blob: 829b26cd4549a4e07ccdf30ea87d902424ce737b [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;
Wei Liuf7b50c42014-06-23 10:50:17 +0100168 unsigned int num_queues = vif->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
Wei Liue9ce7cb2014-06-04 10:30:42 +0100174 /* Drop the packet if queues are not set up */
175 if (num_queues < 1)
176 goto drop;
177
178 /* Obtain the queue to be used to transmit this packet */
179 index = skb_get_queue_mapping(skb);
180 if (index >= num_queues) {
181 pr_warn_ratelimited("Invalid queue %hu for packet on interface %s\n.",
182 index, vif->dev->name);
183 index %= num_queues;
184 }
185 queue = &vif->queues[index];
186
187 /* Drop the packet if queue is not ready */
188 if (queue->task == NULL ||
189 queue->dealloc_task == NULL ||
Zoltan Kissf53c3fe2014-03-06 21:48:26 +0000190 !xenvif_schedulable(vif))
Ian Campbellf942dc22011-03-15 00:06:18 +0000191 goto drop;
192
Paul Durrant210c34d2015-09-02 17:58:36 +0100193 if (vif->multicast_control && skb->pkt_type == PACKET_MULTICAST) {
194 struct ethhdr *eth = (struct ethhdr *)skb->data;
195
196 if (!xenvif_mcast_match(vif, eth->h_dest))
197 goto drop;
198 }
199
David Vrabelf48da8b2014-10-22 14:08:54 +0100200 cb = XENVIF_RX_CB(skb);
David Vrabel26c0e102014-12-18 11:13:06 +0000201 cb->expires = jiffies + vif->drain_timeout;
Ian Campbellf942dc22011-03-15 00:06:18 +0000202
Paul Durrant912e27e2016-10-07 09:32:31 +0100203 /* If there is no hash algorithm configured then make sure there
204 * is no hash information in the socket buffer otherwise it
205 * would be incorrectly forwarded to the frontend.
206 */
207 if (vif->hash.alg == XEN_NETIF_CTRL_HASH_ALGORITHM_NONE)
208 skb_clear_hash(skb);
209
David Vrabelf48da8b2014-10-22 14:08:54 +0100210 xenvif_rx_queue_tail(queue, skb);
Wei Liue9ce7cb2014-06-04 10:30:42 +0100211 xenvif_kick_thread(queue);
Ian Campbellf942dc22011-03-15 00:06:18 +0000212
213 return NETDEV_TX_OK;
214
215 drop:
216 vif->dev->stats.tx_dropped++;
217 dev_kfree_skb(skb);
218 return NETDEV_TX_OK;
219}
220
Ian Campbellf942dc22011-03-15 00:06:18 +0000221static struct net_device_stats *xenvif_get_stats(struct net_device *dev)
222{
223 struct xenvif *vif = netdev_priv(dev);
Wei Liue9ce7cb2014-06-04 10:30:42 +0100224 struct xenvif_queue *queue = NULL;
Mart van Santenebf692f2017-02-10 12:02:18 +0000225 u64 rx_bytes = 0;
226 u64 rx_packets = 0;
227 u64 tx_bytes = 0;
228 u64 tx_packets = 0;
Wei Liue9ce7cb2014-06-04 10:30:42 +0100229 unsigned int index;
230
Igor Druzhininf16f1df2017-01-17 20:49:38 +0000231 spin_lock(&vif->lock);
Wei Liue9ce7cb2014-06-04 10:30:42 +0100232 if (vif->queues == NULL)
233 goto out;
234
235 /* Aggregate tx and rx stats from each queue */
Igor Druzhininf16f1df2017-01-17 20:49:38 +0000236 for (index = 0; index < vif->num_queues; ++index) {
Wei Liue9ce7cb2014-06-04 10:30:42 +0100237 queue = &vif->queues[index];
238 rx_bytes += queue->stats.rx_bytes;
239 rx_packets += queue->stats.rx_packets;
240 tx_bytes += queue->stats.tx_bytes;
241 tx_packets += queue->stats.tx_packets;
242 }
243
244out:
Igor Druzhininf16f1df2017-01-17 20:49:38 +0000245 spin_unlock(&vif->lock);
246
Wei Liue9ce7cb2014-06-04 10:30:42 +0100247 vif->dev->stats.rx_bytes = rx_bytes;
248 vif->dev->stats.rx_packets = rx_packets;
249 vif->dev->stats.tx_bytes = tx_bytes;
250 vif->dev->stats.tx_packets = tx_packets;
251
Ian Campbellf942dc22011-03-15 00:06:18 +0000252 return &vif->dev->stats;
253}
254
255static void xenvif_up(struct xenvif *vif)
256{
Wei Liue9ce7cb2014-06-04 10:30:42 +0100257 struct xenvif_queue *queue = NULL;
Wei Liuf7b50c42014-06-23 10:50:17 +0100258 unsigned int num_queues = vif->num_queues;
Wei Liue9ce7cb2014-06-04 10:30:42 +0100259 unsigned int queue_index;
260
261 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
262 queue = &vif->queues[queue_index];
263 napi_enable(&queue->napi);
264 enable_irq(queue->tx_irq);
265 if (queue->tx_irq != queue->rx_irq)
266 enable_irq(queue->rx_irq);
267 xenvif_napi_schedule_or_enable_events(queue);
268 }
Ian Campbellf942dc22011-03-15 00:06:18 +0000269}
270
271static void xenvif_down(struct xenvif *vif)
272{
Wei Liue9ce7cb2014-06-04 10:30:42 +0100273 struct xenvif_queue *queue = NULL;
Wei Liuf7b50c42014-06-23 10:50:17 +0100274 unsigned int num_queues = vif->num_queues;
Wei Liue9ce7cb2014-06-04 10:30:42 +0100275 unsigned int queue_index;
276
277 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
278 queue = &vif->queues[queue_index];
Wei Liue9ce7cb2014-06-04 10:30:42 +0100279 disable_irq(queue->tx_irq);
280 if (queue->tx_irq != queue->rx_irq)
281 disable_irq(queue->rx_irq);
Zoltan Kiss8fe78982014-10-28 15:29:30 +0000282 napi_disable(&queue->napi);
Wei Liue9ce7cb2014-06-04 10:30:42 +0100283 del_timer_sync(&queue->credit_timeout);
284 }
Ian Campbellf942dc22011-03-15 00:06:18 +0000285}
286
287static int xenvif_open(struct net_device *dev)
288{
289 struct xenvif *vif = netdev_priv(dev);
Zoltan Kiss3d1af1d2014-08-04 16:20:57 +0100290 if (test_bit(VIF_STATUS_CONNECTED, &vif->status))
Ian Campbellf942dc22011-03-15 00:06:18 +0000291 xenvif_up(vif);
Wei Liue9ce7cb2014-06-04 10:30:42 +0100292 netif_tx_start_all_queues(dev);
Ian Campbellf942dc22011-03-15 00:06:18 +0000293 return 0;
294}
295
296static int xenvif_close(struct net_device *dev)
297{
298 struct xenvif *vif = netdev_priv(dev);
Zoltan Kiss3d1af1d2014-08-04 16:20:57 +0100299 if (test_bit(VIF_STATUS_CONNECTED, &vif->status))
Ian Campbellf942dc22011-03-15 00:06:18 +0000300 xenvif_down(vif);
Wei Liue9ce7cb2014-06-04 10:30:42 +0100301 netif_tx_stop_all_queues(dev);
Ian Campbellf942dc22011-03-15 00:06:18 +0000302 return 0;
303}
304
305static int xenvif_change_mtu(struct net_device *dev, int mtu)
306{
307 struct xenvif *vif = netdev_priv(dev);
Jarod Wilsond0c2c992016-10-20 13:55:21 -0400308 int max = vif->can_sg ? ETH_MAX_MTU - VLAN_ETH_HLEN : ETH_DATA_LEN;
Ian Campbellf942dc22011-03-15 00:06:18 +0000309
310 if (mtu > max)
311 return -EINVAL;
312 dev->mtu = mtu;
313 return 0;
314}
315
Michał Mirosławc8f44af2011-11-15 15:29:55 +0000316static netdev_features_t xenvif_fix_features(struct net_device *dev,
317 netdev_features_t features)
Ian Campbellf942dc22011-03-15 00:06:18 +0000318{
319 struct xenvif *vif = netdev_priv(dev);
Ian Campbellf942dc22011-03-15 00:06:18 +0000320
Michał Mirosław47103042011-04-19 03:35:06 +0000321 if (!vif->can_sg)
322 features &= ~NETIF_F_SG;
Paul Durrantfedbc8c2016-10-04 10:29:13 +0100323 if (~(vif->gso_mask) & GSO_BIT(TCPV4))
Michał Mirosław47103042011-04-19 03:35:06 +0000324 features &= ~NETIF_F_TSO;
Paul Durrantfedbc8c2016-10-04 10:29:13 +0100325 if (~(vif->gso_mask) & GSO_BIT(TCPV6))
Paul Durrant82cada22013-10-16 17:50:32 +0100326 features &= ~NETIF_F_TSO6;
Paul Durrant146c8a72013-10-16 17:50:28 +0100327 if (!vif->ip_csum)
Michał Mirosław47103042011-04-19 03:35:06 +0000328 features &= ~NETIF_F_IP_CSUM;
Paul Durrant146c8a72013-10-16 17:50:28 +0100329 if (!vif->ipv6_csum)
330 features &= ~NETIF_F_IPV6_CSUM;
Ian Campbellf942dc22011-03-15 00:06:18 +0000331
Michał Mirosław47103042011-04-19 03:35:06 +0000332 return features;
Ian Campbellf942dc22011-03-15 00:06:18 +0000333}
334
335static const struct xenvif_stat {
336 char name[ETH_GSTRING_LEN];
337 u16 offset;
338} xenvif_stats[] = {
339 {
340 "rx_gso_checksum_fixup",
Wei Liue9ce7cb2014-06-04 10:30:42 +0100341 offsetof(struct xenvif_stats, rx_gso_checksum_fixup)
Ian Campbellf942dc22011-03-15 00:06:18 +0000342 },
Zoltan Kiss1bb332a2014-03-06 21:48:28 +0000343 /* If (sent != success + fail), there are probably packets never
344 * freed up properly!
345 */
346 {
347 "tx_zerocopy_sent",
Wei Liue9ce7cb2014-06-04 10:30:42 +0100348 offsetof(struct xenvif_stats, tx_zerocopy_sent),
Zoltan Kiss1bb332a2014-03-06 21:48:28 +0000349 },
350 {
351 "tx_zerocopy_success",
Wei Liue9ce7cb2014-06-04 10:30:42 +0100352 offsetof(struct xenvif_stats, tx_zerocopy_success),
Zoltan Kiss1bb332a2014-03-06 21:48:28 +0000353 },
354 {
355 "tx_zerocopy_fail",
Wei Liue9ce7cb2014-06-04 10:30:42 +0100356 offsetof(struct xenvif_stats, tx_zerocopy_fail)
Zoltan Kiss1bb332a2014-03-06 21:48:28 +0000357 },
Zoltan Kisse3377f32014-03-06 21:48:29 +0000358 /* Number of packets exceeding MAX_SKB_FRAG slots. You should use
359 * a guest with the same MAX_SKB_FRAG
360 */
361 {
362 "tx_frag_overflow",
Wei Liue9ce7cb2014-06-04 10:30:42 +0100363 offsetof(struct xenvif_stats, tx_frag_overflow)
Zoltan Kisse3377f32014-03-06 21:48:29 +0000364 },
Ian Campbellf942dc22011-03-15 00:06:18 +0000365};
366
367static int xenvif_get_sset_count(struct net_device *dev, int string_set)
368{
369 switch (string_set) {
370 case ETH_SS_STATS:
371 return ARRAY_SIZE(xenvif_stats);
372 default:
373 return -EINVAL;
374 }
375}
376
377static void xenvif_get_ethtool_stats(struct net_device *dev,
378 struct ethtool_stats *stats, u64 * data)
379{
Wei Liue9ce7cb2014-06-04 10:30:42 +0100380 struct xenvif *vif = netdev_priv(dev);
Wei Liuf7b50c42014-06-23 10:50:17 +0100381 unsigned int num_queues = vif->num_queues;
Ian Campbellf942dc22011-03-15 00:06:18 +0000382 int i;
Wei Liue9ce7cb2014-06-04 10:30:42 +0100383 unsigned int queue_index;
Ian Campbellf942dc22011-03-15 00:06:18 +0000384
Wei Liue9ce7cb2014-06-04 10:30:42 +0100385 for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++) {
386 unsigned long accum = 0;
387 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
David Vrabeld63951d2015-03-04 11:14:46 +0000388 void *vif_stats = &vif->queues[queue_index].stats;
Wei Liue9ce7cb2014-06-04 10:30:42 +0100389 accum += *(unsigned long *)(vif_stats + xenvif_stats[i].offset);
390 }
391 data[i] = accum;
392 }
Ian Campbellf942dc22011-03-15 00:06:18 +0000393}
394
395static void xenvif_get_strings(struct net_device *dev, u32 stringset, u8 * data)
396{
397 int i;
398
399 switch (stringset) {
400 case ETH_SS_STATS:
401 for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++)
402 memcpy(data + i * ETH_GSTRING_LEN,
403 xenvif_stats[i].name, ETH_GSTRING_LEN);
404 break;
405 }
406}
407
stephen hemminger813abbb2012-01-04 11:56:58 +0000408static const struct ethtool_ops xenvif_ethtool_ops = {
Ian Campbellf942dc22011-03-15 00:06:18 +0000409 .get_link = ethtool_op_get_link,
410
411 .get_sset_count = xenvif_get_sset_count,
412 .get_ethtool_stats = xenvif_get_ethtool_stats,
413 .get_strings = xenvif_get_strings,
414};
415
stephen hemminger813abbb2012-01-04 11:56:58 +0000416static const struct net_device_ops xenvif_netdev_ops = {
Paul Durrant40d8abd2016-05-13 09:37:27 +0100417 .ndo_select_queue = xenvif_select_queue,
Ian Campbellf942dc22011-03-15 00:06:18 +0000418 .ndo_start_xmit = xenvif_start_xmit,
419 .ndo_get_stats = xenvif_get_stats,
420 .ndo_open = xenvif_open,
421 .ndo_stop = xenvif_close,
422 .ndo_change_mtu = xenvif_change_mtu,
Michał Mirosław47103042011-04-19 03:35:06 +0000423 .ndo_fix_features = xenvif_fix_features,
Matt Wilson4a633a62013-01-22 08:08:25 +0000424 .ndo_set_mac_address = eth_mac_addr,
425 .ndo_validate_addr = eth_validate_addr,
Ian Campbellf942dc22011-03-15 00:06:18 +0000426};
427
428struct xenvif *xenvif_alloc(struct device *parent, domid_t domid,
429 unsigned int handle)
430{
431 int err;
432 struct net_device *dev;
433 struct xenvif *vif;
434 char name[IFNAMSIZ] = {};
435
436 snprintf(name, IFNAMSIZ - 1, "vif%u.%u", domid, handle);
Andrew J. Bennieston8d3d53b2014-06-04 10:30:43 +0100437 /* Allocate a netdev with the max. supported number of queues.
438 * When the guest selects the desired number, it will be updated
Wei Liuf7b50c42014-06-23 10:50:17 +0100439 * via netif_set_real_num_*_queues().
Andrew J. Bennieston8d3d53b2014-06-04 10:30:43 +0100440 */
Tom Gundersenc835a672014-07-14 16:37:24 +0200441 dev = alloc_netdev_mq(sizeof(struct xenvif), name, NET_NAME_UNKNOWN,
442 ether_setup, xenvif_max_queues);
Ian Campbellf942dc22011-03-15 00:06:18 +0000443 if (dev == NULL) {
Wei Liub3f980b2013-08-26 12:59:38 +0100444 pr_warn("Could not allocate netdev for %s\n", name);
Ian Campbellf942dc22011-03-15 00:06:18 +0000445 return ERR_PTR(-ENOMEM);
446 }
447
448 SET_NETDEV_DEV(dev, parent);
449
450 vif = netdev_priv(dev);
Paul Durrantac3d5ac2013-12-23 09:27:17 +0000451
Ian Campbellf942dc22011-03-15 00:06:18 +0000452 vif->domid = domid;
453 vif->handle = handle;
Ian Campbellf942dc22011-03-15 00:06:18 +0000454 vif->can_sg = 1;
Paul Durrant146c8a72013-10-16 17:50:28 +0100455 vif->ip_csum = 1;
Ian Campbellf942dc22011-03-15 00:06:18 +0000456 vif->dev = dev;
Wei Liue9d8b2c2014-04-01 12:46:12 +0100457 vif->disabled = false;
David Vrabel26c0e102014-12-18 11:13:06 +0000458 vif->drain_timeout = msecs_to_jiffies(rx_drain_timeout_msecs);
459 vif->stall_timeout = msecs_to_jiffies(rx_stall_timeout_msecs);
Wei Liue9d8b2c2014-04-01 12:46:12 +0100460
Wei Liuf7b50c42014-06-23 10:50:17 +0100461 /* Start out with no queues. */
Wei Liue9ce7cb2014-06-04 10:30:42 +0100462 vif->queues = NULL;
Wei Liuf7b50c42014-06-23 10:50:17 +0100463 vif->num_queues = 0;
Zoltan Kiss09350782014-03-06 21:48:30 +0000464
David Vrabelecf08d22014-10-22 14:08:55 +0100465 spin_lock_init(&vif->lock);
Paul Durrant210c34d2015-09-02 17:58:36 +0100466 INIT_LIST_HEAD(&vif->fe_mcast_addr);
David Vrabelecf08d22014-10-22 14:08:55 +0100467
Ian Campbellf942dc22011-03-15 00:06:18 +0000468 dev->netdev_ops = &xenvif_netdev_ops;
Paul Durrant146c8a72013-10-16 17:50:28 +0100469 dev->hw_features = NETIF_F_SG |
470 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
Ross Lagerwall2167ca02016-10-04 10:29:18 +0100471 NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_FRAGLIST;
Paul Durrant7365bcf2013-10-16 17:50:30 +0100472 dev->features = dev->hw_features | NETIF_F_RXCSUM;
Wilfried Klaebe7ad24ea2014-05-11 00:12:32 +0000473 dev->ethtool_ops = &xenvif_ethtool_ops;
Ian Campbellf942dc22011-03-15 00:06:18 +0000474
475 dev->tx_queue_len = XENVIF_QUEUE_LENGTH;
476
Jarod Wilsond0c2c992016-10-20 13:55:21 -0400477 dev->min_mtu = 0;
478 dev->max_mtu = ETH_MAX_MTU - VLAN_ETH_HLEN;
479
Ian Campbellf942dc22011-03-15 00:06:18 +0000480 /*
481 * Initialise a dummy MAC address. We choose the numerically
482 * largest non-broadcast address to prevent the address getting
483 * stolen by an Ethernet bridge for STP purposes.
484 * (FE:FF:FF:FF:FF:FF)
485 */
Joe Perches3b6ed262015-03-02 19:54:51 -0800486 eth_broadcast_addr(dev->dev_addr);
Ian Campbellf942dc22011-03-15 00:06:18 +0000487 dev->dev_addr[0] &= ~0x01;
488
489 netif_carrier_off(dev);
490
491 err = register_netdev(dev);
492 if (err) {
493 netdev_warn(dev, "Could not register device: err=%d\n", err);
494 free_netdev(dev);
495 return ERR_PTR(err);
496 }
497
498 netdev_dbg(dev, "Successfully created xenvif\n");
Paul Durrant279f4382013-09-17 17:46:08 +0100499
500 __module_get(THIS_MODULE);
501
Ian Campbellf942dc22011-03-15 00:06:18 +0000502 return vif;
503}
504
Wei Liue9ce7cb2014-06-04 10:30:42 +0100505int xenvif_init_queue(struct xenvif_queue *queue)
Ian Campbellf942dc22011-03-15 00:06:18 +0000506{
Wei Liue9ce7cb2014-06-04 10:30:42 +0100507 int err, i;
Ian Campbellf942dc22011-03-15 00:06:18 +0000508
Wei Liue9ce7cb2014-06-04 10:30:42 +0100509 queue->credit_bytes = queue->remaining_credit = ~0UL;
510 queue->credit_usec = 0UL;
511 init_timer(&queue->credit_timeout);
Palik, Imreedafc132015-03-19 11:05:42 +0100512 queue->credit_timeout.function = xenvif_tx_credit_callback;
Wei Liue9ce7cb2014-06-04 10:30:42 +0100513 queue->credit_window_start = get_jiffies_64();
Ian Campbellf942dc22011-03-15 00:06:18 +0000514
David Vrabelf48da8b2014-10-22 14:08:54 +0100515 queue->rx_queue_max = XENVIF_RX_QUEUE_BYTES;
516
Wei Liue9ce7cb2014-06-04 10:30:42 +0100517 skb_queue_head_init(&queue->rx_queue);
518 skb_queue_head_init(&queue->tx_queue);
Ian Campbellf942dc22011-03-15 00:06:18 +0000519
Wei Liue9ce7cb2014-06-04 10:30:42 +0100520 queue->pending_cons = 0;
521 queue->pending_prod = MAX_PENDING_REQS;
522 for (i = 0; i < MAX_PENDING_REQS; ++i)
523 queue->pending_ring[i] = i;
Paul Durrantca2f09f2013-12-06 16:36:07 +0000524
Wei Liue9ce7cb2014-06-04 10:30:42 +0100525 spin_lock_init(&queue->callback_lock);
526 spin_lock_init(&queue->response_lock);
Wei Liue1f00a692013-05-22 06:34:45 +0000527
Wei Liue9ce7cb2014-06-04 10:30:42 +0100528 /* If ballooning is disabled, this will consume real memory, so you
529 * better enable it. The long term solution would be to use just a
530 * bunch of valid page descriptors, without dependency on ballooning
531 */
David Vrabelff4b1562015-01-08 18:06:01 +0000532 err = gnttab_alloc_pages(MAX_PENDING_REQS,
533 queue->mmap_pages);
Wei Liue9ce7cb2014-06-04 10:30:42 +0100534 if (err) {
535 netdev_err(queue->vif->dev, "Could not reserve mmap_pages\n");
536 return -ENOMEM;
Wei Liue1f00a692013-05-22 06:34:45 +0000537 }
Ian Campbellf942dc22011-03-15 00:06:18 +0000538
Wei Liue9ce7cb2014-06-04 10:30:42 +0100539 for (i = 0; i < MAX_PENDING_REQS; i++) {
540 queue->pending_tx_info[i].callback_struct = (struct ubuf_info)
541 { .callback = xenvif_zerocopy_callback,
542 .ctx = NULL,
543 .desc = i };
544 queue->grant_tx_handle[i] = NETBACK_INVALID_HANDLE;
Wei Liub3f980b2013-08-26 12:59:38 +0100545 }
Ian Campbellf942dc22011-03-15 00:06:18 +0000546
Wei Liue9ce7cb2014-06-04 10:30:42 +0100547 return 0;
548}
Zoltan Kissf53c3fe2014-03-06 21:48:26 +0000549
Wei Liue9ce7cb2014-06-04 10:30:42 +0100550void xenvif_carrier_on(struct xenvif *vif)
551{
Ian Campbellf942dc22011-03-15 00:06:18 +0000552 rtnl_lock();
Michał Mirosław47103042011-04-19 03:35:06 +0000553 if (!vif->can_sg && vif->dev->mtu > ETH_DATA_LEN)
554 dev_set_mtu(vif->dev, ETH_DATA_LEN);
555 netdev_update_features(vif->dev);
Zoltan Kiss3d1af1d2014-08-04 16:20:57 +0100556 set_bit(VIF_STATUS_CONNECTED, &vif->status);
David Vrabeld0e5d832011-09-30 06:37:51 +0000557 if (netif_running(vif->dev))
558 xenvif_up(vif);
Ian Campbellf942dc22011-03-15 00:06:18 +0000559 rtnl_unlock();
Wei Liue9ce7cb2014-06-04 10:30:42 +0100560}
Ian Campbellf942dc22011-03-15 00:06:18 +0000561
Paul Durrant4e15ee22016-05-13 09:37:26 +0100562int xenvif_connect_ctrl(struct xenvif *vif, grant_ref_t ring_ref,
563 unsigned int evtchn)
564{
565 struct net_device *dev = vif->dev;
566 void *addr;
567 struct xen_netif_ctrl_sring *shared;
Juergen Gross0364a882016-09-22 11:06:25 +0200568 int err;
Paul Durrant4e15ee22016-05-13 09:37:26 +0100569
570 err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(vif),
571 &ring_ref, 1, &addr);
572 if (err)
573 goto err;
574
575 shared = (struct xen_netif_ctrl_sring *)addr;
576 BACK_RING_INIT(&vif->ctrl, shared, XEN_PAGE_SIZE);
577
Juergen Gross0364a882016-09-22 11:06:25 +0200578 err = bind_interdomain_evtchn_to_irq(vif->domid, evtchn);
Paul Durrant4e15ee22016-05-13 09:37:26 +0100579 if (err < 0)
580 goto err_unmap;
581
582 vif->ctrl_irq = err;
583
Paul Durrant40d8abd2016-05-13 09:37:27 +0100584 xenvif_init_hash(vif);
585
Juergen Gross0364a882016-09-22 11:06:25 +0200586 err = request_threaded_irq(vif->ctrl_irq, NULL, xenvif_ctrl_irq_fn,
587 IRQF_ONESHOT, "xen-netback-ctrl", vif);
588 if (err) {
589 pr_warn("Could not setup irq handler for %s\n", dev->name);
Paul Durrant4e15ee22016-05-13 09:37:26 +0100590 goto err_deinit;
591 }
592
Paul Durrant4e15ee22016-05-13 09:37:26 +0100593 return 0;
594
595err_deinit:
Paul Durrant40d8abd2016-05-13 09:37:27 +0100596 xenvif_deinit_hash(vif);
Paul Durrant4e15ee22016-05-13 09:37:26 +0100597 unbind_from_irqhandler(vif->ctrl_irq, vif);
598 vif->ctrl_irq = 0;
599
600err_unmap:
601 xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif),
602 vif->ctrl.sring);
603 vif->ctrl.sring = NULL;
604
605err:
606 return err;
607}
608
609int xenvif_connect_data(struct xenvif_queue *queue,
610 unsigned long tx_ring_ref,
611 unsigned long rx_ring_ref,
612 unsigned int tx_evtchn,
613 unsigned int rx_evtchn)
Wei Liue9ce7cb2014-06-04 10:30:42 +0100614{
615 struct task_struct *task;
616 int err = -ENOMEM;
617
618 BUG_ON(queue->tx_irq);
619 BUG_ON(queue->task);
620 BUG_ON(queue->dealloc_task);
621
Paul Durrant4e15ee22016-05-13 09:37:26 +0100622 err = xenvif_map_frontend_data_rings(queue, tx_ring_ref,
623 rx_ring_ref);
Wei Liue9ce7cb2014-06-04 10:30:42 +0100624 if (err < 0)
625 goto err;
626
627 init_waitqueue_head(&queue->wq);
628 init_waitqueue_head(&queue->dealloc_wq);
Wei Liua64bd932014-08-12 11:48:07 +0100629 atomic_set(&queue->inflight_packets, 0);
Wei Liue9ce7cb2014-06-04 10:30:42 +0100630
Wei Liue24f8192014-08-25 16:44:00 +0100631 netif_napi_add(queue->vif->dev, &queue->napi, xenvif_poll,
632 XENVIF_NAPI_WEIGHT);
633
Wei Liue9ce7cb2014-06-04 10:30:42 +0100634 if (tx_evtchn == rx_evtchn) {
635 /* feature-split-event-channels == 0 */
636 err = bind_interdomain_evtchn_to_irqhandler(
637 queue->vif->domid, tx_evtchn, xenvif_interrupt, 0,
638 queue->name, queue);
639 if (err < 0)
640 goto err_unmap;
641 queue->tx_irq = queue->rx_irq = err;
642 disable_irq(queue->tx_irq);
643 } else {
644 /* feature-split-event-channels == 1 */
645 snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
646 "%s-tx", queue->name);
647 err = bind_interdomain_evtchn_to_irqhandler(
648 queue->vif->domid, tx_evtchn, xenvif_tx_interrupt, 0,
649 queue->tx_irq_name, queue);
650 if (err < 0)
651 goto err_unmap;
652 queue->tx_irq = err;
653 disable_irq(queue->tx_irq);
654
655 snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
656 "%s-rx", queue->name);
657 err = bind_interdomain_evtchn_to_irqhandler(
658 queue->vif->domid, rx_evtchn, xenvif_rx_interrupt, 0,
659 queue->rx_irq_name, queue);
660 if (err < 0)
661 goto err_tx_unbind;
662 queue->rx_irq = err;
663 disable_irq(queue->rx_irq);
664 }
665
David Vrabelecf08d22014-10-22 14:08:55 +0100666 queue->stalled = true;
667
Wei Liue9ce7cb2014-06-04 10:30:42 +0100668 task = kthread_create(xenvif_kthread_guest_rx,
669 (void *)queue, "%s-guest-rx", queue->name);
670 if (IS_ERR(task)) {
671 pr_warn("Could not allocate kthread for %s\n", queue->name);
672 err = PTR_ERR(task);
673 goto err_rx_unbind;
674 }
675 queue->task = task;
David Vrabel42b52122015-02-02 16:57:51 +0000676 get_task_struct(task);
Wei Liue9ce7cb2014-06-04 10:30:42 +0100677
678 task = kthread_create(xenvif_dealloc_kthread,
679 (void *)queue, "%s-dealloc", queue->name);
680 if (IS_ERR(task)) {
681 pr_warn("Could not allocate kthread for %s\n", queue->name);
682 err = PTR_ERR(task);
683 goto err_rx_unbind;
684 }
685 queue->dealloc_task = task;
686
687 wake_up_process(queue->task);
688 wake_up_process(queue->dealloc_task);
Wei Liub3f980b2013-08-26 12:59:38 +0100689
Ian Campbellf942dc22011-03-15 00:06:18 +0000690 return 0;
Wei Liub3f980b2013-08-26 12:59:38 +0100691
692err_rx_unbind:
Wei Liue9ce7cb2014-06-04 10:30:42 +0100693 unbind_from_irqhandler(queue->rx_irq, queue);
694 queue->rx_irq = 0;
Wei Liue1f00a692013-05-22 06:34:45 +0000695err_tx_unbind:
Wei Liue9ce7cb2014-06-04 10:30:42 +0100696 unbind_from_irqhandler(queue->tx_irq, queue);
697 queue->tx_irq = 0;
Ian Campbellf942dc22011-03-15 00:06:18 +0000698err_unmap:
Paul Durrant4e15ee22016-05-13 09:37:26 +0100699 xenvif_unmap_frontend_data_rings(queue);
David Vrabel4a658522016-01-15 14:55:35 +0000700 netif_napi_del(&queue->napi);
Ian Campbellf942dc22011-03-15 00:06:18 +0000701err:
Wei Liub103f352013-05-16 23:26:11 +0000702 module_put(THIS_MODULE);
Ian Campbellf942dc22011-03-15 00:06:18 +0000703 return err;
704}
705
Ian Campbell488562862013-02-06 23:41:35 +0000706void xenvif_carrier_off(struct xenvif *vif)
Ian Campbellf942dc22011-03-15 00:06:18 +0000707{
708 struct net_device *dev = vif->dev;
Ian Campbell488562862013-02-06 23:41:35 +0000709
710 rtnl_lock();
Zoltan Kiss3d1af1d2014-08-04 16:20:57 +0100711 if (test_and_clear_bit(VIF_STATUS_CONNECTED, &vif->status)) {
712 netif_carrier_off(dev); /* discard queued packets */
713 if (netif_running(dev))
714 xenvif_down(vif);
715 }
Ian Campbell488562862013-02-06 23:41:35 +0000716 rtnl_unlock();
Ian Campbell488562862013-02-06 23:41:35 +0000717}
718
Paul Durrant4e15ee22016-05-13 09:37:26 +0100719void xenvif_disconnect_data(struct xenvif *vif)
Ian Campbell488562862013-02-06 23:41:35 +0000720{
Wei Liue9ce7cb2014-06-04 10:30:42 +0100721 struct xenvif_queue *queue = NULL;
Wei Liuf7b50c42014-06-23 10:50:17 +0100722 unsigned int num_queues = vif->num_queues;
Wei Liue9ce7cb2014-06-04 10:30:42 +0100723 unsigned int queue_index;
724
Zoltan Kiss3d1af1d2014-08-04 16:20:57 +0100725 xenvif_carrier_off(vif);
Ian Campbellf942dc22011-03-15 00:06:18 +0000726
Wei Liue9ce7cb2014-06-04 10:30:42 +0100727 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
728 queue = &vif->queues[queue_index];
David Vrabeldb739ef2013-11-21 15:26:09 +0000729
Wei Liuea2c5e12014-08-12 11:48:06 +0100730 netif_napi_del(&queue->napi);
731
Wei Liue9ce7cb2014-06-04 10:30:42 +0100732 if (queue->task) {
Wei Liue9ce7cb2014-06-04 10:30:42 +0100733 kthread_stop(queue->task);
David Vrabel42b52122015-02-02 16:57:51 +0000734 put_task_struct(queue->task);
Wei Liue9ce7cb2014-06-04 10:30:42 +0100735 queue->task = NULL;
Wei Liue1f00a692013-05-22 06:34:45 +0000736 }
Ian Campbellf942dc22011-03-15 00:06:18 +0000737
Wei Liue9ce7cb2014-06-04 10:30:42 +0100738 if (queue->dealloc_task) {
739 kthread_stop(queue->dealloc_task);
740 queue->dealloc_task = NULL;
741 }
742
743 if (queue->tx_irq) {
744 if (queue->tx_irq == queue->rx_irq)
745 unbind_from_irqhandler(queue->tx_irq, queue);
746 else {
747 unbind_from_irqhandler(queue->tx_irq, queue);
748 unbind_from_irqhandler(queue->rx_irq, queue);
749 }
750 queue->tx_irq = 0;
751 }
752
Paul Durrant4e15ee22016-05-13 09:37:26 +0100753 xenvif_unmap_frontend_data_rings(queue);
Wei Liue9ce7cb2014-06-04 10:30:42 +0100754 }
Paul Durrant210c34d2015-09-02 17:58:36 +0100755
756 xenvif_mcast_addr_list_free(vif);
Paul Durrant279f4382013-09-17 17:46:08 +0100757}
758
Paul Durrant4e15ee22016-05-13 09:37:26 +0100759void xenvif_disconnect_ctrl(struct xenvif *vif)
760{
Paul Durrant4e15ee22016-05-13 09:37:26 +0100761 if (vif->ctrl_irq) {
Paul Durrantc0fcded2016-05-18 15:55:42 +0100762 xenvif_deinit_hash(vif);
Paul Durrant4e15ee22016-05-13 09:37:26 +0100763 unbind_from_irqhandler(vif->ctrl_irq, vif);
764 vif->ctrl_irq = 0;
765 }
766
767 if (vif->ctrl.sring) {
768 xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif),
769 vif->ctrl.sring);
770 vif->ctrl.sring = NULL;
771 }
772}
773
Andrew J. Bennieston8d3d53b2014-06-04 10:30:43 +0100774/* Reverse the relevant parts of xenvif_init_queue().
775 * Used for queue teardown from xenvif_free(), and on the
776 * error handling paths in xenbus.c:connect().
777 */
778void xenvif_deinit_queue(struct xenvif_queue *queue)
779{
David Vrabelff4b1562015-01-08 18:06:01 +0000780 gnttab_free_pages(MAX_PENDING_REQS, queue->mmap_pages);
Andrew J. Bennieston8d3d53b2014-06-04 10:30:43 +0100781}
782
Paul Durrant279f4382013-09-17 17:46:08 +0100783void xenvif_free(struct xenvif *vif)
784{
David Vrabel9c6f3ff2016-01-15 14:55:36 +0000785 struct xenvif_queue *queues = vif->queues;
Wei Liuf7b50c42014-06-23 10:50:17 +0100786 unsigned int num_queues = vif->num_queues;
Wei Liue9ce7cb2014-06-04 10:30:42 +0100787 unsigned int queue_index;
Zoltan Kissf53c3fe2014-03-06 21:48:26 +0000788
Wei Liue9ce7cb2014-06-04 10:30:42 +0100789 unregister_netdev(vif->dev);
Ian Campbellf942dc22011-03-15 00:06:18 +0000790 free_netdev(vif->dev);
Wei Liub103f352013-05-16 23:26:11 +0000791
David Vrabel9c6f3ff2016-01-15 14:55:36 +0000792 for (queue_index = 0; queue_index < num_queues; ++queue_index)
793 xenvif_deinit_queue(&queues[queue_index]);
794 vfree(queues);
795
Paul Durrant279f4382013-09-17 17:46:08 +0100796 module_put(THIS_MODULE);
Ian Campbellf942dc22011-03-15 00:06:18 +0000797}