4 * Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
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
17 * * Neither the name of Intel Corporation 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.
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.
38 #include <sys/types.h>
42 #include <linux/binfmts.h>
43 #include <xen/xen-compat.h>
44 #if __XEN_LATEST_INTERFACE_VERSION__ < 0x00040200
49 #include <linux/virtio_ring.h>
52 #include <rte_ethdev.h>
53 #include <rte_malloc.h>
54 #include <rte_memcpy.h>
55 #include <rte_string_fns.h>
57 #include <cmdline_parse.h>
58 #include <cmdline_parse_etheraddr.h>
60 #include "rte_xen_lib.h"
61 #include "virtqueue.h"
62 #include "rte_eth_xenvirt.h"
64 #define VQ_DESC_NUM 256
65 #define VIRTIO_MBUF_BURST_SZ 64
67 /* virtio_idx is increased after new device is created.*/
68 static int virtio_idx = 0;
70 static const char *drivername = "xen dummy virtio PMD";
72 static struct rte_eth_link pmd_link = {
74 .link_duplex = ETH_LINK_FULL_DUPLEX,
78 static inline struct rte_mbuf *
79 rte_rxmbuf_alloc(struct rte_mempool *mp)
83 m = __rte_mbuf_raw_alloc(mp);
84 __rte_mbuf_sanity_check_raw(m, 0);
91 eth_xenvirt_rx(void *q, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
93 struct virtqueue *rxvq = q;
94 struct rte_mbuf *rxm, *new_mbuf;
95 uint16_t nb_used, num;
96 uint32_t len[VIRTIO_MBUF_BURST_SZ];
98 struct pmd_internals *pi = rxvq->internals;
100 nb_used = VIRTQUEUE_NUSED(rxvq);
102 rte_compiler_barrier(); /* rmb */
103 num = (uint16_t)(likely(nb_used <= nb_pkts) ? nb_used : nb_pkts);
104 num = (uint16_t)(likely(num <= VIRTIO_MBUF_BURST_SZ) ? num : VIRTIO_MBUF_BURST_SZ);
105 if (unlikely(num == 0)) return 0;
107 num = virtqueue_dequeue_burst(rxvq, rx_pkts, len, num);
108 PMD_RX_LOG(DEBUG, "used:%d dequeue:%d\n", nb_used, num);
109 for (i = 0; i < num ; i ++) {
111 PMD_RX_LOG(DEBUG, "packet len:%d\n", len[i]);
112 rxm->pkt.next = NULL;
113 rxm->pkt.data = (char *)rxm->buf_addr + RTE_PKTMBUF_HEADROOM;
114 rxm->pkt.data_len = (uint16_t)(len[i] - sizeof(struct virtio_net_hdr));
115 rxm->pkt.nb_segs = 1;
116 rxm->pkt.in_port = pi->port_id;
117 rxm->pkt.pkt_len = (uint32_t)(len[i] - sizeof(struct virtio_net_hdr));
119 /* allocate new mbuf for the used descriptor */
120 while (likely(!virtqueue_full(rxvq))) {
121 new_mbuf = rte_rxmbuf_alloc(rxvq->mpool);
122 if (unlikely(new_mbuf == NULL)) {
125 if (unlikely(virtqueue_enqueue_recv_refill(rxvq, new_mbuf))) {
126 rte_pktmbuf_free_seg(new_mbuf);
130 pi->eth_stats.ipackets += num;
135 eth_xenvirt_tx(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
137 struct virtqueue *txvq = tx_queue;
138 struct rte_mbuf *txm;
139 uint16_t nb_used, nb_tx, num, i;
141 uint32_t len[VIRTIO_MBUF_BURST_SZ];
142 struct rte_mbuf *snd_pkts[VIRTIO_MBUF_BURST_SZ];
143 struct pmd_internals *pi = txvq->internals;
147 if (unlikely(nb_pkts == 0))
150 PMD_TX_LOG(DEBUG, "%d packets to xmit", nb_pkts);
151 nb_used = VIRTQUEUE_NUSED(txvq);
153 rte_compiler_barrier(); /* rmb */
155 num = (uint16_t)(likely(nb_used <= VIRTIO_MBUF_BURST_SZ) ? nb_used : VIRTIO_MBUF_BURST_SZ);
156 num = virtqueue_dequeue_burst(txvq, snd_pkts, len, num);
158 for (i = 0; i < num ; i ++) {
159 /* mergable not supported, one segment only */
160 rte_pktmbuf_free_seg(snd_pkts[i]);
163 while (nb_tx < nb_pkts) {
164 if (likely(!virtqueue_full(txvq))) {
165 /* TODO drop tx_pkts if it contains multiple segments */
166 txm = tx_pkts[nb_tx];
167 error = virtqueue_enqueue_xmit(txvq, txm);
168 if (unlikely(error)) {
170 PMD_TX_LOG(ERR, "virtqueue_enqueue Free count = 0\n");
171 else if (error == EMSGSIZE)
172 PMD_TX_LOG(ERR, "virtqueue_enqueue Free count < 1\n");
174 PMD_TX_LOG(ERR, "virtqueue_enqueue error: %d\n", error);
179 PMD_TX_LOG(ERR, "No free tx descriptors to transmit\n");
180 /* virtqueue_notify not needed in our para-virt solution */
184 pi->eth_stats.opackets += nb_tx;
189 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
191 RTE_LOG(ERR, PMD, "%s\n", __func__);
196 * Create a shared page between guest and host.
197 * Host monitors this page if it is cleared on unmap, and then
198 * do necessary clean up.
201 gntalloc_vring_flag(int vtidx)
203 char key_str[PATH_MAX];
204 char val_str[PATH_MAX];
208 if (grefwatch_from_alloc(&gref_tmp, &ptr)) {
209 RTE_LOG(ERR, PMD, "grefwatch_from_alloc error\n");
213 *(uint8_t *)ptr = MAP_FLAG;
214 snprintf(val_str, sizeof(val_str), "%u", gref_tmp);
215 snprintf(key_str, sizeof(key_str),
216 DPDK_XENSTORE_PATH"%d"VRING_FLAG_STR, vtidx);
217 xenstore_write(key_str, val_str);
221 * Notify host this virtio device is started.
222 * Host could start polling this device.
225 dev_start_notify(int vtidx)
227 char key_str[PATH_MAX];
228 char val_str[PATH_MAX];
230 RTE_LOG(INFO, PMD, "%s: virtio %d is started\n", __func__, vtidx);
231 gntalloc_vring_flag(vtidx);
233 snprintf(key_str, sizeof(key_str), "%s%s%d",
234 DPDK_XENSTORE_PATH, EVENT_TYPE_START_STR,
236 snprintf(val_str, sizeof(val_str), "1");
237 xenstore_write(key_str, val_str);
241 * Notify host this virtio device is stopped.
242 * Host could stop polling this device.
245 dev_stop_notify(int vtidx)
252 update_mac_address(struct ether_addr *mac_addrs, int vtidx)
254 char key_str[PATH_MAX];
255 char val_str[PATH_MAX];
258 if (mac_addrs == NULL) {
259 RTE_LOG(ERR, PMD, "%s: NULL pointer mac specified\n", __func__);
262 rv = snprintf(key_str, sizeof(key_str),
263 DPDK_XENSTORE_PATH"%d_ether_addr", vtidx);
266 rv = snprintf(val_str, sizeof(val_str), "%02x:%02x:%02x:%02x:%02x:%02x",
267 mac_addrs->addr_bytes[0],
268 mac_addrs->addr_bytes[1],
269 mac_addrs->addr_bytes[2],
270 mac_addrs->addr_bytes[3],
271 mac_addrs->addr_bytes[4],
272 mac_addrs->addr_bytes[5]);
275 if (xenstore_write(key_str, val_str))
282 eth_dev_start(struct rte_eth_dev *dev)
284 struct virtqueue *rxvq = dev->data->rx_queues[0];
285 struct virtqueue *txvq = dev->data->tx_queues[0];
287 struct pmd_internals *pi = (struct pmd_internals *)dev->data->dev_private;
290 dev->data->dev_link.link_status = 1;
291 while (!virtqueue_full(rxvq)) {
292 m = rte_rxmbuf_alloc(rxvq->mpool);
295 /* Enqueue allocated buffers. */
296 if (virtqueue_enqueue_recv_refill(rxvq, m)) {
297 rte_pktmbuf_free_seg(m);
302 rxvq->internals = pi;
303 txvq->internals = pi;
305 rv = update_mac_address(dev->data->mac_addrs, pi->virtio_idx);
308 dev_start_notify(pi->virtio_idx);
314 eth_dev_stop(struct rte_eth_dev *dev)
316 struct pmd_internals *pi = (struct pmd_internals *)dev->data->dev_private;
318 dev->data->dev_link.link_status = 0;
319 dev_stop_notify(pi->virtio_idx);
323 * Notify host this virtio device is closed.
324 * Host could do necessary clean up to this device.
327 eth_dev_close(struct rte_eth_dev *dev)
333 eth_dev_info(struct rte_eth_dev *dev,
334 struct rte_eth_dev_info *dev_info)
336 struct pmd_internals *internals = dev->data->dev_private;
338 RTE_SET_USED(internals);
339 dev_info->driver_name = drivername;
340 dev_info->max_mac_addrs = 1;
341 dev_info->max_rx_pktlen = (uint32_t)2048;
342 dev_info->max_rx_queues = (uint16_t)1;
343 dev_info->max_tx_queues = (uint16_t)1;
344 dev_info->min_rx_bufsize = 0;
345 dev_info->pci_dev = NULL;
349 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
351 struct pmd_internals *internals = dev->data->dev_private;
353 rte_memcpy(stats, &internals->eth_stats, sizeof(*stats));
357 eth_stats_reset(struct rte_eth_dev *dev)
359 struct pmd_internals *internals = dev->data->dev_private;
360 /* Reset software totals */
361 memset(&internals->eth_stats, 0, sizeof(internals->eth_stats));
365 eth_queue_release(void *q __rte_unused)
370 eth_link_update(struct rte_eth_dev *dev __rte_unused,
371 int wait_to_complete __rte_unused)
377 * Create shared vring between guest and host.
378 * Memory is allocated through grant alloc driver, so it is not physical continuous.
381 gntalloc_vring_create(int queue_type, uint32_t size, int vtidx)
383 char key_str[PATH_MAX] = {0};
384 char val_str[PATH_MAX] = {0};
388 uint32_t *gref_arr = NULL;
389 phys_addr_t *pa_arr = NULL;
390 uint64_t start_index;
393 pg_size = getpagesize();
394 size = RTE_ALIGN_CEIL(size, pg_size);
395 pg_num = size / pg_size;
397 gref_arr = calloc(pg_num, sizeof(gref_arr[0]));
398 pa_arr = calloc(pg_num, sizeof(pa_arr[0]));
400 if (gref_arr == NULL || pa_arr == NULL) {
401 RTE_LOG(ERR, PMD, "%s: calloc failed\n", __func__);
405 va = gntalloc(size, gref_arr, &start_index);
407 RTE_LOG(ERR, PMD, "%s: gntalloc failed\n", __func__);
411 if (get_phys_map(va, pa_arr, pg_num, pg_size))
414 /* write in xenstore gref and pfn for each page of vring */
415 if (grant_node_create(pg_num, gref_arr, pa_arr, val_str, sizeof(val_str))) {
416 gntfree(va, size, start_index);
421 if (queue_type == VTNET_RQ)
422 rv = snprintf(key_str, sizeof(key_str), DPDK_XENSTORE_PATH"%d"RXVRING_XENSTORE_STR, vtidx);
424 rv = snprintf(key_str, sizeof(key_str), DPDK_XENSTORE_PATH"%d"TXVRING_XENSTORE_STR, vtidx);
425 if (rv == -1 || xenstore_write(key_str, val_str) == -1) {
426 gntfree(va, size, start_index);
440 static struct virtqueue *
441 virtio_queue_setup(struct rte_eth_dev *dev, int queue_type)
443 struct virtqueue *vq = NULL;
444 uint16_t vq_size = VQ_DESC_NUM;
446 char vq_name[VIRTQUEUE_MAX_NAME_SZ];
450 /* Allocate memory for virtqueue. */
451 if (queue_type == VTNET_RQ) {
452 snprintf(vq_name, sizeof(vq_name), "port%d_rvq",
454 vq = rte_zmalloc(vq_name, sizeof(struct virtqueue) +
455 vq_size * sizeof(struct vq_desc_extra), CACHE_LINE_SIZE);
457 RTE_LOG(ERR, PMD, "%s: unabled to allocate virtqueue\n", __func__);
460 memcpy(vq->vq_name, vq_name, sizeof(vq->vq_name));
461 } else if(queue_type == VTNET_TQ) {
462 snprintf(vq_name, sizeof(vq_name), "port%d_tvq",
464 vq = rte_zmalloc(vq_name, sizeof(struct virtqueue) +
465 vq_size * sizeof(struct vq_desc_extra), CACHE_LINE_SIZE);
467 RTE_LOG(ERR, PMD, "%s: unabled to allocate virtqueue\n", __func__);
470 memcpy(vq->vq_name, vq_name, sizeof(vq->vq_name));
473 memcpy(vq->vq_name, vq_name, sizeof(vq->vq_name));
475 vq->vq_alignment = VIRTIO_PCI_VRING_ALIGN;
476 vq->vq_nentries = vq_size;
477 vq->vq_free_cnt = vq_size;
478 /* Calcuate vring size according to virtio spec */
479 size = vring_size(vq_size, VIRTIO_PCI_VRING_ALIGN);
480 vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
481 /* Allocate memory for virtio vring through gntalloc driver*/
482 vq->vq_ring_virt_mem = gntalloc_vring_create(queue_type, vq->vq_ring_size,
483 ((struct pmd_internals *)dev->data->dev_private)->virtio_idx);
484 memset(vq->vq_ring_virt_mem, 0, vq->vq_ring_size);
486 vring_init(vr, vq_size, vq->vq_ring_virt_mem, vq->vq_alignment);
488 * Locally maintained last consumed index, this idex trails
491 vq->vq_used_cons_idx = 0;
492 vq->vq_desc_head_idx = 0;
493 vq->vq_free_cnt = vq->vq_nentries;
494 memset(vq->vq_descx, 0, sizeof(struct vq_desc_extra) * vq->vq_nentries);
496 /* Chain all the descriptors in the ring with an END */
497 for (i = 0; i < vq_size - 1; i++)
498 vr->desc[i].next = (uint16_t)(i + 1);
499 vr->desc[i].next = VQ_RING_DESC_CHAIN_END;
505 eth_rx_queue_setup(struct rte_eth_dev *dev,uint16_t rx_queue_id,
506 uint16_t nb_rx_desc __rte_unused,
507 unsigned int socket_id __rte_unused,
508 const struct rte_eth_rxconf *rx_conf __rte_unused,
509 struct rte_mempool *mb_pool)
511 struct virtqueue *vq;
512 vq = dev->data->rx_queues[rx_queue_id] = virtio_queue_setup(dev, VTNET_RQ);
518 eth_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
519 uint16_t nb_tx_desc __rte_unused,
520 unsigned int socket_id __rte_unused,
521 const struct rte_eth_txconf *tx_conf __rte_unused)
523 dev->data->tx_queues[tx_queue_id] = virtio_queue_setup(dev, VTNET_TQ);
529 static struct eth_dev_ops ops = {
530 .dev_start = eth_dev_start,
531 .dev_stop = eth_dev_stop,
532 .dev_close = eth_dev_close,
533 .dev_configure = eth_dev_configure,
534 .dev_infos_get = eth_dev_info,
535 .rx_queue_setup = eth_rx_queue_setup,
536 .tx_queue_setup = eth_tx_queue_setup,
537 .rx_queue_release = eth_queue_release,
538 .tx_queue_release = eth_queue_release,
539 .link_update = eth_link_update,
540 .stats_get = eth_stats_get,
541 .stats_reset = eth_stats_reset,
546 rte_eth_xenvirt_parse_args(struct xenvirt_dict *dict,
547 const char *name, const char *params)
550 char *pairs[RTE_ETH_XENVIRT_MAX_ARGS];
559 args = rte_zmalloc(NULL, strlen(params) + 1, CACHE_LINE_SIZE);
561 RTE_LOG(ERR, PMD, "Couldn't parse %s device \n", name);
564 rte_memcpy(args, params, strlen(params));
566 num_of_pairs = rte_strsplit(args, strnlen(args, MAX_ARG_STRLEN),
568 RTE_ETH_XENVIRT_MAX_ARGS ,
569 RTE_ETH_XENVIRT_PAIRS_DELIM);
571 for (i = 0; i < num_of_pairs; i++) {
574 rte_strsplit(pairs[i], strnlen(pairs[i], MAX_ARG_STRLEN),
576 RTE_ETH_XENVIRT_KEY_VALUE_DELIM);
578 if (pair[0] == NULL || pair[1] == NULL || pair[0][0] == 0
579 || pair[1][0] == 0) {
581 "Couldn't parse %s device,"
582 "wrong key or value \n", name);
586 if (!strncmp(pair[0], RTE_ETH_XENVIRT_MAC_PARAM,
587 sizeof(RTE_ETH_XENVIRT_MAC_PARAM))) {
588 if (cmdline_parse_etheraddr(NULL,
592 "Invalid %s device ether address\n",
597 dict->addr_valid = 1;
614 eth_dev_xenvirt_create(const char *name, const char *params,
615 const unsigned numa_node,
616 enum dev_action action)
618 struct rte_eth_dev_data *data = NULL;
619 struct rte_pci_device *pci_dev = NULL;
620 struct pmd_internals *internals = NULL;
621 struct rte_eth_dev *eth_dev = NULL;
622 struct xenvirt_dict dict;
623 bzero(&dict, sizeof(struct xenvirt_dict));
625 RTE_LOG(INFO, PMD, "Creating virtio rings backed ethdev on numa socket %u\n",
627 RTE_SET_USED(action);
629 if (rte_eth_xenvirt_parse_args(&dict, name, params) < 0) {
630 RTE_LOG(ERR, PMD, "%s: Failed to parse ethdev parameters\n", __func__);
634 /* now do all data allocation - for eth_dev structure, dummy pci driver
635 * and internal (private) data
637 data = rte_zmalloc_socket(name, sizeof(*data), 0, numa_node);
641 pci_dev = rte_zmalloc_socket(name, sizeof(*pci_dev), 0, numa_node);
645 internals = rte_zmalloc_socket(name, sizeof(*internals), 0, numa_node);
646 if (internals == NULL)
649 /* reserve an ethdev entry */
650 eth_dev = rte_eth_dev_allocate(name);
654 pci_dev->numa_node = numa_node;
656 data->dev_private = internals;
657 data->port_id = eth_dev->data->port_id;
658 data->nb_rx_queues = (uint16_t)1;
659 data->nb_tx_queues = (uint16_t)1;
660 data->dev_link = pmd_link;
661 data->mac_addrs = rte_zmalloc("xen_virtio", ETHER_ADDR_LEN, 0);
664 memcpy(&data->mac_addrs->addr_bytes, &dict.addr, sizeof(struct ether_addr));
666 eth_random_addr(&data->mac_addrs->addr_bytes[0]);
668 eth_dev->data = data;
669 eth_dev->dev_ops = &ops;
670 eth_dev->pci_dev = pci_dev;
672 eth_dev->rx_pkt_burst = eth_xenvirt_rx;
673 eth_dev->tx_pkt_burst = eth_xenvirt_tx;
675 internals->virtio_idx = virtio_idx++;
676 internals->port_id = eth_dev->data->port_id;
691 /*TODO: Support multiple process model */
693 rte_pmd_xenvirt_devinit(const char *name, const char *params)
695 if (virtio_idx == 0) {
696 if (xenstore_init() != 0) {
697 RTE_LOG(ERR, PMD, "%s: xenstore init failed\n", __func__);
700 if (gntalloc_open() != 0) {
701 RTE_LOG(ERR, PMD, "%s: grant init failed\n", __func__);
705 eth_dev_xenvirt_create(name, params, rte_socket_id(), DEV_CREATE);
709 static struct rte_driver pmd_xenvirt_drv = {
710 .name = "eth_xenvirt",
712 .init = rte_pmd_xenvirt_devinit,
715 PMD_REGISTER_DRIVER(pmd_xenvirt_drv);