4 * Copyright(c) 2010-2015 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>
43 #define PAGE_SIZE sysconf(_SC_PAGE_SIZE)
45 #include <linux/binfmts.h>
46 #include <xen/xen-compat.h>
47 #if __XEN_LATEST_INTERFACE_VERSION__ < 0x00040200
52 #include <linux/virtio_ring.h>
55 #include <rte_ethdev.h>
56 #include <rte_malloc.h>
57 #include <rte_memcpy.h>
58 #include <rte_string_fns.h>
60 #include <cmdline_parse.h>
61 #include <cmdline_parse_etheraddr.h>
63 #include "rte_xen_lib.h"
64 #include "virtqueue.h"
65 #include "rte_eth_xenvirt.h"
67 #define VQ_DESC_NUM 256
68 #define VIRTIO_MBUF_BURST_SZ 64
70 /* virtio_idx is increased after new device is created.*/
71 static int virtio_idx = 0;
73 static const char *drivername = "xen virtio PMD";
75 static struct rte_eth_link pmd_link = {
76 .link_speed = ETH_SPEED_NUM_10G,
77 .link_duplex = ETH_LINK_FULL_DUPLEX,
78 .link_status = ETH_LINK_DOWN,
79 .link_autoneg = ETH_LINK_SPEED_FIXED
83 eth_xenvirt_free_queues(struct rte_eth_dev *dev);
86 eth_xenvirt_rx(void *q, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
88 struct virtqueue *rxvq = q;
89 struct rte_mbuf *rxm, *new_mbuf;
90 uint16_t nb_used, num;
91 uint32_t len[VIRTIO_MBUF_BURST_SZ];
93 struct pmd_internals *pi = rxvq->internals;
95 nb_used = VIRTQUEUE_NUSED(rxvq);
98 num = (uint16_t)(likely(nb_used <= nb_pkts) ? nb_used : nb_pkts);
99 num = (uint16_t)(likely(num <= VIRTIO_MBUF_BURST_SZ) ? num : VIRTIO_MBUF_BURST_SZ);
100 if (unlikely(num == 0)) return 0;
102 num = virtqueue_dequeue_burst(rxvq, rx_pkts, len, num);
103 PMD_RX_LOG(DEBUG, "used:%d dequeue:%d\n", nb_used, num);
104 for (i = 0; i < num ; i ++) {
106 PMD_RX_LOG(DEBUG, "packet len:%d\n", len[i]);
108 rxm->data_off = RTE_PKTMBUF_HEADROOM;
109 rxm->data_len = (uint16_t)(len[i] - sizeof(struct virtio_net_hdr));
111 rxm->port = pi->port_id;
112 rxm->pkt_len = (uint32_t)(len[i] - sizeof(struct virtio_net_hdr));
114 /* allocate new mbuf for the used descriptor */
115 while (likely(!virtqueue_full(rxvq))) {
116 new_mbuf = rte_mbuf_raw_alloc(rxvq->mpool);
117 if (unlikely(new_mbuf == NULL)) {
120 if (unlikely(virtqueue_enqueue_recv_refill(rxvq, new_mbuf))) {
121 rte_pktmbuf_free_seg(new_mbuf);
125 pi->eth_stats.ipackets += num;
130 eth_xenvirt_tx(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
132 struct virtqueue *txvq = tx_queue;
133 struct rte_mbuf *txm;
134 uint16_t nb_used, nb_tx, num, i;
136 uint32_t len[VIRTIO_MBUF_BURST_SZ];
137 struct rte_mbuf *snd_pkts[VIRTIO_MBUF_BURST_SZ];
138 struct pmd_internals *pi = txvq->internals;
142 if (unlikely(nb_pkts == 0))
145 PMD_TX_LOG(DEBUG, "%d packets to xmit", nb_pkts);
146 nb_used = VIRTQUEUE_NUSED(txvq);
150 num = (uint16_t)(likely(nb_used <= VIRTIO_MBUF_BURST_SZ) ? nb_used : VIRTIO_MBUF_BURST_SZ);
151 num = virtqueue_dequeue_burst(txvq, snd_pkts, len, num);
153 for (i = 0; i < num ; i ++) {
154 /* mergable not supported, one segment only */
155 rte_pktmbuf_free_seg(snd_pkts[i]);
158 while (nb_tx < nb_pkts) {
159 if (likely(!virtqueue_full(txvq))) {
160 /* TODO drop tx_pkts if it contains multiple segments */
161 txm = tx_pkts[nb_tx];
162 error = virtqueue_enqueue_xmit(txvq, txm);
163 if (unlikely(error)) {
165 PMD_TX_LOG(ERR, "virtqueue_enqueue Free count = 0\n");
166 else if (error == EMSGSIZE)
167 PMD_TX_LOG(ERR, "virtqueue_enqueue Free count < 1\n");
169 PMD_TX_LOG(ERR, "virtqueue_enqueue error: %d\n", error);
174 PMD_TX_LOG(ERR, "No free tx descriptors to transmit\n");
175 /* virtqueue_notify not needed in our para-virt solution */
179 pi->eth_stats.opackets += nb_tx;
184 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
186 RTE_LOG(ERR, PMD, "%s\n", __func__);
191 * Create a shared page between guest and host.
192 * Host monitors this page if it is cleared on unmap, and then
193 * do necessary clean up.
196 gntalloc_vring_flag(int vtidx)
198 char key_str[PATH_MAX];
199 char val_str[PATH_MAX];
203 if (grefwatch_from_alloc(&gref_tmp, &ptr)) {
204 RTE_LOG(ERR, PMD, "grefwatch_from_alloc error\n");
208 *(uint8_t *)ptr = MAP_FLAG;
209 snprintf(val_str, sizeof(val_str), "%u", gref_tmp);
210 snprintf(key_str, sizeof(key_str),
211 DPDK_XENSTORE_PATH"%d"VRING_FLAG_STR, vtidx);
212 xenstore_write(key_str, val_str);
216 * Notify host this virtio device is started.
217 * Host could start polling this device.
220 dev_start_notify(int vtidx)
222 char key_str[PATH_MAX];
223 char val_str[PATH_MAX];
225 RTE_LOG(INFO, PMD, "%s: virtio %d is started\n", __func__, vtidx);
226 gntalloc_vring_flag(vtidx);
228 snprintf(key_str, sizeof(key_str), "%s%s%d",
229 DPDK_XENSTORE_PATH, EVENT_TYPE_START_STR,
231 snprintf(val_str, sizeof(val_str), "1");
232 xenstore_write(key_str, val_str);
236 * Notify host this virtio device is stopped.
237 * Host could stop polling this device.
240 dev_stop_notify(int vtidx)
247 update_mac_address(struct ether_addr *mac_addrs, int vtidx)
249 char key_str[PATH_MAX];
250 char val_str[PATH_MAX];
253 if (mac_addrs == NULL) {
254 RTE_LOG(ERR, PMD, "%s: NULL pointer mac specified\n", __func__);
257 rv = snprintf(key_str, sizeof(key_str),
258 DPDK_XENSTORE_PATH"%d_ether_addr", vtidx);
261 rv = snprintf(val_str, sizeof(val_str), "%02x:%02x:%02x:%02x:%02x:%02x",
262 mac_addrs->addr_bytes[0],
263 mac_addrs->addr_bytes[1],
264 mac_addrs->addr_bytes[2],
265 mac_addrs->addr_bytes[3],
266 mac_addrs->addr_bytes[4],
267 mac_addrs->addr_bytes[5]);
270 if (xenstore_write(key_str, val_str))
277 eth_dev_start(struct rte_eth_dev *dev)
279 struct virtqueue *rxvq = dev->data->rx_queues[0];
280 struct virtqueue *txvq = dev->data->tx_queues[0];
282 struct pmd_internals *pi = (struct pmd_internals *)dev->data->dev_private;
285 dev->data->dev_link.link_status = ETH_LINK_UP;
286 while (!virtqueue_full(rxvq)) {
287 m = rte_mbuf_raw_alloc(rxvq->mpool);
290 /* Enqueue allocated buffers. */
291 if (virtqueue_enqueue_recv_refill(rxvq, m)) {
292 rte_pktmbuf_free_seg(m);
297 rxvq->internals = pi;
298 txvq->internals = pi;
300 rv = update_mac_address(dev->data->mac_addrs, pi->virtio_idx);
303 dev_start_notify(pi->virtio_idx);
309 eth_dev_stop(struct rte_eth_dev *dev)
311 struct pmd_internals *pi = (struct pmd_internals *)dev->data->dev_private;
313 dev->data->dev_link.link_status = ETH_LINK_DOWN;
314 dev_stop_notify(pi->virtio_idx);
318 * Notify host this virtio device is closed.
319 * Host could do necessary clean up to this device.
322 eth_dev_close(struct rte_eth_dev *dev)
324 eth_xenvirt_free_queues(dev);
328 eth_dev_info(struct rte_eth_dev *dev,
329 struct rte_eth_dev_info *dev_info)
331 struct pmd_internals *internals = dev->data->dev_private;
333 RTE_SET_USED(internals);
334 dev_info->driver_name = drivername;
335 dev_info->max_mac_addrs = 1;
336 dev_info->max_rx_pktlen = (uint32_t)2048;
337 dev_info->max_rx_queues = (uint16_t)1;
338 dev_info->max_tx_queues = (uint16_t)1;
339 dev_info->min_rx_bufsize = 0;
343 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
345 struct pmd_internals *internals = dev->data->dev_private;
347 rte_memcpy(stats, &internals->eth_stats, sizeof(*stats));
351 eth_stats_reset(struct rte_eth_dev *dev)
353 struct pmd_internals *internals = dev->data->dev_private;
354 /* Reset software totals */
355 memset(&internals->eth_stats, 0, sizeof(internals->eth_stats));
359 eth_queue_release(void *q)
365 eth_link_update(struct rte_eth_dev *dev __rte_unused,
366 int wait_to_complete __rte_unused)
372 * Create shared vring between guest and host.
373 * Memory is allocated through grant alloc driver, so it is not physical continuous.
376 gntalloc_vring_create(int queue_type, uint32_t size, int vtidx)
378 char key_str[PATH_MAX] = {0};
379 char val_str[PATH_MAX] = {0};
383 uint32_t *gref_arr = NULL;
384 phys_addr_t *pa_arr = NULL;
385 uint64_t start_index;
388 pg_size = getpagesize();
389 size = RTE_ALIGN_CEIL(size, pg_size);
390 pg_num = size / pg_size;
392 gref_arr = calloc(pg_num, sizeof(gref_arr[0]));
393 pa_arr = calloc(pg_num, sizeof(pa_arr[0]));
395 if (gref_arr == NULL || pa_arr == NULL) {
396 RTE_LOG(ERR, PMD, "%s: calloc failed\n", __func__);
400 va = gntalloc(size, gref_arr, &start_index);
402 RTE_LOG(ERR, PMD, "%s: gntalloc failed\n", __func__);
406 if (get_phys_map(va, pa_arr, pg_num, pg_size))
409 /* write in xenstore gref and pfn for each page of vring */
410 if (grant_node_create(pg_num, gref_arr, pa_arr, val_str, sizeof(val_str))) {
411 gntfree(va, size, start_index);
416 if (queue_type == VTNET_RQ)
417 rv = snprintf(key_str, sizeof(key_str), DPDK_XENSTORE_PATH"%d"RXVRING_XENSTORE_STR, vtidx);
419 rv = snprintf(key_str, sizeof(key_str), DPDK_XENSTORE_PATH"%d"TXVRING_XENSTORE_STR, vtidx);
420 if (rv == -1 || xenstore_write(key_str, val_str) == -1) {
421 gntfree(va, size, start_index);
433 static struct virtqueue *
434 virtio_queue_setup(struct rte_eth_dev *dev, int queue_type)
436 struct virtqueue *vq = NULL;
437 uint16_t vq_size = VQ_DESC_NUM;
439 char vq_name[VIRTQUEUE_MAX_NAME_SZ];
443 /* Allocate memory for virtqueue. */
444 if (queue_type == VTNET_RQ) {
445 snprintf(vq_name, sizeof(vq_name), "port%d_rvq",
447 vq = rte_zmalloc(vq_name, sizeof(struct virtqueue) +
448 vq_size * sizeof(struct vq_desc_extra), RTE_CACHE_LINE_SIZE);
450 RTE_LOG(ERR, PMD, "%s: unabled to allocate virtqueue\n", __func__);
453 memcpy(vq->vq_name, vq_name, sizeof(vq->vq_name));
454 } else if(queue_type == VTNET_TQ) {
455 snprintf(vq_name, sizeof(vq_name), "port%d_tvq",
457 vq = rte_zmalloc(vq_name, sizeof(struct virtqueue) +
458 vq_size * sizeof(struct vq_desc_extra), RTE_CACHE_LINE_SIZE);
460 RTE_LOG(ERR, PMD, "%s: unabled to allocate virtqueue\n", __func__);
463 memcpy(vq->vq_name, vq_name, sizeof(vq->vq_name));
466 memcpy(vq->vq_name, vq_name, sizeof(vq->vq_name));
468 vq->vq_alignment = VIRTIO_PCI_VRING_ALIGN;
469 vq->vq_nentries = vq_size;
470 vq->vq_free_cnt = vq_size;
471 /* Calcuate vring size according to virtio spec */
472 size = vring_size(vq_size, VIRTIO_PCI_VRING_ALIGN);
473 vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
474 /* Allocate memory for virtio vring through gntalloc driver*/
475 vq->vq_ring_virt_mem = gntalloc_vring_create(queue_type, vq->vq_ring_size,
476 ((struct pmd_internals *)dev->data->dev_private)->virtio_idx);
477 memset(vq->vq_ring_virt_mem, 0, vq->vq_ring_size);
479 vring_init(vr, vq_size, vq->vq_ring_virt_mem, vq->vq_alignment);
481 * Locally maintained last consumed index, this idex trails
484 vq->vq_used_cons_idx = 0;
485 vq->vq_desc_head_idx = 0;
486 vq->vq_free_cnt = vq->vq_nentries;
487 memset(vq->vq_descx, 0, sizeof(struct vq_desc_extra) * vq->vq_nentries);
489 /* Chain all the descriptors in the ring with an END */
490 for (i = 0; i < vq_size - 1; i++)
491 vr->desc[i].next = (uint16_t)(i + 1);
492 vr->desc[i].next = VQ_RING_DESC_CHAIN_END;
498 eth_rx_queue_setup(struct rte_eth_dev *dev,uint16_t rx_queue_id,
499 uint16_t nb_rx_desc __rte_unused,
500 unsigned int socket_id __rte_unused,
501 const struct rte_eth_rxconf *rx_conf __rte_unused,
502 struct rte_mempool *mb_pool)
504 struct virtqueue *vq;
505 vq = dev->data->rx_queues[rx_queue_id] = virtio_queue_setup(dev, VTNET_RQ);
511 eth_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
512 uint16_t nb_tx_desc __rte_unused,
513 unsigned int socket_id __rte_unused,
514 const struct rte_eth_txconf *tx_conf __rte_unused)
516 dev->data->tx_queues[tx_queue_id] = virtio_queue_setup(dev, VTNET_TQ);
521 eth_xenvirt_free_queues(struct rte_eth_dev *dev)
525 for (i = 0; i < dev->data->nb_rx_queues; i++) {
526 eth_queue_release(dev->data->rx_queues[i]);
527 dev->data->rx_queues[i] = NULL;
529 dev->data->nb_rx_queues = 0;
531 for (i = 0; i < dev->data->nb_tx_queues; i++) {
532 eth_queue_release(dev->data->tx_queues[i]);
533 dev->data->tx_queues[i] = NULL;
535 dev->data->nb_tx_queues = 0;
538 static const struct eth_dev_ops ops = {
539 .dev_start = eth_dev_start,
540 .dev_stop = eth_dev_stop,
541 .dev_close = eth_dev_close,
542 .dev_configure = eth_dev_configure,
543 .dev_infos_get = eth_dev_info,
544 .rx_queue_setup = eth_rx_queue_setup,
545 .tx_queue_setup = eth_tx_queue_setup,
546 .rx_queue_release = eth_queue_release,
547 .tx_queue_release = eth_queue_release,
548 .link_update = eth_link_update,
549 .stats_get = eth_stats_get,
550 .stats_reset = eth_stats_reset,
555 rte_eth_xenvirt_parse_args(struct xenvirt_dict *dict,
556 const char *name, const char *params)
559 char *pairs[RTE_ETH_XENVIRT_MAX_ARGS];
568 args = rte_zmalloc(NULL, strlen(params) + 1, RTE_CACHE_LINE_SIZE);
570 RTE_LOG(ERR, PMD, "Couldn't parse %s device \n", name);
573 rte_memcpy(args, params, strlen(params));
575 num_of_pairs = rte_strsplit(args, strnlen(args, MAX_ARG_STRLEN),
577 RTE_ETH_XENVIRT_MAX_ARGS ,
578 RTE_ETH_XENVIRT_PAIRS_DELIM);
580 for (i = 0; i < num_of_pairs; i++) {
583 rte_strsplit(pairs[i], strnlen(pairs[i], MAX_ARG_STRLEN),
585 RTE_ETH_XENVIRT_KEY_VALUE_DELIM);
587 if (pair[0] == NULL || pair[1] == NULL || pair[0][0] == 0
588 || pair[1][0] == 0) {
590 "Couldn't parse %s device,"
591 "wrong key or value \n", name);
595 if (!strncmp(pair[0], RTE_ETH_XENVIRT_MAC_PARAM,
596 sizeof(RTE_ETH_XENVIRT_MAC_PARAM))) {
597 if (cmdline_parse_etheraddr(NULL,
600 sizeof(dict->addr)) < 0) {
602 "Invalid %s device ether address\n",
607 dict->addr_valid = 1;
624 eth_dev_xenvirt_create(const char *name, const char *params,
625 const unsigned numa_node,
626 enum dev_action action)
628 struct rte_eth_dev_data *data = NULL;
629 struct pmd_internals *internals = NULL;
630 struct rte_eth_dev *eth_dev = NULL;
631 struct xenvirt_dict dict;
633 memset(&dict, 0, sizeof(struct xenvirt_dict));
635 RTE_LOG(INFO, PMD, "Creating virtio rings backed ethdev on numa socket %u\n",
637 RTE_SET_USED(action);
639 if (rte_eth_xenvirt_parse_args(&dict, name, params) < 0) {
640 RTE_LOG(ERR, PMD, "%s: Failed to parse ethdev parameters\n", __func__);
644 /* now do all data allocation - for eth_dev structure, dummy pci driver
645 * and internal (private) data
647 data = rte_zmalloc_socket(name, sizeof(*data), 0, numa_node);
651 internals = rte_zmalloc_socket(name, sizeof(*internals), 0, numa_node);
652 if (internals == NULL)
655 /* reserve an ethdev entry */
656 eth_dev = rte_eth_dev_allocate(name);
660 data->dev_private = internals;
661 data->port_id = eth_dev->data->port_id;
662 data->nb_rx_queues = (uint16_t)1;
663 data->nb_tx_queues = (uint16_t)1;
664 data->dev_link = pmd_link;
665 data->mac_addrs = rte_zmalloc("xen_virtio", ETHER_ADDR_LEN, 0);
668 memcpy(&data->mac_addrs->addr_bytes, &dict.addr, sizeof(struct ether_addr));
670 eth_random_addr(&data->mac_addrs->addr_bytes[0]);
672 eth_dev->data = data;
673 eth_dev->dev_ops = &ops;
675 eth_dev->data->dev_flags = RTE_PCI_DRV_DETACHABLE;
676 eth_dev->data->kdrv = RTE_KDRV_NONE;
677 eth_dev->data->drv_name = drivername;
678 eth_dev->driver = NULL;
679 eth_dev->data->numa_node = numa_node;
681 eth_dev->rx_pkt_burst = eth_xenvirt_rx;
682 eth_dev->tx_pkt_burst = eth_xenvirt_tx;
684 internals->virtio_idx = virtio_idx++;
685 internals->port_id = eth_dev->data->port_id;
698 eth_dev_xenvirt_free(const char *name, const unsigned numa_node)
700 struct rte_eth_dev *eth_dev = NULL;
703 "Free virtio rings backed ethdev on numa socket %u\n",
706 /* find an ethdev entry */
707 eth_dev = rte_eth_dev_allocated(name);
711 if (eth_dev->data->dev_started == 1) {
712 eth_dev_stop(eth_dev);
713 eth_dev_close(eth_dev);
716 eth_dev->rx_pkt_burst = NULL;
717 eth_dev->tx_pkt_burst = NULL;
718 eth_dev->dev_ops = NULL;
720 rte_free(eth_dev->data);
721 rte_free(eth_dev->data->dev_private);
722 rte_free(eth_dev->data->mac_addrs);
729 /*TODO: Support multiple process model */
731 rte_pmd_xenvirt_probe(const char *name, const char *params)
733 if (virtio_idx == 0) {
734 if (xenstore_init() != 0) {
735 RTE_LOG(ERR, PMD, "%s: xenstore init failed\n", __func__);
738 if (gntalloc_open() != 0) {
739 RTE_LOG(ERR, PMD, "%s: grant init failed\n", __func__);
743 eth_dev_xenvirt_create(name, params, rte_socket_id(), DEV_CREATE);
748 rte_pmd_xenvirt_remove(const char *name)
750 eth_dev_xenvirt_free(name, rte_socket_id());
752 if (virtio_idx == 0) {
753 if (xenstore_uninit() != 0)
754 RTE_LOG(ERR, PMD, "%s: xenstore uninit failed\n", __func__);
761 static struct rte_vdev_driver pmd_xenvirt_drv = {
762 .probe = rte_pmd_xenvirt_probe,
763 .remove = rte_pmd_xenvirt_remove,
766 RTE_PMD_REGISTER_VDEV(net_xenvirt, pmd_xenvirt_drv);
767 RTE_PMD_REGISTER_ALIAS(net_xenvirt, eth_xenvirt);
768 RTE_PMD_REGISTER_PARAM_STRING(net_xenvirt,