blob: 8397f6c9245158e8b3ff005bc58a419e4250169d [file] [log] [blame]
Ian Campbellf942dc22011-03-15 00:06:18 +00001/*
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
Wei Liub3f980b2013-08-26 12:59:38 +010033#include <linux/kthread.h>
Ingo Molnar0881e7b2017-02-05 15:30:50 +010034#include <linux/sched/task.h>
Ian Campbellf942dc22011-03-15 00:06:18 +000035#include <linux/ethtool.h>
36#include <linux/rtnetlink.h>
37#include <linux/if_vlan.h>
Arnd Bergmanne7b599d2014-06-10 10:34:36 +020038#include <linux/vmalloc.h>
Ian Campbellf942dc22011-03-15 00:06:18 +000039
40#include <xen/events.h>
41#include <asm/xen/hypercall.h>
Zoltan Kissf53c3fe2014-03-06 21:48:26 +000042#include <xen/balloon.h>
Ian Campbellf942dc22011-03-15 00:06:18 +000043
44#define XENVIF_QUEUE_LENGTH 32
Wei Liub3f980b2013-08-26 12:59:38 +010045#define XENVIF_NAPI_WEIGHT 64
Ian Campbellf942dc22011-03-15 00:06:18 +000046
David Vrabelf48da8b2014-10-22 14:08:54 +010047/* 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
Wei Liua64bd932014-08-12 11:48:07 +010050/* 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 */
55void 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
62void xenvif_skb_zerocopy_complete(struct xenvif_queue *queue)
63{
64 atomic_dec(&queue->inflight_packets);
Ross Lagerwall57b22902015-08-04 15:40:59 +010065
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);
Wei Liua64bd932014-08-12 11:48:07 +010071}
72
Ian Campbellf942dc22011-03-15 00:06:18 +000073int xenvif_schedulable(struct xenvif *vif)
74{
Zoltan Kiss3d1af1d2014-08-04 16:20:57 +010075 return netif_running(vif->dev) &&
David Vrabelf48da8b2014-10-22 14:08:54 +010076 test_bit(VIF_STATUS_CONNECTED, &vif->status) &&
77 !vif->disabled;
Ian Campbellf942dc22011-03-15 00:06:18 +000078}
79
Wei Liue1f00a692013-05-22 06:34:45 +000080static irqreturn_t xenvif_tx_interrupt(int irq, void *dev_id)
Ian Campbellf942dc22011-03-15 00:06:18 +000081{
Wei Liue9ce7cb2014-06-04 10:30:42 +010082 struct xenvif_queue *queue = dev_id;
Ian Campbellf942dc22011-03-15 00:06:18 +000083
Wei Liue9ce7cb2014-06-04 10:30:42 +010084 if (RING_HAS_UNCONSUMED_REQUESTS(&queue->tx))
85 napi_schedule(&queue->napi);
Ian Campbellf942dc22011-03-15 00:06:18 +000086
Wei Liue1f00a692013-05-22 06:34:45 +000087 return IRQ_HANDLED;
88}
89
Lad, Prabhakar38741d52015-02-05 13:38:07 +000090static int xenvif_poll(struct napi_struct *napi, int budget)
Wei Liub3f980b2013-08-26 12:59:38 +010091{
Wei Liue9ce7cb2014-06-04 10:30:42 +010092 struct xenvif_queue *queue =
93 container_of(napi, struct xenvif_queue, napi);
Wei Liub3f980b2013-08-26 12:59:38 +010094 int work_done;
95
Wei Liue9d8b2c2014-04-01 12:46:12 +010096 /* 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 */
Zoltan Kiss2561cc12014-08-11 13:01:44 +0100100 if (unlikely(queue->vif->disabled)) {
Wei Liue9d8b2c2014-04-01 12:46:12 +0100101 napi_complete(napi);
102 return 0;
103 }
104
Wei Liue9ce7cb2014-06-04 10:30:42 +0100105 work_done = xenvif_tx_action(queue, budget);
Wei Liub3f980b2013-08-26 12:59:38 +0100106
107 if (work_done < budget) {
Eric Dumazet6ad20162017-01-30 08:22:01 -0800108 napi_complete_done(napi, work_done);
Wei Liue9ce7cb2014-06-04 10:30:42 +0100109 xenvif_napi_schedule_or_enable_events(queue);
Wei Liub3f980b2013-08-26 12:59:38 +0100110 }
111
112 return work_done;
113}
114
Wei Liue1f00a692013-05-22 06:34:45 +0000115static irqreturn_t xenvif_rx_interrupt(int irq, void *dev_id)
116{
Wei Liue9ce7cb2014-06-04 10:30:42 +0100117 struct xenvif_queue *queue = dev_id;
Wei Liue1f00a692013-05-22 06:34:45 +0000118
Wei Liue9ce7cb2014-06-04 10:30:42 +0100119 xenvif_kick_thread(queue);
Ian Campbellf942dc22011-03-15 00:06:18 +0000120
121 return IRQ_HANDLED;
122}
123
Zoltan Kissf51de242014-07-08 19:49:14 +0100124irqreturn_t xenvif_interrupt(int irq, void *dev_id)
Wei Liue1f00a692013-05-22 06:34:45 +0000125{
126 xenvif_tx_interrupt(irq, dev_id);
127 xenvif_rx_interrupt(irq, dev_id);
128
129 return IRQ_HANDLED;
130}
131
Wei Liue9ce7cb2014-06-04 10:30:42 +0100132int xenvif_queue_stopped(struct xenvif_queue *queue)
Zoltan Kiss09350782014-03-06 21:48:30 +0000133{
Wei Liue9ce7cb2014-06-04 10:30:42 +0100134 struct net_device *dev = queue->vif->dev;
135 unsigned int id = queue->id;
136 return netif_tx_queue_stopped(netdev_get_tx_queue(dev, id));
137}
Zoltan Kiss09350782014-03-06 21:48:30 +0000138
Wei Liue9ce7cb2014-06-04 10:30:42 +0100139void xenvif_wake_queue(struct xenvif_queue *queue)
140{
141 struct net_device *dev = queue->vif->dev;
142 unsigned int id = queue->id;
143 netif_tx_wake_queue(netdev_get_tx_queue(dev, id));
144}
145
Paul Durrant40d8abd2016-05-13 09:37:27 +0100146static u16 xenvif_select_queue(struct net_device *dev, struct sk_buff *skb,
147 void *accel_priv,
148 select_queue_fallback_t fallback)
149{
150 struct xenvif *vif = netdev_priv(dev);
151 unsigned int size = vif->hash.size;
152
Paul Durrant912e27e2016-10-07 09:32:31 +0100153 if (vif->hash.alg == XEN_NETIF_CTRL_HASH_ALGORITHM_NONE)
154 return fallback(dev, skb) % dev->real_num_tx_queues;
Paul Durrant40d8abd2016-05-13 09:37:27 +0100155
156 xenvif_set_skb_hash(vif, skb);
157
158 if (size == 0)
159 return skb_get_hash_raw(skb) % dev->real_num_tx_queues;
160
161 return vif->hash.mapping[skb_get_hash_raw(skb) % size];
162}
163
Ian Campbellf942dc22011-03-15 00:06:18 +0000164static int xenvif_start_xmit(struct sk_buff *skb, struct net_device *dev)
165{
166 struct xenvif *vif = netdev_priv(dev);
Wei Liue9ce7cb2014-06-04 10:30:42 +0100167 struct xenvif_queue *queue = NULL;
Igor Druzhininb17075d2017-03-10 21:36:22 +0000168 unsigned int num_queues;
Wei Liue9ce7cb2014-06-04 10:30:42 +0100169 u16 index;
David Vrabelf48da8b2014-10-22 14:08:54 +0100170 struct xenvif_rx_cb *cb;
Ian Campbellf942dc22011-03-15 00:06:18 +0000171
172 BUG_ON(skb->dev != dev);
173
Igor Druzhininb17075d2017-03-10 21:36:22 +0000174 /* Drop the packet if queues are not set up.
175 * This handler should be called inside an RCU read section
176 * so we don't need to enter it here explicitly.
177 */
178 num_queues = READ_ONCE(vif->num_queues);
Wei Liue9ce7cb2014-06-04 10:30:42 +0100179 if (num_queues < 1)
180 goto drop;
181
182 /* Obtain the queue to be used to transmit this packet */
183 index = skb_get_queue_mapping(skb);
184 if (index >= num_queues) {
185 pr_warn_ratelimited("Invalid queue %hu for packet on interface %s\n.",
186 index, vif->dev->name);
187 index %= num_queues;
188 }
189 queue = &vif->queues[index];
190
191 /* Drop the packet if queue is not ready */
192 if (queue->task == NULL ||
193 queue->dealloc_task == NULL ||
Zoltan Kissf53c3fe2014-03-06 21:48:26 +0000194 !xenvif_schedulable(vif))
Ian Campbellf942dc22011-03-15 00:06:18 +0000195 goto drop;
196
Paul Durrant210c34d2015-09-02 17:58:36 +0100197 if (vif->multicast_control && skb->pkt_type == PACKET_MULTICAST) {
198 struct ethhdr *eth = (struct ethhdr *)skb->data;
199
200 if (!xenvif_mcast_match(vif, eth->h_dest))
201 goto drop;
202 }
203
David Vrabelf48da8b2014-10-22 14:08:54 +0100204 cb = XENVIF_RX_CB(skb);
David Vrabel26c0e102014-12-18 11:13:06 +0000205 cb->expires = jiffies + vif->drain_timeout;
Ian Campbellf942dc22011-03-15 00:06:18 +0000206
Paul Durrant912e27e2016-10-07 09:32:31 +0100207 /* If there is no hash algorithm configured then make sure there
208 * is no hash information in the socket buffer otherwise it
209 * would be incorrectly forwarded to the frontend.
210 */
211 if (vif->hash.alg == XEN_NETIF_CTRL_HASH_ALGORITHM_NONE)
212 skb_clear_hash(skb);
213
David Vrabelf48da8b2014-10-22 14:08:54 +0100214 xenvif_rx_queue_tail(queue, skb);
Wei Liue9ce7cb2014-06-04 10:30:42 +0100215 xenvif_kick_thread(queue);
Ian Campbellf942dc22011-03-15 00:06:18 +0000216
217 return NETDEV_TX_OK;
218
219 drop:
220 vif->dev->stats.tx_dropped++;
221 dev_kfree_skb(skb);
222 return NETDEV_TX_OK;
223}
224
Ian Campbellf942dc22011-03-15 00:06:18 +0000225static struct net_device_stats *xenvif_get_stats(struct net_device *dev)
226{
227 struct xenvif *vif = netdev_priv(dev);
Wei Liue9ce7cb2014-06-04 10:30:42 +0100228 struct xenvif_queue *queue = NULL;
Igor Druzhininb17075d2017-03-10 21:36:22 +0000229 unsigned int num_queues;
Mart van Santenebf692f2017-02-10 12:02:18 +0000230 u64 rx_bytes = 0;
231 u64 rx_packets = 0;
232 u64 tx_bytes = 0;
233 u64 tx_packets = 0;
Wei Liue9ce7cb2014-06-04 10:30:42 +0100234 unsigned int index;
235
Igor Druzhininb17075d2017-03-10 21:36:22 +0000236 rcu_read_lock();
237 num_queues = READ_ONCE(vif->num_queues);
Wei Liue9ce7cb2014-06-04 10:30:42 +0100238
239 /* Aggregate tx and rx stats from each queue */
Igor Druzhininb17075d2017-03-10 21:36:22 +0000240 for (index = 0; index < num_queues; ++index) {
Wei Liue9ce7cb2014-06-04 10:30:42 +0100241 queue = &vif->queues[index];
242 rx_bytes += queue->stats.rx_bytes;
243 rx_packets += queue->stats.rx_packets;
244 tx_bytes += queue->stats.tx_bytes;
245 tx_packets += queue->stats.tx_packets;
246 }
247
Igor Druzhininb17075d2017-03-10 21:36:22 +0000248 rcu_read_unlock();
Igor Druzhininf16f1df2017-01-17 20:49:38 +0000249
Wei Liue9ce7cb2014-06-04 10:30:42 +0100250 vif->dev->stats.rx_bytes = rx_bytes;
251 vif->dev->stats.rx_packets = rx_packets;
252 vif->dev->stats.tx_bytes = tx_bytes;
253 vif->dev->stats.tx_packets = tx_packets;
254
Ian Campbellf942dc22011-03-15 00:06:18 +0000255 return &vif->dev->stats;
256}
257
258static void xenvif_up(struct xenvif *vif)
259{
Wei Liue9ce7cb2014-06-04 10:30:42 +0100260 struct xenvif_queue *queue = NULL;
Wei Liuf7b50c42014-06-23 10:50:17 +0100261 unsigned int num_queues = vif->num_queues;
Wei Liue9ce7cb2014-06-04 10:30:42 +0100262 unsigned int queue_index;
263
264 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
265 queue = &vif->queues[queue_index];
266 napi_enable(&queue->napi);
267 enable_irq(queue->tx_irq);
268 if (queue->tx_irq != queue->rx_irq)
269 enable_irq(queue->rx_irq);
270 xenvif_napi_schedule_or_enable_events(queue);
271 }
Ian Campbellf942dc22011-03-15 00:06:18 +0000272}
273
274static void xenvif_down(struct xenvif *vif)
275{
Wei Liue9ce7cb2014-06-04 10:30:42 +0100276 struct xenvif_queue *queue = NULL;
Wei Liuf7b50c42014-06-23 10:50:17 +0100277 unsigned int num_queues = vif->num_queues;
Wei Liue9ce7cb2014-06-04 10:30:42 +0100278 unsigned int queue_index;
279
280 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
281 queue = &vif->queues[queue_index];
Wei Liue9ce7cb2014-06-04 10:30:42 +0100282 disable_irq(queue->tx_irq);
283 if (queue->tx_irq != queue->rx_irq)
284 disable_irq(queue->rx_irq);
Zoltan Kiss8fe78982014-10-28 15:29:30 +0000285 napi_disable(&queue->napi);
Wei Liue9ce7cb2014-06-04 10:30:42 +0100286 del_timer_sync(&queue->credit_timeout);
287 }
Ian Campbellf942dc22011-03-15 00:06:18 +0000288}
289
290static int xenvif_open(struct net_device *dev)
291{
292 struct xenvif *vif = netdev_priv(dev);
Zoltan Kiss3d1af1d2014-08-04 16:20:57 +0100293 if (test_bit(VIF_STATUS_CONNECTED, &vif->status))
Ian Campbellf942dc22011-03-15 00:06:18 +0000294 xenvif_up(vif);
Wei Liue9ce7cb2014-06-04 10:30:42 +0100295 netif_tx_start_all_queues(dev);
Ian Campbellf942dc22011-03-15 00:06:18 +0000296 return 0;
297}
298
299static int xenvif_close(struct net_device *dev)
300{
301 struct xenvif *vif = netdev_priv(dev);
Zoltan Kiss3d1af1d2014-08-04 16:20:57 +0100302 if (test_bit(VIF_STATUS_CONNECTED, &vif->status))
Ian Campbellf942dc22011-03-15 00:06:18 +0000303 xenvif_down(vif);
Wei Liue9ce7cb2014-06-04 10:30:42 +0100304 netif_tx_stop_all_queues(dev);
Ian Campbellf942dc22011-03-15 00:06:18 +0000305 return 0;
306}
307
308static int xenvif_change_mtu(struct net_device *dev, int mtu)
309{
310 struct xenvif *vif = netdev_priv(dev);
Jarod Wilsond0c2c992016-10-20 13:55:21 -0400311 int max = vif->can_sg ? ETH_MAX_MTU - VLAN_ETH_HLEN : ETH_DATA_LEN;
Ian Campbellf942dc22011-03-15 00:06:18 +0000312
313 if (mtu > max)
314 return -EINVAL;
315 dev->mtu = mtu;
316 return 0;
317}
318
Michał Mirosławc8f44af2011-11-15 15:29:55 +0000319static netdev_features_t xenvif_fix_features(struct net_device *dev,
320 netdev_features_t features)
Ian Campbellf942dc22011-03-15 00:06:18 +0000321{
322 struct xenvif *vif = netdev_priv(dev);
Ian Campbellf942dc22011-03-15 00:06:18 +0000323
Michał Mirosław47103042011-04-19 03:35:06 +0000324 if (!vif->can_sg)
325 features &= ~NETIF_F_SG;
Paul Durrantfedbc8c2016-10-04 10:29:13 +0100326 if (~(vif->gso_mask) & GSO_BIT(TCPV4))
Michał Mirosław47103042011-04-19 03:35:06 +0000327 features &= ~NETIF_F_TSO;
Paul Durrantfedbc8c2016-10-04 10:29:13 +0100328 if (~(vif->gso_mask) & GSO_BIT(TCPV6))
Paul Durrant82cada22013-10-16 17:50:32 +0100329 features &= ~NETIF_F_TSO6;
Paul Durrant146c8a72013-10-16 17:50:28 +0100330 if (!vif->ip_csum)
Michał Mirosław47103042011-04-19 03:35:06 +0000331 features &= ~NETIF_F_IP_CSUM;
Paul Durrant146c8a72013-10-16 17:50:28 +0100332 if (!vif->ipv6_csum)
333 features &= ~NETIF_F_IPV6_CSUM;
Ian Campbellf942dc22011-03-15 00:06:18 +0000334
Michał Mirosław47103042011-04-19 03:35:06 +0000335 return features;
Ian Campbellf942dc22011-03-15 00:06:18 +0000336}
337
338static const struct xenvif_stat {
339 char name[ETH_GSTRING_LEN];
340 u16 offset;
341} xenvif_stats[] = {
342 {
343 "rx_gso_checksum_fixup",
Wei Liue9ce7cb2014-06-04 10:30:42 +0100344 offsetof(struct xenvif_stats, rx_gso_checksum_fixup)
Ian Campbellf942dc22011-03-15 00:06:18 +0000345 },
Zoltan Kiss1bb332a2014-03-06 21:48:28 +0000346 /* If (sent != success + fail), there are probably packets never
347 * freed up properly!
348 */
349 {
350 "tx_zerocopy_sent",
Wei Liue9ce7cb2014-06-04 10:30:42 +0100351 offsetof(struct xenvif_stats, tx_zerocopy_sent),
Zoltan Kiss1bb332a2014-03-06 21:48:28 +0000352 },
353 {
354 "tx_zerocopy_success",
Wei Liue9ce7cb2014-06-04 10:30:42 +0100355 offsetof(struct xenvif_stats, tx_zerocopy_success),
Zoltan Kiss1bb332a2014-03-06 21:48:28 +0000356 },
357 {
358 "tx_zerocopy_fail",
Wei Liue9ce7cb2014-06-04 10:30:42 +0100359 offsetof(struct xenvif_stats, tx_zerocopy_fail)
Zoltan Kiss1bb332a2014-03-06 21:48:28 +0000360 },
Zoltan Kisse3377f32014-03-06 21:48:29 +0000361 /* Number of packets exceeding MAX_SKB_FRAG slots. You should use
362 * a guest with the same MAX_SKB_FRAG
363 */
364 {
365 "tx_frag_overflow",
Wei Liue9ce7cb2014-06-04 10:30:42 +0100366 offsetof(struct xenvif_stats, tx_frag_overflow)
Zoltan Kisse3377f32014-03-06 21:48:29 +0000367 },
Ian Campbellf942dc22011-03-15 00:06:18 +0000368};
369
370static int xenvif_get_sset_count(struct net_device *dev, int string_set)
371{
372 switch (string_set) {
373 case ETH_SS_STATS:
374 return ARRAY_SIZE(xenvif_stats);
375 default:
376 return -EINVAL;
377 }
378}
379
380static void xenvif_get_ethtool_stats(struct net_device *dev,
381 struct ethtool_stats *stats, u64 * data)
382{
Wei Liue9ce7cb2014-06-04 10:30:42 +0100383 struct xenvif *vif = netdev_priv(dev);
Igor Druzhininb17075d2017-03-10 21:36:22 +0000384 unsigned int num_queues;
Ian Campbellf942dc22011-03-15 00:06:18 +0000385 int i;
Wei Liue9ce7cb2014-06-04 10:30:42 +0100386 unsigned int queue_index;
Ian Campbellf942dc22011-03-15 00:06:18 +0000387
Igor Druzhininb17075d2017-03-10 21:36:22 +0000388 rcu_read_lock();
389 num_queues = READ_ONCE(vif->num_queues);
390
Wei Liue9ce7cb2014-06-04 10:30:42 +0100391 for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++) {
392 unsigned long accum = 0;
393 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
David Vrabeld63951d2015-03-04 11:14:46 +0000394 void *vif_stats = &vif->queues[queue_index].stats;
Wei Liue9ce7cb2014-06-04 10:30:42 +0100395 accum += *(unsigned long *)(vif_stats + xenvif_stats[i].offset);
396 }
397 data[i] = accum;
398 }
Igor Druzhininb17075d2017-03-10 21:36:22 +0000399
400 rcu_read_unlock();
Ian Campbellf942dc22011-03-15 00:06:18 +0000401}
402
403static void xenvif_get_strings(struct net_device *dev, u32 stringset, u8 * data)
404{
405 int i;
406
407 switch (stringset) {
408 case ETH_SS_STATS:
409 for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++)
410 memcpy(data + i * ETH_GSTRING_LEN,
411 xenvif_stats[i].name, ETH_GSTRING_LEN);
412 break;
413 }
414}
415
stephen hemminger813abbb2012-01-04 11:56:58 +0000416static const struct ethtool_ops xenvif_ethtool_ops = {
Ian Campbellf942dc22011-03-15 00:06:18 +0000417 .get_link = ethtool_op_get_link,
418
419 .get_sset_count = xenvif_get_sset_count,
420 .get_ethtool_stats = xenvif_get_ethtool_stats,
421 .get_strings = xenvif_get_strings,
422};
423
stephen hemminger813abbb2012-01-04 11:56:58 +0000424static const struct net_device_ops xenvif_netdev_ops = {
Paul Durrant40d8abd2016-05-13 09:37:27 +0100425 .ndo_select_queue = xenvif_select_queue,
Ian Campbellf942dc22011-03-15 00:06:18 +0000426 .ndo_start_xmit = xenvif_start_xmit,
427 .ndo_get_stats = xenvif_get_stats,
428 .ndo_open = xenvif_open,
429 .ndo_stop = xenvif_close,
430 .ndo_change_mtu = xenvif_change_mtu,
Michał Mirosław47103042011-04-19 03:35:06 +0000431 .ndo_fix_features = xenvif_fix_features,
Matt Wilson4a633a62013-01-22 08:08:25 +0000432 .ndo_set_mac_address = eth_mac_addr,
433 .ndo_validate_addr = eth_validate_addr,
Ian Campbellf942dc22011-03-15 00:06:18 +0000434};
435
436struct xenvif *xenvif_alloc(struct device *parent, domid_t domid,
437 unsigned int handle)
438{
439 int err;
440 struct net_device *dev;
441 struct xenvif *vif;
442 char name[IFNAMSIZ] = {};
443
444 snprintf(name, IFNAMSIZ - 1, "vif%u.%u", domid, handle);
Andrew J. Bennieston8d3d53b2014-06-04 10:30:43 +0100445 /* Allocate a netdev with the max. supported number of queues.
446 * When the guest selects the desired number, it will be updated
Wei Liuf7b50c42014-06-23 10:50:17 +0100447 * via netif_set_real_num_*_queues().
Andrew J. Bennieston8d3d53b2014-06-04 10:30:43 +0100448 */
Tom Gundersenc835a672014-07-14 16:37:24 +0200449 dev = alloc_netdev_mq(sizeof(struct xenvif), name, NET_NAME_UNKNOWN,
450 ether_setup, xenvif_max_queues);
Ian Campbellf942dc22011-03-15 00:06:18 +0000451 if (dev == NULL) {
Wei Liub3f980b2013-08-26 12:59:38 +0100452 pr_warn("Could not allocate netdev for %s\n", name);
Ian Campbellf942dc22011-03-15 00:06:18 +0000453 return ERR_PTR(-ENOMEM);
454 }
455
456 SET_NETDEV_DEV(dev, parent);
457
458 vif = netdev_priv(dev);
Paul Durrantac3d5ac2013-12-23 09:27:17 +0000459
Ian Campbellf942dc22011-03-15 00:06:18 +0000460 vif->domid = domid;
461 vif->handle = handle;
Ian Campbellf942dc22011-03-15 00:06:18 +0000462 vif->can_sg = 1;
Paul Durrant146c8a72013-10-16 17:50:28 +0100463 vif->ip_csum = 1;
Ian Campbellf942dc22011-03-15 00:06:18 +0000464 vif->dev = dev;
Wei Liue9d8b2c2014-04-01 12:46:12 +0100465 vif->disabled = false;
David Vrabel26c0e102014-12-18 11:13:06 +0000466 vif->drain_timeout = msecs_to_jiffies(rx_drain_timeout_msecs);
467 vif->stall_timeout = msecs_to_jiffies(rx_stall_timeout_msecs);
Wei Liue9d8b2c2014-04-01 12:46:12 +0100468
Wei Liuf7b50c42014-06-23 10:50:17 +0100469 /* Start out with no queues. */
Wei Liue9ce7cb2014-06-04 10:30:42 +0100470 vif->queues = NULL;
Wei Liuf7b50c42014-06-23 10:50:17 +0100471 vif->num_queues = 0;
Zoltan Kiss09350782014-03-06 21:48:30 +0000472
David Vrabelecf08d22014-10-22 14:08:55 +0100473 spin_lock_init(&vif->lock);
Paul Durrant210c34d2015-09-02 17:58:36 +0100474 INIT_LIST_HEAD(&vif->fe_mcast_addr);
David Vrabelecf08d22014-10-22 14:08:55 +0100475
Ian Campbellf942dc22011-03-15 00:06:18 +0000476 dev->netdev_ops = &xenvif_netdev_ops;
Paul Durrant146c8a72013-10-16 17:50:28 +0100477 dev->hw_features = NETIF_F_SG |
478 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
Ross Lagerwall2167ca02016-10-04 10:29:18 +0100479 NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_FRAGLIST;
Paul Durrant7365bcf2013-10-16 17:50:30 +0100480 dev->features = dev->hw_features | NETIF_F_RXCSUM;
Wilfried Klaebe7ad24ea2014-05-11 00:12:32 +0000481 dev->ethtool_ops = &xenvif_ethtool_ops;
Ian Campbellf942dc22011-03-15 00:06:18 +0000482
483 dev->tx_queue_len = XENVIF_QUEUE_LENGTH;
484
Jarod Wilsond0c2c992016-10-20 13:55:21 -0400485 dev->min_mtu = 0;
486 dev->max_mtu = ETH_MAX_MTU - VLAN_ETH_HLEN;
487
Ian Campbellf942dc22011-03-15 00:06:18 +0000488 /*
489 * Initialise a dummy MAC address. We choose the numerically
490 * largest non-broadcast address to prevent the address getting
491 * stolen by an Ethernet bridge for STP purposes.
492 * (FE:FF:FF:FF:FF:FF)
493 */
Joe Perches3b6ed262015-03-02 19:54:51 -0800494 eth_broadcast_addr(dev->dev_addr);
Ian Campbellf942dc22011-03-15 00:06:18 +0000495 dev->dev_addr[0] &= ~0x01;
496
497 netif_carrier_off(dev);
498
499 err = register_netdev(dev);
500 if (err) {
501 netdev_warn(dev, "Could not register device: err=%d\n", err);
502 free_netdev(dev);
503 return ERR_PTR(err);
504 }
505
506 netdev_dbg(dev, "Successfully created xenvif\n");
Paul Durrant279f4382013-09-17 17:46:08 +0100507
508 __module_get(THIS_MODULE);
509
Ian Campbellf942dc22011-03-15 00:06:18 +0000510 return vif;
511}
512
Wei Liue9ce7cb2014-06-04 10:30:42 +0100513int xenvif_init_queue(struct xenvif_queue *queue)
Ian Campbellf942dc22011-03-15 00:06:18 +0000514{
Wei Liue9ce7cb2014-06-04 10:30:42 +0100515 int err, i;
Ian Campbellf942dc22011-03-15 00:06:18 +0000516
Wei Liue9ce7cb2014-06-04 10:30:42 +0100517 queue->credit_bytes = queue->remaining_credit = ~0UL;
518 queue->credit_usec = 0UL;
519 init_timer(&queue->credit_timeout);
Palik, Imreedafc132015-03-19 11:05:42 +0100520 queue->credit_timeout.function = xenvif_tx_credit_callback;
Wei Liue9ce7cb2014-06-04 10:30:42 +0100521 queue->credit_window_start = get_jiffies_64();
Ian Campbellf942dc22011-03-15 00:06:18 +0000522
David Vrabelf48da8b2014-10-22 14:08:54 +0100523 queue->rx_queue_max = XENVIF_RX_QUEUE_BYTES;
524
Wei Liue9ce7cb2014-06-04 10:30:42 +0100525 skb_queue_head_init(&queue->rx_queue);
526 skb_queue_head_init(&queue->tx_queue);
Ian Campbellf942dc22011-03-15 00:06:18 +0000527
Wei Liue9ce7cb2014-06-04 10:30:42 +0100528 queue->pending_cons = 0;
529 queue->pending_prod = MAX_PENDING_REQS;
530 for (i = 0; i < MAX_PENDING_REQS; ++i)
531 queue->pending_ring[i] = i;
Paul Durrantca2f09f2013-12-06 16:36:07 +0000532
Wei Liue9ce7cb2014-06-04 10:30:42 +0100533 spin_lock_init(&queue->callback_lock);
534 spin_lock_init(&queue->response_lock);
Wei Liue1f00a692013-05-22 06:34:45 +0000535
Wei Liue9ce7cb2014-06-04 10:30:42 +0100536 /* If ballooning is disabled, this will consume real memory, so you
537 * better enable it. The long term solution would be to use just a
538 * bunch of valid page descriptors, without dependency on ballooning
539 */
David Vrabelff4b1562015-01-08 18:06:01 +0000540 err = gnttab_alloc_pages(MAX_PENDING_REQS,
541 queue->mmap_pages);
Wei Liue9ce7cb2014-06-04 10:30:42 +0100542 if (err) {
543 netdev_err(queue->vif->dev, "Could not reserve mmap_pages\n");
544 return -ENOMEM;
Wei Liue1f00a692013-05-22 06:34:45 +0000545 }
Ian Campbellf942dc22011-03-15 00:06:18 +0000546
Wei Liue9ce7cb2014-06-04 10:30:42 +0100547 for (i = 0; i < MAX_PENDING_REQS; i++) {
548 queue->pending_tx_info[i].callback_struct = (struct ubuf_info)
549 { .callback = xenvif_zerocopy_callback,
550 .ctx = NULL,
551 .desc = i };
552 queue->grant_tx_handle[i] = NETBACK_INVALID_HANDLE;
Wei Liub3f980b2013-08-26 12:59:38 +0100553 }
Ian Campbellf942dc22011-03-15 00:06:18 +0000554
Wei Liue9ce7cb2014-06-04 10:30:42 +0100555 return 0;
556}
Zoltan Kissf53c3fe2014-03-06 21:48:26 +0000557
Wei Liue9ce7cb2014-06-04 10:30:42 +0100558void xenvif_carrier_on(struct xenvif *vif)
559{
Ian Campbellf942dc22011-03-15 00:06:18 +0000560 rtnl_lock();
Michał Mirosław47103042011-04-19 03:35:06 +0000561 if (!vif->can_sg && vif->dev->mtu > ETH_DATA_LEN)
562 dev_set_mtu(vif->dev, ETH_DATA_LEN);
563 netdev_update_features(vif->dev);
Zoltan Kiss3d1af1d2014-08-04 16:20:57 +0100564 set_bit(VIF_STATUS_CONNECTED, &vif->status);
David Vrabeld0e5d832011-09-30 06:37:51 +0000565 if (netif_running(vif->dev))
566 xenvif_up(vif);
Ian Campbellf942dc22011-03-15 00:06:18 +0000567 rtnl_unlock();
Wei Liue9ce7cb2014-06-04 10:30:42 +0100568}
Ian Campbellf942dc22011-03-15 00:06:18 +0000569
Paul Durrant4e15ee22016-05-13 09:37:26 +0100570int xenvif_connect_ctrl(struct xenvif *vif, grant_ref_t ring_ref,
571 unsigned int evtchn)
572{
573 struct net_device *dev = vif->dev;
574 void *addr;
575 struct xen_netif_ctrl_sring *shared;
Juergen Gross0364a882016-09-22 11:06:25 +0200576 int err;
Paul Durrant4e15ee22016-05-13 09:37:26 +0100577
578 err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(vif),
579 &ring_ref, 1, &addr);
580 if (err)
581 goto err;
582
583 shared = (struct xen_netif_ctrl_sring *)addr;
584 BACK_RING_INIT(&vif->ctrl, shared, XEN_PAGE_SIZE);
585
Juergen Gross0364a882016-09-22 11:06:25 +0200586 err = bind_interdomain_evtchn_to_irq(vif->domid, evtchn);
Paul Durrant4e15ee22016-05-13 09:37:26 +0100587 if (err < 0)
588 goto err_unmap;
589
590 vif->ctrl_irq = err;
591
Paul Durrant40d8abd2016-05-13 09:37:27 +0100592 xenvif_init_hash(vif);
593
Juergen Gross0364a882016-09-22 11:06:25 +0200594 err = request_threaded_irq(vif->ctrl_irq, NULL, xenvif_ctrl_irq_fn,
595 IRQF_ONESHOT, "xen-netback-ctrl", vif);
596 if (err) {
597 pr_warn("Could not setup irq handler for %s\n", dev->name);
Paul Durrant4e15ee22016-05-13 09:37:26 +0100598 goto err_deinit;
599 }
600
Paul Durrant4e15ee22016-05-13 09:37:26 +0100601 return 0;
602
603err_deinit:
Paul Durrant40d8abd2016-05-13 09:37:27 +0100604 xenvif_deinit_hash(vif);
Paul Durrant4e15ee22016-05-13 09:37:26 +0100605 unbind_from_irqhandler(vif->ctrl_irq, vif);
606 vif->ctrl_irq = 0;
607
608err_unmap:
609 xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif),
610 vif->ctrl.sring);
611 vif->ctrl.sring = NULL;
612
613err:
614 return err;
615}
616
617int xenvif_connect_data(struct xenvif_queue *queue,
618 unsigned long tx_ring_ref,
619 unsigned long rx_ring_ref,
620 unsigned int tx_evtchn,
621 unsigned int rx_evtchn)
Wei Liue9ce7cb2014-06-04 10:30:42 +0100622{
623 struct task_struct *task;
624 int err = -ENOMEM;
625
626 BUG_ON(queue->tx_irq);
627 BUG_ON(queue->task);
628 BUG_ON(queue->dealloc_task);
629
Paul Durrant4e15ee22016-05-13 09:37:26 +0100630 err = xenvif_map_frontend_data_rings(queue, tx_ring_ref,
631 rx_ring_ref);
Wei Liue9ce7cb2014-06-04 10:30:42 +0100632 if (err < 0)
633 goto err;
634
635 init_waitqueue_head(&queue->wq);
636 init_waitqueue_head(&queue->dealloc_wq);
Wei Liua64bd932014-08-12 11:48:07 +0100637 atomic_set(&queue->inflight_packets, 0);
Wei Liue9ce7cb2014-06-04 10:30:42 +0100638
Wei Liue24f8192014-08-25 16:44:00 +0100639 netif_napi_add(queue->vif->dev, &queue->napi, xenvif_poll,
640 XENVIF_NAPI_WEIGHT);
641
Wei Liue9ce7cb2014-06-04 10:30:42 +0100642 if (tx_evtchn == rx_evtchn) {
643 /* feature-split-event-channels == 0 */
644 err = bind_interdomain_evtchn_to_irqhandler(
645 queue->vif->domid, tx_evtchn, xenvif_interrupt, 0,
646 queue->name, queue);
647 if (err < 0)
648 goto err_unmap;
649 queue->tx_irq = queue->rx_irq = err;
650 disable_irq(queue->tx_irq);
651 } else {
652 /* feature-split-event-channels == 1 */
653 snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
654 "%s-tx", queue->name);
655 err = bind_interdomain_evtchn_to_irqhandler(
656 queue->vif->domid, tx_evtchn, xenvif_tx_interrupt, 0,
657 queue->tx_irq_name, queue);
658 if (err < 0)
659 goto err_unmap;
660 queue->tx_irq = err;
661 disable_irq(queue->tx_irq);
662
663 snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
664 "%s-rx", queue->name);
665 err = bind_interdomain_evtchn_to_irqhandler(
666 queue->vif->domid, rx_evtchn, xenvif_rx_interrupt, 0,
667 queue->rx_irq_name, queue);
668 if (err < 0)
669 goto err_tx_unbind;
670 queue->rx_irq = err;
671 disable_irq(queue->rx_irq);
672 }
673
David Vrabelecf08d22014-10-22 14:08:55 +0100674 queue->stalled = true;
675
Wei Liue9ce7cb2014-06-04 10:30:42 +0100676 task = kthread_create(xenvif_kthread_guest_rx,
677 (void *)queue, "%s-guest-rx", queue->name);
678 if (IS_ERR(task)) {
679 pr_warn("Could not allocate kthread for %s\n", queue->name);
680 err = PTR_ERR(task);
681 goto err_rx_unbind;
682 }
683 queue->task = task;
David Vrabel42b52122015-02-02 16:57:51 +0000684 get_task_struct(task);
Wei Liue9ce7cb2014-06-04 10:30:42 +0100685
686 task = kthread_create(xenvif_dealloc_kthread,
687 (void *)queue, "%s-dealloc", queue->name);
688 if (IS_ERR(task)) {
689 pr_warn("Could not allocate kthread for %s\n", queue->name);
690 err = PTR_ERR(task);
691 goto err_rx_unbind;
692 }
693 queue->dealloc_task = task;
694
695 wake_up_process(queue->task);
696 wake_up_process(queue->dealloc_task);
Wei Liub3f980b2013-08-26 12:59:38 +0100697
Ian Campbellf942dc22011-03-15 00:06:18 +0000698 return 0;
Wei Liub3f980b2013-08-26 12:59:38 +0100699
700err_rx_unbind:
Wei Liue9ce7cb2014-06-04 10:30:42 +0100701 unbind_from_irqhandler(queue->rx_irq, queue);
702 queue->rx_irq = 0;
Wei Liue1f00a692013-05-22 06:34:45 +0000703err_tx_unbind:
Wei Liue9ce7cb2014-06-04 10:30:42 +0100704 unbind_from_irqhandler(queue->tx_irq, queue);
705 queue->tx_irq = 0;
Ian Campbellf942dc22011-03-15 00:06:18 +0000706err_unmap:
Paul Durrant4e15ee22016-05-13 09:37:26 +0100707 xenvif_unmap_frontend_data_rings(queue);
David Vrabel4a658522016-01-15 14:55:35 +0000708 netif_napi_del(&queue->napi);
Ian Campbellf942dc22011-03-15 00:06:18 +0000709err:
Wei Liub103f352013-05-16 23:26:11 +0000710 module_put(THIS_MODULE);
Ian Campbellf942dc22011-03-15 00:06:18 +0000711 return err;
712}
713
Ian Campbell488562862013-02-06 23:41:35 +0000714void xenvif_carrier_off(struct xenvif *vif)
Ian Campbellf942dc22011-03-15 00:06:18 +0000715{
716 struct net_device *dev = vif->dev;
Ian Campbell488562862013-02-06 23:41:35 +0000717
718 rtnl_lock();
Zoltan Kiss3d1af1d2014-08-04 16:20:57 +0100719 if (test_and_clear_bit(VIF_STATUS_CONNECTED, &vif->status)) {
720 netif_carrier_off(dev); /* discard queued packets */
721 if (netif_running(dev))
722 xenvif_down(vif);
723 }
Ian Campbell488562862013-02-06 23:41:35 +0000724 rtnl_unlock();
Ian Campbell488562862013-02-06 23:41:35 +0000725}
726
Paul Durrant4e15ee22016-05-13 09:37:26 +0100727void xenvif_disconnect_data(struct xenvif *vif)
Ian Campbell488562862013-02-06 23:41:35 +0000728{
Wei Liue9ce7cb2014-06-04 10:30:42 +0100729 struct xenvif_queue *queue = NULL;
Wei Liuf7b50c42014-06-23 10:50:17 +0100730 unsigned int num_queues = vif->num_queues;
Wei Liue9ce7cb2014-06-04 10:30:42 +0100731 unsigned int queue_index;
732
Zoltan Kiss3d1af1d2014-08-04 16:20:57 +0100733 xenvif_carrier_off(vif);
Ian Campbellf942dc22011-03-15 00:06:18 +0000734
Wei Liue9ce7cb2014-06-04 10:30:42 +0100735 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
736 queue = &vif->queues[queue_index];
David Vrabeldb739ef2013-11-21 15:26:09 +0000737
Wei Liuea2c5e12014-08-12 11:48:06 +0100738 netif_napi_del(&queue->napi);
739
Wei Liue9ce7cb2014-06-04 10:30:42 +0100740 if (queue->task) {
Wei Liue9ce7cb2014-06-04 10:30:42 +0100741 kthread_stop(queue->task);
David Vrabel42b52122015-02-02 16:57:51 +0000742 put_task_struct(queue->task);
Wei Liue9ce7cb2014-06-04 10:30:42 +0100743 queue->task = NULL;
Wei Liue1f00a692013-05-22 06:34:45 +0000744 }
Ian Campbellf942dc22011-03-15 00:06:18 +0000745
Wei Liue9ce7cb2014-06-04 10:30:42 +0100746 if (queue->dealloc_task) {
747 kthread_stop(queue->dealloc_task);
748 queue->dealloc_task = NULL;
749 }
750
751 if (queue->tx_irq) {
752 if (queue->tx_irq == queue->rx_irq)
753 unbind_from_irqhandler(queue->tx_irq, queue);
754 else {
755 unbind_from_irqhandler(queue->tx_irq, queue);
756 unbind_from_irqhandler(queue->rx_irq, queue);
757 }
758 queue->tx_irq = 0;
759 }
760
Paul Durrant4e15ee22016-05-13 09:37:26 +0100761 xenvif_unmap_frontend_data_rings(queue);
Wei Liue9ce7cb2014-06-04 10:30:42 +0100762 }
Paul Durrant210c34d2015-09-02 17:58:36 +0100763
764 xenvif_mcast_addr_list_free(vif);
Paul Durrant279f4382013-09-17 17:46:08 +0100765}
766
Paul Durrant4e15ee22016-05-13 09:37:26 +0100767void xenvif_disconnect_ctrl(struct xenvif *vif)
768{
Paul Durrant4e15ee22016-05-13 09:37:26 +0100769 if (vif->ctrl_irq) {
Paul Durrantc0fcded2016-05-18 15:55:42 +0100770 xenvif_deinit_hash(vif);
Paul Durrant4e15ee22016-05-13 09:37:26 +0100771 unbind_from_irqhandler(vif->ctrl_irq, vif);
772 vif->ctrl_irq = 0;
773 }
774
775 if (vif->ctrl.sring) {
776 xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif),
777 vif->ctrl.sring);
778 vif->ctrl.sring = NULL;
779 }
780}
781
Andrew J. Bennieston8d3d53b2014-06-04 10:30:43 +0100782/* Reverse the relevant parts of xenvif_init_queue().
783 * Used for queue teardown from xenvif_free(), and on the
784 * error handling paths in xenbus.c:connect().
785 */
786void xenvif_deinit_queue(struct xenvif_queue *queue)
787{
David Vrabelff4b1562015-01-08 18:06:01 +0000788 gnttab_free_pages(MAX_PENDING_REQS, queue->mmap_pages);
Andrew J. Bennieston8d3d53b2014-06-04 10:30:43 +0100789}
790
Paul Durrant279f4382013-09-17 17:46:08 +0100791void xenvif_free(struct xenvif *vif)
792{
David Vrabel9c6f3ff2016-01-15 14:55:36 +0000793 struct xenvif_queue *queues = vif->queues;
Wei Liuf7b50c42014-06-23 10:50:17 +0100794 unsigned int num_queues = vif->num_queues;
Wei Liue9ce7cb2014-06-04 10:30:42 +0100795 unsigned int queue_index;
Zoltan Kissf53c3fe2014-03-06 21:48:26 +0000796
Wei Liue9ce7cb2014-06-04 10:30:42 +0100797 unregister_netdev(vif->dev);
Ian Campbellf942dc22011-03-15 00:06:18 +0000798 free_netdev(vif->dev);
Wei Liub103f352013-05-16 23:26:11 +0000799
David Vrabel9c6f3ff2016-01-15 14:55:36 +0000800 for (queue_index = 0; queue_index < num_queues; ++queue_index)
801 xenvif_deinit_queue(&queues[queue_index]);
802 vfree(queues);
803
Paul Durrant279f4382013-09-17 17:46:08 +0100804 module_put(THIS_MODULE);
Ian Campbellf942dc22011-03-15 00:06:18 +0000805}