interface.c 20 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797
  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/sched/task.h>
  33. #include <linux/ethtool.h>
  34. #include <linux/rtnetlink.h>
  35. #include <linux/if_vlan.h>
  36. #include <linux/vmalloc.h>
  37. #include <xen/events.h>
  38. #include <asm/xen/hypercall.h>
  39. #include <xen/balloon.h>
  40. #define XENVIF_QUEUE_LENGTH 32
  41. #define XENVIF_NAPI_WEIGHT 64
  42. /* Number of bytes allowed on the internal guest Rx queue. */
  43. #define XENVIF_RX_QUEUE_BYTES (XEN_NETIF_RX_RING_SIZE/2 * PAGE_SIZE)
  44. /* This function is used to set SKBTX_DEV_ZEROCOPY as well as
  45. * increasing the inflight counter. We need to increase the inflight
  46. * counter because core driver calls into xenvif_zerocopy_callback
  47. * which calls xenvif_skb_zerocopy_complete.
  48. */
  49. void xenvif_skb_zerocopy_prepare(struct xenvif_queue *queue,
  50. struct sk_buff *skb)
  51. {
  52. skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
  53. atomic_inc(&queue->inflight_packets);
  54. }
  55. void xenvif_skb_zerocopy_complete(struct xenvif_queue *queue)
  56. {
  57. atomic_dec(&queue->inflight_packets);
  58. /* Wake the dealloc thread _after_ decrementing inflight_packets so
  59. * that if kthread_stop() has already been called, the dealloc thread
  60. * does not wait forever with nothing to wake it.
  61. */
  62. wake_up(&queue->dealloc_wq);
  63. }
  64. int xenvif_schedulable(struct xenvif *vif)
  65. {
  66. return netif_running(vif->dev) &&
  67. test_bit(VIF_STATUS_CONNECTED, &vif->status) &&
  68. !vif->disabled;
  69. }
  70. static irqreturn_t xenvif_tx_interrupt(int irq, void *dev_id)
  71. {
  72. struct xenvif_queue *queue = dev_id;
  73. if (RING_HAS_UNCONSUMED_REQUESTS(&queue->tx))
  74. napi_schedule(&queue->napi);
  75. return IRQ_HANDLED;
  76. }
  77. static int xenvif_poll(struct napi_struct *napi, int budget)
  78. {
  79. struct xenvif_queue *queue =
  80. container_of(napi, struct xenvif_queue, napi);
  81. int work_done;
  82. /* This vif is rogue, we pretend we've there is nothing to do
  83. * for this vif to deschedule it from NAPI. But this interface
  84. * will be turned off in thread context later.
  85. */
  86. if (unlikely(queue->vif->disabled)) {
  87. napi_complete(napi);
  88. return 0;
  89. }
  90. work_done = xenvif_tx_action(queue, budget);
  91. if (work_done < budget) {
  92. napi_complete_done(napi, work_done);
  93. xenvif_napi_schedule_or_enable_events(queue);
  94. }
  95. return work_done;
  96. }
  97. static irqreturn_t xenvif_rx_interrupt(int irq, void *dev_id)
  98. {
  99. struct xenvif_queue *queue = dev_id;
  100. xenvif_kick_thread(queue);
  101. return IRQ_HANDLED;
  102. }
  103. irqreturn_t xenvif_interrupt(int irq, void *dev_id)
  104. {
  105. xenvif_tx_interrupt(irq, dev_id);
  106. xenvif_rx_interrupt(irq, dev_id);
  107. return IRQ_HANDLED;
  108. }
  109. int xenvif_queue_stopped(struct xenvif_queue *queue)
  110. {
  111. struct net_device *dev = queue->vif->dev;
  112. unsigned int id = queue->id;
  113. return netif_tx_queue_stopped(netdev_get_tx_queue(dev, id));
  114. }
  115. void xenvif_wake_queue(struct xenvif_queue *queue)
  116. {
  117. struct net_device *dev = queue->vif->dev;
  118. unsigned int id = queue->id;
  119. netif_tx_wake_queue(netdev_get_tx_queue(dev, id));
  120. }
  121. static u16 xenvif_select_queue(struct net_device *dev, struct sk_buff *skb,
  122. void *accel_priv,
  123. select_queue_fallback_t fallback)
  124. {
  125. struct xenvif *vif = netdev_priv(dev);
  126. unsigned int size = vif->hash.size;
  127. if (vif->hash.alg == XEN_NETIF_CTRL_HASH_ALGORITHM_NONE)
  128. return fallback(dev, skb) % dev->real_num_tx_queues;
  129. xenvif_set_skb_hash(vif, skb);
  130. if (size == 0)
  131. return skb_get_hash_raw(skb) % dev->real_num_tx_queues;
  132. return vif->hash.mapping[skb_get_hash_raw(skb) % size];
  133. }
  134. static int xenvif_start_xmit(struct sk_buff *skb, struct net_device *dev)
  135. {
  136. struct xenvif *vif = netdev_priv(dev);
  137. struct xenvif_queue *queue = NULL;
  138. unsigned int num_queues = vif->num_queues;
  139. u16 index;
  140. struct xenvif_rx_cb *cb;
  141. BUG_ON(skb->dev != dev);
  142. /* Drop the packet if queues are not set up */
  143. if (num_queues < 1)
  144. goto drop;
  145. /* Obtain the queue to be used to transmit this packet */
  146. index = skb_get_queue_mapping(skb);
  147. if (index >= num_queues) {
  148. pr_warn_ratelimited("Invalid queue %hu for packet on interface %s\n.",
  149. index, vif->dev->name);
  150. index %= num_queues;
  151. }
  152. queue = &vif->queues[index];
  153. /* Drop the packet if queue is not ready */
  154. if (queue->task == NULL ||
  155. queue->dealloc_task == NULL ||
  156. !xenvif_schedulable(vif))
  157. goto drop;
  158. if (vif->multicast_control && skb->pkt_type == PACKET_MULTICAST) {
  159. struct ethhdr *eth = (struct ethhdr *)skb->data;
  160. if (!xenvif_mcast_match(vif, eth->h_dest))
  161. goto drop;
  162. }
  163. cb = XENVIF_RX_CB(skb);
  164. cb->expires = jiffies + vif->drain_timeout;
  165. /* If there is no hash algorithm configured then make sure there
  166. * is no hash information in the socket buffer otherwise it
  167. * would be incorrectly forwarded to the frontend.
  168. */
  169. if (vif->hash.alg == XEN_NETIF_CTRL_HASH_ALGORITHM_NONE)
  170. skb_clear_hash(skb);
  171. xenvif_rx_queue_tail(queue, skb);
  172. xenvif_kick_thread(queue);
  173. return NETDEV_TX_OK;
  174. drop:
  175. vif->dev->stats.tx_dropped++;
  176. dev_kfree_skb(skb);
  177. return NETDEV_TX_OK;
  178. }
  179. static struct net_device_stats *xenvif_get_stats(struct net_device *dev)
  180. {
  181. struct xenvif *vif = netdev_priv(dev);
  182. struct xenvif_queue *queue = NULL;
  183. u64 rx_bytes = 0;
  184. u64 rx_packets = 0;
  185. u64 tx_bytes = 0;
  186. u64 tx_packets = 0;
  187. unsigned int index;
  188. spin_lock(&vif->lock);
  189. if (vif->queues == NULL)
  190. goto out;
  191. /* Aggregate tx and rx stats from each queue */
  192. for (index = 0; index < vif->num_queues; ++index) {
  193. queue = &vif->queues[index];
  194. rx_bytes += queue->stats.rx_bytes;
  195. rx_packets += queue->stats.rx_packets;
  196. tx_bytes += queue->stats.tx_bytes;
  197. tx_packets += queue->stats.tx_packets;
  198. }
  199. out:
  200. spin_unlock(&vif->lock);
  201. vif->dev->stats.rx_bytes = rx_bytes;
  202. vif->dev->stats.rx_packets = rx_packets;
  203. vif->dev->stats.tx_bytes = tx_bytes;
  204. vif->dev->stats.tx_packets = tx_packets;
  205. return &vif->dev->stats;
  206. }
  207. static void xenvif_up(struct xenvif *vif)
  208. {
  209. struct xenvif_queue *queue = NULL;
  210. unsigned int num_queues = vif->num_queues;
  211. unsigned int queue_index;
  212. for (queue_index = 0; queue_index < num_queues; ++queue_index) {
  213. queue = &vif->queues[queue_index];
  214. napi_enable(&queue->napi);
  215. enable_irq(queue->tx_irq);
  216. if (queue->tx_irq != queue->rx_irq)
  217. enable_irq(queue->rx_irq);
  218. xenvif_napi_schedule_or_enable_events(queue);
  219. }
  220. }
  221. static void xenvif_down(struct xenvif *vif)
  222. {
  223. struct xenvif_queue *queue = NULL;
  224. unsigned int num_queues = vif->num_queues;
  225. unsigned int queue_index;
  226. for (queue_index = 0; queue_index < num_queues; ++queue_index) {
  227. queue = &vif->queues[queue_index];
  228. disable_irq(queue->tx_irq);
  229. if (queue->tx_irq != queue->rx_irq)
  230. disable_irq(queue->rx_irq);
  231. napi_disable(&queue->napi);
  232. del_timer_sync(&queue->credit_timeout);
  233. }
  234. }
  235. static int xenvif_open(struct net_device *dev)
  236. {
  237. struct xenvif *vif = netdev_priv(dev);
  238. if (test_bit(VIF_STATUS_CONNECTED, &vif->status))
  239. xenvif_up(vif);
  240. netif_tx_start_all_queues(dev);
  241. return 0;
  242. }
  243. static int xenvif_close(struct net_device *dev)
  244. {
  245. struct xenvif *vif = netdev_priv(dev);
  246. if (test_bit(VIF_STATUS_CONNECTED, &vif->status))
  247. xenvif_down(vif);
  248. netif_tx_stop_all_queues(dev);
  249. return 0;
  250. }
  251. static int xenvif_change_mtu(struct net_device *dev, int mtu)
  252. {
  253. struct xenvif *vif = netdev_priv(dev);
  254. int max = vif->can_sg ? ETH_MAX_MTU - VLAN_ETH_HLEN : ETH_DATA_LEN;
  255. if (mtu > max)
  256. return -EINVAL;
  257. dev->mtu = mtu;
  258. return 0;
  259. }
  260. static netdev_features_t xenvif_fix_features(struct net_device *dev,
  261. netdev_features_t features)
  262. {
  263. struct xenvif *vif = netdev_priv(dev);
  264. if (!vif->can_sg)
  265. features &= ~NETIF_F_SG;
  266. if (~(vif->gso_mask) & GSO_BIT(TCPV4))
  267. features &= ~NETIF_F_TSO;
  268. if (~(vif->gso_mask) & GSO_BIT(TCPV6))
  269. features &= ~NETIF_F_TSO6;
  270. if (!vif->ip_csum)
  271. features &= ~NETIF_F_IP_CSUM;
  272. if (!vif->ipv6_csum)
  273. features &= ~NETIF_F_IPV6_CSUM;
  274. return features;
  275. }
  276. static const struct xenvif_stat {
  277. char name[ETH_GSTRING_LEN];
  278. u16 offset;
  279. } xenvif_stats[] = {
  280. {
  281. "rx_gso_checksum_fixup",
  282. offsetof(struct xenvif_stats, rx_gso_checksum_fixup)
  283. },
  284. /* If (sent != success + fail), there are probably packets never
  285. * freed up properly!
  286. */
  287. {
  288. "tx_zerocopy_sent",
  289. offsetof(struct xenvif_stats, tx_zerocopy_sent),
  290. },
  291. {
  292. "tx_zerocopy_success",
  293. offsetof(struct xenvif_stats, tx_zerocopy_success),
  294. },
  295. {
  296. "tx_zerocopy_fail",
  297. offsetof(struct xenvif_stats, tx_zerocopy_fail)
  298. },
  299. /* Number of packets exceeding MAX_SKB_FRAG slots. You should use
  300. * a guest with the same MAX_SKB_FRAG
  301. */
  302. {
  303. "tx_frag_overflow",
  304. offsetof(struct xenvif_stats, tx_frag_overflow)
  305. },
  306. };
  307. static int xenvif_get_sset_count(struct net_device *dev, int string_set)
  308. {
  309. switch (string_set) {
  310. case ETH_SS_STATS:
  311. return ARRAY_SIZE(xenvif_stats);
  312. default:
  313. return -EINVAL;
  314. }
  315. }
  316. static void xenvif_get_ethtool_stats(struct net_device *dev,
  317. struct ethtool_stats *stats, u64 * data)
  318. {
  319. struct xenvif *vif = netdev_priv(dev);
  320. unsigned int num_queues = vif->num_queues;
  321. int i;
  322. unsigned int queue_index;
  323. for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++) {
  324. unsigned long accum = 0;
  325. for (queue_index = 0; queue_index < num_queues; ++queue_index) {
  326. void *vif_stats = &vif->queues[queue_index].stats;
  327. accum += *(unsigned long *)(vif_stats + xenvif_stats[i].offset);
  328. }
  329. data[i] = accum;
  330. }
  331. }
  332. static void xenvif_get_strings(struct net_device *dev, u32 stringset, u8 * data)
  333. {
  334. int i;
  335. switch (stringset) {
  336. case ETH_SS_STATS:
  337. for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++)
  338. memcpy(data + i * ETH_GSTRING_LEN,
  339. xenvif_stats[i].name, ETH_GSTRING_LEN);
  340. break;
  341. }
  342. }
  343. static const struct ethtool_ops xenvif_ethtool_ops = {
  344. .get_link = ethtool_op_get_link,
  345. .get_sset_count = xenvif_get_sset_count,
  346. .get_ethtool_stats = xenvif_get_ethtool_stats,
  347. .get_strings = xenvif_get_strings,
  348. };
  349. static const struct net_device_ops xenvif_netdev_ops = {
  350. .ndo_select_queue = xenvif_select_queue,
  351. .ndo_start_xmit = xenvif_start_xmit,
  352. .ndo_get_stats = xenvif_get_stats,
  353. .ndo_open = xenvif_open,
  354. .ndo_stop = xenvif_close,
  355. .ndo_change_mtu = xenvif_change_mtu,
  356. .ndo_fix_features = xenvif_fix_features,
  357. .ndo_set_mac_address = eth_mac_addr,
  358. .ndo_validate_addr = eth_validate_addr,
  359. };
  360. struct xenvif *xenvif_alloc(struct device *parent, domid_t domid,
  361. unsigned int handle)
  362. {
  363. int err;
  364. struct net_device *dev;
  365. struct xenvif *vif;
  366. char name[IFNAMSIZ] = {};
  367. snprintf(name, IFNAMSIZ - 1, "vif%u.%u", domid, handle);
  368. /* Allocate a netdev with the max. supported number of queues.
  369. * When the guest selects the desired number, it will be updated
  370. * via netif_set_real_num_*_queues().
  371. */
  372. dev = alloc_netdev_mq(sizeof(struct xenvif), name, NET_NAME_UNKNOWN,
  373. ether_setup, xenvif_max_queues);
  374. if (dev == NULL) {
  375. pr_warn("Could not allocate netdev for %s\n", name);
  376. return ERR_PTR(-ENOMEM);
  377. }
  378. SET_NETDEV_DEV(dev, parent);
  379. vif = netdev_priv(dev);
  380. vif->domid = domid;
  381. vif->handle = handle;
  382. vif->can_sg = 1;
  383. vif->ip_csum = 1;
  384. vif->dev = dev;
  385. vif->disabled = false;
  386. vif->drain_timeout = msecs_to_jiffies(rx_drain_timeout_msecs);
  387. vif->stall_timeout = msecs_to_jiffies(rx_stall_timeout_msecs);
  388. /* Start out with no queues. */
  389. vif->queues = NULL;
  390. vif->num_queues = 0;
  391. spin_lock_init(&vif->lock);
  392. INIT_LIST_HEAD(&vif->fe_mcast_addr);
  393. dev->netdev_ops = &xenvif_netdev_ops;
  394. dev->hw_features = NETIF_F_SG |
  395. NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
  396. NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_FRAGLIST;
  397. dev->features = dev->hw_features | NETIF_F_RXCSUM;
  398. dev->ethtool_ops = &xenvif_ethtool_ops;
  399. dev->tx_queue_len = XENVIF_QUEUE_LENGTH;
  400. dev->min_mtu = 0;
  401. dev->max_mtu = ETH_MAX_MTU - VLAN_ETH_HLEN;
  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. eth_broadcast_addr(dev->dev_addr);
  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_timeout.function = xenvif_tx_credit_callback;
  428. queue->credit_window_start = get_jiffies_64();
  429. queue->rx_queue_max = XENVIF_RX_QUEUE_BYTES;
  430. skb_queue_head_init(&queue->rx_queue);
  431. skb_queue_head_init(&queue->tx_queue);
  432. queue->pending_cons = 0;
  433. queue->pending_prod = MAX_PENDING_REQS;
  434. for (i = 0; i < MAX_PENDING_REQS; ++i)
  435. queue->pending_ring[i] = i;
  436. spin_lock_init(&queue->callback_lock);
  437. spin_lock_init(&queue->response_lock);
  438. /* If ballooning is disabled, this will consume real memory, so you
  439. * better enable it. The long term solution would be to use just a
  440. * bunch of valid page descriptors, without dependency on ballooning
  441. */
  442. err = gnttab_alloc_pages(MAX_PENDING_REQS,
  443. queue->mmap_pages);
  444. if (err) {
  445. netdev_err(queue->vif->dev, "Could not reserve mmap_pages\n");
  446. return -ENOMEM;
  447. }
  448. for (i = 0; i < MAX_PENDING_REQS; i++) {
  449. queue->pending_tx_info[i].callback_struct = (struct ubuf_info)
  450. { .callback = xenvif_zerocopy_callback,
  451. .ctx = NULL,
  452. .desc = i };
  453. queue->grant_tx_handle[i] = NETBACK_INVALID_HANDLE;
  454. }
  455. return 0;
  456. }
  457. void xenvif_carrier_on(struct xenvif *vif)
  458. {
  459. rtnl_lock();
  460. if (!vif->can_sg && vif->dev->mtu > ETH_DATA_LEN)
  461. dev_set_mtu(vif->dev, ETH_DATA_LEN);
  462. netdev_update_features(vif->dev);
  463. set_bit(VIF_STATUS_CONNECTED, &vif->status);
  464. if (netif_running(vif->dev))
  465. xenvif_up(vif);
  466. rtnl_unlock();
  467. }
  468. int xenvif_connect_ctrl(struct xenvif *vif, grant_ref_t ring_ref,
  469. unsigned int evtchn)
  470. {
  471. struct net_device *dev = vif->dev;
  472. void *addr;
  473. struct xen_netif_ctrl_sring *shared;
  474. int err;
  475. err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(vif),
  476. &ring_ref, 1, &addr);
  477. if (err)
  478. goto err;
  479. shared = (struct xen_netif_ctrl_sring *)addr;
  480. BACK_RING_INIT(&vif->ctrl, shared, XEN_PAGE_SIZE);
  481. err = bind_interdomain_evtchn_to_irq(vif->domid, evtchn);
  482. if (err < 0)
  483. goto err_unmap;
  484. vif->ctrl_irq = err;
  485. xenvif_init_hash(vif);
  486. err = request_threaded_irq(vif->ctrl_irq, NULL, xenvif_ctrl_irq_fn,
  487. IRQF_ONESHOT, "xen-netback-ctrl", vif);
  488. if (err) {
  489. pr_warn("Could not setup irq handler for %s\n", dev->name);
  490. goto err_deinit;
  491. }
  492. return 0;
  493. err_deinit:
  494. xenvif_deinit_hash(vif);
  495. unbind_from_irqhandler(vif->ctrl_irq, vif);
  496. vif->ctrl_irq = 0;
  497. err_unmap:
  498. xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif),
  499. vif->ctrl.sring);
  500. vif->ctrl.sring = NULL;
  501. err:
  502. return err;
  503. }
  504. int xenvif_connect_data(struct xenvif_queue *queue,
  505. unsigned long tx_ring_ref,
  506. unsigned long rx_ring_ref,
  507. unsigned int tx_evtchn,
  508. unsigned int rx_evtchn)
  509. {
  510. struct task_struct *task;
  511. int err = -ENOMEM;
  512. BUG_ON(queue->tx_irq);
  513. BUG_ON(queue->task);
  514. BUG_ON(queue->dealloc_task);
  515. err = xenvif_map_frontend_data_rings(queue, tx_ring_ref,
  516. rx_ring_ref);
  517. if (err < 0)
  518. goto err;
  519. init_waitqueue_head(&queue->wq);
  520. init_waitqueue_head(&queue->dealloc_wq);
  521. atomic_set(&queue->inflight_packets, 0);
  522. netif_napi_add(queue->vif->dev, &queue->napi, xenvif_poll,
  523. XENVIF_NAPI_WEIGHT);
  524. if (tx_evtchn == rx_evtchn) {
  525. /* feature-split-event-channels == 0 */
  526. err = bind_interdomain_evtchn_to_irqhandler(
  527. queue->vif->domid, tx_evtchn, xenvif_interrupt, 0,
  528. queue->name, queue);
  529. if (err < 0)
  530. goto err_unmap;
  531. queue->tx_irq = queue->rx_irq = err;
  532. disable_irq(queue->tx_irq);
  533. } else {
  534. /* feature-split-event-channels == 1 */
  535. snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
  536. "%s-tx", queue->name);
  537. err = bind_interdomain_evtchn_to_irqhandler(
  538. queue->vif->domid, tx_evtchn, xenvif_tx_interrupt, 0,
  539. queue->tx_irq_name, queue);
  540. if (err < 0)
  541. goto err_unmap;
  542. queue->tx_irq = err;
  543. disable_irq(queue->tx_irq);
  544. snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
  545. "%s-rx", queue->name);
  546. err = bind_interdomain_evtchn_to_irqhandler(
  547. queue->vif->domid, rx_evtchn, xenvif_rx_interrupt, 0,
  548. queue->rx_irq_name, queue);
  549. if (err < 0)
  550. goto err_tx_unbind;
  551. queue->rx_irq = err;
  552. disable_irq(queue->rx_irq);
  553. }
  554. queue->stalled = true;
  555. task = kthread_create(xenvif_kthread_guest_rx,
  556. (void *)queue, "%s-guest-rx", queue->name);
  557. if (IS_ERR(task)) {
  558. pr_warn("Could not allocate kthread for %s\n", queue->name);
  559. err = PTR_ERR(task);
  560. goto err_rx_unbind;
  561. }
  562. queue->task = task;
  563. get_task_struct(task);
  564. task = kthread_create(xenvif_dealloc_kthread,
  565. (void *)queue, "%s-dealloc", queue->name);
  566. if (IS_ERR(task)) {
  567. pr_warn("Could not allocate kthread for %s\n", queue->name);
  568. err = PTR_ERR(task);
  569. goto err_rx_unbind;
  570. }
  571. queue->dealloc_task = task;
  572. wake_up_process(queue->task);
  573. wake_up_process(queue->dealloc_task);
  574. return 0;
  575. err_rx_unbind:
  576. unbind_from_irqhandler(queue->rx_irq, queue);
  577. queue->rx_irq = 0;
  578. err_tx_unbind:
  579. unbind_from_irqhandler(queue->tx_irq, queue);
  580. queue->tx_irq = 0;
  581. err_unmap:
  582. xenvif_unmap_frontend_data_rings(queue);
  583. netif_napi_del(&queue->napi);
  584. err:
  585. module_put(THIS_MODULE);
  586. return err;
  587. }
  588. void xenvif_carrier_off(struct xenvif *vif)
  589. {
  590. struct net_device *dev = vif->dev;
  591. rtnl_lock();
  592. if (test_and_clear_bit(VIF_STATUS_CONNECTED, &vif->status)) {
  593. netif_carrier_off(dev); /* discard queued packets */
  594. if (netif_running(dev))
  595. xenvif_down(vif);
  596. }
  597. rtnl_unlock();
  598. }
  599. void xenvif_disconnect_data(struct xenvif *vif)
  600. {
  601. struct xenvif_queue *queue = NULL;
  602. unsigned int num_queues = vif->num_queues;
  603. unsigned int queue_index;
  604. xenvif_carrier_off(vif);
  605. for (queue_index = 0; queue_index < num_queues; ++queue_index) {
  606. queue = &vif->queues[queue_index];
  607. netif_napi_del(&queue->napi);
  608. if (queue->task) {
  609. kthread_stop(queue->task);
  610. put_task_struct(queue->task);
  611. queue->task = NULL;
  612. }
  613. if (queue->dealloc_task) {
  614. kthread_stop(queue->dealloc_task);
  615. queue->dealloc_task = NULL;
  616. }
  617. if (queue->tx_irq) {
  618. if (queue->tx_irq == queue->rx_irq)
  619. unbind_from_irqhandler(queue->tx_irq, queue);
  620. else {
  621. unbind_from_irqhandler(queue->tx_irq, queue);
  622. unbind_from_irqhandler(queue->rx_irq, queue);
  623. }
  624. queue->tx_irq = 0;
  625. }
  626. xenvif_unmap_frontend_data_rings(queue);
  627. }
  628. xenvif_mcast_addr_list_free(vif);
  629. }
  630. void xenvif_disconnect_ctrl(struct xenvif *vif)
  631. {
  632. if (vif->ctrl_irq) {
  633. xenvif_deinit_hash(vif);
  634. unbind_from_irqhandler(vif->ctrl_irq, vif);
  635. vif->ctrl_irq = 0;
  636. }
  637. if (vif->ctrl.sring) {
  638. xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif),
  639. vif->ctrl.sring);
  640. vif->ctrl.sring = NULL;
  641. }
  642. }
  643. /* Reverse the relevant parts of xenvif_init_queue().
  644. * Used for queue teardown from xenvif_free(), and on the
  645. * error handling paths in xenbus.c:connect().
  646. */
  647. void xenvif_deinit_queue(struct xenvif_queue *queue)
  648. {
  649. gnttab_free_pages(MAX_PENDING_REQS, queue->mmap_pages);
  650. }
  651. void xenvif_free(struct xenvif *vif)
  652. {
  653. struct xenvif_queue *queues = vif->queues;
  654. unsigned int num_queues = vif->num_queues;
  655. unsigned int queue_index;
  656. unregister_netdev(vif->dev);
  657. free_netdev(vif->dev);
  658. for (queue_index = 0; queue_index < num_queues; ++queue_index)
  659. xenvif_deinit_queue(&queues[queue_index]);
  660. vfree(queues);
  661. module_put(THIS_MODULE);
  662. }