]>
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 | |
a884daa6 | 134 | #define TX_PENDING 0xfffe |
0d160211 JF |
135 | grant_ref_t gref_tx_head; |
136 | grant_ref_t grant_tx_ref[NET_TX_RING_SIZE]; | |
cefe0078 | 137 | struct page *grant_tx_page[NET_TX_RING_SIZE]; |
0d160211 | 138 | unsigned tx_skb_freelist; |
a884daa6 | 139 | unsigned int tx_pend_queue; |
0d160211 | 140 | |
84284d3c JF |
141 | spinlock_t rx_lock ____cacheline_aligned_in_smp; |
142 | struct xen_netif_rx_front_ring rx; | |
143 | int rx_ring_ref; | |
144 | ||
84284d3c JF |
145 | struct timer_list rx_refill_timer; |
146 | ||
0d160211 JF |
147 | struct sk_buff *rx_skbs[NET_RX_RING_SIZE]; |
148 | grant_ref_t gref_rx_head; | |
149 | grant_ref_t grant_rx_ref[NET_RX_RING_SIZE]; | |
6c5aa6fc | 150 | |
900e0ac2 JG |
151 | unsigned int rx_rsp_unconsumed; |
152 | spinlock_t rx_cons_lock; | |
153 | ||
6c5aa6fc DK |
154 | struct page_pool *page_pool; |
155 | struct xdp_rxq_info xdp_rxq; | |
2688fcb7 AB |
156 | }; |
157 | ||
158 | struct netfront_info { | |
159 | struct list_head list; | |
160 | struct net_device *netdev; | |
161 | ||
162 | struct xenbus_device *xbdev; | |
163 | ||
164 | /* Multi-queue support */ | |
165 | struct netfront_queue *queues; | |
e0ce4af9 IC |
166 | |
167 | /* Statistics */ | |
900e1833 DV |
168 | struct netfront_stats __percpu *rx_stats; |
169 | struct netfront_stats __percpu *tx_stats; | |
e00f85be | 170 | |
6c5aa6fc DK |
171 | /* XDP state */ |
172 | bool netback_has_xdp_headroom; | |
173 | bool netfront_xdp_enabled; | |
174 | ||
a884daa6 JG |
175 | /* Is device behaving sane? */ |
176 | bool broken; | |
177 | ||
2688fcb7 | 178 | atomic_t rx_gso_checksum_fixup; |
0d160211 JF |
179 | }; |
180 | ||
181 | struct netfront_rx_info { | |
182 | struct xen_netif_rx_response rx; | |
183 | struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1]; | |
184 | }; | |
185 | ||
186 | /* | |
187 | * Access macros for acquiring freeing slots in tx_skbs[]. | |
188 | */ | |
189 | ||
21631d2d JG |
190 | static void add_id_to_list(unsigned *head, unsigned short *list, |
191 | unsigned short id) | |
0d160211 | 192 | { |
21631d2d | 193 | list[id] = *head; |
0d160211 JF |
194 | *head = id; |
195 | } | |
196 | ||
21631d2d | 197 | static unsigned short get_id_from_list(unsigned *head, unsigned short *list) |
0d160211 JF |
198 | { |
199 | unsigned int id = *head; | |
21631d2d JG |
200 | |
201 | if (id != TX_LINK_NONE) { | |
202 | *head = list[id]; | |
203 | list[id] = TX_LINK_NONE; | |
204 | } | |
0d160211 JF |
205 | return id; |
206 | } | |
207 | ||
208 | static int xennet_rxidx(RING_IDX idx) | |
209 | { | |
210 | return idx & (NET_RX_RING_SIZE - 1); | |
211 | } | |
212 | ||
2688fcb7 | 213 | static struct sk_buff *xennet_get_rx_skb(struct netfront_queue *queue, |
0d160211 JF |
214 | RING_IDX ri) |
215 | { | |
216 | int i = xennet_rxidx(ri); | |
2688fcb7 AB |
217 | struct sk_buff *skb = queue->rx_skbs[i]; |
218 | queue->rx_skbs[i] = NULL; | |
0d160211 JF |
219 | return skb; |
220 | } | |
221 | ||
2688fcb7 | 222 | static grant_ref_t xennet_get_rx_ref(struct netfront_queue *queue, |
0d160211 JF |
223 | RING_IDX ri) |
224 | { | |
225 | int i = xennet_rxidx(ri); | |
2688fcb7 AB |
226 | grant_ref_t ref = queue->grant_rx_ref[i]; |
227 | queue->grant_rx_ref[i] = GRANT_INVALID_REF; | |
0d160211 JF |
228 | return ref; |
229 | } | |
230 | ||
231 | #ifdef CONFIG_SYSFS | |
27b917e5 | 232 | static const struct attribute_group xennet_dev_group; |
0d160211 JF |
233 | #endif |
234 | ||
3ad9b358 | 235 | static bool xennet_can_sg(struct net_device *dev) |
0d160211 | 236 | { |
3ad9b358 | 237 | return dev->features & NETIF_F_SG; |
0d160211 JF |
238 | } |
239 | ||
240 | ||
e99e88a9 | 241 | static void rx_refill_timeout(struct timer_list *t) |
0d160211 | 242 | { |
e99e88a9 | 243 | struct netfront_queue *queue = from_timer(queue, t, rx_refill_timer); |
2688fcb7 | 244 | napi_schedule(&queue->napi); |
0d160211 JF |
245 | } |
246 | ||
2688fcb7 | 247 | static int netfront_tx_slot_available(struct netfront_queue *queue) |
0d160211 | 248 | { |
2688fcb7 | 249 | return (queue->tx.req_prod_pvt - queue->tx.rsp_cons) < |
57f230ab | 250 | (NET_TX_RING_SIZE - XEN_NETIF_NR_SLOTS_MIN - 1); |
0d160211 JF |
251 | } |
252 | ||
2688fcb7 | 253 | static void xennet_maybe_wake_tx(struct netfront_queue *queue) |
0d160211 | 254 | { |
2688fcb7 AB |
255 | struct net_device *dev = queue->info->netdev; |
256 | struct netdev_queue *dev_queue = netdev_get_tx_queue(dev, queue->id); | |
0d160211 | 257 | |
2688fcb7 AB |
258 | if (unlikely(netif_tx_queue_stopped(dev_queue)) && |
259 | netfront_tx_slot_available(queue) && | |
0d160211 | 260 | likely(netif_running(dev))) |
2688fcb7 | 261 | netif_tx_wake_queue(netdev_get_tx_queue(dev, queue->id)); |
0d160211 JF |
262 | } |
263 | ||
1f3c2eba DV |
264 | |
265 | static struct sk_buff *xennet_alloc_one_rx_buffer(struct netfront_queue *queue) | |
0d160211 | 266 | { |
0d160211 JF |
267 | struct sk_buff *skb; |
268 | struct page *page; | |
0d160211 | 269 | |
1f3c2eba DV |
270 | skb = __netdev_alloc_skb(queue->info->netdev, |
271 | RX_COPY_THRESHOLD + NET_IP_ALIGN, | |
272 | GFP_ATOMIC | __GFP_NOWARN); | |
273 | if (unlikely(!skb)) | |
274 | return NULL; | |
0d160211 | 275 | |
6240a8a6 RPM |
276 | page = page_pool_alloc_pages(queue->page_pool, |
277 | GFP_ATOMIC | __GFP_NOWARN | __GFP_ZERO); | |
6c5aa6fc | 278 | if (unlikely(!page)) { |
1f3c2eba DV |
279 | kfree_skb(skb); |
280 | return NULL; | |
0d160211 | 281 | } |
1f3c2eba DV |
282 | skb_add_rx_frag(skb, 0, page, 0, 0, PAGE_SIZE); |
283 | ||
284 | /* Align ip header to a 16 bytes boundary */ | |
285 | skb_reserve(skb, NET_IP_ALIGN); | |
286 | skb->dev = queue->info->netdev; | |
287 | ||
288 | return skb; | |
289 | } | |
0d160211 | 290 | |
1f3c2eba DV |
291 | |
292 | static void xennet_alloc_rx_buffers(struct netfront_queue *queue) | |
293 | { | |
294 | RING_IDX req_prod = queue->rx.req_prod_pvt; | |
295 | int notify; | |
538d9291 | 296 | int err = 0; |
1f3c2eba DV |
297 | |
298 | if (unlikely(!netif_carrier_ok(queue->info->netdev))) | |
0d160211 | 299 | return; |
0d160211 | 300 | |
1f3c2eba DV |
301 | for (req_prod = queue->rx.req_prod_pvt; |
302 | req_prod - queue->rx.rsp_cons < NET_RX_RING_SIZE; | |
303 | req_prod++) { | |
304 | struct sk_buff *skb; | |
305 | unsigned short id; | |
306 | grant_ref_t ref; | |
30c5d7f0 | 307 | struct page *page; |
1f3c2eba | 308 | struct xen_netif_rx_request *req; |
0d160211 | 309 | |
1f3c2eba | 310 | skb = xennet_alloc_one_rx_buffer(queue); |
538d9291 VRP |
311 | if (!skb) { |
312 | err = -ENOMEM; | |
0d160211 | 313 | break; |
538d9291 | 314 | } |
0d160211 | 315 | |
1f3c2eba | 316 | id = xennet_rxidx(req_prod); |
0d160211 | 317 | |
2688fcb7 AB |
318 | BUG_ON(queue->rx_skbs[id]); |
319 | queue->rx_skbs[id] = skb; | |
0d160211 | 320 | |
2688fcb7 | 321 | ref = gnttab_claim_grant_reference(&queue->gref_rx_head); |
269ebce4 | 322 | WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref)); |
2688fcb7 | 323 | queue->grant_rx_ref[id] = ref; |
0d160211 | 324 | |
30c5d7f0 | 325 | page = skb_frag_page(&skb_shinfo(skb)->frags[0]); |
0d160211 | 326 | |
1f3c2eba | 327 | req = RING_GET_REQUEST(&queue->rx, req_prod); |
30c5d7f0 JG |
328 | gnttab_page_grant_foreign_access_ref_one(ref, |
329 | queue->info->xbdev->otherend_id, | |
330 | page, | |
331 | 0); | |
0d160211 JF |
332 | req->id = id; |
333 | req->gref = ref; | |
334 | } | |
335 | ||
1f3c2eba DV |
336 | queue->rx.req_prod_pvt = req_prod; |
337 | ||
538d9291 VRP |
338 | /* Try again later if there are not enough requests or skb allocation |
339 | * failed. | |
340 | * Enough requests is quantified as the sum of newly created slots and | |
341 | * the unconsumed slots at the backend. | |
342 | */ | |
343 | if (req_prod - queue->rx.rsp_cons < NET_RX_SLOTS_MIN || | |
344 | unlikely(err)) { | |
1f3c2eba DV |
345 | mod_timer(&queue->rx_refill_timer, jiffies + (HZ/10)); |
346 | return; | |
347 | } | |
348 | ||
2688fcb7 | 349 | RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->rx, notify); |
0d160211 | 350 | if (notify) |
2688fcb7 | 351 | notify_remote_via_irq(queue->rx_irq); |
0d160211 JF |
352 | } |
353 | ||
354 | static int xennet_open(struct net_device *dev) | |
355 | { | |
356 | struct netfront_info *np = netdev_priv(dev); | |
2688fcb7 AB |
357 | unsigned int num_queues = dev->real_num_tx_queues; |
358 | unsigned int i = 0; | |
359 | struct netfront_queue *queue = NULL; | |
360 | ||
a884daa6 | 361 | if (!np->queues || np->broken) |
f599c64f RL |
362 | return -ENODEV; |
363 | ||
2688fcb7 AB |
364 | for (i = 0; i < num_queues; ++i) { |
365 | queue = &np->queues[i]; | |
366 | napi_enable(&queue->napi); | |
367 | ||
368 | spin_lock_bh(&queue->rx_lock); | |
369 | if (netif_carrier_ok(dev)) { | |
370 | xennet_alloc_rx_buffers(queue); | |
371 | queue->rx.sring->rsp_event = queue->rx.rsp_cons + 1; | |
372 | if (RING_HAS_UNCONSUMED_RESPONSES(&queue->rx)) | |
373 | napi_schedule(&queue->napi); | |
374 | } | |
375 | spin_unlock_bh(&queue->rx_lock); | |
0d160211 | 376 | } |
0d160211 | 377 | |
2688fcb7 | 378 | netif_tx_start_all_queues(dev); |
0d160211 JF |
379 | |
380 | return 0; | |
381 | } | |
382 | ||
900e0ac2 | 383 | static bool xennet_tx_buf_gc(struct netfront_queue *queue) |
0d160211 JF |
384 | { |
385 | RING_IDX cons, prod; | |
386 | unsigned short id; | |
0d160211 | 387 | struct sk_buff *skb; |
7d0105b5 | 388 | bool more_to_do; |
900e0ac2 | 389 | bool work_done = false; |
a884daa6 | 390 | const struct device *dev = &queue->info->netdev->dev; |
0d160211 | 391 | |
2688fcb7 | 392 | BUG_ON(!netif_carrier_ok(queue->info->netdev)); |
0d160211 JF |
393 | |
394 | do { | |
2688fcb7 | 395 | prod = queue->tx.sring->rsp_prod; |
a884daa6 JG |
396 | if (RING_RESPONSE_PROD_OVERFLOW(&queue->tx, prod)) { |
397 | dev_alert(dev, "Illegal number of responses %u\n", | |
398 | prod - queue->tx.rsp_cons); | |
399 | goto err; | |
400 | } | |
0d160211 JF |
401 | rmb(); /* Ensure we see responses up to 'rp'. */ |
402 | ||
2688fcb7 | 403 | for (cons = queue->tx.rsp_cons; cons != prod; cons++) { |
8446066b | 404 | struct xen_netif_tx_response txrsp; |
0d160211 | 405 | |
900e0ac2 JG |
406 | work_done = true; |
407 | ||
8446066b JG |
408 | RING_COPY_RESPONSE(&queue->tx, cons, &txrsp); |
409 | if (txrsp.status == XEN_NETIF_RSP_NULL) | |
0d160211 JF |
410 | continue; |
411 | ||
a884daa6 JG |
412 | id = txrsp.id; |
413 | if (id >= RING_SIZE(&queue->tx)) { | |
414 | dev_alert(dev, | |
415 | "Response has incorrect id (%u)\n", | |
416 | id); | |
417 | goto err; | |
418 | } | |
419 | if (queue->tx_link[id] != TX_PENDING) { | |
420 | dev_alert(dev, | |
421 | "Response for inactive request\n"); | |
422 | goto err; | |
423 | } | |
424 | ||
425 | queue->tx_link[id] = TX_LINK_NONE; | |
21631d2d JG |
426 | skb = queue->tx_skbs[id]; |
427 | queue->tx_skbs[id] = NULL; | |
b602c2e0 JG |
428 | if (unlikely(!gnttab_end_foreign_access_ref( |
429 | queue->grant_tx_ref[id], GNTMAP_readonly))) { | |
a884daa6 JG |
430 | dev_alert(dev, |
431 | "Grant still in use by backend domain\n"); | |
432 | goto err; | |
0d160211 | 433 | } |
0d160211 | 434 | gnttab_release_grant_reference( |
2688fcb7 AB |
435 | &queue->gref_tx_head, queue->grant_tx_ref[id]); |
436 | queue->grant_tx_ref[id] = GRANT_INVALID_REF; | |
437 | queue->grant_tx_page[id] = NULL; | |
21631d2d | 438 | add_id_to_list(&queue->tx_skb_freelist, queue->tx_link, id); |
0d160211 JF |
439 | dev_kfree_skb_irq(skb); |
440 | } | |
441 | ||
2688fcb7 | 442 | queue->tx.rsp_cons = prod; |
0d160211 | 443 | |
7d0105b5 MC |
444 | RING_FINAL_CHECK_FOR_RESPONSES(&queue->tx, more_to_do); |
445 | } while (more_to_do); | |
0d160211 | 446 | |
2688fcb7 | 447 | xennet_maybe_wake_tx(queue); |
a884daa6 | 448 | |
900e0ac2 | 449 | return work_done; |
a884daa6 JG |
450 | |
451 | err: | |
452 | queue->info->broken = true; | |
453 | dev_alert(dev, "Disabled for further use\n"); | |
900e0ac2 JG |
454 | |
455 | return work_done; | |
0d160211 JF |
456 | } |
457 | ||
30c5d7f0 JG |
458 | struct xennet_gnttab_make_txreq { |
459 | struct netfront_queue *queue; | |
460 | struct sk_buff *skb; | |
461 | struct page *page; | |
162081ec JG |
462 | struct xen_netif_tx_request *tx; /* Last request on ring page */ |
463 | struct xen_netif_tx_request tx_local; /* Last request local copy*/ | |
30c5d7f0 JG |
464 | unsigned int size; |
465 | }; | |
466 | ||
467 | static void xennet_tx_setup_grant(unsigned long gfn, unsigned int offset, | |
468 | unsigned int len, void *data) | |
0d160211 | 469 | { |
30c5d7f0 | 470 | struct xennet_gnttab_make_txreq *info = data; |
0d160211 | 471 | unsigned int id; |
a55e8bb8 | 472 | struct xen_netif_tx_request *tx; |
0d160211 | 473 | grant_ref_t ref; |
30c5d7f0 JG |
474 | /* convenient aliases */ |
475 | struct page *page = info->page; | |
476 | struct netfront_queue *queue = info->queue; | |
477 | struct sk_buff *skb = info->skb; | |
0d160211 | 478 | |
21631d2d | 479 | id = get_id_from_list(&queue->tx_skb_freelist, queue->tx_link); |
a55e8bb8 DV |
480 | tx = RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++); |
481 | ref = gnttab_claim_grant_reference(&queue->gref_tx_head); | |
269ebce4 | 482 | WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref)); |
0d160211 | 483 | |
30c5d7f0 JG |
484 | gnttab_grant_foreign_access_ref(ref, queue->info->xbdev->otherend_id, |
485 | gfn, GNTMAP_readonly); | |
0d160211 | 486 | |
21631d2d | 487 | queue->tx_skbs[id] = skb; |
a55e8bb8 DV |
488 | queue->grant_tx_page[id] = page; |
489 | queue->grant_tx_ref[id] = ref; | |
0d160211 | 490 | |
162081ec JG |
491 | info->tx_local.id = id; |
492 | info->tx_local.gref = ref; | |
493 | info->tx_local.offset = offset; | |
494 | info->tx_local.size = len; | |
495 | info->tx_local.flags = 0; | |
496 | ||
497 | *tx = info->tx_local; | |
0d160211 | 498 | |
a884daa6 JG |
499 | /* |
500 | * Put the request in the pending queue, it will be set to be pending | |
501 | * when the producer index is about to be raised. | |
502 | */ | |
503 | add_id_to_list(&queue->tx_pend_queue, queue->tx_link, id); | |
504 | ||
30c5d7f0 | 505 | info->tx = tx; |
162081ec | 506 | info->size += info->tx_local.size; |
30c5d7f0 JG |
507 | } |
508 | ||
509 | static struct xen_netif_tx_request *xennet_make_first_txreq( | |
162081ec JG |
510 | struct xennet_gnttab_make_txreq *info, |
511 | unsigned int offset, unsigned int len) | |
30c5d7f0 | 512 | { |
162081ec | 513 | info->size = 0; |
30c5d7f0 | 514 | |
162081ec | 515 | gnttab_for_one_grant(info->page, offset, len, xennet_tx_setup_grant, info); |
30c5d7f0 | 516 | |
162081ec | 517 | return info->tx; |
30c5d7f0 JG |
518 | } |
519 | ||
520 | static void xennet_make_one_txreq(unsigned long gfn, unsigned int offset, | |
521 | unsigned int len, void *data) | |
522 | { | |
523 | struct xennet_gnttab_make_txreq *info = data; | |
524 | ||
525 | info->tx->flags |= XEN_NETTXF_more_data; | |
526 | skb_get(info->skb); | |
527 | xennet_tx_setup_grant(gfn, offset, len, data); | |
a55e8bb8 | 528 | } |
0d160211 | 529 | |
162081ec JG |
530 | static void xennet_make_txreqs( |
531 | struct xennet_gnttab_make_txreq *info, | |
532 | struct page *page, | |
a55e8bb8 DV |
533 | unsigned int offset, unsigned int len) |
534 | { | |
535 | /* Skip unused frames from start of page */ | |
536 | page += offset >> PAGE_SHIFT; | |
537 | offset &= ~PAGE_MASK; | |
0d160211 | 538 | |
a55e8bb8 | 539 | while (len) { |
162081ec JG |
540 | info->page = page; |
541 | info->size = 0; | |
30c5d7f0 JG |
542 | |
543 | gnttab_foreach_grant_in_range(page, offset, len, | |
544 | xennet_make_one_txreq, | |
162081ec | 545 | info); |
30c5d7f0 | 546 | |
a55e8bb8 DV |
547 | page++; |
548 | offset = 0; | |
162081ec | 549 | len -= info->size; |
0d160211 | 550 | } |
0d160211 JF |
551 | } |
552 | ||
f36c3747 | 553 | /* |
e84448d5 DV |
554 | * Count how many ring slots are required to send this skb. Each frag |
555 | * might be a compound page. | |
f36c3747 | 556 | */ |
e84448d5 | 557 | static int xennet_count_skb_slots(struct sk_buff *skb) |
f36c3747 IC |
558 | { |
559 | int i, frags = skb_shinfo(skb)->nr_frags; | |
30c5d7f0 | 560 | int slots; |
e84448d5 | 561 | |
30c5d7f0 JG |
562 | slots = gnttab_count_grant(offset_in_page(skb->data), |
563 | skb_headlen(skb)); | |
f36c3747 IC |
564 | |
565 | for (i = 0; i < frags; i++) { | |
566 | skb_frag_t *frag = skb_shinfo(skb)->frags + i; | |
567 | unsigned long size = skb_frag_size(frag); | |
b54c9d5b | 568 | unsigned long offset = skb_frag_off(frag); |
f36c3747 IC |
569 | |
570 | /* Skip unused frames from start of page */ | |
571 | offset &= ~PAGE_MASK; | |
572 | ||
30c5d7f0 | 573 | slots += gnttab_count_grant(offset, size); |
f36c3747 IC |
574 | } |
575 | ||
30c5d7f0 | 576 | return slots; |
f36c3747 IC |
577 | } |
578 | ||
50ee6061 | 579 | static u16 xennet_select_queue(struct net_device *dev, struct sk_buff *skb, |
a350ecce | 580 | struct net_device *sb_dev) |
2688fcb7 | 581 | { |
50ee6061 AB |
582 | unsigned int num_queues = dev->real_num_tx_queues; |
583 | u32 hash; | |
584 | u16 queue_idx; | |
585 | ||
586 | /* First, check if there is only one queue */ | |
587 | if (num_queues == 1) { | |
588 | queue_idx = 0; | |
589 | } else { | |
590 | hash = skb_get_hash(skb); | |
591 | queue_idx = hash % num_queues; | |
592 | } | |
593 | ||
594 | return queue_idx; | |
2688fcb7 AB |
595 | } |
596 | ||
a884daa6 JG |
597 | static void xennet_mark_tx_pending(struct netfront_queue *queue) |
598 | { | |
599 | unsigned int i; | |
600 | ||
601 | while ((i = get_id_from_list(&queue->tx_pend_queue, queue->tx_link)) != | |
602 | TX_LINK_NONE) | |
603 | queue->tx_link[i] = TX_PENDING; | |
604 | } | |
605 | ||
6c5aa6fc DK |
606 | static int xennet_xdp_xmit_one(struct net_device *dev, |
607 | struct netfront_queue *queue, | |
608 | struct xdp_frame *xdpf) | |
609 | { | |
610 | struct netfront_info *np = netdev_priv(dev); | |
611 | struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats); | |
162081ec JG |
612 | struct xennet_gnttab_make_txreq info = { |
613 | .queue = queue, | |
614 | .skb = NULL, | |
615 | .page = virt_to_page(xdpf->data), | |
616 | }; | |
6c5aa6fc DK |
617 | int notify; |
618 | ||
162081ec | 619 | xennet_make_first_txreq(&info, |
6c5aa6fc DK |
620 | offset_in_page(xdpf->data), |
621 | xdpf->len); | |
622 | ||
a884daa6 JG |
623 | xennet_mark_tx_pending(queue); |
624 | ||
6c5aa6fc DK |
625 | RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify); |
626 | if (notify) | |
627 | notify_remote_via_irq(queue->tx_irq); | |
628 | ||
629 | u64_stats_update_begin(&tx_stats->syncp); | |
630 | tx_stats->bytes += xdpf->len; | |
631 | tx_stats->packets++; | |
632 | u64_stats_update_end(&tx_stats->syncp); | |
633 | ||
634 | xennet_tx_buf_gc(queue); | |
635 | ||
636 | return 0; | |
637 | } | |
638 | ||
639 | static int xennet_xdp_xmit(struct net_device *dev, int n, | |
640 | struct xdp_frame **frames, u32 flags) | |
641 | { | |
642 | unsigned int num_queues = dev->real_num_tx_queues; | |
643 | struct netfront_info *np = netdev_priv(dev); | |
644 | struct netfront_queue *queue = NULL; | |
645 | unsigned long irq_flags; | |
fdc13979 LB |
646 | int nxmit = 0; |
647 | int i; | |
6c5aa6fc | 648 | |
a884daa6 JG |
649 | if (unlikely(np->broken)) |
650 | return -ENODEV; | |
6c5aa6fc DK |
651 | if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK)) |
652 | return -EINVAL; | |
653 | ||
654 | queue = &np->queues[smp_processor_id() % num_queues]; | |
655 | ||
656 | spin_lock_irqsave(&queue->tx_lock, irq_flags); | |
657 | for (i = 0; i < n; i++) { | |
658 | struct xdp_frame *xdpf = frames[i]; | |
659 | ||
660 | if (!xdpf) | |
661 | continue; | |
fdc13979 LB |
662 | if (xennet_xdp_xmit_one(dev, queue, xdpf)) |
663 | break; | |
664 | nxmit++; | |
6c5aa6fc DK |
665 | } |
666 | spin_unlock_irqrestore(&queue->tx_lock, irq_flags); | |
667 | ||
fdc13979 | 668 | return nxmit; |
6c5aa6fc DK |
669 | } |
670 | ||
671 | ||
30c5d7f0 JG |
672 | #define MAX_XEN_SKB_FRAGS (65536 / XEN_PAGE_SIZE + 1) |
673 | ||
24a94b3c | 674 | static netdev_tx_t xennet_start_xmit(struct sk_buff *skb, struct net_device *dev) |
0d160211 | 675 | { |
0d160211 | 676 | struct netfront_info *np = netdev_priv(dev); |
900e1833 | 677 | struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats); |
162081ec | 678 | struct xen_netif_tx_request *first_tx; |
a55e8bb8 | 679 | unsigned int i; |
0d160211 | 680 | int notify; |
f36c3747 | 681 | int slots; |
a55e8bb8 DV |
682 | struct page *page; |
683 | unsigned int offset; | |
684 | unsigned int len; | |
cf66f9d4 | 685 | unsigned long flags; |
2688fcb7 | 686 | struct netfront_queue *queue = NULL; |
162081ec | 687 | struct xennet_gnttab_make_txreq info = { }; |
2688fcb7 AB |
688 | unsigned int num_queues = dev->real_num_tx_queues; |
689 | u16 queue_index; | |
fd07160b | 690 | struct sk_buff *nskb; |
2688fcb7 AB |
691 | |
692 | /* Drop the packet if no queues are set up */ | |
693 | if (num_queues < 1) | |
694 | goto drop; | |
a884daa6 JG |
695 | if (unlikely(np->broken)) |
696 | goto drop; | |
2688fcb7 AB |
697 | /* Determine which queue to transmit this SKB on */ |
698 | queue_index = skb_get_queue_mapping(skb); | |
699 | queue = &np->queues[queue_index]; | |
0d160211 | 700 | |
9ecd1a75 WL |
701 | /* If skb->len is too big for wire format, drop skb and alert |
702 | * user about misconfiguration. | |
703 | */ | |
704 | if (unlikely(skb->len > XEN_NETIF_MAX_TX_SIZE)) { | |
705 | net_alert_ratelimited( | |
706 | "xennet: skb->len = %u, too big for wire format\n", | |
707 | skb->len); | |
708 | goto drop; | |
709 | } | |
710 | ||
e84448d5 | 711 | slots = xennet_count_skb_slots(skb); |
30c5d7f0 | 712 | if (unlikely(slots > MAX_XEN_SKB_FRAGS + 1)) { |
97a6d1bb ZK |
713 | net_dbg_ratelimited("xennet: skb rides the rocket: %d slots, %d bytes\n", |
714 | slots, skb->len); | |
715 | if (skb_linearize(skb)) | |
716 | goto drop; | |
0d160211 JF |
717 | } |
718 | ||
a55e8bb8 DV |
719 | page = virt_to_page(skb->data); |
720 | offset = offset_in_page(skb->data); | |
fd07160b VK |
721 | |
722 | /* The first req should be at least ETH_HLEN size or the packet will be | |
723 | * dropped by netback. | |
724 | */ | |
725 | if (unlikely(PAGE_SIZE - offset < ETH_HLEN)) { | |
726 | nskb = skb_copy(skb, GFP_ATOMIC); | |
727 | if (!nskb) | |
728 | goto drop; | |
62f3250f | 729 | dev_consume_skb_any(skb); |
fd07160b VK |
730 | skb = nskb; |
731 | page = virt_to_page(skb->data); | |
732 | offset = offset_in_page(skb->data); | |
733 | } | |
734 | ||
a55e8bb8 DV |
735 | len = skb_headlen(skb); |
736 | ||
2688fcb7 | 737 | spin_lock_irqsave(&queue->tx_lock, flags); |
0d160211 JF |
738 | |
739 | if (unlikely(!netif_carrier_ok(dev) || | |
f36c3747 | 740 | (slots > 1 && !xennet_can_sg(dev)) || |
8b86a61d | 741 | netif_needs_gso(skb, netif_skb_features(skb)))) { |
2688fcb7 | 742 | spin_unlock_irqrestore(&queue->tx_lock, flags); |
0d160211 JF |
743 | goto drop; |
744 | } | |
745 | ||
a55e8bb8 | 746 | /* First request for the linear area. */ |
162081ec JG |
747 | info.queue = queue; |
748 | info.skb = skb; | |
749 | info.page = page; | |
750 | first_tx = xennet_make_first_txreq(&info, offset, len); | |
751 | offset += info.tx_local.size; | |
30c5d7f0 JG |
752 | if (offset == PAGE_SIZE) { |
753 | page++; | |
754 | offset = 0; | |
755 | } | |
162081ec | 756 | len -= info.tx_local.size; |
0d160211 | 757 | |
0d160211 JF |
758 | if (skb->ip_summed == CHECKSUM_PARTIAL) |
759 | /* local packet? */ | |
162081ec JG |
760 | first_tx->flags |= XEN_NETTXF_csum_blank | |
761 | XEN_NETTXF_data_validated; | |
0d160211 JF |
762 | else if (skb->ip_summed == CHECKSUM_UNNECESSARY) |
763 | /* remote but checksummed. */ | |
162081ec | 764 | first_tx->flags |= XEN_NETTXF_data_validated; |
0d160211 | 765 | |
a55e8bb8 | 766 | /* Optional extra info after the first request. */ |
0d160211 JF |
767 | if (skb_shinfo(skb)->gso_size) { |
768 | struct xen_netif_extra_info *gso; | |
769 | ||
770 | gso = (struct xen_netif_extra_info *) | |
a55e8bb8 | 771 | RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++); |
0d160211 | 772 | |
162081ec | 773 | first_tx->flags |= XEN_NETTXF_extra_info; |
0d160211 JF |
774 | |
775 | gso->u.gso.size = skb_shinfo(skb)->gso_size; | |
2c0057de PD |
776 | gso->u.gso.type = (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) ? |
777 | XEN_NETIF_GSO_TYPE_TCPV6 : | |
778 | XEN_NETIF_GSO_TYPE_TCPV4; | |
0d160211 JF |
779 | gso->u.gso.pad = 0; |
780 | gso->u.gso.features = 0; | |
781 | ||
782 | gso->type = XEN_NETIF_EXTRA_TYPE_GSO; | |
783 | gso->flags = 0; | |
0d160211 JF |
784 | } |
785 | ||
a55e8bb8 | 786 | /* Requests for the rest of the linear area. */ |
162081ec | 787 | xennet_make_txreqs(&info, page, offset, len); |
a55e8bb8 DV |
788 | |
789 | /* Requests for all the frags. */ | |
790 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
791 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
162081ec | 792 | xennet_make_txreqs(&info, skb_frag_page(frag), |
b54c9d5b | 793 | skb_frag_off(frag), |
a55e8bb8 DV |
794 | skb_frag_size(frag)); |
795 | } | |
0d160211 | 796 | |
a55e8bb8 DV |
797 | /* First request has the packet length. */ |
798 | first_tx->size = skb->len; | |
0d160211 | 799 | |
91ffb9d3 DD |
800 | /* timestamp packet in software */ |
801 | skb_tx_timestamp(skb); | |
802 | ||
a884daa6 JG |
803 | xennet_mark_tx_pending(queue); |
804 | ||
2688fcb7 | 805 | RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify); |
0d160211 | 806 | if (notify) |
2688fcb7 | 807 | notify_remote_via_irq(queue->tx_irq); |
0d160211 | 808 | |
900e1833 DV |
809 | u64_stats_update_begin(&tx_stats->syncp); |
810 | tx_stats->bytes += skb->len; | |
811 | tx_stats->packets++; | |
812 | u64_stats_update_end(&tx_stats->syncp); | |
10a273a6 JF |
813 | |
814 | /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */ | |
2688fcb7 | 815 | xennet_tx_buf_gc(queue); |
0d160211 | 816 | |
2688fcb7 AB |
817 | if (!netfront_tx_slot_available(queue)) |
818 | netif_tx_stop_queue(netdev_get_tx_queue(dev, queue->id)); | |
0d160211 | 819 | |
2688fcb7 | 820 | spin_unlock_irqrestore(&queue->tx_lock, flags); |
0d160211 | 821 | |
6ed10654 | 822 | return NETDEV_TX_OK; |
0d160211 JF |
823 | |
824 | drop: | |
09f75cd7 | 825 | dev->stats.tx_dropped++; |
979de8a0 | 826 | dev_kfree_skb_any(skb); |
6ed10654 | 827 | return NETDEV_TX_OK; |
0d160211 JF |
828 | } |
829 | ||
830 | static int xennet_close(struct net_device *dev) | |
831 | { | |
832 | struct netfront_info *np = netdev_priv(dev); | |
2688fcb7 AB |
833 | unsigned int num_queues = dev->real_num_tx_queues; |
834 | unsigned int i; | |
835 | struct netfront_queue *queue; | |
836 | netif_tx_stop_all_queues(np->netdev); | |
837 | for (i = 0; i < num_queues; ++i) { | |
838 | queue = &np->queues[i]; | |
839 | napi_disable(&queue->napi); | |
840 | } | |
0d160211 JF |
841 | return 0; |
842 | } | |
843 | ||
1d0ea124 MMG |
844 | static void xennet_destroy_queues(struct netfront_info *info) |
845 | { | |
846 | unsigned int i; | |
847 | ||
848 | for (i = 0; i < info->netdev->real_num_tx_queues; i++) { | |
849 | struct netfront_queue *queue = &info->queues[i]; | |
850 | ||
851 | if (netif_running(info->netdev)) | |
852 | napi_disable(&queue->napi); | |
853 | netif_napi_del(&queue->napi); | |
854 | } | |
855 | ||
856 | kfree(info->queues); | |
857 | info->queues = NULL; | |
858 | } | |
859 | ||
860 | static void xennet_uninit(struct net_device *dev) | |
861 | { | |
862 | struct netfront_info *np = netdev_priv(dev); | |
863 | xennet_destroy_queues(np); | |
864 | } | |
865 | ||
900e0ac2 JG |
866 | static void xennet_set_rx_rsp_cons(struct netfront_queue *queue, RING_IDX val) |
867 | { | |
868 | unsigned long flags; | |
869 | ||
870 | spin_lock_irqsave(&queue->rx_cons_lock, flags); | |
871 | queue->rx.rsp_cons = val; | |
872 | queue->rx_rsp_unconsumed = RING_HAS_UNCONSUMED_RESPONSES(&queue->rx); | |
873 | spin_unlock_irqrestore(&queue->rx_cons_lock, flags); | |
874 | } | |
875 | ||
2688fcb7 | 876 | static void xennet_move_rx_slot(struct netfront_queue *queue, struct sk_buff *skb, |
0d160211 JF |
877 | grant_ref_t ref) |
878 | { | |
2688fcb7 AB |
879 | int new = xennet_rxidx(queue->rx.req_prod_pvt); |
880 | ||
881 | BUG_ON(queue->rx_skbs[new]); | |
882 | queue->rx_skbs[new] = skb; | |
883 | queue->grant_rx_ref[new] = ref; | |
884 | RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->id = new; | |
885 | RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->gref = ref; | |
886 | queue->rx.req_prod_pvt++; | |
0d160211 JF |
887 | } |
888 | ||
2688fcb7 | 889 | static int xennet_get_extras(struct netfront_queue *queue, |
0d160211 JF |
890 | struct xen_netif_extra_info *extras, |
891 | RING_IDX rp) | |
892 | ||
893 | { | |
8446066b | 894 | struct xen_netif_extra_info extra; |
2688fcb7 AB |
895 | struct device *dev = &queue->info->netdev->dev; |
896 | RING_IDX cons = queue->rx.rsp_cons; | |
0d160211 JF |
897 | int err = 0; |
898 | ||
899 | do { | |
900 | struct sk_buff *skb; | |
901 | grant_ref_t ref; | |
902 | ||
903 | if (unlikely(cons + 1 == rp)) { | |
904 | if (net_ratelimit()) | |
905 | dev_warn(dev, "Missing extra info\n"); | |
906 | err = -EBADR; | |
907 | break; | |
908 | } | |
909 | ||
8446066b | 910 | RING_COPY_RESPONSE(&queue->rx, ++cons, &extra); |
0d160211 | 911 | |
8446066b JG |
912 | if (unlikely(!extra.type || |
913 | extra.type >= XEN_NETIF_EXTRA_TYPE_MAX)) { | |
0d160211 JF |
914 | if (net_ratelimit()) |
915 | dev_warn(dev, "Invalid extra type: %d\n", | |
8446066b | 916 | extra.type); |
0d160211 JF |
917 | err = -EINVAL; |
918 | } else { | |
8446066b | 919 | extras[extra.type - 1] = extra; |
0d160211 JF |
920 | } |
921 | ||
2688fcb7 AB |
922 | skb = xennet_get_rx_skb(queue, cons); |
923 | ref = xennet_get_rx_ref(queue, cons); | |
924 | xennet_move_rx_slot(queue, skb, ref); | |
8446066b | 925 | } while (extra.flags & XEN_NETIF_EXTRA_FLAG_MORE); |
0d160211 | 926 | |
900e0ac2 | 927 | xennet_set_rx_rsp_cons(queue, cons); |
0d160211 JF |
928 | return err; |
929 | } | |
930 | ||
6c5aa6fc DK |
931 | static u32 xennet_run_xdp(struct netfront_queue *queue, struct page *pdata, |
932 | struct xen_netif_rx_response *rx, struct bpf_prog *prog, | |
933 | struct xdp_buff *xdp, bool *need_xdp_flush) | |
934 | { | |
935 | struct xdp_frame *xdpf; | |
936 | u32 len = rx->status; | |
e44f65fd | 937 | u32 act; |
6c5aa6fc DK |
938 | int err; |
939 | ||
43b5169d LB |
940 | xdp_init_buff(xdp, XEN_PAGE_SIZE - XDP_PACKET_HEADROOM, |
941 | &queue->xdp_rxq); | |
be9df4af LB |
942 | xdp_prepare_buff(xdp, page_address(pdata), XDP_PACKET_HEADROOM, |
943 | len, false); | |
6c5aa6fc DK |
944 | |
945 | act = bpf_prog_run_xdp(prog, xdp); | |
946 | switch (act) { | |
947 | case XDP_TX: | |
948 | get_page(pdata); | |
949 | xdpf = xdp_convert_buff_to_frame(xdp); | |
950 | err = xennet_xdp_xmit(queue->info->netdev, 1, &xdpf, 0); | |
fdc13979 LB |
951 | if (unlikely(!err)) |
952 | xdp_return_frame_rx_napi(xdpf); | |
953 | else if (unlikely(err < 0)) | |
6c5aa6fc DK |
954 | trace_xdp_exception(queue->info->netdev, prog, act); |
955 | break; | |
956 | case XDP_REDIRECT: | |
957 | get_page(pdata); | |
958 | err = xdp_do_redirect(queue->info->netdev, xdp, prog); | |
959 | *need_xdp_flush = true; | |
960 | if (unlikely(err)) | |
961 | trace_xdp_exception(queue->info->netdev, prog, act); | |
962 | break; | |
963 | case XDP_PASS: | |
964 | case XDP_DROP: | |
965 | break; | |
966 | ||
967 | case XDP_ABORTED: | |
968 | trace_xdp_exception(queue->info->netdev, prog, act); | |
969 | break; | |
970 | ||
971 | default: | |
972 | bpf_warn_invalid_xdp_action(act); | |
973 | } | |
974 | ||
975 | return act; | |
976 | } | |
977 | ||
2688fcb7 | 978 | static int xennet_get_responses(struct netfront_queue *queue, |
0d160211 | 979 | struct netfront_rx_info *rinfo, RING_IDX rp, |
6c5aa6fc DK |
980 | struct sk_buff_head *list, |
981 | bool *need_xdp_flush) | |
0d160211 | 982 | { |
8446066b | 983 | struct xen_netif_rx_response *rx = &rinfo->rx, rx_local; |
6c5aa6fc | 984 | int max = XEN_NETIF_NR_SLOTS_MIN + (rx->status <= RX_COPY_THRESHOLD); |
2688fcb7 AB |
985 | RING_IDX cons = queue->rx.rsp_cons; |
986 | struct sk_buff *skb = xennet_get_rx_skb(queue, cons); | |
6c5aa6fc | 987 | struct xen_netif_extra_info *extras = rinfo->extras; |
2688fcb7 | 988 | grant_ref_t ref = xennet_get_rx_ref(queue, cons); |
6c5aa6fc DK |
989 | struct device *dev = &queue->info->netdev->dev; |
990 | struct bpf_prog *xdp_prog; | |
991 | struct xdp_buff xdp; | |
7158ff6d | 992 | int slots = 1; |
0d160211 | 993 | int err = 0; |
6c5aa6fc | 994 | u32 verdict; |
0d160211 | 995 | |
f942dc25 | 996 | if (rx->flags & XEN_NETRXF_extra_info) { |
2688fcb7 | 997 | err = xennet_get_extras(queue, extras, rp); |
6c5aa6fc DK |
998 | if (!err) { |
999 | if (extras[XEN_NETIF_EXTRA_TYPE_XDP - 1].type) { | |
1000 | struct xen_netif_extra_info *xdp; | |
1001 | ||
1002 | xdp = &extras[XEN_NETIF_EXTRA_TYPE_XDP - 1]; | |
1003 | rx->offset = xdp->u.xdp.headroom; | |
1004 | } | |
1005 | } | |
2688fcb7 | 1006 | cons = queue->rx.rsp_cons; |
0d160211 JF |
1007 | } |
1008 | ||
1009 | for (;;) { | |
1010 | if (unlikely(rx->status < 0 || | |
30c5d7f0 | 1011 | rx->offset + rx->status > XEN_PAGE_SIZE)) { |
0d160211 | 1012 | if (net_ratelimit()) |
6c10127d | 1013 | dev_warn(dev, "rx->offset: %u, size: %d\n", |
0d160211 | 1014 | rx->offset, rx->status); |
2688fcb7 | 1015 | xennet_move_rx_slot(queue, skb, ref); |
0d160211 JF |
1016 | err = -EINVAL; |
1017 | goto next; | |
1018 | } | |
1019 | ||
1020 | /* | |
1021 | * This definitely indicates a bug, either in this driver or in | |
1022 | * the backend driver. In future this should flag the bad | |
697089dc | 1023 | * situation to the system controller to reboot the backend. |
0d160211 JF |
1024 | */ |
1025 | if (ref == GRANT_INVALID_REF) { | |
1026 | if (net_ratelimit()) | |
1027 | dev_warn(dev, "Bad rx response id %d.\n", | |
1028 | rx->id); | |
1029 | err = -EINVAL; | |
1030 | goto next; | |
1031 | } | |
1032 | ||
60a69176 JG |
1033 | if (!gnttab_end_foreign_access_ref(ref, 0)) { |
1034 | dev_alert(dev, | |
1035 | "Grant still in use by backend domain\n"); | |
1036 | queue->info->broken = true; | |
1037 | dev_alert(dev, "Disabled for further use\n"); | |
1038 | return -EINVAL; | |
1039 | } | |
0d160211 | 1040 | |
2688fcb7 | 1041 | gnttab_release_grant_reference(&queue->gref_rx_head, ref); |
0d160211 | 1042 | |
6c5aa6fc DK |
1043 | rcu_read_lock(); |
1044 | xdp_prog = rcu_dereference(queue->xdp_prog); | |
1045 | if (xdp_prog) { | |
1046 | if (!(rx->flags & XEN_NETRXF_more_data)) { | |
1047 | /* currently only a single page contains data */ | |
1048 | verdict = xennet_run_xdp(queue, | |
1049 | skb_frag_page(&skb_shinfo(skb)->frags[0]), | |
1050 | rx, xdp_prog, &xdp, need_xdp_flush); | |
1051 | if (verdict != XDP_PASS) | |
1052 | err = -EINVAL; | |
1053 | } else { | |
1054 | /* drop the frame */ | |
1055 | err = -EINVAL; | |
1056 | } | |
1057 | } | |
1058 | rcu_read_unlock(); | |
0d160211 | 1059 | next: |
6c5aa6fc | 1060 | __skb_queue_tail(list, skb); |
f942dc25 | 1061 | if (!(rx->flags & XEN_NETRXF_more_data)) |
0d160211 JF |
1062 | break; |
1063 | ||
7158ff6d | 1064 | if (cons + slots == rp) { |
0d160211 | 1065 | if (net_ratelimit()) |
7158ff6d | 1066 | dev_warn(dev, "Need more slots\n"); |
0d160211 JF |
1067 | err = -ENOENT; |
1068 | break; | |
1069 | } | |
1070 | ||
8446066b JG |
1071 | RING_COPY_RESPONSE(&queue->rx, cons + slots, &rx_local); |
1072 | rx = &rx_local; | |
2688fcb7 AB |
1073 | skb = xennet_get_rx_skb(queue, cons + slots); |
1074 | ref = xennet_get_rx_ref(queue, cons + slots); | |
7158ff6d | 1075 | slots++; |
0d160211 JF |
1076 | } |
1077 | ||
7158ff6d | 1078 | if (unlikely(slots > max)) { |
0d160211 | 1079 | if (net_ratelimit()) |
697089dc | 1080 | dev_warn(dev, "Too many slots\n"); |
0d160211 JF |
1081 | err = -E2BIG; |
1082 | } | |
1083 | ||
1084 | if (unlikely(err)) | |
900e0ac2 | 1085 | xennet_set_rx_rsp_cons(queue, cons + slots); |
0d160211 JF |
1086 | |
1087 | return err; | |
1088 | } | |
1089 | ||
1090 | static int xennet_set_skb_gso(struct sk_buff *skb, | |
1091 | struct xen_netif_extra_info *gso) | |
1092 | { | |
1093 | if (!gso->u.gso.size) { | |
1094 | if (net_ratelimit()) | |
383eda32 | 1095 | pr_warn("GSO size must not be zero\n"); |
0d160211 JF |
1096 | return -EINVAL; |
1097 | } | |
1098 | ||
2c0057de PD |
1099 | if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4 && |
1100 | gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV6) { | |
0d160211 | 1101 | if (net_ratelimit()) |
383eda32 | 1102 | pr_warn("Bad GSO type %d\n", gso->u.gso.type); |
0d160211 JF |
1103 | return -EINVAL; |
1104 | } | |
1105 | ||
1106 | skb_shinfo(skb)->gso_size = gso->u.gso.size; | |
2c0057de PD |
1107 | skb_shinfo(skb)->gso_type = |
1108 | (gso->u.gso.type == XEN_NETIF_GSO_TYPE_TCPV4) ? | |
1109 | SKB_GSO_TCPV4 : | |
1110 | SKB_GSO_TCPV6; | |
0d160211 JF |
1111 | |
1112 | /* Header must be checked, and gso_segs computed. */ | |
1113 | skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; | |
1114 | skb_shinfo(skb)->gso_segs = 0; | |
1115 | ||
1116 | return 0; | |
1117 | } | |
1118 | ||
a761129e DZ |
1119 | static int xennet_fill_frags(struct netfront_queue *queue, |
1120 | struct sk_buff *skb, | |
1121 | struct sk_buff_head *list) | |
0d160211 | 1122 | { |
2688fcb7 | 1123 | RING_IDX cons = queue->rx.rsp_cons; |
0d160211 JF |
1124 | struct sk_buff *nskb; |
1125 | ||
1126 | while ((nskb = __skb_dequeue(list))) { | |
8446066b | 1127 | struct xen_netif_rx_response rx; |
01c68026 | 1128 | skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0]; |
0d160211 | 1129 | |
8446066b JG |
1130 | RING_COPY_RESPONSE(&queue->rx, ++cons, &rx); |
1131 | ||
d472b3a6 | 1132 | if (skb_shinfo(skb)->nr_frags == MAX_SKB_FRAGS) { |
093b9c71 | 1133 | unsigned int pull_to = NETFRONT_SKB_CB(skb)->pull_to; |
0d160211 | 1134 | |
d81c5054 | 1135 | BUG_ON(pull_to < skb_headlen(skb)); |
093b9c71 JB |
1136 | __pskb_pull_tail(skb, pull_to - skb_headlen(skb)); |
1137 | } | |
ad4f15dc | 1138 | if (unlikely(skb_shinfo(skb)->nr_frags >= MAX_SKB_FRAGS)) { |
900e0ac2 JG |
1139 | xennet_set_rx_rsp_cons(queue, |
1140 | ++cons + skb_queue_len(list)); | |
ad4f15dc | 1141 | kfree_skb(nskb); |
a761129e | 1142 | return -ENOENT; |
ad4f15dc | 1143 | } |
093b9c71 | 1144 | |
d472b3a6 JG |
1145 | skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, |
1146 | skb_frag_page(nfrag), | |
8446066b | 1147 | rx.offset, rx.status, PAGE_SIZE); |
0d160211 JF |
1148 | |
1149 | skb_shinfo(nskb)->nr_frags = 0; | |
1150 | kfree_skb(nskb); | |
0d160211 JF |
1151 | } |
1152 | ||
900e0ac2 | 1153 | xennet_set_rx_rsp_cons(queue, cons); |
a761129e DZ |
1154 | |
1155 | return 0; | |
0d160211 JF |
1156 | } |
1157 | ||
e0ce4af9 | 1158 | static int checksum_setup(struct net_device *dev, struct sk_buff *skb) |
0d160211 | 1159 | { |
b5cf66cd | 1160 | bool recalculate_partial_csum = false; |
e0ce4af9 IC |
1161 | |
1162 | /* | |
1163 | * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy | |
1164 | * peers can fail to set NETRXF_csum_blank when sending a GSO | |
1165 | * frame. In this case force the SKB to CHECKSUM_PARTIAL and | |
1166 | * recalculate the partial checksum. | |
1167 | */ | |
1168 | if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) { | |
1169 | struct netfront_info *np = netdev_priv(dev); | |
2688fcb7 | 1170 | atomic_inc(&np->rx_gso_checksum_fixup); |
e0ce4af9 | 1171 | skb->ip_summed = CHECKSUM_PARTIAL; |
b5cf66cd | 1172 | recalculate_partial_csum = true; |
e0ce4af9 IC |
1173 | } |
1174 | ||
1175 | /* A non-CHECKSUM_PARTIAL SKB does not require setup. */ | |
1176 | if (skb->ip_summed != CHECKSUM_PARTIAL) | |
1177 | return 0; | |
0d160211 | 1178 | |
b5cf66cd | 1179 | return skb_checksum_setup(skb, recalculate_partial_csum); |
0d160211 JF |
1180 | } |
1181 | ||
2688fcb7 | 1182 | static int handle_incoming_queue(struct netfront_queue *queue, |
09f75cd7 | 1183 | struct sk_buff_head *rxq) |
0d160211 | 1184 | { |
900e1833 | 1185 | struct netfront_stats *rx_stats = this_cpu_ptr(queue->info->rx_stats); |
0d160211 JF |
1186 | int packets_dropped = 0; |
1187 | struct sk_buff *skb; | |
1188 | ||
1189 | while ((skb = __skb_dequeue(rxq)) != NULL) { | |
3683243b | 1190 | int pull_to = NETFRONT_SKB_CB(skb)->pull_to; |
0d160211 | 1191 | |
093b9c71 JB |
1192 | if (pull_to > skb_headlen(skb)) |
1193 | __pskb_pull_tail(skb, pull_to - skb_headlen(skb)); | |
0d160211 JF |
1194 | |
1195 | /* Ethernet work: Delayed to here as it peeks the header. */ | |
2688fcb7 | 1196 | skb->protocol = eth_type_trans(skb, queue->info->netdev); |
d554f73d | 1197 | skb_reset_network_header(skb); |
0d160211 | 1198 | |
2688fcb7 | 1199 | if (checksum_setup(queue->info->netdev, skb)) { |
e0ce4af9 IC |
1200 | kfree_skb(skb); |
1201 | packets_dropped++; | |
2688fcb7 | 1202 | queue->info->netdev->stats.rx_errors++; |
e0ce4af9 | 1203 | continue; |
0d160211 JF |
1204 | } |
1205 | ||
900e1833 DV |
1206 | u64_stats_update_begin(&rx_stats->syncp); |
1207 | rx_stats->packets++; | |
1208 | rx_stats->bytes += skb->len; | |
1209 | u64_stats_update_end(&rx_stats->syncp); | |
0d160211 JF |
1210 | |
1211 | /* Pass it up. */ | |
2688fcb7 | 1212 | napi_gro_receive(&queue->napi, skb); |
0d160211 JF |
1213 | } |
1214 | ||
1215 | return packets_dropped; | |
1216 | } | |
1217 | ||
bea3348e | 1218 | static int xennet_poll(struct napi_struct *napi, int budget) |
0d160211 | 1219 | { |
2688fcb7 AB |
1220 | struct netfront_queue *queue = container_of(napi, struct netfront_queue, napi); |
1221 | struct net_device *dev = queue->info->netdev; | |
0d160211 JF |
1222 | struct sk_buff *skb; |
1223 | struct netfront_rx_info rinfo; | |
1224 | struct xen_netif_rx_response *rx = &rinfo.rx; | |
1225 | struct xen_netif_extra_info *extras = rinfo.extras; | |
1226 | RING_IDX i, rp; | |
bea3348e | 1227 | int work_done; |
0d160211 JF |
1228 | struct sk_buff_head rxq; |
1229 | struct sk_buff_head errq; | |
1230 | struct sk_buff_head tmpq; | |
0d160211 | 1231 | int err; |
6c5aa6fc | 1232 | bool need_xdp_flush = false; |
0d160211 | 1233 | |
2688fcb7 | 1234 | spin_lock(&queue->rx_lock); |
0d160211 | 1235 | |
0d160211 JF |
1236 | skb_queue_head_init(&rxq); |
1237 | skb_queue_head_init(&errq); | |
1238 | skb_queue_head_init(&tmpq); | |
1239 | ||
2688fcb7 | 1240 | rp = queue->rx.sring->rsp_prod; |
a884daa6 JG |
1241 | if (RING_RESPONSE_PROD_OVERFLOW(&queue->rx, rp)) { |
1242 | dev_alert(&dev->dev, "Illegal number of responses %u\n", | |
1243 | rp - queue->rx.rsp_cons); | |
1244 | queue->info->broken = true; | |
1245 | spin_unlock(&queue->rx_lock); | |
1246 | return 0; | |
1247 | } | |
0d160211 JF |
1248 | rmb(); /* Ensure we see queued responses up to 'rp'. */ |
1249 | ||
2688fcb7 | 1250 | i = queue->rx.rsp_cons; |
0d160211 JF |
1251 | work_done = 0; |
1252 | while ((i != rp) && (work_done < budget)) { | |
8446066b | 1253 | RING_COPY_RESPONSE(&queue->rx, i, rx); |
0d160211 JF |
1254 | memset(extras, 0, sizeof(rinfo.extras)); |
1255 | ||
6c5aa6fc DK |
1256 | err = xennet_get_responses(queue, &rinfo, rp, &tmpq, |
1257 | &need_xdp_flush); | |
0d160211 JF |
1258 | |
1259 | if (unlikely(err)) { | |
60a69176 JG |
1260 | if (queue->info->broken) { |
1261 | spin_unlock(&queue->rx_lock); | |
1262 | return 0; | |
1263 | } | |
0d160211 JF |
1264 | err: |
1265 | while ((skb = __skb_dequeue(&tmpq))) | |
1266 | __skb_queue_tail(&errq, skb); | |
09f75cd7 | 1267 | dev->stats.rx_errors++; |
2688fcb7 | 1268 | i = queue->rx.rsp_cons; |
0d160211 JF |
1269 | continue; |
1270 | } | |
1271 | ||
1272 | skb = __skb_dequeue(&tmpq); | |
1273 | ||
1274 | if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) { | |
1275 | struct xen_netif_extra_info *gso; | |
1276 | gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1]; | |
1277 | ||
1278 | if (unlikely(xennet_set_skb_gso(skb, gso))) { | |
1279 | __skb_queue_head(&tmpq, skb); | |
900e0ac2 JG |
1280 | xennet_set_rx_rsp_cons(queue, |
1281 | queue->rx.rsp_cons + | |
1282 | skb_queue_len(&tmpq)); | |
0d160211 JF |
1283 | goto err; |
1284 | } | |
1285 | } | |
1286 | ||
3683243b IC |
1287 | NETFRONT_SKB_CB(skb)->pull_to = rx->status; |
1288 | if (NETFRONT_SKB_CB(skb)->pull_to > RX_COPY_THRESHOLD) | |
1289 | NETFRONT_SKB_CB(skb)->pull_to = RX_COPY_THRESHOLD; | |
0d160211 | 1290 | |
b54c9d5b | 1291 | skb_frag_off_set(&skb_shinfo(skb)->frags[0], rx->offset); |
3683243b IC |
1292 | skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status); |
1293 | skb->data_len = rx->status; | |
093b9c71 | 1294 | skb->len += rx->status; |
0d160211 | 1295 | |
a761129e | 1296 | if (unlikely(xennet_fill_frags(queue, skb, &tmpq))) |
ad4f15dc | 1297 | goto err; |
0d160211 | 1298 | |
f942dc25 | 1299 | if (rx->flags & XEN_NETRXF_csum_blank) |
0d160211 | 1300 | skb->ip_summed = CHECKSUM_PARTIAL; |
f942dc25 | 1301 | else if (rx->flags & XEN_NETRXF_data_validated) |
0d160211 JF |
1302 | skb->ip_summed = CHECKSUM_UNNECESSARY; |
1303 | ||
1304 | __skb_queue_tail(&rxq, skb); | |
1305 | ||
900e0ac2 JG |
1306 | i = queue->rx.rsp_cons + 1; |
1307 | xennet_set_rx_rsp_cons(queue, i); | |
0d160211 JF |
1308 | work_done++; |
1309 | } | |
6c5aa6fc DK |
1310 | if (need_xdp_flush) |
1311 | xdp_do_flush(); | |
0d160211 | 1312 | |
56cfe5d0 | 1313 | __skb_queue_purge(&errq); |
0d160211 | 1314 | |
2688fcb7 | 1315 | work_done -= handle_incoming_queue(queue, &rxq); |
0d160211 | 1316 | |
2688fcb7 | 1317 | xennet_alloc_rx_buffers(queue); |
0d160211 | 1318 | |
0d160211 | 1319 | if (work_done < budget) { |
bea3348e SH |
1320 | int more_to_do = 0; |
1321 | ||
6ad20165 | 1322 | napi_complete_done(napi, work_done); |
0d160211 | 1323 | |
2688fcb7 | 1324 | RING_FINAL_CHECK_FOR_RESPONSES(&queue->rx, more_to_do); |
6a6dc08f DV |
1325 | if (more_to_do) |
1326 | napi_schedule(napi); | |
0d160211 JF |
1327 | } |
1328 | ||
2688fcb7 | 1329 | spin_unlock(&queue->rx_lock); |
0d160211 | 1330 | |
bea3348e | 1331 | return work_done; |
0d160211 JF |
1332 | } |
1333 | ||
1334 | static int xennet_change_mtu(struct net_device *dev, int mtu) | |
1335 | { | |
0c36820e | 1336 | int max = xennet_can_sg(dev) ? XEN_NETIF_MAX_TX_SIZE : ETH_DATA_LEN; |
0d160211 JF |
1337 | |
1338 | if (mtu > max) | |
1339 | return -EINVAL; | |
1340 | dev->mtu = mtu; | |
1341 | return 0; | |
1342 | } | |
1343 | ||
bc1f4470 | 1344 | static void xennet_get_stats64(struct net_device *dev, |
1345 | struct rtnl_link_stats64 *tot) | |
e00f85be | 1346 | { |
1347 | struct netfront_info *np = netdev_priv(dev); | |
1348 | int cpu; | |
1349 | ||
1350 | for_each_possible_cpu(cpu) { | |
900e1833 DV |
1351 | struct netfront_stats *rx_stats = per_cpu_ptr(np->rx_stats, cpu); |
1352 | struct netfront_stats *tx_stats = per_cpu_ptr(np->tx_stats, cpu); | |
e00f85be | 1353 | u64 rx_packets, rx_bytes, tx_packets, tx_bytes; |
1354 | unsigned int start; | |
1355 | ||
1356 | do { | |
900e1833 DV |
1357 | start = u64_stats_fetch_begin_irq(&tx_stats->syncp); |
1358 | tx_packets = tx_stats->packets; | |
1359 | tx_bytes = tx_stats->bytes; | |
1360 | } while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start)); | |
e00f85be | 1361 | |
900e1833 DV |
1362 | do { |
1363 | start = u64_stats_fetch_begin_irq(&rx_stats->syncp); | |
1364 | rx_packets = rx_stats->packets; | |
1365 | rx_bytes = rx_stats->bytes; | |
1366 | } while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start)); | |
e00f85be | 1367 | |
1368 | tot->rx_packets += rx_packets; | |
1369 | tot->tx_packets += tx_packets; | |
1370 | tot->rx_bytes += rx_bytes; | |
1371 | tot->tx_bytes += tx_bytes; | |
1372 | } | |
1373 | ||
1374 | tot->rx_errors = dev->stats.rx_errors; | |
1375 | tot->tx_dropped = dev->stats.tx_dropped; | |
e00f85be | 1376 | } |
1377 | ||
2688fcb7 | 1378 | static void xennet_release_tx_bufs(struct netfront_queue *queue) |
0d160211 JF |
1379 | { |
1380 | struct sk_buff *skb; | |
1381 | int i; | |
1382 | ||
1383 | for (i = 0; i < NET_TX_RING_SIZE; i++) { | |
1384 | /* Skip over entries which are actually freelist references */ | |
21631d2d | 1385 | if (!queue->tx_skbs[i]) |
0d160211 JF |
1386 | continue; |
1387 | ||
21631d2d JG |
1388 | skb = queue->tx_skbs[i]; |
1389 | queue->tx_skbs[i] = NULL; | |
2688fcb7 AB |
1390 | get_page(queue->grant_tx_page[i]); |
1391 | gnttab_end_foreign_access(queue->grant_tx_ref[i], | |
cefe0078 | 1392 | GNTMAP_readonly, |
2688fcb7 AB |
1393 | (unsigned long)page_address(queue->grant_tx_page[i])); |
1394 | queue->grant_tx_page[i] = NULL; | |
1395 | queue->grant_tx_ref[i] = GRANT_INVALID_REF; | |
21631d2d | 1396 | add_id_to_list(&queue->tx_skb_freelist, queue->tx_link, i); |
0d160211 JF |
1397 | dev_kfree_skb_irq(skb); |
1398 | } | |
1399 | } | |
1400 | ||
2688fcb7 | 1401 | static void xennet_release_rx_bufs(struct netfront_queue *queue) |
0d160211 | 1402 | { |
0d160211 JF |
1403 | int id, ref; |
1404 | ||
2688fcb7 | 1405 | spin_lock_bh(&queue->rx_lock); |
0d160211 JF |
1406 | |
1407 | for (id = 0; id < NET_RX_RING_SIZE; id++) { | |
cefe0078 AL |
1408 | struct sk_buff *skb; |
1409 | struct page *page; | |
0d160211 | 1410 | |
2688fcb7 | 1411 | skb = queue->rx_skbs[id]; |
cefe0078 | 1412 | if (!skb) |
0d160211 | 1413 | continue; |
0d160211 | 1414 | |
2688fcb7 | 1415 | ref = queue->grant_rx_ref[id]; |
cefe0078 AL |
1416 | if (ref == GRANT_INVALID_REF) |
1417 | continue; | |
0d160211 | 1418 | |
cefe0078 | 1419 | page = skb_frag_page(&skb_shinfo(skb)->frags[0]); |
0d160211 | 1420 | |
cefe0078 AL |
1421 | /* gnttab_end_foreign_access() needs a page ref until |
1422 | * foreign access is ended (which may be deferred). | |
1423 | */ | |
1424 | get_page(page); | |
1425 | gnttab_end_foreign_access(ref, 0, | |
1426 | (unsigned long)page_address(page)); | |
2688fcb7 | 1427 | queue->grant_rx_ref[id] = GRANT_INVALID_REF; |
0d160211 | 1428 | |
cefe0078 | 1429 | kfree_skb(skb); |
0d160211 JF |
1430 | } |
1431 | ||
2688fcb7 | 1432 | spin_unlock_bh(&queue->rx_lock); |
0d160211 JF |
1433 | } |
1434 | ||
c8f44aff MM |
1435 | static netdev_features_t xennet_fix_features(struct net_device *dev, |
1436 | netdev_features_t features) | |
8f7b01a1 ED |
1437 | { |
1438 | struct netfront_info *np = netdev_priv(dev); | |
8f7b01a1 | 1439 | |
2890ea5c JG |
1440 | if (features & NETIF_F_SG && |
1441 | !xenbus_read_unsigned(np->xbdev->otherend, "feature-sg", 0)) | |
1442 | features &= ~NETIF_F_SG; | |
8f7b01a1 | 1443 | |
2890ea5c JG |
1444 | if (features & NETIF_F_IPV6_CSUM && |
1445 | !xenbus_read_unsigned(np->xbdev->otherend, | |
1446 | "feature-ipv6-csum-offload", 0)) | |
1447 | features &= ~NETIF_F_IPV6_CSUM; | |
8f7b01a1 | 1448 | |
2890ea5c JG |
1449 | if (features & NETIF_F_TSO && |
1450 | !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv4", 0)) | |
1451 | features &= ~NETIF_F_TSO; | |
8f7b01a1 | 1452 | |
2890ea5c JG |
1453 | if (features & NETIF_F_TSO6 && |
1454 | !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv6", 0)) | |
1455 | features &= ~NETIF_F_TSO6; | |
2c0057de | 1456 | |
8f7b01a1 ED |
1457 | return features; |
1458 | } | |
1459 | ||
c8f44aff MM |
1460 | static int xennet_set_features(struct net_device *dev, |
1461 | netdev_features_t features) | |
8f7b01a1 ED |
1462 | { |
1463 | if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) { | |
1464 | netdev_info(dev, "Reducing MTU because no SG offload"); | |
1465 | dev->mtu = ETH_DATA_LEN; | |
1466 | } | |
1467 | ||
1468 | return 0; | |
1469 | } | |
1470 | ||
900e0ac2 | 1471 | static bool xennet_handle_tx(struct netfront_queue *queue, unsigned int *eoi) |
cf66f9d4 | 1472 | { |
cf66f9d4 KRW |
1473 | unsigned long flags; |
1474 | ||
900e0ac2 JG |
1475 | if (unlikely(queue->info->broken)) |
1476 | return false; | |
a884daa6 | 1477 | |
2688fcb7 | 1478 | spin_lock_irqsave(&queue->tx_lock, flags); |
900e0ac2 JG |
1479 | if (xennet_tx_buf_gc(queue)) |
1480 | *eoi = 0; | |
2688fcb7 | 1481 | spin_unlock_irqrestore(&queue->tx_lock, flags); |
cf66f9d4 | 1482 | |
900e0ac2 JG |
1483 | return true; |
1484 | } | |
1485 | ||
1486 | static irqreturn_t xennet_tx_interrupt(int irq, void *dev_id) | |
1487 | { | |
1488 | unsigned int eoiflag = XEN_EOI_FLAG_SPURIOUS; | |
1489 | ||
1490 | if (likely(xennet_handle_tx(dev_id, &eoiflag))) | |
1491 | xen_irq_lateeoi(irq, eoiflag); | |
1492 | ||
d634bf2c WL |
1493 | return IRQ_HANDLED; |
1494 | } | |
1495 | ||
900e0ac2 | 1496 | static bool xennet_handle_rx(struct netfront_queue *queue, unsigned int *eoi) |
d634bf2c | 1497 | { |
900e0ac2 JG |
1498 | unsigned int work_queued; |
1499 | unsigned long flags; | |
d634bf2c | 1500 | |
900e0ac2 JG |
1501 | if (unlikely(queue->info->broken)) |
1502 | return false; | |
1503 | ||
1504 | spin_lock_irqsave(&queue->rx_cons_lock, flags); | |
1505 | work_queued = RING_HAS_UNCONSUMED_RESPONSES(&queue->rx); | |
1506 | if (work_queued > queue->rx_rsp_unconsumed) { | |
1507 | queue->rx_rsp_unconsumed = work_queued; | |
1508 | *eoi = 0; | |
1509 | } else if (unlikely(work_queued < queue->rx_rsp_unconsumed)) { | |
1510 | const struct device *dev = &queue->info->netdev->dev; | |
1511 | ||
1512 | spin_unlock_irqrestore(&queue->rx_cons_lock, flags); | |
1513 | dev_alert(dev, "RX producer index going backwards\n"); | |
1514 | dev_alert(dev, "Disabled for further use\n"); | |
1515 | queue->info->broken = true; | |
1516 | return false; | |
1517 | } | |
1518 | spin_unlock_irqrestore(&queue->rx_cons_lock, flags); | |
a884daa6 | 1519 | |
900e0ac2 | 1520 | if (likely(netif_carrier_ok(queue->info->netdev) && work_queued)) |
76541869 | 1521 | napi_schedule(&queue->napi); |
cf66f9d4 | 1522 | |
900e0ac2 JG |
1523 | return true; |
1524 | } | |
1525 | ||
1526 | static irqreturn_t xennet_rx_interrupt(int irq, void *dev_id) | |
1527 | { | |
1528 | unsigned int eoiflag = XEN_EOI_FLAG_SPURIOUS; | |
1529 | ||
1530 | if (likely(xennet_handle_rx(dev_id, &eoiflag))) | |
1531 | xen_irq_lateeoi(irq, eoiflag); | |
1532 | ||
d634bf2c WL |
1533 | return IRQ_HANDLED; |
1534 | } | |
cf66f9d4 | 1535 | |
d634bf2c WL |
1536 | static irqreturn_t xennet_interrupt(int irq, void *dev_id) |
1537 | { | |
900e0ac2 JG |
1538 | unsigned int eoiflag = XEN_EOI_FLAG_SPURIOUS; |
1539 | ||
1540 | if (xennet_handle_tx(dev_id, &eoiflag) && | |
1541 | xennet_handle_rx(dev_id, &eoiflag)) | |
1542 | xen_irq_lateeoi(irq, eoiflag); | |
1543 | ||
cf66f9d4 KRW |
1544 | return IRQ_HANDLED; |
1545 | } | |
1546 | ||
1547 | #ifdef CONFIG_NET_POLL_CONTROLLER | |
1548 | static void xennet_poll_controller(struct net_device *dev) | |
1549 | { | |
2688fcb7 AB |
1550 | /* Poll each queue */ |
1551 | struct netfront_info *info = netdev_priv(dev); | |
1552 | unsigned int num_queues = dev->real_num_tx_queues; | |
1553 | unsigned int i; | |
a884daa6 JG |
1554 | |
1555 | if (info->broken) | |
1556 | return; | |
1557 | ||
2688fcb7 AB |
1558 | for (i = 0; i < num_queues; ++i) |
1559 | xennet_interrupt(0, &info->queues[i]); | |
cf66f9d4 KRW |
1560 | } |
1561 | #endif | |
1562 | ||
6c5aa6fc DK |
1563 | #define NETBACK_XDP_HEADROOM_DISABLE 0 |
1564 | #define NETBACK_XDP_HEADROOM_ENABLE 1 | |
1565 | ||
1566 | static int talk_to_netback_xdp(struct netfront_info *np, int xdp) | |
1567 | { | |
1568 | int err; | |
1569 | unsigned short headroom; | |
1570 | ||
1571 | headroom = xdp ? XDP_PACKET_HEADROOM : 0; | |
1572 | err = xenbus_printf(XBT_NIL, np->xbdev->nodename, | |
1573 | "xdp-headroom", "%hu", | |
1574 | headroom); | |
1575 | if (err) | |
1576 | pr_warn("Error writing xdp-headroom\n"); | |
1577 | ||
1578 | return err; | |
1579 | } | |
1580 | ||
1581 | static int xennet_xdp_set(struct net_device *dev, struct bpf_prog *prog, | |
1582 | struct netlink_ext_ack *extack) | |
1583 | { | |
1584 | unsigned long max_mtu = XEN_PAGE_SIZE - XDP_PACKET_HEADROOM; | |
1585 | struct netfront_info *np = netdev_priv(dev); | |
1586 | struct bpf_prog *old_prog; | |
1587 | unsigned int i, err; | |
1588 | ||
1589 | if (dev->mtu > max_mtu) { | |
1590 | netdev_warn(dev, "XDP requires MTU less than %lu\n", max_mtu); | |
1591 | return -EINVAL; | |
1592 | } | |
1593 | ||
1594 | if (!np->netback_has_xdp_headroom) | |
1595 | return 0; | |
1596 | ||
1597 | xenbus_switch_state(np->xbdev, XenbusStateReconfiguring); | |
1598 | ||
1599 | err = talk_to_netback_xdp(np, prog ? NETBACK_XDP_HEADROOM_ENABLE : | |
1600 | NETBACK_XDP_HEADROOM_DISABLE); | |
1601 | if (err) | |
1602 | return err; | |
1603 | ||
1604 | /* avoid the race with XDP headroom adjustment */ | |
1605 | wait_event(module_wq, | |
1606 | xenbus_read_driver_state(np->xbdev->otherend) == | |
1607 | XenbusStateReconfigured); | |
1608 | np->netfront_xdp_enabled = true; | |
1609 | ||
1610 | old_prog = rtnl_dereference(np->queues[0].xdp_prog); | |
1611 | ||
1612 | if (prog) | |
1613 | bpf_prog_add(prog, dev->real_num_tx_queues); | |
1614 | ||
1615 | for (i = 0; i < dev->real_num_tx_queues; ++i) | |
1616 | rcu_assign_pointer(np->queues[i].xdp_prog, prog); | |
1617 | ||
1618 | if (old_prog) | |
1619 | for (i = 0; i < dev->real_num_tx_queues; ++i) | |
1620 | bpf_prog_put(old_prog); | |
1621 | ||
1622 | xenbus_switch_state(np->xbdev, XenbusStateConnected); | |
1623 | ||
1624 | return 0; | |
1625 | } | |
1626 | ||
6c5aa6fc DK |
1627 | static int xennet_xdp(struct net_device *dev, struct netdev_bpf *xdp) |
1628 | { | |
a884daa6 JG |
1629 | struct netfront_info *np = netdev_priv(dev); |
1630 | ||
1631 | if (np->broken) | |
1632 | return -ENODEV; | |
1633 | ||
6c5aa6fc DK |
1634 | switch (xdp->command) { |
1635 | case XDP_SETUP_PROG: | |
1636 | return xennet_xdp_set(dev, xdp->prog, xdp->extack); | |
6c5aa6fc DK |
1637 | default: |
1638 | return -EINVAL; | |
1639 | } | |
1640 | } | |
1641 | ||
0a0b9d2e | 1642 | static const struct net_device_ops xennet_netdev_ops = { |
1d0ea124 | 1643 | .ndo_uninit = xennet_uninit, |
0a0b9d2e | 1644 | .ndo_open = xennet_open, |
0a0b9d2e SH |
1645 | .ndo_stop = xennet_close, |
1646 | .ndo_start_xmit = xennet_start_xmit, | |
1647 | .ndo_change_mtu = xennet_change_mtu, | |
e00f85be | 1648 | .ndo_get_stats64 = xennet_get_stats64, |
0a0b9d2e SH |
1649 | .ndo_set_mac_address = eth_mac_addr, |
1650 | .ndo_validate_addr = eth_validate_addr, | |
fb507934 MM |
1651 | .ndo_fix_features = xennet_fix_features, |
1652 | .ndo_set_features = xennet_set_features, | |
2688fcb7 | 1653 | .ndo_select_queue = xennet_select_queue, |
6c5aa6fc DK |
1654 | .ndo_bpf = xennet_xdp, |
1655 | .ndo_xdp_xmit = xennet_xdp_xmit, | |
cf66f9d4 KRW |
1656 | #ifdef CONFIG_NET_POLL_CONTROLLER |
1657 | .ndo_poll_controller = xennet_poll_controller, | |
1658 | #endif | |
0a0b9d2e SH |
1659 | }; |
1660 | ||
900e1833 DV |
1661 | static void xennet_free_netdev(struct net_device *netdev) |
1662 | { | |
1663 | struct netfront_info *np = netdev_priv(netdev); | |
1664 | ||
1665 | free_percpu(np->rx_stats); | |
1666 | free_percpu(np->tx_stats); | |
1667 | free_netdev(netdev); | |
1668 | } | |
1669 | ||
8e0e46bb | 1670 | static struct net_device *xennet_create_dev(struct xenbus_device *dev) |
0d160211 | 1671 | { |
2688fcb7 | 1672 | int err; |
0d160211 JF |
1673 | struct net_device *netdev; |
1674 | struct netfront_info *np; | |
1675 | ||
50ee6061 | 1676 | netdev = alloc_etherdev_mq(sizeof(struct netfront_info), xennet_max_queues); |
41de8d4c | 1677 | if (!netdev) |
0d160211 | 1678 | return ERR_PTR(-ENOMEM); |
0d160211 JF |
1679 | |
1680 | np = netdev_priv(netdev); | |
1681 | np->xbdev = dev; | |
1682 | ||
2688fcb7 | 1683 | np->queues = NULL; |
0d160211 | 1684 | |
e00f85be | 1685 | err = -ENOMEM; |
900e1833 DV |
1686 | np->rx_stats = netdev_alloc_pcpu_stats(struct netfront_stats); |
1687 | if (np->rx_stats == NULL) | |
1688 | goto exit; | |
1689 | np->tx_stats = netdev_alloc_pcpu_stats(struct netfront_stats); | |
1690 | if (np->tx_stats == NULL) | |
e00f85be | 1691 | goto exit; |
1692 | ||
0a0b9d2e SH |
1693 | netdev->netdev_ops = &xennet_netdev_ops; |
1694 | ||
fb507934 MM |
1695 | netdev->features = NETIF_F_IP_CSUM | NETIF_F_RXCSUM | |
1696 | NETIF_F_GSO_ROBUST; | |
2c0057de PD |
1697 | netdev->hw_features = NETIF_F_SG | |
1698 | NETIF_F_IPV6_CSUM | | |
1699 | NETIF_F_TSO | NETIF_F_TSO6; | |
0d160211 | 1700 | |
fc3e5941 IC |
1701 | /* |
1702 | * Assume that all hw features are available for now. This set | |
1703 | * will be adjusted by the call to netdev_update_features() in | |
1704 | * xennet_connect() which is the earliest point where we can | |
1705 | * negotiate with the backend regarding supported features. | |
1706 | */ | |
1707 | netdev->features |= netdev->hw_features; | |
1708 | ||
7ad24ea4 | 1709 | netdev->ethtool_ops = &xennet_ethtool_ops; |
e1043a4b | 1710 | netdev->min_mtu = ETH_MIN_MTU; |
d0c2c997 | 1711 | netdev->max_mtu = XEN_NETIF_MAX_TX_SIZE; |
0d160211 JF |
1712 | SET_NETDEV_DEV(netdev, &dev->dev); |
1713 | ||
1714 | np->netdev = netdev; | |
6c5aa6fc | 1715 | np->netfront_xdp_enabled = false; |
0d160211 JF |
1716 | |
1717 | netif_carrier_off(netdev); | |
1718 | ||
c2c63310 AR |
1719 | do { |
1720 | xenbus_switch_state(dev, XenbusStateInitialising); | |
1721 | err = wait_event_timeout(module_wq, | |
1722 | xenbus_read_driver_state(dev->otherend) != | |
1723 | XenbusStateClosed && | |
1724 | xenbus_read_driver_state(dev->otherend) != | |
1725 | XenbusStateUnknown, XENNET_TIMEOUT); | |
1726 | } while (!err); | |
1727 | ||
0d160211 JF |
1728 | return netdev; |
1729 | ||
0d160211 | 1730 | exit: |
900e1833 | 1731 | xennet_free_netdev(netdev); |
0d160211 JF |
1732 | return ERR_PTR(err); |
1733 | } | |
1734 | ||
80708602 | 1735 | /* |
0d160211 JF |
1736 | * Entry point to this code when a new device is created. Allocate the basic |
1737 | * structures and the ring buffers for communication with the backend, and | |
1738 | * inform the backend of the appropriate details for those. | |
1739 | */ | |
8e0e46bb | 1740 | static int netfront_probe(struct xenbus_device *dev, |
1dd06ae8 | 1741 | const struct xenbus_device_id *id) |
0d160211 JF |
1742 | { |
1743 | int err; | |
1744 | struct net_device *netdev; | |
1745 | struct netfront_info *info; | |
1746 | ||
1747 | netdev = xennet_create_dev(dev); | |
1748 | if (IS_ERR(netdev)) { | |
1749 | err = PTR_ERR(netdev); | |
1750 | xenbus_dev_fatal(dev, err, "creating netdev"); | |
1751 | return err; | |
1752 | } | |
1753 | ||
1754 | info = netdev_priv(netdev); | |
1b713e00 | 1755 | dev_set_drvdata(&dev->dev, info); |
27b917e5 TI |
1756 | #ifdef CONFIG_SYSFS |
1757 | info->netdev->sysfs_groups[0] = &xennet_dev_group; | |
1758 | #endif | |
0d160211 | 1759 | |
0d160211 | 1760 | return 0; |
0d160211 JF |
1761 | } |
1762 | ||
1763 | static void xennet_end_access(int ref, void *page) | |
1764 | { | |
1765 | /* This frees the page as a side-effect */ | |
1766 | if (ref != GRANT_INVALID_REF) | |
1767 | gnttab_end_foreign_access(ref, 0, (unsigned long)page); | |
1768 | } | |
1769 | ||
1770 | static void xennet_disconnect_backend(struct netfront_info *info) | |
1771 | { | |
2688fcb7 | 1772 | unsigned int i = 0; |
2688fcb7 AB |
1773 | unsigned int num_queues = info->netdev->real_num_tx_queues; |
1774 | ||
f9feb1e6 DV |
1775 | netif_carrier_off(info->netdev); |
1776 | ||
9a873c71 | 1777 | for (i = 0; i < num_queues && info->queues; ++i) { |
76541869 DV |
1778 | struct netfront_queue *queue = &info->queues[i]; |
1779 | ||
74470954 BO |
1780 | del_timer_sync(&queue->rx_refill_timer); |
1781 | ||
2688fcb7 AB |
1782 | if (queue->tx_irq && (queue->tx_irq == queue->rx_irq)) |
1783 | unbind_from_irqhandler(queue->tx_irq, queue); | |
1784 | if (queue->tx_irq && (queue->tx_irq != queue->rx_irq)) { | |
1785 | unbind_from_irqhandler(queue->tx_irq, queue); | |
1786 | unbind_from_irqhandler(queue->rx_irq, queue); | |
1787 | } | |
1788 | queue->tx_evtchn = queue->rx_evtchn = 0; | |
1789 | queue->tx_irq = queue->rx_irq = 0; | |
0d160211 | 1790 | |
274b0455 CW |
1791 | if (netif_running(info->netdev)) |
1792 | napi_synchronize(&queue->napi); | |
f9feb1e6 | 1793 | |
a5b5dc3c DV |
1794 | xennet_release_tx_bufs(queue); |
1795 | xennet_release_rx_bufs(queue); | |
1796 | gnttab_free_grant_references(queue->gref_tx_head); | |
1797 | gnttab_free_grant_references(queue->gref_rx_head); | |
1798 | ||
2688fcb7 AB |
1799 | /* End access and free the pages */ |
1800 | xennet_end_access(queue->tx_ring_ref, queue->tx.sring); | |
1801 | xennet_end_access(queue->rx_ring_ref, queue->rx.sring); | |
0d160211 | 1802 | |
2688fcb7 AB |
1803 | queue->tx_ring_ref = GRANT_INVALID_REF; |
1804 | queue->rx_ring_ref = GRANT_INVALID_REF; | |
1805 | queue->tx.sring = NULL; | |
1806 | queue->rx.sring = NULL; | |
6c5aa6fc DK |
1807 | |
1808 | page_pool_destroy(queue->page_pool); | |
2688fcb7 | 1809 | } |
0d160211 JF |
1810 | } |
1811 | ||
80708602 | 1812 | /* |
0d160211 JF |
1813 | * We are reconnecting to the backend, due to a suspend/resume, or a backend |
1814 | * driver restart. We tear down our netif structure and recreate it, but | |
1815 | * leave the device-layer structures intact so that this is transparent to the | |
1816 | * rest of the kernel. | |
1817 | */ | |
1818 | static int netfront_resume(struct xenbus_device *dev) | |
1819 | { | |
1b713e00 | 1820 | struct netfront_info *info = dev_get_drvdata(&dev->dev); |
0d160211 JF |
1821 | |
1822 | dev_dbg(&dev->dev, "%s\n", dev->nodename); | |
1823 | ||
042b2046 DZ |
1824 | netif_tx_lock_bh(info->netdev); |
1825 | netif_device_detach(info->netdev); | |
1826 | netif_tx_unlock_bh(info->netdev); | |
1827 | ||
0d160211 JF |
1828 | xennet_disconnect_backend(info); |
1829 | return 0; | |
1830 | } | |
1831 | ||
1832 | static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[]) | |
1833 | { | |
1834 | char *s, *e, *macstr; | |
1835 | int i; | |
1836 | ||
1837 | macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL); | |
1838 | if (IS_ERR(macstr)) | |
1839 | return PTR_ERR(macstr); | |
1840 | ||
1841 | for (i = 0; i < ETH_ALEN; i++) { | |
1842 | mac[i] = simple_strtoul(s, &e, 16); | |
1843 | if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) { | |
1844 | kfree(macstr); | |
1845 | return -ENOENT; | |
1846 | } | |
1847 | s = e+1; | |
1848 | } | |
1849 | ||
1850 | kfree(macstr); | |
1851 | return 0; | |
1852 | } | |
1853 | ||
2688fcb7 | 1854 | static int setup_netfront_single(struct netfront_queue *queue) |
d634bf2c WL |
1855 | { |
1856 | int err; | |
1857 | ||
2688fcb7 | 1858 | err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn); |
d634bf2c WL |
1859 | if (err < 0) |
1860 | goto fail; | |
1861 | ||
900e0ac2 JG |
1862 | err = bind_evtchn_to_irqhandler_lateeoi(queue->tx_evtchn, |
1863 | xennet_interrupt, 0, | |
1864 | queue->info->netdev->name, | |
1865 | queue); | |
d634bf2c WL |
1866 | if (err < 0) |
1867 | goto bind_fail; | |
2688fcb7 AB |
1868 | queue->rx_evtchn = queue->tx_evtchn; |
1869 | queue->rx_irq = queue->tx_irq = err; | |
d634bf2c WL |
1870 | |
1871 | return 0; | |
1872 | ||
1873 | bind_fail: | |
2688fcb7 AB |
1874 | xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn); |
1875 | queue->tx_evtchn = 0; | |
d634bf2c WL |
1876 | fail: |
1877 | return err; | |
1878 | } | |
1879 | ||
2688fcb7 | 1880 | static int setup_netfront_split(struct netfront_queue *queue) |
d634bf2c WL |
1881 | { |
1882 | int err; | |
1883 | ||
2688fcb7 | 1884 | err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn); |
d634bf2c WL |
1885 | if (err < 0) |
1886 | goto fail; | |
2688fcb7 | 1887 | err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->rx_evtchn); |
d634bf2c WL |
1888 | if (err < 0) |
1889 | goto alloc_rx_evtchn_fail; | |
1890 | ||
2688fcb7 AB |
1891 | snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name), |
1892 | "%s-tx", queue->name); | |
900e0ac2 JG |
1893 | err = bind_evtchn_to_irqhandler_lateeoi(queue->tx_evtchn, |
1894 | xennet_tx_interrupt, 0, | |
1895 | queue->tx_irq_name, queue); | |
d634bf2c WL |
1896 | if (err < 0) |
1897 | goto bind_tx_fail; | |
2688fcb7 | 1898 | queue->tx_irq = err; |
d634bf2c | 1899 | |
2688fcb7 AB |
1900 | snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name), |
1901 | "%s-rx", queue->name); | |
900e0ac2 JG |
1902 | err = bind_evtchn_to_irqhandler_lateeoi(queue->rx_evtchn, |
1903 | xennet_rx_interrupt, 0, | |
1904 | queue->rx_irq_name, queue); | |
d634bf2c WL |
1905 | if (err < 0) |
1906 | goto bind_rx_fail; | |
2688fcb7 | 1907 | queue->rx_irq = err; |
d634bf2c WL |
1908 | |
1909 | return 0; | |
1910 | ||
1911 | bind_rx_fail: | |
2688fcb7 AB |
1912 | unbind_from_irqhandler(queue->tx_irq, queue); |
1913 | queue->tx_irq = 0; | |
d634bf2c | 1914 | bind_tx_fail: |
2688fcb7 AB |
1915 | xenbus_free_evtchn(queue->info->xbdev, queue->rx_evtchn); |
1916 | queue->rx_evtchn = 0; | |
d634bf2c | 1917 | alloc_rx_evtchn_fail: |
2688fcb7 AB |
1918 | xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn); |
1919 | queue->tx_evtchn = 0; | |
d634bf2c WL |
1920 | fail: |
1921 | return err; | |
1922 | } | |
1923 | ||
2688fcb7 AB |
1924 | static int setup_netfront(struct xenbus_device *dev, |
1925 | struct netfront_queue *queue, unsigned int feature_split_evtchn) | |
0d160211 JF |
1926 | { |
1927 | struct xen_netif_tx_sring *txs; | |
60a69176 | 1928 | struct xen_netif_rx_sring *rxs = NULL; |
ccc9d90a | 1929 | grant_ref_t gref; |
0d160211 | 1930 | int err; |
0d160211 | 1931 | |
2688fcb7 AB |
1932 | queue->tx_ring_ref = GRANT_INVALID_REF; |
1933 | queue->rx_ring_ref = GRANT_INVALID_REF; | |
1934 | queue->rx.sring = NULL; | |
1935 | queue->tx.sring = NULL; | |
0d160211 | 1936 | |
a144ff09 | 1937 | txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH); |
0d160211 JF |
1938 | if (!txs) { |
1939 | err = -ENOMEM; | |
1940 | xenbus_dev_fatal(dev, err, "allocating tx ring page"); | |
1941 | goto fail; | |
1942 | } | |
1943 | SHARED_RING_INIT(txs); | |
30c5d7f0 | 1944 | FRONT_RING_INIT(&queue->tx, txs, XEN_PAGE_SIZE); |
0d160211 | 1945 | |
ccc9d90a | 1946 | err = xenbus_grant_ring(dev, txs, 1, &gref); |
1ca2983a | 1947 | if (err < 0) |
60a69176 | 1948 | goto fail; |
ccc9d90a | 1949 | queue->tx_ring_ref = gref; |
0d160211 | 1950 | |
a144ff09 | 1951 | rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH); |
0d160211 JF |
1952 | if (!rxs) { |
1953 | err = -ENOMEM; | |
1954 | xenbus_dev_fatal(dev, err, "allocating rx ring page"); | |
60a69176 | 1955 | goto fail; |
0d160211 JF |
1956 | } |
1957 | SHARED_RING_INIT(rxs); | |
30c5d7f0 | 1958 | FRONT_RING_INIT(&queue->rx, rxs, XEN_PAGE_SIZE); |
0d160211 | 1959 | |
ccc9d90a | 1960 | err = xenbus_grant_ring(dev, rxs, 1, &gref); |
1ca2983a | 1961 | if (err < 0) |
60a69176 | 1962 | goto fail; |
ccc9d90a | 1963 | queue->rx_ring_ref = gref; |
0d160211 | 1964 | |
d634bf2c | 1965 | if (feature_split_evtchn) |
2688fcb7 | 1966 | err = setup_netfront_split(queue); |
d634bf2c WL |
1967 | /* setup single event channel if |
1968 | * a) feature-split-event-channels == 0 | |
1969 | * b) feature-split-event-channels == 1 but failed to setup | |
1970 | */ | |
e93fac3b | 1971 | if (!feature_split_evtchn || err) |
2688fcb7 | 1972 | err = setup_netfront_single(queue); |
d634bf2c | 1973 | |
0d160211 | 1974 | if (err) |
60a69176 | 1975 | goto fail; |
0d160211 | 1976 | |
0d160211 JF |
1977 | return 0; |
1978 | ||
1ca2983a WL |
1979 | /* If we fail to setup netfront, it is safe to just revoke access to |
1980 | * granted pages because backend is not accessing it at this point. | |
1981 | */ | |
60a69176 JG |
1982 | fail: |
1983 | if (queue->rx_ring_ref != GRANT_INVALID_REF) { | |
1984 | gnttab_end_foreign_access(queue->rx_ring_ref, 0, | |
1985 | (unsigned long)rxs); | |
1986 | queue->rx_ring_ref = GRANT_INVALID_REF; | |
1987 | } else { | |
1988 | free_page((unsigned long)rxs); | |
1989 | } | |
1990 | if (queue->tx_ring_ref != GRANT_INVALID_REF) { | |
1991 | gnttab_end_foreign_access(queue->tx_ring_ref, 0, | |
1992 | (unsigned long)txs); | |
1993 | queue->tx_ring_ref = GRANT_INVALID_REF; | |
1994 | } else { | |
1995 | free_page((unsigned long)txs); | |
1996 | } | |
0d160211 JF |
1997 | return err; |
1998 | } | |
1999 | ||
2688fcb7 AB |
2000 | /* Queue-specific initialisation |
2001 | * This used to be done in xennet_create_dev() but must now | |
2002 | * be run per-queue. | |
2003 | */ | |
2004 | static int xennet_init_queue(struct netfront_queue *queue) | |
2005 | { | |
2006 | unsigned short i; | |
2007 | int err = 0; | |
21f2706b | 2008 | char *devid; |
2688fcb7 AB |
2009 | |
2010 | spin_lock_init(&queue->tx_lock); | |
2011 | spin_lock_init(&queue->rx_lock); | |
900e0ac2 | 2012 | spin_lock_init(&queue->rx_cons_lock); |
2688fcb7 | 2013 | |
e99e88a9 | 2014 | timer_setup(&queue->rx_refill_timer, rx_refill_timeout, 0); |
2688fcb7 | 2015 | |
21f2706b XL |
2016 | devid = strrchr(queue->info->xbdev->nodename, '/') + 1; |
2017 | snprintf(queue->name, sizeof(queue->name), "vif%s-q%u", | |
2018 | devid, queue->id); | |
8b715010 | 2019 | |
21631d2d | 2020 | /* Initialise tx_skb_freelist as a free chain containing every entry. */ |
2688fcb7 | 2021 | queue->tx_skb_freelist = 0; |
a884daa6 | 2022 | queue->tx_pend_queue = TX_LINK_NONE; |
2688fcb7 | 2023 | for (i = 0; i < NET_TX_RING_SIZE; i++) { |
21631d2d | 2024 | queue->tx_link[i] = i + 1; |
2688fcb7 AB |
2025 | queue->grant_tx_ref[i] = GRANT_INVALID_REF; |
2026 | queue->grant_tx_page[i] = NULL; | |
2027 | } | |
21631d2d | 2028 | queue->tx_link[NET_TX_RING_SIZE - 1] = TX_LINK_NONE; |
2688fcb7 AB |
2029 | |
2030 | /* Clear out rx_skbs */ | |
2031 | for (i = 0; i < NET_RX_RING_SIZE; i++) { | |
2032 | queue->rx_skbs[i] = NULL; | |
2033 | queue->grant_rx_ref[i] = GRANT_INVALID_REF; | |
2034 | } | |
2035 | ||
2036 | /* A grant for every tx ring slot */ | |
1f3c2eba | 2037 | if (gnttab_alloc_grant_references(NET_TX_RING_SIZE, |
2688fcb7 AB |
2038 | &queue->gref_tx_head) < 0) { |
2039 | pr_alert("can't alloc tx grant refs\n"); | |
2040 | err = -ENOMEM; | |
2041 | goto exit; | |
2042 | } | |
2043 | ||
2044 | /* A grant for every rx ring slot */ | |
1f3c2eba | 2045 | if (gnttab_alloc_grant_references(NET_RX_RING_SIZE, |
2688fcb7 AB |
2046 | &queue->gref_rx_head) < 0) { |
2047 | pr_alert("can't alloc rx grant refs\n"); | |
2048 | err = -ENOMEM; | |
2049 | goto exit_free_tx; | |
2050 | } | |
2051 | ||
2688fcb7 AB |
2052 | return 0; |
2053 | ||
2054 | exit_free_tx: | |
2055 | gnttab_free_grant_references(queue->gref_tx_head); | |
2056 | exit: | |
2057 | return err; | |
2058 | } | |
2059 | ||
50ee6061 AB |
2060 | static int write_queue_xenstore_keys(struct netfront_queue *queue, |
2061 | struct xenbus_transaction *xbt, int write_hierarchical) | |
2062 | { | |
2063 | /* Write the queue-specific keys into XenStore in the traditional | |
2064 | * way for a single queue, or in a queue subkeys for multiple | |
2065 | * queues. | |
2066 | */ | |
2067 | struct xenbus_device *dev = queue->info->xbdev; | |
2068 | int err; | |
2069 | const char *message; | |
2070 | char *path; | |
2071 | size_t pathsize; | |
2072 | ||
2073 | /* Choose the correct place to write the keys */ | |
2074 | if (write_hierarchical) { | |
2075 | pathsize = strlen(dev->nodename) + 10; | |
2076 | path = kzalloc(pathsize, GFP_KERNEL); | |
2077 | if (!path) { | |
2078 | err = -ENOMEM; | |
2079 | message = "out of memory while writing ring references"; | |
2080 | goto error; | |
2081 | } | |
2082 | snprintf(path, pathsize, "%s/queue-%u", | |
2083 | dev->nodename, queue->id); | |
2084 | } else { | |
2085 | path = (char *)dev->nodename; | |
2086 | } | |
2087 | ||
2088 | /* Write ring references */ | |
2089 | err = xenbus_printf(*xbt, path, "tx-ring-ref", "%u", | |
2090 | queue->tx_ring_ref); | |
2091 | if (err) { | |
2092 | message = "writing tx-ring-ref"; | |
2093 | goto error; | |
2094 | } | |
2095 | ||
2096 | err = xenbus_printf(*xbt, path, "rx-ring-ref", "%u", | |
2097 | queue->rx_ring_ref); | |
2098 | if (err) { | |
2099 | message = "writing rx-ring-ref"; | |
2100 | goto error; | |
2101 | } | |
2102 | ||
2103 | /* Write event channels; taking into account both shared | |
2104 | * and split event channel scenarios. | |
2105 | */ | |
2106 | if (queue->tx_evtchn == queue->rx_evtchn) { | |
2107 | /* Shared event channel */ | |
2108 | err = xenbus_printf(*xbt, path, | |
2109 | "event-channel", "%u", queue->tx_evtchn); | |
2110 | if (err) { | |
2111 | message = "writing event-channel"; | |
2112 | goto error; | |
2113 | } | |
2114 | } else { | |
2115 | /* Split event channels */ | |
2116 | err = xenbus_printf(*xbt, path, | |
2117 | "event-channel-tx", "%u", queue->tx_evtchn); | |
2118 | if (err) { | |
2119 | message = "writing event-channel-tx"; | |
2120 | goto error; | |
2121 | } | |
2122 | ||
2123 | err = xenbus_printf(*xbt, path, | |
2124 | "event-channel-rx", "%u", queue->rx_evtchn); | |
2125 | if (err) { | |
2126 | message = "writing event-channel-rx"; | |
2127 | goto error; | |
2128 | } | |
2129 | } | |
2130 | ||
2131 | if (write_hierarchical) | |
2132 | kfree(path); | |
2133 | return 0; | |
2134 | ||
2135 | error: | |
2136 | if (write_hierarchical) | |
2137 | kfree(path); | |
2138 | xenbus_dev_fatal(dev, err, "%s", message); | |
2139 | return err; | |
2140 | } | |
2141 | ||
6c5aa6fc DK |
2142 | |
2143 | ||
2144 | static int xennet_create_page_pool(struct netfront_queue *queue) | |
2145 | { | |
2146 | int err; | |
2147 | struct page_pool_params pp_params = { | |
2148 | .order = 0, | |
2149 | .flags = 0, | |
2150 | .pool_size = NET_RX_RING_SIZE, | |
2151 | .nid = NUMA_NO_NODE, | |
2152 | .dev = &queue->info->netdev->dev, | |
2153 | .offset = XDP_PACKET_HEADROOM, | |
2154 | .max_len = XEN_PAGE_SIZE - XDP_PACKET_HEADROOM, | |
2155 | }; | |
2156 | ||
2157 | queue->page_pool = page_pool_create(&pp_params); | |
2158 | if (IS_ERR(queue->page_pool)) { | |
2159 | err = PTR_ERR(queue->page_pool); | |
2160 | queue->page_pool = NULL; | |
2161 | return err; | |
2162 | } | |
2163 | ||
2164 | err = xdp_rxq_info_reg(&queue->xdp_rxq, queue->info->netdev, | |
b02e5a0e | 2165 | queue->id, 0); |
6c5aa6fc DK |
2166 | if (err) { |
2167 | netdev_err(queue->info->netdev, "xdp_rxq_info_reg failed\n"); | |
2168 | goto err_free_pp; | |
2169 | } | |
2170 | ||
2171 | err = xdp_rxq_info_reg_mem_model(&queue->xdp_rxq, | |
2172 | MEM_TYPE_PAGE_POOL, queue->page_pool); | |
2173 | if (err) { | |
2174 | netdev_err(queue->info->netdev, "xdp_rxq_info_reg_mem_model failed\n"); | |
2175 | goto err_unregister_rxq; | |
2176 | } | |
2177 | return 0; | |
2178 | ||
2179 | err_unregister_rxq: | |
2180 | xdp_rxq_info_unreg(&queue->xdp_rxq); | |
2181 | err_free_pp: | |
2182 | page_pool_destroy(queue->page_pool); | |
2183 | queue->page_pool = NULL; | |
2184 | return err; | |
2185 | } | |
2186 | ||
ce58725f | 2187 | static int xennet_create_queues(struct netfront_info *info, |
ca88ea12 | 2188 | unsigned int *num_queues) |
ce58725f DV |
2189 | { |
2190 | unsigned int i; | |
2191 | int ret; | |
2192 | ||
ca88ea12 | 2193 | info->queues = kcalloc(*num_queues, sizeof(struct netfront_queue), |
ce58725f DV |
2194 | GFP_KERNEL); |
2195 | if (!info->queues) | |
2196 | return -ENOMEM; | |
2197 | ||
ca88ea12 | 2198 | for (i = 0; i < *num_queues; i++) { |
ce58725f DV |
2199 | struct netfront_queue *queue = &info->queues[i]; |
2200 | ||
2201 | queue->id = i; | |
2202 | queue->info = info; | |
2203 | ||
2204 | ret = xennet_init_queue(queue); | |
2205 | if (ret < 0) { | |
f599c64f | 2206 | dev_warn(&info->xbdev->dev, |
69cb8524 | 2207 | "only created %d queues\n", i); |
ca88ea12 | 2208 | *num_queues = i; |
ce58725f DV |
2209 | break; |
2210 | } | |
2211 | ||
6c5aa6fc DK |
2212 | /* use page pool recycling instead of buddy allocator */ |
2213 | ret = xennet_create_page_pool(queue); | |
2214 | if (ret < 0) { | |
2215 | dev_err(&info->xbdev->dev, "can't allocate page pool\n"); | |
2216 | *num_queues = i; | |
2217 | return ret; | |
2218 | } | |
2219 | ||
ce58725f DV |
2220 | netif_napi_add(queue->info->netdev, &queue->napi, |
2221 | xennet_poll, 64); | |
2222 | if (netif_running(info->netdev)) | |
2223 | napi_enable(&queue->napi); | |
2224 | } | |
2225 | ||
ca88ea12 | 2226 | netif_set_real_num_tx_queues(info->netdev, *num_queues); |
ce58725f | 2227 | |
ca88ea12 | 2228 | if (*num_queues == 0) { |
f599c64f | 2229 | dev_err(&info->xbdev->dev, "no queues\n"); |
ce58725f DV |
2230 | return -EINVAL; |
2231 | } | |
2232 | return 0; | |
2233 | } | |
2234 | ||
0d160211 | 2235 | /* Common code used when first setting up, and when resuming. */ |
f502bf2b | 2236 | static int talk_to_netback(struct xenbus_device *dev, |
0d160211 JF |
2237 | struct netfront_info *info) |
2238 | { | |
2239 | const char *message; | |
2240 | struct xenbus_transaction xbt; | |
2241 | int err; | |
2688fcb7 AB |
2242 | unsigned int feature_split_evtchn; |
2243 | unsigned int i = 0; | |
50ee6061 | 2244 | unsigned int max_queues = 0; |
2688fcb7 AB |
2245 | struct netfront_queue *queue = NULL; |
2246 | unsigned int num_queues = 1; | |
0d160211 | 2247 | |
2688fcb7 AB |
2248 | info->netdev->irq = 0; |
2249 | ||
50ee6061 | 2250 | /* Check if backend supports multiple queues */ |
2890ea5c JG |
2251 | max_queues = xenbus_read_unsigned(info->xbdev->otherend, |
2252 | "multi-queue-max-queues", 1); | |
50ee6061 AB |
2253 | num_queues = min(max_queues, xennet_max_queues); |
2254 | ||
2688fcb7 | 2255 | /* Check feature-split-event-channels */ |
2890ea5c JG |
2256 | feature_split_evtchn = xenbus_read_unsigned(info->xbdev->otherend, |
2257 | "feature-split-event-channels", 0); | |
2688fcb7 AB |
2258 | |
2259 | /* Read mac addr. */ | |
2260 | err = xen_net_read_mac(dev, info->netdev->dev_addr); | |
2261 | if (err) { | |
2262 | xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename); | |
cb257783 | 2263 | goto out_unlocked; |
2688fcb7 AB |
2264 | } |
2265 | ||
6c5aa6fc DK |
2266 | info->netback_has_xdp_headroom = xenbus_read_unsigned(info->xbdev->otherend, |
2267 | "feature-xdp-headroom", 0); | |
2268 | if (info->netback_has_xdp_headroom) { | |
2269 | /* set the current xen-netfront xdp state */ | |
2270 | err = talk_to_netback_xdp(info, info->netfront_xdp_enabled ? | |
2271 | NETBACK_XDP_HEADROOM_ENABLE : | |
2272 | NETBACK_XDP_HEADROOM_DISABLE); | |
2273 | if (err) | |
2274 | goto out_unlocked; | |
2275 | } | |
2276 | ||
f599c64f | 2277 | rtnl_lock(); |
ce58725f DV |
2278 | if (info->queues) |
2279 | xennet_destroy_queues(info); | |
2280 | ||
a884daa6 JG |
2281 | /* For the case of a reconnect reset the "broken" indicator. */ |
2282 | info->broken = false; | |
2283 | ||
ca88ea12 | 2284 | err = xennet_create_queues(info, &num_queues); |
e2e004ac RL |
2285 | if (err < 0) { |
2286 | xenbus_dev_fatal(dev, err, "creating queues"); | |
2287 | kfree(info->queues); | |
2288 | info->queues = NULL; | |
2289 | goto out; | |
2290 | } | |
f599c64f | 2291 | rtnl_unlock(); |
2688fcb7 AB |
2292 | |
2293 | /* Create shared ring, alloc event channel -- for each queue */ | |
2294 | for (i = 0; i < num_queues; ++i) { | |
2295 | queue = &info->queues[i]; | |
2688fcb7 | 2296 | err = setup_netfront(dev, queue, feature_split_evtchn); |
e2e004ac RL |
2297 | if (err) |
2298 | goto destroy_ring; | |
2688fcb7 | 2299 | } |
0d160211 JF |
2300 | |
2301 | again: | |
2302 | err = xenbus_transaction_start(&xbt); | |
2303 | if (err) { | |
2304 | xenbus_dev_fatal(dev, err, "starting transaction"); | |
2305 | goto destroy_ring; | |
2306 | } | |
2307 | ||
812494d9 | 2308 | if (xenbus_exists(XBT_NIL, |
2309 | info->xbdev->otherend, "multi-queue-max-queues")) { | |
50ee6061 | 2310 | /* Write the number of queues */ |
812494d9 | 2311 | err = xenbus_printf(xbt, dev->nodename, |
2312 | "multi-queue-num-queues", "%u", num_queues); | |
d634bf2c | 2313 | if (err) { |
50ee6061 AB |
2314 | message = "writing multi-queue-num-queues"; |
2315 | goto abort_transaction_no_dev_fatal; | |
d634bf2c | 2316 | } |
812494d9 | 2317 | } |
50ee6061 | 2318 | |
812494d9 | 2319 | if (num_queues == 1) { |
2320 | err = write_queue_xenstore_keys(&info->queues[0], &xbt, 0); /* flat */ | |
2321 | if (err) | |
2322 | goto abort_transaction_no_dev_fatal; | |
2323 | } else { | |
50ee6061 AB |
2324 | /* Write the keys for each queue */ |
2325 | for (i = 0; i < num_queues; ++i) { | |
2326 | queue = &info->queues[i]; | |
2327 | err = write_queue_xenstore_keys(queue, &xbt, 1); /* hierarchical */ | |
2328 | if (err) | |
2329 | goto abort_transaction_no_dev_fatal; | |
d634bf2c | 2330 | } |
0d160211 JF |
2331 | } |
2332 | ||
50ee6061 | 2333 | /* The remaining keys are not queue-specific */ |
0d160211 JF |
2334 | err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u", |
2335 | 1); | |
2336 | if (err) { | |
2337 | message = "writing request-rx-copy"; | |
2338 | goto abort_transaction; | |
2339 | } | |
2340 | ||
2341 | err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1); | |
2342 | if (err) { | |
2343 | message = "writing feature-rx-notify"; | |
2344 | goto abort_transaction; | |
2345 | } | |
2346 | ||
2347 | err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1); | |
2348 | if (err) { | |
2349 | message = "writing feature-sg"; | |
2350 | goto abort_transaction; | |
2351 | } | |
2352 | ||
2353 | err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1); | |
2354 | if (err) { | |
2355 | message = "writing feature-gso-tcpv4"; | |
2356 | goto abort_transaction; | |
2357 | } | |
2358 | ||
2c0057de PD |
2359 | err = xenbus_write(xbt, dev->nodename, "feature-gso-tcpv6", "1"); |
2360 | if (err) { | |
2361 | message = "writing feature-gso-tcpv6"; | |
2362 | goto abort_transaction; | |
2363 | } | |
2364 | ||
2365 | err = xenbus_write(xbt, dev->nodename, "feature-ipv6-csum-offload", | |
2366 | "1"); | |
2367 | if (err) { | |
2368 | message = "writing feature-ipv6-csum-offload"; | |
2369 | goto abort_transaction; | |
2370 | } | |
2371 | ||
0d160211 JF |
2372 | err = xenbus_transaction_end(xbt, 0); |
2373 | if (err) { | |
2374 | if (err == -EAGAIN) | |
2375 | goto again; | |
2376 | xenbus_dev_fatal(dev, err, "completing transaction"); | |
2377 | goto destroy_ring; | |
2378 | } | |
2379 | ||
2380 | return 0; | |
2381 | ||
2382 | abort_transaction: | |
0d160211 | 2383 | xenbus_dev_fatal(dev, err, "%s", message); |
50ee6061 AB |
2384 | abort_transaction_no_dev_fatal: |
2385 | xenbus_transaction_end(xbt, 1); | |
0d160211 JF |
2386 | destroy_ring: |
2387 | xennet_disconnect_backend(info); | |
f599c64f | 2388 | rtnl_lock(); |
e2e004ac | 2389 | xennet_destroy_queues(info); |
0d160211 | 2390 | out: |
f599c64f | 2391 | rtnl_unlock(); |
cb257783 | 2392 | out_unlocked: |
d86b5672 | 2393 | device_unregister(&dev->dev); |
0d160211 JF |
2394 | return err; |
2395 | } | |
2396 | ||
0d160211 JF |
2397 | static int xennet_connect(struct net_device *dev) |
2398 | { | |
2399 | struct netfront_info *np = netdev_priv(dev); | |
2688fcb7 | 2400 | unsigned int num_queues = 0; |
a5b5dc3c | 2401 | int err; |
2688fcb7 AB |
2402 | unsigned int j = 0; |
2403 | struct netfront_queue *queue = NULL; | |
0d160211 | 2404 | |
2890ea5c | 2405 | if (!xenbus_read_unsigned(np->xbdev->otherend, "feature-rx-copy", 0)) { |
0d160211 | 2406 | dev_info(&dev->dev, |
898eb71c | 2407 | "backend does not support copying receive path\n"); |
0d160211 JF |
2408 | return -ENODEV; |
2409 | } | |
2410 | ||
f502bf2b | 2411 | err = talk_to_netback(np->xbdev, np); |
0d160211 JF |
2412 | if (err) |
2413 | return err; | |
6c5aa6fc DK |
2414 | if (np->netback_has_xdp_headroom) |
2415 | pr_info("backend supports XDP headroom\n"); | |
0d160211 | 2416 | |
2688fcb7 AB |
2417 | /* talk_to_netback() sets the correct number of queues */ |
2418 | num_queues = dev->real_num_tx_queues; | |
2419 | ||
f599c64f RL |
2420 | if (dev->reg_state == NETREG_UNINITIALIZED) { |
2421 | err = register_netdev(dev); | |
2422 | if (err) { | |
2423 | pr_warn("%s: register_netdev err=%d\n", __func__, err); | |
2424 | device_unregister(&np->xbdev->dev); | |
2425 | return err; | |
2426 | } | |
2427 | } | |
2428 | ||
45c8184c RL |
2429 | rtnl_lock(); |
2430 | netdev_update_features(dev); | |
2431 | rtnl_unlock(); | |
2432 | ||
0d160211 | 2433 | /* |
a5b5dc3c | 2434 | * All public and private state should now be sane. Get |
0d160211 JF |
2435 | * ready to start sending and receiving packets and give the driver |
2436 | * domain a kick because we've probably just requeued some | |
2437 | * packets. | |
2438 | */ | |
042b2046 DZ |
2439 | netif_tx_lock_bh(np->netdev); |
2440 | netif_device_attach(np->netdev); | |
2441 | netif_tx_unlock_bh(np->netdev); | |
2442 | ||
0d160211 | 2443 | netif_carrier_on(np->netdev); |
2688fcb7 AB |
2444 | for (j = 0; j < num_queues; ++j) { |
2445 | queue = &np->queues[j]; | |
f50b4076 | 2446 | |
2688fcb7 AB |
2447 | notify_remote_via_irq(queue->tx_irq); |
2448 | if (queue->tx_irq != queue->rx_irq) | |
2449 | notify_remote_via_irq(queue->rx_irq); | |
2688fcb7 | 2450 | |
f50b4076 DV |
2451 | spin_lock_irq(&queue->tx_lock); |
2452 | xennet_tx_buf_gc(queue); | |
2688fcb7 | 2453 | spin_unlock_irq(&queue->tx_lock); |
f50b4076 DV |
2454 | |
2455 | spin_lock_bh(&queue->rx_lock); | |
2456 | xennet_alloc_rx_buffers(queue); | |
2688fcb7 AB |
2457 | spin_unlock_bh(&queue->rx_lock); |
2458 | } | |
0d160211 JF |
2459 | |
2460 | return 0; | |
2461 | } | |
2462 | ||
80708602 | 2463 | /* |
0d160211 JF |
2464 | * Callback received when the backend's state changes. |
2465 | */ | |
f502bf2b | 2466 | static void netback_changed(struct xenbus_device *dev, |
0d160211 JF |
2467 | enum xenbus_state backend_state) |
2468 | { | |
1b713e00 | 2469 | struct netfront_info *np = dev_get_drvdata(&dev->dev); |
0d160211 JF |
2470 | struct net_device *netdev = np->netdev; |
2471 | ||
2472 | dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state)); | |
2473 | ||
8edfe2e9 JG |
2474 | wake_up_all(&module_wq); |
2475 | ||
0d160211 JF |
2476 | switch (backend_state) { |
2477 | case XenbusStateInitialising: | |
2478 | case XenbusStateInitialised: | |
b78c9512 NI |
2479 | case XenbusStateReconfiguring: |
2480 | case XenbusStateReconfigured: | |
0d160211 | 2481 | case XenbusStateUnknown: |
0d160211 JF |
2482 | break; |
2483 | ||
2484 | case XenbusStateInitWait: | |
2485 | if (dev->state != XenbusStateInitialising) | |
2486 | break; | |
2487 | if (xennet_connect(netdev) != 0) | |
2488 | break; | |
2489 | xenbus_switch_state(dev, XenbusStateConnected); | |
08e34eb1 LE |
2490 | break; |
2491 | ||
2492 | case XenbusStateConnected: | |
ee89bab1 | 2493 | netdev_notify_peers(netdev); |
0d160211 JF |
2494 | break; |
2495 | ||
bce3ea81 DV |
2496 | case XenbusStateClosed: |
2497 | if (dev->state == XenbusStateClosed) | |
2498 | break; | |
df561f66 | 2499 | fallthrough; /* Missed the backend's CLOSING state */ |
0d160211 JF |
2500 | case XenbusStateClosing: |
2501 | xenbus_frontend_closed(dev); | |
2502 | break; | |
2503 | } | |
2504 | } | |
2505 | ||
e0ce4af9 IC |
2506 | static const struct xennet_stat { |
2507 | char name[ETH_GSTRING_LEN]; | |
2508 | u16 offset; | |
2509 | } xennet_stats[] = { | |
2510 | { | |
2511 | "rx_gso_checksum_fixup", | |
2512 | offsetof(struct netfront_info, rx_gso_checksum_fixup) | |
2513 | }, | |
2514 | }; | |
2515 | ||
2516 | static int xennet_get_sset_count(struct net_device *dev, int string_set) | |
2517 | { | |
2518 | switch (string_set) { | |
2519 | case ETH_SS_STATS: | |
2520 | return ARRAY_SIZE(xennet_stats); | |
2521 | default: | |
2522 | return -EINVAL; | |
2523 | } | |
2524 | } | |
2525 | ||
2526 | static void xennet_get_ethtool_stats(struct net_device *dev, | |
2527 | struct ethtool_stats *stats, u64 * data) | |
2528 | { | |
2529 | void *np = netdev_priv(dev); | |
2530 | int i; | |
2531 | ||
2532 | for (i = 0; i < ARRAY_SIZE(xennet_stats); i++) | |
2688fcb7 | 2533 | data[i] = atomic_read((atomic_t *)(np + xennet_stats[i].offset)); |
e0ce4af9 IC |
2534 | } |
2535 | ||
2536 | static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data) | |
2537 | { | |
2538 | int i; | |
2539 | ||
2540 | switch (stringset) { | |
2541 | case ETH_SS_STATS: | |
2542 | for (i = 0; i < ARRAY_SIZE(xennet_stats); i++) | |
2543 | memcpy(data + i * ETH_GSTRING_LEN, | |
2544 | xennet_stats[i].name, ETH_GSTRING_LEN); | |
2545 | break; | |
2546 | } | |
2547 | } | |
2548 | ||
0fc0b732 | 2549 | static const struct ethtool_ops xennet_ethtool_ops = |
0d160211 | 2550 | { |
0d160211 | 2551 | .get_link = ethtool_op_get_link, |
e0ce4af9 IC |
2552 | |
2553 | .get_sset_count = xennet_get_sset_count, | |
2554 | .get_ethtool_stats = xennet_get_ethtool_stats, | |
2555 | .get_strings = xennet_get_strings, | |
91ffb9d3 | 2556 | .get_ts_info = ethtool_op_get_ts_info, |
0d160211 JF |
2557 | }; |
2558 | ||
2559 | #ifdef CONFIG_SYSFS | |
1f3c2eba DV |
2560 | static ssize_t show_rxbuf(struct device *dev, |
2561 | struct device_attribute *attr, char *buf) | |
0d160211 | 2562 | { |
1f3c2eba | 2563 | return sprintf(buf, "%lu\n", NET_RX_RING_SIZE); |
0d160211 JF |
2564 | } |
2565 | ||
1f3c2eba DV |
2566 | static ssize_t store_rxbuf(struct device *dev, |
2567 | struct device_attribute *attr, | |
2568 | const char *buf, size_t len) | |
0d160211 | 2569 | { |
0d160211 | 2570 | char *endp; |
0d160211 JF |
2571 | |
2572 | if (!capable(CAP_NET_ADMIN)) | |
2573 | return -EPERM; | |
2574 | ||
8ed7ec13 | 2575 | simple_strtoul(buf, &endp, 0); |
0d160211 JF |
2576 | if (endp == buf) |
2577 | return -EBADMSG; | |
2578 | ||
1f3c2eba | 2579 | /* rxbuf_min and rxbuf_max are no longer configurable. */ |
0d160211 | 2580 | |
0d160211 JF |
2581 | return len; |
2582 | } | |
2583 | ||
d61e4038 JP |
2584 | static DEVICE_ATTR(rxbuf_min, 0644, show_rxbuf, store_rxbuf); |
2585 | static DEVICE_ATTR(rxbuf_max, 0644, show_rxbuf, store_rxbuf); | |
2586 | static DEVICE_ATTR(rxbuf_cur, 0444, show_rxbuf, NULL); | |
0d160211 | 2587 | |
27b917e5 TI |
2588 | static struct attribute *xennet_dev_attrs[] = { |
2589 | &dev_attr_rxbuf_min.attr, | |
2590 | &dev_attr_rxbuf_max.attr, | |
2591 | &dev_attr_rxbuf_cur.attr, | |
2592 | NULL | |
2593 | }; | |
0d160211 | 2594 | |
27b917e5 TI |
2595 | static const struct attribute_group xennet_dev_group = { |
2596 | .attrs = xennet_dev_attrs | |
2597 | }; | |
0d160211 JF |
2598 | #endif /* CONFIG_SYSFS */ |
2599 | ||
c2c63310 | 2600 | static void xennet_bus_close(struct xenbus_device *dev) |
0d160211 | 2601 | { |
c2c63310 | 2602 | int ret; |
0d160211 | 2603 | |
c2c63310 AR |
2604 | if (xenbus_read_driver_state(dev->otherend) == XenbusStateClosed) |
2605 | return; | |
2606 | do { | |
5b5971df | 2607 | xenbus_switch_state(dev, XenbusStateClosing); |
c2c63310 AR |
2608 | ret = wait_event_timeout(module_wq, |
2609 | xenbus_read_driver_state(dev->otherend) == | |
2610 | XenbusStateClosing || | |
2611 | xenbus_read_driver_state(dev->otherend) == | |
2612 | XenbusStateClosed || | |
2613 | xenbus_read_driver_state(dev->otherend) == | |
2614 | XenbusStateUnknown, | |
2615 | XENNET_TIMEOUT); | |
2616 | } while (!ret); | |
2617 | ||
2618 | if (xenbus_read_driver_state(dev->otherend) == XenbusStateClosed) | |
2619 | return; | |
5b5971df | 2620 | |
c2c63310 | 2621 | do { |
5b5971df | 2622 | xenbus_switch_state(dev, XenbusStateClosed); |
c2c63310 AR |
2623 | ret = wait_event_timeout(module_wq, |
2624 | xenbus_read_driver_state(dev->otherend) == | |
2625 | XenbusStateClosed || | |
2626 | xenbus_read_driver_state(dev->otherend) == | |
2627 | XenbusStateUnknown, | |
2628 | XENNET_TIMEOUT); | |
2629 | } while (!ret); | |
2630 | } | |
2631 | ||
2632 | static int xennet_remove(struct xenbus_device *dev) | |
2633 | { | |
2634 | struct netfront_info *info = dev_get_drvdata(&dev->dev); | |
5b5971df | 2635 | |
c2c63310 | 2636 | xennet_bus_close(dev); |
0d160211 JF |
2637 | xennet_disconnect_backend(info); |
2638 | ||
f599c64f RL |
2639 | if (info->netdev->reg_state == NETREG_REGISTERED) |
2640 | unregister_netdev(info->netdev); | |
6bc96d04 | 2641 | |
f599c64f RL |
2642 | if (info->queues) { |
2643 | rtnl_lock(); | |
9a873c71 | 2644 | xennet_destroy_queues(info); |
f599c64f RL |
2645 | rtnl_unlock(); |
2646 | } | |
900e1833 | 2647 | xennet_free_netdev(info->netdev); |
0d160211 JF |
2648 | |
2649 | return 0; | |
2650 | } | |
2651 | ||
95afae48 DV |
2652 | static const struct xenbus_device_id netfront_ids[] = { |
2653 | { "vif" }, | |
2654 | { "" } | |
2655 | }; | |
2656 | ||
2657 | static struct xenbus_driver netfront_driver = { | |
2658 | .ids = netfront_ids, | |
0d160211 | 2659 | .probe = netfront_probe, |
8e0e46bb | 2660 | .remove = xennet_remove, |
0d160211 | 2661 | .resume = netfront_resume, |
f502bf2b | 2662 | .otherend_changed = netback_changed, |
95afae48 | 2663 | }; |
0d160211 JF |
2664 | |
2665 | static int __init netif_init(void) | |
2666 | { | |
6e833587 | 2667 | if (!xen_domain()) |
0d160211 JF |
2668 | return -ENODEV; |
2669 | ||
51c71a3b | 2670 | if (!xen_has_pv_nic_devices()) |
b9136d20 IM |
2671 | return -ENODEV; |
2672 | ||
383eda32 | 2673 | pr_info("Initialising Xen virtual ethernet driver\n"); |
0d160211 | 2674 | |
034702a6 | 2675 | /* Allow as many queues as there are CPUs inut max. 8 if user has not |
32a84405 WL |
2676 | * specified a value. |
2677 | */ | |
2678 | if (xennet_max_queues == 0) | |
034702a6 JG |
2679 | xennet_max_queues = min_t(unsigned int, MAX_QUEUES_DEFAULT, |
2680 | num_online_cpus()); | |
50ee6061 | 2681 | |
ffb78a26 | 2682 | return xenbus_register_frontend(&netfront_driver); |
0d160211 JF |
2683 | } |
2684 | module_init(netif_init); | |
2685 | ||
2686 | ||
2687 | static void __exit netif_exit(void) | |
2688 | { | |
ffb78a26 | 2689 | xenbus_unregister_driver(&netfront_driver); |
0d160211 JF |
2690 | } |
2691 | module_exit(netif_exit); | |
2692 | ||
2693 | MODULE_DESCRIPTION("Xen virtual network device frontend"); | |
2694 | MODULE_LICENSE("GPL"); | |
d2f0c52b | 2695 | MODULE_ALIAS("xen:vif"); |
4f93f09b | 2696 | MODULE_ALIAS("xennet"); |