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>
56 #include <cmdline_parse.h>
57 #include <cmdline_parse_etheraddr.h>
59 #include "rte_xen_lib.h"
60 #include "virtqueue.h"
61 #include "rte_eth_xenvirt.h"
63 #define VQ_DESC_NUM 256
64 #define VIRTIO_MBUF_BURST_SZ 64
66 /* virtio_idx is increased after new device is created.*/
67 static int virtio_idx = 0;
69 static const char *drivername = "xen dummy virtio PMD";
71 static struct rte_eth_link pmd_link = {
73 .link_duplex = ETH_LINK_FULL_DUPLEX,
77 static inline struct rte_mbuf *
78 rte_rxmbuf_alloc(struct rte_mempool *mp)
82 m = __rte_mbuf_raw_alloc(mp);
83 __rte_mbuf_sanity_check_raw(m, RTE_MBUF_PKT, 0);
90 eth_xenvirt_rx(void *q, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
92 struct virtqueue *rxvq = q;
93 struct rte_mbuf *rxm, *new_mbuf;
94 uint16_t nb_used, num;
95 uint32_t len[VIRTIO_MBUF_BURST_SZ];
97 struct pmd_internals *pi = rxvq->internals;
99 nb_used = VIRTQUEUE_NUSED(rxvq);
101 rte_compiler_barrier(); /* rmb */
102 num = (uint16_t)(likely(nb_used <= nb_pkts) ? nb_used : nb_pkts);
103 num = (uint16_t)(likely(num <= VIRTIO_MBUF_BURST_SZ) ? num : VIRTIO_MBUF_BURST_SZ);
104 if (unlikely(num == 0)) return 0;
106 num = virtqueue_dequeue_burst(rxvq, rx_pkts, len, num);
107 PMD_RX_LOG(DEBUG, "used:%d dequeue:%d\n", nb_used, num);
108 for (i = 0; i < num ; i ++) {
110 PMD_RX_LOG(DEBUG, "packet len:%d\n", len[i]);
111 rxm->pkt.next = NULL;
112 rxm->pkt.data = (char *)rxm->buf_addr + RTE_PKTMBUF_HEADROOM;
113 rxm->pkt.data_len = (uint16_t)(len[i] - sizeof(struct virtio_net_hdr));
114 rxm->pkt.nb_segs = 1;
115 rxm->pkt.in_port = pi->port_id;
116 rxm->pkt.pkt_len = (uint32_t)(len[i] - sizeof(struct virtio_net_hdr));
118 /* allocate new mbuf for the used descriptor */
119 while (likely(!virtqueue_full(rxvq))) {
120 new_mbuf = rte_rxmbuf_alloc(rxvq->mpool);
121 if (unlikely(new_mbuf == NULL)) {
124 if (unlikely(virtqueue_enqueue_recv_refill(rxvq, new_mbuf))) {
125 rte_pktmbuf_free_seg(new_mbuf);
129 pi->eth_stats.ipackets += num;
134 eth_xenvirt_tx(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
136 struct virtqueue *txvq = tx_queue;
137 struct rte_mbuf *txm;
138 uint16_t nb_used, nb_tx, num, i;
140 uint32_t len[VIRTIO_MBUF_BURST_SZ];
141 struct rte_mbuf *snd_pkts[VIRTIO_MBUF_BURST_SZ];
142 struct pmd_internals *pi = txvq->internals;
146 if (unlikely(nb_pkts == 0))
149 PMD_TX_LOG(DEBUG, "%d packets to xmit", nb_pkts);
150 nb_used = VIRTQUEUE_NUSED(txvq);
152 rte_compiler_barrier(); /* rmb */
154 num = (uint16_t)(likely(nb_used <= VIRTIO_MBUF_BURST_SZ) ? nb_used : VIRTIO_MBUF_BURST_SZ);
155 num = virtqueue_dequeue_burst(txvq, snd_pkts, len, num);
157 for (i = 0; i < num ; i ++) {
158 /* mergable not supported, one segment only */
159 rte_pktmbuf_free_seg(snd_pkts[i]);
162 while (nb_tx < nb_pkts) {
163 if (likely(!virtqueue_full(txvq))) {
164 /* TODO drop tx_pkts if it contains multiple segments */
165 txm = tx_pkts[nb_tx];
166 error = virtqueue_enqueue_xmit(txvq, txm);
167 if (unlikely(error)) {
169 PMD_TX_LOG(ERR, "virtqueue_enqueue Free count = 0\n");
170 else if (error == EMSGSIZE)
171 PMD_TX_LOG(ERR, "virtqueue_enqueue Free count < 1\n");
173 PMD_TX_LOG(ERR, "virtqueue_enqueue error: %d\n", error);
178 PMD_TX_LOG(ERR, "No free tx descriptors to transmit\n");
179 /* virtqueue_notify not needed in our para-virt solution */
183 pi->eth_stats.opackets += nb_tx;
188 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
190 RTE_LOG(ERR, PMD, "%s\n", __func__);
195 * Create a shared page between guest and host.
196 * Host monitors this page if it is cleared on unmap, and then
197 * do necessary clean up.
200 gntalloc_vring_flag(int vtidx)
202 char key_str[PATH_MAX];
203 char val_str[PATH_MAX];
207 if (grefwatch_from_alloc(&gref_tmp, &ptr)) {
208 RTE_LOG(ERR, PMD, "grefwatch_from_alloc error\n");
212 *(uint8_t *)ptr = MAP_FLAG;
213 rte_snprintf(val_str, sizeof(val_str), "%u", gref_tmp);
214 rte_snprintf(key_str, sizeof(key_str),
215 DPDK_XENSTORE_PATH"%d"VRING_FLAG_STR, vtidx);
216 xenstore_write(key_str, val_str);
220 * Notify host this virtio device is started.
221 * Host could start polling this device.
224 dev_start_notify(int vtidx)
226 char key_str[PATH_MAX];
227 char val_str[PATH_MAX];
229 RTE_LOG(INFO, PMD, "%s: virtio %d is started\n", __func__, vtidx);
230 gntalloc_vring_flag(vtidx);
232 rte_snprintf(key_str, sizeof(key_str), "%s%s%d",
233 DPDK_XENSTORE_PATH, EVENT_TYPE_START_STR,
235 rte_snprintf(val_str, sizeof(val_str), "1");
236 xenstore_write(key_str, val_str);
240 * Notify host this virtio device is stopped.
241 * Host could stop polling this device.
244 dev_stop_notify(int vtidx)
251 update_mac_address(struct ether_addr *mac_addrs, int vtidx)
253 char key_str[PATH_MAX];
254 char val_str[PATH_MAX];
257 if (mac_addrs == NULL) {
258 RTE_LOG(ERR, PMD, "%s: NULL pointer mac specified\n", __func__);
261 rv = rte_snprintf(key_str, sizeof(key_str),
262 DPDK_XENSTORE_PATH"%d_ether_addr", vtidx);
265 rv = rte_snprintf(val_str, sizeof(val_str), "%02x:%02x:%02x:%02x:%02x:%02x",
266 mac_addrs->addr_bytes[0],
267 mac_addrs->addr_bytes[1],
268 mac_addrs->addr_bytes[2],
269 mac_addrs->addr_bytes[3],
270 mac_addrs->addr_bytes[4],
271 mac_addrs->addr_bytes[5]);
274 if (xenstore_write(key_str, val_str))
281 eth_dev_start(struct rte_eth_dev *dev)
283 struct virtqueue *rxvq = dev->data->rx_queues[0];
284 struct virtqueue *txvq = dev->data->tx_queues[0];
286 struct pmd_internals *pi = (struct pmd_internals *)dev->data->dev_private;
289 dev->data->dev_link.link_status = 1;
290 while (!virtqueue_full(rxvq)) {
291 m = rte_rxmbuf_alloc(rxvq->mpool);
294 /* Enqueue allocated buffers. */
295 if (virtqueue_enqueue_recv_refill(rxvq, m)) {
296 rte_pktmbuf_free_seg(m);
301 rxvq->internals = pi;
302 txvq->internals = pi;
304 rv = update_mac_address(dev->data->mac_addrs, pi->virtio_idx);
307 dev_start_notify(pi->virtio_idx);
313 eth_dev_stop(struct rte_eth_dev *dev)
315 struct pmd_internals *pi = (struct pmd_internals *)dev->data->dev_private;
317 dev->data->dev_link.link_status = 0;
318 dev_stop_notify(pi->virtio_idx);
322 * Notify host this virtio device is closed.
323 * Host could do necessary clean up to this device.
326 eth_dev_close(struct rte_eth_dev *dev)
332 eth_dev_info(struct rte_eth_dev *dev,
333 struct rte_eth_dev_info *dev_info)
335 struct pmd_internals *internals = dev->data->dev_private;
337 RTE_SET_USED(internals);
338 dev_info->driver_name = drivername;
339 dev_info->max_mac_addrs = 1;
340 dev_info->max_rx_pktlen = (uint32_t)2048;
341 dev_info->max_rx_queues = (uint16_t)1;
342 dev_info->max_tx_queues = (uint16_t)1;
343 dev_info->min_rx_bufsize = 0;
344 dev_info->pci_dev = NULL;
348 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
350 struct pmd_internals *internals = dev->data->dev_private;
352 rte_memcpy(stats, &internals->eth_stats, sizeof(*stats));
356 eth_stats_reset(struct rte_eth_dev *dev)
358 struct pmd_internals *internals = dev->data->dev_private;
359 /* Reset software totals */
360 memset(&internals->eth_stats, 0, sizeof(internals->eth_stats));
364 eth_queue_release(void *q __rte_unused)
369 eth_link_update(struct rte_eth_dev *dev __rte_unused,
370 int wait_to_complete __rte_unused)
376 * Create shared vring between guest and host.
377 * Memory is allocated through grant alloc driver, so it is not physical continous.
380 gntalloc_vring_create(int queue_type, uint32_t size, int vtidx)
382 char key_str[PATH_MAX] = {0};
383 char val_str[PATH_MAX] = {0};
387 uint32_t *gref_arr = NULL;
388 phys_addr_t *pa_arr = NULL;
389 uint64_t start_index;
392 pg_size = getpagesize();
393 size = RTE_ALIGN_CEIL(size, pg_size);
394 pg_num = size / pg_size;
396 gref_arr = calloc(pg_num, sizeof(gref_arr[0]));
397 pa_arr = calloc(pg_num, sizeof(pa_arr[0]));
399 if (gref_arr == NULL || pa_arr == NULL) {
400 RTE_LOG(ERR, PMD, "%s: calloc failed\n", __func__);
404 va = gntalloc(size, gref_arr, &start_index);
406 RTE_LOG(ERR, PMD, "%s: gntalloc failed\n", __func__);
410 if (get_phys_map(va, pa_arr, pg_num, pg_size))
413 /* write in xenstore gref and pfn for each page of vring */
414 if (grant_node_create(pg_num, gref_arr, pa_arr, val_str, sizeof(val_str))) {
415 gntfree(va, size, start_index);
420 if (queue_type == VTNET_RQ)
421 rv = rte_snprintf(key_str, sizeof(key_str), DPDK_XENSTORE_PATH"%d"RXVRING_XENSTORE_STR, vtidx);
423 rv = rte_snprintf(key_str, sizeof(key_str), DPDK_XENSTORE_PATH"%d"TXVRING_XENSTORE_STR, vtidx);
424 if (rv == -1 || xenstore_write(key_str, val_str) == -1) {
425 gntfree(va, size, start_index);
439 static struct virtqueue *
440 virtio_queue_setup(struct rte_eth_dev *dev, int queue_type)
442 struct virtqueue *vq = NULL;
443 uint16_t vq_size = VQ_DESC_NUM;
445 char vq_name[VIRTQUEUE_MAX_NAME_SZ];
449 /* Allocate memory for virtqueue. */
450 if (queue_type == VTNET_RQ) {
451 rte_snprintf(vq_name, sizeof(vq_name), "port%d_rvq",
453 vq = rte_zmalloc(vq_name, sizeof(struct virtqueue) +
454 vq_size * sizeof(struct vq_desc_extra), CACHE_LINE_SIZE);
456 RTE_LOG(ERR, PMD, "%s: unabled to allocate virtqueue\n", __func__);
459 memcpy(vq->vq_name, vq_name, sizeof(vq->vq_name));
460 } else if(queue_type == VTNET_TQ) {
461 rte_snprintf(vq_name, sizeof(vq_name), "port%d_tvq",
463 vq = rte_zmalloc(vq_name, sizeof(struct virtqueue) +
464 vq_size * sizeof(struct vq_desc_extra), CACHE_LINE_SIZE);
466 RTE_LOG(ERR, PMD, "%s: unabled to allocate virtqueue\n", __func__);
469 memcpy(vq->vq_name, vq_name, sizeof(vq->vq_name));
472 memcpy(vq->vq_name, vq_name, sizeof(vq->vq_name));
474 vq->vq_alignment = VIRTIO_PCI_VRING_ALIGN;
475 vq->vq_nentries = vq_size;
476 vq->vq_free_cnt = vq_size;
477 /* Calcuate vring size according to virtio spec */
478 size = vring_size(vq_size, VIRTIO_PCI_VRING_ALIGN);
479 vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
480 /* Allocate memory for virtio vring through gntalloc driver*/
481 vq->vq_ring_virt_mem = gntalloc_vring_create(queue_type, vq->vq_ring_size,
482 ((struct pmd_internals *)dev->data->dev_private)->virtio_idx);
483 memset(vq->vq_ring_virt_mem, 0, vq->vq_ring_size);
485 vring_init(vr, vq_size, vq->vq_ring_virt_mem, vq->vq_alignment);
487 * Locally maintained last consumed index, this idex trails
490 vq->vq_used_cons_idx = 0;
491 vq->vq_desc_head_idx = 0;
492 vq->vq_free_cnt = vq->vq_nentries;
493 memset(vq->vq_descx, 0, sizeof(struct vq_desc_extra) * vq->vq_nentries);
495 /* Chain all the descriptors in the ring with an END */
496 for (i = 0; i < vq_size - 1; i++)
497 vr->desc[i].next = (uint16_t)(i + 1);
498 vr->desc[i].next = VQ_RING_DESC_CHAIN_END;
504 eth_rx_queue_setup(struct rte_eth_dev *dev,uint16_t rx_queue_id,
505 uint16_t nb_rx_desc __rte_unused,
506 unsigned int socket_id __rte_unused,
507 const struct rte_eth_rxconf *rx_conf __rte_unused,
508 struct rte_mempool *mb_pool)
510 struct virtqueue *vq;
511 vq = dev->data->rx_queues[rx_queue_id] = virtio_queue_setup(dev, VTNET_RQ);
517 eth_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
518 uint16_t nb_tx_desc __rte_unused,
519 unsigned int socket_id __rte_unused,
520 const struct rte_eth_txconf *tx_conf __rte_unused)
522 dev->data->tx_queues[tx_queue_id] = virtio_queue_setup(dev, VTNET_TQ);
528 static struct eth_dev_ops ops = {
529 .dev_start = eth_dev_start,
530 .dev_stop = eth_dev_stop,
531 .dev_close = eth_dev_close,
532 .dev_configure = eth_dev_configure,
533 .dev_infos_get = eth_dev_info,
534 .rx_queue_setup = eth_rx_queue_setup,
535 .tx_queue_setup = eth_tx_queue_setup,
536 .rx_queue_release = eth_queue_release,
537 .tx_queue_release = eth_queue_release,
538 .link_update = eth_link_update,
539 .stats_get = eth_stats_get,
540 .stats_reset = eth_stats_reset,
545 rte_eth_xenvirt_parse_args(struct xenvirt_dict *dict,
546 const char *name, const char *params)
549 char *pairs[RTE_ETH_XENVIRT_MAX_ARGS];
558 args = rte_zmalloc(NULL, strlen(params) + 1, CACHE_LINE_SIZE);
560 RTE_LOG(ERR, PMD, "Couldn't parse %s device \n", name);
563 rte_memcpy(args, params, strlen(params));
565 num_of_pairs = rte_strsplit(args, strnlen(args, MAX_ARG_STRLEN),
567 RTE_ETH_XENVIRT_MAX_ARGS ,
568 RTE_ETH_XENVIRT_PAIRS_DELIM);
570 for (i = 0; i < num_of_pairs; i++) {
573 rte_strsplit(pairs[i], strnlen(pairs[i], MAX_ARG_STRLEN),
575 RTE_ETH_XENVIRT_KEY_VALUE_DELIM);
577 if (pair[0] == NULL || pair[1] == NULL || pair[0][0] == 0
578 || pair[1][0] == 0) {
580 "Couldn't parse %s device,"
581 "wrong key or value \n", name);
585 if (!strncmp(pair[0], RTE_ETH_XENVIRT_MAC_PARAM,
586 sizeof(RTE_ETH_XENVIRT_MAC_PARAM))) {
587 if (cmdline_parse_etheraddr(NULL,
591 "Invalid %s device ether address\n",
596 dict->addr_valid = 1;
613 eth_dev_xenvirt_create(const char *name, const char *params,
614 const unsigned numa_node,
615 enum dev_action action)
617 struct rte_eth_dev_data *data = NULL;
618 struct rte_pci_device *pci_dev = NULL;
619 struct pmd_internals *internals = NULL;
620 struct rte_eth_dev *eth_dev = NULL;
621 struct xenvirt_dict dict;
622 bzero(&dict, sizeof(struct xenvirt_dict));
624 RTE_LOG(INFO, PMD, "Creating virtio rings backed ethdev on numa socket %u\n",
626 RTE_SET_USED(action);
628 if (rte_eth_xenvirt_parse_args(&dict, name, params) < 0) {
629 RTE_LOG(ERR, PMD, "%s: Failed to parse ethdev parameters\n", __func__);
633 /* now do all data allocation - for eth_dev structure, dummy pci driver
634 * and internal (private) data
636 data = rte_zmalloc_socket(name, sizeof(*data), 0, numa_node);
640 pci_dev = rte_zmalloc_socket(name, sizeof(*pci_dev), 0, numa_node);
644 internals = rte_zmalloc_socket(name, sizeof(*internals), 0, numa_node);
645 if (internals == NULL)
648 /* reserve an ethdev entry */
649 eth_dev = rte_eth_dev_allocate();
653 pci_dev->numa_node = numa_node;
655 data->dev_private = internals;
656 data->port_id = eth_dev->data->port_id;
657 data->nb_rx_queues = (uint16_t)1;
658 data->nb_tx_queues = (uint16_t)1;
659 data->dev_link = pmd_link;
660 data->mac_addrs = rte_zmalloc("xen_virtio", ETHER_ADDR_LEN, 0);
663 memcpy(&data->mac_addrs->addr_bytes, &dict.addr, sizeof(struct ether_addr));
665 eth_random_addr(&data->mac_addrs->addr_bytes[0]);
667 eth_dev->data = data;
668 eth_dev->dev_ops = &ops;
669 eth_dev->pci_dev = pci_dev;
671 eth_dev->rx_pkt_burst = eth_xenvirt_rx;
672 eth_dev->tx_pkt_burst = eth_xenvirt_tx;
674 internals->virtio_idx = virtio_idx++;
675 internals->port_id = eth_dev->data->port_id;
690 /*TODO: Support multiple process model */
692 rte_pmd_xenvirt_init(const char *name, const char *params)
694 if (virtio_idx == 0) {
695 if (xenstore_init() != 0) {
696 RTE_LOG(ERR, PMD, "%s: xenstore init failed\n", __func__);
699 if (gntalloc_open() != 0) {
700 RTE_LOG(ERR, PMD, "%s: grant init failed\n", __func__);
704 eth_dev_xenvirt_create(name, params, rte_socket_id(), DEV_CREATE);