interface.c 20 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798
  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. #include "common.h"
  31. #include <linux/kthread.h>
  32. #include <linux/ethtool.h>
  33. #include <linux/rtnetlink.h>
  34. #include <linux/if_vlan.h>
  35. #include <linux/vmalloc.h>
  36. #include <xen/events.h>
  37. #include <asm/xen/hypercall.h>
  38. #include <xen/balloon.h>
  39. #define XENVIF_QUEUE_LENGTH 32
  40. #define XENVIF_NAPI_WEIGHT 64
  41. /* Number of bytes allowed on the internal guest Rx queue. */
  42. #define XENVIF_RX_QUEUE_BYTES (XEN_NETIF_RX_RING_SIZE/2 * PAGE_SIZE)
  43. /* This function is used to set SKBTX_DEV_ZEROCOPY as well as
  44. * increasing the inflight counter. We need to increase the inflight
  45. * counter because core driver calls into xenvif_zerocopy_callback
  46. * which calls xenvif_skb_zerocopy_complete.
  47. */
  48. void xenvif_skb_zerocopy_prepare(struct xenvif_queue *queue,
  49. struct sk_buff *skb)
  50. {
  51. skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
  52. atomic_inc(&queue->inflight_packets);
  53. }
  54. void xenvif_skb_zerocopy_complete(struct xenvif_queue *queue)
  55. {
  56. atomic_dec(&queue->inflight_packets);
  57. /* Wake the dealloc thread _after_ decrementing inflight_packets so
  58. * that if kthread_stop() has already been called, the dealloc thread
  59. * does not wait forever with nothing to wake it.
  60. */
  61. wake_up(&queue->dealloc_wq);
  62. }
  63. int xenvif_schedulable(struct xenvif *vif)
  64. {
  65. return netif_running(vif->dev) &&
  66. test_bit(VIF_STATUS_CONNECTED, &vif->status) &&
  67. !vif->disabled;
  68. }
  69. static irqreturn_t xenvif_tx_interrupt(int irq, void *dev_id)
  70. {
  71. struct xenvif_queue *queue = dev_id;
  72. if (RING_HAS_UNCONSUMED_REQUESTS(&queue->tx))
  73. napi_schedule(&queue->napi);
  74. return IRQ_HANDLED;
  75. }
  76. static int xenvif_poll(struct napi_struct *napi, int budget)
  77. {
  78. struct xenvif_queue *queue =
  79. container_of(napi, struct xenvif_queue, napi);
  80. int work_done;
  81. /* This vif is rogue, we pretend we've there is nothing to do
  82. * for this vif to deschedule it from NAPI. But this interface
  83. * will be turned off in thread context later.
  84. */
  85. if (unlikely(queue->vif->disabled)) {
  86. napi_complete(napi);
  87. return 0;
  88. }
  89. work_done = xenvif_tx_action(queue, budget);
  90. if (work_done < budget) {
  91. napi_complete(napi);
  92. /* If the queue is rate-limited, it shall be
  93. * rescheduled in the timer callback.
  94. */
  95. if (likely(!queue->rate_limited))
  96. xenvif_napi_schedule_or_enable_events(queue);
  97. }
  98. return work_done;
  99. }
  100. static irqreturn_t xenvif_rx_interrupt(int irq, void *dev_id)
  101. {
  102. struct xenvif_queue *queue = dev_id;
  103. xenvif_kick_thread(queue);
  104. return IRQ_HANDLED;
  105. }
  106. irqreturn_t xenvif_interrupt(int irq, void *dev_id)
  107. {
  108. xenvif_tx_interrupt(irq, dev_id);
  109. xenvif_rx_interrupt(irq, dev_id);
  110. return IRQ_HANDLED;
  111. }
  112. int xenvif_queue_stopped(struct xenvif_queue *queue)
  113. {
  114. struct net_device *dev = queue->vif->dev;
  115. unsigned int id = queue->id;
  116. return netif_tx_queue_stopped(netdev_get_tx_queue(dev, id));
  117. }
  118. void xenvif_wake_queue(struct xenvif_queue *queue)
  119. {
  120. struct net_device *dev = queue->vif->dev;
  121. unsigned int id = queue->id;
  122. netif_tx_wake_queue(netdev_get_tx_queue(dev, id));
  123. }
  124. static u16 xenvif_select_queue(struct net_device *dev, struct sk_buff *skb,
  125. void *accel_priv,
  126. select_queue_fallback_t fallback)
  127. {
  128. struct xenvif *vif = netdev_priv(dev);
  129. unsigned int size = vif->hash.size;
  130. if (vif->hash.alg == XEN_NETIF_CTRL_HASH_ALGORITHM_NONE)
  131. return fallback(dev, skb) % dev->real_num_tx_queues;
  132. xenvif_set_skb_hash(vif, skb);
  133. if (size == 0)
  134. return skb_get_hash_raw(skb) % dev->real_num_tx_queues;
  135. return vif->hash.mapping[skb_get_hash_raw(skb) % size];
  136. }
  137. static int xenvif_start_xmit(struct sk_buff *skb, struct net_device *dev)
  138. {
  139. struct xenvif *vif = netdev_priv(dev);
  140. struct xenvif_queue *queue = NULL;
  141. unsigned int num_queues = vif->num_queues;
  142. u16 index;
  143. struct xenvif_rx_cb *cb;
  144. BUG_ON(skb->dev != dev);
  145. /* Drop the packet if queues are not set up */
  146. if (num_queues < 1)
  147. goto drop;
  148. /* Obtain the queue to be used to transmit this packet */
  149. index = skb_get_queue_mapping(skb);
  150. if (index >= num_queues) {
  151. pr_warn_ratelimited("Invalid queue %hu for packet on interface %s\n.",
  152. index, vif->dev->name);
  153. index %= num_queues;
  154. }
  155. queue = &vif->queues[index];
  156. /* Drop the packet if queue is not ready */
  157. if (queue->task == NULL ||
  158. queue->dealloc_task == NULL ||
  159. !xenvif_schedulable(vif))
  160. goto drop;
  161. if (vif->multicast_control && skb->pkt_type == PACKET_MULTICAST) {
  162. struct ethhdr *eth = (struct ethhdr *)skb->data;
  163. if (!xenvif_mcast_match(vif, eth->h_dest))
  164. goto drop;
  165. }
  166. cb = XENVIF_RX_CB(skb);
  167. cb->expires = jiffies + vif->drain_timeout;
  168. /* If there is no hash algorithm configured then make sure there
  169. * is no hash information in the socket buffer otherwise it
  170. * would be incorrectly forwarded to the frontend.
  171. */
  172. if (vif->hash.alg == XEN_NETIF_CTRL_HASH_ALGORITHM_NONE)
  173. skb_clear_hash(skb);
  174. xenvif_rx_queue_tail(queue, skb);
  175. xenvif_kick_thread(queue);
  176. return NETDEV_TX_OK;
  177. drop:
  178. vif->dev->stats.tx_dropped++;
  179. dev_kfree_skb(skb);
  180. return NETDEV_TX_OK;
  181. }
  182. static struct net_device_stats *xenvif_get_stats(struct net_device *dev)
  183. {
  184. struct xenvif *vif = netdev_priv(dev);
  185. struct xenvif_queue *queue = NULL;
  186. u64 rx_bytes = 0;
  187. u64 rx_packets = 0;
  188. u64 tx_bytes = 0;
  189. u64 tx_packets = 0;
  190. unsigned int index;
  191. spin_lock(&vif->lock);
  192. if (vif->queues == NULL)
  193. goto out;
  194. /* Aggregate tx and rx stats from each queue */
  195. for (index = 0; index < vif->num_queues; ++index) {
  196. queue = &vif->queues[index];
  197. rx_bytes += queue->stats.rx_bytes;
  198. rx_packets += queue->stats.rx_packets;
  199. tx_bytes += queue->stats.tx_bytes;
  200. tx_packets += queue->stats.tx_packets;
  201. }
  202. out:
  203. spin_unlock(&vif->lock);
  204. vif->dev->stats.rx_bytes = rx_bytes;
  205. vif->dev->stats.rx_packets = rx_packets;
  206. vif->dev->stats.tx_bytes = tx_bytes;
  207. vif->dev->stats.tx_packets = tx_packets;
  208. return &vif->dev->stats;
  209. }
  210. static void xenvif_up(struct xenvif *vif)
  211. {
  212. struct xenvif_queue *queue = NULL;
  213. unsigned int num_queues = vif->num_queues;
  214. unsigned int queue_index;
  215. for (queue_index = 0; queue_index < num_queues; ++queue_index) {
  216. queue = &vif->queues[queue_index];
  217. napi_enable(&queue->napi);
  218. enable_irq(queue->tx_irq);
  219. if (queue->tx_irq != queue->rx_irq)
  220. enable_irq(queue->rx_irq);
  221. xenvif_napi_schedule_or_enable_events(queue);
  222. }
  223. }
  224. static void xenvif_down(struct xenvif *vif)
  225. {
  226. struct xenvif_queue *queue = NULL;
  227. unsigned int num_queues = vif->num_queues;
  228. unsigned int queue_index;
  229. for (queue_index = 0; queue_index < num_queues; ++queue_index) {
  230. queue = &vif->queues[queue_index];
  231. disable_irq(queue->tx_irq);
  232. if (queue->tx_irq != queue->rx_irq)
  233. disable_irq(queue->rx_irq);
  234. napi_disable(&queue->napi);
  235. del_timer_sync(&queue->credit_timeout);
  236. }
  237. }
  238. static int xenvif_open(struct net_device *dev)
  239. {
  240. struct xenvif *vif = netdev_priv(dev);
  241. if (test_bit(VIF_STATUS_CONNECTED, &vif->status))
  242. xenvif_up(vif);
  243. netif_tx_start_all_queues(dev);
  244. return 0;
  245. }
  246. static int xenvif_close(struct net_device *dev)
  247. {
  248. struct xenvif *vif = netdev_priv(dev);
  249. if (test_bit(VIF_STATUS_CONNECTED, &vif->status))
  250. xenvif_down(vif);
  251. netif_tx_stop_all_queues(dev);
  252. return 0;
  253. }
  254. static int xenvif_change_mtu(struct net_device *dev, int mtu)
  255. {
  256. struct xenvif *vif = netdev_priv(dev);
  257. int max = vif->can_sg ? 65535 - VLAN_ETH_HLEN : ETH_DATA_LEN;
  258. if (mtu > max)
  259. return -EINVAL;
  260. dev->mtu = mtu;
  261. return 0;
  262. }
  263. static netdev_features_t xenvif_fix_features(struct net_device *dev,
  264. netdev_features_t features)
  265. {
  266. struct xenvif *vif = netdev_priv(dev);
  267. if (!vif->can_sg)
  268. features &= ~NETIF_F_SG;
  269. if (~(vif->gso_mask) & GSO_BIT(TCPV4))
  270. features &= ~NETIF_F_TSO;
  271. if (~(vif->gso_mask) & GSO_BIT(TCPV6))
  272. features &= ~NETIF_F_TSO6;
  273. if (!vif->ip_csum)
  274. features &= ~NETIF_F_IP_CSUM;
  275. if (!vif->ipv6_csum)
  276. features &= ~NETIF_F_IPV6_CSUM;
  277. return features;
  278. }
  279. static const struct xenvif_stat {
  280. char name[ETH_GSTRING_LEN];
  281. u16 offset;
  282. } xenvif_stats[] = {
  283. {
  284. "rx_gso_checksum_fixup",
  285. offsetof(struct xenvif_stats, rx_gso_checksum_fixup)
  286. },
  287. /* If (sent != success + fail), there are probably packets never
  288. * freed up properly!
  289. */
  290. {
  291. "tx_zerocopy_sent",
  292. offsetof(struct xenvif_stats, tx_zerocopy_sent),
  293. },
  294. {
  295. "tx_zerocopy_success",
  296. offsetof(struct xenvif_stats, tx_zerocopy_success),
  297. },
  298. {
  299. "tx_zerocopy_fail",
  300. offsetof(struct xenvif_stats, tx_zerocopy_fail)
  301. },
  302. /* Number of packets exceeding MAX_SKB_FRAG slots. You should use
  303. * a guest with the same MAX_SKB_FRAG
  304. */
  305. {
  306. "tx_frag_overflow",
  307. offsetof(struct xenvif_stats, tx_frag_overflow)
  308. },
  309. };
  310. static int xenvif_get_sset_count(struct net_device *dev, int string_set)
  311. {
  312. switch (string_set) {
  313. case ETH_SS_STATS:
  314. return ARRAY_SIZE(xenvif_stats);
  315. default:
  316. return -EINVAL;
  317. }
  318. }
  319. static void xenvif_get_ethtool_stats(struct net_device *dev,
  320. struct ethtool_stats *stats, u64 * data)
  321. {
  322. struct xenvif *vif = netdev_priv(dev);
  323. unsigned int num_queues = vif->num_queues;
  324. int i;
  325. unsigned int queue_index;
  326. for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++) {
  327. unsigned long accum = 0;
  328. for (queue_index = 0; queue_index < num_queues; ++queue_index) {
  329. void *vif_stats = &vif->queues[queue_index].stats;
  330. accum += *(unsigned long *)(vif_stats + xenvif_stats[i].offset);
  331. }
  332. data[i] = accum;
  333. }
  334. }
  335. static void xenvif_get_strings(struct net_device *dev, u32 stringset, u8 * data)
  336. {
  337. int i;
  338. switch (stringset) {
  339. case ETH_SS_STATS:
  340. for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++)
  341. memcpy(data + i * ETH_GSTRING_LEN,
  342. xenvif_stats[i].name, ETH_GSTRING_LEN);
  343. break;
  344. }
  345. }
  346. static const struct ethtool_ops xenvif_ethtool_ops = {
  347. .get_link = ethtool_op_get_link,
  348. .get_sset_count = xenvif_get_sset_count,
  349. .get_ethtool_stats = xenvif_get_ethtool_stats,
  350. .get_strings = xenvif_get_strings,
  351. };
  352. static const struct net_device_ops xenvif_netdev_ops = {
  353. .ndo_select_queue = xenvif_select_queue,
  354. .ndo_start_xmit = xenvif_start_xmit,
  355. .ndo_get_stats = xenvif_get_stats,
  356. .ndo_open = xenvif_open,
  357. .ndo_stop = xenvif_close,
  358. .ndo_change_mtu = xenvif_change_mtu,
  359. .ndo_fix_features = xenvif_fix_features,
  360. .ndo_set_mac_address = eth_mac_addr,
  361. .ndo_validate_addr = eth_validate_addr,
  362. };
  363. struct xenvif *xenvif_alloc(struct device *parent, domid_t domid,
  364. unsigned int handle)
  365. {
  366. int err;
  367. struct net_device *dev;
  368. struct xenvif *vif;
  369. char name[IFNAMSIZ] = {};
  370. snprintf(name, IFNAMSIZ - 1, "vif%u.%u", domid, handle);
  371. /* Allocate a netdev with the max. supported number of queues.
  372. * When the guest selects the desired number, it will be updated
  373. * via netif_set_real_num_*_queues().
  374. */
  375. dev = alloc_netdev_mq(sizeof(struct xenvif), name, NET_NAME_UNKNOWN,
  376. ether_setup, xenvif_max_queues);
  377. if (dev == NULL) {
  378. pr_warn("Could not allocate netdev for %s\n", name);
  379. return ERR_PTR(-ENOMEM);
  380. }
  381. SET_NETDEV_DEV(dev, parent);
  382. vif = netdev_priv(dev);
  383. vif->domid = domid;
  384. vif->handle = handle;
  385. vif->can_sg = 1;
  386. vif->ip_csum = 1;
  387. vif->dev = dev;
  388. vif->disabled = false;
  389. vif->drain_timeout = msecs_to_jiffies(rx_drain_timeout_msecs);
  390. vif->stall_timeout = msecs_to_jiffies(rx_stall_timeout_msecs);
  391. /* Start out with no queues. */
  392. vif->queues = NULL;
  393. vif->num_queues = 0;
  394. spin_lock_init(&vif->lock);
  395. INIT_LIST_HEAD(&vif->fe_mcast_addr);
  396. dev->netdev_ops = &xenvif_netdev_ops;
  397. dev->hw_features = NETIF_F_SG |
  398. NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
  399. NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_FRAGLIST;
  400. dev->features = dev->hw_features | NETIF_F_RXCSUM;
  401. dev->ethtool_ops = &xenvif_ethtool_ops;
  402. dev->tx_queue_len = XENVIF_QUEUE_LENGTH;
  403. /*
  404. * Initialise a dummy MAC address. We choose the numerically
  405. * largest non-broadcast address to prevent the address getting
  406. * stolen by an Ethernet bridge for STP purposes.
  407. * (FE:FF:FF:FF:FF:FF)
  408. */
  409. eth_broadcast_addr(dev->dev_addr);
  410. dev->dev_addr[0] &= ~0x01;
  411. netif_carrier_off(dev);
  412. err = register_netdev(dev);
  413. if (err) {
  414. netdev_warn(dev, "Could not register device: err=%d\n", err);
  415. free_netdev(dev);
  416. return ERR_PTR(err);
  417. }
  418. netdev_dbg(dev, "Successfully created xenvif\n");
  419. __module_get(THIS_MODULE);
  420. return vif;
  421. }
  422. int xenvif_init_queue(struct xenvif_queue *queue)
  423. {
  424. int err, i;
  425. queue->credit_bytes = queue->remaining_credit = ~0UL;
  426. queue->credit_usec = 0UL;
  427. init_timer(&queue->credit_timeout);
  428. queue->credit_timeout.function = xenvif_tx_credit_callback;
  429. queue->credit_window_start = get_jiffies_64();
  430. queue->rx_queue_max = XENVIF_RX_QUEUE_BYTES;
  431. skb_queue_head_init(&queue->rx_queue);
  432. skb_queue_head_init(&queue->tx_queue);
  433. queue->pending_cons = 0;
  434. queue->pending_prod = MAX_PENDING_REQS;
  435. for (i = 0; i < MAX_PENDING_REQS; ++i)
  436. queue->pending_ring[i] = i;
  437. spin_lock_init(&queue->callback_lock);
  438. spin_lock_init(&queue->response_lock);
  439. /* If ballooning is disabled, this will consume real memory, so you
  440. * better enable it. The long term solution would be to use just a
  441. * bunch of valid page descriptors, without dependency on ballooning
  442. */
  443. err = gnttab_alloc_pages(MAX_PENDING_REQS,
  444. queue->mmap_pages);
  445. if (err) {
  446. netdev_err(queue->vif->dev, "Could not reserve mmap_pages\n");
  447. return -ENOMEM;
  448. }
  449. for (i = 0; i < MAX_PENDING_REQS; i++) {
  450. queue->pending_tx_info[i].callback_struct = (struct ubuf_info)
  451. { .callback = xenvif_zerocopy_callback,
  452. .ctx = NULL,
  453. .desc = i };
  454. queue->grant_tx_handle[i] = NETBACK_INVALID_HANDLE;
  455. }
  456. return 0;
  457. }
  458. void xenvif_carrier_on(struct xenvif *vif)
  459. {
  460. rtnl_lock();
  461. if (!vif->can_sg && vif->dev->mtu > ETH_DATA_LEN)
  462. dev_set_mtu(vif->dev, ETH_DATA_LEN);
  463. netdev_update_features(vif->dev);
  464. set_bit(VIF_STATUS_CONNECTED, &vif->status);
  465. if (netif_running(vif->dev))
  466. xenvif_up(vif);
  467. rtnl_unlock();
  468. }
  469. int xenvif_connect_ctrl(struct xenvif *vif, grant_ref_t ring_ref,
  470. unsigned int evtchn)
  471. {
  472. struct net_device *dev = vif->dev;
  473. void *addr;
  474. struct xen_netif_ctrl_sring *shared;
  475. int err;
  476. err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(vif),
  477. &ring_ref, 1, &addr);
  478. if (err)
  479. goto err;
  480. shared = (struct xen_netif_ctrl_sring *)addr;
  481. BACK_RING_INIT(&vif->ctrl, shared, XEN_PAGE_SIZE);
  482. err = bind_interdomain_evtchn_to_irq(vif->domid, evtchn);
  483. if (err < 0)
  484. goto err_unmap;
  485. vif->ctrl_irq = err;
  486. xenvif_init_hash(vif);
  487. err = request_threaded_irq(vif->ctrl_irq, NULL, xenvif_ctrl_irq_fn,
  488. IRQF_ONESHOT, "xen-netback-ctrl", vif);
  489. if (err) {
  490. pr_warn("Could not setup irq handler for %s\n", dev->name);
  491. goto err_deinit;
  492. }
  493. return 0;
  494. err_deinit:
  495. xenvif_deinit_hash(vif);
  496. unbind_from_irqhandler(vif->ctrl_irq, vif);
  497. vif->ctrl_irq = 0;
  498. err_unmap:
  499. xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif),
  500. vif->ctrl.sring);
  501. vif->ctrl.sring = NULL;
  502. err:
  503. return err;
  504. }
  505. int xenvif_connect_data(struct xenvif_queue *queue,
  506. unsigned long tx_ring_ref,
  507. unsigned long rx_ring_ref,
  508. unsigned int tx_evtchn,
  509. unsigned int rx_evtchn)
  510. {
  511. struct task_struct *task;
  512. int err = -ENOMEM;
  513. BUG_ON(queue->tx_irq);
  514. BUG_ON(queue->task);
  515. BUG_ON(queue->dealloc_task);
  516. err = xenvif_map_frontend_data_rings(queue, tx_ring_ref,
  517. rx_ring_ref);
  518. if (err < 0)
  519. goto err;
  520. init_waitqueue_head(&queue->wq);
  521. init_waitqueue_head(&queue->dealloc_wq);
  522. atomic_set(&queue->inflight_packets, 0);
  523. netif_napi_add(queue->vif->dev, &queue->napi, xenvif_poll,
  524. XENVIF_NAPI_WEIGHT);
  525. if (tx_evtchn == rx_evtchn) {
  526. /* feature-split-event-channels == 0 */
  527. err = bind_interdomain_evtchn_to_irqhandler(
  528. queue->vif->domid, tx_evtchn, xenvif_interrupt, 0,
  529. queue->name, queue);
  530. if (err < 0)
  531. goto err_unmap;
  532. queue->tx_irq = queue->rx_irq = err;
  533. disable_irq(queue->tx_irq);
  534. } else {
  535. /* feature-split-event-channels == 1 */
  536. snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
  537. "%s-tx", queue->name);
  538. err = bind_interdomain_evtchn_to_irqhandler(
  539. queue->vif->domid, tx_evtchn, xenvif_tx_interrupt, 0,
  540. queue->tx_irq_name, queue);
  541. if (err < 0)
  542. goto err_unmap;
  543. queue->tx_irq = err;
  544. disable_irq(queue->tx_irq);
  545. snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
  546. "%s-rx", queue->name);
  547. err = bind_interdomain_evtchn_to_irqhandler(
  548. queue->vif->domid, rx_evtchn, xenvif_rx_interrupt, 0,
  549. queue->rx_irq_name, queue);
  550. if (err < 0)
  551. goto err_tx_unbind;
  552. queue->rx_irq = err;
  553. disable_irq(queue->rx_irq);
  554. }
  555. queue->stalled = true;
  556. task = kthread_create(xenvif_kthread_guest_rx,
  557. (void *)queue, "%s-guest-rx", queue->name);
  558. if (IS_ERR(task)) {
  559. pr_warn("Could not allocate kthread for %s\n", queue->name);
  560. err = PTR_ERR(task);
  561. goto err_rx_unbind;
  562. }
  563. queue->task = task;
  564. get_task_struct(task);
  565. task = kthread_create(xenvif_dealloc_kthread,
  566. (void *)queue, "%s-dealloc", queue->name);
  567. if (IS_ERR(task)) {
  568. pr_warn("Could not allocate kthread for %s\n", queue->name);
  569. err = PTR_ERR(task);
  570. goto err_rx_unbind;
  571. }
  572. queue->dealloc_task = task;
  573. wake_up_process(queue->task);
  574. wake_up_process(queue->dealloc_task);
  575. return 0;
  576. err_rx_unbind:
  577. unbind_from_irqhandler(queue->rx_irq, queue);
  578. queue->rx_irq = 0;
  579. err_tx_unbind:
  580. unbind_from_irqhandler(queue->tx_irq, queue);
  581. queue->tx_irq = 0;
  582. err_unmap:
  583. xenvif_unmap_frontend_data_rings(queue);
  584. netif_napi_del(&queue->napi);
  585. err:
  586. module_put(THIS_MODULE);
  587. return err;
  588. }
  589. void xenvif_carrier_off(struct xenvif *vif)
  590. {
  591. struct net_device *dev = vif->dev;
  592. rtnl_lock();
  593. if (test_and_clear_bit(VIF_STATUS_CONNECTED, &vif->status)) {
  594. netif_carrier_off(dev); /* discard queued packets */
  595. if (netif_running(dev))
  596. xenvif_down(vif);
  597. }
  598. rtnl_unlock();
  599. }
  600. void xenvif_disconnect_data(struct xenvif *vif)
  601. {
  602. struct xenvif_queue *queue = NULL;
  603. unsigned int num_queues = vif->num_queues;
  604. unsigned int queue_index;
  605. xenvif_carrier_off(vif);
  606. for (queue_index = 0; queue_index < num_queues; ++queue_index) {
  607. queue = &vif->queues[queue_index];
  608. netif_napi_del(&queue->napi);
  609. if (queue->task) {
  610. kthread_stop(queue->task);
  611. put_task_struct(queue->task);
  612. queue->task = NULL;
  613. }
  614. if (queue->dealloc_task) {
  615. kthread_stop(queue->dealloc_task);
  616. queue->dealloc_task = NULL;
  617. }
  618. if (queue->tx_irq) {
  619. if (queue->tx_irq == queue->rx_irq)
  620. unbind_from_irqhandler(queue->tx_irq, queue);
  621. else {
  622. unbind_from_irqhandler(queue->tx_irq, queue);
  623. unbind_from_irqhandler(queue->rx_irq, queue);
  624. }
  625. queue->tx_irq = 0;
  626. }
  627. xenvif_unmap_frontend_data_rings(queue);
  628. }
  629. xenvif_mcast_addr_list_free(vif);
  630. }
  631. void xenvif_disconnect_ctrl(struct xenvif *vif)
  632. {
  633. if (vif->ctrl_irq) {
  634. xenvif_deinit_hash(vif);
  635. unbind_from_irqhandler(vif->ctrl_irq, vif);
  636. vif->ctrl_irq = 0;
  637. }
  638. if (vif->ctrl.sring) {
  639. xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif),
  640. vif->ctrl.sring);
  641. vif->ctrl.sring = NULL;
  642. }
  643. }
  644. /* Reverse the relevant parts of xenvif_init_queue().
  645. * Used for queue teardown from xenvif_free(), and on the
  646. * error handling paths in xenbus.c:connect().
  647. */
  648. void xenvif_deinit_queue(struct xenvif_queue *queue)
  649. {
  650. gnttab_free_pages(MAX_PENDING_REQS, queue->mmap_pages);
  651. }
  652. void xenvif_free(struct xenvif *vif)
  653. {
  654. struct xenvif_queue *queues = vif->queues;
  655. unsigned int num_queues = vif->num_queues;
  656. unsigned int queue_index;
  657. unregister_netdev(vif->dev);
  658. free_netdev(vif->dev);
  659. for (queue_index = 0; queue_index < num_queues; ++queue_index)
  660. xenvif_deinit_queue(&queues[queue_index]);
  661. vfree(queues);
  662. module_put(THIS_MODULE);
  663. }