1 /*
2 * This program is free software; you can redistribute it and/or
3 * modify it under the terms of the GNU General Public License version 2
4 * as published by the Free Software Foundation; or, when distributed
5 * separately from the Linux kernel or incorporated into other
6 * software packages, subject to the following license:
7 *
8 * Permission is hereby granted, free of charge, to any person obtaining a copy
9 * of this source file (the "Software"), to deal in the Software without
10 * restriction, including without limitation the rights to use, copy, modify,
11 * merge, publish, distribute, sublicense, and/or sell copies of the Software,
12 * and to permit persons to whom the Software is furnished to do so, subject to
13 * the following conditions:
14 *
15 * The above copyright notice and this permission notice shall be included in
16 * all copies or substantial portions of the Software.
17 *
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
21 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
22 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
23 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
24 * IN THE SOFTWARE.
25 */
26
27 #ifndef __XEN_NETBACK__COMMON_H__
28 #define __XEN_NETBACK__COMMON_H__
29
30 #define pr_fmt(fmt) KBUILD_MODNAME ":%s: " fmt, __func__
31
32 #include <linux/module.h>
33 #include <linux/interrupt.h>
34 #include <linux/slab.h>
35 #include <linux/ip.h>
36 #include <linux/in.h>
37 #include <linux/io.h>
38 #include <linux/netdevice.h>
39 #include <linux/etherdevice.h>
40 #include <linux/wait.h>
41 #include <linux/sched.h>
42
43 #include <xen/interface/io/netif.h>
44 #include <xen/interface/grant_table.h>
45 #include <xen/grant_table.h>
46 #include <xen/xenbus.h>
47 #include <xen/page.h>
48 #include <linux/debugfs.h>
49
50 typedef unsigned int pending_ring_idx_t;
51 #define INVALID_PENDING_RING_IDX (~0U)
52
53 struct pending_tx_info {
54 struct xen_netif_tx_request req; /* tx request */
55 unsigned int extra_count;
56 /* Callback data for released SKBs. The callback is always
57 * xenvif_zerocopy_callback, desc contains the pending_idx, which is
58 * also an index in pending_tx_info array. It is initialized in
59 * xenvif_alloc and it never changes.
60 * skb_shinfo(skb)->destructor_arg points to the first mapped slot's
61 * callback_struct in this array of struct pending_tx_info's, then ctx
62 * to the next, or NULL if there is no more slot for this skb.
63 * ubuf_to_vif is a helper which finds the struct xenvif from a pointer
64 * to this field.
65 */
66 struct ubuf_info callback_struct;
67 };
68
69 #define XEN_NETIF_TX_RING_SIZE __CONST_RING_SIZE(xen_netif_tx, XEN_PAGE_SIZE)
70 #define XEN_NETIF_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, XEN_PAGE_SIZE)
71
72 struct xenvif_rx_meta {
73 int id;
74 int size;
75 int gso_type;
76 int gso_size;
77 };
78
79 #define GSO_BIT(type) \
80 (1 << XEN_NETIF_GSO_TYPE_ ## type)
81
82 /* Discriminate from any valid pending_idx value. */
83 #define INVALID_PENDING_IDX 0xFFFF
84
85 #define MAX_BUFFER_OFFSET XEN_PAGE_SIZE
86
87 #define MAX_PENDING_REQS XEN_NETIF_TX_RING_SIZE
88
89 /* The maximum number of frags is derived from the size of a grant (same
90 * as a Xen page size for now).
91 */
92 #define MAX_XEN_SKB_FRAGS (65536 / XEN_PAGE_SIZE + 1)
93
94 #define NETBACK_INVALID_HANDLE -1
95
96 /* To avoid confusion, we define XEN_NETBK_LEGACY_SLOTS_MAX indicating
97 * the maximum slots a valid packet can use. Now this value is defined
98 * to be XEN_NETIF_NR_SLOTS_MIN, which is supposed to be supported by
99 * all backend.
100 */
101 #define XEN_NETBK_LEGACY_SLOTS_MAX XEN_NETIF_NR_SLOTS_MIN
102
103 /* Queue name is interface name with "-qNNN" appended */
104 #define QUEUE_NAME_SIZE (IFNAMSIZ + 5)
105
106 /* IRQ name is queue name with "-tx" or "-rx" appended */
107 #define IRQ_NAME_SIZE (QUEUE_NAME_SIZE + 3)
108
109 struct xenvif;
110
111 struct xenvif_stats {
112 /* Stats fields to be updated per-queue.
113 * A subset of struct net_device_stats that contains only the
114 * fields that are updated in netback.c for each queue.
115 */
116 u64 rx_bytes;
117 u64 rx_packets;
118 u64 tx_bytes;
119 u64 tx_packets;
120
121 /* Additional stats used by xenvif */
122 unsigned long rx_gso_checksum_fixup;
123 unsigned long tx_zerocopy_sent;
124 unsigned long tx_zerocopy_success;
125 unsigned long tx_zerocopy_fail;
126 unsigned long tx_frag_overflow;
127 };
128
129 #define COPY_BATCH_SIZE 64
130
131 struct xenvif_copy_state {
132 struct gnttab_copy op[COPY_BATCH_SIZE];
133 RING_IDX idx[COPY_BATCH_SIZE];
134 unsigned int num;
135 struct sk_buff_head *completed;
136 };
137
138 struct xenvif_queue { /* Per-queue data for xenvif */
139 unsigned int id; /* Queue ID, 0-based */
140 char name[QUEUE_NAME_SIZE]; /* DEVNAME-qN */
141 struct xenvif *vif; /* Parent VIF */
142
143 /* Use NAPI for guest TX */
144 struct napi_struct napi;
145 /* When feature-split-event-channels = 0, tx_irq = rx_irq. */
146 unsigned int tx_irq;
147 /* Only used when feature-split-event-channels = 1 */
148 char tx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-tx */
149 struct xen_netif_tx_back_ring tx;
150 struct sk_buff_head tx_queue;
151 struct page *mmap_pages[MAX_PENDING_REQS];
152 pending_ring_idx_t pending_prod;
153 pending_ring_idx_t pending_cons;
154 u16 pending_ring[MAX_PENDING_REQS];
155 struct pending_tx_info pending_tx_info[MAX_PENDING_REQS];
156 grant_handle_t grant_tx_handle[MAX_PENDING_REQS];
157
158 struct gnttab_copy tx_copy_ops[MAX_PENDING_REQS];
159 struct gnttab_map_grant_ref tx_map_ops[MAX_PENDING_REQS];
160 struct gnttab_unmap_grant_ref tx_unmap_ops[MAX_PENDING_REQS];
161 /* passed to gnttab_[un]map_refs with pages under (un)mapping */
162 struct page *pages_to_map[MAX_PENDING_REQS];
163 struct page *pages_to_unmap[MAX_PENDING_REQS];
164
165 /* This prevents zerocopy callbacks to race over dealloc_ring */
166 spinlock_t callback_lock;
167 /* This prevents dealloc thread and NAPI instance to race over response
168 * creation and pending_ring in xenvif_idx_release. In xenvif_tx_err
169 * it only protect response creation
170 */
171 spinlock_t response_lock;
172 pending_ring_idx_t dealloc_prod;
173 pending_ring_idx_t dealloc_cons;
174 u16 dealloc_ring[MAX_PENDING_REQS];
175 struct task_struct *dealloc_task;
176 wait_queue_head_t dealloc_wq;
177 atomic_t inflight_packets;
178
179 /* Use kthread for guest RX */
180 struct task_struct *task;
181 wait_queue_head_t wq;
182 /* When feature-split-event-channels = 0, tx_irq = rx_irq. */
183 unsigned int rx_irq;
184 /* Only used when feature-split-event-channels = 1 */
185 char rx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-rx */
186 struct xen_netif_rx_back_ring rx;
187 struct sk_buff_head rx_queue;
188
189 unsigned int rx_queue_max;
190 unsigned int rx_queue_len;
191 unsigned long last_rx_time;
192 bool stalled;
193
194 struct xenvif_copy_state rx_copy;
195
196 /* Transmit shaping: allow 'credit_bytes' every 'credit_usec'. */
197 unsigned long credit_bytes;
198 unsigned long credit_usec;
199 unsigned long remaining_credit;
200 struct timer_list credit_timeout;
201 u64 credit_window_start;
202 bool rate_limited;
203
204 /* Statistics */
205 struct xenvif_stats stats;
206 };
207
208 enum state_bit_shift {
209 /* This bit marks that the vif is connected */
210 VIF_STATUS_CONNECTED,
211 };
212
213 struct xenvif_mcast_addr {
214 struct list_head entry;
215 struct rcu_head rcu;
216 u8 addr[6];
217 };
218
219 #define XEN_NETBK_MCAST_MAX 64
220
221 #define XEN_NETBK_MAX_HASH_KEY_SIZE 40
222 #define XEN_NETBK_MAX_HASH_MAPPING_SIZE 128
223 #define XEN_NETBK_HASH_TAG_SIZE 40
224
225 struct xenvif_hash_cache_entry {
226 struct list_head link;
227 struct rcu_head rcu;
228 u8 tag[XEN_NETBK_HASH_TAG_SIZE];
229 unsigned int len;
230 u32 val;
231 int seq;
232 };
233
234 struct xenvif_hash_cache {
235 spinlock_t lock;
236 struct list_head list;
237 unsigned int count;
238 atomic_t seq;
239 };
240
241 struct xenvif_hash {
242 unsigned int alg;
243 u32 flags;
244 u8 key[XEN_NETBK_MAX_HASH_KEY_SIZE];
245 u32 mapping[XEN_NETBK_MAX_HASH_MAPPING_SIZE];
246 unsigned int size;
247 struct xenvif_hash_cache cache;
248 };
249
250 struct xenvif {
251 /* Unique identifier for this interface. */
252 domid_t domid;
253 unsigned int handle;
254
255 u8 fe_dev_addr[6];
256 struct list_head fe_mcast_addr;
257 unsigned int fe_mcast_count;
258
259 /* Frontend feature information. */
260 int gso_mask;
261
262 u8 can_sg:1;
263 u8 ip_csum:1;
264 u8 ipv6_csum:1;
265 u8 multicast_control:1;
266
267 /* Is this interface disabled? True when backend discovers
268 * frontend is rogue.
269 */
270 bool disabled;
271 unsigned long status;
272 unsigned long drain_timeout;
273 unsigned long stall_timeout;
274
275 /* Queues */
276 struct xenvif_queue *queues;
277 unsigned int num_queues; /* active queues, resource allocated */
278 unsigned int stalled_queues;
279
280 struct xenvif_hash hash;
281
282 struct xenbus_watch credit_watch;
283 struct xenbus_watch mcast_ctrl_watch;
284
285 spinlock_t lock;
286
287 #ifdef CONFIG_DEBUG_FS
288 struct dentry *xenvif_dbg_root;
289 #endif
290
291 struct xen_netif_ctrl_back_ring ctrl;
292 unsigned int ctrl_irq;
293
294 /* Miscellaneous private stuff. */
295 struct net_device *dev;
296 };
297
298 struct xenvif_rx_cb {
299 unsigned long expires;
300 int meta_slots_used;
301 };
302
303 #define XENVIF_RX_CB(skb) ((struct xenvif_rx_cb *)(skb)->cb)
304
xenvif_to_xenbus_device(struct xenvif * vif)305 static inline struct xenbus_device *xenvif_to_xenbus_device(struct xenvif *vif)
306 {
307 return to_xenbus_device(vif->dev->dev.parent);
308 }
309
310 void xenvif_tx_credit_callback(unsigned long data);
311
312 struct xenvif *xenvif_alloc(struct device *parent,
313 domid_t domid,
314 unsigned int handle);
315
316 int xenvif_init_queue(struct xenvif_queue *queue);
317 void xenvif_deinit_queue(struct xenvif_queue *queue);
318
319 int xenvif_connect_data(struct xenvif_queue *queue,
320 unsigned long tx_ring_ref,
321 unsigned long rx_ring_ref,
322 unsigned int tx_evtchn,
323 unsigned int rx_evtchn);
324 void xenvif_disconnect_data(struct xenvif *vif);
325 int xenvif_connect_ctrl(struct xenvif *vif, grant_ref_t ring_ref,
326 unsigned int evtchn);
327 void xenvif_disconnect_ctrl(struct xenvif *vif);
328 void xenvif_free(struct xenvif *vif);
329
330 int xenvif_xenbus_init(void);
331 void xenvif_xenbus_fini(void);
332
333 int xenvif_schedulable(struct xenvif *vif);
334
335 int xenvif_queue_stopped(struct xenvif_queue *queue);
336 void xenvif_wake_queue(struct xenvif_queue *queue);
337
338 /* (Un)Map communication rings. */
339 void xenvif_unmap_frontend_data_rings(struct xenvif_queue *queue);
340 int xenvif_map_frontend_data_rings(struct xenvif_queue *queue,
341 grant_ref_t tx_ring_ref,
342 grant_ref_t rx_ring_ref);
343
344 /* Check for SKBs from frontend and schedule backend processing */
345 void xenvif_napi_schedule_or_enable_events(struct xenvif_queue *queue);
346
347 /* Prevent the device from generating any further traffic. */
348 void xenvif_carrier_off(struct xenvif *vif);
349
350 int xenvif_tx_action(struct xenvif_queue *queue, int budget);
351
352 int xenvif_kthread_guest_rx(void *data);
353 void xenvif_kick_thread(struct xenvif_queue *queue);
354
355 int xenvif_dealloc_kthread(void *data);
356
357 irqreturn_t xenvif_ctrl_irq_fn(int irq, void *data);
358
359 void xenvif_rx_action(struct xenvif_queue *queue);
360 void xenvif_rx_queue_tail(struct xenvif_queue *queue, struct sk_buff *skb);
361
362 void xenvif_carrier_on(struct xenvif *vif);
363
364 /* Callback from stack when TX packet can be released */
365 void xenvif_zerocopy_callback(struct ubuf_info *ubuf, bool zerocopy_success);
366
367 /* Unmap a pending page and release it back to the guest */
368 void xenvif_idx_unmap(struct xenvif_queue *queue, u16 pending_idx);
369
nr_pending_reqs(struct xenvif_queue * queue)370 static inline pending_ring_idx_t nr_pending_reqs(struct xenvif_queue *queue)
371 {
372 return MAX_PENDING_REQS -
373 queue->pending_prod + queue->pending_cons;
374 }
375
376 irqreturn_t xenvif_interrupt(int irq, void *dev_id);
377
378 extern bool separate_tx_rx_irq;
379
380 extern unsigned int rx_drain_timeout_msecs;
381 extern unsigned int rx_stall_timeout_msecs;
382 extern unsigned int xenvif_max_queues;
383 extern unsigned int xenvif_hash_cache_size;
384
385 #ifdef CONFIG_DEBUG_FS
386 extern struct dentry *xen_netback_dbg_root;
387 #endif
388
389 void xenvif_skb_zerocopy_prepare(struct xenvif_queue *queue,
390 struct sk_buff *skb);
391 void xenvif_skb_zerocopy_complete(struct xenvif_queue *queue);
392
393 /* Multicast control */
394 bool xenvif_mcast_match(struct xenvif *vif, const u8 *addr);
395 void xenvif_mcast_addr_list_free(struct xenvif *vif);
396
397 /* Hash */
398 void xenvif_init_hash(struct xenvif *vif);
399 void xenvif_deinit_hash(struct xenvif *vif);
400
401 u32 xenvif_set_hash_alg(struct xenvif *vif, u32 alg);
402 u32 xenvif_get_hash_flags(struct xenvif *vif, u32 *flags);
403 u32 xenvif_set_hash_flags(struct xenvif *vif, u32 flags);
404 u32 xenvif_set_hash_key(struct xenvif *vif, u32 gref, u32 len);
405 u32 xenvif_set_hash_mapping_size(struct xenvif *vif, u32 size);
406 u32 xenvif_set_hash_mapping(struct xenvif *vif, u32 gref, u32 len,
407 u32 off);
408
409 void xenvif_set_skb_hash(struct xenvif *vif, struct sk_buff *skb);
410
411 #ifdef CONFIG_DEBUG_FS
412 void xenvif_dump_hash_info(struct xenvif *vif, struct seq_file *m);
413 #endif
414
415 #endif /* __XEN_NETBACK__COMMON_H__ */
416