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> | |
39 | #include <linux/tcp.h> | |
40 | #include <linux/udp.h> | |
41 | #include <linux/moduleparam.h> | |
42 | #include <linux/mm.h> | |
43 | #include <net/ip.h> | |
44 | ||
45 | #include <xen/xenbus.h> | |
46 | #include <xen/events.h> | |
47 | #include <xen/page.h> | |
48 | #include <xen/grant_table.h> | |
49 | ||
50 | #include <xen/interface/io/netif.h> | |
51 | #include <xen/interface/memory.h> | |
52 | #include <xen/interface/grant_table.h> | |
53 | ||
54 | static struct ethtool_ops xennet_ethtool_ops; | |
55 | ||
56 | struct netfront_cb { | |
57 | struct page *page; | |
58 | unsigned offset; | |
59 | }; | |
60 | ||
61 | #define NETFRONT_SKB_CB(skb) ((struct netfront_cb *)((skb)->cb)) | |
62 | ||
63 | #define RX_COPY_THRESHOLD 256 | |
64 | ||
65 | #define GRANT_INVALID_REF 0 | |
66 | ||
67 | #define NET_TX_RING_SIZE __RING_SIZE((struct xen_netif_tx_sring *)0, PAGE_SIZE) | |
68 | #define NET_RX_RING_SIZE __RING_SIZE((struct xen_netif_rx_sring *)0, PAGE_SIZE) | |
69 | #define TX_MAX_TARGET min_t(int, NET_RX_RING_SIZE, 256) | |
70 | ||
71 | struct netfront_info { | |
72 | struct list_head list; | |
73 | struct net_device *netdev; | |
74 | ||
bea3348e | 75 | struct napi_struct napi; |
0d160211 | 76 | |
0d160211 | 77 | unsigned int evtchn; |
84284d3c | 78 | struct xenbus_device *xbdev; |
0d160211 | 79 | |
84284d3c JF |
80 | spinlock_t tx_lock; |
81 | struct xen_netif_tx_front_ring tx; | |
82 | int tx_ring_ref; | |
0d160211 JF |
83 | |
84 | /* | |
85 | * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries | |
86 | * are linked from tx_skb_freelist through skb_entry.link. | |
87 | * | |
88 | * NB. Freelist index entries are always going to be less than | |
89 | * PAGE_OFFSET, whereas pointers to skbs will always be equal or | |
90 | * greater than PAGE_OFFSET: we use this property to distinguish | |
91 | * them. | |
92 | */ | |
93 | union skb_entry { | |
94 | struct sk_buff *skb; | |
95 | unsigned link; | |
96 | } tx_skbs[NET_TX_RING_SIZE]; | |
97 | grant_ref_t gref_tx_head; | |
98 | grant_ref_t grant_tx_ref[NET_TX_RING_SIZE]; | |
99 | unsigned tx_skb_freelist; | |
100 | ||
84284d3c JF |
101 | spinlock_t rx_lock ____cacheline_aligned_in_smp; |
102 | struct xen_netif_rx_front_ring rx; | |
103 | int rx_ring_ref; | |
104 | ||
105 | /* Receive-ring batched refills. */ | |
106 | #define RX_MIN_TARGET 8 | |
107 | #define RX_DFL_MIN_TARGET 64 | |
108 | #define RX_MAX_TARGET min_t(int, NET_RX_RING_SIZE, 256) | |
109 | unsigned rx_min_target, rx_max_target, rx_target; | |
110 | struct sk_buff_head rx_batch; | |
111 | ||
112 | struct timer_list rx_refill_timer; | |
113 | ||
0d160211 JF |
114 | struct sk_buff *rx_skbs[NET_RX_RING_SIZE]; |
115 | grant_ref_t gref_rx_head; | |
116 | grant_ref_t grant_rx_ref[NET_RX_RING_SIZE]; | |
117 | ||
0d160211 JF |
118 | unsigned long rx_pfn_array[NET_RX_RING_SIZE]; |
119 | struct multicall_entry rx_mcl[NET_RX_RING_SIZE+1]; | |
120 | struct mmu_update rx_mmu[NET_RX_RING_SIZE]; | |
121 | }; | |
122 | ||
123 | struct netfront_rx_info { | |
124 | struct xen_netif_rx_response rx; | |
125 | struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1]; | |
126 | }; | |
127 | ||
128 | /* | |
129 | * Access macros for acquiring freeing slots in tx_skbs[]. | |
130 | */ | |
131 | ||
132 | static void add_id_to_freelist(unsigned *head, union skb_entry *list, | |
133 | unsigned short id) | |
134 | { | |
135 | list[id].link = *head; | |
136 | *head = id; | |
137 | } | |
138 | ||
139 | static unsigned short get_id_from_freelist(unsigned *head, | |
140 | union skb_entry *list) | |
141 | { | |
142 | unsigned int id = *head; | |
143 | *head = list[id].link; | |
144 | return id; | |
145 | } | |
146 | ||
147 | static int xennet_rxidx(RING_IDX idx) | |
148 | { | |
149 | return idx & (NET_RX_RING_SIZE - 1); | |
150 | } | |
151 | ||
152 | static struct sk_buff *xennet_get_rx_skb(struct netfront_info *np, | |
153 | RING_IDX ri) | |
154 | { | |
155 | int i = xennet_rxidx(ri); | |
156 | struct sk_buff *skb = np->rx_skbs[i]; | |
157 | np->rx_skbs[i] = NULL; | |
158 | return skb; | |
159 | } | |
160 | ||
161 | static grant_ref_t xennet_get_rx_ref(struct netfront_info *np, | |
162 | RING_IDX ri) | |
163 | { | |
164 | int i = xennet_rxidx(ri); | |
165 | grant_ref_t ref = np->grant_rx_ref[i]; | |
166 | np->grant_rx_ref[i] = GRANT_INVALID_REF; | |
167 | return ref; | |
168 | } | |
169 | ||
170 | #ifdef CONFIG_SYSFS | |
171 | static int xennet_sysfs_addif(struct net_device *netdev); | |
172 | static void xennet_sysfs_delif(struct net_device *netdev); | |
173 | #else /* !CONFIG_SYSFS */ | |
174 | #define xennet_sysfs_addif(dev) (0) | |
175 | #define xennet_sysfs_delif(dev) do { } while (0) | |
176 | #endif | |
177 | ||
178 | static int xennet_can_sg(struct net_device *dev) | |
179 | { | |
180 | return dev->features & NETIF_F_SG; | |
181 | } | |
182 | ||
183 | ||
184 | static void rx_refill_timeout(unsigned long data) | |
185 | { | |
186 | struct net_device *dev = (struct net_device *)data; | |
bea3348e SH |
187 | struct netfront_info *np = netdev_priv(dev); |
188 | netif_rx_schedule(dev, &np->napi); | |
0d160211 JF |
189 | } |
190 | ||
191 | static int netfront_tx_slot_available(struct netfront_info *np) | |
192 | { | |
193 | return ((np->tx.req_prod_pvt - np->tx.rsp_cons) < | |
194 | (TX_MAX_TARGET - MAX_SKB_FRAGS - 2)); | |
195 | } | |
196 | ||
197 | static void xennet_maybe_wake_tx(struct net_device *dev) | |
198 | { | |
199 | struct netfront_info *np = netdev_priv(dev); | |
200 | ||
201 | if (unlikely(netif_queue_stopped(dev)) && | |
202 | netfront_tx_slot_available(np) && | |
203 | likely(netif_running(dev))) | |
204 | netif_wake_queue(dev); | |
205 | } | |
206 | ||
207 | static void xennet_alloc_rx_buffers(struct net_device *dev) | |
208 | { | |
209 | unsigned short id; | |
210 | struct netfront_info *np = netdev_priv(dev); | |
211 | struct sk_buff *skb; | |
212 | struct page *page; | |
213 | int i, batch_target, notify; | |
214 | RING_IDX req_prod = np->rx.req_prod_pvt; | |
0d160211 JF |
215 | grant_ref_t ref; |
216 | unsigned long pfn; | |
217 | void *vaddr; | |
0d160211 JF |
218 | struct xen_netif_rx_request *req; |
219 | ||
220 | if (unlikely(!netif_carrier_ok(dev))) | |
221 | return; | |
222 | ||
223 | /* | |
224 | * Allocate skbuffs greedily, even though we batch updates to the | |
225 | * receive ring. This creates a less bursty demand on the memory | |
226 | * allocator, so should reduce the chance of failed allocation requests | |
227 | * both for ourself and for other kernel subsystems. | |
228 | */ | |
229 | batch_target = np->rx_target - (req_prod - np->rx.rsp_cons); | |
230 | for (i = skb_queue_len(&np->rx_batch); i < batch_target; i++) { | |
231 | skb = __netdev_alloc_skb(dev, RX_COPY_THRESHOLD, | |
232 | GFP_ATOMIC | __GFP_NOWARN); | |
233 | if (unlikely(!skb)) | |
234 | goto no_skb; | |
235 | ||
236 | page = alloc_page(GFP_ATOMIC | __GFP_NOWARN); | |
237 | if (!page) { | |
238 | kfree_skb(skb); | |
239 | no_skb: | |
240 | /* Any skbuffs queued for refill? Force them out. */ | |
241 | if (i != 0) | |
242 | goto refill; | |
243 | /* Could not allocate any skbuffs. Try again later. */ | |
244 | mod_timer(&np->rx_refill_timer, | |
245 | jiffies + (HZ/10)); | |
246 | break; | |
247 | } | |
248 | ||
249 | skb_shinfo(skb)->frags[0].page = page; | |
250 | skb_shinfo(skb)->nr_frags = 1; | |
251 | __skb_queue_tail(&np->rx_batch, skb); | |
252 | } | |
253 | ||
254 | /* Is the batch large enough to be worthwhile? */ | |
255 | if (i < (np->rx_target/2)) { | |
256 | if (req_prod > np->rx.sring->req_prod) | |
257 | goto push; | |
258 | return; | |
259 | } | |
260 | ||
261 | /* Adjust our fill target if we risked running out of buffers. */ | |
262 | if (((req_prod - np->rx.sring->rsp_prod) < (np->rx_target / 4)) && | |
263 | ((np->rx_target *= 2) > np->rx_max_target)) | |
264 | np->rx_target = np->rx_max_target; | |
265 | ||
266 | refill: | |
5dcddfae | 267 | for (i = 0; ; i++) { |
0d160211 JF |
268 | skb = __skb_dequeue(&np->rx_batch); |
269 | if (skb == NULL) | |
270 | break; | |
271 | ||
272 | skb->dev = dev; | |
273 | ||
274 | id = xennet_rxidx(req_prod + i); | |
275 | ||
276 | BUG_ON(np->rx_skbs[id]); | |
277 | np->rx_skbs[id] = skb; | |
278 | ||
279 | ref = gnttab_claim_grant_reference(&np->gref_rx_head); | |
280 | BUG_ON((signed short)ref < 0); | |
281 | np->grant_rx_ref[id] = ref; | |
282 | ||
283 | pfn = page_to_pfn(skb_shinfo(skb)->frags[0].page); | |
284 | vaddr = page_address(skb_shinfo(skb)->frags[0].page); | |
285 | ||
286 | req = RING_GET_REQUEST(&np->rx, req_prod + i); | |
287 | gnttab_grant_foreign_access_ref(ref, | |
288 | np->xbdev->otherend_id, | |
289 | pfn_to_mfn(pfn), | |
290 | 0); | |
291 | ||
292 | req->id = id; | |
293 | req->gref = ref; | |
294 | } | |
295 | ||
5dcddfae | 296 | wmb(); /* barrier so backend seens requests */ |
0d160211 JF |
297 | |
298 | /* Above is a suitable barrier to ensure backend will see requests. */ | |
299 | np->rx.req_prod_pvt = req_prod + i; | |
300 | push: | |
301 | RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np->rx, notify); | |
302 | if (notify) | |
303 | notify_remote_via_irq(np->netdev->irq); | |
304 | } | |
305 | ||
306 | static int xennet_open(struct net_device *dev) | |
307 | { | |
308 | struct netfront_info *np = netdev_priv(dev); | |
309 | ||
bea3348e SH |
310 | napi_enable(&np->napi); |
311 | ||
0d160211 JF |
312 | spin_lock_bh(&np->rx_lock); |
313 | if (netif_carrier_ok(dev)) { | |
314 | xennet_alloc_rx_buffers(dev); | |
315 | np->rx.sring->rsp_event = np->rx.rsp_cons + 1; | |
316 | if (RING_HAS_UNCONSUMED_RESPONSES(&np->rx)) | |
bea3348e | 317 | netif_rx_schedule(dev, &np->napi); |
0d160211 JF |
318 | } |
319 | spin_unlock_bh(&np->rx_lock); | |
320 | ||
321 | xennet_maybe_wake_tx(dev); | |
322 | ||
323 | return 0; | |
324 | } | |
325 | ||
326 | static void xennet_tx_buf_gc(struct net_device *dev) | |
327 | { | |
328 | RING_IDX cons, prod; | |
329 | unsigned short id; | |
330 | struct netfront_info *np = netdev_priv(dev); | |
331 | struct sk_buff *skb; | |
332 | ||
333 | BUG_ON(!netif_carrier_ok(dev)); | |
334 | ||
335 | do { | |
336 | prod = np->tx.sring->rsp_prod; | |
337 | rmb(); /* Ensure we see responses up to 'rp'. */ | |
338 | ||
339 | for (cons = np->tx.rsp_cons; cons != prod; cons++) { | |
340 | struct xen_netif_tx_response *txrsp; | |
341 | ||
342 | txrsp = RING_GET_RESPONSE(&np->tx, cons); | |
343 | if (txrsp->status == NETIF_RSP_NULL) | |
344 | continue; | |
345 | ||
346 | id = txrsp->id; | |
347 | skb = np->tx_skbs[id].skb; | |
348 | if (unlikely(gnttab_query_foreign_access( | |
349 | np->grant_tx_ref[id]) != 0)) { | |
350 | printk(KERN_ALERT "xennet_tx_buf_gc: warning " | |
351 | "-- grant still in use by backend " | |
352 | "domain.\n"); | |
353 | BUG(); | |
354 | } | |
355 | gnttab_end_foreign_access_ref( | |
356 | np->grant_tx_ref[id], GNTMAP_readonly); | |
357 | gnttab_release_grant_reference( | |
358 | &np->gref_tx_head, np->grant_tx_ref[id]); | |
359 | np->grant_tx_ref[id] = GRANT_INVALID_REF; | |
360 | add_id_to_freelist(&np->tx_skb_freelist, np->tx_skbs, id); | |
361 | dev_kfree_skb_irq(skb); | |
362 | } | |
363 | ||
364 | np->tx.rsp_cons = prod; | |
365 | ||
366 | /* | |
367 | * Set a new event, then check for race with update of tx_cons. | |
368 | * Note that it is essential to schedule a callback, no matter | |
369 | * how few buffers are pending. Even if there is space in the | |
370 | * transmit ring, higher layers may be blocked because too much | |
371 | * data is outstanding: in such cases notification from Xen is | |
372 | * likely to be the only kick that we'll get. | |
373 | */ | |
374 | np->tx.sring->rsp_event = | |
375 | prod + ((np->tx.sring->req_prod - prod) >> 1) + 1; | |
376 | mb(); /* update shared area */ | |
377 | } while ((cons == prod) && (prod != np->tx.sring->rsp_prod)); | |
378 | ||
379 | xennet_maybe_wake_tx(dev); | |
380 | } | |
381 | ||
382 | static void xennet_make_frags(struct sk_buff *skb, struct net_device *dev, | |
383 | struct xen_netif_tx_request *tx) | |
384 | { | |
385 | struct netfront_info *np = netdev_priv(dev); | |
386 | char *data = skb->data; | |
387 | unsigned long mfn; | |
388 | RING_IDX prod = np->tx.req_prod_pvt; | |
389 | int frags = skb_shinfo(skb)->nr_frags; | |
390 | unsigned int offset = offset_in_page(data); | |
391 | unsigned int len = skb_headlen(skb); | |
392 | unsigned int id; | |
393 | grant_ref_t ref; | |
394 | int i; | |
395 | ||
396 | /* While the header overlaps a page boundary (including being | |
397 | larger than a page), split it it into page-sized chunks. */ | |
398 | while (len > PAGE_SIZE - offset) { | |
399 | tx->size = PAGE_SIZE - offset; | |
400 | tx->flags |= NETTXF_more_data; | |
401 | len -= tx->size; | |
402 | data += tx->size; | |
403 | offset = 0; | |
404 | ||
405 | id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs); | |
406 | np->tx_skbs[id].skb = skb_get(skb); | |
407 | tx = RING_GET_REQUEST(&np->tx, prod++); | |
408 | tx->id = id; | |
409 | ref = gnttab_claim_grant_reference(&np->gref_tx_head); | |
410 | BUG_ON((signed short)ref < 0); | |
411 | ||
412 | mfn = virt_to_mfn(data); | |
413 | gnttab_grant_foreign_access_ref(ref, np->xbdev->otherend_id, | |
414 | mfn, GNTMAP_readonly); | |
415 | ||
416 | tx->gref = np->grant_tx_ref[id] = ref; | |
417 | tx->offset = offset; | |
418 | tx->size = len; | |
419 | tx->flags = 0; | |
420 | } | |
421 | ||
422 | /* Grant backend access to each skb fragment page. */ | |
423 | for (i = 0; i < frags; i++) { | |
424 | skb_frag_t *frag = skb_shinfo(skb)->frags + i; | |
425 | ||
426 | tx->flags |= NETTXF_more_data; | |
427 | ||
428 | id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs); | |
429 | np->tx_skbs[id].skb = skb_get(skb); | |
430 | tx = RING_GET_REQUEST(&np->tx, prod++); | |
431 | tx->id = id; | |
432 | ref = gnttab_claim_grant_reference(&np->gref_tx_head); | |
433 | BUG_ON((signed short)ref < 0); | |
434 | ||
435 | mfn = pfn_to_mfn(page_to_pfn(frag->page)); | |
436 | gnttab_grant_foreign_access_ref(ref, np->xbdev->otherend_id, | |
437 | mfn, GNTMAP_readonly); | |
438 | ||
439 | tx->gref = np->grant_tx_ref[id] = ref; | |
440 | tx->offset = frag->page_offset; | |
441 | tx->size = frag->size; | |
442 | tx->flags = 0; | |
443 | } | |
444 | ||
445 | np->tx.req_prod_pvt = prod; | |
446 | } | |
447 | ||
448 | static int xennet_start_xmit(struct sk_buff *skb, struct net_device *dev) | |
449 | { | |
450 | unsigned short id; | |
451 | struct netfront_info *np = netdev_priv(dev); | |
452 | struct xen_netif_tx_request *tx; | |
453 | struct xen_netif_extra_info *extra; | |
454 | char *data = skb->data; | |
455 | RING_IDX i; | |
456 | grant_ref_t ref; | |
457 | unsigned long mfn; | |
458 | int notify; | |
459 | int frags = skb_shinfo(skb)->nr_frags; | |
460 | unsigned int offset = offset_in_page(data); | |
461 | unsigned int len = skb_headlen(skb); | |
462 | ||
463 | frags += (offset + len + PAGE_SIZE - 1) / PAGE_SIZE; | |
464 | if (unlikely(frags > MAX_SKB_FRAGS + 1)) { | |
465 | printk(KERN_ALERT "xennet: skb rides the rocket: %d frags\n", | |
466 | frags); | |
467 | dump_stack(); | |
468 | goto drop; | |
469 | } | |
470 | ||
471 | spin_lock_irq(&np->tx_lock); | |
472 | ||
473 | if (unlikely(!netif_carrier_ok(dev) || | |
474 | (frags > 1 && !xennet_can_sg(dev)) || | |
475 | netif_needs_gso(dev, skb))) { | |
476 | spin_unlock_irq(&np->tx_lock); | |
477 | goto drop; | |
478 | } | |
479 | ||
480 | i = np->tx.req_prod_pvt; | |
481 | ||
482 | id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs); | |
483 | np->tx_skbs[id].skb = skb; | |
484 | ||
485 | tx = RING_GET_REQUEST(&np->tx, i); | |
486 | ||
487 | tx->id = id; | |
488 | ref = gnttab_claim_grant_reference(&np->gref_tx_head); | |
489 | BUG_ON((signed short)ref < 0); | |
490 | mfn = virt_to_mfn(data); | |
491 | gnttab_grant_foreign_access_ref( | |
492 | ref, np->xbdev->otherend_id, mfn, GNTMAP_readonly); | |
493 | tx->gref = np->grant_tx_ref[id] = ref; | |
494 | tx->offset = offset; | |
495 | tx->size = len; | |
496 | extra = NULL; | |
497 | ||
498 | tx->flags = 0; | |
499 | if (skb->ip_summed == CHECKSUM_PARTIAL) | |
500 | /* local packet? */ | |
501 | tx->flags |= NETTXF_csum_blank | NETTXF_data_validated; | |
502 | else if (skb->ip_summed == CHECKSUM_UNNECESSARY) | |
503 | /* remote but checksummed. */ | |
504 | tx->flags |= NETTXF_data_validated; | |
505 | ||
506 | if (skb_shinfo(skb)->gso_size) { | |
507 | struct xen_netif_extra_info *gso; | |
508 | ||
509 | gso = (struct xen_netif_extra_info *) | |
510 | RING_GET_REQUEST(&np->tx, ++i); | |
511 | ||
512 | if (extra) | |
513 | extra->flags |= XEN_NETIF_EXTRA_FLAG_MORE; | |
514 | else | |
515 | tx->flags |= NETTXF_extra_info; | |
516 | ||
517 | gso->u.gso.size = skb_shinfo(skb)->gso_size; | |
518 | gso->u.gso.type = XEN_NETIF_GSO_TYPE_TCPV4; | |
519 | gso->u.gso.pad = 0; | |
520 | gso->u.gso.features = 0; | |
521 | ||
522 | gso->type = XEN_NETIF_EXTRA_TYPE_GSO; | |
523 | gso->flags = 0; | |
524 | extra = gso; | |
525 | } | |
526 | ||
527 | np->tx.req_prod_pvt = i + 1; | |
528 | ||
529 | xennet_make_frags(skb, dev, tx); | |
530 | tx->size = skb->len; | |
531 | ||
532 | RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np->tx, notify); | |
533 | if (notify) | |
534 | notify_remote_via_irq(np->netdev->irq); | |
535 | ||
09f75cd7 JG |
536 | dev->stats.tx_bytes += skb->len; |
537 | dev->stats.tx_packets++; | |
10a273a6 JF |
538 | |
539 | /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */ | |
0d160211 JF |
540 | xennet_tx_buf_gc(dev); |
541 | ||
542 | if (!netfront_tx_slot_available(np)) | |
543 | netif_stop_queue(dev); | |
544 | ||
545 | spin_unlock_irq(&np->tx_lock); | |
546 | ||
0d160211 JF |
547 | return 0; |
548 | ||
549 | drop: | |
09f75cd7 | 550 | dev->stats.tx_dropped++; |
0d160211 JF |
551 | dev_kfree_skb(skb); |
552 | return 0; | |
553 | } | |
554 | ||
555 | static int xennet_close(struct net_device *dev) | |
556 | { | |
557 | struct netfront_info *np = netdev_priv(dev); | |
558 | netif_stop_queue(np->netdev); | |
bea3348e | 559 | napi_disable(&np->napi); |
0d160211 JF |
560 | return 0; |
561 | } | |
562 | ||
0d160211 JF |
563 | static void xennet_move_rx_slot(struct netfront_info *np, struct sk_buff *skb, |
564 | grant_ref_t ref) | |
565 | { | |
566 | int new = xennet_rxidx(np->rx.req_prod_pvt); | |
567 | ||
568 | BUG_ON(np->rx_skbs[new]); | |
569 | np->rx_skbs[new] = skb; | |
570 | np->grant_rx_ref[new] = ref; | |
571 | RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->id = new; | |
572 | RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->gref = ref; | |
573 | np->rx.req_prod_pvt++; | |
574 | } | |
575 | ||
576 | static int xennet_get_extras(struct netfront_info *np, | |
577 | struct xen_netif_extra_info *extras, | |
578 | RING_IDX rp) | |
579 | ||
580 | { | |
581 | struct xen_netif_extra_info *extra; | |
582 | struct device *dev = &np->netdev->dev; | |
583 | RING_IDX cons = np->rx.rsp_cons; | |
584 | int err = 0; | |
585 | ||
586 | do { | |
587 | struct sk_buff *skb; | |
588 | grant_ref_t ref; | |
589 | ||
590 | if (unlikely(cons + 1 == rp)) { | |
591 | if (net_ratelimit()) | |
592 | dev_warn(dev, "Missing extra info\n"); | |
593 | err = -EBADR; | |
594 | break; | |
595 | } | |
596 | ||
597 | extra = (struct xen_netif_extra_info *) | |
598 | RING_GET_RESPONSE(&np->rx, ++cons); | |
599 | ||
600 | if (unlikely(!extra->type || | |
601 | extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) { | |
602 | if (net_ratelimit()) | |
603 | dev_warn(dev, "Invalid extra type: %d\n", | |
604 | extra->type); | |
605 | err = -EINVAL; | |
606 | } else { | |
607 | memcpy(&extras[extra->type - 1], extra, | |
608 | sizeof(*extra)); | |
609 | } | |
610 | ||
611 | skb = xennet_get_rx_skb(np, cons); | |
612 | ref = xennet_get_rx_ref(np, cons); | |
613 | xennet_move_rx_slot(np, skb, ref); | |
614 | } while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE); | |
615 | ||
616 | np->rx.rsp_cons = cons; | |
617 | return err; | |
618 | } | |
619 | ||
620 | static int xennet_get_responses(struct netfront_info *np, | |
621 | struct netfront_rx_info *rinfo, RING_IDX rp, | |
622 | struct sk_buff_head *list) | |
623 | { | |
624 | struct xen_netif_rx_response *rx = &rinfo->rx; | |
625 | struct xen_netif_extra_info *extras = rinfo->extras; | |
626 | struct device *dev = &np->netdev->dev; | |
627 | RING_IDX cons = np->rx.rsp_cons; | |
628 | struct sk_buff *skb = xennet_get_rx_skb(np, cons); | |
629 | grant_ref_t ref = xennet_get_rx_ref(np, cons); | |
630 | int max = MAX_SKB_FRAGS + (rx->status <= RX_COPY_THRESHOLD); | |
631 | int frags = 1; | |
632 | int err = 0; | |
633 | unsigned long ret; | |
634 | ||
635 | if (rx->flags & NETRXF_extra_info) { | |
636 | err = xennet_get_extras(np, extras, rp); | |
637 | cons = np->rx.rsp_cons; | |
638 | } | |
639 | ||
640 | for (;;) { | |
641 | if (unlikely(rx->status < 0 || | |
642 | rx->offset + rx->status > PAGE_SIZE)) { | |
643 | if (net_ratelimit()) | |
644 | dev_warn(dev, "rx->offset: %x, size: %u\n", | |
645 | rx->offset, rx->status); | |
646 | xennet_move_rx_slot(np, skb, ref); | |
647 | err = -EINVAL; | |
648 | goto next; | |
649 | } | |
650 | ||
651 | /* | |
652 | * This definitely indicates a bug, either in this driver or in | |
653 | * the backend driver. In future this should flag the bad | |
654 | * situation to the system controller to reboot the backed. | |
655 | */ | |
656 | if (ref == GRANT_INVALID_REF) { | |
657 | if (net_ratelimit()) | |
658 | dev_warn(dev, "Bad rx response id %d.\n", | |
659 | rx->id); | |
660 | err = -EINVAL; | |
661 | goto next; | |
662 | } | |
663 | ||
664 | ret = gnttab_end_foreign_access_ref(ref, 0); | |
665 | BUG_ON(!ret); | |
666 | ||
667 | gnttab_release_grant_reference(&np->gref_rx_head, ref); | |
668 | ||
669 | __skb_queue_tail(list, skb); | |
670 | ||
671 | next: | |
672 | if (!(rx->flags & NETRXF_more_data)) | |
673 | break; | |
674 | ||
675 | if (cons + frags == rp) { | |
676 | if (net_ratelimit()) | |
677 | dev_warn(dev, "Need more frags\n"); | |
678 | err = -ENOENT; | |
679 | break; | |
680 | } | |
681 | ||
682 | rx = RING_GET_RESPONSE(&np->rx, cons + frags); | |
683 | skb = xennet_get_rx_skb(np, cons + frags); | |
684 | ref = xennet_get_rx_ref(np, cons + frags); | |
685 | frags++; | |
686 | } | |
687 | ||
688 | if (unlikely(frags > max)) { | |
689 | if (net_ratelimit()) | |
690 | dev_warn(dev, "Too many frags\n"); | |
691 | err = -E2BIG; | |
692 | } | |
693 | ||
694 | if (unlikely(err)) | |
695 | np->rx.rsp_cons = cons + frags; | |
696 | ||
697 | return err; | |
698 | } | |
699 | ||
700 | static int xennet_set_skb_gso(struct sk_buff *skb, | |
701 | struct xen_netif_extra_info *gso) | |
702 | { | |
703 | if (!gso->u.gso.size) { | |
704 | if (net_ratelimit()) | |
705 | printk(KERN_WARNING "GSO size must not be zero.\n"); | |
706 | return -EINVAL; | |
707 | } | |
708 | ||
709 | /* Currently only TCPv4 S.O. is supported. */ | |
710 | if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4) { | |
711 | if (net_ratelimit()) | |
712 | printk(KERN_WARNING "Bad GSO type %d.\n", gso->u.gso.type); | |
713 | return -EINVAL; | |
714 | } | |
715 | ||
716 | skb_shinfo(skb)->gso_size = gso->u.gso.size; | |
717 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4; | |
718 | ||
719 | /* Header must be checked, and gso_segs computed. */ | |
720 | skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; | |
721 | skb_shinfo(skb)->gso_segs = 0; | |
722 | ||
723 | return 0; | |
724 | } | |
725 | ||
726 | static RING_IDX xennet_fill_frags(struct netfront_info *np, | |
727 | struct sk_buff *skb, | |
728 | struct sk_buff_head *list) | |
729 | { | |
730 | struct skb_shared_info *shinfo = skb_shinfo(skb); | |
731 | int nr_frags = shinfo->nr_frags; | |
732 | RING_IDX cons = np->rx.rsp_cons; | |
733 | skb_frag_t *frag = shinfo->frags + nr_frags; | |
734 | struct sk_buff *nskb; | |
735 | ||
736 | while ((nskb = __skb_dequeue(list))) { | |
737 | struct xen_netif_rx_response *rx = | |
738 | RING_GET_RESPONSE(&np->rx, ++cons); | |
739 | ||
740 | frag->page = skb_shinfo(nskb)->frags[0].page; | |
741 | frag->page_offset = rx->offset; | |
742 | frag->size = rx->status; | |
743 | ||
744 | skb->data_len += rx->status; | |
745 | ||
746 | skb_shinfo(nskb)->nr_frags = 0; | |
747 | kfree_skb(nskb); | |
748 | ||
749 | frag++; | |
750 | nr_frags++; | |
751 | } | |
752 | ||
753 | shinfo->nr_frags = nr_frags; | |
754 | return cons; | |
755 | } | |
756 | ||
757 | static int skb_checksum_setup(struct sk_buff *skb) | |
758 | { | |
759 | struct iphdr *iph; | |
760 | unsigned char *th; | |
761 | int err = -EPROTO; | |
762 | ||
763 | if (skb->protocol != htons(ETH_P_IP)) | |
764 | goto out; | |
765 | ||
766 | iph = (void *)skb->data; | |
767 | th = skb->data + 4 * iph->ihl; | |
768 | if (th >= skb_tail_pointer(skb)) | |
769 | goto out; | |
770 | ||
771 | skb->csum_start = th - skb->head; | |
772 | switch (iph->protocol) { | |
773 | case IPPROTO_TCP: | |
774 | skb->csum_offset = offsetof(struct tcphdr, check); | |
775 | break; | |
776 | case IPPROTO_UDP: | |
777 | skb->csum_offset = offsetof(struct udphdr, check); | |
778 | break; | |
779 | default: | |
780 | if (net_ratelimit()) | |
781 | printk(KERN_ERR "Attempting to checksum a non-" | |
782 | "TCP/UDP packet, dropping a protocol" | |
783 | " %d packet", iph->protocol); | |
784 | goto out; | |
785 | } | |
786 | ||
787 | if ((th + skb->csum_offset + 2) > skb_tail_pointer(skb)) | |
788 | goto out; | |
789 | ||
790 | err = 0; | |
791 | ||
792 | out: | |
793 | return err; | |
794 | } | |
795 | ||
796 | static int handle_incoming_queue(struct net_device *dev, | |
09f75cd7 | 797 | struct sk_buff_head *rxq) |
0d160211 | 798 | { |
0d160211 JF |
799 | int packets_dropped = 0; |
800 | struct sk_buff *skb; | |
801 | ||
802 | while ((skb = __skb_dequeue(rxq)) != NULL) { | |
803 | struct page *page = NETFRONT_SKB_CB(skb)->page; | |
804 | void *vaddr = page_address(page); | |
805 | unsigned offset = NETFRONT_SKB_CB(skb)->offset; | |
806 | ||
807 | memcpy(skb->data, vaddr + offset, | |
808 | skb_headlen(skb)); | |
809 | ||
810 | if (page != skb_shinfo(skb)->frags[0].page) | |
811 | __free_page(page); | |
812 | ||
813 | /* Ethernet work: Delayed to here as it peeks the header. */ | |
814 | skb->protocol = eth_type_trans(skb, dev); | |
815 | ||
816 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
817 | if (skb_checksum_setup(skb)) { | |
818 | kfree_skb(skb); | |
819 | packets_dropped++; | |
09f75cd7 | 820 | dev->stats.rx_errors++; |
0d160211 JF |
821 | continue; |
822 | } | |
823 | } | |
824 | ||
09f75cd7 JG |
825 | dev->stats.rx_packets++; |
826 | dev->stats.rx_bytes += skb->len; | |
0d160211 JF |
827 | |
828 | /* Pass it up. */ | |
829 | netif_receive_skb(skb); | |
830 | dev->last_rx = jiffies; | |
831 | } | |
832 | ||
833 | return packets_dropped; | |
834 | } | |
835 | ||
bea3348e | 836 | static int xennet_poll(struct napi_struct *napi, int budget) |
0d160211 | 837 | { |
bea3348e SH |
838 | struct netfront_info *np = container_of(napi, struct netfront_info, napi); |
839 | struct net_device *dev = np->netdev; | |
0d160211 JF |
840 | struct sk_buff *skb; |
841 | struct netfront_rx_info rinfo; | |
842 | struct xen_netif_rx_response *rx = &rinfo.rx; | |
843 | struct xen_netif_extra_info *extras = rinfo.extras; | |
844 | RING_IDX i, rp; | |
bea3348e | 845 | int work_done; |
0d160211 JF |
846 | struct sk_buff_head rxq; |
847 | struct sk_buff_head errq; | |
848 | struct sk_buff_head tmpq; | |
849 | unsigned long flags; | |
850 | unsigned int len; | |
851 | int err; | |
852 | ||
853 | spin_lock(&np->rx_lock); | |
854 | ||
0d160211 JF |
855 | skb_queue_head_init(&rxq); |
856 | skb_queue_head_init(&errq); | |
857 | skb_queue_head_init(&tmpq); | |
858 | ||
0d160211 JF |
859 | rp = np->rx.sring->rsp_prod; |
860 | rmb(); /* Ensure we see queued responses up to 'rp'. */ | |
861 | ||
862 | i = np->rx.rsp_cons; | |
863 | work_done = 0; | |
864 | while ((i != rp) && (work_done < budget)) { | |
865 | memcpy(rx, RING_GET_RESPONSE(&np->rx, i), sizeof(*rx)); | |
866 | memset(extras, 0, sizeof(rinfo.extras)); | |
867 | ||
868 | err = xennet_get_responses(np, &rinfo, rp, &tmpq); | |
869 | ||
870 | if (unlikely(err)) { | |
871 | err: | |
872 | while ((skb = __skb_dequeue(&tmpq))) | |
873 | __skb_queue_tail(&errq, skb); | |
09f75cd7 | 874 | dev->stats.rx_errors++; |
0d160211 JF |
875 | i = np->rx.rsp_cons; |
876 | continue; | |
877 | } | |
878 | ||
879 | skb = __skb_dequeue(&tmpq); | |
880 | ||
881 | if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) { | |
882 | struct xen_netif_extra_info *gso; | |
883 | gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1]; | |
884 | ||
885 | if (unlikely(xennet_set_skb_gso(skb, gso))) { | |
886 | __skb_queue_head(&tmpq, skb); | |
887 | np->rx.rsp_cons += skb_queue_len(&tmpq); | |
888 | goto err; | |
889 | } | |
890 | } | |
891 | ||
892 | NETFRONT_SKB_CB(skb)->page = skb_shinfo(skb)->frags[0].page; | |
893 | NETFRONT_SKB_CB(skb)->offset = rx->offset; | |
894 | ||
895 | len = rx->status; | |
896 | if (len > RX_COPY_THRESHOLD) | |
897 | len = RX_COPY_THRESHOLD; | |
898 | skb_put(skb, len); | |
899 | ||
900 | if (rx->status > len) { | |
901 | skb_shinfo(skb)->frags[0].page_offset = | |
902 | rx->offset + len; | |
903 | skb_shinfo(skb)->frags[0].size = rx->status - len; | |
904 | skb->data_len = rx->status - len; | |
905 | } else { | |
906 | skb_shinfo(skb)->frags[0].page = NULL; | |
907 | skb_shinfo(skb)->nr_frags = 0; | |
908 | } | |
909 | ||
910 | i = xennet_fill_frags(np, skb, &tmpq); | |
911 | ||
912 | /* | |
913 | * Truesize approximates the size of true data plus | |
914 | * any supervisor overheads. Adding hypervisor | |
915 | * overheads has been shown to significantly reduce | |
916 | * achievable bandwidth with the default receive | |
917 | * buffer size. It is therefore not wise to account | |
918 | * for it here. | |
919 | * | |
920 | * After alloc_skb(RX_COPY_THRESHOLD), truesize is set | |
921 | * to RX_COPY_THRESHOLD + the supervisor | |
922 | * overheads. Here, we add the size of the data pulled | |
923 | * in xennet_fill_frags(). | |
924 | * | |
925 | * We also adjust for any unused space in the main | |
926 | * data area by subtracting (RX_COPY_THRESHOLD - | |
927 | * len). This is especially important with drivers | |
928 | * which split incoming packets into header and data, | |
929 | * using only 66 bytes of the main data area (see the | |
930 | * e1000 driver for example.) On such systems, | |
931 | * without this last adjustement, our achievable | |
932 | * receive throughout using the standard receive | |
933 | * buffer size was cut by 25%(!!!). | |
934 | */ | |
935 | skb->truesize += skb->data_len - (RX_COPY_THRESHOLD - len); | |
936 | skb->len += skb->data_len; | |
937 | ||
938 | if (rx->flags & NETRXF_csum_blank) | |
939 | skb->ip_summed = CHECKSUM_PARTIAL; | |
940 | else if (rx->flags & NETRXF_data_validated) | |
941 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
942 | ||
943 | __skb_queue_tail(&rxq, skb); | |
944 | ||
945 | np->rx.rsp_cons = ++i; | |
946 | work_done++; | |
947 | } | |
948 | ||
949 | while ((skb = __skb_dequeue(&errq))) | |
950 | kfree_skb(skb); | |
951 | ||
952 | work_done -= handle_incoming_queue(dev, &rxq); | |
953 | ||
954 | /* If we get a callback with very few responses, reduce fill target. */ | |
955 | /* NB. Note exponential increase, linear decrease. */ | |
956 | if (((np->rx.req_prod_pvt - np->rx.sring->rsp_prod) > | |
957 | ((3*np->rx_target) / 4)) && | |
958 | (--np->rx_target < np->rx_min_target)) | |
959 | np->rx_target = np->rx_min_target; | |
960 | ||
961 | xennet_alloc_rx_buffers(dev); | |
962 | ||
0d160211 | 963 | if (work_done < budget) { |
bea3348e SH |
964 | int more_to_do = 0; |
965 | ||
0d160211 JF |
966 | local_irq_save(flags); |
967 | ||
968 | RING_FINAL_CHECK_FOR_RESPONSES(&np->rx, more_to_do); | |
969 | if (!more_to_do) | |
bea3348e | 970 | __netif_rx_complete(dev, napi); |
0d160211 JF |
971 | |
972 | local_irq_restore(flags); | |
973 | } | |
974 | ||
975 | spin_unlock(&np->rx_lock); | |
976 | ||
bea3348e | 977 | return work_done; |
0d160211 JF |
978 | } |
979 | ||
980 | static int xennet_change_mtu(struct net_device *dev, int mtu) | |
981 | { | |
982 | int max = xennet_can_sg(dev) ? 65535 - ETH_HLEN : ETH_DATA_LEN; | |
983 | ||
984 | if (mtu > max) | |
985 | return -EINVAL; | |
986 | dev->mtu = mtu; | |
987 | return 0; | |
988 | } | |
989 | ||
990 | static void xennet_release_tx_bufs(struct netfront_info *np) | |
991 | { | |
992 | struct sk_buff *skb; | |
993 | int i; | |
994 | ||
995 | for (i = 0; i < NET_TX_RING_SIZE; i++) { | |
996 | /* Skip over entries which are actually freelist references */ | |
997 | if ((unsigned long)np->tx_skbs[i].skb < PAGE_OFFSET) | |
998 | continue; | |
999 | ||
1000 | skb = np->tx_skbs[i].skb; | |
1001 | gnttab_end_foreign_access_ref(np->grant_tx_ref[i], | |
1002 | GNTMAP_readonly); | |
1003 | gnttab_release_grant_reference(&np->gref_tx_head, | |
1004 | np->grant_tx_ref[i]); | |
1005 | np->grant_tx_ref[i] = GRANT_INVALID_REF; | |
1006 | add_id_to_freelist(&np->tx_skb_freelist, np->tx_skbs, i); | |
1007 | dev_kfree_skb_irq(skb); | |
1008 | } | |
1009 | } | |
1010 | ||
1011 | static void xennet_release_rx_bufs(struct netfront_info *np) | |
1012 | { | |
1013 | struct mmu_update *mmu = np->rx_mmu; | |
1014 | struct multicall_entry *mcl = np->rx_mcl; | |
1015 | struct sk_buff_head free_list; | |
1016 | struct sk_buff *skb; | |
1017 | unsigned long mfn; | |
1018 | int xfer = 0, noxfer = 0, unused = 0; | |
1019 | int id, ref; | |
1020 | ||
1021 | dev_warn(&np->netdev->dev, "%s: fix me for copying receiver.\n", | |
1022 | __func__); | |
1023 | return; | |
1024 | ||
1025 | skb_queue_head_init(&free_list); | |
1026 | ||
1027 | spin_lock_bh(&np->rx_lock); | |
1028 | ||
1029 | for (id = 0; id < NET_RX_RING_SIZE; id++) { | |
1030 | ref = np->grant_rx_ref[id]; | |
1031 | if (ref == GRANT_INVALID_REF) { | |
1032 | unused++; | |
1033 | continue; | |
1034 | } | |
1035 | ||
1036 | skb = np->rx_skbs[id]; | |
1037 | mfn = gnttab_end_foreign_transfer_ref(ref); | |
1038 | gnttab_release_grant_reference(&np->gref_rx_head, ref); | |
1039 | np->grant_rx_ref[id] = GRANT_INVALID_REF; | |
1040 | ||
1041 | if (0 == mfn) { | |
1042 | skb_shinfo(skb)->nr_frags = 0; | |
1043 | dev_kfree_skb(skb); | |
1044 | noxfer++; | |
1045 | continue; | |
1046 | } | |
1047 | ||
1048 | if (!xen_feature(XENFEAT_auto_translated_physmap)) { | |
1049 | /* Remap the page. */ | |
1050 | struct page *page = skb_shinfo(skb)->frags[0].page; | |
1051 | unsigned long pfn = page_to_pfn(page); | |
1052 | void *vaddr = page_address(page); | |
1053 | ||
1054 | MULTI_update_va_mapping(mcl, (unsigned long)vaddr, | |
1055 | mfn_pte(mfn, PAGE_KERNEL), | |
1056 | 0); | |
1057 | mcl++; | |
1058 | mmu->ptr = ((u64)mfn << PAGE_SHIFT) | |
1059 | | MMU_MACHPHYS_UPDATE; | |
1060 | mmu->val = pfn; | |
1061 | mmu++; | |
1062 | ||
1063 | set_phys_to_machine(pfn, mfn); | |
1064 | } | |
1065 | __skb_queue_tail(&free_list, skb); | |
1066 | xfer++; | |
1067 | } | |
1068 | ||
1069 | dev_info(&np->netdev->dev, "%s: %d xfer, %d noxfer, %d unused\n", | |
1070 | __func__, xfer, noxfer, unused); | |
1071 | ||
1072 | if (xfer) { | |
1073 | if (!xen_feature(XENFEAT_auto_translated_physmap)) { | |
1074 | /* Do all the remapping work and M2P updates. */ | |
1075 | MULTI_mmu_update(mcl, np->rx_mmu, mmu - np->rx_mmu, | |
79ea13ce | 1076 | NULL, DOMID_SELF); |
0d160211 JF |
1077 | mcl++; |
1078 | HYPERVISOR_multicall(np->rx_mcl, mcl - np->rx_mcl); | |
1079 | } | |
1080 | } | |
1081 | ||
1082 | while ((skb = __skb_dequeue(&free_list)) != NULL) | |
1083 | dev_kfree_skb(skb); | |
1084 | ||
1085 | spin_unlock_bh(&np->rx_lock); | |
1086 | } | |
1087 | ||
1088 | static void xennet_uninit(struct net_device *dev) | |
1089 | { | |
1090 | struct netfront_info *np = netdev_priv(dev); | |
1091 | xennet_release_tx_bufs(np); | |
1092 | xennet_release_rx_bufs(np); | |
1093 | gnttab_free_grant_references(np->gref_tx_head); | |
1094 | gnttab_free_grant_references(np->gref_rx_head); | |
1095 | } | |
1096 | ||
1097 | static struct net_device * __devinit xennet_create_dev(struct xenbus_device *dev) | |
1098 | { | |
1099 | int i, err; | |
1100 | struct net_device *netdev; | |
1101 | struct netfront_info *np; | |
1102 | ||
1103 | netdev = alloc_etherdev(sizeof(struct netfront_info)); | |
1104 | if (!netdev) { | |
1105 | printk(KERN_WARNING "%s> alloc_etherdev failed.\n", | |
1106 | __func__); | |
1107 | return ERR_PTR(-ENOMEM); | |
1108 | } | |
1109 | ||
1110 | np = netdev_priv(netdev); | |
1111 | np->xbdev = dev; | |
1112 | ||
1113 | spin_lock_init(&np->tx_lock); | |
1114 | spin_lock_init(&np->rx_lock); | |
1115 | ||
1116 | skb_queue_head_init(&np->rx_batch); | |
1117 | np->rx_target = RX_DFL_MIN_TARGET; | |
1118 | np->rx_min_target = RX_DFL_MIN_TARGET; | |
1119 | np->rx_max_target = RX_MAX_TARGET; | |
1120 | ||
1121 | init_timer(&np->rx_refill_timer); | |
1122 | np->rx_refill_timer.data = (unsigned long)netdev; | |
1123 | np->rx_refill_timer.function = rx_refill_timeout; | |
1124 | ||
1125 | /* Initialise tx_skbs as a free chain containing every entry. */ | |
1126 | np->tx_skb_freelist = 0; | |
1127 | for (i = 0; i < NET_TX_RING_SIZE; i++) { | |
1128 | np->tx_skbs[i].link = i+1; | |
1129 | np->grant_tx_ref[i] = GRANT_INVALID_REF; | |
1130 | } | |
1131 | ||
1132 | /* Clear out rx_skbs */ | |
1133 | for (i = 0; i < NET_RX_RING_SIZE; i++) { | |
1134 | np->rx_skbs[i] = NULL; | |
1135 | np->grant_rx_ref[i] = GRANT_INVALID_REF; | |
1136 | } | |
1137 | ||
1138 | /* A grant for every tx ring slot */ | |
1139 | if (gnttab_alloc_grant_references(TX_MAX_TARGET, | |
1140 | &np->gref_tx_head) < 0) { | |
1141 | printk(KERN_ALERT "#### netfront can't alloc tx grant refs\n"); | |
1142 | err = -ENOMEM; | |
1143 | goto exit; | |
1144 | } | |
1145 | /* A grant for every rx ring slot */ | |
1146 | if (gnttab_alloc_grant_references(RX_MAX_TARGET, | |
1147 | &np->gref_rx_head) < 0) { | |
1148 | printk(KERN_ALERT "#### netfront can't alloc rx grant refs\n"); | |
1149 | err = -ENOMEM; | |
1150 | goto exit_free_tx; | |
1151 | } | |
1152 | ||
1153 | netdev->open = xennet_open; | |
1154 | netdev->hard_start_xmit = xennet_start_xmit; | |
1155 | netdev->stop = xennet_close; | |
bea3348e | 1156 | netif_napi_add(netdev, &np->napi, xennet_poll, 64); |
0d160211 JF |
1157 | netdev->uninit = xennet_uninit; |
1158 | netdev->change_mtu = xennet_change_mtu; | |
0d160211 JF |
1159 | netdev->features = NETIF_F_IP_CSUM; |
1160 | ||
1161 | SET_ETHTOOL_OPS(netdev, &xennet_ethtool_ops); | |
0d160211 JF |
1162 | SET_NETDEV_DEV(netdev, &dev->dev); |
1163 | ||
1164 | np->netdev = netdev; | |
1165 | ||
1166 | netif_carrier_off(netdev); | |
1167 | ||
1168 | return netdev; | |
1169 | ||
1170 | exit_free_tx: | |
1171 | gnttab_free_grant_references(np->gref_tx_head); | |
1172 | exit: | |
1173 | free_netdev(netdev); | |
1174 | return ERR_PTR(err); | |
1175 | } | |
1176 | ||
1177 | /** | |
1178 | * Entry point to this code when a new device is created. Allocate the basic | |
1179 | * structures and the ring buffers for communication with the backend, and | |
1180 | * inform the backend of the appropriate details for those. | |
1181 | */ | |
1182 | static int __devinit netfront_probe(struct xenbus_device *dev, | |
1183 | const struct xenbus_device_id *id) | |
1184 | { | |
1185 | int err; | |
1186 | struct net_device *netdev; | |
1187 | struct netfront_info *info; | |
1188 | ||
1189 | netdev = xennet_create_dev(dev); | |
1190 | if (IS_ERR(netdev)) { | |
1191 | err = PTR_ERR(netdev); | |
1192 | xenbus_dev_fatal(dev, err, "creating netdev"); | |
1193 | return err; | |
1194 | } | |
1195 | ||
1196 | info = netdev_priv(netdev); | |
1197 | dev->dev.driver_data = info; | |
1198 | ||
1199 | err = register_netdev(info->netdev); | |
1200 | if (err) { | |
1201 | printk(KERN_WARNING "%s: register_netdev err=%d\n", | |
1202 | __func__, err); | |
1203 | goto fail; | |
1204 | } | |
1205 | ||
1206 | err = xennet_sysfs_addif(info->netdev); | |
1207 | if (err) { | |
1208 | unregister_netdev(info->netdev); | |
1209 | printk(KERN_WARNING "%s: add sysfs failed err=%d\n", | |
1210 | __func__, err); | |
1211 | goto fail; | |
1212 | } | |
1213 | ||
1214 | return 0; | |
1215 | ||
1216 | fail: | |
1217 | free_netdev(netdev); | |
1218 | dev->dev.driver_data = NULL; | |
1219 | return err; | |
1220 | } | |
1221 | ||
1222 | static void xennet_end_access(int ref, void *page) | |
1223 | { | |
1224 | /* This frees the page as a side-effect */ | |
1225 | if (ref != GRANT_INVALID_REF) | |
1226 | gnttab_end_foreign_access(ref, 0, (unsigned long)page); | |
1227 | } | |
1228 | ||
1229 | static void xennet_disconnect_backend(struct netfront_info *info) | |
1230 | { | |
1231 | /* Stop old i/f to prevent errors whilst we rebuild the state. */ | |
1232 | spin_lock_bh(&info->rx_lock); | |
1233 | spin_lock_irq(&info->tx_lock); | |
1234 | netif_carrier_off(info->netdev); | |
1235 | spin_unlock_irq(&info->tx_lock); | |
1236 | spin_unlock_bh(&info->rx_lock); | |
1237 | ||
1238 | if (info->netdev->irq) | |
1239 | unbind_from_irqhandler(info->netdev->irq, info->netdev); | |
1240 | info->evtchn = info->netdev->irq = 0; | |
1241 | ||
1242 | /* End access and free the pages */ | |
1243 | xennet_end_access(info->tx_ring_ref, info->tx.sring); | |
1244 | xennet_end_access(info->rx_ring_ref, info->rx.sring); | |
1245 | ||
1246 | info->tx_ring_ref = GRANT_INVALID_REF; | |
1247 | info->rx_ring_ref = GRANT_INVALID_REF; | |
1248 | info->tx.sring = NULL; | |
1249 | info->rx.sring = NULL; | |
1250 | } | |
1251 | ||
1252 | /** | |
1253 | * We are reconnecting to the backend, due to a suspend/resume, or a backend | |
1254 | * driver restart. We tear down our netif structure and recreate it, but | |
1255 | * leave the device-layer structures intact so that this is transparent to the | |
1256 | * rest of the kernel. | |
1257 | */ | |
1258 | static int netfront_resume(struct xenbus_device *dev) | |
1259 | { | |
1260 | struct netfront_info *info = dev->dev.driver_data; | |
1261 | ||
1262 | dev_dbg(&dev->dev, "%s\n", dev->nodename); | |
1263 | ||
1264 | xennet_disconnect_backend(info); | |
1265 | return 0; | |
1266 | } | |
1267 | ||
1268 | static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[]) | |
1269 | { | |
1270 | char *s, *e, *macstr; | |
1271 | int i; | |
1272 | ||
1273 | macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL); | |
1274 | if (IS_ERR(macstr)) | |
1275 | return PTR_ERR(macstr); | |
1276 | ||
1277 | for (i = 0; i < ETH_ALEN; i++) { | |
1278 | mac[i] = simple_strtoul(s, &e, 16); | |
1279 | if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) { | |
1280 | kfree(macstr); | |
1281 | return -ENOENT; | |
1282 | } | |
1283 | s = e+1; | |
1284 | } | |
1285 | ||
1286 | kfree(macstr); | |
1287 | return 0; | |
1288 | } | |
1289 | ||
1290 | static irqreturn_t xennet_interrupt(int irq, void *dev_id) | |
1291 | { | |
1292 | struct net_device *dev = dev_id; | |
1293 | struct netfront_info *np = netdev_priv(dev); | |
1294 | unsigned long flags; | |
1295 | ||
1296 | spin_lock_irqsave(&np->tx_lock, flags); | |
1297 | ||
1298 | if (likely(netif_carrier_ok(dev))) { | |
1299 | xennet_tx_buf_gc(dev); | |
1300 | /* Under tx_lock: protects access to rx shared-ring indexes. */ | |
1301 | if (RING_HAS_UNCONSUMED_RESPONSES(&np->rx)) | |
bea3348e | 1302 | netif_rx_schedule(dev, &np->napi); |
0d160211 JF |
1303 | } |
1304 | ||
1305 | spin_unlock_irqrestore(&np->tx_lock, flags); | |
1306 | ||
1307 | return IRQ_HANDLED; | |
1308 | } | |
1309 | ||
1310 | static int setup_netfront(struct xenbus_device *dev, struct netfront_info *info) | |
1311 | { | |
1312 | struct xen_netif_tx_sring *txs; | |
1313 | struct xen_netif_rx_sring *rxs; | |
1314 | int err; | |
1315 | struct net_device *netdev = info->netdev; | |
1316 | ||
1317 | info->tx_ring_ref = GRANT_INVALID_REF; | |
1318 | info->rx_ring_ref = GRANT_INVALID_REF; | |
1319 | info->rx.sring = NULL; | |
1320 | info->tx.sring = NULL; | |
1321 | netdev->irq = 0; | |
1322 | ||
1323 | err = xen_net_read_mac(dev, netdev->dev_addr); | |
1324 | if (err) { | |
1325 | xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename); | |
1326 | goto fail; | |
1327 | } | |
1328 | ||
1329 | txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_KERNEL); | |
1330 | if (!txs) { | |
1331 | err = -ENOMEM; | |
1332 | xenbus_dev_fatal(dev, err, "allocating tx ring page"); | |
1333 | goto fail; | |
1334 | } | |
1335 | SHARED_RING_INIT(txs); | |
1336 | FRONT_RING_INIT(&info->tx, txs, PAGE_SIZE); | |
1337 | ||
1338 | err = xenbus_grant_ring(dev, virt_to_mfn(txs)); | |
1339 | if (err < 0) { | |
1340 | free_page((unsigned long)txs); | |
1341 | goto fail; | |
1342 | } | |
1343 | ||
1344 | info->tx_ring_ref = err; | |
1345 | rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_KERNEL); | |
1346 | if (!rxs) { | |
1347 | err = -ENOMEM; | |
1348 | xenbus_dev_fatal(dev, err, "allocating rx ring page"); | |
1349 | goto fail; | |
1350 | } | |
1351 | SHARED_RING_INIT(rxs); | |
1352 | FRONT_RING_INIT(&info->rx, rxs, PAGE_SIZE); | |
1353 | ||
1354 | err = xenbus_grant_ring(dev, virt_to_mfn(rxs)); | |
1355 | if (err < 0) { | |
1356 | free_page((unsigned long)rxs); | |
1357 | goto fail; | |
1358 | } | |
1359 | info->rx_ring_ref = err; | |
1360 | ||
1361 | err = xenbus_alloc_evtchn(dev, &info->evtchn); | |
1362 | if (err) | |
1363 | goto fail; | |
1364 | ||
1365 | err = bind_evtchn_to_irqhandler(info->evtchn, xennet_interrupt, | |
1366 | IRQF_SAMPLE_RANDOM, netdev->name, | |
1367 | netdev); | |
1368 | if (err < 0) | |
1369 | goto fail; | |
1370 | netdev->irq = err; | |
1371 | return 0; | |
1372 | ||
1373 | fail: | |
1374 | return err; | |
1375 | } | |
1376 | ||
1377 | /* Common code used when first setting up, and when resuming. */ | |
1378 | static int talk_to_backend(struct xenbus_device *dev, | |
1379 | struct netfront_info *info) | |
1380 | { | |
1381 | const char *message; | |
1382 | struct xenbus_transaction xbt; | |
1383 | int err; | |
1384 | ||
1385 | /* Create shared ring, alloc event channel. */ | |
1386 | err = setup_netfront(dev, info); | |
1387 | if (err) | |
1388 | goto out; | |
1389 | ||
1390 | again: | |
1391 | err = xenbus_transaction_start(&xbt); | |
1392 | if (err) { | |
1393 | xenbus_dev_fatal(dev, err, "starting transaction"); | |
1394 | goto destroy_ring; | |
1395 | } | |
1396 | ||
1397 | err = xenbus_printf(xbt, dev->nodename, "tx-ring-ref", "%u", | |
1398 | info->tx_ring_ref); | |
1399 | if (err) { | |
1400 | message = "writing tx ring-ref"; | |
1401 | goto abort_transaction; | |
1402 | } | |
1403 | err = xenbus_printf(xbt, dev->nodename, "rx-ring-ref", "%u", | |
1404 | info->rx_ring_ref); | |
1405 | if (err) { | |
1406 | message = "writing rx ring-ref"; | |
1407 | goto abort_transaction; | |
1408 | } | |
1409 | err = xenbus_printf(xbt, dev->nodename, | |
1410 | "event-channel", "%u", info->evtchn); | |
1411 | if (err) { | |
1412 | message = "writing event-channel"; | |
1413 | goto abort_transaction; | |
1414 | } | |
1415 | ||
1416 | err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u", | |
1417 | 1); | |
1418 | if (err) { | |
1419 | message = "writing request-rx-copy"; | |
1420 | goto abort_transaction; | |
1421 | } | |
1422 | ||
1423 | err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1); | |
1424 | if (err) { | |
1425 | message = "writing feature-rx-notify"; | |
1426 | goto abort_transaction; | |
1427 | } | |
1428 | ||
1429 | err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1); | |
1430 | if (err) { | |
1431 | message = "writing feature-sg"; | |
1432 | goto abort_transaction; | |
1433 | } | |
1434 | ||
1435 | err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1); | |
1436 | if (err) { | |
1437 | message = "writing feature-gso-tcpv4"; | |
1438 | goto abort_transaction; | |
1439 | } | |
1440 | ||
1441 | err = xenbus_transaction_end(xbt, 0); | |
1442 | if (err) { | |
1443 | if (err == -EAGAIN) | |
1444 | goto again; | |
1445 | xenbus_dev_fatal(dev, err, "completing transaction"); | |
1446 | goto destroy_ring; | |
1447 | } | |
1448 | ||
1449 | return 0; | |
1450 | ||
1451 | abort_transaction: | |
1452 | xenbus_transaction_end(xbt, 1); | |
1453 | xenbus_dev_fatal(dev, err, "%s", message); | |
1454 | destroy_ring: | |
1455 | xennet_disconnect_backend(info); | |
1456 | out: | |
1457 | return err; | |
1458 | } | |
1459 | ||
1460 | static int xennet_set_sg(struct net_device *dev, u32 data) | |
1461 | { | |
1462 | if (data) { | |
1463 | struct netfront_info *np = netdev_priv(dev); | |
1464 | int val; | |
1465 | ||
1466 | if (xenbus_scanf(XBT_NIL, np->xbdev->otherend, "feature-sg", | |
1467 | "%d", &val) < 0) | |
1468 | val = 0; | |
1469 | if (!val) | |
1470 | return -ENOSYS; | |
1471 | } else if (dev->mtu > ETH_DATA_LEN) | |
1472 | dev->mtu = ETH_DATA_LEN; | |
1473 | ||
1474 | return ethtool_op_set_sg(dev, data); | |
1475 | } | |
1476 | ||
1477 | static int xennet_set_tso(struct net_device *dev, u32 data) | |
1478 | { | |
1479 | if (data) { | |
1480 | struct netfront_info *np = netdev_priv(dev); | |
1481 | int val; | |
1482 | ||
1483 | if (xenbus_scanf(XBT_NIL, np->xbdev->otherend, | |
1484 | "feature-gso-tcpv4", "%d", &val) < 0) | |
1485 | val = 0; | |
1486 | if (!val) | |
1487 | return -ENOSYS; | |
1488 | } | |
1489 | ||
1490 | return ethtool_op_set_tso(dev, data); | |
1491 | } | |
1492 | ||
1493 | static void xennet_set_features(struct net_device *dev) | |
1494 | { | |
1495 | /* Turn off all GSO bits except ROBUST. */ | |
1496 | dev->features &= (1 << NETIF_F_GSO_SHIFT) - 1; | |
1497 | dev->features |= NETIF_F_GSO_ROBUST; | |
1498 | xennet_set_sg(dev, 0); | |
1499 | ||
1500 | /* We need checksum offload to enable scatter/gather and TSO. */ | |
1501 | if (!(dev->features & NETIF_F_IP_CSUM)) | |
1502 | return; | |
1503 | ||
1504 | if (!xennet_set_sg(dev, 1)) | |
1505 | xennet_set_tso(dev, 1); | |
1506 | } | |
1507 | ||
1508 | static int xennet_connect(struct net_device *dev) | |
1509 | { | |
1510 | struct netfront_info *np = netdev_priv(dev); | |
1511 | int i, requeue_idx, err; | |
1512 | struct sk_buff *skb; | |
1513 | grant_ref_t ref; | |
1514 | struct xen_netif_rx_request *req; | |
1515 | unsigned int feature_rx_copy; | |
1516 | ||
1517 | err = xenbus_scanf(XBT_NIL, np->xbdev->otherend, | |
1518 | "feature-rx-copy", "%u", &feature_rx_copy); | |
1519 | if (err != 1) | |
1520 | feature_rx_copy = 0; | |
1521 | ||
1522 | if (!feature_rx_copy) { | |
1523 | dev_info(&dev->dev, | |
898eb71c | 1524 | "backend does not support copying receive path\n"); |
0d160211 JF |
1525 | return -ENODEV; |
1526 | } | |
1527 | ||
1528 | err = talk_to_backend(np->xbdev, np); | |
1529 | if (err) | |
1530 | return err; | |
1531 | ||
1532 | xennet_set_features(dev); | |
1533 | ||
1534 | spin_lock_bh(&np->rx_lock); | |
1535 | spin_lock_irq(&np->tx_lock); | |
1536 | ||
1537 | /* Step 1: Discard all pending TX packet fragments. */ | |
1538 | xennet_release_tx_bufs(np); | |
1539 | ||
1540 | /* Step 2: Rebuild the RX buffer freelist and the RX ring itself. */ | |
1541 | for (requeue_idx = 0, i = 0; i < NET_RX_RING_SIZE; i++) { | |
1542 | if (!np->rx_skbs[i]) | |
1543 | continue; | |
1544 | ||
1545 | skb = np->rx_skbs[requeue_idx] = xennet_get_rx_skb(np, i); | |
1546 | ref = np->grant_rx_ref[requeue_idx] = xennet_get_rx_ref(np, i); | |
1547 | req = RING_GET_REQUEST(&np->rx, requeue_idx); | |
1548 | ||
1549 | gnttab_grant_foreign_access_ref( | |
1550 | ref, np->xbdev->otherend_id, | |
1551 | pfn_to_mfn(page_to_pfn(skb_shinfo(skb)-> | |
1552 | frags->page)), | |
1553 | 0); | |
1554 | req->gref = ref; | |
1555 | req->id = requeue_idx; | |
1556 | ||
1557 | requeue_idx++; | |
1558 | } | |
1559 | ||
1560 | np->rx.req_prod_pvt = requeue_idx; | |
1561 | ||
1562 | /* | |
1563 | * Step 3: All public and private state should now be sane. Get | |
1564 | * ready to start sending and receiving packets and give the driver | |
1565 | * domain a kick because we've probably just requeued some | |
1566 | * packets. | |
1567 | */ | |
1568 | netif_carrier_on(np->netdev); | |
1569 | notify_remote_via_irq(np->netdev->irq); | |
1570 | xennet_tx_buf_gc(dev); | |
1571 | xennet_alloc_rx_buffers(dev); | |
1572 | ||
1573 | spin_unlock_irq(&np->tx_lock); | |
1574 | spin_unlock_bh(&np->rx_lock); | |
1575 | ||
1576 | return 0; | |
1577 | } | |
1578 | ||
1579 | /** | |
1580 | * Callback received when the backend's state changes. | |
1581 | */ | |
1582 | static void backend_changed(struct xenbus_device *dev, | |
1583 | enum xenbus_state backend_state) | |
1584 | { | |
1585 | struct netfront_info *np = dev->dev.driver_data; | |
1586 | struct net_device *netdev = np->netdev; | |
1587 | ||
1588 | dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state)); | |
1589 | ||
1590 | switch (backend_state) { | |
1591 | case XenbusStateInitialising: | |
1592 | case XenbusStateInitialised: | |
1593 | case XenbusStateConnected: | |
1594 | case XenbusStateUnknown: | |
1595 | case XenbusStateClosed: | |
1596 | break; | |
1597 | ||
1598 | case XenbusStateInitWait: | |
1599 | if (dev->state != XenbusStateInitialising) | |
1600 | break; | |
1601 | if (xennet_connect(netdev) != 0) | |
1602 | break; | |
1603 | xenbus_switch_state(dev, XenbusStateConnected); | |
1604 | break; | |
1605 | ||
1606 | case XenbusStateClosing: | |
1607 | xenbus_frontend_closed(dev); | |
1608 | break; | |
1609 | } | |
1610 | } | |
1611 | ||
1612 | static struct ethtool_ops xennet_ethtool_ops = | |
1613 | { | |
0d160211 | 1614 | .set_tx_csum = ethtool_op_set_tx_csum, |
0d160211 | 1615 | .set_sg = xennet_set_sg, |
0d160211 JF |
1616 | .set_tso = xennet_set_tso, |
1617 | .get_link = ethtool_op_get_link, | |
1618 | }; | |
1619 | ||
1620 | #ifdef CONFIG_SYSFS | |
1621 | static ssize_t show_rxbuf_min(struct device *dev, | |
1622 | struct device_attribute *attr, char *buf) | |
1623 | { | |
1624 | struct net_device *netdev = to_net_dev(dev); | |
1625 | struct netfront_info *info = netdev_priv(netdev); | |
1626 | ||
1627 | return sprintf(buf, "%u\n", info->rx_min_target); | |
1628 | } | |
1629 | ||
1630 | static ssize_t store_rxbuf_min(struct device *dev, | |
1631 | struct device_attribute *attr, | |
1632 | const char *buf, size_t len) | |
1633 | { | |
1634 | struct net_device *netdev = to_net_dev(dev); | |
1635 | struct netfront_info *np = netdev_priv(netdev); | |
1636 | char *endp; | |
1637 | unsigned long target; | |
1638 | ||
1639 | if (!capable(CAP_NET_ADMIN)) | |
1640 | return -EPERM; | |
1641 | ||
1642 | target = simple_strtoul(buf, &endp, 0); | |
1643 | if (endp == buf) | |
1644 | return -EBADMSG; | |
1645 | ||
1646 | if (target < RX_MIN_TARGET) | |
1647 | target = RX_MIN_TARGET; | |
1648 | if (target > RX_MAX_TARGET) | |
1649 | target = RX_MAX_TARGET; | |
1650 | ||
1651 | spin_lock_bh(&np->rx_lock); | |
1652 | if (target > np->rx_max_target) | |
1653 | np->rx_max_target = target; | |
1654 | np->rx_min_target = target; | |
1655 | if (target > np->rx_target) | |
1656 | np->rx_target = target; | |
1657 | ||
1658 | xennet_alloc_rx_buffers(netdev); | |
1659 | ||
1660 | spin_unlock_bh(&np->rx_lock); | |
1661 | return len; | |
1662 | } | |
1663 | ||
1664 | static ssize_t show_rxbuf_max(struct device *dev, | |
1665 | struct device_attribute *attr, char *buf) | |
1666 | { | |
1667 | struct net_device *netdev = to_net_dev(dev); | |
1668 | struct netfront_info *info = netdev_priv(netdev); | |
1669 | ||
1670 | return sprintf(buf, "%u\n", info->rx_max_target); | |
1671 | } | |
1672 | ||
1673 | static ssize_t store_rxbuf_max(struct device *dev, | |
1674 | struct device_attribute *attr, | |
1675 | const char *buf, size_t len) | |
1676 | { | |
1677 | struct net_device *netdev = to_net_dev(dev); | |
1678 | struct netfront_info *np = netdev_priv(netdev); | |
1679 | char *endp; | |
1680 | unsigned long target; | |
1681 | ||
1682 | if (!capable(CAP_NET_ADMIN)) | |
1683 | return -EPERM; | |
1684 | ||
1685 | target = simple_strtoul(buf, &endp, 0); | |
1686 | if (endp == buf) | |
1687 | return -EBADMSG; | |
1688 | ||
1689 | if (target < RX_MIN_TARGET) | |
1690 | target = RX_MIN_TARGET; | |
1691 | if (target > RX_MAX_TARGET) | |
1692 | target = RX_MAX_TARGET; | |
1693 | ||
1694 | spin_lock_bh(&np->rx_lock); | |
1695 | if (target < np->rx_min_target) | |
1696 | np->rx_min_target = target; | |
1697 | np->rx_max_target = target; | |
1698 | if (target < np->rx_target) | |
1699 | np->rx_target = target; | |
1700 | ||
1701 | xennet_alloc_rx_buffers(netdev); | |
1702 | ||
1703 | spin_unlock_bh(&np->rx_lock); | |
1704 | return len; | |
1705 | } | |
1706 | ||
1707 | static ssize_t show_rxbuf_cur(struct device *dev, | |
1708 | struct device_attribute *attr, char *buf) | |
1709 | { | |
1710 | struct net_device *netdev = to_net_dev(dev); | |
1711 | struct netfront_info *info = netdev_priv(netdev); | |
1712 | ||
1713 | return sprintf(buf, "%u\n", info->rx_target); | |
1714 | } | |
1715 | ||
1716 | static struct device_attribute xennet_attrs[] = { | |
1717 | __ATTR(rxbuf_min, S_IRUGO|S_IWUSR, show_rxbuf_min, store_rxbuf_min), | |
1718 | __ATTR(rxbuf_max, S_IRUGO|S_IWUSR, show_rxbuf_max, store_rxbuf_max), | |
1719 | __ATTR(rxbuf_cur, S_IRUGO, show_rxbuf_cur, NULL), | |
1720 | }; | |
1721 | ||
1722 | static int xennet_sysfs_addif(struct net_device *netdev) | |
1723 | { | |
1724 | int i; | |
1725 | int err; | |
1726 | ||
1727 | for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++) { | |
1728 | err = device_create_file(&netdev->dev, | |
1729 | &xennet_attrs[i]); | |
1730 | if (err) | |
1731 | goto fail; | |
1732 | } | |
1733 | return 0; | |
1734 | ||
1735 | fail: | |
1736 | while (--i >= 0) | |
1737 | device_remove_file(&netdev->dev, &xennet_attrs[i]); | |
1738 | return err; | |
1739 | } | |
1740 | ||
1741 | static void xennet_sysfs_delif(struct net_device *netdev) | |
1742 | { | |
1743 | int i; | |
1744 | ||
1745 | for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++) | |
1746 | device_remove_file(&netdev->dev, &xennet_attrs[i]); | |
1747 | } | |
1748 | ||
1749 | #endif /* CONFIG_SYSFS */ | |
1750 | ||
1751 | static struct xenbus_device_id netfront_ids[] = { | |
1752 | { "vif" }, | |
1753 | { "" } | |
1754 | }; | |
1755 | ||
1756 | ||
1757 | static int __devexit xennet_remove(struct xenbus_device *dev) | |
1758 | { | |
1759 | struct netfront_info *info = dev->dev.driver_data; | |
1760 | ||
1761 | dev_dbg(&dev->dev, "%s\n", dev->nodename); | |
1762 | ||
1763 | unregister_netdev(info->netdev); | |
1764 | ||
1765 | xennet_disconnect_backend(info); | |
1766 | ||
1767 | del_timer_sync(&info->rx_refill_timer); | |
1768 | ||
1769 | xennet_sysfs_delif(info->netdev); | |
1770 | ||
1771 | free_netdev(info->netdev); | |
1772 | ||
1773 | return 0; | |
1774 | } | |
1775 | ||
1776 | static struct xenbus_driver netfront = { | |
1777 | .name = "vif", | |
1778 | .owner = THIS_MODULE, | |
1779 | .ids = netfront_ids, | |
1780 | .probe = netfront_probe, | |
1781 | .remove = __devexit_p(xennet_remove), | |
1782 | .resume = netfront_resume, | |
1783 | .otherend_changed = backend_changed, | |
1784 | }; | |
1785 | ||
1786 | static int __init netif_init(void) | |
1787 | { | |
1788 | if (!is_running_on_xen()) | |
1789 | return -ENODEV; | |
1790 | ||
1791 | if (is_initial_xendomain()) | |
1792 | return 0; | |
1793 | ||
1794 | printk(KERN_INFO "Initialising Xen virtual ethernet driver.\n"); | |
1795 | ||
1796 | return xenbus_register_frontend(&netfront); | |
1797 | } | |
1798 | module_init(netif_init); | |
1799 | ||
1800 | ||
1801 | static void __exit netif_exit(void) | |
1802 | { | |
1803 | if (is_initial_xendomain()) | |
1804 | return; | |
1805 | ||
ab59859d | 1806 | xenbus_unregister_driver(&netfront); |
0d160211 JF |
1807 | } |
1808 | module_exit(netif_exit); | |
1809 | ||
1810 | MODULE_DESCRIPTION("Xen virtual network device frontend"); | |
1811 | MODULE_LICENSE("GPL"); | |
d2f0c52b | 1812 | MODULE_ALIAS("xen:vif"); |
4f93f09b | 1813 | MODULE_ALIAS("xennet"); |