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 static inline void xenvif_stop_queue(struct xenvif_queue *queue)
65 struct net_device *dev = queue->vif->dev;
67 if (!queue->vif->can_queue)
70 netif_tx_stop_queue(netdev_get_tx_queue(dev, queue->id));
73 int xenvif_schedulable(struct xenvif *vif)
75 return netif_running(vif->dev) &&
76 test_bit(VIF_STATUS_CONNECTED, &vif->status);
79 static irqreturn_t xenvif_tx_interrupt(int irq, void *dev_id)
81 struct xenvif_queue *queue = dev_id;
83 if (RING_HAS_UNCONSUMED_REQUESTS(&queue->tx))
84 napi_schedule(&queue->napi);
89 int xenvif_poll(struct napi_struct *napi, int budget)
91 struct xenvif_queue *queue =
92 container_of(napi, struct xenvif_queue, napi);
95 /* This vif is rogue, we pretend we've there is nothing to do
96 * for this vif to deschedule it from NAPI. But this interface
97 * will be turned off in thread context later.
99 if (unlikely(queue->vif->disabled)) {
104 work_done = xenvif_tx_action(queue, budget);
106 if (work_done < budget) {
108 xenvif_napi_schedule_or_enable_events(queue);
114 static irqreturn_t xenvif_rx_interrupt(int irq, void *dev_id)
116 struct xenvif_queue *queue = dev_id;
117 struct netdev_queue *net_queue =
118 netdev_get_tx_queue(queue->vif->dev, queue->id);
120 /* QUEUE_STATUS_RX_PURGE_EVENT is only set if either QDisc was off OR
121 * the carrier went down and this queue was previously blocked
123 if (unlikely(netif_tx_queue_stopped(net_queue) ||
124 (!netif_carrier_ok(queue->vif->dev) &&
125 test_bit(QUEUE_STATUS_RX_STALLED, &queue->status))))
126 set_bit(QUEUE_STATUS_RX_PURGE_EVENT, &queue->status);
127 xenvif_kick_thread(queue);
132 irqreturn_t xenvif_interrupt(int irq, void *dev_id)
134 xenvif_tx_interrupt(irq, dev_id);
135 xenvif_rx_interrupt(irq, dev_id);
140 int xenvif_queue_stopped(struct xenvif_queue *queue)
142 struct net_device *dev = queue->vif->dev;
143 unsigned int id = queue->id;
144 return netif_tx_queue_stopped(netdev_get_tx_queue(dev, id));
147 void xenvif_wake_queue(struct xenvif_queue *queue)
149 struct net_device *dev = queue->vif->dev;
150 unsigned int id = queue->id;
151 netif_tx_wake_queue(netdev_get_tx_queue(dev, id));
154 /* Callback to wake the queue's thread and turn the carrier off on timeout */
155 static void xenvif_rx_stalled(unsigned long data)
157 struct xenvif_queue *queue = (struct xenvif_queue *)data;
159 if (xenvif_queue_stopped(queue)) {
160 set_bit(QUEUE_STATUS_RX_PURGE_EVENT, &queue->status);
161 xenvif_kick_thread(queue);
165 static int xenvif_start_xmit(struct sk_buff *skb, struct net_device *dev)
167 struct xenvif *vif = netdev_priv(dev);
168 struct xenvif_queue *queue = NULL;
169 unsigned int num_queues = vif->num_queues;
171 int min_slots_needed;
173 BUG_ON(skb->dev != dev);
175 /* Drop the packet if queues are not set up */
179 /* Obtain the queue to be used to transmit this packet */
180 index = skb_get_queue_mapping(skb);
181 if (index >= num_queues) {
182 pr_warn_ratelimited("Invalid queue %hu for packet on interface %s\n.",
183 index, vif->dev->name);
186 queue = &vif->queues[index];
188 /* Drop the packet if queue is not ready */
189 if (queue->task == NULL ||
190 queue->dealloc_task == NULL ||
191 !xenvif_schedulable(vif))
194 /* At best we'll need one slot for the header and one for each
197 min_slots_needed = 1 + skb_shinfo(skb)->nr_frags;
199 /* If the skb is GSO then we'll also need an extra slot for the
205 /* If the skb can't possibly fit in the remaining slots
206 * then turn off the queue to give the ring a chance to
209 if (!xenvif_rx_ring_slots_available(queue, min_slots_needed)) {
210 queue->rx_stalled.function = xenvif_rx_stalled;
211 queue->rx_stalled.data = (unsigned long)queue;
212 xenvif_stop_queue(queue);
213 mod_timer(&queue->rx_stalled,
214 jiffies + rx_drain_timeout_jiffies);
217 skb_queue_tail(&queue->rx_queue, skb);
218 xenvif_kick_thread(queue);
223 vif->dev->stats.tx_dropped++;
228 static struct net_device_stats *xenvif_get_stats(struct net_device *dev)
230 struct xenvif *vif = netdev_priv(dev);
231 struct xenvif_queue *queue = NULL;
232 unsigned int num_queues = vif->num_queues;
233 unsigned long rx_bytes = 0;
234 unsigned long rx_packets = 0;
235 unsigned long tx_bytes = 0;
236 unsigned long tx_packets = 0;
239 if (vif->queues == NULL)
242 /* Aggregate tx and rx stats from each queue */
243 for (index = 0; index < num_queues; ++index) {
244 queue = &vif->queues[index];
245 rx_bytes += queue->stats.rx_bytes;
246 rx_packets += queue->stats.rx_packets;
247 tx_bytes += queue->stats.tx_bytes;
248 tx_packets += queue->stats.tx_packets;
252 vif->dev->stats.rx_bytes = rx_bytes;
253 vif->dev->stats.rx_packets = rx_packets;
254 vif->dev->stats.tx_bytes = tx_bytes;
255 vif->dev->stats.tx_packets = tx_packets;
257 return &vif->dev->stats;
260 static void xenvif_up(struct xenvif *vif)
262 struct xenvif_queue *queue = NULL;
263 unsigned int num_queues = vif->num_queues;
264 unsigned int queue_index;
266 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
267 queue = &vif->queues[queue_index];
268 napi_enable(&queue->napi);
269 enable_irq(queue->tx_irq);
270 if (queue->tx_irq != queue->rx_irq)
271 enable_irq(queue->rx_irq);
272 xenvif_napi_schedule_or_enable_events(queue);
276 static void xenvif_down(struct xenvif *vif)
278 struct xenvif_queue *queue = NULL;
279 unsigned int num_queues = vif->num_queues;
280 unsigned int queue_index;
282 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
283 queue = &vif->queues[queue_index];
284 napi_disable(&queue->napi);
285 disable_irq(queue->tx_irq);
286 if (queue->tx_irq != queue->rx_irq)
287 disable_irq(queue->rx_irq);
288 del_timer_sync(&queue->credit_timeout);
292 static int xenvif_open(struct net_device *dev)
294 struct xenvif *vif = netdev_priv(dev);
295 if (test_bit(VIF_STATUS_CONNECTED, &vif->status))
297 netif_tx_start_all_queues(dev);
301 static int xenvif_close(struct net_device *dev)
303 struct xenvif *vif = netdev_priv(dev);
304 if (test_bit(VIF_STATUS_CONNECTED, &vif->status))
306 netif_tx_stop_all_queues(dev);
310 static int xenvif_change_mtu(struct net_device *dev, int mtu)
312 struct xenvif *vif = netdev_priv(dev);
313 int max = vif->can_sg ? 65535 - VLAN_ETH_HLEN : ETH_DATA_LEN;
321 static netdev_features_t xenvif_fix_features(struct net_device *dev,
322 netdev_features_t features)
324 struct xenvif *vif = netdev_priv(dev);
327 features &= ~NETIF_F_SG;
328 if (~(vif->gso_mask | vif->gso_prefix_mask) & GSO_BIT(TCPV4))
329 features &= ~NETIF_F_TSO;
330 if (~(vif->gso_mask | vif->gso_prefix_mask) & GSO_BIT(TCPV6))
331 features &= ~NETIF_F_TSO6;
333 features &= ~NETIF_F_IP_CSUM;
335 features &= ~NETIF_F_IPV6_CSUM;
340 static const struct xenvif_stat {
341 char name[ETH_GSTRING_LEN];
345 "rx_gso_checksum_fixup",
346 offsetof(struct xenvif_stats, rx_gso_checksum_fixup)
348 /* If (sent != success + fail), there are probably packets never
353 offsetof(struct xenvif_stats, tx_zerocopy_sent),
356 "tx_zerocopy_success",
357 offsetof(struct xenvif_stats, tx_zerocopy_success),
361 offsetof(struct xenvif_stats, tx_zerocopy_fail)
363 /* Number of packets exceeding MAX_SKB_FRAG slots. You should use
364 * a guest with the same MAX_SKB_FRAG
368 offsetof(struct xenvif_stats, tx_frag_overflow)
372 static int xenvif_get_sset_count(struct net_device *dev, int string_set)
374 switch (string_set) {
376 return ARRAY_SIZE(xenvif_stats);
382 static void xenvif_get_ethtool_stats(struct net_device *dev,
383 struct ethtool_stats *stats, u64 * data)
385 struct xenvif *vif = netdev_priv(dev);
386 unsigned int num_queues = vif->num_queues;
388 unsigned int queue_index;
389 struct xenvif_stats *vif_stats;
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 vif_stats = &vif->queues[queue_index].stats;
395 accum += *(unsigned long *)(vif_stats + xenvif_stats[i].offset);
401 static void xenvif_get_strings(struct net_device *dev, u32 stringset, u8 * data)
407 for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++)
408 memcpy(data + i * ETH_GSTRING_LEN,
409 xenvif_stats[i].name, ETH_GSTRING_LEN);
414 static const struct ethtool_ops xenvif_ethtool_ops = {
415 .get_link = ethtool_op_get_link,
417 .get_sset_count = xenvif_get_sset_count,
418 .get_ethtool_stats = xenvif_get_ethtool_stats,
419 .get_strings = xenvif_get_strings,
422 static const struct net_device_ops xenvif_netdev_ops = {
423 .ndo_start_xmit = xenvif_start_xmit,
424 .ndo_get_stats = xenvif_get_stats,
425 .ndo_open = xenvif_open,
426 .ndo_stop = xenvif_close,
427 .ndo_change_mtu = xenvif_change_mtu,
428 .ndo_fix_features = xenvif_fix_features,
429 .ndo_set_mac_address = eth_mac_addr,
430 .ndo_validate_addr = eth_validate_addr,
433 struct xenvif *xenvif_alloc(struct device *parent, domid_t domid,
437 struct net_device *dev;
439 char name[IFNAMSIZ] = {};
441 snprintf(name, IFNAMSIZ - 1, "vif%u.%u", domid, handle);
442 /* Allocate a netdev with the max. supported number of queues.
443 * When the guest selects the desired number, it will be updated
444 * via netif_set_real_num_*_queues().
446 dev = alloc_netdev_mq(sizeof(struct xenvif), name, NET_NAME_UNKNOWN,
447 ether_setup, xenvif_max_queues);
449 pr_warn("Could not allocate netdev for %s\n", name);
450 return ERR_PTR(-ENOMEM);
453 SET_NETDEV_DEV(dev, parent);
455 vif = netdev_priv(dev);
458 vif->handle = handle;
462 vif->disabled = false;
464 /* Start out with no queues. */
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;
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 * Initialise a dummy MAC address. We choose the numerically
479 * largest non-broadcast address to prevent the address getting
480 * stolen by an Ethernet bridge for STP purposes.
481 * (FE:FF:FF:FF:FF:FF)
483 memset(dev->dev_addr, 0xFF, ETH_ALEN);
484 dev->dev_addr[0] &= ~0x01;
486 netif_carrier_off(dev);
488 err = register_netdev(dev);
490 netdev_warn(dev, "Could not register device: err=%d\n", err);
495 netdev_dbg(dev, "Successfully created xenvif\n");
497 __module_get(THIS_MODULE);
502 int xenvif_init_queue(struct xenvif_queue *queue)
506 queue->credit_bytes = queue->remaining_credit = ~0UL;
507 queue->credit_usec = 0UL;
508 init_timer(&queue->credit_timeout);
509 queue->credit_window_start = get_jiffies_64();
511 skb_queue_head_init(&queue->rx_queue);
512 skb_queue_head_init(&queue->tx_queue);
514 queue->pending_cons = 0;
515 queue->pending_prod = MAX_PENDING_REQS;
516 for (i = 0; i < MAX_PENDING_REQS; ++i)
517 queue->pending_ring[i] = i;
519 spin_lock_init(&queue->callback_lock);
520 spin_lock_init(&queue->response_lock);
522 /* If ballooning is disabled, this will consume real memory, so you
523 * better enable it. The long term solution would be to use just a
524 * bunch of valid page descriptors, without dependency on ballooning
526 err = alloc_xenballooned_pages(MAX_PENDING_REQS,
530 netdev_err(queue->vif->dev, "Could not reserve mmap_pages\n");
534 for (i = 0; i < MAX_PENDING_REQS; i++) {
535 queue->pending_tx_info[i].callback_struct = (struct ubuf_info)
536 { .callback = xenvif_zerocopy_callback,
539 queue->grant_tx_handle[i] = NETBACK_INVALID_HANDLE;
542 init_timer(&queue->rx_stalled);
547 void xenvif_carrier_on(struct xenvif *vif)
550 if (!vif->can_sg && vif->dev->mtu > ETH_DATA_LEN)
551 dev_set_mtu(vif->dev, ETH_DATA_LEN);
552 netdev_update_features(vif->dev);
553 set_bit(VIF_STATUS_CONNECTED, &vif->status);
554 netif_carrier_on(vif->dev);
555 if (netif_running(vif->dev))
560 int xenvif_connect(struct xenvif_queue *queue, unsigned long tx_ring_ref,
561 unsigned long rx_ring_ref, unsigned int tx_evtchn,
562 unsigned int rx_evtchn)
564 struct task_struct *task;
567 BUG_ON(queue->tx_irq);
569 BUG_ON(queue->dealloc_task);
571 err = xenvif_map_frontend_rings(queue, tx_ring_ref, rx_ring_ref);
575 init_waitqueue_head(&queue->wq);
576 init_waitqueue_head(&queue->dealloc_wq);
577 atomic_set(&queue->inflight_packets, 0);
579 netif_napi_add(queue->vif->dev, &queue->napi, xenvif_poll,
582 if (tx_evtchn == rx_evtchn) {
583 /* feature-split-event-channels == 0 */
584 err = bind_interdomain_evtchn_to_irqhandler(
585 queue->vif->domid, tx_evtchn, xenvif_interrupt, 0,
589 queue->tx_irq = queue->rx_irq = err;
590 disable_irq(queue->tx_irq);
592 /* feature-split-event-channels == 1 */
593 snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
594 "%s-tx", queue->name);
595 err = bind_interdomain_evtchn_to_irqhandler(
596 queue->vif->domid, tx_evtchn, xenvif_tx_interrupt, 0,
597 queue->tx_irq_name, queue);
601 disable_irq(queue->tx_irq);
603 snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
604 "%s-rx", queue->name);
605 err = bind_interdomain_evtchn_to_irqhandler(
606 queue->vif->domid, rx_evtchn, xenvif_rx_interrupt, 0,
607 queue->rx_irq_name, queue);
611 disable_irq(queue->rx_irq);
614 task = kthread_create(xenvif_kthread_guest_rx,
615 (void *)queue, "%s-guest-rx", queue->name);
617 pr_warn("Could not allocate kthread for %s\n", queue->name);
623 task = kthread_create(xenvif_dealloc_kthread,
624 (void *)queue, "%s-dealloc", queue->name);
626 pr_warn("Could not allocate kthread for %s\n", queue->name);
630 queue->dealloc_task = task;
632 wake_up_process(queue->task);
633 wake_up_process(queue->dealloc_task);
638 unbind_from_irqhandler(queue->rx_irq, queue);
641 unbind_from_irqhandler(queue->tx_irq, queue);
644 xenvif_unmap_frontend_rings(queue);
646 module_put(THIS_MODULE);
650 void xenvif_carrier_off(struct xenvif *vif)
652 struct net_device *dev = vif->dev;
655 if (test_and_clear_bit(VIF_STATUS_CONNECTED, &vif->status)) {
656 netif_carrier_off(dev); /* discard queued packets */
657 if (netif_running(dev))
663 void xenvif_disconnect(struct xenvif *vif)
665 struct xenvif_queue *queue = NULL;
666 unsigned int num_queues = vif->num_queues;
667 unsigned int queue_index;
669 xenvif_carrier_off(vif);
671 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
672 queue = &vif->queues[queue_index];
674 netif_napi_del(&queue->napi);
677 del_timer_sync(&queue->rx_stalled);
678 kthread_stop(queue->task);
682 if (queue->dealloc_task) {
683 kthread_stop(queue->dealloc_task);
684 queue->dealloc_task = NULL;
688 if (queue->tx_irq == queue->rx_irq)
689 unbind_from_irqhandler(queue->tx_irq, queue);
691 unbind_from_irqhandler(queue->tx_irq, queue);
692 unbind_from_irqhandler(queue->rx_irq, queue);
697 xenvif_unmap_frontend_rings(queue);
701 /* Reverse the relevant parts of xenvif_init_queue().
702 * Used for queue teardown from xenvif_free(), and on the
703 * error handling paths in xenbus.c:connect().
705 void xenvif_deinit_queue(struct xenvif_queue *queue)
707 free_xenballooned_pages(MAX_PENDING_REQS, queue->mmap_pages);
710 void xenvif_free(struct xenvif *vif)
712 struct xenvif_queue *queue = NULL;
713 unsigned int num_queues = vif->num_queues;
714 unsigned int queue_index;
716 unregister_netdev(vif->dev);
718 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
719 queue = &vif->queues[queue_index];
720 xenvif_deinit_queue(queue);
727 free_netdev(vif->dev);
729 module_put(THIS_MODULE);