?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 4.19.133
[linux/fpc-iii.git] / drivers / net / xen-netfront.c
blob6b4675a9494b2c3b8471c33a9da37278de418b74
1 /*
2 * Virtual network driver for conversing with remote driver backends.
3 *
4 * Copyright (c) 2002-2005, K A Fraser
5 * Copyright (c) 2005, XenSource Ltd
6 *
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License version 2
9 * as published by the Free Software Foundation; or, when distributed
10 * separately from the Linux kernel or incorporated into other
11 * software packages, subject to the following license:
12 *
13 * Permission is hereby granted, free of charge, to any person obtaining a copy
14 * of this source file (the "Software"), to deal in the Software without
15 * restriction, including without limitation the rights to use, copy, modify,
16 * merge, publish, distribute, sublicense, and/or sell copies of the Software,
17 * and to permit persons to whom the Software is furnished to do so, subject to
18 * the following conditions:
19 *
20 * The above copyright notice and this permission notice shall be included in
21 * all copies or substantial portions of the Software.
22 *
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
24 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
25 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
26 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
27 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
28 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
29 * IN THE SOFTWARE.
30 */
31
32 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
33
34 #include <linux/module.h>
35 #include <linux/kernel.h>
36 #include <linux/netdevice.h>
37 #include <linux/etherdevice.h>
38 #include <linux/skbuff.h>
39 #include <linux/ethtool.h>
40 #include <linux/if_ether.h>
41 #include <net/tcp.h>
42 #include <linux/udp.h>
43 #include <linux/moduleparam.h>
44 #include <linux/mm.h>
45 #include <linux/slab.h>
46 #include <net/ip.h>
47
48 #include <xen/xen.h>
49 #include <xen/xenbus.h>
50 #include <xen/events.h>
51 #include <xen/page.h>
52 #include <xen/platform_pci.h>
53 #include <xen/grant_table.h>
54
55 #include <xen/interface/io/netif.h>
56 #include <xen/interface/memory.h>
57 #include <xen/interface/grant_table.h>
58
59 /* Module parameters */
60 #define MAX_QUEUES_DEFAULT 8
61 static unsigned int xennet_max_queues;
62 module_param_named(max_queues, xennet_max_queues, uint, 0644);
63 MODULE_PARM_DESC(max_queues,
64 "Maximum number of queues per virtual interface");
65
66 static const struct ethtool_ops xennet_ethtool_ops;
67
68 struct netfront_cb {
69 int pull_to;
70 };
71
72 #define NETFRONT_SKB_CB(skb) ((struct netfront_cb *)((skb)->cb))
73
74 #define RX_COPY_THRESHOLD 256
75
76 #define GRANT_INVALID_REF 0
77
78 #define NET_TX_RING_SIZE __CONST_RING_SIZE(xen_netif_tx, XEN_PAGE_SIZE)
79 #define NET_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, XEN_PAGE_SIZE)
80
81 /* Minimum number of Rx slots (includes slot for GSO metadata). */
82 #define NET_RX_SLOTS_MIN (XEN_NETIF_NR_SLOTS_MIN + 1)
83
84 /* Queue name is interface name with "-qNNN" appended */
85 #define QUEUE_NAME_SIZE (IFNAMSIZ + 6)
86
87 /* IRQ name is queue name with "-tx" or "-rx" appended */
88 #define IRQ_NAME_SIZE (QUEUE_NAME_SIZE + 3)
89
90 static DECLARE_WAIT_QUEUE_HEAD(module_wq);
91
92 struct netfront_stats {
93 u64 packets;
94 u64 bytes;
95 struct u64_stats_sync syncp;
96 };
97
98 struct netfront_info;
99
100 struct netfront_queue {
101 unsigned int id; /* Queue ID, 0-based */
102 char name[QUEUE_NAME_SIZE]; /* DEVNAME-qN */
103 struct netfront_info *info;
104
105 struct napi_struct napi;
106
107 /* Split event channels support, tx_* == rx_* when using
108 * single event channel.
109 */
110 unsigned int tx_evtchn, rx_evtchn;
111 unsigned int tx_irq, rx_irq;
112 /* Only used when split event channels support is enabled */
113 char tx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-tx */
114 char rx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-rx */
115
116 spinlock_t tx_lock;
117 struct xen_netif_tx_front_ring tx;
118 int tx_ring_ref;
119
120 /*
121 * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries
122 * are linked from tx_skb_freelist through skb_entry.link.
123 *
124 * NB. Freelist index entries are always going to be less than
125 * PAGE_OFFSET, whereas pointers to skbs will always be equal or
126 * greater than PAGE_OFFSET: we use this property to distinguish
127 * them.
128 */
129 union skb_entry {
130 struct sk_buff *skb;
131 unsigned long link;
132 } tx_skbs[NET_TX_RING_SIZE];
133 grant_ref_t gref_tx_head;
134 grant_ref_t grant_tx_ref[NET_TX_RING_SIZE];
135 struct page *grant_tx_page[NET_TX_RING_SIZE];
136 unsigned tx_skb_freelist;
137
138 spinlock_t rx_lock ____cacheline_aligned_in_smp;
139 struct xen_netif_rx_front_ring rx;
140 int rx_ring_ref;
141
142 struct timer_list rx_refill_timer;
143
144 struct sk_buff *rx_skbs[NET_RX_RING_SIZE];
145 grant_ref_t gref_rx_head;
146 grant_ref_t grant_rx_ref[NET_RX_RING_SIZE];
147 };
148
149 struct netfront_info {
150 struct list_head list;
151 struct net_device *netdev;
152
153 struct xenbus_device *xbdev;
154
155 /* Multi-queue support */
156 struct netfront_queue *queues;
157
158 /* Statistics */
159 struct netfront_stats __percpu *rx_stats;
160 struct netfront_stats __percpu *tx_stats;
161
162 atomic_t rx_gso_checksum_fixup;
163 };
164
165 struct netfront_rx_info {
166 struct xen_netif_rx_response rx;
167 struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1];
168 };
169
170 static void skb_entry_set_link(union skb_entry *list, unsigned short id)
171 {
172 list->link = id;
173 }
174
175 static int skb_entry_is_link(const union skb_entry *list)
176 {
177 BUILD_BUG_ON(sizeof(list->skb) != sizeof(list->link));
178 return (unsigned long)list->skb < PAGE_OFFSET;
179 }
180
181 /*
182 * Access macros for acquiring freeing slots in tx_skbs[].
183 */
184
185 static void add_id_to_freelist(unsigned *head, union skb_entry *list,
186 unsigned short id)
187 {
188 skb_entry_set_link(&list[id], *head);
189 *head = id;
190 }
191
192 static unsigned short get_id_from_freelist(unsigned *head,
193 union skb_entry *list)
194 {
195 unsigned int id = *head;
196 *head = list[id].link;
197 return id;
198 }
199
200 static int xennet_rxidx(RING_IDX idx)
201 {
202 return idx & (NET_RX_RING_SIZE - 1);
203 }
204
205 static struct sk_buff *xennet_get_rx_skb(struct netfront_queue *queue,
206 RING_IDX ri)
207 {
208 int i = xennet_rxidx(ri);
209 struct sk_buff *skb = queue->rx_skbs[i];
210 queue->rx_skbs[i] = NULL;
211 return skb;
212 }
213
214 static grant_ref_t xennet_get_rx_ref(struct netfront_queue *queue,
215 RING_IDX ri)
216 {
217 int i = xennet_rxidx(ri);
218 grant_ref_t ref = queue->grant_rx_ref[i];
219 queue->grant_rx_ref[i] = GRANT_INVALID_REF;
220 return ref;
221 }
222
223 #ifdef CONFIG_SYSFS
224 static const struct attribute_group xennet_dev_group;
225 #endif
226
227 static bool xennet_can_sg(struct net_device *dev)
228 {
229 return dev->features & NETIF_F_SG;
230 }
231
232
233 static void rx_refill_timeout(struct timer_list *t)
234 {
235 struct netfront_queue *queue = from_timer(queue, t, rx_refill_timer);
236 napi_schedule(&queue->napi);
237 }
238
239 static int netfront_tx_slot_available(struct netfront_queue *queue)
240 {
241 return (queue->tx.req_prod_pvt - queue->tx.rsp_cons) <
242 (NET_TX_RING_SIZE - XEN_NETIF_NR_SLOTS_MIN - 1);
243 }
244
245 static void xennet_maybe_wake_tx(struct netfront_queue *queue)
246 {
247 struct net_device *dev = queue->info->netdev;
248 struct netdev_queue *dev_queue = netdev_get_tx_queue(dev, queue->id);
249
250 if (unlikely(netif_tx_queue_stopped(dev_queue)) &&
251 netfront_tx_slot_available(queue) &&
252 likely(netif_running(dev)))
253 netif_tx_wake_queue(netdev_get_tx_queue(dev, queue->id));
254 }
255
256
257 static struct sk_buff *xennet_alloc_one_rx_buffer(struct netfront_queue *queue)
258 {
259 struct sk_buff *skb;
260 struct page *page;
261
262 skb = __netdev_alloc_skb(queue->info->netdev,
263 RX_COPY_THRESHOLD + NET_IP_ALIGN,
264 GFP_ATOMIC | __GFP_NOWARN);
265 if (unlikely(!skb))
266 return NULL;
267
268 page = alloc_page(GFP_ATOMIC | __GFP_NOWARN);
269 if (!page) {
270 kfree_skb(skb);
271 return NULL;
272 }
273 skb_add_rx_frag(skb, 0, page, 0, 0, PAGE_SIZE);
274
275 /* Align ip header to a 16 bytes boundary */
276 skb_reserve(skb, NET_IP_ALIGN);
277 skb->dev = queue->info->netdev;
278
279 return skb;
280 }
281
282
283 static void xennet_alloc_rx_buffers(struct netfront_queue *queue)
284 {
285 RING_IDX req_prod = queue->rx.req_prod_pvt;
286 int notify;
287 int err = 0;
288
289 if (unlikely(!netif_carrier_ok(queue->info->netdev)))
290 return;
291
292 for (req_prod = queue->rx.req_prod_pvt;
293 req_prod - queue->rx.rsp_cons < NET_RX_RING_SIZE;
294 req_prod++) {
295 struct sk_buff *skb;
296 unsigned short id;
297 grant_ref_t ref;
298 struct page *page;
299 struct xen_netif_rx_request *req;
300
301 skb = xennet_alloc_one_rx_buffer(queue);
302 if (!skb) {
303 err = -ENOMEM;
304 break;
305 }
306
307 id = xennet_rxidx(req_prod);
308
309 BUG_ON(queue->rx_skbs[id]);
310 queue->rx_skbs[id] = skb;
311
312 ref = gnttab_claim_grant_reference(&queue->gref_rx_head);
313 WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref));
314 queue->grant_rx_ref[id] = ref;
315
316 page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
317
318 req = RING_GET_REQUEST(&queue->rx, req_prod);
319 gnttab_page_grant_foreign_access_ref_one(ref,
320 queue->info->xbdev->otherend_id,
321 page,
322 0);
323 req->id = id;
324 req->gref = ref;
325 }
326
327 queue->rx.req_prod_pvt = req_prod;
328
329 /* Try again later if there are not enough requests or skb allocation
330 * failed.
331 * Enough requests is quantified as the sum of newly created slots and
332 * the unconsumed slots at the backend.
333 */
334 if (req_prod - queue->rx.rsp_cons < NET_RX_SLOTS_MIN ||
335 unlikely(err)) {
336 mod_timer(&queue->rx_refill_timer, jiffies + (HZ/10));
337 return;
338 }
339
340 wmb(); /* barrier so backend seens requests */
341
342 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->rx, notify);
343 if (notify)
344 notify_remote_via_irq(queue->rx_irq);
345 }
346
347 static int xennet_open(struct net_device *dev)
348 {
349 struct netfront_info *np = netdev_priv(dev);
350 unsigned int num_queues = dev->real_num_tx_queues;
351 unsigned int i = 0;
352 struct netfront_queue *queue = NULL;
353
354 if (!np->queues)
355 return -ENODEV;
356
357 for (i = 0; i < num_queues; ++i) {
358 queue = &np->queues[i];
359 napi_enable(&queue->napi);
360
361 spin_lock_bh(&queue->rx_lock);
362 if (netif_carrier_ok(dev)) {
363 xennet_alloc_rx_buffers(queue);
364 queue->rx.sring->rsp_event = queue->rx.rsp_cons + 1;
365 if (RING_HAS_UNCONSUMED_RESPONSES(&queue->rx))
366 napi_schedule(&queue->napi);
367 }
368 spin_unlock_bh(&queue->rx_lock);
369 }
370
371 netif_tx_start_all_queues(dev);
372
373 return 0;
374 }
375
376 static void xennet_tx_buf_gc(struct netfront_queue *queue)
377 {
378 RING_IDX cons, prod;
379 unsigned short id;
380 struct sk_buff *skb;
381 bool more_to_do;
382
383 BUG_ON(!netif_carrier_ok(queue->info->netdev));
384
385 do {
386 prod = queue->tx.sring->rsp_prod;
387 rmb(); /* Ensure we see responses up to 'rp'. */
388
389 for (cons = queue->tx.rsp_cons; cons != prod; cons++) {
390 struct xen_netif_tx_response *txrsp;
391
392 txrsp = RING_GET_RESPONSE(&queue->tx, cons);
393 if (txrsp->status == XEN_NETIF_RSP_NULL)
394 continue;
395
396 id = txrsp->id;
397 skb = queue->tx_skbs[id].skb;
398 if (unlikely(gnttab_query_foreign_access(
399 queue->grant_tx_ref[id]) != 0)) {
400 pr_alert("%s: warning -- grant still in use by backend domain\n",
401 __func__);
402 BUG();
403 }
404 gnttab_end_foreign_access_ref(
405 queue->grant_tx_ref[id], GNTMAP_readonly);
406 gnttab_release_grant_reference(
407 &queue->gref_tx_head, queue->grant_tx_ref[id]);
408 queue->grant_tx_ref[id] = GRANT_INVALID_REF;
409 queue->grant_tx_page[id] = NULL;
410 add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, id);
411 dev_kfree_skb_irq(skb);
412 }
413
414 queue->tx.rsp_cons = prod;
415
416 RING_FINAL_CHECK_FOR_RESPONSES(&queue->tx, more_to_do);
417 } while (more_to_do);
418
419 xennet_maybe_wake_tx(queue);
420 }
421
422 struct xennet_gnttab_make_txreq {
423 struct netfront_queue *queue;
424 struct sk_buff *skb;
425 struct page *page;
426 struct xen_netif_tx_request *tx; /* Last request */
427 unsigned int size;
428 };
429
430 static void xennet_tx_setup_grant(unsigned long gfn, unsigned int offset,
431 unsigned int len, void *data)
432 {
433 struct xennet_gnttab_make_txreq *info = data;
434 unsigned int id;
435 struct xen_netif_tx_request *tx;
436 grant_ref_t ref;
437 /* convenient aliases */
438 struct page *page = info->page;
439 struct netfront_queue *queue = info->queue;
440 struct sk_buff *skb = info->skb;
441
442 id = get_id_from_freelist(&queue->tx_skb_freelist, queue->tx_skbs);
443 tx = RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++);
444 ref = gnttab_claim_grant_reference(&queue->gref_tx_head);
445 WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref));
446
447 gnttab_grant_foreign_access_ref(ref, queue->info->xbdev->otherend_id,
448 gfn, GNTMAP_readonly);
449
450 queue->tx_skbs[id].skb = skb;
451 queue->grant_tx_page[id] = page;
452 queue->grant_tx_ref[id] = ref;
453
454 tx->id = id;
455 tx->gref = ref;
456 tx->offset = offset;
457 tx->size = len;
458 tx->flags = 0;
459
460 info->tx = tx;
461 info->size += tx->size;
462 }
463
464 static struct xen_netif_tx_request *xennet_make_first_txreq(
465 struct netfront_queue *queue, struct sk_buff *skb,
466 struct page *page, unsigned int offset, unsigned int len)
467 {
468 struct xennet_gnttab_make_txreq info = {
469 .queue = queue,
470 .skb = skb,
471 .page = page,
472 .size = 0,
473 };
474
475 gnttab_for_one_grant(page, offset, len, xennet_tx_setup_grant, &info);
476
477 return info.tx;
478 }
479
480 static void xennet_make_one_txreq(unsigned long gfn, unsigned int offset,
481 unsigned int len, void *data)
482 {
483 struct xennet_gnttab_make_txreq *info = data;
484
485 info->tx->flags |= XEN_NETTXF_more_data;
486 skb_get(info->skb);
487 xennet_tx_setup_grant(gfn, offset, len, data);
488 }
489
490 static struct xen_netif_tx_request *xennet_make_txreqs(
491 struct netfront_queue *queue, struct xen_netif_tx_request *tx,
492 struct sk_buff *skb, struct page *page,
493 unsigned int offset, unsigned int len)
494 {
495 struct xennet_gnttab_make_txreq info = {
496 .queue = queue,
497 .skb = skb,
498 .tx = tx,
499 };
500
501 /* Skip unused frames from start of page */
502 page += offset >> PAGE_SHIFT;
503 offset &= ~PAGE_MASK;
504
505 while (len) {
506 info.page = page;
507 info.size = 0;
508
509 gnttab_foreach_grant_in_range(page, offset, len,
510 xennet_make_one_txreq,
511 &info);
512
513 page++;
514 offset = 0;
515 len -= info.size;
516 }
517
518 return info.tx;
519 }
520
521 /*
522 * Count how many ring slots are required to send this skb. Each frag
523 * might be a compound page.
524 */
525 static int xennet_count_skb_slots(struct sk_buff *skb)
526 {
527 int i, frags = skb_shinfo(skb)->nr_frags;
528 int slots;
529
530 slots = gnttab_count_grant(offset_in_page(skb->data),
531 skb_headlen(skb));
532
533 for (i = 0; i < frags; i++) {
534 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
535 unsigned long size = skb_frag_size(frag);
536 unsigned long offset = frag->page_offset;
537
538 /* Skip unused frames from start of page */
539 offset &= ~PAGE_MASK;
540
541 slots += gnttab_count_grant(offset, size);
542 }
543
544 return slots;
545 }
546
547 static u16 xennet_select_queue(struct net_device *dev, struct sk_buff *skb,
548 struct net_device *sb_dev,
549 select_queue_fallback_t fallback)
550 {
551 unsigned int num_queues = dev->real_num_tx_queues;
552 u32 hash;
553 u16 queue_idx;
554
555 /* First, check if there is only one queue */
556 if (num_queues == 1) {
557 queue_idx = 0;
558 } else {
559 hash = skb_get_hash(skb);
560 queue_idx = hash % num_queues;
561 }
562
563 return queue_idx;
564 }
565
566 #define MAX_XEN_SKB_FRAGS (65536 / XEN_PAGE_SIZE + 1)
567
568 static netdev_tx_t xennet_start_xmit(struct sk_buff *skb, struct net_device *dev)
569 {
570 struct netfront_info *np = netdev_priv(dev);
571 struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats);
572 struct xen_netif_tx_request *tx, *first_tx;
573 unsigned int i;
574 int notify;
575 int slots;
576 struct page *page;
577 unsigned int offset;
578 unsigned int len;
579 unsigned long flags;
580 struct netfront_queue *queue = NULL;
581 unsigned int num_queues = dev->real_num_tx_queues;
582 u16 queue_index;
583 struct sk_buff *nskb;
584
585 /* Drop the packet if no queues are set up */
586 if (num_queues < 1)
587 goto drop;
588 /* Determine which queue to transmit this SKB on */
589 queue_index = skb_get_queue_mapping(skb);
590 queue = &np->queues[queue_index];
591
592 /* If skb->len is too big for wire format, drop skb and alert
593 * user about misconfiguration.
594 */
595 if (unlikely(skb->len > XEN_NETIF_MAX_TX_SIZE)) {
596 net_alert_ratelimited(
597 "xennet: skb->len = %u, too big for wire format\n",
598 skb->len);
599 goto drop;
600 }
601
602 slots = xennet_count_skb_slots(skb);
603 if (unlikely(slots > MAX_XEN_SKB_FRAGS + 1)) {
604 net_dbg_ratelimited("xennet: skb rides the rocket: %d slots, %d bytes\n",
605 slots, skb->len);
606 if (skb_linearize(skb))
607 goto drop;
608 }
609
610 page = virt_to_page(skb->data);
611 offset = offset_in_page(skb->data);
612
613 /* The first req should be at least ETH_HLEN size or the packet will be
614 * dropped by netback.
615 */
616 if (unlikely(PAGE_SIZE - offset < ETH_HLEN)) {
617 nskb = skb_copy(skb, GFP_ATOMIC);
618 if (!nskb)
619 goto drop;
620 dev_consume_skb_any(skb);
621 skb = nskb;
622 page = virt_to_page(skb->data);
623 offset = offset_in_page(skb->data);
624 }
625
626 len = skb_headlen(skb);
627
628 spin_lock_irqsave(&queue->tx_lock, flags);
629
630 if (unlikely(!netif_carrier_ok(dev) ||
631 (slots > 1 && !xennet_can_sg(dev)) ||
632 netif_needs_gso(skb, netif_skb_features(skb)))) {
633 spin_unlock_irqrestore(&queue->tx_lock, flags);
634 goto drop;
635 }
636
637 /* First request for the linear area. */
638 first_tx = tx = xennet_make_first_txreq(queue, skb,
639 page, offset, len);
640 offset += tx->size;
641 if (offset == PAGE_SIZE) {
642 page++;
643 offset = 0;
644 }
645 len -= tx->size;
646
647 if (skb->ip_summed == CHECKSUM_PARTIAL)
648 /* local packet? */
649 tx->flags |= XEN_NETTXF_csum_blank | XEN_NETTXF_data_validated;
650 else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
651 /* remote but checksummed. */
652 tx->flags |= XEN_NETTXF_data_validated;
653
654 /* Optional extra info after the first request. */
655 if (skb_shinfo(skb)->gso_size) {
656 struct xen_netif_extra_info *gso;
657
658 gso = (struct xen_netif_extra_info *)
659 RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++);
660
661 tx->flags |= XEN_NETTXF_extra_info;
662
663 gso->u.gso.size = skb_shinfo(skb)->gso_size;
664 gso->u.gso.type = (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) ?
665 XEN_NETIF_GSO_TYPE_TCPV6 :
666 XEN_NETIF_GSO_TYPE_TCPV4;
667 gso->u.gso.pad = 0;
668 gso->u.gso.features = 0;
669
670 gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
671 gso->flags = 0;
672 }
673
674 /* Requests for the rest of the linear area. */
675 tx = xennet_make_txreqs(queue, tx, skb, page, offset, len);
676
677 /* Requests for all the frags. */
678 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
679 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
680 tx = xennet_make_txreqs(queue, tx, skb,
681 skb_frag_page(frag), frag->page_offset,
682 skb_frag_size(frag));
683 }
684
685 /* First request has the packet length. */
686 first_tx->size = skb->len;
687
688 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify);
689 if (notify)
690 notify_remote_via_irq(queue->tx_irq);
691
692 u64_stats_update_begin(&tx_stats->syncp);
693 tx_stats->bytes += skb->len;
694 tx_stats->packets++;
695 u64_stats_update_end(&tx_stats->syncp);
696
697 /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
698 xennet_tx_buf_gc(queue);
699
700 if (!netfront_tx_slot_available(queue))
701 netif_tx_stop_queue(netdev_get_tx_queue(dev, queue->id));
702
703 spin_unlock_irqrestore(&queue->tx_lock, flags);
704
705 return NETDEV_TX_OK;
706
707 drop:
708 dev->stats.tx_dropped++;
709 dev_kfree_skb_any(skb);
710 return NETDEV_TX_OK;
711 }
712
713 static int xennet_close(struct net_device *dev)
714 {
715 struct netfront_info *np = netdev_priv(dev);
716 unsigned int num_queues = dev->real_num_tx_queues;
717 unsigned int i;
718 struct netfront_queue *queue;
719 netif_tx_stop_all_queues(np->netdev);
720 for (i = 0; i < num_queues; ++i) {
721 queue = &np->queues[i];
722 napi_disable(&queue->napi);
723 }
724 return 0;
725 }
726
727 static void xennet_move_rx_slot(struct netfront_queue *queue, struct sk_buff *skb,
728 grant_ref_t ref)
729 {
730 int new = xennet_rxidx(queue->rx.req_prod_pvt);
731
732 BUG_ON(queue->rx_skbs[new]);
733 queue->rx_skbs[new] = skb;
734 queue->grant_rx_ref[new] = ref;
735 RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->id = new;
736 RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->gref = ref;
737 queue->rx.req_prod_pvt++;
738 }
739
740 static int xennet_get_extras(struct netfront_queue *queue,
741 struct xen_netif_extra_info *extras,
742 RING_IDX rp)
743
744 {
745 struct xen_netif_extra_info *extra;
746 struct device *dev = &queue->info->netdev->dev;
747 RING_IDX cons = queue->rx.rsp_cons;
748 int err = 0;
749
750 do {
751 struct sk_buff *skb;
752 grant_ref_t ref;
753
754 if (unlikely(cons + 1 == rp)) {
755 if (net_ratelimit())
756 dev_warn(dev, "Missing extra info\n");
757 err = -EBADR;
758 break;
759 }
760
761 extra = (struct xen_netif_extra_info *)
762 RING_GET_RESPONSE(&queue->rx, ++cons);
763
764 if (unlikely(!extra->type ||
765 extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
766 if (net_ratelimit())
767 dev_warn(dev, "Invalid extra type: %d\n",
768 extra->type);
769 err = -EINVAL;
770 } else {
771 memcpy(&extras[extra->type - 1], extra,
772 sizeof(*extra));
773 }
774
775 skb = xennet_get_rx_skb(queue, cons);
776 ref = xennet_get_rx_ref(queue, cons);
777 xennet_move_rx_slot(queue, skb, ref);
778 } while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE);
779
780 queue->rx.rsp_cons = cons;
781 return err;
782 }
783
784 static int xennet_get_responses(struct netfront_queue *queue,
785 struct netfront_rx_info *rinfo, RING_IDX rp,
786 struct sk_buff_head *list)
787 {
788 struct xen_netif_rx_response *rx = &rinfo->rx;
789 struct xen_netif_extra_info *extras = rinfo->extras;
790 struct device *dev = &queue->info->netdev->dev;
791 RING_IDX cons = queue->rx.rsp_cons;
792 struct sk_buff *skb = xennet_get_rx_skb(queue, cons);
793 grant_ref_t ref = xennet_get_rx_ref(queue, cons);
794 int max = XEN_NETIF_NR_SLOTS_MIN + (rx->status <= RX_COPY_THRESHOLD);
795 int slots = 1;
796 int err = 0;
797 unsigned long ret;
798
799 if (rx->flags & XEN_NETRXF_extra_info) {
800 err = xennet_get_extras(queue, extras, rp);
801 cons = queue->rx.rsp_cons;
802 }
803
804 for (;;) {
805 if (unlikely(rx->status < 0 ||
806 rx->offset + rx->status > XEN_PAGE_SIZE)) {
807 if (net_ratelimit())
808 dev_warn(dev, "rx->offset: %u, size: %d\n",
809 rx->offset, rx->status);
810 xennet_move_rx_slot(queue, skb, ref);
811 err = -EINVAL;
812 goto next;
813 }
814
815 /*
816 * This definitely indicates a bug, either in this driver or in
817 * the backend driver. In future this should flag the bad
818 * situation to the system controller to reboot the backend.
819 */
820 if (ref == GRANT_INVALID_REF) {
821 if (net_ratelimit())
822 dev_warn(dev, "Bad rx response id %d.\n",
823 rx->id);
824 err = -EINVAL;
825 goto next;
826 }
827
828 ret = gnttab_end_foreign_access_ref(ref, 0);
829 BUG_ON(!ret);
830
831 gnttab_release_grant_reference(&queue->gref_rx_head, ref);
832
833 __skb_queue_tail(list, skb);
834
835 next:
836 if (!(rx->flags & XEN_NETRXF_more_data))
837 break;
838
839 if (cons + slots == rp) {
840 if (net_ratelimit())
841 dev_warn(dev, "Need more slots\n");
842 err = -ENOENT;
843 break;
844 }
845
846 rx = RING_GET_RESPONSE(&queue->rx, cons + slots);
847 skb = xennet_get_rx_skb(queue, cons + slots);
848 ref = xennet_get_rx_ref(queue, cons + slots);
849 slots++;
850 }
851
852 if (unlikely(slots > max)) {
853 if (net_ratelimit())
854 dev_warn(dev, "Too many slots\n");
855 err = -E2BIG;
856 }
857
858 if (unlikely(err))
859 queue->rx.rsp_cons = cons + slots;
860
861 return err;
862 }
863
864 static int xennet_set_skb_gso(struct sk_buff *skb,
865 struct xen_netif_extra_info *gso)
866 {
867 if (!gso->u.gso.size) {
868 if (net_ratelimit())
869 pr_warn("GSO size must not be zero\n");
870 return -EINVAL;
871 }
872
873 if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4 &&
874 gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV6) {
875 if (net_ratelimit())
876 pr_warn("Bad GSO type %d\n", gso->u.gso.type);
877 return -EINVAL;
878 }
879
880 skb_shinfo(skb)->gso_size = gso->u.gso.size;
881 skb_shinfo(skb)->gso_type =
882 (gso->u.gso.type == XEN_NETIF_GSO_TYPE_TCPV4) ?
883 SKB_GSO_TCPV4 :
884 SKB_GSO_TCPV6;
885
886 /* Header must be checked, and gso_segs computed. */
887 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
888 skb_shinfo(skb)->gso_segs = 0;
889
890 return 0;
891 }
892
893 static int xennet_fill_frags(struct netfront_queue *queue,
894 struct sk_buff *skb,
895 struct sk_buff_head *list)
896 {
897 RING_IDX cons = queue->rx.rsp_cons;
898 struct sk_buff *nskb;
899
900 while ((nskb = __skb_dequeue(list))) {
901 struct xen_netif_rx_response *rx =
902 RING_GET_RESPONSE(&queue->rx, ++cons);
903 skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0];
904
905 if (skb_shinfo(skb)->nr_frags == MAX_SKB_FRAGS) {
906 unsigned int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
907
908 BUG_ON(pull_to < skb_headlen(skb));
909 __pskb_pull_tail(skb, pull_to - skb_headlen(skb));
910 }
911 if (unlikely(skb_shinfo(skb)->nr_frags >= MAX_SKB_FRAGS)) {
912 queue->rx.rsp_cons = ++cons + skb_queue_len(list);
913 kfree_skb(nskb);
914 return -ENOENT;
915 }
916
917 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags,
918 skb_frag_page(nfrag),
919 rx->offset, rx->status, PAGE_SIZE);
920
921 skb_shinfo(nskb)->nr_frags = 0;
922 kfree_skb(nskb);
923 }
924
925 queue->rx.rsp_cons = cons;
926
927 return 0;
928 }
929
930 static int checksum_setup(struct net_device *dev, struct sk_buff *skb)
931 {
932 bool recalculate_partial_csum = false;
933
934 /*
935 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
936 * peers can fail to set NETRXF_csum_blank when sending a GSO
937 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
938 * recalculate the partial checksum.
939 */
940 if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
941 struct netfront_info *np = netdev_priv(dev);
942 atomic_inc(&np->rx_gso_checksum_fixup);
943 skb->ip_summed = CHECKSUM_PARTIAL;
944 recalculate_partial_csum = true;
945 }
946
947 /* A non-CHECKSUM_PARTIAL SKB does not require setup. */
948 if (skb->ip_summed != CHECKSUM_PARTIAL)
949 return 0;
950
951 return skb_checksum_setup(skb, recalculate_partial_csum);
952 }
953
954 static int handle_incoming_queue(struct netfront_queue *queue,
955 struct sk_buff_head *rxq)
956 {
957 struct netfront_stats *rx_stats = this_cpu_ptr(queue->info->rx_stats);
958 int packets_dropped = 0;
959 struct sk_buff *skb;
960
961 while ((skb = __skb_dequeue(rxq)) != NULL) {
962 int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
963
964 if (pull_to > skb_headlen(skb))
965 __pskb_pull_tail(skb, pull_to - skb_headlen(skb));
966
967 /* Ethernet work: Delayed to here as it peeks the header. */
968 skb->protocol = eth_type_trans(skb, queue->info->netdev);
969 skb_reset_network_header(skb);
970
971 if (checksum_setup(queue->info->netdev, skb)) {
972 kfree_skb(skb);
973 packets_dropped++;
974 queue->info->netdev->stats.rx_errors++;
975 continue;
976 }
977
978 u64_stats_update_begin(&rx_stats->syncp);
979 rx_stats->packets++;
980 rx_stats->bytes += skb->len;
981 u64_stats_update_end(&rx_stats->syncp);
982
983 /* Pass it up. */
984 napi_gro_receive(&queue->napi, skb);
985 }
986
987 return packets_dropped;
988 }
989
990 static int xennet_poll(struct napi_struct *napi, int budget)
991 {
992 struct netfront_queue *queue = container_of(napi, struct netfront_queue, napi);
993 struct net_device *dev = queue->info->netdev;
994 struct sk_buff *skb;
995 struct netfront_rx_info rinfo;
996 struct xen_netif_rx_response *rx = &rinfo.rx;
997 struct xen_netif_extra_info *extras = rinfo.extras;
998 RING_IDX i, rp;
999 int work_done;
1000 struct sk_buff_head rxq;
1001 struct sk_buff_head errq;
1002 struct sk_buff_head tmpq;
1003 int err;
1004
1005 spin_lock(&queue->rx_lock);
1006
1007 skb_queue_head_init(&rxq);
1008 skb_queue_head_init(&errq);
1009 skb_queue_head_init(&tmpq);
1010
1011 rp = queue->rx.sring->rsp_prod;
1012 rmb(); /* Ensure we see queued responses up to 'rp'. */
1013
1014 i = queue->rx.rsp_cons;
1015 work_done = 0;
1016 while ((i != rp) && (work_done < budget)) {
1017 memcpy(rx, RING_GET_RESPONSE(&queue->rx, i), sizeof(*rx));
1018 memset(extras, 0, sizeof(rinfo.extras));
1019
1020 err = xennet_get_responses(queue, &rinfo, rp, &tmpq);
1021
1022 if (unlikely(err)) {
1023 err:
1024 while ((skb = __skb_dequeue(&tmpq)))
1025 __skb_queue_tail(&errq, skb);
1026 dev->stats.rx_errors++;
1027 i = queue->rx.rsp_cons;
1028 continue;
1029 }
1030
1031 skb = __skb_dequeue(&tmpq);
1032
1033 if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
1034 struct xen_netif_extra_info *gso;
1035 gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
1036
1037 if (unlikely(xennet_set_skb_gso(skb, gso))) {
1038 __skb_queue_head(&tmpq, skb);
1039 queue->rx.rsp_cons += skb_queue_len(&tmpq);
1040 goto err;
1041 }
1042 }
1043
1044 NETFRONT_SKB_CB(skb)->pull_to = rx->status;
1045 if (NETFRONT_SKB_CB(skb)->pull_to > RX_COPY_THRESHOLD)
1046 NETFRONT_SKB_CB(skb)->pull_to = RX_COPY_THRESHOLD;
1047
1048 skb_shinfo(skb)->frags[0].page_offset = rx->offset;
1049 skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status);
1050 skb->data_len = rx->status;
1051 skb->len += rx->status;
1052
1053 if (unlikely(xennet_fill_frags(queue, skb, &tmpq)))
1054 goto err;
1055
1056 if (rx->flags & XEN_NETRXF_csum_blank)
1057 skb->ip_summed = CHECKSUM_PARTIAL;
1058 else if (rx->flags & XEN_NETRXF_data_validated)
1059 skb->ip_summed = CHECKSUM_UNNECESSARY;
1060
1061 __skb_queue_tail(&rxq, skb);
1062
1063 i = ++queue->rx.rsp_cons;
1064 work_done++;
1065 }
1066
1067 __skb_queue_purge(&errq);
1068
1069 work_done -= handle_incoming_queue(queue, &rxq);
1070
1071 xennet_alloc_rx_buffers(queue);
1072
1073 if (work_done < budget) {
1074 int more_to_do = 0;
1075
1076 napi_complete_done(napi, work_done);
1077
1078 RING_FINAL_CHECK_FOR_RESPONSES(&queue->rx, more_to_do);
1079 if (more_to_do)
1080 napi_schedule(napi);
1081 }
1082
1083 spin_unlock(&queue->rx_lock);
1084
1085 return work_done;
1086 }
1087
1088 static int xennet_change_mtu(struct net_device *dev, int mtu)
1089 {
1090 int max = xennet_can_sg(dev) ? XEN_NETIF_MAX_TX_SIZE : ETH_DATA_LEN;
1091
1092 if (mtu > max)
1093 return -EINVAL;
1094 dev->mtu = mtu;
1095 return 0;
1096 }
1097
1098 static void xennet_get_stats64(struct net_device *dev,
1099 struct rtnl_link_stats64 *tot)
1100 {
1101 struct netfront_info *np = netdev_priv(dev);
1102 int cpu;
1103
1104 for_each_possible_cpu(cpu) {
1105 struct netfront_stats *rx_stats = per_cpu_ptr(np->rx_stats, cpu);
1106 struct netfront_stats *tx_stats = per_cpu_ptr(np->tx_stats, cpu);
1107 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1108 unsigned int start;
1109
1110 do {
1111 start = u64_stats_fetch_begin_irq(&tx_stats->syncp);
1112 tx_packets = tx_stats->packets;
1113 tx_bytes = tx_stats->bytes;
1114 } while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start));
1115
1116 do {
1117 start = u64_stats_fetch_begin_irq(&rx_stats->syncp);
1118 rx_packets = rx_stats->packets;
1119 rx_bytes = rx_stats->bytes;
1120 } while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start));
1121
1122 tot->rx_packets += rx_packets;
1123 tot->tx_packets += tx_packets;
1124 tot->rx_bytes += rx_bytes;
1125 tot->tx_bytes += tx_bytes;
1126 }
1127
1128 tot->rx_errors = dev->stats.rx_errors;
1129 tot->tx_dropped = dev->stats.tx_dropped;
1130 }
1131
1132 static void xennet_release_tx_bufs(struct netfront_queue *queue)
1133 {
1134 struct sk_buff *skb;
1135 int i;
1136
1137 for (i = 0; i < NET_TX_RING_SIZE; i++) {
1138 /* Skip over entries which are actually freelist references */
1139 if (skb_entry_is_link(&queue->tx_skbs[i]))
1140 continue;
1141
1142 skb = queue->tx_skbs[i].skb;
1143 get_page(queue->grant_tx_page[i]);
1144 gnttab_end_foreign_access(queue->grant_tx_ref[i],
1145 GNTMAP_readonly,
1146 (unsigned long)page_address(queue->grant_tx_page[i]));
1147 queue->grant_tx_page[i] = NULL;
1148 queue->grant_tx_ref[i] = GRANT_INVALID_REF;
1149 add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, i);
1150 dev_kfree_skb_irq(skb);
1151 }
1152 }
1153
1154 static void xennet_release_rx_bufs(struct netfront_queue *queue)
1155 {
1156 int id, ref;
1157
1158 spin_lock_bh(&queue->rx_lock);
1159
1160 for (id = 0; id < NET_RX_RING_SIZE; id++) {
1161 struct sk_buff *skb;
1162 struct page *page;
1163
1164 skb = queue->rx_skbs[id];
1165 if (!skb)
1166 continue;
1167
1168 ref = queue->grant_rx_ref[id];
1169 if (ref == GRANT_INVALID_REF)
1170 continue;
1171
1172 page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
1173
1174 /* gnttab_end_foreign_access() needs a page ref until
1175 * foreign access is ended (which may be deferred).
1176 */
1177 get_page(page);
1178 gnttab_end_foreign_access(ref, 0,
1179 (unsigned long)page_address(page));
1180 queue->grant_rx_ref[id] = GRANT_INVALID_REF;
1181
1182 kfree_skb(skb);
1183 }
1184
1185 spin_unlock_bh(&queue->rx_lock);
1186 }
1187
1188 static netdev_features_t xennet_fix_features(struct net_device *dev,
1189 netdev_features_t features)
1190 {
1191 struct netfront_info *np = netdev_priv(dev);
1192
1193 if (features & NETIF_F_SG &&
1194 !xenbus_read_unsigned(np->xbdev->otherend, "feature-sg", 0))
1195 features &= ~NETIF_F_SG;
1196
1197 if (features & NETIF_F_IPV6_CSUM &&
1198 !xenbus_read_unsigned(np->xbdev->otherend,
1199 "feature-ipv6-csum-offload", 0))
1200 features &= ~NETIF_F_IPV6_CSUM;
1201
1202 if (features & NETIF_F_TSO &&
1203 !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv4", 0))
1204 features &= ~NETIF_F_TSO;
1205
1206 if (features & NETIF_F_TSO6 &&
1207 !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv6", 0))
1208 features &= ~NETIF_F_TSO6;
1209
1210 return features;
1211 }
1212
1213 static int xennet_set_features(struct net_device *dev,
1214 netdev_features_t features)
1215 {
1216 if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) {
1217 netdev_info(dev, "Reducing MTU because no SG offload");
1218 dev->mtu = ETH_DATA_LEN;
1219 }
1220
1221 return 0;
1222 }
1223
1224 static irqreturn_t xennet_tx_interrupt(int irq, void *dev_id)
1225 {
1226 struct netfront_queue *queue = dev_id;
1227 unsigned long flags;
1228
1229 spin_lock_irqsave(&queue->tx_lock, flags);
1230 xennet_tx_buf_gc(queue);
1231 spin_unlock_irqrestore(&queue->tx_lock, flags);
1232
1233 return IRQ_HANDLED;
1234 }
1235
1236 static irqreturn_t xennet_rx_interrupt(int irq, void *dev_id)
1237 {
1238 struct netfront_queue *queue = dev_id;
1239 struct net_device *dev = queue->info->netdev;
1240
1241 if (likely(netif_carrier_ok(dev) &&
1242 RING_HAS_UNCONSUMED_RESPONSES(&queue->rx)))
1243 napi_schedule(&queue->napi);
1244
1245 return IRQ_HANDLED;
1246 }
1247
1248 static irqreturn_t xennet_interrupt(int irq, void *dev_id)
1249 {
1250 xennet_tx_interrupt(irq, dev_id);
1251 xennet_rx_interrupt(irq, dev_id);
1252 return IRQ_HANDLED;
1253 }
1254
1255 #ifdef CONFIG_NET_POLL_CONTROLLER
1256 static void xennet_poll_controller(struct net_device *dev)
1257 {
1258 /* Poll each queue */
1259 struct netfront_info *info = netdev_priv(dev);
1260 unsigned int num_queues = dev->real_num_tx_queues;
1261 unsigned int i;
1262 for (i = 0; i < num_queues; ++i)
1263 xennet_interrupt(0, &info->queues[i]);
1264 }
1265 #endif
1266
1267 static const struct net_device_ops xennet_netdev_ops = {
1268 .ndo_open = xennet_open,
1269 .ndo_stop = xennet_close,
1270 .ndo_start_xmit = xennet_start_xmit,
1271 .ndo_change_mtu = xennet_change_mtu,
1272 .ndo_get_stats64 = xennet_get_stats64,
1273 .ndo_set_mac_address = eth_mac_addr,
1274 .ndo_validate_addr = eth_validate_addr,
1275 .ndo_fix_features = xennet_fix_features,
1276 .ndo_set_features = xennet_set_features,
1277 .ndo_select_queue = xennet_select_queue,
1278 #ifdef CONFIG_NET_POLL_CONTROLLER
1279 .ndo_poll_controller = xennet_poll_controller,
1280 #endif
1281 };
1282
1283 static void xennet_free_netdev(struct net_device *netdev)
1284 {
1285 struct netfront_info *np = netdev_priv(netdev);
1286
1287 free_percpu(np->rx_stats);
1288 free_percpu(np->tx_stats);
1289 free_netdev(netdev);
1290 }
1291
1292 static struct net_device *xennet_create_dev(struct xenbus_device *dev)
1293 {
1294 int err;
1295 struct net_device *netdev;
1296 struct netfront_info *np;
1297
1298 netdev = alloc_etherdev_mq(sizeof(struct netfront_info), xennet_max_queues);
1299 if (!netdev)
1300 return ERR_PTR(-ENOMEM);
1301
1302 np = netdev_priv(netdev);
1303 np->xbdev = dev;
1304
1305 np->queues = NULL;
1306
1307 err = -ENOMEM;
1308 np->rx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1309 if (np->rx_stats == NULL)
1310 goto exit;
1311 np->tx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1312 if (np->tx_stats == NULL)
1313 goto exit;
1314
1315 netdev->netdev_ops = &xennet_netdev_ops;
1316
1317 netdev->features = NETIF_F_IP_CSUM | NETIF_F_RXCSUM |
1318 NETIF_F_GSO_ROBUST;
1319 netdev->hw_features = NETIF_F_SG |
1320 NETIF_F_IPV6_CSUM |
1321 NETIF_F_TSO | NETIF_F_TSO6;
1322
1323 /*
1324 * Assume that all hw features are available for now. This set
1325 * will be adjusted by the call to netdev_update_features() in
1326 * xennet_connect() which is the earliest point where we can
1327 * negotiate with the backend regarding supported features.
1328 */
1329 netdev->features |= netdev->hw_features;
1330
1331 netdev->ethtool_ops = &xennet_ethtool_ops;
1332 netdev->min_mtu = ETH_MIN_MTU;
1333 netdev->max_mtu = XEN_NETIF_MAX_TX_SIZE;
1334 SET_NETDEV_DEV(netdev, &dev->dev);
1335
1336 np->netdev = netdev;
1337
1338 netif_carrier_off(netdev);
1339
1340 xenbus_switch_state(dev, XenbusStateInitialising);
1341 wait_event(module_wq,
1342 xenbus_read_driver_state(dev->otherend) !=
1343 XenbusStateClosed &&
1344 xenbus_read_driver_state(dev->otherend) !=
1345 XenbusStateUnknown);
1346 return netdev;
1347
1348 exit:
1349 xennet_free_netdev(netdev);
1350 return ERR_PTR(err);
1351 }
1352
1353 /**
1354 * Entry point to this code when a new device is created. Allocate the basic
1355 * structures and the ring buffers for communication with the backend, and
1356 * inform the backend of the appropriate details for those.
1357 */
1358 static int netfront_probe(struct xenbus_device *dev,
1359 const struct xenbus_device_id *id)
1360 {
1361 int err;
1362 struct net_device *netdev;
1363 struct netfront_info *info;
1364
1365 netdev = xennet_create_dev(dev);
1366 if (IS_ERR(netdev)) {
1367 err = PTR_ERR(netdev);
1368 xenbus_dev_fatal(dev, err, "creating netdev");
1369 return err;
1370 }
1371
1372 info = netdev_priv(netdev);
1373 dev_set_drvdata(&dev->dev, info);
1374 #ifdef CONFIG_SYSFS
1375 info->netdev->sysfs_groups[0] = &xennet_dev_group;
1376 #endif
1377
1378 return 0;
1379 }
1380
1381 static void xennet_end_access(int ref, void *page)
1382 {
1383 /* This frees the page as a side-effect */
1384 if (ref != GRANT_INVALID_REF)
1385 gnttab_end_foreign_access(ref, 0, (unsigned long)page);
1386 }
1387
1388 static void xennet_disconnect_backend(struct netfront_info *info)
1389 {
1390 unsigned int i = 0;
1391 unsigned int num_queues = info->netdev->real_num_tx_queues;
1392
1393 netif_carrier_off(info->netdev);
1394
1395 for (i = 0; i < num_queues && info->queues; ++i) {
1396 struct netfront_queue *queue = &info->queues[i];
1397
1398 del_timer_sync(&queue->rx_refill_timer);
1399
1400 if (queue->tx_irq && (queue->tx_irq == queue->rx_irq))
1401 unbind_from_irqhandler(queue->tx_irq, queue);
1402 if (queue->tx_irq && (queue->tx_irq != queue->rx_irq)) {
1403 unbind_from_irqhandler(queue->tx_irq, queue);
1404 unbind_from_irqhandler(queue->rx_irq, queue);
1405 }
1406 queue->tx_evtchn = queue->rx_evtchn = 0;
1407 queue->tx_irq = queue->rx_irq = 0;
1408
1409 if (netif_running(info->netdev))
1410 napi_synchronize(&queue->napi);
1411
1412 xennet_release_tx_bufs(queue);
1413 xennet_release_rx_bufs(queue);
1414 gnttab_free_grant_references(queue->gref_tx_head);
1415 gnttab_free_grant_references(queue->gref_rx_head);
1416
1417 /* End access and free the pages */
1418 xennet_end_access(queue->tx_ring_ref, queue->tx.sring);
1419 xennet_end_access(queue->rx_ring_ref, queue->rx.sring);
1420
1421 queue->tx_ring_ref = GRANT_INVALID_REF;
1422 queue->rx_ring_ref = GRANT_INVALID_REF;
1423 queue->tx.sring = NULL;
1424 queue->rx.sring = NULL;
1425 }
1426 }
1427
1428 /**
1429 * We are reconnecting to the backend, due to a suspend/resume, or a backend
1430 * driver restart. We tear down our netif structure and recreate it, but
1431 * leave the device-layer structures intact so that this is transparent to the
1432 * rest of the kernel.
1433 */
1434 static int netfront_resume(struct xenbus_device *dev)
1435 {
1436 struct netfront_info *info = dev_get_drvdata(&dev->dev);
1437
1438 dev_dbg(&dev->dev, "%s\n", dev->nodename);
1439
1440 xennet_disconnect_backend(info);
1441 return 0;
1442 }
1443
1444 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
1445 {
1446 char *s, *e, *macstr;
1447 int i;
1448
1449 macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
1450 if (IS_ERR(macstr))
1451 return PTR_ERR(macstr);
1452
1453 for (i = 0; i < ETH_ALEN; i++) {
1454 mac[i] = simple_strtoul(s, &e, 16);
1455 if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
1456 kfree(macstr);
1457 return -ENOENT;
1458 }
1459 s = e+1;
1460 }
1461
1462 kfree(macstr);
1463 return 0;
1464 }
1465
1466 static int setup_netfront_single(struct netfront_queue *queue)
1467 {
1468 int err;
1469
1470 err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1471 if (err < 0)
1472 goto fail;
1473
1474 err = bind_evtchn_to_irqhandler(queue->tx_evtchn,
1475 xennet_interrupt,
1476 0, queue->info->netdev->name, queue);
1477 if (err < 0)
1478 goto bind_fail;
1479 queue->rx_evtchn = queue->tx_evtchn;
1480 queue->rx_irq = queue->tx_irq = err;
1481
1482 return 0;
1483
1484 bind_fail:
1485 xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1486 queue->tx_evtchn = 0;
1487 fail:
1488 return err;
1489 }
1490
1491 static int setup_netfront_split(struct netfront_queue *queue)
1492 {
1493 int err;
1494
1495 err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1496 if (err < 0)
1497 goto fail;
1498 err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->rx_evtchn);
1499 if (err < 0)
1500 goto alloc_rx_evtchn_fail;
1501
1502 snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
1503 "%s-tx", queue->name);
1504 err = bind_evtchn_to_irqhandler(queue->tx_evtchn,
1505 xennet_tx_interrupt,
1506 0, queue->tx_irq_name, queue);
1507 if (err < 0)
1508 goto bind_tx_fail;
1509 queue->tx_irq = err;
1510
1511 snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
1512 "%s-rx", queue->name);
1513 err = bind_evtchn_to_irqhandler(queue->rx_evtchn,
1514 xennet_rx_interrupt,
1515 0, queue->rx_irq_name, queue);
1516 if (err < 0)
1517 goto bind_rx_fail;
1518 queue->rx_irq = err;
1519
1520 return 0;
1521
1522 bind_rx_fail:
1523 unbind_from_irqhandler(queue->tx_irq, queue);
1524 queue->tx_irq = 0;
1525 bind_tx_fail:
1526 xenbus_free_evtchn(queue->info->xbdev, queue->rx_evtchn);
1527 queue->rx_evtchn = 0;
1528 alloc_rx_evtchn_fail:
1529 xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1530 queue->tx_evtchn = 0;
1531 fail:
1532 return err;
1533 }
1534
1535 static int setup_netfront(struct xenbus_device *dev,
1536 struct netfront_queue *queue, unsigned int feature_split_evtchn)
1537 {
1538 struct xen_netif_tx_sring *txs;
1539 struct xen_netif_rx_sring *rxs;
1540 grant_ref_t gref;
1541 int err;
1542
1543 queue->tx_ring_ref = GRANT_INVALID_REF;
1544 queue->rx_ring_ref = GRANT_INVALID_REF;
1545 queue->rx.sring = NULL;
1546 queue->tx.sring = NULL;
1547
1548 txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1549 if (!txs) {
1550 err = -ENOMEM;
1551 xenbus_dev_fatal(dev, err, "allocating tx ring page");
1552 goto fail;
1553 }
1554 SHARED_RING_INIT(txs);
1555 FRONT_RING_INIT(&queue->tx, txs, XEN_PAGE_SIZE);
1556
1557 err = xenbus_grant_ring(dev, txs, 1, &gref);
1558 if (err < 0)
1559 goto grant_tx_ring_fail;
1560 queue->tx_ring_ref = gref;
1561
1562 rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1563 if (!rxs) {
1564 err = -ENOMEM;
1565 xenbus_dev_fatal(dev, err, "allocating rx ring page");
1566 goto alloc_rx_ring_fail;
1567 }
1568 SHARED_RING_INIT(rxs);
1569 FRONT_RING_INIT(&queue->rx, rxs, XEN_PAGE_SIZE);
1570
1571 err = xenbus_grant_ring(dev, rxs, 1, &gref);
1572 if (err < 0)
1573 goto grant_rx_ring_fail;
1574 queue->rx_ring_ref = gref;
1575
1576 if (feature_split_evtchn)
1577 err = setup_netfront_split(queue);
1578 /* setup single event channel if
1579 * a) feature-split-event-channels == 0
1580 * b) feature-split-event-channels == 1 but failed to setup
1581 */
1582 if (!feature_split_evtchn || (feature_split_evtchn && err))
1583 err = setup_netfront_single(queue);
1584
1585 if (err)
1586 goto alloc_evtchn_fail;
1587
1588 return 0;
1589
1590 /* If we fail to setup netfront, it is safe to just revoke access to
1591 * granted pages because backend is not accessing it at this point.
1592 */
1593 alloc_evtchn_fail:
1594 gnttab_end_foreign_access_ref(queue->rx_ring_ref, 0);
1595 grant_rx_ring_fail:
1596 free_page((unsigned long)rxs);
1597 alloc_rx_ring_fail:
1598 gnttab_end_foreign_access_ref(queue->tx_ring_ref, 0);
1599 grant_tx_ring_fail:
1600 free_page((unsigned long)txs);
1601 fail:
1602 return err;
1603 }
1604
1605 /* Queue-specific initialisation
1606 * This used to be done in xennet_create_dev() but must now
1607 * be run per-queue.
1608 */
1609 static int xennet_init_queue(struct netfront_queue *queue)
1610 {
1611 unsigned short i;
1612 int err = 0;
1613 char *devid;
1614
1615 spin_lock_init(&queue->tx_lock);
1616 spin_lock_init(&queue->rx_lock);
1617
1618 timer_setup(&queue->rx_refill_timer, rx_refill_timeout, 0);
1619
1620 devid = strrchr(queue->info->xbdev->nodename, '/') + 1;
1621 snprintf(queue->name, sizeof(queue->name), "vif%s-q%u",
1622 devid, queue->id);
1623
1624 /* Initialise tx_skbs as a free chain containing every entry. */
1625 queue->tx_skb_freelist = 0;
1626 for (i = 0; i < NET_TX_RING_SIZE; i++) {
1627 skb_entry_set_link(&queue->tx_skbs[i], i+1);
1628 queue->grant_tx_ref[i] = GRANT_INVALID_REF;
1629 queue->grant_tx_page[i] = NULL;
1630 }
1631
1632 /* Clear out rx_skbs */
1633 for (i = 0; i < NET_RX_RING_SIZE; i++) {
1634 queue->rx_skbs[i] = NULL;
1635 queue->grant_rx_ref[i] = GRANT_INVALID_REF;
1636 }
1637
1638 /* A grant for every tx ring slot */
1639 if (gnttab_alloc_grant_references(NET_TX_RING_SIZE,
1640 &queue->gref_tx_head) < 0) {
1641 pr_alert("can't alloc tx grant refs\n");
1642 err = -ENOMEM;
1643 goto exit;
1644 }
1645
1646 /* A grant for every rx ring slot */
1647 if (gnttab_alloc_grant_references(NET_RX_RING_SIZE,
1648 &queue->gref_rx_head) < 0) {
1649 pr_alert("can't alloc rx grant refs\n");
1650 err = -ENOMEM;
1651 goto exit_free_tx;
1652 }
1653
1654 return 0;
1655
1656 exit_free_tx:
1657 gnttab_free_grant_references(queue->gref_tx_head);
1658 exit:
1659 return err;
1660 }
1661
1662 static int write_queue_xenstore_keys(struct netfront_queue *queue,
1663 struct xenbus_transaction *xbt, int write_hierarchical)
1664 {
1665 /* Write the queue-specific keys into XenStore in the traditional
1666 * way for a single queue, or in a queue subkeys for multiple
1667 * queues.
1668 */
1669 struct xenbus_device *dev = queue->info->xbdev;
1670 int err;
1671 const char *message;
1672 char *path;
1673 size_t pathsize;
1674
1675 /* Choose the correct place to write the keys */
1676 if (write_hierarchical) {
1677 pathsize = strlen(dev->nodename) + 10;
1678 path = kzalloc(pathsize, GFP_KERNEL);
1679 if (!path) {
1680 err = -ENOMEM;
1681 message = "out of memory while writing ring references";
1682 goto error;
1683 }
1684 snprintf(path, pathsize, "%s/queue-%u",
1685 dev->nodename, queue->id);
1686 } else {
1687 path = (char *)dev->nodename;
1688 }
1689
1690 /* Write ring references */
1691 err = xenbus_printf(*xbt, path, "tx-ring-ref", "%u",
1692 queue->tx_ring_ref);
1693 if (err) {
1694 message = "writing tx-ring-ref";
1695 goto error;
1696 }
1697
1698 err = xenbus_printf(*xbt, path, "rx-ring-ref", "%u",
1699 queue->rx_ring_ref);
1700 if (err) {
1701 message = "writing rx-ring-ref";
1702 goto error;
1703 }
1704
1705 /* Write event channels; taking into account both shared
1706 * and split event channel scenarios.
1707 */
1708 if (queue->tx_evtchn == queue->rx_evtchn) {
1709 /* Shared event channel */
1710 err = xenbus_printf(*xbt, path,
1711 "event-channel", "%u", queue->tx_evtchn);
1712 if (err) {
1713 message = "writing event-channel";
1714 goto error;
1715 }
1716 } else {
1717 /* Split event channels */
1718 err = xenbus_printf(*xbt, path,
1719 "event-channel-tx", "%u", queue->tx_evtchn);
1720 if (err) {
1721 message = "writing event-channel-tx";
1722 goto error;
1723 }
1724
1725 err = xenbus_printf(*xbt, path,
1726 "event-channel-rx", "%u", queue->rx_evtchn);
1727 if (err) {
1728 message = "writing event-channel-rx";
1729 goto error;
1730 }
1731 }
1732
1733 if (write_hierarchical)
1734 kfree(path);
1735 return 0;
1736
1737 error:
1738 if (write_hierarchical)
1739 kfree(path);
1740 xenbus_dev_fatal(dev, err, "%s", message);
1741 return err;
1742 }
1743
1744 static void xennet_destroy_queues(struct netfront_info *info)
1745 {
1746 unsigned int i;
1747
1748 for (i = 0; i < info->netdev->real_num_tx_queues; i++) {
1749 struct netfront_queue *queue = &info->queues[i];
1750
1751 if (netif_running(info->netdev))
1752 napi_disable(&queue->napi);
1753 netif_napi_del(&queue->napi);
1754 }
1755
1756 kfree(info->queues);
1757 info->queues = NULL;
1758 }
1759
1760 static int xennet_create_queues(struct netfront_info *info,
1761 unsigned int *num_queues)
1762 {
1763 unsigned int i;
1764 int ret;
1765
1766 info->queues = kcalloc(*num_queues, sizeof(struct netfront_queue),
1767 GFP_KERNEL);
1768 if (!info->queues)
1769 return -ENOMEM;
1770
1771 for (i = 0; i < *num_queues; i++) {
1772 struct netfront_queue *queue = &info->queues[i];
1773
1774 queue->id = i;
1775 queue->info = info;
1776
1777 ret = xennet_init_queue(queue);
1778 if (ret < 0) {
1779 dev_warn(&info->xbdev->dev,
1780 "only created %d queues\n", i);
1781 *num_queues = i;
1782 break;
1783 }
1784
1785 netif_napi_add(queue->info->netdev, &queue->napi,
1786 xennet_poll, 64);
1787 if (netif_running(info->netdev))
1788 napi_enable(&queue->napi);
1789 }
1790
1791 netif_set_real_num_tx_queues(info->netdev, *num_queues);
1792
1793 if (*num_queues == 0) {
1794 dev_err(&info->xbdev->dev, "no queues\n");
1795 return -EINVAL;
1796 }
1797 return 0;
1798 }
1799
1800 /* Common code used when first setting up, and when resuming. */
1801 static int talk_to_netback(struct xenbus_device *dev,
1802 struct netfront_info *info)
1803 {
1804 const char *message;
1805 struct xenbus_transaction xbt;
1806 int err;
1807 unsigned int feature_split_evtchn;
1808 unsigned int i = 0;
1809 unsigned int max_queues = 0;
1810 struct netfront_queue *queue = NULL;
1811 unsigned int num_queues = 1;
1812
1813 info->netdev->irq = 0;
1814
1815 /* Check if backend supports multiple queues */
1816 max_queues = xenbus_read_unsigned(info->xbdev->otherend,
1817 "multi-queue-max-queues", 1);
1818 num_queues = min(max_queues, xennet_max_queues);
1819
1820 /* Check feature-split-event-channels */
1821 feature_split_evtchn = xenbus_read_unsigned(info->xbdev->otherend,
1822 "feature-split-event-channels", 0);
1823
1824 /* Read mac addr. */
1825 err = xen_net_read_mac(dev, info->netdev->dev_addr);
1826 if (err) {
1827 xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
1828 goto out_unlocked;
1829 }
1830
1831 rtnl_lock();
1832 if (info->queues)
1833 xennet_destroy_queues(info);
1834
1835 err = xennet_create_queues(info, &num_queues);
1836 if (err < 0) {
1837 xenbus_dev_fatal(dev, err, "creating queues");
1838 kfree(info->queues);
1839 info->queues = NULL;
1840 goto out;
1841 }
1842 rtnl_unlock();
1843
1844 /* Create shared ring, alloc event channel -- for each queue */
1845 for (i = 0; i < num_queues; ++i) {
1846 queue = &info->queues[i];
1847 err = setup_netfront(dev, queue, feature_split_evtchn);
1848 if (err)
1849 goto destroy_ring;
1850 }
1851
1852 again:
1853 err = xenbus_transaction_start(&xbt);
1854 if (err) {
1855 xenbus_dev_fatal(dev, err, "starting transaction");
1856 goto destroy_ring;
1857 }
1858
1859 if (xenbus_exists(XBT_NIL,
1860 info->xbdev->otherend, "multi-queue-max-queues")) {
1861 /* Write the number of queues */
1862 err = xenbus_printf(xbt, dev->nodename,
1863 "multi-queue-num-queues", "%u", num_queues);
1864 if (err) {
1865 message = "writing multi-queue-num-queues";
1866 goto abort_transaction_no_dev_fatal;
1867 }
1868 }
1869
1870 if (num_queues == 1) {
1871 err = write_queue_xenstore_keys(&info->queues[0], &xbt, 0); /* flat */
1872 if (err)
1873 goto abort_transaction_no_dev_fatal;
1874 } else {
1875 /* Write the keys for each queue */
1876 for (i = 0; i < num_queues; ++i) {
1877 queue = &info->queues[i];
1878 err = write_queue_xenstore_keys(queue, &xbt, 1); /* hierarchical */
1879 if (err)
1880 goto abort_transaction_no_dev_fatal;
1881 }
1882 }
1883
1884 /* The remaining keys are not queue-specific */
1885 err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u",
1886 1);
1887 if (err) {
1888 message = "writing request-rx-copy";
1889 goto abort_transaction;
1890 }
1891
1892 err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1);
1893 if (err) {
1894 message = "writing feature-rx-notify";
1895 goto abort_transaction;
1896 }
1897
1898 err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1);
1899 if (err) {
1900 message = "writing feature-sg";
1901 goto abort_transaction;
1902 }
1903
1904 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1);
1905 if (err) {
1906 message = "writing feature-gso-tcpv4";
1907 goto abort_transaction;
1908 }
1909
1910 err = xenbus_write(xbt, dev->nodename, "feature-gso-tcpv6", "1");
1911 if (err) {
1912 message = "writing feature-gso-tcpv6";
1913 goto abort_transaction;
1914 }
1915
1916 err = xenbus_write(xbt, dev->nodename, "feature-ipv6-csum-offload",
1917 "1");
1918 if (err) {
1919 message = "writing feature-ipv6-csum-offload";
1920 goto abort_transaction;
1921 }
1922
1923 err = xenbus_transaction_end(xbt, 0);
1924 if (err) {
1925 if (err == -EAGAIN)
1926 goto again;
1927 xenbus_dev_fatal(dev, err, "completing transaction");
1928 goto destroy_ring;
1929 }
1930
1931 return 0;
1932
1933 abort_transaction:
1934 xenbus_dev_fatal(dev, err, "%s", message);
1935 abort_transaction_no_dev_fatal:
1936 xenbus_transaction_end(xbt, 1);
1937 destroy_ring:
1938 xennet_disconnect_backend(info);
1939 rtnl_lock();
1940 xennet_destroy_queues(info);
1941 out:
1942 rtnl_unlock();
1943 out_unlocked:
1944 device_unregister(&dev->dev);
1945 return err;
1946 }
1947
1948 static int xennet_connect(struct net_device *dev)
1949 {
1950 struct netfront_info *np = netdev_priv(dev);
1951 unsigned int num_queues = 0;
1952 int err;
1953 unsigned int j = 0;
1954 struct netfront_queue *queue = NULL;
1955
1956 if (!xenbus_read_unsigned(np->xbdev->otherend, "feature-rx-copy", 0)) {
1957 dev_info(&dev->dev,
1958 "backend does not support copying receive path\n");
1959 return -ENODEV;
1960 }
1961
1962 err = talk_to_netback(np->xbdev, np);
1963 if (err)
1964 return err;
1965
1966 /* talk_to_netback() sets the correct number of queues */
1967 num_queues = dev->real_num_tx_queues;
1968
1969 if (dev->reg_state == NETREG_UNINITIALIZED) {
1970 err = register_netdev(dev);
1971 if (err) {
1972 pr_warn("%s: register_netdev err=%d\n", __func__, err);
1973 device_unregister(&np->xbdev->dev);
1974 return err;
1975 }
1976 }
1977
1978 rtnl_lock();
1979 netdev_update_features(dev);
1980 rtnl_unlock();
1981
1982 /*
1983 * All public and private state should now be sane. Get
1984 * ready to start sending and receiving packets and give the driver
1985 * domain a kick because we've probably just requeued some
1986 * packets.
1987 */
1988 netif_carrier_on(np->netdev);
1989 for (j = 0; j < num_queues; ++j) {
1990 queue = &np->queues[j];
1991
1992 notify_remote_via_irq(queue->tx_irq);
1993 if (queue->tx_irq != queue->rx_irq)
1994 notify_remote_via_irq(queue->rx_irq);
1995
1996 spin_lock_irq(&queue->tx_lock);
1997 xennet_tx_buf_gc(queue);
1998 spin_unlock_irq(&queue->tx_lock);
1999
2000 spin_lock_bh(&queue->rx_lock);
2001 xennet_alloc_rx_buffers(queue);
2002 spin_unlock_bh(&queue->rx_lock);
2003 }
2004
2005 return 0;
2006 }
2007
2008 /**
2009 * Callback received when the backend's state changes.
2010 */
2011 static void netback_changed(struct xenbus_device *dev,
2012 enum xenbus_state backend_state)
2013 {
2014 struct netfront_info *np = dev_get_drvdata(&dev->dev);
2015 struct net_device *netdev = np->netdev;
2016
2017 dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state));
2018
2019 wake_up_all(&module_wq);
2020
2021 switch (backend_state) {
2022 case XenbusStateInitialising:
2023 case XenbusStateInitialised:
2024 case XenbusStateReconfiguring:
2025 case XenbusStateReconfigured:
2026 case XenbusStateUnknown:
2027 break;
2028
2029 case XenbusStateInitWait:
2030 if (dev->state != XenbusStateInitialising)
2031 break;
2032 if (xennet_connect(netdev) != 0)
2033 break;
2034 xenbus_switch_state(dev, XenbusStateConnected);
2035 break;
2036
2037 case XenbusStateConnected:
2038 netdev_notify_peers(netdev);
2039 break;
2040
2041 case XenbusStateClosed:
2042 if (dev->state == XenbusStateClosed)
2043 break;
2044 /* Missed the backend's CLOSING state -- fallthrough */
2045 case XenbusStateClosing:
2046 xenbus_frontend_closed(dev);
2047 break;
2048 }
2049 }
2050
2051 static const struct xennet_stat {
2052 char name[ETH_GSTRING_LEN];
2053 u16 offset;
2054 } xennet_stats[] = {
2055 {
2056 "rx_gso_checksum_fixup",
2057 offsetof(struct netfront_info, rx_gso_checksum_fixup)
2058 },
2059 };
2060
2061 static int xennet_get_sset_count(struct net_device *dev, int string_set)
2062 {
2063 switch (string_set) {
2064 case ETH_SS_STATS:
2065 return ARRAY_SIZE(xennet_stats);
2066 default:
2067 return -EINVAL;
2068 }
2069 }
2070
2071 static void xennet_get_ethtool_stats(struct net_device *dev,
2072 struct ethtool_stats *stats, u64 * data)
2073 {
2074 void *np = netdev_priv(dev);
2075 int i;
2076
2077 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2078 data[i] = atomic_read((atomic_t *)(np + xennet_stats[i].offset));
2079 }
2080
2081 static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data)
2082 {
2083 int i;
2084
2085 switch (stringset) {
2086 case ETH_SS_STATS:
2087 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2088 memcpy(data + i * ETH_GSTRING_LEN,
2089 xennet_stats[i].name, ETH_GSTRING_LEN);
2090 break;
2091 }
2092 }
2093
2094 static const struct ethtool_ops xennet_ethtool_ops =
2095 {
2096 .get_link = ethtool_op_get_link,
2097
2098 .get_sset_count = xennet_get_sset_count,
2099 .get_ethtool_stats = xennet_get_ethtool_stats,
2100 .get_strings = xennet_get_strings,
2101 };
2102
2103 #ifdef CONFIG_SYSFS
2104 static ssize_t show_rxbuf(struct device *dev,
2105 struct device_attribute *attr, char *buf)
2106 {
2107 return sprintf(buf, "%lu\n", NET_RX_RING_SIZE);
2108 }
2109
2110 static ssize_t store_rxbuf(struct device *dev,
2111 struct device_attribute *attr,
2112 const char *buf, size_t len)
2113 {
2114 char *endp;
2115 unsigned long target;
2116
2117 if (!capable(CAP_NET_ADMIN))
2118 return -EPERM;
2119
2120 target = simple_strtoul(buf, &endp, 0);
2121 if (endp == buf)
2122 return -EBADMSG;
2123
2124 /* rxbuf_min and rxbuf_max are no longer configurable. */
2125
2126 return len;
2127 }
2128
2129 static DEVICE_ATTR(rxbuf_min, 0644, show_rxbuf, store_rxbuf);
2130 static DEVICE_ATTR(rxbuf_max, 0644, show_rxbuf, store_rxbuf);
2131 static DEVICE_ATTR(rxbuf_cur, 0444, show_rxbuf, NULL);
2132
2133 static struct attribute *xennet_dev_attrs[] = {
2134 &dev_attr_rxbuf_min.attr,
2135 &dev_attr_rxbuf_max.attr,
2136 &dev_attr_rxbuf_cur.attr,
2137 NULL
2138 };
2139
2140 static const struct attribute_group xennet_dev_group = {
2141 .attrs = xennet_dev_attrs
2142 };
2143 #endif /* CONFIG_SYSFS */
2144
2145 static int xennet_remove(struct xenbus_device *dev)
2146 {
2147 struct netfront_info *info = dev_get_drvdata(&dev->dev);
2148
2149 dev_dbg(&dev->dev, "%s\n", dev->nodename);
2150
2151 if (xenbus_read_driver_state(dev->otherend) != XenbusStateClosed) {
2152 xenbus_switch_state(dev, XenbusStateClosing);
2153 wait_event(module_wq,
2154 xenbus_read_driver_state(dev->otherend) ==
2155 XenbusStateClosing ||
2156 xenbus_read_driver_state(dev->otherend) ==
2157 XenbusStateUnknown);
2158
2159 xenbus_switch_state(dev, XenbusStateClosed);
2160 wait_event(module_wq,
2161 xenbus_read_driver_state(dev->otherend) ==
2162 XenbusStateClosed ||
2163 xenbus_read_driver_state(dev->otherend) ==
2164 XenbusStateUnknown);
2165 }
2166
2167 xennet_disconnect_backend(info);
2168
2169 if (info->netdev->reg_state == NETREG_REGISTERED)
2170 unregister_netdev(info->netdev);
2171
2172 if (info->queues) {
2173 rtnl_lock();
2174 xennet_destroy_queues(info);
2175 rtnl_unlock();
2176 }
2177 xennet_free_netdev(info->netdev);
2178
2179 return 0;
2180 }
2181
2182 static const struct xenbus_device_id netfront_ids[] = {
2183 { "vif" },
2184 { "" }
2185 };
2186
2187 static struct xenbus_driver netfront_driver = {
2188 .ids = netfront_ids,
2189 .probe = netfront_probe,
2190 .remove = xennet_remove,
2191 .resume = netfront_resume,
2192 .otherend_changed = netback_changed,
2193 };
2194
2195 static int __init netif_init(void)
2196 {
2197 if (!xen_domain())
2198 return -ENODEV;
2199
2200 if (!xen_has_pv_nic_devices())
2201 return -ENODEV;
2202
2203 pr_info("Initialising Xen virtual ethernet driver\n");
2204
2205 /* Allow as many queues as there are CPUs inut max. 8 if user has not
2206 * specified a value.
2207 */
2208 if (xennet_max_queues == 0)
2209 xennet_max_queues = min_t(unsigned int, MAX_QUEUES_DEFAULT,
2210 num_online_cpus());
2211
2212 return xenbus_register_frontend(&netfront_driver);
2213 }
2214 module_init(netif_init);
2215
2216
2217 static void __exit netif_exit(void)
2218 {
2219 xenbus_unregister_driver(&netfront_driver);
2220 }
2221 module_exit(netif_exit);
2222
2223 MODULE_DESCRIPTION("Xen virtual network device frontend");
2224 MODULE_LICENSE("GPL");
2225 MODULE_ALIAS("xen:vif");
2226 MODULE_ALIAS("xennet");
Linux with added FPC-III support