2 * Network-device interface management.
4 * Copyright (c) 2004-2005, Keir Fraser
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:
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:
19 * The above copyright notice and this permission notice shall be included in
20 * all copies or substantial portions of the Software.
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
33 #include <linux/kthread.h>
34 #include <linux/ethtool.h>
35 #include <linux/rtnetlink.h>
36 #include <linux/if_vlan.h>
37 #include <linux/vmalloc.h>
39 #include <xen/events.h>
40 #include <asm/xen/hypercall.h>
41 #include <xen/balloon.h>
43 #define XENVIF_QUEUE_LENGTH 32
44 #define XENVIF_NAPI_WEIGHT 64
46 int xenvif_schedulable(struct xenvif *vif)
48 return netif_running(vif->dev) && netif_carrier_ok(vif->dev);
51 static irqreturn_t xenvif_tx_interrupt(int irq, void *dev_id)
53 struct xenvif *vif = dev_id;
55 if (RING_HAS_UNCONSUMED_REQUESTS(&vif->tx))
56 napi_schedule(&vif->napi);
61 static int xenvif_poll(struct napi_struct *napi, int budget)
63 struct xenvif *vif = container_of(napi, struct xenvif, napi);
66 /* This vif is rogue, we pretend we've there is nothing to do
67 * for this vif to deschedule it from NAPI. But this interface
68 * will be turned off in thread context later.
70 if (unlikely(vif->disabled)) {
75 work_done = xenvif_tx_action(vif, budget);
77 if (work_done < budget) {
79 xenvif_napi_schedule_or_enable_events(vif);
85 static irqreturn_t xenvif_rx_interrupt(int irq, void *dev_id)
87 struct xenvif *vif = dev_id;
89 xenvif_kick_thread(vif);
94 static irqreturn_t xenvif_interrupt(int irq, void *dev_id)
96 xenvif_tx_interrupt(irq, dev_id);
97 xenvif_rx_interrupt(irq, dev_id);
102 static void xenvif_wake_queue(unsigned long data)
104 struct xenvif *vif = (struct xenvif *)data;
106 if (netif_queue_stopped(vif->dev)) {
107 netdev_err(vif->dev, "draining TX queue\n");
108 vif->rx_queue_purge = true;
109 xenvif_kick_thread(vif);
110 netif_wake_queue(vif->dev);
114 static int xenvif_start_xmit(struct sk_buff *skb, struct net_device *dev)
116 struct xenvif *vif = netdev_priv(dev);
117 int min_slots_needed;
119 BUG_ON(skb->dev != dev);
121 /* Drop the packet if vif is not ready */
122 if (vif->task == NULL ||
123 vif->dealloc_task == NULL ||
124 !xenvif_schedulable(vif))
127 /* At best we'll need one slot for the header and one for each
130 min_slots_needed = 1 + skb_shinfo(skb)->nr_frags;
132 /* If the skb is GSO then we'll also need an extra slot for the
138 /* If the skb can't possibly fit in the remaining slots
139 * then turn off the queue to give the ring a chance to
142 if (!xenvif_rx_ring_slots_available(vif, min_slots_needed)) {
143 vif->wake_queue.function = xenvif_wake_queue;
144 vif->wake_queue.data = (unsigned long)vif;
145 xenvif_stop_queue(vif);
146 mod_timer(&vif->wake_queue,
147 jiffies + rx_drain_timeout_jiffies);
150 skb_queue_tail(&vif->rx_queue, skb);
151 xenvif_kick_thread(vif);
156 vif->dev->stats.tx_dropped++;
161 static struct net_device_stats *xenvif_get_stats(struct net_device *dev)
163 struct xenvif *vif = netdev_priv(dev);
164 return &vif->dev->stats;
167 static void xenvif_up(struct xenvif *vif)
169 napi_enable(&vif->napi);
170 enable_irq(vif->tx_irq);
171 if (vif->tx_irq != vif->rx_irq)
172 enable_irq(vif->rx_irq);
173 xenvif_napi_schedule_or_enable_events(vif);
176 static void xenvif_down(struct xenvif *vif)
178 napi_disable(&vif->napi);
179 disable_irq(vif->tx_irq);
180 if (vif->tx_irq != vif->rx_irq)
181 disable_irq(vif->rx_irq);
182 del_timer_sync(&vif->credit_timeout);
185 static int xenvif_open(struct net_device *dev)
187 struct xenvif *vif = netdev_priv(dev);
188 if (netif_carrier_ok(dev))
190 netif_start_queue(dev);
194 static int xenvif_close(struct net_device *dev)
196 struct xenvif *vif = netdev_priv(dev);
197 if (netif_carrier_ok(dev))
199 netif_stop_queue(dev);
203 static int xenvif_change_mtu(struct net_device *dev, int mtu)
205 struct xenvif *vif = netdev_priv(dev);
206 int max = vif->can_sg ? 65535 - VLAN_ETH_HLEN : ETH_DATA_LEN;
214 static netdev_features_t xenvif_fix_features(struct net_device *dev,
215 netdev_features_t features)
217 struct xenvif *vif = netdev_priv(dev);
220 features &= ~NETIF_F_SG;
221 if (~(vif->gso_mask | vif->gso_prefix_mask) & GSO_BIT(TCPV4))
222 features &= ~NETIF_F_TSO;
223 if (~(vif->gso_mask | vif->gso_prefix_mask) & GSO_BIT(TCPV6))
224 features &= ~NETIF_F_TSO6;
226 features &= ~NETIF_F_IP_CSUM;
228 features &= ~NETIF_F_IPV6_CSUM;
233 static const struct xenvif_stat {
234 char name[ETH_GSTRING_LEN];
238 "rx_gso_checksum_fixup",
239 offsetof(struct xenvif, rx_gso_checksum_fixup)
241 /* If (sent != success + fail), there are probably packets never
246 offsetof(struct xenvif, tx_zerocopy_sent),
249 "tx_zerocopy_success",
250 offsetof(struct xenvif, tx_zerocopy_success),
254 offsetof(struct xenvif, tx_zerocopy_fail)
256 /* Number of packets exceeding MAX_SKB_FRAG slots. You should use
257 * a guest with the same MAX_SKB_FRAG
261 offsetof(struct xenvif, tx_frag_overflow)
265 static int xenvif_get_sset_count(struct net_device *dev, int string_set)
267 switch (string_set) {
269 return ARRAY_SIZE(xenvif_stats);
275 static void xenvif_get_ethtool_stats(struct net_device *dev,
276 struct ethtool_stats *stats, u64 * data)
278 void *vif = netdev_priv(dev);
281 for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++)
282 data[i] = *(unsigned long *)(vif + xenvif_stats[i].offset);
285 static void xenvif_get_strings(struct net_device *dev, u32 stringset, u8 * data)
291 for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++)
292 memcpy(data + i * ETH_GSTRING_LEN,
293 xenvif_stats[i].name, ETH_GSTRING_LEN);
298 static const struct ethtool_ops xenvif_ethtool_ops = {
299 .get_link = ethtool_op_get_link,
301 .get_sset_count = xenvif_get_sset_count,
302 .get_ethtool_stats = xenvif_get_ethtool_stats,
303 .get_strings = xenvif_get_strings,
306 static const struct net_device_ops xenvif_netdev_ops = {
307 .ndo_start_xmit = xenvif_start_xmit,
308 .ndo_get_stats = xenvif_get_stats,
309 .ndo_open = xenvif_open,
310 .ndo_stop = xenvif_close,
311 .ndo_change_mtu = xenvif_change_mtu,
312 .ndo_fix_features = xenvif_fix_features,
313 .ndo_set_mac_address = eth_mac_addr,
314 .ndo_validate_addr = eth_validate_addr,
317 struct xenvif *xenvif_alloc(struct device *parent, domid_t domid,
321 struct net_device *dev;
323 char name[IFNAMSIZ] = {};
326 snprintf(name, IFNAMSIZ - 1, "vif%u.%u", domid, handle);
327 dev = alloc_netdev(sizeof(struct xenvif), name, ether_setup);
329 pr_warn("Could not allocate netdev for %s\n", name);
330 return ERR_PTR(-ENOMEM);
333 SET_NETDEV_DEV(dev, parent);
335 vif = netdev_priv(dev);
338 vif->handle = handle;
343 vif->disabled = false;
345 vif->credit_bytes = vif->remaining_credit = ~0UL;
346 vif->credit_usec = 0UL;
347 init_timer(&vif->credit_timeout);
348 vif->credit_window_start = get_jiffies_64();
350 init_timer(&vif->wake_queue);
352 dev->netdev_ops = &xenvif_netdev_ops;
353 dev->hw_features = NETIF_F_SG |
354 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
355 NETIF_F_TSO | NETIF_F_TSO6;
356 dev->features = dev->hw_features | NETIF_F_RXCSUM;
357 dev->ethtool_ops = &xenvif_ethtool_ops;
359 dev->tx_queue_len = XENVIF_QUEUE_LENGTH;
361 skb_queue_head_init(&vif->rx_queue);
362 skb_queue_head_init(&vif->tx_queue);
364 vif->pending_cons = 0;
365 vif->pending_prod = MAX_PENDING_REQS;
366 for (i = 0; i < MAX_PENDING_REQS; i++)
367 vif->pending_ring[i] = i;
368 spin_lock_init(&vif->callback_lock);
369 spin_lock_init(&vif->response_lock);
370 /* If ballooning is disabled, this will consume real memory, so you
371 * better enable it. The long term solution would be to use just a
372 * bunch of valid page descriptors, without dependency on ballooning
374 err = alloc_xenballooned_pages(MAX_PENDING_REQS,
378 netdev_err(dev, "Could not reserve mmap_pages\n");
379 return ERR_PTR(-ENOMEM);
381 for (i = 0; i < MAX_PENDING_REQS; i++) {
382 vif->pending_tx_info[i].callback_struct = (struct ubuf_info)
383 { .callback = xenvif_zerocopy_callback,
386 vif->grant_tx_handle[i] = NETBACK_INVALID_HANDLE;
390 * Initialise a dummy MAC address. We choose the numerically
391 * largest non-broadcast address to prevent the address getting
392 * stolen by an Ethernet bridge for STP purposes.
393 * (FE:FF:FF:FF:FF:FF)
395 memset(dev->dev_addr, 0xFF, ETH_ALEN);
396 dev->dev_addr[0] &= ~0x01;
398 netif_napi_add(dev, &vif->napi, xenvif_poll, XENVIF_NAPI_WEIGHT);
400 netif_carrier_off(dev);
402 err = register_netdev(dev);
404 netdev_warn(dev, "Could not register device: err=%d\n", err);
409 netdev_dbg(dev, "Successfully created xenvif\n");
411 __module_get(THIS_MODULE);
416 int xenvif_connect(struct xenvif *vif, unsigned long tx_ring_ref,
417 unsigned long rx_ring_ref, unsigned int tx_evtchn,
418 unsigned int rx_evtchn)
420 struct task_struct *task;
425 BUG_ON(vif->dealloc_task);
427 err = xenvif_map_frontend_rings(vif, tx_ring_ref, rx_ring_ref);
431 init_waitqueue_head(&vif->wq);
432 init_waitqueue_head(&vif->dealloc_wq);
434 if (tx_evtchn == rx_evtchn) {
435 /* feature-split-event-channels == 0 */
436 err = bind_interdomain_evtchn_to_irqhandler(
437 vif->domid, tx_evtchn, xenvif_interrupt, 0,
438 vif->dev->name, vif);
441 vif->tx_irq = vif->rx_irq = err;
442 disable_irq(vif->tx_irq);
444 /* feature-split-event-channels == 1 */
445 snprintf(vif->tx_irq_name, sizeof(vif->tx_irq_name),
446 "%s-tx", vif->dev->name);
447 err = bind_interdomain_evtchn_to_irqhandler(
448 vif->domid, tx_evtchn, xenvif_tx_interrupt, 0,
449 vif->tx_irq_name, vif);
453 disable_irq(vif->tx_irq);
455 snprintf(vif->rx_irq_name, sizeof(vif->rx_irq_name),
456 "%s-rx", vif->dev->name);
457 err = bind_interdomain_evtchn_to_irqhandler(
458 vif->domid, rx_evtchn, xenvif_rx_interrupt, 0,
459 vif->rx_irq_name, vif);
463 disable_irq(vif->rx_irq);
466 task = kthread_create(xenvif_kthread_guest_rx,
467 (void *)vif, "%s-guest-rx", vif->dev->name);
469 pr_warn("Could not allocate kthread for %s\n", vif->dev->name);
476 task = kthread_create(xenvif_dealloc_kthread,
477 (void *)vif, "%s-dealloc", vif->dev->name);
479 pr_warn("Could not allocate kthread for %s\n", vif->dev->name);
484 vif->dealloc_task = task;
487 if (!vif->can_sg && vif->dev->mtu > ETH_DATA_LEN)
488 dev_set_mtu(vif->dev, ETH_DATA_LEN);
489 netdev_update_features(vif->dev);
490 netif_carrier_on(vif->dev);
491 if (netif_running(vif->dev))
495 wake_up_process(vif->task);
496 wake_up_process(vif->dealloc_task);
501 unbind_from_irqhandler(vif->rx_irq, vif);
504 unbind_from_irqhandler(vif->tx_irq, vif);
507 xenvif_unmap_frontend_rings(vif);
509 module_put(THIS_MODULE);
513 void xenvif_carrier_off(struct xenvif *vif)
515 struct net_device *dev = vif->dev;
518 netif_carrier_off(dev); /* discard queued packets */
519 if (netif_running(dev))
524 void xenvif_disconnect(struct xenvif *vif)
526 if (netif_carrier_ok(vif->dev))
527 xenvif_carrier_off(vif);
530 del_timer_sync(&vif->wake_queue);
531 kthread_stop(vif->task);
535 if (vif->dealloc_task) {
536 kthread_stop(vif->dealloc_task);
537 vif->dealloc_task = NULL;
541 if (vif->tx_irq == vif->rx_irq)
542 unbind_from_irqhandler(vif->tx_irq, vif);
544 unbind_from_irqhandler(vif->tx_irq, vif);
545 unbind_from_irqhandler(vif->rx_irq, vif);
550 xenvif_unmap_frontend_rings(vif);
553 void xenvif_free(struct xenvif *vif)
555 int i, unmap_timeout = 0;
556 /* Here we want to avoid timeout messages if an skb can be legitimately
557 * stuck somewhere else. Realistically this could be an another vif's
558 * internal or QDisc queue. That another vif also has this
559 * rx_drain_timeout_msecs timeout, but the timer only ditches the
560 * internal queue. After that, the QDisc queue can put in worst case
561 * XEN_NETIF_RX_RING_SIZE / MAX_SKB_FRAGS skbs into that another vif's
562 * internal queue, so we need several rounds of such timeouts until we
563 * can be sure that no another vif should have skb's from us. We are
564 * not sending more skb's, so newly stuck packets are not interesting
567 unsigned int worst_case_skb_lifetime = (rx_drain_timeout_msecs/1000) *
568 DIV_ROUND_UP(XENVIF_QUEUE_LENGTH, (XEN_NETIF_RX_RING_SIZE / MAX_SKB_FRAGS));
570 for (i = 0; i < MAX_PENDING_REQS; ++i) {
571 if (vif->grant_tx_handle[i] != NETBACK_INVALID_HANDLE) {
573 schedule_timeout(msecs_to_jiffies(1000));
574 if (unmap_timeout > worst_case_skb_lifetime &&
577 "Page still granted! Index: %x\n",
579 /* If there are still unmapped pages, reset the loop to
580 * start checking again. We shouldn't exit here until
581 * dealloc thread and NAPI instance release all the
582 * pages. If a kernel bug causes the skbs to stall
583 * somewhere, the interface cannot be brought down
590 free_xenballooned_pages(MAX_PENDING_REQS, vif->mmap_pages);
592 netif_napi_del(&vif->napi);
594 unregister_netdev(vif->dev);
596 free_netdev(vif->dev);
598 module_put(THIS_MODULE);