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 | ||
ca981633 | 48 | #include <asm/xen/page.h> |
1ccbf534 | 49 | #include <xen/xen.h> |
0d160211 JF |
50 | #include <xen/xenbus.h> |
51 | #include <xen/events.h> | |
52 | #include <xen/page.h> | |
b9136d20 | 53 | #include <xen/platform_pci.h> |
0d160211 JF |
54 | #include <xen/grant_table.h> |
55 | ||
56 | #include <xen/interface/io/netif.h> | |
57 | #include <xen/interface/memory.h> | |
58 | #include <xen/interface/grant_table.h> | |
59 | ||
50ee6061 AB |
60 | /* Module parameters */ |
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 | ||
667c78af JF |
78 | #define NET_TX_RING_SIZE __CONST_RING_SIZE(xen_netif_tx, PAGE_SIZE) |
79 | #define NET_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, 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; | |
285 | ||
286 | if (unlikely(!netif_carrier_ok(queue->info->netdev))) | |
0d160211 | 287 | return; |
0d160211 | 288 | |
1f3c2eba DV |
289 | for (req_prod = queue->rx.req_prod_pvt; |
290 | req_prod - queue->rx.rsp_cons < NET_RX_RING_SIZE; | |
291 | req_prod++) { | |
292 | struct sk_buff *skb; | |
293 | unsigned short id; | |
294 | grant_ref_t ref; | |
295 | unsigned long pfn; | |
296 | struct xen_netif_rx_request *req; | |
0d160211 | 297 | |
1f3c2eba DV |
298 | skb = xennet_alloc_one_rx_buffer(queue); |
299 | if (!skb) | |
0d160211 JF |
300 | break; |
301 | ||
1f3c2eba | 302 | id = xennet_rxidx(req_prod); |
0d160211 | 303 | |
2688fcb7 AB |
304 | BUG_ON(queue->rx_skbs[id]); |
305 | queue->rx_skbs[id] = skb; | |
0d160211 | 306 | |
2688fcb7 | 307 | ref = gnttab_claim_grant_reference(&queue->gref_rx_head); |
0d160211 | 308 | BUG_ON((signed short)ref < 0); |
2688fcb7 | 309 | queue->grant_rx_ref[id] = ref; |
0d160211 | 310 | |
01c68026 | 311 | pfn = page_to_pfn(skb_frag_page(&skb_shinfo(skb)->frags[0])); |
0d160211 | 312 | |
1f3c2eba | 313 | req = RING_GET_REQUEST(&queue->rx, req_prod); |
0d160211 | 314 | gnttab_grant_foreign_access_ref(ref, |
2688fcb7 | 315 | queue->info->xbdev->otherend_id, |
0d160211 JF |
316 | pfn_to_mfn(pfn), |
317 | 0); | |
318 | ||
319 | req->id = id; | |
320 | req->gref = ref; | |
321 | } | |
322 | ||
1f3c2eba DV |
323 | queue->rx.req_prod_pvt = req_prod; |
324 | ||
325 | /* Not enough requests? Try again later. */ | |
326 | if (req_prod - queue->rx.rsp_cons < NET_RX_SLOTS_MIN) { | |
327 | mod_timer(&queue->rx_refill_timer, jiffies + (HZ/10)); | |
328 | return; | |
329 | } | |
330 | ||
5dcddfae | 331 | wmb(); /* barrier so backend seens requests */ |
0d160211 | 332 | |
2688fcb7 | 333 | RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->rx, notify); |
0d160211 | 334 | if (notify) |
2688fcb7 | 335 | notify_remote_via_irq(queue->rx_irq); |
0d160211 JF |
336 | } |
337 | ||
338 | static int xennet_open(struct net_device *dev) | |
339 | { | |
340 | struct netfront_info *np = netdev_priv(dev); | |
2688fcb7 AB |
341 | unsigned int num_queues = dev->real_num_tx_queues; |
342 | unsigned int i = 0; | |
343 | struct netfront_queue *queue = NULL; | |
344 | ||
345 | for (i = 0; i < num_queues; ++i) { | |
346 | queue = &np->queues[i]; | |
347 | napi_enable(&queue->napi); | |
348 | ||
349 | spin_lock_bh(&queue->rx_lock); | |
350 | if (netif_carrier_ok(dev)) { | |
351 | xennet_alloc_rx_buffers(queue); | |
352 | queue->rx.sring->rsp_event = queue->rx.rsp_cons + 1; | |
353 | if (RING_HAS_UNCONSUMED_RESPONSES(&queue->rx)) | |
354 | napi_schedule(&queue->napi); | |
355 | } | |
356 | spin_unlock_bh(&queue->rx_lock); | |
0d160211 | 357 | } |
0d160211 | 358 | |
2688fcb7 | 359 | netif_tx_start_all_queues(dev); |
0d160211 JF |
360 | |
361 | return 0; | |
362 | } | |
363 | ||
2688fcb7 | 364 | static void xennet_tx_buf_gc(struct netfront_queue *queue) |
0d160211 JF |
365 | { |
366 | RING_IDX cons, prod; | |
367 | unsigned short id; | |
0d160211 JF |
368 | struct sk_buff *skb; |
369 | ||
2688fcb7 | 370 | BUG_ON(!netif_carrier_ok(queue->info->netdev)); |
0d160211 JF |
371 | |
372 | do { | |
2688fcb7 | 373 | prod = queue->tx.sring->rsp_prod; |
0d160211 JF |
374 | rmb(); /* Ensure we see responses up to 'rp'. */ |
375 | ||
2688fcb7 | 376 | for (cons = queue->tx.rsp_cons; cons != prod; cons++) { |
0d160211 JF |
377 | struct xen_netif_tx_response *txrsp; |
378 | ||
2688fcb7 | 379 | txrsp = RING_GET_RESPONSE(&queue->tx, cons); |
f942dc25 | 380 | if (txrsp->status == XEN_NETIF_RSP_NULL) |
0d160211 JF |
381 | continue; |
382 | ||
383 | id = txrsp->id; | |
2688fcb7 | 384 | skb = queue->tx_skbs[id].skb; |
0d160211 | 385 | if (unlikely(gnttab_query_foreign_access( |
2688fcb7 | 386 | queue->grant_tx_ref[id]) != 0)) { |
383eda32 JP |
387 | pr_alert("%s: warning -- grant still in use by backend domain\n", |
388 | __func__); | |
0d160211 JF |
389 | BUG(); |
390 | } | |
391 | gnttab_end_foreign_access_ref( | |
2688fcb7 | 392 | queue->grant_tx_ref[id], GNTMAP_readonly); |
0d160211 | 393 | gnttab_release_grant_reference( |
2688fcb7 AB |
394 | &queue->gref_tx_head, queue->grant_tx_ref[id]); |
395 | queue->grant_tx_ref[id] = GRANT_INVALID_REF; | |
396 | queue->grant_tx_page[id] = NULL; | |
397 | add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, id); | |
0d160211 JF |
398 | dev_kfree_skb_irq(skb); |
399 | } | |
400 | ||
2688fcb7 | 401 | queue->tx.rsp_cons = prod; |
0d160211 JF |
402 | |
403 | /* | |
404 | * Set a new event, then check for race with update of tx_cons. | |
405 | * Note that it is essential to schedule a callback, no matter | |
406 | * how few buffers are pending. Even if there is space in the | |
407 | * transmit ring, higher layers may be blocked because too much | |
408 | * data is outstanding: in such cases notification from Xen is | |
409 | * likely to be the only kick that we'll get. | |
410 | */ | |
2688fcb7 AB |
411 | queue->tx.sring->rsp_event = |
412 | prod + ((queue->tx.sring->req_prod - prod) >> 1) + 1; | |
0d160211 | 413 | mb(); /* update shared area */ |
2688fcb7 | 414 | } while ((cons == prod) && (prod != queue->tx.sring->rsp_prod)); |
0d160211 | 415 | |
2688fcb7 | 416 | xennet_maybe_wake_tx(queue); |
0d160211 JF |
417 | } |
418 | ||
a55e8bb8 DV |
419 | static struct xen_netif_tx_request *xennet_make_one_txreq( |
420 | struct netfront_queue *queue, struct sk_buff *skb, | |
421 | struct page *page, unsigned int offset, unsigned int len) | |
0d160211 | 422 | { |
0d160211 | 423 | unsigned int id; |
a55e8bb8 | 424 | struct xen_netif_tx_request *tx; |
0d160211 | 425 | grant_ref_t ref; |
0d160211 | 426 | |
a55e8bb8 | 427 | len = min_t(unsigned int, PAGE_SIZE - offset, len); |
0d160211 | 428 | |
a55e8bb8 DV |
429 | id = get_id_from_freelist(&queue->tx_skb_freelist, queue->tx_skbs); |
430 | tx = RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++); | |
431 | ref = gnttab_claim_grant_reference(&queue->gref_tx_head); | |
432 | BUG_ON((signed short)ref < 0); | |
0d160211 | 433 | |
a55e8bb8 DV |
434 | gnttab_grant_foreign_access_ref(ref, queue->info->xbdev->otherend_id, |
435 | page_to_mfn(page), GNTMAP_readonly); | |
0d160211 | 436 | |
a55e8bb8 DV |
437 | queue->tx_skbs[id].skb = skb; |
438 | queue->grant_tx_page[id] = page; | |
439 | queue->grant_tx_ref[id] = ref; | |
0d160211 | 440 | |
a55e8bb8 DV |
441 | tx->id = id; |
442 | tx->gref = ref; | |
443 | tx->offset = offset; | |
444 | tx->size = len; | |
445 | tx->flags = 0; | |
0d160211 | 446 | |
a55e8bb8 DV |
447 | return tx; |
448 | } | |
0d160211 | 449 | |
a55e8bb8 DV |
450 | static struct xen_netif_tx_request *xennet_make_txreqs( |
451 | struct netfront_queue *queue, struct xen_netif_tx_request *tx, | |
452 | struct sk_buff *skb, struct page *page, | |
453 | unsigned int offset, unsigned int len) | |
454 | { | |
455 | /* Skip unused frames from start of page */ | |
456 | page += offset >> PAGE_SHIFT; | |
457 | offset &= ~PAGE_MASK; | |
0d160211 | 458 | |
a55e8bb8 DV |
459 | while (len) { |
460 | tx->flags |= XEN_NETTXF_more_data; | |
461 | tx = xennet_make_one_txreq(queue, skb_get(skb), | |
462 | page, offset, len); | |
463 | page++; | |
464 | offset = 0; | |
465 | len -= tx->size; | |
0d160211 JF |
466 | } |
467 | ||
a55e8bb8 | 468 | return tx; |
0d160211 JF |
469 | } |
470 | ||
f36c3747 | 471 | /* |
e84448d5 DV |
472 | * Count how many ring slots are required to send this skb. Each frag |
473 | * might be a compound page. | |
f36c3747 | 474 | */ |
e84448d5 | 475 | static int xennet_count_skb_slots(struct sk_buff *skb) |
f36c3747 IC |
476 | { |
477 | int i, frags = skb_shinfo(skb)->nr_frags; | |
e84448d5 DV |
478 | int pages; |
479 | ||
480 | pages = PFN_UP(offset_in_page(skb->data) + skb_headlen(skb)); | |
f36c3747 IC |
481 | |
482 | for (i = 0; i < frags; i++) { | |
483 | skb_frag_t *frag = skb_shinfo(skb)->frags + i; | |
484 | unsigned long size = skb_frag_size(frag); | |
485 | unsigned long offset = frag->page_offset; | |
486 | ||
487 | /* Skip unused frames from start of page */ | |
488 | offset &= ~PAGE_MASK; | |
489 | ||
490 | pages += PFN_UP(offset + size); | |
491 | } | |
492 | ||
493 | return pages; | |
494 | } | |
495 | ||
50ee6061 AB |
496 | static u16 xennet_select_queue(struct net_device *dev, struct sk_buff *skb, |
497 | void *accel_priv, select_queue_fallback_t fallback) | |
2688fcb7 | 498 | { |
50ee6061 AB |
499 | unsigned int num_queues = dev->real_num_tx_queues; |
500 | u32 hash; | |
501 | u16 queue_idx; | |
502 | ||
503 | /* First, check if there is only one queue */ | |
504 | if (num_queues == 1) { | |
505 | queue_idx = 0; | |
506 | } else { | |
507 | hash = skb_get_hash(skb); | |
508 | queue_idx = hash % num_queues; | |
509 | } | |
510 | ||
511 | return queue_idx; | |
2688fcb7 AB |
512 | } |
513 | ||
0d160211 JF |
514 | static int xennet_start_xmit(struct sk_buff *skb, struct net_device *dev) |
515 | { | |
0d160211 | 516 | struct netfront_info *np = netdev_priv(dev); |
900e1833 | 517 | struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats); |
a55e8bb8 DV |
518 | struct xen_netif_tx_request *tx, *first_tx; |
519 | unsigned int i; | |
0d160211 | 520 | int notify; |
f36c3747 | 521 | int slots; |
a55e8bb8 DV |
522 | struct page *page; |
523 | unsigned int offset; | |
524 | unsigned int len; | |
cf66f9d4 | 525 | unsigned long flags; |
2688fcb7 AB |
526 | struct netfront_queue *queue = NULL; |
527 | unsigned int num_queues = dev->real_num_tx_queues; | |
528 | u16 queue_index; | |
529 | ||
530 | /* Drop the packet if no queues are set up */ | |
531 | if (num_queues < 1) | |
532 | goto drop; | |
533 | /* Determine which queue to transmit this SKB on */ | |
534 | queue_index = skb_get_queue_mapping(skb); | |
535 | queue = &np->queues[queue_index]; | |
0d160211 | 536 | |
9ecd1a75 WL |
537 | /* If skb->len is too big for wire format, drop skb and alert |
538 | * user about misconfiguration. | |
539 | */ | |
540 | if (unlikely(skb->len > XEN_NETIF_MAX_TX_SIZE)) { | |
541 | net_alert_ratelimited( | |
542 | "xennet: skb->len = %u, too big for wire format\n", | |
543 | skb->len); | |
544 | goto drop; | |
545 | } | |
546 | ||
e84448d5 | 547 | slots = xennet_count_skb_slots(skb); |
f36c3747 | 548 | if (unlikely(slots > MAX_SKB_FRAGS + 1)) { |
97a6d1bb ZK |
549 | net_dbg_ratelimited("xennet: skb rides the rocket: %d slots, %d bytes\n", |
550 | slots, skb->len); | |
551 | if (skb_linearize(skb)) | |
552 | goto drop; | |
0d160211 JF |
553 | } |
554 | ||
a55e8bb8 DV |
555 | page = virt_to_page(skb->data); |
556 | offset = offset_in_page(skb->data); | |
557 | len = skb_headlen(skb); | |
558 | ||
2688fcb7 | 559 | spin_lock_irqsave(&queue->tx_lock, flags); |
0d160211 JF |
560 | |
561 | if (unlikely(!netif_carrier_ok(dev) || | |
f36c3747 | 562 | (slots > 1 && !xennet_can_sg(dev)) || |
04ffcb25 | 563 | netif_needs_gso(dev, skb, netif_skb_features(skb)))) { |
2688fcb7 | 564 | spin_unlock_irqrestore(&queue->tx_lock, flags); |
0d160211 JF |
565 | goto drop; |
566 | } | |
567 | ||
a55e8bb8 DV |
568 | /* First request for the linear area. */ |
569 | first_tx = tx = xennet_make_one_txreq(queue, skb, | |
570 | page, offset, len); | |
571 | page++; | |
572 | offset = 0; | |
573 | len -= tx->size; | |
0d160211 | 574 | |
0d160211 JF |
575 | if (skb->ip_summed == CHECKSUM_PARTIAL) |
576 | /* local packet? */ | |
f942dc25 | 577 | tx->flags |= XEN_NETTXF_csum_blank | XEN_NETTXF_data_validated; |
0d160211 JF |
578 | else if (skb->ip_summed == CHECKSUM_UNNECESSARY) |
579 | /* remote but checksummed. */ | |
f942dc25 | 580 | tx->flags |= XEN_NETTXF_data_validated; |
0d160211 | 581 | |
a55e8bb8 | 582 | /* Optional extra info after the first request. */ |
0d160211 JF |
583 | if (skb_shinfo(skb)->gso_size) { |
584 | struct xen_netif_extra_info *gso; | |
585 | ||
586 | gso = (struct xen_netif_extra_info *) | |
a55e8bb8 | 587 | RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++); |
0d160211 | 588 | |
e2d617c0 | 589 | tx->flags |= XEN_NETTXF_extra_info; |
0d160211 JF |
590 | |
591 | gso->u.gso.size = skb_shinfo(skb)->gso_size; | |
2c0057de PD |
592 | gso->u.gso.type = (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) ? |
593 | XEN_NETIF_GSO_TYPE_TCPV6 : | |
594 | XEN_NETIF_GSO_TYPE_TCPV4; | |
0d160211 JF |
595 | gso->u.gso.pad = 0; |
596 | gso->u.gso.features = 0; | |
597 | ||
598 | gso->type = XEN_NETIF_EXTRA_TYPE_GSO; | |
599 | gso->flags = 0; | |
0d160211 JF |
600 | } |
601 | ||
a55e8bb8 DV |
602 | /* Requests for the rest of the linear area. */ |
603 | tx = xennet_make_txreqs(queue, tx, skb, page, offset, len); | |
604 | ||
605 | /* Requests for all the frags. */ | |
606 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
607 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
608 | tx = xennet_make_txreqs(queue, tx, skb, | |
609 | skb_frag_page(frag), frag->page_offset, | |
610 | skb_frag_size(frag)); | |
611 | } | |
0d160211 | 612 | |
a55e8bb8 DV |
613 | /* First request has the packet length. */ |
614 | first_tx->size = skb->len; | |
0d160211 | 615 | |
2688fcb7 | 616 | RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify); |
0d160211 | 617 | if (notify) |
2688fcb7 | 618 | notify_remote_via_irq(queue->tx_irq); |
0d160211 | 619 | |
900e1833 DV |
620 | u64_stats_update_begin(&tx_stats->syncp); |
621 | tx_stats->bytes += skb->len; | |
622 | tx_stats->packets++; | |
623 | u64_stats_update_end(&tx_stats->syncp); | |
10a273a6 JF |
624 | |
625 | /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */ | |
2688fcb7 | 626 | xennet_tx_buf_gc(queue); |
0d160211 | 627 | |
2688fcb7 AB |
628 | if (!netfront_tx_slot_available(queue)) |
629 | netif_tx_stop_queue(netdev_get_tx_queue(dev, queue->id)); | |
0d160211 | 630 | |
2688fcb7 | 631 | spin_unlock_irqrestore(&queue->tx_lock, flags); |
0d160211 | 632 | |
6ed10654 | 633 | return NETDEV_TX_OK; |
0d160211 JF |
634 | |
635 | drop: | |
09f75cd7 | 636 | dev->stats.tx_dropped++; |
979de8a0 | 637 | dev_kfree_skb_any(skb); |
6ed10654 | 638 | return NETDEV_TX_OK; |
0d160211 JF |
639 | } |
640 | ||
641 | static int xennet_close(struct net_device *dev) | |
642 | { | |
643 | struct netfront_info *np = netdev_priv(dev); | |
2688fcb7 AB |
644 | unsigned int num_queues = dev->real_num_tx_queues; |
645 | unsigned int i; | |
646 | struct netfront_queue *queue; | |
647 | netif_tx_stop_all_queues(np->netdev); | |
648 | for (i = 0; i < num_queues; ++i) { | |
649 | queue = &np->queues[i]; | |
650 | napi_disable(&queue->napi); | |
651 | } | |
0d160211 JF |
652 | return 0; |
653 | } | |
654 | ||
2688fcb7 | 655 | static void xennet_move_rx_slot(struct netfront_queue *queue, struct sk_buff *skb, |
0d160211 JF |
656 | grant_ref_t ref) |
657 | { | |
2688fcb7 AB |
658 | int new = xennet_rxidx(queue->rx.req_prod_pvt); |
659 | ||
660 | BUG_ON(queue->rx_skbs[new]); | |
661 | queue->rx_skbs[new] = skb; | |
662 | queue->grant_rx_ref[new] = ref; | |
663 | RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->id = new; | |
664 | RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->gref = ref; | |
665 | queue->rx.req_prod_pvt++; | |
0d160211 JF |
666 | } |
667 | ||
2688fcb7 | 668 | static int xennet_get_extras(struct netfront_queue *queue, |
0d160211 JF |
669 | struct xen_netif_extra_info *extras, |
670 | RING_IDX rp) | |
671 | ||
672 | { | |
673 | struct xen_netif_extra_info *extra; | |
2688fcb7 AB |
674 | struct device *dev = &queue->info->netdev->dev; |
675 | RING_IDX cons = queue->rx.rsp_cons; | |
0d160211 JF |
676 | int err = 0; |
677 | ||
678 | do { | |
679 | struct sk_buff *skb; | |
680 | grant_ref_t ref; | |
681 | ||
682 | if (unlikely(cons + 1 == rp)) { | |
683 | if (net_ratelimit()) | |
684 | dev_warn(dev, "Missing extra info\n"); | |
685 | err = -EBADR; | |
686 | break; | |
687 | } | |
688 | ||
689 | extra = (struct xen_netif_extra_info *) | |
2688fcb7 | 690 | RING_GET_RESPONSE(&queue->rx, ++cons); |
0d160211 JF |
691 | |
692 | if (unlikely(!extra->type || | |
693 | extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) { | |
694 | if (net_ratelimit()) | |
695 | dev_warn(dev, "Invalid extra type: %d\n", | |
696 | extra->type); | |
697 | err = -EINVAL; | |
698 | } else { | |
699 | memcpy(&extras[extra->type - 1], extra, | |
700 | sizeof(*extra)); | |
701 | } | |
702 | ||
2688fcb7 AB |
703 | skb = xennet_get_rx_skb(queue, cons); |
704 | ref = xennet_get_rx_ref(queue, cons); | |
705 | xennet_move_rx_slot(queue, skb, ref); | |
0d160211 JF |
706 | } while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE); |
707 | ||
2688fcb7 | 708 | queue->rx.rsp_cons = cons; |
0d160211 JF |
709 | return err; |
710 | } | |
711 | ||
2688fcb7 | 712 | static int xennet_get_responses(struct netfront_queue *queue, |
0d160211 JF |
713 | struct netfront_rx_info *rinfo, RING_IDX rp, |
714 | struct sk_buff_head *list) | |
715 | { | |
716 | struct xen_netif_rx_response *rx = &rinfo->rx; | |
717 | struct xen_netif_extra_info *extras = rinfo->extras; | |
2688fcb7 AB |
718 | struct device *dev = &queue->info->netdev->dev; |
719 | RING_IDX cons = queue->rx.rsp_cons; | |
720 | struct sk_buff *skb = xennet_get_rx_skb(queue, cons); | |
721 | grant_ref_t ref = xennet_get_rx_ref(queue, cons); | |
0d160211 | 722 | int max = MAX_SKB_FRAGS + (rx->status <= RX_COPY_THRESHOLD); |
7158ff6d | 723 | int slots = 1; |
0d160211 JF |
724 | int err = 0; |
725 | unsigned long ret; | |
726 | ||
f942dc25 | 727 | if (rx->flags & XEN_NETRXF_extra_info) { |
2688fcb7 AB |
728 | err = xennet_get_extras(queue, extras, rp); |
729 | cons = queue->rx.rsp_cons; | |
0d160211 JF |
730 | } |
731 | ||
732 | for (;;) { | |
733 | if (unlikely(rx->status < 0 || | |
734 | rx->offset + rx->status > PAGE_SIZE)) { | |
735 | if (net_ratelimit()) | |
736 | dev_warn(dev, "rx->offset: %x, size: %u\n", | |
737 | rx->offset, rx->status); | |
2688fcb7 | 738 | xennet_move_rx_slot(queue, skb, ref); |
0d160211 JF |
739 | err = -EINVAL; |
740 | goto next; | |
741 | } | |
742 | ||
743 | /* | |
744 | * This definitely indicates a bug, either in this driver or in | |
745 | * the backend driver. In future this should flag the bad | |
697089dc | 746 | * situation to the system controller to reboot the backend. |
0d160211 JF |
747 | */ |
748 | if (ref == GRANT_INVALID_REF) { | |
749 | if (net_ratelimit()) | |
750 | dev_warn(dev, "Bad rx response id %d.\n", | |
751 | rx->id); | |
752 | err = -EINVAL; | |
753 | goto next; | |
754 | } | |
755 | ||
756 | ret = gnttab_end_foreign_access_ref(ref, 0); | |
757 | BUG_ON(!ret); | |
758 | ||
2688fcb7 | 759 | gnttab_release_grant_reference(&queue->gref_rx_head, ref); |
0d160211 JF |
760 | |
761 | __skb_queue_tail(list, skb); | |
762 | ||
763 | next: | |
f942dc25 | 764 | if (!(rx->flags & XEN_NETRXF_more_data)) |
0d160211 JF |
765 | break; |
766 | ||
7158ff6d | 767 | if (cons + slots == rp) { |
0d160211 | 768 | if (net_ratelimit()) |
7158ff6d | 769 | dev_warn(dev, "Need more slots\n"); |
0d160211 JF |
770 | err = -ENOENT; |
771 | break; | |
772 | } | |
773 | ||
2688fcb7 AB |
774 | rx = RING_GET_RESPONSE(&queue->rx, cons + slots); |
775 | skb = xennet_get_rx_skb(queue, cons + slots); | |
776 | ref = xennet_get_rx_ref(queue, cons + slots); | |
7158ff6d | 777 | slots++; |
0d160211 JF |
778 | } |
779 | ||
7158ff6d | 780 | if (unlikely(slots > max)) { |
0d160211 | 781 | if (net_ratelimit()) |
697089dc | 782 | dev_warn(dev, "Too many slots\n"); |
0d160211 JF |
783 | err = -E2BIG; |
784 | } | |
785 | ||
786 | if (unlikely(err)) | |
2688fcb7 | 787 | queue->rx.rsp_cons = cons + slots; |
0d160211 JF |
788 | |
789 | return err; | |
790 | } | |
791 | ||
792 | static int xennet_set_skb_gso(struct sk_buff *skb, | |
793 | struct xen_netif_extra_info *gso) | |
794 | { | |
795 | if (!gso->u.gso.size) { | |
796 | if (net_ratelimit()) | |
383eda32 | 797 | pr_warn("GSO size must not be zero\n"); |
0d160211 JF |
798 | return -EINVAL; |
799 | } | |
800 | ||
2c0057de PD |
801 | if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4 && |
802 | gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV6) { | |
0d160211 | 803 | if (net_ratelimit()) |
383eda32 | 804 | pr_warn("Bad GSO type %d\n", gso->u.gso.type); |
0d160211 JF |
805 | return -EINVAL; |
806 | } | |
807 | ||
808 | skb_shinfo(skb)->gso_size = gso->u.gso.size; | |
2c0057de PD |
809 | skb_shinfo(skb)->gso_type = |
810 | (gso->u.gso.type == XEN_NETIF_GSO_TYPE_TCPV4) ? | |
811 | SKB_GSO_TCPV4 : | |
812 | SKB_GSO_TCPV6; | |
0d160211 JF |
813 | |
814 | /* Header must be checked, and gso_segs computed. */ | |
815 | skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; | |
816 | skb_shinfo(skb)->gso_segs = 0; | |
817 | ||
818 | return 0; | |
819 | } | |
820 | ||
2688fcb7 | 821 | static RING_IDX xennet_fill_frags(struct netfront_queue *queue, |
0d160211 JF |
822 | struct sk_buff *skb, |
823 | struct sk_buff_head *list) | |
824 | { | |
825 | struct skb_shared_info *shinfo = skb_shinfo(skb); | |
2688fcb7 | 826 | RING_IDX cons = queue->rx.rsp_cons; |
0d160211 JF |
827 | struct sk_buff *nskb; |
828 | ||
829 | while ((nskb = __skb_dequeue(list))) { | |
830 | struct xen_netif_rx_response *rx = | |
2688fcb7 | 831 | RING_GET_RESPONSE(&queue->rx, ++cons); |
01c68026 | 832 | skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0]; |
0d160211 | 833 | |
093b9c71 JB |
834 | if (shinfo->nr_frags == MAX_SKB_FRAGS) { |
835 | unsigned int pull_to = NETFRONT_SKB_CB(skb)->pull_to; | |
0d160211 | 836 | |
093b9c71 JB |
837 | BUG_ON(pull_to <= skb_headlen(skb)); |
838 | __pskb_pull_tail(skb, pull_to - skb_headlen(skb)); | |
839 | } | |
840 | BUG_ON(shinfo->nr_frags >= MAX_SKB_FRAGS); | |
841 | ||
842 | skb_add_rx_frag(skb, shinfo->nr_frags, skb_frag_page(nfrag), | |
843 | rx->offset, rx->status, PAGE_SIZE); | |
0d160211 JF |
844 | |
845 | skb_shinfo(nskb)->nr_frags = 0; | |
846 | kfree_skb(nskb); | |
0d160211 JF |
847 | } |
848 | ||
0d160211 JF |
849 | return cons; |
850 | } | |
851 | ||
e0ce4af9 | 852 | static int checksum_setup(struct net_device *dev, struct sk_buff *skb) |
0d160211 | 853 | { |
b5cf66cd | 854 | bool recalculate_partial_csum = false; |
e0ce4af9 IC |
855 | |
856 | /* | |
857 | * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy | |
858 | * peers can fail to set NETRXF_csum_blank when sending a GSO | |
859 | * frame. In this case force the SKB to CHECKSUM_PARTIAL and | |
860 | * recalculate the partial checksum. | |
861 | */ | |
862 | if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) { | |
863 | struct netfront_info *np = netdev_priv(dev); | |
2688fcb7 | 864 | atomic_inc(&np->rx_gso_checksum_fixup); |
e0ce4af9 | 865 | skb->ip_summed = CHECKSUM_PARTIAL; |
b5cf66cd | 866 | recalculate_partial_csum = true; |
e0ce4af9 IC |
867 | } |
868 | ||
869 | /* A non-CHECKSUM_PARTIAL SKB does not require setup. */ | |
870 | if (skb->ip_summed != CHECKSUM_PARTIAL) | |
871 | return 0; | |
0d160211 | 872 | |
b5cf66cd | 873 | return skb_checksum_setup(skb, recalculate_partial_csum); |
0d160211 JF |
874 | } |
875 | ||
2688fcb7 | 876 | static int handle_incoming_queue(struct netfront_queue *queue, |
09f75cd7 | 877 | struct sk_buff_head *rxq) |
0d160211 | 878 | { |
900e1833 | 879 | struct netfront_stats *rx_stats = this_cpu_ptr(queue->info->rx_stats); |
0d160211 JF |
880 | int packets_dropped = 0; |
881 | struct sk_buff *skb; | |
882 | ||
883 | while ((skb = __skb_dequeue(rxq)) != NULL) { | |
3683243b | 884 | int pull_to = NETFRONT_SKB_CB(skb)->pull_to; |
0d160211 | 885 | |
093b9c71 JB |
886 | if (pull_to > skb_headlen(skb)) |
887 | __pskb_pull_tail(skb, pull_to - skb_headlen(skb)); | |
0d160211 JF |
888 | |
889 | /* Ethernet work: Delayed to here as it peeks the header. */ | |
2688fcb7 | 890 | skb->protocol = eth_type_trans(skb, queue->info->netdev); |
d554f73d | 891 | skb_reset_network_header(skb); |
0d160211 | 892 | |
2688fcb7 | 893 | if (checksum_setup(queue->info->netdev, skb)) { |
e0ce4af9 IC |
894 | kfree_skb(skb); |
895 | packets_dropped++; | |
2688fcb7 | 896 | queue->info->netdev->stats.rx_errors++; |
e0ce4af9 | 897 | continue; |
0d160211 JF |
898 | } |
899 | ||
900e1833 DV |
900 | u64_stats_update_begin(&rx_stats->syncp); |
901 | rx_stats->packets++; | |
902 | rx_stats->bytes += skb->len; | |
903 | u64_stats_update_end(&rx_stats->syncp); | |
0d160211 JF |
904 | |
905 | /* Pass it up. */ | |
2688fcb7 | 906 | napi_gro_receive(&queue->napi, skb); |
0d160211 JF |
907 | } |
908 | ||
909 | return packets_dropped; | |
910 | } | |
911 | ||
bea3348e | 912 | static int xennet_poll(struct napi_struct *napi, int budget) |
0d160211 | 913 | { |
2688fcb7 AB |
914 | struct netfront_queue *queue = container_of(napi, struct netfront_queue, napi); |
915 | struct net_device *dev = queue->info->netdev; | |
0d160211 JF |
916 | struct sk_buff *skb; |
917 | struct netfront_rx_info rinfo; | |
918 | struct xen_netif_rx_response *rx = &rinfo.rx; | |
919 | struct xen_netif_extra_info *extras = rinfo.extras; | |
920 | RING_IDX i, rp; | |
bea3348e | 921 | int work_done; |
0d160211 JF |
922 | struct sk_buff_head rxq; |
923 | struct sk_buff_head errq; | |
924 | struct sk_buff_head tmpq; | |
0d160211 JF |
925 | int err; |
926 | ||
2688fcb7 | 927 | spin_lock(&queue->rx_lock); |
0d160211 | 928 | |
0d160211 JF |
929 | skb_queue_head_init(&rxq); |
930 | skb_queue_head_init(&errq); | |
931 | skb_queue_head_init(&tmpq); | |
932 | ||
2688fcb7 | 933 | rp = queue->rx.sring->rsp_prod; |
0d160211 JF |
934 | rmb(); /* Ensure we see queued responses up to 'rp'. */ |
935 | ||
2688fcb7 | 936 | i = queue->rx.rsp_cons; |
0d160211 JF |
937 | work_done = 0; |
938 | while ((i != rp) && (work_done < budget)) { | |
2688fcb7 | 939 | memcpy(rx, RING_GET_RESPONSE(&queue->rx, i), sizeof(*rx)); |
0d160211 JF |
940 | memset(extras, 0, sizeof(rinfo.extras)); |
941 | ||
2688fcb7 | 942 | err = xennet_get_responses(queue, &rinfo, rp, &tmpq); |
0d160211 JF |
943 | |
944 | if (unlikely(err)) { | |
945 | err: | |
946 | while ((skb = __skb_dequeue(&tmpq))) | |
947 | __skb_queue_tail(&errq, skb); | |
09f75cd7 | 948 | dev->stats.rx_errors++; |
2688fcb7 | 949 | i = queue->rx.rsp_cons; |
0d160211 JF |
950 | continue; |
951 | } | |
952 | ||
953 | skb = __skb_dequeue(&tmpq); | |
954 | ||
955 | if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) { | |
956 | struct xen_netif_extra_info *gso; | |
957 | gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1]; | |
958 | ||
959 | if (unlikely(xennet_set_skb_gso(skb, gso))) { | |
960 | __skb_queue_head(&tmpq, skb); | |
2688fcb7 | 961 | queue->rx.rsp_cons += skb_queue_len(&tmpq); |
0d160211 JF |
962 | goto err; |
963 | } | |
964 | } | |
965 | ||
3683243b IC |
966 | NETFRONT_SKB_CB(skb)->pull_to = rx->status; |
967 | if (NETFRONT_SKB_CB(skb)->pull_to > RX_COPY_THRESHOLD) | |
968 | NETFRONT_SKB_CB(skb)->pull_to = RX_COPY_THRESHOLD; | |
0d160211 | 969 | |
3683243b IC |
970 | skb_shinfo(skb)->frags[0].page_offset = rx->offset; |
971 | skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status); | |
972 | skb->data_len = rx->status; | |
093b9c71 | 973 | skb->len += rx->status; |
0d160211 | 974 | |
2688fcb7 | 975 | i = xennet_fill_frags(queue, skb, &tmpq); |
0d160211 | 976 | |
f942dc25 | 977 | if (rx->flags & XEN_NETRXF_csum_blank) |
0d160211 | 978 | skb->ip_summed = CHECKSUM_PARTIAL; |
f942dc25 | 979 | else if (rx->flags & XEN_NETRXF_data_validated) |
0d160211 JF |
980 | skb->ip_summed = CHECKSUM_UNNECESSARY; |
981 | ||
982 | __skb_queue_tail(&rxq, skb); | |
983 | ||
2688fcb7 | 984 | queue->rx.rsp_cons = ++i; |
0d160211 JF |
985 | work_done++; |
986 | } | |
987 | ||
56cfe5d0 | 988 | __skb_queue_purge(&errq); |
0d160211 | 989 | |
2688fcb7 | 990 | work_done -= handle_incoming_queue(queue, &rxq); |
0d160211 | 991 | |
2688fcb7 | 992 | xennet_alloc_rx_buffers(queue); |
0d160211 | 993 | |
0d160211 | 994 | if (work_done < budget) { |
bea3348e SH |
995 | int more_to_do = 0; |
996 | ||
6a6dc08f | 997 | napi_complete(napi); |
0d160211 | 998 | |
2688fcb7 | 999 | RING_FINAL_CHECK_FOR_RESPONSES(&queue->rx, more_to_do); |
6a6dc08f DV |
1000 | if (more_to_do) |
1001 | napi_schedule(napi); | |
0d160211 JF |
1002 | } |
1003 | ||
2688fcb7 | 1004 | spin_unlock(&queue->rx_lock); |
0d160211 | 1005 | |
bea3348e | 1006 | return work_done; |
0d160211 JF |
1007 | } |
1008 | ||
1009 | static int xennet_change_mtu(struct net_device *dev, int mtu) | |
1010 | { | |
9ecd1a75 WL |
1011 | int max = xennet_can_sg(dev) ? |
1012 | XEN_NETIF_MAX_TX_SIZE - MAX_TCP_HEADER : ETH_DATA_LEN; | |
0d160211 JF |
1013 | |
1014 | if (mtu > max) | |
1015 | return -EINVAL; | |
1016 | dev->mtu = mtu; | |
1017 | return 0; | |
1018 | } | |
1019 | ||
e00f85be | 1020 | static struct rtnl_link_stats64 *xennet_get_stats64(struct net_device *dev, |
1021 | struct rtnl_link_stats64 *tot) | |
1022 | { | |
1023 | struct netfront_info *np = netdev_priv(dev); | |
1024 | int cpu; | |
1025 | ||
1026 | for_each_possible_cpu(cpu) { | |
900e1833 DV |
1027 | struct netfront_stats *rx_stats = per_cpu_ptr(np->rx_stats, cpu); |
1028 | struct netfront_stats *tx_stats = per_cpu_ptr(np->tx_stats, cpu); | |
e00f85be | 1029 | u64 rx_packets, rx_bytes, tx_packets, tx_bytes; |
1030 | unsigned int start; | |
1031 | ||
1032 | do { | |
900e1833 DV |
1033 | start = u64_stats_fetch_begin_irq(&tx_stats->syncp); |
1034 | tx_packets = tx_stats->packets; | |
1035 | tx_bytes = tx_stats->bytes; | |
1036 | } while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start)); | |
e00f85be | 1037 | |
900e1833 DV |
1038 | do { |
1039 | start = u64_stats_fetch_begin_irq(&rx_stats->syncp); | |
1040 | rx_packets = rx_stats->packets; | |
1041 | rx_bytes = rx_stats->bytes; | |
1042 | } while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start)); | |
e00f85be | 1043 | |
1044 | tot->rx_packets += rx_packets; | |
1045 | tot->tx_packets += tx_packets; | |
1046 | tot->rx_bytes += rx_bytes; | |
1047 | tot->tx_bytes += tx_bytes; | |
1048 | } | |
1049 | ||
1050 | tot->rx_errors = dev->stats.rx_errors; | |
1051 | tot->tx_dropped = dev->stats.tx_dropped; | |
1052 | ||
1053 | return tot; | |
1054 | } | |
1055 | ||
2688fcb7 | 1056 | static void xennet_release_tx_bufs(struct netfront_queue *queue) |
0d160211 JF |
1057 | { |
1058 | struct sk_buff *skb; | |
1059 | int i; | |
1060 | ||
1061 | for (i = 0; i < NET_TX_RING_SIZE; i++) { | |
1062 | /* Skip over entries which are actually freelist references */ | |
2688fcb7 | 1063 | if (skb_entry_is_link(&queue->tx_skbs[i])) |
0d160211 JF |
1064 | continue; |
1065 | ||
2688fcb7 AB |
1066 | skb = queue->tx_skbs[i].skb; |
1067 | get_page(queue->grant_tx_page[i]); | |
1068 | gnttab_end_foreign_access(queue->grant_tx_ref[i], | |
cefe0078 | 1069 | GNTMAP_readonly, |
2688fcb7 AB |
1070 | (unsigned long)page_address(queue->grant_tx_page[i])); |
1071 | queue->grant_tx_page[i] = NULL; | |
1072 | queue->grant_tx_ref[i] = GRANT_INVALID_REF; | |
1073 | add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, i); | |
0d160211 JF |
1074 | dev_kfree_skb_irq(skb); |
1075 | } | |
1076 | } | |
1077 | ||
2688fcb7 | 1078 | static void xennet_release_rx_bufs(struct netfront_queue *queue) |
0d160211 | 1079 | { |
0d160211 JF |
1080 | int id, ref; |
1081 | ||
2688fcb7 | 1082 | spin_lock_bh(&queue->rx_lock); |
0d160211 JF |
1083 | |
1084 | for (id = 0; id < NET_RX_RING_SIZE; id++) { | |
cefe0078 AL |
1085 | struct sk_buff *skb; |
1086 | struct page *page; | |
0d160211 | 1087 | |
2688fcb7 | 1088 | skb = queue->rx_skbs[id]; |
cefe0078 | 1089 | if (!skb) |
0d160211 | 1090 | continue; |
0d160211 | 1091 | |
2688fcb7 | 1092 | ref = queue->grant_rx_ref[id]; |
cefe0078 AL |
1093 | if (ref == GRANT_INVALID_REF) |
1094 | continue; | |
0d160211 | 1095 | |
cefe0078 | 1096 | page = skb_frag_page(&skb_shinfo(skb)->frags[0]); |
0d160211 | 1097 | |
cefe0078 AL |
1098 | /* gnttab_end_foreign_access() needs a page ref until |
1099 | * foreign access is ended (which may be deferred). | |
1100 | */ | |
1101 | get_page(page); | |
1102 | gnttab_end_foreign_access(ref, 0, | |
1103 | (unsigned long)page_address(page)); | |
2688fcb7 | 1104 | queue->grant_rx_ref[id] = GRANT_INVALID_REF; |
0d160211 | 1105 | |
cefe0078 | 1106 | kfree_skb(skb); |
0d160211 JF |
1107 | } |
1108 | ||
2688fcb7 | 1109 | spin_unlock_bh(&queue->rx_lock); |
0d160211 JF |
1110 | } |
1111 | ||
c8f44aff MM |
1112 | static netdev_features_t xennet_fix_features(struct net_device *dev, |
1113 | netdev_features_t features) | |
8f7b01a1 ED |
1114 | { |
1115 | struct netfront_info *np = netdev_priv(dev); | |
1116 | int val; | |
1117 | ||
1118 | if (features & NETIF_F_SG) { | |
1119 | if (xenbus_scanf(XBT_NIL, np->xbdev->otherend, "feature-sg", | |
1120 | "%d", &val) < 0) | |
1121 | val = 0; | |
1122 | ||
1123 | if (!val) | |
1124 | features &= ~NETIF_F_SG; | |
1125 | } | |
1126 | ||
2c0057de PD |
1127 | if (features & NETIF_F_IPV6_CSUM) { |
1128 | if (xenbus_scanf(XBT_NIL, np->xbdev->otherend, | |
1129 | "feature-ipv6-csum-offload", "%d", &val) < 0) | |
1130 | val = 0; | |
1131 | ||
1132 | if (!val) | |
1133 | features &= ~NETIF_F_IPV6_CSUM; | |
1134 | } | |
1135 | ||
8f7b01a1 ED |
1136 | if (features & NETIF_F_TSO) { |
1137 | if (xenbus_scanf(XBT_NIL, np->xbdev->otherend, | |
1138 | "feature-gso-tcpv4", "%d", &val) < 0) | |
1139 | val = 0; | |
1140 | ||
1141 | if (!val) | |
1142 | features &= ~NETIF_F_TSO; | |
1143 | } | |
1144 | ||
2c0057de PD |
1145 | if (features & NETIF_F_TSO6) { |
1146 | if (xenbus_scanf(XBT_NIL, np->xbdev->otherend, | |
1147 | "feature-gso-tcpv6", "%d", &val) < 0) | |
1148 | val = 0; | |
1149 | ||
1150 | if (!val) | |
1151 | features &= ~NETIF_F_TSO6; | |
1152 | } | |
1153 | ||
8f7b01a1 ED |
1154 | return features; |
1155 | } | |
1156 | ||
c8f44aff MM |
1157 | static int xennet_set_features(struct net_device *dev, |
1158 | netdev_features_t features) | |
8f7b01a1 ED |
1159 | { |
1160 | if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) { | |
1161 | netdev_info(dev, "Reducing MTU because no SG offload"); | |
1162 | dev->mtu = ETH_DATA_LEN; | |
1163 | } | |
1164 | ||
1165 | return 0; | |
1166 | } | |
1167 | ||
d634bf2c | 1168 | static irqreturn_t xennet_tx_interrupt(int irq, void *dev_id) |
cf66f9d4 | 1169 | { |
2688fcb7 | 1170 | struct netfront_queue *queue = dev_id; |
cf66f9d4 KRW |
1171 | unsigned long flags; |
1172 | ||
2688fcb7 AB |
1173 | spin_lock_irqsave(&queue->tx_lock, flags); |
1174 | xennet_tx_buf_gc(queue); | |
1175 | spin_unlock_irqrestore(&queue->tx_lock, flags); | |
cf66f9d4 | 1176 | |
d634bf2c WL |
1177 | return IRQ_HANDLED; |
1178 | } | |
1179 | ||
1180 | static irqreturn_t xennet_rx_interrupt(int irq, void *dev_id) | |
1181 | { | |
2688fcb7 AB |
1182 | struct netfront_queue *queue = dev_id; |
1183 | struct net_device *dev = queue->info->netdev; | |
d634bf2c WL |
1184 | |
1185 | if (likely(netif_carrier_ok(dev) && | |
2688fcb7 | 1186 | RING_HAS_UNCONSUMED_RESPONSES(&queue->rx))) |
76541869 | 1187 | napi_schedule(&queue->napi); |
cf66f9d4 | 1188 | |
d634bf2c WL |
1189 | return IRQ_HANDLED; |
1190 | } | |
cf66f9d4 | 1191 | |
d634bf2c WL |
1192 | static irqreturn_t xennet_interrupt(int irq, void *dev_id) |
1193 | { | |
1194 | xennet_tx_interrupt(irq, dev_id); | |
1195 | xennet_rx_interrupt(irq, dev_id); | |
cf66f9d4 KRW |
1196 | return IRQ_HANDLED; |
1197 | } | |
1198 | ||
1199 | #ifdef CONFIG_NET_POLL_CONTROLLER | |
1200 | static void xennet_poll_controller(struct net_device *dev) | |
1201 | { | |
2688fcb7 AB |
1202 | /* Poll each queue */ |
1203 | struct netfront_info *info = netdev_priv(dev); | |
1204 | unsigned int num_queues = dev->real_num_tx_queues; | |
1205 | unsigned int i; | |
1206 | for (i = 0; i < num_queues; ++i) | |
1207 | xennet_interrupt(0, &info->queues[i]); | |
cf66f9d4 KRW |
1208 | } |
1209 | #endif | |
1210 | ||
0a0b9d2e SH |
1211 | static const struct net_device_ops xennet_netdev_ops = { |
1212 | .ndo_open = xennet_open, | |
0a0b9d2e SH |
1213 | .ndo_stop = xennet_close, |
1214 | .ndo_start_xmit = xennet_start_xmit, | |
1215 | .ndo_change_mtu = xennet_change_mtu, | |
e00f85be | 1216 | .ndo_get_stats64 = xennet_get_stats64, |
0a0b9d2e SH |
1217 | .ndo_set_mac_address = eth_mac_addr, |
1218 | .ndo_validate_addr = eth_validate_addr, | |
fb507934 MM |
1219 | .ndo_fix_features = xennet_fix_features, |
1220 | .ndo_set_features = xennet_set_features, | |
2688fcb7 | 1221 | .ndo_select_queue = xennet_select_queue, |
cf66f9d4 KRW |
1222 | #ifdef CONFIG_NET_POLL_CONTROLLER |
1223 | .ndo_poll_controller = xennet_poll_controller, | |
1224 | #endif | |
0a0b9d2e SH |
1225 | }; |
1226 | ||
900e1833 DV |
1227 | static void xennet_free_netdev(struct net_device *netdev) |
1228 | { | |
1229 | struct netfront_info *np = netdev_priv(netdev); | |
1230 | ||
1231 | free_percpu(np->rx_stats); | |
1232 | free_percpu(np->tx_stats); | |
1233 | free_netdev(netdev); | |
1234 | } | |
1235 | ||
8e0e46bb | 1236 | static struct net_device *xennet_create_dev(struct xenbus_device *dev) |
0d160211 | 1237 | { |
2688fcb7 | 1238 | int err; |
0d160211 JF |
1239 | struct net_device *netdev; |
1240 | struct netfront_info *np; | |
1241 | ||
50ee6061 | 1242 | netdev = alloc_etherdev_mq(sizeof(struct netfront_info), xennet_max_queues); |
41de8d4c | 1243 | if (!netdev) |
0d160211 | 1244 | return ERR_PTR(-ENOMEM); |
0d160211 JF |
1245 | |
1246 | np = netdev_priv(netdev); | |
1247 | np->xbdev = dev; | |
1248 | ||
2688fcb7 AB |
1249 | /* No need to use rtnl_lock() before the call below as it |
1250 | * happens before register_netdev(). | |
1251 | */ | |
1252 | netif_set_real_num_tx_queues(netdev, 0); | |
1253 | np->queues = NULL; | |
0d160211 | 1254 | |
e00f85be | 1255 | err = -ENOMEM; |
900e1833 DV |
1256 | np->rx_stats = netdev_alloc_pcpu_stats(struct netfront_stats); |
1257 | if (np->rx_stats == NULL) | |
1258 | goto exit; | |
1259 | np->tx_stats = netdev_alloc_pcpu_stats(struct netfront_stats); | |
1260 | if (np->tx_stats == NULL) | |
e00f85be | 1261 | goto exit; |
1262 | ||
0a0b9d2e SH |
1263 | netdev->netdev_ops = &xennet_netdev_ops; |
1264 | ||
fb507934 MM |
1265 | netdev->features = NETIF_F_IP_CSUM | NETIF_F_RXCSUM | |
1266 | NETIF_F_GSO_ROBUST; | |
2c0057de PD |
1267 | netdev->hw_features = NETIF_F_SG | |
1268 | NETIF_F_IPV6_CSUM | | |
1269 | NETIF_F_TSO | NETIF_F_TSO6; | |
0d160211 | 1270 | |
fc3e5941 IC |
1271 | /* |
1272 | * Assume that all hw features are available for now. This set | |
1273 | * will be adjusted by the call to netdev_update_features() in | |
1274 | * xennet_connect() which is the earliest point where we can | |
1275 | * negotiate with the backend regarding supported features. | |
1276 | */ | |
1277 | netdev->features |= netdev->hw_features; | |
1278 | ||
7ad24ea4 | 1279 | netdev->ethtool_ops = &xennet_ethtool_ops; |
0d160211 JF |
1280 | SET_NETDEV_DEV(netdev, &dev->dev); |
1281 | ||
9ecd1a75 WL |
1282 | netif_set_gso_max_size(netdev, XEN_NETIF_MAX_TX_SIZE - MAX_TCP_HEADER); |
1283 | ||
0d160211 JF |
1284 | np->netdev = netdev; |
1285 | ||
1286 | netif_carrier_off(netdev); | |
1287 | ||
1288 | return netdev; | |
1289 | ||
0d160211 | 1290 | exit: |
900e1833 | 1291 | xennet_free_netdev(netdev); |
0d160211 JF |
1292 | return ERR_PTR(err); |
1293 | } | |
1294 | ||
1295 | /** | |
1296 | * Entry point to this code when a new device is created. Allocate the basic | |
1297 | * structures and the ring buffers for communication with the backend, and | |
1298 | * inform the backend of the appropriate details for those. | |
1299 | */ | |
8e0e46bb | 1300 | static int netfront_probe(struct xenbus_device *dev, |
1dd06ae8 | 1301 | const struct xenbus_device_id *id) |
0d160211 JF |
1302 | { |
1303 | int err; | |
1304 | struct net_device *netdev; | |
1305 | struct netfront_info *info; | |
1306 | ||
1307 | netdev = xennet_create_dev(dev); | |
1308 | if (IS_ERR(netdev)) { | |
1309 | err = PTR_ERR(netdev); | |
1310 | xenbus_dev_fatal(dev, err, "creating netdev"); | |
1311 | return err; | |
1312 | } | |
1313 | ||
1314 | info = netdev_priv(netdev); | |
1b713e00 | 1315 | dev_set_drvdata(&dev->dev, info); |
27b917e5 TI |
1316 | #ifdef CONFIG_SYSFS |
1317 | info->netdev->sysfs_groups[0] = &xennet_dev_group; | |
1318 | #endif | |
0d160211 JF |
1319 | err = register_netdev(info->netdev); |
1320 | if (err) { | |
383eda32 | 1321 | pr_warn("%s: register_netdev err=%d\n", __func__, err); |
0d160211 JF |
1322 | goto fail; |
1323 | } | |
1324 | ||
0d160211 JF |
1325 | return 0; |
1326 | ||
1327 | fail: | |
900e1833 | 1328 | xennet_free_netdev(netdev); |
1b713e00 | 1329 | dev_set_drvdata(&dev->dev, NULL); |
0d160211 JF |
1330 | return err; |
1331 | } | |
1332 | ||
1333 | static void xennet_end_access(int ref, void *page) | |
1334 | { | |
1335 | /* This frees the page as a side-effect */ | |
1336 | if (ref != GRANT_INVALID_REF) | |
1337 | gnttab_end_foreign_access(ref, 0, (unsigned long)page); | |
1338 | } | |
1339 | ||
1340 | static void xennet_disconnect_backend(struct netfront_info *info) | |
1341 | { | |
2688fcb7 | 1342 | unsigned int i = 0; |
2688fcb7 AB |
1343 | unsigned int num_queues = info->netdev->real_num_tx_queues; |
1344 | ||
f9feb1e6 DV |
1345 | netif_carrier_off(info->netdev); |
1346 | ||
2688fcb7 | 1347 | for (i = 0; i < num_queues; ++i) { |
76541869 DV |
1348 | struct netfront_queue *queue = &info->queues[i]; |
1349 | ||
2688fcb7 AB |
1350 | if (queue->tx_irq && (queue->tx_irq == queue->rx_irq)) |
1351 | unbind_from_irqhandler(queue->tx_irq, queue); | |
1352 | if (queue->tx_irq && (queue->tx_irq != queue->rx_irq)) { | |
1353 | unbind_from_irqhandler(queue->tx_irq, queue); | |
1354 | unbind_from_irqhandler(queue->rx_irq, queue); | |
1355 | } | |
1356 | queue->tx_evtchn = queue->rx_evtchn = 0; | |
1357 | queue->tx_irq = queue->rx_irq = 0; | |
0d160211 | 1358 | |
f9feb1e6 DV |
1359 | napi_synchronize(&queue->napi); |
1360 | ||
a5b5dc3c DV |
1361 | xennet_release_tx_bufs(queue); |
1362 | xennet_release_rx_bufs(queue); | |
1363 | gnttab_free_grant_references(queue->gref_tx_head); | |
1364 | gnttab_free_grant_references(queue->gref_rx_head); | |
1365 | ||
2688fcb7 AB |
1366 | /* End access and free the pages */ |
1367 | xennet_end_access(queue->tx_ring_ref, queue->tx.sring); | |
1368 | xennet_end_access(queue->rx_ring_ref, queue->rx.sring); | |
0d160211 | 1369 | |
2688fcb7 AB |
1370 | queue->tx_ring_ref = GRANT_INVALID_REF; |
1371 | queue->rx_ring_ref = GRANT_INVALID_REF; | |
1372 | queue->tx.sring = NULL; | |
1373 | queue->rx.sring = NULL; | |
1374 | } | |
0d160211 JF |
1375 | } |
1376 | ||
1377 | /** | |
1378 | * We are reconnecting to the backend, due to a suspend/resume, or a backend | |
1379 | * driver restart. We tear down our netif structure and recreate it, but | |
1380 | * leave the device-layer structures intact so that this is transparent to the | |
1381 | * rest of the kernel. | |
1382 | */ | |
1383 | static int netfront_resume(struct xenbus_device *dev) | |
1384 | { | |
1b713e00 | 1385 | struct netfront_info *info = dev_get_drvdata(&dev->dev); |
0d160211 JF |
1386 | |
1387 | dev_dbg(&dev->dev, "%s\n", dev->nodename); | |
1388 | ||
1389 | xennet_disconnect_backend(info); | |
1390 | return 0; | |
1391 | } | |
1392 | ||
1393 | static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[]) | |
1394 | { | |
1395 | char *s, *e, *macstr; | |
1396 | int i; | |
1397 | ||
1398 | macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL); | |
1399 | if (IS_ERR(macstr)) | |
1400 | return PTR_ERR(macstr); | |
1401 | ||
1402 | for (i = 0; i < ETH_ALEN; i++) { | |
1403 | mac[i] = simple_strtoul(s, &e, 16); | |
1404 | if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) { | |
1405 | kfree(macstr); | |
1406 | return -ENOENT; | |
1407 | } | |
1408 | s = e+1; | |
1409 | } | |
1410 | ||
1411 | kfree(macstr); | |
1412 | return 0; | |
1413 | } | |
1414 | ||
2688fcb7 | 1415 | static int setup_netfront_single(struct netfront_queue *queue) |
d634bf2c WL |
1416 | { |
1417 | int err; | |
1418 | ||
2688fcb7 | 1419 | err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn); |
d634bf2c WL |
1420 | if (err < 0) |
1421 | goto fail; | |
1422 | ||
2688fcb7 | 1423 | err = bind_evtchn_to_irqhandler(queue->tx_evtchn, |
d634bf2c | 1424 | xennet_interrupt, |
2688fcb7 | 1425 | 0, queue->info->netdev->name, queue); |
d634bf2c WL |
1426 | if (err < 0) |
1427 | goto bind_fail; | |
2688fcb7 AB |
1428 | queue->rx_evtchn = queue->tx_evtchn; |
1429 | queue->rx_irq = queue->tx_irq = err; | |
d634bf2c WL |
1430 | |
1431 | return 0; | |
1432 | ||
1433 | bind_fail: | |
2688fcb7 AB |
1434 | xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn); |
1435 | queue->tx_evtchn = 0; | |
d634bf2c WL |
1436 | fail: |
1437 | return err; | |
1438 | } | |
1439 | ||
2688fcb7 | 1440 | static int setup_netfront_split(struct netfront_queue *queue) |
d634bf2c WL |
1441 | { |
1442 | int err; | |
1443 | ||
2688fcb7 | 1444 | err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn); |
d634bf2c WL |
1445 | if (err < 0) |
1446 | goto fail; | |
2688fcb7 | 1447 | err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->rx_evtchn); |
d634bf2c WL |
1448 | if (err < 0) |
1449 | goto alloc_rx_evtchn_fail; | |
1450 | ||
2688fcb7 AB |
1451 | snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name), |
1452 | "%s-tx", queue->name); | |
1453 | err = bind_evtchn_to_irqhandler(queue->tx_evtchn, | |
d634bf2c | 1454 | xennet_tx_interrupt, |
2688fcb7 | 1455 | 0, queue->tx_irq_name, queue); |
d634bf2c WL |
1456 | if (err < 0) |
1457 | goto bind_tx_fail; | |
2688fcb7 | 1458 | queue->tx_irq = err; |
d634bf2c | 1459 | |
2688fcb7 AB |
1460 | snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name), |
1461 | "%s-rx", queue->name); | |
1462 | err = bind_evtchn_to_irqhandler(queue->rx_evtchn, | |
d634bf2c | 1463 | xennet_rx_interrupt, |
2688fcb7 | 1464 | 0, queue->rx_irq_name, queue); |
d634bf2c WL |
1465 | if (err < 0) |
1466 | goto bind_rx_fail; | |
2688fcb7 | 1467 | queue->rx_irq = err; |
d634bf2c WL |
1468 | |
1469 | return 0; | |
1470 | ||
1471 | bind_rx_fail: | |
2688fcb7 AB |
1472 | unbind_from_irqhandler(queue->tx_irq, queue); |
1473 | queue->tx_irq = 0; | |
d634bf2c | 1474 | bind_tx_fail: |
2688fcb7 AB |
1475 | xenbus_free_evtchn(queue->info->xbdev, queue->rx_evtchn); |
1476 | queue->rx_evtchn = 0; | |
d634bf2c | 1477 | alloc_rx_evtchn_fail: |
2688fcb7 AB |
1478 | xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn); |
1479 | queue->tx_evtchn = 0; | |
d634bf2c WL |
1480 | fail: |
1481 | return err; | |
1482 | } | |
1483 | ||
2688fcb7 AB |
1484 | static int setup_netfront(struct xenbus_device *dev, |
1485 | struct netfront_queue *queue, unsigned int feature_split_evtchn) | |
0d160211 JF |
1486 | { |
1487 | struct xen_netif_tx_sring *txs; | |
1488 | struct xen_netif_rx_sring *rxs; | |
1489 | int err; | |
0d160211 | 1490 | |
2688fcb7 AB |
1491 | queue->tx_ring_ref = GRANT_INVALID_REF; |
1492 | queue->rx_ring_ref = GRANT_INVALID_REF; | |
1493 | queue->rx.sring = NULL; | |
1494 | queue->tx.sring = NULL; | |
0d160211 | 1495 | |
a144ff09 | 1496 | txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH); |
0d160211 JF |
1497 | if (!txs) { |
1498 | err = -ENOMEM; | |
1499 | xenbus_dev_fatal(dev, err, "allocating tx ring page"); | |
1500 | goto fail; | |
1501 | } | |
1502 | SHARED_RING_INIT(txs); | |
2688fcb7 | 1503 | FRONT_RING_INIT(&queue->tx, txs, PAGE_SIZE); |
0d160211 JF |
1504 | |
1505 | err = xenbus_grant_ring(dev, virt_to_mfn(txs)); | |
1ca2983a WL |
1506 | if (err < 0) |
1507 | goto grant_tx_ring_fail; | |
2688fcb7 | 1508 | queue->tx_ring_ref = err; |
0d160211 | 1509 | |
a144ff09 | 1510 | rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH); |
0d160211 JF |
1511 | if (!rxs) { |
1512 | err = -ENOMEM; | |
1513 | xenbus_dev_fatal(dev, err, "allocating rx ring page"); | |
1ca2983a | 1514 | goto alloc_rx_ring_fail; |
0d160211 JF |
1515 | } |
1516 | SHARED_RING_INIT(rxs); | |
2688fcb7 | 1517 | FRONT_RING_INIT(&queue->rx, rxs, PAGE_SIZE); |
0d160211 JF |
1518 | |
1519 | err = xenbus_grant_ring(dev, virt_to_mfn(rxs)); | |
1ca2983a WL |
1520 | if (err < 0) |
1521 | goto grant_rx_ring_fail; | |
2688fcb7 | 1522 | queue->rx_ring_ref = err; |
0d160211 | 1523 | |
d634bf2c | 1524 | if (feature_split_evtchn) |
2688fcb7 | 1525 | err = setup_netfront_split(queue); |
d634bf2c WL |
1526 | /* setup single event channel if |
1527 | * a) feature-split-event-channels == 0 | |
1528 | * b) feature-split-event-channels == 1 but failed to setup | |
1529 | */ | |
1530 | if (!feature_split_evtchn || (feature_split_evtchn && err)) | |
2688fcb7 | 1531 | err = setup_netfront_single(queue); |
d634bf2c | 1532 | |
0d160211 | 1533 | if (err) |
1ca2983a | 1534 | goto alloc_evtchn_fail; |
0d160211 | 1535 | |
0d160211 JF |
1536 | return 0; |
1537 | ||
1ca2983a WL |
1538 | /* If we fail to setup netfront, it is safe to just revoke access to |
1539 | * granted pages because backend is not accessing it at this point. | |
1540 | */ | |
1ca2983a | 1541 | alloc_evtchn_fail: |
2688fcb7 | 1542 | gnttab_end_foreign_access_ref(queue->rx_ring_ref, 0); |
1ca2983a WL |
1543 | grant_rx_ring_fail: |
1544 | free_page((unsigned long)rxs); | |
1545 | alloc_rx_ring_fail: | |
2688fcb7 | 1546 | gnttab_end_foreign_access_ref(queue->tx_ring_ref, 0); |
1ca2983a WL |
1547 | grant_tx_ring_fail: |
1548 | free_page((unsigned long)txs); | |
1549 | fail: | |
0d160211 JF |
1550 | return err; |
1551 | } | |
1552 | ||
2688fcb7 AB |
1553 | /* Queue-specific initialisation |
1554 | * This used to be done in xennet_create_dev() but must now | |
1555 | * be run per-queue. | |
1556 | */ | |
1557 | static int xennet_init_queue(struct netfront_queue *queue) | |
1558 | { | |
1559 | unsigned short i; | |
1560 | int err = 0; | |
1561 | ||
1562 | spin_lock_init(&queue->tx_lock); | |
1563 | spin_lock_init(&queue->rx_lock); | |
1564 | ||
2688fcb7 AB |
1565 | init_timer(&queue->rx_refill_timer); |
1566 | queue->rx_refill_timer.data = (unsigned long)queue; | |
1567 | queue->rx_refill_timer.function = rx_refill_timeout; | |
1568 | ||
8b715010 WL |
1569 | snprintf(queue->name, sizeof(queue->name), "%s-q%u", |
1570 | queue->info->netdev->name, queue->id); | |
1571 | ||
2688fcb7 AB |
1572 | /* Initialise tx_skbs as a free chain containing every entry. */ |
1573 | queue->tx_skb_freelist = 0; | |
1574 | for (i = 0; i < NET_TX_RING_SIZE; i++) { | |
1575 | skb_entry_set_link(&queue->tx_skbs[i], i+1); | |
1576 | queue->grant_tx_ref[i] = GRANT_INVALID_REF; | |
1577 | queue->grant_tx_page[i] = NULL; | |
1578 | } | |
1579 | ||
1580 | /* Clear out rx_skbs */ | |
1581 | for (i = 0; i < NET_RX_RING_SIZE; i++) { | |
1582 | queue->rx_skbs[i] = NULL; | |
1583 | queue->grant_rx_ref[i] = GRANT_INVALID_REF; | |
1584 | } | |
1585 | ||
1586 | /* A grant for every tx ring slot */ | |
1f3c2eba | 1587 | if (gnttab_alloc_grant_references(NET_TX_RING_SIZE, |
2688fcb7 AB |
1588 | &queue->gref_tx_head) < 0) { |
1589 | pr_alert("can't alloc tx grant refs\n"); | |
1590 | err = -ENOMEM; | |
1591 | goto exit; | |
1592 | } | |
1593 | ||
1594 | /* A grant for every rx ring slot */ | |
1f3c2eba | 1595 | if (gnttab_alloc_grant_references(NET_RX_RING_SIZE, |
2688fcb7 AB |
1596 | &queue->gref_rx_head) < 0) { |
1597 | pr_alert("can't alloc rx grant refs\n"); | |
1598 | err = -ENOMEM; | |
1599 | goto exit_free_tx; | |
1600 | } | |
1601 | ||
2688fcb7 AB |
1602 | return 0; |
1603 | ||
1604 | exit_free_tx: | |
1605 | gnttab_free_grant_references(queue->gref_tx_head); | |
1606 | exit: | |
1607 | return err; | |
1608 | } | |
1609 | ||
50ee6061 AB |
1610 | static int write_queue_xenstore_keys(struct netfront_queue *queue, |
1611 | struct xenbus_transaction *xbt, int write_hierarchical) | |
1612 | { | |
1613 | /* Write the queue-specific keys into XenStore in the traditional | |
1614 | * way for a single queue, or in a queue subkeys for multiple | |
1615 | * queues. | |
1616 | */ | |
1617 | struct xenbus_device *dev = queue->info->xbdev; | |
1618 | int err; | |
1619 | const char *message; | |
1620 | char *path; | |
1621 | size_t pathsize; | |
1622 | ||
1623 | /* Choose the correct place to write the keys */ | |
1624 | if (write_hierarchical) { | |
1625 | pathsize = strlen(dev->nodename) + 10; | |
1626 | path = kzalloc(pathsize, GFP_KERNEL); | |
1627 | if (!path) { | |
1628 | err = -ENOMEM; | |
1629 | message = "out of memory while writing ring references"; | |
1630 | goto error; | |
1631 | } | |
1632 | snprintf(path, pathsize, "%s/queue-%u", | |
1633 | dev->nodename, queue->id); | |
1634 | } else { | |
1635 | path = (char *)dev->nodename; | |
1636 | } | |
1637 | ||
1638 | /* Write ring references */ | |
1639 | err = xenbus_printf(*xbt, path, "tx-ring-ref", "%u", | |
1640 | queue->tx_ring_ref); | |
1641 | if (err) { | |
1642 | message = "writing tx-ring-ref"; | |
1643 | goto error; | |
1644 | } | |
1645 | ||
1646 | err = xenbus_printf(*xbt, path, "rx-ring-ref", "%u", | |
1647 | queue->rx_ring_ref); | |
1648 | if (err) { | |
1649 | message = "writing rx-ring-ref"; | |
1650 | goto error; | |
1651 | } | |
1652 | ||
1653 | /* Write event channels; taking into account both shared | |
1654 | * and split event channel scenarios. | |
1655 | */ | |
1656 | if (queue->tx_evtchn == queue->rx_evtchn) { | |
1657 | /* Shared event channel */ | |
1658 | err = xenbus_printf(*xbt, path, | |
1659 | "event-channel", "%u", queue->tx_evtchn); | |
1660 | if (err) { | |
1661 | message = "writing event-channel"; | |
1662 | goto error; | |
1663 | } | |
1664 | } else { | |
1665 | /* Split event channels */ | |
1666 | err = xenbus_printf(*xbt, path, | |
1667 | "event-channel-tx", "%u", queue->tx_evtchn); | |
1668 | if (err) { | |
1669 | message = "writing event-channel-tx"; | |
1670 | goto error; | |
1671 | } | |
1672 | ||
1673 | err = xenbus_printf(*xbt, path, | |
1674 | "event-channel-rx", "%u", queue->rx_evtchn); | |
1675 | if (err) { | |
1676 | message = "writing event-channel-rx"; | |
1677 | goto error; | |
1678 | } | |
1679 | } | |
1680 | ||
1681 | if (write_hierarchical) | |
1682 | kfree(path); | |
1683 | return 0; | |
1684 | ||
1685 | error: | |
1686 | if (write_hierarchical) | |
1687 | kfree(path); | |
1688 | xenbus_dev_fatal(dev, err, "%s", message); | |
1689 | return err; | |
1690 | } | |
1691 | ||
ce58725f DV |
1692 | static void xennet_destroy_queues(struct netfront_info *info) |
1693 | { | |
1694 | unsigned int i; | |
1695 | ||
1696 | rtnl_lock(); | |
1697 | ||
1698 | for (i = 0; i < info->netdev->real_num_tx_queues; i++) { | |
1699 | struct netfront_queue *queue = &info->queues[i]; | |
1700 | ||
1701 | if (netif_running(info->netdev)) | |
1702 | napi_disable(&queue->napi); | |
1703 | netif_napi_del(&queue->napi); | |
1704 | } | |
1705 | ||
1706 | rtnl_unlock(); | |
1707 | ||
1708 | kfree(info->queues); | |
1709 | info->queues = NULL; | |
1710 | } | |
1711 | ||
1712 | static int xennet_create_queues(struct netfront_info *info, | |
1713 | unsigned int num_queues) | |
1714 | { | |
1715 | unsigned int i; | |
1716 | int ret; | |
1717 | ||
1718 | info->queues = kcalloc(num_queues, sizeof(struct netfront_queue), | |
1719 | GFP_KERNEL); | |
1720 | if (!info->queues) | |
1721 | return -ENOMEM; | |
1722 | ||
1723 | rtnl_lock(); | |
1724 | ||
1725 | for (i = 0; i < num_queues; i++) { | |
1726 | struct netfront_queue *queue = &info->queues[i]; | |
1727 | ||
1728 | queue->id = i; | |
1729 | queue->info = info; | |
1730 | ||
1731 | ret = xennet_init_queue(queue); | |
1732 | if (ret < 0) { | |
69cb8524 DV |
1733 | dev_warn(&info->netdev->dev, |
1734 | "only created %d queues\n", i); | |
ce58725f DV |
1735 | num_queues = i; |
1736 | break; | |
1737 | } | |
1738 | ||
1739 | netif_napi_add(queue->info->netdev, &queue->napi, | |
1740 | xennet_poll, 64); | |
1741 | if (netif_running(info->netdev)) | |
1742 | napi_enable(&queue->napi); | |
1743 | } | |
1744 | ||
1745 | netif_set_real_num_tx_queues(info->netdev, num_queues); | |
1746 | ||
1747 | rtnl_unlock(); | |
1748 | ||
1749 | if (num_queues == 0) { | |
1750 | dev_err(&info->netdev->dev, "no queues\n"); | |
1751 | return -EINVAL; | |
1752 | } | |
1753 | return 0; | |
1754 | } | |
1755 | ||
0d160211 | 1756 | /* Common code used when first setting up, and when resuming. */ |
f502bf2b | 1757 | static int talk_to_netback(struct xenbus_device *dev, |
0d160211 JF |
1758 | struct netfront_info *info) |
1759 | { | |
1760 | const char *message; | |
1761 | struct xenbus_transaction xbt; | |
1762 | int err; | |
2688fcb7 AB |
1763 | unsigned int feature_split_evtchn; |
1764 | unsigned int i = 0; | |
50ee6061 | 1765 | unsigned int max_queues = 0; |
2688fcb7 AB |
1766 | struct netfront_queue *queue = NULL; |
1767 | unsigned int num_queues = 1; | |
0d160211 | 1768 | |
2688fcb7 AB |
1769 | info->netdev->irq = 0; |
1770 | ||
50ee6061 AB |
1771 | /* Check if backend supports multiple queues */ |
1772 | err = xenbus_scanf(XBT_NIL, info->xbdev->otherend, | |
1773 | "multi-queue-max-queues", "%u", &max_queues); | |
1774 | if (err < 0) | |
1775 | max_queues = 1; | |
1776 | num_queues = min(max_queues, xennet_max_queues); | |
1777 | ||
2688fcb7 AB |
1778 | /* Check feature-split-event-channels */ |
1779 | err = xenbus_scanf(XBT_NIL, info->xbdev->otherend, | |
1780 | "feature-split-event-channels", "%u", | |
1781 | &feature_split_evtchn); | |
1782 | if (err < 0) | |
1783 | feature_split_evtchn = 0; | |
1784 | ||
1785 | /* Read mac addr. */ | |
1786 | err = xen_net_read_mac(dev, info->netdev->dev_addr); | |
1787 | if (err) { | |
1788 | xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename); | |
1789 | goto out; | |
1790 | } | |
1791 | ||
ce58725f DV |
1792 | if (info->queues) |
1793 | xennet_destroy_queues(info); | |
1794 | ||
1795 | err = xennet_create_queues(info, num_queues); | |
1796 | if (err < 0) | |
1797 | goto destroy_ring; | |
2688fcb7 AB |
1798 | |
1799 | /* Create shared ring, alloc event channel -- for each queue */ | |
1800 | for (i = 0; i < num_queues; ++i) { | |
1801 | queue = &info->queues[i]; | |
2688fcb7 AB |
1802 | err = setup_netfront(dev, queue, feature_split_evtchn); |
1803 | if (err) { | |
ce58725f DV |
1804 | /* setup_netfront() will tidy up the current |
1805 | * queue on error, but we need to clean up | |
2688fcb7 AB |
1806 | * those already allocated. |
1807 | */ | |
1808 | if (i > 0) { | |
1809 | rtnl_lock(); | |
1810 | netif_set_real_num_tx_queues(info->netdev, i); | |
1811 | rtnl_unlock(); | |
1812 | goto destroy_ring; | |
1813 | } else { | |
1814 | goto out; | |
1815 | } | |
1816 | } | |
1817 | } | |
0d160211 JF |
1818 | |
1819 | again: | |
1820 | err = xenbus_transaction_start(&xbt); | |
1821 | if (err) { | |
1822 | xenbus_dev_fatal(dev, err, "starting transaction"); | |
1823 | goto destroy_ring; | |
1824 | } | |
1825 | ||
50ee6061 AB |
1826 | if (num_queues == 1) { |
1827 | err = write_queue_xenstore_keys(&info->queues[0], &xbt, 0); /* flat */ | |
1828 | if (err) | |
1829 | goto abort_transaction_no_dev_fatal; | |
d634bf2c | 1830 | } else { |
50ee6061 AB |
1831 | /* Write the number of queues */ |
1832 | err = xenbus_printf(xbt, dev->nodename, "multi-queue-num-queues", | |
1833 | "%u", num_queues); | |
d634bf2c | 1834 | if (err) { |
50ee6061 AB |
1835 | message = "writing multi-queue-num-queues"; |
1836 | goto abort_transaction_no_dev_fatal; | |
d634bf2c | 1837 | } |
50ee6061 AB |
1838 | |
1839 | /* Write the keys for each queue */ | |
1840 | for (i = 0; i < num_queues; ++i) { | |
1841 | queue = &info->queues[i]; | |
1842 | err = write_queue_xenstore_keys(queue, &xbt, 1); /* hierarchical */ | |
1843 | if (err) | |
1844 | goto abort_transaction_no_dev_fatal; | |
d634bf2c | 1845 | } |
0d160211 JF |
1846 | } |
1847 | ||
50ee6061 | 1848 | /* The remaining keys are not queue-specific */ |
0d160211 JF |
1849 | err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u", |
1850 | 1); | |
1851 | if (err) { | |
1852 | message = "writing request-rx-copy"; | |
1853 | goto abort_transaction; | |
1854 | } | |
1855 | ||
1856 | err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1); | |
1857 | if (err) { | |
1858 | message = "writing feature-rx-notify"; | |
1859 | goto abort_transaction; | |
1860 | } | |
1861 | ||
1862 | err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1); | |
1863 | if (err) { | |
1864 | message = "writing feature-sg"; | |
1865 | goto abort_transaction; | |
1866 | } | |
1867 | ||
1868 | err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1); | |
1869 | if (err) { | |
1870 | message = "writing feature-gso-tcpv4"; | |
1871 | goto abort_transaction; | |
1872 | } | |
1873 | ||
2c0057de PD |
1874 | err = xenbus_write(xbt, dev->nodename, "feature-gso-tcpv6", "1"); |
1875 | if (err) { | |
1876 | message = "writing feature-gso-tcpv6"; | |
1877 | goto abort_transaction; | |
1878 | } | |
1879 | ||
1880 | err = xenbus_write(xbt, dev->nodename, "feature-ipv6-csum-offload", | |
1881 | "1"); | |
1882 | if (err) { | |
1883 | message = "writing feature-ipv6-csum-offload"; | |
1884 | goto abort_transaction; | |
1885 | } | |
1886 | ||
0d160211 JF |
1887 | err = xenbus_transaction_end(xbt, 0); |
1888 | if (err) { | |
1889 | if (err == -EAGAIN) | |
1890 | goto again; | |
1891 | xenbus_dev_fatal(dev, err, "completing transaction"); | |
1892 | goto destroy_ring; | |
1893 | } | |
1894 | ||
1895 | return 0; | |
1896 | ||
1897 | abort_transaction: | |
0d160211 | 1898 | xenbus_dev_fatal(dev, err, "%s", message); |
50ee6061 AB |
1899 | abort_transaction_no_dev_fatal: |
1900 | xenbus_transaction_end(xbt, 1); | |
0d160211 JF |
1901 | destroy_ring: |
1902 | xennet_disconnect_backend(info); | |
2688fcb7 AB |
1903 | kfree(info->queues); |
1904 | info->queues = NULL; | |
1905 | rtnl_lock(); | |
1906 | netif_set_real_num_tx_queues(info->netdev, 0); | |
db8c8ab6 | 1907 | rtnl_unlock(); |
0d160211 JF |
1908 | out: |
1909 | return err; | |
1910 | } | |
1911 | ||
0d160211 JF |
1912 | static int xennet_connect(struct net_device *dev) |
1913 | { | |
1914 | struct netfront_info *np = netdev_priv(dev); | |
2688fcb7 | 1915 | unsigned int num_queues = 0; |
a5b5dc3c | 1916 | int err; |
0d160211 | 1917 | unsigned int feature_rx_copy; |
2688fcb7 AB |
1918 | unsigned int j = 0; |
1919 | struct netfront_queue *queue = NULL; | |
0d160211 JF |
1920 | |
1921 | err = xenbus_scanf(XBT_NIL, np->xbdev->otherend, | |
1922 | "feature-rx-copy", "%u", &feature_rx_copy); | |
1923 | if (err != 1) | |
1924 | feature_rx_copy = 0; | |
1925 | ||
1926 | if (!feature_rx_copy) { | |
1927 | dev_info(&dev->dev, | |
898eb71c | 1928 | "backend does not support copying receive path\n"); |
0d160211 JF |
1929 | return -ENODEV; |
1930 | } | |
1931 | ||
f502bf2b | 1932 | err = talk_to_netback(np->xbdev, np); |
0d160211 JF |
1933 | if (err) |
1934 | return err; | |
1935 | ||
2688fcb7 AB |
1936 | /* talk_to_netback() sets the correct number of queues */ |
1937 | num_queues = dev->real_num_tx_queues; | |
1938 | ||
1ba37c51 | 1939 | rtnl_lock(); |
fb507934 | 1940 | netdev_update_features(dev); |
1ba37c51 | 1941 | rtnl_unlock(); |
0d160211 | 1942 | |
0d160211 | 1943 | /* |
a5b5dc3c | 1944 | * All public and private state should now be sane. Get |
0d160211 JF |
1945 | * ready to start sending and receiving packets and give the driver |
1946 | * domain a kick because we've probably just requeued some | |
1947 | * packets. | |
1948 | */ | |
1949 | netif_carrier_on(np->netdev); | |
2688fcb7 AB |
1950 | for (j = 0; j < num_queues; ++j) { |
1951 | queue = &np->queues[j]; | |
f50b4076 | 1952 | |
2688fcb7 AB |
1953 | notify_remote_via_irq(queue->tx_irq); |
1954 | if (queue->tx_irq != queue->rx_irq) | |
1955 | notify_remote_via_irq(queue->rx_irq); | |
2688fcb7 | 1956 | |
f50b4076 DV |
1957 | spin_lock_irq(&queue->tx_lock); |
1958 | xennet_tx_buf_gc(queue); | |
2688fcb7 | 1959 | spin_unlock_irq(&queue->tx_lock); |
f50b4076 DV |
1960 | |
1961 | spin_lock_bh(&queue->rx_lock); | |
1962 | xennet_alloc_rx_buffers(queue); | |
2688fcb7 AB |
1963 | spin_unlock_bh(&queue->rx_lock); |
1964 | } | |
0d160211 JF |
1965 | |
1966 | return 0; | |
1967 | } | |
1968 | ||
1969 | /** | |
1970 | * Callback received when the backend's state changes. | |
1971 | */ | |
f502bf2b | 1972 | static void netback_changed(struct xenbus_device *dev, |
0d160211 JF |
1973 | enum xenbus_state backend_state) |
1974 | { | |
1b713e00 | 1975 | struct netfront_info *np = dev_get_drvdata(&dev->dev); |
0d160211 JF |
1976 | struct net_device *netdev = np->netdev; |
1977 | ||
1978 | dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state)); | |
1979 | ||
1980 | switch (backend_state) { | |
1981 | case XenbusStateInitialising: | |
1982 | case XenbusStateInitialised: | |
b78c9512 NI |
1983 | case XenbusStateReconfiguring: |
1984 | case XenbusStateReconfigured: | |
0d160211 | 1985 | case XenbusStateUnknown: |
0d160211 JF |
1986 | break; |
1987 | ||
1988 | case XenbusStateInitWait: | |
1989 | if (dev->state != XenbusStateInitialising) | |
1990 | break; | |
1991 | if (xennet_connect(netdev) != 0) | |
1992 | break; | |
1993 | xenbus_switch_state(dev, XenbusStateConnected); | |
08e34eb1 LE |
1994 | break; |
1995 | ||
1996 | case XenbusStateConnected: | |
ee89bab1 | 1997 | netdev_notify_peers(netdev); |
0d160211 JF |
1998 | break; |
1999 | ||
bce3ea81 DV |
2000 | case XenbusStateClosed: |
2001 | if (dev->state == XenbusStateClosed) | |
2002 | break; | |
2003 | /* Missed the backend's CLOSING state -- fallthrough */ | |
0d160211 JF |
2004 | case XenbusStateClosing: |
2005 | xenbus_frontend_closed(dev); | |
2006 | break; | |
2007 | } | |
2008 | } | |
2009 | ||
e0ce4af9 IC |
2010 | static const struct xennet_stat { |
2011 | char name[ETH_GSTRING_LEN]; | |
2012 | u16 offset; | |
2013 | } xennet_stats[] = { | |
2014 | { | |
2015 | "rx_gso_checksum_fixup", | |
2016 | offsetof(struct netfront_info, rx_gso_checksum_fixup) | |
2017 | }, | |
2018 | }; | |
2019 | ||
2020 | static int xennet_get_sset_count(struct net_device *dev, int string_set) | |
2021 | { | |
2022 | switch (string_set) { | |
2023 | case ETH_SS_STATS: | |
2024 | return ARRAY_SIZE(xennet_stats); | |
2025 | default: | |
2026 | return -EINVAL; | |
2027 | } | |
2028 | } | |
2029 | ||
2030 | static void xennet_get_ethtool_stats(struct net_device *dev, | |
2031 | struct ethtool_stats *stats, u64 * data) | |
2032 | { | |
2033 | void *np = netdev_priv(dev); | |
2034 | int i; | |
2035 | ||
2036 | for (i = 0; i < ARRAY_SIZE(xennet_stats); i++) | |
2688fcb7 | 2037 | data[i] = atomic_read((atomic_t *)(np + xennet_stats[i].offset)); |
e0ce4af9 IC |
2038 | } |
2039 | ||
2040 | static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data) | |
2041 | { | |
2042 | int i; | |
2043 | ||
2044 | switch (stringset) { | |
2045 | case ETH_SS_STATS: | |
2046 | for (i = 0; i < ARRAY_SIZE(xennet_stats); i++) | |
2047 | memcpy(data + i * ETH_GSTRING_LEN, | |
2048 | xennet_stats[i].name, ETH_GSTRING_LEN); | |
2049 | break; | |
2050 | } | |
2051 | } | |
2052 | ||
0fc0b732 | 2053 | static const struct ethtool_ops xennet_ethtool_ops = |
0d160211 | 2054 | { |
0d160211 | 2055 | .get_link = ethtool_op_get_link, |
e0ce4af9 IC |
2056 | |
2057 | .get_sset_count = xennet_get_sset_count, | |
2058 | .get_ethtool_stats = xennet_get_ethtool_stats, | |
2059 | .get_strings = xennet_get_strings, | |
0d160211 JF |
2060 | }; |
2061 | ||
2062 | #ifdef CONFIG_SYSFS | |
1f3c2eba DV |
2063 | static ssize_t show_rxbuf(struct device *dev, |
2064 | struct device_attribute *attr, char *buf) | |
0d160211 | 2065 | { |
1f3c2eba | 2066 | return sprintf(buf, "%lu\n", NET_RX_RING_SIZE); |
0d160211 JF |
2067 | } |
2068 | ||
1f3c2eba DV |
2069 | static ssize_t store_rxbuf(struct device *dev, |
2070 | struct device_attribute *attr, | |
2071 | const char *buf, size_t len) | |
0d160211 | 2072 | { |
0d160211 JF |
2073 | char *endp; |
2074 | unsigned long target; | |
2075 | ||
2076 | if (!capable(CAP_NET_ADMIN)) | |
2077 | return -EPERM; | |
2078 | ||
2079 | target = simple_strtoul(buf, &endp, 0); | |
2080 | if (endp == buf) | |
2081 | return -EBADMSG; | |
2082 | ||
1f3c2eba | 2083 | /* rxbuf_min and rxbuf_max are no longer configurable. */ |
0d160211 | 2084 | |
0d160211 JF |
2085 | return len; |
2086 | } | |
2087 | ||
27b917e5 TI |
2088 | static DEVICE_ATTR(rxbuf_min, S_IRUGO|S_IWUSR, show_rxbuf, store_rxbuf); |
2089 | static DEVICE_ATTR(rxbuf_max, S_IRUGO|S_IWUSR, show_rxbuf, store_rxbuf); | |
2090 | static DEVICE_ATTR(rxbuf_cur, S_IRUGO, show_rxbuf, NULL); | |
0d160211 | 2091 | |
27b917e5 TI |
2092 | static struct attribute *xennet_dev_attrs[] = { |
2093 | &dev_attr_rxbuf_min.attr, | |
2094 | &dev_attr_rxbuf_max.attr, | |
2095 | &dev_attr_rxbuf_cur.attr, | |
2096 | NULL | |
2097 | }; | |
0d160211 | 2098 | |
27b917e5 TI |
2099 | static const struct attribute_group xennet_dev_group = { |
2100 | .attrs = xennet_dev_attrs | |
2101 | }; | |
0d160211 JF |
2102 | #endif /* CONFIG_SYSFS */ |
2103 | ||
8e0e46bb | 2104 | static int xennet_remove(struct xenbus_device *dev) |
0d160211 | 2105 | { |
1b713e00 | 2106 | struct netfront_info *info = dev_get_drvdata(&dev->dev); |
2688fcb7 AB |
2107 | unsigned int num_queues = info->netdev->real_num_tx_queues; |
2108 | struct netfront_queue *queue = NULL; | |
2109 | unsigned int i = 0; | |
0d160211 JF |
2110 | |
2111 | dev_dbg(&dev->dev, "%s\n", dev->nodename); | |
2112 | ||
0d160211 JF |
2113 | xennet_disconnect_backend(info); |
2114 | ||
6bc96d04 IC |
2115 | unregister_netdev(info->netdev); |
2116 | ||
2688fcb7 AB |
2117 | for (i = 0; i < num_queues; ++i) { |
2118 | queue = &info->queues[i]; | |
2119 | del_timer_sync(&queue->rx_refill_timer); | |
2120 | } | |
2121 | ||
2122 | if (num_queues) { | |
2123 | kfree(info->queues); | |
2124 | info->queues = NULL; | |
2125 | } | |
6bc96d04 | 2126 | |
900e1833 | 2127 | xennet_free_netdev(info->netdev); |
0d160211 JF |
2128 | |
2129 | return 0; | |
2130 | } | |
2131 | ||
95afae48 DV |
2132 | static const struct xenbus_device_id netfront_ids[] = { |
2133 | { "vif" }, | |
2134 | { "" } | |
2135 | }; | |
2136 | ||
2137 | static struct xenbus_driver netfront_driver = { | |
2138 | .ids = netfront_ids, | |
0d160211 | 2139 | .probe = netfront_probe, |
8e0e46bb | 2140 | .remove = xennet_remove, |
0d160211 | 2141 | .resume = netfront_resume, |
f502bf2b | 2142 | .otherend_changed = netback_changed, |
95afae48 | 2143 | }; |
0d160211 JF |
2144 | |
2145 | static int __init netif_init(void) | |
2146 | { | |
6e833587 | 2147 | if (!xen_domain()) |
0d160211 JF |
2148 | return -ENODEV; |
2149 | ||
51c71a3b | 2150 | if (!xen_has_pv_nic_devices()) |
b9136d20 IM |
2151 | return -ENODEV; |
2152 | ||
383eda32 | 2153 | pr_info("Initialising Xen virtual ethernet driver\n"); |
0d160211 | 2154 | |
50ee6061 AB |
2155 | /* Allow as many queues as there are CPUs, by default */ |
2156 | xennet_max_queues = num_online_cpus(); | |
2157 | ||
ffb78a26 | 2158 | return xenbus_register_frontend(&netfront_driver); |
0d160211 JF |
2159 | } |
2160 | module_init(netif_init); | |
2161 | ||
2162 | ||
2163 | static void __exit netif_exit(void) | |
2164 | { | |
ffb78a26 | 2165 | xenbus_unregister_driver(&netfront_driver); |
0d160211 JF |
2166 | } |
2167 | module_exit(netif_exit); | |
2168 | ||
2169 | MODULE_DESCRIPTION("Xen virtual network device frontend"); | |
2170 | MODULE_LICENSE("GPL"); | |
d2f0c52b | 2171 | MODULE_ALIAS("xen:vif"); |
4f93f09b | 2172 | MODULE_ALIAS("xennet"); |