Commit | Line | Data |
---|---|---|
f942dc25 IC |
1 | /* |
2 | * Back-end of the driver for virtual network devices. This portion of the | |
3 | * driver exports a 'unified' network-device interface that can be accessed | |
4 | * by any operating system that implements a compatible front end. A | |
5 | * reference front-end implementation can be found in: | |
6 | * drivers/net/xen-netfront.c | |
7 | * | |
8 | * Copyright (c) 2002-2005, K A Fraser | |
9 | * | |
10 | * This program is free software; you can redistribute it and/or | |
11 | * modify it under the terms of the GNU General Public License version 2 | |
12 | * as published by the Free Software Foundation; or, when distributed | |
13 | * separately from the Linux kernel or incorporated into other | |
14 | * software packages, subject to the following license: | |
15 | * | |
16 | * Permission is hereby granted, free of charge, to any person obtaining a copy | |
17 | * of this source file (the "Software"), to deal in the Software without | |
18 | * restriction, including without limitation the rights to use, copy, modify, | |
19 | * merge, publish, distribute, sublicense, and/or sell copies of the Software, | |
20 | * and to permit persons to whom the Software is furnished to do so, subject to | |
21 | * the following conditions: | |
22 | * | |
23 | * The above copyright notice and this permission notice shall be included in | |
24 | * all copies or substantial portions of the Software. | |
25 | * | |
26 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | |
27 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | |
28 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE | |
29 | * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER | |
30 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING | |
31 | * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS | |
32 | * IN THE SOFTWARE. | |
33 | */ | |
34 | ||
35 | #include "common.h" | |
36 | ||
37 | #include <linux/kthread.h> | |
38 | #include <linux/if_vlan.h> | |
39 | #include <linux/udp.h> | |
e3377f36 | 40 | #include <linux/highmem.h> |
f942dc25 IC |
41 | |
42 | #include <net/tcp.h> | |
43 | ||
ca981633 | 44 | #include <xen/xen.h> |
f942dc25 IC |
45 | #include <xen/events.h> |
46 | #include <xen/interface/memory.h> | |
47 | ||
48 | #include <asm/xen/hypercall.h> | |
49 | #include <asm/xen/page.h> | |
50 | ||
e1f00a69 WL |
51 | /* Provide an option to disable split event channels at load time as |
52 | * event channels are limited resource. Split event channels are | |
53 | * enabled by default. | |
54 | */ | |
55 | bool separate_tx_rx_irq = 1; | |
56 | module_param(separate_tx_rx_irq, bool, 0644); | |
57 | ||
09350788 | 58 | /* When guest ring is filled up, qdisc queues the packets for us, but we have |
0e59a4a5 | 59 | * to timeout them, otherwise other guests' packets can get stuck there |
09350788 ZK |
60 | */ |
61 | unsigned int rx_drain_timeout_msecs = 10000; | |
62 | module_param(rx_drain_timeout_msecs, uint, 0444); | |
63 | unsigned int rx_drain_timeout_jiffies; | |
64 | ||
8d3d53b3 AB |
65 | unsigned int xenvif_max_queues; |
66 | module_param_named(max_queues, xenvif_max_queues, uint, 0644); | |
67 | MODULE_PARM_DESC(max_queues, | |
68 | "Maximum number of queues per virtual interface"); | |
69 | ||
2810e5b9 WL |
70 | /* |
71 | * This is the maximum slots a skb can have. If a guest sends a skb | |
72 | * which exceeds this limit it is considered malicious. | |
73 | */ | |
37641494 WL |
74 | #define FATAL_SKB_SLOTS_DEFAULT 20 |
75 | static unsigned int fatal_skb_slots = FATAL_SKB_SLOTS_DEFAULT; | |
76 | module_param(fatal_skb_slots, uint, 0444); | |
77 | ||
e9ce7cb6 | 78 | static void xenvif_idx_release(struct xenvif_queue *queue, u16 pending_idx, |
7376419a WL |
79 | u8 status); |
80 | ||
e9ce7cb6 | 81 | static void make_tx_response(struct xenvif_queue *queue, |
f942dc25 IC |
82 | struct xen_netif_tx_request *txp, |
83 | s8 st); | |
b3f980bd | 84 | |
e9ce7cb6 WL |
85 | static inline int tx_work_todo(struct xenvif_queue *queue); |
86 | static inline int rx_work_todo(struct xenvif_queue *queue); | |
b3f980bd | 87 | |
e9ce7cb6 | 88 | static struct xen_netif_rx_response *make_rx_response(struct xenvif_queue *queue, |
f942dc25 IC |
89 | u16 id, |
90 | s8 st, | |
91 | u16 offset, | |
92 | u16 size, | |
93 | u16 flags); | |
94 | ||
e9ce7cb6 | 95 | static inline unsigned long idx_to_pfn(struct xenvif_queue *queue, |
ea066ad1 | 96 | u16 idx) |
f942dc25 | 97 | { |
e9ce7cb6 | 98 | return page_to_pfn(queue->mmap_pages[idx]); |
f942dc25 IC |
99 | } |
100 | ||
e9ce7cb6 | 101 | static inline unsigned long idx_to_kaddr(struct xenvif_queue *queue, |
ea066ad1 | 102 | u16 idx) |
f942dc25 | 103 | { |
e9ce7cb6 | 104 | return (unsigned long)pfn_to_kaddr(idx_to_pfn(queue, idx)); |
f942dc25 IC |
105 | } |
106 | ||
7aceb47a ZK |
107 | #define callback_param(vif, pending_idx) \ |
108 | (vif->pending_tx_info[pending_idx].callback_struct) | |
109 | ||
f53c3fe8 ZK |
110 | /* Find the containing VIF's structure from a pointer in pending_tx_info array |
111 | */ | |
e9ce7cb6 | 112 | static inline struct xenvif_queue *ubuf_to_queue(const struct ubuf_info *ubuf) |
3e2234b3 | 113 | { |
f53c3fe8 ZK |
114 | u16 pending_idx = ubuf->desc; |
115 | struct pending_tx_info *temp = | |
116 | container_of(ubuf, struct pending_tx_info, callback_struct); | |
117 | return container_of(temp - pending_idx, | |
e9ce7cb6 | 118 | struct xenvif_queue, |
f53c3fe8 | 119 | pending_tx_info[0]); |
3e2234b3 | 120 | } |
f53c3fe8 | 121 | |
2eba61d5 PD |
122 | /* This is a miniumum size for the linear area to avoid lots of |
123 | * calls to __pskb_pull_tail() as we set up checksum offsets. The | |
124 | * value 128 was chosen as it covers all IPv4 and most likely | |
125 | * IPv6 headers. | |
f942dc25 | 126 | */ |
2eba61d5 | 127 | #define PKT_PROT_LEN 128 |
f942dc25 | 128 | |
ea066ad1 IC |
129 | static u16 frag_get_pending_idx(skb_frag_t *frag) |
130 | { | |
131 | return (u16)frag->page_offset; | |
132 | } | |
133 | ||
134 | static void frag_set_pending_idx(skb_frag_t *frag, u16 pending_idx) | |
135 | { | |
136 | frag->page_offset = pending_idx; | |
137 | } | |
138 | ||
f942dc25 IC |
139 | static inline pending_ring_idx_t pending_index(unsigned i) |
140 | { | |
141 | return i & (MAX_PENDING_REQS-1); | |
142 | } | |
143 | ||
e9ce7cb6 | 144 | bool xenvif_rx_ring_slots_available(struct xenvif_queue *queue, int needed) |
f942dc25 | 145 | { |
ca2f09f2 | 146 | RING_IDX prod, cons; |
f942dc25 | 147 | |
ca2f09f2 | 148 | do { |
e9ce7cb6 WL |
149 | prod = queue->rx.sring->req_prod; |
150 | cons = queue->rx.req_cons; | |
f942dc25 | 151 | |
ca2f09f2 PD |
152 | if (prod - cons >= needed) |
153 | return true; | |
f942dc25 | 154 | |
e9ce7cb6 | 155 | queue->rx.sring->req_event = prod + 1; |
f942dc25 | 156 | |
ca2f09f2 PD |
157 | /* Make sure event is visible before we check prod |
158 | * again. | |
159 | */ | |
160 | mb(); | |
e9ce7cb6 | 161 | } while (queue->rx.sring->req_prod != prod); |
f942dc25 | 162 | |
ca2f09f2 | 163 | return false; |
f942dc25 IC |
164 | } |
165 | ||
166 | /* | |
167 | * Returns true if we should start a new receive buffer instead of | |
168 | * adding 'size' bytes to a buffer which currently contains 'offset' | |
169 | * bytes. | |
170 | */ | |
171 | static bool start_new_rx_buffer(int offset, unsigned long size, int head) | |
172 | { | |
173 | /* simple case: we have completely filled the current buffer. */ | |
174 | if (offset == MAX_BUFFER_OFFSET) | |
175 | return true; | |
176 | ||
177 | /* | |
178 | * complex case: start a fresh buffer if the current frag | |
179 | * would overflow the current buffer but only if: | |
180 | * (i) this frag would fit completely in the next buffer | |
181 | * and (ii) there is already some data in the current buffer | |
182 | * and (iii) this is not the head buffer. | |
183 | * | |
184 | * Where: | |
185 | * - (i) stops us splitting a frag into two copies | |
186 | * unless the frag is too large for a single buffer. | |
187 | * - (ii) stops us from leaving a buffer pointlessly empty. | |
188 | * - (iii) stops us leaving the first buffer | |
189 | * empty. Strictly speaking this is already covered | |
190 | * by (ii) but is explicitly checked because | |
191 | * netfront relies on the first buffer being | |
192 | * non-empty and can crash otherwise. | |
193 | * | |
194 | * This means we will effectively linearise small | |
195 | * frags but do not needlessly split large buffers | |
196 | * into multiple copies tend to give large frags their | |
197 | * own buffers as before. | |
198 | */ | |
0576eddf PD |
199 | BUG_ON(size > MAX_BUFFER_OFFSET); |
200 | if ((offset + size > MAX_BUFFER_OFFSET) && offset && !head) | |
f942dc25 IC |
201 | return true; |
202 | ||
203 | return false; | |
204 | } | |
205 | ||
f942dc25 IC |
206 | struct netrx_pending_operations { |
207 | unsigned copy_prod, copy_cons; | |
208 | unsigned meta_prod, meta_cons; | |
209 | struct gnttab_copy *copy; | |
b3f980bd | 210 | struct xenvif_rx_meta *meta; |
f942dc25 IC |
211 | int copy_off; |
212 | grant_ref_t copy_gref; | |
213 | }; | |
214 | ||
e9ce7cb6 | 215 | static struct xenvif_rx_meta *get_next_rx_buffer(struct xenvif_queue *queue, |
b3f980bd | 216 | struct netrx_pending_operations *npo) |
f942dc25 | 217 | { |
b3f980bd | 218 | struct xenvif_rx_meta *meta; |
f942dc25 IC |
219 | struct xen_netif_rx_request *req; |
220 | ||
e9ce7cb6 | 221 | req = RING_GET_REQUEST(&queue->rx, queue->rx.req_cons++); |
f942dc25 IC |
222 | |
223 | meta = npo->meta + npo->meta_prod++; | |
82cada22 | 224 | meta->gso_type = XEN_NETIF_GSO_TYPE_NONE; |
f942dc25 IC |
225 | meta->gso_size = 0; |
226 | meta->size = 0; | |
227 | meta->id = req->id; | |
228 | ||
229 | npo->copy_off = 0; | |
230 | npo->copy_gref = req->gref; | |
231 | ||
232 | return meta; | |
233 | } | |
234 | ||
33bc801d WL |
235 | /* |
236 | * Set up the grant operations for this fragment. If it's a flipping | |
237 | * interface, we also set up the unmap request from here. | |
238 | */ | |
e9ce7cb6 | 239 | static void xenvif_gop_frag_copy(struct xenvif_queue *queue, struct sk_buff *skb, |
7376419a WL |
240 | struct netrx_pending_operations *npo, |
241 | struct page *page, unsigned long size, | |
3e2234b3 | 242 | unsigned long offset, int *head, |
e9ce7cb6 | 243 | struct xenvif_queue *foreign_queue, |
3e2234b3 | 244 | grant_ref_t foreign_gref) |
f942dc25 IC |
245 | { |
246 | struct gnttab_copy *copy_gop; | |
b3f980bd | 247 | struct xenvif_rx_meta *meta; |
f942dc25 | 248 | unsigned long bytes; |
5bd07670 | 249 | int gso_type = XEN_NETIF_GSO_TYPE_NONE; |
f942dc25 IC |
250 | |
251 | /* Data must not cross a page boundary. */ | |
6a8ed462 | 252 | BUG_ON(size + offset > PAGE_SIZE<<compound_order(page)); |
f942dc25 IC |
253 | |
254 | meta = npo->meta + npo->meta_prod - 1; | |
255 | ||
6a8ed462 IC |
256 | /* Skip unused frames from start of page */ |
257 | page += offset >> PAGE_SHIFT; | |
258 | offset &= ~PAGE_MASK; | |
259 | ||
f942dc25 | 260 | while (size > 0) { |
6a8ed462 | 261 | BUG_ON(offset >= PAGE_SIZE); |
f942dc25 IC |
262 | BUG_ON(npo->copy_off > MAX_BUFFER_OFFSET); |
263 | ||
6a8ed462 IC |
264 | bytes = PAGE_SIZE - offset; |
265 | ||
266 | if (bytes > size) | |
267 | bytes = size; | |
268 | ||
33bc801d | 269 | if (start_new_rx_buffer(npo->copy_off, bytes, *head)) { |
f942dc25 IC |
270 | /* |
271 | * Netfront requires there to be some data in the head | |
272 | * buffer. | |
273 | */ | |
33bc801d | 274 | BUG_ON(*head); |
f942dc25 | 275 | |
e9ce7cb6 | 276 | meta = get_next_rx_buffer(queue, npo); |
f942dc25 IC |
277 | } |
278 | ||
f942dc25 IC |
279 | if (npo->copy_off + bytes > MAX_BUFFER_OFFSET) |
280 | bytes = MAX_BUFFER_OFFSET - npo->copy_off; | |
281 | ||
282 | copy_gop = npo->copy + npo->copy_prod++; | |
283 | copy_gop->flags = GNTCOPY_dest_gref; | |
b3f980bd WL |
284 | copy_gop->len = bytes; |
285 | ||
e9ce7cb6 WL |
286 | if (foreign_queue) { |
287 | copy_gop->source.domid = foreign_queue->vif->domid; | |
3e2234b3 ZK |
288 | copy_gop->source.u.ref = foreign_gref; |
289 | copy_gop->flags |= GNTCOPY_source_gref; | |
290 | } else { | |
291 | copy_gop->source.domid = DOMID_SELF; | |
292 | copy_gop->source.u.gmfn = | |
293 | virt_to_mfn(page_address(page)); | |
294 | } | |
f942dc25 | 295 | copy_gop->source.offset = offset; |
f942dc25 | 296 | |
e9ce7cb6 | 297 | copy_gop->dest.domid = queue->vif->domid; |
f942dc25 IC |
298 | copy_gop->dest.offset = npo->copy_off; |
299 | copy_gop->dest.u.ref = npo->copy_gref; | |
f942dc25 IC |
300 | |
301 | npo->copy_off += bytes; | |
302 | meta->size += bytes; | |
303 | ||
304 | offset += bytes; | |
305 | size -= bytes; | |
306 | ||
6a8ed462 IC |
307 | /* Next frame */ |
308 | if (offset == PAGE_SIZE && size) { | |
309 | BUG_ON(!PageCompound(page)); | |
310 | page++; | |
311 | offset = 0; | |
312 | } | |
313 | ||
f942dc25 | 314 | /* Leave a gap for the GSO descriptor. */ |
5bd07670 AL |
315 | if (skb_is_gso(skb)) { |
316 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) | |
317 | gso_type = XEN_NETIF_GSO_TYPE_TCPV4; | |
318 | else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) | |
319 | gso_type = XEN_NETIF_GSO_TYPE_TCPV6; | |
320 | } | |
82cada22 | 321 | |
e9ce7cb6 WL |
322 | if (*head && ((1 << gso_type) & queue->vif->gso_mask)) |
323 | queue->rx.req_cons++; | |
f942dc25 | 324 | |
33bc801d | 325 | *head = 0; /* There must be something in this buffer now. */ |
f942dc25 IC |
326 | |
327 | } | |
328 | } | |
329 | ||
58375744 ZK |
330 | /* |
331 | * Find the grant ref for a given frag in a chain of struct ubuf_info's | |
332 | * skb: the skb itself | |
333 | * i: the frag's number | |
334 | * ubuf: a pointer to an element in the chain. It should not be NULL | |
335 | * | |
336 | * Returns a pointer to the element in the chain where the page were found. If | |
337 | * not found, returns NULL. | |
338 | * See the definition of callback_struct in common.h for more details about | |
339 | * the chain. | |
340 | */ | |
341 | static const struct ubuf_info *xenvif_find_gref(const struct sk_buff *const skb, | |
342 | const int i, | |
343 | const struct ubuf_info *ubuf) | |
344 | { | |
e9ce7cb6 | 345 | struct xenvif_queue *foreign_queue = ubuf_to_queue(ubuf); |
58375744 ZK |
346 | |
347 | do { | |
348 | u16 pending_idx = ubuf->desc; | |
349 | ||
350 | if (skb_shinfo(skb)->frags[i].page.p == | |
e9ce7cb6 | 351 | foreign_queue->mmap_pages[pending_idx]) |
58375744 ZK |
352 | break; |
353 | ubuf = (struct ubuf_info *) ubuf->ctx; | |
354 | } while (ubuf); | |
355 | ||
356 | return ubuf; | |
357 | } | |
358 | ||
f942dc25 IC |
359 | /* |
360 | * Prepare an SKB to be transmitted to the frontend. | |
361 | * | |
362 | * This function is responsible for allocating grant operations, meta | |
363 | * structures, etc. | |
364 | * | |
365 | * It returns the number of meta structures consumed. The number of | |
366 | * ring slots used is always equal to the number of meta slots used | |
367 | * plus the number of GSO descriptors used. Currently, we use either | |
368 | * zero GSO descriptors (for non-GSO packets) or one descriptor (for | |
369 | * frontend-side LRO). | |
370 | */ | |
7376419a | 371 | static int xenvif_gop_skb(struct sk_buff *skb, |
e9ce7cb6 WL |
372 | struct netrx_pending_operations *npo, |
373 | struct xenvif_queue *queue) | |
f942dc25 IC |
374 | { |
375 | struct xenvif *vif = netdev_priv(skb->dev); | |
376 | int nr_frags = skb_shinfo(skb)->nr_frags; | |
377 | int i; | |
378 | struct xen_netif_rx_request *req; | |
b3f980bd | 379 | struct xenvif_rx_meta *meta; |
f942dc25 | 380 | unsigned char *data; |
33bc801d | 381 | int head = 1; |
f942dc25 | 382 | int old_meta_prod; |
82cada22 | 383 | int gso_type; |
58375744 ZK |
384 | const struct ubuf_info *ubuf = skb_shinfo(skb)->destructor_arg; |
385 | const struct ubuf_info *const head_ubuf = ubuf; | |
f942dc25 IC |
386 | |
387 | old_meta_prod = npo->meta_prod; | |
388 | ||
5bd07670 AL |
389 | gso_type = XEN_NETIF_GSO_TYPE_NONE; |
390 | if (skb_is_gso(skb)) { | |
391 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) | |
392 | gso_type = XEN_NETIF_GSO_TYPE_TCPV4; | |
393 | else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) | |
394 | gso_type = XEN_NETIF_GSO_TYPE_TCPV6; | |
82cada22 PD |
395 | } |
396 | ||
f942dc25 | 397 | /* Set up a GSO prefix descriptor, if necessary */ |
a3314f3d | 398 | if ((1 << gso_type) & vif->gso_prefix_mask) { |
e9ce7cb6 | 399 | req = RING_GET_REQUEST(&queue->rx, queue->rx.req_cons++); |
f942dc25 | 400 | meta = npo->meta + npo->meta_prod++; |
82cada22 | 401 | meta->gso_type = gso_type; |
5bd07670 | 402 | meta->gso_size = skb_shinfo(skb)->gso_size; |
f942dc25 IC |
403 | meta->size = 0; |
404 | meta->id = req->id; | |
405 | } | |
406 | ||
e9ce7cb6 | 407 | req = RING_GET_REQUEST(&queue->rx, queue->rx.req_cons++); |
f942dc25 IC |
408 | meta = npo->meta + npo->meta_prod++; |
409 | ||
82cada22 PD |
410 | if ((1 << gso_type) & vif->gso_mask) { |
411 | meta->gso_type = gso_type; | |
5bd07670 | 412 | meta->gso_size = skb_shinfo(skb)->gso_size; |
82cada22 PD |
413 | } else { |
414 | meta->gso_type = XEN_NETIF_GSO_TYPE_NONE; | |
f942dc25 | 415 | meta->gso_size = 0; |
82cada22 | 416 | } |
f942dc25 IC |
417 | |
418 | meta->size = 0; | |
419 | meta->id = req->id; | |
420 | npo->copy_off = 0; | |
421 | npo->copy_gref = req->gref; | |
422 | ||
423 | data = skb->data; | |
424 | while (data < skb_tail_pointer(skb)) { | |
425 | unsigned int offset = offset_in_page(data); | |
426 | unsigned int len = PAGE_SIZE - offset; | |
427 | ||
428 | if (data + len > skb_tail_pointer(skb)) | |
429 | len = skb_tail_pointer(skb) - data; | |
430 | ||
e9ce7cb6 | 431 | xenvif_gop_frag_copy(queue, skb, npo, |
3e2234b3 ZK |
432 | virt_to_page(data), len, offset, &head, |
433 | NULL, | |
434 | 0); | |
f942dc25 IC |
435 | data += len; |
436 | } | |
437 | ||
438 | for (i = 0; i < nr_frags; i++) { | |
58375744 ZK |
439 | /* This variable also signals whether foreign_gref has a real |
440 | * value or not. | |
441 | */ | |
e9ce7cb6 | 442 | struct xenvif_queue *foreign_queue = NULL; |
58375744 ZK |
443 | grant_ref_t foreign_gref; |
444 | ||
445 | if ((skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) && | |
446 | (ubuf->callback == &xenvif_zerocopy_callback)) { | |
447 | const struct ubuf_info *const startpoint = ubuf; | |
448 | ||
449 | /* Ideally ubuf points to the chain element which | |
450 | * belongs to this frag. Or if frags were removed from | |
451 | * the beginning, then shortly before it. | |
452 | */ | |
453 | ubuf = xenvif_find_gref(skb, i, ubuf); | |
454 | ||
455 | /* Try again from the beginning of the list, if we | |
456 | * haven't tried from there. This only makes sense in | |
457 | * the unlikely event of reordering the original frags. | |
458 | * For injected local pages it's an unnecessary second | |
459 | * run. | |
460 | */ | |
461 | if (unlikely(!ubuf) && startpoint != head_ubuf) | |
462 | ubuf = xenvif_find_gref(skb, i, head_ubuf); | |
463 | ||
464 | if (likely(ubuf)) { | |
465 | u16 pending_idx = ubuf->desc; | |
466 | ||
e9ce7cb6 WL |
467 | foreign_queue = ubuf_to_queue(ubuf); |
468 | foreign_gref = | |
469 | foreign_queue->pending_tx_info[pending_idx].req.gref; | |
58375744 ZK |
470 | /* Just a safety measure. If this was the last |
471 | * element on the list, the for loop will | |
472 | * iterate again if a local page were added to | |
473 | * the end. Using head_ubuf here prevents the | |
474 | * second search on the chain. Or the original | |
475 | * frags changed order, but that's less likely. | |
476 | * In any way, ubuf shouldn't be NULL. | |
477 | */ | |
478 | ubuf = ubuf->ctx ? | |
479 | (struct ubuf_info *) ubuf->ctx : | |
480 | head_ubuf; | |
481 | } else | |
482 | /* This frag was a local page, added to the | |
483 | * array after the skb left netback. | |
484 | */ | |
485 | ubuf = head_ubuf; | |
486 | } | |
e9ce7cb6 | 487 | xenvif_gop_frag_copy(queue, skb, npo, |
7376419a WL |
488 | skb_frag_page(&skb_shinfo(skb)->frags[i]), |
489 | skb_frag_size(&skb_shinfo(skb)->frags[i]), | |
490 | skb_shinfo(skb)->frags[i].page_offset, | |
3e2234b3 | 491 | &head, |
e9ce7cb6 WL |
492 | foreign_queue, |
493 | foreign_queue ? foreign_gref : UINT_MAX); | |
f942dc25 IC |
494 | } |
495 | ||
496 | return npo->meta_prod - old_meta_prod; | |
497 | } | |
498 | ||
499 | /* | |
7376419a | 500 | * This is a twin to xenvif_gop_skb. Assume that xenvif_gop_skb was |
f942dc25 IC |
501 | * used to set up the operations on the top of |
502 | * netrx_pending_operations, which have since been done. Check that | |
503 | * they didn't give any errors and advance over them. | |
504 | */ | |
7376419a WL |
505 | static int xenvif_check_gop(struct xenvif *vif, int nr_meta_slots, |
506 | struct netrx_pending_operations *npo) | |
f942dc25 IC |
507 | { |
508 | struct gnttab_copy *copy_op; | |
509 | int status = XEN_NETIF_RSP_OKAY; | |
510 | int i; | |
511 | ||
512 | for (i = 0; i < nr_meta_slots; i++) { | |
513 | copy_op = npo->copy + npo->copy_cons++; | |
514 | if (copy_op->status != GNTST_okay) { | |
515 | netdev_dbg(vif->dev, | |
516 | "Bad status %d from copy to DOM%d.\n", | |
517 | copy_op->status, vif->domid); | |
518 | status = XEN_NETIF_RSP_ERROR; | |
519 | } | |
520 | } | |
521 | ||
522 | return status; | |
523 | } | |
524 | ||
e9ce7cb6 | 525 | static void xenvif_add_frag_responses(struct xenvif_queue *queue, int status, |
7376419a WL |
526 | struct xenvif_rx_meta *meta, |
527 | int nr_meta_slots) | |
f942dc25 IC |
528 | { |
529 | int i; | |
530 | unsigned long offset; | |
531 | ||
532 | /* No fragments used */ | |
533 | if (nr_meta_slots <= 1) | |
534 | return; | |
535 | ||
536 | nr_meta_slots--; | |
537 | ||
538 | for (i = 0; i < nr_meta_slots; i++) { | |
539 | int flags; | |
540 | if (i == nr_meta_slots - 1) | |
541 | flags = 0; | |
542 | else | |
543 | flags = XEN_NETRXF_more_data; | |
544 | ||
545 | offset = 0; | |
e9ce7cb6 | 546 | make_rx_response(queue, meta[i].id, status, offset, |
f942dc25 IC |
547 | meta[i].size, flags); |
548 | } | |
549 | } | |
550 | ||
8f13dd96 | 551 | struct xenvif_rx_cb { |
33bc801d WL |
552 | int meta_slots_used; |
553 | }; | |
554 | ||
8f13dd96 ZK |
555 | #define XENVIF_RX_CB(skb) ((struct xenvif_rx_cb *)(skb)->cb) |
556 | ||
e9ce7cb6 | 557 | void xenvif_kick_thread(struct xenvif_queue *queue) |
b3f980bd | 558 | { |
e9ce7cb6 | 559 | wake_up(&queue->wq); |
b3f980bd WL |
560 | } |
561 | ||
e9ce7cb6 | 562 | static void xenvif_rx_action(struct xenvif_queue *queue) |
f942dc25 | 563 | { |
f942dc25 | 564 | s8 status; |
e1f00a69 | 565 | u16 flags; |
f942dc25 IC |
566 | struct xen_netif_rx_response *resp; |
567 | struct sk_buff_head rxq; | |
568 | struct sk_buff *skb; | |
569 | LIST_HEAD(notify); | |
570 | int ret; | |
f942dc25 | 571 | unsigned long offset; |
11b57f90 | 572 | bool need_to_notify = false; |
f942dc25 IC |
573 | |
574 | struct netrx_pending_operations npo = { | |
e9ce7cb6 WL |
575 | .copy = queue->grant_copy_op, |
576 | .meta = queue->meta, | |
f942dc25 IC |
577 | }; |
578 | ||
579 | skb_queue_head_init(&rxq); | |
580 | ||
e9ce7cb6 | 581 | while ((skb = skb_dequeue(&queue->rx_queue)) != NULL) { |
9ab9831b | 582 | RING_IDX max_slots_needed; |
1425c7a4 PD |
583 | RING_IDX old_req_cons; |
584 | RING_IDX ring_slots_used; | |
ca2f09f2 PD |
585 | int i; |
586 | ||
587 | /* We need a cheap worse case estimate for the number of | |
588 | * slots we'll use. | |
589 | */ | |
590 | ||
591 | max_slots_needed = DIV_ROUND_UP(offset_in_page(skb->data) + | |
592 | skb_headlen(skb), | |
593 | PAGE_SIZE); | |
594 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
595 | unsigned int size; | |
a02eb473 PD |
596 | unsigned int offset; |
597 | ||
ca2f09f2 | 598 | size = skb_frag_size(&skb_shinfo(skb)->frags[i]); |
a02eb473 PD |
599 | offset = skb_shinfo(skb)->frags[i].page_offset; |
600 | ||
601 | /* For a worse-case estimate we need to factor in | |
602 | * the fragment page offset as this will affect the | |
603 | * number of times xenvif_gop_frag_copy() will | |
604 | * call start_new_rx_buffer(). | |
605 | */ | |
606 | max_slots_needed += DIV_ROUND_UP(offset + size, | |
607 | PAGE_SIZE); | |
ca2f09f2 | 608 | } |
a02eb473 PD |
609 | |
610 | /* To avoid the estimate becoming too pessimal for some | |
611 | * frontends that limit posted rx requests, cap the estimate | |
612 | * at MAX_SKB_FRAGS. | |
613 | */ | |
614 | if (max_slots_needed > MAX_SKB_FRAGS) | |
615 | max_slots_needed = MAX_SKB_FRAGS; | |
616 | ||
617 | /* We may need one more slot for GSO metadata */ | |
5bd07670 AL |
618 | if (skb_is_gso(skb) && |
619 | (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4 || | |
620 | skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)) | |
ca2f09f2 PD |
621 | max_slots_needed++; |
622 | ||
623 | /* If the skb may not fit then bail out now */ | |
e9ce7cb6 WL |
624 | if (!xenvif_rx_ring_slots_available(queue, max_slots_needed)) { |
625 | skb_queue_head(&queue->rx_queue, skb); | |
11b57f90 | 626 | need_to_notify = true; |
e9ce7cb6 | 627 | queue->rx_last_skb_slots = max_slots_needed; |
ca2f09f2 | 628 | break; |
9ab9831b | 629 | } else |
e9ce7cb6 | 630 | queue->rx_last_skb_slots = 0; |
f942dc25 | 631 | |
e9ce7cb6 WL |
632 | old_req_cons = queue->rx.req_cons; |
633 | XENVIF_RX_CB(skb)->meta_slots_used = xenvif_gop_skb(skb, &npo, queue); | |
634 | ring_slots_used = queue->rx.req_cons - old_req_cons; | |
1425c7a4 PD |
635 | |
636 | BUG_ON(ring_slots_used > max_slots_needed); | |
f942dc25 IC |
637 | |
638 | __skb_queue_tail(&rxq, skb); | |
f942dc25 IC |
639 | } |
640 | ||
e9ce7cb6 | 641 | BUG_ON(npo.meta_prod > ARRAY_SIZE(queue->meta)); |
f942dc25 IC |
642 | |
643 | if (!npo.copy_prod) | |
ca2f09f2 | 644 | goto done; |
f942dc25 | 645 | |
ac3d5ac2 | 646 | BUG_ON(npo.copy_prod > MAX_GRANT_COPY_OPS); |
e9ce7cb6 | 647 | gnttab_batch_copy(queue->grant_copy_op, npo.copy_prod); |
f942dc25 IC |
648 | |
649 | while ((skb = __skb_dequeue(&rxq)) != NULL) { | |
f942dc25 | 650 | |
e9ce7cb6 WL |
651 | if ((1 << queue->meta[npo.meta_cons].gso_type) & |
652 | queue->vif->gso_prefix_mask) { | |
653 | resp = RING_GET_RESPONSE(&queue->rx, | |
654 | queue->rx.rsp_prod_pvt++); | |
f942dc25 IC |
655 | |
656 | resp->flags = XEN_NETRXF_gso_prefix | XEN_NETRXF_more_data; | |
657 | ||
e9ce7cb6 WL |
658 | resp->offset = queue->meta[npo.meta_cons].gso_size; |
659 | resp->id = queue->meta[npo.meta_cons].id; | |
8f13dd96 | 660 | resp->status = XENVIF_RX_CB(skb)->meta_slots_used; |
f942dc25 IC |
661 | |
662 | npo.meta_cons++; | |
8f13dd96 | 663 | XENVIF_RX_CB(skb)->meta_slots_used--; |
f942dc25 IC |
664 | } |
665 | ||
666 | ||
e9ce7cb6 WL |
667 | queue->stats.tx_bytes += skb->len; |
668 | queue->stats.tx_packets++; | |
f942dc25 | 669 | |
e9ce7cb6 | 670 | status = xenvif_check_gop(queue->vif, |
8f13dd96 ZK |
671 | XENVIF_RX_CB(skb)->meta_slots_used, |
672 | &npo); | |
f942dc25 | 673 | |
8f13dd96 | 674 | if (XENVIF_RX_CB(skb)->meta_slots_used == 1) |
f942dc25 IC |
675 | flags = 0; |
676 | else | |
677 | flags = XEN_NETRXF_more_data; | |
678 | ||
679 | if (skb->ip_summed == CHECKSUM_PARTIAL) /* local packet? */ | |
680 | flags |= XEN_NETRXF_csum_blank | XEN_NETRXF_data_validated; | |
681 | else if (skb->ip_summed == CHECKSUM_UNNECESSARY) | |
682 | /* remote but checksummed. */ | |
683 | flags |= XEN_NETRXF_data_validated; | |
684 | ||
685 | offset = 0; | |
e9ce7cb6 | 686 | resp = make_rx_response(queue, queue->meta[npo.meta_cons].id, |
f942dc25 | 687 | status, offset, |
e9ce7cb6 | 688 | queue->meta[npo.meta_cons].size, |
f942dc25 IC |
689 | flags); |
690 | ||
e9ce7cb6 WL |
691 | if ((1 << queue->meta[npo.meta_cons].gso_type) & |
692 | queue->vif->gso_mask) { | |
f942dc25 IC |
693 | struct xen_netif_extra_info *gso = |
694 | (struct xen_netif_extra_info *) | |
e9ce7cb6 WL |
695 | RING_GET_RESPONSE(&queue->rx, |
696 | queue->rx.rsp_prod_pvt++); | |
f942dc25 IC |
697 | |
698 | resp->flags |= XEN_NETRXF_extra_info; | |
699 | ||
e9ce7cb6 WL |
700 | gso->u.gso.type = queue->meta[npo.meta_cons].gso_type; |
701 | gso->u.gso.size = queue->meta[npo.meta_cons].gso_size; | |
f942dc25 IC |
702 | gso->u.gso.pad = 0; |
703 | gso->u.gso.features = 0; | |
704 | ||
705 | gso->type = XEN_NETIF_EXTRA_TYPE_GSO; | |
706 | gso->flags = 0; | |
707 | } | |
708 | ||
e9ce7cb6 WL |
709 | xenvif_add_frag_responses(queue, status, |
710 | queue->meta + npo.meta_cons + 1, | |
8f13dd96 | 711 | XENVIF_RX_CB(skb)->meta_slots_used); |
f942dc25 | 712 | |
e9ce7cb6 | 713 | RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&queue->rx, ret); |
f942dc25 | 714 | |
11b57f90 | 715 | need_to_notify |= !!ret; |
b3f980bd | 716 | |
8f13dd96 | 717 | npo.meta_cons += XENVIF_RX_CB(skb)->meta_slots_used; |
f942dc25 IC |
718 | dev_kfree_skb(skb); |
719 | } | |
720 | ||
ca2f09f2 | 721 | done: |
b3f980bd | 722 | if (need_to_notify) |
e9ce7cb6 | 723 | notify_remote_via_irq(queue->rx_irq); |
f942dc25 IC |
724 | } |
725 | ||
e9ce7cb6 | 726 | void xenvif_napi_schedule_or_enable_events(struct xenvif_queue *queue) |
f942dc25 IC |
727 | { |
728 | int more_to_do; | |
729 | ||
e9ce7cb6 | 730 | RING_FINAL_CHECK_FOR_REQUESTS(&queue->tx, more_to_do); |
f942dc25 IC |
731 | |
732 | if (more_to_do) | |
e9ce7cb6 | 733 | napi_schedule(&queue->napi); |
f942dc25 IC |
734 | } |
735 | ||
e9ce7cb6 | 736 | static void tx_add_credit(struct xenvif_queue *queue) |
f942dc25 IC |
737 | { |
738 | unsigned long max_burst, max_credit; | |
739 | ||
740 | /* | |
741 | * Allow a burst big enough to transmit a jumbo packet of up to 128kB. | |
742 | * Otherwise the interface can seize up due to insufficient credit. | |
743 | */ | |
e9ce7cb6 | 744 | max_burst = RING_GET_REQUEST(&queue->tx, queue->tx.req_cons)->size; |
f942dc25 | 745 | max_burst = min(max_burst, 131072UL); |
e9ce7cb6 | 746 | max_burst = max(max_burst, queue->credit_bytes); |
f942dc25 IC |
747 | |
748 | /* Take care that adding a new chunk of credit doesn't wrap to zero. */ | |
e9ce7cb6 WL |
749 | max_credit = queue->remaining_credit + queue->credit_bytes; |
750 | if (max_credit < queue->remaining_credit) | |
f942dc25 IC |
751 | max_credit = ULONG_MAX; /* wrapped: clamp to ULONG_MAX */ |
752 | ||
e9ce7cb6 | 753 | queue->remaining_credit = min(max_credit, max_burst); |
f942dc25 IC |
754 | } |
755 | ||
756 | static void tx_credit_callback(unsigned long data) | |
757 | { | |
e9ce7cb6 WL |
758 | struct xenvif_queue *queue = (struct xenvif_queue *)data; |
759 | tx_add_credit(queue); | |
760 | xenvif_napi_schedule_or_enable_events(queue); | |
f942dc25 IC |
761 | } |
762 | ||
e9ce7cb6 | 763 | static void xenvif_tx_err(struct xenvif_queue *queue, |
7376419a | 764 | struct xen_netif_tx_request *txp, RING_IDX end) |
f942dc25 | 765 | { |
e9ce7cb6 | 766 | RING_IDX cons = queue->tx.req_cons; |
f53c3fe8 | 767 | unsigned long flags; |
f942dc25 IC |
768 | |
769 | do { | |
e9ce7cb6 WL |
770 | spin_lock_irqsave(&queue->response_lock, flags); |
771 | make_tx_response(queue, txp, XEN_NETIF_RSP_ERROR); | |
772 | spin_unlock_irqrestore(&queue->response_lock, flags); | |
b9149729 | 773 | if (cons == end) |
f942dc25 | 774 | break; |
e9ce7cb6 | 775 | txp = RING_GET_REQUEST(&queue->tx, cons++); |
f942dc25 | 776 | } while (1); |
e9ce7cb6 | 777 | queue->tx.req_cons = cons; |
f942dc25 IC |
778 | } |
779 | ||
7376419a | 780 | static void xenvif_fatal_tx_err(struct xenvif *vif) |
48856286 IC |
781 | { |
782 | netdev_err(vif->dev, "fatal error; disabling device\n"); | |
e9d8b2c2 | 783 | vif->disabled = true; |
e9ce7cb6 WL |
784 | /* Disable the vif from queue 0's kthread */ |
785 | if (vif->queues) | |
786 | xenvif_kick_thread(&vif->queues[0]); | |
48856286 IC |
787 | } |
788 | ||
e9ce7cb6 | 789 | static int xenvif_count_requests(struct xenvif_queue *queue, |
7376419a WL |
790 | struct xen_netif_tx_request *first, |
791 | struct xen_netif_tx_request *txp, | |
792 | int work_to_do) | |
f942dc25 | 793 | { |
e9ce7cb6 | 794 | RING_IDX cons = queue->tx.req_cons; |
2810e5b9 WL |
795 | int slots = 0; |
796 | int drop_err = 0; | |
59ccb4eb | 797 | int more_data; |
f942dc25 IC |
798 | |
799 | if (!(first->flags & XEN_NETTXF_more_data)) | |
800 | return 0; | |
801 | ||
802 | do { | |
59ccb4eb WL |
803 | struct xen_netif_tx_request dropped_tx = { 0 }; |
804 | ||
2810e5b9 | 805 | if (slots >= work_to_do) { |
e9ce7cb6 | 806 | netdev_err(queue->vif->dev, |
2810e5b9 WL |
807 | "Asked for %d slots but exceeds this limit\n", |
808 | work_to_do); | |
e9ce7cb6 | 809 | xenvif_fatal_tx_err(queue->vif); |
35876b5f | 810 | return -ENODATA; |
f942dc25 IC |
811 | } |
812 | ||
2810e5b9 WL |
813 | /* This guest is really using too many slots and |
814 | * considered malicious. | |
815 | */ | |
37641494 | 816 | if (unlikely(slots >= fatal_skb_slots)) { |
e9ce7cb6 | 817 | netdev_err(queue->vif->dev, |
2810e5b9 | 818 | "Malicious frontend using %d slots, threshold %u\n", |
37641494 | 819 | slots, fatal_skb_slots); |
e9ce7cb6 | 820 | xenvif_fatal_tx_err(queue->vif); |
35876b5f | 821 | return -E2BIG; |
f942dc25 IC |
822 | } |
823 | ||
2810e5b9 | 824 | /* Xen network protocol had implicit dependency on |
37641494 WL |
825 | * MAX_SKB_FRAGS. XEN_NETBK_LEGACY_SLOTS_MAX is set to |
826 | * the historical MAX_SKB_FRAGS value 18 to honor the | |
827 | * same behavior as before. Any packet using more than | |
828 | * 18 slots but less than fatal_skb_slots slots is | |
829 | * dropped | |
2810e5b9 | 830 | */ |
37641494 | 831 | if (!drop_err && slots >= XEN_NETBK_LEGACY_SLOTS_MAX) { |
2810e5b9 | 832 | if (net_ratelimit()) |
e9ce7cb6 | 833 | netdev_dbg(queue->vif->dev, |
2810e5b9 | 834 | "Too many slots (%d) exceeding limit (%d), dropping packet\n", |
37641494 | 835 | slots, XEN_NETBK_LEGACY_SLOTS_MAX); |
2810e5b9 WL |
836 | drop_err = -E2BIG; |
837 | } | |
838 | ||
59ccb4eb WL |
839 | if (drop_err) |
840 | txp = &dropped_tx; | |
841 | ||
e9ce7cb6 | 842 | memcpy(txp, RING_GET_REQUEST(&queue->tx, cons + slots), |
f942dc25 | 843 | sizeof(*txp)); |
03393fd5 WL |
844 | |
845 | /* If the guest submitted a frame >= 64 KiB then | |
846 | * first->size overflowed and following slots will | |
847 | * appear to be larger than the frame. | |
848 | * | |
849 | * This cannot be fatal error as there are buggy | |
850 | * frontends that do this. | |
851 | * | |
852 | * Consume all slots and drop the packet. | |
853 | */ | |
854 | if (!drop_err && txp->size > first->size) { | |
855 | if (net_ratelimit()) | |
e9ce7cb6 | 856 | netdev_dbg(queue->vif->dev, |
03393fd5 WL |
857 | "Invalid tx request, slot size %u > remaining size %u\n", |
858 | txp->size, first->size); | |
859 | drop_err = -EIO; | |
f942dc25 IC |
860 | } |
861 | ||
862 | first->size -= txp->size; | |
2810e5b9 | 863 | slots++; |
f942dc25 IC |
864 | |
865 | if (unlikely((txp->offset + txp->size) > PAGE_SIZE)) { | |
e9ce7cb6 | 866 | netdev_err(queue->vif->dev, "Cross page boundary, txp->offset: %x, size: %u\n", |
f942dc25 | 867 | txp->offset, txp->size); |
e9ce7cb6 | 868 | xenvif_fatal_tx_err(queue->vif); |
35876b5f | 869 | return -EINVAL; |
f942dc25 | 870 | } |
59ccb4eb WL |
871 | |
872 | more_data = txp->flags & XEN_NETTXF_more_data; | |
873 | ||
874 | if (!drop_err) | |
875 | txp++; | |
876 | ||
877 | } while (more_data); | |
2810e5b9 WL |
878 | |
879 | if (drop_err) { | |
e9ce7cb6 | 880 | xenvif_tx_err(queue, first, cons + slots); |
2810e5b9 WL |
881 | return drop_err; |
882 | } | |
883 | ||
884 | return slots; | |
f942dc25 IC |
885 | } |
886 | ||
8f13dd96 ZK |
887 | |
888 | struct xenvif_tx_cb { | |
889 | u16 pending_idx; | |
890 | }; | |
891 | ||
892 | #define XENVIF_TX_CB(skb) ((struct xenvif_tx_cb *)(skb)->cb) | |
893 | ||
e9ce7cb6 | 894 | static inline void xenvif_tx_create_map_op(struct xenvif_queue *queue, |
9074ce24 ZK |
895 | u16 pending_idx, |
896 | struct xen_netif_tx_request *txp, | |
897 | struct gnttab_map_grant_ref *mop) | |
f53c3fe8 | 898 | { |
e9ce7cb6 WL |
899 | queue->pages_to_map[mop-queue->tx_map_ops] = queue->mmap_pages[pending_idx]; |
900 | gnttab_set_map_op(mop, idx_to_kaddr(queue, pending_idx), | |
f53c3fe8 | 901 | GNTMAP_host_map | GNTMAP_readonly, |
e9ce7cb6 | 902 | txp->gref, queue->vif->domid); |
f53c3fe8 | 903 | |
e9ce7cb6 | 904 | memcpy(&queue->pending_tx_info[pending_idx].req, txp, |
f53c3fe8 ZK |
905 | sizeof(*txp)); |
906 | } | |
907 | ||
e3377f36 ZK |
908 | static inline struct sk_buff *xenvif_alloc_skb(unsigned int size) |
909 | { | |
910 | struct sk_buff *skb = | |
911 | alloc_skb(size + NET_SKB_PAD + NET_IP_ALIGN, | |
912 | GFP_ATOMIC | __GFP_NOWARN); | |
913 | if (unlikely(skb == NULL)) | |
914 | return NULL; | |
915 | ||
916 | /* Packets passed to netif_rx() must have some headroom. */ | |
917 | skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN); | |
918 | ||
919 | /* Initialize it here to avoid later surprises */ | |
920 | skb_shinfo(skb)->destructor_arg = NULL; | |
921 | ||
922 | return skb; | |
923 | } | |
924 | ||
e9ce7cb6 | 925 | static struct gnttab_map_grant_ref *xenvif_get_requests(struct xenvif_queue *queue, |
f53c3fe8 ZK |
926 | struct sk_buff *skb, |
927 | struct xen_netif_tx_request *txp, | |
928 | struct gnttab_map_grant_ref *gop) | |
f942dc25 IC |
929 | { |
930 | struct skb_shared_info *shinfo = skb_shinfo(skb); | |
931 | skb_frag_t *frags = shinfo->frags; | |
8f13dd96 | 932 | u16 pending_idx = XENVIF_TX_CB(skb)->pending_idx; |
62bad319 ZK |
933 | int start; |
934 | pending_ring_idx_t index; | |
e3377f36 | 935 | unsigned int nr_slots, frag_overflow = 0; |
2810e5b9 WL |
936 | |
937 | /* At this point shinfo->nr_frags is in fact the number of | |
37641494 | 938 | * slots, which can be as large as XEN_NETBK_LEGACY_SLOTS_MAX. |
2810e5b9 | 939 | */ |
e3377f36 ZK |
940 | if (shinfo->nr_frags > MAX_SKB_FRAGS) { |
941 | frag_overflow = shinfo->nr_frags - MAX_SKB_FRAGS; | |
942 | BUG_ON(frag_overflow > MAX_SKB_FRAGS); | |
943 | shinfo->nr_frags = MAX_SKB_FRAGS; | |
944 | } | |
2810e5b9 | 945 | nr_slots = shinfo->nr_frags; |
f942dc25 IC |
946 | |
947 | /* Skip first skb fragment if it is on same page as header fragment. */ | |
ea066ad1 | 948 | start = (frag_get_pending_idx(&shinfo->frags[0]) == pending_idx); |
f942dc25 | 949 | |
f53c3fe8 ZK |
950 | for (shinfo->nr_frags = start; shinfo->nr_frags < nr_slots; |
951 | shinfo->nr_frags++, txp++, gop++) { | |
e9ce7cb6 WL |
952 | index = pending_index(queue->pending_cons++); |
953 | pending_idx = queue->pending_ring[index]; | |
954 | xenvif_tx_create_map_op(queue, pending_idx, txp, gop); | |
f53c3fe8 | 955 | frag_set_pending_idx(&frags[shinfo->nr_frags], pending_idx); |
f942dc25 IC |
956 | } |
957 | ||
e3377f36 ZK |
958 | if (frag_overflow) { |
959 | struct sk_buff *nskb = xenvif_alloc_skb(0); | |
960 | if (unlikely(nskb == NULL)) { | |
961 | if (net_ratelimit()) | |
e9ce7cb6 | 962 | netdev_err(queue->vif->dev, |
e3377f36 ZK |
963 | "Can't allocate the frag_list skb.\n"); |
964 | return NULL; | |
965 | } | |
966 | ||
967 | shinfo = skb_shinfo(nskb); | |
968 | frags = shinfo->frags; | |
969 | ||
970 | for (shinfo->nr_frags = 0; shinfo->nr_frags < frag_overflow; | |
971 | shinfo->nr_frags++, txp++, gop++) { | |
e9ce7cb6 WL |
972 | index = pending_index(queue->pending_cons++); |
973 | pending_idx = queue->pending_ring[index]; | |
974 | xenvif_tx_create_map_op(queue, pending_idx, txp, gop); | |
e3377f36 ZK |
975 | frag_set_pending_idx(&frags[shinfo->nr_frags], |
976 | pending_idx); | |
977 | } | |
978 | ||
979 | skb_shinfo(skb)->frag_list = nskb; | |
980 | } | |
2810e5b9 | 981 | |
f942dc25 IC |
982 | return gop; |
983 | } | |
984 | ||
e9ce7cb6 | 985 | static inline void xenvif_grant_handle_set(struct xenvif_queue *queue, |
f53c3fe8 ZK |
986 | u16 pending_idx, |
987 | grant_handle_t handle) | |
988 | { | |
e9ce7cb6 | 989 | if (unlikely(queue->grant_tx_handle[pending_idx] != |
f53c3fe8 | 990 | NETBACK_INVALID_HANDLE)) { |
e9ce7cb6 | 991 | netdev_err(queue->vif->dev, |
f53c3fe8 ZK |
992 | "Trying to overwrite active handle! pending_idx: %x\n", |
993 | pending_idx); | |
994 | BUG(); | |
995 | } | |
e9ce7cb6 | 996 | queue->grant_tx_handle[pending_idx] = handle; |
f53c3fe8 ZK |
997 | } |
998 | ||
e9ce7cb6 | 999 | static inline void xenvif_grant_handle_reset(struct xenvif_queue *queue, |
f53c3fe8 ZK |
1000 | u16 pending_idx) |
1001 | { | |
e9ce7cb6 | 1002 | if (unlikely(queue->grant_tx_handle[pending_idx] == |
f53c3fe8 | 1003 | NETBACK_INVALID_HANDLE)) { |
e9ce7cb6 | 1004 | netdev_err(queue->vif->dev, |
f53c3fe8 ZK |
1005 | "Trying to unmap invalid handle! pending_idx: %x\n", |
1006 | pending_idx); | |
1007 | BUG(); | |
1008 | } | |
e9ce7cb6 | 1009 | queue->grant_tx_handle[pending_idx] = NETBACK_INVALID_HANDLE; |
f53c3fe8 ZK |
1010 | } |
1011 | ||
e9ce7cb6 | 1012 | static int xenvif_tx_check_gop(struct xenvif_queue *queue, |
7376419a | 1013 | struct sk_buff *skb, |
bdab8275 ZK |
1014 | struct gnttab_map_grant_ref **gopp_map, |
1015 | struct gnttab_copy **gopp_copy) | |
f942dc25 | 1016 | { |
9074ce24 | 1017 | struct gnttab_map_grant_ref *gop_map = *gopp_map; |
8f13dd96 | 1018 | u16 pending_idx = XENVIF_TX_CB(skb)->pending_idx; |
f942dc25 IC |
1019 | struct skb_shared_info *shinfo = skb_shinfo(skb); |
1020 | int nr_frags = shinfo->nr_frags; | |
bdab8275 | 1021 | int i, err; |
e3377f36 | 1022 | struct sk_buff *first_skb = NULL; |
f942dc25 IC |
1023 | |
1024 | /* Check status of header. */ | |
bdab8275 ZK |
1025 | err = (*gopp_copy)->status; |
1026 | (*gopp_copy)++; | |
1027 | if (unlikely(err)) { | |
1028 | if (net_ratelimit()) | |
e9ce7cb6 | 1029 | netdev_dbg(queue->vif->dev, |
00aefceb | 1030 | "Grant copy of header failed! status: %d pending_idx: %u ref: %u\n", |
bdab8275 ZK |
1031 | (*gopp_copy)->status, |
1032 | pending_idx, | |
1033 | (*gopp_copy)->source.u.ref); | |
e9ce7cb6 | 1034 | xenvif_idx_release(queue, pending_idx, XEN_NETIF_RSP_ERROR); |
bdab8275 | 1035 | } |
f942dc25 | 1036 | |
e3377f36 | 1037 | check_frags: |
bdab8275 | 1038 | for (i = 0; i < nr_frags; i++, gop_map++) { |
f942dc25 | 1039 | int j, newerr; |
f942dc25 | 1040 | |
ea066ad1 | 1041 | pending_idx = frag_get_pending_idx(&shinfo->frags[i]); |
f942dc25 IC |
1042 | |
1043 | /* Check error status: if okay then remember grant handle. */ | |
bdab8275 | 1044 | newerr = gop_map->status; |
2810e5b9 | 1045 | |
f942dc25 | 1046 | if (likely(!newerr)) { |
e9ce7cb6 | 1047 | xenvif_grant_handle_set(queue, |
9074ce24 ZK |
1048 | pending_idx, |
1049 | gop_map->handle); | |
f942dc25 IC |
1050 | /* Had a previous error? Invalidate this fragment. */ |
1051 | if (unlikely(err)) | |
e9ce7cb6 | 1052 | xenvif_idx_unmap(queue, pending_idx); |
f942dc25 IC |
1053 | continue; |
1054 | } | |
1055 | ||
1056 | /* Error on this fragment: respond to client with an error. */ | |
bdab8275 | 1057 | if (net_ratelimit()) |
e9ce7cb6 | 1058 | netdev_dbg(queue->vif->dev, |
00aefceb | 1059 | "Grant map of %d. frag failed! status: %d pending_idx: %u ref: %u\n", |
bdab8275 ZK |
1060 | i, |
1061 | gop_map->status, | |
1062 | pending_idx, | |
1063 | gop_map->ref); | |
e9ce7cb6 | 1064 | xenvif_idx_release(queue, pending_idx, XEN_NETIF_RSP_ERROR); |
f942dc25 IC |
1065 | |
1066 | /* Not the first error? Preceding frags already invalidated. */ | |
1067 | if (err) | |
1068 | continue; | |
bdab8275 ZK |
1069 | /* First error: invalidate preceding fragments. */ |
1070 | for (j = 0; j < i; j++) { | |
5ccb3ea7 | 1071 | pending_idx = frag_get_pending_idx(&shinfo->frags[j]); |
e9ce7cb6 | 1072 | xenvif_idx_unmap(queue, pending_idx); |
f942dc25 IC |
1073 | } |
1074 | ||
1075 | /* Remember the error: invalidate all subsequent fragments. */ | |
1076 | err = newerr; | |
1077 | } | |
1078 | ||
e3377f36 ZK |
1079 | if (skb_has_frag_list(skb)) { |
1080 | first_skb = skb; | |
1081 | skb = shinfo->frag_list; | |
1082 | shinfo = skb_shinfo(skb); | |
1083 | nr_frags = shinfo->nr_frags; | |
e3377f36 ZK |
1084 | |
1085 | goto check_frags; | |
1086 | } | |
1087 | ||
1088 | /* There was a mapping error in the frag_list skb. We have to unmap | |
1089 | * the first skb's frags | |
1090 | */ | |
1091 | if (first_skb && err) { | |
1092 | int j; | |
1093 | shinfo = skb_shinfo(first_skb); | |
bdab8275 | 1094 | for (j = 0; j < shinfo->nr_frags; j++) { |
e3377f36 | 1095 | pending_idx = frag_get_pending_idx(&shinfo->frags[j]); |
e9ce7cb6 | 1096 | xenvif_idx_unmap(queue, pending_idx); |
e3377f36 ZK |
1097 | } |
1098 | } | |
1099 | ||
bdab8275 | 1100 | *gopp_map = gop_map; |
f942dc25 IC |
1101 | return err; |
1102 | } | |
1103 | ||
e9ce7cb6 | 1104 | static void xenvif_fill_frags(struct xenvif_queue *queue, struct sk_buff *skb) |
f942dc25 IC |
1105 | { |
1106 | struct skb_shared_info *shinfo = skb_shinfo(skb); | |
1107 | int nr_frags = shinfo->nr_frags; | |
1108 | int i; | |
f53c3fe8 ZK |
1109 | u16 prev_pending_idx = INVALID_PENDING_IDX; |
1110 | ||
f942dc25 IC |
1111 | for (i = 0; i < nr_frags; i++) { |
1112 | skb_frag_t *frag = shinfo->frags + i; | |
1113 | struct xen_netif_tx_request *txp; | |
ea066ad1 IC |
1114 | struct page *page; |
1115 | u16 pending_idx; | |
f942dc25 | 1116 | |
ea066ad1 | 1117 | pending_idx = frag_get_pending_idx(frag); |
f942dc25 | 1118 | |
f53c3fe8 | 1119 | /* If this is not the first frag, chain it to the previous*/ |
bdab8275 | 1120 | if (prev_pending_idx == INVALID_PENDING_IDX) |
f53c3fe8 | 1121 | skb_shinfo(skb)->destructor_arg = |
e9ce7cb6 | 1122 | &callback_param(queue, pending_idx); |
bdab8275 | 1123 | else |
e9ce7cb6 WL |
1124 | callback_param(queue, prev_pending_idx).ctx = |
1125 | &callback_param(queue, pending_idx); | |
f53c3fe8 | 1126 | |
e9ce7cb6 | 1127 | callback_param(queue, pending_idx).ctx = NULL; |
f53c3fe8 ZK |
1128 | prev_pending_idx = pending_idx; |
1129 | ||
e9ce7cb6 WL |
1130 | txp = &queue->pending_tx_info[pending_idx].req; |
1131 | page = virt_to_page(idx_to_kaddr(queue, pending_idx)); | |
ea066ad1 | 1132 | __skb_fill_page_desc(skb, i, page, txp->offset, txp->size); |
f942dc25 IC |
1133 | skb->len += txp->size; |
1134 | skb->data_len += txp->size; | |
1135 | skb->truesize += txp->size; | |
1136 | ||
f53c3fe8 | 1137 | /* Take an extra reference to offset network stack's put_page */ |
e9ce7cb6 | 1138 | get_page(queue->mmap_pages[pending_idx]); |
f942dc25 | 1139 | } |
f53c3fe8 ZK |
1140 | /* FIXME: __skb_fill_page_desc set this to true because page->pfmemalloc |
1141 | * overlaps with "index", and "mapping" is not set. I think mapping | |
1142 | * should be set. If delivered to local stack, it would drop this | |
1143 | * skb in sk_filter unless the socket has the right to use it. | |
1144 | */ | |
1145 | skb->pfmemalloc = false; | |
f942dc25 IC |
1146 | } |
1147 | ||
e9ce7cb6 | 1148 | static int xenvif_get_extras(struct xenvif_queue *queue, |
f942dc25 IC |
1149 | struct xen_netif_extra_info *extras, |
1150 | int work_to_do) | |
1151 | { | |
1152 | struct xen_netif_extra_info extra; | |
e9ce7cb6 | 1153 | RING_IDX cons = queue->tx.req_cons; |
f942dc25 IC |
1154 | |
1155 | do { | |
1156 | if (unlikely(work_to_do-- <= 0)) { | |
e9ce7cb6 WL |
1157 | netdev_err(queue->vif->dev, "Missing extra info\n"); |
1158 | xenvif_fatal_tx_err(queue->vif); | |
f942dc25 IC |
1159 | return -EBADR; |
1160 | } | |
1161 | ||
e9ce7cb6 | 1162 | memcpy(&extra, RING_GET_REQUEST(&queue->tx, cons), |
f942dc25 IC |
1163 | sizeof(extra)); |
1164 | if (unlikely(!extra.type || | |
1165 | extra.type >= XEN_NETIF_EXTRA_TYPE_MAX)) { | |
e9ce7cb6 WL |
1166 | queue->tx.req_cons = ++cons; |
1167 | netdev_err(queue->vif->dev, | |
f942dc25 | 1168 | "Invalid extra type: %d\n", extra.type); |
e9ce7cb6 | 1169 | xenvif_fatal_tx_err(queue->vif); |
f942dc25 IC |
1170 | return -EINVAL; |
1171 | } | |
1172 | ||
1173 | memcpy(&extras[extra.type - 1], &extra, sizeof(extra)); | |
e9ce7cb6 | 1174 | queue->tx.req_cons = ++cons; |
f942dc25 IC |
1175 | } while (extra.flags & XEN_NETIF_EXTRA_FLAG_MORE); |
1176 | ||
1177 | return work_to_do; | |
1178 | } | |
1179 | ||
7376419a WL |
1180 | static int xenvif_set_skb_gso(struct xenvif *vif, |
1181 | struct sk_buff *skb, | |
1182 | struct xen_netif_extra_info *gso) | |
f942dc25 IC |
1183 | { |
1184 | if (!gso->u.gso.size) { | |
48856286 | 1185 | netdev_err(vif->dev, "GSO size must not be zero.\n"); |
7376419a | 1186 | xenvif_fatal_tx_err(vif); |
f942dc25 IC |
1187 | return -EINVAL; |
1188 | } | |
1189 | ||
a9468587 PD |
1190 | switch (gso->u.gso.type) { |
1191 | case XEN_NETIF_GSO_TYPE_TCPV4: | |
1192 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4; | |
1193 | break; | |
1194 | case XEN_NETIF_GSO_TYPE_TCPV6: | |
1195 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6; | |
1196 | break; | |
1197 | default: | |
48856286 | 1198 | netdev_err(vif->dev, "Bad GSO type %d.\n", gso->u.gso.type); |
7376419a | 1199 | xenvif_fatal_tx_err(vif); |
f942dc25 IC |
1200 | return -EINVAL; |
1201 | } | |
1202 | ||
1203 | skb_shinfo(skb)->gso_size = gso->u.gso.size; | |
b89587a7 | 1204 | /* gso_segs will be calculated later */ |
f942dc25 IC |
1205 | |
1206 | return 0; | |
1207 | } | |
1208 | ||
e9ce7cb6 | 1209 | static int checksum_setup(struct xenvif_queue *queue, struct sk_buff *skb) |
2eba61d5 | 1210 | { |
2721637c | 1211 | bool recalculate_partial_csum = false; |
2eba61d5 PD |
1212 | |
1213 | /* A GSO SKB must be CHECKSUM_PARTIAL. However some buggy | |
1214 | * peers can fail to set NETRXF_csum_blank when sending a GSO | |
1215 | * frame. In this case force the SKB to CHECKSUM_PARTIAL and | |
1216 | * recalculate the partial checksum. | |
1217 | */ | |
1218 | if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) { | |
e9ce7cb6 | 1219 | queue->stats.rx_gso_checksum_fixup++; |
2eba61d5 | 1220 | skb->ip_summed = CHECKSUM_PARTIAL; |
2721637c | 1221 | recalculate_partial_csum = true; |
2eba61d5 PD |
1222 | } |
1223 | ||
1224 | /* A non-CHECKSUM_PARTIAL SKB does not require setup. */ | |
1225 | if (skb->ip_summed != CHECKSUM_PARTIAL) | |
1226 | return 0; | |
1227 | ||
2721637c | 1228 | return skb_checksum_setup(skb, recalculate_partial_csum); |
2eba61d5 PD |
1229 | } |
1230 | ||
e9ce7cb6 | 1231 | static bool tx_credit_exceeded(struct xenvif_queue *queue, unsigned size) |
f942dc25 | 1232 | { |
059dfa6a | 1233 | u64 now = get_jiffies_64(); |
e9ce7cb6 WL |
1234 | u64 next_credit = queue->credit_window_start + |
1235 | msecs_to_jiffies(queue->credit_usec / 1000); | |
f942dc25 IC |
1236 | |
1237 | /* Timer could already be pending in rare cases. */ | |
e9ce7cb6 | 1238 | if (timer_pending(&queue->credit_timeout)) |
f942dc25 IC |
1239 | return true; |
1240 | ||
1241 | /* Passed the point where we can replenish credit? */ | |
059dfa6a | 1242 | if (time_after_eq64(now, next_credit)) { |
e9ce7cb6 WL |
1243 | queue->credit_window_start = now; |
1244 | tx_add_credit(queue); | |
f942dc25 IC |
1245 | } |
1246 | ||
1247 | /* Still too big to send right now? Set a callback. */ | |
e9ce7cb6 WL |
1248 | if (size > queue->remaining_credit) { |
1249 | queue->credit_timeout.data = | |
1250 | (unsigned long)queue; | |
1251 | queue->credit_timeout.function = | |
f942dc25 | 1252 | tx_credit_callback; |
e9ce7cb6 | 1253 | mod_timer(&queue->credit_timeout, |
f942dc25 | 1254 | next_credit); |
e9ce7cb6 | 1255 | queue->credit_window_start = next_credit; |
f942dc25 IC |
1256 | |
1257 | return true; | |
1258 | } | |
1259 | ||
1260 | return false; | |
1261 | } | |
1262 | ||
e9ce7cb6 | 1263 | static void xenvif_tx_build_gops(struct xenvif_queue *queue, |
bdab8275 ZK |
1264 | int budget, |
1265 | unsigned *copy_ops, | |
1266 | unsigned *map_ops) | |
f942dc25 | 1267 | { |
e9ce7cb6 | 1268 | struct gnttab_map_grant_ref *gop = queue->tx_map_ops, *request_gop; |
f942dc25 IC |
1269 | struct sk_buff *skb; |
1270 | int ret; | |
1271 | ||
e9ce7cb6 | 1272 | while (skb_queue_len(&queue->tx_queue) < budget) { |
f942dc25 | 1273 | struct xen_netif_tx_request txreq; |
37641494 | 1274 | struct xen_netif_tx_request txfrags[XEN_NETBK_LEGACY_SLOTS_MAX]; |
f942dc25 IC |
1275 | struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX-1]; |
1276 | u16 pending_idx; | |
1277 | RING_IDX idx; | |
1278 | int work_to_do; | |
1279 | unsigned int data_len; | |
1280 | pending_ring_idx_t index; | |
1281 | ||
e9ce7cb6 | 1282 | if (queue->tx.sring->req_prod - queue->tx.req_cons > |
48856286 | 1283 | XEN_NETIF_TX_RING_SIZE) { |
e9ce7cb6 | 1284 | netdev_err(queue->vif->dev, |
48856286 IC |
1285 | "Impossible number of requests. " |
1286 | "req_prod %d, req_cons %d, size %ld\n", | |
e9ce7cb6 | 1287 | queue->tx.sring->req_prod, queue->tx.req_cons, |
48856286 | 1288 | XEN_NETIF_TX_RING_SIZE); |
e9ce7cb6 | 1289 | xenvif_fatal_tx_err(queue->vif); |
e9d8b2c2 | 1290 | break; |
48856286 IC |
1291 | } |
1292 | ||
e9ce7cb6 | 1293 | work_to_do = RING_HAS_UNCONSUMED_REQUESTS(&queue->tx); |
b3f980bd WL |
1294 | if (!work_to_do) |
1295 | break; | |
f942dc25 | 1296 | |
e9ce7cb6 | 1297 | idx = queue->tx.req_cons; |
f942dc25 | 1298 | rmb(); /* Ensure that we see the request before we copy it. */ |
e9ce7cb6 | 1299 | memcpy(&txreq, RING_GET_REQUEST(&queue->tx, idx), sizeof(txreq)); |
f942dc25 IC |
1300 | |
1301 | /* Credit-based scheduling. */ | |
e9ce7cb6 WL |
1302 | if (txreq.size > queue->remaining_credit && |
1303 | tx_credit_exceeded(queue, txreq.size)) | |
b3f980bd | 1304 | break; |
f942dc25 | 1305 | |
e9ce7cb6 | 1306 | queue->remaining_credit -= txreq.size; |
f942dc25 IC |
1307 | |
1308 | work_to_do--; | |
e9ce7cb6 | 1309 | queue->tx.req_cons = ++idx; |
f942dc25 IC |
1310 | |
1311 | memset(extras, 0, sizeof(extras)); | |
1312 | if (txreq.flags & XEN_NETTXF_extra_info) { | |
e9ce7cb6 | 1313 | work_to_do = xenvif_get_extras(queue, extras, |
7376419a | 1314 | work_to_do); |
e9ce7cb6 | 1315 | idx = queue->tx.req_cons; |
48856286 | 1316 | if (unlikely(work_to_do < 0)) |
b3f980bd | 1317 | break; |
f942dc25 IC |
1318 | } |
1319 | ||
e9ce7cb6 | 1320 | ret = xenvif_count_requests(queue, &txreq, txfrags, work_to_do); |
48856286 | 1321 | if (unlikely(ret < 0)) |
b3f980bd | 1322 | break; |
48856286 | 1323 | |
f942dc25 IC |
1324 | idx += ret; |
1325 | ||
1326 | if (unlikely(txreq.size < ETH_HLEN)) { | |
e9ce7cb6 | 1327 | netdev_dbg(queue->vif->dev, |
f942dc25 | 1328 | "Bad packet size: %d\n", txreq.size); |
e9ce7cb6 | 1329 | xenvif_tx_err(queue, &txreq, idx); |
b3f980bd | 1330 | break; |
f942dc25 IC |
1331 | } |
1332 | ||
1333 | /* No crossing a page as the payload mustn't fragment. */ | |
1334 | if (unlikely((txreq.offset + txreq.size) > PAGE_SIZE)) { | |
e9ce7cb6 | 1335 | netdev_err(queue->vif->dev, |
f942dc25 IC |
1336 | "txreq.offset: %x, size: %u, end: %lu\n", |
1337 | txreq.offset, txreq.size, | |
1338 | (txreq.offset&~PAGE_MASK) + txreq.size); | |
e9ce7cb6 | 1339 | xenvif_fatal_tx_err(queue->vif); |
b3f980bd | 1340 | break; |
f942dc25 IC |
1341 | } |
1342 | ||
e9ce7cb6 WL |
1343 | index = pending_index(queue->pending_cons); |
1344 | pending_idx = queue->pending_ring[index]; | |
f942dc25 IC |
1345 | |
1346 | data_len = (txreq.size > PKT_PROT_LEN && | |
37641494 | 1347 | ret < XEN_NETBK_LEGACY_SLOTS_MAX) ? |
f942dc25 IC |
1348 | PKT_PROT_LEN : txreq.size; |
1349 | ||
e3377f36 | 1350 | skb = xenvif_alloc_skb(data_len); |
f942dc25 | 1351 | if (unlikely(skb == NULL)) { |
e9ce7cb6 | 1352 | netdev_dbg(queue->vif->dev, |
f942dc25 | 1353 | "Can't allocate a skb in start_xmit.\n"); |
e9ce7cb6 | 1354 | xenvif_tx_err(queue, &txreq, idx); |
f942dc25 IC |
1355 | break; |
1356 | } | |
1357 | ||
f942dc25 IC |
1358 | if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) { |
1359 | struct xen_netif_extra_info *gso; | |
1360 | gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1]; | |
1361 | ||
e9ce7cb6 | 1362 | if (xenvif_set_skb_gso(queue->vif, skb, gso)) { |
7376419a | 1363 | /* Failure in xenvif_set_skb_gso is fatal. */ |
f942dc25 | 1364 | kfree_skb(skb); |
b3f980bd | 1365 | break; |
f942dc25 IC |
1366 | } |
1367 | } | |
1368 | ||
8f13dd96 | 1369 | XENVIF_TX_CB(skb)->pending_idx = pending_idx; |
f942dc25 IC |
1370 | |
1371 | __skb_put(skb, data_len); | |
e9ce7cb6 WL |
1372 | queue->tx_copy_ops[*copy_ops].source.u.ref = txreq.gref; |
1373 | queue->tx_copy_ops[*copy_ops].source.domid = queue->vif->domid; | |
1374 | queue->tx_copy_ops[*copy_ops].source.offset = txreq.offset; | |
bdab8275 | 1375 | |
e9ce7cb6 | 1376 | queue->tx_copy_ops[*copy_ops].dest.u.gmfn = |
bdab8275 | 1377 | virt_to_mfn(skb->data); |
e9ce7cb6 WL |
1378 | queue->tx_copy_ops[*copy_ops].dest.domid = DOMID_SELF; |
1379 | queue->tx_copy_ops[*copy_ops].dest.offset = | |
bdab8275 ZK |
1380 | offset_in_page(skb->data); |
1381 | ||
e9ce7cb6 WL |
1382 | queue->tx_copy_ops[*copy_ops].len = data_len; |
1383 | queue->tx_copy_ops[*copy_ops].flags = GNTCOPY_source_gref; | |
bdab8275 ZK |
1384 | |
1385 | (*copy_ops)++; | |
f942dc25 IC |
1386 | |
1387 | skb_shinfo(skb)->nr_frags = ret; | |
1388 | if (data_len < txreq.size) { | |
1389 | skb_shinfo(skb)->nr_frags++; | |
ea066ad1 IC |
1390 | frag_set_pending_idx(&skb_shinfo(skb)->frags[0], |
1391 | pending_idx); | |
e9ce7cb6 | 1392 | xenvif_tx_create_map_op(queue, pending_idx, &txreq, gop); |
bdab8275 | 1393 | gop++; |
f942dc25 | 1394 | } else { |
ea066ad1 IC |
1395 | frag_set_pending_idx(&skb_shinfo(skb)->frags[0], |
1396 | INVALID_PENDING_IDX); | |
e9ce7cb6 | 1397 | memcpy(&queue->pending_tx_info[pending_idx].req, &txreq, |
bdab8275 | 1398 | sizeof(txreq)); |
f942dc25 IC |
1399 | } |
1400 | ||
e9ce7cb6 | 1401 | queue->pending_cons++; |
f942dc25 | 1402 | |
e9ce7cb6 | 1403 | request_gop = xenvif_get_requests(queue, skb, txfrags, gop); |
f942dc25 IC |
1404 | if (request_gop == NULL) { |
1405 | kfree_skb(skb); | |
e9ce7cb6 | 1406 | xenvif_tx_err(queue, &txreq, idx); |
b3f980bd | 1407 | break; |
f942dc25 IC |
1408 | } |
1409 | gop = request_gop; | |
1410 | ||
e9ce7cb6 | 1411 | __skb_queue_tail(&queue->tx_queue, skb); |
1e0b6eac | 1412 | |
e9ce7cb6 | 1413 | queue->tx.req_cons = idx; |
f942dc25 | 1414 | |
e9ce7cb6 WL |
1415 | if (((gop-queue->tx_map_ops) >= ARRAY_SIZE(queue->tx_map_ops)) || |
1416 | (*copy_ops >= ARRAY_SIZE(queue->tx_copy_ops))) | |
f942dc25 IC |
1417 | break; |
1418 | } | |
1419 | ||
e9ce7cb6 | 1420 | (*map_ops) = gop - queue->tx_map_ops; |
bdab8275 | 1421 | return; |
f942dc25 IC |
1422 | } |
1423 | ||
e3377f36 ZK |
1424 | /* Consolidate skb with a frag_list into a brand new one with local pages on |
1425 | * frags. Returns 0 or -ENOMEM if can't allocate new pages. | |
1426 | */ | |
e9ce7cb6 | 1427 | static int xenvif_handle_frag_list(struct xenvif_queue *queue, struct sk_buff *skb) |
e3377f36 ZK |
1428 | { |
1429 | unsigned int offset = skb_headlen(skb); | |
1430 | skb_frag_t frags[MAX_SKB_FRAGS]; | |
1431 | int i; | |
1432 | struct ubuf_info *uarg; | |
1433 | struct sk_buff *nskb = skb_shinfo(skb)->frag_list; | |
1434 | ||
e9ce7cb6 WL |
1435 | queue->stats.tx_zerocopy_sent += 2; |
1436 | queue->stats.tx_frag_overflow++; | |
e3377f36 | 1437 | |
e9ce7cb6 | 1438 | xenvif_fill_frags(queue, nskb); |
e3377f36 ZK |
1439 | /* Subtract frags size, we will correct it later */ |
1440 | skb->truesize -= skb->data_len; | |
1441 | skb->len += nskb->len; | |
1442 | skb->data_len += nskb->len; | |
1443 | ||
1444 | /* create a brand new frags array and coalesce there */ | |
1445 | for (i = 0; offset < skb->len; i++) { | |
1446 | struct page *page; | |
1447 | unsigned int len; | |
1448 | ||
1449 | BUG_ON(i >= MAX_SKB_FRAGS); | |
1450 | page = alloc_page(GFP_ATOMIC|__GFP_COLD); | |
1451 | if (!page) { | |
1452 | int j; | |
1453 | skb->truesize += skb->data_len; | |
1454 | for (j = 0; j < i; j++) | |
1455 | put_page(frags[j].page.p); | |
1456 | return -ENOMEM; | |
1457 | } | |
1458 | ||
1459 | if (offset + PAGE_SIZE < skb->len) | |
1460 | len = PAGE_SIZE; | |
1461 | else | |
1462 | len = skb->len - offset; | |
1463 | if (skb_copy_bits(skb, offset, page_address(page), len)) | |
1464 | BUG(); | |
1465 | ||
1466 | offset += len; | |
1467 | frags[i].page.p = page; | |
1468 | frags[i].page_offset = 0; | |
1469 | skb_frag_size_set(&frags[i], len); | |
1470 | } | |
1471 | /* swap out with old one */ | |
1472 | memcpy(skb_shinfo(skb)->frags, | |
1473 | frags, | |
1474 | i * sizeof(skb_frag_t)); | |
1475 | skb_shinfo(skb)->nr_frags = i; | |
1476 | skb->truesize += i * PAGE_SIZE; | |
1477 | ||
1478 | /* remove traces of mapped pages and frag_list */ | |
1479 | skb_frag_list_init(skb); | |
1480 | uarg = skb_shinfo(skb)->destructor_arg; | |
1481 | uarg->callback(uarg, true); | |
1482 | skb_shinfo(skb)->destructor_arg = NULL; | |
1483 | ||
1484 | skb_shinfo(nskb)->tx_flags |= SKBTX_DEV_ZEROCOPY; | |
1485 | kfree_skb(nskb); | |
1486 | ||
1487 | return 0; | |
1488 | } | |
b3f980bd | 1489 | |
e9ce7cb6 | 1490 | static int xenvif_tx_submit(struct xenvif_queue *queue) |
f942dc25 | 1491 | { |
e9ce7cb6 WL |
1492 | struct gnttab_map_grant_ref *gop_map = queue->tx_map_ops; |
1493 | struct gnttab_copy *gop_copy = queue->tx_copy_ops; | |
f942dc25 | 1494 | struct sk_buff *skb; |
b3f980bd | 1495 | int work_done = 0; |
f942dc25 | 1496 | |
e9ce7cb6 | 1497 | while ((skb = __skb_dequeue(&queue->tx_queue)) != NULL) { |
f942dc25 | 1498 | struct xen_netif_tx_request *txp; |
f942dc25 IC |
1499 | u16 pending_idx; |
1500 | unsigned data_len; | |
1501 | ||
8f13dd96 | 1502 | pending_idx = XENVIF_TX_CB(skb)->pending_idx; |
e9ce7cb6 | 1503 | txp = &queue->pending_tx_info[pending_idx].req; |
f942dc25 IC |
1504 | |
1505 | /* Check the remap error code. */ | |
e9ce7cb6 | 1506 | if (unlikely(xenvif_tx_check_gop(queue, skb, &gop_map, &gop_copy))) { |
f942dc25 IC |
1507 | skb_shinfo(skb)->nr_frags = 0; |
1508 | kfree_skb(skb); | |
1509 | continue; | |
1510 | } | |
1511 | ||
1512 | data_len = skb->len; | |
e9ce7cb6 | 1513 | callback_param(queue, pending_idx).ctx = NULL; |
f942dc25 IC |
1514 | if (data_len < txp->size) { |
1515 | /* Append the packet payload as a fragment. */ | |
1516 | txp->offset += data_len; | |
1517 | txp->size -= data_len; | |
1518 | } else { | |
1519 | /* Schedule a response immediately. */ | |
e9ce7cb6 | 1520 | xenvif_idx_release(queue, pending_idx, |
bdab8275 | 1521 | XEN_NETIF_RSP_OKAY); |
f942dc25 IC |
1522 | } |
1523 | ||
1524 | if (txp->flags & XEN_NETTXF_csum_blank) | |
1525 | skb->ip_summed = CHECKSUM_PARTIAL; | |
1526 | else if (txp->flags & XEN_NETTXF_data_validated) | |
1527 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
1528 | ||
e9ce7cb6 | 1529 | xenvif_fill_frags(queue, skb); |
f942dc25 | 1530 | |
e3377f36 | 1531 | if (unlikely(skb_has_frag_list(skb))) { |
e9ce7cb6 | 1532 | if (xenvif_handle_frag_list(queue, skb)) { |
e3377f36 | 1533 | if (net_ratelimit()) |
e9ce7cb6 | 1534 | netdev_err(queue->vif->dev, |
e3377f36 ZK |
1535 | "Not enough memory to consolidate frag_list!\n"); |
1536 | skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY; | |
1537 | kfree_skb(skb); | |
1538 | continue; | |
1539 | } | |
1540 | } | |
1541 | ||
2eba61d5 | 1542 | if (skb_is_nonlinear(skb) && skb_headlen(skb) < PKT_PROT_LEN) { |
f942dc25 IC |
1543 | int target = min_t(int, skb->len, PKT_PROT_LEN); |
1544 | __pskb_pull_tail(skb, target - skb_headlen(skb)); | |
1545 | } | |
1546 | ||
e9ce7cb6 | 1547 | skb->dev = queue->vif->dev; |
f942dc25 | 1548 | skb->protocol = eth_type_trans(skb, skb->dev); |
f9ca8f74 | 1549 | skb_reset_network_header(skb); |
f942dc25 | 1550 | |
e9ce7cb6 WL |
1551 | if (checksum_setup(queue, skb)) { |
1552 | netdev_dbg(queue->vif->dev, | |
f942dc25 | 1553 | "Can't setup checksum in net_tx_action\n"); |
f53c3fe8 ZK |
1554 | /* We have to set this flag to trigger the callback */ |
1555 | if (skb_shinfo(skb)->destructor_arg) | |
1556 | skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY; | |
f942dc25 IC |
1557 | kfree_skb(skb); |
1558 | continue; | |
1559 | } | |
1560 | ||
40893fd0 | 1561 | skb_probe_transport_header(skb, 0); |
f9ca8f74 | 1562 | |
b89587a7 PD |
1563 | /* If the packet is GSO then we will have just set up the |
1564 | * transport header offset in checksum_setup so it's now | |
1565 | * straightforward to calculate gso_segs. | |
1566 | */ | |
1567 | if (skb_is_gso(skb)) { | |
1568 | int mss = skb_shinfo(skb)->gso_size; | |
1569 | int hdrlen = skb_transport_header(skb) - | |
1570 | skb_mac_header(skb) + | |
1571 | tcp_hdrlen(skb); | |
1572 | ||
1573 | skb_shinfo(skb)->gso_segs = | |
1574 | DIV_ROUND_UP(skb->len - hdrlen, mss); | |
1575 | } | |
1576 | ||
e9ce7cb6 WL |
1577 | queue->stats.rx_bytes += skb->len; |
1578 | queue->stats.rx_packets++; | |
f942dc25 | 1579 | |
b3f980bd WL |
1580 | work_done++; |
1581 | ||
f53c3fe8 ZK |
1582 | /* Set this flag right before netif_receive_skb, otherwise |
1583 | * someone might think this packet already left netback, and | |
1584 | * do a skb_copy_ubufs while we are still in control of the | |
1585 | * skb. E.g. the __pskb_pull_tail earlier can do such thing. | |
1586 | */ | |
1bb332af | 1587 | if (skb_shinfo(skb)->destructor_arg) { |
f53c3fe8 | 1588 | skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY; |
e9ce7cb6 | 1589 | queue->stats.tx_zerocopy_sent++; |
1bb332af | 1590 | } |
f53c3fe8 | 1591 | |
b3f980bd | 1592 | netif_receive_skb(skb); |
f942dc25 | 1593 | } |
b3f980bd WL |
1594 | |
1595 | return work_done; | |
f942dc25 IC |
1596 | } |
1597 | ||
3e2234b3 ZK |
1598 | void xenvif_zerocopy_callback(struct ubuf_info *ubuf, bool zerocopy_success) |
1599 | { | |
f53c3fe8 ZK |
1600 | unsigned long flags; |
1601 | pending_ring_idx_t index; | |
e9ce7cb6 | 1602 | struct xenvif_queue *queue = ubuf_to_queue(ubuf); |
f53c3fe8 ZK |
1603 | |
1604 | /* This is the only place where we grab this lock, to protect callbacks | |
1605 | * from each other. | |
1606 | */ | |
e9ce7cb6 | 1607 | spin_lock_irqsave(&queue->callback_lock, flags); |
f53c3fe8 ZK |
1608 | do { |
1609 | u16 pending_idx = ubuf->desc; | |
1610 | ubuf = (struct ubuf_info *) ubuf->ctx; | |
e9ce7cb6 | 1611 | BUG_ON(queue->dealloc_prod - queue->dealloc_cons >= |
f53c3fe8 | 1612 | MAX_PENDING_REQS); |
e9ce7cb6 WL |
1613 | index = pending_index(queue->dealloc_prod); |
1614 | queue->dealloc_ring[index] = pending_idx; | |
f53c3fe8 ZK |
1615 | /* Sync with xenvif_tx_dealloc_action: |
1616 | * insert idx then incr producer. | |
1617 | */ | |
1618 | smp_wmb(); | |
e9ce7cb6 | 1619 | queue->dealloc_prod++; |
f53c3fe8 | 1620 | } while (ubuf); |
e9ce7cb6 WL |
1621 | wake_up(&queue->dealloc_wq); |
1622 | spin_unlock_irqrestore(&queue->callback_lock, flags); | |
f53c3fe8 | 1623 | |
1bb332af | 1624 | if (likely(zerocopy_success)) |
e9ce7cb6 | 1625 | queue->stats.tx_zerocopy_success++; |
1bb332af | 1626 | else |
e9ce7cb6 | 1627 | queue->stats.tx_zerocopy_fail++; |
f53c3fe8 ZK |
1628 | } |
1629 | ||
e9ce7cb6 | 1630 | static inline void xenvif_tx_dealloc_action(struct xenvif_queue *queue) |
f53c3fe8 ZK |
1631 | { |
1632 | struct gnttab_unmap_grant_ref *gop; | |
1633 | pending_ring_idx_t dc, dp; | |
1634 | u16 pending_idx, pending_idx_release[MAX_PENDING_REQS]; | |
1635 | unsigned int i = 0; | |
1636 | ||
e9ce7cb6 WL |
1637 | dc = queue->dealloc_cons; |
1638 | gop = queue->tx_unmap_ops; | |
f53c3fe8 ZK |
1639 | |
1640 | /* Free up any grants we have finished using */ | |
1641 | do { | |
e9ce7cb6 | 1642 | dp = queue->dealloc_prod; |
f53c3fe8 ZK |
1643 | |
1644 | /* Ensure we see all indices enqueued by all | |
1645 | * xenvif_zerocopy_callback(). | |
1646 | */ | |
1647 | smp_rmb(); | |
1648 | ||
1649 | while (dc != dp) { | |
e9ce7cb6 | 1650 | BUG_ON(gop - queue->tx_unmap_ops > MAX_PENDING_REQS); |
f53c3fe8 | 1651 | pending_idx = |
e9ce7cb6 | 1652 | queue->dealloc_ring[pending_index(dc++)]; |
f53c3fe8 | 1653 | |
e9ce7cb6 | 1654 | pending_idx_release[gop-queue->tx_unmap_ops] = |
f53c3fe8 | 1655 | pending_idx; |
e9ce7cb6 WL |
1656 | queue->pages_to_unmap[gop-queue->tx_unmap_ops] = |
1657 | queue->mmap_pages[pending_idx]; | |
f53c3fe8 | 1658 | gnttab_set_unmap_op(gop, |
e9ce7cb6 | 1659 | idx_to_kaddr(queue, pending_idx), |
f53c3fe8 | 1660 | GNTMAP_host_map, |
e9ce7cb6 WL |
1661 | queue->grant_tx_handle[pending_idx]); |
1662 | xenvif_grant_handle_reset(queue, pending_idx); | |
f53c3fe8 ZK |
1663 | ++gop; |
1664 | } | |
1665 | ||
e9ce7cb6 | 1666 | } while (dp != queue->dealloc_prod); |
f53c3fe8 | 1667 | |
e9ce7cb6 | 1668 | queue->dealloc_cons = dc; |
f53c3fe8 | 1669 | |
e9ce7cb6 | 1670 | if (gop - queue->tx_unmap_ops > 0) { |
f53c3fe8 | 1671 | int ret; |
e9ce7cb6 | 1672 | ret = gnttab_unmap_refs(queue->tx_unmap_ops, |
f53c3fe8 | 1673 | NULL, |
e9ce7cb6 WL |
1674 | queue->pages_to_unmap, |
1675 | gop - queue->tx_unmap_ops); | |
f53c3fe8 | 1676 | if (ret) { |
e9ce7cb6 WL |
1677 | netdev_err(queue->vif->dev, "Unmap fail: nr_ops %tx ret %d\n", |
1678 | gop - queue->tx_unmap_ops, ret); | |
1679 | for (i = 0; i < gop - queue->tx_unmap_ops; ++i) { | |
f53c3fe8 | 1680 | if (gop[i].status != GNTST_okay) |
e9ce7cb6 | 1681 | netdev_err(queue->vif->dev, |
f53c3fe8 ZK |
1682 | " host_addr: %llx handle: %x status: %d\n", |
1683 | gop[i].host_addr, | |
1684 | gop[i].handle, | |
1685 | gop[i].status); | |
1686 | } | |
1687 | BUG(); | |
1688 | } | |
1689 | } | |
1690 | ||
e9ce7cb6 WL |
1691 | for (i = 0; i < gop - queue->tx_unmap_ops; ++i) |
1692 | xenvif_idx_release(queue, pending_idx_release[i], | |
f53c3fe8 | 1693 | XEN_NETIF_RSP_OKAY); |
3e2234b3 ZK |
1694 | } |
1695 | ||
f53c3fe8 | 1696 | |
f942dc25 | 1697 | /* Called after netfront has transmitted */ |
e9ce7cb6 | 1698 | int xenvif_tx_action(struct xenvif_queue *queue, int budget) |
f942dc25 | 1699 | { |
bdab8275 | 1700 | unsigned nr_mops, nr_cops = 0; |
f53c3fe8 | 1701 | int work_done, ret; |
f942dc25 | 1702 | |
e9ce7cb6 | 1703 | if (unlikely(!tx_work_todo(queue))) |
b3f980bd WL |
1704 | return 0; |
1705 | ||
e9ce7cb6 | 1706 | xenvif_tx_build_gops(queue, budget, &nr_cops, &nr_mops); |
f942dc25 | 1707 | |
bdab8275 | 1708 | if (nr_cops == 0) |
b3f980bd WL |
1709 | return 0; |
1710 | ||
e9ce7cb6 | 1711 | gnttab_batch_copy(queue->tx_copy_ops, nr_cops); |
bdab8275 | 1712 | if (nr_mops != 0) { |
e9ce7cb6 | 1713 | ret = gnttab_map_refs(queue->tx_map_ops, |
bdab8275 | 1714 | NULL, |
e9ce7cb6 | 1715 | queue->pages_to_map, |
bdab8275 ZK |
1716 | nr_mops); |
1717 | BUG_ON(ret); | |
1718 | } | |
f942dc25 | 1719 | |
e9ce7cb6 | 1720 | work_done = xenvif_tx_submit(queue); |
f942dc25 | 1721 | |
b3f980bd | 1722 | return work_done; |
f942dc25 IC |
1723 | } |
1724 | ||
e9ce7cb6 | 1725 | static void xenvif_idx_release(struct xenvif_queue *queue, u16 pending_idx, |
7376419a | 1726 | u8 status) |
f942dc25 | 1727 | { |
f942dc25 | 1728 | struct pending_tx_info *pending_tx_info; |
f53c3fe8 | 1729 | pending_ring_idx_t index; |
f53c3fe8 | 1730 | unsigned long flags; |
2810e5b9 | 1731 | |
e9ce7cb6 WL |
1732 | pending_tx_info = &queue->pending_tx_info[pending_idx]; |
1733 | spin_lock_irqsave(&queue->response_lock, flags); | |
1734 | make_tx_response(queue, &pending_tx_info->req, status); | |
1735 | index = pending_index(queue->pending_prod); | |
1736 | queue->pending_ring[index] = pending_idx; | |
62bad319 ZK |
1737 | /* TX shouldn't use the index before we give it back here */ |
1738 | mb(); | |
e9ce7cb6 WL |
1739 | queue->pending_prod++; |
1740 | spin_unlock_irqrestore(&queue->response_lock, flags); | |
f942dc25 IC |
1741 | } |
1742 | ||
2810e5b9 | 1743 | |
e9ce7cb6 | 1744 | static void make_tx_response(struct xenvif_queue *queue, |
f942dc25 IC |
1745 | struct xen_netif_tx_request *txp, |
1746 | s8 st) | |
1747 | { | |
e9ce7cb6 | 1748 | RING_IDX i = queue->tx.rsp_prod_pvt; |
f942dc25 IC |
1749 | struct xen_netif_tx_response *resp; |
1750 | int notify; | |
1751 | ||
e9ce7cb6 | 1752 | resp = RING_GET_RESPONSE(&queue->tx, i); |
f942dc25 IC |
1753 | resp->id = txp->id; |
1754 | resp->status = st; | |
1755 | ||
1756 | if (txp->flags & XEN_NETTXF_extra_info) | |
e9ce7cb6 | 1757 | RING_GET_RESPONSE(&queue->tx, ++i)->status = XEN_NETIF_RSP_NULL; |
f942dc25 | 1758 | |
e9ce7cb6 WL |
1759 | queue->tx.rsp_prod_pvt = ++i; |
1760 | RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&queue->tx, notify); | |
f942dc25 | 1761 | if (notify) |
e9ce7cb6 | 1762 | notify_remote_via_irq(queue->tx_irq); |
f942dc25 IC |
1763 | } |
1764 | ||
e9ce7cb6 | 1765 | static struct xen_netif_rx_response *make_rx_response(struct xenvif_queue *queue, |
f942dc25 IC |
1766 | u16 id, |
1767 | s8 st, | |
1768 | u16 offset, | |
1769 | u16 size, | |
1770 | u16 flags) | |
1771 | { | |
e9ce7cb6 | 1772 | RING_IDX i = queue->rx.rsp_prod_pvt; |
f942dc25 IC |
1773 | struct xen_netif_rx_response *resp; |
1774 | ||
e9ce7cb6 | 1775 | resp = RING_GET_RESPONSE(&queue->rx, i); |
f942dc25 IC |
1776 | resp->offset = offset; |
1777 | resp->flags = flags; | |
1778 | resp->id = id; | |
1779 | resp->status = (s16)size; | |
1780 | if (st < 0) | |
1781 | resp->status = (s16)st; | |
1782 | ||
e9ce7cb6 | 1783 | queue->rx.rsp_prod_pvt = ++i; |
f942dc25 IC |
1784 | |
1785 | return resp; | |
1786 | } | |
1787 | ||
e9ce7cb6 | 1788 | void xenvif_idx_unmap(struct xenvif_queue *queue, u16 pending_idx) |
f53c3fe8 ZK |
1789 | { |
1790 | int ret; | |
1791 | struct gnttab_unmap_grant_ref tx_unmap_op; | |
1792 | ||
1793 | gnttab_set_unmap_op(&tx_unmap_op, | |
e9ce7cb6 | 1794 | idx_to_kaddr(queue, pending_idx), |
f53c3fe8 | 1795 | GNTMAP_host_map, |
e9ce7cb6 WL |
1796 | queue->grant_tx_handle[pending_idx]); |
1797 | xenvif_grant_handle_reset(queue, pending_idx); | |
f53c3fe8 ZK |
1798 | |
1799 | ret = gnttab_unmap_refs(&tx_unmap_op, NULL, | |
e9ce7cb6 | 1800 | &queue->mmap_pages[pending_idx], 1); |
7aceb47a | 1801 | if (ret) { |
e9ce7cb6 | 1802 | netdev_err(queue->vif->dev, |
7aceb47a ZK |
1803 | "Unmap fail: ret: %d pending_idx: %d host_addr: %llx handle: %x status: %d\n", |
1804 | ret, | |
1805 | pending_idx, | |
1806 | tx_unmap_op.host_addr, | |
1807 | tx_unmap_op.handle, | |
1808 | tx_unmap_op.status); | |
1809 | BUG(); | |
1810 | } | |
f53c3fe8 | 1811 | |
e9ce7cb6 | 1812 | xenvif_idx_release(queue, pending_idx, XEN_NETIF_RSP_OKAY); |
f53c3fe8 ZK |
1813 | } |
1814 | ||
e9ce7cb6 | 1815 | static inline int rx_work_todo(struct xenvif_queue *queue) |
f942dc25 | 1816 | { |
e9ce7cb6 WL |
1817 | return (!skb_queue_empty(&queue->rx_queue) && |
1818 | xenvif_rx_ring_slots_available(queue, queue->rx_last_skb_slots)) || | |
1819 | queue->rx_queue_purge; | |
f942dc25 IC |
1820 | } |
1821 | ||
e9ce7cb6 | 1822 | static inline int tx_work_todo(struct xenvif_queue *queue) |
f942dc25 | 1823 | { |
e9ce7cb6 | 1824 | if (likely(RING_HAS_UNCONSUMED_REQUESTS(&queue->tx))) |
f942dc25 IC |
1825 | return 1; |
1826 | ||
1827 | return 0; | |
1828 | } | |
1829 | ||
e9ce7cb6 | 1830 | static inline bool tx_dealloc_work_todo(struct xenvif_queue *queue) |
f53c3fe8 | 1831 | { |
e9ce7cb6 | 1832 | return queue->dealloc_cons != queue->dealloc_prod; |
f53c3fe8 ZK |
1833 | } |
1834 | ||
e9ce7cb6 | 1835 | void xenvif_unmap_frontend_rings(struct xenvif_queue *queue) |
f942dc25 | 1836 | { |
e9ce7cb6 WL |
1837 | if (queue->tx.sring) |
1838 | xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(queue->vif), | |
1839 | queue->tx.sring); | |
1840 | if (queue->rx.sring) | |
1841 | xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(queue->vif), | |
1842 | queue->rx.sring); | |
f942dc25 IC |
1843 | } |
1844 | ||
e9ce7cb6 | 1845 | int xenvif_map_frontend_rings(struct xenvif_queue *queue, |
7376419a WL |
1846 | grant_ref_t tx_ring_ref, |
1847 | grant_ref_t rx_ring_ref) | |
f942dc25 | 1848 | { |
c9d63699 | 1849 | void *addr; |
f942dc25 IC |
1850 | struct xen_netif_tx_sring *txs; |
1851 | struct xen_netif_rx_sring *rxs; | |
1852 | ||
1853 | int err = -ENOMEM; | |
1854 | ||
e9ce7cb6 | 1855 | err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(queue->vif), |
c9d63699 DV |
1856 | tx_ring_ref, &addr); |
1857 | if (err) | |
f942dc25 IC |
1858 | goto err; |
1859 | ||
c9d63699 | 1860 | txs = (struct xen_netif_tx_sring *)addr; |
e9ce7cb6 | 1861 | BACK_RING_INIT(&queue->tx, txs, PAGE_SIZE); |
f942dc25 | 1862 | |
e9ce7cb6 | 1863 | err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(queue->vif), |
c9d63699 DV |
1864 | rx_ring_ref, &addr); |
1865 | if (err) | |
f942dc25 | 1866 | goto err; |
f942dc25 | 1867 | |
c9d63699 | 1868 | rxs = (struct xen_netif_rx_sring *)addr; |
e9ce7cb6 | 1869 | BACK_RING_INIT(&queue->rx, rxs, PAGE_SIZE); |
f942dc25 IC |
1870 | |
1871 | return 0; | |
1872 | ||
1873 | err: | |
e9ce7cb6 | 1874 | xenvif_unmap_frontend_rings(queue); |
f942dc25 IC |
1875 | return err; |
1876 | } | |
1877 | ||
e9ce7cb6 | 1878 | static void xenvif_start_queue(struct xenvif_queue *queue) |
ca2f09f2 | 1879 | { |
e9ce7cb6 WL |
1880 | if (xenvif_schedulable(queue->vif)) |
1881 | xenvif_wake_queue(queue); | |
ca2f09f2 PD |
1882 | } |
1883 | ||
121fa4b7 | 1884 | int xenvif_kthread_guest_rx(void *data) |
b3f980bd | 1885 | { |
e9ce7cb6 | 1886 | struct xenvif_queue *queue = data; |
ca2f09f2 | 1887 | struct sk_buff *skb; |
b3f980bd WL |
1888 | |
1889 | while (!kthread_should_stop()) { | |
e9ce7cb6 WL |
1890 | wait_event_interruptible(queue->wq, |
1891 | rx_work_todo(queue) || | |
1892 | queue->vif->disabled || | |
b3f980bd | 1893 | kthread_should_stop()); |
e9d8b2c2 WL |
1894 | |
1895 | /* This frontend is found to be rogue, disable it in | |
1896 | * kthread context. Currently this is only set when | |
1897 | * netback finds out frontend sends malformed packet, | |
1898 | * but we cannot disable the interface in softirq | |
e9ce7cb6 WL |
1899 | * context so we defer it here, if this thread is |
1900 | * associated with queue 0. | |
e9d8b2c2 | 1901 | */ |
e9ce7cb6 WL |
1902 | if (unlikely(queue->vif->disabled && netif_carrier_ok(queue->vif->dev) && queue->id == 0)) |
1903 | xenvif_carrier_off(queue->vif); | |
e9d8b2c2 | 1904 | |
b3f980bd WL |
1905 | if (kthread_should_stop()) |
1906 | break; | |
1907 | ||
e9ce7cb6 WL |
1908 | if (queue->rx_queue_purge) { |
1909 | skb_queue_purge(&queue->rx_queue); | |
1910 | queue->rx_queue_purge = false; | |
09350788 ZK |
1911 | } |
1912 | ||
e9ce7cb6 WL |
1913 | if (!skb_queue_empty(&queue->rx_queue)) |
1914 | xenvif_rx_action(queue); | |
b3f980bd | 1915 | |
e9ce7cb6 WL |
1916 | if (skb_queue_empty(&queue->rx_queue) && |
1917 | xenvif_queue_stopped(queue)) { | |
1918 | del_timer_sync(&queue->wake_queue); | |
1919 | xenvif_start_queue(queue); | |
09350788 | 1920 | } |
ca2f09f2 | 1921 | |
b3f980bd WL |
1922 | cond_resched(); |
1923 | } | |
1924 | ||
ca2f09f2 | 1925 | /* Bin any remaining skbs */ |
e9ce7cb6 | 1926 | while ((skb = skb_dequeue(&queue->rx_queue)) != NULL) |
ca2f09f2 PD |
1927 | dev_kfree_skb(skb); |
1928 | ||
b3f980bd WL |
1929 | return 0; |
1930 | } | |
1931 | ||
f53c3fe8 ZK |
1932 | int xenvif_dealloc_kthread(void *data) |
1933 | { | |
e9ce7cb6 | 1934 | struct xenvif_queue *queue = data; |
f53c3fe8 ZK |
1935 | |
1936 | while (!kthread_should_stop()) { | |
e9ce7cb6 WL |
1937 | wait_event_interruptible(queue->dealloc_wq, |
1938 | tx_dealloc_work_todo(queue) || | |
f53c3fe8 ZK |
1939 | kthread_should_stop()); |
1940 | if (kthread_should_stop()) | |
1941 | break; | |
1942 | ||
e9ce7cb6 | 1943 | xenvif_tx_dealloc_action(queue); |
f53c3fe8 ZK |
1944 | cond_resched(); |
1945 | } | |
1946 | ||
1947 | /* Unmap anything remaining*/ | |
e9ce7cb6 WL |
1948 | if (tx_dealloc_work_todo(queue)) |
1949 | xenvif_tx_dealloc_action(queue); | |
f53c3fe8 ZK |
1950 | |
1951 | return 0; | |
1952 | } | |
1953 | ||
f942dc25 IC |
1954 | static int __init netback_init(void) |
1955 | { | |
f942dc25 | 1956 | int rc = 0; |
f942dc25 | 1957 | |
2a14b244 | 1958 | if (!xen_domain()) |
f942dc25 IC |
1959 | return -ENODEV; |
1960 | ||
8d3d53b3 AB |
1961 | /* Allow as many queues as there are CPUs, by default */ |
1962 | xenvif_max_queues = num_online_cpus(); | |
1963 | ||
37641494 | 1964 | if (fatal_skb_slots < XEN_NETBK_LEGACY_SLOTS_MAX) { |
383eda32 JP |
1965 | pr_info("fatal_skb_slots too small (%d), bump it to XEN_NETBK_LEGACY_SLOTS_MAX (%d)\n", |
1966 | fatal_skb_slots, XEN_NETBK_LEGACY_SLOTS_MAX); | |
37641494 | 1967 | fatal_skb_slots = XEN_NETBK_LEGACY_SLOTS_MAX; |
2810e5b9 WL |
1968 | } |
1969 | ||
f942dc25 IC |
1970 | rc = xenvif_xenbus_init(); |
1971 | if (rc) | |
1972 | goto failed_init; | |
1973 | ||
09350788 ZK |
1974 | rx_drain_timeout_jiffies = msecs_to_jiffies(rx_drain_timeout_msecs); |
1975 | ||
f942dc25 IC |
1976 | return 0; |
1977 | ||
1978 | failed_init: | |
f942dc25 | 1979 | return rc; |
f942dc25 IC |
1980 | } |
1981 | ||
1982 | module_init(netback_init); | |
1983 | ||
b103f358 WL |
1984 | static void __exit netback_fini(void) |
1985 | { | |
b103f358 | 1986 | xenvif_xenbus_fini(); |
b103f358 WL |
1987 | } |
1988 | module_exit(netback_fini); | |
1989 | ||
f942dc25 | 1990 | MODULE_LICENSE("Dual BSD/GPL"); |
f984cec6 | 1991 | MODULE_ALIAS("xen-backend:vif"); |