]>
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 | |
21631d2d | 129 | * are linked from tx_skb_freelist through tx_link. |
0d160211 | 130 | */ |
21631d2d JG |
131 | struct sk_buff *tx_skbs[NET_TX_RING_SIZE]; |
132 | unsigned short tx_link[NET_TX_RING_SIZE]; | |
133 | #define TX_LINK_NONE 0xffff | |
0d160211 JF |
134 | grant_ref_t gref_tx_head; |
135 | grant_ref_t grant_tx_ref[NET_TX_RING_SIZE]; | |
cefe0078 | 136 | struct page *grant_tx_page[NET_TX_RING_SIZE]; |
0d160211 JF |
137 | unsigned tx_skb_freelist; |
138 | ||
84284d3c JF |
139 | spinlock_t rx_lock ____cacheline_aligned_in_smp; |
140 | struct xen_netif_rx_front_ring rx; | |
141 | int rx_ring_ref; | |
142 | ||
84284d3c JF |
143 | struct timer_list rx_refill_timer; |
144 | ||
0d160211 JF |
145 | struct sk_buff *rx_skbs[NET_RX_RING_SIZE]; |
146 | grant_ref_t gref_rx_head; | |
147 | grant_ref_t grant_rx_ref[NET_RX_RING_SIZE]; | |
6c5aa6fc DK |
148 | |
149 | struct page_pool *page_pool; | |
150 | struct xdp_rxq_info xdp_rxq; | |
2688fcb7 AB |
151 | }; |
152 | ||
153 | struct netfront_info { | |
154 | struct list_head list; | |
155 | struct net_device *netdev; | |
156 | ||
157 | struct xenbus_device *xbdev; | |
158 | ||
159 | /* Multi-queue support */ | |
160 | struct netfront_queue *queues; | |
e0ce4af9 IC |
161 | |
162 | /* Statistics */ | |
900e1833 DV |
163 | struct netfront_stats __percpu *rx_stats; |
164 | struct netfront_stats __percpu *tx_stats; | |
e00f85be | 165 | |
6c5aa6fc DK |
166 | /* XDP state */ |
167 | bool netback_has_xdp_headroom; | |
168 | bool netfront_xdp_enabled; | |
169 | ||
2688fcb7 | 170 | atomic_t rx_gso_checksum_fixup; |
0d160211 JF |
171 | }; |
172 | ||
173 | struct netfront_rx_info { | |
174 | struct xen_netif_rx_response rx; | |
175 | struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1]; | |
176 | }; | |
177 | ||
178 | /* | |
179 | * Access macros for acquiring freeing slots in tx_skbs[]. | |
180 | */ | |
181 | ||
21631d2d JG |
182 | static void add_id_to_list(unsigned *head, unsigned short *list, |
183 | unsigned short id) | |
0d160211 | 184 | { |
21631d2d | 185 | list[id] = *head; |
0d160211 JF |
186 | *head = id; |
187 | } | |
188 | ||
21631d2d | 189 | static unsigned short get_id_from_list(unsigned *head, unsigned short *list) |
0d160211 JF |
190 | { |
191 | unsigned int id = *head; | |
21631d2d JG |
192 | |
193 | if (id != TX_LINK_NONE) { | |
194 | *head = list[id]; | |
195 | list[id] = TX_LINK_NONE; | |
196 | } | |
0d160211 JF |
197 | return id; |
198 | } | |
199 | ||
200 | static int xennet_rxidx(RING_IDX idx) | |
201 | { | |
202 | return idx & (NET_RX_RING_SIZE - 1); | |
203 | } | |
204 | ||
2688fcb7 | 205 | static struct sk_buff *xennet_get_rx_skb(struct netfront_queue *queue, |
0d160211 JF |
206 | RING_IDX ri) |
207 | { | |
208 | int i = xennet_rxidx(ri); | |
2688fcb7 AB |
209 | struct sk_buff *skb = queue->rx_skbs[i]; |
210 | queue->rx_skbs[i] = NULL; | |
0d160211 JF |
211 | return skb; |
212 | } | |
213 | ||
2688fcb7 | 214 | static grant_ref_t xennet_get_rx_ref(struct netfront_queue *queue, |
0d160211 JF |
215 | RING_IDX ri) |
216 | { | |
217 | int i = xennet_rxidx(ri); | |
2688fcb7 AB |
218 | grant_ref_t ref = queue->grant_rx_ref[i]; |
219 | queue->grant_rx_ref[i] = GRANT_INVALID_REF; | |
0d160211 JF |
220 | return ref; |
221 | } | |
222 | ||
223 | #ifdef CONFIG_SYSFS | |
27b917e5 | 224 | static const struct attribute_group xennet_dev_group; |
0d160211 JF |
225 | #endif |
226 | ||
3ad9b358 | 227 | static bool xennet_can_sg(struct net_device *dev) |
0d160211 | 228 | { |
3ad9b358 | 229 | return dev->features & NETIF_F_SG; |
0d160211 JF |
230 | } |
231 | ||
232 | ||
e99e88a9 | 233 | static void rx_refill_timeout(struct timer_list *t) |
0d160211 | 234 | { |
e99e88a9 | 235 | struct netfront_queue *queue = from_timer(queue, t, rx_refill_timer); |
2688fcb7 | 236 | napi_schedule(&queue->napi); |
0d160211 JF |
237 | } |
238 | ||
2688fcb7 | 239 | static int netfront_tx_slot_available(struct netfront_queue *queue) |
0d160211 | 240 | { |
2688fcb7 | 241 | return (queue->tx.req_prod_pvt - queue->tx.rsp_cons) < |
57f230ab | 242 | (NET_TX_RING_SIZE - XEN_NETIF_NR_SLOTS_MIN - 1); |
0d160211 JF |
243 | } |
244 | ||
2688fcb7 | 245 | static void xennet_maybe_wake_tx(struct netfront_queue *queue) |
0d160211 | 246 | { |
2688fcb7 AB |
247 | struct net_device *dev = queue->info->netdev; |
248 | struct netdev_queue *dev_queue = netdev_get_tx_queue(dev, queue->id); | |
0d160211 | 249 | |
2688fcb7 AB |
250 | if (unlikely(netif_tx_queue_stopped(dev_queue)) && |
251 | netfront_tx_slot_available(queue) && | |
0d160211 | 252 | likely(netif_running(dev))) |
2688fcb7 | 253 | netif_tx_wake_queue(netdev_get_tx_queue(dev, queue->id)); |
0d160211 JF |
254 | } |
255 | ||
1f3c2eba DV |
256 | |
257 | static struct sk_buff *xennet_alloc_one_rx_buffer(struct netfront_queue *queue) | |
0d160211 | 258 | { |
0d160211 JF |
259 | struct sk_buff *skb; |
260 | struct page *page; | |
0d160211 | 261 | |
1f3c2eba DV |
262 | skb = __netdev_alloc_skb(queue->info->netdev, |
263 | RX_COPY_THRESHOLD + NET_IP_ALIGN, | |
264 | GFP_ATOMIC | __GFP_NOWARN); | |
265 | if (unlikely(!skb)) | |
266 | return NULL; | |
0d160211 | 267 | |
6c5aa6fc DK |
268 | page = page_pool_dev_alloc_pages(queue->page_pool); |
269 | if (unlikely(!page)) { | |
1f3c2eba DV |
270 | kfree_skb(skb); |
271 | return NULL; | |
0d160211 | 272 | } |
1f3c2eba DV |
273 | skb_add_rx_frag(skb, 0, page, 0, 0, PAGE_SIZE); |
274 | ||
275 | /* Align ip header to a 16 bytes boundary */ | |
276 | skb_reserve(skb, NET_IP_ALIGN); | |
277 | skb->dev = queue->info->netdev; | |
278 | ||
279 | return skb; | |
280 | } | |
0d160211 | 281 | |
1f3c2eba DV |
282 | |
283 | static void xennet_alloc_rx_buffers(struct netfront_queue *queue) | |
284 | { | |
285 | RING_IDX req_prod = queue->rx.req_prod_pvt; | |
286 | int notify; | |
538d9291 | 287 | int err = 0; |
1f3c2eba DV |
288 | |
289 | if (unlikely(!netif_carrier_ok(queue->info->netdev))) | |
0d160211 | 290 | return; |
0d160211 | 291 | |
1f3c2eba DV |
292 | for (req_prod = queue->rx.req_prod_pvt; |
293 | req_prod - queue->rx.rsp_cons < NET_RX_RING_SIZE; | |
294 | req_prod++) { | |
295 | struct sk_buff *skb; | |
296 | unsigned short id; | |
297 | grant_ref_t ref; | |
30c5d7f0 | 298 | struct page *page; |
1f3c2eba | 299 | struct xen_netif_rx_request *req; |
0d160211 | 300 | |
1f3c2eba | 301 | skb = xennet_alloc_one_rx_buffer(queue); |
538d9291 VRP |
302 | if (!skb) { |
303 | err = -ENOMEM; | |
0d160211 | 304 | break; |
538d9291 | 305 | } |
0d160211 | 306 | |
1f3c2eba | 307 | id = xennet_rxidx(req_prod); |
0d160211 | 308 | |
2688fcb7 AB |
309 | BUG_ON(queue->rx_skbs[id]); |
310 | queue->rx_skbs[id] = skb; | |
0d160211 | 311 | |
2688fcb7 | 312 | ref = gnttab_claim_grant_reference(&queue->gref_rx_head); |
269ebce4 | 313 | WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref)); |
2688fcb7 | 314 | queue->grant_rx_ref[id] = ref; |
0d160211 | 315 | |
30c5d7f0 | 316 | page = skb_frag_page(&skb_shinfo(skb)->frags[0]); |
0d160211 | 317 | |
1f3c2eba | 318 | req = RING_GET_REQUEST(&queue->rx, req_prod); |
30c5d7f0 JG |
319 | gnttab_page_grant_foreign_access_ref_one(ref, |
320 | queue->info->xbdev->otherend_id, | |
321 | page, | |
322 | 0); | |
0d160211 JF |
323 | req->id = id; |
324 | req->gref = ref; | |
325 | } | |
326 | ||
1f3c2eba DV |
327 | queue->rx.req_prod_pvt = req_prod; |
328 | ||
538d9291 VRP |
329 | /* Try again later if there are not enough requests or skb allocation |
330 | * failed. | |
331 | * Enough requests is quantified as the sum of newly created slots and | |
332 | * the unconsumed slots at the backend. | |
333 | */ | |
334 | if (req_prod - queue->rx.rsp_cons < NET_RX_SLOTS_MIN || | |
335 | unlikely(err)) { | |
1f3c2eba DV |
336 | mod_timer(&queue->rx_refill_timer, jiffies + (HZ/10)); |
337 | return; | |
338 | } | |
339 | ||
2688fcb7 | 340 | RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->rx, notify); |
0d160211 | 341 | if (notify) |
2688fcb7 | 342 | notify_remote_via_irq(queue->rx_irq); |
0d160211 JF |
343 | } |
344 | ||
345 | static int xennet_open(struct net_device *dev) | |
346 | { | |
347 | struct netfront_info *np = netdev_priv(dev); | |
2688fcb7 AB |
348 | unsigned int num_queues = dev->real_num_tx_queues; |
349 | unsigned int i = 0; | |
350 | struct netfront_queue *queue = NULL; | |
351 | ||
f599c64f RL |
352 | if (!np->queues) |
353 | return -ENODEV; | |
354 | ||
2688fcb7 AB |
355 | for (i = 0; i < num_queues; ++i) { |
356 | queue = &np->queues[i]; | |
357 | napi_enable(&queue->napi); | |
358 | ||
359 | spin_lock_bh(&queue->rx_lock); | |
360 | if (netif_carrier_ok(dev)) { | |
361 | xennet_alloc_rx_buffers(queue); | |
362 | queue->rx.sring->rsp_event = queue->rx.rsp_cons + 1; | |
363 | if (RING_HAS_UNCONSUMED_RESPONSES(&queue->rx)) | |
364 | napi_schedule(&queue->napi); | |
365 | } | |
366 | spin_unlock_bh(&queue->rx_lock); | |
0d160211 | 367 | } |
0d160211 | 368 | |
2688fcb7 | 369 | netif_tx_start_all_queues(dev); |
0d160211 JF |
370 | |
371 | return 0; | |
372 | } | |
373 | ||
2688fcb7 | 374 | static void xennet_tx_buf_gc(struct netfront_queue *queue) |
0d160211 JF |
375 | { |
376 | RING_IDX cons, prod; | |
377 | unsigned short id; | |
0d160211 | 378 | struct sk_buff *skb; |
7d0105b5 | 379 | bool more_to_do; |
0d160211 | 380 | |
2688fcb7 | 381 | BUG_ON(!netif_carrier_ok(queue->info->netdev)); |
0d160211 JF |
382 | |
383 | do { | |
2688fcb7 | 384 | prod = queue->tx.sring->rsp_prod; |
0d160211 JF |
385 | rmb(); /* Ensure we see responses up to 'rp'. */ |
386 | ||
2688fcb7 | 387 | for (cons = queue->tx.rsp_cons; cons != prod; cons++) { |
8446066b | 388 | struct xen_netif_tx_response txrsp; |
0d160211 | 389 | |
8446066b JG |
390 | RING_COPY_RESPONSE(&queue->tx, cons, &txrsp); |
391 | if (txrsp.status == XEN_NETIF_RSP_NULL) | |
0d160211 JF |
392 | continue; |
393 | ||
8446066b | 394 | id = txrsp.id; |
21631d2d JG |
395 | skb = queue->tx_skbs[id]; |
396 | queue->tx_skbs[id] = NULL; | |
0d160211 | 397 | if (unlikely(gnttab_query_foreign_access( |
2688fcb7 | 398 | queue->grant_tx_ref[id]) != 0)) { |
383eda32 JP |
399 | pr_alert("%s: warning -- grant still in use by backend domain\n", |
400 | __func__); | |
0d160211 JF |
401 | BUG(); |
402 | } | |
403 | gnttab_end_foreign_access_ref( | |
2688fcb7 | 404 | queue->grant_tx_ref[id], GNTMAP_readonly); |
0d160211 | 405 | gnttab_release_grant_reference( |
2688fcb7 AB |
406 | &queue->gref_tx_head, queue->grant_tx_ref[id]); |
407 | queue->grant_tx_ref[id] = GRANT_INVALID_REF; | |
408 | queue->grant_tx_page[id] = NULL; | |
21631d2d | 409 | add_id_to_list(&queue->tx_skb_freelist, queue->tx_link, id); |
0d160211 JF |
410 | dev_kfree_skb_irq(skb); |
411 | } | |
412 | ||
2688fcb7 | 413 | queue->tx.rsp_cons = prod; |
0d160211 | 414 | |
7d0105b5 MC |
415 | RING_FINAL_CHECK_FOR_RESPONSES(&queue->tx, more_to_do); |
416 | } while (more_to_do); | |
0d160211 | 417 | |
2688fcb7 | 418 | xennet_maybe_wake_tx(queue); |
0d160211 JF |
419 | } |
420 | ||
30c5d7f0 JG |
421 | struct xennet_gnttab_make_txreq { |
422 | struct netfront_queue *queue; | |
423 | struct sk_buff *skb; | |
424 | struct page *page; | |
162081ec JG |
425 | struct xen_netif_tx_request *tx; /* Last request on ring page */ |
426 | struct xen_netif_tx_request tx_local; /* Last request local copy*/ | |
30c5d7f0 JG |
427 | unsigned int size; |
428 | }; | |
429 | ||
430 | static void xennet_tx_setup_grant(unsigned long gfn, unsigned int offset, | |
431 | unsigned int len, void *data) | |
0d160211 | 432 | { |
30c5d7f0 | 433 | struct xennet_gnttab_make_txreq *info = data; |
0d160211 | 434 | unsigned int id; |
a55e8bb8 | 435 | struct xen_netif_tx_request *tx; |
0d160211 | 436 | grant_ref_t ref; |
30c5d7f0 JG |
437 | /* convenient aliases */ |
438 | struct page *page = info->page; | |
439 | struct netfront_queue *queue = info->queue; | |
440 | struct sk_buff *skb = info->skb; | |
0d160211 | 441 | |
21631d2d | 442 | id = get_id_from_list(&queue->tx_skb_freelist, queue->tx_link); |
a55e8bb8 DV |
443 | tx = RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++); |
444 | ref = gnttab_claim_grant_reference(&queue->gref_tx_head); | |
269ebce4 | 445 | WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref)); |
0d160211 | 446 | |
30c5d7f0 JG |
447 | gnttab_grant_foreign_access_ref(ref, queue->info->xbdev->otherend_id, |
448 | gfn, GNTMAP_readonly); | |
0d160211 | 449 | |
21631d2d | 450 | queue->tx_skbs[id] = skb; |
a55e8bb8 DV |
451 | queue->grant_tx_page[id] = page; |
452 | queue->grant_tx_ref[id] = ref; | |
0d160211 | 453 | |
162081ec JG |
454 | info->tx_local.id = id; |
455 | info->tx_local.gref = ref; | |
456 | info->tx_local.offset = offset; | |
457 | info->tx_local.size = len; | |
458 | info->tx_local.flags = 0; | |
459 | ||
460 | *tx = info->tx_local; | |
0d160211 | 461 | |
30c5d7f0 | 462 | info->tx = tx; |
162081ec | 463 | info->size += info->tx_local.size; |
30c5d7f0 JG |
464 | } |
465 | ||
466 | static struct xen_netif_tx_request *xennet_make_first_txreq( | |
162081ec JG |
467 | struct xennet_gnttab_make_txreq *info, |
468 | unsigned int offset, unsigned int len) | |
30c5d7f0 | 469 | { |
162081ec | 470 | info->size = 0; |
30c5d7f0 | 471 | |
162081ec | 472 | gnttab_for_one_grant(info->page, offset, len, xennet_tx_setup_grant, info); |
30c5d7f0 | 473 | |
162081ec | 474 | return info->tx; |
30c5d7f0 JG |
475 | } |
476 | ||
477 | static void xennet_make_one_txreq(unsigned long gfn, unsigned int offset, | |
478 | unsigned int len, void *data) | |
479 | { | |
480 | struct xennet_gnttab_make_txreq *info = data; | |
481 | ||
482 | info->tx->flags |= XEN_NETTXF_more_data; | |
483 | skb_get(info->skb); | |
484 | xennet_tx_setup_grant(gfn, offset, len, data); | |
a55e8bb8 | 485 | } |
0d160211 | 486 | |
162081ec JG |
487 | static void xennet_make_txreqs( |
488 | struct xennet_gnttab_make_txreq *info, | |
489 | struct page *page, | |
a55e8bb8 DV |
490 | unsigned int offset, unsigned int len) |
491 | { | |
492 | /* Skip unused frames from start of page */ | |
493 | page += offset >> PAGE_SHIFT; | |
494 | offset &= ~PAGE_MASK; | |
0d160211 | 495 | |
a55e8bb8 | 496 | while (len) { |
162081ec JG |
497 | info->page = page; |
498 | info->size = 0; | |
30c5d7f0 JG |
499 | |
500 | gnttab_foreach_grant_in_range(page, offset, len, | |
501 | xennet_make_one_txreq, | |
162081ec | 502 | info); |
30c5d7f0 | 503 | |
a55e8bb8 DV |
504 | page++; |
505 | offset = 0; | |
162081ec | 506 | len -= info->size; |
0d160211 | 507 | } |
0d160211 JF |
508 | } |
509 | ||
f36c3747 | 510 | /* |
e84448d5 DV |
511 | * Count how many ring slots are required to send this skb. Each frag |
512 | * might be a compound page. | |
f36c3747 | 513 | */ |
e84448d5 | 514 | static int xennet_count_skb_slots(struct sk_buff *skb) |
f36c3747 IC |
515 | { |
516 | int i, frags = skb_shinfo(skb)->nr_frags; | |
30c5d7f0 | 517 | int slots; |
e84448d5 | 518 | |
30c5d7f0 JG |
519 | slots = gnttab_count_grant(offset_in_page(skb->data), |
520 | skb_headlen(skb)); | |
f36c3747 IC |
521 | |
522 | for (i = 0; i < frags; i++) { | |
523 | skb_frag_t *frag = skb_shinfo(skb)->frags + i; | |
524 | unsigned long size = skb_frag_size(frag); | |
b54c9d5b | 525 | unsigned long offset = skb_frag_off(frag); |
f36c3747 IC |
526 | |
527 | /* Skip unused frames from start of page */ | |
528 | offset &= ~PAGE_MASK; | |
529 | ||
30c5d7f0 | 530 | slots += gnttab_count_grant(offset, size); |
f36c3747 IC |
531 | } |
532 | ||
30c5d7f0 | 533 | return slots; |
f36c3747 IC |
534 | } |
535 | ||
50ee6061 | 536 | static u16 xennet_select_queue(struct net_device *dev, struct sk_buff *skb, |
a350ecce | 537 | struct net_device *sb_dev) |
2688fcb7 | 538 | { |
50ee6061 AB |
539 | unsigned int num_queues = dev->real_num_tx_queues; |
540 | u32 hash; | |
541 | u16 queue_idx; | |
542 | ||
543 | /* First, check if there is only one queue */ | |
544 | if (num_queues == 1) { | |
545 | queue_idx = 0; | |
546 | } else { | |
547 | hash = skb_get_hash(skb); | |
548 | queue_idx = hash % num_queues; | |
549 | } | |
550 | ||
551 | return queue_idx; | |
2688fcb7 AB |
552 | } |
553 | ||
6c5aa6fc DK |
554 | static int xennet_xdp_xmit_one(struct net_device *dev, |
555 | struct netfront_queue *queue, | |
556 | struct xdp_frame *xdpf) | |
557 | { | |
558 | struct netfront_info *np = netdev_priv(dev); | |
559 | struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats); | |
162081ec JG |
560 | struct xennet_gnttab_make_txreq info = { |
561 | .queue = queue, | |
562 | .skb = NULL, | |
563 | .page = virt_to_page(xdpf->data), | |
564 | }; | |
6c5aa6fc DK |
565 | int notify; |
566 | ||
162081ec | 567 | xennet_make_first_txreq(&info, |
6c5aa6fc DK |
568 | offset_in_page(xdpf->data), |
569 | xdpf->len); | |
570 | ||
571 | RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify); | |
572 | if (notify) | |
573 | notify_remote_via_irq(queue->tx_irq); | |
574 | ||
575 | u64_stats_update_begin(&tx_stats->syncp); | |
576 | tx_stats->bytes += xdpf->len; | |
577 | tx_stats->packets++; | |
578 | u64_stats_update_end(&tx_stats->syncp); | |
579 | ||
580 | xennet_tx_buf_gc(queue); | |
581 | ||
582 | return 0; | |
583 | } | |
584 | ||
585 | static int xennet_xdp_xmit(struct net_device *dev, int n, | |
586 | struct xdp_frame **frames, u32 flags) | |
587 | { | |
588 | unsigned int num_queues = dev->real_num_tx_queues; | |
589 | struct netfront_info *np = netdev_priv(dev); | |
590 | struct netfront_queue *queue = NULL; | |
591 | unsigned long irq_flags; | |
fdc13979 LB |
592 | int nxmit = 0; |
593 | int i; | |
6c5aa6fc DK |
594 | |
595 | if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK)) | |
596 | return -EINVAL; | |
597 | ||
598 | queue = &np->queues[smp_processor_id() % num_queues]; | |
599 | ||
600 | spin_lock_irqsave(&queue->tx_lock, irq_flags); | |
601 | for (i = 0; i < n; i++) { | |
602 | struct xdp_frame *xdpf = frames[i]; | |
603 | ||
604 | if (!xdpf) | |
605 | continue; | |
fdc13979 LB |
606 | if (xennet_xdp_xmit_one(dev, queue, xdpf)) |
607 | break; | |
608 | nxmit++; | |
6c5aa6fc DK |
609 | } |
610 | spin_unlock_irqrestore(&queue->tx_lock, irq_flags); | |
611 | ||
fdc13979 | 612 | return nxmit; |
6c5aa6fc DK |
613 | } |
614 | ||
615 | ||
30c5d7f0 JG |
616 | #define MAX_XEN_SKB_FRAGS (65536 / XEN_PAGE_SIZE + 1) |
617 | ||
24a94b3c | 618 | static netdev_tx_t xennet_start_xmit(struct sk_buff *skb, struct net_device *dev) |
0d160211 | 619 | { |
0d160211 | 620 | struct netfront_info *np = netdev_priv(dev); |
900e1833 | 621 | struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats); |
162081ec | 622 | struct xen_netif_tx_request *first_tx; |
a55e8bb8 | 623 | unsigned int i; |
0d160211 | 624 | int notify; |
f36c3747 | 625 | int slots; |
a55e8bb8 DV |
626 | struct page *page; |
627 | unsigned int offset; | |
628 | unsigned int len; | |
cf66f9d4 | 629 | unsigned long flags; |
2688fcb7 | 630 | struct netfront_queue *queue = NULL; |
162081ec | 631 | struct xennet_gnttab_make_txreq info = { }; |
2688fcb7 AB |
632 | unsigned int num_queues = dev->real_num_tx_queues; |
633 | u16 queue_index; | |
fd07160b | 634 | struct sk_buff *nskb; |
2688fcb7 AB |
635 | |
636 | /* Drop the packet if no queues are set up */ | |
637 | if (num_queues < 1) | |
638 | goto drop; | |
639 | /* Determine which queue to transmit this SKB on */ | |
640 | queue_index = skb_get_queue_mapping(skb); | |
641 | queue = &np->queues[queue_index]; | |
0d160211 | 642 | |
9ecd1a75 WL |
643 | /* If skb->len is too big for wire format, drop skb and alert |
644 | * user about misconfiguration. | |
645 | */ | |
646 | if (unlikely(skb->len > XEN_NETIF_MAX_TX_SIZE)) { | |
647 | net_alert_ratelimited( | |
648 | "xennet: skb->len = %u, too big for wire format\n", | |
649 | skb->len); | |
650 | goto drop; | |
651 | } | |
652 | ||
e84448d5 | 653 | slots = xennet_count_skb_slots(skb); |
30c5d7f0 | 654 | if (unlikely(slots > MAX_XEN_SKB_FRAGS + 1)) { |
97a6d1bb ZK |
655 | net_dbg_ratelimited("xennet: skb rides the rocket: %d slots, %d bytes\n", |
656 | slots, skb->len); | |
657 | if (skb_linearize(skb)) | |
658 | goto drop; | |
0d160211 JF |
659 | } |
660 | ||
a55e8bb8 DV |
661 | page = virt_to_page(skb->data); |
662 | offset = offset_in_page(skb->data); | |
fd07160b VK |
663 | |
664 | /* The first req should be at least ETH_HLEN size or the packet will be | |
665 | * dropped by netback. | |
666 | */ | |
667 | if (unlikely(PAGE_SIZE - offset < ETH_HLEN)) { | |
668 | nskb = skb_copy(skb, GFP_ATOMIC); | |
669 | if (!nskb) | |
670 | goto drop; | |
62f3250f | 671 | dev_consume_skb_any(skb); |
fd07160b VK |
672 | skb = nskb; |
673 | page = virt_to_page(skb->data); | |
674 | offset = offset_in_page(skb->data); | |
675 | } | |
676 | ||
a55e8bb8 DV |
677 | len = skb_headlen(skb); |
678 | ||
2688fcb7 | 679 | spin_lock_irqsave(&queue->tx_lock, flags); |
0d160211 JF |
680 | |
681 | if (unlikely(!netif_carrier_ok(dev) || | |
f36c3747 | 682 | (slots > 1 && !xennet_can_sg(dev)) || |
8b86a61d | 683 | netif_needs_gso(skb, netif_skb_features(skb)))) { |
2688fcb7 | 684 | spin_unlock_irqrestore(&queue->tx_lock, flags); |
0d160211 JF |
685 | goto drop; |
686 | } | |
687 | ||
a55e8bb8 | 688 | /* First request for the linear area. */ |
162081ec JG |
689 | info.queue = queue; |
690 | info.skb = skb; | |
691 | info.page = page; | |
692 | first_tx = xennet_make_first_txreq(&info, offset, len); | |
693 | offset += info.tx_local.size; | |
30c5d7f0 JG |
694 | if (offset == PAGE_SIZE) { |
695 | page++; | |
696 | offset = 0; | |
697 | } | |
162081ec | 698 | len -= info.tx_local.size; |
0d160211 | 699 | |
0d160211 JF |
700 | if (skb->ip_summed == CHECKSUM_PARTIAL) |
701 | /* local packet? */ | |
162081ec JG |
702 | first_tx->flags |= XEN_NETTXF_csum_blank | |
703 | XEN_NETTXF_data_validated; | |
0d160211 JF |
704 | else if (skb->ip_summed == CHECKSUM_UNNECESSARY) |
705 | /* remote but checksummed. */ | |
162081ec | 706 | first_tx->flags |= XEN_NETTXF_data_validated; |
0d160211 | 707 | |
a55e8bb8 | 708 | /* Optional extra info after the first request. */ |
0d160211 JF |
709 | if (skb_shinfo(skb)->gso_size) { |
710 | struct xen_netif_extra_info *gso; | |
711 | ||
712 | gso = (struct xen_netif_extra_info *) | |
a55e8bb8 | 713 | RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++); |
0d160211 | 714 | |
162081ec | 715 | first_tx->flags |= XEN_NETTXF_extra_info; |
0d160211 JF |
716 | |
717 | gso->u.gso.size = skb_shinfo(skb)->gso_size; | |
2c0057de PD |
718 | gso->u.gso.type = (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) ? |
719 | XEN_NETIF_GSO_TYPE_TCPV6 : | |
720 | XEN_NETIF_GSO_TYPE_TCPV4; | |
0d160211 JF |
721 | gso->u.gso.pad = 0; |
722 | gso->u.gso.features = 0; | |
723 | ||
724 | gso->type = XEN_NETIF_EXTRA_TYPE_GSO; | |
725 | gso->flags = 0; | |
0d160211 JF |
726 | } |
727 | ||
a55e8bb8 | 728 | /* Requests for the rest of the linear area. */ |
162081ec | 729 | xennet_make_txreqs(&info, page, offset, len); |
a55e8bb8 DV |
730 | |
731 | /* Requests for all the frags. */ | |
732 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
733 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
162081ec | 734 | xennet_make_txreqs(&info, skb_frag_page(frag), |
b54c9d5b | 735 | skb_frag_off(frag), |
a55e8bb8 DV |
736 | skb_frag_size(frag)); |
737 | } | |
0d160211 | 738 | |
a55e8bb8 DV |
739 | /* First request has the packet length. */ |
740 | first_tx->size = skb->len; | |
0d160211 | 741 | |
91ffb9d3 DD |
742 | /* timestamp packet in software */ |
743 | skb_tx_timestamp(skb); | |
744 | ||
2688fcb7 | 745 | RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify); |
0d160211 | 746 | if (notify) |
2688fcb7 | 747 | notify_remote_via_irq(queue->tx_irq); |
0d160211 | 748 | |
900e1833 DV |
749 | u64_stats_update_begin(&tx_stats->syncp); |
750 | tx_stats->bytes += skb->len; | |
751 | tx_stats->packets++; | |
752 | u64_stats_update_end(&tx_stats->syncp); | |
10a273a6 JF |
753 | |
754 | /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */ | |
2688fcb7 | 755 | xennet_tx_buf_gc(queue); |
0d160211 | 756 | |
2688fcb7 AB |
757 | if (!netfront_tx_slot_available(queue)) |
758 | netif_tx_stop_queue(netdev_get_tx_queue(dev, queue->id)); | |
0d160211 | 759 | |
2688fcb7 | 760 | spin_unlock_irqrestore(&queue->tx_lock, flags); |
0d160211 | 761 | |
6ed10654 | 762 | return NETDEV_TX_OK; |
0d160211 JF |
763 | |
764 | drop: | |
09f75cd7 | 765 | dev->stats.tx_dropped++; |
979de8a0 | 766 | dev_kfree_skb_any(skb); |
6ed10654 | 767 | return NETDEV_TX_OK; |
0d160211 JF |
768 | } |
769 | ||
770 | static int xennet_close(struct net_device *dev) | |
771 | { | |
772 | struct netfront_info *np = netdev_priv(dev); | |
2688fcb7 AB |
773 | unsigned int num_queues = dev->real_num_tx_queues; |
774 | unsigned int i; | |
775 | struct netfront_queue *queue; | |
776 | netif_tx_stop_all_queues(np->netdev); | |
777 | for (i = 0; i < num_queues; ++i) { | |
778 | queue = &np->queues[i]; | |
779 | napi_disable(&queue->napi); | |
780 | } | |
0d160211 JF |
781 | return 0; |
782 | } | |
783 | ||
2688fcb7 | 784 | static void xennet_move_rx_slot(struct netfront_queue *queue, struct sk_buff *skb, |
0d160211 JF |
785 | grant_ref_t ref) |
786 | { | |
2688fcb7 AB |
787 | int new = xennet_rxidx(queue->rx.req_prod_pvt); |
788 | ||
789 | BUG_ON(queue->rx_skbs[new]); | |
790 | queue->rx_skbs[new] = skb; | |
791 | queue->grant_rx_ref[new] = ref; | |
792 | RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->id = new; | |
793 | RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->gref = ref; | |
794 | queue->rx.req_prod_pvt++; | |
0d160211 JF |
795 | } |
796 | ||
2688fcb7 | 797 | static int xennet_get_extras(struct netfront_queue *queue, |
0d160211 JF |
798 | struct xen_netif_extra_info *extras, |
799 | RING_IDX rp) | |
800 | ||
801 | { | |
8446066b | 802 | struct xen_netif_extra_info extra; |
2688fcb7 AB |
803 | struct device *dev = &queue->info->netdev->dev; |
804 | RING_IDX cons = queue->rx.rsp_cons; | |
0d160211 JF |
805 | int err = 0; |
806 | ||
807 | do { | |
808 | struct sk_buff *skb; | |
809 | grant_ref_t ref; | |
810 | ||
811 | if (unlikely(cons + 1 == rp)) { | |
812 | if (net_ratelimit()) | |
813 | dev_warn(dev, "Missing extra info\n"); | |
814 | err = -EBADR; | |
815 | break; | |
816 | } | |
817 | ||
8446066b | 818 | RING_COPY_RESPONSE(&queue->rx, ++cons, &extra); |
0d160211 | 819 | |
8446066b JG |
820 | if (unlikely(!extra.type || |
821 | extra.type >= XEN_NETIF_EXTRA_TYPE_MAX)) { | |
0d160211 JF |
822 | if (net_ratelimit()) |
823 | dev_warn(dev, "Invalid extra type: %d\n", | |
8446066b | 824 | extra.type); |
0d160211 JF |
825 | err = -EINVAL; |
826 | } else { | |
8446066b | 827 | extras[extra.type - 1] = extra; |
0d160211 JF |
828 | } |
829 | ||
2688fcb7 AB |
830 | skb = xennet_get_rx_skb(queue, cons); |
831 | ref = xennet_get_rx_ref(queue, cons); | |
832 | xennet_move_rx_slot(queue, skb, ref); | |
8446066b | 833 | } while (extra.flags & XEN_NETIF_EXTRA_FLAG_MORE); |
0d160211 | 834 | |
2688fcb7 | 835 | queue->rx.rsp_cons = cons; |
0d160211 JF |
836 | return err; |
837 | } | |
838 | ||
6c5aa6fc DK |
839 | static u32 xennet_run_xdp(struct netfront_queue *queue, struct page *pdata, |
840 | struct xen_netif_rx_response *rx, struct bpf_prog *prog, | |
841 | struct xdp_buff *xdp, bool *need_xdp_flush) | |
842 | { | |
843 | struct xdp_frame *xdpf; | |
844 | u32 len = rx->status; | |
e44f65fd | 845 | u32 act; |
6c5aa6fc DK |
846 | int err; |
847 | ||
43b5169d LB |
848 | xdp_init_buff(xdp, XEN_PAGE_SIZE - XDP_PACKET_HEADROOM, |
849 | &queue->xdp_rxq); | |
be9df4af LB |
850 | xdp_prepare_buff(xdp, page_address(pdata), XDP_PACKET_HEADROOM, |
851 | len, false); | |
6c5aa6fc DK |
852 | |
853 | act = bpf_prog_run_xdp(prog, xdp); | |
854 | switch (act) { | |
855 | case XDP_TX: | |
856 | get_page(pdata); | |
857 | xdpf = xdp_convert_buff_to_frame(xdp); | |
858 | err = xennet_xdp_xmit(queue->info->netdev, 1, &xdpf, 0); | |
fdc13979 LB |
859 | if (unlikely(!err)) |
860 | xdp_return_frame_rx_napi(xdpf); | |
861 | else if (unlikely(err < 0)) | |
6c5aa6fc DK |
862 | trace_xdp_exception(queue->info->netdev, prog, act); |
863 | break; | |
864 | case XDP_REDIRECT: | |
865 | get_page(pdata); | |
866 | err = xdp_do_redirect(queue->info->netdev, xdp, prog); | |
867 | *need_xdp_flush = true; | |
868 | if (unlikely(err)) | |
869 | trace_xdp_exception(queue->info->netdev, prog, act); | |
870 | break; | |
871 | case XDP_PASS: | |
872 | case XDP_DROP: | |
873 | break; | |
874 | ||
875 | case XDP_ABORTED: | |
876 | trace_xdp_exception(queue->info->netdev, prog, act); | |
877 | break; | |
878 | ||
879 | default: | |
880 | bpf_warn_invalid_xdp_action(act); | |
881 | } | |
882 | ||
883 | return act; | |
884 | } | |
885 | ||
2688fcb7 | 886 | static int xennet_get_responses(struct netfront_queue *queue, |
0d160211 | 887 | struct netfront_rx_info *rinfo, RING_IDX rp, |
6c5aa6fc DK |
888 | struct sk_buff_head *list, |
889 | bool *need_xdp_flush) | |
0d160211 | 890 | { |
8446066b | 891 | struct xen_netif_rx_response *rx = &rinfo->rx, rx_local; |
6c5aa6fc | 892 | int max = XEN_NETIF_NR_SLOTS_MIN + (rx->status <= RX_COPY_THRESHOLD); |
2688fcb7 AB |
893 | RING_IDX cons = queue->rx.rsp_cons; |
894 | struct sk_buff *skb = xennet_get_rx_skb(queue, cons); | |
6c5aa6fc | 895 | struct xen_netif_extra_info *extras = rinfo->extras; |
2688fcb7 | 896 | grant_ref_t ref = xennet_get_rx_ref(queue, cons); |
6c5aa6fc DK |
897 | struct device *dev = &queue->info->netdev->dev; |
898 | struct bpf_prog *xdp_prog; | |
899 | struct xdp_buff xdp; | |
900 | unsigned long ret; | |
7158ff6d | 901 | int slots = 1; |
0d160211 | 902 | int err = 0; |
6c5aa6fc | 903 | u32 verdict; |
0d160211 | 904 | |
f942dc25 | 905 | if (rx->flags & XEN_NETRXF_extra_info) { |
2688fcb7 | 906 | err = xennet_get_extras(queue, extras, rp); |
6c5aa6fc DK |
907 | if (!err) { |
908 | if (extras[XEN_NETIF_EXTRA_TYPE_XDP - 1].type) { | |
909 | struct xen_netif_extra_info *xdp; | |
910 | ||
911 | xdp = &extras[XEN_NETIF_EXTRA_TYPE_XDP - 1]; | |
912 | rx->offset = xdp->u.xdp.headroom; | |
913 | } | |
914 | } | |
2688fcb7 | 915 | cons = queue->rx.rsp_cons; |
0d160211 JF |
916 | } |
917 | ||
918 | for (;;) { | |
919 | if (unlikely(rx->status < 0 || | |
30c5d7f0 | 920 | rx->offset + rx->status > XEN_PAGE_SIZE)) { |
0d160211 | 921 | if (net_ratelimit()) |
6c10127d | 922 | dev_warn(dev, "rx->offset: %u, size: %d\n", |
0d160211 | 923 | rx->offset, rx->status); |
2688fcb7 | 924 | xennet_move_rx_slot(queue, skb, ref); |
0d160211 JF |
925 | err = -EINVAL; |
926 | goto next; | |
927 | } | |
928 | ||
929 | /* | |
930 | * This definitely indicates a bug, either in this driver or in | |
931 | * the backend driver. In future this should flag the bad | |
697089dc | 932 | * situation to the system controller to reboot the backend. |
0d160211 JF |
933 | */ |
934 | if (ref == GRANT_INVALID_REF) { | |
935 | if (net_ratelimit()) | |
936 | dev_warn(dev, "Bad rx response id %d.\n", | |
937 | rx->id); | |
938 | err = -EINVAL; | |
939 | goto next; | |
940 | } | |
941 | ||
942 | ret = gnttab_end_foreign_access_ref(ref, 0); | |
943 | BUG_ON(!ret); | |
944 | ||
2688fcb7 | 945 | gnttab_release_grant_reference(&queue->gref_rx_head, ref); |
0d160211 | 946 | |
6c5aa6fc DK |
947 | rcu_read_lock(); |
948 | xdp_prog = rcu_dereference(queue->xdp_prog); | |
949 | if (xdp_prog) { | |
950 | if (!(rx->flags & XEN_NETRXF_more_data)) { | |
951 | /* currently only a single page contains data */ | |
952 | verdict = xennet_run_xdp(queue, | |
953 | skb_frag_page(&skb_shinfo(skb)->frags[0]), | |
954 | rx, xdp_prog, &xdp, need_xdp_flush); | |
955 | if (verdict != XDP_PASS) | |
956 | err = -EINVAL; | |
957 | } else { | |
958 | /* drop the frame */ | |
959 | err = -EINVAL; | |
960 | } | |
961 | } | |
962 | rcu_read_unlock(); | |
0d160211 | 963 | next: |
6c5aa6fc | 964 | __skb_queue_tail(list, skb); |
f942dc25 | 965 | if (!(rx->flags & XEN_NETRXF_more_data)) |
0d160211 JF |
966 | break; |
967 | ||
7158ff6d | 968 | if (cons + slots == rp) { |
0d160211 | 969 | if (net_ratelimit()) |
7158ff6d | 970 | dev_warn(dev, "Need more slots\n"); |
0d160211 JF |
971 | err = -ENOENT; |
972 | break; | |
973 | } | |
974 | ||
8446066b JG |
975 | RING_COPY_RESPONSE(&queue->rx, cons + slots, &rx_local); |
976 | rx = &rx_local; | |
2688fcb7 AB |
977 | skb = xennet_get_rx_skb(queue, cons + slots); |
978 | ref = xennet_get_rx_ref(queue, cons + slots); | |
7158ff6d | 979 | slots++; |
0d160211 JF |
980 | } |
981 | ||
7158ff6d | 982 | if (unlikely(slots > max)) { |
0d160211 | 983 | if (net_ratelimit()) |
697089dc | 984 | dev_warn(dev, "Too many slots\n"); |
0d160211 JF |
985 | err = -E2BIG; |
986 | } | |
987 | ||
988 | if (unlikely(err)) | |
2688fcb7 | 989 | queue->rx.rsp_cons = cons + slots; |
0d160211 JF |
990 | |
991 | return err; | |
992 | } | |
993 | ||
994 | static int xennet_set_skb_gso(struct sk_buff *skb, | |
995 | struct xen_netif_extra_info *gso) | |
996 | { | |
997 | if (!gso->u.gso.size) { | |
998 | if (net_ratelimit()) | |
383eda32 | 999 | pr_warn("GSO size must not be zero\n"); |
0d160211 JF |
1000 | return -EINVAL; |
1001 | } | |
1002 | ||
2c0057de PD |
1003 | if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4 && |
1004 | gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV6) { | |
0d160211 | 1005 | if (net_ratelimit()) |
383eda32 | 1006 | pr_warn("Bad GSO type %d\n", gso->u.gso.type); |
0d160211 JF |
1007 | return -EINVAL; |
1008 | } | |
1009 | ||
1010 | skb_shinfo(skb)->gso_size = gso->u.gso.size; | |
2c0057de PD |
1011 | skb_shinfo(skb)->gso_type = |
1012 | (gso->u.gso.type == XEN_NETIF_GSO_TYPE_TCPV4) ? | |
1013 | SKB_GSO_TCPV4 : | |
1014 | SKB_GSO_TCPV6; | |
0d160211 JF |
1015 | |
1016 | /* Header must be checked, and gso_segs computed. */ | |
1017 | skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; | |
1018 | skb_shinfo(skb)->gso_segs = 0; | |
1019 | ||
1020 | return 0; | |
1021 | } | |
1022 | ||
a761129e DZ |
1023 | static int xennet_fill_frags(struct netfront_queue *queue, |
1024 | struct sk_buff *skb, | |
1025 | struct sk_buff_head *list) | |
0d160211 | 1026 | { |
2688fcb7 | 1027 | RING_IDX cons = queue->rx.rsp_cons; |
0d160211 JF |
1028 | struct sk_buff *nskb; |
1029 | ||
1030 | while ((nskb = __skb_dequeue(list))) { | |
8446066b | 1031 | struct xen_netif_rx_response rx; |
01c68026 | 1032 | skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0]; |
0d160211 | 1033 | |
8446066b JG |
1034 | RING_COPY_RESPONSE(&queue->rx, ++cons, &rx); |
1035 | ||
d472b3a6 | 1036 | if (skb_shinfo(skb)->nr_frags == MAX_SKB_FRAGS) { |
093b9c71 | 1037 | unsigned int pull_to = NETFRONT_SKB_CB(skb)->pull_to; |
0d160211 | 1038 | |
d81c5054 | 1039 | BUG_ON(pull_to < skb_headlen(skb)); |
093b9c71 JB |
1040 | __pskb_pull_tail(skb, pull_to - skb_headlen(skb)); |
1041 | } | |
ad4f15dc | 1042 | if (unlikely(skb_shinfo(skb)->nr_frags >= MAX_SKB_FRAGS)) { |
00b36850 | 1043 | queue->rx.rsp_cons = ++cons + skb_queue_len(list); |
ad4f15dc | 1044 | kfree_skb(nskb); |
a761129e | 1045 | return -ENOENT; |
ad4f15dc | 1046 | } |
093b9c71 | 1047 | |
d472b3a6 JG |
1048 | skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, |
1049 | skb_frag_page(nfrag), | |
8446066b | 1050 | rx.offset, rx.status, PAGE_SIZE); |
0d160211 JF |
1051 | |
1052 | skb_shinfo(nskb)->nr_frags = 0; | |
1053 | kfree_skb(nskb); | |
0d160211 JF |
1054 | } |
1055 | ||
a761129e DZ |
1056 | queue->rx.rsp_cons = cons; |
1057 | ||
1058 | return 0; | |
0d160211 JF |
1059 | } |
1060 | ||
e0ce4af9 | 1061 | static int checksum_setup(struct net_device *dev, struct sk_buff *skb) |
0d160211 | 1062 | { |
b5cf66cd | 1063 | bool recalculate_partial_csum = false; |
e0ce4af9 IC |
1064 | |
1065 | /* | |
1066 | * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy | |
1067 | * peers can fail to set NETRXF_csum_blank when sending a GSO | |
1068 | * frame. In this case force the SKB to CHECKSUM_PARTIAL and | |
1069 | * recalculate the partial checksum. | |
1070 | */ | |
1071 | if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) { | |
1072 | struct netfront_info *np = netdev_priv(dev); | |
2688fcb7 | 1073 | atomic_inc(&np->rx_gso_checksum_fixup); |
e0ce4af9 | 1074 | skb->ip_summed = CHECKSUM_PARTIAL; |
b5cf66cd | 1075 | recalculate_partial_csum = true; |
e0ce4af9 IC |
1076 | } |
1077 | ||
1078 | /* A non-CHECKSUM_PARTIAL SKB does not require setup. */ | |
1079 | if (skb->ip_summed != CHECKSUM_PARTIAL) | |
1080 | return 0; | |
0d160211 | 1081 | |
b5cf66cd | 1082 | return skb_checksum_setup(skb, recalculate_partial_csum); |
0d160211 JF |
1083 | } |
1084 | ||
2688fcb7 | 1085 | static int handle_incoming_queue(struct netfront_queue *queue, |
09f75cd7 | 1086 | struct sk_buff_head *rxq) |
0d160211 | 1087 | { |
900e1833 | 1088 | struct netfront_stats *rx_stats = this_cpu_ptr(queue->info->rx_stats); |
0d160211 JF |
1089 | int packets_dropped = 0; |
1090 | struct sk_buff *skb; | |
1091 | ||
1092 | while ((skb = __skb_dequeue(rxq)) != NULL) { | |
3683243b | 1093 | int pull_to = NETFRONT_SKB_CB(skb)->pull_to; |
0d160211 | 1094 | |
093b9c71 JB |
1095 | if (pull_to > skb_headlen(skb)) |
1096 | __pskb_pull_tail(skb, pull_to - skb_headlen(skb)); | |
0d160211 JF |
1097 | |
1098 | /* Ethernet work: Delayed to here as it peeks the header. */ | |
2688fcb7 | 1099 | skb->protocol = eth_type_trans(skb, queue->info->netdev); |
d554f73d | 1100 | skb_reset_network_header(skb); |
0d160211 | 1101 | |
2688fcb7 | 1102 | if (checksum_setup(queue->info->netdev, skb)) { |
e0ce4af9 IC |
1103 | kfree_skb(skb); |
1104 | packets_dropped++; | |
2688fcb7 | 1105 | queue->info->netdev->stats.rx_errors++; |
e0ce4af9 | 1106 | continue; |
0d160211 JF |
1107 | } |
1108 | ||
900e1833 DV |
1109 | u64_stats_update_begin(&rx_stats->syncp); |
1110 | rx_stats->packets++; | |
1111 | rx_stats->bytes += skb->len; | |
1112 | u64_stats_update_end(&rx_stats->syncp); | |
0d160211 JF |
1113 | |
1114 | /* Pass it up. */ | |
2688fcb7 | 1115 | napi_gro_receive(&queue->napi, skb); |
0d160211 JF |
1116 | } |
1117 | ||
1118 | return packets_dropped; | |
1119 | } | |
1120 | ||
bea3348e | 1121 | static int xennet_poll(struct napi_struct *napi, int budget) |
0d160211 | 1122 | { |
2688fcb7 AB |
1123 | struct netfront_queue *queue = container_of(napi, struct netfront_queue, napi); |
1124 | struct net_device *dev = queue->info->netdev; | |
0d160211 JF |
1125 | struct sk_buff *skb; |
1126 | struct netfront_rx_info rinfo; | |
1127 | struct xen_netif_rx_response *rx = &rinfo.rx; | |
1128 | struct xen_netif_extra_info *extras = rinfo.extras; | |
1129 | RING_IDX i, rp; | |
bea3348e | 1130 | int work_done; |
0d160211 JF |
1131 | struct sk_buff_head rxq; |
1132 | struct sk_buff_head errq; | |
1133 | struct sk_buff_head tmpq; | |
0d160211 | 1134 | int err; |
6c5aa6fc | 1135 | bool need_xdp_flush = false; |
0d160211 | 1136 | |
2688fcb7 | 1137 | spin_lock(&queue->rx_lock); |
0d160211 | 1138 | |
0d160211 JF |
1139 | skb_queue_head_init(&rxq); |
1140 | skb_queue_head_init(&errq); | |
1141 | skb_queue_head_init(&tmpq); | |
1142 | ||
2688fcb7 | 1143 | rp = queue->rx.sring->rsp_prod; |
0d160211 JF |
1144 | rmb(); /* Ensure we see queued responses up to 'rp'. */ |
1145 | ||
2688fcb7 | 1146 | i = queue->rx.rsp_cons; |
0d160211 JF |
1147 | work_done = 0; |
1148 | while ((i != rp) && (work_done < budget)) { | |
8446066b | 1149 | RING_COPY_RESPONSE(&queue->rx, i, rx); |
0d160211 JF |
1150 | memset(extras, 0, sizeof(rinfo.extras)); |
1151 | ||
6c5aa6fc DK |
1152 | err = xennet_get_responses(queue, &rinfo, rp, &tmpq, |
1153 | &need_xdp_flush); | |
0d160211 JF |
1154 | |
1155 | if (unlikely(err)) { | |
1156 | err: | |
1157 | while ((skb = __skb_dequeue(&tmpq))) | |
1158 | __skb_queue_tail(&errq, skb); | |
09f75cd7 | 1159 | dev->stats.rx_errors++; |
2688fcb7 | 1160 | i = queue->rx.rsp_cons; |
0d160211 JF |
1161 | continue; |
1162 | } | |
1163 | ||
1164 | skb = __skb_dequeue(&tmpq); | |
1165 | ||
1166 | if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) { | |
1167 | struct xen_netif_extra_info *gso; | |
1168 | gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1]; | |
1169 | ||
1170 | if (unlikely(xennet_set_skb_gso(skb, gso))) { | |
1171 | __skb_queue_head(&tmpq, skb); | |
2688fcb7 | 1172 | queue->rx.rsp_cons += skb_queue_len(&tmpq); |
0d160211 JF |
1173 | goto err; |
1174 | } | |
1175 | } | |
1176 | ||
3683243b IC |
1177 | NETFRONT_SKB_CB(skb)->pull_to = rx->status; |
1178 | if (NETFRONT_SKB_CB(skb)->pull_to > RX_COPY_THRESHOLD) | |
1179 | NETFRONT_SKB_CB(skb)->pull_to = RX_COPY_THRESHOLD; | |
0d160211 | 1180 | |
b54c9d5b | 1181 | skb_frag_off_set(&skb_shinfo(skb)->frags[0], rx->offset); |
3683243b IC |
1182 | skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status); |
1183 | skb->data_len = rx->status; | |
093b9c71 | 1184 | skb->len += rx->status; |
0d160211 | 1185 | |
a761129e | 1186 | if (unlikely(xennet_fill_frags(queue, skb, &tmpq))) |
ad4f15dc | 1187 | goto err; |
0d160211 | 1188 | |
f942dc25 | 1189 | if (rx->flags & XEN_NETRXF_csum_blank) |
0d160211 | 1190 | skb->ip_summed = CHECKSUM_PARTIAL; |
f942dc25 | 1191 | else if (rx->flags & XEN_NETRXF_data_validated) |
0d160211 JF |
1192 | skb->ip_summed = CHECKSUM_UNNECESSARY; |
1193 | ||
1194 | __skb_queue_tail(&rxq, skb); | |
1195 | ||
a761129e | 1196 | i = ++queue->rx.rsp_cons; |
0d160211 JF |
1197 | work_done++; |
1198 | } | |
6c5aa6fc DK |
1199 | if (need_xdp_flush) |
1200 | xdp_do_flush(); | |
0d160211 | 1201 | |
56cfe5d0 | 1202 | __skb_queue_purge(&errq); |
0d160211 | 1203 | |
2688fcb7 | 1204 | work_done -= handle_incoming_queue(queue, &rxq); |
0d160211 | 1205 | |
2688fcb7 | 1206 | xennet_alloc_rx_buffers(queue); |
0d160211 | 1207 | |
0d160211 | 1208 | if (work_done < budget) { |
bea3348e SH |
1209 | int more_to_do = 0; |
1210 | ||
6ad20165 | 1211 | napi_complete_done(napi, work_done); |
0d160211 | 1212 | |
2688fcb7 | 1213 | RING_FINAL_CHECK_FOR_RESPONSES(&queue->rx, more_to_do); |
6a6dc08f DV |
1214 | if (more_to_do) |
1215 | napi_schedule(napi); | |
0d160211 JF |
1216 | } |
1217 | ||
2688fcb7 | 1218 | spin_unlock(&queue->rx_lock); |
0d160211 | 1219 | |
bea3348e | 1220 | return work_done; |
0d160211 JF |
1221 | } |
1222 | ||
1223 | static int xennet_change_mtu(struct net_device *dev, int mtu) | |
1224 | { | |
0c36820e | 1225 | int max = xennet_can_sg(dev) ? XEN_NETIF_MAX_TX_SIZE : ETH_DATA_LEN; |
0d160211 JF |
1226 | |
1227 | if (mtu > max) | |
1228 | return -EINVAL; | |
1229 | dev->mtu = mtu; | |
1230 | return 0; | |
1231 | } | |
1232 | ||
bc1f4470 | 1233 | static void xennet_get_stats64(struct net_device *dev, |
1234 | struct rtnl_link_stats64 *tot) | |
e00f85be | 1235 | { |
1236 | struct netfront_info *np = netdev_priv(dev); | |
1237 | int cpu; | |
1238 | ||
1239 | for_each_possible_cpu(cpu) { | |
900e1833 DV |
1240 | struct netfront_stats *rx_stats = per_cpu_ptr(np->rx_stats, cpu); |
1241 | struct netfront_stats *tx_stats = per_cpu_ptr(np->tx_stats, cpu); | |
e00f85be | 1242 | u64 rx_packets, rx_bytes, tx_packets, tx_bytes; |
1243 | unsigned int start; | |
1244 | ||
1245 | do { | |
900e1833 DV |
1246 | start = u64_stats_fetch_begin_irq(&tx_stats->syncp); |
1247 | tx_packets = tx_stats->packets; | |
1248 | tx_bytes = tx_stats->bytes; | |
1249 | } while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start)); | |
e00f85be | 1250 | |
900e1833 DV |
1251 | do { |
1252 | start = u64_stats_fetch_begin_irq(&rx_stats->syncp); | |
1253 | rx_packets = rx_stats->packets; | |
1254 | rx_bytes = rx_stats->bytes; | |
1255 | } while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start)); | |
e00f85be | 1256 | |
1257 | tot->rx_packets += rx_packets; | |
1258 | tot->tx_packets += tx_packets; | |
1259 | tot->rx_bytes += rx_bytes; | |
1260 | tot->tx_bytes += tx_bytes; | |
1261 | } | |
1262 | ||
1263 | tot->rx_errors = dev->stats.rx_errors; | |
1264 | tot->tx_dropped = dev->stats.tx_dropped; | |
e00f85be | 1265 | } |
1266 | ||
2688fcb7 | 1267 | static void xennet_release_tx_bufs(struct netfront_queue *queue) |
0d160211 JF |
1268 | { |
1269 | struct sk_buff *skb; | |
1270 | int i; | |
1271 | ||
1272 | for (i = 0; i < NET_TX_RING_SIZE; i++) { | |
1273 | /* Skip over entries which are actually freelist references */ | |
21631d2d | 1274 | if (!queue->tx_skbs[i]) |
0d160211 JF |
1275 | continue; |
1276 | ||
21631d2d JG |
1277 | skb = queue->tx_skbs[i]; |
1278 | queue->tx_skbs[i] = NULL; | |
2688fcb7 AB |
1279 | get_page(queue->grant_tx_page[i]); |
1280 | gnttab_end_foreign_access(queue->grant_tx_ref[i], | |
cefe0078 | 1281 | GNTMAP_readonly, |
2688fcb7 AB |
1282 | (unsigned long)page_address(queue->grant_tx_page[i])); |
1283 | queue->grant_tx_page[i] = NULL; | |
1284 | queue->grant_tx_ref[i] = GRANT_INVALID_REF; | |
21631d2d | 1285 | add_id_to_list(&queue->tx_skb_freelist, queue->tx_link, i); |
0d160211 JF |
1286 | dev_kfree_skb_irq(skb); |
1287 | } | |
1288 | } | |
1289 | ||
2688fcb7 | 1290 | static void xennet_release_rx_bufs(struct netfront_queue *queue) |
0d160211 | 1291 | { |
0d160211 JF |
1292 | int id, ref; |
1293 | ||
2688fcb7 | 1294 | spin_lock_bh(&queue->rx_lock); |
0d160211 JF |
1295 | |
1296 | for (id = 0; id < NET_RX_RING_SIZE; id++) { | |
cefe0078 AL |
1297 | struct sk_buff *skb; |
1298 | struct page *page; | |
0d160211 | 1299 | |
2688fcb7 | 1300 | skb = queue->rx_skbs[id]; |
cefe0078 | 1301 | if (!skb) |
0d160211 | 1302 | continue; |
0d160211 | 1303 | |
2688fcb7 | 1304 | ref = queue->grant_rx_ref[id]; |
cefe0078 AL |
1305 | if (ref == GRANT_INVALID_REF) |
1306 | continue; | |
0d160211 | 1307 | |
cefe0078 | 1308 | page = skb_frag_page(&skb_shinfo(skb)->frags[0]); |
0d160211 | 1309 | |
cefe0078 AL |
1310 | /* gnttab_end_foreign_access() needs a page ref until |
1311 | * foreign access is ended (which may be deferred). | |
1312 | */ | |
1313 | get_page(page); | |
1314 | gnttab_end_foreign_access(ref, 0, | |
1315 | (unsigned long)page_address(page)); | |
2688fcb7 | 1316 | queue->grant_rx_ref[id] = GRANT_INVALID_REF; |
0d160211 | 1317 | |
cefe0078 | 1318 | kfree_skb(skb); |
0d160211 JF |
1319 | } |
1320 | ||
2688fcb7 | 1321 | spin_unlock_bh(&queue->rx_lock); |
0d160211 JF |
1322 | } |
1323 | ||
c8f44aff MM |
1324 | static netdev_features_t xennet_fix_features(struct net_device *dev, |
1325 | netdev_features_t features) | |
8f7b01a1 ED |
1326 | { |
1327 | struct netfront_info *np = netdev_priv(dev); | |
8f7b01a1 | 1328 | |
2890ea5c JG |
1329 | if (features & NETIF_F_SG && |
1330 | !xenbus_read_unsigned(np->xbdev->otherend, "feature-sg", 0)) | |
1331 | features &= ~NETIF_F_SG; | |
8f7b01a1 | 1332 | |
2890ea5c JG |
1333 | if (features & NETIF_F_IPV6_CSUM && |
1334 | !xenbus_read_unsigned(np->xbdev->otherend, | |
1335 | "feature-ipv6-csum-offload", 0)) | |
1336 | features &= ~NETIF_F_IPV6_CSUM; | |
8f7b01a1 | 1337 | |
2890ea5c JG |
1338 | if (features & NETIF_F_TSO && |
1339 | !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv4", 0)) | |
1340 | features &= ~NETIF_F_TSO; | |
8f7b01a1 | 1341 | |
2890ea5c JG |
1342 | if (features & NETIF_F_TSO6 && |
1343 | !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv6", 0)) | |
1344 | features &= ~NETIF_F_TSO6; | |
2c0057de | 1345 | |
8f7b01a1 ED |
1346 | return features; |
1347 | } | |
1348 | ||
c8f44aff MM |
1349 | static int xennet_set_features(struct net_device *dev, |
1350 | netdev_features_t features) | |
8f7b01a1 ED |
1351 | { |
1352 | if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) { | |
1353 | netdev_info(dev, "Reducing MTU because no SG offload"); | |
1354 | dev->mtu = ETH_DATA_LEN; | |
1355 | } | |
1356 | ||
1357 | return 0; | |
1358 | } | |
1359 | ||
d634bf2c | 1360 | static irqreturn_t xennet_tx_interrupt(int irq, void *dev_id) |
cf66f9d4 | 1361 | { |
2688fcb7 | 1362 | struct netfront_queue *queue = dev_id; |
cf66f9d4 KRW |
1363 | unsigned long flags; |
1364 | ||
2688fcb7 AB |
1365 | spin_lock_irqsave(&queue->tx_lock, flags); |
1366 | xennet_tx_buf_gc(queue); | |
1367 | spin_unlock_irqrestore(&queue->tx_lock, flags); | |
cf66f9d4 | 1368 | |
d634bf2c WL |
1369 | return IRQ_HANDLED; |
1370 | } | |
1371 | ||
1372 | static irqreturn_t xennet_rx_interrupt(int irq, void *dev_id) | |
1373 | { | |
2688fcb7 AB |
1374 | struct netfront_queue *queue = dev_id; |
1375 | struct net_device *dev = queue->info->netdev; | |
d634bf2c WL |
1376 | |
1377 | if (likely(netif_carrier_ok(dev) && | |
2688fcb7 | 1378 | RING_HAS_UNCONSUMED_RESPONSES(&queue->rx))) |
76541869 | 1379 | napi_schedule(&queue->napi); |
cf66f9d4 | 1380 | |
d634bf2c WL |
1381 | return IRQ_HANDLED; |
1382 | } | |
cf66f9d4 | 1383 | |
d634bf2c WL |
1384 | static irqreturn_t xennet_interrupt(int irq, void *dev_id) |
1385 | { | |
1386 | xennet_tx_interrupt(irq, dev_id); | |
1387 | xennet_rx_interrupt(irq, dev_id); | |
cf66f9d4 KRW |
1388 | return IRQ_HANDLED; |
1389 | } | |
1390 | ||
1391 | #ifdef CONFIG_NET_POLL_CONTROLLER | |
1392 | static void xennet_poll_controller(struct net_device *dev) | |
1393 | { | |
2688fcb7 AB |
1394 | /* Poll each queue */ |
1395 | struct netfront_info *info = netdev_priv(dev); | |
1396 | unsigned int num_queues = dev->real_num_tx_queues; | |
1397 | unsigned int i; | |
1398 | for (i = 0; i < num_queues; ++i) | |
1399 | xennet_interrupt(0, &info->queues[i]); | |
cf66f9d4 KRW |
1400 | } |
1401 | #endif | |
1402 | ||
6c5aa6fc DK |
1403 | #define NETBACK_XDP_HEADROOM_DISABLE 0 |
1404 | #define NETBACK_XDP_HEADROOM_ENABLE 1 | |
1405 | ||
1406 | static int talk_to_netback_xdp(struct netfront_info *np, int xdp) | |
1407 | { | |
1408 | int err; | |
1409 | unsigned short headroom; | |
1410 | ||
1411 | headroom = xdp ? XDP_PACKET_HEADROOM : 0; | |
1412 | err = xenbus_printf(XBT_NIL, np->xbdev->nodename, | |
1413 | "xdp-headroom", "%hu", | |
1414 | headroom); | |
1415 | if (err) | |
1416 | pr_warn("Error writing xdp-headroom\n"); | |
1417 | ||
1418 | return err; | |
1419 | } | |
1420 | ||
1421 | static int xennet_xdp_set(struct net_device *dev, struct bpf_prog *prog, | |
1422 | struct netlink_ext_ack *extack) | |
1423 | { | |
1424 | unsigned long max_mtu = XEN_PAGE_SIZE - XDP_PACKET_HEADROOM; | |
1425 | struct netfront_info *np = netdev_priv(dev); | |
1426 | struct bpf_prog *old_prog; | |
1427 | unsigned int i, err; | |
1428 | ||
1429 | if (dev->mtu > max_mtu) { | |
1430 | netdev_warn(dev, "XDP requires MTU less than %lu\n", max_mtu); | |
1431 | return -EINVAL; | |
1432 | } | |
1433 | ||
1434 | if (!np->netback_has_xdp_headroom) | |
1435 | return 0; | |
1436 | ||
1437 | xenbus_switch_state(np->xbdev, XenbusStateReconfiguring); | |
1438 | ||
1439 | err = talk_to_netback_xdp(np, prog ? NETBACK_XDP_HEADROOM_ENABLE : | |
1440 | NETBACK_XDP_HEADROOM_DISABLE); | |
1441 | if (err) | |
1442 | return err; | |
1443 | ||
1444 | /* avoid the race with XDP headroom adjustment */ | |
1445 | wait_event(module_wq, | |
1446 | xenbus_read_driver_state(np->xbdev->otherend) == | |
1447 | XenbusStateReconfigured); | |
1448 | np->netfront_xdp_enabled = true; | |
1449 | ||
1450 | old_prog = rtnl_dereference(np->queues[0].xdp_prog); | |
1451 | ||
1452 | if (prog) | |
1453 | bpf_prog_add(prog, dev->real_num_tx_queues); | |
1454 | ||
1455 | for (i = 0; i < dev->real_num_tx_queues; ++i) | |
1456 | rcu_assign_pointer(np->queues[i].xdp_prog, prog); | |
1457 | ||
1458 | if (old_prog) | |
1459 | for (i = 0; i < dev->real_num_tx_queues; ++i) | |
1460 | bpf_prog_put(old_prog); | |
1461 | ||
1462 | xenbus_switch_state(np->xbdev, XenbusStateConnected); | |
1463 | ||
1464 | return 0; | |
1465 | } | |
1466 | ||
6c5aa6fc DK |
1467 | static int xennet_xdp(struct net_device *dev, struct netdev_bpf *xdp) |
1468 | { | |
1469 | switch (xdp->command) { | |
1470 | case XDP_SETUP_PROG: | |
1471 | return xennet_xdp_set(dev, xdp->prog, xdp->extack); | |
6c5aa6fc DK |
1472 | default: |
1473 | return -EINVAL; | |
1474 | } | |
1475 | } | |
1476 | ||
0a0b9d2e SH |
1477 | static const struct net_device_ops xennet_netdev_ops = { |
1478 | .ndo_open = xennet_open, | |
0a0b9d2e SH |
1479 | .ndo_stop = xennet_close, |
1480 | .ndo_start_xmit = xennet_start_xmit, | |
1481 | .ndo_change_mtu = xennet_change_mtu, | |
e00f85be | 1482 | .ndo_get_stats64 = xennet_get_stats64, |
0a0b9d2e SH |
1483 | .ndo_set_mac_address = eth_mac_addr, |
1484 | .ndo_validate_addr = eth_validate_addr, | |
fb507934 MM |
1485 | .ndo_fix_features = xennet_fix_features, |
1486 | .ndo_set_features = xennet_set_features, | |
2688fcb7 | 1487 | .ndo_select_queue = xennet_select_queue, |
6c5aa6fc DK |
1488 | .ndo_bpf = xennet_xdp, |
1489 | .ndo_xdp_xmit = xennet_xdp_xmit, | |
cf66f9d4 KRW |
1490 | #ifdef CONFIG_NET_POLL_CONTROLLER |
1491 | .ndo_poll_controller = xennet_poll_controller, | |
1492 | #endif | |
0a0b9d2e SH |
1493 | }; |
1494 | ||
900e1833 DV |
1495 | static void xennet_free_netdev(struct net_device *netdev) |
1496 | { | |
1497 | struct netfront_info *np = netdev_priv(netdev); | |
1498 | ||
1499 | free_percpu(np->rx_stats); | |
1500 | free_percpu(np->tx_stats); | |
1501 | free_netdev(netdev); | |
1502 | } | |
1503 | ||
8e0e46bb | 1504 | static struct net_device *xennet_create_dev(struct xenbus_device *dev) |
0d160211 | 1505 | { |
2688fcb7 | 1506 | int err; |
0d160211 JF |
1507 | struct net_device *netdev; |
1508 | struct netfront_info *np; | |
1509 | ||
50ee6061 | 1510 | netdev = alloc_etherdev_mq(sizeof(struct netfront_info), xennet_max_queues); |
41de8d4c | 1511 | if (!netdev) |
0d160211 | 1512 | return ERR_PTR(-ENOMEM); |
0d160211 JF |
1513 | |
1514 | np = netdev_priv(netdev); | |
1515 | np->xbdev = dev; | |
1516 | ||
2688fcb7 | 1517 | np->queues = NULL; |
0d160211 | 1518 | |
e00f85be | 1519 | err = -ENOMEM; |
900e1833 DV |
1520 | np->rx_stats = netdev_alloc_pcpu_stats(struct netfront_stats); |
1521 | if (np->rx_stats == NULL) | |
1522 | goto exit; | |
1523 | np->tx_stats = netdev_alloc_pcpu_stats(struct netfront_stats); | |
1524 | if (np->tx_stats == NULL) | |
e00f85be | 1525 | goto exit; |
1526 | ||
0a0b9d2e SH |
1527 | netdev->netdev_ops = &xennet_netdev_ops; |
1528 | ||
fb507934 MM |
1529 | netdev->features = NETIF_F_IP_CSUM | NETIF_F_RXCSUM | |
1530 | NETIF_F_GSO_ROBUST; | |
2c0057de PD |
1531 | netdev->hw_features = NETIF_F_SG | |
1532 | NETIF_F_IPV6_CSUM | | |
1533 | NETIF_F_TSO | NETIF_F_TSO6; | |
0d160211 | 1534 | |
fc3e5941 IC |
1535 | /* |
1536 | * Assume that all hw features are available for now. This set | |
1537 | * will be adjusted by the call to netdev_update_features() in | |
1538 | * xennet_connect() which is the earliest point where we can | |
1539 | * negotiate with the backend regarding supported features. | |
1540 | */ | |
1541 | netdev->features |= netdev->hw_features; | |
1542 | ||
7ad24ea4 | 1543 | netdev->ethtool_ops = &xennet_ethtool_ops; |
e1043a4b | 1544 | netdev->min_mtu = ETH_MIN_MTU; |
d0c2c997 | 1545 | netdev->max_mtu = XEN_NETIF_MAX_TX_SIZE; |
0d160211 JF |
1546 | SET_NETDEV_DEV(netdev, &dev->dev); |
1547 | ||
1548 | np->netdev = netdev; | |
6c5aa6fc | 1549 | np->netfront_xdp_enabled = false; |
0d160211 JF |
1550 | |
1551 | netif_carrier_off(netdev); | |
1552 | ||
c2c63310 AR |
1553 | do { |
1554 | xenbus_switch_state(dev, XenbusStateInitialising); | |
1555 | err = wait_event_timeout(module_wq, | |
1556 | xenbus_read_driver_state(dev->otherend) != | |
1557 | XenbusStateClosed && | |
1558 | xenbus_read_driver_state(dev->otherend) != | |
1559 | XenbusStateUnknown, XENNET_TIMEOUT); | |
1560 | } while (!err); | |
1561 | ||
0d160211 JF |
1562 | return netdev; |
1563 | ||
0d160211 | 1564 | exit: |
900e1833 | 1565 | xennet_free_netdev(netdev); |
0d160211 JF |
1566 | return ERR_PTR(err); |
1567 | } | |
1568 | ||
80708602 | 1569 | /* |
0d160211 JF |
1570 | * Entry point to this code when a new device is created. Allocate the basic |
1571 | * structures and the ring buffers for communication with the backend, and | |
1572 | * inform the backend of the appropriate details for those. | |
1573 | */ | |
8e0e46bb | 1574 | static int netfront_probe(struct xenbus_device *dev, |
1dd06ae8 | 1575 | const struct xenbus_device_id *id) |
0d160211 JF |
1576 | { |
1577 | int err; | |
1578 | struct net_device *netdev; | |
1579 | struct netfront_info *info; | |
1580 | ||
1581 | netdev = xennet_create_dev(dev); | |
1582 | if (IS_ERR(netdev)) { | |
1583 | err = PTR_ERR(netdev); | |
1584 | xenbus_dev_fatal(dev, err, "creating netdev"); | |
1585 | return err; | |
1586 | } | |
1587 | ||
1588 | info = netdev_priv(netdev); | |
1b713e00 | 1589 | dev_set_drvdata(&dev->dev, info); |
27b917e5 TI |
1590 | #ifdef CONFIG_SYSFS |
1591 | info->netdev->sysfs_groups[0] = &xennet_dev_group; | |
1592 | #endif | |
0d160211 | 1593 | |
0d160211 | 1594 | return 0; |
0d160211 JF |
1595 | } |
1596 | ||
1597 | static void xennet_end_access(int ref, void *page) | |
1598 | { | |
1599 | /* This frees the page as a side-effect */ | |
1600 | if (ref != GRANT_INVALID_REF) | |
1601 | gnttab_end_foreign_access(ref, 0, (unsigned long)page); | |
1602 | } | |
1603 | ||
1604 | static void xennet_disconnect_backend(struct netfront_info *info) | |
1605 | { | |
2688fcb7 | 1606 | unsigned int i = 0; |
2688fcb7 AB |
1607 | unsigned int num_queues = info->netdev->real_num_tx_queues; |
1608 | ||
f9feb1e6 DV |
1609 | netif_carrier_off(info->netdev); |
1610 | ||
9a873c71 | 1611 | for (i = 0; i < num_queues && info->queues; ++i) { |
76541869 DV |
1612 | struct netfront_queue *queue = &info->queues[i]; |
1613 | ||
74470954 BO |
1614 | del_timer_sync(&queue->rx_refill_timer); |
1615 | ||
2688fcb7 AB |
1616 | if (queue->tx_irq && (queue->tx_irq == queue->rx_irq)) |
1617 | unbind_from_irqhandler(queue->tx_irq, queue); | |
1618 | if (queue->tx_irq && (queue->tx_irq != queue->rx_irq)) { | |
1619 | unbind_from_irqhandler(queue->tx_irq, queue); | |
1620 | unbind_from_irqhandler(queue->rx_irq, queue); | |
1621 | } | |
1622 | queue->tx_evtchn = queue->rx_evtchn = 0; | |
1623 | queue->tx_irq = queue->rx_irq = 0; | |
0d160211 | 1624 | |
274b0455 CW |
1625 | if (netif_running(info->netdev)) |
1626 | napi_synchronize(&queue->napi); | |
f9feb1e6 | 1627 | |
a5b5dc3c DV |
1628 | xennet_release_tx_bufs(queue); |
1629 | xennet_release_rx_bufs(queue); | |
1630 | gnttab_free_grant_references(queue->gref_tx_head); | |
1631 | gnttab_free_grant_references(queue->gref_rx_head); | |
1632 | ||
2688fcb7 AB |
1633 | /* End access and free the pages */ |
1634 | xennet_end_access(queue->tx_ring_ref, queue->tx.sring); | |
1635 | xennet_end_access(queue->rx_ring_ref, queue->rx.sring); | |
0d160211 | 1636 | |
2688fcb7 AB |
1637 | queue->tx_ring_ref = GRANT_INVALID_REF; |
1638 | queue->rx_ring_ref = GRANT_INVALID_REF; | |
1639 | queue->tx.sring = NULL; | |
1640 | queue->rx.sring = NULL; | |
6c5aa6fc DK |
1641 | |
1642 | page_pool_destroy(queue->page_pool); | |
2688fcb7 | 1643 | } |
0d160211 JF |
1644 | } |
1645 | ||
80708602 | 1646 | /* |
0d160211 JF |
1647 | * We are reconnecting to the backend, due to a suspend/resume, or a backend |
1648 | * driver restart. We tear down our netif structure and recreate it, but | |
1649 | * leave the device-layer structures intact so that this is transparent to the | |
1650 | * rest of the kernel. | |
1651 | */ | |
1652 | static int netfront_resume(struct xenbus_device *dev) | |
1653 | { | |
1b713e00 | 1654 | struct netfront_info *info = dev_get_drvdata(&dev->dev); |
0d160211 JF |
1655 | |
1656 | dev_dbg(&dev->dev, "%s\n", dev->nodename); | |
1657 | ||
1658 | xennet_disconnect_backend(info); | |
1659 | return 0; | |
1660 | } | |
1661 | ||
1662 | static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[]) | |
1663 | { | |
1664 | char *s, *e, *macstr; | |
1665 | int i; | |
1666 | ||
1667 | macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL); | |
1668 | if (IS_ERR(macstr)) | |
1669 | return PTR_ERR(macstr); | |
1670 | ||
1671 | for (i = 0; i < ETH_ALEN; i++) { | |
1672 | mac[i] = simple_strtoul(s, &e, 16); | |
1673 | if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) { | |
1674 | kfree(macstr); | |
1675 | return -ENOENT; | |
1676 | } | |
1677 | s = e+1; | |
1678 | } | |
1679 | ||
1680 | kfree(macstr); | |
1681 | return 0; | |
1682 | } | |
1683 | ||
2688fcb7 | 1684 | static int setup_netfront_single(struct netfront_queue *queue) |
d634bf2c WL |
1685 | { |
1686 | int err; | |
1687 | ||
2688fcb7 | 1688 | err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn); |
d634bf2c WL |
1689 | if (err < 0) |
1690 | goto fail; | |
1691 | ||
2688fcb7 | 1692 | err = bind_evtchn_to_irqhandler(queue->tx_evtchn, |
d634bf2c | 1693 | xennet_interrupt, |
2688fcb7 | 1694 | 0, queue->info->netdev->name, queue); |
d634bf2c WL |
1695 | if (err < 0) |
1696 | goto bind_fail; | |
2688fcb7 AB |
1697 | queue->rx_evtchn = queue->tx_evtchn; |
1698 | queue->rx_irq = queue->tx_irq = err; | |
d634bf2c WL |
1699 | |
1700 | return 0; | |
1701 | ||
1702 | bind_fail: | |
2688fcb7 AB |
1703 | xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn); |
1704 | queue->tx_evtchn = 0; | |
d634bf2c WL |
1705 | fail: |
1706 | return err; | |
1707 | } | |
1708 | ||
2688fcb7 | 1709 | static int setup_netfront_split(struct netfront_queue *queue) |
d634bf2c WL |
1710 | { |
1711 | int err; | |
1712 | ||
2688fcb7 | 1713 | err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn); |
d634bf2c WL |
1714 | if (err < 0) |
1715 | goto fail; | |
2688fcb7 | 1716 | err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->rx_evtchn); |
d634bf2c WL |
1717 | if (err < 0) |
1718 | goto alloc_rx_evtchn_fail; | |
1719 | ||
2688fcb7 AB |
1720 | snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name), |
1721 | "%s-tx", queue->name); | |
1722 | err = bind_evtchn_to_irqhandler(queue->tx_evtchn, | |
d634bf2c | 1723 | xennet_tx_interrupt, |
2688fcb7 | 1724 | 0, queue->tx_irq_name, queue); |
d634bf2c WL |
1725 | if (err < 0) |
1726 | goto bind_tx_fail; | |
2688fcb7 | 1727 | queue->tx_irq = err; |
d634bf2c | 1728 | |
2688fcb7 AB |
1729 | snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name), |
1730 | "%s-rx", queue->name); | |
1731 | err = bind_evtchn_to_irqhandler(queue->rx_evtchn, | |
d634bf2c | 1732 | xennet_rx_interrupt, |
2688fcb7 | 1733 | 0, queue->rx_irq_name, queue); |
d634bf2c WL |
1734 | if (err < 0) |
1735 | goto bind_rx_fail; | |
2688fcb7 | 1736 | queue->rx_irq = err; |
d634bf2c WL |
1737 | |
1738 | return 0; | |
1739 | ||
1740 | bind_rx_fail: | |
2688fcb7 AB |
1741 | unbind_from_irqhandler(queue->tx_irq, queue); |
1742 | queue->tx_irq = 0; | |
d634bf2c | 1743 | bind_tx_fail: |
2688fcb7 AB |
1744 | xenbus_free_evtchn(queue->info->xbdev, queue->rx_evtchn); |
1745 | queue->rx_evtchn = 0; | |
d634bf2c | 1746 | alloc_rx_evtchn_fail: |
2688fcb7 AB |
1747 | xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn); |
1748 | queue->tx_evtchn = 0; | |
d634bf2c WL |
1749 | fail: |
1750 | return err; | |
1751 | } | |
1752 | ||
2688fcb7 AB |
1753 | static int setup_netfront(struct xenbus_device *dev, |
1754 | struct netfront_queue *queue, unsigned int feature_split_evtchn) | |
0d160211 JF |
1755 | { |
1756 | struct xen_netif_tx_sring *txs; | |
1757 | struct xen_netif_rx_sring *rxs; | |
ccc9d90a | 1758 | grant_ref_t gref; |
0d160211 | 1759 | int err; |
0d160211 | 1760 | |
2688fcb7 AB |
1761 | queue->tx_ring_ref = GRANT_INVALID_REF; |
1762 | queue->rx_ring_ref = GRANT_INVALID_REF; | |
1763 | queue->rx.sring = NULL; | |
1764 | queue->tx.sring = NULL; | |
0d160211 | 1765 | |
a144ff09 | 1766 | txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH); |
0d160211 JF |
1767 | if (!txs) { |
1768 | err = -ENOMEM; | |
1769 | xenbus_dev_fatal(dev, err, "allocating tx ring page"); | |
1770 | goto fail; | |
1771 | } | |
1772 | SHARED_RING_INIT(txs); | |
30c5d7f0 | 1773 | FRONT_RING_INIT(&queue->tx, txs, XEN_PAGE_SIZE); |
0d160211 | 1774 | |
ccc9d90a | 1775 | err = xenbus_grant_ring(dev, txs, 1, &gref); |
1ca2983a WL |
1776 | if (err < 0) |
1777 | goto grant_tx_ring_fail; | |
ccc9d90a | 1778 | queue->tx_ring_ref = gref; |
0d160211 | 1779 | |
a144ff09 | 1780 | rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH); |
0d160211 JF |
1781 | if (!rxs) { |
1782 | err = -ENOMEM; | |
1783 | xenbus_dev_fatal(dev, err, "allocating rx ring page"); | |
1ca2983a | 1784 | goto alloc_rx_ring_fail; |
0d160211 JF |
1785 | } |
1786 | SHARED_RING_INIT(rxs); | |
30c5d7f0 | 1787 | FRONT_RING_INIT(&queue->rx, rxs, XEN_PAGE_SIZE); |
0d160211 | 1788 | |
ccc9d90a | 1789 | err = xenbus_grant_ring(dev, rxs, 1, &gref); |
1ca2983a WL |
1790 | if (err < 0) |
1791 | goto grant_rx_ring_fail; | |
ccc9d90a | 1792 | queue->rx_ring_ref = gref; |
0d160211 | 1793 | |
d634bf2c | 1794 | if (feature_split_evtchn) |
2688fcb7 | 1795 | err = setup_netfront_split(queue); |
d634bf2c WL |
1796 | /* setup single event channel if |
1797 | * a) feature-split-event-channels == 0 | |
1798 | * b) feature-split-event-channels == 1 but failed to setup | |
1799 | */ | |
e93fac3b | 1800 | if (!feature_split_evtchn || err) |
2688fcb7 | 1801 | err = setup_netfront_single(queue); |
d634bf2c | 1802 | |
0d160211 | 1803 | if (err) |
1ca2983a | 1804 | goto alloc_evtchn_fail; |
0d160211 | 1805 | |
0d160211 JF |
1806 | return 0; |
1807 | ||
1ca2983a WL |
1808 | /* If we fail to setup netfront, it is safe to just revoke access to |
1809 | * granted pages because backend is not accessing it at this point. | |
1810 | */ | |
1ca2983a | 1811 | alloc_evtchn_fail: |
2688fcb7 | 1812 | gnttab_end_foreign_access_ref(queue->rx_ring_ref, 0); |
1ca2983a WL |
1813 | grant_rx_ring_fail: |
1814 | free_page((unsigned long)rxs); | |
1815 | alloc_rx_ring_fail: | |
2688fcb7 | 1816 | gnttab_end_foreign_access_ref(queue->tx_ring_ref, 0); |
1ca2983a WL |
1817 | grant_tx_ring_fail: |
1818 | free_page((unsigned long)txs); | |
1819 | fail: | |
0d160211 JF |
1820 | return err; |
1821 | } | |
1822 | ||
2688fcb7 AB |
1823 | /* Queue-specific initialisation |
1824 | * This used to be done in xennet_create_dev() but must now | |
1825 | * be run per-queue. | |
1826 | */ | |
1827 | static int xennet_init_queue(struct netfront_queue *queue) | |
1828 | { | |
1829 | unsigned short i; | |
1830 | int err = 0; | |
21f2706b | 1831 | char *devid; |
2688fcb7 AB |
1832 | |
1833 | spin_lock_init(&queue->tx_lock); | |
1834 | spin_lock_init(&queue->rx_lock); | |
1835 | ||
e99e88a9 | 1836 | timer_setup(&queue->rx_refill_timer, rx_refill_timeout, 0); |
2688fcb7 | 1837 | |
21f2706b XL |
1838 | devid = strrchr(queue->info->xbdev->nodename, '/') + 1; |
1839 | snprintf(queue->name, sizeof(queue->name), "vif%s-q%u", | |
1840 | devid, queue->id); | |
8b715010 | 1841 | |
21631d2d | 1842 | /* Initialise tx_skb_freelist as a free chain containing every entry. */ |
2688fcb7 AB |
1843 | queue->tx_skb_freelist = 0; |
1844 | for (i = 0; i < NET_TX_RING_SIZE; i++) { | |
21631d2d | 1845 | queue->tx_link[i] = i + 1; |
2688fcb7 AB |
1846 | queue->grant_tx_ref[i] = GRANT_INVALID_REF; |
1847 | queue->grant_tx_page[i] = NULL; | |
1848 | } | |
21631d2d | 1849 | queue->tx_link[NET_TX_RING_SIZE - 1] = TX_LINK_NONE; |
2688fcb7 AB |
1850 | |
1851 | /* Clear out rx_skbs */ | |
1852 | for (i = 0; i < NET_RX_RING_SIZE; i++) { | |
1853 | queue->rx_skbs[i] = NULL; | |
1854 | queue->grant_rx_ref[i] = GRANT_INVALID_REF; | |
1855 | } | |
1856 | ||
1857 | /* A grant for every tx ring slot */ | |
1f3c2eba | 1858 | if (gnttab_alloc_grant_references(NET_TX_RING_SIZE, |
2688fcb7 AB |
1859 | &queue->gref_tx_head) < 0) { |
1860 | pr_alert("can't alloc tx grant refs\n"); | |
1861 | err = -ENOMEM; | |
1862 | goto exit; | |
1863 | } | |
1864 | ||
1865 | /* A grant for every rx ring slot */ | |
1f3c2eba | 1866 | if (gnttab_alloc_grant_references(NET_RX_RING_SIZE, |
2688fcb7 AB |
1867 | &queue->gref_rx_head) < 0) { |
1868 | pr_alert("can't alloc rx grant refs\n"); | |
1869 | err = -ENOMEM; | |
1870 | goto exit_free_tx; | |
1871 | } | |
1872 | ||
2688fcb7 AB |
1873 | return 0; |
1874 | ||
1875 | exit_free_tx: | |
1876 | gnttab_free_grant_references(queue->gref_tx_head); | |
1877 | exit: | |
1878 | return err; | |
1879 | } | |
1880 | ||
50ee6061 AB |
1881 | static int write_queue_xenstore_keys(struct netfront_queue *queue, |
1882 | struct xenbus_transaction *xbt, int write_hierarchical) | |
1883 | { | |
1884 | /* Write the queue-specific keys into XenStore in the traditional | |
1885 | * way for a single queue, or in a queue subkeys for multiple | |
1886 | * queues. | |
1887 | */ | |
1888 | struct xenbus_device *dev = queue->info->xbdev; | |
1889 | int err; | |
1890 | const char *message; | |
1891 | char *path; | |
1892 | size_t pathsize; | |
1893 | ||
1894 | /* Choose the correct place to write the keys */ | |
1895 | if (write_hierarchical) { | |
1896 | pathsize = strlen(dev->nodename) + 10; | |
1897 | path = kzalloc(pathsize, GFP_KERNEL); | |
1898 | if (!path) { | |
1899 | err = -ENOMEM; | |
1900 | message = "out of memory while writing ring references"; | |
1901 | goto error; | |
1902 | } | |
1903 | snprintf(path, pathsize, "%s/queue-%u", | |
1904 | dev->nodename, queue->id); | |
1905 | } else { | |
1906 | path = (char *)dev->nodename; | |
1907 | } | |
1908 | ||
1909 | /* Write ring references */ | |
1910 | err = xenbus_printf(*xbt, path, "tx-ring-ref", "%u", | |
1911 | queue->tx_ring_ref); | |
1912 | if (err) { | |
1913 | message = "writing tx-ring-ref"; | |
1914 | goto error; | |
1915 | } | |
1916 | ||
1917 | err = xenbus_printf(*xbt, path, "rx-ring-ref", "%u", | |
1918 | queue->rx_ring_ref); | |
1919 | if (err) { | |
1920 | message = "writing rx-ring-ref"; | |
1921 | goto error; | |
1922 | } | |
1923 | ||
1924 | /* Write event channels; taking into account both shared | |
1925 | * and split event channel scenarios. | |
1926 | */ | |
1927 | if (queue->tx_evtchn == queue->rx_evtchn) { | |
1928 | /* Shared event channel */ | |
1929 | err = xenbus_printf(*xbt, path, | |
1930 | "event-channel", "%u", queue->tx_evtchn); | |
1931 | if (err) { | |
1932 | message = "writing event-channel"; | |
1933 | goto error; | |
1934 | } | |
1935 | } else { | |
1936 | /* Split event channels */ | |
1937 | err = xenbus_printf(*xbt, path, | |
1938 | "event-channel-tx", "%u", queue->tx_evtchn); | |
1939 | if (err) { | |
1940 | message = "writing event-channel-tx"; | |
1941 | goto error; | |
1942 | } | |
1943 | ||
1944 | err = xenbus_printf(*xbt, path, | |
1945 | "event-channel-rx", "%u", queue->rx_evtchn); | |
1946 | if (err) { | |
1947 | message = "writing event-channel-rx"; | |
1948 | goto error; | |
1949 | } | |
1950 | } | |
1951 | ||
1952 | if (write_hierarchical) | |
1953 | kfree(path); | |
1954 | return 0; | |
1955 | ||
1956 | error: | |
1957 | if (write_hierarchical) | |
1958 | kfree(path); | |
1959 | xenbus_dev_fatal(dev, err, "%s", message); | |
1960 | return err; | |
1961 | } | |
1962 | ||
ce58725f DV |
1963 | static void xennet_destroy_queues(struct netfront_info *info) |
1964 | { | |
1965 | unsigned int i; | |
1966 | ||
ce58725f DV |
1967 | for (i = 0; i < info->netdev->real_num_tx_queues; i++) { |
1968 | struct netfront_queue *queue = &info->queues[i]; | |
1969 | ||
1970 | if (netif_running(info->netdev)) | |
1971 | napi_disable(&queue->napi); | |
1972 | netif_napi_del(&queue->napi); | |
1973 | } | |
1974 | ||
ce58725f DV |
1975 | kfree(info->queues); |
1976 | info->queues = NULL; | |
1977 | } | |
1978 | ||
6c5aa6fc DK |
1979 | |
1980 | ||
1981 | static int xennet_create_page_pool(struct netfront_queue *queue) | |
1982 | { | |
1983 | int err; | |
1984 | struct page_pool_params pp_params = { | |
1985 | .order = 0, | |
1986 | .flags = 0, | |
1987 | .pool_size = NET_RX_RING_SIZE, | |
1988 | .nid = NUMA_NO_NODE, | |
1989 | .dev = &queue->info->netdev->dev, | |
1990 | .offset = XDP_PACKET_HEADROOM, | |
1991 | .max_len = XEN_PAGE_SIZE - XDP_PACKET_HEADROOM, | |
1992 | }; | |
1993 | ||
1994 | queue->page_pool = page_pool_create(&pp_params); | |
1995 | if (IS_ERR(queue->page_pool)) { | |
1996 | err = PTR_ERR(queue->page_pool); | |
1997 | queue->page_pool = NULL; | |
1998 | return err; | |
1999 | } | |
2000 | ||
2001 | err = xdp_rxq_info_reg(&queue->xdp_rxq, queue->info->netdev, | |
b02e5a0e | 2002 | queue->id, 0); |
6c5aa6fc DK |
2003 | if (err) { |
2004 | netdev_err(queue->info->netdev, "xdp_rxq_info_reg failed\n"); | |
2005 | goto err_free_pp; | |
2006 | } | |
2007 | ||
2008 | err = xdp_rxq_info_reg_mem_model(&queue->xdp_rxq, | |
2009 | MEM_TYPE_PAGE_POOL, queue->page_pool); | |
2010 | if (err) { | |
2011 | netdev_err(queue->info->netdev, "xdp_rxq_info_reg_mem_model failed\n"); | |
2012 | goto err_unregister_rxq; | |
2013 | } | |
2014 | return 0; | |
2015 | ||
2016 | err_unregister_rxq: | |
2017 | xdp_rxq_info_unreg(&queue->xdp_rxq); | |
2018 | err_free_pp: | |
2019 | page_pool_destroy(queue->page_pool); | |
2020 | queue->page_pool = NULL; | |
2021 | return err; | |
2022 | } | |
2023 | ||
ce58725f | 2024 | static int xennet_create_queues(struct netfront_info *info, |
ca88ea12 | 2025 | unsigned int *num_queues) |
ce58725f DV |
2026 | { |
2027 | unsigned int i; | |
2028 | int ret; | |
2029 | ||
ca88ea12 | 2030 | info->queues = kcalloc(*num_queues, sizeof(struct netfront_queue), |
ce58725f DV |
2031 | GFP_KERNEL); |
2032 | if (!info->queues) | |
2033 | return -ENOMEM; | |
2034 | ||
ca88ea12 | 2035 | for (i = 0; i < *num_queues; i++) { |
ce58725f DV |
2036 | struct netfront_queue *queue = &info->queues[i]; |
2037 | ||
2038 | queue->id = i; | |
2039 | queue->info = info; | |
2040 | ||
2041 | ret = xennet_init_queue(queue); | |
2042 | if (ret < 0) { | |
f599c64f | 2043 | dev_warn(&info->xbdev->dev, |
69cb8524 | 2044 | "only created %d queues\n", i); |
ca88ea12 | 2045 | *num_queues = i; |
ce58725f DV |
2046 | break; |
2047 | } | |
2048 | ||
6c5aa6fc DK |
2049 | /* use page pool recycling instead of buddy allocator */ |
2050 | ret = xennet_create_page_pool(queue); | |
2051 | if (ret < 0) { | |
2052 | dev_err(&info->xbdev->dev, "can't allocate page pool\n"); | |
2053 | *num_queues = i; | |
2054 | return ret; | |
2055 | } | |
2056 | ||
ce58725f DV |
2057 | netif_napi_add(queue->info->netdev, &queue->napi, |
2058 | xennet_poll, 64); | |
2059 | if (netif_running(info->netdev)) | |
2060 | napi_enable(&queue->napi); | |
2061 | } | |
2062 | ||
ca88ea12 | 2063 | netif_set_real_num_tx_queues(info->netdev, *num_queues); |
ce58725f | 2064 | |
ca88ea12 | 2065 | if (*num_queues == 0) { |
f599c64f | 2066 | dev_err(&info->xbdev->dev, "no queues\n"); |
ce58725f DV |
2067 | return -EINVAL; |
2068 | } | |
2069 | return 0; | |
2070 | } | |
2071 | ||
0d160211 | 2072 | /* Common code used when first setting up, and when resuming. */ |
f502bf2b | 2073 | static int talk_to_netback(struct xenbus_device *dev, |
0d160211 JF |
2074 | struct netfront_info *info) |
2075 | { | |
2076 | const char *message; | |
2077 | struct xenbus_transaction xbt; | |
2078 | int err; | |
2688fcb7 AB |
2079 | unsigned int feature_split_evtchn; |
2080 | unsigned int i = 0; | |
50ee6061 | 2081 | unsigned int max_queues = 0; |
2688fcb7 AB |
2082 | struct netfront_queue *queue = NULL; |
2083 | unsigned int num_queues = 1; | |
0d160211 | 2084 | |
2688fcb7 AB |
2085 | info->netdev->irq = 0; |
2086 | ||
50ee6061 | 2087 | /* Check if backend supports multiple queues */ |
2890ea5c JG |
2088 | max_queues = xenbus_read_unsigned(info->xbdev->otherend, |
2089 | "multi-queue-max-queues", 1); | |
50ee6061 AB |
2090 | num_queues = min(max_queues, xennet_max_queues); |
2091 | ||
2688fcb7 | 2092 | /* Check feature-split-event-channels */ |
2890ea5c JG |
2093 | feature_split_evtchn = xenbus_read_unsigned(info->xbdev->otherend, |
2094 | "feature-split-event-channels", 0); | |
2688fcb7 AB |
2095 | |
2096 | /* Read mac addr. */ | |
2097 | err = xen_net_read_mac(dev, info->netdev->dev_addr); | |
2098 | if (err) { | |
2099 | xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename); | |
cb257783 | 2100 | goto out_unlocked; |
2688fcb7 AB |
2101 | } |
2102 | ||
6c5aa6fc DK |
2103 | info->netback_has_xdp_headroom = xenbus_read_unsigned(info->xbdev->otherend, |
2104 | "feature-xdp-headroom", 0); | |
2105 | if (info->netback_has_xdp_headroom) { | |
2106 | /* set the current xen-netfront xdp state */ | |
2107 | err = talk_to_netback_xdp(info, info->netfront_xdp_enabled ? | |
2108 | NETBACK_XDP_HEADROOM_ENABLE : | |
2109 | NETBACK_XDP_HEADROOM_DISABLE); | |
2110 | if (err) | |
2111 | goto out_unlocked; | |
2112 | } | |
2113 | ||
f599c64f | 2114 | rtnl_lock(); |
ce58725f DV |
2115 | if (info->queues) |
2116 | xennet_destroy_queues(info); | |
2117 | ||
ca88ea12 | 2118 | err = xennet_create_queues(info, &num_queues); |
e2e004ac RL |
2119 | if (err < 0) { |
2120 | xenbus_dev_fatal(dev, err, "creating queues"); | |
2121 | kfree(info->queues); | |
2122 | info->queues = NULL; | |
2123 | goto out; | |
2124 | } | |
f599c64f | 2125 | rtnl_unlock(); |
2688fcb7 AB |
2126 | |
2127 | /* Create shared ring, alloc event channel -- for each queue */ | |
2128 | for (i = 0; i < num_queues; ++i) { | |
2129 | queue = &info->queues[i]; | |
2688fcb7 | 2130 | err = setup_netfront(dev, queue, feature_split_evtchn); |
e2e004ac RL |
2131 | if (err) |
2132 | goto destroy_ring; | |
2688fcb7 | 2133 | } |
0d160211 JF |
2134 | |
2135 | again: | |
2136 | err = xenbus_transaction_start(&xbt); | |
2137 | if (err) { | |
2138 | xenbus_dev_fatal(dev, err, "starting transaction"); | |
2139 | goto destroy_ring; | |
2140 | } | |
2141 | ||
812494d9 | 2142 | if (xenbus_exists(XBT_NIL, |
2143 | info->xbdev->otherend, "multi-queue-max-queues")) { | |
50ee6061 | 2144 | /* Write the number of queues */ |
812494d9 | 2145 | err = xenbus_printf(xbt, dev->nodename, |
2146 | "multi-queue-num-queues", "%u", num_queues); | |
d634bf2c | 2147 | if (err) { |
50ee6061 AB |
2148 | message = "writing multi-queue-num-queues"; |
2149 | goto abort_transaction_no_dev_fatal; | |
d634bf2c | 2150 | } |
812494d9 | 2151 | } |
50ee6061 | 2152 | |
812494d9 | 2153 | if (num_queues == 1) { |
2154 | err = write_queue_xenstore_keys(&info->queues[0], &xbt, 0); /* flat */ | |
2155 | if (err) | |
2156 | goto abort_transaction_no_dev_fatal; | |
2157 | } else { | |
50ee6061 AB |
2158 | /* Write the keys for each queue */ |
2159 | for (i = 0; i < num_queues; ++i) { | |
2160 | queue = &info->queues[i]; | |
2161 | err = write_queue_xenstore_keys(queue, &xbt, 1); /* hierarchical */ | |
2162 | if (err) | |
2163 | goto abort_transaction_no_dev_fatal; | |
d634bf2c | 2164 | } |
0d160211 JF |
2165 | } |
2166 | ||
50ee6061 | 2167 | /* The remaining keys are not queue-specific */ |
0d160211 JF |
2168 | err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u", |
2169 | 1); | |
2170 | if (err) { | |
2171 | message = "writing request-rx-copy"; | |
2172 | goto abort_transaction; | |
2173 | } | |
2174 | ||
2175 | err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1); | |
2176 | if (err) { | |
2177 | message = "writing feature-rx-notify"; | |
2178 | goto abort_transaction; | |
2179 | } | |
2180 | ||
2181 | err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1); | |
2182 | if (err) { | |
2183 | message = "writing feature-sg"; | |
2184 | goto abort_transaction; | |
2185 | } | |
2186 | ||
2187 | err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1); | |
2188 | if (err) { | |
2189 | message = "writing feature-gso-tcpv4"; | |
2190 | goto abort_transaction; | |
2191 | } | |
2192 | ||
2c0057de PD |
2193 | err = xenbus_write(xbt, dev->nodename, "feature-gso-tcpv6", "1"); |
2194 | if (err) { | |
2195 | message = "writing feature-gso-tcpv6"; | |
2196 | goto abort_transaction; | |
2197 | } | |
2198 | ||
2199 | err = xenbus_write(xbt, dev->nodename, "feature-ipv6-csum-offload", | |
2200 | "1"); | |
2201 | if (err) { | |
2202 | message = "writing feature-ipv6-csum-offload"; | |
2203 | goto abort_transaction; | |
2204 | } | |
2205 | ||
0d160211 JF |
2206 | err = xenbus_transaction_end(xbt, 0); |
2207 | if (err) { | |
2208 | if (err == -EAGAIN) | |
2209 | goto again; | |
2210 | xenbus_dev_fatal(dev, err, "completing transaction"); | |
2211 | goto destroy_ring; | |
2212 | } | |
2213 | ||
2214 | return 0; | |
2215 | ||
2216 | abort_transaction: | |
0d160211 | 2217 | xenbus_dev_fatal(dev, err, "%s", message); |
50ee6061 AB |
2218 | abort_transaction_no_dev_fatal: |
2219 | xenbus_transaction_end(xbt, 1); | |
0d160211 JF |
2220 | destroy_ring: |
2221 | xennet_disconnect_backend(info); | |
f599c64f | 2222 | rtnl_lock(); |
e2e004ac | 2223 | xennet_destroy_queues(info); |
0d160211 | 2224 | out: |
f599c64f | 2225 | rtnl_unlock(); |
cb257783 | 2226 | out_unlocked: |
d86b5672 | 2227 | device_unregister(&dev->dev); |
0d160211 JF |
2228 | return err; |
2229 | } | |
2230 | ||
0d160211 JF |
2231 | static int xennet_connect(struct net_device *dev) |
2232 | { | |
2233 | struct netfront_info *np = netdev_priv(dev); | |
2688fcb7 | 2234 | unsigned int num_queues = 0; |
a5b5dc3c | 2235 | int err; |
2688fcb7 AB |
2236 | unsigned int j = 0; |
2237 | struct netfront_queue *queue = NULL; | |
0d160211 | 2238 | |
2890ea5c | 2239 | if (!xenbus_read_unsigned(np->xbdev->otherend, "feature-rx-copy", 0)) { |
0d160211 | 2240 | dev_info(&dev->dev, |
898eb71c | 2241 | "backend does not support copying receive path\n"); |
0d160211 JF |
2242 | return -ENODEV; |
2243 | } | |
2244 | ||
f502bf2b | 2245 | err = talk_to_netback(np->xbdev, np); |
0d160211 JF |
2246 | if (err) |
2247 | return err; | |
6c5aa6fc DK |
2248 | if (np->netback_has_xdp_headroom) |
2249 | pr_info("backend supports XDP headroom\n"); | |
0d160211 | 2250 | |
2688fcb7 AB |
2251 | /* talk_to_netback() sets the correct number of queues */ |
2252 | num_queues = dev->real_num_tx_queues; | |
2253 | ||
f599c64f RL |
2254 | if (dev->reg_state == NETREG_UNINITIALIZED) { |
2255 | err = register_netdev(dev); | |
2256 | if (err) { | |
2257 | pr_warn("%s: register_netdev err=%d\n", __func__, err); | |
2258 | device_unregister(&np->xbdev->dev); | |
2259 | return err; | |
2260 | } | |
2261 | } | |
2262 | ||
45c8184c RL |
2263 | rtnl_lock(); |
2264 | netdev_update_features(dev); | |
2265 | rtnl_unlock(); | |
2266 | ||
0d160211 | 2267 | /* |
a5b5dc3c | 2268 | * All public and private state should now be sane. Get |
0d160211 JF |
2269 | * ready to start sending and receiving packets and give the driver |
2270 | * domain a kick because we've probably just requeued some | |
2271 | * packets. | |
2272 | */ | |
2273 | netif_carrier_on(np->netdev); | |
2688fcb7 AB |
2274 | for (j = 0; j < num_queues; ++j) { |
2275 | queue = &np->queues[j]; | |
f50b4076 | 2276 | |
2688fcb7 AB |
2277 | notify_remote_via_irq(queue->tx_irq); |
2278 | if (queue->tx_irq != queue->rx_irq) | |
2279 | notify_remote_via_irq(queue->rx_irq); | |
2688fcb7 | 2280 | |
f50b4076 DV |
2281 | spin_lock_irq(&queue->tx_lock); |
2282 | xennet_tx_buf_gc(queue); | |
2688fcb7 | 2283 | spin_unlock_irq(&queue->tx_lock); |
f50b4076 DV |
2284 | |
2285 | spin_lock_bh(&queue->rx_lock); | |
2286 | xennet_alloc_rx_buffers(queue); | |
2688fcb7 AB |
2287 | spin_unlock_bh(&queue->rx_lock); |
2288 | } | |
0d160211 JF |
2289 | |
2290 | return 0; | |
2291 | } | |
2292 | ||
80708602 | 2293 | /* |
0d160211 JF |
2294 | * Callback received when the backend's state changes. |
2295 | */ | |
f502bf2b | 2296 | static void netback_changed(struct xenbus_device *dev, |
0d160211 JF |
2297 | enum xenbus_state backend_state) |
2298 | { | |
1b713e00 | 2299 | struct netfront_info *np = dev_get_drvdata(&dev->dev); |
0d160211 JF |
2300 | struct net_device *netdev = np->netdev; |
2301 | ||
2302 | dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state)); | |
2303 | ||
8edfe2e9 JG |
2304 | wake_up_all(&module_wq); |
2305 | ||
0d160211 JF |
2306 | switch (backend_state) { |
2307 | case XenbusStateInitialising: | |
2308 | case XenbusStateInitialised: | |
b78c9512 NI |
2309 | case XenbusStateReconfiguring: |
2310 | case XenbusStateReconfigured: | |
0d160211 | 2311 | case XenbusStateUnknown: |
0d160211 JF |
2312 | break; |
2313 | ||
2314 | case XenbusStateInitWait: | |
2315 | if (dev->state != XenbusStateInitialising) | |
2316 | break; | |
2317 | if (xennet_connect(netdev) != 0) | |
2318 | break; | |
2319 | xenbus_switch_state(dev, XenbusStateConnected); | |
08e34eb1 LE |
2320 | break; |
2321 | ||
2322 | case XenbusStateConnected: | |
ee89bab1 | 2323 | netdev_notify_peers(netdev); |
0d160211 JF |
2324 | break; |
2325 | ||
bce3ea81 DV |
2326 | case XenbusStateClosed: |
2327 | if (dev->state == XenbusStateClosed) | |
2328 | break; | |
df561f66 | 2329 | fallthrough; /* Missed the backend's CLOSING state */ |
0d160211 JF |
2330 | case XenbusStateClosing: |
2331 | xenbus_frontend_closed(dev); | |
2332 | break; | |
2333 | } | |
2334 | } | |
2335 | ||
e0ce4af9 IC |
2336 | static const struct xennet_stat { |
2337 | char name[ETH_GSTRING_LEN]; | |
2338 | u16 offset; | |
2339 | } xennet_stats[] = { | |
2340 | { | |
2341 | "rx_gso_checksum_fixup", | |
2342 | offsetof(struct netfront_info, rx_gso_checksum_fixup) | |
2343 | }, | |
2344 | }; | |
2345 | ||
2346 | static int xennet_get_sset_count(struct net_device *dev, int string_set) | |
2347 | { | |
2348 | switch (string_set) { | |
2349 | case ETH_SS_STATS: | |
2350 | return ARRAY_SIZE(xennet_stats); | |
2351 | default: | |
2352 | return -EINVAL; | |
2353 | } | |
2354 | } | |
2355 | ||
2356 | static void xennet_get_ethtool_stats(struct net_device *dev, | |
2357 | struct ethtool_stats *stats, u64 * data) | |
2358 | { | |
2359 | void *np = netdev_priv(dev); | |
2360 | int i; | |
2361 | ||
2362 | for (i = 0; i < ARRAY_SIZE(xennet_stats); i++) | |
2688fcb7 | 2363 | data[i] = atomic_read((atomic_t *)(np + xennet_stats[i].offset)); |
e0ce4af9 IC |
2364 | } |
2365 | ||
2366 | static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data) | |
2367 | { | |
2368 | int i; | |
2369 | ||
2370 | switch (stringset) { | |
2371 | case ETH_SS_STATS: | |
2372 | for (i = 0; i < ARRAY_SIZE(xennet_stats); i++) | |
2373 | memcpy(data + i * ETH_GSTRING_LEN, | |
2374 | xennet_stats[i].name, ETH_GSTRING_LEN); | |
2375 | break; | |
2376 | } | |
2377 | } | |
2378 | ||
0fc0b732 | 2379 | static const struct ethtool_ops xennet_ethtool_ops = |
0d160211 | 2380 | { |
0d160211 | 2381 | .get_link = ethtool_op_get_link, |
e0ce4af9 IC |
2382 | |
2383 | .get_sset_count = xennet_get_sset_count, | |
2384 | .get_ethtool_stats = xennet_get_ethtool_stats, | |
2385 | .get_strings = xennet_get_strings, | |
91ffb9d3 | 2386 | .get_ts_info = ethtool_op_get_ts_info, |
0d160211 JF |
2387 | }; |
2388 | ||
2389 | #ifdef CONFIG_SYSFS | |
1f3c2eba DV |
2390 | static ssize_t show_rxbuf(struct device *dev, |
2391 | struct device_attribute *attr, char *buf) | |
0d160211 | 2392 | { |
1f3c2eba | 2393 | return sprintf(buf, "%lu\n", NET_RX_RING_SIZE); |
0d160211 JF |
2394 | } |
2395 | ||
1f3c2eba DV |
2396 | static ssize_t store_rxbuf(struct device *dev, |
2397 | struct device_attribute *attr, | |
2398 | const char *buf, size_t len) | |
0d160211 | 2399 | { |
0d160211 | 2400 | char *endp; |
0d160211 JF |
2401 | |
2402 | if (!capable(CAP_NET_ADMIN)) | |
2403 | return -EPERM; | |
2404 | ||
8ed7ec13 | 2405 | simple_strtoul(buf, &endp, 0); |
0d160211 JF |
2406 | if (endp == buf) |
2407 | return -EBADMSG; | |
2408 | ||
1f3c2eba | 2409 | /* rxbuf_min and rxbuf_max are no longer configurable. */ |
0d160211 | 2410 | |
0d160211 JF |
2411 | return len; |
2412 | } | |
2413 | ||
d61e4038 JP |
2414 | static DEVICE_ATTR(rxbuf_min, 0644, show_rxbuf, store_rxbuf); |
2415 | static DEVICE_ATTR(rxbuf_max, 0644, show_rxbuf, store_rxbuf); | |
2416 | static DEVICE_ATTR(rxbuf_cur, 0444, show_rxbuf, NULL); | |
0d160211 | 2417 | |
27b917e5 TI |
2418 | static struct attribute *xennet_dev_attrs[] = { |
2419 | &dev_attr_rxbuf_min.attr, | |
2420 | &dev_attr_rxbuf_max.attr, | |
2421 | &dev_attr_rxbuf_cur.attr, | |
2422 | NULL | |
2423 | }; | |
0d160211 | 2424 | |
27b917e5 TI |
2425 | static const struct attribute_group xennet_dev_group = { |
2426 | .attrs = xennet_dev_attrs | |
2427 | }; | |
0d160211 JF |
2428 | #endif /* CONFIG_SYSFS */ |
2429 | ||
c2c63310 | 2430 | static void xennet_bus_close(struct xenbus_device *dev) |
0d160211 | 2431 | { |
c2c63310 | 2432 | int ret; |
0d160211 | 2433 | |
c2c63310 AR |
2434 | if (xenbus_read_driver_state(dev->otherend) == XenbusStateClosed) |
2435 | return; | |
2436 | do { | |
5b5971df | 2437 | xenbus_switch_state(dev, XenbusStateClosing); |
c2c63310 AR |
2438 | ret = wait_event_timeout(module_wq, |
2439 | xenbus_read_driver_state(dev->otherend) == | |
2440 | XenbusStateClosing || | |
2441 | xenbus_read_driver_state(dev->otherend) == | |
2442 | XenbusStateClosed || | |
2443 | xenbus_read_driver_state(dev->otherend) == | |
2444 | XenbusStateUnknown, | |
2445 | XENNET_TIMEOUT); | |
2446 | } while (!ret); | |
2447 | ||
2448 | if (xenbus_read_driver_state(dev->otherend) == XenbusStateClosed) | |
2449 | return; | |
5b5971df | 2450 | |
c2c63310 | 2451 | do { |
5b5971df | 2452 | xenbus_switch_state(dev, XenbusStateClosed); |
c2c63310 AR |
2453 | ret = wait_event_timeout(module_wq, |
2454 | xenbus_read_driver_state(dev->otherend) == | |
2455 | XenbusStateClosed || | |
2456 | xenbus_read_driver_state(dev->otherend) == | |
2457 | XenbusStateUnknown, | |
2458 | XENNET_TIMEOUT); | |
2459 | } while (!ret); | |
2460 | } | |
2461 | ||
2462 | static int xennet_remove(struct xenbus_device *dev) | |
2463 | { | |
2464 | struct netfront_info *info = dev_get_drvdata(&dev->dev); | |
5b5971df | 2465 | |
c2c63310 | 2466 | xennet_bus_close(dev); |
0d160211 JF |
2467 | xennet_disconnect_backend(info); |
2468 | ||
f599c64f RL |
2469 | if (info->netdev->reg_state == NETREG_REGISTERED) |
2470 | unregister_netdev(info->netdev); | |
6bc96d04 | 2471 | |
f599c64f RL |
2472 | if (info->queues) { |
2473 | rtnl_lock(); | |
9a873c71 | 2474 | xennet_destroy_queues(info); |
f599c64f RL |
2475 | rtnl_unlock(); |
2476 | } | |
900e1833 | 2477 | xennet_free_netdev(info->netdev); |
0d160211 JF |
2478 | |
2479 | return 0; | |
2480 | } | |
2481 | ||
95afae48 DV |
2482 | static const struct xenbus_device_id netfront_ids[] = { |
2483 | { "vif" }, | |
2484 | { "" } | |
2485 | }; | |
2486 | ||
2487 | static struct xenbus_driver netfront_driver = { | |
2488 | .ids = netfront_ids, | |
0d160211 | 2489 | .probe = netfront_probe, |
8e0e46bb | 2490 | .remove = xennet_remove, |
0d160211 | 2491 | .resume = netfront_resume, |
f502bf2b | 2492 | .otherend_changed = netback_changed, |
95afae48 | 2493 | }; |
0d160211 JF |
2494 | |
2495 | static int __init netif_init(void) | |
2496 | { | |
6e833587 | 2497 | if (!xen_domain()) |
0d160211 JF |
2498 | return -ENODEV; |
2499 | ||
51c71a3b | 2500 | if (!xen_has_pv_nic_devices()) |
b9136d20 IM |
2501 | return -ENODEV; |
2502 | ||
383eda32 | 2503 | pr_info("Initialising Xen virtual ethernet driver\n"); |
0d160211 | 2504 | |
034702a6 | 2505 | /* Allow as many queues as there are CPUs inut max. 8 if user has not |
32a84405 WL |
2506 | * specified a value. |
2507 | */ | |
2508 | if (xennet_max_queues == 0) | |
034702a6 JG |
2509 | xennet_max_queues = min_t(unsigned int, MAX_QUEUES_DEFAULT, |
2510 | num_online_cpus()); | |
50ee6061 | 2511 | |
ffb78a26 | 2512 | return xenbus_register_frontend(&netfront_driver); |
0d160211 JF |
2513 | } |
2514 | module_init(netif_init); | |
2515 | ||
2516 | ||
2517 | static void __exit netif_exit(void) | |
2518 | { | |
ffb78a26 | 2519 | xenbus_unregister_driver(&netfront_driver); |
0d160211 JF |
2520 | } |
2521 | module_exit(netif_exit); | |
2522 | ||
2523 | MODULE_DESCRIPTION("Xen virtual network device frontend"); | |
2524 | MODULE_LICENSE("GPL"); | |
d2f0c52b | 2525 | MODULE_ALIAS("xen:vif"); |
4f93f09b | 2526 | MODULE_ALIAS("xennet"); |