1
2
3
4
5
6
7
8
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_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 sysfs_remove_link(&virtfn->dev.kobj, "physfn");
177
178 mutex_lock(&iov->dev->sriov->lock);
179 pci_remove_bus_device(virtfn);
180 virtfn_remove_bus(dev->bus, virtfn_bus(dev, id));
181 mutex_unlock(&iov->dev->sriov->lock);
182
183 pci_dev_put(dev);
184}
185
186static int sriov_migration(struct pci_dev *dev)
187{
188 u16 status;
189 struct pci_sriov *iov = dev->sriov;
190
191 if (!iov->nr_virtfn)
192 return 0;
193
194 if (!(iov->cap & PCI_SRIOV_CAP_VFM))
195 return 0;
196
197 pci_read_config_word(dev, iov->pos + PCI_SRIOV_STATUS, &status);
198 if (!(status & PCI_SRIOV_STATUS_VFM))
199 return 0;
200
201 schedule_work(&iov->mtask);
202
203 return 1;
204}
205
206static void sriov_migration_task(struct work_struct *work)
207{
208 int i;
209 u8 state;
210 u16 status;
211 struct pci_sriov *iov = container_of(work, struct pci_sriov, mtask);
212
213 for (i = iov->initial; i < iov->nr_virtfn; i++) {
214 state = readb(iov->mstate + i);
215 if (state == PCI_SRIOV_VFM_MI) {
216 writeb(PCI_SRIOV_VFM_AV, iov->mstate + i);
217 state = readb(iov->mstate + i);
218 if (state == PCI_SRIOV_VFM_AV)
219 virtfn_add(iov->self, i, 1);
220 } else if (state == PCI_SRIOV_VFM_MO) {
221 virtfn_remove(iov->self, i, 1);
222 writeb(PCI_SRIOV_VFM_UA, iov->mstate + i);
223 state = readb(iov->mstate + i);
224 if (state == PCI_SRIOV_VFM_AV)
225 virtfn_add(iov->self, i, 0);
226 }
227 }
228
229 pci_read_config_word(iov->self, iov->pos + PCI_SRIOV_STATUS, &status);
230 status &= ~PCI_SRIOV_STATUS_VFM;
231 pci_write_config_word(iov->self, iov->pos + PCI_SRIOV_STATUS, status);
232}
233
234static int sriov_enable_migration(struct pci_dev *dev, int nr_virtfn)
235{
236 int bir;
237 u32 table;
238 resource_size_t pa;
239 struct pci_sriov *iov = dev->sriov;
240
241 if (nr_virtfn <= iov->initial)
242 return 0;
243
244 pci_read_config_dword(dev, iov->pos + PCI_SRIOV_VFM, &table);
245 bir = PCI_SRIOV_VFM_BIR(table);
246 if (bir > PCI_STD_RESOURCE_END)
247 return -EIO;
248
249 table = PCI_SRIOV_VFM_OFFSET(table);
250 if (table + nr_virtfn > pci_resource_len(dev, bir))
251 return -EIO;
252
253 pa = pci_resource_start(dev, bir) + table;
254 iov->mstate = ioremap(pa, nr_virtfn);
255 if (!iov->mstate)
256 return -ENOMEM;
257
258 INIT_WORK(&iov->mtask, sriov_migration_task);
259
260 iov->ctrl |= PCI_SRIOV_CTRL_VFM | PCI_SRIOV_CTRL_INTR;
261 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
262
263 return 0;
264}
265
266static void sriov_disable_migration(struct pci_dev *dev)
267{
268 struct pci_sriov *iov = dev->sriov;
269
270 iov->ctrl &= ~(PCI_SRIOV_CTRL_VFM | PCI_SRIOV_CTRL_INTR);
271 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
272
273 cancel_work_sync(&iov->mtask);
274 iounmap(iov->mstate);
275}
276
277static int sriov_enable(struct pci_dev *dev, int nr_virtfn)
278{
279 int rc;
280 int i, j;
281 int nres;
282 u16 offset, stride, initial;
283 struct resource *res;
284 struct pci_dev *pdev;
285 struct pci_sriov *iov = dev->sriov;
286 int bars = 0;
287
288 if (!nr_virtfn)
289 return 0;
290
291 if (iov->nr_virtfn)
292 return -EINVAL;
293
294 pci_read_config_word(dev, iov->pos + PCI_SRIOV_INITIAL_VF, &initial);
295 if (initial > iov->total ||
296 (!(iov->cap & PCI_SRIOV_CAP_VFM) && (initial != iov->total)))
297 return -EIO;
298
299 if (nr_virtfn < 0 || nr_virtfn > iov->total ||
300 (!(iov->cap & PCI_SRIOV_CAP_VFM) && (nr_virtfn > initial)))
301 return -EINVAL;
302
303 pci_write_config_word(dev, iov->pos + PCI_SRIOV_NUM_VF, nr_virtfn);
304 pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_OFFSET, &offset);
305 pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_STRIDE, &stride);
306 if (!offset || (nr_virtfn > 1 && !stride))
307 return -EIO;
308
309 nres = 0;
310 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
311 bars |= (1 << (i + PCI_IOV_RESOURCES));
312 res = dev->resource + PCI_IOV_RESOURCES + i;
313 if (res->parent)
314 nres++;
315 }
316 if (nres != iov->nres) {
317 dev_err(&dev->dev, "not enough MMIO resources for SR-IOV\n");
318 return -ENOMEM;
319 }
320
321 iov->offset = offset;
322 iov->stride = stride;
323
324 if (virtfn_bus(dev, nr_virtfn - 1) > dev->bus->subordinate) {
325 dev_err(&dev->dev, "SR-IOV: bus number out of range\n");
326 return -ENOMEM;
327 }
328
329 if (pci_enable_resources(dev, bars)) {
330 dev_err(&dev->dev, "SR-IOV: IOV BARS not allocated\n");
331 return -ENOMEM;
332 }
333
334 if (iov->link != dev->devfn) {
335 pdev = pci_get_slot(dev->bus, iov->link);
336 if (!pdev)
337 return -ENODEV;
338
339 pci_dev_put(pdev);
340
341 if (!pdev->is_physfn)
342 return -ENODEV;
343
344 rc = sysfs_create_link(&dev->dev.kobj,
345 &pdev->dev.kobj, "dep_link");
346 if (rc)
347 return rc;
348 }
349
350 iov->ctrl |= PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE;
351 pci_block_user_cfg_access(dev);
352 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
353 msleep(100);
354 pci_unblock_user_cfg_access(dev);
355
356 iov->initial = initial;
357 if (nr_virtfn < initial)
358 initial = nr_virtfn;
359
360 for (i = 0; i < initial; i++) {
361 rc = virtfn_add(dev, i, 0);
362 if (rc)
363 goto failed;
364 }
365
366 if (iov->cap & PCI_SRIOV_CAP_VFM) {
367 rc = sriov_enable_migration(dev, nr_virtfn);
368 if (rc)
369 goto failed;
370 }
371
372 kobject_uevent(&dev->dev.kobj, KOBJ_CHANGE);
373 iov->nr_virtfn = nr_virtfn;
374
375 return 0;
376
377failed:
378 for (j = 0; j < i; j++)
379 virtfn_remove(dev, j, 0);
380
381 iov->ctrl &= ~(PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE);
382 pci_block_user_cfg_access(dev);
383 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
384 ssleep(1);
385 pci_unblock_user_cfg_access(dev);
386
387 if (iov->link != dev->devfn)
388 sysfs_remove_link(&dev->dev.kobj, "dep_link");
389
390 return rc;
391}
392
393static void sriov_disable(struct pci_dev *dev)
394{
395 int i;
396 struct pci_sriov *iov = dev->sriov;
397
398 if (!iov->nr_virtfn)
399 return;
400
401 if (iov->cap & PCI_SRIOV_CAP_VFM)
402 sriov_disable_migration(dev);
403
404 for (i = 0; i < iov->nr_virtfn; i++)
405 virtfn_remove(dev, i, 0);
406
407 iov->ctrl &= ~(PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE);
408 pci_block_user_cfg_access(dev);
409 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
410 ssleep(1);
411 pci_unblock_user_cfg_access(dev);
412
413 if (iov->link != dev->devfn)
414 sysfs_remove_link(&dev->dev.kobj, "dep_link");
415
416 iov->nr_virtfn = 0;
417}
418
419static int sriov_init(struct pci_dev *dev, int pos)
420{
421 int i;
422 int rc;
423 int nres;
424 u32 pgsz;
425 u16 ctrl, total, offset, stride;
426 struct pci_sriov *iov;
427 struct resource *res;
428 struct pci_dev *pdev;
429
430 if (dev->pcie_type != PCI_EXP_TYPE_RC_END &&
431 dev->pcie_type != PCI_EXP_TYPE_ENDPOINT)
432 return -ENODEV;
433
434 pci_read_config_word(dev, pos + PCI_SRIOV_CTRL, &ctrl);
435 if (ctrl & PCI_SRIOV_CTRL_VFE) {
436 pci_write_config_word(dev, pos + PCI_SRIOV_CTRL, 0);
437 ssleep(1);
438 }
439
440 pci_read_config_word(dev, pos + PCI_SRIOV_TOTAL_VF, &total);
441 if (!total)
442 return 0;
443
444 ctrl = 0;
445 list_for_each_entry(pdev, &dev->bus->devices, bus_list)
446 if (pdev->is_physfn)
447 goto found;
448
449 pdev = NULL;
450 if (pci_ari_enabled(dev->bus))
451 ctrl |= PCI_SRIOV_CTRL_ARI;
452
453found:
454 pci_write_config_word(dev, pos + PCI_SRIOV_CTRL, ctrl);
455 pci_write_config_word(dev, pos + PCI_SRIOV_NUM_VF, total);
456 pci_read_config_word(dev, pos + PCI_SRIOV_VF_OFFSET, &offset);
457 pci_read_config_word(dev, pos + PCI_SRIOV_VF_STRIDE, &stride);
458 if (!offset || (total > 1 && !stride))
459 return -EIO;
460
461 pci_read_config_dword(dev, pos + PCI_SRIOV_SUP_PGSIZE, &pgsz);
462 i = PAGE_SHIFT > 12 ? PAGE_SHIFT - 12 : 0;
463 pgsz &= ~((1 << i) - 1);
464 if (!pgsz)
465 return -EIO;
466
467 pgsz &= ~(pgsz - 1);
468 pci_write_config_dword(dev, pos + PCI_SRIOV_SYS_PGSIZE, pgsz);
469
470 nres = 0;
471 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
472 res = dev->resource + PCI_IOV_RESOURCES + i;
473 i += __pci_read_base(dev, pci_bar_unknown, res,
474 pos + PCI_SRIOV_BAR + i * 4);
475 if (!res->flags)
476 continue;
477 if (resource_size(res) & (PAGE_SIZE - 1)) {
478 rc = -EIO;
479 goto failed;
480 }
481 res->end = res->start + resource_size(res) * total - 1;
482 nres++;
483 }
484
485 iov = kzalloc(sizeof(*iov), GFP_KERNEL);
486 if (!iov) {
487 rc = -ENOMEM;
488 goto failed;
489 }
490
491 iov->pos = pos;
492 iov->nres = nres;
493 iov->ctrl = ctrl;
494 iov->total = total;
495 iov->offset = offset;
496 iov->stride = stride;
497 iov->pgsz = pgsz;
498 iov->self = dev;
499 pci_read_config_dword(dev, pos + PCI_SRIOV_CAP, &iov->cap);
500 pci_read_config_byte(dev, pos + PCI_SRIOV_FUNC_LINK, &iov->link);
501 if (dev->pcie_type == PCI_EXP_TYPE_RC_END)
502 iov->link = PCI_DEVFN(PCI_SLOT(dev->devfn), iov->link);
503
504 if (pdev)
505 iov->dev = pci_dev_get(pdev);
506 else
507 iov->dev = dev;
508
509 mutex_init(&iov->lock);
510
511 dev->sriov = iov;
512 dev->is_physfn = 1;
513
514 return 0;
515
516failed:
517 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
518 res = dev->resource + PCI_IOV_RESOURCES + i;
519 res->flags = 0;
520 }
521
522 return rc;
523}
524
525static void sriov_release(struct pci_dev *dev)
526{
527 BUG_ON(dev->sriov->nr_virtfn);
528
529 if (dev != dev->sriov->dev)
530 pci_dev_put(dev->sriov->dev);
531
532 mutex_destroy(&dev->sriov->lock);
533
534 kfree(dev->sriov);
535 dev->sriov = NULL;
536}
537
538static void sriov_restore_state(struct pci_dev *dev)
539{
540 int i;
541 u16 ctrl;
542 struct pci_sriov *iov = dev->sriov;
543
544 pci_read_config_word(dev, iov->pos + PCI_SRIOV_CTRL, &ctrl);
545 if (ctrl & PCI_SRIOV_CTRL_VFE)
546 return;
547
548 for (i = PCI_IOV_RESOURCES; i <= PCI_IOV_RESOURCE_END; i++)
549 pci_update_resource(dev, i);
550
551 pci_write_config_dword(dev, iov->pos + PCI_SRIOV_SYS_PGSIZE, iov->pgsz);
552 pci_write_config_word(dev, iov->pos + PCI_SRIOV_NUM_VF, iov->nr_virtfn);
553 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
554 if (iov->ctrl & PCI_SRIOV_CTRL_VFE)
555 msleep(100);
556}
557
558
559
560
561
562
563
564int pci_iov_init(struct pci_dev *dev)
565{
566 int pos;
567
568 if (!pci_is_pcie(dev))
569 return -ENODEV;
570
571 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_SRIOV);
572 if (pos)
573 return sriov_init(dev, pos);
574
575 return -ENODEV;
576}
577
578
579
580
581
582void pci_iov_release(struct pci_dev *dev)
583{
584 if (dev->is_physfn)
585 sriov_release(dev);
586}
587
588
589
590
591
592
593
594
595
596int pci_iov_resource_bar(struct pci_dev *dev, int resno,
597 enum pci_bar_type *type)
598{
599 if (resno < PCI_IOV_RESOURCES || resno > PCI_IOV_RESOURCE_END)
600 return 0;
601
602 BUG_ON(!dev->is_physfn);
603
604 *type = pci_bar_unknown;
605
606 return dev->sriov->pos + PCI_SRIOV_BAR +
607 4 * (resno - PCI_IOV_RESOURCES);
608}
609
610
611
612
613
614
615
616
617
618
619
620resource_size_t pci_sriov_resource_alignment(struct pci_dev *dev, int resno)
621{
622 struct resource tmp;
623 enum pci_bar_type type;
624 int reg = pci_iov_resource_bar(dev, resno, &type);
625
626 if (!reg)
627 return 0;
628
629 __pci_read_base(dev, type, &tmp, reg);
630 return resource_alignment(&tmp);
631}
632
633
634
635
636
637void pci_restore_iov_state(struct pci_dev *dev)
638{
639 if (dev->is_physfn)
640 sriov_restore_state(dev);
641}
642
643
644
645
646
647
648
649
650int pci_iov_bus_range(struct pci_bus *bus)
651{
652 int max = 0;
653 u8 busnr;
654 struct pci_dev *dev;
655
656 list_for_each_entry(dev, &bus->devices, bus_list) {
657 if (!dev->is_physfn)
658 continue;
659 busnr = virtfn_bus(dev, dev->sriov->total - 1);
660 if (busnr > max)
661 max = busnr;
662 }
663
664 return max ? max - bus->number : 0;
665}
666
667
668
669
670
671
672
673
674int pci_enable_sriov(struct pci_dev *dev, int nr_virtfn)
675{
676 might_sleep();
677
678 if (!dev->is_physfn)
679 return -ENODEV;
680
681 return sriov_enable(dev, nr_virtfn);
682}
683EXPORT_SYMBOL_GPL(pci_enable_sriov);
684
685
686
687
688
689void pci_disable_sriov(struct pci_dev *dev)
690{
691 might_sleep();
692
693 if (!dev->is_physfn)
694 return;
695
696 sriov_disable(dev);
697}
698EXPORT_SYMBOL_GPL(pci_disable_sriov);
699
700
701
702
703
704
705
706
707
708
709
710irqreturn_t pci_sriov_migration(struct pci_dev *dev)
711{
712 if (!dev->is_physfn)
713 return IRQ_NONE;
714
715 return sriov_migration(dev) ? IRQ_HANDLED : IRQ_NONE;
716}
717EXPORT_SYMBOL_GPL(pci_sriov_migration);
718
719
720
721
722
723
724
725int pci_num_vf(struct pci_dev *dev)
726{
727 if (!dev || !dev->is_physfn)
728 return 0;
729 else
730 return dev->sriov->nr_virtfn;
731}
732EXPORT_SYMBOL_GPL(pci_num_vf);
733
