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/ethtool.h>
35 #include <linux/rtnetlink.h>
36 #include <linux/if_vlan.h>
37 #include <linux/vmalloc.h>
39 #include <xen/events.h>
40 #include <asm/xen/hypercall.h>
41 #include <xen/balloon.h>
43 #define XENVIF_QUEUE_LENGTH 32
44 #define XENVIF_NAPI_WEIGHT 64
46 static inline void xenvif_stop_queue(struct xenvif_queue *queue)
48 struct net_device *dev = queue->vif->dev;
50 if (!queue->vif->can_queue)
53 netif_tx_stop_queue(netdev_get_tx_queue(dev, queue->id));
56 int xenvif_schedulable(struct xenvif *vif)
58 return netif_running(vif->dev) &&
59 test_bit(VIF_STATUS_CONNECTED, &vif->status);
62 static irqreturn_t xenvif_tx_interrupt(int irq, void *dev_id)
64 struct xenvif_queue *queue = dev_id;
66 if (RING_HAS_UNCONSUMED_REQUESTS(&queue->tx))
67 napi_schedule(&queue->napi);
72 int xenvif_poll(struct napi_struct *napi, int budget)
74 struct xenvif_queue *queue =
75 container_of(napi, struct xenvif_queue, napi);
78 /* This vif is rogue, we pretend we've there is nothing to do
79 * for this vif to deschedule it from NAPI. But this interface
80 * will be turned off in thread context later.
82 if (unlikely(queue->vif->disabled)) {
87 work_done = xenvif_tx_action(queue, budget);
89 if (work_done < budget) {
91 xenvif_napi_schedule_or_enable_events(queue);
97 static irqreturn_t xenvif_rx_interrupt(int irq, void *dev_id)
99 struct xenvif_queue *queue = dev_id;
101 xenvif_kick_thread(queue);
106 irqreturn_t xenvif_interrupt(int irq, void *dev_id)
108 xenvif_tx_interrupt(irq, dev_id);
109 xenvif_rx_interrupt(irq, dev_id);
114 int xenvif_queue_stopped(struct xenvif_queue *queue)
116 struct net_device *dev = queue->vif->dev;
117 unsigned int id = queue->id;
118 return netif_tx_queue_stopped(netdev_get_tx_queue(dev, id));
121 void xenvif_wake_queue(struct xenvif_queue *queue)
123 struct net_device *dev = queue->vif->dev;
124 unsigned int id = queue->id;
125 netif_tx_wake_queue(netdev_get_tx_queue(dev, id));
128 /* Callback to wake the queue and drain it on timeout */
129 static void xenvif_wake_queue_callback(unsigned long data)
131 struct xenvif_queue *queue = (struct xenvif_queue *)data;
133 if (xenvif_queue_stopped(queue)) {
134 netdev_err(queue->vif->dev, "draining TX queue\n");
135 queue->rx_queue_purge = true;
136 xenvif_kick_thread(queue);
137 xenvif_wake_queue(queue);
141 static int xenvif_start_xmit(struct sk_buff *skb, struct net_device *dev)
143 struct xenvif *vif = netdev_priv(dev);
144 struct xenvif_queue *queue = NULL;
145 unsigned int num_queues = vif->num_queues;
147 int min_slots_needed;
149 BUG_ON(skb->dev != dev);
151 /* Drop the packet if queues are not set up */
155 /* Obtain the queue to be used to transmit this packet */
156 index = skb_get_queue_mapping(skb);
157 if (index >= num_queues) {
158 pr_warn_ratelimited("Invalid queue %hu for packet on interface %s\n.",
159 index, vif->dev->name);
162 queue = &vif->queues[index];
164 /* Drop the packet if queue is not ready */
165 if (queue->task == NULL ||
166 queue->dealloc_task == NULL ||
167 !xenvif_schedulable(vif))
170 /* At best we'll need one slot for the header and one for each
173 min_slots_needed = 1 + skb_shinfo(skb)->nr_frags;
175 /* If the skb is GSO then we'll also need an extra slot for the
181 /* If the skb can't possibly fit in the remaining slots
182 * then turn off the queue to give the ring a chance to
185 if (!xenvif_rx_ring_slots_available(queue, min_slots_needed)) {
186 queue->wake_queue.function = xenvif_wake_queue_callback;
187 queue->wake_queue.data = (unsigned long)queue;
188 xenvif_stop_queue(queue);
189 mod_timer(&queue->wake_queue,
190 jiffies + rx_drain_timeout_jiffies);
193 skb_queue_tail(&queue->rx_queue, skb);
194 xenvif_kick_thread(queue);
199 vif->dev->stats.tx_dropped++;
204 static struct net_device_stats *xenvif_get_stats(struct net_device *dev)
206 struct xenvif *vif = netdev_priv(dev);
207 struct xenvif_queue *queue = NULL;
208 unsigned int num_queues = vif->num_queues;
209 unsigned long rx_bytes = 0;
210 unsigned long rx_packets = 0;
211 unsigned long tx_bytes = 0;
212 unsigned long tx_packets = 0;
215 if (vif->queues == NULL)
218 /* Aggregate tx and rx stats from each queue */
219 for (index = 0; index < num_queues; ++index) {
220 queue = &vif->queues[index];
221 rx_bytes += queue->stats.rx_bytes;
222 rx_packets += queue->stats.rx_packets;
223 tx_bytes += queue->stats.tx_bytes;
224 tx_packets += queue->stats.tx_packets;
228 vif->dev->stats.rx_bytes = rx_bytes;
229 vif->dev->stats.rx_packets = rx_packets;
230 vif->dev->stats.tx_bytes = tx_bytes;
231 vif->dev->stats.tx_packets = tx_packets;
233 return &vif->dev->stats;
236 static void xenvif_up(struct xenvif *vif)
238 struct xenvif_queue *queue = NULL;
239 unsigned int num_queues = vif->num_queues;
240 unsigned int queue_index;
242 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
243 queue = &vif->queues[queue_index];
244 napi_enable(&queue->napi);
245 enable_irq(queue->tx_irq);
246 if (queue->tx_irq != queue->rx_irq)
247 enable_irq(queue->rx_irq);
248 xenvif_napi_schedule_or_enable_events(queue);
252 static void xenvif_down(struct xenvif *vif)
254 struct xenvif_queue *queue = NULL;
255 unsigned int num_queues = vif->num_queues;
256 unsigned int queue_index;
258 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
259 queue = &vif->queues[queue_index];
260 napi_disable(&queue->napi);
261 disable_irq(queue->tx_irq);
262 if (queue->tx_irq != queue->rx_irq)
263 disable_irq(queue->rx_irq);
264 del_timer_sync(&queue->credit_timeout);
268 static int xenvif_open(struct net_device *dev)
270 struct xenvif *vif = netdev_priv(dev);
271 if (test_bit(VIF_STATUS_CONNECTED, &vif->status))
273 netif_tx_start_all_queues(dev);
277 static int xenvif_close(struct net_device *dev)
279 struct xenvif *vif = netdev_priv(dev);
280 if (test_bit(VIF_STATUS_CONNECTED, &vif->status))
282 netif_tx_stop_all_queues(dev);
286 static int xenvif_change_mtu(struct net_device *dev, int mtu)
288 struct xenvif *vif = netdev_priv(dev);
289 int max = vif->can_sg ? 65535 - VLAN_ETH_HLEN : ETH_DATA_LEN;
297 static netdev_features_t xenvif_fix_features(struct net_device *dev,
298 netdev_features_t features)
300 struct xenvif *vif = netdev_priv(dev);
303 features &= ~NETIF_F_SG;
304 if (~(vif->gso_mask | vif->gso_prefix_mask) & GSO_BIT(TCPV4))
305 features &= ~NETIF_F_TSO;
306 if (~(vif->gso_mask | vif->gso_prefix_mask) & GSO_BIT(TCPV6))
307 features &= ~NETIF_F_TSO6;
309 features &= ~NETIF_F_IP_CSUM;
311 features &= ~NETIF_F_IPV6_CSUM;
316 static const struct xenvif_stat {
317 char name[ETH_GSTRING_LEN];
321 "rx_gso_checksum_fixup",
322 offsetof(struct xenvif_stats, rx_gso_checksum_fixup)
324 /* If (sent != success + fail), there are probably packets never
329 offsetof(struct xenvif_stats, tx_zerocopy_sent),
332 "tx_zerocopy_success",
333 offsetof(struct xenvif_stats, tx_zerocopy_success),
337 offsetof(struct xenvif_stats, tx_zerocopy_fail)
339 /* Number of packets exceeding MAX_SKB_FRAG slots. You should use
340 * a guest with the same MAX_SKB_FRAG
344 offsetof(struct xenvif_stats, tx_frag_overflow)
348 static int xenvif_get_sset_count(struct net_device *dev, int string_set)
350 switch (string_set) {
352 return ARRAY_SIZE(xenvif_stats);
358 static void xenvif_get_ethtool_stats(struct net_device *dev,
359 struct ethtool_stats *stats, u64 * data)
361 struct xenvif *vif = netdev_priv(dev);
362 unsigned int num_queues = vif->num_queues;
364 unsigned int queue_index;
365 struct xenvif_stats *vif_stats;
367 for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++) {
368 unsigned long accum = 0;
369 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
370 vif_stats = &vif->queues[queue_index].stats;
371 accum += *(unsigned long *)(vif_stats + xenvif_stats[i].offset);
377 static void xenvif_get_strings(struct net_device *dev, u32 stringset, u8 * data)
383 for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++)
384 memcpy(data + i * ETH_GSTRING_LEN,
385 xenvif_stats[i].name, ETH_GSTRING_LEN);
390 static const struct ethtool_ops xenvif_ethtool_ops = {
391 .get_link = ethtool_op_get_link,
393 .get_sset_count = xenvif_get_sset_count,
394 .get_ethtool_stats = xenvif_get_ethtool_stats,
395 .get_strings = xenvif_get_strings,
398 static const struct net_device_ops xenvif_netdev_ops = {
399 .ndo_start_xmit = xenvif_start_xmit,
400 .ndo_get_stats = xenvif_get_stats,
401 .ndo_open = xenvif_open,
402 .ndo_stop = xenvif_close,
403 .ndo_change_mtu = xenvif_change_mtu,
404 .ndo_fix_features = xenvif_fix_features,
405 .ndo_set_mac_address = eth_mac_addr,
406 .ndo_validate_addr = eth_validate_addr,
409 struct xenvif *xenvif_alloc(struct device *parent, domid_t domid,
413 struct net_device *dev;
415 char name[IFNAMSIZ] = {};
417 snprintf(name, IFNAMSIZ - 1, "vif%u.%u", domid, handle);
418 /* Allocate a netdev with the max. supported number of queues.
419 * When the guest selects the desired number, it will be updated
420 * via netif_set_real_num_*_queues().
422 dev = alloc_netdev_mq(sizeof(struct xenvif), name, NET_NAME_UNKNOWN,
423 ether_setup, xenvif_max_queues);
425 pr_warn("Could not allocate netdev for %s\n", name);
426 return ERR_PTR(-ENOMEM);
429 SET_NETDEV_DEV(dev, parent);
431 vif = netdev_priv(dev);
434 vif->handle = handle;
438 vif->disabled = false;
440 /* Start out with no queues. */
444 dev->netdev_ops = &xenvif_netdev_ops;
445 dev->hw_features = NETIF_F_SG |
446 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
447 NETIF_F_TSO | NETIF_F_TSO6;
448 dev->features = dev->hw_features | NETIF_F_RXCSUM;
449 dev->ethtool_ops = &xenvif_ethtool_ops;
451 dev->tx_queue_len = XENVIF_QUEUE_LENGTH;
454 * Initialise a dummy MAC address. We choose the numerically
455 * largest non-broadcast address to prevent the address getting
456 * stolen by an Ethernet bridge for STP purposes.
457 * (FE:FF:FF:FF:FF:FF)
459 memset(dev->dev_addr, 0xFF, ETH_ALEN);
460 dev->dev_addr[0] &= ~0x01;
462 netif_carrier_off(dev);
464 err = register_netdev(dev);
466 netdev_warn(dev, "Could not register device: err=%d\n", err);
471 netdev_dbg(dev, "Successfully created xenvif\n");
473 __module_get(THIS_MODULE);
478 int xenvif_init_queue(struct xenvif_queue *queue)
482 queue->credit_bytes = queue->remaining_credit = ~0UL;
483 queue->credit_usec = 0UL;
484 init_timer(&queue->credit_timeout);
485 queue->credit_window_start = get_jiffies_64();
487 skb_queue_head_init(&queue->rx_queue);
488 skb_queue_head_init(&queue->tx_queue);
490 queue->pending_cons = 0;
491 queue->pending_prod = MAX_PENDING_REQS;
492 for (i = 0; i < MAX_PENDING_REQS; ++i)
493 queue->pending_ring[i] = i;
495 spin_lock_init(&queue->callback_lock);
496 spin_lock_init(&queue->response_lock);
498 /* If ballooning is disabled, this will consume real memory, so you
499 * better enable it. The long term solution would be to use just a
500 * bunch of valid page descriptors, without dependency on ballooning
502 err = alloc_xenballooned_pages(MAX_PENDING_REQS,
506 netdev_err(queue->vif->dev, "Could not reserve mmap_pages\n");
510 for (i = 0; i < MAX_PENDING_REQS; i++) {
511 queue->pending_tx_info[i].callback_struct = (struct ubuf_info)
512 { .callback = xenvif_zerocopy_callback,
515 queue->grant_tx_handle[i] = NETBACK_INVALID_HANDLE;
518 init_timer(&queue->wake_queue);
520 netif_napi_add(queue->vif->dev, &queue->napi, xenvif_poll,
526 void xenvif_carrier_on(struct xenvif *vif)
529 if (!vif->can_sg && vif->dev->mtu > ETH_DATA_LEN)
530 dev_set_mtu(vif->dev, ETH_DATA_LEN);
531 netdev_update_features(vif->dev);
532 set_bit(VIF_STATUS_CONNECTED, &vif->status);
533 netif_carrier_on(vif->dev);
534 if (netif_running(vif->dev))
539 int xenvif_connect(struct xenvif_queue *queue, unsigned long tx_ring_ref,
540 unsigned long rx_ring_ref, unsigned int tx_evtchn,
541 unsigned int rx_evtchn)
543 struct task_struct *task;
546 BUG_ON(queue->tx_irq);
548 BUG_ON(queue->dealloc_task);
550 err = xenvif_map_frontend_rings(queue, tx_ring_ref, rx_ring_ref);
554 init_waitqueue_head(&queue->wq);
555 init_waitqueue_head(&queue->dealloc_wq);
557 if (tx_evtchn == rx_evtchn) {
558 /* feature-split-event-channels == 0 */
559 err = bind_interdomain_evtchn_to_irqhandler(
560 queue->vif->domid, tx_evtchn, xenvif_interrupt, 0,
564 queue->tx_irq = queue->rx_irq = err;
565 disable_irq(queue->tx_irq);
567 /* feature-split-event-channels == 1 */
568 snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
569 "%s-tx", queue->name);
570 err = bind_interdomain_evtchn_to_irqhandler(
571 queue->vif->domid, tx_evtchn, xenvif_tx_interrupt, 0,
572 queue->tx_irq_name, queue);
576 disable_irq(queue->tx_irq);
578 snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
579 "%s-rx", queue->name);
580 err = bind_interdomain_evtchn_to_irqhandler(
581 queue->vif->domid, rx_evtchn, xenvif_rx_interrupt, 0,
582 queue->rx_irq_name, queue);
586 disable_irq(queue->rx_irq);
589 task = kthread_create(xenvif_kthread_guest_rx,
590 (void *)queue, "%s-guest-rx", queue->name);
592 pr_warn("Could not allocate kthread for %s\n", queue->name);
598 task = kthread_create(xenvif_dealloc_kthread,
599 (void *)queue, "%s-dealloc", queue->name);
601 pr_warn("Could not allocate kthread for %s\n", queue->name);
605 queue->dealloc_task = task;
607 wake_up_process(queue->task);
608 wake_up_process(queue->dealloc_task);
613 unbind_from_irqhandler(queue->rx_irq, queue);
616 unbind_from_irqhandler(queue->tx_irq, queue);
619 xenvif_unmap_frontend_rings(queue);
621 module_put(THIS_MODULE);
625 void xenvif_carrier_off(struct xenvif *vif)
627 struct net_device *dev = vif->dev;
630 if (test_and_clear_bit(VIF_STATUS_CONNECTED, &vif->status)) {
631 netif_carrier_off(dev); /* discard queued packets */
632 if (netif_running(dev))
638 static void xenvif_wait_unmap_timeout(struct xenvif_queue *queue,
639 unsigned int worst_case_skb_lifetime)
641 int i, unmap_timeout = 0;
643 for (i = 0; i < MAX_PENDING_REQS; ++i) {
644 if (queue->grant_tx_handle[i] != NETBACK_INVALID_HANDLE) {
646 schedule_timeout(msecs_to_jiffies(1000));
647 if (unmap_timeout > worst_case_skb_lifetime &&
649 netdev_err(queue->vif->dev,
650 "Page still granted! Index: %x\n",
657 void xenvif_disconnect(struct xenvif *vif)
659 struct xenvif_queue *queue = NULL;
660 unsigned int num_queues = vif->num_queues;
661 unsigned int queue_index;
663 xenvif_carrier_off(vif);
665 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
666 queue = &vif->queues[queue_index];
669 del_timer_sync(&queue->wake_queue);
670 kthread_stop(queue->task);
674 if (queue->dealloc_task) {
675 kthread_stop(queue->dealloc_task);
676 queue->dealloc_task = NULL;
680 if (queue->tx_irq == queue->rx_irq)
681 unbind_from_irqhandler(queue->tx_irq, queue);
683 unbind_from_irqhandler(queue->tx_irq, queue);
684 unbind_from_irqhandler(queue->rx_irq, queue);
689 xenvif_unmap_frontend_rings(queue);
693 /* Reverse the relevant parts of xenvif_init_queue().
694 * Used for queue teardown from xenvif_free(), and on the
695 * error handling paths in xenbus.c:connect().
697 void xenvif_deinit_queue(struct xenvif_queue *queue)
699 free_xenballooned_pages(MAX_PENDING_REQS, queue->mmap_pages);
700 netif_napi_del(&queue->napi);
703 void xenvif_free(struct xenvif *vif)
705 struct xenvif_queue *queue = NULL;
706 unsigned int num_queues = vif->num_queues;
707 unsigned int queue_index;
708 /* Here we want to avoid timeout messages if an skb can be legitimately
709 * stuck somewhere else. Realistically this could be an another vif's
710 * internal or QDisc queue. That another vif also has this
711 * rx_drain_timeout_msecs timeout, but the timer only ditches the
712 * internal queue. After that, the QDisc queue can put in worst case
713 * XEN_NETIF_RX_RING_SIZE / MAX_SKB_FRAGS skbs into that another vif's
714 * internal queue, so we need several rounds of such timeouts until we
715 * can be sure that no another vif should have skb's from us. We are
716 * not sending more skb's, so newly stuck packets are not interesting
719 unsigned int worst_case_skb_lifetime = (rx_drain_timeout_msecs/1000) *
720 DIV_ROUND_UP(XENVIF_QUEUE_LENGTH, (XEN_NETIF_RX_RING_SIZE / MAX_SKB_FRAGS));
722 unregister_netdev(vif->dev);
724 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
725 queue = &vif->queues[queue_index];
726 xenvif_wait_unmap_timeout(queue, worst_case_skb_lifetime);
727 xenvif_deinit_queue(queue);
734 free_netdev(vif->dev);
736 module_put(THIS_MODULE);