xref: /dpdk/lib/vhost/vdpa.c (revision 665b49c51639a10c553433bc2bcd85c7331c631e)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2018 Intel Corporation
3  */
4 
5 /**
6  * @file
7  *
8  * Device specific vhost lib
9  */
10 
11 #include <sys/queue.h>
12 
13 #include <dev_driver.h>
14 #include <rte_class.h>
15 #include <rte_malloc.h>
16 #include <rte_spinlock.h>
17 #include <rte_tailq.h>
18 
19 #include "rte_vdpa.h"
20 #include "vdpa_driver.h"
21 #include "vhost.h"
22 
23 /** Double linked list of vDPA devices. */
24 TAILQ_HEAD(vdpa_device_list, rte_vdpa_device);
25 
26 static struct vdpa_device_list vdpa_device_list__ =
27 	TAILQ_HEAD_INITIALIZER(vdpa_device_list__);
28 static rte_spinlock_t vdpa_device_list_lock = RTE_SPINLOCK_INITIALIZER;
29 static struct vdpa_device_list * const vdpa_device_list
30 	__rte_guarded_by(&vdpa_device_list_lock) = &vdpa_device_list__;
31 
32 static struct rte_vdpa_device *
33 __vdpa_find_device_by_name(const char *name)
34 	__rte_exclusive_locks_required(&vdpa_device_list_lock)
35 {
36 	struct rte_vdpa_device *dev, *ret = NULL;
37 
38 	if (name == NULL)
39 		return NULL;
40 
41 	TAILQ_FOREACH(dev, vdpa_device_list, next) {
42 		if (!strncmp(dev->device->name, name, RTE_DEV_NAME_MAX_LEN)) {
43 			ret = dev;
44 			break;
45 		}
46 	}
47 
48 	return ret;
49 }
50 
51 struct rte_vdpa_device *
52 rte_vdpa_find_device_by_name(const char *name)
53 {
54 	struct rte_vdpa_device *dev;
55 
56 	rte_spinlock_lock(&vdpa_device_list_lock);
57 	dev = __vdpa_find_device_by_name(name);
58 	rte_spinlock_unlock(&vdpa_device_list_lock);
59 
60 	return dev;
61 }
62 
63 struct rte_device *
64 rte_vdpa_get_rte_device(struct rte_vdpa_device *vdpa_dev)
65 {
66 	if (vdpa_dev == NULL)
67 		return NULL;
68 
69 	return vdpa_dev->device;
70 }
71 
72 struct rte_vdpa_device *
73 rte_vdpa_register_device(struct rte_device *rte_dev,
74 		struct rte_vdpa_dev_ops *ops)
75 {
76 	struct rte_vdpa_device *dev;
77 	int ret = 0;
78 
79 	if (ops == NULL)
80 		return NULL;
81 
82 	/* Check mandatory ops are implemented */
83 	if (!ops->get_queue_num || !ops->get_features ||
84 			!ops->get_protocol_features || !ops->dev_conf ||
85 			!ops->dev_close || !ops->set_vring_state ||
86 			!ops->set_features) {
87 		VHOST_LOG_CONFIG(rte_dev->name, ERR,
88 			"Some mandatory vDPA ops aren't implemented\n");
89 		return NULL;
90 	}
91 
92 	rte_spinlock_lock(&vdpa_device_list_lock);
93 	/* Check the device hasn't been register already */
94 	dev = __vdpa_find_device_by_name(rte_dev->name);
95 	if (dev) {
96 		dev = NULL;
97 		goto out_unlock;
98 	}
99 
100 	dev = rte_zmalloc(NULL, sizeof(*dev), 0);
101 	if (!dev)
102 		goto out_unlock;
103 
104 	dev->device = rte_dev;
105 	dev->ops = ops;
106 
107 	if (ops->get_dev_type) {
108 		ret = ops->get_dev_type(dev, &dev->type);
109 		if (ret) {
110 			VHOST_LOG_CONFIG(rte_dev->name, ERR,
111 					 "Failed to get vdpa dev type.\n");
112 			ret = -1;
113 			goto out_unlock;
114 		}
115 	} else {
116 		/** by default, we assume vdpa device is a net device */
117 		dev->type = RTE_VHOST_VDPA_DEVICE_TYPE_NET;
118 	}
119 
120 	TAILQ_INSERT_TAIL(vdpa_device_list, dev, next);
121 out_unlock:
122 	rte_spinlock_unlock(&vdpa_device_list_lock);
123 
124 	return dev;
125 }
126 
127 int
128 rte_vdpa_unregister_device(struct rte_vdpa_device *dev)
129 {
130 	struct rte_vdpa_device *cur_dev, *tmp_dev;
131 	int ret = -1;
132 
133 	rte_spinlock_lock(&vdpa_device_list_lock);
134 	RTE_TAILQ_FOREACH_SAFE(cur_dev, vdpa_device_list, next, tmp_dev) {
135 		if (dev != cur_dev)
136 			continue;
137 
138 		TAILQ_REMOVE(vdpa_device_list, dev, next);
139 		rte_free(dev);
140 		ret = 0;
141 		break;
142 	}
143 	rte_spinlock_unlock(&vdpa_device_list_lock);
144 
145 	return ret;
146 }
147 
148 int
149 rte_vdpa_relay_vring_used(int vid, uint16_t qid, void *vring_m)
150 	__rte_no_thread_safety_analysis /* FIXME: requires iotlb_lock? */
151 {
152 	struct virtio_net *dev = get_device(vid);
153 	uint16_t idx, idx_m, desc_id;
154 	struct vhost_virtqueue *vq;
155 	struct vring_desc desc;
156 	struct vring_desc *desc_ring;
157 	struct vring_desc *idesc = NULL;
158 	struct vring *s_vring;
159 	uint64_t dlen;
160 	uint32_t nr_descs;
161 	int ret;
162 
163 	if (!dev || !vring_m)
164 		return -1;
165 
166 	if (qid >= dev->nr_vring)
167 		return -1;
168 
169 	if (vq_is_packed(dev))
170 		return -1;
171 
172 	s_vring = (struct vring *)vring_m;
173 	vq = dev->virtqueue[qid];
174 	idx = vq->used->idx;
175 	idx_m = s_vring->used->idx;
176 	ret = (uint16_t)(idx_m - idx);
177 
178 	while (idx != idx_m) {
179 		/* copy used entry, used ring logging is not covered here */
180 		vq->used->ring[idx & (vq->size - 1)] =
181 			s_vring->used->ring[idx & (vq->size - 1)];
182 
183 		desc_id = vq->used->ring[idx & (vq->size - 1)].id;
184 		desc_ring = vq->desc;
185 		nr_descs = vq->size;
186 
187 		if (unlikely(desc_id >= vq->size))
188 			return -1;
189 
190 		if (vq->desc[desc_id].flags & VRING_DESC_F_INDIRECT) {
191 			dlen = vq->desc[desc_id].len;
192 			nr_descs = dlen / sizeof(struct vring_desc);
193 			if (unlikely(nr_descs > vq->size))
194 				return -1;
195 
196 			desc_ring = (struct vring_desc *)(uintptr_t)
197 				vhost_iova_to_vva(dev, vq,
198 						vq->desc[desc_id].addr, &dlen,
199 						VHOST_ACCESS_RO);
200 			if (unlikely(!desc_ring))
201 				return -1;
202 
203 			if (unlikely(dlen < vq->desc[desc_id].len)) {
204 				idesc = vhost_alloc_copy_ind_table(dev, vq,
205 						vq->desc[desc_id].addr,
206 						vq->desc[desc_id].len);
207 				if (unlikely(!idesc))
208 					return -1;
209 
210 				desc_ring = idesc;
211 			}
212 
213 			desc_id = 0;
214 		}
215 
216 		/* dirty page logging for DMA writeable buffer */
217 		do {
218 			if (unlikely(desc_id >= vq->size))
219 				goto fail;
220 			if (unlikely(nr_descs-- == 0))
221 				goto fail;
222 			desc = desc_ring[desc_id];
223 			if (desc.flags & VRING_DESC_F_WRITE)
224 				vhost_log_write_iova(dev, vq, desc.addr,
225 						     desc.len);
226 			desc_id = desc.next;
227 		} while (desc.flags & VRING_DESC_F_NEXT);
228 
229 		if (unlikely(idesc)) {
230 			free_ind_table(idesc);
231 			idesc = NULL;
232 		}
233 
234 		idx++;
235 	}
236 
237 	/* used idx is the synchronization point for the split vring */
238 	__atomic_store_n(&vq->used->idx, idx_m, __ATOMIC_RELEASE);
239 
240 	if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
241 		vring_used_event(s_vring) = idx_m;
242 
243 	return ret;
244 
245 fail:
246 	if (unlikely(idesc))
247 		free_ind_table(idesc);
248 	return -1;
249 }
250 
251 int
252 rte_vdpa_get_queue_num(struct rte_vdpa_device *dev, uint32_t *queue_num)
253 {
254 	if (dev == NULL || dev->ops == NULL || dev->ops->get_queue_num == NULL)
255 		return -1;
256 
257 	return dev->ops->get_queue_num(dev, queue_num);
258 }
259 
260 int
261 rte_vdpa_get_features(struct rte_vdpa_device *dev, uint64_t *features)
262 {
263 	if (dev == NULL || dev->ops == NULL || dev->ops->get_features == NULL)
264 		return -1;
265 
266 	return dev->ops->get_features(dev, features);
267 }
268 
269 int
270 rte_vdpa_get_protocol_features(struct rte_vdpa_device *dev, uint64_t *features)
271 {
272 	if (dev == NULL || dev->ops == NULL ||
273 			dev->ops->get_protocol_features == NULL)
274 		return -1;
275 
276 	return dev->ops->get_protocol_features(dev, features);
277 }
278 
279 int
280 rte_vdpa_get_stats_names(struct rte_vdpa_device *dev,
281 		struct rte_vdpa_stat_name *stats_names,
282 		unsigned int size)
283 {
284 	if (!dev)
285 		return -EINVAL;
286 
287 	if (dev->ops->get_stats_names == NULL)
288 		return -ENOTSUP;
289 
290 	return dev->ops->get_stats_names(dev, stats_names, size);
291 }
292 
293 int
294 rte_vdpa_get_stats(struct rte_vdpa_device *dev, uint16_t qid,
295 		struct rte_vdpa_stat *stats, unsigned int n)
296 {
297 	if (!dev || !stats || !n)
298 		return -EINVAL;
299 
300 	if (dev->ops->get_stats == NULL)
301 		return -ENOTSUP;
302 
303 	return dev->ops->get_stats(dev, qid, stats, n);
304 }
305 
306 int
307 rte_vdpa_reset_stats(struct rte_vdpa_device *dev, uint16_t qid)
308 {
309 	if (!dev)
310 		return -EINVAL;
311 
312 	if (dev->ops->reset_stats == NULL)
313 		return -ENOTSUP;
314 
315 	return dev->ops->reset_stats(dev, qid);
316 }
317 
318 static int
319 vdpa_dev_match(struct rte_vdpa_device *dev,
320 	      const struct rte_device *rte_dev)
321 {
322 	if (dev->device == rte_dev)
323 		return 0;
324 
325 	return -1;
326 }
327 
328 /* Generic rte_vdpa_dev comparison function. */
329 typedef int (*rte_vdpa_cmp_t)(struct rte_vdpa_device *,
330 		const struct rte_device *rte_dev);
331 
332 static struct rte_vdpa_device *
333 vdpa_find_device(const struct rte_vdpa_device *start, rte_vdpa_cmp_t cmp,
334 		struct rte_device *rte_dev)
335 {
336 	struct rte_vdpa_device *dev;
337 
338 	rte_spinlock_lock(&vdpa_device_list_lock);
339 	if (start == NULL)
340 		dev = TAILQ_FIRST(vdpa_device_list);
341 	else
342 		dev = TAILQ_NEXT(start, next);
343 
344 	while (dev != NULL) {
345 		if (cmp(dev, rte_dev) == 0)
346 			break;
347 
348 		dev = TAILQ_NEXT(dev, next);
349 	}
350 	rte_spinlock_unlock(&vdpa_device_list_lock);
351 
352 	return dev;
353 }
354 
355 static void *
356 vdpa_dev_iterate(const void *start,
357 		const char *str,
358 		const struct rte_dev_iterator *it)
359 {
360 	struct rte_vdpa_device *vdpa_dev = NULL;
361 
362 	RTE_SET_USED(str);
363 
364 	vdpa_dev = vdpa_find_device(start, vdpa_dev_match, it->device);
365 
366 	return vdpa_dev;
367 }
368 
369 static struct rte_class rte_class_vdpa = {
370 	.dev_iterate = vdpa_dev_iterate,
371 };
372 
373 RTE_REGISTER_CLASS(vdpa, rte_class_vdpa);
374