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/ethtool.h>
17 #include <linux/skbuff.h>
18 #include <linux/kthread.h>
19 #include <linux/delay.h>
21 #include <rte_kni_common.h>
27 #define WD_TIMEOUT 5 /*jiffies */
29 #define KNI_WAIT_RESPONSE_TIMEOUT 300 /* 3 seconds */
31 /* typedef for rx function */
32 typedef void (*kni_net_rx_t)(struct kni_dev *kni);
34 static void kni_net_rx_normal(struct kni_dev *kni);
36 /* kni rx function pointer, with default to normal rx */
37 static kni_net_rx_t kni_net_rx_func = kni_net_rx_normal;
39 /* physical address to kernel virtual address */
43 return phys_to_virt((unsigned long)pa);
46 /* physical address to virtual address */
48 pa2va(void *pa, struct rte_kni_mbuf *m)
52 va = (void *)((unsigned long)pa +
53 (unsigned long)m->buf_addr -
54 (unsigned long)m->buf_physaddr);
58 /* mbuf data kernel virtual address from mbuf kernel virtual address */
60 kva2data_kva(struct rte_kni_mbuf *m)
62 return phys_to_virt(m->buf_physaddr + m->data_off);
66 * It can be called to process the request.
69 kni_net_process_request(struct kni_dev *kni, struct rte_kni_request *req)
77 pr_err("No kni instance or request\n");
81 mutex_lock(&kni->sync_lock);
84 memcpy(kni->sync_kva, req, sizeof(struct rte_kni_request));
85 num = kni_fifo_put(kni->req_q, &kni->sync_va, 1);
87 pr_err("Cannot send to req_q\n");
92 ret_val = wait_event_interruptible_timeout(kni->wq,
93 kni_fifo_count(kni->resp_q), 3 * HZ);
94 if (signal_pending(current) || ret_val <= 0) {
98 num = kni_fifo_get(kni->resp_q, (void **)&resp_va, 1);
99 if (num != 1 || resp_va != kni->sync_va) {
100 /* This should never happen */
101 pr_err("No data in resp_q\n");
106 memcpy(req, kni->sync_kva, sizeof(struct rte_kni_request));
110 mutex_unlock(&kni->sync_lock);
118 kni_net_open(struct net_device *dev)
121 struct rte_kni_request req;
122 struct kni_dev *kni = netdev_priv(dev);
124 netif_start_queue(dev);
125 if (dflt_carrier == 1)
126 netif_carrier_on(dev);
128 netif_carrier_off(dev);
130 memset(&req, 0, sizeof(req));
131 req.req_id = RTE_KNI_REQ_CFG_NETWORK_IF;
133 /* Setting if_up to non-zero means up */
135 ret = kni_net_process_request(kni, &req);
137 return (ret == 0) ? req.result : ret;
141 kni_net_release(struct net_device *dev)
144 struct rte_kni_request req;
145 struct kni_dev *kni = netdev_priv(dev);
147 netif_stop_queue(dev); /* can't transmit any more */
148 netif_carrier_off(dev);
150 memset(&req, 0, sizeof(req));
151 req.req_id = RTE_KNI_REQ_CFG_NETWORK_IF;
153 /* Setting if_up to 0 means down */
155 ret = kni_net_process_request(kni, &req);
157 return (ret == 0) ? req.result : ret;
161 kni_fifo_trans_pa2va(struct kni_dev *kni,
162 struct rte_kni_fifo *src_pa, struct rte_kni_fifo *dst_va)
164 uint32_t ret, i, num_dst, num_rx;
165 struct rte_kni_mbuf *kva, *prev_kva;
170 num_dst = kni_fifo_free_count(dst_va);
174 num_rx = min_t(uint32_t, num_dst, MBUF_BURST_SZ);
176 num_rx = kni_fifo_get(src_pa, kni->pa, num_rx);
180 for (i = 0; i < num_rx; i++) {
181 kva = pa2kva(kni->pa[i]);
182 kni->va[i] = pa2va(kni->pa[i], kva);
184 kva_nb_segs = kva->nb_segs;
185 for (nb_segs = 0; nb_segs < kva_nb_segs; nb_segs++) {
190 kva = pa2kva(kva->next);
191 /* Convert physical address to virtual address */
192 prev_kva->next = pa2va(prev_kva->next, kva);
196 ret = kni_fifo_put(dst_va, kni->va, num_rx);
198 /* Failing should not happen */
199 pr_err("Fail to enqueue entries into dst_va\n");
205 /* Try to release mbufs when kni release */
206 void kni_net_release_fifo_phy(struct kni_dev *kni)
208 /* release rx_q first, because it can't release in userspace */
209 kni_fifo_trans_pa2va(kni, kni->rx_q, kni->free_q);
210 /* release alloc_q for speeding up kni release in userspace */
211 kni_fifo_trans_pa2va(kni, kni->alloc_q, kni->free_q);
215 * Configuration changes (passed on by ifconfig)
218 kni_net_config(struct net_device *dev, struct ifmap *map)
220 if (dev->flags & IFF_UP) /* can't act on a running interface */
223 /* ignore other fields */
228 * Transmit a packet (called by the kernel)
231 kni_net_tx(struct sk_buff *skb, struct net_device *dev)
235 struct kni_dev *kni = netdev_priv(dev);
236 struct rte_kni_mbuf *pkt_kva = NULL;
240 /* save the timestamp */
241 #ifdef HAVE_TRANS_START_HELPER
242 netif_trans_update(dev);
244 dev->trans_start = jiffies;
247 /* Check if the length of skb is less than mbuf size */
248 if (skb->len > kni->mbuf_size)
252 * Check if it has at least one free entry in tx_q and
253 * one entry in alloc_q.
255 if (kni_fifo_free_count(kni->tx_q) == 0 ||
256 kni_fifo_count(kni->alloc_q) == 0) {
258 * If no free entry in tx_q or no entry in alloc_q,
259 * drops skb and goes out.
264 /* dequeue a mbuf from alloc_q */
265 ret = kni_fifo_get(kni->alloc_q, &pkt_pa, 1);
266 if (likely(ret == 1)) {
269 pkt_kva = pa2kva(pkt_pa);
270 data_kva = kva2data_kva(pkt_kva);
271 pkt_va = pa2va(pkt_pa, pkt_kva);
274 memcpy(data_kva, skb->data, len);
275 if (unlikely(len < ETH_ZLEN)) {
276 memset(data_kva + len, 0, ETH_ZLEN - len);
279 pkt_kva->pkt_len = len;
280 pkt_kva->data_len = len;
282 /* enqueue mbuf into tx_q */
283 ret = kni_fifo_put(kni->tx_q, &pkt_va, 1);
284 if (unlikely(ret != 1)) {
285 /* Failing should not happen */
286 pr_err("Fail to enqueue mbuf into tx_q\n");
290 /* Failing should not happen */
291 pr_err("Fail to dequeue mbuf from alloc_q\n");
295 /* Free skb and update statistics */
297 dev->stats.tx_bytes += len;
298 dev->stats.tx_packets++;
303 /* Free skb and update statistics */
305 dev->stats.tx_dropped++;
311 * RX: normal working mode
314 kni_net_rx_normal(struct kni_dev *kni)
318 uint32_t i, num_rx, num_fq;
319 struct rte_kni_mbuf *kva, *prev_kva;
322 struct net_device *dev = kni->net_dev;
324 /* Get the number of free entries in free_q */
325 num_fq = kni_fifo_free_count(kni->free_q);
327 /* No room on the free_q, bail out */
331 /* Calculate the number of entries to dequeue from rx_q */
332 num_rx = min_t(uint32_t, num_fq, MBUF_BURST_SZ);
334 /* Burst dequeue from rx_q */
335 num_rx = kni_fifo_get(kni->rx_q, kni->pa, num_rx);
339 /* Transfer received packets to netif */
340 for (i = 0; i < num_rx; i++) {
341 kva = pa2kva(kni->pa[i]);
343 data_kva = kva2data_kva(kva);
344 kni->va[i] = pa2va(kni->pa[i], kva);
346 skb = netdev_alloc_skb(dev, len);
348 /* Update statistics */
349 dev->stats.rx_dropped++;
353 if (kva->nb_segs == 1) {
354 memcpy(skb_put(skb, len), data_kva, len);
357 int kva_nb_segs = kva->nb_segs;
359 for (nb_segs = 0; nb_segs < kva_nb_segs; nb_segs++) {
360 memcpy(skb_put(skb, kva->data_len),
361 data_kva, kva->data_len);
367 kva = pa2kva(kva->next);
368 data_kva = kva2data_kva(kva);
369 /* Convert physical address to virtual address */
370 prev_kva->next = pa2va(prev_kva->next, kva);
374 skb->protocol = eth_type_trans(skb, dev);
375 skb->ip_summed = CHECKSUM_UNNECESSARY;
377 /* Call netif interface */
380 /* Update statistics */
381 dev->stats.rx_bytes += len;
382 dev->stats.rx_packets++;
385 /* Burst enqueue mbufs into free_q */
386 ret = kni_fifo_put(kni->free_q, kni->va, num_rx);
388 /* Failing should not happen */
389 pr_err("Fail to enqueue entries into free_q\n");
393 * RX: loopback with enqueue/dequeue fifos.
396 kni_net_rx_lo_fifo(struct kni_dev *kni)
400 uint32_t i, num, num_rq, num_tq, num_aq, num_fq;
401 struct rte_kni_mbuf *kva, *next_kva;
403 struct rte_kni_mbuf *alloc_kva;
404 void *alloc_data_kva;
405 struct net_device *dev = kni->net_dev;
407 /* Get the number of entries in rx_q */
408 num_rq = kni_fifo_count(kni->rx_q);
410 /* Get the number of free entries in tx_q */
411 num_tq = kni_fifo_free_count(kni->tx_q);
413 /* Get the number of entries in alloc_q */
414 num_aq = kni_fifo_count(kni->alloc_q);
416 /* Get the number of free entries in free_q */
417 num_fq = kni_fifo_free_count(kni->free_q);
419 /* Calculate the number of entries to be dequeued from rx_q */
420 num = min(num_rq, num_tq);
421 num = min(num, num_aq);
422 num = min(num, num_fq);
423 num = min_t(uint32_t, num, MBUF_BURST_SZ);
425 /* Return if no entry to dequeue from rx_q */
429 /* Burst dequeue from rx_q */
430 ret = kni_fifo_get(kni->rx_q, kni->pa, num);
432 return; /* Failing should not happen */
434 /* Dequeue entries from alloc_q */
435 ret = kni_fifo_get(kni->alloc_q, kni->alloc_pa, num);
439 for (i = 0; i < num; i++) {
440 kva = pa2kva(kni->pa[i]);
442 data_kva = kva2data_kva(kva);
443 kni->va[i] = pa2va(kni->pa[i], kva);
446 next_kva = pa2kva(kva->next);
447 /* Convert physical address to virtual address */
448 kva->next = pa2va(kva->next, next_kva);
452 alloc_kva = pa2kva(kni->alloc_pa[i]);
453 alloc_data_kva = kva2data_kva(alloc_kva);
454 kni->alloc_va[i] = pa2va(kni->alloc_pa[i], alloc_kva);
456 memcpy(alloc_data_kva, data_kva, len);
457 alloc_kva->pkt_len = len;
458 alloc_kva->data_len = len;
460 dev->stats.tx_bytes += len;
461 dev->stats.rx_bytes += len;
464 /* Burst enqueue mbufs into tx_q */
465 ret = kni_fifo_put(kni->tx_q, kni->alloc_va, num);
467 /* Failing should not happen */
468 pr_err("Fail to enqueue mbufs into tx_q\n");
471 /* Burst enqueue mbufs into free_q */
472 ret = kni_fifo_put(kni->free_q, kni->va, num);
474 /* Failing should not happen */
475 pr_err("Fail to enqueue mbufs into free_q\n");
478 * Update statistic, and enqueue/dequeue failure is impossible,
479 * as all queues are checked at first.
481 dev->stats.tx_packets += num;
482 dev->stats.rx_packets += num;
486 * RX: loopback with enqueue/dequeue fifos and sk buffer copies.
489 kni_net_rx_lo_fifo_skb(struct kni_dev *kni)
493 uint32_t i, num_rq, num_fq, num;
494 struct rte_kni_mbuf *kva, *prev_kva;
497 struct net_device *dev = kni->net_dev;
499 /* Get the number of entries in rx_q */
500 num_rq = kni_fifo_count(kni->rx_q);
502 /* Get the number of free entries in free_q */
503 num_fq = kni_fifo_free_count(kni->free_q);
505 /* Calculate the number of entries to dequeue from rx_q */
506 num = min(num_rq, num_fq);
507 num = min_t(uint32_t, num, MBUF_BURST_SZ);
509 /* Return if no entry to dequeue from rx_q */
513 /* Burst dequeue mbufs from rx_q */
514 ret = kni_fifo_get(kni->rx_q, kni->pa, num);
518 /* Copy mbufs to sk buffer and then call tx interface */
519 for (i = 0; i < num; i++) {
520 kva = pa2kva(kni->pa[i]);
522 data_kva = kva2data_kva(kva);
523 kni->va[i] = pa2va(kni->pa[i], kva);
525 skb = netdev_alloc_skb(dev, len);
527 memcpy(skb_put(skb, len), data_kva, len);
528 skb->ip_summed = CHECKSUM_UNNECESSARY;
532 /* Simulate real usage, allocate/copy skb twice */
533 skb = netdev_alloc_skb(dev, len);
535 dev->stats.rx_dropped++;
539 if (kva->nb_segs == 1) {
540 memcpy(skb_put(skb, len), data_kva, len);
543 int kva_nb_segs = kva->nb_segs;
545 for (nb_segs = 0; nb_segs < kva_nb_segs; nb_segs++) {
546 memcpy(skb_put(skb, kva->data_len),
547 data_kva, kva->data_len);
553 kva = pa2kva(kva->next);
554 data_kva = kva2data_kva(kva);
555 /* Convert physical address to virtual address */
556 prev_kva->next = pa2va(prev_kva->next, kva);
560 skb->ip_summed = CHECKSUM_UNNECESSARY;
562 dev->stats.rx_bytes += len;
563 dev->stats.rx_packets++;
565 /* call tx interface */
566 kni_net_tx(skb, dev);
569 /* enqueue all the mbufs from rx_q into free_q */
570 ret = kni_fifo_put(kni->free_q, kni->va, num);
572 /* Failing should not happen */
573 pr_err("Fail to enqueue mbufs into free_q\n");
578 kni_net_rx(struct kni_dev *kni)
581 * It doesn't need to check if it is NULL pointer,
582 * as it has a default value
584 (*kni_net_rx_func)(kni);
588 * Deal with a transmit timeout.
591 kni_net_tx_timeout(struct net_device *dev)
593 pr_debug("Transmit timeout at %ld, latency %ld\n", jiffies,
594 jiffies - dev_trans_start(dev));
596 dev->stats.tx_errors++;
597 netif_wake_queue(dev);
601 kni_net_change_mtu(struct net_device *dev, int new_mtu)
604 struct rte_kni_request req;
605 struct kni_dev *kni = netdev_priv(dev);
607 pr_debug("kni_net_change_mtu new mtu %d to be set\n", new_mtu);
609 memset(&req, 0, sizeof(req));
610 req.req_id = RTE_KNI_REQ_CHANGE_MTU;
611 req.new_mtu = new_mtu;
612 ret = kni_net_process_request(kni, &req);
613 if (ret == 0 && req.result == 0)
616 return (ret == 0) ? req.result : ret;
620 kni_net_set_promiscusity(struct net_device *netdev, int flags)
622 struct rte_kni_request req;
623 struct kni_dev *kni = netdev_priv(netdev);
625 memset(&req, 0, sizeof(req));
626 req.req_id = RTE_KNI_REQ_CHANGE_PROMISC;
628 if (netdev->flags & IFF_PROMISC)
629 req.promiscusity = 1;
631 req.promiscusity = 0;
632 kni_net_process_request(kni, &req);
636 * Checks if the user space application provided the resp message
639 kni_net_poll_resp(struct kni_dev *kni)
641 if (kni_fifo_count(kni->resp_q))
642 wake_up_interruptible(&kni->wq);
646 * Fill the eth header
649 kni_net_header(struct sk_buff *skb, struct net_device *dev,
650 unsigned short type, const void *daddr,
651 const void *saddr, uint32_t len)
653 struct ethhdr *eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
655 memcpy(eth->h_source, saddr ? saddr : dev->dev_addr, dev->addr_len);
656 memcpy(eth->h_dest, daddr ? daddr : dev->dev_addr, dev->addr_len);
657 eth->h_proto = htons(type);
659 return dev->hard_header_len;
663 * Re-fill the eth header
665 #ifdef HAVE_REBUILD_HEADER
667 kni_net_rebuild_header(struct sk_buff *skb)
669 struct net_device *dev = skb->dev;
670 struct ethhdr *eth = (struct ethhdr *) skb->data;
672 memcpy(eth->h_source, dev->dev_addr, dev->addr_len);
673 memcpy(eth->h_dest, dev->dev_addr, dev->addr_len);
680 * kni_net_set_mac - Change the Ethernet Address of the KNI NIC
681 * @netdev: network interface device structure
682 * @p: pointer to an address structure
684 * Returns 0 on success, negative on failure
687 kni_net_set_mac(struct net_device *netdev, void *p)
690 struct rte_kni_request req;
692 struct sockaddr *addr = p;
694 memset(&req, 0, sizeof(req));
695 req.req_id = RTE_KNI_REQ_CHANGE_MAC_ADDR;
697 if (!is_valid_ether_addr((unsigned char *)(addr->sa_data)))
698 return -EADDRNOTAVAIL;
700 memcpy(req.mac_addr, addr->sa_data, netdev->addr_len);
701 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
703 kni = netdev_priv(netdev);
704 ret = kni_net_process_request(kni, &req);
706 return (ret == 0 ? req.result : ret);
709 #ifdef HAVE_CHANGE_CARRIER_CB
711 kni_net_change_carrier(struct net_device *dev, bool new_carrier)
714 netif_carrier_on(dev);
716 netif_carrier_off(dev);
721 static const struct header_ops kni_net_header_ops = {
722 .create = kni_net_header,
723 .parse = eth_header_parse,
724 #ifdef HAVE_REBUILD_HEADER
725 .rebuild = kni_net_rebuild_header,
727 .cache = NULL, /* disable caching */
730 static const struct net_device_ops kni_net_netdev_ops = {
731 .ndo_open = kni_net_open,
732 .ndo_stop = kni_net_release,
733 .ndo_set_config = kni_net_config,
734 .ndo_change_rx_flags = kni_net_set_promiscusity,
735 .ndo_start_xmit = kni_net_tx,
736 .ndo_change_mtu = kni_net_change_mtu,
737 .ndo_tx_timeout = kni_net_tx_timeout,
738 .ndo_set_mac_address = kni_net_set_mac,
739 #ifdef HAVE_CHANGE_CARRIER_CB
740 .ndo_change_carrier = kni_net_change_carrier,
744 static void kni_get_drvinfo(struct net_device *dev,
745 struct ethtool_drvinfo *info)
747 strlcpy(info->version, KNI_VERSION, sizeof(info->version));
748 strlcpy(info->driver, "kni", sizeof(info->driver));
751 static const struct ethtool_ops kni_net_ethtool_ops = {
752 .get_drvinfo = kni_get_drvinfo,
753 .get_link = ethtool_op_get_link,
757 kni_net_init(struct net_device *dev)
759 struct kni_dev *kni = netdev_priv(dev);
761 init_waitqueue_head(&kni->wq);
762 mutex_init(&kni->sync_lock);
764 ether_setup(dev); /* assign some of the fields */
765 dev->netdev_ops = &kni_net_netdev_ops;
766 dev->header_ops = &kni_net_header_ops;
767 dev->ethtool_ops = &kni_net_ethtool_ops;
768 dev->watchdog_timeo = WD_TIMEOUT;
772 kni_net_config_lo_mode(char *lo_str)
775 pr_debug("loopback disabled");
779 if (!strcmp(lo_str, "lo_mode_none"))
780 pr_debug("loopback disabled");
781 else if (!strcmp(lo_str, "lo_mode_fifo")) {
782 pr_debug("loopback mode=lo_mode_fifo enabled");
783 kni_net_rx_func = kni_net_rx_lo_fifo;
784 } else if (!strcmp(lo_str, "lo_mode_fifo_skb")) {
785 pr_debug("loopback mode=lo_mode_fifo_skb enabled");
786 kni_net_rx_func = kni_net_rx_lo_fifo_skb;
788 pr_debug("Unknown loopback parameter, disabled");