blob: 6929bcb61cf00cec88c5cad2ecb45878cc06475d [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>
Ian Campbellf942dc22011-03-15 00:06:18 +000034#include <linux/ethtool.h>
35#include <linux/rtnetlink.h>
36#include <linux/if_vlan.h>
37
38#include <xen/events.h>
39#include <asm/xen/hypercall.h>
Zoltan Kissf53c3fe2014-03-06 21:48:26 +000040#include <xen/balloon.h>
Ian Campbellf942dc22011-03-15 00:06:18 +000041
42#define XENVIF_QUEUE_LENGTH 32
Wei Liub3f980b2013-08-26 12:59:38 +010043#define XENVIF_NAPI_WEIGHT 64
Ian Campbellf942dc22011-03-15 00:06:18 +000044
Wei Liue9ce7cb2014-06-04 10:30:42 +010045static inline void xenvif_stop_queue(struct xenvif_queue *queue)
46{
47 struct net_device *dev = queue->vif->dev;
48
49 if (!queue->vif->can_queue)
50 return;
51
52 netif_tx_stop_queue(netdev_get_tx_queue(dev, queue->id));
53}
54
Ian Campbellf942dc22011-03-15 00:06:18 +000055int xenvif_schedulable(struct xenvif *vif)
56{
57 return netif_running(vif->dev) && netif_carrier_ok(vif->dev);
58}
59
Wei Liue1f00a692013-05-22 06:34:45 +000060static irqreturn_t xenvif_tx_interrupt(int irq, void *dev_id)
Ian Campbellf942dc22011-03-15 00:06:18 +000061{
Wei Liue9ce7cb2014-06-04 10:30:42 +010062 struct xenvif_queue *queue = dev_id;
Ian Campbellf942dc22011-03-15 00:06:18 +000063
Wei Liue9ce7cb2014-06-04 10:30:42 +010064 if (RING_HAS_UNCONSUMED_REQUESTS(&queue->tx))
65 napi_schedule(&queue->napi);
Ian Campbellf942dc22011-03-15 00:06:18 +000066
Wei Liue1f00a692013-05-22 06:34:45 +000067 return IRQ_HANDLED;
68}
69
Wei Liue9ce7cb2014-06-04 10:30:42 +010070int xenvif_poll(struct napi_struct *napi, int budget)
Wei Liub3f980b2013-08-26 12:59:38 +010071{
Wei Liue9ce7cb2014-06-04 10:30:42 +010072 struct xenvif_queue *queue =
73 container_of(napi, struct xenvif_queue, napi);
Wei Liub3f980b2013-08-26 12:59:38 +010074 int work_done;
75
Wei Liue9d8b2c2014-04-01 12:46:12 +010076 /* This vif is rogue, we pretend we've there is nothing to do
77 * for this vif to deschedule it from NAPI. But this interface
78 * will be turned off in thread context later.
79 */
Wei Liue9ce7cb2014-06-04 10:30:42 +010080 if (unlikely(queue->vif->disabled)) {
Wei Liue9d8b2c2014-04-01 12:46:12 +010081 napi_complete(napi);
82 return 0;
83 }
84
Wei Liue9ce7cb2014-06-04 10:30:42 +010085 work_done = xenvif_tx_action(queue, budget);
Wei Liub3f980b2013-08-26 12:59:38 +010086
87 if (work_done < budget) {
David Vrabel0d08fce2014-05-16 12:26:04 +010088 napi_complete(napi);
Wei Liue9ce7cb2014-06-04 10:30:42 +010089 xenvif_napi_schedule_or_enable_events(queue);
Wei Liub3f980b2013-08-26 12:59:38 +010090 }
91
92 return work_done;
93}
94
Wei Liue1f00a692013-05-22 06:34:45 +000095static irqreturn_t xenvif_rx_interrupt(int irq, void *dev_id)
96{
Wei Liue9ce7cb2014-06-04 10:30:42 +010097 struct xenvif_queue *queue = dev_id;
Wei Liue1f00a692013-05-22 06:34:45 +000098
Wei Liue9ce7cb2014-06-04 10:30:42 +010099 xenvif_kick_thread(queue);
Ian Campbellf942dc22011-03-15 00:06:18 +0000100
101 return IRQ_HANDLED;
102}
103
Wei Liue1f00a692013-05-22 06:34:45 +0000104static irqreturn_t xenvif_interrupt(int irq, void *dev_id)
105{
106 xenvif_tx_interrupt(irq, dev_id);
107 xenvif_rx_interrupt(irq, dev_id);
108
109 return IRQ_HANDLED;
110}
111
Wei Liue9ce7cb2014-06-04 10:30:42 +0100112int xenvif_queue_stopped(struct xenvif_queue *queue)
Zoltan Kiss09350782014-03-06 21:48:30 +0000113{
Wei Liue9ce7cb2014-06-04 10:30:42 +0100114 struct net_device *dev = queue->vif->dev;
115 unsigned int id = queue->id;
116 return netif_tx_queue_stopped(netdev_get_tx_queue(dev, id));
117}
Zoltan Kiss09350782014-03-06 21:48:30 +0000118
Wei Liue9ce7cb2014-06-04 10:30:42 +0100119void xenvif_wake_queue(struct xenvif_queue *queue)
120{
121 struct net_device *dev = queue->vif->dev;
122 unsigned int id = queue->id;
123 netif_tx_wake_queue(netdev_get_tx_queue(dev, id));
124}
125
126/* Callback to wake the queue and drain it on timeout */
127static void xenvif_wake_queue_callback(unsigned long data)
128{
129 struct xenvif_queue *queue = (struct xenvif_queue *)data;
130
131 if (xenvif_queue_stopped(queue)) {
132 netdev_err(queue->vif->dev, "draining TX queue\n");
133 queue->rx_queue_purge = true;
134 xenvif_kick_thread(queue);
135 xenvif_wake_queue(queue);
Zoltan Kiss09350782014-03-06 21:48:30 +0000136 }
137}
138
Wei Liue9ce7cb2014-06-04 10:30:42 +0100139static u16 xenvif_select_queue(struct net_device *dev, struct sk_buff *skb,
140 void *accel_priv, select_queue_fallback_t fallback)
141{
Wei Liue9ce7cb2014-06-04 10:30:42 +0100142 unsigned int num_queues = dev->real_num_tx_queues;
143 u32 hash;
144 u16 queue_index;
145
146 /* First, check if there is only one queue to optimise the
147 * single-queue or old frontend scenario.
148 */
149 if (num_queues == 1) {
150 queue_index = 0;
151 } else {
152 /* Use skb_get_hash to obtain an L4 hash if available */
153 hash = skb_get_hash(skb);
154 queue_index = hash % num_queues;
155 }
156
157 return queue_index;
158}
159
Ian Campbellf942dc22011-03-15 00:06:18 +0000160static int xenvif_start_xmit(struct sk_buff *skb, struct net_device *dev)
161{
162 struct xenvif *vif = netdev_priv(dev);
Wei Liue9ce7cb2014-06-04 10:30:42 +0100163 struct xenvif_queue *queue = NULL;
164 unsigned int num_queues = dev->real_num_tx_queues;
165 u16 index;
Paul Durrantca2f09f2013-12-06 16:36:07 +0000166 int min_slots_needed;
Ian Campbellf942dc22011-03-15 00:06:18 +0000167
168 BUG_ON(skb->dev != dev);
169
Wei Liue9ce7cb2014-06-04 10:30:42 +0100170 /* Drop the packet if queues are not set up */
171 if (num_queues < 1)
172 goto drop;
173
174 /* Obtain the queue to be used to transmit this packet */
175 index = skb_get_queue_mapping(skb);
176 if (index >= num_queues) {
177 pr_warn_ratelimited("Invalid queue %hu for packet on interface %s\n.",
178 index, vif->dev->name);
179 index %= num_queues;
180 }
181 queue = &vif->queues[index];
182
183 /* Drop the packet if queue is not ready */
184 if (queue->task == NULL ||
185 queue->dealloc_task == NULL ||
Zoltan Kissf53c3fe2014-03-06 21:48:26 +0000186 !xenvif_schedulable(vif))
Ian Campbellf942dc22011-03-15 00:06:18 +0000187 goto drop;
188
Paul Durrantca2f09f2013-12-06 16:36:07 +0000189 /* At best we'll need one slot for the header and one for each
190 * frag.
191 */
192 min_slots_needed = 1 + skb_shinfo(skb)->nr_frags;
Ian Campbellf942dc22011-03-15 00:06:18 +0000193
Paul Durrantca2f09f2013-12-06 16:36:07 +0000194 /* If the skb is GSO then we'll also need an extra slot for the
195 * metadata.
196 */
Wei Liu836fbaf2014-03-11 12:45:32 +0000197 if (skb_is_gso(skb))
Paul Durrantca2f09f2013-12-06 16:36:07 +0000198 min_slots_needed++;
Ian Campbellf942dc22011-03-15 00:06:18 +0000199
Paul Durrantca2f09f2013-12-06 16:36:07 +0000200 /* If the skb can't possibly fit in the remaining slots
201 * then turn off the queue to give the ring a chance to
202 * drain.
203 */
Wei Liue9ce7cb2014-06-04 10:30:42 +0100204 if (!xenvif_rx_ring_slots_available(queue, min_slots_needed)) {
205 queue->wake_queue.function = xenvif_wake_queue_callback;
206 queue->wake_queue.data = (unsigned long)queue;
207 xenvif_stop_queue(queue);
208 mod_timer(&queue->wake_queue,
Zoltan Kiss09350782014-03-06 21:48:30 +0000209 jiffies + rx_drain_timeout_jiffies);
210 }
Ian Campbellf942dc22011-03-15 00:06:18 +0000211
Wei Liue9ce7cb2014-06-04 10:30:42 +0100212 skb_queue_tail(&queue->rx_queue, skb);
213 xenvif_kick_thread(queue);
Ian Campbellf942dc22011-03-15 00:06:18 +0000214
215 return NETDEV_TX_OK;
216
217 drop:
218 vif->dev->stats.tx_dropped++;
219 dev_kfree_skb(skb);
220 return NETDEV_TX_OK;
221}
222
Ian Campbellf942dc22011-03-15 00:06:18 +0000223static struct net_device_stats *xenvif_get_stats(struct net_device *dev)
224{
225 struct xenvif *vif = netdev_priv(dev);
Wei Liue9ce7cb2014-06-04 10:30:42 +0100226 struct xenvif_queue *queue = NULL;
227 unsigned int num_queues = dev->real_num_tx_queues;
228 unsigned long rx_bytes = 0;
229 unsigned long rx_packets = 0;
230 unsigned long tx_bytes = 0;
231 unsigned long tx_packets = 0;
232 unsigned int index;
233
234 if (vif->queues == NULL)
235 goto out;
236
237 /* Aggregate tx and rx stats from each queue */
238 for (index = 0; index < num_queues; ++index) {
239 queue = &vif->queues[index];
240 rx_bytes += queue->stats.rx_bytes;
241 rx_packets += queue->stats.rx_packets;
242 tx_bytes += queue->stats.tx_bytes;
243 tx_packets += queue->stats.tx_packets;
244 }
245
246out:
247 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;
258 unsigned int num_queues = vif->dev->real_num_tx_queues;
259 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;
274 unsigned int num_queues = vif->dev->real_num_tx_queues;
275 unsigned int queue_index;
276
277 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
278 queue = &vif->queues[queue_index];
279 napi_disable(&queue->napi);
280 disable_irq(queue->tx_irq);
281 if (queue->tx_irq != queue->rx_irq)
282 disable_irq(queue->rx_irq);
283 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);
290 if (netif_carrier_ok(dev))
291 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);
299 if (netif_carrier_ok(dev))
300 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);
308 int max = vif->can_sg ? 65535 - VLAN_ETH_HLEN : ETH_DATA_LEN;
309
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 Durrant82cada22013-10-16 17:50:32 +0100323 if (~(vif->gso_mask | vif->gso_prefix_mask) & GSO_BIT(TCPV4))
Michał Mirosław47103042011-04-19 03:35:06 +0000324 features &= ~NETIF_F_TSO;
Paul Durrant82cada22013-10-16 17:50:32 +0100325 if (~(vif->gso_mask | vif->gso_prefix_mask) & GSO_BIT(TCPV6))
326 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);
381 unsigned int num_queues = dev->real_num_tx_queues;
Ian Campbellf942dc22011-03-15 00:06:18 +0000382 int i;
Wei Liue9ce7cb2014-06-04 10:30:42 +0100383 unsigned int queue_index;
384 struct xenvif_stats *vif_stats;
Ian Campbellf942dc22011-03-15 00:06:18 +0000385
Wei Liue9ce7cb2014-06-04 10:30:42 +0100386 for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++) {
387 unsigned long accum = 0;
388 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
389 vif_stats = &vif->queues[queue_index].stats;
390 accum += *(unsigned long *)(vif_stats + xenvif_stats[i].offset);
391 }
392 data[i] = accum;
393 }
Ian Campbellf942dc22011-03-15 00:06:18 +0000394}
395
396static void xenvif_get_strings(struct net_device *dev, u32 stringset, u8 * data)
397{
398 int i;
399
400 switch (stringset) {
401 case ETH_SS_STATS:
402 for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++)
403 memcpy(data + i * ETH_GSTRING_LEN,
404 xenvif_stats[i].name, ETH_GSTRING_LEN);
405 break;
406 }
407}
408
stephen hemminger813abbb2012-01-04 11:56:58 +0000409static const struct ethtool_ops xenvif_ethtool_ops = {
Ian Campbellf942dc22011-03-15 00:06:18 +0000410 .get_link = ethtool_op_get_link,
411
412 .get_sset_count = xenvif_get_sset_count,
413 .get_ethtool_stats = xenvif_get_ethtool_stats,
414 .get_strings = xenvif_get_strings,
415};
416
stephen hemminger813abbb2012-01-04 11:56:58 +0000417static const struct net_device_ops xenvif_netdev_ops = {
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,
Wei Liue9ce7cb2014-06-04 10:30:42 +0100426 .ndo_select_queue = xenvif_select_queue,
Ian Campbellf942dc22011-03-15 00:06:18 +0000427};
428
429struct xenvif *xenvif_alloc(struct device *parent, domid_t domid,
430 unsigned int handle)
431{
432 int err;
433 struct net_device *dev;
434 struct xenvif *vif;
435 char name[IFNAMSIZ] = {};
436
437 snprintf(name, IFNAMSIZ - 1, "vif%u.%u", domid, handle);
Andrew J. Bennieston8d3d53b2014-06-04 10:30:43 +0100438 /* Allocate a netdev with the max. supported number of queues.
439 * When the guest selects the desired number, it will be updated
440 * via netif_set_real_num_tx_queues().
441 */
442 dev = alloc_netdev_mq(sizeof(struct xenvif), name, ether_setup,
443 xenvif_max_queues);
Ian Campbellf942dc22011-03-15 00:06:18 +0000444 if (dev == NULL) {
Wei Liub3f980b2013-08-26 12:59:38 +0100445 pr_warn("Could not allocate netdev for %s\n", name);
Ian Campbellf942dc22011-03-15 00:06:18 +0000446 return ERR_PTR(-ENOMEM);
447 }
448
449 SET_NETDEV_DEV(dev, parent);
450
451 vif = netdev_priv(dev);
Paul Durrantac3d5ac2013-12-23 09:27:17 +0000452
Ian Campbellf942dc22011-03-15 00:06:18 +0000453 vif->domid = domid;
454 vif->handle = handle;
Ian Campbellf942dc22011-03-15 00:06:18 +0000455 vif->can_sg = 1;
Paul Durrant146c8a72013-10-16 17:50:28 +0100456 vif->ip_csum = 1;
Ian Campbellf942dc22011-03-15 00:06:18 +0000457 vif->dev = dev;
Wei Liue9d8b2c2014-04-01 12:46:12 +0100458 vif->disabled = false;
459
Wei Liue9ce7cb2014-06-04 10:30:42 +0100460 /* Start out with no queues. The call below does not require
461 * rtnl_lock() as it happens before register_netdev().
462 */
463 vif->queues = NULL;
464 netif_set_real_num_tx_queues(dev, 0);
Zoltan Kiss09350782014-03-06 21:48:30 +0000465
Ian Campbellf942dc22011-03-15 00:06:18 +0000466 dev->netdev_ops = &xenvif_netdev_ops;
Paul Durrant146c8a72013-10-16 17:50:28 +0100467 dev->hw_features = NETIF_F_SG |
468 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
Paul Durrant82cada22013-10-16 17:50:32 +0100469 NETIF_F_TSO | NETIF_F_TSO6;
Paul Durrant7365bcf2013-10-16 17:50:30 +0100470 dev->features = dev->hw_features | NETIF_F_RXCSUM;
Wilfried Klaebe7ad24ea2014-05-11 00:12:32 +0000471 dev->ethtool_ops = &xenvif_ethtool_ops;
Ian Campbellf942dc22011-03-15 00:06:18 +0000472
473 dev->tx_queue_len = XENVIF_QUEUE_LENGTH;
474
475 /*
476 * Initialise a dummy MAC address. We choose the numerically
477 * largest non-broadcast address to prevent the address getting
478 * stolen by an Ethernet bridge for STP purposes.
479 * (FE:FF:FF:FF:FF:FF)
480 */
481 memset(dev->dev_addr, 0xFF, ETH_ALEN);
482 dev->dev_addr[0] &= ~0x01;
483
484 netif_carrier_off(dev);
485
486 err = register_netdev(dev);
487 if (err) {
488 netdev_warn(dev, "Could not register device: err=%d\n", err);
489 free_netdev(dev);
490 return ERR_PTR(err);
491 }
492
493 netdev_dbg(dev, "Successfully created xenvif\n");
Paul Durrant279f4382013-09-17 17:46:08 +0100494
495 __module_get(THIS_MODULE);
496
Ian Campbellf942dc22011-03-15 00:06:18 +0000497 return vif;
498}
499
Wei Liue9ce7cb2014-06-04 10:30:42 +0100500int xenvif_init_queue(struct xenvif_queue *queue)
Ian Campbellf942dc22011-03-15 00:06:18 +0000501{
Wei Liue9ce7cb2014-06-04 10:30:42 +0100502 int err, i;
Ian Campbellf942dc22011-03-15 00:06:18 +0000503
Wei Liue9ce7cb2014-06-04 10:30:42 +0100504 queue->credit_bytes = queue->remaining_credit = ~0UL;
505 queue->credit_usec = 0UL;
506 init_timer(&queue->credit_timeout);
507 queue->credit_window_start = get_jiffies_64();
Ian Campbellf942dc22011-03-15 00:06:18 +0000508
Wei Liue9ce7cb2014-06-04 10:30:42 +0100509 skb_queue_head_init(&queue->rx_queue);
510 skb_queue_head_init(&queue->tx_queue);
Ian Campbellf942dc22011-03-15 00:06:18 +0000511
Wei Liue9ce7cb2014-06-04 10:30:42 +0100512 queue->pending_cons = 0;
513 queue->pending_prod = MAX_PENDING_REQS;
514 for (i = 0; i < MAX_PENDING_REQS; ++i)
515 queue->pending_ring[i] = i;
Paul Durrantca2f09f2013-12-06 16:36:07 +0000516
Wei Liue9ce7cb2014-06-04 10:30:42 +0100517 spin_lock_init(&queue->callback_lock);
518 spin_lock_init(&queue->response_lock);
Wei Liue1f00a692013-05-22 06:34:45 +0000519
Wei Liue9ce7cb2014-06-04 10:30:42 +0100520 /* If ballooning is disabled, this will consume real memory, so you
521 * better enable it. The long term solution would be to use just a
522 * bunch of valid page descriptors, without dependency on ballooning
523 */
524 err = alloc_xenballooned_pages(MAX_PENDING_REQS,
525 queue->mmap_pages,
526 false);
527 if (err) {
528 netdev_err(queue->vif->dev, "Could not reserve mmap_pages\n");
529 return -ENOMEM;
Wei Liue1f00a692013-05-22 06:34:45 +0000530 }
Ian Campbellf942dc22011-03-15 00:06:18 +0000531
Wei Liue9ce7cb2014-06-04 10:30:42 +0100532 for (i = 0; i < MAX_PENDING_REQS; i++) {
533 queue->pending_tx_info[i].callback_struct = (struct ubuf_info)
534 { .callback = xenvif_zerocopy_callback,
535 .ctx = NULL,
536 .desc = i };
537 queue->grant_tx_handle[i] = NETBACK_INVALID_HANDLE;
Wei Liub3f980b2013-08-26 12:59:38 +0100538 }
Ian Campbellf942dc22011-03-15 00:06:18 +0000539
Wei Liue9ce7cb2014-06-04 10:30:42 +0100540 init_timer(&queue->wake_queue);
Paul Durrant67fa3662013-12-03 14:06:25 +0000541
Wei Liue9ce7cb2014-06-04 10:30:42 +0100542 netif_napi_add(queue->vif->dev, &queue->napi, xenvif_poll,
543 XENVIF_NAPI_WEIGHT);
Zoltan Kissf53c3fe2014-03-06 21:48:26 +0000544
Wei Liue9ce7cb2014-06-04 10:30:42 +0100545 return 0;
546}
Zoltan Kissf53c3fe2014-03-06 21:48:26 +0000547
Wei Liue9ce7cb2014-06-04 10:30:42 +0100548void xenvif_carrier_on(struct xenvif *vif)
549{
Ian Campbellf942dc22011-03-15 00:06:18 +0000550 rtnl_lock();
Michał Mirosław47103042011-04-19 03:35:06 +0000551 if (!vif->can_sg && vif->dev->mtu > ETH_DATA_LEN)
552 dev_set_mtu(vif->dev, ETH_DATA_LEN);
553 netdev_update_features(vif->dev);
554 netif_carrier_on(vif->dev);
David Vrabeld0e5d832011-09-30 06:37:51 +0000555 if (netif_running(vif->dev))
556 xenvif_up(vif);
Ian Campbellf942dc22011-03-15 00:06:18 +0000557 rtnl_unlock();
Wei Liue9ce7cb2014-06-04 10:30:42 +0100558}
Ian Campbellf942dc22011-03-15 00:06:18 +0000559
Wei Liue9ce7cb2014-06-04 10:30:42 +0100560int xenvif_connect(struct xenvif_queue *queue, unsigned long tx_ring_ref,
561 unsigned long rx_ring_ref, unsigned int tx_evtchn,
562 unsigned int rx_evtchn)
563{
564 struct task_struct *task;
565 int err = -ENOMEM;
566
567 BUG_ON(queue->tx_irq);
568 BUG_ON(queue->task);
569 BUG_ON(queue->dealloc_task);
570
571 err = xenvif_map_frontend_rings(queue, tx_ring_ref, rx_ring_ref);
572 if (err < 0)
573 goto err;
574
575 init_waitqueue_head(&queue->wq);
576 init_waitqueue_head(&queue->dealloc_wq);
577
578 if (tx_evtchn == rx_evtchn) {
579 /* feature-split-event-channels == 0 */
580 err = bind_interdomain_evtchn_to_irqhandler(
581 queue->vif->domid, tx_evtchn, xenvif_interrupt, 0,
582 queue->name, queue);
583 if (err < 0)
584 goto err_unmap;
585 queue->tx_irq = queue->rx_irq = err;
586 disable_irq(queue->tx_irq);
587 } else {
588 /* feature-split-event-channels == 1 */
589 snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
590 "%s-tx", queue->name);
591 err = bind_interdomain_evtchn_to_irqhandler(
592 queue->vif->domid, tx_evtchn, xenvif_tx_interrupt, 0,
593 queue->tx_irq_name, queue);
594 if (err < 0)
595 goto err_unmap;
596 queue->tx_irq = err;
597 disable_irq(queue->tx_irq);
598
599 snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
600 "%s-rx", queue->name);
601 err = bind_interdomain_evtchn_to_irqhandler(
602 queue->vif->domid, rx_evtchn, xenvif_rx_interrupt, 0,
603 queue->rx_irq_name, queue);
604 if (err < 0)
605 goto err_tx_unbind;
606 queue->rx_irq = err;
607 disable_irq(queue->rx_irq);
608 }
609
610 task = kthread_create(xenvif_kthread_guest_rx,
611 (void *)queue, "%s-guest-rx", queue->name);
612 if (IS_ERR(task)) {
613 pr_warn("Could not allocate kthread for %s\n", queue->name);
614 err = PTR_ERR(task);
615 goto err_rx_unbind;
616 }
617 queue->task = task;
618
619 task = kthread_create(xenvif_dealloc_kthread,
620 (void *)queue, "%s-dealloc", queue->name);
621 if (IS_ERR(task)) {
622 pr_warn("Could not allocate kthread for %s\n", queue->name);
623 err = PTR_ERR(task);
624 goto err_rx_unbind;
625 }
626 queue->dealloc_task = task;
627
628 wake_up_process(queue->task);
629 wake_up_process(queue->dealloc_task);
Wei Liub3f980b2013-08-26 12:59:38 +0100630
Ian Campbellf942dc22011-03-15 00:06:18 +0000631 return 0;
Wei Liub3f980b2013-08-26 12:59:38 +0100632
633err_rx_unbind:
Wei Liue9ce7cb2014-06-04 10:30:42 +0100634 unbind_from_irqhandler(queue->rx_irq, queue);
635 queue->rx_irq = 0;
Wei Liue1f00a692013-05-22 06:34:45 +0000636err_tx_unbind:
Wei Liue9ce7cb2014-06-04 10:30:42 +0100637 unbind_from_irqhandler(queue->tx_irq, queue);
638 queue->tx_irq = 0;
Ian Campbellf942dc22011-03-15 00:06:18 +0000639err_unmap:
Wei Liue9ce7cb2014-06-04 10:30:42 +0100640 xenvif_unmap_frontend_rings(queue);
Ian Campbellf942dc22011-03-15 00:06:18 +0000641err:
Wei Liub103f352013-05-16 23:26:11 +0000642 module_put(THIS_MODULE);
Ian Campbellf942dc22011-03-15 00:06:18 +0000643 return err;
644}
645
Ian Campbell488562862013-02-06 23:41:35 +0000646void xenvif_carrier_off(struct xenvif *vif)
Ian Campbellf942dc22011-03-15 00:06:18 +0000647{
648 struct net_device *dev = vif->dev;
Ian Campbell488562862013-02-06 23:41:35 +0000649
650 rtnl_lock();
651 netif_carrier_off(dev); /* discard queued packets */
652 if (netif_running(dev))
653 xenvif_down(vif);
654 rtnl_unlock();
Ian Campbell488562862013-02-06 23:41:35 +0000655}
656
Wei Liue9ce7cb2014-06-04 10:30:42 +0100657static void xenvif_wait_unmap_timeout(struct xenvif_queue *queue,
658 unsigned int worst_case_skb_lifetime)
659{
660 int i, unmap_timeout = 0;
661
662 for (i = 0; i < MAX_PENDING_REQS; ++i) {
663 if (queue->grant_tx_handle[i] != NETBACK_INVALID_HANDLE) {
664 unmap_timeout++;
665 schedule_timeout(msecs_to_jiffies(1000));
666 if (unmap_timeout > worst_case_skb_lifetime &&
667 net_ratelimit())
668 netdev_err(queue->vif->dev,
669 "Page still granted! Index: %x\n",
670 i);
671 i = -1;
672 }
673 }
674}
675
Ian Campbell488562862013-02-06 23:41:35 +0000676void xenvif_disconnect(struct xenvif *vif)
677{
Wei Liue9ce7cb2014-06-04 10:30:42 +0100678 struct xenvif_queue *queue = NULL;
679 unsigned int num_queues = vif->dev->real_num_tx_queues;
680 unsigned int queue_index;
681
Ian Campbell488562862013-02-06 23:41:35 +0000682 if (netif_carrier_ok(vif->dev))
683 xenvif_carrier_off(vif);
Ian Campbellf942dc22011-03-15 00:06:18 +0000684
Wei Liue9ce7cb2014-06-04 10:30:42 +0100685 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
686 queue = &vif->queues[queue_index];
David Vrabeldb739ef2013-11-21 15:26:09 +0000687
Wei Liue9ce7cb2014-06-04 10:30:42 +0100688 if (queue->task) {
689 del_timer_sync(&queue->wake_queue);
690 kthread_stop(queue->task);
691 queue->task = NULL;
Wei Liue1f00a692013-05-22 06:34:45 +0000692 }
Ian Campbellf942dc22011-03-15 00:06:18 +0000693
Wei Liue9ce7cb2014-06-04 10:30:42 +0100694 if (queue->dealloc_task) {
695 kthread_stop(queue->dealloc_task);
696 queue->dealloc_task = NULL;
697 }
698
699 if (queue->tx_irq) {
700 if (queue->tx_irq == queue->rx_irq)
701 unbind_from_irqhandler(queue->tx_irq, queue);
702 else {
703 unbind_from_irqhandler(queue->tx_irq, queue);
704 unbind_from_irqhandler(queue->rx_irq, queue);
705 }
706 queue->tx_irq = 0;
707 }
708
709 xenvif_unmap_frontend_rings(queue);
710 }
Paul Durrant279f4382013-09-17 17:46:08 +0100711}
712
Andrew J. Bennieston8d3d53b2014-06-04 10:30:43 +0100713/* Reverse the relevant parts of xenvif_init_queue().
714 * Used for queue teardown from xenvif_free(), and on the
715 * error handling paths in xenbus.c:connect().
716 */
717void xenvif_deinit_queue(struct xenvif_queue *queue)
718{
719 free_xenballooned_pages(MAX_PENDING_REQS, queue->mmap_pages);
720 netif_napi_del(&queue->napi);
721}
722
Paul Durrant279f4382013-09-17 17:46:08 +0100723void xenvif_free(struct xenvif *vif)
724{
Wei Liue9ce7cb2014-06-04 10:30:42 +0100725 struct xenvif_queue *queue = NULL;
726 unsigned int num_queues = vif->dev->real_num_tx_queues;
727 unsigned int queue_index;
Zoltan Kiss0e59a4a2014-03-24 23:59:50 +0000728 /* Here we want to avoid timeout messages if an skb can be legitimately
729 * stuck somewhere else. Realistically this could be an another vif's
Zoltan Kiss09350782014-03-06 21:48:30 +0000730 * internal or QDisc queue. That another vif also has this
731 * rx_drain_timeout_msecs timeout, but the timer only ditches the
732 * internal queue. After that, the QDisc queue can put in worst case
733 * XEN_NETIF_RX_RING_SIZE / MAX_SKB_FRAGS skbs into that another vif's
734 * internal queue, so we need several rounds of such timeouts until we
735 * can be sure that no another vif should have skb's from us. We are
Zoltan Kiss0e59a4a2014-03-24 23:59:50 +0000736 * not sending more skb's, so newly stuck packets are not interesting
Zoltan Kiss09350782014-03-06 21:48:30 +0000737 * for us here.
738 */
739 unsigned int worst_case_skb_lifetime = (rx_drain_timeout_msecs/1000) *
740 DIV_ROUND_UP(XENVIF_QUEUE_LENGTH, (XEN_NETIF_RX_RING_SIZE / MAX_SKB_FRAGS));
Zoltan Kissf53c3fe2014-03-06 21:48:26 +0000741
Wei Liue9ce7cb2014-06-04 10:30:42 +0100742 unregister_netdev(vif->dev);
743
744 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
745 queue = &vif->queues[queue_index];
Wei Liue9ce7cb2014-06-04 10:30:42 +0100746 xenvif_wait_unmap_timeout(queue, worst_case_skb_lifetime);
Andrew J. Bennieston8d3d53b2014-06-04 10:30:43 +0100747 xenvif_deinit_queue(queue);
Zoltan Kissf53c3fe2014-03-06 21:48:26 +0000748 }
749
Wei Liue9ce7cb2014-06-04 10:30:42 +0100750 /* Free the array of queues. The call below does not require
751 * rtnl_lock() because it happens after unregister_netdev().
752 */
753 netif_set_real_num_tx_queues(vif->dev, 0);
754 vfree(vif->queues);
755 vif->queues = NULL;
Ian Campbellf942dc22011-03-15 00:06:18 +0000756
Ian Campbellf942dc22011-03-15 00:06:18 +0000757 free_netdev(vif->dev);
Wei Liub103f352013-05-16 23:26:11 +0000758
Paul Durrant279f4382013-09-17 17:46:08 +0100759 module_put(THIS_MODULE);
Ian Campbellf942dc22011-03-15 00:06:18 +0000760}