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/ethtool.h>
35#include <linux/rtnetlink.h>
36#include <linux/if_vlan.h>
37#include <linux/vmalloc.h>
38
39#include <xen/events.h>
40#include <asm/xen/hypercall.h>
41#include <xen/balloon.h>
42
43#define XENVIF_QUEUE_LENGTH 32
44#define XENVIF_NAPI_WEIGHT 64
45
46int xenvif_schedulable(struct xenvif *vif)
47{
48 return netif_running(vif->dev) && netif_carrier_ok(vif->dev);
49}
50
51static irqreturn_t xenvif_tx_interrupt(int irq, void *dev_id)
52{
53 struct xenvif *vif = dev_id;
54
55 if (RING_HAS_UNCONSUMED_REQUESTS(&vif->tx))
56 napi_schedule(&vif->napi);
57
58 return IRQ_HANDLED;
59}
60
61static int xenvif_poll(struct napi_struct *napi, int budget)
62{
63 struct xenvif *vif = container_of(napi, struct xenvif, napi);
64 int work_done;
65
66 /* This vif is rogue, we pretend we've there is nothing to do
67 * for this vif to deschedule it from NAPI. But this interface
68 * will be turned off in thread context later.
69 */
70 if (unlikely(vif->disabled)) {
71 napi_complete(napi);
72 return 0;
73 }
74
75 work_done = xenvif_tx_action(vif, budget);
76
77 if (work_done < budget) {
78 napi_complete(napi);
79 xenvif_napi_schedule_or_enable_events(vif);
80 }
81
82 return work_done;
83}
84
85static irqreturn_t xenvif_rx_interrupt(int irq, void *dev_id)
86{
87 struct xenvif *vif = dev_id;
88
89 xenvif_kick_thread(vif);
90
91 return IRQ_HANDLED;
92}
93
94static irqreturn_t xenvif_interrupt(int irq, void *dev_id)
95{
96 xenvif_tx_interrupt(irq, dev_id);
97 xenvif_rx_interrupt(irq, dev_id);
98
99 return IRQ_HANDLED;
100}
101
102static void xenvif_wake_queue(unsigned long data)
103{
104 struct xenvif *vif = (struct xenvif *)data;
105
106 if (netif_queue_stopped(vif->dev)) {
107 netdev_err(vif->dev, "draining TX queue\n");
108 vif->rx_queue_purge = true;
109 xenvif_kick_thread(vif);
110 netif_wake_queue(vif->dev);
111 }
112}
113
114static int xenvif_start_xmit(struct sk_buff *skb, struct net_device *dev)
115{
116 struct xenvif *vif = netdev_priv(dev);
117 int min_slots_needed;
118
119 BUG_ON(skb->dev != dev);
120
121 /* Drop the packet if vif is not ready */
122 if (vif->task == NULL ||
123 vif->dealloc_task == NULL ||
124 !xenvif_schedulable(vif))
125 goto drop;
126
127 /* At best we'll need one slot for the header and one for each
128 * frag.
129 */
130 min_slots_needed = 1 + skb_shinfo(skb)->nr_frags;
131
132 /* If the skb is GSO then we'll also need an extra slot for the
133 * metadata.
134 */
135 if (skb_is_gso(skb))
136 min_slots_needed++;
137
138 /* If the skb can't possibly fit in the remaining slots
139 * then turn off the queue to give the ring a chance to
140 * drain.
141 */
142 if (!xenvif_rx_ring_slots_available(vif, min_slots_needed)) {
143 vif->wake_queue.function = xenvif_wake_queue;
144 vif->wake_queue.data = (unsigned long)vif;
145 xenvif_stop_queue(vif);
146 mod_timer(&vif->wake_queue,
147 jiffies + rx_drain_timeout_jiffies);
148 }
149
150 skb_queue_tail(&vif->rx_queue, skb);
151 xenvif_kick_thread(vif);
152
153 return NETDEV_TX_OK;
154
155 drop:
156 vif->dev->stats.tx_dropped++;
157 dev_kfree_skb(skb);
158 return NETDEV_TX_OK;
159}
160
161static struct net_device_stats *xenvif_get_stats(struct net_device *dev)
162{
163 struct xenvif *vif = netdev_priv(dev);
164 return &vif->dev->stats;
165}
166
167static void xenvif_up(struct xenvif *vif)
168{
169 napi_enable(&vif->napi);
170 enable_irq(vif->tx_irq);
171 if (vif->tx_irq != vif->rx_irq)
172 enable_irq(vif->rx_irq);
173 xenvif_napi_schedule_or_enable_events(vif);
174}
175
176static void xenvif_down(struct xenvif *vif)
177{
178 napi_disable(&vif->napi);
179 disable_irq(vif->tx_irq);
180 if (vif->tx_irq != vif->rx_irq)
181 disable_irq(vif->rx_irq);
182 del_timer_sync(&vif->credit_timeout);
183}
184
185static int xenvif_open(struct net_device *dev)
186{
187 struct xenvif *vif = netdev_priv(dev);
188 if (netif_carrier_ok(dev))
189 xenvif_up(vif);
190 netif_start_queue(dev);
191 return 0;
192}
193
194static int xenvif_close(struct net_device *dev)
195{
196 struct xenvif *vif = netdev_priv(dev);
197 if (netif_carrier_ok(dev))
198 xenvif_down(vif);
199 netif_stop_queue(dev);
200 return 0;
201}
202
203static int xenvif_change_mtu(struct net_device *dev, int mtu)
204{
205 struct xenvif *vif = netdev_priv(dev);
206 int max = vif->can_sg ? 65535 - VLAN_ETH_HLEN : ETH_DATA_LEN;
207
208 if (mtu > max)
209 return -EINVAL;
210 dev->mtu = mtu;
211 return 0;
212}
213
214static netdev_features_t xenvif_fix_features(struct net_device *dev,
215 netdev_features_t features)
216{
217 struct xenvif *vif = netdev_priv(dev);
218
219 if (!vif->can_sg)
220 features &= ~NETIF_F_SG;
221 if (~(vif->gso_mask | vif->gso_prefix_mask) & GSO_BIT(TCPV4))
222 features &= ~NETIF_F_TSO;
223 if (~(vif->gso_mask | vif->gso_prefix_mask) & GSO_BIT(TCPV6))
224 features &= ~NETIF_F_TSO6;
225 if (!vif->ip_csum)
226 features &= ~NETIF_F_IP_CSUM;
227 if (!vif->ipv6_csum)
228 features &= ~NETIF_F_IPV6_CSUM;
229
230 return features;
231}
232
233static const struct xenvif_stat {
234 char name[ETH_GSTRING_LEN];
235 u16 offset;
236} xenvif_stats[] = {
237 {
238 "rx_gso_checksum_fixup",
239 offsetof(struct xenvif, rx_gso_checksum_fixup)
240 },
241 /* If (sent != success + fail), there are probably packets never
242 * freed up properly!
243 */
244 {
245 "tx_zerocopy_sent",
246 offsetof(struct xenvif, tx_zerocopy_sent),
247 },
248 {
249 "tx_zerocopy_success",
250 offsetof(struct xenvif, tx_zerocopy_success),
251 },
252 {
253 "tx_zerocopy_fail",
254 offsetof(struct xenvif, tx_zerocopy_fail)
255 },
256 /* Number of packets exceeding MAX_SKB_FRAG slots. You should use
257 * a guest with the same MAX_SKB_FRAG
258 */
259 {
260 "tx_frag_overflow",
261 offsetof(struct xenvif, tx_frag_overflow)
262 },
263};
264
265static int xenvif_get_sset_count(struct net_device *dev, int string_set)
266{
267 switch (string_set) {
268 case ETH_SS_STATS:
269 return ARRAY_SIZE(xenvif_stats);
270 default:
271 return -EINVAL;
272 }
273}
274
275static void xenvif_get_ethtool_stats(struct net_device *dev,
276 struct ethtool_stats *stats, u64 * data)
277{
278 void *vif = netdev_priv(dev);
279 int i;
280
281 for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++)
282 data[i] = *(unsigned long *)(vif + xenvif_stats[i].offset);
283}
284
285static void xenvif_get_strings(struct net_device *dev, u32 stringset, u8 * data)
286{
287 int i;
288
289 switch (stringset) {
290 case ETH_SS_STATS:
291 for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++)
292 memcpy(data + i * ETH_GSTRING_LEN,
293 xenvif_stats[i].name, ETH_GSTRING_LEN);
294 break;
295 }
296}
297
298static const struct ethtool_ops xenvif_ethtool_ops = {
299 .get_link = ethtool_op_get_link,
300
301 .get_sset_count = xenvif_get_sset_count,
302 .get_ethtool_stats = xenvif_get_ethtool_stats,
303 .get_strings = xenvif_get_strings,
304};
305
306static const struct net_device_ops xenvif_netdev_ops = {
307 .ndo_start_xmit = xenvif_start_xmit,
308 .ndo_get_stats = xenvif_get_stats,
309 .ndo_open = xenvif_open,
310 .ndo_stop = xenvif_close,
311 .ndo_change_mtu = xenvif_change_mtu,
312 .ndo_fix_features = xenvif_fix_features,
313 .ndo_set_mac_address = eth_mac_addr,
314 .ndo_validate_addr = eth_validate_addr,
315};
316
317struct xenvif *xenvif_alloc(struct device *parent, domid_t domid,
318 unsigned int handle)
319{
320 int err;
321 struct net_device *dev;
322 struct xenvif *vif;
323 char name[IFNAMSIZ] = {};
324 int i;
325
326 snprintf(name, IFNAMSIZ - 1, "vif%u.%u", domid, handle);
327 dev = alloc_netdev(sizeof(struct xenvif), name, ether_setup);
328 if (dev == NULL) {
329 pr_warn("Could not allocate netdev for %s\n", name);
330 return ERR_PTR(-ENOMEM);
331 }
332
333 SET_NETDEV_DEV(dev, parent);
334
335 vif = netdev_priv(dev);
336
337 vif->grant_copy_op = vmalloc(sizeof(struct gnttab_copy) *
338 MAX_GRANT_COPY_OPS);
339 if (vif->grant_copy_op == NULL) {
340 pr_warn("Could not allocate grant copy space for %s\n", name);
341 free_netdev(dev);
342 return ERR_PTR(-ENOMEM);
343 }
344
345 vif->domid = domid;
346 vif->handle = handle;
347 vif->can_sg = 1;
348 vif->ip_csum = 1;
349 vif->dev = dev;
350
351 vif->disabled = false;
352
353 vif->credit_bytes = vif->remaining_credit = ~0UL;
354 vif->credit_usec = 0UL;
355 init_timer(&vif->credit_timeout);
356 vif->credit_window_start = get_jiffies_64();
357
358 init_timer(&vif->wake_queue);
359
360 dev->netdev_ops = &xenvif_netdev_ops;
361 dev->hw_features = NETIF_F_SG |
362 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
363 NETIF_F_TSO | NETIF_F_TSO6;
364 dev->features = dev->hw_features | NETIF_F_RXCSUM;
365 SET_ETHTOOL_OPS(dev, &xenvif_ethtool_ops);
366
367 dev->tx_queue_len = XENVIF_QUEUE_LENGTH;
368
369 skb_queue_head_init(&vif->rx_queue);
370 skb_queue_head_init(&vif->tx_queue);
371
372 vif->pending_cons = 0;
373 vif->pending_prod = MAX_PENDING_REQS;
374 for (i = 0; i < MAX_PENDING_REQS; i++)
375 vif->pending_ring[i] = i;
376 spin_lock_init(&vif->callback_lock);
377 spin_lock_init(&vif->response_lock);
378 /* If ballooning is disabled, this will consume real memory, so you
379 * better enable it. The long term solution would be to use just a
380 * bunch of valid page descriptors, without dependency on ballooning
381 */
382 err = alloc_xenballooned_pages(MAX_PENDING_REQS,
383 vif->mmap_pages,
384 false);
385 if (err) {
386 netdev_err(dev, "Could not reserve mmap_pages\n");
387 return ERR_PTR(-ENOMEM);
388 }
389 for (i = 0; i < MAX_PENDING_REQS; i++) {
390 vif->pending_tx_info[i].callback_struct = (struct ubuf_info)
391 { .callback = xenvif_zerocopy_callback,
392 .ctx = NULL,
393 .desc = i };
394 vif->grant_tx_handle[i] = NETBACK_INVALID_HANDLE;
395 }
396
397 /*
398 * Initialise a dummy MAC address. We choose the numerically
399 * largest non-broadcast address to prevent the address getting
400 * stolen by an Ethernet bridge for STP purposes.
401 * (FE:FF:FF:FF:FF:FF)
402 */
403 memset(dev->dev_addr, 0xFF, ETH_ALEN);
404 dev->dev_addr[0] &= ~0x01;
405
406 netif_napi_add(dev, &vif->napi, xenvif_poll, XENVIF_NAPI_WEIGHT);
407
408 netif_carrier_off(dev);
409
410 err = register_netdev(dev);
411 if (err) {
412 netdev_warn(dev, "Could not register device: err=%d\n", err);
413 free_netdev(dev);
414 return ERR_PTR(err);
415 }
416
417 netdev_dbg(dev, "Successfully created xenvif\n");
418
419 __module_get(THIS_MODULE);
420
421 return vif;
422}
423
424int xenvif_connect(struct xenvif *vif, unsigned long tx_ring_ref,
425 unsigned long rx_ring_ref, unsigned int tx_evtchn,
426 unsigned int rx_evtchn)
427{
428 struct task_struct *task;
429 int err = -ENOMEM;
430
431 BUG_ON(vif->tx_irq);
432 BUG_ON(vif->task);
433 BUG_ON(vif->dealloc_task);
434
435 err = xenvif_map_frontend_rings(vif, tx_ring_ref, rx_ring_ref);
436 if (err < 0)
437 goto err;
438
439 init_waitqueue_head(&vif->wq);
440 init_waitqueue_head(&vif->dealloc_wq);
441
442 if (tx_evtchn == rx_evtchn) {
443 /* feature-split-event-channels == 0 */
444 err = bind_interdomain_evtchn_to_irqhandler(
445 vif->domid, tx_evtchn, xenvif_interrupt, 0,
446 vif->dev->name, vif);
447 if (err < 0)
448 goto err_unmap;
449 vif->tx_irq = vif->rx_irq = err;
450 disable_irq(vif->tx_irq);
451 } else {
452 /* feature-split-event-channels == 1 */
453 snprintf(vif->tx_irq_name, sizeof(vif->tx_irq_name),
454 "%s-tx", vif->dev->name);
455 err = bind_interdomain_evtchn_to_irqhandler(
456 vif->domid, tx_evtchn, xenvif_tx_interrupt, 0,
457 vif->tx_irq_name, vif);
458 if (err < 0)
459 goto err_unmap;
460 vif->tx_irq = err;
461 disable_irq(vif->tx_irq);
462
463 snprintf(vif->rx_irq_name, sizeof(vif->rx_irq_name),
464 "%s-rx", vif->dev->name);
465 err = bind_interdomain_evtchn_to_irqhandler(
466 vif->domid, rx_evtchn, xenvif_rx_interrupt, 0,
467 vif->rx_irq_name, vif);
468 if (err < 0)
469 goto err_tx_unbind;
470 vif->rx_irq = err;
471 disable_irq(vif->rx_irq);
472 }
473
474 task = kthread_create(xenvif_kthread_guest_rx,
475 (void *)vif, "%s-guest-rx", vif->dev->name);
476 if (IS_ERR(task)) {
477 pr_warn("Could not allocate kthread for %s\n", vif->dev->name);
478 err = PTR_ERR(task);
479 goto err_rx_unbind;
480 }
481
482 vif->task = task;
483
484 task = kthread_create(xenvif_dealloc_kthread,
485 (void *)vif, "%s-dealloc", vif->dev->name);
486 if (IS_ERR(task)) {
487 pr_warn("Could not allocate kthread for %s\n", vif->dev->name);
488 err = PTR_ERR(task);
489 goto err_rx_unbind;
490 }
491
492 vif->dealloc_task = task;
493
494 rtnl_lock();
495 if (!vif->can_sg && vif->dev->mtu > ETH_DATA_LEN)
496 dev_set_mtu(vif->dev, ETH_DATA_LEN);
497 netdev_update_features(vif->dev);
498 netif_carrier_on(vif->dev);
499 if (netif_running(vif->dev))
500 xenvif_up(vif);
501 rtnl_unlock();
502
503 wake_up_process(vif->task);
504 wake_up_process(vif->dealloc_task);
505
506 return 0;
507
508err_rx_unbind:
509 unbind_from_irqhandler(vif->rx_irq, vif);
510 vif->rx_irq = 0;
511err_tx_unbind:
512 unbind_from_irqhandler(vif->tx_irq, vif);
513 vif->tx_irq = 0;
514err_unmap:
515 xenvif_unmap_frontend_rings(vif);
516err:
517 module_put(THIS_MODULE);
518 return err;
519}
520
521void xenvif_carrier_off(struct xenvif *vif)
522{
523 struct net_device *dev = vif->dev;
524
525 rtnl_lock();
526 netif_carrier_off(dev); /* discard queued packets */
527 if (netif_running(dev))
528 xenvif_down(vif);
529 rtnl_unlock();
530}
531
532void xenvif_disconnect(struct xenvif *vif)
533{
534 if (netif_carrier_ok(vif->dev))
535 xenvif_carrier_off(vif);
536
537 if (vif->task) {
538 del_timer_sync(&vif->wake_queue);
539 kthread_stop(vif->task);
540 vif->task = NULL;
541 }
542
543 if (vif->dealloc_task) {
544 kthread_stop(vif->dealloc_task);
545 vif->dealloc_task = NULL;
546 }
547
548 if (vif->tx_irq) {
549 if (vif->tx_irq == vif->rx_irq)
550 unbind_from_irqhandler(vif->tx_irq, vif);
551 else {
552 unbind_from_irqhandler(vif->tx_irq, vif);
553 unbind_from_irqhandler(vif->rx_irq, vif);
554 }
555 vif->tx_irq = 0;
556 }
557
558 xenvif_unmap_frontend_rings(vif);
559}
560
561void xenvif_free(struct xenvif *vif)
562{
563 int i, unmap_timeout = 0;
564 /* Here we want to avoid timeout messages if an skb can be legitimately
565 * stuck somewhere else. Realistically this could be an another vif's
566 * internal or QDisc queue. That another vif also has this
567 * rx_drain_timeout_msecs timeout, but the timer only ditches the
568 * internal queue. After that, the QDisc queue can put in worst case
569 * XEN_NETIF_RX_RING_SIZE / MAX_SKB_FRAGS skbs into that another vif's
570 * internal queue, so we need several rounds of such timeouts until we
571 * can be sure that no another vif should have skb's from us. We are
572 * not sending more skb's, so newly stuck packets are not interesting
573 * for us here.
574 */
575 unsigned int worst_case_skb_lifetime = (rx_drain_timeout_msecs/1000) *
576 DIV_ROUND_UP(XENVIF_QUEUE_LENGTH, (XEN_NETIF_RX_RING_SIZE / MAX_SKB_FRAGS));
577
578 for (i = 0; i < MAX_PENDING_REQS; ++i) {
579 if (vif->grant_tx_handle[i] != NETBACK_INVALID_HANDLE) {
580 unmap_timeout++;
581 schedule_timeout(msecs_to_jiffies(1000));
582 if (unmap_timeout > worst_case_skb_lifetime &&
583 net_ratelimit())
584 netdev_err(vif->dev,
585 "Page still granted! Index: %x\n",
586 i);
587 /* If there are still unmapped pages, reset the loop to
588 * start checking again. We shouldn't exit here until
589 * dealloc thread and NAPI instance release all the
590 * pages. If a kernel bug causes the skbs to stall
591 * somewhere, the interface cannot be brought down
592 * properly.
593 */
594 i = -1;
595 }
596 }
597
598 free_xenballooned_pages(MAX_PENDING_REQS, vif->mmap_pages);
599
600 netif_napi_del(&vif->napi);
601
602 unregister_netdev(vif->dev);
603
604 vfree(vif->grant_copy_op);
605 free_netdev(vif->dev);
606
607 module_put(THIS_MODULE);
608}
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/ethtool.h>
34#include <linux/rtnetlink.h>
35#include <linux/if_vlan.h>
36
37#include <xen/events.h>
38#include <asm/xen/hypercall.h>
39
40#define XENVIF_QUEUE_LENGTH 32
41
42void xenvif_get(struct xenvif *vif)
43{
44 atomic_inc(&vif->refcnt);
45}
46
47void xenvif_put(struct xenvif *vif)
48{
49 if (atomic_dec_and_test(&vif->refcnt))
50 wake_up(&vif->waiting_to_free);
51}
52
53int xenvif_schedulable(struct xenvif *vif)
54{
55 return netif_running(vif->dev) && netif_carrier_ok(vif->dev);
56}
57
58static int xenvif_rx_schedulable(struct xenvif *vif)
59{
60 return xenvif_schedulable(vif) && !xen_netbk_rx_ring_full(vif);
61}
62
63static irqreturn_t xenvif_interrupt(int irq, void *dev_id)
64{
65 struct xenvif *vif = dev_id;
66
67 if (vif->netbk == NULL)
68 return IRQ_NONE;
69
70 xen_netbk_schedule_xenvif(vif);
71
72 if (xenvif_rx_schedulable(vif))
73 netif_wake_queue(vif->dev);
74
75 return IRQ_HANDLED;
76}
77
78static int xenvif_start_xmit(struct sk_buff *skb, struct net_device *dev)
79{
80 struct xenvif *vif = netdev_priv(dev);
81
82 BUG_ON(skb->dev != dev);
83
84 if (vif->netbk == NULL)
85 goto drop;
86
87 /* Drop the packet if the target domain has no receive buffers. */
88 if (!xenvif_rx_schedulable(vif))
89 goto drop;
90
91 /* Reserve ring slots for the worst-case number of fragments. */
92 vif->rx_req_cons_peek += xen_netbk_count_skb_slots(vif, skb);
93 xenvif_get(vif);
94
95 if (vif->can_queue && xen_netbk_must_stop_queue(vif))
96 netif_stop_queue(dev);
97
98 xen_netbk_queue_tx_skb(vif, skb);
99
100 return NETDEV_TX_OK;
101
102 drop:
103 vif->dev->stats.tx_dropped++;
104 dev_kfree_skb(skb);
105 return NETDEV_TX_OK;
106}
107
108void xenvif_receive_skb(struct xenvif *vif, struct sk_buff *skb)
109{
110 netif_rx_ni(skb);
111}
112
113void xenvif_notify_tx_completion(struct xenvif *vif)
114{
115 if (netif_queue_stopped(vif->dev) && xenvif_rx_schedulable(vif))
116 netif_wake_queue(vif->dev);
117}
118
119static struct net_device_stats *xenvif_get_stats(struct net_device *dev)
120{
121 struct xenvif *vif = netdev_priv(dev);
122 return &vif->dev->stats;
123}
124
125static void xenvif_up(struct xenvif *vif)
126{
127 xen_netbk_add_xenvif(vif);
128 enable_irq(vif->irq);
129 xen_netbk_check_rx_xenvif(vif);
130}
131
132static void xenvif_down(struct xenvif *vif)
133{
134 disable_irq(vif->irq);
135 xen_netbk_deschedule_xenvif(vif);
136 xen_netbk_remove_xenvif(vif);
137}
138
139static int xenvif_open(struct net_device *dev)
140{
141 struct xenvif *vif = netdev_priv(dev);
142 if (netif_carrier_ok(dev))
143 xenvif_up(vif);
144 netif_start_queue(dev);
145 return 0;
146}
147
148static int xenvif_close(struct net_device *dev)
149{
150 struct xenvif *vif = netdev_priv(dev);
151 if (netif_carrier_ok(dev))
152 xenvif_down(vif);
153 netif_stop_queue(dev);
154 return 0;
155}
156
157static int xenvif_change_mtu(struct net_device *dev, int mtu)
158{
159 struct xenvif *vif = netdev_priv(dev);
160 int max = vif->can_sg ? 65535 - VLAN_ETH_HLEN : ETH_DATA_LEN;
161
162 if (mtu > max)
163 return -EINVAL;
164 dev->mtu = mtu;
165 return 0;
166}
167
168static netdev_features_t xenvif_fix_features(struct net_device *dev,
169 netdev_features_t features)
170{
171 struct xenvif *vif = netdev_priv(dev);
172
173 if (!vif->can_sg)
174 features &= ~NETIF_F_SG;
175 if (!vif->gso && !vif->gso_prefix)
176 features &= ~NETIF_F_TSO;
177 if (!vif->csum)
178 features &= ~NETIF_F_IP_CSUM;
179
180 return features;
181}
182
183static const struct xenvif_stat {
184 char name[ETH_GSTRING_LEN];
185 u16 offset;
186} xenvif_stats[] = {
187 {
188 "rx_gso_checksum_fixup",
189 offsetof(struct xenvif, rx_gso_checksum_fixup)
190 },
191};
192
193static int xenvif_get_sset_count(struct net_device *dev, int string_set)
194{
195 switch (string_set) {
196 case ETH_SS_STATS:
197 return ARRAY_SIZE(xenvif_stats);
198 default:
199 return -EINVAL;
200 }
201}
202
203static void xenvif_get_ethtool_stats(struct net_device *dev,
204 struct ethtool_stats *stats, u64 * data)
205{
206 void *vif = netdev_priv(dev);
207 int i;
208
209 for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++)
210 data[i] = *(unsigned long *)(vif + xenvif_stats[i].offset);
211}
212
213static void xenvif_get_strings(struct net_device *dev, u32 stringset, u8 * data)
214{
215 int i;
216
217 switch (stringset) {
218 case ETH_SS_STATS:
219 for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++)
220 memcpy(data + i * ETH_GSTRING_LEN,
221 xenvif_stats[i].name, ETH_GSTRING_LEN);
222 break;
223 }
224}
225
226static const struct ethtool_ops xenvif_ethtool_ops = {
227 .get_link = ethtool_op_get_link,
228
229 .get_sset_count = xenvif_get_sset_count,
230 .get_ethtool_stats = xenvif_get_ethtool_stats,
231 .get_strings = xenvif_get_strings,
232};
233
234static const struct net_device_ops xenvif_netdev_ops = {
235 .ndo_start_xmit = xenvif_start_xmit,
236 .ndo_get_stats = xenvif_get_stats,
237 .ndo_open = xenvif_open,
238 .ndo_stop = xenvif_close,
239 .ndo_change_mtu = xenvif_change_mtu,
240 .ndo_fix_features = xenvif_fix_features,
241};
242
243struct xenvif *xenvif_alloc(struct device *parent, domid_t domid,
244 unsigned int handle)
245{
246 int err;
247 struct net_device *dev;
248 struct xenvif *vif;
249 char name[IFNAMSIZ] = {};
250
251 snprintf(name, IFNAMSIZ - 1, "vif%u.%u", domid, handle);
252 dev = alloc_netdev(sizeof(struct xenvif), name, ether_setup);
253 if (dev == NULL) {
254 pr_warn("Could not allocate netdev\n");
255 return ERR_PTR(-ENOMEM);
256 }
257
258 SET_NETDEV_DEV(dev, parent);
259
260 vif = netdev_priv(dev);
261 vif->domid = domid;
262 vif->handle = handle;
263 vif->netbk = NULL;
264 vif->can_sg = 1;
265 vif->csum = 1;
266 atomic_set(&vif->refcnt, 1);
267 init_waitqueue_head(&vif->waiting_to_free);
268 vif->dev = dev;
269 INIT_LIST_HEAD(&vif->schedule_list);
270 INIT_LIST_HEAD(&vif->notify_list);
271
272 vif->credit_bytes = vif->remaining_credit = ~0UL;
273 vif->credit_usec = 0UL;
274 init_timer(&vif->credit_timeout);
275 /* Initialize 'expires' now: it's used to track the credit window. */
276 vif->credit_timeout.expires = jiffies;
277
278 dev->netdev_ops = &xenvif_netdev_ops;
279 dev->hw_features = NETIF_F_SG | NETIF_F_IP_CSUM | NETIF_F_TSO;
280 dev->features = dev->hw_features;
281 SET_ETHTOOL_OPS(dev, &xenvif_ethtool_ops);
282
283 dev->tx_queue_len = XENVIF_QUEUE_LENGTH;
284
285 /*
286 * Initialise a dummy MAC address. We choose the numerically
287 * largest non-broadcast address to prevent the address getting
288 * stolen by an Ethernet bridge for STP purposes.
289 * (FE:FF:FF:FF:FF:FF)
290 */
291 memset(dev->dev_addr, 0xFF, ETH_ALEN);
292 dev->dev_addr[0] &= ~0x01;
293
294 netif_carrier_off(dev);
295
296 err = register_netdev(dev);
297 if (err) {
298 netdev_warn(dev, "Could not register device: err=%d\n", err);
299 free_netdev(dev);
300 return ERR_PTR(err);
301 }
302
303 netdev_dbg(dev, "Successfully created xenvif\n");
304 return vif;
305}
306
307int xenvif_connect(struct xenvif *vif, unsigned long tx_ring_ref,
308 unsigned long rx_ring_ref, unsigned int evtchn)
309{
310 int err = -ENOMEM;
311
312 /* Already connected through? */
313 if (vif->irq)
314 return 0;
315
316 err = xen_netbk_map_frontend_rings(vif, tx_ring_ref, rx_ring_ref);
317 if (err < 0)
318 goto err;
319
320 err = bind_interdomain_evtchn_to_irqhandler(
321 vif->domid, evtchn, xenvif_interrupt, 0,
322 vif->dev->name, vif);
323 if (err < 0)
324 goto err_unmap;
325 vif->irq = err;
326 disable_irq(vif->irq);
327
328 xenvif_get(vif);
329
330 rtnl_lock();
331 if (!vif->can_sg && vif->dev->mtu > ETH_DATA_LEN)
332 dev_set_mtu(vif->dev, ETH_DATA_LEN);
333 netdev_update_features(vif->dev);
334 netif_carrier_on(vif->dev);
335 if (netif_running(vif->dev))
336 xenvif_up(vif);
337 rtnl_unlock();
338
339 return 0;
340err_unmap:
341 xen_netbk_unmap_frontend_rings(vif);
342err:
343 return err;
344}
345
346void xenvif_disconnect(struct xenvif *vif)
347{
348 struct net_device *dev = vif->dev;
349 if (netif_carrier_ok(dev)) {
350 rtnl_lock();
351 netif_carrier_off(dev); /* discard queued packets */
352 if (netif_running(dev))
353 xenvif_down(vif);
354 rtnl_unlock();
355 xenvif_put(vif);
356 }
357
358 atomic_dec(&vif->refcnt);
359 wait_event(vif->waiting_to_free, atomic_read(&vif->refcnt) == 0);
360
361 del_timer_sync(&vif->credit_timeout);
362
363 if (vif->irq)
364 unbind_from_irqhandler(vif->irq, vif);
365
366 unregister_netdev(vif->dev);
367
368 xen_netbk_unmap_frontend_rings(vif);
369
370 free_netdev(vif->dev);
371}