Commit | Line | Data |
---|---|---|
0d160211 JF |
1 | /* |
2 | * Virtual network driver for conversing with remote driver backends. | |
3 | * | |
4 | * Copyright (c) 2002-2005, K A Fraser | |
5 | * Copyright (c) 2005, XenSource Ltd | |
6 | * | |
7 | * This program is free software; you can redistribute it and/or | |
8 | * modify it under the terms of the GNU General Public License version 2 | |
9 | * as published by the Free Software Foundation; or, when distributed | |
10 | * separately from the Linux kernel or incorporated into other | |
11 | * software packages, subject to the following license: | |
12 | * | |
13 | * Permission is hereby granted, free of charge, to any person obtaining a copy | |
14 | * of this source file (the "Software"), to deal in the Software without | |
15 | * restriction, including without limitation the rights to use, copy, modify, | |
16 | * merge, publish, distribute, sublicense, and/or sell copies of the Software, | |
17 | * and to permit persons to whom the Software is furnished to do so, subject to | |
18 | * the following conditions: | |
19 | * | |
20 | * The above copyright notice and this permission notice shall be included in | |
21 | * all copies or substantial portions of the Software. | |
22 | * | |
23 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | |
24 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | |
25 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE | |
26 | * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER | |
27 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING | |
28 | * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS | |
29 | * IN THE SOFTWARE. | |
30 | */ | |
31 | ||
383eda32 JP |
32 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
33 | ||
0d160211 JF |
34 | #include <linux/module.h> |
35 | #include <linux/kernel.h> | |
36 | #include <linux/netdevice.h> | |
37 | #include <linux/etherdevice.h> | |
38 | #include <linux/skbuff.h> | |
39 | #include <linux/ethtool.h> | |
40 | #include <linux/if_ether.h> | |
9ecd1a75 | 41 | #include <net/tcp.h> |
0d160211 JF |
42 | #include <linux/udp.h> |
43 | #include <linux/moduleparam.h> | |
44 | #include <linux/mm.h> | |
5a0e3ad6 | 45 | #include <linux/slab.h> |
0d160211 JF |
46 | #include <net/ip.h> |
47 | ||
1ccbf534 | 48 | #include <xen/xen.h> |
0d160211 JF |
49 | #include <xen/xenbus.h> |
50 | #include <xen/events.h> | |
51 | #include <xen/page.h> | |
b9136d20 | 52 | #include <xen/platform_pci.h> |
0d160211 JF |
53 | #include <xen/grant_table.h> |
54 | ||
55 | #include <xen/interface/io/netif.h> | |
56 | #include <xen/interface/memory.h> | |
57 | #include <xen/interface/grant_table.h> | |
58 | ||
50ee6061 | 59 | /* Module parameters */ |
034702a6 | 60 | #define MAX_QUEUES_DEFAULT 8 |
50ee6061 AB |
61 | static unsigned int xennet_max_queues; |
62 | module_param_named(max_queues, xennet_max_queues, uint, 0644); | |
63 | MODULE_PARM_DESC(max_queues, | |
64 | "Maximum number of queues per virtual interface"); | |
65 | ||
0fc0b732 | 66 | static const struct ethtool_ops xennet_ethtool_ops; |
0d160211 JF |
67 | |
68 | struct netfront_cb { | |
3683243b | 69 | int pull_to; |
0d160211 JF |
70 | }; |
71 | ||
72 | #define NETFRONT_SKB_CB(skb) ((struct netfront_cb *)((skb)->cb)) | |
73 | ||
74 | #define RX_COPY_THRESHOLD 256 | |
75 | ||
76 | #define GRANT_INVALID_REF 0 | |
77 | ||
30c5d7f0 JG |
78 | #define NET_TX_RING_SIZE __CONST_RING_SIZE(xen_netif_tx, XEN_PAGE_SIZE) |
79 | #define NET_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, XEN_PAGE_SIZE) | |
1f3c2eba DV |
80 | |
81 | /* Minimum number of Rx slots (includes slot for GSO metadata). */ | |
82 | #define NET_RX_SLOTS_MIN (XEN_NETIF_NR_SLOTS_MIN + 1) | |
0d160211 | 83 | |
2688fcb7 AB |
84 | /* Queue name is interface name with "-qNNN" appended */ |
85 | #define QUEUE_NAME_SIZE (IFNAMSIZ + 6) | |
86 | ||
87 | /* IRQ name is queue name with "-tx" or "-rx" appended */ | |
88 | #define IRQ_NAME_SIZE (QUEUE_NAME_SIZE + 3) | |
89 | ||
e00f85be | 90 | struct netfront_stats { |
900e1833 DV |
91 | u64 packets; |
92 | u64 bytes; | |
e00f85be | 93 | struct u64_stats_sync syncp; |
94 | }; | |
95 | ||
2688fcb7 AB |
96 | struct netfront_info; |
97 | ||
98 | struct netfront_queue { | |
99 | unsigned int id; /* Queue ID, 0-based */ | |
100 | char name[QUEUE_NAME_SIZE]; /* DEVNAME-qN */ | |
101 | struct netfront_info *info; | |
0d160211 | 102 | |
bea3348e | 103 | struct napi_struct napi; |
0d160211 | 104 | |
d634bf2c WL |
105 | /* Split event channels support, tx_* == rx_* when using |
106 | * single event channel. | |
107 | */ | |
108 | unsigned int tx_evtchn, rx_evtchn; | |
109 | unsigned int tx_irq, rx_irq; | |
110 | /* Only used when split event channels support is enabled */ | |
2688fcb7 AB |
111 | char tx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-tx */ |
112 | char rx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-rx */ | |
0d160211 | 113 | |
84284d3c JF |
114 | spinlock_t tx_lock; |
115 | struct xen_netif_tx_front_ring tx; | |
116 | int tx_ring_ref; | |
0d160211 JF |
117 | |
118 | /* | |
119 | * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries | |
120 | * are linked from tx_skb_freelist through skb_entry.link. | |
121 | * | |
122 | * NB. Freelist index entries are always going to be less than | |
123 | * PAGE_OFFSET, whereas pointers to skbs will always be equal or | |
124 | * greater than PAGE_OFFSET: we use this property to distinguish | |
125 | * them. | |
126 | */ | |
127 | union skb_entry { | |
128 | struct sk_buff *skb; | |
1ffb40b8 | 129 | unsigned long link; |
0d160211 JF |
130 | } tx_skbs[NET_TX_RING_SIZE]; |
131 | grant_ref_t gref_tx_head; | |
132 | grant_ref_t grant_tx_ref[NET_TX_RING_SIZE]; | |
cefe0078 | 133 | struct page *grant_tx_page[NET_TX_RING_SIZE]; |
0d160211 JF |
134 | unsigned tx_skb_freelist; |
135 | ||
84284d3c JF |
136 | spinlock_t rx_lock ____cacheline_aligned_in_smp; |
137 | struct xen_netif_rx_front_ring rx; | |
138 | int rx_ring_ref; | |
139 | ||
84284d3c JF |
140 | struct timer_list rx_refill_timer; |
141 | ||
0d160211 JF |
142 | struct sk_buff *rx_skbs[NET_RX_RING_SIZE]; |
143 | grant_ref_t gref_rx_head; | |
144 | grant_ref_t grant_rx_ref[NET_RX_RING_SIZE]; | |
2688fcb7 AB |
145 | }; |
146 | ||
147 | struct netfront_info { | |
148 | struct list_head list; | |
149 | struct net_device *netdev; | |
150 | ||
151 | struct xenbus_device *xbdev; | |
152 | ||
153 | /* Multi-queue support */ | |
154 | struct netfront_queue *queues; | |
e0ce4af9 IC |
155 | |
156 | /* Statistics */ | |
900e1833 DV |
157 | struct netfront_stats __percpu *rx_stats; |
158 | struct netfront_stats __percpu *tx_stats; | |
e00f85be | 159 | |
2688fcb7 | 160 | atomic_t rx_gso_checksum_fixup; |
0d160211 JF |
161 | }; |
162 | ||
163 | struct netfront_rx_info { | |
164 | struct xen_netif_rx_response rx; | |
165 | struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1]; | |
166 | }; | |
167 | ||
1ffb40b8 IY |
168 | static void skb_entry_set_link(union skb_entry *list, unsigned short id) |
169 | { | |
170 | list->link = id; | |
171 | } | |
172 | ||
173 | static int skb_entry_is_link(const union skb_entry *list) | |
174 | { | |
175 | BUILD_BUG_ON(sizeof(list->skb) != sizeof(list->link)); | |
807540ba | 176 | return (unsigned long)list->skb < PAGE_OFFSET; |
1ffb40b8 IY |
177 | } |
178 | ||
0d160211 JF |
179 | /* |
180 | * Access macros for acquiring freeing slots in tx_skbs[]. | |
181 | */ | |
182 | ||
183 | static void add_id_to_freelist(unsigned *head, union skb_entry *list, | |
184 | unsigned short id) | |
185 | { | |
1ffb40b8 | 186 | skb_entry_set_link(&list[id], *head); |
0d160211 JF |
187 | *head = id; |
188 | } | |
189 | ||
190 | static unsigned short get_id_from_freelist(unsigned *head, | |
191 | union skb_entry *list) | |
192 | { | |
193 | unsigned int id = *head; | |
194 | *head = list[id].link; | |
195 | return id; | |
196 | } | |
197 | ||
198 | static int xennet_rxidx(RING_IDX idx) | |
199 | { | |
200 | return idx & (NET_RX_RING_SIZE - 1); | |
201 | } | |
202 | ||
2688fcb7 | 203 | static struct sk_buff *xennet_get_rx_skb(struct netfront_queue *queue, |
0d160211 JF |
204 | RING_IDX ri) |
205 | { | |
206 | int i = xennet_rxidx(ri); | |
2688fcb7 AB |
207 | struct sk_buff *skb = queue->rx_skbs[i]; |
208 | queue->rx_skbs[i] = NULL; | |
0d160211 JF |
209 | return skb; |
210 | } | |
211 | ||
2688fcb7 | 212 | static grant_ref_t xennet_get_rx_ref(struct netfront_queue *queue, |
0d160211 JF |
213 | RING_IDX ri) |
214 | { | |
215 | int i = xennet_rxidx(ri); | |
2688fcb7 AB |
216 | grant_ref_t ref = queue->grant_rx_ref[i]; |
217 | queue->grant_rx_ref[i] = GRANT_INVALID_REF; | |
0d160211 JF |
218 | return ref; |
219 | } | |
220 | ||
221 | #ifdef CONFIG_SYSFS | |
27b917e5 | 222 | static const struct attribute_group xennet_dev_group; |
0d160211 JF |
223 | #endif |
224 | ||
3ad9b358 | 225 | static bool xennet_can_sg(struct net_device *dev) |
0d160211 | 226 | { |
3ad9b358 | 227 | return dev->features & NETIF_F_SG; |
0d160211 JF |
228 | } |
229 | ||
230 | ||
231 | static void rx_refill_timeout(unsigned long data) | |
232 | { | |
2688fcb7 AB |
233 | struct netfront_queue *queue = (struct netfront_queue *)data; |
234 | napi_schedule(&queue->napi); | |
0d160211 JF |
235 | } |
236 | ||
2688fcb7 | 237 | static int netfront_tx_slot_available(struct netfront_queue *queue) |
0d160211 | 238 | { |
2688fcb7 | 239 | return (queue->tx.req_prod_pvt - queue->tx.rsp_cons) < |
1f3c2eba | 240 | (NET_TX_RING_SIZE - MAX_SKB_FRAGS - 2); |
0d160211 JF |
241 | } |
242 | ||
2688fcb7 | 243 | static void xennet_maybe_wake_tx(struct netfront_queue *queue) |
0d160211 | 244 | { |
2688fcb7 AB |
245 | struct net_device *dev = queue->info->netdev; |
246 | struct netdev_queue *dev_queue = netdev_get_tx_queue(dev, queue->id); | |
0d160211 | 247 | |
2688fcb7 AB |
248 | if (unlikely(netif_tx_queue_stopped(dev_queue)) && |
249 | netfront_tx_slot_available(queue) && | |
0d160211 | 250 | likely(netif_running(dev))) |
2688fcb7 | 251 | netif_tx_wake_queue(netdev_get_tx_queue(dev, queue->id)); |
0d160211 JF |
252 | } |
253 | ||
1f3c2eba DV |
254 | |
255 | static struct sk_buff *xennet_alloc_one_rx_buffer(struct netfront_queue *queue) | |
0d160211 | 256 | { |
0d160211 JF |
257 | struct sk_buff *skb; |
258 | struct page *page; | |
0d160211 | 259 | |
1f3c2eba DV |
260 | skb = __netdev_alloc_skb(queue->info->netdev, |
261 | RX_COPY_THRESHOLD + NET_IP_ALIGN, | |
262 | GFP_ATOMIC | __GFP_NOWARN); | |
263 | if (unlikely(!skb)) | |
264 | return NULL; | |
0d160211 | 265 | |
1f3c2eba DV |
266 | page = alloc_page(GFP_ATOMIC | __GFP_NOWARN); |
267 | if (!page) { | |
268 | kfree_skb(skb); | |
269 | return NULL; | |
0d160211 | 270 | } |
1f3c2eba DV |
271 | skb_add_rx_frag(skb, 0, page, 0, 0, PAGE_SIZE); |
272 | ||
273 | /* Align ip header to a 16 bytes boundary */ | |
274 | skb_reserve(skb, NET_IP_ALIGN); | |
275 | skb->dev = queue->info->netdev; | |
276 | ||
277 | return skb; | |
278 | } | |
0d160211 | 279 | |
1f3c2eba DV |
280 | |
281 | static void xennet_alloc_rx_buffers(struct netfront_queue *queue) | |
282 | { | |
283 | RING_IDX req_prod = queue->rx.req_prod_pvt; | |
284 | int notify; | |
538d9291 | 285 | int err = 0; |
1f3c2eba DV |
286 | |
287 | if (unlikely(!netif_carrier_ok(queue->info->netdev))) | |
0d160211 | 288 | return; |
0d160211 | 289 | |
1f3c2eba DV |
290 | for (req_prod = queue->rx.req_prod_pvt; |
291 | req_prod - queue->rx.rsp_cons < NET_RX_RING_SIZE; | |
292 | req_prod++) { | |
293 | struct sk_buff *skb; | |
294 | unsigned short id; | |
295 | grant_ref_t ref; | |
30c5d7f0 | 296 | struct page *page; |
1f3c2eba | 297 | struct xen_netif_rx_request *req; |
0d160211 | 298 | |
1f3c2eba | 299 | skb = xennet_alloc_one_rx_buffer(queue); |
538d9291 VRP |
300 | if (!skb) { |
301 | err = -ENOMEM; | |
0d160211 | 302 | break; |
538d9291 | 303 | } |
0d160211 | 304 | |
1f3c2eba | 305 | id = xennet_rxidx(req_prod); |
0d160211 | 306 | |
2688fcb7 AB |
307 | BUG_ON(queue->rx_skbs[id]); |
308 | queue->rx_skbs[id] = skb; | |
0d160211 | 309 | |
2688fcb7 | 310 | ref = gnttab_claim_grant_reference(&queue->gref_rx_head); |
269ebce4 | 311 | WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref)); |
2688fcb7 | 312 | queue->grant_rx_ref[id] = ref; |
0d160211 | 313 | |
30c5d7f0 | 314 | page = skb_frag_page(&skb_shinfo(skb)->frags[0]); |
0d160211 | 315 | |
1f3c2eba | 316 | req = RING_GET_REQUEST(&queue->rx, req_prod); |
30c5d7f0 JG |
317 | gnttab_page_grant_foreign_access_ref_one(ref, |
318 | queue->info->xbdev->otherend_id, | |
319 | page, | |
320 | 0); | |
0d160211 JF |
321 | req->id = id; |
322 | req->gref = ref; | |
323 | } | |
324 | ||
1f3c2eba DV |
325 | queue->rx.req_prod_pvt = req_prod; |
326 | ||
538d9291 VRP |
327 | /* Try again later if there are not enough requests or skb allocation |
328 | * failed. | |
329 | * Enough requests is quantified as the sum of newly created slots and | |
330 | * the unconsumed slots at the backend. | |
331 | */ | |
332 | if (req_prod - queue->rx.rsp_cons < NET_RX_SLOTS_MIN || | |
333 | unlikely(err)) { | |
1f3c2eba DV |
334 | mod_timer(&queue->rx_refill_timer, jiffies + (HZ/10)); |
335 | return; | |
336 | } | |
337 | ||
5dcddfae | 338 | wmb(); /* barrier so backend seens requests */ |
0d160211 | 339 | |
2688fcb7 | 340 | RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->rx, notify); |
0d160211 | 341 | if (notify) |
2688fcb7 | 342 | notify_remote_via_irq(queue->rx_irq); |
0d160211 JF |
343 | } |
344 | ||
345 | static int xennet_open(struct net_device *dev) | |
346 | { | |
347 | struct netfront_info *np = netdev_priv(dev); | |
2688fcb7 AB |
348 | unsigned int num_queues = dev->real_num_tx_queues; |
349 | unsigned int i = 0; | |
350 | struct netfront_queue *queue = NULL; | |
351 | ||
352 | for (i = 0; i < num_queues; ++i) { | |
353 | queue = &np->queues[i]; | |
354 | napi_enable(&queue->napi); | |
355 | ||
356 | spin_lock_bh(&queue->rx_lock); | |
357 | if (netif_carrier_ok(dev)) { | |
358 | xennet_alloc_rx_buffers(queue); | |
359 | queue->rx.sring->rsp_event = queue->rx.rsp_cons + 1; | |
360 | if (RING_HAS_UNCONSUMED_RESPONSES(&queue->rx)) | |
361 | napi_schedule(&queue->napi); | |
362 | } | |
363 | spin_unlock_bh(&queue->rx_lock); | |
0d160211 | 364 | } |
0d160211 | 365 | |
2688fcb7 | 366 | netif_tx_start_all_queues(dev); |
0d160211 JF |
367 | |
368 | return 0; | |
369 | } | |
370 | ||
2688fcb7 | 371 | static void xennet_tx_buf_gc(struct netfront_queue *queue) |
0d160211 JF |
372 | { |
373 | RING_IDX cons, prod; | |
374 | unsigned short id; | |
0d160211 | 375 | struct sk_buff *skb; |
7d0105b5 | 376 | bool more_to_do; |
0d160211 | 377 | |
2688fcb7 | 378 | BUG_ON(!netif_carrier_ok(queue->info->netdev)); |
0d160211 JF |
379 | |
380 | do { | |
2688fcb7 | 381 | prod = queue->tx.sring->rsp_prod; |
0d160211 JF |
382 | rmb(); /* Ensure we see responses up to 'rp'. */ |
383 | ||
2688fcb7 | 384 | for (cons = queue->tx.rsp_cons; cons != prod; cons++) { |
0d160211 JF |
385 | struct xen_netif_tx_response *txrsp; |
386 | ||
2688fcb7 | 387 | txrsp = RING_GET_RESPONSE(&queue->tx, cons); |
f942dc25 | 388 | if (txrsp->status == XEN_NETIF_RSP_NULL) |
0d160211 JF |
389 | continue; |
390 | ||
391 | id = txrsp->id; | |
2688fcb7 | 392 | skb = queue->tx_skbs[id].skb; |
0d160211 | 393 | if (unlikely(gnttab_query_foreign_access( |
2688fcb7 | 394 | queue->grant_tx_ref[id]) != 0)) { |
383eda32 JP |
395 | pr_alert("%s: warning -- grant still in use by backend domain\n", |
396 | __func__); | |
0d160211 JF |
397 | BUG(); |
398 | } | |
399 | gnttab_end_foreign_access_ref( | |
2688fcb7 | 400 | queue->grant_tx_ref[id], GNTMAP_readonly); |
0d160211 | 401 | gnttab_release_grant_reference( |
2688fcb7 AB |
402 | &queue->gref_tx_head, queue->grant_tx_ref[id]); |
403 | queue->grant_tx_ref[id] = GRANT_INVALID_REF; | |
404 | queue->grant_tx_page[id] = NULL; | |
405 | add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, id); | |
0d160211 JF |
406 | dev_kfree_skb_irq(skb); |
407 | } | |
408 | ||
2688fcb7 | 409 | queue->tx.rsp_cons = prod; |
0d160211 | 410 | |
7d0105b5 MC |
411 | RING_FINAL_CHECK_FOR_RESPONSES(&queue->tx, more_to_do); |
412 | } while (more_to_do); | |
0d160211 | 413 | |
2688fcb7 | 414 | xennet_maybe_wake_tx(queue); |
0d160211 JF |
415 | } |
416 | ||
30c5d7f0 JG |
417 | struct xennet_gnttab_make_txreq { |
418 | struct netfront_queue *queue; | |
419 | struct sk_buff *skb; | |
420 | struct page *page; | |
421 | struct xen_netif_tx_request *tx; /* Last request */ | |
422 | unsigned int size; | |
423 | }; | |
424 | ||
425 | static void xennet_tx_setup_grant(unsigned long gfn, unsigned int offset, | |
426 | unsigned int len, void *data) | |
0d160211 | 427 | { |
30c5d7f0 | 428 | struct xennet_gnttab_make_txreq *info = data; |
0d160211 | 429 | unsigned int id; |
a55e8bb8 | 430 | struct xen_netif_tx_request *tx; |
0d160211 | 431 | grant_ref_t ref; |
30c5d7f0 JG |
432 | /* convenient aliases */ |
433 | struct page *page = info->page; | |
434 | struct netfront_queue *queue = info->queue; | |
435 | struct sk_buff *skb = info->skb; | |
0d160211 | 436 | |
a55e8bb8 DV |
437 | id = get_id_from_freelist(&queue->tx_skb_freelist, queue->tx_skbs); |
438 | tx = RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++); | |
439 | ref = gnttab_claim_grant_reference(&queue->gref_tx_head); | |
269ebce4 | 440 | WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref)); |
0d160211 | 441 | |
30c5d7f0 JG |
442 | gnttab_grant_foreign_access_ref(ref, queue->info->xbdev->otherend_id, |
443 | gfn, GNTMAP_readonly); | |
0d160211 | 444 | |
a55e8bb8 DV |
445 | queue->tx_skbs[id].skb = skb; |
446 | queue->grant_tx_page[id] = page; | |
447 | queue->grant_tx_ref[id] = ref; | |
0d160211 | 448 | |
a55e8bb8 DV |
449 | tx->id = id; |
450 | tx->gref = ref; | |
451 | tx->offset = offset; | |
452 | tx->size = len; | |
453 | tx->flags = 0; | |
0d160211 | 454 | |
30c5d7f0 JG |
455 | info->tx = tx; |
456 | info->size += tx->size; | |
457 | } | |
458 | ||
459 | static struct xen_netif_tx_request *xennet_make_first_txreq( | |
460 | struct netfront_queue *queue, struct sk_buff *skb, | |
461 | struct page *page, unsigned int offset, unsigned int len) | |
462 | { | |
463 | struct xennet_gnttab_make_txreq info = { | |
464 | .queue = queue, | |
465 | .skb = skb, | |
466 | .page = page, | |
467 | .size = 0, | |
468 | }; | |
469 | ||
470 | gnttab_for_one_grant(page, offset, len, xennet_tx_setup_grant, &info); | |
471 | ||
472 | return info.tx; | |
473 | } | |
474 | ||
475 | static void xennet_make_one_txreq(unsigned long gfn, unsigned int offset, | |
476 | unsigned int len, void *data) | |
477 | { | |
478 | struct xennet_gnttab_make_txreq *info = data; | |
479 | ||
480 | info->tx->flags |= XEN_NETTXF_more_data; | |
481 | skb_get(info->skb); | |
482 | xennet_tx_setup_grant(gfn, offset, len, data); | |
a55e8bb8 | 483 | } |
0d160211 | 484 | |
a55e8bb8 DV |
485 | static struct xen_netif_tx_request *xennet_make_txreqs( |
486 | struct netfront_queue *queue, struct xen_netif_tx_request *tx, | |
487 | struct sk_buff *skb, struct page *page, | |
488 | unsigned int offset, unsigned int len) | |
489 | { | |
30c5d7f0 JG |
490 | struct xennet_gnttab_make_txreq info = { |
491 | .queue = queue, | |
492 | .skb = skb, | |
493 | .tx = tx, | |
494 | }; | |
495 | ||
a55e8bb8 DV |
496 | /* Skip unused frames from start of page */ |
497 | page += offset >> PAGE_SHIFT; | |
498 | offset &= ~PAGE_MASK; | |
0d160211 | 499 | |
a55e8bb8 | 500 | while (len) { |
30c5d7f0 JG |
501 | info.page = page; |
502 | info.size = 0; | |
503 | ||
504 | gnttab_foreach_grant_in_range(page, offset, len, | |
505 | xennet_make_one_txreq, | |
506 | &info); | |
507 | ||
a55e8bb8 DV |
508 | page++; |
509 | offset = 0; | |
30c5d7f0 | 510 | len -= info.size; |
0d160211 JF |
511 | } |
512 | ||
30c5d7f0 | 513 | return info.tx; |
0d160211 JF |
514 | } |
515 | ||
f36c3747 | 516 | /* |
e84448d5 DV |
517 | * Count how many ring slots are required to send this skb. Each frag |
518 | * might be a compound page. | |
f36c3747 | 519 | */ |
e84448d5 | 520 | static int xennet_count_skb_slots(struct sk_buff *skb) |
f36c3747 IC |
521 | { |
522 | int i, frags = skb_shinfo(skb)->nr_frags; | |
30c5d7f0 | 523 | int slots; |
e84448d5 | 524 | |
30c5d7f0 JG |
525 | slots = gnttab_count_grant(offset_in_page(skb->data), |
526 | skb_headlen(skb)); | |
f36c3747 IC |
527 | |
528 | for (i = 0; i < frags; i++) { | |
529 | skb_frag_t *frag = skb_shinfo(skb)->frags + i; | |
530 | unsigned long size = skb_frag_size(frag); | |
531 | unsigned long offset = frag->page_offset; | |
532 | ||
533 | /* Skip unused frames from start of page */ | |
534 | offset &= ~PAGE_MASK; | |
535 | ||
30c5d7f0 | 536 | slots += gnttab_count_grant(offset, size); |
f36c3747 IC |
537 | } |
538 | ||
30c5d7f0 | 539 | return slots; |
f36c3747 IC |
540 | } |
541 | ||
50ee6061 AB |
542 | static u16 xennet_select_queue(struct net_device *dev, struct sk_buff *skb, |
543 | void *accel_priv, select_queue_fallback_t fallback) | |
2688fcb7 | 544 | { |
50ee6061 AB |
545 | unsigned int num_queues = dev->real_num_tx_queues; |
546 | u32 hash; | |
547 | u16 queue_idx; | |
548 | ||
549 | /* First, check if there is only one queue */ | |
550 | if (num_queues == 1) { | |
551 | queue_idx = 0; | |
552 | } else { | |
553 | hash = skb_get_hash(skb); | |
554 | queue_idx = hash % num_queues; | |
555 | } | |
556 | ||
557 | return queue_idx; | |
2688fcb7 AB |
558 | } |
559 | ||
30c5d7f0 JG |
560 | #define MAX_XEN_SKB_FRAGS (65536 / XEN_PAGE_SIZE + 1) |
561 | ||
0d160211 JF |
562 | static int xennet_start_xmit(struct sk_buff *skb, struct net_device *dev) |
563 | { | |
0d160211 | 564 | struct netfront_info *np = netdev_priv(dev); |
900e1833 | 565 | struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats); |
a55e8bb8 DV |
566 | struct xen_netif_tx_request *tx, *first_tx; |
567 | unsigned int i; | |
0d160211 | 568 | int notify; |
f36c3747 | 569 | int slots; |
a55e8bb8 DV |
570 | struct page *page; |
571 | unsigned int offset; | |
572 | unsigned int len; | |
cf66f9d4 | 573 | unsigned long flags; |
2688fcb7 AB |
574 | struct netfront_queue *queue = NULL; |
575 | unsigned int num_queues = dev->real_num_tx_queues; | |
576 | u16 queue_index; | |
fd07160b | 577 | struct sk_buff *nskb; |
2688fcb7 AB |
578 | |
579 | /* Drop the packet if no queues are set up */ | |
580 | if (num_queues < 1) | |
581 | goto drop; | |
582 | /* Determine which queue to transmit this SKB on */ | |
583 | queue_index = skb_get_queue_mapping(skb); | |
584 | queue = &np->queues[queue_index]; | |
0d160211 | 585 | |
9ecd1a75 WL |
586 | /* If skb->len is too big for wire format, drop skb and alert |
587 | * user about misconfiguration. | |
588 | */ | |
589 | if (unlikely(skb->len > XEN_NETIF_MAX_TX_SIZE)) { | |
590 | net_alert_ratelimited( | |
591 | "xennet: skb->len = %u, too big for wire format\n", | |
592 | skb->len); | |
593 | goto drop; | |
594 | } | |
595 | ||
e84448d5 | 596 | slots = xennet_count_skb_slots(skb); |
30c5d7f0 | 597 | if (unlikely(slots > MAX_XEN_SKB_FRAGS + 1)) { |
97a6d1bb ZK |
598 | net_dbg_ratelimited("xennet: skb rides the rocket: %d slots, %d bytes\n", |
599 | slots, skb->len); | |
600 | if (skb_linearize(skb)) | |
601 | goto drop; | |
0d160211 JF |
602 | } |
603 | ||
a55e8bb8 DV |
604 | page = virt_to_page(skb->data); |
605 | offset = offset_in_page(skb->data); | |
fd07160b VK |
606 | |
607 | /* The first req should be at least ETH_HLEN size or the packet will be | |
608 | * dropped by netback. | |
609 | */ | |
610 | if (unlikely(PAGE_SIZE - offset < ETH_HLEN)) { | |
611 | nskb = skb_copy(skb, GFP_ATOMIC); | |
612 | if (!nskb) | |
613 | goto drop; | |
614 | dev_kfree_skb_any(skb); | |
615 | skb = nskb; | |
616 | page = virt_to_page(skb->data); | |
617 | offset = offset_in_page(skb->data); | |
618 | } | |
619 | ||
a55e8bb8 DV |
620 | len = skb_headlen(skb); |
621 | ||
2688fcb7 | 622 | spin_lock_irqsave(&queue->tx_lock, flags); |
0d160211 JF |
623 | |
624 | if (unlikely(!netif_carrier_ok(dev) || | |
f36c3747 | 625 | (slots > 1 && !xennet_can_sg(dev)) || |
8b86a61d | 626 | netif_needs_gso(skb, netif_skb_features(skb)))) { |
2688fcb7 | 627 | spin_unlock_irqrestore(&queue->tx_lock, flags); |
0d160211 JF |
628 | goto drop; |
629 | } | |
630 | ||
a55e8bb8 | 631 | /* First request for the linear area. */ |
30c5d7f0 JG |
632 | first_tx = tx = xennet_make_first_txreq(queue, skb, |
633 | page, offset, len); | |
634 | offset += tx->size; | |
635 | if (offset == PAGE_SIZE) { | |
636 | page++; | |
637 | offset = 0; | |
638 | } | |
a55e8bb8 | 639 | len -= tx->size; |
0d160211 | 640 | |
0d160211 JF |
641 | if (skb->ip_summed == CHECKSUM_PARTIAL) |
642 | /* local packet? */ | |
f942dc25 | 643 | tx->flags |= XEN_NETTXF_csum_blank | XEN_NETTXF_data_validated; |
0d160211 JF |
644 | else if (skb->ip_summed == CHECKSUM_UNNECESSARY) |
645 | /* remote but checksummed. */ | |
f942dc25 | 646 | tx->flags |= XEN_NETTXF_data_validated; |
0d160211 | 647 | |
a55e8bb8 | 648 | /* Optional extra info after the first request. */ |
0d160211 JF |
649 | if (skb_shinfo(skb)->gso_size) { |
650 | struct xen_netif_extra_info *gso; | |
651 | ||
652 | gso = (struct xen_netif_extra_info *) | |
a55e8bb8 | 653 | RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++); |
0d160211 | 654 | |
e2d617c0 | 655 | tx->flags |= XEN_NETTXF_extra_info; |
0d160211 JF |
656 | |
657 | gso->u.gso.size = skb_shinfo(skb)->gso_size; | |
2c0057de PD |
658 | gso->u.gso.type = (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) ? |
659 | XEN_NETIF_GSO_TYPE_TCPV6 : | |
660 | XEN_NETIF_GSO_TYPE_TCPV4; | |
0d160211 JF |
661 | gso->u.gso.pad = 0; |
662 | gso->u.gso.features = 0; | |
663 | ||
664 | gso->type = XEN_NETIF_EXTRA_TYPE_GSO; | |
665 | gso->flags = 0; | |
0d160211 JF |
666 | } |
667 | ||
a55e8bb8 DV |
668 | /* Requests for the rest of the linear area. */ |
669 | tx = xennet_make_txreqs(queue, tx, skb, page, offset, len); | |
670 | ||
671 | /* Requests for all the frags. */ | |
672 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
673 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
674 | tx = xennet_make_txreqs(queue, tx, skb, | |
675 | skb_frag_page(frag), frag->page_offset, | |
676 | skb_frag_size(frag)); | |
677 | } | |
0d160211 | 678 | |
a55e8bb8 DV |
679 | /* First request has the packet length. */ |
680 | first_tx->size = skb->len; | |
0d160211 | 681 | |
2688fcb7 | 682 | RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify); |
0d160211 | 683 | if (notify) |
2688fcb7 | 684 | notify_remote_via_irq(queue->tx_irq); |
0d160211 | 685 | |
900e1833 DV |
686 | u64_stats_update_begin(&tx_stats->syncp); |
687 | tx_stats->bytes += skb->len; | |
688 | tx_stats->packets++; | |
689 | u64_stats_update_end(&tx_stats->syncp); | |
10a273a6 JF |
690 | |
691 | /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */ | |
2688fcb7 | 692 | xennet_tx_buf_gc(queue); |
0d160211 | 693 | |
2688fcb7 AB |
694 | if (!netfront_tx_slot_available(queue)) |
695 | netif_tx_stop_queue(netdev_get_tx_queue(dev, queue->id)); | |
0d160211 | 696 | |
2688fcb7 | 697 | spin_unlock_irqrestore(&queue->tx_lock, flags); |
0d160211 | 698 | |
6ed10654 | 699 | return NETDEV_TX_OK; |
0d160211 JF |
700 | |
701 | drop: | |
09f75cd7 | 702 | dev->stats.tx_dropped++; |
979de8a0 | 703 | dev_kfree_skb_any(skb); |
6ed10654 | 704 | return NETDEV_TX_OK; |
0d160211 JF |
705 | } |
706 | ||
707 | static int xennet_close(struct net_device *dev) | |
708 | { | |
709 | struct netfront_info *np = netdev_priv(dev); | |
2688fcb7 AB |
710 | unsigned int num_queues = dev->real_num_tx_queues; |
711 | unsigned int i; | |
712 | struct netfront_queue *queue; | |
713 | netif_tx_stop_all_queues(np->netdev); | |
714 | for (i = 0; i < num_queues; ++i) { | |
715 | queue = &np->queues[i]; | |
716 | napi_disable(&queue->napi); | |
717 | } | |
0d160211 JF |
718 | return 0; |
719 | } | |
720 | ||
2688fcb7 | 721 | static void xennet_move_rx_slot(struct netfront_queue *queue, struct sk_buff *skb, |
0d160211 JF |
722 | grant_ref_t ref) |
723 | { | |
2688fcb7 AB |
724 | int new = xennet_rxidx(queue->rx.req_prod_pvt); |
725 | ||
726 | BUG_ON(queue->rx_skbs[new]); | |
727 | queue->rx_skbs[new] = skb; | |
728 | queue->grant_rx_ref[new] = ref; | |
729 | RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->id = new; | |
730 | RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->gref = ref; | |
731 | queue->rx.req_prod_pvt++; | |
0d160211 JF |
732 | } |
733 | ||
2688fcb7 | 734 | static int xennet_get_extras(struct netfront_queue *queue, |
0d160211 JF |
735 | struct xen_netif_extra_info *extras, |
736 | RING_IDX rp) | |
737 | ||
738 | { | |
739 | struct xen_netif_extra_info *extra; | |
2688fcb7 AB |
740 | struct device *dev = &queue->info->netdev->dev; |
741 | RING_IDX cons = queue->rx.rsp_cons; | |
0d160211 JF |
742 | int err = 0; |
743 | ||
744 | do { | |
745 | struct sk_buff *skb; | |
746 | grant_ref_t ref; | |
747 | ||
748 | if (unlikely(cons + 1 == rp)) { | |
749 | if (net_ratelimit()) | |
750 | dev_warn(dev, "Missing extra info\n"); | |
751 | err = -EBADR; | |
752 | break; | |
753 | } | |
754 | ||
755 | extra = (struct xen_netif_extra_info *) | |
2688fcb7 | 756 | RING_GET_RESPONSE(&queue->rx, ++cons); |
0d160211 JF |
757 | |
758 | if (unlikely(!extra->type || | |
759 | extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) { | |
760 | if (net_ratelimit()) | |
761 | dev_warn(dev, "Invalid extra type: %d\n", | |
762 | extra->type); | |
763 | err = -EINVAL; | |
764 | } else { | |
765 | memcpy(&extras[extra->type - 1], extra, | |
766 | sizeof(*extra)); | |
767 | } | |
768 | ||
2688fcb7 AB |
769 | skb = xennet_get_rx_skb(queue, cons); |
770 | ref = xennet_get_rx_ref(queue, cons); | |
771 | xennet_move_rx_slot(queue, skb, ref); | |
0d160211 JF |
772 | } while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE); |
773 | ||
2688fcb7 | 774 | queue->rx.rsp_cons = cons; |
0d160211 JF |
775 | return err; |
776 | } | |
777 | ||
2688fcb7 | 778 | static int xennet_get_responses(struct netfront_queue *queue, |
0d160211 JF |
779 | struct netfront_rx_info *rinfo, RING_IDX rp, |
780 | struct sk_buff_head *list) | |
781 | { | |
782 | struct xen_netif_rx_response *rx = &rinfo->rx; | |
783 | struct xen_netif_extra_info *extras = rinfo->extras; | |
2688fcb7 AB |
784 | struct device *dev = &queue->info->netdev->dev; |
785 | RING_IDX cons = queue->rx.rsp_cons; | |
786 | struct sk_buff *skb = xennet_get_rx_skb(queue, cons); | |
787 | grant_ref_t ref = xennet_get_rx_ref(queue, cons); | |
0d160211 | 788 | int max = MAX_SKB_FRAGS + (rx->status <= RX_COPY_THRESHOLD); |
7158ff6d | 789 | int slots = 1; |
0d160211 JF |
790 | int err = 0; |
791 | unsigned long ret; | |
792 | ||
f942dc25 | 793 | if (rx->flags & XEN_NETRXF_extra_info) { |
2688fcb7 AB |
794 | err = xennet_get_extras(queue, extras, rp); |
795 | cons = queue->rx.rsp_cons; | |
0d160211 JF |
796 | } |
797 | ||
798 | for (;;) { | |
799 | if (unlikely(rx->status < 0 || | |
30c5d7f0 | 800 | rx->offset + rx->status > XEN_PAGE_SIZE)) { |
0d160211 | 801 | if (net_ratelimit()) |
6c10127d | 802 | dev_warn(dev, "rx->offset: %u, size: %d\n", |
0d160211 | 803 | rx->offset, rx->status); |
2688fcb7 | 804 | xennet_move_rx_slot(queue, skb, ref); |
0d160211 JF |
805 | err = -EINVAL; |
806 | goto next; | |
807 | } | |
808 | ||
809 | /* | |
810 | * This definitely indicates a bug, either in this driver or in | |
811 | * the backend driver. In future this should flag the bad | |
697089dc | 812 | * situation to the system controller to reboot the backend. |
0d160211 JF |
813 | */ |
814 | if (ref == GRANT_INVALID_REF) { | |
815 | if (net_ratelimit()) | |
816 | dev_warn(dev, "Bad rx response id %d.\n", | |
817 | rx->id); | |
818 | err = -EINVAL; | |
819 | goto next; | |
820 | } | |
821 | ||
822 | ret = gnttab_end_foreign_access_ref(ref, 0); | |
823 | BUG_ON(!ret); | |
824 | ||
2688fcb7 | 825 | gnttab_release_grant_reference(&queue->gref_rx_head, ref); |
0d160211 JF |
826 | |
827 | __skb_queue_tail(list, skb); | |
828 | ||
829 | next: | |
f942dc25 | 830 | if (!(rx->flags & XEN_NETRXF_more_data)) |
0d160211 JF |
831 | break; |
832 | ||
7158ff6d | 833 | if (cons + slots == rp) { |
0d160211 | 834 | if (net_ratelimit()) |
7158ff6d | 835 | dev_warn(dev, "Need more slots\n"); |
0d160211 JF |
836 | err = -ENOENT; |
837 | break; | |
838 | } | |
839 | ||
2688fcb7 AB |
840 | rx = RING_GET_RESPONSE(&queue->rx, cons + slots); |
841 | skb = xennet_get_rx_skb(queue, cons + slots); | |
842 | ref = xennet_get_rx_ref(queue, cons + slots); | |
7158ff6d | 843 | slots++; |
0d160211 JF |
844 | } |
845 | ||
7158ff6d | 846 | if (unlikely(slots > max)) { |
0d160211 | 847 | if (net_ratelimit()) |
697089dc | 848 | dev_warn(dev, "Too many slots\n"); |
0d160211 JF |
849 | err = -E2BIG; |
850 | } | |
851 | ||
852 | if (unlikely(err)) | |
2688fcb7 | 853 | queue->rx.rsp_cons = cons + slots; |
0d160211 JF |
854 | |
855 | return err; | |
856 | } | |
857 | ||
858 | static int xennet_set_skb_gso(struct sk_buff *skb, | |
859 | struct xen_netif_extra_info *gso) | |
860 | { | |
861 | if (!gso->u.gso.size) { | |
862 | if (net_ratelimit()) | |
383eda32 | 863 | pr_warn("GSO size must not be zero\n"); |
0d160211 JF |
864 | return -EINVAL; |
865 | } | |
866 | ||
2c0057de PD |
867 | if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4 && |
868 | gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV6) { | |
0d160211 | 869 | if (net_ratelimit()) |
383eda32 | 870 | pr_warn("Bad GSO type %d\n", gso->u.gso.type); |
0d160211 JF |
871 | return -EINVAL; |
872 | } | |
873 | ||
874 | skb_shinfo(skb)->gso_size = gso->u.gso.size; | |
2c0057de PD |
875 | skb_shinfo(skb)->gso_type = |
876 | (gso->u.gso.type == XEN_NETIF_GSO_TYPE_TCPV4) ? | |
877 | SKB_GSO_TCPV4 : | |
878 | SKB_GSO_TCPV6; | |
0d160211 JF |
879 | |
880 | /* Header must be checked, and gso_segs computed. */ | |
881 | skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; | |
882 | skb_shinfo(skb)->gso_segs = 0; | |
883 | ||
884 | return 0; | |
885 | } | |
886 | ||
2688fcb7 | 887 | static RING_IDX xennet_fill_frags(struct netfront_queue *queue, |
0d160211 JF |
888 | struct sk_buff *skb, |
889 | struct sk_buff_head *list) | |
890 | { | |
891 | struct skb_shared_info *shinfo = skb_shinfo(skb); | |
2688fcb7 | 892 | RING_IDX cons = queue->rx.rsp_cons; |
0d160211 JF |
893 | struct sk_buff *nskb; |
894 | ||
895 | while ((nskb = __skb_dequeue(list))) { | |
896 | struct xen_netif_rx_response *rx = | |
2688fcb7 | 897 | RING_GET_RESPONSE(&queue->rx, ++cons); |
01c68026 | 898 | skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0]; |
0d160211 | 899 | |
093b9c71 JB |
900 | if (shinfo->nr_frags == MAX_SKB_FRAGS) { |
901 | unsigned int pull_to = NETFRONT_SKB_CB(skb)->pull_to; | |
0d160211 | 902 | |
093b9c71 JB |
903 | BUG_ON(pull_to <= skb_headlen(skb)); |
904 | __pskb_pull_tail(skb, pull_to - skb_headlen(skb)); | |
905 | } | |
906 | BUG_ON(shinfo->nr_frags >= MAX_SKB_FRAGS); | |
907 | ||
908 | skb_add_rx_frag(skb, shinfo->nr_frags, skb_frag_page(nfrag), | |
909 | rx->offset, rx->status, PAGE_SIZE); | |
0d160211 JF |
910 | |
911 | skb_shinfo(nskb)->nr_frags = 0; | |
912 | kfree_skb(nskb); | |
0d160211 JF |
913 | } |
914 | ||
0d160211 JF |
915 | return cons; |
916 | } | |
917 | ||
e0ce4af9 | 918 | static int checksum_setup(struct net_device *dev, struct sk_buff *skb) |
0d160211 | 919 | { |
b5cf66cd | 920 | bool recalculate_partial_csum = false; |
e0ce4af9 IC |
921 | |
922 | /* | |
923 | * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy | |
924 | * peers can fail to set NETRXF_csum_blank when sending a GSO | |
925 | * frame. In this case force the SKB to CHECKSUM_PARTIAL and | |
926 | * recalculate the partial checksum. | |
927 | */ | |
928 | if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) { | |
929 | struct netfront_info *np = netdev_priv(dev); | |
2688fcb7 | 930 | atomic_inc(&np->rx_gso_checksum_fixup); |
e0ce4af9 | 931 | skb->ip_summed = CHECKSUM_PARTIAL; |
b5cf66cd | 932 | recalculate_partial_csum = true; |
e0ce4af9 IC |
933 | } |
934 | ||
935 | /* A non-CHECKSUM_PARTIAL SKB does not require setup. */ | |
936 | if (skb->ip_summed != CHECKSUM_PARTIAL) | |
937 | return 0; | |
0d160211 | 938 | |
b5cf66cd | 939 | return skb_checksum_setup(skb, recalculate_partial_csum); |
0d160211 JF |
940 | } |
941 | ||
2688fcb7 | 942 | static int handle_incoming_queue(struct netfront_queue *queue, |
09f75cd7 | 943 | struct sk_buff_head *rxq) |
0d160211 | 944 | { |
900e1833 | 945 | struct netfront_stats *rx_stats = this_cpu_ptr(queue->info->rx_stats); |
0d160211 JF |
946 | int packets_dropped = 0; |
947 | struct sk_buff *skb; | |
948 | ||
949 | while ((skb = __skb_dequeue(rxq)) != NULL) { | |
3683243b | 950 | int pull_to = NETFRONT_SKB_CB(skb)->pull_to; |
0d160211 | 951 | |
093b9c71 JB |
952 | if (pull_to > skb_headlen(skb)) |
953 | __pskb_pull_tail(skb, pull_to - skb_headlen(skb)); | |
0d160211 JF |
954 | |
955 | /* Ethernet work: Delayed to here as it peeks the header. */ | |
2688fcb7 | 956 | skb->protocol = eth_type_trans(skb, queue->info->netdev); |
d554f73d | 957 | skb_reset_network_header(skb); |
0d160211 | 958 | |
2688fcb7 | 959 | if (checksum_setup(queue->info->netdev, skb)) { |
e0ce4af9 IC |
960 | kfree_skb(skb); |
961 | packets_dropped++; | |
2688fcb7 | 962 | queue->info->netdev->stats.rx_errors++; |
e0ce4af9 | 963 | continue; |
0d160211 JF |
964 | } |
965 | ||
900e1833 DV |
966 | u64_stats_update_begin(&rx_stats->syncp); |
967 | rx_stats->packets++; | |
968 | rx_stats->bytes += skb->len; | |
969 | u64_stats_update_end(&rx_stats->syncp); | |
0d160211 JF |
970 | |
971 | /* Pass it up. */ | |
2688fcb7 | 972 | napi_gro_receive(&queue->napi, skb); |
0d160211 JF |
973 | } |
974 | ||
975 | return packets_dropped; | |
976 | } | |
977 | ||
bea3348e | 978 | static int xennet_poll(struct napi_struct *napi, int budget) |
0d160211 | 979 | { |
2688fcb7 AB |
980 | struct netfront_queue *queue = container_of(napi, struct netfront_queue, napi); |
981 | struct net_device *dev = queue->info->netdev; | |
0d160211 JF |
982 | struct sk_buff *skb; |
983 | struct netfront_rx_info rinfo; | |
984 | struct xen_netif_rx_response *rx = &rinfo.rx; | |
985 | struct xen_netif_extra_info *extras = rinfo.extras; | |
986 | RING_IDX i, rp; | |
bea3348e | 987 | int work_done; |
0d160211 JF |
988 | struct sk_buff_head rxq; |
989 | struct sk_buff_head errq; | |
990 | struct sk_buff_head tmpq; | |
0d160211 JF |
991 | int err; |
992 | ||
2688fcb7 | 993 | spin_lock(&queue->rx_lock); |
0d160211 | 994 | |
0d160211 JF |
995 | skb_queue_head_init(&rxq); |
996 | skb_queue_head_init(&errq); | |
997 | skb_queue_head_init(&tmpq); | |
998 | ||
2688fcb7 | 999 | rp = queue->rx.sring->rsp_prod; |
0d160211 JF |
1000 | rmb(); /* Ensure we see queued responses up to 'rp'. */ |
1001 | ||
2688fcb7 | 1002 | i = queue->rx.rsp_cons; |
0d160211 JF |
1003 | work_done = 0; |
1004 | while ((i != rp) && (work_done < budget)) { | |
2688fcb7 | 1005 | memcpy(rx, RING_GET_RESPONSE(&queue->rx, i), sizeof(*rx)); |
0d160211 JF |
1006 | memset(extras, 0, sizeof(rinfo.extras)); |
1007 | ||
2688fcb7 | 1008 | err = xennet_get_responses(queue, &rinfo, rp, &tmpq); |
0d160211 JF |
1009 | |
1010 | if (unlikely(err)) { | |
1011 | err: | |
1012 | while ((skb = __skb_dequeue(&tmpq))) | |
1013 | __skb_queue_tail(&errq, skb); | |
09f75cd7 | 1014 | dev->stats.rx_errors++; |
2688fcb7 | 1015 | i = queue->rx.rsp_cons; |
0d160211 JF |
1016 | continue; |
1017 | } | |
1018 | ||
1019 | skb = __skb_dequeue(&tmpq); | |
1020 | ||
1021 | if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) { | |
1022 | struct xen_netif_extra_info *gso; | |
1023 | gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1]; | |
1024 | ||
1025 | if (unlikely(xennet_set_skb_gso(skb, gso))) { | |
1026 | __skb_queue_head(&tmpq, skb); | |
2688fcb7 | 1027 | queue->rx.rsp_cons += skb_queue_len(&tmpq); |
0d160211 JF |
1028 | goto err; |
1029 | } | |
1030 | } | |
1031 | ||
3683243b IC |
1032 | NETFRONT_SKB_CB(skb)->pull_to = rx->status; |
1033 | if (NETFRONT_SKB_CB(skb)->pull_to > RX_COPY_THRESHOLD) | |
1034 | NETFRONT_SKB_CB(skb)->pull_to = RX_COPY_THRESHOLD; | |
0d160211 | 1035 | |
3683243b IC |
1036 | skb_shinfo(skb)->frags[0].page_offset = rx->offset; |
1037 | skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status); | |
1038 | skb->data_len = rx->status; | |
093b9c71 | 1039 | skb->len += rx->status; |
0d160211 | 1040 | |
2688fcb7 | 1041 | i = xennet_fill_frags(queue, skb, &tmpq); |
0d160211 | 1042 | |
f942dc25 | 1043 | if (rx->flags & XEN_NETRXF_csum_blank) |
0d160211 | 1044 | skb->ip_summed = CHECKSUM_PARTIAL; |
f942dc25 | 1045 | else if (rx->flags & XEN_NETRXF_data_validated) |
0d160211 JF |
1046 | skb->ip_summed = CHECKSUM_UNNECESSARY; |
1047 | ||
1048 | __skb_queue_tail(&rxq, skb); | |
1049 | ||
2688fcb7 | 1050 | queue->rx.rsp_cons = ++i; |
0d160211 JF |
1051 | work_done++; |
1052 | } | |
1053 | ||
56cfe5d0 | 1054 | __skb_queue_purge(&errq); |
0d160211 | 1055 | |
2688fcb7 | 1056 | work_done -= handle_incoming_queue(queue, &rxq); |
0d160211 | 1057 | |
2688fcb7 | 1058 | xennet_alloc_rx_buffers(queue); |
0d160211 | 1059 | |
0d160211 | 1060 | if (work_done < budget) { |
bea3348e SH |
1061 | int more_to_do = 0; |
1062 | ||
6ad20165 | 1063 | napi_complete_done(napi, work_done); |
0d160211 | 1064 | |
2688fcb7 | 1065 | RING_FINAL_CHECK_FOR_RESPONSES(&queue->rx, more_to_do); |
6a6dc08f DV |
1066 | if (more_to_do) |
1067 | napi_schedule(napi); | |
0d160211 JF |
1068 | } |
1069 | ||
2688fcb7 | 1070 | spin_unlock(&queue->rx_lock); |
0d160211 | 1071 | |
bea3348e | 1072 | return work_done; |
0d160211 JF |
1073 | } |
1074 | ||
1075 | static int xennet_change_mtu(struct net_device *dev, int mtu) | |
1076 | { | |
0c36820e | 1077 | int max = xennet_can_sg(dev) ? XEN_NETIF_MAX_TX_SIZE : ETH_DATA_LEN; |
0d160211 JF |
1078 | |
1079 | if (mtu > max) | |
1080 | return -EINVAL; | |
1081 | dev->mtu = mtu; | |
1082 | return 0; | |
1083 | } | |
1084 | ||
bc1f4470 | 1085 | static void xennet_get_stats64(struct net_device *dev, |
1086 | struct rtnl_link_stats64 *tot) | |
e00f85be | 1087 | { |
1088 | struct netfront_info *np = netdev_priv(dev); | |
1089 | int cpu; | |
1090 | ||
1091 | for_each_possible_cpu(cpu) { | |
900e1833 DV |
1092 | struct netfront_stats *rx_stats = per_cpu_ptr(np->rx_stats, cpu); |
1093 | struct netfront_stats *tx_stats = per_cpu_ptr(np->tx_stats, cpu); | |
e00f85be | 1094 | u64 rx_packets, rx_bytes, tx_packets, tx_bytes; |
1095 | unsigned int start; | |
1096 | ||
1097 | do { | |
900e1833 DV |
1098 | start = u64_stats_fetch_begin_irq(&tx_stats->syncp); |
1099 | tx_packets = tx_stats->packets; | |
1100 | tx_bytes = tx_stats->bytes; | |
1101 | } while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start)); | |
e00f85be | 1102 | |
900e1833 DV |
1103 | do { |
1104 | start = u64_stats_fetch_begin_irq(&rx_stats->syncp); | |
1105 | rx_packets = rx_stats->packets; | |
1106 | rx_bytes = rx_stats->bytes; | |
1107 | } while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start)); | |
e00f85be | 1108 | |
1109 | tot->rx_packets += rx_packets; | |
1110 | tot->tx_packets += tx_packets; | |
1111 | tot->rx_bytes += rx_bytes; | |
1112 | tot->tx_bytes += tx_bytes; | |
1113 | } | |
1114 | ||
1115 | tot->rx_errors = dev->stats.rx_errors; | |
1116 | tot->tx_dropped = dev->stats.tx_dropped; | |
e00f85be | 1117 | } |
1118 | ||
2688fcb7 | 1119 | static void xennet_release_tx_bufs(struct netfront_queue *queue) |
0d160211 JF |
1120 | { |
1121 | struct sk_buff *skb; | |
1122 | int i; | |
1123 | ||
1124 | for (i = 0; i < NET_TX_RING_SIZE; i++) { | |
1125 | /* Skip over entries which are actually freelist references */ | |
2688fcb7 | 1126 | if (skb_entry_is_link(&queue->tx_skbs[i])) |
0d160211 JF |
1127 | continue; |
1128 | ||
2688fcb7 AB |
1129 | skb = queue->tx_skbs[i].skb; |
1130 | get_page(queue->grant_tx_page[i]); | |
1131 | gnttab_end_foreign_access(queue->grant_tx_ref[i], | |
cefe0078 | 1132 | GNTMAP_readonly, |
2688fcb7 AB |
1133 | (unsigned long)page_address(queue->grant_tx_page[i])); |
1134 | queue->grant_tx_page[i] = NULL; | |
1135 | queue->grant_tx_ref[i] = GRANT_INVALID_REF; | |
1136 | add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, i); | |
0d160211 JF |
1137 | dev_kfree_skb_irq(skb); |
1138 | } | |
1139 | } | |
1140 | ||
2688fcb7 | 1141 | static void xennet_release_rx_bufs(struct netfront_queue *queue) |
0d160211 | 1142 | { |
0d160211 JF |
1143 | int id, ref; |
1144 | ||
2688fcb7 | 1145 | spin_lock_bh(&queue->rx_lock); |
0d160211 JF |
1146 | |
1147 | for (id = 0; id < NET_RX_RING_SIZE; id++) { | |
cefe0078 AL |
1148 | struct sk_buff *skb; |
1149 | struct page *page; | |
0d160211 | 1150 | |
2688fcb7 | 1151 | skb = queue->rx_skbs[id]; |
cefe0078 | 1152 | if (!skb) |
0d160211 | 1153 | continue; |
0d160211 | 1154 | |
2688fcb7 | 1155 | ref = queue->grant_rx_ref[id]; |
cefe0078 AL |
1156 | if (ref == GRANT_INVALID_REF) |
1157 | continue; | |
0d160211 | 1158 | |
cefe0078 | 1159 | page = skb_frag_page(&skb_shinfo(skb)->frags[0]); |
0d160211 | 1160 | |
cefe0078 AL |
1161 | /* gnttab_end_foreign_access() needs a page ref until |
1162 | * foreign access is ended (which may be deferred). | |
1163 | */ | |
1164 | get_page(page); | |
1165 | gnttab_end_foreign_access(ref, 0, | |
1166 | (unsigned long)page_address(page)); | |
2688fcb7 | 1167 | queue->grant_rx_ref[id] = GRANT_INVALID_REF; |
0d160211 | 1168 | |
cefe0078 | 1169 | kfree_skb(skb); |
0d160211 JF |
1170 | } |
1171 | ||
2688fcb7 | 1172 | spin_unlock_bh(&queue->rx_lock); |
0d160211 JF |
1173 | } |
1174 | ||
c8f44aff MM |
1175 | static netdev_features_t xennet_fix_features(struct net_device *dev, |
1176 | netdev_features_t features) | |
8f7b01a1 ED |
1177 | { |
1178 | struct netfront_info *np = netdev_priv(dev); | |
8f7b01a1 | 1179 | |
2890ea5c JG |
1180 | if (features & NETIF_F_SG && |
1181 | !xenbus_read_unsigned(np->xbdev->otherend, "feature-sg", 0)) | |
1182 | features &= ~NETIF_F_SG; | |
8f7b01a1 | 1183 | |
2890ea5c JG |
1184 | if (features & NETIF_F_IPV6_CSUM && |
1185 | !xenbus_read_unsigned(np->xbdev->otherend, | |
1186 | "feature-ipv6-csum-offload", 0)) | |
1187 | features &= ~NETIF_F_IPV6_CSUM; | |
8f7b01a1 | 1188 | |
2890ea5c JG |
1189 | if (features & NETIF_F_TSO && |
1190 | !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv4", 0)) | |
1191 | features &= ~NETIF_F_TSO; | |
8f7b01a1 | 1192 | |
2890ea5c JG |
1193 | if (features & NETIF_F_TSO6 && |
1194 | !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv6", 0)) | |
1195 | features &= ~NETIF_F_TSO6; | |
2c0057de | 1196 | |
8f7b01a1 ED |
1197 | return features; |
1198 | } | |
1199 | ||
c8f44aff MM |
1200 | static int xennet_set_features(struct net_device *dev, |
1201 | netdev_features_t features) | |
8f7b01a1 ED |
1202 | { |
1203 | if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) { | |
1204 | netdev_info(dev, "Reducing MTU because no SG offload"); | |
1205 | dev->mtu = ETH_DATA_LEN; | |
1206 | } | |
1207 | ||
1208 | return 0; | |
1209 | } | |
1210 | ||
d634bf2c | 1211 | static irqreturn_t xennet_tx_interrupt(int irq, void *dev_id) |
cf66f9d4 | 1212 | { |
2688fcb7 | 1213 | struct netfront_queue *queue = dev_id; |
cf66f9d4 KRW |
1214 | unsigned long flags; |
1215 | ||
2688fcb7 AB |
1216 | spin_lock_irqsave(&queue->tx_lock, flags); |
1217 | xennet_tx_buf_gc(queue); | |
1218 | spin_unlock_irqrestore(&queue->tx_lock, flags); | |
cf66f9d4 | 1219 | |
d634bf2c WL |
1220 | return IRQ_HANDLED; |
1221 | } | |
1222 | ||
1223 | static irqreturn_t xennet_rx_interrupt(int irq, void *dev_id) | |
1224 | { | |
2688fcb7 AB |
1225 | struct netfront_queue *queue = dev_id; |
1226 | struct net_device *dev = queue->info->netdev; | |
d634bf2c WL |
1227 | |
1228 | if (likely(netif_carrier_ok(dev) && | |
2688fcb7 | 1229 | RING_HAS_UNCONSUMED_RESPONSES(&queue->rx))) |
76541869 | 1230 | napi_schedule(&queue->napi); |
cf66f9d4 | 1231 | |
d634bf2c WL |
1232 | return IRQ_HANDLED; |
1233 | } | |
cf66f9d4 | 1234 | |
d634bf2c WL |
1235 | static irqreturn_t xennet_interrupt(int irq, void *dev_id) |
1236 | { | |
1237 | xennet_tx_interrupt(irq, dev_id); | |
1238 | xennet_rx_interrupt(irq, dev_id); | |
cf66f9d4 KRW |
1239 | return IRQ_HANDLED; |
1240 | } | |
1241 | ||
1242 | #ifdef CONFIG_NET_POLL_CONTROLLER | |
1243 | static void xennet_poll_controller(struct net_device *dev) | |
1244 | { | |
2688fcb7 AB |
1245 | /* Poll each queue */ |
1246 | struct netfront_info *info = netdev_priv(dev); | |
1247 | unsigned int num_queues = dev->real_num_tx_queues; | |
1248 | unsigned int i; | |
1249 | for (i = 0; i < num_queues; ++i) | |
1250 | xennet_interrupt(0, &info->queues[i]); | |
cf66f9d4 KRW |
1251 | } |
1252 | #endif | |
1253 | ||
0a0b9d2e SH |
1254 | static const struct net_device_ops xennet_netdev_ops = { |
1255 | .ndo_open = xennet_open, | |
0a0b9d2e SH |
1256 | .ndo_stop = xennet_close, |
1257 | .ndo_start_xmit = xennet_start_xmit, | |
1258 | .ndo_change_mtu = xennet_change_mtu, | |
e00f85be | 1259 | .ndo_get_stats64 = xennet_get_stats64, |
0a0b9d2e SH |
1260 | .ndo_set_mac_address = eth_mac_addr, |
1261 | .ndo_validate_addr = eth_validate_addr, | |
fb507934 MM |
1262 | .ndo_fix_features = xennet_fix_features, |
1263 | .ndo_set_features = xennet_set_features, | |
2688fcb7 | 1264 | .ndo_select_queue = xennet_select_queue, |
cf66f9d4 KRW |
1265 | #ifdef CONFIG_NET_POLL_CONTROLLER |
1266 | .ndo_poll_controller = xennet_poll_controller, | |
1267 | #endif | |
0a0b9d2e SH |
1268 | }; |
1269 | ||
900e1833 DV |
1270 | static void xennet_free_netdev(struct net_device *netdev) |
1271 | { | |
1272 | struct netfront_info *np = netdev_priv(netdev); | |
1273 | ||
1274 | free_percpu(np->rx_stats); | |
1275 | free_percpu(np->tx_stats); | |
1276 | free_netdev(netdev); | |
1277 | } | |
1278 | ||
8e0e46bb | 1279 | static struct net_device *xennet_create_dev(struct xenbus_device *dev) |
0d160211 | 1280 | { |
2688fcb7 | 1281 | int err; |
0d160211 JF |
1282 | struct net_device *netdev; |
1283 | struct netfront_info *np; | |
1284 | ||
50ee6061 | 1285 | netdev = alloc_etherdev_mq(sizeof(struct netfront_info), xennet_max_queues); |
41de8d4c | 1286 | if (!netdev) |
0d160211 | 1287 | return ERR_PTR(-ENOMEM); |
0d160211 JF |
1288 | |
1289 | np = netdev_priv(netdev); | |
1290 | np->xbdev = dev; | |
1291 | ||
2688fcb7 | 1292 | np->queues = NULL; |
0d160211 | 1293 | |
e00f85be | 1294 | err = -ENOMEM; |
900e1833 DV |
1295 | np->rx_stats = netdev_alloc_pcpu_stats(struct netfront_stats); |
1296 | if (np->rx_stats == NULL) | |
1297 | goto exit; | |
1298 | np->tx_stats = netdev_alloc_pcpu_stats(struct netfront_stats); | |
1299 | if (np->tx_stats == NULL) | |
e00f85be | 1300 | goto exit; |
1301 | ||
0a0b9d2e SH |
1302 | netdev->netdev_ops = &xennet_netdev_ops; |
1303 | ||
fb507934 MM |
1304 | netdev->features = NETIF_F_IP_CSUM | NETIF_F_RXCSUM | |
1305 | NETIF_F_GSO_ROBUST; | |
2c0057de PD |
1306 | netdev->hw_features = NETIF_F_SG | |
1307 | NETIF_F_IPV6_CSUM | | |
1308 | NETIF_F_TSO | NETIF_F_TSO6; | |
0d160211 | 1309 | |
fc3e5941 IC |
1310 | /* |
1311 | * Assume that all hw features are available for now. This set | |
1312 | * will be adjusted by the call to netdev_update_features() in | |
1313 | * xennet_connect() which is the earliest point where we can | |
1314 | * negotiate with the backend regarding supported features. | |
1315 | */ | |
1316 | netdev->features |= netdev->hw_features; | |
1317 | ||
7ad24ea4 | 1318 | netdev->ethtool_ops = &xennet_ethtool_ops; |
d0c2c997 JW |
1319 | netdev->min_mtu = 0; |
1320 | netdev->max_mtu = XEN_NETIF_MAX_TX_SIZE; | |
0d160211 JF |
1321 | SET_NETDEV_DEV(netdev, &dev->dev); |
1322 | ||
1323 | np->netdev = netdev; | |
1324 | ||
1325 | netif_carrier_off(netdev); | |
1326 | ||
1327 | return netdev; | |
1328 | ||
0d160211 | 1329 | exit: |
900e1833 | 1330 | xennet_free_netdev(netdev); |
0d160211 JF |
1331 | return ERR_PTR(err); |
1332 | } | |
1333 | ||
1334 | /** | |
1335 | * Entry point to this code when a new device is created. Allocate the basic | |
1336 | * structures and the ring buffers for communication with the backend, and | |
1337 | * inform the backend of the appropriate details for those. | |
1338 | */ | |
8e0e46bb | 1339 | static int netfront_probe(struct xenbus_device *dev, |
1dd06ae8 | 1340 | const struct xenbus_device_id *id) |
0d160211 JF |
1341 | { |
1342 | int err; | |
1343 | struct net_device *netdev; | |
1344 | struct netfront_info *info; | |
1345 | ||
1346 | netdev = xennet_create_dev(dev); | |
1347 | if (IS_ERR(netdev)) { | |
1348 | err = PTR_ERR(netdev); | |
1349 | xenbus_dev_fatal(dev, err, "creating netdev"); | |
1350 | return err; | |
1351 | } | |
1352 | ||
1353 | info = netdev_priv(netdev); | |
1b713e00 | 1354 | dev_set_drvdata(&dev->dev, info); |
27b917e5 TI |
1355 | #ifdef CONFIG_SYSFS |
1356 | info->netdev->sysfs_groups[0] = &xennet_dev_group; | |
1357 | #endif | |
0d160211 JF |
1358 | err = register_netdev(info->netdev); |
1359 | if (err) { | |
383eda32 | 1360 | pr_warn("%s: register_netdev err=%d\n", __func__, err); |
0d160211 JF |
1361 | goto fail; |
1362 | } | |
1363 | ||
0d160211 JF |
1364 | return 0; |
1365 | ||
1366 | fail: | |
900e1833 | 1367 | xennet_free_netdev(netdev); |
1b713e00 | 1368 | dev_set_drvdata(&dev->dev, NULL); |
0d160211 JF |
1369 | return err; |
1370 | } | |
1371 | ||
1372 | static void xennet_end_access(int ref, void *page) | |
1373 | { | |
1374 | /* This frees the page as a side-effect */ | |
1375 | if (ref != GRANT_INVALID_REF) | |
1376 | gnttab_end_foreign_access(ref, 0, (unsigned long)page); | |
1377 | } | |
1378 | ||
1379 | static void xennet_disconnect_backend(struct netfront_info *info) | |
1380 | { | |
2688fcb7 | 1381 | unsigned int i = 0; |
2688fcb7 AB |
1382 | unsigned int num_queues = info->netdev->real_num_tx_queues; |
1383 | ||
f9feb1e6 DV |
1384 | netif_carrier_off(info->netdev); |
1385 | ||
9a873c71 | 1386 | for (i = 0; i < num_queues && info->queues; ++i) { |
76541869 DV |
1387 | struct netfront_queue *queue = &info->queues[i]; |
1388 | ||
74470954 BO |
1389 | del_timer_sync(&queue->rx_refill_timer); |
1390 | ||
2688fcb7 AB |
1391 | if (queue->tx_irq && (queue->tx_irq == queue->rx_irq)) |
1392 | unbind_from_irqhandler(queue->tx_irq, queue); | |
1393 | if (queue->tx_irq && (queue->tx_irq != queue->rx_irq)) { | |
1394 | unbind_from_irqhandler(queue->tx_irq, queue); | |
1395 | unbind_from_irqhandler(queue->rx_irq, queue); | |
1396 | } | |
1397 | queue->tx_evtchn = queue->rx_evtchn = 0; | |
1398 | queue->tx_irq = queue->rx_irq = 0; | |
0d160211 | 1399 | |
274b0455 CW |
1400 | if (netif_running(info->netdev)) |
1401 | napi_synchronize(&queue->napi); | |
f9feb1e6 | 1402 | |
a5b5dc3c DV |
1403 | xennet_release_tx_bufs(queue); |
1404 | xennet_release_rx_bufs(queue); | |
1405 | gnttab_free_grant_references(queue->gref_tx_head); | |
1406 | gnttab_free_grant_references(queue->gref_rx_head); | |
1407 | ||
2688fcb7 AB |
1408 | /* End access and free the pages */ |
1409 | xennet_end_access(queue->tx_ring_ref, queue->tx.sring); | |
1410 | xennet_end_access(queue->rx_ring_ref, queue->rx.sring); | |
0d160211 | 1411 | |
2688fcb7 AB |
1412 | queue->tx_ring_ref = GRANT_INVALID_REF; |
1413 | queue->rx_ring_ref = GRANT_INVALID_REF; | |
1414 | queue->tx.sring = NULL; | |
1415 | queue->rx.sring = NULL; | |
1416 | } | |
0d160211 JF |
1417 | } |
1418 | ||
1419 | /** | |
1420 | * We are reconnecting to the backend, due to a suspend/resume, or a backend | |
1421 | * driver restart. We tear down our netif structure and recreate it, but | |
1422 | * leave the device-layer structures intact so that this is transparent to the | |
1423 | * rest of the kernel. | |
1424 | */ | |
1425 | static int netfront_resume(struct xenbus_device *dev) | |
1426 | { | |
1b713e00 | 1427 | struct netfront_info *info = dev_get_drvdata(&dev->dev); |
0d160211 JF |
1428 | |
1429 | dev_dbg(&dev->dev, "%s\n", dev->nodename); | |
1430 | ||
1431 | xennet_disconnect_backend(info); | |
1432 | return 0; | |
1433 | } | |
1434 | ||
1435 | static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[]) | |
1436 | { | |
1437 | char *s, *e, *macstr; | |
1438 | int i; | |
1439 | ||
1440 | macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL); | |
1441 | if (IS_ERR(macstr)) | |
1442 | return PTR_ERR(macstr); | |
1443 | ||
1444 | for (i = 0; i < ETH_ALEN; i++) { | |
1445 | mac[i] = simple_strtoul(s, &e, 16); | |
1446 | if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) { | |
1447 | kfree(macstr); | |
1448 | return -ENOENT; | |
1449 | } | |
1450 | s = e+1; | |
1451 | } | |
1452 | ||
1453 | kfree(macstr); | |
1454 | return 0; | |
1455 | } | |
1456 | ||
2688fcb7 | 1457 | static int setup_netfront_single(struct netfront_queue *queue) |
d634bf2c WL |
1458 | { |
1459 | int err; | |
1460 | ||
2688fcb7 | 1461 | err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn); |
d634bf2c WL |
1462 | if (err < 0) |
1463 | goto fail; | |
1464 | ||
2688fcb7 | 1465 | err = bind_evtchn_to_irqhandler(queue->tx_evtchn, |
d634bf2c | 1466 | xennet_interrupt, |
2688fcb7 | 1467 | 0, queue->info->netdev->name, queue); |
d634bf2c WL |
1468 | if (err < 0) |
1469 | goto bind_fail; | |
2688fcb7 AB |
1470 | queue->rx_evtchn = queue->tx_evtchn; |
1471 | queue->rx_irq = queue->tx_irq = err; | |
d634bf2c WL |
1472 | |
1473 | return 0; | |
1474 | ||
1475 | bind_fail: | |
2688fcb7 AB |
1476 | xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn); |
1477 | queue->tx_evtchn = 0; | |
d634bf2c WL |
1478 | fail: |
1479 | return err; | |
1480 | } | |
1481 | ||
2688fcb7 | 1482 | static int setup_netfront_split(struct netfront_queue *queue) |
d634bf2c WL |
1483 | { |
1484 | int err; | |
1485 | ||
2688fcb7 | 1486 | err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn); |
d634bf2c WL |
1487 | if (err < 0) |
1488 | goto fail; | |
2688fcb7 | 1489 | err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->rx_evtchn); |
d634bf2c WL |
1490 | if (err < 0) |
1491 | goto alloc_rx_evtchn_fail; | |
1492 | ||
2688fcb7 AB |
1493 | snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name), |
1494 | "%s-tx", queue->name); | |
1495 | err = bind_evtchn_to_irqhandler(queue->tx_evtchn, | |
d634bf2c | 1496 | xennet_tx_interrupt, |
2688fcb7 | 1497 | 0, queue->tx_irq_name, queue); |
d634bf2c WL |
1498 | if (err < 0) |
1499 | goto bind_tx_fail; | |
2688fcb7 | 1500 | queue->tx_irq = err; |
d634bf2c | 1501 | |
2688fcb7 AB |
1502 | snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name), |
1503 | "%s-rx", queue->name); | |
1504 | err = bind_evtchn_to_irqhandler(queue->rx_evtchn, | |
d634bf2c | 1505 | xennet_rx_interrupt, |
2688fcb7 | 1506 | 0, queue->rx_irq_name, queue); |
d634bf2c WL |
1507 | if (err < 0) |
1508 | goto bind_rx_fail; | |
2688fcb7 | 1509 | queue->rx_irq = err; |
d634bf2c WL |
1510 | |
1511 | return 0; | |
1512 | ||
1513 | bind_rx_fail: | |
2688fcb7 AB |
1514 | unbind_from_irqhandler(queue->tx_irq, queue); |
1515 | queue->tx_irq = 0; | |
d634bf2c | 1516 | bind_tx_fail: |
2688fcb7 AB |
1517 | xenbus_free_evtchn(queue->info->xbdev, queue->rx_evtchn); |
1518 | queue->rx_evtchn = 0; | |
d634bf2c | 1519 | alloc_rx_evtchn_fail: |
2688fcb7 AB |
1520 | xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn); |
1521 | queue->tx_evtchn = 0; | |
d634bf2c WL |
1522 | fail: |
1523 | return err; | |
1524 | } | |
1525 | ||
2688fcb7 AB |
1526 | static int setup_netfront(struct xenbus_device *dev, |
1527 | struct netfront_queue *queue, unsigned int feature_split_evtchn) | |
0d160211 JF |
1528 | { |
1529 | struct xen_netif_tx_sring *txs; | |
1530 | struct xen_netif_rx_sring *rxs; | |
ccc9d90a | 1531 | grant_ref_t gref; |
0d160211 | 1532 | int err; |
0d160211 | 1533 | |
2688fcb7 AB |
1534 | queue->tx_ring_ref = GRANT_INVALID_REF; |
1535 | queue->rx_ring_ref = GRANT_INVALID_REF; | |
1536 | queue->rx.sring = NULL; | |
1537 | queue->tx.sring = NULL; | |
0d160211 | 1538 | |
a144ff09 | 1539 | txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH); |
0d160211 JF |
1540 | if (!txs) { |
1541 | err = -ENOMEM; | |
1542 | xenbus_dev_fatal(dev, err, "allocating tx ring page"); | |
1543 | goto fail; | |
1544 | } | |
1545 | SHARED_RING_INIT(txs); | |
30c5d7f0 | 1546 | FRONT_RING_INIT(&queue->tx, txs, XEN_PAGE_SIZE); |
0d160211 | 1547 | |
ccc9d90a | 1548 | err = xenbus_grant_ring(dev, txs, 1, &gref); |
1ca2983a WL |
1549 | if (err < 0) |
1550 | goto grant_tx_ring_fail; | |
ccc9d90a | 1551 | queue->tx_ring_ref = gref; |
0d160211 | 1552 | |
a144ff09 | 1553 | rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH); |
0d160211 JF |
1554 | if (!rxs) { |
1555 | err = -ENOMEM; | |
1556 | xenbus_dev_fatal(dev, err, "allocating rx ring page"); | |
1ca2983a | 1557 | goto alloc_rx_ring_fail; |
0d160211 JF |
1558 | } |
1559 | SHARED_RING_INIT(rxs); | |
30c5d7f0 | 1560 | FRONT_RING_INIT(&queue->rx, rxs, XEN_PAGE_SIZE); |
0d160211 | 1561 | |
ccc9d90a | 1562 | err = xenbus_grant_ring(dev, rxs, 1, &gref); |
1ca2983a WL |
1563 | if (err < 0) |
1564 | goto grant_rx_ring_fail; | |
ccc9d90a | 1565 | queue->rx_ring_ref = gref; |
0d160211 | 1566 | |
d634bf2c | 1567 | if (feature_split_evtchn) |
2688fcb7 | 1568 | err = setup_netfront_split(queue); |
d634bf2c WL |
1569 | /* setup single event channel if |
1570 | * a) feature-split-event-channels == 0 | |
1571 | * b) feature-split-event-channels == 1 but failed to setup | |
1572 | */ | |
1573 | if (!feature_split_evtchn || (feature_split_evtchn && err)) | |
2688fcb7 | 1574 | err = setup_netfront_single(queue); |
d634bf2c | 1575 | |
0d160211 | 1576 | if (err) |
1ca2983a | 1577 | goto alloc_evtchn_fail; |
0d160211 | 1578 | |
0d160211 JF |
1579 | return 0; |
1580 | ||
1ca2983a WL |
1581 | /* If we fail to setup netfront, it is safe to just revoke access to |
1582 | * granted pages because backend is not accessing it at this point. | |
1583 | */ | |
1ca2983a | 1584 | alloc_evtchn_fail: |
2688fcb7 | 1585 | gnttab_end_foreign_access_ref(queue->rx_ring_ref, 0); |
1ca2983a WL |
1586 | grant_rx_ring_fail: |
1587 | free_page((unsigned long)rxs); | |
1588 | alloc_rx_ring_fail: | |
2688fcb7 | 1589 | gnttab_end_foreign_access_ref(queue->tx_ring_ref, 0); |
1ca2983a WL |
1590 | grant_tx_ring_fail: |
1591 | free_page((unsigned long)txs); | |
1592 | fail: | |
0d160211 JF |
1593 | return err; |
1594 | } | |
1595 | ||
2688fcb7 AB |
1596 | /* Queue-specific initialisation |
1597 | * This used to be done in xennet_create_dev() but must now | |
1598 | * be run per-queue. | |
1599 | */ | |
1600 | static int xennet_init_queue(struct netfront_queue *queue) | |
1601 | { | |
1602 | unsigned short i; | |
1603 | int err = 0; | |
1604 | ||
1605 | spin_lock_init(&queue->tx_lock); | |
1606 | spin_lock_init(&queue->rx_lock); | |
1607 | ||
493be55a VT |
1608 | setup_timer(&queue->rx_refill_timer, rx_refill_timeout, |
1609 | (unsigned long)queue); | |
2688fcb7 | 1610 | |
8b715010 WL |
1611 | snprintf(queue->name, sizeof(queue->name), "%s-q%u", |
1612 | queue->info->netdev->name, queue->id); | |
1613 | ||
2688fcb7 AB |
1614 | /* Initialise tx_skbs as a free chain containing every entry. */ |
1615 | queue->tx_skb_freelist = 0; | |
1616 | for (i = 0; i < NET_TX_RING_SIZE; i++) { | |
1617 | skb_entry_set_link(&queue->tx_skbs[i], i+1); | |
1618 | queue->grant_tx_ref[i] = GRANT_INVALID_REF; | |
1619 | queue->grant_tx_page[i] = NULL; | |
1620 | } | |
1621 | ||
1622 | /* Clear out rx_skbs */ | |
1623 | for (i = 0; i < NET_RX_RING_SIZE; i++) { | |
1624 | queue->rx_skbs[i] = NULL; | |
1625 | queue->grant_rx_ref[i] = GRANT_INVALID_REF; | |
1626 | } | |
1627 | ||
1628 | /* A grant for every tx ring slot */ | |
1f3c2eba | 1629 | if (gnttab_alloc_grant_references(NET_TX_RING_SIZE, |
2688fcb7 AB |
1630 | &queue->gref_tx_head) < 0) { |
1631 | pr_alert("can't alloc tx grant refs\n"); | |
1632 | err = -ENOMEM; | |
1633 | goto exit; | |
1634 | } | |
1635 | ||
1636 | /* A grant for every rx ring slot */ | |
1f3c2eba | 1637 | if (gnttab_alloc_grant_references(NET_RX_RING_SIZE, |
2688fcb7 AB |
1638 | &queue->gref_rx_head) < 0) { |
1639 | pr_alert("can't alloc rx grant refs\n"); | |
1640 | err = -ENOMEM; | |
1641 | goto exit_free_tx; | |
1642 | } | |
1643 | ||
2688fcb7 AB |
1644 | return 0; |
1645 | ||
1646 | exit_free_tx: | |
1647 | gnttab_free_grant_references(queue->gref_tx_head); | |
1648 | exit: | |
1649 | return err; | |
1650 | } | |
1651 | ||
50ee6061 AB |
1652 | static int write_queue_xenstore_keys(struct netfront_queue *queue, |
1653 | struct xenbus_transaction *xbt, int write_hierarchical) | |
1654 | { | |
1655 | /* Write the queue-specific keys into XenStore in the traditional | |
1656 | * way for a single queue, or in a queue subkeys for multiple | |
1657 | * queues. | |
1658 | */ | |
1659 | struct xenbus_device *dev = queue->info->xbdev; | |
1660 | int err; | |
1661 | const char *message; | |
1662 | char *path; | |
1663 | size_t pathsize; | |
1664 | ||
1665 | /* Choose the correct place to write the keys */ | |
1666 | if (write_hierarchical) { | |
1667 | pathsize = strlen(dev->nodename) + 10; | |
1668 | path = kzalloc(pathsize, GFP_KERNEL); | |
1669 | if (!path) { | |
1670 | err = -ENOMEM; | |
1671 | message = "out of memory while writing ring references"; | |
1672 | goto error; | |
1673 | } | |
1674 | snprintf(path, pathsize, "%s/queue-%u", | |
1675 | dev->nodename, queue->id); | |
1676 | } else { | |
1677 | path = (char *)dev->nodename; | |
1678 | } | |
1679 | ||
1680 | /* Write ring references */ | |
1681 | err = xenbus_printf(*xbt, path, "tx-ring-ref", "%u", | |
1682 | queue->tx_ring_ref); | |
1683 | if (err) { | |
1684 | message = "writing tx-ring-ref"; | |
1685 | goto error; | |
1686 | } | |
1687 | ||
1688 | err = xenbus_printf(*xbt, path, "rx-ring-ref", "%u", | |
1689 | queue->rx_ring_ref); | |
1690 | if (err) { | |
1691 | message = "writing rx-ring-ref"; | |
1692 | goto error; | |
1693 | } | |
1694 | ||
1695 | /* Write event channels; taking into account both shared | |
1696 | * and split event channel scenarios. | |
1697 | */ | |
1698 | if (queue->tx_evtchn == queue->rx_evtchn) { | |
1699 | /* Shared event channel */ | |
1700 | err = xenbus_printf(*xbt, path, | |
1701 | "event-channel", "%u", queue->tx_evtchn); | |
1702 | if (err) { | |
1703 | message = "writing event-channel"; | |
1704 | goto error; | |
1705 | } | |
1706 | } else { | |
1707 | /* Split event channels */ | |
1708 | err = xenbus_printf(*xbt, path, | |
1709 | "event-channel-tx", "%u", queue->tx_evtchn); | |
1710 | if (err) { | |
1711 | message = "writing event-channel-tx"; | |
1712 | goto error; | |
1713 | } | |
1714 | ||
1715 | err = xenbus_printf(*xbt, path, | |
1716 | "event-channel-rx", "%u", queue->rx_evtchn); | |
1717 | if (err) { | |
1718 | message = "writing event-channel-rx"; | |
1719 | goto error; | |
1720 | } | |
1721 | } | |
1722 | ||
1723 | if (write_hierarchical) | |
1724 | kfree(path); | |
1725 | return 0; | |
1726 | ||
1727 | error: | |
1728 | if (write_hierarchical) | |
1729 | kfree(path); | |
1730 | xenbus_dev_fatal(dev, err, "%s", message); | |
1731 | return err; | |
1732 | } | |
1733 | ||
ce58725f DV |
1734 | static void xennet_destroy_queues(struct netfront_info *info) |
1735 | { | |
1736 | unsigned int i; | |
1737 | ||
1738 | rtnl_lock(); | |
1739 | ||
1740 | for (i = 0; i < info->netdev->real_num_tx_queues; i++) { | |
1741 | struct netfront_queue *queue = &info->queues[i]; | |
1742 | ||
1743 | if (netif_running(info->netdev)) | |
1744 | napi_disable(&queue->napi); | |
1745 | netif_napi_del(&queue->napi); | |
1746 | } | |
1747 | ||
1748 | rtnl_unlock(); | |
1749 | ||
1750 | kfree(info->queues); | |
1751 | info->queues = NULL; | |
1752 | } | |
1753 | ||
1754 | static int xennet_create_queues(struct netfront_info *info, | |
ca88ea12 | 1755 | unsigned int *num_queues) |
ce58725f DV |
1756 | { |
1757 | unsigned int i; | |
1758 | int ret; | |
1759 | ||
ca88ea12 | 1760 | info->queues = kcalloc(*num_queues, sizeof(struct netfront_queue), |
ce58725f DV |
1761 | GFP_KERNEL); |
1762 | if (!info->queues) | |
1763 | return -ENOMEM; | |
1764 | ||
1765 | rtnl_lock(); | |
1766 | ||
ca88ea12 | 1767 | for (i = 0; i < *num_queues; i++) { |
ce58725f DV |
1768 | struct netfront_queue *queue = &info->queues[i]; |
1769 | ||
1770 | queue->id = i; | |
1771 | queue->info = info; | |
1772 | ||
1773 | ret = xennet_init_queue(queue); | |
1774 | if (ret < 0) { | |
69cb8524 DV |
1775 | dev_warn(&info->netdev->dev, |
1776 | "only created %d queues\n", i); | |
ca88ea12 | 1777 | *num_queues = i; |
ce58725f DV |
1778 | break; |
1779 | } | |
1780 | ||
1781 | netif_napi_add(queue->info->netdev, &queue->napi, | |
1782 | xennet_poll, 64); | |
1783 | if (netif_running(info->netdev)) | |
1784 | napi_enable(&queue->napi); | |
1785 | } | |
1786 | ||
ca88ea12 | 1787 | netif_set_real_num_tx_queues(info->netdev, *num_queues); |
ce58725f DV |
1788 | |
1789 | rtnl_unlock(); | |
1790 | ||
ca88ea12 | 1791 | if (*num_queues == 0) { |
ce58725f DV |
1792 | dev_err(&info->netdev->dev, "no queues\n"); |
1793 | return -EINVAL; | |
1794 | } | |
1795 | return 0; | |
1796 | } | |
1797 | ||
0d160211 | 1798 | /* Common code used when first setting up, and when resuming. */ |
f502bf2b | 1799 | static int talk_to_netback(struct xenbus_device *dev, |
0d160211 JF |
1800 | struct netfront_info *info) |
1801 | { | |
1802 | const char *message; | |
1803 | struct xenbus_transaction xbt; | |
1804 | int err; | |
2688fcb7 AB |
1805 | unsigned int feature_split_evtchn; |
1806 | unsigned int i = 0; | |
50ee6061 | 1807 | unsigned int max_queues = 0; |
2688fcb7 AB |
1808 | struct netfront_queue *queue = NULL; |
1809 | unsigned int num_queues = 1; | |
0d160211 | 1810 | |
2688fcb7 AB |
1811 | info->netdev->irq = 0; |
1812 | ||
50ee6061 | 1813 | /* Check if backend supports multiple queues */ |
2890ea5c JG |
1814 | max_queues = xenbus_read_unsigned(info->xbdev->otherend, |
1815 | "multi-queue-max-queues", 1); | |
50ee6061 AB |
1816 | num_queues = min(max_queues, xennet_max_queues); |
1817 | ||
2688fcb7 | 1818 | /* Check feature-split-event-channels */ |
2890ea5c JG |
1819 | feature_split_evtchn = xenbus_read_unsigned(info->xbdev->otherend, |
1820 | "feature-split-event-channels", 0); | |
2688fcb7 AB |
1821 | |
1822 | /* Read mac addr. */ | |
1823 | err = xen_net_read_mac(dev, info->netdev->dev_addr); | |
1824 | if (err) { | |
1825 | xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename); | |
1826 | goto out; | |
1827 | } | |
1828 | ||
ce58725f DV |
1829 | if (info->queues) |
1830 | xennet_destroy_queues(info); | |
1831 | ||
ca88ea12 | 1832 | err = xennet_create_queues(info, &num_queues); |
e2e004ac RL |
1833 | if (err < 0) { |
1834 | xenbus_dev_fatal(dev, err, "creating queues"); | |
1835 | kfree(info->queues); | |
1836 | info->queues = NULL; | |
1837 | goto out; | |
1838 | } | |
2688fcb7 AB |
1839 | |
1840 | /* Create shared ring, alloc event channel -- for each queue */ | |
1841 | for (i = 0; i < num_queues; ++i) { | |
1842 | queue = &info->queues[i]; | |
2688fcb7 | 1843 | err = setup_netfront(dev, queue, feature_split_evtchn); |
e2e004ac RL |
1844 | if (err) |
1845 | goto destroy_ring; | |
2688fcb7 | 1846 | } |
0d160211 JF |
1847 | |
1848 | again: | |
1849 | err = xenbus_transaction_start(&xbt); | |
1850 | if (err) { | |
1851 | xenbus_dev_fatal(dev, err, "starting transaction"); | |
1852 | goto destroy_ring; | |
1853 | } | |
1854 | ||
812494d9 | 1855 | if (xenbus_exists(XBT_NIL, |
1856 | info->xbdev->otherend, "multi-queue-max-queues")) { | |
50ee6061 | 1857 | /* Write the number of queues */ |
812494d9 | 1858 | err = xenbus_printf(xbt, dev->nodename, |
1859 | "multi-queue-num-queues", "%u", num_queues); | |
d634bf2c | 1860 | if (err) { |
50ee6061 AB |
1861 | message = "writing multi-queue-num-queues"; |
1862 | goto abort_transaction_no_dev_fatal; | |
d634bf2c | 1863 | } |
812494d9 | 1864 | } |
50ee6061 | 1865 | |
812494d9 | 1866 | if (num_queues == 1) { |
1867 | err = write_queue_xenstore_keys(&info->queues[0], &xbt, 0); /* flat */ | |
1868 | if (err) | |
1869 | goto abort_transaction_no_dev_fatal; | |
1870 | } else { | |
50ee6061 AB |
1871 | /* Write the keys for each queue */ |
1872 | for (i = 0; i < num_queues; ++i) { | |
1873 | queue = &info->queues[i]; | |
1874 | err = write_queue_xenstore_keys(queue, &xbt, 1); /* hierarchical */ | |
1875 | if (err) | |
1876 | goto abort_transaction_no_dev_fatal; | |
d634bf2c | 1877 | } |
0d160211 JF |
1878 | } |
1879 | ||
50ee6061 | 1880 | /* The remaining keys are not queue-specific */ |
0d160211 JF |
1881 | err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u", |
1882 | 1); | |
1883 | if (err) { | |
1884 | message = "writing request-rx-copy"; | |
1885 | goto abort_transaction; | |
1886 | } | |
1887 | ||
1888 | err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1); | |
1889 | if (err) { | |
1890 | message = "writing feature-rx-notify"; | |
1891 | goto abort_transaction; | |
1892 | } | |
1893 | ||
1894 | err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1); | |
1895 | if (err) { | |
1896 | message = "writing feature-sg"; | |
1897 | goto abort_transaction; | |
1898 | } | |
1899 | ||
1900 | err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1); | |
1901 | if (err) { | |
1902 | message = "writing feature-gso-tcpv4"; | |
1903 | goto abort_transaction; | |
1904 | } | |
1905 | ||
2c0057de PD |
1906 | err = xenbus_write(xbt, dev->nodename, "feature-gso-tcpv6", "1"); |
1907 | if (err) { | |
1908 | message = "writing feature-gso-tcpv6"; | |
1909 | goto abort_transaction; | |
1910 | } | |
1911 | ||
1912 | err = xenbus_write(xbt, dev->nodename, "feature-ipv6-csum-offload", | |
1913 | "1"); | |
1914 | if (err) { | |
1915 | message = "writing feature-ipv6-csum-offload"; | |
1916 | goto abort_transaction; | |
1917 | } | |
1918 | ||
0d160211 JF |
1919 | err = xenbus_transaction_end(xbt, 0); |
1920 | if (err) { | |
1921 | if (err == -EAGAIN) | |
1922 | goto again; | |
1923 | xenbus_dev_fatal(dev, err, "completing transaction"); | |
1924 | goto destroy_ring; | |
1925 | } | |
1926 | ||
1927 | return 0; | |
1928 | ||
1929 | abort_transaction: | |
0d160211 | 1930 | xenbus_dev_fatal(dev, err, "%s", message); |
50ee6061 AB |
1931 | abort_transaction_no_dev_fatal: |
1932 | xenbus_transaction_end(xbt, 1); | |
0d160211 JF |
1933 | destroy_ring: |
1934 | xennet_disconnect_backend(info); | |
e2e004ac | 1935 | xennet_destroy_queues(info); |
0d160211 | 1936 | out: |
d86b5672 | 1937 | device_unregister(&dev->dev); |
0d160211 JF |
1938 | return err; |
1939 | } | |
1940 | ||
0d160211 JF |
1941 | static int xennet_connect(struct net_device *dev) |
1942 | { | |
1943 | struct netfront_info *np = netdev_priv(dev); | |
2688fcb7 | 1944 | unsigned int num_queues = 0; |
a5b5dc3c | 1945 | int err; |
2688fcb7 AB |
1946 | unsigned int j = 0; |
1947 | struct netfront_queue *queue = NULL; | |
0d160211 | 1948 | |
2890ea5c | 1949 | if (!xenbus_read_unsigned(np->xbdev->otherend, "feature-rx-copy", 0)) { |
0d160211 | 1950 | dev_info(&dev->dev, |
898eb71c | 1951 | "backend does not support copying receive path\n"); |
0d160211 JF |
1952 | return -ENODEV; |
1953 | } | |
1954 | ||
f502bf2b | 1955 | err = talk_to_netback(np->xbdev, np); |
0d160211 JF |
1956 | if (err) |
1957 | return err; | |
1958 | ||
2688fcb7 AB |
1959 | /* talk_to_netback() sets the correct number of queues */ |
1960 | num_queues = dev->real_num_tx_queues; | |
1961 | ||
1ba37c51 | 1962 | rtnl_lock(); |
fb507934 | 1963 | netdev_update_features(dev); |
1ba37c51 | 1964 | rtnl_unlock(); |
0d160211 | 1965 | |
0d160211 | 1966 | /* |
a5b5dc3c | 1967 | * All public and private state should now be sane. Get |
0d160211 JF |
1968 | * ready to start sending and receiving packets and give the driver |
1969 | * domain a kick because we've probably just requeued some | |
1970 | * packets. | |
1971 | */ | |
1972 | netif_carrier_on(np->netdev); | |
2688fcb7 AB |
1973 | for (j = 0; j < num_queues; ++j) { |
1974 | queue = &np->queues[j]; | |
f50b4076 | 1975 | |
2688fcb7 AB |
1976 | notify_remote_via_irq(queue->tx_irq); |
1977 | if (queue->tx_irq != queue->rx_irq) | |
1978 | notify_remote_via_irq(queue->rx_irq); | |
2688fcb7 | 1979 | |
f50b4076 DV |
1980 | spin_lock_irq(&queue->tx_lock); |
1981 | xennet_tx_buf_gc(queue); | |
2688fcb7 | 1982 | spin_unlock_irq(&queue->tx_lock); |
f50b4076 DV |
1983 | |
1984 | spin_lock_bh(&queue->rx_lock); | |
1985 | xennet_alloc_rx_buffers(queue); | |
2688fcb7 AB |
1986 | spin_unlock_bh(&queue->rx_lock); |
1987 | } | |
0d160211 JF |
1988 | |
1989 | return 0; | |
1990 | } | |
1991 | ||
1992 | /** | |
1993 | * Callback received when the backend's state changes. | |
1994 | */ | |
f502bf2b | 1995 | static void netback_changed(struct xenbus_device *dev, |
0d160211 JF |
1996 | enum xenbus_state backend_state) |
1997 | { | |
1b713e00 | 1998 | struct netfront_info *np = dev_get_drvdata(&dev->dev); |
0d160211 JF |
1999 | struct net_device *netdev = np->netdev; |
2000 | ||
2001 | dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state)); | |
2002 | ||
2003 | switch (backend_state) { | |
2004 | case XenbusStateInitialising: | |
2005 | case XenbusStateInitialised: | |
b78c9512 NI |
2006 | case XenbusStateReconfiguring: |
2007 | case XenbusStateReconfigured: | |
0d160211 | 2008 | case XenbusStateUnknown: |
0d160211 JF |
2009 | break; |
2010 | ||
2011 | case XenbusStateInitWait: | |
2012 | if (dev->state != XenbusStateInitialising) | |
2013 | break; | |
2014 | if (xennet_connect(netdev) != 0) | |
2015 | break; | |
2016 | xenbus_switch_state(dev, XenbusStateConnected); | |
08e34eb1 LE |
2017 | break; |
2018 | ||
2019 | case XenbusStateConnected: | |
ee89bab1 | 2020 | netdev_notify_peers(netdev); |
0d160211 JF |
2021 | break; |
2022 | ||
bce3ea81 DV |
2023 | case XenbusStateClosed: |
2024 | if (dev->state == XenbusStateClosed) | |
2025 | break; | |
2026 | /* Missed the backend's CLOSING state -- fallthrough */ | |
0d160211 JF |
2027 | case XenbusStateClosing: |
2028 | xenbus_frontend_closed(dev); | |
2029 | break; | |
2030 | } | |
2031 | } | |
2032 | ||
e0ce4af9 IC |
2033 | static const struct xennet_stat { |
2034 | char name[ETH_GSTRING_LEN]; | |
2035 | u16 offset; | |
2036 | } xennet_stats[] = { | |
2037 | { | |
2038 | "rx_gso_checksum_fixup", | |
2039 | offsetof(struct netfront_info, rx_gso_checksum_fixup) | |
2040 | }, | |
2041 | }; | |
2042 | ||
2043 | static int xennet_get_sset_count(struct net_device *dev, int string_set) | |
2044 | { | |
2045 | switch (string_set) { | |
2046 | case ETH_SS_STATS: | |
2047 | return ARRAY_SIZE(xennet_stats); | |
2048 | default: | |
2049 | return -EINVAL; | |
2050 | } | |
2051 | } | |
2052 | ||
2053 | static void xennet_get_ethtool_stats(struct net_device *dev, | |
2054 | struct ethtool_stats *stats, u64 * data) | |
2055 | { | |
2056 | void *np = netdev_priv(dev); | |
2057 | int i; | |
2058 | ||
2059 | for (i = 0; i < ARRAY_SIZE(xennet_stats); i++) | |
2688fcb7 | 2060 | data[i] = atomic_read((atomic_t *)(np + xennet_stats[i].offset)); |
e0ce4af9 IC |
2061 | } |
2062 | ||
2063 | static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data) | |
2064 | { | |
2065 | int i; | |
2066 | ||
2067 | switch (stringset) { | |
2068 | case ETH_SS_STATS: | |
2069 | for (i = 0; i < ARRAY_SIZE(xennet_stats); i++) | |
2070 | memcpy(data + i * ETH_GSTRING_LEN, | |
2071 | xennet_stats[i].name, ETH_GSTRING_LEN); | |
2072 | break; | |
2073 | } | |
2074 | } | |
2075 | ||
0fc0b732 | 2076 | static const struct ethtool_ops xennet_ethtool_ops = |
0d160211 | 2077 | { |
0d160211 | 2078 | .get_link = ethtool_op_get_link, |
e0ce4af9 IC |
2079 | |
2080 | .get_sset_count = xennet_get_sset_count, | |
2081 | .get_ethtool_stats = xennet_get_ethtool_stats, | |
2082 | .get_strings = xennet_get_strings, | |
0d160211 JF |
2083 | }; |
2084 | ||
2085 | #ifdef CONFIG_SYSFS | |
1f3c2eba DV |
2086 | static ssize_t show_rxbuf(struct device *dev, |
2087 | struct device_attribute *attr, char *buf) | |
0d160211 | 2088 | { |
1f3c2eba | 2089 | return sprintf(buf, "%lu\n", NET_RX_RING_SIZE); |
0d160211 JF |
2090 | } |
2091 | ||
1f3c2eba DV |
2092 | static ssize_t store_rxbuf(struct device *dev, |
2093 | struct device_attribute *attr, | |
2094 | const char *buf, size_t len) | |
0d160211 | 2095 | { |
0d160211 JF |
2096 | char *endp; |
2097 | unsigned long target; | |
2098 | ||
2099 | if (!capable(CAP_NET_ADMIN)) | |
2100 | return -EPERM; | |
2101 | ||
2102 | target = simple_strtoul(buf, &endp, 0); | |
2103 | if (endp == buf) | |
2104 | return -EBADMSG; | |
2105 | ||
1f3c2eba | 2106 | /* rxbuf_min and rxbuf_max are no longer configurable. */ |
0d160211 | 2107 | |
0d160211 JF |
2108 | return len; |
2109 | } | |
2110 | ||
27b917e5 TI |
2111 | static DEVICE_ATTR(rxbuf_min, S_IRUGO|S_IWUSR, show_rxbuf, store_rxbuf); |
2112 | static DEVICE_ATTR(rxbuf_max, S_IRUGO|S_IWUSR, show_rxbuf, store_rxbuf); | |
2113 | static DEVICE_ATTR(rxbuf_cur, S_IRUGO, show_rxbuf, NULL); | |
0d160211 | 2114 | |
27b917e5 TI |
2115 | static struct attribute *xennet_dev_attrs[] = { |
2116 | &dev_attr_rxbuf_min.attr, | |
2117 | &dev_attr_rxbuf_max.attr, | |
2118 | &dev_attr_rxbuf_cur.attr, | |
2119 | NULL | |
2120 | }; | |
0d160211 | 2121 | |
27b917e5 TI |
2122 | static const struct attribute_group xennet_dev_group = { |
2123 | .attrs = xennet_dev_attrs | |
2124 | }; | |
0d160211 JF |
2125 | #endif /* CONFIG_SYSFS */ |
2126 | ||
8e0e46bb | 2127 | static int xennet_remove(struct xenbus_device *dev) |
0d160211 | 2128 | { |
1b713e00 | 2129 | struct netfront_info *info = dev_get_drvdata(&dev->dev); |
0d160211 JF |
2130 | |
2131 | dev_dbg(&dev->dev, "%s\n", dev->nodename); | |
2132 | ||
0d160211 JF |
2133 | xennet_disconnect_backend(info); |
2134 | ||
6bc96d04 IC |
2135 | unregister_netdev(info->netdev); |
2136 | ||
9a873c71 CW |
2137 | if (info->queues) |
2138 | xennet_destroy_queues(info); | |
900e1833 | 2139 | xennet_free_netdev(info->netdev); |
0d160211 JF |
2140 | |
2141 | return 0; | |
2142 | } | |
2143 | ||
95afae48 DV |
2144 | static const struct xenbus_device_id netfront_ids[] = { |
2145 | { "vif" }, | |
2146 | { "" } | |
2147 | }; | |
2148 | ||
2149 | static struct xenbus_driver netfront_driver = { | |
2150 | .ids = netfront_ids, | |
0d160211 | 2151 | .probe = netfront_probe, |
8e0e46bb | 2152 | .remove = xennet_remove, |
0d160211 | 2153 | .resume = netfront_resume, |
f502bf2b | 2154 | .otherend_changed = netback_changed, |
95afae48 | 2155 | }; |
0d160211 JF |
2156 | |
2157 | static int __init netif_init(void) | |
2158 | { | |
6e833587 | 2159 | if (!xen_domain()) |
0d160211 JF |
2160 | return -ENODEV; |
2161 | ||
51c71a3b | 2162 | if (!xen_has_pv_nic_devices()) |
b9136d20 IM |
2163 | return -ENODEV; |
2164 | ||
383eda32 | 2165 | pr_info("Initialising Xen virtual ethernet driver\n"); |
0d160211 | 2166 | |
034702a6 | 2167 | /* Allow as many queues as there are CPUs inut max. 8 if user has not |
32a84405 WL |
2168 | * specified a value. |
2169 | */ | |
2170 | if (xennet_max_queues == 0) | |
034702a6 JG |
2171 | xennet_max_queues = min_t(unsigned int, MAX_QUEUES_DEFAULT, |
2172 | num_online_cpus()); | |
50ee6061 | 2173 | |
ffb78a26 | 2174 | return xenbus_register_frontend(&netfront_driver); |
0d160211 JF |
2175 | } |
2176 | module_init(netif_init); | |
2177 | ||
2178 | ||
2179 | static void __exit netif_exit(void) | |
2180 | { | |
ffb78a26 | 2181 | xenbus_unregister_driver(&netfront_driver); |
0d160211 JF |
2182 | } |
2183 | module_exit(netif_exit); | |
2184 | ||
2185 | MODULE_DESCRIPTION("Xen virtual network device frontend"); | |
2186 | MODULE_LICENSE("GPL"); | |
d2f0c52b | 2187 | MODULE_ALIAS("xen:vif"); |
4f93f09b | 2188 | MODULE_ALIAS("xennet"); |