]>
Commit | Line | Data |
---|---|---|
f942dc25 IC |
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 | ||
b3f980bd | 33 | #include <linux/kthread.h> |
f942dc25 IC |
34 | #include <linux/ethtool.h> |
35 | #include <linux/rtnetlink.h> | |
36 | #include <linux/if_vlan.h> | |
e7b599d7 | 37 | #include <linux/vmalloc.h> |
f942dc25 IC |
38 | |
39 | #include <xen/events.h> | |
40 | #include <asm/xen/hypercall.h> | |
f53c3fe8 | 41 | #include <xen/balloon.h> |
f942dc25 IC |
42 | |
43 | #define XENVIF_QUEUE_LENGTH 32 | |
b3f980bd | 44 | #define XENVIF_NAPI_WEIGHT 64 |
f942dc25 | 45 | |
a64bd934 WL |
46 | /* This function is used to set SKBTX_DEV_ZEROCOPY as well as |
47 | * increasing the inflight counter. We need to increase the inflight | |
48 | * counter because core driver calls into xenvif_zerocopy_callback | |
49 | * which calls xenvif_skb_zerocopy_complete. | |
50 | */ | |
51 | void xenvif_skb_zerocopy_prepare(struct xenvif_queue *queue, | |
52 | struct sk_buff *skb) | |
53 | { | |
54 | skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY; | |
55 | atomic_inc(&queue->inflight_packets); | |
56 | } | |
57 | ||
58 | void xenvif_skb_zerocopy_complete(struct xenvif_queue *queue) | |
59 | { | |
60 | atomic_dec(&queue->inflight_packets); | |
61 | } | |
62 | ||
e9ce7cb6 WL |
63 | static inline void xenvif_stop_queue(struct xenvif_queue *queue) |
64 | { | |
65 | struct net_device *dev = queue->vif->dev; | |
66 | ||
67 | if (!queue->vif->can_queue) | |
68 | return; | |
69 | ||
70 | netif_tx_stop_queue(netdev_get_tx_queue(dev, queue->id)); | |
71 | } | |
72 | ||
f942dc25 IC |
73 | int xenvif_schedulable(struct xenvif *vif) |
74 | { | |
3d1af1df ZK |
75 | return netif_running(vif->dev) && |
76 | test_bit(VIF_STATUS_CONNECTED, &vif->status); | |
f942dc25 IC |
77 | } |
78 | ||
e1f00a69 | 79 | static irqreturn_t xenvif_tx_interrupt(int irq, void *dev_id) |
f942dc25 | 80 | { |
e9ce7cb6 | 81 | struct xenvif_queue *queue = dev_id; |
f942dc25 | 82 | |
e9ce7cb6 WL |
83 | if (RING_HAS_UNCONSUMED_REQUESTS(&queue->tx)) |
84 | napi_schedule(&queue->napi); | |
f942dc25 | 85 | |
e1f00a69 WL |
86 | return IRQ_HANDLED; |
87 | } | |
88 | ||
e9ce7cb6 | 89 | int xenvif_poll(struct napi_struct *napi, int budget) |
b3f980bd | 90 | { |
e9ce7cb6 WL |
91 | struct xenvif_queue *queue = |
92 | container_of(napi, struct xenvif_queue, napi); | |
b3f980bd WL |
93 | int work_done; |
94 | ||
e9d8b2c2 WL |
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. | |
98 | */ | |
2561cc15 | 99 | if (unlikely(queue->vif->disabled)) { |
e9d8b2c2 WL |
100 | napi_complete(napi); |
101 | return 0; | |
102 | } | |
103 | ||
e9ce7cb6 | 104 | work_done = xenvif_tx_action(queue, budget); |
b3f980bd WL |
105 | |
106 | if (work_done < budget) { | |
0d08fceb | 107 | napi_complete(napi); |
e9ce7cb6 | 108 | xenvif_napi_schedule_or_enable_events(queue); |
b3f980bd WL |
109 | } |
110 | ||
111 | return work_done; | |
112 | } | |
113 | ||
e1f00a69 WL |
114 | static irqreturn_t xenvif_rx_interrupt(int irq, void *dev_id) |
115 | { | |
e9ce7cb6 | 116 | struct xenvif_queue *queue = dev_id; |
f34a4cf9 ZK |
117 | struct netdev_queue *net_queue = |
118 | netdev_get_tx_queue(queue->vif->dev, queue->id); | |
e1f00a69 | 119 | |
f34a4cf9 ZK |
120 | /* QUEUE_STATUS_RX_PURGE_EVENT is only set if either QDisc was off OR |
121 | * the carrier went down and this queue was previously blocked | |
122 | */ | |
123 | if (unlikely(netif_tx_queue_stopped(net_queue) || | |
124 | (!netif_carrier_ok(queue->vif->dev) && | |
125 | test_bit(QUEUE_STATUS_RX_STALLED, &queue->status)))) | |
126 | set_bit(QUEUE_STATUS_RX_PURGE_EVENT, &queue->status); | |
e9ce7cb6 | 127 | xenvif_kick_thread(queue); |
f942dc25 IC |
128 | |
129 | return IRQ_HANDLED; | |
130 | } | |
131 | ||
f51de243 | 132 | irqreturn_t xenvif_interrupt(int irq, void *dev_id) |
e1f00a69 WL |
133 | { |
134 | xenvif_tx_interrupt(irq, dev_id); | |
135 | xenvif_rx_interrupt(irq, dev_id); | |
136 | ||
137 | return IRQ_HANDLED; | |
138 | } | |
139 | ||
e9ce7cb6 WL |
140 | int xenvif_queue_stopped(struct xenvif_queue *queue) |
141 | { | |
142 | struct net_device *dev = queue->vif->dev; | |
143 | unsigned int id = queue->id; | |
144 | return netif_tx_queue_stopped(netdev_get_tx_queue(dev, id)); | |
145 | } | |
146 | ||
147 | void xenvif_wake_queue(struct xenvif_queue *queue) | |
148 | { | |
149 | struct net_device *dev = queue->vif->dev; | |
150 | unsigned int id = queue->id; | |
151 | netif_tx_wake_queue(netdev_get_tx_queue(dev, id)); | |
152 | } | |
153 | ||
f34a4cf9 ZK |
154 | /* Callback to wake the queue's thread and turn the carrier off on timeout */ |
155 | static void xenvif_rx_stalled(unsigned long data) | |
09350788 | 156 | { |
e9ce7cb6 WL |
157 | struct xenvif_queue *queue = (struct xenvif_queue *)data; |
158 | ||
159 | if (xenvif_queue_stopped(queue)) { | |
f34a4cf9 | 160 | set_bit(QUEUE_STATUS_RX_PURGE_EVENT, &queue->status); |
e9ce7cb6 | 161 | xenvif_kick_thread(queue); |
e9ce7cb6 WL |
162 | } |
163 | } | |
09350788 | 164 | |
f942dc25 IC |
165 | static int xenvif_start_xmit(struct sk_buff *skb, struct net_device *dev) |
166 | { | |
167 | struct xenvif *vif = netdev_priv(dev); | |
e9ce7cb6 | 168 | struct xenvif_queue *queue = NULL; |
f7b50c4e | 169 | unsigned int num_queues = vif->num_queues; |
e9ce7cb6 | 170 | u16 index; |
ca2f09f2 | 171 | int min_slots_needed; |
f942dc25 IC |
172 | |
173 | BUG_ON(skb->dev != dev); | |
174 | ||
e9ce7cb6 WL |
175 | /* Drop the packet if queues are not set up */ |
176 | if (num_queues < 1) | |
177 | goto drop; | |
178 | ||
179 | /* Obtain the queue to be used to transmit this packet */ | |
180 | index = skb_get_queue_mapping(skb); | |
181 | if (index >= num_queues) { | |
182 | pr_warn_ratelimited("Invalid queue %hu for packet on interface %s\n.", | |
183 | index, vif->dev->name); | |
184 | index %= num_queues; | |
185 | } | |
186 | queue = &vif->queues[index]; | |
187 | ||
188 | /* Drop the packet if queue is not ready */ | |
189 | if (queue->task == NULL || | |
190 | queue->dealloc_task == NULL || | |
f53c3fe8 | 191 | !xenvif_schedulable(vif)) |
f942dc25 IC |
192 | goto drop; |
193 | ||
ca2f09f2 PD |
194 | /* At best we'll need one slot for the header and one for each |
195 | * frag. | |
196 | */ | |
197 | min_slots_needed = 1 + skb_shinfo(skb)->nr_frags; | |
f942dc25 | 198 | |
ca2f09f2 PD |
199 | /* If the skb is GSO then we'll also need an extra slot for the |
200 | * metadata. | |
201 | */ | |
836fbaf4 | 202 | if (skb_is_gso(skb)) |
ca2f09f2 | 203 | min_slots_needed++; |
f942dc25 | 204 | |
ca2f09f2 PD |
205 | /* If the skb can't possibly fit in the remaining slots |
206 | * then turn off the queue to give the ring a chance to | |
207 | * drain. | |
208 | */ | |
e9ce7cb6 | 209 | if (!xenvif_rx_ring_slots_available(queue, min_slots_needed)) { |
f34a4cf9 ZK |
210 | queue->rx_stalled.function = xenvif_rx_stalled; |
211 | queue->rx_stalled.data = (unsigned long)queue; | |
e9ce7cb6 | 212 | xenvif_stop_queue(queue); |
f34a4cf9 ZK |
213 | mod_timer(&queue->rx_stalled, |
214 | jiffies + rx_drain_timeout_jiffies); | |
09350788 | 215 | } |
f942dc25 | 216 | |
e9ce7cb6 WL |
217 | skb_queue_tail(&queue->rx_queue, skb); |
218 | xenvif_kick_thread(queue); | |
f942dc25 IC |
219 | |
220 | return NETDEV_TX_OK; | |
221 | ||
222 | drop: | |
223 | vif->dev->stats.tx_dropped++; | |
224 | dev_kfree_skb(skb); | |
225 | return NETDEV_TX_OK; | |
226 | } | |
227 | ||
f942dc25 IC |
228 | static struct net_device_stats *xenvif_get_stats(struct net_device *dev) |
229 | { | |
230 | struct xenvif *vif = netdev_priv(dev); | |
e9ce7cb6 | 231 | struct xenvif_queue *queue = NULL; |
f7b50c4e | 232 | unsigned int num_queues = vif->num_queues; |
e9ce7cb6 WL |
233 | unsigned long rx_bytes = 0; |
234 | unsigned long rx_packets = 0; | |
235 | unsigned long tx_bytes = 0; | |
236 | unsigned long tx_packets = 0; | |
237 | unsigned int index; | |
238 | ||
239 | if (vif->queues == NULL) | |
240 | goto out; | |
241 | ||
242 | /* Aggregate tx and rx stats from each queue */ | |
243 | for (index = 0; index < num_queues; ++index) { | |
244 | queue = &vif->queues[index]; | |
245 | rx_bytes += queue->stats.rx_bytes; | |
246 | rx_packets += queue->stats.rx_packets; | |
247 | tx_bytes += queue->stats.tx_bytes; | |
248 | tx_packets += queue->stats.tx_packets; | |
249 | } | |
250 | ||
251 | out: | |
252 | vif->dev->stats.rx_bytes = rx_bytes; | |
253 | vif->dev->stats.rx_packets = rx_packets; | |
254 | vif->dev->stats.tx_bytes = tx_bytes; | |
255 | vif->dev->stats.tx_packets = tx_packets; | |
256 | ||
f942dc25 IC |
257 | return &vif->dev->stats; |
258 | } | |
259 | ||
260 | static void xenvif_up(struct xenvif *vif) | |
261 | { | |
e9ce7cb6 | 262 | struct xenvif_queue *queue = NULL; |
f7b50c4e | 263 | unsigned int num_queues = vif->num_queues; |
e9ce7cb6 WL |
264 | unsigned int queue_index; |
265 | ||
266 | for (queue_index = 0; queue_index < num_queues; ++queue_index) { | |
267 | queue = &vif->queues[queue_index]; | |
268 | napi_enable(&queue->napi); | |
269 | enable_irq(queue->tx_irq); | |
270 | if (queue->tx_irq != queue->rx_irq) | |
271 | enable_irq(queue->rx_irq); | |
272 | xenvif_napi_schedule_or_enable_events(queue); | |
273 | } | |
f942dc25 IC |
274 | } |
275 | ||
276 | static void xenvif_down(struct xenvif *vif) | |
277 | { | |
e9ce7cb6 | 278 | struct xenvif_queue *queue = NULL; |
f7b50c4e | 279 | unsigned int num_queues = vif->num_queues; |
e9ce7cb6 WL |
280 | unsigned int queue_index; |
281 | ||
282 | for (queue_index = 0; queue_index < num_queues; ++queue_index) { | |
283 | queue = &vif->queues[queue_index]; | |
284 | napi_disable(&queue->napi); | |
285 | disable_irq(queue->tx_irq); | |
286 | if (queue->tx_irq != queue->rx_irq) | |
287 | disable_irq(queue->rx_irq); | |
288 | del_timer_sync(&queue->credit_timeout); | |
289 | } | |
f942dc25 IC |
290 | } |
291 | ||
292 | static int xenvif_open(struct net_device *dev) | |
293 | { | |
294 | struct xenvif *vif = netdev_priv(dev); | |
3d1af1df | 295 | if (test_bit(VIF_STATUS_CONNECTED, &vif->status)) |
f942dc25 | 296 | xenvif_up(vif); |
e9ce7cb6 | 297 | netif_tx_start_all_queues(dev); |
f942dc25 IC |
298 | return 0; |
299 | } | |
300 | ||
301 | static int xenvif_close(struct net_device *dev) | |
302 | { | |
303 | struct xenvif *vif = netdev_priv(dev); | |
3d1af1df | 304 | if (test_bit(VIF_STATUS_CONNECTED, &vif->status)) |
f942dc25 | 305 | xenvif_down(vif); |
e9ce7cb6 | 306 | netif_tx_stop_all_queues(dev); |
f942dc25 IC |
307 | return 0; |
308 | } | |
309 | ||
310 | static int xenvif_change_mtu(struct net_device *dev, int mtu) | |
311 | { | |
312 | struct xenvif *vif = netdev_priv(dev); | |
313 | int max = vif->can_sg ? 65535 - VLAN_ETH_HLEN : ETH_DATA_LEN; | |
314 | ||
315 | if (mtu > max) | |
316 | return -EINVAL; | |
317 | dev->mtu = mtu; | |
318 | return 0; | |
319 | } | |
320 | ||
c8f44aff MM |
321 | static netdev_features_t xenvif_fix_features(struct net_device *dev, |
322 | netdev_features_t features) | |
f942dc25 IC |
323 | { |
324 | struct xenvif *vif = netdev_priv(dev); | |
f942dc25 | 325 | |
47103041 MM |
326 | if (!vif->can_sg) |
327 | features &= ~NETIF_F_SG; | |
82cada22 | 328 | if (~(vif->gso_mask | vif->gso_prefix_mask) & GSO_BIT(TCPV4)) |
47103041 | 329 | features &= ~NETIF_F_TSO; |
82cada22 PD |
330 | if (~(vif->gso_mask | vif->gso_prefix_mask) & GSO_BIT(TCPV6)) |
331 | features &= ~NETIF_F_TSO6; | |
146c8a77 | 332 | if (!vif->ip_csum) |
47103041 | 333 | features &= ~NETIF_F_IP_CSUM; |
146c8a77 PD |
334 | if (!vif->ipv6_csum) |
335 | features &= ~NETIF_F_IPV6_CSUM; | |
f942dc25 | 336 | |
47103041 | 337 | return features; |
f942dc25 IC |
338 | } |
339 | ||
340 | static const struct xenvif_stat { | |
341 | char name[ETH_GSTRING_LEN]; | |
342 | u16 offset; | |
343 | } xenvif_stats[] = { | |
344 | { | |
345 | "rx_gso_checksum_fixup", | |
e9ce7cb6 | 346 | offsetof(struct xenvif_stats, rx_gso_checksum_fixup) |
f942dc25 | 347 | }, |
1bb332af ZK |
348 | /* If (sent != success + fail), there are probably packets never |
349 | * freed up properly! | |
350 | */ | |
351 | { | |
352 | "tx_zerocopy_sent", | |
e9ce7cb6 | 353 | offsetof(struct xenvif_stats, tx_zerocopy_sent), |
1bb332af ZK |
354 | }, |
355 | { | |
356 | "tx_zerocopy_success", | |
e9ce7cb6 | 357 | offsetof(struct xenvif_stats, tx_zerocopy_success), |
1bb332af ZK |
358 | }, |
359 | { | |
360 | "tx_zerocopy_fail", | |
e9ce7cb6 | 361 | offsetof(struct xenvif_stats, tx_zerocopy_fail) |
1bb332af | 362 | }, |
e3377f36 ZK |
363 | /* Number of packets exceeding MAX_SKB_FRAG slots. You should use |
364 | * a guest with the same MAX_SKB_FRAG | |
365 | */ | |
366 | { | |
367 | "tx_frag_overflow", | |
e9ce7cb6 | 368 | offsetof(struct xenvif_stats, tx_frag_overflow) |
e3377f36 | 369 | }, |
f942dc25 IC |
370 | }; |
371 | ||
372 | static int xenvif_get_sset_count(struct net_device *dev, int string_set) | |
373 | { | |
374 | switch (string_set) { | |
375 | case ETH_SS_STATS: | |
376 | return ARRAY_SIZE(xenvif_stats); | |
377 | default: | |
378 | return -EINVAL; | |
379 | } | |
380 | } | |
381 | ||
382 | static void xenvif_get_ethtool_stats(struct net_device *dev, | |
383 | struct ethtool_stats *stats, u64 * data) | |
384 | { | |
e9ce7cb6 | 385 | struct xenvif *vif = netdev_priv(dev); |
f7b50c4e | 386 | unsigned int num_queues = vif->num_queues; |
f942dc25 | 387 | int i; |
e9ce7cb6 WL |
388 | unsigned int queue_index; |
389 | struct xenvif_stats *vif_stats; | |
390 | ||
391 | for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++) { | |
392 | unsigned long accum = 0; | |
393 | for (queue_index = 0; queue_index < num_queues; ++queue_index) { | |
394 | vif_stats = &vif->queues[queue_index].stats; | |
395 | accum += *(unsigned long *)(vif_stats + xenvif_stats[i].offset); | |
396 | } | |
397 | data[i] = accum; | |
398 | } | |
f942dc25 IC |
399 | } |
400 | ||
401 | static void xenvif_get_strings(struct net_device *dev, u32 stringset, u8 * data) | |
402 | { | |
403 | int i; | |
404 | ||
405 | switch (stringset) { | |
406 | case ETH_SS_STATS: | |
407 | for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++) | |
408 | memcpy(data + i * ETH_GSTRING_LEN, | |
409 | xenvif_stats[i].name, ETH_GSTRING_LEN); | |
410 | break; | |
411 | } | |
412 | } | |
413 | ||
813abbba | 414 | static const struct ethtool_ops xenvif_ethtool_ops = { |
f942dc25 IC |
415 | .get_link = ethtool_op_get_link, |
416 | ||
417 | .get_sset_count = xenvif_get_sset_count, | |
418 | .get_ethtool_stats = xenvif_get_ethtool_stats, | |
419 | .get_strings = xenvif_get_strings, | |
420 | }; | |
421 | ||
813abbba | 422 | static const struct net_device_ops xenvif_netdev_ops = { |
f942dc25 IC |
423 | .ndo_start_xmit = xenvif_start_xmit, |
424 | .ndo_get_stats = xenvif_get_stats, | |
425 | .ndo_open = xenvif_open, | |
426 | .ndo_stop = xenvif_close, | |
427 | .ndo_change_mtu = xenvif_change_mtu, | |
47103041 | 428 | .ndo_fix_features = xenvif_fix_features, |
4a633a60 MW |
429 | .ndo_set_mac_address = eth_mac_addr, |
430 | .ndo_validate_addr = eth_validate_addr, | |
f942dc25 IC |
431 | }; |
432 | ||
433 | struct xenvif *xenvif_alloc(struct device *parent, domid_t domid, | |
434 | unsigned int handle) | |
435 | { | |
436 | int err; | |
437 | struct net_device *dev; | |
438 | struct xenvif *vif; | |
439 | char name[IFNAMSIZ] = {}; | |
440 | ||
441 | snprintf(name, IFNAMSIZ - 1, "vif%u.%u", domid, handle); | |
8d3d53b3 AB |
442 | /* Allocate a netdev with the max. supported number of queues. |
443 | * When the guest selects the desired number, it will be updated | |
f7b50c4e | 444 | * via netif_set_real_num_*_queues(). |
8d3d53b3 | 445 | */ |
c835a677 TG |
446 | dev = alloc_netdev_mq(sizeof(struct xenvif), name, NET_NAME_UNKNOWN, |
447 | ether_setup, xenvif_max_queues); | |
f942dc25 | 448 | if (dev == NULL) { |
b3f980bd | 449 | pr_warn("Could not allocate netdev for %s\n", name); |
f942dc25 IC |
450 | return ERR_PTR(-ENOMEM); |
451 | } | |
452 | ||
453 | SET_NETDEV_DEV(dev, parent); | |
454 | ||
455 | vif = netdev_priv(dev); | |
ac3d5ac2 | 456 | |
f942dc25 IC |
457 | vif->domid = domid; |
458 | vif->handle = handle; | |
f942dc25 | 459 | vif->can_sg = 1; |
146c8a77 | 460 | vif->ip_csum = 1; |
f942dc25 | 461 | vif->dev = dev; |
e9d8b2c2 WL |
462 | vif->disabled = false; |
463 | ||
f7b50c4e | 464 | /* Start out with no queues. */ |
e9ce7cb6 | 465 | vif->queues = NULL; |
f7b50c4e | 466 | vif->num_queues = 0; |
09350788 | 467 | |
f942dc25 | 468 | dev->netdev_ops = &xenvif_netdev_ops; |
146c8a77 PD |
469 | dev->hw_features = NETIF_F_SG | |
470 | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | | |
82cada22 | 471 | NETIF_F_TSO | NETIF_F_TSO6; |
7365bcfa | 472 | dev->features = dev->hw_features | NETIF_F_RXCSUM; |
7ad24ea4 | 473 | dev->ethtool_ops = &xenvif_ethtool_ops; |
f942dc25 IC |
474 | |
475 | dev->tx_queue_len = XENVIF_QUEUE_LENGTH; | |
476 | ||
477 | /* | |
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) | |
482 | */ | |
483 | memset(dev->dev_addr, 0xFF, ETH_ALEN); | |
484 | dev->dev_addr[0] &= ~0x01; | |
485 | ||
486 | netif_carrier_off(dev); | |
487 | ||
488 | err = register_netdev(dev); | |
489 | if (err) { | |
490 | netdev_warn(dev, "Could not register device: err=%d\n", err); | |
491 | free_netdev(dev); | |
492 | return ERR_PTR(err); | |
493 | } | |
494 | ||
495 | netdev_dbg(dev, "Successfully created xenvif\n"); | |
279f438e PD |
496 | |
497 | __module_get(THIS_MODULE); | |
498 | ||
f942dc25 IC |
499 | return vif; |
500 | } | |
501 | ||
e9ce7cb6 WL |
502 | int xenvif_init_queue(struct xenvif_queue *queue) |
503 | { | |
504 | int err, i; | |
505 | ||
506 | queue->credit_bytes = queue->remaining_credit = ~0UL; | |
507 | queue->credit_usec = 0UL; | |
508 | init_timer(&queue->credit_timeout); | |
509 | queue->credit_window_start = get_jiffies_64(); | |
510 | ||
511 | skb_queue_head_init(&queue->rx_queue); | |
512 | skb_queue_head_init(&queue->tx_queue); | |
513 | ||
514 | queue->pending_cons = 0; | |
515 | queue->pending_prod = MAX_PENDING_REQS; | |
516 | for (i = 0; i < MAX_PENDING_REQS; ++i) | |
517 | queue->pending_ring[i] = i; | |
518 | ||
519 | spin_lock_init(&queue->callback_lock); | |
520 | spin_lock_init(&queue->response_lock); | |
521 | ||
522 | /* If ballooning is disabled, this will consume real memory, so you | |
523 | * better enable it. The long term solution would be to use just a | |
524 | * bunch of valid page descriptors, without dependency on ballooning | |
525 | */ | |
526 | err = alloc_xenballooned_pages(MAX_PENDING_REQS, | |
527 | queue->mmap_pages, | |
528 | false); | |
529 | if (err) { | |
530 | netdev_err(queue->vif->dev, "Could not reserve mmap_pages\n"); | |
531 | return -ENOMEM; | |
532 | } | |
533 | ||
534 | for (i = 0; i < MAX_PENDING_REQS; i++) { | |
535 | queue->pending_tx_info[i].callback_struct = (struct ubuf_info) | |
536 | { .callback = xenvif_zerocopy_callback, | |
537 | .ctx = NULL, | |
538 | .desc = i }; | |
539 | queue->grant_tx_handle[i] = NETBACK_INVALID_HANDLE; | |
540 | } | |
541 | ||
f34a4cf9 | 542 | init_timer(&queue->rx_stalled); |
e9ce7cb6 | 543 | |
e9ce7cb6 WL |
544 | return 0; |
545 | } | |
546 | ||
547 | void xenvif_carrier_on(struct xenvif *vif) | |
548 | { | |
549 | rtnl_lock(); | |
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); | |
3d1af1df | 553 | set_bit(VIF_STATUS_CONNECTED, &vif->status); |
e9ce7cb6 WL |
554 | netif_carrier_on(vif->dev); |
555 | if (netif_running(vif->dev)) | |
556 | xenvif_up(vif); | |
557 | rtnl_unlock(); | |
558 | } | |
559 | ||
560 | int xenvif_connect(struct xenvif_queue *queue, unsigned long tx_ring_ref, | |
e1f00a69 WL |
561 | unsigned long rx_ring_ref, unsigned int tx_evtchn, |
562 | unsigned int rx_evtchn) | |
f942dc25 | 563 | { |
67fa3660 | 564 | struct task_struct *task; |
f942dc25 IC |
565 | int err = -ENOMEM; |
566 | ||
e9ce7cb6 WL |
567 | BUG_ON(queue->tx_irq); |
568 | BUG_ON(queue->task); | |
569 | BUG_ON(queue->dealloc_task); | |
f942dc25 | 570 | |
e9ce7cb6 | 571 | err = xenvif_map_frontend_rings(queue, tx_ring_ref, rx_ring_ref); |
f942dc25 IC |
572 | if (err < 0) |
573 | goto err; | |
574 | ||
e9ce7cb6 WL |
575 | init_waitqueue_head(&queue->wq); |
576 | init_waitqueue_head(&queue->dealloc_wq); | |
a64bd934 | 577 | atomic_set(&queue->inflight_packets, 0); |
ca2f09f2 | 578 | |
e1f00a69 WL |
579 | if (tx_evtchn == rx_evtchn) { |
580 | /* feature-split-event-channels == 0 */ | |
581 | err = bind_interdomain_evtchn_to_irqhandler( | |
e9ce7cb6 WL |
582 | queue->vif->domid, tx_evtchn, xenvif_interrupt, 0, |
583 | queue->name, queue); | |
e1f00a69 WL |
584 | if (err < 0) |
585 | goto err_unmap; | |
e9ce7cb6 WL |
586 | queue->tx_irq = queue->rx_irq = err; |
587 | disable_irq(queue->tx_irq); | |
e1f00a69 WL |
588 | } else { |
589 | /* feature-split-event-channels == 1 */ | |
e9ce7cb6 WL |
590 | snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name), |
591 | "%s-tx", queue->name); | |
e1f00a69 | 592 | err = bind_interdomain_evtchn_to_irqhandler( |
e9ce7cb6 WL |
593 | queue->vif->domid, tx_evtchn, xenvif_tx_interrupt, 0, |
594 | queue->tx_irq_name, queue); | |
e1f00a69 WL |
595 | if (err < 0) |
596 | goto err_unmap; | |
e9ce7cb6 WL |
597 | queue->tx_irq = err; |
598 | disable_irq(queue->tx_irq); | |
e1f00a69 | 599 | |
e9ce7cb6 WL |
600 | snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name), |
601 | "%s-rx", queue->name); | |
e1f00a69 | 602 | err = bind_interdomain_evtchn_to_irqhandler( |
e9ce7cb6 WL |
603 | queue->vif->domid, rx_evtchn, xenvif_rx_interrupt, 0, |
604 | queue->rx_irq_name, queue); | |
e1f00a69 WL |
605 | if (err < 0) |
606 | goto err_tx_unbind; | |
e9ce7cb6 WL |
607 | queue->rx_irq = err; |
608 | disable_irq(queue->rx_irq); | |
e1f00a69 | 609 | } |
f942dc25 | 610 | |
121fa4b7 | 611 | task = kthread_create(xenvif_kthread_guest_rx, |
e9ce7cb6 | 612 | (void *)queue, "%s-guest-rx", queue->name); |
67fa3660 | 613 | if (IS_ERR(task)) { |
e9ce7cb6 | 614 | pr_warn("Could not allocate kthread for %s\n", queue->name); |
67fa3660 | 615 | err = PTR_ERR(task); |
b3f980bd WL |
616 | goto err_rx_unbind; |
617 | } | |
e9ce7cb6 | 618 | queue->task = task; |
67fa3660 | 619 | |
f53c3fe8 | 620 | task = kthread_create(xenvif_dealloc_kthread, |
e9ce7cb6 | 621 | (void *)queue, "%s-dealloc", queue->name); |
f53c3fe8 | 622 | if (IS_ERR(task)) { |
e9ce7cb6 | 623 | pr_warn("Could not allocate kthread for %s\n", queue->name); |
f53c3fe8 ZK |
624 | err = PTR_ERR(task); |
625 | goto err_rx_unbind; | |
626 | } | |
e9ce7cb6 | 627 | queue->dealloc_task = task; |
f53c3fe8 | 628 | |
e9ce7cb6 WL |
629 | wake_up_process(queue->task); |
630 | wake_up_process(queue->dealloc_task); | |
b3f980bd | 631 | |
ea2c5e13 WL |
632 | netif_napi_add(queue->vif->dev, &queue->napi, xenvif_poll, |
633 | XENVIF_NAPI_WEIGHT); | |
634 | ||
f942dc25 | 635 | return 0; |
b3f980bd WL |
636 | |
637 | err_rx_unbind: | |
e9ce7cb6 WL |
638 | unbind_from_irqhandler(queue->rx_irq, queue); |
639 | queue->rx_irq = 0; | |
e1f00a69 | 640 | err_tx_unbind: |
e9ce7cb6 WL |
641 | unbind_from_irqhandler(queue->tx_irq, queue); |
642 | queue->tx_irq = 0; | |
f942dc25 | 643 | err_unmap: |
e9ce7cb6 | 644 | xenvif_unmap_frontend_rings(queue); |
f942dc25 | 645 | err: |
b103f358 | 646 | module_put(THIS_MODULE); |
f942dc25 IC |
647 | return err; |
648 | } | |
649 | ||
48856286 | 650 | void xenvif_carrier_off(struct xenvif *vif) |
f942dc25 IC |
651 | { |
652 | struct net_device *dev = vif->dev; | |
48856286 IC |
653 | |
654 | rtnl_lock(); | |
3d1af1df ZK |
655 | if (test_and_clear_bit(VIF_STATUS_CONNECTED, &vif->status)) { |
656 | netif_carrier_off(dev); /* discard queued packets */ | |
657 | if (netif_running(dev)) | |
658 | xenvif_down(vif); | |
659 | } | |
48856286 | 660 | rtnl_unlock(); |
48856286 IC |
661 | } |
662 | ||
663 | void xenvif_disconnect(struct xenvif *vif) | |
664 | { | |
e9ce7cb6 | 665 | struct xenvif_queue *queue = NULL; |
f7b50c4e | 666 | unsigned int num_queues = vif->num_queues; |
e9ce7cb6 WL |
667 | unsigned int queue_index; |
668 | ||
3d1af1df | 669 | xenvif_carrier_off(vif); |
f942dc25 | 670 | |
e9ce7cb6 WL |
671 | for (queue_index = 0; queue_index < num_queues; ++queue_index) { |
672 | queue = &vif->queues[queue_index]; | |
db739ef3 | 673 | |
ea2c5e13 WL |
674 | netif_napi_del(&queue->napi); |
675 | ||
e9ce7cb6 | 676 | if (queue->task) { |
f34a4cf9 | 677 | del_timer_sync(&queue->rx_stalled); |
e9ce7cb6 WL |
678 | kthread_stop(queue->task); |
679 | queue->task = NULL; | |
680 | } | |
f53c3fe8 | 681 | |
e9ce7cb6 WL |
682 | if (queue->dealloc_task) { |
683 | kthread_stop(queue->dealloc_task); | |
684 | queue->dealloc_task = NULL; | |
685 | } | |
686 | ||
687 | if (queue->tx_irq) { | |
688 | if (queue->tx_irq == queue->rx_irq) | |
689 | unbind_from_irqhandler(queue->tx_irq, queue); | |
690 | else { | |
691 | unbind_from_irqhandler(queue->tx_irq, queue); | |
692 | unbind_from_irqhandler(queue->rx_irq, queue); | |
693 | } | |
694 | queue->tx_irq = 0; | |
e1f00a69 | 695 | } |
f942dc25 | 696 | |
e9ce7cb6 WL |
697 | xenvif_unmap_frontend_rings(queue); |
698 | } | |
279f438e PD |
699 | } |
700 | ||
8d3d53b3 AB |
701 | /* Reverse the relevant parts of xenvif_init_queue(). |
702 | * Used for queue teardown from xenvif_free(), and on the | |
703 | * error handling paths in xenbus.c:connect(). | |
704 | */ | |
705 | void xenvif_deinit_queue(struct xenvif_queue *queue) | |
706 | { | |
707 | free_xenballooned_pages(MAX_PENDING_REQS, queue->mmap_pages); | |
8d3d53b3 AB |
708 | } |
709 | ||
279f438e PD |
710 | void xenvif_free(struct xenvif *vif) |
711 | { | |
e9ce7cb6 | 712 | struct xenvif_queue *queue = NULL; |
f7b50c4e | 713 | unsigned int num_queues = vif->num_queues; |
e9ce7cb6 | 714 | unsigned int queue_index; |
f53c3fe8 | 715 | |
e9ce7cb6 | 716 | unregister_netdev(vif->dev); |
f53c3fe8 | 717 | |
e9ce7cb6 WL |
718 | for (queue_index = 0; queue_index < num_queues; ++queue_index) { |
719 | queue = &vif->queues[queue_index]; | |
8d3d53b3 | 720 | xenvif_deinit_queue(queue); |
e9ce7cb6 WL |
721 | } |
722 | ||
e9ce7cb6 WL |
723 | vfree(vif->queues); |
724 | vif->queues = NULL; | |
f7b50c4e | 725 | vif->num_queues = 0; |
f942dc25 | 726 | |
f942dc25 | 727 | free_netdev(vif->dev); |
b103f358 | 728 | |
279f438e | 729 | module_put(THIS_MODULE); |
f942dc25 | 730 | } |