]>
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 | ||
8edfe2e9 | 90 | static DECLARE_WAIT_QUEUE_HEAD(module_wq); |
5b5971df | 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) < |
57f230ab | 242 | (NET_TX_RING_SIZE - XEN_NETIF_NR_SLOTS_MIN - 1); |
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 | 547 | static u16 xennet_select_queue(struct net_device *dev, struct sk_buff *skb, |
4f49dec9 AD |
548 | struct net_device *sb_dev, |
549 | select_queue_fallback_t fallback) | |
2688fcb7 | 550 | { |
50ee6061 AB |
551 | unsigned int num_queues = dev->real_num_tx_queues; |
552 | u32 hash; | |
553 | u16 queue_idx; | |
554 | ||
555 | /* First, check if there is only one queue */ | |
556 | if (num_queues == 1) { | |
557 | queue_idx = 0; | |
558 | } else { | |
559 | hash = skb_get_hash(skb); | |
560 | queue_idx = hash % num_queues; | |
561 | } | |
562 | ||
563 | return queue_idx; | |
2688fcb7 AB |
564 | } |
565 | ||
30c5d7f0 JG |
566 | #define MAX_XEN_SKB_FRAGS (65536 / XEN_PAGE_SIZE + 1) |
567 | ||
24a94b3c | 568 | static netdev_tx_t xennet_start_xmit(struct sk_buff *skb, struct net_device *dev) |
0d160211 | 569 | { |
0d160211 | 570 | struct netfront_info *np = netdev_priv(dev); |
900e1833 | 571 | struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats); |
a55e8bb8 DV |
572 | struct xen_netif_tx_request *tx, *first_tx; |
573 | unsigned int i; | |
0d160211 | 574 | int notify; |
f36c3747 | 575 | int slots; |
a55e8bb8 DV |
576 | struct page *page; |
577 | unsigned int offset; | |
578 | unsigned int len; | |
cf66f9d4 | 579 | unsigned long flags; |
2688fcb7 AB |
580 | struct netfront_queue *queue = NULL; |
581 | unsigned int num_queues = dev->real_num_tx_queues; | |
582 | u16 queue_index; | |
fd07160b | 583 | struct sk_buff *nskb; |
2688fcb7 AB |
584 | |
585 | /* Drop the packet if no queues are set up */ | |
586 | if (num_queues < 1) | |
587 | goto drop; | |
588 | /* Determine which queue to transmit this SKB on */ | |
589 | queue_index = skb_get_queue_mapping(skb); | |
590 | queue = &np->queues[queue_index]; | |
0d160211 | 591 | |
9ecd1a75 WL |
592 | /* If skb->len is too big for wire format, drop skb and alert |
593 | * user about misconfiguration. | |
594 | */ | |
595 | if (unlikely(skb->len > XEN_NETIF_MAX_TX_SIZE)) { | |
596 | net_alert_ratelimited( | |
597 | "xennet: skb->len = %u, too big for wire format\n", | |
598 | skb->len); | |
599 | goto drop; | |
600 | } | |
601 | ||
e84448d5 | 602 | slots = xennet_count_skb_slots(skb); |
30c5d7f0 | 603 | if (unlikely(slots > MAX_XEN_SKB_FRAGS + 1)) { |
97a6d1bb ZK |
604 | net_dbg_ratelimited("xennet: skb rides the rocket: %d slots, %d bytes\n", |
605 | slots, skb->len); | |
606 | if (skb_linearize(skb)) | |
607 | goto drop; | |
0d160211 JF |
608 | } |
609 | ||
a55e8bb8 DV |
610 | page = virt_to_page(skb->data); |
611 | offset = offset_in_page(skb->data); | |
fd07160b VK |
612 | |
613 | /* The first req should be at least ETH_HLEN size or the packet will be | |
614 | * dropped by netback. | |
615 | */ | |
616 | if (unlikely(PAGE_SIZE - offset < ETH_HLEN)) { | |
617 | nskb = skb_copy(skb, GFP_ATOMIC); | |
618 | if (!nskb) | |
619 | goto drop; | |
62f3250f | 620 | dev_consume_skb_any(skb); |
fd07160b VK |
621 | skb = nskb; |
622 | page = virt_to_page(skb->data); | |
623 | offset = offset_in_page(skb->data); | |
624 | } | |
625 | ||
a55e8bb8 DV |
626 | len = skb_headlen(skb); |
627 | ||
2688fcb7 | 628 | spin_lock_irqsave(&queue->tx_lock, flags); |
0d160211 JF |
629 | |
630 | if (unlikely(!netif_carrier_ok(dev) || | |
f36c3747 | 631 | (slots > 1 && !xennet_can_sg(dev)) || |
8b86a61d | 632 | netif_needs_gso(skb, netif_skb_features(skb)))) { |
2688fcb7 | 633 | spin_unlock_irqrestore(&queue->tx_lock, flags); |
0d160211 JF |
634 | goto drop; |
635 | } | |
636 | ||
a55e8bb8 | 637 | /* First request for the linear area. */ |
30c5d7f0 JG |
638 | first_tx = tx = xennet_make_first_txreq(queue, skb, |
639 | page, offset, len); | |
640 | offset += tx->size; | |
641 | if (offset == PAGE_SIZE) { | |
642 | page++; | |
643 | offset = 0; | |
644 | } | |
a55e8bb8 | 645 | len -= tx->size; |
0d160211 | 646 | |
0d160211 JF |
647 | if (skb->ip_summed == CHECKSUM_PARTIAL) |
648 | /* local packet? */ | |
f942dc25 | 649 | tx->flags |= XEN_NETTXF_csum_blank | XEN_NETTXF_data_validated; |
0d160211 JF |
650 | else if (skb->ip_summed == CHECKSUM_UNNECESSARY) |
651 | /* remote but checksummed. */ | |
f942dc25 | 652 | tx->flags |= XEN_NETTXF_data_validated; |
0d160211 | 653 | |
a55e8bb8 | 654 | /* Optional extra info after the first request. */ |
0d160211 JF |
655 | if (skb_shinfo(skb)->gso_size) { |
656 | struct xen_netif_extra_info *gso; | |
657 | ||
658 | gso = (struct xen_netif_extra_info *) | |
a55e8bb8 | 659 | RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++); |
0d160211 | 660 | |
e2d617c0 | 661 | tx->flags |= XEN_NETTXF_extra_info; |
0d160211 JF |
662 | |
663 | gso->u.gso.size = skb_shinfo(skb)->gso_size; | |
2c0057de PD |
664 | gso->u.gso.type = (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) ? |
665 | XEN_NETIF_GSO_TYPE_TCPV6 : | |
666 | XEN_NETIF_GSO_TYPE_TCPV4; | |
0d160211 JF |
667 | gso->u.gso.pad = 0; |
668 | gso->u.gso.features = 0; | |
669 | ||
670 | gso->type = XEN_NETIF_EXTRA_TYPE_GSO; | |
671 | gso->flags = 0; | |
0d160211 JF |
672 | } |
673 | ||
a55e8bb8 DV |
674 | /* Requests for the rest of the linear area. */ |
675 | tx = xennet_make_txreqs(queue, tx, skb, page, offset, len); | |
676 | ||
677 | /* Requests for all the frags. */ | |
678 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
679 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
680 | tx = xennet_make_txreqs(queue, tx, skb, | |
681 | skb_frag_page(frag), frag->page_offset, | |
682 | skb_frag_size(frag)); | |
683 | } | |
0d160211 | 684 | |
a55e8bb8 DV |
685 | /* First request has the packet length. */ |
686 | first_tx->size = skb->len; | |
0d160211 | 687 | |
2688fcb7 | 688 | RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify); |
0d160211 | 689 | if (notify) |
2688fcb7 | 690 | notify_remote_via_irq(queue->tx_irq); |
0d160211 | 691 | |
900e1833 DV |
692 | u64_stats_update_begin(&tx_stats->syncp); |
693 | tx_stats->bytes += skb->len; | |
694 | tx_stats->packets++; | |
695 | u64_stats_update_end(&tx_stats->syncp); | |
10a273a6 JF |
696 | |
697 | /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */ | |
2688fcb7 | 698 | xennet_tx_buf_gc(queue); |
0d160211 | 699 | |
2688fcb7 AB |
700 | if (!netfront_tx_slot_available(queue)) |
701 | netif_tx_stop_queue(netdev_get_tx_queue(dev, queue->id)); | |
0d160211 | 702 | |
2688fcb7 | 703 | spin_unlock_irqrestore(&queue->tx_lock, flags); |
0d160211 | 704 | |
6ed10654 | 705 | return NETDEV_TX_OK; |
0d160211 JF |
706 | |
707 | drop: | |
09f75cd7 | 708 | dev->stats.tx_dropped++; |
979de8a0 | 709 | dev_kfree_skb_any(skb); |
6ed10654 | 710 | return NETDEV_TX_OK; |
0d160211 JF |
711 | } |
712 | ||
713 | static int xennet_close(struct net_device *dev) | |
714 | { | |
715 | struct netfront_info *np = netdev_priv(dev); | |
2688fcb7 AB |
716 | unsigned int num_queues = dev->real_num_tx_queues; |
717 | unsigned int i; | |
718 | struct netfront_queue *queue; | |
719 | netif_tx_stop_all_queues(np->netdev); | |
720 | for (i = 0; i < num_queues; ++i) { | |
721 | queue = &np->queues[i]; | |
722 | napi_disable(&queue->napi); | |
723 | } | |
0d160211 JF |
724 | return 0; |
725 | } | |
726 | ||
2688fcb7 | 727 | static void xennet_move_rx_slot(struct netfront_queue *queue, struct sk_buff *skb, |
0d160211 JF |
728 | grant_ref_t ref) |
729 | { | |
2688fcb7 AB |
730 | int new = xennet_rxidx(queue->rx.req_prod_pvt); |
731 | ||
732 | BUG_ON(queue->rx_skbs[new]); | |
733 | queue->rx_skbs[new] = skb; | |
734 | queue->grant_rx_ref[new] = ref; | |
735 | RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->id = new; | |
736 | RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->gref = ref; | |
737 | queue->rx.req_prod_pvt++; | |
0d160211 JF |
738 | } |
739 | ||
2688fcb7 | 740 | static int xennet_get_extras(struct netfront_queue *queue, |
0d160211 JF |
741 | struct xen_netif_extra_info *extras, |
742 | RING_IDX rp) | |
743 | ||
744 | { | |
745 | struct xen_netif_extra_info *extra; | |
2688fcb7 AB |
746 | struct device *dev = &queue->info->netdev->dev; |
747 | RING_IDX cons = queue->rx.rsp_cons; | |
0d160211 JF |
748 | int err = 0; |
749 | ||
750 | do { | |
751 | struct sk_buff *skb; | |
752 | grant_ref_t ref; | |
753 | ||
754 | if (unlikely(cons + 1 == rp)) { | |
755 | if (net_ratelimit()) | |
756 | dev_warn(dev, "Missing extra info\n"); | |
757 | err = -EBADR; | |
758 | break; | |
759 | } | |
760 | ||
761 | extra = (struct xen_netif_extra_info *) | |
2688fcb7 | 762 | RING_GET_RESPONSE(&queue->rx, ++cons); |
0d160211 JF |
763 | |
764 | if (unlikely(!extra->type || | |
765 | extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) { | |
766 | if (net_ratelimit()) | |
767 | dev_warn(dev, "Invalid extra type: %d\n", | |
768 | extra->type); | |
769 | err = -EINVAL; | |
770 | } else { | |
771 | memcpy(&extras[extra->type - 1], extra, | |
772 | sizeof(*extra)); | |
773 | } | |
774 | ||
2688fcb7 AB |
775 | skb = xennet_get_rx_skb(queue, cons); |
776 | ref = xennet_get_rx_ref(queue, cons); | |
777 | xennet_move_rx_slot(queue, skb, ref); | |
0d160211 JF |
778 | } while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE); |
779 | ||
2688fcb7 | 780 | queue->rx.rsp_cons = cons; |
0d160211 JF |
781 | return err; |
782 | } | |
783 | ||
2688fcb7 | 784 | static int xennet_get_responses(struct netfront_queue *queue, |
0d160211 JF |
785 | struct netfront_rx_info *rinfo, RING_IDX rp, |
786 | struct sk_buff_head *list) | |
787 | { | |
788 | struct xen_netif_rx_response *rx = &rinfo->rx; | |
789 | struct xen_netif_extra_info *extras = rinfo->extras; | |
2688fcb7 AB |
790 | struct device *dev = &queue->info->netdev->dev; |
791 | RING_IDX cons = queue->rx.rsp_cons; | |
792 | struct sk_buff *skb = xennet_get_rx_skb(queue, cons); | |
793 | grant_ref_t ref = xennet_get_rx_ref(queue, cons); | |
57f230ab | 794 | int max = XEN_NETIF_NR_SLOTS_MIN + (rx->status <= RX_COPY_THRESHOLD); |
7158ff6d | 795 | int slots = 1; |
0d160211 JF |
796 | int err = 0; |
797 | unsigned long ret; | |
798 | ||
f942dc25 | 799 | if (rx->flags & XEN_NETRXF_extra_info) { |
2688fcb7 AB |
800 | err = xennet_get_extras(queue, extras, rp); |
801 | cons = queue->rx.rsp_cons; | |
0d160211 JF |
802 | } |
803 | ||
804 | for (;;) { | |
805 | if (unlikely(rx->status < 0 || | |
30c5d7f0 | 806 | rx->offset + rx->status > XEN_PAGE_SIZE)) { |
0d160211 | 807 | if (net_ratelimit()) |
6c10127d | 808 | dev_warn(dev, "rx->offset: %u, size: %d\n", |
0d160211 | 809 | rx->offset, rx->status); |
2688fcb7 | 810 | xennet_move_rx_slot(queue, skb, ref); |
0d160211 JF |
811 | err = -EINVAL; |
812 | goto next; | |
813 | } | |
814 | ||
815 | /* | |
816 | * This definitely indicates a bug, either in this driver or in | |
817 | * the backend driver. In future this should flag the bad | |
697089dc | 818 | * situation to the system controller to reboot the backend. |
0d160211 JF |
819 | */ |
820 | if (ref == GRANT_INVALID_REF) { | |
821 | if (net_ratelimit()) | |
822 | dev_warn(dev, "Bad rx response id %d.\n", | |
823 | rx->id); | |
824 | err = -EINVAL; | |
825 | goto next; | |
826 | } | |
827 | ||
828 | ret = gnttab_end_foreign_access_ref(ref, 0); | |
829 | BUG_ON(!ret); | |
830 | ||
2688fcb7 | 831 | gnttab_release_grant_reference(&queue->gref_rx_head, ref); |
0d160211 JF |
832 | |
833 | __skb_queue_tail(list, skb); | |
834 | ||
835 | next: | |
f942dc25 | 836 | if (!(rx->flags & XEN_NETRXF_more_data)) |
0d160211 JF |
837 | break; |
838 | ||
7158ff6d | 839 | if (cons + slots == rp) { |
0d160211 | 840 | if (net_ratelimit()) |
7158ff6d | 841 | dev_warn(dev, "Need more slots\n"); |
0d160211 JF |
842 | err = -ENOENT; |
843 | break; | |
844 | } | |
845 | ||
2688fcb7 AB |
846 | rx = RING_GET_RESPONSE(&queue->rx, cons + slots); |
847 | skb = xennet_get_rx_skb(queue, cons + slots); | |
848 | ref = xennet_get_rx_ref(queue, cons + slots); | |
7158ff6d | 849 | slots++; |
0d160211 JF |
850 | } |
851 | ||
7158ff6d | 852 | if (unlikely(slots > max)) { |
0d160211 | 853 | if (net_ratelimit()) |
697089dc | 854 | dev_warn(dev, "Too many slots\n"); |
0d160211 JF |
855 | err = -E2BIG; |
856 | } | |
857 | ||
858 | if (unlikely(err)) | |
2688fcb7 | 859 | queue->rx.rsp_cons = cons + slots; |
0d160211 JF |
860 | |
861 | return err; | |
862 | } | |
863 | ||
864 | static int xennet_set_skb_gso(struct sk_buff *skb, | |
865 | struct xen_netif_extra_info *gso) | |
866 | { | |
867 | if (!gso->u.gso.size) { | |
868 | if (net_ratelimit()) | |
383eda32 | 869 | pr_warn("GSO size must not be zero\n"); |
0d160211 JF |
870 | return -EINVAL; |
871 | } | |
872 | ||
2c0057de PD |
873 | if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4 && |
874 | gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV6) { | |
0d160211 | 875 | if (net_ratelimit()) |
383eda32 | 876 | pr_warn("Bad GSO type %d\n", gso->u.gso.type); |
0d160211 JF |
877 | return -EINVAL; |
878 | } | |
879 | ||
880 | skb_shinfo(skb)->gso_size = gso->u.gso.size; | |
2c0057de PD |
881 | skb_shinfo(skb)->gso_type = |
882 | (gso->u.gso.type == XEN_NETIF_GSO_TYPE_TCPV4) ? | |
883 | SKB_GSO_TCPV4 : | |
884 | SKB_GSO_TCPV6; | |
0d160211 JF |
885 | |
886 | /* Header must be checked, and gso_segs computed. */ | |
887 | skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; | |
888 | skb_shinfo(skb)->gso_segs = 0; | |
889 | ||
890 | return 0; | |
891 | } | |
892 | ||
2688fcb7 | 893 | static RING_IDX xennet_fill_frags(struct netfront_queue *queue, |
0d160211 JF |
894 | struct sk_buff *skb, |
895 | struct sk_buff_head *list) | |
896 | { | |
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 | |
d472b3a6 | 905 | if (skb_shinfo(skb)->nr_frags == MAX_SKB_FRAGS) { |
093b9c71 | 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 | } | |
d472b3a6 | 911 | BUG_ON(skb_shinfo(skb)->nr_frags >= MAX_SKB_FRAGS); |
093b9c71 | 912 | |
d472b3a6 JG |
913 | skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, |
914 | skb_frag_page(nfrag), | |
093b9c71 | 915 | rx->offset, rx->status, PAGE_SIZE); |
0d160211 JF |
916 | |
917 | skb_shinfo(nskb)->nr_frags = 0; | |
918 | kfree_skb(nskb); | |
0d160211 JF |
919 | } |
920 | ||
0d160211 JF |
921 | return cons; |
922 | } | |
923 | ||
e0ce4af9 | 924 | static int checksum_setup(struct net_device *dev, struct sk_buff *skb) |
0d160211 | 925 | { |
b5cf66cd | 926 | bool recalculate_partial_csum = false; |
e0ce4af9 IC |
927 | |
928 | /* | |
929 | * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy | |
930 | * peers can fail to set NETRXF_csum_blank when sending a GSO | |
931 | * frame. In this case force the SKB to CHECKSUM_PARTIAL and | |
932 | * recalculate the partial checksum. | |
933 | */ | |
934 | if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) { | |
935 | struct netfront_info *np = netdev_priv(dev); | |
2688fcb7 | 936 | atomic_inc(&np->rx_gso_checksum_fixup); |
e0ce4af9 | 937 | skb->ip_summed = CHECKSUM_PARTIAL; |
b5cf66cd | 938 | recalculate_partial_csum = true; |
e0ce4af9 IC |
939 | } |
940 | ||
941 | /* A non-CHECKSUM_PARTIAL SKB does not require setup. */ | |
942 | if (skb->ip_summed != CHECKSUM_PARTIAL) | |
943 | return 0; | |
0d160211 | 944 | |
b5cf66cd | 945 | return skb_checksum_setup(skb, recalculate_partial_csum); |
0d160211 JF |
946 | } |
947 | ||
2688fcb7 | 948 | static int handle_incoming_queue(struct netfront_queue *queue, |
09f75cd7 | 949 | struct sk_buff_head *rxq) |
0d160211 | 950 | { |
900e1833 | 951 | struct netfront_stats *rx_stats = this_cpu_ptr(queue->info->rx_stats); |
0d160211 JF |
952 | int packets_dropped = 0; |
953 | struct sk_buff *skb; | |
954 | ||
955 | while ((skb = __skb_dequeue(rxq)) != NULL) { | |
3683243b | 956 | int pull_to = NETFRONT_SKB_CB(skb)->pull_to; |
0d160211 | 957 | |
093b9c71 JB |
958 | if (pull_to > skb_headlen(skb)) |
959 | __pskb_pull_tail(skb, pull_to - skb_headlen(skb)); | |
0d160211 JF |
960 | |
961 | /* Ethernet work: Delayed to here as it peeks the header. */ | |
2688fcb7 | 962 | skb->protocol = eth_type_trans(skb, queue->info->netdev); |
d554f73d | 963 | skb_reset_network_header(skb); |
0d160211 | 964 | |
2688fcb7 | 965 | if (checksum_setup(queue->info->netdev, skb)) { |
e0ce4af9 IC |
966 | kfree_skb(skb); |
967 | packets_dropped++; | |
2688fcb7 | 968 | queue->info->netdev->stats.rx_errors++; |
e0ce4af9 | 969 | continue; |
0d160211 JF |
970 | } |
971 | ||
900e1833 DV |
972 | u64_stats_update_begin(&rx_stats->syncp); |
973 | rx_stats->packets++; | |
974 | rx_stats->bytes += skb->len; | |
975 | u64_stats_update_end(&rx_stats->syncp); | |
0d160211 JF |
976 | |
977 | /* Pass it up. */ | |
2688fcb7 | 978 | napi_gro_receive(&queue->napi, skb); |
0d160211 JF |
979 | } |
980 | ||
981 | return packets_dropped; | |
982 | } | |
983 | ||
bea3348e | 984 | static int xennet_poll(struct napi_struct *napi, int budget) |
0d160211 | 985 | { |
2688fcb7 AB |
986 | struct netfront_queue *queue = container_of(napi, struct netfront_queue, napi); |
987 | struct net_device *dev = queue->info->netdev; | |
0d160211 JF |
988 | struct sk_buff *skb; |
989 | struct netfront_rx_info rinfo; | |
990 | struct xen_netif_rx_response *rx = &rinfo.rx; | |
991 | struct xen_netif_extra_info *extras = rinfo.extras; | |
992 | RING_IDX i, rp; | |
bea3348e | 993 | int work_done; |
0d160211 JF |
994 | struct sk_buff_head rxq; |
995 | struct sk_buff_head errq; | |
996 | struct sk_buff_head tmpq; | |
0d160211 JF |
997 | int err; |
998 | ||
2688fcb7 | 999 | spin_lock(&queue->rx_lock); |
0d160211 | 1000 | |
0d160211 JF |
1001 | skb_queue_head_init(&rxq); |
1002 | skb_queue_head_init(&errq); | |
1003 | skb_queue_head_init(&tmpq); | |
1004 | ||
2688fcb7 | 1005 | rp = queue->rx.sring->rsp_prod; |
0d160211 JF |
1006 | rmb(); /* Ensure we see queued responses up to 'rp'. */ |
1007 | ||
2688fcb7 | 1008 | i = queue->rx.rsp_cons; |
0d160211 JF |
1009 | work_done = 0; |
1010 | while ((i != rp) && (work_done < budget)) { | |
2688fcb7 | 1011 | memcpy(rx, RING_GET_RESPONSE(&queue->rx, i), sizeof(*rx)); |
0d160211 JF |
1012 | memset(extras, 0, sizeof(rinfo.extras)); |
1013 | ||
2688fcb7 | 1014 | err = xennet_get_responses(queue, &rinfo, rp, &tmpq); |
0d160211 JF |
1015 | |
1016 | if (unlikely(err)) { | |
1017 | err: | |
1018 | while ((skb = __skb_dequeue(&tmpq))) | |
1019 | __skb_queue_tail(&errq, skb); | |
09f75cd7 | 1020 | dev->stats.rx_errors++; |
2688fcb7 | 1021 | i = queue->rx.rsp_cons; |
0d160211 JF |
1022 | continue; |
1023 | } | |
1024 | ||
1025 | skb = __skb_dequeue(&tmpq); | |
1026 | ||
1027 | if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) { | |
1028 | struct xen_netif_extra_info *gso; | |
1029 | gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1]; | |
1030 | ||
1031 | if (unlikely(xennet_set_skb_gso(skb, gso))) { | |
1032 | __skb_queue_head(&tmpq, skb); | |
2688fcb7 | 1033 | queue->rx.rsp_cons += skb_queue_len(&tmpq); |
0d160211 JF |
1034 | goto err; |
1035 | } | |
1036 | } | |
1037 | ||
3683243b IC |
1038 | NETFRONT_SKB_CB(skb)->pull_to = rx->status; |
1039 | if (NETFRONT_SKB_CB(skb)->pull_to > RX_COPY_THRESHOLD) | |
1040 | NETFRONT_SKB_CB(skb)->pull_to = RX_COPY_THRESHOLD; | |
0d160211 | 1041 | |
3683243b IC |
1042 | skb_shinfo(skb)->frags[0].page_offset = rx->offset; |
1043 | skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status); | |
1044 | skb->data_len = rx->status; | |
093b9c71 | 1045 | skb->len += rx->status; |
0d160211 | 1046 | |
2688fcb7 | 1047 | i = xennet_fill_frags(queue, skb, &tmpq); |
0d160211 | 1048 | |
f942dc25 | 1049 | if (rx->flags & XEN_NETRXF_csum_blank) |
0d160211 | 1050 | skb->ip_summed = CHECKSUM_PARTIAL; |
f942dc25 | 1051 | else if (rx->flags & XEN_NETRXF_data_validated) |
0d160211 JF |
1052 | skb->ip_summed = CHECKSUM_UNNECESSARY; |
1053 | ||
1054 | __skb_queue_tail(&rxq, skb); | |
1055 | ||
2688fcb7 | 1056 | queue->rx.rsp_cons = ++i; |
0d160211 JF |
1057 | work_done++; |
1058 | } | |
1059 | ||
56cfe5d0 | 1060 | __skb_queue_purge(&errq); |
0d160211 | 1061 | |
2688fcb7 | 1062 | work_done -= handle_incoming_queue(queue, &rxq); |
0d160211 | 1063 | |
2688fcb7 | 1064 | xennet_alloc_rx_buffers(queue); |
0d160211 | 1065 | |
0d160211 | 1066 | if (work_done < budget) { |
bea3348e SH |
1067 | int more_to_do = 0; |
1068 | ||
6ad20165 | 1069 | napi_complete_done(napi, work_done); |
0d160211 | 1070 | |
2688fcb7 | 1071 | RING_FINAL_CHECK_FOR_RESPONSES(&queue->rx, more_to_do); |
6a6dc08f DV |
1072 | if (more_to_do) |
1073 | napi_schedule(napi); | |
0d160211 JF |
1074 | } |
1075 | ||
2688fcb7 | 1076 | spin_unlock(&queue->rx_lock); |
0d160211 | 1077 | |
bea3348e | 1078 | return work_done; |
0d160211 JF |
1079 | } |
1080 | ||
1081 | static int xennet_change_mtu(struct net_device *dev, int mtu) | |
1082 | { | |
0c36820e | 1083 | int max = xennet_can_sg(dev) ? XEN_NETIF_MAX_TX_SIZE : ETH_DATA_LEN; |
0d160211 JF |
1084 | |
1085 | if (mtu > max) | |
1086 | return -EINVAL; | |
1087 | dev->mtu = mtu; | |
1088 | return 0; | |
1089 | } | |
1090 | ||
bc1f4470 | 1091 | static void xennet_get_stats64(struct net_device *dev, |
1092 | struct rtnl_link_stats64 *tot) | |
e00f85be | 1093 | { |
1094 | struct netfront_info *np = netdev_priv(dev); | |
1095 | int cpu; | |
1096 | ||
1097 | for_each_possible_cpu(cpu) { | |
900e1833 DV |
1098 | struct netfront_stats *rx_stats = per_cpu_ptr(np->rx_stats, cpu); |
1099 | struct netfront_stats *tx_stats = per_cpu_ptr(np->tx_stats, cpu); | |
e00f85be | 1100 | u64 rx_packets, rx_bytes, tx_packets, tx_bytes; |
1101 | unsigned int start; | |
1102 | ||
1103 | do { | |
900e1833 DV |
1104 | start = u64_stats_fetch_begin_irq(&tx_stats->syncp); |
1105 | tx_packets = tx_stats->packets; | |
1106 | tx_bytes = tx_stats->bytes; | |
1107 | } while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start)); | |
e00f85be | 1108 | |
900e1833 DV |
1109 | do { |
1110 | start = u64_stats_fetch_begin_irq(&rx_stats->syncp); | |
1111 | rx_packets = rx_stats->packets; | |
1112 | rx_bytes = rx_stats->bytes; | |
1113 | } while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start)); | |
e00f85be | 1114 | |
1115 | tot->rx_packets += rx_packets; | |
1116 | tot->tx_packets += tx_packets; | |
1117 | tot->rx_bytes += rx_bytes; | |
1118 | tot->tx_bytes += tx_bytes; | |
1119 | } | |
1120 | ||
1121 | tot->rx_errors = dev->stats.rx_errors; | |
1122 | tot->tx_dropped = dev->stats.tx_dropped; | |
e00f85be | 1123 | } |
1124 | ||
2688fcb7 | 1125 | static void xennet_release_tx_bufs(struct netfront_queue *queue) |
0d160211 JF |
1126 | { |
1127 | struct sk_buff *skb; | |
1128 | int i; | |
1129 | ||
1130 | for (i = 0; i < NET_TX_RING_SIZE; i++) { | |
1131 | /* Skip over entries which are actually freelist references */ | |
2688fcb7 | 1132 | if (skb_entry_is_link(&queue->tx_skbs[i])) |
0d160211 JF |
1133 | continue; |
1134 | ||
2688fcb7 AB |
1135 | skb = queue->tx_skbs[i].skb; |
1136 | get_page(queue->grant_tx_page[i]); | |
1137 | gnttab_end_foreign_access(queue->grant_tx_ref[i], | |
cefe0078 | 1138 | GNTMAP_readonly, |
2688fcb7 AB |
1139 | (unsigned long)page_address(queue->grant_tx_page[i])); |
1140 | queue->grant_tx_page[i] = NULL; | |
1141 | queue->grant_tx_ref[i] = GRANT_INVALID_REF; | |
1142 | add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, i); | |
0d160211 JF |
1143 | dev_kfree_skb_irq(skb); |
1144 | } | |
1145 | } | |
1146 | ||
2688fcb7 | 1147 | static void xennet_release_rx_bufs(struct netfront_queue *queue) |
0d160211 | 1148 | { |
0d160211 JF |
1149 | int id, ref; |
1150 | ||
2688fcb7 | 1151 | spin_lock_bh(&queue->rx_lock); |
0d160211 JF |
1152 | |
1153 | for (id = 0; id < NET_RX_RING_SIZE; id++) { | |
cefe0078 AL |
1154 | struct sk_buff *skb; |
1155 | struct page *page; | |
0d160211 | 1156 | |
2688fcb7 | 1157 | skb = queue->rx_skbs[id]; |
cefe0078 | 1158 | if (!skb) |
0d160211 | 1159 | continue; |
0d160211 | 1160 | |
2688fcb7 | 1161 | ref = queue->grant_rx_ref[id]; |
cefe0078 AL |
1162 | if (ref == GRANT_INVALID_REF) |
1163 | continue; | |
0d160211 | 1164 | |
cefe0078 | 1165 | page = skb_frag_page(&skb_shinfo(skb)->frags[0]); |
0d160211 | 1166 | |
cefe0078 AL |
1167 | /* gnttab_end_foreign_access() needs a page ref until |
1168 | * foreign access is ended (which may be deferred). | |
1169 | */ | |
1170 | get_page(page); | |
1171 | gnttab_end_foreign_access(ref, 0, | |
1172 | (unsigned long)page_address(page)); | |
2688fcb7 | 1173 | queue->grant_rx_ref[id] = GRANT_INVALID_REF; |
0d160211 | 1174 | |
cefe0078 | 1175 | kfree_skb(skb); |
0d160211 JF |
1176 | } |
1177 | ||
2688fcb7 | 1178 | spin_unlock_bh(&queue->rx_lock); |
0d160211 JF |
1179 | } |
1180 | ||
c8f44aff MM |
1181 | static netdev_features_t xennet_fix_features(struct net_device *dev, |
1182 | netdev_features_t features) | |
8f7b01a1 ED |
1183 | { |
1184 | struct netfront_info *np = netdev_priv(dev); | |
8f7b01a1 | 1185 | |
2890ea5c JG |
1186 | if (features & NETIF_F_SG && |
1187 | !xenbus_read_unsigned(np->xbdev->otherend, "feature-sg", 0)) | |
1188 | features &= ~NETIF_F_SG; | |
8f7b01a1 | 1189 | |
2890ea5c JG |
1190 | if (features & NETIF_F_IPV6_CSUM && |
1191 | !xenbus_read_unsigned(np->xbdev->otherend, | |
1192 | "feature-ipv6-csum-offload", 0)) | |
1193 | features &= ~NETIF_F_IPV6_CSUM; | |
8f7b01a1 | 1194 | |
2890ea5c JG |
1195 | if (features & NETIF_F_TSO && |
1196 | !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv4", 0)) | |
1197 | features &= ~NETIF_F_TSO; | |
8f7b01a1 | 1198 | |
2890ea5c JG |
1199 | if (features & NETIF_F_TSO6 && |
1200 | !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv6", 0)) | |
1201 | features &= ~NETIF_F_TSO6; | |
2c0057de | 1202 | |
8f7b01a1 ED |
1203 | return features; |
1204 | } | |
1205 | ||
c8f44aff MM |
1206 | static int xennet_set_features(struct net_device *dev, |
1207 | netdev_features_t features) | |
8f7b01a1 ED |
1208 | { |
1209 | if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) { | |
1210 | netdev_info(dev, "Reducing MTU because no SG offload"); | |
1211 | dev->mtu = ETH_DATA_LEN; | |
1212 | } | |
1213 | ||
1214 | return 0; | |
1215 | } | |
1216 | ||
d634bf2c | 1217 | static irqreturn_t xennet_tx_interrupt(int irq, void *dev_id) |
cf66f9d4 | 1218 | { |
2688fcb7 | 1219 | struct netfront_queue *queue = dev_id; |
cf66f9d4 KRW |
1220 | unsigned long flags; |
1221 | ||
2688fcb7 AB |
1222 | spin_lock_irqsave(&queue->tx_lock, flags); |
1223 | xennet_tx_buf_gc(queue); | |
1224 | spin_unlock_irqrestore(&queue->tx_lock, flags); | |
cf66f9d4 | 1225 | |
d634bf2c WL |
1226 | return IRQ_HANDLED; |
1227 | } | |
1228 | ||
1229 | static irqreturn_t xennet_rx_interrupt(int irq, void *dev_id) | |
1230 | { | |
2688fcb7 AB |
1231 | struct netfront_queue *queue = dev_id; |
1232 | struct net_device *dev = queue->info->netdev; | |
d634bf2c WL |
1233 | |
1234 | if (likely(netif_carrier_ok(dev) && | |
2688fcb7 | 1235 | RING_HAS_UNCONSUMED_RESPONSES(&queue->rx))) |
76541869 | 1236 | napi_schedule(&queue->napi); |
cf66f9d4 | 1237 | |
d634bf2c WL |
1238 | return IRQ_HANDLED; |
1239 | } | |
cf66f9d4 | 1240 | |
d634bf2c WL |
1241 | static irqreturn_t xennet_interrupt(int irq, void *dev_id) |
1242 | { | |
1243 | xennet_tx_interrupt(irq, dev_id); | |
1244 | xennet_rx_interrupt(irq, dev_id); | |
cf66f9d4 KRW |
1245 | return IRQ_HANDLED; |
1246 | } | |
1247 | ||
1248 | #ifdef CONFIG_NET_POLL_CONTROLLER | |
1249 | static void xennet_poll_controller(struct net_device *dev) | |
1250 | { | |
2688fcb7 AB |
1251 | /* Poll each queue */ |
1252 | struct netfront_info *info = netdev_priv(dev); | |
1253 | unsigned int num_queues = dev->real_num_tx_queues; | |
1254 | unsigned int i; | |
1255 | for (i = 0; i < num_queues; ++i) | |
1256 | xennet_interrupt(0, &info->queues[i]); | |
cf66f9d4 KRW |
1257 | } |
1258 | #endif | |
1259 | ||
0a0b9d2e SH |
1260 | static const struct net_device_ops xennet_netdev_ops = { |
1261 | .ndo_open = xennet_open, | |
0a0b9d2e SH |
1262 | .ndo_stop = xennet_close, |
1263 | .ndo_start_xmit = xennet_start_xmit, | |
1264 | .ndo_change_mtu = xennet_change_mtu, | |
e00f85be | 1265 | .ndo_get_stats64 = xennet_get_stats64, |
0a0b9d2e SH |
1266 | .ndo_set_mac_address = eth_mac_addr, |
1267 | .ndo_validate_addr = eth_validate_addr, | |
fb507934 MM |
1268 | .ndo_fix_features = xennet_fix_features, |
1269 | .ndo_set_features = xennet_set_features, | |
2688fcb7 | 1270 | .ndo_select_queue = xennet_select_queue, |
cf66f9d4 KRW |
1271 | #ifdef CONFIG_NET_POLL_CONTROLLER |
1272 | .ndo_poll_controller = xennet_poll_controller, | |
1273 | #endif | |
0a0b9d2e SH |
1274 | }; |
1275 | ||
900e1833 DV |
1276 | static void xennet_free_netdev(struct net_device *netdev) |
1277 | { | |
1278 | struct netfront_info *np = netdev_priv(netdev); | |
1279 | ||
1280 | free_percpu(np->rx_stats); | |
1281 | free_percpu(np->tx_stats); | |
1282 | free_netdev(netdev); | |
1283 | } | |
1284 | ||
8e0e46bb | 1285 | static struct net_device *xennet_create_dev(struct xenbus_device *dev) |
0d160211 | 1286 | { |
2688fcb7 | 1287 | int err; |
0d160211 JF |
1288 | struct net_device *netdev; |
1289 | struct netfront_info *np; | |
1290 | ||
50ee6061 | 1291 | netdev = alloc_etherdev_mq(sizeof(struct netfront_info), xennet_max_queues); |
41de8d4c | 1292 | if (!netdev) |
0d160211 | 1293 | return ERR_PTR(-ENOMEM); |
0d160211 JF |
1294 | |
1295 | np = netdev_priv(netdev); | |
1296 | np->xbdev = dev; | |
1297 | ||
2688fcb7 | 1298 | np->queues = NULL; |
0d160211 | 1299 | |
e00f85be | 1300 | err = -ENOMEM; |
900e1833 DV |
1301 | np->rx_stats = netdev_alloc_pcpu_stats(struct netfront_stats); |
1302 | if (np->rx_stats == NULL) | |
1303 | goto exit; | |
1304 | np->tx_stats = netdev_alloc_pcpu_stats(struct netfront_stats); | |
1305 | if (np->tx_stats == NULL) | |
e00f85be | 1306 | goto exit; |
1307 | ||
0a0b9d2e SH |
1308 | netdev->netdev_ops = &xennet_netdev_ops; |
1309 | ||
fb507934 MM |
1310 | netdev->features = NETIF_F_IP_CSUM | NETIF_F_RXCSUM | |
1311 | NETIF_F_GSO_ROBUST; | |
2c0057de PD |
1312 | netdev->hw_features = NETIF_F_SG | |
1313 | NETIF_F_IPV6_CSUM | | |
1314 | NETIF_F_TSO | NETIF_F_TSO6; | |
0d160211 | 1315 | |
fc3e5941 IC |
1316 | /* |
1317 | * Assume that all hw features are available for now. This set | |
1318 | * will be adjusted by the call to netdev_update_features() in | |
1319 | * xennet_connect() which is the earliest point where we can | |
1320 | * negotiate with the backend regarding supported features. | |
1321 | */ | |
1322 | netdev->features |= netdev->hw_features; | |
1323 | ||
7ad24ea4 | 1324 | netdev->ethtool_ops = &xennet_ethtool_ops; |
e1043a4b | 1325 | netdev->min_mtu = ETH_MIN_MTU; |
d0c2c997 | 1326 | netdev->max_mtu = XEN_NETIF_MAX_TX_SIZE; |
0d160211 JF |
1327 | SET_NETDEV_DEV(netdev, &dev->dev); |
1328 | ||
1329 | np->netdev = netdev; | |
1330 | ||
1331 | netif_carrier_off(netdev); | |
1332 | ||
b707fda2 | 1333 | xenbus_switch_state(dev, XenbusStateInitialising); |
8edfe2e9 JG |
1334 | wait_event(module_wq, |
1335 | xenbus_read_driver_state(dev->otherend) != | |
1336 | XenbusStateClosed && | |
1337 | xenbus_read_driver_state(dev->otherend) != | |
1338 | XenbusStateUnknown); | |
0d160211 JF |
1339 | return netdev; |
1340 | ||
0d160211 | 1341 | exit: |
900e1833 | 1342 | xennet_free_netdev(netdev); |
0d160211 JF |
1343 | return ERR_PTR(err); |
1344 | } | |
1345 | ||
1346 | /** | |
1347 | * Entry point to this code when a new device is created. Allocate the basic | |
1348 | * structures and the ring buffers for communication with the backend, and | |
1349 | * inform the backend of the appropriate details for those. | |
1350 | */ | |
8e0e46bb | 1351 | static int netfront_probe(struct xenbus_device *dev, |
1dd06ae8 | 1352 | const struct xenbus_device_id *id) |
0d160211 JF |
1353 | { |
1354 | int err; | |
1355 | struct net_device *netdev; | |
1356 | struct netfront_info *info; | |
1357 | ||
1358 | netdev = xennet_create_dev(dev); | |
1359 | if (IS_ERR(netdev)) { | |
1360 | err = PTR_ERR(netdev); | |
1361 | xenbus_dev_fatal(dev, err, "creating netdev"); | |
1362 | return err; | |
1363 | } | |
1364 | ||
1365 | info = netdev_priv(netdev); | |
1b713e00 | 1366 | dev_set_drvdata(&dev->dev, info); |
27b917e5 TI |
1367 | #ifdef CONFIG_SYSFS |
1368 | info->netdev->sysfs_groups[0] = &xennet_dev_group; | |
1369 | #endif | |
0d160211 | 1370 | |
0d160211 | 1371 | return 0; |
0d160211 JF |
1372 | } |
1373 | ||
1374 | static void xennet_end_access(int ref, void *page) | |
1375 | { | |
1376 | /* This frees the page as a side-effect */ | |
1377 | if (ref != GRANT_INVALID_REF) | |
1378 | gnttab_end_foreign_access(ref, 0, (unsigned long)page); | |
1379 | } | |
1380 | ||
1381 | static void xennet_disconnect_backend(struct netfront_info *info) | |
1382 | { | |
2688fcb7 | 1383 | unsigned int i = 0; |
2688fcb7 AB |
1384 | unsigned int num_queues = info->netdev->real_num_tx_queues; |
1385 | ||
f9feb1e6 DV |
1386 | netif_carrier_off(info->netdev); |
1387 | ||
9a873c71 | 1388 | for (i = 0; i < num_queues && info->queues; ++i) { |
76541869 DV |
1389 | struct netfront_queue *queue = &info->queues[i]; |
1390 | ||
74470954 BO |
1391 | del_timer_sync(&queue->rx_refill_timer); |
1392 | ||
2688fcb7 AB |
1393 | if (queue->tx_irq && (queue->tx_irq == queue->rx_irq)) |
1394 | unbind_from_irqhandler(queue->tx_irq, queue); | |
1395 | if (queue->tx_irq && (queue->tx_irq != queue->rx_irq)) { | |
1396 | unbind_from_irqhandler(queue->tx_irq, queue); | |
1397 | unbind_from_irqhandler(queue->rx_irq, queue); | |
1398 | } | |
1399 | queue->tx_evtchn = queue->rx_evtchn = 0; | |
1400 | queue->tx_irq = queue->rx_irq = 0; | |
0d160211 | 1401 | |
274b0455 CW |
1402 | if (netif_running(info->netdev)) |
1403 | napi_synchronize(&queue->napi); | |
f9feb1e6 | 1404 | |
a5b5dc3c DV |
1405 | xennet_release_tx_bufs(queue); |
1406 | xennet_release_rx_bufs(queue); | |
1407 | gnttab_free_grant_references(queue->gref_tx_head); | |
1408 | gnttab_free_grant_references(queue->gref_rx_head); | |
1409 | ||
2688fcb7 AB |
1410 | /* End access and free the pages */ |
1411 | xennet_end_access(queue->tx_ring_ref, queue->tx.sring); | |
1412 | xennet_end_access(queue->rx_ring_ref, queue->rx.sring); | |
0d160211 | 1413 | |
2688fcb7 AB |
1414 | queue->tx_ring_ref = GRANT_INVALID_REF; |
1415 | queue->rx_ring_ref = GRANT_INVALID_REF; | |
1416 | queue->tx.sring = NULL; | |
1417 | queue->rx.sring = NULL; | |
1418 | } | |
0d160211 JF |
1419 | } |
1420 | ||
1421 | /** | |
1422 | * We are reconnecting to the backend, due to a suspend/resume, or a backend | |
1423 | * driver restart. We tear down our netif structure and recreate it, but | |
1424 | * leave the device-layer structures intact so that this is transparent to the | |
1425 | * rest of the kernel. | |
1426 | */ | |
1427 | static int netfront_resume(struct xenbus_device *dev) | |
1428 | { | |
1b713e00 | 1429 | struct netfront_info *info = dev_get_drvdata(&dev->dev); |
0d160211 JF |
1430 | |
1431 | dev_dbg(&dev->dev, "%s\n", dev->nodename); | |
1432 | ||
1433 | xennet_disconnect_backend(info); | |
1434 | return 0; | |
1435 | } | |
1436 | ||
1437 | static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[]) | |
1438 | { | |
1439 | char *s, *e, *macstr; | |
1440 | int i; | |
1441 | ||
1442 | macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL); | |
1443 | if (IS_ERR(macstr)) | |
1444 | return PTR_ERR(macstr); | |
1445 | ||
1446 | for (i = 0; i < ETH_ALEN; i++) { | |
1447 | mac[i] = simple_strtoul(s, &e, 16); | |
1448 | if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) { | |
1449 | kfree(macstr); | |
1450 | return -ENOENT; | |
1451 | } | |
1452 | s = e+1; | |
1453 | } | |
1454 | ||
1455 | kfree(macstr); | |
1456 | return 0; | |
1457 | } | |
1458 | ||
2688fcb7 | 1459 | static int setup_netfront_single(struct netfront_queue *queue) |
d634bf2c WL |
1460 | { |
1461 | int err; | |
1462 | ||
2688fcb7 | 1463 | err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn); |
d634bf2c WL |
1464 | if (err < 0) |
1465 | goto fail; | |
1466 | ||
2688fcb7 | 1467 | err = bind_evtchn_to_irqhandler(queue->tx_evtchn, |
d634bf2c | 1468 | xennet_interrupt, |
2688fcb7 | 1469 | 0, queue->info->netdev->name, queue); |
d634bf2c WL |
1470 | if (err < 0) |
1471 | goto bind_fail; | |
2688fcb7 AB |
1472 | queue->rx_evtchn = queue->tx_evtchn; |
1473 | queue->rx_irq = queue->tx_irq = err; | |
d634bf2c WL |
1474 | |
1475 | return 0; | |
1476 | ||
1477 | bind_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 | 1484 | static int setup_netfront_split(struct netfront_queue *queue) |
d634bf2c WL |
1485 | { |
1486 | int err; | |
1487 | ||
2688fcb7 | 1488 | err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn); |
d634bf2c WL |
1489 | if (err < 0) |
1490 | goto fail; | |
2688fcb7 | 1491 | err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->rx_evtchn); |
d634bf2c WL |
1492 | if (err < 0) |
1493 | goto alloc_rx_evtchn_fail; | |
1494 | ||
2688fcb7 AB |
1495 | snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name), |
1496 | "%s-tx", queue->name); | |
1497 | err = bind_evtchn_to_irqhandler(queue->tx_evtchn, | |
d634bf2c | 1498 | xennet_tx_interrupt, |
2688fcb7 | 1499 | 0, queue->tx_irq_name, queue); |
d634bf2c WL |
1500 | if (err < 0) |
1501 | goto bind_tx_fail; | |
2688fcb7 | 1502 | queue->tx_irq = err; |
d634bf2c | 1503 | |
2688fcb7 AB |
1504 | snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name), |
1505 | "%s-rx", queue->name); | |
1506 | err = bind_evtchn_to_irqhandler(queue->rx_evtchn, | |
d634bf2c | 1507 | xennet_rx_interrupt, |
2688fcb7 | 1508 | 0, queue->rx_irq_name, queue); |
d634bf2c WL |
1509 | if (err < 0) |
1510 | goto bind_rx_fail; | |
2688fcb7 | 1511 | queue->rx_irq = err; |
d634bf2c WL |
1512 | |
1513 | return 0; | |
1514 | ||
1515 | bind_rx_fail: | |
2688fcb7 AB |
1516 | unbind_from_irqhandler(queue->tx_irq, queue); |
1517 | queue->tx_irq = 0; | |
d634bf2c | 1518 | bind_tx_fail: |
2688fcb7 AB |
1519 | xenbus_free_evtchn(queue->info->xbdev, queue->rx_evtchn); |
1520 | queue->rx_evtchn = 0; | |
d634bf2c | 1521 | alloc_rx_evtchn_fail: |
2688fcb7 AB |
1522 | xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn); |
1523 | queue->tx_evtchn = 0; | |
d634bf2c WL |
1524 | fail: |
1525 | return err; | |
1526 | } | |
1527 | ||
2688fcb7 AB |
1528 | static int setup_netfront(struct xenbus_device *dev, |
1529 | struct netfront_queue *queue, unsigned int feature_split_evtchn) | |
0d160211 JF |
1530 | { |
1531 | struct xen_netif_tx_sring *txs; | |
1532 | struct xen_netif_rx_sring *rxs; | |
ccc9d90a | 1533 | grant_ref_t gref; |
0d160211 | 1534 | int err; |
0d160211 | 1535 | |
2688fcb7 AB |
1536 | queue->tx_ring_ref = GRANT_INVALID_REF; |
1537 | queue->rx_ring_ref = GRANT_INVALID_REF; | |
1538 | queue->rx.sring = NULL; | |
1539 | queue->tx.sring = NULL; | |
0d160211 | 1540 | |
a144ff09 | 1541 | txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH); |
0d160211 JF |
1542 | if (!txs) { |
1543 | err = -ENOMEM; | |
1544 | xenbus_dev_fatal(dev, err, "allocating tx ring page"); | |
1545 | goto fail; | |
1546 | } | |
1547 | SHARED_RING_INIT(txs); | |
30c5d7f0 | 1548 | FRONT_RING_INIT(&queue->tx, txs, XEN_PAGE_SIZE); |
0d160211 | 1549 | |
ccc9d90a | 1550 | err = xenbus_grant_ring(dev, txs, 1, &gref); |
1ca2983a WL |
1551 | if (err < 0) |
1552 | goto grant_tx_ring_fail; | |
ccc9d90a | 1553 | queue->tx_ring_ref = gref; |
0d160211 | 1554 | |
a144ff09 | 1555 | rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH); |
0d160211 JF |
1556 | if (!rxs) { |
1557 | err = -ENOMEM; | |
1558 | xenbus_dev_fatal(dev, err, "allocating rx ring page"); | |
1ca2983a | 1559 | goto alloc_rx_ring_fail; |
0d160211 JF |
1560 | } |
1561 | SHARED_RING_INIT(rxs); | |
30c5d7f0 | 1562 | FRONT_RING_INIT(&queue->rx, rxs, XEN_PAGE_SIZE); |
0d160211 | 1563 | |
ccc9d90a | 1564 | err = xenbus_grant_ring(dev, rxs, 1, &gref); |
1ca2983a WL |
1565 | if (err < 0) |
1566 | goto grant_rx_ring_fail; | |
ccc9d90a | 1567 | queue->rx_ring_ref = gref; |
0d160211 | 1568 | |
d634bf2c | 1569 | if (feature_split_evtchn) |
2688fcb7 | 1570 | err = setup_netfront_split(queue); |
d634bf2c WL |
1571 | /* setup single event channel if |
1572 | * a) feature-split-event-channels == 0 | |
1573 | * b) feature-split-event-channels == 1 but failed to setup | |
1574 | */ | |
1575 | if (!feature_split_evtchn || (feature_split_evtchn && err)) | |
2688fcb7 | 1576 | err = setup_netfront_single(queue); |
d634bf2c | 1577 | |
0d160211 | 1578 | if (err) |
1ca2983a | 1579 | goto alloc_evtchn_fail; |
0d160211 | 1580 | |
0d160211 JF |
1581 | return 0; |
1582 | ||
1ca2983a WL |
1583 | /* If we fail to setup netfront, it is safe to just revoke access to |
1584 | * granted pages because backend is not accessing it at this point. | |
1585 | */ | |
1ca2983a | 1586 | alloc_evtchn_fail: |
2688fcb7 | 1587 | gnttab_end_foreign_access_ref(queue->rx_ring_ref, 0); |
1ca2983a WL |
1588 | grant_rx_ring_fail: |
1589 | free_page((unsigned long)rxs); | |
1590 | alloc_rx_ring_fail: | |
2688fcb7 | 1591 | gnttab_end_foreign_access_ref(queue->tx_ring_ref, 0); |
1ca2983a WL |
1592 | grant_tx_ring_fail: |
1593 | free_page((unsigned long)txs); | |
1594 | fail: | |
0d160211 JF |
1595 | return err; |
1596 | } | |
1597 | ||
2688fcb7 AB |
1598 | /* Queue-specific initialisation |
1599 | * This used to be done in xennet_create_dev() but must now | |
1600 | * be run per-queue. | |
1601 | */ | |
1602 | static int xennet_init_queue(struct netfront_queue *queue) | |
1603 | { | |
1604 | unsigned short i; | |
1605 | int err = 0; | |
21f2706b | 1606 | char *devid; |
2688fcb7 AB |
1607 | |
1608 | spin_lock_init(&queue->tx_lock); | |
1609 | spin_lock_init(&queue->rx_lock); | |
1610 | ||
e99e88a9 | 1611 | timer_setup(&queue->rx_refill_timer, rx_refill_timeout, 0); |
2688fcb7 | 1612 | |
21f2706b XL |
1613 | devid = strrchr(queue->info->xbdev->nodename, '/') + 1; |
1614 | snprintf(queue->name, sizeof(queue->name), "vif%s-q%u", | |
1615 | devid, queue->id); | |
8b715010 | 1616 | |
2688fcb7 AB |
1617 | /* Initialise tx_skbs as a free chain containing every entry. */ |
1618 | queue->tx_skb_freelist = 0; | |
1619 | for (i = 0; i < NET_TX_RING_SIZE; i++) { | |
1620 | skb_entry_set_link(&queue->tx_skbs[i], i+1); | |
1621 | queue->grant_tx_ref[i] = GRANT_INVALID_REF; | |
1622 | queue->grant_tx_page[i] = NULL; | |
1623 | } | |
1624 | ||
1625 | /* Clear out rx_skbs */ | |
1626 | for (i = 0; i < NET_RX_RING_SIZE; i++) { | |
1627 | queue->rx_skbs[i] = NULL; | |
1628 | queue->grant_rx_ref[i] = GRANT_INVALID_REF; | |
1629 | } | |
1630 | ||
1631 | /* A grant for every tx ring slot */ | |
1f3c2eba | 1632 | if (gnttab_alloc_grant_references(NET_TX_RING_SIZE, |
2688fcb7 AB |
1633 | &queue->gref_tx_head) < 0) { |
1634 | pr_alert("can't alloc tx grant refs\n"); | |
1635 | err = -ENOMEM; | |
1636 | goto exit; | |
1637 | } | |
1638 | ||
1639 | /* A grant for every rx ring slot */ | |
1f3c2eba | 1640 | if (gnttab_alloc_grant_references(NET_RX_RING_SIZE, |
2688fcb7 AB |
1641 | &queue->gref_rx_head) < 0) { |
1642 | pr_alert("can't alloc rx grant refs\n"); | |
1643 | err = -ENOMEM; | |
1644 | goto exit_free_tx; | |
1645 | } | |
1646 | ||
2688fcb7 AB |
1647 | return 0; |
1648 | ||
1649 | exit_free_tx: | |
1650 | gnttab_free_grant_references(queue->gref_tx_head); | |
1651 | exit: | |
1652 | return err; | |
1653 | } | |
1654 | ||
50ee6061 AB |
1655 | static int write_queue_xenstore_keys(struct netfront_queue *queue, |
1656 | struct xenbus_transaction *xbt, int write_hierarchical) | |
1657 | { | |
1658 | /* Write the queue-specific keys into XenStore in the traditional | |
1659 | * way for a single queue, or in a queue subkeys for multiple | |
1660 | * queues. | |
1661 | */ | |
1662 | struct xenbus_device *dev = queue->info->xbdev; | |
1663 | int err; | |
1664 | const char *message; | |
1665 | char *path; | |
1666 | size_t pathsize; | |
1667 | ||
1668 | /* Choose the correct place to write the keys */ | |
1669 | if (write_hierarchical) { | |
1670 | pathsize = strlen(dev->nodename) + 10; | |
1671 | path = kzalloc(pathsize, GFP_KERNEL); | |
1672 | if (!path) { | |
1673 | err = -ENOMEM; | |
1674 | message = "out of memory while writing ring references"; | |
1675 | goto error; | |
1676 | } | |
1677 | snprintf(path, pathsize, "%s/queue-%u", | |
1678 | dev->nodename, queue->id); | |
1679 | } else { | |
1680 | path = (char *)dev->nodename; | |
1681 | } | |
1682 | ||
1683 | /* Write ring references */ | |
1684 | err = xenbus_printf(*xbt, path, "tx-ring-ref", "%u", | |
1685 | queue->tx_ring_ref); | |
1686 | if (err) { | |
1687 | message = "writing tx-ring-ref"; | |
1688 | goto error; | |
1689 | } | |
1690 | ||
1691 | err = xenbus_printf(*xbt, path, "rx-ring-ref", "%u", | |
1692 | queue->rx_ring_ref); | |
1693 | if (err) { | |
1694 | message = "writing rx-ring-ref"; | |
1695 | goto error; | |
1696 | } | |
1697 | ||
1698 | /* Write event channels; taking into account both shared | |
1699 | * and split event channel scenarios. | |
1700 | */ | |
1701 | if (queue->tx_evtchn == queue->rx_evtchn) { | |
1702 | /* Shared event channel */ | |
1703 | err = xenbus_printf(*xbt, path, | |
1704 | "event-channel", "%u", queue->tx_evtchn); | |
1705 | if (err) { | |
1706 | message = "writing event-channel"; | |
1707 | goto error; | |
1708 | } | |
1709 | } else { | |
1710 | /* Split event channels */ | |
1711 | err = xenbus_printf(*xbt, path, | |
1712 | "event-channel-tx", "%u", queue->tx_evtchn); | |
1713 | if (err) { | |
1714 | message = "writing event-channel-tx"; | |
1715 | goto error; | |
1716 | } | |
1717 | ||
1718 | err = xenbus_printf(*xbt, path, | |
1719 | "event-channel-rx", "%u", queue->rx_evtchn); | |
1720 | if (err) { | |
1721 | message = "writing event-channel-rx"; | |
1722 | goto error; | |
1723 | } | |
1724 | } | |
1725 | ||
1726 | if (write_hierarchical) | |
1727 | kfree(path); | |
1728 | return 0; | |
1729 | ||
1730 | error: | |
1731 | if (write_hierarchical) | |
1732 | kfree(path); | |
1733 | xenbus_dev_fatal(dev, err, "%s", message); | |
1734 | return err; | |
1735 | } | |
1736 | ||
ce58725f DV |
1737 | static void xennet_destroy_queues(struct netfront_info *info) |
1738 | { | |
1739 | unsigned int i; | |
1740 | ||
ce58725f DV |
1741 | for (i = 0; i < info->netdev->real_num_tx_queues; i++) { |
1742 | struct netfront_queue *queue = &info->queues[i]; | |
1743 | ||
1744 | if (netif_running(info->netdev)) | |
1745 | napi_disable(&queue->napi); | |
1746 | netif_napi_del(&queue->napi); | |
1747 | } | |
1748 | ||
ce58725f DV |
1749 | kfree(info->queues); |
1750 | info->queues = NULL; | |
1751 | } | |
1752 | ||
1753 | static int xennet_create_queues(struct netfront_info *info, | |
ca88ea12 | 1754 | unsigned int *num_queues) |
ce58725f DV |
1755 | { |
1756 | unsigned int i; | |
1757 | int ret; | |
1758 | ||
ca88ea12 | 1759 | info->queues = kcalloc(*num_queues, sizeof(struct netfront_queue), |
ce58725f DV |
1760 | GFP_KERNEL); |
1761 | if (!info->queues) | |
1762 | return -ENOMEM; | |
1763 | ||
ca88ea12 | 1764 | for (i = 0; i < *num_queues; i++) { |
ce58725f DV |
1765 | struct netfront_queue *queue = &info->queues[i]; |
1766 | ||
1767 | queue->id = i; | |
1768 | queue->info = info; | |
1769 | ||
1770 | ret = xennet_init_queue(queue); | |
1771 | if (ret < 0) { | |
f599c64f | 1772 | dev_warn(&info->xbdev->dev, |
69cb8524 | 1773 | "only created %d queues\n", i); |
ca88ea12 | 1774 | *num_queues = i; |
ce58725f DV |
1775 | break; |
1776 | } | |
1777 | ||
1778 | netif_napi_add(queue->info->netdev, &queue->napi, | |
1779 | xennet_poll, 64); | |
1780 | if (netif_running(info->netdev)) | |
1781 | napi_enable(&queue->napi); | |
1782 | } | |
1783 | ||
ca88ea12 | 1784 | netif_set_real_num_tx_queues(info->netdev, *num_queues); |
ce58725f | 1785 | |
ca88ea12 | 1786 | if (*num_queues == 0) { |
f599c64f | 1787 | dev_err(&info->xbdev->dev, "no queues\n"); |
ce58725f DV |
1788 | return -EINVAL; |
1789 | } | |
1790 | return 0; | |
1791 | } | |
1792 | ||
0d160211 | 1793 | /* Common code used when first setting up, and when resuming. */ |
f502bf2b | 1794 | static int talk_to_netback(struct xenbus_device *dev, |
0d160211 JF |
1795 | struct netfront_info *info) |
1796 | { | |
1797 | const char *message; | |
1798 | struct xenbus_transaction xbt; | |
1799 | int err; | |
2688fcb7 AB |
1800 | unsigned int feature_split_evtchn; |
1801 | unsigned int i = 0; | |
50ee6061 | 1802 | unsigned int max_queues = 0; |
2688fcb7 AB |
1803 | struct netfront_queue *queue = NULL; |
1804 | unsigned int num_queues = 1; | |
0d160211 | 1805 | |
2688fcb7 AB |
1806 | info->netdev->irq = 0; |
1807 | ||
50ee6061 | 1808 | /* Check if backend supports multiple queues */ |
2890ea5c JG |
1809 | max_queues = xenbus_read_unsigned(info->xbdev->otherend, |
1810 | "multi-queue-max-queues", 1); | |
50ee6061 AB |
1811 | num_queues = min(max_queues, xennet_max_queues); |
1812 | ||
2688fcb7 | 1813 | /* Check feature-split-event-channels */ |
2890ea5c JG |
1814 | feature_split_evtchn = xenbus_read_unsigned(info->xbdev->otherend, |
1815 | "feature-split-event-channels", 0); | |
2688fcb7 AB |
1816 | |
1817 | /* Read mac addr. */ | |
1818 | err = xen_net_read_mac(dev, info->netdev->dev_addr); | |
1819 | if (err) { | |
1820 | xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename); | |
cb257783 | 1821 | goto out_unlocked; |
2688fcb7 AB |
1822 | } |
1823 | ||
f599c64f | 1824 | rtnl_lock(); |
ce58725f DV |
1825 | if (info->queues) |
1826 | xennet_destroy_queues(info); | |
1827 | ||
ca88ea12 | 1828 | err = xennet_create_queues(info, &num_queues); |
e2e004ac RL |
1829 | if (err < 0) { |
1830 | xenbus_dev_fatal(dev, err, "creating queues"); | |
1831 | kfree(info->queues); | |
1832 | info->queues = NULL; | |
1833 | goto out; | |
1834 | } | |
f599c64f | 1835 | rtnl_unlock(); |
2688fcb7 AB |
1836 | |
1837 | /* Create shared ring, alloc event channel -- for each queue */ | |
1838 | for (i = 0; i < num_queues; ++i) { | |
1839 | queue = &info->queues[i]; | |
2688fcb7 | 1840 | err = setup_netfront(dev, queue, feature_split_evtchn); |
e2e004ac RL |
1841 | if (err) |
1842 | goto destroy_ring; | |
2688fcb7 | 1843 | } |
0d160211 JF |
1844 | |
1845 | again: | |
1846 | err = xenbus_transaction_start(&xbt); | |
1847 | if (err) { | |
1848 | xenbus_dev_fatal(dev, err, "starting transaction"); | |
1849 | goto destroy_ring; | |
1850 | } | |
1851 | ||
812494d9 | 1852 | if (xenbus_exists(XBT_NIL, |
1853 | info->xbdev->otherend, "multi-queue-max-queues")) { | |
50ee6061 | 1854 | /* Write the number of queues */ |
812494d9 | 1855 | err = xenbus_printf(xbt, dev->nodename, |
1856 | "multi-queue-num-queues", "%u", num_queues); | |
d634bf2c | 1857 | if (err) { |
50ee6061 AB |
1858 | message = "writing multi-queue-num-queues"; |
1859 | goto abort_transaction_no_dev_fatal; | |
d634bf2c | 1860 | } |
812494d9 | 1861 | } |
50ee6061 | 1862 | |
812494d9 | 1863 | if (num_queues == 1) { |
1864 | err = write_queue_xenstore_keys(&info->queues[0], &xbt, 0); /* flat */ | |
1865 | if (err) | |
1866 | goto abort_transaction_no_dev_fatal; | |
1867 | } else { | |
50ee6061 AB |
1868 | /* Write the keys for each queue */ |
1869 | for (i = 0; i < num_queues; ++i) { | |
1870 | queue = &info->queues[i]; | |
1871 | err = write_queue_xenstore_keys(queue, &xbt, 1); /* hierarchical */ | |
1872 | if (err) | |
1873 | goto abort_transaction_no_dev_fatal; | |
d634bf2c | 1874 | } |
0d160211 JF |
1875 | } |
1876 | ||
50ee6061 | 1877 | /* The remaining keys are not queue-specific */ |
0d160211 JF |
1878 | err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u", |
1879 | 1); | |
1880 | if (err) { | |
1881 | message = "writing request-rx-copy"; | |
1882 | goto abort_transaction; | |
1883 | } | |
1884 | ||
1885 | err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1); | |
1886 | if (err) { | |
1887 | message = "writing feature-rx-notify"; | |
1888 | goto abort_transaction; | |
1889 | } | |
1890 | ||
1891 | err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1); | |
1892 | if (err) { | |
1893 | message = "writing feature-sg"; | |
1894 | goto abort_transaction; | |
1895 | } | |
1896 | ||
1897 | err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1); | |
1898 | if (err) { | |
1899 | message = "writing feature-gso-tcpv4"; | |
1900 | goto abort_transaction; | |
1901 | } | |
1902 | ||
2c0057de PD |
1903 | err = xenbus_write(xbt, dev->nodename, "feature-gso-tcpv6", "1"); |
1904 | if (err) { | |
1905 | message = "writing feature-gso-tcpv6"; | |
1906 | goto abort_transaction; | |
1907 | } | |
1908 | ||
1909 | err = xenbus_write(xbt, dev->nodename, "feature-ipv6-csum-offload", | |
1910 | "1"); | |
1911 | if (err) { | |
1912 | message = "writing feature-ipv6-csum-offload"; | |
1913 | goto abort_transaction; | |
1914 | } | |
1915 | ||
0d160211 JF |
1916 | err = xenbus_transaction_end(xbt, 0); |
1917 | if (err) { | |
1918 | if (err == -EAGAIN) | |
1919 | goto again; | |
1920 | xenbus_dev_fatal(dev, err, "completing transaction"); | |
1921 | goto destroy_ring; | |
1922 | } | |
1923 | ||
1924 | return 0; | |
1925 | ||
1926 | abort_transaction: | |
0d160211 | 1927 | xenbus_dev_fatal(dev, err, "%s", message); |
50ee6061 AB |
1928 | abort_transaction_no_dev_fatal: |
1929 | xenbus_transaction_end(xbt, 1); | |
0d160211 JF |
1930 | destroy_ring: |
1931 | xennet_disconnect_backend(info); | |
f599c64f | 1932 | rtnl_lock(); |
e2e004ac | 1933 | xennet_destroy_queues(info); |
0d160211 | 1934 | out: |
f599c64f | 1935 | rtnl_unlock(); |
cb257783 | 1936 | out_unlocked: |
d86b5672 | 1937 | device_unregister(&dev->dev); |
0d160211 JF |
1938 | return err; |
1939 | } | |
1940 | ||
0d160211 JF |
1941 | static int xennet_connect(struct net_device *dev) |
1942 | { | |
1943 | struct netfront_info *np = netdev_priv(dev); | |
2688fcb7 | 1944 | unsigned int num_queues = 0; |
a5b5dc3c | 1945 | int err; |
2688fcb7 AB |
1946 | unsigned int j = 0; |
1947 | struct netfront_queue *queue = NULL; | |
0d160211 | 1948 | |
2890ea5c | 1949 | if (!xenbus_read_unsigned(np->xbdev->otherend, "feature-rx-copy", 0)) { |
0d160211 | 1950 | dev_info(&dev->dev, |
898eb71c | 1951 | "backend does not support copying receive path\n"); |
0d160211 JF |
1952 | return -ENODEV; |
1953 | } | |
1954 | ||
f502bf2b | 1955 | err = talk_to_netback(np->xbdev, np); |
0d160211 JF |
1956 | if (err) |
1957 | return err; | |
1958 | ||
2688fcb7 AB |
1959 | /* talk_to_netback() sets the correct number of queues */ |
1960 | num_queues = dev->real_num_tx_queues; | |
1961 | ||
f599c64f RL |
1962 | if (dev->reg_state == NETREG_UNINITIALIZED) { |
1963 | err = register_netdev(dev); | |
1964 | if (err) { | |
1965 | pr_warn("%s: register_netdev err=%d\n", __func__, err); | |
1966 | device_unregister(&np->xbdev->dev); | |
1967 | return err; | |
1968 | } | |
1969 | } | |
1970 | ||
45c8184c RL |
1971 | rtnl_lock(); |
1972 | netdev_update_features(dev); | |
1973 | rtnl_unlock(); | |
1974 | ||
0d160211 | 1975 | /* |
a5b5dc3c | 1976 | * All public and private state should now be sane. Get |
0d160211 JF |
1977 | * ready to start sending and receiving packets and give the driver |
1978 | * domain a kick because we've probably just requeued some | |
1979 | * packets. | |
1980 | */ | |
1981 | netif_carrier_on(np->netdev); | |
2688fcb7 AB |
1982 | for (j = 0; j < num_queues; ++j) { |
1983 | queue = &np->queues[j]; | |
f50b4076 | 1984 | |
2688fcb7 AB |
1985 | notify_remote_via_irq(queue->tx_irq); |
1986 | if (queue->tx_irq != queue->rx_irq) | |
1987 | notify_remote_via_irq(queue->rx_irq); | |
2688fcb7 | 1988 | |
f50b4076 DV |
1989 | spin_lock_irq(&queue->tx_lock); |
1990 | xennet_tx_buf_gc(queue); | |
2688fcb7 | 1991 | spin_unlock_irq(&queue->tx_lock); |
f50b4076 DV |
1992 | |
1993 | spin_lock_bh(&queue->rx_lock); | |
1994 | xennet_alloc_rx_buffers(queue); | |
2688fcb7 AB |
1995 | spin_unlock_bh(&queue->rx_lock); |
1996 | } | |
0d160211 JF |
1997 | |
1998 | return 0; | |
1999 | } | |
2000 | ||
2001 | /** | |
2002 | * Callback received when the backend's state changes. | |
2003 | */ | |
f502bf2b | 2004 | static void netback_changed(struct xenbus_device *dev, |
0d160211 JF |
2005 | enum xenbus_state backend_state) |
2006 | { | |
1b713e00 | 2007 | struct netfront_info *np = dev_get_drvdata(&dev->dev); |
0d160211 JF |
2008 | struct net_device *netdev = np->netdev; |
2009 | ||
2010 | dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state)); | |
2011 | ||
8edfe2e9 JG |
2012 | wake_up_all(&module_wq); |
2013 | ||
0d160211 JF |
2014 | switch (backend_state) { |
2015 | case XenbusStateInitialising: | |
2016 | case XenbusStateInitialised: | |
b78c9512 NI |
2017 | case XenbusStateReconfiguring: |
2018 | case XenbusStateReconfigured: | |
0d160211 | 2019 | case XenbusStateUnknown: |
0d160211 JF |
2020 | break; |
2021 | ||
2022 | case XenbusStateInitWait: | |
2023 | if (dev->state != XenbusStateInitialising) | |
2024 | break; | |
2025 | if (xennet_connect(netdev) != 0) | |
2026 | break; | |
2027 | xenbus_switch_state(dev, XenbusStateConnected); | |
08e34eb1 LE |
2028 | break; |
2029 | ||
2030 | case XenbusStateConnected: | |
ee89bab1 | 2031 | netdev_notify_peers(netdev); |
0d160211 JF |
2032 | break; |
2033 | ||
bce3ea81 DV |
2034 | case XenbusStateClosed: |
2035 | if (dev->state == XenbusStateClosed) | |
2036 | break; | |
2037 | /* Missed the backend's CLOSING state -- fallthrough */ | |
0d160211 JF |
2038 | case XenbusStateClosing: |
2039 | xenbus_frontend_closed(dev); | |
2040 | break; | |
2041 | } | |
2042 | } | |
2043 | ||
e0ce4af9 IC |
2044 | static const struct xennet_stat { |
2045 | char name[ETH_GSTRING_LEN]; | |
2046 | u16 offset; | |
2047 | } xennet_stats[] = { | |
2048 | { | |
2049 | "rx_gso_checksum_fixup", | |
2050 | offsetof(struct netfront_info, rx_gso_checksum_fixup) | |
2051 | }, | |
2052 | }; | |
2053 | ||
2054 | static int xennet_get_sset_count(struct net_device *dev, int string_set) | |
2055 | { | |
2056 | switch (string_set) { | |
2057 | case ETH_SS_STATS: | |
2058 | return ARRAY_SIZE(xennet_stats); | |
2059 | default: | |
2060 | return -EINVAL; | |
2061 | } | |
2062 | } | |
2063 | ||
2064 | static void xennet_get_ethtool_stats(struct net_device *dev, | |
2065 | struct ethtool_stats *stats, u64 * data) | |
2066 | { | |
2067 | void *np = netdev_priv(dev); | |
2068 | int i; | |
2069 | ||
2070 | for (i = 0; i < ARRAY_SIZE(xennet_stats); i++) | |
2688fcb7 | 2071 | data[i] = atomic_read((atomic_t *)(np + xennet_stats[i].offset)); |
e0ce4af9 IC |
2072 | } |
2073 | ||
2074 | static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data) | |
2075 | { | |
2076 | int i; | |
2077 | ||
2078 | switch (stringset) { | |
2079 | case ETH_SS_STATS: | |
2080 | for (i = 0; i < ARRAY_SIZE(xennet_stats); i++) | |
2081 | memcpy(data + i * ETH_GSTRING_LEN, | |
2082 | xennet_stats[i].name, ETH_GSTRING_LEN); | |
2083 | break; | |
2084 | } | |
2085 | } | |
2086 | ||
0fc0b732 | 2087 | static const struct ethtool_ops xennet_ethtool_ops = |
0d160211 | 2088 | { |
0d160211 | 2089 | .get_link = ethtool_op_get_link, |
e0ce4af9 IC |
2090 | |
2091 | .get_sset_count = xennet_get_sset_count, | |
2092 | .get_ethtool_stats = xennet_get_ethtool_stats, | |
2093 | .get_strings = xennet_get_strings, | |
0d160211 JF |
2094 | }; |
2095 | ||
2096 | #ifdef CONFIG_SYSFS | |
1f3c2eba DV |
2097 | static ssize_t show_rxbuf(struct device *dev, |
2098 | struct device_attribute *attr, char *buf) | |
0d160211 | 2099 | { |
1f3c2eba | 2100 | return sprintf(buf, "%lu\n", NET_RX_RING_SIZE); |
0d160211 JF |
2101 | } |
2102 | ||
1f3c2eba DV |
2103 | static ssize_t store_rxbuf(struct device *dev, |
2104 | struct device_attribute *attr, | |
2105 | const char *buf, size_t len) | |
0d160211 | 2106 | { |
0d160211 JF |
2107 | char *endp; |
2108 | unsigned long target; | |
2109 | ||
2110 | if (!capable(CAP_NET_ADMIN)) | |
2111 | return -EPERM; | |
2112 | ||
2113 | target = simple_strtoul(buf, &endp, 0); | |
2114 | if (endp == buf) | |
2115 | return -EBADMSG; | |
2116 | ||
1f3c2eba | 2117 | /* rxbuf_min and rxbuf_max are no longer configurable. */ |
0d160211 | 2118 | |
0d160211 JF |
2119 | return len; |
2120 | } | |
2121 | ||
d61e4038 JP |
2122 | static DEVICE_ATTR(rxbuf_min, 0644, show_rxbuf, store_rxbuf); |
2123 | static DEVICE_ATTR(rxbuf_max, 0644, show_rxbuf, store_rxbuf); | |
2124 | static DEVICE_ATTR(rxbuf_cur, 0444, show_rxbuf, NULL); | |
0d160211 | 2125 | |
27b917e5 TI |
2126 | static struct attribute *xennet_dev_attrs[] = { |
2127 | &dev_attr_rxbuf_min.attr, | |
2128 | &dev_attr_rxbuf_max.attr, | |
2129 | &dev_attr_rxbuf_cur.attr, | |
2130 | NULL | |
2131 | }; | |
0d160211 | 2132 | |
27b917e5 TI |
2133 | static const struct attribute_group xennet_dev_group = { |
2134 | .attrs = xennet_dev_attrs | |
2135 | }; | |
0d160211 JF |
2136 | #endif /* CONFIG_SYSFS */ |
2137 | ||
8e0e46bb | 2138 | static int xennet_remove(struct xenbus_device *dev) |
0d160211 | 2139 | { |
1b713e00 | 2140 | struct netfront_info *info = dev_get_drvdata(&dev->dev); |
0d160211 JF |
2141 | |
2142 | dev_dbg(&dev->dev, "%s\n", dev->nodename); | |
2143 | ||
5b5971df EO |
2144 | if (xenbus_read_driver_state(dev->otherend) != XenbusStateClosed) { |
2145 | xenbus_switch_state(dev, XenbusStateClosing); | |
8edfe2e9 | 2146 | wait_event(module_wq, |
5b5971df | 2147 | xenbus_read_driver_state(dev->otherend) == |
c2d2e673 JA |
2148 | XenbusStateClosing || |
2149 | xenbus_read_driver_state(dev->otherend) == | |
2150 | XenbusStateUnknown); | |
5b5971df EO |
2151 | |
2152 | xenbus_switch_state(dev, XenbusStateClosed); | |
8edfe2e9 | 2153 | wait_event(module_wq, |
5b5971df EO |
2154 | xenbus_read_driver_state(dev->otherend) == |
2155 | XenbusStateClosed || | |
2156 | xenbus_read_driver_state(dev->otherend) == | |
2157 | XenbusStateUnknown); | |
2158 | } | |
2159 | ||
0d160211 JF |
2160 | xennet_disconnect_backend(info); |
2161 | ||
f599c64f RL |
2162 | if (info->netdev->reg_state == NETREG_REGISTERED) |
2163 | unregister_netdev(info->netdev); | |
6bc96d04 | 2164 | |
f599c64f RL |
2165 | if (info->queues) { |
2166 | rtnl_lock(); | |
9a873c71 | 2167 | xennet_destroy_queues(info); |
f599c64f RL |
2168 | rtnl_unlock(); |
2169 | } | |
900e1833 | 2170 | xennet_free_netdev(info->netdev); |
0d160211 JF |
2171 | |
2172 | return 0; | |
2173 | } | |
2174 | ||
95afae48 DV |
2175 | static const struct xenbus_device_id netfront_ids[] = { |
2176 | { "vif" }, | |
2177 | { "" } | |
2178 | }; | |
2179 | ||
2180 | static struct xenbus_driver netfront_driver = { | |
2181 | .ids = netfront_ids, | |
0d160211 | 2182 | .probe = netfront_probe, |
8e0e46bb | 2183 | .remove = xennet_remove, |
0d160211 | 2184 | .resume = netfront_resume, |
f502bf2b | 2185 | .otherend_changed = netback_changed, |
95afae48 | 2186 | }; |
0d160211 JF |
2187 | |
2188 | static int __init netif_init(void) | |
2189 | { | |
6e833587 | 2190 | if (!xen_domain()) |
0d160211 JF |
2191 | return -ENODEV; |
2192 | ||
51c71a3b | 2193 | if (!xen_has_pv_nic_devices()) |
b9136d20 IM |
2194 | return -ENODEV; |
2195 | ||
383eda32 | 2196 | pr_info("Initialising Xen virtual ethernet driver\n"); |
0d160211 | 2197 | |
034702a6 | 2198 | /* Allow as many queues as there are CPUs inut max. 8 if user has not |
32a84405 WL |
2199 | * specified a value. |
2200 | */ | |
2201 | if (xennet_max_queues == 0) | |
034702a6 JG |
2202 | xennet_max_queues = min_t(unsigned int, MAX_QUEUES_DEFAULT, |
2203 | num_online_cpus()); | |
50ee6061 | 2204 | |
ffb78a26 | 2205 | return xenbus_register_frontend(&netfront_driver); |
0d160211 JF |
2206 | } |
2207 | module_init(netif_init); | |
2208 | ||
2209 | ||
2210 | static void __exit netif_exit(void) | |
2211 | { | |
ffb78a26 | 2212 | xenbus_unregister_driver(&netfront_driver); |
0d160211 JF |
2213 | } |
2214 | module_exit(netif_exit); | |
2215 | ||
2216 | MODULE_DESCRIPTION("Xen virtual network device frontend"); | |
2217 | MODULE_LICENSE("GPL"); | |
d2f0c52b | 2218 | MODULE_ALIAS("xen:vif"); |
4f93f09b | 2219 | MODULE_ALIAS("xennet"); |