1 /*
2 * Virtio PCI driver - common functionality for all device versions
3 *
4 * This module allows virtio devices to be used over a virtual PCI device.
5 * This can be used with QEMU based VMMs like KVM or Xen.
6 *
7 * Copyright IBM Corp. 2007
8 * Copyright Red Hat, Inc. 2014
9 *
10 * Authors:
11 * Anthony Liguori <aliguori@us.ibm.com>
12 * Rusty Russell <rusty@rustcorp.com.au>
13 * Michael S. Tsirkin <mst@redhat.com>
14 *
15 * This work is licensed under the terms of the GNU GPL, version 2 or later.
16 * See the COPYING file in the top-level directory.
17 *
18 */
19
20 #include "virtio_pci_common.h"
21
22 static bool force_legacy = false;
23
24 #if IS_ENABLED(CONFIG_VIRTIO_PCI_LEGACY)
25 module_param(force_legacy, bool, 0444);
26 MODULE_PARM_DESC(force_legacy,
27 "Force legacy mode for transitional virtio 1 devices");
28 #endif
29
30 /* wait for pending irq handlers */
vp_synchronize_vectors(struct virtio_device * vdev)31 void vp_synchronize_vectors(struct virtio_device *vdev)
32 {
33 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
34 int i;
35
36 if (vp_dev->intx_enabled)
37 synchronize_irq(vp_dev->pci_dev->irq);
38
39 for (i = 0; i < vp_dev->msix_vectors; ++i)
40 synchronize_irq(pci_irq_vector(vp_dev->pci_dev, i));
41 }
42
43 /* the notify function used when creating a virt queue */
vp_notify(struct virtqueue * vq)44 bool vp_notify(struct virtqueue *vq)
45 {
46 /* we write the queue's selector into the notification register to
47 * signal the other end */
48 iowrite16(vq->index, (void __iomem *)vq->priv);
49 return true;
50 }
51
52 /* Handle a configuration change: Tell driver if it wants to know. */
vp_config_changed(int irq,void * opaque)53 static irqreturn_t vp_config_changed(int irq, void *opaque)
54 {
55 struct virtio_pci_device *vp_dev = opaque;
56
57 virtio_config_changed(&vp_dev->vdev);
58 return IRQ_HANDLED;
59 }
60
61 /* Notify all virtqueues on an interrupt. */
vp_vring_interrupt(int irq,void * opaque)62 static irqreturn_t vp_vring_interrupt(int irq, void *opaque)
63 {
64 struct virtio_pci_device *vp_dev = opaque;
65 struct virtio_pci_vq_info *info;
66 irqreturn_t ret = IRQ_NONE;
67 unsigned long flags;
68
69 spin_lock_irqsave(&vp_dev->lock, flags);
70 list_for_each_entry(info, &vp_dev->virtqueues, node) {
71 if (vring_interrupt(irq, info->vq) == IRQ_HANDLED)
72 ret = IRQ_HANDLED;
73 }
74 spin_unlock_irqrestore(&vp_dev->lock, flags);
75
76 return ret;
77 }
78
79 /* A small wrapper to also acknowledge the interrupt when it's handled.
80 * I really need an EIO hook for the vring so I can ack the interrupt once we
81 * know that we'll be handling the IRQ but before we invoke the callback since
82 * the callback may notify the host which results in the host attempting to
83 * raise an interrupt that we would then mask once we acknowledged the
84 * interrupt. */
vp_interrupt(int irq,void * opaque)85 static irqreturn_t vp_interrupt(int irq, void *opaque)
86 {
87 struct virtio_pci_device *vp_dev = opaque;
88 u8 isr;
89
90 /* reading the ISR has the effect of also clearing it so it's very
91 * important to save off the value. */
92 isr = ioread8(vp_dev->isr);
93
94 /* It's definitely not us if the ISR was not high */
95 if (!isr)
96 return IRQ_NONE;
97
98 /* Configuration change? Tell driver if it wants to know. */
99 if (isr & VIRTIO_PCI_ISR_CONFIG)
100 vp_config_changed(irq, opaque);
101
102 return vp_vring_interrupt(irq, opaque);
103 }
104
vp_request_msix_vectors(struct virtio_device * vdev,int nvectors,bool per_vq_vectors,struct irq_affinity * desc)105 static int vp_request_msix_vectors(struct virtio_device *vdev, int nvectors,
106 bool per_vq_vectors, struct irq_affinity *desc)
107 {
108 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
109 const char *name = dev_name(&vp_dev->vdev.dev);
110 unsigned flags = PCI_IRQ_MSIX;
111 unsigned i, v;
112 int err = -ENOMEM;
113
114 vp_dev->msix_vectors = nvectors;
115
116 vp_dev->msix_names = kmalloc(nvectors * sizeof *vp_dev->msix_names,
117 GFP_KERNEL);
118 if (!vp_dev->msix_names)
119 goto error;
120 vp_dev->msix_affinity_masks
121 = kzalloc(nvectors * sizeof *vp_dev->msix_affinity_masks,
122 GFP_KERNEL);
123 if (!vp_dev->msix_affinity_masks)
124 goto error;
125 for (i = 0; i < nvectors; ++i)
126 if (!alloc_cpumask_var(&vp_dev->msix_affinity_masks[i],
127 GFP_KERNEL))
128 goto error;
129
130 if (desc) {
131 flags |= PCI_IRQ_AFFINITY;
132 desc->pre_vectors++; /* virtio config vector */
133 }
134
135 err = pci_alloc_irq_vectors_affinity(vp_dev->pci_dev, nvectors,
136 nvectors, flags, desc);
137 if (err < 0)
138 goto error;
139 vp_dev->msix_enabled = 1;
140
141 /* Set the vector used for configuration */
142 v = vp_dev->msix_used_vectors;
143 snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names,
144 "%s-config", name);
145 err = request_irq(pci_irq_vector(vp_dev->pci_dev, v),
146 vp_config_changed, 0, vp_dev->msix_names[v],
147 vp_dev);
148 if (err)
149 goto error;
150 ++vp_dev->msix_used_vectors;
151
152 v = vp_dev->config_vector(vp_dev, v);
153 /* Verify we had enough resources to assign the vector */
154 if (v == VIRTIO_MSI_NO_VECTOR) {
155 err = -EBUSY;
156 goto error;
157 }
158
159 if (!per_vq_vectors) {
160 /* Shared vector for all VQs */
161 v = vp_dev->msix_used_vectors;
162 snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names,
163 "%s-virtqueues", name);
164 err = request_irq(pci_irq_vector(vp_dev->pci_dev, v),
165 vp_vring_interrupt, 0, vp_dev->msix_names[v],
166 vp_dev);
167 if (err)
168 goto error;
169 ++vp_dev->msix_used_vectors;
170 }
171 return 0;
172 error:
173 return err;
174 }
175
vp_setup_vq(struct virtio_device * vdev,unsigned index,void (* callback)(struct virtqueue * vq),const char * name,bool ctx,u16 msix_vec)176 static struct virtqueue *vp_setup_vq(struct virtio_device *vdev, unsigned index,
177 void (*callback)(struct virtqueue *vq),
178 const char *name,
179 bool ctx,
180 u16 msix_vec)
181 {
182 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
183 struct virtio_pci_vq_info *info = kmalloc(sizeof *info, GFP_KERNEL);
184 struct virtqueue *vq;
185 unsigned long flags;
186
187 /* fill out our structure that represents an active queue */
188 if (!info)
189 return ERR_PTR(-ENOMEM);
190
191 vq = vp_dev->setup_vq(vp_dev, info, index, callback, name, ctx,
192 msix_vec);
193 if (IS_ERR(vq))
194 goto out_info;
195
196 info->vq = vq;
197 if (callback) {
198 spin_lock_irqsave(&vp_dev->lock, flags);
199 list_add(&info->node, &vp_dev->virtqueues);
200 spin_unlock_irqrestore(&vp_dev->lock, flags);
201 } else {
202 INIT_LIST_HEAD(&info->node);
203 }
204
205 vp_dev->vqs[index] = info;
206 return vq;
207
208 out_info:
209 kfree(info);
210 return vq;
211 }
212
vp_del_vq(struct virtqueue * vq)213 static void vp_del_vq(struct virtqueue *vq)
214 {
215 struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
216 struct virtio_pci_vq_info *info = vp_dev->vqs[vq->index];
217 unsigned long flags;
218
219 spin_lock_irqsave(&vp_dev->lock, flags);
220 list_del(&info->node);
221 spin_unlock_irqrestore(&vp_dev->lock, flags);
222
223 vp_dev->del_vq(info);
224 kfree(info);
225 }
226
227 /* the config->del_vqs() implementation */
vp_del_vqs(struct virtio_device * vdev)228 void vp_del_vqs(struct virtio_device *vdev)
229 {
230 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
231 struct virtqueue *vq, *n;
232 int i;
233
234 list_for_each_entry_safe(vq, n, &vdev->vqs, list) {
235 if (vp_dev->per_vq_vectors) {
236 int v = vp_dev->vqs[vq->index]->msix_vector;
237
238 if (v != VIRTIO_MSI_NO_VECTOR) {
239 int irq = pci_irq_vector(vp_dev->pci_dev, v);
240
241 irq_set_affinity_hint(irq, NULL);
242 free_irq(irq, vq);
243 }
244 }
245 vp_del_vq(vq);
246 }
247 vp_dev->per_vq_vectors = false;
248
249 if (vp_dev->intx_enabled) {
250 free_irq(vp_dev->pci_dev->irq, vp_dev);
251 vp_dev->intx_enabled = 0;
252 }
253
254 for (i = 0; i < vp_dev->msix_used_vectors; ++i)
255 free_irq(pci_irq_vector(vp_dev->pci_dev, i), vp_dev);
256
257 if (vp_dev->msix_affinity_masks) {
258 for (i = 0; i < vp_dev->msix_vectors; i++)
259 if (vp_dev->msix_affinity_masks[i])
260 free_cpumask_var(vp_dev->msix_affinity_masks[i]);
261 }
262
263 if (vp_dev->msix_enabled) {
264 /* Disable the vector used for configuration */
265 vp_dev->config_vector(vp_dev, VIRTIO_MSI_NO_VECTOR);
266
267 pci_free_irq_vectors(vp_dev->pci_dev);
268 vp_dev->msix_enabled = 0;
269 }
270
271 vp_dev->msix_vectors = 0;
272 vp_dev->msix_used_vectors = 0;
273 kfree(vp_dev->msix_names);
274 vp_dev->msix_names = NULL;
275 kfree(vp_dev->msix_affinity_masks);
276 vp_dev->msix_affinity_masks = NULL;
277 kfree(vp_dev->vqs);
278 vp_dev->vqs = NULL;
279 }
280
vp_find_vqs_msix(struct virtio_device * vdev,unsigned nvqs,struct virtqueue * vqs[],vq_callback_t * callbacks[],const char * const names[],bool per_vq_vectors,const bool * ctx,struct irq_affinity * desc)281 static int vp_find_vqs_msix(struct virtio_device *vdev, unsigned nvqs,
282 struct virtqueue *vqs[], vq_callback_t *callbacks[],
283 const char * const names[], bool per_vq_vectors,
284 const bool *ctx,
285 struct irq_affinity *desc)
286 {
287 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
288 u16 msix_vec;
289 int i, err, nvectors, allocated_vectors;
290
291 vp_dev->vqs = kcalloc(nvqs, sizeof(*vp_dev->vqs), GFP_KERNEL);
292 if (!vp_dev->vqs)
293 return -ENOMEM;
294
295 if (per_vq_vectors) {
296 /* Best option: one for change interrupt, one per vq. */
297 nvectors = 1;
298 for (i = 0; i < nvqs; ++i)
299 if (callbacks[i])
300 ++nvectors;
301 } else {
302 /* Second best: one for change, shared for all vqs. */
303 nvectors = 2;
304 }
305
306 err = vp_request_msix_vectors(vdev, nvectors, per_vq_vectors,
307 per_vq_vectors ? desc : NULL);
308 if (err)
309 goto error_find;
310
311 vp_dev->per_vq_vectors = per_vq_vectors;
312 allocated_vectors = vp_dev->msix_used_vectors;
313 for (i = 0; i < nvqs; ++i) {
314 if (!names[i]) {
315 vqs[i] = NULL;
316 continue;
317 }
318
319 if (!callbacks[i])
320 msix_vec = VIRTIO_MSI_NO_VECTOR;
321 else if (vp_dev->per_vq_vectors)
322 msix_vec = allocated_vectors++;
323 else
324 msix_vec = VP_MSIX_VQ_VECTOR;
325 vqs[i] = vp_setup_vq(vdev, i, callbacks[i], names[i],
326 ctx ? ctx[i] : false,
327 msix_vec);
328 if (IS_ERR(vqs[i])) {
329 err = PTR_ERR(vqs[i]);
330 goto error_find;
331 }
332
333 if (!vp_dev->per_vq_vectors || msix_vec == VIRTIO_MSI_NO_VECTOR)
334 continue;
335
336 /* allocate per-vq irq if available and necessary */
337 snprintf(vp_dev->msix_names[msix_vec],
338 sizeof *vp_dev->msix_names,
339 "%s-%s",
340 dev_name(&vp_dev->vdev.dev), names[i]);
341 err = request_irq(pci_irq_vector(vp_dev->pci_dev, msix_vec),
342 vring_interrupt, 0,
343 vp_dev->msix_names[msix_vec],
344 vqs[i]);
345 if (err)
346 goto error_find;
347 }
348 return 0;
349
350 error_find:
351 vp_del_vqs(vdev);
352 return err;
353 }
354
vp_find_vqs_intx(struct virtio_device * vdev,unsigned nvqs,struct virtqueue * vqs[],vq_callback_t * callbacks[],const char * const names[],const bool * ctx)355 static int vp_find_vqs_intx(struct virtio_device *vdev, unsigned nvqs,
356 struct virtqueue *vqs[], vq_callback_t *callbacks[],
357 const char * const names[], const bool *ctx)
358 {
359 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
360 int i, err;
361
362 vp_dev->vqs = kcalloc(nvqs, sizeof(*vp_dev->vqs), GFP_KERNEL);
363 if (!vp_dev->vqs)
364 return -ENOMEM;
365
366 err = request_irq(vp_dev->pci_dev->irq, vp_interrupt, IRQF_SHARED,
367 dev_name(&vdev->dev), vp_dev);
368 if (err)
369 goto out_del_vqs;
370
371 vp_dev->intx_enabled = 1;
372 vp_dev->per_vq_vectors = false;
373 for (i = 0; i < nvqs; ++i) {
374 if (!names[i]) {
375 vqs[i] = NULL;
376 continue;
377 }
378 vqs[i] = vp_setup_vq(vdev, i, callbacks[i], names[i],
379 ctx ? ctx[i] : false,
380 VIRTIO_MSI_NO_VECTOR);
381 if (IS_ERR(vqs[i])) {
382 err = PTR_ERR(vqs[i]);
383 goto out_del_vqs;
384 }
385 }
386
387 return 0;
388 out_del_vqs:
389 vp_del_vqs(vdev);
390 return err;
391 }
392
393 /* the config->find_vqs() implementation */
vp_find_vqs(struct virtio_device * vdev,unsigned nvqs,struct virtqueue * vqs[],vq_callback_t * callbacks[],const char * const names[],const bool * ctx,struct irq_affinity * desc)394 int vp_find_vqs(struct virtio_device *vdev, unsigned nvqs,
395 struct virtqueue *vqs[], vq_callback_t *callbacks[],
396 const char * const names[], const bool *ctx,
397 struct irq_affinity *desc)
398 {
399 int err;
400
401 /* Try MSI-X with one vector per queue. */
402 err = vp_find_vqs_msix(vdev, nvqs, vqs, callbacks, names, true, ctx, desc);
403 if (!err)
404 return 0;
405 /* Fallback: MSI-X with one vector for config, one shared for queues. */
406 err = vp_find_vqs_msix(vdev, nvqs, vqs, callbacks, names, false, ctx, desc);
407 if (!err)
408 return 0;
409 /* Finally fall back to regular interrupts. */
410 return vp_find_vqs_intx(vdev, nvqs, vqs, callbacks, names, ctx);
411 }
412
vp_bus_name(struct virtio_device * vdev)413 const char *vp_bus_name(struct virtio_device *vdev)
414 {
415 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
416
417 return pci_name(vp_dev->pci_dev);
418 }
419
420 /* Setup the affinity for a virtqueue:
421 * - force the affinity for per vq vector
422 * - OR over all affinities for shared MSI
423 * - ignore the affinity request if we're using INTX
424 */
vp_set_vq_affinity(struct virtqueue * vq,int cpu)425 int vp_set_vq_affinity(struct virtqueue *vq, int cpu)
426 {
427 struct virtio_device *vdev = vq->vdev;
428 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
429 struct virtio_pci_vq_info *info = vp_dev->vqs[vq->index];
430 struct cpumask *mask;
431 unsigned int irq;
432
433 if (!vq->callback)
434 return -EINVAL;
435
436 if (vp_dev->msix_enabled) {
437 mask = vp_dev->msix_affinity_masks[info->msix_vector];
438 irq = pci_irq_vector(vp_dev->pci_dev, info->msix_vector);
439 if (cpu == -1)
440 irq_set_affinity_hint(irq, NULL);
441 else {
442 cpumask_clear(mask);
443 cpumask_set_cpu(cpu, mask);
444 irq_set_affinity_hint(irq, mask);
445 }
446 }
447 return 0;
448 }
449
vp_get_vq_affinity(struct virtio_device * vdev,int index)450 const struct cpumask *vp_get_vq_affinity(struct virtio_device *vdev, int index)
451 {
452 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
453
454 if (!vp_dev->per_vq_vectors ||
455 vp_dev->vqs[index]->msix_vector == VIRTIO_MSI_NO_VECTOR)
456 return NULL;
457
458 return pci_irq_get_affinity(vp_dev->pci_dev,
459 vp_dev->vqs[index]->msix_vector);
460 }
461
462 #ifdef CONFIG_PM_SLEEP
virtio_pci_freeze(struct device * dev)463 static int virtio_pci_freeze(struct device *dev)
464 {
465 struct pci_dev *pci_dev = to_pci_dev(dev);
466 struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
467 int ret;
468
469 ret = virtio_device_freeze(&vp_dev->vdev);
470
471 if (!ret)
472 pci_disable_device(pci_dev);
473 return ret;
474 }
475
virtio_pci_restore(struct device * dev)476 static int virtio_pci_restore(struct device *dev)
477 {
478 struct pci_dev *pci_dev = to_pci_dev(dev);
479 struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
480 int ret;
481
482 ret = pci_enable_device(pci_dev);
483 if (ret)
484 return ret;
485
486 pci_set_master(pci_dev);
487 return virtio_device_restore(&vp_dev->vdev);
488 }
489
490 static const struct dev_pm_ops virtio_pci_pm_ops = {
491 SET_SYSTEM_SLEEP_PM_OPS(virtio_pci_freeze, virtio_pci_restore)
492 };
493 #endif
494
495
496 /* Qumranet donated their vendor ID for devices 0x1000 thru 0x10FF. */
497 static const struct pci_device_id virtio_pci_id_table[] = {
498 { PCI_DEVICE(PCI_VENDOR_ID_REDHAT_QUMRANET, PCI_ANY_ID) },
499 { 0 }
500 };
501
502 MODULE_DEVICE_TABLE(pci, virtio_pci_id_table);
503
virtio_pci_release_dev(struct device * _d)504 static void virtio_pci_release_dev(struct device *_d)
505 {
506 struct virtio_device *vdev = dev_to_virtio(_d);
507 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
508
509 /* As struct device is a kobject, it's not safe to
510 * free the memory (including the reference counter itself)
511 * until it's release callback. */
512 kfree(vp_dev);
513 }
514
virtio_pci_probe(struct pci_dev * pci_dev,const struct pci_device_id * id)515 static int virtio_pci_probe(struct pci_dev *pci_dev,
516 const struct pci_device_id *id)
517 {
518 struct virtio_pci_device *vp_dev;
519 int rc;
520
521 /* allocate our structure and fill it out */
522 vp_dev = kzalloc(sizeof(struct virtio_pci_device), GFP_KERNEL);
523 if (!vp_dev)
524 return -ENOMEM;
525
526 pci_set_drvdata(pci_dev, vp_dev);
527 vp_dev->vdev.dev.parent = &pci_dev->dev;
528 vp_dev->vdev.dev.release = virtio_pci_release_dev;
529 vp_dev->pci_dev = pci_dev;
530 INIT_LIST_HEAD(&vp_dev->virtqueues);
531 spin_lock_init(&vp_dev->lock);
532
533 /* enable the device */
534 rc = pci_enable_device(pci_dev);
535 if (rc)
536 goto err_enable_device;
537
538 if (force_legacy) {
539 rc = virtio_pci_legacy_probe(vp_dev);
540 /* Also try modern mode if we can't map BAR0 (no IO space). */
541 if (rc == -ENODEV || rc == -ENOMEM)
542 rc = virtio_pci_modern_probe(vp_dev);
543 if (rc)
544 goto err_probe;
545 } else {
546 rc = virtio_pci_modern_probe(vp_dev);
547 if (rc == -ENODEV)
548 rc = virtio_pci_legacy_probe(vp_dev);
549 if (rc)
550 goto err_probe;
551 }
552
553 pci_set_master(pci_dev);
554
555 rc = register_virtio_device(&vp_dev->vdev);
556 if (rc)
557 goto err_register;
558
559 return 0;
560
561 err_register:
562 if (vp_dev->ioaddr)
563 virtio_pci_legacy_remove(vp_dev);
564 else
565 virtio_pci_modern_remove(vp_dev);
566 err_probe:
567 pci_disable_device(pci_dev);
568 err_enable_device:
569 kfree(vp_dev);
570 return rc;
571 }
572
virtio_pci_remove(struct pci_dev * pci_dev)573 static void virtio_pci_remove(struct pci_dev *pci_dev)
574 {
575 struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
576 struct device *dev = get_device(&vp_dev->vdev.dev);
577
578 unregister_virtio_device(&vp_dev->vdev);
579
580 if (vp_dev->ioaddr)
581 virtio_pci_legacy_remove(vp_dev);
582 else
583 virtio_pci_modern_remove(vp_dev);
584
585 pci_disable_device(pci_dev);
586 put_device(dev);
587 }
588
589 static struct pci_driver virtio_pci_driver = {
590 .name = "virtio-pci",
591 .id_table = virtio_pci_id_table,
592 .probe = virtio_pci_probe,
593 .remove = virtio_pci_remove,
594 #ifdef CONFIG_PM_SLEEP
595 .driver.pm = &virtio_pci_pm_ops,
596 #endif
597 };
598
599 module_pci_driver(virtio_pci_driver);
600
601 MODULE_AUTHOR("Anthony Liguori <aliguori@us.ibm.com>");
602 MODULE_DESCRIPTION("virtio-pci");
603 MODULE_LICENSE("GPL");
604 MODULE_VERSION("1");
605