net/virtio: split virtio-user start
[dpdk.git] / drivers / net / virtio / virtio_user / virtio_user_dev.c
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2010-2016 Intel Corporation
3  */
4
5 #include <stdint.h>
6 #include <stdio.h>
7 #include <fcntl.h>
8 #include <string.h>
9 #include <errno.h>
10 #include <sys/mman.h>
11 #include <unistd.h>
12 #include <sys/eventfd.h>
13 #include <sys/types.h>
14 #include <sys/stat.h>
15
16 #include <rte_string_fns.h>
17 #include <rte_eal_memconfig.h>
18
19 #include "vhost.h"
20 #include "virtio_user_dev.h"
21 #include "../virtio_ethdev.h"
22
23 #define VIRTIO_USER_MEM_EVENT_CLB_NAME "virtio_user_mem_event_clb"
24
25 static int
26 virtio_user_create_queue(struct virtio_user_dev *dev, uint32_t queue_sel)
27 {
28         /* Of all per virtqueue MSGs, make sure VHOST_SET_VRING_CALL come
29          * firstly because vhost depends on this msg to allocate virtqueue
30          * pair.
31          */
32         struct vhost_vring_file file;
33
34         file.index = queue_sel;
35         file.fd = dev->callfds[queue_sel];
36         dev->ops->send_request(dev, VHOST_USER_SET_VRING_CALL, &file);
37
38         return 0;
39 }
40
41 static int
42 virtio_user_kick_queue(struct virtio_user_dev *dev, uint32_t queue_sel)
43 {
44         struct vhost_vring_file file;
45         struct vhost_vring_state state;
46         struct vring *vring = &dev->vrings[queue_sel];
47         struct vring_packed *pq_vring = &dev->packed_vrings[queue_sel];
48         struct vhost_vring_addr addr = {
49                 .index = queue_sel,
50                 .log_guest_addr = 0,
51                 .flags = 0, /* disable log */
52         };
53
54         if (dev->features & (1ULL << VIRTIO_F_RING_PACKED)) {
55                 addr.desc_user_addr =
56                         (uint64_t)(uintptr_t)pq_vring->desc;
57                 addr.avail_user_addr =
58                         (uint64_t)(uintptr_t)pq_vring->driver;
59                 addr.used_user_addr =
60                         (uint64_t)(uintptr_t)pq_vring->device;
61         } else {
62                 addr.desc_user_addr = (uint64_t)(uintptr_t)vring->desc;
63                 addr.avail_user_addr = (uint64_t)(uintptr_t)vring->avail;
64                 addr.used_user_addr = (uint64_t)(uintptr_t)vring->used;
65         }
66
67         state.index = queue_sel;
68         state.num = vring->num;
69         dev->ops->send_request(dev, VHOST_USER_SET_VRING_NUM, &state);
70
71         state.index = queue_sel;
72         state.num = 0; /* no reservation */
73         if (dev->features & (1ULL << VIRTIO_F_RING_PACKED))
74                 state.num |= (1 << 15);
75         dev->ops->send_request(dev, VHOST_USER_SET_VRING_BASE, &state);
76
77         dev->ops->send_request(dev, VHOST_USER_SET_VRING_ADDR, &addr);
78
79         /* Of all per virtqueue MSGs, make sure VHOST_USER_SET_VRING_KICK comes
80          * lastly because vhost depends on this msg to judge if
81          * virtio is ready.
82          */
83         file.index = queue_sel;
84         file.fd = dev->kickfds[queue_sel];
85         dev->ops->send_request(dev, VHOST_USER_SET_VRING_KICK, &file);
86
87         return 0;
88 }
89
90 static int
91 virtio_user_queue_setup(struct virtio_user_dev *dev,
92                         int (*fn)(struct virtio_user_dev *, uint32_t))
93 {
94         uint32_t i, queue_sel;
95
96         for (i = 0; i < dev->max_queue_pairs; ++i) {
97                 queue_sel = 2 * i + VTNET_SQ_RQ_QUEUE_IDX;
98                 if (fn(dev, queue_sel) < 0) {
99                         PMD_DRV_LOG(INFO, "setup rx vq fails: %u", i);
100                         return -1;
101                 }
102         }
103         for (i = 0; i < dev->max_queue_pairs; ++i) {
104                 queue_sel = 2 * i + VTNET_SQ_TQ_QUEUE_IDX;
105                 if (fn(dev, queue_sel) < 0) {
106                         PMD_DRV_LOG(INFO, "setup tx vq fails: %u", i);
107                         return -1;
108                 }
109         }
110
111         return 0;
112 }
113
114 int
115 virtio_user_dev_set_features(struct virtio_user_dev *dev)
116 {
117         uint64_t features;
118         int ret = -1;
119
120         pthread_mutex_lock(&dev->mutex);
121
122         if (dev->backend_type == VIRTIO_USER_BACKEND_VHOST_USER &&
123                         dev->vhostfd < 0)
124                 goto error;
125
126         /* Step 0: tell vhost to create queues */
127         if (virtio_user_queue_setup(dev, virtio_user_create_queue) < 0)
128                 goto error;
129
130         features = dev->features;
131
132         /* Strip VIRTIO_NET_F_MAC, as MAC address is handled in vdev init */
133         features &= ~(1ull << VIRTIO_NET_F_MAC);
134         /* Strip VIRTIO_NET_F_CTRL_VQ, as devices do not really need to know */
135         features &= ~(1ull << VIRTIO_NET_F_CTRL_VQ);
136         features &= ~(1ull << VIRTIO_NET_F_STATUS);
137         ret = dev->ops->send_request(dev, VHOST_USER_SET_FEATURES, &features);
138         if (ret < 0)
139                 goto error;
140         PMD_DRV_LOG(INFO, "set features: %" PRIx64, features);
141 error:
142         pthread_mutex_unlock(&dev->mutex);
143
144         return ret;
145 }
146
147 int
148 virtio_user_start_device(struct virtio_user_dev *dev)
149 {
150         int ret;
151
152         /*
153          * XXX workaround!
154          *
155          * We need to make sure that the locks will be
156          * taken in the correct order to avoid deadlocks.
157          *
158          * Before releasing this lock, this thread should
159          * not trigger any memory hotplug events.
160          *
161          * This is a temporary workaround, and should be
162          * replaced when we get proper supports from the
163          * memory subsystem in the future.
164          */
165         rte_mcfg_mem_read_lock();
166         pthread_mutex_lock(&dev->mutex);
167
168         if (dev->backend_type == VIRTIO_USER_BACKEND_VHOST_USER &&
169                         dev->vhostfd < 0)
170                 goto error;
171
172         /* Step 2: share memory regions */
173         ret = dev->ops->send_request(dev, VHOST_USER_SET_MEM_TABLE, NULL);
174         if (ret < 0)
175                 goto error;
176
177         /* Step 3: kick queues */
178         if (virtio_user_queue_setup(dev, virtio_user_kick_queue) < 0)
179                 goto error;
180
181         /* Step 4: enable queues
182          * we enable the 1st queue pair by default.
183          */
184         dev->ops->enable_qp(dev, 0, 1);
185
186         dev->started = true;
187         pthread_mutex_unlock(&dev->mutex);
188         rte_mcfg_mem_read_unlock();
189
190         return 0;
191 error:
192         pthread_mutex_unlock(&dev->mutex);
193         rte_mcfg_mem_read_unlock();
194         /* TODO: free resource here or caller to check */
195         return -1;
196 }
197
198 int virtio_user_stop_device(struct virtio_user_dev *dev)
199 {
200         struct vhost_vring_state state;
201         uint32_t i;
202         int error = 0;
203
204         pthread_mutex_lock(&dev->mutex);
205         if (!dev->started)
206                 goto out;
207
208         for (i = 0; i < dev->max_queue_pairs; ++i)
209                 dev->ops->enable_qp(dev, i, 0);
210
211         /* Stop the backend. */
212         for (i = 0; i < dev->max_queue_pairs * 2; ++i) {
213                 state.index = i;
214                 if (dev->ops->send_request(dev, VHOST_USER_GET_VRING_BASE,
215                                            &state) < 0) {
216                         PMD_DRV_LOG(ERR, "get_vring_base failed, index=%u\n",
217                                     i);
218                         error = -1;
219                         goto out;
220                 }
221         }
222
223         dev->started = false;
224 out:
225         pthread_mutex_unlock(&dev->mutex);
226
227         return error;
228 }
229
230 static inline void
231 parse_mac(struct virtio_user_dev *dev, const char *mac)
232 {
233         struct rte_ether_addr tmp;
234
235         if (!mac)
236                 return;
237
238         if (rte_ether_unformat_addr(mac, &tmp) == 0) {
239                 memcpy(dev->mac_addr, &tmp, RTE_ETHER_ADDR_LEN);
240                 dev->mac_specified = 1;
241         } else {
242                 /* ignore the wrong mac, use random mac */
243                 PMD_DRV_LOG(ERR, "wrong format of mac: %s", mac);
244         }
245 }
246
247 static int
248 virtio_user_dev_init_notify(struct virtio_user_dev *dev)
249 {
250         uint32_t i, j;
251         int callfd;
252         int kickfd;
253
254         for (i = 0; i < VIRTIO_MAX_VIRTQUEUES; ++i) {
255                 if (i >= dev->max_queue_pairs * 2) {
256                         dev->kickfds[i] = -1;
257                         dev->callfds[i] = -1;
258                         continue;
259                 }
260
261                 /* May use invalid flag, but some backend uses kickfd and
262                  * callfd as criteria to judge if dev is alive. so finally we
263                  * use real event_fd.
264                  */
265                 callfd = eventfd(0, EFD_CLOEXEC | EFD_NONBLOCK);
266                 if (callfd < 0) {
267                         PMD_DRV_LOG(ERR, "callfd error, %s", strerror(errno));
268                         break;
269                 }
270                 kickfd = eventfd(0, EFD_CLOEXEC | EFD_NONBLOCK);
271                 if (kickfd < 0) {
272                         PMD_DRV_LOG(ERR, "kickfd error, %s", strerror(errno));
273                         break;
274                 }
275                 dev->callfds[i] = callfd;
276                 dev->kickfds[i] = kickfd;
277         }
278
279         if (i < VIRTIO_MAX_VIRTQUEUES) {
280                 for (j = 0; j <= i; ++j) {
281                         close(dev->callfds[j]);
282                         close(dev->kickfds[j]);
283                 }
284
285                 return -1;
286         }
287
288         return 0;
289 }
290
291 static int
292 virtio_user_fill_intr_handle(struct virtio_user_dev *dev)
293 {
294         uint32_t i;
295         struct rte_eth_dev *eth_dev = &rte_eth_devices[dev->port_id];
296
297         if (!eth_dev->intr_handle) {
298                 eth_dev->intr_handle = malloc(sizeof(*eth_dev->intr_handle));
299                 if (!eth_dev->intr_handle) {
300                         PMD_DRV_LOG(ERR, "fail to allocate intr_handle");
301                         return -1;
302                 }
303                 memset(eth_dev->intr_handle, 0, sizeof(*eth_dev->intr_handle));
304         }
305
306         for (i = 0; i < dev->max_queue_pairs; ++i)
307                 eth_dev->intr_handle->efds[i] = dev->callfds[i];
308         eth_dev->intr_handle->nb_efd = dev->max_queue_pairs;
309         eth_dev->intr_handle->max_intr = dev->max_queue_pairs + 1;
310         eth_dev->intr_handle->type = RTE_INTR_HANDLE_VDEV;
311         /* For virtio vdev, no need to read counter for clean */
312         eth_dev->intr_handle->efd_counter_size = 0;
313         eth_dev->intr_handle->fd = -1;
314         if (dev->vhostfd >= 0)
315                 eth_dev->intr_handle->fd = dev->vhostfd;
316         else if (dev->is_server)
317                 eth_dev->intr_handle->fd = dev->listenfd;
318
319         return 0;
320 }
321
322 static void
323 virtio_user_mem_event_cb(enum rte_mem_event type __rte_unused,
324                          const void *addr,
325                          size_t len __rte_unused,
326                          void *arg)
327 {
328         struct virtio_user_dev *dev = arg;
329         struct rte_memseg_list *msl;
330         uint16_t i;
331
332         /* ignore externally allocated memory */
333         msl = rte_mem_virt2memseg_list(addr);
334         if (msl->external)
335                 return;
336
337         pthread_mutex_lock(&dev->mutex);
338
339         if (dev->started == false)
340                 goto exit;
341
342         /* Step 1: pause the active queues */
343         for (i = 0; i < dev->queue_pairs; i++)
344                 dev->ops->enable_qp(dev, i, 0);
345
346         /* Step 2: update memory regions */
347         dev->ops->send_request(dev, VHOST_USER_SET_MEM_TABLE, NULL);
348
349         /* Step 3: resume the active queues */
350         for (i = 0; i < dev->queue_pairs; i++)
351                 dev->ops->enable_qp(dev, i, 1);
352
353 exit:
354         pthread_mutex_unlock(&dev->mutex);
355 }
356
357 static int
358 virtio_user_dev_setup(struct virtio_user_dev *dev)
359 {
360         uint32_t q;
361
362         dev->vhostfd = -1;
363         dev->vhostfds = NULL;
364         dev->tapfds = NULL;
365
366         if (dev->is_server) {
367                 if (dev->backend_type != VIRTIO_USER_BACKEND_VHOST_USER) {
368                         PMD_DRV_LOG(ERR, "Server mode only supports vhost-user!");
369                         return -1;
370                 }
371                 dev->ops = &virtio_ops_user;
372         } else {
373                 if (dev->backend_type == VIRTIO_USER_BACKEND_VHOST_USER) {
374                         dev->ops = &virtio_ops_user;
375                 } else if (dev->backend_type ==
376                                         VIRTIO_USER_BACKEND_VHOST_KERNEL) {
377                         dev->ops = &virtio_ops_kernel;
378
379                         dev->vhostfds = malloc(dev->max_queue_pairs *
380                                                sizeof(int));
381                         dev->tapfds = malloc(dev->max_queue_pairs *
382                                              sizeof(int));
383                         if (!dev->vhostfds || !dev->tapfds) {
384                                 PMD_INIT_LOG(ERR, "Failed to malloc");
385                                 return -1;
386                         }
387
388                         for (q = 0; q < dev->max_queue_pairs; ++q) {
389                                 dev->vhostfds[q] = -1;
390                                 dev->tapfds[q] = -1;
391                         }
392                 }
393         }
394
395         if (dev->ops->setup(dev) < 0)
396                 return -1;
397
398         if (virtio_user_dev_init_notify(dev) < 0)
399                 return -1;
400
401         if (virtio_user_fill_intr_handle(dev) < 0)
402                 return -1;
403
404         return 0;
405 }
406
407 /* Use below macro to filter features from vhost backend */
408 #define VIRTIO_USER_SUPPORTED_FEATURES                  \
409         (1ULL << VIRTIO_NET_F_MAC               |       \
410          1ULL << VIRTIO_NET_F_STATUS            |       \
411          1ULL << VIRTIO_NET_F_MQ                |       \
412          1ULL << VIRTIO_NET_F_CTRL_MAC_ADDR     |       \
413          1ULL << VIRTIO_NET_F_CTRL_VQ           |       \
414          1ULL << VIRTIO_NET_F_CTRL_RX           |       \
415          1ULL << VIRTIO_NET_F_CTRL_VLAN         |       \
416          1ULL << VIRTIO_NET_F_CSUM              |       \
417          1ULL << VIRTIO_NET_F_HOST_TSO4         |       \
418          1ULL << VIRTIO_NET_F_HOST_TSO6         |       \
419          1ULL << VIRTIO_NET_F_MRG_RXBUF         |       \
420          1ULL << VIRTIO_RING_F_INDIRECT_DESC    |       \
421          1ULL << VIRTIO_NET_F_GUEST_CSUM        |       \
422          1ULL << VIRTIO_NET_F_GUEST_TSO4        |       \
423          1ULL << VIRTIO_NET_F_GUEST_TSO6        |       \
424          1ULL << VIRTIO_F_IN_ORDER              |       \
425          1ULL << VIRTIO_F_VERSION_1             |       \
426          1ULL << VIRTIO_F_RING_PACKED           |       \
427          1ULL << VHOST_USER_F_PROTOCOL_FEATURES)
428
429 #define VIRTIO_USER_SUPPORTED_PROTOCOL_FEATURES         \
430         (1ULL << VHOST_USER_PROTOCOL_F_MQ |             \
431          1ULL << VHOST_USER_PROTOCOL_F_REPLY_ACK |      \
432          1ULL << VHOST_USER_PROTOCOL_F_STATUS)
433
434 int
435 virtio_user_dev_init(struct virtio_user_dev *dev, char *path, int queues,
436                      int cq, int queue_size, const char *mac, char **ifname,
437                      int server, int mrg_rxbuf, int in_order, int packed_vq,
438                      enum virtio_user_backend_type backend_type)
439 {
440         uint64_t protocol_features = 0;
441
442         pthread_mutex_init(&dev->mutex, NULL);
443         strlcpy(dev->path, path, PATH_MAX);
444         dev->started = 0;
445         dev->max_queue_pairs = queues;
446         dev->queue_pairs = 1; /* mq disabled by default */
447         dev->queue_size = queue_size;
448         dev->is_server = server;
449         dev->mac_specified = 0;
450         dev->frontend_features = 0;
451         dev->unsupported_features = ~VIRTIO_USER_SUPPORTED_FEATURES;
452         dev->protocol_features = VIRTIO_USER_SUPPORTED_PROTOCOL_FEATURES;
453         dev->backend_type = backend_type;
454
455         parse_mac(dev, mac);
456
457         if (*ifname) {
458                 dev->ifname = *ifname;
459                 *ifname = NULL;
460         }
461
462         if (virtio_user_dev_setup(dev) < 0) {
463                 PMD_INIT_LOG(ERR, "backend set up fails");
464                 return -1;
465         }
466
467         if (dev->backend_type != VIRTIO_USER_BACKEND_VHOST_USER)
468                 dev->unsupported_features |=
469                         (1ULL << VHOST_USER_F_PROTOCOL_FEATURES);
470
471         if (!dev->is_server) {
472                 if (dev->ops->send_request(dev, VHOST_USER_SET_OWNER,
473                                            NULL) < 0) {
474                         PMD_INIT_LOG(ERR, "set_owner fails: %s",
475                                      strerror(errno));
476                         return -1;
477                 }
478
479                 if (dev->ops->send_request(dev, VHOST_USER_GET_FEATURES,
480                                            &dev->device_features) < 0) {
481                         PMD_INIT_LOG(ERR, "get_features failed: %s",
482                                      strerror(errno));
483                         return -1;
484                 }
485
486
487                 if (dev->device_features &
488                                 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)) {
489                         if (dev->ops->send_request(dev,
490                                         VHOST_USER_GET_PROTOCOL_FEATURES,
491                                         &protocol_features))
492                                 return -1;
493
494                         dev->protocol_features &= protocol_features;
495
496                         if (dev->ops->send_request(dev,
497                                         VHOST_USER_SET_PROTOCOL_FEATURES,
498                                         &dev->protocol_features))
499                                 return -1;
500
501                         if (!(dev->protocol_features &
502                                         (1ULL << VHOST_USER_PROTOCOL_F_MQ)))
503                                 dev->unsupported_features |=
504                                         (1ull << VIRTIO_NET_F_MQ);
505                 }
506         } else {
507                 /* We just pretend vhost-user can support all these features.
508                  * Note that this could be problematic that if some feature is
509                  * negotiated but not supported by the vhost-user which comes
510                  * later.
511                  */
512                 dev->device_features = VIRTIO_USER_SUPPORTED_FEATURES;
513         }
514
515
516
517         if (!mrg_rxbuf)
518                 dev->unsupported_features |= (1ull << VIRTIO_NET_F_MRG_RXBUF);
519
520         if (!in_order)
521                 dev->unsupported_features |= (1ull << VIRTIO_F_IN_ORDER);
522
523         if (!packed_vq)
524                 dev->unsupported_features |= (1ull << VIRTIO_F_RING_PACKED);
525
526         if (dev->mac_specified)
527                 dev->frontend_features |= (1ull << VIRTIO_NET_F_MAC);
528         else
529                 dev->unsupported_features |= (1ull << VIRTIO_NET_F_MAC);
530
531         if (cq) {
532                 /* device does not really need to know anything about CQ,
533                  * so if necessary, we just claim to support CQ
534                  */
535                 dev->frontend_features |= (1ull << VIRTIO_NET_F_CTRL_VQ);
536         } else {
537                 dev->unsupported_features |= (1ull << VIRTIO_NET_F_CTRL_VQ);
538                 /* Also disable features that depend on VIRTIO_NET_F_CTRL_VQ */
539                 dev->unsupported_features |= (1ull << VIRTIO_NET_F_CTRL_RX);
540                 dev->unsupported_features |= (1ull << VIRTIO_NET_F_CTRL_VLAN);
541                 dev->unsupported_features |=
542                         (1ull << VIRTIO_NET_F_GUEST_ANNOUNCE);
543                 dev->unsupported_features |= (1ull << VIRTIO_NET_F_MQ);
544                 dev->unsupported_features |=
545                         (1ull << VIRTIO_NET_F_CTRL_MAC_ADDR);
546         }
547
548         /* The backend will not report this feature, we add it explicitly */
549         if (dev->backend_type == VIRTIO_USER_BACKEND_VHOST_USER)
550                 dev->frontend_features |= (1ull << VIRTIO_NET_F_STATUS);
551
552         /*
553          * Device features =
554          *     (frontend_features | backend_features) & ~unsupported_features;
555          */
556         dev->device_features |= dev->frontend_features;
557         dev->device_features &= ~dev->unsupported_features;
558
559         if (rte_mem_event_callback_register(VIRTIO_USER_MEM_EVENT_CLB_NAME,
560                                 virtio_user_mem_event_cb, dev)) {
561                 if (rte_errno != ENOTSUP) {
562                         PMD_INIT_LOG(ERR, "Failed to register mem event"
563                                         " callback\n");
564                         return -1;
565                 }
566         }
567
568         return 0;
569 }
570
571 void
572 virtio_user_dev_uninit(struct virtio_user_dev *dev)
573 {
574         uint32_t i;
575
576         virtio_user_stop_device(dev);
577
578         rte_mem_event_callback_unregister(VIRTIO_USER_MEM_EVENT_CLB_NAME, dev);
579
580         for (i = 0; i < dev->max_queue_pairs * 2; ++i) {
581                 close(dev->callfds[i]);
582                 close(dev->kickfds[i]);
583         }
584
585         if (dev->vhostfd >= 0)
586                 close(dev->vhostfd);
587
588         if (dev->is_server && dev->listenfd >= 0) {
589                 close(dev->listenfd);
590                 dev->listenfd = -1;
591         }
592
593         if (dev->vhostfds) {
594                 for (i = 0; i < dev->max_queue_pairs; ++i) {
595                         close(dev->vhostfds[i]);
596                         if (dev->tapfds[i] >= 0)
597                                 close(dev->tapfds[i]);
598                 }
599                 free(dev->vhostfds);
600                 free(dev->tapfds);
601         }
602
603         free(dev->ifname);
604
605         if (dev->is_server)
606                 unlink(dev->path);
607 }
608
609 uint8_t
610 virtio_user_handle_mq(struct virtio_user_dev *dev, uint16_t q_pairs)
611 {
612         uint16_t i;
613         uint8_t ret = 0;
614
615         if (q_pairs > dev->max_queue_pairs) {
616                 PMD_INIT_LOG(ERR, "multi-q config %u, but only %u supported",
617                              q_pairs, dev->max_queue_pairs);
618                 return -1;
619         }
620
621         /* Server mode can't enable queue pairs if vhostfd is invalid,
622          * always return 0 in this case.
623          */
624         if (!dev->is_server || dev->vhostfd >= 0) {
625                 for (i = 0; i < q_pairs; ++i)
626                         ret |= dev->ops->enable_qp(dev, i, 1);
627                 for (i = q_pairs; i < dev->max_queue_pairs; ++i)
628                         ret |= dev->ops->enable_qp(dev, i, 0);
629         }
630         dev->queue_pairs = q_pairs;
631
632         return ret;
633 }
634
635 static uint32_t
636 virtio_user_handle_ctrl_msg(struct virtio_user_dev *dev, struct vring *vring,
637                             uint16_t idx_hdr)
638 {
639         struct virtio_net_ctrl_hdr *hdr;
640         virtio_net_ctrl_ack status = ~0;
641         uint16_t i, idx_data, idx_status;
642         uint32_t n_descs = 0;
643
644         /* locate desc for header, data, and status */
645         idx_data = vring->desc[idx_hdr].next;
646         n_descs++;
647
648         i = idx_data;
649         while (vring->desc[i].flags == VRING_DESC_F_NEXT) {
650                 i = vring->desc[i].next;
651                 n_descs++;
652         }
653
654         /* locate desc for status */
655         idx_status = i;
656         n_descs++;
657
658         hdr = (void *)(uintptr_t)vring->desc[idx_hdr].addr;
659         if (hdr->class == VIRTIO_NET_CTRL_MQ &&
660             hdr->cmd == VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) {
661                 uint16_t queues;
662
663                 queues = *(uint16_t *)(uintptr_t)vring->desc[idx_data].addr;
664                 status = virtio_user_handle_mq(dev, queues);
665         } else if (hdr->class == VIRTIO_NET_CTRL_RX  ||
666                    hdr->class == VIRTIO_NET_CTRL_MAC ||
667                    hdr->class == VIRTIO_NET_CTRL_VLAN) {
668                 status = 0;
669         }
670
671         /* Update status */
672         *(virtio_net_ctrl_ack *)(uintptr_t)vring->desc[idx_status].addr = status;
673
674         return n_descs;
675 }
676
677 static inline int
678 desc_is_avail(struct vring_packed_desc *desc, bool wrap_counter)
679 {
680         uint16_t flags = __atomic_load_n(&desc->flags, __ATOMIC_ACQUIRE);
681
682         return wrap_counter == !!(flags & VRING_PACKED_DESC_F_AVAIL) &&
683                 wrap_counter != !!(flags & VRING_PACKED_DESC_F_USED);
684 }
685
686 static uint32_t
687 virtio_user_handle_ctrl_msg_packed(struct virtio_user_dev *dev,
688                                    struct vring_packed *vring,
689                                    uint16_t idx_hdr)
690 {
691         struct virtio_net_ctrl_hdr *hdr;
692         virtio_net_ctrl_ack status = ~0;
693         uint16_t idx_data, idx_status;
694         /* initialize to one, header is first */
695         uint32_t n_descs = 1;
696
697         /* locate desc for header, data, and status */
698         idx_data = idx_hdr + 1;
699         if (idx_data >= dev->queue_size)
700                 idx_data -= dev->queue_size;
701
702         n_descs++;
703
704         idx_status = idx_data;
705         while (vring->desc[idx_status].flags & VRING_DESC_F_NEXT) {
706                 idx_status++;
707                 if (idx_status >= dev->queue_size)
708                         idx_status -= dev->queue_size;
709                 n_descs++;
710         }
711
712         hdr = (void *)(uintptr_t)vring->desc[idx_hdr].addr;
713         if (hdr->class == VIRTIO_NET_CTRL_MQ &&
714             hdr->cmd == VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) {
715                 uint16_t queues;
716
717                 queues = *(uint16_t *)(uintptr_t)
718                                 vring->desc[idx_data].addr;
719                 status = virtio_user_handle_mq(dev, queues);
720         } else if (hdr->class == VIRTIO_NET_CTRL_RX  ||
721                    hdr->class == VIRTIO_NET_CTRL_MAC ||
722                    hdr->class == VIRTIO_NET_CTRL_VLAN) {
723                 status = 0;
724         }
725
726         /* Update status */
727         *(virtio_net_ctrl_ack *)(uintptr_t)
728                 vring->desc[idx_status].addr = status;
729
730         /* Update used descriptor */
731         vring->desc[idx_hdr].id = vring->desc[idx_status].id;
732         vring->desc[idx_hdr].len = sizeof(status);
733
734         return n_descs;
735 }
736
737 void
738 virtio_user_handle_cq_packed(struct virtio_user_dev *dev, uint16_t queue_idx)
739 {
740         struct virtio_user_queue *vq = &dev->packed_queues[queue_idx];
741         struct vring_packed *vring = &dev->packed_vrings[queue_idx];
742         uint16_t n_descs, flags;
743
744         /* Perform a load-acquire barrier in desc_is_avail to
745          * enforce the ordering between desc flags and desc
746          * content.
747          */
748         while (desc_is_avail(&vring->desc[vq->used_idx],
749                              vq->used_wrap_counter)) {
750
751                 n_descs = virtio_user_handle_ctrl_msg_packed(dev, vring,
752                                 vq->used_idx);
753
754                 flags = VRING_DESC_F_WRITE;
755                 if (vq->used_wrap_counter)
756                         flags |= VRING_PACKED_DESC_F_AVAIL_USED;
757
758                 __atomic_store_n(&vring->desc[vq->used_idx].flags, flags,
759                                  __ATOMIC_RELEASE);
760
761                 vq->used_idx += n_descs;
762                 if (vq->used_idx >= dev->queue_size) {
763                         vq->used_idx -= dev->queue_size;
764                         vq->used_wrap_counter ^= 1;
765                 }
766         }
767 }
768
769 void
770 virtio_user_handle_cq(struct virtio_user_dev *dev, uint16_t queue_idx)
771 {
772         uint16_t avail_idx, desc_idx;
773         struct vring_used_elem *uep;
774         uint32_t n_descs;
775         struct vring *vring = &dev->vrings[queue_idx];
776
777         /* Consume avail ring, using used ring idx as first one */
778         while (__atomic_load_n(&vring->used->idx, __ATOMIC_RELAXED)
779                != vring->avail->idx) {
780                 avail_idx = __atomic_load_n(&vring->used->idx, __ATOMIC_RELAXED)
781                             & (vring->num - 1);
782                 desc_idx = vring->avail->ring[avail_idx];
783
784                 n_descs = virtio_user_handle_ctrl_msg(dev, vring, desc_idx);
785
786                 /* Update used ring */
787                 uep = &vring->used->ring[avail_idx];
788                 uep->id = desc_idx;
789                 uep->len = n_descs;
790
791                 __atomic_add_fetch(&vring->used->idx, 1, __ATOMIC_RELAXED);
792         }
793 }
794
795 int
796 virtio_user_send_status_update(struct virtio_user_dev *dev, uint8_t status)
797 {
798         int ret;
799         uint64_t arg = status;
800
801         if (dev->backend_type == VIRTIO_USER_BACKEND_VHOST_USER)
802                 ret = dev->ops->send_request(dev,
803                                 VHOST_USER_SET_STATUS, &arg);
804         else if (dev->backend_type == VIRTIO_USER_BACKEND_VHOST_VDPA)
805                 ret = dev->ops->send_request(dev,
806                                 VHOST_USER_SET_STATUS, &status);
807         else
808                 return 0;
809
810         if (ret) {
811                 PMD_INIT_LOG(ERR, "VHOST_USER_SET_STATUS failed (%d): %s", ret,
812                              strerror(errno));
813                 return -1;
814         }
815
816         return 0;
817 }
818
819 int
820 virtio_user_update_status(struct virtio_user_dev *dev)
821 {
822         uint64_t ret;
823         uint8_t status;
824         int err;
825
826         if (dev->backend_type == VIRTIO_USER_BACKEND_VHOST_USER) {
827                 err = dev->ops->send_request(dev, VHOST_USER_GET_STATUS, &ret);
828                 if (!err && ret > UINT8_MAX) {
829                         PMD_INIT_LOG(ERR, "Invalid VHOST_USER_GET_STATUS "
830                                         "response 0x%" PRIx64 "\n", ret);
831                         return -1;
832                 }
833
834                 status = ret;
835         } else if (dev->backend_type == VIRTIO_USER_BACKEND_VHOST_VDPA) {
836                 err = dev->ops->send_request(dev, VHOST_USER_GET_STATUS,
837                                 &status);
838         } else {
839                 return 0;
840         }
841
842         if (err) {
843                 PMD_INIT_LOG(ERR, "VHOST_USER_GET_STATUS failed (%d): %s", err,
844                              strerror(errno));
845                 return -1;
846         }
847
848         dev->status = status;
849         PMD_INIT_LOG(DEBUG, "Updated Device Status(0x%08x):\n"
850                         "\t-RESET: %u\n"
851                         "\t-ACKNOWLEDGE: %u\n"
852                         "\t-DRIVER: %u\n"
853                         "\t-DRIVER_OK: %u\n"
854                         "\t-FEATURES_OK: %u\n"
855                         "\t-DEVICE_NEED_RESET: %u\n"
856                         "\t-FAILED: %u\n",
857                         dev->status,
858                         (dev->status == VIRTIO_CONFIG_STATUS_RESET),
859                         !!(dev->status & VIRTIO_CONFIG_STATUS_ACK),
860                         !!(dev->status & VIRTIO_CONFIG_STATUS_DRIVER),
861                         !!(dev->status & VIRTIO_CONFIG_STATUS_DRIVER_OK),
862                         !!(dev->status & VIRTIO_CONFIG_STATUS_FEATURES_OK),
863                         !!(dev->status & VIRTIO_CONFIG_STATUS_DEV_NEED_RESET),
864                         !!(dev->status & VIRTIO_CONFIG_STATUS_FAILED));
865         return 0;
866 }