interface.c 19 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740
  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. static inline void xenvif_stop_queue(struct xenvif_queue *queue)
  42. {
  43. struct net_device *dev = queue->vif->dev;
  44. if (!queue->vif->can_queue)
  45. return;
  46. netif_tx_stop_queue(netdev_get_tx_queue(dev, queue->id));
  47. }
  48. int xenvif_schedulable(struct xenvif *vif)
  49. {
  50. return netif_running(vif->dev) &&
  51. test_bit(VIF_STATUS_CONNECTED, &vif->status);
  52. }
  53. static irqreturn_t xenvif_tx_interrupt(int irq, void *dev_id)
  54. {
  55. struct xenvif_queue *queue = dev_id;
  56. if (RING_HAS_UNCONSUMED_REQUESTS(&queue->tx))
  57. napi_schedule(&queue->napi);
  58. return IRQ_HANDLED;
  59. }
  60. int xenvif_poll(struct napi_struct *napi, int budget)
  61. {
  62. struct xenvif_queue *queue =
  63. container_of(napi, struct xenvif_queue, napi);
  64. int work_done;
  65. /* This vif is rogue, we pretend we've there is nothing to do
  66. * for this vif to deschedule it from NAPI. But this interface
  67. * will be turned off in thread context later.
  68. */
  69. if (unlikely(queue->vif->disabled)) {
  70. napi_complete(napi);
  71. return 0;
  72. }
  73. work_done = xenvif_tx_action(queue, budget);
  74. if (work_done < budget) {
  75. napi_complete(napi);
  76. xenvif_napi_schedule_or_enable_events(queue);
  77. }
  78. return work_done;
  79. }
  80. static irqreturn_t xenvif_rx_interrupt(int irq, void *dev_id)
  81. {
  82. struct xenvif_queue *queue = dev_id;
  83. struct netdev_queue *net_queue =
  84. netdev_get_tx_queue(queue->vif->dev, queue->id);
  85. /* QUEUE_STATUS_RX_PURGE_EVENT is only set if either QDisc was off OR
  86. * the carrier went down and this queue was previously blocked
  87. */
  88. if (unlikely(netif_tx_queue_stopped(net_queue) ||
  89. (!netif_carrier_ok(queue->vif->dev) &&
  90. test_bit(QUEUE_STATUS_RX_STALLED, &queue->status))))
  91. set_bit(QUEUE_STATUS_RX_PURGE_EVENT, &queue->status);
  92. xenvif_kick_thread(queue);
  93. return IRQ_HANDLED;
  94. }
  95. irqreturn_t xenvif_interrupt(int irq, void *dev_id)
  96. {
  97. xenvif_tx_interrupt(irq, dev_id);
  98. xenvif_rx_interrupt(irq, dev_id);
  99. return IRQ_HANDLED;
  100. }
  101. int xenvif_queue_stopped(struct xenvif_queue *queue)
  102. {
  103. struct net_device *dev = queue->vif->dev;
  104. unsigned int id = queue->id;
  105. return netif_tx_queue_stopped(netdev_get_tx_queue(dev, id));
  106. }
  107. void xenvif_wake_queue(struct xenvif_queue *queue)
  108. {
  109. struct net_device *dev = queue->vif->dev;
  110. unsigned int id = queue->id;
  111. netif_tx_wake_queue(netdev_get_tx_queue(dev, id));
  112. }
  113. /* Callback to wake the queue's thread and turn the carrier off on timeout */
  114. static void xenvif_rx_stalled(unsigned long data)
  115. {
  116. struct xenvif_queue *queue = (struct xenvif_queue *)data;
  117. if (xenvif_queue_stopped(queue)) {
  118. set_bit(QUEUE_STATUS_RX_PURGE_EVENT, &queue->status);
  119. xenvif_kick_thread(queue);
  120. }
  121. }
  122. static int xenvif_start_xmit(struct sk_buff *skb, struct net_device *dev)
  123. {
  124. struct xenvif *vif = netdev_priv(dev);
  125. struct xenvif_queue *queue = NULL;
  126. unsigned int num_queues = vif->num_queues;
  127. u16 index;
  128. int min_slots_needed;
  129. BUG_ON(skb->dev != dev);
  130. /* Drop the packet if queues are not set up */
  131. if (num_queues < 1)
  132. goto drop;
  133. /* Obtain the queue to be used to transmit this packet */
  134. index = skb_get_queue_mapping(skb);
  135. if (index >= num_queues) {
  136. pr_warn_ratelimited("Invalid queue %hu for packet on interface %s\n.",
  137. index, vif->dev->name);
  138. index %= num_queues;
  139. }
  140. queue = &vif->queues[index];
  141. /* Drop the packet if queue is not ready */
  142. if (queue->task == NULL ||
  143. queue->dealloc_task == NULL ||
  144. !xenvif_schedulable(vif))
  145. goto drop;
  146. /* At best we'll need one slot for the header and one for each
  147. * frag.
  148. */
  149. min_slots_needed = 1 + skb_shinfo(skb)->nr_frags;
  150. /* If the skb is GSO then we'll also need an extra slot for the
  151. * metadata.
  152. */
  153. if (skb_is_gso(skb))
  154. min_slots_needed++;
  155. /* If the skb can't possibly fit in the remaining slots
  156. * then turn off the queue to give the ring a chance to
  157. * drain.
  158. */
  159. if (!xenvif_rx_ring_slots_available(queue, min_slots_needed)) {
  160. queue->rx_stalled.function = xenvif_rx_stalled;
  161. queue->rx_stalled.data = (unsigned long)queue;
  162. xenvif_stop_queue(queue);
  163. mod_timer(&queue->rx_stalled,
  164. jiffies + rx_drain_timeout_jiffies);
  165. }
  166. skb_queue_tail(&queue->rx_queue, skb);
  167. xenvif_kick_thread(queue);
  168. return NETDEV_TX_OK;
  169. drop:
  170. vif->dev->stats.tx_dropped++;
  171. dev_kfree_skb(skb);
  172. return NETDEV_TX_OK;
  173. }
  174. static struct net_device_stats *xenvif_get_stats(struct net_device *dev)
  175. {
  176. struct xenvif *vif = netdev_priv(dev);
  177. struct xenvif_queue *queue = NULL;
  178. unsigned int num_queues = vif->num_queues;
  179. unsigned long rx_bytes = 0;
  180. unsigned long rx_packets = 0;
  181. unsigned long tx_bytes = 0;
  182. unsigned long tx_packets = 0;
  183. unsigned int index;
  184. if (vif->queues == NULL)
  185. goto out;
  186. /* Aggregate tx and rx stats from each queue */
  187. for (index = 0; index < num_queues; ++index) {
  188. queue = &vif->queues[index];
  189. rx_bytes += queue->stats.rx_bytes;
  190. rx_packets += queue->stats.rx_packets;
  191. tx_bytes += queue->stats.tx_bytes;
  192. tx_packets += queue->stats.tx_packets;
  193. }
  194. out:
  195. vif->dev->stats.rx_bytes = rx_bytes;
  196. vif->dev->stats.rx_packets = rx_packets;
  197. vif->dev->stats.tx_bytes = tx_bytes;
  198. vif->dev->stats.tx_packets = tx_packets;
  199. return &vif->dev->stats;
  200. }
  201. static void xenvif_up(struct xenvif *vif)
  202. {
  203. struct xenvif_queue *queue = NULL;
  204. unsigned int num_queues = vif->num_queues;
  205. unsigned int queue_index;
  206. for (queue_index = 0; queue_index < num_queues; ++queue_index) {
  207. queue = &vif->queues[queue_index];
  208. napi_enable(&queue->napi);
  209. enable_irq(queue->tx_irq);
  210. if (queue->tx_irq != queue->rx_irq)
  211. enable_irq(queue->rx_irq);
  212. xenvif_napi_schedule_or_enable_events(queue);
  213. }
  214. }
  215. static void xenvif_down(struct xenvif *vif)
  216. {
  217. struct xenvif_queue *queue = NULL;
  218. unsigned int num_queues = vif->num_queues;
  219. unsigned int queue_index;
  220. for (queue_index = 0; queue_index < num_queues; ++queue_index) {
  221. queue = &vif->queues[queue_index];
  222. napi_disable(&queue->napi);
  223. disable_irq(queue->tx_irq);
  224. if (queue->tx_irq != queue->rx_irq)
  225. disable_irq(queue->rx_irq);
  226. del_timer_sync(&queue->credit_timeout);
  227. }
  228. }
  229. static int xenvif_open(struct net_device *dev)
  230. {
  231. struct xenvif *vif = netdev_priv(dev);
  232. if (test_bit(VIF_STATUS_CONNECTED, &vif->status))
  233. xenvif_up(vif);
  234. netif_tx_start_all_queues(dev);
  235. return 0;
  236. }
  237. static int xenvif_close(struct net_device *dev)
  238. {
  239. struct xenvif *vif = netdev_priv(dev);
  240. if (test_bit(VIF_STATUS_CONNECTED, &vif->status))
  241. xenvif_down(vif);
  242. netif_tx_stop_all_queues(dev);
  243. return 0;
  244. }
  245. static int xenvif_change_mtu(struct net_device *dev, int mtu)
  246. {
  247. struct xenvif *vif = netdev_priv(dev);
  248. int max = vif->can_sg ? 65535 - VLAN_ETH_HLEN : ETH_DATA_LEN;
  249. if (mtu > max)
  250. return -EINVAL;
  251. dev->mtu = mtu;
  252. return 0;
  253. }
  254. static netdev_features_t xenvif_fix_features(struct net_device *dev,
  255. netdev_features_t features)
  256. {
  257. struct xenvif *vif = netdev_priv(dev);
  258. if (!vif->can_sg)
  259. features &= ~NETIF_F_SG;
  260. if (~(vif->gso_mask | vif->gso_prefix_mask) & GSO_BIT(TCPV4))
  261. features &= ~NETIF_F_TSO;
  262. if (~(vif->gso_mask | vif->gso_prefix_mask) & GSO_BIT(TCPV6))
  263. features &= ~NETIF_F_TSO6;
  264. if (!vif->ip_csum)
  265. features &= ~NETIF_F_IP_CSUM;
  266. if (!vif->ipv6_csum)
  267. features &= ~NETIF_F_IPV6_CSUM;
  268. return features;
  269. }
  270. static const struct xenvif_stat {
  271. char name[ETH_GSTRING_LEN];
  272. u16 offset;
  273. } xenvif_stats[] = {
  274. {
  275. "rx_gso_checksum_fixup",
  276. offsetof(struct xenvif_stats, rx_gso_checksum_fixup)
  277. },
  278. /* If (sent != success + fail), there are probably packets never
  279. * freed up properly!
  280. */
  281. {
  282. "tx_zerocopy_sent",
  283. offsetof(struct xenvif_stats, tx_zerocopy_sent),
  284. },
  285. {
  286. "tx_zerocopy_success",
  287. offsetof(struct xenvif_stats, tx_zerocopy_success),
  288. },
  289. {
  290. "tx_zerocopy_fail",
  291. offsetof(struct xenvif_stats, tx_zerocopy_fail)
  292. },
  293. /* Number of packets exceeding MAX_SKB_FRAG slots. You should use
  294. * a guest with the same MAX_SKB_FRAG
  295. */
  296. {
  297. "tx_frag_overflow",
  298. offsetof(struct xenvif_stats, tx_frag_overflow)
  299. },
  300. };
  301. static int xenvif_get_sset_count(struct net_device *dev, int string_set)
  302. {
  303. switch (string_set) {
  304. case ETH_SS_STATS:
  305. return ARRAY_SIZE(xenvif_stats);
  306. default:
  307. return -EINVAL;
  308. }
  309. }
  310. static void xenvif_get_ethtool_stats(struct net_device *dev,
  311. struct ethtool_stats *stats, u64 * data)
  312. {
  313. struct xenvif *vif = netdev_priv(dev);
  314. unsigned int num_queues = vif->num_queues;
  315. int i;
  316. unsigned int queue_index;
  317. struct xenvif_stats *vif_stats;
  318. for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++) {
  319. unsigned long accum = 0;
  320. for (queue_index = 0; queue_index < num_queues; ++queue_index) {
  321. vif_stats = &vif->queues[queue_index].stats;
  322. accum += *(unsigned long *)(vif_stats + xenvif_stats[i].offset);
  323. }
  324. data[i] = accum;
  325. }
  326. }
  327. static void xenvif_get_strings(struct net_device *dev, u32 stringset, u8 * data)
  328. {
  329. int i;
  330. switch (stringset) {
  331. case ETH_SS_STATS:
  332. for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++)
  333. memcpy(data + i * ETH_GSTRING_LEN,
  334. xenvif_stats[i].name, ETH_GSTRING_LEN);
  335. break;
  336. }
  337. }
  338. static const struct ethtool_ops xenvif_ethtool_ops = {
  339. .get_link = ethtool_op_get_link,
  340. .get_sset_count = xenvif_get_sset_count,
  341. .get_ethtool_stats = xenvif_get_ethtool_stats,
  342. .get_strings = xenvif_get_strings,
  343. };
  344. static const struct net_device_ops xenvif_netdev_ops = {
  345. .ndo_start_xmit = xenvif_start_xmit,
  346. .ndo_get_stats = xenvif_get_stats,
  347. .ndo_open = xenvif_open,
  348. .ndo_stop = xenvif_close,
  349. .ndo_change_mtu = xenvif_change_mtu,
  350. .ndo_fix_features = xenvif_fix_features,
  351. .ndo_set_mac_address = eth_mac_addr,
  352. .ndo_validate_addr = eth_validate_addr,
  353. };
  354. struct xenvif *xenvif_alloc(struct device *parent, domid_t domid,
  355. unsigned int handle)
  356. {
  357. int err;
  358. struct net_device *dev;
  359. struct xenvif *vif;
  360. char name[IFNAMSIZ] = {};
  361. snprintf(name, IFNAMSIZ - 1, "vif%u.%u", domid, handle);
  362. /* Allocate a netdev with the max. supported number of queues.
  363. * When the guest selects the desired number, it will be updated
  364. * via netif_set_real_num_*_queues().
  365. */
  366. dev = alloc_netdev_mq(sizeof(struct xenvif), name, NET_NAME_UNKNOWN,
  367. ether_setup, xenvif_max_queues);
  368. if (dev == NULL) {
  369. pr_warn("Could not allocate netdev for %s\n", name);
  370. return ERR_PTR(-ENOMEM);
  371. }
  372. SET_NETDEV_DEV(dev, parent);
  373. vif = netdev_priv(dev);
  374. vif->domid = domid;
  375. vif->handle = handle;
  376. vif->can_sg = 1;
  377. vif->ip_csum = 1;
  378. vif->dev = dev;
  379. vif->disabled = false;
  380. /* Start out with no queues. */
  381. vif->queues = NULL;
  382. vif->num_queues = 0;
  383. dev->netdev_ops = &xenvif_netdev_ops;
  384. dev->hw_features = NETIF_F_SG |
  385. NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
  386. NETIF_F_TSO | NETIF_F_TSO6;
  387. dev->features = dev->hw_features | NETIF_F_RXCSUM;
  388. dev->ethtool_ops = &xenvif_ethtool_ops;
  389. dev->tx_queue_len = XENVIF_QUEUE_LENGTH;
  390. /*
  391. * Initialise a dummy MAC address. We choose the numerically
  392. * largest non-broadcast address to prevent the address getting
  393. * stolen by an Ethernet bridge for STP purposes.
  394. * (FE:FF:FF:FF:FF:FF)
  395. */
  396. memset(dev->dev_addr, 0xFF, ETH_ALEN);
  397. dev->dev_addr[0] &= ~0x01;
  398. netif_carrier_off(dev);
  399. err = register_netdev(dev);
  400. if (err) {
  401. netdev_warn(dev, "Could not register device: err=%d\n", err);
  402. free_netdev(dev);
  403. return ERR_PTR(err);
  404. }
  405. netdev_dbg(dev, "Successfully created xenvif\n");
  406. __module_get(THIS_MODULE);
  407. return vif;
  408. }
  409. int xenvif_init_queue(struct xenvif_queue *queue)
  410. {
  411. int err, i;
  412. queue->credit_bytes = queue->remaining_credit = ~0UL;
  413. queue->credit_usec = 0UL;
  414. init_timer(&queue->credit_timeout);
  415. queue->credit_window_start = get_jiffies_64();
  416. skb_queue_head_init(&queue->rx_queue);
  417. skb_queue_head_init(&queue->tx_queue);
  418. queue->pending_cons = 0;
  419. queue->pending_prod = MAX_PENDING_REQS;
  420. for (i = 0; i < MAX_PENDING_REQS; ++i)
  421. queue->pending_ring[i] = i;
  422. spin_lock_init(&queue->callback_lock);
  423. spin_lock_init(&queue->response_lock);
  424. /* If ballooning is disabled, this will consume real memory, so you
  425. * better enable it. The long term solution would be to use just a
  426. * bunch of valid page descriptors, without dependency on ballooning
  427. */
  428. err = alloc_xenballooned_pages(MAX_PENDING_REQS,
  429. queue->mmap_pages,
  430. false);
  431. if (err) {
  432. netdev_err(queue->vif->dev, "Could not reserve mmap_pages\n");
  433. return -ENOMEM;
  434. }
  435. for (i = 0; i < MAX_PENDING_REQS; i++) {
  436. queue->pending_tx_info[i].callback_struct = (struct ubuf_info)
  437. { .callback = xenvif_zerocopy_callback,
  438. .ctx = NULL,
  439. .desc = i };
  440. queue->grant_tx_handle[i] = NETBACK_INVALID_HANDLE;
  441. }
  442. init_timer(&queue->rx_stalled);
  443. netif_napi_add(queue->vif->dev, &queue->napi, xenvif_poll,
  444. XENVIF_NAPI_WEIGHT);
  445. return 0;
  446. }
  447. void xenvif_carrier_on(struct xenvif *vif)
  448. {
  449. rtnl_lock();
  450. if (!vif->can_sg && vif->dev->mtu > ETH_DATA_LEN)
  451. dev_set_mtu(vif->dev, ETH_DATA_LEN);
  452. netdev_update_features(vif->dev);
  453. set_bit(VIF_STATUS_CONNECTED, &vif->status);
  454. netif_carrier_on(vif->dev);
  455. if (netif_running(vif->dev))
  456. xenvif_up(vif);
  457. rtnl_unlock();
  458. }
  459. int xenvif_connect(struct xenvif_queue *queue, unsigned long tx_ring_ref,
  460. unsigned long rx_ring_ref, unsigned int tx_evtchn,
  461. unsigned int rx_evtchn)
  462. {
  463. struct task_struct *task;
  464. int err = -ENOMEM;
  465. BUG_ON(queue->tx_irq);
  466. BUG_ON(queue->task);
  467. BUG_ON(queue->dealloc_task);
  468. err = xenvif_map_frontend_rings(queue, tx_ring_ref, rx_ring_ref);
  469. if (err < 0)
  470. goto err;
  471. init_waitqueue_head(&queue->wq);
  472. init_waitqueue_head(&queue->dealloc_wq);
  473. if (tx_evtchn == rx_evtchn) {
  474. /* feature-split-event-channels == 0 */
  475. err = bind_interdomain_evtchn_to_irqhandler(
  476. queue->vif->domid, tx_evtchn, xenvif_interrupt, 0,
  477. queue->name, queue);
  478. if (err < 0)
  479. goto err_unmap;
  480. queue->tx_irq = queue->rx_irq = err;
  481. disable_irq(queue->tx_irq);
  482. } else {
  483. /* feature-split-event-channels == 1 */
  484. snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
  485. "%s-tx", queue->name);
  486. err = bind_interdomain_evtchn_to_irqhandler(
  487. queue->vif->domid, tx_evtchn, xenvif_tx_interrupt, 0,
  488. queue->tx_irq_name, queue);
  489. if (err < 0)
  490. goto err_unmap;
  491. queue->tx_irq = err;
  492. disable_irq(queue->tx_irq);
  493. snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
  494. "%s-rx", queue->name);
  495. err = bind_interdomain_evtchn_to_irqhandler(
  496. queue->vif->domid, rx_evtchn, xenvif_rx_interrupt, 0,
  497. queue->rx_irq_name, queue);
  498. if (err < 0)
  499. goto err_tx_unbind;
  500. queue->rx_irq = err;
  501. disable_irq(queue->rx_irq);
  502. }
  503. task = kthread_create(xenvif_kthread_guest_rx,
  504. (void *)queue, "%s-guest-rx", queue->name);
  505. if (IS_ERR(task)) {
  506. pr_warn("Could not allocate kthread for %s\n", queue->name);
  507. err = PTR_ERR(task);
  508. goto err_rx_unbind;
  509. }
  510. queue->task = task;
  511. task = kthread_create(xenvif_dealloc_kthread,
  512. (void *)queue, "%s-dealloc", queue->name);
  513. if (IS_ERR(task)) {
  514. pr_warn("Could not allocate kthread for %s\n", queue->name);
  515. err = PTR_ERR(task);
  516. goto err_rx_unbind;
  517. }
  518. queue->dealloc_task = task;
  519. wake_up_process(queue->task);
  520. wake_up_process(queue->dealloc_task);
  521. return 0;
  522. err_rx_unbind:
  523. unbind_from_irqhandler(queue->rx_irq, queue);
  524. queue->rx_irq = 0;
  525. err_tx_unbind:
  526. unbind_from_irqhandler(queue->tx_irq, queue);
  527. queue->tx_irq = 0;
  528. err_unmap:
  529. xenvif_unmap_frontend_rings(queue);
  530. err:
  531. module_put(THIS_MODULE);
  532. return err;
  533. }
  534. void xenvif_carrier_off(struct xenvif *vif)
  535. {
  536. struct net_device *dev = vif->dev;
  537. rtnl_lock();
  538. if (test_and_clear_bit(VIF_STATUS_CONNECTED, &vif->status)) {
  539. netif_carrier_off(dev); /* discard queued packets */
  540. if (netif_running(dev))
  541. xenvif_down(vif);
  542. }
  543. rtnl_unlock();
  544. }
  545. static void xenvif_wait_unmap_timeout(struct xenvif_queue *queue,
  546. unsigned int worst_case_skb_lifetime)
  547. {
  548. int i, unmap_timeout = 0;
  549. for (i = 0; i < MAX_PENDING_REQS; ++i) {
  550. if (queue->grant_tx_handle[i] != NETBACK_INVALID_HANDLE) {
  551. unmap_timeout++;
  552. schedule_timeout(msecs_to_jiffies(1000));
  553. if (unmap_timeout > worst_case_skb_lifetime &&
  554. net_ratelimit())
  555. netdev_err(queue->vif->dev,
  556. "Page still granted! Index: %x\n",
  557. i);
  558. i = -1;
  559. }
  560. }
  561. }
  562. void xenvif_disconnect(struct xenvif *vif)
  563. {
  564. struct xenvif_queue *queue = NULL;
  565. unsigned int num_queues = vif->num_queues;
  566. unsigned int queue_index;
  567. xenvif_carrier_off(vif);
  568. for (queue_index = 0; queue_index < num_queues; ++queue_index) {
  569. queue = &vif->queues[queue_index];
  570. if (queue->task) {
  571. del_timer_sync(&queue->rx_stalled);
  572. kthread_stop(queue->task);
  573. queue->task = NULL;
  574. }
  575. if (queue->dealloc_task) {
  576. kthread_stop(queue->dealloc_task);
  577. queue->dealloc_task = NULL;
  578. }
  579. if (queue->tx_irq) {
  580. if (queue->tx_irq == queue->rx_irq)
  581. unbind_from_irqhandler(queue->tx_irq, queue);
  582. else {
  583. unbind_from_irqhandler(queue->tx_irq, queue);
  584. unbind_from_irqhandler(queue->rx_irq, queue);
  585. }
  586. queue->tx_irq = 0;
  587. }
  588. xenvif_unmap_frontend_rings(queue);
  589. }
  590. }
  591. /* Reverse the relevant parts of xenvif_init_queue().
  592. * Used for queue teardown from xenvif_free(), and on the
  593. * error handling paths in xenbus.c:connect().
  594. */
  595. void xenvif_deinit_queue(struct xenvif_queue *queue)
  596. {
  597. free_xenballooned_pages(MAX_PENDING_REQS, queue->mmap_pages);
  598. netif_napi_del(&queue->napi);
  599. }
  600. void xenvif_free(struct xenvif *vif)
  601. {
  602. struct xenvif_queue *queue = NULL;
  603. unsigned int num_queues = vif->num_queues;
  604. unsigned int queue_index;
  605. /* Here we want to avoid timeout messages if an skb can be legitimately
  606. * stuck somewhere else. Realistically this could be an another vif's
  607. * internal or QDisc queue. That another vif also has this
  608. * rx_drain_timeout_msecs timeout, so give it time to drain out.
  609. * Although if that other guest wakes up just before its timeout happens
  610. * and takes only one skb from QDisc, it can hold onto other skbs for a
  611. * longer period.
  612. */
  613. unsigned int worst_case_skb_lifetime = (rx_drain_timeout_msecs/1000);
  614. unregister_netdev(vif->dev);
  615. for (queue_index = 0; queue_index < num_queues; ++queue_index) {
  616. queue = &vif->queues[queue_index];
  617. xenvif_wait_unmap_timeout(queue, worst_case_skb_lifetime);
  618. xenvif_deinit_queue(queue);
  619. }
  620. vfree(vif->queues);
  621. vif->queues = NULL;
  622. vif->num_queues = 0;
  623. free_netdev(vif->dev);
  624. module_put(THIS_MODULE);
  625. }