1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (C) 2018-2020 Intel Corporation.
4 * Copyright (C) 2020 Red Hat, Inc.
5 *
6 * Author: Tiwei Bie <tiwei.bie@intel.com>
7 * Jason Wang <jasowang@redhat.com>
8 *
9 * Thanks Michael S. Tsirkin for the valuable comments and
10 * suggestions. And thanks to Cunming Liang and Zhihong Wang for all
11 * their supports.
12 */
13
14 #include <linux/kernel.h>
15 #include <linux/module.h>
16 #include <linux/cdev.h>
17 #include <linux/device.h>
18 #include <linux/mm.h>
19 #include <linux/slab.h>
20 #include <linux/iommu.h>
21 #include <linux/uuid.h>
22 #include <linux/vdpa.h>
23 #include <linux/nospec.h>
24 #include <linux/vhost.h>
25
26 #include "vhost.h"
27
28 enum {
29 VHOST_VDPA_BACKEND_FEATURES =
30 (1ULL << VHOST_BACKEND_F_IOTLB_MSG_V2) |
31 (1ULL << VHOST_BACKEND_F_IOTLB_BATCH) |
32 (1ULL << VHOST_BACKEND_F_IOTLB_ASID),
33 };
34
35 #define VHOST_VDPA_DEV_MAX (1U << MINORBITS)
36
37 #define VHOST_VDPA_IOTLB_BUCKETS 16
38
39 struct vhost_vdpa_as {
40 struct hlist_node hash_link;
41 struct vhost_iotlb iotlb;
42 u32 id;
43 };
44
45 struct vhost_vdpa {
46 struct vhost_dev vdev;
47 struct iommu_domain *domain;
48 struct vhost_virtqueue *vqs;
49 struct completion completion;
50 struct vdpa_device *vdpa;
51 struct hlist_head as[VHOST_VDPA_IOTLB_BUCKETS];
52 struct device dev;
53 struct cdev cdev;
54 atomic_t opened;
55 u32 nvqs;
56 int virtio_id;
57 int minor;
58 struct eventfd_ctx *config_ctx;
59 int in_batch;
60 struct vdpa_iova_range range;
61 u32 batch_asid;
62 };
63
64 static DEFINE_IDA(vhost_vdpa_ida);
65
66 static dev_t vhost_vdpa_major;
67
68 static void vhost_vdpa_iotlb_unmap(struct vhost_vdpa *v,
69 struct vhost_iotlb *iotlb, u64 start,
70 u64 last, u32 asid);
71
iotlb_to_asid(struct vhost_iotlb * iotlb)72 static inline u32 iotlb_to_asid(struct vhost_iotlb *iotlb)
73 {
74 struct vhost_vdpa_as *as = container_of(iotlb, struct
75 vhost_vdpa_as, iotlb);
76 return as->id;
77 }
78
asid_to_as(struct vhost_vdpa * v,u32 asid)79 static struct vhost_vdpa_as *asid_to_as(struct vhost_vdpa *v, u32 asid)
80 {
81 struct hlist_head *head = &v->as[asid % VHOST_VDPA_IOTLB_BUCKETS];
82 struct vhost_vdpa_as *as;
83
84 hlist_for_each_entry(as, head, hash_link)
85 if (as->id == asid)
86 return as;
87
88 return NULL;
89 }
90
asid_to_iotlb(struct vhost_vdpa * v,u32 asid)91 static struct vhost_iotlb *asid_to_iotlb(struct vhost_vdpa *v, u32 asid)
92 {
93 struct vhost_vdpa_as *as = asid_to_as(v, asid);
94
95 if (!as)
96 return NULL;
97
98 return &as->iotlb;
99 }
100
vhost_vdpa_alloc_as(struct vhost_vdpa * v,u32 asid)101 static struct vhost_vdpa_as *vhost_vdpa_alloc_as(struct vhost_vdpa *v, u32 asid)
102 {
103 struct hlist_head *head = &v->as[asid % VHOST_VDPA_IOTLB_BUCKETS];
104 struct vhost_vdpa_as *as;
105
106 if (asid_to_as(v, asid))
107 return NULL;
108
109 if (asid >= v->vdpa->nas)
110 return NULL;
111
112 as = kmalloc(sizeof(*as), GFP_KERNEL);
113 if (!as)
114 return NULL;
115
116 vhost_iotlb_init(&as->iotlb, 0, 0);
117 as->id = asid;
118 hlist_add_head(&as->hash_link, head);
119
120 return as;
121 }
122
vhost_vdpa_find_alloc_as(struct vhost_vdpa * v,u32 asid)123 static struct vhost_vdpa_as *vhost_vdpa_find_alloc_as(struct vhost_vdpa *v,
124 u32 asid)
125 {
126 struct vhost_vdpa_as *as = asid_to_as(v, asid);
127
128 if (as)
129 return as;
130
131 return vhost_vdpa_alloc_as(v, asid);
132 }
133
vhost_vdpa_remove_as(struct vhost_vdpa * v,u32 asid)134 static int vhost_vdpa_remove_as(struct vhost_vdpa *v, u32 asid)
135 {
136 struct vhost_vdpa_as *as = asid_to_as(v, asid);
137
138 if (!as)
139 return -EINVAL;
140
141 hlist_del(&as->hash_link);
142 vhost_vdpa_iotlb_unmap(v, &as->iotlb, 0ULL, 0ULL - 1, asid);
143 kfree(as);
144
145 return 0;
146 }
147
handle_vq_kick(struct vhost_work * work)148 static void handle_vq_kick(struct vhost_work *work)
149 {
150 struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
151 poll.work);
152 struct vhost_vdpa *v = container_of(vq->dev, struct vhost_vdpa, vdev);
153 const struct vdpa_config_ops *ops = v->vdpa->config;
154
155 ops->kick_vq(v->vdpa, vq - v->vqs);
156 }
157
vhost_vdpa_virtqueue_cb(void * private)158 static irqreturn_t vhost_vdpa_virtqueue_cb(void *private)
159 {
160 struct vhost_virtqueue *vq = private;
161 struct eventfd_ctx *call_ctx = vq->call_ctx.ctx;
162
163 if (call_ctx)
164 eventfd_signal(call_ctx, 1);
165
166 return IRQ_HANDLED;
167 }
168
vhost_vdpa_config_cb(void * private)169 static irqreturn_t vhost_vdpa_config_cb(void *private)
170 {
171 struct vhost_vdpa *v = private;
172 struct eventfd_ctx *config_ctx = v->config_ctx;
173
174 if (config_ctx)
175 eventfd_signal(config_ctx, 1);
176
177 return IRQ_HANDLED;
178 }
179
vhost_vdpa_setup_vq_irq(struct vhost_vdpa * v,u16 qid)180 static void vhost_vdpa_setup_vq_irq(struct vhost_vdpa *v, u16 qid)
181 {
182 struct vhost_virtqueue *vq = &v->vqs[qid];
183 const struct vdpa_config_ops *ops = v->vdpa->config;
184 struct vdpa_device *vdpa = v->vdpa;
185 int ret, irq;
186
187 if (!ops->get_vq_irq)
188 return;
189
190 irq = ops->get_vq_irq(vdpa, qid);
191 if (irq < 0)
192 return;
193
194 if (!vq->call_ctx.ctx)
195 return;
196
197 vq->call_ctx.producer.irq = irq;
198 ret = irq_bypass_register_producer(&vq->call_ctx.producer);
199 if (unlikely(ret))
200 dev_info(&v->dev, "vq %u, irq bypass producer (token %p) registration fails, ret = %d\n",
201 qid, vq->call_ctx.producer.token, ret);
202 }
203
vhost_vdpa_unsetup_vq_irq(struct vhost_vdpa * v,u16 qid)204 static void vhost_vdpa_unsetup_vq_irq(struct vhost_vdpa *v, u16 qid)
205 {
206 struct vhost_virtqueue *vq = &v->vqs[qid];
207
208 irq_bypass_unregister_producer(&vq->call_ctx.producer);
209 }
210
vhost_vdpa_reset(struct vhost_vdpa * v)211 static int vhost_vdpa_reset(struct vhost_vdpa *v)
212 {
213 struct vdpa_device *vdpa = v->vdpa;
214
215 v->in_batch = 0;
216
217 return vdpa_reset(vdpa);
218 }
219
vhost_vdpa_bind_mm(struct vhost_vdpa * v)220 static long vhost_vdpa_bind_mm(struct vhost_vdpa *v)
221 {
222 struct vdpa_device *vdpa = v->vdpa;
223 const struct vdpa_config_ops *ops = vdpa->config;
224
225 if (!vdpa->use_va || !ops->bind_mm)
226 return 0;
227
228 return ops->bind_mm(vdpa, v->vdev.mm);
229 }
230
vhost_vdpa_unbind_mm(struct vhost_vdpa * v)231 static void vhost_vdpa_unbind_mm(struct vhost_vdpa *v)
232 {
233 struct vdpa_device *vdpa = v->vdpa;
234 const struct vdpa_config_ops *ops = vdpa->config;
235
236 if (!vdpa->use_va || !ops->unbind_mm)
237 return;
238
239 ops->unbind_mm(vdpa);
240 }
241
vhost_vdpa_get_device_id(struct vhost_vdpa * v,u8 __user * argp)242 static long vhost_vdpa_get_device_id(struct vhost_vdpa *v, u8 __user *argp)
243 {
244 struct vdpa_device *vdpa = v->vdpa;
245 const struct vdpa_config_ops *ops = vdpa->config;
246 u32 device_id;
247
248 device_id = ops->get_device_id(vdpa);
249
250 if (copy_to_user(argp, &device_id, sizeof(device_id)))
251 return -EFAULT;
252
253 return 0;
254 }
255
vhost_vdpa_get_status(struct vhost_vdpa * v,u8 __user * statusp)256 static long vhost_vdpa_get_status(struct vhost_vdpa *v, u8 __user *statusp)
257 {
258 struct vdpa_device *vdpa = v->vdpa;
259 const struct vdpa_config_ops *ops = vdpa->config;
260 u8 status;
261
262 status = ops->get_status(vdpa);
263
264 if (copy_to_user(statusp, &status, sizeof(status)))
265 return -EFAULT;
266
267 return 0;
268 }
269
vhost_vdpa_set_status(struct vhost_vdpa * v,u8 __user * statusp)270 static long vhost_vdpa_set_status(struct vhost_vdpa *v, u8 __user *statusp)
271 {
272 struct vdpa_device *vdpa = v->vdpa;
273 const struct vdpa_config_ops *ops = vdpa->config;
274 u8 status, status_old;
275 u32 nvqs = v->nvqs;
276 int ret;
277 u16 i;
278
279 if (copy_from_user(&status, statusp, sizeof(status)))
280 return -EFAULT;
281
282 status_old = ops->get_status(vdpa);
283
284 /*
285 * Userspace shouldn't remove status bits unless reset the
286 * status to 0.
287 */
288 if (status != 0 && (status_old & ~status) != 0)
289 return -EINVAL;
290
291 if ((status_old & VIRTIO_CONFIG_S_DRIVER_OK) && !(status & VIRTIO_CONFIG_S_DRIVER_OK))
292 for (i = 0; i < nvqs; i++)
293 vhost_vdpa_unsetup_vq_irq(v, i);
294
295 if (status == 0) {
296 ret = vdpa_reset(vdpa);
297 if (ret)
298 return ret;
299 } else
300 vdpa_set_status(vdpa, status);
301
302 if ((status & VIRTIO_CONFIG_S_DRIVER_OK) && !(status_old & VIRTIO_CONFIG_S_DRIVER_OK))
303 for (i = 0; i < nvqs; i++)
304 vhost_vdpa_setup_vq_irq(v, i);
305
306 return 0;
307 }
308
vhost_vdpa_config_validate(struct vhost_vdpa * v,struct vhost_vdpa_config * c)309 static int vhost_vdpa_config_validate(struct vhost_vdpa *v,
310 struct vhost_vdpa_config *c)
311 {
312 struct vdpa_device *vdpa = v->vdpa;
313 size_t size = vdpa->config->get_config_size(vdpa);
314
315 if (c->len == 0 || c->off > size)
316 return -EINVAL;
317
318 if (c->len > size - c->off)
319 return -E2BIG;
320
321 return 0;
322 }
323
vhost_vdpa_get_config(struct vhost_vdpa * v,struct vhost_vdpa_config __user * c)324 static long vhost_vdpa_get_config(struct vhost_vdpa *v,
325 struct vhost_vdpa_config __user *c)
326 {
327 struct vdpa_device *vdpa = v->vdpa;
328 struct vhost_vdpa_config config;
329 unsigned long size = offsetof(struct vhost_vdpa_config, buf);
330 u8 *buf;
331
332 if (copy_from_user(&config, c, size))
333 return -EFAULT;
334 if (vhost_vdpa_config_validate(v, &config))
335 return -EINVAL;
336 buf = kvzalloc(config.len, GFP_KERNEL);
337 if (!buf)
338 return -ENOMEM;
339
340 vdpa_get_config(vdpa, config.off, buf, config.len);
341
342 if (copy_to_user(c->buf, buf, config.len)) {
343 kvfree(buf);
344 return -EFAULT;
345 }
346
347 kvfree(buf);
348 return 0;
349 }
350
vhost_vdpa_set_config(struct vhost_vdpa * v,struct vhost_vdpa_config __user * c)351 static long vhost_vdpa_set_config(struct vhost_vdpa *v,
352 struct vhost_vdpa_config __user *c)
353 {
354 struct vdpa_device *vdpa = v->vdpa;
355 struct vhost_vdpa_config config;
356 unsigned long size = offsetof(struct vhost_vdpa_config, buf);
357 u8 *buf;
358
359 if (copy_from_user(&config, c, size))
360 return -EFAULT;
361 if (vhost_vdpa_config_validate(v, &config))
362 return -EINVAL;
363
364 buf = vmemdup_user(c->buf, config.len);
365 if (IS_ERR(buf))
366 return PTR_ERR(buf);
367
368 vdpa_set_config(vdpa, config.off, buf, config.len);
369
370 kvfree(buf);
371 return 0;
372 }
373
vhost_vdpa_can_suspend(const struct vhost_vdpa * v)374 static bool vhost_vdpa_can_suspend(const struct vhost_vdpa *v)
375 {
376 struct vdpa_device *vdpa = v->vdpa;
377 const struct vdpa_config_ops *ops = vdpa->config;
378
379 return ops->suspend;
380 }
381
vhost_vdpa_can_resume(const struct vhost_vdpa * v)382 static bool vhost_vdpa_can_resume(const struct vhost_vdpa *v)
383 {
384 struct vdpa_device *vdpa = v->vdpa;
385 const struct vdpa_config_ops *ops = vdpa->config;
386
387 return ops->resume;
388 }
389
vhost_vdpa_get_features(struct vhost_vdpa * v,u64 __user * featurep)390 static long vhost_vdpa_get_features(struct vhost_vdpa *v, u64 __user *featurep)
391 {
392 struct vdpa_device *vdpa = v->vdpa;
393 const struct vdpa_config_ops *ops = vdpa->config;
394 u64 features;
395
396 features = ops->get_device_features(vdpa);
397
398 if (copy_to_user(featurep, &features, sizeof(features)))
399 return -EFAULT;
400
401 return 0;
402 }
403
vhost_vdpa_get_backend_features(const struct vhost_vdpa * v)404 static u64 vhost_vdpa_get_backend_features(const struct vhost_vdpa *v)
405 {
406 struct vdpa_device *vdpa = v->vdpa;
407 const struct vdpa_config_ops *ops = vdpa->config;
408
409 if (!ops->get_backend_features)
410 return 0;
411 else
412 return ops->get_backend_features(vdpa);
413 }
414
vhost_vdpa_set_features(struct vhost_vdpa * v,u64 __user * featurep)415 static long vhost_vdpa_set_features(struct vhost_vdpa *v, u64 __user *featurep)
416 {
417 struct vdpa_device *vdpa = v->vdpa;
418 const struct vdpa_config_ops *ops = vdpa->config;
419 struct vhost_dev *d = &v->vdev;
420 u64 actual_features;
421 u64 features;
422 int i;
423
424 /*
425 * It's not allowed to change the features after they have
426 * been negotiated.
427 */
428 if (ops->get_status(vdpa) & VIRTIO_CONFIG_S_FEATURES_OK)
429 return -EBUSY;
430
431 if (copy_from_user(&features, featurep, sizeof(features)))
432 return -EFAULT;
433
434 if (vdpa_set_features(vdpa, features))
435 return -EINVAL;
436
437 /* let the vqs know what has been configured */
438 actual_features = ops->get_driver_features(vdpa);
439 for (i = 0; i < d->nvqs; ++i) {
440 struct vhost_virtqueue *vq = d->vqs[i];
441
442 mutex_lock(&vq->mutex);
443 vq->acked_features = actual_features;
444 mutex_unlock(&vq->mutex);
445 }
446
447 return 0;
448 }
449
vhost_vdpa_get_vring_num(struct vhost_vdpa * v,u16 __user * argp)450 static long vhost_vdpa_get_vring_num(struct vhost_vdpa *v, u16 __user *argp)
451 {
452 struct vdpa_device *vdpa = v->vdpa;
453 const struct vdpa_config_ops *ops = vdpa->config;
454 u16 num;
455
456 num = ops->get_vq_num_max(vdpa);
457
458 if (copy_to_user(argp, &num, sizeof(num)))
459 return -EFAULT;
460
461 return 0;
462 }
463
vhost_vdpa_config_put(struct vhost_vdpa * v)464 static void vhost_vdpa_config_put(struct vhost_vdpa *v)
465 {
466 if (v->config_ctx) {
467 eventfd_ctx_put(v->config_ctx);
468 v->config_ctx = NULL;
469 }
470 }
471
vhost_vdpa_set_config_call(struct vhost_vdpa * v,u32 __user * argp)472 static long vhost_vdpa_set_config_call(struct vhost_vdpa *v, u32 __user *argp)
473 {
474 struct vdpa_callback cb;
475 int fd;
476 struct eventfd_ctx *ctx;
477
478 cb.callback = vhost_vdpa_config_cb;
479 cb.private = v;
480 if (copy_from_user(&fd, argp, sizeof(fd)))
481 return -EFAULT;
482
483 ctx = fd == VHOST_FILE_UNBIND ? NULL : eventfd_ctx_fdget(fd);
484 swap(ctx, v->config_ctx);
485
486 if (!IS_ERR_OR_NULL(ctx))
487 eventfd_ctx_put(ctx);
488
489 if (IS_ERR(v->config_ctx)) {
490 long ret = PTR_ERR(v->config_ctx);
491
492 v->config_ctx = NULL;
493 return ret;
494 }
495
496 v->vdpa->config->set_config_cb(v->vdpa, &cb);
497
498 return 0;
499 }
500
vhost_vdpa_get_iova_range(struct vhost_vdpa * v,u32 __user * argp)501 static long vhost_vdpa_get_iova_range(struct vhost_vdpa *v, u32 __user *argp)
502 {
503 struct vhost_vdpa_iova_range range = {
504 .first = v->range.first,
505 .last = v->range.last,
506 };
507
508 if (copy_to_user(argp, &range, sizeof(range)))
509 return -EFAULT;
510 return 0;
511 }
512
vhost_vdpa_get_config_size(struct vhost_vdpa * v,u32 __user * argp)513 static long vhost_vdpa_get_config_size(struct vhost_vdpa *v, u32 __user *argp)
514 {
515 struct vdpa_device *vdpa = v->vdpa;
516 const struct vdpa_config_ops *ops = vdpa->config;
517 u32 size;
518
519 size = ops->get_config_size(vdpa);
520
521 if (copy_to_user(argp, &size, sizeof(size)))
522 return -EFAULT;
523
524 return 0;
525 }
526
vhost_vdpa_get_vqs_count(struct vhost_vdpa * v,u32 __user * argp)527 static long vhost_vdpa_get_vqs_count(struct vhost_vdpa *v, u32 __user *argp)
528 {
529 struct vdpa_device *vdpa = v->vdpa;
530
531 if (copy_to_user(argp, &vdpa->nvqs, sizeof(vdpa->nvqs)))
532 return -EFAULT;
533
534 return 0;
535 }
536
537 /* After a successful return of ioctl the device must not process more
538 * virtqueue descriptors. The device can answer to read or writes of config
539 * fields as if it were not suspended. In particular, writing to "queue_enable"
540 * with a value of 1 will not make the device start processing buffers.
541 */
vhost_vdpa_suspend(struct vhost_vdpa * v)542 static long vhost_vdpa_suspend(struct vhost_vdpa *v)
543 {
544 struct vdpa_device *vdpa = v->vdpa;
545 const struct vdpa_config_ops *ops = vdpa->config;
546
547 if (!ops->suspend)
548 return -EOPNOTSUPP;
549
550 return ops->suspend(vdpa);
551 }
552
553 /* After a successful return of this ioctl the device resumes processing
554 * virtqueue descriptors. The device becomes fully operational the same way it
555 * was before it was suspended.
556 */
vhost_vdpa_resume(struct vhost_vdpa * v)557 static long vhost_vdpa_resume(struct vhost_vdpa *v)
558 {
559 struct vdpa_device *vdpa = v->vdpa;
560 const struct vdpa_config_ops *ops = vdpa->config;
561
562 if (!ops->resume)
563 return -EOPNOTSUPP;
564
565 return ops->resume(vdpa);
566 }
567
vhost_vdpa_vring_ioctl(struct vhost_vdpa * v,unsigned int cmd,void __user * argp)568 static long vhost_vdpa_vring_ioctl(struct vhost_vdpa *v, unsigned int cmd,
569 void __user *argp)
570 {
571 struct vdpa_device *vdpa = v->vdpa;
572 const struct vdpa_config_ops *ops = vdpa->config;
573 struct vdpa_vq_state vq_state;
574 struct vdpa_callback cb;
575 struct vhost_virtqueue *vq;
576 struct vhost_vring_state s;
577 u32 idx;
578 long r;
579
580 r = get_user(idx, (u32 __user *)argp);
581 if (r < 0)
582 return r;
583
584 if (idx >= v->nvqs)
585 return -ENOBUFS;
586
587 idx = array_index_nospec(idx, v->nvqs);
588 vq = &v->vqs[idx];
589
590 switch (cmd) {
591 case VHOST_VDPA_SET_VRING_ENABLE:
592 if (copy_from_user(&s, argp, sizeof(s)))
593 return -EFAULT;
594 ops->set_vq_ready(vdpa, idx, s.num);
595 return 0;
596 case VHOST_VDPA_GET_VRING_GROUP:
597 if (!ops->get_vq_group)
598 return -EOPNOTSUPP;
599 s.index = idx;
600 s.num = ops->get_vq_group(vdpa, idx);
601 if (s.num >= vdpa->ngroups)
602 return -EIO;
603 else if (copy_to_user(argp, &s, sizeof(s)))
604 return -EFAULT;
605 return 0;
606 case VHOST_VDPA_SET_GROUP_ASID:
607 if (copy_from_user(&s, argp, sizeof(s)))
608 return -EFAULT;
609 if (s.num >= vdpa->nas)
610 return -EINVAL;
611 if (!ops->set_group_asid)
612 return -EOPNOTSUPP;
613 return ops->set_group_asid(vdpa, idx, s.num);
614 case VHOST_GET_VRING_BASE:
615 r = ops->get_vq_state(v->vdpa, idx, &vq_state);
616 if (r)
617 return r;
618
619 if (vhost_has_feature(vq, VIRTIO_F_RING_PACKED)) {
620 vq->last_avail_idx = vq_state.packed.last_avail_idx |
621 (vq_state.packed.last_avail_counter << 15);
622 vq->last_used_idx = vq_state.packed.last_used_idx |
623 (vq_state.packed.last_used_counter << 15);
624 } else {
625 vq->last_avail_idx = vq_state.split.avail_index;
626 }
627 break;
628 case VHOST_SET_VRING_CALL:
629 if (vq->call_ctx.ctx) {
630 if (ops->get_status(vdpa) &
631 VIRTIO_CONFIG_S_DRIVER_OK)
632 vhost_vdpa_unsetup_vq_irq(v, idx);
633 vq->call_ctx.producer.token = NULL;
634 }
635 break;
636 }
637
638 r = vhost_vring_ioctl(&v->vdev, cmd, argp);
639 if (r)
640 return r;
641
642 switch (cmd) {
643 case VHOST_SET_VRING_ADDR:
644 if (ops->set_vq_address(vdpa, idx,
645 (u64)(uintptr_t)vq->desc,
646 (u64)(uintptr_t)vq->avail,
647 (u64)(uintptr_t)vq->used))
648 r = -EINVAL;
649 break;
650
651 case VHOST_SET_VRING_BASE:
652 if (vhost_has_feature(vq, VIRTIO_F_RING_PACKED)) {
653 vq_state.packed.last_avail_idx = vq->last_avail_idx & 0x7fff;
654 vq_state.packed.last_avail_counter = !!(vq->last_avail_idx & 0x8000);
655 vq_state.packed.last_used_idx = vq->last_used_idx & 0x7fff;
656 vq_state.packed.last_used_counter = !!(vq->last_used_idx & 0x8000);
657 } else {
658 vq_state.split.avail_index = vq->last_avail_idx;
659 }
660 r = ops->set_vq_state(vdpa, idx, &vq_state);
661 break;
662
663 case VHOST_SET_VRING_CALL:
664 if (vq->call_ctx.ctx) {
665 cb.callback = vhost_vdpa_virtqueue_cb;
666 cb.private = vq;
667 cb.trigger = vq->call_ctx.ctx;
668 vq->call_ctx.producer.token = vq->call_ctx.ctx;
669 if (ops->get_status(vdpa) &
670 VIRTIO_CONFIG_S_DRIVER_OK)
671 vhost_vdpa_setup_vq_irq(v, idx);
672 } else {
673 cb.callback = NULL;
674 cb.private = NULL;
675 cb.trigger = NULL;
676 }
677 ops->set_vq_cb(vdpa, idx, &cb);
678 break;
679
680 case VHOST_SET_VRING_NUM:
681 ops->set_vq_num(vdpa, idx, vq->num);
682 break;
683 }
684
685 return r;
686 }
687
vhost_vdpa_unlocked_ioctl(struct file * filep,unsigned int cmd,unsigned long arg)688 static long vhost_vdpa_unlocked_ioctl(struct file *filep,
689 unsigned int cmd, unsigned long arg)
690 {
691 struct vhost_vdpa *v = filep->private_data;
692 struct vhost_dev *d = &v->vdev;
693 void __user *argp = (void __user *)arg;
694 u64 __user *featurep = argp;
695 u64 features;
696 long r = 0;
697
698 if (cmd == VHOST_SET_BACKEND_FEATURES) {
699 if (copy_from_user(&features, featurep, sizeof(features)))
700 return -EFAULT;
701 if (features & ~(VHOST_VDPA_BACKEND_FEATURES |
702 BIT_ULL(VHOST_BACKEND_F_SUSPEND) |
703 BIT_ULL(VHOST_BACKEND_F_RESUME) |
704 BIT_ULL(VHOST_BACKEND_F_ENABLE_AFTER_DRIVER_OK)))
705 return -EOPNOTSUPP;
706 if ((features & BIT_ULL(VHOST_BACKEND_F_SUSPEND)) &&
707 !vhost_vdpa_can_suspend(v))
708 return -EOPNOTSUPP;
709 if ((features & BIT_ULL(VHOST_BACKEND_F_RESUME)) &&
710 !vhost_vdpa_can_resume(v))
711 return -EOPNOTSUPP;
712 vhost_set_backend_features(&v->vdev, features);
713 return 0;
714 }
715
716 mutex_lock(&d->mutex);
717
718 switch (cmd) {
719 case VHOST_VDPA_GET_DEVICE_ID:
720 r = vhost_vdpa_get_device_id(v, argp);
721 break;
722 case VHOST_VDPA_GET_STATUS:
723 r = vhost_vdpa_get_status(v, argp);
724 break;
725 case VHOST_VDPA_SET_STATUS:
726 r = vhost_vdpa_set_status(v, argp);
727 break;
728 case VHOST_VDPA_GET_CONFIG:
729 r = vhost_vdpa_get_config(v, argp);
730 break;
731 case VHOST_VDPA_SET_CONFIG:
732 r = vhost_vdpa_set_config(v, argp);
733 break;
734 case VHOST_GET_FEATURES:
735 r = vhost_vdpa_get_features(v, argp);
736 break;
737 case VHOST_SET_FEATURES:
738 r = vhost_vdpa_set_features(v, argp);
739 break;
740 case VHOST_VDPA_GET_VRING_NUM:
741 r = vhost_vdpa_get_vring_num(v, argp);
742 break;
743 case VHOST_VDPA_GET_GROUP_NUM:
744 if (copy_to_user(argp, &v->vdpa->ngroups,
745 sizeof(v->vdpa->ngroups)))
746 r = -EFAULT;
747 break;
748 case VHOST_VDPA_GET_AS_NUM:
749 if (copy_to_user(argp, &v->vdpa->nas, sizeof(v->vdpa->nas)))
750 r = -EFAULT;
751 break;
752 case VHOST_SET_LOG_BASE:
753 case VHOST_SET_LOG_FD:
754 r = -ENOIOCTLCMD;
755 break;
756 case VHOST_VDPA_SET_CONFIG_CALL:
757 r = vhost_vdpa_set_config_call(v, argp);
758 break;
759 case VHOST_GET_BACKEND_FEATURES:
760 features = VHOST_VDPA_BACKEND_FEATURES;
761 if (vhost_vdpa_can_suspend(v))
762 features |= BIT_ULL(VHOST_BACKEND_F_SUSPEND);
763 if (vhost_vdpa_can_resume(v))
764 features |= BIT_ULL(VHOST_BACKEND_F_RESUME);
765 features |= vhost_vdpa_get_backend_features(v);
766 if (copy_to_user(featurep, &features, sizeof(features)))
767 r = -EFAULT;
768 break;
769 case VHOST_VDPA_GET_IOVA_RANGE:
770 r = vhost_vdpa_get_iova_range(v, argp);
771 break;
772 case VHOST_VDPA_GET_CONFIG_SIZE:
773 r = vhost_vdpa_get_config_size(v, argp);
774 break;
775 case VHOST_VDPA_GET_VQS_COUNT:
776 r = vhost_vdpa_get_vqs_count(v, argp);
777 break;
778 case VHOST_VDPA_SUSPEND:
779 r = vhost_vdpa_suspend(v);
780 break;
781 case VHOST_VDPA_RESUME:
782 r = vhost_vdpa_resume(v);
783 break;
784 default:
785 r = vhost_dev_ioctl(&v->vdev, cmd, argp);
786 if (r == -ENOIOCTLCMD)
787 r = vhost_vdpa_vring_ioctl(v, cmd, argp);
788 break;
789 }
790
791 if (r)
792 goto out;
793
794 switch (cmd) {
795 case VHOST_SET_OWNER:
796 r = vhost_vdpa_bind_mm(v);
797 if (r)
798 vhost_dev_reset_owner(d, NULL);
799 break;
800 }
801 out:
802 mutex_unlock(&d->mutex);
803 return r;
804 }
vhost_vdpa_general_unmap(struct vhost_vdpa * v,struct vhost_iotlb_map * map,u32 asid)805 static void vhost_vdpa_general_unmap(struct vhost_vdpa *v,
806 struct vhost_iotlb_map *map, u32 asid)
807 {
808 struct vdpa_device *vdpa = v->vdpa;
809 const struct vdpa_config_ops *ops = vdpa->config;
810 if (ops->dma_map) {
811 ops->dma_unmap(vdpa, asid, map->start, map->size);
812 } else if (ops->set_map == NULL) {
813 iommu_unmap(v->domain, map->start, map->size);
814 }
815 }
816
vhost_vdpa_pa_unmap(struct vhost_vdpa * v,struct vhost_iotlb * iotlb,u64 start,u64 last,u32 asid)817 static void vhost_vdpa_pa_unmap(struct vhost_vdpa *v, struct vhost_iotlb *iotlb,
818 u64 start, u64 last, u32 asid)
819 {
820 struct vhost_dev *dev = &v->vdev;
821 struct vhost_iotlb_map *map;
822 struct page *page;
823 unsigned long pfn, pinned;
824
825 while ((map = vhost_iotlb_itree_first(iotlb, start, last)) != NULL) {
826 pinned = PFN_DOWN(map->size);
827 for (pfn = PFN_DOWN(map->addr);
828 pinned > 0; pfn++, pinned--) {
829 page = pfn_to_page(pfn);
830 if (map->perm & VHOST_ACCESS_WO)
831 set_page_dirty_lock(page);
832 unpin_user_page(page);
833 }
834 atomic64_sub(PFN_DOWN(map->size), &dev->mm->pinned_vm);
835 vhost_vdpa_general_unmap(v, map, asid);
836 vhost_iotlb_map_free(iotlb, map);
837 }
838 }
839
vhost_vdpa_va_unmap(struct vhost_vdpa * v,struct vhost_iotlb * iotlb,u64 start,u64 last,u32 asid)840 static void vhost_vdpa_va_unmap(struct vhost_vdpa *v, struct vhost_iotlb *iotlb,
841 u64 start, u64 last, u32 asid)
842 {
843 struct vhost_iotlb_map *map;
844 struct vdpa_map_file *map_file;
845
846 while ((map = vhost_iotlb_itree_first(iotlb, start, last)) != NULL) {
847 map_file = (struct vdpa_map_file *)map->opaque;
848 fput(map_file->file);
849 kfree(map_file);
850 vhost_vdpa_general_unmap(v, map, asid);
851 vhost_iotlb_map_free(iotlb, map);
852 }
853 }
854
vhost_vdpa_iotlb_unmap(struct vhost_vdpa * v,struct vhost_iotlb * iotlb,u64 start,u64 last,u32 asid)855 static void vhost_vdpa_iotlb_unmap(struct vhost_vdpa *v,
856 struct vhost_iotlb *iotlb, u64 start,
857 u64 last, u32 asid)
858 {
859 struct vdpa_device *vdpa = v->vdpa;
860
861 if (vdpa->use_va)
862 return vhost_vdpa_va_unmap(v, iotlb, start, last, asid);
863
864 return vhost_vdpa_pa_unmap(v, iotlb, start, last, asid);
865 }
866
perm_to_iommu_flags(u32 perm)867 static int perm_to_iommu_flags(u32 perm)
868 {
869 int flags = 0;
870
871 switch (perm) {
872 case VHOST_ACCESS_WO:
873 flags |= IOMMU_WRITE;
874 break;
875 case VHOST_ACCESS_RO:
876 flags |= IOMMU_READ;
877 break;
878 case VHOST_ACCESS_RW:
879 flags |= (IOMMU_WRITE | IOMMU_READ);
880 break;
881 default:
882 WARN(1, "invalidate vhost IOTLB permission\n");
883 break;
884 }
885
886 return flags | IOMMU_CACHE;
887 }
888
vhost_vdpa_map(struct vhost_vdpa * v,struct vhost_iotlb * iotlb,u64 iova,u64 size,u64 pa,u32 perm,void * opaque)889 static int vhost_vdpa_map(struct vhost_vdpa *v, struct vhost_iotlb *iotlb,
890 u64 iova, u64 size, u64 pa, u32 perm, void *opaque)
891 {
892 struct vhost_dev *dev = &v->vdev;
893 struct vdpa_device *vdpa = v->vdpa;
894 const struct vdpa_config_ops *ops = vdpa->config;
895 u32 asid = iotlb_to_asid(iotlb);
896 int r = 0;
897
898 r = vhost_iotlb_add_range_ctx(iotlb, iova, iova + size - 1,
899 pa, perm, opaque);
900 if (r)
901 return r;
902
903 if (ops->dma_map) {
904 r = ops->dma_map(vdpa, asid, iova, size, pa, perm, opaque);
905 } else if (ops->set_map) {
906 if (!v->in_batch)
907 r = ops->set_map(vdpa, asid, iotlb);
908 } else {
909 r = iommu_map(v->domain, iova, pa, size,
910 perm_to_iommu_flags(perm), GFP_KERNEL);
911 }
912 if (r) {
913 vhost_iotlb_del_range(iotlb, iova, iova + size - 1);
914 return r;
915 }
916
917 if (!vdpa->use_va)
918 atomic64_add(PFN_DOWN(size), &dev->mm->pinned_vm);
919
920 return 0;
921 }
922
vhost_vdpa_unmap(struct vhost_vdpa * v,struct vhost_iotlb * iotlb,u64 iova,u64 size)923 static void vhost_vdpa_unmap(struct vhost_vdpa *v,
924 struct vhost_iotlb *iotlb,
925 u64 iova, u64 size)
926 {
927 struct vdpa_device *vdpa = v->vdpa;
928 const struct vdpa_config_ops *ops = vdpa->config;
929 u32 asid = iotlb_to_asid(iotlb);
930
931 vhost_vdpa_iotlb_unmap(v, iotlb, iova, iova + size - 1, asid);
932
933 if (ops->set_map) {
934 if (!v->in_batch)
935 ops->set_map(vdpa, asid, iotlb);
936 }
937
938 }
939
vhost_vdpa_va_map(struct vhost_vdpa * v,struct vhost_iotlb * iotlb,u64 iova,u64 size,u64 uaddr,u32 perm)940 static int vhost_vdpa_va_map(struct vhost_vdpa *v,
941 struct vhost_iotlb *iotlb,
942 u64 iova, u64 size, u64 uaddr, u32 perm)
943 {
944 struct vhost_dev *dev = &v->vdev;
945 u64 offset, map_size, map_iova = iova;
946 struct vdpa_map_file *map_file;
947 struct vm_area_struct *vma;
948 int ret = 0;
949
950 mmap_read_lock(dev->mm);
951
952 while (size) {
953 vma = find_vma(dev->mm, uaddr);
954 if (!vma) {
955 ret = -EINVAL;
956 break;
957 }
958 map_size = min(size, vma->vm_end - uaddr);
959 if (!(vma->vm_file && (vma->vm_flags & VM_SHARED) &&
960 !(vma->vm_flags & (VM_IO | VM_PFNMAP))))
961 goto next;
962
963 map_file = kzalloc(sizeof(*map_file), GFP_KERNEL);
964 if (!map_file) {
965 ret = -ENOMEM;
966 break;
967 }
968 offset = (vma->vm_pgoff << PAGE_SHIFT) + uaddr - vma->vm_start;
969 map_file->offset = offset;
970 map_file->file = get_file(vma->vm_file);
971 ret = vhost_vdpa_map(v, iotlb, map_iova, map_size, uaddr,
972 perm, map_file);
973 if (ret) {
974 fput(map_file->file);
975 kfree(map_file);
976 break;
977 }
978 next:
979 size -= map_size;
980 uaddr += map_size;
981 map_iova += map_size;
982 }
983 if (ret)
984 vhost_vdpa_unmap(v, iotlb, iova, map_iova - iova);
985
986 mmap_read_unlock(dev->mm);
987
988 return ret;
989 }
990
vhost_vdpa_pa_map(struct vhost_vdpa * v,struct vhost_iotlb * iotlb,u64 iova,u64 size,u64 uaddr,u32 perm)991 static int vhost_vdpa_pa_map(struct vhost_vdpa *v,
992 struct vhost_iotlb *iotlb,
993 u64 iova, u64 size, u64 uaddr, u32 perm)
994 {
995 struct vhost_dev *dev = &v->vdev;
996 struct page **page_list;
997 unsigned long list_size = PAGE_SIZE / sizeof(struct page *);
998 unsigned int gup_flags = FOLL_LONGTERM;
999 unsigned long npages, cur_base, map_pfn, last_pfn = 0;
1000 unsigned long lock_limit, sz2pin, nchunks, i;
1001 u64 start = iova;
1002 long pinned;
1003 int ret = 0;
1004
1005 /* Limit the use of memory for bookkeeping */
1006 page_list = (struct page **) __get_free_page(GFP_KERNEL);
1007 if (!page_list)
1008 return -ENOMEM;
1009
1010 if (perm & VHOST_ACCESS_WO)
1011 gup_flags |= FOLL_WRITE;
1012
1013 npages = PFN_UP(size + (iova & ~PAGE_MASK));
1014 if (!npages) {
1015 ret = -EINVAL;
1016 goto free;
1017 }
1018
1019 mmap_read_lock(dev->mm);
1020
1021 lock_limit = PFN_DOWN(rlimit(RLIMIT_MEMLOCK));
1022 if (npages + atomic64_read(&dev->mm->pinned_vm) > lock_limit) {
1023 ret = -ENOMEM;
1024 goto unlock;
1025 }
1026
1027 cur_base = uaddr & PAGE_MASK;
1028 iova &= PAGE_MASK;
1029 nchunks = 0;
1030
1031 while (npages) {
1032 sz2pin = min_t(unsigned long, npages, list_size);
1033 pinned = pin_user_pages(cur_base, sz2pin,
1034 gup_flags, page_list);
1035 if (sz2pin != pinned) {
1036 if (pinned < 0) {
1037 ret = pinned;
1038 } else {
1039 unpin_user_pages(page_list, pinned);
1040 ret = -ENOMEM;
1041 }
1042 goto out;
1043 }
1044 nchunks++;
1045
1046 if (!last_pfn)
1047 map_pfn = page_to_pfn(page_list[0]);
1048
1049 for (i = 0; i < pinned; i++) {
1050 unsigned long this_pfn = page_to_pfn(page_list[i]);
1051 u64 csize;
1052
1053 if (last_pfn && (this_pfn != last_pfn + 1)) {
1054 /* Pin a contiguous chunk of memory */
1055 csize = PFN_PHYS(last_pfn - map_pfn + 1);
1056 ret = vhost_vdpa_map(v, iotlb, iova, csize,
1057 PFN_PHYS(map_pfn),
1058 perm, NULL);
1059 if (ret) {
1060 /*
1061 * Unpin the pages that are left unmapped
1062 * from this point on in the current
1063 * page_list. The remaining outstanding
1064 * ones which may stride across several
1065 * chunks will be covered in the common
1066 * error path subsequently.
1067 */
1068 unpin_user_pages(&page_list[i],
1069 pinned - i);
1070 goto out;
1071 }
1072
1073 map_pfn = this_pfn;
1074 iova += csize;
1075 nchunks = 0;
1076 }
1077
1078 last_pfn = this_pfn;
1079 }
1080
1081 cur_base += PFN_PHYS(pinned);
1082 npages -= pinned;
1083 }
1084
1085 /* Pin the rest chunk */
1086 ret = vhost_vdpa_map(v, iotlb, iova, PFN_PHYS(last_pfn - map_pfn + 1),
1087 PFN_PHYS(map_pfn), perm, NULL);
1088 out:
1089 if (ret) {
1090 if (nchunks) {
1091 unsigned long pfn;
1092
1093 /*
1094 * Unpin the outstanding pages which are yet to be
1095 * mapped but haven't due to vdpa_map() or
1096 * pin_user_pages() failure.
1097 *
1098 * Mapped pages are accounted in vdpa_map(), hence
1099 * the corresponding unpinning will be handled by
1100 * vdpa_unmap().
1101 */
1102 WARN_ON(!last_pfn);
1103 for (pfn = map_pfn; pfn <= last_pfn; pfn++)
1104 unpin_user_page(pfn_to_page(pfn));
1105 }
1106 vhost_vdpa_unmap(v, iotlb, start, size);
1107 }
1108 unlock:
1109 mmap_read_unlock(dev->mm);
1110 free:
1111 free_page((unsigned long)page_list);
1112 return ret;
1113
1114 }
1115
vhost_vdpa_process_iotlb_update(struct vhost_vdpa * v,struct vhost_iotlb * iotlb,struct vhost_iotlb_msg * msg)1116 static int vhost_vdpa_process_iotlb_update(struct vhost_vdpa *v,
1117 struct vhost_iotlb *iotlb,
1118 struct vhost_iotlb_msg *msg)
1119 {
1120 struct vdpa_device *vdpa = v->vdpa;
1121
1122 if (msg->iova < v->range.first || !msg->size ||
1123 msg->iova > U64_MAX - msg->size + 1 ||
1124 msg->iova + msg->size - 1 > v->range.last)
1125 return -EINVAL;
1126
1127 if (vhost_iotlb_itree_first(iotlb, msg->iova,
1128 msg->iova + msg->size - 1))
1129 return -EEXIST;
1130
1131 if (vdpa->use_va)
1132 return vhost_vdpa_va_map(v, iotlb, msg->iova, msg->size,
1133 msg->uaddr, msg->perm);
1134
1135 return vhost_vdpa_pa_map(v, iotlb, msg->iova, msg->size, msg->uaddr,
1136 msg->perm);
1137 }
1138
vhost_vdpa_process_iotlb_msg(struct vhost_dev * dev,u32 asid,struct vhost_iotlb_msg * msg)1139 static int vhost_vdpa_process_iotlb_msg(struct vhost_dev *dev, u32 asid,
1140 struct vhost_iotlb_msg *msg)
1141 {
1142 struct vhost_vdpa *v = container_of(dev, struct vhost_vdpa, vdev);
1143 struct vdpa_device *vdpa = v->vdpa;
1144 const struct vdpa_config_ops *ops = vdpa->config;
1145 struct vhost_iotlb *iotlb = NULL;
1146 struct vhost_vdpa_as *as = NULL;
1147 int r = 0;
1148
1149 mutex_lock(&dev->mutex);
1150
1151 r = vhost_dev_check_owner(dev);
1152 if (r)
1153 goto unlock;
1154
1155 if (msg->type == VHOST_IOTLB_UPDATE ||
1156 msg->type == VHOST_IOTLB_BATCH_BEGIN) {
1157 as = vhost_vdpa_find_alloc_as(v, asid);
1158 if (!as) {
1159 dev_err(&v->dev, "can't find and alloc asid %d\n",
1160 asid);
1161 r = -EINVAL;
1162 goto unlock;
1163 }
1164 iotlb = &as->iotlb;
1165 } else
1166 iotlb = asid_to_iotlb(v, asid);
1167
1168 if ((v->in_batch && v->batch_asid != asid) || !iotlb) {
1169 if (v->in_batch && v->batch_asid != asid) {
1170 dev_info(&v->dev, "batch id %d asid %d\n",
1171 v->batch_asid, asid);
1172 }
1173 if (!iotlb)
1174 dev_err(&v->dev, "no iotlb for asid %d\n", asid);
1175 r = -EINVAL;
1176 goto unlock;
1177 }
1178
1179 switch (msg->type) {
1180 case VHOST_IOTLB_UPDATE:
1181 r = vhost_vdpa_process_iotlb_update(v, iotlb, msg);
1182 break;
1183 case VHOST_IOTLB_INVALIDATE:
1184 vhost_vdpa_unmap(v, iotlb, msg->iova, msg->size);
1185 break;
1186 case VHOST_IOTLB_BATCH_BEGIN:
1187 v->batch_asid = asid;
1188 v->in_batch = true;
1189 break;
1190 case VHOST_IOTLB_BATCH_END:
1191 if (v->in_batch && ops->set_map)
1192 ops->set_map(vdpa, asid, iotlb);
1193 v->in_batch = false;
1194 break;
1195 default:
1196 r = -EINVAL;
1197 break;
1198 }
1199 unlock:
1200 mutex_unlock(&dev->mutex);
1201
1202 return r;
1203 }
1204
vhost_vdpa_chr_write_iter(struct kiocb * iocb,struct iov_iter * from)1205 static ssize_t vhost_vdpa_chr_write_iter(struct kiocb *iocb,
1206 struct iov_iter *from)
1207 {
1208 struct file *file = iocb->ki_filp;
1209 struct vhost_vdpa *v = file->private_data;
1210 struct vhost_dev *dev = &v->vdev;
1211
1212 return vhost_chr_write_iter(dev, from);
1213 }
1214
vhost_vdpa_alloc_domain(struct vhost_vdpa * v)1215 static int vhost_vdpa_alloc_domain(struct vhost_vdpa *v)
1216 {
1217 struct vdpa_device *vdpa = v->vdpa;
1218 const struct vdpa_config_ops *ops = vdpa->config;
1219 struct device *dma_dev = vdpa_get_dma_dev(vdpa);
1220 const struct bus_type *bus;
1221 int ret;
1222
1223 /* Device want to do DMA by itself */
1224 if (ops->set_map || ops->dma_map)
1225 return 0;
1226
1227 bus = dma_dev->bus;
1228 if (!bus)
1229 return -EFAULT;
1230
1231 if (!device_iommu_capable(dma_dev, IOMMU_CAP_CACHE_COHERENCY)) {
1232 dev_warn_once(&v->dev,
1233 "Failed to allocate domain, device is not IOMMU cache coherent capable\n");
1234 return -ENOTSUPP;
1235 }
1236
1237 v->domain = iommu_domain_alloc(bus);
1238 if (!v->domain)
1239 return -EIO;
1240
1241 ret = iommu_attach_device(v->domain, dma_dev);
1242 if (ret)
1243 goto err_attach;
1244
1245 return 0;
1246
1247 err_attach:
1248 iommu_domain_free(v->domain);
1249 v->domain = NULL;
1250 return ret;
1251 }
1252
vhost_vdpa_free_domain(struct vhost_vdpa * v)1253 static void vhost_vdpa_free_domain(struct vhost_vdpa *v)
1254 {
1255 struct vdpa_device *vdpa = v->vdpa;
1256 struct device *dma_dev = vdpa_get_dma_dev(vdpa);
1257
1258 if (v->domain) {
1259 iommu_detach_device(v->domain, dma_dev);
1260 iommu_domain_free(v->domain);
1261 }
1262
1263 v->domain = NULL;
1264 }
1265
vhost_vdpa_set_iova_range(struct vhost_vdpa * v)1266 static void vhost_vdpa_set_iova_range(struct vhost_vdpa *v)
1267 {
1268 struct vdpa_iova_range *range = &v->range;
1269 struct vdpa_device *vdpa = v->vdpa;
1270 const struct vdpa_config_ops *ops = vdpa->config;
1271
1272 if (ops->get_iova_range) {
1273 *range = ops->get_iova_range(vdpa);
1274 } else if (v->domain && v->domain->geometry.force_aperture) {
1275 range->first = v->domain->geometry.aperture_start;
1276 range->last = v->domain->geometry.aperture_end;
1277 } else {
1278 range->first = 0;
1279 range->last = ULLONG_MAX;
1280 }
1281 }
1282
vhost_vdpa_cleanup(struct vhost_vdpa * v)1283 static void vhost_vdpa_cleanup(struct vhost_vdpa *v)
1284 {
1285 struct vhost_vdpa_as *as;
1286 u32 asid;
1287
1288 for (asid = 0; asid < v->vdpa->nas; asid++) {
1289 as = asid_to_as(v, asid);
1290 if (as)
1291 vhost_vdpa_remove_as(v, asid);
1292 }
1293
1294 vhost_vdpa_free_domain(v);
1295 vhost_dev_cleanup(&v->vdev);
1296 kfree(v->vdev.vqs);
1297 }
1298
vhost_vdpa_open(struct inode * inode,struct file * filep)1299 static int vhost_vdpa_open(struct inode *inode, struct file *filep)
1300 {
1301 struct vhost_vdpa *v;
1302 struct vhost_dev *dev;
1303 struct vhost_virtqueue **vqs;
1304 int r, opened;
1305 u32 i, nvqs;
1306
1307 v = container_of(inode->i_cdev, struct vhost_vdpa, cdev);
1308
1309 opened = atomic_cmpxchg(&v->opened, 0, 1);
1310 if (opened)
1311 return -EBUSY;
1312
1313 nvqs = v->nvqs;
1314 r = vhost_vdpa_reset(v);
1315 if (r)
1316 goto err;
1317
1318 vqs = kmalloc_array(nvqs, sizeof(*vqs), GFP_KERNEL);
1319 if (!vqs) {
1320 r = -ENOMEM;
1321 goto err;
1322 }
1323
1324 dev = &v->vdev;
1325 for (i = 0; i < nvqs; i++) {
1326 vqs[i] = &v->vqs[i];
1327 vqs[i]->handle_kick = handle_vq_kick;
1328 vqs[i]->call_ctx.ctx = NULL;
1329 }
1330 vhost_dev_init(dev, vqs, nvqs, 0, 0, 0, false,
1331 vhost_vdpa_process_iotlb_msg);
1332
1333 r = vhost_vdpa_alloc_domain(v);
1334 if (r)
1335 goto err_alloc_domain;
1336
1337 vhost_vdpa_set_iova_range(v);
1338
1339 filep->private_data = v;
1340
1341 return 0;
1342
1343 err_alloc_domain:
1344 vhost_vdpa_cleanup(v);
1345 err:
1346 atomic_dec(&v->opened);
1347 return r;
1348 }
1349
vhost_vdpa_clean_irq(struct vhost_vdpa * v)1350 static void vhost_vdpa_clean_irq(struct vhost_vdpa *v)
1351 {
1352 u32 i;
1353
1354 for (i = 0; i < v->nvqs; i++)
1355 vhost_vdpa_unsetup_vq_irq(v, i);
1356 }
1357
vhost_vdpa_release(struct inode * inode,struct file * filep)1358 static int vhost_vdpa_release(struct inode *inode, struct file *filep)
1359 {
1360 struct vhost_vdpa *v = filep->private_data;
1361 struct vhost_dev *d = &v->vdev;
1362
1363 mutex_lock(&d->mutex);
1364 filep->private_data = NULL;
1365 vhost_vdpa_clean_irq(v);
1366 vhost_vdpa_reset(v);
1367 vhost_dev_stop(&v->vdev);
1368 vhost_vdpa_unbind_mm(v);
1369 vhost_vdpa_config_put(v);
1370 vhost_vdpa_cleanup(v);
1371 mutex_unlock(&d->mutex);
1372
1373 atomic_dec(&v->opened);
1374 complete(&v->completion);
1375
1376 return 0;
1377 }
1378
1379 #ifdef CONFIG_MMU
vhost_vdpa_fault(struct vm_fault * vmf)1380 static vm_fault_t vhost_vdpa_fault(struct vm_fault *vmf)
1381 {
1382 struct vhost_vdpa *v = vmf->vma->vm_file->private_data;
1383 struct vdpa_device *vdpa = v->vdpa;
1384 const struct vdpa_config_ops *ops = vdpa->config;
1385 struct vdpa_notification_area notify;
1386 struct vm_area_struct *vma = vmf->vma;
1387 u16 index = vma->vm_pgoff;
1388
1389 notify = ops->get_vq_notification(vdpa, index);
1390
1391 return vmf_insert_pfn(vma, vmf->address & PAGE_MASK, PFN_DOWN(notify.addr));
1392 }
1393
1394 static const struct vm_operations_struct vhost_vdpa_vm_ops = {
1395 .fault = vhost_vdpa_fault,
1396 };
1397
vhost_vdpa_mmap(struct file * file,struct vm_area_struct * vma)1398 static int vhost_vdpa_mmap(struct file *file, struct vm_area_struct *vma)
1399 {
1400 struct vhost_vdpa *v = vma->vm_file->private_data;
1401 struct vdpa_device *vdpa = v->vdpa;
1402 const struct vdpa_config_ops *ops = vdpa->config;
1403 struct vdpa_notification_area notify;
1404 unsigned long index = vma->vm_pgoff;
1405
1406 if (vma->vm_end - vma->vm_start != PAGE_SIZE)
1407 return -EINVAL;
1408 if ((vma->vm_flags & VM_SHARED) == 0)
1409 return -EINVAL;
1410 if (vma->vm_flags & VM_READ)
1411 return -EINVAL;
1412 if (index > 65535)
1413 return -EINVAL;
1414 if (!ops->get_vq_notification)
1415 return -ENOTSUPP;
1416
1417 /* To be safe and easily modelled by userspace, We only
1418 * support the doorbell which sits on the page boundary and
1419 * does not share the page with other registers.
1420 */
1421 notify = ops->get_vq_notification(vdpa, index);
1422 if (notify.addr & (PAGE_SIZE - 1))
1423 return -EINVAL;
1424 if (vma->vm_end - vma->vm_start != notify.size)
1425 return -ENOTSUPP;
1426
1427 vm_flags_set(vma, VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP);
1428 vma->vm_ops = &vhost_vdpa_vm_ops;
1429 return 0;
1430 }
1431 #endif /* CONFIG_MMU */
1432
1433 static const struct file_operations vhost_vdpa_fops = {
1434 .owner = THIS_MODULE,
1435 .open = vhost_vdpa_open,
1436 .release = vhost_vdpa_release,
1437 .write_iter = vhost_vdpa_chr_write_iter,
1438 .unlocked_ioctl = vhost_vdpa_unlocked_ioctl,
1439 #ifdef CONFIG_MMU
1440 .mmap = vhost_vdpa_mmap,
1441 #endif /* CONFIG_MMU */
1442 .compat_ioctl = compat_ptr_ioctl,
1443 };
1444
vhost_vdpa_release_dev(struct device * device)1445 static void vhost_vdpa_release_dev(struct device *device)
1446 {
1447 struct vhost_vdpa *v =
1448 container_of(device, struct vhost_vdpa, dev);
1449
1450 ida_simple_remove(&vhost_vdpa_ida, v->minor);
1451 kfree(v->vqs);
1452 kfree(v);
1453 }
1454
vhost_vdpa_probe(struct vdpa_device * vdpa)1455 static int vhost_vdpa_probe(struct vdpa_device *vdpa)
1456 {
1457 const struct vdpa_config_ops *ops = vdpa->config;
1458 struct vhost_vdpa *v;
1459 int minor;
1460 int i, r;
1461
1462 /* We can't support platform IOMMU device with more than 1
1463 * group or as
1464 */
1465 if (!ops->set_map && !ops->dma_map &&
1466 (vdpa->ngroups > 1 || vdpa->nas > 1))
1467 return -EOPNOTSUPP;
1468
1469 v = kzalloc(sizeof(*v), GFP_KERNEL | __GFP_RETRY_MAYFAIL);
1470 if (!v)
1471 return -ENOMEM;
1472
1473 minor = ida_simple_get(&vhost_vdpa_ida, 0,
1474 VHOST_VDPA_DEV_MAX, GFP_KERNEL);
1475 if (minor < 0) {
1476 kfree(v);
1477 return minor;
1478 }
1479
1480 atomic_set(&v->opened, 0);
1481 v->minor = minor;
1482 v->vdpa = vdpa;
1483 v->nvqs = vdpa->nvqs;
1484 v->virtio_id = ops->get_device_id(vdpa);
1485
1486 device_initialize(&v->dev);
1487 v->dev.release = vhost_vdpa_release_dev;
1488 v->dev.parent = &vdpa->dev;
1489 v->dev.devt = MKDEV(MAJOR(vhost_vdpa_major), minor);
1490 v->vqs = kmalloc_array(v->nvqs, sizeof(struct vhost_virtqueue),
1491 GFP_KERNEL);
1492 if (!v->vqs) {
1493 r = -ENOMEM;
1494 goto err;
1495 }
1496
1497 r = dev_set_name(&v->dev, "vhost-vdpa-%u", minor);
1498 if (r)
1499 goto err;
1500
1501 cdev_init(&v->cdev, &vhost_vdpa_fops);
1502 v->cdev.owner = THIS_MODULE;
1503
1504 r = cdev_device_add(&v->cdev, &v->dev);
1505 if (r)
1506 goto err;
1507
1508 init_completion(&v->completion);
1509 vdpa_set_drvdata(vdpa, v);
1510
1511 for (i = 0; i < VHOST_VDPA_IOTLB_BUCKETS; i++)
1512 INIT_HLIST_HEAD(&v->as[i]);
1513
1514 return 0;
1515
1516 err:
1517 put_device(&v->dev);
1518 return r;
1519 }
1520
vhost_vdpa_remove(struct vdpa_device * vdpa)1521 static void vhost_vdpa_remove(struct vdpa_device *vdpa)
1522 {
1523 struct vhost_vdpa *v = vdpa_get_drvdata(vdpa);
1524 int opened;
1525
1526 cdev_device_del(&v->cdev, &v->dev);
1527
1528 do {
1529 opened = atomic_cmpxchg(&v->opened, 0, 1);
1530 if (!opened)
1531 break;
1532 wait_for_completion(&v->completion);
1533 } while (1);
1534
1535 put_device(&v->dev);
1536 }
1537
1538 static struct vdpa_driver vhost_vdpa_driver = {
1539 .driver = {
1540 .name = "vhost_vdpa",
1541 },
1542 .probe = vhost_vdpa_probe,
1543 .remove = vhost_vdpa_remove,
1544 };
1545
vhost_vdpa_init(void)1546 static int __init vhost_vdpa_init(void)
1547 {
1548 int r;
1549
1550 r = alloc_chrdev_region(&vhost_vdpa_major, 0, VHOST_VDPA_DEV_MAX,
1551 "vhost-vdpa");
1552 if (r)
1553 goto err_alloc_chrdev;
1554
1555 r = vdpa_register_driver(&vhost_vdpa_driver);
1556 if (r)
1557 goto err_vdpa_register_driver;
1558
1559 return 0;
1560
1561 err_vdpa_register_driver:
1562 unregister_chrdev_region(vhost_vdpa_major, VHOST_VDPA_DEV_MAX);
1563 err_alloc_chrdev:
1564 return r;
1565 }
1566 module_init(vhost_vdpa_init);
1567
vhost_vdpa_exit(void)1568 static void __exit vhost_vdpa_exit(void)
1569 {
1570 vdpa_unregister_driver(&vhost_vdpa_driver);
1571 unregister_chrdev_region(vhost_vdpa_major, VHOST_VDPA_DEV_MAX);
1572 }
1573 module_exit(vhost_vdpa_exit);
1574
1575 MODULE_VERSION("0.0.1");
1576 MODULE_LICENSE("GPL v2");
1577 MODULE_AUTHOR("Intel Corporation");
1578 MODULE_DESCRIPTION("vDPA-based vhost backend for virtio");
1579