]>
Commit | Line | Data |
---|---|---|
f942dc25 IC |
1 | /* |
2 | * Network-device interface management. | |
3 | * | |
4 | * Copyright (c) 2004-2005, Keir Fraser | |
5 | * | |
6 | * This program is free software; you can redistribute it and/or | |
7 | * modify it under the terms of the GNU General Public License version 2 | |
8 | * as published by the Free Software Foundation; or, when distributed | |
9 | * separately from the Linux kernel or incorporated into other | |
10 | * software packages, subject to the following license: | |
11 | * | |
12 | * Permission is hereby granted, free of charge, to any person obtaining a copy | |
13 | * of this source file (the "Software"), to deal in the Software without | |
14 | * restriction, including without limitation the rights to use, copy, modify, | |
15 | * merge, publish, distribute, sublicense, and/or sell copies of the Software, | |
16 | * and to permit persons to whom the Software is furnished to do so, subject to | |
17 | * the following conditions: | |
18 | * | |
19 | * The above copyright notice and this permission notice shall be included in | |
20 | * all copies or substantial portions of the Software. | |
21 | * | |
22 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | |
23 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | |
24 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE | |
25 | * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER | |
26 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING | |
27 | * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS | |
28 | * IN THE SOFTWARE. | |
29 | */ | |
30 | ||
31 | #include "common.h" | |
32 | ||
b3f980bd | 33 | #include <linux/kthread.h> |
0881e7bd | 34 | #include <linux/sched/task.h> |
f942dc25 IC |
35 | #include <linux/ethtool.h> |
36 | #include <linux/rtnetlink.h> | |
37 | #include <linux/if_vlan.h> | |
e7b599d7 | 38 | #include <linux/vmalloc.h> |
f942dc25 IC |
39 | |
40 | #include <xen/events.h> | |
41 | #include <asm/xen/hypercall.h> | |
f53c3fe8 | 42 | #include <xen/balloon.h> |
f942dc25 IC |
43 | |
44 | #define XENVIF_QUEUE_LENGTH 32 | |
b3f980bd | 45 | #define XENVIF_NAPI_WEIGHT 64 |
f942dc25 | 46 | |
f48da8b1 DV |
47 | /* Number of bytes allowed on the internal guest Rx queue. */ |
48 | #define XENVIF_RX_QUEUE_BYTES (XEN_NETIF_RX_RING_SIZE/2 * PAGE_SIZE) | |
49 | ||
a64bd934 WL |
50 | /* This function is used to set SKBTX_DEV_ZEROCOPY as well as |
51 | * increasing the inflight counter. We need to increase the inflight | |
52 | * counter because core driver calls into xenvif_zerocopy_callback | |
53 | * which calls xenvif_skb_zerocopy_complete. | |
54 | */ | |
55 | void xenvif_skb_zerocopy_prepare(struct xenvif_queue *queue, | |
56 | struct sk_buff *skb) | |
57 | { | |
58 | skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY; | |
59 | atomic_inc(&queue->inflight_packets); | |
60 | } | |
61 | ||
62 | void xenvif_skb_zerocopy_complete(struct xenvif_queue *queue) | |
63 | { | |
64 | atomic_dec(&queue->inflight_packets); | |
57b22906 RL |
65 | |
66 | /* Wake the dealloc thread _after_ decrementing inflight_packets so | |
67 | * that if kthread_stop() has already been called, the dealloc thread | |
68 | * does not wait forever with nothing to wake it. | |
69 | */ | |
70 | wake_up(&queue->dealloc_wq); | |
a64bd934 WL |
71 | } |
72 | ||
f942dc25 IC |
73 | int xenvif_schedulable(struct xenvif *vif) |
74 | { | |
3d1af1df | 75 | return netif_running(vif->dev) && |
f48da8b1 DV |
76 | test_bit(VIF_STATUS_CONNECTED, &vif->status) && |
77 | !vif->disabled; | |
f942dc25 IC |
78 | } |
79 | ||
e1f00a69 | 80 | static irqreturn_t xenvif_tx_interrupt(int irq, void *dev_id) |
f942dc25 | 81 | { |
e9ce7cb6 | 82 | struct xenvif_queue *queue = dev_id; |
f942dc25 | 83 | |
e9ce7cb6 WL |
84 | if (RING_HAS_UNCONSUMED_REQUESTS(&queue->tx)) |
85 | napi_schedule(&queue->napi); | |
f942dc25 | 86 | |
e1f00a69 WL |
87 | return IRQ_HANDLED; |
88 | } | |
89 | ||
38741d50 | 90 | static int xenvif_poll(struct napi_struct *napi, int budget) |
b3f980bd | 91 | { |
e9ce7cb6 WL |
92 | struct xenvif_queue *queue = |
93 | container_of(napi, struct xenvif_queue, napi); | |
b3f980bd WL |
94 | int work_done; |
95 | ||
e9d8b2c2 WL |
96 | /* This vif is rogue, we pretend we've there is nothing to do |
97 | * for this vif to deschedule it from NAPI. But this interface | |
98 | * will be turned off in thread context later. | |
99 | */ | |
2561cc15 | 100 | if (unlikely(queue->vif->disabled)) { |
e9d8b2c2 WL |
101 | napi_complete(napi); |
102 | return 0; | |
103 | } | |
104 | ||
e9ce7cb6 | 105 | work_done = xenvif_tx_action(queue, budget); |
b3f980bd WL |
106 | |
107 | if (work_done < budget) { | |
6ad20165 | 108 | napi_complete_done(napi, work_done); |
e9ce7cb6 | 109 | xenvif_napi_schedule_or_enable_events(queue); |
b3f980bd WL |
110 | } |
111 | ||
112 | return work_done; | |
113 | } | |
114 | ||
e1f00a69 WL |
115 | static irqreturn_t xenvif_rx_interrupt(int irq, void *dev_id) |
116 | { | |
e9ce7cb6 | 117 | struct xenvif_queue *queue = dev_id; |
e1f00a69 | 118 | |
e9ce7cb6 | 119 | xenvif_kick_thread(queue); |
f942dc25 IC |
120 | |
121 | return IRQ_HANDLED; | |
122 | } | |
123 | ||
f51de243 | 124 | irqreturn_t xenvif_interrupt(int irq, void *dev_id) |
e1f00a69 WL |
125 | { |
126 | xenvif_tx_interrupt(irq, dev_id); | |
127 | xenvif_rx_interrupt(irq, dev_id); | |
128 | ||
129 | return IRQ_HANDLED; | |
130 | } | |
131 | ||
e9ce7cb6 WL |
132 | int xenvif_queue_stopped(struct xenvif_queue *queue) |
133 | { | |
134 | struct net_device *dev = queue->vif->dev; | |
135 | unsigned int id = queue->id; | |
136 | return netif_tx_queue_stopped(netdev_get_tx_queue(dev, id)); | |
137 | } | |
138 | ||
139 | void xenvif_wake_queue(struct xenvif_queue *queue) | |
140 | { | |
141 | struct net_device *dev = queue->vif->dev; | |
142 | unsigned int id = queue->id; | |
143 | netif_tx_wake_queue(netdev_get_tx_queue(dev, id)); | |
144 | } | |
145 | ||
40d8abde PD |
146 | static u16 xenvif_select_queue(struct net_device *dev, struct sk_buff *skb, |
147 | void *accel_priv, | |
148 | select_queue_fallback_t fallback) | |
149 | { | |
150 | struct xenvif *vif = netdev_priv(dev); | |
151 | unsigned int size = vif->hash.size; | |
152 | ||
912e27e8 PD |
153 | if (vif->hash.alg == XEN_NETIF_CTRL_HASH_ALGORITHM_NONE) |
154 | return fallback(dev, skb) % dev->real_num_tx_queues; | |
40d8abde PD |
155 | |
156 | xenvif_set_skb_hash(vif, skb); | |
157 | ||
158 | if (size == 0) | |
159 | return skb_get_hash_raw(skb) % dev->real_num_tx_queues; | |
160 | ||
161 | return vif->hash.mapping[skb_get_hash_raw(skb) % size]; | |
162 | } | |
163 | ||
f942dc25 IC |
164 | static int xenvif_start_xmit(struct sk_buff *skb, struct net_device *dev) |
165 | { | |
166 | struct xenvif *vif = netdev_priv(dev); | |
e9ce7cb6 | 167 | struct xenvif_queue *queue = NULL; |
f7b50c4e | 168 | unsigned int num_queues = vif->num_queues; |
e9ce7cb6 | 169 | u16 index; |
f48da8b1 | 170 | struct xenvif_rx_cb *cb; |
f942dc25 IC |
171 | |
172 | BUG_ON(skb->dev != dev); | |
173 | ||
e9ce7cb6 WL |
174 | /* Drop the packet if queues are not set up */ |
175 | if (num_queues < 1) | |
176 | goto drop; | |
177 | ||
178 | /* Obtain the queue to be used to transmit this packet */ | |
179 | index = skb_get_queue_mapping(skb); | |
180 | if (index >= num_queues) { | |
181 | pr_warn_ratelimited("Invalid queue %hu for packet on interface %s\n.", | |
182 | index, vif->dev->name); | |
183 | index %= num_queues; | |
184 | } | |
185 | queue = &vif->queues[index]; | |
186 | ||
187 | /* Drop the packet if queue is not ready */ | |
188 | if (queue->task == NULL || | |
189 | queue->dealloc_task == NULL || | |
f53c3fe8 | 190 | !xenvif_schedulable(vif)) |
f942dc25 IC |
191 | goto drop; |
192 | ||
210c34dc PD |
193 | if (vif->multicast_control && skb->pkt_type == PACKET_MULTICAST) { |
194 | struct ethhdr *eth = (struct ethhdr *)skb->data; | |
195 | ||
196 | if (!xenvif_mcast_match(vif, eth->h_dest)) | |
197 | goto drop; | |
198 | } | |
199 | ||
f48da8b1 | 200 | cb = XENVIF_RX_CB(skb); |
26c0e102 | 201 | cb->expires = jiffies + vif->drain_timeout; |
f942dc25 | 202 | |
912e27e8 PD |
203 | /* If there is no hash algorithm configured then make sure there |
204 | * is no hash information in the socket buffer otherwise it | |
205 | * would be incorrectly forwarded to the frontend. | |
206 | */ | |
207 | if (vif->hash.alg == XEN_NETIF_CTRL_HASH_ALGORITHM_NONE) | |
208 | skb_clear_hash(skb); | |
209 | ||
f48da8b1 | 210 | xenvif_rx_queue_tail(queue, skb); |
e9ce7cb6 | 211 | xenvif_kick_thread(queue); |
f942dc25 IC |
212 | |
213 | return NETDEV_TX_OK; | |
214 | ||
215 | drop: | |
216 | vif->dev->stats.tx_dropped++; | |
217 | dev_kfree_skb(skb); | |
218 | return NETDEV_TX_OK; | |
219 | } | |
220 | ||
f942dc25 IC |
221 | static struct net_device_stats *xenvif_get_stats(struct net_device *dev) |
222 | { | |
223 | struct xenvif *vif = netdev_priv(dev); | |
e9ce7cb6 | 224 | struct xenvif_queue *queue = NULL; |
ebf692f8 MS |
225 | u64 rx_bytes = 0; |
226 | u64 rx_packets = 0; | |
227 | u64 tx_bytes = 0; | |
228 | u64 tx_packets = 0; | |
e9ce7cb6 WL |
229 | unsigned int index; |
230 | ||
f16f1df6 | 231 | spin_lock(&vif->lock); |
e9ce7cb6 WL |
232 | if (vif->queues == NULL) |
233 | goto out; | |
234 | ||
235 | /* Aggregate tx and rx stats from each queue */ | |
f16f1df6 | 236 | for (index = 0; index < vif->num_queues; ++index) { |
e9ce7cb6 WL |
237 | queue = &vif->queues[index]; |
238 | rx_bytes += queue->stats.rx_bytes; | |
239 | rx_packets += queue->stats.rx_packets; | |
240 | tx_bytes += queue->stats.tx_bytes; | |
241 | tx_packets += queue->stats.tx_packets; | |
242 | } | |
243 | ||
244 | out: | |
f16f1df6 ID |
245 | spin_unlock(&vif->lock); |
246 | ||
e9ce7cb6 WL |
247 | vif->dev->stats.rx_bytes = rx_bytes; |
248 | vif->dev->stats.rx_packets = rx_packets; | |
249 | vif->dev->stats.tx_bytes = tx_bytes; | |
250 | vif->dev->stats.tx_packets = tx_packets; | |
251 | ||
f942dc25 IC |
252 | return &vif->dev->stats; |
253 | } | |
254 | ||
255 | static void xenvif_up(struct xenvif *vif) | |
256 | { | |
e9ce7cb6 | 257 | struct xenvif_queue *queue = NULL; |
f7b50c4e | 258 | unsigned int num_queues = vif->num_queues; |
e9ce7cb6 WL |
259 | unsigned int queue_index; |
260 | ||
261 | for (queue_index = 0; queue_index < num_queues; ++queue_index) { | |
262 | queue = &vif->queues[queue_index]; | |
263 | napi_enable(&queue->napi); | |
264 | enable_irq(queue->tx_irq); | |
265 | if (queue->tx_irq != queue->rx_irq) | |
266 | enable_irq(queue->rx_irq); | |
267 | xenvif_napi_schedule_or_enable_events(queue); | |
268 | } | |
f942dc25 IC |
269 | } |
270 | ||
271 | static void xenvif_down(struct xenvif *vif) | |
272 | { | |
e9ce7cb6 | 273 | struct xenvif_queue *queue = NULL; |
f7b50c4e | 274 | unsigned int num_queues = vif->num_queues; |
e9ce7cb6 WL |
275 | unsigned int queue_index; |
276 | ||
277 | for (queue_index = 0; queue_index < num_queues; ++queue_index) { | |
278 | queue = &vif->queues[queue_index]; | |
e9ce7cb6 WL |
279 | disable_irq(queue->tx_irq); |
280 | if (queue->tx_irq != queue->rx_irq) | |
281 | disable_irq(queue->rx_irq); | |
8fe78989 | 282 | napi_disable(&queue->napi); |
e9ce7cb6 WL |
283 | del_timer_sync(&queue->credit_timeout); |
284 | } | |
f942dc25 IC |
285 | } |
286 | ||
287 | static int xenvif_open(struct net_device *dev) | |
288 | { | |
289 | struct xenvif *vif = netdev_priv(dev); | |
3d1af1df | 290 | if (test_bit(VIF_STATUS_CONNECTED, &vif->status)) |
f942dc25 | 291 | xenvif_up(vif); |
e9ce7cb6 | 292 | netif_tx_start_all_queues(dev); |
f942dc25 IC |
293 | return 0; |
294 | } | |
295 | ||
296 | static int xenvif_close(struct net_device *dev) | |
297 | { | |
298 | struct xenvif *vif = netdev_priv(dev); | |
3d1af1df | 299 | if (test_bit(VIF_STATUS_CONNECTED, &vif->status)) |
f942dc25 | 300 | xenvif_down(vif); |
e9ce7cb6 | 301 | netif_tx_stop_all_queues(dev); |
f942dc25 IC |
302 | return 0; |
303 | } | |
304 | ||
305 | static int xenvif_change_mtu(struct net_device *dev, int mtu) | |
306 | { | |
307 | struct xenvif *vif = netdev_priv(dev); | |
d0c2c997 | 308 | int max = vif->can_sg ? ETH_MAX_MTU - VLAN_ETH_HLEN : ETH_DATA_LEN; |
f942dc25 IC |
309 | |
310 | if (mtu > max) | |
311 | return -EINVAL; | |
312 | dev->mtu = mtu; | |
313 | return 0; | |
314 | } | |
315 | ||
c8f44aff MM |
316 | static netdev_features_t xenvif_fix_features(struct net_device *dev, |
317 | netdev_features_t features) | |
f942dc25 IC |
318 | { |
319 | struct xenvif *vif = netdev_priv(dev); | |
f942dc25 | 320 | |
47103041 MM |
321 | if (!vif->can_sg) |
322 | features &= ~NETIF_F_SG; | |
fedbc8c1 | 323 | if (~(vif->gso_mask) & GSO_BIT(TCPV4)) |
47103041 | 324 | features &= ~NETIF_F_TSO; |
fedbc8c1 | 325 | if (~(vif->gso_mask) & GSO_BIT(TCPV6)) |
82cada22 | 326 | features &= ~NETIF_F_TSO6; |
146c8a77 | 327 | if (!vif->ip_csum) |
47103041 | 328 | features &= ~NETIF_F_IP_CSUM; |
146c8a77 PD |
329 | if (!vif->ipv6_csum) |
330 | features &= ~NETIF_F_IPV6_CSUM; | |
f942dc25 | 331 | |
47103041 | 332 | return features; |
f942dc25 IC |
333 | } |
334 | ||
335 | static const struct xenvif_stat { | |
336 | char name[ETH_GSTRING_LEN]; | |
337 | u16 offset; | |
338 | } xenvif_stats[] = { | |
339 | { | |
340 | "rx_gso_checksum_fixup", | |
e9ce7cb6 | 341 | offsetof(struct xenvif_stats, rx_gso_checksum_fixup) |
f942dc25 | 342 | }, |
1bb332af ZK |
343 | /* If (sent != success + fail), there are probably packets never |
344 | * freed up properly! | |
345 | */ | |
346 | { | |
347 | "tx_zerocopy_sent", | |
e9ce7cb6 | 348 | offsetof(struct xenvif_stats, tx_zerocopy_sent), |
1bb332af ZK |
349 | }, |
350 | { | |
351 | "tx_zerocopy_success", | |
e9ce7cb6 | 352 | offsetof(struct xenvif_stats, tx_zerocopy_success), |
1bb332af ZK |
353 | }, |
354 | { | |
355 | "tx_zerocopy_fail", | |
e9ce7cb6 | 356 | offsetof(struct xenvif_stats, tx_zerocopy_fail) |
1bb332af | 357 | }, |
e3377f36 ZK |
358 | /* Number of packets exceeding MAX_SKB_FRAG slots. You should use |
359 | * a guest with the same MAX_SKB_FRAG | |
360 | */ | |
361 | { | |
362 | "tx_frag_overflow", | |
e9ce7cb6 | 363 | offsetof(struct xenvif_stats, tx_frag_overflow) |
e3377f36 | 364 | }, |
f942dc25 IC |
365 | }; |
366 | ||
367 | static int xenvif_get_sset_count(struct net_device *dev, int string_set) | |
368 | { | |
369 | switch (string_set) { | |
370 | case ETH_SS_STATS: | |
371 | return ARRAY_SIZE(xenvif_stats); | |
372 | default: | |
373 | return -EINVAL; | |
374 | } | |
375 | } | |
376 | ||
377 | static void xenvif_get_ethtool_stats(struct net_device *dev, | |
378 | struct ethtool_stats *stats, u64 * data) | |
379 | { | |
e9ce7cb6 | 380 | struct xenvif *vif = netdev_priv(dev); |
f7b50c4e | 381 | unsigned int num_queues = vif->num_queues; |
f942dc25 | 382 | int i; |
e9ce7cb6 | 383 | unsigned int queue_index; |
e9ce7cb6 WL |
384 | |
385 | for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++) { | |
386 | unsigned long accum = 0; | |
387 | for (queue_index = 0; queue_index < num_queues; ++queue_index) { | |
d63951d7 | 388 | void *vif_stats = &vif->queues[queue_index].stats; |
e9ce7cb6 WL |
389 | accum += *(unsigned long *)(vif_stats + xenvif_stats[i].offset); |
390 | } | |
391 | data[i] = accum; | |
392 | } | |
f942dc25 IC |
393 | } |
394 | ||
395 | static void xenvif_get_strings(struct net_device *dev, u32 stringset, u8 * data) | |
396 | { | |
397 | int i; | |
398 | ||
399 | switch (stringset) { | |
400 | case ETH_SS_STATS: | |
401 | for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++) | |
402 | memcpy(data + i * ETH_GSTRING_LEN, | |
403 | xenvif_stats[i].name, ETH_GSTRING_LEN); | |
404 | break; | |
405 | } | |
406 | } | |
407 | ||
813abbba | 408 | static const struct ethtool_ops xenvif_ethtool_ops = { |
f942dc25 IC |
409 | .get_link = ethtool_op_get_link, |
410 | ||
411 | .get_sset_count = xenvif_get_sset_count, | |
412 | .get_ethtool_stats = xenvif_get_ethtool_stats, | |
413 | .get_strings = xenvif_get_strings, | |
414 | }; | |
415 | ||
813abbba | 416 | static const struct net_device_ops xenvif_netdev_ops = { |
40d8abde | 417 | .ndo_select_queue = xenvif_select_queue, |
f942dc25 IC |
418 | .ndo_start_xmit = xenvif_start_xmit, |
419 | .ndo_get_stats = xenvif_get_stats, | |
420 | .ndo_open = xenvif_open, | |
421 | .ndo_stop = xenvif_close, | |
422 | .ndo_change_mtu = xenvif_change_mtu, | |
47103041 | 423 | .ndo_fix_features = xenvif_fix_features, |
4a633a60 MW |
424 | .ndo_set_mac_address = eth_mac_addr, |
425 | .ndo_validate_addr = eth_validate_addr, | |
f942dc25 IC |
426 | }; |
427 | ||
428 | struct xenvif *xenvif_alloc(struct device *parent, domid_t domid, | |
429 | unsigned int handle) | |
430 | { | |
431 | int err; | |
432 | struct net_device *dev; | |
433 | struct xenvif *vif; | |
434 | char name[IFNAMSIZ] = {}; | |
435 | ||
436 | snprintf(name, IFNAMSIZ - 1, "vif%u.%u", domid, handle); | |
8d3d53b3 AB |
437 | /* Allocate a netdev with the max. supported number of queues. |
438 | * When the guest selects the desired number, it will be updated | |
f7b50c4e | 439 | * via netif_set_real_num_*_queues(). |
8d3d53b3 | 440 | */ |
c835a677 TG |
441 | dev = alloc_netdev_mq(sizeof(struct xenvif), name, NET_NAME_UNKNOWN, |
442 | ether_setup, xenvif_max_queues); | |
f942dc25 | 443 | if (dev == NULL) { |
b3f980bd | 444 | pr_warn("Could not allocate netdev for %s\n", name); |
f942dc25 IC |
445 | return ERR_PTR(-ENOMEM); |
446 | } | |
447 | ||
448 | SET_NETDEV_DEV(dev, parent); | |
449 | ||
450 | vif = netdev_priv(dev); | |
ac3d5ac2 | 451 | |
f942dc25 IC |
452 | vif->domid = domid; |
453 | vif->handle = handle; | |
f942dc25 | 454 | vif->can_sg = 1; |
146c8a77 | 455 | vif->ip_csum = 1; |
f942dc25 | 456 | vif->dev = dev; |
e9d8b2c2 | 457 | vif->disabled = false; |
26c0e102 DV |
458 | vif->drain_timeout = msecs_to_jiffies(rx_drain_timeout_msecs); |
459 | vif->stall_timeout = msecs_to_jiffies(rx_stall_timeout_msecs); | |
e9d8b2c2 | 460 | |
f7b50c4e | 461 | /* Start out with no queues. */ |
e9ce7cb6 | 462 | vif->queues = NULL; |
f7b50c4e | 463 | vif->num_queues = 0; |
09350788 | 464 | |
ecf08d2d | 465 | spin_lock_init(&vif->lock); |
210c34dc | 466 | INIT_LIST_HEAD(&vif->fe_mcast_addr); |
ecf08d2d | 467 | |
f942dc25 | 468 | dev->netdev_ops = &xenvif_netdev_ops; |
146c8a77 PD |
469 | dev->hw_features = NETIF_F_SG | |
470 | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | | |
2167ca02 | 471 | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_FRAGLIST; |
7365bcfa | 472 | dev->features = dev->hw_features | NETIF_F_RXCSUM; |
7ad24ea4 | 473 | dev->ethtool_ops = &xenvif_ethtool_ops; |
f942dc25 IC |
474 | |
475 | dev->tx_queue_len = XENVIF_QUEUE_LENGTH; | |
476 | ||
d0c2c997 JW |
477 | dev->min_mtu = 0; |
478 | dev->max_mtu = ETH_MAX_MTU - VLAN_ETH_HLEN; | |
479 | ||
f942dc25 IC |
480 | /* |
481 | * Initialise a dummy MAC address. We choose the numerically | |
482 | * largest non-broadcast address to prevent the address getting | |
483 | * stolen by an Ethernet bridge for STP purposes. | |
484 | * (FE:FF:FF:FF:FF:FF) | |
485 | */ | |
3b6ed26d | 486 | eth_broadcast_addr(dev->dev_addr); |
f942dc25 IC |
487 | dev->dev_addr[0] &= ~0x01; |
488 | ||
489 | netif_carrier_off(dev); | |
490 | ||
491 | err = register_netdev(dev); | |
492 | if (err) { | |
493 | netdev_warn(dev, "Could not register device: err=%d\n", err); | |
494 | free_netdev(dev); | |
495 | return ERR_PTR(err); | |
496 | } | |
497 | ||
498 | netdev_dbg(dev, "Successfully created xenvif\n"); | |
279f438e PD |
499 | |
500 | __module_get(THIS_MODULE); | |
501 | ||
f942dc25 IC |
502 | return vif; |
503 | } | |
504 | ||
e9ce7cb6 WL |
505 | int xenvif_init_queue(struct xenvif_queue *queue) |
506 | { | |
507 | int err, i; | |
508 | ||
509 | queue->credit_bytes = queue->remaining_credit = ~0UL; | |
510 | queue->credit_usec = 0UL; | |
511 | init_timer(&queue->credit_timeout); | |
edafc132 | 512 | queue->credit_timeout.function = xenvif_tx_credit_callback; |
e9ce7cb6 WL |
513 | queue->credit_window_start = get_jiffies_64(); |
514 | ||
f48da8b1 DV |
515 | queue->rx_queue_max = XENVIF_RX_QUEUE_BYTES; |
516 | ||
e9ce7cb6 WL |
517 | skb_queue_head_init(&queue->rx_queue); |
518 | skb_queue_head_init(&queue->tx_queue); | |
519 | ||
520 | queue->pending_cons = 0; | |
521 | queue->pending_prod = MAX_PENDING_REQS; | |
522 | for (i = 0; i < MAX_PENDING_REQS; ++i) | |
523 | queue->pending_ring[i] = i; | |
524 | ||
525 | spin_lock_init(&queue->callback_lock); | |
526 | spin_lock_init(&queue->response_lock); | |
527 | ||
528 | /* If ballooning is disabled, this will consume real memory, so you | |
529 | * better enable it. The long term solution would be to use just a | |
530 | * bunch of valid page descriptors, without dependency on ballooning | |
531 | */ | |
ff4b156f DV |
532 | err = gnttab_alloc_pages(MAX_PENDING_REQS, |
533 | queue->mmap_pages); | |
e9ce7cb6 WL |
534 | if (err) { |
535 | netdev_err(queue->vif->dev, "Could not reserve mmap_pages\n"); | |
536 | return -ENOMEM; | |
537 | } | |
538 | ||
539 | for (i = 0; i < MAX_PENDING_REQS; i++) { | |
540 | queue->pending_tx_info[i].callback_struct = (struct ubuf_info) | |
541 | { .callback = xenvif_zerocopy_callback, | |
542 | .ctx = NULL, | |
543 | .desc = i }; | |
544 | queue->grant_tx_handle[i] = NETBACK_INVALID_HANDLE; | |
545 | } | |
546 | ||
e9ce7cb6 WL |
547 | return 0; |
548 | } | |
549 | ||
550 | void xenvif_carrier_on(struct xenvif *vif) | |
551 | { | |
552 | rtnl_lock(); | |
553 | if (!vif->can_sg && vif->dev->mtu > ETH_DATA_LEN) | |
554 | dev_set_mtu(vif->dev, ETH_DATA_LEN); | |
555 | netdev_update_features(vif->dev); | |
3d1af1df | 556 | set_bit(VIF_STATUS_CONNECTED, &vif->status); |
e9ce7cb6 WL |
557 | if (netif_running(vif->dev)) |
558 | xenvif_up(vif); | |
559 | rtnl_unlock(); | |
560 | } | |
561 | ||
4e15ee2c PD |
562 | int xenvif_connect_ctrl(struct xenvif *vif, grant_ref_t ring_ref, |
563 | unsigned int evtchn) | |
564 | { | |
565 | struct net_device *dev = vif->dev; | |
566 | void *addr; | |
567 | struct xen_netif_ctrl_sring *shared; | |
0364a882 | 568 | int err; |
4e15ee2c PD |
569 | |
570 | err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(vif), | |
571 | &ring_ref, 1, &addr); | |
572 | if (err) | |
573 | goto err; | |
574 | ||
575 | shared = (struct xen_netif_ctrl_sring *)addr; | |
576 | BACK_RING_INIT(&vif->ctrl, shared, XEN_PAGE_SIZE); | |
577 | ||
0364a882 | 578 | err = bind_interdomain_evtchn_to_irq(vif->domid, evtchn); |
4e15ee2c PD |
579 | if (err < 0) |
580 | goto err_unmap; | |
581 | ||
582 | vif->ctrl_irq = err; | |
583 | ||
40d8abde PD |
584 | xenvif_init_hash(vif); |
585 | ||
0364a882 JG |
586 | err = request_threaded_irq(vif->ctrl_irq, NULL, xenvif_ctrl_irq_fn, |
587 | IRQF_ONESHOT, "xen-netback-ctrl", vif); | |
588 | if (err) { | |
589 | pr_warn("Could not setup irq handler for %s\n", dev->name); | |
4e15ee2c PD |
590 | goto err_deinit; |
591 | } | |
592 | ||
4e15ee2c PD |
593 | return 0; |
594 | ||
595 | err_deinit: | |
40d8abde | 596 | xenvif_deinit_hash(vif); |
4e15ee2c PD |
597 | unbind_from_irqhandler(vif->ctrl_irq, vif); |
598 | vif->ctrl_irq = 0; | |
599 | ||
600 | err_unmap: | |
601 | xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif), | |
602 | vif->ctrl.sring); | |
603 | vif->ctrl.sring = NULL; | |
604 | ||
605 | err: | |
606 | return err; | |
607 | } | |
608 | ||
609 | int xenvif_connect_data(struct xenvif_queue *queue, | |
610 | unsigned long tx_ring_ref, | |
611 | unsigned long rx_ring_ref, | |
612 | unsigned int tx_evtchn, | |
613 | unsigned int rx_evtchn) | |
f942dc25 | 614 | { |
67fa3660 | 615 | struct task_struct *task; |
f942dc25 IC |
616 | int err = -ENOMEM; |
617 | ||
e9ce7cb6 WL |
618 | BUG_ON(queue->tx_irq); |
619 | BUG_ON(queue->task); | |
620 | BUG_ON(queue->dealloc_task); | |
f942dc25 | 621 | |
4e15ee2c PD |
622 | err = xenvif_map_frontend_data_rings(queue, tx_ring_ref, |
623 | rx_ring_ref); | |
f942dc25 IC |
624 | if (err < 0) |
625 | goto err; | |
626 | ||
e9ce7cb6 WL |
627 | init_waitqueue_head(&queue->wq); |
628 | init_waitqueue_head(&queue->dealloc_wq); | |
a64bd934 | 629 | atomic_set(&queue->inflight_packets, 0); |
ca2f09f2 | 630 | |
e24f8191 WL |
631 | netif_napi_add(queue->vif->dev, &queue->napi, xenvif_poll, |
632 | XENVIF_NAPI_WEIGHT); | |
633 | ||
e1f00a69 WL |
634 | if (tx_evtchn == rx_evtchn) { |
635 | /* feature-split-event-channels == 0 */ | |
636 | err = bind_interdomain_evtchn_to_irqhandler( | |
e9ce7cb6 WL |
637 | queue->vif->domid, tx_evtchn, xenvif_interrupt, 0, |
638 | queue->name, queue); | |
e1f00a69 WL |
639 | if (err < 0) |
640 | goto err_unmap; | |
e9ce7cb6 WL |
641 | queue->tx_irq = queue->rx_irq = err; |
642 | disable_irq(queue->tx_irq); | |
e1f00a69 WL |
643 | } else { |
644 | /* feature-split-event-channels == 1 */ | |
e9ce7cb6 WL |
645 | snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name), |
646 | "%s-tx", queue->name); | |
e1f00a69 | 647 | err = bind_interdomain_evtchn_to_irqhandler( |
e9ce7cb6 WL |
648 | queue->vif->domid, tx_evtchn, xenvif_tx_interrupt, 0, |
649 | queue->tx_irq_name, queue); | |
e1f00a69 WL |
650 | if (err < 0) |
651 | goto err_unmap; | |
e9ce7cb6 WL |
652 | queue->tx_irq = err; |
653 | disable_irq(queue->tx_irq); | |
e1f00a69 | 654 | |
e9ce7cb6 WL |
655 | snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name), |
656 | "%s-rx", queue->name); | |
e1f00a69 | 657 | err = bind_interdomain_evtchn_to_irqhandler( |
e9ce7cb6 WL |
658 | queue->vif->domid, rx_evtchn, xenvif_rx_interrupt, 0, |
659 | queue->rx_irq_name, queue); | |
e1f00a69 WL |
660 | if (err < 0) |
661 | goto err_tx_unbind; | |
e9ce7cb6 WL |
662 | queue->rx_irq = err; |
663 | disable_irq(queue->rx_irq); | |
e1f00a69 | 664 | } |
f942dc25 | 665 | |
ecf08d2d DV |
666 | queue->stalled = true; |
667 | ||
121fa4b7 | 668 | task = kthread_create(xenvif_kthread_guest_rx, |
e9ce7cb6 | 669 | (void *)queue, "%s-guest-rx", queue->name); |
67fa3660 | 670 | if (IS_ERR(task)) { |
e9ce7cb6 | 671 | pr_warn("Could not allocate kthread for %s\n", queue->name); |
67fa3660 | 672 | err = PTR_ERR(task); |
b3f980bd WL |
673 | goto err_rx_unbind; |
674 | } | |
e9ce7cb6 | 675 | queue->task = task; |
42b5212f | 676 | get_task_struct(task); |
67fa3660 | 677 | |
f53c3fe8 | 678 | task = kthread_create(xenvif_dealloc_kthread, |
e9ce7cb6 | 679 | (void *)queue, "%s-dealloc", queue->name); |
f53c3fe8 | 680 | if (IS_ERR(task)) { |
e9ce7cb6 | 681 | pr_warn("Could not allocate kthread for %s\n", queue->name); |
f53c3fe8 ZK |
682 | err = PTR_ERR(task); |
683 | goto err_rx_unbind; | |
684 | } | |
e9ce7cb6 | 685 | queue->dealloc_task = task; |
f53c3fe8 | 686 | |
e9ce7cb6 WL |
687 | wake_up_process(queue->task); |
688 | wake_up_process(queue->dealloc_task); | |
b3f980bd | 689 | |
f942dc25 | 690 | return 0; |
b3f980bd WL |
691 | |
692 | err_rx_unbind: | |
e9ce7cb6 WL |
693 | unbind_from_irqhandler(queue->rx_irq, queue); |
694 | queue->rx_irq = 0; | |
e1f00a69 | 695 | err_tx_unbind: |
e9ce7cb6 WL |
696 | unbind_from_irqhandler(queue->tx_irq, queue); |
697 | queue->tx_irq = 0; | |
f942dc25 | 698 | err_unmap: |
4e15ee2c | 699 | xenvif_unmap_frontend_data_rings(queue); |
4a658527 | 700 | netif_napi_del(&queue->napi); |
f942dc25 | 701 | err: |
b103f358 | 702 | module_put(THIS_MODULE); |
f942dc25 IC |
703 | return err; |
704 | } | |
705 | ||
48856286 | 706 | void xenvif_carrier_off(struct xenvif *vif) |
f942dc25 IC |
707 | { |
708 | struct net_device *dev = vif->dev; | |
48856286 IC |
709 | |
710 | rtnl_lock(); | |
3d1af1df ZK |
711 | if (test_and_clear_bit(VIF_STATUS_CONNECTED, &vif->status)) { |
712 | netif_carrier_off(dev); /* discard queued packets */ | |
713 | if (netif_running(dev)) | |
714 | xenvif_down(vif); | |
715 | } | |
48856286 | 716 | rtnl_unlock(); |
48856286 IC |
717 | } |
718 | ||
4e15ee2c | 719 | void xenvif_disconnect_data(struct xenvif *vif) |
48856286 | 720 | { |
e9ce7cb6 | 721 | struct xenvif_queue *queue = NULL; |
f7b50c4e | 722 | unsigned int num_queues = vif->num_queues; |
e9ce7cb6 WL |
723 | unsigned int queue_index; |
724 | ||
3d1af1df | 725 | xenvif_carrier_off(vif); |
f942dc25 | 726 | |
e9ce7cb6 WL |
727 | for (queue_index = 0; queue_index < num_queues; ++queue_index) { |
728 | queue = &vif->queues[queue_index]; | |
db739ef3 | 729 | |
ea2c5e13 WL |
730 | netif_napi_del(&queue->napi); |
731 | ||
e9ce7cb6 | 732 | if (queue->task) { |
e9ce7cb6 | 733 | kthread_stop(queue->task); |
42b5212f | 734 | put_task_struct(queue->task); |
e9ce7cb6 WL |
735 | queue->task = NULL; |
736 | } | |
f53c3fe8 | 737 | |
e9ce7cb6 WL |
738 | if (queue->dealloc_task) { |
739 | kthread_stop(queue->dealloc_task); | |
740 | queue->dealloc_task = NULL; | |
741 | } | |
742 | ||
743 | if (queue->tx_irq) { | |
744 | if (queue->tx_irq == queue->rx_irq) | |
745 | unbind_from_irqhandler(queue->tx_irq, queue); | |
746 | else { | |
747 | unbind_from_irqhandler(queue->tx_irq, queue); | |
748 | unbind_from_irqhandler(queue->rx_irq, queue); | |
749 | } | |
750 | queue->tx_irq = 0; | |
e1f00a69 | 751 | } |
f942dc25 | 752 | |
4e15ee2c | 753 | xenvif_unmap_frontend_data_rings(queue); |
e9ce7cb6 | 754 | } |
210c34dc PD |
755 | |
756 | xenvif_mcast_addr_list_free(vif); | |
279f438e PD |
757 | } |
758 | ||
4e15ee2c PD |
759 | void xenvif_disconnect_ctrl(struct xenvif *vif) |
760 | { | |
4e15ee2c | 761 | if (vif->ctrl_irq) { |
c0fcded2 | 762 | xenvif_deinit_hash(vif); |
4e15ee2c PD |
763 | unbind_from_irqhandler(vif->ctrl_irq, vif); |
764 | vif->ctrl_irq = 0; | |
765 | } | |
766 | ||
767 | if (vif->ctrl.sring) { | |
768 | xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif), | |
769 | vif->ctrl.sring); | |
770 | vif->ctrl.sring = NULL; | |
771 | } | |
772 | } | |
773 | ||
8d3d53b3 AB |
774 | /* Reverse the relevant parts of xenvif_init_queue(). |
775 | * Used for queue teardown from xenvif_free(), and on the | |
776 | * error handling paths in xenbus.c:connect(). | |
777 | */ | |
778 | void xenvif_deinit_queue(struct xenvif_queue *queue) | |
779 | { | |
ff4b156f | 780 | gnttab_free_pages(MAX_PENDING_REQS, queue->mmap_pages); |
8d3d53b3 AB |
781 | } |
782 | ||
279f438e PD |
783 | void xenvif_free(struct xenvif *vif) |
784 | { | |
9c6f3ffe | 785 | struct xenvif_queue *queues = vif->queues; |
f7b50c4e | 786 | unsigned int num_queues = vif->num_queues; |
e9ce7cb6 | 787 | unsigned int queue_index; |
f53c3fe8 | 788 | |
e9ce7cb6 | 789 | unregister_netdev(vif->dev); |
f942dc25 | 790 | free_netdev(vif->dev); |
b103f358 | 791 | |
9c6f3ffe DV |
792 | for (queue_index = 0; queue_index < num_queues; ++queue_index) |
793 | xenvif_deinit_queue(&queues[queue_index]); | |
794 | vfree(queues); | |
795 | ||
279f438e | 796 | module_put(THIS_MODULE); |
f942dc25 | 797 | } |