]>
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 | ||
32 | #include <linux/module.h> | |
33 | #include <linux/kernel.h> | |
34 | #include <linux/netdevice.h> | |
35 | #include <linux/etherdevice.h> | |
36 | #include <linux/skbuff.h> | |
37 | #include <linux/ethtool.h> | |
38 | #include <linux/if_ether.h> | |
9ecd1a75 | 39 | #include <net/tcp.h> |
0d160211 JF |
40 | #include <linux/udp.h> |
41 | #include <linux/moduleparam.h> | |
42 | #include <linux/mm.h> | |
5a0e3ad6 | 43 | #include <linux/slab.h> |
0d160211 JF |
44 | #include <net/ip.h> |
45 | ||
ca981633 | 46 | #include <asm/xen/page.h> |
1ccbf534 | 47 | #include <xen/xen.h> |
0d160211 JF |
48 | #include <xen/xenbus.h> |
49 | #include <xen/events.h> | |
50 | #include <xen/page.h> | |
b9136d20 | 51 | #include <xen/platform_pci.h> |
0d160211 JF |
52 | #include <xen/grant_table.h> |
53 | ||
54 | #include <xen/interface/io/netif.h> | |
55 | #include <xen/interface/memory.h> | |
56 | #include <xen/interface/grant_table.h> | |
57 | ||
0fc0b732 | 58 | static const struct ethtool_ops xennet_ethtool_ops; |
0d160211 JF |
59 | |
60 | struct netfront_cb { | |
3683243b | 61 | int pull_to; |
0d160211 JF |
62 | }; |
63 | ||
64 | #define NETFRONT_SKB_CB(skb) ((struct netfront_cb *)((skb)->cb)) | |
65 | ||
66 | #define RX_COPY_THRESHOLD 256 | |
67 | ||
68 | #define GRANT_INVALID_REF 0 | |
69 | ||
667c78af JF |
70 | #define NET_TX_RING_SIZE __CONST_RING_SIZE(xen_netif_tx, PAGE_SIZE) |
71 | #define NET_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, PAGE_SIZE) | |
40206dd9 | 72 | #define TX_MAX_TARGET min_t(int, NET_TX_RING_SIZE, 256) |
0d160211 | 73 | |
e00f85be | 74 | struct netfront_stats { |
75 | u64 rx_packets; | |
76 | u64 tx_packets; | |
77 | u64 rx_bytes; | |
78 | u64 tx_bytes; | |
79 | struct u64_stats_sync syncp; | |
80 | }; | |
81 | ||
0d160211 JF |
82 | struct netfront_info { |
83 | struct list_head list; | |
84 | struct net_device *netdev; | |
85 | ||
bea3348e | 86 | struct napi_struct napi; |
0d160211 | 87 | |
d634bf2c WL |
88 | /* Split event channels support, tx_* == rx_* when using |
89 | * single event channel. | |
90 | */ | |
91 | unsigned int tx_evtchn, rx_evtchn; | |
92 | unsigned int tx_irq, rx_irq; | |
93 | /* Only used when split event channels support is enabled */ | |
94 | char tx_irq_name[IFNAMSIZ+4]; /* DEVNAME-tx */ | |
95 | char rx_irq_name[IFNAMSIZ+4]; /* DEVNAME-rx */ | |
96 | ||
84284d3c | 97 | struct xenbus_device *xbdev; |
0d160211 | 98 | |
84284d3c JF |
99 | spinlock_t tx_lock; |
100 | struct xen_netif_tx_front_ring tx; | |
101 | int tx_ring_ref; | |
0d160211 JF |
102 | |
103 | /* | |
104 | * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries | |
105 | * are linked from tx_skb_freelist through skb_entry.link. | |
106 | * | |
107 | * NB. Freelist index entries are always going to be less than | |
108 | * PAGE_OFFSET, whereas pointers to skbs will always be equal or | |
109 | * greater than PAGE_OFFSET: we use this property to distinguish | |
110 | * them. | |
111 | */ | |
112 | union skb_entry { | |
113 | struct sk_buff *skb; | |
1ffb40b8 | 114 | unsigned long link; |
0d160211 JF |
115 | } tx_skbs[NET_TX_RING_SIZE]; |
116 | grant_ref_t gref_tx_head; | |
117 | grant_ref_t grant_tx_ref[NET_TX_RING_SIZE]; | |
118 | unsigned tx_skb_freelist; | |
119 | ||
84284d3c JF |
120 | spinlock_t rx_lock ____cacheline_aligned_in_smp; |
121 | struct xen_netif_rx_front_ring rx; | |
122 | int rx_ring_ref; | |
123 | ||
124 | /* Receive-ring batched refills. */ | |
125 | #define RX_MIN_TARGET 8 | |
126 | #define RX_DFL_MIN_TARGET 64 | |
127 | #define RX_MAX_TARGET min_t(int, NET_RX_RING_SIZE, 256) | |
128 | unsigned rx_min_target, rx_max_target, rx_target; | |
129 | struct sk_buff_head rx_batch; | |
130 | ||
131 | struct timer_list rx_refill_timer; | |
132 | ||
0d160211 JF |
133 | struct sk_buff *rx_skbs[NET_RX_RING_SIZE]; |
134 | grant_ref_t gref_rx_head; | |
135 | grant_ref_t grant_rx_ref[NET_RX_RING_SIZE]; | |
136 | ||
0d160211 JF |
137 | unsigned long rx_pfn_array[NET_RX_RING_SIZE]; |
138 | struct multicall_entry rx_mcl[NET_RX_RING_SIZE+1]; | |
139 | struct mmu_update rx_mmu[NET_RX_RING_SIZE]; | |
e0ce4af9 IC |
140 | |
141 | /* Statistics */ | |
e00f85be | 142 | struct netfront_stats __percpu *stats; |
143 | ||
e9a799ea | 144 | unsigned long rx_gso_checksum_fixup; |
0d160211 JF |
145 | }; |
146 | ||
147 | struct netfront_rx_info { | |
148 | struct xen_netif_rx_response rx; | |
149 | struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1]; | |
150 | }; | |
151 | ||
1ffb40b8 IY |
152 | static void skb_entry_set_link(union skb_entry *list, unsigned short id) |
153 | { | |
154 | list->link = id; | |
155 | } | |
156 | ||
157 | static int skb_entry_is_link(const union skb_entry *list) | |
158 | { | |
159 | BUILD_BUG_ON(sizeof(list->skb) != sizeof(list->link)); | |
807540ba | 160 | return (unsigned long)list->skb < PAGE_OFFSET; |
1ffb40b8 IY |
161 | } |
162 | ||
0d160211 JF |
163 | /* |
164 | * Access macros for acquiring freeing slots in tx_skbs[]. | |
165 | */ | |
166 | ||
167 | static void add_id_to_freelist(unsigned *head, union skb_entry *list, | |
168 | unsigned short id) | |
169 | { | |
1ffb40b8 | 170 | skb_entry_set_link(&list[id], *head); |
0d160211 JF |
171 | *head = id; |
172 | } | |
173 | ||
174 | static unsigned short get_id_from_freelist(unsigned *head, | |
175 | union skb_entry *list) | |
176 | { | |
177 | unsigned int id = *head; | |
178 | *head = list[id].link; | |
179 | return id; | |
180 | } | |
181 | ||
182 | static int xennet_rxidx(RING_IDX idx) | |
183 | { | |
184 | return idx & (NET_RX_RING_SIZE - 1); | |
185 | } | |
186 | ||
187 | static struct sk_buff *xennet_get_rx_skb(struct netfront_info *np, | |
188 | RING_IDX ri) | |
189 | { | |
190 | int i = xennet_rxidx(ri); | |
191 | struct sk_buff *skb = np->rx_skbs[i]; | |
192 | np->rx_skbs[i] = NULL; | |
193 | return skb; | |
194 | } | |
195 | ||
196 | static grant_ref_t xennet_get_rx_ref(struct netfront_info *np, | |
197 | RING_IDX ri) | |
198 | { | |
199 | int i = xennet_rxidx(ri); | |
200 | grant_ref_t ref = np->grant_rx_ref[i]; | |
201 | np->grant_rx_ref[i] = GRANT_INVALID_REF; | |
202 | return ref; | |
203 | } | |
204 | ||
205 | #ifdef CONFIG_SYSFS | |
206 | static int xennet_sysfs_addif(struct net_device *netdev); | |
207 | static void xennet_sysfs_delif(struct net_device *netdev); | |
208 | #else /* !CONFIG_SYSFS */ | |
209 | #define xennet_sysfs_addif(dev) (0) | |
210 | #define xennet_sysfs_delif(dev) do { } while (0) | |
211 | #endif | |
212 | ||
3ad9b358 | 213 | static bool xennet_can_sg(struct net_device *dev) |
0d160211 | 214 | { |
3ad9b358 | 215 | return dev->features & NETIF_F_SG; |
0d160211 JF |
216 | } |
217 | ||
218 | ||
219 | static void rx_refill_timeout(unsigned long data) | |
220 | { | |
221 | struct net_device *dev = (struct net_device *)data; | |
bea3348e | 222 | struct netfront_info *np = netdev_priv(dev); |
288379f0 | 223 | napi_schedule(&np->napi); |
0d160211 JF |
224 | } |
225 | ||
226 | static int netfront_tx_slot_available(struct netfront_info *np) | |
227 | { | |
807540ba ED |
228 | return (np->tx.req_prod_pvt - np->tx.rsp_cons) < |
229 | (TX_MAX_TARGET - MAX_SKB_FRAGS - 2); | |
0d160211 JF |
230 | } |
231 | ||
232 | static void xennet_maybe_wake_tx(struct net_device *dev) | |
233 | { | |
234 | struct netfront_info *np = netdev_priv(dev); | |
235 | ||
236 | if (unlikely(netif_queue_stopped(dev)) && | |
237 | netfront_tx_slot_available(np) && | |
238 | likely(netif_running(dev))) | |
239 | netif_wake_queue(dev); | |
240 | } | |
241 | ||
242 | static void xennet_alloc_rx_buffers(struct net_device *dev) | |
243 | { | |
244 | unsigned short id; | |
245 | struct netfront_info *np = netdev_priv(dev); | |
246 | struct sk_buff *skb; | |
247 | struct page *page; | |
248 | int i, batch_target, notify; | |
249 | RING_IDX req_prod = np->rx.req_prod_pvt; | |
0d160211 JF |
250 | grant_ref_t ref; |
251 | unsigned long pfn; | |
252 | void *vaddr; | |
0d160211 JF |
253 | struct xen_netif_rx_request *req; |
254 | ||
255 | if (unlikely(!netif_carrier_ok(dev))) | |
256 | return; | |
257 | ||
258 | /* | |
259 | * Allocate skbuffs greedily, even though we batch updates to the | |
260 | * receive ring. This creates a less bursty demand on the memory | |
261 | * allocator, so should reduce the chance of failed allocation requests | |
262 | * both for ourself and for other kernel subsystems. | |
263 | */ | |
264 | batch_target = np->rx_target - (req_prod - np->rx.rsp_cons); | |
265 | for (i = skb_queue_len(&np->rx_batch); i < batch_target; i++) { | |
617a20bb | 266 | skb = __netdev_alloc_skb(dev, RX_COPY_THRESHOLD + NET_IP_ALIGN, |
0d160211 JF |
267 | GFP_ATOMIC | __GFP_NOWARN); |
268 | if (unlikely(!skb)) | |
269 | goto no_skb; | |
270 | ||
617a20bb IY |
271 | /* Align ip header to a 16 bytes boundary */ |
272 | skb_reserve(skb, NET_IP_ALIGN); | |
273 | ||
0d160211 JF |
274 | page = alloc_page(GFP_ATOMIC | __GFP_NOWARN); |
275 | if (!page) { | |
276 | kfree_skb(skb); | |
277 | no_skb: | |
278 | /* Any skbuffs queued for refill? Force them out. */ | |
279 | if (i != 0) | |
280 | goto refill; | |
281 | /* Could not allocate any skbuffs. Try again later. */ | |
282 | mod_timer(&np->rx_refill_timer, | |
283 | jiffies + (HZ/10)); | |
284 | break; | |
285 | } | |
286 | ||
01c68026 | 287 | __skb_fill_page_desc(skb, 0, page, 0, 0); |
0d160211 JF |
288 | skb_shinfo(skb)->nr_frags = 1; |
289 | __skb_queue_tail(&np->rx_batch, skb); | |
290 | } | |
291 | ||
292 | /* Is the batch large enough to be worthwhile? */ | |
293 | if (i < (np->rx_target/2)) { | |
294 | if (req_prod > np->rx.sring->req_prod) | |
295 | goto push; | |
296 | return; | |
297 | } | |
298 | ||
299 | /* Adjust our fill target if we risked running out of buffers. */ | |
300 | if (((req_prod - np->rx.sring->rsp_prod) < (np->rx_target / 4)) && | |
301 | ((np->rx_target *= 2) > np->rx_max_target)) | |
302 | np->rx_target = np->rx_max_target; | |
303 | ||
304 | refill: | |
5dcddfae | 305 | for (i = 0; ; i++) { |
0d160211 JF |
306 | skb = __skb_dequeue(&np->rx_batch); |
307 | if (skb == NULL) | |
308 | break; | |
309 | ||
310 | skb->dev = dev; | |
311 | ||
312 | id = xennet_rxidx(req_prod + i); | |
313 | ||
314 | BUG_ON(np->rx_skbs[id]); | |
315 | np->rx_skbs[id] = skb; | |
316 | ||
317 | ref = gnttab_claim_grant_reference(&np->gref_rx_head); | |
318 | BUG_ON((signed short)ref < 0); | |
319 | np->grant_rx_ref[id] = ref; | |
320 | ||
01c68026 IC |
321 | pfn = page_to_pfn(skb_frag_page(&skb_shinfo(skb)->frags[0])); |
322 | vaddr = page_address(skb_frag_page(&skb_shinfo(skb)->frags[0])); | |
0d160211 JF |
323 | |
324 | req = RING_GET_REQUEST(&np->rx, req_prod + i); | |
325 | gnttab_grant_foreign_access_ref(ref, | |
326 | np->xbdev->otherend_id, | |
327 | pfn_to_mfn(pfn), | |
328 | 0); | |
329 | ||
330 | req->id = id; | |
331 | req->gref = ref; | |
332 | } | |
333 | ||
5dcddfae | 334 | wmb(); /* barrier so backend seens requests */ |
0d160211 JF |
335 | |
336 | /* Above is a suitable barrier to ensure backend will see requests. */ | |
337 | np->rx.req_prod_pvt = req_prod + i; | |
338 | push: | |
339 | RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np->rx, notify); | |
340 | if (notify) | |
d634bf2c | 341 | notify_remote_via_irq(np->rx_irq); |
0d160211 JF |
342 | } |
343 | ||
344 | static int xennet_open(struct net_device *dev) | |
345 | { | |
346 | struct netfront_info *np = netdev_priv(dev); | |
347 | ||
bea3348e SH |
348 | napi_enable(&np->napi); |
349 | ||
0d160211 JF |
350 | spin_lock_bh(&np->rx_lock); |
351 | if (netif_carrier_ok(dev)) { | |
352 | xennet_alloc_rx_buffers(dev); | |
353 | np->rx.sring->rsp_event = np->rx.rsp_cons + 1; | |
354 | if (RING_HAS_UNCONSUMED_RESPONSES(&np->rx)) | |
288379f0 | 355 | napi_schedule(&np->napi); |
0d160211 JF |
356 | } |
357 | spin_unlock_bh(&np->rx_lock); | |
358 | ||
0b1ab1b8 | 359 | netif_start_queue(dev); |
0d160211 JF |
360 | |
361 | return 0; | |
362 | } | |
363 | ||
364 | static void xennet_tx_buf_gc(struct net_device *dev) | |
365 | { | |
366 | RING_IDX cons, prod; | |
367 | unsigned short id; | |
368 | struct netfront_info *np = netdev_priv(dev); | |
369 | struct sk_buff *skb; | |
370 | ||
371 | BUG_ON(!netif_carrier_ok(dev)); | |
372 | ||
373 | do { | |
374 | prod = np->tx.sring->rsp_prod; | |
375 | rmb(); /* Ensure we see responses up to 'rp'. */ | |
376 | ||
377 | for (cons = np->tx.rsp_cons; cons != prod; cons++) { | |
378 | struct xen_netif_tx_response *txrsp; | |
379 | ||
380 | txrsp = RING_GET_RESPONSE(&np->tx, cons); | |
f942dc25 | 381 | if (txrsp->status == XEN_NETIF_RSP_NULL) |
0d160211 JF |
382 | continue; |
383 | ||
384 | id = txrsp->id; | |
385 | skb = np->tx_skbs[id].skb; | |
386 | if (unlikely(gnttab_query_foreign_access( | |
387 | np->grant_tx_ref[id]) != 0)) { | |
388 | printk(KERN_ALERT "xennet_tx_buf_gc: warning " | |
389 | "-- grant still in use by backend " | |
390 | "domain.\n"); | |
391 | BUG(); | |
392 | } | |
393 | gnttab_end_foreign_access_ref( | |
394 | np->grant_tx_ref[id], GNTMAP_readonly); | |
395 | gnttab_release_grant_reference( | |
396 | &np->gref_tx_head, np->grant_tx_ref[id]); | |
397 | np->grant_tx_ref[id] = GRANT_INVALID_REF; | |
398 | add_id_to_freelist(&np->tx_skb_freelist, np->tx_skbs, id); | |
399 | dev_kfree_skb_irq(skb); | |
400 | } | |
401 | ||
402 | np->tx.rsp_cons = prod; | |
403 | ||
404 | /* | |
405 | * Set a new event, then check for race with update of tx_cons. | |
406 | * Note that it is essential to schedule a callback, no matter | |
407 | * how few buffers are pending. Even if there is space in the | |
408 | * transmit ring, higher layers may be blocked because too much | |
409 | * data is outstanding: in such cases notification from Xen is | |
410 | * likely to be the only kick that we'll get. | |
411 | */ | |
412 | np->tx.sring->rsp_event = | |
413 | prod + ((np->tx.sring->req_prod - prod) >> 1) + 1; | |
414 | mb(); /* update shared area */ | |
415 | } while ((cons == prod) && (prod != np->tx.sring->rsp_prod)); | |
416 | ||
417 | xennet_maybe_wake_tx(dev); | |
418 | } | |
419 | ||
420 | static void xennet_make_frags(struct sk_buff *skb, struct net_device *dev, | |
421 | struct xen_netif_tx_request *tx) | |
422 | { | |
423 | struct netfront_info *np = netdev_priv(dev); | |
424 | char *data = skb->data; | |
425 | unsigned long mfn; | |
426 | RING_IDX prod = np->tx.req_prod_pvt; | |
427 | int frags = skb_shinfo(skb)->nr_frags; | |
428 | unsigned int offset = offset_in_page(data); | |
429 | unsigned int len = skb_headlen(skb); | |
430 | unsigned int id; | |
431 | grant_ref_t ref; | |
432 | int i; | |
433 | ||
434 | /* While the header overlaps a page boundary (including being | |
435 | larger than a page), split it it into page-sized chunks. */ | |
436 | while (len > PAGE_SIZE - offset) { | |
437 | tx->size = PAGE_SIZE - offset; | |
f942dc25 | 438 | tx->flags |= XEN_NETTXF_more_data; |
0d160211 JF |
439 | len -= tx->size; |
440 | data += tx->size; | |
441 | offset = 0; | |
442 | ||
443 | id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs); | |
444 | np->tx_skbs[id].skb = skb_get(skb); | |
445 | tx = RING_GET_REQUEST(&np->tx, prod++); | |
446 | tx->id = id; | |
447 | ref = gnttab_claim_grant_reference(&np->gref_tx_head); | |
448 | BUG_ON((signed short)ref < 0); | |
449 | ||
450 | mfn = virt_to_mfn(data); | |
451 | gnttab_grant_foreign_access_ref(ref, np->xbdev->otherend_id, | |
452 | mfn, GNTMAP_readonly); | |
453 | ||
454 | tx->gref = np->grant_tx_ref[id] = ref; | |
455 | tx->offset = offset; | |
456 | tx->size = len; | |
457 | tx->flags = 0; | |
458 | } | |
459 | ||
460 | /* Grant backend access to each skb fragment page. */ | |
461 | for (i = 0; i < frags; i++) { | |
462 | skb_frag_t *frag = skb_shinfo(skb)->frags + i; | |
f36c3747 | 463 | struct page *page = skb_frag_page(frag); |
0d160211 | 464 | |
f36c3747 IC |
465 | len = skb_frag_size(frag); |
466 | offset = frag->page_offset; | |
0d160211 | 467 | |
f36c3747 IC |
468 | /* Data must not cross a page boundary. */ |
469 | BUG_ON(len + offset > PAGE_SIZE<<compound_order(page)); | |
0d160211 | 470 | |
f36c3747 IC |
471 | /* Skip unused frames from start of page */ |
472 | page += offset >> PAGE_SHIFT; | |
473 | offset &= ~PAGE_MASK; | |
0d160211 | 474 | |
f36c3747 IC |
475 | while (len > 0) { |
476 | unsigned long bytes; | |
477 | ||
478 | BUG_ON(offset >= PAGE_SIZE); | |
479 | ||
480 | bytes = PAGE_SIZE - offset; | |
481 | if (bytes > len) | |
482 | bytes = len; | |
483 | ||
484 | tx->flags |= XEN_NETTXF_more_data; | |
485 | ||
486 | id = get_id_from_freelist(&np->tx_skb_freelist, | |
487 | np->tx_skbs); | |
488 | np->tx_skbs[id].skb = skb_get(skb); | |
489 | tx = RING_GET_REQUEST(&np->tx, prod++); | |
490 | tx->id = id; | |
491 | ref = gnttab_claim_grant_reference(&np->gref_tx_head); | |
492 | BUG_ON((signed short)ref < 0); | |
493 | ||
494 | mfn = pfn_to_mfn(page_to_pfn(page)); | |
495 | gnttab_grant_foreign_access_ref(ref, | |
496 | np->xbdev->otherend_id, | |
497 | mfn, GNTMAP_readonly); | |
498 | ||
499 | tx->gref = np->grant_tx_ref[id] = ref; | |
500 | tx->offset = offset; | |
501 | tx->size = bytes; | |
502 | tx->flags = 0; | |
503 | ||
504 | offset += bytes; | |
505 | len -= bytes; | |
506 | ||
507 | /* Next frame */ | |
508 | if (offset == PAGE_SIZE && len) { | |
509 | BUG_ON(!PageCompound(page)); | |
510 | page++; | |
511 | offset = 0; | |
512 | } | |
513 | } | |
0d160211 JF |
514 | } |
515 | ||
516 | np->tx.req_prod_pvt = prod; | |
517 | } | |
518 | ||
f36c3747 IC |
519 | /* |
520 | * Count how many ring slots are required to send the frags of this | |
521 | * skb. Each frag might be a compound page. | |
522 | */ | |
523 | static int xennet_count_skb_frag_slots(struct sk_buff *skb) | |
524 | { | |
525 | int i, frags = skb_shinfo(skb)->nr_frags; | |
526 | int pages = 0; | |
527 | ||
528 | for (i = 0; i < frags; i++) { | |
529 | skb_frag_t *frag = skb_shinfo(skb)->frags + i; | |
530 | unsigned long size = skb_frag_size(frag); | |
531 | unsigned long offset = frag->page_offset; | |
532 | ||
533 | /* Skip unused frames from start of page */ | |
534 | offset &= ~PAGE_MASK; | |
535 | ||
536 | pages += PFN_UP(offset + size); | |
537 | } | |
538 | ||
539 | return pages; | |
540 | } | |
541 | ||
0d160211 JF |
542 | static int xennet_start_xmit(struct sk_buff *skb, struct net_device *dev) |
543 | { | |
544 | unsigned short id; | |
545 | struct netfront_info *np = netdev_priv(dev); | |
e00f85be | 546 | struct netfront_stats *stats = this_cpu_ptr(np->stats); |
0d160211 | 547 | struct xen_netif_tx_request *tx; |
0d160211 JF |
548 | char *data = skb->data; |
549 | RING_IDX i; | |
550 | grant_ref_t ref; | |
551 | unsigned long mfn; | |
552 | int notify; | |
f36c3747 | 553 | int slots; |
0d160211 JF |
554 | unsigned int offset = offset_in_page(data); |
555 | unsigned int len = skb_headlen(skb); | |
cf66f9d4 | 556 | unsigned long flags; |
0d160211 | 557 | |
9ecd1a75 WL |
558 | /* If skb->len is too big for wire format, drop skb and alert |
559 | * user about misconfiguration. | |
560 | */ | |
561 | if (unlikely(skb->len > XEN_NETIF_MAX_TX_SIZE)) { | |
562 | net_alert_ratelimited( | |
563 | "xennet: skb->len = %u, too big for wire format\n", | |
564 | skb->len); | |
565 | goto drop; | |
566 | } | |
567 | ||
f36c3747 IC |
568 | slots = DIV_ROUND_UP(offset + len, PAGE_SIZE) + |
569 | xennet_count_skb_frag_slots(skb); | |
570 | if (unlikely(slots > MAX_SKB_FRAGS + 1)) { | |
571 | net_alert_ratelimited( | |
572 | "xennet: skb rides the rocket: %d slots\n", slots); | |
0d160211 JF |
573 | goto drop; |
574 | } | |
575 | ||
cf66f9d4 | 576 | spin_lock_irqsave(&np->tx_lock, flags); |
0d160211 JF |
577 | |
578 | if (unlikely(!netif_carrier_ok(dev) || | |
f36c3747 | 579 | (slots > 1 && !xennet_can_sg(dev)) || |
fc741216 | 580 | netif_needs_gso(skb, netif_skb_features(skb)))) { |
cf66f9d4 | 581 | spin_unlock_irqrestore(&np->tx_lock, flags); |
0d160211 JF |
582 | goto drop; |
583 | } | |
584 | ||
585 | i = np->tx.req_prod_pvt; | |
586 | ||
587 | id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs); | |
588 | np->tx_skbs[id].skb = skb; | |
589 | ||
590 | tx = RING_GET_REQUEST(&np->tx, i); | |
591 | ||
592 | tx->id = id; | |
593 | ref = gnttab_claim_grant_reference(&np->gref_tx_head); | |
594 | BUG_ON((signed short)ref < 0); | |
595 | mfn = virt_to_mfn(data); | |
596 | gnttab_grant_foreign_access_ref( | |
597 | ref, np->xbdev->otherend_id, mfn, GNTMAP_readonly); | |
598 | tx->gref = np->grant_tx_ref[id] = ref; | |
599 | tx->offset = offset; | |
600 | tx->size = len; | |
0d160211 JF |
601 | |
602 | tx->flags = 0; | |
603 | if (skb->ip_summed == CHECKSUM_PARTIAL) | |
604 | /* local packet? */ | |
f942dc25 | 605 | tx->flags |= XEN_NETTXF_csum_blank | XEN_NETTXF_data_validated; |
0d160211 JF |
606 | else if (skb->ip_summed == CHECKSUM_UNNECESSARY) |
607 | /* remote but checksummed. */ | |
f942dc25 | 608 | tx->flags |= XEN_NETTXF_data_validated; |
0d160211 JF |
609 | |
610 | if (skb_shinfo(skb)->gso_size) { | |
611 | struct xen_netif_extra_info *gso; | |
612 | ||
613 | gso = (struct xen_netif_extra_info *) | |
614 | RING_GET_REQUEST(&np->tx, ++i); | |
615 | ||
e2d617c0 | 616 | tx->flags |= XEN_NETTXF_extra_info; |
0d160211 JF |
617 | |
618 | gso->u.gso.size = skb_shinfo(skb)->gso_size; | |
619 | gso->u.gso.type = XEN_NETIF_GSO_TYPE_TCPV4; | |
620 | gso->u.gso.pad = 0; | |
621 | gso->u.gso.features = 0; | |
622 | ||
623 | gso->type = XEN_NETIF_EXTRA_TYPE_GSO; | |
624 | gso->flags = 0; | |
0d160211 JF |
625 | } |
626 | ||
627 | np->tx.req_prod_pvt = i + 1; | |
628 | ||
629 | xennet_make_frags(skb, dev, tx); | |
630 | tx->size = skb->len; | |
631 | ||
632 | RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np->tx, notify); | |
633 | if (notify) | |
d634bf2c | 634 | notify_remote_via_irq(np->tx_irq); |
0d160211 | 635 | |
e00f85be | 636 | u64_stats_update_begin(&stats->syncp); |
637 | stats->tx_bytes += skb->len; | |
638 | stats->tx_packets++; | |
639 | u64_stats_update_end(&stats->syncp); | |
10a273a6 JF |
640 | |
641 | /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */ | |
0d160211 JF |
642 | xennet_tx_buf_gc(dev); |
643 | ||
644 | if (!netfront_tx_slot_available(np)) | |
645 | netif_stop_queue(dev); | |
646 | ||
cf66f9d4 | 647 | spin_unlock_irqrestore(&np->tx_lock, flags); |
0d160211 | 648 | |
6ed10654 | 649 | return NETDEV_TX_OK; |
0d160211 JF |
650 | |
651 | drop: | |
09f75cd7 | 652 | dev->stats.tx_dropped++; |
0d160211 | 653 | dev_kfree_skb(skb); |
6ed10654 | 654 | return NETDEV_TX_OK; |
0d160211 JF |
655 | } |
656 | ||
657 | static int xennet_close(struct net_device *dev) | |
658 | { | |
659 | struct netfront_info *np = netdev_priv(dev); | |
660 | netif_stop_queue(np->netdev); | |
bea3348e | 661 | napi_disable(&np->napi); |
0d160211 JF |
662 | return 0; |
663 | } | |
664 | ||
0d160211 JF |
665 | static void xennet_move_rx_slot(struct netfront_info *np, struct sk_buff *skb, |
666 | grant_ref_t ref) | |
667 | { | |
668 | int new = xennet_rxidx(np->rx.req_prod_pvt); | |
669 | ||
670 | BUG_ON(np->rx_skbs[new]); | |
671 | np->rx_skbs[new] = skb; | |
672 | np->grant_rx_ref[new] = ref; | |
673 | RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->id = new; | |
674 | RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->gref = ref; | |
675 | np->rx.req_prod_pvt++; | |
676 | } | |
677 | ||
678 | static int xennet_get_extras(struct netfront_info *np, | |
679 | struct xen_netif_extra_info *extras, | |
680 | RING_IDX rp) | |
681 | ||
682 | { | |
683 | struct xen_netif_extra_info *extra; | |
684 | struct device *dev = &np->netdev->dev; | |
685 | RING_IDX cons = np->rx.rsp_cons; | |
686 | int err = 0; | |
687 | ||
688 | do { | |
689 | struct sk_buff *skb; | |
690 | grant_ref_t ref; | |
691 | ||
692 | if (unlikely(cons + 1 == rp)) { | |
693 | if (net_ratelimit()) | |
694 | dev_warn(dev, "Missing extra info\n"); | |
695 | err = -EBADR; | |
696 | break; | |
697 | } | |
698 | ||
699 | extra = (struct xen_netif_extra_info *) | |
700 | RING_GET_RESPONSE(&np->rx, ++cons); | |
701 | ||
702 | if (unlikely(!extra->type || | |
703 | extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) { | |
704 | if (net_ratelimit()) | |
705 | dev_warn(dev, "Invalid extra type: %d\n", | |
706 | extra->type); | |
707 | err = -EINVAL; | |
708 | } else { | |
709 | memcpy(&extras[extra->type - 1], extra, | |
710 | sizeof(*extra)); | |
711 | } | |
712 | ||
713 | skb = xennet_get_rx_skb(np, cons); | |
714 | ref = xennet_get_rx_ref(np, cons); | |
715 | xennet_move_rx_slot(np, skb, ref); | |
716 | } while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE); | |
717 | ||
718 | np->rx.rsp_cons = cons; | |
719 | return err; | |
720 | } | |
721 | ||
722 | static int xennet_get_responses(struct netfront_info *np, | |
723 | struct netfront_rx_info *rinfo, RING_IDX rp, | |
724 | struct sk_buff_head *list) | |
725 | { | |
726 | struct xen_netif_rx_response *rx = &rinfo->rx; | |
727 | struct xen_netif_extra_info *extras = rinfo->extras; | |
728 | struct device *dev = &np->netdev->dev; | |
729 | RING_IDX cons = np->rx.rsp_cons; | |
730 | struct sk_buff *skb = xennet_get_rx_skb(np, cons); | |
731 | grant_ref_t ref = xennet_get_rx_ref(np, cons); | |
732 | int max = MAX_SKB_FRAGS + (rx->status <= RX_COPY_THRESHOLD); | |
7158ff6d | 733 | int slots = 1; |
0d160211 JF |
734 | int err = 0; |
735 | unsigned long ret; | |
736 | ||
f942dc25 | 737 | if (rx->flags & XEN_NETRXF_extra_info) { |
0d160211 JF |
738 | err = xennet_get_extras(np, extras, rp); |
739 | cons = np->rx.rsp_cons; | |
740 | } | |
741 | ||
742 | for (;;) { | |
743 | if (unlikely(rx->status < 0 || | |
744 | rx->offset + rx->status > PAGE_SIZE)) { | |
745 | if (net_ratelimit()) | |
746 | dev_warn(dev, "rx->offset: %x, size: %u\n", | |
747 | rx->offset, rx->status); | |
748 | xennet_move_rx_slot(np, skb, ref); | |
749 | err = -EINVAL; | |
750 | goto next; | |
751 | } | |
752 | ||
753 | /* | |
754 | * This definitely indicates a bug, either in this driver or in | |
755 | * the backend driver. In future this should flag the bad | |
697089dc | 756 | * situation to the system controller to reboot the backend. |
0d160211 JF |
757 | */ |
758 | if (ref == GRANT_INVALID_REF) { | |
759 | if (net_ratelimit()) | |
760 | dev_warn(dev, "Bad rx response id %d.\n", | |
761 | rx->id); | |
762 | err = -EINVAL; | |
763 | goto next; | |
764 | } | |
765 | ||
766 | ret = gnttab_end_foreign_access_ref(ref, 0); | |
767 | BUG_ON(!ret); | |
768 | ||
769 | gnttab_release_grant_reference(&np->gref_rx_head, ref); | |
770 | ||
771 | __skb_queue_tail(list, skb); | |
772 | ||
773 | next: | |
f942dc25 | 774 | if (!(rx->flags & XEN_NETRXF_more_data)) |
0d160211 JF |
775 | break; |
776 | ||
7158ff6d | 777 | if (cons + slots == rp) { |
0d160211 | 778 | if (net_ratelimit()) |
7158ff6d | 779 | dev_warn(dev, "Need more slots\n"); |
0d160211 JF |
780 | err = -ENOENT; |
781 | break; | |
782 | } | |
783 | ||
7158ff6d WL |
784 | rx = RING_GET_RESPONSE(&np->rx, cons + slots); |
785 | skb = xennet_get_rx_skb(np, cons + slots); | |
786 | ref = xennet_get_rx_ref(np, cons + slots); | |
787 | slots++; | |
0d160211 JF |
788 | } |
789 | ||
7158ff6d | 790 | if (unlikely(slots > max)) { |
0d160211 | 791 | if (net_ratelimit()) |
697089dc | 792 | dev_warn(dev, "Too many slots\n"); |
0d160211 JF |
793 | err = -E2BIG; |
794 | } | |
795 | ||
796 | if (unlikely(err)) | |
7158ff6d | 797 | np->rx.rsp_cons = cons + slots; |
0d160211 JF |
798 | |
799 | return err; | |
800 | } | |
801 | ||
802 | static int xennet_set_skb_gso(struct sk_buff *skb, | |
803 | struct xen_netif_extra_info *gso) | |
804 | { | |
805 | if (!gso->u.gso.size) { | |
806 | if (net_ratelimit()) | |
807 | printk(KERN_WARNING "GSO size must not be zero.\n"); | |
808 | return -EINVAL; | |
809 | } | |
810 | ||
811 | /* Currently only TCPv4 S.O. is supported. */ | |
812 | if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4) { | |
813 | if (net_ratelimit()) | |
814 | printk(KERN_WARNING "Bad GSO type %d.\n", gso->u.gso.type); | |
815 | return -EINVAL; | |
816 | } | |
817 | ||
818 | skb_shinfo(skb)->gso_size = gso->u.gso.size; | |
819 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4; | |
820 | ||
821 | /* Header must be checked, and gso_segs computed. */ | |
822 | skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; | |
823 | skb_shinfo(skb)->gso_segs = 0; | |
824 | ||
825 | return 0; | |
826 | } | |
827 | ||
828 | static RING_IDX xennet_fill_frags(struct netfront_info *np, | |
829 | struct sk_buff *skb, | |
830 | struct sk_buff_head *list) | |
831 | { | |
832 | struct skb_shared_info *shinfo = skb_shinfo(skb); | |
833 | int nr_frags = shinfo->nr_frags; | |
834 | RING_IDX cons = np->rx.rsp_cons; | |
0d160211 JF |
835 | struct sk_buff *nskb; |
836 | ||
837 | while ((nskb = __skb_dequeue(list))) { | |
838 | struct xen_netif_rx_response *rx = | |
839 | RING_GET_RESPONSE(&np->rx, ++cons); | |
01c68026 | 840 | skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0]; |
0d160211 | 841 | |
01c68026 IC |
842 | __skb_fill_page_desc(skb, nr_frags, |
843 | skb_frag_page(nfrag), | |
844 | rx->offset, rx->status); | |
0d160211 JF |
845 | |
846 | skb->data_len += rx->status; | |
847 | ||
848 | skb_shinfo(nskb)->nr_frags = 0; | |
849 | kfree_skb(nskb); | |
850 | ||
0d160211 JF |
851 | nr_frags++; |
852 | } | |
853 | ||
854 | shinfo->nr_frags = nr_frags; | |
855 | return cons; | |
856 | } | |
857 | ||
e0ce4af9 | 858 | static int checksum_setup(struct net_device *dev, struct sk_buff *skb) |
0d160211 JF |
859 | { |
860 | struct iphdr *iph; | |
0d160211 | 861 | int err = -EPROTO; |
e0ce4af9 IC |
862 | int recalculate_partial_csum = 0; |
863 | ||
864 | /* | |
865 | * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy | |
866 | * peers can fail to set NETRXF_csum_blank when sending a GSO | |
867 | * frame. In this case force the SKB to CHECKSUM_PARTIAL and | |
868 | * recalculate the partial checksum. | |
869 | */ | |
870 | if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) { | |
871 | struct netfront_info *np = netdev_priv(dev); | |
872 | np->rx_gso_checksum_fixup++; | |
873 | skb->ip_summed = CHECKSUM_PARTIAL; | |
874 | recalculate_partial_csum = 1; | |
875 | } | |
876 | ||
877 | /* A non-CHECKSUM_PARTIAL SKB does not require setup. */ | |
878 | if (skb->ip_summed != CHECKSUM_PARTIAL) | |
879 | return 0; | |
0d160211 JF |
880 | |
881 | if (skb->protocol != htons(ETH_P_IP)) | |
882 | goto out; | |
883 | ||
884 | iph = (void *)skb->data; | |
0d160211 | 885 | |
0d160211 JF |
886 | switch (iph->protocol) { |
887 | case IPPROTO_TCP: | |
8249152c LR |
888 | if (!skb_partial_csum_set(skb, 4 * iph->ihl, |
889 | offsetof(struct tcphdr, check))) | |
890 | goto out; | |
e0ce4af9 IC |
891 | |
892 | if (recalculate_partial_csum) { | |
8249152c | 893 | struct tcphdr *tcph = tcp_hdr(skb); |
e0ce4af9 IC |
894 | tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, |
895 | skb->len - iph->ihl*4, | |
896 | IPPROTO_TCP, 0); | |
897 | } | |
0d160211 JF |
898 | break; |
899 | case IPPROTO_UDP: | |
8249152c LR |
900 | if (!skb_partial_csum_set(skb, 4 * iph->ihl, |
901 | offsetof(struct udphdr, check))) | |
902 | goto out; | |
e0ce4af9 IC |
903 | |
904 | if (recalculate_partial_csum) { | |
8249152c | 905 | struct udphdr *udph = udp_hdr(skb); |
e0ce4af9 IC |
906 | udph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, |
907 | skb->len - iph->ihl*4, | |
908 | IPPROTO_UDP, 0); | |
909 | } | |
0d160211 JF |
910 | break; |
911 | default: | |
912 | if (net_ratelimit()) | |
913 | printk(KERN_ERR "Attempting to checksum a non-" | |
914 | "TCP/UDP packet, dropping a protocol" | |
915 | " %d packet", iph->protocol); | |
916 | goto out; | |
917 | } | |
918 | ||
0d160211 JF |
919 | err = 0; |
920 | ||
921 | out: | |
922 | return err; | |
923 | } | |
924 | ||
925 | static int handle_incoming_queue(struct net_device *dev, | |
09f75cd7 | 926 | struct sk_buff_head *rxq) |
0d160211 | 927 | { |
e00f85be | 928 | struct netfront_info *np = netdev_priv(dev); |
929 | struct netfront_stats *stats = this_cpu_ptr(np->stats); | |
0d160211 JF |
930 | int packets_dropped = 0; |
931 | struct sk_buff *skb; | |
932 | ||
933 | while ((skb = __skb_dequeue(rxq)) != NULL) { | |
3683243b | 934 | int pull_to = NETFRONT_SKB_CB(skb)->pull_to; |
0d160211 | 935 | |
3683243b | 936 | __pskb_pull_tail(skb, pull_to - skb_headlen(skb)); |
0d160211 JF |
937 | |
938 | /* Ethernet work: Delayed to here as it peeks the header. */ | |
939 | skb->protocol = eth_type_trans(skb, dev); | |
940 | ||
e0ce4af9 IC |
941 | if (checksum_setup(dev, skb)) { |
942 | kfree_skb(skb); | |
943 | packets_dropped++; | |
944 | dev->stats.rx_errors++; | |
945 | continue; | |
0d160211 JF |
946 | } |
947 | ||
e00f85be | 948 | u64_stats_update_begin(&stats->syncp); |
949 | stats->rx_packets++; | |
950 | stats->rx_bytes += skb->len; | |
951 | u64_stats_update_end(&stats->syncp); | |
0d160211 JF |
952 | |
953 | /* Pass it up. */ | |
954 | netif_receive_skb(skb); | |
0d160211 JF |
955 | } |
956 | ||
957 | return packets_dropped; | |
958 | } | |
959 | ||
bea3348e | 960 | static int xennet_poll(struct napi_struct *napi, int budget) |
0d160211 | 961 | { |
bea3348e SH |
962 | struct netfront_info *np = container_of(napi, struct netfront_info, napi); |
963 | struct net_device *dev = np->netdev; | |
0d160211 JF |
964 | struct sk_buff *skb; |
965 | struct netfront_rx_info rinfo; | |
966 | struct xen_netif_rx_response *rx = &rinfo.rx; | |
967 | struct xen_netif_extra_info *extras = rinfo.extras; | |
968 | RING_IDX i, rp; | |
bea3348e | 969 | int work_done; |
0d160211 JF |
970 | struct sk_buff_head rxq; |
971 | struct sk_buff_head errq; | |
972 | struct sk_buff_head tmpq; | |
973 | unsigned long flags; | |
0d160211 JF |
974 | int err; |
975 | ||
976 | spin_lock(&np->rx_lock); | |
977 | ||
0d160211 JF |
978 | skb_queue_head_init(&rxq); |
979 | skb_queue_head_init(&errq); | |
980 | skb_queue_head_init(&tmpq); | |
981 | ||
0d160211 JF |
982 | rp = np->rx.sring->rsp_prod; |
983 | rmb(); /* Ensure we see queued responses up to 'rp'. */ | |
984 | ||
985 | i = np->rx.rsp_cons; | |
986 | work_done = 0; | |
987 | while ((i != rp) && (work_done < budget)) { | |
988 | memcpy(rx, RING_GET_RESPONSE(&np->rx, i), sizeof(*rx)); | |
989 | memset(extras, 0, sizeof(rinfo.extras)); | |
990 | ||
991 | err = xennet_get_responses(np, &rinfo, rp, &tmpq); | |
992 | ||
993 | if (unlikely(err)) { | |
994 | err: | |
995 | while ((skb = __skb_dequeue(&tmpq))) | |
996 | __skb_queue_tail(&errq, skb); | |
09f75cd7 | 997 | dev->stats.rx_errors++; |
0d160211 JF |
998 | i = np->rx.rsp_cons; |
999 | continue; | |
1000 | } | |
1001 | ||
1002 | skb = __skb_dequeue(&tmpq); | |
1003 | ||
1004 | if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) { | |
1005 | struct xen_netif_extra_info *gso; | |
1006 | gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1]; | |
1007 | ||
1008 | if (unlikely(xennet_set_skb_gso(skb, gso))) { | |
1009 | __skb_queue_head(&tmpq, skb); | |
1010 | np->rx.rsp_cons += skb_queue_len(&tmpq); | |
1011 | goto err; | |
1012 | } | |
1013 | } | |
1014 | ||
3683243b IC |
1015 | NETFRONT_SKB_CB(skb)->pull_to = rx->status; |
1016 | if (NETFRONT_SKB_CB(skb)->pull_to > RX_COPY_THRESHOLD) | |
1017 | NETFRONT_SKB_CB(skb)->pull_to = RX_COPY_THRESHOLD; | |
0d160211 | 1018 | |
3683243b IC |
1019 | skb_shinfo(skb)->frags[0].page_offset = rx->offset; |
1020 | skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status); | |
1021 | skb->data_len = rx->status; | |
0d160211 JF |
1022 | |
1023 | i = xennet_fill_frags(np, skb, &tmpq); | |
1024 | ||
1025 | /* | |
d9a58a78 IC |
1026 | * Truesize is the actual allocation size, even if the |
1027 | * allocation is only partially used. | |
1028 | */ | |
1029 | skb->truesize += PAGE_SIZE * skb_shinfo(skb)->nr_frags; | |
0d160211 JF |
1030 | skb->len += skb->data_len; |
1031 | ||
f942dc25 | 1032 | if (rx->flags & XEN_NETRXF_csum_blank) |
0d160211 | 1033 | skb->ip_summed = CHECKSUM_PARTIAL; |
f942dc25 | 1034 | else if (rx->flags & XEN_NETRXF_data_validated) |
0d160211 JF |
1035 | skb->ip_summed = CHECKSUM_UNNECESSARY; |
1036 | ||
1037 | __skb_queue_tail(&rxq, skb); | |
1038 | ||
1039 | np->rx.rsp_cons = ++i; | |
1040 | work_done++; | |
1041 | } | |
1042 | ||
56cfe5d0 | 1043 | __skb_queue_purge(&errq); |
0d160211 JF |
1044 | |
1045 | work_done -= handle_incoming_queue(dev, &rxq); | |
1046 | ||
1047 | /* If we get a callback with very few responses, reduce fill target. */ | |
1048 | /* NB. Note exponential increase, linear decrease. */ | |
1049 | if (((np->rx.req_prod_pvt - np->rx.sring->rsp_prod) > | |
1050 | ((3*np->rx_target) / 4)) && | |
1051 | (--np->rx_target < np->rx_min_target)) | |
1052 | np->rx_target = np->rx_min_target; | |
1053 | ||
1054 | xennet_alloc_rx_buffers(dev); | |
1055 | ||
0d160211 | 1056 | if (work_done < budget) { |
bea3348e SH |
1057 | int more_to_do = 0; |
1058 | ||
0d160211 JF |
1059 | local_irq_save(flags); |
1060 | ||
1061 | RING_FINAL_CHECK_FOR_RESPONSES(&np->rx, more_to_do); | |
1062 | if (!more_to_do) | |
288379f0 | 1063 | __napi_complete(napi); |
0d160211 JF |
1064 | |
1065 | local_irq_restore(flags); | |
1066 | } | |
1067 | ||
1068 | spin_unlock(&np->rx_lock); | |
1069 | ||
bea3348e | 1070 | return work_done; |
0d160211 JF |
1071 | } |
1072 | ||
1073 | static int xennet_change_mtu(struct net_device *dev, int mtu) | |
1074 | { | |
9ecd1a75 WL |
1075 | int max = xennet_can_sg(dev) ? |
1076 | XEN_NETIF_MAX_TX_SIZE - MAX_TCP_HEADER : ETH_DATA_LEN; | |
0d160211 JF |
1077 | |
1078 | if (mtu > max) | |
1079 | return -EINVAL; | |
1080 | dev->mtu = mtu; | |
1081 | return 0; | |
1082 | } | |
1083 | ||
e00f85be | 1084 | static struct rtnl_link_stats64 *xennet_get_stats64(struct net_device *dev, |
1085 | struct rtnl_link_stats64 *tot) | |
1086 | { | |
1087 | struct netfront_info *np = netdev_priv(dev); | |
1088 | int cpu; | |
1089 | ||
1090 | for_each_possible_cpu(cpu) { | |
1091 | struct netfront_stats *stats = per_cpu_ptr(np->stats, cpu); | |
1092 | u64 rx_packets, rx_bytes, tx_packets, tx_bytes; | |
1093 | unsigned int start; | |
1094 | ||
1095 | do { | |
1096 | start = u64_stats_fetch_begin_bh(&stats->syncp); | |
1097 | ||
1098 | rx_packets = stats->rx_packets; | |
1099 | tx_packets = stats->tx_packets; | |
1100 | rx_bytes = stats->rx_bytes; | |
1101 | tx_bytes = stats->tx_bytes; | |
1102 | } while (u64_stats_fetch_retry_bh(&stats->syncp, start)); | |
1103 | ||
1104 | tot->rx_packets += rx_packets; | |
1105 | tot->tx_packets += tx_packets; | |
1106 | tot->rx_bytes += rx_bytes; | |
1107 | tot->tx_bytes += tx_bytes; | |
1108 | } | |
1109 | ||
1110 | tot->rx_errors = dev->stats.rx_errors; | |
1111 | tot->tx_dropped = dev->stats.tx_dropped; | |
1112 | ||
1113 | return tot; | |
1114 | } | |
1115 | ||
0d160211 JF |
1116 | static void xennet_release_tx_bufs(struct netfront_info *np) |
1117 | { | |
1118 | struct sk_buff *skb; | |
1119 | int i; | |
1120 | ||
1121 | for (i = 0; i < NET_TX_RING_SIZE; i++) { | |
1122 | /* Skip over entries which are actually freelist references */ | |
1ffb40b8 | 1123 | if (skb_entry_is_link(&np->tx_skbs[i])) |
0d160211 JF |
1124 | continue; |
1125 | ||
1126 | skb = np->tx_skbs[i].skb; | |
1127 | gnttab_end_foreign_access_ref(np->grant_tx_ref[i], | |
1128 | GNTMAP_readonly); | |
1129 | gnttab_release_grant_reference(&np->gref_tx_head, | |
1130 | np->grant_tx_ref[i]); | |
1131 | np->grant_tx_ref[i] = GRANT_INVALID_REF; | |
1132 | add_id_to_freelist(&np->tx_skb_freelist, np->tx_skbs, i); | |
1133 | dev_kfree_skb_irq(skb); | |
1134 | } | |
1135 | } | |
1136 | ||
1137 | static void xennet_release_rx_bufs(struct netfront_info *np) | |
1138 | { | |
1139 | struct mmu_update *mmu = np->rx_mmu; | |
1140 | struct multicall_entry *mcl = np->rx_mcl; | |
1141 | struct sk_buff_head free_list; | |
1142 | struct sk_buff *skb; | |
1143 | unsigned long mfn; | |
1144 | int xfer = 0, noxfer = 0, unused = 0; | |
1145 | int id, ref; | |
1146 | ||
1147 | dev_warn(&np->netdev->dev, "%s: fix me for copying receiver.\n", | |
1148 | __func__); | |
1149 | return; | |
1150 | ||
1151 | skb_queue_head_init(&free_list); | |
1152 | ||
1153 | spin_lock_bh(&np->rx_lock); | |
1154 | ||
1155 | for (id = 0; id < NET_RX_RING_SIZE; id++) { | |
1156 | ref = np->grant_rx_ref[id]; | |
1157 | if (ref == GRANT_INVALID_REF) { | |
1158 | unused++; | |
1159 | continue; | |
1160 | } | |
1161 | ||
1162 | skb = np->rx_skbs[id]; | |
1163 | mfn = gnttab_end_foreign_transfer_ref(ref); | |
1164 | gnttab_release_grant_reference(&np->gref_rx_head, ref); | |
1165 | np->grant_rx_ref[id] = GRANT_INVALID_REF; | |
1166 | ||
1167 | if (0 == mfn) { | |
1168 | skb_shinfo(skb)->nr_frags = 0; | |
1169 | dev_kfree_skb(skb); | |
1170 | noxfer++; | |
1171 | continue; | |
1172 | } | |
1173 | ||
1174 | if (!xen_feature(XENFEAT_auto_translated_physmap)) { | |
1175 | /* Remap the page. */ | |
01c68026 IC |
1176 | const struct page *page = |
1177 | skb_frag_page(&skb_shinfo(skb)->frags[0]); | |
0d160211 JF |
1178 | unsigned long pfn = page_to_pfn(page); |
1179 | void *vaddr = page_address(page); | |
1180 | ||
1181 | MULTI_update_va_mapping(mcl, (unsigned long)vaddr, | |
1182 | mfn_pte(mfn, PAGE_KERNEL), | |
1183 | 0); | |
1184 | mcl++; | |
1185 | mmu->ptr = ((u64)mfn << PAGE_SHIFT) | |
1186 | | MMU_MACHPHYS_UPDATE; | |
1187 | mmu->val = pfn; | |
1188 | mmu++; | |
1189 | ||
1190 | set_phys_to_machine(pfn, mfn); | |
1191 | } | |
1192 | __skb_queue_tail(&free_list, skb); | |
1193 | xfer++; | |
1194 | } | |
1195 | ||
1196 | dev_info(&np->netdev->dev, "%s: %d xfer, %d noxfer, %d unused\n", | |
1197 | __func__, xfer, noxfer, unused); | |
1198 | ||
1199 | if (xfer) { | |
1200 | if (!xen_feature(XENFEAT_auto_translated_physmap)) { | |
1201 | /* Do all the remapping work and M2P updates. */ | |
1202 | MULTI_mmu_update(mcl, np->rx_mmu, mmu - np->rx_mmu, | |
79ea13ce | 1203 | NULL, DOMID_SELF); |
0d160211 JF |
1204 | mcl++; |
1205 | HYPERVISOR_multicall(np->rx_mcl, mcl - np->rx_mcl); | |
1206 | } | |
1207 | } | |
1208 | ||
56cfe5d0 | 1209 | __skb_queue_purge(&free_list); |
0d160211 JF |
1210 | |
1211 | spin_unlock_bh(&np->rx_lock); | |
1212 | } | |
1213 | ||
1214 | static void xennet_uninit(struct net_device *dev) | |
1215 | { | |
1216 | struct netfront_info *np = netdev_priv(dev); | |
1217 | xennet_release_tx_bufs(np); | |
1218 | xennet_release_rx_bufs(np); | |
1219 | gnttab_free_grant_references(np->gref_tx_head); | |
1220 | gnttab_free_grant_references(np->gref_rx_head); | |
1221 | } | |
1222 | ||
c8f44aff MM |
1223 | static netdev_features_t xennet_fix_features(struct net_device *dev, |
1224 | netdev_features_t features) | |
8f7b01a1 ED |
1225 | { |
1226 | struct netfront_info *np = netdev_priv(dev); | |
1227 | int val; | |
1228 | ||
1229 | if (features & NETIF_F_SG) { | |
1230 | if (xenbus_scanf(XBT_NIL, np->xbdev->otherend, "feature-sg", | |
1231 | "%d", &val) < 0) | |
1232 | val = 0; | |
1233 | ||
1234 | if (!val) | |
1235 | features &= ~NETIF_F_SG; | |
1236 | } | |
1237 | ||
1238 | if (features & NETIF_F_TSO) { | |
1239 | if (xenbus_scanf(XBT_NIL, np->xbdev->otherend, | |
1240 | "feature-gso-tcpv4", "%d", &val) < 0) | |
1241 | val = 0; | |
1242 | ||
1243 | if (!val) | |
1244 | features &= ~NETIF_F_TSO; | |
1245 | } | |
1246 | ||
1247 | return features; | |
1248 | } | |
1249 | ||
c8f44aff MM |
1250 | static int xennet_set_features(struct net_device *dev, |
1251 | netdev_features_t features) | |
8f7b01a1 ED |
1252 | { |
1253 | if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) { | |
1254 | netdev_info(dev, "Reducing MTU because no SG offload"); | |
1255 | dev->mtu = ETH_DATA_LEN; | |
1256 | } | |
1257 | ||
1258 | return 0; | |
1259 | } | |
1260 | ||
d634bf2c | 1261 | static irqreturn_t xennet_tx_interrupt(int irq, void *dev_id) |
cf66f9d4 | 1262 | { |
d634bf2c WL |
1263 | struct netfront_info *np = dev_id; |
1264 | struct net_device *dev = np->netdev; | |
cf66f9d4 KRW |
1265 | unsigned long flags; |
1266 | ||
1267 | spin_lock_irqsave(&np->tx_lock, flags); | |
d634bf2c WL |
1268 | xennet_tx_buf_gc(dev); |
1269 | spin_unlock_irqrestore(&np->tx_lock, flags); | |
cf66f9d4 | 1270 | |
d634bf2c WL |
1271 | return IRQ_HANDLED; |
1272 | } | |
1273 | ||
1274 | static irqreturn_t xennet_rx_interrupt(int irq, void *dev_id) | |
1275 | { | |
1276 | struct netfront_info *np = dev_id; | |
1277 | struct net_device *dev = np->netdev; | |
1278 | ||
1279 | if (likely(netif_carrier_ok(dev) && | |
1280 | RING_HAS_UNCONSUMED_RESPONSES(&np->rx))) | |
cf66f9d4 | 1281 | napi_schedule(&np->napi); |
cf66f9d4 | 1282 | |
d634bf2c WL |
1283 | return IRQ_HANDLED; |
1284 | } | |
cf66f9d4 | 1285 | |
d634bf2c WL |
1286 | static irqreturn_t xennet_interrupt(int irq, void *dev_id) |
1287 | { | |
1288 | xennet_tx_interrupt(irq, dev_id); | |
1289 | xennet_rx_interrupt(irq, dev_id); | |
cf66f9d4 KRW |
1290 | return IRQ_HANDLED; |
1291 | } | |
1292 | ||
1293 | #ifdef CONFIG_NET_POLL_CONTROLLER | |
1294 | static void xennet_poll_controller(struct net_device *dev) | |
1295 | { | |
1296 | xennet_interrupt(0, dev); | |
1297 | } | |
1298 | #endif | |
1299 | ||
0a0b9d2e SH |
1300 | static const struct net_device_ops xennet_netdev_ops = { |
1301 | .ndo_open = xennet_open, | |
1302 | .ndo_uninit = xennet_uninit, | |
1303 | .ndo_stop = xennet_close, | |
1304 | .ndo_start_xmit = xennet_start_xmit, | |
1305 | .ndo_change_mtu = xennet_change_mtu, | |
e00f85be | 1306 | .ndo_get_stats64 = xennet_get_stats64, |
0a0b9d2e SH |
1307 | .ndo_set_mac_address = eth_mac_addr, |
1308 | .ndo_validate_addr = eth_validate_addr, | |
fb507934 MM |
1309 | .ndo_fix_features = xennet_fix_features, |
1310 | .ndo_set_features = xennet_set_features, | |
cf66f9d4 KRW |
1311 | #ifdef CONFIG_NET_POLL_CONTROLLER |
1312 | .ndo_poll_controller = xennet_poll_controller, | |
1313 | #endif | |
0a0b9d2e SH |
1314 | }; |
1315 | ||
8e0e46bb | 1316 | static struct net_device *xennet_create_dev(struct xenbus_device *dev) |
0d160211 JF |
1317 | { |
1318 | int i, err; | |
1319 | struct net_device *netdev; | |
1320 | struct netfront_info *np; | |
1321 | ||
1322 | netdev = alloc_etherdev(sizeof(struct netfront_info)); | |
41de8d4c | 1323 | if (!netdev) |
0d160211 | 1324 | return ERR_PTR(-ENOMEM); |
0d160211 JF |
1325 | |
1326 | np = netdev_priv(netdev); | |
1327 | np->xbdev = dev; | |
1328 | ||
1329 | spin_lock_init(&np->tx_lock); | |
1330 | spin_lock_init(&np->rx_lock); | |
1331 | ||
1332 | skb_queue_head_init(&np->rx_batch); | |
1333 | np->rx_target = RX_DFL_MIN_TARGET; | |
1334 | np->rx_min_target = RX_DFL_MIN_TARGET; | |
1335 | np->rx_max_target = RX_MAX_TARGET; | |
1336 | ||
1337 | init_timer(&np->rx_refill_timer); | |
1338 | np->rx_refill_timer.data = (unsigned long)netdev; | |
1339 | np->rx_refill_timer.function = rx_refill_timeout; | |
1340 | ||
e00f85be | 1341 | err = -ENOMEM; |
1342 | np->stats = alloc_percpu(struct netfront_stats); | |
1343 | if (np->stats == NULL) | |
1344 | goto exit; | |
1345 | ||
0d160211 JF |
1346 | /* Initialise tx_skbs as a free chain containing every entry. */ |
1347 | np->tx_skb_freelist = 0; | |
1348 | for (i = 0; i < NET_TX_RING_SIZE; i++) { | |
1ffb40b8 | 1349 | skb_entry_set_link(&np->tx_skbs[i], i+1); |
0d160211 JF |
1350 | np->grant_tx_ref[i] = GRANT_INVALID_REF; |
1351 | } | |
1352 | ||
1353 | /* Clear out rx_skbs */ | |
1354 | for (i = 0; i < NET_RX_RING_SIZE; i++) { | |
1355 | np->rx_skbs[i] = NULL; | |
1356 | np->grant_rx_ref[i] = GRANT_INVALID_REF; | |
1357 | } | |
1358 | ||
1359 | /* A grant for every tx ring slot */ | |
1360 | if (gnttab_alloc_grant_references(TX_MAX_TARGET, | |
1361 | &np->gref_tx_head) < 0) { | |
1362 | printk(KERN_ALERT "#### netfront can't alloc tx grant refs\n"); | |
1363 | err = -ENOMEM; | |
e00f85be | 1364 | goto exit_free_stats; |
0d160211 JF |
1365 | } |
1366 | /* A grant for every rx ring slot */ | |
1367 | if (gnttab_alloc_grant_references(RX_MAX_TARGET, | |
1368 | &np->gref_rx_head) < 0) { | |
1369 | printk(KERN_ALERT "#### netfront can't alloc rx grant refs\n"); | |
1370 | err = -ENOMEM; | |
1371 | goto exit_free_tx; | |
1372 | } | |
1373 | ||
0a0b9d2e SH |
1374 | netdev->netdev_ops = &xennet_netdev_ops; |
1375 | ||
bea3348e | 1376 | netif_napi_add(netdev, &np->napi, xennet_poll, 64); |
fb507934 MM |
1377 | netdev->features = NETIF_F_IP_CSUM | NETIF_F_RXCSUM | |
1378 | NETIF_F_GSO_ROBUST; | |
1379 | netdev->hw_features = NETIF_F_IP_CSUM | NETIF_F_SG | NETIF_F_TSO; | |
0d160211 | 1380 | |
fc3e5941 IC |
1381 | /* |
1382 | * Assume that all hw features are available for now. This set | |
1383 | * will be adjusted by the call to netdev_update_features() in | |
1384 | * xennet_connect() which is the earliest point where we can | |
1385 | * negotiate with the backend regarding supported features. | |
1386 | */ | |
1387 | netdev->features |= netdev->hw_features; | |
1388 | ||
0d160211 | 1389 | SET_ETHTOOL_OPS(netdev, &xennet_ethtool_ops); |
0d160211 JF |
1390 | SET_NETDEV_DEV(netdev, &dev->dev); |
1391 | ||
9ecd1a75 WL |
1392 | netif_set_gso_max_size(netdev, XEN_NETIF_MAX_TX_SIZE - MAX_TCP_HEADER); |
1393 | ||
0d160211 JF |
1394 | np->netdev = netdev; |
1395 | ||
1396 | netif_carrier_off(netdev); | |
1397 | ||
1398 | return netdev; | |
1399 | ||
1400 | exit_free_tx: | |
1401 | gnttab_free_grant_references(np->gref_tx_head); | |
e00f85be | 1402 | exit_free_stats: |
1403 | free_percpu(np->stats); | |
0d160211 JF |
1404 | exit: |
1405 | free_netdev(netdev); | |
1406 | return ERR_PTR(err); | |
1407 | } | |
1408 | ||
1409 | /** | |
1410 | * Entry point to this code when a new device is created. Allocate the basic | |
1411 | * structures and the ring buffers for communication with the backend, and | |
1412 | * inform the backend of the appropriate details for those. | |
1413 | */ | |
8e0e46bb | 1414 | static int netfront_probe(struct xenbus_device *dev, |
1dd06ae8 | 1415 | const struct xenbus_device_id *id) |
0d160211 JF |
1416 | { |
1417 | int err; | |
1418 | struct net_device *netdev; | |
1419 | struct netfront_info *info; | |
1420 | ||
1421 | netdev = xennet_create_dev(dev); | |
1422 | if (IS_ERR(netdev)) { | |
1423 | err = PTR_ERR(netdev); | |
1424 | xenbus_dev_fatal(dev, err, "creating netdev"); | |
1425 | return err; | |
1426 | } | |
1427 | ||
1428 | info = netdev_priv(netdev); | |
1b713e00 | 1429 | dev_set_drvdata(&dev->dev, info); |
0d160211 JF |
1430 | |
1431 | err = register_netdev(info->netdev); | |
1432 | if (err) { | |
1433 | printk(KERN_WARNING "%s: register_netdev err=%d\n", | |
1434 | __func__, err); | |
1435 | goto fail; | |
1436 | } | |
1437 | ||
1438 | err = xennet_sysfs_addif(info->netdev); | |
1439 | if (err) { | |
1440 | unregister_netdev(info->netdev); | |
1441 | printk(KERN_WARNING "%s: add sysfs failed err=%d\n", | |
1442 | __func__, err); | |
1443 | goto fail; | |
1444 | } | |
1445 | ||
1446 | return 0; | |
1447 | ||
1448 | fail: | |
1449 | free_netdev(netdev); | |
1b713e00 | 1450 | dev_set_drvdata(&dev->dev, NULL); |
0d160211 JF |
1451 | return err; |
1452 | } | |
1453 | ||
1454 | static void xennet_end_access(int ref, void *page) | |
1455 | { | |
1456 | /* This frees the page as a side-effect */ | |
1457 | if (ref != GRANT_INVALID_REF) | |
1458 | gnttab_end_foreign_access(ref, 0, (unsigned long)page); | |
1459 | } | |
1460 | ||
1461 | static void xennet_disconnect_backend(struct netfront_info *info) | |
1462 | { | |
1463 | /* Stop old i/f to prevent errors whilst we rebuild the state. */ | |
1464 | spin_lock_bh(&info->rx_lock); | |
1465 | spin_lock_irq(&info->tx_lock); | |
1466 | netif_carrier_off(info->netdev); | |
1467 | spin_unlock_irq(&info->tx_lock); | |
1468 | spin_unlock_bh(&info->rx_lock); | |
1469 | ||
d634bf2c WL |
1470 | if (info->tx_irq && (info->tx_irq == info->rx_irq)) |
1471 | unbind_from_irqhandler(info->tx_irq, info); | |
1472 | if (info->tx_irq && (info->tx_irq != info->rx_irq)) { | |
1473 | unbind_from_irqhandler(info->tx_irq, info); | |
1474 | unbind_from_irqhandler(info->rx_irq, info); | |
1475 | } | |
1476 | info->tx_evtchn = info->rx_evtchn = 0; | |
1477 | info->tx_irq = info->rx_irq = 0; | |
0d160211 JF |
1478 | |
1479 | /* End access and free the pages */ | |
1480 | xennet_end_access(info->tx_ring_ref, info->tx.sring); | |
1481 | xennet_end_access(info->rx_ring_ref, info->rx.sring); | |
1482 | ||
1483 | info->tx_ring_ref = GRANT_INVALID_REF; | |
1484 | info->rx_ring_ref = GRANT_INVALID_REF; | |
1485 | info->tx.sring = NULL; | |
1486 | info->rx.sring = NULL; | |
1487 | } | |
1488 | ||
1489 | /** | |
1490 | * We are reconnecting to the backend, due to a suspend/resume, or a backend | |
1491 | * driver restart. We tear down our netif structure and recreate it, but | |
1492 | * leave the device-layer structures intact so that this is transparent to the | |
1493 | * rest of the kernel. | |
1494 | */ | |
1495 | static int netfront_resume(struct xenbus_device *dev) | |
1496 | { | |
1b713e00 | 1497 | struct netfront_info *info = dev_get_drvdata(&dev->dev); |
0d160211 JF |
1498 | |
1499 | dev_dbg(&dev->dev, "%s\n", dev->nodename); | |
1500 | ||
1501 | xennet_disconnect_backend(info); | |
1502 | return 0; | |
1503 | } | |
1504 | ||
1505 | static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[]) | |
1506 | { | |
1507 | char *s, *e, *macstr; | |
1508 | int i; | |
1509 | ||
1510 | macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL); | |
1511 | if (IS_ERR(macstr)) | |
1512 | return PTR_ERR(macstr); | |
1513 | ||
1514 | for (i = 0; i < ETH_ALEN; i++) { | |
1515 | mac[i] = simple_strtoul(s, &e, 16); | |
1516 | if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) { | |
1517 | kfree(macstr); | |
1518 | return -ENOENT; | |
1519 | } | |
1520 | s = e+1; | |
1521 | } | |
1522 | ||
1523 | kfree(macstr); | |
1524 | return 0; | |
1525 | } | |
1526 | ||
d634bf2c WL |
1527 | static int setup_netfront_single(struct netfront_info *info) |
1528 | { | |
1529 | int err; | |
1530 | ||
1531 | err = xenbus_alloc_evtchn(info->xbdev, &info->tx_evtchn); | |
1532 | if (err < 0) | |
1533 | goto fail; | |
1534 | ||
1535 | err = bind_evtchn_to_irqhandler(info->tx_evtchn, | |
1536 | xennet_interrupt, | |
1537 | 0, info->netdev->name, info); | |
1538 | if (err < 0) | |
1539 | goto bind_fail; | |
1540 | info->rx_evtchn = info->tx_evtchn; | |
1541 | info->rx_irq = info->tx_irq = err; | |
1542 | ||
1543 | return 0; | |
1544 | ||
1545 | bind_fail: | |
1546 | xenbus_free_evtchn(info->xbdev, info->tx_evtchn); | |
1547 | info->tx_evtchn = 0; | |
1548 | fail: | |
1549 | return err; | |
1550 | } | |
1551 | ||
1552 | static int setup_netfront_split(struct netfront_info *info) | |
1553 | { | |
1554 | int err; | |
1555 | ||
1556 | err = xenbus_alloc_evtchn(info->xbdev, &info->tx_evtchn); | |
1557 | if (err < 0) | |
1558 | goto fail; | |
1559 | err = xenbus_alloc_evtchn(info->xbdev, &info->rx_evtchn); | |
1560 | if (err < 0) | |
1561 | goto alloc_rx_evtchn_fail; | |
1562 | ||
1563 | snprintf(info->tx_irq_name, sizeof(info->tx_irq_name), | |
1564 | "%s-tx", info->netdev->name); | |
1565 | err = bind_evtchn_to_irqhandler(info->tx_evtchn, | |
1566 | xennet_tx_interrupt, | |
1567 | 0, info->tx_irq_name, info); | |
1568 | if (err < 0) | |
1569 | goto bind_tx_fail; | |
1570 | info->tx_irq = err; | |
1571 | ||
1572 | snprintf(info->rx_irq_name, sizeof(info->rx_irq_name), | |
1573 | "%s-rx", info->netdev->name); | |
1574 | err = bind_evtchn_to_irqhandler(info->rx_evtchn, | |
1575 | xennet_rx_interrupt, | |
1576 | 0, info->rx_irq_name, info); | |
1577 | if (err < 0) | |
1578 | goto bind_rx_fail; | |
1579 | info->rx_irq = err; | |
1580 | ||
1581 | return 0; | |
1582 | ||
1583 | bind_rx_fail: | |
1584 | unbind_from_irqhandler(info->tx_irq, info); | |
1585 | info->tx_irq = 0; | |
1586 | bind_tx_fail: | |
1587 | xenbus_free_evtchn(info->xbdev, info->rx_evtchn); | |
1588 | info->rx_evtchn = 0; | |
1589 | alloc_rx_evtchn_fail: | |
1590 | xenbus_free_evtchn(info->xbdev, info->tx_evtchn); | |
1591 | info->tx_evtchn = 0; | |
1592 | fail: | |
1593 | return err; | |
1594 | } | |
1595 | ||
0d160211 JF |
1596 | static int setup_netfront(struct xenbus_device *dev, struct netfront_info *info) |
1597 | { | |
1598 | struct xen_netif_tx_sring *txs; | |
1599 | struct xen_netif_rx_sring *rxs; | |
1600 | int err; | |
1601 | struct net_device *netdev = info->netdev; | |
d634bf2c | 1602 | unsigned int feature_split_evtchn; |
0d160211 JF |
1603 | |
1604 | info->tx_ring_ref = GRANT_INVALID_REF; | |
1605 | info->rx_ring_ref = GRANT_INVALID_REF; | |
1606 | info->rx.sring = NULL; | |
1607 | info->tx.sring = NULL; | |
1608 | netdev->irq = 0; | |
1609 | ||
d634bf2c WL |
1610 | err = xenbus_scanf(XBT_NIL, info->xbdev->otherend, |
1611 | "feature-split-event-channels", "%u", | |
1612 | &feature_split_evtchn); | |
1613 | if (err < 0) | |
1614 | feature_split_evtchn = 0; | |
1615 | ||
0d160211 JF |
1616 | err = xen_net_read_mac(dev, netdev->dev_addr); |
1617 | if (err) { | |
1618 | xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename); | |
1619 | goto fail; | |
1620 | } | |
1621 | ||
a144ff09 | 1622 | txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH); |
0d160211 JF |
1623 | if (!txs) { |
1624 | err = -ENOMEM; | |
1625 | xenbus_dev_fatal(dev, err, "allocating tx ring page"); | |
1626 | goto fail; | |
1627 | } | |
1628 | SHARED_RING_INIT(txs); | |
1629 | FRONT_RING_INIT(&info->tx, txs, PAGE_SIZE); | |
1630 | ||
1631 | err = xenbus_grant_ring(dev, virt_to_mfn(txs)); | |
1ca2983a WL |
1632 | if (err < 0) |
1633 | goto grant_tx_ring_fail; | |
0d160211 JF |
1634 | |
1635 | info->tx_ring_ref = err; | |
a144ff09 | 1636 | rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH); |
0d160211 JF |
1637 | if (!rxs) { |
1638 | err = -ENOMEM; | |
1639 | xenbus_dev_fatal(dev, err, "allocating rx ring page"); | |
1ca2983a | 1640 | goto alloc_rx_ring_fail; |
0d160211 JF |
1641 | } |
1642 | SHARED_RING_INIT(rxs); | |
1643 | FRONT_RING_INIT(&info->rx, rxs, PAGE_SIZE); | |
1644 | ||
1645 | err = xenbus_grant_ring(dev, virt_to_mfn(rxs)); | |
1ca2983a WL |
1646 | if (err < 0) |
1647 | goto grant_rx_ring_fail; | |
0d160211 JF |
1648 | info->rx_ring_ref = err; |
1649 | ||
d634bf2c WL |
1650 | if (feature_split_evtchn) |
1651 | err = setup_netfront_split(info); | |
1652 | /* setup single event channel if | |
1653 | * a) feature-split-event-channels == 0 | |
1654 | * b) feature-split-event-channels == 1 but failed to setup | |
1655 | */ | |
1656 | if (!feature_split_evtchn || (feature_split_evtchn && err)) | |
1657 | err = setup_netfront_single(info); | |
1658 | ||
0d160211 | 1659 | if (err) |
1ca2983a | 1660 | goto alloc_evtchn_fail; |
0d160211 | 1661 | |
0d160211 JF |
1662 | return 0; |
1663 | ||
1ca2983a WL |
1664 | /* If we fail to setup netfront, it is safe to just revoke access to |
1665 | * granted pages because backend is not accessing it at this point. | |
1666 | */ | |
1ca2983a WL |
1667 | alloc_evtchn_fail: |
1668 | gnttab_end_foreign_access_ref(info->rx_ring_ref, 0); | |
1669 | grant_rx_ring_fail: | |
1670 | free_page((unsigned long)rxs); | |
1671 | alloc_rx_ring_fail: | |
1672 | gnttab_end_foreign_access_ref(info->tx_ring_ref, 0); | |
1673 | grant_tx_ring_fail: | |
1674 | free_page((unsigned long)txs); | |
1675 | fail: | |
0d160211 JF |
1676 | return err; |
1677 | } | |
1678 | ||
1679 | /* Common code used when first setting up, and when resuming. */ | |
f502bf2b | 1680 | static int talk_to_netback(struct xenbus_device *dev, |
0d160211 JF |
1681 | struct netfront_info *info) |
1682 | { | |
1683 | const char *message; | |
1684 | struct xenbus_transaction xbt; | |
1685 | int err; | |
1686 | ||
1687 | /* Create shared ring, alloc event channel. */ | |
1688 | err = setup_netfront(dev, info); | |
1689 | if (err) | |
1690 | goto out; | |
1691 | ||
1692 | again: | |
1693 | err = xenbus_transaction_start(&xbt); | |
1694 | if (err) { | |
1695 | xenbus_dev_fatal(dev, err, "starting transaction"); | |
1696 | goto destroy_ring; | |
1697 | } | |
1698 | ||
1699 | err = xenbus_printf(xbt, dev->nodename, "tx-ring-ref", "%u", | |
1700 | info->tx_ring_ref); | |
1701 | if (err) { | |
1702 | message = "writing tx ring-ref"; | |
1703 | goto abort_transaction; | |
1704 | } | |
1705 | err = xenbus_printf(xbt, dev->nodename, "rx-ring-ref", "%u", | |
1706 | info->rx_ring_ref); | |
1707 | if (err) { | |
1708 | message = "writing rx ring-ref"; | |
1709 | goto abort_transaction; | |
1710 | } | |
d634bf2c WL |
1711 | |
1712 | if (info->tx_evtchn == info->rx_evtchn) { | |
1713 | err = xenbus_printf(xbt, dev->nodename, | |
1714 | "event-channel", "%u", info->tx_evtchn); | |
1715 | if (err) { | |
1716 | message = "writing event-channel"; | |
1717 | goto abort_transaction; | |
1718 | } | |
1719 | } else { | |
1720 | err = xenbus_printf(xbt, dev->nodename, | |
1721 | "event-channel-tx", "%u", info->tx_evtchn); | |
1722 | if (err) { | |
1723 | message = "writing event-channel-tx"; | |
1724 | goto abort_transaction; | |
1725 | } | |
1726 | err = xenbus_printf(xbt, dev->nodename, | |
1727 | "event-channel-rx", "%u", info->rx_evtchn); | |
1728 | if (err) { | |
1729 | message = "writing event-channel-rx"; | |
1730 | goto abort_transaction; | |
1731 | } | |
0d160211 JF |
1732 | } |
1733 | ||
1734 | err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u", | |
1735 | 1); | |
1736 | if (err) { | |
1737 | message = "writing request-rx-copy"; | |
1738 | goto abort_transaction; | |
1739 | } | |
1740 | ||
1741 | err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1); | |
1742 | if (err) { | |
1743 | message = "writing feature-rx-notify"; | |
1744 | goto abort_transaction; | |
1745 | } | |
1746 | ||
1747 | err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1); | |
1748 | if (err) { | |
1749 | message = "writing feature-sg"; | |
1750 | goto abort_transaction; | |
1751 | } | |
1752 | ||
1753 | err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1); | |
1754 | if (err) { | |
1755 | message = "writing feature-gso-tcpv4"; | |
1756 | goto abort_transaction; | |
1757 | } | |
1758 | ||
1759 | err = xenbus_transaction_end(xbt, 0); | |
1760 | if (err) { | |
1761 | if (err == -EAGAIN) | |
1762 | goto again; | |
1763 | xenbus_dev_fatal(dev, err, "completing transaction"); | |
1764 | goto destroy_ring; | |
1765 | } | |
1766 | ||
1767 | return 0; | |
1768 | ||
1769 | abort_transaction: | |
1770 | xenbus_transaction_end(xbt, 1); | |
1771 | xenbus_dev_fatal(dev, err, "%s", message); | |
1772 | destroy_ring: | |
1773 | xennet_disconnect_backend(info); | |
1774 | out: | |
1775 | return err; | |
1776 | } | |
1777 | ||
0d160211 JF |
1778 | static int xennet_connect(struct net_device *dev) |
1779 | { | |
1780 | struct netfront_info *np = netdev_priv(dev); | |
1781 | int i, requeue_idx, err; | |
1782 | struct sk_buff *skb; | |
1783 | grant_ref_t ref; | |
1784 | struct xen_netif_rx_request *req; | |
1785 | unsigned int feature_rx_copy; | |
1786 | ||
1787 | err = xenbus_scanf(XBT_NIL, np->xbdev->otherend, | |
1788 | "feature-rx-copy", "%u", &feature_rx_copy); | |
1789 | if (err != 1) | |
1790 | feature_rx_copy = 0; | |
1791 | ||
1792 | if (!feature_rx_copy) { | |
1793 | dev_info(&dev->dev, | |
898eb71c | 1794 | "backend does not support copying receive path\n"); |
0d160211 JF |
1795 | return -ENODEV; |
1796 | } | |
1797 | ||
f502bf2b | 1798 | err = talk_to_netback(np->xbdev, np); |
0d160211 JF |
1799 | if (err) |
1800 | return err; | |
1801 | ||
1ba37c51 | 1802 | rtnl_lock(); |
fb507934 | 1803 | netdev_update_features(dev); |
1ba37c51 | 1804 | rtnl_unlock(); |
0d160211 JF |
1805 | |
1806 | spin_lock_bh(&np->rx_lock); | |
1807 | spin_lock_irq(&np->tx_lock); | |
1808 | ||
1809 | /* Step 1: Discard all pending TX packet fragments. */ | |
1810 | xennet_release_tx_bufs(np); | |
1811 | ||
1812 | /* Step 2: Rebuild the RX buffer freelist and the RX ring itself. */ | |
1813 | for (requeue_idx = 0, i = 0; i < NET_RX_RING_SIZE; i++) { | |
01c68026 IC |
1814 | skb_frag_t *frag; |
1815 | const struct page *page; | |
0d160211 JF |
1816 | if (!np->rx_skbs[i]) |
1817 | continue; | |
1818 | ||
1819 | skb = np->rx_skbs[requeue_idx] = xennet_get_rx_skb(np, i); | |
1820 | ref = np->grant_rx_ref[requeue_idx] = xennet_get_rx_ref(np, i); | |
1821 | req = RING_GET_REQUEST(&np->rx, requeue_idx); | |
1822 | ||
01c68026 IC |
1823 | frag = &skb_shinfo(skb)->frags[0]; |
1824 | page = skb_frag_page(frag); | |
0d160211 JF |
1825 | gnttab_grant_foreign_access_ref( |
1826 | ref, np->xbdev->otherend_id, | |
01c68026 | 1827 | pfn_to_mfn(page_to_pfn(page)), |
0d160211 JF |
1828 | 0); |
1829 | req->gref = ref; | |
1830 | req->id = requeue_idx; | |
1831 | ||
1832 | requeue_idx++; | |
1833 | } | |
1834 | ||
1835 | np->rx.req_prod_pvt = requeue_idx; | |
1836 | ||
1837 | /* | |
1838 | * Step 3: All public and private state should now be sane. Get | |
1839 | * ready to start sending and receiving packets and give the driver | |
1840 | * domain a kick because we've probably just requeued some | |
1841 | * packets. | |
1842 | */ | |
1843 | netif_carrier_on(np->netdev); | |
d634bf2c WL |
1844 | notify_remote_via_irq(np->tx_irq); |
1845 | if (np->tx_irq != np->rx_irq) | |
1846 | notify_remote_via_irq(np->rx_irq); | |
0d160211 JF |
1847 | xennet_tx_buf_gc(dev); |
1848 | xennet_alloc_rx_buffers(dev); | |
1849 | ||
1850 | spin_unlock_irq(&np->tx_lock); | |
1851 | spin_unlock_bh(&np->rx_lock); | |
1852 | ||
1853 | return 0; | |
1854 | } | |
1855 | ||
1856 | /** | |
1857 | * Callback received when the backend's state changes. | |
1858 | */ | |
f502bf2b | 1859 | static void netback_changed(struct xenbus_device *dev, |
0d160211 JF |
1860 | enum xenbus_state backend_state) |
1861 | { | |
1b713e00 | 1862 | struct netfront_info *np = dev_get_drvdata(&dev->dev); |
0d160211 JF |
1863 | struct net_device *netdev = np->netdev; |
1864 | ||
1865 | dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state)); | |
1866 | ||
1867 | switch (backend_state) { | |
1868 | case XenbusStateInitialising: | |
1869 | case XenbusStateInitialised: | |
b78c9512 NI |
1870 | case XenbusStateReconfiguring: |
1871 | case XenbusStateReconfigured: | |
0d160211 JF |
1872 | case XenbusStateUnknown: |
1873 | case XenbusStateClosed: | |
1874 | break; | |
1875 | ||
1876 | case XenbusStateInitWait: | |
1877 | if (dev->state != XenbusStateInitialising) | |
1878 | break; | |
1879 | if (xennet_connect(netdev) != 0) | |
1880 | break; | |
1881 | xenbus_switch_state(dev, XenbusStateConnected); | |
08e34eb1 LE |
1882 | break; |
1883 | ||
1884 | case XenbusStateConnected: | |
ee89bab1 | 1885 | netdev_notify_peers(netdev); |
0d160211 JF |
1886 | break; |
1887 | ||
1888 | case XenbusStateClosing: | |
1889 | xenbus_frontend_closed(dev); | |
1890 | break; | |
1891 | } | |
1892 | } | |
1893 | ||
e0ce4af9 IC |
1894 | static const struct xennet_stat { |
1895 | char name[ETH_GSTRING_LEN]; | |
1896 | u16 offset; | |
1897 | } xennet_stats[] = { | |
1898 | { | |
1899 | "rx_gso_checksum_fixup", | |
1900 | offsetof(struct netfront_info, rx_gso_checksum_fixup) | |
1901 | }, | |
1902 | }; | |
1903 | ||
1904 | static int xennet_get_sset_count(struct net_device *dev, int string_set) | |
1905 | { | |
1906 | switch (string_set) { | |
1907 | case ETH_SS_STATS: | |
1908 | return ARRAY_SIZE(xennet_stats); | |
1909 | default: | |
1910 | return -EINVAL; | |
1911 | } | |
1912 | } | |
1913 | ||
1914 | static void xennet_get_ethtool_stats(struct net_device *dev, | |
1915 | struct ethtool_stats *stats, u64 * data) | |
1916 | { | |
1917 | void *np = netdev_priv(dev); | |
1918 | int i; | |
1919 | ||
1920 | for (i = 0; i < ARRAY_SIZE(xennet_stats); i++) | |
48f26d51 | 1921 | data[i] = *(unsigned long *)(np + xennet_stats[i].offset); |
e0ce4af9 IC |
1922 | } |
1923 | ||
1924 | static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data) | |
1925 | { | |
1926 | int i; | |
1927 | ||
1928 | switch (stringset) { | |
1929 | case ETH_SS_STATS: | |
1930 | for (i = 0; i < ARRAY_SIZE(xennet_stats); i++) | |
1931 | memcpy(data + i * ETH_GSTRING_LEN, | |
1932 | xennet_stats[i].name, ETH_GSTRING_LEN); | |
1933 | break; | |
1934 | } | |
1935 | } | |
1936 | ||
0fc0b732 | 1937 | static const struct ethtool_ops xennet_ethtool_ops = |
0d160211 | 1938 | { |
0d160211 | 1939 | .get_link = ethtool_op_get_link, |
e0ce4af9 IC |
1940 | |
1941 | .get_sset_count = xennet_get_sset_count, | |
1942 | .get_ethtool_stats = xennet_get_ethtool_stats, | |
1943 | .get_strings = xennet_get_strings, | |
0d160211 JF |
1944 | }; |
1945 | ||
1946 | #ifdef CONFIG_SYSFS | |
1947 | static ssize_t show_rxbuf_min(struct device *dev, | |
1948 | struct device_attribute *attr, char *buf) | |
1949 | { | |
1950 | struct net_device *netdev = to_net_dev(dev); | |
1951 | struct netfront_info *info = netdev_priv(netdev); | |
1952 | ||
1953 | return sprintf(buf, "%u\n", info->rx_min_target); | |
1954 | } | |
1955 | ||
1956 | static ssize_t store_rxbuf_min(struct device *dev, | |
1957 | struct device_attribute *attr, | |
1958 | const char *buf, size_t len) | |
1959 | { | |
1960 | struct net_device *netdev = to_net_dev(dev); | |
1961 | struct netfront_info *np = netdev_priv(netdev); | |
1962 | char *endp; | |
1963 | unsigned long target; | |
1964 | ||
1965 | if (!capable(CAP_NET_ADMIN)) | |
1966 | return -EPERM; | |
1967 | ||
1968 | target = simple_strtoul(buf, &endp, 0); | |
1969 | if (endp == buf) | |
1970 | return -EBADMSG; | |
1971 | ||
1972 | if (target < RX_MIN_TARGET) | |
1973 | target = RX_MIN_TARGET; | |
1974 | if (target > RX_MAX_TARGET) | |
1975 | target = RX_MAX_TARGET; | |
1976 | ||
1977 | spin_lock_bh(&np->rx_lock); | |
1978 | if (target > np->rx_max_target) | |
1979 | np->rx_max_target = target; | |
1980 | np->rx_min_target = target; | |
1981 | if (target > np->rx_target) | |
1982 | np->rx_target = target; | |
1983 | ||
1984 | xennet_alloc_rx_buffers(netdev); | |
1985 | ||
1986 | spin_unlock_bh(&np->rx_lock); | |
1987 | return len; | |
1988 | } | |
1989 | ||
1990 | static ssize_t show_rxbuf_max(struct device *dev, | |
1991 | struct device_attribute *attr, char *buf) | |
1992 | { | |
1993 | struct net_device *netdev = to_net_dev(dev); | |
1994 | struct netfront_info *info = netdev_priv(netdev); | |
1995 | ||
1996 | return sprintf(buf, "%u\n", info->rx_max_target); | |
1997 | } | |
1998 | ||
1999 | static ssize_t store_rxbuf_max(struct device *dev, | |
2000 | struct device_attribute *attr, | |
2001 | const char *buf, size_t len) | |
2002 | { | |
2003 | struct net_device *netdev = to_net_dev(dev); | |
2004 | struct netfront_info *np = netdev_priv(netdev); | |
2005 | char *endp; | |
2006 | unsigned long target; | |
2007 | ||
2008 | if (!capable(CAP_NET_ADMIN)) | |
2009 | return -EPERM; | |
2010 | ||
2011 | target = simple_strtoul(buf, &endp, 0); | |
2012 | if (endp == buf) | |
2013 | return -EBADMSG; | |
2014 | ||
2015 | if (target < RX_MIN_TARGET) | |
2016 | target = RX_MIN_TARGET; | |
2017 | if (target > RX_MAX_TARGET) | |
2018 | target = RX_MAX_TARGET; | |
2019 | ||
2020 | spin_lock_bh(&np->rx_lock); | |
2021 | if (target < np->rx_min_target) | |
2022 | np->rx_min_target = target; | |
2023 | np->rx_max_target = target; | |
2024 | if (target < np->rx_target) | |
2025 | np->rx_target = target; | |
2026 | ||
2027 | xennet_alloc_rx_buffers(netdev); | |
2028 | ||
2029 | spin_unlock_bh(&np->rx_lock); | |
2030 | return len; | |
2031 | } | |
2032 | ||
2033 | static ssize_t show_rxbuf_cur(struct device *dev, | |
2034 | struct device_attribute *attr, char *buf) | |
2035 | { | |
2036 | struct net_device *netdev = to_net_dev(dev); | |
2037 | struct netfront_info *info = netdev_priv(netdev); | |
2038 | ||
2039 | return sprintf(buf, "%u\n", info->rx_target); | |
2040 | } | |
2041 | ||
2042 | static struct device_attribute xennet_attrs[] = { | |
2043 | __ATTR(rxbuf_min, S_IRUGO|S_IWUSR, show_rxbuf_min, store_rxbuf_min), | |
2044 | __ATTR(rxbuf_max, S_IRUGO|S_IWUSR, show_rxbuf_max, store_rxbuf_max), | |
2045 | __ATTR(rxbuf_cur, S_IRUGO, show_rxbuf_cur, NULL), | |
2046 | }; | |
2047 | ||
2048 | static int xennet_sysfs_addif(struct net_device *netdev) | |
2049 | { | |
2050 | int i; | |
2051 | int err; | |
2052 | ||
2053 | for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++) { | |
2054 | err = device_create_file(&netdev->dev, | |
2055 | &xennet_attrs[i]); | |
2056 | if (err) | |
2057 | goto fail; | |
2058 | } | |
2059 | return 0; | |
2060 | ||
2061 | fail: | |
2062 | while (--i >= 0) | |
2063 | device_remove_file(&netdev->dev, &xennet_attrs[i]); | |
2064 | return err; | |
2065 | } | |
2066 | ||
2067 | static void xennet_sysfs_delif(struct net_device *netdev) | |
2068 | { | |
2069 | int i; | |
2070 | ||
2071 | for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++) | |
2072 | device_remove_file(&netdev->dev, &xennet_attrs[i]); | |
2073 | } | |
2074 | ||
2075 | #endif /* CONFIG_SYSFS */ | |
2076 | ||
73db144b | 2077 | static const struct xenbus_device_id netfront_ids[] = { |
0d160211 JF |
2078 | { "vif" }, |
2079 | { "" } | |
2080 | }; | |
2081 | ||
2082 | ||
8e0e46bb | 2083 | static int xennet_remove(struct xenbus_device *dev) |
0d160211 | 2084 | { |
1b713e00 | 2085 | struct netfront_info *info = dev_get_drvdata(&dev->dev); |
0d160211 JF |
2086 | |
2087 | dev_dbg(&dev->dev, "%s\n", dev->nodename); | |
2088 | ||
0d160211 JF |
2089 | xennet_disconnect_backend(info); |
2090 | ||
0d160211 JF |
2091 | xennet_sysfs_delif(info->netdev); |
2092 | ||
6bc96d04 IC |
2093 | unregister_netdev(info->netdev); |
2094 | ||
2095 | del_timer_sync(&info->rx_refill_timer); | |
2096 | ||
e00f85be | 2097 | free_percpu(info->stats); |
2098 | ||
0d160211 JF |
2099 | free_netdev(info->netdev); |
2100 | ||
2101 | return 0; | |
2102 | } | |
2103 | ||
73db144b | 2104 | static DEFINE_XENBUS_DRIVER(netfront, , |
0d160211 | 2105 | .probe = netfront_probe, |
8e0e46bb | 2106 | .remove = xennet_remove, |
0d160211 | 2107 | .resume = netfront_resume, |
f502bf2b | 2108 | .otherend_changed = netback_changed, |
73db144b | 2109 | ); |
0d160211 JF |
2110 | |
2111 | static int __init netif_init(void) | |
2112 | { | |
6e833587 | 2113 | if (!xen_domain()) |
0d160211 JF |
2114 | return -ENODEV; |
2115 | ||
e95ae5a4 | 2116 | if (xen_hvm_domain() && !xen_platform_pci_unplug) |
b9136d20 IM |
2117 | return -ENODEV; |
2118 | ||
0d160211 JF |
2119 | printk(KERN_INFO "Initialising Xen virtual ethernet driver.\n"); |
2120 | ||
ffb78a26 | 2121 | return xenbus_register_frontend(&netfront_driver); |
0d160211 JF |
2122 | } |
2123 | module_init(netif_init); | |
2124 | ||
2125 | ||
2126 | static void __exit netif_exit(void) | |
2127 | { | |
ffb78a26 | 2128 | xenbus_unregister_driver(&netfront_driver); |
0d160211 JF |
2129 | } |
2130 | module_exit(netif_exit); | |
2131 | ||
2132 | MODULE_DESCRIPTION("Xen virtual network device frontend"); | |
2133 | MODULE_LICENSE("GPL"); | |
d2f0c52b | 2134 | MODULE_ALIAS("xen:vif"); |
4f93f09b | 2135 | MODULE_ALIAS("xennet"); |