Linux Audio

Check our new training course

Linux kernel drivers training

Mar 31-Apr 9, 2025, special US time zones
Register
Loading...
v3.5.6
  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
 22static 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
 28static 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
 34static struct pci_bus *virtfn_add_bus(struct pci_bus *bus, int busnr)
 35{
 36	int rc;
 37	struct pci_bus *child;
 38
 39	if (bus->number == busnr)
 40		return bus;
 41
 42	child = pci_find_bus(pci_domain_nr(bus), busnr);
 43	if (child)
 44		return child;
 45
 46	child = pci_add_new_bus(bus, NULL, busnr);
 47	if (!child)
 48		return NULL;
 49
 50	child->subordinate = busnr;
 51	child->dev.parent = bus->bridge;
 52	rc = pci_bus_add_child(child);
 53	if (rc) {
 54		pci_remove_bus(child);
 55		return NULL;
 56	}
 57
 58	return child;
 59}
 60
 61static void virtfn_remove_bus(struct pci_bus *bus, int busnr)
 62{
 63	struct pci_bus *child;
 64
 65	if (bus->number == busnr)
 66		return;
 67
 68	child = pci_find_bus(pci_domain_nr(bus), busnr);
 69	BUG_ON(!child);
 70
 71	if (list_empty(&child->devices))
 72		pci_remove_bus(child);
 73}
 74
 75static int virtfn_add(struct pci_dev *dev, int id, int reset)
 76{
 77	int i;
 78	int rc;
 79	u64 size;
 80	char buf[VIRTFN_ID_LEN];
 81	struct pci_dev *virtfn;
 82	struct resource *res;
 83	struct pci_sriov *iov = dev->sriov;
 84
 85	virtfn = alloc_pci_dev();
 86	if (!virtfn)
 87		return -ENOMEM;
 88
 89	mutex_lock(&iov->dev->sriov->lock);
 90	virtfn->bus = virtfn_add_bus(dev->bus, virtfn_bus(dev, id));
 91	if (!virtfn->bus) {
 92		kfree(virtfn);
 93		mutex_unlock(&iov->dev->sriov->lock);
 94		return -ENOMEM;
 95	}
 96	virtfn->devfn = virtfn_devfn(dev, id);
 97	virtfn->vendor = dev->vendor;
 98	pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_DID, &virtfn->device);
 99	pci_setup_device(virtfn);
100	virtfn->dev.parent = dev->dev.parent;
101
102	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
103		res = dev->resource + PCI_IOV_RESOURCES + i;
104		if (!res->parent)
105			continue;
106		virtfn->resource[i].name = pci_name(virtfn);
107		virtfn->resource[i].flags = res->flags;
108		size = resource_size(res);
109		do_div(size, iov->total);
110		virtfn->resource[i].start = res->start + size * id;
111		virtfn->resource[i].end = virtfn->resource[i].start + size - 1;
112		rc = request_resource(res, &virtfn->resource[i]);
113		BUG_ON(rc);
114	}
115
116	if (reset)
117		__pci_reset_function(virtfn);
118
119	pci_device_add(virtfn, virtfn->bus);
120	mutex_unlock(&iov->dev->sriov->lock);
121
122	virtfn->physfn = pci_dev_get(dev);
123	virtfn->is_virtfn = 1;
124
125	rc = pci_bus_add_device(virtfn);
126	if (rc)
127		goto failed1;
128	sprintf(buf, "virtfn%u", id);
129	rc = sysfs_create_link(&dev->dev.kobj, &virtfn->dev.kobj, buf);
130	if (rc)
131		goto failed1;
132	rc = sysfs_create_link(&virtfn->dev.kobj, &dev->dev.kobj, "physfn");
133	if (rc)
134		goto failed2;
135
136	kobject_uevent(&virtfn->dev.kobj, KOBJ_CHANGE);
137
138	return 0;
139
140failed2:
141	sysfs_remove_link(&dev->dev.kobj, buf);
142failed1:
143	pci_dev_put(dev);
144	mutex_lock(&iov->dev->sriov->lock);
145	pci_stop_and_remove_bus_device(virtfn);
146	virtfn_remove_bus(dev->bus, virtfn_bus(dev, id));
147	mutex_unlock(&iov->dev->sriov->lock);
148
149	return rc;
150}
151
152static void virtfn_remove(struct pci_dev *dev, int id, int reset)
153{
154	char buf[VIRTFN_ID_LEN];
155	struct pci_bus *bus;
156	struct pci_dev *virtfn;
157	struct pci_sriov *iov = dev->sriov;
158
159	bus = pci_find_bus(pci_domain_nr(dev->bus), virtfn_bus(dev, id));
160	if (!bus)
161		return;
162
163	virtfn = pci_get_slot(bus, virtfn_devfn(dev, id));
164	if (!virtfn)
165		return;
166
167	pci_dev_put(virtfn);
168
169	if (reset) {
170		device_release_driver(&virtfn->dev);
171		__pci_reset_function(virtfn);
172	}
173
174	sprintf(buf, "virtfn%u", id);
175	sysfs_remove_link(&dev->dev.kobj, buf);
176	/*
177	 * pci_stop_dev() could have been called for this virtfn already,
178	 * so the directory for the virtfn may have been removed before.
179	 * Double check to avoid spurious sysfs warnings.
180	 */
181	if (virtfn->dev.kobj.sd)
182		sysfs_remove_link(&virtfn->dev.kobj, "physfn");
183
184	mutex_lock(&iov->dev->sriov->lock);
185	pci_stop_and_remove_bus_device(virtfn);
186	virtfn_remove_bus(dev->bus, virtfn_bus(dev, id));
187	mutex_unlock(&iov->dev->sriov->lock);
188
189	pci_dev_put(dev);
190}
191
192static int sriov_migration(struct pci_dev *dev)
193{
194	u16 status;
195	struct pci_sriov *iov = dev->sriov;
196
197	if (!iov->nr_virtfn)
198		return 0;
199
200	if (!(iov->cap & PCI_SRIOV_CAP_VFM))
201		return 0;
202
203	pci_read_config_word(dev, iov->pos + PCI_SRIOV_STATUS, &status);
204	if (!(status & PCI_SRIOV_STATUS_VFM))
205		return 0;
206
207	schedule_work(&iov->mtask);
208
209	return 1;
210}
211
212static void sriov_migration_task(struct work_struct *work)
213{
214	int i;
215	u8 state;
216	u16 status;
217	struct pci_sriov *iov = container_of(work, struct pci_sriov, mtask);
218
219	for (i = iov->initial; i < iov->nr_virtfn; i++) {
220		state = readb(iov->mstate + i);
221		if (state == PCI_SRIOV_VFM_MI) {
222			writeb(PCI_SRIOV_VFM_AV, iov->mstate + i);
223			state = readb(iov->mstate + i);
224			if (state == PCI_SRIOV_VFM_AV)
225				virtfn_add(iov->self, i, 1);
226		} else if (state == PCI_SRIOV_VFM_MO) {
227			virtfn_remove(iov->self, i, 1);
228			writeb(PCI_SRIOV_VFM_UA, iov->mstate + i);
229			state = readb(iov->mstate + i);
230			if (state == PCI_SRIOV_VFM_AV)
231				virtfn_add(iov->self, i, 0);
232		}
233	}
234
235	pci_read_config_word(iov->self, iov->pos + PCI_SRIOV_STATUS, &status);
236	status &= ~PCI_SRIOV_STATUS_VFM;
237	pci_write_config_word(iov->self, iov->pos + PCI_SRIOV_STATUS, status);
238}
239
240static int sriov_enable_migration(struct pci_dev *dev, int nr_virtfn)
241{
242	int bir;
243	u32 table;
244	resource_size_t pa;
245	struct pci_sriov *iov = dev->sriov;
246
247	if (nr_virtfn <= iov->initial)
248		return 0;
249
250	pci_read_config_dword(dev, iov->pos + PCI_SRIOV_VFM, &table);
251	bir = PCI_SRIOV_VFM_BIR(table);
252	if (bir > PCI_STD_RESOURCE_END)
253		return -EIO;
254
255	table = PCI_SRIOV_VFM_OFFSET(table);
256	if (table + nr_virtfn > pci_resource_len(dev, bir))
257		return -EIO;
258
259	pa = pci_resource_start(dev, bir) + table;
260	iov->mstate = ioremap(pa, nr_virtfn);
261	if (!iov->mstate)
262		return -ENOMEM;
263
264	INIT_WORK(&iov->mtask, sriov_migration_task);
265
266	iov->ctrl |= PCI_SRIOV_CTRL_VFM | PCI_SRIOV_CTRL_INTR;
267	pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
268
269	return 0;
270}
271
272static void sriov_disable_migration(struct pci_dev *dev)
273{
274	struct pci_sriov *iov = dev->sriov;
275
276	iov->ctrl &= ~(PCI_SRIOV_CTRL_VFM | PCI_SRIOV_CTRL_INTR);
277	pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
278
279	cancel_work_sync(&iov->mtask);
280	iounmap(iov->mstate);
281}
282
283static int sriov_enable(struct pci_dev *dev, int nr_virtfn)
284{
285	int rc;
286	int i, j;
287	int nres;
288	u16 offset, stride, initial;
289	struct resource *res;
290	struct pci_dev *pdev;
291	struct pci_sriov *iov = dev->sriov;
292	int bars = 0;
293
294	if (!nr_virtfn)
295		return 0;
296
297	if (iov->nr_virtfn)
298		return -EINVAL;
299
300	pci_read_config_word(dev, iov->pos + PCI_SRIOV_INITIAL_VF, &initial);
301	if (initial > iov->total ||
302	    (!(iov->cap & PCI_SRIOV_CAP_VFM) && (initial != iov->total)))
303		return -EIO;
304
305	if (nr_virtfn < 0 || nr_virtfn > iov->total ||
306	    (!(iov->cap & PCI_SRIOV_CAP_VFM) && (nr_virtfn > initial)))
307		return -EINVAL;
308
309	pci_write_config_word(dev, iov->pos + PCI_SRIOV_NUM_VF, nr_virtfn);
310	pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_OFFSET, &offset);
311	pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_STRIDE, &stride);
312	if (!offset || (nr_virtfn > 1 && !stride))
313		return -EIO;
314
315	nres = 0;
316	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
317		bars |= (1 << (i + PCI_IOV_RESOURCES));
318		res = dev->resource + PCI_IOV_RESOURCES + i;
319		if (res->parent)
320			nres++;
321	}
322	if (nres != iov->nres) {
323		dev_err(&dev->dev, "not enough MMIO resources for SR-IOV\n");
324		return -ENOMEM;
325	}
326
327	iov->offset = offset;
328	iov->stride = stride;
329
330	if (virtfn_bus(dev, nr_virtfn - 1) > dev->bus->subordinate) {
331		dev_err(&dev->dev, "SR-IOV: bus number out of range\n");
332		return -ENOMEM;
333	}
334
335	if (pci_enable_resources(dev, bars)) {
336		dev_err(&dev->dev, "SR-IOV: IOV BARS not allocated\n");
337		return -ENOMEM;
338	}
339
340	if (iov->link != dev->devfn) {
341		pdev = pci_get_slot(dev->bus, iov->link);
342		if (!pdev)
343			return -ENODEV;
344
345		pci_dev_put(pdev);
346
347		if (!pdev->is_physfn)
348			return -ENODEV;
349
350		rc = sysfs_create_link(&dev->dev.kobj,
351					&pdev->dev.kobj, "dep_link");
352		if (rc)
353			return rc;
354	}
355
356	iov->ctrl |= PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE;
357	pci_cfg_access_lock(dev);
358	pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
359	msleep(100);
360	pci_cfg_access_unlock(dev);
361
362	iov->initial = initial;
363	if (nr_virtfn < initial)
364		initial = nr_virtfn;
365
366	for (i = 0; i < initial; i++) {
367		rc = virtfn_add(dev, i, 0);
368		if (rc)
369			goto failed;
370	}
371
372	if (iov->cap & PCI_SRIOV_CAP_VFM) {
373		rc = sriov_enable_migration(dev, nr_virtfn);
374		if (rc)
375			goto failed;
376	}
377
378	kobject_uevent(&dev->dev.kobj, KOBJ_CHANGE);
379	iov->nr_virtfn = nr_virtfn;
380
381	return 0;
382
383failed:
384	for (j = 0; j < i; j++)
385		virtfn_remove(dev, j, 0);
386
387	iov->ctrl &= ~(PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE);
388	pci_cfg_access_lock(dev);
389	pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
390	ssleep(1);
391	pci_cfg_access_unlock(dev);
392
393	if (iov->link != dev->devfn)
394		sysfs_remove_link(&dev->dev.kobj, "dep_link");
395
396	return rc;
397}
398
399static void sriov_disable(struct pci_dev *dev)
400{
401	int i;
402	struct pci_sriov *iov = dev->sriov;
403
404	if (!iov->nr_virtfn)
405		return;
406
407	if (iov->cap & PCI_SRIOV_CAP_VFM)
408		sriov_disable_migration(dev);
409
410	for (i = 0; i < iov->nr_virtfn; i++)
411		virtfn_remove(dev, i, 0);
412
413	iov->ctrl &= ~(PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE);
414	pci_cfg_access_lock(dev);
415	pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
416	ssleep(1);
417	pci_cfg_access_unlock(dev);
418
419	if (iov->link != dev->devfn)
420		sysfs_remove_link(&dev->dev.kobj, "dep_link");
421
422	iov->nr_virtfn = 0;
423}
424
425static int sriov_init(struct pci_dev *dev, int pos)
426{
427	int i;
428	int rc;
429	int nres;
430	u32 pgsz;
431	u16 ctrl, total, offset, stride;
432	struct pci_sriov *iov;
433	struct resource *res;
434	struct pci_dev *pdev;
435
436	if (dev->pcie_type != PCI_EXP_TYPE_RC_END &&
437	    dev->pcie_type != PCI_EXP_TYPE_ENDPOINT)
438		return -ENODEV;
439
440	pci_read_config_word(dev, pos + PCI_SRIOV_CTRL, &ctrl);
441	if (ctrl & PCI_SRIOV_CTRL_VFE) {
442		pci_write_config_word(dev, pos + PCI_SRIOV_CTRL, 0);
443		ssleep(1);
444	}
445
446	pci_read_config_word(dev, pos + PCI_SRIOV_TOTAL_VF, &total);
447	if (!total)
448		return 0;
449
450	ctrl = 0;
451	list_for_each_entry(pdev, &dev->bus->devices, bus_list)
452		if (pdev->is_physfn)
453			goto found;
454
455	pdev = NULL;
456	if (pci_ari_enabled(dev->bus))
457		ctrl |= PCI_SRIOV_CTRL_ARI;
458
459found:
460	pci_write_config_word(dev, pos + PCI_SRIOV_CTRL, ctrl);
 
461	pci_read_config_word(dev, pos + PCI_SRIOV_VF_OFFSET, &offset);
462	pci_read_config_word(dev, pos + PCI_SRIOV_VF_STRIDE, &stride);
463	if (!offset || (total > 1 && !stride))
464		return -EIO;
465
466	pci_read_config_dword(dev, pos + PCI_SRIOV_SUP_PGSIZE, &pgsz);
467	i = PAGE_SHIFT > 12 ? PAGE_SHIFT - 12 : 0;
468	pgsz &= ~((1 << i) - 1);
469	if (!pgsz)
470		return -EIO;
471
472	pgsz &= ~(pgsz - 1);
473	pci_write_config_dword(dev, pos + PCI_SRIOV_SYS_PGSIZE, pgsz);
474
475	nres = 0;
476	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
477		res = dev->resource + PCI_IOV_RESOURCES + i;
478		i += __pci_read_base(dev, pci_bar_unknown, res,
479				     pos + PCI_SRIOV_BAR + i * 4);
480		if (!res->flags)
481			continue;
482		if (resource_size(res) & (PAGE_SIZE - 1)) {
483			rc = -EIO;
484			goto failed;
485		}
486		res->end = res->start + resource_size(res) * total - 1;
487		nres++;
488	}
489
490	iov = kzalloc(sizeof(*iov), GFP_KERNEL);
491	if (!iov) {
492		rc = -ENOMEM;
493		goto failed;
494	}
495
496	iov->pos = pos;
497	iov->nres = nres;
498	iov->ctrl = ctrl;
499	iov->total = total;
500	iov->offset = offset;
501	iov->stride = stride;
502	iov->pgsz = pgsz;
503	iov->self = dev;
504	pci_read_config_dword(dev, pos + PCI_SRIOV_CAP, &iov->cap);
505	pci_read_config_byte(dev, pos + PCI_SRIOV_FUNC_LINK, &iov->link);
506	if (dev->pcie_type == PCI_EXP_TYPE_RC_END)
507		iov->link = PCI_DEVFN(PCI_SLOT(dev->devfn), iov->link);
508
509	if (pdev)
510		iov->dev = pci_dev_get(pdev);
511	else
512		iov->dev = dev;
513
514	mutex_init(&iov->lock);
515
516	dev->sriov = iov;
517	dev->is_physfn = 1;
518
519	return 0;
520
521failed:
522	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
523		res = dev->resource + PCI_IOV_RESOURCES + i;
524		res->flags = 0;
525	}
526
527	return rc;
528}
529
530static void sriov_release(struct pci_dev *dev)
531{
532	BUG_ON(dev->sriov->nr_virtfn);
533
534	if (dev != dev->sriov->dev)
535		pci_dev_put(dev->sriov->dev);
536
537	mutex_destroy(&dev->sriov->lock);
538
539	kfree(dev->sriov);
540	dev->sriov = NULL;
541}
542
543static void sriov_restore_state(struct pci_dev *dev)
544{
545	int i;
546	u16 ctrl;
547	struct pci_sriov *iov = dev->sriov;
548
549	pci_read_config_word(dev, iov->pos + PCI_SRIOV_CTRL, &ctrl);
550	if (ctrl & PCI_SRIOV_CTRL_VFE)
551		return;
552
553	for (i = PCI_IOV_RESOURCES; i <= PCI_IOV_RESOURCE_END; i++)
554		pci_update_resource(dev, i);
555
556	pci_write_config_dword(dev, iov->pos + PCI_SRIOV_SYS_PGSIZE, iov->pgsz);
557	pci_write_config_word(dev, iov->pos + PCI_SRIOV_NUM_VF, iov->nr_virtfn);
558	pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
559	if (iov->ctrl & PCI_SRIOV_CTRL_VFE)
560		msleep(100);
561}
562
563/**
564 * pci_iov_init - initialize the IOV capability
565 * @dev: the PCI device
566 *
567 * Returns 0 on success, or negative on failure.
568 */
569int pci_iov_init(struct pci_dev *dev)
570{
571	int pos;
572
573	if (!pci_is_pcie(dev))
574		return -ENODEV;
575
576	pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_SRIOV);
577	if (pos)
578		return sriov_init(dev, pos);
579
580	return -ENODEV;
581}
582
583/**
584 * pci_iov_release - release resources used by the IOV capability
585 * @dev: the PCI device
586 */
587void pci_iov_release(struct pci_dev *dev)
588{
589	if (dev->is_physfn)
590		sriov_release(dev);
591}
592
593/**
594 * pci_iov_resource_bar - get position of the SR-IOV BAR
595 * @dev: the PCI device
596 * @resno: the resource number
597 * @type: the BAR type to be filled in
598 *
599 * Returns position of the BAR encapsulated in the SR-IOV capability.
600 */
601int pci_iov_resource_bar(struct pci_dev *dev, int resno,
602			 enum pci_bar_type *type)
603{
604	if (resno < PCI_IOV_RESOURCES || resno > PCI_IOV_RESOURCE_END)
605		return 0;
606
607	BUG_ON(!dev->is_physfn);
608
609	*type = pci_bar_unknown;
610
611	return dev->sriov->pos + PCI_SRIOV_BAR +
612		4 * (resno - PCI_IOV_RESOURCES);
613}
614
615/**
616 * pci_sriov_resource_alignment - get resource alignment for VF BAR
617 * @dev: the PCI device
618 * @resno: the resource number
619 *
620 * Returns the alignment of the VF BAR found in the SR-IOV capability.
621 * This is not the same as the resource size which is defined as
622 * the VF BAR size multiplied by the number of VFs.  The alignment
623 * is just the VF BAR size.
624 */
625resource_size_t pci_sriov_resource_alignment(struct pci_dev *dev, int resno)
626{
627	struct resource tmp;
628	enum pci_bar_type type;
629	int reg = pci_iov_resource_bar(dev, resno, &type);
630	
631	if (!reg)
632		return 0;
633
634	 __pci_read_base(dev, type, &tmp, reg);
635	return resource_alignment(&tmp);
636}
637
638/**
639 * pci_restore_iov_state - restore the state of the IOV capability
640 * @dev: the PCI device
641 */
642void pci_restore_iov_state(struct pci_dev *dev)
643{
644	if (dev->is_physfn)
645		sriov_restore_state(dev);
646}
647
648/**
649 * pci_iov_bus_range - find bus range used by Virtual Function
650 * @bus: the PCI bus
651 *
652 * Returns max number of buses (exclude current one) used by Virtual
653 * Functions.
654 */
655int pci_iov_bus_range(struct pci_bus *bus)
656{
657	int max = 0;
658	u8 busnr;
659	struct pci_dev *dev;
660
661	list_for_each_entry(dev, &bus->devices, bus_list) {
662		if (!dev->is_physfn)
663			continue;
664		busnr = virtfn_bus(dev, dev->sriov->total - 1);
665		if (busnr > max)
666			max = busnr;
667	}
668
669	return max ? max - bus->number : 0;
670}
671
672/**
673 * pci_enable_sriov - enable the SR-IOV capability
674 * @dev: the PCI device
675 * @nr_virtfn: number of virtual functions to enable
676 *
677 * Returns 0 on success, or negative on failure.
678 */
679int pci_enable_sriov(struct pci_dev *dev, int nr_virtfn)
680{
681	might_sleep();
682
683	if (!dev->is_physfn)
684		return -ENODEV;
685
686	return sriov_enable(dev, nr_virtfn);
687}
688EXPORT_SYMBOL_GPL(pci_enable_sriov);
689
690/**
691 * pci_disable_sriov - disable the SR-IOV capability
692 * @dev: the PCI device
693 */
694void pci_disable_sriov(struct pci_dev *dev)
695{
696	might_sleep();
697
698	if (!dev->is_physfn)
699		return;
700
701	sriov_disable(dev);
702}
703EXPORT_SYMBOL_GPL(pci_disable_sriov);
704
705/**
706 * pci_sriov_migration - notify SR-IOV core of Virtual Function Migration
707 * @dev: the PCI device
708 *
709 * Returns IRQ_HANDLED if the IRQ is handled, or IRQ_NONE if not.
710 *
711 * Physical Function driver is responsible to register IRQ handler using
712 * VF Migration Interrupt Message Number, and call this function when the
713 * interrupt is generated by the hardware.
714 */
715irqreturn_t pci_sriov_migration(struct pci_dev *dev)
716{
717	if (!dev->is_physfn)
718		return IRQ_NONE;
719
720	return sriov_migration(dev) ? IRQ_HANDLED : IRQ_NONE;
721}
722EXPORT_SYMBOL_GPL(pci_sriov_migration);
723
724/**
725 * pci_num_vf - return number of VFs associated with a PF device_release_driver
726 * @dev: the PCI device
727 *
728 * Returns number of VFs, or 0 if SR-IOV is not enabled.
729 */
730int pci_num_vf(struct pci_dev *dev)
731{
732	if (!dev || !dev->is_physfn)
733		return 0;
734	else
735		return dev->sriov->nr_virtfn;
736}
737EXPORT_SYMBOL_GPL(pci_num_vf);
v3.1
  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/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
 21static inline u8 virtfn_bus(struct pci_dev *dev, int id)
 22{
 23	return dev->bus->number + ((dev->devfn + dev->sriov->offset +
 24				    dev->sriov->stride * id) >> 8);
 25}
 26
 27static inline u8 virtfn_devfn(struct pci_dev *dev, int id)
 28{
 29	return (dev->devfn + dev->sriov->offset +
 30		dev->sriov->stride * id) & 0xff;
 31}
 32
 33static struct pci_bus *virtfn_add_bus(struct pci_bus *bus, int busnr)
 34{
 35	int rc;
 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	child->subordinate = busnr;
 50	child->dev.parent = bus->bridge;
 51	rc = pci_bus_add_child(child);
 52	if (rc) {
 53		pci_remove_bus(child);
 54		return NULL;
 55	}
 56
 57	return child;
 58}
 59
 60static void virtfn_remove_bus(struct pci_bus *bus, int busnr)
 61{
 62	struct pci_bus *child;
 63
 64	if (bus->number == busnr)
 65		return;
 66
 67	child = pci_find_bus(pci_domain_nr(bus), busnr);
 68	BUG_ON(!child);
 69
 70	if (list_empty(&child->devices))
 71		pci_remove_bus(child);
 72}
 73
 74static int virtfn_add(struct pci_dev *dev, int id, int reset)
 75{
 76	int i;
 77	int rc;
 78	u64 size;
 79	char buf[VIRTFN_ID_LEN];
 80	struct pci_dev *virtfn;
 81	struct resource *res;
 82	struct pci_sriov *iov = dev->sriov;
 83
 84	virtfn = alloc_pci_dev();
 85	if (!virtfn)
 86		return -ENOMEM;
 87
 88	mutex_lock(&iov->dev->sriov->lock);
 89	virtfn->bus = virtfn_add_bus(dev->bus, virtfn_bus(dev, id));
 90	if (!virtfn->bus) {
 91		kfree(virtfn);
 92		mutex_unlock(&iov->dev->sriov->lock);
 93		return -ENOMEM;
 94	}
 95	virtfn->devfn = virtfn_devfn(dev, id);
 96	virtfn->vendor = dev->vendor;
 97	pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_DID, &virtfn->device);
 98	pci_setup_device(virtfn);
 99	virtfn->dev.parent = dev->dev.parent;
100
101	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
102		res = dev->resource + PCI_IOV_RESOURCES + i;
103		if (!res->parent)
104			continue;
105		virtfn->resource[i].name = pci_name(virtfn);
106		virtfn->resource[i].flags = res->flags;
107		size = resource_size(res);
108		do_div(size, iov->total);
109		virtfn->resource[i].start = res->start + size * id;
110		virtfn->resource[i].end = virtfn->resource[i].start + size - 1;
111		rc = request_resource(res, &virtfn->resource[i]);
112		BUG_ON(rc);
113	}
114
115	if (reset)
116		__pci_reset_function(virtfn);
117
118	pci_device_add(virtfn, virtfn->bus);
119	mutex_unlock(&iov->dev->sriov->lock);
120
121	virtfn->physfn = pci_dev_get(dev);
122	virtfn->is_virtfn = 1;
123
124	rc = pci_bus_add_device(virtfn);
125	if (rc)
126		goto failed1;
127	sprintf(buf, "virtfn%u", id);
128	rc = sysfs_create_link(&dev->dev.kobj, &virtfn->dev.kobj, buf);
129	if (rc)
130		goto failed1;
131	rc = sysfs_create_link(&virtfn->dev.kobj, &dev->dev.kobj, "physfn");
132	if (rc)
133		goto failed2;
134
135	kobject_uevent(&virtfn->dev.kobj, KOBJ_CHANGE);
136
137	return 0;
138
139failed2:
140	sysfs_remove_link(&dev->dev.kobj, buf);
141failed1:
142	pci_dev_put(dev);
143	mutex_lock(&iov->dev->sriov->lock);
144	pci_remove_bus_device(virtfn);
145	virtfn_remove_bus(dev->bus, virtfn_bus(dev, id));
146	mutex_unlock(&iov->dev->sriov->lock);
147
148	return rc;
149}
150
151static void virtfn_remove(struct pci_dev *dev, int id, int reset)
152{
153	char buf[VIRTFN_ID_LEN];
154	struct pci_bus *bus;
155	struct pci_dev *virtfn;
156	struct pci_sriov *iov = dev->sriov;
157
158	bus = pci_find_bus(pci_domain_nr(dev->bus), virtfn_bus(dev, id));
159	if (!bus)
160		return;
161
162	virtfn = pci_get_slot(bus, virtfn_devfn(dev, id));
163	if (!virtfn)
164		return;
165
166	pci_dev_put(virtfn);
167
168	if (reset) {
169		device_release_driver(&virtfn->dev);
170		__pci_reset_function(virtfn);
171	}
172
173	sprintf(buf, "virtfn%u", id);
174	sysfs_remove_link(&dev->dev.kobj, buf);
175	sysfs_remove_link(&virtfn->dev.kobj, "physfn");
 
 
 
 
 
 
176
177	mutex_lock(&iov->dev->sriov->lock);
178	pci_remove_bus_device(virtfn);
179	virtfn_remove_bus(dev->bus, virtfn_bus(dev, id));
180	mutex_unlock(&iov->dev->sriov->lock);
181
182	pci_dev_put(dev);
183}
184
185static int sriov_migration(struct pci_dev *dev)
186{
187	u16 status;
188	struct pci_sriov *iov = dev->sriov;
189
190	if (!iov->nr_virtfn)
191		return 0;
192
193	if (!(iov->cap & PCI_SRIOV_CAP_VFM))
194		return 0;
195
196	pci_read_config_word(dev, iov->pos + PCI_SRIOV_STATUS, &status);
197	if (!(status & PCI_SRIOV_STATUS_VFM))
198		return 0;
199
200	schedule_work(&iov->mtask);
201
202	return 1;
203}
204
205static void sriov_migration_task(struct work_struct *work)
206{
207	int i;
208	u8 state;
209	u16 status;
210	struct pci_sriov *iov = container_of(work, struct pci_sriov, mtask);
211
212	for (i = iov->initial; i < iov->nr_virtfn; i++) {
213		state = readb(iov->mstate + i);
214		if (state == PCI_SRIOV_VFM_MI) {
215			writeb(PCI_SRIOV_VFM_AV, iov->mstate + i);
216			state = readb(iov->mstate + i);
217			if (state == PCI_SRIOV_VFM_AV)
218				virtfn_add(iov->self, i, 1);
219		} else if (state == PCI_SRIOV_VFM_MO) {
220			virtfn_remove(iov->self, i, 1);
221			writeb(PCI_SRIOV_VFM_UA, iov->mstate + i);
222			state = readb(iov->mstate + i);
223			if (state == PCI_SRIOV_VFM_AV)
224				virtfn_add(iov->self, i, 0);
225		}
226	}
227
228	pci_read_config_word(iov->self, iov->pos + PCI_SRIOV_STATUS, &status);
229	status &= ~PCI_SRIOV_STATUS_VFM;
230	pci_write_config_word(iov->self, iov->pos + PCI_SRIOV_STATUS, status);
231}
232
233static int sriov_enable_migration(struct pci_dev *dev, int nr_virtfn)
234{
235	int bir;
236	u32 table;
237	resource_size_t pa;
238	struct pci_sriov *iov = dev->sriov;
239
240	if (nr_virtfn <= iov->initial)
241		return 0;
242
243	pci_read_config_dword(dev, iov->pos + PCI_SRIOV_VFM, &table);
244	bir = PCI_SRIOV_VFM_BIR(table);
245	if (bir > PCI_STD_RESOURCE_END)
246		return -EIO;
247
248	table = PCI_SRIOV_VFM_OFFSET(table);
249	if (table + nr_virtfn > pci_resource_len(dev, bir))
250		return -EIO;
251
252	pa = pci_resource_start(dev, bir) + table;
253	iov->mstate = ioremap(pa, nr_virtfn);
254	if (!iov->mstate)
255		return -ENOMEM;
256
257	INIT_WORK(&iov->mtask, sriov_migration_task);
258
259	iov->ctrl |= PCI_SRIOV_CTRL_VFM | PCI_SRIOV_CTRL_INTR;
260	pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
261
262	return 0;
263}
264
265static void sriov_disable_migration(struct pci_dev *dev)
266{
267	struct pci_sriov *iov = dev->sriov;
268
269	iov->ctrl &= ~(PCI_SRIOV_CTRL_VFM | PCI_SRIOV_CTRL_INTR);
270	pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
271
272	cancel_work_sync(&iov->mtask);
273	iounmap(iov->mstate);
274}
275
276static int sriov_enable(struct pci_dev *dev, int nr_virtfn)
277{
278	int rc;
279	int i, j;
280	int nres;
281	u16 offset, stride, initial;
282	struct resource *res;
283	struct pci_dev *pdev;
284	struct pci_sriov *iov = dev->sriov;
 
285
286	if (!nr_virtfn)
287		return 0;
288
289	if (iov->nr_virtfn)
290		return -EINVAL;
291
292	pci_read_config_word(dev, iov->pos + PCI_SRIOV_INITIAL_VF, &initial);
293	if (initial > iov->total ||
294	    (!(iov->cap & PCI_SRIOV_CAP_VFM) && (initial != iov->total)))
295		return -EIO;
296
297	if (nr_virtfn < 0 || nr_virtfn > iov->total ||
298	    (!(iov->cap & PCI_SRIOV_CAP_VFM) && (nr_virtfn > initial)))
299		return -EINVAL;
300
301	pci_write_config_word(dev, iov->pos + PCI_SRIOV_NUM_VF, nr_virtfn);
302	pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_OFFSET, &offset);
303	pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_STRIDE, &stride);
304	if (!offset || (nr_virtfn > 1 && !stride))
305		return -EIO;
306
307	nres = 0;
308	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
 
309		res = dev->resource + PCI_IOV_RESOURCES + i;
310		if (res->parent)
311			nres++;
312	}
313	if (nres != iov->nres) {
314		dev_err(&dev->dev, "not enough MMIO resources for SR-IOV\n");
315		return -ENOMEM;
316	}
317
318	iov->offset = offset;
319	iov->stride = stride;
320
321	if (virtfn_bus(dev, nr_virtfn - 1) > dev->bus->subordinate) {
322		dev_err(&dev->dev, "SR-IOV: bus number out of range\n");
323		return -ENOMEM;
324	}
325
 
 
 
 
 
326	if (iov->link != dev->devfn) {
327		pdev = pci_get_slot(dev->bus, iov->link);
328		if (!pdev)
329			return -ENODEV;
330
331		pci_dev_put(pdev);
332
333		if (!pdev->is_physfn)
334			return -ENODEV;
335
336		rc = sysfs_create_link(&dev->dev.kobj,
337					&pdev->dev.kobj, "dep_link");
338		if (rc)
339			return rc;
340	}
341
342	iov->ctrl |= PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE;
343	pci_block_user_cfg_access(dev);
344	pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
345	msleep(100);
346	pci_unblock_user_cfg_access(dev);
347
348	iov->initial = initial;
349	if (nr_virtfn < initial)
350		initial = nr_virtfn;
351
352	for (i = 0; i < initial; i++) {
353		rc = virtfn_add(dev, i, 0);
354		if (rc)
355			goto failed;
356	}
357
358	if (iov->cap & PCI_SRIOV_CAP_VFM) {
359		rc = sriov_enable_migration(dev, nr_virtfn);
360		if (rc)
361			goto failed;
362	}
363
364	kobject_uevent(&dev->dev.kobj, KOBJ_CHANGE);
365	iov->nr_virtfn = nr_virtfn;
366
367	return 0;
368
369failed:
370	for (j = 0; j < i; j++)
371		virtfn_remove(dev, j, 0);
372
373	iov->ctrl &= ~(PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE);
374	pci_block_user_cfg_access(dev);
375	pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
376	ssleep(1);
377	pci_unblock_user_cfg_access(dev);
378
379	if (iov->link != dev->devfn)
380		sysfs_remove_link(&dev->dev.kobj, "dep_link");
381
382	return rc;
383}
384
385static void sriov_disable(struct pci_dev *dev)
386{
387	int i;
388	struct pci_sriov *iov = dev->sriov;
389
390	if (!iov->nr_virtfn)
391		return;
392
393	if (iov->cap & PCI_SRIOV_CAP_VFM)
394		sriov_disable_migration(dev);
395
396	for (i = 0; i < iov->nr_virtfn; i++)
397		virtfn_remove(dev, i, 0);
398
399	iov->ctrl &= ~(PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE);
400	pci_block_user_cfg_access(dev);
401	pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
402	ssleep(1);
403	pci_unblock_user_cfg_access(dev);
404
405	if (iov->link != dev->devfn)
406		sysfs_remove_link(&dev->dev.kobj, "dep_link");
407
408	iov->nr_virtfn = 0;
409}
410
411static int sriov_init(struct pci_dev *dev, int pos)
412{
413	int i;
414	int rc;
415	int nres;
416	u32 pgsz;
417	u16 ctrl, total, offset, stride;
418	struct pci_sriov *iov;
419	struct resource *res;
420	struct pci_dev *pdev;
421
422	if (dev->pcie_type != PCI_EXP_TYPE_RC_END &&
423	    dev->pcie_type != PCI_EXP_TYPE_ENDPOINT)
424		return -ENODEV;
425
426	pci_read_config_word(dev, pos + PCI_SRIOV_CTRL, &ctrl);
427	if (ctrl & PCI_SRIOV_CTRL_VFE) {
428		pci_write_config_word(dev, pos + PCI_SRIOV_CTRL, 0);
429		ssleep(1);
430	}
431
432	pci_read_config_word(dev, pos + PCI_SRIOV_TOTAL_VF, &total);
433	if (!total)
434		return 0;
435
436	ctrl = 0;
437	list_for_each_entry(pdev, &dev->bus->devices, bus_list)
438		if (pdev->is_physfn)
439			goto found;
440
441	pdev = NULL;
442	if (pci_ari_enabled(dev->bus))
443		ctrl |= PCI_SRIOV_CTRL_ARI;
444
445found:
446	pci_write_config_word(dev, pos + PCI_SRIOV_CTRL, ctrl);
447	pci_write_config_word(dev, pos + PCI_SRIOV_NUM_VF, total);
448	pci_read_config_word(dev, pos + PCI_SRIOV_VF_OFFSET, &offset);
449	pci_read_config_word(dev, pos + PCI_SRIOV_VF_STRIDE, &stride);
450	if (!offset || (total > 1 && !stride))
451		return -EIO;
452
453	pci_read_config_dword(dev, pos + PCI_SRIOV_SUP_PGSIZE, &pgsz);
454	i = PAGE_SHIFT > 12 ? PAGE_SHIFT - 12 : 0;
455	pgsz &= ~((1 << i) - 1);
456	if (!pgsz)
457		return -EIO;
458
459	pgsz &= ~(pgsz - 1);
460	pci_write_config_dword(dev, pos + PCI_SRIOV_SYS_PGSIZE, pgsz);
461
462	nres = 0;
463	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
464		res = dev->resource + PCI_IOV_RESOURCES + i;
465		i += __pci_read_base(dev, pci_bar_unknown, res,
466				     pos + PCI_SRIOV_BAR + i * 4);
467		if (!res->flags)
468			continue;
469		if (resource_size(res) & (PAGE_SIZE - 1)) {
470			rc = -EIO;
471			goto failed;
472		}
473		res->end = res->start + resource_size(res) * total - 1;
474		nres++;
475	}
476
477	iov = kzalloc(sizeof(*iov), GFP_KERNEL);
478	if (!iov) {
479		rc = -ENOMEM;
480		goto failed;
481	}
482
483	iov->pos = pos;
484	iov->nres = nres;
485	iov->ctrl = ctrl;
486	iov->total = total;
487	iov->offset = offset;
488	iov->stride = stride;
489	iov->pgsz = pgsz;
490	iov->self = dev;
491	pci_read_config_dword(dev, pos + PCI_SRIOV_CAP, &iov->cap);
492	pci_read_config_byte(dev, pos + PCI_SRIOV_FUNC_LINK, &iov->link);
493	if (dev->pcie_type == PCI_EXP_TYPE_RC_END)
494		iov->link = PCI_DEVFN(PCI_SLOT(dev->devfn), iov->link);
495
496	if (pdev)
497		iov->dev = pci_dev_get(pdev);
498	else
499		iov->dev = dev;
500
501	mutex_init(&iov->lock);
502
503	dev->sriov = iov;
504	dev->is_physfn = 1;
505
506	return 0;
507
508failed:
509	for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
510		res = dev->resource + PCI_IOV_RESOURCES + i;
511		res->flags = 0;
512	}
513
514	return rc;
515}
516
517static void sriov_release(struct pci_dev *dev)
518{
519	BUG_ON(dev->sriov->nr_virtfn);
520
521	if (dev != dev->sriov->dev)
522		pci_dev_put(dev->sriov->dev);
523
524	mutex_destroy(&dev->sriov->lock);
525
526	kfree(dev->sriov);
527	dev->sriov = NULL;
528}
529
530static void sriov_restore_state(struct pci_dev *dev)
531{
532	int i;
533	u16 ctrl;
534	struct pci_sriov *iov = dev->sriov;
535
536	pci_read_config_word(dev, iov->pos + PCI_SRIOV_CTRL, &ctrl);
537	if (ctrl & PCI_SRIOV_CTRL_VFE)
538		return;
539
540	for (i = PCI_IOV_RESOURCES; i <= PCI_IOV_RESOURCE_END; i++)
541		pci_update_resource(dev, i);
542
543	pci_write_config_dword(dev, iov->pos + PCI_SRIOV_SYS_PGSIZE, iov->pgsz);
544	pci_write_config_word(dev, iov->pos + PCI_SRIOV_NUM_VF, iov->nr_virtfn);
545	pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
546	if (iov->ctrl & PCI_SRIOV_CTRL_VFE)
547		msleep(100);
548}
549
550/**
551 * pci_iov_init - initialize the IOV capability
552 * @dev: the PCI device
553 *
554 * Returns 0 on success, or negative on failure.
555 */
556int pci_iov_init(struct pci_dev *dev)
557{
558	int pos;
559
560	if (!pci_is_pcie(dev))
561		return -ENODEV;
562
563	pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_SRIOV);
564	if (pos)
565		return sriov_init(dev, pos);
566
567	return -ENODEV;
568}
569
570/**
571 * pci_iov_release - release resources used by the IOV capability
572 * @dev: the PCI device
573 */
574void pci_iov_release(struct pci_dev *dev)
575{
576	if (dev->is_physfn)
577		sriov_release(dev);
578}
579
580/**
581 * pci_iov_resource_bar - get position of the SR-IOV BAR
582 * @dev: the PCI device
583 * @resno: the resource number
584 * @type: the BAR type to be filled in
585 *
586 * Returns position of the BAR encapsulated in the SR-IOV capability.
587 */
588int pci_iov_resource_bar(struct pci_dev *dev, int resno,
589			 enum pci_bar_type *type)
590{
591	if (resno < PCI_IOV_RESOURCES || resno > PCI_IOV_RESOURCE_END)
592		return 0;
593
594	BUG_ON(!dev->is_physfn);
595
596	*type = pci_bar_unknown;
597
598	return dev->sriov->pos + PCI_SRIOV_BAR +
599		4 * (resno - PCI_IOV_RESOURCES);
600}
601
602/**
603 * pci_sriov_resource_alignment - get resource alignment for VF BAR
604 * @dev: the PCI device
605 * @resno: the resource number
606 *
607 * Returns the alignment of the VF BAR found in the SR-IOV capability.
608 * This is not the same as the resource size which is defined as
609 * the VF BAR size multiplied by the number of VFs.  The alignment
610 * is just the VF BAR size.
611 */
612resource_size_t pci_sriov_resource_alignment(struct pci_dev *dev, int resno)
613{
614	struct resource tmp;
615	enum pci_bar_type type;
616	int reg = pci_iov_resource_bar(dev, resno, &type);
617	
618	if (!reg)
619		return 0;
620
621	 __pci_read_base(dev, type, &tmp, reg);
622	return resource_alignment(&tmp);
623}
624
625/**
626 * pci_restore_iov_state - restore the state of the IOV capability
627 * @dev: the PCI device
628 */
629void pci_restore_iov_state(struct pci_dev *dev)
630{
631	if (dev->is_physfn)
632		sriov_restore_state(dev);
633}
634
635/**
636 * pci_iov_bus_range - find bus range used by Virtual Function
637 * @bus: the PCI bus
638 *
639 * Returns max number of buses (exclude current one) used by Virtual
640 * Functions.
641 */
642int pci_iov_bus_range(struct pci_bus *bus)
643{
644	int max = 0;
645	u8 busnr;
646	struct pci_dev *dev;
647
648	list_for_each_entry(dev, &bus->devices, bus_list) {
649		if (!dev->is_physfn)
650			continue;
651		busnr = virtfn_bus(dev, dev->sriov->total - 1);
652		if (busnr > max)
653			max = busnr;
654	}
655
656	return max ? max - bus->number : 0;
657}
658
659/**
660 * pci_enable_sriov - enable the SR-IOV capability
661 * @dev: the PCI device
662 * @nr_virtfn: number of virtual functions to enable
663 *
664 * Returns 0 on success, or negative on failure.
665 */
666int pci_enable_sriov(struct pci_dev *dev, int nr_virtfn)
667{
668	might_sleep();
669
670	if (!dev->is_physfn)
671		return -ENODEV;
672
673	return sriov_enable(dev, nr_virtfn);
674}
675EXPORT_SYMBOL_GPL(pci_enable_sriov);
676
677/**
678 * pci_disable_sriov - disable the SR-IOV capability
679 * @dev: the PCI device
680 */
681void pci_disable_sriov(struct pci_dev *dev)
682{
683	might_sleep();
684
685	if (!dev->is_physfn)
686		return;
687
688	sriov_disable(dev);
689}
690EXPORT_SYMBOL_GPL(pci_disable_sriov);
691
692/**
693 * pci_sriov_migration - notify SR-IOV core of Virtual Function Migration
694 * @dev: the PCI device
695 *
696 * Returns IRQ_HANDLED if the IRQ is handled, or IRQ_NONE if not.
697 *
698 * Physical Function driver is responsible to register IRQ handler using
699 * VF Migration Interrupt Message Number, and call this function when the
700 * interrupt is generated by the hardware.
701 */
702irqreturn_t pci_sriov_migration(struct pci_dev *dev)
703{
704	if (!dev->is_physfn)
705		return IRQ_NONE;
706
707	return sriov_migration(dev) ? IRQ_HANDLED : IRQ_NONE;
708}
709EXPORT_SYMBOL_GPL(pci_sriov_migration);
710
711/**
712 * pci_num_vf - return number of VFs associated with a PF device_release_driver
713 * @dev: the PCI device
714 *
715 * Returns number of VFs, or 0 if SR-IOV is not enabled.
716 */
717int pci_num_vf(struct pci_dev *dev)
718{
719	if (!dev || !dev->is_physfn)
720		return 0;
721	else
722		return dev->sriov->nr_virtfn;
723}
724EXPORT_SYMBOL_GPL(pci_num_vf);
725
726static int ats_alloc_one(struct pci_dev *dev, int ps)
727{
728	int pos;
729	u16 cap;
730	struct pci_ats *ats;
731
732	pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ATS);
733	if (!pos)
734		return -ENODEV;
735
736	ats = kzalloc(sizeof(*ats), GFP_KERNEL);
737	if (!ats)
738		return -ENOMEM;
739
740	ats->pos = pos;
741	ats->stu = ps;
742	pci_read_config_word(dev, pos + PCI_ATS_CAP, &cap);
743	ats->qdep = PCI_ATS_CAP_QDEP(cap) ? PCI_ATS_CAP_QDEP(cap) :
744					    PCI_ATS_MAX_QDEP;
745	dev->ats = ats;
746
747	return 0;
748}
749
750static void ats_free_one(struct pci_dev *dev)
751{
752	kfree(dev->ats);
753	dev->ats = NULL;
754}
755
756/**
757 * pci_enable_ats - enable the ATS capability
758 * @dev: the PCI device
759 * @ps: the IOMMU page shift
760 *
761 * Returns 0 on success, or negative on failure.
762 */
763int pci_enable_ats(struct pci_dev *dev, int ps)
764{
765	int rc;
766	u16 ctrl;
767
768	BUG_ON(dev->ats && dev->ats->is_enabled);
769
770	if (ps < PCI_ATS_MIN_STU)
771		return -EINVAL;
772
773	if (dev->is_physfn || dev->is_virtfn) {
774		struct pci_dev *pdev = dev->is_physfn ? dev : dev->physfn;
775
776		mutex_lock(&pdev->sriov->lock);
777		if (pdev->ats)
778			rc = pdev->ats->stu == ps ? 0 : -EINVAL;
779		else
780			rc = ats_alloc_one(pdev, ps);
781
782		if (!rc)
783			pdev->ats->ref_cnt++;
784		mutex_unlock(&pdev->sriov->lock);
785		if (rc)
786			return rc;
787	}
788
789	if (!dev->is_physfn) {
790		rc = ats_alloc_one(dev, ps);
791		if (rc)
792			return rc;
793	}
794
795	ctrl = PCI_ATS_CTRL_ENABLE;
796	if (!dev->is_virtfn)
797		ctrl |= PCI_ATS_CTRL_STU(ps - PCI_ATS_MIN_STU);
798	pci_write_config_word(dev, dev->ats->pos + PCI_ATS_CTRL, ctrl);
799
800	dev->ats->is_enabled = 1;
801
802	return 0;
803}
804
805/**
806 * pci_disable_ats - disable the ATS capability
807 * @dev: the PCI device
808 */
809void pci_disable_ats(struct pci_dev *dev)
810{
811	u16 ctrl;
812
813	BUG_ON(!dev->ats || !dev->ats->is_enabled);
814
815	pci_read_config_word(dev, dev->ats->pos + PCI_ATS_CTRL, &ctrl);
816	ctrl &= ~PCI_ATS_CTRL_ENABLE;
817	pci_write_config_word(dev, dev->ats->pos + PCI_ATS_CTRL, ctrl);
818
819	dev->ats->is_enabled = 0;
820
821	if (dev->is_physfn || dev->is_virtfn) {
822		struct pci_dev *pdev = dev->is_physfn ? dev : dev->physfn;
823
824		mutex_lock(&pdev->sriov->lock);
825		pdev->ats->ref_cnt--;
826		if (!pdev->ats->ref_cnt)
827			ats_free_one(pdev);
828		mutex_unlock(&pdev->sriov->lock);
829	}
830
831	if (!dev->is_physfn)
832		ats_free_one(dev);
833}
834
835/**
836 * pci_ats_queue_depth - query the ATS Invalidate Queue Depth
837 * @dev: the PCI device
838 *
839 * Returns the queue depth on success, or negative on failure.
840 *
841 * The ATS spec uses 0 in the Invalidate Queue Depth field to
842 * indicate that the function can accept 32 Invalidate Request.
843 * But here we use the `real' values (i.e. 1~32) for the Queue
844 * Depth; and 0 indicates the function shares the Queue with
845 * other functions (doesn't exclusively own a Queue).
846 */
847int pci_ats_queue_depth(struct pci_dev *dev)
848{
849	int pos;
850	u16 cap;
851
852	if (dev->is_virtfn)
853		return 0;
854
855	if (dev->ats)
856		return dev->ats->qdep;
857
858	pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ATS);
859	if (!pos)
860		return -ENODEV;
861
862	pci_read_config_word(dev, pos + PCI_ATS_CAP, &cap);
863
864	return PCI_ATS_CAP_QDEP(cap) ? PCI_ATS_CAP_QDEP(cap) :
865				       PCI_ATS_MAX_QDEP;
866}