]>
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> | |
a9fd60e2 | 47 | #include <xen/page.h> |
f942dc25 IC |
48 | |
49 | #include <asm/xen/hypercall.h> | |
f942dc25 | 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 | */ | |
c489dbb1 | 55 | bool separate_tx_rx_irq = true; |
e1f00a69 WL |
56 | module_param(separate_tx_rx_irq, bool, 0644); |
57 | ||
f48da8b1 DV |
58 | /* The time that packets can stay on the guest Rx internal queue |
59 | * before they are dropped. | |
09350788 ZK |
60 | */ |
61 | unsigned int rx_drain_timeout_msecs = 10000; | |
62 | module_param(rx_drain_timeout_msecs, uint, 0444); | |
09350788 | 63 | |
ecf08d2d DV |
64 | /* The length of time before the frontend is considered unresponsive |
65 | * because it isn't providing Rx slots. | |
66 | */ | |
26c0e102 | 67 | unsigned int rx_stall_timeout_msecs = 60000; |
ecf08d2d | 68 | module_param(rx_stall_timeout_msecs, uint, 0444); |
ecf08d2d | 69 | |
56dd5af9 | 70 | #define MAX_QUEUES_DEFAULT 8 |
8d3d53b3 AB |
71 | unsigned int xenvif_max_queues; |
72 | module_param_named(max_queues, xenvif_max_queues, uint, 0644); | |
73 | MODULE_PARM_DESC(max_queues, | |
74 | "Maximum number of queues per virtual interface"); | |
75 | ||
2810e5b9 WL |
76 | /* |
77 | * This is the maximum slots a skb can have. If a guest sends a skb | |
78 | * which exceeds this limit it is considered malicious. | |
79 | */ | |
37641494 WL |
80 | #define FATAL_SKB_SLOTS_DEFAULT 20 |
81 | static unsigned int fatal_skb_slots = FATAL_SKB_SLOTS_DEFAULT; | |
82 | module_param(fatal_skb_slots, uint, 0444); | |
83 | ||
7e5d7753 MC |
84 | /* The amount to copy out of the first guest Tx slot into the skb's |
85 | * linear area. If the first slot has more data, it will be mapped | |
86 | * and put into the first frag. | |
87 | * | |
88 | * This is sized to avoid pulling headers from the frags for most | |
89 | * TCP/IP packets. | |
90 | */ | |
91 | #define XEN_NETBACK_TX_COPY_LEN 128 | |
92 | ||
40d8abde PD |
93 | /* This is the maximum number of flows in the hash cache. */ |
94 | #define XENVIF_HASH_CACHE_SIZE_DEFAULT 64 | |
95 | unsigned int xenvif_hash_cache_size = XENVIF_HASH_CACHE_SIZE_DEFAULT; | |
96 | module_param_named(hash_cache_size, xenvif_hash_cache_size, uint, 0644); | |
97 | MODULE_PARM_DESC(hash_cache_size, "Number of flows in the hash cache"); | |
7e5d7753 | 98 | |
e9ce7cb6 | 99 | static void xenvif_idx_release(struct xenvif_queue *queue, u16 pending_idx, |
7376419a WL |
100 | u8 status); |
101 | ||
e9ce7cb6 | 102 | static void make_tx_response(struct xenvif_queue *queue, |
f942dc25 | 103 | struct xen_netif_tx_request *txp, |
562abd39 | 104 | unsigned int extra_count, |
f942dc25 | 105 | s8 st); |
c8a4d299 | 106 | static void push_tx_responses(struct xenvif_queue *queue); |
b3f980bd | 107 | |
e9ce7cb6 | 108 | static inline int tx_work_todo(struct xenvif_queue *queue); |
b3f980bd | 109 | |
e9ce7cb6 | 110 | static inline unsigned long idx_to_pfn(struct xenvif_queue *queue, |
ea066ad1 | 111 | u16 idx) |
f942dc25 | 112 | { |
e9ce7cb6 | 113 | return page_to_pfn(queue->mmap_pages[idx]); |
f942dc25 IC |
114 | } |
115 | ||
e9ce7cb6 | 116 | static inline unsigned long idx_to_kaddr(struct xenvif_queue *queue, |
ea066ad1 | 117 | u16 idx) |
f942dc25 | 118 | { |
e9ce7cb6 | 119 | return (unsigned long)pfn_to_kaddr(idx_to_pfn(queue, idx)); |
f942dc25 IC |
120 | } |
121 | ||
7aceb47a ZK |
122 | #define callback_param(vif, pending_idx) \ |
123 | (vif->pending_tx_info[pending_idx].callback_struct) | |
124 | ||
f53c3fe8 ZK |
125 | /* Find the containing VIF's structure from a pointer in pending_tx_info array |
126 | */ | |
e9ce7cb6 | 127 | static inline struct xenvif_queue *ubuf_to_queue(const struct ubuf_info *ubuf) |
3e2234b3 | 128 | { |
f53c3fe8 ZK |
129 | u16 pending_idx = ubuf->desc; |
130 | struct pending_tx_info *temp = | |
131 | container_of(ubuf, struct pending_tx_info, callback_struct); | |
132 | return container_of(temp - pending_idx, | |
e9ce7cb6 | 133 | struct xenvif_queue, |
f53c3fe8 | 134 | pending_tx_info[0]); |
3e2234b3 | 135 | } |
f53c3fe8 | 136 | |
ea066ad1 IC |
137 | static u16 frag_get_pending_idx(skb_frag_t *frag) |
138 | { | |
139 | return (u16)frag->page_offset; | |
140 | } | |
141 | ||
142 | static void frag_set_pending_idx(skb_frag_t *frag, u16 pending_idx) | |
143 | { | |
144 | frag->page_offset = pending_idx; | |
145 | } | |
146 | ||
f942dc25 IC |
147 | static inline pending_ring_idx_t pending_index(unsigned i) |
148 | { | |
149 | return i & (MAX_PENDING_REQS-1); | |
150 | } | |
151 | ||
e9ce7cb6 | 152 | void xenvif_kick_thread(struct xenvif_queue *queue) |
b3f980bd | 153 | { |
e9ce7cb6 | 154 | wake_up(&queue->wq); |
b3f980bd WL |
155 | } |
156 | ||
e9ce7cb6 | 157 | void xenvif_napi_schedule_or_enable_events(struct xenvif_queue *queue) |
f942dc25 IC |
158 | { |
159 | int more_to_do; | |
160 | ||
e9ce7cb6 | 161 | RING_FINAL_CHECK_FOR_REQUESTS(&queue->tx, more_to_do); |
f942dc25 IC |
162 | |
163 | if (more_to_do) | |
e9ce7cb6 | 164 | napi_schedule(&queue->napi); |
f942dc25 IC |
165 | } |
166 | ||
e9ce7cb6 | 167 | static void tx_add_credit(struct xenvif_queue *queue) |
f942dc25 IC |
168 | { |
169 | unsigned long max_burst, max_credit; | |
170 | ||
171 | /* | |
172 | * Allow a burst big enough to transmit a jumbo packet of up to 128kB. | |
173 | * Otherwise the interface can seize up due to insufficient credit. | |
174 | */ | |
0f589967 | 175 | max_burst = max(131072UL, queue->credit_bytes); |
f942dc25 IC |
176 | |
177 | /* Take care that adding a new chunk of credit doesn't wrap to zero. */ | |
e9ce7cb6 WL |
178 | max_credit = queue->remaining_credit + queue->credit_bytes; |
179 | if (max_credit < queue->remaining_credit) | |
f942dc25 IC |
180 | max_credit = ULONG_MAX; /* wrapped: clamp to ULONG_MAX */ |
181 | ||
e9ce7cb6 | 182 | queue->remaining_credit = min(max_credit, max_burst); |
f942dc25 IC |
183 | } |
184 | ||
edafc132 | 185 | void xenvif_tx_credit_callback(unsigned long data) |
f942dc25 | 186 | { |
e9ce7cb6 WL |
187 | struct xenvif_queue *queue = (struct xenvif_queue *)data; |
188 | tx_add_credit(queue); | |
189 | xenvif_napi_schedule_or_enable_events(queue); | |
f942dc25 IC |
190 | } |
191 | ||
e9ce7cb6 | 192 | static void xenvif_tx_err(struct xenvif_queue *queue, |
562abd39 PD |
193 | struct xen_netif_tx_request *txp, |
194 | unsigned int extra_count, RING_IDX end) | |
f942dc25 | 195 | { |
e9ce7cb6 | 196 | RING_IDX cons = queue->tx.req_cons; |
f53c3fe8 | 197 | unsigned long flags; |
f942dc25 IC |
198 | |
199 | do { | |
e9ce7cb6 | 200 | spin_lock_irqsave(&queue->response_lock, flags); |
562abd39 | 201 | make_tx_response(queue, txp, extra_count, XEN_NETIF_RSP_ERROR); |
c8a4d299 | 202 | push_tx_responses(queue); |
e9ce7cb6 | 203 | spin_unlock_irqrestore(&queue->response_lock, flags); |
b9149729 | 204 | if (cons == end) |
f942dc25 | 205 | break; |
68a33bfd | 206 | RING_COPY_REQUEST(&queue->tx, cons++, txp); |
72eec92a | 207 | extra_count = 0; /* only the first frag can have extras */ |
f942dc25 | 208 | } while (1); |
e9ce7cb6 | 209 | queue->tx.req_cons = cons; |
f942dc25 IC |
210 | } |
211 | ||
7376419a | 212 | static void xenvif_fatal_tx_err(struct xenvif *vif) |
48856286 IC |
213 | { |
214 | netdev_err(vif->dev, "fatal error; disabling device\n"); | |
e9d8b2c2 | 215 | vif->disabled = true; |
e9ce7cb6 | 216 | /* Disable the vif from queue 0's kthread */ |
b17075d5 | 217 | if (vif->num_queues) |
e9ce7cb6 | 218 | xenvif_kick_thread(&vif->queues[0]); |
48856286 IC |
219 | } |
220 | ||
e9ce7cb6 | 221 | static int xenvif_count_requests(struct xenvif_queue *queue, |
7376419a | 222 | struct xen_netif_tx_request *first, |
562abd39 | 223 | unsigned int extra_count, |
7376419a WL |
224 | struct xen_netif_tx_request *txp, |
225 | int work_to_do) | |
f942dc25 | 226 | { |
e9ce7cb6 | 227 | RING_IDX cons = queue->tx.req_cons; |
2810e5b9 WL |
228 | int slots = 0; |
229 | int drop_err = 0; | |
59ccb4eb | 230 | int more_data; |
f942dc25 IC |
231 | |
232 | if (!(first->flags & XEN_NETTXF_more_data)) | |
233 | return 0; | |
234 | ||
235 | do { | |
59ccb4eb WL |
236 | struct xen_netif_tx_request dropped_tx = { 0 }; |
237 | ||
2810e5b9 | 238 | if (slots >= work_to_do) { |
e9ce7cb6 | 239 | netdev_err(queue->vif->dev, |
2810e5b9 WL |
240 | "Asked for %d slots but exceeds this limit\n", |
241 | work_to_do); | |
e9ce7cb6 | 242 | xenvif_fatal_tx_err(queue->vif); |
35876b5f | 243 | return -ENODATA; |
f942dc25 IC |
244 | } |
245 | ||
2810e5b9 WL |
246 | /* This guest is really using too many slots and |
247 | * considered malicious. | |
248 | */ | |
37641494 | 249 | if (unlikely(slots >= fatal_skb_slots)) { |
e9ce7cb6 | 250 | netdev_err(queue->vif->dev, |
2810e5b9 | 251 | "Malicious frontend using %d slots, threshold %u\n", |
37641494 | 252 | slots, fatal_skb_slots); |
e9ce7cb6 | 253 | xenvif_fatal_tx_err(queue->vif); |
35876b5f | 254 | return -E2BIG; |
f942dc25 IC |
255 | } |
256 | ||
2810e5b9 | 257 | /* Xen network protocol had implicit dependency on |
37641494 WL |
258 | * MAX_SKB_FRAGS. XEN_NETBK_LEGACY_SLOTS_MAX is set to |
259 | * the historical MAX_SKB_FRAGS value 18 to honor the | |
260 | * same behavior as before. Any packet using more than | |
261 | * 18 slots but less than fatal_skb_slots slots is | |
262 | * dropped | |
2810e5b9 | 263 | */ |
37641494 | 264 | if (!drop_err && slots >= XEN_NETBK_LEGACY_SLOTS_MAX) { |
2810e5b9 | 265 | if (net_ratelimit()) |
e9ce7cb6 | 266 | netdev_dbg(queue->vif->dev, |
2810e5b9 | 267 | "Too many slots (%d) exceeding limit (%d), dropping packet\n", |
37641494 | 268 | slots, XEN_NETBK_LEGACY_SLOTS_MAX); |
2810e5b9 WL |
269 | drop_err = -E2BIG; |
270 | } | |
271 | ||
59ccb4eb WL |
272 | if (drop_err) |
273 | txp = &dropped_tx; | |
274 | ||
68a33bfd | 275 | RING_COPY_REQUEST(&queue->tx, cons + slots, txp); |
03393fd5 WL |
276 | |
277 | /* If the guest submitted a frame >= 64 KiB then | |
278 | * first->size overflowed and following slots will | |
279 | * appear to be larger than the frame. | |
280 | * | |
281 | * This cannot be fatal error as there are buggy | |
282 | * frontends that do this. | |
283 | * | |
284 | * Consume all slots and drop the packet. | |
285 | */ | |
286 | if (!drop_err && txp->size > first->size) { | |
287 | if (net_ratelimit()) | |
e9ce7cb6 | 288 | netdev_dbg(queue->vif->dev, |
03393fd5 WL |
289 | "Invalid tx request, slot size %u > remaining size %u\n", |
290 | txp->size, first->size); | |
291 | drop_err = -EIO; | |
f942dc25 IC |
292 | } |
293 | ||
294 | first->size -= txp->size; | |
2810e5b9 | 295 | slots++; |
f942dc25 | 296 | |
d0089e8a | 297 | if (unlikely((txp->offset + txp->size) > XEN_PAGE_SIZE)) { |
68946159 | 298 | netdev_err(queue->vif->dev, "Cross page boundary, txp->offset: %u, size: %u\n", |
f942dc25 | 299 | txp->offset, txp->size); |
e9ce7cb6 | 300 | xenvif_fatal_tx_err(queue->vif); |
35876b5f | 301 | return -EINVAL; |
f942dc25 | 302 | } |
59ccb4eb WL |
303 | |
304 | more_data = txp->flags & XEN_NETTXF_more_data; | |
305 | ||
306 | if (!drop_err) | |
307 | txp++; | |
308 | ||
309 | } while (more_data); | |
2810e5b9 WL |
310 | |
311 | if (drop_err) { | |
562abd39 | 312 | xenvif_tx_err(queue, first, extra_count, cons + slots); |
2810e5b9 WL |
313 | return drop_err; |
314 | } | |
315 | ||
316 | return slots; | |
f942dc25 IC |
317 | } |
318 | ||
8f13dd96 ZK |
319 | |
320 | struct xenvif_tx_cb { | |
321 | u16 pending_idx; | |
322 | }; | |
323 | ||
324 | #define XENVIF_TX_CB(skb) ((struct xenvif_tx_cb *)(skb)->cb) | |
325 | ||
e9ce7cb6 | 326 | static inline void xenvif_tx_create_map_op(struct xenvif_queue *queue, |
562abd39 PD |
327 | u16 pending_idx, |
328 | struct xen_netif_tx_request *txp, | |
329 | unsigned int extra_count, | |
330 | struct gnttab_map_grant_ref *mop) | |
f53c3fe8 | 331 | { |
e9ce7cb6 WL |
332 | queue->pages_to_map[mop-queue->tx_map_ops] = queue->mmap_pages[pending_idx]; |
333 | gnttab_set_map_op(mop, idx_to_kaddr(queue, pending_idx), | |
f53c3fe8 | 334 | GNTMAP_host_map | GNTMAP_readonly, |
e9ce7cb6 | 335 | txp->gref, queue->vif->domid); |
f53c3fe8 | 336 | |
e9ce7cb6 | 337 | memcpy(&queue->pending_tx_info[pending_idx].req, txp, |
f53c3fe8 | 338 | sizeof(*txp)); |
562abd39 | 339 | queue->pending_tx_info[pending_idx].extra_count = extra_count; |
f53c3fe8 ZK |
340 | } |
341 | ||
e3377f36 ZK |
342 | static inline struct sk_buff *xenvif_alloc_skb(unsigned int size) |
343 | { | |
344 | struct sk_buff *skb = | |
345 | alloc_skb(size + NET_SKB_PAD + NET_IP_ALIGN, | |
346 | GFP_ATOMIC | __GFP_NOWARN); | |
347 | if (unlikely(skb == NULL)) | |
348 | return NULL; | |
349 | ||
350 | /* Packets passed to netif_rx() must have some headroom. */ | |
351 | skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN); | |
352 | ||
353 | /* Initialize it here to avoid later surprises */ | |
354 | skb_shinfo(skb)->destructor_arg = NULL; | |
355 | ||
356 | return skb; | |
357 | } | |
358 | ||
e9ce7cb6 | 359 | static struct gnttab_map_grant_ref *xenvif_get_requests(struct xenvif_queue *queue, |
f53c3fe8 ZK |
360 | struct sk_buff *skb, |
361 | struct xen_netif_tx_request *txp, | |
2475b225 RL |
362 | struct gnttab_map_grant_ref *gop, |
363 | unsigned int frag_overflow, | |
364 | struct sk_buff *nskb) | |
f942dc25 IC |
365 | { |
366 | struct skb_shared_info *shinfo = skb_shinfo(skb); | |
367 | skb_frag_t *frags = shinfo->frags; | |
8f13dd96 | 368 | u16 pending_idx = XENVIF_TX_CB(skb)->pending_idx; |
62bad319 ZK |
369 | int start; |
370 | pending_ring_idx_t index; | |
2475b225 | 371 | unsigned int nr_slots; |
2810e5b9 | 372 | |
2810e5b9 | 373 | nr_slots = shinfo->nr_frags; |
f942dc25 IC |
374 | |
375 | /* Skip first skb fragment if it is on same page as header fragment. */ | |
ea066ad1 | 376 | start = (frag_get_pending_idx(&shinfo->frags[0]) == pending_idx); |
f942dc25 | 377 | |
f53c3fe8 ZK |
378 | for (shinfo->nr_frags = start; shinfo->nr_frags < nr_slots; |
379 | shinfo->nr_frags++, txp++, gop++) { | |
e9ce7cb6 WL |
380 | index = pending_index(queue->pending_cons++); |
381 | pending_idx = queue->pending_ring[index]; | |
562abd39 | 382 | xenvif_tx_create_map_op(queue, pending_idx, txp, 0, gop); |
f53c3fe8 | 383 | frag_set_pending_idx(&frags[shinfo->nr_frags], pending_idx); |
f942dc25 IC |
384 | } |
385 | ||
e3377f36 | 386 | if (frag_overflow) { |
e3377f36 ZK |
387 | |
388 | shinfo = skb_shinfo(nskb); | |
389 | frags = shinfo->frags; | |
390 | ||
391 | for (shinfo->nr_frags = 0; shinfo->nr_frags < frag_overflow; | |
392 | shinfo->nr_frags++, txp++, gop++) { | |
e9ce7cb6 WL |
393 | index = pending_index(queue->pending_cons++); |
394 | pending_idx = queue->pending_ring[index]; | |
562abd39 PD |
395 | xenvif_tx_create_map_op(queue, pending_idx, txp, 0, |
396 | gop); | |
e3377f36 ZK |
397 | frag_set_pending_idx(&frags[shinfo->nr_frags], |
398 | pending_idx); | |
399 | } | |
400 | ||
401 | skb_shinfo(skb)->frag_list = nskb; | |
402 | } | |
2810e5b9 | 403 | |
f942dc25 IC |
404 | return gop; |
405 | } | |
406 | ||
e9ce7cb6 | 407 | static inline void xenvif_grant_handle_set(struct xenvif_queue *queue, |
f53c3fe8 ZK |
408 | u16 pending_idx, |
409 | grant_handle_t handle) | |
410 | { | |
e9ce7cb6 | 411 | if (unlikely(queue->grant_tx_handle[pending_idx] != |
f53c3fe8 | 412 | NETBACK_INVALID_HANDLE)) { |
e9ce7cb6 | 413 | netdev_err(queue->vif->dev, |
68946159 | 414 | "Trying to overwrite active handle! pending_idx: 0x%x\n", |
f53c3fe8 ZK |
415 | pending_idx); |
416 | BUG(); | |
417 | } | |
e9ce7cb6 | 418 | queue->grant_tx_handle[pending_idx] = handle; |
f53c3fe8 ZK |
419 | } |
420 | ||
e9ce7cb6 | 421 | static inline void xenvif_grant_handle_reset(struct xenvif_queue *queue, |
f53c3fe8 ZK |
422 | u16 pending_idx) |
423 | { | |
e9ce7cb6 | 424 | if (unlikely(queue->grant_tx_handle[pending_idx] == |
f53c3fe8 | 425 | NETBACK_INVALID_HANDLE)) { |
e9ce7cb6 | 426 | netdev_err(queue->vif->dev, |
68946159 | 427 | "Trying to unmap invalid handle! pending_idx: 0x%x\n", |
f53c3fe8 ZK |
428 | pending_idx); |
429 | BUG(); | |
430 | } | |
e9ce7cb6 | 431 | queue->grant_tx_handle[pending_idx] = NETBACK_INVALID_HANDLE; |
f53c3fe8 ZK |
432 | } |
433 | ||
e9ce7cb6 | 434 | static int xenvif_tx_check_gop(struct xenvif_queue *queue, |
7376419a | 435 | struct sk_buff *skb, |
bdab8275 ZK |
436 | struct gnttab_map_grant_ref **gopp_map, |
437 | struct gnttab_copy **gopp_copy) | |
f942dc25 | 438 | { |
9074ce24 | 439 | struct gnttab_map_grant_ref *gop_map = *gopp_map; |
8f13dd96 | 440 | u16 pending_idx = XENVIF_TX_CB(skb)->pending_idx; |
1a998d3e ZK |
441 | /* This always points to the shinfo of the skb being checked, which |
442 | * could be either the first or the one on the frag_list | |
443 | */ | |
f942dc25 | 444 | struct skb_shared_info *shinfo = skb_shinfo(skb); |
1a998d3e ZK |
445 | /* If this is non-NULL, we are currently checking the frag_list skb, and |
446 | * this points to the shinfo of the first one | |
447 | */ | |
448 | struct skb_shared_info *first_shinfo = NULL; | |
f942dc25 | 449 | int nr_frags = shinfo->nr_frags; |
1b860da0 ZK |
450 | const bool sharedslot = nr_frags && |
451 | frag_get_pending_idx(&shinfo->frags[0]) == pending_idx; | |
bdab8275 | 452 | int i, err; |
f942dc25 IC |
453 | |
454 | /* Check status of header. */ | |
bdab8275 | 455 | err = (*gopp_copy)->status; |
bdab8275 ZK |
456 | if (unlikely(err)) { |
457 | if (net_ratelimit()) | |
e9ce7cb6 | 458 | netdev_dbg(queue->vif->dev, |
00aefceb | 459 | "Grant copy of header failed! status: %d pending_idx: %u ref: %u\n", |
bdab8275 ZK |
460 | (*gopp_copy)->status, |
461 | pending_idx, | |
462 | (*gopp_copy)->source.u.ref); | |
1b860da0 ZK |
463 | /* The first frag might still have this slot mapped */ |
464 | if (!sharedslot) | |
465 | xenvif_idx_release(queue, pending_idx, | |
466 | XEN_NETIF_RSP_ERROR); | |
bdab8275 | 467 | } |
d8cfbfc4 | 468 | (*gopp_copy)++; |
f942dc25 | 469 | |
e3377f36 | 470 | check_frags: |
bdab8275 | 471 | for (i = 0; i < nr_frags; i++, gop_map++) { |
f942dc25 | 472 | int j, newerr; |
f942dc25 | 473 | |
ea066ad1 | 474 | pending_idx = frag_get_pending_idx(&shinfo->frags[i]); |
f942dc25 IC |
475 | |
476 | /* Check error status: if okay then remember grant handle. */ | |
bdab8275 | 477 | newerr = gop_map->status; |
2810e5b9 | 478 | |
f942dc25 | 479 | if (likely(!newerr)) { |
e9ce7cb6 | 480 | xenvif_grant_handle_set(queue, |
9074ce24 ZK |
481 | pending_idx, |
482 | gop_map->handle); | |
f942dc25 | 483 | /* Had a previous error? Invalidate this fragment. */ |
1b860da0 | 484 | if (unlikely(err)) { |
e9ce7cb6 | 485 | xenvif_idx_unmap(queue, pending_idx); |
1b860da0 ZK |
486 | /* If the mapping of the first frag was OK, but |
487 | * the header's copy failed, and they are | |
488 | * sharing a slot, send an error | |
489 | */ | |
490 | if (i == 0 && sharedslot) | |
491 | xenvif_idx_release(queue, pending_idx, | |
492 | XEN_NETIF_RSP_ERROR); | |
493 | else | |
494 | xenvif_idx_release(queue, pending_idx, | |
495 | XEN_NETIF_RSP_OKAY); | |
496 | } | |
f942dc25 IC |
497 | continue; |
498 | } | |
499 | ||
500 | /* Error on this fragment: respond to client with an error. */ | |
bdab8275 | 501 | if (net_ratelimit()) |
e9ce7cb6 | 502 | netdev_dbg(queue->vif->dev, |
00aefceb | 503 | "Grant map of %d. frag failed! status: %d pending_idx: %u ref: %u\n", |
bdab8275 ZK |
504 | i, |
505 | gop_map->status, | |
506 | pending_idx, | |
507 | gop_map->ref); | |
1b860da0 | 508 | |
e9ce7cb6 | 509 | xenvif_idx_release(queue, pending_idx, XEN_NETIF_RSP_ERROR); |
f942dc25 IC |
510 | |
511 | /* Not the first error? Preceding frags already invalidated. */ | |
512 | if (err) | |
513 | continue; | |
1b860da0 ZK |
514 | |
515 | /* First error: if the header haven't shared a slot with the | |
516 | * first frag, release it as well. | |
517 | */ | |
518 | if (!sharedslot) | |
519 | xenvif_idx_release(queue, | |
520 | XENVIF_TX_CB(skb)->pending_idx, | |
521 | XEN_NETIF_RSP_OKAY); | |
522 | ||
523 | /* Invalidate preceding fragments of this skb. */ | |
bdab8275 | 524 | for (j = 0; j < i; j++) { |
5ccb3ea7 | 525 | pending_idx = frag_get_pending_idx(&shinfo->frags[j]); |
e9ce7cb6 | 526 | xenvif_idx_unmap(queue, pending_idx); |
1b860da0 ZK |
527 | xenvif_idx_release(queue, pending_idx, |
528 | XEN_NETIF_RSP_OKAY); | |
f942dc25 IC |
529 | } |
530 | ||
1a998d3e ZK |
531 | /* And if we found the error while checking the frag_list, unmap |
532 | * the first skb's frags | |
533 | */ | |
534 | if (first_shinfo) { | |
535 | for (j = 0; j < first_shinfo->nr_frags; j++) { | |
536 | pending_idx = frag_get_pending_idx(&first_shinfo->frags[j]); | |
537 | xenvif_idx_unmap(queue, pending_idx); | |
1b860da0 ZK |
538 | xenvif_idx_release(queue, pending_idx, |
539 | XEN_NETIF_RSP_OKAY); | |
1a998d3e | 540 | } |
f942dc25 IC |
541 | } |
542 | ||
543 | /* Remember the error: invalidate all subsequent fragments. */ | |
544 | err = newerr; | |
545 | } | |
546 | ||
1a998d3e ZK |
547 | if (skb_has_frag_list(skb) && !first_shinfo) { |
548 | first_shinfo = skb_shinfo(skb); | |
549 | shinfo = skb_shinfo(skb_shinfo(skb)->frag_list); | |
e3377f36 | 550 | nr_frags = shinfo->nr_frags; |
e3377f36 ZK |
551 | |
552 | goto check_frags; | |
553 | } | |
554 | ||
bdab8275 | 555 | *gopp_map = gop_map; |
f942dc25 IC |
556 | return err; |
557 | } | |
558 | ||
e9ce7cb6 | 559 | static void xenvif_fill_frags(struct xenvif_queue *queue, struct sk_buff *skb) |
f942dc25 IC |
560 | { |
561 | struct skb_shared_info *shinfo = skb_shinfo(skb); | |
562 | int nr_frags = shinfo->nr_frags; | |
563 | int i; | |
f53c3fe8 ZK |
564 | u16 prev_pending_idx = INVALID_PENDING_IDX; |
565 | ||
f942dc25 IC |
566 | for (i = 0; i < nr_frags; i++) { |
567 | skb_frag_t *frag = shinfo->frags + i; | |
568 | struct xen_netif_tx_request *txp; | |
ea066ad1 IC |
569 | struct page *page; |
570 | u16 pending_idx; | |
f942dc25 | 571 | |
ea066ad1 | 572 | pending_idx = frag_get_pending_idx(frag); |
f942dc25 | 573 | |
f53c3fe8 | 574 | /* If this is not the first frag, chain it to the previous*/ |
bdab8275 | 575 | if (prev_pending_idx == INVALID_PENDING_IDX) |
f53c3fe8 | 576 | skb_shinfo(skb)->destructor_arg = |
e9ce7cb6 | 577 | &callback_param(queue, pending_idx); |
bdab8275 | 578 | else |
e9ce7cb6 WL |
579 | callback_param(queue, prev_pending_idx).ctx = |
580 | &callback_param(queue, pending_idx); | |
f53c3fe8 | 581 | |
e9ce7cb6 | 582 | callback_param(queue, pending_idx).ctx = NULL; |
f53c3fe8 ZK |
583 | prev_pending_idx = pending_idx; |
584 | ||
e9ce7cb6 WL |
585 | txp = &queue->pending_tx_info[pending_idx].req; |
586 | page = virt_to_page(idx_to_kaddr(queue, pending_idx)); | |
ea066ad1 | 587 | __skb_fill_page_desc(skb, i, page, txp->offset, txp->size); |
f942dc25 IC |
588 | skb->len += txp->size; |
589 | skb->data_len += txp->size; | |
590 | skb->truesize += txp->size; | |
591 | ||
f53c3fe8 | 592 | /* Take an extra reference to offset network stack's put_page */ |
e9ce7cb6 | 593 | get_page(queue->mmap_pages[pending_idx]); |
f942dc25 IC |
594 | } |
595 | } | |
596 | ||
e9ce7cb6 | 597 | static int xenvif_get_extras(struct xenvif_queue *queue, |
562abd39 PD |
598 | struct xen_netif_extra_info *extras, |
599 | unsigned int *extra_count, | |
600 | int work_to_do) | |
f942dc25 IC |
601 | { |
602 | struct xen_netif_extra_info extra; | |
e9ce7cb6 | 603 | RING_IDX cons = queue->tx.req_cons; |
f942dc25 IC |
604 | |
605 | do { | |
606 | if (unlikely(work_to_do-- <= 0)) { | |
e9ce7cb6 WL |
607 | netdev_err(queue->vif->dev, "Missing extra info\n"); |
608 | xenvif_fatal_tx_err(queue->vif); | |
f942dc25 IC |
609 | return -EBADR; |
610 | } | |
611 | ||
68a33bfd | 612 | RING_COPY_REQUEST(&queue->tx, cons, &extra); |
562abd39 PD |
613 | |
614 | queue->tx.req_cons = ++cons; | |
615 | (*extra_count)++; | |
616 | ||
f942dc25 IC |
617 | if (unlikely(!extra.type || |
618 | extra.type >= XEN_NETIF_EXTRA_TYPE_MAX)) { | |
e9ce7cb6 | 619 | netdev_err(queue->vif->dev, |
f942dc25 | 620 | "Invalid extra type: %d\n", extra.type); |
e9ce7cb6 | 621 | xenvif_fatal_tx_err(queue->vif); |
f942dc25 IC |
622 | return -EINVAL; |
623 | } | |
624 | ||
625 | memcpy(&extras[extra.type - 1], &extra, sizeof(extra)); | |
f942dc25 IC |
626 | } while (extra.flags & XEN_NETIF_EXTRA_FLAG_MORE); |
627 | ||
628 | return work_to_do; | |
629 | } | |
630 | ||
7376419a WL |
631 | static int xenvif_set_skb_gso(struct xenvif *vif, |
632 | struct sk_buff *skb, | |
633 | struct xen_netif_extra_info *gso) | |
f942dc25 IC |
634 | { |
635 | if (!gso->u.gso.size) { | |
48856286 | 636 | netdev_err(vif->dev, "GSO size must not be zero.\n"); |
7376419a | 637 | xenvif_fatal_tx_err(vif); |
f942dc25 IC |
638 | return -EINVAL; |
639 | } | |
640 | ||
a9468587 PD |
641 | switch (gso->u.gso.type) { |
642 | case XEN_NETIF_GSO_TYPE_TCPV4: | |
643 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4; | |
644 | break; | |
645 | case XEN_NETIF_GSO_TYPE_TCPV6: | |
646 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6; | |
647 | break; | |
648 | default: | |
48856286 | 649 | netdev_err(vif->dev, "Bad GSO type %d.\n", gso->u.gso.type); |
7376419a | 650 | xenvif_fatal_tx_err(vif); |
f942dc25 IC |
651 | return -EINVAL; |
652 | } | |
653 | ||
654 | skb_shinfo(skb)->gso_size = gso->u.gso.size; | |
b89587a7 | 655 | /* gso_segs will be calculated later */ |
f942dc25 IC |
656 | |
657 | return 0; | |
658 | } | |
659 | ||
e9ce7cb6 | 660 | static int checksum_setup(struct xenvif_queue *queue, struct sk_buff *skb) |
2eba61d5 | 661 | { |
2721637c | 662 | bool recalculate_partial_csum = false; |
2eba61d5 PD |
663 | |
664 | /* A GSO SKB must be CHECKSUM_PARTIAL. However some buggy | |
665 | * peers can fail to set NETRXF_csum_blank when sending a GSO | |
666 | * frame. In this case force the SKB to CHECKSUM_PARTIAL and | |
667 | * recalculate the partial checksum. | |
668 | */ | |
669 | if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) { | |
e9ce7cb6 | 670 | queue->stats.rx_gso_checksum_fixup++; |
2eba61d5 | 671 | skb->ip_summed = CHECKSUM_PARTIAL; |
2721637c | 672 | recalculate_partial_csum = true; |
2eba61d5 PD |
673 | } |
674 | ||
675 | /* A non-CHECKSUM_PARTIAL SKB does not require setup. */ | |
676 | if (skb->ip_summed != CHECKSUM_PARTIAL) | |
677 | return 0; | |
678 | ||
2721637c | 679 | return skb_checksum_setup(skb, recalculate_partial_csum); |
2eba61d5 PD |
680 | } |
681 | ||
e9ce7cb6 | 682 | static bool tx_credit_exceeded(struct xenvif_queue *queue, unsigned size) |
f942dc25 | 683 | { |
059dfa6a | 684 | u64 now = get_jiffies_64(); |
e9ce7cb6 WL |
685 | u64 next_credit = queue->credit_window_start + |
686 | msecs_to_jiffies(queue->credit_usec / 1000); | |
f942dc25 IC |
687 | |
688 | /* Timer could already be pending in rare cases. */ | |
e9ce7cb6 | 689 | if (timer_pending(&queue->credit_timeout)) |
f942dc25 IC |
690 | return true; |
691 | ||
692 | /* Passed the point where we can replenish credit? */ | |
059dfa6a | 693 | if (time_after_eq64(now, next_credit)) { |
e9ce7cb6 WL |
694 | queue->credit_window_start = now; |
695 | tx_add_credit(queue); | |
f942dc25 IC |
696 | } |
697 | ||
698 | /* Still too big to send right now? Set a callback. */ | |
e9ce7cb6 WL |
699 | if (size > queue->remaining_credit) { |
700 | queue->credit_timeout.data = | |
701 | (unsigned long)queue; | |
e9ce7cb6 | 702 | mod_timer(&queue->credit_timeout, |
f942dc25 | 703 | next_credit); |
e9ce7cb6 | 704 | queue->credit_window_start = next_credit; |
f942dc25 IC |
705 | |
706 | return true; | |
707 | } | |
708 | ||
709 | return false; | |
710 | } | |
711 | ||
210c34dc PD |
712 | /* No locking is required in xenvif_mcast_add/del() as they are |
713 | * only ever invoked from NAPI poll. An RCU list is used because | |
714 | * xenvif_mcast_match() is called asynchronously, during start_xmit. | |
715 | */ | |
716 | ||
717 | static int xenvif_mcast_add(struct xenvif *vif, const u8 *addr) | |
718 | { | |
719 | struct xenvif_mcast_addr *mcast; | |
720 | ||
721 | if (vif->fe_mcast_count == XEN_NETBK_MCAST_MAX) { | |
722 | if (net_ratelimit()) | |
723 | netdev_err(vif->dev, | |
724 | "Too many multicast addresses\n"); | |
725 | return -ENOSPC; | |
726 | } | |
727 | ||
728 | mcast = kzalloc(sizeof(*mcast), GFP_ATOMIC); | |
729 | if (!mcast) | |
730 | return -ENOMEM; | |
731 | ||
732 | ether_addr_copy(mcast->addr, addr); | |
733 | list_add_tail_rcu(&mcast->entry, &vif->fe_mcast_addr); | |
734 | vif->fe_mcast_count++; | |
735 | ||
736 | return 0; | |
737 | } | |
738 | ||
739 | static void xenvif_mcast_del(struct xenvif *vif, const u8 *addr) | |
740 | { | |
741 | struct xenvif_mcast_addr *mcast; | |
742 | ||
743 | list_for_each_entry_rcu(mcast, &vif->fe_mcast_addr, entry) { | |
744 | if (ether_addr_equal(addr, mcast->addr)) { | |
745 | --vif->fe_mcast_count; | |
746 | list_del_rcu(&mcast->entry); | |
747 | kfree_rcu(mcast, rcu); | |
748 | break; | |
749 | } | |
750 | } | |
751 | } | |
752 | ||
753 | bool xenvif_mcast_match(struct xenvif *vif, const u8 *addr) | |
754 | { | |
755 | struct xenvif_mcast_addr *mcast; | |
756 | ||
757 | rcu_read_lock(); | |
758 | list_for_each_entry_rcu(mcast, &vif->fe_mcast_addr, entry) { | |
759 | if (ether_addr_equal(addr, mcast->addr)) { | |
760 | rcu_read_unlock(); | |
761 | return true; | |
762 | } | |
763 | } | |
764 | rcu_read_unlock(); | |
765 | ||
766 | return false; | |
767 | } | |
768 | ||
769 | void xenvif_mcast_addr_list_free(struct xenvif *vif) | |
770 | { | |
771 | /* No need for locking or RCU here. NAPI poll and TX queue | |
772 | * are stopped. | |
773 | */ | |
774 | while (!list_empty(&vif->fe_mcast_addr)) { | |
775 | struct xenvif_mcast_addr *mcast; | |
776 | ||
777 | mcast = list_first_entry(&vif->fe_mcast_addr, | |
778 | struct xenvif_mcast_addr, | |
779 | entry); | |
780 | --vif->fe_mcast_count; | |
781 | list_del(&mcast->entry); | |
782 | kfree(mcast); | |
783 | } | |
784 | } | |
785 | ||
e9ce7cb6 | 786 | static void xenvif_tx_build_gops(struct xenvif_queue *queue, |
bdab8275 ZK |
787 | int budget, |
788 | unsigned *copy_ops, | |
789 | unsigned *map_ops) | |
f942dc25 | 790 | { |
2475b225 RL |
791 | struct gnttab_map_grant_ref *gop = queue->tx_map_ops; |
792 | struct sk_buff *skb, *nskb; | |
f942dc25 | 793 | int ret; |
2475b225 | 794 | unsigned int frag_overflow; |
f942dc25 | 795 | |
e9ce7cb6 | 796 | while (skb_queue_len(&queue->tx_queue) < budget) { |
f942dc25 | 797 | struct xen_netif_tx_request txreq; |
37641494 | 798 | struct xen_netif_tx_request txfrags[XEN_NETBK_LEGACY_SLOTS_MAX]; |
f942dc25 | 799 | struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX-1]; |
562abd39 | 800 | unsigned int extra_count; |
f942dc25 IC |
801 | u16 pending_idx; |
802 | RING_IDX idx; | |
803 | int work_to_do; | |
804 | unsigned int data_len; | |
805 | pending_ring_idx_t index; | |
806 | ||
e9ce7cb6 | 807 | if (queue->tx.sring->req_prod - queue->tx.req_cons > |
48856286 | 808 | XEN_NETIF_TX_RING_SIZE) { |
e9ce7cb6 | 809 | netdev_err(queue->vif->dev, |
48856286 IC |
810 | "Impossible number of requests. " |
811 | "req_prod %d, req_cons %d, size %ld\n", | |
e9ce7cb6 | 812 | queue->tx.sring->req_prod, queue->tx.req_cons, |
48856286 | 813 | XEN_NETIF_TX_RING_SIZE); |
e9ce7cb6 | 814 | xenvif_fatal_tx_err(queue->vif); |
e9d8b2c2 | 815 | break; |
48856286 IC |
816 | } |
817 | ||
e9ce7cb6 | 818 | work_to_do = RING_HAS_UNCONSUMED_REQUESTS(&queue->tx); |
b3f980bd WL |
819 | if (!work_to_do) |
820 | break; | |
f942dc25 | 821 | |
e9ce7cb6 | 822 | idx = queue->tx.req_cons; |
f942dc25 | 823 | rmb(); /* Ensure that we see the request before we copy it. */ |
68a33bfd | 824 | RING_COPY_REQUEST(&queue->tx, idx, &txreq); |
f942dc25 IC |
825 | |
826 | /* Credit-based scheduling. */ | |
e9ce7cb6 WL |
827 | if (txreq.size > queue->remaining_credit && |
828 | tx_credit_exceeded(queue, txreq.size)) | |
b3f980bd | 829 | break; |
f942dc25 | 830 | |
e9ce7cb6 | 831 | queue->remaining_credit -= txreq.size; |
f942dc25 IC |
832 | |
833 | work_to_do--; | |
e9ce7cb6 | 834 | queue->tx.req_cons = ++idx; |
f942dc25 IC |
835 | |
836 | memset(extras, 0, sizeof(extras)); | |
562abd39 | 837 | extra_count = 0; |
f942dc25 | 838 | if (txreq.flags & XEN_NETTXF_extra_info) { |
e9ce7cb6 | 839 | work_to_do = xenvif_get_extras(queue, extras, |
562abd39 | 840 | &extra_count, |
7376419a | 841 | work_to_do); |
e9ce7cb6 | 842 | idx = queue->tx.req_cons; |
48856286 | 843 | if (unlikely(work_to_do < 0)) |
b3f980bd | 844 | break; |
f942dc25 IC |
845 | } |
846 | ||
210c34dc PD |
847 | if (extras[XEN_NETIF_EXTRA_TYPE_MCAST_ADD - 1].type) { |
848 | struct xen_netif_extra_info *extra; | |
849 | ||
850 | extra = &extras[XEN_NETIF_EXTRA_TYPE_MCAST_ADD - 1]; | |
851 | ret = xenvif_mcast_add(queue->vif, extra->u.mcast.addr); | |
852 | ||
562abd39 | 853 | make_tx_response(queue, &txreq, extra_count, |
210c34dc PD |
854 | (ret == 0) ? |
855 | XEN_NETIF_RSP_OKAY : | |
856 | XEN_NETIF_RSP_ERROR); | |
857 | push_tx_responses(queue); | |
858 | continue; | |
859 | } | |
860 | ||
861 | if (extras[XEN_NETIF_EXTRA_TYPE_MCAST_DEL - 1].type) { | |
862 | struct xen_netif_extra_info *extra; | |
863 | ||
864 | extra = &extras[XEN_NETIF_EXTRA_TYPE_MCAST_DEL - 1]; | |
865 | xenvif_mcast_del(queue->vif, extra->u.mcast.addr); | |
866 | ||
562abd39 PD |
867 | make_tx_response(queue, &txreq, extra_count, |
868 | XEN_NETIF_RSP_OKAY); | |
210c34dc PD |
869 | push_tx_responses(queue); |
870 | continue; | |
871 | } | |
872 | ||
562abd39 PD |
873 | ret = xenvif_count_requests(queue, &txreq, extra_count, |
874 | txfrags, work_to_do); | |
48856286 | 875 | if (unlikely(ret < 0)) |
b3f980bd | 876 | break; |
48856286 | 877 | |
f942dc25 IC |
878 | idx += ret; |
879 | ||
880 | if (unlikely(txreq.size < ETH_HLEN)) { | |
e9ce7cb6 | 881 | netdev_dbg(queue->vif->dev, |
f942dc25 | 882 | "Bad packet size: %d\n", txreq.size); |
562abd39 | 883 | xenvif_tx_err(queue, &txreq, extra_count, idx); |
b3f980bd | 884 | break; |
f942dc25 IC |
885 | } |
886 | ||
887 | /* No crossing a page as the payload mustn't fragment. */ | |
d0089e8a | 888 | if (unlikely((txreq.offset + txreq.size) > XEN_PAGE_SIZE)) { |
e9ce7cb6 | 889 | netdev_err(queue->vif->dev, |
68946159 | 890 | "txreq.offset: %u, size: %u, end: %lu\n", |
f942dc25 | 891 | txreq.offset, txreq.size, |
d0089e8a | 892 | (unsigned long)(txreq.offset&~XEN_PAGE_MASK) + txreq.size); |
e9ce7cb6 | 893 | xenvif_fatal_tx_err(queue->vif); |
b3f980bd | 894 | break; |
f942dc25 IC |
895 | } |
896 | ||
e9ce7cb6 WL |
897 | index = pending_index(queue->pending_cons); |
898 | pending_idx = queue->pending_ring[index]; | |
f942dc25 | 899 | |
7e5d7753 | 900 | data_len = (txreq.size > XEN_NETBACK_TX_COPY_LEN && |
37641494 | 901 | ret < XEN_NETBK_LEGACY_SLOTS_MAX) ? |
7e5d7753 | 902 | XEN_NETBACK_TX_COPY_LEN : txreq.size; |
f942dc25 | 903 | |
e3377f36 | 904 | skb = xenvif_alloc_skb(data_len); |
f942dc25 | 905 | if (unlikely(skb == NULL)) { |
e9ce7cb6 | 906 | netdev_dbg(queue->vif->dev, |
f942dc25 | 907 | "Can't allocate a skb in start_xmit.\n"); |
562abd39 | 908 | xenvif_tx_err(queue, &txreq, extra_count, idx); |
f942dc25 IC |
909 | break; |
910 | } | |
911 | ||
2475b225 RL |
912 | skb_shinfo(skb)->nr_frags = ret; |
913 | if (data_len < txreq.size) | |
914 | skb_shinfo(skb)->nr_frags++; | |
915 | /* At this point shinfo->nr_frags is in fact the number of | |
916 | * slots, which can be as large as XEN_NETBK_LEGACY_SLOTS_MAX. | |
917 | */ | |
918 | frag_overflow = 0; | |
919 | nskb = NULL; | |
920 | if (skb_shinfo(skb)->nr_frags > MAX_SKB_FRAGS) { | |
921 | frag_overflow = skb_shinfo(skb)->nr_frags - MAX_SKB_FRAGS; | |
922 | BUG_ON(frag_overflow > MAX_SKB_FRAGS); | |
923 | skb_shinfo(skb)->nr_frags = MAX_SKB_FRAGS; | |
924 | nskb = xenvif_alloc_skb(0); | |
925 | if (unlikely(nskb == NULL)) { | |
926 | kfree_skb(skb); | |
562abd39 | 927 | xenvif_tx_err(queue, &txreq, extra_count, idx); |
2475b225 RL |
928 | if (net_ratelimit()) |
929 | netdev_err(queue->vif->dev, | |
930 | "Can't allocate the frag_list skb.\n"); | |
931 | break; | |
932 | } | |
933 | } | |
934 | ||
f942dc25 IC |
935 | if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) { |
936 | struct xen_netif_extra_info *gso; | |
937 | gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1]; | |
938 | ||
e9ce7cb6 | 939 | if (xenvif_set_skb_gso(queue->vif, skb, gso)) { |
7376419a | 940 | /* Failure in xenvif_set_skb_gso is fatal. */ |
f942dc25 | 941 | kfree_skb(skb); |
2475b225 | 942 | kfree_skb(nskb); |
b3f980bd | 943 | break; |
f942dc25 IC |
944 | } |
945 | } | |
946 | ||
c2d09fde PD |
947 | if (extras[XEN_NETIF_EXTRA_TYPE_HASH - 1].type) { |
948 | struct xen_netif_extra_info *extra; | |
949 | enum pkt_hash_types type = PKT_HASH_TYPE_NONE; | |
950 | ||
951 | extra = &extras[XEN_NETIF_EXTRA_TYPE_HASH - 1]; | |
952 | ||
953 | switch (extra->u.hash.type) { | |
954 | case _XEN_NETIF_CTRL_HASH_TYPE_IPV4: | |
955 | case _XEN_NETIF_CTRL_HASH_TYPE_IPV6: | |
956 | type = PKT_HASH_TYPE_L3; | |
957 | break; | |
958 | ||
959 | case _XEN_NETIF_CTRL_HASH_TYPE_IPV4_TCP: | |
960 | case _XEN_NETIF_CTRL_HASH_TYPE_IPV6_TCP: | |
961 | type = PKT_HASH_TYPE_L4; | |
962 | break; | |
963 | ||
964 | default: | |
965 | break; | |
966 | } | |
967 | ||
968 | if (type != PKT_HASH_TYPE_NONE) | |
969 | skb_set_hash(skb, | |
970 | *(u32 *)extra->u.hash.value, | |
971 | type); | |
972 | } | |
973 | ||
8f13dd96 | 974 | XENVIF_TX_CB(skb)->pending_idx = pending_idx; |
f942dc25 IC |
975 | |
976 | __skb_put(skb, data_len); | |
e9ce7cb6 WL |
977 | queue->tx_copy_ops[*copy_ops].source.u.ref = txreq.gref; |
978 | queue->tx_copy_ops[*copy_ops].source.domid = queue->vif->domid; | |
979 | queue->tx_copy_ops[*copy_ops].source.offset = txreq.offset; | |
bdab8275 | 980 | |
e9ce7cb6 | 981 | queue->tx_copy_ops[*copy_ops].dest.u.gmfn = |
0df4f266 | 982 | virt_to_gfn(skb->data); |
e9ce7cb6 WL |
983 | queue->tx_copy_ops[*copy_ops].dest.domid = DOMID_SELF; |
984 | queue->tx_copy_ops[*copy_ops].dest.offset = | |
d0089e8a | 985 | offset_in_page(skb->data) & ~XEN_PAGE_MASK; |
bdab8275 | 986 | |
e9ce7cb6 WL |
987 | queue->tx_copy_ops[*copy_ops].len = data_len; |
988 | queue->tx_copy_ops[*copy_ops].flags = GNTCOPY_source_gref; | |
bdab8275 ZK |
989 | |
990 | (*copy_ops)++; | |
f942dc25 | 991 | |
f942dc25 | 992 | if (data_len < txreq.size) { |
ea066ad1 IC |
993 | frag_set_pending_idx(&skb_shinfo(skb)->frags[0], |
994 | pending_idx); | |
562abd39 PD |
995 | xenvif_tx_create_map_op(queue, pending_idx, &txreq, |
996 | extra_count, gop); | |
bdab8275 | 997 | gop++; |
f942dc25 | 998 | } else { |
ea066ad1 IC |
999 | frag_set_pending_idx(&skb_shinfo(skb)->frags[0], |
1000 | INVALID_PENDING_IDX); | |
562abd39 PD |
1001 | memcpy(&queue->pending_tx_info[pending_idx].req, |
1002 | &txreq, sizeof(txreq)); | |
1003 | queue->pending_tx_info[pending_idx].extra_count = | |
1004 | extra_count; | |
f942dc25 IC |
1005 | } |
1006 | ||
e9ce7cb6 | 1007 | queue->pending_cons++; |
f942dc25 | 1008 | |
2475b225 RL |
1009 | gop = xenvif_get_requests(queue, skb, txfrags, gop, |
1010 | frag_overflow, nskb); | |
f942dc25 | 1011 | |
e9ce7cb6 | 1012 | __skb_queue_tail(&queue->tx_queue, skb); |
1e0b6eac | 1013 | |
e9ce7cb6 | 1014 | queue->tx.req_cons = idx; |
f942dc25 | 1015 | |
e9ce7cb6 WL |
1016 | if (((gop-queue->tx_map_ops) >= ARRAY_SIZE(queue->tx_map_ops)) || |
1017 | (*copy_ops >= ARRAY_SIZE(queue->tx_copy_ops))) | |
f942dc25 IC |
1018 | break; |
1019 | } | |
1020 | ||
e9ce7cb6 | 1021 | (*map_ops) = gop - queue->tx_map_ops; |
bdab8275 | 1022 | return; |
f942dc25 IC |
1023 | } |
1024 | ||
e3377f36 ZK |
1025 | /* Consolidate skb with a frag_list into a brand new one with local pages on |
1026 | * frags. Returns 0 or -ENOMEM if can't allocate new pages. | |
1027 | */ | |
e9ce7cb6 | 1028 | static int xenvif_handle_frag_list(struct xenvif_queue *queue, struct sk_buff *skb) |
e3377f36 ZK |
1029 | { |
1030 | unsigned int offset = skb_headlen(skb); | |
1031 | skb_frag_t frags[MAX_SKB_FRAGS]; | |
49d9991a | 1032 | int i, f; |
e3377f36 ZK |
1033 | struct ubuf_info *uarg; |
1034 | struct sk_buff *nskb = skb_shinfo(skb)->frag_list; | |
1035 | ||
e9ce7cb6 WL |
1036 | queue->stats.tx_zerocopy_sent += 2; |
1037 | queue->stats.tx_frag_overflow++; | |
e3377f36 | 1038 | |
e9ce7cb6 | 1039 | xenvif_fill_frags(queue, nskb); |
e3377f36 ZK |
1040 | /* Subtract frags size, we will correct it later */ |
1041 | skb->truesize -= skb->data_len; | |
1042 | skb->len += nskb->len; | |
1043 | skb->data_len += nskb->len; | |
1044 | ||
1045 | /* create a brand new frags array and coalesce there */ | |
1046 | for (i = 0; offset < skb->len; i++) { | |
1047 | struct page *page; | |
1048 | unsigned int len; | |
1049 | ||
1050 | BUG_ON(i >= MAX_SKB_FRAGS); | |
44cc8ed1 | 1051 | page = alloc_page(GFP_ATOMIC); |
e3377f36 ZK |
1052 | if (!page) { |
1053 | int j; | |
1054 | skb->truesize += skb->data_len; | |
1055 | for (j = 0; j < i; j++) | |
1056 | put_page(frags[j].page.p); | |
1057 | return -ENOMEM; | |
1058 | } | |
1059 | ||
1060 | if (offset + PAGE_SIZE < skb->len) | |
1061 | len = PAGE_SIZE; | |
1062 | else | |
1063 | len = skb->len - offset; | |
1064 | if (skb_copy_bits(skb, offset, page_address(page), len)) | |
1065 | BUG(); | |
1066 | ||
1067 | offset += len; | |
1068 | frags[i].page.p = page; | |
1069 | frags[i].page_offset = 0; | |
1070 | skb_frag_size_set(&frags[i], len); | |
1071 | } | |
49d9991a | 1072 | |
b0c21bad DV |
1073 | /* Copied all the bits from the frag list -- free it. */ |
1074 | skb_frag_list_init(skb); | |
1075 | xenvif_skb_zerocopy_prepare(queue, nskb); | |
1076 | kfree_skb(nskb); | |
1077 | ||
49d9991a DV |
1078 | /* Release all the original (foreign) frags. */ |
1079 | for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) | |
1080 | skb_frag_unref(skb, f); | |
e3377f36 | 1081 | uarg = skb_shinfo(skb)->destructor_arg; |
a64bd934 WL |
1082 | /* increase inflight counter to offset decrement in callback */ |
1083 | atomic_inc(&queue->inflight_packets); | |
e3377f36 ZK |
1084 | uarg->callback(uarg, true); |
1085 | skb_shinfo(skb)->destructor_arg = NULL; | |
1086 | ||
b0c21bad DV |
1087 | /* Fill the skb with the new (local) frags. */ |
1088 | memcpy(skb_shinfo(skb)->frags, frags, i * sizeof(skb_frag_t)); | |
1089 | skb_shinfo(skb)->nr_frags = i; | |
1090 | skb->truesize += i * PAGE_SIZE; | |
e3377f36 ZK |
1091 | |
1092 | return 0; | |
1093 | } | |
b3f980bd | 1094 | |
e9ce7cb6 | 1095 | static int xenvif_tx_submit(struct xenvif_queue *queue) |
f942dc25 | 1096 | { |
e9ce7cb6 WL |
1097 | struct gnttab_map_grant_ref *gop_map = queue->tx_map_ops; |
1098 | struct gnttab_copy *gop_copy = queue->tx_copy_ops; | |
f942dc25 | 1099 | struct sk_buff *skb; |
b3f980bd | 1100 | int work_done = 0; |
f942dc25 | 1101 | |
e9ce7cb6 | 1102 | while ((skb = __skb_dequeue(&queue->tx_queue)) != NULL) { |
f942dc25 | 1103 | struct xen_netif_tx_request *txp; |
f942dc25 IC |
1104 | u16 pending_idx; |
1105 | unsigned data_len; | |
1106 | ||
8f13dd96 | 1107 | pending_idx = XENVIF_TX_CB(skb)->pending_idx; |
e9ce7cb6 | 1108 | txp = &queue->pending_tx_info[pending_idx].req; |
f942dc25 IC |
1109 | |
1110 | /* Check the remap error code. */ | |
e9ce7cb6 | 1111 | if (unlikely(xenvif_tx_check_gop(queue, skb, &gop_map, &gop_copy))) { |
b42cc6e4 ZK |
1112 | /* If there was an error, xenvif_tx_check_gop is |
1113 | * expected to release all the frags which were mapped, | |
1114 | * so kfree_skb shouldn't do it again | |
1115 | */ | |
f942dc25 | 1116 | skb_shinfo(skb)->nr_frags = 0; |
b42cc6e4 ZK |
1117 | if (skb_has_frag_list(skb)) { |
1118 | struct sk_buff *nskb = | |
1119 | skb_shinfo(skb)->frag_list; | |
1120 | skb_shinfo(nskb)->nr_frags = 0; | |
1121 | } | |
f942dc25 IC |
1122 | kfree_skb(skb); |
1123 | continue; | |
1124 | } | |
1125 | ||
1126 | data_len = skb->len; | |
e9ce7cb6 | 1127 | callback_param(queue, pending_idx).ctx = NULL; |
f942dc25 IC |
1128 | if (data_len < txp->size) { |
1129 | /* Append the packet payload as a fragment. */ | |
1130 | txp->offset += data_len; | |
1131 | txp->size -= data_len; | |
1132 | } else { | |
1133 | /* Schedule a response immediately. */ | |
e9ce7cb6 | 1134 | xenvif_idx_release(queue, pending_idx, |
bdab8275 | 1135 | XEN_NETIF_RSP_OKAY); |
f942dc25 IC |
1136 | } |
1137 | ||
1138 | if (txp->flags & XEN_NETTXF_csum_blank) | |
1139 | skb->ip_summed = CHECKSUM_PARTIAL; | |
1140 | else if (txp->flags & XEN_NETTXF_data_validated) | |
1141 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
1142 | ||
e9ce7cb6 | 1143 | xenvif_fill_frags(queue, skb); |
f942dc25 | 1144 | |
e3377f36 | 1145 | if (unlikely(skb_has_frag_list(skb))) { |
e9ce7cb6 | 1146 | if (xenvif_handle_frag_list(queue, skb)) { |
e3377f36 | 1147 | if (net_ratelimit()) |
e9ce7cb6 | 1148 | netdev_err(queue->vif->dev, |
e3377f36 | 1149 | "Not enough memory to consolidate frag_list!\n"); |
a64bd934 | 1150 | xenvif_skb_zerocopy_prepare(queue, skb); |
e3377f36 ZK |
1151 | kfree_skb(skb); |
1152 | continue; | |
1153 | } | |
1154 | } | |
1155 | ||
e9ce7cb6 | 1156 | skb->dev = queue->vif->dev; |
f942dc25 | 1157 | skb->protocol = eth_type_trans(skb, skb->dev); |
f9ca8f74 | 1158 | skb_reset_network_header(skb); |
f942dc25 | 1159 | |
e9ce7cb6 WL |
1160 | if (checksum_setup(queue, skb)) { |
1161 | netdev_dbg(queue->vif->dev, | |
f942dc25 | 1162 | "Can't setup checksum in net_tx_action\n"); |
f53c3fe8 ZK |
1163 | /* We have to set this flag to trigger the callback */ |
1164 | if (skb_shinfo(skb)->destructor_arg) | |
a64bd934 | 1165 | xenvif_skb_zerocopy_prepare(queue, skb); |
f942dc25 IC |
1166 | kfree_skb(skb); |
1167 | continue; | |
1168 | } | |
1169 | ||
40893fd0 | 1170 | skb_probe_transport_header(skb, 0); |
f9ca8f74 | 1171 | |
b89587a7 PD |
1172 | /* If the packet is GSO then we will have just set up the |
1173 | * transport header offset in checksum_setup so it's now | |
1174 | * straightforward to calculate gso_segs. | |
1175 | */ | |
1176 | if (skb_is_gso(skb)) { | |
1177 | int mss = skb_shinfo(skb)->gso_size; | |
1178 | int hdrlen = skb_transport_header(skb) - | |
1179 | skb_mac_header(skb) + | |
1180 | tcp_hdrlen(skb); | |
1181 | ||
1182 | skb_shinfo(skb)->gso_segs = | |
1183 | DIV_ROUND_UP(skb->len - hdrlen, mss); | |
1184 | } | |
1185 | ||
e9ce7cb6 WL |
1186 | queue->stats.rx_bytes += skb->len; |
1187 | queue->stats.rx_packets++; | |
f942dc25 | 1188 | |
b3f980bd WL |
1189 | work_done++; |
1190 | ||
f53c3fe8 ZK |
1191 | /* Set this flag right before netif_receive_skb, otherwise |
1192 | * someone might think this packet already left netback, and | |
1193 | * do a skb_copy_ubufs while we are still in control of the | |
1194 | * skb. E.g. the __pskb_pull_tail earlier can do such thing. | |
1195 | */ | |
1bb332af | 1196 | if (skb_shinfo(skb)->destructor_arg) { |
a64bd934 | 1197 | xenvif_skb_zerocopy_prepare(queue, skb); |
e9ce7cb6 | 1198 | queue->stats.tx_zerocopy_sent++; |
1bb332af | 1199 | } |
f53c3fe8 | 1200 | |
b3f980bd | 1201 | netif_receive_skb(skb); |
f942dc25 | 1202 | } |
b3f980bd WL |
1203 | |
1204 | return work_done; | |
f942dc25 IC |
1205 | } |
1206 | ||
3e2234b3 ZK |
1207 | void xenvif_zerocopy_callback(struct ubuf_info *ubuf, bool zerocopy_success) |
1208 | { | |
f53c3fe8 ZK |
1209 | unsigned long flags; |
1210 | pending_ring_idx_t index; | |
e9ce7cb6 | 1211 | struct xenvif_queue *queue = ubuf_to_queue(ubuf); |
f53c3fe8 ZK |
1212 | |
1213 | /* This is the only place where we grab this lock, to protect callbacks | |
1214 | * from each other. | |
1215 | */ | |
e9ce7cb6 | 1216 | spin_lock_irqsave(&queue->callback_lock, flags); |
f53c3fe8 ZK |
1217 | do { |
1218 | u16 pending_idx = ubuf->desc; | |
1219 | ubuf = (struct ubuf_info *) ubuf->ctx; | |
e9ce7cb6 | 1220 | BUG_ON(queue->dealloc_prod - queue->dealloc_cons >= |
f53c3fe8 | 1221 | MAX_PENDING_REQS); |
e9ce7cb6 WL |
1222 | index = pending_index(queue->dealloc_prod); |
1223 | queue->dealloc_ring[index] = pending_idx; | |
f53c3fe8 ZK |
1224 | /* Sync with xenvif_tx_dealloc_action: |
1225 | * insert idx then incr producer. | |
1226 | */ | |
1227 | smp_wmb(); | |
e9ce7cb6 | 1228 | queue->dealloc_prod++; |
f53c3fe8 | 1229 | } while (ubuf); |
e9ce7cb6 | 1230 | spin_unlock_irqrestore(&queue->callback_lock, flags); |
f53c3fe8 | 1231 | |
1bb332af | 1232 | if (likely(zerocopy_success)) |
e9ce7cb6 | 1233 | queue->stats.tx_zerocopy_success++; |
1bb332af | 1234 | else |
e9ce7cb6 | 1235 | queue->stats.tx_zerocopy_fail++; |
a64bd934 | 1236 | xenvif_skb_zerocopy_complete(queue); |
f53c3fe8 ZK |
1237 | } |
1238 | ||
e9ce7cb6 | 1239 | static inline void xenvif_tx_dealloc_action(struct xenvif_queue *queue) |
f53c3fe8 ZK |
1240 | { |
1241 | struct gnttab_unmap_grant_ref *gop; | |
1242 | pending_ring_idx_t dc, dp; | |
1243 | u16 pending_idx, pending_idx_release[MAX_PENDING_REQS]; | |
1244 | unsigned int i = 0; | |
1245 | ||
e9ce7cb6 WL |
1246 | dc = queue->dealloc_cons; |
1247 | gop = queue->tx_unmap_ops; | |
f53c3fe8 ZK |
1248 | |
1249 | /* Free up any grants we have finished using */ | |
1250 | do { | |
e9ce7cb6 | 1251 | dp = queue->dealloc_prod; |
f53c3fe8 ZK |
1252 | |
1253 | /* Ensure we see all indices enqueued by all | |
1254 | * xenvif_zerocopy_callback(). | |
1255 | */ | |
1256 | smp_rmb(); | |
1257 | ||
1258 | while (dc != dp) { | |
50c2e4dd | 1259 | BUG_ON(gop - queue->tx_unmap_ops >= MAX_PENDING_REQS); |
f53c3fe8 | 1260 | pending_idx = |
e9ce7cb6 | 1261 | queue->dealloc_ring[pending_index(dc++)]; |
f53c3fe8 | 1262 | |
50c2e4dd | 1263 | pending_idx_release[gop - queue->tx_unmap_ops] = |
f53c3fe8 | 1264 | pending_idx; |
50c2e4dd | 1265 | queue->pages_to_unmap[gop - queue->tx_unmap_ops] = |
e9ce7cb6 | 1266 | queue->mmap_pages[pending_idx]; |
f53c3fe8 | 1267 | gnttab_set_unmap_op(gop, |
e9ce7cb6 | 1268 | idx_to_kaddr(queue, pending_idx), |
f53c3fe8 | 1269 | GNTMAP_host_map, |
e9ce7cb6 WL |
1270 | queue->grant_tx_handle[pending_idx]); |
1271 | xenvif_grant_handle_reset(queue, pending_idx); | |
f53c3fe8 ZK |
1272 | ++gop; |
1273 | } | |
1274 | ||
e9ce7cb6 | 1275 | } while (dp != queue->dealloc_prod); |
f53c3fe8 | 1276 | |
e9ce7cb6 | 1277 | queue->dealloc_cons = dc; |
f53c3fe8 | 1278 | |
e9ce7cb6 | 1279 | if (gop - queue->tx_unmap_ops > 0) { |
f53c3fe8 | 1280 | int ret; |
e9ce7cb6 | 1281 | ret = gnttab_unmap_refs(queue->tx_unmap_ops, |
f53c3fe8 | 1282 | NULL, |
e9ce7cb6 WL |
1283 | queue->pages_to_unmap, |
1284 | gop - queue->tx_unmap_ops); | |
f53c3fe8 | 1285 | if (ret) { |
68946159 | 1286 | netdev_err(queue->vif->dev, "Unmap fail: nr_ops %tu ret %d\n", |
e9ce7cb6 WL |
1287 | gop - queue->tx_unmap_ops, ret); |
1288 | for (i = 0; i < gop - queue->tx_unmap_ops; ++i) { | |
f53c3fe8 | 1289 | if (gop[i].status != GNTST_okay) |
e9ce7cb6 | 1290 | netdev_err(queue->vif->dev, |
68946159 | 1291 | " host_addr: 0x%llx handle: 0x%x status: %d\n", |
f53c3fe8 ZK |
1292 | gop[i].host_addr, |
1293 | gop[i].handle, | |
1294 | gop[i].status); | |
1295 | } | |
1296 | BUG(); | |
1297 | } | |
1298 | } | |
1299 | ||
e9ce7cb6 WL |
1300 | for (i = 0; i < gop - queue->tx_unmap_ops; ++i) |
1301 | xenvif_idx_release(queue, pending_idx_release[i], | |
f53c3fe8 | 1302 | XEN_NETIF_RSP_OKAY); |
3e2234b3 ZK |
1303 | } |
1304 | ||
f53c3fe8 | 1305 | |
f942dc25 | 1306 | /* Called after netfront has transmitted */ |
e9ce7cb6 | 1307 | int xenvif_tx_action(struct xenvif_queue *queue, int budget) |
f942dc25 | 1308 | { |
bdab8275 | 1309 | unsigned nr_mops, nr_cops = 0; |
f53c3fe8 | 1310 | int work_done, ret; |
f942dc25 | 1311 | |
e9ce7cb6 | 1312 | if (unlikely(!tx_work_todo(queue))) |
b3f980bd WL |
1313 | return 0; |
1314 | ||
e9ce7cb6 | 1315 | xenvif_tx_build_gops(queue, budget, &nr_cops, &nr_mops); |
f942dc25 | 1316 | |
bdab8275 | 1317 | if (nr_cops == 0) |
b3f980bd WL |
1318 | return 0; |
1319 | ||
e9ce7cb6 | 1320 | gnttab_batch_copy(queue->tx_copy_ops, nr_cops); |
bdab8275 | 1321 | if (nr_mops != 0) { |
e9ce7cb6 | 1322 | ret = gnttab_map_refs(queue->tx_map_ops, |
bdab8275 | 1323 | NULL, |
e9ce7cb6 | 1324 | queue->pages_to_map, |
bdab8275 ZK |
1325 | nr_mops); |
1326 | BUG_ON(ret); | |
1327 | } | |
f942dc25 | 1328 | |
e9ce7cb6 | 1329 | work_done = xenvif_tx_submit(queue); |
f942dc25 | 1330 | |
b3f980bd | 1331 | return work_done; |
f942dc25 IC |
1332 | } |
1333 | ||
e9ce7cb6 | 1334 | static void xenvif_idx_release(struct xenvif_queue *queue, u16 pending_idx, |
7376419a | 1335 | u8 status) |
f942dc25 | 1336 | { |
f942dc25 | 1337 | struct pending_tx_info *pending_tx_info; |
f53c3fe8 | 1338 | pending_ring_idx_t index; |
f53c3fe8 | 1339 | unsigned long flags; |
2810e5b9 | 1340 | |
e9ce7cb6 | 1341 | pending_tx_info = &queue->pending_tx_info[pending_idx]; |
7fbb9d84 | 1342 | |
e9ce7cb6 | 1343 | spin_lock_irqsave(&queue->response_lock, flags); |
7fbb9d84 | 1344 | |
562abd39 PD |
1345 | make_tx_response(queue, &pending_tx_info->req, |
1346 | pending_tx_info->extra_count, status); | |
7fbb9d84 DV |
1347 | |
1348 | /* Release the pending index before pusing the Tx response so | |
1349 | * its available before a new Tx request is pushed by the | |
1350 | * frontend. | |
1351 | */ | |
1352 | index = pending_index(queue->pending_prod++); | |
e9ce7cb6 | 1353 | queue->pending_ring[index] = pending_idx; |
7fbb9d84 | 1354 | |
c8a4d299 | 1355 | push_tx_responses(queue); |
7fbb9d84 | 1356 | |
e9ce7cb6 | 1357 | spin_unlock_irqrestore(&queue->response_lock, flags); |
f942dc25 IC |
1358 | } |
1359 | ||
2810e5b9 | 1360 | |
e9ce7cb6 | 1361 | static void make_tx_response(struct xenvif_queue *queue, |
f942dc25 | 1362 | struct xen_netif_tx_request *txp, |
562abd39 | 1363 | unsigned int extra_count, |
f942dc25 IC |
1364 | s8 st) |
1365 | { | |
e9ce7cb6 | 1366 | RING_IDX i = queue->tx.rsp_prod_pvt; |
f942dc25 | 1367 | struct xen_netif_tx_response *resp; |
f942dc25 | 1368 | |
e9ce7cb6 | 1369 | resp = RING_GET_RESPONSE(&queue->tx, i); |
f942dc25 IC |
1370 | resp->id = txp->id; |
1371 | resp->status = st; | |
1372 | ||
562abd39 | 1373 | while (extra_count-- != 0) |
e9ce7cb6 | 1374 | RING_GET_RESPONSE(&queue->tx, ++i)->status = XEN_NETIF_RSP_NULL; |
f942dc25 | 1375 | |
e9ce7cb6 | 1376 | queue->tx.rsp_prod_pvt = ++i; |
f942dc25 IC |
1377 | } |
1378 | ||
c8a4d299 DV |
1379 | static void push_tx_responses(struct xenvif_queue *queue) |
1380 | { | |
1381 | int notify; | |
1382 | ||
1383 | RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&queue->tx, notify); | |
1384 | if (notify) | |
1385 | notify_remote_via_irq(queue->tx_irq); | |
1386 | } | |
1387 | ||
e9ce7cb6 | 1388 | void xenvif_idx_unmap(struct xenvif_queue *queue, u16 pending_idx) |
f53c3fe8 ZK |
1389 | { |
1390 | int ret; | |
1391 | struct gnttab_unmap_grant_ref tx_unmap_op; | |
1392 | ||
1393 | gnttab_set_unmap_op(&tx_unmap_op, | |
e9ce7cb6 | 1394 | idx_to_kaddr(queue, pending_idx), |
f53c3fe8 | 1395 | GNTMAP_host_map, |
e9ce7cb6 WL |
1396 | queue->grant_tx_handle[pending_idx]); |
1397 | xenvif_grant_handle_reset(queue, pending_idx); | |
f53c3fe8 ZK |
1398 | |
1399 | ret = gnttab_unmap_refs(&tx_unmap_op, NULL, | |
e9ce7cb6 | 1400 | &queue->mmap_pages[pending_idx], 1); |
7aceb47a | 1401 | if (ret) { |
e9ce7cb6 | 1402 | netdev_err(queue->vif->dev, |
68946159 | 1403 | "Unmap fail: ret: %d pending_idx: %d host_addr: %llx handle: 0x%x status: %d\n", |
7aceb47a ZK |
1404 | ret, |
1405 | pending_idx, | |
1406 | tx_unmap_op.host_addr, | |
1407 | tx_unmap_op.handle, | |
1408 | tx_unmap_op.status); | |
1409 | BUG(); | |
1410 | } | |
f53c3fe8 ZK |
1411 | } |
1412 | ||
e9ce7cb6 | 1413 | static inline int tx_work_todo(struct xenvif_queue *queue) |
f942dc25 | 1414 | { |
e9ce7cb6 | 1415 | if (likely(RING_HAS_UNCONSUMED_REQUESTS(&queue->tx))) |
f942dc25 IC |
1416 | return 1; |
1417 | ||
1418 | return 0; | |
1419 | } | |
1420 | ||
e9ce7cb6 | 1421 | static inline bool tx_dealloc_work_todo(struct xenvif_queue *queue) |
f53c3fe8 | 1422 | { |
e9ce7cb6 | 1423 | return queue->dealloc_cons != queue->dealloc_prod; |
f53c3fe8 ZK |
1424 | } |
1425 | ||
4e15ee2c | 1426 | void xenvif_unmap_frontend_data_rings(struct xenvif_queue *queue) |
f942dc25 | 1427 | { |
e9ce7cb6 WL |
1428 | if (queue->tx.sring) |
1429 | xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(queue->vif), | |
1430 | queue->tx.sring); | |
1431 | if (queue->rx.sring) | |
1432 | xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(queue->vif), | |
1433 | queue->rx.sring); | |
f942dc25 IC |
1434 | } |
1435 | ||
4e15ee2c PD |
1436 | int xenvif_map_frontend_data_rings(struct xenvif_queue *queue, |
1437 | grant_ref_t tx_ring_ref, | |
1438 | grant_ref_t rx_ring_ref) | |
f942dc25 | 1439 | { |
c9d63699 | 1440 | void *addr; |
f942dc25 IC |
1441 | struct xen_netif_tx_sring *txs; |
1442 | struct xen_netif_rx_sring *rxs; | |
1443 | ||
1444 | int err = -ENOMEM; | |
1445 | ||
e9ce7cb6 | 1446 | err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(queue->vif), |
ccc9d90a | 1447 | &tx_ring_ref, 1, &addr); |
c9d63699 | 1448 | if (err) |
f942dc25 IC |
1449 | goto err; |
1450 | ||
c9d63699 | 1451 | txs = (struct xen_netif_tx_sring *)addr; |
d0089e8a | 1452 | BACK_RING_INIT(&queue->tx, txs, XEN_PAGE_SIZE); |
f942dc25 | 1453 | |
e9ce7cb6 | 1454 | err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(queue->vif), |
ccc9d90a | 1455 | &rx_ring_ref, 1, &addr); |
c9d63699 | 1456 | if (err) |
f942dc25 | 1457 | goto err; |
f942dc25 | 1458 | |
c9d63699 | 1459 | rxs = (struct xen_netif_rx_sring *)addr; |
d0089e8a | 1460 | BACK_RING_INIT(&queue->rx, rxs, XEN_PAGE_SIZE); |
f942dc25 IC |
1461 | |
1462 | return 0; | |
1463 | ||
1464 | err: | |
4e15ee2c | 1465 | xenvif_unmap_frontend_data_rings(queue); |
f942dc25 IC |
1466 | return err; |
1467 | } | |
1468 | ||
a64bd934 WL |
1469 | static bool xenvif_dealloc_kthread_should_stop(struct xenvif_queue *queue) |
1470 | { | |
1471 | /* Dealloc thread must remain running until all inflight | |
1472 | * packets complete. | |
1473 | */ | |
1474 | return kthread_should_stop() && | |
1475 | !atomic_read(&queue->inflight_packets); | |
1476 | } | |
1477 | ||
f53c3fe8 ZK |
1478 | int xenvif_dealloc_kthread(void *data) |
1479 | { | |
e9ce7cb6 | 1480 | struct xenvif_queue *queue = data; |
f53c3fe8 | 1481 | |
a64bd934 | 1482 | for (;;) { |
e9ce7cb6 WL |
1483 | wait_event_interruptible(queue->dealloc_wq, |
1484 | tx_dealloc_work_todo(queue) || | |
a64bd934 WL |
1485 | xenvif_dealloc_kthread_should_stop(queue)); |
1486 | if (xenvif_dealloc_kthread_should_stop(queue)) | |
f53c3fe8 ZK |
1487 | break; |
1488 | ||
e9ce7cb6 | 1489 | xenvif_tx_dealloc_action(queue); |
f53c3fe8 ZK |
1490 | cond_resched(); |
1491 | } | |
1492 | ||
1493 | /* Unmap anything remaining*/ | |
e9ce7cb6 WL |
1494 | if (tx_dealloc_work_todo(queue)) |
1495 | xenvif_tx_dealloc_action(queue); | |
f53c3fe8 ZK |
1496 | |
1497 | return 0; | |
1498 | } | |
1499 | ||
4e15ee2c PD |
1500 | static void make_ctrl_response(struct xenvif *vif, |
1501 | const struct xen_netif_ctrl_request *req, | |
1502 | u32 status, u32 data) | |
1503 | { | |
1504 | RING_IDX idx = vif->ctrl.rsp_prod_pvt; | |
1505 | struct xen_netif_ctrl_response rsp = { | |
1506 | .id = req->id, | |
1507 | .type = req->type, | |
1508 | .status = status, | |
1509 | .data = data, | |
1510 | }; | |
1511 | ||
1512 | *RING_GET_RESPONSE(&vif->ctrl, idx) = rsp; | |
1513 | vif->ctrl.rsp_prod_pvt = ++idx; | |
1514 | } | |
1515 | ||
1516 | static void push_ctrl_response(struct xenvif *vif) | |
1517 | { | |
1518 | int notify; | |
1519 | ||
1520 | RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&vif->ctrl, notify); | |
1521 | if (notify) | |
1522 | notify_remote_via_irq(vif->ctrl_irq); | |
1523 | } | |
1524 | ||
1525 | static void process_ctrl_request(struct xenvif *vif, | |
1526 | const struct xen_netif_ctrl_request *req) | |
1527 | { | |
40d8abde PD |
1528 | u32 status = XEN_NETIF_CTRL_STATUS_NOT_SUPPORTED; |
1529 | u32 data = 0; | |
1530 | ||
1531 | switch (req->type) { | |
1532 | case XEN_NETIF_CTRL_TYPE_SET_HASH_ALGORITHM: | |
1533 | status = xenvif_set_hash_alg(vif, req->data[0]); | |
1534 | break; | |
1535 | ||
1536 | case XEN_NETIF_CTRL_TYPE_GET_HASH_FLAGS: | |
1537 | status = xenvif_get_hash_flags(vif, &data); | |
1538 | break; | |
1539 | ||
1540 | case XEN_NETIF_CTRL_TYPE_SET_HASH_FLAGS: | |
1541 | status = xenvif_set_hash_flags(vif, req->data[0]); | |
1542 | break; | |
1543 | ||
1544 | case XEN_NETIF_CTRL_TYPE_SET_HASH_KEY: | |
1545 | status = xenvif_set_hash_key(vif, req->data[0], | |
1546 | req->data[1]); | |
1547 | break; | |
1548 | ||
1549 | case XEN_NETIF_CTRL_TYPE_GET_HASH_MAPPING_SIZE: | |
1550 | status = XEN_NETIF_CTRL_STATUS_SUCCESS; | |
1551 | data = XEN_NETBK_MAX_HASH_MAPPING_SIZE; | |
1552 | break; | |
1553 | ||
1554 | case XEN_NETIF_CTRL_TYPE_SET_HASH_MAPPING_SIZE: | |
1555 | status = xenvif_set_hash_mapping_size(vif, | |
1556 | req->data[0]); | |
1557 | break; | |
1558 | ||
1559 | case XEN_NETIF_CTRL_TYPE_SET_HASH_MAPPING: | |
1560 | status = xenvif_set_hash_mapping(vif, req->data[0], | |
1561 | req->data[1], | |
1562 | req->data[2]); | |
1563 | break; | |
1564 | ||
1565 | default: | |
1566 | break; | |
1567 | } | |
1568 | ||
1569 | make_ctrl_response(vif, req, status, data); | |
4e15ee2c PD |
1570 | push_ctrl_response(vif); |
1571 | } | |
1572 | ||
1573 | static void xenvif_ctrl_action(struct xenvif *vif) | |
1574 | { | |
1575 | for (;;) { | |
1576 | RING_IDX req_prod, req_cons; | |
1577 | ||
1578 | req_prod = vif->ctrl.sring->req_prod; | |
1579 | req_cons = vif->ctrl.req_cons; | |
1580 | ||
1581 | /* Make sure we can see requests before we process them. */ | |
1582 | rmb(); | |
1583 | ||
1584 | if (req_cons == req_prod) | |
1585 | break; | |
1586 | ||
1587 | while (req_cons != req_prod) { | |
1588 | struct xen_netif_ctrl_request req; | |
1589 | ||
1590 | RING_COPY_REQUEST(&vif->ctrl, req_cons, &req); | |
1591 | req_cons++; | |
1592 | ||
1593 | process_ctrl_request(vif, &req); | |
1594 | } | |
1595 | ||
1596 | vif->ctrl.req_cons = req_cons; | |
1597 | vif->ctrl.sring->req_event = req_cons + 1; | |
1598 | } | |
1599 | } | |
1600 | ||
1601 | static bool xenvif_ctrl_work_todo(struct xenvif *vif) | |
1602 | { | |
1603 | if (likely(RING_HAS_UNCONSUMED_REQUESTS(&vif->ctrl))) | |
1604 | return 1; | |
1605 | ||
1606 | return 0; | |
1607 | } | |
1608 | ||
0364a882 | 1609 | irqreturn_t xenvif_ctrl_irq_fn(int irq, void *data) |
4e15ee2c PD |
1610 | { |
1611 | struct xenvif *vif = data; | |
1612 | ||
0364a882 JG |
1613 | while (xenvif_ctrl_work_todo(vif)) |
1614 | xenvif_ctrl_action(vif); | |
4e15ee2c | 1615 | |
0364a882 | 1616 | return IRQ_HANDLED; |
4e15ee2c PD |
1617 | } |
1618 | ||
f942dc25 IC |
1619 | static int __init netback_init(void) |
1620 | { | |
f942dc25 | 1621 | int rc = 0; |
f942dc25 | 1622 | |
2a14b244 | 1623 | if (!xen_domain()) |
f942dc25 IC |
1624 | return -ENODEV; |
1625 | ||
56dd5af9 | 1626 | /* Allow as many queues as there are CPUs but max. 8 if user has not |
4c82ac3c WL |
1627 | * specified a value. |
1628 | */ | |
1629 | if (xenvif_max_queues == 0) | |
56dd5af9 JG |
1630 | xenvif_max_queues = min_t(unsigned int, MAX_QUEUES_DEFAULT, |
1631 | num_online_cpus()); | |
8d3d53b3 | 1632 | |
37641494 | 1633 | if (fatal_skb_slots < XEN_NETBK_LEGACY_SLOTS_MAX) { |
383eda32 JP |
1634 | pr_info("fatal_skb_slots too small (%d), bump it to XEN_NETBK_LEGACY_SLOTS_MAX (%d)\n", |
1635 | fatal_skb_slots, XEN_NETBK_LEGACY_SLOTS_MAX); | |
37641494 | 1636 | fatal_skb_slots = XEN_NETBK_LEGACY_SLOTS_MAX; |
2810e5b9 WL |
1637 | } |
1638 | ||
f942dc25 IC |
1639 | rc = xenvif_xenbus_init(); |
1640 | if (rc) | |
1641 | goto failed_init; | |
1642 | ||
f51de243 ZK |
1643 | #ifdef CONFIG_DEBUG_FS |
1644 | xen_netback_dbg_root = debugfs_create_dir("xen-netback", NULL); | |
1645 | if (IS_ERR_OR_NULL(xen_netback_dbg_root)) | |
1646 | pr_warn("Init of debugfs returned %ld!\n", | |
1647 | PTR_ERR(xen_netback_dbg_root)); | |
1648 | #endif /* CONFIG_DEBUG_FS */ | |
1649 | ||
f942dc25 IC |
1650 | return 0; |
1651 | ||
1652 | failed_init: | |
f942dc25 | 1653 | return rc; |
f942dc25 IC |
1654 | } |
1655 | ||
1656 | module_init(netback_init); | |
1657 | ||
b103f358 WL |
1658 | static void __exit netback_fini(void) |
1659 | { | |
f51de243 ZK |
1660 | #ifdef CONFIG_DEBUG_FS |
1661 | if (!IS_ERR_OR_NULL(xen_netback_dbg_root)) | |
1662 | debugfs_remove_recursive(xen_netback_dbg_root); | |
1663 | #endif /* CONFIG_DEBUG_FS */ | |
b103f358 | 1664 | xenvif_xenbus_fini(); |
b103f358 WL |
1665 | } |
1666 | module_exit(netback_fini); | |
1667 | ||
f942dc25 | 1668 | MODULE_LICENSE("Dual BSD/GPL"); |
f984cec6 | 1669 | MODULE_ALIAS("xen-backend:vif"); |