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> | |
5a0e3ad6 | 43 | #include <linux/slab.h> |
0d160211 JF |
44 | #include <net/ip.h> |
45 | ||
1ccbf534 | 46 | #include <xen/xen.h> |
0d160211 JF |
47 | #include <xen/xenbus.h> |
48 | #include <xen/events.h> | |
49 | #include <xen/page.h> | |
50 | #include <xen/grant_table.h> | |
51 | ||
52 | #include <xen/interface/io/netif.h> | |
53 | #include <xen/interface/memory.h> | |
54 | #include <xen/interface/grant_table.h> | |
55 | ||
0fc0b732 | 56 | static const struct ethtool_ops xennet_ethtool_ops; |
0d160211 JF |
57 | |
58 | struct netfront_cb { | |
59 | struct page *page; | |
60 | unsigned offset; | |
61 | }; | |
62 | ||
63 | #define NETFRONT_SKB_CB(skb) ((struct netfront_cb *)((skb)->cb)) | |
64 | ||
65 | #define RX_COPY_THRESHOLD 256 | |
66 | ||
67 | #define GRANT_INVALID_REF 0 | |
68 | ||
667c78af JF |
69 | #define NET_TX_RING_SIZE __CONST_RING_SIZE(xen_netif_tx, PAGE_SIZE) |
70 | #define NET_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, PAGE_SIZE) | |
0d160211 JF |
71 | #define TX_MAX_TARGET min_t(int, NET_RX_RING_SIZE, 256) |
72 | ||
e00f85be | 73 | struct netfront_stats { |
74 | u64 rx_packets; | |
75 | u64 tx_packets; | |
76 | u64 rx_bytes; | |
77 | u64 tx_bytes; | |
78 | struct u64_stats_sync syncp; | |
79 | }; | |
80 | ||
0d160211 JF |
81 | struct netfront_info { |
82 | struct list_head list; | |
83 | struct net_device *netdev; | |
84 | ||
bea3348e | 85 | struct napi_struct napi; |
0d160211 | 86 | |
0d160211 | 87 | unsigned int evtchn; |
84284d3c | 88 | struct xenbus_device *xbdev; |
0d160211 | 89 | |
84284d3c JF |
90 | spinlock_t tx_lock; |
91 | struct xen_netif_tx_front_ring tx; | |
92 | int tx_ring_ref; | |
0d160211 JF |
93 | |
94 | /* | |
95 | * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries | |
96 | * are linked from tx_skb_freelist through skb_entry.link. | |
97 | * | |
98 | * NB. Freelist index entries are always going to be less than | |
99 | * PAGE_OFFSET, whereas pointers to skbs will always be equal or | |
100 | * greater than PAGE_OFFSET: we use this property to distinguish | |
101 | * them. | |
102 | */ | |
103 | union skb_entry { | |
104 | struct sk_buff *skb; | |
1ffb40b8 | 105 | unsigned long link; |
0d160211 JF |
106 | } tx_skbs[NET_TX_RING_SIZE]; |
107 | grant_ref_t gref_tx_head; | |
108 | grant_ref_t grant_tx_ref[NET_TX_RING_SIZE]; | |
109 | unsigned tx_skb_freelist; | |
110 | ||
84284d3c JF |
111 | spinlock_t rx_lock ____cacheline_aligned_in_smp; |
112 | struct xen_netif_rx_front_ring rx; | |
113 | int rx_ring_ref; | |
114 | ||
115 | /* Receive-ring batched refills. */ | |
116 | #define RX_MIN_TARGET 8 | |
117 | #define RX_DFL_MIN_TARGET 64 | |
118 | #define RX_MAX_TARGET min_t(int, NET_RX_RING_SIZE, 256) | |
119 | unsigned rx_min_target, rx_max_target, rx_target; | |
120 | struct sk_buff_head rx_batch; | |
121 | ||
122 | struct timer_list rx_refill_timer; | |
123 | ||
0d160211 JF |
124 | struct sk_buff *rx_skbs[NET_RX_RING_SIZE]; |
125 | grant_ref_t gref_rx_head; | |
126 | grant_ref_t grant_rx_ref[NET_RX_RING_SIZE]; | |
127 | ||
0d160211 JF |
128 | unsigned long rx_pfn_array[NET_RX_RING_SIZE]; |
129 | struct multicall_entry rx_mcl[NET_RX_RING_SIZE+1]; | |
130 | struct mmu_update rx_mmu[NET_RX_RING_SIZE]; | |
e0ce4af9 IC |
131 | |
132 | /* Statistics */ | |
e00f85be | 133 | struct netfront_stats __percpu *stats; |
134 | ||
e9a799ea | 135 | unsigned long rx_gso_checksum_fixup; |
0d160211 JF |
136 | }; |
137 | ||
138 | struct netfront_rx_info { | |
139 | struct xen_netif_rx_response rx; | |
140 | struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1]; | |
141 | }; | |
142 | ||
1ffb40b8 IY |
143 | static void skb_entry_set_link(union skb_entry *list, unsigned short id) |
144 | { | |
145 | list->link = id; | |
146 | } | |
147 | ||
148 | static int skb_entry_is_link(const union skb_entry *list) | |
149 | { | |
150 | BUILD_BUG_ON(sizeof(list->skb) != sizeof(list->link)); | |
807540ba | 151 | return (unsigned long)list->skb < PAGE_OFFSET; |
1ffb40b8 IY |
152 | } |
153 | ||
0d160211 JF |
154 | /* |
155 | * Access macros for acquiring freeing slots in tx_skbs[]. | |
156 | */ | |
157 | ||
158 | static void add_id_to_freelist(unsigned *head, union skb_entry *list, | |
159 | unsigned short id) | |
160 | { | |
1ffb40b8 | 161 | skb_entry_set_link(&list[id], *head); |
0d160211 JF |
162 | *head = id; |
163 | } | |
164 | ||
165 | static unsigned short get_id_from_freelist(unsigned *head, | |
166 | union skb_entry *list) | |
167 | { | |
168 | unsigned int id = *head; | |
169 | *head = list[id].link; | |
170 | return id; | |
171 | } | |
172 | ||
173 | static int xennet_rxidx(RING_IDX idx) | |
174 | { | |
175 | return idx & (NET_RX_RING_SIZE - 1); | |
176 | } | |
177 | ||
178 | static struct sk_buff *xennet_get_rx_skb(struct netfront_info *np, | |
179 | RING_IDX ri) | |
180 | { | |
181 | int i = xennet_rxidx(ri); | |
182 | struct sk_buff *skb = np->rx_skbs[i]; | |
183 | np->rx_skbs[i] = NULL; | |
184 | return skb; | |
185 | } | |
186 | ||
187 | static grant_ref_t xennet_get_rx_ref(struct netfront_info *np, | |
188 | RING_IDX ri) | |
189 | { | |
190 | int i = xennet_rxidx(ri); | |
191 | grant_ref_t ref = np->grant_rx_ref[i]; | |
192 | np->grant_rx_ref[i] = GRANT_INVALID_REF; | |
193 | return ref; | |
194 | } | |
195 | ||
196 | #ifdef CONFIG_SYSFS | |
197 | static int xennet_sysfs_addif(struct net_device *netdev); | |
198 | static void xennet_sysfs_delif(struct net_device *netdev); | |
199 | #else /* !CONFIG_SYSFS */ | |
200 | #define xennet_sysfs_addif(dev) (0) | |
201 | #define xennet_sysfs_delif(dev) do { } while (0) | |
202 | #endif | |
203 | ||
204 | static int xennet_can_sg(struct net_device *dev) | |
205 | { | |
206 | return dev->features & NETIF_F_SG; | |
207 | } | |
208 | ||
209 | ||
210 | static void rx_refill_timeout(unsigned long data) | |
211 | { | |
212 | struct net_device *dev = (struct net_device *)data; | |
bea3348e | 213 | struct netfront_info *np = netdev_priv(dev); |
288379f0 | 214 | napi_schedule(&np->napi); |
0d160211 JF |
215 | } |
216 | ||
217 | static int netfront_tx_slot_available(struct netfront_info *np) | |
218 | { | |
807540ba ED |
219 | return (np->tx.req_prod_pvt - np->tx.rsp_cons) < |
220 | (TX_MAX_TARGET - MAX_SKB_FRAGS - 2); | |
0d160211 JF |
221 | } |
222 | ||
223 | static void xennet_maybe_wake_tx(struct net_device *dev) | |
224 | { | |
225 | struct netfront_info *np = netdev_priv(dev); | |
226 | ||
227 | if (unlikely(netif_queue_stopped(dev)) && | |
228 | netfront_tx_slot_available(np) && | |
229 | likely(netif_running(dev))) | |
230 | netif_wake_queue(dev); | |
231 | } | |
232 | ||
233 | static void xennet_alloc_rx_buffers(struct net_device *dev) | |
234 | { | |
235 | unsigned short id; | |
236 | struct netfront_info *np = netdev_priv(dev); | |
237 | struct sk_buff *skb; | |
238 | struct page *page; | |
239 | int i, batch_target, notify; | |
240 | RING_IDX req_prod = np->rx.req_prod_pvt; | |
0d160211 JF |
241 | grant_ref_t ref; |
242 | unsigned long pfn; | |
243 | void *vaddr; | |
0d160211 JF |
244 | struct xen_netif_rx_request *req; |
245 | ||
246 | if (unlikely(!netif_carrier_ok(dev))) | |
247 | return; | |
248 | ||
249 | /* | |
250 | * Allocate skbuffs greedily, even though we batch updates to the | |
251 | * receive ring. This creates a less bursty demand on the memory | |
252 | * allocator, so should reduce the chance of failed allocation requests | |
253 | * both for ourself and for other kernel subsystems. | |
254 | */ | |
255 | batch_target = np->rx_target - (req_prod - np->rx.rsp_cons); | |
256 | for (i = skb_queue_len(&np->rx_batch); i < batch_target; i++) { | |
617a20bb | 257 | skb = __netdev_alloc_skb(dev, RX_COPY_THRESHOLD + NET_IP_ALIGN, |
0d160211 JF |
258 | GFP_ATOMIC | __GFP_NOWARN); |
259 | if (unlikely(!skb)) | |
260 | goto no_skb; | |
261 | ||
617a20bb IY |
262 | /* Align ip header to a 16 bytes boundary */ |
263 | skb_reserve(skb, NET_IP_ALIGN); | |
264 | ||
0d160211 JF |
265 | page = alloc_page(GFP_ATOMIC | __GFP_NOWARN); |
266 | if (!page) { | |
267 | kfree_skb(skb); | |
268 | no_skb: | |
269 | /* Any skbuffs queued for refill? Force them out. */ | |
270 | if (i != 0) | |
271 | goto refill; | |
272 | /* Could not allocate any skbuffs. Try again later. */ | |
273 | mod_timer(&np->rx_refill_timer, | |
274 | jiffies + (HZ/10)); | |
275 | break; | |
276 | } | |
277 | ||
01c68026 | 278 | __skb_fill_page_desc(skb, 0, page, 0, 0); |
0d160211 JF |
279 | skb_shinfo(skb)->nr_frags = 1; |
280 | __skb_queue_tail(&np->rx_batch, skb); | |
281 | } | |
282 | ||
283 | /* Is the batch large enough to be worthwhile? */ | |
284 | if (i < (np->rx_target/2)) { | |
285 | if (req_prod > np->rx.sring->req_prod) | |
286 | goto push; | |
287 | return; | |
288 | } | |
289 | ||
290 | /* Adjust our fill target if we risked running out of buffers. */ | |
291 | if (((req_prod - np->rx.sring->rsp_prod) < (np->rx_target / 4)) && | |
292 | ((np->rx_target *= 2) > np->rx_max_target)) | |
293 | np->rx_target = np->rx_max_target; | |
294 | ||
295 | refill: | |
5dcddfae | 296 | for (i = 0; ; i++) { |
0d160211 JF |
297 | skb = __skb_dequeue(&np->rx_batch); |
298 | if (skb == NULL) | |
299 | break; | |
300 | ||
301 | skb->dev = dev; | |
302 | ||
303 | id = xennet_rxidx(req_prod + i); | |
304 | ||
305 | BUG_ON(np->rx_skbs[id]); | |
306 | np->rx_skbs[id] = skb; | |
307 | ||
308 | ref = gnttab_claim_grant_reference(&np->gref_rx_head); | |
309 | BUG_ON((signed short)ref < 0); | |
310 | np->grant_rx_ref[id] = ref; | |
311 | ||
01c68026 IC |
312 | pfn = page_to_pfn(skb_frag_page(&skb_shinfo(skb)->frags[0])); |
313 | vaddr = page_address(skb_frag_page(&skb_shinfo(skb)->frags[0])); | |
0d160211 JF |
314 | |
315 | req = RING_GET_REQUEST(&np->rx, req_prod + i); | |
316 | gnttab_grant_foreign_access_ref(ref, | |
317 | np->xbdev->otherend_id, | |
318 | pfn_to_mfn(pfn), | |
319 | 0); | |
320 | ||
321 | req->id = id; | |
322 | req->gref = ref; | |
323 | } | |
324 | ||
5dcddfae | 325 | wmb(); /* barrier so backend seens requests */ |
0d160211 JF |
326 | |
327 | /* Above is a suitable barrier to ensure backend will see requests. */ | |
328 | np->rx.req_prod_pvt = req_prod + i; | |
329 | push: | |
330 | RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np->rx, notify); | |
331 | if (notify) | |
332 | notify_remote_via_irq(np->netdev->irq); | |
333 | } | |
334 | ||
335 | static int xennet_open(struct net_device *dev) | |
336 | { | |
337 | struct netfront_info *np = netdev_priv(dev); | |
338 | ||
bea3348e SH |
339 | napi_enable(&np->napi); |
340 | ||
0d160211 JF |
341 | spin_lock_bh(&np->rx_lock); |
342 | if (netif_carrier_ok(dev)) { | |
343 | xennet_alloc_rx_buffers(dev); | |
344 | np->rx.sring->rsp_event = np->rx.rsp_cons + 1; | |
345 | if (RING_HAS_UNCONSUMED_RESPONSES(&np->rx)) | |
288379f0 | 346 | napi_schedule(&np->napi); |
0d160211 JF |
347 | } |
348 | spin_unlock_bh(&np->rx_lock); | |
349 | ||
0b1ab1b8 | 350 | netif_start_queue(dev); |
0d160211 JF |
351 | |
352 | return 0; | |
353 | } | |
354 | ||
355 | static void xennet_tx_buf_gc(struct net_device *dev) | |
356 | { | |
357 | RING_IDX cons, prod; | |
358 | unsigned short id; | |
359 | struct netfront_info *np = netdev_priv(dev); | |
360 | struct sk_buff *skb; | |
361 | ||
362 | BUG_ON(!netif_carrier_ok(dev)); | |
363 | ||
364 | do { | |
365 | prod = np->tx.sring->rsp_prod; | |
366 | rmb(); /* Ensure we see responses up to 'rp'. */ | |
367 | ||
368 | for (cons = np->tx.rsp_cons; cons != prod; cons++) { | |
369 | struct xen_netif_tx_response *txrsp; | |
370 | ||
371 | txrsp = RING_GET_RESPONSE(&np->tx, cons); | |
f942dc25 | 372 | if (txrsp->status == XEN_NETIF_RSP_NULL) |
0d160211 JF |
373 | continue; |
374 | ||
375 | id = txrsp->id; | |
376 | skb = np->tx_skbs[id].skb; | |
377 | if (unlikely(gnttab_query_foreign_access( | |
378 | np->grant_tx_ref[id]) != 0)) { | |
379 | printk(KERN_ALERT "xennet_tx_buf_gc: warning " | |
380 | "-- grant still in use by backend " | |
381 | "domain.\n"); | |
382 | BUG(); | |
383 | } | |
384 | gnttab_end_foreign_access_ref( | |
385 | np->grant_tx_ref[id], GNTMAP_readonly); | |
386 | gnttab_release_grant_reference( | |
387 | &np->gref_tx_head, np->grant_tx_ref[id]); | |
388 | np->grant_tx_ref[id] = GRANT_INVALID_REF; | |
389 | add_id_to_freelist(&np->tx_skb_freelist, np->tx_skbs, id); | |
390 | dev_kfree_skb_irq(skb); | |
391 | } | |
392 | ||
393 | np->tx.rsp_cons = prod; | |
394 | ||
395 | /* | |
396 | * Set a new event, then check for race with update of tx_cons. | |
397 | * Note that it is essential to schedule a callback, no matter | |
398 | * how few buffers are pending. Even if there is space in the | |
399 | * transmit ring, higher layers may be blocked because too much | |
400 | * data is outstanding: in such cases notification from Xen is | |
401 | * likely to be the only kick that we'll get. | |
402 | */ | |
403 | np->tx.sring->rsp_event = | |
404 | prod + ((np->tx.sring->req_prod - prod) >> 1) + 1; | |
405 | mb(); /* update shared area */ | |
406 | } while ((cons == prod) && (prod != np->tx.sring->rsp_prod)); | |
407 | ||
408 | xennet_maybe_wake_tx(dev); | |
409 | } | |
410 | ||
411 | static void xennet_make_frags(struct sk_buff *skb, struct net_device *dev, | |
412 | struct xen_netif_tx_request *tx) | |
413 | { | |
414 | struct netfront_info *np = netdev_priv(dev); | |
415 | char *data = skb->data; | |
416 | unsigned long mfn; | |
417 | RING_IDX prod = np->tx.req_prod_pvt; | |
418 | int frags = skb_shinfo(skb)->nr_frags; | |
419 | unsigned int offset = offset_in_page(data); | |
420 | unsigned int len = skb_headlen(skb); | |
421 | unsigned int id; | |
422 | grant_ref_t ref; | |
423 | int i; | |
424 | ||
425 | /* While the header overlaps a page boundary (including being | |
426 | larger than a page), split it it into page-sized chunks. */ | |
427 | while (len > PAGE_SIZE - offset) { | |
428 | tx->size = PAGE_SIZE - offset; | |
f942dc25 | 429 | tx->flags |= XEN_NETTXF_more_data; |
0d160211 JF |
430 | len -= tx->size; |
431 | data += tx->size; | |
432 | offset = 0; | |
433 | ||
434 | id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs); | |
435 | np->tx_skbs[id].skb = skb_get(skb); | |
436 | tx = RING_GET_REQUEST(&np->tx, prod++); | |
437 | tx->id = id; | |
438 | ref = gnttab_claim_grant_reference(&np->gref_tx_head); | |
439 | BUG_ON((signed short)ref < 0); | |
440 | ||
441 | mfn = virt_to_mfn(data); | |
442 | gnttab_grant_foreign_access_ref(ref, np->xbdev->otherend_id, | |
443 | mfn, GNTMAP_readonly); | |
444 | ||
445 | tx->gref = np->grant_tx_ref[id] = ref; | |
446 | tx->offset = offset; | |
447 | tx->size = len; | |
448 | tx->flags = 0; | |
449 | } | |
450 | ||
451 | /* Grant backend access to each skb fragment page. */ | |
452 | for (i = 0; i < frags; i++) { | |
453 | skb_frag_t *frag = skb_shinfo(skb)->frags + i; | |
454 | ||
f942dc25 | 455 | tx->flags |= XEN_NETTXF_more_data; |
0d160211 JF |
456 | |
457 | id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs); | |
458 | np->tx_skbs[id].skb = skb_get(skb); | |
459 | tx = RING_GET_REQUEST(&np->tx, prod++); | |
460 | tx->id = id; | |
461 | ref = gnttab_claim_grant_reference(&np->gref_tx_head); | |
462 | BUG_ON((signed short)ref < 0); | |
463 | ||
01c68026 | 464 | mfn = pfn_to_mfn(page_to_pfn(skb_frag_page(frag))); |
0d160211 JF |
465 | gnttab_grant_foreign_access_ref(ref, np->xbdev->otherend_id, |
466 | mfn, GNTMAP_readonly); | |
467 | ||
468 | tx->gref = np->grant_tx_ref[id] = ref; | |
469 | tx->offset = frag->page_offset; | |
9e903e08 | 470 | tx->size = skb_frag_size(frag); |
0d160211 JF |
471 | tx->flags = 0; |
472 | } | |
473 | ||
474 | np->tx.req_prod_pvt = prod; | |
475 | } | |
476 | ||
477 | static int xennet_start_xmit(struct sk_buff *skb, struct net_device *dev) | |
478 | { | |
479 | unsigned short id; | |
480 | struct netfront_info *np = netdev_priv(dev); | |
e00f85be | 481 | struct netfront_stats *stats = this_cpu_ptr(np->stats); |
0d160211 JF |
482 | struct xen_netif_tx_request *tx; |
483 | struct xen_netif_extra_info *extra; | |
484 | char *data = skb->data; | |
485 | RING_IDX i; | |
486 | grant_ref_t ref; | |
487 | unsigned long mfn; | |
488 | int notify; | |
489 | int frags = skb_shinfo(skb)->nr_frags; | |
490 | unsigned int offset = offset_in_page(data); | |
491 | unsigned int len = skb_headlen(skb); | |
492 | ||
43526370 | 493 | frags += DIV_ROUND_UP(offset + len, PAGE_SIZE); |
0d160211 JF |
494 | if (unlikely(frags > MAX_SKB_FRAGS + 1)) { |
495 | printk(KERN_ALERT "xennet: skb rides the rocket: %d frags\n", | |
496 | frags); | |
497 | dump_stack(); | |
498 | goto drop; | |
499 | } | |
500 | ||
501 | spin_lock_irq(&np->tx_lock); | |
502 | ||
503 | if (unlikely(!netif_carrier_ok(dev) || | |
504 | (frags > 1 && !xennet_can_sg(dev)) || | |
fc741216 | 505 | netif_needs_gso(skb, netif_skb_features(skb)))) { |
0d160211 JF |
506 | spin_unlock_irq(&np->tx_lock); |
507 | goto drop; | |
508 | } | |
509 | ||
510 | i = np->tx.req_prod_pvt; | |
511 | ||
512 | id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs); | |
513 | np->tx_skbs[id].skb = skb; | |
514 | ||
515 | tx = RING_GET_REQUEST(&np->tx, i); | |
516 | ||
517 | tx->id = id; | |
518 | ref = gnttab_claim_grant_reference(&np->gref_tx_head); | |
519 | BUG_ON((signed short)ref < 0); | |
520 | mfn = virt_to_mfn(data); | |
521 | gnttab_grant_foreign_access_ref( | |
522 | ref, np->xbdev->otherend_id, mfn, GNTMAP_readonly); | |
523 | tx->gref = np->grant_tx_ref[id] = ref; | |
524 | tx->offset = offset; | |
525 | tx->size = len; | |
526 | extra = NULL; | |
527 | ||
528 | tx->flags = 0; | |
529 | if (skb->ip_summed == CHECKSUM_PARTIAL) | |
530 | /* local packet? */ | |
f942dc25 | 531 | tx->flags |= XEN_NETTXF_csum_blank | XEN_NETTXF_data_validated; |
0d160211 JF |
532 | else if (skb->ip_summed == CHECKSUM_UNNECESSARY) |
533 | /* remote but checksummed. */ | |
f942dc25 | 534 | tx->flags |= XEN_NETTXF_data_validated; |
0d160211 JF |
535 | |
536 | if (skb_shinfo(skb)->gso_size) { | |
537 | struct xen_netif_extra_info *gso; | |
538 | ||
539 | gso = (struct xen_netif_extra_info *) | |
540 | RING_GET_REQUEST(&np->tx, ++i); | |
541 | ||
542 | if (extra) | |
543 | extra->flags |= XEN_NETIF_EXTRA_FLAG_MORE; | |
544 | else | |
f942dc25 | 545 | tx->flags |= XEN_NETTXF_extra_info; |
0d160211 JF |
546 | |
547 | gso->u.gso.size = skb_shinfo(skb)->gso_size; | |
548 | gso->u.gso.type = XEN_NETIF_GSO_TYPE_TCPV4; | |
549 | gso->u.gso.pad = 0; | |
550 | gso->u.gso.features = 0; | |
551 | ||
552 | gso->type = XEN_NETIF_EXTRA_TYPE_GSO; | |
553 | gso->flags = 0; | |
554 | extra = gso; | |
555 | } | |
556 | ||
557 | np->tx.req_prod_pvt = i + 1; | |
558 | ||
559 | xennet_make_frags(skb, dev, tx); | |
560 | tx->size = skb->len; | |
561 | ||
562 | RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np->tx, notify); | |
563 | if (notify) | |
564 | notify_remote_via_irq(np->netdev->irq); | |
565 | ||
e00f85be | 566 | u64_stats_update_begin(&stats->syncp); |
567 | stats->tx_bytes += skb->len; | |
568 | stats->tx_packets++; | |
569 | u64_stats_update_end(&stats->syncp); | |
10a273a6 JF |
570 | |
571 | /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */ | |
0d160211 JF |
572 | xennet_tx_buf_gc(dev); |
573 | ||
574 | if (!netfront_tx_slot_available(np)) | |
575 | netif_stop_queue(dev); | |
576 | ||
577 | spin_unlock_irq(&np->tx_lock); | |
578 | ||
6ed10654 | 579 | return NETDEV_TX_OK; |
0d160211 JF |
580 | |
581 | drop: | |
09f75cd7 | 582 | dev->stats.tx_dropped++; |
0d160211 | 583 | dev_kfree_skb(skb); |
6ed10654 | 584 | return NETDEV_TX_OK; |
0d160211 JF |
585 | } |
586 | ||
587 | static int xennet_close(struct net_device *dev) | |
588 | { | |
589 | struct netfront_info *np = netdev_priv(dev); | |
590 | netif_stop_queue(np->netdev); | |
bea3348e | 591 | napi_disable(&np->napi); |
0d160211 JF |
592 | return 0; |
593 | } | |
594 | ||
0d160211 JF |
595 | static void xennet_move_rx_slot(struct netfront_info *np, struct sk_buff *skb, |
596 | grant_ref_t ref) | |
597 | { | |
598 | int new = xennet_rxidx(np->rx.req_prod_pvt); | |
599 | ||
600 | BUG_ON(np->rx_skbs[new]); | |
601 | np->rx_skbs[new] = skb; | |
602 | np->grant_rx_ref[new] = ref; | |
603 | RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->id = new; | |
604 | RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->gref = ref; | |
605 | np->rx.req_prod_pvt++; | |
606 | } | |
607 | ||
608 | static int xennet_get_extras(struct netfront_info *np, | |
609 | struct xen_netif_extra_info *extras, | |
610 | RING_IDX rp) | |
611 | ||
612 | { | |
613 | struct xen_netif_extra_info *extra; | |
614 | struct device *dev = &np->netdev->dev; | |
615 | RING_IDX cons = np->rx.rsp_cons; | |
616 | int err = 0; | |
617 | ||
618 | do { | |
619 | struct sk_buff *skb; | |
620 | grant_ref_t ref; | |
621 | ||
622 | if (unlikely(cons + 1 == rp)) { | |
623 | if (net_ratelimit()) | |
624 | dev_warn(dev, "Missing extra info\n"); | |
625 | err = -EBADR; | |
626 | break; | |
627 | } | |
628 | ||
629 | extra = (struct xen_netif_extra_info *) | |
630 | RING_GET_RESPONSE(&np->rx, ++cons); | |
631 | ||
632 | if (unlikely(!extra->type || | |
633 | extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) { | |
634 | if (net_ratelimit()) | |
635 | dev_warn(dev, "Invalid extra type: %d\n", | |
636 | extra->type); | |
637 | err = -EINVAL; | |
638 | } else { | |
639 | memcpy(&extras[extra->type - 1], extra, | |
640 | sizeof(*extra)); | |
641 | } | |
642 | ||
643 | skb = xennet_get_rx_skb(np, cons); | |
644 | ref = xennet_get_rx_ref(np, cons); | |
645 | xennet_move_rx_slot(np, skb, ref); | |
646 | } while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE); | |
647 | ||
648 | np->rx.rsp_cons = cons; | |
649 | return err; | |
650 | } | |
651 | ||
652 | static int xennet_get_responses(struct netfront_info *np, | |
653 | struct netfront_rx_info *rinfo, RING_IDX rp, | |
654 | struct sk_buff_head *list) | |
655 | { | |
656 | struct xen_netif_rx_response *rx = &rinfo->rx; | |
657 | struct xen_netif_extra_info *extras = rinfo->extras; | |
658 | struct device *dev = &np->netdev->dev; | |
659 | RING_IDX cons = np->rx.rsp_cons; | |
660 | struct sk_buff *skb = xennet_get_rx_skb(np, cons); | |
661 | grant_ref_t ref = xennet_get_rx_ref(np, cons); | |
662 | int max = MAX_SKB_FRAGS + (rx->status <= RX_COPY_THRESHOLD); | |
663 | int frags = 1; | |
664 | int err = 0; | |
665 | unsigned long ret; | |
666 | ||
f942dc25 | 667 | if (rx->flags & XEN_NETRXF_extra_info) { |
0d160211 JF |
668 | err = xennet_get_extras(np, extras, rp); |
669 | cons = np->rx.rsp_cons; | |
670 | } | |
671 | ||
672 | for (;;) { | |
673 | if (unlikely(rx->status < 0 || | |
674 | rx->offset + rx->status > PAGE_SIZE)) { | |
675 | if (net_ratelimit()) | |
676 | dev_warn(dev, "rx->offset: %x, size: %u\n", | |
677 | rx->offset, rx->status); | |
678 | xennet_move_rx_slot(np, skb, ref); | |
679 | err = -EINVAL; | |
680 | goto next; | |
681 | } | |
682 | ||
683 | /* | |
684 | * This definitely indicates a bug, either in this driver or in | |
685 | * the backend driver. In future this should flag the bad | |
686 | * situation to the system controller to reboot the backed. | |
687 | */ | |
688 | if (ref == GRANT_INVALID_REF) { | |
689 | if (net_ratelimit()) | |
690 | dev_warn(dev, "Bad rx response id %d.\n", | |
691 | rx->id); | |
692 | err = -EINVAL; | |
693 | goto next; | |
694 | } | |
695 | ||
696 | ret = gnttab_end_foreign_access_ref(ref, 0); | |
697 | BUG_ON(!ret); | |
698 | ||
699 | gnttab_release_grant_reference(&np->gref_rx_head, ref); | |
700 | ||
701 | __skb_queue_tail(list, skb); | |
702 | ||
703 | next: | |
f942dc25 | 704 | if (!(rx->flags & XEN_NETRXF_more_data)) |
0d160211 JF |
705 | break; |
706 | ||
707 | if (cons + frags == rp) { | |
708 | if (net_ratelimit()) | |
709 | dev_warn(dev, "Need more frags\n"); | |
710 | err = -ENOENT; | |
711 | break; | |
712 | } | |
713 | ||
714 | rx = RING_GET_RESPONSE(&np->rx, cons + frags); | |
715 | skb = xennet_get_rx_skb(np, cons + frags); | |
716 | ref = xennet_get_rx_ref(np, cons + frags); | |
717 | frags++; | |
718 | } | |
719 | ||
720 | if (unlikely(frags > max)) { | |
721 | if (net_ratelimit()) | |
722 | dev_warn(dev, "Too many frags\n"); | |
723 | err = -E2BIG; | |
724 | } | |
725 | ||
726 | if (unlikely(err)) | |
727 | np->rx.rsp_cons = cons + frags; | |
728 | ||
729 | return err; | |
730 | } | |
731 | ||
732 | static int xennet_set_skb_gso(struct sk_buff *skb, | |
733 | struct xen_netif_extra_info *gso) | |
734 | { | |
735 | if (!gso->u.gso.size) { | |
736 | if (net_ratelimit()) | |
737 | printk(KERN_WARNING "GSO size must not be zero.\n"); | |
738 | return -EINVAL; | |
739 | } | |
740 | ||
741 | /* Currently only TCPv4 S.O. is supported. */ | |
742 | if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4) { | |
743 | if (net_ratelimit()) | |
744 | printk(KERN_WARNING "Bad GSO type %d.\n", gso->u.gso.type); | |
745 | return -EINVAL; | |
746 | } | |
747 | ||
748 | skb_shinfo(skb)->gso_size = gso->u.gso.size; | |
749 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4; | |
750 | ||
751 | /* Header must be checked, and gso_segs computed. */ | |
752 | skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; | |
753 | skb_shinfo(skb)->gso_segs = 0; | |
754 | ||
755 | return 0; | |
756 | } | |
757 | ||
758 | static RING_IDX xennet_fill_frags(struct netfront_info *np, | |
759 | struct sk_buff *skb, | |
760 | struct sk_buff_head *list) | |
761 | { | |
762 | struct skb_shared_info *shinfo = skb_shinfo(skb); | |
763 | int nr_frags = shinfo->nr_frags; | |
764 | RING_IDX cons = np->rx.rsp_cons; | |
0d160211 JF |
765 | struct sk_buff *nskb; |
766 | ||
767 | while ((nskb = __skb_dequeue(list))) { | |
768 | struct xen_netif_rx_response *rx = | |
769 | RING_GET_RESPONSE(&np->rx, ++cons); | |
01c68026 | 770 | skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0]; |
0d160211 | 771 | |
01c68026 IC |
772 | __skb_fill_page_desc(skb, nr_frags, |
773 | skb_frag_page(nfrag), | |
774 | rx->offset, rx->status); | |
0d160211 JF |
775 | |
776 | skb->data_len += rx->status; | |
777 | ||
778 | skb_shinfo(nskb)->nr_frags = 0; | |
779 | kfree_skb(nskb); | |
780 | ||
0d160211 JF |
781 | nr_frags++; |
782 | } | |
783 | ||
784 | shinfo->nr_frags = nr_frags; | |
785 | return cons; | |
786 | } | |
787 | ||
e0ce4af9 | 788 | static int checksum_setup(struct net_device *dev, struct sk_buff *skb) |
0d160211 JF |
789 | { |
790 | struct iphdr *iph; | |
791 | unsigned char *th; | |
792 | int err = -EPROTO; | |
e0ce4af9 IC |
793 | int recalculate_partial_csum = 0; |
794 | ||
795 | /* | |
796 | * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy | |
797 | * peers can fail to set NETRXF_csum_blank when sending a GSO | |
798 | * frame. In this case force the SKB to CHECKSUM_PARTIAL and | |
799 | * recalculate the partial checksum. | |
800 | */ | |
801 | if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) { | |
802 | struct netfront_info *np = netdev_priv(dev); | |
803 | np->rx_gso_checksum_fixup++; | |
804 | skb->ip_summed = CHECKSUM_PARTIAL; | |
805 | recalculate_partial_csum = 1; | |
806 | } | |
807 | ||
808 | /* A non-CHECKSUM_PARTIAL SKB does not require setup. */ | |
809 | if (skb->ip_summed != CHECKSUM_PARTIAL) | |
810 | return 0; | |
0d160211 JF |
811 | |
812 | if (skb->protocol != htons(ETH_P_IP)) | |
813 | goto out; | |
814 | ||
815 | iph = (void *)skb->data; | |
816 | th = skb->data + 4 * iph->ihl; | |
817 | if (th >= skb_tail_pointer(skb)) | |
818 | goto out; | |
819 | ||
820 | skb->csum_start = th - skb->head; | |
821 | switch (iph->protocol) { | |
822 | case IPPROTO_TCP: | |
823 | skb->csum_offset = offsetof(struct tcphdr, check); | |
e0ce4af9 IC |
824 | |
825 | if (recalculate_partial_csum) { | |
826 | struct tcphdr *tcph = (struct tcphdr *)th; | |
827 | tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, | |
828 | skb->len - iph->ihl*4, | |
829 | IPPROTO_TCP, 0); | |
830 | } | |
0d160211 JF |
831 | break; |
832 | case IPPROTO_UDP: | |
833 | skb->csum_offset = offsetof(struct udphdr, check); | |
e0ce4af9 IC |
834 | |
835 | if (recalculate_partial_csum) { | |
836 | struct udphdr *udph = (struct udphdr *)th; | |
837 | udph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, | |
838 | skb->len - iph->ihl*4, | |
839 | IPPROTO_UDP, 0); | |
840 | } | |
0d160211 JF |
841 | break; |
842 | default: | |
843 | if (net_ratelimit()) | |
844 | printk(KERN_ERR "Attempting to checksum a non-" | |
845 | "TCP/UDP packet, dropping a protocol" | |
846 | " %d packet", iph->protocol); | |
847 | goto out; | |
848 | } | |
849 | ||
850 | if ((th + skb->csum_offset + 2) > skb_tail_pointer(skb)) | |
851 | goto out; | |
852 | ||
853 | err = 0; | |
854 | ||
855 | out: | |
856 | return err; | |
857 | } | |
858 | ||
859 | static int handle_incoming_queue(struct net_device *dev, | |
09f75cd7 | 860 | struct sk_buff_head *rxq) |
0d160211 | 861 | { |
e00f85be | 862 | struct netfront_info *np = netdev_priv(dev); |
863 | struct netfront_stats *stats = this_cpu_ptr(np->stats); | |
0d160211 JF |
864 | int packets_dropped = 0; |
865 | struct sk_buff *skb; | |
866 | ||
867 | while ((skb = __skb_dequeue(rxq)) != NULL) { | |
868 | struct page *page = NETFRONT_SKB_CB(skb)->page; | |
869 | void *vaddr = page_address(page); | |
870 | unsigned offset = NETFRONT_SKB_CB(skb)->offset; | |
871 | ||
872 | memcpy(skb->data, vaddr + offset, | |
873 | skb_headlen(skb)); | |
874 | ||
01c68026 | 875 | if (page != skb_frag_page(&skb_shinfo(skb)->frags[0])) |
0d160211 JF |
876 | __free_page(page); |
877 | ||
878 | /* Ethernet work: Delayed to here as it peeks the header. */ | |
879 | skb->protocol = eth_type_trans(skb, dev); | |
880 | ||
e0ce4af9 IC |
881 | if (checksum_setup(dev, skb)) { |
882 | kfree_skb(skb); | |
883 | packets_dropped++; | |
884 | dev->stats.rx_errors++; | |
885 | continue; | |
0d160211 JF |
886 | } |
887 | ||
e00f85be | 888 | u64_stats_update_begin(&stats->syncp); |
889 | stats->rx_packets++; | |
890 | stats->rx_bytes += skb->len; | |
891 | u64_stats_update_end(&stats->syncp); | |
0d160211 JF |
892 | |
893 | /* Pass it up. */ | |
894 | netif_receive_skb(skb); | |
0d160211 JF |
895 | } |
896 | ||
897 | return packets_dropped; | |
898 | } | |
899 | ||
bea3348e | 900 | static int xennet_poll(struct napi_struct *napi, int budget) |
0d160211 | 901 | { |
bea3348e SH |
902 | struct netfront_info *np = container_of(napi, struct netfront_info, napi); |
903 | struct net_device *dev = np->netdev; | |
0d160211 JF |
904 | struct sk_buff *skb; |
905 | struct netfront_rx_info rinfo; | |
906 | struct xen_netif_rx_response *rx = &rinfo.rx; | |
907 | struct xen_netif_extra_info *extras = rinfo.extras; | |
908 | RING_IDX i, rp; | |
bea3348e | 909 | int work_done; |
0d160211 JF |
910 | struct sk_buff_head rxq; |
911 | struct sk_buff_head errq; | |
912 | struct sk_buff_head tmpq; | |
913 | unsigned long flags; | |
914 | unsigned int len; | |
915 | int err; | |
916 | ||
917 | spin_lock(&np->rx_lock); | |
918 | ||
0d160211 JF |
919 | skb_queue_head_init(&rxq); |
920 | skb_queue_head_init(&errq); | |
921 | skb_queue_head_init(&tmpq); | |
922 | ||
0d160211 JF |
923 | rp = np->rx.sring->rsp_prod; |
924 | rmb(); /* Ensure we see queued responses up to 'rp'. */ | |
925 | ||
926 | i = np->rx.rsp_cons; | |
927 | work_done = 0; | |
928 | while ((i != rp) && (work_done < budget)) { | |
929 | memcpy(rx, RING_GET_RESPONSE(&np->rx, i), sizeof(*rx)); | |
930 | memset(extras, 0, sizeof(rinfo.extras)); | |
931 | ||
932 | err = xennet_get_responses(np, &rinfo, rp, &tmpq); | |
933 | ||
934 | if (unlikely(err)) { | |
935 | err: | |
936 | while ((skb = __skb_dequeue(&tmpq))) | |
937 | __skb_queue_tail(&errq, skb); | |
09f75cd7 | 938 | dev->stats.rx_errors++; |
0d160211 JF |
939 | i = np->rx.rsp_cons; |
940 | continue; | |
941 | } | |
942 | ||
943 | skb = __skb_dequeue(&tmpq); | |
944 | ||
945 | if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) { | |
946 | struct xen_netif_extra_info *gso; | |
947 | gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1]; | |
948 | ||
949 | if (unlikely(xennet_set_skb_gso(skb, gso))) { | |
950 | __skb_queue_head(&tmpq, skb); | |
951 | np->rx.rsp_cons += skb_queue_len(&tmpq); | |
952 | goto err; | |
953 | } | |
954 | } | |
955 | ||
01c68026 IC |
956 | NETFRONT_SKB_CB(skb)->page = |
957 | skb_frag_page(&skb_shinfo(skb)->frags[0]); | |
0d160211 JF |
958 | NETFRONT_SKB_CB(skb)->offset = rx->offset; |
959 | ||
960 | len = rx->status; | |
961 | if (len > RX_COPY_THRESHOLD) | |
962 | len = RX_COPY_THRESHOLD; | |
963 | skb_put(skb, len); | |
964 | ||
965 | if (rx->status > len) { | |
966 | skb_shinfo(skb)->frags[0].page_offset = | |
967 | rx->offset + len; | |
9e903e08 | 968 | skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status - len); |
0d160211 JF |
969 | skb->data_len = rx->status - len; |
970 | } else { | |
01c68026 | 971 | __skb_fill_page_desc(skb, 0, NULL, 0, 0); |
0d160211 JF |
972 | skb_shinfo(skb)->nr_frags = 0; |
973 | } | |
974 | ||
975 | i = xennet_fill_frags(np, skb, &tmpq); | |
976 | ||
977 | /* | |
978 | * Truesize approximates the size of true data plus | |
979 | * any supervisor overheads. Adding hypervisor | |
980 | * overheads has been shown to significantly reduce | |
981 | * achievable bandwidth with the default receive | |
982 | * buffer size. It is therefore not wise to account | |
983 | * for it here. | |
984 | * | |
985 | * After alloc_skb(RX_COPY_THRESHOLD), truesize is set | |
986 | * to RX_COPY_THRESHOLD + the supervisor | |
987 | * overheads. Here, we add the size of the data pulled | |
988 | * in xennet_fill_frags(). | |
989 | * | |
990 | * We also adjust for any unused space in the main | |
991 | * data area by subtracting (RX_COPY_THRESHOLD - | |
992 | * len). This is especially important with drivers | |
993 | * which split incoming packets into header and data, | |
994 | * using only 66 bytes of the main data area (see the | |
995 | * e1000 driver for example.) On such systems, | |
996 | * without this last adjustement, our achievable | |
997 | * receive throughout using the standard receive | |
998 | * buffer size was cut by 25%(!!!). | |
999 | */ | |
1000 | skb->truesize += skb->data_len - (RX_COPY_THRESHOLD - len); | |
1001 | skb->len += skb->data_len; | |
1002 | ||
f942dc25 | 1003 | if (rx->flags & XEN_NETRXF_csum_blank) |
0d160211 | 1004 | skb->ip_summed = CHECKSUM_PARTIAL; |
f942dc25 | 1005 | else if (rx->flags & XEN_NETRXF_data_validated) |
0d160211 JF |
1006 | skb->ip_summed = CHECKSUM_UNNECESSARY; |
1007 | ||
1008 | __skb_queue_tail(&rxq, skb); | |
1009 | ||
1010 | np->rx.rsp_cons = ++i; | |
1011 | work_done++; | |
1012 | } | |
1013 | ||
56cfe5d0 | 1014 | __skb_queue_purge(&errq); |
0d160211 JF |
1015 | |
1016 | work_done -= handle_incoming_queue(dev, &rxq); | |
1017 | ||
1018 | /* If we get a callback with very few responses, reduce fill target. */ | |
1019 | /* NB. Note exponential increase, linear decrease. */ | |
1020 | if (((np->rx.req_prod_pvt - np->rx.sring->rsp_prod) > | |
1021 | ((3*np->rx_target) / 4)) && | |
1022 | (--np->rx_target < np->rx_min_target)) | |
1023 | np->rx_target = np->rx_min_target; | |
1024 | ||
1025 | xennet_alloc_rx_buffers(dev); | |
1026 | ||
0d160211 | 1027 | if (work_done < budget) { |
bea3348e SH |
1028 | int more_to_do = 0; |
1029 | ||
0d160211 JF |
1030 | local_irq_save(flags); |
1031 | ||
1032 | RING_FINAL_CHECK_FOR_RESPONSES(&np->rx, more_to_do); | |
1033 | if (!more_to_do) | |
288379f0 | 1034 | __napi_complete(napi); |
0d160211 JF |
1035 | |
1036 | local_irq_restore(flags); | |
1037 | } | |
1038 | ||
1039 | spin_unlock(&np->rx_lock); | |
1040 | ||
bea3348e | 1041 | return work_done; |
0d160211 JF |
1042 | } |
1043 | ||
1044 | static int xennet_change_mtu(struct net_device *dev, int mtu) | |
1045 | { | |
1046 | int max = xennet_can_sg(dev) ? 65535 - ETH_HLEN : ETH_DATA_LEN; | |
1047 | ||
1048 | if (mtu > max) | |
1049 | return -EINVAL; | |
1050 | dev->mtu = mtu; | |
1051 | return 0; | |
1052 | } | |
1053 | ||
e00f85be | 1054 | static struct rtnl_link_stats64 *xennet_get_stats64(struct net_device *dev, |
1055 | struct rtnl_link_stats64 *tot) | |
1056 | { | |
1057 | struct netfront_info *np = netdev_priv(dev); | |
1058 | int cpu; | |
1059 | ||
1060 | for_each_possible_cpu(cpu) { | |
1061 | struct netfront_stats *stats = per_cpu_ptr(np->stats, cpu); | |
1062 | u64 rx_packets, rx_bytes, tx_packets, tx_bytes; | |
1063 | unsigned int start; | |
1064 | ||
1065 | do { | |
1066 | start = u64_stats_fetch_begin_bh(&stats->syncp); | |
1067 | ||
1068 | rx_packets = stats->rx_packets; | |
1069 | tx_packets = stats->tx_packets; | |
1070 | rx_bytes = stats->rx_bytes; | |
1071 | tx_bytes = stats->tx_bytes; | |
1072 | } while (u64_stats_fetch_retry_bh(&stats->syncp, start)); | |
1073 | ||
1074 | tot->rx_packets += rx_packets; | |
1075 | tot->tx_packets += tx_packets; | |
1076 | tot->rx_bytes += rx_bytes; | |
1077 | tot->tx_bytes += tx_bytes; | |
1078 | } | |
1079 | ||
1080 | tot->rx_errors = dev->stats.rx_errors; | |
1081 | tot->tx_dropped = dev->stats.tx_dropped; | |
1082 | ||
1083 | return tot; | |
1084 | } | |
1085 | ||
0d160211 JF |
1086 | static void xennet_release_tx_bufs(struct netfront_info *np) |
1087 | { | |
1088 | struct sk_buff *skb; | |
1089 | int i; | |
1090 | ||
1091 | for (i = 0; i < NET_TX_RING_SIZE; i++) { | |
1092 | /* Skip over entries which are actually freelist references */ | |
1ffb40b8 | 1093 | if (skb_entry_is_link(&np->tx_skbs[i])) |
0d160211 JF |
1094 | continue; |
1095 | ||
1096 | skb = np->tx_skbs[i].skb; | |
1097 | gnttab_end_foreign_access_ref(np->grant_tx_ref[i], | |
1098 | GNTMAP_readonly); | |
1099 | gnttab_release_grant_reference(&np->gref_tx_head, | |
1100 | np->grant_tx_ref[i]); | |
1101 | np->grant_tx_ref[i] = GRANT_INVALID_REF; | |
1102 | add_id_to_freelist(&np->tx_skb_freelist, np->tx_skbs, i); | |
1103 | dev_kfree_skb_irq(skb); | |
1104 | } | |
1105 | } | |
1106 | ||
1107 | static void xennet_release_rx_bufs(struct netfront_info *np) | |
1108 | { | |
1109 | struct mmu_update *mmu = np->rx_mmu; | |
1110 | struct multicall_entry *mcl = np->rx_mcl; | |
1111 | struct sk_buff_head free_list; | |
1112 | struct sk_buff *skb; | |
1113 | unsigned long mfn; | |
1114 | int xfer = 0, noxfer = 0, unused = 0; | |
1115 | int id, ref; | |
1116 | ||
1117 | dev_warn(&np->netdev->dev, "%s: fix me for copying receiver.\n", | |
1118 | __func__); | |
1119 | return; | |
1120 | ||
1121 | skb_queue_head_init(&free_list); | |
1122 | ||
1123 | spin_lock_bh(&np->rx_lock); | |
1124 | ||
1125 | for (id = 0; id < NET_RX_RING_SIZE; id++) { | |
1126 | ref = np->grant_rx_ref[id]; | |
1127 | if (ref == GRANT_INVALID_REF) { | |
1128 | unused++; | |
1129 | continue; | |
1130 | } | |
1131 | ||
1132 | skb = np->rx_skbs[id]; | |
1133 | mfn = gnttab_end_foreign_transfer_ref(ref); | |
1134 | gnttab_release_grant_reference(&np->gref_rx_head, ref); | |
1135 | np->grant_rx_ref[id] = GRANT_INVALID_REF; | |
1136 | ||
1137 | if (0 == mfn) { | |
1138 | skb_shinfo(skb)->nr_frags = 0; | |
1139 | dev_kfree_skb(skb); | |
1140 | noxfer++; | |
1141 | continue; | |
1142 | } | |
1143 | ||
1144 | if (!xen_feature(XENFEAT_auto_translated_physmap)) { | |
1145 | /* Remap the page. */ | |
01c68026 IC |
1146 | const struct page *page = |
1147 | skb_frag_page(&skb_shinfo(skb)->frags[0]); | |
0d160211 JF |
1148 | unsigned long pfn = page_to_pfn(page); |
1149 | void *vaddr = page_address(page); | |
1150 | ||
1151 | MULTI_update_va_mapping(mcl, (unsigned long)vaddr, | |
1152 | mfn_pte(mfn, PAGE_KERNEL), | |
1153 | 0); | |
1154 | mcl++; | |
1155 | mmu->ptr = ((u64)mfn << PAGE_SHIFT) | |
1156 | | MMU_MACHPHYS_UPDATE; | |
1157 | mmu->val = pfn; | |
1158 | mmu++; | |
1159 | ||
1160 | set_phys_to_machine(pfn, mfn); | |
1161 | } | |
1162 | __skb_queue_tail(&free_list, skb); | |
1163 | xfer++; | |
1164 | } | |
1165 | ||
1166 | dev_info(&np->netdev->dev, "%s: %d xfer, %d noxfer, %d unused\n", | |
1167 | __func__, xfer, noxfer, unused); | |
1168 | ||
1169 | if (xfer) { | |
1170 | if (!xen_feature(XENFEAT_auto_translated_physmap)) { | |
1171 | /* Do all the remapping work and M2P updates. */ | |
1172 | MULTI_mmu_update(mcl, np->rx_mmu, mmu - np->rx_mmu, | |
79ea13ce | 1173 | NULL, DOMID_SELF); |
0d160211 JF |
1174 | mcl++; |
1175 | HYPERVISOR_multicall(np->rx_mcl, mcl - np->rx_mcl); | |
1176 | } | |
1177 | } | |
1178 | ||
56cfe5d0 | 1179 | __skb_queue_purge(&free_list); |
0d160211 JF |
1180 | |
1181 | spin_unlock_bh(&np->rx_lock); | |
1182 | } | |
1183 | ||
1184 | static void xennet_uninit(struct net_device *dev) | |
1185 | { | |
1186 | struct netfront_info *np = netdev_priv(dev); | |
1187 | xennet_release_tx_bufs(np); | |
1188 | xennet_release_rx_bufs(np); | |
1189 | gnttab_free_grant_references(np->gref_tx_head); | |
1190 | gnttab_free_grant_references(np->gref_rx_head); | |
1191 | } | |
1192 | ||
8f7b01a1 ED |
1193 | static u32 xennet_fix_features(struct net_device *dev, u32 features) |
1194 | { | |
1195 | struct netfront_info *np = netdev_priv(dev); | |
1196 | int val; | |
1197 | ||
1198 | if (features & NETIF_F_SG) { | |
1199 | if (xenbus_scanf(XBT_NIL, np->xbdev->otherend, "feature-sg", | |
1200 | "%d", &val) < 0) | |
1201 | val = 0; | |
1202 | ||
1203 | if (!val) | |
1204 | features &= ~NETIF_F_SG; | |
1205 | } | |
1206 | ||
1207 | if (features & NETIF_F_TSO) { | |
1208 | if (xenbus_scanf(XBT_NIL, np->xbdev->otherend, | |
1209 | "feature-gso-tcpv4", "%d", &val) < 0) | |
1210 | val = 0; | |
1211 | ||
1212 | if (!val) | |
1213 | features &= ~NETIF_F_TSO; | |
1214 | } | |
1215 | ||
1216 | return features; | |
1217 | } | |
1218 | ||
1219 | static int xennet_set_features(struct net_device *dev, u32 features) | |
1220 | { | |
1221 | if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) { | |
1222 | netdev_info(dev, "Reducing MTU because no SG offload"); | |
1223 | dev->mtu = ETH_DATA_LEN; | |
1224 | } | |
1225 | ||
1226 | return 0; | |
1227 | } | |
1228 | ||
0a0b9d2e SH |
1229 | static const struct net_device_ops xennet_netdev_ops = { |
1230 | .ndo_open = xennet_open, | |
1231 | .ndo_uninit = xennet_uninit, | |
1232 | .ndo_stop = xennet_close, | |
1233 | .ndo_start_xmit = xennet_start_xmit, | |
1234 | .ndo_change_mtu = xennet_change_mtu, | |
e00f85be | 1235 | .ndo_get_stats64 = xennet_get_stats64, |
0a0b9d2e SH |
1236 | .ndo_set_mac_address = eth_mac_addr, |
1237 | .ndo_validate_addr = eth_validate_addr, | |
fb507934 MM |
1238 | .ndo_fix_features = xennet_fix_features, |
1239 | .ndo_set_features = xennet_set_features, | |
0a0b9d2e SH |
1240 | }; |
1241 | ||
0d160211 JF |
1242 | static struct net_device * __devinit xennet_create_dev(struct xenbus_device *dev) |
1243 | { | |
1244 | int i, err; | |
1245 | struct net_device *netdev; | |
1246 | struct netfront_info *np; | |
1247 | ||
1248 | netdev = alloc_etherdev(sizeof(struct netfront_info)); | |
1249 | if (!netdev) { | |
1250 | printk(KERN_WARNING "%s> alloc_etherdev failed.\n", | |
1251 | __func__); | |
1252 | return ERR_PTR(-ENOMEM); | |
1253 | } | |
1254 | ||
1255 | np = netdev_priv(netdev); | |
1256 | np->xbdev = dev; | |
1257 | ||
1258 | spin_lock_init(&np->tx_lock); | |
1259 | spin_lock_init(&np->rx_lock); | |
1260 | ||
1261 | skb_queue_head_init(&np->rx_batch); | |
1262 | np->rx_target = RX_DFL_MIN_TARGET; | |
1263 | np->rx_min_target = RX_DFL_MIN_TARGET; | |
1264 | np->rx_max_target = RX_MAX_TARGET; | |
1265 | ||
1266 | init_timer(&np->rx_refill_timer); | |
1267 | np->rx_refill_timer.data = (unsigned long)netdev; | |
1268 | np->rx_refill_timer.function = rx_refill_timeout; | |
1269 | ||
e00f85be | 1270 | err = -ENOMEM; |
1271 | np->stats = alloc_percpu(struct netfront_stats); | |
1272 | if (np->stats == NULL) | |
1273 | goto exit; | |
1274 | ||
0d160211 JF |
1275 | /* Initialise tx_skbs as a free chain containing every entry. */ |
1276 | np->tx_skb_freelist = 0; | |
1277 | for (i = 0; i < NET_TX_RING_SIZE; i++) { | |
1ffb40b8 | 1278 | skb_entry_set_link(&np->tx_skbs[i], i+1); |
0d160211 JF |
1279 | np->grant_tx_ref[i] = GRANT_INVALID_REF; |
1280 | } | |
1281 | ||
1282 | /* Clear out rx_skbs */ | |
1283 | for (i = 0; i < NET_RX_RING_SIZE; i++) { | |
1284 | np->rx_skbs[i] = NULL; | |
1285 | np->grant_rx_ref[i] = GRANT_INVALID_REF; | |
1286 | } | |
1287 | ||
1288 | /* A grant for every tx ring slot */ | |
1289 | if (gnttab_alloc_grant_references(TX_MAX_TARGET, | |
1290 | &np->gref_tx_head) < 0) { | |
1291 | printk(KERN_ALERT "#### netfront can't alloc tx grant refs\n"); | |
1292 | err = -ENOMEM; | |
e00f85be | 1293 | goto exit_free_stats; |
0d160211 JF |
1294 | } |
1295 | /* A grant for every rx ring slot */ | |
1296 | if (gnttab_alloc_grant_references(RX_MAX_TARGET, | |
1297 | &np->gref_rx_head) < 0) { | |
1298 | printk(KERN_ALERT "#### netfront can't alloc rx grant refs\n"); | |
1299 | err = -ENOMEM; | |
1300 | goto exit_free_tx; | |
1301 | } | |
1302 | ||
0a0b9d2e SH |
1303 | netdev->netdev_ops = &xennet_netdev_ops; |
1304 | ||
bea3348e | 1305 | netif_napi_add(netdev, &np->napi, xennet_poll, 64); |
fb507934 MM |
1306 | netdev->features = NETIF_F_IP_CSUM | NETIF_F_RXCSUM | |
1307 | NETIF_F_GSO_ROBUST; | |
1308 | netdev->hw_features = NETIF_F_IP_CSUM | NETIF_F_SG | NETIF_F_TSO; | |
0d160211 | 1309 | |
fc3e5941 IC |
1310 | /* |
1311 | * Assume that all hw features are available for now. This set | |
1312 | * will be adjusted by the call to netdev_update_features() in | |
1313 | * xennet_connect() which is the earliest point where we can | |
1314 | * negotiate with the backend regarding supported features. | |
1315 | */ | |
1316 | netdev->features |= netdev->hw_features; | |
1317 | ||
0d160211 | 1318 | SET_ETHTOOL_OPS(netdev, &xennet_ethtool_ops); |
0d160211 JF |
1319 | SET_NETDEV_DEV(netdev, &dev->dev); |
1320 | ||
1321 | np->netdev = netdev; | |
1322 | ||
1323 | netif_carrier_off(netdev); | |
1324 | ||
1325 | return netdev; | |
1326 | ||
1327 | exit_free_tx: | |
1328 | gnttab_free_grant_references(np->gref_tx_head); | |
e00f85be | 1329 | exit_free_stats: |
1330 | free_percpu(np->stats); | |
0d160211 JF |
1331 | exit: |
1332 | free_netdev(netdev); | |
1333 | return ERR_PTR(err); | |
1334 | } | |
1335 | ||
1336 | /** | |
1337 | * Entry point to this code when a new device is created. Allocate the basic | |
1338 | * structures and the ring buffers for communication with the backend, and | |
1339 | * inform the backend of the appropriate details for those. | |
1340 | */ | |
1341 | static int __devinit netfront_probe(struct xenbus_device *dev, | |
1342 | const struct xenbus_device_id *id) | |
1343 | { | |
1344 | int err; | |
1345 | struct net_device *netdev; | |
1346 | struct netfront_info *info; | |
1347 | ||
1348 | netdev = xennet_create_dev(dev); | |
1349 | if (IS_ERR(netdev)) { | |
1350 | err = PTR_ERR(netdev); | |
1351 | xenbus_dev_fatal(dev, err, "creating netdev"); | |
1352 | return err; | |
1353 | } | |
1354 | ||
1355 | info = netdev_priv(netdev); | |
1b713e00 | 1356 | dev_set_drvdata(&dev->dev, info); |
0d160211 JF |
1357 | |
1358 | err = register_netdev(info->netdev); | |
1359 | if (err) { | |
1360 | printk(KERN_WARNING "%s: register_netdev err=%d\n", | |
1361 | __func__, err); | |
1362 | goto fail; | |
1363 | } | |
1364 | ||
1365 | err = xennet_sysfs_addif(info->netdev); | |
1366 | if (err) { | |
1367 | unregister_netdev(info->netdev); | |
1368 | printk(KERN_WARNING "%s: add sysfs failed err=%d\n", | |
1369 | __func__, err); | |
1370 | goto fail; | |
1371 | } | |
1372 | ||
1373 | return 0; | |
1374 | ||
1375 | fail: | |
1376 | free_netdev(netdev); | |
1b713e00 | 1377 | dev_set_drvdata(&dev->dev, NULL); |
0d160211 JF |
1378 | return err; |
1379 | } | |
1380 | ||
1381 | static void xennet_end_access(int ref, void *page) | |
1382 | { | |
1383 | /* This frees the page as a side-effect */ | |
1384 | if (ref != GRANT_INVALID_REF) | |
1385 | gnttab_end_foreign_access(ref, 0, (unsigned long)page); | |
1386 | } | |
1387 | ||
1388 | static void xennet_disconnect_backend(struct netfront_info *info) | |
1389 | { | |
1390 | /* Stop old i/f to prevent errors whilst we rebuild the state. */ | |
1391 | spin_lock_bh(&info->rx_lock); | |
1392 | spin_lock_irq(&info->tx_lock); | |
1393 | netif_carrier_off(info->netdev); | |
1394 | spin_unlock_irq(&info->tx_lock); | |
1395 | spin_unlock_bh(&info->rx_lock); | |
1396 | ||
1397 | if (info->netdev->irq) | |
1398 | unbind_from_irqhandler(info->netdev->irq, info->netdev); | |
1399 | info->evtchn = info->netdev->irq = 0; | |
1400 | ||
1401 | /* End access and free the pages */ | |
1402 | xennet_end_access(info->tx_ring_ref, info->tx.sring); | |
1403 | xennet_end_access(info->rx_ring_ref, info->rx.sring); | |
1404 | ||
1405 | info->tx_ring_ref = GRANT_INVALID_REF; | |
1406 | info->rx_ring_ref = GRANT_INVALID_REF; | |
1407 | info->tx.sring = NULL; | |
1408 | info->rx.sring = NULL; | |
1409 | } | |
1410 | ||
1411 | /** | |
1412 | * We are reconnecting to the backend, due to a suspend/resume, or a backend | |
1413 | * driver restart. We tear down our netif structure and recreate it, but | |
1414 | * leave the device-layer structures intact so that this is transparent to the | |
1415 | * rest of the kernel. | |
1416 | */ | |
1417 | static int netfront_resume(struct xenbus_device *dev) | |
1418 | { | |
1b713e00 | 1419 | struct netfront_info *info = dev_get_drvdata(&dev->dev); |
0d160211 JF |
1420 | |
1421 | dev_dbg(&dev->dev, "%s\n", dev->nodename); | |
1422 | ||
1423 | xennet_disconnect_backend(info); | |
1424 | return 0; | |
1425 | } | |
1426 | ||
1427 | static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[]) | |
1428 | { | |
1429 | char *s, *e, *macstr; | |
1430 | int i; | |
1431 | ||
1432 | macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL); | |
1433 | if (IS_ERR(macstr)) | |
1434 | return PTR_ERR(macstr); | |
1435 | ||
1436 | for (i = 0; i < ETH_ALEN; i++) { | |
1437 | mac[i] = simple_strtoul(s, &e, 16); | |
1438 | if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) { | |
1439 | kfree(macstr); | |
1440 | return -ENOENT; | |
1441 | } | |
1442 | s = e+1; | |
1443 | } | |
1444 | ||
1445 | kfree(macstr); | |
1446 | return 0; | |
1447 | } | |
1448 | ||
1449 | static irqreturn_t xennet_interrupt(int irq, void *dev_id) | |
1450 | { | |
1451 | struct net_device *dev = dev_id; | |
1452 | struct netfront_info *np = netdev_priv(dev); | |
1453 | unsigned long flags; | |
1454 | ||
1455 | spin_lock_irqsave(&np->tx_lock, flags); | |
1456 | ||
1457 | if (likely(netif_carrier_ok(dev))) { | |
1458 | xennet_tx_buf_gc(dev); | |
1459 | /* Under tx_lock: protects access to rx shared-ring indexes. */ | |
1460 | if (RING_HAS_UNCONSUMED_RESPONSES(&np->rx)) | |
288379f0 | 1461 | napi_schedule(&np->napi); |
0d160211 JF |
1462 | } |
1463 | ||
1464 | spin_unlock_irqrestore(&np->tx_lock, flags); | |
1465 | ||
1466 | return IRQ_HANDLED; | |
1467 | } | |
1468 | ||
1469 | static int setup_netfront(struct xenbus_device *dev, struct netfront_info *info) | |
1470 | { | |
1471 | struct xen_netif_tx_sring *txs; | |
1472 | struct xen_netif_rx_sring *rxs; | |
1473 | int err; | |
1474 | struct net_device *netdev = info->netdev; | |
1475 | ||
1476 | info->tx_ring_ref = GRANT_INVALID_REF; | |
1477 | info->rx_ring_ref = GRANT_INVALID_REF; | |
1478 | info->rx.sring = NULL; | |
1479 | info->tx.sring = NULL; | |
1480 | netdev->irq = 0; | |
1481 | ||
1482 | err = xen_net_read_mac(dev, netdev->dev_addr); | |
1483 | if (err) { | |
1484 | xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename); | |
1485 | goto fail; | |
1486 | } | |
1487 | ||
a144ff09 | 1488 | txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH); |
0d160211 JF |
1489 | if (!txs) { |
1490 | err = -ENOMEM; | |
1491 | xenbus_dev_fatal(dev, err, "allocating tx ring page"); | |
1492 | goto fail; | |
1493 | } | |
1494 | SHARED_RING_INIT(txs); | |
1495 | FRONT_RING_INIT(&info->tx, txs, PAGE_SIZE); | |
1496 | ||
1497 | err = xenbus_grant_ring(dev, virt_to_mfn(txs)); | |
1498 | if (err < 0) { | |
1499 | free_page((unsigned long)txs); | |
1500 | goto fail; | |
1501 | } | |
1502 | ||
1503 | info->tx_ring_ref = err; | |
a144ff09 | 1504 | rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH); |
0d160211 JF |
1505 | if (!rxs) { |
1506 | err = -ENOMEM; | |
1507 | xenbus_dev_fatal(dev, err, "allocating rx ring page"); | |
1508 | goto fail; | |
1509 | } | |
1510 | SHARED_RING_INIT(rxs); | |
1511 | FRONT_RING_INIT(&info->rx, rxs, PAGE_SIZE); | |
1512 | ||
1513 | err = xenbus_grant_ring(dev, virt_to_mfn(rxs)); | |
1514 | if (err < 0) { | |
1515 | free_page((unsigned long)rxs); | |
1516 | goto fail; | |
1517 | } | |
1518 | info->rx_ring_ref = err; | |
1519 | ||
1520 | err = xenbus_alloc_evtchn(dev, &info->evtchn); | |
1521 | if (err) | |
1522 | goto fail; | |
1523 | ||
1524 | err = bind_evtchn_to_irqhandler(info->evtchn, xennet_interrupt, | |
ab392d2d | 1525 | 0, netdev->name, netdev); |
0d160211 JF |
1526 | if (err < 0) |
1527 | goto fail; | |
1528 | netdev->irq = err; | |
1529 | return 0; | |
1530 | ||
1531 | fail: | |
1532 | return err; | |
1533 | } | |
1534 | ||
1535 | /* Common code used when first setting up, and when resuming. */ | |
f502bf2b | 1536 | static int talk_to_netback(struct xenbus_device *dev, |
0d160211 JF |
1537 | struct netfront_info *info) |
1538 | { | |
1539 | const char *message; | |
1540 | struct xenbus_transaction xbt; | |
1541 | int err; | |
1542 | ||
1543 | /* Create shared ring, alloc event channel. */ | |
1544 | err = setup_netfront(dev, info); | |
1545 | if (err) | |
1546 | goto out; | |
1547 | ||
1548 | again: | |
1549 | err = xenbus_transaction_start(&xbt); | |
1550 | if (err) { | |
1551 | xenbus_dev_fatal(dev, err, "starting transaction"); | |
1552 | goto destroy_ring; | |
1553 | } | |
1554 | ||
1555 | err = xenbus_printf(xbt, dev->nodename, "tx-ring-ref", "%u", | |
1556 | info->tx_ring_ref); | |
1557 | if (err) { | |
1558 | message = "writing tx ring-ref"; | |
1559 | goto abort_transaction; | |
1560 | } | |
1561 | err = xenbus_printf(xbt, dev->nodename, "rx-ring-ref", "%u", | |
1562 | info->rx_ring_ref); | |
1563 | if (err) { | |
1564 | message = "writing rx ring-ref"; | |
1565 | goto abort_transaction; | |
1566 | } | |
1567 | err = xenbus_printf(xbt, dev->nodename, | |
1568 | "event-channel", "%u", info->evtchn); | |
1569 | if (err) { | |
1570 | message = "writing event-channel"; | |
1571 | goto abort_transaction; | |
1572 | } | |
1573 | ||
1574 | err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u", | |
1575 | 1); | |
1576 | if (err) { | |
1577 | message = "writing request-rx-copy"; | |
1578 | goto abort_transaction; | |
1579 | } | |
1580 | ||
1581 | err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1); | |
1582 | if (err) { | |
1583 | message = "writing feature-rx-notify"; | |
1584 | goto abort_transaction; | |
1585 | } | |
1586 | ||
1587 | err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1); | |
1588 | if (err) { | |
1589 | message = "writing feature-sg"; | |
1590 | goto abort_transaction; | |
1591 | } | |
1592 | ||
1593 | err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1); | |
1594 | if (err) { | |
1595 | message = "writing feature-gso-tcpv4"; | |
1596 | goto abort_transaction; | |
1597 | } | |
1598 | ||
1599 | err = xenbus_transaction_end(xbt, 0); | |
1600 | if (err) { | |
1601 | if (err == -EAGAIN) | |
1602 | goto again; | |
1603 | xenbus_dev_fatal(dev, err, "completing transaction"); | |
1604 | goto destroy_ring; | |
1605 | } | |
1606 | ||
1607 | return 0; | |
1608 | ||
1609 | abort_transaction: | |
1610 | xenbus_transaction_end(xbt, 1); | |
1611 | xenbus_dev_fatal(dev, err, "%s", message); | |
1612 | destroy_ring: | |
1613 | xennet_disconnect_backend(info); | |
1614 | out: | |
1615 | return err; | |
1616 | } | |
1617 | ||
0d160211 JF |
1618 | static int xennet_connect(struct net_device *dev) |
1619 | { | |
1620 | struct netfront_info *np = netdev_priv(dev); | |
1621 | int i, requeue_idx, err; | |
1622 | struct sk_buff *skb; | |
1623 | grant_ref_t ref; | |
1624 | struct xen_netif_rx_request *req; | |
1625 | unsigned int feature_rx_copy; | |
1626 | ||
1627 | err = xenbus_scanf(XBT_NIL, np->xbdev->otherend, | |
1628 | "feature-rx-copy", "%u", &feature_rx_copy); | |
1629 | if (err != 1) | |
1630 | feature_rx_copy = 0; | |
1631 | ||
1632 | if (!feature_rx_copy) { | |
1633 | dev_info(&dev->dev, | |
898eb71c | 1634 | "backend does not support copying receive path\n"); |
0d160211 JF |
1635 | return -ENODEV; |
1636 | } | |
1637 | ||
f502bf2b | 1638 | err = talk_to_netback(np->xbdev, np); |
0d160211 JF |
1639 | if (err) |
1640 | return err; | |
1641 | ||
1ba37c51 | 1642 | rtnl_lock(); |
fb507934 | 1643 | netdev_update_features(dev); |
1ba37c51 | 1644 | rtnl_unlock(); |
0d160211 JF |
1645 | |
1646 | spin_lock_bh(&np->rx_lock); | |
1647 | spin_lock_irq(&np->tx_lock); | |
1648 | ||
1649 | /* Step 1: Discard all pending TX packet fragments. */ | |
1650 | xennet_release_tx_bufs(np); | |
1651 | ||
1652 | /* Step 2: Rebuild the RX buffer freelist and the RX ring itself. */ | |
1653 | for (requeue_idx = 0, i = 0; i < NET_RX_RING_SIZE; i++) { | |
01c68026 IC |
1654 | skb_frag_t *frag; |
1655 | const struct page *page; | |
0d160211 JF |
1656 | if (!np->rx_skbs[i]) |
1657 | continue; | |
1658 | ||
1659 | skb = np->rx_skbs[requeue_idx] = xennet_get_rx_skb(np, i); | |
1660 | ref = np->grant_rx_ref[requeue_idx] = xennet_get_rx_ref(np, i); | |
1661 | req = RING_GET_REQUEST(&np->rx, requeue_idx); | |
1662 | ||
01c68026 IC |
1663 | frag = &skb_shinfo(skb)->frags[0]; |
1664 | page = skb_frag_page(frag); | |
0d160211 JF |
1665 | gnttab_grant_foreign_access_ref( |
1666 | ref, np->xbdev->otherend_id, | |
01c68026 | 1667 | pfn_to_mfn(page_to_pfn(page)), |
0d160211 JF |
1668 | 0); |
1669 | req->gref = ref; | |
1670 | req->id = requeue_idx; | |
1671 | ||
1672 | requeue_idx++; | |
1673 | } | |
1674 | ||
1675 | np->rx.req_prod_pvt = requeue_idx; | |
1676 | ||
1677 | /* | |
1678 | * Step 3: All public and private state should now be sane. Get | |
1679 | * ready to start sending and receiving packets and give the driver | |
1680 | * domain a kick because we've probably just requeued some | |
1681 | * packets. | |
1682 | */ | |
1683 | netif_carrier_on(np->netdev); | |
1684 | notify_remote_via_irq(np->netdev->irq); | |
1685 | xennet_tx_buf_gc(dev); | |
1686 | xennet_alloc_rx_buffers(dev); | |
1687 | ||
1688 | spin_unlock_irq(&np->tx_lock); | |
1689 | spin_unlock_bh(&np->rx_lock); | |
1690 | ||
1691 | return 0; | |
1692 | } | |
1693 | ||
1694 | /** | |
1695 | * Callback received when the backend's state changes. | |
1696 | */ | |
f502bf2b | 1697 | static void netback_changed(struct xenbus_device *dev, |
0d160211 JF |
1698 | enum xenbus_state backend_state) |
1699 | { | |
1b713e00 | 1700 | struct netfront_info *np = dev_get_drvdata(&dev->dev); |
0d160211 JF |
1701 | struct net_device *netdev = np->netdev; |
1702 | ||
1703 | dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state)); | |
1704 | ||
1705 | switch (backend_state) { | |
1706 | case XenbusStateInitialising: | |
1707 | case XenbusStateInitialised: | |
b78c9512 NI |
1708 | case XenbusStateReconfiguring: |
1709 | case XenbusStateReconfigured: | |
0d160211 JF |
1710 | case XenbusStateConnected: |
1711 | case XenbusStateUnknown: | |
1712 | case XenbusStateClosed: | |
1713 | break; | |
1714 | ||
1715 | case XenbusStateInitWait: | |
1716 | if (dev->state != XenbusStateInitialising) | |
1717 | break; | |
1718 | if (xennet_connect(netdev) != 0) | |
1719 | break; | |
1720 | xenbus_switch_state(dev, XenbusStateConnected); | |
59297067 | 1721 | netif_notify_peers(netdev); |
0d160211 JF |
1722 | break; |
1723 | ||
1724 | case XenbusStateClosing: | |
1725 | xenbus_frontend_closed(dev); | |
1726 | break; | |
1727 | } | |
1728 | } | |
1729 | ||
e0ce4af9 IC |
1730 | static const struct xennet_stat { |
1731 | char name[ETH_GSTRING_LEN]; | |
1732 | u16 offset; | |
1733 | } xennet_stats[] = { | |
1734 | { | |
1735 | "rx_gso_checksum_fixup", | |
1736 | offsetof(struct netfront_info, rx_gso_checksum_fixup) | |
1737 | }, | |
1738 | }; | |
1739 | ||
1740 | static int xennet_get_sset_count(struct net_device *dev, int string_set) | |
1741 | { | |
1742 | switch (string_set) { | |
1743 | case ETH_SS_STATS: | |
1744 | return ARRAY_SIZE(xennet_stats); | |
1745 | default: | |
1746 | return -EINVAL; | |
1747 | } | |
1748 | } | |
1749 | ||
1750 | static void xennet_get_ethtool_stats(struct net_device *dev, | |
1751 | struct ethtool_stats *stats, u64 * data) | |
1752 | { | |
1753 | void *np = netdev_priv(dev); | |
1754 | int i; | |
1755 | ||
1756 | for (i = 0; i < ARRAY_SIZE(xennet_stats); i++) | |
48f26d51 | 1757 | data[i] = *(unsigned long *)(np + xennet_stats[i].offset); |
e0ce4af9 IC |
1758 | } |
1759 | ||
1760 | static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data) | |
1761 | { | |
1762 | int i; | |
1763 | ||
1764 | switch (stringset) { | |
1765 | case ETH_SS_STATS: | |
1766 | for (i = 0; i < ARRAY_SIZE(xennet_stats); i++) | |
1767 | memcpy(data + i * ETH_GSTRING_LEN, | |
1768 | xennet_stats[i].name, ETH_GSTRING_LEN); | |
1769 | break; | |
1770 | } | |
1771 | } | |
1772 | ||
0fc0b732 | 1773 | static const struct ethtool_ops xennet_ethtool_ops = |
0d160211 | 1774 | { |
0d160211 | 1775 | .get_link = ethtool_op_get_link, |
e0ce4af9 IC |
1776 | |
1777 | .get_sset_count = xennet_get_sset_count, | |
1778 | .get_ethtool_stats = xennet_get_ethtool_stats, | |
1779 | .get_strings = xennet_get_strings, | |
0d160211 JF |
1780 | }; |
1781 | ||
1782 | #ifdef CONFIG_SYSFS | |
1783 | static ssize_t show_rxbuf_min(struct device *dev, | |
1784 | struct device_attribute *attr, char *buf) | |
1785 | { | |
1786 | struct net_device *netdev = to_net_dev(dev); | |
1787 | struct netfront_info *info = netdev_priv(netdev); | |
1788 | ||
1789 | return sprintf(buf, "%u\n", info->rx_min_target); | |
1790 | } | |
1791 | ||
1792 | static ssize_t store_rxbuf_min(struct device *dev, | |
1793 | struct device_attribute *attr, | |
1794 | const char *buf, size_t len) | |
1795 | { | |
1796 | struct net_device *netdev = to_net_dev(dev); | |
1797 | struct netfront_info *np = netdev_priv(netdev); | |
1798 | char *endp; | |
1799 | unsigned long target; | |
1800 | ||
1801 | if (!capable(CAP_NET_ADMIN)) | |
1802 | return -EPERM; | |
1803 | ||
1804 | target = simple_strtoul(buf, &endp, 0); | |
1805 | if (endp == buf) | |
1806 | return -EBADMSG; | |
1807 | ||
1808 | if (target < RX_MIN_TARGET) | |
1809 | target = RX_MIN_TARGET; | |
1810 | if (target > RX_MAX_TARGET) | |
1811 | target = RX_MAX_TARGET; | |
1812 | ||
1813 | spin_lock_bh(&np->rx_lock); | |
1814 | if (target > np->rx_max_target) | |
1815 | np->rx_max_target = target; | |
1816 | np->rx_min_target = target; | |
1817 | if (target > np->rx_target) | |
1818 | np->rx_target = target; | |
1819 | ||
1820 | xennet_alloc_rx_buffers(netdev); | |
1821 | ||
1822 | spin_unlock_bh(&np->rx_lock); | |
1823 | return len; | |
1824 | } | |
1825 | ||
1826 | static ssize_t show_rxbuf_max(struct device *dev, | |
1827 | struct device_attribute *attr, char *buf) | |
1828 | { | |
1829 | struct net_device *netdev = to_net_dev(dev); | |
1830 | struct netfront_info *info = netdev_priv(netdev); | |
1831 | ||
1832 | return sprintf(buf, "%u\n", info->rx_max_target); | |
1833 | } | |
1834 | ||
1835 | static ssize_t store_rxbuf_max(struct device *dev, | |
1836 | struct device_attribute *attr, | |
1837 | const char *buf, size_t len) | |
1838 | { | |
1839 | struct net_device *netdev = to_net_dev(dev); | |
1840 | struct netfront_info *np = netdev_priv(netdev); | |
1841 | char *endp; | |
1842 | unsigned long target; | |
1843 | ||
1844 | if (!capable(CAP_NET_ADMIN)) | |
1845 | return -EPERM; | |
1846 | ||
1847 | target = simple_strtoul(buf, &endp, 0); | |
1848 | if (endp == buf) | |
1849 | return -EBADMSG; | |
1850 | ||
1851 | if (target < RX_MIN_TARGET) | |
1852 | target = RX_MIN_TARGET; | |
1853 | if (target > RX_MAX_TARGET) | |
1854 | target = RX_MAX_TARGET; | |
1855 | ||
1856 | spin_lock_bh(&np->rx_lock); | |
1857 | if (target < np->rx_min_target) | |
1858 | np->rx_min_target = target; | |
1859 | np->rx_max_target = target; | |
1860 | if (target < np->rx_target) | |
1861 | np->rx_target = target; | |
1862 | ||
1863 | xennet_alloc_rx_buffers(netdev); | |
1864 | ||
1865 | spin_unlock_bh(&np->rx_lock); | |
1866 | return len; | |
1867 | } | |
1868 | ||
1869 | static ssize_t show_rxbuf_cur(struct device *dev, | |
1870 | struct device_attribute *attr, char *buf) | |
1871 | { | |
1872 | struct net_device *netdev = to_net_dev(dev); | |
1873 | struct netfront_info *info = netdev_priv(netdev); | |
1874 | ||
1875 | return sprintf(buf, "%u\n", info->rx_target); | |
1876 | } | |
1877 | ||
1878 | static struct device_attribute xennet_attrs[] = { | |
1879 | __ATTR(rxbuf_min, S_IRUGO|S_IWUSR, show_rxbuf_min, store_rxbuf_min), | |
1880 | __ATTR(rxbuf_max, S_IRUGO|S_IWUSR, show_rxbuf_max, store_rxbuf_max), | |
1881 | __ATTR(rxbuf_cur, S_IRUGO, show_rxbuf_cur, NULL), | |
1882 | }; | |
1883 | ||
1884 | static int xennet_sysfs_addif(struct net_device *netdev) | |
1885 | { | |
1886 | int i; | |
1887 | int err; | |
1888 | ||
1889 | for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++) { | |
1890 | err = device_create_file(&netdev->dev, | |
1891 | &xennet_attrs[i]); | |
1892 | if (err) | |
1893 | goto fail; | |
1894 | } | |
1895 | return 0; | |
1896 | ||
1897 | fail: | |
1898 | while (--i >= 0) | |
1899 | device_remove_file(&netdev->dev, &xennet_attrs[i]); | |
1900 | return err; | |
1901 | } | |
1902 | ||
1903 | static void xennet_sysfs_delif(struct net_device *netdev) | |
1904 | { | |
1905 | int i; | |
1906 | ||
1907 | for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++) | |
1908 | device_remove_file(&netdev->dev, &xennet_attrs[i]); | |
1909 | } | |
1910 | ||
1911 | #endif /* CONFIG_SYSFS */ | |
1912 | ||
1913 | static struct xenbus_device_id netfront_ids[] = { | |
1914 | { "vif" }, | |
1915 | { "" } | |
1916 | }; | |
1917 | ||
1918 | ||
1919 | static int __devexit xennet_remove(struct xenbus_device *dev) | |
1920 | { | |
1b713e00 | 1921 | struct netfront_info *info = dev_get_drvdata(&dev->dev); |
0d160211 JF |
1922 | |
1923 | dev_dbg(&dev->dev, "%s\n", dev->nodename); | |
1924 | ||
1925 | unregister_netdev(info->netdev); | |
1926 | ||
1927 | xennet_disconnect_backend(info); | |
1928 | ||
1929 | del_timer_sync(&info->rx_refill_timer); | |
1930 | ||
1931 | xennet_sysfs_delif(info->netdev); | |
1932 | ||
e00f85be | 1933 | free_percpu(info->stats); |
1934 | ||
0d160211 JF |
1935 | free_netdev(info->netdev); |
1936 | ||
1937 | return 0; | |
1938 | } | |
1939 | ||
ffb78a26 | 1940 | static struct xenbus_driver netfront_driver = { |
0d160211 JF |
1941 | .name = "vif", |
1942 | .owner = THIS_MODULE, | |
1943 | .ids = netfront_ids, | |
1944 | .probe = netfront_probe, | |
1945 | .remove = __devexit_p(xennet_remove), | |
1946 | .resume = netfront_resume, | |
f502bf2b | 1947 | .otherend_changed = netback_changed, |
0d160211 JF |
1948 | }; |
1949 | ||
1950 | static int __init netif_init(void) | |
1951 | { | |
6e833587 | 1952 | if (!xen_domain()) |
0d160211 JF |
1953 | return -ENODEV; |
1954 | ||
6e833587 | 1955 | if (xen_initial_domain()) |
0d160211 JF |
1956 | return 0; |
1957 | ||
1958 | printk(KERN_INFO "Initialising Xen virtual ethernet driver.\n"); | |
1959 | ||
ffb78a26 | 1960 | return xenbus_register_frontend(&netfront_driver); |
0d160211 JF |
1961 | } |
1962 | module_init(netif_init); | |
1963 | ||
1964 | ||
1965 | static void __exit netif_exit(void) | |
1966 | { | |
6e833587 | 1967 | if (xen_initial_domain()) |
0d160211 JF |
1968 | return; |
1969 | ||
ffb78a26 | 1970 | xenbus_unregister_driver(&netfront_driver); |
0d160211 JF |
1971 | } |
1972 | module_exit(netif_exit); | |
1973 | ||
1974 | MODULE_DESCRIPTION("Xen virtual network device frontend"); | |
1975 | MODULE_LICENSE("GPL"); | |
d2f0c52b | 1976 | MODULE_ALIAS("xen:vif"); |
4f93f09b | 1977 | MODULE_ALIAS("xennet"); |