Loading...
1/*
2 * Network-device interface management.
3 *
4 * Copyright (c) 2004-2005, Keir Fraser
5 *
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:
11 *
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:
18 *
19 * The above copyright notice and this permission notice shall be included in
20 * all copies or substantial portions of the Software.
21 *
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
28 * IN THE SOFTWARE.
29 */
30
31#include "common.h"
32
33#include <linux/kthread.h>
34#include <linux/sched/task.h>
35#include <linux/ethtool.h>
36#include <linux/rtnetlink.h>
37#include <linux/if_vlan.h>
38#include <linux/vmalloc.h>
39
40#include <xen/events.h>
41#include <asm/xen/hypercall.h>
42#include <xen/balloon.h>
43
44#define XENVIF_QUEUE_LENGTH 32
45#define XENVIF_NAPI_WEIGHT 64
46
47/* Number of bytes allowed on the internal guest Rx queue. */
48#define XENVIF_RX_QUEUE_BYTES (XEN_NETIF_RX_RING_SIZE/2 * PAGE_SIZE)
49
50/* This function is used to set SKBTX_DEV_ZEROCOPY as well as
51 * increasing the inflight counter. We need to increase the inflight
52 * counter because core driver calls into xenvif_zerocopy_callback
53 * which calls xenvif_skb_zerocopy_complete.
54 */
55void xenvif_skb_zerocopy_prepare(struct xenvif_queue *queue,
56 struct sk_buff *skb)
57{
58 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
59 atomic_inc(&queue->inflight_packets);
60}
61
62void xenvif_skb_zerocopy_complete(struct xenvif_queue *queue)
63{
64 atomic_dec(&queue->inflight_packets);
65
66 /* Wake the dealloc thread _after_ decrementing inflight_packets so
67 * that if kthread_stop() has already been called, the dealloc thread
68 * does not wait forever with nothing to wake it.
69 */
70 wake_up(&queue->dealloc_wq);
71}
72
73int xenvif_schedulable(struct xenvif *vif)
74{
75 return netif_running(vif->dev) &&
76 test_bit(VIF_STATUS_CONNECTED, &vif->status) &&
77 !vif->disabled;
78}
79
80static irqreturn_t xenvif_tx_interrupt(int irq, void *dev_id)
81{
82 struct xenvif_queue *queue = dev_id;
83
84 if (RING_HAS_UNCONSUMED_REQUESTS(&queue->tx))
85 napi_schedule(&queue->napi);
86
87 return IRQ_HANDLED;
88}
89
90static int xenvif_poll(struct napi_struct *napi, int budget)
91{
92 struct xenvif_queue *queue =
93 container_of(napi, struct xenvif_queue, napi);
94 int work_done;
95
96 /* This vif is rogue, we pretend we've there is nothing to do
97 * for this vif to deschedule it from NAPI. But this interface
98 * will be turned off in thread context later.
99 */
100 if (unlikely(queue->vif->disabled)) {
101 napi_complete(napi);
102 return 0;
103 }
104
105 work_done = xenvif_tx_action(queue, budget);
106
107 if (work_done < budget) {
108 napi_complete_done(napi, work_done);
109 /* If the queue is rate-limited, it shall be
110 * rescheduled in the timer callback.
111 */
112 if (likely(!queue->rate_limited))
113 xenvif_napi_schedule_or_enable_events(queue);
114 }
115
116 return work_done;
117}
118
119static irqreturn_t xenvif_rx_interrupt(int irq, void *dev_id)
120{
121 struct xenvif_queue *queue = dev_id;
122
123 xenvif_kick_thread(queue);
124
125 return IRQ_HANDLED;
126}
127
128irqreturn_t xenvif_interrupt(int irq, void *dev_id)
129{
130 xenvif_tx_interrupt(irq, dev_id);
131 xenvif_rx_interrupt(irq, dev_id);
132
133 return IRQ_HANDLED;
134}
135
136int xenvif_queue_stopped(struct xenvif_queue *queue)
137{
138 struct net_device *dev = queue->vif->dev;
139 unsigned int id = queue->id;
140 return netif_tx_queue_stopped(netdev_get_tx_queue(dev, id));
141}
142
143void xenvif_wake_queue(struct xenvif_queue *queue)
144{
145 struct net_device *dev = queue->vif->dev;
146 unsigned int id = queue->id;
147 netif_tx_wake_queue(netdev_get_tx_queue(dev, id));
148}
149
150static u16 xenvif_select_queue(struct net_device *dev, struct sk_buff *skb,
151 void *accel_priv,
152 select_queue_fallback_t fallback)
153{
154 struct xenvif *vif = netdev_priv(dev);
155 unsigned int size = vif->hash.size;
156
157 if (vif->hash.alg == XEN_NETIF_CTRL_HASH_ALGORITHM_NONE)
158 return fallback(dev, skb) % dev->real_num_tx_queues;
159
160 xenvif_set_skb_hash(vif, skb);
161
162 if (size == 0)
163 return skb_get_hash_raw(skb) % dev->real_num_tx_queues;
164
165 return vif->hash.mapping[skb_get_hash_raw(skb) % size];
166}
167
168static int xenvif_start_xmit(struct sk_buff *skb, struct net_device *dev)
169{
170 struct xenvif *vif = netdev_priv(dev);
171 struct xenvif_queue *queue = NULL;
172 unsigned int num_queues;
173 u16 index;
174 struct xenvif_rx_cb *cb;
175
176 BUG_ON(skb->dev != dev);
177
178 /* Drop the packet if queues are not set up.
179 * This handler should be called inside an RCU read section
180 * so we don't need to enter it here explicitly.
181 */
182 num_queues = READ_ONCE(vif->num_queues);
183 if (num_queues < 1)
184 goto drop;
185
186 /* Obtain the queue to be used to transmit this packet */
187 index = skb_get_queue_mapping(skb);
188 if (index >= num_queues) {
189 pr_warn_ratelimited("Invalid queue %hu for packet on interface %s\n",
190 index, vif->dev->name);
191 index %= num_queues;
192 }
193 queue = &vif->queues[index];
194
195 /* Drop the packet if queue is not ready */
196 if (queue->task == NULL ||
197 queue->dealloc_task == NULL ||
198 !xenvif_schedulable(vif))
199 goto drop;
200
201 if (vif->multicast_control && skb->pkt_type == PACKET_MULTICAST) {
202 struct ethhdr *eth = (struct ethhdr *)skb->data;
203
204 if (!xenvif_mcast_match(vif, eth->h_dest))
205 goto drop;
206 }
207
208 cb = XENVIF_RX_CB(skb);
209 cb->expires = jiffies + vif->drain_timeout;
210
211 /* If there is no hash algorithm configured then make sure there
212 * is no hash information in the socket buffer otherwise it
213 * would be incorrectly forwarded to the frontend.
214 */
215 if (vif->hash.alg == XEN_NETIF_CTRL_HASH_ALGORITHM_NONE)
216 skb_clear_hash(skb);
217
218 xenvif_rx_queue_tail(queue, skb);
219 xenvif_kick_thread(queue);
220
221 return NETDEV_TX_OK;
222
223 drop:
224 vif->dev->stats.tx_dropped++;
225 dev_kfree_skb(skb);
226 return NETDEV_TX_OK;
227}
228
229static struct net_device_stats *xenvif_get_stats(struct net_device *dev)
230{
231 struct xenvif *vif = netdev_priv(dev);
232 struct xenvif_queue *queue = NULL;
233 unsigned int num_queues;
234 u64 rx_bytes = 0;
235 u64 rx_packets = 0;
236 u64 tx_bytes = 0;
237 u64 tx_packets = 0;
238 unsigned int index;
239
240 rcu_read_lock();
241 num_queues = READ_ONCE(vif->num_queues);
242
243 /* Aggregate tx and rx stats from each queue */
244 for (index = 0; index < num_queues; ++index) {
245 queue = &vif->queues[index];
246 rx_bytes += queue->stats.rx_bytes;
247 rx_packets += queue->stats.rx_packets;
248 tx_bytes += queue->stats.tx_bytes;
249 tx_packets += queue->stats.tx_packets;
250 }
251
252 rcu_read_unlock();
253
254 vif->dev->stats.rx_bytes = rx_bytes;
255 vif->dev->stats.rx_packets = rx_packets;
256 vif->dev->stats.tx_bytes = tx_bytes;
257 vif->dev->stats.tx_packets = tx_packets;
258
259 return &vif->dev->stats;
260}
261
262static void xenvif_up(struct xenvif *vif)
263{
264 struct xenvif_queue *queue = NULL;
265 unsigned int num_queues = vif->num_queues;
266 unsigned int queue_index;
267
268 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
269 queue = &vif->queues[queue_index];
270 napi_enable(&queue->napi);
271 enable_irq(queue->tx_irq);
272 if (queue->tx_irq != queue->rx_irq)
273 enable_irq(queue->rx_irq);
274 xenvif_napi_schedule_or_enable_events(queue);
275 }
276}
277
278static void xenvif_down(struct xenvif *vif)
279{
280 struct xenvif_queue *queue = NULL;
281 unsigned int num_queues = vif->num_queues;
282 unsigned int queue_index;
283
284 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
285 queue = &vif->queues[queue_index];
286 disable_irq(queue->tx_irq);
287 if (queue->tx_irq != queue->rx_irq)
288 disable_irq(queue->rx_irq);
289 napi_disable(&queue->napi);
290 del_timer_sync(&queue->credit_timeout);
291 }
292}
293
294static int xenvif_open(struct net_device *dev)
295{
296 struct xenvif *vif = netdev_priv(dev);
297 if (test_bit(VIF_STATUS_CONNECTED, &vif->status))
298 xenvif_up(vif);
299 netif_tx_start_all_queues(dev);
300 return 0;
301}
302
303static int xenvif_close(struct net_device *dev)
304{
305 struct xenvif *vif = netdev_priv(dev);
306 if (test_bit(VIF_STATUS_CONNECTED, &vif->status))
307 xenvif_down(vif);
308 netif_tx_stop_all_queues(dev);
309 return 0;
310}
311
312static int xenvif_change_mtu(struct net_device *dev, int mtu)
313{
314 struct xenvif *vif = netdev_priv(dev);
315 int max = vif->can_sg ? ETH_MAX_MTU - VLAN_ETH_HLEN : ETH_DATA_LEN;
316
317 if (mtu > max)
318 return -EINVAL;
319 dev->mtu = mtu;
320 return 0;
321}
322
323static netdev_features_t xenvif_fix_features(struct net_device *dev,
324 netdev_features_t features)
325{
326 struct xenvif *vif = netdev_priv(dev);
327
328 if (!vif->can_sg)
329 features &= ~NETIF_F_SG;
330 if (~(vif->gso_mask) & GSO_BIT(TCPV4))
331 features &= ~NETIF_F_TSO;
332 if (~(vif->gso_mask) & GSO_BIT(TCPV6))
333 features &= ~NETIF_F_TSO6;
334 if (!vif->ip_csum)
335 features &= ~NETIF_F_IP_CSUM;
336 if (!vif->ipv6_csum)
337 features &= ~NETIF_F_IPV6_CSUM;
338
339 return features;
340}
341
342static const struct xenvif_stat {
343 char name[ETH_GSTRING_LEN];
344 u16 offset;
345} xenvif_stats[] = {
346 {
347 "rx_gso_checksum_fixup",
348 offsetof(struct xenvif_stats, rx_gso_checksum_fixup)
349 },
350 /* If (sent != success + fail), there are probably packets never
351 * freed up properly!
352 */
353 {
354 "tx_zerocopy_sent",
355 offsetof(struct xenvif_stats, tx_zerocopy_sent),
356 },
357 {
358 "tx_zerocopy_success",
359 offsetof(struct xenvif_stats, tx_zerocopy_success),
360 },
361 {
362 "tx_zerocopy_fail",
363 offsetof(struct xenvif_stats, tx_zerocopy_fail)
364 },
365 /* Number of packets exceeding MAX_SKB_FRAG slots. You should use
366 * a guest with the same MAX_SKB_FRAG
367 */
368 {
369 "tx_frag_overflow",
370 offsetof(struct xenvif_stats, tx_frag_overflow)
371 },
372};
373
374static int xenvif_get_sset_count(struct net_device *dev, int string_set)
375{
376 switch (string_set) {
377 case ETH_SS_STATS:
378 return ARRAY_SIZE(xenvif_stats);
379 default:
380 return -EINVAL;
381 }
382}
383
384static void xenvif_get_ethtool_stats(struct net_device *dev,
385 struct ethtool_stats *stats, u64 * data)
386{
387 struct xenvif *vif = netdev_priv(dev);
388 unsigned int num_queues;
389 int i;
390 unsigned int queue_index;
391
392 rcu_read_lock();
393 num_queues = READ_ONCE(vif->num_queues);
394
395 for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++) {
396 unsigned long accum = 0;
397 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
398 void *vif_stats = &vif->queues[queue_index].stats;
399 accum += *(unsigned long *)(vif_stats + xenvif_stats[i].offset);
400 }
401 data[i] = accum;
402 }
403
404 rcu_read_unlock();
405}
406
407static void xenvif_get_strings(struct net_device *dev, u32 stringset, u8 * data)
408{
409 int i;
410
411 switch (stringset) {
412 case ETH_SS_STATS:
413 for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++)
414 memcpy(data + i * ETH_GSTRING_LEN,
415 xenvif_stats[i].name, ETH_GSTRING_LEN);
416 break;
417 }
418}
419
420static const struct ethtool_ops xenvif_ethtool_ops = {
421 .get_link = ethtool_op_get_link,
422
423 .get_sset_count = xenvif_get_sset_count,
424 .get_ethtool_stats = xenvif_get_ethtool_stats,
425 .get_strings = xenvif_get_strings,
426};
427
428static const struct net_device_ops xenvif_netdev_ops = {
429 .ndo_select_queue = xenvif_select_queue,
430 .ndo_start_xmit = xenvif_start_xmit,
431 .ndo_get_stats = xenvif_get_stats,
432 .ndo_open = xenvif_open,
433 .ndo_stop = xenvif_close,
434 .ndo_change_mtu = xenvif_change_mtu,
435 .ndo_fix_features = xenvif_fix_features,
436 .ndo_set_mac_address = eth_mac_addr,
437 .ndo_validate_addr = eth_validate_addr,
438};
439
440struct xenvif *xenvif_alloc(struct device *parent, domid_t domid,
441 unsigned int handle)
442{
443 int err;
444 struct net_device *dev;
445 struct xenvif *vif;
446 char name[IFNAMSIZ] = {};
447
448 snprintf(name, IFNAMSIZ - 1, "vif%u.%u", domid, handle);
449 /* Allocate a netdev with the max. supported number of queues.
450 * When the guest selects the desired number, it will be updated
451 * via netif_set_real_num_*_queues().
452 */
453 dev = alloc_netdev_mq(sizeof(struct xenvif), name, NET_NAME_UNKNOWN,
454 ether_setup, xenvif_max_queues);
455 if (dev == NULL) {
456 pr_warn("Could not allocate netdev for %s\n", name);
457 return ERR_PTR(-ENOMEM);
458 }
459
460 SET_NETDEV_DEV(dev, parent);
461
462 vif = netdev_priv(dev);
463
464 vif->domid = domid;
465 vif->handle = handle;
466 vif->can_sg = 1;
467 vif->ip_csum = 1;
468 vif->dev = dev;
469 vif->disabled = false;
470 vif->drain_timeout = msecs_to_jiffies(rx_drain_timeout_msecs);
471 vif->stall_timeout = msecs_to_jiffies(rx_stall_timeout_msecs);
472
473 /* Start out with no queues. */
474 vif->queues = NULL;
475 vif->num_queues = 0;
476
477 spin_lock_init(&vif->lock);
478 INIT_LIST_HEAD(&vif->fe_mcast_addr);
479
480 dev->netdev_ops = &xenvif_netdev_ops;
481 dev->hw_features = NETIF_F_SG |
482 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
483 NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_FRAGLIST;
484 dev->features = dev->hw_features | NETIF_F_RXCSUM;
485 dev->ethtool_ops = &xenvif_ethtool_ops;
486
487 dev->tx_queue_len = XENVIF_QUEUE_LENGTH;
488
489 dev->min_mtu = ETH_MIN_MTU;
490 dev->max_mtu = ETH_MAX_MTU - VLAN_ETH_HLEN;
491
492 /*
493 * Initialise a dummy MAC address. We choose the numerically
494 * largest non-broadcast address to prevent the address getting
495 * stolen by an Ethernet bridge for STP purposes.
496 * (FE:FF:FF:FF:FF:FF)
497 */
498 eth_broadcast_addr(dev->dev_addr);
499 dev->dev_addr[0] &= ~0x01;
500
501 netif_carrier_off(dev);
502
503 err = register_netdev(dev);
504 if (err) {
505 netdev_warn(dev, "Could not register device: err=%d\n", err);
506 free_netdev(dev);
507 return ERR_PTR(err);
508 }
509
510 netdev_dbg(dev, "Successfully created xenvif\n");
511
512 __module_get(THIS_MODULE);
513
514 return vif;
515}
516
517int xenvif_init_queue(struct xenvif_queue *queue)
518{
519 int err, i;
520
521 queue->credit_bytes = queue->remaining_credit = ~0UL;
522 queue->credit_usec = 0UL;
523 timer_setup(&queue->credit_timeout, xenvif_tx_credit_callback, 0);
524 queue->credit_window_start = get_jiffies_64();
525
526 queue->rx_queue_max = XENVIF_RX_QUEUE_BYTES;
527
528 skb_queue_head_init(&queue->rx_queue);
529 skb_queue_head_init(&queue->tx_queue);
530
531 queue->pending_cons = 0;
532 queue->pending_prod = MAX_PENDING_REQS;
533 for (i = 0; i < MAX_PENDING_REQS; ++i)
534 queue->pending_ring[i] = i;
535
536 spin_lock_init(&queue->callback_lock);
537 spin_lock_init(&queue->response_lock);
538
539 /* If ballooning is disabled, this will consume real memory, so you
540 * better enable it. The long term solution would be to use just a
541 * bunch of valid page descriptors, without dependency on ballooning
542 */
543 err = gnttab_alloc_pages(MAX_PENDING_REQS,
544 queue->mmap_pages);
545 if (err) {
546 netdev_err(queue->vif->dev, "Could not reserve mmap_pages\n");
547 return -ENOMEM;
548 }
549
550 for (i = 0; i < MAX_PENDING_REQS; i++) {
551 queue->pending_tx_info[i].callback_struct = (struct ubuf_info)
552 { .callback = xenvif_zerocopy_callback,
553 { { .ctx = NULL,
554 .desc = i } } };
555 queue->grant_tx_handle[i] = NETBACK_INVALID_HANDLE;
556 }
557
558 return 0;
559}
560
561void xenvif_carrier_on(struct xenvif *vif)
562{
563 rtnl_lock();
564 if (!vif->can_sg && vif->dev->mtu > ETH_DATA_LEN)
565 dev_set_mtu(vif->dev, ETH_DATA_LEN);
566 netdev_update_features(vif->dev);
567 set_bit(VIF_STATUS_CONNECTED, &vif->status);
568 if (netif_running(vif->dev))
569 xenvif_up(vif);
570 rtnl_unlock();
571}
572
573int xenvif_connect_ctrl(struct xenvif *vif, grant_ref_t ring_ref,
574 unsigned int evtchn)
575{
576 struct net_device *dev = vif->dev;
577 void *addr;
578 struct xen_netif_ctrl_sring *shared;
579 int err;
580
581 err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(vif),
582 &ring_ref, 1, &addr);
583 if (err)
584 goto err;
585
586 shared = (struct xen_netif_ctrl_sring *)addr;
587 BACK_RING_INIT(&vif->ctrl, shared, XEN_PAGE_SIZE);
588
589 err = bind_interdomain_evtchn_to_irq(vif->domid, evtchn);
590 if (err < 0)
591 goto err_unmap;
592
593 vif->ctrl_irq = err;
594
595 xenvif_init_hash(vif);
596
597 err = request_threaded_irq(vif->ctrl_irq, NULL, xenvif_ctrl_irq_fn,
598 IRQF_ONESHOT, "xen-netback-ctrl", vif);
599 if (err) {
600 pr_warn("Could not setup irq handler for %s\n", dev->name);
601 goto err_deinit;
602 }
603
604 return 0;
605
606err_deinit:
607 xenvif_deinit_hash(vif);
608 unbind_from_irqhandler(vif->ctrl_irq, vif);
609 vif->ctrl_irq = 0;
610
611err_unmap:
612 xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif),
613 vif->ctrl.sring);
614 vif->ctrl.sring = NULL;
615
616err:
617 return err;
618}
619
620int xenvif_connect_data(struct xenvif_queue *queue,
621 unsigned long tx_ring_ref,
622 unsigned long rx_ring_ref,
623 unsigned int tx_evtchn,
624 unsigned int rx_evtchn)
625{
626 struct task_struct *task;
627 int err = -ENOMEM;
628
629 BUG_ON(queue->tx_irq);
630 BUG_ON(queue->task);
631 BUG_ON(queue->dealloc_task);
632
633 err = xenvif_map_frontend_data_rings(queue, tx_ring_ref,
634 rx_ring_ref);
635 if (err < 0)
636 goto err;
637
638 init_waitqueue_head(&queue->wq);
639 init_waitqueue_head(&queue->dealloc_wq);
640 atomic_set(&queue->inflight_packets, 0);
641
642 netif_napi_add(queue->vif->dev, &queue->napi, xenvif_poll,
643 XENVIF_NAPI_WEIGHT);
644
645 if (tx_evtchn == rx_evtchn) {
646 /* feature-split-event-channels == 0 */
647 err = bind_interdomain_evtchn_to_irqhandler(
648 queue->vif->domid, tx_evtchn, xenvif_interrupt, 0,
649 queue->name, queue);
650 if (err < 0)
651 goto err_unmap;
652 queue->tx_irq = queue->rx_irq = err;
653 disable_irq(queue->tx_irq);
654 } else {
655 /* feature-split-event-channels == 1 */
656 snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
657 "%s-tx", queue->name);
658 err = bind_interdomain_evtchn_to_irqhandler(
659 queue->vif->domid, tx_evtchn, xenvif_tx_interrupt, 0,
660 queue->tx_irq_name, queue);
661 if (err < 0)
662 goto err_unmap;
663 queue->tx_irq = err;
664 disable_irq(queue->tx_irq);
665
666 snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
667 "%s-rx", queue->name);
668 err = bind_interdomain_evtchn_to_irqhandler(
669 queue->vif->domid, rx_evtchn, xenvif_rx_interrupt, 0,
670 queue->rx_irq_name, queue);
671 if (err < 0)
672 goto err_tx_unbind;
673 queue->rx_irq = err;
674 disable_irq(queue->rx_irq);
675 }
676
677 queue->stalled = true;
678
679 task = kthread_create(xenvif_kthread_guest_rx,
680 (void *)queue, "%s-guest-rx", queue->name);
681 if (IS_ERR(task)) {
682 pr_warn("Could not allocate kthread for %s\n", queue->name);
683 err = PTR_ERR(task);
684 goto err_rx_unbind;
685 }
686 queue->task = task;
687 get_task_struct(task);
688
689 task = kthread_create(xenvif_dealloc_kthread,
690 (void *)queue, "%s-dealloc", queue->name);
691 if (IS_ERR(task)) {
692 pr_warn("Could not allocate kthread for %s\n", queue->name);
693 err = PTR_ERR(task);
694 goto err_rx_unbind;
695 }
696 queue->dealloc_task = task;
697
698 wake_up_process(queue->task);
699 wake_up_process(queue->dealloc_task);
700
701 return 0;
702
703err_rx_unbind:
704 unbind_from_irqhandler(queue->rx_irq, queue);
705 queue->rx_irq = 0;
706err_tx_unbind:
707 unbind_from_irqhandler(queue->tx_irq, queue);
708 queue->tx_irq = 0;
709err_unmap:
710 xenvif_unmap_frontend_data_rings(queue);
711 netif_napi_del(&queue->napi);
712err:
713 module_put(THIS_MODULE);
714 return err;
715}
716
717void xenvif_carrier_off(struct xenvif *vif)
718{
719 struct net_device *dev = vif->dev;
720
721 rtnl_lock();
722 if (test_and_clear_bit(VIF_STATUS_CONNECTED, &vif->status)) {
723 netif_carrier_off(dev); /* discard queued packets */
724 if (netif_running(dev))
725 xenvif_down(vif);
726 }
727 rtnl_unlock();
728}
729
730void xenvif_disconnect_data(struct xenvif *vif)
731{
732 struct xenvif_queue *queue = NULL;
733 unsigned int num_queues = vif->num_queues;
734 unsigned int queue_index;
735
736 xenvif_carrier_off(vif);
737
738 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
739 queue = &vif->queues[queue_index];
740
741 netif_napi_del(&queue->napi);
742
743 if (queue->task) {
744 kthread_stop(queue->task);
745 put_task_struct(queue->task);
746 queue->task = NULL;
747 }
748
749 if (queue->dealloc_task) {
750 kthread_stop(queue->dealloc_task);
751 queue->dealloc_task = NULL;
752 }
753
754 if (queue->tx_irq) {
755 if (queue->tx_irq == queue->rx_irq)
756 unbind_from_irqhandler(queue->tx_irq, queue);
757 else {
758 unbind_from_irqhandler(queue->tx_irq, queue);
759 unbind_from_irqhandler(queue->rx_irq, queue);
760 }
761 queue->tx_irq = 0;
762 }
763
764 xenvif_unmap_frontend_data_rings(queue);
765 }
766
767 xenvif_mcast_addr_list_free(vif);
768}
769
770void xenvif_disconnect_ctrl(struct xenvif *vif)
771{
772 if (vif->ctrl_irq) {
773 xenvif_deinit_hash(vif);
774 unbind_from_irqhandler(vif->ctrl_irq, vif);
775 vif->ctrl_irq = 0;
776 }
777
778 if (vif->ctrl.sring) {
779 xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif),
780 vif->ctrl.sring);
781 vif->ctrl.sring = NULL;
782 }
783}
784
785/* Reverse the relevant parts of xenvif_init_queue().
786 * Used for queue teardown from xenvif_free(), and on the
787 * error handling paths in xenbus.c:connect().
788 */
789void xenvif_deinit_queue(struct xenvif_queue *queue)
790{
791 gnttab_free_pages(MAX_PENDING_REQS, queue->mmap_pages);
792}
793
794void xenvif_free(struct xenvif *vif)
795{
796 struct xenvif_queue *queues = vif->queues;
797 unsigned int num_queues = vif->num_queues;
798 unsigned int queue_index;
799
800 unregister_netdev(vif->dev);
801 free_netdev(vif->dev);
802
803 for (queue_index = 0; queue_index < num_queues; ++queue_index)
804 xenvif_deinit_queue(&queues[queue_index]);
805 vfree(queues);
806
807 module_put(THIS_MODULE);
808}
1/*
2 * Network-device interface management.
3 *
4 * Copyright (c) 2004-2005, Keir Fraser
5 *
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:
11 *
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:
18 *
19 * The above copyright notice and this permission notice shall be included in
20 * all copies or substantial portions of the Software.
21 *
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
28 * IN THE SOFTWARE.
29 */
30
31#include "common.h"
32
33#include <linux/kthread.h>
34#include <linux/sched/task.h>
35#include <linux/ethtool.h>
36#include <linux/rtnetlink.h>
37#include <linux/if_vlan.h>
38#include <linux/vmalloc.h>
39
40#include <xen/events.h>
41#include <asm/xen/hypercall.h>
42#include <xen/balloon.h>
43
44#define XENVIF_QUEUE_LENGTH 32
45#define XENVIF_NAPI_WEIGHT 64
46
47/* Number of bytes allowed on the internal guest Rx queue. */
48#define XENVIF_RX_QUEUE_BYTES (XEN_NETIF_RX_RING_SIZE/2 * PAGE_SIZE)
49
50/* This function is used to set SKBFL_ZEROCOPY_ENABLE as well as
51 * increasing the inflight counter. We need to increase the inflight
52 * counter because core driver calls into xenvif_zerocopy_callback
53 * which calls xenvif_skb_zerocopy_complete.
54 */
55void xenvif_skb_zerocopy_prepare(struct xenvif_queue *queue,
56 struct sk_buff *skb)
57{
58 skb_shinfo(skb)->flags |= SKBFL_ZEROCOPY_ENABLE;
59 atomic_inc(&queue->inflight_packets);
60}
61
62void xenvif_skb_zerocopy_complete(struct xenvif_queue *queue)
63{
64 atomic_dec(&queue->inflight_packets);
65
66 /* Wake the dealloc thread _after_ decrementing inflight_packets so
67 * that if kthread_stop() has already been called, the dealloc thread
68 * does not wait forever with nothing to wake it.
69 */
70 wake_up(&queue->dealloc_wq);
71}
72
73int xenvif_schedulable(struct xenvif *vif)
74{
75 return netif_running(vif->dev) &&
76 test_bit(VIF_STATUS_CONNECTED, &vif->status) &&
77 !vif->disabled;
78}
79
80static bool xenvif_handle_tx_interrupt(struct xenvif_queue *queue)
81{
82 bool rc;
83
84 rc = RING_HAS_UNCONSUMED_REQUESTS(&queue->tx);
85 if (rc)
86 napi_schedule(&queue->napi);
87 return rc;
88}
89
90static irqreturn_t xenvif_tx_interrupt(int irq, void *dev_id)
91{
92 struct xenvif_queue *queue = dev_id;
93 int old;
94
95 old = atomic_fetch_or(NETBK_TX_EOI, &queue->eoi_pending);
96 WARN(old & NETBK_TX_EOI, "Interrupt while EOI pending\n");
97
98 if (!xenvif_handle_tx_interrupt(queue)) {
99 atomic_andnot(NETBK_TX_EOI, &queue->eoi_pending);
100 xen_irq_lateeoi(irq, XEN_EOI_FLAG_SPURIOUS);
101 }
102
103 return IRQ_HANDLED;
104}
105
106static int xenvif_poll(struct napi_struct *napi, int budget)
107{
108 struct xenvif_queue *queue =
109 container_of(napi, struct xenvif_queue, napi);
110 int work_done;
111
112 /* This vif is rogue, we pretend we've there is nothing to do
113 * for this vif to deschedule it from NAPI. But this interface
114 * will be turned off in thread context later.
115 */
116 if (unlikely(queue->vif->disabled)) {
117 napi_complete(napi);
118 return 0;
119 }
120
121 work_done = xenvif_tx_action(queue, budget);
122
123 if (work_done < budget) {
124 napi_complete_done(napi, work_done);
125 /* If the queue is rate-limited, it shall be
126 * rescheduled in the timer callback.
127 */
128 if (likely(!queue->rate_limited))
129 xenvif_napi_schedule_or_enable_events(queue);
130 }
131
132 return work_done;
133}
134
135static bool xenvif_handle_rx_interrupt(struct xenvif_queue *queue)
136{
137 bool rc;
138
139 rc = xenvif_have_rx_work(queue, false);
140 if (rc)
141 xenvif_kick_thread(queue);
142 return rc;
143}
144
145static irqreturn_t xenvif_rx_interrupt(int irq, void *dev_id)
146{
147 struct xenvif_queue *queue = dev_id;
148 int old;
149
150 old = atomic_fetch_or(NETBK_RX_EOI, &queue->eoi_pending);
151 WARN(old & NETBK_RX_EOI, "Interrupt while EOI pending\n");
152
153 if (!xenvif_handle_rx_interrupt(queue)) {
154 atomic_andnot(NETBK_RX_EOI, &queue->eoi_pending);
155 xen_irq_lateeoi(irq, XEN_EOI_FLAG_SPURIOUS);
156 }
157
158 return IRQ_HANDLED;
159}
160
161irqreturn_t xenvif_interrupt(int irq, void *dev_id)
162{
163 struct xenvif_queue *queue = dev_id;
164 int old;
165 bool has_rx, has_tx;
166
167 old = atomic_fetch_or(NETBK_COMMON_EOI, &queue->eoi_pending);
168 WARN(old, "Interrupt while EOI pending\n");
169
170 has_tx = xenvif_handle_tx_interrupt(queue);
171 has_rx = xenvif_handle_rx_interrupt(queue);
172
173 if (!has_rx && !has_tx) {
174 atomic_andnot(NETBK_COMMON_EOI, &queue->eoi_pending);
175 xen_irq_lateeoi(irq, XEN_EOI_FLAG_SPURIOUS);
176 }
177
178 return IRQ_HANDLED;
179}
180
181int xenvif_queue_stopped(struct xenvif_queue *queue)
182{
183 struct net_device *dev = queue->vif->dev;
184 unsigned int id = queue->id;
185 return netif_tx_queue_stopped(netdev_get_tx_queue(dev, id));
186}
187
188void xenvif_wake_queue(struct xenvif_queue *queue)
189{
190 struct net_device *dev = queue->vif->dev;
191 unsigned int id = queue->id;
192 netif_tx_wake_queue(netdev_get_tx_queue(dev, id));
193}
194
195static u16 xenvif_select_queue(struct net_device *dev, struct sk_buff *skb,
196 struct net_device *sb_dev)
197{
198 struct xenvif *vif = netdev_priv(dev);
199 unsigned int size = vif->hash.size;
200 unsigned int num_queues;
201
202 /* If queues are not set up internally - always return 0
203 * as the packet going to be dropped anyway */
204 num_queues = READ_ONCE(vif->num_queues);
205 if (num_queues < 1)
206 return 0;
207
208 if (vif->hash.alg == XEN_NETIF_CTRL_HASH_ALGORITHM_NONE)
209 return netdev_pick_tx(dev, skb, NULL) %
210 dev->real_num_tx_queues;
211
212 xenvif_set_skb_hash(vif, skb);
213
214 if (size == 0)
215 return skb_get_hash_raw(skb) % dev->real_num_tx_queues;
216
217 return vif->hash.mapping[vif->hash.mapping_sel]
218 [skb_get_hash_raw(skb) % size];
219}
220
221static netdev_tx_t
222xenvif_start_xmit(struct sk_buff *skb, struct net_device *dev)
223{
224 struct xenvif *vif = netdev_priv(dev);
225 struct xenvif_queue *queue = NULL;
226 unsigned int num_queues;
227 u16 index;
228 struct xenvif_rx_cb *cb;
229
230 BUG_ON(skb->dev != dev);
231
232 /* Drop the packet if queues are not set up.
233 * This handler should be called inside an RCU read section
234 * so we don't need to enter it here explicitly.
235 */
236 num_queues = READ_ONCE(vif->num_queues);
237 if (num_queues < 1)
238 goto drop;
239
240 /* Obtain the queue to be used to transmit this packet */
241 index = skb_get_queue_mapping(skb);
242 if (index >= num_queues) {
243 pr_warn_ratelimited("Invalid queue %hu for packet on interface %s\n",
244 index, vif->dev->name);
245 index %= num_queues;
246 }
247 queue = &vif->queues[index];
248
249 /* Drop the packet if queue is not ready */
250 if (queue->task == NULL ||
251 queue->dealloc_task == NULL ||
252 !xenvif_schedulable(vif))
253 goto drop;
254
255 if (vif->multicast_control && skb->pkt_type == PACKET_MULTICAST) {
256 struct ethhdr *eth = (struct ethhdr *)skb->data;
257
258 if (!xenvif_mcast_match(vif, eth->h_dest))
259 goto drop;
260 }
261
262 cb = XENVIF_RX_CB(skb);
263 cb->expires = jiffies + vif->drain_timeout;
264
265 /* If there is no hash algorithm configured then make sure there
266 * is no hash information in the socket buffer otherwise it
267 * would be incorrectly forwarded to the frontend.
268 */
269 if (vif->hash.alg == XEN_NETIF_CTRL_HASH_ALGORITHM_NONE)
270 skb_clear_hash(skb);
271
272 xenvif_rx_queue_tail(queue, skb);
273 xenvif_kick_thread(queue);
274
275 return NETDEV_TX_OK;
276
277 drop:
278 vif->dev->stats.tx_dropped++;
279 dev_kfree_skb(skb);
280 return NETDEV_TX_OK;
281}
282
283static struct net_device_stats *xenvif_get_stats(struct net_device *dev)
284{
285 struct xenvif *vif = netdev_priv(dev);
286 struct xenvif_queue *queue = NULL;
287 unsigned int num_queues;
288 u64 rx_bytes = 0;
289 u64 rx_packets = 0;
290 u64 tx_bytes = 0;
291 u64 tx_packets = 0;
292 unsigned int index;
293
294 rcu_read_lock();
295 num_queues = READ_ONCE(vif->num_queues);
296
297 /* Aggregate tx and rx stats from each queue */
298 for (index = 0; index < num_queues; ++index) {
299 queue = &vif->queues[index];
300 rx_bytes += queue->stats.rx_bytes;
301 rx_packets += queue->stats.rx_packets;
302 tx_bytes += queue->stats.tx_bytes;
303 tx_packets += queue->stats.tx_packets;
304 }
305
306 rcu_read_unlock();
307
308 vif->dev->stats.rx_bytes = rx_bytes;
309 vif->dev->stats.rx_packets = rx_packets;
310 vif->dev->stats.tx_bytes = tx_bytes;
311 vif->dev->stats.tx_packets = tx_packets;
312
313 return &vif->dev->stats;
314}
315
316static void xenvif_up(struct xenvif *vif)
317{
318 struct xenvif_queue *queue = NULL;
319 unsigned int num_queues = vif->num_queues;
320 unsigned int queue_index;
321
322 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
323 queue = &vif->queues[queue_index];
324 napi_enable(&queue->napi);
325 enable_irq(queue->tx_irq);
326 if (queue->tx_irq != queue->rx_irq)
327 enable_irq(queue->rx_irq);
328 xenvif_napi_schedule_or_enable_events(queue);
329 }
330}
331
332static void xenvif_down(struct xenvif *vif)
333{
334 struct xenvif_queue *queue = NULL;
335 unsigned int num_queues = vif->num_queues;
336 unsigned int queue_index;
337
338 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
339 queue = &vif->queues[queue_index];
340 disable_irq(queue->tx_irq);
341 if (queue->tx_irq != queue->rx_irq)
342 disable_irq(queue->rx_irq);
343 napi_disable(&queue->napi);
344 del_timer_sync(&queue->credit_timeout);
345 }
346}
347
348static int xenvif_open(struct net_device *dev)
349{
350 struct xenvif *vif = netdev_priv(dev);
351 if (test_bit(VIF_STATUS_CONNECTED, &vif->status))
352 xenvif_up(vif);
353 netif_tx_start_all_queues(dev);
354 return 0;
355}
356
357static int xenvif_close(struct net_device *dev)
358{
359 struct xenvif *vif = netdev_priv(dev);
360 if (test_bit(VIF_STATUS_CONNECTED, &vif->status))
361 xenvif_down(vif);
362 netif_tx_stop_all_queues(dev);
363 return 0;
364}
365
366static int xenvif_change_mtu(struct net_device *dev, int mtu)
367{
368 struct xenvif *vif = netdev_priv(dev);
369 int max = vif->can_sg ? ETH_MAX_MTU - VLAN_ETH_HLEN : ETH_DATA_LEN;
370
371 if (mtu > max)
372 return -EINVAL;
373 dev->mtu = mtu;
374 return 0;
375}
376
377static netdev_features_t xenvif_fix_features(struct net_device *dev,
378 netdev_features_t features)
379{
380 struct xenvif *vif = netdev_priv(dev);
381
382 if (!vif->can_sg)
383 features &= ~NETIF_F_SG;
384 if (~(vif->gso_mask) & GSO_BIT(TCPV4))
385 features &= ~NETIF_F_TSO;
386 if (~(vif->gso_mask) & GSO_BIT(TCPV6))
387 features &= ~NETIF_F_TSO6;
388 if (!vif->ip_csum)
389 features &= ~NETIF_F_IP_CSUM;
390 if (!vif->ipv6_csum)
391 features &= ~NETIF_F_IPV6_CSUM;
392
393 return features;
394}
395
396static const struct xenvif_stat {
397 char name[ETH_GSTRING_LEN];
398 u16 offset;
399} xenvif_stats[] = {
400 {
401 "rx_gso_checksum_fixup",
402 offsetof(struct xenvif_stats, rx_gso_checksum_fixup)
403 },
404 /* If (sent != success + fail), there are probably packets never
405 * freed up properly!
406 */
407 {
408 "tx_zerocopy_sent",
409 offsetof(struct xenvif_stats, tx_zerocopy_sent),
410 },
411 {
412 "tx_zerocopy_success",
413 offsetof(struct xenvif_stats, tx_zerocopy_success),
414 },
415 {
416 "tx_zerocopy_fail",
417 offsetof(struct xenvif_stats, tx_zerocopy_fail)
418 },
419 /* Number of packets exceeding MAX_SKB_FRAG slots. You should use
420 * a guest with the same MAX_SKB_FRAG
421 */
422 {
423 "tx_frag_overflow",
424 offsetof(struct xenvif_stats, tx_frag_overflow)
425 },
426};
427
428static int xenvif_get_sset_count(struct net_device *dev, int string_set)
429{
430 switch (string_set) {
431 case ETH_SS_STATS:
432 return ARRAY_SIZE(xenvif_stats);
433 default:
434 return -EINVAL;
435 }
436}
437
438static void xenvif_get_ethtool_stats(struct net_device *dev,
439 struct ethtool_stats *stats, u64 * data)
440{
441 struct xenvif *vif = netdev_priv(dev);
442 unsigned int num_queues;
443 int i;
444 unsigned int queue_index;
445
446 rcu_read_lock();
447 num_queues = READ_ONCE(vif->num_queues);
448
449 for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++) {
450 unsigned long accum = 0;
451 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
452 void *vif_stats = &vif->queues[queue_index].stats;
453 accum += *(unsigned long *)(vif_stats + xenvif_stats[i].offset);
454 }
455 data[i] = accum;
456 }
457
458 rcu_read_unlock();
459}
460
461static void xenvif_get_strings(struct net_device *dev, u32 stringset, u8 * data)
462{
463 int i;
464
465 switch (stringset) {
466 case ETH_SS_STATS:
467 for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++)
468 memcpy(data + i * ETH_GSTRING_LEN,
469 xenvif_stats[i].name, ETH_GSTRING_LEN);
470 break;
471 }
472}
473
474static const struct ethtool_ops xenvif_ethtool_ops = {
475 .get_link = ethtool_op_get_link,
476
477 .get_sset_count = xenvif_get_sset_count,
478 .get_ethtool_stats = xenvif_get_ethtool_stats,
479 .get_strings = xenvif_get_strings,
480};
481
482static const struct net_device_ops xenvif_netdev_ops = {
483 .ndo_select_queue = xenvif_select_queue,
484 .ndo_start_xmit = xenvif_start_xmit,
485 .ndo_get_stats = xenvif_get_stats,
486 .ndo_open = xenvif_open,
487 .ndo_stop = xenvif_close,
488 .ndo_change_mtu = xenvif_change_mtu,
489 .ndo_fix_features = xenvif_fix_features,
490 .ndo_set_mac_address = eth_mac_addr,
491 .ndo_validate_addr = eth_validate_addr,
492};
493
494struct xenvif *xenvif_alloc(struct device *parent, domid_t domid,
495 unsigned int handle)
496{
497 int err;
498 struct net_device *dev;
499 struct xenvif *vif;
500 char name[IFNAMSIZ] = {};
501
502 snprintf(name, IFNAMSIZ - 1, "vif%u.%u", domid, handle);
503 /* Allocate a netdev with the max. supported number of queues.
504 * When the guest selects the desired number, it will be updated
505 * via netif_set_real_num_*_queues().
506 */
507 dev = alloc_netdev_mq(sizeof(struct xenvif), name, NET_NAME_UNKNOWN,
508 ether_setup, xenvif_max_queues);
509 if (dev == NULL) {
510 pr_warn("Could not allocate netdev for %s\n", name);
511 return ERR_PTR(-ENOMEM);
512 }
513
514 SET_NETDEV_DEV(dev, parent);
515
516 vif = netdev_priv(dev);
517
518 vif->domid = domid;
519 vif->handle = handle;
520 vif->can_sg = 1;
521 vif->ip_csum = 1;
522 vif->dev = dev;
523 vif->disabled = false;
524 vif->drain_timeout = msecs_to_jiffies(rx_drain_timeout_msecs);
525 vif->stall_timeout = msecs_to_jiffies(rx_stall_timeout_msecs);
526
527 /* Start out with no queues. */
528 vif->queues = NULL;
529 vif->num_queues = 0;
530
531 vif->xdp_headroom = 0;
532
533 spin_lock_init(&vif->lock);
534 INIT_LIST_HEAD(&vif->fe_mcast_addr);
535
536 dev->netdev_ops = &xenvif_netdev_ops;
537 dev->hw_features = NETIF_F_SG |
538 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
539 NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_FRAGLIST;
540 dev->features = dev->hw_features | NETIF_F_RXCSUM;
541 dev->ethtool_ops = &xenvif_ethtool_ops;
542
543 dev->tx_queue_len = XENVIF_QUEUE_LENGTH;
544
545 dev->min_mtu = ETH_MIN_MTU;
546 dev->max_mtu = ETH_MAX_MTU - VLAN_ETH_HLEN;
547
548 /*
549 * Initialise a dummy MAC address. We choose the numerically
550 * largest non-broadcast address to prevent the address getting
551 * stolen by an Ethernet bridge for STP purposes.
552 * (FE:FF:FF:FF:FF:FF)
553 */
554 eth_broadcast_addr(dev->dev_addr);
555 dev->dev_addr[0] &= ~0x01;
556
557 netif_carrier_off(dev);
558
559 err = register_netdev(dev);
560 if (err) {
561 netdev_warn(dev, "Could not register device: err=%d\n", err);
562 free_netdev(dev);
563 return ERR_PTR(err);
564 }
565
566 netdev_dbg(dev, "Successfully created xenvif\n");
567
568 __module_get(THIS_MODULE);
569
570 return vif;
571}
572
573int xenvif_init_queue(struct xenvif_queue *queue)
574{
575 int err, i;
576
577 queue->credit_bytes = queue->remaining_credit = ~0UL;
578 queue->credit_usec = 0UL;
579 timer_setup(&queue->credit_timeout, xenvif_tx_credit_callback, 0);
580 queue->credit_window_start = get_jiffies_64();
581
582 queue->rx_queue_max = XENVIF_RX_QUEUE_BYTES;
583
584 skb_queue_head_init(&queue->rx_queue);
585 skb_queue_head_init(&queue->tx_queue);
586
587 queue->pending_cons = 0;
588 queue->pending_prod = MAX_PENDING_REQS;
589 for (i = 0; i < MAX_PENDING_REQS; ++i)
590 queue->pending_ring[i] = i;
591
592 spin_lock_init(&queue->callback_lock);
593 spin_lock_init(&queue->response_lock);
594
595 /* If ballooning is disabled, this will consume real memory, so you
596 * better enable it. The long term solution would be to use just a
597 * bunch of valid page descriptors, without dependency on ballooning
598 */
599 err = gnttab_alloc_pages(MAX_PENDING_REQS,
600 queue->mmap_pages);
601 if (err) {
602 netdev_err(queue->vif->dev, "Could not reserve mmap_pages\n");
603 return -ENOMEM;
604 }
605
606 for (i = 0; i < MAX_PENDING_REQS; i++) {
607 queue->pending_tx_info[i].callback_struct = (struct ubuf_info)
608 { .callback = xenvif_zerocopy_callback,
609 { { .ctx = NULL,
610 .desc = i } } };
611 queue->grant_tx_handle[i] = NETBACK_INVALID_HANDLE;
612 }
613
614 return 0;
615}
616
617void xenvif_carrier_on(struct xenvif *vif)
618{
619 rtnl_lock();
620 if (!vif->can_sg && vif->dev->mtu > ETH_DATA_LEN)
621 dev_set_mtu(vif->dev, ETH_DATA_LEN);
622 netdev_update_features(vif->dev);
623 set_bit(VIF_STATUS_CONNECTED, &vif->status);
624 if (netif_running(vif->dev))
625 xenvif_up(vif);
626 rtnl_unlock();
627}
628
629int xenvif_connect_ctrl(struct xenvif *vif, grant_ref_t ring_ref,
630 unsigned int evtchn)
631{
632 struct net_device *dev = vif->dev;
633 struct xenbus_device *xendev = xenvif_to_xenbus_device(vif);
634 void *addr;
635 struct xen_netif_ctrl_sring *shared;
636 RING_IDX rsp_prod, req_prod;
637 int err;
638
639 err = xenbus_map_ring_valloc(xendev, &ring_ref, 1, &addr);
640 if (err)
641 goto err;
642
643 shared = (struct xen_netif_ctrl_sring *)addr;
644 rsp_prod = READ_ONCE(shared->rsp_prod);
645 req_prod = READ_ONCE(shared->req_prod);
646
647 BACK_RING_ATTACH(&vif->ctrl, shared, rsp_prod, XEN_PAGE_SIZE);
648
649 err = -EIO;
650 if (req_prod - rsp_prod > RING_SIZE(&vif->ctrl))
651 goto err_unmap;
652
653 err = bind_interdomain_evtchn_to_irq_lateeoi(xendev, evtchn);
654 if (err < 0)
655 goto err_unmap;
656
657 vif->ctrl_irq = err;
658
659 xenvif_init_hash(vif);
660
661 err = request_threaded_irq(vif->ctrl_irq, NULL, xenvif_ctrl_irq_fn,
662 IRQF_ONESHOT, "xen-netback-ctrl", vif);
663 if (err) {
664 pr_warn("Could not setup irq handler for %s\n", dev->name);
665 goto err_deinit;
666 }
667
668 return 0;
669
670err_deinit:
671 xenvif_deinit_hash(vif);
672 unbind_from_irqhandler(vif->ctrl_irq, vif);
673 vif->ctrl_irq = 0;
674
675err_unmap:
676 xenbus_unmap_ring_vfree(xendev, vif->ctrl.sring);
677 vif->ctrl.sring = NULL;
678
679err:
680 return err;
681}
682
683static void xenvif_disconnect_queue(struct xenvif_queue *queue)
684{
685 if (queue->task) {
686 kthread_stop(queue->task);
687 put_task_struct(queue->task);
688 queue->task = NULL;
689 }
690
691 if (queue->dealloc_task) {
692 kthread_stop(queue->dealloc_task);
693 queue->dealloc_task = NULL;
694 }
695
696 if (queue->napi.poll) {
697 netif_napi_del(&queue->napi);
698 queue->napi.poll = NULL;
699 }
700
701 if (queue->tx_irq) {
702 unbind_from_irqhandler(queue->tx_irq, queue);
703 if (queue->tx_irq == queue->rx_irq)
704 queue->rx_irq = 0;
705 queue->tx_irq = 0;
706 }
707
708 if (queue->rx_irq) {
709 unbind_from_irqhandler(queue->rx_irq, queue);
710 queue->rx_irq = 0;
711 }
712
713 xenvif_unmap_frontend_data_rings(queue);
714}
715
716int xenvif_connect_data(struct xenvif_queue *queue,
717 unsigned long tx_ring_ref,
718 unsigned long rx_ring_ref,
719 unsigned int tx_evtchn,
720 unsigned int rx_evtchn)
721{
722 struct xenbus_device *dev = xenvif_to_xenbus_device(queue->vif);
723 struct task_struct *task;
724 int err;
725
726 BUG_ON(queue->tx_irq);
727 BUG_ON(queue->task);
728 BUG_ON(queue->dealloc_task);
729
730 err = xenvif_map_frontend_data_rings(queue, tx_ring_ref,
731 rx_ring_ref);
732 if (err < 0)
733 goto err;
734
735 init_waitqueue_head(&queue->wq);
736 init_waitqueue_head(&queue->dealloc_wq);
737 atomic_set(&queue->inflight_packets, 0);
738
739 netif_napi_add(queue->vif->dev, &queue->napi, xenvif_poll,
740 XENVIF_NAPI_WEIGHT);
741
742 queue->stalled = true;
743
744 task = kthread_run(xenvif_kthread_guest_rx, queue,
745 "%s-guest-rx", queue->name);
746 if (IS_ERR(task))
747 goto kthread_err;
748 queue->task = task;
749 /*
750 * Take a reference to the task in order to prevent it from being freed
751 * if the thread function returns before kthread_stop is called.
752 */
753 get_task_struct(task);
754
755 task = kthread_run(xenvif_dealloc_kthread, queue,
756 "%s-dealloc", queue->name);
757 if (IS_ERR(task))
758 goto kthread_err;
759 queue->dealloc_task = task;
760
761 if (tx_evtchn == rx_evtchn) {
762 /* feature-split-event-channels == 0 */
763 err = bind_interdomain_evtchn_to_irqhandler_lateeoi(
764 dev, tx_evtchn, xenvif_interrupt, 0,
765 queue->name, queue);
766 if (err < 0)
767 goto err;
768 queue->tx_irq = queue->rx_irq = err;
769 disable_irq(queue->tx_irq);
770 } else {
771 /* feature-split-event-channels == 1 */
772 snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
773 "%s-tx", queue->name);
774 err = bind_interdomain_evtchn_to_irqhandler_lateeoi(
775 dev, tx_evtchn, xenvif_tx_interrupt, 0,
776 queue->tx_irq_name, queue);
777 if (err < 0)
778 goto err;
779 queue->tx_irq = err;
780 disable_irq(queue->tx_irq);
781
782 snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
783 "%s-rx", queue->name);
784 err = bind_interdomain_evtchn_to_irqhandler_lateeoi(
785 dev, rx_evtchn, xenvif_rx_interrupt, 0,
786 queue->rx_irq_name, queue);
787 if (err < 0)
788 goto err;
789 queue->rx_irq = err;
790 disable_irq(queue->rx_irq);
791 }
792
793 return 0;
794
795kthread_err:
796 pr_warn("Could not allocate kthread for %s\n", queue->name);
797 err = PTR_ERR(task);
798err:
799 xenvif_disconnect_queue(queue);
800 return err;
801}
802
803void xenvif_carrier_off(struct xenvif *vif)
804{
805 struct net_device *dev = vif->dev;
806
807 rtnl_lock();
808 if (test_and_clear_bit(VIF_STATUS_CONNECTED, &vif->status)) {
809 netif_carrier_off(dev); /* discard queued packets */
810 if (netif_running(dev))
811 xenvif_down(vif);
812 }
813 rtnl_unlock();
814}
815
816void xenvif_disconnect_data(struct xenvif *vif)
817{
818 struct xenvif_queue *queue = NULL;
819 unsigned int num_queues = vif->num_queues;
820 unsigned int queue_index;
821
822 xenvif_carrier_off(vif);
823
824 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
825 queue = &vif->queues[queue_index];
826
827 xenvif_disconnect_queue(queue);
828 }
829
830 xenvif_mcast_addr_list_free(vif);
831}
832
833void xenvif_disconnect_ctrl(struct xenvif *vif)
834{
835 if (vif->ctrl_irq) {
836 xenvif_deinit_hash(vif);
837 unbind_from_irqhandler(vif->ctrl_irq, vif);
838 vif->ctrl_irq = 0;
839 }
840
841 if (vif->ctrl.sring) {
842 xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif),
843 vif->ctrl.sring);
844 vif->ctrl.sring = NULL;
845 }
846}
847
848/* Reverse the relevant parts of xenvif_init_queue().
849 * Used for queue teardown from xenvif_free(), and on the
850 * error handling paths in xenbus.c:connect().
851 */
852void xenvif_deinit_queue(struct xenvif_queue *queue)
853{
854 gnttab_free_pages(MAX_PENDING_REQS, queue->mmap_pages);
855}
856
857void xenvif_free(struct xenvif *vif)
858{
859 struct xenvif_queue *queues = vif->queues;
860 unsigned int num_queues = vif->num_queues;
861 unsigned int queue_index;
862
863 unregister_netdev(vif->dev);
864 free_netdev(vif->dev);
865
866 for (queue_index = 0; queue_index < num_queues; ++queue_index)
867 xenvif_deinit_queue(&queues[queue_index]);
868 vfree(queues);
869
870 module_put(THIS_MODULE);
871}