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 int xenvif_queue_stopped(struct xenvif_queue *queue)
133 struct net_device *dev = queue->vif->dev;
134 unsigned int id = queue->id;
135 return netif_tx_queue_stopped(netdev_get_tx_queue(dev, id));
138 void xenvif_wake_queue(struct xenvif_queue *queue)
140 struct net_device *dev = queue->vif->dev;
141 unsigned int id = queue->id;
142 netif_tx_wake_queue(netdev_get_tx_queue(dev, id));
145 static u16 xenvif_select_queue(struct net_device *dev, struct sk_buff *skb,
147 select_queue_fallback_t fallback)
149 struct xenvif *vif = netdev_priv(dev);
150 unsigned int size = vif->hash.size;
152 if (vif->hash.alg == XEN_NETIF_CTRL_HASH_ALGORITHM_NONE)
153 return fallback(dev, skb) % dev->real_num_tx_queues;
155 xenvif_set_skb_hash(vif, skb);
158 return skb_get_hash_raw(skb) % dev->real_num_tx_queues;
160 return vif->hash.mapping[skb_get_hash_raw(skb) % size];
163 static int xenvif_start_xmit(struct sk_buff *skb, struct net_device *dev)
165 struct xenvif *vif = netdev_priv(dev);
166 struct xenvif_queue *queue = NULL;
167 unsigned int num_queues = vif->num_queues;
169 struct xenvif_rx_cb *cb;
171 BUG_ON(skb->dev != dev);
173 /* Drop the packet if queues are not set up */
177 /* Obtain the queue to be used to transmit this packet */
178 index = skb_get_queue_mapping(skb);
179 if (index >= num_queues) {
180 pr_warn_ratelimited("Invalid queue %hu for packet on interface %s\n.",
181 index, vif->dev->name);
184 queue = &vif->queues[index];
186 /* Drop the packet if queue is not ready */
187 if (queue->task == NULL ||
188 queue->dealloc_task == NULL ||
189 !xenvif_schedulable(vif))
192 if (vif->multicast_control && skb->pkt_type == PACKET_MULTICAST) {
193 struct ethhdr *eth = (struct ethhdr *)skb->data;
195 if (!xenvif_mcast_match(vif, eth->h_dest))
199 cb = XENVIF_RX_CB(skb);
200 cb->expires = jiffies + vif->drain_timeout;
202 /* If there is no hash algorithm configured then make sure there
203 * is no hash information in the socket buffer otherwise it
204 * would be incorrectly forwarded to the frontend.
206 if (vif->hash.alg == XEN_NETIF_CTRL_HASH_ALGORITHM_NONE)
209 xenvif_rx_queue_tail(queue, skb);
210 xenvif_kick_thread(queue);
215 vif->dev->stats.tx_dropped++;
220 static struct net_device_stats *xenvif_get_stats(struct net_device *dev)
222 struct xenvif *vif = netdev_priv(dev);
223 struct xenvif_queue *queue = NULL;
224 unsigned int num_queues = vif->num_queues;
225 unsigned long rx_bytes = 0;
226 unsigned long rx_packets = 0;
227 unsigned long tx_bytes = 0;
228 unsigned long tx_packets = 0;
231 if (vif->queues == NULL)
234 /* Aggregate tx and rx stats from each queue */
235 for (index = 0; index < num_queues; ++index) {
236 queue = &vif->queues[index];
237 rx_bytes += queue->stats.rx_bytes;
238 rx_packets += queue->stats.rx_packets;
239 tx_bytes += queue->stats.tx_bytes;
240 tx_packets += queue->stats.tx_packets;
244 vif->dev->stats.rx_bytes = rx_bytes;
245 vif->dev->stats.rx_packets = rx_packets;
246 vif->dev->stats.tx_bytes = tx_bytes;
247 vif->dev->stats.tx_packets = tx_packets;
249 return &vif->dev->stats;
252 static void xenvif_up(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_enable(&queue->napi);
261 enable_irq(queue->tx_irq);
262 if (queue->tx_irq != queue->rx_irq)
263 enable_irq(queue->rx_irq);
264 xenvif_napi_schedule_or_enable_events(queue);
268 static void xenvif_down(struct xenvif *vif)
270 struct xenvif_queue *queue = NULL;
271 unsigned int num_queues = vif->num_queues;
272 unsigned int queue_index;
274 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
275 queue = &vif->queues[queue_index];
276 disable_irq(queue->tx_irq);
277 if (queue->tx_irq != queue->rx_irq)
278 disable_irq(queue->rx_irq);
279 napi_disable(&queue->napi);
280 del_timer_sync(&queue->credit_timeout);
284 static int xenvif_open(struct net_device *dev)
286 struct xenvif *vif = netdev_priv(dev);
287 if (test_bit(VIF_STATUS_CONNECTED, &vif->status))
289 netif_tx_start_all_queues(dev);
293 static int xenvif_close(struct net_device *dev)
295 struct xenvif *vif = netdev_priv(dev);
296 if (test_bit(VIF_STATUS_CONNECTED, &vif->status))
298 netif_tx_stop_all_queues(dev);
302 static int xenvif_change_mtu(struct net_device *dev, int mtu)
304 struct xenvif *vif = netdev_priv(dev);
305 int max = vif->can_sg ? ETH_MAX_MTU - VLAN_ETH_HLEN : ETH_DATA_LEN;
313 static netdev_features_t xenvif_fix_features(struct net_device *dev,
314 netdev_features_t features)
316 struct xenvif *vif = netdev_priv(dev);
319 features &= ~NETIF_F_SG;
320 if (~(vif->gso_mask) & GSO_BIT(TCPV4))
321 features &= ~NETIF_F_TSO;
322 if (~(vif->gso_mask) & GSO_BIT(TCPV6))
323 features &= ~NETIF_F_TSO6;
325 features &= ~NETIF_F_IP_CSUM;
327 features &= ~NETIF_F_IPV6_CSUM;
332 static const struct xenvif_stat {
333 char name[ETH_GSTRING_LEN];
337 "rx_gso_checksum_fixup",
338 offsetof(struct xenvif_stats, rx_gso_checksum_fixup)
340 /* If (sent != success + fail), there are probably packets never
345 offsetof(struct xenvif_stats, tx_zerocopy_sent),
348 "tx_zerocopy_success",
349 offsetof(struct xenvif_stats, tx_zerocopy_success),
353 offsetof(struct xenvif_stats, tx_zerocopy_fail)
355 /* Number of packets exceeding MAX_SKB_FRAG slots. You should use
356 * a guest with the same MAX_SKB_FRAG
360 offsetof(struct xenvif_stats, tx_frag_overflow)
364 static int xenvif_get_sset_count(struct net_device *dev, int string_set)
366 switch (string_set) {
368 return ARRAY_SIZE(xenvif_stats);
374 static void xenvif_get_ethtool_stats(struct net_device *dev,
375 struct ethtool_stats *stats, u64 * data)
377 struct xenvif *vif = netdev_priv(dev);
378 unsigned int num_queues = vif->num_queues;
380 unsigned int queue_index;
382 for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++) {
383 unsigned long accum = 0;
384 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
385 void *vif_stats = &vif->queues[queue_index].stats;
386 accum += *(unsigned long *)(vif_stats + xenvif_stats[i].offset);
392 static void xenvif_get_strings(struct net_device *dev, u32 stringset, u8 * data)
398 for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++)
399 memcpy(data + i * ETH_GSTRING_LEN,
400 xenvif_stats[i].name, ETH_GSTRING_LEN);
405 static const struct ethtool_ops xenvif_ethtool_ops = {
406 .get_link = ethtool_op_get_link,
408 .get_sset_count = xenvif_get_sset_count,
409 .get_ethtool_stats = xenvif_get_ethtool_stats,
410 .get_strings = xenvif_get_strings,
413 static const struct net_device_ops xenvif_netdev_ops = {
414 .ndo_select_queue = xenvif_select_queue,
415 .ndo_start_xmit = xenvif_start_xmit,
416 .ndo_get_stats = xenvif_get_stats,
417 .ndo_open = xenvif_open,
418 .ndo_stop = xenvif_close,
419 .ndo_change_mtu = xenvif_change_mtu,
420 .ndo_fix_features = xenvif_fix_features,
421 .ndo_set_mac_address = eth_mac_addr,
422 .ndo_validate_addr = eth_validate_addr,
425 struct xenvif *xenvif_alloc(struct device *parent, domid_t domid,
429 struct net_device *dev;
431 char name[IFNAMSIZ] = {};
433 snprintf(name, IFNAMSIZ - 1, "vif%u.%u", domid, handle);
434 /* Allocate a netdev with the max. supported number of queues.
435 * When the guest selects the desired number, it will be updated
436 * via netif_set_real_num_*_queues().
438 dev = alloc_netdev_mq(sizeof(struct xenvif), name, NET_NAME_UNKNOWN,
439 ether_setup, xenvif_max_queues);
441 pr_warn("Could not allocate netdev for %s\n", name);
442 return ERR_PTR(-ENOMEM);
445 SET_NETDEV_DEV(dev, parent);
447 vif = netdev_priv(dev);
450 vif->handle = handle;
454 vif->disabled = false;
455 vif->drain_timeout = msecs_to_jiffies(rx_drain_timeout_msecs);
456 vif->stall_timeout = msecs_to_jiffies(rx_stall_timeout_msecs);
458 /* Start out with no queues. */
462 spin_lock_init(&vif->lock);
463 INIT_LIST_HEAD(&vif->fe_mcast_addr);
465 dev->netdev_ops = &xenvif_netdev_ops;
466 dev->hw_features = NETIF_F_SG |
467 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
468 NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_FRAGLIST;
469 dev->features = dev->hw_features | NETIF_F_RXCSUM;
470 dev->ethtool_ops = &xenvif_ethtool_ops;
472 dev->tx_queue_len = XENVIF_QUEUE_LENGTH;
475 dev->max_mtu = ETH_MAX_MTU - VLAN_ETH_HLEN;
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 eth_broadcast_addr(dev->dev_addr);
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_timeout.function = xenvif_tx_credit_callback;
510 queue->credit_window_start = get_jiffies_64();
512 queue->rx_queue_max = XENVIF_RX_QUEUE_BYTES;
514 skb_queue_head_init(&queue->rx_queue);
515 skb_queue_head_init(&queue->tx_queue);
517 queue->pending_cons = 0;
518 queue->pending_prod = MAX_PENDING_REQS;
519 for (i = 0; i < MAX_PENDING_REQS; ++i)
520 queue->pending_ring[i] = i;
522 spin_lock_init(&queue->callback_lock);
523 spin_lock_init(&queue->response_lock);
525 /* If ballooning is disabled, this will consume real memory, so you
526 * better enable it. The long term solution would be to use just a
527 * bunch of valid page descriptors, without dependency on ballooning
529 err = gnttab_alloc_pages(MAX_PENDING_REQS,
532 netdev_err(queue->vif->dev, "Could not reserve mmap_pages\n");
536 for (i = 0; i < MAX_PENDING_REQS; i++) {
537 queue->pending_tx_info[i].callback_struct = (struct ubuf_info)
538 { .callback = xenvif_zerocopy_callback,
541 queue->grant_tx_handle[i] = NETBACK_INVALID_HANDLE;
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 if (netif_running(vif->dev))
559 int xenvif_connect_ctrl(struct xenvif *vif, grant_ref_t ring_ref,
562 struct net_device *dev = vif->dev;
564 struct xen_netif_ctrl_sring *shared;
567 err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(vif),
568 &ring_ref, 1, &addr);
572 shared = (struct xen_netif_ctrl_sring *)addr;
573 BACK_RING_INIT(&vif->ctrl, shared, XEN_PAGE_SIZE);
575 err = bind_interdomain_evtchn_to_irq(vif->domid, evtchn);
581 xenvif_init_hash(vif);
583 err = request_threaded_irq(vif->ctrl_irq, NULL, xenvif_ctrl_irq_fn,
584 IRQF_ONESHOT, "xen-netback-ctrl", vif);
586 pr_warn("Could not setup irq handler for %s\n", dev->name);
593 xenvif_deinit_hash(vif);
594 unbind_from_irqhandler(vif->ctrl_irq, vif);
598 xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif),
600 vif->ctrl.sring = NULL;
606 int xenvif_connect_data(struct xenvif_queue *queue,
607 unsigned long tx_ring_ref,
608 unsigned long rx_ring_ref,
609 unsigned int tx_evtchn,
610 unsigned int rx_evtchn)
612 struct task_struct *task;
615 BUG_ON(queue->tx_irq);
617 BUG_ON(queue->dealloc_task);
619 err = xenvif_map_frontend_data_rings(queue, tx_ring_ref,
624 init_waitqueue_head(&queue->wq);
625 init_waitqueue_head(&queue->dealloc_wq);
626 atomic_set(&queue->inflight_packets, 0);
628 netif_napi_add(queue->vif->dev, &queue->napi, xenvif_poll,
631 if (tx_evtchn == rx_evtchn) {
632 /* feature-split-event-channels == 0 */
633 err = bind_interdomain_evtchn_to_irqhandler(
634 queue->vif->domid, tx_evtchn, xenvif_interrupt, 0,
638 queue->tx_irq = queue->rx_irq = err;
639 disable_irq(queue->tx_irq);
641 /* feature-split-event-channels == 1 */
642 snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
643 "%s-tx", queue->name);
644 err = bind_interdomain_evtchn_to_irqhandler(
645 queue->vif->domid, tx_evtchn, xenvif_tx_interrupt, 0,
646 queue->tx_irq_name, queue);
650 disable_irq(queue->tx_irq);
652 snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
653 "%s-rx", queue->name);
654 err = bind_interdomain_evtchn_to_irqhandler(
655 queue->vif->domid, rx_evtchn, xenvif_rx_interrupt, 0,
656 queue->rx_irq_name, queue);
660 disable_irq(queue->rx_irq);
663 queue->stalled = true;
665 task = kthread_create(xenvif_kthread_guest_rx,
666 (void *)queue, "%s-guest-rx", queue->name);
668 pr_warn("Could not allocate kthread for %s\n", queue->name);
673 get_task_struct(task);
675 task = kthread_create(xenvif_dealloc_kthread,
676 (void *)queue, "%s-dealloc", queue->name);
678 pr_warn("Could not allocate kthread for %s\n", queue->name);
682 queue->dealloc_task = task;
684 wake_up_process(queue->task);
685 wake_up_process(queue->dealloc_task);
690 unbind_from_irqhandler(queue->rx_irq, queue);
693 unbind_from_irqhandler(queue->tx_irq, queue);
696 xenvif_unmap_frontend_data_rings(queue);
697 netif_napi_del(&queue->napi);
699 module_put(THIS_MODULE);
703 void xenvif_carrier_off(struct xenvif *vif)
705 struct net_device *dev = vif->dev;
708 if (test_and_clear_bit(VIF_STATUS_CONNECTED, &vif->status)) {
709 netif_carrier_off(dev); /* discard queued packets */
710 if (netif_running(dev))
716 void xenvif_disconnect_data(struct xenvif *vif)
718 struct xenvif_queue *queue = NULL;
719 unsigned int num_queues = vif->num_queues;
720 unsigned int queue_index;
722 xenvif_carrier_off(vif);
724 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
725 queue = &vif->queues[queue_index];
727 netif_napi_del(&queue->napi);
730 kthread_stop(queue->task);
731 put_task_struct(queue->task);
735 if (queue->dealloc_task) {
736 kthread_stop(queue->dealloc_task);
737 queue->dealloc_task = NULL;
741 if (queue->tx_irq == queue->rx_irq)
742 unbind_from_irqhandler(queue->tx_irq, queue);
744 unbind_from_irqhandler(queue->tx_irq, queue);
745 unbind_from_irqhandler(queue->rx_irq, queue);
750 xenvif_unmap_frontend_data_rings(queue);
753 xenvif_mcast_addr_list_free(vif);
756 void xenvif_disconnect_ctrl(struct xenvif *vif)
759 xenvif_deinit_hash(vif);
760 unbind_from_irqhandler(vif->ctrl_irq, vif);
764 if (vif->ctrl.sring) {
765 xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif),
767 vif->ctrl.sring = NULL;
771 /* Reverse the relevant parts of xenvif_init_queue().
772 * Used for queue teardown from xenvif_free(), and on the
773 * error handling paths in xenbus.c:connect().
775 void xenvif_deinit_queue(struct xenvif_queue *queue)
777 gnttab_free_pages(MAX_PENDING_REQS, queue->mmap_pages);
780 void xenvif_free(struct xenvif *vif)
782 struct xenvif_queue *queues = vif->queues;
783 unsigned int num_queues = vif->num_queues;
784 unsigned int queue_index;
786 unregister_netdev(vif->dev);
787 free_netdev(vif->dev);
789 for (queue_index = 0; queue_index < num_queues; ++queue_index)
790 xenvif_deinit_queue(&queues[queue_index]);
793 module_put(THIS_MODULE);