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 | ||
f07f9893 PD |
161 | if (vif->hash.alg == XEN_NETIF_CTRL_HASH_ALGORITHM_NONE) { |
162 | u16 index = fallback(dev, skb) % dev->real_num_tx_queues; | |
163 | ||
164 | /* Make sure there is no hash information in the socket | |
165 | * buffer otherwise it would be incorrectly forwarded | |
166 | * to the frontend. | |
167 | */ | |
168 | skb_clear_hash(skb); | |
169 | ||
170 | return index; | |
171 | } | |
40d8abde PD |
172 | |
173 | xenvif_set_skb_hash(vif, skb); | |
174 | ||
175 | if (size == 0) | |
176 | return skb_get_hash_raw(skb) % dev->real_num_tx_queues; | |
177 | ||
178 | return vif->hash.mapping[skb_get_hash_raw(skb) % size]; | |
179 | } | |
180 | ||
f942dc25 IC |
181 | static int xenvif_start_xmit(struct sk_buff *skb, struct net_device *dev) |
182 | { | |
183 | struct xenvif *vif = netdev_priv(dev); | |
e9ce7cb6 | 184 | struct xenvif_queue *queue = NULL; |
f7b50c4e | 185 | unsigned int num_queues = vif->num_queues; |
e9ce7cb6 | 186 | u16 index; |
f48da8b1 | 187 | struct xenvif_rx_cb *cb; |
f942dc25 IC |
188 | |
189 | BUG_ON(skb->dev != dev); | |
190 | ||
e9ce7cb6 WL |
191 | /* Drop the packet if queues are not set up */ |
192 | if (num_queues < 1) | |
193 | goto drop; | |
194 | ||
195 | /* Obtain the queue to be used to transmit this packet */ | |
196 | index = skb_get_queue_mapping(skb); | |
197 | if (index >= num_queues) { | |
198 | pr_warn_ratelimited("Invalid queue %hu for packet on interface %s\n.", | |
199 | index, vif->dev->name); | |
200 | index %= num_queues; | |
201 | } | |
202 | queue = &vif->queues[index]; | |
203 | ||
204 | /* Drop the packet if queue is not ready */ | |
205 | if (queue->task == NULL || | |
206 | queue->dealloc_task == NULL || | |
f53c3fe8 | 207 | !xenvif_schedulable(vif)) |
f942dc25 IC |
208 | goto drop; |
209 | ||
210c34dc PD |
210 | if (vif->multicast_control && skb->pkt_type == PACKET_MULTICAST) { |
211 | struct ethhdr *eth = (struct ethhdr *)skb->data; | |
212 | ||
213 | if (!xenvif_mcast_match(vif, eth->h_dest)) | |
214 | goto drop; | |
215 | } | |
216 | ||
f48da8b1 | 217 | cb = XENVIF_RX_CB(skb); |
26c0e102 | 218 | cb->expires = jiffies + vif->drain_timeout; |
f942dc25 | 219 | |
f48da8b1 | 220 | xenvif_rx_queue_tail(queue, skb); |
e9ce7cb6 | 221 | xenvif_kick_thread(queue); |
f942dc25 IC |
222 | |
223 | return NETDEV_TX_OK; | |
224 | ||
225 | drop: | |
226 | vif->dev->stats.tx_dropped++; | |
227 | dev_kfree_skb(skb); | |
228 | return NETDEV_TX_OK; | |
229 | } | |
230 | ||
f942dc25 IC |
231 | static struct net_device_stats *xenvif_get_stats(struct net_device *dev) |
232 | { | |
233 | struct xenvif *vif = netdev_priv(dev); | |
e9ce7cb6 | 234 | struct xenvif_queue *queue = NULL; |
f7b50c4e | 235 | unsigned int num_queues = vif->num_queues; |
e9ce7cb6 WL |
236 | unsigned long rx_bytes = 0; |
237 | unsigned long rx_packets = 0; | |
238 | unsigned long tx_bytes = 0; | |
239 | unsigned long tx_packets = 0; | |
240 | unsigned int index; | |
241 | ||
242 | if (vif->queues == NULL) | |
243 | goto out; | |
244 | ||
245 | /* Aggregate tx and rx stats from each queue */ | |
246 | for (index = 0; index < num_queues; ++index) { | |
247 | queue = &vif->queues[index]; | |
248 | rx_bytes += queue->stats.rx_bytes; | |
249 | rx_packets += queue->stats.rx_packets; | |
250 | tx_bytes += queue->stats.tx_bytes; | |
251 | tx_packets += queue->stats.tx_packets; | |
252 | } | |
253 | ||
254 | out: | |
255 | vif->dev->stats.rx_bytes = rx_bytes; | |
256 | vif->dev->stats.rx_packets = rx_packets; | |
257 | vif->dev->stats.tx_bytes = tx_bytes; | |
258 | vif->dev->stats.tx_packets = tx_packets; | |
259 | ||
f942dc25 IC |
260 | return &vif->dev->stats; |
261 | } | |
262 | ||
263 | static void xenvif_up(struct xenvif *vif) | |
264 | { | |
e9ce7cb6 | 265 | struct xenvif_queue *queue = NULL; |
f7b50c4e | 266 | unsigned int num_queues = vif->num_queues; |
e9ce7cb6 WL |
267 | unsigned int queue_index; |
268 | ||
269 | for (queue_index = 0; queue_index < num_queues; ++queue_index) { | |
270 | queue = &vif->queues[queue_index]; | |
271 | napi_enable(&queue->napi); | |
272 | enable_irq(queue->tx_irq); | |
273 | if (queue->tx_irq != queue->rx_irq) | |
274 | enable_irq(queue->rx_irq); | |
275 | xenvif_napi_schedule_or_enable_events(queue); | |
276 | } | |
f942dc25 IC |
277 | } |
278 | ||
279 | static void xenvif_down(struct xenvif *vif) | |
280 | { | |
e9ce7cb6 | 281 | struct xenvif_queue *queue = NULL; |
f7b50c4e | 282 | unsigned int num_queues = vif->num_queues; |
e9ce7cb6 WL |
283 | unsigned int queue_index; |
284 | ||
285 | for (queue_index = 0; queue_index < num_queues; ++queue_index) { | |
286 | queue = &vif->queues[queue_index]; | |
e9ce7cb6 WL |
287 | disable_irq(queue->tx_irq); |
288 | if (queue->tx_irq != queue->rx_irq) | |
289 | disable_irq(queue->rx_irq); | |
8fe78989 | 290 | napi_disable(&queue->napi); |
e9ce7cb6 WL |
291 | del_timer_sync(&queue->credit_timeout); |
292 | } | |
f942dc25 IC |
293 | } |
294 | ||
295 | static int xenvif_open(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_up(vif); |
e9ce7cb6 | 300 | netif_tx_start_all_queues(dev); |
f942dc25 IC |
301 | return 0; |
302 | } | |
303 | ||
304 | static int xenvif_close(struct net_device *dev) | |
305 | { | |
306 | struct xenvif *vif = netdev_priv(dev); | |
3d1af1df | 307 | if (test_bit(VIF_STATUS_CONNECTED, &vif->status)) |
f942dc25 | 308 | xenvif_down(vif); |
e9ce7cb6 | 309 | netif_tx_stop_all_queues(dev); |
f942dc25 IC |
310 | return 0; |
311 | } | |
312 | ||
313 | static int xenvif_change_mtu(struct net_device *dev, int mtu) | |
314 | { | |
315 | struct xenvif *vif = netdev_priv(dev); | |
316 | int max = vif->can_sg ? 65535 - VLAN_ETH_HLEN : ETH_DATA_LEN; | |
317 | ||
318 | if (mtu > max) | |
319 | return -EINVAL; | |
320 | dev->mtu = mtu; | |
321 | return 0; | |
322 | } | |
323 | ||
c8f44aff MM |
324 | static netdev_features_t xenvif_fix_features(struct net_device *dev, |
325 | netdev_features_t features) | |
f942dc25 IC |
326 | { |
327 | struct xenvif *vif = netdev_priv(dev); | |
f942dc25 | 328 | |
47103041 MM |
329 | if (!vif->can_sg) |
330 | features &= ~NETIF_F_SG; | |
82cada22 | 331 | if (~(vif->gso_mask | vif->gso_prefix_mask) & GSO_BIT(TCPV4)) |
47103041 | 332 | features &= ~NETIF_F_TSO; |
82cada22 PD |
333 | if (~(vif->gso_mask | vif->gso_prefix_mask) & GSO_BIT(TCPV6)) |
334 | features &= ~NETIF_F_TSO6; | |
146c8a77 | 335 | if (!vif->ip_csum) |
47103041 | 336 | features &= ~NETIF_F_IP_CSUM; |
146c8a77 PD |
337 | if (!vif->ipv6_csum) |
338 | features &= ~NETIF_F_IPV6_CSUM; | |
f942dc25 | 339 | |
47103041 | 340 | return features; |
f942dc25 IC |
341 | } |
342 | ||
343 | static const struct xenvif_stat { | |
344 | char name[ETH_GSTRING_LEN]; | |
345 | u16 offset; | |
346 | } xenvif_stats[] = { | |
347 | { | |
348 | "rx_gso_checksum_fixup", | |
e9ce7cb6 | 349 | offsetof(struct xenvif_stats, rx_gso_checksum_fixup) |
f942dc25 | 350 | }, |
1bb332af ZK |
351 | /* If (sent != success + fail), there are probably packets never |
352 | * freed up properly! | |
353 | */ | |
354 | { | |
355 | "tx_zerocopy_sent", | |
e9ce7cb6 | 356 | offsetof(struct xenvif_stats, tx_zerocopy_sent), |
1bb332af ZK |
357 | }, |
358 | { | |
359 | "tx_zerocopy_success", | |
e9ce7cb6 | 360 | offsetof(struct xenvif_stats, tx_zerocopy_success), |
1bb332af ZK |
361 | }, |
362 | { | |
363 | "tx_zerocopy_fail", | |
e9ce7cb6 | 364 | offsetof(struct xenvif_stats, tx_zerocopy_fail) |
1bb332af | 365 | }, |
e3377f36 ZK |
366 | /* Number of packets exceeding MAX_SKB_FRAG slots. You should use |
367 | * a guest with the same MAX_SKB_FRAG | |
368 | */ | |
369 | { | |
370 | "tx_frag_overflow", | |
e9ce7cb6 | 371 | offsetof(struct xenvif_stats, tx_frag_overflow) |
e3377f36 | 372 | }, |
f942dc25 IC |
373 | }; |
374 | ||
375 | static int xenvif_get_sset_count(struct net_device *dev, int string_set) | |
376 | { | |
377 | switch (string_set) { | |
378 | case ETH_SS_STATS: | |
379 | return ARRAY_SIZE(xenvif_stats); | |
380 | default: | |
381 | return -EINVAL; | |
382 | } | |
383 | } | |
384 | ||
385 | static void xenvif_get_ethtool_stats(struct net_device *dev, | |
386 | struct ethtool_stats *stats, u64 * data) | |
387 | { | |
e9ce7cb6 | 388 | struct xenvif *vif = netdev_priv(dev); |
f7b50c4e | 389 | unsigned int num_queues = vif->num_queues; |
f942dc25 | 390 | int i; |
e9ce7cb6 | 391 | unsigned int queue_index; |
e9ce7cb6 WL |
392 | |
393 | for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++) { | |
394 | unsigned long accum = 0; | |
395 | for (queue_index = 0; queue_index < num_queues; ++queue_index) { | |
d63951d7 | 396 | void *vif_stats = &vif->queues[queue_index].stats; |
e9ce7cb6 WL |
397 | accum += *(unsigned long *)(vif_stats + xenvif_stats[i].offset); |
398 | } | |
399 | data[i] = accum; | |
400 | } | |
f942dc25 IC |
401 | } |
402 | ||
403 | static void xenvif_get_strings(struct net_device *dev, u32 stringset, u8 * data) | |
404 | { | |
405 | int i; | |
406 | ||
407 | switch (stringset) { | |
408 | case ETH_SS_STATS: | |
409 | for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++) | |
410 | memcpy(data + i * ETH_GSTRING_LEN, | |
411 | xenvif_stats[i].name, ETH_GSTRING_LEN); | |
412 | break; | |
413 | } | |
414 | } | |
415 | ||
813abbba | 416 | static const struct ethtool_ops xenvif_ethtool_ops = { |
f942dc25 IC |
417 | .get_link = ethtool_op_get_link, |
418 | ||
419 | .get_sset_count = xenvif_get_sset_count, | |
420 | .get_ethtool_stats = xenvif_get_ethtool_stats, | |
421 | .get_strings = xenvif_get_strings, | |
422 | }; | |
423 | ||
813abbba | 424 | static const struct net_device_ops xenvif_netdev_ops = { |
40d8abde | 425 | .ndo_select_queue = xenvif_select_queue, |
f942dc25 IC |
426 | .ndo_start_xmit = xenvif_start_xmit, |
427 | .ndo_get_stats = xenvif_get_stats, | |
428 | .ndo_open = xenvif_open, | |
429 | .ndo_stop = xenvif_close, | |
430 | .ndo_change_mtu = xenvif_change_mtu, | |
47103041 | 431 | .ndo_fix_features = xenvif_fix_features, |
4a633a60 MW |
432 | .ndo_set_mac_address = eth_mac_addr, |
433 | .ndo_validate_addr = eth_validate_addr, | |
f942dc25 IC |
434 | }; |
435 | ||
436 | struct xenvif *xenvif_alloc(struct device *parent, domid_t domid, | |
437 | unsigned int handle) | |
438 | { | |
439 | int err; | |
440 | struct net_device *dev; | |
441 | struct xenvif *vif; | |
442 | char name[IFNAMSIZ] = {}; | |
443 | ||
444 | snprintf(name, IFNAMSIZ - 1, "vif%u.%u", domid, handle); | |
8d3d53b3 AB |
445 | /* Allocate a netdev with the max. supported number of queues. |
446 | * When the guest selects the desired number, it will be updated | |
f7b50c4e | 447 | * via netif_set_real_num_*_queues(). |
8d3d53b3 | 448 | */ |
c835a677 TG |
449 | dev = alloc_netdev_mq(sizeof(struct xenvif), name, NET_NAME_UNKNOWN, |
450 | ether_setup, xenvif_max_queues); | |
f942dc25 | 451 | if (dev == NULL) { |
b3f980bd | 452 | pr_warn("Could not allocate netdev for %s\n", name); |
f942dc25 IC |
453 | return ERR_PTR(-ENOMEM); |
454 | } | |
455 | ||
456 | SET_NETDEV_DEV(dev, parent); | |
457 | ||
458 | vif = netdev_priv(dev); | |
ac3d5ac2 | 459 | |
f942dc25 IC |
460 | vif->domid = domid; |
461 | vif->handle = handle; | |
f942dc25 | 462 | vif->can_sg = 1; |
146c8a77 | 463 | vif->ip_csum = 1; |
f942dc25 | 464 | vif->dev = dev; |
e9d8b2c2 | 465 | vif->disabled = false; |
26c0e102 DV |
466 | vif->drain_timeout = msecs_to_jiffies(rx_drain_timeout_msecs); |
467 | vif->stall_timeout = msecs_to_jiffies(rx_stall_timeout_msecs); | |
e9d8b2c2 | 468 | |
f7b50c4e | 469 | /* Start out with no queues. */ |
e9ce7cb6 | 470 | vif->queues = NULL; |
f7b50c4e | 471 | vif->num_queues = 0; |
09350788 | 472 | |
ecf08d2d | 473 | spin_lock_init(&vif->lock); |
210c34dc | 474 | INIT_LIST_HEAD(&vif->fe_mcast_addr); |
ecf08d2d | 475 | |
f942dc25 | 476 | dev->netdev_ops = &xenvif_netdev_ops; |
146c8a77 PD |
477 | dev->hw_features = NETIF_F_SG | |
478 | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | | |
82cada22 | 479 | NETIF_F_TSO | NETIF_F_TSO6; |
7365bcfa | 480 | dev->features = dev->hw_features | NETIF_F_RXCSUM; |
7ad24ea4 | 481 | dev->ethtool_ops = &xenvif_ethtool_ops; |
f942dc25 IC |
482 | |
483 | dev->tx_queue_len = XENVIF_QUEUE_LENGTH; | |
484 | ||
485 | /* | |
486 | * Initialise a dummy MAC address. We choose the numerically | |
487 | * largest non-broadcast address to prevent the address getting | |
488 | * stolen by an Ethernet bridge for STP purposes. | |
489 | * (FE:FF:FF:FF:FF:FF) | |
490 | */ | |
3b6ed26d | 491 | eth_broadcast_addr(dev->dev_addr); |
f942dc25 IC |
492 | dev->dev_addr[0] &= ~0x01; |
493 | ||
494 | netif_carrier_off(dev); | |
495 | ||
496 | err = register_netdev(dev); | |
497 | if (err) { | |
498 | netdev_warn(dev, "Could not register device: err=%d\n", err); | |
499 | free_netdev(dev); | |
500 | return ERR_PTR(err); | |
501 | } | |
502 | ||
503 | netdev_dbg(dev, "Successfully created xenvif\n"); | |
279f438e PD |
504 | |
505 | __module_get(THIS_MODULE); | |
506 | ||
f942dc25 IC |
507 | return vif; |
508 | } | |
509 | ||
e9ce7cb6 WL |
510 | int xenvif_init_queue(struct xenvif_queue *queue) |
511 | { | |
512 | int err, i; | |
513 | ||
514 | queue->credit_bytes = queue->remaining_credit = ~0UL; | |
515 | queue->credit_usec = 0UL; | |
516 | init_timer(&queue->credit_timeout); | |
edafc132 | 517 | queue->credit_timeout.function = xenvif_tx_credit_callback; |
e9ce7cb6 WL |
518 | queue->credit_window_start = get_jiffies_64(); |
519 | ||
f48da8b1 DV |
520 | queue->rx_queue_max = XENVIF_RX_QUEUE_BYTES; |
521 | ||
e9ce7cb6 WL |
522 | skb_queue_head_init(&queue->rx_queue); |
523 | skb_queue_head_init(&queue->tx_queue); | |
524 | ||
525 | queue->pending_cons = 0; | |
526 | queue->pending_prod = MAX_PENDING_REQS; | |
527 | for (i = 0; i < MAX_PENDING_REQS; ++i) | |
528 | queue->pending_ring[i] = i; | |
529 | ||
530 | spin_lock_init(&queue->callback_lock); | |
531 | spin_lock_init(&queue->response_lock); | |
532 | ||
533 | /* If ballooning is disabled, this will consume real memory, so you | |
534 | * better enable it. The long term solution would be to use just a | |
535 | * bunch of valid page descriptors, without dependency on ballooning | |
536 | */ | |
ff4b156f DV |
537 | err = gnttab_alloc_pages(MAX_PENDING_REQS, |
538 | queue->mmap_pages); | |
e9ce7cb6 WL |
539 | if (err) { |
540 | netdev_err(queue->vif->dev, "Could not reserve mmap_pages\n"); | |
541 | return -ENOMEM; | |
542 | } | |
543 | ||
544 | for (i = 0; i < MAX_PENDING_REQS; i++) { | |
545 | queue->pending_tx_info[i].callback_struct = (struct ubuf_info) | |
546 | { .callback = xenvif_zerocopy_callback, | |
547 | .ctx = NULL, | |
548 | .desc = i }; | |
549 | queue->grant_tx_handle[i] = NETBACK_INVALID_HANDLE; | |
550 | } | |
551 | ||
e9ce7cb6 WL |
552 | return 0; |
553 | } | |
554 | ||
555 | void xenvif_carrier_on(struct xenvif *vif) | |
556 | { | |
557 | rtnl_lock(); | |
558 | if (!vif->can_sg && vif->dev->mtu > ETH_DATA_LEN) | |
559 | dev_set_mtu(vif->dev, ETH_DATA_LEN); | |
560 | netdev_update_features(vif->dev); | |
3d1af1df | 561 | set_bit(VIF_STATUS_CONNECTED, &vif->status); |
e9ce7cb6 WL |
562 | if (netif_running(vif->dev)) |
563 | xenvif_up(vif); | |
564 | rtnl_unlock(); | |
565 | } | |
566 | ||
4e15ee2c PD |
567 | int xenvif_connect_ctrl(struct xenvif *vif, grant_ref_t ring_ref, |
568 | unsigned int evtchn) | |
569 | { | |
570 | struct net_device *dev = vif->dev; | |
571 | void *addr; | |
572 | struct xen_netif_ctrl_sring *shared; | |
573 | struct task_struct *task; | |
574 | int err = -ENOMEM; | |
575 | ||
576 | err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(vif), | |
577 | &ring_ref, 1, &addr); | |
578 | if (err) | |
579 | goto err; | |
580 | ||
581 | shared = (struct xen_netif_ctrl_sring *)addr; | |
582 | BACK_RING_INIT(&vif->ctrl, shared, XEN_PAGE_SIZE); | |
583 | ||
584 | init_waitqueue_head(&vif->ctrl_wq); | |
585 | ||
586 | err = bind_interdomain_evtchn_to_irqhandler(vif->domid, evtchn, | |
587 | xenvif_ctrl_interrupt, | |
588 | 0, dev->name, vif); | |
589 | if (err < 0) | |
590 | goto err_unmap; | |
591 | ||
592 | vif->ctrl_irq = err; | |
593 | ||
40d8abde PD |
594 | xenvif_init_hash(vif); |
595 | ||
4e15ee2c PD |
596 | task = kthread_create(xenvif_ctrl_kthread, (void *)vif, |
597 | "%s-control", dev->name); | |
598 | if (IS_ERR(task)) { | |
599 | pr_warn("Could not allocate kthread for %s\n", dev->name); | |
600 | err = PTR_ERR(task); | |
601 | goto err_deinit; | |
602 | } | |
603 | ||
604 | get_task_struct(task); | |
605 | vif->ctrl_task = task; | |
606 | ||
607 | wake_up_process(vif->ctrl_task); | |
608 | ||
609 | return 0; | |
610 | ||
611 | err_deinit: | |
40d8abde | 612 | xenvif_deinit_hash(vif); |
4e15ee2c PD |
613 | unbind_from_irqhandler(vif->ctrl_irq, vif); |
614 | vif->ctrl_irq = 0; | |
615 | ||
616 | err_unmap: | |
617 | xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif), | |
618 | vif->ctrl.sring); | |
619 | vif->ctrl.sring = NULL; | |
620 | ||
621 | err: | |
622 | return err; | |
623 | } | |
624 | ||
625 | int xenvif_connect_data(struct xenvif_queue *queue, | |
626 | unsigned long tx_ring_ref, | |
627 | unsigned long rx_ring_ref, | |
628 | unsigned int tx_evtchn, | |
629 | unsigned int rx_evtchn) | |
f942dc25 | 630 | { |
67fa3660 | 631 | struct task_struct *task; |
f942dc25 IC |
632 | int err = -ENOMEM; |
633 | ||
e9ce7cb6 WL |
634 | BUG_ON(queue->tx_irq); |
635 | BUG_ON(queue->task); | |
636 | BUG_ON(queue->dealloc_task); | |
f942dc25 | 637 | |
4e15ee2c PD |
638 | err = xenvif_map_frontend_data_rings(queue, tx_ring_ref, |
639 | rx_ring_ref); | |
f942dc25 IC |
640 | if (err < 0) |
641 | goto err; | |
642 | ||
e9ce7cb6 WL |
643 | init_waitqueue_head(&queue->wq); |
644 | init_waitqueue_head(&queue->dealloc_wq); | |
a64bd934 | 645 | atomic_set(&queue->inflight_packets, 0); |
ca2f09f2 | 646 | |
e24f8191 WL |
647 | netif_napi_add(queue->vif->dev, &queue->napi, xenvif_poll, |
648 | XENVIF_NAPI_WEIGHT); | |
649 | ||
e1f00a69 WL |
650 | if (tx_evtchn == rx_evtchn) { |
651 | /* feature-split-event-channels == 0 */ | |
652 | err = bind_interdomain_evtchn_to_irqhandler( | |
e9ce7cb6 WL |
653 | queue->vif->domid, tx_evtchn, xenvif_interrupt, 0, |
654 | queue->name, queue); | |
e1f00a69 WL |
655 | if (err < 0) |
656 | goto err_unmap; | |
e9ce7cb6 WL |
657 | queue->tx_irq = queue->rx_irq = err; |
658 | disable_irq(queue->tx_irq); | |
e1f00a69 WL |
659 | } else { |
660 | /* feature-split-event-channels == 1 */ | |
e9ce7cb6 WL |
661 | snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name), |
662 | "%s-tx", queue->name); | |
e1f00a69 | 663 | err = bind_interdomain_evtchn_to_irqhandler( |
e9ce7cb6 WL |
664 | queue->vif->domid, tx_evtchn, xenvif_tx_interrupt, 0, |
665 | queue->tx_irq_name, queue); | |
e1f00a69 WL |
666 | if (err < 0) |
667 | goto err_unmap; | |
e9ce7cb6 WL |
668 | queue->tx_irq = err; |
669 | disable_irq(queue->tx_irq); | |
e1f00a69 | 670 | |
e9ce7cb6 WL |
671 | snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name), |
672 | "%s-rx", queue->name); | |
e1f00a69 | 673 | err = bind_interdomain_evtchn_to_irqhandler( |
e9ce7cb6 WL |
674 | queue->vif->domid, rx_evtchn, xenvif_rx_interrupt, 0, |
675 | queue->rx_irq_name, queue); | |
e1f00a69 WL |
676 | if (err < 0) |
677 | goto err_tx_unbind; | |
e9ce7cb6 WL |
678 | queue->rx_irq = err; |
679 | disable_irq(queue->rx_irq); | |
e1f00a69 | 680 | } |
f942dc25 | 681 | |
ecf08d2d DV |
682 | queue->stalled = true; |
683 | ||
121fa4b7 | 684 | task = kthread_create(xenvif_kthread_guest_rx, |
e9ce7cb6 | 685 | (void *)queue, "%s-guest-rx", queue->name); |
67fa3660 | 686 | if (IS_ERR(task)) { |
e9ce7cb6 | 687 | pr_warn("Could not allocate kthread for %s\n", queue->name); |
67fa3660 | 688 | err = PTR_ERR(task); |
b3f980bd WL |
689 | goto err_rx_unbind; |
690 | } | |
e9ce7cb6 | 691 | queue->task = task; |
42b5212f | 692 | get_task_struct(task); |
67fa3660 | 693 | |
f53c3fe8 | 694 | task = kthread_create(xenvif_dealloc_kthread, |
e9ce7cb6 | 695 | (void *)queue, "%s-dealloc", queue->name); |
f53c3fe8 | 696 | if (IS_ERR(task)) { |
e9ce7cb6 | 697 | pr_warn("Could not allocate kthread for %s\n", queue->name); |
f53c3fe8 ZK |
698 | err = PTR_ERR(task); |
699 | goto err_rx_unbind; | |
700 | } | |
e9ce7cb6 | 701 | queue->dealloc_task = task; |
f53c3fe8 | 702 | |
e9ce7cb6 WL |
703 | wake_up_process(queue->task); |
704 | wake_up_process(queue->dealloc_task); | |
b3f980bd | 705 | |
f942dc25 | 706 | return 0; |
b3f980bd WL |
707 | |
708 | err_rx_unbind: | |
e9ce7cb6 WL |
709 | unbind_from_irqhandler(queue->rx_irq, queue); |
710 | queue->rx_irq = 0; | |
e1f00a69 | 711 | err_tx_unbind: |
e9ce7cb6 WL |
712 | unbind_from_irqhandler(queue->tx_irq, queue); |
713 | queue->tx_irq = 0; | |
f942dc25 | 714 | err_unmap: |
4e15ee2c | 715 | xenvif_unmap_frontend_data_rings(queue); |
4a658527 | 716 | netif_napi_del(&queue->napi); |
f942dc25 | 717 | err: |
b103f358 | 718 | module_put(THIS_MODULE); |
f942dc25 IC |
719 | return err; |
720 | } | |
721 | ||
48856286 | 722 | void xenvif_carrier_off(struct xenvif *vif) |
f942dc25 IC |
723 | { |
724 | struct net_device *dev = vif->dev; | |
48856286 IC |
725 | |
726 | rtnl_lock(); | |
3d1af1df ZK |
727 | if (test_and_clear_bit(VIF_STATUS_CONNECTED, &vif->status)) { |
728 | netif_carrier_off(dev); /* discard queued packets */ | |
729 | if (netif_running(dev)) | |
730 | xenvif_down(vif); | |
731 | } | |
48856286 | 732 | rtnl_unlock(); |
48856286 IC |
733 | } |
734 | ||
4e15ee2c | 735 | void xenvif_disconnect_data(struct xenvif *vif) |
48856286 | 736 | { |
e9ce7cb6 | 737 | struct xenvif_queue *queue = NULL; |
f7b50c4e | 738 | unsigned int num_queues = vif->num_queues; |
e9ce7cb6 WL |
739 | unsigned int queue_index; |
740 | ||
3d1af1df | 741 | xenvif_carrier_off(vif); |
f942dc25 | 742 | |
e9ce7cb6 WL |
743 | for (queue_index = 0; queue_index < num_queues; ++queue_index) { |
744 | queue = &vif->queues[queue_index]; | |
db739ef3 | 745 | |
ea2c5e13 WL |
746 | netif_napi_del(&queue->napi); |
747 | ||
e9ce7cb6 | 748 | if (queue->task) { |
e9ce7cb6 | 749 | kthread_stop(queue->task); |
42b5212f | 750 | put_task_struct(queue->task); |
e9ce7cb6 WL |
751 | queue->task = NULL; |
752 | } | |
f53c3fe8 | 753 | |
e9ce7cb6 WL |
754 | if (queue->dealloc_task) { |
755 | kthread_stop(queue->dealloc_task); | |
756 | queue->dealloc_task = NULL; | |
757 | } | |
758 | ||
759 | if (queue->tx_irq) { | |
760 | if (queue->tx_irq == queue->rx_irq) | |
761 | unbind_from_irqhandler(queue->tx_irq, queue); | |
762 | else { | |
763 | unbind_from_irqhandler(queue->tx_irq, queue); | |
764 | unbind_from_irqhandler(queue->rx_irq, queue); | |
765 | } | |
766 | queue->tx_irq = 0; | |
e1f00a69 | 767 | } |
f942dc25 | 768 | |
4e15ee2c | 769 | xenvif_unmap_frontend_data_rings(queue); |
e9ce7cb6 | 770 | } |
210c34dc PD |
771 | |
772 | xenvif_mcast_addr_list_free(vif); | |
279f438e PD |
773 | } |
774 | ||
4e15ee2c PD |
775 | void xenvif_disconnect_ctrl(struct xenvif *vif) |
776 | { | |
777 | if (vif->ctrl_task) { | |
778 | kthread_stop(vif->ctrl_task); | |
779 | put_task_struct(vif->ctrl_task); | |
780 | vif->ctrl_task = NULL; | |
781 | } | |
782 | ||
783 | if (vif->ctrl_irq) { | |
c0fcded2 | 784 | xenvif_deinit_hash(vif); |
4e15ee2c PD |
785 | unbind_from_irqhandler(vif->ctrl_irq, vif); |
786 | vif->ctrl_irq = 0; | |
787 | } | |
788 | ||
789 | if (vif->ctrl.sring) { | |
790 | xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif), | |
791 | vif->ctrl.sring); | |
792 | vif->ctrl.sring = NULL; | |
793 | } | |
794 | } | |
795 | ||
8d3d53b3 AB |
796 | /* Reverse the relevant parts of xenvif_init_queue(). |
797 | * Used for queue teardown from xenvif_free(), and on the | |
798 | * error handling paths in xenbus.c:connect(). | |
799 | */ | |
800 | void xenvif_deinit_queue(struct xenvif_queue *queue) | |
801 | { | |
ff4b156f | 802 | gnttab_free_pages(MAX_PENDING_REQS, queue->mmap_pages); |
8d3d53b3 AB |
803 | } |
804 | ||
279f438e PD |
805 | void xenvif_free(struct xenvif *vif) |
806 | { | |
9c6f3ffe | 807 | struct xenvif_queue *queues = vif->queues; |
f7b50c4e | 808 | unsigned int num_queues = vif->num_queues; |
e9ce7cb6 | 809 | unsigned int queue_index; |
f53c3fe8 | 810 | |
e9ce7cb6 | 811 | unregister_netdev(vif->dev); |
f942dc25 | 812 | free_netdev(vif->dev); |
b103f358 | 813 | |
9c6f3ffe DV |
814 | for (queue_index = 0; queue_index < num_queues; ++queue_index) |
815 | xenvif_deinit_queue(&queues[queue_index]); | |
816 | vfree(queues); | |
817 | ||
279f438e | 818 | module_put(THIS_MODULE); |
f942dc25 | 819 | } |