]>
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 | |
e9ce7cb6 WL |
46 | static inline void xenvif_stop_queue(struct xenvif_queue *queue) |
47 | { | |
48 | struct net_device *dev = queue->vif->dev; | |
49 | ||
50 | if (!queue->vif->can_queue) | |
51 | return; | |
52 | ||
53 | netif_tx_stop_queue(netdev_get_tx_queue(dev, queue->id)); | |
54 | } | |
55 | ||
f942dc25 IC |
56 | int xenvif_schedulable(struct xenvif *vif) |
57 | { | |
3d1af1df ZK |
58 | return netif_running(vif->dev) && |
59 | test_bit(VIF_STATUS_CONNECTED, &vif->status); | |
f942dc25 IC |
60 | } |
61 | ||
e1f00a69 | 62 | static irqreturn_t xenvif_tx_interrupt(int irq, void *dev_id) |
f942dc25 | 63 | { |
e9ce7cb6 | 64 | struct xenvif_queue *queue = dev_id; |
f942dc25 | 65 | |
e9ce7cb6 WL |
66 | if (RING_HAS_UNCONSUMED_REQUESTS(&queue->tx)) |
67 | napi_schedule(&queue->napi); | |
f942dc25 | 68 | |
e1f00a69 WL |
69 | return IRQ_HANDLED; |
70 | } | |
71 | ||
e9ce7cb6 | 72 | int xenvif_poll(struct napi_struct *napi, int budget) |
b3f980bd | 73 | { |
e9ce7cb6 WL |
74 | struct xenvif_queue *queue = |
75 | container_of(napi, struct xenvif_queue, napi); | |
b3f980bd WL |
76 | int work_done; |
77 | ||
e9d8b2c2 WL |
78 | /* This vif is rogue, we pretend we've there is nothing to do |
79 | * for this vif to deschedule it from NAPI. But this interface | |
80 | * will be turned off in thread context later. | |
81 | */ | |
2561cc15 | 82 | if (unlikely(queue->vif->disabled)) { |
e9d8b2c2 WL |
83 | napi_complete(napi); |
84 | return 0; | |
85 | } | |
86 | ||
e9ce7cb6 | 87 | work_done = xenvif_tx_action(queue, budget); |
b3f980bd WL |
88 | |
89 | if (work_done < budget) { | |
0d08fceb | 90 | napi_complete(napi); |
e9ce7cb6 | 91 | xenvif_napi_schedule_or_enable_events(queue); |
b3f980bd WL |
92 | } |
93 | ||
94 | return work_done; | |
95 | } | |
96 | ||
e1f00a69 WL |
97 | static irqreturn_t xenvif_rx_interrupt(int irq, void *dev_id) |
98 | { | |
e9ce7cb6 | 99 | struct xenvif_queue *queue = dev_id; |
f34a4cf9 ZK |
100 | struct netdev_queue *net_queue = |
101 | netdev_get_tx_queue(queue->vif->dev, queue->id); | |
e1f00a69 | 102 | |
f34a4cf9 ZK |
103 | /* QUEUE_STATUS_RX_PURGE_EVENT is only set if either QDisc was off OR |
104 | * the carrier went down and this queue was previously blocked | |
105 | */ | |
106 | if (unlikely(netif_tx_queue_stopped(net_queue) || | |
107 | (!netif_carrier_ok(queue->vif->dev) && | |
108 | test_bit(QUEUE_STATUS_RX_STALLED, &queue->status)))) | |
109 | set_bit(QUEUE_STATUS_RX_PURGE_EVENT, &queue->status); | |
e9ce7cb6 | 110 | xenvif_kick_thread(queue); |
f942dc25 IC |
111 | |
112 | return IRQ_HANDLED; | |
113 | } | |
114 | ||
f51de243 | 115 | irqreturn_t xenvif_interrupt(int irq, void *dev_id) |
e1f00a69 WL |
116 | { |
117 | xenvif_tx_interrupt(irq, dev_id); | |
118 | xenvif_rx_interrupt(irq, dev_id); | |
119 | ||
120 | return IRQ_HANDLED; | |
121 | } | |
122 | ||
e9ce7cb6 WL |
123 | int xenvif_queue_stopped(struct xenvif_queue *queue) |
124 | { | |
125 | struct net_device *dev = queue->vif->dev; | |
126 | unsigned int id = queue->id; | |
127 | return netif_tx_queue_stopped(netdev_get_tx_queue(dev, id)); | |
128 | } | |
129 | ||
130 | void xenvif_wake_queue(struct xenvif_queue *queue) | |
131 | { | |
132 | struct net_device *dev = queue->vif->dev; | |
133 | unsigned int id = queue->id; | |
134 | netif_tx_wake_queue(netdev_get_tx_queue(dev, id)); | |
135 | } | |
136 | ||
f34a4cf9 ZK |
137 | /* Callback to wake the queue's thread and turn the carrier off on timeout */ |
138 | static void xenvif_rx_stalled(unsigned long data) | |
09350788 | 139 | { |
e9ce7cb6 WL |
140 | struct xenvif_queue *queue = (struct xenvif_queue *)data; |
141 | ||
142 | if (xenvif_queue_stopped(queue)) { | |
f34a4cf9 | 143 | set_bit(QUEUE_STATUS_RX_PURGE_EVENT, &queue->status); |
e9ce7cb6 | 144 | xenvif_kick_thread(queue); |
e9ce7cb6 WL |
145 | } |
146 | } | |
09350788 | 147 | |
f942dc25 IC |
148 | static int xenvif_start_xmit(struct sk_buff *skb, struct net_device *dev) |
149 | { | |
150 | struct xenvif *vif = netdev_priv(dev); | |
e9ce7cb6 | 151 | struct xenvif_queue *queue = NULL; |
f7b50c4e | 152 | unsigned int num_queues = vif->num_queues; |
e9ce7cb6 | 153 | u16 index; |
ca2f09f2 | 154 | int min_slots_needed; |
f942dc25 IC |
155 | |
156 | BUG_ON(skb->dev != dev); | |
157 | ||
e9ce7cb6 WL |
158 | /* Drop the packet if queues are not set up */ |
159 | if (num_queues < 1) | |
160 | goto drop; | |
161 | ||
162 | /* Obtain the queue to be used to transmit this packet */ | |
163 | index = skb_get_queue_mapping(skb); | |
164 | if (index >= num_queues) { | |
165 | pr_warn_ratelimited("Invalid queue %hu for packet on interface %s\n.", | |
166 | index, vif->dev->name); | |
167 | index %= num_queues; | |
168 | } | |
169 | queue = &vif->queues[index]; | |
170 | ||
171 | /* Drop the packet if queue is not ready */ | |
172 | if (queue->task == NULL || | |
173 | queue->dealloc_task == NULL || | |
f53c3fe8 | 174 | !xenvif_schedulable(vif)) |
f942dc25 IC |
175 | goto drop; |
176 | ||
ca2f09f2 PD |
177 | /* At best we'll need one slot for the header and one for each |
178 | * frag. | |
179 | */ | |
180 | min_slots_needed = 1 + skb_shinfo(skb)->nr_frags; | |
f942dc25 | 181 | |
ca2f09f2 PD |
182 | /* If the skb is GSO then we'll also need an extra slot for the |
183 | * metadata. | |
184 | */ | |
836fbaf4 | 185 | if (skb_is_gso(skb)) |
ca2f09f2 | 186 | min_slots_needed++; |
f942dc25 | 187 | |
ca2f09f2 PD |
188 | /* If the skb can't possibly fit in the remaining slots |
189 | * then turn off the queue to give the ring a chance to | |
190 | * drain. | |
191 | */ | |
e9ce7cb6 | 192 | if (!xenvif_rx_ring_slots_available(queue, min_slots_needed)) { |
f34a4cf9 ZK |
193 | queue->rx_stalled.function = xenvif_rx_stalled; |
194 | queue->rx_stalled.data = (unsigned long)queue; | |
e9ce7cb6 | 195 | xenvif_stop_queue(queue); |
f34a4cf9 ZK |
196 | mod_timer(&queue->rx_stalled, |
197 | jiffies + rx_drain_timeout_jiffies); | |
09350788 | 198 | } |
f942dc25 | 199 | |
e9ce7cb6 WL |
200 | skb_queue_tail(&queue->rx_queue, skb); |
201 | xenvif_kick_thread(queue); | |
f942dc25 IC |
202 | |
203 | return NETDEV_TX_OK; | |
204 | ||
205 | drop: | |
206 | vif->dev->stats.tx_dropped++; | |
207 | dev_kfree_skb(skb); | |
208 | return NETDEV_TX_OK; | |
209 | } | |
210 | ||
f942dc25 IC |
211 | static struct net_device_stats *xenvif_get_stats(struct net_device *dev) |
212 | { | |
213 | struct xenvif *vif = netdev_priv(dev); | |
e9ce7cb6 | 214 | struct xenvif_queue *queue = NULL; |
f7b50c4e | 215 | unsigned int num_queues = vif->num_queues; |
e9ce7cb6 WL |
216 | unsigned long rx_bytes = 0; |
217 | unsigned long rx_packets = 0; | |
218 | unsigned long tx_bytes = 0; | |
219 | unsigned long tx_packets = 0; | |
220 | unsigned int index; | |
221 | ||
222 | if (vif->queues == NULL) | |
223 | goto out; | |
224 | ||
225 | /* Aggregate tx and rx stats from each queue */ | |
226 | for (index = 0; index < num_queues; ++index) { | |
227 | queue = &vif->queues[index]; | |
228 | rx_bytes += queue->stats.rx_bytes; | |
229 | rx_packets += queue->stats.rx_packets; | |
230 | tx_bytes += queue->stats.tx_bytes; | |
231 | tx_packets += queue->stats.tx_packets; | |
232 | } | |
233 | ||
234 | out: | |
235 | vif->dev->stats.rx_bytes = rx_bytes; | |
236 | vif->dev->stats.rx_packets = rx_packets; | |
237 | vif->dev->stats.tx_bytes = tx_bytes; | |
238 | vif->dev->stats.tx_packets = tx_packets; | |
239 | ||
f942dc25 IC |
240 | return &vif->dev->stats; |
241 | } | |
242 | ||
243 | static void xenvif_up(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]; | |
251 | napi_enable(&queue->napi); | |
252 | enable_irq(queue->tx_irq); | |
253 | if (queue->tx_irq != queue->rx_irq) | |
254 | enable_irq(queue->rx_irq); | |
255 | xenvif_napi_schedule_or_enable_events(queue); | |
256 | } | |
f942dc25 IC |
257 | } |
258 | ||
259 | static void xenvif_down(struct xenvif *vif) | |
260 | { | |
e9ce7cb6 | 261 | struct xenvif_queue *queue = NULL; |
f7b50c4e | 262 | unsigned int num_queues = vif->num_queues; |
e9ce7cb6 WL |
263 | unsigned int queue_index; |
264 | ||
265 | for (queue_index = 0; queue_index < num_queues; ++queue_index) { | |
266 | queue = &vif->queues[queue_index]; | |
267 | napi_disable(&queue->napi); | |
268 | disable_irq(queue->tx_irq); | |
269 | if (queue->tx_irq != queue->rx_irq) | |
270 | disable_irq(queue->rx_irq); | |
271 | del_timer_sync(&queue->credit_timeout); | |
272 | } | |
f942dc25 IC |
273 | } |
274 | ||
275 | static int xenvif_open(struct net_device *dev) | |
276 | { | |
277 | struct xenvif *vif = netdev_priv(dev); | |
3d1af1df | 278 | if (test_bit(VIF_STATUS_CONNECTED, &vif->status)) |
f942dc25 | 279 | xenvif_up(vif); |
e9ce7cb6 | 280 | netif_tx_start_all_queues(dev); |
f942dc25 IC |
281 | return 0; |
282 | } | |
283 | ||
284 | static int xenvif_close(struct net_device *dev) | |
285 | { | |
286 | struct xenvif *vif = netdev_priv(dev); | |
3d1af1df | 287 | if (test_bit(VIF_STATUS_CONNECTED, &vif->status)) |
f942dc25 | 288 | xenvif_down(vif); |
e9ce7cb6 | 289 | netif_tx_stop_all_queues(dev); |
f942dc25 IC |
290 | return 0; |
291 | } | |
292 | ||
293 | static int xenvif_change_mtu(struct net_device *dev, int mtu) | |
294 | { | |
295 | struct xenvif *vif = netdev_priv(dev); | |
296 | int max = vif->can_sg ? 65535 - VLAN_ETH_HLEN : ETH_DATA_LEN; | |
297 | ||
298 | if (mtu > max) | |
299 | return -EINVAL; | |
300 | dev->mtu = mtu; | |
301 | return 0; | |
302 | } | |
303 | ||
c8f44aff MM |
304 | static netdev_features_t xenvif_fix_features(struct net_device *dev, |
305 | netdev_features_t features) | |
f942dc25 IC |
306 | { |
307 | struct xenvif *vif = netdev_priv(dev); | |
f942dc25 | 308 | |
47103041 MM |
309 | if (!vif->can_sg) |
310 | features &= ~NETIF_F_SG; | |
82cada22 | 311 | if (~(vif->gso_mask | vif->gso_prefix_mask) & GSO_BIT(TCPV4)) |
47103041 | 312 | features &= ~NETIF_F_TSO; |
82cada22 PD |
313 | if (~(vif->gso_mask | vif->gso_prefix_mask) & GSO_BIT(TCPV6)) |
314 | features &= ~NETIF_F_TSO6; | |
146c8a77 | 315 | if (!vif->ip_csum) |
47103041 | 316 | features &= ~NETIF_F_IP_CSUM; |
146c8a77 PD |
317 | if (!vif->ipv6_csum) |
318 | features &= ~NETIF_F_IPV6_CSUM; | |
f942dc25 | 319 | |
47103041 | 320 | return features; |
f942dc25 IC |
321 | } |
322 | ||
323 | static const struct xenvif_stat { | |
324 | char name[ETH_GSTRING_LEN]; | |
325 | u16 offset; | |
326 | } xenvif_stats[] = { | |
327 | { | |
328 | "rx_gso_checksum_fixup", | |
e9ce7cb6 | 329 | offsetof(struct xenvif_stats, rx_gso_checksum_fixup) |
f942dc25 | 330 | }, |
1bb332af ZK |
331 | /* If (sent != success + fail), there are probably packets never |
332 | * freed up properly! | |
333 | */ | |
334 | { | |
335 | "tx_zerocopy_sent", | |
e9ce7cb6 | 336 | offsetof(struct xenvif_stats, tx_zerocopy_sent), |
1bb332af ZK |
337 | }, |
338 | { | |
339 | "tx_zerocopy_success", | |
e9ce7cb6 | 340 | offsetof(struct xenvif_stats, tx_zerocopy_success), |
1bb332af ZK |
341 | }, |
342 | { | |
343 | "tx_zerocopy_fail", | |
e9ce7cb6 | 344 | offsetof(struct xenvif_stats, tx_zerocopy_fail) |
1bb332af | 345 | }, |
e3377f36 ZK |
346 | /* Number of packets exceeding MAX_SKB_FRAG slots. You should use |
347 | * a guest with the same MAX_SKB_FRAG | |
348 | */ | |
349 | { | |
350 | "tx_frag_overflow", | |
e9ce7cb6 | 351 | offsetof(struct xenvif_stats, tx_frag_overflow) |
e3377f36 | 352 | }, |
f942dc25 IC |
353 | }; |
354 | ||
355 | static int xenvif_get_sset_count(struct net_device *dev, int string_set) | |
356 | { | |
357 | switch (string_set) { | |
358 | case ETH_SS_STATS: | |
359 | return ARRAY_SIZE(xenvif_stats); | |
360 | default: | |
361 | return -EINVAL; | |
362 | } | |
363 | } | |
364 | ||
365 | static void xenvif_get_ethtool_stats(struct net_device *dev, | |
366 | struct ethtool_stats *stats, u64 * data) | |
367 | { | |
e9ce7cb6 | 368 | struct xenvif *vif = netdev_priv(dev); |
f7b50c4e | 369 | unsigned int num_queues = vif->num_queues; |
f942dc25 | 370 | int i; |
e9ce7cb6 WL |
371 | unsigned int queue_index; |
372 | struct xenvif_stats *vif_stats; | |
373 | ||
374 | for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++) { | |
375 | unsigned long accum = 0; | |
376 | for (queue_index = 0; queue_index < num_queues; ++queue_index) { | |
377 | vif_stats = &vif->queues[queue_index].stats; | |
378 | accum += *(unsigned long *)(vif_stats + xenvif_stats[i].offset); | |
379 | } | |
380 | data[i] = accum; | |
381 | } | |
f942dc25 IC |
382 | } |
383 | ||
384 | static void xenvif_get_strings(struct net_device *dev, u32 stringset, u8 * data) | |
385 | { | |
386 | int i; | |
387 | ||
388 | switch (stringset) { | |
389 | case ETH_SS_STATS: | |
390 | for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++) | |
391 | memcpy(data + i * ETH_GSTRING_LEN, | |
392 | xenvif_stats[i].name, ETH_GSTRING_LEN); | |
393 | break; | |
394 | } | |
395 | } | |
396 | ||
813abbba | 397 | static const struct ethtool_ops xenvif_ethtool_ops = { |
f942dc25 IC |
398 | .get_link = ethtool_op_get_link, |
399 | ||
400 | .get_sset_count = xenvif_get_sset_count, | |
401 | .get_ethtool_stats = xenvif_get_ethtool_stats, | |
402 | .get_strings = xenvif_get_strings, | |
403 | }; | |
404 | ||
813abbba | 405 | static const struct net_device_ops xenvif_netdev_ops = { |
f942dc25 IC |
406 | .ndo_start_xmit = xenvif_start_xmit, |
407 | .ndo_get_stats = xenvif_get_stats, | |
408 | .ndo_open = xenvif_open, | |
409 | .ndo_stop = xenvif_close, | |
410 | .ndo_change_mtu = xenvif_change_mtu, | |
47103041 | 411 | .ndo_fix_features = xenvif_fix_features, |
4a633a60 MW |
412 | .ndo_set_mac_address = eth_mac_addr, |
413 | .ndo_validate_addr = eth_validate_addr, | |
f942dc25 IC |
414 | }; |
415 | ||
416 | struct xenvif *xenvif_alloc(struct device *parent, domid_t domid, | |
417 | unsigned int handle) | |
418 | { | |
419 | int err; | |
420 | struct net_device *dev; | |
421 | struct xenvif *vif; | |
422 | char name[IFNAMSIZ] = {}; | |
423 | ||
424 | snprintf(name, IFNAMSIZ - 1, "vif%u.%u", domid, handle); | |
8d3d53b3 AB |
425 | /* Allocate a netdev with the max. supported number of queues. |
426 | * When the guest selects the desired number, it will be updated | |
f7b50c4e | 427 | * via netif_set_real_num_*_queues(). |
8d3d53b3 | 428 | */ |
c835a677 TG |
429 | dev = alloc_netdev_mq(sizeof(struct xenvif), name, NET_NAME_UNKNOWN, |
430 | ether_setup, xenvif_max_queues); | |
f942dc25 | 431 | if (dev == NULL) { |
b3f980bd | 432 | pr_warn("Could not allocate netdev for %s\n", name); |
f942dc25 IC |
433 | return ERR_PTR(-ENOMEM); |
434 | } | |
435 | ||
436 | SET_NETDEV_DEV(dev, parent); | |
437 | ||
438 | vif = netdev_priv(dev); | |
ac3d5ac2 | 439 | |
f942dc25 IC |
440 | vif->domid = domid; |
441 | vif->handle = handle; | |
f942dc25 | 442 | vif->can_sg = 1; |
146c8a77 | 443 | vif->ip_csum = 1; |
f942dc25 | 444 | vif->dev = dev; |
e9d8b2c2 WL |
445 | vif->disabled = false; |
446 | ||
f7b50c4e | 447 | /* Start out with no queues. */ |
e9ce7cb6 | 448 | vif->queues = NULL; |
f7b50c4e | 449 | vif->num_queues = 0; |
09350788 | 450 | |
f942dc25 | 451 | dev->netdev_ops = &xenvif_netdev_ops; |
146c8a77 PD |
452 | dev->hw_features = NETIF_F_SG | |
453 | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | | |
82cada22 | 454 | NETIF_F_TSO | NETIF_F_TSO6; |
7365bcfa | 455 | dev->features = dev->hw_features | NETIF_F_RXCSUM; |
7ad24ea4 | 456 | dev->ethtool_ops = &xenvif_ethtool_ops; |
f942dc25 IC |
457 | |
458 | dev->tx_queue_len = XENVIF_QUEUE_LENGTH; | |
459 | ||
460 | /* | |
461 | * Initialise a dummy MAC address. We choose the numerically | |
462 | * largest non-broadcast address to prevent the address getting | |
463 | * stolen by an Ethernet bridge for STP purposes. | |
464 | * (FE:FF:FF:FF:FF:FF) | |
465 | */ | |
466 | memset(dev->dev_addr, 0xFF, ETH_ALEN); | |
467 | dev->dev_addr[0] &= ~0x01; | |
468 | ||
469 | netif_carrier_off(dev); | |
470 | ||
471 | err = register_netdev(dev); | |
472 | if (err) { | |
473 | netdev_warn(dev, "Could not register device: err=%d\n", err); | |
474 | free_netdev(dev); | |
475 | return ERR_PTR(err); | |
476 | } | |
477 | ||
478 | netdev_dbg(dev, "Successfully created xenvif\n"); | |
279f438e PD |
479 | |
480 | __module_get(THIS_MODULE); | |
481 | ||
f942dc25 IC |
482 | return vif; |
483 | } | |
484 | ||
e9ce7cb6 WL |
485 | int xenvif_init_queue(struct xenvif_queue *queue) |
486 | { | |
487 | int err, i; | |
488 | ||
489 | queue->credit_bytes = queue->remaining_credit = ~0UL; | |
490 | queue->credit_usec = 0UL; | |
491 | init_timer(&queue->credit_timeout); | |
492 | queue->credit_window_start = get_jiffies_64(); | |
493 | ||
494 | skb_queue_head_init(&queue->rx_queue); | |
495 | skb_queue_head_init(&queue->tx_queue); | |
496 | ||
497 | queue->pending_cons = 0; | |
498 | queue->pending_prod = MAX_PENDING_REQS; | |
499 | for (i = 0; i < MAX_PENDING_REQS; ++i) | |
500 | queue->pending_ring[i] = i; | |
501 | ||
502 | spin_lock_init(&queue->callback_lock); | |
503 | spin_lock_init(&queue->response_lock); | |
504 | ||
505 | /* If ballooning is disabled, this will consume real memory, so you | |
506 | * better enable it. The long term solution would be to use just a | |
507 | * bunch of valid page descriptors, without dependency on ballooning | |
508 | */ | |
509 | err = alloc_xenballooned_pages(MAX_PENDING_REQS, | |
510 | queue->mmap_pages, | |
511 | false); | |
512 | if (err) { | |
513 | netdev_err(queue->vif->dev, "Could not reserve mmap_pages\n"); | |
514 | return -ENOMEM; | |
515 | } | |
516 | ||
517 | for (i = 0; i < MAX_PENDING_REQS; i++) { | |
518 | queue->pending_tx_info[i].callback_struct = (struct ubuf_info) | |
519 | { .callback = xenvif_zerocopy_callback, | |
520 | .ctx = NULL, | |
521 | .desc = i }; | |
522 | queue->grant_tx_handle[i] = NETBACK_INVALID_HANDLE; | |
523 | } | |
524 | ||
f34a4cf9 | 525 | init_timer(&queue->rx_stalled); |
e9ce7cb6 | 526 | |
e9ce7cb6 WL |
527 | return 0; |
528 | } | |
529 | ||
530 | void xenvif_carrier_on(struct xenvif *vif) | |
531 | { | |
532 | rtnl_lock(); | |
533 | if (!vif->can_sg && vif->dev->mtu > ETH_DATA_LEN) | |
534 | dev_set_mtu(vif->dev, ETH_DATA_LEN); | |
535 | netdev_update_features(vif->dev); | |
3d1af1df | 536 | set_bit(VIF_STATUS_CONNECTED, &vif->status); |
e9ce7cb6 WL |
537 | netif_carrier_on(vif->dev); |
538 | if (netif_running(vif->dev)) | |
539 | xenvif_up(vif); | |
540 | rtnl_unlock(); | |
541 | } | |
542 | ||
543 | int xenvif_connect(struct xenvif_queue *queue, unsigned long tx_ring_ref, | |
e1f00a69 WL |
544 | unsigned long rx_ring_ref, unsigned int tx_evtchn, |
545 | unsigned int rx_evtchn) | |
f942dc25 | 546 | { |
67fa3660 | 547 | struct task_struct *task; |
f942dc25 IC |
548 | int err = -ENOMEM; |
549 | ||
e9ce7cb6 WL |
550 | BUG_ON(queue->tx_irq); |
551 | BUG_ON(queue->task); | |
552 | BUG_ON(queue->dealloc_task); | |
f942dc25 | 553 | |
e9ce7cb6 | 554 | err = xenvif_map_frontend_rings(queue, tx_ring_ref, rx_ring_ref); |
f942dc25 IC |
555 | if (err < 0) |
556 | goto err; | |
557 | ||
e9ce7cb6 WL |
558 | init_waitqueue_head(&queue->wq); |
559 | init_waitqueue_head(&queue->dealloc_wq); | |
ca2f09f2 | 560 | |
e1f00a69 WL |
561 | if (tx_evtchn == rx_evtchn) { |
562 | /* feature-split-event-channels == 0 */ | |
563 | err = bind_interdomain_evtchn_to_irqhandler( | |
e9ce7cb6 WL |
564 | queue->vif->domid, tx_evtchn, xenvif_interrupt, 0, |
565 | queue->name, queue); | |
e1f00a69 WL |
566 | if (err < 0) |
567 | goto err_unmap; | |
e9ce7cb6 WL |
568 | queue->tx_irq = queue->rx_irq = err; |
569 | disable_irq(queue->tx_irq); | |
e1f00a69 WL |
570 | } else { |
571 | /* feature-split-event-channels == 1 */ | |
e9ce7cb6 WL |
572 | snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name), |
573 | "%s-tx", queue->name); | |
e1f00a69 | 574 | err = bind_interdomain_evtchn_to_irqhandler( |
e9ce7cb6 WL |
575 | queue->vif->domid, tx_evtchn, xenvif_tx_interrupt, 0, |
576 | queue->tx_irq_name, queue); | |
e1f00a69 WL |
577 | if (err < 0) |
578 | goto err_unmap; | |
e9ce7cb6 WL |
579 | queue->tx_irq = err; |
580 | disable_irq(queue->tx_irq); | |
e1f00a69 | 581 | |
e9ce7cb6 WL |
582 | snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name), |
583 | "%s-rx", queue->name); | |
e1f00a69 | 584 | err = bind_interdomain_evtchn_to_irqhandler( |
e9ce7cb6 WL |
585 | queue->vif->domid, rx_evtchn, xenvif_rx_interrupt, 0, |
586 | queue->rx_irq_name, queue); | |
e1f00a69 WL |
587 | if (err < 0) |
588 | goto err_tx_unbind; | |
e9ce7cb6 WL |
589 | queue->rx_irq = err; |
590 | disable_irq(queue->rx_irq); | |
e1f00a69 | 591 | } |
f942dc25 | 592 | |
121fa4b7 | 593 | task = kthread_create(xenvif_kthread_guest_rx, |
e9ce7cb6 | 594 | (void *)queue, "%s-guest-rx", queue->name); |
67fa3660 | 595 | if (IS_ERR(task)) { |
e9ce7cb6 | 596 | pr_warn("Could not allocate kthread for %s\n", queue->name); |
67fa3660 | 597 | err = PTR_ERR(task); |
b3f980bd WL |
598 | goto err_rx_unbind; |
599 | } | |
e9ce7cb6 | 600 | queue->task = task; |
67fa3660 | 601 | |
f53c3fe8 | 602 | task = kthread_create(xenvif_dealloc_kthread, |
e9ce7cb6 | 603 | (void *)queue, "%s-dealloc", queue->name); |
f53c3fe8 | 604 | if (IS_ERR(task)) { |
e9ce7cb6 | 605 | pr_warn("Could not allocate kthread for %s\n", queue->name); |
f53c3fe8 ZK |
606 | err = PTR_ERR(task); |
607 | goto err_rx_unbind; | |
608 | } | |
e9ce7cb6 | 609 | queue->dealloc_task = task; |
f53c3fe8 | 610 | |
e9ce7cb6 WL |
611 | wake_up_process(queue->task); |
612 | wake_up_process(queue->dealloc_task); | |
b3f980bd | 613 | |
ea2c5e13 WL |
614 | netif_napi_add(queue->vif->dev, &queue->napi, xenvif_poll, |
615 | XENVIF_NAPI_WEIGHT); | |
616 | ||
f942dc25 | 617 | return 0; |
b3f980bd WL |
618 | |
619 | err_rx_unbind: | |
e9ce7cb6 WL |
620 | unbind_from_irqhandler(queue->rx_irq, queue); |
621 | queue->rx_irq = 0; | |
e1f00a69 | 622 | err_tx_unbind: |
e9ce7cb6 WL |
623 | unbind_from_irqhandler(queue->tx_irq, queue); |
624 | queue->tx_irq = 0; | |
f942dc25 | 625 | err_unmap: |
e9ce7cb6 | 626 | xenvif_unmap_frontend_rings(queue); |
f942dc25 | 627 | err: |
b103f358 | 628 | module_put(THIS_MODULE); |
f942dc25 IC |
629 | return err; |
630 | } | |
631 | ||
48856286 | 632 | void xenvif_carrier_off(struct xenvif *vif) |
f942dc25 IC |
633 | { |
634 | struct net_device *dev = vif->dev; | |
48856286 IC |
635 | |
636 | rtnl_lock(); | |
3d1af1df ZK |
637 | if (test_and_clear_bit(VIF_STATUS_CONNECTED, &vif->status)) { |
638 | netif_carrier_off(dev); /* discard queued packets */ | |
639 | if (netif_running(dev)) | |
640 | xenvif_down(vif); | |
641 | } | |
48856286 | 642 | rtnl_unlock(); |
48856286 IC |
643 | } |
644 | ||
e9ce7cb6 WL |
645 | static void xenvif_wait_unmap_timeout(struct xenvif_queue *queue, |
646 | unsigned int worst_case_skb_lifetime) | |
647 | { | |
648 | int i, unmap_timeout = 0; | |
649 | ||
650 | for (i = 0; i < MAX_PENDING_REQS; ++i) { | |
651 | if (queue->grant_tx_handle[i] != NETBACK_INVALID_HANDLE) { | |
652 | unmap_timeout++; | |
653 | schedule_timeout(msecs_to_jiffies(1000)); | |
654 | if (unmap_timeout > worst_case_skb_lifetime && | |
655 | net_ratelimit()) | |
656 | netdev_err(queue->vif->dev, | |
657 | "Page still granted! Index: %x\n", | |
658 | i); | |
659 | i = -1; | |
660 | } | |
661 | } | |
662 | } | |
663 | ||
48856286 IC |
664 | void xenvif_disconnect(struct xenvif *vif) |
665 | { | |
e9ce7cb6 | 666 | struct xenvif_queue *queue = NULL; |
f7b50c4e | 667 | unsigned int num_queues = vif->num_queues; |
e9ce7cb6 WL |
668 | unsigned int queue_index; |
669 | ||
3d1af1df | 670 | xenvif_carrier_off(vif); |
f942dc25 | 671 | |
e9ce7cb6 WL |
672 | for (queue_index = 0; queue_index < num_queues; ++queue_index) { |
673 | queue = &vif->queues[queue_index]; | |
db739ef3 | 674 | |
ea2c5e13 WL |
675 | netif_napi_del(&queue->napi); |
676 | ||
e9ce7cb6 | 677 | if (queue->task) { |
f34a4cf9 | 678 | del_timer_sync(&queue->rx_stalled); |
e9ce7cb6 WL |
679 | kthread_stop(queue->task); |
680 | queue->task = NULL; | |
681 | } | |
f53c3fe8 | 682 | |
e9ce7cb6 WL |
683 | if (queue->dealloc_task) { |
684 | kthread_stop(queue->dealloc_task); | |
685 | queue->dealloc_task = NULL; | |
686 | } | |
687 | ||
688 | if (queue->tx_irq) { | |
689 | if (queue->tx_irq == queue->rx_irq) | |
690 | unbind_from_irqhandler(queue->tx_irq, queue); | |
691 | else { | |
692 | unbind_from_irqhandler(queue->tx_irq, queue); | |
693 | unbind_from_irqhandler(queue->rx_irq, queue); | |
694 | } | |
695 | queue->tx_irq = 0; | |
e1f00a69 | 696 | } |
f942dc25 | 697 | |
e9ce7cb6 WL |
698 | xenvif_unmap_frontend_rings(queue); |
699 | } | |
279f438e PD |
700 | } |
701 | ||
8d3d53b3 AB |
702 | /* Reverse the relevant parts of xenvif_init_queue(). |
703 | * Used for queue teardown from xenvif_free(), and on the | |
704 | * error handling paths in xenbus.c:connect(). | |
705 | */ | |
706 | void xenvif_deinit_queue(struct xenvif_queue *queue) | |
707 | { | |
708 | free_xenballooned_pages(MAX_PENDING_REQS, queue->mmap_pages); | |
8d3d53b3 AB |
709 | } |
710 | ||
279f438e PD |
711 | void xenvif_free(struct xenvif *vif) |
712 | { | |
e9ce7cb6 | 713 | struct xenvif_queue *queue = NULL; |
f7b50c4e | 714 | unsigned int num_queues = vif->num_queues; |
e9ce7cb6 | 715 | unsigned int queue_index; |
0e59a4a5 ZK |
716 | /* Here we want to avoid timeout messages if an skb can be legitimately |
717 | * stuck somewhere else. Realistically this could be an another vif's | |
09350788 | 718 | * internal or QDisc queue. That another vif also has this |
f34a4cf9 ZK |
719 | * rx_drain_timeout_msecs timeout, so give it time to drain out. |
720 | * Although if that other guest wakes up just before its timeout happens | |
721 | * and takes only one skb from QDisc, it can hold onto other skbs for a | |
722 | * longer period. | |
09350788 | 723 | */ |
f34a4cf9 | 724 | unsigned int worst_case_skb_lifetime = (rx_drain_timeout_msecs/1000); |
f53c3fe8 | 725 | |
e9ce7cb6 | 726 | unregister_netdev(vif->dev); |
f53c3fe8 | 727 | |
e9ce7cb6 WL |
728 | for (queue_index = 0; queue_index < num_queues; ++queue_index) { |
729 | queue = &vif->queues[queue_index]; | |
e9ce7cb6 | 730 | xenvif_wait_unmap_timeout(queue, worst_case_skb_lifetime); |
8d3d53b3 | 731 | xenvif_deinit_queue(queue); |
e9ce7cb6 WL |
732 | } |
733 | ||
e9ce7cb6 WL |
734 | vfree(vif->queues); |
735 | vif->queues = NULL; | |
f7b50c4e | 736 | vif->num_queues = 0; |
f942dc25 | 737 | |
f942dc25 | 738 | free_netdev(vif->dev); |
b103f358 | 739 | |
279f438e | 740 | module_put(THIS_MODULE); |
f942dc25 | 741 | } |