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 kni_fifo_trans_pa2va(struct kni_dev *kni,
168 struct rte_kni_fifo *src_pa, struct rte_kni_fifo *dst_va)
170 uint32_t ret, i, num_dst, num_rx;
173 num_dst = kni_fifo_free_count(dst_va);
177 num_rx = min_t(uint32_t, num_dst, MBUF_BURST_SZ);
179 num_rx = kni_fifo_get(src_pa, kni->pa, num_rx);
183 for (i = 0; i < num_rx; i++) {
184 kva = pa2kva(kni->pa[i]);
185 kni->va[i] = pa2va(kni->pa[i], kva);
188 ret = kni_fifo_put(dst_va, kni->va, num_rx);
190 /* Failing should not happen */
191 pr_err("Fail to enqueue entries into dst_va\n");
197 /* Try to release mbufs when kni release */
198 void kni_net_release_fifo_phy(struct kni_dev *kni)
200 /* release rx_q first, because it can't release in userspace */
201 kni_fifo_trans_pa2va(kni, kni->rx_q, kni->free_q);
202 /* release alloc_q for speeding up kni release in userspace */
203 kni_fifo_trans_pa2va(kni, kni->alloc_q, kni->free_q);
207 * Configuration changes (passed on by ifconfig)
210 kni_net_config(struct net_device *dev, struct ifmap *map)
212 if (dev->flags & IFF_UP) /* can't act on a running interface */
215 /* ignore other fields */
220 * Transmit a packet (called by the kernel)
223 kni_net_tx(struct sk_buff *skb, struct net_device *dev)
227 struct kni_dev *kni = netdev_priv(dev);
228 struct rte_kni_mbuf *pkt_kva = NULL;
232 /* save the timestamp */
233 #ifdef HAVE_TRANS_START_HELPER
234 netif_trans_update(dev);
236 dev->trans_start = jiffies;
239 /* Check if the length of skb is less than mbuf size */
240 if (skb->len > kni->mbuf_size)
244 * Check if it has at least one free entry in tx_q and
245 * one entry in alloc_q.
247 if (kni_fifo_free_count(kni->tx_q) == 0 ||
248 kni_fifo_count(kni->alloc_q) == 0) {
250 * If no free entry in tx_q or no entry in alloc_q,
251 * drops skb and goes out.
256 /* dequeue a mbuf from alloc_q */
257 ret = kni_fifo_get(kni->alloc_q, &pkt_pa, 1);
258 if (likely(ret == 1)) {
261 pkt_kva = pa2kva(pkt_pa);
262 data_kva = kva2data_kva(pkt_kva);
263 pkt_va = pa2va(pkt_pa, pkt_kva);
266 memcpy(data_kva, skb->data, len);
267 if (unlikely(len < ETH_ZLEN)) {
268 memset(data_kva + len, 0, ETH_ZLEN - len);
271 pkt_kva->pkt_len = len;
272 pkt_kva->data_len = len;
274 /* enqueue mbuf into tx_q */
275 ret = kni_fifo_put(kni->tx_q, &pkt_va, 1);
276 if (unlikely(ret != 1)) {
277 /* Failing should not happen */
278 pr_err("Fail to enqueue mbuf into tx_q\n");
282 /* Failing should not happen */
283 pr_err("Fail to dequeue mbuf from alloc_q\n");
287 /* Free skb and update statistics */
289 kni->stats.tx_bytes += len;
290 kni->stats.tx_packets++;
295 /* Free skb and update statistics */
297 kni->stats.tx_dropped++;
303 * RX: normal working mode
306 kni_net_rx_normal(struct kni_dev *kni)
310 uint32_t i, num_rx, num_fq;
311 struct rte_kni_mbuf *kva;
314 struct net_device *dev = kni->net_dev;
316 /* Get the number of free entries in free_q */
317 num_fq = kni_fifo_free_count(kni->free_q);
319 /* No room on the free_q, bail out */
323 /* Calculate the number of entries to dequeue from rx_q */
324 num_rx = min_t(uint32_t, num_fq, MBUF_BURST_SZ);
326 /* Burst dequeue from rx_q */
327 num_rx = kni_fifo_get(kni->rx_q, kni->pa, num_rx);
331 /* Transfer received packets to netif */
332 for (i = 0; i < num_rx; i++) {
333 kva = pa2kva(kni->pa[i]);
335 data_kva = kva2data_kva(kva);
336 kni->va[i] = pa2va(kni->pa[i], kva);
338 skb = dev_alloc_skb(len + 2);
340 /* Update statistics */
341 kni->stats.rx_dropped++;
345 /* Align IP on 16B boundary */
348 if (kva->nb_segs == 1) {
349 memcpy(skb_put(skb, len), data_kva, len);
352 int kva_nb_segs = kva->nb_segs;
354 for (nb_segs = 0; nb_segs < kva_nb_segs; nb_segs++) {
355 memcpy(skb_put(skb, kva->data_len),
356 data_kva, kva->data_len);
361 kva = pa2kva(va2pa(kva->next, kva));
362 data_kva = kva2data_kva(kva);
367 skb->protocol = eth_type_trans(skb, dev);
368 skb->ip_summed = CHECKSUM_UNNECESSARY;
370 /* Call netif interface */
373 /* Update statistics */
374 kni->stats.rx_bytes += len;
375 kni->stats.rx_packets++;
378 /* Burst enqueue mbufs into free_q */
379 ret = kni_fifo_put(kni->free_q, kni->va, num_rx);
381 /* Failing should not happen */
382 pr_err("Fail to enqueue entries into free_q\n");
386 * RX: loopback with enqueue/dequeue fifos.
389 kni_net_rx_lo_fifo(struct kni_dev *kni)
393 uint32_t i, num, num_rq, num_tq, num_aq, num_fq;
394 struct rte_kni_mbuf *kva;
396 struct rte_kni_mbuf *alloc_kva;
397 void *alloc_data_kva;
399 /* Get the number of entries in rx_q */
400 num_rq = kni_fifo_count(kni->rx_q);
402 /* Get the number of free entrie in tx_q */
403 num_tq = kni_fifo_free_count(kni->tx_q);
405 /* Get the number of entries in alloc_q */
406 num_aq = kni_fifo_count(kni->alloc_q);
408 /* Get the number of free entries in free_q */
409 num_fq = kni_fifo_free_count(kni->free_q);
411 /* Calculate the number of entries to be dequeued from rx_q */
412 num = min(num_rq, num_tq);
413 num = min(num, num_aq);
414 num = min(num, num_fq);
415 num = min_t(uint32_t, num, MBUF_BURST_SZ);
417 /* Return if no entry to dequeue from rx_q */
421 /* Burst dequeue from rx_q */
422 ret = kni_fifo_get(kni->rx_q, kni->pa, num);
424 return; /* Failing should not happen */
426 /* Dequeue entries from alloc_q */
427 ret = kni_fifo_get(kni->alloc_q, kni->alloc_pa, num);
431 for (i = 0; i < num; i++) {
432 kva = pa2kva(kni->pa[i]);
434 data_kva = kva2data_kva(kva);
435 kni->va[i] = pa2va(kni->pa[i], kva);
437 alloc_kva = pa2kva(kni->alloc_pa[i]);
438 alloc_data_kva = kva2data_kva(alloc_kva);
439 kni->alloc_va[i] = pa2va(kni->alloc_pa[i], alloc_kva);
441 memcpy(alloc_data_kva, data_kva, len);
442 alloc_kva->pkt_len = len;
443 alloc_kva->data_len = len;
445 kni->stats.tx_bytes += len;
446 kni->stats.rx_bytes += len;
449 /* Burst enqueue mbufs into tx_q */
450 ret = kni_fifo_put(kni->tx_q, kni->alloc_va, num);
452 /* Failing should not happen */
453 pr_err("Fail to enqueue mbufs into tx_q\n");
456 /* Burst enqueue mbufs into free_q */
457 ret = kni_fifo_put(kni->free_q, kni->va, num);
459 /* Failing should not happen */
460 pr_err("Fail to enqueue mbufs into free_q\n");
463 * Update statistic, and enqueue/dequeue failure is impossible,
464 * as all queues are checked at first.
466 kni->stats.tx_packets += num;
467 kni->stats.rx_packets += num;
471 * RX: loopback with enqueue/dequeue fifos and sk buffer copies.
474 kni_net_rx_lo_fifo_skb(struct kni_dev *kni)
478 uint32_t i, num_rq, num_fq, num;
479 struct rte_kni_mbuf *kva;
482 struct net_device *dev = kni->net_dev;
484 /* Get the number of entries in rx_q */
485 num_rq = kni_fifo_count(kni->rx_q);
487 /* Get the number of free entries in free_q */
488 num_fq = kni_fifo_free_count(kni->free_q);
490 /* Calculate the number of entries to dequeue from rx_q */
491 num = min(num_rq, num_fq);
492 num = min_t(uint32_t, num, MBUF_BURST_SZ);
494 /* Return if no entry to dequeue from rx_q */
498 /* Burst dequeue mbufs from rx_q */
499 ret = kni_fifo_get(kni->rx_q, kni->pa, num);
503 /* Copy mbufs to sk buffer and then call tx interface */
504 for (i = 0; i < num; i++) {
505 kva = pa2kva(kni->pa[i]);
507 data_kva = kva2data_kva(kva);
508 kni->va[i] = pa2va(kni->pa[i], kva);
510 skb = dev_alloc_skb(len + 2);
512 /* Align IP on 16B boundary */
514 memcpy(skb_put(skb, len), data_kva, len);
516 skb->ip_summed = CHECKSUM_UNNECESSARY;
520 /* Simulate real usage, allocate/copy skb twice */
521 skb = dev_alloc_skb(len + 2);
523 kni->stats.rx_dropped++;
527 /* Align IP on 16B boundary */
530 if (kva->nb_segs == 1) {
531 memcpy(skb_put(skb, len), data_kva, len);
534 int kva_nb_segs = kva->nb_segs;
536 for (nb_segs = 0; nb_segs < kva_nb_segs; nb_segs++) {
537 memcpy(skb_put(skb, kva->data_len),
538 data_kva, kva->data_len);
543 kva = pa2kva(va2pa(kva->next, kva));
544 data_kva = kva2data_kva(kva);
549 skb->ip_summed = CHECKSUM_UNNECESSARY;
551 kni->stats.rx_bytes += len;
552 kni->stats.rx_packets++;
554 /* call tx interface */
555 kni_net_tx(skb, dev);
558 /* enqueue all the mbufs from rx_q into free_q */
559 ret = kni_fifo_put(kni->free_q, kni->va, num);
561 /* Failing should not happen */
562 pr_err("Fail to enqueue mbufs into free_q\n");
567 kni_net_rx(struct kni_dev *kni)
570 * It doesn't need to check if it is NULL pointer,
571 * as it has a default value
573 (*kni_net_rx_func)(kni);
577 * Deal with a transmit timeout.
580 kni_net_tx_timeout(struct net_device *dev)
582 struct kni_dev *kni = netdev_priv(dev);
584 pr_debug("Transmit timeout at %ld, latency %ld\n", jiffies,
585 jiffies - dev_trans_start(dev));
587 kni->stats.tx_errors++;
588 netif_wake_queue(dev);
595 kni_net_ioctl(struct net_device *dev, struct ifreq *rq, int cmd)
597 pr_debug("kni_net_ioctl group:%d cmd:%d\n",
598 ((struct kni_dev *)netdev_priv(dev))->group_id, cmd);
604 kni_net_set_rx_mode(struct net_device *dev)
609 kni_net_change_mtu(struct net_device *dev, int new_mtu)
612 struct rte_kni_request req;
613 struct kni_dev *kni = netdev_priv(dev);
615 pr_debug("kni_net_change_mtu new mtu %d to be set\n", new_mtu);
617 memset(&req, 0, sizeof(req));
618 req.req_id = RTE_KNI_REQ_CHANGE_MTU;
619 req.new_mtu = new_mtu;
620 ret = kni_net_process_request(kni, &req);
621 if (ret == 0 && req.result == 0)
624 return (ret == 0) ? req.result : ret;
628 kni_net_set_promiscusity(struct net_device *netdev, int flags)
630 struct rte_kni_request req;
631 struct kni_dev *kni = netdev_priv(netdev);
633 memset(&req, 0, sizeof(req));
634 req.req_id = RTE_KNI_REQ_CHANGE_PROMISC;
636 if (netdev->flags & IFF_PROMISC)
637 req.promiscusity = 1;
639 req.promiscusity = 0;
640 kni_net_process_request(kni, &req);
644 * Checks if the user space application provided the resp message
647 kni_net_poll_resp(struct kni_dev *kni)
649 if (kni_fifo_count(kni->resp_q))
650 wake_up_interruptible(&kni->wq);
654 * Return statistics to the caller
656 static struct net_device_stats *
657 kni_net_stats(struct net_device *dev)
659 struct kni_dev *kni = netdev_priv(dev);
665 * Fill the eth header
668 kni_net_header(struct sk_buff *skb, struct net_device *dev,
669 unsigned short type, const void *daddr,
670 const void *saddr, uint32_t len)
672 struct ethhdr *eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
674 memcpy(eth->h_source, saddr ? saddr : dev->dev_addr, dev->addr_len);
675 memcpy(eth->h_dest, daddr ? daddr : dev->dev_addr, dev->addr_len);
676 eth->h_proto = htons(type);
678 return dev->hard_header_len;
682 * Re-fill the eth header
684 #ifdef HAVE_REBUILD_HEADER
686 kni_net_rebuild_header(struct sk_buff *skb)
688 struct net_device *dev = skb->dev;
689 struct ethhdr *eth = (struct ethhdr *) skb->data;
691 memcpy(eth->h_source, dev->dev_addr, dev->addr_len);
692 memcpy(eth->h_dest, dev->dev_addr, dev->addr_len);
699 * kni_net_set_mac - Change the Ethernet Address of the KNI NIC
700 * @netdev: network interface device structure
701 * @p: pointer to an address structure
703 * Returns 0 on success, negative on failure
706 kni_net_set_mac(struct net_device *netdev, void *p)
709 struct rte_kni_request req;
711 struct sockaddr *addr = p;
713 memset(&req, 0, sizeof(req));
714 req.req_id = RTE_KNI_REQ_CHANGE_MAC_ADDR;
716 if (!is_valid_ether_addr((unsigned char *)(addr->sa_data)))
717 return -EADDRNOTAVAIL;
719 memcpy(req.mac_addr, addr->sa_data, netdev->addr_len);
720 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
722 kni = netdev_priv(netdev);
723 ret = kni_net_process_request(kni, &req);
725 return (ret == 0 ? req.result : ret);
728 #ifdef HAVE_CHANGE_CARRIER_CB
730 kni_net_change_carrier(struct net_device *dev, bool new_carrier)
733 netif_carrier_on(dev);
735 netif_carrier_off(dev);
740 static const struct header_ops kni_net_header_ops = {
741 .create = kni_net_header,
742 #ifdef HAVE_REBUILD_HEADER
743 .rebuild = kni_net_rebuild_header,
745 .cache = NULL, /* disable caching */
748 static const struct net_device_ops kni_net_netdev_ops = {
749 .ndo_open = kni_net_open,
750 .ndo_stop = kni_net_release,
751 .ndo_set_config = kni_net_config,
752 .ndo_change_rx_flags = kni_net_set_promiscusity,
753 .ndo_start_xmit = kni_net_tx,
754 .ndo_change_mtu = kni_net_change_mtu,
755 .ndo_do_ioctl = kni_net_ioctl,
756 .ndo_set_rx_mode = kni_net_set_rx_mode,
757 .ndo_get_stats = kni_net_stats,
758 .ndo_tx_timeout = kni_net_tx_timeout,
759 .ndo_set_mac_address = kni_net_set_mac,
760 #ifdef HAVE_CHANGE_CARRIER_CB
761 .ndo_change_carrier = kni_net_change_carrier,
766 kni_net_init(struct net_device *dev)
768 struct kni_dev *kni = netdev_priv(dev);
770 init_waitqueue_head(&kni->wq);
771 mutex_init(&kni->sync_lock);
773 ether_setup(dev); /* assign some of the fields */
774 dev->netdev_ops = &kni_net_netdev_ops;
775 dev->header_ops = &kni_net_header_ops;
776 dev->watchdog_timeo = WD_TIMEOUT;
780 kni_net_config_lo_mode(char *lo_str)
783 pr_debug("loopback disabled");
787 if (!strcmp(lo_str, "lo_mode_none"))
788 pr_debug("loopback disabled");
789 else if (!strcmp(lo_str, "lo_mode_fifo")) {
790 pr_debug("loopback mode=lo_mode_fifo enabled");
791 kni_net_rx_func = kni_net_rx_lo_fifo;
792 } else if (!strcmp(lo_str, "lo_mode_fifo_skb")) {
793 pr_debug("loopback mode=lo_mode_fifo_skb enabled");
794 kni_net_rx_func = kni_net_rx_lo_fifo_skb;
796 pr_debug("Incognizant parameter, loopback disabled");