Loading...
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}
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
19int 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
27int 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
35/*
36 * Per SR-IOV spec sec 3.3.10 and 3.3.11, First VF Offset and VF Stride may
37 * change when NumVFs changes.
38 *
39 * Update iov->offset and iov->stride when NumVFs is written.
40 */
41static inline void pci_iov_set_numvfs(struct pci_dev *dev, int nr_virtfn)
42{
43 struct pci_sriov *iov = dev->sriov;
44
45 pci_write_config_word(dev, iov->pos + PCI_SRIOV_NUM_VF, nr_virtfn);
46 pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_OFFSET, &iov->offset);
47 pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_STRIDE, &iov->stride);
48}
49
50/*
51 * The PF consumes one bus number. NumVFs, First VF Offset, and VF Stride
52 * determine how many additional bus numbers will be consumed by VFs.
53 *
54 * Iterate over all valid NumVFs, validate offset and stride, and calculate
55 * the maximum number of bus numbers that could ever be required.
56 */
57static int compute_max_vf_buses(struct pci_dev *dev)
58{
59 struct pci_sriov *iov = dev->sriov;
60 int nr_virtfn, busnr, rc = 0;
61
62 for (nr_virtfn = iov->total_VFs; nr_virtfn; nr_virtfn--) {
63 pci_iov_set_numvfs(dev, nr_virtfn);
64 if (!iov->offset || (nr_virtfn > 1 && !iov->stride)) {
65 rc = -EIO;
66 goto out;
67 }
68
69 busnr = pci_iov_virtfn_bus(dev, nr_virtfn - 1);
70 if (busnr > iov->max_VF_buses)
71 iov->max_VF_buses = busnr;
72 }
73
74out:
75 pci_iov_set_numvfs(dev, 0);
76 return rc;
77}
78
79static struct pci_bus *virtfn_add_bus(struct pci_bus *bus, int busnr)
80{
81 struct pci_bus *child;
82
83 if (bus->number == busnr)
84 return bus;
85
86 child = pci_find_bus(pci_domain_nr(bus), busnr);
87 if (child)
88 return child;
89
90 child = pci_add_new_bus(bus, NULL, busnr);
91 if (!child)
92 return NULL;
93
94 pci_bus_insert_busn_res(child, busnr, busnr);
95
96 return child;
97}
98
99static void virtfn_remove_bus(struct pci_bus *physbus, struct pci_bus *virtbus)
100{
101 if (physbus != virtbus && list_empty(&virtbus->devices))
102 pci_remove_bus(virtbus);
103}
104
105resource_size_t pci_iov_resource_size(struct pci_dev *dev, int resno)
106{
107 if (!dev->is_physfn)
108 return 0;
109
110 return dev->sriov->barsz[resno - PCI_IOV_RESOURCES];
111}
112
113static void pci_read_vf_config_common(struct pci_dev *virtfn)
114{
115 struct pci_dev *physfn = virtfn->physfn;
116
117 /*
118 * Some config registers are the same across all associated VFs.
119 * Read them once from VF0 so we can skip reading them from the
120 * other VFs.
121 *
122 * PCIe r4.0, sec 9.3.4.1, technically doesn't require all VFs to
123 * have the same Revision ID and Subsystem ID, but we assume they
124 * do.
125 */
126 pci_read_config_dword(virtfn, PCI_CLASS_REVISION,
127 &physfn->sriov->class);
128 pci_read_config_byte(virtfn, PCI_HEADER_TYPE,
129 &physfn->sriov->hdr_type);
130 pci_read_config_word(virtfn, PCI_SUBSYSTEM_VENDOR_ID,
131 &physfn->sriov->subsystem_vendor);
132 pci_read_config_word(virtfn, PCI_SUBSYSTEM_ID,
133 &physfn->sriov->subsystem_device);
134}
135
136int pci_iov_sysfs_link(struct pci_dev *dev,
137 struct pci_dev *virtfn, int id)
138{
139 char buf[VIRTFN_ID_LEN];
140 int rc;
141
142 sprintf(buf, "virtfn%u", id);
143 rc = sysfs_create_link(&dev->dev.kobj, &virtfn->dev.kobj, buf);
144 if (rc)
145 goto failed;
146 rc = sysfs_create_link(&virtfn->dev.kobj, &dev->dev.kobj, "physfn");
147 if (rc)
148 goto failed1;
149
150 kobject_uevent(&virtfn->dev.kobj, KOBJ_CHANGE);
151
152 return 0;
153
154failed1:
155 sysfs_remove_link(&dev->dev.kobj, buf);
156failed:
157 return rc;
158}
159
160int pci_iov_add_virtfn(struct pci_dev *dev, int id)
161{
162 int i;
163 int rc = -ENOMEM;
164 u64 size;
165 struct pci_dev *virtfn;
166 struct resource *res;
167 struct pci_sriov *iov = dev->sriov;
168 struct pci_bus *bus;
169
170 bus = virtfn_add_bus(dev->bus, pci_iov_virtfn_bus(dev, id));
171 if (!bus)
172 goto failed;
173
174 virtfn = pci_alloc_dev(bus);
175 if (!virtfn)
176 goto failed0;
177
178 virtfn->devfn = pci_iov_virtfn_devfn(dev, id);
179 virtfn->vendor = dev->vendor;
180 virtfn->device = iov->vf_device;
181 virtfn->is_virtfn = 1;
182 virtfn->physfn = pci_dev_get(dev);
183
184 if (id == 0)
185 pci_read_vf_config_common(virtfn);
186
187 rc = pci_setup_device(virtfn);
188 if (rc)
189 goto failed1;
190
191 virtfn->dev.parent = dev->dev.parent;
192 virtfn->multifunction = 0;
193
194 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
195 res = &dev->resource[i + PCI_IOV_RESOURCES];
196 if (!res->parent)
197 continue;
198 virtfn->resource[i].name = pci_name(virtfn);
199 virtfn->resource[i].flags = res->flags;
200 size = pci_iov_resource_size(dev, i + PCI_IOV_RESOURCES);
201 virtfn->resource[i].start = res->start + size * id;
202 virtfn->resource[i].end = virtfn->resource[i].start + size - 1;
203 rc = request_resource(res, &virtfn->resource[i]);
204 BUG_ON(rc);
205 }
206
207 pci_device_add(virtfn, virtfn->bus);
208 rc = pci_iov_sysfs_link(dev, virtfn, id);
209 if (rc)
210 goto failed1;
211
212 pci_bus_add_device(virtfn);
213
214 return 0;
215
216failed1:
217 pci_stop_and_remove_bus_device(virtfn);
218 pci_dev_put(dev);
219failed0:
220 virtfn_remove_bus(dev->bus, bus);
221failed:
222
223 return rc;
224}
225
226void pci_iov_remove_virtfn(struct pci_dev *dev, int id)
227{
228 char buf[VIRTFN_ID_LEN];
229 struct pci_dev *virtfn;
230
231 virtfn = pci_get_domain_bus_and_slot(pci_domain_nr(dev->bus),
232 pci_iov_virtfn_bus(dev, id),
233 pci_iov_virtfn_devfn(dev, id));
234 if (!virtfn)
235 return;
236
237 sprintf(buf, "virtfn%u", id);
238 sysfs_remove_link(&dev->dev.kobj, buf);
239 /*
240 * pci_stop_dev() could have been called for this virtfn already,
241 * so the directory for the virtfn may have been removed before.
242 * Double check to avoid spurious sysfs warnings.
243 */
244 if (virtfn->dev.kobj.sd)
245 sysfs_remove_link(&virtfn->dev.kobj, "physfn");
246
247 pci_stop_and_remove_bus_device(virtfn);
248 virtfn_remove_bus(dev->bus, virtfn->bus);
249
250 /* balance pci_get_domain_bus_and_slot() */
251 pci_dev_put(virtfn);
252 pci_dev_put(dev);
253}
254
255static ssize_t sriov_totalvfs_show(struct device *dev,
256 struct device_attribute *attr,
257 char *buf)
258{
259 struct pci_dev *pdev = to_pci_dev(dev);
260
261 return sprintf(buf, "%u\n", pci_sriov_get_totalvfs(pdev));
262}
263
264static ssize_t sriov_numvfs_show(struct device *dev,
265 struct device_attribute *attr,
266 char *buf)
267{
268 struct pci_dev *pdev = to_pci_dev(dev);
269 u16 num_vfs;
270
271 /* Serialize vs sriov_numvfs_store() so readers see valid num_VFs */
272 device_lock(&pdev->dev);
273 num_vfs = pdev->sriov->num_VFs;
274 device_unlock(&pdev->dev);
275
276 return sprintf(buf, "%u\n", num_vfs);
277}
278
279/*
280 * num_vfs > 0; number of VFs to enable
281 * num_vfs = 0; disable all VFs
282 *
283 * Note: SRIOV spec does not allow partial VF
284 * disable, so it's all or none.
285 */
286static ssize_t sriov_numvfs_store(struct device *dev,
287 struct device_attribute *attr,
288 const char *buf, size_t count)
289{
290 struct pci_dev *pdev = to_pci_dev(dev);
291 int ret;
292 u16 num_vfs;
293
294 ret = kstrtou16(buf, 0, &num_vfs);
295 if (ret < 0)
296 return ret;
297
298 if (num_vfs > pci_sriov_get_totalvfs(pdev))
299 return -ERANGE;
300
301 device_lock(&pdev->dev);
302
303 if (num_vfs == pdev->sriov->num_VFs)
304 goto exit;
305
306 /* is PF driver loaded w/callback */
307 if (!pdev->driver || !pdev->driver->sriov_configure) {
308 pci_info(pdev, "Driver does not support SRIOV configuration via sysfs\n");
309 ret = -ENOENT;
310 goto exit;
311 }
312
313 if (num_vfs == 0) {
314 /* disable VFs */
315 ret = pdev->driver->sriov_configure(pdev, 0);
316 goto exit;
317 }
318
319 /* enable VFs */
320 if (pdev->sriov->num_VFs) {
321 pci_warn(pdev, "%d VFs already enabled. Disable before enabling %d VFs\n",
322 pdev->sriov->num_VFs, num_vfs);
323 ret = -EBUSY;
324 goto exit;
325 }
326
327 ret = pdev->driver->sriov_configure(pdev, num_vfs);
328 if (ret < 0)
329 goto exit;
330
331 if (ret != num_vfs)
332 pci_warn(pdev, "%d VFs requested; only %d enabled\n",
333 num_vfs, ret);
334
335exit:
336 device_unlock(&pdev->dev);
337
338 if (ret < 0)
339 return ret;
340
341 return count;
342}
343
344static ssize_t sriov_offset_show(struct device *dev,
345 struct device_attribute *attr,
346 char *buf)
347{
348 struct pci_dev *pdev = to_pci_dev(dev);
349
350 return sprintf(buf, "%u\n", pdev->sriov->offset);
351}
352
353static ssize_t sriov_stride_show(struct device *dev,
354 struct device_attribute *attr,
355 char *buf)
356{
357 struct pci_dev *pdev = to_pci_dev(dev);
358
359 return sprintf(buf, "%u\n", pdev->sriov->stride);
360}
361
362static ssize_t sriov_vf_device_show(struct device *dev,
363 struct device_attribute *attr,
364 char *buf)
365{
366 struct pci_dev *pdev = to_pci_dev(dev);
367
368 return sprintf(buf, "%x\n", pdev->sriov->vf_device);
369}
370
371static ssize_t sriov_drivers_autoprobe_show(struct device *dev,
372 struct device_attribute *attr,
373 char *buf)
374{
375 struct pci_dev *pdev = to_pci_dev(dev);
376
377 return sprintf(buf, "%u\n", pdev->sriov->drivers_autoprobe);
378}
379
380static ssize_t sriov_drivers_autoprobe_store(struct device *dev,
381 struct device_attribute *attr,
382 const char *buf, size_t count)
383{
384 struct pci_dev *pdev = to_pci_dev(dev);
385 bool drivers_autoprobe;
386
387 if (kstrtobool(buf, &drivers_autoprobe) < 0)
388 return -EINVAL;
389
390 pdev->sriov->drivers_autoprobe = drivers_autoprobe;
391
392 return count;
393}
394
395static DEVICE_ATTR_RO(sriov_totalvfs);
396static DEVICE_ATTR_RW(sriov_numvfs);
397static DEVICE_ATTR_RO(sriov_offset);
398static DEVICE_ATTR_RO(sriov_stride);
399static DEVICE_ATTR_RO(sriov_vf_device);
400static DEVICE_ATTR_RW(sriov_drivers_autoprobe);
401
402static struct attribute *sriov_dev_attrs[] = {
403 &dev_attr_sriov_totalvfs.attr,
404 &dev_attr_sriov_numvfs.attr,
405 &dev_attr_sriov_offset.attr,
406 &dev_attr_sriov_stride.attr,
407 &dev_attr_sriov_vf_device.attr,
408 &dev_attr_sriov_drivers_autoprobe.attr,
409 NULL,
410};
411
412static umode_t sriov_attrs_are_visible(struct kobject *kobj,
413 struct attribute *a, int n)
414{
415 struct device *dev = kobj_to_dev(kobj);
416
417 if (!dev_is_pf(dev))
418 return 0;
419
420 return a->mode;
421}
422
423const struct attribute_group sriov_dev_attr_group = {
424 .attrs = sriov_dev_attrs,
425 .is_visible = sriov_attrs_are_visible,
426};
427
428int __weak pcibios_sriov_enable(struct pci_dev *pdev, u16 num_vfs)
429{
430 return 0;
431}
432
433int __weak pcibios_sriov_disable(struct pci_dev *pdev)
434{
435 return 0;
436}
437
438static int sriov_add_vfs(struct pci_dev *dev, u16 num_vfs)
439{
440 unsigned int i;
441 int rc;
442
443 if (dev->no_vf_scan)
444 return 0;
445
446 for (i = 0; i < num_vfs; i++) {
447 rc = pci_iov_add_virtfn(dev, i);
448 if (rc)
449 goto failed;
450 }
451 return 0;
452failed:
453 while (i--)
454 pci_iov_remove_virtfn(dev, i);
455
456 return rc;
457}
458
459static int sriov_enable(struct pci_dev *dev, int nr_virtfn)
460{
461 int rc;
462 int i;
463 int nres;
464 u16 initial;
465 struct resource *res;
466 struct pci_dev *pdev;
467 struct pci_sriov *iov = dev->sriov;
468 int bars = 0;
469 int bus;
470
471 if (!nr_virtfn)
472 return 0;
473
474 if (iov->num_VFs)
475 return -EINVAL;
476
477 pci_read_config_word(dev, iov->pos + PCI_SRIOV_INITIAL_VF, &initial);
478 if (initial > iov->total_VFs ||
479 (!(iov->cap & PCI_SRIOV_CAP_VFM) && (initial != iov->total_VFs)))
480 return -EIO;
481
482 if (nr_virtfn < 0 || nr_virtfn > iov->total_VFs ||
483 (!(iov->cap & PCI_SRIOV_CAP_VFM) && (nr_virtfn > initial)))
484 return -EINVAL;
485
486 nres = 0;
487 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
488 bars |= (1 << (i + PCI_IOV_RESOURCES));
489 res = &dev->resource[i + PCI_IOV_RESOURCES];
490 if (res->parent)
491 nres++;
492 }
493 if (nres != iov->nres) {
494 pci_err(dev, "not enough MMIO resources for SR-IOV\n");
495 return -ENOMEM;
496 }
497
498 bus = pci_iov_virtfn_bus(dev, nr_virtfn - 1);
499 if (bus > dev->bus->busn_res.end) {
500 pci_err(dev, "can't enable %d VFs (bus %02x out of range of %pR)\n",
501 nr_virtfn, bus, &dev->bus->busn_res);
502 return -ENOMEM;
503 }
504
505 if (pci_enable_resources(dev, bars)) {
506 pci_err(dev, "SR-IOV: IOV BARS not allocated\n");
507 return -ENOMEM;
508 }
509
510 if (iov->link != dev->devfn) {
511 pdev = pci_get_slot(dev->bus, iov->link);
512 if (!pdev)
513 return -ENODEV;
514
515 if (!pdev->is_physfn) {
516 pci_dev_put(pdev);
517 return -ENOSYS;
518 }
519
520 rc = sysfs_create_link(&dev->dev.kobj,
521 &pdev->dev.kobj, "dep_link");
522 pci_dev_put(pdev);
523 if (rc)
524 return rc;
525 }
526
527 iov->initial_VFs = initial;
528 if (nr_virtfn < initial)
529 initial = nr_virtfn;
530
531 rc = pcibios_sriov_enable(dev, initial);
532 if (rc) {
533 pci_err(dev, "failure %d from pcibios_sriov_enable()\n", rc);
534 goto err_pcibios;
535 }
536
537 pci_iov_set_numvfs(dev, nr_virtfn);
538 iov->ctrl |= PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE;
539 pci_cfg_access_lock(dev);
540 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
541 msleep(100);
542 pci_cfg_access_unlock(dev);
543
544 rc = sriov_add_vfs(dev, initial);
545 if (rc)
546 goto err_pcibios;
547
548 kobject_uevent(&dev->dev.kobj, KOBJ_CHANGE);
549 iov->num_VFs = nr_virtfn;
550
551 return 0;
552
553err_pcibios:
554 iov->ctrl &= ~(PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE);
555 pci_cfg_access_lock(dev);
556 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
557 ssleep(1);
558 pci_cfg_access_unlock(dev);
559
560 pcibios_sriov_disable(dev);
561
562 if (iov->link != dev->devfn)
563 sysfs_remove_link(&dev->dev.kobj, "dep_link");
564
565 pci_iov_set_numvfs(dev, 0);
566 return rc;
567}
568
569static void sriov_del_vfs(struct pci_dev *dev)
570{
571 struct pci_sriov *iov = dev->sriov;
572 int i;
573
574 for (i = 0; i < iov->num_VFs; i++)
575 pci_iov_remove_virtfn(dev, i);
576}
577
578static void sriov_disable(struct pci_dev *dev)
579{
580 struct pci_sriov *iov = dev->sriov;
581
582 if (!iov->num_VFs)
583 return;
584
585 sriov_del_vfs(dev);
586 iov->ctrl &= ~(PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE);
587 pci_cfg_access_lock(dev);
588 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
589 ssleep(1);
590 pci_cfg_access_unlock(dev);
591
592 pcibios_sriov_disable(dev);
593
594 if (iov->link != dev->devfn)
595 sysfs_remove_link(&dev->dev.kobj, "dep_link");
596
597 iov->num_VFs = 0;
598 pci_iov_set_numvfs(dev, 0);
599}
600
601static int sriov_init(struct pci_dev *dev, int pos)
602{
603 int i, bar64;
604 int rc;
605 int nres;
606 u32 pgsz;
607 u16 ctrl, total;
608 struct pci_sriov *iov;
609 struct resource *res;
610 struct pci_dev *pdev;
611
612 pci_read_config_word(dev, pos + PCI_SRIOV_CTRL, &ctrl);
613 if (ctrl & PCI_SRIOV_CTRL_VFE) {
614 pci_write_config_word(dev, pos + PCI_SRIOV_CTRL, 0);
615 ssleep(1);
616 }
617
618 ctrl = 0;
619 list_for_each_entry(pdev, &dev->bus->devices, bus_list)
620 if (pdev->is_physfn)
621 goto found;
622
623 pdev = NULL;
624 if (pci_ari_enabled(dev->bus))
625 ctrl |= PCI_SRIOV_CTRL_ARI;
626
627found:
628 pci_write_config_word(dev, pos + PCI_SRIOV_CTRL, ctrl);
629
630 pci_read_config_word(dev, pos + PCI_SRIOV_TOTAL_VF, &total);
631 if (!total)
632 return 0;
633
634 pci_read_config_dword(dev, pos + PCI_SRIOV_SUP_PGSIZE, &pgsz);
635 i = PAGE_SHIFT > 12 ? PAGE_SHIFT - 12 : 0;
636 pgsz &= ~((1 << i) - 1);
637 if (!pgsz)
638 return -EIO;
639
640 pgsz &= ~(pgsz - 1);
641 pci_write_config_dword(dev, pos + PCI_SRIOV_SYS_PGSIZE, pgsz);
642
643 iov = kzalloc(sizeof(*iov), GFP_KERNEL);
644 if (!iov)
645 return -ENOMEM;
646
647 nres = 0;
648 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
649 res = &dev->resource[i + PCI_IOV_RESOURCES];
650 /*
651 * If it is already FIXED, don't change it, something
652 * (perhaps EA or header fixups) wants it this way.
653 */
654 if (res->flags & IORESOURCE_PCI_FIXED)
655 bar64 = (res->flags & IORESOURCE_MEM_64) ? 1 : 0;
656 else
657 bar64 = __pci_read_base(dev, pci_bar_unknown, res,
658 pos + PCI_SRIOV_BAR + i * 4);
659 if (!res->flags)
660 continue;
661 if (resource_size(res) & (PAGE_SIZE - 1)) {
662 rc = -EIO;
663 goto failed;
664 }
665 iov->barsz[i] = resource_size(res);
666 res->end = res->start + resource_size(res) * total - 1;
667 pci_info(dev, "VF(n) BAR%d space: %pR (contains BAR%d for %d VFs)\n",
668 i, res, i, total);
669 i += bar64;
670 nres++;
671 }
672
673 iov->pos = pos;
674 iov->nres = nres;
675 iov->ctrl = ctrl;
676 iov->total_VFs = total;
677 iov->driver_max_VFs = total;
678 pci_read_config_word(dev, pos + PCI_SRIOV_VF_DID, &iov->vf_device);
679 iov->pgsz = pgsz;
680 iov->self = dev;
681 iov->drivers_autoprobe = true;
682 pci_read_config_dword(dev, pos + PCI_SRIOV_CAP, &iov->cap);
683 pci_read_config_byte(dev, pos + PCI_SRIOV_FUNC_LINK, &iov->link);
684 if (pci_pcie_type(dev) == PCI_EXP_TYPE_RC_END)
685 iov->link = PCI_DEVFN(PCI_SLOT(dev->devfn), iov->link);
686
687 if (pdev)
688 iov->dev = pci_dev_get(pdev);
689 else
690 iov->dev = dev;
691
692 dev->sriov = iov;
693 dev->is_physfn = 1;
694 rc = compute_max_vf_buses(dev);
695 if (rc)
696 goto fail_max_buses;
697
698 return 0;
699
700fail_max_buses:
701 dev->sriov = NULL;
702 dev->is_physfn = 0;
703failed:
704 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
705 res = &dev->resource[i + PCI_IOV_RESOURCES];
706 res->flags = 0;
707 }
708
709 kfree(iov);
710 return rc;
711}
712
713static void sriov_release(struct pci_dev *dev)
714{
715 BUG_ON(dev->sriov->num_VFs);
716
717 if (dev != dev->sriov->dev)
718 pci_dev_put(dev->sriov->dev);
719
720 kfree(dev->sriov);
721 dev->sriov = NULL;
722}
723
724static void sriov_restore_state(struct pci_dev *dev)
725{
726 int i;
727 u16 ctrl;
728 struct pci_sriov *iov = dev->sriov;
729
730 pci_read_config_word(dev, iov->pos + PCI_SRIOV_CTRL, &ctrl);
731 if (ctrl & PCI_SRIOV_CTRL_VFE)
732 return;
733
734 /*
735 * Restore PCI_SRIOV_CTRL_ARI before pci_iov_set_numvfs() because
736 * it reads offset & stride, which depend on PCI_SRIOV_CTRL_ARI.
737 */
738 ctrl &= ~PCI_SRIOV_CTRL_ARI;
739 ctrl |= iov->ctrl & PCI_SRIOV_CTRL_ARI;
740 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, ctrl);
741
742 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++)
743 pci_update_resource(dev, i + PCI_IOV_RESOURCES);
744
745 pci_write_config_dword(dev, iov->pos + PCI_SRIOV_SYS_PGSIZE, iov->pgsz);
746 pci_iov_set_numvfs(dev, iov->num_VFs);
747 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
748 if (iov->ctrl & PCI_SRIOV_CTRL_VFE)
749 msleep(100);
750}
751
752/**
753 * pci_iov_init - initialize the IOV capability
754 * @dev: the PCI device
755 *
756 * Returns 0 on success, or negative on failure.
757 */
758int pci_iov_init(struct pci_dev *dev)
759{
760 int pos;
761
762 if (!pci_is_pcie(dev))
763 return -ENODEV;
764
765 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_SRIOV);
766 if (pos)
767 return sriov_init(dev, pos);
768
769 return -ENODEV;
770}
771
772/**
773 * pci_iov_release - release resources used by the IOV capability
774 * @dev: the PCI device
775 */
776void pci_iov_release(struct pci_dev *dev)
777{
778 if (dev->is_physfn)
779 sriov_release(dev);
780}
781
782/**
783 * pci_iov_remove - clean up SR-IOV state after PF driver is detached
784 * @dev: the PCI device
785 */
786void pci_iov_remove(struct pci_dev *dev)
787{
788 struct pci_sriov *iov = dev->sriov;
789
790 if (!dev->is_physfn)
791 return;
792
793 iov->driver_max_VFs = iov->total_VFs;
794 if (iov->num_VFs)
795 pci_warn(dev, "driver left SR-IOV enabled after remove\n");
796}
797
798/**
799 * pci_iov_update_resource - update a VF BAR
800 * @dev: the PCI device
801 * @resno: the resource number
802 *
803 * Update a VF BAR in the SR-IOV capability of a PF.
804 */
805void pci_iov_update_resource(struct pci_dev *dev, int resno)
806{
807 struct pci_sriov *iov = dev->is_physfn ? dev->sriov : NULL;
808 struct resource *res = dev->resource + resno;
809 int vf_bar = resno - PCI_IOV_RESOURCES;
810 struct pci_bus_region region;
811 u16 cmd;
812 u32 new;
813 int reg;
814
815 /*
816 * The generic pci_restore_bars() path calls this for all devices,
817 * including VFs and non-SR-IOV devices. If this is not a PF, we
818 * have nothing to do.
819 */
820 if (!iov)
821 return;
822
823 pci_read_config_word(dev, iov->pos + PCI_SRIOV_CTRL, &cmd);
824 if ((cmd & PCI_SRIOV_CTRL_VFE) && (cmd & PCI_SRIOV_CTRL_MSE)) {
825 dev_WARN(&dev->dev, "can't update enabled VF BAR%d %pR\n",
826 vf_bar, res);
827 return;
828 }
829
830 /*
831 * Ignore unimplemented BARs, unused resource slots for 64-bit
832 * BARs, and non-movable resources, e.g., those described via
833 * Enhanced Allocation.
834 */
835 if (!res->flags)
836 return;
837
838 if (res->flags & IORESOURCE_UNSET)
839 return;
840
841 if (res->flags & IORESOURCE_PCI_FIXED)
842 return;
843
844 pcibios_resource_to_bus(dev->bus, ®ion, res);
845 new = region.start;
846 new |= res->flags & ~PCI_BASE_ADDRESS_MEM_MASK;
847
848 reg = iov->pos + PCI_SRIOV_BAR + 4 * vf_bar;
849 pci_write_config_dword(dev, reg, new);
850 if (res->flags & IORESOURCE_MEM_64) {
851 new = region.start >> 16 >> 16;
852 pci_write_config_dword(dev, reg + 4, new);
853 }
854}
855
856resource_size_t __weak pcibios_iov_resource_alignment(struct pci_dev *dev,
857 int resno)
858{
859 return pci_iov_resource_size(dev, resno);
860}
861
862/**
863 * pci_sriov_resource_alignment - get resource alignment for VF BAR
864 * @dev: the PCI device
865 * @resno: the resource number
866 *
867 * Returns the alignment of the VF BAR found in the SR-IOV capability.
868 * This is not the same as the resource size which is defined as
869 * the VF BAR size multiplied by the number of VFs. The alignment
870 * is just the VF BAR size.
871 */
872resource_size_t pci_sriov_resource_alignment(struct pci_dev *dev, int resno)
873{
874 return pcibios_iov_resource_alignment(dev, resno);
875}
876
877/**
878 * pci_restore_iov_state - restore the state of the IOV capability
879 * @dev: the PCI device
880 */
881void pci_restore_iov_state(struct pci_dev *dev)
882{
883 if (dev->is_physfn)
884 sriov_restore_state(dev);
885}
886
887/**
888 * pci_vf_drivers_autoprobe - set PF property drivers_autoprobe for VFs
889 * @dev: the PCI device
890 * @auto_probe: set VF drivers auto probe flag
891 */
892void pci_vf_drivers_autoprobe(struct pci_dev *dev, bool auto_probe)
893{
894 if (dev->is_physfn)
895 dev->sriov->drivers_autoprobe = auto_probe;
896}
897
898/**
899 * pci_iov_bus_range - find bus range used by Virtual Function
900 * @bus: the PCI bus
901 *
902 * Returns max number of buses (exclude current one) used by Virtual
903 * Functions.
904 */
905int pci_iov_bus_range(struct pci_bus *bus)
906{
907 int max = 0;
908 struct pci_dev *dev;
909
910 list_for_each_entry(dev, &bus->devices, bus_list) {
911 if (!dev->is_physfn)
912 continue;
913 if (dev->sriov->max_VF_buses > max)
914 max = dev->sriov->max_VF_buses;
915 }
916
917 return max ? max - bus->number : 0;
918}
919
920/**
921 * pci_enable_sriov - enable the SR-IOV capability
922 * @dev: the PCI device
923 * @nr_virtfn: number of virtual functions to enable
924 *
925 * Returns 0 on success, or negative on failure.
926 */
927int pci_enable_sriov(struct pci_dev *dev, int nr_virtfn)
928{
929 might_sleep();
930
931 if (!dev->is_physfn)
932 return -ENOSYS;
933
934 return sriov_enable(dev, nr_virtfn);
935}
936EXPORT_SYMBOL_GPL(pci_enable_sriov);
937
938/**
939 * pci_disable_sriov - disable the SR-IOV capability
940 * @dev: the PCI device
941 */
942void pci_disable_sriov(struct pci_dev *dev)
943{
944 might_sleep();
945
946 if (!dev->is_physfn)
947 return;
948
949 sriov_disable(dev);
950}
951EXPORT_SYMBOL_GPL(pci_disable_sriov);
952
953/**
954 * pci_num_vf - return number of VFs associated with a PF device_release_driver
955 * @dev: the PCI device
956 *
957 * Returns number of VFs, or 0 if SR-IOV is not enabled.
958 */
959int pci_num_vf(struct pci_dev *dev)
960{
961 if (!dev->is_physfn)
962 return 0;
963
964 return dev->sriov->num_VFs;
965}
966EXPORT_SYMBOL_GPL(pci_num_vf);
967
968/**
969 * pci_vfs_assigned - returns number of VFs are assigned to a guest
970 * @dev: the PCI device
971 *
972 * Returns number of VFs belonging to this device that are assigned to a guest.
973 * If device is not a physical function returns 0.
974 */
975int pci_vfs_assigned(struct pci_dev *dev)
976{
977 struct pci_dev *vfdev;
978 unsigned int vfs_assigned = 0;
979 unsigned short dev_id;
980
981 /* only search if we are a PF */
982 if (!dev->is_physfn)
983 return 0;
984
985 /*
986 * determine the device ID for the VFs, the vendor ID will be the
987 * same as the PF so there is no need to check for that one
988 */
989 dev_id = dev->sriov->vf_device;
990
991 /* loop through all the VFs to see if we own any that are assigned */
992 vfdev = pci_get_device(dev->vendor, dev_id, NULL);
993 while (vfdev) {
994 /*
995 * It is considered assigned if it is a virtual function with
996 * our dev as the physical function and the assigned bit is set
997 */
998 if (vfdev->is_virtfn && (vfdev->physfn == dev) &&
999 pci_is_dev_assigned(vfdev))
1000 vfs_assigned++;
1001
1002 vfdev = pci_get_device(dev->vendor, dev_id, vfdev);
1003 }
1004
1005 return vfs_assigned;
1006}
1007EXPORT_SYMBOL_GPL(pci_vfs_assigned);
1008
1009/**
1010 * pci_sriov_set_totalvfs -- reduce the TotalVFs available
1011 * @dev: the PCI PF device
1012 * @numvfs: number that should be used for TotalVFs supported
1013 *
1014 * Should be called from PF driver's probe routine with
1015 * device's mutex held.
1016 *
1017 * Returns 0 if PF is an SRIOV-capable device and
1018 * value of numvfs valid. If not a PF return -ENOSYS;
1019 * if numvfs is invalid return -EINVAL;
1020 * if VFs already enabled, return -EBUSY.
1021 */
1022int pci_sriov_set_totalvfs(struct pci_dev *dev, u16 numvfs)
1023{
1024 if (!dev->is_physfn)
1025 return -ENOSYS;
1026
1027 if (numvfs > dev->sriov->total_VFs)
1028 return -EINVAL;
1029
1030 /* Shouldn't change if VFs already enabled */
1031 if (dev->sriov->ctrl & PCI_SRIOV_CTRL_VFE)
1032 return -EBUSY;
1033
1034 dev->sriov->driver_max_VFs = numvfs;
1035 return 0;
1036}
1037EXPORT_SYMBOL_GPL(pci_sriov_set_totalvfs);
1038
1039/**
1040 * pci_sriov_get_totalvfs -- get total VFs supported on this device
1041 * @dev: the PCI PF device
1042 *
1043 * For a PCIe device with SRIOV support, return the PCIe
1044 * SRIOV capability value of TotalVFs or the value of driver_max_VFs
1045 * if the driver reduced it. Otherwise 0.
1046 */
1047int pci_sriov_get_totalvfs(struct pci_dev *dev)
1048{
1049 if (!dev->is_physfn)
1050 return 0;
1051
1052 return dev->sriov->driver_max_VFs;
1053}
1054EXPORT_SYMBOL_GPL(pci_sriov_get_totalvfs);
1055
1056/**
1057 * pci_sriov_configure_simple - helper to configure SR-IOV
1058 * @dev: the PCI device
1059 * @nr_virtfn: number of virtual functions to enable, 0 to disable
1060 *
1061 * Enable or disable SR-IOV for devices that don't require any PF setup
1062 * before enabling SR-IOV. Return value is negative on error, or number of
1063 * VFs allocated on success.
1064 */
1065int pci_sriov_configure_simple(struct pci_dev *dev, int nr_virtfn)
1066{
1067 int rc;
1068
1069 might_sleep();
1070
1071 if (!dev->is_physfn)
1072 return -ENODEV;
1073
1074 if (pci_vfs_assigned(dev)) {
1075 pci_warn(dev, "Cannot modify SR-IOV while VFs are assigned\n");
1076 return -EPERM;
1077 }
1078
1079 if (nr_virtfn == 0) {
1080 sriov_disable(dev);
1081 return 0;
1082 }
1083
1084 rc = sriov_enable(dev, nr_virtfn);
1085 if (rc < 0)
1086 return rc;
1087
1088 return nr_virtfn;
1089}
1090EXPORT_SYMBOL_GPL(pci_sriov_configure_simple);