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 | ||
5b5971df EO |
90 | static DECLARE_WAIT_QUEUE_HEAD(module_unload_q); |
91 | ||
e00f85be | 92 | struct netfront_stats { |
900e1833 DV |
93 | u64 packets; |
94 | u64 bytes; | |
e00f85be | 95 | struct u64_stats_sync syncp; |
96 | }; | |
97 | ||
2688fcb7 AB |
98 | struct netfront_info; |
99 | ||
100 | struct netfront_queue { | |
101 | unsigned int id; /* Queue ID, 0-based */ | |
102 | char name[QUEUE_NAME_SIZE]; /* DEVNAME-qN */ | |
103 | struct netfront_info *info; | |
0d160211 | 104 | |
bea3348e | 105 | struct napi_struct napi; |
0d160211 | 106 | |
d634bf2c WL |
107 | /* Split event channels support, tx_* == rx_* when using |
108 | * single event channel. | |
109 | */ | |
110 | unsigned int tx_evtchn, rx_evtchn; | |
111 | unsigned int tx_irq, rx_irq; | |
112 | /* Only used when split event channels support is enabled */ | |
2688fcb7 AB |
113 | char tx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-tx */ |
114 | char rx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-rx */ | |
0d160211 | 115 | |
84284d3c JF |
116 | spinlock_t tx_lock; |
117 | struct xen_netif_tx_front_ring tx; | |
118 | int tx_ring_ref; | |
0d160211 JF |
119 | |
120 | /* | |
121 | * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries | |
122 | * are linked from tx_skb_freelist through skb_entry.link. | |
123 | * | |
124 | * NB. Freelist index entries are always going to be less than | |
125 | * PAGE_OFFSET, whereas pointers to skbs will always be equal or | |
126 | * greater than PAGE_OFFSET: we use this property to distinguish | |
127 | * them. | |
128 | */ | |
129 | union skb_entry { | |
130 | struct sk_buff *skb; | |
1ffb40b8 | 131 | unsigned long link; |
0d160211 JF |
132 | } tx_skbs[NET_TX_RING_SIZE]; |
133 | grant_ref_t gref_tx_head; | |
134 | grant_ref_t grant_tx_ref[NET_TX_RING_SIZE]; | |
cefe0078 | 135 | struct page *grant_tx_page[NET_TX_RING_SIZE]; |
0d160211 JF |
136 | unsigned tx_skb_freelist; |
137 | ||
84284d3c JF |
138 | spinlock_t rx_lock ____cacheline_aligned_in_smp; |
139 | struct xen_netif_rx_front_ring rx; | |
140 | int rx_ring_ref; | |
141 | ||
84284d3c JF |
142 | struct timer_list rx_refill_timer; |
143 | ||
0d160211 JF |
144 | struct sk_buff *rx_skbs[NET_RX_RING_SIZE]; |
145 | grant_ref_t gref_rx_head; | |
146 | grant_ref_t grant_rx_ref[NET_RX_RING_SIZE]; | |
2688fcb7 AB |
147 | }; |
148 | ||
149 | struct netfront_info { | |
150 | struct list_head list; | |
151 | struct net_device *netdev; | |
152 | ||
153 | struct xenbus_device *xbdev; | |
154 | ||
155 | /* Multi-queue support */ | |
156 | struct netfront_queue *queues; | |
e0ce4af9 IC |
157 | |
158 | /* Statistics */ | |
900e1833 DV |
159 | struct netfront_stats __percpu *rx_stats; |
160 | struct netfront_stats __percpu *tx_stats; | |
e00f85be | 161 | |
2688fcb7 | 162 | atomic_t rx_gso_checksum_fixup; |
0d160211 JF |
163 | }; |
164 | ||
165 | struct netfront_rx_info { | |
166 | struct xen_netif_rx_response rx; | |
167 | struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1]; | |
168 | }; | |
169 | ||
1ffb40b8 IY |
170 | static void skb_entry_set_link(union skb_entry *list, unsigned short id) |
171 | { | |
172 | list->link = id; | |
173 | } | |
174 | ||
175 | static int skb_entry_is_link(const union skb_entry *list) | |
176 | { | |
177 | BUILD_BUG_ON(sizeof(list->skb) != sizeof(list->link)); | |
807540ba | 178 | return (unsigned long)list->skb < PAGE_OFFSET; |
1ffb40b8 IY |
179 | } |
180 | ||
0d160211 JF |
181 | /* |
182 | * Access macros for acquiring freeing slots in tx_skbs[]. | |
183 | */ | |
184 | ||
185 | static void add_id_to_freelist(unsigned *head, union skb_entry *list, | |
186 | unsigned short id) | |
187 | { | |
1ffb40b8 | 188 | skb_entry_set_link(&list[id], *head); |
0d160211 JF |
189 | *head = id; |
190 | } | |
191 | ||
192 | static unsigned short get_id_from_freelist(unsigned *head, | |
193 | union skb_entry *list) | |
194 | { | |
195 | unsigned int id = *head; | |
196 | *head = list[id].link; | |
197 | return id; | |
198 | } | |
199 | ||
200 | static int xennet_rxidx(RING_IDX idx) | |
201 | { | |
202 | return idx & (NET_RX_RING_SIZE - 1); | |
203 | } | |
204 | ||
2688fcb7 | 205 | static struct sk_buff *xennet_get_rx_skb(struct netfront_queue *queue, |
0d160211 JF |
206 | RING_IDX ri) |
207 | { | |
208 | int i = xennet_rxidx(ri); | |
2688fcb7 AB |
209 | struct sk_buff *skb = queue->rx_skbs[i]; |
210 | queue->rx_skbs[i] = NULL; | |
0d160211 JF |
211 | return skb; |
212 | } | |
213 | ||
2688fcb7 | 214 | static grant_ref_t xennet_get_rx_ref(struct netfront_queue *queue, |
0d160211 JF |
215 | RING_IDX ri) |
216 | { | |
217 | int i = xennet_rxidx(ri); | |
2688fcb7 AB |
218 | grant_ref_t ref = queue->grant_rx_ref[i]; |
219 | queue->grant_rx_ref[i] = GRANT_INVALID_REF; | |
0d160211 JF |
220 | return ref; |
221 | } | |
222 | ||
223 | #ifdef CONFIG_SYSFS | |
27b917e5 | 224 | static const struct attribute_group xennet_dev_group; |
0d160211 JF |
225 | #endif |
226 | ||
3ad9b358 | 227 | static bool xennet_can_sg(struct net_device *dev) |
0d160211 | 228 | { |
3ad9b358 | 229 | return dev->features & NETIF_F_SG; |
0d160211 JF |
230 | } |
231 | ||
232 | ||
e99e88a9 | 233 | static void rx_refill_timeout(struct timer_list *t) |
0d160211 | 234 | { |
e99e88a9 | 235 | struct netfront_queue *queue = from_timer(queue, t, rx_refill_timer); |
2688fcb7 | 236 | napi_schedule(&queue->napi); |
0d160211 JF |
237 | } |
238 | ||
2688fcb7 | 239 | static int netfront_tx_slot_available(struct netfront_queue *queue) |
0d160211 | 240 | { |
2688fcb7 | 241 | return (queue->tx.req_prod_pvt - queue->tx.rsp_cons) < |
1f3c2eba | 242 | (NET_TX_RING_SIZE - MAX_SKB_FRAGS - 2); |
0d160211 JF |
243 | } |
244 | ||
2688fcb7 | 245 | static void xennet_maybe_wake_tx(struct netfront_queue *queue) |
0d160211 | 246 | { |
2688fcb7 AB |
247 | struct net_device *dev = queue->info->netdev; |
248 | struct netdev_queue *dev_queue = netdev_get_tx_queue(dev, queue->id); | |
0d160211 | 249 | |
2688fcb7 AB |
250 | if (unlikely(netif_tx_queue_stopped(dev_queue)) && |
251 | netfront_tx_slot_available(queue) && | |
0d160211 | 252 | likely(netif_running(dev))) |
2688fcb7 | 253 | netif_tx_wake_queue(netdev_get_tx_queue(dev, queue->id)); |
0d160211 JF |
254 | } |
255 | ||
1f3c2eba DV |
256 | |
257 | static struct sk_buff *xennet_alloc_one_rx_buffer(struct netfront_queue *queue) | |
0d160211 | 258 | { |
0d160211 JF |
259 | struct sk_buff *skb; |
260 | struct page *page; | |
0d160211 | 261 | |
1f3c2eba DV |
262 | skb = __netdev_alloc_skb(queue->info->netdev, |
263 | RX_COPY_THRESHOLD + NET_IP_ALIGN, | |
264 | GFP_ATOMIC | __GFP_NOWARN); | |
265 | if (unlikely(!skb)) | |
266 | return NULL; | |
0d160211 | 267 | |
1f3c2eba DV |
268 | page = alloc_page(GFP_ATOMIC | __GFP_NOWARN); |
269 | if (!page) { | |
270 | kfree_skb(skb); | |
271 | return NULL; | |
0d160211 | 272 | } |
1f3c2eba DV |
273 | skb_add_rx_frag(skb, 0, page, 0, 0, PAGE_SIZE); |
274 | ||
275 | /* Align ip header to a 16 bytes boundary */ | |
276 | skb_reserve(skb, NET_IP_ALIGN); | |
277 | skb->dev = queue->info->netdev; | |
278 | ||
279 | return skb; | |
280 | } | |
0d160211 | 281 | |
1f3c2eba DV |
282 | |
283 | static void xennet_alloc_rx_buffers(struct netfront_queue *queue) | |
284 | { | |
285 | RING_IDX req_prod = queue->rx.req_prod_pvt; | |
286 | int notify; | |
538d9291 | 287 | int err = 0; |
1f3c2eba DV |
288 | |
289 | if (unlikely(!netif_carrier_ok(queue->info->netdev))) | |
0d160211 | 290 | return; |
0d160211 | 291 | |
1f3c2eba DV |
292 | for (req_prod = queue->rx.req_prod_pvt; |
293 | req_prod - queue->rx.rsp_cons < NET_RX_RING_SIZE; | |
294 | req_prod++) { | |
295 | struct sk_buff *skb; | |
296 | unsigned short id; | |
297 | grant_ref_t ref; | |
30c5d7f0 | 298 | struct page *page; |
1f3c2eba | 299 | struct xen_netif_rx_request *req; |
0d160211 | 300 | |
1f3c2eba | 301 | skb = xennet_alloc_one_rx_buffer(queue); |
538d9291 VRP |
302 | if (!skb) { |
303 | err = -ENOMEM; | |
0d160211 | 304 | break; |
538d9291 | 305 | } |
0d160211 | 306 | |
1f3c2eba | 307 | id = xennet_rxidx(req_prod); |
0d160211 | 308 | |
2688fcb7 AB |
309 | BUG_ON(queue->rx_skbs[id]); |
310 | queue->rx_skbs[id] = skb; | |
0d160211 | 311 | |
2688fcb7 | 312 | ref = gnttab_claim_grant_reference(&queue->gref_rx_head); |
269ebce4 | 313 | WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref)); |
2688fcb7 | 314 | queue->grant_rx_ref[id] = ref; |
0d160211 | 315 | |
30c5d7f0 | 316 | page = skb_frag_page(&skb_shinfo(skb)->frags[0]); |
0d160211 | 317 | |
1f3c2eba | 318 | req = RING_GET_REQUEST(&queue->rx, req_prod); |
30c5d7f0 JG |
319 | gnttab_page_grant_foreign_access_ref_one(ref, |
320 | queue->info->xbdev->otherend_id, | |
321 | page, | |
322 | 0); | |
0d160211 JF |
323 | req->id = id; |
324 | req->gref = ref; | |
325 | } | |
326 | ||
1f3c2eba DV |
327 | queue->rx.req_prod_pvt = req_prod; |
328 | ||
538d9291 VRP |
329 | /* Try again later if there are not enough requests or skb allocation |
330 | * failed. | |
331 | * Enough requests is quantified as the sum of newly created slots and | |
332 | * the unconsumed slots at the backend. | |
333 | */ | |
334 | if (req_prod - queue->rx.rsp_cons < NET_RX_SLOTS_MIN || | |
335 | unlikely(err)) { | |
1f3c2eba DV |
336 | mod_timer(&queue->rx_refill_timer, jiffies + (HZ/10)); |
337 | return; | |
338 | } | |
339 | ||
5dcddfae | 340 | wmb(); /* barrier so backend seens requests */ |
0d160211 | 341 | |
2688fcb7 | 342 | RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->rx, notify); |
0d160211 | 343 | if (notify) |
2688fcb7 | 344 | notify_remote_via_irq(queue->rx_irq); |
0d160211 JF |
345 | } |
346 | ||
347 | static int xennet_open(struct net_device *dev) | |
348 | { | |
349 | struct netfront_info *np = netdev_priv(dev); | |
2688fcb7 AB |
350 | unsigned int num_queues = dev->real_num_tx_queues; |
351 | unsigned int i = 0; | |
352 | struct netfront_queue *queue = NULL; | |
353 | ||
f599c64f RL |
354 | if (!np->queues) |
355 | return -ENODEV; | |
356 | ||
2688fcb7 AB |
357 | for (i = 0; i < num_queues; ++i) { |
358 | queue = &np->queues[i]; | |
359 | napi_enable(&queue->napi); | |
360 | ||
361 | spin_lock_bh(&queue->rx_lock); | |
362 | if (netif_carrier_ok(dev)) { | |
363 | xennet_alloc_rx_buffers(queue); | |
364 | queue->rx.sring->rsp_event = queue->rx.rsp_cons + 1; | |
365 | if (RING_HAS_UNCONSUMED_RESPONSES(&queue->rx)) | |
366 | napi_schedule(&queue->napi); | |
367 | } | |
368 | spin_unlock_bh(&queue->rx_lock); | |
0d160211 | 369 | } |
0d160211 | 370 | |
2688fcb7 | 371 | netif_tx_start_all_queues(dev); |
0d160211 JF |
372 | |
373 | return 0; | |
374 | } | |
375 | ||
2688fcb7 | 376 | static void xennet_tx_buf_gc(struct netfront_queue *queue) |
0d160211 JF |
377 | { |
378 | RING_IDX cons, prod; | |
379 | unsigned short id; | |
0d160211 | 380 | struct sk_buff *skb; |
7d0105b5 | 381 | bool more_to_do; |
0d160211 | 382 | |
2688fcb7 | 383 | BUG_ON(!netif_carrier_ok(queue->info->netdev)); |
0d160211 JF |
384 | |
385 | do { | |
2688fcb7 | 386 | prod = queue->tx.sring->rsp_prod; |
0d160211 JF |
387 | rmb(); /* Ensure we see responses up to 'rp'. */ |
388 | ||
2688fcb7 | 389 | for (cons = queue->tx.rsp_cons; cons != prod; cons++) { |
0d160211 JF |
390 | struct xen_netif_tx_response *txrsp; |
391 | ||
2688fcb7 | 392 | txrsp = RING_GET_RESPONSE(&queue->tx, cons); |
f942dc25 | 393 | if (txrsp->status == XEN_NETIF_RSP_NULL) |
0d160211 JF |
394 | continue; |
395 | ||
396 | id = txrsp->id; | |
2688fcb7 | 397 | skb = queue->tx_skbs[id].skb; |
0d160211 | 398 | if (unlikely(gnttab_query_foreign_access( |
2688fcb7 | 399 | queue->grant_tx_ref[id]) != 0)) { |
383eda32 JP |
400 | pr_alert("%s: warning -- grant still in use by backend domain\n", |
401 | __func__); | |
0d160211 JF |
402 | BUG(); |
403 | } | |
404 | gnttab_end_foreign_access_ref( | |
2688fcb7 | 405 | queue->grant_tx_ref[id], GNTMAP_readonly); |
0d160211 | 406 | gnttab_release_grant_reference( |
2688fcb7 AB |
407 | &queue->gref_tx_head, queue->grant_tx_ref[id]); |
408 | queue->grant_tx_ref[id] = GRANT_INVALID_REF; | |
409 | queue->grant_tx_page[id] = NULL; | |
410 | add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, id); | |
0d160211 JF |
411 | dev_kfree_skb_irq(skb); |
412 | } | |
413 | ||
2688fcb7 | 414 | queue->tx.rsp_cons = prod; |
0d160211 | 415 | |
7d0105b5 MC |
416 | RING_FINAL_CHECK_FOR_RESPONSES(&queue->tx, more_to_do); |
417 | } while (more_to_do); | |
0d160211 | 418 | |
2688fcb7 | 419 | xennet_maybe_wake_tx(queue); |
0d160211 JF |
420 | } |
421 | ||
30c5d7f0 JG |
422 | struct xennet_gnttab_make_txreq { |
423 | struct netfront_queue *queue; | |
424 | struct sk_buff *skb; | |
425 | struct page *page; | |
426 | struct xen_netif_tx_request *tx; /* Last request */ | |
427 | unsigned int size; | |
428 | }; | |
429 | ||
430 | static void xennet_tx_setup_grant(unsigned long gfn, unsigned int offset, | |
431 | unsigned int len, void *data) | |
0d160211 | 432 | { |
30c5d7f0 | 433 | struct xennet_gnttab_make_txreq *info = data; |
0d160211 | 434 | unsigned int id; |
a55e8bb8 | 435 | struct xen_netif_tx_request *tx; |
0d160211 | 436 | grant_ref_t ref; |
30c5d7f0 JG |
437 | /* convenient aliases */ |
438 | struct page *page = info->page; | |
439 | struct netfront_queue *queue = info->queue; | |
440 | struct sk_buff *skb = info->skb; | |
0d160211 | 441 | |
a55e8bb8 DV |
442 | id = get_id_from_freelist(&queue->tx_skb_freelist, queue->tx_skbs); |
443 | tx = RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++); | |
444 | ref = gnttab_claim_grant_reference(&queue->gref_tx_head); | |
269ebce4 | 445 | WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref)); |
0d160211 | 446 | |
30c5d7f0 JG |
447 | gnttab_grant_foreign_access_ref(ref, queue->info->xbdev->otherend_id, |
448 | gfn, GNTMAP_readonly); | |
0d160211 | 449 | |
a55e8bb8 DV |
450 | queue->tx_skbs[id].skb = skb; |
451 | queue->grant_tx_page[id] = page; | |
452 | queue->grant_tx_ref[id] = ref; | |
0d160211 | 453 | |
a55e8bb8 DV |
454 | tx->id = id; |
455 | tx->gref = ref; | |
456 | tx->offset = offset; | |
457 | tx->size = len; | |
458 | tx->flags = 0; | |
0d160211 | 459 | |
30c5d7f0 JG |
460 | info->tx = tx; |
461 | info->size += tx->size; | |
462 | } | |
463 | ||
464 | static struct xen_netif_tx_request *xennet_make_first_txreq( | |
465 | struct netfront_queue *queue, struct sk_buff *skb, | |
466 | struct page *page, unsigned int offset, unsigned int len) | |
467 | { | |
468 | struct xennet_gnttab_make_txreq info = { | |
469 | .queue = queue, | |
470 | .skb = skb, | |
471 | .page = page, | |
472 | .size = 0, | |
473 | }; | |
474 | ||
475 | gnttab_for_one_grant(page, offset, len, xennet_tx_setup_grant, &info); | |
476 | ||
477 | return info.tx; | |
478 | } | |
479 | ||
480 | static void xennet_make_one_txreq(unsigned long gfn, unsigned int offset, | |
481 | unsigned int len, void *data) | |
482 | { | |
483 | struct xennet_gnttab_make_txreq *info = data; | |
484 | ||
485 | info->tx->flags |= XEN_NETTXF_more_data; | |
486 | skb_get(info->skb); | |
487 | xennet_tx_setup_grant(gfn, offset, len, data); | |
a55e8bb8 | 488 | } |
0d160211 | 489 | |
a55e8bb8 DV |
490 | static struct xen_netif_tx_request *xennet_make_txreqs( |
491 | struct netfront_queue *queue, struct xen_netif_tx_request *tx, | |
492 | struct sk_buff *skb, struct page *page, | |
493 | unsigned int offset, unsigned int len) | |
494 | { | |
30c5d7f0 JG |
495 | struct xennet_gnttab_make_txreq info = { |
496 | .queue = queue, | |
497 | .skb = skb, | |
498 | .tx = tx, | |
499 | }; | |
500 | ||
a55e8bb8 DV |
501 | /* Skip unused frames from start of page */ |
502 | page += offset >> PAGE_SHIFT; | |
503 | offset &= ~PAGE_MASK; | |
0d160211 | 504 | |
a55e8bb8 | 505 | while (len) { |
30c5d7f0 JG |
506 | info.page = page; |
507 | info.size = 0; | |
508 | ||
509 | gnttab_foreach_grant_in_range(page, offset, len, | |
510 | xennet_make_one_txreq, | |
511 | &info); | |
512 | ||
a55e8bb8 DV |
513 | page++; |
514 | offset = 0; | |
30c5d7f0 | 515 | len -= info.size; |
0d160211 JF |
516 | } |
517 | ||
30c5d7f0 | 518 | return info.tx; |
0d160211 JF |
519 | } |
520 | ||
f36c3747 | 521 | /* |
e84448d5 DV |
522 | * Count how many ring slots are required to send this skb. Each frag |
523 | * might be a compound page. | |
f36c3747 | 524 | */ |
e84448d5 | 525 | static int xennet_count_skb_slots(struct sk_buff *skb) |
f36c3747 IC |
526 | { |
527 | int i, frags = skb_shinfo(skb)->nr_frags; | |
30c5d7f0 | 528 | int slots; |
e84448d5 | 529 | |
30c5d7f0 JG |
530 | slots = gnttab_count_grant(offset_in_page(skb->data), |
531 | skb_headlen(skb)); | |
f36c3747 IC |
532 | |
533 | for (i = 0; i < frags; i++) { | |
534 | skb_frag_t *frag = skb_shinfo(skb)->frags + i; | |
535 | unsigned long size = skb_frag_size(frag); | |
536 | unsigned long offset = frag->page_offset; | |
537 | ||
538 | /* Skip unused frames from start of page */ | |
539 | offset &= ~PAGE_MASK; | |
540 | ||
30c5d7f0 | 541 | slots += gnttab_count_grant(offset, size); |
f36c3747 IC |
542 | } |
543 | ||
30c5d7f0 | 544 | return slots; |
f36c3747 IC |
545 | } |
546 | ||
50ee6061 AB |
547 | static u16 xennet_select_queue(struct net_device *dev, struct sk_buff *skb, |
548 | void *accel_priv, select_queue_fallback_t fallback) | |
2688fcb7 | 549 | { |
50ee6061 AB |
550 | unsigned int num_queues = dev->real_num_tx_queues; |
551 | u32 hash; | |
552 | u16 queue_idx; | |
553 | ||
554 | /* First, check if there is only one queue */ | |
555 | if (num_queues == 1) { | |
556 | queue_idx = 0; | |
557 | } else { | |
558 | hash = skb_get_hash(skb); | |
559 | queue_idx = hash % num_queues; | |
560 | } | |
561 | ||
562 | return queue_idx; | |
2688fcb7 AB |
563 | } |
564 | ||
30c5d7f0 JG |
565 | #define MAX_XEN_SKB_FRAGS (65536 / XEN_PAGE_SIZE + 1) |
566 | ||
0d160211 JF |
567 | static int xennet_start_xmit(struct sk_buff *skb, struct net_device *dev) |
568 | { | |
0d160211 | 569 | struct netfront_info *np = netdev_priv(dev); |
900e1833 | 570 | struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats); |
a55e8bb8 DV |
571 | struct xen_netif_tx_request *tx, *first_tx; |
572 | unsigned int i; | |
0d160211 | 573 | int notify; |
f36c3747 | 574 | int slots; |
a55e8bb8 DV |
575 | struct page *page; |
576 | unsigned int offset; | |
577 | unsigned int len; | |
cf66f9d4 | 578 | unsigned long flags; |
2688fcb7 AB |
579 | struct netfront_queue *queue = NULL; |
580 | unsigned int num_queues = dev->real_num_tx_queues; | |
581 | u16 queue_index; | |
fd07160b | 582 | struct sk_buff *nskb; |
2688fcb7 AB |
583 | |
584 | /* Drop the packet if no queues are set up */ | |
585 | if (num_queues < 1) | |
586 | goto drop; | |
587 | /* Determine which queue to transmit this SKB on */ | |
588 | queue_index = skb_get_queue_mapping(skb); | |
589 | queue = &np->queues[queue_index]; | |
0d160211 | 590 | |
9ecd1a75 WL |
591 | /* If skb->len is too big for wire format, drop skb and alert |
592 | * user about misconfiguration. | |
593 | */ | |
594 | if (unlikely(skb->len > XEN_NETIF_MAX_TX_SIZE)) { | |
595 | net_alert_ratelimited( | |
596 | "xennet: skb->len = %u, too big for wire format\n", | |
597 | skb->len); | |
598 | goto drop; | |
599 | } | |
600 | ||
e84448d5 | 601 | slots = xennet_count_skb_slots(skb); |
30c5d7f0 | 602 | if (unlikely(slots > MAX_XEN_SKB_FRAGS + 1)) { |
97a6d1bb ZK |
603 | net_dbg_ratelimited("xennet: skb rides the rocket: %d slots, %d bytes\n", |
604 | slots, skb->len); | |
605 | if (skb_linearize(skb)) | |
606 | goto drop; | |
0d160211 JF |
607 | } |
608 | ||
a55e8bb8 DV |
609 | page = virt_to_page(skb->data); |
610 | offset = offset_in_page(skb->data); | |
fd07160b VK |
611 | |
612 | /* The first req should be at least ETH_HLEN size or the packet will be | |
613 | * dropped by netback. | |
614 | */ | |
615 | if (unlikely(PAGE_SIZE - offset < ETH_HLEN)) { | |
616 | nskb = skb_copy(skb, GFP_ATOMIC); | |
617 | if (!nskb) | |
618 | goto drop; | |
62f3250f | 619 | dev_consume_skb_any(skb); |
fd07160b VK |
620 | skb = nskb; |
621 | page = virt_to_page(skb->data); | |
622 | offset = offset_in_page(skb->data); | |
623 | } | |
624 | ||
a55e8bb8 DV |
625 | len = skb_headlen(skb); |
626 | ||
2688fcb7 | 627 | spin_lock_irqsave(&queue->tx_lock, flags); |
0d160211 JF |
628 | |
629 | if (unlikely(!netif_carrier_ok(dev) || | |
f36c3747 | 630 | (slots > 1 && !xennet_can_sg(dev)) || |
8b86a61d | 631 | netif_needs_gso(skb, netif_skb_features(skb)))) { |
2688fcb7 | 632 | spin_unlock_irqrestore(&queue->tx_lock, flags); |
0d160211 JF |
633 | goto drop; |
634 | } | |
635 | ||
a55e8bb8 | 636 | /* First request for the linear area. */ |
30c5d7f0 JG |
637 | first_tx = tx = xennet_make_first_txreq(queue, skb, |
638 | page, offset, len); | |
639 | offset += tx->size; | |
640 | if (offset == PAGE_SIZE) { | |
641 | page++; | |
642 | offset = 0; | |
643 | } | |
a55e8bb8 | 644 | len -= tx->size; |
0d160211 | 645 | |
0d160211 JF |
646 | if (skb->ip_summed == CHECKSUM_PARTIAL) |
647 | /* local packet? */ | |
f942dc25 | 648 | tx->flags |= XEN_NETTXF_csum_blank | XEN_NETTXF_data_validated; |
0d160211 JF |
649 | else if (skb->ip_summed == CHECKSUM_UNNECESSARY) |
650 | /* remote but checksummed. */ | |
f942dc25 | 651 | tx->flags |= XEN_NETTXF_data_validated; |
0d160211 | 652 | |
a55e8bb8 | 653 | /* Optional extra info after the first request. */ |
0d160211 JF |
654 | if (skb_shinfo(skb)->gso_size) { |
655 | struct xen_netif_extra_info *gso; | |
656 | ||
657 | gso = (struct xen_netif_extra_info *) | |
a55e8bb8 | 658 | RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++); |
0d160211 | 659 | |
e2d617c0 | 660 | tx->flags |= XEN_NETTXF_extra_info; |
0d160211 JF |
661 | |
662 | gso->u.gso.size = skb_shinfo(skb)->gso_size; | |
2c0057de PD |
663 | gso->u.gso.type = (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) ? |
664 | XEN_NETIF_GSO_TYPE_TCPV6 : | |
665 | XEN_NETIF_GSO_TYPE_TCPV4; | |
0d160211 JF |
666 | gso->u.gso.pad = 0; |
667 | gso->u.gso.features = 0; | |
668 | ||
669 | gso->type = XEN_NETIF_EXTRA_TYPE_GSO; | |
670 | gso->flags = 0; | |
0d160211 JF |
671 | } |
672 | ||
a55e8bb8 DV |
673 | /* Requests for the rest of the linear area. */ |
674 | tx = xennet_make_txreqs(queue, tx, skb, page, offset, len); | |
675 | ||
676 | /* Requests for all the frags. */ | |
677 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
678 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
679 | tx = xennet_make_txreqs(queue, tx, skb, | |
680 | skb_frag_page(frag), frag->page_offset, | |
681 | skb_frag_size(frag)); | |
682 | } | |
0d160211 | 683 | |
a55e8bb8 DV |
684 | /* First request has the packet length. */ |
685 | first_tx->size = skb->len; | |
0d160211 | 686 | |
2688fcb7 | 687 | RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify); |
0d160211 | 688 | if (notify) |
2688fcb7 | 689 | notify_remote_via_irq(queue->tx_irq); |
0d160211 | 690 | |
900e1833 DV |
691 | u64_stats_update_begin(&tx_stats->syncp); |
692 | tx_stats->bytes += skb->len; | |
693 | tx_stats->packets++; | |
694 | u64_stats_update_end(&tx_stats->syncp); | |
10a273a6 JF |
695 | |
696 | /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */ | |
2688fcb7 | 697 | xennet_tx_buf_gc(queue); |
0d160211 | 698 | |
2688fcb7 AB |
699 | if (!netfront_tx_slot_available(queue)) |
700 | netif_tx_stop_queue(netdev_get_tx_queue(dev, queue->id)); | |
0d160211 | 701 | |
2688fcb7 | 702 | spin_unlock_irqrestore(&queue->tx_lock, flags); |
0d160211 | 703 | |
6ed10654 | 704 | return NETDEV_TX_OK; |
0d160211 JF |
705 | |
706 | drop: | |
09f75cd7 | 707 | dev->stats.tx_dropped++; |
979de8a0 | 708 | dev_kfree_skb_any(skb); |
6ed10654 | 709 | return NETDEV_TX_OK; |
0d160211 JF |
710 | } |
711 | ||
712 | static int xennet_close(struct net_device *dev) | |
713 | { | |
714 | struct netfront_info *np = netdev_priv(dev); | |
2688fcb7 AB |
715 | unsigned int num_queues = dev->real_num_tx_queues; |
716 | unsigned int i; | |
717 | struct netfront_queue *queue; | |
718 | netif_tx_stop_all_queues(np->netdev); | |
719 | for (i = 0; i < num_queues; ++i) { | |
720 | queue = &np->queues[i]; | |
721 | napi_disable(&queue->napi); | |
722 | } | |
0d160211 JF |
723 | return 0; |
724 | } | |
725 | ||
2688fcb7 | 726 | static void xennet_move_rx_slot(struct netfront_queue *queue, struct sk_buff *skb, |
0d160211 JF |
727 | grant_ref_t ref) |
728 | { | |
2688fcb7 AB |
729 | int new = xennet_rxidx(queue->rx.req_prod_pvt); |
730 | ||
731 | BUG_ON(queue->rx_skbs[new]); | |
732 | queue->rx_skbs[new] = skb; | |
733 | queue->grant_rx_ref[new] = ref; | |
734 | RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->id = new; | |
735 | RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->gref = ref; | |
736 | queue->rx.req_prod_pvt++; | |
0d160211 JF |
737 | } |
738 | ||
2688fcb7 | 739 | static int xennet_get_extras(struct netfront_queue *queue, |
0d160211 JF |
740 | struct xen_netif_extra_info *extras, |
741 | RING_IDX rp) | |
742 | ||
743 | { | |
744 | struct xen_netif_extra_info *extra; | |
2688fcb7 AB |
745 | struct device *dev = &queue->info->netdev->dev; |
746 | RING_IDX cons = queue->rx.rsp_cons; | |
0d160211 JF |
747 | int err = 0; |
748 | ||
749 | do { | |
750 | struct sk_buff *skb; | |
751 | grant_ref_t ref; | |
752 | ||
753 | if (unlikely(cons + 1 == rp)) { | |
754 | if (net_ratelimit()) | |
755 | dev_warn(dev, "Missing extra info\n"); | |
756 | err = -EBADR; | |
757 | break; | |
758 | } | |
759 | ||
760 | extra = (struct xen_netif_extra_info *) | |
2688fcb7 | 761 | RING_GET_RESPONSE(&queue->rx, ++cons); |
0d160211 JF |
762 | |
763 | if (unlikely(!extra->type || | |
764 | extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) { | |
765 | if (net_ratelimit()) | |
766 | dev_warn(dev, "Invalid extra type: %d\n", | |
767 | extra->type); | |
768 | err = -EINVAL; | |
769 | } else { | |
770 | memcpy(&extras[extra->type - 1], extra, | |
771 | sizeof(*extra)); | |
772 | } | |
773 | ||
2688fcb7 AB |
774 | skb = xennet_get_rx_skb(queue, cons); |
775 | ref = xennet_get_rx_ref(queue, cons); | |
776 | xennet_move_rx_slot(queue, skb, ref); | |
0d160211 JF |
777 | } while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE); |
778 | ||
2688fcb7 | 779 | queue->rx.rsp_cons = cons; |
0d160211 JF |
780 | return err; |
781 | } | |
782 | ||
2688fcb7 | 783 | static int xennet_get_responses(struct netfront_queue *queue, |
0d160211 JF |
784 | struct netfront_rx_info *rinfo, RING_IDX rp, |
785 | struct sk_buff_head *list) | |
786 | { | |
787 | struct xen_netif_rx_response *rx = &rinfo->rx; | |
788 | struct xen_netif_extra_info *extras = rinfo->extras; | |
2688fcb7 AB |
789 | struct device *dev = &queue->info->netdev->dev; |
790 | RING_IDX cons = queue->rx.rsp_cons; | |
791 | struct sk_buff *skb = xennet_get_rx_skb(queue, cons); | |
792 | grant_ref_t ref = xennet_get_rx_ref(queue, cons); | |
0d160211 | 793 | int max = MAX_SKB_FRAGS + (rx->status <= RX_COPY_THRESHOLD); |
7158ff6d | 794 | int slots = 1; |
0d160211 JF |
795 | int err = 0; |
796 | unsigned long ret; | |
797 | ||
f942dc25 | 798 | if (rx->flags & XEN_NETRXF_extra_info) { |
2688fcb7 AB |
799 | err = xennet_get_extras(queue, extras, rp); |
800 | cons = queue->rx.rsp_cons; | |
0d160211 JF |
801 | } |
802 | ||
803 | for (;;) { | |
804 | if (unlikely(rx->status < 0 || | |
30c5d7f0 | 805 | rx->offset + rx->status > XEN_PAGE_SIZE)) { |
0d160211 | 806 | if (net_ratelimit()) |
6c10127d | 807 | dev_warn(dev, "rx->offset: %u, size: %d\n", |
0d160211 | 808 | rx->offset, rx->status); |
2688fcb7 | 809 | xennet_move_rx_slot(queue, skb, ref); |
0d160211 JF |
810 | err = -EINVAL; |
811 | goto next; | |
812 | } | |
813 | ||
814 | /* | |
815 | * This definitely indicates a bug, either in this driver or in | |
816 | * the backend driver. In future this should flag the bad | |
697089dc | 817 | * situation to the system controller to reboot the backend. |
0d160211 JF |
818 | */ |
819 | if (ref == GRANT_INVALID_REF) { | |
820 | if (net_ratelimit()) | |
821 | dev_warn(dev, "Bad rx response id %d.\n", | |
822 | rx->id); | |
823 | err = -EINVAL; | |
824 | goto next; | |
825 | } | |
826 | ||
827 | ret = gnttab_end_foreign_access_ref(ref, 0); | |
828 | BUG_ON(!ret); | |
829 | ||
2688fcb7 | 830 | gnttab_release_grant_reference(&queue->gref_rx_head, ref); |
0d160211 JF |
831 | |
832 | __skb_queue_tail(list, skb); | |
833 | ||
834 | next: | |
f942dc25 | 835 | if (!(rx->flags & XEN_NETRXF_more_data)) |
0d160211 JF |
836 | break; |
837 | ||
7158ff6d | 838 | if (cons + slots == rp) { |
0d160211 | 839 | if (net_ratelimit()) |
7158ff6d | 840 | dev_warn(dev, "Need more slots\n"); |
0d160211 JF |
841 | err = -ENOENT; |
842 | break; | |
843 | } | |
844 | ||
2688fcb7 AB |
845 | rx = RING_GET_RESPONSE(&queue->rx, cons + slots); |
846 | skb = xennet_get_rx_skb(queue, cons + slots); | |
847 | ref = xennet_get_rx_ref(queue, cons + slots); | |
7158ff6d | 848 | slots++; |
0d160211 JF |
849 | } |
850 | ||
7158ff6d | 851 | if (unlikely(slots > max)) { |
0d160211 | 852 | if (net_ratelimit()) |
697089dc | 853 | dev_warn(dev, "Too many slots\n"); |
0d160211 JF |
854 | err = -E2BIG; |
855 | } | |
856 | ||
857 | if (unlikely(err)) | |
2688fcb7 | 858 | queue->rx.rsp_cons = cons + slots; |
0d160211 JF |
859 | |
860 | return err; | |
861 | } | |
862 | ||
863 | static int xennet_set_skb_gso(struct sk_buff *skb, | |
864 | struct xen_netif_extra_info *gso) | |
865 | { | |
866 | if (!gso->u.gso.size) { | |
867 | if (net_ratelimit()) | |
383eda32 | 868 | pr_warn("GSO size must not be zero\n"); |
0d160211 JF |
869 | return -EINVAL; |
870 | } | |
871 | ||
2c0057de PD |
872 | if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4 && |
873 | gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV6) { | |
0d160211 | 874 | if (net_ratelimit()) |
383eda32 | 875 | pr_warn("Bad GSO type %d\n", gso->u.gso.type); |
0d160211 JF |
876 | return -EINVAL; |
877 | } | |
878 | ||
879 | skb_shinfo(skb)->gso_size = gso->u.gso.size; | |
2c0057de PD |
880 | skb_shinfo(skb)->gso_type = |
881 | (gso->u.gso.type == XEN_NETIF_GSO_TYPE_TCPV4) ? | |
882 | SKB_GSO_TCPV4 : | |
883 | SKB_GSO_TCPV6; | |
0d160211 JF |
884 | |
885 | /* Header must be checked, and gso_segs computed. */ | |
886 | skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; | |
887 | skb_shinfo(skb)->gso_segs = 0; | |
888 | ||
889 | return 0; | |
890 | } | |
891 | ||
2688fcb7 | 892 | static RING_IDX xennet_fill_frags(struct netfront_queue *queue, |
0d160211 JF |
893 | struct sk_buff *skb, |
894 | struct sk_buff_head *list) | |
895 | { | |
896 | struct skb_shared_info *shinfo = skb_shinfo(skb); | |
2688fcb7 | 897 | RING_IDX cons = queue->rx.rsp_cons; |
0d160211 JF |
898 | struct sk_buff *nskb; |
899 | ||
900 | while ((nskb = __skb_dequeue(list))) { | |
901 | struct xen_netif_rx_response *rx = | |
2688fcb7 | 902 | RING_GET_RESPONSE(&queue->rx, ++cons); |
01c68026 | 903 | skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0]; |
0d160211 | 904 | |
093b9c71 JB |
905 | if (shinfo->nr_frags == MAX_SKB_FRAGS) { |
906 | unsigned int pull_to = NETFRONT_SKB_CB(skb)->pull_to; | |
0d160211 | 907 | |
093b9c71 JB |
908 | BUG_ON(pull_to <= skb_headlen(skb)); |
909 | __pskb_pull_tail(skb, pull_to - skb_headlen(skb)); | |
910 | } | |
911 | BUG_ON(shinfo->nr_frags >= MAX_SKB_FRAGS); | |
912 | ||
913 | skb_add_rx_frag(skb, shinfo->nr_frags, skb_frag_page(nfrag), | |
914 | rx->offset, rx->status, PAGE_SIZE); | |
0d160211 JF |
915 | |
916 | skb_shinfo(nskb)->nr_frags = 0; | |
917 | kfree_skb(nskb); | |
0d160211 JF |
918 | } |
919 | ||
0d160211 JF |
920 | return cons; |
921 | } | |
922 | ||
e0ce4af9 | 923 | static int checksum_setup(struct net_device *dev, struct sk_buff *skb) |
0d160211 | 924 | { |
b5cf66cd | 925 | bool recalculate_partial_csum = false; |
e0ce4af9 IC |
926 | |
927 | /* | |
928 | * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy | |
929 | * peers can fail to set NETRXF_csum_blank when sending a GSO | |
930 | * frame. In this case force the SKB to CHECKSUM_PARTIAL and | |
931 | * recalculate the partial checksum. | |
932 | */ | |
933 | if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) { | |
934 | struct netfront_info *np = netdev_priv(dev); | |
2688fcb7 | 935 | atomic_inc(&np->rx_gso_checksum_fixup); |
e0ce4af9 | 936 | skb->ip_summed = CHECKSUM_PARTIAL; |
b5cf66cd | 937 | recalculate_partial_csum = true; |
e0ce4af9 IC |
938 | } |
939 | ||
940 | /* A non-CHECKSUM_PARTIAL SKB does not require setup. */ | |
941 | if (skb->ip_summed != CHECKSUM_PARTIAL) | |
942 | return 0; | |
0d160211 | 943 | |
b5cf66cd | 944 | return skb_checksum_setup(skb, recalculate_partial_csum); |
0d160211 JF |
945 | } |
946 | ||
2688fcb7 | 947 | static int handle_incoming_queue(struct netfront_queue *queue, |
09f75cd7 | 948 | struct sk_buff_head *rxq) |
0d160211 | 949 | { |
900e1833 | 950 | struct netfront_stats *rx_stats = this_cpu_ptr(queue->info->rx_stats); |
0d160211 JF |
951 | int packets_dropped = 0; |
952 | struct sk_buff *skb; | |
953 | ||
954 | while ((skb = __skb_dequeue(rxq)) != NULL) { | |
3683243b | 955 | int pull_to = NETFRONT_SKB_CB(skb)->pull_to; |
0d160211 | 956 | |
093b9c71 JB |
957 | if (pull_to > skb_headlen(skb)) |
958 | __pskb_pull_tail(skb, pull_to - skb_headlen(skb)); | |
0d160211 JF |
959 | |
960 | /* Ethernet work: Delayed to here as it peeks the header. */ | |
2688fcb7 | 961 | skb->protocol = eth_type_trans(skb, queue->info->netdev); |
d554f73d | 962 | skb_reset_network_header(skb); |
0d160211 | 963 | |
2688fcb7 | 964 | if (checksum_setup(queue->info->netdev, skb)) { |
e0ce4af9 IC |
965 | kfree_skb(skb); |
966 | packets_dropped++; | |
2688fcb7 | 967 | queue->info->netdev->stats.rx_errors++; |
e0ce4af9 | 968 | continue; |
0d160211 JF |
969 | } |
970 | ||
900e1833 DV |
971 | u64_stats_update_begin(&rx_stats->syncp); |
972 | rx_stats->packets++; | |
973 | rx_stats->bytes += skb->len; | |
974 | u64_stats_update_end(&rx_stats->syncp); | |
0d160211 JF |
975 | |
976 | /* Pass it up. */ | |
2688fcb7 | 977 | napi_gro_receive(&queue->napi, skb); |
0d160211 JF |
978 | } |
979 | ||
980 | return packets_dropped; | |
981 | } | |
982 | ||
bea3348e | 983 | static int xennet_poll(struct napi_struct *napi, int budget) |
0d160211 | 984 | { |
2688fcb7 AB |
985 | struct netfront_queue *queue = container_of(napi, struct netfront_queue, napi); |
986 | struct net_device *dev = queue->info->netdev; | |
0d160211 JF |
987 | struct sk_buff *skb; |
988 | struct netfront_rx_info rinfo; | |
989 | struct xen_netif_rx_response *rx = &rinfo.rx; | |
990 | struct xen_netif_extra_info *extras = rinfo.extras; | |
991 | RING_IDX i, rp; | |
bea3348e | 992 | int work_done; |
0d160211 JF |
993 | struct sk_buff_head rxq; |
994 | struct sk_buff_head errq; | |
995 | struct sk_buff_head tmpq; | |
0d160211 JF |
996 | int err; |
997 | ||
2688fcb7 | 998 | spin_lock(&queue->rx_lock); |
0d160211 | 999 | |
0d160211 JF |
1000 | skb_queue_head_init(&rxq); |
1001 | skb_queue_head_init(&errq); | |
1002 | skb_queue_head_init(&tmpq); | |
1003 | ||
2688fcb7 | 1004 | rp = queue->rx.sring->rsp_prod; |
0d160211 JF |
1005 | rmb(); /* Ensure we see queued responses up to 'rp'. */ |
1006 | ||
2688fcb7 | 1007 | i = queue->rx.rsp_cons; |
0d160211 JF |
1008 | work_done = 0; |
1009 | while ((i != rp) && (work_done < budget)) { | |
2688fcb7 | 1010 | memcpy(rx, RING_GET_RESPONSE(&queue->rx, i), sizeof(*rx)); |
0d160211 JF |
1011 | memset(extras, 0, sizeof(rinfo.extras)); |
1012 | ||
2688fcb7 | 1013 | err = xennet_get_responses(queue, &rinfo, rp, &tmpq); |
0d160211 JF |
1014 | |
1015 | if (unlikely(err)) { | |
1016 | err: | |
1017 | while ((skb = __skb_dequeue(&tmpq))) | |
1018 | __skb_queue_tail(&errq, skb); | |
09f75cd7 | 1019 | dev->stats.rx_errors++; |
2688fcb7 | 1020 | i = queue->rx.rsp_cons; |
0d160211 JF |
1021 | continue; |
1022 | } | |
1023 | ||
1024 | skb = __skb_dequeue(&tmpq); | |
1025 | ||
1026 | if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) { | |
1027 | struct xen_netif_extra_info *gso; | |
1028 | gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1]; | |
1029 | ||
1030 | if (unlikely(xennet_set_skb_gso(skb, gso))) { | |
1031 | __skb_queue_head(&tmpq, skb); | |
2688fcb7 | 1032 | queue->rx.rsp_cons += skb_queue_len(&tmpq); |
0d160211 JF |
1033 | goto err; |
1034 | } | |
1035 | } | |
1036 | ||
3683243b IC |
1037 | NETFRONT_SKB_CB(skb)->pull_to = rx->status; |
1038 | if (NETFRONT_SKB_CB(skb)->pull_to > RX_COPY_THRESHOLD) | |
1039 | NETFRONT_SKB_CB(skb)->pull_to = RX_COPY_THRESHOLD; | |
0d160211 | 1040 | |
3683243b IC |
1041 | skb_shinfo(skb)->frags[0].page_offset = rx->offset; |
1042 | skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status); | |
1043 | skb->data_len = rx->status; | |
093b9c71 | 1044 | skb->len += rx->status; |
0d160211 | 1045 | |
2688fcb7 | 1046 | i = xennet_fill_frags(queue, skb, &tmpq); |
0d160211 | 1047 | |
f942dc25 | 1048 | if (rx->flags & XEN_NETRXF_csum_blank) |
0d160211 | 1049 | skb->ip_summed = CHECKSUM_PARTIAL; |
f942dc25 | 1050 | else if (rx->flags & XEN_NETRXF_data_validated) |
0d160211 JF |
1051 | skb->ip_summed = CHECKSUM_UNNECESSARY; |
1052 | ||
1053 | __skb_queue_tail(&rxq, skb); | |
1054 | ||
2688fcb7 | 1055 | queue->rx.rsp_cons = ++i; |
0d160211 JF |
1056 | work_done++; |
1057 | } | |
1058 | ||
56cfe5d0 | 1059 | __skb_queue_purge(&errq); |
0d160211 | 1060 | |
2688fcb7 | 1061 | work_done -= handle_incoming_queue(queue, &rxq); |
0d160211 | 1062 | |
2688fcb7 | 1063 | xennet_alloc_rx_buffers(queue); |
0d160211 | 1064 | |
0d160211 | 1065 | if (work_done < budget) { |
bea3348e SH |
1066 | int more_to_do = 0; |
1067 | ||
6ad20165 | 1068 | napi_complete_done(napi, work_done); |
0d160211 | 1069 | |
2688fcb7 | 1070 | RING_FINAL_CHECK_FOR_RESPONSES(&queue->rx, more_to_do); |
6a6dc08f DV |
1071 | if (more_to_do) |
1072 | napi_schedule(napi); | |
0d160211 JF |
1073 | } |
1074 | ||
2688fcb7 | 1075 | spin_unlock(&queue->rx_lock); |
0d160211 | 1076 | |
bea3348e | 1077 | return work_done; |
0d160211 JF |
1078 | } |
1079 | ||
1080 | static int xennet_change_mtu(struct net_device *dev, int mtu) | |
1081 | { | |
0c36820e | 1082 | int max = xennet_can_sg(dev) ? XEN_NETIF_MAX_TX_SIZE : ETH_DATA_LEN; |
0d160211 JF |
1083 | |
1084 | if (mtu > max) | |
1085 | return -EINVAL; | |
1086 | dev->mtu = mtu; | |
1087 | return 0; | |
1088 | } | |
1089 | ||
bc1f4470 | 1090 | static void xennet_get_stats64(struct net_device *dev, |
1091 | struct rtnl_link_stats64 *tot) | |
e00f85be | 1092 | { |
1093 | struct netfront_info *np = netdev_priv(dev); | |
1094 | int cpu; | |
1095 | ||
1096 | for_each_possible_cpu(cpu) { | |
900e1833 DV |
1097 | struct netfront_stats *rx_stats = per_cpu_ptr(np->rx_stats, cpu); |
1098 | struct netfront_stats *tx_stats = per_cpu_ptr(np->tx_stats, cpu); | |
e00f85be | 1099 | u64 rx_packets, rx_bytes, tx_packets, tx_bytes; |
1100 | unsigned int start; | |
1101 | ||
1102 | do { | |
900e1833 DV |
1103 | start = u64_stats_fetch_begin_irq(&tx_stats->syncp); |
1104 | tx_packets = tx_stats->packets; | |
1105 | tx_bytes = tx_stats->bytes; | |
1106 | } while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start)); | |
e00f85be | 1107 | |
900e1833 DV |
1108 | do { |
1109 | start = u64_stats_fetch_begin_irq(&rx_stats->syncp); | |
1110 | rx_packets = rx_stats->packets; | |
1111 | rx_bytes = rx_stats->bytes; | |
1112 | } while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start)); | |
e00f85be | 1113 | |
1114 | tot->rx_packets += rx_packets; | |
1115 | tot->tx_packets += tx_packets; | |
1116 | tot->rx_bytes += rx_bytes; | |
1117 | tot->tx_bytes += tx_bytes; | |
1118 | } | |
1119 | ||
1120 | tot->rx_errors = dev->stats.rx_errors; | |
1121 | tot->tx_dropped = dev->stats.tx_dropped; | |
e00f85be | 1122 | } |
1123 | ||
2688fcb7 | 1124 | static void xennet_release_tx_bufs(struct netfront_queue *queue) |
0d160211 JF |
1125 | { |
1126 | struct sk_buff *skb; | |
1127 | int i; | |
1128 | ||
1129 | for (i = 0; i < NET_TX_RING_SIZE; i++) { | |
1130 | /* Skip over entries which are actually freelist references */ | |
2688fcb7 | 1131 | if (skb_entry_is_link(&queue->tx_skbs[i])) |
0d160211 JF |
1132 | continue; |
1133 | ||
2688fcb7 AB |
1134 | skb = queue->tx_skbs[i].skb; |
1135 | get_page(queue->grant_tx_page[i]); | |
1136 | gnttab_end_foreign_access(queue->grant_tx_ref[i], | |
cefe0078 | 1137 | GNTMAP_readonly, |
2688fcb7 AB |
1138 | (unsigned long)page_address(queue->grant_tx_page[i])); |
1139 | queue->grant_tx_page[i] = NULL; | |
1140 | queue->grant_tx_ref[i] = GRANT_INVALID_REF; | |
1141 | add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, i); | |
0d160211 JF |
1142 | dev_kfree_skb_irq(skb); |
1143 | } | |
1144 | } | |
1145 | ||
2688fcb7 | 1146 | static void xennet_release_rx_bufs(struct netfront_queue *queue) |
0d160211 | 1147 | { |
0d160211 JF |
1148 | int id, ref; |
1149 | ||
2688fcb7 | 1150 | spin_lock_bh(&queue->rx_lock); |
0d160211 JF |
1151 | |
1152 | for (id = 0; id < NET_RX_RING_SIZE; id++) { | |
cefe0078 AL |
1153 | struct sk_buff *skb; |
1154 | struct page *page; | |
0d160211 | 1155 | |
2688fcb7 | 1156 | skb = queue->rx_skbs[id]; |
cefe0078 | 1157 | if (!skb) |
0d160211 | 1158 | continue; |
0d160211 | 1159 | |
2688fcb7 | 1160 | ref = queue->grant_rx_ref[id]; |
cefe0078 AL |
1161 | if (ref == GRANT_INVALID_REF) |
1162 | continue; | |
0d160211 | 1163 | |
cefe0078 | 1164 | page = skb_frag_page(&skb_shinfo(skb)->frags[0]); |
0d160211 | 1165 | |
cefe0078 AL |
1166 | /* gnttab_end_foreign_access() needs a page ref until |
1167 | * foreign access is ended (which may be deferred). | |
1168 | */ | |
1169 | get_page(page); | |
1170 | gnttab_end_foreign_access(ref, 0, | |
1171 | (unsigned long)page_address(page)); | |
2688fcb7 | 1172 | queue->grant_rx_ref[id] = GRANT_INVALID_REF; |
0d160211 | 1173 | |
cefe0078 | 1174 | kfree_skb(skb); |
0d160211 JF |
1175 | } |
1176 | ||
2688fcb7 | 1177 | spin_unlock_bh(&queue->rx_lock); |
0d160211 JF |
1178 | } |
1179 | ||
c8f44aff MM |
1180 | static netdev_features_t xennet_fix_features(struct net_device *dev, |
1181 | netdev_features_t features) | |
8f7b01a1 ED |
1182 | { |
1183 | struct netfront_info *np = netdev_priv(dev); | |
8f7b01a1 | 1184 | |
2890ea5c JG |
1185 | if (features & NETIF_F_SG && |
1186 | !xenbus_read_unsigned(np->xbdev->otherend, "feature-sg", 0)) | |
1187 | features &= ~NETIF_F_SG; | |
8f7b01a1 | 1188 | |
2890ea5c JG |
1189 | if (features & NETIF_F_IPV6_CSUM && |
1190 | !xenbus_read_unsigned(np->xbdev->otherend, | |
1191 | "feature-ipv6-csum-offload", 0)) | |
1192 | features &= ~NETIF_F_IPV6_CSUM; | |
8f7b01a1 | 1193 | |
2890ea5c JG |
1194 | if (features & NETIF_F_TSO && |
1195 | !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv4", 0)) | |
1196 | features &= ~NETIF_F_TSO; | |
8f7b01a1 | 1197 | |
2890ea5c JG |
1198 | if (features & NETIF_F_TSO6 && |
1199 | !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv6", 0)) | |
1200 | features &= ~NETIF_F_TSO6; | |
2c0057de | 1201 | |
8f7b01a1 ED |
1202 | return features; |
1203 | } | |
1204 | ||
c8f44aff MM |
1205 | static int xennet_set_features(struct net_device *dev, |
1206 | netdev_features_t features) | |
8f7b01a1 ED |
1207 | { |
1208 | if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) { | |
1209 | netdev_info(dev, "Reducing MTU because no SG offload"); | |
1210 | dev->mtu = ETH_DATA_LEN; | |
1211 | } | |
1212 | ||
1213 | return 0; | |
1214 | } | |
1215 | ||
d634bf2c | 1216 | static irqreturn_t xennet_tx_interrupt(int irq, void *dev_id) |
cf66f9d4 | 1217 | { |
2688fcb7 | 1218 | struct netfront_queue *queue = dev_id; |
cf66f9d4 KRW |
1219 | unsigned long flags; |
1220 | ||
2688fcb7 AB |
1221 | spin_lock_irqsave(&queue->tx_lock, flags); |
1222 | xennet_tx_buf_gc(queue); | |
1223 | spin_unlock_irqrestore(&queue->tx_lock, flags); | |
cf66f9d4 | 1224 | |
d634bf2c WL |
1225 | return IRQ_HANDLED; |
1226 | } | |
1227 | ||
1228 | static irqreturn_t xennet_rx_interrupt(int irq, void *dev_id) | |
1229 | { | |
2688fcb7 AB |
1230 | struct netfront_queue *queue = dev_id; |
1231 | struct net_device *dev = queue->info->netdev; | |
d634bf2c WL |
1232 | |
1233 | if (likely(netif_carrier_ok(dev) && | |
2688fcb7 | 1234 | RING_HAS_UNCONSUMED_RESPONSES(&queue->rx))) |
76541869 | 1235 | napi_schedule(&queue->napi); |
cf66f9d4 | 1236 | |
d634bf2c WL |
1237 | return IRQ_HANDLED; |
1238 | } | |
cf66f9d4 | 1239 | |
d634bf2c WL |
1240 | static irqreturn_t xennet_interrupt(int irq, void *dev_id) |
1241 | { | |
1242 | xennet_tx_interrupt(irq, dev_id); | |
1243 | xennet_rx_interrupt(irq, dev_id); | |
cf66f9d4 KRW |
1244 | return IRQ_HANDLED; |
1245 | } | |
1246 | ||
1247 | #ifdef CONFIG_NET_POLL_CONTROLLER | |
1248 | static void xennet_poll_controller(struct net_device *dev) | |
1249 | { | |
2688fcb7 AB |
1250 | /* Poll each queue */ |
1251 | struct netfront_info *info = netdev_priv(dev); | |
1252 | unsigned int num_queues = dev->real_num_tx_queues; | |
1253 | unsigned int i; | |
1254 | for (i = 0; i < num_queues; ++i) | |
1255 | xennet_interrupt(0, &info->queues[i]); | |
cf66f9d4 KRW |
1256 | } |
1257 | #endif | |
1258 | ||
0a0b9d2e SH |
1259 | static const struct net_device_ops xennet_netdev_ops = { |
1260 | .ndo_open = xennet_open, | |
0a0b9d2e SH |
1261 | .ndo_stop = xennet_close, |
1262 | .ndo_start_xmit = xennet_start_xmit, | |
1263 | .ndo_change_mtu = xennet_change_mtu, | |
e00f85be | 1264 | .ndo_get_stats64 = xennet_get_stats64, |
0a0b9d2e SH |
1265 | .ndo_set_mac_address = eth_mac_addr, |
1266 | .ndo_validate_addr = eth_validate_addr, | |
fb507934 MM |
1267 | .ndo_fix_features = xennet_fix_features, |
1268 | .ndo_set_features = xennet_set_features, | |
2688fcb7 | 1269 | .ndo_select_queue = xennet_select_queue, |
cf66f9d4 KRW |
1270 | #ifdef CONFIG_NET_POLL_CONTROLLER |
1271 | .ndo_poll_controller = xennet_poll_controller, | |
1272 | #endif | |
0a0b9d2e SH |
1273 | }; |
1274 | ||
900e1833 DV |
1275 | static void xennet_free_netdev(struct net_device *netdev) |
1276 | { | |
1277 | struct netfront_info *np = netdev_priv(netdev); | |
1278 | ||
1279 | free_percpu(np->rx_stats); | |
1280 | free_percpu(np->tx_stats); | |
1281 | free_netdev(netdev); | |
1282 | } | |
1283 | ||
8e0e46bb | 1284 | static struct net_device *xennet_create_dev(struct xenbus_device *dev) |
0d160211 | 1285 | { |
2688fcb7 | 1286 | int err; |
0d160211 JF |
1287 | struct net_device *netdev; |
1288 | struct netfront_info *np; | |
1289 | ||
50ee6061 | 1290 | netdev = alloc_etherdev_mq(sizeof(struct netfront_info), xennet_max_queues); |
41de8d4c | 1291 | if (!netdev) |
0d160211 | 1292 | return ERR_PTR(-ENOMEM); |
0d160211 JF |
1293 | |
1294 | np = netdev_priv(netdev); | |
1295 | np->xbdev = dev; | |
1296 | ||
2688fcb7 | 1297 | np->queues = NULL; |
0d160211 | 1298 | |
e00f85be | 1299 | err = -ENOMEM; |
900e1833 DV |
1300 | np->rx_stats = netdev_alloc_pcpu_stats(struct netfront_stats); |
1301 | if (np->rx_stats == NULL) | |
1302 | goto exit; | |
1303 | np->tx_stats = netdev_alloc_pcpu_stats(struct netfront_stats); | |
1304 | if (np->tx_stats == NULL) | |
e00f85be | 1305 | goto exit; |
1306 | ||
0a0b9d2e SH |
1307 | netdev->netdev_ops = &xennet_netdev_ops; |
1308 | ||
fb507934 MM |
1309 | netdev->features = NETIF_F_IP_CSUM | NETIF_F_RXCSUM | |
1310 | NETIF_F_GSO_ROBUST; | |
2c0057de PD |
1311 | netdev->hw_features = NETIF_F_SG | |
1312 | NETIF_F_IPV6_CSUM | | |
1313 | NETIF_F_TSO | NETIF_F_TSO6; | |
0d160211 | 1314 | |
fc3e5941 IC |
1315 | /* |
1316 | * Assume that all hw features are available for now. This set | |
1317 | * will be adjusted by the call to netdev_update_features() in | |
1318 | * xennet_connect() which is the earliest point where we can | |
1319 | * negotiate with the backend regarding supported features. | |
1320 | */ | |
1321 | netdev->features |= netdev->hw_features; | |
1322 | ||
7ad24ea4 | 1323 | netdev->ethtool_ops = &xennet_ethtool_ops; |
e1043a4b | 1324 | netdev->min_mtu = ETH_MIN_MTU; |
d0c2c997 | 1325 | netdev->max_mtu = XEN_NETIF_MAX_TX_SIZE; |
0d160211 JF |
1326 | SET_NETDEV_DEV(netdev, &dev->dev); |
1327 | ||
1328 | np->netdev = netdev; | |
1329 | ||
1330 | netif_carrier_off(netdev); | |
1331 | ||
b707fda2 | 1332 | xenbus_switch_state(dev, XenbusStateInitialising); |
0d160211 JF |
1333 | return netdev; |
1334 | ||
0d160211 | 1335 | exit: |
900e1833 | 1336 | xennet_free_netdev(netdev); |
0d160211 JF |
1337 | return ERR_PTR(err); |
1338 | } | |
1339 | ||
1340 | /** | |
1341 | * Entry point to this code when a new device is created. Allocate the basic | |
1342 | * structures and the ring buffers for communication with the backend, and | |
1343 | * inform the backend of the appropriate details for those. | |
1344 | */ | |
8e0e46bb | 1345 | static int netfront_probe(struct xenbus_device *dev, |
1dd06ae8 | 1346 | const struct xenbus_device_id *id) |
0d160211 JF |
1347 | { |
1348 | int err; | |
1349 | struct net_device *netdev; | |
1350 | struct netfront_info *info; | |
1351 | ||
1352 | netdev = xennet_create_dev(dev); | |
1353 | if (IS_ERR(netdev)) { | |
1354 | err = PTR_ERR(netdev); | |
1355 | xenbus_dev_fatal(dev, err, "creating netdev"); | |
1356 | return err; | |
1357 | } | |
1358 | ||
1359 | info = netdev_priv(netdev); | |
1b713e00 | 1360 | dev_set_drvdata(&dev->dev, info); |
27b917e5 TI |
1361 | #ifdef CONFIG_SYSFS |
1362 | info->netdev->sysfs_groups[0] = &xennet_dev_group; | |
1363 | #endif | |
0d160211 | 1364 | |
0d160211 | 1365 | return 0; |
0d160211 JF |
1366 | } |
1367 | ||
1368 | static void xennet_end_access(int ref, void *page) | |
1369 | { | |
1370 | /* This frees the page as a side-effect */ | |
1371 | if (ref != GRANT_INVALID_REF) | |
1372 | gnttab_end_foreign_access(ref, 0, (unsigned long)page); | |
1373 | } | |
1374 | ||
1375 | static void xennet_disconnect_backend(struct netfront_info *info) | |
1376 | { | |
2688fcb7 | 1377 | unsigned int i = 0; |
2688fcb7 AB |
1378 | unsigned int num_queues = info->netdev->real_num_tx_queues; |
1379 | ||
f9feb1e6 DV |
1380 | netif_carrier_off(info->netdev); |
1381 | ||
9a873c71 | 1382 | for (i = 0; i < num_queues && info->queues; ++i) { |
76541869 DV |
1383 | struct netfront_queue *queue = &info->queues[i]; |
1384 | ||
74470954 BO |
1385 | del_timer_sync(&queue->rx_refill_timer); |
1386 | ||
2688fcb7 AB |
1387 | if (queue->tx_irq && (queue->tx_irq == queue->rx_irq)) |
1388 | unbind_from_irqhandler(queue->tx_irq, queue); | |
1389 | if (queue->tx_irq && (queue->tx_irq != queue->rx_irq)) { | |
1390 | unbind_from_irqhandler(queue->tx_irq, queue); | |
1391 | unbind_from_irqhandler(queue->rx_irq, queue); | |
1392 | } | |
1393 | queue->tx_evtchn = queue->rx_evtchn = 0; | |
1394 | queue->tx_irq = queue->rx_irq = 0; | |
0d160211 | 1395 | |
274b0455 CW |
1396 | if (netif_running(info->netdev)) |
1397 | napi_synchronize(&queue->napi); | |
f9feb1e6 | 1398 | |
a5b5dc3c DV |
1399 | xennet_release_tx_bufs(queue); |
1400 | xennet_release_rx_bufs(queue); | |
1401 | gnttab_free_grant_references(queue->gref_tx_head); | |
1402 | gnttab_free_grant_references(queue->gref_rx_head); | |
1403 | ||
2688fcb7 AB |
1404 | /* End access and free the pages */ |
1405 | xennet_end_access(queue->tx_ring_ref, queue->tx.sring); | |
1406 | xennet_end_access(queue->rx_ring_ref, queue->rx.sring); | |
0d160211 | 1407 | |
2688fcb7 AB |
1408 | queue->tx_ring_ref = GRANT_INVALID_REF; |
1409 | queue->rx_ring_ref = GRANT_INVALID_REF; | |
1410 | queue->tx.sring = NULL; | |
1411 | queue->rx.sring = NULL; | |
1412 | } | |
0d160211 JF |
1413 | } |
1414 | ||
1415 | /** | |
1416 | * We are reconnecting to the backend, due to a suspend/resume, or a backend | |
1417 | * driver restart. We tear down our netif structure and recreate it, but | |
1418 | * leave the device-layer structures intact so that this is transparent to the | |
1419 | * rest of the kernel. | |
1420 | */ | |
1421 | static int netfront_resume(struct xenbus_device *dev) | |
1422 | { | |
1b713e00 | 1423 | struct netfront_info *info = dev_get_drvdata(&dev->dev); |
0d160211 JF |
1424 | |
1425 | dev_dbg(&dev->dev, "%s\n", dev->nodename); | |
1426 | ||
1427 | xennet_disconnect_backend(info); | |
1428 | return 0; | |
1429 | } | |
1430 | ||
1431 | static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[]) | |
1432 | { | |
1433 | char *s, *e, *macstr; | |
1434 | int i; | |
1435 | ||
1436 | macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL); | |
1437 | if (IS_ERR(macstr)) | |
1438 | return PTR_ERR(macstr); | |
1439 | ||
1440 | for (i = 0; i < ETH_ALEN; i++) { | |
1441 | mac[i] = simple_strtoul(s, &e, 16); | |
1442 | if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) { | |
1443 | kfree(macstr); | |
1444 | return -ENOENT; | |
1445 | } | |
1446 | s = e+1; | |
1447 | } | |
1448 | ||
1449 | kfree(macstr); | |
1450 | return 0; | |
1451 | } | |
1452 | ||
2688fcb7 | 1453 | static int setup_netfront_single(struct netfront_queue *queue) |
d634bf2c WL |
1454 | { |
1455 | int err; | |
1456 | ||
2688fcb7 | 1457 | err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn); |
d634bf2c WL |
1458 | if (err < 0) |
1459 | goto fail; | |
1460 | ||
2688fcb7 | 1461 | err = bind_evtchn_to_irqhandler(queue->tx_evtchn, |
d634bf2c | 1462 | xennet_interrupt, |
2688fcb7 | 1463 | 0, queue->info->netdev->name, queue); |
d634bf2c WL |
1464 | if (err < 0) |
1465 | goto bind_fail; | |
2688fcb7 AB |
1466 | queue->rx_evtchn = queue->tx_evtchn; |
1467 | queue->rx_irq = queue->tx_irq = err; | |
d634bf2c WL |
1468 | |
1469 | return 0; | |
1470 | ||
1471 | bind_fail: | |
2688fcb7 AB |
1472 | xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn); |
1473 | queue->tx_evtchn = 0; | |
d634bf2c WL |
1474 | fail: |
1475 | return err; | |
1476 | } | |
1477 | ||
2688fcb7 | 1478 | static int setup_netfront_split(struct netfront_queue *queue) |
d634bf2c WL |
1479 | { |
1480 | int err; | |
1481 | ||
2688fcb7 | 1482 | err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn); |
d634bf2c WL |
1483 | if (err < 0) |
1484 | goto fail; | |
2688fcb7 | 1485 | err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->rx_evtchn); |
d634bf2c WL |
1486 | if (err < 0) |
1487 | goto alloc_rx_evtchn_fail; | |
1488 | ||
2688fcb7 AB |
1489 | snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name), |
1490 | "%s-tx", queue->name); | |
1491 | err = bind_evtchn_to_irqhandler(queue->tx_evtchn, | |
d634bf2c | 1492 | xennet_tx_interrupt, |
2688fcb7 | 1493 | 0, queue->tx_irq_name, queue); |
d634bf2c WL |
1494 | if (err < 0) |
1495 | goto bind_tx_fail; | |
2688fcb7 | 1496 | queue->tx_irq = err; |
d634bf2c | 1497 | |
2688fcb7 AB |
1498 | snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name), |
1499 | "%s-rx", queue->name); | |
1500 | err = bind_evtchn_to_irqhandler(queue->rx_evtchn, | |
d634bf2c | 1501 | xennet_rx_interrupt, |
2688fcb7 | 1502 | 0, queue->rx_irq_name, queue); |
d634bf2c WL |
1503 | if (err < 0) |
1504 | goto bind_rx_fail; | |
2688fcb7 | 1505 | queue->rx_irq = err; |
d634bf2c WL |
1506 | |
1507 | return 0; | |
1508 | ||
1509 | bind_rx_fail: | |
2688fcb7 AB |
1510 | unbind_from_irqhandler(queue->tx_irq, queue); |
1511 | queue->tx_irq = 0; | |
d634bf2c | 1512 | bind_tx_fail: |
2688fcb7 AB |
1513 | xenbus_free_evtchn(queue->info->xbdev, queue->rx_evtchn); |
1514 | queue->rx_evtchn = 0; | |
d634bf2c | 1515 | alloc_rx_evtchn_fail: |
2688fcb7 AB |
1516 | xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn); |
1517 | queue->tx_evtchn = 0; | |
d634bf2c WL |
1518 | fail: |
1519 | return err; | |
1520 | } | |
1521 | ||
2688fcb7 AB |
1522 | static int setup_netfront(struct xenbus_device *dev, |
1523 | struct netfront_queue *queue, unsigned int feature_split_evtchn) | |
0d160211 JF |
1524 | { |
1525 | struct xen_netif_tx_sring *txs; | |
1526 | struct xen_netif_rx_sring *rxs; | |
ccc9d90a | 1527 | grant_ref_t gref; |
0d160211 | 1528 | int err; |
0d160211 | 1529 | |
2688fcb7 AB |
1530 | queue->tx_ring_ref = GRANT_INVALID_REF; |
1531 | queue->rx_ring_ref = GRANT_INVALID_REF; | |
1532 | queue->rx.sring = NULL; | |
1533 | queue->tx.sring = NULL; | |
0d160211 | 1534 | |
a144ff09 | 1535 | txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH); |
0d160211 JF |
1536 | if (!txs) { |
1537 | err = -ENOMEM; | |
1538 | xenbus_dev_fatal(dev, err, "allocating tx ring page"); | |
1539 | goto fail; | |
1540 | } | |
1541 | SHARED_RING_INIT(txs); | |
30c5d7f0 | 1542 | FRONT_RING_INIT(&queue->tx, txs, XEN_PAGE_SIZE); |
0d160211 | 1543 | |
ccc9d90a | 1544 | err = xenbus_grant_ring(dev, txs, 1, &gref); |
1ca2983a WL |
1545 | if (err < 0) |
1546 | goto grant_tx_ring_fail; | |
ccc9d90a | 1547 | queue->tx_ring_ref = gref; |
0d160211 | 1548 | |
a144ff09 | 1549 | rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH); |
0d160211 JF |
1550 | if (!rxs) { |
1551 | err = -ENOMEM; | |
1552 | xenbus_dev_fatal(dev, err, "allocating rx ring page"); | |
1ca2983a | 1553 | goto alloc_rx_ring_fail; |
0d160211 JF |
1554 | } |
1555 | SHARED_RING_INIT(rxs); | |
30c5d7f0 | 1556 | FRONT_RING_INIT(&queue->rx, rxs, XEN_PAGE_SIZE); |
0d160211 | 1557 | |
ccc9d90a | 1558 | err = xenbus_grant_ring(dev, rxs, 1, &gref); |
1ca2983a WL |
1559 | if (err < 0) |
1560 | goto grant_rx_ring_fail; | |
ccc9d90a | 1561 | queue->rx_ring_ref = gref; |
0d160211 | 1562 | |
d634bf2c | 1563 | if (feature_split_evtchn) |
2688fcb7 | 1564 | err = setup_netfront_split(queue); |
d634bf2c WL |
1565 | /* setup single event channel if |
1566 | * a) feature-split-event-channels == 0 | |
1567 | * b) feature-split-event-channels == 1 but failed to setup | |
1568 | */ | |
1569 | if (!feature_split_evtchn || (feature_split_evtchn && err)) | |
2688fcb7 | 1570 | err = setup_netfront_single(queue); |
d634bf2c | 1571 | |
0d160211 | 1572 | if (err) |
1ca2983a | 1573 | goto alloc_evtchn_fail; |
0d160211 | 1574 | |
0d160211 JF |
1575 | return 0; |
1576 | ||
1ca2983a WL |
1577 | /* If we fail to setup netfront, it is safe to just revoke access to |
1578 | * granted pages because backend is not accessing it at this point. | |
1579 | */ | |
1ca2983a | 1580 | alloc_evtchn_fail: |
2688fcb7 | 1581 | gnttab_end_foreign_access_ref(queue->rx_ring_ref, 0); |
1ca2983a WL |
1582 | grant_rx_ring_fail: |
1583 | free_page((unsigned long)rxs); | |
1584 | alloc_rx_ring_fail: | |
2688fcb7 | 1585 | gnttab_end_foreign_access_ref(queue->tx_ring_ref, 0); |
1ca2983a WL |
1586 | grant_tx_ring_fail: |
1587 | free_page((unsigned long)txs); | |
1588 | fail: | |
0d160211 JF |
1589 | return err; |
1590 | } | |
1591 | ||
2688fcb7 AB |
1592 | /* Queue-specific initialisation |
1593 | * This used to be done in xennet_create_dev() but must now | |
1594 | * be run per-queue. | |
1595 | */ | |
1596 | static int xennet_init_queue(struct netfront_queue *queue) | |
1597 | { | |
1598 | unsigned short i; | |
1599 | int err = 0; | |
1600 | ||
1601 | spin_lock_init(&queue->tx_lock); | |
1602 | spin_lock_init(&queue->rx_lock); | |
1603 | ||
e99e88a9 | 1604 | timer_setup(&queue->rx_refill_timer, rx_refill_timeout, 0); |
2688fcb7 | 1605 | |
8b715010 WL |
1606 | snprintf(queue->name, sizeof(queue->name), "%s-q%u", |
1607 | queue->info->netdev->name, queue->id); | |
1608 | ||
2688fcb7 AB |
1609 | /* Initialise tx_skbs as a free chain containing every entry. */ |
1610 | queue->tx_skb_freelist = 0; | |
1611 | for (i = 0; i < NET_TX_RING_SIZE; i++) { | |
1612 | skb_entry_set_link(&queue->tx_skbs[i], i+1); | |
1613 | queue->grant_tx_ref[i] = GRANT_INVALID_REF; | |
1614 | queue->grant_tx_page[i] = NULL; | |
1615 | } | |
1616 | ||
1617 | /* Clear out rx_skbs */ | |
1618 | for (i = 0; i < NET_RX_RING_SIZE; i++) { | |
1619 | queue->rx_skbs[i] = NULL; | |
1620 | queue->grant_rx_ref[i] = GRANT_INVALID_REF; | |
1621 | } | |
1622 | ||
1623 | /* A grant for every tx ring slot */ | |
1f3c2eba | 1624 | if (gnttab_alloc_grant_references(NET_TX_RING_SIZE, |
2688fcb7 AB |
1625 | &queue->gref_tx_head) < 0) { |
1626 | pr_alert("can't alloc tx grant refs\n"); | |
1627 | err = -ENOMEM; | |
1628 | goto exit; | |
1629 | } | |
1630 | ||
1631 | /* A grant for every rx ring slot */ | |
1f3c2eba | 1632 | if (gnttab_alloc_grant_references(NET_RX_RING_SIZE, |
2688fcb7 AB |
1633 | &queue->gref_rx_head) < 0) { |
1634 | pr_alert("can't alloc rx grant refs\n"); | |
1635 | err = -ENOMEM; | |
1636 | goto exit_free_tx; | |
1637 | } | |
1638 | ||
2688fcb7 AB |
1639 | return 0; |
1640 | ||
1641 | exit_free_tx: | |
1642 | gnttab_free_grant_references(queue->gref_tx_head); | |
1643 | exit: | |
1644 | return err; | |
1645 | } | |
1646 | ||
50ee6061 AB |
1647 | static int write_queue_xenstore_keys(struct netfront_queue *queue, |
1648 | struct xenbus_transaction *xbt, int write_hierarchical) | |
1649 | { | |
1650 | /* Write the queue-specific keys into XenStore in the traditional | |
1651 | * way for a single queue, or in a queue subkeys for multiple | |
1652 | * queues. | |
1653 | */ | |
1654 | struct xenbus_device *dev = queue->info->xbdev; | |
1655 | int err; | |
1656 | const char *message; | |
1657 | char *path; | |
1658 | size_t pathsize; | |
1659 | ||
1660 | /* Choose the correct place to write the keys */ | |
1661 | if (write_hierarchical) { | |
1662 | pathsize = strlen(dev->nodename) + 10; | |
1663 | path = kzalloc(pathsize, GFP_KERNEL); | |
1664 | if (!path) { | |
1665 | err = -ENOMEM; | |
1666 | message = "out of memory while writing ring references"; | |
1667 | goto error; | |
1668 | } | |
1669 | snprintf(path, pathsize, "%s/queue-%u", | |
1670 | dev->nodename, queue->id); | |
1671 | } else { | |
1672 | path = (char *)dev->nodename; | |
1673 | } | |
1674 | ||
1675 | /* Write ring references */ | |
1676 | err = xenbus_printf(*xbt, path, "tx-ring-ref", "%u", | |
1677 | queue->tx_ring_ref); | |
1678 | if (err) { | |
1679 | message = "writing tx-ring-ref"; | |
1680 | goto error; | |
1681 | } | |
1682 | ||
1683 | err = xenbus_printf(*xbt, path, "rx-ring-ref", "%u", | |
1684 | queue->rx_ring_ref); | |
1685 | if (err) { | |
1686 | message = "writing rx-ring-ref"; | |
1687 | goto error; | |
1688 | } | |
1689 | ||
1690 | /* Write event channels; taking into account both shared | |
1691 | * and split event channel scenarios. | |
1692 | */ | |
1693 | if (queue->tx_evtchn == queue->rx_evtchn) { | |
1694 | /* Shared event channel */ | |
1695 | err = xenbus_printf(*xbt, path, | |
1696 | "event-channel", "%u", queue->tx_evtchn); | |
1697 | if (err) { | |
1698 | message = "writing event-channel"; | |
1699 | goto error; | |
1700 | } | |
1701 | } else { | |
1702 | /* Split event channels */ | |
1703 | err = xenbus_printf(*xbt, path, | |
1704 | "event-channel-tx", "%u", queue->tx_evtchn); | |
1705 | if (err) { | |
1706 | message = "writing event-channel-tx"; | |
1707 | goto error; | |
1708 | } | |
1709 | ||
1710 | err = xenbus_printf(*xbt, path, | |
1711 | "event-channel-rx", "%u", queue->rx_evtchn); | |
1712 | if (err) { | |
1713 | message = "writing event-channel-rx"; | |
1714 | goto error; | |
1715 | } | |
1716 | } | |
1717 | ||
1718 | if (write_hierarchical) | |
1719 | kfree(path); | |
1720 | return 0; | |
1721 | ||
1722 | error: | |
1723 | if (write_hierarchical) | |
1724 | kfree(path); | |
1725 | xenbus_dev_fatal(dev, err, "%s", message); | |
1726 | return err; | |
1727 | } | |
1728 | ||
ce58725f DV |
1729 | static void xennet_destroy_queues(struct netfront_info *info) |
1730 | { | |
1731 | unsigned int i; | |
1732 | ||
ce58725f DV |
1733 | for (i = 0; i < info->netdev->real_num_tx_queues; i++) { |
1734 | struct netfront_queue *queue = &info->queues[i]; | |
1735 | ||
1736 | if (netif_running(info->netdev)) | |
1737 | napi_disable(&queue->napi); | |
1738 | netif_napi_del(&queue->napi); | |
1739 | } | |
1740 | ||
ce58725f DV |
1741 | kfree(info->queues); |
1742 | info->queues = NULL; | |
1743 | } | |
1744 | ||
1745 | static int xennet_create_queues(struct netfront_info *info, | |
ca88ea12 | 1746 | unsigned int *num_queues) |
ce58725f DV |
1747 | { |
1748 | unsigned int i; | |
1749 | int ret; | |
1750 | ||
ca88ea12 | 1751 | info->queues = kcalloc(*num_queues, sizeof(struct netfront_queue), |
ce58725f DV |
1752 | GFP_KERNEL); |
1753 | if (!info->queues) | |
1754 | return -ENOMEM; | |
1755 | ||
ca88ea12 | 1756 | for (i = 0; i < *num_queues; i++) { |
ce58725f DV |
1757 | struct netfront_queue *queue = &info->queues[i]; |
1758 | ||
1759 | queue->id = i; | |
1760 | queue->info = info; | |
1761 | ||
1762 | ret = xennet_init_queue(queue); | |
1763 | if (ret < 0) { | |
f599c64f | 1764 | dev_warn(&info->xbdev->dev, |
69cb8524 | 1765 | "only created %d queues\n", i); |
ca88ea12 | 1766 | *num_queues = i; |
ce58725f DV |
1767 | break; |
1768 | } | |
1769 | ||
1770 | netif_napi_add(queue->info->netdev, &queue->napi, | |
1771 | xennet_poll, 64); | |
1772 | if (netif_running(info->netdev)) | |
1773 | napi_enable(&queue->napi); | |
1774 | } | |
1775 | ||
ca88ea12 | 1776 | netif_set_real_num_tx_queues(info->netdev, *num_queues); |
ce58725f | 1777 | |
ca88ea12 | 1778 | if (*num_queues == 0) { |
f599c64f | 1779 | dev_err(&info->xbdev->dev, "no queues\n"); |
ce58725f DV |
1780 | return -EINVAL; |
1781 | } | |
1782 | return 0; | |
1783 | } | |
1784 | ||
0d160211 | 1785 | /* Common code used when first setting up, and when resuming. */ |
f502bf2b | 1786 | static int talk_to_netback(struct xenbus_device *dev, |
0d160211 JF |
1787 | struct netfront_info *info) |
1788 | { | |
1789 | const char *message; | |
1790 | struct xenbus_transaction xbt; | |
1791 | int err; | |
2688fcb7 AB |
1792 | unsigned int feature_split_evtchn; |
1793 | unsigned int i = 0; | |
50ee6061 | 1794 | unsigned int max_queues = 0; |
2688fcb7 AB |
1795 | struct netfront_queue *queue = NULL; |
1796 | unsigned int num_queues = 1; | |
0d160211 | 1797 | |
2688fcb7 AB |
1798 | info->netdev->irq = 0; |
1799 | ||
50ee6061 | 1800 | /* Check if backend supports multiple queues */ |
2890ea5c JG |
1801 | max_queues = xenbus_read_unsigned(info->xbdev->otherend, |
1802 | "multi-queue-max-queues", 1); | |
50ee6061 AB |
1803 | num_queues = min(max_queues, xennet_max_queues); |
1804 | ||
2688fcb7 | 1805 | /* Check feature-split-event-channels */ |
2890ea5c JG |
1806 | feature_split_evtchn = xenbus_read_unsigned(info->xbdev->otherend, |
1807 | "feature-split-event-channels", 0); | |
2688fcb7 AB |
1808 | |
1809 | /* Read mac addr. */ | |
1810 | err = xen_net_read_mac(dev, info->netdev->dev_addr); | |
1811 | if (err) { | |
1812 | xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename); | |
1813 | goto out; | |
1814 | } | |
1815 | ||
f599c64f | 1816 | rtnl_lock(); |
ce58725f DV |
1817 | if (info->queues) |
1818 | xennet_destroy_queues(info); | |
1819 | ||
ca88ea12 | 1820 | err = xennet_create_queues(info, &num_queues); |
e2e004ac RL |
1821 | if (err < 0) { |
1822 | xenbus_dev_fatal(dev, err, "creating queues"); | |
1823 | kfree(info->queues); | |
1824 | info->queues = NULL; | |
1825 | goto out; | |
1826 | } | |
f599c64f | 1827 | rtnl_unlock(); |
2688fcb7 AB |
1828 | |
1829 | /* Create shared ring, alloc event channel -- for each queue */ | |
1830 | for (i = 0; i < num_queues; ++i) { | |
1831 | queue = &info->queues[i]; | |
2688fcb7 | 1832 | err = setup_netfront(dev, queue, feature_split_evtchn); |
e2e004ac RL |
1833 | if (err) |
1834 | goto destroy_ring; | |
2688fcb7 | 1835 | } |
0d160211 JF |
1836 | |
1837 | again: | |
1838 | err = xenbus_transaction_start(&xbt); | |
1839 | if (err) { | |
1840 | xenbus_dev_fatal(dev, err, "starting transaction"); | |
1841 | goto destroy_ring; | |
1842 | } | |
1843 | ||
812494d9 | 1844 | if (xenbus_exists(XBT_NIL, |
1845 | info->xbdev->otherend, "multi-queue-max-queues")) { | |
50ee6061 | 1846 | /* Write the number of queues */ |
812494d9 | 1847 | err = xenbus_printf(xbt, dev->nodename, |
1848 | "multi-queue-num-queues", "%u", num_queues); | |
d634bf2c | 1849 | if (err) { |
50ee6061 AB |
1850 | message = "writing multi-queue-num-queues"; |
1851 | goto abort_transaction_no_dev_fatal; | |
d634bf2c | 1852 | } |
812494d9 | 1853 | } |
50ee6061 | 1854 | |
812494d9 | 1855 | if (num_queues == 1) { |
1856 | err = write_queue_xenstore_keys(&info->queues[0], &xbt, 0); /* flat */ | |
1857 | if (err) | |
1858 | goto abort_transaction_no_dev_fatal; | |
1859 | } else { | |
50ee6061 AB |
1860 | /* Write the keys for each queue */ |
1861 | for (i = 0; i < num_queues; ++i) { | |
1862 | queue = &info->queues[i]; | |
1863 | err = write_queue_xenstore_keys(queue, &xbt, 1); /* hierarchical */ | |
1864 | if (err) | |
1865 | goto abort_transaction_no_dev_fatal; | |
d634bf2c | 1866 | } |
0d160211 JF |
1867 | } |
1868 | ||
50ee6061 | 1869 | /* The remaining keys are not queue-specific */ |
0d160211 JF |
1870 | err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u", |
1871 | 1); | |
1872 | if (err) { | |
1873 | message = "writing request-rx-copy"; | |
1874 | goto abort_transaction; | |
1875 | } | |
1876 | ||
1877 | err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1); | |
1878 | if (err) { | |
1879 | message = "writing feature-rx-notify"; | |
1880 | goto abort_transaction; | |
1881 | } | |
1882 | ||
1883 | err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1); | |
1884 | if (err) { | |
1885 | message = "writing feature-sg"; | |
1886 | goto abort_transaction; | |
1887 | } | |
1888 | ||
1889 | err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1); | |
1890 | if (err) { | |
1891 | message = "writing feature-gso-tcpv4"; | |
1892 | goto abort_transaction; | |
1893 | } | |
1894 | ||
2c0057de PD |
1895 | err = xenbus_write(xbt, dev->nodename, "feature-gso-tcpv6", "1"); |
1896 | if (err) { | |
1897 | message = "writing feature-gso-tcpv6"; | |
1898 | goto abort_transaction; | |
1899 | } | |
1900 | ||
1901 | err = xenbus_write(xbt, dev->nodename, "feature-ipv6-csum-offload", | |
1902 | "1"); | |
1903 | if (err) { | |
1904 | message = "writing feature-ipv6-csum-offload"; | |
1905 | goto abort_transaction; | |
1906 | } | |
1907 | ||
0d160211 JF |
1908 | err = xenbus_transaction_end(xbt, 0); |
1909 | if (err) { | |
1910 | if (err == -EAGAIN) | |
1911 | goto again; | |
1912 | xenbus_dev_fatal(dev, err, "completing transaction"); | |
1913 | goto destroy_ring; | |
1914 | } | |
1915 | ||
1916 | return 0; | |
1917 | ||
1918 | abort_transaction: | |
0d160211 | 1919 | xenbus_dev_fatal(dev, err, "%s", message); |
50ee6061 AB |
1920 | abort_transaction_no_dev_fatal: |
1921 | xenbus_transaction_end(xbt, 1); | |
0d160211 JF |
1922 | destroy_ring: |
1923 | xennet_disconnect_backend(info); | |
f599c64f | 1924 | rtnl_lock(); |
e2e004ac | 1925 | xennet_destroy_queues(info); |
0d160211 | 1926 | out: |
f599c64f | 1927 | rtnl_unlock(); |
d86b5672 | 1928 | device_unregister(&dev->dev); |
0d160211 JF |
1929 | return err; |
1930 | } | |
1931 | ||
0d160211 JF |
1932 | static int xennet_connect(struct net_device *dev) |
1933 | { | |
1934 | struct netfront_info *np = netdev_priv(dev); | |
2688fcb7 | 1935 | unsigned int num_queues = 0; |
a5b5dc3c | 1936 | int err; |
2688fcb7 AB |
1937 | unsigned int j = 0; |
1938 | struct netfront_queue *queue = NULL; | |
0d160211 | 1939 | |
2890ea5c | 1940 | if (!xenbus_read_unsigned(np->xbdev->otherend, "feature-rx-copy", 0)) { |
0d160211 | 1941 | dev_info(&dev->dev, |
898eb71c | 1942 | "backend does not support copying receive path\n"); |
0d160211 JF |
1943 | return -ENODEV; |
1944 | } | |
1945 | ||
f502bf2b | 1946 | err = talk_to_netback(np->xbdev, np); |
0d160211 JF |
1947 | if (err) |
1948 | return err; | |
1949 | ||
2688fcb7 AB |
1950 | /* talk_to_netback() sets the correct number of queues */ |
1951 | num_queues = dev->real_num_tx_queues; | |
1952 | ||
1ba37c51 | 1953 | rtnl_lock(); |
fb507934 | 1954 | netdev_update_features(dev); |
1ba37c51 | 1955 | rtnl_unlock(); |
0d160211 | 1956 | |
f599c64f RL |
1957 | if (dev->reg_state == NETREG_UNINITIALIZED) { |
1958 | err = register_netdev(dev); | |
1959 | if (err) { | |
1960 | pr_warn("%s: register_netdev err=%d\n", __func__, err); | |
1961 | device_unregister(&np->xbdev->dev); | |
1962 | return err; | |
1963 | } | |
1964 | } | |
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: | |
c2d2e673 JA |
2008 | break; |
2009 | ||
0d160211 | 2010 | case XenbusStateUnknown: |
c2d2e673 | 2011 | wake_up_all(&module_unload_q); |
0d160211 JF |
2012 | break; |
2013 | ||
2014 | case XenbusStateInitWait: | |
2015 | if (dev->state != XenbusStateInitialising) | |
2016 | break; | |
2017 | if (xennet_connect(netdev) != 0) | |
2018 | break; | |
2019 | xenbus_switch_state(dev, XenbusStateConnected); | |
08e34eb1 LE |
2020 | break; |
2021 | ||
2022 | case XenbusStateConnected: | |
ee89bab1 | 2023 | netdev_notify_peers(netdev); |
0d160211 JF |
2024 | break; |
2025 | ||
bce3ea81 | 2026 | case XenbusStateClosed: |
5b5971df | 2027 | wake_up_all(&module_unload_q); |
bce3ea81 DV |
2028 | if (dev->state == XenbusStateClosed) |
2029 | break; | |
2030 | /* Missed the backend's CLOSING state -- fallthrough */ | |
0d160211 | 2031 | case XenbusStateClosing: |
5b5971df | 2032 | wake_up_all(&module_unload_q); |
0d160211 JF |
2033 | xenbus_frontend_closed(dev); |
2034 | break; | |
2035 | } | |
2036 | } | |
2037 | ||
e0ce4af9 IC |
2038 | static const struct xennet_stat { |
2039 | char name[ETH_GSTRING_LEN]; | |
2040 | u16 offset; | |
2041 | } xennet_stats[] = { | |
2042 | { | |
2043 | "rx_gso_checksum_fixup", | |
2044 | offsetof(struct netfront_info, rx_gso_checksum_fixup) | |
2045 | }, | |
2046 | }; | |
2047 | ||
2048 | static int xennet_get_sset_count(struct net_device *dev, int string_set) | |
2049 | { | |
2050 | switch (string_set) { | |
2051 | case ETH_SS_STATS: | |
2052 | return ARRAY_SIZE(xennet_stats); | |
2053 | default: | |
2054 | return -EINVAL; | |
2055 | } | |
2056 | } | |
2057 | ||
2058 | static void xennet_get_ethtool_stats(struct net_device *dev, | |
2059 | struct ethtool_stats *stats, u64 * data) | |
2060 | { | |
2061 | void *np = netdev_priv(dev); | |
2062 | int i; | |
2063 | ||
2064 | for (i = 0; i < ARRAY_SIZE(xennet_stats); i++) | |
2688fcb7 | 2065 | data[i] = atomic_read((atomic_t *)(np + xennet_stats[i].offset)); |
e0ce4af9 IC |
2066 | } |
2067 | ||
2068 | static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data) | |
2069 | { | |
2070 | int i; | |
2071 | ||
2072 | switch (stringset) { | |
2073 | case ETH_SS_STATS: | |
2074 | for (i = 0; i < ARRAY_SIZE(xennet_stats); i++) | |
2075 | memcpy(data + i * ETH_GSTRING_LEN, | |
2076 | xennet_stats[i].name, ETH_GSTRING_LEN); | |
2077 | break; | |
2078 | } | |
2079 | } | |
2080 | ||
0fc0b732 | 2081 | static const struct ethtool_ops xennet_ethtool_ops = |
0d160211 | 2082 | { |
0d160211 | 2083 | .get_link = ethtool_op_get_link, |
e0ce4af9 IC |
2084 | |
2085 | .get_sset_count = xennet_get_sset_count, | |
2086 | .get_ethtool_stats = xennet_get_ethtool_stats, | |
2087 | .get_strings = xennet_get_strings, | |
0d160211 JF |
2088 | }; |
2089 | ||
2090 | #ifdef CONFIG_SYSFS | |
1f3c2eba DV |
2091 | static ssize_t show_rxbuf(struct device *dev, |
2092 | struct device_attribute *attr, char *buf) | |
0d160211 | 2093 | { |
1f3c2eba | 2094 | return sprintf(buf, "%lu\n", NET_RX_RING_SIZE); |
0d160211 JF |
2095 | } |
2096 | ||
1f3c2eba DV |
2097 | static ssize_t store_rxbuf(struct device *dev, |
2098 | struct device_attribute *attr, | |
2099 | const char *buf, size_t len) | |
0d160211 | 2100 | { |
0d160211 JF |
2101 | char *endp; |
2102 | unsigned long target; | |
2103 | ||
2104 | if (!capable(CAP_NET_ADMIN)) | |
2105 | return -EPERM; | |
2106 | ||
2107 | target = simple_strtoul(buf, &endp, 0); | |
2108 | if (endp == buf) | |
2109 | return -EBADMSG; | |
2110 | ||
1f3c2eba | 2111 | /* rxbuf_min and rxbuf_max are no longer configurable. */ |
0d160211 | 2112 | |
0d160211 JF |
2113 | return len; |
2114 | } | |
2115 | ||
27b917e5 TI |
2116 | static DEVICE_ATTR(rxbuf_min, S_IRUGO|S_IWUSR, show_rxbuf, store_rxbuf); |
2117 | static DEVICE_ATTR(rxbuf_max, S_IRUGO|S_IWUSR, show_rxbuf, store_rxbuf); | |
2118 | static DEVICE_ATTR(rxbuf_cur, S_IRUGO, show_rxbuf, NULL); | |
0d160211 | 2119 | |
27b917e5 TI |
2120 | static struct attribute *xennet_dev_attrs[] = { |
2121 | &dev_attr_rxbuf_min.attr, | |
2122 | &dev_attr_rxbuf_max.attr, | |
2123 | &dev_attr_rxbuf_cur.attr, | |
2124 | NULL | |
2125 | }; | |
0d160211 | 2126 | |
27b917e5 TI |
2127 | static const struct attribute_group xennet_dev_group = { |
2128 | .attrs = xennet_dev_attrs | |
2129 | }; | |
0d160211 JF |
2130 | #endif /* CONFIG_SYSFS */ |
2131 | ||
8e0e46bb | 2132 | static int xennet_remove(struct xenbus_device *dev) |
0d160211 | 2133 | { |
1b713e00 | 2134 | struct netfront_info *info = dev_get_drvdata(&dev->dev); |
0d160211 JF |
2135 | |
2136 | dev_dbg(&dev->dev, "%s\n", dev->nodename); | |
2137 | ||
5b5971df EO |
2138 | if (xenbus_read_driver_state(dev->otherend) != XenbusStateClosed) { |
2139 | xenbus_switch_state(dev, XenbusStateClosing); | |
2140 | wait_event(module_unload_q, | |
2141 | xenbus_read_driver_state(dev->otherend) == | |
c2d2e673 JA |
2142 | XenbusStateClosing || |
2143 | xenbus_read_driver_state(dev->otherend) == | |
2144 | XenbusStateUnknown); | |
5b5971df EO |
2145 | |
2146 | xenbus_switch_state(dev, XenbusStateClosed); | |
2147 | wait_event(module_unload_q, | |
2148 | xenbus_read_driver_state(dev->otherend) == | |
2149 | XenbusStateClosed || | |
2150 | xenbus_read_driver_state(dev->otherend) == | |
2151 | XenbusStateUnknown); | |
2152 | } | |
2153 | ||
0d160211 JF |
2154 | xennet_disconnect_backend(info); |
2155 | ||
f599c64f RL |
2156 | if (info->netdev->reg_state == NETREG_REGISTERED) |
2157 | unregister_netdev(info->netdev); | |
6bc96d04 | 2158 | |
f599c64f RL |
2159 | if (info->queues) { |
2160 | rtnl_lock(); | |
9a873c71 | 2161 | xennet_destroy_queues(info); |
f599c64f RL |
2162 | rtnl_unlock(); |
2163 | } | |
900e1833 | 2164 | xennet_free_netdev(info->netdev); |
0d160211 JF |
2165 | |
2166 | return 0; | |
2167 | } | |
2168 | ||
95afae48 DV |
2169 | static const struct xenbus_device_id netfront_ids[] = { |
2170 | { "vif" }, | |
2171 | { "" } | |
2172 | }; | |
2173 | ||
2174 | static struct xenbus_driver netfront_driver = { | |
2175 | .ids = netfront_ids, | |
0d160211 | 2176 | .probe = netfront_probe, |
8e0e46bb | 2177 | .remove = xennet_remove, |
0d160211 | 2178 | .resume = netfront_resume, |
f502bf2b | 2179 | .otherend_changed = netback_changed, |
95afae48 | 2180 | }; |
0d160211 JF |
2181 | |
2182 | static int __init netif_init(void) | |
2183 | { | |
6e833587 | 2184 | if (!xen_domain()) |
0d160211 JF |
2185 | return -ENODEV; |
2186 | ||
51c71a3b | 2187 | if (!xen_has_pv_nic_devices()) |
b9136d20 IM |
2188 | return -ENODEV; |
2189 | ||
383eda32 | 2190 | pr_info("Initialising Xen virtual ethernet driver\n"); |
0d160211 | 2191 | |
034702a6 | 2192 | /* Allow as many queues as there are CPUs inut max. 8 if user has not |
32a84405 WL |
2193 | * specified a value. |
2194 | */ | |
2195 | if (xennet_max_queues == 0) | |
034702a6 JG |
2196 | xennet_max_queues = min_t(unsigned int, MAX_QUEUES_DEFAULT, |
2197 | num_online_cpus()); | |
50ee6061 | 2198 | |
ffb78a26 | 2199 | return xenbus_register_frontend(&netfront_driver); |
0d160211 JF |
2200 | } |
2201 | module_init(netif_init); | |
2202 | ||
2203 | ||
2204 | static void __exit netif_exit(void) | |
2205 | { | |
ffb78a26 | 2206 | xenbus_unregister_driver(&netfront_driver); |
0d160211 JF |
2207 | } |
2208 | module_exit(netif_exit); | |
2209 | ||
2210 | MODULE_DESCRIPTION("Xen virtual network device frontend"); | |
2211 | MODULE_LICENSE("GPL"); | |
d2f0c52b | 2212 | MODULE_ALIAS("xen:vif"); |
4f93f09b | 2213 | MODULE_ALIAS("xennet"); |