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 | */ | |
59ae9fc6 ZK |
171 | static bool start_new_rx_buffer(int offset, unsigned long size, int head, |
172 | bool full_coalesce) | |
f942dc25 IC |
173 | { |
174 | /* simple case: we have completely filled the current buffer. */ | |
175 | if (offset == MAX_BUFFER_OFFSET) | |
176 | return true; | |
177 | ||
178 | /* | |
179 | * complex case: start a fresh buffer if the current frag | |
180 | * would overflow the current buffer but only if: | |
181 | * (i) this frag would fit completely in the next buffer | |
182 | * and (ii) there is already some data in the current buffer | |
183 | * and (iii) this is not the head buffer. | |
59ae9fc6 | 184 | * and (iv) there is no need to fully utilize the buffers |
f942dc25 IC |
185 | * |
186 | * Where: | |
187 | * - (i) stops us splitting a frag into two copies | |
188 | * unless the frag is too large for a single buffer. | |
189 | * - (ii) stops us from leaving a buffer pointlessly empty. | |
190 | * - (iii) stops us leaving the first buffer | |
191 | * empty. Strictly speaking this is already covered | |
192 | * by (ii) but is explicitly checked because | |
193 | * netfront relies on the first buffer being | |
194 | * non-empty and can crash otherwise. | |
59ae9fc6 ZK |
195 | * - (iv) is needed for skbs which can use up more than MAX_SKB_FRAGS |
196 | * slot | |
f942dc25 IC |
197 | * |
198 | * This means we will effectively linearise small | |
199 | * frags but do not needlessly split large buffers | |
200 | * into multiple copies tend to give large frags their | |
201 | * own buffers as before. | |
202 | */ | |
0576eddf | 203 | BUG_ON(size > MAX_BUFFER_OFFSET); |
59ae9fc6 ZK |
204 | if ((offset + size > MAX_BUFFER_OFFSET) && offset && !head && |
205 | !full_coalesce) | |
f942dc25 IC |
206 | return true; |
207 | ||
208 | return false; | |
209 | } | |
210 | ||
f942dc25 IC |
211 | struct netrx_pending_operations { |
212 | unsigned copy_prod, copy_cons; | |
213 | unsigned meta_prod, meta_cons; | |
214 | struct gnttab_copy *copy; | |
b3f980bd | 215 | struct xenvif_rx_meta *meta; |
f942dc25 IC |
216 | int copy_off; |
217 | grant_ref_t copy_gref; | |
218 | }; | |
219 | ||
e9ce7cb6 | 220 | static struct xenvif_rx_meta *get_next_rx_buffer(struct xenvif_queue *queue, |
b3f980bd | 221 | struct netrx_pending_operations *npo) |
f942dc25 | 222 | { |
b3f980bd | 223 | struct xenvif_rx_meta *meta; |
f942dc25 IC |
224 | struct xen_netif_rx_request *req; |
225 | ||
e9ce7cb6 | 226 | req = RING_GET_REQUEST(&queue->rx, queue->rx.req_cons++); |
f942dc25 IC |
227 | |
228 | meta = npo->meta + npo->meta_prod++; | |
82cada22 | 229 | meta->gso_type = XEN_NETIF_GSO_TYPE_NONE; |
f942dc25 IC |
230 | meta->gso_size = 0; |
231 | meta->size = 0; | |
232 | meta->id = req->id; | |
233 | ||
234 | npo->copy_off = 0; | |
235 | npo->copy_gref = req->gref; | |
236 | ||
237 | return meta; | |
238 | } | |
239 | ||
59ae9fc6 ZK |
240 | struct xenvif_rx_cb { |
241 | int meta_slots_used; | |
242 | bool full_coalesce; | |
243 | }; | |
244 | ||
245 | #define XENVIF_RX_CB(skb) ((struct xenvif_rx_cb *)(skb)->cb) | |
246 | ||
33bc801d WL |
247 | /* |
248 | * Set up the grant operations for this fragment. If it's a flipping | |
249 | * interface, we also set up the unmap request from here. | |
250 | */ | |
e9ce7cb6 | 251 | static void xenvif_gop_frag_copy(struct xenvif_queue *queue, struct sk_buff *skb, |
7376419a WL |
252 | struct netrx_pending_operations *npo, |
253 | struct page *page, unsigned long size, | |
3e2234b3 | 254 | unsigned long offset, int *head, |
e9ce7cb6 | 255 | struct xenvif_queue *foreign_queue, |
3e2234b3 | 256 | grant_ref_t foreign_gref) |
f942dc25 IC |
257 | { |
258 | struct gnttab_copy *copy_gop; | |
b3f980bd | 259 | struct xenvif_rx_meta *meta; |
f942dc25 | 260 | unsigned long bytes; |
5bd07670 | 261 | int gso_type = XEN_NETIF_GSO_TYPE_NONE; |
f942dc25 IC |
262 | |
263 | /* Data must not cross a page boundary. */ | |
6a8ed462 | 264 | BUG_ON(size + offset > PAGE_SIZE<<compound_order(page)); |
f942dc25 IC |
265 | |
266 | meta = npo->meta + npo->meta_prod - 1; | |
267 | ||
6a8ed462 IC |
268 | /* Skip unused frames from start of page */ |
269 | page += offset >> PAGE_SHIFT; | |
270 | offset &= ~PAGE_MASK; | |
271 | ||
f942dc25 | 272 | while (size > 0) { |
6a8ed462 | 273 | BUG_ON(offset >= PAGE_SIZE); |
f942dc25 IC |
274 | BUG_ON(npo->copy_off > MAX_BUFFER_OFFSET); |
275 | ||
6a8ed462 IC |
276 | bytes = PAGE_SIZE - offset; |
277 | ||
278 | if (bytes > size) | |
279 | bytes = size; | |
280 | ||
59ae9fc6 ZK |
281 | if (start_new_rx_buffer(npo->copy_off, |
282 | bytes, | |
283 | *head, | |
284 | XENVIF_RX_CB(skb)->full_coalesce)) { | |
f942dc25 IC |
285 | /* |
286 | * Netfront requires there to be some data in the head | |
287 | * buffer. | |
288 | */ | |
33bc801d | 289 | BUG_ON(*head); |
f942dc25 | 290 | |
e9ce7cb6 | 291 | meta = get_next_rx_buffer(queue, npo); |
f942dc25 IC |
292 | } |
293 | ||
f942dc25 IC |
294 | if (npo->copy_off + bytes > MAX_BUFFER_OFFSET) |
295 | bytes = MAX_BUFFER_OFFSET - npo->copy_off; | |
296 | ||
297 | copy_gop = npo->copy + npo->copy_prod++; | |
298 | copy_gop->flags = GNTCOPY_dest_gref; | |
b3f980bd WL |
299 | copy_gop->len = bytes; |
300 | ||
e9ce7cb6 WL |
301 | if (foreign_queue) { |
302 | copy_gop->source.domid = foreign_queue->vif->domid; | |
3e2234b3 ZK |
303 | copy_gop->source.u.ref = foreign_gref; |
304 | copy_gop->flags |= GNTCOPY_source_gref; | |
305 | } else { | |
306 | copy_gop->source.domid = DOMID_SELF; | |
307 | copy_gop->source.u.gmfn = | |
308 | virt_to_mfn(page_address(page)); | |
309 | } | |
f942dc25 | 310 | copy_gop->source.offset = offset; |
f942dc25 | 311 | |
e9ce7cb6 | 312 | copy_gop->dest.domid = queue->vif->domid; |
f942dc25 IC |
313 | copy_gop->dest.offset = npo->copy_off; |
314 | copy_gop->dest.u.ref = npo->copy_gref; | |
f942dc25 IC |
315 | |
316 | npo->copy_off += bytes; | |
317 | meta->size += bytes; | |
318 | ||
319 | offset += bytes; | |
320 | size -= bytes; | |
321 | ||
6a8ed462 IC |
322 | /* Next frame */ |
323 | if (offset == PAGE_SIZE && size) { | |
324 | BUG_ON(!PageCompound(page)); | |
325 | page++; | |
326 | offset = 0; | |
327 | } | |
328 | ||
f942dc25 | 329 | /* Leave a gap for the GSO descriptor. */ |
5bd07670 AL |
330 | if (skb_is_gso(skb)) { |
331 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) | |
332 | gso_type = XEN_NETIF_GSO_TYPE_TCPV4; | |
333 | else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) | |
334 | gso_type = XEN_NETIF_GSO_TYPE_TCPV6; | |
335 | } | |
82cada22 | 336 | |
e9ce7cb6 WL |
337 | if (*head && ((1 << gso_type) & queue->vif->gso_mask)) |
338 | queue->rx.req_cons++; | |
f942dc25 | 339 | |
33bc801d | 340 | *head = 0; /* There must be something in this buffer now. */ |
f942dc25 IC |
341 | |
342 | } | |
343 | } | |
344 | ||
58375744 ZK |
345 | /* |
346 | * Find the grant ref for a given frag in a chain of struct ubuf_info's | |
347 | * skb: the skb itself | |
348 | * i: the frag's number | |
349 | * ubuf: a pointer to an element in the chain. It should not be NULL | |
350 | * | |
351 | * Returns a pointer to the element in the chain where the page were found. If | |
352 | * not found, returns NULL. | |
353 | * See the definition of callback_struct in common.h for more details about | |
354 | * the chain. | |
355 | */ | |
356 | static const struct ubuf_info *xenvif_find_gref(const struct sk_buff *const skb, | |
357 | const int i, | |
358 | const struct ubuf_info *ubuf) | |
359 | { | |
e9ce7cb6 | 360 | struct xenvif_queue *foreign_queue = ubuf_to_queue(ubuf); |
58375744 ZK |
361 | |
362 | do { | |
363 | u16 pending_idx = ubuf->desc; | |
364 | ||
365 | if (skb_shinfo(skb)->frags[i].page.p == | |
e9ce7cb6 | 366 | foreign_queue->mmap_pages[pending_idx]) |
58375744 ZK |
367 | break; |
368 | ubuf = (struct ubuf_info *) ubuf->ctx; | |
369 | } while (ubuf); | |
370 | ||
371 | return ubuf; | |
372 | } | |
373 | ||
f942dc25 IC |
374 | /* |
375 | * Prepare an SKB to be transmitted to the frontend. | |
376 | * | |
377 | * This function is responsible for allocating grant operations, meta | |
378 | * structures, etc. | |
379 | * | |
380 | * It returns the number of meta structures consumed. The number of | |
381 | * ring slots used is always equal to the number of meta slots used | |
382 | * plus the number of GSO descriptors used. Currently, we use either | |
383 | * zero GSO descriptors (for non-GSO packets) or one descriptor (for | |
384 | * frontend-side LRO). | |
385 | */ | |
7376419a | 386 | static int xenvif_gop_skb(struct sk_buff *skb, |
e9ce7cb6 WL |
387 | struct netrx_pending_operations *npo, |
388 | struct xenvif_queue *queue) | |
f942dc25 IC |
389 | { |
390 | struct xenvif *vif = netdev_priv(skb->dev); | |
391 | int nr_frags = skb_shinfo(skb)->nr_frags; | |
392 | int i; | |
393 | struct xen_netif_rx_request *req; | |
b3f980bd | 394 | struct xenvif_rx_meta *meta; |
f942dc25 | 395 | unsigned char *data; |
33bc801d | 396 | int head = 1; |
f942dc25 | 397 | int old_meta_prod; |
82cada22 | 398 | int gso_type; |
58375744 ZK |
399 | const struct ubuf_info *ubuf = skb_shinfo(skb)->destructor_arg; |
400 | const struct ubuf_info *const head_ubuf = ubuf; | |
f942dc25 IC |
401 | |
402 | old_meta_prod = npo->meta_prod; | |
403 | ||
5bd07670 AL |
404 | gso_type = XEN_NETIF_GSO_TYPE_NONE; |
405 | if (skb_is_gso(skb)) { | |
406 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) | |
407 | gso_type = XEN_NETIF_GSO_TYPE_TCPV4; | |
408 | else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) | |
409 | gso_type = XEN_NETIF_GSO_TYPE_TCPV6; | |
82cada22 PD |
410 | } |
411 | ||
f942dc25 | 412 | /* Set up a GSO prefix descriptor, if necessary */ |
a3314f3d | 413 | if ((1 << gso_type) & vif->gso_prefix_mask) { |
e9ce7cb6 | 414 | req = RING_GET_REQUEST(&queue->rx, queue->rx.req_cons++); |
f942dc25 | 415 | meta = npo->meta + npo->meta_prod++; |
82cada22 | 416 | meta->gso_type = gso_type; |
5bd07670 | 417 | meta->gso_size = skb_shinfo(skb)->gso_size; |
f942dc25 IC |
418 | meta->size = 0; |
419 | meta->id = req->id; | |
420 | } | |
421 | ||
e9ce7cb6 | 422 | req = RING_GET_REQUEST(&queue->rx, queue->rx.req_cons++); |
f942dc25 IC |
423 | meta = npo->meta + npo->meta_prod++; |
424 | ||
82cada22 PD |
425 | if ((1 << gso_type) & vif->gso_mask) { |
426 | meta->gso_type = gso_type; | |
5bd07670 | 427 | meta->gso_size = skb_shinfo(skb)->gso_size; |
82cada22 PD |
428 | } else { |
429 | meta->gso_type = XEN_NETIF_GSO_TYPE_NONE; | |
f942dc25 | 430 | meta->gso_size = 0; |
82cada22 | 431 | } |
f942dc25 IC |
432 | |
433 | meta->size = 0; | |
434 | meta->id = req->id; | |
435 | npo->copy_off = 0; | |
436 | npo->copy_gref = req->gref; | |
437 | ||
438 | data = skb->data; | |
439 | while (data < skb_tail_pointer(skb)) { | |
440 | unsigned int offset = offset_in_page(data); | |
441 | unsigned int len = PAGE_SIZE - offset; | |
442 | ||
443 | if (data + len > skb_tail_pointer(skb)) | |
444 | len = skb_tail_pointer(skb) - data; | |
445 | ||
e9ce7cb6 | 446 | xenvif_gop_frag_copy(queue, skb, npo, |
3e2234b3 ZK |
447 | virt_to_page(data), len, offset, &head, |
448 | NULL, | |
449 | 0); | |
f942dc25 IC |
450 | data += len; |
451 | } | |
452 | ||
453 | for (i = 0; i < nr_frags; i++) { | |
58375744 ZK |
454 | /* This variable also signals whether foreign_gref has a real |
455 | * value or not. | |
456 | */ | |
e9ce7cb6 | 457 | struct xenvif_queue *foreign_queue = NULL; |
58375744 ZK |
458 | grant_ref_t foreign_gref; |
459 | ||
460 | if ((skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) && | |
461 | (ubuf->callback == &xenvif_zerocopy_callback)) { | |
462 | const struct ubuf_info *const startpoint = ubuf; | |
463 | ||
464 | /* Ideally ubuf points to the chain element which | |
465 | * belongs to this frag. Or if frags were removed from | |
466 | * the beginning, then shortly before it. | |
467 | */ | |
468 | ubuf = xenvif_find_gref(skb, i, ubuf); | |
469 | ||
470 | /* Try again from the beginning of the list, if we | |
471 | * haven't tried from there. This only makes sense in | |
472 | * the unlikely event of reordering the original frags. | |
473 | * For injected local pages it's an unnecessary second | |
474 | * run. | |
475 | */ | |
476 | if (unlikely(!ubuf) && startpoint != head_ubuf) | |
477 | ubuf = xenvif_find_gref(skb, i, head_ubuf); | |
478 | ||
479 | if (likely(ubuf)) { | |
480 | u16 pending_idx = ubuf->desc; | |
481 | ||
e9ce7cb6 WL |
482 | foreign_queue = ubuf_to_queue(ubuf); |
483 | foreign_gref = | |
484 | foreign_queue->pending_tx_info[pending_idx].req.gref; | |
58375744 ZK |
485 | /* Just a safety measure. If this was the last |
486 | * element on the list, the for loop will | |
487 | * iterate again if a local page were added to | |
488 | * the end. Using head_ubuf here prevents the | |
489 | * second search on the chain. Or the original | |
490 | * frags changed order, but that's less likely. | |
491 | * In any way, ubuf shouldn't be NULL. | |
492 | */ | |
493 | ubuf = ubuf->ctx ? | |
494 | (struct ubuf_info *) ubuf->ctx : | |
495 | head_ubuf; | |
496 | } else | |
497 | /* This frag was a local page, added to the | |
498 | * array after the skb left netback. | |
499 | */ | |
500 | ubuf = head_ubuf; | |
501 | } | |
e9ce7cb6 | 502 | xenvif_gop_frag_copy(queue, skb, npo, |
7376419a WL |
503 | skb_frag_page(&skb_shinfo(skb)->frags[i]), |
504 | skb_frag_size(&skb_shinfo(skb)->frags[i]), | |
505 | skb_shinfo(skb)->frags[i].page_offset, | |
3e2234b3 | 506 | &head, |
e9ce7cb6 WL |
507 | foreign_queue, |
508 | foreign_queue ? foreign_gref : UINT_MAX); | |
f942dc25 IC |
509 | } |
510 | ||
511 | return npo->meta_prod - old_meta_prod; | |
512 | } | |
513 | ||
514 | /* | |
7376419a | 515 | * This is a twin to xenvif_gop_skb. Assume that xenvif_gop_skb was |
f942dc25 IC |
516 | * used to set up the operations on the top of |
517 | * netrx_pending_operations, which have since been done. Check that | |
518 | * they didn't give any errors and advance over them. | |
519 | */ | |
7376419a WL |
520 | static int xenvif_check_gop(struct xenvif *vif, int nr_meta_slots, |
521 | struct netrx_pending_operations *npo) | |
f942dc25 IC |
522 | { |
523 | struct gnttab_copy *copy_op; | |
524 | int status = XEN_NETIF_RSP_OKAY; | |
525 | int i; | |
526 | ||
527 | for (i = 0; i < nr_meta_slots; i++) { | |
528 | copy_op = npo->copy + npo->copy_cons++; | |
529 | if (copy_op->status != GNTST_okay) { | |
530 | netdev_dbg(vif->dev, | |
531 | "Bad status %d from copy to DOM%d.\n", | |
532 | copy_op->status, vif->domid); | |
533 | status = XEN_NETIF_RSP_ERROR; | |
534 | } | |
535 | } | |
536 | ||
537 | return status; | |
538 | } | |
539 | ||
e9ce7cb6 | 540 | static void xenvif_add_frag_responses(struct xenvif_queue *queue, int status, |
7376419a WL |
541 | struct xenvif_rx_meta *meta, |
542 | int nr_meta_slots) | |
f942dc25 IC |
543 | { |
544 | int i; | |
545 | unsigned long offset; | |
546 | ||
547 | /* No fragments used */ | |
548 | if (nr_meta_slots <= 1) | |
549 | return; | |
550 | ||
551 | nr_meta_slots--; | |
552 | ||
553 | for (i = 0; i < nr_meta_slots; i++) { | |
554 | int flags; | |
555 | if (i == nr_meta_slots - 1) | |
556 | flags = 0; | |
557 | else | |
558 | flags = XEN_NETRXF_more_data; | |
559 | ||
560 | offset = 0; | |
e9ce7cb6 | 561 | make_rx_response(queue, meta[i].id, status, offset, |
f942dc25 IC |
562 | meta[i].size, flags); |
563 | } | |
564 | } | |
565 | ||
e9ce7cb6 | 566 | void xenvif_kick_thread(struct xenvif_queue *queue) |
b3f980bd | 567 | { |
e9ce7cb6 | 568 | wake_up(&queue->wq); |
b3f980bd WL |
569 | } |
570 | ||
e9ce7cb6 | 571 | static void xenvif_rx_action(struct xenvif_queue *queue) |
f942dc25 | 572 | { |
f942dc25 | 573 | s8 status; |
e1f00a69 | 574 | u16 flags; |
f942dc25 IC |
575 | struct xen_netif_rx_response *resp; |
576 | struct sk_buff_head rxq; | |
577 | struct sk_buff *skb; | |
578 | LIST_HEAD(notify); | |
579 | int ret; | |
f942dc25 | 580 | unsigned long offset; |
11b57f90 | 581 | bool need_to_notify = false; |
f942dc25 IC |
582 | |
583 | struct netrx_pending_operations npo = { | |
e9ce7cb6 WL |
584 | .copy = queue->grant_copy_op, |
585 | .meta = queue->meta, | |
f942dc25 IC |
586 | }; |
587 | ||
588 | skb_queue_head_init(&rxq); | |
589 | ||
e9ce7cb6 | 590 | while ((skb = skb_dequeue(&queue->rx_queue)) != NULL) { |
9ab9831b | 591 | RING_IDX max_slots_needed; |
1425c7a4 PD |
592 | RING_IDX old_req_cons; |
593 | RING_IDX ring_slots_used; | |
ca2f09f2 PD |
594 | int i; |
595 | ||
596 | /* We need a cheap worse case estimate for the number of | |
597 | * slots we'll use. | |
598 | */ | |
599 | ||
600 | max_slots_needed = DIV_ROUND_UP(offset_in_page(skb->data) + | |
601 | skb_headlen(skb), | |
602 | PAGE_SIZE); | |
603 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
604 | unsigned int size; | |
a02eb473 PD |
605 | unsigned int offset; |
606 | ||
ca2f09f2 | 607 | size = skb_frag_size(&skb_shinfo(skb)->frags[i]); |
a02eb473 PD |
608 | offset = skb_shinfo(skb)->frags[i].page_offset; |
609 | ||
610 | /* For a worse-case estimate we need to factor in | |
611 | * the fragment page offset as this will affect the | |
612 | * number of times xenvif_gop_frag_copy() will | |
613 | * call start_new_rx_buffer(). | |
614 | */ | |
615 | max_slots_needed += DIV_ROUND_UP(offset + size, | |
616 | PAGE_SIZE); | |
ca2f09f2 | 617 | } |
a02eb473 PD |
618 | |
619 | /* To avoid the estimate becoming too pessimal for some | |
620 | * frontends that limit posted rx requests, cap the estimate | |
59ae9fc6 ZK |
621 | * at MAX_SKB_FRAGS. In this case netback will fully coalesce |
622 | * the skb into the provided slots. | |
a02eb473 | 623 | */ |
59ae9fc6 | 624 | if (max_slots_needed > MAX_SKB_FRAGS) { |
a02eb473 | 625 | max_slots_needed = MAX_SKB_FRAGS; |
59ae9fc6 ZK |
626 | XENVIF_RX_CB(skb)->full_coalesce = true; |
627 | } else { | |
628 | XENVIF_RX_CB(skb)->full_coalesce = false; | |
629 | } | |
a02eb473 PD |
630 | |
631 | /* We may need one more slot for GSO metadata */ | |
5bd07670 AL |
632 | if (skb_is_gso(skb) && |
633 | (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4 || | |
634 | skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)) | |
ca2f09f2 PD |
635 | max_slots_needed++; |
636 | ||
637 | /* If the skb may not fit then bail out now */ | |
e9ce7cb6 WL |
638 | if (!xenvif_rx_ring_slots_available(queue, max_slots_needed)) { |
639 | skb_queue_head(&queue->rx_queue, skb); | |
11b57f90 | 640 | need_to_notify = true; |
e9ce7cb6 | 641 | queue->rx_last_skb_slots = max_slots_needed; |
ca2f09f2 | 642 | break; |
9ab9831b | 643 | } else |
e9ce7cb6 | 644 | queue->rx_last_skb_slots = 0; |
f942dc25 | 645 | |
e9ce7cb6 WL |
646 | old_req_cons = queue->rx.req_cons; |
647 | XENVIF_RX_CB(skb)->meta_slots_used = xenvif_gop_skb(skb, &npo, queue); | |
648 | ring_slots_used = queue->rx.req_cons - old_req_cons; | |
1425c7a4 PD |
649 | |
650 | BUG_ON(ring_slots_used > max_slots_needed); | |
f942dc25 IC |
651 | |
652 | __skb_queue_tail(&rxq, skb); | |
f942dc25 IC |
653 | } |
654 | ||
e9ce7cb6 | 655 | BUG_ON(npo.meta_prod > ARRAY_SIZE(queue->meta)); |
f942dc25 IC |
656 | |
657 | if (!npo.copy_prod) | |
ca2f09f2 | 658 | goto done; |
f942dc25 | 659 | |
ac3d5ac2 | 660 | BUG_ON(npo.copy_prod > MAX_GRANT_COPY_OPS); |
e9ce7cb6 | 661 | gnttab_batch_copy(queue->grant_copy_op, npo.copy_prod); |
f942dc25 IC |
662 | |
663 | while ((skb = __skb_dequeue(&rxq)) != NULL) { | |
f942dc25 | 664 | |
e9ce7cb6 WL |
665 | if ((1 << queue->meta[npo.meta_cons].gso_type) & |
666 | queue->vif->gso_prefix_mask) { | |
667 | resp = RING_GET_RESPONSE(&queue->rx, | |
668 | queue->rx.rsp_prod_pvt++); | |
f942dc25 IC |
669 | |
670 | resp->flags = XEN_NETRXF_gso_prefix | XEN_NETRXF_more_data; | |
671 | ||
e9ce7cb6 WL |
672 | resp->offset = queue->meta[npo.meta_cons].gso_size; |
673 | resp->id = queue->meta[npo.meta_cons].id; | |
8f13dd96 | 674 | resp->status = XENVIF_RX_CB(skb)->meta_slots_used; |
f942dc25 IC |
675 | |
676 | npo.meta_cons++; | |
8f13dd96 | 677 | XENVIF_RX_CB(skb)->meta_slots_used--; |
f942dc25 IC |
678 | } |
679 | ||
680 | ||
e9ce7cb6 WL |
681 | queue->stats.tx_bytes += skb->len; |
682 | queue->stats.tx_packets++; | |
f942dc25 | 683 | |
e9ce7cb6 | 684 | status = xenvif_check_gop(queue->vif, |
8f13dd96 ZK |
685 | XENVIF_RX_CB(skb)->meta_slots_used, |
686 | &npo); | |
f942dc25 | 687 | |
8f13dd96 | 688 | if (XENVIF_RX_CB(skb)->meta_slots_used == 1) |
f942dc25 IC |
689 | flags = 0; |
690 | else | |
691 | flags = XEN_NETRXF_more_data; | |
692 | ||
693 | if (skb->ip_summed == CHECKSUM_PARTIAL) /* local packet? */ | |
694 | flags |= XEN_NETRXF_csum_blank | XEN_NETRXF_data_validated; | |
695 | else if (skb->ip_summed == CHECKSUM_UNNECESSARY) | |
696 | /* remote but checksummed. */ | |
697 | flags |= XEN_NETRXF_data_validated; | |
698 | ||
699 | offset = 0; | |
e9ce7cb6 | 700 | resp = make_rx_response(queue, queue->meta[npo.meta_cons].id, |
f942dc25 | 701 | status, offset, |
e9ce7cb6 | 702 | queue->meta[npo.meta_cons].size, |
f942dc25 IC |
703 | flags); |
704 | ||
e9ce7cb6 WL |
705 | if ((1 << queue->meta[npo.meta_cons].gso_type) & |
706 | queue->vif->gso_mask) { | |
f942dc25 IC |
707 | struct xen_netif_extra_info *gso = |
708 | (struct xen_netif_extra_info *) | |
e9ce7cb6 WL |
709 | RING_GET_RESPONSE(&queue->rx, |
710 | queue->rx.rsp_prod_pvt++); | |
f942dc25 IC |
711 | |
712 | resp->flags |= XEN_NETRXF_extra_info; | |
713 | ||
e9ce7cb6 WL |
714 | gso->u.gso.type = queue->meta[npo.meta_cons].gso_type; |
715 | gso->u.gso.size = queue->meta[npo.meta_cons].gso_size; | |
f942dc25 IC |
716 | gso->u.gso.pad = 0; |
717 | gso->u.gso.features = 0; | |
718 | ||
719 | gso->type = XEN_NETIF_EXTRA_TYPE_GSO; | |
720 | gso->flags = 0; | |
721 | } | |
722 | ||
e9ce7cb6 WL |
723 | xenvif_add_frag_responses(queue, status, |
724 | queue->meta + npo.meta_cons + 1, | |
8f13dd96 | 725 | XENVIF_RX_CB(skb)->meta_slots_used); |
f942dc25 | 726 | |
e9ce7cb6 | 727 | RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&queue->rx, ret); |
f942dc25 | 728 | |
11b57f90 | 729 | need_to_notify |= !!ret; |
b3f980bd | 730 | |
8f13dd96 | 731 | npo.meta_cons += XENVIF_RX_CB(skb)->meta_slots_used; |
f942dc25 IC |
732 | dev_kfree_skb(skb); |
733 | } | |
734 | ||
ca2f09f2 | 735 | done: |
b3f980bd | 736 | if (need_to_notify) |
e9ce7cb6 | 737 | notify_remote_via_irq(queue->rx_irq); |
f942dc25 IC |
738 | } |
739 | ||
e9ce7cb6 | 740 | void xenvif_napi_schedule_or_enable_events(struct xenvif_queue *queue) |
f942dc25 IC |
741 | { |
742 | int more_to_do; | |
743 | ||
e9ce7cb6 | 744 | RING_FINAL_CHECK_FOR_REQUESTS(&queue->tx, more_to_do); |
f942dc25 IC |
745 | |
746 | if (more_to_do) | |
e9ce7cb6 | 747 | napi_schedule(&queue->napi); |
f942dc25 IC |
748 | } |
749 | ||
e9ce7cb6 | 750 | static void tx_add_credit(struct xenvif_queue *queue) |
f942dc25 IC |
751 | { |
752 | unsigned long max_burst, max_credit; | |
753 | ||
754 | /* | |
755 | * Allow a burst big enough to transmit a jumbo packet of up to 128kB. | |
756 | * Otherwise the interface can seize up due to insufficient credit. | |
757 | */ | |
e9ce7cb6 | 758 | max_burst = RING_GET_REQUEST(&queue->tx, queue->tx.req_cons)->size; |
f942dc25 | 759 | max_burst = min(max_burst, 131072UL); |
e9ce7cb6 | 760 | max_burst = max(max_burst, queue->credit_bytes); |
f942dc25 IC |
761 | |
762 | /* Take care that adding a new chunk of credit doesn't wrap to zero. */ | |
e9ce7cb6 WL |
763 | max_credit = queue->remaining_credit + queue->credit_bytes; |
764 | if (max_credit < queue->remaining_credit) | |
f942dc25 IC |
765 | max_credit = ULONG_MAX; /* wrapped: clamp to ULONG_MAX */ |
766 | ||
e9ce7cb6 | 767 | queue->remaining_credit = min(max_credit, max_burst); |
f942dc25 IC |
768 | } |
769 | ||
770 | static void tx_credit_callback(unsigned long data) | |
771 | { | |
e9ce7cb6 WL |
772 | struct xenvif_queue *queue = (struct xenvif_queue *)data; |
773 | tx_add_credit(queue); | |
774 | xenvif_napi_schedule_or_enable_events(queue); | |
f942dc25 IC |
775 | } |
776 | ||
e9ce7cb6 | 777 | static void xenvif_tx_err(struct xenvif_queue *queue, |
7376419a | 778 | struct xen_netif_tx_request *txp, RING_IDX end) |
f942dc25 | 779 | { |
e9ce7cb6 | 780 | RING_IDX cons = queue->tx.req_cons; |
f53c3fe8 | 781 | unsigned long flags; |
f942dc25 IC |
782 | |
783 | do { | |
e9ce7cb6 WL |
784 | spin_lock_irqsave(&queue->response_lock, flags); |
785 | make_tx_response(queue, txp, XEN_NETIF_RSP_ERROR); | |
786 | spin_unlock_irqrestore(&queue->response_lock, flags); | |
b9149729 | 787 | if (cons == end) |
f942dc25 | 788 | break; |
e9ce7cb6 | 789 | txp = RING_GET_REQUEST(&queue->tx, cons++); |
f942dc25 | 790 | } while (1); |
e9ce7cb6 | 791 | queue->tx.req_cons = cons; |
f942dc25 IC |
792 | } |
793 | ||
7376419a | 794 | static void xenvif_fatal_tx_err(struct xenvif *vif) |
48856286 IC |
795 | { |
796 | netdev_err(vif->dev, "fatal error; disabling device\n"); | |
e9d8b2c2 | 797 | vif->disabled = true; |
e9ce7cb6 WL |
798 | /* Disable the vif from queue 0's kthread */ |
799 | if (vif->queues) | |
800 | xenvif_kick_thread(&vif->queues[0]); | |
48856286 IC |
801 | } |
802 | ||
e9ce7cb6 | 803 | static int xenvif_count_requests(struct xenvif_queue *queue, |
7376419a WL |
804 | struct xen_netif_tx_request *first, |
805 | struct xen_netif_tx_request *txp, | |
806 | int work_to_do) | |
f942dc25 | 807 | { |
e9ce7cb6 | 808 | RING_IDX cons = queue->tx.req_cons; |
2810e5b9 WL |
809 | int slots = 0; |
810 | int drop_err = 0; | |
59ccb4eb | 811 | int more_data; |
f942dc25 IC |
812 | |
813 | if (!(first->flags & XEN_NETTXF_more_data)) | |
814 | return 0; | |
815 | ||
816 | do { | |
59ccb4eb WL |
817 | struct xen_netif_tx_request dropped_tx = { 0 }; |
818 | ||
2810e5b9 | 819 | if (slots >= work_to_do) { |
e9ce7cb6 | 820 | netdev_err(queue->vif->dev, |
2810e5b9 WL |
821 | "Asked for %d slots but exceeds this limit\n", |
822 | work_to_do); | |
e9ce7cb6 | 823 | xenvif_fatal_tx_err(queue->vif); |
35876b5f | 824 | return -ENODATA; |
f942dc25 IC |
825 | } |
826 | ||
2810e5b9 WL |
827 | /* This guest is really using too many slots and |
828 | * considered malicious. | |
829 | */ | |
37641494 | 830 | if (unlikely(slots >= fatal_skb_slots)) { |
e9ce7cb6 | 831 | netdev_err(queue->vif->dev, |
2810e5b9 | 832 | "Malicious frontend using %d slots, threshold %u\n", |
37641494 | 833 | slots, fatal_skb_slots); |
e9ce7cb6 | 834 | xenvif_fatal_tx_err(queue->vif); |
35876b5f | 835 | return -E2BIG; |
f942dc25 IC |
836 | } |
837 | ||
2810e5b9 | 838 | /* Xen network protocol had implicit dependency on |
37641494 WL |
839 | * MAX_SKB_FRAGS. XEN_NETBK_LEGACY_SLOTS_MAX is set to |
840 | * the historical MAX_SKB_FRAGS value 18 to honor the | |
841 | * same behavior as before. Any packet using more than | |
842 | * 18 slots but less than fatal_skb_slots slots is | |
843 | * dropped | |
2810e5b9 | 844 | */ |
37641494 | 845 | if (!drop_err && slots >= XEN_NETBK_LEGACY_SLOTS_MAX) { |
2810e5b9 | 846 | if (net_ratelimit()) |
e9ce7cb6 | 847 | netdev_dbg(queue->vif->dev, |
2810e5b9 | 848 | "Too many slots (%d) exceeding limit (%d), dropping packet\n", |
37641494 | 849 | slots, XEN_NETBK_LEGACY_SLOTS_MAX); |
2810e5b9 WL |
850 | drop_err = -E2BIG; |
851 | } | |
852 | ||
59ccb4eb WL |
853 | if (drop_err) |
854 | txp = &dropped_tx; | |
855 | ||
e9ce7cb6 | 856 | memcpy(txp, RING_GET_REQUEST(&queue->tx, cons + slots), |
f942dc25 | 857 | sizeof(*txp)); |
03393fd5 WL |
858 | |
859 | /* If the guest submitted a frame >= 64 KiB then | |
860 | * first->size overflowed and following slots will | |
861 | * appear to be larger than the frame. | |
862 | * | |
863 | * This cannot be fatal error as there are buggy | |
864 | * frontends that do this. | |
865 | * | |
866 | * Consume all slots and drop the packet. | |
867 | */ | |
868 | if (!drop_err && txp->size > first->size) { | |
869 | if (net_ratelimit()) | |
e9ce7cb6 | 870 | netdev_dbg(queue->vif->dev, |
03393fd5 WL |
871 | "Invalid tx request, slot size %u > remaining size %u\n", |
872 | txp->size, first->size); | |
873 | drop_err = -EIO; | |
f942dc25 IC |
874 | } |
875 | ||
876 | first->size -= txp->size; | |
2810e5b9 | 877 | slots++; |
f942dc25 IC |
878 | |
879 | if (unlikely((txp->offset + txp->size) > PAGE_SIZE)) { | |
e9ce7cb6 | 880 | netdev_err(queue->vif->dev, "Cross page boundary, txp->offset: %x, size: %u\n", |
f942dc25 | 881 | txp->offset, txp->size); |
e9ce7cb6 | 882 | xenvif_fatal_tx_err(queue->vif); |
35876b5f | 883 | return -EINVAL; |
f942dc25 | 884 | } |
59ccb4eb WL |
885 | |
886 | more_data = txp->flags & XEN_NETTXF_more_data; | |
887 | ||
888 | if (!drop_err) | |
889 | txp++; | |
890 | ||
891 | } while (more_data); | |
2810e5b9 WL |
892 | |
893 | if (drop_err) { | |
e9ce7cb6 | 894 | xenvif_tx_err(queue, first, cons + slots); |
2810e5b9 WL |
895 | return drop_err; |
896 | } | |
897 | ||
898 | return slots; | |
f942dc25 IC |
899 | } |
900 | ||
8f13dd96 ZK |
901 | |
902 | struct xenvif_tx_cb { | |
903 | u16 pending_idx; | |
904 | }; | |
905 | ||
906 | #define XENVIF_TX_CB(skb) ((struct xenvif_tx_cb *)(skb)->cb) | |
907 | ||
e9ce7cb6 | 908 | static inline void xenvif_tx_create_map_op(struct xenvif_queue *queue, |
9074ce24 ZK |
909 | u16 pending_idx, |
910 | struct xen_netif_tx_request *txp, | |
911 | struct gnttab_map_grant_ref *mop) | |
f53c3fe8 | 912 | { |
e9ce7cb6 WL |
913 | queue->pages_to_map[mop-queue->tx_map_ops] = queue->mmap_pages[pending_idx]; |
914 | gnttab_set_map_op(mop, idx_to_kaddr(queue, pending_idx), | |
f53c3fe8 | 915 | GNTMAP_host_map | GNTMAP_readonly, |
e9ce7cb6 | 916 | txp->gref, queue->vif->domid); |
f53c3fe8 | 917 | |
e9ce7cb6 | 918 | memcpy(&queue->pending_tx_info[pending_idx].req, txp, |
f53c3fe8 ZK |
919 | sizeof(*txp)); |
920 | } | |
921 | ||
e3377f36 ZK |
922 | static inline struct sk_buff *xenvif_alloc_skb(unsigned int size) |
923 | { | |
924 | struct sk_buff *skb = | |
925 | alloc_skb(size + NET_SKB_PAD + NET_IP_ALIGN, | |
926 | GFP_ATOMIC | __GFP_NOWARN); | |
927 | if (unlikely(skb == NULL)) | |
928 | return NULL; | |
929 | ||
930 | /* Packets passed to netif_rx() must have some headroom. */ | |
931 | skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN); | |
932 | ||
933 | /* Initialize it here to avoid later surprises */ | |
934 | skb_shinfo(skb)->destructor_arg = NULL; | |
935 | ||
936 | return skb; | |
937 | } | |
938 | ||
e9ce7cb6 | 939 | static struct gnttab_map_grant_ref *xenvif_get_requests(struct xenvif_queue *queue, |
f53c3fe8 ZK |
940 | struct sk_buff *skb, |
941 | struct xen_netif_tx_request *txp, | |
942 | struct gnttab_map_grant_ref *gop) | |
f942dc25 IC |
943 | { |
944 | struct skb_shared_info *shinfo = skb_shinfo(skb); | |
945 | skb_frag_t *frags = shinfo->frags; | |
8f13dd96 | 946 | u16 pending_idx = XENVIF_TX_CB(skb)->pending_idx; |
62bad319 ZK |
947 | int start; |
948 | pending_ring_idx_t index; | |
e3377f36 | 949 | unsigned int nr_slots, frag_overflow = 0; |
2810e5b9 WL |
950 | |
951 | /* At this point shinfo->nr_frags is in fact the number of | |
37641494 | 952 | * slots, which can be as large as XEN_NETBK_LEGACY_SLOTS_MAX. |
2810e5b9 | 953 | */ |
e3377f36 ZK |
954 | if (shinfo->nr_frags > MAX_SKB_FRAGS) { |
955 | frag_overflow = shinfo->nr_frags - MAX_SKB_FRAGS; | |
956 | BUG_ON(frag_overflow > MAX_SKB_FRAGS); | |
957 | shinfo->nr_frags = MAX_SKB_FRAGS; | |
958 | } | |
2810e5b9 | 959 | nr_slots = shinfo->nr_frags; |
f942dc25 IC |
960 | |
961 | /* Skip first skb fragment if it is on same page as header fragment. */ | |
ea066ad1 | 962 | start = (frag_get_pending_idx(&shinfo->frags[0]) == pending_idx); |
f942dc25 | 963 | |
f53c3fe8 ZK |
964 | for (shinfo->nr_frags = start; shinfo->nr_frags < nr_slots; |
965 | shinfo->nr_frags++, txp++, gop++) { | |
e9ce7cb6 WL |
966 | index = pending_index(queue->pending_cons++); |
967 | pending_idx = queue->pending_ring[index]; | |
968 | xenvif_tx_create_map_op(queue, pending_idx, txp, gop); | |
f53c3fe8 | 969 | frag_set_pending_idx(&frags[shinfo->nr_frags], pending_idx); |
f942dc25 IC |
970 | } |
971 | ||
e3377f36 ZK |
972 | if (frag_overflow) { |
973 | struct sk_buff *nskb = xenvif_alloc_skb(0); | |
974 | if (unlikely(nskb == NULL)) { | |
975 | if (net_ratelimit()) | |
e9ce7cb6 | 976 | netdev_err(queue->vif->dev, |
e3377f36 ZK |
977 | "Can't allocate the frag_list skb.\n"); |
978 | return NULL; | |
979 | } | |
980 | ||
981 | shinfo = skb_shinfo(nskb); | |
982 | frags = shinfo->frags; | |
983 | ||
984 | for (shinfo->nr_frags = 0; shinfo->nr_frags < frag_overflow; | |
985 | shinfo->nr_frags++, txp++, gop++) { | |
e9ce7cb6 WL |
986 | index = pending_index(queue->pending_cons++); |
987 | pending_idx = queue->pending_ring[index]; | |
988 | xenvif_tx_create_map_op(queue, pending_idx, txp, gop); | |
e3377f36 ZK |
989 | frag_set_pending_idx(&frags[shinfo->nr_frags], |
990 | pending_idx); | |
991 | } | |
992 | ||
993 | skb_shinfo(skb)->frag_list = nskb; | |
994 | } | |
2810e5b9 | 995 | |
f942dc25 IC |
996 | return gop; |
997 | } | |
998 | ||
e9ce7cb6 | 999 | static inline void xenvif_grant_handle_set(struct xenvif_queue *queue, |
f53c3fe8 ZK |
1000 | u16 pending_idx, |
1001 | grant_handle_t handle) | |
1002 | { | |
e9ce7cb6 | 1003 | if (unlikely(queue->grant_tx_handle[pending_idx] != |
f53c3fe8 | 1004 | NETBACK_INVALID_HANDLE)) { |
e9ce7cb6 | 1005 | netdev_err(queue->vif->dev, |
f53c3fe8 ZK |
1006 | "Trying to overwrite active handle! pending_idx: %x\n", |
1007 | pending_idx); | |
1008 | BUG(); | |
1009 | } | |
e9ce7cb6 | 1010 | queue->grant_tx_handle[pending_idx] = handle; |
f53c3fe8 ZK |
1011 | } |
1012 | ||
e9ce7cb6 | 1013 | static inline void xenvif_grant_handle_reset(struct xenvif_queue *queue, |
f53c3fe8 ZK |
1014 | u16 pending_idx) |
1015 | { | |
e9ce7cb6 | 1016 | if (unlikely(queue->grant_tx_handle[pending_idx] == |
f53c3fe8 | 1017 | NETBACK_INVALID_HANDLE)) { |
e9ce7cb6 | 1018 | netdev_err(queue->vif->dev, |
f53c3fe8 ZK |
1019 | "Trying to unmap invalid handle! pending_idx: %x\n", |
1020 | pending_idx); | |
1021 | BUG(); | |
1022 | } | |
e9ce7cb6 | 1023 | queue->grant_tx_handle[pending_idx] = NETBACK_INVALID_HANDLE; |
f53c3fe8 ZK |
1024 | } |
1025 | ||
e9ce7cb6 | 1026 | static int xenvif_tx_check_gop(struct xenvif_queue *queue, |
7376419a | 1027 | struct sk_buff *skb, |
bdab8275 ZK |
1028 | struct gnttab_map_grant_ref **gopp_map, |
1029 | struct gnttab_copy **gopp_copy) | |
f942dc25 | 1030 | { |
9074ce24 | 1031 | struct gnttab_map_grant_ref *gop_map = *gopp_map; |
8f13dd96 | 1032 | u16 pending_idx = XENVIF_TX_CB(skb)->pending_idx; |
1a998d3e ZK |
1033 | /* This always points to the shinfo of the skb being checked, which |
1034 | * could be either the first or the one on the frag_list | |
1035 | */ | |
f942dc25 | 1036 | struct skb_shared_info *shinfo = skb_shinfo(skb); |
1a998d3e ZK |
1037 | /* If this is non-NULL, we are currently checking the frag_list skb, and |
1038 | * this points to the shinfo of the first one | |
1039 | */ | |
1040 | struct skb_shared_info *first_shinfo = NULL; | |
f942dc25 | 1041 | int nr_frags = shinfo->nr_frags; |
1b860da0 ZK |
1042 | const bool sharedslot = nr_frags && |
1043 | frag_get_pending_idx(&shinfo->frags[0]) == pending_idx; | |
bdab8275 | 1044 | int i, err; |
f942dc25 IC |
1045 | |
1046 | /* Check status of header. */ | |
bdab8275 | 1047 | err = (*gopp_copy)->status; |
bdab8275 ZK |
1048 | if (unlikely(err)) { |
1049 | if (net_ratelimit()) | |
e9ce7cb6 | 1050 | netdev_dbg(queue->vif->dev, |
00aefceb | 1051 | "Grant copy of header failed! status: %d pending_idx: %u ref: %u\n", |
bdab8275 ZK |
1052 | (*gopp_copy)->status, |
1053 | pending_idx, | |
1054 | (*gopp_copy)->source.u.ref); | |
1b860da0 ZK |
1055 | /* The first frag might still have this slot mapped */ |
1056 | if (!sharedslot) | |
1057 | xenvif_idx_release(queue, pending_idx, | |
1058 | XEN_NETIF_RSP_ERROR); | |
bdab8275 | 1059 | } |
d8cfbfc4 | 1060 | (*gopp_copy)++; |
f942dc25 | 1061 | |
e3377f36 | 1062 | check_frags: |
bdab8275 | 1063 | for (i = 0; i < nr_frags; i++, gop_map++) { |
f942dc25 | 1064 | int j, newerr; |
f942dc25 | 1065 | |
ea066ad1 | 1066 | pending_idx = frag_get_pending_idx(&shinfo->frags[i]); |
f942dc25 IC |
1067 | |
1068 | /* Check error status: if okay then remember grant handle. */ | |
bdab8275 | 1069 | newerr = gop_map->status; |
2810e5b9 | 1070 | |
f942dc25 | 1071 | if (likely(!newerr)) { |
e9ce7cb6 | 1072 | xenvif_grant_handle_set(queue, |
9074ce24 ZK |
1073 | pending_idx, |
1074 | gop_map->handle); | |
f942dc25 | 1075 | /* Had a previous error? Invalidate this fragment. */ |
1b860da0 | 1076 | if (unlikely(err)) { |
e9ce7cb6 | 1077 | xenvif_idx_unmap(queue, pending_idx); |
1b860da0 ZK |
1078 | /* If the mapping of the first frag was OK, but |
1079 | * the header's copy failed, and they are | |
1080 | * sharing a slot, send an error | |
1081 | */ | |
1082 | if (i == 0 && sharedslot) | |
1083 | xenvif_idx_release(queue, pending_idx, | |
1084 | XEN_NETIF_RSP_ERROR); | |
1085 | else | |
1086 | xenvif_idx_release(queue, pending_idx, | |
1087 | XEN_NETIF_RSP_OKAY); | |
1088 | } | |
f942dc25 IC |
1089 | continue; |
1090 | } | |
1091 | ||
1092 | /* Error on this fragment: respond to client with an error. */ | |
bdab8275 | 1093 | if (net_ratelimit()) |
e9ce7cb6 | 1094 | netdev_dbg(queue->vif->dev, |
00aefceb | 1095 | "Grant map of %d. frag failed! status: %d pending_idx: %u ref: %u\n", |
bdab8275 ZK |
1096 | i, |
1097 | gop_map->status, | |
1098 | pending_idx, | |
1099 | gop_map->ref); | |
1b860da0 | 1100 | |
e9ce7cb6 | 1101 | xenvif_idx_release(queue, pending_idx, XEN_NETIF_RSP_ERROR); |
f942dc25 IC |
1102 | |
1103 | /* Not the first error? Preceding frags already invalidated. */ | |
1104 | if (err) | |
1105 | continue; | |
1b860da0 ZK |
1106 | |
1107 | /* First error: if the header haven't shared a slot with the | |
1108 | * first frag, release it as well. | |
1109 | */ | |
1110 | if (!sharedslot) | |
1111 | xenvif_idx_release(queue, | |
1112 | XENVIF_TX_CB(skb)->pending_idx, | |
1113 | XEN_NETIF_RSP_OKAY); | |
1114 | ||
1115 | /* Invalidate preceding fragments of this skb. */ | |
bdab8275 | 1116 | for (j = 0; j < i; j++) { |
5ccb3ea7 | 1117 | pending_idx = frag_get_pending_idx(&shinfo->frags[j]); |
e9ce7cb6 | 1118 | xenvif_idx_unmap(queue, pending_idx); |
1b860da0 ZK |
1119 | xenvif_idx_release(queue, pending_idx, |
1120 | XEN_NETIF_RSP_OKAY); | |
f942dc25 IC |
1121 | } |
1122 | ||
1a998d3e ZK |
1123 | /* And if we found the error while checking the frag_list, unmap |
1124 | * the first skb's frags | |
1125 | */ | |
1126 | if (first_shinfo) { | |
1127 | for (j = 0; j < first_shinfo->nr_frags; j++) { | |
1128 | pending_idx = frag_get_pending_idx(&first_shinfo->frags[j]); | |
1129 | xenvif_idx_unmap(queue, pending_idx); | |
1b860da0 ZK |
1130 | xenvif_idx_release(queue, pending_idx, |
1131 | XEN_NETIF_RSP_OKAY); | |
1a998d3e ZK |
1132 | } |
1133 | } | |
1134 | ||
f942dc25 IC |
1135 | /* Remember the error: invalidate all subsequent fragments. */ |
1136 | err = newerr; | |
1137 | } | |
1138 | ||
1a998d3e ZK |
1139 | if (skb_has_frag_list(skb) && !first_shinfo) { |
1140 | first_shinfo = skb_shinfo(skb); | |
1141 | shinfo = skb_shinfo(skb_shinfo(skb)->frag_list); | |
e3377f36 | 1142 | nr_frags = shinfo->nr_frags; |
e3377f36 ZK |
1143 | |
1144 | goto check_frags; | |
1145 | } | |
1146 | ||
bdab8275 | 1147 | *gopp_map = gop_map; |
f942dc25 IC |
1148 | return err; |
1149 | } | |
1150 | ||
e9ce7cb6 | 1151 | static void xenvif_fill_frags(struct xenvif_queue *queue, struct sk_buff *skb) |
f942dc25 IC |
1152 | { |
1153 | struct skb_shared_info *shinfo = skb_shinfo(skb); | |
1154 | int nr_frags = shinfo->nr_frags; | |
1155 | int i; | |
f53c3fe8 ZK |
1156 | u16 prev_pending_idx = INVALID_PENDING_IDX; |
1157 | ||
f942dc25 IC |
1158 | for (i = 0; i < nr_frags; i++) { |
1159 | skb_frag_t *frag = shinfo->frags + i; | |
1160 | struct xen_netif_tx_request *txp; | |
ea066ad1 IC |
1161 | struct page *page; |
1162 | u16 pending_idx; | |
f942dc25 | 1163 | |
ea066ad1 | 1164 | pending_idx = frag_get_pending_idx(frag); |
f942dc25 | 1165 | |
f53c3fe8 | 1166 | /* If this is not the first frag, chain it to the previous*/ |
bdab8275 | 1167 | if (prev_pending_idx == INVALID_PENDING_IDX) |
f53c3fe8 | 1168 | skb_shinfo(skb)->destructor_arg = |
e9ce7cb6 | 1169 | &callback_param(queue, pending_idx); |
bdab8275 | 1170 | else |
e9ce7cb6 WL |
1171 | callback_param(queue, prev_pending_idx).ctx = |
1172 | &callback_param(queue, pending_idx); | |
f53c3fe8 | 1173 | |
e9ce7cb6 | 1174 | callback_param(queue, pending_idx).ctx = NULL; |
f53c3fe8 ZK |
1175 | prev_pending_idx = pending_idx; |
1176 | ||
e9ce7cb6 WL |
1177 | txp = &queue->pending_tx_info[pending_idx].req; |
1178 | page = virt_to_page(idx_to_kaddr(queue, pending_idx)); | |
ea066ad1 | 1179 | __skb_fill_page_desc(skb, i, page, txp->offset, txp->size); |
f942dc25 IC |
1180 | skb->len += txp->size; |
1181 | skb->data_len += txp->size; | |
1182 | skb->truesize += txp->size; | |
1183 | ||
f53c3fe8 | 1184 | /* Take an extra reference to offset network stack's put_page */ |
e9ce7cb6 | 1185 | get_page(queue->mmap_pages[pending_idx]); |
f942dc25 | 1186 | } |
f53c3fe8 ZK |
1187 | /* FIXME: __skb_fill_page_desc set this to true because page->pfmemalloc |
1188 | * overlaps with "index", and "mapping" is not set. I think mapping | |
1189 | * should be set. If delivered to local stack, it would drop this | |
1190 | * skb in sk_filter unless the socket has the right to use it. | |
1191 | */ | |
1192 | skb->pfmemalloc = false; | |
f942dc25 IC |
1193 | } |
1194 | ||
e9ce7cb6 | 1195 | static int xenvif_get_extras(struct xenvif_queue *queue, |
f942dc25 IC |
1196 | struct xen_netif_extra_info *extras, |
1197 | int work_to_do) | |
1198 | { | |
1199 | struct xen_netif_extra_info extra; | |
e9ce7cb6 | 1200 | RING_IDX cons = queue->tx.req_cons; |
f942dc25 IC |
1201 | |
1202 | do { | |
1203 | if (unlikely(work_to_do-- <= 0)) { | |
e9ce7cb6 WL |
1204 | netdev_err(queue->vif->dev, "Missing extra info\n"); |
1205 | xenvif_fatal_tx_err(queue->vif); | |
f942dc25 IC |
1206 | return -EBADR; |
1207 | } | |
1208 | ||
e9ce7cb6 | 1209 | memcpy(&extra, RING_GET_REQUEST(&queue->tx, cons), |
f942dc25 IC |
1210 | sizeof(extra)); |
1211 | if (unlikely(!extra.type || | |
1212 | extra.type >= XEN_NETIF_EXTRA_TYPE_MAX)) { | |
e9ce7cb6 WL |
1213 | queue->tx.req_cons = ++cons; |
1214 | netdev_err(queue->vif->dev, | |
f942dc25 | 1215 | "Invalid extra type: %d\n", extra.type); |
e9ce7cb6 | 1216 | xenvif_fatal_tx_err(queue->vif); |
f942dc25 IC |
1217 | return -EINVAL; |
1218 | } | |
1219 | ||
1220 | memcpy(&extras[extra.type - 1], &extra, sizeof(extra)); | |
e9ce7cb6 | 1221 | queue->tx.req_cons = ++cons; |
f942dc25 IC |
1222 | } while (extra.flags & XEN_NETIF_EXTRA_FLAG_MORE); |
1223 | ||
1224 | return work_to_do; | |
1225 | } | |
1226 | ||
7376419a WL |
1227 | static int xenvif_set_skb_gso(struct xenvif *vif, |
1228 | struct sk_buff *skb, | |
1229 | struct xen_netif_extra_info *gso) | |
f942dc25 IC |
1230 | { |
1231 | if (!gso->u.gso.size) { | |
48856286 | 1232 | netdev_err(vif->dev, "GSO size must not be zero.\n"); |
7376419a | 1233 | xenvif_fatal_tx_err(vif); |
f942dc25 IC |
1234 | return -EINVAL; |
1235 | } | |
1236 | ||
a9468587 PD |
1237 | switch (gso->u.gso.type) { |
1238 | case XEN_NETIF_GSO_TYPE_TCPV4: | |
1239 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4; | |
1240 | break; | |
1241 | case XEN_NETIF_GSO_TYPE_TCPV6: | |
1242 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6; | |
1243 | break; | |
1244 | default: | |
48856286 | 1245 | netdev_err(vif->dev, "Bad GSO type %d.\n", gso->u.gso.type); |
7376419a | 1246 | xenvif_fatal_tx_err(vif); |
f942dc25 IC |
1247 | return -EINVAL; |
1248 | } | |
1249 | ||
1250 | skb_shinfo(skb)->gso_size = gso->u.gso.size; | |
b89587a7 | 1251 | /* gso_segs will be calculated later */ |
f942dc25 IC |
1252 | |
1253 | return 0; | |
1254 | } | |
1255 | ||
e9ce7cb6 | 1256 | static int checksum_setup(struct xenvif_queue *queue, struct sk_buff *skb) |
2eba61d5 | 1257 | { |
2721637c | 1258 | bool recalculate_partial_csum = false; |
2eba61d5 PD |
1259 | |
1260 | /* A GSO SKB must be CHECKSUM_PARTIAL. However some buggy | |
1261 | * peers can fail to set NETRXF_csum_blank when sending a GSO | |
1262 | * frame. In this case force the SKB to CHECKSUM_PARTIAL and | |
1263 | * recalculate the partial checksum. | |
1264 | */ | |
1265 | if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) { | |
e9ce7cb6 | 1266 | queue->stats.rx_gso_checksum_fixup++; |
2eba61d5 | 1267 | skb->ip_summed = CHECKSUM_PARTIAL; |
2721637c | 1268 | recalculate_partial_csum = true; |
2eba61d5 PD |
1269 | } |
1270 | ||
1271 | /* A non-CHECKSUM_PARTIAL SKB does not require setup. */ | |
1272 | if (skb->ip_summed != CHECKSUM_PARTIAL) | |
1273 | return 0; | |
1274 | ||
2721637c | 1275 | return skb_checksum_setup(skb, recalculate_partial_csum); |
2eba61d5 PD |
1276 | } |
1277 | ||
e9ce7cb6 | 1278 | static bool tx_credit_exceeded(struct xenvif_queue *queue, unsigned size) |
f942dc25 | 1279 | { |
059dfa6a | 1280 | u64 now = get_jiffies_64(); |
e9ce7cb6 WL |
1281 | u64 next_credit = queue->credit_window_start + |
1282 | msecs_to_jiffies(queue->credit_usec / 1000); | |
f942dc25 IC |
1283 | |
1284 | /* Timer could already be pending in rare cases. */ | |
e9ce7cb6 | 1285 | if (timer_pending(&queue->credit_timeout)) |
f942dc25 IC |
1286 | return true; |
1287 | ||
1288 | /* Passed the point where we can replenish credit? */ | |
059dfa6a | 1289 | if (time_after_eq64(now, next_credit)) { |
e9ce7cb6 WL |
1290 | queue->credit_window_start = now; |
1291 | tx_add_credit(queue); | |
f942dc25 IC |
1292 | } |
1293 | ||
1294 | /* Still too big to send right now? Set a callback. */ | |
e9ce7cb6 WL |
1295 | if (size > queue->remaining_credit) { |
1296 | queue->credit_timeout.data = | |
1297 | (unsigned long)queue; | |
1298 | queue->credit_timeout.function = | |
f942dc25 | 1299 | tx_credit_callback; |
e9ce7cb6 | 1300 | mod_timer(&queue->credit_timeout, |
f942dc25 | 1301 | next_credit); |
e9ce7cb6 | 1302 | queue->credit_window_start = next_credit; |
f942dc25 IC |
1303 | |
1304 | return true; | |
1305 | } | |
1306 | ||
1307 | return false; | |
1308 | } | |
1309 | ||
e9ce7cb6 | 1310 | static void xenvif_tx_build_gops(struct xenvif_queue *queue, |
bdab8275 ZK |
1311 | int budget, |
1312 | unsigned *copy_ops, | |
1313 | unsigned *map_ops) | |
f942dc25 | 1314 | { |
e9ce7cb6 | 1315 | struct gnttab_map_grant_ref *gop = queue->tx_map_ops, *request_gop; |
f942dc25 IC |
1316 | struct sk_buff *skb; |
1317 | int ret; | |
1318 | ||
e9ce7cb6 | 1319 | while (skb_queue_len(&queue->tx_queue) < budget) { |
f942dc25 | 1320 | struct xen_netif_tx_request txreq; |
37641494 | 1321 | struct xen_netif_tx_request txfrags[XEN_NETBK_LEGACY_SLOTS_MAX]; |
f942dc25 IC |
1322 | struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX-1]; |
1323 | u16 pending_idx; | |
1324 | RING_IDX idx; | |
1325 | int work_to_do; | |
1326 | unsigned int data_len; | |
1327 | pending_ring_idx_t index; | |
1328 | ||
e9ce7cb6 | 1329 | if (queue->tx.sring->req_prod - queue->tx.req_cons > |
48856286 | 1330 | XEN_NETIF_TX_RING_SIZE) { |
e9ce7cb6 | 1331 | netdev_err(queue->vif->dev, |
48856286 IC |
1332 | "Impossible number of requests. " |
1333 | "req_prod %d, req_cons %d, size %ld\n", | |
e9ce7cb6 | 1334 | queue->tx.sring->req_prod, queue->tx.req_cons, |
48856286 | 1335 | XEN_NETIF_TX_RING_SIZE); |
e9ce7cb6 | 1336 | xenvif_fatal_tx_err(queue->vif); |
e9d8b2c2 | 1337 | break; |
48856286 IC |
1338 | } |
1339 | ||
e9ce7cb6 | 1340 | work_to_do = RING_HAS_UNCONSUMED_REQUESTS(&queue->tx); |
b3f980bd WL |
1341 | if (!work_to_do) |
1342 | break; | |
f942dc25 | 1343 | |
e9ce7cb6 | 1344 | idx = queue->tx.req_cons; |
f942dc25 | 1345 | rmb(); /* Ensure that we see the request before we copy it. */ |
e9ce7cb6 | 1346 | memcpy(&txreq, RING_GET_REQUEST(&queue->tx, idx), sizeof(txreq)); |
f942dc25 IC |
1347 | |
1348 | /* Credit-based scheduling. */ | |
e9ce7cb6 WL |
1349 | if (txreq.size > queue->remaining_credit && |
1350 | tx_credit_exceeded(queue, txreq.size)) | |
b3f980bd | 1351 | break; |
f942dc25 | 1352 | |
e9ce7cb6 | 1353 | queue->remaining_credit -= txreq.size; |
f942dc25 IC |
1354 | |
1355 | work_to_do--; | |
e9ce7cb6 | 1356 | queue->tx.req_cons = ++idx; |
f942dc25 IC |
1357 | |
1358 | memset(extras, 0, sizeof(extras)); | |
1359 | if (txreq.flags & XEN_NETTXF_extra_info) { | |
e9ce7cb6 | 1360 | work_to_do = xenvif_get_extras(queue, extras, |
7376419a | 1361 | work_to_do); |
e9ce7cb6 | 1362 | idx = queue->tx.req_cons; |
48856286 | 1363 | if (unlikely(work_to_do < 0)) |
b3f980bd | 1364 | break; |
f942dc25 IC |
1365 | } |
1366 | ||
e9ce7cb6 | 1367 | ret = xenvif_count_requests(queue, &txreq, txfrags, work_to_do); |
48856286 | 1368 | if (unlikely(ret < 0)) |
b3f980bd | 1369 | break; |
48856286 | 1370 | |
f942dc25 IC |
1371 | idx += ret; |
1372 | ||
1373 | if (unlikely(txreq.size < ETH_HLEN)) { | |
e9ce7cb6 | 1374 | netdev_dbg(queue->vif->dev, |
f942dc25 | 1375 | "Bad packet size: %d\n", txreq.size); |
e9ce7cb6 | 1376 | xenvif_tx_err(queue, &txreq, idx); |
b3f980bd | 1377 | break; |
f942dc25 IC |
1378 | } |
1379 | ||
1380 | /* No crossing a page as the payload mustn't fragment. */ | |
1381 | if (unlikely((txreq.offset + txreq.size) > PAGE_SIZE)) { | |
e9ce7cb6 | 1382 | netdev_err(queue->vif->dev, |
f942dc25 IC |
1383 | "txreq.offset: %x, size: %u, end: %lu\n", |
1384 | txreq.offset, txreq.size, | |
1385 | (txreq.offset&~PAGE_MASK) + txreq.size); | |
e9ce7cb6 | 1386 | xenvif_fatal_tx_err(queue->vif); |
b3f980bd | 1387 | break; |
f942dc25 IC |
1388 | } |
1389 | ||
e9ce7cb6 WL |
1390 | index = pending_index(queue->pending_cons); |
1391 | pending_idx = queue->pending_ring[index]; | |
f942dc25 IC |
1392 | |
1393 | data_len = (txreq.size > PKT_PROT_LEN && | |
37641494 | 1394 | ret < XEN_NETBK_LEGACY_SLOTS_MAX) ? |
f942dc25 IC |
1395 | PKT_PROT_LEN : txreq.size; |
1396 | ||
e3377f36 | 1397 | skb = xenvif_alloc_skb(data_len); |
f942dc25 | 1398 | if (unlikely(skb == NULL)) { |
e9ce7cb6 | 1399 | netdev_dbg(queue->vif->dev, |
f942dc25 | 1400 | "Can't allocate a skb in start_xmit.\n"); |
e9ce7cb6 | 1401 | xenvif_tx_err(queue, &txreq, idx); |
f942dc25 IC |
1402 | break; |
1403 | } | |
1404 | ||
f942dc25 IC |
1405 | if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) { |
1406 | struct xen_netif_extra_info *gso; | |
1407 | gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1]; | |
1408 | ||
e9ce7cb6 | 1409 | if (xenvif_set_skb_gso(queue->vif, skb, gso)) { |
7376419a | 1410 | /* Failure in xenvif_set_skb_gso is fatal. */ |
f942dc25 | 1411 | kfree_skb(skb); |
b3f980bd | 1412 | break; |
f942dc25 IC |
1413 | } |
1414 | } | |
1415 | ||
8f13dd96 | 1416 | XENVIF_TX_CB(skb)->pending_idx = pending_idx; |
f942dc25 IC |
1417 | |
1418 | __skb_put(skb, data_len); | |
e9ce7cb6 WL |
1419 | queue->tx_copy_ops[*copy_ops].source.u.ref = txreq.gref; |
1420 | queue->tx_copy_ops[*copy_ops].source.domid = queue->vif->domid; | |
1421 | queue->tx_copy_ops[*copy_ops].source.offset = txreq.offset; | |
bdab8275 | 1422 | |
e9ce7cb6 | 1423 | queue->tx_copy_ops[*copy_ops].dest.u.gmfn = |
bdab8275 | 1424 | virt_to_mfn(skb->data); |
e9ce7cb6 WL |
1425 | queue->tx_copy_ops[*copy_ops].dest.domid = DOMID_SELF; |
1426 | queue->tx_copy_ops[*copy_ops].dest.offset = | |
bdab8275 ZK |
1427 | offset_in_page(skb->data); |
1428 | ||
e9ce7cb6 WL |
1429 | queue->tx_copy_ops[*copy_ops].len = data_len; |
1430 | queue->tx_copy_ops[*copy_ops].flags = GNTCOPY_source_gref; | |
bdab8275 ZK |
1431 | |
1432 | (*copy_ops)++; | |
f942dc25 IC |
1433 | |
1434 | skb_shinfo(skb)->nr_frags = ret; | |
1435 | if (data_len < txreq.size) { | |
1436 | skb_shinfo(skb)->nr_frags++; | |
ea066ad1 IC |
1437 | frag_set_pending_idx(&skb_shinfo(skb)->frags[0], |
1438 | pending_idx); | |
e9ce7cb6 | 1439 | xenvif_tx_create_map_op(queue, pending_idx, &txreq, gop); |
bdab8275 | 1440 | gop++; |
f942dc25 | 1441 | } else { |
ea066ad1 IC |
1442 | frag_set_pending_idx(&skb_shinfo(skb)->frags[0], |
1443 | INVALID_PENDING_IDX); | |
e9ce7cb6 | 1444 | memcpy(&queue->pending_tx_info[pending_idx].req, &txreq, |
bdab8275 | 1445 | sizeof(txreq)); |
f942dc25 IC |
1446 | } |
1447 | ||
e9ce7cb6 | 1448 | queue->pending_cons++; |
f942dc25 | 1449 | |
e9ce7cb6 | 1450 | request_gop = xenvif_get_requests(queue, skb, txfrags, gop); |
f942dc25 IC |
1451 | if (request_gop == NULL) { |
1452 | kfree_skb(skb); | |
e9ce7cb6 | 1453 | xenvif_tx_err(queue, &txreq, idx); |
b3f980bd | 1454 | break; |
f942dc25 IC |
1455 | } |
1456 | gop = request_gop; | |
1457 | ||
e9ce7cb6 | 1458 | __skb_queue_tail(&queue->tx_queue, skb); |
1e0b6eac | 1459 | |
e9ce7cb6 | 1460 | queue->tx.req_cons = idx; |
f942dc25 | 1461 | |
e9ce7cb6 WL |
1462 | if (((gop-queue->tx_map_ops) >= ARRAY_SIZE(queue->tx_map_ops)) || |
1463 | (*copy_ops >= ARRAY_SIZE(queue->tx_copy_ops))) | |
f942dc25 IC |
1464 | break; |
1465 | } | |
1466 | ||
e9ce7cb6 | 1467 | (*map_ops) = gop - queue->tx_map_ops; |
bdab8275 | 1468 | return; |
f942dc25 IC |
1469 | } |
1470 | ||
e3377f36 ZK |
1471 | /* Consolidate skb with a frag_list into a brand new one with local pages on |
1472 | * frags. Returns 0 or -ENOMEM if can't allocate new pages. | |
1473 | */ | |
e9ce7cb6 | 1474 | static int xenvif_handle_frag_list(struct xenvif_queue *queue, struct sk_buff *skb) |
e3377f36 ZK |
1475 | { |
1476 | unsigned int offset = skb_headlen(skb); | |
1477 | skb_frag_t frags[MAX_SKB_FRAGS]; | |
1478 | int i; | |
1479 | struct ubuf_info *uarg; | |
1480 | struct sk_buff *nskb = skb_shinfo(skb)->frag_list; | |
1481 | ||
e9ce7cb6 WL |
1482 | queue->stats.tx_zerocopy_sent += 2; |
1483 | queue->stats.tx_frag_overflow++; | |
e3377f36 | 1484 | |
e9ce7cb6 | 1485 | xenvif_fill_frags(queue, nskb); |
e3377f36 ZK |
1486 | /* Subtract frags size, we will correct it later */ |
1487 | skb->truesize -= skb->data_len; | |
1488 | skb->len += nskb->len; | |
1489 | skb->data_len += nskb->len; | |
1490 | ||
1491 | /* create a brand new frags array and coalesce there */ | |
1492 | for (i = 0; offset < skb->len; i++) { | |
1493 | struct page *page; | |
1494 | unsigned int len; | |
1495 | ||
1496 | BUG_ON(i >= MAX_SKB_FRAGS); | |
1497 | page = alloc_page(GFP_ATOMIC|__GFP_COLD); | |
1498 | if (!page) { | |
1499 | int j; | |
1500 | skb->truesize += skb->data_len; | |
1501 | for (j = 0; j < i; j++) | |
1502 | put_page(frags[j].page.p); | |
1503 | return -ENOMEM; | |
1504 | } | |
1505 | ||
1506 | if (offset + PAGE_SIZE < skb->len) | |
1507 | len = PAGE_SIZE; | |
1508 | else | |
1509 | len = skb->len - offset; | |
1510 | if (skb_copy_bits(skb, offset, page_address(page), len)) | |
1511 | BUG(); | |
1512 | ||
1513 | offset += len; | |
1514 | frags[i].page.p = page; | |
1515 | frags[i].page_offset = 0; | |
1516 | skb_frag_size_set(&frags[i], len); | |
1517 | } | |
1518 | /* swap out with old one */ | |
1519 | memcpy(skb_shinfo(skb)->frags, | |
1520 | frags, | |
1521 | i * sizeof(skb_frag_t)); | |
1522 | skb_shinfo(skb)->nr_frags = i; | |
1523 | skb->truesize += i * PAGE_SIZE; | |
1524 | ||
1525 | /* remove traces of mapped pages and frag_list */ | |
1526 | skb_frag_list_init(skb); | |
1527 | uarg = skb_shinfo(skb)->destructor_arg; | |
1528 | uarg->callback(uarg, true); | |
1529 | skb_shinfo(skb)->destructor_arg = NULL; | |
1530 | ||
1531 | skb_shinfo(nskb)->tx_flags |= SKBTX_DEV_ZEROCOPY; | |
1532 | kfree_skb(nskb); | |
1533 | ||
1534 | return 0; | |
1535 | } | |
b3f980bd | 1536 | |
e9ce7cb6 | 1537 | static int xenvif_tx_submit(struct xenvif_queue *queue) |
f942dc25 | 1538 | { |
e9ce7cb6 WL |
1539 | struct gnttab_map_grant_ref *gop_map = queue->tx_map_ops; |
1540 | struct gnttab_copy *gop_copy = queue->tx_copy_ops; | |
f942dc25 | 1541 | struct sk_buff *skb; |
b3f980bd | 1542 | int work_done = 0; |
f942dc25 | 1543 | |
e9ce7cb6 | 1544 | while ((skb = __skb_dequeue(&queue->tx_queue)) != NULL) { |
f942dc25 | 1545 | struct xen_netif_tx_request *txp; |
f942dc25 IC |
1546 | u16 pending_idx; |
1547 | unsigned data_len; | |
1548 | ||
8f13dd96 | 1549 | pending_idx = XENVIF_TX_CB(skb)->pending_idx; |
e9ce7cb6 | 1550 | txp = &queue->pending_tx_info[pending_idx].req; |
f942dc25 IC |
1551 | |
1552 | /* Check the remap error code. */ | |
e9ce7cb6 | 1553 | if (unlikely(xenvif_tx_check_gop(queue, skb, &gop_map, &gop_copy))) { |
b42cc6e4 ZK |
1554 | /* If there was an error, xenvif_tx_check_gop is |
1555 | * expected to release all the frags which were mapped, | |
1556 | * so kfree_skb shouldn't do it again | |
1557 | */ | |
f942dc25 | 1558 | skb_shinfo(skb)->nr_frags = 0; |
b42cc6e4 ZK |
1559 | if (skb_has_frag_list(skb)) { |
1560 | struct sk_buff *nskb = | |
1561 | skb_shinfo(skb)->frag_list; | |
1562 | skb_shinfo(nskb)->nr_frags = 0; | |
1563 | } | |
f942dc25 IC |
1564 | kfree_skb(skb); |
1565 | continue; | |
1566 | } | |
1567 | ||
1568 | data_len = skb->len; | |
e9ce7cb6 | 1569 | callback_param(queue, pending_idx).ctx = NULL; |
f942dc25 IC |
1570 | if (data_len < txp->size) { |
1571 | /* Append the packet payload as a fragment. */ | |
1572 | txp->offset += data_len; | |
1573 | txp->size -= data_len; | |
1574 | } else { | |
1575 | /* Schedule a response immediately. */ | |
e9ce7cb6 | 1576 | xenvif_idx_release(queue, pending_idx, |
bdab8275 | 1577 | XEN_NETIF_RSP_OKAY); |
f942dc25 IC |
1578 | } |
1579 | ||
1580 | if (txp->flags & XEN_NETTXF_csum_blank) | |
1581 | skb->ip_summed = CHECKSUM_PARTIAL; | |
1582 | else if (txp->flags & XEN_NETTXF_data_validated) | |
1583 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
1584 | ||
e9ce7cb6 | 1585 | xenvif_fill_frags(queue, skb); |
f942dc25 | 1586 | |
e3377f36 | 1587 | if (unlikely(skb_has_frag_list(skb))) { |
e9ce7cb6 | 1588 | if (xenvif_handle_frag_list(queue, skb)) { |
e3377f36 | 1589 | if (net_ratelimit()) |
e9ce7cb6 | 1590 | netdev_err(queue->vif->dev, |
e3377f36 ZK |
1591 | "Not enough memory to consolidate frag_list!\n"); |
1592 | skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY; | |
1593 | kfree_skb(skb); | |
1594 | continue; | |
1595 | } | |
1596 | } | |
1597 | ||
2eba61d5 | 1598 | if (skb_is_nonlinear(skb) && skb_headlen(skb) < PKT_PROT_LEN) { |
f942dc25 IC |
1599 | int target = min_t(int, skb->len, PKT_PROT_LEN); |
1600 | __pskb_pull_tail(skb, target - skb_headlen(skb)); | |
1601 | } | |
1602 | ||
e9ce7cb6 | 1603 | skb->dev = queue->vif->dev; |
f942dc25 | 1604 | skb->protocol = eth_type_trans(skb, skb->dev); |
f9ca8f74 | 1605 | skb_reset_network_header(skb); |
f942dc25 | 1606 | |
e9ce7cb6 WL |
1607 | if (checksum_setup(queue, skb)) { |
1608 | netdev_dbg(queue->vif->dev, | |
f942dc25 | 1609 | "Can't setup checksum in net_tx_action\n"); |
f53c3fe8 ZK |
1610 | /* We have to set this flag to trigger the callback */ |
1611 | if (skb_shinfo(skb)->destructor_arg) | |
1612 | skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY; | |
f942dc25 IC |
1613 | kfree_skb(skb); |
1614 | continue; | |
1615 | } | |
1616 | ||
40893fd0 | 1617 | skb_probe_transport_header(skb, 0); |
f9ca8f74 | 1618 | |
b89587a7 PD |
1619 | /* If the packet is GSO then we will have just set up the |
1620 | * transport header offset in checksum_setup so it's now | |
1621 | * straightforward to calculate gso_segs. | |
1622 | */ | |
1623 | if (skb_is_gso(skb)) { | |
1624 | int mss = skb_shinfo(skb)->gso_size; | |
1625 | int hdrlen = skb_transport_header(skb) - | |
1626 | skb_mac_header(skb) + | |
1627 | tcp_hdrlen(skb); | |
1628 | ||
1629 | skb_shinfo(skb)->gso_segs = | |
1630 | DIV_ROUND_UP(skb->len - hdrlen, mss); | |
1631 | } | |
1632 | ||
e9ce7cb6 WL |
1633 | queue->stats.rx_bytes += skb->len; |
1634 | queue->stats.rx_packets++; | |
f942dc25 | 1635 | |
b3f980bd WL |
1636 | work_done++; |
1637 | ||
f53c3fe8 ZK |
1638 | /* Set this flag right before netif_receive_skb, otherwise |
1639 | * someone might think this packet already left netback, and | |
1640 | * do a skb_copy_ubufs while we are still in control of the | |
1641 | * skb. E.g. the __pskb_pull_tail earlier can do such thing. | |
1642 | */ | |
1bb332af | 1643 | if (skb_shinfo(skb)->destructor_arg) { |
f53c3fe8 | 1644 | skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY; |
e9ce7cb6 | 1645 | queue->stats.tx_zerocopy_sent++; |
1bb332af | 1646 | } |
f53c3fe8 | 1647 | |
b3f980bd | 1648 | netif_receive_skb(skb); |
f942dc25 | 1649 | } |
b3f980bd WL |
1650 | |
1651 | return work_done; | |
f942dc25 IC |
1652 | } |
1653 | ||
3e2234b3 ZK |
1654 | void xenvif_zerocopy_callback(struct ubuf_info *ubuf, bool zerocopy_success) |
1655 | { | |
f53c3fe8 ZK |
1656 | unsigned long flags; |
1657 | pending_ring_idx_t index; | |
e9ce7cb6 | 1658 | struct xenvif_queue *queue = ubuf_to_queue(ubuf); |
f53c3fe8 ZK |
1659 | |
1660 | /* This is the only place where we grab this lock, to protect callbacks | |
1661 | * from each other. | |
1662 | */ | |
e9ce7cb6 | 1663 | spin_lock_irqsave(&queue->callback_lock, flags); |
f53c3fe8 ZK |
1664 | do { |
1665 | u16 pending_idx = ubuf->desc; | |
1666 | ubuf = (struct ubuf_info *) ubuf->ctx; | |
e9ce7cb6 | 1667 | BUG_ON(queue->dealloc_prod - queue->dealloc_cons >= |
f53c3fe8 | 1668 | MAX_PENDING_REQS); |
e9ce7cb6 WL |
1669 | index = pending_index(queue->dealloc_prod); |
1670 | queue->dealloc_ring[index] = pending_idx; | |
f53c3fe8 ZK |
1671 | /* Sync with xenvif_tx_dealloc_action: |
1672 | * insert idx then incr producer. | |
1673 | */ | |
1674 | smp_wmb(); | |
e9ce7cb6 | 1675 | queue->dealloc_prod++; |
f53c3fe8 | 1676 | } while (ubuf); |
e9ce7cb6 WL |
1677 | wake_up(&queue->dealloc_wq); |
1678 | spin_unlock_irqrestore(&queue->callback_lock, flags); | |
f53c3fe8 | 1679 | |
1bb332af | 1680 | if (likely(zerocopy_success)) |
e9ce7cb6 | 1681 | queue->stats.tx_zerocopy_success++; |
1bb332af | 1682 | else |
e9ce7cb6 | 1683 | queue->stats.tx_zerocopy_fail++; |
f53c3fe8 ZK |
1684 | } |
1685 | ||
e9ce7cb6 | 1686 | static inline void xenvif_tx_dealloc_action(struct xenvif_queue *queue) |
f53c3fe8 ZK |
1687 | { |
1688 | struct gnttab_unmap_grant_ref *gop; | |
1689 | pending_ring_idx_t dc, dp; | |
1690 | u16 pending_idx, pending_idx_release[MAX_PENDING_REQS]; | |
1691 | unsigned int i = 0; | |
1692 | ||
e9ce7cb6 WL |
1693 | dc = queue->dealloc_cons; |
1694 | gop = queue->tx_unmap_ops; | |
f53c3fe8 ZK |
1695 | |
1696 | /* Free up any grants we have finished using */ | |
1697 | do { | |
e9ce7cb6 | 1698 | dp = queue->dealloc_prod; |
f53c3fe8 ZK |
1699 | |
1700 | /* Ensure we see all indices enqueued by all | |
1701 | * xenvif_zerocopy_callback(). | |
1702 | */ | |
1703 | smp_rmb(); | |
1704 | ||
1705 | while (dc != dp) { | |
e9ce7cb6 | 1706 | BUG_ON(gop - queue->tx_unmap_ops > MAX_PENDING_REQS); |
f53c3fe8 | 1707 | pending_idx = |
e9ce7cb6 | 1708 | queue->dealloc_ring[pending_index(dc++)]; |
f53c3fe8 | 1709 | |
e9ce7cb6 | 1710 | pending_idx_release[gop-queue->tx_unmap_ops] = |
f53c3fe8 | 1711 | pending_idx; |
e9ce7cb6 WL |
1712 | queue->pages_to_unmap[gop-queue->tx_unmap_ops] = |
1713 | queue->mmap_pages[pending_idx]; | |
f53c3fe8 | 1714 | gnttab_set_unmap_op(gop, |
e9ce7cb6 | 1715 | idx_to_kaddr(queue, pending_idx), |
f53c3fe8 | 1716 | GNTMAP_host_map, |
e9ce7cb6 WL |
1717 | queue->grant_tx_handle[pending_idx]); |
1718 | xenvif_grant_handle_reset(queue, pending_idx); | |
f53c3fe8 ZK |
1719 | ++gop; |
1720 | } | |
1721 | ||
e9ce7cb6 | 1722 | } while (dp != queue->dealloc_prod); |
f53c3fe8 | 1723 | |
e9ce7cb6 | 1724 | queue->dealloc_cons = dc; |
f53c3fe8 | 1725 | |
e9ce7cb6 | 1726 | if (gop - queue->tx_unmap_ops > 0) { |
f53c3fe8 | 1727 | int ret; |
e9ce7cb6 | 1728 | ret = gnttab_unmap_refs(queue->tx_unmap_ops, |
f53c3fe8 | 1729 | NULL, |
e9ce7cb6 WL |
1730 | queue->pages_to_unmap, |
1731 | gop - queue->tx_unmap_ops); | |
f53c3fe8 | 1732 | if (ret) { |
e9ce7cb6 WL |
1733 | netdev_err(queue->vif->dev, "Unmap fail: nr_ops %tx ret %d\n", |
1734 | gop - queue->tx_unmap_ops, ret); | |
1735 | for (i = 0; i < gop - queue->tx_unmap_ops; ++i) { | |
f53c3fe8 | 1736 | if (gop[i].status != GNTST_okay) |
e9ce7cb6 | 1737 | netdev_err(queue->vif->dev, |
f53c3fe8 ZK |
1738 | " host_addr: %llx handle: %x status: %d\n", |
1739 | gop[i].host_addr, | |
1740 | gop[i].handle, | |
1741 | gop[i].status); | |
1742 | } | |
1743 | BUG(); | |
1744 | } | |
1745 | } | |
1746 | ||
e9ce7cb6 WL |
1747 | for (i = 0; i < gop - queue->tx_unmap_ops; ++i) |
1748 | xenvif_idx_release(queue, pending_idx_release[i], | |
f53c3fe8 | 1749 | XEN_NETIF_RSP_OKAY); |
3e2234b3 ZK |
1750 | } |
1751 | ||
f53c3fe8 | 1752 | |
f942dc25 | 1753 | /* Called after netfront has transmitted */ |
e9ce7cb6 | 1754 | int xenvif_tx_action(struct xenvif_queue *queue, int budget) |
f942dc25 | 1755 | { |
bdab8275 | 1756 | unsigned nr_mops, nr_cops = 0; |
f53c3fe8 | 1757 | int work_done, ret; |
f942dc25 | 1758 | |
e9ce7cb6 | 1759 | if (unlikely(!tx_work_todo(queue))) |
b3f980bd WL |
1760 | return 0; |
1761 | ||
e9ce7cb6 | 1762 | xenvif_tx_build_gops(queue, budget, &nr_cops, &nr_mops); |
f942dc25 | 1763 | |
bdab8275 | 1764 | if (nr_cops == 0) |
b3f980bd WL |
1765 | return 0; |
1766 | ||
e9ce7cb6 | 1767 | gnttab_batch_copy(queue->tx_copy_ops, nr_cops); |
bdab8275 | 1768 | if (nr_mops != 0) { |
e9ce7cb6 | 1769 | ret = gnttab_map_refs(queue->tx_map_ops, |
bdab8275 | 1770 | NULL, |
e9ce7cb6 | 1771 | queue->pages_to_map, |
bdab8275 ZK |
1772 | nr_mops); |
1773 | BUG_ON(ret); | |
1774 | } | |
f942dc25 | 1775 | |
e9ce7cb6 | 1776 | work_done = xenvif_tx_submit(queue); |
f942dc25 | 1777 | |
b3f980bd | 1778 | return work_done; |
f942dc25 IC |
1779 | } |
1780 | ||
e9ce7cb6 | 1781 | static void xenvif_idx_release(struct xenvif_queue *queue, u16 pending_idx, |
7376419a | 1782 | u8 status) |
f942dc25 | 1783 | { |
f942dc25 | 1784 | struct pending_tx_info *pending_tx_info; |
f53c3fe8 | 1785 | pending_ring_idx_t index; |
f53c3fe8 | 1786 | unsigned long flags; |
2810e5b9 | 1787 | |
e9ce7cb6 WL |
1788 | pending_tx_info = &queue->pending_tx_info[pending_idx]; |
1789 | spin_lock_irqsave(&queue->response_lock, flags); | |
1790 | make_tx_response(queue, &pending_tx_info->req, status); | |
1791 | index = pending_index(queue->pending_prod); | |
1792 | queue->pending_ring[index] = pending_idx; | |
62bad319 ZK |
1793 | /* TX shouldn't use the index before we give it back here */ |
1794 | mb(); | |
e9ce7cb6 WL |
1795 | queue->pending_prod++; |
1796 | spin_unlock_irqrestore(&queue->response_lock, flags); | |
f942dc25 IC |
1797 | } |
1798 | ||
2810e5b9 | 1799 | |
e9ce7cb6 | 1800 | static void make_tx_response(struct xenvif_queue *queue, |
f942dc25 IC |
1801 | struct xen_netif_tx_request *txp, |
1802 | s8 st) | |
1803 | { | |
e9ce7cb6 | 1804 | RING_IDX i = queue->tx.rsp_prod_pvt; |
f942dc25 IC |
1805 | struct xen_netif_tx_response *resp; |
1806 | int notify; | |
1807 | ||
e9ce7cb6 | 1808 | resp = RING_GET_RESPONSE(&queue->tx, i); |
f942dc25 IC |
1809 | resp->id = txp->id; |
1810 | resp->status = st; | |
1811 | ||
1812 | if (txp->flags & XEN_NETTXF_extra_info) | |
e9ce7cb6 | 1813 | RING_GET_RESPONSE(&queue->tx, ++i)->status = XEN_NETIF_RSP_NULL; |
f942dc25 | 1814 | |
e9ce7cb6 WL |
1815 | queue->tx.rsp_prod_pvt = ++i; |
1816 | RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&queue->tx, notify); | |
f942dc25 | 1817 | if (notify) |
e9ce7cb6 | 1818 | notify_remote_via_irq(queue->tx_irq); |
f942dc25 IC |
1819 | } |
1820 | ||
e9ce7cb6 | 1821 | static struct xen_netif_rx_response *make_rx_response(struct xenvif_queue *queue, |
f942dc25 IC |
1822 | u16 id, |
1823 | s8 st, | |
1824 | u16 offset, | |
1825 | u16 size, | |
1826 | u16 flags) | |
1827 | { | |
e9ce7cb6 | 1828 | RING_IDX i = queue->rx.rsp_prod_pvt; |
f942dc25 IC |
1829 | struct xen_netif_rx_response *resp; |
1830 | ||
e9ce7cb6 | 1831 | resp = RING_GET_RESPONSE(&queue->rx, i); |
f942dc25 IC |
1832 | resp->offset = offset; |
1833 | resp->flags = flags; | |
1834 | resp->id = id; | |
1835 | resp->status = (s16)size; | |
1836 | if (st < 0) | |
1837 | resp->status = (s16)st; | |
1838 | ||
e9ce7cb6 | 1839 | queue->rx.rsp_prod_pvt = ++i; |
f942dc25 IC |
1840 | |
1841 | return resp; | |
1842 | } | |
1843 | ||
e9ce7cb6 | 1844 | void xenvif_idx_unmap(struct xenvif_queue *queue, u16 pending_idx) |
f53c3fe8 ZK |
1845 | { |
1846 | int ret; | |
1847 | struct gnttab_unmap_grant_ref tx_unmap_op; | |
1848 | ||
1849 | gnttab_set_unmap_op(&tx_unmap_op, | |
e9ce7cb6 | 1850 | idx_to_kaddr(queue, pending_idx), |
f53c3fe8 | 1851 | GNTMAP_host_map, |
e9ce7cb6 WL |
1852 | queue->grant_tx_handle[pending_idx]); |
1853 | xenvif_grant_handle_reset(queue, pending_idx); | |
f53c3fe8 ZK |
1854 | |
1855 | ret = gnttab_unmap_refs(&tx_unmap_op, NULL, | |
e9ce7cb6 | 1856 | &queue->mmap_pages[pending_idx], 1); |
7aceb47a | 1857 | if (ret) { |
e9ce7cb6 | 1858 | netdev_err(queue->vif->dev, |
7aceb47a ZK |
1859 | "Unmap fail: ret: %d pending_idx: %d host_addr: %llx handle: %x status: %d\n", |
1860 | ret, | |
1861 | pending_idx, | |
1862 | tx_unmap_op.host_addr, | |
1863 | tx_unmap_op.handle, | |
1864 | tx_unmap_op.status); | |
1865 | BUG(); | |
1866 | } | |
f53c3fe8 ZK |
1867 | } |
1868 | ||
e9ce7cb6 | 1869 | static inline int rx_work_todo(struct xenvif_queue *queue) |
f942dc25 | 1870 | { |
e9ce7cb6 WL |
1871 | return (!skb_queue_empty(&queue->rx_queue) && |
1872 | xenvif_rx_ring_slots_available(queue, queue->rx_last_skb_slots)) || | |
1873 | queue->rx_queue_purge; | |
f942dc25 IC |
1874 | } |
1875 | ||
e9ce7cb6 | 1876 | static inline int tx_work_todo(struct xenvif_queue *queue) |
f942dc25 | 1877 | { |
e9ce7cb6 | 1878 | if (likely(RING_HAS_UNCONSUMED_REQUESTS(&queue->tx))) |
f942dc25 IC |
1879 | return 1; |
1880 | ||
1881 | return 0; | |
1882 | } | |
1883 | ||
e9ce7cb6 | 1884 | static inline bool tx_dealloc_work_todo(struct xenvif_queue *queue) |
f53c3fe8 | 1885 | { |
e9ce7cb6 | 1886 | return queue->dealloc_cons != queue->dealloc_prod; |
f53c3fe8 ZK |
1887 | } |
1888 | ||
e9ce7cb6 | 1889 | void xenvif_unmap_frontend_rings(struct xenvif_queue *queue) |
f942dc25 | 1890 | { |
e9ce7cb6 WL |
1891 | if (queue->tx.sring) |
1892 | xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(queue->vif), | |
1893 | queue->tx.sring); | |
1894 | if (queue->rx.sring) | |
1895 | xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(queue->vif), | |
1896 | queue->rx.sring); | |
f942dc25 IC |
1897 | } |
1898 | ||
e9ce7cb6 | 1899 | int xenvif_map_frontend_rings(struct xenvif_queue *queue, |
7376419a WL |
1900 | grant_ref_t tx_ring_ref, |
1901 | grant_ref_t rx_ring_ref) | |
f942dc25 | 1902 | { |
c9d63699 | 1903 | void *addr; |
f942dc25 IC |
1904 | struct xen_netif_tx_sring *txs; |
1905 | struct xen_netif_rx_sring *rxs; | |
1906 | ||
1907 | int err = -ENOMEM; | |
1908 | ||
e9ce7cb6 | 1909 | err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(queue->vif), |
c9d63699 DV |
1910 | tx_ring_ref, &addr); |
1911 | if (err) | |
f942dc25 IC |
1912 | goto err; |
1913 | ||
c9d63699 | 1914 | txs = (struct xen_netif_tx_sring *)addr; |
e9ce7cb6 | 1915 | BACK_RING_INIT(&queue->tx, txs, PAGE_SIZE); |
f942dc25 | 1916 | |
e9ce7cb6 | 1917 | err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(queue->vif), |
c9d63699 DV |
1918 | rx_ring_ref, &addr); |
1919 | if (err) | |
f942dc25 | 1920 | goto err; |
f942dc25 | 1921 | |
c9d63699 | 1922 | rxs = (struct xen_netif_rx_sring *)addr; |
e9ce7cb6 | 1923 | BACK_RING_INIT(&queue->rx, rxs, PAGE_SIZE); |
f942dc25 IC |
1924 | |
1925 | return 0; | |
1926 | ||
1927 | err: | |
e9ce7cb6 | 1928 | xenvif_unmap_frontend_rings(queue); |
f942dc25 IC |
1929 | return err; |
1930 | } | |
1931 | ||
e9ce7cb6 | 1932 | static void xenvif_start_queue(struct xenvif_queue *queue) |
ca2f09f2 | 1933 | { |
e9ce7cb6 WL |
1934 | if (xenvif_schedulable(queue->vif)) |
1935 | xenvif_wake_queue(queue); | |
ca2f09f2 PD |
1936 | } |
1937 | ||
121fa4b7 | 1938 | int xenvif_kthread_guest_rx(void *data) |
b3f980bd | 1939 | { |
e9ce7cb6 | 1940 | struct xenvif_queue *queue = data; |
ca2f09f2 | 1941 | struct sk_buff *skb; |
b3f980bd WL |
1942 | |
1943 | while (!kthread_should_stop()) { | |
e9ce7cb6 WL |
1944 | wait_event_interruptible(queue->wq, |
1945 | rx_work_todo(queue) || | |
1946 | queue->vif->disabled || | |
b3f980bd | 1947 | kthread_should_stop()); |
e9d8b2c2 WL |
1948 | |
1949 | /* This frontend is found to be rogue, disable it in | |
1950 | * kthread context. Currently this is only set when | |
1951 | * netback finds out frontend sends malformed packet, | |
1952 | * but we cannot disable the interface in softirq | |
e9ce7cb6 WL |
1953 | * context so we defer it here, if this thread is |
1954 | * associated with queue 0. | |
e9d8b2c2 | 1955 | */ |
e9ce7cb6 WL |
1956 | if (unlikely(queue->vif->disabled && netif_carrier_ok(queue->vif->dev) && queue->id == 0)) |
1957 | xenvif_carrier_off(queue->vif); | |
e9d8b2c2 | 1958 | |
b3f980bd WL |
1959 | if (kthread_should_stop()) |
1960 | break; | |
1961 | ||
e9ce7cb6 WL |
1962 | if (queue->rx_queue_purge) { |
1963 | skb_queue_purge(&queue->rx_queue); | |
1964 | queue->rx_queue_purge = false; | |
09350788 ZK |
1965 | } |
1966 | ||
e9ce7cb6 WL |
1967 | if (!skb_queue_empty(&queue->rx_queue)) |
1968 | xenvif_rx_action(queue); | |
b3f980bd | 1969 | |
e9ce7cb6 WL |
1970 | if (skb_queue_empty(&queue->rx_queue) && |
1971 | xenvif_queue_stopped(queue)) { | |
1972 | del_timer_sync(&queue->wake_queue); | |
1973 | xenvif_start_queue(queue); | |
09350788 | 1974 | } |
ca2f09f2 | 1975 | |
b3f980bd WL |
1976 | cond_resched(); |
1977 | } | |
1978 | ||
ca2f09f2 | 1979 | /* Bin any remaining skbs */ |
e9ce7cb6 | 1980 | while ((skb = skb_dequeue(&queue->rx_queue)) != NULL) |
ca2f09f2 PD |
1981 | dev_kfree_skb(skb); |
1982 | ||
b3f980bd WL |
1983 | return 0; |
1984 | } | |
1985 | ||
f53c3fe8 ZK |
1986 | int xenvif_dealloc_kthread(void *data) |
1987 | { | |
e9ce7cb6 | 1988 | struct xenvif_queue *queue = data; |
f53c3fe8 ZK |
1989 | |
1990 | while (!kthread_should_stop()) { | |
e9ce7cb6 WL |
1991 | wait_event_interruptible(queue->dealloc_wq, |
1992 | tx_dealloc_work_todo(queue) || | |
f53c3fe8 ZK |
1993 | kthread_should_stop()); |
1994 | if (kthread_should_stop()) | |
1995 | break; | |
1996 | ||
e9ce7cb6 | 1997 | xenvif_tx_dealloc_action(queue); |
f53c3fe8 ZK |
1998 | cond_resched(); |
1999 | } | |
2000 | ||
2001 | /* Unmap anything remaining*/ | |
e9ce7cb6 WL |
2002 | if (tx_dealloc_work_todo(queue)) |
2003 | xenvif_tx_dealloc_action(queue); | |
f53c3fe8 ZK |
2004 | |
2005 | return 0; | |
2006 | } | |
2007 | ||
f942dc25 IC |
2008 | static int __init netback_init(void) |
2009 | { | |
f942dc25 | 2010 | int rc = 0; |
f942dc25 | 2011 | |
2a14b244 | 2012 | if (!xen_domain()) |
f942dc25 IC |
2013 | return -ENODEV; |
2014 | ||
8d3d53b3 AB |
2015 | /* Allow as many queues as there are CPUs, by default */ |
2016 | xenvif_max_queues = num_online_cpus(); | |
2017 | ||
37641494 | 2018 | if (fatal_skb_slots < XEN_NETBK_LEGACY_SLOTS_MAX) { |
383eda32 JP |
2019 | pr_info("fatal_skb_slots too small (%d), bump it to XEN_NETBK_LEGACY_SLOTS_MAX (%d)\n", |
2020 | fatal_skb_slots, XEN_NETBK_LEGACY_SLOTS_MAX); | |
37641494 | 2021 | fatal_skb_slots = XEN_NETBK_LEGACY_SLOTS_MAX; |
2810e5b9 WL |
2022 | } |
2023 | ||
f942dc25 IC |
2024 | rc = xenvif_xenbus_init(); |
2025 | if (rc) | |
2026 | goto failed_init; | |
2027 | ||
09350788 ZK |
2028 | rx_drain_timeout_jiffies = msecs_to_jiffies(rx_drain_timeout_msecs); |
2029 | ||
f942dc25 IC |
2030 | return 0; |
2031 | ||
2032 | failed_init: | |
f942dc25 | 2033 | return rc; |
f942dc25 IC |
2034 | } |
2035 | ||
2036 | module_init(netback_init); | |
2037 | ||
b103f358 WL |
2038 | static void __exit netback_fini(void) |
2039 | { | |
b103f358 | 2040 | xenvif_xenbus_fini(); |
b103f358 WL |
2041 | } |
2042 | module_exit(netback_fini); | |
2043 | ||
f942dc25 | 2044 | MODULE_LICENSE("Dual BSD/GPL"); |
f984cec6 | 2045 | MODULE_ALIAS("xen-backend:vif"); |