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 /* Number of bytes allowed on the internal guest Rx queue. */
47 #define XENVIF_RX_QUEUE_BYTES (XEN_NETIF_RX_RING_SIZE/2 * PAGE_SIZE)
49 /* This function is used to set SKBTX_DEV_ZEROCOPY as well as
50 * increasing the inflight counter. We need to increase the inflight
51 * counter because core driver calls into xenvif_zerocopy_callback
52 * which calls xenvif_skb_zerocopy_complete.
54 void xenvif_skb_zerocopy_prepare(struct xenvif_queue *queue,
57 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
58 atomic_inc(&queue->inflight_packets);
61 void xenvif_skb_zerocopy_complete(struct xenvif_queue *queue)
63 atomic_dec(&queue->inflight_packets);
65 /* Wake the dealloc thread _after_ decrementing inflight_packets so
66 * that if kthread_stop() has already been called, the dealloc thread
67 * does not wait forever with nothing to wake it.
69 wake_up(&queue->dealloc_wq);
72 int xenvif_schedulable(struct xenvif *vif)
74 return netif_running(vif->dev) &&
75 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 static 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;
118 xenvif_kick_thread(queue);
123 irqreturn_t xenvif_interrupt(int irq, void *dev_id)
125 xenvif_tx_interrupt(irq, dev_id);
126 xenvif_rx_interrupt(irq, dev_id);
131 irqreturn_t xenvif_ctrl_interrupt(int irq, void *dev_id)
133 struct xenvif *vif = dev_id;
135 wake_up(&vif->ctrl_wq);
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 static int xenvif_start_xmit(struct sk_buff *skb, struct net_device *dev)
156 struct xenvif *vif = netdev_priv(dev);
157 struct xenvif_queue *queue = NULL;
158 unsigned int num_queues = vif->num_queues;
160 struct xenvif_rx_cb *cb;
162 BUG_ON(skb->dev != dev);
164 /* Drop the packet if queues are not set up */
168 /* Obtain the queue to be used to transmit this packet */
169 index = skb_get_queue_mapping(skb);
170 if (index >= num_queues) {
171 pr_warn_ratelimited("Invalid queue %hu for packet on interface %s\n.",
172 index, vif->dev->name);
175 queue = &vif->queues[index];
177 /* Drop the packet if queue is not ready */
178 if (queue->task == NULL ||
179 queue->dealloc_task == NULL ||
180 !xenvif_schedulable(vif))
183 if (vif->multicast_control && skb->pkt_type == PACKET_MULTICAST) {
184 struct ethhdr *eth = (struct ethhdr *)skb->data;
186 if (!xenvif_mcast_match(vif, eth->h_dest))
190 cb = XENVIF_RX_CB(skb);
191 cb->expires = jiffies + vif->drain_timeout;
193 xenvif_rx_queue_tail(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 disable_irq(queue->tx_irq);
261 if (queue->tx_irq != queue->rx_irq)
262 disable_irq(queue->rx_irq);
263 napi_disable(&queue->napi);
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;
366 for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++) {
367 unsigned long accum = 0;
368 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
369 void *vif_stats = &vif->queues[queue_index].stats;
370 accum += *(unsigned long *)(vif_stats + xenvif_stats[i].offset);
376 static void xenvif_get_strings(struct net_device *dev, u32 stringset, u8 * data)
382 for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++)
383 memcpy(data + i * ETH_GSTRING_LEN,
384 xenvif_stats[i].name, ETH_GSTRING_LEN);
389 static const struct ethtool_ops xenvif_ethtool_ops = {
390 .get_link = ethtool_op_get_link,
392 .get_sset_count = xenvif_get_sset_count,
393 .get_ethtool_stats = xenvif_get_ethtool_stats,
394 .get_strings = xenvif_get_strings,
397 static const struct net_device_ops xenvif_netdev_ops = {
398 .ndo_start_xmit = xenvif_start_xmit,
399 .ndo_get_stats = xenvif_get_stats,
400 .ndo_open = xenvif_open,
401 .ndo_stop = xenvif_close,
402 .ndo_change_mtu = xenvif_change_mtu,
403 .ndo_fix_features = xenvif_fix_features,
404 .ndo_set_mac_address = eth_mac_addr,
405 .ndo_validate_addr = eth_validate_addr,
408 struct xenvif *xenvif_alloc(struct device *parent, domid_t domid,
412 struct net_device *dev;
414 char name[IFNAMSIZ] = {};
416 snprintf(name, IFNAMSIZ - 1, "vif%u.%u", domid, handle);
417 /* Allocate a netdev with the max. supported number of queues.
418 * When the guest selects the desired number, it will be updated
419 * via netif_set_real_num_*_queues().
421 dev = alloc_netdev_mq(sizeof(struct xenvif), name, NET_NAME_UNKNOWN,
422 ether_setup, xenvif_max_queues);
424 pr_warn("Could not allocate netdev for %s\n", name);
425 return ERR_PTR(-ENOMEM);
428 SET_NETDEV_DEV(dev, parent);
430 vif = netdev_priv(dev);
433 vif->handle = handle;
437 vif->disabled = false;
438 vif->drain_timeout = msecs_to_jiffies(rx_drain_timeout_msecs);
439 vif->stall_timeout = msecs_to_jiffies(rx_stall_timeout_msecs);
441 /* Start out with no queues. */
445 spin_lock_init(&vif->lock);
446 INIT_LIST_HEAD(&vif->fe_mcast_addr);
448 dev->netdev_ops = &xenvif_netdev_ops;
449 dev->hw_features = NETIF_F_SG |
450 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
451 NETIF_F_TSO | NETIF_F_TSO6;
452 dev->features = dev->hw_features | NETIF_F_RXCSUM;
453 dev->ethtool_ops = &xenvif_ethtool_ops;
455 dev->tx_queue_len = XENVIF_QUEUE_LENGTH;
458 * Initialise a dummy MAC address. We choose the numerically
459 * largest non-broadcast address to prevent the address getting
460 * stolen by an Ethernet bridge for STP purposes.
461 * (FE:FF:FF:FF:FF:FF)
463 eth_broadcast_addr(dev->dev_addr);
464 dev->dev_addr[0] &= ~0x01;
466 netif_carrier_off(dev);
468 err = register_netdev(dev);
470 netdev_warn(dev, "Could not register device: err=%d\n", err);
475 netdev_dbg(dev, "Successfully created xenvif\n");
477 __module_get(THIS_MODULE);
482 int xenvif_init_queue(struct xenvif_queue *queue)
486 queue->credit_bytes = queue->remaining_credit = ~0UL;
487 queue->credit_usec = 0UL;
488 init_timer(&queue->credit_timeout);
489 queue->credit_timeout.function = xenvif_tx_credit_callback;
490 queue->credit_window_start = get_jiffies_64();
492 queue->rx_queue_max = XENVIF_RX_QUEUE_BYTES;
494 skb_queue_head_init(&queue->rx_queue);
495 skb_queue_head_init(&queue->tx_queue);
497 queue->pending_cons = 0;
498 queue->pending_prod = MAX_PENDING_REQS;
499 for (i = 0; i < MAX_PENDING_REQS; ++i)
500 queue->pending_ring[i] = i;
502 spin_lock_init(&queue->callback_lock);
503 spin_lock_init(&queue->response_lock);
505 /* If ballooning is disabled, this will consume real memory, so you
506 * better enable it. The long term solution would be to use just a
507 * bunch of valid page descriptors, without dependency on ballooning
509 err = gnttab_alloc_pages(MAX_PENDING_REQS,
512 netdev_err(queue->vif->dev, "Could not reserve mmap_pages\n");
516 for (i = 0; i < MAX_PENDING_REQS; i++) {
517 queue->pending_tx_info[i].callback_struct = (struct ubuf_info)
518 { .callback = xenvif_zerocopy_callback,
521 queue->grant_tx_handle[i] = NETBACK_INVALID_HANDLE;
527 void xenvif_carrier_on(struct xenvif *vif)
530 if (!vif->can_sg && vif->dev->mtu > ETH_DATA_LEN)
531 dev_set_mtu(vif->dev, ETH_DATA_LEN);
532 netdev_update_features(vif->dev);
533 set_bit(VIF_STATUS_CONNECTED, &vif->status);
534 if (netif_running(vif->dev))
539 int xenvif_connect_ctrl(struct xenvif *vif, grant_ref_t ring_ref,
542 struct net_device *dev = vif->dev;
544 struct xen_netif_ctrl_sring *shared;
545 struct task_struct *task;
548 err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(vif),
549 &ring_ref, 1, &addr);
553 shared = (struct xen_netif_ctrl_sring *)addr;
554 BACK_RING_INIT(&vif->ctrl, shared, XEN_PAGE_SIZE);
556 init_waitqueue_head(&vif->ctrl_wq);
558 err = bind_interdomain_evtchn_to_irqhandler(vif->domid, evtchn,
559 xenvif_ctrl_interrupt,
566 task = kthread_create(xenvif_ctrl_kthread, (void *)vif,
567 "%s-control", dev->name);
569 pr_warn("Could not allocate kthread for %s\n", dev->name);
574 get_task_struct(task);
575 vif->ctrl_task = task;
577 wake_up_process(vif->ctrl_task);
582 unbind_from_irqhandler(vif->ctrl_irq, vif);
586 xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif),
588 vif->ctrl.sring = NULL;
594 int xenvif_connect_data(struct xenvif_queue *queue,
595 unsigned long tx_ring_ref,
596 unsigned long rx_ring_ref,
597 unsigned int tx_evtchn,
598 unsigned int rx_evtchn)
600 struct task_struct *task;
603 BUG_ON(queue->tx_irq);
605 BUG_ON(queue->dealloc_task);
607 err = xenvif_map_frontend_data_rings(queue, tx_ring_ref,
612 init_waitqueue_head(&queue->wq);
613 init_waitqueue_head(&queue->dealloc_wq);
614 atomic_set(&queue->inflight_packets, 0);
616 netif_napi_add(queue->vif->dev, &queue->napi, xenvif_poll,
619 if (tx_evtchn == rx_evtchn) {
620 /* feature-split-event-channels == 0 */
621 err = bind_interdomain_evtchn_to_irqhandler(
622 queue->vif->domid, tx_evtchn, xenvif_interrupt, 0,
626 queue->tx_irq = queue->rx_irq = err;
627 disable_irq(queue->tx_irq);
629 /* feature-split-event-channels == 1 */
630 snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
631 "%s-tx", queue->name);
632 err = bind_interdomain_evtchn_to_irqhandler(
633 queue->vif->domid, tx_evtchn, xenvif_tx_interrupt, 0,
634 queue->tx_irq_name, queue);
638 disable_irq(queue->tx_irq);
640 snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
641 "%s-rx", queue->name);
642 err = bind_interdomain_evtchn_to_irqhandler(
643 queue->vif->domid, rx_evtchn, xenvif_rx_interrupt, 0,
644 queue->rx_irq_name, queue);
648 disable_irq(queue->rx_irq);
651 queue->stalled = true;
653 task = kthread_create(xenvif_kthread_guest_rx,
654 (void *)queue, "%s-guest-rx", queue->name);
656 pr_warn("Could not allocate kthread for %s\n", queue->name);
661 get_task_struct(task);
663 task = kthread_create(xenvif_dealloc_kthread,
664 (void *)queue, "%s-dealloc", queue->name);
666 pr_warn("Could not allocate kthread for %s\n", queue->name);
670 queue->dealloc_task = task;
672 wake_up_process(queue->task);
673 wake_up_process(queue->dealloc_task);
678 unbind_from_irqhandler(queue->rx_irq, queue);
681 unbind_from_irqhandler(queue->tx_irq, queue);
684 xenvif_unmap_frontend_data_rings(queue);
685 netif_napi_del(&queue->napi);
687 module_put(THIS_MODULE);
691 void xenvif_carrier_off(struct xenvif *vif)
693 struct net_device *dev = vif->dev;
696 if (test_and_clear_bit(VIF_STATUS_CONNECTED, &vif->status)) {
697 netif_carrier_off(dev); /* discard queued packets */
698 if (netif_running(dev))
704 void xenvif_disconnect_data(struct xenvif *vif)
706 struct xenvif_queue *queue = NULL;
707 unsigned int num_queues = vif->num_queues;
708 unsigned int queue_index;
710 xenvif_carrier_off(vif);
712 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
713 queue = &vif->queues[queue_index];
715 netif_napi_del(&queue->napi);
718 kthread_stop(queue->task);
719 put_task_struct(queue->task);
723 if (queue->dealloc_task) {
724 kthread_stop(queue->dealloc_task);
725 queue->dealloc_task = NULL;
729 if (queue->tx_irq == queue->rx_irq)
730 unbind_from_irqhandler(queue->tx_irq, queue);
732 unbind_from_irqhandler(queue->tx_irq, queue);
733 unbind_from_irqhandler(queue->rx_irq, queue);
738 xenvif_unmap_frontend_data_rings(queue);
741 xenvif_mcast_addr_list_free(vif);
744 void xenvif_disconnect_ctrl(struct xenvif *vif)
746 if (vif->ctrl_task) {
747 kthread_stop(vif->ctrl_task);
748 put_task_struct(vif->ctrl_task);
749 vif->ctrl_task = NULL;
753 unbind_from_irqhandler(vif->ctrl_irq, vif);
757 if (vif->ctrl.sring) {
758 xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif),
760 vif->ctrl.sring = NULL;
764 /* Reverse the relevant parts of xenvif_init_queue().
765 * Used for queue teardown from xenvif_free(), and on the
766 * error handling paths in xenbus.c:connect().
768 void xenvif_deinit_queue(struct xenvif_queue *queue)
770 gnttab_free_pages(MAX_PENDING_REQS, queue->mmap_pages);
773 void xenvif_free(struct xenvif *vif)
775 struct xenvif_queue *queues = vif->queues;
776 unsigned int num_queues = vif->num_queues;
777 unsigned int queue_index;
779 unregister_netdev(vif->dev);
780 free_netdev(vif->dev);
782 for (queue_index = 0; queue_index < num_queues; ++queue_index)
783 xenvif_deinit_queue(&queues[queue_index]);
786 module_put(THIS_MODULE);