xref: /openbmc/linux/drivers/vhost/vdpa.c (revision 2fa5ebe3) 
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 
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 
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 
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 
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 
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 
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 
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 
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 
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 
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 	irq_bypass_unregister_producer(&vq->call_ctx.producer);
195 	if (!vq->call_ctx.ctx)
196 		return;
197 
198 	vq->call_ctx.producer.token = vq->call_ctx.ctx;
199 	vq->call_ctx.producer.irq = irq;
200 	ret = irq_bypass_register_producer(&vq->call_ctx.producer);
201 	if (unlikely(ret))
202 		dev_info(&v->dev, "vq %u, irq bypass producer (token %p) registration fails, ret =  %d\n",
203 			 qid, vq->call_ctx.producer.token, ret);
204 }
205 
206 static void vhost_vdpa_unsetup_vq_irq(struct vhost_vdpa *v, u16 qid)
207 {
208 	struct vhost_virtqueue *vq = &v->vqs[qid];
209 
210 	irq_bypass_unregister_producer(&vq->call_ctx.producer);
211 }
212 
213 static int vhost_vdpa_reset(struct vhost_vdpa *v)
214 {
215 	struct vdpa_device *vdpa = v->vdpa;
216 
217 	v->in_batch = 0;
218 
219 	return vdpa_reset(vdpa);
220 }
221 
222 static long vhost_vdpa_get_device_id(struct vhost_vdpa *v, u8 __user *argp)
223 {
224 	struct vdpa_device *vdpa = v->vdpa;
225 	const struct vdpa_config_ops *ops = vdpa->config;
226 	u32 device_id;
227 
228 	device_id = ops->get_device_id(vdpa);
229 
230 	if (copy_to_user(argp, &device_id, sizeof(device_id)))
231 		return -EFAULT;
232 
233 	return 0;
234 }
235 
236 static long vhost_vdpa_get_status(struct vhost_vdpa *v, u8 __user *statusp)
237 {
238 	struct vdpa_device *vdpa = v->vdpa;
239 	const struct vdpa_config_ops *ops = vdpa->config;
240 	u8 status;
241 
242 	status = ops->get_status(vdpa);
243 
244 	if (copy_to_user(statusp, &status, sizeof(status)))
245 		return -EFAULT;
246 
247 	return 0;
248 }
249 
250 static long vhost_vdpa_set_status(struct vhost_vdpa *v, u8 __user *statusp)
251 {
252 	struct vdpa_device *vdpa = v->vdpa;
253 	const struct vdpa_config_ops *ops = vdpa->config;
254 	u8 status, status_old;
255 	u32 nvqs = v->nvqs;
256 	int ret;
257 	u16 i;
258 
259 	if (copy_from_user(&status, statusp, sizeof(status)))
260 		return -EFAULT;
261 
262 	status_old = ops->get_status(vdpa);
263 
264 	/*
265 	 * Userspace shouldn't remove status bits unless reset the
266 	 * status to 0.
267 	 */
268 	if (status != 0 && (status_old & ~status) != 0)
269 		return -EINVAL;
270 
271 	if ((status_old & VIRTIO_CONFIG_S_DRIVER_OK) && !(status & VIRTIO_CONFIG_S_DRIVER_OK))
272 		for (i = 0; i < nvqs; i++)
273 			vhost_vdpa_unsetup_vq_irq(v, i);
274 
275 	if (status == 0) {
276 		ret = vdpa_reset(vdpa);
277 		if (ret)
278 			return ret;
279 	} else
280 		vdpa_set_status(vdpa, status);
281 
282 	if ((status & VIRTIO_CONFIG_S_DRIVER_OK) && !(status_old & VIRTIO_CONFIG_S_DRIVER_OK))
283 		for (i = 0; i < nvqs; i++)
284 			vhost_vdpa_setup_vq_irq(v, i);
285 
286 	return 0;
287 }
288 
289 static int vhost_vdpa_config_validate(struct vhost_vdpa *v,
290 				      struct vhost_vdpa_config *c)
291 {
292 	struct vdpa_device *vdpa = v->vdpa;
293 	size_t size = vdpa->config->get_config_size(vdpa);
294 
295 	if (c->len == 0 || c->off > size)
296 		return -EINVAL;
297 
298 	if (c->len > size - c->off)
299 		return -E2BIG;
300 
301 	return 0;
302 }
303 
304 static long vhost_vdpa_get_config(struct vhost_vdpa *v,
305 				  struct vhost_vdpa_config __user *c)
306 {
307 	struct vdpa_device *vdpa = v->vdpa;
308 	struct vhost_vdpa_config config;
309 	unsigned long size = offsetof(struct vhost_vdpa_config, buf);
310 	u8 *buf;
311 
312 	if (copy_from_user(&config, c, size))
313 		return -EFAULT;
314 	if (vhost_vdpa_config_validate(v, &config))
315 		return -EINVAL;
316 	buf = kvzalloc(config.len, GFP_KERNEL);
317 	if (!buf)
318 		return -ENOMEM;
319 
320 	vdpa_get_config(vdpa, config.off, buf, config.len);
321 
322 	if (copy_to_user(c->buf, buf, config.len)) {
323 		kvfree(buf);
324 		return -EFAULT;
325 	}
326 
327 	kvfree(buf);
328 	return 0;
329 }
330 
331 static long vhost_vdpa_set_config(struct vhost_vdpa *v,
332 				  struct vhost_vdpa_config __user *c)
333 {
334 	struct vdpa_device *vdpa = v->vdpa;
335 	struct vhost_vdpa_config config;
336 	unsigned long size = offsetof(struct vhost_vdpa_config, buf);
337 	u8 *buf;
338 
339 	if (copy_from_user(&config, c, size))
340 		return -EFAULT;
341 	if (vhost_vdpa_config_validate(v, &config))
342 		return -EINVAL;
343 
344 	buf = vmemdup_user(c->buf, config.len);
345 	if (IS_ERR(buf))
346 		return PTR_ERR(buf);
347 
348 	vdpa_set_config(vdpa, config.off, buf, config.len);
349 
350 	kvfree(buf);
351 	return 0;
352 }
353 
354 static bool vhost_vdpa_can_suspend(const struct vhost_vdpa *v)
355 {
356 	struct vdpa_device *vdpa = v->vdpa;
357 	const struct vdpa_config_ops *ops = vdpa->config;
358 
359 	return ops->suspend;
360 }
361 
362 static bool vhost_vdpa_can_resume(const struct vhost_vdpa *v)
363 {
364 	struct vdpa_device *vdpa = v->vdpa;
365 	const struct vdpa_config_ops *ops = vdpa->config;
366 
367 	return ops->resume;
368 }
369 
370 static long vhost_vdpa_get_features(struct vhost_vdpa *v, u64 __user *featurep)
371 {
372 	struct vdpa_device *vdpa = v->vdpa;
373 	const struct vdpa_config_ops *ops = vdpa->config;
374 	u64 features;
375 
376 	features = ops->get_device_features(vdpa);
377 
378 	if (copy_to_user(featurep, &features, sizeof(features)))
379 		return -EFAULT;
380 
381 	return 0;
382 }
383 
384 static long vhost_vdpa_set_features(struct vhost_vdpa *v, u64 __user *featurep)
385 {
386 	struct vdpa_device *vdpa = v->vdpa;
387 	const struct vdpa_config_ops *ops = vdpa->config;
388 	u64 features;
389 
390 	/*
391 	 * It's not allowed to change the features after they have
392 	 * been negotiated.
393 	 */
394 	if (ops->get_status(vdpa) & VIRTIO_CONFIG_S_FEATURES_OK)
395 		return -EBUSY;
396 
397 	if (copy_from_user(&features, featurep, sizeof(features)))
398 		return -EFAULT;
399 
400 	if (vdpa_set_features(vdpa, features))
401 		return -EINVAL;
402 
403 	return 0;
404 }
405 
406 static long vhost_vdpa_get_vring_num(struct vhost_vdpa *v, u16 __user *argp)
407 {
408 	struct vdpa_device *vdpa = v->vdpa;
409 	const struct vdpa_config_ops *ops = vdpa->config;
410 	u16 num;
411 
412 	num = ops->get_vq_num_max(vdpa);
413 
414 	if (copy_to_user(argp, &num, sizeof(num)))
415 		return -EFAULT;
416 
417 	return 0;
418 }
419 
420 static void vhost_vdpa_config_put(struct vhost_vdpa *v)
421 {
422 	if (v->config_ctx) {
423 		eventfd_ctx_put(v->config_ctx);
424 		v->config_ctx = NULL;
425 	}
426 }
427 
428 static long vhost_vdpa_set_config_call(struct vhost_vdpa *v, u32 __user *argp)
429 {
430 	struct vdpa_callback cb;
431 	int fd;
432 	struct eventfd_ctx *ctx;
433 
434 	cb.callback = vhost_vdpa_config_cb;
435 	cb.private = v;
436 	if (copy_from_user(&fd, argp, sizeof(fd)))
437 		return  -EFAULT;
438 
439 	ctx = fd == VHOST_FILE_UNBIND ? NULL : eventfd_ctx_fdget(fd);
440 	swap(ctx, v->config_ctx);
441 
442 	if (!IS_ERR_OR_NULL(ctx))
443 		eventfd_ctx_put(ctx);
444 
445 	if (IS_ERR(v->config_ctx)) {
446 		long ret = PTR_ERR(v->config_ctx);
447 
448 		v->config_ctx = NULL;
449 		return ret;
450 	}
451 
452 	v->vdpa->config->set_config_cb(v->vdpa, &cb);
453 
454 	return 0;
455 }
456 
457 static long vhost_vdpa_get_iova_range(struct vhost_vdpa *v, u32 __user *argp)
458 {
459 	struct vhost_vdpa_iova_range range = {
460 		.first = v->range.first,
461 		.last = v->range.last,
462 	};
463 
464 	if (copy_to_user(argp, &range, sizeof(range)))
465 		return -EFAULT;
466 	return 0;
467 }
468 
469 static long vhost_vdpa_get_config_size(struct vhost_vdpa *v, u32 __user *argp)
470 {
471 	struct vdpa_device *vdpa = v->vdpa;
472 	const struct vdpa_config_ops *ops = vdpa->config;
473 	u32 size;
474 
475 	size = ops->get_config_size(vdpa);
476 
477 	if (copy_to_user(argp, &size, sizeof(size)))
478 		return -EFAULT;
479 
480 	return 0;
481 }
482 
483 static long vhost_vdpa_get_vqs_count(struct vhost_vdpa *v, u32 __user *argp)
484 {
485 	struct vdpa_device *vdpa = v->vdpa;
486 
487 	if (copy_to_user(argp, &vdpa->nvqs, sizeof(vdpa->nvqs)))
488 		return -EFAULT;
489 
490 	return 0;
491 }
492 
493 /* After a successful return of ioctl the device must not process more
494  * virtqueue descriptors. The device can answer to read or writes of config
495  * fields as if it were not suspended. In particular, writing to "queue_enable"
496  * with a value of 1 will not make the device start processing buffers.
497  */
498 static long vhost_vdpa_suspend(struct vhost_vdpa *v)
499 {
500 	struct vdpa_device *vdpa = v->vdpa;
501 	const struct vdpa_config_ops *ops = vdpa->config;
502 
503 	if (!ops->suspend)
504 		return -EOPNOTSUPP;
505 
506 	return ops->suspend(vdpa);
507 }
508 
509 /* After a successful return of this ioctl the device resumes processing
510  * virtqueue descriptors. The device becomes fully operational the same way it
511  * was before it was suspended.
512  */
513 static long vhost_vdpa_resume(struct vhost_vdpa *v)
514 {
515 	struct vdpa_device *vdpa = v->vdpa;
516 	const struct vdpa_config_ops *ops = vdpa->config;
517 
518 	if (!ops->resume)
519 		return -EOPNOTSUPP;
520 
521 	return ops->resume(vdpa);
522 }
523 
524 static long vhost_vdpa_vring_ioctl(struct vhost_vdpa *v, unsigned int cmd,
525 				   void __user *argp)
526 {
527 	struct vdpa_device *vdpa = v->vdpa;
528 	const struct vdpa_config_ops *ops = vdpa->config;
529 	struct vdpa_vq_state vq_state;
530 	struct vdpa_callback cb;
531 	struct vhost_virtqueue *vq;
532 	struct vhost_vring_state s;
533 	u32 idx;
534 	long r;
535 
536 	r = get_user(idx, (u32 __user *)argp);
537 	if (r < 0)
538 		return r;
539 
540 	if (idx >= v->nvqs)
541 		return -ENOBUFS;
542 
543 	idx = array_index_nospec(idx, v->nvqs);
544 	vq = &v->vqs[idx];
545 
546 	switch (cmd) {
547 	case VHOST_VDPA_SET_VRING_ENABLE:
548 		if (copy_from_user(&s, argp, sizeof(s)))
549 			return -EFAULT;
550 		ops->set_vq_ready(vdpa, idx, s.num);
551 		return 0;
552 	case VHOST_VDPA_GET_VRING_GROUP:
553 		if (!ops->get_vq_group)
554 			return -EOPNOTSUPP;
555 		s.index = idx;
556 		s.num = ops->get_vq_group(vdpa, idx);
557 		if (s.num >= vdpa->ngroups)
558 			return -EIO;
559 		else if (copy_to_user(argp, &s, sizeof(s)))
560 			return -EFAULT;
561 		return 0;
562 	case VHOST_VDPA_SET_GROUP_ASID:
563 		if (copy_from_user(&s, argp, sizeof(s)))
564 			return -EFAULT;
565 		if (s.num >= vdpa->nas)
566 			return -EINVAL;
567 		if (!ops->set_group_asid)
568 			return -EOPNOTSUPP;
569 		return ops->set_group_asid(vdpa, idx, s.num);
570 	case VHOST_GET_VRING_BASE:
571 		r = ops->get_vq_state(v->vdpa, idx, &vq_state);
572 		if (r)
573 			return r;
574 
575 		vq->last_avail_idx = vq_state.split.avail_index;
576 		break;
577 	}
578 
579 	r = vhost_vring_ioctl(&v->vdev, cmd, argp);
580 	if (r)
581 		return r;
582 
583 	switch (cmd) {
584 	case VHOST_SET_VRING_ADDR:
585 		if (ops->set_vq_address(vdpa, idx,
586 					(u64)(uintptr_t)vq->desc,
587 					(u64)(uintptr_t)vq->avail,
588 					(u64)(uintptr_t)vq->used))
589 			r = -EINVAL;
590 		break;
591 
592 	case VHOST_SET_VRING_BASE:
593 		vq_state.split.avail_index = vq->last_avail_idx;
594 		if (ops->set_vq_state(vdpa, idx, &vq_state))
595 			r = -EINVAL;
596 		break;
597 
598 	case VHOST_SET_VRING_CALL:
599 		if (vq->call_ctx.ctx) {
600 			cb.callback = vhost_vdpa_virtqueue_cb;
601 			cb.private = vq;
602 		} else {
603 			cb.callback = NULL;
604 			cb.private = NULL;
605 		}
606 		ops->set_vq_cb(vdpa, idx, &cb);
607 		vhost_vdpa_setup_vq_irq(v, idx);
608 		break;
609 
610 	case VHOST_SET_VRING_NUM:
611 		ops->set_vq_num(vdpa, idx, vq->num);
612 		break;
613 	}
614 
615 	return r;
616 }
617 
618 static long vhost_vdpa_unlocked_ioctl(struct file *filep,
619 				      unsigned int cmd, unsigned long arg)
620 {
621 	struct vhost_vdpa *v = filep->private_data;
622 	struct vhost_dev *d = &v->vdev;
623 	void __user *argp = (void __user *)arg;
624 	u64 __user *featurep = argp;
625 	u64 features;
626 	long r = 0;
627 
628 	if (cmd == VHOST_SET_BACKEND_FEATURES) {
629 		if (copy_from_user(&features, featurep, sizeof(features)))
630 			return -EFAULT;
631 		if (features & ~(VHOST_VDPA_BACKEND_FEATURES |
632 				 BIT_ULL(VHOST_BACKEND_F_SUSPEND) |
633 				 BIT_ULL(VHOST_BACKEND_F_RESUME)))
634 			return -EOPNOTSUPP;
635 		if ((features & BIT_ULL(VHOST_BACKEND_F_SUSPEND)) &&
636 		     !vhost_vdpa_can_suspend(v))
637 			return -EOPNOTSUPP;
638 		if ((features & BIT_ULL(VHOST_BACKEND_F_RESUME)) &&
639 		     !vhost_vdpa_can_resume(v))
640 			return -EOPNOTSUPP;
641 		vhost_set_backend_features(&v->vdev, features);
642 		return 0;
643 	}
644 
645 	mutex_lock(&d->mutex);
646 
647 	switch (cmd) {
648 	case VHOST_VDPA_GET_DEVICE_ID:
649 		r = vhost_vdpa_get_device_id(v, argp);
650 		break;
651 	case VHOST_VDPA_GET_STATUS:
652 		r = vhost_vdpa_get_status(v, argp);
653 		break;
654 	case VHOST_VDPA_SET_STATUS:
655 		r = vhost_vdpa_set_status(v, argp);
656 		break;
657 	case VHOST_VDPA_GET_CONFIG:
658 		r = vhost_vdpa_get_config(v, argp);
659 		break;
660 	case VHOST_VDPA_SET_CONFIG:
661 		r = vhost_vdpa_set_config(v, argp);
662 		break;
663 	case VHOST_GET_FEATURES:
664 		r = vhost_vdpa_get_features(v, argp);
665 		break;
666 	case VHOST_SET_FEATURES:
667 		r = vhost_vdpa_set_features(v, argp);
668 		break;
669 	case VHOST_VDPA_GET_VRING_NUM:
670 		r = vhost_vdpa_get_vring_num(v, argp);
671 		break;
672 	case VHOST_VDPA_GET_GROUP_NUM:
673 		if (copy_to_user(argp, &v->vdpa->ngroups,
674 				 sizeof(v->vdpa->ngroups)))
675 			r = -EFAULT;
676 		break;
677 	case VHOST_VDPA_GET_AS_NUM:
678 		if (copy_to_user(argp, &v->vdpa->nas, sizeof(v->vdpa->nas)))
679 			r = -EFAULT;
680 		break;
681 	case VHOST_SET_LOG_BASE:
682 	case VHOST_SET_LOG_FD:
683 		r = -ENOIOCTLCMD;
684 		break;
685 	case VHOST_VDPA_SET_CONFIG_CALL:
686 		r = vhost_vdpa_set_config_call(v, argp);
687 		break;
688 	case VHOST_GET_BACKEND_FEATURES:
689 		features = VHOST_VDPA_BACKEND_FEATURES;
690 		if (vhost_vdpa_can_suspend(v))
691 			features |= BIT_ULL(VHOST_BACKEND_F_SUSPEND);
692 		if (vhost_vdpa_can_resume(v))
693 			features |= BIT_ULL(VHOST_BACKEND_F_RESUME);
694 		if (copy_to_user(featurep, &features, sizeof(features)))
695 			r = -EFAULT;
696 		break;
697 	case VHOST_VDPA_GET_IOVA_RANGE:
698 		r = vhost_vdpa_get_iova_range(v, argp);
699 		break;
700 	case VHOST_VDPA_GET_CONFIG_SIZE:
701 		r = vhost_vdpa_get_config_size(v, argp);
702 		break;
703 	case VHOST_VDPA_GET_VQS_COUNT:
704 		r = vhost_vdpa_get_vqs_count(v, argp);
705 		break;
706 	case VHOST_VDPA_SUSPEND:
707 		r = vhost_vdpa_suspend(v);
708 		break;
709 	case VHOST_VDPA_RESUME:
710 		r = vhost_vdpa_resume(v);
711 		break;
712 	default:
713 		r = vhost_dev_ioctl(&v->vdev, cmd, argp);
714 		if (r == -ENOIOCTLCMD)
715 			r = vhost_vdpa_vring_ioctl(v, cmd, argp);
716 		break;
717 	}
718 
719 	mutex_unlock(&d->mutex);
720 	return r;
721 }
722 static void vhost_vdpa_general_unmap(struct vhost_vdpa *v,
723 				     struct vhost_iotlb_map *map, u32 asid)
724 {
725 	struct vdpa_device *vdpa = v->vdpa;
726 	const struct vdpa_config_ops *ops = vdpa->config;
727 	if (ops->dma_map) {
728 		ops->dma_unmap(vdpa, asid, map->start, map->size);
729 	} else if (ops->set_map == NULL) {
730 		iommu_unmap(v->domain, map->start, map->size);
731 	}
732 }
733 
734 static void vhost_vdpa_pa_unmap(struct vhost_vdpa *v, struct vhost_iotlb *iotlb,
735 				u64 start, u64 last, u32 asid)
736 {
737 	struct vhost_dev *dev = &v->vdev;
738 	struct vhost_iotlb_map *map;
739 	struct page *page;
740 	unsigned long pfn, pinned;
741 
742 	while ((map = vhost_iotlb_itree_first(iotlb, start, last)) != NULL) {
743 		pinned = PFN_DOWN(map->size);
744 		for (pfn = PFN_DOWN(map->addr);
745 		     pinned > 0; pfn++, pinned--) {
746 			page = pfn_to_page(pfn);
747 			if (map->perm & VHOST_ACCESS_WO)
748 				set_page_dirty_lock(page);
749 			unpin_user_page(page);
750 		}
751 		atomic64_sub(PFN_DOWN(map->size), &dev->mm->pinned_vm);
752 		vhost_vdpa_general_unmap(v, map, asid);
753 		vhost_iotlb_map_free(iotlb, map);
754 	}
755 }
756 
757 static void vhost_vdpa_va_unmap(struct vhost_vdpa *v, struct vhost_iotlb *iotlb,
758 				u64 start, u64 last, u32 asid)
759 {
760 	struct vhost_iotlb_map *map;
761 	struct vdpa_map_file *map_file;
762 
763 	while ((map = vhost_iotlb_itree_first(iotlb, start, last)) != NULL) {
764 		map_file = (struct vdpa_map_file *)map->opaque;
765 		fput(map_file->file);
766 		kfree(map_file);
767 		vhost_vdpa_general_unmap(v, map, asid);
768 		vhost_iotlb_map_free(iotlb, map);
769 	}
770 }
771 
772 static void vhost_vdpa_iotlb_unmap(struct vhost_vdpa *v,
773 				   struct vhost_iotlb *iotlb, u64 start,
774 				   u64 last, u32 asid)
775 {
776 	struct vdpa_device *vdpa = v->vdpa;
777 
778 	if (vdpa->use_va)
779 		return vhost_vdpa_va_unmap(v, iotlb, start, last, asid);
780 
781 	return vhost_vdpa_pa_unmap(v, iotlb, start, last, asid);
782 }
783 
784 static int perm_to_iommu_flags(u32 perm)
785 {
786 	int flags = 0;
787 
788 	switch (perm) {
789 	case VHOST_ACCESS_WO:
790 		flags |= IOMMU_WRITE;
791 		break;
792 	case VHOST_ACCESS_RO:
793 		flags |= IOMMU_READ;
794 		break;
795 	case VHOST_ACCESS_RW:
796 		flags |= (IOMMU_WRITE | IOMMU_READ);
797 		break;
798 	default:
799 		WARN(1, "invalidate vhost IOTLB permission\n");
800 		break;
801 	}
802 
803 	return flags | IOMMU_CACHE;
804 }
805 
806 static int vhost_vdpa_map(struct vhost_vdpa *v, struct vhost_iotlb *iotlb,
807 			  u64 iova, u64 size, u64 pa, u32 perm, void *opaque)
808 {
809 	struct vhost_dev *dev = &v->vdev;
810 	struct vdpa_device *vdpa = v->vdpa;
811 	const struct vdpa_config_ops *ops = vdpa->config;
812 	u32 asid = iotlb_to_asid(iotlb);
813 	int r = 0;
814 
815 	r = vhost_iotlb_add_range_ctx(iotlb, iova, iova + size - 1,
816 				      pa, perm, opaque);
817 	if (r)
818 		return r;
819 
820 	if (ops->dma_map) {
821 		r = ops->dma_map(vdpa, asid, iova, size, pa, perm, opaque);
822 	} else if (ops->set_map) {
823 		if (!v->in_batch)
824 			r = ops->set_map(vdpa, asid, iotlb);
825 	} else {
826 		r = iommu_map(v->domain, iova, pa, size,
827 			      perm_to_iommu_flags(perm), GFP_KERNEL);
828 	}
829 	if (r) {
830 		vhost_iotlb_del_range(iotlb, iova, iova + size - 1);
831 		return r;
832 	}
833 
834 	if (!vdpa->use_va)
835 		atomic64_add(PFN_DOWN(size), &dev->mm->pinned_vm);
836 
837 	return 0;
838 }
839 
840 static void vhost_vdpa_unmap(struct vhost_vdpa *v,
841 			     struct vhost_iotlb *iotlb,
842 			     u64 iova, u64 size)
843 {
844 	struct vdpa_device *vdpa = v->vdpa;
845 	const struct vdpa_config_ops *ops = vdpa->config;
846 	u32 asid = iotlb_to_asid(iotlb);
847 
848 	vhost_vdpa_iotlb_unmap(v, iotlb, iova, iova + size - 1, asid);
849 
850 	if (ops->set_map) {
851 		if (!v->in_batch)
852 			ops->set_map(vdpa, asid, iotlb);
853 	}
854 	/* If we are in the middle of batch processing, delay the free
855 	 * of AS until BATCH_END.
856 	 */
857 	if (!v->in_batch && !iotlb->nmaps)
858 		vhost_vdpa_remove_as(v, asid);
859 }
860 
861 static int vhost_vdpa_va_map(struct vhost_vdpa *v,
862 			     struct vhost_iotlb *iotlb,
863 			     u64 iova, u64 size, u64 uaddr, u32 perm)
864 {
865 	struct vhost_dev *dev = &v->vdev;
866 	u64 offset, map_size, map_iova = iova;
867 	struct vdpa_map_file *map_file;
868 	struct vm_area_struct *vma;
869 	int ret = 0;
870 
871 	mmap_read_lock(dev->mm);
872 
873 	while (size) {
874 		vma = find_vma(dev->mm, uaddr);
875 		if (!vma) {
876 			ret = -EINVAL;
877 			break;
878 		}
879 		map_size = min(size, vma->vm_end - uaddr);
880 		if (!(vma->vm_file && (vma->vm_flags & VM_SHARED) &&
881 			!(vma->vm_flags & (VM_IO | VM_PFNMAP))))
882 			goto next;
883 
884 		map_file = kzalloc(sizeof(*map_file), GFP_KERNEL);
885 		if (!map_file) {
886 			ret = -ENOMEM;
887 			break;
888 		}
889 		offset = (vma->vm_pgoff << PAGE_SHIFT) + uaddr - vma->vm_start;
890 		map_file->offset = offset;
891 		map_file->file = get_file(vma->vm_file);
892 		ret = vhost_vdpa_map(v, iotlb, map_iova, map_size, uaddr,
893 				     perm, map_file);
894 		if (ret) {
895 			fput(map_file->file);
896 			kfree(map_file);
897 			break;
898 		}
899 next:
900 		size -= map_size;
901 		uaddr += map_size;
902 		map_iova += map_size;
903 	}
904 	if (ret)
905 		vhost_vdpa_unmap(v, iotlb, iova, map_iova - iova);
906 
907 	mmap_read_unlock(dev->mm);
908 
909 	return ret;
910 }
911 
912 static int vhost_vdpa_pa_map(struct vhost_vdpa *v,
913 			     struct vhost_iotlb *iotlb,
914 			     u64 iova, u64 size, u64 uaddr, u32 perm)
915 {
916 	struct vhost_dev *dev = &v->vdev;
917 	struct page **page_list;
918 	unsigned long list_size = PAGE_SIZE / sizeof(struct page *);
919 	unsigned int gup_flags = FOLL_LONGTERM;
920 	unsigned long npages, cur_base, map_pfn, last_pfn = 0;
921 	unsigned long lock_limit, sz2pin, nchunks, i;
922 	u64 start = iova;
923 	long pinned;
924 	int ret = 0;
925 
926 	/* Limit the use of memory for bookkeeping */
927 	page_list = (struct page **) __get_free_page(GFP_KERNEL);
928 	if (!page_list)
929 		return -ENOMEM;
930 
931 	if (perm & VHOST_ACCESS_WO)
932 		gup_flags |= FOLL_WRITE;
933 
934 	npages = PFN_UP(size + (iova & ~PAGE_MASK));
935 	if (!npages) {
936 		ret = -EINVAL;
937 		goto free;
938 	}
939 
940 	mmap_read_lock(dev->mm);
941 
942 	lock_limit = PFN_DOWN(rlimit(RLIMIT_MEMLOCK));
943 	if (npages + atomic64_read(&dev->mm->pinned_vm) > lock_limit) {
944 		ret = -ENOMEM;
945 		goto unlock;
946 	}
947 
948 	cur_base = uaddr & PAGE_MASK;
949 	iova &= PAGE_MASK;
950 	nchunks = 0;
951 
952 	while (npages) {
953 		sz2pin = min_t(unsigned long, npages, list_size);
954 		pinned = pin_user_pages(cur_base, sz2pin,
955 					gup_flags, page_list, NULL);
956 		if (sz2pin != pinned) {
957 			if (pinned < 0) {
958 				ret = pinned;
959 			} else {
960 				unpin_user_pages(page_list, pinned);
961 				ret = -ENOMEM;
962 			}
963 			goto out;
964 		}
965 		nchunks++;
966 
967 		if (!last_pfn)
968 			map_pfn = page_to_pfn(page_list[0]);
969 
970 		for (i = 0; i < pinned; i++) {
971 			unsigned long this_pfn = page_to_pfn(page_list[i]);
972 			u64 csize;
973 
974 			if (last_pfn && (this_pfn != last_pfn + 1)) {
975 				/* Pin a contiguous chunk of memory */
976 				csize = PFN_PHYS(last_pfn - map_pfn + 1);
977 				ret = vhost_vdpa_map(v, iotlb, iova, csize,
978 						     PFN_PHYS(map_pfn),
979 						     perm, NULL);
980 				if (ret) {
981 					/*
982 					 * Unpin the pages that are left unmapped
983 					 * from this point on in the current
984 					 * page_list. The remaining outstanding
985 					 * ones which may stride across several
986 					 * chunks will be covered in the common
987 					 * error path subsequently.
988 					 */
989 					unpin_user_pages(&page_list[i],
990 							 pinned - i);
991 					goto out;
992 				}
993 
994 				map_pfn = this_pfn;
995 				iova += csize;
996 				nchunks = 0;
997 			}
998 
999 			last_pfn = this_pfn;
1000 		}
1001 
1002 		cur_base += PFN_PHYS(pinned);
1003 		npages -= pinned;
1004 	}
1005 
1006 	/* Pin the rest chunk */
1007 	ret = vhost_vdpa_map(v, iotlb, iova, PFN_PHYS(last_pfn - map_pfn + 1),
1008 			     PFN_PHYS(map_pfn), perm, NULL);
1009 out:
1010 	if (ret) {
1011 		if (nchunks) {
1012 			unsigned long pfn;
1013 
1014 			/*
1015 			 * Unpin the outstanding pages which are yet to be
1016 			 * mapped but haven't due to vdpa_map() or
1017 			 * pin_user_pages() failure.
1018 			 *
1019 			 * Mapped pages are accounted in vdpa_map(), hence
1020 			 * the corresponding unpinning will be handled by
1021 			 * vdpa_unmap().
1022 			 */
1023 			WARN_ON(!last_pfn);
1024 			for (pfn = map_pfn; pfn <= last_pfn; pfn++)
1025 				unpin_user_page(pfn_to_page(pfn));
1026 		}
1027 		vhost_vdpa_unmap(v, iotlb, start, size);
1028 	}
1029 unlock:
1030 	mmap_read_unlock(dev->mm);
1031 free:
1032 	free_page((unsigned long)page_list);
1033 	return ret;
1034 
1035 }
1036 
1037 static int vhost_vdpa_process_iotlb_update(struct vhost_vdpa *v,
1038 					   struct vhost_iotlb *iotlb,
1039 					   struct vhost_iotlb_msg *msg)
1040 {
1041 	struct vdpa_device *vdpa = v->vdpa;
1042 
1043 	if (msg->iova < v->range.first || !msg->size ||
1044 	    msg->iova > U64_MAX - msg->size + 1 ||
1045 	    msg->iova + msg->size - 1 > v->range.last)
1046 		return -EINVAL;
1047 
1048 	if (vhost_iotlb_itree_first(iotlb, msg->iova,
1049 				    msg->iova + msg->size - 1))
1050 		return -EEXIST;
1051 
1052 	if (vdpa->use_va)
1053 		return vhost_vdpa_va_map(v, iotlb, msg->iova, msg->size,
1054 					 msg->uaddr, msg->perm);
1055 
1056 	return vhost_vdpa_pa_map(v, iotlb, msg->iova, msg->size, msg->uaddr,
1057 				 msg->perm);
1058 }
1059 
1060 static int vhost_vdpa_process_iotlb_msg(struct vhost_dev *dev, u32 asid,
1061 					struct vhost_iotlb_msg *msg)
1062 {
1063 	struct vhost_vdpa *v = container_of(dev, struct vhost_vdpa, vdev);
1064 	struct vdpa_device *vdpa = v->vdpa;
1065 	const struct vdpa_config_ops *ops = vdpa->config;
1066 	struct vhost_iotlb *iotlb = NULL;
1067 	struct vhost_vdpa_as *as = NULL;
1068 	int r = 0;
1069 
1070 	mutex_lock(&dev->mutex);
1071 
1072 	r = vhost_dev_check_owner(dev);
1073 	if (r)
1074 		goto unlock;
1075 
1076 	if (msg->type == VHOST_IOTLB_UPDATE ||
1077 	    msg->type == VHOST_IOTLB_BATCH_BEGIN) {
1078 		as = vhost_vdpa_find_alloc_as(v, asid);
1079 		if (!as) {
1080 			dev_err(&v->dev, "can't find and alloc asid %d\n",
1081 				asid);
1082 			r = -EINVAL;
1083 			goto unlock;
1084 		}
1085 		iotlb = &as->iotlb;
1086 	} else
1087 		iotlb = asid_to_iotlb(v, asid);
1088 
1089 	if ((v->in_batch && v->batch_asid != asid) || !iotlb) {
1090 		if (v->in_batch && v->batch_asid != asid) {
1091 			dev_info(&v->dev, "batch id %d asid %d\n",
1092 				 v->batch_asid, asid);
1093 		}
1094 		if (!iotlb)
1095 			dev_err(&v->dev, "no iotlb for asid %d\n", asid);
1096 		r = -EINVAL;
1097 		goto unlock;
1098 	}
1099 
1100 	switch (msg->type) {
1101 	case VHOST_IOTLB_UPDATE:
1102 		r = vhost_vdpa_process_iotlb_update(v, iotlb, msg);
1103 		break;
1104 	case VHOST_IOTLB_INVALIDATE:
1105 		vhost_vdpa_unmap(v, iotlb, msg->iova, msg->size);
1106 		break;
1107 	case VHOST_IOTLB_BATCH_BEGIN:
1108 		v->batch_asid = asid;
1109 		v->in_batch = true;
1110 		break;
1111 	case VHOST_IOTLB_BATCH_END:
1112 		if (v->in_batch && ops->set_map)
1113 			ops->set_map(vdpa, asid, iotlb);
1114 		v->in_batch = false;
1115 		if (!iotlb->nmaps)
1116 			vhost_vdpa_remove_as(v, asid);
1117 		break;
1118 	default:
1119 		r = -EINVAL;
1120 		break;
1121 	}
1122 unlock:
1123 	mutex_unlock(&dev->mutex);
1124 
1125 	return r;
1126 }
1127 
1128 static ssize_t vhost_vdpa_chr_write_iter(struct kiocb *iocb,
1129 					 struct iov_iter *from)
1130 {
1131 	struct file *file = iocb->ki_filp;
1132 	struct vhost_vdpa *v = file->private_data;
1133 	struct vhost_dev *dev = &v->vdev;
1134 
1135 	return vhost_chr_write_iter(dev, from);
1136 }
1137 
1138 static int vhost_vdpa_alloc_domain(struct vhost_vdpa *v)
1139 {
1140 	struct vdpa_device *vdpa = v->vdpa;
1141 	const struct vdpa_config_ops *ops = vdpa->config;
1142 	struct device *dma_dev = vdpa_get_dma_dev(vdpa);
1143 	struct bus_type *bus;
1144 	int ret;
1145 
1146 	/* Device want to do DMA by itself */
1147 	if (ops->set_map || ops->dma_map)
1148 		return 0;
1149 
1150 	bus = dma_dev->bus;
1151 	if (!bus)
1152 		return -EFAULT;
1153 
1154 	if (!device_iommu_capable(dma_dev, IOMMU_CAP_CACHE_COHERENCY)) {
1155 		dev_warn_once(&v->dev,
1156 			      "Failed to allocate domain, device is not IOMMU cache coherent capable\n");
1157 		return -ENOTSUPP;
1158 	}
1159 
1160 	v->domain = iommu_domain_alloc(bus);
1161 	if (!v->domain)
1162 		return -EIO;
1163 
1164 	ret = iommu_attach_device(v->domain, dma_dev);
1165 	if (ret)
1166 		goto err_attach;
1167 
1168 	return 0;
1169 
1170 err_attach:
1171 	iommu_domain_free(v->domain);
1172 	v->domain = NULL;
1173 	return ret;
1174 }
1175 
1176 static void vhost_vdpa_free_domain(struct vhost_vdpa *v)
1177 {
1178 	struct vdpa_device *vdpa = v->vdpa;
1179 	struct device *dma_dev = vdpa_get_dma_dev(vdpa);
1180 
1181 	if (v->domain) {
1182 		iommu_detach_device(v->domain, dma_dev);
1183 		iommu_domain_free(v->domain);
1184 	}
1185 
1186 	v->domain = NULL;
1187 }
1188 
1189 static void vhost_vdpa_set_iova_range(struct vhost_vdpa *v)
1190 {
1191 	struct vdpa_iova_range *range = &v->range;
1192 	struct vdpa_device *vdpa = v->vdpa;
1193 	const struct vdpa_config_ops *ops = vdpa->config;
1194 
1195 	if (ops->get_iova_range) {
1196 		*range = ops->get_iova_range(vdpa);
1197 	} else if (v->domain && v->domain->geometry.force_aperture) {
1198 		range->first = v->domain->geometry.aperture_start;
1199 		range->last = v->domain->geometry.aperture_end;
1200 	} else {
1201 		range->first = 0;
1202 		range->last = ULLONG_MAX;
1203 	}
1204 }
1205 
1206 static void vhost_vdpa_cleanup(struct vhost_vdpa *v)
1207 {
1208 	struct vhost_vdpa_as *as;
1209 	u32 asid;
1210 
1211 	for (asid = 0; asid < v->vdpa->nas; asid++) {
1212 		as = asid_to_as(v, asid);
1213 		if (as)
1214 			vhost_vdpa_remove_as(v, asid);
1215 	}
1216 
1217 	vhost_vdpa_free_domain(v);
1218 	vhost_dev_cleanup(&v->vdev);
1219 	kfree(v->vdev.vqs);
1220 }
1221 
1222 static int vhost_vdpa_open(struct inode *inode, struct file *filep)
1223 {
1224 	struct vhost_vdpa *v;
1225 	struct vhost_dev *dev;
1226 	struct vhost_virtqueue **vqs;
1227 	int r, opened;
1228 	u32 i, nvqs;
1229 
1230 	v = container_of(inode->i_cdev, struct vhost_vdpa, cdev);
1231 
1232 	opened = atomic_cmpxchg(&v->opened, 0, 1);
1233 	if (opened)
1234 		return -EBUSY;
1235 
1236 	nvqs = v->nvqs;
1237 	r = vhost_vdpa_reset(v);
1238 	if (r)
1239 		goto err;
1240 
1241 	vqs = kmalloc_array(nvqs, sizeof(*vqs), GFP_KERNEL);
1242 	if (!vqs) {
1243 		r = -ENOMEM;
1244 		goto err;
1245 	}
1246 
1247 	dev = &v->vdev;
1248 	for (i = 0; i < nvqs; i++) {
1249 		vqs[i] = &v->vqs[i];
1250 		vqs[i]->handle_kick = handle_vq_kick;
1251 	}
1252 	vhost_dev_init(dev, vqs, nvqs, 0, 0, 0, false,
1253 		       vhost_vdpa_process_iotlb_msg);
1254 
1255 	r = vhost_vdpa_alloc_domain(v);
1256 	if (r)
1257 		goto err_alloc_domain;
1258 
1259 	vhost_vdpa_set_iova_range(v);
1260 
1261 	filep->private_data = v;
1262 
1263 	return 0;
1264 
1265 err_alloc_domain:
1266 	vhost_vdpa_cleanup(v);
1267 err:
1268 	atomic_dec(&v->opened);
1269 	return r;
1270 }
1271 
1272 static void vhost_vdpa_clean_irq(struct vhost_vdpa *v)
1273 {
1274 	u32 i;
1275 
1276 	for (i = 0; i < v->nvqs; i++)
1277 		vhost_vdpa_unsetup_vq_irq(v, i);
1278 }
1279 
1280 static int vhost_vdpa_release(struct inode *inode, struct file *filep)
1281 {
1282 	struct vhost_vdpa *v = filep->private_data;
1283 	struct vhost_dev *d = &v->vdev;
1284 
1285 	mutex_lock(&d->mutex);
1286 	filep->private_data = NULL;
1287 	vhost_vdpa_clean_irq(v);
1288 	vhost_vdpa_reset(v);
1289 	vhost_dev_stop(&v->vdev);
1290 	vhost_vdpa_config_put(v);
1291 	vhost_vdpa_cleanup(v);
1292 	mutex_unlock(&d->mutex);
1293 
1294 	atomic_dec(&v->opened);
1295 	complete(&v->completion);
1296 
1297 	return 0;
1298 }
1299 
1300 #ifdef CONFIG_MMU
1301 static vm_fault_t vhost_vdpa_fault(struct vm_fault *vmf)
1302 {
1303 	struct vhost_vdpa *v = vmf->vma->vm_file->private_data;
1304 	struct vdpa_device *vdpa = v->vdpa;
1305 	const struct vdpa_config_ops *ops = vdpa->config;
1306 	struct vdpa_notification_area notify;
1307 	struct vm_area_struct *vma = vmf->vma;
1308 	u16 index = vma->vm_pgoff;
1309 
1310 	notify = ops->get_vq_notification(vdpa, index);
1311 
1312 	vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1313 	if (remap_pfn_range(vma, vmf->address & PAGE_MASK,
1314 			    PFN_DOWN(notify.addr), PAGE_SIZE,
1315 			    vma->vm_page_prot))
1316 		return VM_FAULT_SIGBUS;
1317 
1318 	return VM_FAULT_NOPAGE;
1319 }
1320 
1321 static const struct vm_operations_struct vhost_vdpa_vm_ops = {
1322 	.fault = vhost_vdpa_fault,
1323 };
1324 
1325 static int vhost_vdpa_mmap(struct file *file, struct vm_area_struct *vma)
1326 {
1327 	struct vhost_vdpa *v = vma->vm_file->private_data;
1328 	struct vdpa_device *vdpa = v->vdpa;
1329 	const struct vdpa_config_ops *ops = vdpa->config;
1330 	struct vdpa_notification_area notify;
1331 	unsigned long index = vma->vm_pgoff;
1332 
1333 	if (vma->vm_end - vma->vm_start != PAGE_SIZE)
1334 		return -EINVAL;
1335 	if ((vma->vm_flags & VM_SHARED) == 0)
1336 		return -EINVAL;
1337 	if (vma->vm_flags & VM_READ)
1338 		return -EINVAL;
1339 	if (index > 65535)
1340 		return -EINVAL;
1341 	if (!ops->get_vq_notification)
1342 		return -ENOTSUPP;
1343 
1344 	/* To be safe and easily modelled by userspace, We only
1345 	 * support the doorbell which sits on the page boundary and
1346 	 * does not share the page with other registers.
1347 	 */
1348 	notify = ops->get_vq_notification(vdpa, index);
1349 	if (notify.addr & (PAGE_SIZE - 1))
1350 		return -EINVAL;
1351 	if (vma->vm_end - vma->vm_start != notify.size)
1352 		return -ENOTSUPP;
1353 
1354 	vm_flags_set(vma, VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP);
1355 	vma->vm_ops = &vhost_vdpa_vm_ops;
1356 	return 0;
1357 }
1358 #endif /* CONFIG_MMU */
1359 
1360 static const struct file_operations vhost_vdpa_fops = {
1361 	.owner		= THIS_MODULE,
1362 	.open		= vhost_vdpa_open,
1363 	.release	= vhost_vdpa_release,
1364 	.write_iter	= vhost_vdpa_chr_write_iter,
1365 	.unlocked_ioctl	= vhost_vdpa_unlocked_ioctl,
1366 #ifdef CONFIG_MMU
1367 	.mmap		= vhost_vdpa_mmap,
1368 #endif /* CONFIG_MMU */
1369 	.compat_ioctl	= compat_ptr_ioctl,
1370 };
1371 
1372 static void vhost_vdpa_release_dev(struct device *device)
1373 {
1374 	struct vhost_vdpa *v =
1375 	       container_of(device, struct vhost_vdpa, dev);
1376 
1377 	ida_simple_remove(&vhost_vdpa_ida, v->minor);
1378 	kfree(v->vqs);
1379 	kfree(v);
1380 }
1381 
1382 static int vhost_vdpa_probe(struct vdpa_device *vdpa)
1383 {
1384 	const struct vdpa_config_ops *ops = vdpa->config;
1385 	struct vhost_vdpa *v;
1386 	int minor;
1387 	int i, r;
1388 
1389 	/* We can't support platform IOMMU device with more than 1
1390 	 * group or as
1391 	 */
1392 	if (!ops->set_map && !ops->dma_map &&
1393 	    (vdpa->ngroups > 1 || vdpa->nas > 1))
1394 		return -EOPNOTSUPP;
1395 
1396 	v = kzalloc(sizeof(*v), GFP_KERNEL | __GFP_RETRY_MAYFAIL);
1397 	if (!v)
1398 		return -ENOMEM;
1399 
1400 	minor = ida_simple_get(&vhost_vdpa_ida, 0,
1401 			       VHOST_VDPA_DEV_MAX, GFP_KERNEL);
1402 	if (minor < 0) {
1403 		kfree(v);
1404 		return minor;
1405 	}
1406 
1407 	atomic_set(&v->opened, 0);
1408 	v->minor = minor;
1409 	v->vdpa = vdpa;
1410 	v->nvqs = vdpa->nvqs;
1411 	v->virtio_id = ops->get_device_id(vdpa);
1412 
1413 	device_initialize(&v->dev);
1414 	v->dev.release = vhost_vdpa_release_dev;
1415 	v->dev.parent = &vdpa->dev;
1416 	v->dev.devt = MKDEV(MAJOR(vhost_vdpa_major), minor);
1417 	v->vqs = kmalloc_array(v->nvqs, sizeof(struct vhost_virtqueue),
1418 			       GFP_KERNEL);
1419 	if (!v->vqs) {
1420 		r = -ENOMEM;
1421 		goto err;
1422 	}
1423 
1424 	r = dev_set_name(&v->dev, "vhost-vdpa-%u", minor);
1425 	if (r)
1426 		goto err;
1427 
1428 	cdev_init(&v->cdev, &vhost_vdpa_fops);
1429 	v->cdev.owner = THIS_MODULE;
1430 
1431 	r = cdev_device_add(&v->cdev, &v->dev);
1432 	if (r)
1433 		goto err;
1434 
1435 	init_completion(&v->completion);
1436 	vdpa_set_drvdata(vdpa, v);
1437 
1438 	for (i = 0; i < VHOST_VDPA_IOTLB_BUCKETS; i++)
1439 		INIT_HLIST_HEAD(&v->as[i]);
1440 
1441 	return 0;
1442 
1443 err:
1444 	put_device(&v->dev);
1445 	ida_simple_remove(&vhost_vdpa_ida, v->minor);
1446 	return r;
1447 }
1448 
1449 static void vhost_vdpa_remove(struct vdpa_device *vdpa)
1450 {
1451 	struct vhost_vdpa *v = vdpa_get_drvdata(vdpa);
1452 	int opened;
1453 
1454 	cdev_device_del(&v->cdev, &v->dev);
1455 
1456 	do {
1457 		opened = atomic_cmpxchg(&v->opened, 0, 1);
1458 		if (!opened)
1459 			break;
1460 		wait_for_completion(&v->completion);
1461 	} while (1);
1462 
1463 	put_device(&v->dev);
1464 }
1465 
1466 static struct vdpa_driver vhost_vdpa_driver = {
1467 	.driver = {
1468 		.name	= "vhost_vdpa",
1469 	},
1470 	.probe	= vhost_vdpa_probe,
1471 	.remove	= vhost_vdpa_remove,
1472 };
1473 
1474 static int __init vhost_vdpa_init(void)
1475 {
1476 	int r;
1477 
1478 	r = alloc_chrdev_region(&vhost_vdpa_major, 0, VHOST_VDPA_DEV_MAX,
1479 				"vhost-vdpa");
1480 	if (r)
1481 		goto err_alloc_chrdev;
1482 
1483 	r = vdpa_register_driver(&vhost_vdpa_driver);
1484 	if (r)
1485 		goto err_vdpa_register_driver;
1486 
1487 	return 0;
1488 
1489 err_vdpa_register_driver:
1490 	unregister_chrdev_region(vhost_vdpa_major, VHOST_VDPA_DEV_MAX);
1491 err_alloc_chrdev:
1492 	return r;
1493 }
1494 module_init(vhost_vdpa_init);
1495 
1496 static void __exit vhost_vdpa_exit(void)
1497 {
1498 	vdpa_unregister_driver(&vhost_vdpa_driver);
1499 	unregister_chrdev_region(vhost_vdpa_major, VHOST_VDPA_DEV_MAX);
1500 }
1501 module_exit(vhost_vdpa_exit);
1502 
1503 MODULE_VERSION("0.0.1");
1504 MODULE_LICENSE("GPL v2");
1505 MODULE_AUTHOR("Intel Corporation");
1506 MODULE_DESCRIPTION("vDPA-based vhost backend for virtio");
1507