]>
Commit | Line | Data |
---|---|---|
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 |
64 | static unsigned int xennet_max_queues; |
65 | module_param_named(max_queues, xennet_max_queues, uint, 0644); | |
66 | MODULE_PARM_DESC(max_queues, | |
67 | "Maximum number of queues per virtual interface"); | |
68 | ||
c2c63310 AR |
69 | #define XENNET_TIMEOUT (5 * HZ) |
70 | ||
0fc0b732 | 71 | static const struct ethtool_ops xennet_ethtool_ops; |
0d160211 JF |
72 | |
73 | struct netfront_cb { | |
3683243b | 74 | int pull_to; |
0d160211 JF |
75 | }; |
76 | ||
77 | #define NETFRONT_SKB_CB(skb) ((struct netfront_cb *)((skb)->cb)) | |
78 | ||
79 | #define RX_COPY_THRESHOLD 256 | |
80 | ||
81 | #define GRANT_INVALID_REF 0 | |
82 | ||
30c5d7f0 JG |
83 | #define NET_TX_RING_SIZE __CONST_RING_SIZE(xen_netif_tx, XEN_PAGE_SIZE) |
84 | #define NET_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, XEN_PAGE_SIZE) | |
1f3c2eba DV |
85 | |
86 | /* Minimum number of Rx slots (includes slot for GSO metadata). */ | |
87 | #define NET_RX_SLOTS_MIN (XEN_NETIF_NR_SLOTS_MIN + 1) | |
0d160211 | 88 | |
2688fcb7 AB |
89 | /* Queue name is interface name with "-qNNN" appended */ |
90 | #define QUEUE_NAME_SIZE (IFNAMSIZ + 6) | |
91 | ||
92 | /* IRQ name is queue name with "-tx" or "-rx" appended */ | |
93 | #define IRQ_NAME_SIZE (QUEUE_NAME_SIZE + 3) | |
94 | ||
8edfe2e9 | 95 | static DECLARE_WAIT_QUEUE_HEAD(module_wq); |
5b5971df | 96 | |
e00f85be | 97 | struct netfront_stats { |
900e1833 DV |
98 | u64 packets; |
99 | u64 bytes; | |
e00f85be | 100 | struct u64_stats_sync syncp; |
101 | }; | |
102 | ||
2688fcb7 AB |
103 | struct netfront_info; |
104 | ||
105 | struct netfront_queue { | |
106 | unsigned int id; /* Queue ID, 0-based */ | |
107 | char name[QUEUE_NAME_SIZE]; /* DEVNAME-qN */ | |
108 | struct netfront_info *info; | |
0d160211 | 109 | |
6c5aa6fc DK |
110 | struct bpf_prog __rcu *xdp_prog; |
111 | ||
bea3348e | 112 | struct napi_struct napi; |
0d160211 | 113 | |
d634bf2c WL |
114 | /* Split event channels support, tx_* == rx_* when using |
115 | * single event channel. | |
116 | */ | |
117 | unsigned int tx_evtchn, rx_evtchn; | |
118 | unsigned int tx_irq, rx_irq; | |
119 | /* Only used when split event channels support is enabled */ | |
2688fcb7 AB |
120 | char tx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-tx */ |
121 | char rx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-rx */ | |
0d160211 | 122 | |
84284d3c JF |
123 | spinlock_t tx_lock; |
124 | struct xen_netif_tx_front_ring tx; | |
125 | int tx_ring_ref; | |
0d160211 JF |
126 | |
127 | /* | |
128 | * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries | |
129 | * are linked from tx_skb_freelist through skb_entry.link. | |
130 | * | |
131 | * NB. Freelist index entries are always going to be less than | |
132 | * PAGE_OFFSET, whereas pointers to skbs will always be equal or | |
133 | * greater than PAGE_OFFSET: we use this property to distinguish | |
134 | * them. | |
135 | */ | |
136 | union skb_entry { | |
137 | struct sk_buff *skb; | |
1ffb40b8 | 138 | unsigned long link; |
0d160211 JF |
139 | } tx_skbs[NET_TX_RING_SIZE]; |
140 | grant_ref_t gref_tx_head; | |
141 | grant_ref_t grant_tx_ref[NET_TX_RING_SIZE]; | |
cefe0078 | 142 | struct page *grant_tx_page[NET_TX_RING_SIZE]; |
0d160211 JF |
143 | unsigned tx_skb_freelist; |
144 | ||
84284d3c JF |
145 | spinlock_t rx_lock ____cacheline_aligned_in_smp; |
146 | struct xen_netif_rx_front_ring rx; | |
147 | int rx_ring_ref; | |
148 | ||
84284d3c JF |
149 | struct timer_list rx_refill_timer; |
150 | ||
0d160211 JF |
151 | struct sk_buff *rx_skbs[NET_RX_RING_SIZE]; |
152 | grant_ref_t gref_rx_head; | |
153 | grant_ref_t grant_rx_ref[NET_RX_RING_SIZE]; | |
6c5aa6fc DK |
154 | |
155 | struct page_pool *page_pool; | |
156 | struct xdp_rxq_info xdp_rxq; | |
2688fcb7 AB |
157 | }; |
158 | ||
159 | struct netfront_info { | |
160 | struct list_head list; | |
161 | struct net_device *netdev; | |
162 | ||
163 | struct xenbus_device *xbdev; | |
164 | ||
165 | /* Multi-queue support */ | |
166 | struct netfront_queue *queues; | |
e0ce4af9 IC |
167 | |
168 | /* Statistics */ | |
900e1833 DV |
169 | struct netfront_stats __percpu *rx_stats; |
170 | struct netfront_stats __percpu *tx_stats; | |
e00f85be | 171 | |
6c5aa6fc DK |
172 | /* XDP state */ |
173 | bool netback_has_xdp_headroom; | |
174 | bool netfront_xdp_enabled; | |
175 | ||
2688fcb7 | 176 | atomic_t rx_gso_checksum_fixup; |
0d160211 JF |
177 | }; |
178 | ||
179 | struct netfront_rx_info { | |
180 | struct xen_netif_rx_response rx; | |
181 | struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1]; | |
182 | }; | |
183 | ||
1ffb40b8 IY |
184 | static void skb_entry_set_link(union skb_entry *list, unsigned short id) |
185 | { | |
186 | list->link = id; | |
187 | } | |
188 | ||
189 | static int skb_entry_is_link(const union skb_entry *list) | |
190 | { | |
191 | BUILD_BUG_ON(sizeof(list->skb) != sizeof(list->link)); | |
807540ba | 192 | return (unsigned long)list->skb < PAGE_OFFSET; |
1ffb40b8 IY |
193 | } |
194 | ||
0d160211 JF |
195 | /* |
196 | * Access macros for acquiring freeing slots in tx_skbs[]. | |
197 | */ | |
198 | ||
199 | static void add_id_to_freelist(unsigned *head, union skb_entry *list, | |
200 | unsigned short id) | |
201 | { | |
1ffb40b8 | 202 | skb_entry_set_link(&list[id], *head); |
0d160211 JF |
203 | *head = id; |
204 | } | |
205 | ||
206 | static unsigned short get_id_from_freelist(unsigned *head, | |
207 | union skb_entry *list) | |
208 | { | |
209 | unsigned int id = *head; | |
210 | *head = list[id].link; | |
211 | return id; | |
212 | } | |
213 | ||
214 | static int xennet_rxidx(RING_IDX idx) | |
215 | { | |
216 | return idx & (NET_RX_RING_SIZE - 1); | |
217 | } | |
218 | ||
2688fcb7 | 219 | static struct sk_buff *xennet_get_rx_skb(struct netfront_queue *queue, |
0d160211 JF |
220 | RING_IDX ri) |
221 | { | |
222 | int i = xennet_rxidx(ri); | |
2688fcb7 AB |
223 | struct sk_buff *skb = queue->rx_skbs[i]; |
224 | queue->rx_skbs[i] = NULL; | |
0d160211 JF |
225 | return skb; |
226 | } | |
227 | ||
2688fcb7 | 228 | static grant_ref_t xennet_get_rx_ref(struct netfront_queue *queue, |
0d160211 JF |
229 | RING_IDX ri) |
230 | { | |
231 | int i = xennet_rxidx(ri); | |
2688fcb7 AB |
232 | grant_ref_t ref = queue->grant_rx_ref[i]; |
233 | queue->grant_rx_ref[i] = GRANT_INVALID_REF; | |
0d160211 JF |
234 | return ref; |
235 | } | |
236 | ||
237 | #ifdef CONFIG_SYSFS | |
27b917e5 | 238 | static const struct attribute_group xennet_dev_group; |
0d160211 JF |
239 | #endif |
240 | ||
3ad9b358 | 241 | static bool xennet_can_sg(struct net_device *dev) |
0d160211 | 242 | { |
3ad9b358 | 243 | return dev->features & NETIF_F_SG; |
0d160211 JF |
244 | } |
245 | ||
246 | ||
e99e88a9 | 247 | static void rx_refill_timeout(struct timer_list *t) |
0d160211 | 248 | { |
e99e88a9 | 249 | struct netfront_queue *queue = from_timer(queue, t, rx_refill_timer); |
2688fcb7 | 250 | napi_schedule(&queue->napi); |
0d160211 JF |
251 | } |
252 | ||
2688fcb7 | 253 | static int netfront_tx_slot_available(struct netfront_queue *queue) |
0d160211 | 254 | { |
2688fcb7 | 255 | return (queue->tx.req_prod_pvt - queue->tx.rsp_cons) < |
57f230ab | 256 | (NET_TX_RING_SIZE - XEN_NETIF_NR_SLOTS_MIN - 1); |
0d160211 JF |
257 | } |
258 | ||
2688fcb7 | 259 | static void xennet_maybe_wake_tx(struct netfront_queue *queue) |
0d160211 | 260 | { |
2688fcb7 AB |
261 | struct net_device *dev = queue->info->netdev; |
262 | struct netdev_queue *dev_queue = netdev_get_tx_queue(dev, queue->id); | |
0d160211 | 263 | |
2688fcb7 AB |
264 | if (unlikely(netif_tx_queue_stopped(dev_queue)) && |
265 | netfront_tx_slot_available(queue) && | |
0d160211 | 266 | likely(netif_running(dev))) |
2688fcb7 | 267 | netif_tx_wake_queue(netdev_get_tx_queue(dev, queue->id)); |
0d160211 JF |
268 | } |
269 | ||
1f3c2eba DV |
270 | |
271 | static struct sk_buff *xennet_alloc_one_rx_buffer(struct netfront_queue *queue) | |
0d160211 | 272 | { |
0d160211 JF |
273 | struct sk_buff *skb; |
274 | struct page *page; | |
0d160211 | 275 | |
1f3c2eba DV |
276 | skb = __netdev_alloc_skb(queue->info->netdev, |
277 | RX_COPY_THRESHOLD + NET_IP_ALIGN, | |
278 | GFP_ATOMIC | __GFP_NOWARN); | |
279 | if (unlikely(!skb)) | |
280 | return NULL; | |
0d160211 | 281 | |
6c5aa6fc DK |
282 | page = page_pool_dev_alloc_pages(queue->page_pool); |
283 | if (unlikely(!page)) { | |
1f3c2eba DV |
284 | kfree_skb(skb); |
285 | return NULL; | |
0d160211 | 286 | } |
1f3c2eba DV |
287 | skb_add_rx_frag(skb, 0, page, 0, 0, PAGE_SIZE); |
288 | ||
289 | /* Align ip header to a 16 bytes boundary */ | |
290 | skb_reserve(skb, NET_IP_ALIGN); | |
291 | skb->dev = queue->info->netdev; | |
292 | ||
293 | return skb; | |
294 | } | |
0d160211 | 295 | |
1f3c2eba DV |
296 | |
297 | static void xennet_alloc_rx_buffers(struct netfront_queue *queue) | |
298 | { | |
299 | RING_IDX req_prod = queue->rx.req_prod_pvt; | |
300 | int notify; | |
538d9291 | 301 | int err = 0; |
1f3c2eba DV |
302 | |
303 | if (unlikely(!netif_carrier_ok(queue->info->netdev))) | |
0d160211 | 304 | return; |
0d160211 | 305 | |
1f3c2eba DV |
306 | for (req_prod = queue->rx.req_prod_pvt; |
307 | req_prod - queue->rx.rsp_cons < NET_RX_RING_SIZE; | |
308 | req_prod++) { | |
309 | struct sk_buff *skb; | |
310 | unsigned short id; | |
311 | grant_ref_t ref; | |
30c5d7f0 | 312 | struct page *page; |
1f3c2eba | 313 | struct xen_netif_rx_request *req; |
0d160211 | 314 | |
1f3c2eba | 315 | skb = xennet_alloc_one_rx_buffer(queue); |
538d9291 VRP |
316 | if (!skb) { |
317 | err = -ENOMEM; | |
0d160211 | 318 | break; |
538d9291 | 319 | } |
0d160211 | 320 | |
1f3c2eba | 321 | id = xennet_rxidx(req_prod); |
0d160211 | 322 | |
2688fcb7 AB |
323 | BUG_ON(queue->rx_skbs[id]); |
324 | queue->rx_skbs[id] = skb; | |
0d160211 | 325 | |
2688fcb7 | 326 | ref = gnttab_claim_grant_reference(&queue->gref_rx_head); |
269ebce4 | 327 | WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref)); |
2688fcb7 | 328 | queue->grant_rx_ref[id] = ref; |
0d160211 | 329 | |
30c5d7f0 | 330 | page = skb_frag_page(&skb_shinfo(skb)->frags[0]); |
0d160211 | 331 | |
1f3c2eba | 332 | req = RING_GET_REQUEST(&queue->rx, req_prod); |
30c5d7f0 JG |
333 | gnttab_page_grant_foreign_access_ref_one(ref, |
334 | queue->info->xbdev->otherend_id, | |
335 | page, | |
336 | 0); | |
0d160211 JF |
337 | req->id = id; |
338 | req->gref = ref; | |
339 | } | |
340 | ||
1f3c2eba DV |
341 | queue->rx.req_prod_pvt = req_prod; |
342 | ||
538d9291 VRP |
343 | /* Try again later if there are not enough requests or skb allocation |
344 | * failed. | |
345 | * Enough requests is quantified as the sum of newly created slots and | |
346 | * the unconsumed slots at the backend. | |
347 | */ | |
348 | if (req_prod - queue->rx.rsp_cons < NET_RX_SLOTS_MIN || | |
349 | unlikely(err)) { | |
1f3c2eba DV |
350 | mod_timer(&queue->rx_refill_timer, jiffies + (HZ/10)); |
351 | return; | |
352 | } | |
353 | ||
2688fcb7 | 354 | RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->rx, notify); |
0d160211 | 355 | if (notify) |
2688fcb7 | 356 | notify_remote_via_irq(queue->rx_irq); |
0d160211 JF |
357 | } |
358 | ||
359 | static int xennet_open(struct net_device *dev) | |
360 | { | |
361 | struct netfront_info *np = netdev_priv(dev); | |
2688fcb7 AB |
362 | unsigned int num_queues = dev->real_num_tx_queues; |
363 | unsigned int i = 0; | |
364 | struct netfront_queue *queue = NULL; | |
365 | ||
f599c64f RL |
366 | if (!np->queues) |
367 | return -ENODEV; | |
368 | ||
2688fcb7 AB |
369 | for (i = 0; i < num_queues; ++i) { |
370 | queue = &np->queues[i]; | |
371 | napi_enable(&queue->napi); | |
372 | ||
373 | spin_lock_bh(&queue->rx_lock); | |
374 | if (netif_carrier_ok(dev)) { | |
375 | xennet_alloc_rx_buffers(queue); | |
376 | queue->rx.sring->rsp_event = queue->rx.rsp_cons + 1; | |
377 | if (RING_HAS_UNCONSUMED_RESPONSES(&queue->rx)) | |
378 | napi_schedule(&queue->napi); | |
379 | } | |
380 | spin_unlock_bh(&queue->rx_lock); | |
0d160211 | 381 | } |
0d160211 | 382 | |
2688fcb7 | 383 | netif_tx_start_all_queues(dev); |
0d160211 JF |
384 | |
385 | return 0; | |
386 | } | |
387 | ||
2688fcb7 | 388 | static void xennet_tx_buf_gc(struct netfront_queue *queue) |
0d160211 JF |
389 | { |
390 | RING_IDX cons, prod; | |
391 | unsigned short id; | |
0d160211 | 392 | struct sk_buff *skb; |
7d0105b5 | 393 | bool more_to_do; |
0d160211 | 394 | |
2688fcb7 | 395 | BUG_ON(!netif_carrier_ok(queue->info->netdev)); |
0d160211 JF |
396 | |
397 | do { | |
2688fcb7 | 398 | prod = queue->tx.sring->rsp_prod; |
0d160211 JF |
399 | rmb(); /* Ensure we see responses up to 'rp'. */ |
400 | ||
2688fcb7 | 401 | for (cons = queue->tx.rsp_cons; cons != prod; cons++) { |
8446066b | 402 | struct xen_netif_tx_response txrsp; |
0d160211 | 403 | |
8446066b JG |
404 | RING_COPY_RESPONSE(&queue->tx, cons, &txrsp); |
405 | if (txrsp.status == XEN_NETIF_RSP_NULL) | |
0d160211 JF |
406 | continue; |
407 | ||
8446066b | 408 | id = txrsp.id; |
2688fcb7 | 409 | skb = queue->tx_skbs[id].skb; |
0d160211 | 410 | if (unlikely(gnttab_query_foreign_access( |
2688fcb7 | 411 | queue->grant_tx_ref[id]) != 0)) { |
383eda32 JP |
412 | pr_alert("%s: warning -- grant still in use by backend domain\n", |
413 | __func__); | |
0d160211 JF |
414 | BUG(); |
415 | } | |
416 | gnttab_end_foreign_access_ref( | |
2688fcb7 | 417 | queue->grant_tx_ref[id], GNTMAP_readonly); |
0d160211 | 418 | gnttab_release_grant_reference( |
2688fcb7 AB |
419 | &queue->gref_tx_head, queue->grant_tx_ref[id]); |
420 | queue->grant_tx_ref[id] = GRANT_INVALID_REF; | |
421 | queue->grant_tx_page[id] = NULL; | |
422 | add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, id); | |
0d160211 JF |
423 | dev_kfree_skb_irq(skb); |
424 | } | |
425 | ||
2688fcb7 | 426 | queue->tx.rsp_cons = prod; |
0d160211 | 427 | |
7d0105b5 MC |
428 | RING_FINAL_CHECK_FOR_RESPONSES(&queue->tx, more_to_do); |
429 | } while (more_to_do); | |
0d160211 | 430 | |
2688fcb7 | 431 | xennet_maybe_wake_tx(queue); |
0d160211 JF |
432 | } |
433 | ||
30c5d7f0 JG |
434 | struct xennet_gnttab_make_txreq { |
435 | struct netfront_queue *queue; | |
436 | struct sk_buff *skb; | |
437 | struct page *page; | |
162081ec JG |
438 | struct xen_netif_tx_request *tx; /* Last request on ring page */ |
439 | struct xen_netif_tx_request tx_local; /* Last request local copy*/ | |
30c5d7f0 JG |
440 | unsigned int size; |
441 | }; | |
442 | ||
443 | static void xennet_tx_setup_grant(unsigned long gfn, unsigned int offset, | |
444 | unsigned int len, void *data) | |
0d160211 | 445 | { |
30c5d7f0 | 446 | struct xennet_gnttab_make_txreq *info = data; |
0d160211 | 447 | unsigned int id; |
a55e8bb8 | 448 | struct xen_netif_tx_request *tx; |
0d160211 | 449 | grant_ref_t ref; |
30c5d7f0 JG |
450 | /* convenient aliases */ |
451 | struct page *page = info->page; | |
452 | struct netfront_queue *queue = info->queue; | |
453 | struct sk_buff *skb = info->skb; | |
0d160211 | 454 | |
a55e8bb8 DV |
455 | id = get_id_from_freelist(&queue->tx_skb_freelist, queue->tx_skbs); |
456 | tx = RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++); | |
457 | ref = gnttab_claim_grant_reference(&queue->gref_tx_head); | |
269ebce4 | 458 | WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref)); |
0d160211 | 459 | |
30c5d7f0 JG |
460 | gnttab_grant_foreign_access_ref(ref, queue->info->xbdev->otherend_id, |
461 | gfn, GNTMAP_readonly); | |
0d160211 | 462 | |
a55e8bb8 DV |
463 | queue->tx_skbs[id].skb = skb; |
464 | queue->grant_tx_page[id] = page; | |
465 | queue->grant_tx_ref[id] = ref; | |
0d160211 | 466 | |
162081ec JG |
467 | info->tx_local.id = id; |
468 | info->tx_local.gref = ref; | |
469 | info->tx_local.offset = offset; | |
470 | info->tx_local.size = len; | |
471 | info->tx_local.flags = 0; | |
472 | ||
473 | *tx = info->tx_local; | |
0d160211 | 474 | |
30c5d7f0 | 475 | info->tx = tx; |
162081ec | 476 | info->size += info->tx_local.size; |
30c5d7f0 JG |
477 | } |
478 | ||
479 | static struct xen_netif_tx_request *xennet_make_first_txreq( | |
162081ec JG |
480 | struct xennet_gnttab_make_txreq *info, |
481 | unsigned int offset, unsigned int len) | |
30c5d7f0 | 482 | { |
162081ec | 483 | info->size = 0; |
30c5d7f0 | 484 | |
162081ec | 485 | gnttab_for_one_grant(info->page, offset, len, xennet_tx_setup_grant, info); |
30c5d7f0 | 486 | |
162081ec | 487 | return info->tx; |
30c5d7f0 JG |
488 | } |
489 | ||
490 | static 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 | |
162081ec JG |
500 | static void xennet_make_txreqs( |
501 | struct xennet_gnttab_make_txreq *info, | |
502 | struct page *page, | |
a55e8bb8 DV |
503 | unsigned int offset, unsigned int len) |
504 | { | |
505 | /* Skip unused frames from start of page */ | |
506 | page += offset >> PAGE_SHIFT; | |
507 | offset &= ~PAGE_MASK; | |
0d160211 | 508 | |
a55e8bb8 | 509 | while (len) { |
162081ec JG |
510 | info->page = page; |
511 | info->size = 0; | |
30c5d7f0 JG |
512 | |
513 | gnttab_foreach_grant_in_range(page, offset, len, | |
514 | xennet_make_one_txreq, | |
162081ec | 515 | info); |
30c5d7f0 | 516 | |
a55e8bb8 DV |
517 | page++; |
518 | offset = 0; | |
162081ec | 519 | len -= info->size; |
0d160211 | 520 | } |
0d160211 JF |
521 | } |
522 | ||
f36c3747 | 523 | /* |
e84448d5 DV |
524 | * Count how many ring slots are required to send this skb. Each frag |
525 | * might be a compound page. | |
f36c3747 | 526 | */ |
e84448d5 | 527 | static int xennet_count_skb_slots(struct sk_buff *skb) |
f36c3747 IC |
528 | { |
529 | int i, frags = skb_shinfo(skb)->nr_frags; | |
30c5d7f0 | 530 | int slots; |
e84448d5 | 531 | |
30c5d7f0 JG |
532 | slots = gnttab_count_grant(offset_in_page(skb->data), |
533 | skb_headlen(skb)); | |
f36c3747 IC |
534 | |
535 | for (i = 0; i < frags; i++) { | |
536 | skb_frag_t *frag = skb_shinfo(skb)->frags + i; | |
537 | unsigned long size = skb_frag_size(frag); | |
b54c9d5b | 538 | unsigned long offset = skb_frag_off(frag); |
f36c3747 IC |
539 | |
540 | /* Skip unused frames from start of page */ | |
541 | offset &= ~PAGE_MASK; | |
542 | ||
30c5d7f0 | 543 | slots += gnttab_count_grant(offset, size); |
f36c3747 IC |
544 | } |
545 | ||
30c5d7f0 | 546 | return slots; |
f36c3747 IC |
547 | } |
548 | ||
50ee6061 | 549 | static u16 xennet_select_queue(struct net_device *dev, struct sk_buff *skb, |
a350ecce | 550 | struct net_device *sb_dev) |
2688fcb7 | 551 | { |
50ee6061 AB |
552 | unsigned int num_queues = dev->real_num_tx_queues; |
553 | u32 hash; | |
554 | u16 queue_idx; | |
555 | ||
556 | /* First, check if there is only one queue */ | |
557 | if (num_queues == 1) { | |
558 | queue_idx = 0; | |
559 | } else { | |
560 | hash = skb_get_hash(skb); | |
561 | queue_idx = hash % num_queues; | |
562 | } | |
563 | ||
564 | return queue_idx; | |
2688fcb7 AB |
565 | } |
566 | ||
6c5aa6fc DK |
567 | static int xennet_xdp_xmit_one(struct net_device *dev, |
568 | struct netfront_queue *queue, | |
569 | struct xdp_frame *xdpf) | |
570 | { | |
571 | struct netfront_info *np = netdev_priv(dev); | |
572 | struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats); | |
162081ec JG |
573 | struct xennet_gnttab_make_txreq info = { |
574 | .queue = queue, | |
575 | .skb = NULL, | |
576 | .page = virt_to_page(xdpf->data), | |
577 | }; | |
6c5aa6fc DK |
578 | int notify; |
579 | ||
162081ec | 580 | xennet_make_first_txreq(&info, |
6c5aa6fc DK |
581 | offset_in_page(xdpf->data), |
582 | xdpf->len); | |
583 | ||
584 | RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify); | |
585 | if (notify) | |
586 | notify_remote_via_irq(queue->tx_irq); | |
587 | ||
588 | u64_stats_update_begin(&tx_stats->syncp); | |
589 | tx_stats->bytes += xdpf->len; | |
590 | tx_stats->packets++; | |
591 | u64_stats_update_end(&tx_stats->syncp); | |
592 | ||
593 | xennet_tx_buf_gc(queue); | |
594 | ||
595 | return 0; | |
596 | } | |
597 | ||
598 | static int xennet_xdp_xmit(struct net_device *dev, int n, | |
599 | struct xdp_frame **frames, u32 flags) | |
600 | { | |
601 | unsigned int num_queues = dev->real_num_tx_queues; | |
602 | struct netfront_info *np = netdev_priv(dev); | |
603 | struct netfront_queue *queue = NULL; | |
604 | unsigned long irq_flags; | |
fdc13979 LB |
605 | int nxmit = 0; |
606 | int i; | |
6c5aa6fc DK |
607 | |
608 | if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK)) | |
609 | return -EINVAL; | |
610 | ||
611 | queue = &np->queues[smp_processor_id() % num_queues]; | |
612 | ||
613 | spin_lock_irqsave(&queue->tx_lock, irq_flags); | |
614 | for (i = 0; i < n; i++) { | |
615 | struct xdp_frame *xdpf = frames[i]; | |
616 | ||
617 | if (!xdpf) | |
618 | continue; | |
fdc13979 LB |
619 | if (xennet_xdp_xmit_one(dev, queue, xdpf)) |
620 | break; | |
621 | nxmit++; | |
6c5aa6fc DK |
622 | } |
623 | spin_unlock_irqrestore(&queue->tx_lock, irq_flags); | |
624 | ||
fdc13979 | 625 | return nxmit; |
6c5aa6fc DK |
626 | } |
627 | ||
628 | ||
30c5d7f0 JG |
629 | #define MAX_XEN_SKB_FRAGS (65536 / XEN_PAGE_SIZE + 1) |
630 | ||
24a94b3c | 631 | static netdev_tx_t xennet_start_xmit(struct sk_buff *skb, struct net_device *dev) |
0d160211 | 632 | { |
0d160211 | 633 | struct netfront_info *np = netdev_priv(dev); |
900e1833 | 634 | struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats); |
162081ec | 635 | struct xen_netif_tx_request *first_tx; |
a55e8bb8 | 636 | unsigned int i; |
0d160211 | 637 | int notify; |
f36c3747 | 638 | int slots; |
a55e8bb8 DV |
639 | struct page *page; |
640 | unsigned int offset; | |
641 | unsigned int len; | |
cf66f9d4 | 642 | unsigned long flags; |
2688fcb7 | 643 | struct netfront_queue *queue = NULL; |
162081ec | 644 | struct xennet_gnttab_make_txreq info = { }; |
2688fcb7 AB |
645 | unsigned int num_queues = dev->real_num_tx_queues; |
646 | u16 queue_index; | |
fd07160b | 647 | struct sk_buff *nskb; |
2688fcb7 AB |
648 | |
649 | /* Drop the packet if no queues are set up */ | |
650 | if (num_queues < 1) | |
651 | goto drop; | |
652 | /* Determine which queue to transmit this SKB on */ | |
653 | queue_index = skb_get_queue_mapping(skb); | |
654 | queue = &np->queues[queue_index]; | |
0d160211 | 655 | |
9ecd1a75 WL |
656 | /* If skb->len is too big for wire format, drop skb and alert |
657 | * user about misconfiguration. | |
658 | */ | |
659 | if (unlikely(skb->len > XEN_NETIF_MAX_TX_SIZE)) { | |
660 | net_alert_ratelimited( | |
661 | "xennet: skb->len = %u, too big for wire format\n", | |
662 | skb->len); | |
663 | goto drop; | |
664 | } | |
665 | ||
e84448d5 | 666 | slots = xennet_count_skb_slots(skb); |
30c5d7f0 | 667 | if (unlikely(slots > MAX_XEN_SKB_FRAGS + 1)) { |
97a6d1bb ZK |
668 | net_dbg_ratelimited("xennet: skb rides the rocket: %d slots, %d bytes\n", |
669 | slots, skb->len); | |
670 | if (skb_linearize(skb)) | |
671 | goto drop; | |
0d160211 JF |
672 | } |
673 | ||
a55e8bb8 DV |
674 | page = virt_to_page(skb->data); |
675 | offset = offset_in_page(skb->data); | |
fd07160b VK |
676 | |
677 | /* The first req should be at least ETH_HLEN size or the packet will be | |
678 | * dropped by netback. | |
679 | */ | |
680 | if (unlikely(PAGE_SIZE - offset < ETH_HLEN)) { | |
681 | nskb = skb_copy(skb, GFP_ATOMIC); | |
682 | if (!nskb) | |
683 | goto drop; | |
62f3250f | 684 | dev_consume_skb_any(skb); |
fd07160b VK |
685 | skb = nskb; |
686 | page = virt_to_page(skb->data); | |
687 | offset = offset_in_page(skb->data); | |
688 | } | |
689 | ||
a55e8bb8 DV |
690 | len = skb_headlen(skb); |
691 | ||
2688fcb7 | 692 | spin_lock_irqsave(&queue->tx_lock, flags); |
0d160211 JF |
693 | |
694 | if (unlikely(!netif_carrier_ok(dev) || | |
f36c3747 | 695 | (slots > 1 && !xennet_can_sg(dev)) || |
8b86a61d | 696 | netif_needs_gso(skb, netif_skb_features(skb)))) { |
2688fcb7 | 697 | spin_unlock_irqrestore(&queue->tx_lock, flags); |
0d160211 JF |
698 | goto drop; |
699 | } | |
700 | ||
a55e8bb8 | 701 | /* First request for the linear area. */ |
162081ec JG |
702 | info.queue = queue; |
703 | info.skb = skb; | |
704 | info.page = page; | |
705 | first_tx = xennet_make_first_txreq(&info, offset, len); | |
706 | offset += info.tx_local.size; | |
30c5d7f0 JG |
707 | if (offset == PAGE_SIZE) { |
708 | page++; | |
709 | offset = 0; | |
710 | } | |
162081ec | 711 | len -= info.tx_local.size; |
0d160211 | 712 | |
0d160211 JF |
713 | if (skb->ip_summed == CHECKSUM_PARTIAL) |
714 | /* local packet? */ | |
162081ec JG |
715 | first_tx->flags |= XEN_NETTXF_csum_blank | |
716 | XEN_NETTXF_data_validated; | |
0d160211 JF |
717 | else if (skb->ip_summed == CHECKSUM_UNNECESSARY) |
718 | /* remote but checksummed. */ | |
162081ec | 719 | first_tx->flags |= XEN_NETTXF_data_validated; |
0d160211 | 720 | |
a55e8bb8 | 721 | /* Optional extra info after the first request. */ |
0d160211 JF |
722 | if (skb_shinfo(skb)->gso_size) { |
723 | struct xen_netif_extra_info *gso; | |
724 | ||
725 | gso = (struct xen_netif_extra_info *) | |
a55e8bb8 | 726 | RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++); |
0d160211 | 727 | |
162081ec | 728 | first_tx->flags |= XEN_NETTXF_extra_info; |
0d160211 JF |
729 | |
730 | gso->u.gso.size = skb_shinfo(skb)->gso_size; | |
2c0057de PD |
731 | gso->u.gso.type = (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) ? |
732 | XEN_NETIF_GSO_TYPE_TCPV6 : | |
733 | XEN_NETIF_GSO_TYPE_TCPV4; | |
0d160211 JF |
734 | gso->u.gso.pad = 0; |
735 | gso->u.gso.features = 0; | |
736 | ||
737 | gso->type = XEN_NETIF_EXTRA_TYPE_GSO; | |
738 | gso->flags = 0; | |
0d160211 JF |
739 | } |
740 | ||
a55e8bb8 | 741 | /* Requests for the rest of the linear area. */ |
162081ec | 742 | xennet_make_txreqs(&info, page, offset, len); |
a55e8bb8 DV |
743 | |
744 | /* Requests for all the frags. */ | |
745 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
746 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
162081ec | 747 | xennet_make_txreqs(&info, skb_frag_page(frag), |
b54c9d5b | 748 | skb_frag_off(frag), |
a55e8bb8 DV |
749 | skb_frag_size(frag)); |
750 | } | |
0d160211 | 751 | |
a55e8bb8 DV |
752 | /* First request has the packet length. */ |
753 | first_tx->size = skb->len; | |
0d160211 | 754 | |
91ffb9d3 DD |
755 | /* timestamp packet in software */ |
756 | skb_tx_timestamp(skb); | |
757 | ||
2688fcb7 | 758 | RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify); |
0d160211 | 759 | if (notify) |
2688fcb7 | 760 | notify_remote_via_irq(queue->tx_irq); |
0d160211 | 761 | |
900e1833 DV |
762 | u64_stats_update_begin(&tx_stats->syncp); |
763 | tx_stats->bytes += skb->len; | |
764 | tx_stats->packets++; | |
765 | u64_stats_update_end(&tx_stats->syncp); | |
10a273a6 JF |
766 | |
767 | /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */ | |
2688fcb7 | 768 | xennet_tx_buf_gc(queue); |
0d160211 | 769 | |
2688fcb7 AB |
770 | if (!netfront_tx_slot_available(queue)) |
771 | netif_tx_stop_queue(netdev_get_tx_queue(dev, queue->id)); | |
0d160211 | 772 | |
2688fcb7 | 773 | spin_unlock_irqrestore(&queue->tx_lock, flags); |
0d160211 | 774 | |
6ed10654 | 775 | return NETDEV_TX_OK; |
0d160211 JF |
776 | |
777 | drop: | |
09f75cd7 | 778 | dev->stats.tx_dropped++; |
979de8a0 | 779 | dev_kfree_skb_any(skb); |
6ed10654 | 780 | return NETDEV_TX_OK; |
0d160211 JF |
781 | } |
782 | ||
783 | static int xennet_close(struct net_device *dev) | |
784 | { | |
785 | struct netfront_info *np = netdev_priv(dev); | |
2688fcb7 AB |
786 | unsigned int num_queues = dev->real_num_tx_queues; |
787 | unsigned int i; | |
788 | struct netfront_queue *queue; | |
789 | netif_tx_stop_all_queues(np->netdev); | |
790 | for (i = 0; i < num_queues; ++i) { | |
791 | queue = &np->queues[i]; | |
792 | napi_disable(&queue->napi); | |
793 | } | |
0d160211 JF |
794 | return 0; |
795 | } | |
796 | ||
2688fcb7 | 797 | static void xennet_move_rx_slot(struct netfront_queue *queue, struct sk_buff *skb, |
0d160211 JF |
798 | grant_ref_t ref) |
799 | { | |
2688fcb7 AB |
800 | int new = xennet_rxidx(queue->rx.req_prod_pvt); |
801 | ||
802 | BUG_ON(queue->rx_skbs[new]); | |
803 | queue->rx_skbs[new] = skb; | |
804 | queue->grant_rx_ref[new] = ref; | |
805 | RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->id = new; | |
806 | RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->gref = ref; | |
807 | queue->rx.req_prod_pvt++; | |
0d160211 JF |
808 | } |
809 | ||
2688fcb7 | 810 | static int xennet_get_extras(struct netfront_queue *queue, |
0d160211 JF |
811 | struct xen_netif_extra_info *extras, |
812 | RING_IDX rp) | |
813 | ||
814 | { | |
8446066b | 815 | struct xen_netif_extra_info extra; |
2688fcb7 AB |
816 | struct device *dev = &queue->info->netdev->dev; |
817 | RING_IDX cons = queue->rx.rsp_cons; | |
0d160211 JF |
818 | int err = 0; |
819 | ||
820 | do { | |
821 | struct sk_buff *skb; | |
822 | grant_ref_t ref; | |
823 | ||
824 | if (unlikely(cons + 1 == rp)) { | |
825 | if (net_ratelimit()) | |
826 | dev_warn(dev, "Missing extra info\n"); | |
827 | err = -EBADR; | |
828 | break; | |
829 | } | |
830 | ||
8446066b | 831 | RING_COPY_RESPONSE(&queue->rx, ++cons, &extra); |
0d160211 | 832 | |
8446066b JG |
833 | if (unlikely(!extra.type || |
834 | extra.type >= XEN_NETIF_EXTRA_TYPE_MAX)) { | |
0d160211 JF |
835 | if (net_ratelimit()) |
836 | dev_warn(dev, "Invalid extra type: %d\n", | |
8446066b | 837 | extra.type); |
0d160211 JF |
838 | err = -EINVAL; |
839 | } else { | |
8446066b | 840 | extras[extra.type - 1] = extra; |
0d160211 JF |
841 | } |
842 | ||
2688fcb7 AB |
843 | skb = xennet_get_rx_skb(queue, cons); |
844 | ref = xennet_get_rx_ref(queue, cons); | |
845 | xennet_move_rx_slot(queue, skb, ref); | |
8446066b | 846 | } while (extra.flags & XEN_NETIF_EXTRA_FLAG_MORE); |
0d160211 | 847 | |
2688fcb7 | 848 | queue->rx.rsp_cons = cons; |
0d160211 JF |
849 | return err; |
850 | } | |
851 | ||
6c5aa6fc DK |
852 | static u32 xennet_run_xdp(struct netfront_queue *queue, struct page *pdata, |
853 | struct xen_netif_rx_response *rx, struct bpf_prog *prog, | |
854 | struct xdp_buff *xdp, bool *need_xdp_flush) | |
855 | { | |
856 | struct xdp_frame *xdpf; | |
857 | u32 len = rx->status; | |
e44f65fd | 858 | u32 act; |
6c5aa6fc DK |
859 | int err; |
860 | ||
43b5169d LB |
861 | xdp_init_buff(xdp, XEN_PAGE_SIZE - XDP_PACKET_HEADROOM, |
862 | &queue->xdp_rxq); | |
be9df4af LB |
863 | xdp_prepare_buff(xdp, page_address(pdata), XDP_PACKET_HEADROOM, |
864 | len, false); | |
6c5aa6fc DK |
865 | |
866 | act = bpf_prog_run_xdp(prog, xdp); | |
867 | switch (act) { | |
868 | case XDP_TX: | |
869 | get_page(pdata); | |
870 | xdpf = xdp_convert_buff_to_frame(xdp); | |
871 | err = xennet_xdp_xmit(queue->info->netdev, 1, &xdpf, 0); | |
fdc13979 LB |
872 | if (unlikely(!err)) |
873 | xdp_return_frame_rx_napi(xdpf); | |
874 | else if (unlikely(err < 0)) | |
6c5aa6fc DK |
875 | trace_xdp_exception(queue->info->netdev, prog, act); |
876 | break; | |
877 | case XDP_REDIRECT: | |
878 | get_page(pdata); | |
879 | err = xdp_do_redirect(queue->info->netdev, xdp, prog); | |
880 | *need_xdp_flush = true; | |
881 | if (unlikely(err)) | |
882 | trace_xdp_exception(queue->info->netdev, prog, act); | |
883 | break; | |
884 | case XDP_PASS: | |
885 | case XDP_DROP: | |
886 | break; | |
887 | ||
888 | case XDP_ABORTED: | |
889 | trace_xdp_exception(queue->info->netdev, prog, act); | |
890 | break; | |
891 | ||
892 | default: | |
893 | bpf_warn_invalid_xdp_action(act); | |
894 | } | |
895 | ||
896 | return act; | |
897 | } | |
898 | ||
2688fcb7 | 899 | static int xennet_get_responses(struct netfront_queue *queue, |
0d160211 | 900 | struct netfront_rx_info *rinfo, RING_IDX rp, |
6c5aa6fc DK |
901 | struct sk_buff_head *list, |
902 | bool *need_xdp_flush) | |
0d160211 | 903 | { |
8446066b | 904 | struct xen_netif_rx_response *rx = &rinfo->rx, rx_local; |
6c5aa6fc | 905 | int max = XEN_NETIF_NR_SLOTS_MIN + (rx->status <= RX_COPY_THRESHOLD); |
2688fcb7 AB |
906 | RING_IDX cons = queue->rx.rsp_cons; |
907 | struct sk_buff *skb = xennet_get_rx_skb(queue, cons); | |
6c5aa6fc | 908 | struct xen_netif_extra_info *extras = rinfo->extras; |
2688fcb7 | 909 | grant_ref_t ref = xennet_get_rx_ref(queue, cons); |
6c5aa6fc DK |
910 | struct device *dev = &queue->info->netdev->dev; |
911 | struct bpf_prog *xdp_prog; | |
912 | struct xdp_buff xdp; | |
913 | unsigned long ret; | |
7158ff6d | 914 | int slots = 1; |
0d160211 | 915 | int err = 0; |
6c5aa6fc | 916 | u32 verdict; |
0d160211 | 917 | |
f942dc25 | 918 | if (rx->flags & XEN_NETRXF_extra_info) { |
2688fcb7 | 919 | err = xennet_get_extras(queue, extras, rp); |
6c5aa6fc DK |
920 | if (!err) { |
921 | if (extras[XEN_NETIF_EXTRA_TYPE_XDP - 1].type) { | |
922 | struct xen_netif_extra_info *xdp; | |
923 | ||
924 | xdp = &extras[XEN_NETIF_EXTRA_TYPE_XDP - 1]; | |
925 | rx->offset = xdp->u.xdp.headroom; | |
926 | } | |
927 | } | |
2688fcb7 | 928 | cons = queue->rx.rsp_cons; |
0d160211 JF |
929 | } |
930 | ||
931 | for (;;) { | |
932 | if (unlikely(rx->status < 0 || | |
30c5d7f0 | 933 | rx->offset + rx->status > XEN_PAGE_SIZE)) { |
0d160211 | 934 | if (net_ratelimit()) |
6c10127d | 935 | dev_warn(dev, "rx->offset: %u, size: %d\n", |
0d160211 | 936 | rx->offset, rx->status); |
2688fcb7 | 937 | xennet_move_rx_slot(queue, skb, ref); |
0d160211 JF |
938 | err = -EINVAL; |
939 | goto next; | |
940 | } | |
941 | ||
942 | /* | |
943 | * This definitely indicates a bug, either in this driver or in | |
944 | * the backend driver. In future this should flag the bad | |
697089dc | 945 | * situation to the system controller to reboot the backend. |
0d160211 JF |
946 | */ |
947 | if (ref == GRANT_INVALID_REF) { | |
948 | if (net_ratelimit()) | |
949 | dev_warn(dev, "Bad rx response id %d.\n", | |
950 | rx->id); | |
951 | err = -EINVAL; | |
952 | goto next; | |
953 | } | |
954 | ||
955 | ret = gnttab_end_foreign_access_ref(ref, 0); | |
956 | BUG_ON(!ret); | |
957 | ||
2688fcb7 | 958 | gnttab_release_grant_reference(&queue->gref_rx_head, ref); |
0d160211 | 959 | |
6c5aa6fc DK |
960 | rcu_read_lock(); |
961 | xdp_prog = rcu_dereference(queue->xdp_prog); | |
962 | if (xdp_prog) { | |
963 | if (!(rx->flags & XEN_NETRXF_more_data)) { | |
964 | /* currently only a single page contains data */ | |
965 | verdict = xennet_run_xdp(queue, | |
966 | skb_frag_page(&skb_shinfo(skb)->frags[0]), | |
967 | rx, xdp_prog, &xdp, need_xdp_flush); | |
968 | if (verdict != XDP_PASS) | |
969 | err = -EINVAL; | |
970 | } else { | |
971 | /* drop the frame */ | |
972 | err = -EINVAL; | |
973 | } | |
974 | } | |
975 | rcu_read_unlock(); | |
0d160211 | 976 | next: |
6c5aa6fc | 977 | __skb_queue_tail(list, skb); |
f942dc25 | 978 | if (!(rx->flags & XEN_NETRXF_more_data)) |
0d160211 JF |
979 | break; |
980 | ||
7158ff6d | 981 | if (cons + slots == rp) { |
0d160211 | 982 | if (net_ratelimit()) |
7158ff6d | 983 | dev_warn(dev, "Need more slots\n"); |
0d160211 JF |
984 | err = -ENOENT; |
985 | break; | |
986 | } | |
987 | ||
8446066b JG |
988 | RING_COPY_RESPONSE(&queue->rx, cons + slots, &rx_local); |
989 | rx = &rx_local; | |
2688fcb7 AB |
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 | ||
1007 | static 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 |
1036 | static 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))) { | |
8446066b | 1044 | struct xen_netif_rx_response rx; |
01c68026 | 1045 | skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0]; |
0d160211 | 1046 | |
8446066b JG |
1047 | RING_COPY_RESPONSE(&queue->rx, ++cons, &rx); |
1048 | ||
d472b3a6 | 1049 | if (skb_shinfo(skb)->nr_frags == MAX_SKB_FRAGS) { |
093b9c71 | 1050 | unsigned int pull_to = NETFRONT_SKB_CB(skb)->pull_to; |
0d160211 | 1051 | |
d81c5054 | 1052 | BUG_ON(pull_to < skb_headlen(skb)); |
093b9c71 JB |
1053 | __pskb_pull_tail(skb, pull_to - skb_headlen(skb)); |
1054 | } | |
ad4f15dc | 1055 | if (unlikely(skb_shinfo(skb)->nr_frags >= MAX_SKB_FRAGS)) { |
00b36850 | 1056 | queue->rx.rsp_cons = ++cons + skb_queue_len(list); |
ad4f15dc | 1057 | kfree_skb(nskb); |
a761129e | 1058 | return -ENOENT; |
ad4f15dc | 1059 | } |
093b9c71 | 1060 | |
d472b3a6 JG |
1061 | skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, |
1062 | skb_frag_page(nfrag), | |
8446066b | 1063 | rx.offset, rx.status, PAGE_SIZE); |
0d160211 JF |
1064 | |
1065 | skb_shinfo(nskb)->nr_frags = 0; | |
1066 | kfree_skb(nskb); | |
0d160211 JF |
1067 | } |
1068 | ||
a761129e DZ |
1069 | queue->rx.rsp_cons = cons; |
1070 | ||
1071 | return 0; | |
0d160211 JF |
1072 | } |
1073 | ||
e0ce4af9 | 1074 | static int checksum_setup(struct net_device *dev, struct sk_buff *skb) |
0d160211 | 1075 | { |
b5cf66cd | 1076 | bool recalculate_partial_csum = false; |
e0ce4af9 IC |
1077 | |
1078 | /* | |
1079 | * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy | |
1080 | * peers can fail to set NETRXF_csum_blank when sending a GSO | |
1081 | * frame. In this case force the SKB to CHECKSUM_PARTIAL and | |
1082 | * recalculate the partial checksum. | |
1083 | */ | |
1084 | if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) { | |
1085 | struct netfront_info *np = netdev_priv(dev); | |
2688fcb7 | 1086 | atomic_inc(&np->rx_gso_checksum_fixup); |
e0ce4af9 | 1087 | skb->ip_summed = CHECKSUM_PARTIAL; |
b5cf66cd | 1088 | recalculate_partial_csum = true; |
e0ce4af9 IC |
1089 | } |
1090 | ||
1091 | /* A non-CHECKSUM_PARTIAL SKB does not require setup. */ | |
1092 | if (skb->ip_summed != CHECKSUM_PARTIAL) | |
1093 | return 0; | |
0d160211 | 1094 | |
b5cf66cd | 1095 | return skb_checksum_setup(skb, recalculate_partial_csum); |
0d160211 JF |
1096 | } |
1097 | ||
2688fcb7 | 1098 | static int handle_incoming_queue(struct netfront_queue *queue, |
09f75cd7 | 1099 | struct sk_buff_head *rxq) |
0d160211 | 1100 | { |
900e1833 | 1101 | struct netfront_stats *rx_stats = this_cpu_ptr(queue->info->rx_stats); |
0d160211 JF |
1102 | int packets_dropped = 0; |
1103 | struct sk_buff *skb; | |
1104 | ||
1105 | while ((skb = __skb_dequeue(rxq)) != NULL) { | |
3683243b | 1106 | int pull_to = NETFRONT_SKB_CB(skb)->pull_to; |
0d160211 | 1107 | |
093b9c71 JB |
1108 | if (pull_to > skb_headlen(skb)) |
1109 | __pskb_pull_tail(skb, pull_to - skb_headlen(skb)); | |
0d160211 JF |
1110 | |
1111 | /* Ethernet work: Delayed to here as it peeks the header. */ | |
2688fcb7 | 1112 | skb->protocol = eth_type_trans(skb, queue->info->netdev); |
d554f73d | 1113 | skb_reset_network_header(skb); |
0d160211 | 1114 | |
2688fcb7 | 1115 | if (checksum_setup(queue->info->netdev, skb)) { |
e0ce4af9 IC |
1116 | kfree_skb(skb); |
1117 | packets_dropped++; | |
2688fcb7 | 1118 | queue->info->netdev->stats.rx_errors++; |
e0ce4af9 | 1119 | continue; |
0d160211 JF |
1120 | } |
1121 | ||
900e1833 DV |
1122 | u64_stats_update_begin(&rx_stats->syncp); |
1123 | rx_stats->packets++; | |
1124 | rx_stats->bytes += skb->len; | |
1125 | u64_stats_update_end(&rx_stats->syncp); | |
0d160211 JF |
1126 | |
1127 | /* Pass it up. */ | |
2688fcb7 | 1128 | napi_gro_receive(&queue->napi, skb); |
0d160211 JF |
1129 | } |
1130 | ||
1131 | return packets_dropped; | |
1132 | } | |
1133 | ||
bea3348e | 1134 | static int xennet_poll(struct napi_struct *napi, int budget) |
0d160211 | 1135 | { |
2688fcb7 AB |
1136 | struct netfront_queue *queue = container_of(napi, struct netfront_queue, napi); |
1137 | struct net_device *dev = queue->info->netdev; | |
0d160211 JF |
1138 | struct sk_buff *skb; |
1139 | struct netfront_rx_info rinfo; | |
1140 | struct xen_netif_rx_response *rx = &rinfo.rx; | |
1141 | struct xen_netif_extra_info *extras = rinfo.extras; | |
1142 | RING_IDX i, rp; | |
bea3348e | 1143 | int work_done; |
0d160211 JF |
1144 | struct sk_buff_head rxq; |
1145 | struct sk_buff_head errq; | |
1146 | struct sk_buff_head tmpq; | |
0d160211 | 1147 | int err; |
6c5aa6fc | 1148 | bool need_xdp_flush = false; |
0d160211 | 1149 | |
2688fcb7 | 1150 | spin_lock(&queue->rx_lock); |
0d160211 | 1151 | |
0d160211 JF |
1152 | skb_queue_head_init(&rxq); |
1153 | skb_queue_head_init(&errq); | |
1154 | skb_queue_head_init(&tmpq); | |
1155 | ||
2688fcb7 | 1156 | rp = queue->rx.sring->rsp_prod; |
0d160211 JF |
1157 | rmb(); /* Ensure we see queued responses up to 'rp'. */ |
1158 | ||
2688fcb7 | 1159 | i = queue->rx.rsp_cons; |
0d160211 JF |
1160 | work_done = 0; |
1161 | while ((i != rp) && (work_done < budget)) { | |
8446066b | 1162 | RING_COPY_RESPONSE(&queue->rx, i, rx); |
0d160211 JF |
1163 | memset(extras, 0, sizeof(rinfo.extras)); |
1164 | ||
6c5aa6fc DK |
1165 | err = xennet_get_responses(queue, &rinfo, rp, &tmpq, |
1166 | &need_xdp_flush); | |
0d160211 JF |
1167 | |
1168 | if (unlikely(err)) { | |
1169 | err: | |
1170 | while ((skb = __skb_dequeue(&tmpq))) | |
1171 | __skb_queue_tail(&errq, skb); | |
09f75cd7 | 1172 | dev->stats.rx_errors++; |
2688fcb7 | 1173 | i = queue->rx.rsp_cons; |
0d160211 JF |
1174 | continue; |
1175 | } | |
1176 | ||
1177 | skb = __skb_dequeue(&tmpq); | |
1178 | ||
1179 | if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) { | |
1180 | struct xen_netif_extra_info *gso; | |
1181 | gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1]; | |
1182 | ||
1183 | if (unlikely(xennet_set_skb_gso(skb, gso))) { | |
1184 | __skb_queue_head(&tmpq, skb); | |
2688fcb7 | 1185 | queue->rx.rsp_cons += skb_queue_len(&tmpq); |
0d160211 JF |
1186 | goto err; |
1187 | } | |
1188 | } | |
1189 | ||
3683243b IC |
1190 | NETFRONT_SKB_CB(skb)->pull_to = rx->status; |
1191 | if (NETFRONT_SKB_CB(skb)->pull_to > RX_COPY_THRESHOLD) | |
1192 | NETFRONT_SKB_CB(skb)->pull_to = RX_COPY_THRESHOLD; | |
0d160211 | 1193 | |
b54c9d5b | 1194 | skb_frag_off_set(&skb_shinfo(skb)->frags[0], rx->offset); |
3683243b IC |
1195 | skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status); |
1196 | skb->data_len = rx->status; | |
093b9c71 | 1197 | skb->len += rx->status; |
0d160211 | 1198 | |
a761129e | 1199 | if (unlikely(xennet_fill_frags(queue, skb, &tmpq))) |
ad4f15dc | 1200 | goto err; |
0d160211 | 1201 | |
f942dc25 | 1202 | if (rx->flags & XEN_NETRXF_csum_blank) |
0d160211 | 1203 | skb->ip_summed = CHECKSUM_PARTIAL; |
f942dc25 | 1204 | else if (rx->flags & XEN_NETRXF_data_validated) |
0d160211 JF |
1205 | skb->ip_summed = CHECKSUM_UNNECESSARY; |
1206 | ||
1207 | __skb_queue_tail(&rxq, skb); | |
1208 | ||
a761129e | 1209 | i = ++queue->rx.rsp_cons; |
0d160211 JF |
1210 | work_done++; |
1211 | } | |
6c5aa6fc DK |
1212 | if (need_xdp_flush) |
1213 | xdp_do_flush(); | |
0d160211 | 1214 | |
56cfe5d0 | 1215 | __skb_queue_purge(&errq); |
0d160211 | 1216 | |
2688fcb7 | 1217 | work_done -= handle_incoming_queue(queue, &rxq); |
0d160211 | 1218 | |
2688fcb7 | 1219 | xennet_alloc_rx_buffers(queue); |
0d160211 | 1220 | |
0d160211 | 1221 | if (work_done < budget) { |
bea3348e SH |
1222 | int more_to_do = 0; |
1223 | ||
6ad20165 | 1224 | napi_complete_done(napi, work_done); |
0d160211 | 1225 | |
2688fcb7 | 1226 | RING_FINAL_CHECK_FOR_RESPONSES(&queue->rx, more_to_do); |
6a6dc08f DV |
1227 | if (more_to_do) |
1228 | napi_schedule(napi); | |
0d160211 JF |
1229 | } |
1230 | ||
2688fcb7 | 1231 | spin_unlock(&queue->rx_lock); |
0d160211 | 1232 | |
bea3348e | 1233 | return work_done; |
0d160211 JF |
1234 | } |
1235 | ||
1236 | static int xennet_change_mtu(struct net_device *dev, int mtu) | |
1237 | { | |
0c36820e | 1238 | int max = xennet_can_sg(dev) ? XEN_NETIF_MAX_TX_SIZE : ETH_DATA_LEN; |
0d160211 JF |
1239 | |
1240 | if (mtu > max) | |
1241 | return -EINVAL; | |
1242 | dev->mtu = mtu; | |
1243 | return 0; | |
1244 | } | |
1245 | ||
bc1f4470 | 1246 | static void xennet_get_stats64(struct net_device *dev, |
1247 | struct rtnl_link_stats64 *tot) | |
e00f85be | 1248 | { |
1249 | struct netfront_info *np = netdev_priv(dev); | |
1250 | int cpu; | |
1251 | ||
1252 | for_each_possible_cpu(cpu) { | |
900e1833 DV |
1253 | struct netfront_stats *rx_stats = per_cpu_ptr(np->rx_stats, cpu); |
1254 | struct netfront_stats *tx_stats = per_cpu_ptr(np->tx_stats, cpu); | |
e00f85be | 1255 | u64 rx_packets, rx_bytes, tx_packets, tx_bytes; |
1256 | unsigned int start; | |
1257 | ||
1258 | do { | |
900e1833 DV |
1259 | start = u64_stats_fetch_begin_irq(&tx_stats->syncp); |
1260 | tx_packets = tx_stats->packets; | |
1261 | tx_bytes = tx_stats->bytes; | |
1262 | } while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start)); | |
e00f85be | 1263 | |
900e1833 DV |
1264 | do { |
1265 | start = u64_stats_fetch_begin_irq(&rx_stats->syncp); | |
1266 | rx_packets = rx_stats->packets; | |
1267 | rx_bytes = rx_stats->bytes; | |
1268 | } while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start)); | |
e00f85be | 1269 | |
1270 | tot->rx_packets += rx_packets; | |
1271 | tot->tx_packets += tx_packets; | |
1272 | tot->rx_bytes += rx_bytes; | |
1273 | tot->tx_bytes += tx_bytes; | |
1274 | } | |
1275 | ||
1276 | tot->rx_errors = dev->stats.rx_errors; | |
1277 | tot->tx_dropped = dev->stats.tx_dropped; | |
e00f85be | 1278 | } |
1279 | ||
2688fcb7 | 1280 | static void xennet_release_tx_bufs(struct netfront_queue *queue) |
0d160211 JF |
1281 | { |
1282 | struct sk_buff *skb; | |
1283 | int i; | |
1284 | ||
1285 | for (i = 0; i < NET_TX_RING_SIZE; i++) { | |
1286 | /* Skip over entries which are actually freelist references */ | |
2688fcb7 | 1287 | if (skb_entry_is_link(&queue->tx_skbs[i])) |
0d160211 JF |
1288 | continue; |
1289 | ||
2688fcb7 AB |
1290 | skb = queue->tx_skbs[i].skb; |
1291 | get_page(queue->grant_tx_page[i]); | |
1292 | gnttab_end_foreign_access(queue->grant_tx_ref[i], | |
cefe0078 | 1293 | GNTMAP_readonly, |
2688fcb7 AB |
1294 | (unsigned long)page_address(queue->grant_tx_page[i])); |
1295 | queue->grant_tx_page[i] = NULL; | |
1296 | queue->grant_tx_ref[i] = GRANT_INVALID_REF; | |
1297 | add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, i); | |
0d160211 JF |
1298 | dev_kfree_skb_irq(skb); |
1299 | } | |
1300 | } | |
1301 | ||
2688fcb7 | 1302 | static void xennet_release_rx_bufs(struct netfront_queue *queue) |
0d160211 | 1303 | { |
0d160211 JF |
1304 | int id, ref; |
1305 | ||
2688fcb7 | 1306 | spin_lock_bh(&queue->rx_lock); |
0d160211 JF |
1307 | |
1308 | for (id = 0; id < NET_RX_RING_SIZE; id++) { | |
cefe0078 AL |
1309 | struct sk_buff *skb; |
1310 | struct page *page; | |
0d160211 | 1311 | |
2688fcb7 | 1312 | skb = queue->rx_skbs[id]; |
cefe0078 | 1313 | if (!skb) |
0d160211 | 1314 | continue; |
0d160211 | 1315 | |
2688fcb7 | 1316 | ref = queue->grant_rx_ref[id]; |
cefe0078 AL |
1317 | if (ref == GRANT_INVALID_REF) |
1318 | continue; | |
0d160211 | 1319 | |
cefe0078 | 1320 | page = skb_frag_page(&skb_shinfo(skb)->frags[0]); |
0d160211 | 1321 | |
cefe0078 AL |
1322 | /* gnttab_end_foreign_access() needs a page ref until |
1323 | * foreign access is ended (which may be deferred). | |
1324 | */ | |
1325 | get_page(page); | |
1326 | gnttab_end_foreign_access(ref, 0, | |
1327 | (unsigned long)page_address(page)); | |
2688fcb7 | 1328 | queue->grant_rx_ref[id] = GRANT_INVALID_REF; |
0d160211 | 1329 | |
cefe0078 | 1330 | kfree_skb(skb); |
0d160211 JF |
1331 | } |
1332 | ||
2688fcb7 | 1333 | spin_unlock_bh(&queue->rx_lock); |
0d160211 JF |
1334 | } |
1335 | ||
c8f44aff MM |
1336 | static netdev_features_t xennet_fix_features(struct net_device *dev, |
1337 | netdev_features_t features) | |
8f7b01a1 ED |
1338 | { |
1339 | struct netfront_info *np = netdev_priv(dev); | |
8f7b01a1 | 1340 | |
2890ea5c JG |
1341 | if (features & NETIF_F_SG && |
1342 | !xenbus_read_unsigned(np->xbdev->otherend, "feature-sg", 0)) | |
1343 | features &= ~NETIF_F_SG; | |
8f7b01a1 | 1344 | |
2890ea5c JG |
1345 | if (features & NETIF_F_IPV6_CSUM && |
1346 | !xenbus_read_unsigned(np->xbdev->otherend, | |
1347 | "feature-ipv6-csum-offload", 0)) | |
1348 | features &= ~NETIF_F_IPV6_CSUM; | |
8f7b01a1 | 1349 | |
2890ea5c JG |
1350 | if (features & NETIF_F_TSO && |
1351 | !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv4", 0)) | |
1352 | features &= ~NETIF_F_TSO; | |
8f7b01a1 | 1353 | |
2890ea5c JG |
1354 | if (features & NETIF_F_TSO6 && |
1355 | !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv6", 0)) | |
1356 | features &= ~NETIF_F_TSO6; | |
2c0057de | 1357 | |
8f7b01a1 ED |
1358 | return features; |
1359 | } | |
1360 | ||
c8f44aff MM |
1361 | static int xennet_set_features(struct net_device *dev, |
1362 | netdev_features_t features) | |
8f7b01a1 ED |
1363 | { |
1364 | if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) { | |
1365 | netdev_info(dev, "Reducing MTU because no SG offload"); | |
1366 | dev->mtu = ETH_DATA_LEN; | |
1367 | } | |
1368 | ||
1369 | return 0; | |
1370 | } | |
1371 | ||
d634bf2c | 1372 | static irqreturn_t xennet_tx_interrupt(int irq, void *dev_id) |
cf66f9d4 | 1373 | { |
2688fcb7 | 1374 | struct netfront_queue *queue = dev_id; |
cf66f9d4 KRW |
1375 | unsigned long flags; |
1376 | ||
2688fcb7 AB |
1377 | spin_lock_irqsave(&queue->tx_lock, flags); |
1378 | xennet_tx_buf_gc(queue); | |
1379 | spin_unlock_irqrestore(&queue->tx_lock, flags); | |
cf66f9d4 | 1380 | |
d634bf2c WL |
1381 | return IRQ_HANDLED; |
1382 | } | |
1383 | ||
1384 | static irqreturn_t xennet_rx_interrupt(int irq, void *dev_id) | |
1385 | { | |
2688fcb7 AB |
1386 | struct netfront_queue *queue = dev_id; |
1387 | struct net_device *dev = queue->info->netdev; | |
d634bf2c WL |
1388 | |
1389 | if (likely(netif_carrier_ok(dev) && | |
2688fcb7 | 1390 | RING_HAS_UNCONSUMED_RESPONSES(&queue->rx))) |
76541869 | 1391 | napi_schedule(&queue->napi); |
cf66f9d4 | 1392 | |
d634bf2c WL |
1393 | return IRQ_HANDLED; |
1394 | } | |
cf66f9d4 | 1395 | |
d634bf2c WL |
1396 | static irqreturn_t xennet_interrupt(int irq, void *dev_id) |
1397 | { | |
1398 | xennet_tx_interrupt(irq, dev_id); | |
1399 | xennet_rx_interrupt(irq, dev_id); | |
cf66f9d4 KRW |
1400 | return IRQ_HANDLED; |
1401 | } | |
1402 | ||
1403 | #ifdef CONFIG_NET_POLL_CONTROLLER | |
1404 | static void xennet_poll_controller(struct net_device *dev) | |
1405 | { | |
2688fcb7 AB |
1406 | /* Poll each queue */ |
1407 | struct netfront_info *info = netdev_priv(dev); | |
1408 | unsigned int num_queues = dev->real_num_tx_queues; | |
1409 | unsigned int i; | |
1410 | for (i = 0; i < num_queues; ++i) | |
1411 | xennet_interrupt(0, &info->queues[i]); | |
cf66f9d4 KRW |
1412 | } |
1413 | #endif | |
1414 | ||
6c5aa6fc DK |
1415 | #define NETBACK_XDP_HEADROOM_DISABLE 0 |
1416 | #define NETBACK_XDP_HEADROOM_ENABLE 1 | |
1417 | ||
1418 | static int talk_to_netback_xdp(struct netfront_info *np, int xdp) | |
1419 | { | |
1420 | int err; | |
1421 | unsigned short headroom; | |
1422 | ||
1423 | headroom = xdp ? XDP_PACKET_HEADROOM : 0; | |
1424 | err = xenbus_printf(XBT_NIL, np->xbdev->nodename, | |
1425 | "xdp-headroom", "%hu", | |
1426 | headroom); | |
1427 | if (err) | |
1428 | pr_warn("Error writing xdp-headroom\n"); | |
1429 | ||
1430 | return err; | |
1431 | } | |
1432 | ||
1433 | static int xennet_xdp_set(struct net_device *dev, struct bpf_prog *prog, | |
1434 | struct netlink_ext_ack *extack) | |
1435 | { | |
1436 | unsigned long max_mtu = XEN_PAGE_SIZE - XDP_PACKET_HEADROOM; | |
1437 | struct netfront_info *np = netdev_priv(dev); | |
1438 | struct bpf_prog *old_prog; | |
1439 | unsigned int i, err; | |
1440 | ||
1441 | if (dev->mtu > max_mtu) { | |
1442 | netdev_warn(dev, "XDP requires MTU less than %lu\n", max_mtu); | |
1443 | return -EINVAL; | |
1444 | } | |
1445 | ||
1446 | if (!np->netback_has_xdp_headroom) | |
1447 | return 0; | |
1448 | ||
1449 | xenbus_switch_state(np->xbdev, XenbusStateReconfiguring); | |
1450 | ||
1451 | err = talk_to_netback_xdp(np, prog ? NETBACK_XDP_HEADROOM_ENABLE : | |
1452 | NETBACK_XDP_HEADROOM_DISABLE); | |
1453 | if (err) | |
1454 | return err; | |
1455 | ||
1456 | /* avoid the race with XDP headroom adjustment */ | |
1457 | wait_event(module_wq, | |
1458 | xenbus_read_driver_state(np->xbdev->otherend) == | |
1459 | XenbusStateReconfigured); | |
1460 | np->netfront_xdp_enabled = true; | |
1461 | ||
1462 | old_prog = rtnl_dereference(np->queues[0].xdp_prog); | |
1463 | ||
1464 | if (prog) | |
1465 | bpf_prog_add(prog, dev->real_num_tx_queues); | |
1466 | ||
1467 | for (i = 0; i < dev->real_num_tx_queues; ++i) | |
1468 | rcu_assign_pointer(np->queues[i].xdp_prog, prog); | |
1469 | ||
1470 | if (old_prog) | |
1471 | for (i = 0; i < dev->real_num_tx_queues; ++i) | |
1472 | bpf_prog_put(old_prog); | |
1473 | ||
1474 | xenbus_switch_state(np->xbdev, XenbusStateConnected); | |
1475 | ||
1476 | return 0; | |
1477 | } | |
1478 | ||
6c5aa6fc DK |
1479 | static int xennet_xdp(struct net_device *dev, struct netdev_bpf *xdp) |
1480 | { | |
1481 | switch (xdp->command) { | |
1482 | case XDP_SETUP_PROG: | |
1483 | return xennet_xdp_set(dev, xdp->prog, xdp->extack); | |
6c5aa6fc DK |
1484 | default: |
1485 | return -EINVAL; | |
1486 | } | |
1487 | } | |
1488 | ||
0a0b9d2e SH |
1489 | static const struct net_device_ops xennet_netdev_ops = { |
1490 | .ndo_open = xennet_open, | |
0a0b9d2e SH |
1491 | .ndo_stop = xennet_close, |
1492 | .ndo_start_xmit = xennet_start_xmit, | |
1493 | .ndo_change_mtu = xennet_change_mtu, | |
e00f85be | 1494 | .ndo_get_stats64 = xennet_get_stats64, |
0a0b9d2e SH |
1495 | .ndo_set_mac_address = eth_mac_addr, |
1496 | .ndo_validate_addr = eth_validate_addr, | |
fb507934 MM |
1497 | .ndo_fix_features = xennet_fix_features, |
1498 | .ndo_set_features = xennet_set_features, | |
2688fcb7 | 1499 | .ndo_select_queue = xennet_select_queue, |
6c5aa6fc DK |
1500 | .ndo_bpf = xennet_xdp, |
1501 | .ndo_xdp_xmit = xennet_xdp_xmit, | |
cf66f9d4 KRW |
1502 | #ifdef CONFIG_NET_POLL_CONTROLLER |
1503 | .ndo_poll_controller = xennet_poll_controller, | |
1504 | #endif | |
0a0b9d2e SH |
1505 | }; |
1506 | ||
900e1833 DV |
1507 | static void xennet_free_netdev(struct net_device *netdev) |
1508 | { | |
1509 | struct netfront_info *np = netdev_priv(netdev); | |
1510 | ||
1511 | free_percpu(np->rx_stats); | |
1512 | free_percpu(np->tx_stats); | |
1513 | free_netdev(netdev); | |
1514 | } | |
1515 | ||
8e0e46bb | 1516 | static struct net_device *xennet_create_dev(struct xenbus_device *dev) |
0d160211 | 1517 | { |
2688fcb7 | 1518 | int err; |
0d160211 JF |
1519 | struct net_device *netdev; |
1520 | struct netfront_info *np; | |
1521 | ||
50ee6061 | 1522 | netdev = alloc_etherdev_mq(sizeof(struct netfront_info), xennet_max_queues); |
41de8d4c | 1523 | if (!netdev) |
0d160211 | 1524 | return ERR_PTR(-ENOMEM); |
0d160211 JF |
1525 | |
1526 | np = netdev_priv(netdev); | |
1527 | np->xbdev = dev; | |
1528 | ||
2688fcb7 | 1529 | np->queues = NULL; |
0d160211 | 1530 | |
e00f85be | 1531 | err = -ENOMEM; |
900e1833 DV |
1532 | np->rx_stats = netdev_alloc_pcpu_stats(struct netfront_stats); |
1533 | if (np->rx_stats == NULL) | |
1534 | goto exit; | |
1535 | np->tx_stats = netdev_alloc_pcpu_stats(struct netfront_stats); | |
1536 | if (np->tx_stats == NULL) | |
e00f85be | 1537 | goto exit; |
1538 | ||
0a0b9d2e SH |
1539 | netdev->netdev_ops = &xennet_netdev_ops; |
1540 | ||
fb507934 MM |
1541 | netdev->features = NETIF_F_IP_CSUM | NETIF_F_RXCSUM | |
1542 | NETIF_F_GSO_ROBUST; | |
2c0057de PD |
1543 | netdev->hw_features = NETIF_F_SG | |
1544 | NETIF_F_IPV6_CSUM | | |
1545 | NETIF_F_TSO | NETIF_F_TSO6; | |
0d160211 | 1546 | |
fc3e5941 IC |
1547 | /* |
1548 | * Assume that all hw features are available for now. This set | |
1549 | * will be adjusted by the call to netdev_update_features() in | |
1550 | * xennet_connect() which is the earliest point where we can | |
1551 | * negotiate with the backend regarding supported features. | |
1552 | */ | |
1553 | netdev->features |= netdev->hw_features; | |
1554 | ||
7ad24ea4 | 1555 | netdev->ethtool_ops = &xennet_ethtool_ops; |
e1043a4b | 1556 | netdev->min_mtu = ETH_MIN_MTU; |
d0c2c997 | 1557 | netdev->max_mtu = XEN_NETIF_MAX_TX_SIZE; |
0d160211 JF |
1558 | SET_NETDEV_DEV(netdev, &dev->dev); |
1559 | ||
1560 | np->netdev = netdev; | |
6c5aa6fc | 1561 | np->netfront_xdp_enabled = false; |
0d160211 JF |
1562 | |
1563 | netif_carrier_off(netdev); | |
1564 | ||
c2c63310 AR |
1565 | do { |
1566 | xenbus_switch_state(dev, XenbusStateInitialising); | |
1567 | err = wait_event_timeout(module_wq, | |
1568 | xenbus_read_driver_state(dev->otherend) != | |
1569 | XenbusStateClosed && | |
1570 | xenbus_read_driver_state(dev->otherend) != | |
1571 | XenbusStateUnknown, XENNET_TIMEOUT); | |
1572 | } while (!err); | |
1573 | ||
0d160211 JF |
1574 | return netdev; |
1575 | ||
0d160211 | 1576 | exit: |
900e1833 | 1577 | xennet_free_netdev(netdev); |
0d160211 JF |
1578 | return ERR_PTR(err); |
1579 | } | |
1580 | ||
80708602 | 1581 | /* |
0d160211 JF |
1582 | * Entry point to this code when a new device is created. Allocate the basic |
1583 | * structures and the ring buffers for communication with the backend, and | |
1584 | * inform the backend of the appropriate details for those. | |
1585 | */ | |
8e0e46bb | 1586 | static int netfront_probe(struct xenbus_device *dev, |
1dd06ae8 | 1587 | const struct xenbus_device_id *id) |
0d160211 JF |
1588 | { |
1589 | int err; | |
1590 | struct net_device *netdev; | |
1591 | struct netfront_info *info; | |
1592 | ||
1593 | netdev = xennet_create_dev(dev); | |
1594 | if (IS_ERR(netdev)) { | |
1595 | err = PTR_ERR(netdev); | |
1596 | xenbus_dev_fatal(dev, err, "creating netdev"); | |
1597 | return err; | |
1598 | } | |
1599 | ||
1600 | info = netdev_priv(netdev); | |
1b713e00 | 1601 | dev_set_drvdata(&dev->dev, info); |
27b917e5 TI |
1602 | #ifdef CONFIG_SYSFS |
1603 | info->netdev->sysfs_groups[0] = &xennet_dev_group; | |
1604 | #endif | |
0d160211 | 1605 | |
0d160211 | 1606 | return 0; |
0d160211 JF |
1607 | } |
1608 | ||
1609 | static void xennet_end_access(int ref, void *page) | |
1610 | { | |
1611 | /* This frees the page as a side-effect */ | |
1612 | if (ref != GRANT_INVALID_REF) | |
1613 | gnttab_end_foreign_access(ref, 0, (unsigned long)page); | |
1614 | } | |
1615 | ||
1616 | static void xennet_disconnect_backend(struct netfront_info *info) | |
1617 | { | |
2688fcb7 | 1618 | unsigned int i = 0; |
2688fcb7 AB |
1619 | unsigned int num_queues = info->netdev->real_num_tx_queues; |
1620 | ||
f9feb1e6 DV |
1621 | netif_carrier_off(info->netdev); |
1622 | ||
9a873c71 | 1623 | for (i = 0; i < num_queues && info->queues; ++i) { |
76541869 DV |
1624 | struct netfront_queue *queue = &info->queues[i]; |
1625 | ||
74470954 BO |
1626 | del_timer_sync(&queue->rx_refill_timer); |
1627 | ||
2688fcb7 AB |
1628 | if (queue->tx_irq && (queue->tx_irq == queue->rx_irq)) |
1629 | unbind_from_irqhandler(queue->tx_irq, queue); | |
1630 | if (queue->tx_irq && (queue->tx_irq != queue->rx_irq)) { | |
1631 | unbind_from_irqhandler(queue->tx_irq, queue); | |
1632 | unbind_from_irqhandler(queue->rx_irq, queue); | |
1633 | } | |
1634 | queue->tx_evtchn = queue->rx_evtchn = 0; | |
1635 | queue->tx_irq = queue->rx_irq = 0; | |
0d160211 | 1636 | |
274b0455 CW |
1637 | if (netif_running(info->netdev)) |
1638 | napi_synchronize(&queue->napi); | |
f9feb1e6 | 1639 | |
a5b5dc3c DV |
1640 | xennet_release_tx_bufs(queue); |
1641 | xennet_release_rx_bufs(queue); | |
1642 | gnttab_free_grant_references(queue->gref_tx_head); | |
1643 | gnttab_free_grant_references(queue->gref_rx_head); | |
1644 | ||
2688fcb7 AB |
1645 | /* End access and free the pages */ |
1646 | xennet_end_access(queue->tx_ring_ref, queue->tx.sring); | |
1647 | xennet_end_access(queue->rx_ring_ref, queue->rx.sring); | |
0d160211 | 1648 | |
2688fcb7 AB |
1649 | queue->tx_ring_ref = GRANT_INVALID_REF; |
1650 | queue->rx_ring_ref = GRANT_INVALID_REF; | |
1651 | queue->tx.sring = NULL; | |
1652 | queue->rx.sring = NULL; | |
6c5aa6fc DK |
1653 | |
1654 | page_pool_destroy(queue->page_pool); | |
2688fcb7 | 1655 | } |
0d160211 JF |
1656 | } |
1657 | ||
80708602 | 1658 | /* |
0d160211 JF |
1659 | * We are reconnecting to the backend, due to a suspend/resume, or a backend |
1660 | * driver restart. We tear down our netif structure and recreate it, but | |
1661 | * leave the device-layer structures intact so that this is transparent to the | |
1662 | * rest of the kernel. | |
1663 | */ | |
1664 | static int netfront_resume(struct xenbus_device *dev) | |
1665 | { | |
1b713e00 | 1666 | struct netfront_info *info = dev_get_drvdata(&dev->dev); |
0d160211 JF |
1667 | |
1668 | dev_dbg(&dev->dev, "%s\n", dev->nodename); | |
1669 | ||
1670 | xennet_disconnect_backend(info); | |
1671 | return 0; | |
1672 | } | |
1673 | ||
1674 | static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[]) | |
1675 | { | |
1676 | char *s, *e, *macstr; | |
1677 | int i; | |
1678 | ||
1679 | macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL); | |
1680 | if (IS_ERR(macstr)) | |
1681 | return PTR_ERR(macstr); | |
1682 | ||
1683 | for (i = 0; i < ETH_ALEN; i++) { | |
1684 | mac[i] = simple_strtoul(s, &e, 16); | |
1685 | if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) { | |
1686 | kfree(macstr); | |
1687 | return -ENOENT; | |
1688 | } | |
1689 | s = e+1; | |
1690 | } | |
1691 | ||
1692 | kfree(macstr); | |
1693 | return 0; | |
1694 | } | |
1695 | ||
2688fcb7 | 1696 | static int setup_netfront_single(struct netfront_queue *queue) |
d634bf2c WL |
1697 | { |
1698 | int err; | |
1699 | ||
2688fcb7 | 1700 | err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn); |
d634bf2c WL |
1701 | if (err < 0) |
1702 | goto fail; | |
1703 | ||
2688fcb7 | 1704 | err = bind_evtchn_to_irqhandler(queue->tx_evtchn, |
d634bf2c | 1705 | xennet_interrupt, |
2688fcb7 | 1706 | 0, queue->info->netdev->name, queue); |
d634bf2c WL |
1707 | if (err < 0) |
1708 | goto bind_fail; | |
2688fcb7 AB |
1709 | queue->rx_evtchn = queue->tx_evtchn; |
1710 | queue->rx_irq = queue->tx_irq = err; | |
d634bf2c WL |
1711 | |
1712 | return 0; | |
1713 | ||
1714 | bind_fail: | |
2688fcb7 AB |
1715 | xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn); |
1716 | queue->tx_evtchn = 0; | |
d634bf2c WL |
1717 | fail: |
1718 | return err; | |
1719 | } | |
1720 | ||
2688fcb7 | 1721 | static int setup_netfront_split(struct netfront_queue *queue) |
d634bf2c WL |
1722 | { |
1723 | int err; | |
1724 | ||
2688fcb7 | 1725 | err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn); |
d634bf2c WL |
1726 | if (err < 0) |
1727 | goto fail; | |
2688fcb7 | 1728 | err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->rx_evtchn); |
d634bf2c WL |
1729 | if (err < 0) |
1730 | goto alloc_rx_evtchn_fail; | |
1731 | ||
2688fcb7 AB |
1732 | snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name), |
1733 | "%s-tx", queue->name); | |
1734 | err = bind_evtchn_to_irqhandler(queue->tx_evtchn, | |
d634bf2c | 1735 | xennet_tx_interrupt, |
2688fcb7 | 1736 | 0, queue->tx_irq_name, queue); |
d634bf2c WL |
1737 | if (err < 0) |
1738 | goto bind_tx_fail; | |
2688fcb7 | 1739 | queue->tx_irq = err; |
d634bf2c | 1740 | |
2688fcb7 AB |
1741 | snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name), |
1742 | "%s-rx", queue->name); | |
1743 | err = bind_evtchn_to_irqhandler(queue->rx_evtchn, | |
d634bf2c | 1744 | xennet_rx_interrupt, |
2688fcb7 | 1745 | 0, queue->rx_irq_name, queue); |
d634bf2c WL |
1746 | if (err < 0) |
1747 | goto bind_rx_fail; | |
2688fcb7 | 1748 | queue->rx_irq = err; |
d634bf2c WL |
1749 | |
1750 | return 0; | |
1751 | ||
1752 | bind_rx_fail: | |
2688fcb7 AB |
1753 | unbind_from_irqhandler(queue->tx_irq, queue); |
1754 | queue->tx_irq = 0; | |
d634bf2c | 1755 | bind_tx_fail: |
2688fcb7 AB |
1756 | xenbus_free_evtchn(queue->info->xbdev, queue->rx_evtchn); |
1757 | queue->rx_evtchn = 0; | |
d634bf2c | 1758 | alloc_rx_evtchn_fail: |
2688fcb7 AB |
1759 | xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn); |
1760 | queue->tx_evtchn = 0; | |
d634bf2c WL |
1761 | fail: |
1762 | return err; | |
1763 | } | |
1764 | ||
2688fcb7 AB |
1765 | static int setup_netfront(struct xenbus_device *dev, |
1766 | struct netfront_queue *queue, unsigned int feature_split_evtchn) | |
0d160211 JF |
1767 | { |
1768 | struct xen_netif_tx_sring *txs; | |
1769 | struct xen_netif_rx_sring *rxs; | |
ccc9d90a | 1770 | grant_ref_t gref; |
0d160211 | 1771 | int err; |
0d160211 | 1772 | |
2688fcb7 AB |
1773 | queue->tx_ring_ref = GRANT_INVALID_REF; |
1774 | queue->rx_ring_ref = GRANT_INVALID_REF; | |
1775 | queue->rx.sring = NULL; | |
1776 | queue->tx.sring = NULL; | |
0d160211 | 1777 | |
a144ff09 | 1778 | txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH); |
0d160211 JF |
1779 | if (!txs) { |
1780 | err = -ENOMEM; | |
1781 | xenbus_dev_fatal(dev, err, "allocating tx ring page"); | |
1782 | goto fail; | |
1783 | } | |
1784 | SHARED_RING_INIT(txs); | |
30c5d7f0 | 1785 | FRONT_RING_INIT(&queue->tx, txs, XEN_PAGE_SIZE); |
0d160211 | 1786 | |
ccc9d90a | 1787 | err = xenbus_grant_ring(dev, txs, 1, &gref); |
1ca2983a WL |
1788 | if (err < 0) |
1789 | goto grant_tx_ring_fail; | |
ccc9d90a | 1790 | queue->tx_ring_ref = gref; |
0d160211 | 1791 | |
a144ff09 | 1792 | rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH); |
0d160211 JF |
1793 | if (!rxs) { |
1794 | err = -ENOMEM; | |
1795 | xenbus_dev_fatal(dev, err, "allocating rx ring page"); | |
1ca2983a | 1796 | goto alloc_rx_ring_fail; |
0d160211 JF |
1797 | } |
1798 | SHARED_RING_INIT(rxs); | |
30c5d7f0 | 1799 | FRONT_RING_INIT(&queue->rx, rxs, XEN_PAGE_SIZE); |
0d160211 | 1800 | |
ccc9d90a | 1801 | err = xenbus_grant_ring(dev, rxs, 1, &gref); |
1ca2983a WL |
1802 | if (err < 0) |
1803 | goto grant_rx_ring_fail; | |
ccc9d90a | 1804 | queue->rx_ring_ref = gref; |
0d160211 | 1805 | |
d634bf2c | 1806 | if (feature_split_evtchn) |
2688fcb7 | 1807 | err = setup_netfront_split(queue); |
d634bf2c WL |
1808 | /* setup single event channel if |
1809 | * a) feature-split-event-channels == 0 | |
1810 | * b) feature-split-event-channels == 1 but failed to setup | |
1811 | */ | |
e93fac3b | 1812 | if (!feature_split_evtchn || err) |
2688fcb7 | 1813 | err = setup_netfront_single(queue); |
d634bf2c | 1814 | |
0d160211 | 1815 | if (err) |
1ca2983a | 1816 | goto alloc_evtchn_fail; |
0d160211 | 1817 | |
0d160211 JF |
1818 | return 0; |
1819 | ||
1ca2983a WL |
1820 | /* If we fail to setup netfront, it is safe to just revoke access to |
1821 | * granted pages because backend is not accessing it at this point. | |
1822 | */ | |
1ca2983a | 1823 | alloc_evtchn_fail: |
2688fcb7 | 1824 | gnttab_end_foreign_access_ref(queue->rx_ring_ref, 0); |
1ca2983a WL |
1825 | grant_rx_ring_fail: |
1826 | free_page((unsigned long)rxs); | |
1827 | alloc_rx_ring_fail: | |
2688fcb7 | 1828 | gnttab_end_foreign_access_ref(queue->tx_ring_ref, 0); |
1ca2983a WL |
1829 | grant_tx_ring_fail: |
1830 | free_page((unsigned long)txs); | |
1831 | fail: | |
0d160211 JF |
1832 | return err; |
1833 | } | |
1834 | ||
2688fcb7 AB |
1835 | /* Queue-specific initialisation |
1836 | * This used to be done in xennet_create_dev() but must now | |
1837 | * be run per-queue. | |
1838 | */ | |
1839 | static int xennet_init_queue(struct netfront_queue *queue) | |
1840 | { | |
1841 | unsigned short i; | |
1842 | int err = 0; | |
21f2706b | 1843 | char *devid; |
2688fcb7 AB |
1844 | |
1845 | spin_lock_init(&queue->tx_lock); | |
1846 | spin_lock_init(&queue->rx_lock); | |
1847 | ||
e99e88a9 | 1848 | timer_setup(&queue->rx_refill_timer, rx_refill_timeout, 0); |
2688fcb7 | 1849 | |
21f2706b XL |
1850 | devid = strrchr(queue->info->xbdev->nodename, '/') + 1; |
1851 | snprintf(queue->name, sizeof(queue->name), "vif%s-q%u", | |
1852 | devid, queue->id); | |
8b715010 | 1853 | |
2688fcb7 AB |
1854 | /* Initialise tx_skbs as a free chain containing every entry. */ |
1855 | queue->tx_skb_freelist = 0; | |
1856 | for (i = 0; i < NET_TX_RING_SIZE; i++) { | |
1857 | skb_entry_set_link(&queue->tx_skbs[i], i+1); | |
1858 | queue->grant_tx_ref[i] = GRANT_INVALID_REF; | |
1859 | queue->grant_tx_page[i] = NULL; | |
1860 | } | |
1861 | ||
1862 | /* Clear out rx_skbs */ | |
1863 | for (i = 0; i < NET_RX_RING_SIZE; i++) { | |
1864 | queue->rx_skbs[i] = NULL; | |
1865 | queue->grant_rx_ref[i] = GRANT_INVALID_REF; | |
1866 | } | |
1867 | ||
1868 | /* A grant for every tx ring slot */ | |
1f3c2eba | 1869 | if (gnttab_alloc_grant_references(NET_TX_RING_SIZE, |
2688fcb7 AB |
1870 | &queue->gref_tx_head) < 0) { |
1871 | pr_alert("can't alloc tx grant refs\n"); | |
1872 | err = -ENOMEM; | |
1873 | goto exit; | |
1874 | } | |
1875 | ||
1876 | /* A grant for every rx ring slot */ | |
1f3c2eba | 1877 | if (gnttab_alloc_grant_references(NET_RX_RING_SIZE, |
2688fcb7 AB |
1878 | &queue->gref_rx_head) < 0) { |
1879 | pr_alert("can't alloc rx grant refs\n"); | |
1880 | err = -ENOMEM; | |
1881 | goto exit_free_tx; | |
1882 | } | |
1883 | ||
2688fcb7 AB |
1884 | return 0; |
1885 | ||
1886 | exit_free_tx: | |
1887 | gnttab_free_grant_references(queue->gref_tx_head); | |
1888 | exit: | |
1889 | return err; | |
1890 | } | |
1891 | ||
50ee6061 AB |
1892 | static int write_queue_xenstore_keys(struct netfront_queue *queue, |
1893 | struct xenbus_transaction *xbt, int write_hierarchical) | |
1894 | { | |
1895 | /* Write the queue-specific keys into XenStore in the traditional | |
1896 | * way for a single queue, or in a queue subkeys for multiple | |
1897 | * queues. | |
1898 | */ | |
1899 | struct xenbus_device *dev = queue->info->xbdev; | |
1900 | int err; | |
1901 | const char *message; | |
1902 | char *path; | |
1903 | size_t pathsize; | |
1904 | ||
1905 | /* Choose the correct place to write the keys */ | |
1906 | if (write_hierarchical) { | |
1907 | pathsize = strlen(dev->nodename) + 10; | |
1908 | path = kzalloc(pathsize, GFP_KERNEL); | |
1909 | if (!path) { | |
1910 | err = -ENOMEM; | |
1911 | message = "out of memory while writing ring references"; | |
1912 | goto error; | |
1913 | } | |
1914 | snprintf(path, pathsize, "%s/queue-%u", | |
1915 | dev->nodename, queue->id); | |
1916 | } else { | |
1917 | path = (char *)dev->nodename; | |
1918 | } | |
1919 | ||
1920 | /* Write ring references */ | |
1921 | err = xenbus_printf(*xbt, path, "tx-ring-ref", "%u", | |
1922 | queue->tx_ring_ref); | |
1923 | if (err) { | |
1924 | message = "writing tx-ring-ref"; | |
1925 | goto error; | |
1926 | } | |
1927 | ||
1928 | err = xenbus_printf(*xbt, path, "rx-ring-ref", "%u", | |
1929 | queue->rx_ring_ref); | |
1930 | if (err) { | |
1931 | message = "writing rx-ring-ref"; | |
1932 | goto error; | |
1933 | } | |
1934 | ||
1935 | /* Write event channels; taking into account both shared | |
1936 | * and split event channel scenarios. | |
1937 | */ | |
1938 | if (queue->tx_evtchn == queue->rx_evtchn) { | |
1939 | /* Shared event channel */ | |
1940 | err = xenbus_printf(*xbt, path, | |
1941 | "event-channel", "%u", queue->tx_evtchn); | |
1942 | if (err) { | |
1943 | message = "writing event-channel"; | |
1944 | goto error; | |
1945 | } | |
1946 | } else { | |
1947 | /* Split event channels */ | |
1948 | err = xenbus_printf(*xbt, path, | |
1949 | "event-channel-tx", "%u", queue->tx_evtchn); | |
1950 | if (err) { | |
1951 | message = "writing event-channel-tx"; | |
1952 | goto error; | |
1953 | } | |
1954 | ||
1955 | err = xenbus_printf(*xbt, path, | |
1956 | "event-channel-rx", "%u", queue->rx_evtchn); | |
1957 | if (err) { | |
1958 | message = "writing event-channel-rx"; | |
1959 | goto error; | |
1960 | } | |
1961 | } | |
1962 | ||
1963 | if (write_hierarchical) | |
1964 | kfree(path); | |
1965 | return 0; | |
1966 | ||
1967 | error: | |
1968 | if (write_hierarchical) | |
1969 | kfree(path); | |
1970 | xenbus_dev_fatal(dev, err, "%s", message); | |
1971 | return err; | |
1972 | } | |
1973 | ||
ce58725f DV |
1974 | static void xennet_destroy_queues(struct netfront_info *info) |
1975 | { | |
1976 | unsigned int i; | |
1977 | ||
ce58725f DV |
1978 | for (i = 0; i < info->netdev->real_num_tx_queues; i++) { |
1979 | struct netfront_queue *queue = &info->queues[i]; | |
1980 | ||
1981 | if (netif_running(info->netdev)) | |
1982 | napi_disable(&queue->napi); | |
1983 | netif_napi_del(&queue->napi); | |
1984 | } | |
1985 | ||
ce58725f DV |
1986 | kfree(info->queues); |
1987 | info->queues = NULL; | |
1988 | } | |
1989 | ||
6c5aa6fc DK |
1990 | |
1991 | ||
1992 | static int xennet_create_page_pool(struct netfront_queue *queue) | |
1993 | { | |
1994 | int err; | |
1995 | struct page_pool_params pp_params = { | |
1996 | .order = 0, | |
1997 | .flags = 0, | |
1998 | .pool_size = NET_RX_RING_SIZE, | |
1999 | .nid = NUMA_NO_NODE, | |
2000 | .dev = &queue->info->netdev->dev, | |
2001 | .offset = XDP_PACKET_HEADROOM, | |
2002 | .max_len = XEN_PAGE_SIZE - XDP_PACKET_HEADROOM, | |
2003 | }; | |
2004 | ||
2005 | queue->page_pool = page_pool_create(&pp_params); | |
2006 | if (IS_ERR(queue->page_pool)) { | |
2007 | err = PTR_ERR(queue->page_pool); | |
2008 | queue->page_pool = NULL; | |
2009 | return err; | |
2010 | } | |
2011 | ||
2012 | err = xdp_rxq_info_reg(&queue->xdp_rxq, queue->info->netdev, | |
b02e5a0e | 2013 | queue->id, 0); |
6c5aa6fc DK |
2014 | if (err) { |
2015 | netdev_err(queue->info->netdev, "xdp_rxq_info_reg failed\n"); | |
2016 | goto err_free_pp; | |
2017 | } | |
2018 | ||
2019 | err = xdp_rxq_info_reg_mem_model(&queue->xdp_rxq, | |
2020 | MEM_TYPE_PAGE_POOL, queue->page_pool); | |
2021 | if (err) { | |
2022 | netdev_err(queue->info->netdev, "xdp_rxq_info_reg_mem_model failed\n"); | |
2023 | goto err_unregister_rxq; | |
2024 | } | |
2025 | return 0; | |
2026 | ||
2027 | err_unregister_rxq: | |
2028 | xdp_rxq_info_unreg(&queue->xdp_rxq); | |
2029 | err_free_pp: | |
2030 | page_pool_destroy(queue->page_pool); | |
2031 | queue->page_pool = NULL; | |
2032 | return err; | |
2033 | } | |
2034 | ||
ce58725f | 2035 | static int xennet_create_queues(struct netfront_info *info, |
ca88ea12 | 2036 | unsigned int *num_queues) |
ce58725f DV |
2037 | { |
2038 | unsigned int i; | |
2039 | int ret; | |
2040 | ||
ca88ea12 | 2041 | info->queues = kcalloc(*num_queues, sizeof(struct netfront_queue), |
ce58725f DV |
2042 | GFP_KERNEL); |
2043 | if (!info->queues) | |
2044 | return -ENOMEM; | |
2045 | ||
ca88ea12 | 2046 | for (i = 0; i < *num_queues; i++) { |
ce58725f DV |
2047 | struct netfront_queue *queue = &info->queues[i]; |
2048 | ||
2049 | queue->id = i; | |
2050 | queue->info = info; | |
2051 | ||
2052 | ret = xennet_init_queue(queue); | |
2053 | if (ret < 0) { | |
f599c64f | 2054 | dev_warn(&info->xbdev->dev, |
69cb8524 | 2055 | "only created %d queues\n", i); |
ca88ea12 | 2056 | *num_queues = i; |
ce58725f DV |
2057 | break; |
2058 | } | |
2059 | ||
6c5aa6fc DK |
2060 | /* use page pool recycling instead of buddy allocator */ |
2061 | ret = xennet_create_page_pool(queue); | |
2062 | if (ret < 0) { | |
2063 | dev_err(&info->xbdev->dev, "can't allocate page pool\n"); | |
2064 | *num_queues = i; | |
2065 | return ret; | |
2066 | } | |
2067 | ||
ce58725f DV |
2068 | netif_napi_add(queue->info->netdev, &queue->napi, |
2069 | xennet_poll, 64); | |
2070 | if (netif_running(info->netdev)) | |
2071 | napi_enable(&queue->napi); | |
2072 | } | |
2073 | ||
ca88ea12 | 2074 | netif_set_real_num_tx_queues(info->netdev, *num_queues); |
ce58725f | 2075 | |
ca88ea12 | 2076 | if (*num_queues == 0) { |
f599c64f | 2077 | dev_err(&info->xbdev->dev, "no queues\n"); |
ce58725f DV |
2078 | return -EINVAL; |
2079 | } | |
2080 | return 0; | |
2081 | } | |
2082 | ||
0d160211 | 2083 | /* Common code used when first setting up, and when resuming. */ |
f502bf2b | 2084 | static int talk_to_netback(struct xenbus_device *dev, |
0d160211 JF |
2085 | struct netfront_info *info) |
2086 | { | |
2087 | const char *message; | |
2088 | struct xenbus_transaction xbt; | |
2089 | int err; | |
2688fcb7 AB |
2090 | unsigned int feature_split_evtchn; |
2091 | unsigned int i = 0; | |
50ee6061 | 2092 | unsigned int max_queues = 0; |
2688fcb7 AB |
2093 | struct netfront_queue *queue = NULL; |
2094 | unsigned int num_queues = 1; | |
0d160211 | 2095 | |
2688fcb7 AB |
2096 | info->netdev->irq = 0; |
2097 | ||
50ee6061 | 2098 | /* Check if backend supports multiple queues */ |
2890ea5c JG |
2099 | max_queues = xenbus_read_unsigned(info->xbdev->otherend, |
2100 | "multi-queue-max-queues", 1); | |
50ee6061 AB |
2101 | num_queues = min(max_queues, xennet_max_queues); |
2102 | ||
2688fcb7 | 2103 | /* Check feature-split-event-channels */ |
2890ea5c JG |
2104 | feature_split_evtchn = xenbus_read_unsigned(info->xbdev->otherend, |
2105 | "feature-split-event-channels", 0); | |
2688fcb7 AB |
2106 | |
2107 | /* Read mac addr. */ | |
2108 | err = xen_net_read_mac(dev, info->netdev->dev_addr); | |
2109 | if (err) { | |
2110 | xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename); | |
cb257783 | 2111 | goto out_unlocked; |
2688fcb7 AB |
2112 | } |
2113 | ||
6c5aa6fc DK |
2114 | info->netback_has_xdp_headroom = xenbus_read_unsigned(info->xbdev->otherend, |
2115 | "feature-xdp-headroom", 0); | |
2116 | if (info->netback_has_xdp_headroom) { | |
2117 | /* set the current xen-netfront xdp state */ | |
2118 | err = talk_to_netback_xdp(info, info->netfront_xdp_enabled ? | |
2119 | NETBACK_XDP_HEADROOM_ENABLE : | |
2120 | NETBACK_XDP_HEADROOM_DISABLE); | |
2121 | if (err) | |
2122 | goto out_unlocked; | |
2123 | } | |
2124 | ||
f599c64f | 2125 | rtnl_lock(); |
ce58725f DV |
2126 | if (info->queues) |
2127 | xennet_destroy_queues(info); | |
2128 | ||
ca88ea12 | 2129 | err = xennet_create_queues(info, &num_queues); |
e2e004ac RL |
2130 | if (err < 0) { |
2131 | xenbus_dev_fatal(dev, err, "creating queues"); | |
2132 | kfree(info->queues); | |
2133 | info->queues = NULL; | |
2134 | goto out; | |
2135 | } | |
f599c64f | 2136 | rtnl_unlock(); |
2688fcb7 AB |
2137 | |
2138 | /* Create shared ring, alloc event channel -- for each queue */ | |
2139 | for (i = 0; i < num_queues; ++i) { | |
2140 | queue = &info->queues[i]; | |
2688fcb7 | 2141 | err = setup_netfront(dev, queue, feature_split_evtchn); |
e2e004ac RL |
2142 | if (err) |
2143 | goto destroy_ring; | |
2688fcb7 | 2144 | } |
0d160211 JF |
2145 | |
2146 | again: | |
2147 | err = xenbus_transaction_start(&xbt); | |
2148 | if (err) { | |
2149 | xenbus_dev_fatal(dev, err, "starting transaction"); | |
2150 | goto destroy_ring; | |
2151 | } | |
2152 | ||
812494d9 | 2153 | if (xenbus_exists(XBT_NIL, |
2154 | info->xbdev->otherend, "multi-queue-max-queues")) { | |
50ee6061 | 2155 | /* Write the number of queues */ |
812494d9 | 2156 | err = xenbus_printf(xbt, dev->nodename, |
2157 | "multi-queue-num-queues", "%u", num_queues); | |
d634bf2c | 2158 | if (err) { |
50ee6061 AB |
2159 | message = "writing multi-queue-num-queues"; |
2160 | goto abort_transaction_no_dev_fatal; | |
d634bf2c | 2161 | } |
812494d9 | 2162 | } |
50ee6061 | 2163 | |
812494d9 | 2164 | if (num_queues == 1) { |
2165 | err = write_queue_xenstore_keys(&info->queues[0], &xbt, 0); /* flat */ | |
2166 | if (err) | |
2167 | goto abort_transaction_no_dev_fatal; | |
2168 | } else { | |
50ee6061 AB |
2169 | /* Write the keys for each queue */ |
2170 | for (i = 0; i < num_queues; ++i) { | |
2171 | queue = &info->queues[i]; | |
2172 | err = write_queue_xenstore_keys(queue, &xbt, 1); /* hierarchical */ | |
2173 | if (err) | |
2174 | goto abort_transaction_no_dev_fatal; | |
d634bf2c | 2175 | } |
0d160211 JF |
2176 | } |
2177 | ||
50ee6061 | 2178 | /* The remaining keys are not queue-specific */ |
0d160211 JF |
2179 | err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u", |
2180 | 1); | |
2181 | if (err) { | |
2182 | message = "writing request-rx-copy"; | |
2183 | goto abort_transaction; | |
2184 | } | |
2185 | ||
2186 | err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1); | |
2187 | if (err) { | |
2188 | message = "writing feature-rx-notify"; | |
2189 | goto abort_transaction; | |
2190 | } | |
2191 | ||
2192 | err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1); | |
2193 | if (err) { | |
2194 | message = "writing feature-sg"; | |
2195 | goto abort_transaction; | |
2196 | } | |
2197 | ||
2198 | err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1); | |
2199 | if (err) { | |
2200 | message = "writing feature-gso-tcpv4"; | |
2201 | goto abort_transaction; | |
2202 | } | |
2203 | ||
2c0057de PD |
2204 | err = xenbus_write(xbt, dev->nodename, "feature-gso-tcpv6", "1"); |
2205 | if (err) { | |
2206 | message = "writing feature-gso-tcpv6"; | |
2207 | goto abort_transaction; | |
2208 | } | |
2209 | ||
2210 | err = xenbus_write(xbt, dev->nodename, "feature-ipv6-csum-offload", | |
2211 | "1"); | |
2212 | if (err) { | |
2213 | message = "writing feature-ipv6-csum-offload"; | |
2214 | goto abort_transaction; | |
2215 | } | |
2216 | ||
0d160211 JF |
2217 | err = xenbus_transaction_end(xbt, 0); |
2218 | if (err) { | |
2219 | if (err == -EAGAIN) | |
2220 | goto again; | |
2221 | xenbus_dev_fatal(dev, err, "completing transaction"); | |
2222 | goto destroy_ring; | |
2223 | } | |
2224 | ||
2225 | return 0; | |
2226 | ||
2227 | abort_transaction: | |
0d160211 | 2228 | xenbus_dev_fatal(dev, err, "%s", message); |
50ee6061 AB |
2229 | abort_transaction_no_dev_fatal: |
2230 | xenbus_transaction_end(xbt, 1); | |
0d160211 JF |
2231 | destroy_ring: |
2232 | xennet_disconnect_backend(info); | |
f599c64f | 2233 | rtnl_lock(); |
e2e004ac | 2234 | xennet_destroy_queues(info); |
0d160211 | 2235 | out: |
f599c64f | 2236 | rtnl_unlock(); |
cb257783 | 2237 | out_unlocked: |
d86b5672 | 2238 | device_unregister(&dev->dev); |
0d160211 JF |
2239 | return err; |
2240 | } | |
2241 | ||
0d160211 JF |
2242 | static int xennet_connect(struct net_device *dev) |
2243 | { | |
2244 | struct netfront_info *np = netdev_priv(dev); | |
2688fcb7 | 2245 | unsigned int num_queues = 0; |
a5b5dc3c | 2246 | int err; |
2688fcb7 AB |
2247 | unsigned int j = 0; |
2248 | struct netfront_queue *queue = NULL; | |
0d160211 | 2249 | |
2890ea5c | 2250 | if (!xenbus_read_unsigned(np->xbdev->otherend, "feature-rx-copy", 0)) { |
0d160211 | 2251 | dev_info(&dev->dev, |
898eb71c | 2252 | "backend does not support copying receive path\n"); |
0d160211 JF |
2253 | return -ENODEV; |
2254 | } | |
2255 | ||
f502bf2b | 2256 | err = talk_to_netback(np->xbdev, np); |
0d160211 JF |
2257 | if (err) |
2258 | return err; | |
6c5aa6fc DK |
2259 | if (np->netback_has_xdp_headroom) |
2260 | pr_info("backend supports XDP headroom\n"); | |
0d160211 | 2261 | |
2688fcb7 AB |
2262 | /* talk_to_netback() sets the correct number of queues */ |
2263 | num_queues = dev->real_num_tx_queues; | |
2264 | ||
f599c64f RL |
2265 | if (dev->reg_state == NETREG_UNINITIALIZED) { |
2266 | err = register_netdev(dev); | |
2267 | if (err) { | |
2268 | pr_warn("%s: register_netdev err=%d\n", __func__, err); | |
2269 | device_unregister(&np->xbdev->dev); | |
2270 | return err; | |
2271 | } | |
2272 | } | |
2273 | ||
45c8184c RL |
2274 | rtnl_lock(); |
2275 | netdev_update_features(dev); | |
2276 | rtnl_unlock(); | |
2277 | ||
0d160211 | 2278 | /* |
a5b5dc3c | 2279 | * All public and private state should now be sane. Get |
0d160211 JF |
2280 | * ready to start sending and receiving packets and give the driver |
2281 | * domain a kick because we've probably just requeued some | |
2282 | * packets. | |
2283 | */ | |
2284 | netif_carrier_on(np->netdev); | |
2688fcb7 AB |
2285 | for (j = 0; j < num_queues; ++j) { |
2286 | queue = &np->queues[j]; | |
f50b4076 | 2287 | |
2688fcb7 AB |
2288 | notify_remote_via_irq(queue->tx_irq); |
2289 | if (queue->tx_irq != queue->rx_irq) | |
2290 | notify_remote_via_irq(queue->rx_irq); | |
2688fcb7 | 2291 | |
f50b4076 DV |
2292 | spin_lock_irq(&queue->tx_lock); |
2293 | xennet_tx_buf_gc(queue); | |
2688fcb7 | 2294 | spin_unlock_irq(&queue->tx_lock); |
f50b4076 DV |
2295 | |
2296 | spin_lock_bh(&queue->rx_lock); | |
2297 | xennet_alloc_rx_buffers(queue); | |
2688fcb7 AB |
2298 | spin_unlock_bh(&queue->rx_lock); |
2299 | } | |
0d160211 JF |
2300 | |
2301 | return 0; | |
2302 | } | |
2303 | ||
80708602 | 2304 | /* |
0d160211 JF |
2305 | * Callback received when the backend's state changes. |
2306 | */ | |
f502bf2b | 2307 | static void netback_changed(struct xenbus_device *dev, |
0d160211 JF |
2308 | enum xenbus_state backend_state) |
2309 | { | |
1b713e00 | 2310 | struct netfront_info *np = dev_get_drvdata(&dev->dev); |
0d160211 JF |
2311 | struct net_device *netdev = np->netdev; |
2312 | ||
2313 | dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state)); | |
2314 | ||
8edfe2e9 JG |
2315 | wake_up_all(&module_wq); |
2316 | ||
0d160211 JF |
2317 | switch (backend_state) { |
2318 | case XenbusStateInitialising: | |
2319 | case XenbusStateInitialised: | |
b78c9512 NI |
2320 | case XenbusStateReconfiguring: |
2321 | case XenbusStateReconfigured: | |
0d160211 | 2322 | case XenbusStateUnknown: |
0d160211 JF |
2323 | break; |
2324 | ||
2325 | case XenbusStateInitWait: | |
2326 | if (dev->state != XenbusStateInitialising) | |
2327 | break; | |
2328 | if (xennet_connect(netdev) != 0) | |
2329 | break; | |
2330 | xenbus_switch_state(dev, XenbusStateConnected); | |
08e34eb1 LE |
2331 | break; |
2332 | ||
2333 | case XenbusStateConnected: | |
ee89bab1 | 2334 | netdev_notify_peers(netdev); |
0d160211 JF |
2335 | break; |
2336 | ||
bce3ea81 DV |
2337 | case XenbusStateClosed: |
2338 | if (dev->state == XenbusStateClosed) | |
2339 | break; | |
df561f66 | 2340 | fallthrough; /* Missed the backend's CLOSING state */ |
0d160211 JF |
2341 | case XenbusStateClosing: |
2342 | xenbus_frontend_closed(dev); | |
2343 | break; | |
2344 | } | |
2345 | } | |
2346 | ||
e0ce4af9 IC |
2347 | static const struct xennet_stat { |
2348 | char name[ETH_GSTRING_LEN]; | |
2349 | u16 offset; | |
2350 | } xennet_stats[] = { | |
2351 | { | |
2352 | "rx_gso_checksum_fixup", | |
2353 | offsetof(struct netfront_info, rx_gso_checksum_fixup) | |
2354 | }, | |
2355 | }; | |
2356 | ||
2357 | static int xennet_get_sset_count(struct net_device *dev, int string_set) | |
2358 | { | |
2359 | switch (string_set) { | |
2360 | case ETH_SS_STATS: | |
2361 | return ARRAY_SIZE(xennet_stats); | |
2362 | default: | |
2363 | return -EINVAL; | |
2364 | } | |
2365 | } | |
2366 | ||
2367 | static void xennet_get_ethtool_stats(struct net_device *dev, | |
2368 | struct ethtool_stats *stats, u64 * data) | |
2369 | { | |
2370 | void *np = netdev_priv(dev); | |
2371 | int i; | |
2372 | ||
2373 | for (i = 0; i < ARRAY_SIZE(xennet_stats); i++) | |
2688fcb7 | 2374 | data[i] = atomic_read((atomic_t *)(np + xennet_stats[i].offset)); |
e0ce4af9 IC |
2375 | } |
2376 | ||
2377 | static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data) | |
2378 | { | |
2379 | int i; | |
2380 | ||
2381 | switch (stringset) { | |
2382 | case ETH_SS_STATS: | |
2383 | for (i = 0; i < ARRAY_SIZE(xennet_stats); i++) | |
2384 | memcpy(data + i * ETH_GSTRING_LEN, | |
2385 | xennet_stats[i].name, ETH_GSTRING_LEN); | |
2386 | break; | |
2387 | } | |
2388 | } | |
2389 | ||
0fc0b732 | 2390 | static const struct ethtool_ops xennet_ethtool_ops = |
0d160211 | 2391 | { |
0d160211 | 2392 | .get_link = ethtool_op_get_link, |
e0ce4af9 IC |
2393 | |
2394 | .get_sset_count = xennet_get_sset_count, | |
2395 | .get_ethtool_stats = xennet_get_ethtool_stats, | |
2396 | .get_strings = xennet_get_strings, | |
91ffb9d3 | 2397 | .get_ts_info = ethtool_op_get_ts_info, |
0d160211 JF |
2398 | }; |
2399 | ||
2400 | #ifdef CONFIG_SYSFS | |
1f3c2eba DV |
2401 | static ssize_t show_rxbuf(struct device *dev, |
2402 | struct device_attribute *attr, char *buf) | |
0d160211 | 2403 | { |
1f3c2eba | 2404 | return sprintf(buf, "%lu\n", NET_RX_RING_SIZE); |
0d160211 JF |
2405 | } |
2406 | ||
1f3c2eba DV |
2407 | static ssize_t store_rxbuf(struct device *dev, |
2408 | struct device_attribute *attr, | |
2409 | const char *buf, size_t len) | |
0d160211 | 2410 | { |
0d160211 | 2411 | char *endp; |
0d160211 JF |
2412 | |
2413 | if (!capable(CAP_NET_ADMIN)) | |
2414 | return -EPERM; | |
2415 | ||
8ed7ec13 | 2416 | simple_strtoul(buf, &endp, 0); |
0d160211 JF |
2417 | if (endp == buf) |
2418 | return -EBADMSG; | |
2419 | ||
1f3c2eba | 2420 | /* rxbuf_min and rxbuf_max are no longer configurable. */ |
0d160211 | 2421 | |
0d160211 JF |
2422 | return len; |
2423 | } | |
2424 | ||
d61e4038 JP |
2425 | static DEVICE_ATTR(rxbuf_min, 0644, show_rxbuf, store_rxbuf); |
2426 | static DEVICE_ATTR(rxbuf_max, 0644, show_rxbuf, store_rxbuf); | |
2427 | static DEVICE_ATTR(rxbuf_cur, 0444, show_rxbuf, NULL); | |
0d160211 | 2428 | |
27b917e5 TI |
2429 | static struct attribute *xennet_dev_attrs[] = { |
2430 | &dev_attr_rxbuf_min.attr, | |
2431 | &dev_attr_rxbuf_max.attr, | |
2432 | &dev_attr_rxbuf_cur.attr, | |
2433 | NULL | |
2434 | }; | |
0d160211 | 2435 | |
27b917e5 TI |
2436 | static const struct attribute_group xennet_dev_group = { |
2437 | .attrs = xennet_dev_attrs | |
2438 | }; | |
0d160211 JF |
2439 | #endif /* CONFIG_SYSFS */ |
2440 | ||
c2c63310 | 2441 | static void xennet_bus_close(struct xenbus_device *dev) |
0d160211 | 2442 | { |
c2c63310 | 2443 | int ret; |
0d160211 | 2444 | |
c2c63310 AR |
2445 | if (xenbus_read_driver_state(dev->otherend) == XenbusStateClosed) |
2446 | return; | |
2447 | do { | |
5b5971df | 2448 | xenbus_switch_state(dev, XenbusStateClosing); |
c2c63310 AR |
2449 | ret = wait_event_timeout(module_wq, |
2450 | xenbus_read_driver_state(dev->otherend) == | |
2451 | XenbusStateClosing || | |
2452 | xenbus_read_driver_state(dev->otherend) == | |
2453 | XenbusStateClosed || | |
2454 | xenbus_read_driver_state(dev->otherend) == | |
2455 | XenbusStateUnknown, | |
2456 | XENNET_TIMEOUT); | |
2457 | } while (!ret); | |
2458 | ||
2459 | if (xenbus_read_driver_state(dev->otherend) == XenbusStateClosed) | |
2460 | return; | |
5b5971df | 2461 | |
c2c63310 | 2462 | do { |
5b5971df | 2463 | xenbus_switch_state(dev, XenbusStateClosed); |
c2c63310 AR |
2464 | ret = wait_event_timeout(module_wq, |
2465 | xenbus_read_driver_state(dev->otherend) == | |
2466 | XenbusStateClosed || | |
2467 | xenbus_read_driver_state(dev->otherend) == | |
2468 | XenbusStateUnknown, | |
2469 | XENNET_TIMEOUT); | |
2470 | } while (!ret); | |
2471 | } | |
2472 | ||
2473 | static int xennet_remove(struct xenbus_device *dev) | |
2474 | { | |
2475 | struct netfront_info *info = dev_get_drvdata(&dev->dev); | |
5b5971df | 2476 | |
c2c63310 | 2477 | xennet_bus_close(dev); |
0d160211 JF |
2478 | xennet_disconnect_backend(info); |
2479 | ||
f599c64f RL |
2480 | if (info->netdev->reg_state == NETREG_REGISTERED) |
2481 | unregister_netdev(info->netdev); | |
6bc96d04 | 2482 | |
f599c64f RL |
2483 | if (info->queues) { |
2484 | rtnl_lock(); | |
9a873c71 | 2485 | xennet_destroy_queues(info); |
f599c64f RL |
2486 | rtnl_unlock(); |
2487 | } | |
900e1833 | 2488 | xennet_free_netdev(info->netdev); |
0d160211 JF |
2489 | |
2490 | return 0; | |
2491 | } | |
2492 | ||
95afae48 DV |
2493 | static const struct xenbus_device_id netfront_ids[] = { |
2494 | { "vif" }, | |
2495 | { "" } | |
2496 | }; | |
2497 | ||
2498 | static struct xenbus_driver netfront_driver = { | |
2499 | .ids = netfront_ids, | |
0d160211 | 2500 | .probe = netfront_probe, |
8e0e46bb | 2501 | .remove = xennet_remove, |
0d160211 | 2502 | .resume = netfront_resume, |
f502bf2b | 2503 | .otherend_changed = netback_changed, |
95afae48 | 2504 | }; |
0d160211 JF |
2505 | |
2506 | static int __init netif_init(void) | |
2507 | { | |
6e833587 | 2508 | if (!xen_domain()) |
0d160211 JF |
2509 | return -ENODEV; |
2510 | ||
51c71a3b | 2511 | if (!xen_has_pv_nic_devices()) |
b9136d20 IM |
2512 | return -ENODEV; |
2513 | ||
383eda32 | 2514 | pr_info("Initialising Xen virtual ethernet driver\n"); |
0d160211 | 2515 | |
034702a6 | 2516 | /* Allow as many queues as there are CPUs inut max. 8 if user has not |
32a84405 WL |
2517 | * specified a value. |
2518 | */ | |
2519 | if (xennet_max_queues == 0) | |
034702a6 JG |
2520 | xennet_max_queues = min_t(unsigned int, MAX_QUEUES_DEFAULT, |
2521 | num_online_cpus()); | |
50ee6061 | 2522 | |
ffb78a26 | 2523 | return xenbus_register_frontend(&netfront_driver); |
0d160211 JF |
2524 | } |
2525 | module_init(netif_init); | |
2526 | ||
2527 | ||
2528 | static void __exit netif_exit(void) | |
2529 | { | |
ffb78a26 | 2530 | xenbus_unregister_driver(&netfront_driver); |
0d160211 JF |
2531 | } |
2532 | module_exit(netif_exit); | |
2533 | ||
2534 | MODULE_DESCRIPTION("Xen virtual network device frontend"); | |
2535 | MODULE_LICENSE("GPL"); | |
d2f0c52b | 2536 | MODULE_ALIAS("xen:vif"); |
4f93f09b | 2537 | MODULE_ALIAS("xennet"); |