projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge branch 'net-ipa-simplify-endpoint-programming'
[linux-block.git] / drivers / net / xen-netfront.c
CommitLineData
0d160211
JF		 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
383eda32
JP		 32#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
33
0d160211
JF		 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>
9ecd1a75 41#include <net/tcp.h>
0d160211
JF		 42#include <linux/udp.h>
43#include <linux/moduleparam.h>
44#include <linux/mm.h>
5a0e3ad6 45#include <linux/slab.h>
0d160211 46#include <net/ip.h>
6c5aa6fc
DK		 47#include <linux/bpf.h>
48#include <net/page_pool.h>
49#include <linux/bpf_trace.h>
0d160211 50
1ccbf534 51#include <xen/xen.h>
0d160211
JF		 52#include <xen/xenbus.h>
53#include <xen/events.h>
54#include <xen/page.h>
b9136d20 55#include <xen/platform_pci.h>
0d160211
JF		 56#include <xen/grant_table.h>
57
58#include <xen/interface/io/netif.h>
59#include <xen/interface/memory.h>
60#include <xen/interface/grant_table.h>
61
50ee6061 62/* Module parameters */
034702a6 63#define MAX_QUEUES_DEFAULT 8
50ee6061
AB		 64static unsigned int xennet_max_queues;
65module_param_named(max_queues, xennet_max_queues, uint, 0644);
66MODULE_PARM_DESC(max_queues,
67 "Maximum number of queues per virtual interface");
68
0fc0b732 69static const struct ethtool_ops xennet_ethtool_ops;
0d160211
JF		 70
71struct netfront_cb {
3683243b 72 int pull_to;
0d160211
JF		 73};
74
75#define NETFRONT_SKB_CB(skb) ((struct netfront_cb *)((skb)->cb))
76
77#define RX_COPY_THRESHOLD 256
78
79#define GRANT_INVALID_REF 0
80
30c5d7f0
JG		 81#define NET_TX_RING_SIZE __CONST_RING_SIZE(xen_netif_tx, XEN_PAGE_SIZE)
82#define NET_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, XEN_PAGE_SIZE)
1f3c2eba
DV		 83
84/* Minimum number of Rx slots (includes slot for GSO metadata). */
85#define NET_RX_SLOTS_MIN (XEN_NETIF_NR_SLOTS_MIN + 1)
0d160211 86
2688fcb7
AB		 87/* Queue name is interface name with "-qNNN" appended */
88#define QUEUE_NAME_SIZE (IFNAMSIZ + 6)
89
90/* IRQ name is queue name with "-tx" or "-rx" appended */
91#define IRQ_NAME_SIZE (QUEUE_NAME_SIZE + 3)
92
8edfe2e9 93static DECLARE_WAIT_QUEUE_HEAD(module_wq);
5b5971df 94
e00f85be 95struct netfront_stats {
900e1833
DV		 96 u64 packets;
97 u64 bytes;
e00f85be 98 struct u64_stats_sync syncp;
99};
100
2688fcb7
AB		 101struct netfront_info;
102
103struct netfront_queue {
104 unsigned int id; /* Queue ID, 0-based */
105 char name[QUEUE_NAME_SIZE]; /* DEVNAME-qN */
106 struct netfront_info *info;
0d160211 107
6c5aa6fc
DK		 108 struct bpf_prog __rcu *xdp_prog;
109
bea3348e 110 struct napi_struct napi;
0d160211 111
d634bf2c
WL		 112 /* Split event channels support, tx_* == rx_* when using
113 * single event channel.
114 */
115 unsigned int tx_evtchn, rx_evtchn;
116 unsigned int tx_irq, rx_irq;
117 /* Only used when split event channels support is enabled */
2688fcb7
AB		 118 char tx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-tx */
119 char rx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-rx */
0d160211 120
84284d3c
JF		 121 spinlock_t tx_lock;
122 struct xen_netif_tx_front_ring tx;
123 int tx_ring_ref;
0d160211
JF		 124
125 /*
126 * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries
127 * are linked from tx_skb_freelist through skb_entry.link.
128 *
129 * NB. Freelist index entries are always going to be less than
130 * PAGE_OFFSET, whereas pointers to skbs will always be equal or
131 * greater than PAGE_OFFSET: we use this property to distinguish
132 * them.
133 */
134 union skb_entry {
135 struct sk_buff *skb;
1ffb40b8 136 unsigned long link;
0d160211
JF		 137 } tx_skbs[NET_TX_RING_SIZE];
138 grant_ref_t gref_tx_head;
139 grant_ref_t grant_tx_ref[NET_TX_RING_SIZE];
cefe0078 140 struct page *grant_tx_page[NET_TX_RING_SIZE];
0d160211
JF		 141 unsigned tx_skb_freelist;
142
84284d3c
JF		 143 spinlock_t rx_lock ____cacheline_aligned_in_smp;
144 struct xen_netif_rx_front_ring rx;
145 int rx_ring_ref;
146
84284d3c
JF		 147 struct timer_list rx_refill_timer;
148
0d160211
JF		 149 struct sk_buff *rx_skbs[NET_RX_RING_SIZE];
150 grant_ref_t gref_rx_head;
151 grant_ref_t grant_rx_ref[NET_RX_RING_SIZE];
6c5aa6fc
DK		 152
153 struct page_pool *page_pool;
154 struct xdp_rxq_info xdp_rxq;
2688fcb7
AB		 155};
156
157struct netfront_info {
158 struct list_head list;
159 struct net_device *netdev;
160
161 struct xenbus_device *xbdev;
162
163 /* Multi-queue support */
164 struct netfront_queue *queues;
e0ce4af9
IC		 165
166 /* Statistics */
900e1833
DV		 167 struct netfront_stats __percpu *rx_stats;
168 struct netfront_stats __percpu *tx_stats;
e00f85be 169
6c5aa6fc
DK		 170 /* XDP state */
171 bool netback_has_xdp_headroom;
172 bool netfront_xdp_enabled;
173
2688fcb7 174 atomic_t rx_gso_checksum_fixup;
0d160211
JF		 175};
176
177struct netfront_rx_info {
178 struct xen_netif_rx_response rx;
179 struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1];
180};
181
1ffb40b8
IY		 182static void skb_entry_set_link(union skb_entry *list, unsigned short id)
183{
184 list->link = id;
185}
186
187static int skb_entry_is_link(const union skb_entry *list)
188{
189 BUILD_BUG_ON(sizeof(list->skb) != sizeof(list->link));
807540ba 190 return (unsigned long)list->skb < PAGE_OFFSET;
1ffb40b8
IY		 191}
192
0d160211
JF		 193/*
194 * Access macros for acquiring freeing slots in tx_skbs[].
195 */
196
197static void add_id_to_freelist(unsigned *head, union skb_entry *list,
198 unsigned short id)
199{
1ffb40b8 200 skb_entry_set_link(&list[id], *head);
0d160211
JF		 201 *head = id;
202}
203
204static unsigned short get_id_from_freelist(unsigned *head,
205 union skb_entry *list)
206{
207 unsigned int id = *head;
208 *head = list[id].link;
209 return id;
210}
211
212static int xennet_rxidx(RING_IDX idx)
213{
214 return idx & (NET_RX_RING_SIZE - 1);
215}
216
2688fcb7 217static struct sk_buff *xennet_get_rx_skb(struct netfront_queue *queue,
0d160211
JF		 218 RING_IDX ri)
219{
220 int i = xennet_rxidx(ri);
2688fcb7
AB		 221 struct sk_buff *skb = queue->rx_skbs[i];
222 queue->rx_skbs[i] = NULL;
0d160211
JF		 223 return skb;
224}
225
2688fcb7 226static grant_ref_t xennet_get_rx_ref(struct netfront_queue *queue,
0d160211
JF		 227 RING_IDX ri)
228{
229 int i = xennet_rxidx(ri);
2688fcb7
AB		 230 grant_ref_t ref = queue->grant_rx_ref[i];
231 queue->grant_rx_ref[i] = GRANT_INVALID_REF;
0d160211
JF		 232 return ref;
233}
234
235#ifdef CONFIG_SYSFS
27b917e5 236static const struct attribute_group xennet_dev_group;
0d160211
JF		 237#endif
238
3ad9b358 239static bool xennet_can_sg(struct net_device *dev)
0d160211 240{
3ad9b358 241 return dev->features & NETIF_F_SG;
0d160211
JF		 242}
243
244
e99e88a9 245static void rx_refill_timeout(struct timer_list *t)
0d160211 246{
e99e88a9 247 struct netfront_queue *queue = from_timer(queue, t, rx_refill_timer);
2688fcb7 248 napi_schedule(&queue->napi);
0d160211
JF		 249}
250
2688fcb7 251static int netfront_tx_slot_available(struct netfront_queue *queue)
0d160211 252{
2688fcb7 253 return (queue->tx.req_prod_pvt - queue->tx.rsp_cons) <
57f230ab 254 (NET_TX_RING_SIZE - XEN_NETIF_NR_SLOTS_MIN - 1);
0d160211
JF		 255}
256
2688fcb7 257static void xennet_maybe_wake_tx(struct netfront_queue *queue)
0d160211 258{
2688fcb7
AB		 259 struct net_device *dev = queue->info->netdev;
260 struct netdev_queue *dev_queue = netdev_get_tx_queue(dev, queue->id);
0d160211 261
2688fcb7
AB		 262 if (unlikely(netif_tx_queue_stopped(dev_queue)) &&
263 netfront_tx_slot_available(queue) &&
0d160211 264 likely(netif_running(dev)))
2688fcb7 265 netif_tx_wake_queue(netdev_get_tx_queue(dev, queue->id));
0d160211
JF		 266}
267
1f3c2eba
DV		 268
269static struct sk_buff *xennet_alloc_one_rx_buffer(struct netfront_queue *queue)
0d160211 270{
0d160211
JF		 271 struct sk_buff *skb;
272 struct page *page;
0d160211 273
1f3c2eba
DV		 274 skb = __netdev_alloc_skb(queue->info->netdev,
275 RX_COPY_THRESHOLD + NET_IP_ALIGN,
276 GFP_ATOMIC | __GFP_NOWARN);
277 if (unlikely(!skb))
278 return NULL;
0d160211 279
6c5aa6fc
DK		 280 page = page_pool_dev_alloc_pages(queue->page_pool);
281 if (unlikely(!page)) {
1f3c2eba
DV		 282 kfree_skb(skb);
283 return NULL;
0d160211 284 }
1f3c2eba
DV		 285 skb_add_rx_frag(skb, 0, page, 0, 0, PAGE_SIZE);
286
287 /* Align ip header to a 16 bytes boundary */
288 skb_reserve(skb, NET_IP_ALIGN);
289 skb->dev = queue->info->netdev;
290
291 return skb;
292}
0d160211 293
1f3c2eba
DV		 294
295static void xennet_alloc_rx_buffers(struct netfront_queue *queue)
296{
297 RING_IDX req_prod = queue->rx.req_prod_pvt;
298 int notify;
538d9291 299 int err = 0;
1f3c2eba
DV		 300
301 if (unlikely(!netif_carrier_ok(queue->info->netdev)))
0d160211 302 return;
0d160211 303
1f3c2eba
DV		 304 for (req_prod = queue->rx.req_prod_pvt;
305 req_prod - queue->rx.rsp_cons < NET_RX_RING_SIZE;
306 req_prod++) {
307 struct sk_buff *skb;
308 unsigned short id;
309 grant_ref_t ref;
30c5d7f0 310 struct page *page;
1f3c2eba 311 struct xen_netif_rx_request *req;
0d160211 312
1f3c2eba 313 skb = xennet_alloc_one_rx_buffer(queue);
538d9291
VRP		 314 if (!skb) {
315 err = -ENOMEM;
0d160211 316 break;
538d9291 317 }
0d160211 318
1f3c2eba 319 id = xennet_rxidx(req_prod);
0d160211 320
2688fcb7
AB		 321 BUG_ON(queue->rx_skbs[id]);
322 queue->rx_skbs[id] = skb;
0d160211 323
2688fcb7 324 ref = gnttab_claim_grant_reference(&queue->gref_rx_head);
269ebce4 325 WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref));
2688fcb7 326 queue->grant_rx_ref[id] = ref;
0d160211 327
30c5d7f0 328 page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
0d160211 329
1f3c2eba 330 req = RING_GET_REQUEST(&queue->rx, req_prod);
30c5d7f0
JG		 331 gnttab_page_grant_foreign_access_ref_one(ref,
332 queue->info->xbdev->otherend_id,
333 page,
334 0);
0d160211
JF		 335 req->id = id;
336 req->gref = ref;
337 }
338
1f3c2eba
DV		 339 queue->rx.req_prod_pvt = req_prod;
340
538d9291
VRP		 341 /* Try again later if there are not enough requests or skb allocation
342 * failed.
343 * Enough requests is quantified as the sum of newly created slots and
344 * the unconsumed slots at the backend.
345 */
346 if (req_prod - queue->rx.rsp_cons < NET_RX_SLOTS_MIN ||
347 unlikely(err)) {
1f3c2eba
DV		 348 mod_timer(&queue->rx_refill_timer, jiffies + (HZ/10));
349 return;
350 }
351
2688fcb7 352 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->rx, notify);
0d160211 353 if (notify)
2688fcb7 354 notify_remote_via_irq(queue->rx_irq);
0d160211
JF		 355}
356
357static int xennet_open(struct net_device *dev)
358{
359 struct netfront_info *np = netdev_priv(dev);
2688fcb7
AB		 360 unsigned int num_queues = dev->real_num_tx_queues;
361 unsigned int i = 0;
362 struct netfront_queue *queue = NULL;
363
f599c64f
RL		 364 if (!np->queues)
365 return -ENODEV;
366
2688fcb7
AB		 367 for (i = 0; i < num_queues; ++i) {
368 queue = &np->queues[i];
369 napi_enable(&queue->napi);
370
371 spin_lock_bh(&queue->rx_lock);
372 if (netif_carrier_ok(dev)) {
373 xennet_alloc_rx_buffers(queue);
374 queue->rx.sring->rsp_event = queue->rx.rsp_cons + 1;
375 if (RING_HAS_UNCONSUMED_RESPONSES(&queue->rx))
376 napi_schedule(&queue->napi);
377 }
378 spin_unlock_bh(&queue->rx_lock);
0d160211 379 }
0d160211 380
2688fcb7 381 netif_tx_start_all_queues(dev);
0d160211
JF		 382
383 return 0;
384}
385
2688fcb7 386static void xennet_tx_buf_gc(struct netfront_queue *queue)
0d160211
JF		 387{
388 RING_IDX cons, prod;
389 unsigned short id;
0d160211 390 struct sk_buff *skb;
7d0105b5 391 bool more_to_do;
0d160211 392
2688fcb7 393 BUG_ON(!netif_carrier_ok(queue->info->netdev));
0d160211
JF		 394
395 do {
2688fcb7 396 prod = queue->tx.sring->rsp_prod;
0d160211
JF		 397 rmb(); /* Ensure we see responses up to 'rp'. */
398
2688fcb7 399 for (cons = queue->tx.rsp_cons; cons != prod; cons++) {
0d160211
JF		 400 struct xen_netif_tx_response *txrsp;
401
2688fcb7 402 txrsp = RING_GET_RESPONSE(&queue->tx, cons);
f942dc25 403 if (txrsp->status == XEN_NETIF_RSP_NULL)
0d160211
JF		 404 continue;
405
406 id = txrsp->id;
2688fcb7 407 skb = queue->tx_skbs[id].skb;
0d160211 408 if (unlikely(gnttab_query_foreign_access(
2688fcb7 409 queue->grant_tx_ref[id]) != 0)) {
383eda32
JP		 410 pr_alert("%s: warning -- grant still in use by backend domain\n",
411 __func__);
0d160211
JF		 412 BUG();
413 }
414 gnttab_end_foreign_access_ref(
2688fcb7 415 queue->grant_tx_ref[id], GNTMAP_readonly);
0d160211 416 gnttab_release_grant_reference(
2688fcb7
AB		 417 &queue->gref_tx_head, queue->grant_tx_ref[id]);
418 queue->grant_tx_ref[id] = GRANT_INVALID_REF;
419 queue->grant_tx_page[id] = NULL;
420 add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, id);
0d160211
JF		 421 dev_kfree_skb_irq(skb);
422 }
423
2688fcb7 424 queue->tx.rsp_cons = prod;
0d160211 425
7d0105b5
MC		 426 RING_FINAL_CHECK_FOR_RESPONSES(&queue->tx, more_to_do);
427 } while (more_to_do);
0d160211 428
2688fcb7 429 xennet_maybe_wake_tx(queue);
0d160211
JF		 430}
431
30c5d7f0
JG		 432struct xennet_gnttab_make_txreq {
433 struct netfront_queue *queue;
434 struct sk_buff *skb;
435 struct page *page;
436 struct xen_netif_tx_request *tx; /* Last request */
437 unsigned int size;
438};
439
440static void xennet_tx_setup_grant(unsigned long gfn, unsigned int offset,
441 unsigned int len, void *data)
0d160211 442{
30c5d7f0 443 struct xennet_gnttab_make_txreq *info = data;
0d160211 444 unsigned int id;
a55e8bb8 445 struct xen_netif_tx_request *tx;
0d160211 446 grant_ref_t ref;
30c5d7f0
JG		 447 /* convenient aliases */
448 struct page *page = info->page;
449 struct netfront_queue *queue = info->queue;
450 struct sk_buff *skb = info->skb;
0d160211 451
a55e8bb8
DV		 452 id = get_id_from_freelist(&queue->tx_skb_freelist, queue->tx_skbs);
453 tx = RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++);
454 ref = gnttab_claim_grant_reference(&queue->gref_tx_head);
269ebce4 455 WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref));
0d160211 456
30c5d7f0
JG		 457 gnttab_grant_foreign_access_ref(ref, queue->info->xbdev->otherend_id,
458 gfn, GNTMAP_readonly);
0d160211 459
a55e8bb8
DV		 460 queue->tx_skbs[id].skb = skb;
461 queue->grant_tx_page[id] = page;
462 queue->grant_tx_ref[id] = ref;
0d160211 463
a55e8bb8
DV		 464 tx->id = id;
465 tx->gref = ref;
466 tx->offset = offset;
467 tx->size = len;
468 tx->flags = 0;
0d160211 469
30c5d7f0
JG		 470 info->tx = tx;
471 info->size += tx->size;
472}
473
474static struct xen_netif_tx_request *xennet_make_first_txreq(
475 struct netfront_queue *queue, struct sk_buff *skb,
476 struct page *page, unsigned int offset, unsigned int len)
477{
478 struct xennet_gnttab_make_txreq info = {
479 .queue = queue,
480 .skb = skb,
481 .page = page,
482 .size = 0,
483 };
484
485 gnttab_for_one_grant(page, offset, len, xennet_tx_setup_grant, &info);
486
487 return info.tx;
488}
489
490static void xennet_make_one_txreq(unsigned long gfn, unsigned int offset,
491 unsigned int len, void *data)
492{
493 struct xennet_gnttab_make_txreq *info = data;
494
495 info->tx->flags |= XEN_NETTXF_more_data;
496 skb_get(info->skb);
497 xennet_tx_setup_grant(gfn, offset, len, data);
a55e8bb8 498}
0d160211 499
a55e8bb8
DV		 500static struct xen_netif_tx_request *xennet_make_txreqs(
501 struct netfront_queue *queue, struct xen_netif_tx_request *tx,
502 struct sk_buff *skb, struct page *page,
503 unsigned int offset, unsigned int len)
504{
30c5d7f0
JG		 505 struct xennet_gnttab_make_txreq info = {
506 .queue = queue,
507 .skb = skb,
508 .tx = tx,
509 };
510
a55e8bb8
DV		 511 /* Skip unused frames from start of page */
512 page += offset >> PAGE_SHIFT;
513 offset &= ~PAGE_MASK;
0d160211 514
a55e8bb8 515 while (len) {
30c5d7f0
JG		 516 info.page = page;
517 info.size = 0;
518
519 gnttab_foreach_grant_in_range(page, offset, len,
520 xennet_make_one_txreq,
521 &info);
522
a55e8bb8
DV		 523 page++;
524 offset = 0;
30c5d7f0 525 len -= info.size;
0d160211
JF		 526 }
527
30c5d7f0 528 return info.tx;
0d160211
JF		 529}
530
f36c3747 531/*
e84448d5
DV		 532 * Count how many ring slots are required to send this skb. Each frag
533 * might be a compound page.
f36c3747 534 */
e84448d5 535static int xennet_count_skb_slots(struct sk_buff *skb)
f36c3747
IC		 536{
537 int i, frags = skb_shinfo(skb)->nr_frags;
30c5d7f0 538 int slots;
e84448d5 539
30c5d7f0
JG		 540 slots = gnttab_count_grant(offset_in_page(skb->data),
541 skb_headlen(skb));
f36c3747
IC		 542
543 for (i = 0; i < frags; i++) {
544 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
545 unsigned long size = skb_frag_size(frag);
b54c9d5b 546 unsigned long offset = skb_frag_off(frag);
f36c3747
IC		 547
548 /* Skip unused frames from start of page */
549 offset &= ~PAGE_MASK;
550
30c5d7f0 551 slots += gnttab_count_grant(offset, size);
f36c3747
IC		 552 }
553
30c5d7f0 554 return slots;
f36c3747
IC		 555}
556
50ee6061 557static u16 xennet_select_queue(struct net_device *dev, struct sk_buff *skb,
a350ecce 558 struct net_device *sb_dev)
2688fcb7 559{
50ee6061
AB		 560 unsigned int num_queues = dev->real_num_tx_queues;
561 u32 hash;
562 u16 queue_idx;
563
564 /* First, check if there is only one queue */
565 if (num_queues == 1) {
566 queue_idx = 0;
567 } else {
568 hash = skb_get_hash(skb);
569 queue_idx = hash % num_queues;
570 }
571
572 return queue_idx;
2688fcb7
AB		 573}
574
6c5aa6fc
DK		 575static int xennet_xdp_xmit_one(struct net_device *dev,
576 struct netfront_queue *queue,
577 struct xdp_frame *xdpf)
578{
579 struct netfront_info *np = netdev_priv(dev);
580 struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats);
581 int notify;
582
583 xennet_make_first_txreq(queue, NULL,
584 virt_to_page(xdpf->data),
585 offset_in_page(xdpf->data),
586 xdpf->len);
587
588 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify);
589 if (notify)
590 notify_remote_via_irq(queue->tx_irq);
591
592 u64_stats_update_begin(&tx_stats->syncp);
593 tx_stats->bytes += xdpf->len;
594 tx_stats->packets++;
595 u64_stats_update_end(&tx_stats->syncp);
596
597 xennet_tx_buf_gc(queue);
598
599 return 0;
600}
601
602static int xennet_xdp_xmit(struct net_device *dev, int n,
603 struct xdp_frame **frames, u32 flags)
604{
605 unsigned int num_queues = dev->real_num_tx_queues;
606 struct netfront_info *np = netdev_priv(dev);
607 struct netfront_queue *queue = NULL;
608 unsigned long irq_flags;
609 int drops = 0;
610 int i, err;
611
612 if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
613 return -EINVAL;
614
615 queue = &np->queues[smp_processor_id() % num_queues];
616
617 spin_lock_irqsave(&queue->tx_lock, irq_flags);
618 for (i = 0; i < n; i++) {
619 struct xdp_frame *xdpf = frames[i];
620
621 if (!xdpf)
622 continue;
623 err = xennet_xdp_xmit_one(dev, queue, xdpf);
624 if (err) {
625 xdp_return_frame_rx_napi(xdpf);
626 drops++;
627 }
628 }
629 spin_unlock_irqrestore(&queue->tx_lock, irq_flags);
630
631 return n - drops;
632}
633
634
30c5d7f0
JG		 635#define MAX_XEN_SKB_FRAGS (65536 / XEN_PAGE_SIZE + 1)
636
24a94b3c 637static netdev_tx_t xennet_start_xmit(struct sk_buff *skb, struct net_device *dev)
0d160211 638{
0d160211 639 struct netfront_info *np = netdev_priv(dev);
900e1833 640 struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats);
a55e8bb8
DV		 641 struct xen_netif_tx_request *tx, *first_tx;
642 unsigned int i;
0d160211 643 int notify;
f36c3747 644 int slots;
a55e8bb8
DV		 645 struct page *page;
646 unsigned int offset;
647 unsigned int len;
cf66f9d4 648 unsigned long flags;
2688fcb7
AB		 649 struct netfront_queue *queue = NULL;
650 unsigned int num_queues = dev->real_num_tx_queues;
651 u16 queue_index;
fd07160b 652 struct sk_buff *nskb;
2688fcb7
AB		 653
654 /* Drop the packet if no queues are set up */
655 if (num_queues < 1)
656 goto drop;
657 /* Determine which queue to transmit this SKB on */
658 queue_index = skb_get_queue_mapping(skb);
659 queue = &np->queues[queue_index];
0d160211 660
9ecd1a75
WL		 661 /* If skb->len is too big for wire format, drop skb and alert
662 * user about misconfiguration.
663 */
664 if (unlikely(skb->len > XEN_NETIF_MAX_TX_SIZE)) {
665 net_alert_ratelimited(
666 "xennet: skb->len = %u, too big for wire format\n",
667 skb->len);
668 goto drop;
669 }
670
e84448d5 671 slots = xennet_count_skb_slots(skb);
30c5d7f0 672 if (unlikely(slots > MAX_XEN_SKB_FRAGS + 1)) {
97a6d1bb
ZK		 673 net_dbg_ratelimited("xennet: skb rides the rocket: %d slots, %d bytes\n",
674 slots, skb->len);
675 if (skb_linearize(skb))
676 goto drop;
0d160211
JF		 677 }
678
a55e8bb8
DV		 679 page = virt_to_page(skb->data);
680 offset = offset_in_page(skb->data);
fd07160b
VK		 681
682 /* The first req should be at least ETH_HLEN size or the packet will be
683 * dropped by netback.
684 */
685 if (unlikely(PAGE_SIZE - offset < ETH_HLEN)) {
686 nskb = skb_copy(skb, GFP_ATOMIC);
687 if (!nskb)
688 goto drop;
62f3250f 689 dev_consume_skb_any(skb);
fd07160b
VK		 690 skb = nskb;
691 page = virt_to_page(skb->data);
692 offset = offset_in_page(skb->data);
693 }
694
a55e8bb8
DV		 695 len = skb_headlen(skb);
696
2688fcb7 697 spin_lock_irqsave(&queue->tx_lock, flags);
0d160211
JF		 698
699 if (unlikely(!netif_carrier_ok(dev) ||
f36c3747 700 (slots > 1 && !xennet_can_sg(dev)) ||
8b86a61d 701 netif_needs_gso(skb, netif_skb_features(skb)))) {
2688fcb7 702 spin_unlock_irqrestore(&queue->tx_lock, flags);
0d160211
JF		 703 goto drop;
704 }
705
a55e8bb8 706 /* First request for the linear area. */
30c5d7f0
JG		 707 first_tx = tx = xennet_make_first_txreq(queue, skb,
708 page, offset, len);
709 offset += tx->size;
710 if (offset == PAGE_SIZE) {
711 page++;
712 offset = 0;
713 }
a55e8bb8 714 len -= tx->size;
0d160211 715
0d160211
JF		 716 if (skb->ip_summed == CHECKSUM_PARTIAL)
717 /* local packet? */
f942dc25 718 tx->flags |= XEN_NETTXF_csum_blank | XEN_NETTXF_data_validated;
0d160211
JF		 719 else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
720 /* remote but checksummed. */
f942dc25 721 tx->flags |= XEN_NETTXF_data_validated;
0d160211 722
a55e8bb8 723 /* Optional extra info after the first request. */
0d160211
JF		 724 if (skb_shinfo(skb)->gso_size) {
725 struct xen_netif_extra_info *gso;
726
727 gso = (struct xen_netif_extra_info *)
a55e8bb8 728 RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++);
0d160211 729
e2d617c0 730 tx->flags |= XEN_NETTXF_extra_info;
0d160211
JF		 731
732 gso->u.gso.size = skb_shinfo(skb)->gso_size;
2c0057de
PD		 733 gso->u.gso.type = (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) ?
734 XEN_NETIF_GSO_TYPE_TCPV6 :
735 XEN_NETIF_GSO_TYPE_TCPV4;
0d160211
JF		 736 gso->u.gso.pad = 0;
737 gso->u.gso.features = 0;
738
739 gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
740 gso->flags = 0;
0d160211
JF		 741 }
742
a55e8bb8
DV		 743 /* Requests for the rest of the linear area. */
744 tx = xennet_make_txreqs(queue, tx, skb, page, offset, len);
745
746 /* Requests for all the frags. */
747 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
748 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
b54c9d5b
JL		 749 tx = xennet_make_txreqs(queue, tx, skb, skb_frag_page(frag),
750 skb_frag_off(frag),
a55e8bb8
DV		 751 skb_frag_size(frag));
752 }
0d160211 753
a55e8bb8
DV		 754 /* First request has the packet length. */
755 first_tx->size = skb->len;
0d160211 756
2688fcb7 757 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify);
0d160211 758 if (notify)
2688fcb7 759 notify_remote_via_irq(queue->tx_irq);
0d160211 760
900e1833
DV		 761 u64_stats_update_begin(&tx_stats->syncp);
762 tx_stats->bytes += skb->len;
763 tx_stats->packets++;
764 u64_stats_update_end(&tx_stats->syncp);
10a273a6
JF		 765
766 /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
2688fcb7 767 xennet_tx_buf_gc(queue);
0d160211 768
2688fcb7
AB		 769 if (!netfront_tx_slot_available(queue))
770 netif_tx_stop_queue(netdev_get_tx_queue(dev, queue->id));
0d160211 771
2688fcb7 772 spin_unlock_irqrestore(&queue->tx_lock, flags);
0d160211 773
6ed10654 774 return NETDEV_TX_OK;
0d160211
JF		 775
776 drop:
09f75cd7 777 dev->stats.tx_dropped++;
979de8a0 778 dev_kfree_skb_any(skb);
6ed10654 779 return NETDEV_TX_OK;
0d160211
JF		 780}
781
782static int xennet_close(struct net_device *dev)
783{
784 struct netfront_info *np = netdev_priv(dev);
2688fcb7
AB		 785 unsigned int num_queues = dev->real_num_tx_queues;
786 unsigned int i;
787 struct netfront_queue *queue;
788 netif_tx_stop_all_queues(np->netdev);
789 for (i = 0; i < num_queues; ++i) {
790 queue = &np->queues[i];
791 napi_disable(&queue->napi);
792 }
0d160211
JF		 793 return 0;
794}
795
2688fcb7 796static void xennet_move_rx_slot(struct netfront_queue *queue, struct sk_buff *skb,
0d160211
JF		 797 grant_ref_t ref)
798{
2688fcb7
AB		 799 int new = xennet_rxidx(queue->rx.req_prod_pvt);
800
801 BUG_ON(queue->rx_skbs[new]);
802 queue->rx_skbs[new] = skb;
803 queue->grant_rx_ref[new] = ref;
804 RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->id = new;
805 RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->gref = ref;
806 queue->rx.req_prod_pvt++;
0d160211
JF		 807}
808
2688fcb7 809static int xennet_get_extras(struct netfront_queue *queue,
0d160211
JF		 810 struct xen_netif_extra_info *extras,
811 RING_IDX rp)
812
813{
814 struct xen_netif_extra_info *extra;
2688fcb7
AB		 815 struct device *dev = &queue->info->netdev->dev;
816 RING_IDX cons = queue->rx.rsp_cons;
0d160211
JF		 817 int err = 0;
818
819 do {
820 struct sk_buff *skb;
821 grant_ref_t ref;
822
823 if (unlikely(cons + 1 == rp)) {
824 if (net_ratelimit())
825 dev_warn(dev, "Missing extra info\n");
826 err = -EBADR;
827 break;
828 }
829
830 extra = (struct xen_netif_extra_info *)
2688fcb7 831 RING_GET_RESPONSE(&queue->rx, ++cons);
0d160211
JF		 832
833 if (unlikely(!extra->type ||
834 extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
835 if (net_ratelimit())
836 dev_warn(dev, "Invalid extra type: %d\n",
837 extra->type);
838 err = -EINVAL;
839 } else {
840 memcpy(&extras[extra->type - 1], extra,
841 sizeof(*extra));
842 }
843
2688fcb7
AB		 844 skb = xennet_get_rx_skb(queue, cons);
845 ref = xennet_get_rx_ref(queue, cons);
846 xennet_move_rx_slot(queue, skb, ref);
0d160211
JF		 847 } while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE);
848
2688fcb7 849 queue->rx.rsp_cons = cons;
0d160211
JF		 850 return err;
851}
852
6c5aa6fc
DK		 853static u32 xennet_run_xdp(struct netfront_queue *queue, struct page *pdata,
854 struct xen_netif_rx_response *rx, struct bpf_prog *prog,
855 struct xdp_buff *xdp, bool *need_xdp_flush)
856{
857 struct xdp_frame *xdpf;
858 u32 len = rx->status;
859 u32 act = XDP_PASS;
860 int err;
861
862 xdp->data_hard_start = page_address(pdata);
863 xdp->data = xdp->data_hard_start + XDP_PACKET_HEADROOM;
864 xdp_set_data_meta_invalid(xdp);
865 xdp->data_end = xdp->data + len;
866 xdp->rxq = &queue->xdp_rxq;
867 xdp->frame_sz = XEN_PAGE_SIZE - XDP_PACKET_HEADROOM;
868
869 act = bpf_prog_run_xdp(prog, xdp);
870 switch (act) {
871 case XDP_TX:
872 get_page(pdata);
873 xdpf = xdp_convert_buff_to_frame(xdp);
874 err = xennet_xdp_xmit(queue->info->netdev, 1, &xdpf, 0);
875 if (unlikely(err < 0))
876 trace_xdp_exception(queue->info->netdev, prog, act);
877 break;
878 case XDP_REDIRECT:
879 get_page(pdata);
880 err = xdp_do_redirect(queue->info->netdev, xdp, prog);
881 *need_xdp_flush = true;
882 if (unlikely(err))
883 trace_xdp_exception(queue->info->netdev, prog, act);
884 break;
885 case XDP_PASS:
886 case XDP_DROP:
887 break;
888
889 case XDP_ABORTED:
890 trace_xdp_exception(queue->info->netdev, prog, act);
891 break;
892
893 default:
894 bpf_warn_invalid_xdp_action(act);
895 }
896
897 return act;
898}
899
2688fcb7 900static int xennet_get_responses(struct netfront_queue *queue,
0d160211 901 struct netfront_rx_info *rinfo, RING_IDX rp,
6c5aa6fc
DK		 902 struct sk_buff_head *list,
903 bool *need_xdp_flush)
0d160211
JF		 904{
905 struct xen_netif_rx_response *rx = &rinfo->rx;
6c5aa6fc 906 int max = XEN_NETIF_NR_SLOTS_MIN + (rx->status <= RX_COPY_THRESHOLD);
2688fcb7
AB		 907 RING_IDX cons = queue->rx.rsp_cons;
908 struct sk_buff *skb = xennet_get_rx_skb(queue, cons);
6c5aa6fc 909 struct xen_netif_extra_info *extras = rinfo->extras;
2688fcb7 910 grant_ref_t ref = xennet_get_rx_ref(queue, cons);
6c5aa6fc
DK		 911 struct device *dev = &queue->info->netdev->dev;
912 struct bpf_prog *xdp_prog;
913 struct xdp_buff xdp;
914 unsigned long ret;
7158ff6d 915 int slots = 1;
0d160211 916 int err = 0;
6c5aa6fc 917 u32 verdict;
0d160211 918
f942dc25 919 if (rx->flags & XEN_NETRXF_extra_info) {
2688fcb7 920 err = xennet_get_extras(queue, extras, rp);
6c5aa6fc
DK		 921 if (!err) {
922 if (extras[XEN_NETIF_EXTRA_TYPE_XDP - 1].type) {
923 struct xen_netif_extra_info *xdp;
924
925 xdp = &extras[XEN_NETIF_EXTRA_TYPE_XDP - 1];
926 rx->offset = xdp->u.xdp.headroom;
927 }
928 }
2688fcb7 929 cons = queue->rx.rsp_cons;
0d160211
JF		 930 }
931
932 for (;;) {
933 if (unlikely(rx->status < 0 ||
30c5d7f0 934 rx->offset + rx->status > XEN_PAGE_SIZE)) {
0d160211 935 if (net_ratelimit())
6c10127d 936 dev_warn(dev, "rx->offset: %u, size: %d\n",
0d160211 937 rx->offset, rx->status);
2688fcb7 938 xennet_move_rx_slot(queue, skb, ref);
0d160211
JF		 939 err = -EINVAL;
940 goto next;
941 }
942
943 /*
944 * This definitely indicates a bug, either in this driver or in
945 * the backend driver. In future this should flag the bad
697089dc 946 * situation to the system controller to reboot the backend.
0d160211
JF		 947 */
948 if (ref == GRANT_INVALID_REF) {
949 if (net_ratelimit())
950 dev_warn(dev, "Bad rx response id %d.\n",
951 rx->id);
952 err = -EINVAL;
953 goto next;
954 }
955
956 ret = gnttab_end_foreign_access_ref(ref, 0);
957 BUG_ON(!ret);
958
2688fcb7 959 gnttab_release_grant_reference(&queue->gref_rx_head, ref);
0d160211 960
6c5aa6fc
DK		 961 rcu_read_lock();
962 xdp_prog = rcu_dereference(queue->xdp_prog);
963 if (xdp_prog) {
964 if (!(rx->flags & XEN_NETRXF_more_data)) {
965 /* currently only a single page contains data */
966 verdict = xennet_run_xdp(queue,
967 skb_frag_page(&skb_shinfo(skb)->frags[0]),
968 rx, xdp_prog, &xdp, need_xdp_flush);
969 if (verdict != XDP_PASS)
970 err = -EINVAL;
971 } else {
972 /* drop the frame */
973 err = -EINVAL;
974 }
975 }
976 rcu_read_unlock();
0d160211 977next:
6c5aa6fc 978 __skb_queue_tail(list, skb);
f942dc25 979 if (!(rx->flags & XEN_NETRXF_more_data))
0d160211
JF		 980 break;
981
7158ff6d 982 if (cons + slots == rp) {
0d160211 983 if (net_ratelimit())
7158ff6d 984 dev_warn(dev, "Need more slots\n");
0d160211
JF		 985 err = -ENOENT;
986 break;
987 }
988
2688fcb7
AB		 989 rx = RING_GET_RESPONSE(&queue->rx, cons + slots);
990 skb = xennet_get_rx_skb(queue, cons + slots);
991 ref = xennet_get_rx_ref(queue, cons + slots);
7158ff6d 992 slots++;
0d160211
JF		 993 }
994
7158ff6d 995 if (unlikely(slots > max)) {
0d160211 996 if (net_ratelimit())
697089dc 997 dev_warn(dev, "Too many slots\n");
0d160211
JF		 998 err = -E2BIG;
999 }
1000
1001 if (unlikely(err))
2688fcb7 1002 queue->rx.rsp_cons = cons + slots;
0d160211
JF		 1003
1004 return err;
1005}
1006
1007static int xennet_set_skb_gso(struct sk_buff *skb,
1008 struct xen_netif_extra_info *gso)
1009{
1010 if (!gso->u.gso.size) {
1011 if (net_ratelimit())
383eda32 1012 pr_warn("GSO size must not be zero\n");
0d160211
JF		 1013 return -EINVAL;
1014 }
1015
2c0057de
PD		 1016 if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4 &&
1017 gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV6) {
0d160211 1018 if (net_ratelimit())
383eda32 1019 pr_warn("Bad GSO type %d\n", gso->u.gso.type);
0d160211
JF		 1020 return -EINVAL;
1021 }
1022
1023 skb_shinfo(skb)->gso_size = gso->u.gso.size;
2c0057de
PD		 1024 skb_shinfo(skb)->gso_type =
1025 (gso->u.gso.type == XEN_NETIF_GSO_TYPE_TCPV4) ?
1026 SKB_GSO_TCPV4 :
1027 SKB_GSO_TCPV6;
0d160211
JF		 1028
1029 /* Header must be checked, and gso_segs computed. */
1030 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1031 skb_shinfo(skb)->gso_segs = 0;
1032
1033 return 0;
1034}
1035
a761129e
DZ		 1036static int xennet_fill_frags(struct netfront_queue *queue,
1037 struct sk_buff *skb,
1038 struct sk_buff_head *list)
0d160211 1039{
2688fcb7 1040 RING_IDX cons = queue->rx.rsp_cons;
0d160211
JF		 1041 struct sk_buff *nskb;
1042
1043 while ((nskb = __skb_dequeue(list))) {
1044 struct xen_netif_rx_response *rx =
2688fcb7 1045 RING_GET_RESPONSE(&queue->rx, ++cons);
01c68026 1046 skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0];
0d160211 1047
d472b3a6 1048 if (skb_shinfo(skb)->nr_frags == MAX_SKB_FRAGS) {
093b9c71 1049 unsigned int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
0d160211 1050
d81c5054 1051 BUG_ON(pull_to < skb_headlen(skb));
093b9c71
JB		 1052 __pskb_pull_tail(skb, pull_to - skb_headlen(skb));
1053 }
ad4f15dc 1054 if (unlikely(skb_shinfo(skb)->nr_frags >= MAX_SKB_FRAGS)) {
00b36850 1055 queue->rx.rsp_cons = ++cons + skb_queue_len(list);
ad4f15dc 1056 kfree_skb(nskb);
a761129e 1057 return -ENOENT;
ad4f15dc 1058 }
093b9c71 1059
d472b3a6
JG		 1060 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags,
1061 skb_frag_page(nfrag),
093b9c71 1062 rx->offset, rx->status, PAGE_SIZE);
0d160211
JF		 1063
1064 skb_shinfo(nskb)->nr_frags = 0;
1065 kfree_skb(nskb);
0d160211
JF		 1066 }
1067
a761129e
DZ		 1068 queue->rx.rsp_cons = cons;
1069
1070 return 0;
0d160211
JF		 1071}
1072
e0ce4af9 1073static int checksum_setup(struct net_device *dev, struct sk_buff *skb)
0d160211 1074{
b5cf66cd 1075 bool recalculate_partial_csum = false;
e0ce4af9
IC		 1076
1077 /*
1078 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
1079 * peers can fail to set NETRXF_csum_blank when sending a GSO
1080 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
1081 * recalculate the partial checksum.
1082 */
1083 if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
1084 struct netfront_info *np = netdev_priv(dev);
2688fcb7 1085 atomic_inc(&np->rx_gso_checksum_fixup);
e0ce4af9 1086 skb->ip_summed = CHECKSUM_PARTIAL;
b5cf66cd 1087 recalculate_partial_csum = true;
e0ce4af9
IC		 1088 }
1089
1090 /* A non-CHECKSUM_PARTIAL SKB does not require setup. */
1091 if (skb->ip_summed != CHECKSUM_PARTIAL)
1092 return 0;
0d160211 1093
b5cf66cd 1094 return skb_checksum_setup(skb, recalculate_partial_csum);
0d160211
JF		 1095}
1096
2688fcb7 1097static int handle_incoming_queue(struct netfront_queue *queue,
09f75cd7 1098 struct sk_buff_head *rxq)
0d160211 1099{
900e1833 1100 struct netfront_stats *rx_stats = this_cpu_ptr(queue->info->rx_stats);
0d160211
JF		 1101 int packets_dropped = 0;
1102 struct sk_buff *skb;
1103
1104 while ((skb = __skb_dequeue(rxq)) != NULL) {
3683243b 1105 int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
0d160211 1106
093b9c71
JB		 1107 if (pull_to > skb_headlen(skb))
1108 __pskb_pull_tail(skb, pull_to - skb_headlen(skb));
0d160211
JF		 1109
1110 /* Ethernet work: Delayed to here as it peeks the header. */
2688fcb7 1111 skb->protocol = eth_type_trans(skb, queue->info->netdev);
d554f73d 1112 skb_reset_network_header(skb);
0d160211 1113
2688fcb7 1114 if (checksum_setup(queue->info->netdev, skb)) {
e0ce4af9
IC		 1115 kfree_skb(skb);
1116 packets_dropped++;
2688fcb7 1117 queue->info->netdev->stats.rx_errors++;
e0ce4af9 1118 continue;
0d160211
JF		 1119 }
1120
900e1833
DV		 1121 u64_stats_update_begin(&rx_stats->syncp);
1122 rx_stats->packets++;
1123 rx_stats->bytes += skb->len;
1124 u64_stats_update_end(&rx_stats->syncp);
0d160211
JF		 1125
1126 /* Pass it up. */
2688fcb7 1127 napi_gro_receive(&queue->napi, skb);
0d160211
JF		 1128 }
1129
1130 return packets_dropped;
1131}
1132
bea3348e 1133static int xennet_poll(struct napi_struct *napi, int budget)
0d160211 1134{
2688fcb7
AB		 1135 struct netfront_queue *queue = container_of(napi, struct netfront_queue, napi);
1136 struct net_device *dev = queue->info->netdev;
0d160211
JF		 1137 struct sk_buff *skb;
1138 struct netfront_rx_info rinfo;
1139 struct xen_netif_rx_response *rx = &rinfo.rx;
1140 struct xen_netif_extra_info *extras = rinfo.extras;
1141 RING_IDX i, rp;
bea3348e 1142 int work_done;
0d160211
JF		 1143 struct sk_buff_head rxq;
1144 struct sk_buff_head errq;
1145 struct sk_buff_head tmpq;
0d160211 1146 int err;
6c5aa6fc 1147 bool need_xdp_flush = false;
0d160211 1148
2688fcb7 1149 spin_lock(&queue->rx_lock);
0d160211 1150
0d160211
JF		 1151 skb_queue_head_init(&rxq);
1152 skb_queue_head_init(&errq);
1153 skb_queue_head_init(&tmpq);
1154
2688fcb7 1155 rp = queue->rx.sring->rsp_prod;
0d160211
JF		 1156 rmb(); /* Ensure we see queued responses up to 'rp'. */
1157
2688fcb7 1158 i = queue->rx.rsp_cons;
0d160211
JF		 1159 work_done = 0;
1160 while ((i != rp) && (work_done < budget)) {
2688fcb7 1161 memcpy(rx, RING_GET_RESPONSE(&queue->rx, i), sizeof(*rx));
0d160211
JF		 1162 memset(extras, 0, sizeof(rinfo.extras));
1163
6c5aa6fc
DK		 1164 err = xennet_get_responses(queue, &rinfo, rp, &tmpq,
1165 &need_xdp_flush);
0d160211
JF		 1166
1167 if (unlikely(err)) {
1168err:
1169 while ((skb = __skb_dequeue(&tmpq)))
1170 __skb_queue_tail(&errq, skb);
09f75cd7 1171 dev->stats.rx_errors++;
2688fcb7 1172 i = queue->rx.rsp_cons;
0d160211
JF		 1173 continue;
1174 }
1175
1176 skb = __skb_dequeue(&tmpq);
1177
1178 if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
1179 struct xen_netif_extra_info *gso;
1180 gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
1181
1182 if (unlikely(xennet_set_skb_gso(skb, gso))) {
1183 __skb_queue_head(&tmpq, skb);
2688fcb7 1184 queue->rx.rsp_cons += skb_queue_len(&tmpq);
0d160211
JF		 1185 goto err;
1186 }
1187 }
1188
3683243b
IC		 1189 NETFRONT_SKB_CB(skb)->pull_to = rx->status;
1190 if (NETFRONT_SKB_CB(skb)->pull_to > RX_COPY_THRESHOLD)
1191 NETFRONT_SKB_CB(skb)->pull_to = RX_COPY_THRESHOLD;
0d160211 1192
b54c9d5b 1193 skb_frag_off_set(&skb_shinfo(skb)->frags[0], rx->offset);
3683243b
IC		 1194 skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status);
1195 skb->data_len = rx->status;
093b9c71 1196 skb->len += rx->status;
0d160211 1197
a761129e 1198 if (unlikely(xennet_fill_frags(queue, skb, &tmpq)))
ad4f15dc 1199 goto err;
0d160211 1200
f942dc25 1201 if (rx->flags & XEN_NETRXF_csum_blank)
0d160211 1202 skb->ip_summed = CHECKSUM_PARTIAL;
f942dc25 1203 else if (rx->flags & XEN_NETRXF_data_validated)
0d160211
JF		 1204 skb->ip_summed = CHECKSUM_UNNECESSARY;
1205
1206 __skb_queue_tail(&rxq, skb);
1207
a761129e 1208 i = ++queue->rx.rsp_cons;
0d160211
JF		 1209 work_done++;
1210 }
6c5aa6fc
DK		 1211 if (need_xdp_flush)
1212 xdp_do_flush();
0d160211 1213
56cfe5d0 1214 __skb_queue_purge(&errq);
0d160211 1215
2688fcb7 1216 work_done -= handle_incoming_queue(queue, &rxq);
0d160211 1217
2688fcb7 1218 xennet_alloc_rx_buffers(queue);
0d160211 1219
0d160211 1220 if (work_done < budget) {
bea3348e
SH		 1221 int more_to_do = 0;
1222
6ad20165 1223 napi_complete_done(napi, work_done);
0d160211 1224
2688fcb7 1225 RING_FINAL_CHECK_FOR_RESPONSES(&queue->rx, more_to_do);
6a6dc08f
DV		 1226 if (more_to_do)
1227 napi_schedule(napi);
0d160211
JF		 1228 }
1229
2688fcb7 1230 spin_unlock(&queue->rx_lock);
0d160211 1231
bea3348e 1232 return work_done;
0d160211
JF		 1233}
1234
1235static int xennet_change_mtu(struct net_device *dev, int mtu)
1236{
0c36820e 1237 int max = xennet_can_sg(dev) ? XEN_NETIF_MAX_TX_SIZE : ETH_DATA_LEN;
0d160211
JF		 1238
1239 if (mtu > max)
1240 return -EINVAL;
1241 dev->mtu = mtu;
1242 return 0;
1243}
1244
bc1f4470 1245static void xennet_get_stats64(struct net_device *dev,
1246 struct rtnl_link_stats64 *tot)
e00f85be 1247{
1248 struct netfront_info *np = netdev_priv(dev);
1249 int cpu;
1250
1251 for_each_possible_cpu(cpu) {
900e1833
DV		 1252 struct netfront_stats *rx_stats = per_cpu_ptr(np->rx_stats, cpu);
1253 struct netfront_stats *tx_stats = per_cpu_ptr(np->tx_stats, cpu);
e00f85be 1254 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1255 unsigned int start;
1256
1257 do {
900e1833
DV		 1258 start = u64_stats_fetch_begin_irq(&tx_stats->syncp);
1259 tx_packets = tx_stats->packets;
1260 tx_bytes = tx_stats->bytes;
1261 } while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start));
e00f85be 1262
900e1833
DV		 1263 do {
1264 start = u64_stats_fetch_begin_irq(&rx_stats->syncp);
1265 rx_packets = rx_stats->packets;
1266 rx_bytes = rx_stats->bytes;
1267 } while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start));
e00f85be 1268
1269 tot->rx_packets += rx_packets;
1270 tot->tx_packets += tx_packets;
1271 tot->rx_bytes += rx_bytes;
1272 tot->tx_bytes += tx_bytes;
1273 }
1274
1275 tot->rx_errors = dev->stats.rx_errors;
1276 tot->tx_dropped = dev->stats.tx_dropped;
e00f85be 1277}
1278
2688fcb7 1279static void xennet_release_tx_bufs(struct netfront_queue *queue)
0d160211
JF		 1280{
1281 struct sk_buff *skb;
1282 int i;
1283
1284 for (i = 0; i < NET_TX_RING_SIZE; i++) {
1285 /* Skip over entries which are actually freelist references */
2688fcb7 1286 if (skb_entry_is_link(&queue->tx_skbs[i]))
0d160211
JF		 1287 continue;
1288
2688fcb7
AB		 1289 skb = queue->tx_skbs[i].skb;
1290 get_page(queue->grant_tx_page[i]);
1291 gnttab_end_foreign_access(queue->grant_tx_ref[i],
cefe0078 1292 GNTMAP_readonly,
2688fcb7
AB		 1293 (unsigned long)page_address(queue->grant_tx_page[i]));
1294 queue->grant_tx_page[i] = NULL;
1295 queue->grant_tx_ref[i] = GRANT_INVALID_REF;
1296 add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, i);
0d160211
JF		 1297 dev_kfree_skb_irq(skb);
1298 }
1299}
1300
2688fcb7 1301static void xennet_release_rx_bufs(struct netfront_queue *queue)
0d160211 1302{
0d160211
JF		 1303 int id, ref;
1304
2688fcb7 1305 spin_lock_bh(&queue->rx_lock);
0d160211
JF		 1306
1307 for (id = 0; id < NET_RX_RING_SIZE; id++) {
cefe0078
AL		 1308 struct sk_buff *skb;
1309 struct page *page;
0d160211 1310
2688fcb7 1311 skb = queue->rx_skbs[id];
cefe0078 1312 if (!skb)
0d160211 1313 continue;
0d160211 1314
2688fcb7 1315 ref = queue->grant_rx_ref[id];
cefe0078
AL		 1316 if (ref == GRANT_INVALID_REF)
1317 continue;
0d160211 1318
cefe0078 1319 page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
0d160211 1320
cefe0078
AL		 1321 /* gnttab_end_foreign_access() needs a page ref until
1322 * foreign access is ended (which may be deferred).
1323 */
1324 get_page(page);
1325 gnttab_end_foreign_access(ref, 0,
1326 (unsigned long)page_address(page));
2688fcb7 1327 queue->grant_rx_ref[id] = GRANT_INVALID_REF;
0d160211 1328
cefe0078 1329 kfree_skb(skb);
0d160211
JF		 1330 }
1331
2688fcb7 1332 spin_unlock_bh(&queue->rx_lock);
0d160211
JF		 1333}
1334
c8f44aff
MM		 1335static netdev_features_t xennet_fix_features(struct net_device *dev,
1336 netdev_features_t features)
8f7b01a1
ED		 1337{
1338 struct netfront_info *np = netdev_priv(dev);
8f7b01a1 1339
2890ea5c
JG		 1340 if (features & NETIF_F_SG &&
1341 !xenbus_read_unsigned(np->xbdev->otherend, "feature-sg", 0))
1342 features &= ~NETIF_F_SG;
8f7b01a1 1343
2890ea5c
JG		 1344 if (features & NETIF_F_IPV6_CSUM &&
1345 !xenbus_read_unsigned(np->xbdev->otherend,
1346 "feature-ipv6-csum-offload", 0))
1347 features &= ~NETIF_F_IPV6_CSUM;
8f7b01a1 1348
2890ea5c
JG		 1349 if (features & NETIF_F_TSO &&
1350 !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv4", 0))
1351 features &= ~NETIF_F_TSO;
8f7b01a1 1352
2890ea5c
JG		 1353 if (features & NETIF_F_TSO6 &&
1354 !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv6", 0))
1355 features &= ~NETIF_F_TSO6;
2c0057de 1356
8f7b01a1
ED		 1357 return features;
1358}
1359
c8f44aff
MM		 1360static int xennet_set_features(struct net_device *dev,
1361 netdev_features_t features)
8f7b01a1
ED		 1362{
1363 if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) {
1364 netdev_info(dev, "Reducing MTU because no SG offload");
1365 dev->mtu = ETH_DATA_LEN;
1366 }
1367
1368 return 0;
1369}
1370
d634bf2c 1371static irqreturn_t xennet_tx_interrupt(int irq, void *dev_id)
cf66f9d4 1372{
2688fcb7 1373 struct netfront_queue *queue = dev_id;
cf66f9d4
KRW		 1374 unsigned long flags;
1375
2688fcb7
AB		 1376 spin_lock_irqsave(&queue->tx_lock, flags);
1377 xennet_tx_buf_gc(queue);
1378 spin_unlock_irqrestore(&queue->tx_lock, flags);
cf66f9d4 1379
d634bf2c
WL		 1380 return IRQ_HANDLED;
1381}
1382
1383static irqreturn_t xennet_rx_interrupt(int irq, void *dev_id)
1384{
2688fcb7
AB		 1385 struct netfront_queue *queue = dev_id;
1386 struct net_device *dev = queue->info->netdev;
d634bf2c
WL		 1387
1388 if (likely(netif_carrier_ok(dev) &&
2688fcb7 1389 RING_HAS_UNCONSUMED_RESPONSES(&queue->rx)))
76541869 1390 napi_schedule(&queue->napi);
cf66f9d4 1391
d634bf2c
WL		 1392 return IRQ_HANDLED;
1393}
cf66f9d4 1394
d634bf2c
WL		 1395static irqreturn_t xennet_interrupt(int irq, void *dev_id)
1396{
1397 xennet_tx_interrupt(irq, dev_id);
1398 xennet_rx_interrupt(irq, dev_id);
cf66f9d4
KRW		 1399 return IRQ_HANDLED;
1400}
1401
1402#ifdef CONFIG_NET_POLL_CONTROLLER
1403static void xennet_poll_controller(struct net_device *dev)
1404{
2688fcb7
AB		 1405 /* Poll each queue */
1406 struct netfront_info *info = netdev_priv(dev);
1407 unsigned int num_queues = dev->real_num_tx_queues;
1408 unsigned int i;
1409 for (i = 0; i < num_queues; ++i)
1410 xennet_interrupt(0, &info->queues[i]);
cf66f9d4
KRW		 1411}
1412#endif
1413
6c5aa6fc
DK		 1414#define NETBACK_XDP_HEADROOM_DISABLE 0
1415#define NETBACK_XDP_HEADROOM_ENABLE 1
1416
1417static int talk_to_netback_xdp(struct netfront_info *np, int xdp)
1418{
1419 int err;
1420 unsigned short headroom;
1421
1422 headroom = xdp ? XDP_PACKET_HEADROOM : 0;
1423 err = xenbus_printf(XBT_NIL, np->xbdev->nodename,
1424 "xdp-headroom", "%hu",
1425 headroom);
1426 if (err)
1427 pr_warn("Error writing xdp-headroom\n");
1428
1429 return err;
1430}
1431
1432static int xennet_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1433 struct netlink_ext_ack *extack)
1434{
1435 unsigned long max_mtu = XEN_PAGE_SIZE - XDP_PACKET_HEADROOM;
1436 struct netfront_info *np = netdev_priv(dev);
1437 struct bpf_prog *old_prog;
1438 unsigned int i, err;
1439
1440 if (dev->mtu > max_mtu) {
1441 netdev_warn(dev, "XDP requires MTU less than %lu\n", max_mtu);
1442 return -EINVAL;
1443 }
1444
1445 if (!np->netback_has_xdp_headroom)
1446 return 0;
1447
1448 xenbus_switch_state(np->xbdev, XenbusStateReconfiguring);
1449
1450 err = talk_to_netback_xdp(np, prog ? NETBACK_XDP_HEADROOM_ENABLE :
1451 NETBACK_XDP_HEADROOM_DISABLE);
1452 if (err)
1453 return err;
1454
1455 /* avoid the race with XDP headroom adjustment */
1456 wait_event(module_wq,
1457 xenbus_read_driver_state(np->xbdev->otherend) ==
1458 XenbusStateReconfigured);
1459 np->netfront_xdp_enabled = true;
1460
1461 old_prog = rtnl_dereference(np->queues[0].xdp_prog);
1462
1463 if (prog)
1464 bpf_prog_add(prog, dev->real_num_tx_queues);
1465
1466 for (i = 0; i < dev->real_num_tx_queues; ++i)
1467 rcu_assign_pointer(np->queues[i].xdp_prog, prog);
1468
1469 if (old_prog)
1470 for (i = 0; i < dev->real_num_tx_queues; ++i)
1471 bpf_prog_put(old_prog);
1472
1473 xenbus_switch_state(np->xbdev, XenbusStateConnected);
1474
1475 return 0;
1476}
1477
1478static u32 xennet_xdp_query(struct net_device *dev)
1479{
1480 unsigned int num_queues = dev->real_num_tx_queues;
1481 struct netfront_info *np = netdev_priv(dev);
1482 const struct bpf_prog *xdp_prog;
1483 struct netfront_queue *queue;
1484 unsigned int i;
1485
1486 for (i = 0; i < num_queues; ++i) {
1487 queue = &np->queues[i];
1488 xdp_prog = rtnl_dereference(queue->xdp_prog);
1489 if (xdp_prog)
1490 return xdp_prog->aux->id;
1491 }
1492
1493 return 0;
1494}
1495
1496static int xennet_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1497{
1498 switch (xdp->command) {
1499 case XDP_SETUP_PROG:
1500 return xennet_xdp_set(dev, xdp->prog, xdp->extack);
1501 case XDP_QUERY_PROG:
1502 xdp->prog_id = xennet_xdp_query(dev);
1503 return 0;
1504 default:
1505 return -EINVAL;
1506 }
1507}
1508
0a0b9d2e
SH		 1509static const struct net_device_ops xennet_netdev_ops = {
1510 .ndo_open = xennet_open,
0a0b9d2e
SH		 1511 .ndo_stop = xennet_close,
1512 .ndo_start_xmit = xennet_start_xmit,
1513 .ndo_change_mtu = xennet_change_mtu,
e00f85be 1514 .ndo_get_stats64 = xennet_get_stats64,
0a0b9d2e
SH		 1515 .ndo_set_mac_address = eth_mac_addr,
1516 .ndo_validate_addr = eth_validate_addr,
fb507934
MM		 1517 .ndo_fix_features = xennet_fix_features,
1518 .ndo_set_features = xennet_set_features,
2688fcb7 1519 .ndo_select_queue = xennet_select_queue,
6c5aa6fc
DK		 1520 .ndo_bpf = xennet_xdp,
1521 .ndo_xdp_xmit = xennet_xdp_xmit,
cf66f9d4
KRW		 1522#ifdef CONFIG_NET_POLL_CONTROLLER
1523 .ndo_poll_controller = xennet_poll_controller,
1524#endif
0a0b9d2e
SH		 1525};
1526
900e1833
DV		 1527static void xennet_free_netdev(struct net_device *netdev)
1528{
1529 struct netfront_info *np = netdev_priv(netdev);
1530
1531 free_percpu(np->rx_stats);
1532 free_percpu(np->tx_stats);
1533 free_netdev(netdev);
1534}
1535
8e0e46bb 1536static struct net_device *xennet_create_dev(struct xenbus_device *dev)
0d160211 1537{
2688fcb7 1538 int err;
0d160211
JF		 1539 struct net_device *netdev;
1540 struct netfront_info *np;
1541
50ee6061 1542 netdev = alloc_etherdev_mq(sizeof(struct netfront_info), xennet_max_queues);
41de8d4c 1543 if (!netdev)
0d160211 1544 return ERR_PTR(-ENOMEM);
0d160211
JF		 1545
1546 np = netdev_priv(netdev);
1547 np->xbdev = dev;
1548
2688fcb7 1549 np->queues = NULL;
0d160211 1550
e00f85be 1551 err = -ENOMEM;
900e1833
DV		 1552 np->rx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1553 if (np->rx_stats == NULL)
1554 goto exit;
1555 np->tx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1556 if (np->tx_stats == NULL)
e00f85be 1557 goto exit;
1558
0a0b9d2e
SH		 1559 netdev->netdev_ops = &xennet_netdev_ops;
1560
fb507934
MM		 1561 netdev->features = NETIF_F_IP_CSUM | NETIF_F_RXCSUM |
1562 NETIF_F_GSO_ROBUST;
2c0057de
PD		 1563 netdev->hw_features = NETIF_F_SG |
1564 NETIF_F_IPV6_CSUM |
1565 NETIF_F_TSO | NETIF_F_TSO6;
0d160211 1566
fc3e5941
IC		 1567 /*
1568 * Assume that all hw features are available for now. This set
1569 * will be adjusted by the call to netdev_update_features() in
1570 * xennet_connect() which is the earliest point where we can
1571 * negotiate with the backend regarding supported features.
1572 */
1573 netdev->features |= netdev->hw_features;
1574
7ad24ea4 1575 netdev->ethtool_ops = &xennet_ethtool_ops;
e1043a4b 1576 netdev->min_mtu = ETH_MIN_MTU;
d0c2c997 1577 netdev->max_mtu = XEN_NETIF_MAX_TX_SIZE;
0d160211
JF		 1578 SET_NETDEV_DEV(netdev, &dev->dev);
1579
1580 np->netdev = netdev;
6c5aa6fc 1581 np->netfront_xdp_enabled = false;
0d160211
JF		 1582
1583 netif_carrier_off(netdev);
1584
b707fda2 1585 xenbus_switch_state(dev, XenbusStateInitialising);
8edfe2e9
JG		 1586 wait_event(module_wq,
1587 xenbus_read_driver_state(dev->otherend) !=
1588 XenbusStateClosed &&
1589 xenbus_read_driver_state(dev->otherend) !=
1590 XenbusStateUnknown);
0d160211
JF		 1591 return netdev;
1592
0d160211 1593 exit:
900e1833 1594 xennet_free_netdev(netdev);
0d160211
JF		 1595 return ERR_PTR(err);
1596}
1597
1598/**
1599 * Entry point to this code when a new device is created. Allocate the basic
1600 * structures and the ring buffers for communication with the backend, and
1601 * inform the backend of the appropriate details for those.
1602 */
8e0e46bb 1603static int netfront_probe(struct xenbus_device *dev,
1dd06ae8 1604 const struct xenbus_device_id *id)
0d160211
JF		 1605{
1606 int err;
1607 struct net_device *netdev;
1608 struct netfront_info *info;
1609
1610 netdev = xennet_create_dev(dev);
1611 if (IS_ERR(netdev)) {
1612 err = PTR_ERR(netdev);
1613 xenbus_dev_fatal(dev, err, "creating netdev");
1614 return err;
1615 }
1616
1617 info = netdev_priv(netdev);
1b713e00 1618 dev_set_drvdata(&dev->dev, info);
27b917e5
TI		 1619#ifdef CONFIG_SYSFS
1620 info->netdev->sysfs_groups[0] = &xennet_dev_group;
1621#endif
0d160211 1622
0d160211 1623 return 0;
0d160211
JF		 1624}
1625
1626static void xennet_end_access(int ref, void *page)
1627{
1628 /* This frees the page as a side-effect */
1629 if (ref != GRANT_INVALID_REF)
1630 gnttab_end_foreign_access(ref, 0, (unsigned long)page);
1631}
1632
1633static void xennet_disconnect_backend(struct netfront_info *info)
1634{
2688fcb7 1635 unsigned int i = 0;
2688fcb7
AB		 1636 unsigned int num_queues = info->netdev->real_num_tx_queues;
1637
f9feb1e6
DV		 1638 netif_carrier_off(info->netdev);
1639
9a873c71 1640 for (i = 0; i < num_queues && info->queues; ++i) {
76541869
DV		 1641 struct netfront_queue *queue = &info->queues[i];
1642
74470954
BO		 1643 del_timer_sync(&queue->rx_refill_timer);
1644
2688fcb7
AB		 1645 if (queue->tx_irq && (queue->tx_irq == queue->rx_irq))
1646 unbind_from_irqhandler(queue->tx_irq, queue);
1647 if (queue->tx_irq && (queue->tx_irq != queue->rx_irq)) {
1648 unbind_from_irqhandler(queue->tx_irq, queue);
1649 unbind_from_irqhandler(queue->rx_irq, queue);
1650 }
1651 queue->tx_evtchn = queue->rx_evtchn = 0;
1652 queue->tx_irq = queue->rx_irq = 0;
0d160211 1653
274b0455
CW		 1654 if (netif_running(info->netdev))
1655 napi_synchronize(&queue->napi);
f9feb1e6 1656
a5b5dc3c
DV		 1657 xennet_release_tx_bufs(queue);
1658 xennet_release_rx_bufs(queue);
1659 gnttab_free_grant_references(queue->gref_tx_head);
1660 gnttab_free_grant_references(queue->gref_rx_head);
1661
2688fcb7
AB		 1662 /* End access and free the pages */
1663 xennet_end_access(queue->tx_ring_ref, queue->tx.sring);
1664 xennet_end_access(queue->rx_ring_ref, queue->rx.sring);
0d160211 1665
2688fcb7
AB		 1666 queue->tx_ring_ref = GRANT_INVALID_REF;
1667 queue->rx_ring_ref = GRANT_INVALID_REF;
1668 queue->tx.sring = NULL;
1669 queue->rx.sring = NULL;
6c5aa6fc
DK		 1670
1671 page_pool_destroy(queue->page_pool);
2688fcb7 1672 }
0d160211
JF		 1673}
1674
1675/**
1676 * We are reconnecting to the backend, due to a suspend/resume, or a backend
1677 * driver restart. We tear down our netif structure and recreate it, but
1678 * leave the device-layer structures intact so that this is transparent to the
1679 * rest of the kernel.
1680 */
1681static int netfront_resume(struct xenbus_device *dev)
1682{
1b713e00 1683 struct netfront_info *info = dev_get_drvdata(&dev->dev);
0d160211
JF		 1684
1685 dev_dbg(&dev->dev, "%s\n", dev->nodename);
1686
1687 xennet_disconnect_backend(info);
1688 return 0;
1689}
1690
1691static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
1692{
1693 char *s, *e, *macstr;
1694 int i;
1695
1696 macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
1697 if (IS_ERR(macstr))
1698 return PTR_ERR(macstr);
1699
1700 for (i = 0; i < ETH_ALEN; i++) {
1701 mac[i] = simple_strtoul(s, &e, 16);
1702 if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
1703 kfree(macstr);
1704 return -ENOENT;
1705 }
1706 s = e+1;
1707 }
1708
1709 kfree(macstr);
1710 return 0;
1711}
1712
2688fcb7 1713static int setup_netfront_single(struct netfront_queue *queue)
d634bf2c
WL		 1714{
1715 int err;
1716
2688fcb7 1717 err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
d634bf2c
WL		 1718 if (err < 0)
1719 goto fail;
1720
2688fcb7 1721 err = bind_evtchn_to_irqhandler(queue->tx_evtchn,
d634bf2c 1722 xennet_interrupt,
2688fcb7 1723 0, queue->info->netdev->name, queue);
d634bf2c
WL		 1724 if (err < 0)
1725 goto bind_fail;
2688fcb7
AB		 1726 queue->rx_evtchn = queue->tx_evtchn;
1727 queue->rx_irq = queue->tx_irq = err;
d634bf2c
WL		 1728
1729 return 0;
1730
1731bind_fail:
2688fcb7
AB		 1732 xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1733 queue->tx_evtchn = 0;
d634bf2c
WL		 1734fail:
1735 return err;
1736}
1737
2688fcb7 1738static int setup_netfront_split(struct netfront_queue *queue)
d634bf2c
WL		 1739{
1740 int err;
1741
2688fcb7 1742 err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
d634bf2c
WL		 1743 if (err < 0)
1744 goto fail;
2688fcb7 1745 err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->rx_evtchn);
d634bf2c
WL		 1746 if (err < 0)
1747 goto alloc_rx_evtchn_fail;
1748
2688fcb7
AB		 1749 snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
1750 "%s-tx", queue->name);
1751 err = bind_evtchn_to_irqhandler(queue->tx_evtchn,
d634bf2c 1752 xennet_tx_interrupt,
2688fcb7 1753 0, queue->tx_irq_name, queue);
d634bf2c
WL		 1754 if (err < 0)
1755 goto bind_tx_fail;
2688fcb7 1756 queue->tx_irq = err;
d634bf2c 1757
2688fcb7
AB		 1758 snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
1759 "%s-rx", queue->name);
1760 err = bind_evtchn_to_irqhandler(queue->rx_evtchn,
d634bf2c 1761 xennet_rx_interrupt,
2688fcb7 1762 0, queue->rx_irq_name, queue);
d634bf2c
WL		 1763 if (err < 0)
1764 goto bind_rx_fail;
2688fcb7 1765 queue->rx_irq = err;
d634bf2c
WL		 1766
1767 return 0;
1768
1769bind_rx_fail:
2688fcb7
AB		 1770 unbind_from_irqhandler(queue->tx_irq, queue);
1771 queue->tx_irq = 0;
d634bf2c 1772bind_tx_fail:
2688fcb7
AB		 1773 xenbus_free_evtchn(queue->info->xbdev, queue->rx_evtchn);
1774 queue->rx_evtchn = 0;
d634bf2c 1775alloc_rx_evtchn_fail:
2688fcb7
AB		 1776 xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1777 queue->tx_evtchn = 0;
d634bf2c
WL		 1778fail:
1779 return err;
1780}
1781
2688fcb7
AB		 1782static int setup_netfront(struct xenbus_device *dev,
1783 struct netfront_queue *queue, unsigned int feature_split_evtchn)
0d160211
JF		 1784{
1785 struct xen_netif_tx_sring *txs;
1786 struct xen_netif_rx_sring *rxs;
ccc9d90a 1787 grant_ref_t gref;
0d160211 1788 int err;
0d160211 1789
2688fcb7
AB		 1790 queue->tx_ring_ref = GRANT_INVALID_REF;
1791 queue->rx_ring_ref = GRANT_INVALID_REF;
1792 queue->rx.sring = NULL;
1793 queue->tx.sring = NULL;
0d160211 1794
a144ff09 1795 txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
0d160211
JF		 1796 if (!txs) {
1797 err = -ENOMEM;
1798 xenbus_dev_fatal(dev, err, "allocating tx ring page");
1799 goto fail;
1800 }
1801 SHARED_RING_INIT(txs);
30c5d7f0 1802 FRONT_RING_INIT(&queue->tx, txs, XEN_PAGE_SIZE);
0d160211 1803
ccc9d90a 1804 err = xenbus_grant_ring(dev, txs, 1, &gref);
1ca2983a
WL		 1805 if (err < 0)
1806 goto grant_tx_ring_fail;
ccc9d90a 1807 queue->tx_ring_ref = gref;
0d160211 1808
a144ff09 1809 rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
0d160211
JF		 1810 if (!rxs) {
1811 err = -ENOMEM;
1812 xenbus_dev_fatal(dev, err, "allocating rx ring page");
1ca2983a 1813 goto alloc_rx_ring_fail;
0d160211
JF		 1814 }
1815 SHARED_RING_INIT(rxs);
30c5d7f0 1816 FRONT_RING_INIT(&queue->rx, rxs, XEN_PAGE_SIZE);
0d160211 1817
ccc9d90a 1818 err = xenbus_grant_ring(dev, rxs, 1, &gref);
1ca2983a
WL		 1819 if (err < 0)
1820 goto grant_rx_ring_fail;
ccc9d90a 1821 queue->rx_ring_ref = gref;
0d160211 1822
d634bf2c 1823 if (feature_split_evtchn)
2688fcb7 1824 err = setup_netfront_split(queue);
d634bf2c
WL		 1825 /* setup single event channel if
1826 * a) feature-split-event-channels == 0
1827 * b) feature-split-event-channels == 1 but failed to setup
1828 */
1829 if (!feature_split_evtchn || (feature_split_evtchn && err))
2688fcb7 1830 err = setup_netfront_single(queue);
d634bf2c 1831
0d160211 1832 if (err)
1ca2983a 1833 goto alloc_evtchn_fail;
0d160211 1834
0d160211
JF		 1835 return 0;
1836
1ca2983a
WL		 1837 /* If we fail to setup netfront, it is safe to just revoke access to
1838 * granted pages because backend is not accessing it at this point.
1839 */
1ca2983a 1840alloc_evtchn_fail:
2688fcb7 1841 gnttab_end_foreign_access_ref(queue->rx_ring_ref, 0);
1ca2983a
WL		 1842grant_rx_ring_fail:
1843 free_page((unsigned long)rxs);
1844alloc_rx_ring_fail:
2688fcb7 1845 gnttab_end_foreign_access_ref(queue->tx_ring_ref, 0);
1ca2983a
WL		 1846grant_tx_ring_fail:
1847 free_page((unsigned long)txs);
1848fail:
0d160211
JF		 1849 return err;
1850}
1851
2688fcb7
AB		 1852/* Queue-specific initialisation
1853 * This used to be done in xennet_create_dev() but must now
1854 * be run per-queue.
1855 */
1856static int xennet_init_queue(struct netfront_queue *queue)
1857{
1858 unsigned short i;
1859 int err = 0;
21f2706b 1860 char *devid;
2688fcb7
AB		 1861
1862 spin_lock_init(&queue->tx_lock);
1863 spin_lock_init(&queue->rx_lock);
1864
e99e88a9 1865 timer_setup(&queue->rx_refill_timer, rx_refill_timeout, 0);
2688fcb7 1866
21f2706b
XL		 1867 devid = strrchr(queue->info->xbdev->nodename, '/') + 1;
1868 snprintf(queue->name, sizeof(queue->name), "vif%s-q%u",
1869 devid, queue->id);
8b715010 1870
2688fcb7
AB		 1871 /* Initialise tx_skbs as a free chain containing every entry. */
1872 queue->tx_skb_freelist = 0;
1873 for (i = 0; i < NET_TX_RING_SIZE; i++) {
1874 skb_entry_set_link(&queue->tx_skbs[i], i+1);
1875 queue->grant_tx_ref[i] = GRANT_INVALID_REF;
1876 queue->grant_tx_page[i] = NULL;
1877 }
1878
1879 /* Clear out rx_skbs */
1880 for (i = 0; i < NET_RX_RING_SIZE; i++) {
1881 queue->rx_skbs[i] = NULL;
1882 queue->grant_rx_ref[i] = GRANT_INVALID_REF;
1883 }
1884
1885 /* A grant for every tx ring slot */
1f3c2eba 1886 if (gnttab_alloc_grant_references(NET_TX_RING_SIZE,
2688fcb7
AB		 1887 &queue->gref_tx_head) < 0) {
1888 pr_alert("can't alloc tx grant refs\n");
1889 err = -ENOMEM;
1890 goto exit;
1891 }
1892
1893 /* A grant for every rx ring slot */
1f3c2eba 1894 if (gnttab_alloc_grant_references(NET_RX_RING_SIZE,
2688fcb7
AB		 1895 &queue->gref_rx_head) < 0) {
1896 pr_alert("can't alloc rx grant refs\n");
1897 err = -ENOMEM;
1898 goto exit_free_tx;
1899 }
1900
2688fcb7
AB		 1901 return 0;
1902
1903 exit_free_tx:
1904 gnttab_free_grant_references(queue->gref_tx_head);
1905 exit:
1906 return err;
1907}
1908
50ee6061
AB		 1909static int write_queue_xenstore_keys(struct netfront_queue *queue,
1910 struct xenbus_transaction *xbt, int write_hierarchical)
1911{
1912 /* Write the queue-specific keys into XenStore in the traditional
1913 * way for a single queue, or in a queue subkeys for multiple
1914 * queues.
1915 */
1916 struct xenbus_device *dev = queue->info->xbdev;
1917 int err;
1918 const char *message;
1919 char *path;
1920 size_t pathsize;
1921
1922 /* Choose the correct place to write the keys */
1923 if (write_hierarchical) {
1924 pathsize = strlen(dev->nodename) + 10;
1925 path = kzalloc(pathsize, GFP_KERNEL);
1926 if (!path) {
1927 err = -ENOMEM;
1928 message = "out of memory while writing ring references";
1929 goto error;
1930 }
1931 snprintf(path, pathsize, "%s/queue-%u",
1932 dev->nodename, queue->id);
1933 } else {
1934 path = (char *)dev->nodename;
1935 }
1936
1937 /* Write ring references */
1938 err = xenbus_printf(*xbt, path, "tx-ring-ref", "%u",
1939 queue->tx_ring_ref);
1940 if (err) {
1941 message = "writing tx-ring-ref";
1942 goto error;
1943 }
1944
1945 err = xenbus_printf(*xbt, path, "rx-ring-ref", "%u",
1946 queue->rx_ring_ref);
1947 if (err) {
1948 message = "writing rx-ring-ref";
1949 goto error;
1950 }
1951
1952 /* Write event channels; taking into account both shared
1953 * and split event channel scenarios.
1954 */
1955 if (queue->tx_evtchn == queue->rx_evtchn) {
1956 /* Shared event channel */
1957 err = xenbus_printf(*xbt, path,
1958 "event-channel", "%u", queue->tx_evtchn);
1959 if (err) {
1960 message = "writing event-channel";
1961 goto error;
1962 }
1963 } else {
1964 /* Split event channels */
1965 err = xenbus_printf(*xbt, path,
1966 "event-channel-tx", "%u", queue->tx_evtchn);
1967 if (err) {
1968 message = "writing event-channel-tx";
1969 goto error;
1970 }
1971
1972 err = xenbus_printf(*xbt, path,
1973 "event-channel-rx", "%u", queue->rx_evtchn);
1974 if (err) {
1975 message = "writing event-channel-rx";
1976 goto error;
1977 }
1978 }
1979
1980 if (write_hierarchical)
1981 kfree(path);
1982 return 0;
1983
1984error:
1985 if (write_hierarchical)
1986 kfree(path);
1987 xenbus_dev_fatal(dev, err, "%s", message);
1988 return err;
1989}
1990
ce58725f
DV		 1991static void xennet_destroy_queues(struct netfront_info *info)
1992{
1993 unsigned int i;
1994
ce58725f
DV		 1995 for (i = 0; i < info->netdev->real_num_tx_queues; i++) {
1996 struct netfront_queue *queue = &info->queues[i];
1997
1998 if (netif_running(info->netdev))
1999 napi_disable(&queue->napi);
2000 netif_napi_del(&queue->napi);
2001 }
2002
ce58725f
DV		 2003 kfree(info->queues);
2004 info->queues = NULL;
2005}
2006
6c5aa6fc
DK		 2007
2008
2009static int xennet_create_page_pool(struct netfront_queue *queue)
2010{
2011 int err;
2012 struct page_pool_params pp_params = {
2013 .order = 0,
2014 .flags = 0,
2015 .pool_size = NET_RX_RING_SIZE,
2016 .nid = NUMA_NO_NODE,
2017 .dev = &queue->info->netdev->dev,
2018 .offset = XDP_PACKET_HEADROOM,
2019 .max_len = XEN_PAGE_SIZE - XDP_PACKET_HEADROOM,
2020 };
2021
2022 queue->page_pool = page_pool_create(&pp_params);
2023 if (IS_ERR(queue->page_pool)) {
2024 err = PTR_ERR(queue->page_pool);
2025 queue->page_pool = NULL;
2026 return err;
2027 }
2028
2029 err = xdp_rxq_info_reg(&queue->xdp_rxq, queue->info->netdev,
2030 queue->id);
2031 if (err) {
2032 netdev_err(queue->info->netdev, "xdp_rxq_info_reg failed\n");
2033 goto err_free_pp;
2034 }
2035
2036 err = xdp_rxq_info_reg_mem_model(&queue->xdp_rxq,
2037 MEM_TYPE_PAGE_POOL, queue->page_pool);
2038 if (err) {
2039 netdev_err(queue->info->netdev, "xdp_rxq_info_reg_mem_model failed\n");
2040 goto err_unregister_rxq;
2041 }
2042 return 0;
2043
2044err_unregister_rxq:
2045 xdp_rxq_info_unreg(&queue->xdp_rxq);
2046err_free_pp:
2047 page_pool_destroy(queue->page_pool);
2048 queue->page_pool = NULL;
2049 return err;
2050}
2051
ce58725f 2052static int xennet_create_queues(struct netfront_info *info,
ca88ea12 2053 unsigned int *num_queues)
ce58725f
DV		 2054{
2055 unsigned int i;
2056 int ret;
2057
ca88ea12 2058 info->queues = kcalloc(*num_queues, sizeof(struct netfront_queue),
ce58725f
DV		 2059 GFP_KERNEL);
2060 if (!info->queues)
2061 return -ENOMEM;
2062
ca88ea12 2063 for (i = 0; i < *num_queues; i++) {
ce58725f
DV		 2064 struct netfront_queue *queue = &info->queues[i];
2065
2066 queue->id = i;
2067 queue->info = info;
2068
2069 ret = xennet_init_queue(queue);
2070 if (ret < 0) {
f599c64f 2071 dev_warn(&info->xbdev->dev,
69cb8524 2072 "only created %d queues\n", i);
ca88ea12 2073 *num_queues = i;
ce58725f
DV		 2074 break;
2075 }
2076
6c5aa6fc
DK		 2077 /* use page pool recycling instead of buddy allocator */
2078 ret = xennet_create_page_pool(queue);
2079 if (ret < 0) {
2080 dev_err(&info->xbdev->dev, "can't allocate page pool\n");
2081 *num_queues = i;
2082 return ret;
2083 }
2084
ce58725f
DV		 2085 netif_napi_add(queue->info->netdev, &queue->napi,
2086 xennet_poll, 64);
2087 if (netif_running(info->netdev))
2088 napi_enable(&queue->napi);
2089 }
2090
ca88ea12 2091 netif_set_real_num_tx_queues(info->netdev, *num_queues);
ce58725f 2092
ca88ea12 2093 if (*num_queues == 0) {
f599c64f 2094 dev_err(&info->xbdev->dev, "no queues\n");
ce58725f
DV		 2095 return -EINVAL;
2096 }
2097 return 0;
2098}
2099
0d160211 2100/* Common code used when first setting up, and when resuming. */
f502bf2b 2101static int talk_to_netback(struct xenbus_device *dev,
0d160211
JF		 2102 struct netfront_info *info)
2103{
2104 const char *message;
2105 struct xenbus_transaction xbt;
2106 int err;
2688fcb7
AB		 2107 unsigned int feature_split_evtchn;
2108 unsigned int i = 0;
50ee6061 2109 unsigned int max_queues = 0;
2688fcb7
AB		 2110 struct netfront_queue *queue = NULL;
2111 unsigned int num_queues = 1;
0d160211 2112
2688fcb7
AB		 2113 info->netdev->irq = 0;
2114
50ee6061 2115 /* Check if backend supports multiple queues */
2890ea5c
JG		 2116 max_queues = xenbus_read_unsigned(info->xbdev->otherend,
2117 "multi-queue-max-queues", 1);
50ee6061
AB		 2118 num_queues = min(max_queues, xennet_max_queues);
2119
2688fcb7 2120 /* Check feature-split-event-channels */
2890ea5c
JG		 2121 feature_split_evtchn = xenbus_read_unsigned(info->xbdev->otherend,
2122 "feature-split-event-channels", 0);
2688fcb7
AB		 2123
2124 /* Read mac addr. */
2125 err = xen_net_read_mac(dev, info->netdev->dev_addr);
2126 if (err) {
2127 xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
cb257783 2128 goto out_unlocked;
2688fcb7
AB		 2129 }
2130
6c5aa6fc
DK		 2131 info->netback_has_xdp_headroom = xenbus_read_unsigned(info->xbdev->otherend,
2132 "feature-xdp-headroom", 0);
2133 if (info->netback_has_xdp_headroom) {
2134 /* set the current xen-netfront xdp state */
2135 err = talk_to_netback_xdp(info, info->netfront_xdp_enabled ?
2136 NETBACK_XDP_HEADROOM_ENABLE :
2137 NETBACK_XDP_HEADROOM_DISABLE);
2138 if (err)
2139 goto out_unlocked;
2140 }
2141
f599c64f 2142 rtnl_lock();
ce58725f
DV		 2143 if (info->queues)
2144 xennet_destroy_queues(info);
2145
ca88ea12 2146 err = xennet_create_queues(info, &num_queues);
e2e004ac
RL		 2147 if (err < 0) {
2148 xenbus_dev_fatal(dev, err, "creating queues");
2149 kfree(info->queues);
2150 info->queues = NULL;
2151 goto out;
2152 }
f599c64f 2153 rtnl_unlock();
2688fcb7
AB		 2154
2155 /* Create shared ring, alloc event channel -- for each queue */
2156 for (i = 0; i < num_queues; ++i) {
2157 queue = &info->queues[i];
2688fcb7 2158 err = setup_netfront(dev, queue, feature_split_evtchn);
e2e004ac
RL		 2159 if (err)
2160 goto destroy_ring;
2688fcb7 2161 }
0d160211
JF		 2162
2163again:
2164 err = xenbus_transaction_start(&xbt);
2165 if (err) {
2166 xenbus_dev_fatal(dev, err, "starting transaction");
2167 goto destroy_ring;
2168 }
2169
812494d9 2170 if (xenbus_exists(XBT_NIL,
2171 info->xbdev->otherend, "multi-queue-max-queues")) {
50ee6061 2172 /* Write the number of queues */
812494d9 2173 err = xenbus_printf(xbt, dev->nodename,
2174 "multi-queue-num-queues", "%u", num_queues);
d634bf2c 2175 if (err) {
50ee6061
AB		 2176 message = "writing multi-queue-num-queues";
2177 goto abort_transaction_no_dev_fatal;
d634bf2c 2178 }
812494d9 2179 }
50ee6061 2180
812494d9 2181 if (num_queues == 1) {
2182 err = write_queue_xenstore_keys(&info->queues[0], &xbt, 0); /* flat */
2183 if (err)
2184 goto abort_transaction_no_dev_fatal;
2185 } else {
50ee6061
AB		 2186 /* Write the keys for each queue */
2187 for (i = 0; i < num_queues; ++i) {
2188 queue = &info->queues[i];
2189 err = write_queue_xenstore_keys(queue, &xbt, 1); /* hierarchical */
2190 if (err)
2191 goto abort_transaction_no_dev_fatal;
d634bf2c 2192 }
0d160211
JF		 2193 }
2194
50ee6061 2195 /* The remaining keys are not queue-specific */
0d160211
JF		 2196 err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u",
2197 1);
2198 if (err) {
2199 message = "writing request-rx-copy";
2200 goto abort_transaction;
2201 }
2202
2203 err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1);
2204 if (err) {
2205 message = "writing feature-rx-notify";
2206 goto abort_transaction;
2207 }
2208
2209 err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1);
2210 if (err) {
2211 message = "writing feature-sg";
2212 goto abort_transaction;
2213 }
2214
2215 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1);
2216 if (err) {
2217 message = "writing feature-gso-tcpv4";
2218 goto abort_transaction;
2219 }
2220
2c0057de
PD		 2221 err = xenbus_write(xbt, dev->nodename, "feature-gso-tcpv6", "1");
2222 if (err) {
2223 message = "writing feature-gso-tcpv6";
2224 goto abort_transaction;
2225 }
2226
2227 err = xenbus_write(xbt, dev->nodename, "feature-ipv6-csum-offload",
2228 "1");
2229 if (err) {
2230 message = "writing feature-ipv6-csum-offload";
2231 goto abort_transaction;
2232 }
2233
0d160211
JF		 2234 err = xenbus_transaction_end(xbt, 0);
2235 if (err) {
2236 if (err == -EAGAIN)
2237 goto again;
2238 xenbus_dev_fatal(dev, err, "completing transaction");
2239 goto destroy_ring;
2240 }
2241
2242 return 0;
2243
2244 abort_transaction:
0d160211 2245 xenbus_dev_fatal(dev, err, "%s", message);
50ee6061
AB		 2246abort_transaction_no_dev_fatal:
2247 xenbus_transaction_end(xbt, 1);
0d160211
JF		 2248 destroy_ring:
2249 xennet_disconnect_backend(info);
f599c64f 2250 rtnl_lock();
e2e004ac 2251 xennet_destroy_queues(info);
0d160211 2252 out:
f599c64f 2253 rtnl_unlock();
cb257783 2254out_unlocked:
d86b5672 2255 device_unregister(&dev->dev);
0d160211
JF		 2256 return err;
2257}
2258
0d160211
JF		 2259static int xennet_connect(struct net_device *dev)
2260{
2261 struct netfront_info *np = netdev_priv(dev);
2688fcb7 2262 unsigned int num_queues = 0;
a5b5dc3c 2263 int err;
2688fcb7
AB		 2264 unsigned int j = 0;
2265 struct netfront_queue *queue = NULL;
0d160211 2266
2890ea5c 2267 if (!xenbus_read_unsigned(np->xbdev->otherend, "feature-rx-copy", 0)) {
0d160211 2268 dev_info(&dev->dev,
898eb71c 2269 "backend does not support copying receive path\n");
0d160211
JF		 2270 return -ENODEV;
2271 }
2272
f502bf2b 2273 err = talk_to_netback(np->xbdev, np);
0d160211
JF		 2274 if (err)
2275 return err;
6c5aa6fc
DK		 2276 if (np->netback_has_xdp_headroom)
2277 pr_info("backend supports XDP headroom\n");
0d160211 2278
2688fcb7
AB		 2279 /* talk_to_netback() sets the correct number of queues */
2280 num_queues = dev->real_num_tx_queues;
2281
f599c64f
RL		 2282 if (dev->reg_state == NETREG_UNINITIALIZED) {
2283 err = register_netdev(dev);
2284 if (err) {
2285 pr_warn("%s: register_netdev err=%d\n", __func__, err);
2286 device_unregister(&np->xbdev->dev);
2287 return err;
2288 }
2289 }
2290
45c8184c
RL		 2291 rtnl_lock();
2292 netdev_update_features(dev);
2293 rtnl_unlock();
2294
0d160211 2295 /*
a5b5dc3c 2296 * All public and private state should now be sane. Get
0d160211
JF		 2297 * ready to start sending and receiving packets and give the driver
2298 * domain a kick because we've probably just requeued some
2299 * packets.
2300 */
2301 netif_carrier_on(np->netdev);
2688fcb7
AB		 2302 for (j = 0; j < num_queues; ++j) {
2303 queue = &np->queues[j];
f50b4076 2304
2688fcb7
AB		 2305 notify_remote_via_irq(queue->tx_irq);
2306 if (queue->tx_irq != queue->rx_irq)
2307 notify_remote_via_irq(queue->rx_irq);
2688fcb7 2308
f50b4076
DV		 2309 spin_lock_irq(&queue->tx_lock);
2310 xennet_tx_buf_gc(queue);
2688fcb7 2311 spin_unlock_irq(&queue->tx_lock);
f50b4076
DV		 2312
2313 spin_lock_bh(&queue->rx_lock);
2314 xennet_alloc_rx_buffers(queue);
2688fcb7
AB		 2315 spin_unlock_bh(&queue->rx_lock);
2316 }
0d160211
JF		 2317
2318 return 0;
2319}
2320
2321/**
2322 * Callback received when the backend's state changes.
2323 */
f502bf2b 2324static void netback_changed(struct xenbus_device *dev,
0d160211
JF		 2325 enum xenbus_state backend_state)
2326{
1b713e00 2327 struct netfront_info *np = dev_get_drvdata(&dev->dev);
0d160211
JF		 2328 struct net_device *netdev = np->netdev;
2329
2330 dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state));
2331
8edfe2e9
JG		 2332 wake_up_all(&module_wq);
2333
0d160211
JF		 2334 switch (backend_state) {
2335 case XenbusStateInitialising:
2336 case XenbusStateInitialised:
b78c9512
NI		 2337 case XenbusStateReconfiguring:
2338 case XenbusStateReconfigured:
0d160211 2339 case XenbusStateUnknown:
0d160211
JF		 2340 break;
2341
2342 case XenbusStateInitWait:
2343 if (dev->state != XenbusStateInitialising)
2344 break;
2345 if (xennet_connect(netdev) != 0)
2346 break;
2347 xenbus_switch_state(dev, XenbusStateConnected);
08e34eb1
LE		 2348 break;
2349
2350 case XenbusStateConnected:
ee89bab1 2351 netdev_notify_peers(netdev);
0d160211
JF		 2352 break;
2353
bce3ea81
DV		 2354 case XenbusStateClosed:
2355 if (dev->state == XenbusStateClosed)
2356 break;
a32b9d91 2357 /* Fall through - Missed the backend's CLOSING state. */
0d160211
JF		 2358 case XenbusStateClosing:
2359 xenbus_frontend_closed(dev);
2360 break;
2361 }
2362}
2363
e0ce4af9
IC		 2364static const struct xennet_stat {
2365 char name[ETH_GSTRING_LEN];
2366 u16 offset;
2367} xennet_stats[] = {
2368 {
2369 "rx_gso_checksum_fixup",
2370 offsetof(struct netfront_info, rx_gso_checksum_fixup)
2371 },
2372};
2373
2374static int xennet_get_sset_count(struct net_device *dev, int string_set)
2375{
2376 switch (string_set) {
2377 case ETH_SS_STATS:
2378 return ARRAY_SIZE(xennet_stats);
2379 default:
2380 return -EINVAL;
2381 }
2382}
2383
2384static void xennet_get_ethtool_stats(struct net_device *dev,
2385 struct ethtool_stats *stats, u64 * data)
2386{
2387 void *np = netdev_priv(dev);
2388 int i;
2389
2390 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2688fcb7 2391 data[i] = atomic_read((atomic_t *)(np + xennet_stats[i].offset));
e0ce4af9
IC		 2392}
2393
2394static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data)
2395{
2396 int i;
2397
2398 switch (stringset) {
2399 case ETH_SS_STATS:
2400 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2401 memcpy(data + i * ETH_GSTRING_LEN,
2402 xennet_stats[i].name, ETH_GSTRING_LEN);
2403 break;
2404 }
2405}
2406
0fc0b732 2407static const struct ethtool_ops xennet_ethtool_ops =
0d160211 2408{
0d160211 2409 .get_link = ethtool_op_get_link,
e0ce4af9
IC		 2410
2411 .get_sset_count = xennet_get_sset_count,
2412 .get_ethtool_stats = xennet_get_ethtool_stats,
2413 .get_strings = xennet_get_strings,
0d160211
JF		 2414};
2415
2416#ifdef CONFIG_SYSFS
1f3c2eba
DV		 2417static ssize_t show_rxbuf(struct device *dev,
2418 struct device_attribute *attr, char *buf)
0d160211 2419{
1f3c2eba 2420 return sprintf(buf, "%lu\n", NET_RX_RING_SIZE);
0d160211
JF		 2421}
2422
1f3c2eba
DV		 2423static ssize_t store_rxbuf(struct device *dev,
2424 struct device_attribute *attr,
2425 const char *buf, size_t len)
0d160211 2426{
0d160211
JF		 2427 char *endp;
2428 unsigned long target;
2429
2430 if (!capable(CAP_NET_ADMIN))
2431 return -EPERM;
2432
2433 target = simple_strtoul(buf, &endp, 0);
2434 if (endp == buf)
2435 return -EBADMSG;
2436
1f3c2eba 2437 /* rxbuf_min and rxbuf_max are no longer configurable. */
0d160211 2438
0d160211
JF		 2439 return len;
2440}
2441
d61e4038
JP		 2442static DEVICE_ATTR(rxbuf_min, 0644, show_rxbuf, store_rxbuf);
2443static DEVICE_ATTR(rxbuf_max, 0644, show_rxbuf, store_rxbuf);
2444static DEVICE_ATTR(rxbuf_cur, 0444, show_rxbuf, NULL);
0d160211 2445
27b917e5
TI		 2446static struct attribute *xennet_dev_attrs[] = {
2447 &dev_attr_rxbuf_min.attr,
2448 &dev_attr_rxbuf_max.attr,
2449 &dev_attr_rxbuf_cur.attr,
2450 NULL
2451};
0d160211 2452
27b917e5
TI		 2453static const struct attribute_group xennet_dev_group = {
2454 .attrs = xennet_dev_attrs
2455};
0d160211
JF		 2456#endif /* CONFIG_SYSFS */
2457
8e0e46bb 2458static int xennet_remove(struct xenbus_device *dev)
0d160211 2459{
1b713e00 2460 struct netfront_info *info = dev_get_drvdata(&dev->dev);
0d160211
JF		 2461
2462 dev_dbg(&dev->dev, "%s\n", dev->nodename);
2463
5b5971df
EO		 2464 if (xenbus_read_driver_state(dev->otherend) != XenbusStateClosed) {
2465 xenbus_switch_state(dev, XenbusStateClosing);
8edfe2e9 2466 wait_event(module_wq,
5b5971df 2467 xenbus_read_driver_state(dev->otherend) ==
c2d2e673
JA		 2468 XenbusStateClosing ||
2469 xenbus_read_driver_state(dev->otherend) ==
2470 XenbusStateUnknown);
5b5971df
EO		 2471
2472 xenbus_switch_state(dev, XenbusStateClosed);
8edfe2e9 2473 wait_event(module_wq,
5b5971df
EO		 2474 xenbus_read_driver_state(dev->otherend) ==
2475 XenbusStateClosed ||
2476 xenbus_read_driver_state(dev->otherend) ==
2477 XenbusStateUnknown);
2478 }
2479
0d160211
JF		 2480 xennet_disconnect_backend(info);
2481
f599c64f
RL		 2482 if (info->netdev->reg_state == NETREG_REGISTERED)
2483 unregister_netdev(info->netdev);
6bc96d04 2484
f599c64f
RL		 2485 if (info->queues) {
2486 rtnl_lock();
9a873c71 2487 xennet_destroy_queues(info);
f599c64f
RL		 2488 rtnl_unlock();
2489 }
900e1833 2490 xennet_free_netdev(info->netdev);
0d160211
JF		 2491
2492 return 0;
2493}
2494
95afae48
DV		 2495static const struct xenbus_device_id netfront_ids[] = {
2496 { "vif" },
2497 { "" }
2498};
2499
2500static struct xenbus_driver netfront_driver = {
2501 .ids = netfront_ids,
0d160211 2502 .probe = netfront_probe,
8e0e46bb 2503 .remove = xennet_remove,
0d160211 2504 .resume = netfront_resume,
f502bf2b 2505 .otherend_changed = netback_changed,
95afae48 2506};
0d160211
JF		 2507
2508static int __init netif_init(void)
2509{
6e833587 2510 if (!xen_domain())
0d160211
JF		 2511 return -ENODEV;
2512
51c71a3b 2513 if (!xen_has_pv_nic_devices())
b9136d20
IM		 2514 return -ENODEV;
2515
383eda32 2516 pr_info("Initialising Xen virtual ethernet driver\n");
0d160211 2517
034702a6 2518 /* Allow as many queues as there are CPUs inut max. 8 if user has not
32a84405
WL		 2519 * specified a value.
2520 */
2521 if (xennet_max_queues == 0)
034702a6
JG		 2522 xennet_max_queues = min_t(unsigned int, MAX_QUEUES_DEFAULT,
2523 num_online_cpus());
50ee6061 2524
ffb78a26 2525 return xenbus_register_frontend(&netfront_driver);
0d160211
JF		 2526}
2527module_init(netif_init);
2528
2529
2530static void __exit netif_exit(void)
2531{
ffb78a26 2532 xenbus_unregister_driver(&netfront_driver);
0d160211
JF		 2533}
2534module_exit(netif_exit);
2535
2536MODULE_DESCRIPTION("Xen virtual network device frontend");
2537MODULE_LICENSE("GPL");
d2f0c52b 2538MODULE_ALIAS("xen:vif");
4f93f09b 2539MODULE_ALIAS("xennet");
Linux 6.x block layer and io_uring tree(s)