2 * Network-device interface management.
4 * Copyright (c) 2004-2005, Keir Fraser
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License version 2
8 * as published by the Free Software Foundation; or, when distributed
9 * separately from the Linux kernel or incorporated into other
10 * software packages, subject to the following license:
12 * Permission is hereby granted, free of charge, to any person obtaining a copy
13 * of this source file (the "Software"), to deal in the Software without
14 * restriction, including without limitation the rights to use, copy, modify,
15 * merge, publish, distribute, sublicense, and/or sell copies of the Software,
16 * and to permit persons to whom the Software is furnished to do so, subject to
17 * the following conditions:
19 * The above copyright notice and this permission notice shall be included in
20 * all copies or substantial portions of the Software.
22 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
23 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
24 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
25 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
26 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
27 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
33 #include <linux/kthread.h>
34 #include <linux/sched/task.h>
35 #include <linux/ethtool.h>
36 #include <linux/rtnetlink.h>
37 #include <linux/if_vlan.h>
38 #include <linux/vmalloc.h>
40 #include <xen/events.h>
41 #include <asm/xen/hypercall.h>
42 #include <xen/balloon.h>
44 #define XENVIF_QUEUE_LENGTH 32
45 #define XENVIF_NAPI_WEIGHT 64
47 /* Number of bytes allowed on the internal guest Rx queue. */
48 #define XENVIF_RX_QUEUE_BYTES (XEN_NETIF_RX_RING_SIZE/2 * PAGE_SIZE)
50 /* This function is used to set SKBTX_DEV_ZEROCOPY as well as
51 * increasing the inflight counter. We need to increase the inflight
52 * counter because core driver calls into xenvif_zerocopy_callback
53 * which calls xenvif_skb_zerocopy_complete.
55 void xenvif_skb_zerocopy_prepare(struct xenvif_queue *queue,
58 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
59 atomic_inc(&queue->inflight_packets);
62 void xenvif_skb_zerocopy_complete(struct xenvif_queue *queue)
64 atomic_dec(&queue->inflight_packets);
66 /* Wake the dealloc thread _after_ decrementing inflight_packets so
67 * that if kthread_stop() has already been called, the dealloc thread
68 * does not wait forever with nothing to wake it.
70 wake_up(&queue->dealloc_wq);
73 int xenvif_schedulable(struct xenvif *vif)
75 return netif_running(vif->dev) &&
76 test_bit(VIF_STATUS_CONNECTED, &vif->status) &&
80 static irqreturn_t xenvif_tx_interrupt(int irq, void *dev_id)
82 struct xenvif_queue *queue = dev_id;
84 if (RING_HAS_UNCONSUMED_REQUESTS(&queue->tx))
85 napi_schedule(&queue->napi);
90 static int xenvif_poll(struct napi_struct *napi, int budget)
92 struct xenvif_queue *queue =
93 container_of(napi, struct xenvif_queue, napi);
96 /* This vif is rogue, we pretend we've there is nothing to do
97 * for this vif to deschedule it from NAPI. But this interface
98 * will be turned off in thread context later.
100 if (unlikely(queue->vif->disabled)) {
105 work_done = xenvif_tx_action(queue, budget);
107 if (work_done < budget) {
108 napi_complete_done(napi, work_done);
109 xenvif_napi_schedule_or_enable_events(queue);
115 static irqreturn_t xenvif_rx_interrupt(int irq, void *dev_id)
117 struct xenvif_queue *queue = dev_id;
119 xenvif_kick_thread(queue);
124 irqreturn_t xenvif_interrupt(int irq, void *dev_id)
126 xenvif_tx_interrupt(irq, dev_id);
127 xenvif_rx_interrupt(irq, dev_id);
132 int xenvif_queue_stopped(struct xenvif_queue *queue)
134 struct net_device *dev = queue->vif->dev;
135 unsigned int id = queue->id;
136 return netif_tx_queue_stopped(netdev_get_tx_queue(dev, id));
139 void xenvif_wake_queue(struct xenvif_queue *queue)
141 struct net_device *dev = queue->vif->dev;
142 unsigned int id = queue->id;
143 netif_tx_wake_queue(netdev_get_tx_queue(dev, id));
146 static u16 xenvif_select_queue(struct net_device *dev, struct sk_buff *skb,
148 select_queue_fallback_t fallback)
150 struct xenvif *vif = netdev_priv(dev);
151 unsigned int size = vif->hash.size;
153 if (vif->hash.alg == XEN_NETIF_CTRL_HASH_ALGORITHM_NONE)
154 return fallback(dev, skb) % dev->real_num_tx_queues;
156 xenvif_set_skb_hash(vif, skb);
159 return skb_get_hash_raw(skb) % dev->real_num_tx_queues;
161 return vif->hash.mapping[skb_get_hash_raw(skb) % size];
164 static int xenvif_start_xmit(struct sk_buff *skb, struct net_device *dev)
166 struct xenvif *vif = netdev_priv(dev);
167 struct xenvif_queue *queue = NULL;
168 unsigned int num_queues;
170 struct xenvif_rx_cb *cb;
172 BUG_ON(skb->dev != dev);
174 /* Drop the packet if queues are not set up.
175 * This handler should be called inside an RCU read section
176 * so we don't need to enter it here explicitly.
178 num_queues = READ_ONCE(vif->num_queues);
182 /* Obtain the queue to be used to transmit this packet */
183 index = skb_get_queue_mapping(skb);
184 if (index >= num_queues) {
185 pr_warn_ratelimited("Invalid queue %hu for packet on interface %s\n.",
186 index, vif->dev->name);
189 queue = &vif->queues[index];
191 /* Drop the packet if queue is not ready */
192 if (queue->task == NULL ||
193 queue->dealloc_task == NULL ||
194 !xenvif_schedulable(vif))
197 if (vif->multicast_control && skb->pkt_type == PACKET_MULTICAST) {
198 struct ethhdr *eth = (struct ethhdr *)skb->data;
200 if (!xenvif_mcast_match(vif, eth->h_dest))
204 cb = XENVIF_RX_CB(skb);
205 cb->expires = jiffies + vif->drain_timeout;
207 /* If there is no hash algorithm configured then make sure there
208 * is no hash information in the socket buffer otherwise it
209 * would be incorrectly forwarded to the frontend.
211 if (vif->hash.alg == XEN_NETIF_CTRL_HASH_ALGORITHM_NONE)
214 xenvif_rx_queue_tail(queue, skb);
215 xenvif_kick_thread(queue);
220 vif->dev->stats.tx_dropped++;
225 static struct net_device_stats *xenvif_get_stats(struct net_device *dev)
227 struct xenvif *vif = netdev_priv(dev);
228 struct xenvif_queue *queue = NULL;
229 unsigned int num_queues;
237 num_queues = READ_ONCE(vif->num_queues);
239 /* Aggregate tx and rx stats from each queue */
240 for (index = 0; index < num_queues; ++index) {
241 queue = &vif->queues[index];
242 rx_bytes += queue->stats.rx_bytes;
243 rx_packets += queue->stats.rx_packets;
244 tx_bytes += queue->stats.tx_bytes;
245 tx_packets += queue->stats.tx_packets;
250 vif->dev->stats.rx_bytes = rx_bytes;
251 vif->dev->stats.rx_packets = rx_packets;
252 vif->dev->stats.tx_bytes = tx_bytes;
253 vif->dev->stats.tx_packets = tx_packets;
255 return &vif->dev->stats;
258 static void xenvif_up(struct xenvif *vif)
260 struct xenvif_queue *queue = NULL;
261 unsigned int num_queues = vif->num_queues;
262 unsigned int queue_index;
264 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
265 queue = &vif->queues[queue_index];
266 napi_enable(&queue->napi);
267 enable_irq(queue->tx_irq);
268 if (queue->tx_irq != queue->rx_irq)
269 enable_irq(queue->rx_irq);
270 xenvif_napi_schedule_or_enable_events(queue);
274 static void xenvif_down(struct xenvif *vif)
276 struct xenvif_queue *queue = NULL;
277 unsigned int num_queues = vif->num_queues;
278 unsigned int queue_index;
280 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
281 queue = &vif->queues[queue_index];
282 disable_irq(queue->tx_irq);
283 if (queue->tx_irq != queue->rx_irq)
284 disable_irq(queue->rx_irq);
285 napi_disable(&queue->napi);
286 del_timer_sync(&queue->credit_timeout);
290 static int xenvif_open(struct net_device *dev)
292 struct xenvif *vif = netdev_priv(dev);
293 if (test_bit(VIF_STATUS_CONNECTED, &vif->status))
295 netif_tx_start_all_queues(dev);
299 static int xenvif_close(struct net_device *dev)
301 struct xenvif *vif = netdev_priv(dev);
302 if (test_bit(VIF_STATUS_CONNECTED, &vif->status))
304 netif_tx_stop_all_queues(dev);
308 static int xenvif_change_mtu(struct net_device *dev, int mtu)
310 struct xenvif *vif = netdev_priv(dev);
311 int max = vif->can_sg ? ETH_MAX_MTU - VLAN_ETH_HLEN : ETH_DATA_LEN;
319 static netdev_features_t xenvif_fix_features(struct net_device *dev,
320 netdev_features_t features)
322 struct xenvif *vif = netdev_priv(dev);
325 features &= ~NETIF_F_SG;
326 if (~(vif->gso_mask) & GSO_BIT(TCPV4))
327 features &= ~NETIF_F_TSO;
328 if (~(vif->gso_mask) & GSO_BIT(TCPV6))
329 features &= ~NETIF_F_TSO6;
331 features &= ~NETIF_F_IP_CSUM;
333 features &= ~NETIF_F_IPV6_CSUM;
338 static const struct xenvif_stat {
339 char name[ETH_GSTRING_LEN];
343 "rx_gso_checksum_fixup",
344 offsetof(struct xenvif_stats, rx_gso_checksum_fixup)
346 /* If (sent != success + fail), there are probably packets never
351 offsetof(struct xenvif_stats, tx_zerocopy_sent),
354 "tx_zerocopy_success",
355 offsetof(struct xenvif_stats, tx_zerocopy_success),
359 offsetof(struct xenvif_stats, tx_zerocopy_fail)
361 /* Number of packets exceeding MAX_SKB_FRAG slots. You should use
362 * a guest with the same MAX_SKB_FRAG
366 offsetof(struct xenvif_stats, tx_frag_overflow)
370 static int xenvif_get_sset_count(struct net_device *dev, int string_set)
372 switch (string_set) {
374 return ARRAY_SIZE(xenvif_stats);
380 static void xenvif_get_ethtool_stats(struct net_device *dev,
381 struct ethtool_stats *stats, u64 * data)
383 struct xenvif *vif = netdev_priv(dev);
384 unsigned int num_queues;
386 unsigned int queue_index;
389 num_queues = READ_ONCE(vif->num_queues);
391 for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++) {
392 unsigned long accum = 0;
393 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
394 void *vif_stats = &vif->queues[queue_index].stats;
395 accum += *(unsigned long *)(vif_stats + xenvif_stats[i].offset);
403 static void xenvif_get_strings(struct net_device *dev, u32 stringset, u8 * data)
409 for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++)
410 memcpy(data + i * ETH_GSTRING_LEN,
411 xenvif_stats[i].name, ETH_GSTRING_LEN);
416 static const struct ethtool_ops xenvif_ethtool_ops = {
417 .get_link = ethtool_op_get_link,
419 .get_sset_count = xenvif_get_sset_count,
420 .get_ethtool_stats = xenvif_get_ethtool_stats,
421 .get_strings = xenvif_get_strings,
424 static const struct net_device_ops xenvif_netdev_ops = {
425 .ndo_select_queue = xenvif_select_queue,
426 .ndo_start_xmit = xenvif_start_xmit,
427 .ndo_get_stats = xenvif_get_stats,
428 .ndo_open = xenvif_open,
429 .ndo_stop = xenvif_close,
430 .ndo_change_mtu = xenvif_change_mtu,
431 .ndo_fix_features = xenvif_fix_features,
432 .ndo_set_mac_address = eth_mac_addr,
433 .ndo_validate_addr = eth_validate_addr,
436 struct xenvif *xenvif_alloc(struct device *parent, domid_t domid,
440 struct net_device *dev;
442 char name[IFNAMSIZ] = {};
444 snprintf(name, IFNAMSIZ - 1, "vif%u.%u", domid, handle);
445 /* Allocate a netdev with the max. supported number of queues.
446 * When the guest selects the desired number, it will be updated
447 * via netif_set_real_num_*_queues().
449 dev = alloc_netdev_mq(sizeof(struct xenvif), name, NET_NAME_UNKNOWN,
450 ether_setup, xenvif_max_queues);
452 pr_warn("Could not allocate netdev for %s\n", name);
453 return ERR_PTR(-ENOMEM);
456 SET_NETDEV_DEV(dev, parent);
458 vif = netdev_priv(dev);
461 vif->handle = handle;
465 vif->disabled = false;
466 vif->drain_timeout = msecs_to_jiffies(rx_drain_timeout_msecs);
467 vif->stall_timeout = msecs_to_jiffies(rx_stall_timeout_msecs);
469 /* Start out with no queues. */
473 spin_lock_init(&vif->lock);
474 INIT_LIST_HEAD(&vif->fe_mcast_addr);
476 dev->netdev_ops = &xenvif_netdev_ops;
477 dev->hw_features = NETIF_F_SG |
478 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
479 NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_FRAGLIST;
480 dev->features = dev->hw_features | NETIF_F_RXCSUM;
481 dev->ethtool_ops = &xenvif_ethtool_ops;
483 dev->tx_queue_len = XENVIF_QUEUE_LENGTH;
486 dev->max_mtu = ETH_MAX_MTU - VLAN_ETH_HLEN;
489 * Initialise a dummy MAC address. We choose the numerically
490 * largest non-broadcast address to prevent the address getting
491 * stolen by an Ethernet bridge for STP purposes.
492 * (FE:FF:FF:FF:FF:FF)
494 eth_broadcast_addr(dev->dev_addr);
495 dev->dev_addr[0] &= ~0x01;
497 netif_carrier_off(dev);
499 err = register_netdev(dev);
501 netdev_warn(dev, "Could not register device: err=%d\n", err);
506 netdev_dbg(dev, "Successfully created xenvif\n");
508 __module_get(THIS_MODULE);
513 int xenvif_init_queue(struct xenvif_queue *queue)
517 queue->credit_bytes = queue->remaining_credit = ~0UL;
518 queue->credit_usec = 0UL;
519 init_timer(&queue->credit_timeout);
520 queue->credit_timeout.function = xenvif_tx_credit_callback;
521 queue->credit_window_start = get_jiffies_64();
523 queue->rx_queue_max = XENVIF_RX_QUEUE_BYTES;
525 skb_queue_head_init(&queue->rx_queue);
526 skb_queue_head_init(&queue->tx_queue);
528 queue->pending_cons = 0;
529 queue->pending_prod = MAX_PENDING_REQS;
530 for (i = 0; i < MAX_PENDING_REQS; ++i)
531 queue->pending_ring[i] = i;
533 spin_lock_init(&queue->callback_lock);
534 spin_lock_init(&queue->response_lock);
536 /* If ballooning is disabled, this will consume real memory, so you
537 * better enable it. The long term solution would be to use just a
538 * bunch of valid page descriptors, without dependency on ballooning
540 err = gnttab_alloc_pages(MAX_PENDING_REQS,
543 netdev_err(queue->vif->dev, "Could not reserve mmap_pages\n");
547 for (i = 0; i < MAX_PENDING_REQS; i++) {
548 queue->pending_tx_info[i].callback_struct = (struct ubuf_info)
549 { .callback = xenvif_zerocopy_callback,
552 queue->grant_tx_handle[i] = NETBACK_INVALID_HANDLE;
558 void xenvif_carrier_on(struct xenvif *vif)
561 if (!vif->can_sg && vif->dev->mtu > ETH_DATA_LEN)
562 dev_set_mtu(vif->dev, ETH_DATA_LEN);
563 netdev_update_features(vif->dev);
564 set_bit(VIF_STATUS_CONNECTED, &vif->status);
565 if (netif_running(vif->dev))
570 int xenvif_connect_ctrl(struct xenvif *vif, grant_ref_t ring_ref,
573 struct net_device *dev = vif->dev;
575 struct xen_netif_ctrl_sring *shared;
578 err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(vif),
579 &ring_ref, 1, &addr);
583 shared = (struct xen_netif_ctrl_sring *)addr;
584 BACK_RING_INIT(&vif->ctrl, shared, XEN_PAGE_SIZE);
586 err = bind_interdomain_evtchn_to_irq(vif->domid, evtchn);
592 xenvif_init_hash(vif);
594 err = request_threaded_irq(vif->ctrl_irq, NULL, xenvif_ctrl_irq_fn,
595 IRQF_ONESHOT, "xen-netback-ctrl", vif);
597 pr_warn("Could not setup irq handler for %s\n", dev->name);
604 xenvif_deinit_hash(vif);
605 unbind_from_irqhandler(vif->ctrl_irq, vif);
609 xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif),
611 vif->ctrl.sring = NULL;
617 int xenvif_connect_data(struct xenvif_queue *queue,
618 unsigned long tx_ring_ref,
619 unsigned long rx_ring_ref,
620 unsigned int tx_evtchn,
621 unsigned int rx_evtchn)
623 struct task_struct *task;
626 BUG_ON(queue->tx_irq);
628 BUG_ON(queue->dealloc_task);
630 err = xenvif_map_frontend_data_rings(queue, tx_ring_ref,
635 init_waitqueue_head(&queue->wq);
636 init_waitqueue_head(&queue->dealloc_wq);
637 atomic_set(&queue->inflight_packets, 0);
639 netif_napi_add(queue->vif->dev, &queue->napi, xenvif_poll,
642 if (tx_evtchn == rx_evtchn) {
643 /* feature-split-event-channels == 0 */
644 err = bind_interdomain_evtchn_to_irqhandler(
645 queue->vif->domid, tx_evtchn, xenvif_interrupt, 0,
649 queue->tx_irq = queue->rx_irq = err;
650 disable_irq(queue->tx_irq);
652 /* feature-split-event-channels == 1 */
653 snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
654 "%s-tx", queue->name);
655 err = bind_interdomain_evtchn_to_irqhandler(
656 queue->vif->domid, tx_evtchn, xenvif_tx_interrupt, 0,
657 queue->tx_irq_name, queue);
661 disable_irq(queue->tx_irq);
663 snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
664 "%s-rx", queue->name);
665 err = bind_interdomain_evtchn_to_irqhandler(
666 queue->vif->domid, rx_evtchn, xenvif_rx_interrupt, 0,
667 queue->rx_irq_name, queue);
671 disable_irq(queue->rx_irq);
674 queue->stalled = true;
676 task = kthread_create(xenvif_kthread_guest_rx,
677 (void *)queue, "%s-guest-rx", queue->name);
679 pr_warn("Could not allocate kthread for %s\n", queue->name);
684 get_task_struct(task);
686 task = kthread_create(xenvif_dealloc_kthread,
687 (void *)queue, "%s-dealloc", queue->name);
689 pr_warn("Could not allocate kthread for %s\n", queue->name);
693 queue->dealloc_task = task;
695 wake_up_process(queue->task);
696 wake_up_process(queue->dealloc_task);
701 unbind_from_irqhandler(queue->rx_irq, queue);
704 unbind_from_irqhandler(queue->tx_irq, queue);
707 xenvif_unmap_frontend_data_rings(queue);
708 netif_napi_del(&queue->napi);
710 module_put(THIS_MODULE);
714 void xenvif_carrier_off(struct xenvif *vif)
716 struct net_device *dev = vif->dev;
719 if (test_and_clear_bit(VIF_STATUS_CONNECTED, &vif->status)) {
720 netif_carrier_off(dev); /* discard queued packets */
721 if (netif_running(dev))
727 void xenvif_disconnect_data(struct xenvif *vif)
729 struct xenvif_queue *queue = NULL;
730 unsigned int num_queues = vif->num_queues;
731 unsigned int queue_index;
733 xenvif_carrier_off(vif);
735 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
736 queue = &vif->queues[queue_index];
738 netif_napi_del(&queue->napi);
741 kthread_stop(queue->task);
742 put_task_struct(queue->task);
746 if (queue->dealloc_task) {
747 kthread_stop(queue->dealloc_task);
748 queue->dealloc_task = NULL;
752 if (queue->tx_irq == queue->rx_irq)
753 unbind_from_irqhandler(queue->tx_irq, queue);
755 unbind_from_irqhandler(queue->tx_irq, queue);
756 unbind_from_irqhandler(queue->rx_irq, queue);
761 xenvif_unmap_frontend_data_rings(queue);
764 xenvif_mcast_addr_list_free(vif);
767 void xenvif_disconnect_ctrl(struct xenvif *vif)
770 xenvif_deinit_hash(vif);
771 unbind_from_irqhandler(vif->ctrl_irq, vif);
775 if (vif->ctrl.sring) {
776 xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif),
778 vif->ctrl.sring = NULL;
782 /* Reverse the relevant parts of xenvif_init_queue().
783 * Used for queue teardown from xenvif_free(), and on the
784 * error handling paths in xenbus.c:connect().
786 void xenvif_deinit_queue(struct xenvif_queue *queue)
788 gnttab_free_pages(MAX_PENDING_REQS, queue->mmap_pages);
791 void xenvif_free(struct xenvif *vif)
793 struct xenvif_queue *queues = vif->queues;
794 unsigned int num_queues = vif->num_queues;
795 unsigned int queue_index;
797 unregister_netdev(vif->dev);
798 free_netdev(vif->dev);
800 for (queue_index = 0; queue_index < num_queues; ++queue_index)
801 xenvif_deinit_queue(&queues[queue_index]);
804 module_put(THIS_MODULE);