	new = region.start;
937 	new |= res->flags & ~PCI_BASE_ADDRESS_MEM_MASK;
938 
939 	reg = iov->pos + PCI_SRIOV_BAR + 4 * vf_bar;
940 	pci_write_config_dword(dev, reg, new);
941 	if (res->flags & IORESOURCE_MEM_64) {
942 		new = region.start >> 16 >> 16;
943 		pci_write_config_dword(dev, reg + 4, new);
944 	}
945 }
946 
947 resource_size_t __weak pcibios_iov_resource_alignment(struct pci_dev *dev,
948 						      int resno)
949 {
950 	return pci_iov_resource_size(dev, resno);
951 }
952 
953 /**
954  * pci_sriov_resource_alignment - get resource alignment for VF BAR
955  * @dev: the PCI device
956  * @resno: the resource number
957  *
958  * Returns the alignment of the VF BAR found in the SR-IOV capability.
959  * This is not the same as the resource size which is defined as
960  * the VF BAR size multiplied by the number of VFs.  The alignment
961  * is just the VF BAR size.
962  */
963 resource_size_t pci_sriov_resource_alignment(struct pci_dev *dev, int resno)
964 {
965 	return pcibios_iov_resource_alignment(dev, resno);
966 }
967 
968 /**
969  * pci_restore_iov_state - restore the state of the IOV capability
970  * @dev: the PCI device
971  */
972 void pci_restore_iov_state(struct pci_dev *dev)
973 {
974 	if (dev->is_physfn)
975 		sriov_restore_state(dev);
976 }
977 
978 /**
979  * pci_vf_drivers_autoprobe - set PF property drivers_autoprobe for VFs
980  * @dev: the PCI device
981  * @auto_probe: set VF drivers auto probe flag
982  */
983 void pci_vf_drivers_autoprobe(struct pci_dev *dev, bool auto_probe)
984 {
985 	if (dev->is_physfn)
986 		dev->sriov->drivers_autoprobe = auto_probe;
987 }
988 
989 /**
990  * pci_iov_bus_range - find bus range used by Virtual Function
991  * @bus: the PCI bus
992  *
993  * Returns max number of buses (exclude current one) used by Virtual
994  * Functions.
995  */
996 int pci_iov_bus_range(struct pci_bus *bus)
997 {
998 	int max = 0;
999 	struct pci_dev *dev;
1000 
1001 	list_for_each_entry(dev, &bus->devices, bus_list) {
1002 		if (!dev->is_physfn)
1003 			continue;
1004 		if (dev->sriov->max_VF_buses > max)
1005 			max = dev->sriov->max_VF_buses;
1006 	}
1007 
1008 	return max ? max - bus->number : 0;
1009 }
1010 
1011 /**
1012  * pci_enable_sriov - enable the SR-IOV capability
1013  * @dev: the PCI device
1014  * @nr_virtfn: number of virtual functions to enable
1015  *
1016  * Returns 0 on success, or negative on failure.
1017  */
1018 int pci_enable_sriov(struct pci_dev *dev, int nr_virtfn)
1019 {
1020 	might_sleep();
1021 
1022 	if (!dev->is_physfn)
1023 		return -ENOSYS;
1024 
1025 	return sriov_enable(dev, nr_virtfn);
1026 }
1027 EXPORT_SYMBOL_GPL(pci_enable_sriov);
1028 
1029 /**
1030  * pci_disable_sriov - disable the SR-IOV capability
1031  * @dev: the PCI device
1032  */
1033 void pci_disable_sriov(struct pci_dev *dev)
1034 {
1035 	might_sleep();
1036 
1037 	if (!dev->is_physfn)
1038 		return;
1039 
1040 	sriov_disable(dev);
1041 }
1042 EXPORT_SYMBOL_GPL(pci_disable_sriov);
1043 
1044 /**
1045  * pci_num_vf - return number of VFs associated with a PF device_release_driver
1046  * @dev: the PCI device
1047  *
1048  * Returns number of VFs, or 0 if SR-IOV is not enabled.
1049  */
1050 int pci_num_vf(struct pci_dev *dev)
1051 {
1052 	if (!dev->is_physfn)
1053 		return 0;
1054 
1055 	return dev->sriov->num_VFs;
1056 }
1057 EXPORT_SYMBOL_GPL(pci_num_vf);
1058 
1059 /**
1060  * pci_vfs_assigned - returns number of VFs are assigned to a guest
1061  * @dev: the PCI device
1062  *
1063  * Returns number of VFs belonging to this device that are assigned to a guest.
1064  * If device is not a physical function returns 0.
1065  */
1066 int pci_vfs_assigned(struct pci_dev *dev)
1067 {
1068 	struct pci_dev *vfdev;
1069 	unsigned int vfs_assigned = 0;
1070 	unsigned short dev_id;
1071 
1072 	/* only search if we are a PF */
1073 	if (!dev->is_physfn)
1074 		return 0;
1075 
1076 	/*
1077 	 * determine the device ID for the VFs, the vendor ID will be the
1078 	 * same as the PF so there is no need to check for that one
1079 	 */
1080 	dev_id = dev->sriov->vf_device;
1081 
1082 	/* loop through all the VFs to see if we own any that are assigned */
1083 	vfdev = pci_get_device(dev->vendor, dev_id, NULL);
1084 	while (vfdev) {
1085 		/*
1086 		 * It is considered assigned if it is a virtual function with
1087 		 * our dev as the physical function and the assigned bit is set
1088 		 */
1089 		if (vfdev->is_virtfn && (vfdev->physfn == dev) &&
1090 			pci_is_dev_assigned(vfdev))
1091 			vfs_assigned++;
1092 
1093 		vfdev = pci_get_device(dev->vendor, dev_id, vfdev);
1094 	}
1095 
1096 	return vfs_assigned;
1097 }
1098 EXPORT_SYMBOL_GPL(pci_vfs_assigned);
1099 
1100 /**
1101  * pci_sriov_set_totalvfs -- reduce the TotalVFs available
1102  * @dev: the PCI PF device
1103  * @numvfs: number that should be used for TotalVFs supported
1104  *
1105  * Should be called from PF driver's probe routine with
1106  * device's mutex held.
1107  *
1108  * Returns 0 if PF is an SRIOV-capable device and
1109  * value of numvfs valid. If not a PF return -ENOSYS;
1110  * if numvfs is invalid return -EINVAL;
1111  * if VFs already enabled, return -EBUSY.
1112  */
1113 int pci_sriov_set_totalvfs(struct pci_dev *dev, u16 numvfs)
1114 {
1115 	if (!dev->is_physfn)
1116 		return -ENOSYS;
1117 
1118 	if (numvfs > dev->sriov->total_VFs)
1119 		return -EINVAL;
1120 
1121 	/* Shouldn't change if VFs already enabled */
1122 	if (dev->sriov->ctrl & PCI_SRIOV_CTRL_VFE)
1123 		return -EBUSY;
1124 
1125 	dev->sriov->driver_max_VFs = numvfs;
1126 	return 0;
1127 }
1128 EXPORT_SYMBOL_GPL(pci_sriov_set_totalvfs);
1129 
1130 /**
1131  * pci_sriov_get_totalvfs -- get total VFs supported on this device
1132  * @dev: the PCI PF device
1133  *
1134  * For a PCIe device with SRIOV support, return the PCIe
1135  * SRIOV capability value of TotalVFs or the value of driver_max_VFs
1136  * if the driver reduced it.  Otherwise 0.
1137  */
1138 int pci_sriov_get_totalvfs(struct pci_dev *dev)
1139 {
1140 	if (!dev->is_physfn)
1141 		return 0;
1142 
1143 	return dev->sriov->driver_max_VFs;
1144 }
1145 EXPORT_SYMBOL_GPL(pci_sriov_get_totalvfs);
1146 
1147 /**
1148  * pci_sriov_configure_simple - helper to configure SR-IOV
1149  * @dev: the PCI device
1150  * @nr_virtfn: number of virtual functions to enable, 0 to disable
1151  *
1152  * Enable or disable SR-IOV for devices that don't require any PF setup
1153  * before enabling SR-IOV.  Return value is negative on error, or number of
1154  * VFs allocated on success.
1155  */
1156 int pci_sriov_configure_simple(struct pci_dev *dev, int nr_virtfn)
1157 {
1158 	int rc;
1159 
1160 	might_sleep();
1161 
1162 	if (!dev->is_physfn)
1163 		return -ENODEV;
1164 
1165 	if (pci_vfs_assigned(dev)) {
1166 		pci_warn(dev, "Cannot modify SR-IOV while VFs are assigned\n");
1167 		return -EPERM;
1168 	}
1169 
1170 	if (nr_virtfn == 0) {
1171 		sriov_disable(dev);
1172 		return 0;
1173 	}
1174 
1175 	rc = sriov_enable(dev, nr_virtfn);
1176 	if (rc < 0)
1177 		return rc;
1178 
1179 	return nr_virtfn;
1180 }
1181 EXPORT_SYMBOL_GPL(pci_sriov_configure_simple);
1182