blob: 852da34b89615ae7c663f18ea210d1a5ea046808 [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>
Arnd Bergmanne7b599d2014-06-10 10:34:36 +020037#include <linux/vmalloc.h>
Ian Campbellf942dc22011-03-15 00:06:18 +000038
39#include <xen/events.h>
40#include <asm/xen/hypercall.h>
Zoltan Kissf53c3fe2014-03-06 21:48:26 +000041#include <xen/balloon.h>
Ian Campbellf942dc22011-03-15 00:06:18 +000042
43#define XENVIF_QUEUE_LENGTH 32
Wei Liub3f980b2013-08-26 12:59:38 +010044#define XENVIF_NAPI_WEIGHT 64
Ian Campbellf942dc22011-03-15 00:06:18 +000045
Wei Liue9ce7cb2014-06-04 10:30:42 +010046static inline void xenvif_stop_queue(struct xenvif_queue *queue)
47{
48 struct net_device *dev = queue->vif->dev;
49
50 if (!queue->vif->can_queue)
51 return;
52
53 netif_tx_stop_queue(netdev_get_tx_queue(dev, queue->id));
54}
55
Ian Campbellf942dc22011-03-15 00:06:18 +000056int xenvif_schedulable(struct xenvif *vif)
57{
58 return netif_running(vif->dev) && netif_carrier_ok(vif->dev);
59}
60
Wei Liue1f00a692013-05-22 06:34:45 +000061static irqreturn_t xenvif_tx_interrupt(int irq, void *dev_id)
Ian Campbellf942dc22011-03-15 00:06:18 +000062{
Wei Liue9ce7cb2014-06-04 10:30:42 +010063 struct xenvif_queue *queue = dev_id;
Ian Campbellf942dc22011-03-15 00:06:18 +000064
Wei Liue9ce7cb2014-06-04 10:30:42 +010065 if (RING_HAS_UNCONSUMED_REQUESTS(&queue->tx))
66 napi_schedule(&queue->napi);
Ian Campbellf942dc22011-03-15 00:06:18 +000067
Wei Liue1f00a692013-05-22 06:34:45 +000068 return IRQ_HANDLED;
69}
70
Wei Liue9ce7cb2014-06-04 10:30:42 +010071int xenvif_poll(struct napi_struct *napi, int budget)
Wei Liub3f980b2013-08-26 12:59:38 +010072{
Wei Liue9ce7cb2014-06-04 10:30:42 +010073 struct xenvif_queue *queue =
74 container_of(napi, struct xenvif_queue, napi);
Wei Liub3f980b2013-08-26 12:59:38 +010075 int work_done;
76
Wei Liue9d8b2c2014-04-01 12:46:12 +010077 /* This vif is rogue, we pretend we've there is nothing to do
78 * for this vif to deschedule it from NAPI. But this interface
79 * will be turned off in thread context later.
80 */
Wei Liue9ce7cb2014-06-04 10:30:42 +010081 if (unlikely(queue->vif->disabled)) {
Wei Liue9d8b2c2014-04-01 12:46:12 +010082 napi_complete(napi);
83 return 0;
84 }
85
Wei Liue9ce7cb2014-06-04 10:30:42 +010086 work_done = xenvif_tx_action(queue, budget);
Wei Liub3f980b2013-08-26 12:59:38 +010087
88 if (work_done < budget) {
David Vrabel0d08fce2014-05-16 12:26:04 +010089 napi_complete(napi);
Wei Liue9ce7cb2014-06-04 10:30:42 +010090 xenvif_napi_schedule_or_enable_events(queue);
Wei Liub3f980b2013-08-26 12:59:38 +010091 }
92
93 return work_done;
94}
95
Wei Liue1f00a692013-05-22 06:34:45 +000096static irqreturn_t xenvif_rx_interrupt(int irq, void *dev_id)
97{
Wei Liue9ce7cb2014-06-04 10:30:42 +010098 struct xenvif_queue *queue = dev_id;
Wei Liue1f00a692013-05-22 06:34:45 +000099
Wei Liue9ce7cb2014-06-04 10:30:42 +0100100 xenvif_kick_thread(queue);
Ian Campbellf942dc22011-03-15 00:06:18 +0000101
102 return IRQ_HANDLED;
103}
104
Wei Liue1f00a692013-05-22 06:34:45 +0000105static irqreturn_t xenvif_interrupt(int irq, void *dev_id)
106{
107 xenvif_tx_interrupt(irq, dev_id);
108 xenvif_rx_interrupt(irq, dev_id);
109
110 return IRQ_HANDLED;
111}
112
Wei Liue9ce7cb2014-06-04 10:30:42 +0100113int xenvif_queue_stopped(struct xenvif_queue *queue)
Zoltan Kiss09350782014-03-06 21:48:30 +0000114{
Wei Liue9ce7cb2014-06-04 10:30:42 +0100115 struct net_device *dev = queue->vif->dev;
116 unsigned int id = queue->id;
117 return netif_tx_queue_stopped(netdev_get_tx_queue(dev, id));
118}
Zoltan Kiss09350782014-03-06 21:48:30 +0000119
Wei Liue9ce7cb2014-06-04 10:30:42 +0100120void xenvif_wake_queue(struct xenvif_queue *queue)
121{
122 struct net_device *dev = queue->vif->dev;
123 unsigned int id = queue->id;
124 netif_tx_wake_queue(netdev_get_tx_queue(dev, id));
125}
126
127/* Callback to wake the queue and drain it on timeout */
128static void xenvif_wake_queue_callback(unsigned long data)
129{
130 struct xenvif_queue *queue = (struct xenvif_queue *)data;
131
132 if (xenvif_queue_stopped(queue)) {
133 netdev_err(queue->vif->dev, "draining TX queue\n");
134 queue->rx_queue_purge = true;
135 xenvif_kick_thread(queue);
136 xenvif_wake_queue(queue);
Zoltan Kiss09350782014-03-06 21:48:30 +0000137 }
138}
139
Wei Liue9ce7cb2014-06-04 10:30:42 +0100140static u16 xenvif_select_queue(struct net_device *dev, struct sk_buff *skb,
141 void *accel_priv, select_queue_fallback_t fallback)
142{
Wei Liue9ce7cb2014-06-04 10:30:42 +0100143 unsigned int num_queues = dev->real_num_tx_queues;
144 u32 hash;
145 u16 queue_index;
146
147 /* First, check if there is only one queue to optimise the
148 * single-queue or old frontend scenario.
149 */
150 if (num_queues == 1) {
151 queue_index = 0;
152 } else {
153 /* Use skb_get_hash to obtain an L4 hash if available */
154 hash = skb_get_hash(skb);
155 queue_index = hash % num_queues;
156 }
157
158 return queue_index;
159}
160
Ian Campbellf942dc22011-03-15 00:06:18 +0000161static int xenvif_start_xmit(struct sk_buff *skb, struct net_device *dev)
162{
163 struct xenvif *vif = netdev_priv(dev);
Wei Liue9ce7cb2014-06-04 10:30:42 +0100164 struct xenvif_queue *queue = NULL;
165 unsigned int num_queues = dev->real_num_tx_queues;
166 u16 index;
Paul Durrantca2f09f2013-12-06 16:36:07 +0000167 int min_slots_needed;
Ian Campbellf942dc22011-03-15 00:06:18 +0000168
169 BUG_ON(skb->dev != dev);
170
Wei Liue9ce7cb2014-06-04 10:30:42 +0100171 /* Drop the packet if queues are not set up */
172 if (num_queues < 1)
173 goto drop;
174
175 /* Obtain the queue to be used to transmit this packet */
176 index = skb_get_queue_mapping(skb);
177 if (index >= num_queues) {
178 pr_warn_ratelimited("Invalid queue %hu for packet on interface %s\n.",
179 index, vif->dev->name);
180 index %= num_queues;
181 }
182 queue = &vif->queues[index];
183
184 /* Drop the packet if queue is not ready */
185 if (queue->task == NULL ||
186 queue->dealloc_task == NULL ||
Zoltan Kissf53c3fe2014-03-06 21:48:26 +0000187 !xenvif_schedulable(vif))
Ian Campbellf942dc22011-03-15 00:06:18 +0000188 goto drop;
189
Paul Durrantca2f09f2013-12-06 16:36:07 +0000190 /* At best we'll need one slot for the header and one for each
191 * frag.
192 */
193 min_slots_needed = 1 + skb_shinfo(skb)->nr_frags;
Ian Campbellf942dc22011-03-15 00:06:18 +0000194
Paul Durrantca2f09f2013-12-06 16:36:07 +0000195 /* If the skb is GSO then we'll also need an extra slot for the
196 * metadata.
197 */
Wei Liu836fbaf2014-03-11 12:45:32 +0000198 if (skb_is_gso(skb))
Paul Durrantca2f09f2013-12-06 16:36:07 +0000199 min_slots_needed++;
Ian Campbellf942dc22011-03-15 00:06:18 +0000200
Paul Durrantca2f09f2013-12-06 16:36:07 +0000201 /* If the skb can't possibly fit in the remaining slots
202 * then turn off the queue to give the ring a chance to
203 * drain.
204 */
Wei Liue9ce7cb2014-06-04 10:30:42 +0100205 if (!xenvif_rx_ring_slots_available(queue, min_slots_needed)) {
206 queue->wake_queue.function = xenvif_wake_queue_callback;
207 queue->wake_queue.data = (unsigned long)queue;
208 xenvif_stop_queue(queue);
209 mod_timer(&queue->wake_queue,
Zoltan Kiss09350782014-03-06 21:48:30 +0000210 jiffies + rx_drain_timeout_jiffies);
211 }
Ian Campbellf942dc22011-03-15 00:06:18 +0000212
Wei Liue9ce7cb2014-06-04 10:30:42 +0100213 skb_queue_tail(&queue->rx_queue, skb);
214 xenvif_kick_thread(queue);
Ian Campbellf942dc22011-03-15 00:06:18 +0000215
216 return NETDEV_TX_OK;
217
218 drop:
219 vif->dev->stats.tx_dropped++;
220 dev_kfree_skb(skb);
221 return NETDEV_TX_OK;
222}
223
Ian Campbellf942dc22011-03-15 00:06:18 +0000224static struct net_device_stats *xenvif_get_stats(struct net_device *dev)
225{
226 struct xenvif *vif = netdev_priv(dev);
Wei Liue9ce7cb2014-06-04 10:30:42 +0100227 struct xenvif_queue *queue = NULL;
228 unsigned int num_queues = dev->real_num_tx_queues;
229 unsigned long rx_bytes = 0;
230 unsigned long rx_packets = 0;
231 unsigned long tx_bytes = 0;
232 unsigned long tx_packets = 0;
233 unsigned int index;
234
235 if (vif->queues == NULL)
236 goto out;
237
238 /* Aggregate tx and rx stats from each queue */
239 for (index = 0; index < num_queues; ++index) {
240 queue = &vif->queues[index];
241 rx_bytes += queue->stats.rx_bytes;
242 rx_packets += queue->stats.rx_packets;
243 tx_bytes += queue->stats.tx_bytes;
244 tx_packets += queue->stats.tx_packets;
245 }
246
247out:
248 vif->dev->stats.rx_bytes = rx_bytes;
249 vif->dev->stats.rx_packets = rx_packets;
250 vif->dev->stats.tx_bytes = tx_bytes;
251 vif->dev->stats.tx_packets = tx_packets;
252
Ian Campbellf942dc22011-03-15 00:06:18 +0000253 return &vif->dev->stats;
254}
255
256static void xenvif_up(struct xenvif *vif)
257{
Wei Liue9ce7cb2014-06-04 10:30:42 +0100258 struct xenvif_queue *queue = NULL;
259 unsigned int num_queues = vif->dev->real_num_tx_queues;
260 unsigned int queue_index;
261
262 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
263 queue = &vif->queues[queue_index];
264 napi_enable(&queue->napi);
265 enable_irq(queue->tx_irq);
266 if (queue->tx_irq != queue->rx_irq)
267 enable_irq(queue->rx_irq);
268 xenvif_napi_schedule_or_enable_events(queue);
269 }
Ian Campbellf942dc22011-03-15 00:06:18 +0000270}
271
272static void xenvif_down(struct xenvif *vif)
273{
Wei Liue9ce7cb2014-06-04 10:30:42 +0100274 struct xenvif_queue *queue = NULL;
275 unsigned int num_queues = vif->dev->real_num_tx_queues;
276 unsigned int queue_index;
277
278 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
279 queue = &vif->queues[queue_index];
280 napi_disable(&queue->napi);
281 disable_irq(queue->tx_irq);
282 if (queue->tx_irq != queue->rx_irq)
283 disable_irq(queue->rx_irq);
284 del_timer_sync(&queue->credit_timeout);
285 }
Ian Campbellf942dc22011-03-15 00:06:18 +0000286}
287
288static int xenvif_open(struct net_device *dev)
289{
290 struct xenvif *vif = netdev_priv(dev);
291 if (netif_carrier_ok(dev))
292 xenvif_up(vif);
Wei Liue9ce7cb2014-06-04 10:30:42 +0100293 netif_tx_start_all_queues(dev);
Ian Campbellf942dc22011-03-15 00:06:18 +0000294 return 0;
295}
296
297static int xenvif_close(struct net_device *dev)
298{
299 struct xenvif *vif = netdev_priv(dev);
300 if (netif_carrier_ok(dev))
301 xenvif_down(vif);
Wei Liue9ce7cb2014-06-04 10:30:42 +0100302 netif_tx_stop_all_queues(dev);
Ian Campbellf942dc22011-03-15 00:06:18 +0000303 return 0;
304}
305
306static int xenvif_change_mtu(struct net_device *dev, int mtu)
307{
308 struct xenvif *vif = netdev_priv(dev);
309 int max = vif->can_sg ? 65535 - VLAN_ETH_HLEN : ETH_DATA_LEN;
310
311 if (mtu > max)
312 return -EINVAL;
313 dev->mtu = mtu;
314 return 0;
315}
316
Michał Mirosławc8f44af2011-11-15 15:29:55 +0000317static netdev_features_t xenvif_fix_features(struct net_device *dev,
318 netdev_features_t features)
Ian Campbellf942dc22011-03-15 00:06:18 +0000319{
320 struct xenvif *vif = netdev_priv(dev);
Ian Campbellf942dc22011-03-15 00:06:18 +0000321
Michał Mirosław47103042011-04-19 03:35:06 +0000322 if (!vif->can_sg)
323 features &= ~NETIF_F_SG;
Paul Durrant82cada22013-10-16 17:50:32 +0100324 if (~(vif->gso_mask | vif->gso_prefix_mask) & GSO_BIT(TCPV4))
Michał Mirosław47103042011-04-19 03:35:06 +0000325 features &= ~NETIF_F_TSO;
Paul Durrant82cada22013-10-16 17:50:32 +0100326 if (~(vif->gso_mask | vif->gso_prefix_mask) & GSO_BIT(TCPV6))
327 features &= ~NETIF_F_TSO6;
Paul Durrant146c8a72013-10-16 17:50:28 +0100328 if (!vif->ip_csum)
Michał Mirosław47103042011-04-19 03:35:06 +0000329 features &= ~NETIF_F_IP_CSUM;
Paul Durrant146c8a72013-10-16 17:50:28 +0100330 if (!vif->ipv6_csum)
331 features &= ~NETIF_F_IPV6_CSUM;
Ian Campbellf942dc22011-03-15 00:06:18 +0000332
Michał Mirosław47103042011-04-19 03:35:06 +0000333 return features;
Ian Campbellf942dc22011-03-15 00:06:18 +0000334}
335
336static const struct xenvif_stat {
337 char name[ETH_GSTRING_LEN];
338 u16 offset;
339} xenvif_stats[] = {
340 {
341 "rx_gso_checksum_fixup",
Wei Liue9ce7cb2014-06-04 10:30:42 +0100342 offsetof(struct xenvif_stats, rx_gso_checksum_fixup)
Ian Campbellf942dc22011-03-15 00:06:18 +0000343 },
Zoltan Kiss1bb332a2014-03-06 21:48:28 +0000344 /* If (sent != success + fail), there are probably packets never
345 * freed up properly!
346 */
347 {
348 "tx_zerocopy_sent",
Wei Liue9ce7cb2014-06-04 10:30:42 +0100349 offsetof(struct xenvif_stats, tx_zerocopy_sent),
Zoltan Kiss1bb332a2014-03-06 21:48:28 +0000350 },
351 {
352 "tx_zerocopy_success",
Wei Liue9ce7cb2014-06-04 10:30:42 +0100353 offsetof(struct xenvif_stats, tx_zerocopy_success),
Zoltan Kiss1bb332a2014-03-06 21:48:28 +0000354 },
355 {
356 "tx_zerocopy_fail",
Wei Liue9ce7cb2014-06-04 10:30:42 +0100357 offsetof(struct xenvif_stats, tx_zerocopy_fail)
Zoltan Kiss1bb332a2014-03-06 21:48:28 +0000358 },
Zoltan Kisse3377f32014-03-06 21:48:29 +0000359 /* Number of packets exceeding MAX_SKB_FRAG slots. You should use
360 * a guest with the same MAX_SKB_FRAG
361 */
362 {
363 "tx_frag_overflow",
Wei Liue9ce7cb2014-06-04 10:30:42 +0100364 offsetof(struct xenvif_stats, tx_frag_overflow)
Zoltan Kisse3377f32014-03-06 21:48:29 +0000365 },
Ian Campbellf942dc22011-03-15 00:06:18 +0000366};
367
368static int xenvif_get_sset_count(struct net_device *dev, int string_set)
369{
370 switch (string_set) {
371 case ETH_SS_STATS:
372 return ARRAY_SIZE(xenvif_stats);
373 default:
374 return -EINVAL;
375 }
376}
377
378static void xenvif_get_ethtool_stats(struct net_device *dev,
379 struct ethtool_stats *stats, u64 * data)
380{
Wei Liue9ce7cb2014-06-04 10:30:42 +0100381 struct xenvif *vif = netdev_priv(dev);
382 unsigned int num_queues = dev->real_num_tx_queues;
Ian Campbellf942dc22011-03-15 00:06:18 +0000383 int i;
Wei Liue9ce7cb2014-06-04 10:30:42 +0100384 unsigned int queue_index;
385 struct xenvif_stats *vif_stats;
Ian Campbellf942dc22011-03-15 00:06:18 +0000386
Wei Liue9ce7cb2014-06-04 10:30:42 +0100387 for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++) {
388 unsigned long accum = 0;
389 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
390 vif_stats = &vif->queues[queue_index].stats;
391 accum += *(unsigned long *)(vif_stats + xenvif_stats[i].offset);
392 }
393 data[i] = accum;
394 }
Ian Campbellf942dc22011-03-15 00:06:18 +0000395}
396
397static void xenvif_get_strings(struct net_device *dev, u32 stringset, u8 * data)
398{
399 int i;
400
401 switch (stringset) {
402 case ETH_SS_STATS:
403 for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++)
404 memcpy(data + i * ETH_GSTRING_LEN,
405 xenvif_stats[i].name, ETH_GSTRING_LEN);
406 break;
407 }
408}
409
stephen hemminger813abbb2012-01-04 11:56:58 +0000410static const struct ethtool_ops xenvif_ethtool_ops = {
Ian Campbellf942dc22011-03-15 00:06:18 +0000411 .get_link = ethtool_op_get_link,
412
413 .get_sset_count = xenvif_get_sset_count,
414 .get_ethtool_stats = xenvif_get_ethtool_stats,
415 .get_strings = xenvif_get_strings,
416};
417
stephen hemminger813abbb2012-01-04 11:56:58 +0000418static const struct net_device_ops xenvif_netdev_ops = {
Ian Campbellf942dc22011-03-15 00:06:18 +0000419 .ndo_start_xmit = xenvif_start_xmit,
420 .ndo_get_stats = xenvif_get_stats,
421 .ndo_open = xenvif_open,
422 .ndo_stop = xenvif_close,
423 .ndo_change_mtu = xenvif_change_mtu,
Michał Mirosław47103042011-04-19 03:35:06 +0000424 .ndo_fix_features = xenvif_fix_features,
Matt Wilson4a633a62013-01-22 08:08:25 +0000425 .ndo_set_mac_address = eth_mac_addr,
426 .ndo_validate_addr = eth_validate_addr,
Wei Liue9ce7cb2014-06-04 10:30:42 +0100427 .ndo_select_queue = xenvif_select_queue,
Ian Campbellf942dc22011-03-15 00:06:18 +0000428};
429
430struct xenvif *xenvif_alloc(struct device *parent, domid_t domid,
431 unsigned int handle)
432{
433 int err;
434 struct net_device *dev;
435 struct xenvif *vif;
436 char name[IFNAMSIZ] = {};
437
438 snprintf(name, IFNAMSIZ - 1, "vif%u.%u", domid, handle);
Andrew J. Bennieston8d3d53b2014-06-04 10:30:43 +0100439 /* Allocate a netdev with the max. supported number of queues.
440 * When the guest selects the desired number, it will be updated
441 * via netif_set_real_num_tx_queues().
442 */
443 dev = alloc_netdev_mq(sizeof(struct xenvif), name, ether_setup,
444 xenvif_max_queues);
Ian Campbellf942dc22011-03-15 00:06:18 +0000445 if (dev == NULL) {
Wei Liub3f980b2013-08-26 12:59:38 +0100446 pr_warn("Could not allocate netdev for %s\n", name);
Ian Campbellf942dc22011-03-15 00:06:18 +0000447 return ERR_PTR(-ENOMEM);
448 }
449
450 SET_NETDEV_DEV(dev, parent);
451
452 vif = netdev_priv(dev);
Paul Durrantac3d5ac2013-12-23 09:27:17 +0000453
Ian Campbellf942dc22011-03-15 00:06:18 +0000454 vif->domid = domid;
455 vif->handle = handle;
Ian Campbellf942dc22011-03-15 00:06:18 +0000456 vif->can_sg = 1;
Paul Durrant146c8a72013-10-16 17:50:28 +0100457 vif->ip_csum = 1;
Ian Campbellf942dc22011-03-15 00:06:18 +0000458 vif->dev = dev;
Wei Liue9d8b2c2014-04-01 12:46:12 +0100459 vif->disabled = false;
460
Wei Liue9ce7cb2014-06-04 10:30:42 +0100461 /* Start out with no queues. The call below does not require
462 * rtnl_lock() as it happens before register_netdev().
463 */
464 vif->queues = NULL;
465 netif_set_real_num_tx_queues(dev, 0);
Zoltan Kiss09350782014-03-06 21:48:30 +0000466
Ian Campbellf942dc22011-03-15 00:06:18 +0000467 dev->netdev_ops = &xenvif_netdev_ops;
Paul Durrant146c8a72013-10-16 17:50:28 +0100468 dev->hw_features = NETIF_F_SG |
469 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
Paul Durrant82cada22013-10-16 17:50:32 +0100470 NETIF_F_TSO | NETIF_F_TSO6;
Paul Durrant7365bcf2013-10-16 17:50:30 +0100471 dev->features = dev->hw_features | NETIF_F_RXCSUM;
Wilfried Klaebe7ad24ea2014-05-11 00:12:32 +0000472 dev->ethtool_ops = &xenvif_ethtool_ops;
Ian Campbellf942dc22011-03-15 00:06:18 +0000473
474 dev->tx_queue_len = XENVIF_QUEUE_LENGTH;
475
476 /*
477 * Initialise a dummy MAC address. We choose the numerically
478 * largest non-broadcast address to prevent the address getting
479 * stolen by an Ethernet bridge for STP purposes.
480 * (FE:FF:FF:FF:FF:FF)
481 */
482 memset(dev->dev_addr, 0xFF, ETH_ALEN);
483 dev->dev_addr[0] &= ~0x01;
484
485 netif_carrier_off(dev);
486
487 err = register_netdev(dev);
488 if (err) {
489 netdev_warn(dev, "Could not register device: err=%d\n", err);
490 free_netdev(dev);
491 return ERR_PTR(err);
492 }
493
494 netdev_dbg(dev, "Successfully created xenvif\n");
Paul Durrant279f4382013-09-17 17:46:08 +0100495
496 __module_get(THIS_MODULE);
497
Ian Campbellf942dc22011-03-15 00:06:18 +0000498 return vif;
499}
500
Wei Liue9ce7cb2014-06-04 10:30:42 +0100501int xenvif_init_queue(struct xenvif_queue *queue)
Ian Campbellf942dc22011-03-15 00:06:18 +0000502{
Wei Liue9ce7cb2014-06-04 10:30:42 +0100503 int err, i;
Ian Campbellf942dc22011-03-15 00:06:18 +0000504
Wei Liue9ce7cb2014-06-04 10:30:42 +0100505 queue->credit_bytes = queue->remaining_credit = ~0UL;
506 queue->credit_usec = 0UL;
507 init_timer(&queue->credit_timeout);
508 queue->credit_window_start = get_jiffies_64();
Ian Campbellf942dc22011-03-15 00:06:18 +0000509
Wei Liue9ce7cb2014-06-04 10:30:42 +0100510 skb_queue_head_init(&queue->rx_queue);
511 skb_queue_head_init(&queue->tx_queue);
Ian Campbellf942dc22011-03-15 00:06:18 +0000512
Wei Liue9ce7cb2014-06-04 10:30:42 +0100513 queue->pending_cons = 0;
514 queue->pending_prod = MAX_PENDING_REQS;
515 for (i = 0; i < MAX_PENDING_REQS; ++i)
516 queue->pending_ring[i] = i;
Paul Durrantca2f09f2013-12-06 16:36:07 +0000517
Wei Liue9ce7cb2014-06-04 10:30:42 +0100518 spin_lock_init(&queue->callback_lock);
519 spin_lock_init(&queue->response_lock);
Wei Liue1f00a692013-05-22 06:34:45 +0000520
Wei Liue9ce7cb2014-06-04 10:30:42 +0100521 /* If ballooning is disabled, this will consume real memory, so you
522 * better enable it. The long term solution would be to use just a
523 * bunch of valid page descriptors, without dependency on ballooning
524 */
525 err = alloc_xenballooned_pages(MAX_PENDING_REQS,
526 queue->mmap_pages,
527 false);
528 if (err) {
529 netdev_err(queue->vif->dev, "Could not reserve mmap_pages\n");
530 return -ENOMEM;
Wei Liue1f00a692013-05-22 06:34:45 +0000531 }
Ian Campbellf942dc22011-03-15 00:06:18 +0000532
Wei Liue9ce7cb2014-06-04 10:30:42 +0100533 for (i = 0; i < MAX_PENDING_REQS; i++) {
534 queue->pending_tx_info[i].callback_struct = (struct ubuf_info)
535 { .callback = xenvif_zerocopy_callback,
536 .ctx = NULL,
537 .desc = i };
538 queue->grant_tx_handle[i] = NETBACK_INVALID_HANDLE;
Wei Liub3f980b2013-08-26 12:59:38 +0100539 }
Ian Campbellf942dc22011-03-15 00:06:18 +0000540
Wei Liue9ce7cb2014-06-04 10:30:42 +0100541 init_timer(&queue->wake_queue);
Paul Durrant67fa3662013-12-03 14:06:25 +0000542
Wei Liue9ce7cb2014-06-04 10:30:42 +0100543 netif_napi_add(queue->vif->dev, &queue->napi, xenvif_poll,
544 XENVIF_NAPI_WEIGHT);
Zoltan Kissf53c3fe2014-03-06 21:48:26 +0000545
Wei Liue9ce7cb2014-06-04 10:30:42 +0100546 return 0;
547}
Zoltan Kissf53c3fe2014-03-06 21:48:26 +0000548
Wei Liue9ce7cb2014-06-04 10:30:42 +0100549void xenvif_carrier_on(struct xenvif *vif)
550{
Ian Campbellf942dc22011-03-15 00:06:18 +0000551 rtnl_lock();
Michał Mirosław47103042011-04-19 03:35:06 +0000552 if (!vif->can_sg && vif->dev->mtu > ETH_DATA_LEN)
553 dev_set_mtu(vif->dev, ETH_DATA_LEN);
554 netdev_update_features(vif->dev);
555 netif_carrier_on(vif->dev);
David Vrabeld0e5d832011-09-30 06:37:51 +0000556 if (netif_running(vif->dev))
557 xenvif_up(vif);
Ian Campbellf942dc22011-03-15 00:06:18 +0000558 rtnl_unlock();
Wei Liue9ce7cb2014-06-04 10:30:42 +0100559}
Ian Campbellf942dc22011-03-15 00:06:18 +0000560
Wei Liue9ce7cb2014-06-04 10:30:42 +0100561int xenvif_connect(struct xenvif_queue *queue, unsigned long tx_ring_ref,
562 unsigned long rx_ring_ref, unsigned int tx_evtchn,
563 unsigned int rx_evtchn)
564{
565 struct task_struct *task;
566 int err = -ENOMEM;
567
568 BUG_ON(queue->tx_irq);
569 BUG_ON(queue->task);
570 BUG_ON(queue->dealloc_task);
571
572 err = xenvif_map_frontend_rings(queue, tx_ring_ref, rx_ring_ref);
573 if (err < 0)
574 goto err;
575
576 init_waitqueue_head(&queue->wq);
577 init_waitqueue_head(&queue->dealloc_wq);
578
579 if (tx_evtchn == rx_evtchn) {
580 /* feature-split-event-channels == 0 */
581 err = bind_interdomain_evtchn_to_irqhandler(
582 queue->vif->domid, tx_evtchn, xenvif_interrupt, 0,
583 queue->name, queue);
584 if (err < 0)
585 goto err_unmap;
586 queue->tx_irq = queue->rx_irq = err;
587 disable_irq(queue->tx_irq);
588 } else {
589 /* feature-split-event-channels == 1 */
590 snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
591 "%s-tx", queue->name);
592 err = bind_interdomain_evtchn_to_irqhandler(
593 queue->vif->domid, tx_evtchn, xenvif_tx_interrupt, 0,
594 queue->tx_irq_name, queue);
595 if (err < 0)
596 goto err_unmap;
597 queue->tx_irq = err;
598 disable_irq(queue->tx_irq);
599
600 snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
601 "%s-rx", queue->name);
602 err = bind_interdomain_evtchn_to_irqhandler(
603 queue->vif->domid, rx_evtchn, xenvif_rx_interrupt, 0,
604 queue->rx_irq_name, queue);
605 if (err < 0)
606 goto err_tx_unbind;
607 queue->rx_irq = err;
608 disable_irq(queue->rx_irq);
609 }
610
611 task = kthread_create(xenvif_kthread_guest_rx,
612 (void *)queue, "%s-guest-rx", queue->name);
613 if (IS_ERR(task)) {
614 pr_warn("Could not allocate kthread for %s\n", queue->name);
615 err = PTR_ERR(task);
616 goto err_rx_unbind;
617 }
618 queue->task = task;
619
620 task = kthread_create(xenvif_dealloc_kthread,
621 (void *)queue, "%s-dealloc", queue->name);
622 if (IS_ERR(task)) {
623 pr_warn("Could not allocate kthread for %s\n", queue->name);
624 err = PTR_ERR(task);
625 goto err_rx_unbind;
626 }
627 queue->dealloc_task = task;
628
629 wake_up_process(queue->task);
630 wake_up_process(queue->dealloc_task);
Wei Liub3f980b2013-08-26 12:59:38 +0100631
Ian Campbellf942dc22011-03-15 00:06:18 +0000632 return 0;
Wei Liub3f980b2013-08-26 12:59:38 +0100633
634err_rx_unbind:
Wei Liue9ce7cb2014-06-04 10:30:42 +0100635 unbind_from_irqhandler(queue->rx_irq, queue);
636 queue->rx_irq = 0;
Wei Liue1f00a692013-05-22 06:34:45 +0000637err_tx_unbind:
Wei Liue9ce7cb2014-06-04 10:30:42 +0100638 unbind_from_irqhandler(queue->tx_irq, queue);
639 queue->tx_irq = 0;
Ian Campbellf942dc22011-03-15 00:06:18 +0000640err_unmap:
Wei Liue9ce7cb2014-06-04 10:30:42 +0100641 xenvif_unmap_frontend_rings(queue);
Ian Campbellf942dc22011-03-15 00:06:18 +0000642err:
Wei Liub103f352013-05-16 23:26:11 +0000643 module_put(THIS_MODULE);
Ian Campbellf942dc22011-03-15 00:06:18 +0000644 return err;
645}
646
Ian Campbell488562862013-02-06 23:41:35 +0000647void xenvif_carrier_off(struct xenvif *vif)
Ian Campbellf942dc22011-03-15 00:06:18 +0000648{
649 struct net_device *dev = vif->dev;
Ian Campbell488562862013-02-06 23:41:35 +0000650
651 rtnl_lock();
652 netif_carrier_off(dev); /* discard queued packets */
653 if (netif_running(dev))
654 xenvif_down(vif);
655 rtnl_unlock();
Ian Campbell488562862013-02-06 23:41:35 +0000656}
657
Wei Liue9ce7cb2014-06-04 10:30:42 +0100658static void xenvif_wait_unmap_timeout(struct xenvif_queue *queue,
659 unsigned int worst_case_skb_lifetime)
660{
661 int i, unmap_timeout = 0;
662
663 for (i = 0; i < MAX_PENDING_REQS; ++i) {
664 if (queue->grant_tx_handle[i] != NETBACK_INVALID_HANDLE) {
665 unmap_timeout++;
666 schedule_timeout(msecs_to_jiffies(1000));
667 if (unmap_timeout > worst_case_skb_lifetime &&
668 net_ratelimit())
669 netdev_err(queue->vif->dev,
670 "Page still granted! Index: %x\n",
671 i);
672 i = -1;
673 }
674 }
675}
676
Ian Campbell488562862013-02-06 23:41:35 +0000677void xenvif_disconnect(struct xenvif *vif)
678{
Wei Liue9ce7cb2014-06-04 10:30:42 +0100679 struct xenvif_queue *queue = NULL;
680 unsigned int num_queues = vif->dev->real_num_tx_queues;
681 unsigned int queue_index;
682
Ian Campbell488562862013-02-06 23:41:35 +0000683 if (netif_carrier_ok(vif->dev))
684 xenvif_carrier_off(vif);
Ian Campbellf942dc22011-03-15 00:06:18 +0000685
Wei Liue9ce7cb2014-06-04 10:30:42 +0100686 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
687 queue = &vif->queues[queue_index];
David Vrabeldb739ef2013-11-21 15:26:09 +0000688
Wei Liue9ce7cb2014-06-04 10:30:42 +0100689 if (queue->task) {
690 del_timer_sync(&queue->wake_queue);
691 kthread_stop(queue->task);
692 queue->task = NULL;
Wei Liue1f00a692013-05-22 06:34:45 +0000693 }
Ian Campbellf942dc22011-03-15 00:06:18 +0000694
Wei Liue9ce7cb2014-06-04 10:30:42 +0100695 if (queue->dealloc_task) {
696 kthread_stop(queue->dealloc_task);
697 queue->dealloc_task = NULL;
698 }
699
700 if (queue->tx_irq) {
701 if (queue->tx_irq == queue->rx_irq)
702 unbind_from_irqhandler(queue->tx_irq, queue);
703 else {
704 unbind_from_irqhandler(queue->tx_irq, queue);
705 unbind_from_irqhandler(queue->rx_irq, queue);
706 }
707 queue->tx_irq = 0;
708 }
709
710 xenvif_unmap_frontend_rings(queue);
711 }
Paul Durrant279f4382013-09-17 17:46:08 +0100712}
713
Andrew J. Bennieston8d3d53b2014-06-04 10:30:43 +0100714/* Reverse the relevant parts of xenvif_init_queue().
715 * Used for queue teardown from xenvif_free(), and on the
716 * error handling paths in xenbus.c:connect().
717 */
718void xenvif_deinit_queue(struct xenvif_queue *queue)
719{
720 free_xenballooned_pages(MAX_PENDING_REQS, queue->mmap_pages);
721 netif_napi_del(&queue->napi);
722}
723
Paul Durrant279f4382013-09-17 17:46:08 +0100724void xenvif_free(struct xenvif *vif)
725{
Wei Liue9ce7cb2014-06-04 10:30:42 +0100726 struct xenvif_queue *queue = NULL;
727 unsigned int num_queues = vif->dev->real_num_tx_queues;
728 unsigned int queue_index;
Zoltan Kiss0e59a4a2014-03-24 23:59:50 +0000729 /* Here we want to avoid timeout messages if an skb can be legitimately
730 * stuck somewhere else. Realistically this could be an another vif's
Zoltan Kiss09350782014-03-06 21:48:30 +0000731 * internal or QDisc queue. That another vif also has this
732 * rx_drain_timeout_msecs timeout, but the timer only ditches the
733 * internal queue. After that, the QDisc queue can put in worst case
734 * XEN_NETIF_RX_RING_SIZE / MAX_SKB_FRAGS skbs into that another vif's
735 * internal queue, so we need several rounds of such timeouts until we
736 * can be sure that no another vif should have skb's from us. We are
Zoltan Kiss0e59a4a2014-03-24 23:59:50 +0000737 * not sending more skb's, so newly stuck packets are not interesting
Zoltan Kiss09350782014-03-06 21:48:30 +0000738 * for us here.
739 */
740 unsigned int worst_case_skb_lifetime = (rx_drain_timeout_msecs/1000) *
741 DIV_ROUND_UP(XENVIF_QUEUE_LENGTH, (XEN_NETIF_RX_RING_SIZE / MAX_SKB_FRAGS));
Zoltan Kissf53c3fe2014-03-06 21:48:26 +0000742
Wei Liue9ce7cb2014-06-04 10:30:42 +0100743 unregister_netdev(vif->dev);
744
745 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
746 queue = &vif->queues[queue_index];
Wei Liue9ce7cb2014-06-04 10:30:42 +0100747 xenvif_wait_unmap_timeout(queue, worst_case_skb_lifetime);
Andrew J. Bennieston8d3d53b2014-06-04 10:30:43 +0100748 xenvif_deinit_queue(queue);
Zoltan Kissf53c3fe2014-03-06 21:48:26 +0000749 }
750
Wei Liue9ce7cb2014-06-04 10:30:42 +0100751 /* Free the array of queues. The call below does not require
752 * rtnl_lock() because it happens after unregister_netdev().
753 */
754 netif_set_real_num_tx_queues(vif->dev, 0);
755 vfree(vif->queues);
756 vif->queues = NULL;
Ian Campbellf942dc22011-03-15 00:06:18 +0000757
Ian Campbellf942dc22011-03-15 00:06:18 +0000758 free_netdev(vif->dev);
Wei Liub103f352013-05-16 23:26:11 +0000759
Paul Durrant279f4382013-09-17 17:46:08 +0100760 module_put(THIS_MODULE);
Ian Campbellf942dc22011-03-15 00:06:18 +0000761}