interface.c 20 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760
  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 <xen/events.h>
  36. #include <asm/xen/hypercall.h>
  37. #include <xen/balloon.h>
  38. #define XENVIF_QUEUE_LENGTH 32
  39. #define XENVIF_NAPI_WEIGHT 64
  40. static inline void xenvif_stop_queue(struct xenvif_queue *queue)
  41. {
  42. struct net_device *dev = queue->vif->dev;
  43. if (!queue->vif->can_queue)
  44. return;
  45. netif_tx_stop_queue(netdev_get_tx_queue(dev, queue->id));
  46. }
  47. int xenvif_schedulable(struct xenvif *vif)
  48. {
  49. return netif_running(vif->dev) && netif_carrier_ok(vif->dev);
  50. }
  51. static irqreturn_t xenvif_tx_interrupt(int irq, void *dev_id)
  52. {
  53. struct xenvif_queue *queue = dev_id;
  54. if (RING_HAS_UNCONSUMED_REQUESTS(&queue->tx))
  55. napi_schedule(&queue->napi);
  56. return IRQ_HANDLED;
  57. }
  58. int xenvif_poll(struct napi_struct *napi, int budget)
  59. {
  60. struct xenvif_queue *queue =
  61. container_of(napi, struct xenvif_queue, napi);
  62. int work_done;
  63. /* This vif is rogue, we pretend we've there is nothing to do
  64. * for this vif to deschedule it from NAPI. But this interface
  65. * will be turned off in thread context later.
  66. */
  67. if (unlikely(queue->vif->disabled)) {
  68. napi_complete(napi);
  69. return 0;
  70. }
  71. work_done = xenvif_tx_action(queue, budget);
  72. if (work_done < budget) {
  73. napi_complete(napi);
  74. xenvif_napi_schedule_or_enable_events(queue);
  75. }
  76. return work_done;
  77. }
  78. static irqreturn_t xenvif_rx_interrupt(int irq, void *dev_id)
  79. {
  80. struct xenvif_queue *queue = dev_id;
  81. xenvif_kick_thread(queue);
  82. return IRQ_HANDLED;
  83. }
  84. static irqreturn_t xenvif_interrupt(int irq, void *dev_id)
  85. {
  86. xenvif_tx_interrupt(irq, dev_id);
  87. xenvif_rx_interrupt(irq, dev_id);
  88. return IRQ_HANDLED;
  89. }
  90. int xenvif_queue_stopped(struct xenvif_queue *queue)
  91. {
  92. struct net_device *dev = queue->vif->dev;
  93. unsigned int id = queue->id;
  94. return netif_tx_queue_stopped(netdev_get_tx_queue(dev, id));
  95. }
  96. void xenvif_wake_queue(struct xenvif_queue *queue)
  97. {
  98. struct net_device *dev = queue->vif->dev;
  99. unsigned int id = queue->id;
  100. netif_tx_wake_queue(netdev_get_tx_queue(dev, id));
  101. }
  102. /* Callback to wake the queue and drain it on timeout */
  103. static void xenvif_wake_queue_callback(unsigned long data)
  104. {
  105. struct xenvif_queue *queue = (struct xenvif_queue *)data;
  106. if (xenvif_queue_stopped(queue)) {
  107. netdev_err(queue->vif->dev, "draining TX queue\n");
  108. queue->rx_queue_purge = true;
  109. xenvif_kick_thread(queue);
  110. xenvif_wake_queue(queue);
  111. }
  112. }
  113. static u16 xenvif_select_queue(struct net_device *dev, struct sk_buff *skb,
  114. void *accel_priv, select_queue_fallback_t fallback)
  115. {
  116. unsigned int num_queues = dev->real_num_tx_queues;
  117. u32 hash;
  118. u16 queue_index;
  119. /* First, check if there is only one queue to optimise the
  120. * single-queue or old frontend scenario.
  121. */
  122. if (num_queues == 1) {
  123. queue_index = 0;
  124. } else {
  125. /* Use skb_get_hash to obtain an L4 hash if available */
  126. hash = skb_get_hash(skb);
  127. queue_index = hash % num_queues;
  128. }
  129. return queue_index;
  130. }
  131. static int xenvif_start_xmit(struct sk_buff *skb, struct net_device *dev)
  132. {
  133. struct xenvif *vif = netdev_priv(dev);
  134. struct xenvif_queue *queue = NULL;
  135. unsigned int num_queues = dev->real_num_tx_queues;
  136. u16 index;
  137. int min_slots_needed;
  138. BUG_ON(skb->dev != dev);
  139. /* Drop the packet if queues are not set up */
  140. if (num_queues < 1)
  141. goto drop;
  142. /* Obtain the queue to be used to transmit this packet */
  143. index = skb_get_queue_mapping(skb);
  144. if (index >= num_queues) {
  145. pr_warn_ratelimited("Invalid queue %hu for packet on interface %s\n.",
  146. index, vif->dev->name);
  147. index %= num_queues;
  148. }
  149. queue = &vif->queues[index];
  150. /* Drop the packet if queue is not ready */
  151. if (queue->task == NULL ||
  152. queue->dealloc_task == NULL ||
  153. !xenvif_schedulable(vif))
  154. goto drop;
  155. /* At best we'll need one slot for the header and one for each
  156. * frag.
  157. */
  158. min_slots_needed = 1 + skb_shinfo(skb)->nr_frags;
  159. /* If the skb is GSO then we'll also need an extra slot for the
  160. * metadata.
  161. */
  162. if (skb_is_gso(skb))
  163. min_slots_needed++;
  164. /* If the skb can't possibly fit in the remaining slots
  165. * then turn off the queue to give the ring a chance to
  166. * drain.
  167. */
  168. if (!xenvif_rx_ring_slots_available(queue, min_slots_needed)) {
  169. queue->wake_queue.function = xenvif_wake_queue_callback;
  170. queue->wake_queue.data = (unsigned long)queue;
  171. xenvif_stop_queue(queue);
  172. mod_timer(&queue->wake_queue,
  173. jiffies + rx_drain_timeout_jiffies);
  174. }
  175. skb_queue_tail(&queue->rx_queue, skb);
  176. xenvif_kick_thread(queue);
  177. return NETDEV_TX_OK;
  178. drop:
  179. vif->dev->stats.tx_dropped++;
  180. dev_kfree_skb(skb);
  181. return NETDEV_TX_OK;
  182. }
  183. static struct net_device_stats *xenvif_get_stats(struct net_device *dev)
  184. {
  185. struct xenvif *vif = netdev_priv(dev);
  186. struct xenvif_queue *queue = NULL;
  187. unsigned int num_queues = dev->real_num_tx_queues;
  188. unsigned long rx_bytes = 0;
  189. unsigned long rx_packets = 0;
  190. unsigned long tx_bytes = 0;
  191. unsigned long tx_packets = 0;
  192. unsigned int index;
  193. if (vif->queues == NULL)
  194. goto out;
  195. /* Aggregate tx and rx stats from each queue */
  196. for (index = 0; index < num_queues; ++index) {
  197. queue = &vif->queues[index];
  198. rx_bytes += queue->stats.rx_bytes;
  199. rx_packets += queue->stats.rx_packets;
  200. tx_bytes += queue->stats.tx_bytes;
  201. tx_packets += queue->stats.tx_packets;
  202. }
  203. out:
  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->dev->real_num_tx_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->dev->real_num_tx_queues;
  228. unsigned int queue_index;
  229. for (queue_index = 0; queue_index < num_queues; ++queue_index) {
  230. queue = &vif->queues[queue_index];
  231. napi_disable(&queue->napi);
  232. disable_irq(queue->tx_irq);
  233. if (queue->tx_irq != queue->rx_irq)
  234. disable_irq(queue->rx_irq);
  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 (netif_carrier_ok(dev))
  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 (netif_carrier_ok(dev))
  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 | vif->gso_prefix_mask) & GSO_BIT(TCPV4))
  270. features &= ~NETIF_F_TSO;
  271. if (~(vif->gso_mask | vif->gso_prefix_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 = dev->real_num_tx_queues;
  324. int i;
  325. unsigned int queue_index;
  326. struct xenvif_stats *vif_stats;
  327. for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++) {
  328. unsigned long accum = 0;
  329. for (queue_index = 0; queue_index < num_queues; ++queue_index) {
  330. vif_stats = &vif->queues[queue_index].stats;
  331. accum += *(unsigned long *)(vif_stats + xenvif_stats[i].offset);
  332. }
  333. data[i] = accum;
  334. }
  335. }
  336. static void xenvif_get_strings(struct net_device *dev, u32 stringset, u8 * data)
  337. {
  338. int i;
  339. switch (stringset) {
  340. case ETH_SS_STATS:
  341. for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++)
  342. memcpy(data + i * ETH_GSTRING_LEN,
  343. xenvif_stats[i].name, ETH_GSTRING_LEN);
  344. break;
  345. }
  346. }
  347. static const struct ethtool_ops xenvif_ethtool_ops = {
  348. .get_link = ethtool_op_get_link,
  349. .get_sset_count = xenvif_get_sset_count,
  350. .get_ethtool_stats = xenvif_get_ethtool_stats,
  351. .get_strings = xenvif_get_strings,
  352. };
  353. static const struct net_device_ops xenvif_netdev_ops = {
  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. .ndo_select_queue = xenvif_select_queue,
  363. };
  364. struct xenvif *xenvif_alloc(struct device *parent, domid_t domid,
  365. unsigned int handle)
  366. {
  367. int err;
  368. struct net_device *dev;
  369. struct xenvif *vif;
  370. char name[IFNAMSIZ] = {};
  371. snprintf(name, IFNAMSIZ - 1, "vif%u.%u", domid, handle);
  372. /* Allocate a netdev with the max. supported number of queues.
  373. * When the guest selects the desired number, it will be updated
  374. * via netif_set_real_num_tx_queues().
  375. */
  376. dev = alloc_netdev_mq(sizeof(struct xenvif), name, ether_setup,
  377. xenvif_max_queues);
  378. if (dev == NULL) {
  379. pr_warn("Could not allocate netdev for %s\n", name);
  380. return ERR_PTR(-ENOMEM);
  381. }
  382. SET_NETDEV_DEV(dev, parent);
  383. vif = netdev_priv(dev);
  384. vif->domid = domid;
  385. vif->handle = handle;
  386. vif->can_sg = 1;
  387. vif->ip_csum = 1;
  388. vif->dev = dev;
  389. vif->disabled = false;
  390. /* Start out with no queues. The call below does not require
  391. * rtnl_lock() as it happens before register_netdev().
  392. */
  393. vif->queues = NULL;
  394. netif_set_real_num_tx_queues(dev, 0);
  395. dev->netdev_ops = &xenvif_netdev_ops;
  396. dev->hw_features = NETIF_F_SG |
  397. NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
  398. NETIF_F_TSO | NETIF_F_TSO6;
  399. dev->features = dev->hw_features | NETIF_F_RXCSUM;
  400. dev->ethtool_ops = &xenvif_ethtool_ops;
  401. dev->tx_queue_len = XENVIF_QUEUE_LENGTH;
  402. /*
  403. * Initialise a dummy MAC address. We choose the numerically
  404. * largest non-broadcast address to prevent the address getting
  405. * stolen by an Ethernet bridge for STP purposes.
  406. * (FE:FF:FF:FF:FF:FF)
  407. */
  408. memset(dev->dev_addr, 0xFF, ETH_ALEN);
  409. dev->dev_addr[0] &= ~0x01;
  410. netif_carrier_off(dev);
  411. err = register_netdev(dev);
  412. if (err) {
  413. netdev_warn(dev, "Could not register device: err=%d\n", err);
  414. free_netdev(dev);
  415. return ERR_PTR(err);
  416. }
  417. netdev_dbg(dev, "Successfully created xenvif\n");
  418. __module_get(THIS_MODULE);
  419. return vif;
  420. }
  421. int xenvif_init_queue(struct xenvif_queue *queue)
  422. {
  423. int err, i;
  424. queue->credit_bytes = queue->remaining_credit = ~0UL;
  425. queue->credit_usec = 0UL;
  426. init_timer(&queue->credit_timeout);
  427. queue->credit_window_start = get_jiffies_64();
  428. skb_queue_head_init(&queue->rx_queue);
  429. skb_queue_head_init(&queue->tx_queue);
  430. queue->pending_cons = 0;
  431. queue->pending_prod = MAX_PENDING_REQS;
  432. for (i = 0; i < MAX_PENDING_REQS; ++i)
  433. queue->pending_ring[i] = i;
  434. spin_lock_init(&queue->callback_lock);
  435. spin_lock_init(&queue->response_lock);
  436. /* If ballooning is disabled, this will consume real memory, so you
  437. * better enable it. The long term solution would be to use just a
  438. * bunch of valid page descriptors, without dependency on ballooning
  439. */
  440. err = alloc_xenballooned_pages(MAX_PENDING_REQS,
  441. queue->mmap_pages,
  442. false);
  443. if (err) {
  444. netdev_err(queue->vif->dev, "Could not reserve mmap_pages\n");
  445. return -ENOMEM;
  446. }
  447. for (i = 0; i < MAX_PENDING_REQS; i++) {
  448. queue->pending_tx_info[i].callback_struct = (struct ubuf_info)
  449. { .callback = xenvif_zerocopy_callback,
  450. .ctx = NULL,
  451. .desc = i };
  452. queue->grant_tx_handle[i] = NETBACK_INVALID_HANDLE;
  453. }
  454. init_timer(&queue->wake_queue);
  455. netif_napi_add(queue->vif->dev, &queue->napi, xenvif_poll,
  456. XENVIF_NAPI_WEIGHT);
  457. return 0;
  458. }
  459. void xenvif_carrier_on(struct xenvif *vif)
  460. {
  461. rtnl_lock();
  462. if (!vif->can_sg && vif->dev->mtu > ETH_DATA_LEN)
  463. dev_set_mtu(vif->dev, ETH_DATA_LEN);
  464. netdev_update_features(vif->dev);
  465. netif_carrier_on(vif->dev);
  466. if (netif_running(vif->dev))
  467. xenvif_up(vif);
  468. rtnl_unlock();
  469. }
  470. int xenvif_connect(struct xenvif_queue *queue, unsigned long tx_ring_ref,
  471. unsigned long rx_ring_ref, unsigned int tx_evtchn,
  472. unsigned int rx_evtchn)
  473. {
  474. struct task_struct *task;
  475. int err = -ENOMEM;
  476. BUG_ON(queue->tx_irq);
  477. BUG_ON(queue->task);
  478. BUG_ON(queue->dealloc_task);
  479. err = xenvif_map_frontend_rings(queue, tx_ring_ref, rx_ring_ref);
  480. if (err < 0)
  481. goto err;
  482. init_waitqueue_head(&queue->wq);
  483. init_waitqueue_head(&queue->dealloc_wq);
  484. if (tx_evtchn == rx_evtchn) {
  485. /* feature-split-event-channels == 0 */
  486. err = bind_interdomain_evtchn_to_irqhandler(
  487. queue->vif->domid, tx_evtchn, xenvif_interrupt, 0,
  488. queue->name, queue);
  489. if (err < 0)
  490. goto err_unmap;
  491. queue->tx_irq = queue->rx_irq = err;
  492. disable_irq(queue->tx_irq);
  493. } else {
  494. /* feature-split-event-channels == 1 */
  495. snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
  496. "%s-tx", queue->name);
  497. err = bind_interdomain_evtchn_to_irqhandler(
  498. queue->vif->domid, tx_evtchn, xenvif_tx_interrupt, 0,
  499. queue->tx_irq_name, queue);
  500. if (err < 0)
  501. goto err_unmap;
  502. queue->tx_irq = err;
  503. disable_irq(queue->tx_irq);
  504. snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
  505. "%s-rx", queue->name);
  506. err = bind_interdomain_evtchn_to_irqhandler(
  507. queue->vif->domid, rx_evtchn, xenvif_rx_interrupt, 0,
  508. queue->rx_irq_name, queue);
  509. if (err < 0)
  510. goto err_tx_unbind;
  511. queue->rx_irq = err;
  512. disable_irq(queue->rx_irq);
  513. }
  514. task = kthread_create(xenvif_kthread_guest_rx,
  515. (void *)queue, "%s-guest-rx", queue->name);
  516. if (IS_ERR(task)) {
  517. pr_warn("Could not allocate kthread for %s\n", queue->name);
  518. err = PTR_ERR(task);
  519. goto err_rx_unbind;
  520. }
  521. queue->task = task;
  522. task = kthread_create(xenvif_dealloc_kthread,
  523. (void *)queue, "%s-dealloc", queue->name);
  524. if (IS_ERR(task)) {
  525. pr_warn("Could not allocate kthread for %s\n", queue->name);
  526. err = PTR_ERR(task);
  527. goto err_rx_unbind;
  528. }
  529. queue->dealloc_task = task;
  530. wake_up_process(queue->task);
  531. wake_up_process(queue->dealloc_task);
  532. return 0;
  533. err_rx_unbind:
  534. unbind_from_irqhandler(queue->rx_irq, queue);
  535. queue->rx_irq = 0;
  536. err_tx_unbind:
  537. unbind_from_irqhandler(queue->tx_irq, queue);
  538. queue->tx_irq = 0;
  539. err_unmap:
  540. xenvif_unmap_frontend_rings(queue);
  541. err:
  542. module_put(THIS_MODULE);
  543. return err;
  544. }
  545. void xenvif_carrier_off(struct xenvif *vif)
  546. {
  547. struct net_device *dev = vif->dev;
  548. rtnl_lock();
  549. netif_carrier_off(dev); /* discard queued packets */
  550. if (netif_running(dev))
  551. xenvif_down(vif);
  552. rtnl_unlock();
  553. }
  554. static void xenvif_wait_unmap_timeout(struct xenvif_queue *queue,
  555. unsigned int worst_case_skb_lifetime)
  556. {
  557. int i, unmap_timeout = 0;
  558. for (i = 0; i < MAX_PENDING_REQS; ++i) {
  559. if (queue->grant_tx_handle[i] != NETBACK_INVALID_HANDLE) {
  560. unmap_timeout++;
  561. schedule_timeout(msecs_to_jiffies(1000));
  562. if (unmap_timeout > worst_case_skb_lifetime &&
  563. net_ratelimit())
  564. netdev_err(queue->vif->dev,
  565. "Page still granted! Index: %x\n",
  566. i);
  567. i = -1;
  568. }
  569. }
  570. }
  571. void xenvif_disconnect(struct xenvif *vif)
  572. {
  573. struct xenvif_queue *queue = NULL;
  574. unsigned int num_queues = vif->dev->real_num_tx_queues;
  575. unsigned int queue_index;
  576. if (netif_carrier_ok(vif->dev))
  577. xenvif_carrier_off(vif);
  578. for (queue_index = 0; queue_index < num_queues; ++queue_index) {
  579. queue = &vif->queues[queue_index];
  580. if (queue->task) {
  581. del_timer_sync(&queue->wake_queue);
  582. kthread_stop(queue->task);
  583. queue->task = NULL;
  584. }
  585. if (queue->dealloc_task) {
  586. kthread_stop(queue->dealloc_task);
  587. queue->dealloc_task = NULL;
  588. }
  589. if (queue->tx_irq) {
  590. if (queue->tx_irq == queue->rx_irq)
  591. unbind_from_irqhandler(queue->tx_irq, queue);
  592. else {
  593. unbind_from_irqhandler(queue->tx_irq, queue);
  594. unbind_from_irqhandler(queue->rx_irq, queue);
  595. }
  596. queue->tx_irq = 0;
  597. }
  598. xenvif_unmap_frontend_rings(queue);
  599. }
  600. }
  601. /* Reverse the relevant parts of xenvif_init_queue().
  602. * Used for queue teardown from xenvif_free(), and on the
  603. * error handling paths in xenbus.c:connect().
  604. */
  605. void xenvif_deinit_queue(struct xenvif_queue *queue)
  606. {
  607. free_xenballooned_pages(MAX_PENDING_REQS, queue->mmap_pages);
  608. netif_napi_del(&queue->napi);
  609. }
  610. void xenvif_free(struct xenvif *vif)
  611. {
  612. struct xenvif_queue *queue = NULL;
  613. unsigned int num_queues = vif->dev->real_num_tx_queues;
  614. unsigned int queue_index;
  615. /* Here we want to avoid timeout messages if an skb can be legitimately
  616. * stuck somewhere else. Realistically this could be an another vif's
  617. * internal or QDisc queue. That another vif also has this
  618. * rx_drain_timeout_msecs timeout, but the timer only ditches the
  619. * internal queue. After that, the QDisc queue can put in worst case
  620. * XEN_NETIF_RX_RING_SIZE / MAX_SKB_FRAGS skbs into that another vif's
  621. * internal queue, so we need several rounds of such timeouts until we
  622. * can be sure that no another vif should have skb's from us. We are
  623. * not sending more skb's, so newly stuck packets are not interesting
  624. * for us here.
  625. */
  626. unsigned int worst_case_skb_lifetime = (rx_drain_timeout_msecs/1000) *
  627. DIV_ROUND_UP(XENVIF_QUEUE_LENGTH, (XEN_NETIF_RX_RING_SIZE / MAX_SKB_FRAGS));
  628. unregister_netdev(vif->dev);
  629. for (queue_index = 0; queue_index < num_queues; ++queue_index) {
  630. queue = &vif->queues[queue_index];
  631. xenvif_wait_unmap_timeout(queue, worst_case_skb_lifetime);
  632. xenvif_deinit_queue(queue);
  633. }
  634. /* Free the array of queues. The call below does not require
  635. * rtnl_lock() because it happens after unregister_netdev().
  636. */
  637. netif_set_real_num_tx_queues(vif->dev, 0);
  638. vfree(vif->queues);
  639. vif->queues = NULL;
  640. free_netdev(vif->dev);
  641. module_put(THIS_MODULE);
  642. }