]>
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> |
0881e7bd | 34 | #include <linux/sched/task.h> |
f942dc25 IC |
35 | #include <linux/ethtool.h> |
36 | #include <linux/rtnetlink.h> | |
37 | #include <linux/if_vlan.h> | |
e7b599d7 | 38 | #include <linux/vmalloc.h> |
f942dc25 IC |
39 | |
40 | #include <xen/events.h> | |
41 | #include <asm/xen/hypercall.h> | |
f53c3fe8 | 42 | #include <xen/balloon.h> |
f942dc25 IC |
43 | |
44 | #define XENVIF_QUEUE_LENGTH 32 | |
b3f980bd | 45 | #define XENVIF_NAPI_WEIGHT 64 |
f942dc25 | 46 | |
f48da8b1 DV |
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 | ||
a64bd934 WL |
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 | */ | |
55 | void 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 | ||
62 | void xenvif_skb_zerocopy_complete(struct xenvif_queue *queue) | |
63 | { | |
64 | atomic_dec(&queue->inflight_packets); | |
57b22906 RL |
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); | |
a64bd934 WL |
71 | } |
72 | ||
f942dc25 IC |
73 | int xenvif_schedulable(struct xenvif *vif) |
74 | { | |
3d1af1df | 75 | return netif_running(vif->dev) && |
f48da8b1 DV |
76 | test_bit(VIF_STATUS_CONNECTED, &vif->status) && |
77 | !vif->disabled; | |
f942dc25 IC |
78 | } |
79 | ||
e1f00a69 | 80 | static irqreturn_t xenvif_tx_interrupt(int irq, void *dev_id) |
f942dc25 | 81 | { |
e9ce7cb6 | 82 | struct xenvif_queue *queue = dev_id; |
f942dc25 | 83 | |
e9ce7cb6 WL |
84 | if (RING_HAS_UNCONSUMED_REQUESTS(&queue->tx)) |
85 | napi_schedule(&queue->napi); | |
f942dc25 | 86 | |
e1f00a69 WL |
87 | return IRQ_HANDLED; |
88 | } | |
89 | ||
38741d50 | 90 | static int xenvif_poll(struct napi_struct *napi, int budget) |
b3f980bd | 91 | { |
e9ce7cb6 WL |
92 | struct xenvif_queue *queue = |
93 | container_of(napi, struct xenvif_queue, napi); | |
b3f980bd WL |
94 | int work_done; |
95 | ||
e9d8b2c2 WL |
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 | */ | |
2561cc15 | 100 | if (unlikely(queue->vif->disabled)) { |
e9d8b2c2 WL |
101 | napi_complete(napi); |
102 | return 0; | |
103 | } | |
104 | ||
e9ce7cb6 | 105 | work_done = xenvif_tx_action(queue, budget); |
b3f980bd WL |
106 | |
107 | if (work_done < budget) { | |
6ad20165 | 108 | napi_complete_done(napi, work_done); |
dfa523ae WL |
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); | |
b3f980bd WL |
114 | } |
115 | ||
116 | return work_done; | |
117 | } | |
118 | ||
e1f00a69 WL |
119 | static irqreturn_t xenvif_rx_interrupt(int irq, void *dev_id) |
120 | { | |
e9ce7cb6 | 121 | struct xenvif_queue *queue = dev_id; |
e1f00a69 | 122 | |
e9ce7cb6 | 123 | xenvif_kick_thread(queue); |
f942dc25 IC |
124 | |
125 | return IRQ_HANDLED; | |
126 | } | |
127 | ||
f51de243 | 128 | irqreturn_t xenvif_interrupt(int irq, void *dev_id) |
e1f00a69 WL |
129 | { |
130 | xenvif_tx_interrupt(irq, dev_id); | |
131 | xenvif_rx_interrupt(irq, dev_id); | |
132 | ||
133 | return IRQ_HANDLED; | |
134 | } | |
135 | ||
e9ce7cb6 WL |
136 | int 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 | ||
143 | void 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 | ||
40d8abde PD |
150 | static 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 | ||
912e27e8 PD |
157 | if (vif->hash.alg == XEN_NETIF_CTRL_HASH_ALGORITHM_NONE) |
158 | return fallback(dev, skb) % dev->real_num_tx_queues; | |
40d8abde PD |
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 | ||
f942dc25 IC |
168 | static int xenvif_start_xmit(struct sk_buff *skb, struct net_device *dev) |
169 | { | |
170 | struct xenvif *vif = netdev_priv(dev); | |
e9ce7cb6 | 171 | struct xenvif_queue *queue = NULL; |
b17075d5 | 172 | unsigned int num_queues; |
e9ce7cb6 | 173 | u16 index; |
f48da8b1 | 174 | struct xenvif_rx_cb *cb; |
f942dc25 IC |
175 | |
176 | BUG_ON(skb->dev != dev); | |
177 | ||
b17075d5 ID |
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); | |
e9ce7cb6 WL |
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) { | |
cc10f871 | 189 | pr_warn_ratelimited("Invalid queue %hu for packet on interface %s\n", |
e9ce7cb6 WL |
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 || | |
f53c3fe8 | 198 | !xenvif_schedulable(vif)) |
f942dc25 IC |
199 | goto drop; |
200 | ||
210c34dc PD |
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 | ||
f48da8b1 | 208 | cb = XENVIF_RX_CB(skb); |
26c0e102 | 209 | cb->expires = jiffies + vif->drain_timeout; |
f942dc25 | 210 | |
912e27e8 PD |
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 | ||
f48da8b1 | 218 | xenvif_rx_queue_tail(queue, skb); |
e9ce7cb6 | 219 | xenvif_kick_thread(queue); |
f942dc25 IC |
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 | ||
f942dc25 IC |
229 | static struct net_device_stats *xenvif_get_stats(struct net_device *dev) |
230 | { | |
231 | struct xenvif *vif = netdev_priv(dev); | |
e9ce7cb6 | 232 | struct xenvif_queue *queue = NULL; |
b17075d5 | 233 | unsigned int num_queues; |
ebf692f8 MS |
234 | u64 rx_bytes = 0; |
235 | u64 rx_packets = 0; | |
236 | u64 tx_bytes = 0; | |
237 | u64 tx_packets = 0; | |
e9ce7cb6 WL |
238 | unsigned int index; |
239 | ||
b17075d5 ID |
240 | rcu_read_lock(); |
241 | num_queues = READ_ONCE(vif->num_queues); | |
e9ce7cb6 WL |
242 | |
243 | /* Aggregate tx and rx stats from each queue */ | |
b17075d5 | 244 | for (index = 0; index < num_queues; ++index) { |
e9ce7cb6 WL |
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 | ||
b17075d5 | 252 | rcu_read_unlock(); |
f16f1df6 | 253 | |
e9ce7cb6 WL |
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 | ||
f942dc25 IC |
259 | return &vif->dev->stats; |
260 | } | |
261 | ||
262 | static void xenvif_up(struct xenvif *vif) | |
263 | { | |
e9ce7cb6 | 264 | struct xenvif_queue *queue = NULL; |
f7b50c4e | 265 | unsigned int num_queues = vif->num_queues; |
e9ce7cb6 WL |
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 | } | |
f942dc25 IC |
276 | } |
277 | ||
278 | static void xenvif_down(struct xenvif *vif) | |
279 | { | |
e9ce7cb6 | 280 | struct xenvif_queue *queue = NULL; |
f7b50c4e | 281 | unsigned int num_queues = vif->num_queues; |
e9ce7cb6 WL |
282 | unsigned int queue_index; |
283 | ||
284 | for (queue_index = 0; queue_index < num_queues; ++queue_index) { | |
285 | queue = &vif->queues[queue_index]; | |
e9ce7cb6 WL |
286 | disable_irq(queue->tx_irq); |
287 | if (queue->tx_irq != queue->rx_irq) | |
288 | disable_irq(queue->rx_irq); | |
8fe78989 | 289 | napi_disable(&queue->napi); |
e9ce7cb6 WL |
290 | del_timer_sync(&queue->credit_timeout); |
291 | } | |
f942dc25 IC |
292 | } |
293 | ||
294 | static int xenvif_open(struct net_device *dev) | |
295 | { | |
296 | struct xenvif *vif = netdev_priv(dev); | |
3d1af1df | 297 | if (test_bit(VIF_STATUS_CONNECTED, &vif->status)) |
f942dc25 | 298 | xenvif_up(vif); |
e9ce7cb6 | 299 | netif_tx_start_all_queues(dev); |
f942dc25 IC |
300 | return 0; |
301 | } | |
302 | ||
303 | static int xenvif_close(struct net_device *dev) | |
304 | { | |
305 | struct xenvif *vif = netdev_priv(dev); | |
3d1af1df | 306 | if (test_bit(VIF_STATUS_CONNECTED, &vif->status)) |
f942dc25 | 307 | xenvif_down(vif); |
e9ce7cb6 | 308 | netif_tx_stop_all_queues(dev); |
f942dc25 IC |
309 | return 0; |
310 | } | |
311 | ||
312 | static int xenvif_change_mtu(struct net_device *dev, int mtu) | |
313 | { | |
314 | struct xenvif *vif = netdev_priv(dev); | |
d0c2c997 | 315 | int max = vif->can_sg ? ETH_MAX_MTU - VLAN_ETH_HLEN : ETH_DATA_LEN; |
f942dc25 IC |
316 | |
317 | if (mtu > max) | |
318 | return -EINVAL; | |
319 | dev->mtu = mtu; | |
320 | return 0; | |
321 | } | |
322 | ||
c8f44aff MM |
323 | static netdev_features_t xenvif_fix_features(struct net_device *dev, |
324 | netdev_features_t features) | |
f942dc25 IC |
325 | { |
326 | struct xenvif *vif = netdev_priv(dev); | |
f942dc25 | 327 | |
47103041 MM |
328 | if (!vif->can_sg) |
329 | features &= ~NETIF_F_SG; | |
fedbc8c1 | 330 | if (~(vif->gso_mask) & GSO_BIT(TCPV4)) |
47103041 | 331 | features &= ~NETIF_F_TSO; |
fedbc8c1 | 332 | if (~(vif->gso_mask) & GSO_BIT(TCPV6)) |
82cada22 | 333 | features &= ~NETIF_F_TSO6; |
146c8a77 | 334 | if (!vif->ip_csum) |
47103041 | 335 | features &= ~NETIF_F_IP_CSUM; |
146c8a77 PD |
336 | if (!vif->ipv6_csum) |
337 | features &= ~NETIF_F_IPV6_CSUM; | |
f942dc25 | 338 | |
47103041 | 339 | return features; |
f942dc25 IC |
340 | } |
341 | ||
342 | static const struct xenvif_stat { | |
343 | char name[ETH_GSTRING_LEN]; | |
344 | u16 offset; | |
345 | } xenvif_stats[] = { | |
346 | { | |
347 | "rx_gso_checksum_fixup", | |
e9ce7cb6 | 348 | offsetof(struct xenvif_stats, rx_gso_checksum_fixup) |
f942dc25 | 349 | }, |
1bb332af ZK |
350 | /* If (sent != success + fail), there are probably packets never |
351 | * freed up properly! | |
352 | */ | |
353 | { | |
354 | "tx_zerocopy_sent", | |
e9ce7cb6 | 355 | offsetof(struct xenvif_stats, tx_zerocopy_sent), |
1bb332af ZK |
356 | }, |
357 | { | |
358 | "tx_zerocopy_success", | |
e9ce7cb6 | 359 | offsetof(struct xenvif_stats, tx_zerocopy_success), |
1bb332af ZK |
360 | }, |
361 | { | |
362 | "tx_zerocopy_fail", | |
e9ce7cb6 | 363 | offsetof(struct xenvif_stats, tx_zerocopy_fail) |
1bb332af | 364 | }, |
e3377f36 ZK |
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", | |
e9ce7cb6 | 370 | offsetof(struct xenvif_stats, tx_frag_overflow) |
e3377f36 | 371 | }, |
f942dc25 IC |
372 | }; |
373 | ||
374 | static 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 | ||
384 | static void xenvif_get_ethtool_stats(struct net_device *dev, | |
385 | struct ethtool_stats *stats, u64 * data) | |
386 | { | |
e9ce7cb6 | 387 | struct xenvif *vif = netdev_priv(dev); |
b17075d5 | 388 | unsigned int num_queues; |
f942dc25 | 389 | int i; |
e9ce7cb6 | 390 | unsigned int queue_index; |
e9ce7cb6 | 391 | |
b17075d5 ID |
392 | rcu_read_lock(); |
393 | num_queues = READ_ONCE(vif->num_queues); | |
394 | ||
e9ce7cb6 WL |
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) { | |
d63951d7 | 398 | void *vif_stats = &vif->queues[queue_index].stats; |
e9ce7cb6 WL |
399 | accum += *(unsigned long *)(vif_stats + xenvif_stats[i].offset); |
400 | } | |
401 | data[i] = accum; | |
402 | } | |
b17075d5 ID |
403 | |
404 | rcu_read_unlock(); | |
f942dc25 IC |
405 | } |
406 | ||
407 | static 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 | ||
813abbba | 420 | static const struct ethtool_ops xenvif_ethtool_ops = { |
f942dc25 IC |
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 | ||
813abbba | 428 | static const struct net_device_ops xenvif_netdev_ops = { |
40d8abde | 429 | .ndo_select_queue = xenvif_select_queue, |
f942dc25 IC |
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, | |
47103041 | 435 | .ndo_fix_features = xenvif_fix_features, |
4a633a60 MW |
436 | .ndo_set_mac_address = eth_mac_addr, |
437 | .ndo_validate_addr = eth_validate_addr, | |
f942dc25 IC |
438 | }; |
439 | ||
440 | struct 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); | |
8d3d53b3 AB |
449 | /* Allocate a netdev with the max. supported number of queues. |
450 | * When the guest selects the desired number, it will be updated | |
f7b50c4e | 451 | * via netif_set_real_num_*_queues(). |
8d3d53b3 | 452 | */ |
c835a677 TG |
453 | dev = alloc_netdev_mq(sizeof(struct xenvif), name, NET_NAME_UNKNOWN, |
454 | ether_setup, xenvif_max_queues); | |
f942dc25 | 455 | if (dev == NULL) { |
b3f980bd | 456 | pr_warn("Could not allocate netdev for %s\n", name); |
f942dc25 IC |
457 | return ERR_PTR(-ENOMEM); |
458 | } | |
459 | ||
460 | SET_NETDEV_DEV(dev, parent); | |
461 | ||
462 | vif = netdev_priv(dev); | |
ac3d5ac2 | 463 | |
f942dc25 IC |
464 | vif->domid = domid; |
465 | vif->handle = handle; | |
f942dc25 | 466 | vif->can_sg = 1; |
146c8a77 | 467 | vif->ip_csum = 1; |
f942dc25 | 468 | vif->dev = dev; |
e9d8b2c2 | 469 | vif->disabled = false; |
26c0e102 DV |
470 | vif->drain_timeout = msecs_to_jiffies(rx_drain_timeout_msecs); |
471 | vif->stall_timeout = msecs_to_jiffies(rx_stall_timeout_msecs); | |
e9d8b2c2 | 472 | |
f7b50c4e | 473 | /* Start out with no queues. */ |
e9ce7cb6 | 474 | vif->queues = NULL; |
f7b50c4e | 475 | vif->num_queues = 0; |
09350788 | 476 | |
ecf08d2d | 477 | spin_lock_init(&vif->lock); |
210c34dc | 478 | INIT_LIST_HEAD(&vif->fe_mcast_addr); |
ecf08d2d | 479 | |
f942dc25 | 480 | dev->netdev_ops = &xenvif_netdev_ops; |
146c8a77 PD |
481 | dev->hw_features = NETIF_F_SG | |
482 | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | | |
2167ca02 | 483 | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_FRAGLIST; |
7365bcfa | 484 | dev->features = dev->hw_features | NETIF_F_RXCSUM; |
7ad24ea4 | 485 | dev->ethtool_ops = &xenvif_ethtool_ops; |
f942dc25 IC |
486 | |
487 | dev->tx_queue_len = XENVIF_QUEUE_LENGTH; | |
488 | ||
e1043a4b | 489 | dev->min_mtu = ETH_MIN_MTU; |
d0c2c997 JW |
490 | dev->max_mtu = ETH_MAX_MTU - VLAN_ETH_HLEN; |
491 | ||
f942dc25 IC |
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 | */ | |
3b6ed26d | 498 | eth_broadcast_addr(dev->dev_addr); |
f942dc25 IC |
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"); | |
279f438e PD |
511 | |
512 | __module_get(THIS_MODULE); | |
513 | ||
f942dc25 IC |
514 | return vif; |
515 | } | |
516 | ||
e9ce7cb6 WL |
517 | int 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; | |
cac6a8f9 | 523 | timer_setup(&queue->credit_timeout, xenvif_tx_credit_callback, 0); |
e9ce7cb6 WL |
524 | queue->credit_window_start = get_jiffies_64(); |
525 | ||
f48da8b1 DV |
526 | queue->rx_queue_max = XENVIF_RX_QUEUE_BYTES; |
527 | ||
e9ce7cb6 WL |
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 | */ | |
ff4b156f DV |
543 | err = gnttab_alloc_pages(MAX_PENDING_REQS, |
544 | queue->mmap_pages); | |
e9ce7cb6 WL |
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, | |
cc8737a5 WB |
553 | { { .ctx = NULL, |
554 | .desc = i } } }; | |
e9ce7cb6 WL |
555 | queue->grant_tx_handle[i] = NETBACK_INVALID_HANDLE; |
556 | } | |
557 | ||
e9ce7cb6 WL |
558 | return 0; |
559 | } | |
560 | ||
561 | void 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); | |
3d1af1df | 567 | set_bit(VIF_STATUS_CONNECTED, &vif->status); |
e9ce7cb6 WL |
568 | if (netif_running(vif->dev)) |
569 | xenvif_up(vif); | |
570 | rtnl_unlock(); | |
571 | } | |
572 | ||
4e15ee2c PD |
573 | int 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; | |
0364a882 | 579 | int err; |
4e15ee2c PD |
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 | ||
0364a882 | 589 | err = bind_interdomain_evtchn_to_irq(vif->domid, evtchn); |
4e15ee2c PD |
590 | if (err < 0) |
591 | goto err_unmap; | |
592 | ||
593 | vif->ctrl_irq = err; | |
594 | ||
40d8abde PD |
595 | xenvif_init_hash(vif); |
596 | ||
0364a882 JG |
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); | |
4e15ee2c PD |
601 | goto err_deinit; |
602 | } | |
603 | ||
4e15ee2c PD |
604 | return 0; |
605 | ||
606 | err_deinit: | |
40d8abde | 607 | xenvif_deinit_hash(vif); |
4e15ee2c PD |
608 | unbind_from_irqhandler(vif->ctrl_irq, vif); |
609 | vif->ctrl_irq = 0; | |
610 | ||
611 | err_unmap: | |
612 | xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif), | |
613 | vif->ctrl.sring); | |
614 | vif->ctrl.sring = NULL; | |
615 | ||
616 | err: | |
617 | return err; | |
618 | } | |
619 | ||
620 | int 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) | |
f942dc25 | 625 | { |
67fa3660 | 626 | struct task_struct *task; |
f942dc25 IC |
627 | int err = -ENOMEM; |
628 | ||
e9ce7cb6 WL |
629 | BUG_ON(queue->tx_irq); |
630 | BUG_ON(queue->task); | |
631 | BUG_ON(queue->dealloc_task); | |
f942dc25 | 632 | |
4e15ee2c PD |
633 | err = xenvif_map_frontend_data_rings(queue, tx_ring_ref, |
634 | rx_ring_ref); | |
f942dc25 IC |
635 | if (err < 0) |
636 | goto err; | |
637 | ||
e9ce7cb6 WL |
638 | init_waitqueue_head(&queue->wq); |
639 | init_waitqueue_head(&queue->dealloc_wq); | |
a64bd934 | 640 | atomic_set(&queue->inflight_packets, 0); |
ca2f09f2 | 641 | |
e24f8191 WL |
642 | netif_napi_add(queue->vif->dev, &queue->napi, xenvif_poll, |
643 | XENVIF_NAPI_WEIGHT); | |
644 | ||
e1f00a69 WL |
645 | if (tx_evtchn == rx_evtchn) { |
646 | /* feature-split-event-channels == 0 */ | |
647 | err = bind_interdomain_evtchn_to_irqhandler( | |
e9ce7cb6 WL |
648 | queue->vif->domid, tx_evtchn, xenvif_interrupt, 0, |
649 | queue->name, queue); | |
e1f00a69 WL |
650 | if (err < 0) |
651 | goto err_unmap; | |
e9ce7cb6 WL |
652 | queue->tx_irq = queue->rx_irq = err; |
653 | disable_irq(queue->tx_irq); | |
e1f00a69 WL |
654 | } else { |
655 | /* feature-split-event-channels == 1 */ | |
e9ce7cb6 WL |
656 | snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name), |
657 | "%s-tx", queue->name); | |
e1f00a69 | 658 | err = bind_interdomain_evtchn_to_irqhandler( |
e9ce7cb6 WL |
659 | queue->vif->domid, tx_evtchn, xenvif_tx_interrupt, 0, |
660 | queue->tx_irq_name, queue); | |
e1f00a69 WL |
661 | if (err < 0) |
662 | goto err_unmap; | |
e9ce7cb6 WL |
663 | queue->tx_irq = err; |
664 | disable_irq(queue->tx_irq); | |
e1f00a69 | 665 | |
e9ce7cb6 WL |
666 | snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name), |
667 | "%s-rx", queue->name); | |
e1f00a69 | 668 | err = bind_interdomain_evtchn_to_irqhandler( |
e9ce7cb6 WL |
669 | queue->vif->domid, rx_evtchn, xenvif_rx_interrupt, 0, |
670 | queue->rx_irq_name, queue); | |
e1f00a69 WL |
671 | if (err < 0) |
672 | goto err_tx_unbind; | |
e9ce7cb6 WL |
673 | queue->rx_irq = err; |
674 | disable_irq(queue->rx_irq); | |
e1f00a69 | 675 | } |
f942dc25 | 676 | |
ecf08d2d DV |
677 | queue->stalled = true; |
678 | ||
121fa4b7 | 679 | task = kthread_create(xenvif_kthread_guest_rx, |
e9ce7cb6 | 680 | (void *)queue, "%s-guest-rx", queue->name); |
67fa3660 | 681 | if (IS_ERR(task)) { |
e9ce7cb6 | 682 | pr_warn("Could not allocate kthread for %s\n", queue->name); |
67fa3660 | 683 | err = PTR_ERR(task); |
b3f980bd WL |
684 | goto err_rx_unbind; |
685 | } | |
e9ce7cb6 | 686 | queue->task = task; |
42b5212f | 687 | get_task_struct(task); |
67fa3660 | 688 | |
f53c3fe8 | 689 | task = kthread_create(xenvif_dealloc_kthread, |
e9ce7cb6 | 690 | (void *)queue, "%s-dealloc", queue->name); |
f53c3fe8 | 691 | if (IS_ERR(task)) { |
e9ce7cb6 | 692 | pr_warn("Could not allocate kthread for %s\n", queue->name); |
f53c3fe8 ZK |
693 | err = PTR_ERR(task); |
694 | goto err_rx_unbind; | |
695 | } | |
e9ce7cb6 | 696 | queue->dealloc_task = task; |
f53c3fe8 | 697 | |
e9ce7cb6 WL |
698 | wake_up_process(queue->task); |
699 | wake_up_process(queue->dealloc_task); | |
b3f980bd | 700 | |
f942dc25 | 701 | return 0; |
b3f980bd WL |
702 | |
703 | err_rx_unbind: | |
e9ce7cb6 WL |
704 | unbind_from_irqhandler(queue->rx_irq, queue); |
705 | queue->rx_irq = 0; | |
e1f00a69 | 706 | err_tx_unbind: |
e9ce7cb6 WL |
707 | unbind_from_irqhandler(queue->tx_irq, queue); |
708 | queue->tx_irq = 0; | |
f942dc25 | 709 | err_unmap: |
4e15ee2c | 710 | xenvif_unmap_frontend_data_rings(queue); |
4a658527 | 711 | netif_napi_del(&queue->napi); |
f942dc25 | 712 | err: |
b103f358 | 713 | module_put(THIS_MODULE); |
f942dc25 IC |
714 | return err; |
715 | } | |
716 | ||
48856286 | 717 | void xenvif_carrier_off(struct xenvif *vif) |
f942dc25 IC |
718 | { |
719 | struct net_device *dev = vif->dev; | |
48856286 IC |
720 | |
721 | rtnl_lock(); | |
3d1af1df ZK |
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 | } | |
48856286 | 727 | rtnl_unlock(); |
48856286 IC |
728 | } |
729 | ||
4e15ee2c | 730 | void xenvif_disconnect_data(struct xenvif *vif) |
48856286 | 731 | { |
e9ce7cb6 | 732 | struct xenvif_queue *queue = NULL; |
f7b50c4e | 733 | unsigned int num_queues = vif->num_queues; |
e9ce7cb6 WL |
734 | unsigned int queue_index; |
735 | ||
3d1af1df | 736 | xenvif_carrier_off(vif); |
f942dc25 | 737 | |
e9ce7cb6 WL |
738 | for (queue_index = 0; queue_index < num_queues; ++queue_index) { |
739 | queue = &vif->queues[queue_index]; | |
db739ef3 | 740 | |
ea2c5e13 WL |
741 | netif_napi_del(&queue->napi); |
742 | ||
e9ce7cb6 | 743 | if (queue->task) { |
e9ce7cb6 | 744 | kthread_stop(queue->task); |
42b5212f | 745 | put_task_struct(queue->task); |
e9ce7cb6 WL |
746 | queue->task = NULL; |
747 | } | |
f53c3fe8 | 748 | |
e9ce7cb6 WL |
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; | |
e1f00a69 | 762 | } |
f942dc25 | 763 | |
4e15ee2c | 764 | xenvif_unmap_frontend_data_rings(queue); |
e9ce7cb6 | 765 | } |
210c34dc PD |
766 | |
767 | xenvif_mcast_addr_list_free(vif); | |
279f438e PD |
768 | } |
769 | ||
4e15ee2c PD |
770 | void xenvif_disconnect_ctrl(struct xenvif *vif) |
771 | { | |
4e15ee2c | 772 | if (vif->ctrl_irq) { |
c0fcded2 | 773 | xenvif_deinit_hash(vif); |
4e15ee2c PD |
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 | ||
8d3d53b3 AB |
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 | */ | |
789 | void xenvif_deinit_queue(struct xenvif_queue *queue) | |
790 | { | |
ff4b156f | 791 | gnttab_free_pages(MAX_PENDING_REQS, queue->mmap_pages); |
8d3d53b3 AB |
792 | } |
793 | ||
279f438e PD |
794 | void xenvif_free(struct xenvif *vif) |
795 | { | |
9c6f3ffe | 796 | struct xenvif_queue *queues = vif->queues; |
f7b50c4e | 797 | unsigned int num_queues = vif->num_queues; |
e9ce7cb6 | 798 | unsigned int queue_index; |
f53c3fe8 | 799 | |
e9ce7cb6 | 800 | unregister_netdev(vif->dev); |
f942dc25 | 801 | free_netdev(vif->dev); |
b103f358 | 802 | |
9c6f3ffe DV |
803 | for (queue_index = 0; queue_index < num_queues; ++queue_index) |
804 | xenvif_deinit_queue(&queues[queue_index]); | |
805 | vfree(queues); | |
806 | ||
279f438e | 807 | module_put(THIS_MODULE); |
f942dc25 | 808 | } |