xref: /openbmc/linux/drivers/net/xen-netback/common.h (revision 0edbfea5)
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 /* It's possible for an skb to have a maximal number of frags
95  * but still be less than MAX_BUFFER_OFFSET in size. Thus the
96  * worst-case number of copy operations is MAX_XEN_SKB_FRAGS per
97  * ring slot.
98  */
99 #define MAX_GRANT_COPY_OPS (MAX_XEN_SKB_FRAGS * XEN_NETIF_RX_RING_SIZE)
100 
101 #define NETBACK_INVALID_HANDLE -1
102 
103 /* To avoid confusion, we define XEN_NETBK_LEGACY_SLOTS_MAX indicating
104  * the maximum slots a valid packet can use. Now this value is defined
105  * to be XEN_NETIF_NR_SLOTS_MIN, which is supposed to be supported by
106  * all backend.
107  */
108 #define XEN_NETBK_LEGACY_SLOTS_MAX XEN_NETIF_NR_SLOTS_MIN
109 
110 /* Queue name is interface name with "-qNNN" appended */
111 #define QUEUE_NAME_SIZE (IFNAMSIZ + 5)
112 
113 /* IRQ name is queue name with "-tx" or "-rx" appended */
114 #define IRQ_NAME_SIZE (QUEUE_NAME_SIZE + 3)
115 
116 struct xenvif;
117 
118 struct xenvif_stats {
119 	/* Stats fields to be updated per-queue.
120 	 * A subset of struct net_device_stats that contains only the
121 	 * fields that are updated in netback.c for each queue.
122 	 */
123 	unsigned int rx_bytes;
124 	unsigned int rx_packets;
125 	unsigned int tx_bytes;
126 	unsigned int tx_packets;
127 
128 	/* Additional stats used by xenvif */
129 	unsigned long rx_gso_checksum_fixup;
130 	unsigned long tx_zerocopy_sent;
131 	unsigned long tx_zerocopy_success;
132 	unsigned long tx_zerocopy_fail;
133 	unsigned long tx_frag_overflow;
134 };
135 
136 struct xenvif_queue { /* Per-queue data for xenvif */
137 	unsigned int id; /* Queue ID, 0-based */
138 	char name[QUEUE_NAME_SIZE]; /* DEVNAME-qN */
139 	struct xenvif *vif; /* Parent VIF */
140 
141 	/* Use NAPI for guest TX */
142 	struct napi_struct napi;
143 	/* When feature-split-event-channels = 0, tx_irq = rx_irq. */
144 	unsigned int tx_irq;
145 	/* Only used when feature-split-event-channels = 1 */
146 	char tx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-tx */
147 	struct xen_netif_tx_back_ring tx;
148 	struct sk_buff_head tx_queue;
149 	struct page *mmap_pages[MAX_PENDING_REQS];
150 	pending_ring_idx_t pending_prod;
151 	pending_ring_idx_t pending_cons;
152 	u16 pending_ring[MAX_PENDING_REQS];
153 	struct pending_tx_info pending_tx_info[MAX_PENDING_REQS];
154 	grant_handle_t grant_tx_handle[MAX_PENDING_REQS];
155 
156 	struct gnttab_copy tx_copy_ops[MAX_PENDING_REQS];
157 	struct gnttab_map_grant_ref tx_map_ops[MAX_PENDING_REQS];
158 	struct gnttab_unmap_grant_ref tx_unmap_ops[MAX_PENDING_REQS];
159 	/* passed to gnttab_[un]map_refs with pages under (un)mapping */
160 	struct page *pages_to_map[MAX_PENDING_REQS];
161 	struct page *pages_to_unmap[MAX_PENDING_REQS];
162 
163 	/* This prevents zerocopy callbacks  to race over dealloc_ring */
164 	spinlock_t callback_lock;
165 	/* This prevents dealloc thread and NAPI instance to race over response
166 	 * creation and pending_ring in xenvif_idx_release. In xenvif_tx_err
167 	 * it only protect response creation
168 	 */
169 	spinlock_t response_lock;
170 	pending_ring_idx_t dealloc_prod;
171 	pending_ring_idx_t dealloc_cons;
172 	u16 dealloc_ring[MAX_PENDING_REQS];
173 	struct task_struct *dealloc_task;
174 	wait_queue_head_t dealloc_wq;
175 	atomic_t inflight_packets;
176 
177 	/* Use kthread for guest RX */
178 	struct task_struct *task;
179 	wait_queue_head_t wq;
180 	/* When feature-split-event-channels = 0, tx_irq = rx_irq. */
181 	unsigned int rx_irq;
182 	/* Only used when feature-split-event-channels = 1 */
183 	char rx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-rx */
184 	struct xen_netif_rx_back_ring rx;
185 	struct sk_buff_head rx_queue;
186 
187 	unsigned int rx_queue_max;
188 	unsigned int rx_queue_len;
189 	unsigned long last_rx_time;
190 	bool stalled;
191 
192 	struct gnttab_copy grant_copy_op[MAX_GRANT_COPY_OPS];
193 
194 	/* We create one meta structure per ring request we consume, so
195 	 * the maximum number is the same as the ring size.
196 	 */
197 	struct xenvif_rx_meta meta[XEN_NETIF_RX_RING_SIZE];
198 
199 	/* Transmit shaping: allow 'credit_bytes' every 'credit_usec'. */
200 	unsigned long   credit_bytes;
201 	unsigned long   credit_usec;
202 	unsigned long   remaining_credit;
203 	struct timer_list credit_timeout;
204 	u64 credit_window_start;
205 
206 	/* Statistics */
207 	struct xenvif_stats stats;
208 };
209 
210 enum state_bit_shift {
211 	/* This bit marks that the vif is connected */
212 	VIF_STATUS_CONNECTED,
213 };
214 
215 struct xenvif_mcast_addr {
216 	struct list_head entry;
217 	struct rcu_head rcu;
218 	u8 addr[6];
219 };
220 
221 #define XEN_NETBK_MCAST_MAX 64
222 
223 #define XEN_NETBK_MAX_HASH_KEY_SIZE 40
224 #define XEN_NETBK_MAX_HASH_MAPPING_SIZE 128
225 #define XEN_NETBK_HASH_TAG_SIZE 40
226 
227 struct xenvif_hash_cache_entry {
228 	struct list_head link;
229 	struct rcu_head rcu;
230 	u8 tag[XEN_NETBK_HASH_TAG_SIZE];
231 	unsigned int len;
232 	u32 val;
233 	int seq;
234 };
235 
236 struct xenvif_hash_cache {
237 	spinlock_t lock;
238 	struct list_head list;
239 	unsigned int count;
240 	atomic_t seq;
241 };
242 
243 struct xenvif_hash {
244 	unsigned int alg;
245 	u32 flags;
246 	u8 key[XEN_NETBK_MAX_HASH_KEY_SIZE];
247 	u32 mapping[XEN_NETBK_MAX_HASH_MAPPING_SIZE];
248 	unsigned int size;
249 	struct xenvif_hash_cache cache;
250 };
251 
252 struct xenvif {
253 	/* Unique identifier for this interface. */
254 	domid_t          domid;
255 	unsigned int     handle;
256 
257 	u8               fe_dev_addr[6];
258 	struct list_head fe_mcast_addr;
259 	unsigned int     fe_mcast_count;
260 
261 	/* Frontend feature information. */
262 	int gso_mask;
263 	int gso_prefix_mask;
264 
265 	u8 can_sg:1;
266 	u8 ip_csum:1;
267 	u8 ipv6_csum:1;
268 	u8 multicast_control:1;
269 
270 	/* Is this interface disabled? True when backend discovers
271 	 * frontend is rogue.
272 	 */
273 	bool disabled;
274 	unsigned long status;
275 	unsigned long drain_timeout;
276 	unsigned long stall_timeout;
277 
278 	/* Queues */
279 	struct xenvif_queue *queues;
280 	unsigned int num_queues; /* active queues, resource allocated */
281 	unsigned int stalled_queues;
282 
283 	struct xenvif_hash hash;
284 
285 	struct xenbus_watch credit_watch;
286 	struct xenbus_watch mcast_ctrl_watch;
287 
288 	spinlock_t lock;
289 
290 #ifdef CONFIG_DEBUG_FS
291 	struct dentry *xenvif_dbg_root;
292 #endif
293 
294 	struct xen_netif_ctrl_back_ring ctrl;
295 	struct task_struct *ctrl_task;
296 	wait_queue_head_t ctrl_wq;
297 	unsigned int ctrl_irq;
298 
299 	/* Miscellaneous private stuff. */
300 	struct net_device *dev;
301 };
302 
303 struct xenvif_rx_cb {
304 	unsigned long expires;
305 	int meta_slots_used;
306 };
307 
308 #define XENVIF_RX_CB(skb) ((struct xenvif_rx_cb *)(skb)->cb)
309 
310 static inline struct xenbus_device *xenvif_to_xenbus_device(struct xenvif *vif)
311 {
312 	return to_xenbus_device(vif->dev->dev.parent);
313 }
314 
315 void xenvif_tx_credit_callback(unsigned long data);
316 
317 struct xenvif *xenvif_alloc(struct device *parent,
318 			    domid_t domid,
319 			    unsigned int handle);
320 
321 int xenvif_init_queue(struct xenvif_queue *queue);
322 void xenvif_deinit_queue(struct xenvif_queue *queue);
323 
324 int xenvif_connect_data(struct xenvif_queue *queue,
325 			unsigned long tx_ring_ref,
326 			unsigned long rx_ring_ref,
327 			unsigned int tx_evtchn,
328 			unsigned int rx_evtchn);
329 void xenvif_disconnect_data(struct xenvif *vif);
330 int xenvif_connect_ctrl(struct xenvif *vif, grant_ref_t ring_ref,
331 			unsigned int evtchn);
332 void xenvif_disconnect_ctrl(struct xenvif *vif);
333 void xenvif_free(struct xenvif *vif);
334 
335 int xenvif_xenbus_init(void);
336 void xenvif_xenbus_fini(void);
337 
338 int xenvif_schedulable(struct xenvif *vif);
339 
340 int xenvif_queue_stopped(struct xenvif_queue *queue);
341 void xenvif_wake_queue(struct xenvif_queue *queue);
342 
343 /* (Un)Map communication rings. */
344 void xenvif_unmap_frontend_data_rings(struct xenvif_queue *queue);
345 int xenvif_map_frontend_data_rings(struct xenvif_queue *queue,
346 				   grant_ref_t tx_ring_ref,
347 				   grant_ref_t rx_ring_ref);
348 
349 /* Check for SKBs from frontend and schedule backend processing */
350 void xenvif_napi_schedule_or_enable_events(struct xenvif_queue *queue);
351 
352 /* Prevent the device from generating any further traffic. */
353 void xenvif_carrier_off(struct xenvif *vif);
354 
355 int xenvif_tx_action(struct xenvif_queue *queue, int budget);
356 
357 int xenvif_kthread_guest_rx(void *data);
358 void xenvif_kick_thread(struct xenvif_queue *queue);
359 
360 int xenvif_dealloc_kthread(void *data);
361 
362 int xenvif_ctrl_kthread(void *data);
363 
364 void xenvif_rx_queue_tail(struct xenvif_queue *queue, struct sk_buff *skb);
365 
366 void xenvif_carrier_on(struct xenvif *vif);
367 
368 /* Callback from stack when TX packet can be released */
369 void xenvif_zerocopy_callback(struct ubuf_info *ubuf, bool zerocopy_success);
370 
371 /* Unmap a pending page and release it back to the guest */
372 void xenvif_idx_unmap(struct xenvif_queue *queue, u16 pending_idx);
373 
374 static inline pending_ring_idx_t nr_pending_reqs(struct xenvif_queue *queue)
375 {
376 	return MAX_PENDING_REQS -
377 		queue->pending_prod + queue->pending_cons;
378 }
379 
380 irqreturn_t xenvif_interrupt(int irq, void *dev_id);
381 
382 extern bool separate_tx_rx_irq;
383 
384 extern unsigned int rx_drain_timeout_msecs;
385 extern unsigned int rx_stall_timeout_msecs;
386 extern unsigned int xenvif_max_queues;
387 extern unsigned int xenvif_hash_cache_size;
388 
389 #ifdef CONFIG_DEBUG_FS
390 extern struct dentry *xen_netback_dbg_root;
391 #endif
392 
393 void xenvif_skb_zerocopy_prepare(struct xenvif_queue *queue,
394 				 struct sk_buff *skb);
395 void xenvif_skb_zerocopy_complete(struct xenvif_queue *queue);
396 
397 /* Multicast control */
398 bool xenvif_mcast_match(struct xenvif *vif, const u8 *addr);
399 void xenvif_mcast_addr_list_free(struct xenvif *vif);
400 
401 /* Hash */
402 void xenvif_init_hash(struct xenvif *vif);
403 void xenvif_deinit_hash(struct xenvif *vif);
404 
405 u32 xenvif_set_hash_alg(struct xenvif *vif, u32 alg);
406 u32 xenvif_get_hash_flags(struct xenvif *vif, u32 *flags);
407 u32 xenvif_set_hash_flags(struct xenvif *vif, u32 flags);
408 u32 xenvif_set_hash_key(struct xenvif *vif, u32 gref, u32 len);
409 u32 xenvif_set_hash_mapping_size(struct xenvif *vif, u32 size);
410 u32 xenvif_set_hash_mapping(struct xenvif *vif, u32 gref, u32 len,
411 			    u32 off);
412 
413 void xenvif_set_skb_hash(struct xenvif *vif, struct sk_buff *skb);
414 
415 #endif /* __XEN_NETBACK__COMMON_H__ */
416