Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * Virtio PCI driver - common functionality for all device versions
4 *
5 * This module allows virtio devices to be used over a virtual PCI device.
6 * This can be used with QEMU based VMMs like KVM or Xen.
7 *
8 * Copyright IBM Corp. 2007
9 * Copyright Red Hat, Inc. 2014
10 *
11 * Authors:
12 * Anthony Liguori <aliguori@us.ibm.com>
13 * Rusty Russell <rusty@rustcorp.com.au>
14 * Michael S. Tsirkin <mst@redhat.com>
15 */
16
17#include "virtio_pci_common.h"
18
19static bool force_legacy = false;
20
21#if IS_ENABLED(CONFIG_VIRTIO_PCI_LEGACY)
22module_param(force_legacy, bool, 0444);
23MODULE_PARM_DESC(force_legacy,
24 "Force legacy mode for transitional virtio 1 devices");
25#endif
26
27bool vp_is_avq(struct virtio_device *vdev, unsigned int index)
28{
29 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
30
31 if (!virtio_has_feature(vdev, VIRTIO_F_ADMIN_VQ))
32 return false;
33
34 return index == vp_dev->admin_vq.vq_index;
35}
36
37/* wait for pending irq handlers */
38void vp_synchronize_vectors(struct virtio_device *vdev)
39{
40 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
41 int i;
42
43 if (vp_dev->intx_enabled)
44 synchronize_irq(vp_dev->pci_dev->irq);
45
46 for (i = 0; i < vp_dev->msix_vectors; ++i)
47 synchronize_irq(pci_irq_vector(vp_dev->pci_dev, i));
48}
49
50/* the notify function used when creating a virt queue */
51bool vp_notify(struct virtqueue *vq)
52{
53 /* we write the queue's selector into the notification register to
54 * signal the other end */
55 iowrite16(vq->index, (void __iomem *)vq->priv);
56 return true;
57}
58
59/* Notify all slow path virtqueues on an interrupt. */
60static void vp_vring_slow_path_interrupt(int irq,
61 struct virtio_pci_device *vp_dev)
62{
63 struct virtio_pci_vq_info *info;
64 unsigned long flags;
65
66 spin_lock_irqsave(&vp_dev->lock, flags);
67 list_for_each_entry(info, &vp_dev->slow_virtqueues, node)
68 vring_interrupt(irq, info->vq);
69 spin_unlock_irqrestore(&vp_dev->lock, flags);
70}
71
72/* Handle a configuration change: Tell driver if it wants to know. */
73static irqreturn_t vp_config_changed(int irq, void *opaque)
74{
75 struct virtio_pci_device *vp_dev = opaque;
76
77 virtio_config_changed(&vp_dev->vdev);
78 vp_vring_slow_path_interrupt(irq, vp_dev);
79 return IRQ_HANDLED;
80}
81
82/* Notify all virtqueues on an interrupt. */
83static irqreturn_t vp_vring_interrupt(int irq, void *opaque)
84{
85 struct virtio_pci_device *vp_dev = opaque;
86 struct virtio_pci_vq_info *info;
87 irqreturn_t ret = IRQ_NONE;
88 unsigned long flags;
89
90 spin_lock_irqsave(&vp_dev->lock, flags);
91 list_for_each_entry(info, &vp_dev->virtqueues, node) {
92 if (vring_interrupt(irq, info->vq) == IRQ_HANDLED)
93 ret = IRQ_HANDLED;
94 }
95 spin_unlock_irqrestore(&vp_dev->lock, flags);
96
97 return ret;
98}
99
100/* A small wrapper to also acknowledge the interrupt when it's handled.
101 * I really need an EIO hook for the vring so I can ack the interrupt once we
102 * know that we'll be handling the IRQ but before we invoke the callback since
103 * the callback may notify the host which results in the host attempting to
104 * raise an interrupt that we would then mask once we acknowledged the
105 * interrupt. */
106static irqreturn_t vp_interrupt(int irq, void *opaque)
107{
108 struct virtio_pci_device *vp_dev = opaque;
109 u8 isr;
110
111 /* reading the ISR has the effect of also clearing it so it's very
112 * important to save off the value. */
113 isr = ioread8(vp_dev->isr);
114
115 /* It's definitely not us if the ISR was not high */
116 if (!isr)
117 return IRQ_NONE;
118
119 /* Configuration change? Tell driver if it wants to know. */
120 if (isr & VIRTIO_PCI_ISR_CONFIG)
121 vp_config_changed(irq, opaque);
122
123 return vp_vring_interrupt(irq, opaque);
124}
125
126static int vp_request_msix_vectors(struct virtio_device *vdev, int nvectors,
127 bool per_vq_vectors, struct irq_affinity *desc)
128{
129 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
130 const char *name = dev_name(&vp_dev->vdev.dev);
131 unsigned int flags = PCI_IRQ_MSIX;
132 unsigned int i, v;
133 int err = -ENOMEM;
134
135 vp_dev->msix_vectors = nvectors;
136
137 vp_dev->msix_names = kmalloc_array(nvectors,
138 sizeof(*vp_dev->msix_names),
139 GFP_KERNEL);
140 if (!vp_dev->msix_names)
141 goto error;
142 vp_dev->msix_affinity_masks
143 = kcalloc(nvectors, sizeof(*vp_dev->msix_affinity_masks),
144 GFP_KERNEL);
145 if (!vp_dev->msix_affinity_masks)
146 goto error;
147 for (i = 0; i < nvectors; ++i)
148 if (!alloc_cpumask_var(&vp_dev->msix_affinity_masks[i],
149 GFP_KERNEL))
150 goto error;
151
152 if (!per_vq_vectors)
153 desc = NULL;
154
155 if (desc) {
156 flags |= PCI_IRQ_AFFINITY;
157 desc->pre_vectors++; /* virtio config vector */
158 }
159
160 err = pci_alloc_irq_vectors_affinity(vp_dev->pci_dev, nvectors,
161 nvectors, flags, desc);
162 if (err < 0)
163 goto error;
164 vp_dev->msix_enabled = 1;
165
166 /* Set the vector used for configuration */
167 v = vp_dev->msix_used_vectors;
168 snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names,
169 "%s-config", name);
170 err = request_irq(pci_irq_vector(vp_dev->pci_dev, v),
171 vp_config_changed, 0, vp_dev->msix_names[v],
172 vp_dev);
173 if (err)
174 goto error;
175 ++vp_dev->msix_used_vectors;
176
177 v = vp_dev->config_vector(vp_dev, v);
178 /* Verify we had enough resources to assign the vector */
179 if (v == VIRTIO_MSI_NO_VECTOR) {
180 err = -EBUSY;
181 goto error;
182 }
183
184 if (!per_vq_vectors) {
185 /* Shared vector for all VQs */
186 v = vp_dev->msix_used_vectors;
187 snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names,
188 "%s-virtqueues", name);
189 err = request_irq(pci_irq_vector(vp_dev->pci_dev, v),
190 vp_vring_interrupt, 0, vp_dev->msix_names[v],
191 vp_dev);
192 if (err)
193 goto error;
194 ++vp_dev->msix_used_vectors;
195 }
196 return 0;
197error:
198 return err;
199}
200
201static bool vp_is_slow_path_vector(u16 msix_vec)
202{
203 return msix_vec == VP_MSIX_CONFIG_VECTOR;
204}
205
206static struct virtqueue *vp_setup_vq(struct virtio_device *vdev, unsigned int index,
207 void (*callback)(struct virtqueue *vq),
208 const char *name,
209 bool ctx,
210 u16 msix_vec,
211 struct virtio_pci_vq_info **p_info)
212{
213 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
214 struct virtio_pci_vq_info *info = kmalloc(sizeof *info, GFP_KERNEL);
215 struct virtqueue *vq;
216 unsigned long flags;
217
218 /* fill out our structure that represents an active queue */
219 if (!info)
220 return ERR_PTR(-ENOMEM);
221
222 vq = vp_dev->setup_vq(vp_dev, info, index, callback, name, ctx,
223 msix_vec);
224 if (IS_ERR(vq))
225 goto out_info;
226
227 info->vq = vq;
228 if (callback) {
229 spin_lock_irqsave(&vp_dev->lock, flags);
230 if (!vp_is_slow_path_vector(msix_vec))
231 list_add(&info->node, &vp_dev->virtqueues);
232 else
233 list_add(&info->node, &vp_dev->slow_virtqueues);
234 spin_unlock_irqrestore(&vp_dev->lock, flags);
235 } else {
236 INIT_LIST_HEAD(&info->node);
237 }
238
239 *p_info = info;
240 return vq;
241
242out_info:
243 kfree(info);
244 return vq;
245}
246
247static void vp_del_vq(struct virtqueue *vq, struct virtio_pci_vq_info *info)
248{
249 struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
250 unsigned long flags;
251
252 /*
253 * If it fails during re-enable reset vq. This way we won't rejoin
254 * info->node to the queue. Prevent unexpected irqs.
255 */
256 if (!vq->reset) {
257 spin_lock_irqsave(&vp_dev->lock, flags);
258 list_del(&info->node);
259 spin_unlock_irqrestore(&vp_dev->lock, flags);
260 }
261
262 vp_dev->del_vq(info);
263 kfree(info);
264}
265
266/* the config->del_vqs() implementation */
267void vp_del_vqs(struct virtio_device *vdev)
268{
269 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
270 struct virtio_pci_vq_info *info;
271 struct virtqueue *vq, *n;
272 int i;
273
274 list_for_each_entry_safe(vq, n, &vdev->vqs, list) {
275 info = vp_is_avq(vdev, vq->index) ? vp_dev->admin_vq.info :
276 vp_dev->vqs[vq->index];
277
278 if (vp_dev->per_vq_vectors) {
279 int v = info->msix_vector;
280 if (v != VIRTIO_MSI_NO_VECTOR &&
281 !vp_is_slow_path_vector(v)) {
282 int irq = pci_irq_vector(vp_dev->pci_dev, v);
283
284 irq_update_affinity_hint(irq, NULL);
285 free_irq(irq, vq);
286 }
287 }
288 vp_del_vq(vq, info);
289 }
290 vp_dev->per_vq_vectors = false;
291
292 if (vp_dev->intx_enabled) {
293 free_irq(vp_dev->pci_dev->irq, vp_dev);
294 vp_dev->intx_enabled = 0;
295 }
296
297 for (i = 0; i < vp_dev->msix_used_vectors; ++i)
298 free_irq(pci_irq_vector(vp_dev->pci_dev, i), vp_dev);
299
300 if (vp_dev->msix_affinity_masks) {
301 for (i = 0; i < vp_dev->msix_vectors; i++)
302 free_cpumask_var(vp_dev->msix_affinity_masks[i]);
303 }
304
305 if (vp_dev->msix_enabled) {
306 /* Disable the vector used for configuration */
307 vp_dev->config_vector(vp_dev, VIRTIO_MSI_NO_VECTOR);
308
309 pci_free_irq_vectors(vp_dev->pci_dev);
310 vp_dev->msix_enabled = 0;
311 }
312
313 vp_dev->msix_vectors = 0;
314 vp_dev->msix_used_vectors = 0;
315 kfree(vp_dev->msix_names);
316 vp_dev->msix_names = NULL;
317 kfree(vp_dev->msix_affinity_masks);
318 vp_dev->msix_affinity_masks = NULL;
319 kfree(vp_dev->vqs);
320 vp_dev->vqs = NULL;
321}
322
323enum vp_vq_vector_policy {
324 VP_VQ_VECTOR_POLICY_EACH,
325 VP_VQ_VECTOR_POLICY_SHARED_SLOW,
326 VP_VQ_VECTOR_POLICY_SHARED,
327};
328
329static struct virtqueue *
330vp_find_one_vq_msix(struct virtio_device *vdev, int queue_idx,
331 vq_callback_t *callback, const char *name, bool ctx,
332 bool slow_path, int *allocated_vectors,
333 enum vp_vq_vector_policy vector_policy,
334 struct virtio_pci_vq_info **p_info)
335{
336 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
337 struct virtqueue *vq;
338 u16 msix_vec;
339 int err;
340
341 if (!callback)
342 msix_vec = VIRTIO_MSI_NO_VECTOR;
343 else if (vector_policy == VP_VQ_VECTOR_POLICY_EACH ||
344 (vector_policy == VP_VQ_VECTOR_POLICY_SHARED_SLOW &&
345 !slow_path))
346 msix_vec = (*allocated_vectors)++;
347 else if (vector_policy != VP_VQ_VECTOR_POLICY_EACH &&
348 slow_path)
349 msix_vec = VP_MSIX_CONFIG_VECTOR;
350 else
351 msix_vec = VP_MSIX_VQ_VECTOR;
352 vq = vp_setup_vq(vdev, queue_idx, callback, name, ctx, msix_vec,
353 p_info);
354 if (IS_ERR(vq))
355 return vq;
356
357 if (vector_policy == VP_VQ_VECTOR_POLICY_SHARED ||
358 msix_vec == VIRTIO_MSI_NO_VECTOR ||
359 vp_is_slow_path_vector(msix_vec))
360 return vq;
361
362 /* allocate per-vq irq if available and necessary */
363 snprintf(vp_dev->msix_names[msix_vec], sizeof(*vp_dev->msix_names),
364 "%s-%s", dev_name(&vp_dev->vdev.dev), name);
365 err = request_irq(pci_irq_vector(vp_dev->pci_dev, msix_vec),
366 vring_interrupt, 0,
367 vp_dev->msix_names[msix_vec], vq);
368 if (err) {
369 vp_del_vq(vq, *p_info);
370 return ERR_PTR(err);
371 }
372
373 return vq;
374}
375
376static int vp_find_vqs_msix(struct virtio_device *vdev, unsigned int nvqs,
377 struct virtqueue *vqs[],
378 struct virtqueue_info vqs_info[],
379 enum vp_vq_vector_policy vector_policy,
380 struct irq_affinity *desc)
381{
382 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
383 struct virtio_pci_admin_vq *avq = &vp_dev->admin_vq;
384 struct virtqueue_info *vqi;
385 int i, err, nvectors, allocated_vectors, queue_idx = 0;
386 struct virtqueue *vq;
387 bool per_vq_vectors;
388 u16 avq_num = 0;
389
390 vp_dev->vqs = kcalloc(nvqs, sizeof(*vp_dev->vqs), GFP_KERNEL);
391 if (!vp_dev->vqs)
392 return -ENOMEM;
393
394 if (vp_dev->avq_index) {
395 err = vp_dev->avq_index(vdev, &avq->vq_index, &avq_num);
396 if (err)
397 goto error_find;
398 }
399
400 per_vq_vectors = vector_policy != VP_VQ_VECTOR_POLICY_SHARED;
401
402 if (per_vq_vectors) {
403 /* Best option: one for change interrupt, one per vq. */
404 nvectors = 1;
405 for (i = 0; i < nvqs; ++i) {
406 vqi = &vqs_info[i];
407 if (vqi->name && vqi->callback)
408 ++nvectors;
409 }
410 if (avq_num && vector_policy == VP_VQ_VECTOR_POLICY_EACH)
411 ++nvectors;
412 } else {
413 /* Second best: one for change, shared for all vqs. */
414 nvectors = 2;
415 }
416
417 err = vp_request_msix_vectors(vdev, nvectors, per_vq_vectors, desc);
418 if (err)
419 goto error_find;
420
421 vp_dev->per_vq_vectors = per_vq_vectors;
422 allocated_vectors = vp_dev->msix_used_vectors;
423 for (i = 0; i < nvqs; ++i) {
424 vqi = &vqs_info[i];
425 if (!vqi->name) {
426 vqs[i] = NULL;
427 continue;
428 }
429 vqs[i] = vp_find_one_vq_msix(vdev, queue_idx++, vqi->callback,
430 vqi->name, vqi->ctx, false,
431 &allocated_vectors, vector_policy,
432 &vp_dev->vqs[i]);
433 if (IS_ERR(vqs[i])) {
434 err = PTR_ERR(vqs[i]);
435 goto error_find;
436 }
437 }
438
439 if (!avq_num)
440 return 0;
441 sprintf(avq->name, "avq.%u", avq->vq_index);
442 vq = vp_find_one_vq_msix(vdev, avq->vq_index, vp_modern_avq_done,
443 avq->name, false, true, &allocated_vectors,
444 vector_policy, &vp_dev->admin_vq.info);
445 if (IS_ERR(vq)) {
446 err = PTR_ERR(vq);
447 goto error_find;
448 }
449
450 return 0;
451
452error_find:
453 vp_del_vqs(vdev);
454 return err;
455}
456
457static int vp_find_vqs_intx(struct virtio_device *vdev, unsigned int nvqs,
458 struct virtqueue *vqs[],
459 struct virtqueue_info vqs_info[])
460{
461 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
462 struct virtio_pci_admin_vq *avq = &vp_dev->admin_vq;
463 int i, err, queue_idx = 0;
464 struct virtqueue *vq;
465 u16 avq_num = 0;
466
467 vp_dev->vqs = kcalloc(nvqs, sizeof(*vp_dev->vqs), GFP_KERNEL);
468 if (!vp_dev->vqs)
469 return -ENOMEM;
470
471 if (vp_dev->avq_index) {
472 err = vp_dev->avq_index(vdev, &avq->vq_index, &avq_num);
473 if (err)
474 goto out_del_vqs;
475 }
476
477 err = request_irq(vp_dev->pci_dev->irq, vp_interrupt, IRQF_SHARED,
478 dev_name(&vdev->dev), vp_dev);
479 if (err)
480 goto out_del_vqs;
481
482 vp_dev->intx_enabled = 1;
483 vp_dev->per_vq_vectors = false;
484 for (i = 0; i < nvqs; ++i) {
485 struct virtqueue_info *vqi = &vqs_info[i];
486
487 if (!vqi->name) {
488 vqs[i] = NULL;
489 continue;
490 }
491 vqs[i] = vp_setup_vq(vdev, queue_idx++, vqi->callback,
492 vqi->name, vqi->ctx,
493 VIRTIO_MSI_NO_VECTOR, &vp_dev->vqs[i]);
494 if (IS_ERR(vqs[i])) {
495 err = PTR_ERR(vqs[i]);
496 goto out_del_vqs;
497 }
498 }
499
500 if (!avq_num)
501 return 0;
502 sprintf(avq->name, "avq.%u", avq->vq_index);
503 vq = vp_setup_vq(vdev, queue_idx++, vp_modern_avq_done, avq->name,
504 false, VIRTIO_MSI_NO_VECTOR,
505 &vp_dev->admin_vq.info);
506 if (IS_ERR(vq)) {
507 err = PTR_ERR(vq);
508 goto out_del_vqs;
509 }
510
511 return 0;
512out_del_vqs:
513 vp_del_vqs(vdev);
514 return err;
515}
516
517/* the config->find_vqs() implementation */
518int vp_find_vqs(struct virtio_device *vdev, unsigned int nvqs,
519 struct virtqueue *vqs[], struct virtqueue_info vqs_info[],
520 struct irq_affinity *desc)
521{
522 int err;
523
524 /* Try MSI-X with one vector per queue. */
525 err = vp_find_vqs_msix(vdev, nvqs, vqs, vqs_info,
526 VP_VQ_VECTOR_POLICY_EACH, desc);
527 if (!err)
528 return 0;
529 /* Fallback: MSI-X with one shared vector for config and
530 * slow path queues, one vector per queue for the rest.
531 */
532 err = vp_find_vqs_msix(vdev, nvqs, vqs, vqs_info,
533 VP_VQ_VECTOR_POLICY_SHARED_SLOW, desc);
534 if (!err)
535 return 0;
536 /* Fallback: MSI-X with one vector for config, one shared for queues. */
537 err = vp_find_vqs_msix(vdev, nvqs, vqs, vqs_info,
538 VP_VQ_VECTOR_POLICY_SHARED, desc);
539 if (!err)
540 return 0;
541 /* Is there an interrupt? If not give up. */
542 if (!(to_vp_device(vdev)->pci_dev->irq))
543 return err;
544 /* Finally fall back to regular interrupts. */
545 return vp_find_vqs_intx(vdev, nvqs, vqs, vqs_info);
546}
547
548const char *vp_bus_name(struct virtio_device *vdev)
549{
550 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
551
552 return pci_name(vp_dev->pci_dev);
553}
554
555/* Setup the affinity for a virtqueue:
556 * - force the affinity for per vq vector
557 * - OR over all affinities for shared MSI
558 * - ignore the affinity request if we're using INTX
559 */
560int vp_set_vq_affinity(struct virtqueue *vq, const struct cpumask *cpu_mask)
561{
562 struct virtio_device *vdev = vq->vdev;
563 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
564 struct virtio_pci_vq_info *info = vp_dev->vqs[vq->index];
565 struct cpumask *mask;
566 unsigned int irq;
567
568 if (!vq->callback)
569 return -EINVAL;
570
571 if (vp_dev->msix_enabled) {
572 mask = vp_dev->msix_affinity_masks[info->msix_vector];
573 irq = pci_irq_vector(vp_dev->pci_dev, info->msix_vector);
574 if (!cpu_mask)
575 irq_update_affinity_hint(irq, NULL);
576 else {
577 cpumask_copy(mask, cpu_mask);
578 irq_set_affinity_and_hint(irq, mask);
579 }
580 }
581 return 0;
582}
583
584const struct cpumask *vp_get_vq_affinity(struct virtio_device *vdev, int index)
585{
586 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
587
588 if (!vp_dev->per_vq_vectors ||
589 vp_dev->vqs[index]->msix_vector == VIRTIO_MSI_NO_VECTOR ||
590 vp_is_slow_path_vector(vp_dev->vqs[index]->msix_vector))
591 return NULL;
592
593 return pci_irq_get_affinity(vp_dev->pci_dev,
594 vp_dev->vqs[index]->msix_vector);
595}
596
597#ifdef CONFIG_PM_SLEEP
598static int virtio_pci_freeze(struct device *dev)
599{
600 struct pci_dev *pci_dev = to_pci_dev(dev);
601 struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
602 int ret;
603
604 ret = virtio_device_freeze(&vp_dev->vdev);
605
606 if (!ret)
607 pci_disable_device(pci_dev);
608 return ret;
609}
610
611static int virtio_pci_restore(struct device *dev)
612{
613 struct pci_dev *pci_dev = to_pci_dev(dev);
614 struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
615 int ret;
616
617 ret = pci_enable_device(pci_dev);
618 if (ret)
619 return ret;
620
621 pci_set_master(pci_dev);
622 return virtio_device_restore(&vp_dev->vdev);
623}
624
625static bool vp_supports_pm_no_reset(struct device *dev)
626{
627 struct pci_dev *pci_dev = to_pci_dev(dev);
628 u16 pmcsr;
629
630 if (!pci_dev->pm_cap)
631 return false;
632
633 pci_read_config_word(pci_dev, pci_dev->pm_cap + PCI_PM_CTRL, &pmcsr);
634 if (PCI_POSSIBLE_ERROR(pmcsr)) {
635 dev_err(dev, "Unable to query pmcsr");
636 return false;
637 }
638
639 return pmcsr & PCI_PM_CTRL_NO_SOFT_RESET;
640}
641
642static int virtio_pci_suspend(struct device *dev)
643{
644 return vp_supports_pm_no_reset(dev) ? 0 : virtio_pci_freeze(dev);
645}
646
647static int virtio_pci_resume(struct device *dev)
648{
649 return vp_supports_pm_no_reset(dev) ? 0 : virtio_pci_restore(dev);
650}
651
652static const struct dev_pm_ops virtio_pci_pm_ops = {
653 .suspend = virtio_pci_suspend,
654 .resume = virtio_pci_resume,
655 .freeze = virtio_pci_freeze,
656 .thaw = virtio_pci_restore,
657 .poweroff = virtio_pci_freeze,
658 .restore = virtio_pci_restore,
659};
660#endif
661
662
663/* Qumranet donated their vendor ID for devices 0x1000 thru 0x10FF. */
664static const struct pci_device_id virtio_pci_id_table[] = {
665 { PCI_DEVICE(PCI_VENDOR_ID_REDHAT_QUMRANET, PCI_ANY_ID) },
666 { 0 }
667};
668
669MODULE_DEVICE_TABLE(pci, virtio_pci_id_table);
670
671static void virtio_pci_release_dev(struct device *_d)
672{
673 struct virtio_device *vdev = dev_to_virtio(_d);
674 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
675
676 /* As struct device is a kobject, it's not safe to
677 * free the memory (including the reference counter itself)
678 * until it's release callback. */
679 kfree(vp_dev);
680}
681
682static int virtio_pci_probe(struct pci_dev *pci_dev,
683 const struct pci_device_id *id)
684{
685 struct virtio_pci_device *vp_dev, *reg_dev = NULL;
686 int rc;
687
688 /* allocate our structure and fill it out */
689 vp_dev = kzalloc(sizeof(struct virtio_pci_device), GFP_KERNEL);
690 if (!vp_dev)
691 return -ENOMEM;
692
693 pci_set_drvdata(pci_dev, vp_dev);
694 vp_dev->vdev.dev.parent = &pci_dev->dev;
695 vp_dev->vdev.dev.release = virtio_pci_release_dev;
696 vp_dev->pci_dev = pci_dev;
697 INIT_LIST_HEAD(&vp_dev->virtqueues);
698 INIT_LIST_HEAD(&vp_dev->slow_virtqueues);
699 spin_lock_init(&vp_dev->lock);
700
701 /* enable the device */
702 rc = pci_enable_device(pci_dev);
703 if (rc)
704 goto err_enable_device;
705
706 if (force_legacy) {
707 rc = virtio_pci_legacy_probe(vp_dev);
708 /* Also try modern mode if we can't map BAR0 (no IO space). */
709 if (rc == -ENODEV || rc == -ENOMEM)
710 rc = virtio_pci_modern_probe(vp_dev);
711 if (rc)
712 goto err_probe;
713 } else {
714 rc = virtio_pci_modern_probe(vp_dev);
715 if (rc == -ENODEV)
716 rc = virtio_pci_legacy_probe(vp_dev);
717 if (rc)
718 goto err_probe;
719 }
720
721 pci_set_master(pci_dev);
722
723 rc = register_virtio_device(&vp_dev->vdev);
724 reg_dev = vp_dev;
725 if (rc)
726 goto err_register;
727
728 return 0;
729
730err_register:
731 if (vp_dev->is_legacy)
732 virtio_pci_legacy_remove(vp_dev);
733 else
734 virtio_pci_modern_remove(vp_dev);
735err_probe:
736 pci_disable_device(pci_dev);
737err_enable_device:
738 if (reg_dev)
739 put_device(&vp_dev->vdev.dev);
740 else
741 kfree(vp_dev);
742 return rc;
743}
744
745static void virtio_pci_remove(struct pci_dev *pci_dev)
746{
747 struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
748 struct device *dev = get_device(&vp_dev->vdev.dev);
749
750 /*
751 * Device is marked broken on surprise removal so that virtio upper
752 * layers can abort any ongoing operation.
753 */
754 if (!pci_device_is_present(pci_dev))
755 virtio_break_device(&vp_dev->vdev);
756
757 pci_disable_sriov(pci_dev);
758
759 unregister_virtio_device(&vp_dev->vdev);
760
761 if (vp_dev->is_legacy)
762 virtio_pci_legacy_remove(vp_dev);
763 else
764 virtio_pci_modern_remove(vp_dev);
765
766 pci_disable_device(pci_dev);
767 put_device(dev);
768}
769
770static int virtio_pci_sriov_configure(struct pci_dev *pci_dev, int num_vfs)
771{
772 struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
773 struct virtio_device *vdev = &vp_dev->vdev;
774 int ret;
775
776 if (!(vdev->config->get_status(vdev) & VIRTIO_CONFIG_S_DRIVER_OK))
777 return -EBUSY;
778
779 if (!__virtio_test_bit(vdev, VIRTIO_F_SR_IOV))
780 return -EINVAL;
781
782 if (pci_vfs_assigned(pci_dev))
783 return -EPERM;
784
785 if (num_vfs == 0) {
786 pci_disable_sriov(pci_dev);
787 return 0;
788 }
789
790 ret = pci_enable_sriov(pci_dev, num_vfs);
791 if (ret < 0)
792 return ret;
793
794 return num_vfs;
795}
796
797static struct pci_driver virtio_pci_driver = {
798 .name = "virtio-pci",
799 .id_table = virtio_pci_id_table,
800 .probe = virtio_pci_probe,
801 .remove = virtio_pci_remove,
802#ifdef CONFIG_PM_SLEEP
803 .driver.pm = &virtio_pci_pm_ops,
804#endif
805 .sriov_configure = virtio_pci_sriov_configure,
806};
807
808struct virtio_device *virtio_pci_vf_get_pf_dev(struct pci_dev *pdev)
809{
810 struct virtio_pci_device *pf_vp_dev;
811
812 pf_vp_dev = pci_iov_get_pf_drvdata(pdev, &virtio_pci_driver);
813 if (IS_ERR(pf_vp_dev))
814 return NULL;
815
816 return &pf_vp_dev->vdev;
817}
818
819module_pci_driver(virtio_pci_driver);
820
821MODULE_AUTHOR("Anthony Liguori <aliguori@us.ibm.com>");
822MODULE_DESCRIPTION("virtio-pci");
823MODULE_LICENSE("GPL");
824MODULE_VERSION("1");
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * Virtio PCI driver - common functionality for all device versions
4 *
5 * This module allows virtio devices to be used over a virtual PCI device.
6 * This can be used with QEMU based VMMs like KVM or Xen.
7 *
8 * Copyright IBM Corp. 2007
9 * Copyright Red Hat, Inc. 2014
10 *
11 * Authors:
12 * Anthony Liguori <aliguori@us.ibm.com>
13 * Rusty Russell <rusty@rustcorp.com.au>
14 * Michael S. Tsirkin <mst@redhat.com>
15 */
16
17#include "virtio_pci_common.h"
18
19static bool force_legacy = false;
20
21#if IS_ENABLED(CONFIG_VIRTIO_PCI_LEGACY)
22module_param(force_legacy, bool, 0444);
23MODULE_PARM_DESC(force_legacy,
24 "Force legacy mode for transitional virtio 1 devices");
25#endif
26
27/* wait for pending irq handlers */
28void vp_synchronize_vectors(struct virtio_device *vdev)
29{
30 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
31 int i;
32
33 if (vp_dev->intx_enabled)
34 synchronize_irq(vp_dev->pci_dev->irq);
35
36 for (i = 0; i < vp_dev->msix_vectors; ++i)
37 synchronize_irq(pci_irq_vector(vp_dev->pci_dev, i));
38}
39
40/* the notify function used when creating a virt queue */
41bool vp_notify(struct virtqueue *vq)
42{
43 /* we write the queue's selector into the notification register to
44 * signal the other end */
45 iowrite16(vq->index, (void __iomem *)vq->priv);
46 return true;
47}
48
49/* Handle a configuration change: Tell driver if it wants to know. */
50static irqreturn_t vp_config_changed(int irq, void *opaque)
51{
52 struct virtio_pci_device *vp_dev = opaque;
53
54 virtio_config_changed(&vp_dev->vdev);
55 return IRQ_HANDLED;
56}
57
58/* Notify all virtqueues on an interrupt. */
59static irqreturn_t vp_vring_interrupt(int irq, void *opaque)
60{
61 struct virtio_pci_device *vp_dev = opaque;
62 struct virtio_pci_vq_info *info;
63 irqreturn_t ret = IRQ_NONE;
64 unsigned long flags;
65
66 spin_lock_irqsave(&vp_dev->lock, flags);
67 list_for_each_entry(info, &vp_dev->virtqueues, node) {
68 if (vring_interrupt(irq, info->vq) == IRQ_HANDLED)
69 ret = IRQ_HANDLED;
70 }
71 spin_unlock_irqrestore(&vp_dev->lock, flags);
72
73 return ret;
74}
75
76/* A small wrapper to also acknowledge the interrupt when it's handled.
77 * I really need an EIO hook for the vring so I can ack the interrupt once we
78 * know that we'll be handling the IRQ but before we invoke the callback since
79 * the callback may notify the host which results in the host attempting to
80 * raise an interrupt that we would then mask once we acknowledged the
81 * interrupt. */
82static irqreturn_t vp_interrupt(int irq, void *opaque)
83{
84 struct virtio_pci_device *vp_dev = opaque;
85 u8 isr;
86
87 /* reading the ISR has the effect of also clearing it so it's very
88 * important to save off the value. */
89 isr = ioread8(vp_dev->isr);
90
91 /* It's definitely not us if the ISR was not high */
92 if (!isr)
93 return IRQ_NONE;
94
95 /* Configuration change? Tell driver if it wants to know. */
96 if (isr & VIRTIO_PCI_ISR_CONFIG)
97 vp_config_changed(irq, opaque);
98
99 return vp_vring_interrupt(irq, opaque);
100}
101
102static int vp_request_msix_vectors(struct virtio_device *vdev, int nvectors,
103 bool per_vq_vectors, struct irq_affinity *desc)
104{
105 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
106 const char *name = dev_name(&vp_dev->vdev.dev);
107 unsigned int flags = PCI_IRQ_MSIX;
108 unsigned int i, v;
109 int err = -ENOMEM;
110
111 vp_dev->msix_vectors = nvectors;
112
113 vp_dev->msix_names = kmalloc_array(nvectors,
114 sizeof(*vp_dev->msix_names),
115 GFP_KERNEL);
116 if (!vp_dev->msix_names)
117 goto error;
118 vp_dev->msix_affinity_masks
119 = kcalloc(nvectors, sizeof(*vp_dev->msix_affinity_masks),
120 GFP_KERNEL);
121 if (!vp_dev->msix_affinity_masks)
122 goto error;
123 for (i = 0; i < nvectors; ++i)
124 if (!alloc_cpumask_var(&vp_dev->msix_affinity_masks[i],
125 GFP_KERNEL))
126 goto error;
127
128 if (desc) {
129 flags |= PCI_IRQ_AFFINITY;
130 desc->pre_vectors++; /* virtio config vector */
131 }
132
133 err = pci_alloc_irq_vectors_affinity(vp_dev->pci_dev, nvectors,
134 nvectors, flags, desc);
135 if (err < 0)
136 goto error;
137 vp_dev->msix_enabled = 1;
138
139 /* Set the vector used for configuration */
140 v = vp_dev->msix_used_vectors;
141 snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names,
142 "%s-config", name);
143 err = request_irq(pci_irq_vector(vp_dev->pci_dev, v),
144 vp_config_changed, 0, vp_dev->msix_names[v],
145 vp_dev);
146 if (err)
147 goto error;
148 ++vp_dev->msix_used_vectors;
149
150 v = vp_dev->config_vector(vp_dev, v);
151 /* Verify we had enough resources to assign the vector */
152 if (v == VIRTIO_MSI_NO_VECTOR) {
153 err = -EBUSY;
154 goto error;
155 }
156
157 if (!per_vq_vectors) {
158 /* Shared vector for all VQs */
159 v = vp_dev->msix_used_vectors;
160 snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names,
161 "%s-virtqueues", name);
162 err = request_irq(pci_irq_vector(vp_dev->pci_dev, v),
163 vp_vring_interrupt, 0, vp_dev->msix_names[v],
164 vp_dev);
165 if (err)
166 goto error;
167 ++vp_dev->msix_used_vectors;
168 }
169 return 0;
170error:
171 return err;
172}
173
174static struct virtqueue *vp_setup_vq(struct virtio_device *vdev, unsigned int index,
175 void (*callback)(struct virtqueue *vq),
176 const char *name,
177 bool ctx,
178 u16 msix_vec)
179{
180 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
181 struct virtio_pci_vq_info *info = kmalloc(sizeof *info, GFP_KERNEL);
182 struct virtqueue *vq;
183 unsigned long flags;
184
185 /* fill out our structure that represents an active queue */
186 if (!info)
187 return ERR_PTR(-ENOMEM);
188
189 vq = vp_dev->setup_vq(vp_dev, info, index, callback, name, ctx,
190 msix_vec);
191 if (IS_ERR(vq))
192 goto out_info;
193
194 info->vq = vq;
195 if (callback) {
196 spin_lock_irqsave(&vp_dev->lock, flags);
197 list_add(&info->node, &vp_dev->virtqueues);
198 spin_unlock_irqrestore(&vp_dev->lock, flags);
199 } else {
200 INIT_LIST_HEAD(&info->node);
201 }
202
203 vp_dev->vqs[index] = info;
204 return vq;
205
206out_info:
207 kfree(info);
208 return vq;
209}
210
211static void vp_del_vq(struct virtqueue *vq)
212{
213 struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
214 struct virtio_pci_vq_info *info = vp_dev->vqs[vq->index];
215 unsigned long flags;
216
217 /*
218 * If it fails during re-enable reset vq. This way we won't rejoin
219 * info->node to the queue. Prevent unexpected irqs.
220 */
221 if (!vq->reset) {
222 spin_lock_irqsave(&vp_dev->lock, flags);
223 list_del(&info->node);
224 spin_unlock_irqrestore(&vp_dev->lock, flags);
225 }
226
227 vp_dev->del_vq(info);
228 kfree(info);
229}
230
231/* the config->del_vqs() implementation */
232void vp_del_vqs(struct virtio_device *vdev)
233{
234 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
235 struct virtqueue *vq, *n;
236 int i;
237
238 list_for_each_entry_safe(vq, n, &vdev->vqs, list) {
239 if (vp_dev->is_avq(vdev, vq->index))
240 continue;
241
242 if (vp_dev->per_vq_vectors) {
243 int v = vp_dev->vqs[vq->index]->msix_vector;
244
245 if (v != VIRTIO_MSI_NO_VECTOR) {
246 int irq = pci_irq_vector(vp_dev->pci_dev, v);
247
248 irq_update_affinity_hint(irq, NULL);
249 free_irq(irq, vq);
250 }
251 }
252 vp_del_vq(vq);
253 }
254 vp_dev->per_vq_vectors = false;
255
256 if (vp_dev->intx_enabled) {
257 free_irq(vp_dev->pci_dev->irq, vp_dev);
258 vp_dev->intx_enabled = 0;
259 }
260
261 for (i = 0; i < vp_dev->msix_used_vectors; ++i)
262 free_irq(pci_irq_vector(vp_dev->pci_dev, i), vp_dev);
263
264 if (vp_dev->msix_affinity_masks) {
265 for (i = 0; i < vp_dev->msix_vectors; i++)
266 free_cpumask_var(vp_dev->msix_affinity_masks[i]);
267 }
268
269 if (vp_dev->msix_enabled) {
270 /* Disable the vector used for configuration */
271 vp_dev->config_vector(vp_dev, VIRTIO_MSI_NO_VECTOR);
272
273 pci_free_irq_vectors(vp_dev->pci_dev);
274 vp_dev->msix_enabled = 0;
275 }
276
277 vp_dev->msix_vectors = 0;
278 vp_dev->msix_used_vectors = 0;
279 kfree(vp_dev->msix_names);
280 vp_dev->msix_names = NULL;
281 kfree(vp_dev->msix_affinity_masks);
282 vp_dev->msix_affinity_masks = NULL;
283 kfree(vp_dev->vqs);
284 vp_dev->vqs = NULL;
285}
286
287static int vp_find_vqs_msix(struct virtio_device *vdev, unsigned int nvqs,
288 struct virtqueue *vqs[], vq_callback_t *callbacks[],
289 const char * const names[], bool per_vq_vectors,
290 const bool *ctx,
291 struct irq_affinity *desc)
292{
293 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
294 u16 msix_vec;
295 int i, err, nvectors, allocated_vectors, queue_idx = 0;
296
297 vp_dev->vqs = kcalloc(nvqs, sizeof(*vp_dev->vqs), GFP_KERNEL);
298 if (!vp_dev->vqs)
299 return -ENOMEM;
300
301 if (per_vq_vectors) {
302 /* Best option: one for change interrupt, one per vq. */
303 nvectors = 1;
304 for (i = 0; i < nvqs; ++i)
305 if (names[i] && callbacks[i])
306 ++nvectors;
307 } else {
308 /* Second best: one for change, shared for all vqs. */
309 nvectors = 2;
310 }
311
312 err = vp_request_msix_vectors(vdev, nvectors, per_vq_vectors,
313 per_vq_vectors ? desc : NULL);
314 if (err)
315 goto error_find;
316
317 vp_dev->per_vq_vectors = per_vq_vectors;
318 allocated_vectors = vp_dev->msix_used_vectors;
319 for (i = 0; i < nvqs; ++i) {
320 if (!names[i]) {
321 vqs[i] = NULL;
322 continue;
323 }
324
325 if (!callbacks[i])
326 msix_vec = VIRTIO_MSI_NO_VECTOR;
327 else if (vp_dev->per_vq_vectors)
328 msix_vec = allocated_vectors++;
329 else
330 msix_vec = VP_MSIX_VQ_VECTOR;
331 vqs[i] = vp_setup_vq(vdev, queue_idx++, callbacks[i], names[i],
332 ctx ? ctx[i] : false,
333 msix_vec);
334 if (IS_ERR(vqs[i])) {
335 err = PTR_ERR(vqs[i]);
336 goto error_find;
337 }
338
339 if (!vp_dev->per_vq_vectors || msix_vec == VIRTIO_MSI_NO_VECTOR)
340 continue;
341
342 /* allocate per-vq irq if available and necessary */
343 snprintf(vp_dev->msix_names[msix_vec],
344 sizeof *vp_dev->msix_names,
345 "%s-%s",
346 dev_name(&vp_dev->vdev.dev), names[i]);
347 err = request_irq(pci_irq_vector(vp_dev->pci_dev, msix_vec),
348 vring_interrupt, 0,
349 vp_dev->msix_names[msix_vec],
350 vqs[i]);
351 if (err) {
352 vp_del_vq(vqs[i]);
353 goto error_find;
354 }
355 }
356 return 0;
357
358error_find:
359 vp_del_vqs(vdev);
360 return err;
361}
362
363static int vp_find_vqs_intx(struct virtio_device *vdev, unsigned int nvqs,
364 struct virtqueue *vqs[], vq_callback_t *callbacks[],
365 const char * const names[], const bool *ctx)
366{
367 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
368 int i, err, queue_idx = 0;
369
370 vp_dev->vqs = kcalloc(nvqs, sizeof(*vp_dev->vqs), GFP_KERNEL);
371 if (!vp_dev->vqs)
372 return -ENOMEM;
373
374 err = request_irq(vp_dev->pci_dev->irq, vp_interrupt, IRQF_SHARED,
375 dev_name(&vdev->dev), vp_dev);
376 if (err)
377 goto out_del_vqs;
378
379 vp_dev->intx_enabled = 1;
380 vp_dev->per_vq_vectors = false;
381 for (i = 0; i < nvqs; ++i) {
382 if (!names[i]) {
383 vqs[i] = NULL;
384 continue;
385 }
386 vqs[i] = vp_setup_vq(vdev, queue_idx++, callbacks[i], names[i],
387 ctx ? ctx[i] : false,
388 VIRTIO_MSI_NO_VECTOR);
389 if (IS_ERR(vqs[i])) {
390 err = PTR_ERR(vqs[i]);
391 goto out_del_vqs;
392 }
393 }
394
395 return 0;
396out_del_vqs:
397 vp_del_vqs(vdev);
398 return err;
399}
400
401/* the config->find_vqs() implementation */
402int vp_find_vqs(struct virtio_device *vdev, unsigned int nvqs,
403 struct virtqueue *vqs[], vq_callback_t *callbacks[],
404 const char * const names[], const bool *ctx,
405 struct irq_affinity *desc)
406{
407 int err;
408
409 /* Try MSI-X with one vector per queue. */
410 err = vp_find_vqs_msix(vdev, nvqs, vqs, callbacks, names, true, ctx, desc);
411 if (!err)
412 return 0;
413 /* Fallback: MSI-X with one vector for config, one shared for queues. */
414 err = vp_find_vqs_msix(vdev, nvqs, vqs, callbacks, names, false, ctx, desc);
415 if (!err)
416 return 0;
417 /* Is there an interrupt? If not give up. */
418 if (!(to_vp_device(vdev)->pci_dev->irq))
419 return err;
420 /* Finally fall back to regular interrupts. */
421 return vp_find_vqs_intx(vdev, nvqs, vqs, callbacks, names, ctx);
422}
423
424const char *vp_bus_name(struct virtio_device *vdev)
425{
426 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
427
428 return pci_name(vp_dev->pci_dev);
429}
430
431/* Setup the affinity for a virtqueue:
432 * - force the affinity for per vq vector
433 * - OR over all affinities for shared MSI
434 * - ignore the affinity request if we're using INTX
435 */
436int vp_set_vq_affinity(struct virtqueue *vq, const struct cpumask *cpu_mask)
437{
438 struct virtio_device *vdev = vq->vdev;
439 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
440 struct virtio_pci_vq_info *info = vp_dev->vqs[vq->index];
441 struct cpumask *mask;
442 unsigned int irq;
443
444 if (!vq->callback)
445 return -EINVAL;
446
447 if (vp_dev->msix_enabled) {
448 mask = vp_dev->msix_affinity_masks[info->msix_vector];
449 irq = pci_irq_vector(vp_dev->pci_dev, info->msix_vector);
450 if (!cpu_mask)
451 irq_update_affinity_hint(irq, NULL);
452 else {
453 cpumask_copy(mask, cpu_mask);
454 irq_set_affinity_and_hint(irq, mask);
455 }
456 }
457 return 0;
458}
459
460const struct cpumask *vp_get_vq_affinity(struct virtio_device *vdev, int index)
461{
462 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
463
464 if (!vp_dev->per_vq_vectors ||
465 vp_dev->vqs[index]->msix_vector == VIRTIO_MSI_NO_VECTOR)
466 return NULL;
467
468 return pci_irq_get_affinity(vp_dev->pci_dev,
469 vp_dev->vqs[index]->msix_vector);
470}
471
472#ifdef CONFIG_PM_SLEEP
473static int virtio_pci_freeze(struct device *dev)
474{
475 struct pci_dev *pci_dev = to_pci_dev(dev);
476 struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
477 int ret;
478
479 ret = virtio_device_freeze(&vp_dev->vdev);
480
481 if (!ret)
482 pci_disable_device(pci_dev);
483 return ret;
484}
485
486static int virtio_pci_restore(struct device *dev)
487{
488 struct pci_dev *pci_dev = to_pci_dev(dev);
489 struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
490 int ret;
491
492 ret = pci_enable_device(pci_dev);
493 if (ret)
494 return ret;
495
496 pci_set_master(pci_dev);
497 return virtio_device_restore(&vp_dev->vdev);
498}
499
500static bool vp_supports_pm_no_reset(struct device *dev)
501{
502 struct pci_dev *pci_dev = to_pci_dev(dev);
503 u16 pmcsr;
504
505 if (!pci_dev->pm_cap)
506 return false;
507
508 pci_read_config_word(pci_dev, pci_dev->pm_cap + PCI_PM_CTRL, &pmcsr);
509 if (PCI_POSSIBLE_ERROR(pmcsr)) {
510 dev_err(dev, "Unable to query pmcsr");
511 return false;
512 }
513
514 return pmcsr & PCI_PM_CTRL_NO_SOFT_RESET;
515}
516
517static int virtio_pci_suspend(struct device *dev)
518{
519 return vp_supports_pm_no_reset(dev) ? 0 : virtio_pci_freeze(dev);
520}
521
522static int virtio_pci_resume(struct device *dev)
523{
524 return vp_supports_pm_no_reset(dev) ? 0 : virtio_pci_restore(dev);
525}
526
527static const struct dev_pm_ops virtio_pci_pm_ops = {
528 .suspend = virtio_pci_suspend,
529 .resume = virtio_pci_resume,
530 .freeze = virtio_pci_freeze,
531 .thaw = virtio_pci_restore,
532 .poweroff = virtio_pci_freeze,
533 .restore = virtio_pci_restore,
534};
535#endif
536
537
538/* Qumranet donated their vendor ID for devices 0x1000 thru 0x10FF. */
539static const struct pci_device_id virtio_pci_id_table[] = {
540 { PCI_DEVICE(PCI_VENDOR_ID_REDHAT_QUMRANET, PCI_ANY_ID) },
541 { 0 }
542};
543
544MODULE_DEVICE_TABLE(pci, virtio_pci_id_table);
545
546static void virtio_pci_release_dev(struct device *_d)
547{
548 struct virtio_device *vdev = dev_to_virtio(_d);
549 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
550
551 /* As struct device is a kobject, it's not safe to
552 * free the memory (including the reference counter itself)
553 * until it's release callback. */
554 kfree(vp_dev);
555}
556
557static int virtio_pci_probe(struct pci_dev *pci_dev,
558 const struct pci_device_id *id)
559{
560 struct virtio_pci_device *vp_dev, *reg_dev = NULL;
561 int rc;
562
563 /* allocate our structure and fill it out */
564 vp_dev = kzalloc(sizeof(struct virtio_pci_device), GFP_KERNEL);
565 if (!vp_dev)
566 return -ENOMEM;
567
568 pci_set_drvdata(pci_dev, vp_dev);
569 vp_dev->vdev.dev.parent = &pci_dev->dev;
570 vp_dev->vdev.dev.release = virtio_pci_release_dev;
571 vp_dev->pci_dev = pci_dev;
572 INIT_LIST_HEAD(&vp_dev->virtqueues);
573 spin_lock_init(&vp_dev->lock);
574
575 /* enable the device */
576 rc = pci_enable_device(pci_dev);
577 if (rc)
578 goto err_enable_device;
579
580 if (force_legacy) {
581 rc = virtio_pci_legacy_probe(vp_dev);
582 /* Also try modern mode if we can't map BAR0 (no IO space). */
583 if (rc == -ENODEV || rc == -ENOMEM)
584 rc = virtio_pci_modern_probe(vp_dev);
585 if (rc)
586 goto err_probe;
587 } else {
588 rc = virtio_pci_modern_probe(vp_dev);
589 if (rc == -ENODEV)
590 rc = virtio_pci_legacy_probe(vp_dev);
591 if (rc)
592 goto err_probe;
593 }
594
595 pci_set_master(pci_dev);
596
597 rc = register_virtio_device(&vp_dev->vdev);
598 reg_dev = vp_dev;
599 if (rc)
600 goto err_register;
601
602 return 0;
603
604err_register:
605 if (vp_dev->is_legacy)
606 virtio_pci_legacy_remove(vp_dev);
607 else
608 virtio_pci_modern_remove(vp_dev);
609err_probe:
610 pci_disable_device(pci_dev);
611err_enable_device:
612 if (reg_dev)
613 put_device(&vp_dev->vdev.dev);
614 else
615 kfree(vp_dev);
616 return rc;
617}
618
619static void virtio_pci_remove(struct pci_dev *pci_dev)
620{
621 struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
622 struct device *dev = get_device(&vp_dev->vdev.dev);
623
624 /*
625 * Device is marked broken on surprise removal so that virtio upper
626 * layers can abort any ongoing operation.
627 */
628 if (!pci_device_is_present(pci_dev))
629 virtio_break_device(&vp_dev->vdev);
630
631 pci_disable_sriov(pci_dev);
632
633 unregister_virtio_device(&vp_dev->vdev);
634
635 if (vp_dev->is_legacy)
636 virtio_pci_legacy_remove(vp_dev);
637 else
638 virtio_pci_modern_remove(vp_dev);
639
640 pci_disable_device(pci_dev);
641 put_device(dev);
642}
643
644static int virtio_pci_sriov_configure(struct pci_dev *pci_dev, int num_vfs)
645{
646 struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
647 struct virtio_device *vdev = &vp_dev->vdev;
648 int ret;
649
650 if (!(vdev->config->get_status(vdev) & VIRTIO_CONFIG_S_DRIVER_OK))
651 return -EBUSY;
652
653 if (!__virtio_test_bit(vdev, VIRTIO_F_SR_IOV))
654 return -EINVAL;
655
656 if (pci_vfs_assigned(pci_dev))
657 return -EPERM;
658
659 if (num_vfs == 0) {
660 pci_disable_sriov(pci_dev);
661 return 0;
662 }
663
664 ret = pci_enable_sriov(pci_dev, num_vfs);
665 if (ret < 0)
666 return ret;
667
668 return num_vfs;
669}
670
671static struct pci_driver virtio_pci_driver = {
672 .name = "virtio-pci",
673 .id_table = virtio_pci_id_table,
674 .probe = virtio_pci_probe,
675 .remove = virtio_pci_remove,
676#ifdef CONFIG_PM_SLEEP
677 .driver.pm = &virtio_pci_pm_ops,
678#endif
679 .sriov_configure = virtio_pci_sriov_configure,
680};
681
682struct virtio_device *virtio_pci_vf_get_pf_dev(struct pci_dev *pdev)
683{
684 struct virtio_pci_device *pf_vp_dev;
685
686 pf_vp_dev = pci_iov_get_pf_drvdata(pdev, &virtio_pci_driver);
687 if (IS_ERR(pf_vp_dev))
688 return NULL;
689
690 return &pf_vp_dev->vdev;
691}
692
693module_pci_driver(virtio_pci_driver);
694
695MODULE_AUTHOR("Anthony Liguori <aliguori@us.ibm.com>");
696MODULE_DESCRIPTION("virtio-pci");
697MODULE_LICENSE("GPL");
698MODULE_VERSION("1");