projects / linux-2.6-block.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
Merge tag 'mm-hotfixes-stable-2025-07-11-16-16' of git://git.kernel.org/pub/scm/linux...
[linux-2.6-block.git] / drivers / pci / iov.c
... / ...
CommitLineData
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 17 /* "virtfn%u\0" for 2^32 - 1 */
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}
34EXPORT_SYMBOL_GPL(pci_iov_virtfn_devfn);
35
36int pci_iov_vf_id(struct pci_dev *dev)
37{
38 struct pci_dev *pf;
39
40 if (!dev->is_virtfn)
41 return -EINVAL;
42
43 pf = pci_physfn(dev);
44 return (pci_dev_id(dev) - (pci_dev_id(pf) + pf->sriov->offset)) /
45 pf->sriov->stride;
46}
47EXPORT_SYMBOL_GPL(pci_iov_vf_id);
48
49/**
50 * pci_iov_get_pf_drvdata - Return the drvdata of a PF
51 * @dev: VF pci_dev
52 * @pf_driver: Device driver required to own the PF
53 *
54 * This must be called from a context that ensures that a VF driver is attached.
55 * The value returned is invalid once the VF driver completes its remove()
56 * callback.
57 *
58 * Locking is achieved by the driver core. A VF driver cannot be probed until
59 * pci_enable_sriov() is called and pci_disable_sriov() does not return until
60 * all VF drivers have completed their remove().
61 *
62 * The PF driver must call pci_disable_sriov() before it begins to destroy the
63 * drvdata.
64 */
65void *pci_iov_get_pf_drvdata(struct pci_dev *dev, struct pci_driver *pf_driver)
66{
67 struct pci_dev *pf_dev;
68
69 if (!dev->is_virtfn)
70 return ERR_PTR(-EINVAL);
71 pf_dev = dev->physfn;
72 if (pf_dev->driver != pf_driver)
73 return ERR_PTR(-EINVAL);
74 return pci_get_drvdata(pf_dev);
75}
76EXPORT_SYMBOL_GPL(pci_iov_get_pf_drvdata);
77
78/*
79 * Per SR-IOV spec sec 3.3.10 and 3.3.11, First VF Offset and VF Stride may
80 * change when NumVFs changes.
81 *
82 * Update iov->offset and iov->stride when NumVFs is written.
83 */
84static inline void pci_iov_set_numvfs(struct pci_dev *dev, int nr_virtfn)
85{
86 struct pci_sriov *iov = dev->sriov;
87
88 pci_write_config_word(dev, iov->pos + PCI_SRIOV_NUM_VF, nr_virtfn);
89 pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_OFFSET, &iov->offset);
90 pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_STRIDE, &iov->stride);
91}
92
93/*
94 * The PF consumes one bus number. NumVFs, First VF Offset, and VF Stride
95 * determine how many additional bus numbers will be consumed by VFs.
96 *
97 * Iterate over all valid NumVFs, validate offset and stride, and calculate
98 * the maximum number of bus numbers that could ever be required.
99 */
100static int compute_max_vf_buses(struct pci_dev *dev)
101{
102 struct pci_sriov *iov = dev->sriov;
103 int nr_virtfn, busnr, rc = 0;
104
105 for (nr_virtfn = iov->total_VFs; nr_virtfn; nr_virtfn--) {
106 pci_iov_set_numvfs(dev, nr_virtfn);
107 if (!iov->offset || (nr_virtfn > 1 && !iov->stride)) {
108 rc = -EIO;
109 goto out;
110 }
111
112 busnr = pci_iov_virtfn_bus(dev, nr_virtfn - 1);
113 if (busnr > iov->max_VF_buses)
114 iov->max_VF_buses = busnr;
115 }
116
117out:
118 pci_iov_set_numvfs(dev, 0);
119 return rc;
120}
121
122static struct pci_bus *virtfn_add_bus(struct pci_bus *bus, int busnr)
123{
124 struct pci_bus *child;
125
126 if (bus->number == busnr)
127 return bus;
128
129 child = pci_find_bus(pci_domain_nr(bus), busnr);
130 if (child)
131 return child;
132
133 child = pci_add_new_bus(bus, NULL, busnr);
134 if (!child)
135 return NULL;
136
137 pci_bus_insert_busn_res(child, busnr, busnr);
138
139 return child;
140}
141
142static void virtfn_remove_bus(struct pci_bus *physbus, struct pci_bus *virtbus)
143{
144 if (physbus != virtbus && list_empty(&virtbus->devices))
145 pci_remove_bus(virtbus);
146}
147
148resource_size_t pci_iov_resource_size(struct pci_dev *dev, int resno)
149{
150 if (!dev->is_physfn)
151 return 0;
152
153 return dev->sriov->barsz[resno - PCI_IOV_RESOURCES];
154}
155
156static void pci_read_vf_config_common(struct pci_dev *virtfn)
157{
158 struct pci_dev *physfn = virtfn->physfn;
159
160 /*
161 * Some config registers are the same across all associated VFs.
162 * Read them once from VF0 so we can skip reading them from the
163 * other VFs.
164 *
165 * PCIe r4.0, sec 9.3.4.1, technically doesn't require all VFs to
166 * have the same Revision ID and Subsystem ID, but we assume they
167 * do.
168 */
169 pci_read_config_dword(virtfn, PCI_CLASS_REVISION,
170 &physfn->sriov->class);
171 pci_read_config_byte(virtfn, PCI_HEADER_TYPE,
172 &physfn->sriov->hdr_type);
173 pci_read_config_word(virtfn, PCI_SUBSYSTEM_VENDOR_ID,
174 &physfn->sriov->subsystem_vendor);
175 pci_read_config_word(virtfn, PCI_SUBSYSTEM_ID,
176 &physfn->sriov->subsystem_device);
177}
178
179int pci_iov_sysfs_link(struct pci_dev *dev,
180 struct pci_dev *virtfn, int id)
181{
182 char buf[VIRTFN_ID_LEN];
183 int rc;
184
185 sprintf(buf, "virtfn%u", id);
186 rc = sysfs_create_link(&dev->dev.kobj, &virtfn->dev.kobj, buf);
187 if (rc)
188 goto failed;
189 rc = sysfs_create_link(&virtfn->dev.kobj, &dev->dev.kobj, "physfn");
190 if (rc)
191 goto failed1;
192
193 kobject_uevent(&virtfn->dev.kobj, KOBJ_CHANGE);
194
195 return 0;
196
197failed1:
198 sysfs_remove_link(&dev->dev.kobj, buf);
199failed:
200 return rc;
201}
202
203#ifdef CONFIG_PCI_MSI
204static ssize_t sriov_vf_total_msix_show(struct device *dev,
205 struct device_attribute *attr,
206 char *buf)
207{
208 struct pci_dev *pdev = to_pci_dev(dev);
209 u32 vf_total_msix = 0;
210
211 device_lock(dev);
212 if (!pdev->driver || !pdev->driver->sriov_get_vf_total_msix)
213 goto unlock;
214
215 vf_total_msix = pdev->driver->sriov_get_vf_total_msix(pdev);
216unlock:
217 device_unlock(dev);
218 return sysfs_emit(buf, "%u\n", vf_total_msix);
219}
220static DEVICE_ATTR_RO(sriov_vf_total_msix);
221
222static ssize_t sriov_vf_msix_count_store(struct device *dev,
223 struct device_attribute *attr,
224 const char *buf, size_t count)
225{
226 struct pci_dev *vf_dev = to_pci_dev(dev);
227 struct pci_dev *pdev = pci_physfn(vf_dev);
228 int val, ret = 0;
229
230 if (kstrtoint(buf, 0, &val) < 0)
231 return -EINVAL;
232
233 if (val < 0)
234 return -EINVAL;
235
236 device_lock(&pdev->dev);
237 if (!pdev->driver || !pdev->driver->sriov_set_msix_vec_count) {
238 ret = -EOPNOTSUPP;
239 goto err_pdev;
240 }
241
242 device_lock(&vf_dev->dev);
243 if (vf_dev->driver) {
244 /*
245 * A driver is already attached to this VF and has configured
246 * itself based on the current MSI-X vector count. Changing
247 * the vector size could mess up the driver, so block it.
248 */
249 ret = -EBUSY;
250 goto err_dev;
251 }
252
253 ret = pdev->driver->sriov_set_msix_vec_count(vf_dev, val);
254
255err_dev:
256 device_unlock(&vf_dev->dev);
257err_pdev:
258 device_unlock(&pdev->dev);
259 return ret ? : count;
260}
261static DEVICE_ATTR_WO(sriov_vf_msix_count);
262#endif
263
264static struct attribute *sriov_vf_dev_attrs[] = {
265#ifdef CONFIG_PCI_MSI
266 &dev_attr_sriov_vf_msix_count.attr,
267#endif
268 NULL,
269};
270
271static umode_t sriov_vf_attrs_are_visible(struct kobject *kobj,
272 struct attribute *a, int n)
273{
274 struct device *dev = kobj_to_dev(kobj);
275 struct pci_dev *pdev = to_pci_dev(dev);
276
277 if (!pdev->is_virtfn)
278 return 0;
279
280 return a->mode;
281}
282
283const struct attribute_group sriov_vf_dev_attr_group = {
284 .attrs = sriov_vf_dev_attrs,
285 .is_visible = sriov_vf_attrs_are_visible,
286};
287
288static struct pci_dev *pci_iov_scan_device(struct pci_dev *dev, int id,
289 struct pci_bus *bus)
290{
291 struct pci_sriov *iov = dev->sriov;
292 struct pci_dev *virtfn;
293 int rc;
294
295 virtfn = pci_alloc_dev(bus);
296 if (!virtfn)
297 return ERR_PTR(-ENOMEM);
298
299 virtfn->devfn = pci_iov_virtfn_devfn(dev, id);
300 virtfn->vendor = dev->vendor;
301 virtfn->device = iov->vf_device;
302 virtfn->is_virtfn = 1;
303 virtfn->physfn = pci_dev_get(dev);
304 virtfn->no_command_memory = 1;
305
306 if (id == 0)
307 pci_read_vf_config_common(virtfn);
308
309 rc = pci_setup_device(virtfn);
310 if (rc) {
311 pci_dev_put(dev);
312 pci_bus_put(virtfn->bus);
313 kfree(virtfn);
314 return ERR_PTR(rc);
315 }
316
317 return virtfn;
318}
319
320int pci_iov_add_virtfn(struct pci_dev *dev, int id)
321{
322 struct pci_bus *bus;
323 struct pci_dev *virtfn;
324 struct resource *res;
325 int rc, i;
326 u64 size;
327
328 bus = virtfn_add_bus(dev->bus, pci_iov_virtfn_bus(dev, id));
329 if (!bus) {
330 rc = -ENOMEM;
331 goto failed;
332 }
333
334 virtfn = pci_iov_scan_device(dev, id, bus);
335 if (IS_ERR(virtfn)) {
336 rc = PTR_ERR(virtfn);
337 goto failed0;
338 }
339
340 virtfn->dev.parent = dev->dev.parent;
341 virtfn->multifunction = 0;
342
343 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
344 res = &dev->resource[i + PCI_IOV_RESOURCES];
345 if (!res->parent)
346 continue;
347 virtfn->resource[i].name = pci_name(virtfn);
348 virtfn->resource[i].flags = res->flags;
349 size = pci_iov_resource_size(dev, i + PCI_IOV_RESOURCES);
350 resource_set_range(&virtfn->resource[i],
351 res->start + size * id, size);
352 rc = request_resource(res, &virtfn->resource[i]);
353 BUG_ON(rc);
354 }
355
356 pci_device_add(virtfn, virtfn->bus);
357 rc = pci_iov_sysfs_link(dev, virtfn, id);
358 if (rc)
359 goto failed1;
360
361 pci_bus_add_device(virtfn);
362
363 return 0;
364
365failed1:
366 pci_stop_and_remove_bus_device(virtfn);
367 pci_dev_put(dev);
368failed0:
369 virtfn_remove_bus(dev->bus, bus);
370failed:
371
372 return rc;
373}
374
375void pci_iov_remove_virtfn(struct pci_dev *dev, int id)
376{
377 char buf[VIRTFN_ID_LEN];
378 struct pci_dev *virtfn;
379
380 virtfn = pci_get_domain_bus_and_slot(pci_domain_nr(dev->bus),
381 pci_iov_virtfn_bus(dev, id),
382 pci_iov_virtfn_devfn(dev, id));
383 if (!virtfn)
384 return;
385
386 sprintf(buf, "virtfn%u", id);
387 sysfs_remove_link(&dev->dev.kobj, buf);
388 /*
389 * pci_stop_dev() could have been called for this virtfn already,
390 * so the directory for the virtfn may have been removed before.
391 * Double check to avoid spurious sysfs warnings.
392 */
393 if (virtfn->dev.kobj.sd)
394 sysfs_remove_link(&virtfn->dev.kobj, "physfn");
395
396 pci_stop_and_remove_bus_device(virtfn);
397 virtfn_remove_bus(dev->bus, virtfn->bus);
398
399 /* balance pci_get_domain_bus_and_slot() */
400 pci_dev_put(virtfn);
401 pci_dev_put(dev);
402}
403
404static ssize_t sriov_totalvfs_show(struct device *dev,
405 struct device_attribute *attr,
406 char *buf)
407{
408 struct pci_dev *pdev = to_pci_dev(dev);
409
410 return sysfs_emit(buf, "%u\n", pci_sriov_get_totalvfs(pdev));
411}
412
413static ssize_t sriov_numvfs_show(struct device *dev,
414 struct device_attribute *attr,
415 char *buf)
416{
417 struct pci_dev *pdev = to_pci_dev(dev);
418 u16 num_vfs;
419
420 /* Serialize vs sriov_numvfs_store() so readers see valid num_VFs */
421 device_lock(&pdev->dev);
422 num_vfs = pdev->sriov->num_VFs;
423 device_unlock(&pdev->dev);
424
425 return sysfs_emit(buf, "%u\n", num_vfs);
426}
427
428/*
429 * num_vfs > 0; number of VFs to enable
430 * num_vfs = 0; disable all VFs
431 *
432 * Note: SRIOV spec does not allow partial VF
433 * disable, so it's all or none.
434 */
435static ssize_t sriov_numvfs_store(struct device *dev,
436 struct device_attribute *attr,
437 const char *buf, size_t count)
438{
439 struct pci_dev *pdev = to_pci_dev(dev);
440 int ret = 0;
441 u16 num_vfs;
442
443 if (kstrtou16(buf, 0, &num_vfs) < 0)
444 return -EINVAL;
445
446 if (num_vfs > pci_sriov_get_totalvfs(pdev))
447 return -ERANGE;
448
449 device_lock(&pdev->dev);
450
451 if (num_vfs == pdev->sriov->num_VFs)
452 goto exit;
453
454 /* is PF driver loaded */
455 if (!pdev->driver) {
456 pci_info(pdev, "no driver bound to device; cannot configure SR-IOV\n");
457 ret = -ENOENT;
458 goto exit;
459 }
460
461 /* is PF driver loaded w/callback */
462 if (!pdev->driver->sriov_configure) {
463 pci_info(pdev, "driver does not support SR-IOV configuration via sysfs\n");
464 ret = -ENOENT;
465 goto exit;
466 }
467
468 if (num_vfs == 0) {
469 /* disable VFs */
470 ret = pdev->driver->sriov_configure(pdev, 0);
471 goto exit;
472 }
473
474 /* enable VFs */
475 if (pdev->sriov->num_VFs) {
476 pci_warn(pdev, "%d VFs already enabled. Disable before enabling %d VFs\n",
477 pdev->sriov->num_VFs, num_vfs);
478 ret = -EBUSY;
479 goto exit;
480 }
481
482 ret = pdev->driver->sriov_configure(pdev, num_vfs);
483 if (ret < 0)
484 goto exit;
485
486 if (ret != num_vfs)
487 pci_warn(pdev, "%d VFs requested; only %d enabled\n",
488 num_vfs, ret);
489
490exit:
491 device_unlock(&pdev->dev);
492
493 if (ret < 0)
494 return ret;
495
496 return count;
497}
498
499static ssize_t sriov_offset_show(struct device *dev,
500 struct device_attribute *attr,
501 char *buf)
502{
503 struct pci_dev *pdev = to_pci_dev(dev);
504
505 return sysfs_emit(buf, "%u\n", pdev->sriov->offset);
506}
507
508static ssize_t sriov_stride_show(struct device *dev,
509 struct device_attribute *attr,
510 char *buf)
511{
512 struct pci_dev *pdev = to_pci_dev(dev);
513
514 return sysfs_emit(buf, "%u\n", pdev->sriov->stride);
515}
516
517static ssize_t sriov_vf_device_show(struct device *dev,
518 struct device_attribute *attr,
519 char *buf)
520{
521 struct pci_dev *pdev = to_pci_dev(dev);
522
523 return sysfs_emit(buf, "%x\n", pdev->sriov->vf_device);
524}
525
526static ssize_t sriov_drivers_autoprobe_show(struct device *dev,
527 struct device_attribute *attr,
528 char *buf)
529{
530 struct pci_dev *pdev = to_pci_dev(dev);
531
532 return sysfs_emit(buf, "%u\n", pdev->sriov->drivers_autoprobe);
533}
534
535static ssize_t sriov_drivers_autoprobe_store(struct device *dev,
536 struct device_attribute *attr,
537 const char *buf, size_t count)
538{
539 struct pci_dev *pdev = to_pci_dev(dev);
540 bool drivers_autoprobe;
541
542 if (kstrtobool(buf, &drivers_autoprobe) < 0)
543 return -EINVAL;
544
545 pdev->sriov->drivers_autoprobe = drivers_autoprobe;
546
547 return count;
548}
549
550static DEVICE_ATTR_RO(sriov_totalvfs);
551static DEVICE_ATTR_RW(sriov_numvfs);
552static DEVICE_ATTR_RO(sriov_offset);
553static DEVICE_ATTR_RO(sriov_stride);
554static DEVICE_ATTR_RO(sriov_vf_device);
555static DEVICE_ATTR_RW(sriov_drivers_autoprobe);
556
557static struct attribute *sriov_pf_dev_attrs[] = {
558 &dev_attr_sriov_totalvfs.attr,
559 &dev_attr_sriov_numvfs.attr,
560 &dev_attr_sriov_offset.attr,
561 &dev_attr_sriov_stride.attr,
562 &dev_attr_sriov_vf_device.attr,
563 &dev_attr_sriov_drivers_autoprobe.attr,
564#ifdef CONFIG_PCI_MSI
565 &dev_attr_sriov_vf_total_msix.attr,
566#endif
567 NULL,
568};
569
570static umode_t sriov_pf_attrs_are_visible(struct kobject *kobj,
571 struct attribute *a, int n)
572{
573 struct device *dev = kobj_to_dev(kobj);
574
575 if (!dev_is_pf(dev))
576 return 0;
577
578 return a->mode;
579}
580
581const struct attribute_group sriov_pf_dev_attr_group = {
582 .attrs = sriov_pf_dev_attrs,
583 .is_visible = sriov_pf_attrs_are_visible,
584};
585
586int __weak pcibios_sriov_enable(struct pci_dev *pdev, u16 num_vfs)
587{
588 return 0;
589}
590
591int __weak pcibios_sriov_disable(struct pci_dev *pdev)
592{
593 return 0;
594}
595
596static int sriov_add_vfs(struct pci_dev *dev, u16 num_vfs)
597{
598 unsigned int i;
599 int rc;
600
601 if (dev->no_vf_scan)
602 return 0;
603
604 for (i = 0; i < num_vfs; i++) {
605 rc = pci_iov_add_virtfn(dev, i);
606 if (rc)
607 goto failed;
608 }
609 return 0;
610failed:
611 while (i--)
612 pci_iov_remove_virtfn(dev, i);
613
614 return rc;
615}
616
617static int sriov_enable(struct pci_dev *dev, int nr_virtfn)
618{
619 int rc;
620 int i;
621 int nres;
622 u16 initial;
623 struct resource *res;
624 struct pci_dev *pdev;
625 struct pci_sriov *iov = dev->sriov;
626 int bars = 0;
627 int bus;
628
629 if (!nr_virtfn)
630 return 0;
631
632 if (iov->num_VFs)
633 return -EINVAL;
634
635 pci_read_config_word(dev, iov->pos + PCI_SRIOV_INITIAL_VF, &initial);
636 if (initial > iov->total_VFs ||
637 (!(iov->cap & PCI_SRIOV_CAP_VFM) && (initial != iov->total_VFs)))
638 return -EIO;
639
640 if (nr_virtfn < 0 || nr_virtfn > iov->total_VFs ||
641 (!(iov->cap & PCI_SRIOV_CAP_VFM) && (nr_virtfn > initial)))
642 return -EINVAL;
643
644 nres = 0;
645 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
646 bars |= (1 << (i + PCI_IOV_RESOURCES));
647 res = &dev->resource[i + PCI_IOV_RESOURCES];
648 if (res->parent)
649 nres++;
650 }
651 if (nres != iov->nres) {
652 pci_err(dev, "not enough MMIO resources for SR-IOV\n");
653 return -ENOMEM;
654 }
655
656 bus = pci_iov_virtfn_bus(dev, nr_virtfn - 1);
657 if (bus > dev->bus->busn_res.end) {
658 pci_err(dev, "can't enable %d VFs (bus %02x out of range of %pR)\n",
659 nr_virtfn, bus, &dev->bus->busn_res);
660 return -ENOMEM;
661 }
662
663 if (pci_enable_resources(dev, bars)) {
664 pci_err(dev, "SR-IOV: IOV BARS not allocated\n");
665 return -ENOMEM;
666 }
667
668 if (iov->link != dev->devfn) {
669 pdev = pci_get_slot(dev->bus, iov->link);
670 if (!pdev)
671 return -ENODEV;
672
673 if (!pdev->is_physfn) {
674 pci_dev_put(pdev);
675 return -ENOSYS;
676 }
677
678 rc = sysfs_create_link(&dev->dev.kobj,
679 &pdev->dev.kobj, "dep_link");
680 pci_dev_put(pdev);
681 if (rc)
682 return rc;
683 }
684
685 iov->initial_VFs = initial;
686 if (nr_virtfn < initial)
687 initial = nr_virtfn;
688
689 rc = pcibios_sriov_enable(dev, initial);
690 if (rc) {
691 pci_err(dev, "failure %d from pcibios_sriov_enable()\n", rc);
692 goto err_pcibios;
693 }
694
695 pci_iov_set_numvfs(dev, nr_virtfn);
696 iov->ctrl |= PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE;
697 pci_cfg_access_lock(dev);
698 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
699 msleep(100);
700 pci_cfg_access_unlock(dev);
701
702 rc = sriov_add_vfs(dev, initial);
703 if (rc)
704 goto err_pcibios;
705
706 kobject_uevent(&dev->dev.kobj, KOBJ_CHANGE);
707 iov->num_VFs = nr_virtfn;
708
709 return 0;
710
711err_pcibios:
712 iov->ctrl &= ~(PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE);
713 pci_cfg_access_lock(dev);
714 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
715 ssleep(1);
716 pci_cfg_access_unlock(dev);
717
718 pcibios_sriov_disable(dev);
719
720 if (iov->link != dev->devfn)
721 sysfs_remove_link(&dev->dev.kobj, "dep_link");
722
723 pci_iov_set_numvfs(dev, 0);
724 return rc;
725}
726
727static void sriov_del_vfs(struct pci_dev *dev)
728{
729 struct pci_sriov *iov = dev->sriov;
730 int i;
731
732 for (i = 0; i < iov->num_VFs; i++)
733 pci_iov_remove_virtfn(dev, i);
734}
735
736static void sriov_disable(struct pci_dev *dev)
737{
738 struct pci_sriov *iov = dev->sriov;
739
740 if (!iov->num_VFs)
741 return;
742
743 sriov_del_vfs(dev);
744 iov->ctrl &= ~(PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE);
745 pci_cfg_access_lock(dev);
746 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
747 ssleep(1);
748 pci_cfg_access_unlock(dev);
749
750 pcibios_sriov_disable(dev);
751
752 if (iov->link != dev->devfn)
753 sysfs_remove_link(&dev->dev.kobj, "dep_link");
754
755 iov->num_VFs = 0;
756 pci_iov_set_numvfs(dev, 0);
757}
758
759static int sriov_init(struct pci_dev *dev, int pos)
760{
761 int i, bar64;
762 int rc;
763 int nres;
764 u32 pgsz;
765 u16 ctrl, total;
766 struct pci_sriov *iov;
767 struct resource *res;
768 const char *res_name;
769 struct pci_dev *pdev;
770 u32 sriovbars[PCI_SRIOV_NUM_BARS];
771
772 pci_read_config_word(dev, pos + PCI_SRIOV_CTRL, &ctrl);
773 if (ctrl & PCI_SRIOV_CTRL_VFE) {
774 pci_write_config_word(dev, pos + PCI_SRIOV_CTRL, 0);
775 ssleep(1);
776 }
777
778 ctrl = 0;
779 list_for_each_entry(pdev, &dev->bus->devices, bus_list)
780 if (pdev->is_physfn)
781 goto found;
782
783 pdev = NULL;
784 if (pci_ari_enabled(dev->bus))
785 ctrl |= PCI_SRIOV_CTRL_ARI;
786
787found:
788 pci_write_config_word(dev, pos + PCI_SRIOV_CTRL, ctrl);
789
790 pci_read_config_word(dev, pos + PCI_SRIOV_TOTAL_VF, &total);
791 if (!total)
792 return 0;
793
794 pci_read_config_dword(dev, pos + PCI_SRIOV_SUP_PGSIZE, &pgsz);
795 i = PAGE_SHIFT > 12 ? PAGE_SHIFT - 12 : 0;
796 pgsz &= ~((1 << i) - 1);
797 if (!pgsz)
798 return -EIO;
799
800 pgsz &= ~(pgsz - 1);
801 pci_write_config_dword(dev, pos + PCI_SRIOV_SYS_PGSIZE, pgsz);
802
803 iov = kzalloc(sizeof(*iov), GFP_KERNEL);
804 if (!iov)
805 return -ENOMEM;
806
807 /* Sizing SR-IOV BARs with VF Enable cleared - no decode */
808 __pci_size_stdbars(dev, PCI_SRIOV_NUM_BARS,
809 pos + PCI_SRIOV_BAR, sriovbars);
810
811 nres = 0;
812 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
813 res = &dev->resource[i + PCI_IOV_RESOURCES];
814 res_name = pci_resource_name(dev, i + PCI_IOV_RESOURCES);
815
816 /*
817 * If it is already FIXED, don't change it, something
818 * (perhaps EA or header fixups) wants it this way.
819 */
820 if (res->flags & IORESOURCE_PCI_FIXED)
821 bar64 = (res->flags & IORESOURCE_MEM_64) ? 1 : 0;
822 else
823 bar64 = __pci_read_base(dev, pci_bar_unknown, res,
824 pos + PCI_SRIOV_BAR + i * 4,
825 &sriovbars[i]);
826 if (!res->flags)
827 continue;
828 if (resource_size(res) & (PAGE_SIZE - 1)) {
829 rc = -EIO;
830 goto failed;
831 }
832 iov->barsz[i] = resource_size(res);
833 resource_set_size(res, resource_size(res) * total);
834 pci_info(dev, "%s %pR: contains BAR %d for %d VFs\n",
835 res_name, res, i, total);
836 i += bar64;
837 nres++;
838 }
839
840 iov->pos = pos;
841 iov->nres = nres;
842 iov->ctrl = ctrl;
843 iov->total_VFs = total;
844 iov->driver_max_VFs = total;
845 pci_read_config_word(dev, pos + PCI_SRIOV_VF_DID, &iov->vf_device);
846 iov->pgsz = pgsz;
847 iov->self = dev;
848 iov->drivers_autoprobe = true;
849 pci_read_config_dword(dev, pos + PCI_SRIOV_CAP, &iov->cap);
850 pci_read_config_byte(dev, pos + PCI_SRIOV_FUNC_LINK, &iov->link);
851 if (pci_pcie_type(dev) == PCI_EXP_TYPE_RC_END)
852 iov->link = PCI_DEVFN(PCI_SLOT(dev->devfn), iov->link);
853
854 if (pdev)
855 iov->dev = pci_dev_get(pdev);
856 else
857 iov->dev = dev;
858
859 dev->sriov = iov;
860 dev->is_physfn = 1;
861 rc = compute_max_vf_buses(dev);
862 if (rc)
863 goto fail_max_buses;
864
865 return 0;
866
867fail_max_buses:
868 dev->sriov = NULL;
869 dev->is_physfn = 0;
870failed:
871 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
872 res = &dev->resource[i + PCI_IOV_RESOURCES];
873 res->flags = 0;
874 }
875
876 kfree(iov);
877 return rc;
878}
879
880static void sriov_release(struct pci_dev *dev)
881{
882 BUG_ON(dev->sriov->num_VFs);
883
884 if (dev != dev->sriov->dev)
885 pci_dev_put(dev->sriov->dev);
886
887 kfree(dev->sriov);
888 dev->sriov = NULL;
889}
890
891static void sriov_restore_state(struct pci_dev *dev)
892{
893 int i;
894 u16 ctrl;
895 struct pci_sriov *iov = dev->sriov;
896
897 pci_read_config_word(dev, iov->pos + PCI_SRIOV_CTRL, &ctrl);
898 if (ctrl & PCI_SRIOV_CTRL_VFE)
899 return;
900
901 /*
902 * Restore PCI_SRIOV_CTRL_ARI before pci_iov_set_numvfs() because
903 * it reads offset & stride, which depend on PCI_SRIOV_CTRL_ARI.
904 */
905 ctrl &= ~PCI_SRIOV_CTRL_ARI;
906 ctrl |= iov->ctrl & PCI_SRIOV_CTRL_ARI;
907 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, ctrl);
908
909 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++)
910 pci_update_resource(dev, i + PCI_IOV_RESOURCES);
911
912 pci_write_config_dword(dev, iov->pos + PCI_SRIOV_SYS_PGSIZE, iov->pgsz);
913 pci_iov_set_numvfs(dev, iov->num_VFs);
914 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
915 if (iov->ctrl & PCI_SRIOV_CTRL_VFE)
916 msleep(100);
917}
918
919/**
920 * pci_iov_init - initialize the IOV capability
921 * @dev: the PCI device
922 *
923 * Returns 0 on success, or negative on failure.
924 */
925int pci_iov_init(struct pci_dev *dev)
926{
927 int pos;
928
929 if (!pci_is_pcie(dev))
930 return -ENODEV;
931
932 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_SRIOV);
933 if (pos)
934 return sriov_init(dev, pos);
935
936 return -ENODEV;
937}
938
939/**
940 * pci_iov_release - release resources used by the IOV capability
941 * @dev: the PCI device
942 */
943void pci_iov_release(struct pci_dev *dev)
944{
945 if (dev->is_physfn)
946 sriov_release(dev);
947}
948
949/**
950 * pci_iov_remove - clean up SR-IOV state after PF driver is detached
951 * @dev: the PCI device
952 */
953void pci_iov_remove(struct pci_dev *dev)
954{
955 struct pci_sriov *iov = dev->sriov;
956
957 if (!dev->is_physfn)
958 return;
959
960 iov->driver_max_VFs = iov->total_VFs;
961 if (iov->num_VFs)
962 pci_warn(dev, "driver left SR-IOV enabled after remove\n");
963}
964
965/**
966 * pci_iov_update_resource - update a VF BAR
967 * @dev: the PCI device
968 * @resno: the resource number
969 *
970 * Update a VF BAR in the SR-IOV capability of a PF.
971 */
972void pci_iov_update_resource(struct pci_dev *dev, int resno)
973{
974 struct pci_sriov *iov = dev->is_physfn ? dev->sriov : NULL;
975 struct resource *res = pci_resource_n(dev, resno);
976 int vf_bar = resno - PCI_IOV_RESOURCES;
977 struct pci_bus_region region;
978 u16 cmd;
979 u32 new;
980 int reg;
981
982 /*
983 * The generic pci_restore_bars() path calls this for all devices,
984 * including VFs and non-SR-IOV devices. If this is not a PF, we
985 * have nothing to do.
986 */
987 if (!iov)
988 return;
989
990 pci_read_config_word(dev, iov->pos + PCI_SRIOV_CTRL, &cmd);
991 if ((cmd & PCI_SRIOV_CTRL_VFE) && (cmd & PCI_SRIOV_CTRL_MSE)) {
992 dev_WARN(&dev->dev, "can't update enabled VF BAR%d %pR\n",
993 vf_bar, res);
994 return;
995 }
996
997 /*
998 * Ignore unimplemented BARs, unused resource slots for 64-bit
999 * BARs, and non-movable resources, e.g., those described via
1000 * Enhanced Allocation.
1001 */
1002 if (!res->flags)
1003 return;
1004
1005 if (res->flags & IORESOURCE_UNSET)
1006 return;
1007
1008 if (res->flags & IORESOURCE_PCI_FIXED)
1009 return;
1010
1011 pcibios_resource_to_bus(dev->bus, &region, res);
1012 new = region.start;
1013 new |= res->flags & ~PCI_BASE_ADDRESS_MEM_MASK;
1014
1015 reg = iov->pos + PCI_SRIOV_BAR + 4 * vf_bar;
1016 pci_write_config_dword(dev, reg, new);
1017 if (res->flags & IORESOURCE_MEM_64) {
1018 new = region.start >> 16 >> 16;
1019 pci_write_config_dword(dev, reg + 4, new);
1020 }
1021}
1022
1023resource_size_t __weak pcibios_iov_resource_alignment(struct pci_dev *dev,
1024 int resno)
1025{
1026 return pci_iov_resource_size(dev, resno);
1027}
1028
1029/**
1030 * pci_sriov_resource_alignment - get resource alignment for VF BAR
1031 * @dev: the PCI device
1032 * @resno: the resource number
1033 *
1034 * Returns the alignment of the VF BAR found in the SR-IOV capability.
1035 * This is not the same as the resource size which is defined as
1036 * the VF BAR size multiplied by the number of VFs. The alignment
1037 * is just the VF BAR size.
1038 */
1039resource_size_t pci_sriov_resource_alignment(struct pci_dev *dev, int resno)
1040{
1041 return pcibios_iov_resource_alignment(dev, resno);
1042}
1043
1044/**
1045 * pci_restore_iov_state - restore the state of the IOV capability
1046 * @dev: the PCI device
1047 */
1048void pci_restore_iov_state(struct pci_dev *dev)
1049{
1050 if (dev->is_physfn)
1051 sriov_restore_state(dev);
1052}
1053
1054/**
1055 * pci_vf_drivers_autoprobe - set PF property drivers_autoprobe for VFs
1056 * @dev: the PCI device
1057 * @auto_probe: set VF drivers auto probe flag
1058 */
1059void pci_vf_drivers_autoprobe(struct pci_dev *dev, bool auto_probe)
1060{
1061 if (dev->is_physfn)
1062 dev->sriov->drivers_autoprobe = auto_probe;
1063}
1064
1065/**
1066 * pci_iov_bus_range - find bus range used by Virtual Function
1067 * @bus: the PCI bus
1068 *
1069 * Returns max number of buses (exclude current one) used by Virtual
1070 * Functions.
1071 */
1072int pci_iov_bus_range(struct pci_bus *bus)
1073{
1074 int max = 0;
1075 struct pci_dev *dev;
1076
1077 list_for_each_entry(dev, &bus->devices, bus_list) {
1078 if (!dev->is_physfn)
1079 continue;
1080 if (dev->sriov->max_VF_buses > max)
1081 max = dev->sriov->max_VF_buses;
1082 }
1083
1084 return max ? max - bus->number : 0;
1085}
1086
1087/**
1088 * pci_enable_sriov - enable the SR-IOV capability
1089 * @dev: the PCI device
1090 * @nr_virtfn: number of virtual functions to enable
1091 *
1092 * Returns 0 on success, or negative on failure.
1093 */
1094int pci_enable_sriov(struct pci_dev *dev, int nr_virtfn)
1095{
1096 might_sleep();
1097
1098 if (!dev->is_physfn)
1099 return -ENOSYS;
1100
1101 return sriov_enable(dev, nr_virtfn);
1102}
1103EXPORT_SYMBOL_GPL(pci_enable_sriov);
1104
1105/**
1106 * pci_disable_sriov - disable the SR-IOV capability
1107 * @dev: the PCI device
1108 */
1109void pci_disable_sriov(struct pci_dev *dev)
1110{
1111 might_sleep();
1112
1113 if (!dev->is_physfn)
1114 return;
1115
1116 sriov_disable(dev);
1117}
1118EXPORT_SYMBOL_GPL(pci_disable_sriov);
1119
1120/**
1121 * pci_num_vf - return number of VFs associated with a PF device_release_driver
1122 * @dev: the PCI device
1123 *
1124 * Returns number of VFs, or 0 if SR-IOV is not enabled.
1125 */
1126int pci_num_vf(struct pci_dev *dev)
1127{
1128 if (!dev->is_physfn)
1129 return 0;
1130
1131 return dev->sriov->num_VFs;
1132}
1133EXPORT_SYMBOL_GPL(pci_num_vf);
1134
1135/**
1136 * pci_vfs_assigned - returns number of VFs are assigned to a guest
1137 * @dev: the PCI device
1138 *
1139 * Returns number of VFs belonging to this device that are assigned to a guest.
1140 * If device is not a physical function returns 0.
1141 */
1142int pci_vfs_assigned(struct pci_dev *dev)
1143{
1144 struct pci_dev *vfdev;
1145 unsigned int vfs_assigned = 0;
1146 unsigned short dev_id;
1147
1148 /* only search if we are a PF */
1149 if (!dev->is_physfn)
1150 return 0;
1151
1152 /*
1153 * determine the device ID for the VFs, the vendor ID will be the
1154 * same as the PF so there is no need to check for that one
1155 */
1156 dev_id = dev->sriov->vf_device;
1157
1158 /* loop through all the VFs to see if we own any that are assigned */
1159 vfdev = pci_get_device(dev->vendor, dev_id, NULL);
1160 while (vfdev) {
1161 /*
1162 * It is considered assigned if it is a virtual function with
1163 * our dev as the physical function and the assigned bit is set
1164 */
1165 if (vfdev->is_virtfn && (vfdev->physfn == dev) &&
1166 pci_is_dev_assigned(vfdev))
1167 vfs_assigned++;
1168
1169 vfdev = pci_get_device(dev->vendor, dev_id, vfdev);
1170 }
1171
1172 return vfs_assigned;
1173}
1174EXPORT_SYMBOL_GPL(pci_vfs_assigned);
1175
1176/**
1177 * pci_sriov_set_totalvfs -- reduce the TotalVFs available
1178 * @dev: the PCI PF device
1179 * @numvfs: number that should be used for TotalVFs supported
1180 *
1181 * Should be called from PF driver's probe routine with
1182 * device's mutex held.
1183 *
1184 * Returns 0 if PF is an SRIOV-capable device and
1185 * value of numvfs valid. If not a PF return -ENOSYS;
1186 * if numvfs is invalid return -EINVAL;
1187 * if VFs already enabled, return -EBUSY.
1188 */
1189int pci_sriov_set_totalvfs(struct pci_dev *dev, u16 numvfs)
1190{
1191 if (!dev->is_physfn)
1192 return -ENOSYS;
1193
1194 if (numvfs > dev->sriov->total_VFs)
1195 return -EINVAL;
1196
1197 /* Shouldn't change if VFs already enabled */
1198 if (dev->sriov->ctrl & PCI_SRIOV_CTRL_VFE)
1199 return -EBUSY;
1200
1201 dev->sriov->driver_max_VFs = numvfs;
1202 return 0;
1203}
1204EXPORT_SYMBOL_GPL(pci_sriov_set_totalvfs);
1205
1206/**
1207 * pci_sriov_get_totalvfs -- get total VFs supported on this device
1208 * @dev: the PCI PF device
1209 *
1210 * For a PCIe device with SRIOV support, return the PCIe
1211 * SRIOV capability value of TotalVFs or the value of driver_max_VFs
1212 * if the driver reduced it. Otherwise 0.
1213 */
1214int pci_sriov_get_totalvfs(struct pci_dev *dev)
1215{
1216 if (!dev->is_physfn)
1217 return 0;
1218
1219 return dev->sriov->driver_max_VFs;
1220}
1221EXPORT_SYMBOL_GPL(pci_sriov_get_totalvfs);
1222
1223/**
1224 * pci_sriov_configure_simple - helper to configure SR-IOV
1225 * @dev: the PCI device
1226 * @nr_virtfn: number of virtual functions to enable, 0 to disable
1227 *
1228 * Enable or disable SR-IOV for devices that don't require any PF setup
1229 * before enabling SR-IOV. Return value is negative on error, or number of
1230 * VFs allocated on success.
1231 */
1232int pci_sriov_configure_simple(struct pci_dev *dev, int nr_virtfn)
1233{
1234 int rc;
1235
1236 might_sleep();
1237
1238 if (!dev->is_physfn)
1239 return -ENODEV;
1240
1241 if (pci_vfs_assigned(dev)) {
1242 pci_warn(dev, "Cannot modify SR-IOV while VFs are assigned\n");
1243 return -EPERM;
1244 }
1245
1246 if (nr_virtfn == 0) {
1247 sriov_disable(dev);
1248 return 0;
1249 }
1250
1251 rc = sriov_enable(dev, nr_virtfn);
1252 if (rc < 0)
1253 return rc;
1254
1255 return nr_virtfn;
1256}
1257EXPORT_SYMBOL_GPL(pci_sriov_configure_simple);
Linux 6.x block layer and io_uring tree(s)