• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 	/* Callback data for released SKBs. The callback is always
56 	 * xenvif_zerocopy_callback, desc contains the pending_idx, which is
57 	 * also an index in pending_tx_info array. It is initialized in
58 	 * xenvif_alloc and it never changes.
59 	 * skb_shinfo(skb)->destructor_arg points to the first mapped slot's
60 	 * callback_struct in this array of struct pending_tx_info's, then ctx
61 	 * to the next, or NULL if there is no more slot for this skb.
62 	 * ubuf_to_vif is a helper which finds the struct xenvif from a pointer
63 	 * to this field.
64 	 */
65 	struct ubuf_info callback_struct;
66 };
67 
68 #define XEN_NETIF_TX_RING_SIZE __CONST_RING_SIZE(xen_netif_tx, XEN_PAGE_SIZE)
69 #define XEN_NETIF_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, XEN_PAGE_SIZE)
70 
71 struct xenvif_rx_meta {
72 	int id;
73 	int size;
74 	int gso_type;
75 	int gso_size;
76 };
77 
78 #define GSO_BIT(type) \
79 	(1 << XEN_NETIF_GSO_TYPE_ ## type)
80 
81 /* Discriminate from any valid pending_idx value. */
82 #define INVALID_PENDING_IDX 0xFFFF
83 
84 #define MAX_BUFFER_OFFSET XEN_PAGE_SIZE
85 
86 #define MAX_PENDING_REQS XEN_NETIF_TX_RING_SIZE
87 
88 /* The maximum number of frags is derived from the size of a grant (same
89  * as a Xen page size for now).
90  */
91 #define MAX_XEN_SKB_FRAGS (65536 / XEN_PAGE_SIZE + 1)
92 
93 /* It's possible for an skb to have a maximal number of frags
94  * but still be less than MAX_BUFFER_OFFSET in size. Thus the
95  * worst-case number of copy operations is MAX_XEN_SKB_FRAGS per
96  * ring slot.
97  */
98 #define MAX_GRANT_COPY_OPS (MAX_XEN_SKB_FRAGS * XEN_NETIF_RX_RING_SIZE)
99 
100 #define NETBACK_INVALID_HANDLE -1
101 
102 /* To avoid confusion, we define XEN_NETBK_LEGACY_SLOTS_MAX indicating
103  * the maximum slots a valid packet can use. Now this value is defined
104  * to be XEN_NETIF_NR_SLOTS_MIN, which is supposed to be supported by
105  * all backend.
106  */
107 #define XEN_NETBK_LEGACY_SLOTS_MAX XEN_NETIF_NR_SLOTS_MIN
108 
109 /* Queue name is interface name with "-qNNN" appended */
110 #define QUEUE_NAME_SIZE (IFNAMSIZ + 5)
111 
112 /* IRQ name is queue name with "-tx" or "-rx" appended */
113 #define IRQ_NAME_SIZE (QUEUE_NAME_SIZE + 3)
114 
115 struct xenvif;
116 
117 struct xenvif_stats {
118 	/* Stats fields to be updated per-queue.
119 	 * A subset of struct net_device_stats that contains only the
120 	 * fields that are updated in netback.c for each queue.
121 	 */
122 	unsigned int rx_bytes;
123 	unsigned int rx_packets;
124 	unsigned int tx_bytes;
125 	unsigned int tx_packets;
126 
127 	/* Additional stats used by xenvif */
128 	unsigned long rx_gso_checksum_fixup;
129 	unsigned long tx_zerocopy_sent;
130 	unsigned long tx_zerocopy_success;
131 	unsigned long tx_zerocopy_fail;
132 	unsigned long tx_frag_overflow;
133 };
134 
135 struct xenvif_queue { /* Per-queue data for xenvif */
136 	unsigned int id; /* Queue ID, 0-based */
137 	char name[QUEUE_NAME_SIZE]; /* DEVNAME-qN */
138 	struct xenvif *vif; /* Parent VIF */
139 
140 	/*
141 	 * TX/RX common EOI handling.
142 	 * When feature-split-event-channels = 0, interrupt handler sets
143 	 * NETBK_COMMON_EOI, otherwise NETBK_RX_EOI and NETBK_TX_EOI are set
144 	 * by the RX and TX interrupt handlers.
145 	 * RX and TX handler threads will issue an EOI when either
146 	 * NETBK_COMMON_EOI or their specific bits (NETBK_RX_EOI or
147 	 * NETBK_TX_EOI) are set and they will reset those bits.
148 	 */
149 	atomic_t eoi_pending;
150 #define NETBK_RX_EOI		0x01
151 #define NETBK_TX_EOI		0x02
152 #define NETBK_COMMON_EOI	0x04
153 
154 	/* Use NAPI for guest TX */
155 	struct napi_struct napi;
156 	/* When feature-split-event-channels = 0, tx_irq = rx_irq. */
157 	unsigned int tx_irq;
158 	/* Only used when feature-split-event-channels = 1 */
159 	char tx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-tx */
160 	struct xen_netif_tx_back_ring tx;
161 	struct sk_buff_head tx_queue;
162 	struct page *mmap_pages[MAX_PENDING_REQS];
163 	pending_ring_idx_t pending_prod;
164 	pending_ring_idx_t pending_cons;
165 	u16 pending_ring[MAX_PENDING_REQS];
166 	struct pending_tx_info pending_tx_info[MAX_PENDING_REQS];
167 	grant_handle_t grant_tx_handle[MAX_PENDING_REQS];
168 
169 	struct gnttab_copy tx_copy_ops[MAX_PENDING_REQS];
170 	struct gnttab_map_grant_ref tx_map_ops[MAX_PENDING_REQS];
171 	struct gnttab_unmap_grant_ref tx_unmap_ops[MAX_PENDING_REQS];
172 	/* passed to gnttab_[un]map_refs with pages under (un)mapping */
173 	struct page *pages_to_map[MAX_PENDING_REQS];
174 	struct page *pages_to_unmap[MAX_PENDING_REQS];
175 
176 	/* This prevents zerocopy callbacks  to race over dealloc_ring */
177 	spinlock_t callback_lock;
178 	/* This prevents dealloc thread and NAPI instance to race over response
179 	 * creation and pending_ring in xenvif_idx_release. In xenvif_tx_err
180 	 * it only protect response creation
181 	 */
182 	spinlock_t response_lock;
183 	pending_ring_idx_t dealloc_prod;
184 	pending_ring_idx_t dealloc_cons;
185 	u16 dealloc_ring[MAX_PENDING_REQS];
186 	struct task_struct *dealloc_task;
187 	wait_queue_head_t dealloc_wq;
188 	atomic_t inflight_packets;
189 
190 	/* Use kthread for guest RX */
191 	struct task_struct *task;
192 	wait_queue_head_t wq;
193 	/* When feature-split-event-channels = 0, tx_irq = rx_irq. */
194 	unsigned int rx_irq;
195 	/* Only used when feature-split-event-channels = 1 */
196 	char rx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-rx */
197 	struct xen_netif_rx_back_ring rx;
198 	struct sk_buff_head rx_queue;
199 
200 	unsigned int rx_queue_max;
201 	unsigned int rx_queue_len;
202 	unsigned long last_rx_time;
203 	bool stalled;
204 
205 	struct gnttab_copy grant_copy_op[MAX_GRANT_COPY_OPS];
206 
207 	/* We create one meta structure per ring request we consume, so
208 	 * the maximum number is the same as the ring size.
209 	 */
210 	struct xenvif_rx_meta meta[XEN_NETIF_RX_RING_SIZE];
211 
212 	/* Transmit shaping: allow 'credit_bytes' every 'credit_usec'. */
213 	unsigned long   credit_bytes;
214 	unsigned long   credit_usec;
215 	unsigned long   remaining_credit;
216 	struct timer_list credit_timeout;
217 	u64 credit_window_start;
218 	bool rate_limited;
219 
220 	/* Statistics */
221 	struct xenvif_stats stats;
222 };
223 
224 enum state_bit_shift {
225 	/* This bit marks that the vif is connected */
226 	VIF_STATUS_CONNECTED,
227 };
228 
229 struct xenvif_mcast_addr {
230 	struct list_head entry;
231 	struct rcu_head rcu;
232 	u8 addr[6];
233 };
234 
235 #define XEN_NETBK_MCAST_MAX 64
236 
237 struct xenvif {
238 	/* Unique identifier for this interface. */
239 	domid_t          domid;
240 	unsigned int     handle;
241 
242 	u8               fe_dev_addr[6];
243 	struct list_head fe_mcast_addr;
244 	unsigned int     fe_mcast_count;
245 
246 	/* Frontend feature information. */
247 	int gso_mask;
248 	int gso_prefix_mask;
249 
250 	u8 can_sg:1;
251 	u8 ip_csum:1;
252 	u8 ipv6_csum:1;
253 	u8 multicast_control:1;
254 
255 	/* Is this interface disabled? True when backend discovers
256 	 * frontend is rogue.
257 	 */
258 	bool disabled;
259 	unsigned long status;
260 	unsigned long drain_timeout;
261 	unsigned long stall_timeout;
262 
263 	/* Queues */
264 	struct xenvif_queue *queues;
265 	unsigned int num_queues; /* active queues, resource allocated */
266 	unsigned int stalled_queues;
267 
268 	struct xenbus_watch credit_watch;
269 
270 	spinlock_t lock;
271 
272 #ifdef CONFIG_DEBUG_FS
273 	struct dentry *xenvif_dbg_root;
274 #endif
275 
276 	/* Miscellaneous private stuff. */
277 	struct net_device *dev;
278 };
279 
280 struct xenvif_rx_cb {
281 	unsigned long expires;
282 	int meta_slots_used;
283 };
284 
285 #define XENVIF_RX_CB(skb) ((struct xenvif_rx_cb *)(skb)->cb)
286 
xenvif_to_xenbus_device(struct xenvif * vif)287 static inline struct xenbus_device *xenvif_to_xenbus_device(struct xenvif *vif)
288 {
289 	return to_xenbus_device(vif->dev->dev.parent);
290 }
291 
292 void xenvif_tx_credit_callback(unsigned long data);
293 
294 struct xenvif *xenvif_alloc(struct device *parent,
295 			    domid_t domid,
296 			    unsigned int handle);
297 
298 int xenvif_init_queue(struct xenvif_queue *queue);
299 void xenvif_deinit_queue(struct xenvif_queue *queue);
300 
301 int xenvif_connect(struct xenvif_queue *queue, unsigned long tx_ring_ref,
302 		   unsigned long rx_ring_ref, unsigned int tx_evtchn,
303 		   unsigned int rx_evtchn);
304 void xenvif_disconnect(struct xenvif *vif);
305 void xenvif_free(struct xenvif *vif);
306 
307 int xenvif_xenbus_init(void);
308 void xenvif_xenbus_fini(void);
309 
310 int xenvif_schedulable(struct xenvif *vif);
311 
312 int xenvif_queue_stopped(struct xenvif_queue *queue);
313 void xenvif_wake_queue(struct xenvif_queue *queue);
314 
315 /* (Un)Map communication rings. */
316 void xenvif_unmap_frontend_rings(struct xenvif_queue *queue);
317 int xenvif_map_frontend_rings(struct xenvif_queue *queue,
318 			      grant_ref_t tx_ring_ref,
319 			      grant_ref_t rx_ring_ref);
320 
321 /* Check for SKBs from frontend and schedule backend processing */
322 void xenvif_napi_schedule_or_enable_events(struct xenvif_queue *queue);
323 
324 /* Prevent the device from generating any further traffic. */
325 void xenvif_carrier_off(struct xenvif *vif);
326 
327 int xenvif_tx_action(struct xenvif_queue *queue, int budget);
328 
329 int xenvif_kthread_guest_rx(void *data);
330 void xenvif_kick_thread(struct xenvif_queue *queue);
331 
332 int xenvif_dealloc_kthread(void *data);
333 
334 bool xenvif_have_rx_work(struct xenvif_queue *queue, bool test_kthread);
335 void xenvif_rx_queue_tail(struct xenvif_queue *queue, struct sk_buff *skb);
336 
337 void xenvif_carrier_on(struct xenvif *vif);
338 
339 /* Callback from stack when TX packet can be released */
340 void xenvif_zerocopy_callback(struct ubuf_info *ubuf, bool zerocopy_success);
341 
342 /* Unmap a pending page and release it back to the guest */
343 void xenvif_idx_unmap(struct xenvif_queue *queue, u16 pending_idx);
344 
nr_pending_reqs(struct xenvif_queue * queue)345 static inline pending_ring_idx_t nr_pending_reqs(struct xenvif_queue *queue)
346 {
347 	return MAX_PENDING_REQS -
348 		queue->pending_prod + queue->pending_cons;
349 }
350 
351 irqreturn_t xenvif_interrupt(int irq, void *dev_id);
352 
353 extern bool separate_tx_rx_irq;
354 
355 extern unsigned int rx_drain_timeout_msecs;
356 extern unsigned int rx_stall_timeout_msecs;
357 extern unsigned int xenvif_max_queues;
358 
359 #ifdef CONFIG_DEBUG_FS
360 extern struct dentry *xen_netback_dbg_root;
361 #endif
362 
363 void xenvif_skb_zerocopy_prepare(struct xenvif_queue *queue,
364 				 struct sk_buff *skb);
365 void xenvif_skb_zerocopy_complete(struct xenvif_queue *queue);
366 
367 /* Multicast control */
368 bool xenvif_mcast_match(struct xenvif *vif, const u8 *addr);
369 void xenvif_mcast_addr_list_free(struct xenvif *vif);
370 
371 #include <linux/atomic.h>
372 
xenvif_atomic_fetch_or(int i,atomic_t * v)373 static inline int xenvif_atomic_fetch_or(int i, atomic_t *v)
374 {
375 	int c, old;
376 
377 	c = v->counter;
378 	while ((old = cmpxchg(&v->counter, c, c | i)) != c)
379 		c = old;
380 
381 	return c;
382 }
383 
xenvif_atomic_fetch_andnot(int i,atomic_t * v)384 static inline int xenvif_atomic_fetch_andnot(int i, atomic_t *v)
385 {
386 	int c, old;
387 
388 	c = v->counter;
389 	while ((old = cmpxchg(&v->counter, c, c & ~i)) != c)
390 		c = old;
391 
392 	return c;
393 }
394 
395 #endif /* __XEN_NETBACK__COMMON_H__ */
396