net/bnx2x: get cache line size from build configuration
[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 #ifdef RTE_LIBRTE_VHOST_NUMA
37 #include <numaif.h>
38 #endif
39
40 #include <rte_mbuf.h>
41 #include <rte_ethdev.h>
42 #include <rte_malloc.h>
43 #include <rte_memcpy.h>
44 #include <rte_vdev.h>
45 #include <rte_kvargs.h>
46 #include <rte_virtio_net.h>
47 #include <rte_spinlock.h>
48
49 #include "rte_eth_vhost.h"
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
56 static const char *drivername = "VHOST PMD";
57
58 static const char *valid_arguments[] = {
59         ETH_VHOST_IFACE_ARG,
60         ETH_VHOST_QUEUES_ARG,
61         ETH_VHOST_CLIENT_ARG,
62         ETH_VHOST_DEQUEUE_ZERO_COPY,
63         NULL
64 };
65
66 static struct ether_addr base_eth_addr = {
67         .addr_bytes = {
68                 0x56 /* V */,
69                 0x48 /* H */,
70                 0x4F /* O */,
71                 0x53 /* S */,
72                 0x54 /* T */,
73                 0x00
74         }
75 };
76
77 enum vhost_xstats_pkts {
78         VHOST_UNDERSIZE_PKT = 0,
79         VHOST_64_PKT,
80         VHOST_65_TO_127_PKT,
81         VHOST_128_TO_255_PKT,
82         VHOST_256_TO_511_PKT,
83         VHOST_512_TO_1023_PKT,
84         VHOST_1024_TO_1522_PKT,
85         VHOST_1523_TO_MAX_PKT,
86         VHOST_BROADCAST_PKT,
87         VHOST_MULTICAST_PKT,
88         VHOST_UNICAST_PKT,
89         VHOST_ERRORS_PKT,
90         VHOST_ERRORS_FRAGMENTED,
91         VHOST_ERRORS_JABBER,
92         VHOST_UNKNOWN_PROTOCOL,
93         VHOST_XSTATS_MAX,
94 };
95
96 struct vhost_stats {
97         uint64_t pkts;
98         uint64_t bytes;
99         uint64_t missed_pkts;
100         uint64_t xstats[VHOST_XSTATS_MAX];
101 };
102
103 struct vhost_queue {
104         int vid;
105         rte_atomic32_t allow_queuing;
106         rte_atomic32_t while_queuing;
107         struct pmd_internal *internal;
108         struct rte_mempool *mb_pool;
109         uint8_t port;
110         uint16_t virtqueue_id;
111         struct vhost_stats stats;
112 };
113
114 struct pmd_internal {
115         char *dev_name;
116         char *iface_name;
117         uint16_t max_queues;
118         uint64_t flags;
119
120         volatile uint16_t once;
121 };
122
123 struct internal_list {
124         TAILQ_ENTRY(internal_list) next;
125         struct rte_eth_dev *eth_dev;
126 };
127
128 TAILQ_HEAD(internal_list_head, internal_list);
129 static struct internal_list_head internal_list =
130         TAILQ_HEAD_INITIALIZER(internal_list);
131
132 static pthread_mutex_t internal_list_lock = PTHREAD_MUTEX_INITIALIZER;
133
134 static rte_atomic16_t nb_started_ports;
135 static pthread_t session_th;
136
137 static struct rte_eth_link pmd_link = {
138                 .link_speed = 10000,
139                 .link_duplex = ETH_LINK_FULL_DUPLEX,
140                 .link_status = ETH_LINK_DOWN
141 };
142
143 struct rte_vhost_vring_state {
144         rte_spinlock_t lock;
145
146         bool cur[RTE_MAX_QUEUES_PER_PORT * 2];
147         bool seen[RTE_MAX_QUEUES_PER_PORT * 2];
148         unsigned int index;
149         unsigned int max_vring;
150 };
151
152 static struct rte_vhost_vring_state *vring_states[RTE_MAX_ETHPORTS];
153
154 #define VHOST_XSTATS_NAME_SIZE 64
155
156 struct vhost_xstats_name_off {
157         char name[VHOST_XSTATS_NAME_SIZE];
158         uint64_t offset;
159 };
160
161 /* [rx]_is prepended to the name string here */
162 static const struct vhost_xstats_name_off vhost_rxport_stat_strings[] = {
163         {"good_packets",
164          offsetof(struct vhost_queue, stats.pkts)},
165         {"total_bytes",
166          offsetof(struct vhost_queue, stats.bytes)},
167         {"missed_pkts",
168          offsetof(struct vhost_queue, stats.missed_pkts)},
169         {"broadcast_packets",
170          offsetof(struct vhost_queue, stats.xstats[VHOST_BROADCAST_PKT])},
171         {"multicast_packets",
172          offsetof(struct vhost_queue, stats.xstats[VHOST_MULTICAST_PKT])},
173         {"unicast_packets",
174          offsetof(struct vhost_queue, stats.xstats[VHOST_UNICAST_PKT])},
175          {"undersize_packets",
176          offsetof(struct vhost_queue, stats.xstats[VHOST_UNDERSIZE_PKT])},
177         {"size_64_packets",
178          offsetof(struct vhost_queue, stats.xstats[VHOST_64_PKT])},
179         {"size_65_to_127_packets",
180          offsetof(struct vhost_queue, stats.xstats[VHOST_65_TO_127_PKT])},
181         {"size_128_to_255_packets",
182          offsetof(struct vhost_queue, stats.xstats[VHOST_128_TO_255_PKT])},
183         {"size_256_to_511_packets",
184          offsetof(struct vhost_queue, stats.xstats[VHOST_256_TO_511_PKT])},
185         {"size_512_to_1023_packets",
186          offsetof(struct vhost_queue, stats.xstats[VHOST_512_TO_1023_PKT])},
187         {"size_1024_to_1522_packets",
188          offsetof(struct vhost_queue, stats.xstats[VHOST_1024_TO_1522_PKT])},
189         {"size_1523_to_max_packets",
190          offsetof(struct vhost_queue, stats.xstats[VHOST_1523_TO_MAX_PKT])},
191         {"errors_with_bad_CRC",
192          offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_PKT])},
193         {"fragmented_errors",
194          offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_FRAGMENTED])},
195         {"jabber_errors",
196          offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_JABBER])},
197         {"unknown_protos_packets",
198          offsetof(struct vhost_queue, stats.xstats[VHOST_UNKNOWN_PROTOCOL])},
199 };
200
201 /* [tx]_ is prepended to the name string here */
202 static const struct vhost_xstats_name_off vhost_txport_stat_strings[] = {
203         {"good_packets",
204          offsetof(struct vhost_queue, stats.pkts)},
205         {"total_bytes",
206          offsetof(struct vhost_queue, stats.bytes)},
207         {"missed_pkts",
208          offsetof(struct vhost_queue, stats.missed_pkts)},
209         {"broadcast_packets",
210          offsetof(struct vhost_queue, stats.xstats[VHOST_BROADCAST_PKT])},
211         {"multicast_packets",
212          offsetof(struct vhost_queue, stats.xstats[VHOST_MULTICAST_PKT])},
213         {"unicast_packets",
214          offsetof(struct vhost_queue, stats.xstats[VHOST_UNICAST_PKT])},
215         {"undersize_packets",
216          offsetof(struct vhost_queue, stats.xstats[VHOST_UNDERSIZE_PKT])},
217         {"size_64_packets",
218          offsetof(struct vhost_queue, stats.xstats[VHOST_64_PKT])},
219         {"size_65_to_127_packets",
220          offsetof(struct vhost_queue, stats.xstats[VHOST_65_TO_127_PKT])},
221         {"size_128_to_255_packets",
222          offsetof(struct vhost_queue, stats.xstats[VHOST_128_TO_255_PKT])},
223         {"size_256_to_511_packets",
224          offsetof(struct vhost_queue, stats.xstats[VHOST_256_TO_511_PKT])},
225         {"size_512_to_1023_packets",
226          offsetof(struct vhost_queue, stats.xstats[VHOST_512_TO_1023_PKT])},
227         {"size_1024_to_1522_packets",
228          offsetof(struct vhost_queue, stats.xstats[VHOST_1024_TO_1522_PKT])},
229         {"size_1523_to_max_packets",
230          offsetof(struct vhost_queue, stats.xstats[VHOST_1523_TO_MAX_PKT])},
231         {"errors_with_bad_CRC",
232          offsetof(struct vhost_queue, stats.xstats[VHOST_ERRORS_PKT])},
233 };
234
235 #define VHOST_NB_XSTATS_RXPORT (sizeof(vhost_rxport_stat_strings) / \
236                                 sizeof(vhost_rxport_stat_strings[0]))
237
238 #define VHOST_NB_XSTATS_TXPORT (sizeof(vhost_txport_stat_strings) / \
239                                 sizeof(vhost_txport_stat_strings[0]))
240
241 static void
242 vhost_dev_xstats_reset(struct rte_eth_dev *dev)
243 {
244         struct vhost_queue *vq = NULL;
245         unsigned int i = 0;
246
247         for (i = 0; i < dev->data->nb_rx_queues; i++) {
248                 vq = dev->data->rx_queues[i];
249                 if (!vq)
250                         continue;
251                 memset(&vq->stats, 0, sizeof(vq->stats));
252         }
253         for (i = 0; i < dev->data->nb_tx_queues; i++) {
254                 vq = dev->data->tx_queues[i];
255                 if (!vq)
256                         continue;
257                 memset(&vq->stats, 0, sizeof(vq->stats));
258         }
259 }
260
261 static int
262 vhost_dev_xstats_get_names(struct rte_eth_dev *dev __rte_unused,
263                            struct rte_eth_xstat_name *xstats_names,
264                            unsigned int limit __rte_unused)
265 {
266         unsigned int t = 0;
267         int count = 0;
268         int nstats = VHOST_NB_XSTATS_RXPORT + VHOST_NB_XSTATS_TXPORT;
269
270         if (!xstats_names)
271                 return nstats;
272         for (t = 0; t < VHOST_NB_XSTATS_RXPORT; t++) {
273                 snprintf(xstats_names[count].name,
274                          sizeof(xstats_names[count].name),
275                          "rx_%s", vhost_rxport_stat_strings[t].name);
276                 count++;
277         }
278         for (t = 0; t < VHOST_NB_XSTATS_TXPORT; t++) {
279                 snprintf(xstats_names[count].name,
280                          sizeof(xstats_names[count].name),
281                          "tx_%s", vhost_txport_stat_strings[t].name);
282                 count++;
283         }
284         return count;
285 }
286
287 static int
288 vhost_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
289                      unsigned int n)
290 {
291         unsigned int i;
292         unsigned int t;
293         unsigned int count = 0;
294         struct vhost_queue *vq = NULL;
295         unsigned int nxstats = VHOST_NB_XSTATS_RXPORT + VHOST_NB_XSTATS_TXPORT;
296
297         if (n < nxstats)
298                 return nxstats;
299
300         for (i = 0; i < dev->data->nb_rx_queues; i++) {
301                 vq = dev->data->rx_queues[i];
302                 if (!vq)
303                         continue;
304                 vq->stats.xstats[VHOST_UNICAST_PKT] = vq->stats.pkts
305                                 - (vq->stats.xstats[VHOST_BROADCAST_PKT]
306                                 + vq->stats.xstats[VHOST_MULTICAST_PKT]);
307         }
308         for (i = 0; i < dev->data->nb_tx_queues; i++) {
309                 vq = dev->data->tx_queues[i];
310                 if (!vq)
311                         continue;
312                 vq->stats.xstats[VHOST_UNICAST_PKT] = vq->stats.pkts
313                                 + vq->stats.missed_pkts
314                                 - (vq->stats.xstats[VHOST_BROADCAST_PKT]
315                                 + vq->stats.xstats[VHOST_MULTICAST_PKT]);
316         }
317         for (t = 0; t < VHOST_NB_XSTATS_RXPORT; t++) {
318                 xstats[count].value = 0;
319                 for (i = 0; i < dev->data->nb_rx_queues; i++) {
320                         vq = dev->data->rx_queues[i];
321                         if (!vq)
322                                 continue;
323                         xstats[count].value +=
324                                 *(uint64_t *)(((char *)vq)
325                                 + vhost_rxport_stat_strings[t].offset);
326                 }
327                 count++;
328         }
329         for (t = 0; t < VHOST_NB_XSTATS_TXPORT; t++) {
330                 xstats[count].value = 0;
331                 for (i = 0; i < dev->data->nb_tx_queues; i++) {
332                         vq = dev->data->tx_queues[i];
333                         if (!vq)
334                                 continue;
335                         xstats[count].value +=
336                                 *(uint64_t *)(((char *)vq)
337                                 + vhost_txport_stat_strings[t].offset);
338                 }
339                 count++;
340         }
341         return count;
342 }
343
344 static inline void
345 vhost_count_multicast_broadcast(struct vhost_queue *vq,
346                                 struct rte_mbuf *mbuf)
347 {
348         struct ether_addr *ea = NULL;
349         struct vhost_stats *pstats = &vq->stats;
350
351         ea = rte_pktmbuf_mtod(mbuf, struct ether_addr *);
352         if (is_multicast_ether_addr(ea)) {
353                 if (is_broadcast_ether_addr(ea))
354                         pstats->xstats[VHOST_BROADCAST_PKT]++;
355                 else
356                         pstats->xstats[VHOST_MULTICAST_PKT]++;
357         }
358 }
359
360 static void
361 vhost_update_packet_xstats(struct vhost_queue *vq,
362                            struct rte_mbuf **bufs,
363                            uint16_t count)
364 {
365         uint32_t pkt_len = 0;
366         uint64_t i = 0;
367         uint64_t index;
368         struct vhost_stats *pstats = &vq->stats;
369
370         for (i = 0; i < count ; i++) {
371                 pkt_len = bufs[i]->pkt_len;
372                 if (pkt_len == 64) {
373                         pstats->xstats[VHOST_64_PKT]++;
374                 } else if (pkt_len > 64 && pkt_len < 1024) {
375                         index = (sizeof(pkt_len) * 8)
376                                 - __builtin_clz(pkt_len) - 5;
377                         pstats->xstats[index]++;
378                 } else {
379                         if (pkt_len < 64)
380                                 pstats->xstats[VHOST_UNDERSIZE_PKT]++;
381                         else if (pkt_len <= 1522)
382                                 pstats->xstats[VHOST_1024_TO_1522_PKT]++;
383                         else if (pkt_len > 1522)
384                                 pstats->xstats[VHOST_1523_TO_MAX_PKT]++;
385                 }
386                 vhost_count_multicast_broadcast(vq, bufs[i]);
387         }
388 }
389
390 static uint16_t
391 eth_vhost_rx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
392 {
393         struct vhost_queue *r = q;
394         uint16_t i, nb_rx = 0;
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         nb_rx = rte_vhost_dequeue_burst(r->vid,
406                         r->virtqueue_id, r->mb_pool, bufs, nb_bufs);
407
408         r->stats.pkts += nb_rx;
409
410         for (i = 0; likely(i < nb_rx); i++) {
411                 bufs[i]->port = r->port;
412                 r->stats.bytes += bufs[i]->pkt_len;
413         }
414
415         vhost_update_packet_xstats(r, bufs, nb_rx);
416
417 out:
418         rte_atomic32_set(&r->while_queuing, 0);
419
420         return nb_rx;
421 }
422
423 static uint16_t
424 eth_vhost_tx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
425 {
426         struct vhost_queue *r = q;
427         uint16_t i, nb_tx = 0;
428
429         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
430                 return 0;
431
432         rte_atomic32_set(&r->while_queuing, 1);
433
434         if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
435                 goto out;
436
437         /* Enqueue packets to guest RX queue */
438         nb_tx = rte_vhost_enqueue_burst(r->vid,
439                         r->virtqueue_id, bufs, nb_bufs);
440
441         r->stats.pkts += nb_tx;
442         r->stats.missed_pkts += nb_bufs - nb_tx;
443
444         for (i = 0; likely(i < nb_tx); i++)
445                 r->stats.bytes += bufs[i]->pkt_len;
446
447         vhost_update_packet_xstats(r, bufs, nb_tx);
448
449         /* According to RFC2863 page42 section ifHCOutMulticastPkts and
450          * ifHCOutBroadcastPkts, the counters "multicast" and "broadcast"
451          * are increased when packets are not transmitted successfully.
452          */
453         for (i = nb_tx; i < nb_bufs; i++)
454                 vhost_count_multicast_broadcast(r, bufs[i]);
455
456         for (i = 0; likely(i < nb_tx); i++)
457                 rte_pktmbuf_free(bufs[i]);
458 out:
459         rte_atomic32_set(&r->while_queuing, 0);
460
461         return nb_tx;
462 }
463
464 static int
465 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
466 {
467         return 0;
468 }
469
470 static inline struct internal_list *
471 find_internal_resource(char *ifname)
472 {
473         int found = 0;
474         struct internal_list *list;
475         struct pmd_internal *internal;
476
477         if (ifname == NULL)
478                 return NULL;
479
480         pthread_mutex_lock(&internal_list_lock);
481
482         TAILQ_FOREACH(list, &internal_list, next) {
483                 internal = list->eth_dev->data->dev_private;
484                 if (!strcmp(internal->iface_name, ifname)) {
485                         found = 1;
486                         break;
487                 }
488         }
489
490         pthread_mutex_unlock(&internal_list_lock);
491
492         if (!found)
493                 return NULL;
494
495         return list;
496 }
497
498 static int
499 new_device(int vid)
500 {
501         struct rte_eth_dev *eth_dev;
502         struct internal_list *list;
503         struct pmd_internal *internal;
504         struct vhost_queue *vq;
505         unsigned i;
506         char ifname[PATH_MAX];
507 #ifdef RTE_LIBRTE_VHOST_NUMA
508         int newnode;
509 #endif
510
511         rte_vhost_get_ifname(vid, ifname, sizeof(ifname));
512         list = find_internal_resource(ifname);
513         if (list == NULL) {
514                 RTE_LOG(INFO, PMD, "Invalid device name: %s\n", ifname);
515                 return -1;
516         }
517
518         eth_dev = list->eth_dev;
519         internal = eth_dev->data->dev_private;
520
521 #ifdef RTE_LIBRTE_VHOST_NUMA
522         newnode = rte_vhost_get_numa_node(vid);
523         if (newnode >= 0)
524                 eth_dev->data->numa_node = newnode;
525 #endif
526
527         for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
528                 vq = eth_dev->data->rx_queues[i];
529                 if (vq == NULL)
530                         continue;
531                 vq->vid = vid;
532                 vq->internal = internal;
533                 vq->port = eth_dev->data->port_id;
534         }
535         for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
536                 vq = eth_dev->data->tx_queues[i];
537                 if (vq == NULL)
538                         continue;
539                 vq->vid = vid;
540                 vq->internal = internal;
541                 vq->port = eth_dev->data->port_id;
542         }
543
544         for (i = 0; i < rte_vhost_get_queue_num(vid) * VIRTIO_QNUM; i++)
545                 rte_vhost_enable_guest_notification(vid, i, 0);
546
547         eth_dev->data->dev_link.link_status = ETH_LINK_UP;
548
549         for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
550                 vq = eth_dev->data->rx_queues[i];
551                 if (vq == NULL)
552                         continue;
553                 rte_atomic32_set(&vq->allow_queuing, 1);
554         }
555         for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
556                 vq = eth_dev->data->tx_queues[i];
557                 if (vq == NULL)
558                         continue;
559                 rte_atomic32_set(&vq->allow_queuing, 1);
560         }
561
562         RTE_LOG(INFO, PMD, "New connection established\n");
563
564         _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_INTR_LSC);
565
566         return 0;
567 }
568
569 static void
570 destroy_device(int vid)
571 {
572         struct rte_eth_dev *eth_dev;
573         struct vhost_queue *vq;
574         struct internal_list *list;
575         char ifname[PATH_MAX];
576         unsigned i;
577         struct rte_vhost_vring_state *state;
578
579         rte_vhost_get_ifname(vid, ifname, sizeof(ifname));
580         list = find_internal_resource(ifname);
581         if (list == NULL) {
582                 RTE_LOG(ERR, PMD, "Invalid interface name: %s\n", ifname);
583                 return;
584         }
585         eth_dev = list->eth_dev;
586
587         /* Wait until rx/tx_pkt_burst stops accessing vhost device */
588         for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
589                 vq = eth_dev->data->rx_queues[i];
590                 if (vq == NULL)
591                         continue;
592                 rte_atomic32_set(&vq->allow_queuing, 0);
593                 while (rte_atomic32_read(&vq->while_queuing))
594                         rte_pause();
595         }
596         for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
597                 vq = eth_dev->data->tx_queues[i];
598                 if (vq == NULL)
599                         continue;
600                 rte_atomic32_set(&vq->allow_queuing, 0);
601                 while (rte_atomic32_read(&vq->while_queuing))
602                         rte_pause();
603         }
604
605         eth_dev->data->dev_link.link_status = ETH_LINK_DOWN;
606
607         for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
608                 vq = eth_dev->data->rx_queues[i];
609                 if (vq == NULL)
610                         continue;
611                 vq->vid = -1;
612         }
613         for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
614                 vq = eth_dev->data->tx_queues[i];
615                 if (vq == NULL)
616                         continue;
617                 vq->vid = -1;
618         }
619
620         state = vring_states[eth_dev->data->port_id];
621         rte_spinlock_lock(&state->lock);
622         for (i = 0; i <= state->max_vring; i++) {
623                 state->cur[i] = false;
624                 state->seen[i] = false;
625         }
626         state->max_vring = 0;
627         rte_spinlock_unlock(&state->lock);
628
629         RTE_LOG(INFO, PMD, "Connection closed\n");
630
631         _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_INTR_LSC);
632 }
633
634 static int
635 vring_state_changed(int vid, uint16_t vring, int enable)
636 {
637         struct rte_vhost_vring_state *state;
638         struct rte_eth_dev *eth_dev;
639         struct internal_list *list;
640         char ifname[PATH_MAX];
641
642         rte_vhost_get_ifname(vid, ifname, sizeof(ifname));
643         list = find_internal_resource(ifname);
644         if (list == NULL) {
645                 RTE_LOG(ERR, PMD, "Invalid interface name: %s\n", ifname);
646                 return -1;
647         }
648
649         eth_dev = list->eth_dev;
650         /* won't be NULL */
651         state = vring_states[eth_dev->data->port_id];
652         rte_spinlock_lock(&state->lock);
653         state->cur[vring] = enable;
654         state->max_vring = RTE_MAX(vring, state->max_vring);
655         rte_spinlock_unlock(&state->lock);
656
657         RTE_LOG(INFO, PMD, "vring%u is %s\n",
658                         vring, enable ? "enabled" : "disabled");
659
660         _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_QUEUE_STATE);
661
662         return 0;
663 }
664
665 int
666 rte_eth_vhost_get_queue_event(uint8_t port_id,
667                 struct rte_eth_vhost_queue_event *event)
668 {
669         struct rte_vhost_vring_state *state;
670         unsigned int i;
671         int idx;
672
673         if (port_id >= RTE_MAX_ETHPORTS) {
674                 RTE_LOG(ERR, PMD, "Invalid port id\n");
675                 return -1;
676         }
677
678         state = vring_states[port_id];
679         if (!state) {
680                 RTE_LOG(ERR, PMD, "Unused port\n");
681                 return -1;
682         }
683
684         rte_spinlock_lock(&state->lock);
685         for (i = 0; i <= state->max_vring; i++) {
686                 idx = state->index++ % (state->max_vring + 1);
687
688                 if (state->cur[idx] != state->seen[idx]) {
689                         state->seen[idx] = state->cur[idx];
690                         event->queue_id = idx / 2;
691                         event->rx = idx & 1;
692                         event->enable = state->cur[idx];
693                         rte_spinlock_unlock(&state->lock);
694                         return 0;
695                 }
696         }
697         rte_spinlock_unlock(&state->lock);
698
699         return -1;
700 }
701
702 int
703 rte_eth_vhost_get_vid_from_port_id(uint8_t port_id)
704 {
705         struct internal_list *list;
706         struct rte_eth_dev *eth_dev;
707         struct vhost_queue *vq;
708         int vid = -1;
709
710         if (!rte_eth_dev_is_valid_port(port_id))
711                 return -1;
712
713         pthread_mutex_lock(&internal_list_lock);
714
715         TAILQ_FOREACH(list, &internal_list, next) {
716                 eth_dev = list->eth_dev;
717                 if (eth_dev->data->port_id == port_id) {
718                         vq = eth_dev->data->rx_queues[0];
719                         if (vq) {
720                                 vid = vq->vid;
721                         }
722                         break;
723                 }
724         }
725
726         pthread_mutex_unlock(&internal_list_lock);
727
728         return vid;
729 }
730
731 static void *
732 vhost_driver_session(void *param __rte_unused)
733 {
734         static struct virtio_net_device_ops vhost_ops;
735
736         /* set vhost arguments */
737         vhost_ops.new_device = new_device;
738         vhost_ops.destroy_device = destroy_device;
739         vhost_ops.vring_state_changed = vring_state_changed;
740         if (rte_vhost_driver_callback_register(&vhost_ops) < 0)
741                 RTE_LOG(ERR, PMD, "Can't register callbacks\n");
742
743         /* start event handling */
744         rte_vhost_driver_session_start();
745
746         return NULL;
747 }
748
749 static int
750 vhost_driver_session_start(void)
751 {
752         int ret;
753
754         ret = pthread_create(&session_th,
755                         NULL, vhost_driver_session, NULL);
756         if (ret)
757                 RTE_LOG(ERR, PMD, "Can't create a thread\n");
758
759         return ret;
760 }
761
762 static void
763 vhost_driver_session_stop(void)
764 {
765         int ret;
766
767         ret = pthread_cancel(session_th);
768         if (ret)
769                 RTE_LOG(ERR, PMD, "Can't cancel the thread\n");
770
771         ret = pthread_join(session_th, NULL);
772         if (ret)
773                 RTE_LOG(ERR, PMD, "Can't join the thread\n");
774 }
775
776 static int
777 eth_dev_start(struct rte_eth_dev *dev)
778 {
779         struct pmd_internal *internal = dev->data->dev_private;
780         int ret = 0;
781
782         if (rte_atomic16_cmpset(&internal->once, 0, 1)) {
783                 ret = rte_vhost_driver_register(internal->iface_name,
784                                                 internal->flags);
785                 if (ret)
786                         return ret;
787         }
788
789         /* We need only one message handling thread */
790         if (rte_atomic16_add_return(&nb_started_ports, 1) == 1)
791                 ret = vhost_driver_session_start();
792
793         return ret;
794 }
795
796 static void
797 eth_dev_stop(struct rte_eth_dev *dev)
798 {
799         struct pmd_internal *internal = dev->data->dev_private;
800
801         if (rte_atomic16_cmpset(&internal->once, 1, 0))
802                 rte_vhost_driver_unregister(internal->iface_name);
803
804         if (rte_atomic16_sub_return(&nb_started_ports, 1) == 0)
805                 vhost_driver_session_stop();
806 }
807
808 static int
809 eth_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
810                    uint16_t nb_rx_desc __rte_unused,
811                    unsigned int socket_id,
812                    const struct rte_eth_rxconf *rx_conf __rte_unused,
813                    struct rte_mempool *mb_pool)
814 {
815         struct vhost_queue *vq;
816
817         vq = rte_zmalloc_socket(NULL, sizeof(struct vhost_queue),
818                         RTE_CACHE_LINE_SIZE, socket_id);
819         if (vq == NULL) {
820                 RTE_LOG(ERR, PMD, "Failed to allocate memory for rx queue\n");
821                 return -ENOMEM;
822         }
823
824         vq->mb_pool = mb_pool;
825         vq->virtqueue_id = rx_queue_id * VIRTIO_QNUM + VIRTIO_TXQ;
826         dev->data->rx_queues[rx_queue_id] = vq;
827
828         return 0;
829 }
830
831 static int
832 eth_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
833                    uint16_t nb_tx_desc __rte_unused,
834                    unsigned int socket_id,
835                    const struct rte_eth_txconf *tx_conf __rte_unused)
836 {
837         struct vhost_queue *vq;
838
839         vq = rte_zmalloc_socket(NULL, sizeof(struct vhost_queue),
840                         RTE_CACHE_LINE_SIZE, socket_id);
841         if (vq == NULL) {
842                 RTE_LOG(ERR, PMD, "Failed to allocate memory for tx queue\n");
843                 return -ENOMEM;
844         }
845
846         vq->virtqueue_id = tx_queue_id * VIRTIO_QNUM + VIRTIO_RXQ;
847         dev->data->tx_queues[tx_queue_id] = vq;
848
849         return 0;
850 }
851
852 static void
853 eth_dev_info(struct rte_eth_dev *dev,
854              struct rte_eth_dev_info *dev_info)
855 {
856         struct pmd_internal *internal;
857
858         internal = dev->data->dev_private;
859         if (internal == NULL) {
860                 RTE_LOG(ERR, PMD, "Invalid device specified\n");
861                 return;
862         }
863
864         dev_info->driver_name = drivername;
865         dev_info->max_mac_addrs = 1;
866         dev_info->max_rx_pktlen = (uint32_t)-1;
867         dev_info->max_rx_queues = internal->max_queues;
868         dev_info->max_tx_queues = internal->max_queues;
869         dev_info->min_rx_bufsize = 0;
870 }
871
872 static void
873 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
874 {
875         unsigned i;
876         unsigned long rx_total = 0, tx_total = 0, tx_missed_total = 0;
877         unsigned long rx_total_bytes = 0, tx_total_bytes = 0;
878         struct vhost_queue *vq;
879
880         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
881                         i < dev->data->nb_rx_queues; i++) {
882                 if (dev->data->rx_queues[i] == NULL)
883                         continue;
884                 vq = dev->data->rx_queues[i];
885                 stats->q_ipackets[i] = vq->stats.pkts;
886                 rx_total += stats->q_ipackets[i];
887
888                 stats->q_ibytes[i] = vq->stats.bytes;
889                 rx_total_bytes += stats->q_ibytes[i];
890         }
891
892         for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
893                         i < dev->data->nb_tx_queues; i++) {
894                 if (dev->data->tx_queues[i] == NULL)
895                         continue;
896                 vq = dev->data->tx_queues[i];
897                 stats->q_opackets[i] = vq->stats.pkts;
898                 tx_missed_total += vq->stats.missed_pkts;
899                 tx_total += stats->q_opackets[i];
900
901                 stats->q_obytes[i] = vq->stats.bytes;
902                 tx_total_bytes += stats->q_obytes[i];
903         }
904
905         stats->ipackets = rx_total;
906         stats->opackets = tx_total;
907         stats->oerrors = tx_missed_total;
908         stats->ibytes = rx_total_bytes;
909         stats->obytes = tx_total_bytes;
910 }
911
912 static void
913 eth_stats_reset(struct rte_eth_dev *dev)
914 {
915         struct vhost_queue *vq;
916         unsigned i;
917
918         for (i = 0; i < dev->data->nb_rx_queues; i++) {
919                 if (dev->data->rx_queues[i] == NULL)
920                         continue;
921                 vq = dev->data->rx_queues[i];
922                 vq->stats.pkts = 0;
923                 vq->stats.bytes = 0;
924         }
925         for (i = 0; i < dev->data->nb_tx_queues; i++) {
926                 if (dev->data->tx_queues[i] == NULL)
927                         continue;
928                 vq = dev->data->tx_queues[i];
929                 vq->stats.pkts = 0;
930                 vq->stats.bytes = 0;
931                 vq->stats.missed_pkts = 0;
932         }
933 }
934
935 static void
936 eth_queue_release(void *q)
937 {
938         rte_free(q);
939 }
940
941 static int
942 eth_link_update(struct rte_eth_dev *dev __rte_unused,
943                 int wait_to_complete __rte_unused)
944 {
945         return 0;
946 }
947
948 /**
949  * Disable features in feature_mask. Returns 0 on success.
950  */
951 int
952 rte_eth_vhost_feature_disable(uint64_t feature_mask)
953 {
954         return rte_vhost_feature_disable(feature_mask);
955 }
956
957 /**
958  * Enable features in feature_mask. Returns 0 on success.
959  */
960 int
961 rte_eth_vhost_feature_enable(uint64_t feature_mask)
962 {
963         return rte_vhost_feature_enable(feature_mask);
964 }
965
966 /* Returns currently supported vhost features */
967 uint64_t
968 rte_eth_vhost_feature_get(void)
969 {
970         return rte_vhost_feature_get();
971 }
972
973 static const struct eth_dev_ops ops = {
974         .dev_start = eth_dev_start,
975         .dev_stop = eth_dev_stop,
976         .dev_configure = eth_dev_configure,
977         .dev_infos_get = eth_dev_info,
978         .rx_queue_setup = eth_rx_queue_setup,
979         .tx_queue_setup = eth_tx_queue_setup,
980         .rx_queue_release = eth_queue_release,
981         .tx_queue_release = eth_queue_release,
982         .link_update = eth_link_update,
983         .stats_get = eth_stats_get,
984         .stats_reset = eth_stats_reset,
985         .xstats_reset = vhost_dev_xstats_reset,
986         .xstats_get = vhost_dev_xstats_get,
987         .xstats_get_names = vhost_dev_xstats_get_names,
988 };
989
990 static int
991 eth_dev_vhost_create(const char *name, char *iface_name, int16_t queues,
992                      const unsigned numa_node, uint64_t flags)
993 {
994         struct rte_eth_dev_data *data = NULL;
995         struct pmd_internal *internal = NULL;
996         struct rte_eth_dev *eth_dev = NULL;
997         struct ether_addr *eth_addr = NULL;
998         struct rte_vhost_vring_state *vring_state = NULL;
999         struct internal_list *list = NULL;
1000
1001         RTE_LOG(INFO, PMD, "Creating VHOST-USER backend on numa socket %u\n",
1002                 numa_node);
1003
1004         /* now do all data allocation - for eth_dev structure, dummy pci driver
1005          * and internal (private) data
1006          */
1007         data = rte_zmalloc_socket(name, sizeof(*data), 0, numa_node);
1008         if (data == NULL)
1009                 goto error;
1010
1011         internal = rte_zmalloc_socket(name, sizeof(*internal), 0, numa_node);
1012         if (internal == NULL)
1013                 goto error;
1014
1015         list = rte_zmalloc_socket(name, sizeof(*list), 0, numa_node);
1016         if (list == NULL)
1017                 goto error;
1018
1019         /* reserve an ethdev entry */
1020         eth_dev = rte_eth_dev_allocate(name);
1021         if (eth_dev == NULL)
1022                 goto error;
1023
1024         eth_addr = rte_zmalloc_socket(name, sizeof(*eth_addr), 0, numa_node);
1025         if (eth_addr == NULL)
1026                 goto error;
1027         *eth_addr = base_eth_addr;
1028         eth_addr->addr_bytes[5] = eth_dev->data->port_id;
1029
1030         vring_state = rte_zmalloc_socket(name,
1031                         sizeof(*vring_state), 0, numa_node);
1032         if (vring_state == NULL)
1033                 goto error;
1034
1035         TAILQ_INIT(&eth_dev->link_intr_cbs);
1036
1037         /* now put it all together
1038          * - store queue data in internal,
1039          * - store numa_node info in ethdev data
1040          * - point eth_dev_data to internals
1041          * - and point eth_dev structure to new eth_dev_data structure
1042          */
1043         internal->dev_name = strdup(name);
1044         if (internal->dev_name == NULL)
1045                 goto error;
1046         internal->iface_name = strdup(iface_name);
1047         if (internal->iface_name == NULL)
1048                 goto error;
1049         internal->flags = flags;
1050
1051         list->eth_dev = eth_dev;
1052         pthread_mutex_lock(&internal_list_lock);
1053         TAILQ_INSERT_TAIL(&internal_list, list, next);
1054         pthread_mutex_unlock(&internal_list_lock);
1055
1056         rte_spinlock_init(&vring_state->lock);
1057         vring_states[eth_dev->data->port_id] = vring_state;
1058
1059         data->dev_private = internal;
1060         data->port_id = eth_dev->data->port_id;
1061         memmove(data->name, eth_dev->data->name, sizeof(data->name));
1062         data->nb_rx_queues = queues;
1063         data->nb_tx_queues = queues;
1064         internal->max_queues = queues;
1065         data->dev_link = pmd_link;
1066         data->mac_addrs = eth_addr;
1067
1068         /* We'll replace the 'data' originally allocated by eth_dev. So the
1069          * vhost PMD resources won't be shared between multi processes.
1070          */
1071         eth_dev->data = data;
1072         eth_dev->dev_ops = &ops;
1073         eth_dev->driver = NULL;
1074         data->dev_flags =
1075                 RTE_ETH_DEV_DETACHABLE | RTE_ETH_DEV_INTR_LSC;
1076         data->kdrv = RTE_KDRV_NONE;
1077         data->drv_name = internal->dev_name;
1078         data->numa_node = numa_node;
1079
1080         /* finally assign rx and tx ops */
1081         eth_dev->rx_pkt_burst = eth_vhost_rx;
1082         eth_dev->tx_pkt_burst = eth_vhost_tx;
1083
1084         return data->port_id;
1085
1086 error:
1087         if (internal)
1088                 free(internal->dev_name);
1089         rte_free(vring_state);
1090         rte_free(eth_addr);
1091         if (eth_dev)
1092                 rte_eth_dev_release_port(eth_dev);
1093         rte_free(internal);
1094         rte_free(list);
1095         rte_free(data);
1096
1097         return -1;
1098 }
1099
1100 static inline int
1101 open_iface(const char *key __rte_unused, const char *value, void *extra_args)
1102 {
1103         const char **iface_name = extra_args;
1104
1105         if (value == NULL)
1106                 return -1;
1107
1108         *iface_name = value;
1109
1110         return 0;
1111 }
1112
1113 static inline int
1114 open_int(const char *key __rte_unused, const char *value, void *extra_args)
1115 {
1116         uint16_t *n = extra_args;
1117
1118         if (value == NULL || extra_args == NULL)
1119                 return -EINVAL;
1120
1121         *n = (uint16_t)strtoul(value, NULL, 0);
1122         if (*n == USHRT_MAX && errno == ERANGE)
1123                 return -1;
1124
1125         return 0;
1126 }
1127
1128 static int
1129 rte_pmd_vhost_probe(const char *name, const char *params)
1130 {
1131         struct rte_kvargs *kvlist = NULL;
1132         int ret = 0;
1133         char *iface_name;
1134         uint16_t queues;
1135         uint64_t flags = 0;
1136         int client_mode = 0;
1137         int dequeue_zero_copy = 0;
1138
1139         RTE_LOG(INFO, PMD, "Initializing pmd_vhost for %s\n", name);
1140
1141         kvlist = rte_kvargs_parse(params, valid_arguments);
1142         if (kvlist == NULL)
1143                 return -1;
1144
1145         if (rte_kvargs_count(kvlist, ETH_VHOST_IFACE_ARG) == 1) {
1146                 ret = rte_kvargs_process(kvlist, ETH_VHOST_IFACE_ARG,
1147                                          &open_iface, &iface_name);
1148                 if (ret < 0)
1149                         goto out_free;
1150         } else {
1151                 ret = -1;
1152                 goto out_free;
1153         }
1154
1155         if (rte_kvargs_count(kvlist, ETH_VHOST_QUEUES_ARG) == 1) {
1156                 ret = rte_kvargs_process(kvlist, ETH_VHOST_QUEUES_ARG,
1157                                          &open_int, &queues);
1158                 if (ret < 0 || queues > RTE_MAX_QUEUES_PER_PORT)
1159                         goto out_free;
1160
1161         } else
1162                 queues = 1;
1163
1164         if (rte_kvargs_count(kvlist, ETH_VHOST_CLIENT_ARG) == 1) {
1165                 ret = rte_kvargs_process(kvlist, ETH_VHOST_CLIENT_ARG,
1166                                          &open_int, &client_mode);
1167                 if (ret < 0)
1168                         goto out_free;
1169
1170                 if (client_mode)
1171                         flags |= RTE_VHOST_USER_CLIENT;
1172         }
1173
1174         if (rte_kvargs_count(kvlist, ETH_VHOST_DEQUEUE_ZERO_COPY) == 1) {
1175                 ret = rte_kvargs_process(kvlist, ETH_VHOST_DEQUEUE_ZERO_COPY,
1176                                          &open_int, &dequeue_zero_copy);
1177                 if (ret < 0)
1178                         goto out_free;
1179
1180                 if (dequeue_zero_copy)
1181                         flags |= RTE_VHOST_USER_DEQUEUE_ZERO_COPY;
1182         }
1183
1184         eth_dev_vhost_create(name, iface_name, queues, rte_socket_id(), flags);
1185
1186 out_free:
1187         rte_kvargs_free(kvlist);
1188         return ret;
1189 }
1190
1191 static int
1192 rte_pmd_vhost_remove(const char *name)
1193 {
1194         struct rte_eth_dev *eth_dev = NULL;
1195         struct pmd_internal *internal;
1196         struct internal_list *list;
1197         unsigned int i;
1198
1199         RTE_LOG(INFO, PMD, "Un-Initializing pmd_vhost for %s\n", name);
1200
1201         /* find an ethdev entry */
1202         eth_dev = rte_eth_dev_allocated(name);
1203         if (eth_dev == NULL)
1204                 return -ENODEV;
1205
1206         internal = eth_dev->data->dev_private;
1207         if (internal == NULL)
1208                 return -ENODEV;
1209
1210         list = find_internal_resource(internal->iface_name);
1211         if (list == NULL)
1212                 return -ENODEV;
1213
1214         pthread_mutex_lock(&internal_list_lock);
1215         TAILQ_REMOVE(&internal_list, list, next);
1216         pthread_mutex_unlock(&internal_list_lock);
1217         rte_free(list);
1218
1219         eth_dev_stop(eth_dev);
1220
1221         rte_free(vring_states[eth_dev->data->port_id]);
1222         vring_states[eth_dev->data->port_id] = NULL;
1223
1224         free(internal->dev_name);
1225         free(internal->iface_name);
1226
1227         for (i = 0; i < eth_dev->data->nb_rx_queues; i++)
1228                 rte_free(eth_dev->data->rx_queues[i]);
1229         for (i = 0; i < eth_dev->data->nb_tx_queues; i++)
1230                 rte_free(eth_dev->data->tx_queues[i]);
1231
1232         rte_free(eth_dev->data->mac_addrs);
1233         rte_free(eth_dev->data);
1234         rte_free(internal);
1235
1236         rte_eth_dev_release_port(eth_dev);
1237
1238         return 0;
1239 }
1240
1241 static struct rte_vdev_driver pmd_vhost_drv = {
1242         .probe = rte_pmd_vhost_probe,
1243         .remove = rte_pmd_vhost_remove,
1244 };
1245
1246 DRIVER_REGISTER_VDEV(net_vhost, pmd_vhost_drv);
1247 DRIVER_REGISTER_PARAM_STRING(net_vhost,
1248         "iface=<ifc> "
1249         "queues=<int>");