4 * Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of version 2 of the GNU General Public License as
8 * published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
18 * The full GNU General Public License is included in this distribution
19 * in the file called LICENSE.GPL.
21 * Contact Information:
26 * This code is inspired from the book "Linux Device Drivers" by
27 * Alessandro Rubini and Jonathan Corbet, published by O'Reilly & Associates
30 #include <linux/device.h>
31 #include <linux/module.h>
32 #include <linux/version.h>
33 #include <linux/netdevice.h>
34 #include <linux/etherdevice.h> /* eth_type_trans */
35 #include <linux/skbuff.h>
36 #include <linux/kthread.h>
37 #include <linux/delay.h>
39 #include <exec-env/rte_kni_common.h>
43 #define WD_TIMEOUT 5 /*jiffies */
45 #define MBUF_BURST_SZ 32
47 #define KNI_WAIT_RESPONSE_TIMEOUT 300 /* 3 seconds */
49 /* typedef for rx function */
50 typedef void (*kni_net_rx_t)(struct kni_dev *kni);
52 static int kni_net_tx(struct sk_buff *skb, struct net_device *dev);
53 static void kni_net_rx_normal(struct kni_dev *kni);
54 static void kni_net_rx_lo_fifo(struct kni_dev *kni);
55 static void kni_net_rx_lo_fifo_skb(struct kni_dev *kni);
56 static int kni_net_process_request(struct kni_dev *kni,
57 struct rte_kni_request *req);
59 /* kni rx function pointer, with default to normal rx */
60 static kni_net_rx_t kni_net_rx_func = kni_net_rx_normal;
66 kni_net_open(struct net_device *dev)
69 struct rte_kni_request req;
70 struct kni_dev *kni = netdev_priv(dev);
73 memcpy(dev->dev_addr, kni->lad_dev->dev_addr, ETH_ALEN);
76 * Generate random mac address. eth_random_addr() is the newer
77 * version of generating mac address in linux kernel.
79 random_ether_addr(dev->dev_addr);
81 netif_start_queue(dev);
83 memset(&req, 0, sizeof(req));
84 req.req_id = RTE_KNI_REQ_CFG_NETWORK_IF;
86 /* Setting if_up to non-zero means up */
88 ret = kni_net_process_request(kni, &req);
90 return (ret == 0) ? req.result : ret;
94 kni_net_release(struct net_device *dev)
97 struct rte_kni_request req;
98 struct kni_dev *kni = netdev_priv(dev);
100 netif_stop_queue(dev); /* can't transmit any more */
102 memset(&req, 0, sizeof(req));
103 req.req_id = RTE_KNI_REQ_CFG_NETWORK_IF;
105 /* Setting if_up to 0 means down */
107 ret = kni_net_process_request(kni, &req);
109 return (ret == 0) ? req.result : ret;
113 * Configuration changes (passed on by ifconfig)
116 kni_net_config(struct net_device *dev, struct ifmap *map)
118 if (dev->flags & IFF_UP) /* can't act on a running interface */
121 /* ignore other fields */
126 * RX: normal working mode
129 kni_net_rx_normal(struct kni_dev *kni)
133 unsigned i, num_rx, num_fq;
134 struct rte_kni_mbuf *kva;
135 struct rte_kni_mbuf *va[MBUF_BURST_SZ];
139 struct net_device *dev = kni->net_dev;
141 /* Get the number of free entries in free_q */
142 num_fq = kni_fifo_free_count(kni->free_q);
144 /* No room on the free_q, bail out */
148 /* Calculate the number of entries to dequeue from rx_q */
149 num_rx = min(num_fq, (unsigned)MBUF_BURST_SZ);
151 /* Burst dequeue from rx_q */
152 num_rx = kni_fifo_get(kni->rx_q, (void **)va, num_rx);
156 /* Transfer received packets to netif */
157 for (i = 0; i < num_rx; i++) {
158 kva = (void *)va[i] - kni->mbuf_va + kni->mbuf_kva;
160 data_kva = kva->buf_addr + kva->data_off - kni->mbuf_va
163 skb = dev_alloc_skb(len + 2);
165 KNI_ERR("Out of mem, dropping pkts\n");
166 /* Update statistics */
167 kni->stats.rx_dropped++;
170 /* Align IP on 16B boundary */
172 memcpy(skb_put(skb, len), data_kva, len);
174 skb->protocol = eth_type_trans(skb, dev);
175 skb->ip_summed = CHECKSUM_UNNECESSARY;
177 /* Call netif interface */
180 /* Update statistics */
181 kni->stats.rx_bytes += len;
182 kni->stats.rx_packets++;
186 /* Burst enqueue mbufs into free_q */
187 ret = kni_fifo_put(kni->free_q, (void **)va, num_rx);
189 /* Failing should not happen */
190 KNI_ERR("Fail to enqueue entries into free_q\n");
194 * RX: loopback with enqueue/dequeue fifos.
197 kni_net_rx_lo_fifo(struct kni_dev *kni)
201 unsigned i, num, num_rq, num_tq, num_aq, num_fq;
202 struct rte_kni_mbuf *kva;
203 struct rte_kni_mbuf *va[MBUF_BURST_SZ];
206 struct rte_kni_mbuf *alloc_kva;
207 struct rte_kni_mbuf *alloc_va[MBUF_BURST_SZ];
208 void *alloc_data_kva;
210 /* Get the number of entries in rx_q */
211 num_rq = kni_fifo_count(kni->rx_q);
213 /* Get the number of free entrie in tx_q */
214 num_tq = kni_fifo_free_count(kni->tx_q);
216 /* Get the number of entries in alloc_q */
217 num_aq = kni_fifo_count(kni->alloc_q);
219 /* Get the number of free entries in free_q */
220 num_fq = kni_fifo_free_count(kni->free_q);
222 /* Calculate the number of entries to be dequeued from rx_q */
223 num = min(num_rq, num_tq);
224 num = min(num, num_aq);
225 num = min(num, num_fq);
226 num = min(num, (unsigned)MBUF_BURST_SZ);
228 /* Return if no entry to dequeue from rx_q */
232 /* Burst dequeue from rx_q */
233 ret = kni_fifo_get(kni->rx_q, (void **)va, num);
235 return; /* Failing should not happen */
237 /* Dequeue entries from alloc_q */
238 ret = kni_fifo_get(kni->alloc_q, (void **)alloc_va, num);
242 for (i = 0; i < num; i++) {
243 kva = (void *)va[i] - kni->mbuf_va + kni->mbuf_kva;
245 data_kva = kva->buf_addr + kva->data_off -
246 kni->mbuf_va + kni->mbuf_kva;
248 alloc_kva = (void *)alloc_va[i] - kni->mbuf_va +
250 alloc_data_kva = alloc_kva->buf_addr +
251 alloc_kva->data_off - kni->mbuf_va +
253 memcpy(alloc_data_kva, data_kva, len);
254 alloc_kva->pkt_len = len;
255 alloc_kva->data_len = len;
257 kni->stats.tx_bytes += len;
258 kni->stats.rx_bytes += len;
261 /* Burst enqueue mbufs into tx_q */
262 ret = kni_fifo_put(kni->tx_q, (void **)alloc_va, num);
264 /* Failing should not happen */
265 KNI_ERR("Fail to enqueue mbufs into tx_q\n");
268 /* Burst enqueue mbufs into free_q */
269 ret = kni_fifo_put(kni->free_q, (void **)va, num);
271 /* Failing should not happen */
272 KNI_ERR("Fail to enqueue mbufs into free_q\n");
275 * Update statistic, and enqueue/dequeue failure is impossible,
276 * as all queues are checked at first.
278 kni->stats.tx_packets += num;
279 kni->stats.rx_packets += num;
283 * RX: loopback with enqueue/dequeue fifos and sk buffer copies.
286 kni_net_rx_lo_fifo_skb(struct kni_dev *kni)
290 unsigned i, num_rq, num_fq, num;
291 struct rte_kni_mbuf *kva;
292 struct rte_kni_mbuf *va[MBUF_BURST_SZ];
296 struct net_device *dev = kni->net_dev;
298 /* Get the number of entries in rx_q */
299 num_rq = kni_fifo_count(kni->rx_q);
301 /* Get the number of free entries in free_q */
302 num_fq = kni_fifo_free_count(kni->free_q);
304 /* Calculate the number of entries to dequeue from rx_q */
305 num = min(num_rq, num_fq);
306 num = min(num, (unsigned)MBUF_BURST_SZ);
308 /* Return if no entry to dequeue from rx_q */
312 /* Burst dequeue mbufs from rx_q */
313 ret = kni_fifo_get(kni->rx_q, (void **)va, num);
317 /* Copy mbufs to sk buffer and then call tx interface */
318 for (i = 0; i < num; i++) {
319 kva = (void *)va[i] - kni->mbuf_va + kni->mbuf_kva;
321 data_kva = kva->buf_addr + kva->data_off - kni->mbuf_va +
324 skb = dev_alloc_skb(len + 2);
326 KNI_ERR("Out of mem, dropping pkts\n");
328 /* Align IP on 16B boundary */
330 memcpy(skb_put(skb, len), data_kva, len);
332 skb->ip_summed = CHECKSUM_UNNECESSARY;
336 /* Simulate real usage, allocate/copy skb twice */
337 skb = dev_alloc_skb(len + 2);
339 KNI_ERR("Out of mem, dropping pkts\n");
340 kni->stats.rx_dropped++;
343 /* Align IP on 16B boundary */
345 memcpy(skb_put(skb, len), data_kva, len);
347 skb->ip_summed = CHECKSUM_UNNECESSARY;
349 kni->stats.rx_bytes += len;
350 kni->stats.rx_packets++;
352 /* call tx interface */
353 kni_net_tx(skb, dev);
357 /* enqueue all the mbufs from rx_q into free_q */
358 ret = kni_fifo_put(kni->free_q, (void **)&va, num);
360 /* Failing should not happen */
361 KNI_ERR("Fail to enqueue mbufs into free_q\n");
366 kni_net_rx(struct kni_dev *kni)
369 * It doesn't need to check if it is NULL pointer,
370 * as it has a default value
372 (*kni_net_rx_func)(kni);
376 * Transmit a packet (called by the kernel)
380 kni_net_tx(struct sk_buff *skb, struct net_device *dev)
382 struct kni_dev *kni = netdev_priv(dev);
385 kni->stats.tx_dropped++;
391 kni_net_tx(struct sk_buff *skb, struct net_device *dev)
395 struct kni_dev *kni = netdev_priv(dev);
396 struct rte_kni_mbuf *pkt_kva = NULL;
397 struct rte_kni_mbuf *pkt_va = NULL;
399 dev->trans_start = jiffies; /* save the timestamp */
401 /* Check if the length of skb is less than mbuf size */
402 if (skb->len > kni->mbuf_size)
406 * Check if it has at least one free entry in tx_q and
407 * one entry in alloc_q.
409 if (kni_fifo_free_count(kni->tx_q) == 0 ||
410 kni_fifo_count(kni->alloc_q) == 0) {
412 * If no free entry in tx_q or no entry in alloc_q,
413 * drops skb and goes out.
418 /* dequeue a mbuf from alloc_q */
419 ret = kni_fifo_get(kni->alloc_q, (void **)&pkt_va, 1);
420 if (likely(ret == 1)) {
423 pkt_kva = (void *)pkt_va - kni->mbuf_va + kni->mbuf_kva;
424 data_kva = pkt_kva->buf_addr + pkt_kva->data_off - kni->mbuf_va
428 memcpy(data_kva, skb->data, len);
429 if (unlikely(len < ETH_ZLEN)) {
430 memset(data_kva + len, 0, ETH_ZLEN - len);
433 pkt_kva->pkt_len = len;
434 pkt_kva->data_len = len;
436 /* enqueue mbuf into tx_q */
437 ret = kni_fifo_put(kni->tx_q, (void **)&pkt_va, 1);
438 if (unlikely(ret != 1)) {
439 /* Failing should not happen */
440 KNI_ERR("Fail to enqueue mbuf into tx_q\n");
444 /* Failing should not happen */
445 KNI_ERR("Fail to dequeue mbuf from alloc_q\n");
449 /* Free skb and update statistics */
451 kni->stats.tx_bytes += len;
452 kni->stats.tx_packets++;
457 /* Free skb and update statistics */
459 kni->stats.tx_dropped++;
466 * Deal with a transmit timeout.
469 kni_net_tx_timeout (struct net_device *dev)
471 struct kni_dev *kni = netdev_priv(dev);
473 KNI_DBG("Transmit timeout at %ld, latency %ld\n", jiffies,
474 jiffies - dev->trans_start);
476 kni->stats.tx_errors++;
477 netif_wake_queue(dev);
485 kni_net_ioctl(struct net_device *dev, struct ifreq *rq, int cmd)
487 KNI_DBG("kni_net_ioctl %d\n",
488 ((struct kni_dev *)netdev_priv(dev))->group_id);
494 kni_net_set_rx_mode(struct net_device *dev)
499 kni_net_change_mtu(struct net_device *dev, int new_mtu)
502 struct rte_kni_request req;
503 struct kni_dev *kni = netdev_priv(dev);
505 KNI_DBG("kni_net_change_mtu new mtu %d to be set\n", new_mtu);
507 memset(&req, 0, sizeof(req));
508 req.req_id = RTE_KNI_REQ_CHANGE_MTU;
509 req.new_mtu = new_mtu;
510 ret = kni_net_process_request(kni, &req);
511 if (ret == 0 && req.result == 0)
514 return (ret == 0) ? req.result : ret;
518 * Checks if the user space application provided the resp message
521 kni_net_poll_resp(struct kni_dev *kni)
523 if (kni_fifo_count(kni->resp_q))
524 wake_up_interruptible(&kni->wq);
528 * It can be called to process the request.
531 kni_net_process_request(struct kni_dev *kni, struct rte_kni_request *req)
539 KNI_ERR("No kni instance or request\n");
543 mutex_lock(&kni->sync_lock);
546 memcpy(kni->sync_kva, req, sizeof(struct rte_kni_request));
547 num = kni_fifo_put(kni->req_q, &kni->sync_va, 1);
549 KNI_ERR("Cannot send to req_q\n");
554 ret_val = wait_event_interruptible_timeout(kni->wq,
555 kni_fifo_count(kni->resp_q), 3 * HZ);
556 if (signal_pending(current) || ret_val <= 0) {
560 num = kni_fifo_get(kni->resp_q, (void **)&resp_va, 1);
561 if (num != 1 || resp_va != kni->sync_va) {
562 /* This should never happen */
563 KNI_ERR("No data in resp_q\n");
568 memcpy(req, kni->sync_kva, sizeof(struct rte_kni_request));
572 mutex_unlock(&kni->sync_lock);
577 * Return statistics to the caller
579 static struct net_device_stats *
580 kni_net_stats(struct net_device *dev)
582 struct kni_dev *kni = netdev_priv(dev);
587 * Fill the eth header
590 kni_net_header(struct sk_buff *skb, struct net_device *dev,
591 unsigned short type, const void *daddr,
592 const void *saddr, unsigned int len)
594 struct ethhdr *eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
596 memcpy(eth->h_source, saddr ? saddr : dev->dev_addr, dev->addr_len);
597 memcpy(eth->h_dest, daddr ? daddr : dev->dev_addr, dev->addr_len);
598 eth->h_proto = htons(type);
600 return dev->hard_header_len;
605 * Re-fill the eth header
607 #if (LINUX_VERSION_CODE < KERNEL_VERSION(4, 1, 0))
609 kni_net_rebuild_header(struct sk_buff *skb)
611 struct net_device *dev = skb->dev;
612 struct ethhdr *eth = (struct ethhdr *) skb->data;
614 memcpy(eth->h_source, dev->dev_addr, dev->addr_len);
615 memcpy(eth->h_dest, dev->dev_addr, dev->addr_len);
622 * kni_net_set_mac - Change the Ethernet Address of the KNI NIC
623 * @netdev: network interface device structure
624 * @p: pointer to an address structure
626 * Returns 0 on success, negative on failure
628 static int kni_net_set_mac(struct net_device *netdev, void *p)
630 struct sockaddr *addr = p;
631 if (!is_valid_ether_addr((unsigned char *)(addr->sa_data)))
632 return -EADDRNOTAVAIL;
633 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
637 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(3, 9, 0))
638 static int kni_net_change_carrier(struct net_device *dev, bool new_carrier)
641 netif_carrier_on(dev);
643 netif_carrier_off(dev);
648 static const struct header_ops kni_net_header_ops = {
649 .create = kni_net_header,
650 #if (LINUX_VERSION_CODE < KERNEL_VERSION(4, 1, 0))
651 .rebuild = kni_net_rebuild_header,
653 .cache = NULL, /* disable caching */
656 static const struct net_device_ops kni_net_netdev_ops = {
657 .ndo_open = kni_net_open,
658 .ndo_stop = kni_net_release,
659 .ndo_set_config = kni_net_config,
660 .ndo_start_xmit = kni_net_tx,
661 .ndo_change_mtu = kni_net_change_mtu,
662 .ndo_do_ioctl = kni_net_ioctl,
663 .ndo_set_rx_mode = kni_net_set_rx_mode,
664 .ndo_get_stats = kni_net_stats,
665 .ndo_tx_timeout = kni_net_tx_timeout,
666 .ndo_set_mac_address = kni_net_set_mac,
667 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(3, 9, 0))
668 .ndo_change_carrier = kni_net_change_carrier,
673 kni_net_init(struct net_device *dev)
675 struct kni_dev *kni = netdev_priv(dev);
677 KNI_DBG("kni_net_init\n");
679 init_waitqueue_head(&kni->wq);
680 mutex_init(&kni->sync_lock);
682 ether_setup(dev); /* assign some of the fields */
683 dev->netdev_ops = &kni_net_netdev_ops;
684 dev->header_ops = &kni_net_header_ops;
685 dev->watchdog_timeo = WD_TIMEOUT;
689 kni_net_config_lo_mode(char *lo_str)
692 KNI_PRINT("loopback disabled");
696 if (!strcmp(lo_str, "lo_mode_none"))
697 KNI_PRINT("loopback disabled");
698 else if (!strcmp(lo_str, "lo_mode_fifo")) {
699 KNI_PRINT("loopback mode=lo_mode_fifo enabled");
700 kni_net_rx_func = kni_net_rx_lo_fifo;
701 } else if (!strcmp(lo_str, "lo_mode_fifo_skb")) {
702 KNI_PRINT("loopback mode=lo_mode_fifo_skb enabled");
703 kni_net_rx_func = kni_net_rx_lo_fifo_skb;
705 KNI_PRINT("Incognizant parameter, loopback disabled");