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 | |
0d160211 | 88 | unsigned int evtchn; |
84284d3c | 89 | struct xenbus_device *xbdev; |
0d160211 | 90 | |
84284d3c JF |
91 | spinlock_t tx_lock; |
92 | struct xen_netif_tx_front_ring tx; | |
93 | int tx_ring_ref; | |
0d160211 JF |
94 | |
95 | /* | |
96 | * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries | |
97 | * are linked from tx_skb_freelist through skb_entry.link. | |
98 | * | |
99 | * NB. Freelist index entries are always going to be less than | |
100 | * PAGE_OFFSET, whereas pointers to skbs will always be equal or | |
101 | * greater than PAGE_OFFSET: we use this property to distinguish | |
102 | * them. | |
103 | */ | |
104 | union skb_entry { | |
105 | struct sk_buff *skb; | |
1ffb40b8 | 106 | unsigned long link; |
0d160211 JF |
107 | } tx_skbs[NET_TX_RING_SIZE]; |
108 | grant_ref_t gref_tx_head; | |
109 | grant_ref_t grant_tx_ref[NET_TX_RING_SIZE]; | |
64cee83a | 110 | struct page *grant_tx_page[NET_TX_RING_SIZE]; |
0d160211 JF |
111 | unsigned tx_skb_freelist; |
112 | ||
84284d3c JF |
113 | spinlock_t rx_lock ____cacheline_aligned_in_smp; |
114 | struct xen_netif_rx_front_ring rx; | |
115 | int rx_ring_ref; | |
116 | ||
117 | /* Receive-ring batched refills. */ | |
118 | #define RX_MIN_TARGET 8 | |
119 | #define RX_DFL_MIN_TARGET 64 | |
120 | #define RX_MAX_TARGET min_t(int, NET_RX_RING_SIZE, 256) | |
121 | unsigned rx_min_target, rx_max_target, rx_target; | |
122 | struct sk_buff_head rx_batch; | |
123 | ||
124 | struct timer_list rx_refill_timer; | |
125 | ||
0d160211 JF |
126 | struct sk_buff *rx_skbs[NET_RX_RING_SIZE]; |
127 | grant_ref_t gref_rx_head; | |
128 | grant_ref_t grant_rx_ref[NET_RX_RING_SIZE]; | |
129 | ||
0d160211 JF |
130 | unsigned long rx_pfn_array[NET_RX_RING_SIZE]; |
131 | struct multicall_entry rx_mcl[NET_RX_RING_SIZE+1]; | |
132 | struct mmu_update rx_mmu[NET_RX_RING_SIZE]; | |
e0ce4af9 IC |
133 | |
134 | /* Statistics */ | |
e00f85be | 135 | struct netfront_stats __percpu *stats; |
136 | ||
e9a799ea | 137 | unsigned long rx_gso_checksum_fixup; |
0d160211 JF |
138 | }; |
139 | ||
140 | struct netfront_rx_info { | |
141 | struct xen_netif_rx_response rx; | |
142 | struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1]; | |
143 | }; | |
144 | ||
1ffb40b8 IY |
145 | static void skb_entry_set_link(union skb_entry *list, unsigned short id) |
146 | { | |
147 | list->link = id; | |
148 | } | |
149 | ||
150 | static int skb_entry_is_link(const union skb_entry *list) | |
151 | { | |
152 | BUILD_BUG_ON(sizeof(list->skb) != sizeof(list->link)); | |
807540ba | 153 | return (unsigned long)list->skb < PAGE_OFFSET; |
1ffb40b8 IY |
154 | } |
155 | ||
0d160211 JF |
156 | /* |
157 | * Access macros for acquiring freeing slots in tx_skbs[]. | |
158 | */ | |
159 | ||
160 | static void add_id_to_freelist(unsigned *head, union skb_entry *list, | |
161 | unsigned short id) | |
162 | { | |
1ffb40b8 | 163 | skb_entry_set_link(&list[id], *head); |
0d160211 JF |
164 | *head = id; |
165 | } | |
166 | ||
167 | static unsigned short get_id_from_freelist(unsigned *head, | |
168 | union skb_entry *list) | |
169 | { | |
170 | unsigned int id = *head; | |
171 | *head = list[id].link; | |
172 | return id; | |
173 | } | |
174 | ||
175 | static int xennet_rxidx(RING_IDX idx) | |
176 | { | |
177 | return idx & (NET_RX_RING_SIZE - 1); | |
178 | } | |
179 | ||
180 | static struct sk_buff *xennet_get_rx_skb(struct netfront_info *np, | |
181 | RING_IDX ri) | |
182 | { | |
183 | int i = xennet_rxidx(ri); | |
184 | struct sk_buff *skb = np->rx_skbs[i]; | |
185 | np->rx_skbs[i] = NULL; | |
186 | return skb; | |
187 | } | |
188 | ||
189 | static grant_ref_t xennet_get_rx_ref(struct netfront_info *np, | |
190 | RING_IDX ri) | |
191 | { | |
192 | int i = xennet_rxidx(ri); | |
193 | grant_ref_t ref = np->grant_rx_ref[i]; | |
194 | np->grant_rx_ref[i] = GRANT_INVALID_REF; | |
195 | return ref; | |
196 | } | |
197 | ||
198 | #ifdef CONFIG_SYSFS | |
199 | static int xennet_sysfs_addif(struct net_device *netdev); | |
200 | static void xennet_sysfs_delif(struct net_device *netdev); | |
201 | #else /* !CONFIG_SYSFS */ | |
202 | #define xennet_sysfs_addif(dev) (0) | |
203 | #define xennet_sysfs_delif(dev) do { } while (0) | |
204 | #endif | |
205 | ||
3ad9b358 | 206 | static bool xennet_can_sg(struct net_device *dev) |
0d160211 | 207 | { |
3ad9b358 | 208 | return dev->features & NETIF_F_SG; |
0d160211 JF |
209 | } |
210 | ||
211 | ||
212 | static void rx_refill_timeout(unsigned long data) | |
213 | { | |
214 | struct net_device *dev = (struct net_device *)data; | |
bea3348e | 215 | struct netfront_info *np = netdev_priv(dev); |
288379f0 | 216 | napi_schedule(&np->napi); |
0d160211 JF |
217 | } |
218 | ||
219 | static int netfront_tx_slot_available(struct netfront_info *np) | |
220 | { | |
807540ba ED |
221 | return (np->tx.req_prod_pvt - np->tx.rsp_cons) < |
222 | (TX_MAX_TARGET - MAX_SKB_FRAGS - 2); | |
0d160211 JF |
223 | } |
224 | ||
225 | static void xennet_maybe_wake_tx(struct net_device *dev) | |
226 | { | |
227 | struct netfront_info *np = netdev_priv(dev); | |
228 | ||
229 | if (unlikely(netif_queue_stopped(dev)) && | |
230 | netfront_tx_slot_available(np) && | |
231 | likely(netif_running(dev))) | |
232 | netif_wake_queue(dev); | |
233 | } | |
234 | ||
235 | static void xennet_alloc_rx_buffers(struct net_device *dev) | |
236 | { | |
237 | unsigned short id; | |
238 | struct netfront_info *np = netdev_priv(dev); | |
239 | struct sk_buff *skb; | |
240 | struct page *page; | |
241 | int i, batch_target, notify; | |
242 | RING_IDX req_prod = np->rx.req_prod_pvt; | |
0d160211 JF |
243 | grant_ref_t ref; |
244 | unsigned long pfn; | |
245 | void *vaddr; | |
0d160211 JF |
246 | struct xen_netif_rx_request *req; |
247 | ||
248 | if (unlikely(!netif_carrier_ok(dev))) | |
249 | return; | |
250 | ||
251 | /* | |
252 | * Allocate skbuffs greedily, even though we batch updates to the | |
253 | * receive ring. This creates a less bursty demand on the memory | |
254 | * allocator, so should reduce the chance of failed allocation requests | |
255 | * both for ourself and for other kernel subsystems. | |
256 | */ | |
257 | batch_target = np->rx_target - (req_prod - np->rx.rsp_cons); | |
258 | for (i = skb_queue_len(&np->rx_batch); i < batch_target; i++) { | |
617a20bb | 259 | skb = __netdev_alloc_skb(dev, RX_COPY_THRESHOLD + NET_IP_ALIGN, |
0d160211 JF |
260 | GFP_ATOMIC | __GFP_NOWARN); |
261 | if (unlikely(!skb)) | |
262 | goto no_skb; | |
263 | ||
617a20bb IY |
264 | /* Align ip header to a 16 bytes boundary */ |
265 | skb_reserve(skb, NET_IP_ALIGN); | |
266 | ||
0d160211 JF |
267 | page = alloc_page(GFP_ATOMIC | __GFP_NOWARN); |
268 | if (!page) { | |
269 | kfree_skb(skb); | |
270 | no_skb: | |
271 | /* Any skbuffs queued for refill? Force them out. */ | |
272 | if (i != 0) | |
273 | goto refill; | |
274 | /* Could not allocate any skbuffs. Try again later. */ | |
275 | mod_timer(&np->rx_refill_timer, | |
276 | jiffies + (HZ/10)); | |
277 | break; | |
278 | } | |
279 | ||
f4b5b99f | 280 | skb_add_rx_frag(skb, 0, page, 0, 0, PAGE_SIZE); |
0d160211 JF |
281 | __skb_queue_tail(&np->rx_batch, skb); |
282 | } | |
283 | ||
284 | /* Is the batch large enough to be worthwhile? */ | |
285 | if (i < (np->rx_target/2)) { | |
286 | if (req_prod > np->rx.sring->req_prod) | |
287 | goto push; | |
288 | return; | |
289 | } | |
290 | ||
291 | /* Adjust our fill target if we risked running out of buffers. */ | |
292 | if (((req_prod - np->rx.sring->rsp_prod) < (np->rx_target / 4)) && | |
293 | ((np->rx_target *= 2) > np->rx_max_target)) | |
294 | np->rx_target = np->rx_max_target; | |
295 | ||
296 | refill: | |
5dcddfae | 297 | for (i = 0; ; i++) { |
0d160211 JF |
298 | skb = __skb_dequeue(&np->rx_batch); |
299 | if (skb == NULL) | |
300 | break; | |
301 | ||
302 | skb->dev = dev; | |
303 | ||
304 | id = xennet_rxidx(req_prod + i); | |
305 | ||
306 | BUG_ON(np->rx_skbs[id]); | |
307 | np->rx_skbs[id] = skb; | |
308 | ||
309 | ref = gnttab_claim_grant_reference(&np->gref_rx_head); | |
310 | BUG_ON((signed short)ref < 0); | |
311 | np->grant_rx_ref[id] = ref; | |
312 | ||
01c68026 IC |
313 | pfn = page_to_pfn(skb_frag_page(&skb_shinfo(skb)->frags[0])); |
314 | vaddr = page_address(skb_frag_page(&skb_shinfo(skb)->frags[0])); | |
0d160211 JF |
315 | |
316 | req = RING_GET_REQUEST(&np->rx, req_prod + i); | |
317 | gnttab_grant_foreign_access_ref(ref, | |
318 | np->xbdev->otherend_id, | |
319 | pfn_to_mfn(pfn), | |
320 | 0); | |
321 | ||
322 | req->id = id; | |
323 | req->gref = ref; | |
324 | } | |
325 | ||
5dcddfae | 326 | wmb(); /* barrier so backend seens requests */ |
0d160211 JF |
327 | |
328 | /* Above is a suitable barrier to ensure backend will see requests. */ | |
329 | np->rx.req_prod_pvt = req_prod + i; | |
330 | push: | |
331 | RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np->rx, notify); | |
332 | if (notify) | |
333 | notify_remote_via_irq(np->netdev->irq); | |
334 | } | |
335 | ||
336 | static int xennet_open(struct net_device *dev) | |
337 | { | |
338 | struct netfront_info *np = netdev_priv(dev); | |
339 | ||
bea3348e SH |
340 | napi_enable(&np->napi); |
341 | ||
0d160211 JF |
342 | spin_lock_bh(&np->rx_lock); |
343 | if (netif_carrier_ok(dev)) { | |
344 | xennet_alloc_rx_buffers(dev); | |
345 | np->rx.sring->rsp_event = np->rx.rsp_cons + 1; | |
346 | if (RING_HAS_UNCONSUMED_RESPONSES(&np->rx)) | |
288379f0 | 347 | napi_schedule(&np->napi); |
0d160211 JF |
348 | } |
349 | spin_unlock_bh(&np->rx_lock); | |
350 | ||
0b1ab1b8 | 351 | netif_start_queue(dev); |
0d160211 JF |
352 | |
353 | return 0; | |
354 | } | |
355 | ||
356 | static void xennet_tx_buf_gc(struct net_device *dev) | |
357 | { | |
358 | RING_IDX cons, prod; | |
359 | unsigned short id; | |
360 | struct netfront_info *np = netdev_priv(dev); | |
361 | struct sk_buff *skb; | |
362 | ||
363 | BUG_ON(!netif_carrier_ok(dev)); | |
364 | ||
365 | do { | |
366 | prod = np->tx.sring->rsp_prod; | |
367 | rmb(); /* Ensure we see responses up to 'rp'. */ | |
368 | ||
369 | for (cons = np->tx.rsp_cons; cons != prod; cons++) { | |
370 | struct xen_netif_tx_response *txrsp; | |
371 | ||
372 | txrsp = RING_GET_RESPONSE(&np->tx, cons); | |
f942dc25 | 373 | if (txrsp->status == XEN_NETIF_RSP_NULL) |
0d160211 JF |
374 | continue; |
375 | ||
376 | id = txrsp->id; | |
377 | skb = np->tx_skbs[id].skb; | |
378 | if (unlikely(gnttab_query_foreign_access( | |
379 | np->grant_tx_ref[id]) != 0)) { | |
380 | printk(KERN_ALERT "xennet_tx_buf_gc: warning " | |
381 | "-- grant still in use by backend " | |
382 | "domain.\n"); | |
383 | BUG(); | |
384 | } | |
385 | gnttab_end_foreign_access_ref( | |
386 | np->grant_tx_ref[id], GNTMAP_readonly); | |
387 | gnttab_release_grant_reference( | |
388 | &np->gref_tx_head, np->grant_tx_ref[id]); | |
389 | np->grant_tx_ref[id] = GRANT_INVALID_REF; | |
64cee83a | 390 | np->grant_tx_page[id] = NULL; |
0d160211 JF |
391 | add_id_to_freelist(&np->tx_skb_freelist, np->tx_skbs, id); |
392 | dev_kfree_skb_irq(skb); | |
393 | } | |
394 | ||
395 | np->tx.rsp_cons = prod; | |
396 | ||
397 | /* | |
398 | * Set a new event, then check for race with update of tx_cons. | |
399 | * Note that it is essential to schedule a callback, no matter | |
400 | * how few buffers are pending. Even if there is space in the | |
401 | * transmit ring, higher layers may be blocked because too much | |
402 | * data is outstanding: in such cases notification from Xen is | |
403 | * likely to be the only kick that we'll get. | |
404 | */ | |
405 | np->tx.sring->rsp_event = | |
406 | prod + ((np->tx.sring->req_prod - prod) >> 1) + 1; | |
407 | mb(); /* update shared area */ | |
408 | } while ((cons == prod) && (prod != np->tx.sring->rsp_prod)); | |
409 | ||
410 | xennet_maybe_wake_tx(dev); | |
411 | } | |
412 | ||
413 | static void xennet_make_frags(struct sk_buff *skb, struct net_device *dev, | |
414 | struct xen_netif_tx_request *tx) | |
415 | { | |
416 | struct netfront_info *np = netdev_priv(dev); | |
417 | char *data = skb->data; | |
418 | unsigned long mfn; | |
419 | RING_IDX prod = np->tx.req_prod_pvt; | |
420 | int frags = skb_shinfo(skb)->nr_frags; | |
421 | unsigned int offset = offset_in_page(data); | |
422 | unsigned int len = skb_headlen(skb); | |
423 | unsigned int id; | |
424 | grant_ref_t ref; | |
425 | int i; | |
426 | ||
427 | /* While the header overlaps a page boundary (including being | |
428 | larger than a page), split it it into page-sized chunks. */ | |
429 | while (len > PAGE_SIZE - offset) { | |
430 | tx->size = PAGE_SIZE - offset; | |
f942dc25 | 431 | tx->flags |= XEN_NETTXF_more_data; |
0d160211 JF |
432 | len -= tx->size; |
433 | data += tx->size; | |
434 | offset = 0; | |
435 | ||
436 | id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs); | |
437 | np->tx_skbs[id].skb = skb_get(skb); | |
438 | tx = RING_GET_REQUEST(&np->tx, prod++); | |
439 | tx->id = id; | |
440 | ref = gnttab_claim_grant_reference(&np->gref_tx_head); | |
441 | BUG_ON((signed short)ref < 0); | |
442 | ||
443 | mfn = virt_to_mfn(data); | |
444 | gnttab_grant_foreign_access_ref(ref, np->xbdev->otherend_id, | |
445 | mfn, GNTMAP_readonly); | |
446 | ||
64cee83a | 447 | np->grant_tx_page[id] = virt_to_page(data); |
0d160211 JF |
448 | tx->gref = np->grant_tx_ref[id] = ref; |
449 | tx->offset = offset; | |
450 | tx->size = len; | |
451 | tx->flags = 0; | |
452 | } | |
453 | ||
454 | /* Grant backend access to each skb fragment page. */ | |
455 | for (i = 0; i < frags; i++) { | |
456 | skb_frag_t *frag = skb_shinfo(skb)->frags + i; | |
f36c3747 | 457 | struct page *page = skb_frag_page(frag); |
0d160211 | 458 | |
f36c3747 IC |
459 | len = skb_frag_size(frag); |
460 | offset = frag->page_offset; | |
0d160211 | 461 | |
f36c3747 IC |
462 | /* Skip unused frames from start of page */ |
463 | page += offset >> PAGE_SHIFT; | |
464 | offset &= ~PAGE_MASK; | |
0d160211 | 465 | |
f36c3747 IC |
466 | while (len > 0) { |
467 | unsigned long bytes; | |
468 | ||
f36c3747 IC |
469 | bytes = PAGE_SIZE - offset; |
470 | if (bytes > len) | |
471 | bytes = len; | |
472 | ||
473 | tx->flags |= XEN_NETTXF_more_data; | |
474 | ||
475 | id = get_id_from_freelist(&np->tx_skb_freelist, | |
476 | np->tx_skbs); | |
477 | np->tx_skbs[id].skb = skb_get(skb); | |
478 | tx = RING_GET_REQUEST(&np->tx, prod++); | |
479 | tx->id = id; | |
480 | ref = gnttab_claim_grant_reference(&np->gref_tx_head); | |
481 | BUG_ON((signed short)ref < 0); | |
482 | ||
483 | mfn = pfn_to_mfn(page_to_pfn(page)); | |
484 | gnttab_grant_foreign_access_ref(ref, | |
485 | np->xbdev->otherend_id, | |
486 | mfn, GNTMAP_readonly); | |
487 | ||
64cee83a | 488 | np->grant_tx_page[id] = page; |
f36c3747 IC |
489 | tx->gref = np->grant_tx_ref[id] = ref; |
490 | tx->offset = offset; | |
491 | tx->size = bytes; | |
492 | tx->flags = 0; | |
493 | ||
494 | offset += bytes; | |
495 | len -= bytes; | |
496 | ||
497 | /* Next frame */ | |
498 | if (offset == PAGE_SIZE && len) { | |
499 | BUG_ON(!PageCompound(page)); | |
500 | page++; | |
501 | offset = 0; | |
502 | } | |
503 | } | |
0d160211 JF |
504 | } |
505 | ||
506 | np->tx.req_prod_pvt = prod; | |
507 | } | |
508 | ||
f36c3747 IC |
509 | /* |
510 | * Count how many ring slots are required to send the frags of this | |
511 | * skb. Each frag might be a compound page. | |
512 | */ | |
513 | static int xennet_count_skb_frag_slots(struct sk_buff *skb) | |
514 | { | |
515 | int i, frags = skb_shinfo(skb)->nr_frags; | |
516 | int pages = 0; | |
517 | ||
518 | for (i = 0; i < frags; i++) { | |
519 | skb_frag_t *frag = skb_shinfo(skb)->frags + i; | |
520 | unsigned long size = skb_frag_size(frag); | |
521 | unsigned long offset = frag->page_offset; | |
522 | ||
523 | /* Skip unused frames from start of page */ | |
524 | offset &= ~PAGE_MASK; | |
525 | ||
526 | pages += PFN_UP(offset + size); | |
527 | } | |
528 | ||
529 | return pages; | |
530 | } | |
531 | ||
0d160211 JF |
532 | static int xennet_start_xmit(struct sk_buff *skb, struct net_device *dev) |
533 | { | |
534 | unsigned short id; | |
535 | struct netfront_info *np = netdev_priv(dev); | |
e00f85be | 536 | struct netfront_stats *stats = this_cpu_ptr(np->stats); |
0d160211 | 537 | struct xen_netif_tx_request *tx; |
0d160211 JF |
538 | char *data = skb->data; |
539 | RING_IDX i; | |
540 | grant_ref_t ref; | |
541 | unsigned long mfn; | |
542 | int notify; | |
f36c3747 | 543 | int slots; |
0d160211 JF |
544 | unsigned int offset = offset_in_page(data); |
545 | unsigned int len = skb_headlen(skb); | |
cf66f9d4 | 546 | unsigned long flags; |
0d160211 | 547 | |
9ecd1a75 WL |
548 | /* If skb->len is too big for wire format, drop skb and alert |
549 | * user about misconfiguration. | |
550 | */ | |
551 | if (unlikely(skb->len > XEN_NETIF_MAX_TX_SIZE)) { | |
552 | net_alert_ratelimited( | |
553 | "xennet: skb->len = %u, too big for wire format\n", | |
554 | skb->len); | |
555 | goto drop; | |
556 | } | |
557 | ||
f36c3747 IC |
558 | slots = DIV_ROUND_UP(offset + len, PAGE_SIZE) + |
559 | xennet_count_skb_frag_slots(skb); | |
560 | if (unlikely(slots > MAX_SKB_FRAGS + 1)) { | |
561 | net_alert_ratelimited( | |
562 | "xennet: skb rides the rocket: %d slots\n", slots); | |
0d160211 JF |
563 | goto drop; |
564 | } | |
565 | ||
cf66f9d4 | 566 | spin_lock_irqsave(&np->tx_lock, flags); |
0d160211 JF |
567 | |
568 | if (unlikely(!netif_carrier_ok(dev) || | |
f36c3747 | 569 | (slots > 1 && !xennet_can_sg(dev)) || |
fc741216 | 570 | netif_needs_gso(skb, netif_skb_features(skb)))) { |
cf66f9d4 | 571 | spin_unlock_irqrestore(&np->tx_lock, flags); |
0d160211 JF |
572 | goto drop; |
573 | } | |
574 | ||
575 | i = np->tx.req_prod_pvt; | |
576 | ||
577 | id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs); | |
578 | np->tx_skbs[id].skb = skb; | |
579 | ||
580 | tx = RING_GET_REQUEST(&np->tx, i); | |
581 | ||
582 | tx->id = id; | |
583 | ref = gnttab_claim_grant_reference(&np->gref_tx_head); | |
584 | BUG_ON((signed short)ref < 0); | |
585 | mfn = virt_to_mfn(data); | |
586 | gnttab_grant_foreign_access_ref( | |
587 | ref, np->xbdev->otherend_id, mfn, GNTMAP_readonly); | |
64cee83a | 588 | np->grant_tx_page[id] = virt_to_page(data); |
0d160211 JF |
589 | tx->gref = np->grant_tx_ref[id] = ref; |
590 | tx->offset = offset; | |
591 | tx->size = len; | |
0d160211 JF |
592 | |
593 | tx->flags = 0; | |
594 | if (skb->ip_summed == CHECKSUM_PARTIAL) | |
595 | /* local packet? */ | |
f942dc25 | 596 | tx->flags |= XEN_NETTXF_csum_blank | XEN_NETTXF_data_validated; |
0d160211 JF |
597 | else if (skb->ip_summed == CHECKSUM_UNNECESSARY) |
598 | /* remote but checksummed. */ | |
f942dc25 | 599 | tx->flags |= XEN_NETTXF_data_validated; |
0d160211 JF |
600 | |
601 | if (skb_shinfo(skb)->gso_size) { | |
602 | struct xen_netif_extra_info *gso; | |
603 | ||
604 | gso = (struct xen_netif_extra_info *) | |
605 | RING_GET_REQUEST(&np->tx, ++i); | |
606 | ||
e2d617c0 | 607 | tx->flags |= XEN_NETTXF_extra_info; |
0d160211 JF |
608 | |
609 | gso->u.gso.size = skb_shinfo(skb)->gso_size; | |
610 | gso->u.gso.type = XEN_NETIF_GSO_TYPE_TCPV4; | |
611 | gso->u.gso.pad = 0; | |
612 | gso->u.gso.features = 0; | |
613 | ||
614 | gso->type = XEN_NETIF_EXTRA_TYPE_GSO; | |
615 | gso->flags = 0; | |
0d160211 JF |
616 | } |
617 | ||
618 | np->tx.req_prod_pvt = i + 1; | |
619 | ||
620 | xennet_make_frags(skb, dev, tx); | |
621 | tx->size = skb->len; | |
622 | ||
623 | RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np->tx, notify); | |
624 | if (notify) | |
625 | notify_remote_via_irq(np->netdev->irq); | |
626 | ||
e00f85be | 627 | u64_stats_update_begin(&stats->syncp); |
628 | stats->tx_bytes += skb->len; | |
629 | stats->tx_packets++; | |
630 | u64_stats_update_end(&stats->syncp); | |
10a273a6 JF |
631 | |
632 | /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */ | |
0d160211 JF |
633 | xennet_tx_buf_gc(dev); |
634 | ||
635 | if (!netfront_tx_slot_available(np)) | |
636 | netif_stop_queue(dev); | |
637 | ||
cf66f9d4 | 638 | spin_unlock_irqrestore(&np->tx_lock, flags); |
0d160211 | 639 | |
6ed10654 | 640 | return NETDEV_TX_OK; |
0d160211 JF |
641 | |
642 | drop: | |
09f75cd7 | 643 | dev->stats.tx_dropped++; |
0d160211 | 644 | dev_kfree_skb(skb); |
6ed10654 | 645 | return NETDEV_TX_OK; |
0d160211 JF |
646 | } |
647 | ||
648 | static int xennet_close(struct net_device *dev) | |
649 | { | |
650 | struct netfront_info *np = netdev_priv(dev); | |
651 | netif_stop_queue(np->netdev); | |
bea3348e | 652 | napi_disable(&np->napi); |
0d160211 JF |
653 | return 0; |
654 | } | |
655 | ||
0d160211 JF |
656 | static void xennet_move_rx_slot(struct netfront_info *np, struct sk_buff *skb, |
657 | grant_ref_t ref) | |
658 | { | |
659 | int new = xennet_rxidx(np->rx.req_prod_pvt); | |
660 | ||
661 | BUG_ON(np->rx_skbs[new]); | |
662 | np->rx_skbs[new] = skb; | |
663 | np->grant_rx_ref[new] = ref; | |
664 | RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->id = new; | |
665 | RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->gref = ref; | |
666 | np->rx.req_prod_pvt++; | |
667 | } | |
668 | ||
669 | static int xennet_get_extras(struct netfront_info *np, | |
670 | struct xen_netif_extra_info *extras, | |
671 | RING_IDX rp) | |
672 | ||
673 | { | |
674 | struct xen_netif_extra_info *extra; | |
675 | struct device *dev = &np->netdev->dev; | |
676 | RING_IDX cons = np->rx.rsp_cons; | |
677 | int err = 0; | |
678 | ||
679 | do { | |
680 | struct sk_buff *skb; | |
681 | grant_ref_t ref; | |
682 | ||
683 | if (unlikely(cons + 1 == rp)) { | |
684 | if (net_ratelimit()) | |
685 | dev_warn(dev, "Missing extra info\n"); | |
686 | err = -EBADR; | |
687 | break; | |
688 | } | |
689 | ||
690 | extra = (struct xen_netif_extra_info *) | |
691 | RING_GET_RESPONSE(&np->rx, ++cons); | |
692 | ||
693 | if (unlikely(!extra->type || | |
694 | extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) { | |
695 | if (net_ratelimit()) | |
696 | dev_warn(dev, "Invalid extra type: %d\n", | |
697 | extra->type); | |
698 | err = -EINVAL; | |
699 | } else { | |
700 | memcpy(&extras[extra->type - 1], extra, | |
701 | sizeof(*extra)); | |
702 | } | |
703 | ||
704 | skb = xennet_get_rx_skb(np, cons); | |
705 | ref = xennet_get_rx_ref(np, cons); | |
706 | xennet_move_rx_slot(np, skb, ref); | |
707 | } while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE); | |
708 | ||
709 | np->rx.rsp_cons = cons; | |
710 | return err; | |
711 | } | |
712 | ||
713 | static int xennet_get_responses(struct netfront_info *np, | |
714 | struct netfront_rx_info *rinfo, RING_IDX rp, | |
715 | struct sk_buff_head *list) | |
716 | { | |
717 | struct xen_netif_rx_response *rx = &rinfo->rx; | |
718 | struct xen_netif_extra_info *extras = rinfo->extras; | |
719 | struct device *dev = &np->netdev->dev; | |
720 | RING_IDX cons = np->rx.rsp_cons; | |
721 | struct sk_buff *skb = xennet_get_rx_skb(np, cons); | |
722 | grant_ref_t ref = xennet_get_rx_ref(np, cons); | |
723 | int max = MAX_SKB_FRAGS + (rx->status <= RX_COPY_THRESHOLD); | |
7158ff6d | 724 | int slots = 1; |
0d160211 JF |
725 | int err = 0; |
726 | unsigned long ret; | |
727 | ||
f942dc25 | 728 | if (rx->flags & XEN_NETRXF_extra_info) { |
0d160211 JF |
729 | err = xennet_get_extras(np, extras, rp); |
730 | cons = np->rx.rsp_cons; | |
731 | } | |
732 | ||
733 | for (;;) { | |
734 | if (unlikely(rx->status < 0 || | |
735 | rx->offset + rx->status > PAGE_SIZE)) { | |
736 | if (net_ratelimit()) | |
737 | dev_warn(dev, "rx->offset: %x, size: %u\n", | |
738 | rx->offset, rx->status); | |
739 | xennet_move_rx_slot(np, skb, ref); | |
740 | err = -EINVAL; | |
741 | goto next; | |
742 | } | |
743 | ||
744 | /* | |
745 | * This definitely indicates a bug, either in this driver or in | |
746 | * the backend driver. In future this should flag the bad | |
697089dc | 747 | * situation to the system controller to reboot the backend. |
0d160211 JF |
748 | */ |
749 | if (ref == GRANT_INVALID_REF) { | |
750 | if (net_ratelimit()) | |
751 | dev_warn(dev, "Bad rx response id %d.\n", | |
752 | rx->id); | |
753 | err = -EINVAL; | |
754 | goto next; | |
755 | } | |
756 | ||
757 | ret = gnttab_end_foreign_access_ref(ref, 0); | |
758 | BUG_ON(!ret); | |
759 | ||
760 | gnttab_release_grant_reference(&np->gref_rx_head, ref); | |
761 | ||
762 | __skb_queue_tail(list, skb); | |
763 | ||
764 | next: | |
f942dc25 | 765 | if (!(rx->flags & XEN_NETRXF_more_data)) |
0d160211 JF |
766 | break; |
767 | ||
7158ff6d | 768 | if (cons + slots == rp) { |
0d160211 | 769 | if (net_ratelimit()) |
7158ff6d | 770 | dev_warn(dev, "Need more slots\n"); |
0d160211 JF |
771 | err = -ENOENT; |
772 | break; | |
773 | } | |
774 | ||
7158ff6d WL |
775 | rx = RING_GET_RESPONSE(&np->rx, cons + slots); |
776 | skb = xennet_get_rx_skb(np, cons + slots); | |
777 | ref = xennet_get_rx_ref(np, cons + slots); | |
778 | slots++; | |
0d160211 JF |
779 | } |
780 | ||
7158ff6d | 781 | if (unlikely(slots > max)) { |
0d160211 | 782 | if (net_ratelimit()) |
697089dc | 783 | dev_warn(dev, "Too many slots\n"); |
0d160211 JF |
784 | err = -E2BIG; |
785 | } | |
786 | ||
787 | if (unlikely(err)) | |
7158ff6d | 788 | np->rx.rsp_cons = cons + slots; |
0d160211 JF |
789 | |
790 | return err; | |
791 | } | |
792 | ||
793 | static int xennet_set_skb_gso(struct sk_buff *skb, | |
794 | struct xen_netif_extra_info *gso) | |
795 | { | |
796 | if (!gso->u.gso.size) { | |
797 | if (net_ratelimit()) | |
798 | printk(KERN_WARNING "GSO size must not be zero.\n"); | |
799 | return -EINVAL; | |
800 | } | |
801 | ||
802 | /* Currently only TCPv4 S.O. is supported. */ | |
803 | if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4) { | |
804 | if (net_ratelimit()) | |
805 | printk(KERN_WARNING "Bad GSO type %d.\n", gso->u.gso.type); | |
806 | return -EINVAL; | |
807 | } | |
808 | ||
809 | skb_shinfo(skb)->gso_size = gso->u.gso.size; | |
810 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4; | |
811 | ||
812 | /* Header must be checked, and gso_segs computed. */ | |
813 | skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; | |
814 | skb_shinfo(skb)->gso_segs = 0; | |
815 | ||
816 | return 0; | |
817 | } | |
818 | ||
819 | static RING_IDX xennet_fill_frags(struct netfront_info *np, | |
820 | struct sk_buff *skb, | |
821 | struct sk_buff_head *list) | |
822 | { | |
823 | struct skb_shared_info *shinfo = skb_shinfo(skb); | |
0d160211 | 824 | RING_IDX cons = np->rx.rsp_cons; |
0d160211 JF |
825 | struct sk_buff *nskb; |
826 | ||
827 | while ((nskb = __skb_dequeue(list))) { | |
828 | struct xen_netif_rx_response *rx = | |
829 | RING_GET_RESPONSE(&np->rx, ++cons); | |
01c68026 | 830 | skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0]; |
0d160211 | 831 | |
f4b5b99f JB |
832 | if (shinfo->nr_frags == MAX_SKB_FRAGS) { |
833 | unsigned int pull_to = NETFRONT_SKB_CB(skb)->pull_to; | |
0d160211 | 834 | |
f4b5b99f JB |
835 | BUG_ON(pull_to <= skb_headlen(skb)); |
836 | __pskb_pull_tail(skb, pull_to - skb_headlen(skb)); | |
837 | } | |
838 | BUG_ON(shinfo->nr_frags >= MAX_SKB_FRAGS); | |
839 | ||
840 | skb_add_rx_frag(skb, shinfo->nr_frags, skb_frag_page(nfrag), | |
841 | rx->offset, rx->status, PAGE_SIZE); | |
0d160211 JF |
842 | |
843 | skb_shinfo(nskb)->nr_frags = 0; | |
844 | kfree_skb(nskb); | |
0d160211 JF |
845 | } |
846 | ||
0d160211 JF |
847 | return cons; |
848 | } | |
849 | ||
e0ce4af9 | 850 | static int checksum_setup(struct net_device *dev, struct sk_buff *skb) |
0d160211 JF |
851 | { |
852 | struct iphdr *iph; | |
853 | unsigned char *th; | |
854 | int err = -EPROTO; | |
e0ce4af9 IC |
855 | int recalculate_partial_csum = 0; |
856 | ||
857 | /* | |
858 | * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy | |
859 | * peers can fail to set NETRXF_csum_blank when sending a GSO | |
860 | * frame. In this case force the SKB to CHECKSUM_PARTIAL and | |
861 | * recalculate the partial checksum. | |
862 | */ | |
863 | if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) { | |
864 | struct netfront_info *np = netdev_priv(dev); | |
865 | np->rx_gso_checksum_fixup++; | |
866 | skb->ip_summed = CHECKSUM_PARTIAL; | |
867 | recalculate_partial_csum = 1; | |
868 | } | |
869 | ||
870 | /* A non-CHECKSUM_PARTIAL SKB does not require setup. */ | |
871 | if (skb->ip_summed != CHECKSUM_PARTIAL) | |
872 | return 0; | |
0d160211 JF |
873 | |
874 | if (skb->protocol != htons(ETH_P_IP)) | |
875 | goto out; | |
876 | ||
877 | iph = (void *)skb->data; | |
878 | th = skb->data + 4 * iph->ihl; | |
879 | if (th >= skb_tail_pointer(skb)) | |
880 | goto out; | |
881 | ||
882 | skb->csum_start = th - skb->head; | |
883 | switch (iph->protocol) { | |
884 | case IPPROTO_TCP: | |
885 | skb->csum_offset = offsetof(struct tcphdr, check); | |
e0ce4af9 IC |
886 | |
887 | if (recalculate_partial_csum) { | |
888 | struct tcphdr *tcph = (struct tcphdr *)th; | |
889 | tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, | |
890 | skb->len - iph->ihl*4, | |
891 | IPPROTO_TCP, 0); | |
892 | } | |
0d160211 JF |
893 | break; |
894 | case IPPROTO_UDP: | |
895 | skb->csum_offset = offsetof(struct udphdr, check); | |
e0ce4af9 IC |
896 | |
897 | if (recalculate_partial_csum) { | |
898 | struct udphdr *udph = (struct udphdr *)th; | |
899 | udph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, | |
900 | skb->len - iph->ihl*4, | |
901 | IPPROTO_UDP, 0); | |
902 | } | |
0d160211 JF |
903 | break; |
904 | default: | |
905 | if (net_ratelimit()) | |
906 | printk(KERN_ERR "Attempting to checksum a non-" | |
907 | "TCP/UDP packet, dropping a protocol" | |
908 | " %d packet", iph->protocol); | |
909 | goto out; | |
910 | } | |
911 | ||
912 | if ((th + skb->csum_offset + 2) > skb_tail_pointer(skb)) | |
913 | goto out; | |
914 | ||
915 | err = 0; | |
916 | ||
917 | out: | |
918 | return err; | |
919 | } | |
920 | ||
921 | static int handle_incoming_queue(struct net_device *dev, | |
09f75cd7 | 922 | struct sk_buff_head *rxq) |
0d160211 | 923 | { |
e00f85be | 924 | struct netfront_info *np = netdev_priv(dev); |
925 | struct netfront_stats *stats = this_cpu_ptr(np->stats); | |
0d160211 JF |
926 | int packets_dropped = 0; |
927 | struct sk_buff *skb; | |
928 | ||
929 | while ((skb = __skb_dequeue(rxq)) != NULL) { | |
3683243b | 930 | int pull_to = NETFRONT_SKB_CB(skb)->pull_to; |
0d160211 | 931 | |
f4b5b99f JB |
932 | if (pull_to > skb_headlen(skb)) |
933 | __pskb_pull_tail(skb, pull_to - skb_headlen(skb)); | |
0d160211 JF |
934 | |
935 | /* Ethernet work: Delayed to here as it peeks the header. */ | |
936 | skb->protocol = eth_type_trans(skb, dev); | |
937 | ||
e0ce4af9 IC |
938 | if (checksum_setup(dev, skb)) { |
939 | kfree_skb(skb); | |
940 | packets_dropped++; | |
941 | dev->stats.rx_errors++; | |
942 | continue; | |
0d160211 JF |
943 | } |
944 | ||
e00f85be | 945 | u64_stats_update_begin(&stats->syncp); |
946 | stats->rx_packets++; | |
947 | stats->rx_bytes += skb->len; | |
948 | u64_stats_update_end(&stats->syncp); | |
0d160211 JF |
949 | |
950 | /* Pass it up. */ | |
951 | netif_receive_skb(skb); | |
0d160211 JF |
952 | } |
953 | ||
954 | return packets_dropped; | |
955 | } | |
956 | ||
bea3348e | 957 | static int xennet_poll(struct napi_struct *napi, int budget) |
0d160211 | 958 | { |
bea3348e SH |
959 | struct netfront_info *np = container_of(napi, struct netfront_info, napi); |
960 | struct net_device *dev = np->netdev; | |
0d160211 JF |
961 | struct sk_buff *skb; |
962 | struct netfront_rx_info rinfo; | |
963 | struct xen_netif_rx_response *rx = &rinfo.rx; | |
964 | struct xen_netif_extra_info *extras = rinfo.extras; | |
965 | RING_IDX i, rp; | |
bea3348e | 966 | int work_done; |
0d160211 JF |
967 | struct sk_buff_head rxq; |
968 | struct sk_buff_head errq; | |
969 | struct sk_buff_head tmpq; | |
970 | unsigned long flags; | |
0d160211 JF |
971 | int err; |
972 | ||
973 | spin_lock(&np->rx_lock); | |
974 | ||
0d160211 JF |
975 | skb_queue_head_init(&rxq); |
976 | skb_queue_head_init(&errq); | |
977 | skb_queue_head_init(&tmpq); | |
978 | ||
0d160211 JF |
979 | rp = np->rx.sring->rsp_prod; |
980 | rmb(); /* Ensure we see queued responses up to 'rp'. */ | |
981 | ||
982 | i = np->rx.rsp_cons; | |
983 | work_done = 0; | |
984 | while ((i != rp) && (work_done < budget)) { | |
985 | memcpy(rx, RING_GET_RESPONSE(&np->rx, i), sizeof(*rx)); | |
986 | memset(extras, 0, sizeof(rinfo.extras)); | |
987 | ||
988 | err = xennet_get_responses(np, &rinfo, rp, &tmpq); | |
989 | ||
990 | if (unlikely(err)) { | |
991 | err: | |
992 | while ((skb = __skb_dequeue(&tmpq))) | |
993 | __skb_queue_tail(&errq, skb); | |
09f75cd7 | 994 | dev->stats.rx_errors++; |
0d160211 JF |
995 | i = np->rx.rsp_cons; |
996 | continue; | |
997 | } | |
998 | ||
999 | skb = __skb_dequeue(&tmpq); | |
1000 | ||
1001 | if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) { | |
1002 | struct xen_netif_extra_info *gso; | |
1003 | gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1]; | |
1004 | ||
1005 | if (unlikely(xennet_set_skb_gso(skb, gso))) { | |
1006 | __skb_queue_head(&tmpq, skb); | |
1007 | np->rx.rsp_cons += skb_queue_len(&tmpq); | |
1008 | goto err; | |
1009 | } | |
1010 | } | |
1011 | ||
3683243b IC |
1012 | NETFRONT_SKB_CB(skb)->pull_to = rx->status; |
1013 | if (NETFRONT_SKB_CB(skb)->pull_to > RX_COPY_THRESHOLD) | |
1014 | NETFRONT_SKB_CB(skb)->pull_to = RX_COPY_THRESHOLD; | |
0d160211 | 1015 | |
3683243b IC |
1016 | skb_shinfo(skb)->frags[0].page_offset = rx->offset; |
1017 | skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status); | |
1018 | skb->data_len = rx->status; | |
f4b5b99f | 1019 | skb->len += rx->status; |
0d160211 JF |
1020 | |
1021 | i = xennet_fill_frags(np, skb, &tmpq); | |
1022 | ||
f942dc25 | 1023 | if (rx->flags & XEN_NETRXF_csum_blank) |
0d160211 | 1024 | skb->ip_summed = CHECKSUM_PARTIAL; |
f942dc25 | 1025 | else if (rx->flags & XEN_NETRXF_data_validated) |
0d160211 JF |
1026 | skb->ip_summed = CHECKSUM_UNNECESSARY; |
1027 | ||
1028 | __skb_queue_tail(&rxq, skb); | |
1029 | ||
1030 | np->rx.rsp_cons = ++i; | |
1031 | work_done++; | |
1032 | } | |
1033 | ||
56cfe5d0 | 1034 | __skb_queue_purge(&errq); |
0d160211 JF |
1035 | |
1036 | work_done -= handle_incoming_queue(dev, &rxq); | |
1037 | ||
1038 | /* If we get a callback with very few responses, reduce fill target. */ | |
1039 | /* NB. Note exponential increase, linear decrease. */ | |
1040 | if (((np->rx.req_prod_pvt - np->rx.sring->rsp_prod) > | |
1041 | ((3*np->rx_target) / 4)) && | |
1042 | (--np->rx_target < np->rx_min_target)) | |
1043 | np->rx_target = np->rx_min_target; | |
1044 | ||
1045 | xennet_alloc_rx_buffers(dev); | |
1046 | ||
0d160211 | 1047 | if (work_done < budget) { |
bea3348e SH |
1048 | int more_to_do = 0; |
1049 | ||
0d160211 JF |
1050 | local_irq_save(flags); |
1051 | ||
1052 | RING_FINAL_CHECK_FOR_RESPONSES(&np->rx, more_to_do); | |
1053 | if (!more_to_do) | |
288379f0 | 1054 | __napi_complete(napi); |
0d160211 JF |
1055 | |
1056 | local_irq_restore(flags); | |
1057 | } | |
1058 | ||
1059 | spin_unlock(&np->rx_lock); | |
1060 | ||
bea3348e | 1061 | return work_done; |
0d160211 JF |
1062 | } |
1063 | ||
1064 | static int xennet_change_mtu(struct net_device *dev, int mtu) | |
1065 | { | |
9ecd1a75 WL |
1066 | int max = xennet_can_sg(dev) ? |
1067 | XEN_NETIF_MAX_TX_SIZE - MAX_TCP_HEADER : ETH_DATA_LEN; | |
0d160211 JF |
1068 | |
1069 | if (mtu > max) | |
1070 | return -EINVAL; | |
1071 | dev->mtu = mtu; | |
1072 | return 0; | |
1073 | } | |
1074 | ||
e00f85be | 1075 | static struct rtnl_link_stats64 *xennet_get_stats64(struct net_device *dev, |
1076 | struct rtnl_link_stats64 *tot) | |
1077 | { | |
1078 | struct netfront_info *np = netdev_priv(dev); | |
1079 | int cpu; | |
1080 | ||
1081 | for_each_possible_cpu(cpu) { | |
1082 | struct netfront_stats *stats = per_cpu_ptr(np->stats, cpu); | |
1083 | u64 rx_packets, rx_bytes, tx_packets, tx_bytes; | |
1084 | unsigned int start; | |
1085 | ||
1086 | do { | |
1087 | start = u64_stats_fetch_begin_bh(&stats->syncp); | |
1088 | ||
1089 | rx_packets = stats->rx_packets; | |
1090 | tx_packets = stats->tx_packets; | |
1091 | rx_bytes = stats->rx_bytes; | |
1092 | tx_bytes = stats->tx_bytes; | |
1093 | } while (u64_stats_fetch_retry_bh(&stats->syncp, start)); | |
1094 | ||
1095 | tot->rx_packets += rx_packets; | |
1096 | tot->tx_packets += tx_packets; | |
1097 | tot->rx_bytes += rx_bytes; | |
1098 | tot->tx_bytes += tx_bytes; | |
1099 | } | |
1100 | ||
1101 | tot->rx_errors = dev->stats.rx_errors; | |
1102 | tot->tx_dropped = dev->stats.tx_dropped; | |
1103 | ||
1104 | return tot; | |
1105 | } | |
1106 | ||
0d160211 JF |
1107 | static void xennet_release_tx_bufs(struct netfront_info *np) |
1108 | { | |
1109 | struct sk_buff *skb; | |
1110 | int i; | |
1111 | ||
1112 | for (i = 0; i < NET_TX_RING_SIZE; i++) { | |
1113 | /* Skip over entries which are actually freelist references */ | |
1ffb40b8 | 1114 | if (skb_entry_is_link(&np->tx_skbs[i])) |
0d160211 JF |
1115 | continue; |
1116 | ||
1117 | skb = np->tx_skbs[i].skb; | |
64cee83a AL |
1118 | get_page(np->grant_tx_page[i]); |
1119 | gnttab_end_foreign_access(np->grant_tx_ref[i], | |
1120 | GNTMAP_readonly, | |
1121 | (unsigned long)page_address(np->grant_tx_page[i])); | |
1122 | np->grant_tx_page[i] = NULL; | |
0d160211 JF |
1123 | np->grant_tx_ref[i] = GRANT_INVALID_REF; |
1124 | add_id_to_freelist(&np->tx_skb_freelist, np->tx_skbs, i); | |
1125 | dev_kfree_skb_irq(skb); | |
1126 | } | |
1127 | } | |
1128 | ||
1129 | static void xennet_release_rx_bufs(struct netfront_info *np) | |
1130 | { | |
0d160211 JF |
1131 | int id, ref; |
1132 | ||
0d160211 JF |
1133 | spin_lock_bh(&np->rx_lock); |
1134 | ||
1135 | for (id = 0; id < NET_RX_RING_SIZE; id++) { | |
64cee83a AL |
1136 | struct sk_buff *skb; |
1137 | struct page *page; | |
0d160211 JF |
1138 | |
1139 | skb = np->rx_skbs[id]; | |
64cee83a | 1140 | if (!skb) |
0d160211 | 1141 | continue; |
0d160211 | 1142 | |
64cee83a AL |
1143 | ref = np->grant_rx_ref[id]; |
1144 | if (ref == GRANT_INVALID_REF) | |
1145 | continue; | |
0d160211 | 1146 | |
64cee83a | 1147 | page = skb_frag_page(&skb_shinfo(skb)->frags[0]); |
0d160211 | 1148 | |
64cee83a AL |
1149 | /* gnttab_end_foreign_access() needs a page ref until |
1150 | * foreign access is ended (which may be deferred). | |
1151 | */ | |
1152 | get_page(page); | |
1153 | gnttab_end_foreign_access(ref, 0, | |
1154 | (unsigned long)page_address(page)); | |
1155 | np->grant_rx_ref[id] = GRANT_INVALID_REF; | |
0d160211 | 1156 | |
64cee83a | 1157 | kfree_skb(skb); |
0d160211 JF |
1158 | } |
1159 | ||
0d160211 JF |
1160 | spin_unlock_bh(&np->rx_lock); |
1161 | } | |
1162 | ||
1163 | static void xennet_uninit(struct net_device *dev) | |
1164 | { | |
1165 | struct netfront_info *np = netdev_priv(dev); | |
1166 | xennet_release_tx_bufs(np); | |
1167 | xennet_release_rx_bufs(np); | |
1168 | gnttab_free_grant_references(np->gref_tx_head); | |
1169 | gnttab_free_grant_references(np->gref_rx_head); | |
1170 | } | |
1171 | ||
c8f44aff MM |
1172 | static netdev_features_t xennet_fix_features(struct net_device *dev, |
1173 | netdev_features_t features) | |
8f7b01a1 ED |
1174 | { |
1175 | struct netfront_info *np = netdev_priv(dev); | |
1176 | int val; | |
1177 | ||
1178 | if (features & NETIF_F_SG) { | |
1179 | if (xenbus_scanf(XBT_NIL, np->xbdev->otherend, "feature-sg", | |
1180 | "%d", &val) < 0) | |
1181 | val = 0; | |
1182 | ||
1183 | if (!val) | |
1184 | features &= ~NETIF_F_SG; | |
1185 | } | |
1186 | ||
1187 | if (features & NETIF_F_TSO) { | |
1188 | if (xenbus_scanf(XBT_NIL, np->xbdev->otherend, | |
1189 | "feature-gso-tcpv4", "%d", &val) < 0) | |
1190 | val = 0; | |
1191 | ||
1192 | if (!val) | |
1193 | features &= ~NETIF_F_TSO; | |
1194 | } | |
1195 | ||
1196 | return features; | |
1197 | } | |
1198 | ||
c8f44aff MM |
1199 | static int xennet_set_features(struct net_device *dev, |
1200 | netdev_features_t features) | |
8f7b01a1 ED |
1201 | { |
1202 | if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) { | |
1203 | netdev_info(dev, "Reducing MTU because no SG offload"); | |
1204 | dev->mtu = ETH_DATA_LEN; | |
1205 | } | |
1206 | ||
1207 | return 0; | |
1208 | } | |
1209 | ||
cf66f9d4 KRW |
1210 | static irqreturn_t xennet_interrupt(int irq, void *dev_id) |
1211 | { | |
1212 | struct net_device *dev = dev_id; | |
1213 | struct netfront_info *np = netdev_priv(dev); | |
1214 | unsigned long flags; | |
1215 | ||
1216 | spin_lock_irqsave(&np->tx_lock, flags); | |
1217 | ||
1218 | if (likely(netif_carrier_ok(dev))) { | |
1219 | xennet_tx_buf_gc(dev); | |
1220 | /* Under tx_lock: protects access to rx shared-ring indexes. */ | |
1221 | if (RING_HAS_UNCONSUMED_RESPONSES(&np->rx)) | |
1222 | napi_schedule(&np->napi); | |
1223 | } | |
1224 | ||
1225 | spin_unlock_irqrestore(&np->tx_lock, flags); | |
1226 | ||
1227 | return IRQ_HANDLED; | |
1228 | } | |
1229 | ||
1230 | #ifdef CONFIG_NET_POLL_CONTROLLER | |
1231 | static void xennet_poll_controller(struct net_device *dev) | |
1232 | { | |
1233 | xennet_interrupt(0, dev); | |
1234 | } | |
1235 | #endif | |
1236 | ||
0a0b9d2e SH |
1237 | static const struct net_device_ops xennet_netdev_ops = { |
1238 | .ndo_open = xennet_open, | |
1239 | .ndo_uninit = xennet_uninit, | |
1240 | .ndo_stop = xennet_close, | |
1241 | .ndo_start_xmit = xennet_start_xmit, | |
1242 | .ndo_change_mtu = xennet_change_mtu, | |
e00f85be | 1243 | .ndo_get_stats64 = xennet_get_stats64, |
0a0b9d2e SH |
1244 | .ndo_set_mac_address = eth_mac_addr, |
1245 | .ndo_validate_addr = eth_validate_addr, | |
fb507934 MM |
1246 | .ndo_fix_features = xennet_fix_features, |
1247 | .ndo_set_features = xennet_set_features, | |
cf66f9d4 KRW |
1248 | #ifdef CONFIG_NET_POLL_CONTROLLER |
1249 | .ndo_poll_controller = xennet_poll_controller, | |
1250 | #endif | |
0a0b9d2e SH |
1251 | }; |
1252 | ||
8e0e46bb | 1253 | static struct net_device *xennet_create_dev(struct xenbus_device *dev) |
0d160211 JF |
1254 | { |
1255 | int i, err; | |
1256 | struct net_device *netdev; | |
1257 | struct netfront_info *np; | |
1258 | ||
1259 | netdev = alloc_etherdev(sizeof(struct netfront_info)); | |
41de8d4c | 1260 | if (!netdev) |
0d160211 | 1261 | return ERR_PTR(-ENOMEM); |
0d160211 JF |
1262 | |
1263 | np = netdev_priv(netdev); | |
1264 | np->xbdev = dev; | |
1265 | ||
1266 | spin_lock_init(&np->tx_lock); | |
1267 | spin_lock_init(&np->rx_lock); | |
1268 | ||
1269 | skb_queue_head_init(&np->rx_batch); | |
1270 | np->rx_target = RX_DFL_MIN_TARGET; | |
1271 | np->rx_min_target = RX_DFL_MIN_TARGET; | |
1272 | np->rx_max_target = RX_MAX_TARGET; | |
1273 | ||
1274 | init_timer(&np->rx_refill_timer); | |
1275 | np->rx_refill_timer.data = (unsigned long)netdev; | |
1276 | np->rx_refill_timer.function = rx_refill_timeout; | |
1277 | ||
e00f85be | 1278 | err = -ENOMEM; |
1279 | np->stats = alloc_percpu(struct netfront_stats); | |
1280 | if (np->stats == NULL) | |
1281 | goto exit; | |
1282 | ||
0d160211 JF |
1283 | /* Initialise tx_skbs as a free chain containing every entry. */ |
1284 | np->tx_skb_freelist = 0; | |
1285 | for (i = 0; i < NET_TX_RING_SIZE; i++) { | |
1ffb40b8 | 1286 | skb_entry_set_link(&np->tx_skbs[i], i+1); |
0d160211 JF |
1287 | np->grant_tx_ref[i] = GRANT_INVALID_REF; |
1288 | } | |
1289 | ||
1290 | /* Clear out rx_skbs */ | |
1291 | for (i = 0; i < NET_RX_RING_SIZE; i++) { | |
1292 | np->rx_skbs[i] = NULL; | |
1293 | np->grant_rx_ref[i] = GRANT_INVALID_REF; | |
64cee83a | 1294 | np->grant_tx_page[i] = NULL; |
0d160211 JF |
1295 | } |
1296 | ||
1297 | /* A grant for every tx ring slot */ | |
1298 | if (gnttab_alloc_grant_references(TX_MAX_TARGET, | |
1299 | &np->gref_tx_head) < 0) { | |
1300 | printk(KERN_ALERT "#### netfront can't alloc tx grant refs\n"); | |
1301 | err = -ENOMEM; | |
e00f85be | 1302 | goto exit_free_stats; |
0d160211 JF |
1303 | } |
1304 | /* A grant for every rx ring slot */ | |
1305 | if (gnttab_alloc_grant_references(RX_MAX_TARGET, | |
1306 | &np->gref_rx_head) < 0) { | |
1307 | printk(KERN_ALERT "#### netfront can't alloc rx grant refs\n"); | |
1308 | err = -ENOMEM; | |
1309 | goto exit_free_tx; | |
1310 | } | |
1311 | ||
0a0b9d2e SH |
1312 | netdev->netdev_ops = &xennet_netdev_ops; |
1313 | ||
bea3348e | 1314 | netif_napi_add(netdev, &np->napi, xennet_poll, 64); |
fb507934 MM |
1315 | netdev->features = NETIF_F_IP_CSUM | NETIF_F_RXCSUM | |
1316 | NETIF_F_GSO_ROBUST; | |
1317 | netdev->hw_features = NETIF_F_IP_CSUM | NETIF_F_SG | NETIF_F_TSO; | |
0d160211 | 1318 | |
fc3e5941 IC |
1319 | /* |
1320 | * Assume that all hw features are available for now. This set | |
1321 | * will be adjusted by the call to netdev_update_features() in | |
1322 | * xennet_connect() which is the earliest point where we can | |
1323 | * negotiate with the backend regarding supported features. | |
1324 | */ | |
1325 | netdev->features |= netdev->hw_features; | |
1326 | ||
0d160211 | 1327 | SET_ETHTOOL_OPS(netdev, &xennet_ethtool_ops); |
0d160211 JF |
1328 | SET_NETDEV_DEV(netdev, &dev->dev); |
1329 | ||
9ecd1a75 WL |
1330 | netif_set_gso_max_size(netdev, XEN_NETIF_MAX_TX_SIZE - MAX_TCP_HEADER); |
1331 | ||
0d160211 JF |
1332 | np->netdev = netdev; |
1333 | ||
1334 | netif_carrier_off(netdev); | |
1335 | ||
1336 | return netdev; | |
1337 | ||
1338 | exit_free_tx: | |
1339 | gnttab_free_grant_references(np->gref_tx_head); | |
e00f85be | 1340 | exit_free_stats: |
1341 | free_percpu(np->stats); | |
0d160211 JF |
1342 | exit: |
1343 | free_netdev(netdev); | |
1344 | return ERR_PTR(err); | |
1345 | } | |
1346 | ||
1347 | /** | |
1348 | * Entry point to this code when a new device is created. Allocate the basic | |
1349 | * structures and the ring buffers for communication with the backend, and | |
1350 | * inform the backend of the appropriate details for those. | |
1351 | */ | |
8e0e46bb | 1352 | static int netfront_probe(struct xenbus_device *dev, |
1dd06ae8 | 1353 | const struct xenbus_device_id *id) |
0d160211 JF |
1354 | { |
1355 | int err; | |
1356 | struct net_device *netdev; | |
1357 | struct netfront_info *info; | |
1358 | ||
1359 | netdev = xennet_create_dev(dev); | |
1360 | if (IS_ERR(netdev)) { | |
1361 | err = PTR_ERR(netdev); | |
1362 | xenbus_dev_fatal(dev, err, "creating netdev"); | |
1363 | return err; | |
1364 | } | |
1365 | ||
1366 | info = netdev_priv(netdev); | |
1b713e00 | 1367 | dev_set_drvdata(&dev->dev, info); |
0d160211 JF |
1368 | |
1369 | err = register_netdev(info->netdev); | |
1370 | if (err) { | |
1371 | printk(KERN_WARNING "%s: register_netdev err=%d\n", | |
1372 | __func__, err); | |
1373 | goto fail; | |
1374 | } | |
1375 | ||
1376 | err = xennet_sysfs_addif(info->netdev); | |
1377 | if (err) { | |
1378 | unregister_netdev(info->netdev); | |
1379 | printk(KERN_WARNING "%s: add sysfs failed err=%d\n", | |
1380 | __func__, err); | |
1381 | goto fail; | |
1382 | } | |
1383 | ||
1384 | return 0; | |
1385 | ||
1386 | fail: | |
1387 | free_netdev(netdev); | |
1b713e00 | 1388 | dev_set_drvdata(&dev->dev, NULL); |
0d160211 JF |
1389 | return err; |
1390 | } | |
1391 | ||
1392 | static void xennet_end_access(int ref, void *page) | |
1393 | { | |
1394 | /* This frees the page as a side-effect */ | |
1395 | if (ref != GRANT_INVALID_REF) | |
1396 | gnttab_end_foreign_access(ref, 0, (unsigned long)page); | |
1397 | } | |
1398 | ||
1399 | static void xennet_disconnect_backend(struct netfront_info *info) | |
1400 | { | |
1401 | /* Stop old i/f to prevent errors whilst we rebuild the state. */ | |
1402 | spin_lock_bh(&info->rx_lock); | |
1403 | spin_lock_irq(&info->tx_lock); | |
1404 | netif_carrier_off(info->netdev); | |
1405 | spin_unlock_irq(&info->tx_lock); | |
1406 | spin_unlock_bh(&info->rx_lock); | |
1407 | ||
1408 | if (info->netdev->irq) | |
1409 | unbind_from_irqhandler(info->netdev->irq, info->netdev); | |
1410 | info->evtchn = info->netdev->irq = 0; | |
1411 | ||
1412 | /* End access and free the pages */ | |
1413 | xennet_end_access(info->tx_ring_ref, info->tx.sring); | |
1414 | xennet_end_access(info->rx_ring_ref, info->rx.sring); | |
1415 | ||
1416 | info->tx_ring_ref = GRANT_INVALID_REF; | |
1417 | info->rx_ring_ref = GRANT_INVALID_REF; | |
1418 | info->tx.sring = NULL; | |
1419 | info->rx.sring = NULL; | |
1420 | } | |
1421 | ||
1422 | /** | |
1423 | * We are reconnecting to the backend, due to a suspend/resume, or a backend | |
1424 | * driver restart. We tear down our netif structure and recreate it, but | |
1425 | * leave the device-layer structures intact so that this is transparent to the | |
1426 | * rest of the kernel. | |
1427 | */ | |
1428 | static int netfront_resume(struct xenbus_device *dev) | |
1429 | { | |
1b713e00 | 1430 | struct netfront_info *info = dev_get_drvdata(&dev->dev); |
0d160211 JF |
1431 | |
1432 | dev_dbg(&dev->dev, "%s\n", dev->nodename); | |
1433 | ||
1434 | xennet_disconnect_backend(info); | |
1435 | return 0; | |
1436 | } | |
1437 | ||
1438 | static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[]) | |
1439 | { | |
1440 | char *s, *e, *macstr; | |
1441 | int i; | |
1442 | ||
1443 | macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL); | |
1444 | if (IS_ERR(macstr)) | |
1445 | return PTR_ERR(macstr); | |
1446 | ||
1447 | for (i = 0; i < ETH_ALEN; i++) { | |
1448 | mac[i] = simple_strtoul(s, &e, 16); | |
1449 | if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) { | |
1450 | kfree(macstr); | |
1451 | return -ENOENT; | |
1452 | } | |
1453 | s = e+1; | |
1454 | } | |
1455 | ||
1456 | kfree(macstr); | |
1457 | return 0; | |
1458 | } | |
1459 | ||
0d160211 JF |
1460 | static int setup_netfront(struct xenbus_device *dev, struct netfront_info *info) |
1461 | { | |
1462 | struct xen_netif_tx_sring *txs; | |
1463 | struct xen_netif_rx_sring *rxs; | |
1464 | int err; | |
1465 | struct net_device *netdev = info->netdev; | |
1466 | ||
1467 | info->tx_ring_ref = GRANT_INVALID_REF; | |
1468 | info->rx_ring_ref = GRANT_INVALID_REF; | |
1469 | info->rx.sring = NULL; | |
1470 | info->tx.sring = NULL; | |
1471 | netdev->irq = 0; | |
1472 | ||
1473 | err = xen_net_read_mac(dev, netdev->dev_addr); | |
1474 | if (err) { | |
1475 | xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename); | |
1476 | goto fail; | |
1477 | } | |
1478 | ||
a144ff09 | 1479 | txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH); |
0d160211 JF |
1480 | if (!txs) { |
1481 | err = -ENOMEM; | |
1482 | xenbus_dev_fatal(dev, err, "allocating tx ring page"); | |
1483 | goto fail; | |
1484 | } | |
1485 | SHARED_RING_INIT(txs); | |
1486 | FRONT_RING_INIT(&info->tx, txs, PAGE_SIZE); | |
1487 | ||
1488 | err = xenbus_grant_ring(dev, virt_to_mfn(txs)); | |
1489 | if (err < 0) { | |
1490 | free_page((unsigned long)txs); | |
1491 | goto fail; | |
1492 | } | |
1493 | ||
1494 | info->tx_ring_ref = err; | |
a144ff09 | 1495 | rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH); |
0d160211 JF |
1496 | if (!rxs) { |
1497 | err = -ENOMEM; | |
1498 | xenbus_dev_fatal(dev, err, "allocating rx ring page"); | |
1499 | goto fail; | |
1500 | } | |
1501 | SHARED_RING_INIT(rxs); | |
1502 | FRONT_RING_INIT(&info->rx, rxs, PAGE_SIZE); | |
1503 | ||
1504 | err = xenbus_grant_ring(dev, virt_to_mfn(rxs)); | |
1505 | if (err < 0) { | |
1506 | free_page((unsigned long)rxs); | |
1507 | goto fail; | |
1508 | } | |
1509 | info->rx_ring_ref = err; | |
1510 | ||
1511 | err = xenbus_alloc_evtchn(dev, &info->evtchn); | |
1512 | if (err) | |
1513 | goto fail; | |
1514 | ||
1515 | err = bind_evtchn_to_irqhandler(info->evtchn, xennet_interrupt, | |
ab392d2d | 1516 | 0, netdev->name, netdev); |
0d160211 JF |
1517 | if (err < 0) |
1518 | goto fail; | |
1519 | netdev->irq = err; | |
1520 | return 0; | |
1521 | ||
1522 | fail: | |
1523 | return err; | |
1524 | } | |
1525 | ||
1526 | /* Common code used when first setting up, and when resuming. */ | |
f502bf2b | 1527 | static int talk_to_netback(struct xenbus_device *dev, |
0d160211 JF |
1528 | struct netfront_info *info) |
1529 | { | |
1530 | const char *message; | |
1531 | struct xenbus_transaction xbt; | |
1532 | int err; | |
1533 | ||
1534 | /* Create shared ring, alloc event channel. */ | |
1535 | err = setup_netfront(dev, info); | |
1536 | if (err) | |
1537 | goto out; | |
1538 | ||
1539 | again: | |
1540 | err = xenbus_transaction_start(&xbt); | |
1541 | if (err) { | |
1542 | xenbus_dev_fatal(dev, err, "starting transaction"); | |
1543 | goto destroy_ring; | |
1544 | } | |
1545 | ||
1546 | err = xenbus_printf(xbt, dev->nodename, "tx-ring-ref", "%u", | |
1547 | info->tx_ring_ref); | |
1548 | if (err) { | |
1549 | message = "writing tx ring-ref"; | |
1550 | goto abort_transaction; | |
1551 | } | |
1552 | err = xenbus_printf(xbt, dev->nodename, "rx-ring-ref", "%u", | |
1553 | info->rx_ring_ref); | |
1554 | if (err) { | |
1555 | message = "writing rx ring-ref"; | |
1556 | goto abort_transaction; | |
1557 | } | |
1558 | err = xenbus_printf(xbt, dev->nodename, | |
1559 | "event-channel", "%u", info->evtchn); | |
1560 | if (err) { | |
1561 | message = "writing event-channel"; | |
1562 | goto abort_transaction; | |
1563 | } | |
1564 | ||
1565 | err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u", | |
1566 | 1); | |
1567 | if (err) { | |
1568 | message = "writing request-rx-copy"; | |
1569 | goto abort_transaction; | |
1570 | } | |
1571 | ||
1572 | err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1); | |
1573 | if (err) { | |
1574 | message = "writing feature-rx-notify"; | |
1575 | goto abort_transaction; | |
1576 | } | |
1577 | ||
1578 | err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1); | |
1579 | if (err) { | |
1580 | message = "writing feature-sg"; | |
1581 | goto abort_transaction; | |
1582 | } | |
1583 | ||
1584 | err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1); | |
1585 | if (err) { | |
1586 | message = "writing feature-gso-tcpv4"; | |
1587 | goto abort_transaction; | |
1588 | } | |
1589 | ||
1590 | err = xenbus_transaction_end(xbt, 0); | |
1591 | if (err) { | |
1592 | if (err == -EAGAIN) | |
1593 | goto again; | |
1594 | xenbus_dev_fatal(dev, err, "completing transaction"); | |
1595 | goto destroy_ring; | |
1596 | } | |
1597 | ||
1598 | return 0; | |
1599 | ||
1600 | abort_transaction: | |
1601 | xenbus_transaction_end(xbt, 1); | |
1602 | xenbus_dev_fatal(dev, err, "%s", message); | |
1603 | destroy_ring: | |
1604 | xennet_disconnect_backend(info); | |
1605 | out: | |
1606 | return err; | |
1607 | } | |
1608 | ||
0d160211 JF |
1609 | static int xennet_connect(struct net_device *dev) |
1610 | { | |
1611 | struct netfront_info *np = netdev_priv(dev); | |
1612 | int i, requeue_idx, err; | |
1613 | struct sk_buff *skb; | |
1614 | grant_ref_t ref; | |
1615 | struct xen_netif_rx_request *req; | |
1616 | unsigned int feature_rx_copy; | |
1617 | ||
1618 | err = xenbus_scanf(XBT_NIL, np->xbdev->otherend, | |
1619 | "feature-rx-copy", "%u", &feature_rx_copy); | |
1620 | if (err != 1) | |
1621 | feature_rx_copy = 0; | |
1622 | ||
1623 | if (!feature_rx_copy) { | |
1624 | dev_info(&dev->dev, | |
898eb71c | 1625 | "backend does not support copying receive path\n"); |
0d160211 JF |
1626 | return -ENODEV; |
1627 | } | |
1628 | ||
f502bf2b | 1629 | err = talk_to_netback(np->xbdev, np); |
0d160211 JF |
1630 | if (err) |
1631 | return err; | |
1632 | ||
1ba37c51 | 1633 | rtnl_lock(); |
fb507934 | 1634 | netdev_update_features(dev); |
1ba37c51 | 1635 | rtnl_unlock(); |
0d160211 JF |
1636 | |
1637 | spin_lock_bh(&np->rx_lock); | |
1638 | spin_lock_irq(&np->tx_lock); | |
1639 | ||
1640 | /* Step 1: Discard all pending TX packet fragments. */ | |
1641 | xennet_release_tx_bufs(np); | |
1642 | ||
1643 | /* Step 2: Rebuild the RX buffer freelist and the RX ring itself. */ | |
1644 | for (requeue_idx = 0, i = 0; i < NET_RX_RING_SIZE; i++) { | |
01c68026 IC |
1645 | skb_frag_t *frag; |
1646 | const struct page *page; | |
0d160211 JF |
1647 | if (!np->rx_skbs[i]) |
1648 | continue; | |
1649 | ||
1650 | skb = np->rx_skbs[requeue_idx] = xennet_get_rx_skb(np, i); | |
1651 | ref = np->grant_rx_ref[requeue_idx] = xennet_get_rx_ref(np, i); | |
1652 | req = RING_GET_REQUEST(&np->rx, requeue_idx); | |
1653 | ||
01c68026 IC |
1654 | frag = &skb_shinfo(skb)->frags[0]; |
1655 | page = skb_frag_page(frag); | |
0d160211 JF |
1656 | gnttab_grant_foreign_access_ref( |
1657 | ref, np->xbdev->otherend_id, | |
01c68026 | 1658 | pfn_to_mfn(page_to_pfn(page)), |
0d160211 JF |
1659 | 0); |
1660 | req->gref = ref; | |
1661 | req->id = requeue_idx; | |
1662 | ||
1663 | requeue_idx++; | |
1664 | } | |
1665 | ||
1666 | np->rx.req_prod_pvt = requeue_idx; | |
1667 | ||
1668 | /* | |
1669 | * Step 3: All public and private state should now be sane. Get | |
1670 | * ready to start sending and receiving packets and give the driver | |
1671 | * domain a kick because we've probably just requeued some | |
1672 | * packets. | |
1673 | */ | |
1674 | netif_carrier_on(np->netdev); | |
1675 | notify_remote_via_irq(np->netdev->irq); | |
1676 | xennet_tx_buf_gc(dev); | |
1677 | xennet_alloc_rx_buffers(dev); | |
1678 | ||
1679 | spin_unlock_irq(&np->tx_lock); | |
1680 | spin_unlock_bh(&np->rx_lock); | |
1681 | ||
1682 | return 0; | |
1683 | } | |
1684 | ||
1685 | /** | |
1686 | * Callback received when the backend's state changes. | |
1687 | */ | |
f502bf2b | 1688 | static void netback_changed(struct xenbus_device *dev, |
0d160211 JF |
1689 | enum xenbus_state backend_state) |
1690 | { | |
1b713e00 | 1691 | struct netfront_info *np = dev_get_drvdata(&dev->dev); |
0d160211 JF |
1692 | struct net_device *netdev = np->netdev; |
1693 | ||
1694 | dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state)); | |
1695 | ||
1696 | switch (backend_state) { | |
1697 | case XenbusStateInitialising: | |
1698 | case XenbusStateInitialised: | |
b78c9512 NI |
1699 | case XenbusStateReconfiguring: |
1700 | case XenbusStateReconfigured: | |
0d160211 JF |
1701 | case XenbusStateUnknown: |
1702 | case XenbusStateClosed: | |
1703 | break; | |
1704 | ||
1705 | case XenbusStateInitWait: | |
1706 | if (dev->state != XenbusStateInitialising) | |
1707 | break; | |
1708 | if (xennet_connect(netdev) != 0) | |
1709 | break; | |
1710 | xenbus_switch_state(dev, XenbusStateConnected); | |
08e34eb1 LE |
1711 | break; |
1712 | ||
1713 | case XenbusStateConnected: | |
ee89bab1 | 1714 | netdev_notify_peers(netdev); |
0d160211 JF |
1715 | break; |
1716 | ||
1717 | case XenbusStateClosing: | |
1718 | xenbus_frontend_closed(dev); | |
1719 | break; | |
1720 | } | |
1721 | } | |
1722 | ||
e0ce4af9 IC |
1723 | static const struct xennet_stat { |
1724 | char name[ETH_GSTRING_LEN]; | |
1725 | u16 offset; | |
1726 | } xennet_stats[] = { | |
1727 | { | |
1728 | "rx_gso_checksum_fixup", | |
1729 | offsetof(struct netfront_info, rx_gso_checksum_fixup) | |
1730 | }, | |
1731 | }; | |
1732 | ||
1733 | static int xennet_get_sset_count(struct net_device *dev, int string_set) | |
1734 | { | |
1735 | switch (string_set) { | |
1736 | case ETH_SS_STATS: | |
1737 | return ARRAY_SIZE(xennet_stats); | |
1738 | default: | |
1739 | return -EINVAL; | |
1740 | } | |
1741 | } | |
1742 | ||
1743 | static void xennet_get_ethtool_stats(struct net_device *dev, | |
1744 | struct ethtool_stats *stats, u64 * data) | |
1745 | { | |
1746 | void *np = netdev_priv(dev); | |
1747 | int i; | |
1748 | ||
1749 | for (i = 0; i < ARRAY_SIZE(xennet_stats); i++) | |
48f26d51 | 1750 | data[i] = *(unsigned long *)(np + xennet_stats[i].offset); |
e0ce4af9 IC |
1751 | } |
1752 | ||
1753 | static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data) | |
1754 | { | |
1755 | int i; | |
1756 | ||
1757 | switch (stringset) { | |
1758 | case ETH_SS_STATS: | |
1759 | for (i = 0; i < ARRAY_SIZE(xennet_stats); i++) | |
1760 | memcpy(data + i * ETH_GSTRING_LEN, | |
1761 | xennet_stats[i].name, ETH_GSTRING_LEN); | |
1762 | break; | |
1763 | } | |
1764 | } | |
1765 | ||
0fc0b732 | 1766 | static const struct ethtool_ops xennet_ethtool_ops = |
0d160211 | 1767 | { |
0d160211 | 1768 | .get_link = ethtool_op_get_link, |
e0ce4af9 IC |
1769 | |
1770 | .get_sset_count = xennet_get_sset_count, | |
1771 | .get_ethtool_stats = xennet_get_ethtool_stats, | |
1772 | .get_strings = xennet_get_strings, | |
0d160211 JF |
1773 | }; |
1774 | ||
1775 | #ifdef CONFIG_SYSFS | |
1776 | static ssize_t show_rxbuf_min(struct device *dev, | |
1777 | struct device_attribute *attr, char *buf) | |
1778 | { | |
1779 | struct net_device *netdev = to_net_dev(dev); | |
1780 | struct netfront_info *info = netdev_priv(netdev); | |
1781 | ||
1782 | return sprintf(buf, "%u\n", info->rx_min_target); | |
1783 | } | |
1784 | ||
1785 | static ssize_t store_rxbuf_min(struct device *dev, | |
1786 | struct device_attribute *attr, | |
1787 | const char *buf, size_t len) | |
1788 | { | |
1789 | struct net_device *netdev = to_net_dev(dev); | |
1790 | struct netfront_info *np = netdev_priv(netdev); | |
1791 | char *endp; | |
1792 | unsigned long target; | |
1793 | ||
1794 | if (!capable(CAP_NET_ADMIN)) | |
1795 | return -EPERM; | |
1796 | ||
1797 | target = simple_strtoul(buf, &endp, 0); | |
1798 | if (endp == buf) | |
1799 | return -EBADMSG; | |
1800 | ||
1801 | if (target < RX_MIN_TARGET) | |
1802 | target = RX_MIN_TARGET; | |
1803 | if (target > RX_MAX_TARGET) | |
1804 | target = RX_MAX_TARGET; | |
1805 | ||
1806 | spin_lock_bh(&np->rx_lock); | |
1807 | if (target > np->rx_max_target) | |
1808 | np->rx_max_target = target; | |
1809 | np->rx_min_target = target; | |
1810 | if (target > np->rx_target) | |
1811 | np->rx_target = target; | |
1812 | ||
1813 | xennet_alloc_rx_buffers(netdev); | |
1814 | ||
1815 | spin_unlock_bh(&np->rx_lock); | |
1816 | return len; | |
1817 | } | |
1818 | ||
1819 | static ssize_t show_rxbuf_max(struct device *dev, | |
1820 | struct device_attribute *attr, char *buf) | |
1821 | { | |
1822 | struct net_device *netdev = to_net_dev(dev); | |
1823 | struct netfront_info *info = netdev_priv(netdev); | |
1824 | ||
1825 | return sprintf(buf, "%u\n", info->rx_max_target); | |
1826 | } | |
1827 | ||
1828 | static ssize_t store_rxbuf_max(struct device *dev, | |
1829 | struct device_attribute *attr, | |
1830 | const char *buf, size_t len) | |
1831 | { | |
1832 | struct net_device *netdev = to_net_dev(dev); | |
1833 | struct netfront_info *np = netdev_priv(netdev); | |
1834 | char *endp; | |
1835 | unsigned long target; | |
1836 | ||
1837 | if (!capable(CAP_NET_ADMIN)) | |
1838 | return -EPERM; | |
1839 | ||
1840 | target = simple_strtoul(buf, &endp, 0); | |
1841 | if (endp == buf) | |
1842 | return -EBADMSG; | |
1843 | ||
1844 | if (target < RX_MIN_TARGET) | |
1845 | target = RX_MIN_TARGET; | |
1846 | if (target > RX_MAX_TARGET) | |
1847 | target = RX_MAX_TARGET; | |
1848 | ||
1849 | spin_lock_bh(&np->rx_lock); | |
1850 | if (target < np->rx_min_target) | |
1851 | np->rx_min_target = target; | |
1852 | np->rx_max_target = target; | |
1853 | if (target < np->rx_target) | |
1854 | np->rx_target = target; | |
1855 | ||
1856 | xennet_alloc_rx_buffers(netdev); | |
1857 | ||
1858 | spin_unlock_bh(&np->rx_lock); | |
1859 | return len; | |
1860 | } | |
1861 | ||
1862 | static ssize_t show_rxbuf_cur(struct device *dev, | |
1863 | struct device_attribute *attr, char *buf) | |
1864 | { | |
1865 | struct net_device *netdev = to_net_dev(dev); | |
1866 | struct netfront_info *info = netdev_priv(netdev); | |
1867 | ||
1868 | return sprintf(buf, "%u\n", info->rx_target); | |
1869 | } | |
1870 | ||
1871 | static struct device_attribute xennet_attrs[] = { | |
1872 | __ATTR(rxbuf_min, S_IRUGO|S_IWUSR, show_rxbuf_min, store_rxbuf_min), | |
1873 | __ATTR(rxbuf_max, S_IRUGO|S_IWUSR, show_rxbuf_max, store_rxbuf_max), | |
1874 | __ATTR(rxbuf_cur, S_IRUGO, show_rxbuf_cur, NULL), | |
1875 | }; | |
1876 | ||
1877 | static int xennet_sysfs_addif(struct net_device *netdev) | |
1878 | { | |
1879 | int i; | |
1880 | int err; | |
1881 | ||
1882 | for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++) { | |
1883 | err = device_create_file(&netdev->dev, | |
1884 | &xennet_attrs[i]); | |
1885 | if (err) | |
1886 | goto fail; | |
1887 | } | |
1888 | return 0; | |
1889 | ||
1890 | fail: | |
1891 | while (--i >= 0) | |
1892 | device_remove_file(&netdev->dev, &xennet_attrs[i]); | |
1893 | return err; | |
1894 | } | |
1895 | ||
1896 | static void xennet_sysfs_delif(struct net_device *netdev) | |
1897 | { | |
1898 | int i; | |
1899 | ||
1900 | for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++) | |
1901 | device_remove_file(&netdev->dev, &xennet_attrs[i]); | |
1902 | } | |
1903 | ||
1904 | #endif /* CONFIG_SYSFS */ | |
1905 | ||
73db144b | 1906 | static const struct xenbus_device_id netfront_ids[] = { |
0d160211 JF |
1907 | { "vif" }, |
1908 | { "" } | |
1909 | }; | |
1910 | ||
1911 | ||
8e0e46bb | 1912 | static int xennet_remove(struct xenbus_device *dev) |
0d160211 | 1913 | { |
1b713e00 | 1914 | struct netfront_info *info = dev_get_drvdata(&dev->dev); |
0d160211 JF |
1915 | |
1916 | dev_dbg(&dev->dev, "%s\n", dev->nodename); | |
1917 | ||
0d160211 JF |
1918 | xennet_disconnect_backend(info); |
1919 | ||
0d160211 JF |
1920 | xennet_sysfs_delif(info->netdev); |
1921 | ||
6bc96d04 IC |
1922 | unregister_netdev(info->netdev); |
1923 | ||
1924 | del_timer_sync(&info->rx_refill_timer); | |
1925 | ||
e00f85be | 1926 | free_percpu(info->stats); |
1927 | ||
0d160211 JF |
1928 | free_netdev(info->netdev); |
1929 | ||
1930 | return 0; | |
1931 | } | |
1932 | ||
73db144b | 1933 | static DEFINE_XENBUS_DRIVER(netfront, , |
0d160211 | 1934 | .probe = netfront_probe, |
8e0e46bb | 1935 | .remove = xennet_remove, |
0d160211 | 1936 | .resume = netfront_resume, |
f502bf2b | 1937 | .otherend_changed = netback_changed, |
73db144b | 1938 | ); |
0d160211 JF |
1939 | |
1940 | static int __init netif_init(void) | |
1941 | { | |
6e833587 | 1942 | if (!xen_domain()) |
0d160211 JF |
1943 | return -ENODEV; |
1944 | ||
e95ae5a4 | 1945 | if (xen_hvm_domain() && !xen_platform_pci_unplug) |
b9136d20 IM |
1946 | return -ENODEV; |
1947 | ||
0d160211 JF |
1948 | printk(KERN_INFO "Initialising Xen virtual ethernet driver.\n"); |
1949 | ||
ffb78a26 | 1950 | return xenbus_register_frontend(&netfront_driver); |
0d160211 JF |
1951 | } |
1952 | module_init(netif_init); | |
1953 | ||
1954 | ||
1955 | static void __exit netif_exit(void) | |
1956 | { | |
ffb78a26 | 1957 | xenbus_unregister_driver(&netfront_driver); |
0d160211 JF |
1958 | } |
1959 | module_exit(netif_exit); | |
1960 | ||
1961 | MODULE_DESCRIPTION("Xen virtual network device frontend"); | |
1962 | MODULE_LICENSE("GPL"); | |
d2f0c52b | 1963 | MODULE_ALIAS("xen:vif"); |
4f93f09b | 1964 | MODULE_ALIAS("xennet"); |