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 <rte_config.h>
40 #include <exec-env/rte_kni_common.h>
44 #define WD_TIMEOUT 5 /*jiffies */
46 #define MBUF_BURST_SZ 32
48 #define KNI_WAIT_RESPONSE_TIMEOUT 300 /* 3 seconds */
50 /* typedef for rx function */
51 typedef void (*kni_net_rx_t)(struct kni_dev *kni);
53 static int kni_net_tx(struct sk_buff *skb, struct net_device *dev);
54 static void kni_net_rx_normal(struct kni_dev *kni);
55 static void kni_net_rx_lo_fifo(struct kni_dev *kni);
56 static void kni_net_rx_lo_fifo_skb(struct kni_dev *kni);
57 static int kni_net_process_request(struct kni_dev *kni,
58 struct rte_kni_request *req);
60 /* kni rx function pointer, with default to normal rx */
61 static kni_net_rx_t kni_net_rx_func = kni_net_rx_normal;
67 kni_net_open(struct net_device *dev)
70 struct rte_kni_request req;
71 struct kni_dev *kni = netdev_priv(dev);
74 memcpy(dev->dev_addr, kni->lad_dev->dev_addr, ETH_ALEN);
77 * Generate random mac address. eth_random_addr() is the newer
78 * version of generating mac address in linux kernel.
80 random_ether_addr(dev->dev_addr);
82 netif_start_queue(dev);
84 memset(&req, 0, sizeof(req));
85 req.req_id = RTE_KNI_REQ_CFG_NETWORK_IF;
87 /* Setting if_up to non-zero means up */
89 ret = kni_net_process_request(kni, &req);
91 return (ret == 0 ? req.result : ret);
95 kni_net_release(struct net_device *dev)
98 struct rte_kni_request req;
99 struct kni_dev *kni = netdev_priv(dev);
101 netif_stop_queue(dev); /* can't transmit any more */
103 memset(&req, 0, sizeof(req));
104 req.req_id = RTE_KNI_REQ_CFG_NETWORK_IF;
106 /* Setting if_up to 0 means down */
108 ret = kni_net_process_request(kni, &req);
110 return (ret == 0 ? req.result : ret);
114 * Configuration changes (passed on by ifconfig)
117 kni_net_config(struct net_device *dev, struct ifmap *map)
119 if (dev->flags & IFF_UP) /* can't act on a running interface */
122 /* ignore other fields */
127 * RX: normal working mode
130 kni_net_rx_normal(struct kni_dev *kni)
134 unsigned i, num, num_rq, num_fq;
135 struct rte_kni_mbuf *kva;
136 struct rte_kni_mbuf *va[MBUF_BURST_SZ];
140 struct net_device *dev = kni->net_dev;
142 /* Get the number of entries in rx_q */
143 num_rq = kni_fifo_count(kni->rx_q);
145 /* Get the number of free entries in free_q */
146 num_fq = kni_fifo_free_count(kni->free_q);
148 /* Calculate the number of entries to dequeue in rx_q */
149 num = min(num_rq, num_fq);
150 num = min(num, (unsigned)MBUF_BURST_SZ);
152 /* Return if no entry in rx_q and no free entry in free_q */
156 /* Burst dequeue from rx_q */
157 ret = kni_fifo_get(kni->rx_q, (void **)va, num);
159 return; /* Failing should not happen */
161 /* Transfer received packets to netif */
162 for (i = 0; i < num; i++) {
163 kva = (void *)va[i] - kni->mbuf_va + kni->mbuf_kva;
165 data_kva = kva->buf_addr + kva->data_off - kni->mbuf_va
168 skb = dev_alloc_skb(len + 2);
170 KNI_ERR("Out of mem, dropping pkts\n");
171 /* Update statistics */
172 kni->stats.rx_dropped++;
175 /* Align IP on 16B boundary */
177 memcpy(skb_put(skb, len), data_kva, len);
179 skb->protocol = eth_type_trans(skb, dev);
180 skb->ip_summed = CHECKSUM_UNNECESSARY;
182 /* Call netif interface */
185 /* Update statistics */
186 kni->stats.rx_bytes += len;
187 kni->stats.rx_packets++;
191 /* Burst enqueue mbufs into free_q */
192 ret = kni_fifo_put(kni->free_q, (void **)va, num);
194 /* Failing should not happen */
195 KNI_ERR("Fail to enqueue entries into free_q\n");
199 * RX: loopback with enqueue/dequeue fifos.
202 kni_net_rx_lo_fifo(struct kni_dev *kni)
206 unsigned i, num, num_rq, num_tq, num_aq, num_fq;
207 struct rte_kni_mbuf *kva;
208 struct rte_kni_mbuf *va[MBUF_BURST_SZ];
211 struct rte_kni_mbuf *alloc_kva;
212 struct rte_kni_mbuf *alloc_va[MBUF_BURST_SZ];
213 void *alloc_data_kva;
215 /* Get the number of entries in rx_q */
216 num_rq = kni_fifo_count(kni->rx_q);
218 /* Get the number of free entrie in tx_q */
219 num_tq = kni_fifo_free_count(kni->tx_q);
221 /* Get the number of entries in alloc_q */
222 num_aq = kni_fifo_count(kni->alloc_q);
224 /* Get the number of free entries in free_q */
225 num_fq = kni_fifo_free_count(kni->free_q);
227 /* Calculate the number of entries to be dequeued from rx_q */
228 num = min(num_rq, num_tq);
229 num = min(num, num_aq);
230 num = min(num, num_fq);
231 num = min(num, (unsigned)MBUF_BURST_SZ);
233 /* Return if no entry to dequeue from rx_q */
237 /* Burst dequeue from rx_q */
238 ret = kni_fifo_get(kni->rx_q, (void **)va, num);
240 return; /* Failing should not happen */
242 /* Dequeue entries from alloc_q */
243 ret = kni_fifo_get(kni->alloc_q, (void **)alloc_va, num);
247 for (i = 0; i < num; i++) {
248 kva = (void *)va[i] - kni->mbuf_va + kni->mbuf_kva;
250 data_kva = kva->buf_addr + kva->data_off -
251 kni->mbuf_va + kni->mbuf_kva;
253 alloc_kva = (void *)alloc_va[i] - kni->mbuf_va +
255 alloc_data_kva = alloc_kva->buf_addr +
256 alloc_kva->data_off - kni->mbuf_va +
258 memcpy(alloc_data_kva, data_kva, len);
259 alloc_kva->pkt_len = len;
260 alloc_kva->data_len = len;
262 kni->stats.tx_bytes += len;
263 kni->stats.rx_bytes += len;
266 /* Burst enqueue mbufs into tx_q */
267 ret = kni_fifo_put(kni->tx_q, (void **)alloc_va, num);
269 /* Failing should not happen */
270 KNI_ERR("Fail to enqueue mbufs into tx_q\n");
273 /* Burst enqueue mbufs into free_q */
274 ret = kni_fifo_put(kni->free_q, (void **)va, num);
276 /* Failing should not happen */
277 KNI_ERR("Fail to enqueue mbufs into free_q\n");
280 * Update statistic, and enqueue/dequeue failure is impossible,
281 * as all queues are checked at first.
283 kni->stats.tx_packets += num;
284 kni->stats.rx_packets += num;
288 * RX: loopback with enqueue/dequeue fifos and sk buffer copies.
291 kni_net_rx_lo_fifo_skb(struct kni_dev *kni)
295 unsigned i, num_rq, num_fq, num;
296 struct rte_kni_mbuf *kva;
297 struct rte_kni_mbuf *va[MBUF_BURST_SZ];
301 struct net_device *dev = kni->net_dev;
303 /* Get the number of entries in rx_q */
304 num_rq = kni_fifo_count(kni->rx_q);
306 /* Get the number of free entries in free_q */
307 num_fq = kni_fifo_free_count(kni->free_q);
309 /* Calculate the number of entries to dequeue from rx_q */
310 num = min(num_rq, num_fq);
311 num = min(num, (unsigned)MBUF_BURST_SZ);
313 /* Return if no entry to dequeue from rx_q */
317 /* Burst dequeue mbufs from rx_q */
318 ret = kni_fifo_get(kni->rx_q, (void **)va, num);
322 /* Copy mbufs to sk buffer and then call tx interface */
323 for (i = 0; i < num; i++) {
324 kva = (void *)va[i] - kni->mbuf_va + kni->mbuf_kva;
326 data_kva = kva->buf_addr + kva->data_off - kni->mbuf_va +
329 skb = dev_alloc_skb(len + 2);
331 KNI_ERR("Out of mem, dropping pkts\n");
333 /* Align IP on 16B boundary */
335 memcpy(skb_put(skb, len), data_kva, len);
337 skb->ip_summed = CHECKSUM_UNNECESSARY;
341 /* Simulate real usage, allocate/copy skb twice */
342 skb = dev_alloc_skb(len + 2);
344 KNI_ERR("Out of mem, dropping pkts\n");
345 kni->stats.rx_dropped++;
348 /* Align IP on 16B boundary */
350 memcpy(skb_put(skb, len), data_kva, len);
352 skb->ip_summed = CHECKSUM_UNNECESSARY;
354 kni->stats.rx_bytes += len;
355 kni->stats.rx_packets++;
357 /* call tx interface */
358 kni_net_tx(skb, dev);
362 /* enqueue all the mbufs from rx_q into free_q */
363 ret = kni_fifo_put(kni->free_q, (void **)&va, num);
365 /* Failing should not happen */
366 KNI_ERR("Fail to enqueue mbufs into free_q\n");
371 kni_net_rx(struct kni_dev *kni)
374 * It doesn't need to check if it is NULL pointer,
375 * as it has a default value
377 (*kni_net_rx_func)(kni);
381 * Transmit a packet (called by the kernel)
385 kni_net_tx(struct sk_buff *skb, struct net_device *dev)
387 struct kni_dev *kni = netdev_priv(dev);
390 kni->stats.tx_dropped++;
396 kni_net_tx(struct sk_buff *skb, struct net_device *dev)
400 struct kni_dev *kni = netdev_priv(dev);
401 struct rte_kni_mbuf *pkt_kva = NULL;
402 struct rte_kni_mbuf *pkt_va = NULL;
404 dev->trans_start = jiffies; /* save the timestamp */
406 /* Check if the length of skb is less than mbuf size */
407 if (skb->len > kni->mbuf_size)
411 * Check if it has at least one free entry in tx_q and
412 * one entry in alloc_q.
414 if (kni_fifo_free_count(kni->tx_q) == 0 ||
415 kni_fifo_count(kni->alloc_q) == 0) {
417 * If no free entry in tx_q or no entry in alloc_q,
418 * drops skb and goes out.
423 /* dequeue a mbuf from alloc_q */
424 ret = kni_fifo_get(kni->alloc_q, (void **)&pkt_va, 1);
425 if (likely(ret == 1)) {
428 pkt_kva = (void *)pkt_va - kni->mbuf_va + kni->mbuf_kva;
429 data_kva = pkt_kva->buf_addr + pkt_kva->data_off - kni->mbuf_va
433 memcpy(data_kva, skb->data, len);
434 if (unlikely(len < ETH_ZLEN)) {
435 memset(data_kva + len, 0, ETH_ZLEN - len);
438 pkt_kva->pkt_len = len;
439 pkt_kva->data_len = len;
441 /* enqueue mbuf into tx_q */
442 ret = kni_fifo_put(kni->tx_q, (void **)&pkt_va, 1);
443 if (unlikely(ret != 1)) {
444 /* Failing should not happen */
445 KNI_ERR("Fail to enqueue mbuf into tx_q\n");
449 /* Failing should not happen */
450 KNI_ERR("Fail to dequeue mbuf from alloc_q\n");
454 /* Free skb and update statistics */
456 kni->stats.tx_bytes += len;
457 kni->stats.tx_packets++;
462 /* Free skb and update statistics */
464 kni->stats.tx_dropped++;
471 * Deal with a transmit timeout.
474 kni_net_tx_timeout (struct net_device *dev)
476 struct kni_dev *kni = netdev_priv(dev);
478 KNI_DBG("Transmit timeout at %ld, latency %ld\n", jiffies,
479 jiffies - dev->trans_start);
481 kni->stats.tx_errors++;
482 netif_wake_queue(dev);
490 kni_net_ioctl(struct net_device *dev, struct ifreq *rq, int cmd)
492 KNI_DBG("kni_net_ioctl %d\n",
493 ((struct kni_dev *)netdev_priv(dev))->group_id);
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
608 kni_net_rebuild_header(struct sk_buff *skb)
610 struct net_device *dev = skb->dev;
611 struct ethhdr *eth = (struct ethhdr *) skb->data;
613 memcpy(eth->h_source, dev->dev_addr, dev->addr_len);
614 memcpy(eth->h_dest, dev->dev_addr, dev->addr_len);
620 * kni_net_set_mac - Change the Ethernet Address of the KNI NIC
621 * @netdev: network interface device structure
622 * @p: pointer to an address structure
624 * Returns 0 on success, negative on failure
626 static int kni_net_set_mac(struct net_device *netdev, void *p)
628 struct sockaddr *addr = p;
629 if (!is_valid_ether_addr((unsigned char *)(addr->sa_data)))
630 return -EADDRNOTAVAIL;
631 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
635 static const struct header_ops kni_net_header_ops = {
636 .create = kni_net_header,
637 .rebuild = kni_net_rebuild_header,
638 .cache = NULL, /* disable caching */
641 static const struct net_device_ops kni_net_netdev_ops = {
642 .ndo_open = kni_net_open,
643 .ndo_stop = kni_net_release,
644 .ndo_set_config = kni_net_config,
645 .ndo_start_xmit = kni_net_tx,
646 .ndo_change_mtu = kni_net_change_mtu,
647 .ndo_do_ioctl = kni_net_ioctl,
648 .ndo_get_stats = kni_net_stats,
649 .ndo_tx_timeout = kni_net_tx_timeout,
650 .ndo_set_mac_address = kni_net_set_mac,
654 kni_net_init(struct net_device *dev)
656 struct kni_dev *kni = netdev_priv(dev);
658 KNI_DBG("kni_net_init\n");
660 init_waitqueue_head(&kni->wq);
661 mutex_init(&kni->sync_lock);
663 ether_setup(dev); /* assign some of the fields */
664 dev->netdev_ops = &kni_net_netdev_ops;
665 dev->header_ops = &kni_net_header_ops;
666 dev->watchdog_timeo = WD_TIMEOUT;
670 kni_net_config_lo_mode(char *lo_str)
673 KNI_PRINT("loopback disabled");
677 if (!strcmp(lo_str, "lo_mode_none"))
678 KNI_PRINT("loopback disabled");
679 else if (!strcmp(lo_str, "lo_mode_fifo")) {
680 KNI_PRINT("loopback mode=lo_mode_fifo enabled");
681 kni_net_rx_func = kni_net_rx_lo_fifo;
682 } else if (!strcmp(lo_str, "lo_mode_fifo_skb")) {
683 KNI_PRINT("loopback mode=lo_mode_fifo_skb enabled");
684 kni_net_rx_func = kni_net_rx_lo_fifo_skb;
686 KNI_PRINT("Incognizant parameter, loopback disabled");