Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1 | /* |
| 2 | * Network-device interface management. |
| 3 | * |
| 4 | * Copyright (c) 2004-2005, Keir Fraser |
| 5 | * |
| 6 | * This program is free software; you can redistribute it and/or |
| 7 | * modify it under the terms of the GNU General Public License version 2 |
| 8 | * as published by the Free Software Foundation; or, when distributed |
| 9 | * separately from the Linux kernel or incorporated into other |
| 10 | * software packages, subject to the following license: |
| 11 | * |
| 12 | * Permission is hereby granted, free of charge, to any person obtaining a copy |
| 13 | * of this source file (the "Software"), to deal in the Software without |
| 14 | * restriction, including without limitation the rights to use, copy, modify, |
| 15 | * merge, publish, distribute, sublicense, and/or sell copies of the Software, |
| 16 | * and to permit persons to whom the Software is furnished to do so, subject to |
| 17 | * the following conditions: |
| 18 | * |
| 19 | * The above copyright notice and this permission notice shall be included in |
| 20 | * all copies or substantial portions of the Software. |
| 21 | * |
| 22 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
| 23 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
| 24 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE |
| 25 | * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER |
| 26 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING |
| 27 | * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS |
| 28 | * IN THE SOFTWARE. |
| 29 | */ |
| 30 | |
| 31 | #include "common.h" |
| 32 | |
| 33 | #include <linux/kthread.h> |
| 34 | #include <linux/sched/task.h> |
| 35 | #include <linux/ethtool.h> |
| 36 | #include <linux/rtnetlink.h> |
| 37 | #include <linux/if_vlan.h> |
| 38 | #include <linux/vmalloc.h> |
| 39 | |
| 40 | #include <xen/events.h> |
| 41 | #include <asm/xen/hypercall.h> |
| 42 | #include <xen/balloon.h> |
| 43 | |
| 44 | #define XENVIF_QUEUE_LENGTH 32 |
| 45 | #define XENVIF_NAPI_WEIGHT 64 |
| 46 | |
| 47 | /* Number of bytes allowed on the internal guest Rx queue. */ |
| 48 | #define XENVIF_RX_QUEUE_BYTES (XEN_NETIF_RX_RING_SIZE/2 * PAGE_SIZE) |
| 49 | |
| 50 | /* This function is used to set SKBTX_DEV_ZEROCOPY as well as |
| 51 | * increasing the inflight counter. We need to increase the inflight |
| 52 | * counter because core driver calls into xenvif_zerocopy_callback |
| 53 | * which calls xenvif_skb_zerocopy_complete. |
| 54 | */ |
| 55 | void xenvif_skb_zerocopy_prepare(struct xenvif_queue *queue, |
| 56 | struct sk_buff *skb) |
| 57 | { |
| 58 | skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY; |
| 59 | atomic_inc(&queue->inflight_packets); |
| 60 | } |
| 61 | |
| 62 | void xenvif_skb_zerocopy_complete(struct xenvif_queue *queue) |
| 63 | { |
| 64 | atomic_dec(&queue->inflight_packets); |
| 65 | |
| 66 | /* Wake the dealloc thread _after_ decrementing inflight_packets so |
| 67 | * that if kthread_stop() has already been called, the dealloc thread |
| 68 | * does not wait forever with nothing to wake it. |
| 69 | */ |
| 70 | wake_up(&queue->dealloc_wq); |
| 71 | } |
| 72 | |
| 73 | int xenvif_schedulable(struct xenvif *vif) |
| 74 | { |
| 75 | return netif_running(vif->dev) && |
| 76 | test_bit(VIF_STATUS_CONNECTED, &vif->status) && |
| 77 | !vif->disabled; |
| 78 | } |
| 79 | |
| 80 | static irqreturn_t xenvif_tx_interrupt(int irq, void *dev_id) |
| 81 | { |
| 82 | struct xenvif_queue *queue = dev_id; |
| 83 | |
| 84 | if (RING_HAS_UNCONSUMED_REQUESTS(&queue->tx)) |
| 85 | napi_schedule(&queue->napi); |
| 86 | |
| 87 | return IRQ_HANDLED; |
| 88 | } |
| 89 | |
| 90 | static int xenvif_poll(struct napi_struct *napi, int budget) |
| 91 | { |
| 92 | struct xenvif_queue *queue = |
| 93 | container_of(napi, struct xenvif_queue, napi); |
| 94 | int work_done; |
| 95 | |
| 96 | /* This vif is rogue, we pretend we've there is nothing to do |
| 97 | * for this vif to deschedule it from NAPI. But this interface |
| 98 | * will be turned off in thread context later. |
| 99 | */ |
| 100 | if (unlikely(queue->vif->disabled)) { |
| 101 | napi_complete(napi); |
| 102 | return 0; |
| 103 | } |
| 104 | |
| 105 | work_done = xenvif_tx_action(queue, budget); |
| 106 | |
| 107 | if (work_done < budget) { |
| 108 | napi_complete_done(napi, work_done); |
| 109 | /* If the queue is rate-limited, it shall be |
| 110 | * rescheduled in the timer callback. |
| 111 | */ |
| 112 | if (likely(!queue->rate_limited)) |
| 113 | xenvif_napi_schedule_or_enable_events(queue); |
| 114 | } |
| 115 | |
| 116 | return work_done; |
| 117 | } |
| 118 | |
| 119 | static irqreturn_t xenvif_rx_interrupt(int irq, void *dev_id) |
| 120 | { |
| 121 | struct xenvif_queue *queue = dev_id; |
| 122 | |
| 123 | xenvif_kick_thread(queue); |
| 124 | |
| 125 | return IRQ_HANDLED; |
| 126 | } |
| 127 | |
| 128 | irqreturn_t xenvif_interrupt(int irq, void *dev_id) |
| 129 | { |
| 130 | xenvif_tx_interrupt(irq, dev_id); |
| 131 | xenvif_rx_interrupt(irq, dev_id); |
| 132 | |
| 133 | return IRQ_HANDLED; |
| 134 | } |
| 135 | |
| 136 | int xenvif_queue_stopped(struct xenvif_queue *queue) |
| 137 | { |
| 138 | struct net_device *dev = queue->vif->dev; |
| 139 | unsigned int id = queue->id; |
| 140 | return netif_tx_queue_stopped(netdev_get_tx_queue(dev, id)); |
| 141 | } |
| 142 | |
| 143 | void xenvif_wake_queue(struct xenvif_queue *queue) |
| 144 | { |
| 145 | struct net_device *dev = queue->vif->dev; |
| 146 | unsigned int id = queue->id; |
| 147 | netif_tx_wake_queue(netdev_get_tx_queue(dev, id)); |
| 148 | } |
| 149 | |
| 150 | static u16 xenvif_select_queue(struct net_device *dev, struct sk_buff *skb, |
| 151 | struct net_device *sb_dev, |
| 152 | select_queue_fallback_t fallback) |
| 153 | { |
| 154 | struct xenvif *vif = netdev_priv(dev); |
| 155 | unsigned int size = vif->hash.size; |
| 156 | |
| 157 | if (vif->hash.alg == XEN_NETIF_CTRL_HASH_ALGORITHM_NONE) |
| 158 | return fallback(dev, skb, NULL) % dev->real_num_tx_queues; |
| 159 | |
| 160 | xenvif_set_skb_hash(vif, skb); |
| 161 | |
| 162 | if (size == 0) |
| 163 | return skb_get_hash_raw(skb) % dev->real_num_tx_queues; |
| 164 | |
| 165 | return vif->hash.mapping[vif->hash.mapping_sel] |
| 166 | [skb_get_hash_raw(skb) % size]; |
| 167 | } |
| 168 | |
| 169 | static int xenvif_start_xmit(struct sk_buff *skb, struct net_device *dev) |
| 170 | { |
| 171 | struct xenvif *vif = netdev_priv(dev); |
| 172 | struct xenvif_queue *queue = NULL; |
| 173 | unsigned int num_queues; |
| 174 | u16 index; |
| 175 | struct xenvif_rx_cb *cb; |
| 176 | |
| 177 | BUG_ON(skb->dev != dev); |
| 178 | |
| 179 | /* Drop the packet if queues are not set up. |
| 180 | * This handler should be called inside an RCU read section |
| 181 | * so we don't need to enter it here explicitly. |
| 182 | */ |
| 183 | num_queues = READ_ONCE(vif->num_queues); |
| 184 | if (num_queues < 1) |
| 185 | goto drop; |
| 186 | |
| 187 | /* Obtain the queue to be used to transmit this packet */ |
| 188 | index = skb_get_queue_mapping(skb); |
| 189 | if (index >= num_queues) { |
| 190 | pr_warn_ratelimited("Invalid queue %hu for packet on interface %s\n", |
| 191 | index, vif->dev->name); |
| 192 | index %= num_queues; |
| 193 | } |
| 194 | queue = &vif->queues[index]; |
| 195 | |
| 196 | /* Drop the packet if queue is not ready */ |
| 197 | if (queue->task == NULL || |
| 198 | queue->dealloc_task == NULL || |
| 199 | !xenvif_schedulable(vif)) |
| 200 | goto drop; |
| 201 | |
| 202 | if (vif->multicast_control && skb->pkt_type == PACKET_MULTICAST) { |
| 203 | struct ethhdr *eth = (struct ethhdr *)skb->data; |
| 204 | |
| 205 | if (!xenvif_mcast_match(vif, eth->h_dest)) |
| 206 | goto drop; |
| 207 | } |
| 208 | |
| 209 | cb = XENVIF_RX_CB(skb); |
| 210 | cb->expires = jiffies + vif->drain_timeout; |
| 211 | |
| 212 | /* If there is no hash algorithm configured then make sure there |
| 213 | * is no hash information in the socket buffer otherwise it |
| 214 | * would be incorrectly forwarded to the frontend. |
| 215 | */ |
| 216 | if (vif->hash.alg == XEN_NETIF_CTRL_HASH_ALGORITHM_NONE) |
| 217 | skb_clear_hash(skb); |
| 218 | |
| 219 | xenvif_rx_queue_tail(queue, skb); |
| 220 | xenvif_kick_thread(queue); |
| 221 | |
| 222 | return NETDEV_TX_OK; |
| 223 | |
| 224 | drop: |
| 225 | vif->dev->stats.tx_dropped++; |
| 226 | dev_kfree_skb(skb); |
| 227 | return NETDEV_TX_OK; |
| 228 | } |
| 229 | |
| 230 | static struct net_device_stats *xenvif_get_stats(struct net_device *dev) |
| 231 | { |
| 232 | struct xenvif *vif = netdev_priv(dev); |
| 233 | struct xenvif_queue *queue = NULL; |
| 234 | unsigned int num_queues; |
| 235 | u64 rx_bytes = 0; |
| 236 | u64 rx_packets = 0; |
| 237 | u64 tx_bytes = 0; |
| 238 | u64 tx_packets = 0; |
| 239 | unsigned int index; |
| 240 | |
| 241 | rcu_read_lock(); |
| 242 | num_queues = READ_ONCE(vif->num_queues); |
| 243 | |
| 244 | /* Aggregate tx and rx stats from each queue */ |
| 245 | for (index = 0; index < num_queues; ++index) { |
| 246 | queue = &vif->queues[index]; |
| 247 | rx_bytes += queue->stats.rx_bytes; |
| 248 | rx_packets += queue->stats.rx_packets; |
| 249 | tx_bytes += queue->stats.tx_bytes; |
| 250 | tx_packets += queue->stats.tx_packets; |
| 251 | } |
| 252 | |
| 253 | rcu_read_unlock(); |
| 254 | |
| 255 | vif->dev->stats.rx_bytes = rx_bytes; |
| 256 | vif->dev->stats.rx_packets = rx_packets; |
| 257 | vif->dev->stats.tx_bytes = tx_bytes; |
| 258 | vif->dev->stats.tx_packets = tx_packets; |
| 259 | |
| 260 | return &vif->dev->stats; |
| 261 | } |
| 262 | |
| 263 | static void xenvif_up(struct xenvif *vif) |
| 264 | { |
| 265 | struct xenvif_queue *queue = NULL; |
| 266 | unsigned int num_queues = vif->num_queues; |
| 267 | unsigned int queue_index; |
| 268 | |
| 269 | for (queue_index = 0; queue_index < num_queues; ++queue_index) { |
| 270 | queue = &vif->queues[queue_index]; |
| 271 | napi_enable(&queue->napi); |
| 272 | enable_irq(queue->tx_irq); |
| 273 | if (queue->tx_irq != queue->rx_irq) |
| 274 | enable_irq(queue->rx_irq); |
| 275 | xenvif_napi_schedule_or_enable_events(queue); |
| 276 | } |
| 277 | } |
| 278 | |
| 279 | static void xenvif_down(struct xenvif *vif) |
| 280 | { |
| 281 | struct xenvif_queue *queue = NULL; |
| 282 | unsigned int num_queues = vif->num_queues; |
| 283 | unsigned int queue_index; |
| 284 | |
| 285 | for (queue_index = 0; queue_index < num_queues; ++queue_index) { |
| 286 | queue = &vif->queues[queue_index]; |
| 287 | disable_irq(queue->tx_irq); |
| 288 | if (queue->tx_irq != queue->rx_irq) |
| 289 | disable_irq(queue->rx_irq); |
| 290 | napi_disable(&queue->napi); |
| 291 | del_timer_sync(&queue->credit_timeout); |
| 292 | } |
| 293 | } |
| 294 | |
| 295 | static int xenvif_open(struct net_device *dev) |
| 296 | { |
| 297 | struct xenvif *vif = netdev_priv(dev); |
| 298 | if (test_bit(VIF_STATUS_CONNECTED, &vif->status)) |
| 299 | xenvif_up(vif); |
| 300 | netif_tx_start_all_queues(dev); |
| 301 | return 0; |
| 302 | } |
| 303 | |
| 304 | static int xenvif_close(struct net_device *dev) |
| 305 | { |
| 306 | struct xenvif *vif = netdev_priv(dev); |
| 307 | if (test_bit(VIF_STATUS_CONNECTED, &vif->status)) |
| 308 | xenvif_down(vif); |
| 309 | netif_tx_stop_all_queues(dev); |
| 310 | return 0; |
| 311 | } |
| 312 | |
| 313 | static int xenvif_change_mtu(struct net_device *dev, int mtu) |
| 314 | { |
| 315 | struct xenvif *vif = netdev_priv(dev); |
| 316 | int max = vif->can_sg ? ETH_MAX_MTU - VLAN_ETH_HLEN : ETH_DATA_LEN; |
| 317 | |
| 318 | if (mtu > max) |
| 319 | return -EINVAL; |
| 320 | dev->mtu = mtu; |
| 321 | return 0; |
| 322 | } |
| 323 | |
| 324 | static netdev_features_t xenvif_fix_features(struct net_device *dev, |
| 325 | netdev_features_t features) |
| 326 | { |
| 327 | struct xenvif *vif = netdev_priv(dev); |
| 328 | |
| 329 | if (!vif->can_sg) |
| 330 | features &= ~NETIF_F_SG; |
| 331 | if (~(vif->gso_mask) & GSO_BIT(TCPV4)) |
| 332 | features &= ~NETIF_F_TSO; |
| 333 | if (~(vif->gso_mask) & GSO_BIT(TCPV6)) |
| 334 | features &= ~NETIF_F_TSO6; |
| 335 | if (!vif->ip_csum) |
| 336 | features &= ~NETIF_F_IP_CSUM; |
| 337 | if (!vif->ipv6_csum) |
| 338 | features &= ~NETIF_F_IPV6_CSUM; |
| 339 | |
| 340 | return features; |
| 341 | } |
| 342 | |
| 343 | static const struct xenvif_stat { |
| 344 | char name[ETH_GSTRING_LEN]; |
| 345 | u16 offset; |
| 346 | } xenvif_stats[] = { |
| 347 | { |
| 348 | "rx_gso_checksum_fixup", |
| 349 | offsetof(struct xenvif_stats, rx_gso_checksum_fixup) |
| 350 | }, |
| 351 | /* If (sent != success + fail), there are probably packets never |
| 352 | * freed up properly! |
| 353 | */ |
| 354 | { |
| 355 | "tx_zerocopy_sent", |
| 356 | offsetof(struct xenvif_stats, tx_zerocopy_sent), |
| 357 | }, |
| 358 | { |
| 359 | "tx_zerocopy_success", |
| 360 | offsetof(struct xenvif_stats, tx_zerocopy_success), |
| 361 | }, |
| 362 | { |
| 363 | "tx_zerocopy_fail", |
| 364 | offsetof(struct xenvif_stats, tx_zerocopy_fail) |
| 365 | }, |
| 366 | /* Number of packets exceeding MAX_SKB_FRAG slots. You should use |
| 367 | * a guest with the same MAX_SKB_FRAG |
| 368 | */ |
| 369 | { |
| 370 | "tx_frag_overflow", |
| 371 | offsetof(struct xenvif_stats, tx_frag_overflow) |
| 372 | }, |
| 373 | }; |
| 374 | |
| 375 | static int xenvif_get_sset_count(struct net_device *dev, int string_set) |
| 376 | { |
| 377 | switch (string_set) { |
| 378 | case ETH_SS_STATS: |
| 379 | return ARRAY_SIZE(xenvif_stats); |
| 380 | default: |
| 381 | return -EINVAL; |
| 382 | } |
| 383 | } |
| 384 | |
| 385 | static void xenvif_get_ethtool_stats(struct net_device *dev, |
| 386 | struct ethtool_stats *stats, u64 * data) |
| 387 | { |
| 388 | struct xenvif *vif = netdev_priv(dev); |
| 389 | unsigned int num_queues; |
| 390 | int i; |
| 391 | unsigned int queue_index; |
| 392 | |
| 393 | rcu_read_lock(); |
| 394 | num_queues = READ_ONCE(vif->num_queues); |
| 395 | |
| 396 | for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++) { |
| 397 | unsigned long accum = 0; |
| 398 | for (queue_index = 0; queue_index < num_queues; ++queue_index) { |
| 399 | void *vif_stats = &vif->queues[queue_index].stats; |
| 400 | accum += *(unsigned long *)(vif_stats + xenvif_stats[i].offset); |
| 401 | } |
| 402 | data[i] = accum; |
| 403 | } |
| 404 | |
| 405 | rcu_read_unlock(); |
| 406 | } |
| 407 | |
| 408 | static void xenvif_get_strings(struct net_device *dev, u32 stringset, u8 * data) |
| 409 | { |
| 410 | int i; |
| 411 | |
| 412 | switch (stringset) { |
| 413 | case ETH_SS_STATS: |
| 414 | for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++) |
| 415 | memcpy(data + i * ETH_GSTRING_LEN, |
| 416 | xenvif_stats[i].name, ETH_GSTRING_LEN); |
| 417 | break; |
| 418 | } |
| 419 | } |
| 420 | |
| 421 | static const struct ethtool_ops xenvif_ethtool_ops = { |
| 422 | .get_link = ethtool_op_get_link, |
| 423 | |
| 424 | .get_sset_count = xenvif_get_sset_count, |
| 425 | .get_ethtool_stats = xenvif_get_ethtool_stats, |
| 426 | .get_strings = xenvif_get_strings, |
| 427 | }; |
| 428 | |
| 429 | static const struct net_device_ops xenvif_netdev_ops = { |
| 430 | .ndo_select_queue = xenvif_select_queue, |
| 431 | .ndo_start_xmit = xenvif_start_xmit, |
| 432 | .ndo_get_stats = xenvif_get_stats, |
| 433 | .ndo_open = xenvif_open, |
| 434 | .ndo_stop = xenvif_close, |
| 435 | .ndo_change_mtu = xenvif_change_mtu, |
| 436 | .ndo_fix_features = xenvif_fix_features, |
| 437 | .ndo_set_mac_address = eth_mac_addr, |
| 438 | .ndo_validate_addr = eth_validate_addr, |
| 439 | }; |
| 440 | |
| 441 | struct xenvif *xenvif_alloc(struct device *parent, domid_t domid, |
| 442 | unsigned int handle) |
| 443 | { |
| 444 | int err; |
| 445 | struct net_device *dev; |
| 446 | struct xenvif *vif; |
| 447 | char name[IFNAMSIZ] = {}; |
| 448 | |
| 449 | snprintf(name, IFNAMSIZ - 1, "vif%u.%u", domid, handle); |
| 450 | /* Allocate a netdev with the max. supported number of queues. |
| 451 | * When the guest selects the desired number, it will be updated |
| 452 | * via netif_set_real_num_*_queues(). |
| 453 | */ |
| 454 | dev = alloc_netdev_mq(sizeof(struct xenvif), name, NET_NAME_UNKNOWN, |
| 455 | ether_setup, xenvif_max_queues); |
| 456 | if (dev == NULL) { |
| 457 | pr_warn("Could not allocate netdev for %s\n", name); |
| 458 | return ERR_PTR(-ENOMEM); |
| 459 | } |
| 460 | |
| 461 | SET_NETDEV_DEV(dev, parent); |
| 462 | |
| 463 | vif = netdev_priv(dev); |
| 464 | |
| 465 | vif->domid = domid; |
| 466 | vif->handle = handle; |
| 467 | vif->can_sg = 1; |
| 468 | vif->ip_csum = 1; |
| 469 | vif->dev = dev; |
| 470 | vif->disabled = false; |
| 471 | vif->drain_timeout = msecs_to_jiffies(rx_drain_timeout_msecs); |
| 472 | vif->stall_timeout = msecs_to_jiffies(rx_stall_timeout_msecs); |
| 473 | |
| 474 | /* Start out with no queues. */ |
| 475 | vif->queues = NULL; |
| 476 | vif->num_queues = 0; |
| 477 | |
| 478 | spin_lock_init(&vif->lock); |
| 479 | INIT_LIST_HEAD(&vif->fe_mcast_addr); |
| 480 | |
| 481 | dev->netdev_ops = &xenvif_netdev_ops; |
| 482 | dev->hw_features = NETIF_F_SG | |
| 483 | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM | |
| 484 | NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_FRAGLIST; |
| 485 | dev->features = dev->hw_features | NETIF_F_RXCSUM; |
| 486 | dev->ethtool_ops = &xenvif_ethtool_ops; |
| 487 | |
| 488 | dev->tx_queue_len = XENVIF_QUEUE_LENGTH; |
| 489 | |
| 490 | dev->min_mtu = ETH_MIN_MTU; |
| 491 | dev->max_mtu = ETH_MAX_MTU - VLAN_ETH_HLEN; |
| 492 | |
| 493 | /* |
| 494 | * Initialise a dummy MAC address. We choose the numerically |
| 495 | * largest non-broadcast address to prevent the address getting |
| 496 | * stolen by an Ethernet bridge for STP purposes. |
| 497 | * (FE:FF:FF:FF:FF:FF) |
| 498 | */ |
| 499 | eth_broadcast_addr(dev->dev_addr); |
| 500 | dev->dev_addr[0] &= ~0x01; |
| 501 | |
| 502 | netif_carrier_off(dev); |
| 503 | |
| 504 | err = register_netdev(dev); |
| 505 | if (err) { |
| 506 | netdev_warn(dev, "Could not register device: err=%d\n", err); |
| 507 | free_netdev(dev); |
| 508 | return ERR_PTR(err); |
| 509 | } |
| 510 | |
| 511 | netdev_dbg(dev, "Successfully created xenvif\n"); |
| 512 | |
| 513 | __module_get(THIS_MODULE); |
| 514 | |
| 515 | return vif; |
| 516 | } |
| 517 | |
| 518 | int xenvif_init_queue(struct xenvif_queue *queue) |
| 519 | { |
| 520 | int err, i; |
| 521 | |
| 522 | queue->credit_bytes = queue->remaining_credit = ~0UL; |
| 523 | queue->credit_usec = 0UL; |
| 524 | timer_setup(&queue->credit_timeout, xenvif_tx_credit_callback, 0); |
| 525 | queue->credit_window_start = get_jiffies_64(); |
| 526 | |
| 527 | queue->rx_queue_max = XENVIF_RX_QUEUE_BYTES; |
| 528 | |
| 529 | skb_queue_head_init(&queue->rx_queue); |
| 530 | skb_queue_head_init(&queue->tx_queue); |
| 531 | |
| 532 | queue->pending_cons = 0; |
| 533 | queue->pending_prod = MAX_PENDING_REQS; |
| 534 | for (i = 0; i < MAX_PENDING_REQS; ++i) |
| 535 | queue->pending_ring[i] = i; |
| 536 | |
| 537 | spin_lock_init(&queue->callback_lock); |
| 538 | spin_lock_init(&queue->response_lock); |
| 539 | |
| 540 | /* If ballooning is disabled, this will consume real memory, so you |
| 541 | * better enable it. The long term solution would be to use just a |
| 542 | * bunch of valid page descriptors, without dependency on ballooning |
| 543 | */ |
| 544 | err = gnttab_alloc_pages(MAX_PENDING_REQS, |
| 545 | queue->mmap_pages); |
| 546 | if (err) { |
| 547 | netdev_err(queue->vif->dev, "Could not reserve mmap_pages\n"); |
| 548 | return -ENOMEM; |
| 549 | } |
| 550 | |
| 551 | for (i = 0; i < MAX_PENDING_REQS; i++) { |
| 552 | queue->pending_tx_info[i].callback_struct = (struct ubuf_info) |
| 553 | { .callback = xenvif_zerocopy_callback, |
| 554 | { { .ctx = NULL, |
| 555 | .desc = i } } }; |
| 556 | queue->grant_tx_handle[i] = NETBACK_INVALID_HANDLE; |
| 557 | } |
| 558 | |
| 559 | return 0; |
| 560 | } |
| 561 | |
| 562 | void xenvif_carrier_on(struct xenvif *vif) |
| 563 | { |
| 564 | rtnl_lock(); |
| 565 | if (!vif->can_sg && vif->dev->mtu > ETH_DATA_LEN) |
| 566 | dev_set_mtu(vif->dev, ETH_DATA_LEN); |
| 567 | netdev_update_features(vif->dev); |
| 568 | set_bit(VIF_STATUS_CONNECTED, &vif->status); |
| 569 | if (netif_running(vif->dev)) |
| 570 | xenvif_up(vif); |
| 571 | rtnl_unlock(); |
| 572 | } |
| 573 | |
| 574 | int xenvif_connect_ctrl(struct xenvif *vif, grant_ref_t ring_ref, |
| 575 | unsigned int evtchn) |
| 576 | { |
| 577 | struct net_device *dev = vif->dev; |
| 578 | void *addr; |
| 579 | struct xen_netif_ctrl_sring *shared; |
| 580 | int err; |
| 581 | |
| 582 | err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(vif), |
| 583 | &ring_ref, 1, &addr); |
| 584 | if (err) |
| 585 | goto err; |
| 586 | |
| 587 | shared = (struct xen_netif_ctrl_sring *)addr; |
| 588 | BACK_RING_INIT(&vif->ctrl, shared, XEN_PAGE_SIZE); |
| 589 | |
| 590 | err = bind_interdomain_evtchn_to_irq(vif->domid, evtchn); |
| 591 | if (err < 0) |
| 592 | goto err_unmap; |
| 593 | |
| 594 | vif->ctrl_irq = err; |
| 595 | |
| 596 | xenvif_init_hash(vif); |
| 597 | |
| 598 | err = request_threaded_irq(vif->ctrl_irq, NULL, xenvif_ctrl_irq_fn, |
| 599 | IRQF_ONESHOT, "xen-netback-ctrl", vif); |
| 600 | if (err) { |
| 601 | pr_warn("Could not setup irq handler for %s\n", dev->name); |
| 602 | goto err_deinit; |
| 603 | } |
| 604 | |
| 605 | return 0; |
| 606 | |
| 607 | err_deinit: |
| 608 | xenvif_deinit_hash(vif); |
| 609 | unbind_from_irqhandler(vif->ctrl_irq, vif); |
| 610 | vif->ctrl_irq = 0; |
| 611 | |
| 612 | err_unmap: |
| 613 | xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif), |
| 614 | vif->ctrl.sring); |
| 615 | vif->ctrl.sring = NULL; |
| 616 | |
| 617 | err: |
| 618 | return err; |
| 619 | } |
| 620 | |
| 621 | int xenvif_connect_data(struct xenvif_queue *queue, |
| 622 | unsigned long tx_ring_ref, |
| 623 | unsigned long rx_ring_ref, |
| 624 | unsigned int tx_evtchn, |
| 625 | unsigned int rx_evtchn) |
| 626 | { |
| 627 | struct task_struct *task; |
| 628 | int err = -ENOMEM; |
| 629 | |
| 630 | BUG_ON(queue->tx_irq); |
| 631 | BUG_ON(queue->task); |
| 632 | BUG_ON(queue->dealloc_task); |
| 633 | |
| 634 | err = xenvif_map_frontend_data_rings(queue, tx_ring_ref, |
| 635 | rx_ring_ref); |
| 636 | if (err < 0) |
| 637 | goto err; |
| 638 | |
| 639 | init_waitqueue_head(&queue->wq); |
| 640 | init_waitqueue_head(&queue->dealloc_wq); |
| 641 | atomic_set(&queue->inflight_packets, 0); |
| 642 | |
| 643 | netif_napi_add(queue->vif->dev, &queue->napi, xenvif_poll, |
| 644 | XENVIF_NAPI_WEIGHT); |
| 645 | |
| 646 | if (tx_evtchn == rx_evtchn) { |
| 647 | /* feature-split-event-channels == 0 */ |
| 648 | err = bind_interdomain_evtchn_to_irqhandler( |
| 649 | queue->vif->domid, tx_evtchn, xenvif_interrupt, 0, |
| 650 | queue->name, queue); |
| 651 | if (err < 0) |
| 652 | goto err_unmap; |
| 653 | queue->tx_irq = queue->rx_irq = err; |
| 654 | disable_irq(queue->tx_irq); |
| 655 | } else { |
| 656 | /* feature-split-event-channels == 1 */ |
| 657 | snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name), |
| 658 | "%s-tx", queue->name); |
| 659 | err = bind_interdomain_evtchn_to_irqhandler( |
| 660 | queue->vif->domid, tx_evtchn, xenvif_tx_interrupt, 0, |
| 661 | queue->tx_irq_name, queue); |
| 662 | if (err < 0) |
| 663 | goto err_unmap; |
| 664 | queue->tx_irq = err; |
| 665 | disable_irq(queue->tx_irq); |
| 666 | |
| 667 | snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name), |
| 668 | "%s-rx", queue->name); |
| 669 | err = bind_interdomain_evtchn_to_irqhandler( |
| 670 | queue->vif->domid, rx_evtchn, xenvif_rx_interrupt, 0, |
| 671 | queue->rx_irq_name, queue); |
| 672 | if (err < 0) |
| 673 | goto err_tx_unbind; |
| 674 | queue->rx_irq = err; |
| 675 | disable_irq(queue->rx_irq); |
| 676 | } |
| 677 | |
| 678 | queue->stalled = true; |
| 679 | |
| 680 | task = kthread_create(xenvif_kthread_guest_rx, |
| 681 | (void *)queue, "%s-guest-rx", queue->name); |
| 682 | if (IS_ERR(task)) { |
| 683 | pr_warn("Could not allocate kthread for %s\n", queue->name); |
| 684 | err = PTR_ERR(task); |
| 685 | goto err_rx_unbind; |
| 686 | } |
| 687 | queue->task = task; |
| 688 | get_task_struct(task); |
| 689 | |
| 690 | task = kthread_create(xenvif_dealloc_kthread, |
| 691 | (void *)queue, "%s-dealloc", queue->name); |
| 692 | if (IS_ERR(task)) { |
| 693 | pr_warn("Could not allocate kthread for %s\n", queue->name); |
| 694 | err = PTR_ERR(task); |
| 695 | goto err_rx_unbind; |
| 696 | } |
| 697 | queue->dealloc_task = task; |
| 698 | |
| 699 | wake_up_process(queue->task); |
| 700 | wake_up_process(queue->dealloc_task); |
| 701 | |
| 702 | return 0; |
| 703 | |
| 704 | err_rx_unbind: |
| 705 | unbind_from_irqhandler(queue->rx_irq, queue); |
| 706 | queue->rx_irq = 0; |
| 707 | err_tx_unbind: |
| 708 | unbind_from_irqhandler(queue->tx_irq, queue); |
| 709 | queue->tx_irq = 0; |
| 710 | err_unmap: |
| 711 | xenvif_unmap_frontend_data_rings(queue); |
| 712 | netif_napi_del(&queue->napi); |
| 713 | err: |
| 714 | module_put(THIS_MODULE); |
| 715 | return err; |
| 716 | } |
| 717 | |
| 718 | void xenvif_carrier_off(struct xenvif *vif) |
| 719 | { |
| 720 | struct net_device *dev = vif->dev; |
| 721 | |
| 722 | rtnl_lock(); |
| 723 | if (test_and_clear_bit(VIF_STATUS_CONNECTED, &vif->status)) { |
| 724 | netif_carrier_off(dev); /* discard queued packets */ |
| 725 | if (netif_running(dev)) |
| 726 | xenvif_down(vif); |
| 727 | } |
| 728 | rtnl_unlock(); |
| 729 | } |
| 730 | |
| 731 | void xenvif_disconnect_data(struct xenvif *vif) |
| 732 | { |
| 733 | struct xenvif_queue *queue = NULL; |
| 734 | unsigned int num_queues = vif->num_queues; |
| 735 | unsigned int queue_index; |
| 736 | |
| 737 | xenvif_carrier_off(vif); |
| 738 | |
| 739 | for (queue_index = 0; queue_index < num_queues; ++queue_index) { |
| 740 | queue = &vif->queues[queue_index]; |
| 741 | |
| 742 | netif_napi_del(&queue->napi); |
| 743 | |
| 744 | if (queue->task) { |
| 745 | kthread_stop(queue->task); |
| 746 | put_task_struct(queue->task); |
| 747 | queue->task = NULL; |
| 748 | } |
| 749 | |
| 750 | if (queue->dealloc_task) { |
| 751 | kthread_stop(queue->dealloc_task); |
| 752 | queue->dealloc_task = NULL; |
| 753 | } |
| 754 | |
| 755 | if (queue->tx_irq) { |
| 756 | if (queue->tx_irq == queue->rx_irq) |
| 757 | unbind_from_irqhandler(queue->tx_irq, queue); |
| 758 | else { |
| 759 | unbind_from_irqhandler(queue->tx_irq, queue); |
| 760 | unbind_from_irqhandler(queue->rx_irq, queue); |
| 761 | } |
| 762 | queue->tx_irq = 0; |
| 763 | } |
| 764 | |
| 765 | xenvif_unmap_frontend_data_rings(queue); |
| 766 | } |
| 767 | |
| 768 | xenvif_mcast_addr_list_free(vif); |
| 769 | } |
| 770 | |
| 771 | void xenvif_disconnect_ctrl(struct xenvif *vif) |
| 772 | { |
| 773 | if (vif->ctrl_irq) { |
| 774 | xenvif_deinit_hash(vif); |
| 775 | unbind_from_irqhandler(vif->ctrl_irq, vif); |
| 776 | vif->ctrl_irq = 0; |
| 777 | } |
| 778 | |
| 779 | if (vif->ctrl.sring) { |
| 780 | xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif), |
| 781 | vif->ctrl.sring); |
| 782 | vif->ctrl.sring = NULL; |
| 783 | } |
| 784 | } |
| 785 | |
| 786 | /* Reverse the relevant parts of xenvif_init_queue(). |
| 787 | * Used for queue teardown from xenvif_free(), and on the |
| 788 | * error handling paths in xenbus.c:connect(). |
| 789 | */ |
| 790 | void xenvif_deinit_queue(struct xenvif_queue *queue) |
| 791 | { |
| 792 | gnttab_free_pages(MAX_PENDING_REQS, queue->mmap_pages); |
| 793 | } |
| 794 | |
| 795 | void xenvif_free(struct xenvif *vif) |
| 796 | { |
| 797 | struct xenvif_queue *queues = vif->queues; |
| 798 | unsigned int num_queues = vif->num_queues; |
| 799 | unsigned int queue_index; |
| 800 | |
| 801 | unregister_netdev(vif->dev); |
| 802 | free_netdev(vif->dev); |
| 803 | |
| 804 | for (queue_index = 0; queue_index < num_queues; ++queue_index) |
| 805 | xenvif_deinit_queue(&queues[queue_index]); |
| 806 | vfree(queues); |
| 807 | |
| 808 | module_put(THIS_MODULE); |
| 809 | } |