drivers/net: use ethdev allocation helper for vdev
[dpdk.git] / drivers / net / vhost / rte_eth_vhost.c
1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright (c) 2016 IGEL Co., Ltd.
5  *   All rights reserved.
6  *
7  *   Redistribution and use in source and binary forms, with or without
8  *   modification, are permitted provided that the following conditions
9  *   are met:
10  *
11  *     * Redistributions of source code must retain the above copyright
12  *       notice, this list of conditions and the following disclaimer.
13  *     * Redistributions in binary form must reproduce the above copyright
14  *       notice, this list of conditions and the following disclaimer in
15  *       the documentation and/or other materials provided with the
16  *       distribution.
17  *     * Neither the name of IGEL Co.,Ltd. nor the names of its
18  *       contributors may be used to endorse or promote products derived
19  *       from this software without specific prior written permission.
20  *
21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32  */
33 #include <unistd.h>
34 #include <pthread.h>
35 #include <stdbool.h>
36
37 #include <rte_mbuf.h>
38 #include <rte_ethdev.h>
39 #include <rte_ethdev_vdev.h>
40 #include <rte_malloc.h>
41 #include <rte_memcpy.h>
42 #include <rte_vdev.h>
43 #include <rte_kvargs.h>
44 #include <rte_vhost.h>
45 #include <rte_spinlock.h>
46
47 #include "rte_eth_vhost.h"
48
49 enum {VIRTIO_RXQ, VIRTIO_TXQ, VIRTIO_QNUM};
50
51 #define ETH_VHOST_IFACE_ARG             "iface"
52 #define ETH_VHOST_QUEUES_ARG            "queues"
53 #define ETH_VHOST_CLIENT_ARG            "client"
54 #define ETH_VHOST_DEQUEUE_ZERO_COPY     "dequeue-zero-copy"
55 #define VHOST_MAX_PKT_BURST 32
56
57 static const char *valid_arguments[] = {
58         ETH_VHOST_IFACE_ARG,
59         ETH_VHOST_QUEUES_ARG,
60         ETH_VHOST_CLIENT_ARG,
61         ETH_VHOST_DEQUEUE_ZERO_COPY,
62         NULL
63 };
64
65 static struct ether_addr base_eth_addr = {
66         .addr_bytes = {
67                 0x56 /* V */,
68                 0x48 /* H */,
69                 0x4F /* O */,
70                 0x53 /* S */,
71                 0x54 /* T */,
72                 0x00
73         }
74 };
75
76 enum vhost_xstats_pkts {
77         VHOST_UNDERSIZE_PKT = 0,
78         VHOST_64_PKT,
79         VHOST_65_TO_127_PKT,
80         VHOST_128_TO_255_PKT,
81         VHOST_256_TO_511_PKT,
82         VHOST_512_TO_1023_PKT,
83         VHOST_1024_TO_1522_PKT,
84         VHOST_1523_TO_MAX_PKT,
85         VHOST_BROADCAST_PKT,
86         VHOST_MULTICAST_PKT,
87         VHOST_UNICAST_PKT,
88         VHOST_ERRORS_PKT,
89         VHOST_ERRORS_FRAGMENTED,
90         VHOST_ERRORS_JABBER,
91         VHOST_UNKNOWN_PROTOCOL,
92         VHOST_XSTATS_MAX,
93 };
94
95 struct vhost_stats {
96         uint64_t pkts;
97         uint64_t bytes;
98         uint64_t missed_pkts;
99         uint64_t xstats[VHOST_XSTATS_MAX];
100 };
101
102 struct vhost_queue {
103         int vid;
104         rte_atomic32_t allow_queuing;
105         rte_atomic32_t while_queuing;
106         struct pmd_internal *internal;
107         struct rte_mempool *mb_pool;
108         uint8_t port;
109         uint16_t virtqueue_id;
110         struct vhost_stats stats;
111 };
112
113 struct pmd_internal {
114         rte_atomic32_t dev_attached;
115         char *dev_name;
116         char *iface_name;
117         uint16_t max_queues;
118         rte_atomic32_t started;
119 };
120
121 struct internal_list {
122         TAILQ_ENTRY(internal_list) next;
123         struct rte_eth_dev *eth_dev;
124 };
125
126 TAILQ_HEAD(internal_list_head, internal_list);
127 static struct internal_list_head internal_list =
128         TAILQ_HEAD_INITIALIZER(internal_list);
129
130 static pthread_mutex_t internal_list_lock = PTHREAD_MUTEX_INITIALIZER;
131
132 static struct rte_eth_link pmd_link = {
133                 .link_speed = 10000,
134                 .link_duplex = ETH_LINK_FULL_DUPLEX,
135                 .link_status = ETH_LINK_DOWN
136 };
137
138 struct rte_vhost_vring_state {
139         rte_spinlock_t lock;
140
141         bool cur[RTE_MAX_QUEUES_PER_PORT * 2];
142         bool seen[RTE_MAX_QUEUES_PER_PORT * 2];
143         unsigned int index;
144         unsigned int max_vring;
145 };
146
147 static struct rte_vhost_vring_state *vring_states[RTE_MAX_ETHPORTS];
148
149 #define VHOST_XSTATS_NAME_SIZE 64
150
151 struct vhost_xstats_name_off {
152         char name[VHOST_XSTATS_NAME_SIZE];
153         uint64_t offset;
154 };
155
156 /* [rx]_is prepended to the name string here */
157 static const struct vhost_xstats_name_off vhost_rxport_stat_strings[] = {
158         {"good_packets",
159          offsetof(struct vhost_queue, stats.pkts)},
160         {"total_bytes",
161          offsetof(struct vhost_queue, stats.bytes)},
162         {"missed_pkts",
163          offsetof(struct vhost_queue, stats.missed_pkts)},
164         {"broadcast_packets",
165          offsetof(struct vhost_queue, stats.xstats[VHOST_BROADCAST_PKT])},
166         {"multicast_packets",
167          offsetof(struct vhost_queue, stats.xstats[VHOST_MULTICAST_PKT])},
168         {"unicast_packets",
169          offsetof(struct vhost_queue, stats.xstats[VHOST_UNICAST_PKT])},
170          {"undersize_packets",
171          offsetof(struct vhost_queue, stats.xstats[VHOST_UNDERSIZE_PKT])},
172         {"size_64_packets",
173          offsetof(struct vhost_queue, stats.xstats[VHOST_64_PKT])},
174         {"size_65_to_127_packets",
175          offsetof(struct vhost_queue, stats.xstats[VHOST_65_TO_127_PKT])},
176         {"size_128_to_255_packets",
177          offsetof(struct vhost_queue, stats.xstats[VHOST_128_TO_255_PKT])},
178         {"size_256_to_511_packets",
179          offsetof(struct vhost_queue, stats.xstats[VHOST_256_TO_511_PKT])},
180         {"size_512_to_1023_packets",
181          offsetof(struct vhost_queue, stats.xstats[VHOST_512_TO_1023_PKT])},
182         {"size_1024_to_1522_packets",
183          offsetof(struct vhost_queue, stats.xstats[VHOST_1024_TO_1522_PKT])},
184         {"size_1523_to_max_packets",
185          offsetof(struct vhost_queue, stats.xstats[VHOST_1523_TO_MAX_PKT])},
186         {"errors_with_bad_CRC",
187          offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_PKT])},
188         {"fragmented_errors",
189          offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_FRAGMENTED])},
190         {"jabber_errors",
191          offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_JABBER])},
192         {"unknown_protos_packets",
193          offsetof(struct vhost_queue, stats.xstats[VHOST_UNKNOWN_PROTOCOL])},
194 };
195
196 /* [tx]_ is prepended to the name string here */
197 static const struct vhost_xstats_name_off vhost_txport_stat_strings[] = {
198         {"good_packets",
199          offsetof(struct vhost_queue, stats.pkts)},
200         {"total_bytes",
201          offsetof(struct vhost_queue, stats.bytes)},
202         {"missed_pkts",
203          offsetof(struct vhost_queue, stats.missed_pkts)},
204         {"broadcast_packets",
205          offsetof(struct vhost_queue, stats.xstats[VHOST_BROADCAST_PKT])},
206         {"multicast_packets",
207          offsetof(struct vhost_queue, stats.xstats[VHOST_MULTICAST_PKT])},
208         {"unicast_packets",
209          offsetof(struct vhost_queue, stats.xstats[VHOST_UNICAST_PKT])},
210         {"undersize_packets",
211          offsetof(struct vhost_queue, stats.xstats[VHOST_UNDERSIZE_PKT])},
212         {"size_64_packets",
213          offsetof(struct vhost_queue, stats.xstats[VHOST_64_PKT])},
214         {"size_65_to_127_packets",
215          offsetof(struct vhost_queue, stats.xstats[VHOST_65_TO_127_PKT])},
216         {"size_128_to_255_packets",
217          offsetof(struct vhost_queue, stats.xstats[VHOST_128_TO_255_PKT])},
218         {"size_256_to_511_packets",
219          offsetof(struct vhost_queue, stats.xstats[VHOST_256_TO_511_PKT])},
220         {"size_512_to_1023_packets",
221          offsetof(struct vhost_queue, stats.xstats[VHOST_512_TO_1023_PKT])},
222         {"size_1024_to_1522_packets",
223          offsetof(struct vhost_queue, stats.xstats[VHOST_1024_TO_1522_PKT])},
224         {"size_1523_to_max_packets",
225          offsetof(struct vhost_queue, stats.xstats[VHOST_1523_TO_MAX_PKT])},
226         {"errors_with_bad_CRC",
227          offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_PKT])},
228 };
229
230 #define VHOST_NB_XSTATS_RXPORT (sizeof(vhost_rxport_stat_strings) / \
231                                 sizeof(vhost_rxport_stat_strings[0]))
232
233 #define VHOST_NB_XSTATS_TXPORT (sizeof(vhost_txport_stat_strings) / \
234                                 sizeof(vhost_txport_stat_strings[0]))
235
236 static void
237 vhost_dev_xstats_reset(struct rte_eth_dev *dev)
238 {
239         struct vhost_queue *vq = NULL;
240         unsigned int i = 0;
241
242         for (i = 0; i < dev->data->nb_rx_queues; i++) {
243                 vq = dev->data->rx_queues[i];
244                 if (!vq)
245                         continue;
246                 memset(&vq->stats, 0, sizeof(vq->stats));
247         }
248         for (i = 0; i < dev->data->nb_tx_queues; i++) {
249                 vq = dev->data->tx_queues[i];
250                 if (!vq)
251                         continue;
252                 memset(&vq->stats, 0, sizeof(vq->stats));
253         }
254 }
255
256 static int
257 vhost_dev_xstats_get_names(struct rte_eth_dev *dev __rte_unused,
258                            struct rte_eth_xstat_name *xstats_names,
259                            unsigned int limit __rte_unused)
260 {
261         unsigned int t = 0;
262         int count = 0;
263         int nstats = VHOST_NB_XSTATS_RXPORT + VHOST_NB_XSTATS_TXPORT;
264
265         if (!xstats_names)
266                 return nstats;
267         for (t = 0; t < VHOST_NB_XSTATS_RXPORT; t++) {
268                 snprintf(xstats_names[count].name,
269                          sizeof(xstats_names[count].name),
270                          "rx_%s", vhost_rxport_stat_strings[t].name);
271                 count++;
272         }
273         for (t = 0; t < VHOST_NB_XSTATS_TXPORT; t++) {
274                 snprintf(xstats_names[count].name,
275                          sizeof(xstats_names[count].name),
276                          "tx_%s", vhost_txport_stat_strings[t].name);
277                 count++;
278         }
279         return count;
280 }
281
282 static int
283 vhost_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
284                      unsigned int n)
285 {
286         unsigned int i;
287         unsigned int t;
288         unsigned int count = 0;
289         struct vhost_queue *vq = NULL;
290         unsigned int nxstats = VHOST_NB_XSTATS_RXPORT + VHOST_NB_XSTATS_TXPORT;
291
292         if (n < nxstats)
293                 return nxstats;
294
295         for (i = 0; i < dev->data->nb_rx_queues; i++) {
296                 vq = dev->data->rx_queues[i];
297                 if (!vq)
298                         continue;
299                 vq->stats.xstats[VHOST_UNICAST_PKT] = vq->stats.pkts
300                                 - (vq->stats.xstats[VHOST_BROADCAST_PKT]
301                                 + vq->stats.xstats[VHOST_MULTICAST_PKT]);
302         }
303         for (i = 0; i < dev->data->nb_tx_queues; i++) {
304                 vq = dev->data->tx_queues[i];
305                 if (!vq)
306                         continue;
307                 vq->stats.xstats[VHOST_UNICAST_PKT] = vq->stats.pkts
308                                 + vq->stats.missed_pkts
309                                 - (vq->stats.xstats[VHOST_BROADCAST_PKT]
310                                 + vq->stats.xstats[VHOST_MULTICAST_PKT]);
311         }
312         for (t = 0; t < VHOST_NB_XSTATS_RXPORT; t++) {
313                 xstats[count].value = 0;
314                 for (i = 0; i < dev->data->nb_rx_queues; i++) {
315                         vq = dev->data->rx_queues[i];
316                         if (!vq)
317                                 continue;
318                         xstats[count].value +=
319                                 *(uint64_t *)(((char *)vq)
320                                 + vhost_rxport_stat_strings[t].offset);
321                 }
322                 xstats[count].id = count;
323                 count++;
324         }
325         for (t = 0; t < VHOST_NB_XSTATS_TXPORT; t++) {
326                 xstats[count].value = 0;
327                 for (i = 0; i < dev->data->nb_tx_queues; i++) {
328                         vq = dev->data->tx_queues[i];
329                         if (!vq)
330                                 continue;
331                         xstats[count].value +=
332                                 *(uint64_t *)(((char *)vq)
333                                 + vhost_txport_stat_strings[t].offset);
334                 }
335                 xstats[count].id = count;
336                 count++;
337         }
338         return count;
339 }
340
341 static inline void
342 vhost_count_multicast_broadcast(struct vhost_queue *vq,
343                                 struct rte_mbuf *mbuf)
344 {
345         struct ether_addr *ea = NULL;
346         struct vhost_stats *pstats = &vq->stats;
347
348         ea = rte_pktmbuf_mtod(mbuf, struct ether_addr *);
349         if (is_multicast_ether_addr(ea)) {
350                 if (is_broadcast_ether_addr(ea))
351                         pstats->xstats[VHOST_BROADCAST_PKT]++;
352                 else
353                         pstats->xstats[VHOST_MULTICAST_PKT]++;
354         }
355 }
356
357 static void
358 vhost_update_packet_xstats(struct vhost_queue *vq,
359                            struct rte_mbuf **bufs,
360                            uint16_t count)
361 {
362         uint32_t pkt_len = 0;
363         uint64_t i = 0;
364         uint64_t index;
365         struct vhost_stats *pstats = &vq->stats;
366
367         for (i = 0; i < count ; i++) {
368                 pkt_len = bufs[i]->pkt_len;
369                 if (pkt_len == 64) {
370                         pstats->xstats[VHOST_64_PKT]++;
371                 } else if (pkt_len > 64 && pkt_len < 1024) {
372                         index = (sizeof(pkt_len) * 8)
373                                 - __builtin_clz(pkt_len) - 5;
374                         pstats->xstats[index]++;
375                 } else {
376                         if (pkt_len < 64)
377                                 pstats->xstats[VHOST_UNDERSIZE_PKT]++;
378                         else if (pkt_len <= 1522)
379                                 pstats->xstats[VHOST_1024_TO_1522_PKT]++;
380                         else if (pkt_len > 1522)
381                                 pstats->xstats[VHOST_1523_TO_MAX_PKT]++;
382                 }
383                 vhost_count_multicast_broadcast(vq, bufs[i]);
384         }
385 }
386
387 static uint16_t
388 eth_vhost_rx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
389 {
390         struct vhost_queue *r = q;
391         uint16_t i, nb_rx = 0;
392         uint16_t nb_receive = nb_bufs;
393
394         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
395                 return 0;
396
397         rte_atomic32_set(&r->while_queuing, 1);
398
399         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
400                 goto out;
401
402         /* Dequeue packets from guest TX queue */
403         while (nb_receive) {
404                 uint16_t nb_pkts;
405                 uint16_t num = (uint16_t)RTE_MIN(nb_receive,
406                                                  VHOST_MAX_PKT_BURST);
407
408                 nb_pkts = rte_vhost_dequeue_burst(r->vid, r->virtqueue_id,
409                                                   r->mb_pool, &bufs[nb_rx],
410                                                   num);
411
412                 nb_rx += nb_pkts;
413                 nb_receive -= nb_pkts;
414                 if (nb_pkts < num)
415                         break;
416         }
417
418         r->stats.pkts += nb_rx;
419
420         for (i = 0; likely(i < nb_rx); i++) {
421                 bufs[i]->port = r->port;
422                 r->stats.bytes += bufs[i]->pkt_len;
423         }
424
425         vhost_update_packet_xstats(r, bufs, nb_rx);
426
427 out:
428         rte_atomic32_set(&r->while_queuing, 0);
429
430         return nb_rx;
431 }
432
433 static uint16_t
434 eth_vhost_tx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
435 {
436         struct vhost_queue *r = q;
437         uint16_t i, nb_tx = 0;
438         uint16_t nb_send = nb_bufs;
439
440         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
441                 return 0;
442
443         rte_atomic32_set(&r->while_queuing, 1);
444
445         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
446                 goto out;
447
448         /* Enqueue packets to guest RX queue */
449         while (nb_send) {
450                 uint16_t nb_pkts;
451                 uint16_t num = (uint16_t)RTE_MIN(nb_send,
452                                                  VHOST_MAX_PKT_BURST);
453
454                 nb_pkts = rte_vhost_enqueue_burst(r->vid, r->virtqueue_id,
455                                                   &bufs[nb_tx], num);
456
457                 nb_tx += nb_pkts;
458                 nb_send -= nb_pkts;
459                 if (nb_pkts < num)
460                         break;
461         }
462
463         r->stats.pkts += nb_tx;
464         r->stats.missed_pkts += nb_bufs - nb_tx;
465
466         for (i = 0; likely(i < nb_tx); i++)
467                 r->stats.bytes += bufs[i]->pkt_len;
468
469         vhost_update_packet_xstats(r, bufs, nb_tx);
470
471         /* According to RFC2863 page42 section ifHCOutMulticastPkts and
472          * ifHCOutBroadcastPkts, the counters "multicast" and "broadcast"
473          * are increased when packets are not transmitted successfully.
474          */
475         for (i = nb_tx; i < nb_bufs; i++)
476                 vhost_count_multicast_broadcast(r, bufs[i]);
477
478         for (i = 0; likely(i < nb_tx); i++)
479                 rte_pktmbuf_free(bufs[i]);
480 out:
481         rte_atomic32_set(&r->while_queuing, 0);
482
483         return nb_tx;
484 }
485
486 static int
487 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
488 {
489         return 0;
490 }
491
492 static inline struct internal_list *
493 find_internal_resource(char *ifname)
494 {
495         int found = 0;
496         struct internal_list *list;
497         struct pmd_internal *internal;
498
499         if (ifname == NULL)
500                 return NULL;
501
502         pthread_mutex_lock(&internal_list_lock);
503
504         TAILQ_FOREACH(list, &internal_list, next) {
505                 internal = list->eth_dev->data->dev_private;
506                 if (!strcmp(internal->iface_name, ifname)) {
507                         found = 1;
508                         break;
509                 }
510         }
511
512         pthread_mutex_unlock(&internal_list_lock);
513
514         if (!found)
515                 return NULL;
516
517         return list;
518 }
519
520 static void
521 update_queuing_status(struct rte_eth_dev *dev)
522 {
523         struct pmd_internal *internal = dev->data->dev_private;
524         struct vhost_queue *vq;
525         unsigned int i;
526         int allow_queuing = 1;
527
528         if (rte_atomic32_read(&internal->started) == 0 ||
529             rte_atomic32_read(&internal->dev_attached) == 0)
530                 allow_queuing = 0;
531
532         /* Wait until rx/tx_pkt_burst stops accessing vhost device */
533         for (i = 0; i < dev->data->nb_rx_queues; i++) {
534                 vq = dev->data->rx_queues[i];
535                 if (vq == NULL)
536                         continue;
537                 rte_atomic32_set(&vq->allow_queuing, allow_queuing);
538                 while (rte_atomic32_read(&vq->while_queuing))
539                         rte_pause();
540         }
541
542         for (i = 0; i < dev->data->nb_tx_queues; i++) {
543                 vq = dev->data->tx_queues[i];
544                 if (vq == NULL)
545                         continue;
546                 rte_atomic32_set(&vq->allow_queuing, allow_queuing);
547                 while (rte_atomic32_read(&vq->while_queuing))
548                         rte_pause();
549         }
550 }
551
552 static int
553 new_device(int vid)
554 {
555         struct rte_eth_dev *eth_dev;
556         struct internal_list *list;
557         struct pmd_internal *internal;
558         struct vhost_queue *vq;
559         unsigned i;
560         char ifname[PATH_MAX];
561 #ifdef RTE_LIBRTE_VHOST_NUMA
562         int newnode;
563 #endif
564
565         rte_vhost_get_ifname(vid, ifname, sizeof(ifname));
566         list = find_internal_resource(ifname);
567         if (list == NULL) {
568                 RTE_LOG(INFO, PMD, "Invalid device name: %s\n", ifname);
569                 return -1;
570         }
571
572         eth_dev = list->eth_dev;
573         internal = eth_dev->data->dev_private;
574
575 #ifdef RTE_LIBRTE_VHOST_NUMA
576         newnode = rte_vhost_get_numa_node(vid);
577         if (newnode >= 0)
578                 eth_dev->data->numa_node = newnode;
579 #endif
580
581         for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
582                 vq = eth_dev->data->rx_queues[i];
583                 if (vq == NULL)
584                         continue;
585                 vq->vid = vid;
586                 vq->internal = internal;
587                 vq->port = eth_dev->data->port_id;
588         }
589         for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
590                 vq = eth_dev->data->tx_queues[i];
591                 if (vq == NULL)
592                         continue;
593                 vq->vid = vid;
594                 vq->internal = internal;
595                 vq->port = eth_dev->data->port_id;
596         }
597
598         for (i = 0; i < rte_vhost_get_vring_num(vid); i++)
599                 rte_vhost_enable_guest_notification(vid, i, 0);
600
601         rte_vhost_get_mtu(vid, &eth_dev->data->mtu);
602
603         eth_dev->data->dev_link.link_status = ETH_LINK_UP;
604
605         rte_atomic32_set(&internal->dev_attached, 1);
606         update_queuing_status(eth_dev);
607
608         RTE_LOG(INFO, PMD, "New connection established\n");
609
610         _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_INTR_LSC, NULL);
611
612         return 0;
613 }
614
615 static void
616 destroy_device(int vid)
617 {
618         struct rte_eth_dev *eth_dev;
619         struct pmd_internal *internal;
620         struct vhost_queue *vq;
621         struct internal_list *list;
622         char ifname[PATH_MAX];
623         unsigned i;
624         struct rte_vhost_vring_state *state;
625
626         rte_vhost_get_ifname(vid, ifname, sizeof(ifname));
627         list = find_internal_resource(ifname);
628         if (list == NULL) {
629                 RTE_LOG(ERR, PMD, "Invalid interface name: %s\n", ifname);
630                 return;
631         }
632         eth_dev = list->eth_dev;
633         internal = eth_dev->data->dev_private;
634
635         rte_atomic32_set(&internal->dev_attached, 0);
636         update_queuing_status(eth_dev);
637
638         eth_dev->data->dev_link.link_status = ETH_LINK_DOWN;
639
640         for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
641                 vq = eth_dev->data->rx_queues[i];
642                 if (vq == NULL)
643                         continue;
644                 vq->vid = -1;
645         }
646         for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
647                 vq = eth_dev->data->tx_queues[i];
648                 if (vq == NULL)
649                         continue;
650                 vq->vid = -1;
651         }
652
653         state = vring_states[eth_dev->data->port_id];
654         rte_spinlock_lock(&state->lock);
655         for (i = 0; i <= state->max_vring; i++) {
656                 state->cur[i] = false;
657                 state->seen[i] = false;
658         }
659         state->max_vring = 0;
660         rte_spinlock_unlock(&state->lock);
661
662         RTE_LOG(INFO, PMD, "Connection closed\n");
663
664         _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_INTR_LSC, NULL);
665 }
666
667 static int
668 vring_state_changed(int vid, uint16_t vring, int enable)
669 {
670         struct rte_vhost_vring_state *state;
671         struct rte_eth_dev *eth_dev;
672         struct internal_list *list;
673         char ifname[PATH_MAX];
674
675         rte_vhost_get_ifname(vid, ifname, sizeof(ifname));
676         list = find_internal_resource(ifname);
677         if (list == NULL) {
678                 RTE_LOG(ERR, PMD, "Invalid interface name: %s\n", ifname);
679                 return -1;
680         }
681
682         eth_dev = list->eth_dev;
683         /* won't be NULL */
684         state = vring_states[eth_dev->data->port_id];
685         rte_spinlock_lock(&state->lock);
686         state->cur[vring] = enable;
687         state->max_vring = RTE_MAX(vring, state->max_vring);
688         rte_spinlock_unlock(&state->lock);
689
690         RTE_LOG(INFO, PMD, "vring%u is %s\n",
691                         vring, enable ? "enabled" : "disabled");
692
693         _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_QUEUE_STATE, NULL);
694
695         return 0;
696 }
697
698 static struct vhost_device_ops vhost_ops = {
699         .new_device          = new_device,
700         .destroy_device      = destroy_device,
701         .vring_state_changed = vring_state_changed,
702 };
703
704 int
705 rte_eth_vhost_get_queue_event(uint8_t port_id,
706                 struct rte_eth_vhost_queue_event *event)
707 {
708         struct rte_vhost_vring_state *state;
709         unsigned int i;
710         int idx;
711
712         if (port_id >= RTE_MAX_ETHPORTS) {
713                 RTE_LOG(ERR, PMD, "Invalid port id\n");
714                 return -1;
715         }
716
717         state = vring_states[port_id];
718         if (!state) {
719                 RTE_LOG(ERR, PMD, "Unused port\n");
720                 return -1;
721         }
722
723         rte_spinlock_lock(&state->lock);
724         for (i = 0; i <= state->max_vring; i++) {
725                 idx = state->index++ % (state->max_vring + 1);
726
727                 if (state->cur[idx] != state->seen[idx]) {
728                         state->seen[idx] = state->cur[idx];
729                         event->queue_id = idx / 2;
730                         event->rx = idx & 1;
731                         event->enable = state->cur[idx];
732                         rte_spinlock_unlock(&state->lock);
733                         return 0;
734                 }
735         }
736         rte_spinlock_unlock(&state->lock);
737
738         return -1;
739 }
740
741 int
742 rte_eth_vhost_get_vid_from_port_id(uint8_t port_id)
743 {
744         struct internal_list *list;
745         struct rte_eth_dev *eth_dev;
746         struct vhost_queue *vq;
747         int vid = -1;
748
749         if (!rte_eth_dev_is_valid_port(port_id))
750                 return -1;
751
752         pthread_mutex_lock(&internal_list_lock);
753
754         TAILQ_FOREACH(list, &internal_list, next) {
755                 eth_dev = list->eth_dev;
756                 if (eth_dev->data->port_id == port_id) {
757                         vq = eth_dev->data->rx_queues[0];
758                         if (vq) {
759                                 vid = vq->vid;
760                         }
761                         break;
762                 }
763         }
764
765         pthread_mutex_unlock(&internal_list_lock);
766
767         return vid;
768 }
769
770 static int
771 eth_dev_start(struct rte_eth_dev *dev)
772 {
773         struct pmd_internal *internal = dev->data->dev_private;
774
775         rte_atomic32_set(&internal->started, 1);
776         update_queuing_status(dev);
777
778         return 0;
779 }
780
781 static void
782 eth_dev_stop(struct rte_eth_dev *dev)
783 {
784         struct pmd_internal *internal = dev->data->dev_private;
785
786         rte_atomic32_set(&internal->started, 0);
787         update_queuing_status(dev);
788 }
789
790 static void
791 eth_dev_close(struct rte_eth_dev *dev)
792 {
793         struct pmd_internal *internal;
794         struct internal_list *list;
795
796         internal = dev->data->dev_private;
797         if (!internal)
798                 return;
799
800         rte_vhost_driver_unregister(internal->iface_name);
801
802         list = find_internal_resource(internal->iface_name);
803         if (!list)
804                 return;
805
806         pthread_mutex_lock(&internal_list_lock);
807         TAILQ_REMOVE(&internal_list, list, next);
808         pthread_mutex_unlock(&internal_list_lock);
809         rte_free(list);
810
811         free(internal->dev_name);
812         free(internal->iface_name);
813         rte_free(internal);
814 }
815
816 static int
817 eth_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
818                    uint16_t nb_rx_desc __rte_unused,
819                    unsigned int socket_id,
820                    const struct rte_eth_rxconf *rx_conf __rte_unused,
821                    struct rte_mempool *mb_pool)
822 {
823         struct vhost_queue *vq;
824
825         vq = rte_zmalloc_socket(NULL, sizeof(struct vhost_queue),
826                         RTE_CACHE_LINE_SIZE, socket_id);
827         if (vq == NULL) {
828                 RTE_LOG(ERR, PMD, "Failed to allocate memory for rx queue\n");
829                 return -ENOMEM;
830         }
831
832         vq->mb_pool = mb_pool;
833         vq->virtqueue_id = rx_queue_id * VIRTIO_QNUM + VIRTIO_TXQ;
834         dev->data->rx_queues[rx_queue_id] = vq;
835
836         return 0;
837 }
838
839 static int
840 eth_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
841                    uint16_t nb_tx_desc __rte_unused,
842                    unsigned int socket_id,
843                    const struct rte_eth_txconf *tx_conf __rte_unused)
844 {
845         struct vhost_queue *vq;
846
847         vq = rte_zmalloc_socket(NULL, sizeof(struct vhost_queue),
848                         RTE_CACHE_LINE_SIZE, socket_id);
849         if (vq == NULL) {
850                 RTE_LOG(ERR, PMD, "Failed to allocate memory for tx queue\n");
851                 return -ENOMEM;
852         }
853
854         vq->virtqueue_id = tx_queue_id * VIRTIO_QNUM + VIRTIO_RXQ;
855         dev->data->tx_queues[tx_queue_id] = vq;
856
857         return 0;
858 }
859
860 static void
861 eth_dev_info(struct rte_eth_dev *dev,
862              struct rte_eth_dev_info *dev_info)
863 {
864         struct pmd_internal *internal;
865
866         internal = dev->data->dev_private;
867         if (internal == NULL) {
868                 RTE_LOG(ERR, PMD, "Invalid device specified\n");
869                 return;
870         }
871
872         dev_info->max_mac_addrs = 1;
873         dev_info->max_rx_pktlen = (uint32_t)-1;
874         dev_info->max_rx_queues = internal->max_queues;
875         dev_info->max_tx_queues = internal->max_queues;
876         dev_info->min_rx_bufsize = 0;
877 }
878
879 static void
880 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
881 {
882         unsigned i;
883         unsigned long rx_total = 0, tx_total = 0, tx_missed_total = 0;
884         unsigned long rx_total_bytes = 0, tx_total_bytes = 0;
885         struct vhost_queue *vq;
886
887         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
888                         i < dev->data->nb_rx_queues; i++) {
889                 if (dev->data->rx_queues[i] == NULL)
890                         continue;
891                 vq = dev->data->rx_queues[i];
892                 stats->q_ipackets[i] = vq->stats.pkts;
893                 rx_total += stats->q_ipackets[i];
894
895                 stats->q_ibytes[i] = vq->stats.bytes;
896                 rx_total_bytes += stats->q_ibytes[i];
897         }
898
899         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
900                         i < dev->data->nb_tx_queues; i++) {
901                 if (dev->data->tx_queues[i] == NULL)
902                         continue;
903                 vq = dev->data->tx_queues[i];
904                 stats->q_opackets[i] = vq->stats.pkts;
905                 tx_missed_total += vq->stats.missed_pkts;
906                 tx_total += stats->q_opackets[i];
907
908                 stats->q_obytes[i] = vq->stats.bytes;
909                 tx_total_bytes += stats->q_obytes[i];
910         }
911
912         stats->ipackets = rx_total;
913         stats->opackets = tx_total;
914         stats->oerrors = tx_missed_total;
915         stats->ibytes = rx_total_bytes;
916         stats->obytes = tx_total_bytes;
917 }
918
919 static void
920 eth_stats_reset(struct rte_eth_dev *dev)
921 {
922         struct vhost_queue *vq;
923         unsigned i;
924
925         for (i = 0; i < dev->data->nb_rx_queues; i++) {
926                 if (dev->data->rx_queues[i] == NULL)
927                         continue;
928                 vq = dev->data->rx_queues[i];
929                 vq->stats.pkts = 0;
930                 vq->stats.bytes = 0;
931         }
932         for (i = 0; i < dev->data->nb_tx_queues; i++) {
933                 if (dev->data->tx_queues[i] == NULL)
934                         continue;
935                 vq = dev->data->tx_queues[i];
936                 vq->stats.pkts = 0;
937                 vq->stats.bytes = 0;
938                 vq->stats.missed_pkts = 0;
939         }
940 }
941
942 static void
943 eth_queue_release(void *q)
944 {
945         rte_free(q);
946 }
947
948 static int
949 eth_tx_done_cleanup(void *txq __rte_unused, uint32_t free_cnt __rte_unused)
950 {
951         /*
952          * vHost does not hang onto mbuf. eth_vhost_tx() copies packet data
953          * and releases mbuf, so nothing to cleanup.
954          */
955         return 0;
956 }
957
958 static int
959 eth_link_update(struct rte_eth_dev *dev __rte_unused,
960                 int wait_to_complete __rte_unused)
961 {
962         return 0;
963 }
964
965 static const struct eth_dev_ops ops = {
966         .dev_start = eth_dev_start,
967         .dev_stop = eth_dev_stop,
968         .dev_close = eth_dev_close,
969         .dev_configure = eth_dev_configure,
970         .dev_infos_get = eth_dev_info,
971         .rx_queue_setup = eth_rx_queue_setup,
972         .tx_queue_setup = eth_tx_queue_setup,
973         .rx_queue_release = eth_queue_release,
974         .tx_queue_release = eth_queue_release,
975         .tx_done_cleanup = eth_tx_done_cleanup,
976         .link_update = eth_link_update,
977         .stats_get = eth_stats_get,
978         .stats_reset = eth_stats_reset,
979         .xstats_reset = vhost_dev_xstats_reset,
980         .xstats_get = vhost_dev_xstats_get,
981         .xstats_get_names = vhost_dev_xstats_get_names,
982 };
983
984 static struct rte_vdev_driver pmd_vhost_drv;
985
986 static int
987 eth_dev_vhost_create(struct rte_vdev_device *dev, char *iface_name,
988         int16_t queues, const unsigned int numa_node, uint64_t flags)
989 {
990         const char *name = rte_vdev_device_name(dev);
991         struct rte_eth_dev_data *data = NULL;
992         struct pmd_internal *internal = NULL;
993         struct rte_eth_dev *eth_dev = NULL;
994         struct ether_addr *eth_addr = NULL;
995         struct rte_vhost_vring_state *vring_state = NULL;
996         struct internal_list *list = NULL;
997
998         RTE_LOG(INFO, PMD, "Creating VHOST-USER backend on numa socket %u\n",
999                 numa_node);
1000
1001         /* now do all data allocation - for eth_dev structure and internal
1002          * (private) data
1003          */
1004         data = rte_zmalloc_socket(name, sizeof(*data), 0, numa_node);
1005         if (data == NULL)
1006                 goto error;
1007
1008         list = rte_zmalloc_socket(name, sizeof(*list), 0, numa_node);
1009         if (list == NULL)
1010                 goto error;
1011
1012         /* reserve an ethdev entry */
1013         eth_dev = rte_eth_vdev_allocate(dev, sizeof(*internal));
1014         if (eth_dev == NULL)
1015                 goto error;
1016
1017         eth_addr = rte_zmalloc_socket(name, sizeof(*eth_addr), 0, numa_node);
1018         if (eth_addr == NULL)
1019                 goto error;
1020         *eth_addr = base_eth_addr;
1021         eth_addr->addr_bytes[5] = eth_dev->data->port_id;
1022
1023         vring_state = rte_zmalloc_socket(name,
1024                         sizeof(*vring_state), 0, numa_node);
1025         if (vring_state == NULL)
1026                 goto error;
1027
1028         /* now put it all together
1029          * - store queue data in internal,
1030          * - point eth_dev_data to internals
1031          * - and point eth_dev structure to new eth_dev_data structure
1032          */
1033         internal = eth_dev->data->dev_private;
1034         internal->dev_name = strdup(name);
1035         if (internal->dev_name == NULL)
1036                 goto error;
1037         internal->iface_name = strdup(iface_name);
1038         if (internal->iface_name == NULL)
1039                 goto error;
1040
1041         list->eth_dev = eth_dev;
1042         pthread_mutex_lock(&internal_list_lock);
1043         TAILQ_INSERT_TAIL(&internal_list, list, next);
1044         pthread_mutex_unlock(&internal_list_lock);
1045
1046         rte_spinlock_init(&vring_state->lock);
1047         vring_states[eth_dev->data->port_id] = vring_state;
1048
1049         /* We'll replace the 'data' originally allocated by eth_dev. So the
1050          * vhost PMD resources won't be shared between multi processes.
1051          */
1052         rte_memcpy(data, eth_dev->data, sizeof(*data));
1053         eth_dev->data = data;
1054
1055         data->nb_rx_queues = queues;
1056         data->nb_tx_queues = queues;
1057         internal->max_queues = queues;
1058         data->dev_link = pmd_link;
1059         data->mac_addrs = eth_addr;
1060         data->dev_flags =
1061                 RTE_ETH_DEV_DETACHABLE | RTE_ETH_DEV_INTR_LSC;
1062
1063         eth_dev->dev_ops = &ops;
1064
1065         /* finally assign rx and tx ops */
1066         eth_dev->rx_pkt_burst = eth_vhost_rx;
1067         eth_dev->tx_pkt_burst = eth_vhost_tx;
1068
1069         if (rte_vhost_driver_register(iface_name, flags))
1070                 goto error;
1071
1072         if (rte_vhost_driver_callback_register(iface_name, &vhost_ops) < 0) {
1073                 RTE_LOG(ERR, PMD, "Can't register callbacks\n");
1074                 goto error;
1075         }
1076
1077         if (rte_vhost_driver_start(iface_name) < 0) {
1078                 RTE_LOG(ERR, PMD, "Failed to start driver for %s\n",
1079                         iface_name);
1080                 goto error;
1081         }
1082
1083         return data->port_id;
1084
1085 error:
1086         if (internal) {
1087                 free(internal->iface_name);
1088                 free(internal->dev_name);
1089         }
1090         rte_free(vring_state);
1091         rte_free(eth_addr);
1092         if (eth_dev)
1093                 rte_eth_dev_release_port(eth_dev);
1094         rte_free(internal);
1095         rte_free(list);
1096         rte_free(data);
1097
1098         return -1;
1099 }
1100
1101 static inline int
1102 open_iface(const char *key __rte_unused, const char *value, void *extra_args)
1103 {
1104         const char **iface_name = extra_args;
1105
1106         if (value == NULL)
1107                 return -1;
1108
1109         *iface_name = value;
1110
1111         return 0;
1112 }
1113
1114 static inline int
1115 open_int(const char *key __rte_unused, const char *value, void *extra_args)
1116 {
1117         uint16_t *n = extra_args;
1118
1119         if (value == NULL || extra_args == NULL)
1120                 return -EINVAL;
1121
1122         *n = (uint16_t)strtoul(value, NULL, 0);
1123         if (*n == USHRT_MAX && errno == ERANGE)
1124                 return -1;
1125
1126         return 0;
1127 }
1128
1129 static int
1130 rte_pmd_vhost_probe(struct rte_vdev_device *dev)
1131 {
1132         struct rte_kvargs *kvlist = NULL;
1133         int ret = 0;
1134         char *iface_name;
1135         uint16_t queues;
1136         uint64_t flags = 0;
1137         int client_mode = 0;
1138         int dequeue_zero_copy = 0;
1139
1140         RTE_LOG(INFO, PMD, "Initializing pmd_vhost for %s\n",
1141                 rte_vdev_device_name(dev));
1142
1143         kvlist = rte_kvargs_parse(rte_vdev_device_args(dev), valid_arguments);
1144         if (kvlist == NULL)
1145                 return -1;
1146
1147         if (rte_kvargs_count(kvlist, ETH_VHOST_IFACE_ARG) == 1) {
1148                 ret = rte_kvargs_process(kvlist, ETH_VHOST_IFACE_ARG,
1149                                          &open_iface, &iface_name);
1150                 if (ret < 0)
1151                         goto out_free;
1152         } else {
1153                 ret = -1;
1154                 goto out_free;
1155         }
1156
1157         if (rte_kvargs_count(kvlist, ETH_VHOST_QUEUES_ARG) == 1) {
1158                 ret = rte_kvargs_process(kvlist, ETH_VHOST_QUEUES_ARG,
1159                                          &open_int, &queues);
1160                 if (ret < 0 || queues > RTE_MAX_QUEUES_PER_PORT)
1161                         goto out_free;
1162
1163         } else
1164                 queues = 1;
1165
1166         if (rte_kvargs_count(kvlist, ETH_VHOST_CLIENT_ARG) == 1) {
1167                 ret = rte_kvargs_process(kvlist, ETH_VHOST_CLIENT_ARG,
1168                                          &open_int, &client_mode);
1169                 if (ret < 0)
1170                         goto out_free;
1171
1172                 if (client_mode)
1173                         flags |= RTE_VHOST_USER_CLIENT;
1174         }
1175
1176         if (rte_kvargs_count(kvlist, ETH_VHOST_DEQUEUE_ZERO_COPY) == 1) {
1177                 ret = rte_kvargs_process(kvlist, ETH_VHOST_DEQUEUE_ZERO_COPY,
1178                                          &open_int, &dequeue_zero_copy);
1179                 if (ret < 0)
1180                         goto out_free;
1181
1182                 if (dequeue_zero_copy)
1183                         flags |= RTE_VHOST_USER_DEQUEUE_ZERO_COPY;
1184         }
1185
1186         if (dev->device.numa_node == SOCKET_ID_ANY)
1187                 dev->device.numa_node = rte_socket_id();
1188
1189         eth_dev_vhost_create(dev, iface_name, queues, dev->device.numa_node,
1190                 flags);
1191
1192 out_free:
1193         rte_kvargs_free(kvlist);
1194         return ret;
1195 }
1196
1197 static int
1198 rte_pmd_vhost_remove(struct rte_vdev_device *dev)
1199 {
1200         const char *name;
1201         struct rte_eth_dev *eth_dev = NULL;
1202         unsigned int i;
1203
1204         name = rte_vdev_device_name(dev);
1205         RTE_LOG(INFO, PMD, "Un-Initializing pmd_vhost for %s\n", name);
1206
1207         /* find an ethdev entry */
1208         eth_dev = rte_eth_dev_allocated(name);
1209         if (eth_dev == NULL)
1210                 return -ENODEV;
1211
1212         eth_dev_stop(eth_dev);
1213
1214         eth_dev_close(eth_dev);
1215
1216         rte_free(vring_states[eth_dev->data->port_id]);
1217         vring_states[eth_dev->data->port_id] = NULL;
1218
1219         for (i = 0; i < eth_dev->data->nb_rx_queues; i++)
1220                 rte_free(eth_dev->data->rx_queues[i]);
1221         for (i = 0; i < eth_dev->data->nb_tx_queues; i++)
1222                 rte_free(eth_dev->data->tx_queues[i]);
1223
1224         rte_free(eth_dev->data->mac_addrs);
1225         rte_free(eth_dev->data);
1226
1227         rte_eth_dev_release_port(eth_dev);
1228
1229         return 0;
1230 }
1231
1232 static struct rte_vdev_driver pmd_vhost_drv = {
1233         .probe = rte_pmd_vhost_probe,
1234         .remove = rte_pmd_vhost_remove,
1235 };
1236
1237 RTE_PMD_REGISTER_VDEV(net_vhost, pmd_vhost_drv);
1238 RTE_PMD_REGISTER_ALIAS(net_vhost, eth_vhost);
1239 RTE_PMD_REGISTER_PARAM_STRING(net_vhost,
1240         "iface=<ifc> "
1241         "queues=<int>");