1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright(c) 2010-2014 Intel Corporation.
7 * This code is inspired from the book "Linux Device Drivers" by
8 * Alessandro Rubini and Jonathan Corbet, published by O'Reilly & Associates
11 #include <linux/device.h>
12 #include <linux/module.h>
13 #include <linux/version.h>
14 #include <linux/netdevice.h>
15 #include <linux/etherdevice.h> /* eth_type_trans */
16 #include <linux/skbuff.h>
17 #include <linux/kthread.h>
18 #include <linux/delay.h>
20 #include <exec-env/rte_kni_common.h>
26 #define WD_TIMEOUT 5 /*jiffies */
28 #define KNI_WAIT_RESPONSE_TIMEOUT 300 /* 3 seconds */
30 /* typedef for rx function */
31 typedef void (*kni_net_rx_t)(struct kni_dev *kni);
33 static void kni_net_rx_normal(struct kni_dev *kni);
35 /* kni rx function pointer, with default to normal rx */
36 static kni_net_rx_t kni_net_rx_func = kni_net_rx_normal;
38 /* physical address to kernel virtual address */
42 return phys_to_virt((unsigned long)pa);
45 /* physical address to virtual address */
47 pa2va(void *pa, struct rte_kni_mbuf *m)
51 va = (void *)((unsigned long)pa +
52 (unsigned long)m->buf_addr -
53 (unsigned long)m->buf_physaddr);
57 /* mbuf data kernel virtual address from mbuf kernel virtual address */
59 kva2data_kva(struct rte_kni_mbuf *m)
61 return phys_to_virt(m->buf_physaddr + m->data_off);
64 /* virtual address to physical address */
66 va2pa(void *va, struct rte_kni_mbuf *m)
70 pa = (void *)((unsigned long)va -
71 ((unsigned long)m->buf_addr -
72 (unsigned long)m->buf_physaddr));
77 * It can be called to process the request.
80 kni_net_process_request(struct kni_dev *kni, struct rte_kni_request *req)
88 pr_err("No kni instance or request\n");
92 mutex_lock(&kni->sync_lock);
95 memcpy(kni->sync_kva, req, sizeof(struct rte_kni_request));
96 num = kni_fifo_put(kni->req_q, &kni->sync_va, 1);
98 pr_err("Cannot send to req_q\n");
103 ret_val = wait_event_interruptible_timeout(kni->wq,
104 kni_fifo_count(kni->resp_q), 3 * HZ);
105 if (signal_pending(current) || ret_val <= 0) {
109 num = kni_fifo_get(kni->resp_q, (void **)&resp_va, 1);
110 if (num != 1 || resp_va != kni->sync_va) {
111 /* This should never happen */
112 pr_err("No data in resp_q\n");
117 memcpy(req, kni->sync_kva, sizeof(struct rte_kni_request));
121 mutex_unlock(&kni->sync_lock);
129 kni_net_open(struct net_device *dev)
132 struct rte_kni_request req;
133 struct kni_dev *kni = netdev_priv(dev);
135 netif_start_queue(dev);
137 memset(&req, 0, sizeof(req));
138 req.req_id = RTE_KNI_REQ_CFG_NETWORK_IF;
140 /* Setting if_up to non-zero means up */
142 ret = kni_net_process_request(kni, &req);
144 return (ret == 0) ? req.result : ret;
148 kni_net_release(struct net_device *dev)
151 struct rte_kni_request req;
152 struct kni_dev *kni = netdev_priv(dev);
154 netif_stop_queue(dev); /* can't transmit any more */
156 memset(&req, 0, sizeof(req));
157 req.req_id = RTE_KNI_REQ_CFG_NETWORK_IF;
159 /* Setting if_up to 0 means down */
161 ret = kni_net_process_request(kni, &req);
163 return (ret == 0) ? req.result : ret;
167 * Configuration changes (passed on by ifconfig)
170 kni_net_config(struct net_device *dev, struct ifmap *map)
172 if (dev->flags & IFF_UP) /* can't act on a running interface */
175 /* ignore other fields */
180 * Transmit a packet (called by the kernel)
183 kni_net_tx(struct sk_buff *skb, struct net_device *dev)
187 struct kni_dev *kni = netdev_priv(dev);
188 struct rte_kni_mbuf *pkt_kva = NULL;
192 /* save the timestamp */
193 #ifdef HAVE_TRANS_START_HELPER
194 netif_trans_update(dev);
196 dev->trans_start = jiffies;
199 /* Check if the length of skb is less than mbuf size */
200 if (skb->len > kni->mbuf_size)
204 * Check if it has at least one free entry in tx_q and
205 * one entry in alloc_q.
207 if (kni_fifo_free_count(kni->tx_q) == 0 ||
208 kni_fifo_count(kni->alloc_q) == 0) {
210 * If no free entry in tx_q or no entry in alloc_q,
211 * drops skb and goes out.
216 /* dequeue a mbuf from alloc_q */
217 ret = kni_fifo_get(kni->alloc_q, &pkt_pa, 1);
218 if (likely(ret == 1)) {
221 pkt_kva = pa2kva(pkt_pa);
222 data_kva = kva2data_kva(pkt_kva);
223 pkt_va = pa2va(pkt_pa, pkt_kva);
226 memcpy(data_kva, skb->data, len);
227 if (unlikely(len < ETH_ZLEN)) {
228 memset(data_kva + len, 0, ETH_ZLEN - len);
231 pkt_kva->pkt_len = len;
232 pkt_kva->data_len = len;
234 /* enqueue mbuf into tx_q */
235 ret = kni_fifo_put(kni->tx_q, &pkt_va, 1);
236 if (unlikely(ret != 1)) {
237 /* Failing should not happen */
238 pr_err("Fail to enqueue mbuf into tx_q\n");
242 /* Failing should not happen */
243 pr_err("Fail to dequeue mbuf from alloc_q\n");
247 /* Free skb and update statistics */
249 kni->stats.tx_bytes += len;
250 kni->stats.tx_packets++;
255 /* Free skb and update statistics */
257 kni->stats.tx_dropped++;
263 * RX: normal working mode
266 kni_net_rx_normal(struct kni_dev *kni)
270 uint32_t i, num_rx, num_fq;
271 struct rte_kni_mbuf *kva;
274 struct net_device *dev = kni->net_dev;
276 /* Get the number of free entries in free_q */
277 num_fq = kni_fifo_free_count(kni->free_q);
279 /* No room on the free_q, bail out */
283 /* Calculate the number of entries to dequeue from rx_q */
284 num_rx = min_t(uint32_t, num_fq, MBUF_BURST_SZ);
286 /* Burst dequeue from rx_q */
287 num_rx = kni_fifo_get(kni->rx_q, kni->pa, num_rx);
291 /* Transfer received packets to netif */
292 for (i = 0; i < num_rx; i++) {
293 kva = pa2kva(kni->pa[i]);
295 data_kva = kva2data_kva(kva);
296 kni->va[i] = pa2va(kni->pa[i], kva);
298 skb = dev_alloc_skb(len + 2);
300 /* Update statistics */
301 kni->stats.rx_dropped++;
305 /* Align IP on 16B boundary */
308 if (kva->nb_segs == 1) {
309 memcpy(skb_put(skb, len), data_kva, len);
312 int kva_nb_segs = kva->nb_segs;
314 for (nb_segs = 0; nb_segs < kva_nb_segs; nb_segs++) {
315 memcpy(skb_put(skb, kva->data_len),
316 data_kva, kva->data_len);
321 kva = pa2kva(va2pa(kva->next, kva));
322 data_kva = kva2data_kva(kva);
327 skb->protocol = eth_type_trans(skb, dev);
328 skb->ip_summed = CHECKSUM_UNNECESSARY;
330 /* Call netif interface */
333 /* Update statistics */
334 kni->stats.rx_bytes += len;
335 kni->stats.rx_packets++;
338 /* Burst enqueue mbufs into free_q */
339 ret = kni_fifo_put(kni->free_q, kni->va, num_rx);
341 /* Failing should not happen */
342 pr_err("Fail to enqueue entries into free_q\n");
346 * RX: loopback with enqueue/dequeue fifos.
349 kni_net_rx_lo_fifo(struct kni_dev *kni)
353 uint32_t i, num, num_rq, num_tq, num_aq, num_fq;
354 struct rte_kni_mbuf *kva;
356 struct rte_kni_mbuf *alloc_kva;
357 void *alloc_data_kva;
359 /* Get the number of entries in rx_q */
360 num_rq = kni_fifo_count(kni->rx_q);
362 /* Get the number of free entrie in tx_q */
363 num_tq = kni_fifo_free_count(kni->tx_q);
365 /* Get the number of entries in alloc_q */
366 num_aq = kni_fifo_count(kni->alloc_q);
368 /* Get the number of free entries in free_q */
369 num_fq = kni_fifo_free_count(kni->free_q);
371 /* Calculate the number of entries to be dequeued from rx_q */
372 num = min(num_rq, num_tq);
373 num = min(num, num_aq);
374 num = min(num, num_fq);
375 num = min_t(uint32_t, num, MBUF_BURST_SZ);
377 /* Return if no entry to dequeue from rx_q */
381 /* Burst dequeue from rx_q */
382 ret = kni_fifo_get(kni->rx_q, kni->pa, num);
384 return; /* Failing should not happen */
386 /* Dequeue entries from alloc_q */
387 ret = kni_fifo_get(kni->alloc_q, kni->alloc_pa, num);
391 for (i = 0; i < num; i++) {
392 kva = pa2kva(kni->pa[i]);
394 data_kva = kva2data_kva(kva);
395 kni->va[i] = pa2va(kni->pa[i], kva);
397 alloc_kva = pa2kva(kni->alloc_pa[i]);
398 alloc_data_kva = kva2data_kva(alloc_kva);
399 kni->alloc_va[i] = pa2va(kni->alloc_pa[i], alloc_kva);
401 memcpy(alloc_data_kva, data_kva, len);
402 alloc_kva->pkt_len = len;
403 alloc_kva->data_len = len;
405 kni->stats.tx_bytes += len;
406 kni->stats.rx_bytes += len;
409 /* Burst enqueue mbufs into tx_q */
410 ret = kni_fifo_put(kni->tx_q, kni->alloc_va, num);
412 /* Failing should not happen */
413 pr_err("Fail to enqueue mbufs into tx_q\n");
416 /* Burst enqueue mbufs into free_q */
417 ret = kni_fifo_put(kni->free_q, kni->va, num);
419 /* Failing should not happen */
420 pr_err("Fail to enqueue mbufs into free_q\n");
423 * Update statistic, and enqueue/dequeue failure is impossible,
424 * as all queues are checked at first.
426 kni->stats.tx_packets += num;
427 kni->stats.rx_packets += num;
431 * RX: loopback with enqueue/dequeue fifos and sk buffer copies.
434 kni_net_rx_lo_fifo_skb(struct kni_dev *kni)
438 uint32_t i, num_rq, num_fq, num;
439 struct rte_kni_mbuf *kva;
442 struct net_device *dev = kni->net_dev;
444 /* Get the number of entries in rx_q */
445 num_rq = kni_fifo_count(kni->rx_q);
447 /* Get the number of free entries in free_q */
448 num_fq = kni_fifo_free_count(kni->free_q);
450 /* Calculate the number of entries to dequeue from rx_q */
451 num = min(num_rq, num_fq);
452 num = min_t(uint32_t, num, MBUF_BURST_SZ);
454 /* Return if no entry to dequeue from rx_q */
458 /* Burst dequeue mbufs from rx_q */
459 ret = kni_fifo_get(kni->rx_q, kni->pa, num);
463 /* Copy mbufs to sk buffer and then call tx interface */
464 for (i = 0; i < num; i++) {
465 kva = pa2kva(kni->pa[i]);
467 data_kva = kva2data_kva(kva);
468 kni->va[i] = pa2va(kni->pa[i], kva);
470 skb = dev_alloc_skb(len + 2);
472 /* Align IP on 16B boundary */
474 memcpy(skb_put(skb, len), data_kva, len);
476 skb->ip_summed = CHECKSUM_UNNECESSARY;
480 /* Simulate real usage, allocate/copy skb twice */
481 skb = dev_alloc_skb(len + 2);
483 kni->stats.rx_dropped++;
487 /* Align IP on 16B boundary */
490 if (kva->nb_segs == 1) {
491 memcpy(skb_put(skb, len), data_kva, len);
494 int kva_nb_segs = kva->nb_segs;
496 for (nb_segs = 0; nb_segs < kva_nb_segs; nb_segs++) {
497 memcpy(skb_put(skb, kva->data_len),
498 data_kva, kva->data_len);
503 kva = pa2kva(va2pa(kva->next, kva));
504 data_kva = kva2data_kva(kva);
509 skb->ip_summed = CHECKSUM_UNNECESSARY;
511 kni->stats.rx_bytes += len;
512 kni->stats.rx_packets++;
514 /* call tx interface */
515 kni_net_tx(skb, dev);
518 /* enqueue all the mbufs from rx_q into free_q */
519 ret = kni_fifo_put(kni->free_q, kni->va, num);
521 /* Failing should not happen */
522 pr_err("Fail to enqueue mbufs into free_q\n");
527 kni_net_rx(struct kni_dev *kni)
530 * It doesn't need to check if it is NULL pointer,
531 * as it has a default value
533 (*kni_net_rx_func)(kni);
537 * Deal with a transmit timeout.
540 kni_net_tx_timeout(struct net_device *dev)
542 struct kni_dev *kni = netdev_priv(dev);
544 pr_debug("Transmit timeout at %ld, latency %ld\n", jiffies,
545 jiffies - dev_trans_start(dev));
547 kni->stats.tx_errors++;
548 netif_wake_queue(dev);
555 kni_net_ioctl(struct net_device *dev, struct ifreq *rq, int cmd)
557 pr_debug("kni_net_ioctl group:%d cmd:%d\n",
558 ((struct kni_dev *)netdev_priv(dev))->group_id, cmd);
564 kni_net_set_rx_mode(struct net_device *dev)
569 kni_net_change_mtu(struct net_device *dev, int new_mtu)
572 struct rte_kni_request req;
573 struct kni_dev *kni = netdev_priv(dev);
575 pr_debug("kni_net_change_mtu new mtu %d to be set\n", new_mtu);
577 memset(&req, 0, sizeof(req));
578 req.req_id = RTE_KNI_REQ_CHANGE_MTU;
579 req.new_mtu = new_mtu;
580 ret = kni_net_process_request(kni, &req);
581 if (ret == 0 && req.result == 0)
584 return (ret == 0) ? req.result : ret;
588 kni_net_set_promiscusity(struct net_device *netdev, int flags)
590 struct rte_kni_request req;
591 struct kni_dev *kni = netdev_priv(netdev);
593 memset(&req, 0, sizeof(req));
594 req.req_id = RTE_KNI_REQ_CHANGE_PROMISC;
596 if (netdev->flags & IFF_PROMISC)
597 req.promiscusity = 1;
599 req.promiscusity = 0;
600 kni_net_process_request(kni, &req);
604 * Checks if the user space application provided the resp message
607 kni_net_poll_resp(struct kni_dev *kni)
609 if (kni_fifo_count(kni->resp_q))
610 wake_up_interruptible(&kni->wq);
614 * Return statistics to the caller
616 static struct net_device_stats *
617 kni_net_stats(struct net_device *dev)
619 struct kni_dev *kni = netdev_priv(dev);
625 * Fill the eth header
628 kni_net_header(struct sk_buff *skb, struct net_device *dev,
629 unsigned short type, const void *daddr,
630 const void *saddr, uint32_t len)
632 struct ethhdr *eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
634 memcpy(eth->h_source, saddr ? saddr : dev->dev_addr, dev->addr_len);
635 memcpy(eth->h_dest, daddr ? daddr : dev->dev_addr, dev->addr_len);
636 eth->h_proto = htons(type);
638 return dev->hard_header_len;
642 * Re-fill the eth header
644 #ifdef HAVE_REBUILD_HEADER
646 kni_net_rebuild_header(struct sk_buff *skb)
648 struct net_device *dev = skb->dev;
649 struct ethhdr *eth = (struct ethhdr *) skb->data;
651 memcpy(eth->h_source, dev->dev_addr, dev->addr_len);
652 memcpy(eth->h_dest, dev->dev_addr, dev->addr_len);
659 * kni_net_set_mac - Change the Ethernet Address of the KNI NIC
660 * @netdev: network interface device structure
661 * @p: pointer to an address structure
663 * Returns 0 on success, negative on failure
666 kni_net_set_mac(struct net_device *netdev, void *p)
669 struct rte_kni_request req;
671 struct sockaddr *addr = p;
673 memset(&req, 0, sizeof(req));
674 req.req_id = RTE_KNI_REQ_CHANGE_MAC_ADDR;
676 if (!is_valid_ether_addr((unsigned char *)(addr->sa_data)))
677 return -EADDRNOTAVAIL;
679 memcpy(req.mac_addr, addr->sa_data, netdev->addr_len);
680 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
682 kni = netdev_priv(netdev);
683 ret = kni_net_process_request(kni, &req);
685 return (ret == 0 ? req.result : ret);
688 #ifdef HAVE_CHANGE_CARRIER_CB
690 kni_net_change_carrier(struct net_device *dev, bool new_carrier)
693 netif_carrier_on(dev);
695 netif_carrier_off(dev);
700 static const struct header_ops kni_net_header_ops = {
701 .create = kni_net_header,
702 #ifdef HAVE_REBUILD_HEADER
703 .rebuild = kni_net_rebuild_header,
705 .cache = NULL, /* disable caching */
708 static const struct net_device_ops kni_net_netdev_ops = {
709 .ndo_open = kni_net_open,
710 .ndo_stop = kni_net_release,
711 .ndo_set_config = kni_net_config,
712 .ndo_change_rx_flags = kni_net_set_promiscusity,
713 .ndo_start_xmit = kni_net_tx,
714 .ndo_change_mtu = kni_net_change_mtu,
715 .ndo_do_ioctl = kni_net_ioctl,
716 .ndo_set_rx_mode = kni_net_set_rx_mode,
717 .ndo_get_stats = kni_net_stats,
718 .ndo_tx_timeout = kni_net_tx_timeout,
719 .ndo_set_mac_address = kni_net_set_mac,
720 #ifdef HAVE_CHANGE_CARRIER_CB
721 .ndo_change_carrier = kni_net_change_carrier,
726 kni_net_init(struct net_device *dev)
728 struct kni_dev *kni = netdev_priv(dev);
730 init_waitqueue_head(&kni->wq);
731 mutex_init(&kni->sync_lock);
733 ether_setup(dev); /* assign some of the fields */
734 dev->netdev_ops = &kni_net_netdev_ops;
735 dev->header_ops = &kni_net_header_ops;
736 dev->watchdog_timeo = WD_TIMEOUT;
740 kni_net_config_lo_mode(char *lo_str)
743 pr_debug("loopback disabled");
747 if (!strcmp(lo_str, "lo_mode_none"))
748 pr_debug("loopback disabled");
749 else if (!strcmp(lo_str, "lo_mode_fifo")) {
750 pr_debug("loopback mode=lo_mode_fifo enabled");
751 kni_net_rx_func = kni_net_rx_lo_fifo;
752 } else if (!strcmp(lo_str, "lo_mode_fifo_skb")) {
753 pr_debug("loopback mode=lo_mode_fifo_skb enabled");
754 kni_net_rx_func = kni_net_rx_lo_fifo_skb;
756 pr_debug("Incognizant parameter, loopback disabled");