785882e062ce32ed1725429e3af081385a4d2da1
[dpdk.git] / drivers / net / virtio / virtio_user_ethdev.c
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2010-2016 Intel Corporation
3  */
4
5 #include <stdint.h>
6 #include <sys/types.h>
7 #include <unistd.h>
8 #include <fcntl.h>
9 #include <sys/socket.h>
10
11 #include <rte_malloc.h>
12 #include <rte_kvargs.h>
13 #include <rte_ethdev_vdev.h>
14 #include <rte_bus_vdev.h>
15 #include <rte_alarm.h>
16 #include <rte_cycles.h>
17
18 #include "virtio_ethdev.h"
19 #include "virtio_logs.h"
20 #include "virtio_pci.h"
21 #include "virtqueue.h"
22 #include "virtio_rxtx.h"
23 #include "virtio_user/virtio_user_dev.h"
24 #include "virtio_user/vhost.h"
25
26 #define virtio_user_get_dev(hw) \
27         ((struct virtio_user_dev *)(hw)->virtio_user_dev)
28
29 static void
30 virtio_user_reset_queues_packed(struct rte_eth_dev *dev)
31 {
32         struct virtio_hw *hw = dev->data->dev_private;
33         struct virtnet_rx *rxvq;
34         struct virtnet_tx *txvq;
35         uint16_t i;
36
37         /* Add lock to avoid queue contention. */
38         rte_spinlock_lock(&hw->state_lock);
39         hw->started = 0;
40
41         /*
42          * Waitting for datapath to complete before resetting queues.
43          * 1 ms should be enough for the ongoing Tx/Rx function to finish.
44          */
45         rte_delay_ms(1);
46
47         /* Vring reset for each Tx queue and Rx queue. */
48         for (i = 0; i < dev->data->nb_rx_queues; i++) {
49                 rxvq = dev->data->rx_queues[i];
50                 virtqueue_rxvq_reset_packed(rxvq->vq);
51                 virtio_dev_rx_queue_setup_finish(dev, i);
52         }
53
54         for (i = 0; i < dev->data->nb_tx_queues; i++) {
55                 txvq = dev->data->tx_queues[i];
56                 virtqueue_txvq_reset_packed(txvq->vq);
57         }
58
59         hw->started = 1;
60         rte_spinlock_unlock(&hw->state_lock);
61 }
62
63
64 static int
65 virtio_user_server_reconnect(struct virtio_user_dev *dev)
66 {
67         int ret;
68         int connectfd;
69         struct rte_eth_dev *eth_dev = &rte_eth_devices[dev->port_id];
70         struct virtio_hw *hw = eth_dev->data->dev_private;
71         uint64_t protocol_features;
72
73         connectfd = accept(dev->listenfd, NULL, NULL);
74         if (connectfd < 0)
75                 return -1;
76
77         dev->vhostfd = connectfd;
78         if (dev->ops->send_request(dev, VHOST_USER_GET_FEATURES,
79                                    &dev->device_features) < 0) {
80                 PMD_INIT_LOG(ERR, "get_features failed: %s",
81                              strerror(errno));
82                 return -1;
83         }
84
85         if (dev->device_features &
86                         (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)) {
87                 if (dev->ops->send_request(dev,
88                                         VHOST_USER_GET_PROTOCOL_FEATURES,
89                                         &protocol_features))
90                         return -1;
91
92                 dev->protocol_features &= protocol_features;
93
94                 if (dev->ops->send_request(dev,
95                                         VHOST_USER_SET_PROTOCOL_FEATURES,
96                                         &dev->protocol_features))
97                         return -1;
98
99                 if (!(dev->protocol_features &
100                                 (1ULL << VHOST_USER_PROTOCOL_F_MQ)))
101                         dev->unsupported_features |= (1ull << VIRTIO_NET_F_MQ);
102         }
103
104         dev->device_features |= dev->frontend_features;
105
106         /* umask vhost-user unsupported features */
107         dev->device_features &= ~(dev->unsupported_features);
108
109         dev->features &= dev->device_features;
110
111         /* For packed ring, resetting queues is required in reconnection. */
112         if (vtpci_packed_queue(hw) &&
113            (vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_DRIVER_OK)) {
114                 PMD_INIT_LOG(NOTICE, "Packets on the fly will be dropped"
115                                 " when packed ring reconnecting.");
116                 virtio_user_reset_queues_packed(eth_dev);
117         }
118
119         ret = virtio_user_start_device(dev);
120         if (ret < 0)
121                 return -1;
122
123         if (dev->queue_pairs > 1) {
124                 ret = virtio_user_handle_mq(dev, dev->queue_pairs);
125                 if (ret != 0) {
126                         PMD_INIT_LOG(ERR, "Fails to enable multi-queue pairs!");
127                         return -1;
128                 }
129         }
130         if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
131                 if (rte_intr_disable(eth_dev->intr_handle) < 0) {
132                         PMD_DRV_LOG(ERR, "interrupt disable failed");
133                         return -1;
134                 }
135                 rte_intr_callback_unregister(eth_dev->intr_handle,
136                                              virtio_interrupt_handler,
137                                              eth_dev);
138                 eth_dev->intr_handle->fd = connectfd;
139                 rte_intr_callback_register(eth_dev->intr_handle,
140                                            virtio_interrupt_handler, eth_dev);
141
142                 if (rte_intr_enable(eth_dev->intr_handle) < 0) {
143                         PMD_DRV_LOG(ERR, "interrupt enable failed");
144                         return -1;
145                 }
146         }
147         PMD_INIT_LOG(NOTICE, "server mode virtio-user reconnection succeeds!");
148         return 0;
149 }
150
151 static void
152 virtio_user_delayed_handler(void *param)
153 {
154         struct virtio_hw *hw = (struct virtio_hw *)param;
155         struct rte_eth_dev *eth_dev = &rte_eth_devices[hw->port_id];
156         struct virtio_user_dev *dev = virtio_user_get_dev(hw);
157
158         if (rte_intr_disable(eth_dev->intr_handle) < 0) {
159                 PMD_DRV_LOG(ERR, "interrupt disable failed");
160                 return;
161         }
162         rte_intr_callback_unregister(eth_dev->intr_handle,
163                                      virtio_interrupt_handler, eth_dev);
164         if (dev->is_server) {
165                 if (dev->vhostfd >= 0) {
166                         close(dev->vhostfd);
167                         dev->vhostfd = -1;
168                 }
169                 eth_dev->intr_handle->fd = dev->listenfd;
170                 rte_intr_callback_register(eth_dev->intr_handle,
171                                            virtio_interrupt_handler, eth_dev);
172                 if (rte_intr_enable(eth_dev->intr_handle) < 0) {
173                         PMD_DRV_LOG(ERR, "interrupt enable failed");
174                         return;
175                 }
176         }
177 }
178
179 static void
180 virtio_user_read_dev_config(struct virtio_hw *hw, size_t offset,
181                      void *dst, int length)
182 {
183         int i;
184         struct virtio_user_dev *dev = virtio_user_get_dev(hw);
185
186         if (offset == offsetof(struct virtio_net_config, mac) &&
187             length == RTE_ETHER_ADDR_LEN) {
188                 for (i = 0; i < RTE_ETHER_ADDR_LEN; ++i)
189                         ((uint8_t *)dst)[i] = dev->mac_addr[i];
190                 return;
191         }
192
193         if (offset == offsetof(struct virtio_net_config, status)) {
194                 char buf[128];
195
196                 if (dev->vhostfd >= 0) {
197                         int r;
198                         int flags;
199
200                         flags = fcntl(dev->vhostfd, F_GETFL);
201                         if (fcntl(dev->vhostfd, F_SETFL,
202                                         flags | O_NONBLOCK) == -1) {
203                                 PMD_DRV_LOG(ERR, "error setting O_NONBLOCK flag");
204                                 return;
205                         }
206                         r = recv(dev->vhostfd, buf, 128, MSG_PEEK);
207                         if (r == 0 || (r < 0 && errno != EAGAIN)) {
208                                 dev->net_status &= (~VIRTIO_NET_S_LINK_UP);
209                                 PMD_DRV_LOG(ERR, "virtio-user port %u is down",
210                                             hw->port_id);
211
212                                 /* This function could be called in the process
213                                  * of interrupt handling, callback cannot be
214                                  * unregistered here, set an alarm to do it.
215                                  */
216                                 rte_eal_alarm_set(1,
217                                                   virtio_user_delayed_handler,
218                                                   (void *)hw);
219                         } else {
220                                 dev->net_status |= VIRTIO_NET_S_LINK_UP;
221                         }
222                         if (fcntl(dev->vhostfd, F_SETFL,
223                                         flags & ~O_NONBLOCK) == -1) {
224                                 PMD_DRV_LOG(ERR, "error clearing O_NONBLOCK flag");
225                                 return;
226                         }
227                 } else if (dev->is_server) {
228                         dev->net_status &= (~VIRTIO_NET_S_LINK_UP);
229                         if (virtio_user_server_reconnect(dev) >= 0)
230                                 dev->net_status |= VIRTIO_NET_S_LINK_UP;
231                 }
232
233                 *(uint16_t *)dst = dev->net_status;
234         }
235
236         if (offset == offsetof(struct virtio_net_config, max_virtqueue_pairs))
237                 *(uint16_t *)dst = dev->max_queue_pairs;
238 }
239
240 static void
241 virtio_user_write_dev_config(struct virtio_hw *hw, size_t offset,
242                       const void *src, int length)
243 {
244         int i;
245         struct virtio_user_dev *dev = virtio_user_get_dev(hw);
246
247         if ((offset == offsetof(struct virtio_net_config, mac)) &&
248             (length == RTE_ETHER_ADDR_LEN))
249                 for (i = 0; i < RTE_ETHER_ADDR_LEN; ++i)
250                         dev->mac_addr[i] = ((const uint8_t *)src)[i];
251         else
252                 PMD_DRV_LOG(ERR, "not supported offset=%zu, len=%d",
253                             offset, length);
254 }
255
256 static void
257 virtio_user_reset(struct virtio_hw *hw)
258 {
259         struct virtio_user_dev *dev = virtio_user_get_dev(hw);
260
261         if (dev->status & VIRTIO_CONFIG_STATUS_DRIVER_OK)
262                 virtio_user_stop_device(dev);
263 }
264
265 static void
266 virtio_user_set_status(struct virtio_hw *hw, uint8_t status)
267 {
268         struct virtio_user_dev *dev = virtio_user_get_dev(hw);
269
270         if (status & VIRTIO_CONFIG_STATUS_DRIVER_OK)
271                 virtio_user_start_device(dev);
272         else if (status == VIRTIO_CONFIG_STATUS_RESET)
273                 virtio_user_reset(hw);
274         dev->status = status;
275         virtio_user_send_status_update(dev, status);
276 }
277
278 static uint8_t
279 virtio_user_get_status(struct virtio_hw *hw)
280 {
281         struct virtio_user_dev *dev = virtio_user_get_dev(hw);
282
283         return dev->status;
284 }
285
286 static uint64_t
287 virtio_user_get_features(struct virtio_hw *hw)
288 {
289         struct virtio_user_dev *dev = virtio_user_get_dev(hw);
290
291         /* unmask feature bits defined in vhost user protocol */
292         return dev->device_features & VIRTIO_PMD_SUPPORTED_GUEST_FEATURES;
293 }
294
295 static void
296 virtio_user_set_features(struct virtio_hw *hw, uint64_t features)
297 {
298         struct virtio_user_dev *dev = virtio_user_get_dev(hw);
299
300         dev->features = features & dev->device_features;
301 }
302
303 static uint8_t
304 virtio_user_get_isr(struct virtio_hw *hw __rte_unused)
305 {
306         /* rxq interrupts and config interrupt are separated in virtio-user,
307          * here we only report config change.
308          */
309         return VIRTIO_PCI_ISR_CONFIG;
310 }
311
312 static uint16_t
313 virtio_user_set_config_irq(struct virtio_hw *hw __rte_unused,
314                     uint16_t vec __rte_unused)
315 {
316         return 0;
317 }
318
319 static uint16_t
320 virtio_user_set_queue_irq(struct virtio_hw *hw __rte_unused,
321                           struct virtqueue *vq __rte_unused,
322                           uint16_t vec)
323 {
324         /* pretend we have done that */
325         return vec;
326 }
327
328 /* This function is to get the queue size, aka, number of descs, of a specified
329  * queue. Different with the VHOST_USER_GET_QUEUE_NUM, which is used to get the
330  * max supported queues.
331  */
332 static uint16_t
333 virtio_user_get_queue_num(struct virtio_hw *hw, uint16_t queue_id __rte_unused)
334 {
335         struct virtio_user_dev *dev = virtio_user_get_dev(hw);
336
337         /* Currently, each queue has same queue size */
338         return dev->queue_size;
339 }
340
341 static void
342 virtio_user_setup_queue_packed(struct virtqueue *vq,
343                                struct virtio_user_dev *dev)
344 {
345         uint16_t queue_idx = vq->vq_queue_index;
346         struct vring_packed *vring;
347         uint64_t desc_addr;
348         uint64_t avail_addr;
349         uint64_t used_addr;
350         uint16_t i;
351
352         vring  = &dev->packed_vrings[queue_idx];
353         desc_addr = (uintptr_t)vq->vq_ring_virt_mem;
354         avail_addr = desc_addr + vq->vq_nentries *
355                 sizeof(struct vring_packed_desc);
356         used_addr = RTE_ALIGN_CEIL(avail_addr +
357                            sizeof(struct vring_packed_desc_event),
358                            VIRTIO_PCI_VRING_ALIGN);
359         vring->num = vq->vq_nentries;
360         vring->desc = (void *)(uintptr_t)desc_addr;
361         vring->driver = (void *)(uintptr_t)avail_addr;
362         vring->device = (void *)(uintptr_t)used_addr;
363         dev->packed_queues[queue_idx].avail_wrap_counter = true;
364         dev->packed_queues[queue_idx].used_wrap_counter = true;
365
366         for (i = 0; i < vring->num; i++)
367                 vring->desc[i].flags = 0;
368 }
369
370 static void
371 virtio_user_setup_queue_split(struct virtqueue *vq, struct virtio_user_dev *dev)
372 {
373         uint16_t queue_idx = vq->vq_queue_index;
374         uint64_t desc_addr, avail_addr, used_addr;
375
376         desc_addr = (uintptr_t)vq->vq_ring_virt_mem;
377         avail_addr = desc_addr + vq->vq_nentries * sizeof(struct vring_desc);
378         used_addr = RTE_ALIGN_CEIL(avail_addr + offsetof(struct vring_avail,
379                                                          ring[vq->vq_nentries]),
380                                    VIRTIO_PCI_VRING_ALIGN);
381
382         dev->vrings[queue_idx].num = vq->vq_nentries;
383         dev->vrings[queue_idx].desc = (void *)(uintptr_t)desc_addr;
384         dev->vrings[queue_idx].avail = (void *)(uintptr_t)avail_addr;
385         dev->vrings[queue_idx].used = (void *)(uintptr_t)used_addr;
386 }
387
388 static int
389 virtio_user_setup_queue(struct virtio_hw *hw, struct virtqueue *vq)
390 {
391         struct virtio_user_dev *dev = virtio_user_get_dev(hw);
392
393         if (vtpci_packed_queue(hw))
394                 virtio_user_setup_queue_packed(vq, dev);
395         else
396                 virtio_user_setup_queue_split(vq, dev);
397
398         return 0;
399 }
400
401 static void
402 virtio_user_del_queue(struct virtio_hw *hw, struct virtqueue *vq)
403 {
404         /* For legacy devices, write 0 to VIRTIO_PCI_QUEUE_PFN port, QEMU
405          * correspondingly stops the ioeventfds, and reset the status of
406          * the device.
407          * For modern devices, set queue desc, avail, used in PCI bar to 0,
408          * not see any more behavior in QEMU.
409          *
410          * Here we just care about what information to deliver to vhost-user
411          * or vhost-kernel. So we just close ioeventfd for now.
412          */
413         struct virtio_user_dev *dev = virtio_user_get_dev(hw);
414
415         close(dev->callfds[vq->vq_queue_index]);
416         close(dev->kickfds[vq->vq_queue_index]);
417 }
418
419 static void
420 virtio_user_notify_queue(struct virtio_hw *hw, struct virtqueue *vq)
421 {
422         uint64_t buf = 1;
423         struct virtio_user_dev *dev = virtio_user_get_dev(hw);
424
425         if (hw->cvq && (hw->cvq->vq == vq)) {
426                 if (vtpci_packed_queue(vq->hw))
427                         virtio_user_handle_cq_packed(dev, vq->vq_queue_index);
428                 else
429                         virtio_user_handle_cq(dev, vq->vq_queue_index);
430                 return;
431         }
432
433         if (write(dev->kickfds[vq->vq_queue_index], &buf, sizeof(buf)) < 0)
434                 PMD_DRV_LOG(ERR, "failed to kick backend: %s",
435                             strerror(errno));
436 }
437
438 const struct virtio_pci_ops virtio_user_ops = {
439         .read_dev_cfg   = virtio_user_read_dev_config,
440         .write_dev_cfg  = virtio_user_write_dev_config,
441         .get_status     = virtio_user_get_status,
442         .set_status     = virtio_user_set_status,
443         .get_features   = virtio_user_get_features,
444         .set_features   = virtio_user_set_features,
445         .get_isr        = virtio_user_get_isr,
446         .set_config_irq = virtio_user_set_config_irq,
447         .set_queue_irq  = virtio_user_set_queue_irq,
448         .get_queue_num  = virtio_user_get_queue_num,
449         .setup_queue    = virtio_user_setup_queue,
450         .del_queue      = virtio_user_del_queue,
451         .notify_queue   = virtio_user_notify_queue,
452 };
453
454 static const char *valid_args[] = {
455 #define VIRTIO_USER_ARG_QUEUES_NUM     "queues"
456         VIRTIO_USER_ARG_QUEUES_NUM,
457 #define VIRTIO_USER_ARG_CQ_NUM         "cq"
458         VIRTIO_USER_ARG_CQ_NUM,
459 #define VIRTIO_USER_ARG_MAC            "mac"
460         VIRTIO_USER_ARG_MAC,
461 #define VIRTIO_USER_ARG_PATH           "path"
462         VIRTIO_USER_ARG_PATH,
463 #define VIRTIO_USER_ARG_QUEUE_SIZE     "queue_size"
464         VIRTIO_USER_ARG_QUEUE_SIZE,
465 #define VIRTIO_USER_ARG_INTERFACE_NAME "iface"
466         VIRTIO_USER_ARG_INTERFACE_NAME,
467 #define VIRTIO_USER_ARG_SERVER_MODE    "server"
468         VIRTIO_USER_ARG_SERVER_MODE,
469 #define VIRTIO_USER_ARG_MRG_RXBUF      "mrg_rxbuf"
470         VIRTIO_USER_ARG_MRG_RXBUF,
471 #define VIRTIO_USER_ARG_IN_ORDER       "in_order"
472         VIRTIO_USER_ARG_IN_ORDER,
473 #define VIRTIO_USER_ARG_PACKED_VQ      "packed_vq"
474         VIRTIO_USER_ARG_PACKED_VQ,
475 #define VIRTIO_USER_ARG_SPEED          "speed"
476         VIRTIO_USER_ARG_SPEED,
477 #define VIRTIO_USER_ARG_VECTORIZED     "vectorized"
478         VIRTIO_USER_ARG_VECTORIZED,
479         NULL
480 };
481
482 #define VIRTIO_USER_DEF_CQ_EN   0
483 #define VIRTIO_USER_DEF_Q_NUM   1
484 #define VIRTIO_USER_DEF_Q_SZ    256
485 #define VIRTIO_USER_DEF_SERVER_MODE     0
486
487 static int
488 get_string_arg(const char *key __rte_unused,
489                const char *value, void *extra_args)
490 {
491         if (!value || !extra_args)
492                 return -EINVAL;
493
494         *(char **)extra_args = strdup(value);
495
496         if (!*(char **)extra_args)
497                 return -ENOMEM;
498
499         return 0;
500 }
501
502 static int
503 get_integer_arg(const char *key __rte_unused,
504                 const char *value, void *extra_args)
505 {
506         uint64_t integer = 0;
507         if (!value || !extra_args)
508                 return -EINVAL;
509         errno = 0;
510         integer = strtoull(value, NULL, 0);
511         /* extra_args keeps default value, it should be replaced
512          * only in case of successful parsing of the 'value' arg
513          */
514         if (errno == 0)
515                 *(uint64_t *)extra_args = integer;
516         return -errno;
517 }
518
519 static struct rte_eth_dev *
520 virtio_user_eth_dev_alloc(struct rte_vdev_device *vdev)
521 {
522         struct rte_eth_dev *eth_dev;
523         struct rte_eth_dev_data *data;
524         struct virtio_hw *hw;
525         struct virtio_user_dev *dev;
526
527         eth_dev = rte_eth_vdev_allocate(vdev, sizeof(*hw));
528         if (!eth_dev) {
529                 PMD_INIT_LOG(ERR, "cannot alloc rte_eth_dev");
530                 return NULL;
531         }
532
533         data = eth_dev->data;
534         hw = eth_dev->data->dev_private;
535
536         dev = rte_zmalloc(NULL, sizeof(*dev), 0);
537         if (!dev) {
538                 PMD_INIT_LOG(ERR, "malloc virtio_user_dev failed");
539                 rte_eth_dev_release_port(eth_dev);
540                 return NULL;
541         }
542
543         hw->port_id = data->port_id;
544         dev->port_id = data->port_id;
545         virtio_hw_internal[hw->port_id].vtpci_ops = &virtio_user_ops;
546         /*
547          * MSIX is required to enable LSC (see virtio_init_device).
548          * Here just pretend that we support msix.
549          */
550         hw->use_msix = 1;
551         hw->modern   = 0;
552         hw->use_vec_rx = 0;
553         hw->use_vec_tx = 0;
554         hw->use_inorder_rx = 0;
555         hw->use_inorder_tx = 0;
556         hw->virtio_user_dev = dev;
557         return eth_dev;
558 }
559
560 static void
561 virtio_user_eth_dev_free(struct rte_eth_dev *eth_dev)
562 {
563         struct rte_eth_dev_data *data = eth_dev->data;
564         struct virtio_hw *hw = data->dev_private;
565
566         rte_free(hw->virtio_user_dev);
567         rte_eth_dev_release_port(eth_dev);
568 }
569
570 /* Dev initialization routine. Invoked once for each virtio vdev at
571  * EAL init time, see rte_bus_probe().
572  * Returns 0 on success.
573  */
574 static int
575 virtio_user_pmd_probe(struct rte_vdev_device *dev)
576 {
577         struct rte_kvargs *kvlist = NULL;
578         struct rte_eth_dev *eth_dev;
579         struct virtio_hw *hw;
580         uint64_t queues = VIRTIO_USER_DEF_Q_NUM;
581         uint64_t cq = VIRTIO_USER_DEF_CQ_EN;
582         uint64_t queue_size = VIRTIO_USER_DEF_Q_SZ;
583         uint64_t server_mode = VIRTIO_USER_DEF_SERVER_MODE;
584         uint64_t mrg_rxbuf = 1;
585         uint64_t in_order = 1;
586         uint64_t packed_vq = 0;
587         uint64_t vectorized = 0;
588         char *path = NULL;
589         char *ifname = NULL;
590         char *mac_addr = NULL;
591         int ret = -1;
592
593         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
594                 const char *name = rte_vdev_device_name(dev);
595                 eth_dev = rte_eth_dev_attach_secondary(name);
596                 if (!eth_dev) {
597                         PMD_INIT_LOG(ERR, "Failed to probe %s", name);
598                         return -1;
599                 }
600
601                 if (eth_virtio_dev_init(eth_dev) < 0) {
602                         PMD_INIT_LOG(ERR, "eth_virtio_dev_init fails");
603                         rte_eth_dev_release_port(eth_dev);
604                         return -1;
605                 }
606
607                 eth_dev->dev_ops = &virtio_user_secondary_eth_dev_ops;
608                 eth_dev->device = &dev->device;
609                 rte_eth_dev_probing_finish(eth_dev);
610                 return 0;
611         }
612
613         kvlist = rte_kvargs_parse(rte_vdev_device_args(dev), valid_args);
614         if (!kvlist) {
615                 PMD_INIT_LOG(ERR, "error when parsing param");
616                 goto end;
617         }
618
619         if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_PATH) == 1) {
620                 if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_PATH,
621                                        &get_string_arg, &path) < 0) {
622                         PMD_INIT_LOG(ERR, "error to parse %s",
623                                      VIRTIO_USER_ARG_PATH);
624                         goto end;
625                 }
626         } else {
627                 PMD_INIT_LOG(ERR, "arg %s is mandatory for virtio_user",
628                              VIRTIO_USER_ARG_PATH);
629                 goto end;
630         }
631
632         if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_INTERFACE_NAME) == 1) {
633                 if (is_vhost_user_by_type(path)) {
634                         PMD_INIT_LOG(ERR,
635                                 "arg %s applies only to vhost-kernel backend",
636                                 VIRTIO_USER_ARG_INTERFACE_NAME);
637                         goto end;
638                 }
639
640                 if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_INTERFACE_NAME,
641                                        &get_string_arg, &ifname) < 0) {
642                         PMD_INIT_LOG(ERR, "error to parse %s",
643                                      VIRTIO_USER_ARG_INTERFACE_NAME);
644                         goto end;
645                 }
646         }
647
648         if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_MAC) == 1) {
649                 if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_MAC,
650                                        &get_string_arg, &mac_addr) < 0) {
651                         PMD_INIT_LOG(ERR, "error to parse %s",
652                                      VIRTIO_USER_ARG_MAC);
653                         goto end;
654                 }
655         }
656
657         if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_QUEUE_SIZE) == 1) {
658                 if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_QUEUE_SIZE,
659                                        &get_integer_arg, &queue_size) < 0) {
660                         PMD_INIT_LOG(ERR, "error to parse %s",
661                                      VIRTIO_USER_ARG_QUEUE_SIZE);
662                         goto end;
663                 }
664         }
665
666         if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_QUEUES_NUM) == 1) {
667                 if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_QUEUES_NUM,
668                                        &get_integer_arg, &queues) < 0) {
669                         PMD_INIT_LOG(ERR, "error to parse %s",
670                                      VIRTIO_USER_ARG_QUEUES_NUM);
671                         goto end;
672                 }
673         }
674
675         if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_SERVER_MODE) == 1) {
676                 if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_SERVER_MODE,
677                                        &get_integer_arg, &server_mode) < 0) {
678                         PMD_INIT_LOG(ERR, "error to parse %s",
679                                      VIRTIO_USER_ARG_SERVER_MODE);
680                         goto end;
681                 }
682         }
683
684         if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_CQ_NUM) == 1) {
685                 if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_CQ_NUM,
686                                        &get_integer_arg, &cq) < 0) {
687                         PMD_INIT_LOG(ERR, "error to parse %s",
688                                      VIRTIO_USER_ARG_CQ_NUM);
689                         goto end;
690                 }
691         } else if (queues > 1) {
692                 cq = 1;
693         }
694
695         if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_PACKED_VQ) == 1) {
696                 if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_PACKED_VQ,
697                                        &get_integer_arg, &packed_vq) < 0) {
698                         PMD_INIT_LOG(ERR, "error to parse %s",
699                                      VIRTIO_USER_ARG_PACKED_VQ);
700                         goto end;
701                 }
702         }
703
704         if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_VECTORIZED) == 1) {
705                 if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_VECTORIZED,
706                                        &get_integer_arg, &vectorized) < 0) {
707                         PMD_INIT_LOG(ERR, "error to parse %s",
708                                      VIRTIO_USER_ARG_VECTORIZED);
709                         goto end;
710                 }
711         }
712
713         if (queues > 1 && cq == 0) {
714                 PMD_INIT_LOG(ERR, "multi-q requires ctrl-q");
715                 goto end;
716         }
717
718         if (queues > VIRTIO_MAX_VIRTQUEUE_PAIRS) {
719                 PMD_INIT_LOG(ERR, "arg %s %" PRIu64 " exceeds the limit %u",
720                         VIRTIO_USER_ARG_QUEUES_NUM, queues,
721                         VIRTIO_MAX_VIRTQUEUE_PAIRS);
722                 goto end;
723         }
724
725         if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_MRG_RXBUF) == 1) {
726                 if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_MRG_RXBUF,
727                                        &get_integer_arg, &mrg_rxbuf) < 0) {
728                         PMD_INIT_LOG(ERR, "error to parse %s",
729                                      VIRTIO_USER_ARG_MRG_RXBUF);
730                         goto end;
731                 }
732         }
733
734         if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_IN_ORDER) == 1) {
735                 if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_IN_ORDER,
736                                        &get_integer_arg, &in_order) < 0) {
737                         PMD_INIT_LOG(ERR, "error to parse %s",
738                                      VIRTIO_USER_ARG_IN_ORDER);
739                         goto end;
740                 }
741         }
742
743         eth_dev = virtio_user_eth_dev_alloc(dev);
744         if (!eth_dev) {
745                 PMD_INIT_LOG(ERR, "virtio_user fails to alloc device");
746                 goto end;
747         }
748
749         hw = eth_dev->data->dev_private;
750         if (virtio_user_dev_init(hw->virtio_user_dev, path, queues, cq,
751                          queue_size, mac_addr, &ifname, server_mode,
752                          mrg_rxbuf, in_order, packed_vq) < 0) {
753                 PMD_INIT_LOG(ERR, "virtio_user_dev_init fails");
754                 virtio_user_eth_dev_free(eth_dev);
755                 goto end;
756         }
757
758         /* previously called by pci probing for physical dev */
759         if (eth_virtio_dev_init(eth_dev) < 0) {
760                 PMD_INIT_LOG(ERR, "eth_virtio_dev_init fails");
761                 virtio_user_eth_dev_free(eth_dev);
762                 goto end;
763         }
764
765         if (vectorized) {
766                 if (packed_vq) {
767 #if defined(CC_AVX512_SUPPORT)
768                         hw->use_vec_rx = 1;
769                         hw->use_vec_tx = 1;
770 #else
771                         PMD_INIT_LOG(INFO,
772                                 "building environment do not support packed ring vectorized");
773 #endif
774                 } else {
775                         hw->use_vec_rx = 1;
776                 }
777         }
778
779         rte_eth_dev_probing_finish(eth_dev);
780         ret = 0;
781
782 end:
783         if (kvlist)
784                 rte_kvargs_free(kvlist);
785         if (path)
786                 free(path);
787         if (mac_addr)
788                 free(mac_addr);
789         if (ifname)
790                 free(ifname);
791         return ret;
792 }
793
794 static int
795 virtio_user_pmd_remove(struct rte_vdev_device *vdev)
796 {
797         const char *name;
798         struct rte_eth_dev *eth_dev;
799
800         if (!vdev)
801                 return -EINVAL;
802
803         name = rte_vdev_device_name(vdev);
804         PMD_DRV_LOG(INFO, "Un-Initializing %s", name);
805         eth_dev = rte_eth_dev_allocated(name);
806         /* Port has already been released by close. */
807         if (!eth_dev)
808                 return 0;
809
810         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
811                 return rte_eth_dev_release_port(eth_dev);
812
813         /* make sure the device is stopped, queues freed */
814         rte_eth_dev_close(eth_dev->data->port_id);
815
816         return 0;
817 }
818
819 static struct rte_vdev_driver virtio_user_driver = {
820         .probe = virtio_user_pmd_probe,
821         .remove = virtio_user_pmd_remove,
822 };
823
824 RTE_PMD_REGISTER_VDEV(net_virtio_user, virtio_user_driver);
825 RTE_PMD_REGISTER_ALIAS(net_virtio_user, virtio_user);
826 RTE_PMD_REGISTER_PARAM_STRING(net_virtio_user,
827         "path=<path> "
828         "mac=<mac addr> "
829         "cq=<int> "
830         "queue_size=<int> "
831         "queues=<int> "
832         "iface=<string> "
833         "server=<0|1> "
834         "mrg_rxbuf=<0|1> "
835         "in_order=<0|1> "
836         "packed_vq=<0|1> "
837         "speed=<int> "
838         "vectorized=<0|1>");