?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
x86/ldt: Further fix FPU emulation
[linux/fpc-iii.git] / drivers / net / xen-netback / netback.c
blob7d50711476fe1e88debca95beb790d770261f036
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>
40 #include <linux/highmem.h>
41
42 #include <net/tcp.h>
43
44 #include <xen/xen.h>
45 #include <xen/events.h>
46 #include <xen/interface/memory.h>
47 #include <xen/page.h>
48
49 #include <asm/xen/hypercall.h>
50
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 = true;
56 module_param(separate_tx_rx_irq, bool, 0644);
57
58 /* The time that packets can stay on the guest Rx internal queue
59 * before they are dropped.
60 */
61 unsigned int rx_drain_timeout_msecs = 10000;
62 module_param(rx_drain_timeout_msecs, uint, 0444);
63
64 /* The length of time before the frontend is considered unresponsive
65 * because it isn't providing Rx slots.
66 */
67 unsigned int rx_stall_timeout_msecs = 60000;
68 module_param(rx_stall_timeout_msecs, uint, 0444);
69
70 unsigned int xenvif_max_queues;
71 module_param_named(max_queues, xenvif_max_queues, uint, 0644);
72 MODULE_PARM_DESC(max_queues,
73 "Maximum number of queues per virtual interface");
74
75 /*
76 * This is the maximum slots a skb can have. If a guest sends a skb
77 * which exceeds this limit it is considered malicious.
78 */
79 #define FATAL_SKB_SLOTS_DEFAULT 20
80 static unsigned int fatal_skb_slots = FATAL_SKB_SLOTS_DEFAULT;
81 module_param(fatal_skb_slots, uint, 0444);
82
83 /* The amount to copy out of the first guest Tx slot into the skb's
84 * linear area. If the first slot has more data, it will be mapped
85 * and put into the first frag.
86 *
87 * This is sized to avoid pulling headers from the frags for most
88 * TCP/IP packets.
89 */
90 #define XEN_NETBACK_TX_COPY_LEN 128
91
92
93 static void xenvif_idx_release(struct xenvif_queue *queue, u16 pending_idx,
94 u8 status);
95
96 static void make_tx_response(struct xenvif_queue *queue,
97 struct xen_netif_tx_request *txp,
98 s8 st);
99 static void push_tx_responses(struct xenvif_queue *queue);
100
101 static inline int tx_work_todo(struct xenvif_queue *queue);
102
103 static struct xen_netif_rx_response *make_rx_response(struct xenvif_queue *queue,
104 u16 id,
105 s8 st,
106 u16 offset,
107 u16 size,
108 u16 flags);
109
110 static inline unsigned long idx_to_pfn(struct xenvif_queue *queue,
111 u16 idx)
112 {
113 return page_to_pfn(queue->mmap_pages[idx]);
114 }
115
116 static inline unsigned long idx_to_kaddr(struct xenvif_queue *queue,
117 u16 idx)
118 {
119 return (unsigned long)pfn_to_kaddr(idx_to_pfn(queue, idx));
120 }
121
122 #define callback_param(vif, pending_idx) \
123 (vif->pending_tx_info[pending_idx].callback_struct)
124
125 /* Find the containing VIF's structure from a pointer in pending_tx_info array
126 */
127 static inline struct xenvif_queue *ubuf_to_queue(const struct ubuf_info *ubuf)
128 {
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,
133 struct xenvif_queue,
134 pending_tx_info[0]);
135 }
136
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
147 static inline pending_ring_idx_t pending_index(unsigned i)
148 {
149 return i & (MAX_PENDING_REQS-1);
150 }
151
152 bool xenvif_rx_ring_slots_available(struct xenvif_queue *queue, int needed)
153 {
154 RING_IDX prod, cons;
155
156 do {
157 prod = queue->rx.sring->req_prod;
158 cons = queue->rx.req_cons;
159
160 if (prod - cons >= needed)
161 return true;
162
163 queue->rx.sring->req_event = prod + 1;
164
165 /* Make sure event is visible before we check prod
166 * again.
167 */
168 mb();
169 } while (queue->rx.sring->req_prod != prod);
170
171 return false;
172 }
173
174 void xenvif_rx_queue_tail(struct xenvif_queue *queue, struct sk_buff *skb)
175 {
176 unsigned long flags;
177
178 spin_lock_irqsave(&queue->rx_queue.lock, flags);
179
180 __skb_queue_tail(&queue->rx_queue, skb);
181
182 queue->rx_queue_len += skb->len;
183 if (queue->rx_queue_len > queue->rx_queue_max)
184 netif_tx_stop_queue(netdev_get_tx_queue(queue->vif->dev, queue->id));
185
186 spin_unlock_irqrestore(&queue->rx_queue.lock, flags);
187 }
188
189 static struct sk_buff *xenvif_rx_dequeue(struct xenvif_queue *queue)
190 {
191 struct sk_buff *skb;
192
193 spin_lock_irq(&queue->rx_queue.lock);
194
195 skb = __skb_dequeue(&queue->rx_queue);
196 if (skb)
197 queue->rx_queue_len -= skb->len;
198
199 spin_unlock_irq(&queue->rx_queue.lock);
200
201 return skb;
202 }
203
204 static void xenvif_rx_queue_maybe_wake(struct xenvif_queue *queue)
205 {
206 spin_lock_irq(&queue->rx_queue.lock);
207
208 if (queue->rx_queue_len < queue->rx_queue_max)
209 netif_tx_wake_queue(netdev_get_tx_queue(queue->vif->dev, queue->id));
210
211 spin_unlock_irq(&queue->rx_queue.lock);
212 }
213
214
215 static void xenvif_rx_queue_purge(struct xenvif_queue *queue)
216 {
217 struct sk_buff *skb;
218 while ((skb = xenvif_rx_dequeue(queue)) != NULL)
219 kfree_skb(skb);
220 }
221
222 static void xenvif_rx_queue_drop_expired(struct xenvif_queue *queue)
223 {
224 struct sk_buff *skb;
225
226 for(;;) {
227 skb = skb_peek(&queue->rx_queue);
228 if (!skb)
229 break;
230 if (time_before(jiffies, XENVIF_RX_CB(skb)->expires))
231 break;
232 xenvif_rx_dequeue(queue);
233 kfree_skb(skb);
234 }
235 }
236
237 struct netrx_pending_operations {
238 unsigned copy_prod, copy_cons;
239 unsigned meta_prod, meta_cons;
240 struct gnttab_copy *copy;
241 struct xenvif_rx_meta *meta;
242 int copy_off;
243 grant_ref_t copy_gref;
244 };
245
246 static struct xenvif_rx_meta *get_next_rx_buffer(struct xenvif_queue *queue,
247 struct netrx_pending_operations *npo)
248 {
249 struct xenvif_rx_meta *meta;
250 struct xen_netif_rx_request *req;
251
252 req = RING_GET_REQUEST(&queue->rx, queue->rx.req_cons++);
253
254 meta = npo->meta + npo->meta_prod++;
255 meta->gso_type = XEN_NETIF_GSO_TYPE_NONE;
256 meta->gso_size = 0;
257 meta->size = 0;
258 meta->id = req->id;
259
260 npo->copy_off = 0;
261 npo->copy_gref = req->gref;
262
263 return meta;
264 }
265
266 /*
267 * Set up the grant operations for this fragment. If it's a flipping
268 * interface, we also set up the unmap request from here.
269 */
270 static void xenvif_gop_frag_copy(struct xenvif_queue *queue, struct sk_buff *skb,
271 struct netrx_pending_operations *npo,
272 struct page *page, unsigned long size,
273 unsigned long offset, int *head)
274 {
275 struct gnttab_copy *copy_gop;
276 struct xenvif_rx_meta *meta;
277 unsigned long bytes;
278 int gso_type = XEN_NETIF_GSO_TYPE_NONE;
279
280 /* Data must not cross a page boundary. */
281 BUG_ON(size + offset > PAGE_SIZE<<compound_order(page));
282
283 meta = npo->meta + npo->meta_prod - 1;
284
285 /* Skip unused frames from start of page */
286 page += offset >> PAGE_SHIFT;
287 offset &= ~PAGE_MASK;
288
289 while (size > 0) {
290 struct xen_page_foreign *foreign;
291
292 BUG_ON(offset >= PAGE_SIZE);
293 BUG_ON(npo->copy_off > MAX_BUFFER_OFFSET);
294
295 if (npo->copy_off == MAX_BUFFER_OFFSET)
296 meta = get_next_rx_buffer(queue, npo);
297
298 bytes = PAGE_SIZE - offset;
299 if (bytes > size)
300 bytes = size;
301
302 if (npo->copy_off + bytes > MAX_BUFFER_OFFSET)
303 bytes = MAX_BUFFER_OFFSET - npo->copy_off;
304
305 copy_gop = npo->copy + npo->copy_prod++;
306 copy_gop->flags = GNTCOPY_dest_gref;
307 copy_gop->len = bytes;
308
309 foreign = xen_page_foreign(page);
310 if (foreign) {
311 copy_gop->source.domid = foreign->domid;
312 copy_gop->source.u.ref = foreign->gref;
313 copy_gop->flags |= GNTCOPY_source_gref;
314 } else {
315 copy_gop->source.domid = DOMID_SELF;
316 copy_gop->source.u.gmfn =
317 virt_to_mfn(page_address(page));
318 }
319 copy_gop->source.offset = offset;
320
321 copy_gop->dest.domid = queue->vif->domid;
322 copy_gop->dest.offset = npo->copy_off;
323 copy_gop->dest.u.ref = npo->copy_gref;
324
325 npo->copy_off += bytes;
326 meta->size += bytes;
327
328 offset += bytes;
329 size -= bytes;
330
331 /* Next frame */
332 if (offset == PAGE_SIZE && size) {
333 BUG_ON(!PageCompound(page));
334 page++;
335 offset = 0;
336 }
337
338 /* Leave a gap for the GSO descriptor. */
339 if (skb_is_gso(skb)) {
340 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
341 gso_type = XEN_NETIF_GSO_TYPE_TCPV4;
342 else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
343 gso_type = XEN_NETIF_GSO_TYPE_TCPV6;
344 }
345
346 if (*head && ((1 << gso_type) & queue->vif->gso_mask))
347 queue->rx.req_cons++;
348
349 *head = 0; /* There must be something in this buffer now. */
350
351 }
352 }
353
354 /*
355 * Prepare an SKB to be transmitted to the frontend.
356 *
357 * This function is responsible for allocating grant operations, meta
358 * structures, etc.
359 *
360 * It returns the number of meta structures consumed. The number of
361 * ring slots used is always equal to the number of meta slots used
362 * plus the number of GSO descriptors used. Currently, we use either
363 * zero GSO descriptors (for non-GSO packets) or one descriptor (for
364 * frontend-side LRO).
365 */
366 static int xenvif_gop_skb(struct sk_buff *skb,
367 struct netrx_pending_operations *npo,
368 struct xenvif_queue *queue)
369 {
370 struct xenvif *vif = netdev_priv(skb->dev);
371 int nr_frags = skb_shinfo(skb)->nr_frags;
372 int i;
373 struct xen_netif_rx_request *req;
374 struct xenvif_rx_meta *meta;
375 unsigned char *data;
376 int head = 1;
377 int old_meta_prod;
378 int gso_type;
379
380 old_meta_prod = npo->meta_prod;
381
382 gso_type = XEN_NETIF_GSO_TYPE_NONE;
383 if (skb_is_gso(skb)) {
384 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
385 gso_type = XEN_NETIF_GSO_TYPE_TCPV4;
386 else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
387 gso_type = XEN_NETIF_GSO_TYPE_TCPV6;
388 }
389
390 /* Set up a GSO prefix descriptor, if necessary */
391 if ((1 << gso_type) & vif->gso_prefix_mask) {
392 req = RING_GET_REQUEST(&queue->rx, queue->rx.req_cons++);
393 meta = npo->meta + npo->meta_prod++;
394 meta->gso_type = gso_type;
395 meta->gso_size = skb_shinfo(skb)->gso_size;
396 meta->size = 0;
397 meta->id = req->id;
398 }
399
400 req = RING_GET_REQUEST(&queue->rx, queue->rx.req_cons++);
401 meta = npo->meta + npo->meta_prod++;
402
403 if ((1 << gso_type) & vif->gso_mask) {
404 meta->gso_type = gso_type;
405 meta->gso_size = skb_shinfo(skb)->gso_size;
406 } else {
407 meta->gso_type = XEN_NETIF_GSO_TYPE_NONE;
408 meta->gso_size = 0;
409 }
410
411 meta->size = 0;
412 meta->id = req->id;
413 npo->copy_off = 0;
414 npo->copy_gref = req->gref;
415
416 data = skb->data;
417 while (data < skb_tail_pointer(skb)) {
418 unsigned int offset = offset_in_page(data);
419 unsigned int len = PAGE_SIZE - offset;
420
421 if (data + len > skb_tail_pointer(skb))
422 len = skb_tail_pointer(skb) - data;
423
424 xenvif_gop_frag_copy(queue, skb, npo,
425 virt_to_page(data), len, offset, &head);
426 data += len;
427 }
428
429 for (i = 0; i < nr_frags; i++) {
430 xenvif_gop_frag_copy(queue, skb, npo,
431 skb_frag_page(&skb_shinfo(skb)->frags[i]),
432 skb_frag_size(&skb_shinfo(skb)->frags[i]),
433 skb_shinfo(skb)->frags[i].page_offset,
434 &head);
435 }
436
437 return npo->meta_prod - old_meta_prod;
438 }
439
440 /*
441 * This is a twin to xenvif_gop_skb. Assume that xenvif_gop_skb was
442 * used to set up the operations on the top of
443 * netrx_pending_operations, which have since been done. Check that
444 * they didn't give any errors and advance over them.
445 */
446 static int xenvif_check_gop(struct xenvif *vif, int nr_meta_slots,
447 struct netrx_pending_operations *npo)
448 {
449 struct gnttab_copy *copy_op;
450 int status = XEN_NETIF_RSP_OKAY;
451 int i;
452
453 for (i = 0; i < nr_meta_slots; i++) {
454 copy_op = npo->copy + npo->copy_cons++;
455 if (copy_op->status != GNTST_okay) {
456 netdev_dbg(vif->dev,
457 "Bad status %d from copy to DOM%d.\n",
458 copy_op->status, vif->domid);
459 status = XEN_NETIF_RSP_ERROR;
460 }
461 }
462
463 return status;
464 }
465
466 static void xenvif_add_frag_responses(struct xenvif_queue *queue, int status,
467 struct xenvif_rx_meta *meta,
468 int nr_meta_slots)
469 {
470 int i;
471 unsigned long offset;
472
473 /* No fragments used */
474 if (nr_meta_slots <= 1)
475 return;
476
477 nr_meta_slots--;
478
479 for (i = 0; i < nr_meta_slots; i++) {
480 int flags;
481 if (i == nr_meta_slots - 1)
482 flags = 0;
483 else
484 flags = XEN_NETRXF_more_data;
485
486 offset = 0;
487 make_rx_response(queue, meta[i].id, status, offset,
488 meta[i].size, flags);
489 }
490 }
491
492 void xenvif_kick_thread(struct xenvif_queue *queue)
493 {
494 wake_up(&queue->wq);
495 }
496
497 static void xenvif_rx_action(struct xenvif_queue *queue)
498 {
499 s8 status;
500 u16 flags;
501 struct xen_netif_rx_response *resp;
502 struct sk_buff_head rxq;
503 struct sk_buff *skb;
504 LIST_HEAD(notify);
505 int ret;
506 unsigned long offset;
507 bool need_to_notify = false;
508
509 struct netrx_pending_operations npo = {
510 .copy = queue->grant_copy_op,
511 .meta = queue->meta,
512 };
513
514 skb_queue_head_init(&rxq);
515
516 while (xenvif_rx_ring_slots_available(queue, XEN_NETBK_RX_SLOTS_MAX)
517 && (skb = xenvif_rx_dequeue(queue)) != NULL) {
518 queue->last_rx_time = jiffies;
519
520 XENVIF_RX_CB(skb)->meta_slots_used = xenvif_gop_skb(skb, &npo, queue);
521
522 __skb_queue_tail(&rxq, skb);
523 }
524
525 BUG_ON(npo.meta_prod > ARRAY_SIZE(queue->meta));
526
527 if (!npo.copy_prod)
528 goto done;
529
530 BUG_ON(npo.copy_prod > MAX_GRANT_COPY_OPS);
531 gnttab_batch_copy(queue->grant_copy_op, npo.copy_prod);
532
533 while ((skb = __skb_dequeue(&rxq)) != NULL) {
534
535 if ((1 << queue->meta[npo.meta_cons].gso_type) &
536 queue->vif->gso_prefix_mask) {
537 resp = RING_GET_RESPONSE(&queue->rx,
538 queue->rx.rsp_prod_pvt++);
539
540 resp->flags = XEN_NETRXF_gso_prefix | XEN_NETRXF_more_data;
541
542 resp->offset = queue->meta[npo.meta_cons].gso_size;
543 resp->id = queue->meta[npo.meta_cons].id;
544 resp->status = XENVIF_RX_CB(skb)->meta_slots_used;
545
546 npo.meta_cons++;
547 XENVIF_RX_CB(skb)->meta_slots_used--;
548 }
549
550
551 queue->stats.tx_bytes += skb->len;
552 queue->stats.tx_packets++;
553
554 status = xenvif_check_gop(queue->vif,
555 XENVIF_RX_CB(skb)->meta_slots_used,
556 &npo);
557
558 if (XENVIF_RX_CB(skb)->meta_slots_used == 1)
559 flags = 0;
560 else
561 flags = XEN_NETRXF_more_data;
562
563 if (skb->ip_summed == CHECKSUM_PARTIAL) /* local packet? */
564 flags |= XEN_NETRXF_csum_blank | XEN_NETRXF_data_validated;
565 else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
566 /* remote but checksummed. */
567 flags |= XEN_NETRXF_data_validated;
568
569 offset = 0;
570 resp = make_rx_response(queue, queue->meta[npo.meta_cons].id,
571 status, offset,
572 queue->meta[npo.meta_cons].size,
573 flags);
574
575 if ((1 << queue->meta[npo.meta_cons].gso_type) &
576 queue->vif->gso_mask) {
577 struct xen_netif_extra_info *gso =
578 (struct xen_netif_extra_info *)
579 RING_GET_RESPONSE(&queue->rx,
580 queue->rx.rsp_prod_pvt++);
581
582 resp->flags |= XEN_NETRXF_extra_info;
583
584 gso->u.gso.type = queue->meta[npo.meta_cons].gso_type;
585 gso->u.gso.size = queue->meta[npo.meta_cons].gso_size;
586 gso->u.gso.pad = 0;
587 gso->u.gso.features = 0;
588
589 gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
590 gso->flags = 0;
591 }
592
593 xenvif_add_frag_responses(queue, status,
594 queue->meta + npo.meta_cons + 1,
595 XENVIF_RX_CB(skb)->meta_slots_used);
596
597 RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&queue->rx, ret);
598
599 need_to_notify |= !!ret;
600
601 npo.meta_cons += XENVIF_RX_CB(skb)->meta_slots_used;
602 dev_kfree_skb(skb);
603 }
604
605 done:
606 if (need_to_notify)
607 notify_remote_via_irq(queue->rx_irq);
608 }
609
610 void xenvif_napi_schedule_or_enable_events(struct xenvif_queue *queue)
611 {
612 int more_to_do;
613
614 RING_FINAL_CHECK_FOR_REQUESTS(&queue->tx, more_to_do);
615
616 if (more_to_do)
617 napi_schedule(&queue->napi);
618 }
619
620 static void tx_add_credit(struct xenvif_queue *queue)
621 {
622 unsigned long max_burst, max_credit;
623
624 /*
625 * Allow a burst big enough to transmit a jumbo packet of up to 128kB.
626 * Otherwise the interface can seize up due to insufficient credit.
627 */
628 max_burst = RING_GET_REQUEST(&queue->tx, queue->tx.req_cons)->size;
629 max_burst = min(max_burst, 131072UL);
630 max_burst = max(max_burst, queue->credit_bytes);
631
632 /* Take care that adding a new chunk of credit doesn't wrap to zero. */
633 max_credit = queue->remaining_credit + queue->credit_bytes;
634 if (max_credit < queue->remaining_credit)
635 max_credit = ULONG_MAX; /* wrapped: clamp to ULONG_MAX */
636
637 queue->remaining_credit = min(max_credit, max_burst);
638 }
639
640 void xenvif_tx_credit_callback(unsigned long data)
641 {
642 struct xenvif_queue *queue = (struct xenvif_queue *)data;
643 tx_add_credit(queue);
644 xenvif_napi_schedule_or_enable_events(queue);
645 }
646
647 static void xenvif_tx_err(struct xenvif_queue *queue,
648 struct xen_netif_tx_request *txp, RING_IDX end)
649 {
650 RING_IDX cons = queue->tx.req_cons;
651 unsigned long flags;
652
653 do {
654 spin_lock_irqsave(&queue->response_lock, flags);
655 make_tx_response(queue, txp, XEN_NETIF_RSP_ERROR);
656 push_tx_responses(queue);
657 spin_unlock_irqrestore(&queue->response_lock, flags);
658 if (cons == end)
659 break;
660 txp = RING_GET_REQUEST(&queue->tx, cons++);
661 } while (1);
662 queue->tx.req_cons = cons;
663 }
664
665 static void xenvif_fatal_tx_err(struct xenvif *vif)
666 {
667 netdev_err(vif->dev, "fatal error; disabling device\n");
668 vif->disabled = true;
669 /* Disable the vif from queue 0's kthread */
670 if (vif->queues)
671 xenvif_kick_thread(&vif->queues[0]);
672 }
673
674 static int xenvif_count_requests(struct xenvif_queue *queue,
675 struct xen_netif_tx_request *first,
676 struct xen_netif_tx_request *txp,
677 int work_to_do)
678 {
679 RING_IDX cons = queue->tx.req_cons;
680 int slots = 0;
681 int drop_err = 0;
682 int more_data;
683
684 if (!(first->flags & XEN_NETTXF_more_data))
685 return 0;
686
687 do {
688 struct xen_netif_tx_request dropped_tx = { 0 };
689
690 if (slots >= work_to_do) {
691 netdev_err(queue->vif->dev,
692 "Asked for %d slots but exceeds this limit\n",
693 work_to_do);
694 xenvif_fatal_tx_err(queue->vif);
695 return -ENODATA;
696 }
697
698 /* This guest is really using too many slots and
699 * considered malicious.
700 */
701 if (unlikely(slots >= fatal_skb_slots)) {
702 netdev_err(queue->vif->dev,
703 "Malicious frontend using %d slots, threshold %u\n",
704 slots, fatal_skb_slots);
705 xenvif_fatal_tx_err(queue->vif);
706 return -E2BIG;
707 }
708
709 /* Xen network protocol had implicit dependency on
710 * MAX_SKB_FRAGS. XEN_NETBK_LEGACY_SLOTS_MAX is set to
711 * the historical MAX_SKB_FRAGS value 18 to honor the
712 * same behavior as before. Any packet using more than
713 * 18 slots but less than fatal_skb_slots slots is
714 * dropped
715 */
716 if (!drop_err && slots >= XEN_NETBK_LEGACY_SLOTS_MAX) {
717 if (net_ratelimit())
718 netdev_dbg(queue->vif->dev,
719 "Too many slots (%d) exceeding limit (%d), dropping packet\n",
720 slots, XEN_NETBK_LEGACY_SLOTS_MAX);
721 drop_err = -E2BIG;
722 }
723
724 if (drop_err)
725 txp = &dropped_tx;
726
727 memcpy(txp, RING_GET_REQUEST(&queue->tx, cons + slots),
728 sizeof(*txp));
729
730 /* If the guest submitted a frame >= 64 KiB then
731 * first->size overflowed and following slots will
732 * appear to be larger than the frame.
733 *
734 * This cannot be fatal error as there are buggy
735 * frontends that do this.
736 *
737 * Consume all slots and drop the packet.
738 */
739 if (!drop_err && txp->size > first->size) {
740 if (net_ratelimit())
741 netdev_dbg(queue->vif->dev,
742 "Invalid tx request, slot size %u > remaining size %u\n",
743 txp->size, first->size);
744 drop_err = -EIO;
745 }
746
747 first->size -= txp->size;
748 slots++;
749
750 if (unlikely((txp->offset + txp->size) > PAGE_SIZE)) {
751 netdev_err(queue->vif->dev, "Cross page boundary, txp->offset: %u, size: %u\n",
752 txp->offset, txp->size);
753 xenvif_fatal_tx_err(queue->vif);
754 return -EINVAL;
755 }
756
757 more_data = txp->flags & XEN_NETTXF_more_data;
758
759 if (!drop_err)
760 txp++;
761
762 } while (more_data);
763
764 if (drop_err) {
765 xenvif_tx_err(queue, first, cons + slots);
766 return drop_err;
767 }
768
769 return slots;
770 }
771
772
773 struct xenvif_tx_cb {
774 u16 pending_idx;
775 };
776
777 #define XENVIF_TX_CB(skb) ((struct xenvif_tx_cb *)(skb)->cb)
778
779 static inline void xenvif_tx_create_map_op(struct xenvif_queue *queue,
780 u16 pending_idx,
781 struct xen_netif_tx_request *txp,
782 struct gnttab_map_grant_ref *mop)
783 {
784 queue->pages_to_map[mop-queue->tx_map_ops] = queue->mmap_pages[pending_idx];
785 gnttab_set_map_op(mop, idx_to_kaddr(queue, pending_idx),
786 GNTMAP_host_map | GNTMAP_readonly,
787 txp->gref, queue->vif->domid);
788
789 memcpy(&queue->pending_tx_info[pending_idx].req, txp,
790 sizeof(*txp));
791 }
792
793 static inline struct sk_buff *xenvif_alloc_skb(unsigned int size)
794 {
795 struct sk_buff *skb =
796 alloc_skb(size + NET_SKB_PAD + NET_IP_ALIGN,
797 GFP_ATOMIC | __GFP_NOWARN);
798 if (unlikely(skb == NULL))
799 return NULL;
800
801 /* Packets passed to netif_rx() must have some headroom. */
802 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
803
804 /* Initialize it here to avoid later surprises */
805 skb_shinfo(skb)->destructor_arg = NULL;
806
807 return skb;
808 }
809
810 static struct gnttab_map_grant_ref *xenvif_get_requests(struct xenvif_queue *queue,
811 struct sk_buff *skb,
812 struct xen_netif_tx_request *txp,
813 struct gnttab_map_grant_ref *gop)
814 {
815 struct skb_shared_info *shinfo = skb_shinfo(skb);
816 skb_frag_t *frags = shinfo->frags;
817 u16 pending_idx = XENVIF_TX_CB(skb)->pending_idx;
818 int start;
819 pending_ring_idx_t index;
820 unsigned int nr_slots, frag_overflow = 0;
821
822 /* At this point shinfo->nr_frags is in fact the number of
823 * slots, which can be as large as XEN_NETBK_LEGACY_SLOTS_MAX.
824 */
825 if (shinfo->nr_frags > MAX_SKB_FRAGS) {
826 frag_overflow = shinfo->nr_frags - MAX_SKB_FRAGS;
827 BUG_ON(frag_overflow > MAX_SKB_FRAGS);
828 shinfo->nr_frags = MAX_SKB_FRAGS;
829 }
830 nr_slots = shinfo->nr_frags;
831
832 /* Skip first skb fragment if it is on same page as header fragment. */
833 start = (frag_get_pending_idx(&shinfo->frags[0]) == pending_idx);
834
835 for (shinfo->nr_frags = start; shinfo->nr_frags < nr_slots;
836 shinfo->nr_frags++, txp++, gop++) {
837 index = pending_index(queue->pending_cons++);
838 pending_idx = queue->pending_ring[index];
839 xenvif_tx_create_map_op(queue, pending_idx, txp, gop);
840 frag_set_pending_idx(&frags[shinfo->nr_frags], pending_idx);
841 }
842
843 if (frag_overflow) {
844 struct sk_buff *nskb = xenvif_alloc_skb(0);
845 if (unlikely(nskb == NULL)) {
846 if (net_ratelimit())
847 netdev_err(queue->vif->dev,
848 "Can't allocate the frag_list skb.\n");
849 return NULL;
850 }
851
852 shinfo = skb_shinfo(nskb);
853 frags = shinfo->frags;
854
855 for (shinfo->nr_frags = 0; shinfo->nr_frags < frag_overflow;
856 shinfo->nr_frags++, txp++, gop++) {
857 index = pending_index(queue->pending_cons++);
858 pending_idx = queue->pending_ring[index];
859 xenvif_tx_create_map_op(queue, pending_idx, txp, gop);
860 frag_set_pending_idx(&frags[shinfo->nr_frags],
861 pending_idx);
862 }
863
864 skb_shinfo(skb)->frag_list = nskb;
865 }
866
867 return gop;
868 }
869
870 static inline void xenvif_grant_handle_set(struct xenvif_queue *queue,
871 u16 pending_idx,
872 grant_handle_t handle)
873 {
874 if (unlikely(queue->grant_tx_handle[pending_idx] !=
875 NETBACK_INVALID_HANDLE)) {
876 netdev_err(queue->vif->dev,
877 "Trying to overwrite active handle! pending_idx: 0x%x\n",
878 pending_idx);
879 BUG();
880 }
881 queue->grant_tx_handle[pending_idx] = handle;
882 }
883
884 static inline void xenvif_grant_handle_reset(struct xenvif_queue *queue,
885 u16 pending_idx)
886 {
887 if (unlikely(queue->grant_tx_handle[pending_idx] ==
888 NETBACK_INVALID_HANDLE)) {
889 netdev_err(queue->vif->dev,
890 "Trying to unmap invalid handle! pending_idx: 0x%x\n",
891 pending_idx);
892 BUG();
893 }
894 queue->grant_tx_handle[pending_idx] = NETBACK_INVALID_HANDLE;
895 }
896
897 static int xenvif_tx_check_gop(struct xenvif_queue *queue,
898 struct sk_buff *skb,
899 struct gnttab_map_grant_ref **gopp_map,
900 struct gnttab_copy **gopp_copy)
901 {
902 struct gnttab_map_grant_ref *gop_map = *gopp_map;
903 u16 pending_idx = XENVIF_TX_CB(skb)->pending_idx;
904 /* This always points to the shinfo of the skb being checked, which
905 * could be either the first or the one on the frag_list
906 */
907 struct skb_shared_info *shinfo = skb_shinfo(skb);
908 /* If this is non-NULL, we are currently checking the frag_list skb, and
909 * this points to the shinfo of the first one
910 */
911 struct skb_shared_info *first_shinfo = NULL;
912 int nr_frags = shinfo->nr_frags;
913 const bool sharedslot = nr_frags &&
914 frag_get_pending_idx(&shinfo->frags[0]) == pending_idx;
915 int i, err;
916
917 /* Check status of header. */
918 err = (*gopp_copy)->status;
919 if (unlikely(err)) {
920 if (net_ratelimit())
921 netdev_dbg(queue->vif->dev,
922 "Grant copy of header failed! status: %d pending_idx: %u ref: %u\n",
923 (*gopp_copy)->status,
924 pending_idx,
925 (*gopp_copy)->source.u.ref);
926 /* The first frag might still have this slot mapped */
927 if (!sharedslot)
928 xenvif_idx_release(queue, pending_idx,
929 XEN_NETIF_RSP_ERROR);
930 }
931 (*gopp_copy)++;
932
933 check_frags:
934 for (i = 0; i < nr_frags; i++, gop_map++) {
935 int j, newerr;
936
937 pending_idx = frag_get_pending_idx(&shinfo->frags[i]);
938
939 /* Check error status: if okay then remember grant handle. */
940 newerr = gop_map->status;
941
942 if (likely(!newerr)) {
943 xenvif_grant_handle_set(queue,
944 pending_idx,
945 gop_map->handle);
946 /* Had a previous error? Invalidate this fragment. */
947 if (unlikely(err)) {
948 xenvif_idx_unmap(queue, pending_idx);
949 /* If the mapping of the first frag was OK, but
950 * the header's copy failed, and they are
951 * sharing a slot, send an error
952 */
953 if (i == 0 && sharedslot)
954 xenvif_idx_release(queue, pending_idx,
955 XEN_NETIF_RSP_ERROR);
956 else
957 xenvif_idx_release(queue, pending_idx,
958 XEN_NETIF_RSP_OKAY);
959 }
960 continue;
961 }
962
963 /* Error on this fragment: respond to client with an error. */
964 if (net_ratelimit())
965 netdev_dbg(queue->vif->dev,
966 "Grant map of %d. frag failed! status: %d pending_idx: %u ref: %u\n",
967 i,
968 gop_map->status,
969 pending_idx,
970 gop_map->ref);
971
972 xenvif_idx_release(queue, pending_idx, XEN_NETIF_RSP_ERROR);
973
974 /* Not the first error? Preceding frags already invalidated. */
975 if (err)
976 continue;
977
978 /* First error: if the header haven't shared a slot with the
979 * first frag, release it as well.
980 */
981 if (!sharedslot)
982 xenvif_idx_release(queue,
983 XENVIF_TX_CB(skb)->pending_idx,
984 XEN_NETIF_RSP_OKAY);
985
986 /* Invalidate preceding fragments of this skb. */
987 for (j = 0; j < i; j++) {
988 pending_idx = frag_get_pending_idx(&shinfo->frags[j]);
989 xenvif_idx_unmap(queue, pending_idx);
990 xenvif_idx_release(queue, pending_idx,
991 XEN_NETIF_RSP_OKAY);
992 }
993
994 /* And if we found the error while checking the frag_list, unmap
995 * the first skb's frags
996 */
997 if (first_shinfo) {
998 for (j = 0; j < first_shinfo->nr_frags; j++) {
999 pending_idx = frag_get_pending_idx(&first_shinfo->frags[j]);
1000 xenvif_idx_unmap(queue, pending_idx);
1001 xenvif_idx_release(queue, pending_idx,
1002 XEN_NETIF_RSP_OKAY);
1003 }
1004 }
1005
1006 /* Remember the error: invalidate all subsequent fragments. */
1007 err = newerr;
1008 }
1009
1010 if (skb_has_frag_list(skb) && !first_shinfo) {
1011 first_shinfo = skb_shinfo(skb);
1012 shinfo = skb_shinfo(skb_shinfo(skb)->frag_list);
1013 nr_frags = shinfo->nr_frags;
1014
1015 goto check_frags;
1016 }
1017
1018 *gopp_map = gop_map;
1019 return err;
1020 }
1021
1022 static void xenvif_fill_frags(struct xenvif_queue *queue, struct sk_buff *skb)
1023 {
1024 struct skb_shared_info *shinfo = skb_shinfo(skb);
1025 int nr_frags = shinfo->nr_frags;
1026 int i;
1027 u16 prev_pending_idx = INVALID_PENDING_IDX;
1028
1029 for (i = 0; i < nr_frags; i++) {
1030 skb_frag_t *frag = shinfo->frags + i;
1031 struct xen_netif_tx_request *txp;
1032 struct page *page;
1033 u16 pending_idx;
1034
1035 pending_idx = frag_get_pending_idx(frag);
1036
1037 /* If this is not the first frag, chain it to the previous*/
1038 if (prev_pending_idx == INVALID_PENDING_IDX)
1039 skb_shinfo(skb)->destructor_arg =
1040 &callback_param(queue, pending_idx);
1041 else
1042 callback_param(queue, prev_pending_idx).ctx =
1043 &callback_param(queue, pending_idx);
1044
1045 callback_param(queue, pending_idx).ctx = NULL;
1046 prev_pending_idx = pending_idx;
1047
1048 txp = &queue->pending_tx_info[pending_idx].req;
1049 page = virt_to_page(idx_to_kaddr(queue, pending_idx));
1050 __skb_fill_page_desc(skb, i, page, txp->offset, txp->size);
1051 skb->len += txp->size;
1052 skb->data_len += txp->size;
1053 skb->truesize += txp->size;
1054
1055 /* Take an extra reference to offset network stack's put_page */
1056 get_page(queue->mmap_pages[pending_idx]);
1057 }
1058 }
1059
1060 static int xenvif_get_extras(struct xenvif_queue *queue,
1061 struct xen_netif_extra_info *extras,
1062 int work_to_do)
1063 {
1064 struct xen_netif_extra_info extra;
1065 RING_IDX cons = queue->tx.req_cons;
1066
1067 do {
1068 if (unlikely(work_to_do-- <= 0)) {
1069 netdev_err(queue->vif->dev, "Missing extra info\n");
1070 xenvif_fatal_tx_err(queue->vif);
1071 return -EBADR;
1072 }
1073
1074 memcpy(&extra, RING_GET_REQUEST(&queue->tx, cons),
1075 sizeof(extra));
1076 if (unlikely(!extra.type ||
1077 extra.type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
1078 queue->tx.req_cons = ++cons;
1079 netdev_err(queue->vif->dev,
1080 "Invalid extra type: %d\n", extra.type);
1081 xenvif_fatal_tx_err(queue->vif);
1082 return -EINVAL;
1083 }
1084
1085 memcpy(&extras[extra.type - 1], &extra, sizeof(extra));
1086 queue->tx.req_cons = ++cons;
1087 } while (extra.flags & XEN_NETIF_EXTRA_FLAG_MORE);
1088
1089 return work_to_do;
1090 }
1091
1092 static int xenvif_set_skb_gso(struct xenvif *vif,
1093 struct sk_buff *skb,
1094 struct xen_netif_extra_info *gso)
1095 {
1096 if (!gso->u.gso.size) {
1097 netdev_err(vif->dev, "GSO size must not be zero.\n");
1098 xenvif_fatal_tx_err(vif);
1099 return -EINVAL;
1100 }
1101
1102 switch (gso->u.gso.type) {
1103 case XEN_NETIF_GSO_TYPE_TCPV4:
1104 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1105 break;
1106 case XEN_NETIF_GSO_TYPE_TCPV6:
1107 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1108 break;
1109 default:
1110 netdev_err(vif->dev, "Bad GSO type %d.\n", gso->u.gso.type);
1111 xenvif_fatal_tx_err(vif);
1112 return -EINVAL;
1113 }
1114
1115 skb_shinfo(skb)->gso_size = gso->u.gso.size;
1116 /* gso_segs will be calculated later */
1117
1118 return 0;
1119 }
1120
1121 static int checksum_setup(struct xenvif_queue *queue, struct sk_buff *skb)
1122 {
1123 bool recalculate_partial_csum = false;
1124
1125 /* A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
1126 * peers can fail to set NETRXF_csum_blank when sending a GSO
1127 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
1128 * recalculate the partial checksum.
1129 */
1130 if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
1131 queue->stats.rx_gso_checksum_fixup++;
1132 skb->ip_summed = CHECKSUM_PARTIAL;
1133 recalculate_partial_csum = true;
1134 }
1135
1136 /* A non-CHECKSUM_PARTIAL SKB does not require setup. */
1137 if (skb->ip_summed != CHECKSUM_PARTIAL)
1138 return 0;
1139
1140 return skb_checksum_setup(skb, recalculate_partial_csum);
1141 }
1142
1143 static bool tx_credit_exceeded(struct xenvif_queue *queue, unsigned size)
1144 {
1145 u64 now = get_jiffies_64();
1146 u64 next_credit = queue->credit_window_start +
1147 msecs_to_jiffies(queue->credit_usec / 1000);
1148
1149 /* Timer could already be pending in rare cases. */
1150 if (timer_pending(&queue->credit_timeout))
1151 return true;
1152
1153 /* Passed the point where we can replenish credit? */
1154 if (time_after_eq64(now, next_credit)) {
1155 queue->credit_window_start = now;
1156 tx_add_credit(queue);
1157 }
1158
1159 /* Still too big to send right now? Set a callback. */
1160 if (size > queue->remaining_credit) {
1161 queue->credit_timeout.data =
1162 (unsigned long)queue;
1163 mod_timer(&queue->credit_timeout,
1164 next_credit);
1165 queue->credit_window_start = next_credit;
1166
1167 return true;
1168 }
1169
1170 return false;
1171 }
1172
1173 static void xenvif_tx_build_gops(struct xenvif_queue *queue,
1174 int budget,
1175 unsigned *copy_ops,
1176 unsigned *map_ops)
1177 {
1178 struct gnttab_map_grant_ref *gop = queue->tx_map_ops, *request_gop;
1179 struct sk_buff *skb;
1180 int ret;
1181
1182 while (skb_queue_len(&queue->tx_queue) < budget) {
1183 struct xen_netif_tx_request txreq;
1184 struct xen_netif_tx_request txfrags[XEN_NETBK_LEGACY_SLOTS_MAX];
1185 struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX-1];
1186 u16 pending_idx;
1187 RING_IDX idx;
1188 int work_to_do;
1189 unsigned int data_len;
1190 pending_ring_idx_t index;
1191
1192 if (queue->tx.sring->req_prod - queue->tx.req_cons >
1193 XEN_NETIF_TX_RING_SIZE) {
1194 netdev_err(queue->vif->dev,
1195 "Impossible number of requests. "
1196 "req_prod %d, req_cons %d, size %ld\n",
1197 queue->tx.sring->req_prod, queue->tx.req_cons,
1198 XEN_NETIF_TX_RING_SIZE);
1199 xenvif_fatal_tx_err(queue->vif);
1200 break;
1201 }
1202
1203 work_to_do = RING_HAS_UNCONSUMED_REQUESTS(&queue->tx);
1204 if (!work_to_do)
1205 break;
1206
1207 idx = queue->tx.req_cons;
1208 rmb(); /* Ensure that we see the request before we copy it. */
1209 memcpy(&txreq, RING_GET_REQUEST(&queue->tx, idx), sizeof(txreq));
1210
1211 /* Credit-based scheduling. */
1212 if (txreq.size > queue->remaining_credit &&
1213 tx_credit_exceeded(queue, txreq.size))
1214 break;
1215
1216 queue->remaining_credit -= txreq.size;
1217
1218 work_to_do--;
1219 queue->tx.req_cons = ++idx;
1220
1221 memset(extras, 0, sizeof(extras));
1222 if (txreq.flags & XEN_NETTXF_extra_info) {
1223 work_to_do = xenvif_get_extras(queue, extras,
1224 work_to_do);
1225 idx = queue->tx.req_cons;
1226 if (unlikely(work_to_do < 0))
1227 break;
1228 }
1229
1230 ret = xenvif_count_requests(queue, &txreq, txfrags, work_to_do);
1231 if (unlikely(ret < 0))
1232 break;
1233
1234 idx += ret;
1235
1236 if (unlikely(txreq.size < ETH_HLEN)) {
1237 netdev_dbg(queue->vif->dev,
1238 "Bad packet size: %d\n", txreq.size);
1239 xenvif_tx_err(queue, &txreq, idx);
1240 break;
1241 }
1242
1243 /* No crossing a page as the payload mustn't fragment. */
1244 if (unlikely((txreq.offset + txreq.size) > PAGE_SIZE)) {
1245 netdev_err(queue->vif->dev,
1246 "txreq.offset: %u, size: %u, end: %lu\n",
1247 txreq.offset, txreq.size,
1248 (unsigned long)(txreq.offset&~PAGE_MASK) + txreq.size);
1249 xenvif_fatal_tx_err(queue->vif);
1250 break;
1251 }
1252
1253 index = pending_index(queue->pending_cons);
1254 pending_idx = queue->pending_ring[index];
1255
1256 data_len = (txreq.size > XEN_NETBACK_TX_COPY_LEN &&
1257 ret < XEN_NETBK_LEGACY_SLOTS_MAX) ?
1258 XEN_NETBACK_TX_COPY_LEN : txreq.size;
1259
1260 skb = xenvif_alloc_skb(data_len);
1261 if (unlikely(skb == NULL)) {
1262 netdev_dbg(queue->vif->dev,
1263 "Can't allocate a skb in start_xmit.\n");
1264 xenvif_tx_err(queue, &txreq, idx);
1265 break;
1266 }
1267
1268 if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
1269 struct xen_netif_extra_info *gso;
1270 gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
1271
1272 if (xenvif_set_skb_gso(queue->vif, skb, gso)) {
1273 /* Failure in xenvif_set_skb_gso is fatal. */
1274 kfree_skb(skb);
1275 break;
1276 }
1277 }
1278
1279 XENVIF_TX_CB(skb)->pending_idx = pending_idx;
1280
1281 __skb_put(skb, data_len);
1282 queue->tx_copy_ops[*copy_ops].source.u.ref = txreq.gref;
1283 queue->tx_copy_ops[*copy_ops].source.domid = queue->vif->domid;
1284 queue->tx_copy_ops[*copy_ops].source.offset = txreq.offset;
1285
1286 queue->tx_copy_ops[*copy_ops].dest.u.gmfn =
1287 virt_to_mfn(skb->data);
1288 queue->tx_copy_ops[*copy_ops].dest.domid = DOMID_SELF;
1289 queue->tx_copy_ops[*copy_ops].dest.offset =
1290 offset_in_page(skb->data);
1291
1292 queue->tx_copy_ops[*copy_ops].len = data_len;
1293 queue->tx_copy_ops[*copy_ops].flags = GNTCOPY_source_gref;
1294
1295 (*copy_ops)++;
1296
1297 skb_shinfo(skb)->nr_frags = ret;
1298 if (data_len < txreq.size) {
1299 skb_shinfo(skb)->nr_frags++;
1300 frag_set_pending_idx(&skb_shinfo(skb)->frags[0],
1301 pending_idx);
1302 xenvif_tx_create_map_op(queue, pending_idx, &txreq, gop);
1303 gop++;
1304 } else {
1305 frag_set_pending_idx(&skb_shinfo(skb)->frags[0],
1306 INVALID_PENDING_IDX);
1307 memcpy(&queue->pending_tx_info[pending_idx].req, &txreq,
1308 sizeof(txreq));
1309 }
1310
1311 queue->pending_cons++;
1312
1313 request_gop = xenvif_get_requests(queue, skb, txfrags, gop);
1314 if (request_gop == NULL) {
1315 kfree_skb(skb);
1316 xenvif_tx_err(queue, &txreq, idx);
1317 break;
1318 }
1319 gop = request_gop;
1320
1321 __skb_queue_tail(&queue->tx_queue, skb);
1322
1323 queue->tx.req_cons = idx;
1324
1325 if (((gop-queue->tx_map_ops) >= ARRAY_SIZE(queue->tx_map_ops)) ||
1326 (*copy_ops >= ARRAY_SIZE(queue->tx_copy_ops)))
1327 break;
1328 }
1329
1330 (*map_ops) = gop - queue->tx_map_ops;
1331 return;
1332 }
1333
1334 /* Consolidate skb with a frag_list into a brand new one with local pages on
1335 * frags. Returns 0 or -ENOMEM if can't allocate new pages.
1336 */
1337 static int xenvif_handle_frag_list(struct xenvif_queue *queue, struct sk_buff *skb)
1338 {
1339 unsigned int offset = skb_headlen(skb);
1340 skb_frag_t frags[MAX_SKB_FRAGS];
1341 int i, f;
1342 struct ubuf_info *uarg;
1343 struct sk_buff *nskb = skb_shinfo(skb)->frag_list;
1344
1345 queue->stats.tx_zerocopy_sent += 2;
1346 queue->stats.tx_frag_overflow++;
1347
1348 xenvif_fill_frags(queue, nskb);
1349 /* Subtract frags size, we will correct it later */
1350 skb->truesize -= skb->data_len;
1351 skb->len += nskb->len;
1352 skb->data_len += nskb->len;
1353
1354 /* create a brand new frags array and coalesce there */
1355 for (i = 0; offset < skb->len; i++) {
1356 struct page *page;
1357 unsigned int len;
1358
1359 BUG_ON(i >= MAX_SKB_FRAGS);
1360 page = alloc_page(GFP_ATOMIC);
1361 if (!page) {
1362 int j;
1363 skb->truesize += skb->data_len;
1364 for (j = 0; j < i; j++)
1365 put_page(frags[j].page.p);
1366 return -ENOMEM;
1367 }
1368
1369 if (offset + PAGE_SIZE < skb->len)
1370 len = PAGE_SIZE;
1371 else
1372 len = skb->len - offset;
1373 if (skb_copy_bits(skb, offset, page_address(page), len))
1374 BUG();
1375
1376 offset += len;
1377 frags[i].page.p = page;
1378 frags[i].page_offset = 0;
1379 skb_frag_size_set(&frags[i], len);
1380 }
1381
1382 /* Copied all the bits from the frag list -- free it. */
1383 skb_frag_list_init(skb);
1384 xenvif_skb_zerocopy_prepare(queue, nskb);
1385 kfree_skb(nskb);
1386
1387 /* Release all the original (foreign) frags. */
1388 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
1389 skb_frag_unref(skb, f);
1390 uarg = skb_shinfo(skb)->destructor_arg;
1391 /* increase inflight counter to offset decrement in callback */
1392 atomic_inc(&queue->inflight_packets);
1393 uarg->callback(uarg, true);
1394 skb_shinfo(skb)->destructor_arg = NULL;
1395
1396 /* Fill the skb with the new (local) frags. */
1397 memcpy(skb_shinfo(skb)->frags, frags, i * sizeof(skb_frag_t));
1398 skb_shinfo(skb)->nr_frags = i;
1399 skb->truesize += i * PAGE_SIZE;
1400
1401 return 0;
1402 }
1403
1404 static int xenvif_tx_submit(struct xenvif_queue *queue)
1405 {
1406 struct gnttab_map_grant_ref *gop_map = queue->tx_map_ops;
1407 struct gnttab_copy *gop_copy = queue->tx_copy_ops;
1408 struct sk_buff *skb;
1409 int work_done = 0;
1410
1411 while ((skb = __skb_dequeue(&queue->tx_queue)) != NULL) {
1412 struct xen_netif_tx_request *txp;
1413 u16 pending_idx;
1414 unsigned data_len;
1415
1416 pending_idx = XENVIF_TX_CB(skb)->pending_idx;
1417 txp = &queue->pending_tx_info[pending_idx].req;
1418
1419 /* Check the remap error code. */
1420 if (unlikely(xenvif_tx_check_gop(queue, skb, &gop_map, &gop_copy))) {
1421 /* If there was an error, xenvif_tx_check_gop is
1422 * expected to release all the frags which were mapped,
1423 * so kfree_skb shouldn't do it again
1424 */
1425 skb_shinfo(skb)->nr_frags = 0;
1426 if (skb_has_frag_list(skb)) {
1427 struct sk_buff *nskb =
1428 skb_shinfo(skb)->frag_list;
1429 skb_shinfo(nskb)->nr_frags = 0;
1430 }
1431 kfree_skb(skb);
1432 continue;
1433 }
1434
1435 data_len = skb->len;
1436 callback_param(queue, pending_idx).ctx = NULL;
1437 if (data_len < txp->size) {
1438 /* Append the packet payload as a fragment. */
1439 txp->offset += data_len;
1440 txp->size -= data_len;
1441 } else {
1442 /* Schedule a response immediately. */
1443 xenvif_idx_release(queue, pending_idx,
1444 XEN_NETIF_RSP_OKAY);
1445 }
1446
1447 if (txp->flags & XEN_NETTXF_csum_blank)
1448 skb->ip_summed = CHECKSUM_PARTIAL;
1449 else if (txp->flags & XEN_NETTXF_data_validated)
1450 skb->ip_summed = CHECKSUM_UNNECESSARY;
1451
1452 xenvif_fill_frags(queue, skb);
1453
1454 if (unlikely(skb_has_frag_list(skb))) {
1455 if (xenvif_handle_frag_list(queue, skb)) {
1456 if (net_ratelimit())
1457 netdev_err(queue->vif->dev,
1458 "Not enough memory to consolidate frag_list!\n");
1459 xenvif_skb_zerocopy_prepare(queue, skb);
1460 kfree_skb(skb);
1461 continue;
1462 }
1463 }
1464
1465 skb->dev = queue->vif->dev;
1466 skb->protocol = eth_type_trans(skb, skb->dev);
1467 skb_reset_network_header(skb);
1468
1469 if (checksum_setup(queue, skb)) {
1470 netdev_dbg(queue->vif->dev,
1471 "Can't setup checksum in net_tx_action\n");
1472 /* We have to set this flag to trigger the callback */
1473 if (skb_shinfo(skb)->destructor_arg)
1474 xenvif_skb_zerocopy_prepare(queue, skb);
1475 kfree_skb(skb);
1476 continue;
1477 }
1478
1479 skb_probe_transport_header(skb, 0);
1480
1481 /* If the packet is GSO then we will have just set up the
1482 * transport header offset in checksum_setup so it's now
1483 * straightforward to calculate gso_segs.
1484 */
1485 if (skb_is_gso(skb)) {
1486 int mss = skb_shinfo(skb)->gso_size;
1487 int hdrlen = skb_transport_header(skb) -
1488 skb_mac_header(skb) +
1489 tcp_hdrlen(skb);
1490
1491 skb_shinfo(skb)->gso_segs =
1492 DIV_ROUND_UP(skb->len - hdrlen, mss);
1493 }
1494
1495 queue->stats.rx_bytes += skb->len;
1496 queue->stats.rx_packets++;
1497
1498 work_done++;
1499
1500 /* Set this flag right before netif_receive_skb, otherwise
1501 * someone might think this packet already left netback, and
1502 * do a skb_copy_ubufs while we are still in control of the
1503 * skb. E.g. the __pskb_pull_tail earlier can do such thing.
1504 */
1505 if (skb_shinfo(skb)->destructor_arg) {
1506 xenvif_skb_zerocopy_prepare(queue, skb);
1507 queue->stats.tx_zerocopy_sent++;
1508 }
1509
1510 netif_receive_skb(skb);
1511 }
1512
1513 return work_done;
1514 }
1515
1516 void xenvif_zerocopy_callback(struct ubuf_info *ubuf, bool zerocopy_success)
1517 {
1518 unsigned long flags;
1519 pending_ring_idx_t index;
1520 struct xenvif_queue *queue = ubuf_to_queue(ubuf);
1521
1522 /* This is the only place where we grab this lock, to protect callbacks
1523 * from each other.
1524 */
1525 spin_lock_irqsave(&queue->callback_lock, flags);
1526 do {
1527 u16 pending_idx = ubuf->desc;
1528 ubuf = (struct ubuf_info *) ubuf->ctx;
1529 BUG_ON(queue->dealloc_prod - queue->dealloc_cons >=
1530 MAX_PENDING_REQS);
1531 index = pending_index(queue->dealloc_prod);
1532 queue->dealloc_ring[index] = pending_idx;
1533 /* Sync with xenvif_tx_dealloc_action:
1534 * insert idx then incr producer.
1535 */
1536 smp_wmb();
1537 queue->dealloc_prod++;
1538 } while (ubuf);
1539 wake_up(&queue->dealloc_wq);
1540 spin_unlock_irqrestore(&queue->callback_lock, flags);
1541
1542 if (likely(zerocopy_success))
1543 queue->stats.tx_zerocopy_success++;
1544 else
1545 queue->stats.tx_zerocopy_fail++;
1546 xenvif_skb_zerocopy_complete(queue);
1547 }
1548
1549 static inline void xenvif_tx_dealloc_action(struct xenvif_queue *queue)
1550 {
1551 struct gnttab_unmap_grant_ref *gop;
1552 pending_ring_idx_t dc, dp;
1553 u16 pending_idx, pending_idx_release[MAX_PENDING_REQS];
1554 unsigned int i = 0;
1555
1556 dc = queue->dealloc_cons;
1557 gop = queue->tx_unmap_ops;
1558
1559 /* Free up any grants we have finished using */
1560 do {
1561 dp = queue->dealloc_prod;
1562
1563 /* Ensure we see all indices enqueued by all
1564 * xenvif_zerocopy_callback().
1565 */
1566 smp_rmb();
1567
1568 while (dc != dp) {
1569 BUG_ON(gop - queue->tx_unmap_ops >= MAX_PENDING_REQS);
1570 pending_idx =
1571 queue->dealloc_ring[pending_index(dc++)];
1572
1573 pending_idx_release[gop - queue->tx_unmap_ops] =
1574 pending_idx;
1575 queue->pages_to_unmap[gop - queue->tx_unmap_ops] =
1576 queue->mmap_pages[pending_idx];
1577 gnttab_set_unmap_op(gop,
1578 idx_to_kaddr(queue, pending_idx),
1579 GNTMAP_host_map,
1580 queue->grant_tx_handle[pending_idx]);
1581 xenvif_grant_handle_reset(queue, pending_idx);
1582 ++gop;
1583 }
1584
1585 } while (dp != queue->dealloc_prod);
1586
1587 queue->dealloc_cons = dc;
1588
1589 if (gop - queue->tx_unmap_ops > 0) {
1590 int ret;
1591 ret = gnttab_unmap_refs(queue->tx_unmap_ops,
1592 NULL,
1593 queue->pages_to_unmap,
1594 gop - queue->tx_unmap_ops);
1595 if (ret) {
1596 netdev_err(queue->vif->dev, "Unmap fail: nr_ops %tu ret %d\n",
1597 gop - queue->tx_unmap_ops, ret);
1598 for (i = 0; i < gop - queue->tx_unmap_ops; ++i) {
1599 if (gop[i].status != GNTST_okay)
1600 netdev_err(queue->vif->dev,
1601 " host_addr: 0x%llx handle: 0x%x status: %d\n",
1602 gop[i].host_addr,
1603 gop[i].handle,
1604 gop[i].status);
1605 }
1606 BUG();
1607 }
1608 }
1609
1610 for (i = 0; i < gop - queue->tx_unmap_ops; ++i)
1611 xenvif_idx_release(queue, pending_idx_release[i],
1612 XEN_NETIF_RSP_OKAY);
1613 }
1614
1615
1616 /* Called after netfront has transmitted */
1617 int xenvif_tx_action(struct xenvif_queue *queue, int budget)
1618 {
1619 unsigned nr_mops, nr_cops = 0;
1620 int work_done, ret;
1621
1622 if (unlikely(!tx_work_todo(queue)))
1623 return 0;
1624
1625 xenvif_tx_build_gops(queue, budget, &nr_cops, &nr_mops);
1626
1627 if (nr_cops == 0)
1628 return 0;
1629
1630 gnttab_batch_copy(queue->tx_copy_ops, nr_cops);
1631 if (nr_mops != 0) {
1632 ret = gnttab_map_refs(queue->tx_map_ops,
1633 NULL,
1634 queue->pages_to_map,
1635 nr_mops);
1636 BUG_ON(ret);
1637 }
1638
1639 work_done = xenvif_tx_submit(queue);
1640
1641 return work_done;
1642 }
1643
1644 static void xenvif_idx_release(struct xenvif_queue *queue, u16 pending_idx,
1645 u8 status)
1646 {
1647 struct pending_tx_info *pending_tx_info;
1648 pending_ring_idx_t index;
1649 unsigned long flags;
1650
1651 pending_tx_info = &queue->pending_tx_info[pending_idx];
1652
1653 spin_lock_irqsave(&queue->response_lock, flags);
1654
1655 make_tx_response(queue, &pending_tx_info->req, status);
1656
1657 /* Release the pending index before pusing the Tx response so
1658 * its available before a new Tx request is pushed by the
1659 * frontend.
1660 */
1661 index = pending_index(queue->pending_prod++);
1662 queue->pending_ring[index] = pending_idx;
1663
1664 push_tx_responses(queue);
1665
1666 spin_unlock_irqrestore(&queue->response_lock, flags);
1667 }
1668
1669
1670 static void make_tx_response(struct xenvif_queue *queue,
1671 struct xen_netif_tx_request *txp,
1672 s8 st)
1673 {
1674 RING_IDX i = queue->tx.rsp_prod_pvt;
1675 struct xen_netif_tx_response *resp;
1676
1677 resp = RING_GET_RESPONSE(&queue->tx, i);
1678 resp->id = txp->id;
1679 resp->status = st;
1680
1681 if (txp->flags & XEN_NETTXF_extra_info)
1682 RING_GET_RESPONSE(&queue->tx, ++i)->status = XEN_NETIF_RSP_NULL;
1683
1684 queue->tx.rsp_prod_pvt = ++i;
1685 }
1686
1687 static void push_tx_responses(struct xenvif_queue *queue)
1688 {
1689 int notify;
1690
1691 RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&queue->tx, notify);
1692 if (notify)
1693 notify_remote_via_irq(queue->tx_irq);
1694 }
1695
1696 static struct xen_netif_rx_response *make_rx_response(struct xenvif_queue *queue,
1697 u16 id,
1698 s8 st,
1699 u16 offset,
1700 u16 size,
1701 u16 flags)
1702 {
1703 RING_IDX i = queue->rx.rsp_prod_pvt;
1704 struct xen_netif_rx_response *resp;
1705
1706 resp = RING_GET_RESPONSE(&queue->rx, i);
1707 resp->offset = offset;
1708 resp->flags = flags;
1709 resp->id = id;
1710 resp->status = (s16)size;
1711 if (st < 0)
1712 resp->status = (s16)st;
1713
1714 queue->rx.rsp_prod_pvt = ++i;
1715
1716 return resp;
1717 }
1718
1719 void xenvif_idx_unmap(struct xenvif_queue *queue, u16 pending_idx)
1720 {
1721 int ret;
1722 struct gnttab_unmap_grant_ref tx_unmap_op;
1723
1724 gnttab_set_unmap_op(&tx_unmap_op,
1725 idx_to_kaddr(queue, pending_idx),
1726 GNTMAP_host_map,
1727 queue->grant_tx_handle[pending_idx]);
1728 xenvif_grant_handle_reset(queue, pending_idx);
1729
1730 ret = gnttab_unmap_refs(&tx_unmap_op, NULL,
1731 &queue->mmap_pages[pending_idx], 1);
1732 if (ret) {
1733 netdev_err(queue->vif->dev,
1734 "Unmap fail: ret: %d pending_idx: %d host_addr: %llx handle: 0x%x status: %d\n",
1735 ret,
1736 pending_idx,
1737 tx_unmap_op.host_addr,
1738 tx_unmap_op.handle,
1739 tx_unmap_op.status);
1740 BUG();
1741 }
1742 }
1743
1744 static inline int tx_work_todo(struct xenvif_queue *queue)
1745 {
1746 if (likely(RING_HAS_UNCONSUMED_REQUESTS(&queue->tx)))
1747 return 1;
1748
1749 return 0;
1750 }
1751
1752 static inline bool tx_dealloc_work_todo(struct xenvif_queue *queue)
1753 {
1754 return queue->dealloc_cons != queue->dealloc_prod;
1755 }
1756
1757 void xenvif_unmap_frontend_rings(struct xenvif_queue *queue)
1758 {
1759 if (queue->tx.sring)
1760 xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(queue->vif),
1761 queue->tx.sring);
1762 if (queue->rx.sring)
1763 xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(queue->vif),
1764 queue->rx.sring);
1765 }
1766
1767 int xenvif_map_frontend_rings(struct xenvif_queue *queue,
1768 grant_ref_t tx_ring_ref,
1769 grant_ref_t rx_ring_ref)
1770 {
1771 void *addr;
1772 struct xen_netif_tx_sring *txs;
1773 struct xen_netif_rx_sring *rxs;
1774
1775 int err = -ENOMEM;
1776
1777 err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(queue->vif),
1778 &tx_ring_ref, 1, &addr);
1779 if (err)
1780 goto err;
1781
1782 txs = (struct xen_netif_tx_sring *)addr;
1783 BACK_RING_INIT(&queue->tx, txs, PAGE_SIZE);
1784
1785 err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(queue->vif),
1786 &rx_ring_ref, 1, &addr);
1787 if (err)
1788 goto err;
1789
1790 rxs = (struct xen_netif_rx_sring *)addr;
1791 BACK_RING_INIT(&queue->rx, rxs, PAGE_SIZE);
1792
1793 return 0;
1794
1795 err:
1796 xenvif_unmap_frontend_rings(queue);
1797 return err;
1798 }
1799
1800 static void xenvif_queue_carrier_off(struct xenvif_queue *queue)
1801 {
1802 struct xenvif *vif = queue->vif;
1803
1804 queue->stalled = true;
1805
1806 /* At least one queue has stalled? Disable the carrier. */
1807 spin_lock(&vif->lock);
1808 if (vif->stalled_queues++ == 0) {
1809 netdev_info(vif->dev, "Guest Rx stalled");
1810 netif_carrier_off(vif->dev);
1811 }
1812 spin_unlock(&vif->lock);
1813 }
1814
1815 static void xenvif_queue_carrier_on(struct xenvif_queue *queue)
1816 {
1817 struct xenvif *vif = queue->vif;
1818
1819 queue->last_rx_time = jiffies; /* Reset Rx stall detection. */
1820 queue->stalled = false;
1821
1822 /* All queues are ready? Enable the carrier. */
1823 spin_lock(&vif->lock);
1824 if (--vif->stalled_queues == 0) {
1825 netdev_info(vif->dev, "Guest Rx ready");
1826 netif_carrier_on(vif->dev);
1827 }
1828 spin_unlock(&vif->lock);
1829 }
1830
1831 static bool xenvif_rx_queue_stalled(struct xenvif_queue *queue)
1832 {
1833 RING_IDX prod, cons;
1834
1835 prod = queue->rx.sring->req_prod;
1836 cons = queue->rx.req_cons;
1837
1838 return !queue->stalled
1839 && prod - cons < XEN_NETBK_RX_SLOTS_MAX
1840 && time_after(jiffies,
1841 queue->last_rx_time + queue->vif->stall_timeout);
1842 }
1843
1844 static bool xenvif_rx_queue_ready(struct xenvif_queue *queue)
1845 {
1846 RING_IDX prod, cons;
1847
1848 prod = queue->rx.sring->req_prod;
1849 cons = queue->rx.req_cons;
1850
1851 return queue->stalled
1852 && prod - cons >= XEN_NETBK_RX_SLOTS_MAX;
1853 }
1854
1855 static bool xenvif_have_rx_work(struct xenvif_queue *queue)
1856 {
1857 return (!skb_queue_empty(&queue->rx_queue)
1858 && xenvif_rx_ring_slots_available(queue, XEN_NETBK_RX_SLOTS_MAX))
1859 || (queue->vif->stall_timeout &&
1860 (xenvif_rx_queue_stalled(queue)
1861 || xenvif_rx_queue_ready(queue)))
1862 || kthread_should_stop()
1863 || queue->vif->disabled;
1864 }
1865
1866 static long xenvif_rx_queue_timeout(struct xenvif_queue *queue)
1867 {
1868 struct sk_buff *skb;
1869 long timeout;
1870
1871 skb = skb_peek(&queue->rx_queue);
1872 if (!skb)
1873 return MAX_SCHEDULE_TIMEOUT;
1874
1875 timeout = XENVIF_RX_CB(skb)->expires - jiffies;
1876 return timeout < 0 ? 0 : timeout;
1877 }
1878
1879 /* Wait until the guest Rx thread has work.
1880 *
1881 * The timeout needs to be adjusted based on the current head of the
1882 * queue (and not just the head at the beginning). In particular, if
1883 * the queue is initially empty an infinite timeout is used and this
1884 * needs to be reduced when a skb is queued.
1885 *
1886 * This cannot be done with wait_event_timeout() because it only
1887 * calculates the timeout once.
1888 */
1889 static void xenvif_wait_for_rx_work(struct xenvif_queue *queue)
1890 {
1891 DEFINE_WAIT(wait);
1892
1893 if (xenvif_have_rx_work(queue))
1894 return;
1895
1896 for (;;) {
1897 long ret;
1898
1899 prepare_to_wait(&queue->wq, &wait, TASK_INTERRUPTIBLE);
1900 if (xenvif_have_rx_work(queue))
1901 break;
1902 ret = schedule_timeout(xenvif_rx_queue_timeout(queue));
1903 if (!ret)
1904 break;
1905 }
1906 finish_wait(&queue->wq, &wait);
1907 }
1908
1909 int xenvif_kthread_guest_rx(void *data)
1910 {
1911 struct xenvif_queue *queue = data;
1912 struct xenvif *vif = queue->vif;
1913
1914 if (!vif->stall_timeout)
1915 xenvif_queue_carrier_on(queue);
1916
1917 for (;;) {
1918 xenvif_wait_for_rx_work(queue);
1919
1920 if (kthread_should_stop())
1921 break;
1922
1923 /* This frontend is found to be rogue, disable it in
1924 * kthread context. Currently this is only set when
1925 * netback finds out frontend sends malformed packet,
1926 * but we cannot disable the interface in softirq
1927 * context so we defer it here, if this thread is
1928 * associated with queue 0.
1929 */
1930 if (unlikely(vif->disabled && queue->id == 0)) {
1931 xenvif_carrier_off(vif);
1932 break;
1933 }
1934
1935 if (!skb_queue_empty(&queue->rx_queue))
1936 xenvif_rx_action(queue);
1937
1938 /* If the guest hasn't provided any Rx slots for a
1939 * while it's probably not responsive, drop the
1940 * carrier so packets are dropped earlier.
1941 */
1942 if (vif->stall_timeout) {
1943 if (xenvif_rx_queue_stalled(queue))
1944 xenvif_queue_carrier_off(queue);
1945 else if (xenvif_rx_queue_ready(queue))
1946 xenvif_queue_carrier_on(queue);
1947 }
1948
1949 /* Queued packets may have foreign pages from other
1950 * domains. These cannot be queued indefinitely as
1951 * this would starve guests of grant refs and transmit
1952 * slots.
1953 */
1954 xenvif_rx_queue_drop_expired(queue);
1955
1956 xenvif_rx_queue_maybe_wake(queue);
1957
1958 cond_resched();
1959 }
1960
1961 /* Bin any remaining skbs */
1962 xenvif_rx_queue_purge(queue);
1963
1964 return 0;
1965 }
1966
1967 static bool xenvif_dealloc_kthread_should_stop(struct xenvif_queue *queue)
1968 {
1969 /* Dealloc thread must remain running until all inflight
1970 * packets complete.
1971 */
1972 return kthread_should_stop() &&
1973 !atomic_read(&queue->inflight_packets);
1974 }
1975
1976 int xenvif_dealloc_kthread(void *data)
1977 {
1978 struct xenvif_queue *queue = data;
1979
1980 for (;;) {
1981 wait_event_interruptible(queue->dealloc_wq,
1982 tx_dealloc_work_todo(queue) ||
1983 xenvif_dealloc_kthread_should_stop(queue));
1984 if (xenvif_dealloc_kthread_should_stop(queue))
1985 break;
1986
1987 xenvif_tx_dealloc_action(queue);
1988 cond_resched();
1989 }
1990
1991 /* Unmap anything remaining*/
1992 if (tx_dealloc_work_todo(queue))
1993 xenvif_tx_dealloc_action(queue);
1994
1995 return 0;
1996 }
1997
1998 static int __init netback_init(void)
1999 {
2000 int rc = 0;
2001
2002 if (!xen_domain())
2003 return -ENODEV;
2004
2005 /* Allow as many queues as there are CPUs, by default */
2006 xenvif_max_queues = num_online_cpus();
2007
2008 if (fatal_skb_slots < XEN_NETBK_LEGACY_SLOTS_MAX) {
2009 pr_info("fatal_skb_slots too small (%d), bump it to XEN_NETBK_LEGACY_SLOTS_MAX (%d)\n",
2010 fatal_skb_slots, XEN_NETBK_LEGACY_SLOTS_MAX);
2011 fatal_skb_slots = XEN_NETBK_LEGACY_SLOTS_MAX;
2012 }
2013
2014 rc = xenvif_xenbus_init();
2015 if (rc)
2016 goto failed_init;
2017
2018 #ifdef CONFIG_DEBUG_FS
2019 xen_netback_dbg_root = debugfs_create_dir("xen-netback", NULL);
2020 if (IS_ERR_OR_NULL(xen_netback_dbg_root))
2021 pr_warn("Init of debugfs returned %ld!\n",
2022 PTR_ERR(xen_netback_dbg_root));
2023 #endif /* CONFIG_DEBUG_FS */
2024
2025 return 0;
2026
2027 failed_init:
2028 return rc;
2029 }
2030
2031 module_init(netback_init);
2032
2033 static void __exit netback_fini(void)
2034 {
2035 #ifdef CONFIG_DEBUG_FS
2036 if (!IS_ERR_OR_NULL(xen_netback_dbg_root))
2037 debugfs_remove_recursive(xen_netback_dbg_root);
2038 #endif /* CONFIG_DEBUG_FS */
2039 xenvif_xenbus_fini();
2040 }
2041 module_exit(netback_fini);
2042
2043 MODULE_LICENSE("Dual BSD/GPL");
2044 MODULE_ALIAS("xen-backend:vif");
Linux with added FPC-III support