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 /* This function is used to set SKBTX_DEV_ZEROCOPY as well as
47 * increasing the inflight counter. We need to increase the inflight
48 * counter because core driver calls into xenvif_zerocopy_callback
49 * which calls xenvif_skb_zerocopy_complete.
51 void xenvif_skb_zerocopy_prepare(struct xenvif_queue *queue,
54 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
55 atomic_inc(&queue->inflight_packets);
58 void xenvif_skb_zerocopy_complete(struct xenvif_queue *queue)
60 atomic_dec(&queue->inflight_packets);
63 int xenvif_schedulable(struct xenvif *vif)
65 return netif_running(vif->dev) &&
66 test_bit(VIF_STATUS_CONNECTED, &vif->status);
69 static irqreturn_t xenvif_tx_interrupt(int irq, void *dev_id)
71 struct xenvif_queue *queue = dev_id;
73 if (RING_HAS_UNCONSUMED_REQUESTS(&queue->tx))
74 napi_schedule(&queue->napi);
79 int xenvif_poll(struct napi_struct *napi, int budget)
81 struct xenvif_queue *queue =
82 container_of(napi, struct xenvif_queue, napi);
85 /* This vif is rogue, we pretend we've there is nothing to do
86 * for this vif to deschedule it from NAPI. But this interface
87 * will be turned off in thread context later.
89 if (unlikely(queue->vif->disabled)) {
94 work_done = xenvif_tx_action(queue, budget);
96 if (work_done < budget) {
98 xenvif_napi_schedule_or_enable_events(queue);
104 static irqreturn_t xenvif_rx_interrupt(int irq, void *dev_id)
106 struct xenvif_queue *queue = dev_id;
107 struct netdev_queue *net_queue =
108 netdev_get_tx_queue(queue->vif->dev, queue->id);
110 /* QUEUE_STATUS_RX_PURGE_EVENT is only set if either QDisc was off OR
111 * the carrier went down and this queue was previously blocked
113 if (unlikely(netif_tx_queue_stopped(net_queue) ||
114 (!netif_carrier_ok(queue->vif->dev) &&
115 test_bit(QUEUE_STATUS_RX_STALLED, &queue->status))))
116 set_bit(QUEUE_STATUS_RX_PURGE_EVENT, &queue->status);
117 xenvif_kick_thread(queue);
122 irqreturn_t xenvif_interrupt(int irq, void *dev_id)
124 xenvif_tx_interrupt(irq, dev_id);
125 xenvif_rx_interrupt(irq, dev_id);
130 int xenvif_queue_stopped(struct xenvif_queue *queue)
132 struct net_device *dev = queue->vif->dev;
133 unsigned int id = queue->id;
134 return netif_tx_queue_stopped(netdev_get_tx_queue(dev, id));
137 void xenvif_wake_queue(struct xenvif_queue *queue)
139 struct net_device *dev = queue->vif->dev;
140 unsigned int id = queue->id;
141 netif_tx_wake_queue(netdev_get_tx_queue(dev, id));
144 /* Callback to wake the queue's thread and turn the carrier off on timeout */
145 static void xenvif_rx_stalled(unsigned long data)
147 struct xenvif_queue *queue = (struct xenvif_queue *)data;
149 if (xenvif_queue_stopped(queue)) {
150 set_bit(QUEUE_STATUS_RX_PURGE_EVENT, &queue->status);
151 xenvif_kick_thread(queue);
155 static int xenvif_start_xmit(struct sk_buff *skb, struct net_device *dev)
157 struct xenvif *vif = netdev_priv(dev);
158 struct xenvif_queue *queue = NULL;
159 unsigned int num_queues = vif->num_queues;
161 int min_slots_needed;
163 BUG_ON(skb->dev != dev);
165 /* Drop the packet if queues are not set up */
169 /* Obtain the queue to be used to transmit this packet */
170 index = skb_get_queue_mapping(skb);
171 if (index >= num_queues) {
172 pr_warn_ratelimited("Invalid queue %hu for packet on interface %s\n.",
173 index, vif->dev->name);
176 queue = &vif->queues[index];
178 /* Drop the packet if queue is not ready */
179 if (queue->task == NULL ||
180 queue->dealloc_task == NULL ||
181 !xenvif_schedulable(vif))
184 /* At best we'll need one slot for the header and one for each
187 min_slots_needed = 1 + skb_shinfo(skb)->nr_frags;
189 /* If the skb is GSO then we'll also need an extra slot for the
195 /* If the skb can't possibly fit in the remaining slots
196 * then turn off the queue to give the ring a chance to
199 if (!xenvif_rx_ring_slots_available(queue, min_slots_needed)) {
200 queue->rx_stalled.function = xenvif_rx_stalled;
201 queue->rx_stalled.data = (unsigned long)queue;
202 netif_tx_stop_queue(netdev_get_tx_queue(dev, queue->id));
203 mod_timer(&queue->rx_stalled,
204 jiffies + rx_drain_timeout_jiffies);
207 skb_queue_tail(&queue->rx_queue, skb);
208 xenvif_kick_thread(queue);
213 vif->dev->stats.tx_dropped++;
218 static struct net_device_stats *xenvif_get_stats(struct net_device *dev)
220 struct xenvif *vif = netdev_priv(dev);
221 struct xenvif_queue *queue = NULL;
222 unsigned int num_queues = vif->num_queues;
223 unsigned long rx_bytes = 0;
224 unsigned long rx_packets = 0;
225 unsigned long tx_bytes = 0;
226 unsigned long tx_packets = 0;
229 if (vif->queues == NULL)
232 /* Aggregate tx and rx stats from each queue */
233 for (index = 0; index < num_queues; ++index) {
234 queue = &vif->queues[index];
235 rx_bytes += queue->stats.rx_bytes;
236 rx_packets += queue->stats.rx_packets;
237 tx_bytes += queue->stats.tx_bytes;
238 tx_packets += queue->stats.tx_packets;
242 vif->dev->stats.rx_bytes = rx_bytes;
243 vif->dev->stats.rx_packets = rx_packets;
244 vif->dev->stats.tx_bytes = tx_bytes;
245 vif->dev->stats.tx_packets = tx_packets;
247 return &vif->dev->stats;
250 static void xenvif_up(struct xenvif *vif)
252 struct xenvif_queue *queue = NULL;
253 unsigned int num_queues = vif->num_queues;
254 unsigned int queue_index;
256 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
257 queue = &vif->queues[queue_index];
258 napi_enable(&queue->napi);
259 enable_irq(queue->tx_irq);
260 if (queue->tx_irq != queue->rx_irq)
261 enable_irq(queue->rx_irq);
262 xenvif_napi_schedule_or_enable_events(queue);
266 static void xenvif_down(struct xenvif *vif)
268 struct xenvif_queue *queue = NULL;
269 unsigned int num_queues = vif->num_queues;
270 unsigned int queue_index;
272 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
273 queue = &vif->queues[queue_index];
274 napi_disable(&queue->napi);
275 disable_irq(queue->tx_irq);
276 if (queue->tx_irq != queue->rx_irq)
277 disable_irq(queue->rx_irq);
278 del_timer_sync(&queue->credit_timeout);
282 static int xenvif_open(struct net_device *dev)
284 struct xenvif *vif = netdev_priv(dev);
285 if (test_bit(VIF_STATUS_CONNECTED, &vif->status))
287 netif_tx_start_all_queues(dev);
291 static int xenvif_close(struct net_device *dev)
293 struct xenvif *vif = netdev_priv(dev);
294 if (test_bit(VIF_STATUS_CONNECTED, &vif->status))
296 netif_tx_stop_all_queues(dev);
300 static int xenvif_change_mtu(struct net_device *dev, int mtu)
302 struct xenvif *vif = netdev_priv(dev);
303 int max = vif->can_sg ? 65535 - VLAN_ETH_HLEN : ETH_DATA_LEN;
311 static netdev_features_t xenvif_fix_features(struct net_device *dev,
312 netdev_features_t features)
314 struct xenvif *vif = netdev_priv(dev);
317 features &= ~NETIF_F_SG;
318 if (~(vif->gso_mask | vif->gso_prefix_mask) & GSO_BIT(TCPV4))
319 features &= ~NETIF_F_TSO;
320 if (~(vif->gso_mask | vif->gso_prefix_mask) & GSO_BIT(TCPV6))
321 features &= ~NETIF_F_TSO6;
323 features &= ~NETIF_F_IP_CSUM;
325 features &= ~NETIF_F_IPV6_CSUM;
330 static const struct xenvif_stat {
331 char name[ETH_GSTRING_LEN];
335 "rx_gso_checksum_fixup",
336 offsetof(struct xenvif_stats, rx_gso_checksum_fixup)
338 /* If (sent != success + fail), there are probably packets never
343 offsetof(struct xenvif_stats, tx_zerocopy_sent),
346 "tx_zerocopy_success",
347 offsetof(struct xenvif_stats, tx_zerocopy_success),
351 offsetof(struct xenvif_stats, tx_zerocopy_fail)
353 /* Number of packets exceeding MAX_SKB_FRAG slots. You should use
354 * a guest with the same MAX_SKB_FRAG
358 offsetof(struct xenvif_stats, tx_frag_overflow)
362 static int xenvif_get_sset_count(struct net_device *dev, int string_set)
364 switch (string_set) {
366 return ARRAY_SIZE(xenvif_stats);
372 static void xenvif_get_ethtool_stats(struct net_device *dev,
373 struct ethtool_stats *stats, u64 * data)
375 struct xenvif *vif = netdev_priv(dev);
376 unsigned int num_queues = vif->num_queues;
378 unsigned int queue_index;
379 struct xenvif_stats *vif_stats;
381 for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++) {
382 unsigned long accum = 0;
383 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
384 vif_stats = &vif->queues[queue_index].stats;
385 accum += *(unsigned long *)(vif_stats + xenvif_stats[i].offset);
391 static void xenvif_get_strings(struct net_device *dev, u32 stringset, u8 * data)
397 for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++)
398 memcpy(data + i * ETH_GSTRING_LEN,
399 xenvif_stats[i].name, ETH_GSTRING_LEN);
404 static const struct ethtool_ops xenvif_ethtool_ops = {
405 .get_link = ethtool_op_get_link,
407 .get_sset_count = xenvif_get_sset_count,
408 .get_ethtool_stats = xenvif_get_ethtool_stats,
409 .get_strings = xenvif_get_strings,
412 static const struct net_device_ops xenvif_netdev_ops = {
413 .ndo_start_xmit = xenvif_start_xmit,
414 .ndo_get_stats = xenvif_get_stats,
415 .ndo_open = xenvif_open,
416 .ndo_stop = xenvif_close,
417 .ndo_change_mtu = xenvif_change_mtu,
418 .ndo_fix_features = xenvif_fix_features,
419 .ndo_set_mac_address = eth_mac_addr,
420 .ndo_validate_addr = eth_validate_addr,
423 struct xenvif *xenvif_alloc(struct device *parent, domid_t domid,
427 struct net_device *dev;
429 char name[IFNAMSIZ] = {};
431 snprintf(name, IFNAMSIZ - 1, "vif%u.%u", domid, handle);
432 /* Allocate a netdev with the max. supported number of queues.
433 * When the guest selects the desired number, it will be updated
434 * via netif_set_real_num_*_queues().
436 dev = alloc_netdev_mq(sizeof(struct xenvif), name, NET_NAME_UNKNOWN,
437 ether_setup, xenvif_max_queues);
439 pr_warn("Could not allocate netdev for %s\n", name);
440 return ERR_PTR(-ENOMEM);
443 SET_NETDEV_DEV(dev, parent);
445 vif = netdev_priv(dev);
448 vif->handle = handle;
452 vif->disabled = false;
454 /* Start out with no queues. */
458 dev->netdev_ops = &xenvif_netdev_ops;
459 dev->hw_features = NETIF_F_SG |
460 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
461 NETIF_F_TSO | NETIF_F_TSO6;
462 dev->features = dev->hw_features | NETIF_F_RXCSUM;
463 dev->ethtool_ops = &xenvif_ethtool_ops;
465 dev->tx_queue_len = XENVIF_QUEUE_LENGTH;
468 * Initialise a dummy MAC address. We choose the numerically
469 * largest non-broadcast address to prevent the address getting
470 * stolen by an Ethernet bridge for STP purposes.
471 * (FE:FF:FF:FF:FF:FF)
473 memset(dev->dev_addr, 0xFF, ETH_ALEN);
474 dev->dev_addr[0] &= ~0x01;
476 netif_carrier_off(dev);
478 err = register_netdev(dev);
480 netdev_warn(dev, "Could not register device: err=%d\n", err);
485 netdev_dbg(dev, "Successfully created xenvif\n");
487 __module_get(THIS_MODULE);
492 int xenvif_init_queue(struct xenvif_queue *queue)
496 queue->credit_bytes = queue->remaining_credit = ~0UL;
497 queue->credit_usec = 0UL;
498 init_timer(&queue->credit_timeout);
499 queue->credit_window_start = get_jiffies_64();
501 skb_queue_head_init(&queue->rx_queue);
502 skb_queue_head_init(&queue->tx_queue);
504 queue->pending_cons = 0;
505 queue->pending_prod = MAX_PENDING_REQS;
506 for (i = 0; i < MAX_PENDING_REQS; ++i)
507 queue->pending_ring[i] = i;
509 spin_lock_init(&queue->callback_lock);
510 spin_lock_init(&queue->response_lock);
512 /* If ballooning is disabled, this will consume real memory, so you
513 * better enable it. The long term solution would be to use just a
514 * bunch of valid page descriptors, without dependency on ballooning
516 err = alloc_xenballooned_pages(MAX_PENDING_REQS,
520 netdev_err(queue->vif->dev, "Could not reserve mmap_pages\n");
524 for (i = 0; i < MAX_PENDING_REQS; i++) {
525 queue->pending_tx_info[i].callback_struct = (struct ubuf_info)
526 { .callback = xenvif_zerocopy_callback,
529 queue->grant_tx_handle[i] = NETBACK_INVALID_HANDLE;
532 init_timer(&queue->rx_stalled);
537 void xenvif_carrier_on(struct xenvif *vif)
540 if (!vif->can_sg && vif->dev->mtu > ETH_DATA_LEN)
541 dev_set_mtu(vif->dev, ETH_DATA_LEN);
542 netdev_update_features(vif->dev);
543 set_bit(VIF_STATUS_CONNECTED, &vif->status);
544 netif_carrier_on(vif->dev);
545 if (netif_running(vif->dev))
550 int xenvif_connect(struct xenvif_queue *queue, unsigned long tx_ring_ref,
551 unsigned long rx_ring_ref, unsigned int tx_evtchn,
552 unsigned int rx_evtchn)
554 struct task_struct *task;
557 BUG_ON(queue->tx_irq);
559 BUG_ON(queue->dealloc_task);
561 err = xenvif_map_frontend_rings(queue, tx_ring_ref, rx_ring_ref);
565 init_waitqueue_head(&queue->wq);
566 init_waitqueue_head(&queue->dealloc_wq);
567 atomic_set(&queue->inflight_packets, 0);
569 netif_napi_add(queue->vif->dev, &queue->napi, xenvif_poll,
572 if (tx_evtchn == rx_evtchn) {
573 /* feature-split-event-channels == 0 */
574 err = bind_interdomain_evtchn_to_irqhandler(
575 queue->vif->domid, tx_evtchn, xenvif_interrupt, 0,
579 queue->tx_irq = queue->rx_irq = err;
580 disable_irq(queue->tx_irq);
582 /* feature-split-event-channels == 1 */
583 snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
584 "%s-tx", queue->name);
585 err = bind_interdomain_evtchn_to_irqhandler(
586 queue->vif->domid, tx_evtchn, xenvif_tx_interrupt, 0,
587 queue->tx_irq_name, queue);
591 disable_irq(queue->tx_irq);
593 snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
594 "%s-rx", queue->name);
595 err = bind_interdomain_evtchn_to_irqhandler(
596 queue->vif->domid, rx_evtchn, xenvif_rx_interrupt, 0,
597 queue->rx_irq_name, queue);
601 disable_irq(queue->rx_irq);
604 task = kthread_create(xenvif_kthread_guest_rx,
605 (void *)queue, "%s-guest-rx", queue->name);
607 pr_warn("Could not allocate kthread for %s\n", queue->name);
613 task = kthread_create(xenvif_dealloc_kthread,
614 (void *)queue, "%s-dealloc", queue->name);
616 pr_warn("Could not allocate kthread for %s\n", queue->name);
620 queue->dealloc_task = task;
622 wake_up_process(queue->task);
623 wake_up_process(queue->dealloc_task);
628 unbind_from_irqhandler(queue->rx_irq, queue);
631 unbind_from_irqhandler(queue->tx_irq, queue);
634 xenvif_unmap_frontend_rings(queue);
636 module_put(THIS_MODULE);
640 void xenvif_carrier_off(struct xenvif *vif)
642 struct net_device *dev = vif->dev;
645 if (test_and_clear_bit(VIF_STATUS_CONNECTED, &vif->status)) {
646 netif_carrier_off(dev); /* discard queued packets */
647 if (netif_running(dev))
653 void xenvif_disconnect(struct xenvif *vif)
655 struct xenvif_queue *queue = NULL;
656 unsigned int num_queues = vif->num_queues;
657 unsigned int queue_index;
659 xenvif_carrier_off(vif);
661 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
662 queue = &vif->queues[queue_index];
664 netif_napi_del(&queue->napi);
667 del_timer_sync(&queue->rx_stalled);
668 kthread_stop(queue->task);
672 if (queue->dealloc_task) {
673 kthread_stop(queue->dealloc_task);
674 queue->dealloc_task = NULL;
678 if (queue->tx_irq == queue->rx_irq)
679 unbind_from_irqhandler(queue->tx_irq, queue);
681 unbind_from_irqhandler(queue->tx_irq, queue);
682 unbind_from_irqhandler(queue->rx_irq, queue);
687 xenvif_unmap_frontend_rings(queue);
691 /* Reverse the relevant parts of xenvif_init_queue().
692 * Used for queue teardown from xenvif_free(), and on the
693 * error handling paths in xenbus.c:connect().
695 void xenvif_deinit_queue(struct xenvif_queue *queue)
697 free_xenballooned_pages(MAX_PENDING_REQS, queue->mmap_pages);
700 void xenvif_free(struct xenvif *vif)
702 struct xenvif_queue *queue = NULL;
703 unsigned int num_queues = vif->num_queues;
704 unsigned int queue_index;
706 unregister_netdev(vif->dev);
708 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
709 queue = &vif->queues[queue_index];
710 xenvif_deinit_queue(queue);
717 free_netdev(vif->dev);
719 module_put(THIS_MODULE);