net/virtio: add device config support to vDPA
[dpdk.git] / drivers / net / virtio / virtio_user / vhost_vdpa.c
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2020 Red Hat Inc.
3  */
4
5 #include <sys/ioctl.h>
6 #include <sys/types.h>
7 #include <sys/stat.h>
8 #include <fcntl.h>
9 #include <unistd.h>
10
11 #include <rte_memory.h>
12
13 #include "vhost.h"
14 #include "virtio_user_dev.h"
15
16 struct vhost_vdpa_data {
17         int vhostfd;
18         uint64_t protocol_features;
19 };
20
21 #define VHOST_VDPA_SUPPORTED_BACKEND_FEATURES           \
22         (1ULL << VHOST_BACKEND_F_IOTLB_MSG_V2   |       \
23         1ULL << VHOST_BACKEND_F_IOTLB_BATCH)
24
25 /* vhost kernel & vdpa ioctls */
26 #define VHOST_VIRTIO 0xAF
27 #define VHOST_GET_FEATURES _IOR(VHOST_VIRTIO, 0x00, __u64)
28 #define VHOST_SET_FEATURES _IOW(VHOST_VIRTIO, 0x00, __u64)
29 #define VHOST_SET_OWNER _IO(VHOST_VIRTIO, 0x01)
30 #define VHOST_RESET_OWNER _IO(VHOST_VIRTIO, 0x02)
31 #define VHOST_SET_LOG_BASE _IOW(VHOST_VIRTIO, 0x04, __u64)
32 #define VHOST_SET_LOG_FD _IOW(VHOST_VIRTIO, 0x07, int)
33 #define VHOST_SET_VRING_NUM _IOW(VHOST_VIRTIO, 0x10, struct vhost_vring_state)
34 #define VHOST_SET_VRING_ADDR _IOW(VHOST_VIRTIO, 0x11, struct vhost_vring_addr)
35 #define VHOST_SET_VRING_BASE _IOW(VHOST_VIRTIO, 0x12, struct vhost_vring_state)
36 #define VHOST_GET_VRING_BASE _IOWR(VHOST_VIRTIO, 0x12, struct vhost_vring_state)
37 #define VHOST_SET_VRING_KICK _IOW(VHOST_VIRTIO, 0x20, struct vhost_vring_file)
38 #define VHOST_SET_VRING_CALL _IOW(VHOST_VIRTIO, 0x21, struct vhost_vring_file)
39 #define VHOST_SET_VRING_ERR _IOW(VHOST_VIRTIO, 0x22, struct vhost_vring_file)
40 #define VHOST_NET_SET_BACKEND _IOW(VHOST_VIRTIO, 0x30, struct vhost_vring_file)
41 #define VHOST_VDPA_GET_DEVICE_ID _IOR(VHOST_VIRTIO, 0x70, __u32)
42 #define VHOST_VDPA_GET_STATUS _IOR(VHOST_VIRTIO, 0x71, __u8)
43 #define VHOST_VDPA_SET_STATUS _IOW(VHOST_VIRTIO, 0x72, __u8)
44 #define VHOST_VDPA_GET_CONFIG _IOR(VHOST_VIRTIO, 0x73, struct vhost_vdpa_config)
45 #define VHOST_VDPA_SET_CONFIG _IOW(VHOST_VIRTIO, 0x74, struct vhost_vdpa_config)
46 #define VHOST_VDPA_SET_VRING_ENABLE _IOW(VHOST_VIRTIO, 0x75, struct vhost_vring_state)
47 #define VHOST_SET_BACKEND_FEATURES _IOW(VHOST_VIRTIO, 0x25, __u64)
48 #define VHOST_GET_BACKEND_FEATURES _IOR(VHOST_VIRTIO, 0x26, __u64)
49
50 /* no alignment requirement */
51 struct vhost_iotlb_msg {
52         uint64_t iova;
53         uint64_t size;
54         uint64_t uaddr;
55 #define VHOST_ACCESS_RO      0x1
56 #define VHOST_ACCESS_WO      0x2
57 #define VHOST_ACCESS_RW      0x3
58         uint8_t perm;
59 #define VHOST_IOTLB_MISS           1
60 #define VHOST_IOTLB_UPDATE         2
61 #define VHOST_IOTLB_INVALIDATE     3
62 #define VHOST_IOTLB_ACCESS_FAIL    4
63 #define VHOST_IOTLB_BATCH_BEGIN    5
64 #define VHOST_IOTLB_BATCH_END      6
65         uint8_t type;
66 };
67
68 #define VHOST_IOTLB_MSG_V2 0x2
69
70 struct vhost_vdpa_config {
71         uint32_t off;
72         uint32_t len;
73         uint8_t buf[0];
74 };
75
76 struct vhost_msg {
77         uint32_t type;
78         uint32_t reserved;
79         union {
80                 struct vhost_iotlb_msg iotlb;
81                 uint8_t padding[64];
82         };
83 };
84
85
86 static int
87 vhost_vdpa_ioctl(int fd, uint64_t request, void *arg)
88 {
89         int ret;
90
91         ret = ioctl(fd, request, arg);
92         if (ret) {
93                 PMD_DRV_LOG(ERR, "Vhost-vDPA ioctl %"PRIu64" failed (%s)",
94                                 request, strerror(errno));
95                 return -1;
96         }
97
98         return 0;
99 }
100
101 static int
102 vhost_vdpa_set_owner(struct virtio_user_dev *dev)
103 {
104         struct vhost_vdpa_data *data = dev->backend_data;
105
106         return vhost_vdpa_ioctl(data->vhostfd, VHOST_SET_OWNER, NULL);
107 }
108
109 static int
110 vhost_vdpa_get_protocol_features(struct virtio_user_dev *dev, uint64_t *features)
111 {
112         struct vhost_vdpa_data *data = dev->backend_data;
113
114         return vhost_vdpa_ioctl(data->vhostfd, VHOST_GET_BACKEND_FEATURES, features);
115 }
116
117 static int
118 vhost_vdpa_set_protocol_features(struct virtio_user_dev *dev, uint64_t features)
119 {
120         struct vhost_vdpa_data *data = dev->backend_data;
121
122         return vhost_vdpa_ioctl(data->vhostfd, VHOST_SET_BACKEND_FEATURES, &features);
123 }
124
125 static int
126 vhost_vdpa_get_features(struct virtio_user_dev *dev, uint64_t *features)
127 {
128         struct vhost_vdpa_data *data = dev->backend_data;
129         int ret;
130
131         ret = vhost_vdpa_ioctl(data->vhostfd, VHOST_GET_FEATURES, features);
132         if (ret) {
133                 PMD_DRV_LOG(ERR, "Failed to get features");
134                 return -1;
135         }
136
137         /* Multiqueue not supported for now */
138         *features &= ~(1ULL << VIRTIO_NET_F_MQ);
139
140         /* Negotiated vDPA backend features */
141         ret = vhost_vdpa_get_protocol_features(dev, &data->protocol_features);
142         if (ret < 0) {
143                 PMD_DRV_LOG(ERR, "Failed to get backend features");
144                 return -1;
145         }
146
147         data->protocol_features &= VHOST_VDPA_SUPPORTED_BACKEND_FEATURES;
148
149         ret = vhost_vdpa_set_protocol_features(dev, data->protocol_features);
150         if (ret < 0) {
151                 PMD_DRV_LOG(ERR, "Failed to set backend features");
152                 return -1;
153         }
154
155         return 0;
156 }
157
158 static int
159 vhost_vdpa_set_features(struct virtio_user_dev *dev, uint64_t features)
160 {
161         struct vhost_vdpa_data *data = dev->backend_data;
162
163         /* WORKAROUND */
164         features |= 1ULL << VIRTIO_F_IOMMU_PLATFORM;
165
166         return vhost_vdpa_ioctl(data->vhostfd, VHOST_SET_FEATURES, &features);
167 }
168
169 static int
170 vhost_vdpa_iotlb_batch_begin(struct virtio_user_dev *dev)
171 {
172         struct vhost_vdpa_data *data = dev->backend_data;
173         struct vhost_msg msg = {};
174
175         if (!(data->protocol_features & (1ULL << VHOST_BACKEND_F_IOTLB_BATCH)))
176                 return 0;
177
178         if (!(data->protocol_features & (1ULL << VHOST_BACKEND_F_IOTLB_MSG_V2))) {
179                 PMD_DRV_LOG(ERR, "IOTLB_MSG_V2 not supported by the backend.");
180                 return -1;
181         }
182
183         msg.type = VHOST_IOTLB_MSG_V2;
184         msg.iotlb.type = VHOST_IOTLB_BATCH_BEGIN;
185
186         if (write(data->vhostfd, &msg, sizeof(msg)) != sizeof(msg)) {
187                 PMD_DRV_LOG(ERR, "Failed to send IOTLB batch begin (%s)",
188                                 strerror(errno));
189                 return -1;
190         }
191
192         return 0;
193 }
194
195 static int
196 vhost_vdpa_iotlb_batch_end(struct virtio_user_dev *dev)
197 {
198         struct vhost_vdpa_data *data = dev->backend_data;
199         struct vhost_msg msg = {};
200
201         if (!(data->protocol_features & (1ULL << VHOST_BACKEND_F_IOTLB_BATCH)))
202                 return 0;
203
204         if (!(data->protocol_features & (1ULL << VHOST_BACKEND_F_IOTLB_MSG_V2))) {
205                 PMD_DRV_LOG(ERR, "IOTLB_MSG_V2 not supported by the backend.");
206                 return -1;
207         }
208
209         msg.type = VHOST_IOTLB_MSG_V2;
210         msg.iotlb.type = VHOST_IOTLB_BATCH_END;
211
212         if (write(data->vhostfd, &msg, sizeof(msg)) != sizeof(msg)) {
213                 PMD_DRV_LOG(ERR, "Failed to send IOTLB batch end (%s)",
214                                 strerror(errno));
215                 return -1;
216         }
217
218         return 0;
219 }
220
221 static int
222 vhost_vdpa_dma_map(struct virtio_user_dev *dev, void *addr,
223                                   uint64_t iova, size_t len)
224 {
225         struct vhost_vdpa_data *data = dev->backend_data;
226         struct vhost_msg msg = {};
227
228         if (!(data->protocol_features & (1ULL << VHOST_BACKEND_F_IOTLB_MSG_V2))) {
229                 PMD_DRV_LOG(ERR, "IOTLB_MSG_V2 not supported by the backend.");
230                 return -1;
231         }
232
233         msg.type = VHOST_IOTLB_MSG_V2;
234         msg.iotlb.type = VHOST_IOTLB_UPDATE;
235         msg.iotlb.iova = iova;
236         msg.iotlb.uaddr = (uint64_t)(uintptr_t)addr;
237         msg.iotlb.size = len;
238         msg.iotlb.perm = VHOST_ACCESS_RW;
239
240         PMD_DRV_LOG(DEBUG, "%s: iova: 0x%" PRIx64 ", addr: %p, len: 0x%zx",
241                         __func__, iova, addr, len);
242
243         if (write(data->vhostfd, &msg, sizeof(msg)) != sizeof(msg)) {
244                 PMD_DRV_LOG(ERR, "Failed to send IOTLB update (%s)",
245                                 strerror(errno));
246                 return -1;
247         }
248
249         return 0;
250 }
251
252 static int
253 vhost_vdpa_dma_unmap(struct virtio_user_dev *dev, __rte_unused void *addr,
254                                   uint64_t iova, size_t len)
255 {
256         struct vhost_vdpa_data *data = dev->backend_data;
257         struct vhost_msg msg = {};
258
259         if (!(data->protocol_features & (1ULL << VHOST_BACKEND_F_IOTLB_MSG_V2))) {
260                 PMD_DRV_LOG(ERR, "IOTLB_MSG_V2 not supported by the backend.");
261                 return -1;
262         }
263
264         msg.type = VHOST_IOTLB_MSG_V2;
265         msg.iotlb.type = VHOST_IOTLB_INVALIDATE;
266         msg.iotlb.iova = iova;
267         msg.iotlb.size = len;
268
269         PMD_DRV_LOG(DEBUG, "%s: iova: 0x%" PRIx64 ", len: 0x%zx",
270                         __func__, iova, len);
271
272         if (write(data->vhostfd, &msg, sizeof(msg)) != sizeof(msg)) {
273                 PMD_DRV_LOG(ERR, "Failed to send IOTLB invalidate (%s)",
274                                 strerror(errno));
275                 return -1;
276         }
277
278         return 0;
279 }
280
281 static int
282 vhost_vdpa_dma_map_batch(struct virtio_user_dev *dev, void *addr,
283                                   uint64_t iova, size_t len)
284 {
285         int ret;
286
287         if (vhost_vdpa_iotlb_batch_begin(dev) < 0)
288                 return -1;
289
290         ret = vhost_vdpa_dma_map(dev, addr, iova, len);
291
292         if (vhost_vdpa_iotlb_batch_end(dev) < 0)
293                 return -1;
294
295         return ret;
296 }
297
298 static int
299 vhost_vdpa_dma_unmap_batch(struct virtio_user_dev *dev, void *addr,
300                                   uint64_t iova, size_t len)
301 {
302         int ret;
303
304         if (vhost_vdpa_iotlb_batch_begin(dev) < 0)
305                 return -1;
306
307         ret = vhost_vdpa_dma_unmap(dev, addr, iova, len);
308
309         if (vhost_vdpa_iotlb_batch_end(dev) < 0)
310                 return -1;
311
312         return ret;
313 }
314
315 static int
316 vhost_vdpa_map_contig(const struct rte_memseg_list *msl,
317                 const struct rte_memseg *ms, size_t len, void *arg)
318 {
319         struct virtio_user_dev *dev = arg;
320
321         if (msl->external)
322                 return 0;
323
324         return vhost_vdpa_dma_map(dev, ms->addr, ms->iova, len);
325 }
326
327 static int
328 vhost_vdpa_map(const struct rte_memseg_list *msl, const struct rte_memseg *ms,
329                 void *arg)
330 {
331         struct virtio_user_dev *dev = arg;
332
333         /* skip external memory that isn't a heap */
334         if (msl->external && !msl->heap)
335                 return 0;
336
337         /* skip any segments with invalid IOVA addresses */
338         if (ms->iova == RTE_BAD_IOVA)
339                 return 0;
340
341         /* if IOVA mode is VA, we've already mapped the internal segments */
342         if (!msl->external && rte_eal_iova_mode() == RTE_IOVA_VA)
343                 return 0;
344
345         return vhost_vdpa_dma_map(dev, ms->addr, ms->iova, ms->len);
346 }
347
348 static int
349 vhost_vdpa_set_memory_table(struct virtio_user_dev *dev)
350 {
351         int ret;
352
353         if (vhost_vdpa_iotlb_batch_begin(dev) < 0)
354                 return -1;
355
356         vhost_vdpa_dma_unmap(dev, NULL, 0, SIZE_MAX);
357
358         if (rte_eal_iova_mode() == RTE_IOVA_VA) {
359                 /* with IOVA as VA mode, we can get away with mapping contiguous
360                  * chunks rather than going page-by-page.
361                  */
362                 ret = rte_memseg_contig_walk_thread_unsafe(
363                                 vhost_vdpa_map_contig, dev);
364                 if (ret)
365                         goto batch_end;
366                 /* we have to continue the walk because we've skipped the
367                  * external segments during the config walk.
368                  */
369         }
370         ret = rte_memseg_walk_thread_unsafe(vhost_vdpa_map, dev);
371
372 batch_end:
373         if (vhost_vdpa_iotlb_batch_end(dev) < 0)
374                 return -1;
375
376         return ret;
377 }
378
379 static int
380 vhost_vdpa_set_vring_enable(struct virtio_user_dev *dev, struct vhost_vring_state *state)
381 {
382         struct vhost_vdpa_data *data = dev->backend_data;
383
384         return vhost_vdpa_ioctl(data->vhostfd, VHOST_VDPA_SET_VRING_ENABLE, state);
385 }
386
387 static int
388 vhost_vdpa_set_vring_num(struct virtio_user_dev *dev, struct vhost_vring_state *state)
389 {
390         struct vhost_vdpa_data *data = dev->backend_data;
391
392         return vhost_vdpa_ioctl(data->vhostfd, VHOST_SET_VRING_NUM, state);
393 }
394
395 static int
396 vhost_vdpa_set_vring_base(struct virtio_user_dev *dev, struct vhost_vring_state *state)
397 {
398         struct vhost_vdpa_data *data = dev->backend_data;
399
400         return vhost_vdpa_ioctl(data->vhostfd, VHOST_SET_VRING_BASE, state);
401 }
402
403 static int
404 vhost_vdpa_get_vring_base(struct virtio_user_dev *dev, struct vhost_vring_state *state)
405 {
406         struct vhost_vdpa_data *data = dev->backend_data;
407
408         return vhost_vdpa_ioctl(data->vhostfd, VHOST_GET_VRING_BASE, state);
409 }
410
411 static int
412 vhost_vdpa_set_vring_call(struct virtio_user_dev *dev, struct vhost_vring_file *file)
413 {
414         struct vhost_vdpa_data *data = dev->backend_data;
415
416         return vhost_vdpa_ioctl(data->vhostfd, VHOST_SET_VRING_CALL, file);
417 }
418
419 static int
420 vhost_vdpa_set_vring_kick(struct virtio_user_dev *dev, struct vhost_vring_file *file)
421 {
422         struct vhost_vdpa_data *data = dev->backend_data;
423
424         return vhost_vdpa_ioctl(data->vhostfd, VHOST_SET_VRING_KICK, file);
425 }
426
427 static int
428 vhost_vdpa_set_vring_addr(struct virtio_user_dev *dev, struct vhost_vring_addr *addr)
429 {
430         struct vhost_vdpa_data *data = dev->backend_data;
431
432         return vhost_vdpa_ioctl(data->vhostfd, VHOST_SET_VRING_ADDR, addr);
433 }
434
435 static int
436 vhost_vdpa_get_status(struct virtio_user_dev *dev, uint8_t *status)
437 {
438         struct vhost_vdpa_data *data = dev->backend_data;
439
440         return vhost_vdpa_ioctl(data->vhostfd, VHOST_VDPA_GET_STATUS, status);
441 }
442
443 static int
444 vhost_vdpa_set_status(struct virtio_user_dev *dev, uint8_t status)
445 {
446         struct vhost_vdpa_data *data = dev->backend_data;
447
448         return vhost_vdpa_ioctl(data->vhostfd, VHOST_VDPA_SET_STATUS, &status);
449 }
450
451 static int
452 vhost_vdpa_get_config(struct virtio_user_dev *dev, uint8_t *data, uint32_t off, uint32_t len)
453 {
454         struct vhost_vdpa_data *vdpa_data = dev->backend_data;
455         struct vhost_vdpa_config *config;
456         int ret = 0;
457
458         config = malloc(sizeof(*config) + len);
459         if (!config) {
460                 PMD_DRV_LOG(ERR, "Failed to allocate vDPA config data");
461                 return -1;
462         }
463
464         config->off = off;
465         config->len = len;
466
467         ret = vhost_vdpa_ioctl(vdpa_data->vhostfd, VHOST_VDPA_GET_CONFIG, config);
468         if (ret) {
469                 PMD_DRV_LOG(ERR, "Failed to get vDPA config (offset 0x%x, len 0x%x)", off, len);
470                 ret = -1;
471                 goto out;
472         }
473
474         memcpy(data, config->buf, len);
475 out:
476         free(config);
477
478         return ret;
479 }
480
481 static int
482 vhost_vdpa_set_config(struct virtio_user_dev *dev, const uint8_t *data, uint32_t off, uint32_t len)
483 {
484         struct vhost_vdpa_data *vdpa_data = dev->backend_data;
485         struct vhost_vdpa_config *config;
486         int ret = 0;
487
488         config = malloc(sizeof(*config) + len);
489         if (!config) {
490                 PMD_DRV_LOG(ERR, "Failed to allocate vDPA config data");
491                 return -1;
492         }
493
494         config->off = off;
495         config->len = len;
496
497         memcpy(config->buf, data, len);
498
499         ret = vhost_vdpa_ioctl(vdpa_data->vhostfd, VHOST_VDPA_SET_CONFIG, config);
500         if (ret) {
501                 PMD_DRV_LOG(ERR, "Failed to set vDPA config (offset 0x%x, len 0x%x)", off, len);
502                 ret = -1;
503         }
504
505         free(config);
506
507         return ret;
508 }
509
510 /**
511  * Set up environment to talk with a vhost vdpa backend.
512  *
513  * @return
514  *   - (-1) if fail to set up;
515  *   - (>=0) if successful.
516  */
517 static int
518 vhost_vdpa_setup(struct virtio_user_dev *dev)
519 {
520         struct vhost_vdpa_data *data;
521         uint32_t did = (uint32_t)-1;
522
523         data = malloc(sizeof(*data));
524         if (!data) {
525                 PMD_DRV_LOG(ERR, "(%s) Faidle to allocate backend data", dev->path);
526                 return -1;
527         }
528
529         data->vhostfd = open(dev->path, O_RDWR);
530         if (data->vhostfd < 0) {
531                 PMD_DRV_LOG(ERR, "Failed to open %s: %s\n",
532                                 dev->path, strerror(errno));
533                 free(data);
534                 return -1;
535         }
536
537         if (ioctl(data->vhostfd, VHOST_VDPA_GET_DEVICE_ID, &did) < 0 ||
538                         did != VIRTIO_ID_NETWORK) {
539                 PMD_DRV_LOG(ERR, "Invalid vdpa device ID: %u\n", did);
540                 close(data->vhostfd);
541                 free(data);
542                 return -1;
543         }
544
545         dev->backend_data = data;
546
547         return 0;
548 }
549
550 static int
551 vhost_vdpa_destroy(struct virtio_user_dev *dev)
552 {
553         struct vhost_vdpa_data *data = dev->backend_data;
554
555         if (!data)
556                 return 0;
557
558         close(data->vhostfd);
559
560         free(data);
561         dev->backend_data = NULL;
562
563         return 0;
564 }
565
566 static int
567 vhost_vdpa_enable_queue_pair(struct virtio_user_dev *dev,
568                                uint16_t pair_idx,
569                                int enable)
570 {
571         int i;
572
573         if (dev->qp_enabled[pair_idx] == enable)
574                 return 0;
575
576         for (i = 0; i < 2; ++i) {
577                 struct vhost_vring_state state = {
578                         .index = pair_idx * 2 + i,
579                         .num   = enable,
580                 };
581
582                 if (vhost_vdpa_set_vring_enable(dev, &state))
583                         return -1;
584         }
585
586         dev->qp_enabled[pair_idx] = enable;
587
588         return 0;
589 }
590
591 static int
592 vhost_vdpa_get_backend_features(uint64_t *features)
593 {
594         *features = 0;
595
596         return 0;
597 }
598
599 static int
600 vhost_vdpa_update_link_state(struct virtio_user_dev *dev __rte_unused)
601 {
602         /* Nothing to update (for now?) */
603         return 0;
604 }
605
606 static int
607 vhost_vdpa_get_intr_fd(struct virtio_user_dev *dev __rte_unused)
608 {
609         /* No link state interrupt with Vhost-vDPA */
610         return -1;
611 }
612
613 struct virtio_user_backend_ops virtio_ops_vdpa = {
614         .setup = vhost_vdpa_setup,
615         .destroy = vhost_vdpa_destroy,
616         .get_backend_features = vhost_vdpa_get_backend_features,
617         .set_owner = vhost_vdpa_set_owner,
618         .get_features = vhost_vdpa_get_features,
619         .set_features = vhost_vdpa_set_features,
620         .set_memory_table = vhost_vdpa_set_memory_table,
621         .set_vring_num = vhost_vdpa_set_vring_num,
622         .set_vring_base = vhost_vdpa_set_vring_base,
623         .get_vring_base = vhost_vdpa_get_vring_base,
624         .set_vring_call = vhost_vdpa_set_vring_call,
625         .set_vring_kick = vhost_vdpa_set_vring_kick,
626         .set_vring_addr = vhost_vdpa_set_vring_addr,
627         .get_status = vhost_vdpa_get_status,
628         .set_status = vhost_vdpa_set_status,
629         .get_config = vhost_vdpa_get_config,
630         .set_config = vhost_vdpa_set_config,
631         .enable_qp = vhost_vdpa_enable_queue_pair,
632         .dma_map = vhost_vdpa_dma_map_batch,
633         .dma_unmap = vhost_vdpa_dma_unmap_batch,
634         .update_link_state = vhost_vdpa_update_link_state,
635         .get_intr_fd = vhost_vdpa_get_intr_fd,
636 };