]>
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 | ||
e9ce7cb6 WL |
131 | int xenvif_queue_stopped(struct xenvif_queue *queue) |
132 | { | |
133 | struct net_device *dev = queue->vif->dev; | |
134 | unsigned int id = queue->id; | |
135 | return netif_tx_queue_stopped(netdev_get_tx_queue(dev, id)); | |
136 | } | |
137 | ||
138 | void xenvif_wake_queue(struct xenvif_queue *queue) | |
139 | { | |
140 | struct net_device *dev = queue->vif->dev; | |
141 | unsigned int id = queue->id; | |
142 | netif_tx_wake_queue(netdev_get_tx_queue(dev, id)); | |
143 | } | |
144 | ||
f942dc25 IC |
145 | static int xenvif_start_xmit(struct sk_buff *skb, struct net_device *dev) |
146 | { | |
147 | struct xenvif *vif = netdev_priv(dev); | |
e9ce7cb6 | 148 | struct xenvif_queue *queue = NULL; |
f7b50c4e | 149 | unsigned int num_queues = vif->num_queues; |
e9ce7cb6 | 150 | u16 index; |
f48da8b1 | 151 | struct xenvif_rx_cb *cb; |
f942dc25 IC |
152 | |
153 | BUG_ON(skb->dev != dev); | |
154 | ||
e9ce7cb6 WL |
155 | /* Drop the packet if queues are not set up */ |
156 | if (num_queues < 1) | |
157 | goto drop; | |
158 | ||
159 | /* Obtain the queue to be used to transmit this packet */ | |
160 | index = skb_get_queue_mapping(skb); | |
161 | if (index >= num_queues) { | |
162 | pr_warn_ratelimited("Invalid queue %hu for packet on interface %s\n.", | |
163 | index, vif->dev->name); | |
164 | index %= num_queues; | |
165 | } | |
166 | queue = &vif->queues[index]; | |
167 | ||
168 | /* Drop the packet if queue is not ready */ | |
169 | if (queue->task == NULL || | |
170 | queue->dealloc_task == NULL || | |
f53c3fe8 | 171 | !xenvif_schedulable(vif)) |
f942dc25 IC |
172 | goto drop; |
173 | ||
210c34dc PD |
174 | if (vif->multicast_control && skb->pkt_type == PACKET_MULTICAST) { |
175 | struct ethhdr *eth = (struct ethhdr *)skb->data; | |
176 | ||
177 | if (!xenvif_mcast_match(vif, eth->h_dest)) | |
178 | goto drop; | |
179 | } | |
180 | ||
f48da8b1 | 181 | cb = XENVIF_RX_CB(skb); |
26c0e102 | 182 | cb->expires = jiffies + vif->drain_timeout; |
f942dc25 | 183 | |
f48da8b1 | 184 | xenvif_rx_queue_tail(queue, skb); |
e9ce7cb6 | 185 | xenvif_kick_thread(queue); |
f942dc25 IC |
186 | |
187 | return NETDEV_TX_OK; | |
188 | ||
189 | drop: | |
190 | vif->dev->stats.tx_dropped++; | |
191 | dev_kfree_skb(skb); | |
192 | return NETDEV_TX_OK; | |
193 | } | |
194 | ||
f942dc25 IC |
195 | static struct net_device_stats *xenvif_get_stats(struct net_device *dev) |
196 | { | |
197 | struct xenvif *vif = netdev_priv(dev); | |
e9ce7cb6 | 198 | struct xenvif_queue *queue = NULL; |
f7b50c4e | 199 | unsigned int num_queues = vif->num_queues; |
e9ce7cb6 WL |
200 | unsigned long rx_bytes = 0; |
201 | unsigned long rx_packets = 0; | |
202 | unsigned long tx_bytes = 0; | |
203 | unsigned long tx_packets = 0; | |
204 | unsigned int index; | |
205 | ||
206 | if (vif->queues == NULL) | |
207 | goto out; | |
208 | ||
209 | /* Aggregate tx and rx stats from each queue */ | |
210 | for (index = 0; index < num_queues; ++index) { | |
211 | queue = &vif->queues[index]; | |
212 | rx_bytes += queue->stats.rx_bytes; | |
213 | rx_packets += queue->stats.rx_packets; | |
214 | tx_bytes += queue->stats.tx_bytes; | |
215 | tx_packets += queue->stats.tx_packets; | |
216 | } | |
217 | ||
218 | out: | |
219 | vif->dev->stats.rx_bytes = rx_bytes; | |
220 | vif->dev->stats.rx_packets = rx_packets; | |
221 | vif->dev->stats.tx_bytes = tx_bytes; | |
222 | vif->dev->stats.tx_packets = tx_packets; | |
223 | ||
f942dc25 IC |
224 | return &vif->dev->stats; |
225 | } | |
226 | ||
227 | static void xenvif_up(struct xenvif *vif) | |
228 | { | |
e9ce7cb6 | 229 | struct xenvif_queue *queue = NULL; |
f7b50c4e | 230 | unsigned int num_queues = vif->num_queues; |
e9ce7cb6 WL |
231 | unsigned int queue_index; |
232 | ||
233 | for (queue_index = 0; queue_index < num_queues; ++queue_index) { | |
234 | queue = &vif->queues[queue_index]; | |
235 | napi_enable(&queue->napi); | |
236 | enable_irq(queue->tx_irq); | |
237 | if (queue->tx_irq != queue->rx_irq) | |
238 | enable_irq(queue->rx_irq); | |
239 | xenvif_napi_schedule_or_enable_events(queue); | |
240 | } | |
f942dc25 IC |
241 | } |
242 | ||
243 | static void xenvif_down(struct xenvif *vif) | |
244 | { | |
e9ce7cb6 | 245 | struct xenvif_queue *queue = NULL; |
f7b50c4e | 246 | unsigned int num_queues = vif->num_queues; |
e9ce7cb6 WL |
247 | unsigned int queue_index; |
248 | ||
249 | for (queue_index = 0; queue_index < num_queues; ++queue_index) { | |
250 | queue = &vif->queues[queue_index]; | |
e9ce7cb6 WL |
251 | disable_irq(queue->tx_irq); |
252 | if (queue->tx_irq != queue->rx_irq) | |
253 | disable_irq(queue->rx_irq); | |
8fe78989 | 254 | napi_disable(&queue->napi); |
e9ce7cb6 WL |
255 | del_timer_sync(&queue->credit_timeout); |
256 | } | |
f942dc25 IC |
257 | } |
258 | ||
259 | static int xenvif_open(struct net_device *dev) | |
260 | { | |
261 | struct xenvif *vif = netdev_priv(dev); | |
3d1af1df | 262 | if (test_bit(VIF_STATUS_CONNECTED, &vif->status)) |
f942dc25 | 263 | xenvif_up(vif); |
e9ce7cb6 | 264 | netif_tx_start_all_queues(dev); |
f942dc25 IC |
265 | return 0; |
266 | } | |
267 | ||
268 | static int xenvif_close(struct net_device *dev) | |
269 | { | |
270 | struct xenvif *vif = netdev_priv(dev); | |
3d1af1df | 271 | if (test_bit(VIF_STATUS_CONNECTED, &vif->status)) |
f942dc25 | 272 | xenvif_down(vif); |
e9ce7cb6 | 273 | netif_tx_stop_all_queues(dev); |
f942dc25 IC |
274 | return 0; |
275 | } | |
276 | ||
277 | static int xenvif_change_mtu(struct net_device *dev, int mtu) | |
278 | { | |
279 | struct xenvif *vif = netdev_priv(dev); | |
280 | int max = vif->can_sg ? 65535 - VLAN_ETH_HLEN : ETH_DATA_LEN; | |
281 | ||
282 | if (mtu > max) | |
283 | return -EINVAL; | |
284 | dev->mtu = mtu; | |
285 | return 0; | |
286 | } | |
287 | ||
c8f44aff MM |
288 | static netdev_features_t xenvif_fix_features(struct net_device *dev, |
289 | netdev_features_t features) | |
f942dc25 IC |
290 | { |
291 | struct xenvif *vif = netdev_priv(dev); | |
f942dc25 | 292 | |
47103041 MM |
293 | if (!vif->can_sg) |
294 | features &= ~NETIF_F_SG; | |
82cada22 | 295 | if (~(vif->gso_mask | vif->gso_prefix_mask) & GSO_BIT(TCPV4)) |
47103041 | 296 | features &= ~NETIF_F_TSO; |
82cada22 PD |
297 | if (~(vif->gso_mask | vif->gso_prefix_mask) & GSO_BIT(TCPV6)) |
298 | features &= ~NETIF_F_TSO6; | |
146c8a77 | 299 | if (!vif->ip_csum) |
47103041 | 300 | features &= ~NETIF_F_IP_CSUM; |
146c8a77 PD |
301 | if (!vif->ipv6_csum) |
302 | features &= ~NETIF_F_IPV6_CSUM; | |
f942dc25 | 303 | |
47103041 | 304 | return features; |
f942dc25 IC |
305 | } |
306 | ||
307 | static const struct xenvif_stat { | |
308 | char name[ETH_GSTRING_LEN]; | |
309 | u16 offset; | |
310 | } xenvif_stats[] = { | |
311 | { | |
312 | "rx_gso_checksum_fixup", | |
e9ce7cb6 | 313 | offsetof(struct xenvif_stats, rx_gso_checksum_fixup) |
f942dc25 | 314 | }, |
1bb332af ZK |
315 | /* If (sent != success + fail), there are probably packets never |
316 | * freed up properly! | |
317 | */ | |
318 | { | |
319 | "tx_zerocopy_sent", | |
e9ce7cb6 | 320 | offsetof(struct xenvif_stats, tx_zerocopy_sent), |
1bb332af ZK |
321 | }, |
322 | { | |
323 | "tx_zerocopy_success", | |
e9ce7cb6 | 324 | offsetof(struct xenvif_stats, tx_zerocopy_success), |
1bb332af ZK |
325 | }, |
326 | { | |
327 | "tx_zerocopy_fail", | |
e9ce7cb6 | 328 | offsetof(struct xenvif_stats, tx_zerocopy_fail) |
1bb332af | 329 | }, |
e3377f36 ZK |
330 | /* Number of packets exceeding MAX_SKB_FRAG slots. You should use |
331 | * a guest with the same MAX_SKB_FRAG | |
332 | */ | |
333 | { | |
334 | "tx_frag_overflow", | |
e9ce7cb6 | 335 | offsetof(struct xenvif_stats, tx_frag_overflow) |
e3377f36 | 336 | }, |
f942dc25 IC |
337 | }; |
338 | ||
339 | static int xenvif_get_sset_count(struct net_device *dev, int string_set) | |
340 | { | |
341 | switch (string_set) { | |
342 | case ETH_SS_STATS: | |
343 | return ARRAY_SIZE(xenvif_stats); | |
344 | default: | |
345 | return -EINVAL; | |
346 | } | |
347 | } | |
348 | ||
349 | static void xenvif_get_ethtool_stats(struct net_device *dev, | |
350 | struct ethtool_stats *stats, u64 * data) | |
351 | { | |
e9ce7cb6 | 352 | struct xenvif *vif = netdev_priv(dev); |
f7b50c4e | 353 | unsigned int num_queues = vif->num_queues; |
f942dc25 | 354 | int i; |
e9ce7cb6 | 355 | unsigned int queue_index; |
e9ce7cb6 WL |
356 | |
357 | for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++) { | |
358 | unsigned long accum = 0; | |
359 | for (queue_index = 0; queue_index < num_queues; ++queue_index) { | |
d63951d7 | 360 | void *vif_stats = &vif->queues[queue_index].stats; |
e9ce7cb6 WL |
361 | accum += *(unsigned long *)(vif_stats + xenvif_stats[i].offset); |
362 | } | |
363 | data[i] = accum; | |
364 | } | |
f942dc25 IC |
365 | } |
366 | ||
367 | static void xenvif_get_strings(struct net_device *dev, u32 stringset, u8 * data) | |
368 | { | |
369 | int i; | |
370 | ||
371 | switch (stringset) { | |
372 | case ETH_SS_STATS: | |
373 | for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++) | |
374 | memcpy(data + i * ETH_GSTRING_LEN, | |
375 | xenvif_stats[i].name, ETH_GSTRING_LEN); | |
376 | break; | |
377 | } | |
378 | } | |
379 | ||
813abbba | 380 | static const struct ethtool_ops xenvif_ethtool_ops = { |
f942dc25 IC |
381 | .get_link = ethtool_op_get_link, |
382 | ||
383 | .get_sset_count = xenvif_get_sset_count, | |
384 | .get_ethtool_stats = xenvif_get_ethtool_stats, | |
385 | .get_strings = xenvif_get_strings, | |
386 | }; | |
387 | ||
813abbba | 388 | static const struct net_device_ops xenvif_netdev_ops = { |
f942dc25 IC |
389 | .ndo_start_xmit = xenvif_start_xmit, |
390 | .ndo_get_stats = xenvif_get_stats, | |
391 | .ndo_open = xenvif_open, | |
392 | .ndo_stop = xenvif_close, | |
393 | .ndo_change_mtu = xenvif_change_mtu, | |
47103041 | 394 | .ndo_fix_features = xenvif_fix_features, |
4a633a60 MW |
395 | .ndo_set_mac_address = eth_mac_addr, |
396 | .ndo_validate_addr = eth_validate_addr, | |
f942dc25 IC |
397 | }; |
398 | ||
399 | struct xenvif *xenvif_alloc(struct device *parent, domid_t domid, | |
400 | unsigned int handle) | |
401 | { | |
402 | int err; | |
403 | struct net_device *dev; | |
404 | struct xenvif *vif; | |
405 | char name[IFNAMSIZ] = {}; | |
406 | ||
407 | snprintf(name, IFNAMSIZ - 1, "vif%u.%u", domid, handle); | |
8d3d53b3 AB |
408 | /* Allocate a netdev with the max. supported number of queues. |
409 | * When the guest selects the desired number, it will be updated | |
f7b50c4e | 410 | * via netif_set_real_num_*_queues(). |
8d3d53b3 | 411 | */ |
c835a677 TG |
412 | dev = alloc_netdev_mq(sizeof(struct xenvif), name, NET_NAME_UNKNOWN, |
413 | ether_setup, xenvif_max_queues); | |
f942dc25 | 414 | if (dev == NULL) { |
b3f980bd | 415 | pr_warn("Could not allocate netdev for %s\n", name); |
f942dc25 IC |
416 | return ERR_PTR(-ENOMEM); |
417 | } | |
418 | ||
419 | SET_NETDEV_DEV(dev, parent); | |
420 | ||
421 | vif = netdev_priv(dev); | |
ac3d5ac2 | 422 | |
f942dc25 IC |
423 | vif->domid = domid; |
424 | vif->handle = handle; | |
f942dc25 | 425 | vif->can_sg = 1; |
146c8a77 | 426 | vif->ip_csum = 1; |
f942dc25 | 427 | vif->dev = dev; |
e9d8b2c2 | 428 | vif->disabled = false; |
26c0e102 DV |
429 | vif->drain_timeout = msecs_to_jiffies(rx_drain_timeout_msecs); |
430 | vif->stall_timeout = msecs_to_jiffies(rx_stall_timeout_msecs); | |
e9d8b2c2 | 431 | |
f7b50c4e | 432 | /* Start out with no queues. */ |
e9ce7cb6 | 433 | vif->queues = NULL; |
f7b50c4e | 434 | vif->num_queues = 0; |
09350788 | 435 | |
ecf08d2d | 436 | spin_lock_init(&vif->lock); |
210c34dc | 437 | INIT_LIST_HEAD(&vif->fe_mcast_addr); |
ecf08d2d | 438 | |
f942dc25 | 439 | dev->netdev_ops = &xenvif_netdev_ops; |
146c8a77 PD |
440 | dev->hw_features = NETIF_F_SG | |
441 | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | | |
82cada22 | 442 | NETIF_F_TSO | NETIF_F_TSO6; |
7365bcfa | 443 | dev->features = dev->hw_features | NETIF_F_RXCSUM; |
7ad24ea4 | 444 | dev->ethtool_ops = &xenvif_ethtool_ops; |
f942dc25 IC |
445 | |
446 | dev->tx_queue_len = XENVIF_QUEUE_LENGTH; | |
447 | ||
448 | /* | |
449 | * Initialise a dummy MAC address. We choose the numerically | |
450 | * largest non-broadcast address to prevent the address getting | |
451 | * stolen by an Ethernet bridge for STP purposes. | |
452 | * (FE:FF:FF:FF:FF:FF) | |
453 | */ | |
3b6ed26d | 454 | eth_broadcast_addr(dev->dev_addr); |
f942dc25 IC |
455 | dev->dev_addr[0] &= ~0x01; |
456 | ||
457 | netif_carrier_off(dev); | |
458 | ||
459 | err = register_netdev(dev); | |
460 | if (err) { | |
461 | netdev_warn(dev, "Could not register device: err=%d\n", err); | |
462 | free_netdev(dev); | |
463 | return ERR_PTR(err); | |
464 | } | |
465 | ||
466 | netdev_dbg(dev, "Successfully created xenvif\n"); | |
279f438e PD |
467 | |
468 | __module_get(THIS_MODULE); | |
469 | ||
f942dc25 IC |
470 | return vif; |
471 | } | |
472 | ||
e9ce7cb6 WL |
473 | int xenvif_init_queue(struct xenvif_queue *queue) |
474 | { | |
475 | int err, i; | |
476 | ||
477 | queue->credit_bytes = queue->remaining_credit = ~0UL; | |
478 | queue->credit_usec = 0UL; | |
479 | init_timer(&queue->credit_timeout); | |
edafc132 | 480 | queue->credit_timeout.function = xenvif_tx_credit_callback; |
e9ce7cb6 WL |
481 | queue->credit_window_start = get_jiffies_64(); |
482 | ||
f48da8b1 DV |
483 | queue->rx_queue_max = XENVIF_RX_QUEUE_BYTES; |
484 | ||
e9ce7cb6 WL |
485 | skb_queue_head_init(&queue->rx_queue); |
486 | skb_queue_head_init(&queue->tx_queue); | |
487 | ||
488 | queue->pending_cons = 0; | |
489 | queue->pending_prod = MAX_PENDING_REQS; | |
490 | for (i = 0; i < MAX_PENDING_REQS; ++i) | |
491 | queue->pending_ring[i] = i; | |
492 | ||
493 | spin_lock_init(&queue->callback_lock); | |
494 | spin_lock_init(&queue->response_lock); | |
495 | ||
496 | /* If ballooning is disabled, this will consume real memory, so you | |
497 | * better enable it. The long term solution would be to use just a | |
498 | * bunch of valid page descriptors, without dependency on ballooning | |
499 | */ | |
ff4b156f DV |
500 | err = gnttab_alloc_pages(MAX_PENDING_REQS, |
501 | queue->mmap_pages); | |
e9ce7cb6 WL |
502 | if (err) { |
503 | netdev_err(queue->vif->dev, "Could not reserve mmap_pages\n"); | |
504 | return -ENOMEM; | |
505 | } | |
506 | ||
507 | for (i = 0; i < MAX_PENDING_REQS; i++) { | |
508 | queue->pending_tx_info[i].callback_struct = (struct ubuf_info) | |
509 | { .callback = xenvif_zerocopy_callback, | |
510 | .ctx = NULL, | |
511 | .desc = i }; | |
512 | queue->grant_tx_handle[i] = NETBACK_INVALID_HANDLE; | |
513 | } | |
514 | ||
e9ce7cb6 WL |
515 | return 0; |
516 | } | |
517 | ||
518 | void xenvif_carrier_on(struct xenvif *vif) | |
519 | { | |
520 | rtnl_lock(); | |
521 | if (!vif->can_sg && vif->dev->mtu > ETH_DATA_LEN) | |
522 | dev_set_mtu(vif->dev, ETH_DATA_LEN); | |
523 | netdev_update_features(vif->dev); | |
3d1af1df | 524 | set_bit(VIF_STATUS_CONNECTED, &vif->status); |
e9ce7cb6 WL |
525 | if (netif_running(vif->dev)) |
526 | xenvif_up(vif); | |
527 | rtnl_unlock(); | |
528 | } | |
529 | ||
530 | int xenvif_connect(struct xenvif_queue *queue, unsigned long tx_ring_ref, | |
e1f00a69 WL |
531 | unsigned long rx_ring_ref, unsigned int tx_evtchn, |
532 | unsigned int rx_evtchn) | |
f942dc25 | 533 | { |
67fa3660 | 534 | struct task_struct *task; |
f942dc25 IC |
535 | int err = -ENOMEM; |
536 | ||
e9ce7cb6 WL |
537 | BUG_ON(queue->tx_irq); |
538 | BUG_ON(queue->task); | |
539 | BUG_ON(queue->dealloc_task); | |
f942dc25 | 540 | |
e9ce7cb6 | 541 | err = xenvif_map_frontend_rings(queue, tx_ring_ref, rx_ring_ref); |
f942dc25 IC |
542 | if (err < 0) |
543 | goto err; | |
544 | ||
e9ce7cb6 WL |
545 | init_waitqueue_head(&queue->wq); |
546 | init_waitqueue_head(&queue->dealloc_wq); | |
a64bd934 | 547 | atomic_set(&queue->inflight_packets, 0); |
ca2f09f2 | 548 | |
e24f8191 WL |
549 | netif_napi_add(queue->vif->dev, &queue->napi, xenvif_poll, |
550 | XENVIF_NAPI_WEIGHT); | |
551 | ||
e1f00a69 WL |
552 | if (tx_evtchn == rx_evtchn) { |
553 | /* feature-split-event-channels == 0 */ | |
554 | err = bind_interdomain_evtchn_to_irqhandler( | |
e9ce7cb6 WL |
555 | queue->vif->domid, tx_evtchn, xenvif_interrupt, 0, |
556 | queue->name, queue); | |
e1f00a69 WL |
557 | if (err < 0) |
558 | goto err_unmap; | |
e9ce7cb6 WL |
559 | queue->tx_irq = queue->rx_irq = err; |
560 | disable_irq(queue->tx_irq); | |
e1f00a69 WL |
561 | } else { |
562 | /* feature-split-event-channels == 1 */ | |
e9ce7cb6 WL |
563 | snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name), |
564 | "%s-tx", queue->name); | |
e1f00a69 | 565 | err = bind_interdomain_evtchn_to_irqhandler( |
e9ce7cb6 WL |
566 | queue->vif->domid, tx_evtchn, xenvif_tx_interrupt, 0, |
567 | queue->tx_irq_name, queue); | |
e1f00a69 WL |
568 | if (err < 0) |
569 | goto err_unmap; | |
e9ce7cb6 WL |
570 | queue->tx_irq = err; |
571 | disable_irq(queue->tx_irq); | |
e1f00a69 | 572 | |
e9ce7cb6 WL |
573 | snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name), |
574 | "%s-rx", queue->name); | |
e1f00a69 | 575 | err = bind_interdomain_evtchn_to_irqhandler( |
e9ce7cb6 WL |
576 | queue->vif->domid, rx_evtchn, xenvif_rx_interrupt, 0, |
577 | queue->rx_irq_name, queue); | |
e1f00a69 WL |
578 | if (err < 0) |
579 | goto err_tx_unbind; | |
e9ce7cb6 WL |
580 | queue->rx_irq = err; |
581 | disable_irq(queue->rx_irq); | |
e1f00a69 | 582 | } |
f942dc25 | 583 | |
ecf08d2d DV |
584 | queue->stalled = true; |
585 | ||
121fa4b7 | 586 | task = kthread_create(xenvif_kthread_guest_rx, |
e9ce7cb6 | 587 | (void *)queue, "%s-guest-rx", queue->name); |
67fa3660 | 588 | if (IS_ERR(task)) { |
e9ce7cb6 | 589 | pr_warn("Could not allocate kthread for %s\n", queue->name); |
67fa3660 | 590 | err = PTR_ERR(task); |
b3f980bd WL |
591 | goto err_rx_unbind; |
592 | } | |
e9ce7cb6 | 593 | queue->task = task; |
42b5212f | 594 | get_task_struct(task); |
67fa3660 | 595 | |
f53c3fe8 | 596 | task = kthread_create(xenvif_dealloc_kthread, |
e9ce7cb6 | 597 | (void *)queue, "%s-dealloc", queue->name); |
f53c3fe8 | 598 | if (IS_ERR(task)) { |
e9ce7cb6 | 599 | pr_warn("Could not allocate kthread for %s\n", queue->name); |
f53c3fe8 ZK |
600 | err = PTR_ERR(task); |
601 | goto err_rx_unbind; | |
602 | } | |
e9ce7cb6 | 603 | queue->dealloc_task = task; |
f53c3fe8 | 604 | |
e9ce7cb6 WL |
605 | wake_up_process(queue->task); |
606 | wake_up_process(queue->dealloc_task); | |
b3f980bd | 607 | |
f942dc25 | 608 | return 0; |
b3f980bd WL |
609 | |
610 | err_rx_unbind: | |
e9ce7cb6 WL |
611 | unbind_from_irqhandler(queue->rx_irq, queue); |
612 | queue->rx_irq = 0; | |
e1f00a69 | 613 | err_tx_unbind: |
e9ce7cb6 WL |
614 | unbind_from_irqhandler(queue->tx_irq, queue); |
615 | queue->tx_irq = 0; | |
f942dc25 | 616 | err_unmap: |
e9ce7cb6 | 617 | xenvif_unmap_frontend_rings(queue); |
4a658527 | 618 | netif_napi_del(&queue->napi); |
f942dc25 | 619 | err: |
b103f358 | 620 | module_put(THIS_MODULE); |
f942dc25 IC |
621 | return err; |
622 | } | |
623 | ||
48856286 | 624 | void xenvif_carrier_off(struct xenvif *vif) |
f942dc25 IC |
625 | { |
626 | struct net_device *dev = vif->dev; | |
48856286 IC |
627 | |
628 | rtnl_lock(); | |
3d1af1df ZK |
629 | if (test_and_clear_bit(VIF_STATUS_CONNECTED, &vif->status)) { |
630 | netif_carrier_off(dev); /* discard queued packets */ | |
631 | if (netif_running(dev)) | |
632 | xenvif_down(vif); | |
633 | } | |
48856286 | 634 | rtnl_unlock(); |
48856286 IC |
635 | } |
636 | ||
637 | void xenvif_disconnect(struct xenvif *vif) | |
638 | { | |
e9ce7cb6 | 639 | struct xenvif_queue *queue = NULL; |
f7b50c4e | 640 | unsigned int num_queues = vif->num_queues; |
e9ce7cb6 WL |
641 | unsigned int queue_index; |
642 | ||
3d1af1df | 643 | xenvif_carrier_off(vif); |
f942dc25 | 644 | |
e9ce7cb6 WL |
645 | for (queue_index = 0; queue_index < num_queues; ++queue_index) { |
646 | queue = &vif->queues[queue_index]; | |
db739ef3 | 647 | |
ea2c5e13 WL |
648 | netif_napi_del(&queue->napi); |
649 | ||
e9ce7cb6 | 650 | if (queue->task) { |
e9ce7cb6 | 651 | kthread_stop(queue->task); |
42b5212f | 652 | put_task_struct(queue->task); |
e9ce7cb6 WL |
653 | queue->task = NULL; |
654 | } | |
f53c3fe8 | 655 | |
e9ce7cb6 WL |
656 | if (queue->dealloc_task) { |
657 | kthread_stop(queue->dealloc_task); | |
658 | queue->dealloc_task = NULL; | |
659 | } | |
660 | ||
661 | if (queue->tx_irq) { | |
662 | if (queue->tx_irq == queue->rx_irq) | |
663 | unbind_from_irqhandler(queue->tx_irq, queue); | |
664 | else { | |
665 | unbind_from_irqhandler(queue->tx_irq, queue); | |
666 | unbind_from_irqhandler(queue->rx_irq, queue); | |
667 | } | |
668 | queue->tx_irq = 0; | |
e1f00a69 | 669 | } |
f942dc25 | 670 | |
e9ce7cb6 WL |
671 | xenvif_unmap_frontend_rings(queue); |
672 | } | |
210c34dc PD |
673 | |
674 | xenvif_mcast_addr_list_free(vif); | |
279f438e PD |
675 | } |
676 | ||
8d3d53b3 AB |
677 | /* Reverse the relevant parts of xenvif_init_queue(). |
678 | * Used for queue teardown from xenvif_free(), and on the | |
679 | * error handling paths in xenbus.c:connect(). | |
680 | */ | |
681 | void xenvif_deinit_queue(struct xenvif_queue *queue) | |
682 | { | |
ff4b156f | 683 | gnttab_free_pages(MAX_PENDING_REQS, queue->mmap_pages); |
8d3d53b3 AB |
684 | } |
685 | ||
279f438e PD |
686 | void xenvif_free(struct xenvif *vif) |
687 | { | |
9c6f3ffe | 688 | struct xenvif_queue *queues = vif->queues; |
f7b50c4e | 689 | unsigned int num_queues = vif->num_queues; |
e9ce7cb6 | 690 | unsigned int queue_index; |
f53c3fe8 | 691 | |
e9ce7cb6 | 692 | unregister_netdev(vif->dev); |
f942dc25 | 693 | free_netdev(vif->dev); |
b103f358 | 694 | |
9c6f3ffe DV |
695 | for (queue_index = 0; queue_index < num_queues; ++queue_index) |
696 | xenvif_deinit_queue(&queues[queue_index]); | |
697 | vfree(queues); | |
698 | ||
279f438e | 699 | module_put(THIS_MODULE); |
f942dc25 | 700 | } |