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);
72 netif_start_queue(dev);
74 memset(&req, 0, sizeof(req));
75 req.req_id = RTE_KNI_REQ_CFG_NETWORK_IF;
77 /* Setting if_up to non-zero means up */
79 ret = kni_net_process_request(kni, &req);
81 return (ret == 0) ? req.result : ret;
85 kni_net_release(struct net_device *dev)
88 struct rte_kni_request req;
89 struct kni_dev *kni = netdev_priv(dev);
91 netif_stop_queue(dev); /* can't transmit any more */
93 memset(&req, 0, sizeof(req));
94 req.req_id = RTE_KNI_REQ_CFG_NETWORK_IF;
96 /* Setting if_up to 0 means down */
98 ret = kni_net_process_request(kni, &req);
100 return (ret == 0) ? req.result : ret;
104 * Configuration changes (passed on by ifconfig)
107 kni_net_config(struct net_device *dev, struct ifmap *map)
109 if (dev->flags & IFF_UP) /* can't act on a running interface */
112 /* ignore other fields */
117 * RX: normal working mode
120 kni_net_rx_normal(struct kni_dev *kni)
124 unsigned i, num_rx, num_fq;
125 struct rte_kni_mbuf *kva;
126 struct rte_kni_mbuf *va[MBUF_BURST_SZ];
130 struct net_device *dev = kni->net_dev;
132 /* Get the number of free entries in free_q */
133 num_fq = kni_fifo_free_count(kni->free_q);
135 /* No room on the free_q, bail out */
139 /* Calculate the number of entries to dequeue from rx_q */
140 num_rx = min(num_fq, (unsigned)MBUF_BURST_SZ);
142 /* Burst dequeue from rx_q */
143 num_rx = kni_fifo_get(kni->rx_q, (void **)va, num_rx);
147 /* Transfer received packets to netif */
148 for (i = 0; i < num_rx; i++) {
149 kva = (void *)va[i] - kni->mbuf_va + kni->mbuf_kva;
152 data_kva = kva->buf_addr + kva->data_off - kni->mbuf_va
155 skb = dev_alloc_skb(len + 2);
157 KNI_ERR("Out of mem, dropping pkts\n");
158 /* Update statistics */
159 kni->stats.rx_dropped++;
163 /* Align IP on 16B boundary */
166 if (kva->nb_segs == 1) {
167 memcpy(skb_put(skb, len), data_kva, len);
170 int kva_nb_segs = kva->nb_segs;
172 for (nb_segs = 0; nb_segs < kva_nb_segs; nb_segs++) {
173 memcpy(skb_put(skb, kva->data_len),
174 data_kva, kva->data_len);
179 kva = kva->next - kni->mbuf_va + kni->mbuf_kva;
180 data_kva = kva->buf_addr + kva->data_off
181 - kni->mbuf_va + kni->mbuf_kva;
186 skb->protocol = eth_type_trans(skb, dev);
187 skb->ip_summed = CHECKSUM_UNNECESSARY;
189 /* Call netif interface */
192 /* Update statistics */
193 kni->stats.rx_bytes += len;
194 kni->stats.rx_packets++;
197 /* Burst enqueue mbufs into free_q */
198 ret = kni_fifo_put(kni->free_q, (void **)va, num_rx);
200 /* Failing should not happen */
201 KNI_ERR("Fail to enqueue entries into free_q\n");
205 * RX: loopback with enqueue/dequeue fifos.
208 kni_net_rx_lo_fifo(struct kni_dev *kni)
212 unsigned i, num, num_rq, num_tq, num_aq, num_fq;
213 struct rte_kni_mbuf *kva;
214 struct rte_kni_mbuf *va[MBUF_BURST_SZ];
217 struct rte_kni_mbuf *alloc_kva;
218 struct rte_kni_mbuf *alloc_va[MBUF_BURST_SZ];
219 void *alloc_data_kva;
221 /* Get the number of entries in rx_q */
222 num_rq = kni_fifo_count(kni->rx_q);
224 /* Get the number of free entrie in tx_q */
225 num_tq = kni_fifo_free_count(kni->tx_q);
227 /* Get the number of entries in alloc_q */
228 num_aq = kni_fifo_count(kni->alloc_q);
230 /* Get the number of free entries in free_q */
231 num_fq = kni_fifo_free_count(kni->free_q);
233 /* Calculate the number of entries to be dequeued from rx_q */
234 num = min(num_rq, num_tq);
235 num = min(num, num_aq);
236 num = min(num, num_fq);
237 num = min(num, (unsigned)MBUF_BURST_SZ);
239 /* Return if no entry to dequeue from rx_q */
243 /* Burst dequeue from rx_q */
244 ret = kni_fifo_get(kni->rx_q, (void **)va, num);
246 return; /* Failing should not happen */
248 /* Dequeue entries from alloc_q */
249 ret = kni_fifo_get(kni->alloc_q, (void **)alloc_va, num);
253 for (i = 0; i < num; i++) {
254 kva = (void *)va[i] - kni->mbuf_va + kni->mbuf_kva;
256 data_kva = kva->buf_addr + kva->data_off -
257 kni->mbuf_va + kni->mbuf_kva;
259 alloc_kva = (void *)alloc_va[i] - kni->mbuf_va +
261 alloc_data_kva = alloc_kva->buf_addr +
262 alloc_kva->data_off - kni->mbuf_va +
264 memcpy(alloc_data_kva, data_kva, len);
265 alloc_kva->pkt_len = len;
266 alloc_kva->data_len = len;
268 kni->stats.tx_bytes += len;
269 kni->stats.rx_bytes += len;
272 /* Burst enqueue mbufs into tx_q */
273 ret = kni_fifo_put(kni->tx_q, (void **)alloc_va, num);
275 /* Failing should not happen */
276 KNI_ERR("Fail to enqueue mbufs into tx_q\n");
279 /* Burst enqueue mbufs into free_q */
280 ret = kni_fifo_put(kni->free_q, (void **)va, num);
282 /* Failing should not happen */
283 KNI_ERR("Fail to enqueue mbufs into free_q\n");
286 * Update statistic, and enqueue/dequeue failure is impossible,
287 * as all queues are checked at first.
289 kni->stats.tx_packets += num;
290 kni->stats.rx_packets += num;
294 * RX: loopback with enqueue/dequeue fifos and sk buffer copies.
297 kni_net_rx_lo_fifo_skb(struct kni_dev *kni)
301 unsigned i, num_rq, num_fq, num;
302 struct rte_kni_mbuf *kva;
303 struct rte_kni_mbuf *va[MBUF_BURST_SZ];
307 struct net_device *dev = kni->net_dev;
309 /* Get the number of entries in rx_q */
310 num_rq = kni_fifo_count(kni->rx_q);
312 /* Get the number of free entries in free_q */
313 num_fq = kni_fifo_free_count(kni->free_q);
315 /* Calculate the number of entries to dequeue from rx_q */
316 num = min(num_rq, num_fq);
317 num = min(num, (unsigned)MBUF_BURST_SZ);
319 /* Return if no entry to dequeue from rx_q */
323 /* Burst dequeue mbufs from rx_q */
324 ret = kni_fifo_get(kni->rx_q, (void **)va, num);
328 /* Copy mbufs to sk buffer and then call tx interface */
329 for (i = 0; i < num; i++) {
330 kva = (void *)va[i] - kni->mbuf_va + kni->mbuf_kva;
332 data_kva = kva->buf_addr + kva->data_off - kni->mbuf_va +
335 skb = dev_alloc_skb(len + 2);
337 KNI_ERR("Out of mem, dropping pkts\n");
339 /* Align IP on 16B boundary */
341 memcpy(skb_put(skb, len), data_kva, len);
343 skb->ip_summed = CHECKSUM_UNNECESSARY;
347 /* Simulate real usage, allocate/copy skb twice */
348 skb = dev_alloc_skb(len + 2);
350 KNI_ERR("Out of mem, dropping pkts\n");
351 kni->stats.rx_dropped++;
355 /* Align IP on 16B boundary */
358 if (kva->nb_segs == 1) {
359 memcpy(skb_put(skb, len), data_kva, len);
362 int kva_nb_segs = kva->nb_segs;
364 for (nb_segs = 0; nb_segs < kva_nb_segs; nb_segs++) {
365 memcpy(skb_put(skb, kva->data_len),
366 data_kva, kva->data_len);
371 kva = kva->next - kni->mbuf_va + kni->mbuf_kva;
372 data_kva = kva->buf_addr + kva->data_off
373 - kni->mbuf_va + kni->mbuf_kva;
378 skb->ip_summed = CHECKSUM_UNNECESSARY;
380 kni->stats.rx_bytes += len;
381 kni->stats.rx_packets++;
383 /* call tx interface */
384 kni_net_tx(skb, dev);
387 /* enqueue all the mbufs from rx_q into free_q */
388 ret = kni_fifo_put(kni->free_q, (void **)&va, num);
390 /* Failing should not happen */
391 KNI_ERR("Fail to enqueue mbufs into free_q\n");
396 kni_net_rx(struct kni_dev *kni)
399 * It doesn't need to check if it is NULL pointer,
400 * as it has a default value
402 (*kni_net_rx_func)(kni);
406 * Transmit a packet (called by the kernel)
410 kni_net_tx(struct sk_buff *skb, struct net_device *dev)
412 struct kni_dev *kni = netdev_priv(dev);
415 kni->stats.tx_dropped++;
421 kni_net_tx(struct sk_buff *skb, struct net_device *dev)
425 struct kni_dev *kni = netdev_priv(dev);
426 struct rte_kni_mbuf *pkt_kva = NULL;
427 struct rte_kni_mbuf *pkt_va = NULL;
429 dev->trans_start = jiffies; /* save the timestamp */
431 /* Check if the length of skb is less than mbuf size */
432 if (skb->len > kni->mbuf_size)
436 * Check if it has at least one free entry in tx_q and
437 * one entry in alloc_q.
439 if (kni_fifo_free_count(kni->tx_q) == 0 ||
440 kni_fifo_count(kni->alloc_q) == 0) {
442 * If no free entry in tx_q or no entry in alloc_q,
443 * drops skb and goes out.
448 /* dequeue a mbuf from alloc_q */
449 ret = kni_fifo_get(kni->alloc_q, (void **)&pkt_va, 1);
450 if (likely(ret == 1)) {
453 pkt_kva = (void *)pkt_va - kni->mbuf_va + kni->mbuf_kva;
454 data_kva = pkt_kva->buf_addr + pkt_kva->data_off - kni->mbuf_va
458 memcpy(data_kva, skb->data, len);
459 if (unlikely(len < ETH_ZLEN)) {
460 memset(data_kva + len, 0, ETH_ZLEN - len);
463 pkt_kva->pkt_len = len;
464 pkt_kva->data_len = len;
466 /* enqueue mbuf into tx_q */
467 ret = kni_fifo_put(kni->tx_q, (void **)&pkt_va, 1);
468 if (unlikely(ret != 1)) {
469 /* Failing should not happen */
470 KNI_ERR("Fail to enqueue mbuf into tx_q\n");
474 /* Failing should not happen */
475 KNI_ERR("Fail to dequeue mbuf from alloc_q\n");
479 /* Free skb and update statistics */
481 kni->stats.tx_bytes += len;
482 kni->stats.tx_packets++;
487 /* Free skb and update statistics */
489 kni->stats.tx_dropped++;
496 * Deal with a transmit timeout.
499 kni_net_tx_timeout (struct net_device *dev)
501 struct kni_dev *kni = netdev_priv(dev);
503 KNI_DBG("Transmit timeout at %ld, latency %ld\n", jiffies,
504 jiffies - dev->trans_start);
506 kni->stats.tx_errors++;
507 netif_wake_queue(dev);
515 kni_net_ioctl(struct net_device *dev, struct ifreq *rq, int cmd)
517 KNI_DBG("kni_net_ioctl %d\n",
518 ((struct kni_dev *)netdev_priv(dev))->group_id);
524 kni_net_set_rx_mode(struct net_device *dev)
529 kni_net_change_mtu(struct net_device *dev, int new_mtu)
532 struct rte_kni_request req;
533 struct kni_dev *kni = netdev_priv(dev);
535 KNI_DBG("kni_net_change_mtu new mtu %d to be set\n", new_mtu);
537 memset(&req, 0, sizeof(req));
538 req.req_id = RTE_KNI_REQ_CHANGE_MTU;
539 req.new_mtu = new_mtu;
540 ret = kni_net_process_request(kni, &req);
541 if (ret == 0 && req.result == 0)
544 return (ret == 0) ? req.result : ret;
548 * Checks if the user space application provided the resp message
551 kni_net_poll_resp(struct kni_dev *kni)
553 if (kni_fifo_count(kni->resp_q))
554 wake_up_interruptible(&kni->wq);
558 * It can be called to process the request.
561 kni_net_process_request(struct kni_dev *kni, struct rte_kni_request *req)
569 KNI_ERR("No kni instance or request\n");
573 mutex_lock(&kni->sync_lock);
576 memcpy(kni->sync_kva, req, sizeof(struct rte_kni_request));
577 num = kni_fifo_put(kni->req_q, &kni->sync_va, 1);
579 KNI_ERR("Cannot send to req_q\n");
584 ret_val = wait_event_interruptible_timeout(kni->wq,
585 kni_fifo_count(kni->resp_q), 3 * HZ);
586 if (signal_pending(current) || ret_val <= 0) {
590 num = kni_fifo_get(kni->resp_q, (void **)&resp_va, 1);
591 if (num != 1 || resp_va != kni->sync_va) {
592 /* This should never happen */
593 KNI_ERR("No data in resp_q\n");
598 memcpy(req, kni->sync_kva, sizeof(struct rte_kni_request));
602 mutex_unlock(&kni->sync_lock);
607 * Return statistics to the caller
609 static struct net_device_stats *
610 kni_net_stats(struct net_device *dev)
612 struct kni_dev *kni = netdev_priv(dev);
617 * Fill the eth header
620 kni_net_header(struct sk_buff *skb, struct net_device *dev,
621 unsigned short type, const void *daddr,
622 const void *saddr, unsigned int len)
624 struct ethhdr *eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
626 memcpy(eth->h_source, saddr ? saddr : dev->dev_addr, dev->addr_len);
627 memcpy(eth->h_dest, daddr ? daddr : dev->dev_addr, dev->addr_len);
628 eth->h_proto = htons(type);
630 return dev->hard_header_len;
635 * Re-fill the eth header
637 #if (LINUX_VERSION_CODE < KERNEL_VERSION(4, 1, 0))
639 kni_net_rebuild_header(struct sk_buff *skb)
641 struct net_device *dev = skb->dev;
642 struct ethhdr *eth = (struct ethhdr *) skb->data;
644 memcpy(eth->h_source, dev->dev_addr, dev->addr_len);
645 memcpy(eth->h_dest, dev->dev_addr, dev->addr_len);
652 * kni_net_set_mac - Change the Ethernet Address of the KNI NIC
653 * @netdev: network interface device structure
654 * @p: pointer to an address structure
656 * Returns 0 on success, negative on failure
658 static int kni_net_set_mac(struct net_device *netdev, void *p)
660 struct sockaddr *addr = p;
661 if (!is_valid_ether_addr((unsigned char *)(addr->sa_data)))
662 return -EADDRNOTAVAIL;
663 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
667 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(3, 9, 0))
668 static int kni_net_change_carrier(struct net_device *dev, bool new_carrier)
671 netif_carrier_on(dev);
673 netif_carrier_off(dev);
678 static const struct header_ops kni_net_header_ops = {
679 .create = kni_net_header,
680 #if (LINUX_VERSION_CODE < KERNEL_VERSION(4, 1, 0))
681 .rebuild = kni_net_rebuild_header,
683 .cache = NULL, /* disable caching */
686 static const struct net_device_ops kni_net_netdev_ops = {
687 .ndo_open = kni_net_open,
688 .ndo_stop = kni_net_release,
689 .ndo_set_config = kni_net_config,
690 .ndo_start_xmit = kni_net_tx,
691 .ndo_change_mtu = kni_net_change_mtu,
692 .ndo_do_ioctl = kni_net_ioctl,
693 .ndo_set_rx_mode = kni_net_set_rx_mode,
694 .ndo_get_stats = kni_net_stats,
695 .ndo_tx_timeout = kni_net_tx_timeout,
696 .ndo_set_mac_address = kni_net_set_mac,
697 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(3, 9, 0))
698 .ndo_change_carrier = kni_net_change_carrier,
703 kni_net_init(struct net_device *dev)
705 struct kni_dev *kni = netdev_priv(dev);
707 KNI_DBG("kni_net_init\n");
709 init_waitqueue_head(&kni->wq);
710 mutex_init(&kni->sync_lock);
712 ether_setup(dev); /* assign some of the fields */
713 dev->netdev_ops = &kni_net_netdev_ops;
714 dev->header_ops = &kni_net_header_ops;
715 dev->watchdog_timeo = WD_TIMEOUT;
719 kni_net_config_lo_mode(char *lo_str)
722 KNI_PRINT("loopback disabled");
726 if (!strcmp(lo_str, "lo_mode_none"))
727 KNI_PRINT("loopback disabled");
728 else if (!strcmp(lo_str, "lo_mode_fifo")) {
729 KNI_PRINT("loopback mode=lo_mode_fifo enabled");
730 kni_net_rx_func = kni_net_rx_lo_fifo;
731 } else if (!strcmp(lo_str, "lo_mode_fifo_skb")) {
732 KNI_PRINT("loopback mode=lo_mode_fifo_skb enabled");
733 kni_net_rx_func = kni_net_rx_lo_fifo_skb;
735 KNI_PRINT("Incognizant parameter, loopback disabled");