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 /* If the queue is rate-limited, it shall be
110 * rescheduled in the timer callback.
112 if (likely(!queue->rate_limited))
113 xenvif_napi_schedule_or_enable_events(queue);
119 static irqreturn_t xenvif_rx_interrupt(int irq, void *dev_id)
121 struct xenvif_queue *queue = dev_id;
123 xenvif_kick_thread(queue);
128 irqreturn_t xenvif_interrupt(int irq, void *dev_id)
130 xenvif_tx_interrupt(irq, dev_id);
131 xenvif_rx_interrupt(irq, dev_id);
136 int xenvif_queue_stopped(struct xenvif_queue *queue)
138 struct net_device *dev = queue->vif->dev;
139 unsigned int id = queue->id;
140 return netif_tx_queue_stopped(netdev_get_tx_queue(dev, id));
143 void xenvif_wake_queue(struct xenvif_queue *queue)
145 struct net_device *dev = queue->vif->dev;
146 unsigned int id = queue->id;
147 netif_tx_wake_queue(netdev_get_tx_queue(dev, id));
150 static u16 xenvif_select_queue(struct net_device *dev, struct sk_buff *skb,
151 struct net_device *sb_dev,
152 select_queue_fallback_t fallback)
154 struct xenvif *vif = netdev_priv(dev);
155 unsigned int size = vif->hash.size;
157 if (vif->hash.alg == XEN_NETIF_CTRL_HASH_ALGORITHM_NONE)
158 return fallback(dev, skb, NULL) % dev->real_num_tx_queues;
160 xenvif_set_skb_hash(vif, skb);
163 return skb_get_hash_raw(skb) % dev->real_num_tx_queues;
165 return vif->hash.mapping[vif->hash.mapping_sel]
166 [skb_get_hash_raw(skb) % size];
170 xenvif_start_xmit(struct sk_buff *skb, struct net_device *dev)
172 struct xenvif *vif = netdev_priv(dev);
173 struct xenvif_queue *queue = NULL;
174 unsigned int num_queues;
176 struct xenvif_rx_cb *cb;
178 BUG_ON(skb->dev != dev);
180 /* Drop the packet if queues are not set up.
181 * This handler should be called inside an RCU read section
182 * so we don't need to enter it here explicitly.
184 num_queues = READ_ONCE(vif->num_queues);
188 /* Obtain the queue to be used to transmit this packet */
189 index = skb_get_queue_mapping(skb);
190 if (index >= num_queues) {
191 pr_warn_ratelimited("Invalid queue %hu for packet on interface %s\n",
192 index, vif->dev->name);
195 queue = &vif->queues[index];
197 /* Drop the packet if queue is not ready */
198 if (queue->task == NULL ||
199 queue->dealloc_task == NULL ||
200 !xenvif_schedulable(vif))
203 if (vif->multicast_control && skb->pkt_type == PACKET_MULTICAST) {
204 struct ethhdr *eth = (struct ethhdr *)skb->data;
206 if (!xenvif_mcast_match(vif, eth->h_dest))
210 cb = XENVIF_RX_CB(skb);
211 cb->expires = jiffies + vif->drain_timeout;
213 /* If there is no hash algorithm configured then make sure there
214 * is no hash information in the socket buffer otherwise it
215 * would be incorrectly forwarded to the frontend.
217 if (vif->hash.alg == XEN_NETIF_CTRL_HASH_ALGORITHM_NONE)
220 xenvif_rx_queue_tail(queue, skb);
221 xenvif_kick_thread(queue);
226 vif->dev->stats.tx_dropped++;
231 static struct net_device_stats *xenvif_get_stats(struct net_device *dev)
233 struct xenvif *vif = netdev_priv(dev);
234 struct xenvif_queue *queue = NULL;
235 unsigned int num_queues;
243 num_queues = READ_ONCE(vif->num_queues);
245 /* Aggregate tx and rx stats from each queue */
246 for (index = 0; index < num_queues; ++index) {
247 queue = &vif->queues[index];
248 rx_bytes += queue->stats.rx_bytes;
249 rx_packets += queue->stats.rx_packets;
250 tx_bytes += queue->stats.tx_bytes;
251 tx_packets += queue->stats.tx_packets;
256 vif->dev->stats.rx_bytes = rx_bytes;
257 vif->dev->stats.rx_packets = rx_packets;
258 vif->dev->stats.tx_bytes = tx_bytes;
259 vif->dev->stats.tx_packets = tx_packets;
261 return &vif->dev->stats;
264 static void xenvif_up(struct xenvif *vif)
266 struct xenvif_queue *queue = NULL;
267 unsigned int num_queues = vif->num_queues;
268 unsigned int queue_index;
270 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
271 queue = &vif->queues[queue_index];
272 napi_enable(&queue->napi);
273 enable_irq(queue->tx_irq);
274 if (queue->tx_irq != queue->rx_irq)
275 enable_irq(queue->rx_irq);
276 xenvif_napi_schedule_or_enable_events(queue);
280 static void xenvif_down(struct xenvif *vif)
282 struct xenvif_queue *queue = NULL;
283 unsigned int num_queues = vif->num_queues;
284 unsigned int queue_index;
286 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
287 queue = &vif->queues[queue_index];
288 disable_irq(queue->tx_irq);
289 if (queue->tx_irq != queue->rx_irq)
290 disable_irq(queue->rx_irq);
291 napi_disable(&queue->napi);
292 del_timer_sync(&queue->credit_timeout);
296 static int xenvif_open(struct net_device *dev)
298 struct xenvif *vif = netdev_priv(dev);
299 if (test_bit(VIF_STATUS_CONNECTED, &vif->status))
301 netif_tx_start_all_queues(dev);
305 static int xenvif_close(struct net_device *dev)
307 struct xenvif *vif = netdev_priv(dev);
308 if (test_bit(VIF_STATUS_CONNECTED, &vif->status))
310 netif_tx_stop_all_queues(dev);
314 static int xenvif_change_mtu(struct net_device *dev, int mtu)
316 struct xenvif *vif = netdev_priv(dev);
317 int max = vif->can_sg ? ETH_MAX_MTU - VLAN_ETH_HLEN : ETH_DATA_LEN;
325 static netdev_features_t xenvif_fix_features(struct net_device *dev,
326 netdev_features_t features)
328 struct xenvif *vif = netdev_priv(dev);
331 features &= ~NETIF_F_SG;
332 if (~(vif->gso_mask) & GSO_BIT(TCPV4))
333 features &= ~NETIF_F_TSO;
334 if (~(vif->gso_mask) & GSO_BIT(TCPV6))
335 features &= ~NETIF_F_TSO6;
337 features &= ~NETIF_F_IP_CSUM;
339 features &= ~NETIF_F_IPV6_CSUM;
344 static const struct xenvif_stat {
345 char name[ETH_GSTRING_LEN];
349 "rx_gso_checksum_fixup",
350 offsetof(struct xenvif_stats, rx_gso_checksum_fixup)
352 /* If (sent != success + fail), there are probably packets never
357 offsetof(struct xenvif_stats, tx_zerocopy_sent),
360 "tx_zerocopy_success",
361 offsetof(struct xenvif_stats, tx_zerocopy_success),
365 offsetof(struct xenvif_stats, tx_zerocopy_fail)
367 /* Number of packets exceeding MAX_SKB_FRAG slots. You should use
368 * a guest with the same MAX_SKB_FRAG
372 offsetof(struct xenvif_stats, tx_frag_overflow)
376 static int xenvif_get_sset_count(struct net_device *dev, int string_set)
378 switch (string_set) {
380 return ARRAY_SIZE(xenvif_stats);
386 static void xenvif_get_ethtool_stats(struct net_device *dev,
387 struct ethtool_stats *stats, u64 * data)
389 struct xenvif *vif = netdev_priv(dev);
390 unsigned int num_queues;
392 unsigned int queue_index;
395 num_queues = READ_ONCE(vif->num_queues);
397 for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++) {
398 unsigned long accum = 0;
399 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
400 void *vif_stats = &vif->queues[queue_index].stats;
401 accum += *(unsigned long *)(vif_stats + xenvif_stats[i].offset);
409 static void xenvif_get_strings(struct net_device *dev, u32 stringset, u8 * data)
415 for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++)
416 memcpy(data + i * ETH_GSTRING_LEN,
417 xenvif_stats[i].name, ETH_GSTRING_LEN);
422 static const struct ethtool_ops xenvif_ethtool_ops = {
423 .get_link = ethtool_op_get_link,
425 .get_sset_count = xenvif_get_sset_count,
426 .get_ethtool_stats = xenvif_get_ethtool_stats,
427 .get_strings = xenvif_get_strings,
430 static const struct net_device_ops xenvif_netdev_ops = {
431 .ndo_select_queue = xenvif_select_queue,
432 .ndo_start_xmit = xenvif_start_xmit,
433 .ndo_get_stats = xenvif_get_stats,
434 .ndo_open = xenvif_open,
435 .ndo_stop = xenvif_close,
436 .ndo_change_mtu = xenvif_change_mtu,
437 .ndo_fix_features = xenvif_fix_features,
438 .ndo_set_mac_address = eth_mac_addr,
439 .ndo_validate_addr = eth_validate_addr,
442 struct xenvif *xenvif_alloc(struct device *parent, domid_t domid,
446 struct net_device *dev;
448 char name[IFNAMSIZ] = {};
450 snprintf(name, IFNAMSIZ - 1, "vif%u.%u", domid, handle);
451 /* Allocate a netdev with the max. supported number of queues.
452 * When the guest selects the desired number, it will be updated
453 * via netif_set_real_num_*_queues().
455 dev = alloc_netdev_mq(sizeof(struct xenvif), name, NET_NAME_UNKNOWN,
456 ether_setup, xenvif_max_queues);
458 pr_warn("Could not allocate netdev for %s\n", name);
459 return ERR_PTR(-ENOMEM);
462 SET_NETDEV_DEV(dev, parent);
464 vif = netdev_priv(dev);
467 vif->handle = handle;
471 vif->disabled = false;
472 vif->drain_timeout = msecs_to_jiffies(rx_drain_timeout_msecs);
473 vif->stall_timeout = msecs_to_jiffies(rx_stall_timeout_msecs);
475 /* Start out with no queues. */
479 spin_lock_init(&vif->lock);
480 INIT_LIST_HEAD(&vif->fe_mcast_addr);
482 dev->netdev_ops = &xenvif_netdev_ops;
483 dev->hw_features = NETIF_F_SG |
484 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
485 NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_FRAGLIST;
486 dev->features = dev->hw_features | NETIF_F_RXCSUM;
487 dev->ethtool_ops = &xenvif_ethtool_ops;
489 dev->tx_queue_len = XENVIF_QUEUE_LENGTH;
491 dev->min_mtu = ETH_MIN_MTU;
492 dev->max_mtu = ETH_MAX_MTU - VLAN_ETH_HLEN;
495 * Initialise a dummy MAC address. We choose the numerically
496 * largest non-broadcast address to prevent the address getting
497 * stolen by an Ethernet bridge for STP purposes.
498 * (FE:FF:FF:FF:FF:FF)
500 eth_broadcast_addr(dev->dev_addr);
501 dev->dev_addr[0] &= ~0x01;
503 netif_carrier_off(dev);
505 err = register_netdev(dev);
507 netdev_warn(dev, "Could not register device: err=%d\n", err);
512 netdev_dbg(dev, "Successfully created xenvif\n");
514 __module_get(THIS_MODULE);
519 int xenvif_init_queue(struct xenvif_queue *queue)
523 queue->credit_bytes = queue->remaining_credit = ~0UL;
524 queue->credit_usec = 0UL;
525 timer_setup(&queue->credit_timeout, xenvif_tx_credit_callback, 0);
526 queue->credit_window_start = get_jiffies_64();
528 queue->rx_queue_max = XENVIF_RX_QUEUE_BYTES;
530 skb_queue_head_init(&queue->rx_queue);
531 skb_queue_head_init(&queue->tx_queue);
533 queue->pending_cons = 0;
534 queue->pending_prod = MAX_PENDING_REQS;
535 for (i = 0; i < MAX_PENDING_REQS; ++i)
536 queue->pending_ring[i] = i;
538 spin_lock_init(&queue->callback_lock);
539 spin_lock_init(&queue->response_lock);
541 /* If ballooning is disabled, this will consume real memory, so you
542 * better enable it. The long term solution would be to use just a
543 * bunch of valid page descriptors, without dependency on ballooning
545 err = gnttab_alloc_pages(MAX_PENDING_REQS,
548 netdev_err(queue->vif->dev, "Could not reserve mmap_pages\n");
552 for (i = 0; i < MAX_PENDING_REQS; i++) {
553 queue->pending_tx_info[i].callback_struct = (struct ubuf_info)
554 { .callback = xenvif_zerocopy_callback,
557 queue->grant_tx_handle[i] = NETBACK_INVALID_HANDLE;
563 void xenvif_carrier_on(struct xenvif *vif)
566 if (!vif->can_sg && vif->dev->mtu > ETH_DATA_LEN)
567 dev_set_mtu(vif->dev, ETH_DATA_LEN);
568 netdev_update_features(vif->dev);
569 set_bit(VIF_STATUS_CONNECTED, &vif->status);
570 if (netif_running(vif->dev))
575 int xenvif_connect_ctrl(struct xenvif *vif, grant_ref_t ring_ref,
578 struct net_device *dev = vif->dev;
580 struct xen_netif_ctrl_sring *shared;
583 err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(vif),
584 &ring_ref, 1, &addr);
588 shared = (struct xen_netif_ctrl_sring *)addr;
589 BACK_RING_INIT(&vif->ctrl, shared, XEN_PAGE_SIZE);
591 err = bind_interdomain_evtchn_to_irq(vif->domid, evtchn);
597 xenvif_init_hash(vif);
599 err = request_threaded_irq(vif->ctrl_irq, NULL, xenvif_ctrl_irq_fn,
600 IRQF_ONESHOT, "xen-netback-ctrl", vif);
602 pr_warn("Could not setup irq handler for %s\n", dev->name);
609 xenvif_deinit_hash(vif);
610 unbind_from_irqhandler(vif->ctrl_irq, vif);
614 xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif),
616 vif->ctrl.sring = NULL;
622 int xenvif_connect_data(struct xenvif_queue *queue,
623 unsigned long tx_ring_ref,
624 unsigned long rx_ring_ref,
625 unsigned int tx_evtchn,
626 unsigned int rx_evtchn)
628 struct task_struct *task;
631 BUG_ON(queue->tx_irq);
633 BUG_ON(queue->dealloc_task);
635 err = xenvif_map_frontend_data_rings(queue, tx_ring_ref,
640 init_waitqueue_head(&queue->wq);
641 init_waitqueue_head(&queue->dealloc_wq);
642 atomic_set(&queue->inflight_packets, 0);
644 netif_napi_add(queue->vif->dev, &queue->napi, xenvif_poll,
647 if (tx_evtchn == rx_evtchn) {
648 /* feature-split-event-channels == 0 */
649 err = bind_interdomain_evtchn_to_irqhandler(
650 queue->vif->domid, tx_evtchn, xenvif_interrupt, 0,
654 queue->tx_irq = queue->rx_irq = err;
655 disable_irq(queue->tx_irq);
657 /* feature-split-event-channels == 1 */
658 snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
659 "%s-tx", queue->name);
660 err = bind_interdomain_evtchn_to_irqhandler(
661 queue->vif->domid, tx_evtchn, xenvif_tx_interrupt, 0,
662 queue->tx_irq_name, queue);
666 disable_irq(queue->tx_irq);
668 snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
669 "%s-rx", queue->name);
670 err = bind_interdomain_evtchn_to_irqhandler(
671 queue->vif->domid, rx_evtchn, xenvif_rx_interrupt, 0,
672 queue->rx_irq_name, queue);
676 disable_irq(queue->rx_irq);
679 queue->stalled = true;
681 task = kthread_create(xenvif_kthread_guest_rx,
682 (void *)queue, "%s-guest-rx", queue->name);
684 pr_warn("Could not allocate kthread for %s\n", queue->name);
689 get_task_struct(task);
691 task = kthread_create(xenvif_dealloc_kthread,
692 (void *)queue, "%s-dealloc", queue->name);
694 pr_warn("Could not allocate kthread for %s\n", queue->name);
698 queue->dealloc_task = task;
700 wake_up_process(queue->task);
701 wake_up_process(queue->dealloc_task);
706 unbind_from_irqhandler(queue->rx_irq, queue);
709 unbind_from_irqhandler(queue->tx_irq, queue);
712 xenvif_unmap_frontend_data_rings(queue);
713 netif_napi_del(&queue->napi);
715 module_put(THIS_MODULE);
719 void xenvif_carrier_off(struct xenvif *vif)
721 struct net_device *dev = vif->dev;
724 if (test_and_clear_bit(VIF_STATUS_CONNECTED, &vif->status)) {
725 netif_carrier_off(dev); /* discard queued packets */
726 if (netif_running(dev))
732 void xenvif_disconnect_data(struct xenvif *vif)
734 struct xenvif_queue *queue = NULL;
735 unsigned int num_queues = vif->num_queues;
736 unsigned int queue_index;
738 xenvif_carrier_off(vif);
740 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
741 queue = &vif->queues[queue_index];
743 netif_napi_del(&queue->napi);
746 kthread_stop(queue->task);
747 put_task_struct(queue->task);
751 if (queue->dealloc_task) {
752 kthread_stop(queue->dealloc_task);
753 queue->dealloc_task = NULL;
757 if (queue->tx_irq == queue->rx_irq)
758 unbind_from_irqhandler(queue->tx_irq, queue);
760 unbind_from_irqhandler(queue->tx_irq, queue);
761 unbind_from_irqhandler(queue->rx_irq, queue);
766 xenvif_unmap_frontend_data_rings(queue);
769 xenvif_mcast_addr_list_free(vif);
772 void xenvif_disconnect_ctrl(struct xenvif *vif)
775 xenvif_deinit_hash(vif);
776 unbind_from_irqhandler(vif->ctrl_irq, vif);
780 if (vif->ctrl.sring) {
781 xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif),
783 vif->ctrl.sring = NULL;
787 /* Reverse the relevant parts of xenvif_init_queue().
788 * Used for queue teardown from xenvif_free(), and on the
789 * error handling paths in xenbus.c:connect().
791 void xenvif_deinit_queue(struct xenvif_queue *queue)
793 gnttab_free_pages(MAX_PENDING_REQS, queue->mmap_pages);
796 void xenvif_free(struct xenvif *vif)
798 struct xenvif_queue *queues = vif->queues;
799 unsigned int num_queues = vif->num_queues;
800 unsigned int queue_index;
802 unregister_netdev(vif->dev);
803 free_netdev(vif->dev);
805 for (queue_index = 0; queue_index < num_queues; ++queue_index)
806 xenvif_deinit_queue(&queues[queue_index]);
809 module_put(THIS_MODULE);