b355c526fa16f47b85a541e0ebca2c097320c85c
[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_malloc.h>
40 #include <rte_memcpy.h>
41 #include <rte_vdev.h>
42 #include <rte_kvargs.h>
43 #include <rte_virtio_net.h>
44 #include <rte_spinlock.h>
45
46 #include "rte_eth_vhost.h"
47
48 enum {VIRTIO_RXQ, VIRTIO_TXQ, VIRTIO_QNUM};
49
50 #define ETH_VHOST_IFACE_ARG             "iface"
51 #define ETH_VHOST_QUEUES_ARG            "queues"
52 #define ETH_VHOST_CLIENT_ARG            "client"
53 #define ETH_VHOST_DEQUEUE_ZERO_COPY     "dequeue-zero-copy"
54 #define VHOST_MAX_PKT_BURST 32
55
56 static const char *valid_arguments[] = {
57         ETH_VHOST_IFACE_ARG,
58         ETH_VHOST_QUEUES_ARG,
59         ETH_VHOST_CLIENT_ARG,
60         ETH_VHOST_DEQUEUE_ZERO_COPY,
61         NULL
62 };
63
64 static struct ether_addr base_eth_addr = {
65         .addr_bytes = {
66                 0x56 /* V */,
67                 0x48 /* H */,
68                 0x4F /* O */,
69                 0x53 /* S */,
70                 0x54 /* T */,
71                 0x00
72         }
73 };
74
75 enum vhost_xstats_pkts {
76         VHOST_UNDERSIZE_PKT = 0,
77         VHOST_64_PKT,
78         VHOST_65_TO_127_PKT,
79         VHOST_128_TO_255_PKT,
80         VHOST_256_TO_511_PKT,
81         VHOST_512_TO_1023_PKT,
82         VHOST_1024_TO_1522_PKT,
83         VHOST_1523_TO_MAX_PKT,
84         VHOST_BROADCAST_PKT,
85         VHOST_MULTICAST_PKT,
86         VHOST_UNICAST_PKT,
87         VHOST_ERRORS_PKT,
88         VHOST_ERRORS_FRAGMENTED,
89         VHOST_ERRORS_JABBER,
90         VHOST_UNKNOWN_PROTOCOL,
91         VHOST_XSTATS_MAX,
92 };
93
94 struct vhost_stats {
95         uint64_t pkts;
96         uint64_t bytes;
97         uint64_t missed_pkts;
98         uint64_t xstats[VHOST_XSTATS_MAX];
99 };
100
101 struct vhost_queue {
102         int vid;
103         rte_atomic32_t allow_queuing;
104         rte_atomic32_t while_queuing;
105         struct pmd_internal *internal;
106         struct rte_mempool *mb_pool;
107         uint8_t port;
108         uint16_t virtqueue_id;
109         struct vhost_stats stats;
110 };
111
112 struct pmd_internal {
113         rte_atomic32_t dev_attached;
114         char *dev_name;
115         char *iface_name;
116         uint16_t max_queues;
117         rte_atomic32_t started;
118 };
119
120 struct internal_list {
121         TAILQ_ENTRY(internal_list) next;
122         struct rte_eth_dev *eth_dev;
123 };
124
125 TAILQ_HEAD(internal_list_head, internal_list);
126 static struct internal_list_head internal_list =
127         TAILQ_HEAD_INITIALIZER(internal_list);
128
129 static pthread_mutex_t internal_list_lock = PTHREAD_MUTEX_INITIALIZER;
130
131 static rte_atomic16_t nb_started_ports;
132 static pthread_t session_th;
133
134 static struct rte_eth_link pmd_link = {
135                 .link_speed = 10000,
136                 .link_duplex = ETH_LINK_FULL_DUPLEX,
137                 .link_status = ETH_LINK_DOWN
138 };
139
140 struct rte_vhost_vring_state {
141         rte_spinlock_t lock;
142
143         bool cur[RTE_MAX_QUEUES_PER_PORT * 2];
144         bool seen[RTE_MAX_QUEUES_PER_PORT * 2];
145         unsigned int index;
146         unsigned int max_vring;
147 };
148
149 static struct rte_vhost_vring_state *vring_states[RTE_MAX_ETHPORTS];
150
151 #define VHOST_XSTATS_NAME_SIZE 64
152
153 struct vhost_xstats_name_off {
154         char name[VHOST_XSTATS_NAME_SIZE];
155         uint64_t offset;
156 };
157
158 /* [rx]_is prepended to the name string here */
159 static const struct vhost_xstats_name_off vhost_rxport_stat_strings[] = {
160         {"good_packets",
161          offsetof(struct vhost_queue, stats.pkts)},
162         {"total_bytes",
163          offsetof(struct vhost_queue, stats.bytes)},
164         {"missed_pkts",
165          offsetof(struct vhost_queue, stats.missed_pkts)},
166         {"broadcast_packets",
167          offsetof(struct vhost_queue, stats.xstats[VHOST_BROADCAST_PKT])},
168         {"multicast_packets",
169          offsetof(struct vhost_queue, stats.xstats[VHOST_MULTICAST_PKT])},
170         {"unicast_packets",
171          offsetof(struct vhost_queue, stats.xstats[VHOST_UNICAST_PKT])},
172          {"undersize_packets",
173          offsetof(struct vhost_queue, stats.xstats[VHOST_UNDERSIZE_PKT])},
174         {"size_64_packets",
175          offsetof(struct vhost_queue, stats.xstats[VHOST_64_PKT])},
176         {"size_65_to_127_packets",
177          offsetof(struct vhost_queue, stats.xstats[VHOST_65_TO_127_PKT])},
178         {"size_128_to_255_packets",
179          offsetof(struct vhost_queue, stats.xstats[VHOST_128_TO_255_PKT])},
180         {"size_256_to_511_packets",
181          offsetof(struct vhost_queue, stats.xstats[VHOST_256_TO_511_PKT])},
182         {"size_512_to_1023_packets",
183          offsetof(struct vhost_queue, stats.xstats[VHOST_512_TO_1023_PKT])},
184         {"size_1024_to_1522_packets",
185          offsetof(struct vhost_queue, stats.xstats[VHOST_1024_TO_1522_PKT])},
186         {"size_1523_to_max_packets",
187          offsetof(struct vhost_queue, stats.xstats[VHOST_1523_TO_MAX_PKT])},
188         {"errors_with_bad_CRC",
189          offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_PKT])},
190         {"fragmented_errors",
191          offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_FRAGMENTED])},
192         {"jabber_errors",
193          offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_JABBER])},
194         {"unknown_protos_packets",
195          offsetof(struct vhost_queue, stats.xstats[VHOST_UNKNOWN_PROTOCOL])},
196 };
197
198 /* [tx]_ is prepended to the name string here */
199 static const struct vhost_xstats_name_off vhost_txport_stat_strings[] = {
200         {"good_packets",
201          offsetof(struct vhost_queue, stats.pkts)},
202         {"total_bytes",
203          offsetof(struct vhost_queue, stats.bytes)},
204         {"missed_pkts",
205          offsetof(struct vhost_queue, stats.missed_pkts)},
206         {"broadcast_packets",
207          offsetof(struct vhost_queue, stats.xstats[VHOST_BROADCAST_PKT])},
208         {"multicast_packets",
209          offsetof(struct vhost_queue, stats.xstats[VHOST_MULTICAST_PKT])},
210         {"unicast_packets",
211          offsetof(struct vhost_queue, stats.xstats[VHOST_UNICAST_PKT])},
212         {"undersize_packets",
213          offsetof(struct vhost_queue, stats.xstats[VHOST_UNDERSIZE_PKT])},
214         {"size_64_packets",
215          offsetof(struct vhost_queue, stats.xstats[VHOST_64_PKT])},
216         {"size_65_to_127_packets",
217          offsetof(struct vhost_queue, stats.xstats[VHOST_65_TO_127_PKT])},
218         {"size_128_to_255_packets",
219          offsetof(struct vhost_queue, stats.xstats[VHOST_128_TO_255_PKT])},
220         {"size_256_to_511_packets",
221          offsetof(struct vhost_queue, stats.xstats[VHOST_256_TO_511_PKT])},
222         {"size_512_to_1023_packets",
223          offsetof(struct vhost_queue, stats.xstats[VHOST_512_TO_1023_PKT])},
224         {"size_1024_to_1522_packets",
225          offsetof(struct vhost_queue, stats.xstats[VHOST_1024_TO_1522_PKT])},
226         {"size_1523_to_max_packets",
227          offsetof(struct vhost_queue, stats.xstats[VHOST_1523_TO_MAX_PKT])},
228         {"errors_with_bad_CRC",
229          offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_PKT])},
230 };
231
232 #define VHOST_NB_XSTATS_RXPORT (sizeof(vhost_rxport_stat_strings) / \
233                                 sizeof(vhost_rxport_stat_strings[0]))
234
235 #define VHOST_NB_XSTATS_TXPORT (sizeof(vhost_txport_stat_strings) / \
236                                 sizeof(vhost_txport_stat_strings[0]))
237
238 static void
239 vhost_dev_xstats_reset(struct rte_eth_dev *dev)
240 {
241         struct vhost_queue *vq = NULL;
242         unsigned int i = 0;
243
244         for (i = 0; i < dev->data->nb_rx_queues; i++) {
245                 vq = dev->data->rx_queues[i];
246                 if (!vq)
247                         continue;
248                 memset(&vq->stats, 0, sizeof(vq->stats));
249         }
250         for (i = 0; i < dev->data->nb_tx_queues; i++) {
251                 vq = dev->data->tx_queues[i];
252                 if (!vq)
253                         continue;
254                 memset(&vq->stats, 0, sizeof(vq->stats));
255         }
256 }
257
258 static int
259 vhost_dev_xstats_get_names(struct rte_eth_dev *dev __rte_unused,
260                            struct rte_eth_xstat_name *xstats_names,
261                            unsigned int limit __rte_unused)
262 {
263         unsigned int t = 0;
264         int count = 0;
265         int nstats = VHOST_NB_XSTATS_RXPORT + VHOST_NB_XSTATS_TXPORT;
266
267         if (!xstats_names)
268                 return nstats;
269         for (t = 0; t < VHOST_NB_XSTATS_RXPORT; t++) {
270                 snprintf(xstats_names[count].name,
271                          sizeof(xstats_names[count].name),
272                          "rx_%s", vhost_rxport_stat_strings[t].name);
273                 count++;
274         }
275         for (t = 0; t < VHOST_NB_XSTATS_TXPORT; t++) {
276                 snprintf(xstats_names[count].name,
277                          sizeof(xstats_names[count].name),
278                          "tx_%s", vhost_txport_stat_strings[t].name);
279                 count++;
280         }
281         return count;
282 }
283
284 static int
285 vhost_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
286                      unsigned int n)
287 {
288         unsigned int i;
289         unsigned int t;
290         unsigned int count = 0;
291         struct vhost_queue *vq = NULL;
292         unsigned int nxstats = VHOST_NB_XSTATS_RXPORT + VHOST_NB_XSTATS_TXPORT;
293
294         if (n < nxstats)
295                 return nxstats;
296
297         for (i = 0; i < dev->data->nb_rx_queues; i++) {
298                 vq = dev->data->rx_queues[i];
299                 if (!vq)
300                         continue;
301                 vq->stats.xstats[VHOST_UNICAST_PKT] = vq->stats.pkts
302                                 - (vq->stats.xstats[VHOST_BROADCAST_PKT]
303                                 + vq->stats.xstats[VHOST_MULTICAST_PKT]);
304         }
305         for (i = 0; i < dev->data->nb_tx_queues; i++) {
306                 vq = dev->data->tx_queues[i];
307                 if (!vq)
308                         continue;
309                 vq->stats.xstats[VHOST_UNICAST_PKT] = vq->stats.pkts
310                                 + vq->stats.missed_pkts
311                                 - (vq->stats.xstats[VHOST_BROADCAST_PKT]
312                                 + vq->stats.xstats[VHOST_MULTICAST_PKT]);
313         }
314         for (t = 0; t < VHOST_NB_XSTATS_RXPORT; t++) {
315                 xstats[count].value = 0;
316                 for (i = 0; i < dev->data->nb_rx_queues; i++) {
317                         vq = dev->data->rx_queues[i];
318                         if (!vq)
319                                 continue;
320                         xstats[count].value +=
321                                 *(uint64_t *)(((char *)vq)
322                                 + vhost_rxport_stat_strings[t].offset);
323                 }
324                 xstats[count].id = count;
325                 count++;
326         }
327         for (t = 0; t < VHOST_NB_XSTATS_TXPORT; t++) {
328                 xstats[count].value = 0;
329                 for (i = 0; i < dev->data->nb_tx_queues; i++) {
330                         vq = dev->data->tx_queues[i];
331                         if (!vq)
332                                 continue;
333                         xstats[count].value +=
334                                 *(uint64_t *)(((char *)vq)
335                                 + vhost_txport_stat_strings[t].offset);
336                 }
337                 xstats[count].id = count;
338                 count++;
339         }
340         return count;
341 }
342
343 static inline void
344 vhost_count_multicast_broadcast(struct vhost_queue *vq,
345                                 struct rte_mbuf *mbuf)
346 {
347         struct ether_addr *ea = NULL;
348         struct vhost_stats *pstats = &vq->stats;
349
350         ea = rte_pktmbuf_mtod(mbuf, struct ether_addr *);
351         if (is_multicast_ether_addr(ea)) {
352                 if (is_broadcast_ether_addr(ea))
353                         pstats->xstats[VHOST_BROADCAST_PKT]++;
354                 else
355                         pstats->xstats[VHOST_MULTICAST_PKT]++;
356         }
357 }
358
359 static void
360 vhost_update_packet_xstats(struct vhost_queue *vq,
361                            struct rte_mbuf **bufs,
362                            uint16_t count)
363 {
364         uint32_t pkt_len = 0;
365         uint64_t i = 0;
366         uint64_t index;
367         struct vhost_stats *pstats = &vq->stats;
368
369         for (i = 0; i < count ; i++) {
370                 pkt_len = bufs[i]->pkt_len;
371                 if (pkt_len == 64) {
372                         pstats->xstats[VHOST_64_PKT]++;
373                 } else if (pkt_len > 64 && pkt_len < 1024) {
374                         index = (sizeof(pkt_len) * 8)
375                                 - __builtin_clz(pkt_len) - 5;
376                         pstats->xstats[index]++;
377                 } else {
378                         if (pkt_len < 64)
379                                 pstats->xstats[VHOST_UNDERSIZE_PKT]++;
380                         else if (pkt_len <= 1522)
381                                 pstats->xstats[VHOST_1024_TO_1522_PKT]++;
382                         else if (pkt_len > 1522)
383                                 pstats->xstats[VHOST_1523_TO_MAX_PKT]++;
384                 }
385                 vhost_count_multicast_broadcast(vq, bufs[i]);
386         }
387 }
388
389 static uint16_t
390 eth_vhost_rx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
391 {
392         struct vhost_queue *r = q;
393         uint16_t i, nb_rx = 0;
394         uint16_t nb_receive = nb_bufs;
395
396         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
397                 return 0;
398
399         rte_atomic32_set(&r->while_queuing, 1);
400
401         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
402                 goto out;
403
404         /* Dequeue packets from guest TX queue */
405         while (nb_receive) {
406                 uint16_t nb_pkts;
407                 uint16_t num = (uint16_t)RTE_MIN(nb_receive,
408                                                  VHOST_MAX_PKT_BURST);
409
410                 nb_pkts = rte_vhost_dequeue_burst(r->vid, r->virtqueue_id,
411                                                   r->mb_pool, &bufs[nb_rx],
412                                                   num);
413
414                 nb_rx += nb_pkts;
415                 nb_receive -= nb_pkts;
416                 if (nb_pkts < num)
417                         break;
418         }
419
420         r->stats.pkts += nb_rx;
421
422         for (i = 0; likely(i < nb_rx); i++) {
423                 bufs[i]->port = r->port;
424                 r->stats.bytes += bufs[i]->pkt_len;
425         }
426
427         vhost_update_packet_xstats(r, bufs, nb_rx);
428
429 out:
430         rte_atomic32_set(&r->while_queuing, 0);
431
432         return nb_rx;
433 }
434
435 static uint16_t
436 eth_vhost_tx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
437 {
438         struct vhost_queue *r = q;
439         uint16_t i, nb_tx = 0;
440         uint16_t nb_send = nb_bufs;
441
442         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
443                 return 0;
444
445         rte_atomic32_set(&r->while_queuing, 1);
446
447         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
448                 goto out;
449
450         /* Enqueue packets to guest RX queue */
451         while (nb_send) {
452                 uint16_t nb_pkts;
453                 uint16_t num = (uint16_t)RTE_MIN(nb_send,
454                                                  VHOST_MAX_PKT_BURST);
455
456                 nb_pkts = rte_vhost_enqueue_burst(r->vid, r->virtqueue_id,
457                                                   &bufs[nb_tx], num);
458
459                 nb_tx += nb_pkts;
460                 nb_send -= nb_pkts;
461                 if (nb_pkts < num)
462                         break;
463         }
464
465         r->stats.pkts += nb_tx;
466         r->stats.missed_pkts += nb_bufs - nb_tx;
467
468         for (i = 0; likely(i < nb_tx); i++)
469                 r->stats.bytes += bufs[i]->pkt_len;
470
471         vhost_update_packet_xstats(r, bufs, nb_tx);
472
473         /* According to RFC2863 page42 section ifHCOutMulticastPkts and
474          * ifHCOutBroadcastPkts, the counters "multicast" and "broadcast"
475          * are increased when packets are not transmitted successfully.
476          */
477         for (i = nb_tx; i < nb_bufs; i++)
478                 vhost_count_multicast_broadcast(r, bufs[i]);
479
480         for (i = 0; likely(i < nb_tx); i++)
481                 rte_pktmbuf_free(bufs[i]);
482 out:
483         rte_atomic32_set(&r->while_queuing, 0);
484
485         return nb_tx;
486 }
487
488 static int
489 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
490 {
491         return 0;
492 }
493
494 static inline struct internal_list *
495 find_internal_resource(char *ifname)
496 {
497         int found = 0;
498         struct internal_list *list;
499         struct pmd_internal *internal;
500
501         if (ifname == NULL)
502                 return NULL;
503
504         pthread_mutex_lock(&internal_list_lock);
505
506         TAILQ_FOREACH(list, &internal_list, next) {
507                 internal = list->eth_dev->data->dev_private;
508                 if (!strcmp(internal->iface_name, ifname)) {
509                         found = 1;
510                         break;
511                 }
512         }
513
514         pthread_mutex_unlock(&internal_list_lock);
515
516         if (!found)
517                 return NULL;
518
519         return list;
520 }
521
522 static void
523 update_queuing_status(struct rte_eth_dev *dev)
524 {
525         struct pmd_internal *internal = dev->data->dev_private;
526         struct vhost_queue *vq;
527         unsigned int i;
528         int allow_queuing = 1;
529
530         if (rte_atomic32_read(&internal->started) == 0 ||
531             rte_atomic32_read(&internal->dev_attached) == 0)
532                 allow_queuing = 0;
533
534         /* Wait until rx/tx_pkt_burst stops accessing vhost device */
535         for (i = 0; i < dev->data->nb_rx_queues; i++) {
536                 vq = dev->data->rx_queues[i];
537                 if (vq == NULL)
538                         continue;
539                 rte_atomic32_set(&vq->allow_queuing, allow_queuing);
540                 while (rte_atomic32_read(&vq->while_queuing))
541                         rte_pause();
542         }
543
544         for (i = 0; i < dev->data->nb_tx_queues; i++) {
545                 vq = dev->data->tx_queues[i];
546                 if (vq == NULL)
547                         continue;
548                 rte_atomic32_set(&vq->allow_queuing, allow_queuing);
549                 while (rte_atomic32_read(&vq->while_queuing))
550                         rte_pause();
551         }
552 }
553
554 static int
555 new_device(int vid)
556 {
557         struct rte_eth_dev *eth_dev;
558         struct internal_list *list;
559         struct pmd_internal *internal;
560         struct vhost_queue *vq;
561         unsigned i;
562         char ifname[PATH_MAX];
563 #ifdef RTE_LIBRTE_VHOST_NUMA
564         int newnode;
565 #endif
566
567         rte_vhost_get_ifname(vid, ifname, sizeof(ifname));
568         list = find_internal_resource(ifname);
569         if (list == NULL) {
570                 RTE_LOG(INFO, PMD, "Invalid device name: %s\n", ifname);
571                 return -1;
572         }
573
574         eth_dev = list->eth_dev;
575         internal = eth_dev->data->dev_private;
576
577 #ifdef RTE_LIBRTE_VHOST_NUMA
578         newnode = rte_vhost_get_numa_node(vid);
579         if (newnode >= 0)
580                 eth_dev->data->numa_node = newnode;
581 #endif
582
583         for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
584                 vq = eth_dev->data->rx_queues[i];
585                 if (vq == NULL)
586                         continue;
587                 vq->vid = vid;
588                 vq->internal = internal;
589                 vq->port = eth_dev->data->port_id;
590         }
591         for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
592                 vq = eth_dev->data->tx_queues[i];
593                 if (vq == NULL)
594                         continue;
595                 vq->vid = vid;
596                 vq->internal = internal;
597                 vq->port = eth_dev->data->port_id;
598         }
599
600         for (i = 0; i < rte_vhost_get_vring_num(vid); i++)
601                 rte_vhost_enable_guest_notification(vid, i, 0);
602
603         rte_vhost_get_mtu(vid, &eth_dev->data->mtu);
604
605         eth_dev->data->dev_link.link_status = ETH_LINK_UP;
606
607         rte_atomic32_set(&internal->dev_attached, 1);
608         update_queuing_status(eth_dev);
609
610         RTE_LOG(INFO, PMD, "New connection established\n");
611
612         _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_INTR_LSC, NULL);
613
614         return 0;
615 }
616
617 static void
618 destroy_device(int vid)
619 {
620         struct rte_eth_dev *eth_dev;
621         struct pmd_internal *internal;
622         struct vhost_queue *vq;
623         struct internal_list *list;
624         char ifname[PATH_MAX];
625         unsigned i;
626         struct rte_vhost_vring_state *state;
627
628         rte_vhost_get_ifname(vid, ifname, sizeof(ifname));
629         list = find_internal_resource(ifname);
630         if (list == NULL) {
631                 RTE_LOG(ERR, PMD, "Invalid interface name: %s\n", ifname);
632                 return;
633         }
634         eth_dev = list->eth_dev;
635         internal = eth_dev->data->dev_private;
636
637         rte_atomic32_set(&internal->dev_attached, 0);
638         update_queuing_status(eth_dev);
639
640         eth_dev->data->dev_link.link_status = ETH_LINK_DOWN;
641
642         for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
643                 vq = eth_dev->data->rx_queues[i];
644                 if (vq == NULL)
645                         continue;
646                 vq->vid = -1;
647         }
648         for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
649                 vq = eth_dev->data->tx_queues[i];
650                 if (vq == NULL)
651                         continue;
652                 vq->vid = -1;
653         }
654
655         state = vring_states[eth_dev->data->port_id];
656         rte_spinlock_lock(&state->lock);
657         for (i = 0; i <= state->max_vring; i++) {
658                 state->cur[i] = false;
659                 state->seen[i] = false;
660         }
661         state->max_vring = 0;
662         rte_spinlock_unlock(&state->lock);
663
664         RTE_LOG(INFO, PMD, "Connection closed\n");
665
666         _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_INTR_LSC, NULL);
667 }
668
669 static int
670 vring_state_changed(int vid, uint16_t vring, int enable)
671 {
672         struct rte_vhost_vring_state *state;
673         struct rte_eth_dev *eth_dev;
674         struct internal_list *list;
675         char ifname[PATH_MAX];
676
677         rte_vhost_get_ifname(vid, ifname, sizeof(ifname));
678         list = find_internal_resource(ifname);
679         if (list == NULL) {
680                 RTE_LOG(ERR, PMD, "Invalid interface name: %s\n", ifname);
681                 return -1;
682         }
683
684         eth_dev = list->eth_dev;
685         /* won't be NULL */
686         state = vring_states[eth_dev->data->port_id];
687         rte_spinlock_lock(&state->lock);
688         state->cur[vring] = enable;
689         state->max_vring = RTE_MAX(vring, state->max_vring);
690         rte_spinlock_unlock(&state->lock);
691
692         RTE_LOG(INFO, PMD, "vring%u is %s\n",
693                         vring, enable ? "enabled" : "disabled");
694
695         _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_QUEUE_STATE, NULL);
696
697         return 0;
698 }
699
700 static struct virtio_net_device_ops vhost_ops = {
701         .new_device          = new_device,
702         .destroy_device      = destroy_device,
703         .vring_state_changed = vring_state_changed,
704 };
705
706 int
707 rte_eth_vhost_get_queue_event(uint8_t port_id,
708                 struct rte_eth_vhost_queue_event *event)
709 {
710         struct rte_vhost_vring_state *state;
711         unsigned int i;
712         int idx;
713
714         if (port_id >= RTE_MAX_ETHPORTS) {
715                 RTE_LOG(ERR, PMD, "Invalid port id\n");
716                 return -1;
717         }
718
719         state = vring_states[port_id];
720         if (!state) {
721                 RTE_LOG(ERR, PMD, "Unused port\n");
722                 return -1;
723         }
724
725         rte_spinlock_lock(&state->lock);
726         for (i = 0; i <= state->max_vring; i++) {
727                 idx = state->index++ % (state->max_vring + 1);
728
729                 if (state->cur[idx] != state->seen[idx]) {
730                         state->seen[idx] = state->cur[idx];
731                         event->queue_id = idx / 2;
732                         event->rx = idx & 1;
733                         event->enable = state->cur[idx];
734                         rte_spinlock_unlock(&state->lock);
735                         return 0;
736                 }
737         }
738         rte_spinlock_unlock(&state->lock);
739
740         return -1;
741 }
742
743 int
744 rte_eth_vhost_get_vid_from_port_id(uint8_t port_id)
745 {
746         struct internal_list *list;
747         struct rte_eth_dev *eth_dev;
748         struct vhost_queue *vq;
749         int vid = -1;
750
751         if (!rte_eth_dev_is_valid_port(port_id))
752                 return -1;
753
754         pthread_mutex_lock(&internal_list_lock);
755
756         TAILQ_FOREACH(list, &internal_list, next) {
757                 eth_dev = list->eth_dev;
758                 if (eth_dev->data->port_id == port_id) {
759                         vq = eth_dev->data->rx_queues[0];
760                         if (vq) {
761                                 vid = vq->vid;
762                         }
763                         break;
764                 }
765         }
766
767         pthread_mutex_unlock(&internal_list_lock);
768
769         return vid;
770 }
771
772 static void *
773 vhost_driver_session(void *param __rte_unused)
774 {
775         /* start event handling */
776         rte_vhost_driver_session_start();
777
778         return NULL;
779 }
780
781 static int
782 vhost_driver_session_start(void)
783 {
784         int ret;
785
786         ret = pthread_create(&session_th,
787                         NULL, vhost_driver_session, NULL);
788         if (ret)
789                 RTE_LOG(ERR, PMD, "Can't create a thread\n");
790
791         return ret;
792 }
793
794 static void
795 vhost_driver_session_stop(void)
796 {
797         int ret;
798
799         ret = pthread_cancel(session_th);
800         if (ret)
801                 RTE_LOG(ERR, PMD, "Can't cancel the thread\n");
802
803         ret = pthread_join(session_th, NULL);
804         if (ret)
805                 RTE_LOG(ERR, PMD, "Can't join the thread\n");
806 }
807
808 static int
809 eth_dev_start(struct rte_eth_dev *dev)
810 {
811         struct pmd_internal *internal = dev->data->dev_private;
812
813         rte_atomic32_set(&internal->started, 1);
814         update_queuing_status(dev);
815
816         return 0;
817 }
818
819 static void
820 eth_dev_stop(struct rte_eth_dev *dev)
821 {
822         struct pmd_internal *internal = dev->data->dev_private;
823
824         rte_atomic32_set(&internal->started, 0);
825         update_queuing_status(dev);
826 }
827
828 static void
829 eth_dev_close(struct rte_eth_dev *dev)
830 {
831         struct pmd_internal *internal;
832         struct internal_list *list;
833
834         internal = dev->data->dev_private;
835         if (!internal)
836                 return;
837
838         rte_vhost_driver_unregister(internal->iface_name);
839
840         list = find_internal_resource(internal->iface_name);
841         if (!list)
842                 return;
843
844         pthread_mutex_lock(&internal_list_lock);
845         TAILQ_REMOVE(&internal_list, list, next);
846         pthread_mutex_unlock(&internal_list_lock);
847         rte_free(list);
848
849         free(internal->dev_name);
850         free(internal->iface_name);
851         rte_free(internal);
852 }
853
854 static int
855 eth_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
856                    uint16_t nb_rx_desc __rte_unused,
857                    unsigned int socket_id,
858                    const struct rte_eth_rxconf *rx_conf __rte_unused,
859                    struct rte_mempool *mb_pool)
860 {
861         struct vhost_queue *vq;
862
863         vq = rte_zmalloc_socket(NULL, sizeof(struct vhost_queue),
864                         RTE_CACHE_LINE_SIZE, socket_id);
865         if (vq == NULL) {
866                 RTE_LOG(ERR, PMD, "Failed to allocate memory for rx queue\n");
867                 return -ENOMEM;
868         }
869
870         vq->mb_pool = mb_pool;
871         vq->virtqueue_id = rx_queue_id * VIRTIO_QNUM + VIRTIO_TXQ;
872         dev->data->rx_queues[rx_queue_id] = vq;
873
874         return 0;
875 }
876
877 static int
878 eth_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
879                    uint16_t nb_tx_desc __rte_unused,
880                    unsigned int socket_id,
881                    const struct rte_eth_txconf *tx_conf __rte_unused)
882 {
883         struct vhost_queue *vq;
884
885         vq = rte_zmalloc_socket(NULL, sizeof(struct vhost_queue),
886                         RTE_CACHE_LINE_SIZE, socket_id);
887         if (vq == NULL) {
888                 RTE_LOG(ERR, PMD, "Failed to allocate memory for tx queue\n");
889                 return -ENOMEM;
890         }
891
892         vq->virtqueue_id = tx_queue_id * VIRTIO_QNUM + VIRTIO_RXQ;
893         dev->data->tx_queues[tx_queue_id] = vq;
894
895         return 0;
896 }
897
898 static void
899 eth_dev_info(struct rte_eth_dev *dev,
900              struct rte_eth_dev_info *dev_info)
901 {
902         struct pmd_internal *internal;
903
904         internal = dev->data->dev_private;
905         if (internal == NULL) {
906                 RTE_LOG(ERR, PMD, "Invalid device specified\n");
907                 return;
908         }
909
910         dev_info->max_mac_addrs = 1;
911         dev_info->max_rx_pktlen = (uint32_t)-1;
912         dev_info->max_rx_queues = internal->max_queues;
913         dev_info->max_tx_queues = internal->max_queues;
914         dev_info->min_rx_bufsize = 0;
915 }
916
917 static void
918 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
919 {
920         unsigned i;
921         unsigned long rx_total = 0, tx_total = 0, tx_missed_total = 0;
922         unsigned long rx_total_bytes = 0, tx_total_bytes = 0;
923         struct vhost_queue *vq;
924
925         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
926                         i < dev->data->nb_rx_queues; i++) {
927                 if (dev->data->rx_queues[i] == NULL)
928                         continue;
929                 vq = dev->data->rx_queues[i];
930                 stats->q_ipackets[i] = vq->stats.pkts;
931                 rx_total += stats->q_ipackets[i];
932
933                 stats->q_ibytes[i] = vq->stats.bytes;
934                 rx_total_bytes += stats->q_ibytes[i];
935         }
936
937         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
938                         i < dev->data->nb_tx_queues; i++) {
939                 if (dev->data->tx_queues[i] == NULL)
940                         continue;
941                 vq = dev->data->tx_queues[i];
942                 stats->q_opackets[i] = vq->stats.pkts;
943                 tx_missed_total += vq->stats.missed_pkts;
944                 tx_total += stats->q_opackets[i];
945
946                 stats->q_obytes[i] = vq->stats.bytes;
947                 tx_total_bytes += stats->q_obytes[i];
948         }
949
950         stats->ipackets = rx_total;
951         stats->opackets = tx_total;
952         stats->oerrors = tx_missed_total;
953         stats->ibytes = rx_total_bytes;
954         stats->obytes = tx_total_bytes;
955 }
956
957 static void
958 eth_stats_reset(struct rte_eth_dev *dev)
959 {
960         struct vhost_queue *vq;
961         unsigned i;
962
963         for (i = 0; i < dev->data->nb_rx_queues; i++) {
964                 if (dev->data->rx_queues[i] == NULL)
965                         continue;
966                 vq = dev->data->rx_queues[i];
967                 vq->stats.pkts = 0;
968                 vq->stats.bytes = 0;
969         }
970         for (i = 0; i < dev->data->nb_tx_queues; i++) {
971                 if (dev->data->tx_queues[i] == NULL)
972                         continue;
973                 vq = dev->data->tx_queues[i];
974                 vq->stats.pkts = 0;
975                 vq->stats.bytes = 0;
976                 vq->stats.missed_pkts = 0;
977         }
978 }
979
980 static void
981 eth_queue_release(void *q)
982 {
983         rte_free(q);
984 }
985
986 static int
987 eth_tx_done_cleanup(void *txq __rte_unused, uint32_t free_cnt __rte_unused)
988 {
989         /*
990          * vHost does not hang onto mbuf. eth_vhost_tx() copies packet data
991          * and releases mbuf, so nothing to cleanup.
992          */
993         return 0;
994 }
995
996 static int
997 eth_link_update(struct rte_eth_dev *dev __rte_unused,
998                 int wait_to_complete __rte_unused)
999 {
1000         return 0;
1001 }
1002
1003 static const struct eth_dev_ops ops = {
1004         .dev_start = eth_dev_start,
1005         .dev_stop = eth_dev_stop,
1006         .dev_close = eth_dev_close,
1007         .dev_configure = eth_dev_configure,
1008         .dev_infos_get = eth_dev_info,
1009         .rx_queue_setup = eth_rx_queue_setup,
1010         .tx_queue_setup = eth_tx_queue_setup,
1011         .rx_queue_release = eth_queue_release,
1012         .tx_queue_release = eth_queue_release,
1013         .tx_done_cleanup = eth_tx_done_cleanup,
1014         .link_update = eth_link_update,
1015         .stats_get = eth_stats_get,
1016         .stats_reset = eth_stats_reset,
1017         .xstats_reset = vhost_dev_xstats_reset,
1018         .xstats_get = vhost_dev_xstats_get,
1019         .xstats_get_names = vhost_dev_xstats_get_names,
1020 };
1021
1022 static struct rte_vdev_driver pmd_vhost_drv;
1023
1024 static int
1025 eth_dev_vhost_create(const char *name, char *iface_name, int16_t queues,
1026                      const unsigned numa_node, uint64_t flags)
1027 {
1028         struct rte_eth_dev_data *data = NULL;
1029         struct pmd_internal *internal = NULL;
1030         struct rte_eth_dev *eth_dev = NULL;
1031         struct ether_addr *eth_addr = NULL;
1032         struct rte_vhost_vring_state *vring_state = NULL;
1033         struct internal_list *list = NULL;
1034
1035         RTE_LOG(INFO, PMD, "Creating VHOST-USER backend on numa socket %u\n",
1036                 numa_node);
1037
1038         /* now do all data allocation - for eth_dev structure, dummy pci driver
1039          * and internal (private) data
1040          */
1041         data = rte_zmalloc_socket(name, sizeof(*data), 0, numa_node);
1042         if (data == NULL)
1043                 goto error;
1044
1045         internal = rte_zmalloc_socket(name, sizeof(*internal), 0, numa_node);
1046         if (internal == NULL)
1047                 goto error;
1048
1049         list = rte_zmalloc_socket(name, sizeof(*list), 0, numa_node);
1050         if (list == NULL)
1051                 goto error;
1052
1053         /* reserve an ethdev entry */
1054         eth_dev = rte_eth_dev_allocate(name);
1055         if (eth_dev == NULL)
1056                 goto error;
1057
1058         eth_addr = rte_zmalloc_socket(name, sizeof(*eth_addr), 0, numa_node);
1059         if (eth_addr == NULL)
1060                 goto error;
1061         *eth_addr = base_eth_addr;
1062         eth_addr->addr_bytes[5] = eth_dev->data->port_id;
1063
1064         vring_state = rte_zmalloc_socket(name,
1065                         sizeof(*vring_state), 0, numa_node);
1066         if (vring_state == NULL)
1067                 goto error;
1068
1069         /* now put it all together
1070          * - store queue data in internal,
1071          * - store numa_node info in ethdev data
1072          * - point eth_dev_data to internals
1073          * - and point eth_dev structure to new eth_dev_data structure
1074          */
1075         internal->dev_name = strdup(name);
1076         if (internal->dev_name == NULL)
1077                 goto error;
1078         internal->iface_name = strdup(iface_name);
1079         if (internal->iface_name == NULL)
1080                 goto error;
1081
1082         list->eth_dev = eth_dev;
1083         pthread_mutex_lock(&internal_list_lock);
1084         TAILQ_INSERT_TAIL(&internal_list, list, next);
1085         pthread_mutex_unlock(&internal_list_lock);
1086
1087         rte_spinlock_init(&vring_state->lock);
1088         vring_states[eth_dev->data->port_id] = vring_state;
1089
1090         data->dev_private = internal;
1091         data->port_id = eth_dev->data->port_id;
1092         memmove(data->name, eth_dev->data->name, sizeof(data->name));
1093         data->nb_rx_queues = queues;
1094         data->nb_tx_queues = queues;
1095         internal->max_queues = queues;
1096         data->dev_link = pmd_link;
1097         data->mac_addrs = eth_addr;
1098
1099         /* We'll replace the 'data' originally allocated by eth_dev. So the
1100          * vhost PMD resources won't be shared between multi processes.
1101          */
1102         eth_dev->data = data;
1103         eth_dev->dev_ops = &ops;
1104         eth_dev->driver = NULL;
1105         data->dev_flags =
1106                 RTE_ETH_DEV_DETACHABLE | RTE_ETH_DEV_INTR_LSC;
1107         data->kdrv = RTE_KDRV_NONE;
1108         data->drv_name = pmd_vhost_drv.driver.name;
1109         data->numa_node = numa_node;
1110
1111         /* finally assign rx and tx ops */
1112         eth_dev->rx_pkt_burst = eth_vhost_rx;
1113         eth_dev->tx_pkt_burst = eth_vhost_tx;
1114
1115         if (rte_vhost_driver_register(iface_name, flags))
1116                 goto error;
1117
1118         if (rte_vhost_driver_callback_register(iface_name, &vhost_ops) < 0) {
1119                 RTE_LOG(ERR, PMD, "Can't register callbacks\n");
1120                 goto error;
1121         }
1122
1123         /* We need only one message handling thread */
1124         if (rte_atomic16_add_return(&nb_started_ports, 1) == 1) {
1125                 if (vhost_driver_session_start())
1126                         goto error;
1127         }
1128
1129         return data->port_id;
1130
1131 error:
1132         if (internal)
1133                 free(internal->dev_name);
1134         rte_free(vring_state);
1135         rte_free(eth_addr);
1136         if (eth_dev)
1137                 rte_eth_dev_release_port(eth_dev);
1138         rte_free(internal);
1139         rte_free(list);
1140         rte_free(data);
1141
1142         return -1;
1143 }
1144
1145 static inline int
1146 open_iface(const char *key __rte_unused, const char *value, void *extra_args)
1147 {
1148         const char **iface_name = extra_args;
1149
1150         if (value == NULL)
1151                 return -1;
1152
1153         *iface_name = value;
1154
1155         return 0;
1156 }
1157
1158 static inline int
1159 open_int(const char *key __rte_unused, const char *value, void *extra_args)
1160 {
1161         uint16_t *n = extra_args;
1162
1163         if (value == NULL || extra_args == NULL)
1164                 return -EINVAL;
1165
1166         *n = (uint16_t)strtoul(value, NULL, 0);
1167         if (*n == USHRT_MAX && errno == ERANGE)
1168                 return -1;
1169
1170         return 0;
1171 }
1172
1173 static int
1174 rte_pmd_vhost_probe(const char *name, const char *params)
1175 {
1176         struct rte_kvargs *kvlist = NULL;
1177         int ret = 0;
1178         char *iface_name;
1179         uint16_t queues;
1180         uint64_t flags = 0;
1181         int client_mode = 0;
1182         int dequeue_zero_copy = 0;
1183
1184         RTE_LOG(INFO, PMD, "Initializing pmd_vhost for %s\n", name);
1185
1186         kvlist = rte_kvargs_parse(params, valid_arguments);
1187         if (kvlist == NULL)
1188                 return -1;
1189
1190         if (rte_kvargs_count(kvlist, ETH_VHOST_IFACE_ARG) == 1) {
1191                 ret = rte_kvargs_process(kvlist, ETH_VHOST_IFACE_ARG,
1192                                          &open_iface, &iface_name);
1193                 if (ret < 0)
1194                         goto out_free;
1195         } else {
1196                 ret = -1;
1197                 goto out_free;
1198         }
1199
1200         if (rte_kvargs_count(kvlist, ETH_VHOST_QUEUES_ARG) == 1) {
1201                 ret = rte_kvargs_process(kvlist, ETH_VHOST_QUEUES_ARG,
1202                                          &open_int, &queues);
1203                 if (ret < 0 || queues > RTE_MAX_QUEUES_PER_PORT)
1204                         goto out_free;
1205
1206         } else
1207                 queues = 1;
1208
1209         if (rte_kvargs_count(kvlist, ETH_VHOST_CLIENT_ARG) == 1) {
1210                 ret = rte_kvargs_process(kvlist, ETH_VHOST_CLIENT_ARG,
1211                                          &open_int, &client_mode);
1212                 if (ret < 0)
1213                         goto out_free;
1214
1215                 if (client_mode)
1216                         flags |= RTE_VHOST_USER_CLIENT;
1217         }
1218
1219         if (rte_kvargs_count(kvlist, ETH_VHOST_DEQUEUE_ZERO_COPY) == 1) {
1220                 ret = rte_kvargs_process(kvlist, ETH_VHOST_DEQUEUE_ZERO_COPY,
1221                                          &open_int, &dequeue_zero_copy);
1222                 if (ret < 0)
1223                         goto out_free;
1224
1225                 if (dequeue_zero_copy)
1226                         flags |= RTE_VHOST_USER_DEQUEUE_ZERO_COPY;
1227         }
1228
1229         eth_dev_vhost_create(name, iface_name, queues, rte_socket_id(), flags);
1230
1231 out_free:
1232         rte_kvargs_free(kvlist);
1233         return ret;
1234 }
1235
1236 static int
1237 rte_pmd_vhost_remove(const char *name)
1238 {
1239         struct rte_eth_dev *eth_dev = NULL;
1240         unsigned int i;
1241
1242         RTE_LOG(INFO, PMD, "Un-Initializing pmd_vhost for %s\n", name);
1243
1244         /* find an ethdev entry */
1245         eth_dev = rte_eth_dev_allocated(name);
1246         if (eth_dev == NULL)
1247                 return -ENODEV;
1248
1249         eth_dev_stop(eth_dev);
1250
1251         eth_dev_close(eth_dev);
1252
1253         if (rte_atomic16_sub_return(&nb_started_ports, 1) == 0)
1254                 vhost_driver_session_stop();
1255
1256         rte_free(vring_states[eth_dev->data->port_id]);
1257         vring_states[eth_dev->data->port_id] = NULL;
1258
1259         for (i = 0; i < eth_dev->data->nb_rx_queues; i++)
1260                 rte_free(eth_dev->data->rx_queues[i]);
1261         for (i = 0; i < eth_dev->data->nb_tx_queues; i++)
1262                 rte_free(eth_dev->data->tx_queues[i]);
1263
1264         rte_free(eth_dev->data->mac_addrs);
1265         rte_free(eth_dev->data);
1266
1267         rte_eth_dev_release_port(eth_dev);
1268
1269         return 0;
1270 }
1271
1272 static struct rte_vdev_driver pmd_vhost_drv = {
1273         .probe = rte_pmd_vhost_probe,
1274         .remove = rte_pmd_vhost_remove,
1275 };
1276
1277 RTE_PMD_REGISTER_VDEV(net_vhost, pmd_vhost_drv);
1278 RTE_PMD_REGISTER_ALIAS(net_vhost, eth_vhost);
1279 RTE_PMD_REGISTER_PARAM_STRING(net_vhost,
1280         "iface=<ifc> "
1281         "queues=<int>");