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 = vif->num_queues;
170 struct xenvif_rx_cb *cb;
172 BUG_ON(skb->dev != dev);
174 /* Drop the packet if queues are not set up */
178 /* Obtain the queue to be used to transmit this packet */
179 index = skb_get_queue_mapping(skb);
180 if (index >= num_queues) {
181 pr_warn_ratelimited("Invalid queue %hu for packet on interface %s\n.",
182 index, vif->dev->name);
185 queue = &vif->queues[index];
187 /* Drop the packet if queue is not ready */
188 if (queue->task == NULL ||
189 queue->dealloc_task == NULL ||
190 !xenvif_schedulable(vif))
193 if (vif->multicast_control && skb->pkt_type == PACKET_MULTICAST) {
194 struct ethhdr *eth = (struct ethhdr *)skb->data;
196 if (!xenvif_mcast_match(vif, eth->h_dest))
200 cb = XENVIF_RX_CB(skb);
201 cb->expires = jiffies + vif->drain_timeout;
203 /* If there is no hash algorithm configured then make sure there
204 * is no hash information in the socket buffer otherwise it
205 * would be incorrectly forwarded to the frontend.
207 if (vif->hash.alg == XEN_NETIF_CTRL_HASH_ALGORITHM_NONE)
210 xenvif_rx_queue_tail(queue, skb);
211 xenvif_kick_thread(queue);
216 vif->dev->stats.tx_dropped++;
221 static struct net_device_stats *xenvif_get_stats(struct net_device *dev)
223 struct xenvif *vif = netdev_priv(dev);
224 struct xenvif_queue *queue = NULL;
231 spin_lock(&vif->lock);
232 if (vif->queues == NULL)
235 /* Aggregate tx and rx stats from each queue */
236 for (index = 0; index < vif->num_queues; ++index) {
237 queue = &vif->queues[index];
238 rx_bytes += queue->stats.rx_bytes;
239 rx_packets += queue->stats.rx_packets;
240 tx_bytes += queue->stats.tx_bytes;
241 tx_packets += queue->stats.tx_packets;
245 spin_unlock(&vif->lock);
247 vif->dev->stats.rx_bytes = rx_bytes;
248 vif->dev->stats.rx_packets = rx_packets;
249 vif->dev->stats.tx_bytes = tx_bytes;
250 vif->dev->stats.tx_packets = tx_packets;
252 return &vif->dev->stats;
255 static void xenvif_up(struct xenvif *vif)
257 struct xenvif_queue *queue = NULL;
258 unsigned int num_queues = vif->num_queues;
259 unsigned int queue_index;
261 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
262 queue = &vif->queues[queue_index];
263 napi_enable(&queue->napi);
264 enable_irq(queue->tx_irq);
265 if (queue->tx_irq != queue->rx_irq)
266 enable_irq(queue->rx_irq);
267 xenvif_napi_schedule_or_enable_events(queue);
271 static void xenvif_down(struct xenvif *vif)
273 struct xenvif_queue *queue = NULL;
274 unsigned int num_queues = vif->num_queues;
275 unsigned int queue_index;
277 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
278 queue = &vif->queues[queue_index];
279 disable_irq(queue->tx_irq);
280 if (queue->tx_irq != queue->rx_irq)
281 disable_irq(queue->rx_irq);
282 napi_disable(&queue->napi);
283 del_timer_sync(&queue->credit_timeout);
287 static int xenvif_open(struct net_device *dev)
289 struct xenvif *vif = netdev_priv(dev);
290 if (test_bit(VIF_STATUS_CONNECTED, &vif->status))
292 netif_tx_start_all_queues(dev);
296 static int xenvif_close(struct net_device *dev)
298 struct xenvif *vif = netdev_priv(dev);
299 if (test_bit(VIF_STATUS_CONNECTED, &vif->status))
301 netif_tx_stop_all_queues(dev);
305 static int xenvif_change_mtu(struct net_device *dev, int mtu)
307 struct xenvif *vif = netdev_priv(dev);
308 int max = vif->can_sg ? ETH_MAX_MTU - VLAN_ETH_HLEN : ETH_DATA_LEN;
316 static netdev_features_t xenvif_fix_features(struct net_device *dev,
317 netdev_features_t features)
319 struct xenvif *vif = netdev_priv(dev);
322 features &= ~NETIF_F_SG;
323 if (~(vif->gso_mask) & GSO_BIT(TCPV4))
324 features &= ~NETIF_F_TSO;
325 if (~(vif->gso_mask) & GSO_BIT(TCPV6))
326 features &= ~NETIF_F_TSO6;
328 features &= ~NETIF_F_IP_CSUM;
330 features &= ~NETIF_F_IPV6_CSUM;
335 static const struct xenvif_stat {
336 char name[ETH_GSTRING_LEN];
340 "rx_gso_checksum_fixup",
341 offsetof(struct xenvif_stats, rx_gso_checksum_fixup)
343 /* If (sent != success + fail), there are probably packets never
348 offsetof(struct xenvif_stats, tx_zerocopy_sent),
351 "tx_zerocopy_success",
352 offsetof(struct xenvif_stats, tx_zerocopy_success),
356 offsetof(struct xenvif_stats, tx_zerocopy_fail)
358 /* Number of packets exceeding MAX_SKB_FRAG slots. You should use
359 * a guest with the same MAX_SKB_FRAG
363 offsetof(struct xenvif_stats, tx_frag_overflow)
367 static int xenvif_get_sset_count(struct net_device *dev, int string_set)
369 switch (string_set) {
371 return ARRAY_SIZE(xenvif_stats);
377 static void xenvif_get_ethtool_stats(struct net_device *dev,
378 struct ethtool_stats *stats, u64 * data)
380 struct xenvif *vif = netdev_priv(dev);
381 unsigned int num_queues = vif->num_queues;
383 unsigned int queue_index;
385 for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++) {
386 unsigned long accum = 0;
387 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
388 void *vif_stats = &vif->queues[queue_index].stats;
389 accum += *(unsigned long *)(vif_stats + xenvif_stats[i].offset);
395 static void xenvif_get_strings(struct net_device *dev, u32 stringset, u8 * data)
401 for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++)
402 memcpy(data + i * ETH_GSTRING_LEN,
403 xenvif_stats[i].name, ETH_GSTRING_LEN);
408 static const struct ethtool_ops xenvif_ethtool_ops = {
409 .get_link = ethtool_op_get_link,
411 .get_sset_count = xenvif_get_sset_count,
412 .get_ethtool_stats = xenvif_get_ethtool_stats,
413 .get_strings = xenvif_get_strings,
416 static const struct net_device_ops xenvif_netdev_ops = {
417 .ndo_select_queue = xenvif_select_queue,
418 .ndo_start_xmit = xenvif_start_xmit,
419 .ndo_get_stats = xenvif_get_stats,
420 .ndo_open = xenvif_open,
421 .ndo_stop = xenvif_close,
422 .ndo_change_mtu = xenvif_change_mtu,
423 .ndo_fix_features = xenvif_fix_features,
424 .ndo_set_mac_address = eth_mac_addr,
425 .ndo_validate_addr = eth_validate_addr,
428 struct xenvif *xenvif_alloc(struct device *parent, domid_t domid,
432 struct net_device *dev;
434 char name[IFNAMSIZ] = {};
436 snprintf(name, IFNAMSIZ - 1, "vif%u.%u", domid, handle);
437 /* Allocate a netdev with the max. supported number of queues.
438 * When the guest selects the desired number, it will be updated
439 * via netif_set_real_num_*_queues().
441 dev = alloc_netdev_mq(sizeof(struct xenvif), name, NET_NAME_UNKNOWN,
442 ether_setup, xenvif_max_queues);
444 pr_warn("Could not allocate netdev for %s\n", name);
445 return ERR_PTR(-ENOMEM);
448 SET_NETDEV_DEV(dev, parent);
450 vif = netdev_priv(dev);
453 vif->handle = handle;
457 vif->disabled = false;
458 vif->drain_timeout = msecs_to_jiffies(rx_drain_timeout_msecs);
459 vif->stall_timeout = msecs_to_jiffies(rx_stall_timeout_msecs);
461 /* Start out with no queues. */
465 spin_lock_init(&vif->lock);
466 INIT_LIST_HEAD(&vif->fe_mcast_addr);
468 dev->netdev_ops = &xenvif_netdev_ops;
469 dev->hw_features = NETIF_F_SG |
470 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
471 NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_FRAGLIST;
472 dev->features = dev->hw_features | NETIF_F_RXCSUM;
473 dev->ethtool_ops = &xenvif_ethtool_ops;
475 dev->tx_queue_len = XENVIF_QUEUE_LENGTH;
478 dev->max_mtu = ETH_MAX_MTU - VLAN_ETH_HLEN;
481 * Initialise a dummy MAC address. We choose the numerically
482 * largest non-broadcast address to prevent the address getting
483 * stolen by an Ethernet bridge for STP purposes.
484 * (FE:FF:FF:FF:FF:FF)
486 eth_broadcast_addr(dev->dev_addr);
487 dev->dev_addr[0] &= ~0x01;
489 netif_carrier_off(dev);
491 err = register_netdev(dev);
493 netdev_warn(dev, "Could not register device: err=%d\n", err);
498 netdev_dbg(dev, "Successfully created xenvif\n");
500 __module_get(THIS_MODULE);
505 int xenvif_init_queue(struct xenvif_queue *queue)
509 queue->credit_bytes = queue->remaining_credit = ~0UL;
510 queue->credit_usec = 0UL;
511 init_timer(&queue->credit_timeout);
512 queue->credit_timeout.function = xenvif_tx_credit_callback;
513 queue->credit_window_start = get_jiffies_64();
515 queue->rx_queue_max = XENVIF_RX_QUEUE_BYTES;
517 skb_queue_head_init(&queue->rx_queue);
518 skb_queue_head_init(&queue->tx_queue);
520 queue->pending_cons = 0;
521 queue->pending_prod = MAX_PENDING_REQS;
522 for (i = 0; i < MAX_PENDING_REQS; ++i)
523 queue->pending_ring[i] = i;
525 spin_lock_init(&queue->callback_lock);
526 spin_lock_init(&queue->response_lock);
528 /* If ballooning is disabled, this will consume real memory, so you
529 * better enable it. The long term solution would be to use just a
530 * bunch of valid page descriptors, without dependency on ballooning
532 err = gnttab_alloc_pages(MAX_PENDING_REQS,
535 netdev_err(queue->vif->dev, "Could not reserve mmap_pages\n");
539 for (i = 0; i < MAX_PENDING_REQS; i++) {
540 queue->pending_tx_info[i].callback_struct = (struct ubuf_info)
541 { .callback = xenvif_zerocopy_callback,
544 queue->grant_tx_handle[i] = NETBACK_INVALID_HANDLE;
550 void xenvif_carrier_on(struct xenvif *vif)
553 if (!vif->can_sg && vif->dev->mtu > ETH_DATA_LEN)
554 dev_set_mtu(vif->dev, ETH_DATA_LEN);
555 netdev_update_features(vif->dev);
556 set_bit(VIF_STATUS_CONNECTED, &vif->status);
557 if (netif_running(vif->dev))
562 int xenvif_connect_ctrl(struct xenvif *vif, grant_ref_t ring_ref,
565 struct net_device *dev = vif->dev;
567 struct xen_netif_ctrl_sring *shared;
570 err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(vif),
571 &ring_ref, 1, &addr);
575 shared = (struct xen_netif_ctrl_sring *)addr;
576 BACK_RING_INIT(&vif->ctrl, shared, XEN_PAGE_SIZE);
578 err = bind_interdomain_evtchn_to_irq(vif->domid, evtchn);
584 xenvif_init_hash(vif);
586 err = request_threaded_irq(vif->ctrl_irq, NULL, xenvif_ctrl_irq_fn,
587 IRQF_ONESHOT, "xen-netback-ctrl", vif);
589 pr_warn("Could not setup irq handler for %s\n", dev->name);
596 xenvif_deinit_hash(vif);
597 unbind_from_irqhandler(vif->ctrl_irq, vif);
601 xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif),
603 vif->ctrl.sring = NULL;
609 int xenvif_connect_data(struct xenvif_queue *queue,
610 unsigned long tx_ring_ref,
611 unsigned long rx_ring_ref,
612 unsigned int tx_evtchn,
613 unsigned int rx_evtchn)
615 struct task_struct *task;
618 BUG_ON(queue->tx_irq);
620 BUG_ON(queue->dealloc_task);
622 err = xenvif_map_frontend_data_rings(queue, tx_ring_ref,
627 init_waitqueue_head(&queue->wq);
628 init_waitqueue_head(&queue->dealloc_wq);
629 atomic_set(&queue->inflight_packets, 0);
631 netif_napi_add(queue->vif->dev, &queue->napi, xenvif_poll,
634 if (tx_evtchn == rx_evtchn) {
635 /* feature-split-event-channels == 0 */
636 err = bind_interdomain_evtchn_to_irqhandler(
637 queue->vif->domid, tx_evtchn, xenvif_interrupt, 0,
641 queue->tx_irq = queue->rx_irq = err;
642 disable_irq(queue->tx_irq);
644 /* feature-split-event-channels == 1 */
645 snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
646 "%s-tx", queue->name);
647 err = bind_interdomain_evtchn_to_irqhandler(
648 queue->vif->domid, tx_evtchn, xenvif_tx_interrupt, 0,
649 queue->tx_irq_name, queue);
653 disable_irq(queue->tx_irq);
655 snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
656 "%s-rx", queue->name);
657 err = bind_interdomain_evtchn_to_irqhandler(
658 queue->vif->domid, rx_evtchn, xenvif_rx_interrupt, 0,
659 queue->rx_irq_name, queue);
663 disable_irq(queue->rx_irq);
666 queue->stalled = true;
668 task = kthread_create(xenvif_kthread_guest_rx,
669 (void *)queue, "%s-guest-rx", queue->name);
671 pr_warn("Could not allocate kthread for %s\n", queue->name);
676 get_task_struct(task);
678 task = kthread_create(xenvif_dealloc_kthread,
679 (void *)queue, "%s-dealloc", queue->name);
681 pr_warn("Could not allocate kthread for %s\n", queue->name);
685 queue->dealloc_task = task;
687 wake_up_process(queue->task);
688 wake_up_process(queue->dealloc_task);
693 unbind_from_irqhandler(queue->rx_irq, queue);
696 unbind_from_irqhandler(queue->tx_irq, queue);
699 xenvif_unmap_frontend_data_rings(queue);
700 netif_napi_del(&queue->napi);
702 module_put(THIS_MODULE);
706 void xenvif_carrier_off(struct xenvif *vif)
708 struct net_device *dev = vif->dev;
711 if (test_and_clear_bit(VIF_STATUS_CONNECTED, &vif->status)) {
712 netif_carrier_off(dev); /* discard queued packets */
713 if (netif_running(dev))
719 void xenvif_disconnect_data(struct xenvif *vif)
721 struct xenvif_queue *queue = NULL;
722 unsigned int num_queues = vif->num_queues;
723 unsigned int queue_index;
725 xenvif_carrier_off(vif);
727 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
728 queue = &vif->queues[queue_index];
730 netif_napi_del(&queue->napi);
733 kthread_stop(queue->task);
734 put_task_struct(queue->task);
738 if (queue->dealloc_task) {
739 kthread_stop(queue->dealloc_task);
740 queue->dealloc_task = NULL;
744 if (queue->tx_irq == queue->rx_irq)
745 unbind_from_irqhandler(queue->tx_irq, queue);
747 unbind_from_irqhandler(queue->tx_irq, queue);
748 unbind_from_irqhandler(queue->rx_irq, queue);
753 xenvif_unmap_frontend_data_rings(queue);
756 xenvif_mcast_addr_list_free(vif);
759 void xenvif_disconnect_ctrl(struct xenvif *vif)
762 xenvif_deinit_hash(vif);
763 unbind_from_irqhandler(vif->ctrl_irq, vif);
767 if (vif->ctrl.sring) {
768 xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif),
770 vif->ctrl.sring = NULL;
774 /* Reverse the relevant parts of xenvif_init_queue().
775 * Used for queue teardown from xenvif_free(), and on the
776 * error handling paths in xenbus.c:connect().
778 void xenvif_deinit_queue(struct xenvif_queue *queue)
780 gnttab_free_pages(MAX_PENDING_REQS, queue->mmap_pages);
783 void xenvif_free(struct xenvif *vif)
785 struct xenvif_queue *queues = vif->queues;
786 unsigned int num_queues = vif->num_queues;
787 unsigned int queue_index;
789 unregister_netdev(vif->dev);
790 free_netdev(vif->dev);
792 for (queue_index = 0; queue_index < num_queues; ++queue_index)
793 xenvif_deinit_queue(&queues[queue_index]);
796 module_put(THIS_MODULE);