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 | |
f48da8b1 DV |
46 | /* Number of bytes allowed on the internal guest Rx queue. */ |
47 | #define XENVIF_RX_QUEUE_BYTES (XEN_NETIF_RX_RING_SIZE/2 * PAGE_SIZE) | |
48 | ||
a64bd934 WL |
49 | /* This function is used to set SKBTX_DEV_ZEROCOPY as well as |
50 | * increasing the inflight counter. We need to increase the inflight | |
51 | * counter because core driver calls into xenvif_zerocopy_callback | |
52 | * which calls xenvif_skb_zerocopy_complete. | |
53 | */ | |
54 | void xenvif_skb_zerocopy_prepare(struct xenvif_queue *queue, | |
55 | struct sk_buff *skb) | |
56 | { | |
57 | skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY; | |
58 | atomic_inc(&queue->inflight_packets); | |
59 | } | |
60 | ||
61 | void xenvif_skb_zerocopy_complete(struct xenvif_queue *queue) | |
62 | { | |
63 | atomic_dec(&queue->inflight_packets); | |
57b22906 RL |
64 | |
65 | /* Wake the dealloc thread _after_ decrementing inflight_packets so | |
66 | * that if kthread_stop() has already been called, the dealloc thread | |
67 | * does not wait forever with nothing to wake it. | |
68 | */ | |
69 | wake_up(&queue->dealloc_wq); | |
a64bd934 WL |
70 | } |
71 | ||
f942dc25 IC |
72 | int xenvif_schedulable(struct xenvif *vif) |
73 | { | |
3d1af1df | 74 | return netif_running(vif->dev) && |
f48da8b1 DV |
75 | test_bit(VIF_STATUS_CONNECTED, &vif->status) && |
76 | !vif->disabled; | |
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 | ||
38741d50 | 89 | static 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; |
e1f00a69 | 117 | |
e9ce7cb6 | 118 | xenvif_kick_thread(queue); |
f942dc25 IC |
119 | |
120 | return IRQ_HANDLED; | |
121 | } | |
122 | ||
f51de243 | 123 | irqreturn_t xenvif_interrupt(int irq, void *dev_id) |
e1f00a69 WL |
124 | { |
125 | xenvif_tx_interrupt(irq, dev_id); | |
126 | xenvif_rx_interrupt(irq, dev_id); | |
127 | ||
128 | return IRQ_HANDLED; | |
129 | } | |
130 | ||
4e15ee2c PD |
131 | irqreturn_t xenvif_ctrl_interrupt(int irq, void *dev_id) |
132 | { | |
133 | struct xenvif *vif = dev_id; | |
134 | ||
135 | wake_up(&vif->ctrl_wq); | |
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 | ||
40d8abde PD |
154 | static u16 xenvif_select_queue(struct net_device *dev, struct sk_buff *skb, |
155 | void *accel_priv, | |
156 | select_queue_fallback_t fallback) | |
157 | { | |
158 | struct xenvif *vif = netdev_priv(dev); | |
159 | unsigned int size = vif->hash.size; | |
160 | ||
161 | if (vif->hash.alg == XEN_NETIF_CTRL_HASH_ALGORITHM_NONE) | |
162 | return fallback(dev, skb) % dev->real_num_tx_queues; | |
163 | ||
164 | xenvif_set_skb_hash(vif, skb); | |
165 | ||
166 | if (size == 0) | |
167 | return skb_get_hash_raw(skb) % dev->real_num_tx_queues; | |
168 | ||
169 | return vif->hash.mapping[skb_get_hash_raw(skb) % size]; | |
170 | } | |
171 | ||
f942dc25 IC |
172 | static int xenvif_start_xmit(struct sk_buff *skb, struct net_device *dev) |
173 | { | |
174 | struct xenvif *vif = netdev_priv(dev); | |
e9ce7cb6 | 175 | struct xenvif_queue *queue = NULL; |
f7b50c4e | 176 | unsigned int num_queues = vif->num_queues; |
e9ce7cb6 | 177 | u16 index; |
f48da8b1 | 178 | struct xenvif_rx_cb *cb; |
f942dc25 IC |
179 | |
180 | BUG_ON(skb->dev != dev); | |
181 | ||
e9ce7cb6 WL |
182 | /* Drop the packet if queues are not set up */ |
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 || | |
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 | |
f48da8b1 | 211 | xenvif_rx_queue_tail(queue, skb); |
e9ce7cb6 | 212 | xenvif_kick_thread(queue); |
f942dc25 IC |
213 | |
214 | return NETDEV_TX_OK; | |
215 | ||
216 | drop: | |
217 | vif->dev->stats.tx_dropped++; | |
218 | dev_kfree_skb(skb); | |
219 | return NETDEV_TX_OK; | |
220 | } | |
221 | ||
f942dc25 IC |
222 | static struct net_device_stats *xenvif_get_stats(struct net_device *dev) |
223 | { | |
224 | struct xenvif *vif = netdev_priv(dev); | |
e9ce7cb6 | 225 | struct xenvif_queue *queue = NULL; |
f7b50c4e | 226 | unsigned int num_queues = vif->num_queues; |
e9ce7cb6 WL |
227 | unsigned long rx_bytes = 0; |
228 | unsigned long rx_packets = 0; | |
229 | unsigned long tx_bytes = 0; | |
230 | unsigned long tx_packets = 0; | |
231 | unsigned int index; | |
232 | ||
233 | if (vif->queues == NULL) | |
234 | goto out; | |
235 | ||
236 | /* Aggregate tx and rx stats from each queue */ | |
237 | for (index = 0; index < num_queues; ++index) { | |
238 | queue = &vif->queues[index]; | |
239 | rx_bytes += queue->stats.rx_bytes; | |
240 | rx_packets += queue->stats.rx_packets; | |
241 | tx_bytes += queue->stats.tx_bytes; | |
242 | tx_packets += queue->stats.tx_packets; | |
243 | } | |
244 | ||
245 | out: | |
246 | vif->dev->stats.rx_bytes = rx_bytes; | |
247 | vif->dev->stats.rx_packets = rx_packets; | |
248 | vif->dev->stats.tx_bytes = tx_bytes; | |
249 | vif->dev->stats.tx_packets = tx_packets; | |
250 | ||
f942dc25 IC |
251 | return &vif->dev->stats; |
252 | } | |
253 | ||
254 | static void xenvif_up(struct xenvif *vif) | |
255 | { | |
e9ce7cb6 | 256 | struct xenvif_queue *queue = NULL; |
f7b50c4e | 257 | unsigned int num_queues = vif->num_queues; |
e9ce7cb6 WL |
258 | unsigned int queue_index; |
259 | ||
260 | for (queue_index = 0; queue_index < num_queues; ++queue_index) { | |
261 | queue = &vif->queues[queue_index]; | |
262 | napi_enable(&queue->napi); | |
263 | enable_irq(queue->tx_irq); | |
264 | if (queue->tx_irq != queue->rx_irq) | |
265 | enable_irq(queue->rx_irq); | |
266 | xenvif_napi_schedule_or_enable_events(queue); | |
267 | } | |
f942dc25 IC |
268 | } |
269 | ||
270 | static void xenvif_down(struct xenvif *vif) | |
271 | { | |
e9ce7cb6 | 272 | struct xenvif_queue *queue = NULL; |
f7b50c4e | 273 | unsigned int num_queues = vif->num_queues; |
e9ce7cb6 WL |
274 | unsigned int queue_index; |
275 | ||
276 | for (queue_index = 0; queue_index < num_queues; ++queue_index) { | |
277 | queue = &vif->queues[queue_index]; | |
e9ce7cb6 WL |
278 | disable_irq(queue->tx_irq); |
279 | if (queue->tx_irq != queue->rx_irq) | |
280 | disable_irq(queue->rx_irq); | |
8fe78989 | 281 | napi_disable(&queue->napi); |
e9ce7cb6 WL |
282 | del_timer_sync(&queue->credit_timeout); |
283 | } | |
f942dc25 IC |
284 | } |
285 | ||
286 | static int xenvif_open(struct net_device *dev) | |
287 | { | |
288 | struct xenvif *vif = netdev_priv(dev); | |
3d1af1df | 289 | if (test_bit(VIF_STATUS_CONNECTED, &vif->status)) |
f942dc25 | 290 | xenvif_up(vif); |
e9ce7cb6 | 291 | netif_tx_start_all_queues(dev); |
f942dc25 IC |
292 | return 0; |
293 | } | |
294 | ||
295 | static int xenvif_close(struct net_device *dev) | |
296 | { | |
297 | struct xenvif *vif = netdev_priv(dev); | |
3d1af1df | 298 | if (test_bit(VIF_STATUS_CONNECTED, &vif->status)) |
f942dc25 | 299 | xenvif_down(vif); |
e9ce7cb6 | 300 | netif_tx_stop_all_queues(dev); |
f942dc25 IC |
301 | return 0; |
302 | } | |
303 | ||
304 | static int xenvif_change_mtu(struct net_device *dev, int mtu) | |
305 | { | |
306 | struct xenvif *vif = netdev_priv(dev); | |
307 | int max = vif->can_sg ? 65535 - VLAN_ETH_HLEN : ETH_DATA_LEN; | |
308 | ||
309 | if (mtu > max) | |
310 | return -EINVAL; | |
311 | dev->mtu = mtu; | |
312 | return 0; | |
313 | } | |
314 | ||
c8f44aff MM |
315 | static netdev_features_t xenvif_fix_features(struct net_device *dev, |
316 | netdev_features_t features) | |
f942dc25 IC |
317 | { |
318 | struct xenvif *vif = netdev_priv(dev); | |
f942dc25 | 319 | |
47103041 MM |
320 | if (!vif->can_sg) |
321 | features &= ~NETIF_F_SG; | |
82cada22 | 322 | if (~(vif->gso_mask | vif->gso_prefix_mask) & GSO_BIT(TCPV4)) |
47103041 | 323 | features &= ~NETIF_F_TSO; |
82cada22 PD |
324 | if (~(vif->gso_mask | vif->gso_prefix_mask) & GSO_BIT(TCPV6)) |
325 | features &= ~NETIF_F_TSO6; | |
146c8a77 | 326 | if (!vif->ip_csum) |
47103041 | 327 | features &= ~NETIF_F_IP_CSUM; |
146c8a77 PD |
328 | if (!vif->ipv6_csum) |
329 | features &= ~NETIF_F_IPV6_CSUM; | |
f942dc25 | 330 | |
47103041 | 331 | return features; |
f942dc25 IC |
332 | } |
333 | ||
334 | static const struct xenvif_stat { | |
335 | char name[ETH_GSTRING_LEN]; | |
336 | u16 offset; | |
337 | } xenvif_stats[] = { | |
338 | { | |
339 | "rx_gso_checksum_fixup", | |
e9ce7cb6 | 340 | offsetof(struct xenvif_stats, rx_gso_checksum_fixup) |
f942dc25 | 341 | }, |
1bb332af ZK |
342 | /* If (sent != success + fail), there are probably packets never |
343 | * freed up properly! | |
344 | */ | |
345 | { | |
346 | "tx_zerocopy_sent", | |
e9ce7cb6 | 347 | offsetof(struct xenvif_stats, tx_zerocopy_sent), |
1bb332af ZK |
348 | }, |
349 | { | |
350 | "tx_zerocopy_success", | |
e9ce7cb6 | 351 | offsetof(struct xenvif_stats, tx_zerocopy_success), |
1bb332af ZK |
352 | }, |
353 | { | |
354 | "tx_zerocopy_fail", | |
e9ce7cb6 | 355 | offsetof(struct xenvif_stats, tx_zerocopy_fail) |
1bb332af | 356 | }, |
e3377f36 ZK |
357 | /* Number of packets exceeding MAX_SKB_FRAG slots. You should use |
358 | * a guest with the same MAX_SKB_FRAG | |
359 | */ | |
360 | { | |
361 | "tx_frag_overflow", | |
e9ce7cb6 | 362 | offsetof(struct xenvif_stats, tx_frag_overflow) |
e3377f36 | 363 | }, |
f942dc25 IC |
364 | }; |
365 | ||
366 | static int xenvif_get_sset_count(struct net_device *dev, int string_set) | |
367 | { | |
368 | switch (string_set) { | |
369 | case ETH_SS_STATS: | |
370 | return ARRAY_SIZE(xenvif_stats); | |
371 | default: | |
372 | return -EINVAL; | |
373 | } | |
374 | } | |
375 | ||
376 | static void xenvif_get_ethtool_stats(struct net_device *dev, | |
377 | struct ethtool_stats *stats, u64 * data) | |
378 | { | |
e9ce7cb6 | 379 | struct xenvif *vif = netdev_priv(dev); |
f7b50c4e | 380 | unsigned int num_queues = vif->num_queues; |
f942dc25 | 381 | int i; |
e9ce7cb6 | 382 | unsigned int queue_index; |
e9ce7cb6 WL |
383 | |
384 | for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++) { | |
385 | unsigned long accum = 0; | |
386 | for (queue_index = 0; queue_index < num_queues; ++queue_index) { | |
d63951d7 | 387 | void *vif_stats = &vif->queues[queue_index].stats; |
e9ce7cb6 WL |
388 | accum += *(unsigned long *)(vif_stats + xenvif_stats[i].offset); |
389 | } | |
390 | data[i] = accum; | |
391 | } | |
f942dc25 IC |
392 | } |
393 | ||
394 | static void xenvif_get_strings(struct net_device *dev, u32 stringset, u8 * data) | |
395 | { | |
396 | int i; | |
397 | ||
398 | switch (stringset) { | |
399 | case ETH_SS_STATS: | |
400 | for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++) | |
401 | memcpy(data + i * ETH_GSTRING_LEN, | |
402 | xenvif_stats[i].name, ETH_GSTRING_LEN); | |
403 | break; | |
404 | } | |
405 | } | |
406 | ||
813abbba | 407 | static const struct ethtool_ops xenvif_ethtool_ops = { |
f942dc25 IC |
408 | .get_link = ethtool_op_get_link, |
409 | ||
410 | .get_sset_count = xenvif_get_sset_count, | |
411 | .get_ethtool_stats = xenvif_get_ethtool_stats, | |
412 | .get_strings = xenvif_get_strings, | |
413 | }; | |
414 | ||
813abbba | 415 | static const struct net_device_ops xenvif_netdev_ops = { |
40d8abde | 416 | .ndo_select_queue = xenvif_select_queue, |
f942dc25 IC |
417 | .ndo_start_xmit = xenvif_start_xmit, |
418 | .ndo_get_stats = xenvif_get_stats, | |
419 | .ndo_open = xenvif_open, | |
420 | .ndo_stop = xenvif_close, | |
421 | .ndo_change_mtu = xenvif_change_mtu, | |
47103041 | 422 | .ndo_fix_features = xenvif_fix_features, |
4a633a60 MW |
423 | .ndo_set_mac_address = eth_mac_addr, |
424 | .ndo_validate_addr = eth_validate_addr, | |
f942dc25 IC |
425 | }; |
426 | ||
427 | struct xenvif *xenvif_alloc(struct device *parent, domid_t domid, | |
428 | unsigned int handle) | |
429 | { | |
430 | int err; | |
431 | struct net_device *dev; | |
432 | struct xenvif *vif; | |
433 | char name[IFNAMSIZ] = {}; | |
434 | ||
435 | snprintf(name, IFNAMSIZ - 1, "vif%u.%u", domid, handle); | |
8d3d53b3 AB |
436 | /* Allocate a netdev with the max. supported number of queues. |
437 | * When the guest selects the desired number, it will be updated | |
f7b50c4e | 438 | * via netif_set_real_num_*_queues(). |
8d3d53b3 | 439 | */ |
c835a677 TG |
440 | dev = alloc_netdev_mq(sizeof(struct xenvif), name, NET_NAME_UNKNOWN, |
441 | ether_setup, xenvif_max_queues); | |
f942dc25 | 442 | if (dev == NULL) { |
b3f980bd | 443 | pr_warn("Could not allocate netdev for %s\n", name); |
f942dc25 IC |
444 | return ERR_PTR(-ENOMEM); |
445 | } | |
446 | ||
447 | SET_NETDEV_DEV(dev, parent); | |
448 | ||
449 | vif = netdev_priv(dev); | |
ac3d5ac2 | 450 | |
f942dc25 IC |
451 | vif->domid = domid; |
452 | vif->handle = handle; | |
f942dc25 | 453 | vif->can_sg = 1; |
146c8a77 | 454 | vif->ip_csum = 1; |
f942dc25 | 455 | vif->dev = dev; |
e9d8b2c2 | 456 | vif->disabled = false; |
26c0e102 DV |
457 | vif->drain_timeout = msecs_to_jiffies(rx_drain_timeout_msecs); |
458 | vif->stall_timeout = msecs_to_jiffies(rx_stall_timeout_msecs); | |
e9d8b2c2 | 459 | |
f7b50c4e | 460 | /* Start out with no queues. */ |
e9ce7cb6 | 461 | vif->queues = NULL; |
f7b50c4e | 462 | vif->num_queues = 0; |
09350788 | 463 | |
ecf08d2d | 464 | spin_lock_init(&vif->lock); |
210c34dc | 465 | INIT_LIST_HEAD(&vif->fe_mcast_addr); |
ecf08d2d | 466 | |
f942dc25 | 467 | dev->netdev_ops = &xenvif_netdev_ops; |
146c8a77 PD |
468 | dev->hw_features = NETIF_F_SG | |
469 | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | | |
82cada22 | 470 | NETIF_F_TSO | NETIF_F_TSO6; |
7365bcfa | 471 | dev->features = dev->hw_features | NETIF_F_RXCSUM; |
7ad24ea4 | 472 | dev->ethtool_ops = &xenvif_ethtool_ops; |
f942dc25 IC |
473 | |
474 | dev->tx_queue_len = XENVIF_QUEUE_LENGTH; | |
475 | ||
476 | /* | |
477 | * Initialise a dummy MAC address. We choose the numerically | |
478 | * largest non-broadcast address to prevent the address getting | |
479 | * stolen by an Ethernet bridge for STP purposes. | |
480 | * (FE:FF:FF:FF:FF:FF) | |
481 | */ | |
3b6ed26d | 482 | eth_broadcast_addr(dev->dev_addr); |
f942dc25 IC |
483 | dev->dev_addr[0] &= ~0x01; |
484 | ||
485 | netif_carrier_off(dev); | |
486 | ||
487 | err = register_netdev(dev); | |
488 | if (err) { | |
489 | netdev_warn(dev, "Could not register device: err=%d\n", err); | |
490 | free_netdev(dev); | |
491 | return ERR_PTR(err); | |
492 | } | |
493 | ||
494 | netdev_dbg(dev, "Successfully created xenvif\n"); | |
279f438e PD |
495 | |
496 | __module_get(THIS_MODULE); | |
497 | ||
f942dc25 IC |
498 | return vif; |
499 | } | |
500 | ||
e9ce7cb6 WL |
501 | int xenvif_init_queue(struct xenvif_queue *queue) |
502 | { | |
503 | int err, i; | |
504 | ||
505 | queue->credit_bytes = queue->remaining_credit = ~0UL; | |
506 | queue->credit_usec = 0UL; | |
507 | init_timer(&queue->credit_timeout); | |
edafc132 | 508 | queue->credit_timeout.function = xenvif_tx_credit_callback; |
e9ce7cb6 WL |
509 | queue->credit_window_start = get_jiffies_64(); |
510 | ||
f48da8b1 DV |
511 | queue->rx_queue_max = XENVIF_RX_QUEUE_BYTES; |
512 | ||
e9ce7cb6 WL |
513 | skb_queue_head_init(&queue->rx_queue); |
514 | skb_queue_head_init(&queue->tx_queue); | |
515 | ||
516 | queue->pending_cons = 0; | |
517 | queue->pending_prod = MAX_PENDING_REQS; | |
518 | for (i = 0; i < MAX_PENDING_REQS; ++i) | |
519 | queue->pending_ring[i] = i; | |
520 | ||
521 | spin_lock_init(&queue->callback_lock); | |
522 | spin_lock_init(&queue->response_lock); | |
523 | ||
524 | /* If ballooning is disabled, this will consume real memory, so you | |
525 | * better enable it. The long term solution would be to use just a | |
526 | * bunch of valid page descriptors, without dependency on ballooning | |
527 | */ | |
ff4b156f DV |
528 | err = gnttab_alloc_pages(MAX_PENDING_REQS, |
529 | queue->mmap_pages); | |
e9ce7cb6 WL |
530 | if (err) { |
531 | netdev_err(queue->vif->dev, "Could not reserve mmap_pages\n"); | |
532 | return -ENOMEM; | |
533 | } | |
534 | ||
535 | for (i = 0; i < MAX_PENDING_REQS; i++) { | |
536 | queue->pending_tx_info[i].callback_struct = (struct ubuf_info) | |
537 | { .callback = xenvif_zerocopy_callback, | |
538 | .ctx = NULL, | |
539 | .desc = i }; | |
540 | queue->grant_tx_handle[i] = NETBACK_INVALID_HANDLE; | |
541 | } | |
542 | ||
e9ce7cb6 WL |
543 | return 0; |
544 | } | |
545 | ||
546 | void xenvif_carrier_on(struct xenvif *vif) | |
547 | { | |
548 | rtnl_lock(); | |
549 | if (!vif->can_sg && vif->dev->mtu > ETH_DATA_LEN) | |
550 | dev_set_mtu(vif->dev, ETH_DATA_LEN); | |
551 | netdev_update_features(vif->dev); | |
3d1af1df | 552 | set_bit(VIF_STATUS_CONNECTED, &vif->status); |
e9ce7cb6 WL |
553 | if (netif_running(vif->dev)) |
554 | xenvif_up(vif); | |
555 | rtnl_unlock(); | |
556 | } | |
557 | ||
4e15ee2c PD |
558 | int xenvif_connect_ctrl(struct xenvif *vif, grant_ref_t ring_ref, |
559 | unsigned int evtchn) | |
560 | { | |
561 | struct net_device *dev = vif->dev; | |
562 | void *addr; | |
563 | struct xen_netif_ctrl_sring *shared; | |
564 | struct task_struct *task; | |
565 | int err = -ENOMEM; | |
566 | ||
567 | err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(vif), | |
568 | &ring_ref, 1, &addr); | |
569 | if (err) | |
570 | goto err; | |
571 | ||
572 | shared = (struct xen_netif_ctrl_sring *)addr; | |
573 | BACK_RING_INIT(&vif->ctrl, shared, XEN_PAGE_SIZE); | |
574 | ||
575 | init_waitqueue_head(&vif->ctrl_wq); | |
576 | ||
577 | err = bind_interdomain_evtchn_to_irqhandler(vif->domid, evtchn, | |
578 | xenvif_ctrl_interrupt, | |
579 | 0, dev->name, vif); | |
580 | if (err < 0) | |
581 | goto err_unmap; | |
582 | ||
583 | vif->ctrl_irq = err; | |
584 | ||
40d8abde PD |
585 | xenvif_init_hash(vif); |
586 | ||
4e15ee2c PD |
587 | task = kthread_create(xenvif_ctrl_kthread, (void *)vif, |
588 | "%s-control", dev->name); | |
589 | if (IS_ERR(task)) { | |
590 | pr_warn("Could not allocate kthread for %s\n", dev->name); | |
591 | err = PTR_ERR(task); | |
592 | goto err_deinit; | |
593 | } | |
594 | ||
595 | get_task_struct(task); | |
596 | vif->ctrl_task = task; | |
597 | ||
598 | wake_up_process(vif->ctrl_task); | |
599 | ||
600 | return 0; | |
601 | ||
602 | err_deinit: | |
40d8abde | 603 | xenvif_deinit_hash(vif); |
4e15ee2c PD |
604 | unbind_from_irqhandler(vif->ctrl_irq, vif); |
605 | vif->ctrl_irq = 0; | |
606 | ||
607 | err_unmap: | |
608 | xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif), | |
609 | vif->ctrl.sring); | |
610 | vif->ctrl.sring = NULL; | |
611 | ||
612 | err: | |
613 | return err; | |
614 | } | |
615 | ||
616 | int xenvif_connect_data(struct xenvif_queue *queue, | |
617 | unsigned long tx_ring_ref, | |
618 | unsigned long rx_ring_ref, | |
619 | unsigned int tx_evtchn, | |
620 | unsigned int rx_evtchn) | |
f942dc25 | 621 | { |
67fa3660 | 622 | struct task_struct *task; |
f942dc25 IC |
623 | int err = -ENOMEM; |
624 | ||
e9ce7cb6 WL |
625 | BUG_ON(queue->tx_irq); |
626 | BUG_ON(queue->task); | |
627 | BUG_ON(queue->dealloc_task); | |
f942dc25 | 628 | |
4e15ee2c PD |
629 | err = xenvif_map_frontend_data_rings(queue, tx_ring_ref, |
630 | rx_ring_ref); | |
f942dc25 IC |
631 | if (err < 0) |
632 | goto err; | |
633 | ||
e9ce7cb6 WL |
634 | init_waitqueue_head(&queue->wq); |
635 | init_waitqueue_head(&queue->dealloc_wq); | |
a64bd934 | 636 | atomic_set(&queue->inflight_packets, 0); |
ca2f09f2 | 637 | |
e24f8191 WL |
638 | netif_napi_add(queue->vif->dev, &queue->napi, xenvif_poll, |
639 | XENVIF_NAPI_WEIGHT); | |
640 | ||
e1f00a69 WL |
641 | if (tx_evtchn == rx_evtchn) { |
642 | /* feature-split-event-channels == 0 */ | |
643 | err = bind_interdomain_evtchn_to_irqhandler( | |
e9ce7cb6 WL |
644 | queue->vif->domid, tx_evtchn, xenvif_interrupt, 0, |
645 | queue->name, queue); | |
e1f00a69 WL |
646 | if (err < 0) |
647 | goto err_unmap; | |
e9ce7cb6 WL |
648 | queue->tx_irq = queue->rx_irq = err; |
649 | disable_irq(queue->tx_irq); | |
e1f00a69 WL |
650 | } else { |
651 | /* feature-split-event-channels == 1 */ | |
e9ce7cb6 WL |
652 | snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name), |
653 | "%s-tx", queue->name); | |
e1f00a69 | 654 | err = bind_interdomain_evtchn_to_irqhandler( |
e9ce7cb6 WL |
655 | queue->vif->domid, tx_evtchn, xenvif_tx_interrupt, 0, |
656 | queue->tx_irq_name, queue); | |
e1f00a69 WL |
657 | if (err < 0) |
658 | goto err_unmap; | |
e9ce7cb6 WL |
659 | queue->tx_irq = err; |
660 | disable_irq(queue->tx_irq); | |
e1f00a69 | 661 | |
e9ce7cb6 WL |
662 | snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name), |
663 | "%s-rx", queue->name); | |
e1f00a69 | 664 | err = bind_interdomain_evtchn_to_irqhandler( |
e9ce7cb6 WL |
665 | queue->vif->domid, rx_evtchn, xenvif_rx_interrupt, 0, |
666 | queue->rx_irq_name, queue); | |
e1f00a69 WL |
667 | if (err < 0) |
668 | goto err_tx_unbind; | |
e9ce7cb6 WL |
669 | queue->rx_irq = err; |
670 | disable_irq(queue->rx_irq); | |
e1f00a69 | 671 | } |
f942dc25 | 672 | |
ecf08d2d DV |
673 | queue->stalled = true; |
674 | ||
121fa4b7 | 675 | task = kthread_create(xenvif_kthread_guest_rx, |
e9ce7cb6 | 676 | (void *)queue, "%s-guest-rx", queue->name); |
67fa3660 | 677 | if (IS_ERR(task)) { |
e9ce7cb6 | 678 | pr_warn("Could not allocate kthread for %s\n", queue->name); |
67fa3660 | 679 | err = PTR_ERR(task); |
b3f980bd WL |
680 | goto err_rx_unbind; |
681 | } | |
e9ce7cb6 | 682 | queue->task = task; |
42b5212f | 683 | get_task_struct(task); |
67fa3660 | 684 | |
f53c3fe8 | 685 | task = kthread_create(xenvif_dealloc_kthread, |
e9ce7cb6 | 686 | (void *)queue, "%s-dealloc", queue->name); |
f53c3fe8 | 687 | if (IS_ERR(task)) { |
e9ce7cb6 | 688 | pr_warn("Could not allocate kthread for %s\n", queue->name); |
f53c3fe8 ZK |
689 | err = PTR_ERR(task); |
690 | goto err_rx_unbind; | |
691 | } | |
e9ce7cb6 | 692 | queue->dealloc_task = task; |
f53c3fe8 | 693 | |
e9ce7cb6 WL |
694 | wake_up_process(queue->task); |
695 | wake_up_process(queue->dealloc_task); | |
b3f980bd | 696 | |
f942dc25 | 697 | return 0; |
b3f980bd WL |
698 | |
699 | err_rx_unbind: | |
e9ce7cb6 WL |
700 | unbind_from_irqhandler(queue->rx_irq, queue); |
701 | queue->rx_irq = 0; | |
e1f00a69 | 702 | err_tx_unbind: |
e9ce7cb6 WL |
703 | unbind_from_irqhandler(queue->tx_irq, queue); |
704 | queue->tx_irq = 0; | |
f942dc25 | 705 | err_unmap: |
4e15ee2c | 706 | xenvif_unmap_frontend_data_rings(queue); |
4a658527 | 707 | netif_napi_del(&queue->napi); |
f942dc25 | 708 | err: |
b103f358 | 709 | module_put(THIS_MODULE); |
f942dc25 IC |
710 | return err; |
711 | } | |
712 | ||
48856286 | 713 | void xenvif_carrier_off(struct xenvif *vif) |
f942dc25 IC |
714 | { |
715 | struct net_device *dev = vif->dev; | |
48856286 IC |
716 | |
717 | rtnl_lock(); | |
3d1af1df ZK |
718 | if (test_and_clear_bit(VIF_STATUS_CONNECTED, &vif->status)) { |
719 | netif_carrier_off(dev); /* discard queued packets */ | |
720 | if (netif_running(dev)) | |
721 | xenvif_down(vif); | |
722 | } | |
48856286 | 723 | rtnl_unlock(); |
48856286 IC |
724 | } |
725 | ||
4e15ee2c | 726 | void xenvif_disconnect_data(struct xenvif *vif) |
48856286 | 727 | { |
e9ce7cb6 | 728 | struct xenvif_queue *queue = NULL; |
f7b50c4e | 729 | unsigned int num_queues = vif->num_queues; |
e9ce7cb6 WL |
730 | unsigned int queue_index; |
731 | ||
3d1af1df | 732 | xenvif_carrier_off(vif); |
f942dc25 | 733 | |
e9ce7cb6 WL |
734 | for (queue_index = 0; queue_index < num_queues; ++queue_index) { |
735 | queue = &vif->queues[queue_index]; | |
db739ef3 | 736 | |
ea2c5e13 WL |
737 | netif_napi_del(&queue->napi); |
738 | ||
e9ce7cb6 | 739 | if (queue->task) { |
e9ce7cb6 | 740 | kthread_stop(queue->task); |
42b5212f | 741 | put_task_struct(queue->task); |
e9ce7cb6 WL |
742 | queue->task = NULL; |
743 | } | |
f53c3fe8 | 744 | |
e9ce7cb6 WL |
745 | if (queue->dealloc_task) { |
746 | kthread_stop(queue->dealloc_task); | |
747 | queue->dealloc_task = NULL; | |
748 | } | |
749 | ||
750 | if (queue->tx_irq) { | |
751 | if (queue->tx_irq == queue->rx_irq) | |
752 | unbind_from_irqhandler(queue->tx_irq, queue); | |
753 | else { | |
754 | unbind_from_irqhandler(queue->tx_irq, queue); | |
755 | unbind_from_irqhandler(queue->rx_irq, queue); | |
756 | } | |
757 | queue->tx_irq = 0; | |
e1f00a69 | 758 | } |
f942dc25 | 759 | |
4e15ee2c | 760 | xenvif_unmap_frontend_data_rings(queue); |
e9ce7cb6 | 761 | } |
210c34dc PD |
762 | |
763 | xenvif_mcast_addr_list_free(vif); | |
279f438e PD |
764 | } |
765 | ||
4e15ee2c PD |
766 | void xenvif_disconnect_ctrl(struct xenvif *vif) |
767 | { | |
768 | if (vif->ctrl_task) { | |
769 | kthread_stop(vif->ctrl_task); | |
770 | put_task_struct(vif->ctrl_task); | |
771 | vif->ctrl_task = NULL; | |
772 | } | |
773 | ||
40d8abde PD |
774 | xenvif_deinit_hash(vif); |
775 | ||
4e15ee2c PD |
776 | if (vif->ctrl_irq) { |
777 | unbind_from_irqhandler(vif->ctrl_irq, vif); | |
778 | vif->ctrl_irq = 0; | |
779 | } | |
780 | ||
781 | if (vif->ctrl.sring) { | |
782 | xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif), | |
783 | vif->ctrl.sring); | |
784 | vif->ctrl.sring = NULL; | |
785 | } | |
786 | } | |
787 | ||
8d3d53b3 AB |
788 | /* Reverse the relevant parts of xenvif_init_queue(). |
789 | * Used for queue teardown from xenvif_free(), and on the | |
790 | * error handling paths in xenbus.c:connect(). | |
791 | */ | |
792 | void xenvif_deinit_queue(struct xenvif_queue *queue) | |
793 | { | |
ff4b156f | 794 | gnttab_free_pages(MAX_PENDING_REQS, queue->mmap_pages); |
8d3d53b3 AB |
795 | } |
796 | ||
279f438e PD |
797 | void xenvif_free(struct xenvif *vif) |
798 | { | |
9c6f3ffe | 799 | struct xenvif_queue *queues = vif->queues; |
f7b50c4e | 800 | unsigned int num_queues = vif->num_queues; |
e9ce7cb6 | 801 | unsigned int queue_index; |
f53c3fe8 | 802 | |
e9ce7cb6 | 803 | unregister_netdev(vif->dev); |
f942dc25 | 804 | free_netdev(vif->dev); |
b103f358 | 805 | |
9c6f3ffe DV |
806 | for (queue_index = 0; queue_index < num_queues; ++queue_index) |
807 | xenvif_deinit_queue(&queues[queue_index]); | |
808 | vfree(queues); | |
809 | ||
279f438e | 810 | module_put(THIS_MODULE); |
f942dc25 | 811 | } |