Commit | Line | Data |
---|---|---|
4c8cf318 TB |
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> | |
ddd89d0a | 18 | #include <linux/mm.h> |
4c8cf318 TB |
19 | #include <linux/iommu.h> |
20 | #include <linux/uuid.h> | |
21 | #include <linux/vdpa.h> | |
22 | #include <linux/nospec.h> | |
23 | #include <linux/vhost.h> | |
24 | #include <linux/virtio_net.h> | |
776f3950 | 25 | #include <linux/kernel.h> |
4c8cf318 TB |
26 | |
27 | #include "vhost.h" | |
28 | ||
4c8cf318 TB |
29 | /* Currently, only network backend w/o multiqueue is supported. */ |
30 | #define VHOST_VDPA_VQ_MAX 2 | |
31 | ||
32 | #define VHOST_VDPA_DEV_MAX (1U << MINORBITS) | |
33 | ||
34 | struct vhost_vdpa { | |
35 | struct vhost_dev vdev; | |
36 | struct iommu_domain *domain; | |
37 | struct vhost_virtqueue *vqs; | |
38 | struct completion completion; | |
39 | struct vdpa_device *vdpa; | |
40 | struct device dev; | |
41 | struct cdev cdev; | |
42 | atomic_t opened; | |
43 | int nvqs; | |
44 | int virtio_id; | |
45 | int minor; | |
776f3950 | 46 | struct eventfd_ctx *config_ctx; |
4c8cf318 TB |
47 | }; |
48 | ||
49 | static DEFINE_IDA(vhost_vdpa_ida); | |
50 | ||
51 | static dev_t vhost_vdpa_major; | |
52 | ||
4c8cf318 TB |
53 | static void handle_vq_kick(struct vhost_work *work) |
54 | { | |
55 | struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue, | |
56 | poll.work); | |
57 | struct vhost_vdpa *v = container_of(vq->dev, struct vhost_vdpa, vdev); | |
58 | const struct vdpa_config_ops *ops = v->vdpa->config; | |
59 | ||
60 | ops->kick_vq(v->vdpa, vq - v->vqs); | |
61 | } | |
62 | ||
63 | static irqreturn_t vhost_vdpa_virtqueue_cb(void *private) | |
64 | { | |
65 | struct vhost_virtqueue *vq = private; | |
265a0ad8 | 66 | struct eventfd_ctx *call_ctx = vq->call_ctx.ctx; |
4c8cf318 TB |
67 | |
68 | if (call_ctx) | |
69 | eventfd_signal(call_ctx, 1); | |
70 | ||
71 | return IRQ_HANDLED; | |
72 | } | |
73 | ||
776f3950 ZL |
74 | static irqreturn_t vhost_vdpa_config_cb(void *private) |
75 | { | |
76 | struct vhost_vdpa *v = private; | |
77 | struct eventfd_ctx *config_ctx = v->config_ctx; | |
78 | ||
79 | if (config_ctx) | |
80 | eventfd_signal(config_ctx, 1); | |
81 | ||
82 | return IRQ_HANDLED; | |
83 | } | |
84 | ||
2cf1ba9a ZL |
85 | static void vhost_vdpa_setup_vq_irq(struct vhost_vdpa *v, u16 qid) |
86 | { | |
87 | struct vhost_virtqueue *vq = &v->vqs[qid]; | |
88 | const struct vdpa_config_ops *ops = v->vdpa->config; | |
89 | struct vdpa_device *vdpa = v->vdpa; | |
90 | int ret, irq; | |
91 | ||
92 | if (!ops->get_vq_irq) | |
93 | return; | |
94 | ||
95 | irq = ops->get_vq_irq(vdpa, qid); | |
96 | spin_lock(&vq->call_ctx.ctx_lock); | |
97 | irq_bypass_unregister_producer(&vq->call_ctx.producer); | |
98 | if (!vq->call_ctx.ctx || irq < 0) { | |
99 | spin_unlock(&vq->call_ctx.ctx_lock); | |
100 | return; | |
101 | } | |
102 | ||
103 | vq->call_ctx.producer.token = vq->call_ctx.ctx; | |
104 | vq->call_ctx.producer.irq = irq; | |
105 | ret = irq_bypass_register_producer(&vq->call_ctx.producer); | |
106 | spin_unlock(&vq->call_ctx.ctx_lock); | |
107 | } | |
108 | ||
109 | static void vhost_vdpa_unsetup_vq_irq(struct vhost_vdpa *v, u16 qid) | |
110 | { | |
111 | struct vhost_virtqueue *vq = &v->vqs[qid]; | |
112 | ||
113 | spin_lock(&vq->call_ctx.ctx_lock); | |
114 | irq_bypass_unregister_producer(&vq->call_ctx.producer); | |
115 | spin_unlock(&vq->call_ctx.ctx_lock); | |
116 | } | |
117 | ||
4c8cf318 TB |
118 | static void vhost_vdpa_reset(struct vhost_vdpa *v) |
119 | { | |
120 | struct vdpa_device *vdpa = v->vdpa; | |
4c8cf318 | 121 | |
0d234007 | 122 | vdpa_reset(vdpa); |
4c8cf318 TB |
123 | } |
124 | ||
125 | static long vhost_vdpa_get_device_id(struct vhost_vdpa *v, u8 __user *argp) | |
126 | { | |
127 | struct vdpa_device *vdpa = v->vdpa; | |
128 | const struct vdpa_config_ops *ops = vdpa->config; | |
129 | u32 device_id; | |
130 | ||
131 | device_id = ops->get_device_id(vdpa); | |
132 | ||
133 | if (copy_to_user(argp, &device_id, sizeof(device_id))) | |
134 | return -EFAULT; | |
135 | ||
136 | return 0; | |
137 | } | |
138 | ||
139 | static long vhost_vdpa_get_status(struct vhost_vdpa *v, u8 __user *statusp) | |
140 | { | |
141 | struct vdpa_device *vdpa = v->vdpa; | |
142 | const struct vdpa_config_ops *ops = vdpa->config; | |
143 | u8 status; | |
144 | ||
145 | status = ops->get_status(vdpa); | |
146 | ||
147 | if (copy_to_user(statusp, &status, sizeof(status))) | |
148 | return -EFAULT; | |
149 | ||
150 | return 0; | |
151 | } | |
152 | ||
153 | static long vhost_vdpa_set_status(struct vhost_vdpa *v, u8 __user *statusp) | |
154 | { | |
155 | struct vdpa_device *vdpa = v->vdpa; | |
156 | const struct vdpa_config_ops *ops = vdpa->config; | |
2cf1ba9a ZL |
157 | u8 status, status_old; |
158 | int nvqs = v->nvqs; | |
159 | u16 i; | |
4c8cf318 TB |
160 | |
161 | if (copy_from_user(&status, statusp, sizeof(status))) | |
162 | return -EFAULT; | |
163 | ||
2cf1ba9a ZL |
164 | status_old = ops->get_status(vdpa); |
165 | ||
4c8cf318 TB |
166 | /* |
167 | * Userspace shouldn't remove status bits unless reset the | |
168 | * status to 0. | |
169 | */ | |
170 | if (status != 0 && (ops->get_status(vdpa) & ~status) != 0) | |
171 | return -EINVAL; | |
172 | ||
173 | ops->set_status(vdpa, status); | |
174 | ||
2cf1ba9a ZL |
175 | if ((status & VIRTIO_CONFIG_S_DRIVER_OK) && !(status_old & VIRTIO_CONFIG_S_DRIVER_OK)) |
176 | for (i = 0; i < nvqs; i++) | |
177 | vhost_vdpa_setup_vq_irq(v, i); | |
178 | ||
179 | if ((status_old & VIRTIO_CONFIG_S_DRIVER_OK) && !(status & VIRTIO_CONFIG_S_DRIVER_OK)) | |
180 | for (i = 0; i < nvqs; i++) | |
181 | vhost_vdpa_unsetup_vq_irq(v, i); | |
182 | ||
4c8cf318 TB |
183 | return 0; |
184 | } | |
185 | ||
186 | static int vhost_vdpa_config_validate(struct vhost_vdpa *v, | |
187 | struct vhost_vdpa_config *c) | |
188 | { | |
189 | long size = 0; | |
190 | ||
191 | switch (v->virtio_id) { | |
192 | case VIRTIO_ID_NET: | |
193 | size = sizeof(struct virtio_net_config); | |
194 | break; | |
195 | } | |
196 | ||
197 | if (c->len == 0) | |
198 | return -EINVAL; | |
199 | ||
200 | if (c->len > size - c->off) | |
201 | return -E2BIG; | |
202 | ||
203 | return 0; | |
204 | } | |
205 | ||
206 | static long vhost_vdpa_get_config(struct vhost_vdpa *v, | |
207 | struct vhost_vdpa_config __user *c) | |
208 | { | |
209 | struct vdpa_device *vdpa = v->vdpa; | |
4c8cf318 TB |
210 | struct vhost_vdpa_config config; |
211 | unsigned long size = offsetof(struct vhost_vdpa_config, buf); | |
212 | u8 *buf; | |
213 | ||
214 | if (copy_from_user(&config, c, size)) | |
215 | return -EFAULT; | |
216 | if (vhost_vdpa_config_validate(v, &config)) | |
217 | return -EINVAL; | |
218 | buf = kvzalloc(config.len, GFP_KERNEL); | |
219 | if (!buf) | |
220 | return -ENOMEM; | |
221 | ||
0d234007 | 222 | vdpa_get_config(vdpa, config.off, buf, config.len); |
4c8cf318 TB |
223 | |
224 | if (copy_to_user(c->buf, buf, config.len)) { | |
225 | kvfree(buf); | |
226 | return -EFAULT; | |
227 | } | |
228 | ||
229 | kvfree(buf); | |
230 | return 0; | |
231 | } | |
232 | ||
233 | static long vhost_vdpa_set_config(struct vhost_vdpa *v, | |
234 | struct vhost_vdpa_config __user *c) | |
235 | { | |
236 | struct vdpa_device *vdpa = v->vdpa; | |
237 | const struct vdpa_config_ops *ops = vdpa->config; | |
238 | struct vhost_vdpa_config config; | |
239 | unsigned long size = offsetof(struct vhost_vdpa_config, buf); | |
240 | u8 *buf; | |
241 | ||
242 | if (copy_from_user(&config, c, size)) | |
243 | return -EFAULT; | |
244 | if (vhost_vdpa_config_validate(v, &config)) | |
245 | return -EINVAL; | |
246 | buf = kvzalloc(config.len, GFP_KERNEL); | |
247 | if (!buf) | |
248 | return -ENOMEM; | |
249 | ||
250 | if (copy_from_user(buf, c->buf, config.len)) { | |
251 | kvfree(buf); | |
252 | return -EFAULT; | |
253 | } | |
254 | ||
255 | ops->set_config(vdpa, config.off, buf, config.len); | |
256 | ||
257 | kvfree(buf); | |
258 | return 0; | |
259 | } | |
260 | ||
261 | static long vhost_vdpa_get_features(struct vhost_vdpa *v, u64 __user *featurep) | |
262 | { | |
263 | struct vdpa_device *vdpa = v->vdpa; | |
264 | const struct vdpa_config_ops *ops = vdpa->config; | |
265 | u64 features; | |
266 | ||
267 | features = ops->get_features(vdpa); | |
4c8cf318 TB |
268 | |
269 | if (copy_to_user(featurep, &features, sizeof(features))) | |
270 | return -EFAULT; | |
271 | ||
272 | return 0; | |
273 | } | |
274 | ||
275 | static long vhost_vdpa_set_features(struct vhost_vdpa *v, u64 __user *featurep) | |
276 | { | |
277 | struct vdpa_device *vdpa = v->vdpa; | |
278 | const struct vdpa_config_ops *ops = vdpa->config; | |
279 | u64 features; | |
280 | ||
281 | /* | |
282 | * It's not allowed to change the features after they have | |
283 | * been negotiated. | |
284 | */ | |
285 | if (ops->get_status(vdpa) & VIRTIO_CONFIG_S_FEATURES_OK) | |
286 | return -EBUSY; | |
287 | ||
288 | if (copy_from_user(&features, featurep, sizeof(features))) | |
289 | return -EFAULT; | |
290 | ||
0d234007 | 291 | if (vdpa_set_features(vdpa, features)) |
4c8cf318 TB |
292 | return -EINVAL; |
293 | ||
294 | return 0; | |
295 | } | |
296 | ||
297 | static long vhost_vdpa_get_vring_num(struct vhost_vdpa *v, u16 __user *argp) | |
298 | { | |
299 | struct vdpa_device *vdpa = v->vdpa; | |
300 | const struct vdpa_config_ops *ops = vdpa->config; | |
301 | u16 num; | |
302 | ||
303 | num = ops->get_vq_num_max(vdpa); | |
304 | ||
305 | if (copy_to_user(argp, &num, sizeof(num))) | |
306 | return -EFAULT; | |
307 | ||
308 | return 0; | |
309 | } | |
310 | ||
776f3950 ZL |
311 | static void vhost_vdpa_config_put(struct vhost_vdpa *v) |
312 | { | |
313 | if (v->config_ctx) | |
314 | eventfd_ctx_put(v->config_ctx); | |
315 | } | |
316 | ||
317 | static long vhost_vdpa_set_config_call(struct vhost_vdpa *v, u32 __user *argp) | |
318 | { | |
319 | struct vdpa_callback cb; | |
320 | int fd; | |
321 | struct eventfd_ctx *ctx; | |
322 | ||
323 | cb.callback = vhost_vdpa_config_cb; | |
324 | cb.private = v->vdpa; | |
325 | if (copy_from_user(&fd, argp, sizeof(fd))) | |
326 | return -EFAULT; | |
327 | ||
328 | ctx = fd == VHOST_FILE_UNBIND ? NULL : eventfd_ctx_fdget(fd); | |
329 | swap(ctx, v->config_ctx); | |
330 | ||
331 | if (!IS_ERR_OR_NULL(ctx)) | |
332 | eventfd_ctx_put(ctx); | |
333 | ||
334 | if (IS_ERR(v->config_ctx)) | |
335 | return PTR_ERR(v->config_ctx); | |
336 | ||
337 | v->vdpa->config->set_config_cb(v->vdpa, &cb); | |
338 | ||
339 | return 0; | |
340 | } | |
2cf1ba9a | 341 | |
4c8cf318 TB |
342 | static long vhost_vdpa_vring_ioctl(struct vhost_vdpa *v, unsigned int cmd, |
343 | void __user *argp) | |
344 | { | |
345 | struct vdpa_device *vdpa = v->vdpa; | |
346 | const struct vdpa_config_ops *ops = vdpa->config; | |
347 | struct vdpa_callback cb; | |
348 | struct vhost_virtqueue *vq; | |
349 | struct vhost_vring_state s; | |
4c8cf318 TB |
350 | u32 idx; |
351 | long r; | |
352 | ||
353 | r = get_user(idx, (u32 __user *)argp); | |
354 | if (r < 0) | |
355 | return r; | |
356 | ||
357 | if (idx >= v->nvqs) | |
358 | return -ENOBUFS; | |
359 | ||
360 | idx = array_index_nospec(idx, v->nvqs); | |
361 | vq = &v->vqs[idx]; | |
362 | ||
4c8cf318 TB |
363 | if (cmd == VHOST_VDPA_SET_VRING_ENABLE) { |
364 | if (copy_from_user(&s, argp, sizeof(s))) | |
365 | return -EFAULT; | |
366 | ops->set_vq_ready(vdpa, idx, s.num); | |
367 | return 0; | |
368 | } | |
369 | ||
370 | if (cmd == VHOST_GET_VRING_BASE) | |
371 | vq->last_avail_idx = ops->get_vq_state(v->vdpa, idx); | |
372 | ||
373 | r = vhost_vring_ioctl(&v->vdev, cmd, argp); | |
374 | if (r) | |
375 | return r; | |
376 | ||
377 | switch (cmd) { | |
378 | case VHOST_SET_VRING_ADDR: | |
379 | if (ops->set_vq_address(vdpa, idx, | |
380 | (u64)(uintptr_t)vq->desc, | |
381 | (u64)(uintptr_t)vq->avail, | |
382 | (u64)(uintptr_t)vq->used)) | |
383 | r = -EINVAL; | |
384 | break; | |
385 | ||
386 | case VHOST_SET_VRING_BASE: | |
387 | if (ops->set_vq_state(vdpa, idx, vq->last_avail_idx)) | |
388 | r = -EINVAL; | |
389 | break; | |
390 | ||
391 | case VHOST_SET_VRING_CALL: | |
265a0ad8 | 392 | if (vq->call_ctx.ctx) { |
4c8cf318 TB |
393 | cb.callback = vhost_vdpa_virtqueue_cb; |
394 | cb.private = vq; | |
395 | } else { | |
396 | cb.callback = NULL; | |
397 | cb.private = NULL; | |
398 | } | |
399 | ops->set_vq_cb(vdpa, idx, &cb); | |
2cf1ba9a | 400 | vhost_vdpa_setup_vq_irq(v, idx); |
4c8cf318 TB |
401 | break; |
402 | ||
403 | case VHOST_SET_VRING_NUM: | |
404 | ops->set_vq_num(vdpa, idx, vq->num); | |
405 | break; | |
406 | } | |
407 | ||
408 | return r; | |
409 | } | |
410 | ||
411 | static long vhost_vdpa_unlocked_ioctl(struct file *filep, | |
412 | unsigned int cmd, unsigned long arg) | |
413 | { | |
414 | struct vhost_vdpa *v = filep->private_data; | |
415 | struct vhost_dev *d = &v->vdev; | |
416 | void __user *argp = (void __user *)arg; | |
417 | long r; | |
418 | ||
419 | mutex_lock(&d->mutex); | |
420 | ||
421 | switch (cmd) { | |
422 | case VHOST_VDPA_GET_DEVICE_ID: | |
423 | r = vhost_vdpa_get_device_id(v, argp); | |
424 | break; | |
425 | case VHOST_VDPA_GET_STATUS: | |
426 | r = vhost_vdpa_get_status(v, argp); | |
427 | break; | |
428 | case VHOST_VDPA_SET_STATUS: | |
429 | r = vhost_vdpa_set_status(v, argp); | |
430 | break; | |
431 | case VHOST_VDPA_GET_CONFIG: | |
432 | r = vhost_vdpa_get_config(v, argp); | |
433 | break; | |
434 | case VHOST_VDPA_SET_CONFIG: | |
435 | r = vhost_vdpa_set_config(v, argp); | |
436 | break; | |
437 | case VHOST_GET_FEATURES: | |
438 | r = vhost_vdpa_get_features(v, argp); | |
439 | break; | |
440 | case VHOST_SET_FEATURES: | |
441 | r = vhost_vdpa_set_features(v, argp); | |
442 | break; | |
443 | case VHOST_VDPA_GET_VRING_NUM: | |
444 | r = vhost_vdpa_get_vring_num(v, argp); | |
445 | break; | |
446 | case VHOST_SET_LOG_BASE: | |
447 | case VHOST_SET_LOG_FD: | |
448 | r = -ENOIOCTLCMD; | |
449 | break; | |
776f3950 ZL |
450 | case VHOST_VDPA_SET_CONFIG_CALL: |
451 | r = vhost_vdpa_set_config_call(v, argp); | |
452 | break; | |
4c8cf318 TB |
453 | default: |
454 | r = vhost_dev_ioctl(&v->vdev, cmd, argp); | |
455 | if (r == -ENOIOCTLCMD) | |
456 | r = vhost_vdpa_vring_ioctl(v, cmd, argp); | |
457 | break; | |
458 | } | |
459 | ||
460 | mutex_unlock(&d->mutex); | |
461 | return r; | |
462 | } | |
463 | ||
464 | static void vhost_vdpa_iotlb_unmap(struct vhost_vdpa *v, u64 start, u64 last) | |
465 | { | |
466 | struct vhost_dev *dev = &v->vdev; | |
467 | struct vhost_iotlb *iotlb = dev->iotlb; | |
468 | struct vhost_iotlb_map *map; | |
469 | struct page *page; | |
470 | unsigned long pfn, pinned; | |
471 | ||
472 | while ((map = vhost_iotlb_itree_first(iotlb, start, last)) != NULL) { | |
473 | pinned = map->size >> PAGE_SHIFT; | |
474 | for (pfn = map->addr >> PAGE_SHIFT; | |
475 | pinned > 0; pfn++, pinned--) { | |
476 | page = pfn_to_page(pfn); | |
477 | if (map->perm & VHOST_ACCESS_WO) | |
478 | set_page_dirty_lock(page); | |
479 | unpin_user_page(page); | |
480 | } | |
481 | atomic64_sub(map->size >> PAGE_SHIFT, &dev->mm->pinned_vm); | |
482 | vhost_iotlb_map_free(iotlb, map); | |
483 | } | |
484 | } | |
485 | ||
486 | static void vhost_vdpa_iotlb_free(struct vhost_vdpa *v) | |
487 | { | |
488 | struct vhost_dev *dev = &v->vdev; | |
489 | ||
490 | vhost_vdpa_iotlb_unmap(v, 0ULL, 0ULL - 1); | |
491 | kfree(dev->iotlb); | |
492 | dev->iotlb = NULL; | |
493 | } | |
494 | ||
495 | static int perm_to_iommu_flags(u32 perm) | |
496 | { | |
497 | int flags = 0; | |
498 | ||
499 | switch (perm) { | |
500 | case VHOST_ACCESS_WO: | |
501 | flags |= IOMMU_WRITE; | |
502 | break; | |
503 | case VHOST_ACCESS_RO: | |
504 | flags |= IOMMU_READ; | |
505 | break; | |
506 | case VHOST_ACCESS_RW: | |
507 | flags |= (IOMMU_WRITE | IOMMU_READ); | |
508 | break; | |
509 | default: | |
510 | WARN(1, "invalidate vhost IOTLB permission\n"); | |
511 | break; | |
512 | } | |
513 | ||
514 | return flags | IOMMU_CACHE; | |
515 | } | |
516 | ||
517 | static int vhost_vdpa_map(struct vhost_vdpa *v, | |
518 | u64 iova, u64 size, u64 pa, u32 perm) | |
519 | { | |
520 | struct vhost_dev *dev = &v->vdev; | |
521 | struct vdpa_device *vdpa = v->vdpa; | |
522 | const struct vdpa_config_ops *ops = vdpa->config; | |
523 | int r = 0; | |
524 | ||
525 | r = vhost_iotlb_add_range(dev->iotlb, iova, iova + size - 1, | |
526 | pa, perm); | |
527 | if (r) | |
528 | return r; | |
529 | ||
530 | if (ops->dma_map) | |
531 | r = ops->dma_map(vdpa, iova, size, pa, perm); | |
532 | else if (ops->set_map) | |
533 | r = ops->set_map(vdpa, dev->iotlb); | |
534 | else | |
535 | r = iommu_map(v->domain, iova, pa, size, | |
536 | perm_to_iommu_flags(perm)); | |
537 | ||
538 | return r; | |
539 | } | |
540 | ||
541 | static void vhost_vdpa_unmap(struct vhost_vdpa *v, u64 iova, u64 size) | |
542 | { | |
543 | struct vhost_dev *dev = &v->vdev; | |
544 | struct vdpa_device *vdpa = v->vdpa; | |
545 | const struct vdpa_config_ops *ops = vdpa->config; | |
546 | ||
547 | vhost_vdpa_iotlb_unmap(v, iova, iova + size - 1); | |
548 | ||
549 | if (ops->dma_map) | |
550 | ops->dma_unmap(vdpa, iova, size); | |
551 | else if (ops->set_map) | |
552 | ops->set_map(vdpa, dev->iotlb); | |
553 | else | |
554 | iommu_unmap(v->domain, iova, size); | |
555 | } | |
556 | ||
557 | static int vhost_vdpa_process_iotlb_update(struct vhost_vdpa *v, | |
558 | struct vhost_iotlb_msg *msg) | |
559 | { | |
560 | struct vhost_dev *dev = &v->vdev; | |
561 | struct vhost_iotlb *iotlb = dev->iotlb; | |
562 | struct page **page_list; | |
563 | unsigned long list_size = PAGE_SIZE / sizeof(struct page *); | |
564 | unsigned int gup_flags = FOLL_LONGTERM; | |
565 | unsigned long npages, cur_base, map_pfn, last_pfn = 0; | |
566 | unsigned long locked, lock_limit, pinned, i; | |
567 | u64 iova = msg->iova; | |
568 | int ret = 0; | |
569 | ||
570 | if (vhost_iotlb_itree_first(iotlb, msg->iova, | |
571 | msg->iova + msg->size - 1)) | |
572 | return -EEXIST; | |
573 | ||
574 | page_list = (struct page **) __get_free_page(GFP_KERNEL); | |
575 | if (!page_list) | |
576 | return -ENOMEM; | |
577 | ||
578 | if (msg->perm & VHOST_ACCESS_WO) | |
579 | gup_flags |= FOLL_WRITE; | |
580 | ||
581 | npages = PAGE_ALIGN(msg->size + (iova & ~PAGE_MASK)) >> PAGE_SHIFT; | |
582 | if (!npages) | |
583 | return -EINVAL; | |
584 | ||
d8ed45c5 | 585 | mmap_read_lock(dev->mm); |
4c8cf318 TB |
586 | |
587 | locked = atomic64_add_return(npages, &dev->mm->pinned_vm); | |
588 | lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; | |
589 | ||
590 | if (locked > lock_limit) { | |
591 | ret = -ENOMEM; | |
592 | goto out; | |
593 | } | |
594 | ||
595 | cur_base = msg->uaddr & PAGE_MASK; | |
596 | iova &= PAGE_MASK; | |
597 | ||
598 | while (npages) { | |
599 | pinned = min_t(unsigned long, npages, list_size); | |
600 | ret = pin_user_pages(cur_base, pinned, | |
601 | gup_flags, page_list, NULL); | |
602 | if (ret != pinned) | |
603 | goto out; | |
604 | ||
605 | if (!last_pfn) | |
606 | map_pfn = page_to_pfn(page_list[0]); | |
607 | ||
608 | for (i = 0; i < ret; i++) { | |
609 | unsigned long this_pfn = page_to_pfn(page_list[i]); | |
610 | u64 csize; | |
611 | ||
612 | if (last_pfn && (this_pfn != last_pfn + 1)) { | |
613 | /* Pin a contiguous chunk of memory */ | |
614 | csize = (last_pfn - map_pfn + 1) << PAGE_SHIFT; | |
615 | if (vhost_vdpa_map(v, iova, csize, | |
616 | map_pfn << PAGE_SHIFT, | |
617 | msg->perm)) | |
618 | goto out; | |
619 | map_pfn = this_pfn; | |
620 | iova += csize; | |
621 | } | |
622 | ||
623 | last_pfn = this_pfn; | |
624 | } | |
625 | ||
626 | cur_base += ret << PAGE_SHIFT; | |
627 | npages -= ret; | |
628 | } | |
629 | ||
630 | /* Pin the rest chunk */ | |
631 | ret = vhost_vdpa_map(v, iova, (last_pfn - map_pfn + 1) << PAGE_SHIFT, | |
632 | map_pfn << PAGE_SHIFT, msg->perm); | |
633 | out: | |
634 | if (ret) { | |
635 | vhost_vdpa_unmap(v, msg->iova, msg->size); | |
636 | atomic64_sub(npages, &dev->mm->pinned_vm); | |
637 | } | |
d8ed45c5 | 638 | mmap_read_unlock(dev->mm); |
4c8cf318 TB |
639 | free_page((unsigned long)page_list); |
640 | return ret; | |
641 | } | |
642 | ||
643 | static int vhost_vdpa_process_iotlb_msg(struct vhost_dev *dev, | |
644 | struct vhost_iotlb_msg *msg) | |
645 | { | |
646 | struct vhost_vdpa *v = container_of(dev, struct vhost_vdpa, vdev); | |
647 | int r = 0; | |
648 | ||
649 | r = vhost_dev_check_owner(dev); | |
650 | if (r) | |
651 | return r; | |
652 | ||
653 | switch (msg->type) { | |
654 | case VHOST_IOTLB_UPDATE: | |
655 | r = vhost_vdpa_process_iotlb_update(v, msg); | |
656 | break; | |
657 | case VHOST_IOTLB_INVALIDATE: | |
658 | vhost_vdpa_unmap(v, msg->iova, msg->size); | |
659 | break; | |
660 | default: | |
661 | r = -EINVAL; | |
662 | break; | |
663 | } | |
664 | ||
665 | return r; | |
666 | } | |
667 | ||
668 | static ssize_t vhost_vdpa_chr_write_iter(struct kiocb *iocb, | |
669 | struct iov_iter *from) | |
670 | { | |
671 | struct file *file = iocb->ki_filp; | |
672 | struct vhost_vdpa *v = file->private_data; | |
673 | struct vhost_dev *dev = &v->vdev; | |
674 | ||
675 | return vhost_chr_write_iter(dev, from); | |
676 | } | |
677 | ||
678 | static int vhost_vdpa_alloc_domain(struct vhost_vdpa *v) | |
679 | { | |
680 | struct vdpa_device *vdpa = v->vdpa; | |
681 | const struct vdpa_config_ops *ops = vdpa->config; | |
682 | struct device *dma_dev = vdpa_get_dma_dev(vdpa); | |
683 | struct bus_type *bus; | |
684 | int ret; | |
685 | ||
686 | /* Device want to do DMA by itself */ | |
687 | if (ops->set_map || ops->dma_map) | |
688 | return 0; | |
689 | ||
690 | bus = dma_dev->bus; | |
691 | if (!bus) | |
692 | return -EFAULT; | |
693 | ||
694 | if (!iommu_capable(bus, IOMMU_CAP_CACHE_COHERENCY)) | |
695 | return -ENOTSUPP; | |
696 | ||
697 | v->domain = iommu_domain_alloc(bus); | |
698 | if (!v->domain) | |
699 | return -EIO; | |
700 | ||
701 | ret = iommu_attach_device(v->domain, dma_dev); | |
702 | if (ret) | |
703 | goto err_attach; | |
704 | ||
705 | return 0; | |
706 | ||
707 | err_attach: | |
708 | iommu_domain_free(v->domain); | |
709 | return ret; | |
710 | } | |
711 | ||
712 | static void vhost_vdpa_free_domain(struct vhost_vdpa *v) | |
713 | { | |
714 | struct vdpa_device *vdpa = v->vdpa; | |
715 | struct device *dma_dev = vdpa_get_dma_dev(vdpa); | |
716 | ||
717 | if (v->domain) { | |
718 | iommu_detach_device(v->domain, dma_dev); | |
719 | iommu_domain_free(v->domain); | |
720 | } | |
721 | ||
722 | v->domain = NULL; | |
723 | } | |
724 | ||
725 | static int vhost_vdpa_open(struct inode *inode, struct file *filep) | |
726 | { | |
727 | struct vhost_vdpa *v; | |
728 | struct vhost_dev *dev; | |
729 | struct vhost_virtqueue **vqs; | |
730 | int nvqs, i, r, opened; | |
731 | ||
732 | v = container_of(inode->i_cdev, struct vhost_vdpa, cdev); | |
4c8cf318 TB |
733 | |
734 | opened = atomic_cmpxchg(&v->opened, 0, 1); | |
735 | if (opened) | |
736 | return -EBUSY; | |
737 | ||
738 | nvqs = v->nvqs; | |
739 | vhost_vdpa_reset(v); | |
740 | ||
741 | vqs = kmalloc_array(nvqs, sizeof(*vqs), GFP_KERNEL); | |
742 | if (!vqs) { | |
743 | r = -ENOMEM; | |
744 | goto err; | |
745 | } | |
746 | ||
747 | dev = &v->vdev; | |
748 | for (i = 0; i < nvqs; i++) { | |
749 | vqs[i] = &v->vqs[i]; | |
750 | vqs[i]->handle_kick = handle_vq_kick; | |
751 | } | |
01fcb1cb | 752 | vhost_dev_init(dev, vqs, nvqs, 0, 0, 0, false, |
4c8cf318 TB |
753 | vhost_vdpa_process_iotlb_msg); |
754 | ||
755 | dev->iotlb = vhost_iotlb_alloc(0, 0); | |
756 | if (!dev->iotlb) { | |
757 | r = -ENOMEM; | |
758 | goto err_init_iotlb; | |
759 | } | |
760 | ||
761 | r = vhost_vdpa_alloc_domain(v); | |
762 | if (r) | |
763 | goto err_init_iotlb; | |
764 | ||
765 | filep->private_data = v; | |
766 | ||
767 | return 0; | |
768 | ||
769 | err_init_iotlb: | |
770 | vhost_dev_cleanup(&v->vdev); | |
771 | err: | |
772 | atomic_dec(&v->opened); | |
773 | return r; | |
774 | } | |
775 | ||
2cf1ba9a ZL |
776 | static void vhost_vdpa_clean_irq(struct vhost_vdpa *v) |
777 | { | |
778 | struct vhost_virtqueue *vq; | |
779 | int i; | |
780 | ||
781 | for (i = 0; i < v->nvqs; i++) { | |
782 | vq = &v->vqs[i]; | |
783 | if (vq->call_ctx.producer.irq) | |
784 | irq_bypass_unregister_producer(&vq->call_ctx.producer); | |
785 | } | |
786 | } | |
787 | ||
4c8cf318 TB |
788 | static int vhost_vdpa_release(struct inode *inode, struct file *filep) |
789 | { | |
790 | struct vhost_vdpa *v = filep->private_data; | |
791 | struct vhost_dev *d = &v->vdev; | |
792 | ||
793 | mutex_lock(&d->mutex); | |
794 | filep->private_data = NULL; | |
795 | vhost_vdpa_reset(v); | |
796 | vhost_dev_stop(&v->vdev); | |
797 | vhost_vdpa_iotlb_free(v); | |
798 | vhost_vdpa_free_domain(v); | |
776f3950 | 799 | vhost_vdpa_config_put(v); |
2cf1ba9a | 800 | vhost_vdpa_clean_irq(v); |
4c8cf318 TB |
801 | vhost_dev_cleanup(&v->vdev); |
802 | kfree(v->vdev.vqs); | |
803 | mutex_unlock(&d->mutex); | |
804 | ||
805 | atomic_dec(&v->opened); | |
806 | complete(&v->completion); | |
807 | ||
808 | return 0; | |
809 | } | |
810 | ||
4b4e4867 | 811 | #ifdef CONFIG_MMU |
ddd89d0a JW |
812 | static vm_fault_t vhost_vdpa_fault(struct vm_fault *vmf) |
813 | { | |
814 | struct vhost_vdpa *v = vmf->vma->vm_file->private_data; | |
815 | struct vdpa_device *vdpa = v->vdpa; | |
816 | const struct vdpa_config_ops *ops = vdpa->config; | |
817 | struct vdpa_notification_area notify; | |
818 | struct vm_area_struct *vma = vmf->vma; | |
819 | u16 index = vma->vm_pgoff; | |
820 | ||
821 | notify = ops->get_vq_notification(vdpa, index); | |
822 | ||
823 | vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot); | |
824 | if (remap_pfn_range(vma, vmf->address & PAGE_MASK, | |
825 | notify.addr >> PAGE_SHIFT, PAGE_SIZE, | |
826 | vma->vm_page_prot)) | |
827 | return VM_FAULT_SIGBUS; | |
828 | ||
829 | return VM_FAULT_NOPAGE; | |
830 | } | |
831 | ||
832 | static const struct vm_operations_struct vhost_vdpa_vm_ops = { | |
833 | .fault = vhost_vdpa_fault, | |
834 | }; | |
835 | ||
836 | static int vhost_vdpa_mmap(struct file *file, struct vm_area_struct *vma) | |
837 | { | |
838 | struct vhost_vdpa *v = vma->vm_file->private_data; | |
839 | struct vdpa_device *vdpa = v->vdpa; | |
840 | const struct vdpa_config_ops *ops = vdpa->config; | |
841 | struct vdpa_notification_area notify; | |
c09cc2c3 | 842 | unsigned long index = vma->vm_pgoff; |
ddd89d0a JW |
843 | |
844 | if (vma->vm_end - vma->vm_start != PAGE_SIZE) | |
845 | return -EINVAL; | |
846 | if ((vma->vm_flags & VM_SHARED) == 0) | |
847 | return -EINVAL; | |
848 | if (vma->vm_flags & VM_READ) | |
849 | return -EINVAL; | |
850 | if (index > 65535) | |
851 | return -EINVAL; | |
852 | if (!ops->get_vq_notification) | |
853 | return -ENOTSUPP; | |
854 | ||
855 | /* To be safe and easily modelled by userspace, We only | |
856 | * support the doorbell which sits on the page boundary and | |
857 | * does not share the page with other registers. | |
858 | */ | |
859 | notify = ops->get_vq_notification(vdpa, index); | |
860 | if (notify.addr & (PAGE_SIZE - 1)) | |
861 | return -EINVAL; | |
862 | if (vma->vm_end - vma->vm_start != notify.size) | |
863 | return -ENOTSUPP; | |
864 | ||
865 | vma->vm_ops = &vhost_vdpa_vm_ops; | |
866 | return 0; | |
867 | } | |
4b4e4867 | 868 | #endif /* CONFIG_MMU */ |
ddd89d0a | 869 | |
4c8cf318 TB |
870 | static const struct file_operations vhost_vdpa_fops = { |
871 | .owner = THIS_MODULE, | |
872 | .open = vhost_vdpa_open, | |
873 | .release = vhost_vdpa_release, | |
874 | .write_iter = vhost_vdpa_chr_write_iter, | |
875 | .unlocked_ioctl = vhost_vdpa_unlocked_ioctl, | |
4b4e4867 | 876 | #ifdef CONFIG_MMU |
ddd89d0a | 877 | .mmap = vhost_vdpa_mmap, |
4b4e4867 | 878 | #endif /* CONFIG_MMU */ |
4c8cf318 TB |
879 | .compat_ioctl = compat_ptr_ioctl, |
880 | }; | |
881 | ||
882 | static void vhost_vdpa_release_dev(struct device *device) | |
883 | { | |
884 | struct vhost_vdpa *v = | |
885 | container_of(device, struct vhost_vdpa, dev); | |
886 | ||
887 | ida_simple_remove(&vhost_vdpa_ida, v->minor); | |
888 | kfree(v->vqs); | |
889 | kfree(v); | |
890 | } | |
891 | ||
892 | static int vhost_vdpa_probe(struct vdpa_device *vdpa) | |
893 | { | |
894 | const struct vdpa_config_ops *ops = vdpa->config; | |
895 | struct vhost_vdpa *v; | |
896 | int minor, nvqs = VHOST_VDPA_VQ_MAX; | |
897 | int r; | |
898 | ||
899 | /* Currently, we only accept the network devices. */ | |
900 | if (ops->get_device_id(vdpa) != VIRTIO_ID_NET) | |
901 | return -ENOTSUPP; | |
902 | ||
903 | v = kzalloc(sizeof(*v), GFP_KERNEL | __GFP_RETRY_MAYFAIL); | |
904 | if (!v) | |
905 | return -ENOMEM; | |
906 | ||
907 | minor = ida_simple_get(&vhost_vdpa_ida, 0, | |
908 | VHOST_VDPA_DEV_MAX, GFP_KERNEL); | |
909 | if (minor < 0) { | |
910 | kfree(v); | |
911 | return minor; | |
912 | } | |
913 | ||
914 | atomic_set(&v->opened, 0); | |
915 | v->minor = minor; | |
916 | v->vdpa = vdpa; | |
917 | v->nvqs = nvqs; | |
918 | v->virtio_id = ops->get_device_id(vdpa); | |
919 | ||
920 | device_initialize(&v->dev); | |
921 | v->dev.release = vhost_vdpa_release_dev; | |
922 | v->dev.parent = &vdpa->dev; | |
923 | v->dev.devt = MKDEV(MAJOR(vhost_vdpa_major), minor); | |
924 | v->vqs = kmalloc_array(nvqs, sizeof(struct vhost_virtqueue), | |
925 | GFP_KERNEL); | |
926 | if (!v->vqs) { | |
927 | r = -ENOMEM; | |
928 | goto err; | |
929 | } | |
930 | ||
931 | r = dev_set_name(&v->dev, "vhost-vdpa-%u", minor); | |
932 | if (r) | |
933 | goto err; | |
934 | ||
935 | cdev_init(&v->cdev, &vhost_vdpa_fops); | |
936 | v->cdev.owner = THIS_MODULE; | |
937 | ||
938 | r = cdev_device_add(&v->cdev, &v->dev); | |
939 | if (r) | |
940 | goto err; | |
941 | ||
942 | init_completion(&v->completion); | |
943 | vdpa_set_drvdata(vdpa, v); | |
944 | ||
945 | return 0; | |
946 | ||
947 | err: | |
948 | put_device(&v->dev); | |
949 | return r; | |
950 | } | |
951 | ||
952 | static void vhost_vdpa_remove(struct vdpa_device *vdpa) | |
953 | { | |
954 | struct vhost_vdpa *v = vdpa_get_drvdata(vdpa); | |
955 | int opened; | |
956 | ||
957 | cdev_device_del(&v->cdev, &v->dev); | |
958 | ||
959 | do { | |
960 | opened = atomic_cmpxchg(&v->opened, 0, 1); | |
961 | if (!opened) | |
962 | break; | |
963 | wait_for_completion(&v->completion); | |
964 | } while (1); | |
965 | ||
966 | put_device(&v->dev); | |
967 | } | |
968 | ||
969 | static struct vdpa_driver vhost_vdpa_driver = { | |
970 | .driver = { | |
971 | .name = "vhost_vdpa", | |
972 | }, | |
973 | .probe = vhost_vdpa_probe, | |
974 | .remove = vhost_vdpa_remove, | |
975 | }; | |
976 | ||
977 | static int __init vhost_vdpa_init(void) | |
978 | { | |
979 | int r; | |
980 | ||
981 | r = alloc_chrdev_region(&vhost_vdpa_major, 0, VHOST_VDPA_DEV_MAX, | |
982 | "vhost-vdpa"); | |
983 | if (r) | |
984 | goto err_alloc_chrdev; | |
985 | ||
986 | r = vdpa_register_driver(&vhost_vdpa_driver); | |
987 | if (r) | |
988 | goto err_vdpa_register_driver; | |
989 | ||
990 | return 0; | |
991 | ||
992 | err_vdpa_register_driver: | |
993 | unregister_chrdev_region(vhost_vdpa_major, VHOST_VDPA_DEV_MAX); | |
994 | err_alloc_chrdev: | |
995 | return r; | |
996 | } | |
997 | module_init(vhost_vdpa_init); | |
998 | ||
999 | static void __exit vhost_vdpa_exit(void) | |
1000 | { | |
1001 | vdpa_unregister_driver(&vhost_vdpa_driver); | |
1002 | unregister_chrdev_region(vhost_vdpa_major, VHOST_VDPA_DEV_MAX); | |
1003 | } | |
1004 | module_exit(vhost_vdpa_exit); | |
1005 | ||
1006 | MODULE_VERSION("0.0.1"); | |
1007 | MODULE_LICENSE("GPL v2"); | |
1008 | MODULE_AUTHOR("Intel Corporation"); | |
1009 | MODULE_DESCRIPTION("vDPA-based vhost backend for virtio"); |