xref: /openbmc/linux/drivers/net/xen-netback/common.h (revision 36fe4655)
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 	/*
144 	 * TX/RX common EOI handling.
145 	 * When feature-split-event-channels = 0, interrupt handler sets
146 	 * NETBK_COMMON_EOI, otherwise NETBK_RX_EOI and NETBK_TX_EOI are set
147 	 * by the RX and TX interrupt handlers.
148 	 * RX and TX handler threads will issue an EOI when either
149 	 * NETBK_COMMON_EOI or their specific bits (NETBK_RX_EOI or
150 	 * NETBK_TX_EOI) are set and they will reset those bits.
151 	 */
152 	atomic_t eoi_pending;
153 #define NETBK_RX_EOI		0x01
154 #define NETBK_TX_EOI		0x02
155 #define NETBK_COMMON_EOI	0x04
156 
157 	/* Use NAPI for guest TX */
158 	struct napi_struct napi;
159 	/* When feature-split-event-channels = 0, tx_irq = rx_irq. */
160 	unsigned int tx_irq;
161 	/* Only used when feature-split-event-channels = 1 */
162 	char tx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-tx */
163 	struct xen_netif_tx_back_ring tx;
164 	struct sk_buff_head tx_queue;
165 	struct page *mmap_pages[MAX_PENDING_REQS];
166 	pending_ring_idx_t pending_prod;
167 	pending_ring_idx_t pending_cons;
168 	u16 pending_ring[MAX_PENDING_REQS];
169 	struct pending_tx_info pending_tx_info[MAX_PENDING_REQS];
170 	grant_handle_t grant_tx_handle[MAX_PENDING_REQS];
171 
172 	struct gnttab_copy tx_copy_ops[MAX_PENDING_REQS];
173 	struct gnttab_map_grant_ref tx_map_ops[MAX_PENDING_REQS];
174 	struct gnttab_unmap_grant_ref tx_unmap_ops[MAX_PENDING_REQS];
175 	/* passed to gnttab_[un]map_refs with pages under (un)mapping */
176 	struct page *pages_to_map[MAX_PENDING_REQS];
177 	struct page *pages_to_unmap[MAX_PENDING_REQS];
178 
179 	/* This prevents zerocopy callbacks  to race over dealloc_ring */
180 	spinlock_t callback_lock;
181 	/* This prevents dealloc thread and NAPI instance to race over response
182 	 * creation and pending_ring in xenvif_idx_release. In xenvif_tx_err
183 	 * it only protect response creation
184 	 */
185 	spinlock_t response_lock;
186 	pending_ring_idx_t dealloc_prod;
187 	pending_ring_idx_t dealloc_cons;
188 	u16 dealloc_ring[MAX_PENDING_REQS];
189 	struct task_struct *dealloc_task;
190 	wait_queue_head_t dealloc_wq;
191 	atomic_t inflight_packets;
192 
193 	/* Use kthread for guest RX */
194 	struct task_struct *task;
195 	wait_queue_head_t wq;
196 	/* When feature-split-event-channels = 0, tx_irq = rx_irq. */
197 	unsigned int rx_irq;
198 	/* Only used when feature-split-event-channels = 1 */
199 	char rx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-rx */
200 	struct xen_netif_rx_back_ring rx;
201 	struct sk_buff_head rx_queue;
202 
203 	unsigned int rx_queue_max;
204 	unsigned int rx_queue_len;
205 	unsigned long last_rx_time;
206 	bool stalled;
207 
208 	struct xenvif_copy_state rx_copy;
209 
210 	/* Transmit shaping: allow 'credit_bytes' every 'credit_usec'. */
211 	unsigned long   credit_bytes;
212 	unsigned long   credit_usec;
213 	unsigned long   remaining_credit;
214 	struct timer_list credit_timeout;
215 	u64 credit_window_start;
216 	bool rate_limited;
217 
218 	/* Statistics */
219 	struct xenvif_stats stats;
220 };
221 
222 enum state_bit_shift {
223 	/* This bit marks that the vif is connected */
224 	VIF_STATUS_CONNECTED,
225 };
226 
227 struct xenvif_mcast_addr {
228 	struct list_head entry;
229 	struct rcu_head rcu;
230 	u8 addr[6];
231 };
232 
233 #define XEN_NETBK_MCAST_MAX 64
234 
235 #define XEN_NETBK_MAX_HASH_KEY_SIZE 40
236 #define XEN_NETBK_MAX_HASH_MAPPING_SIZE 128
237 #define XEN_NETBK_HASH_TAG_SIZE 40
238 
239 struct xenvif_hash_cache_entry {
240 	struct list_head link;
241 	struct rcu_head rcu;
242 	u8 tag[XEN_NETBK_HASH_TAG_SIZE];
243 	unsigned int len;
244 	u32 val;
245 	int seq;
246 };
247 
248 struct xenvif_hash_cache {
249 	spinlock_t lock;
250 	struct list_head list;
251 	unsigned int count;
252 	atomic_t seq;
253 };
254 
255 struct xenvif_hash {
256 	unsigned int alg;
257 	u32 flags;
258 	bool mapping_sel;
259 	u8 key[XEN_NETBK_MAX_HASH_KEY_SIZE];
260 	u32 mapping[2][XEN_NETBK_MAX_HASH_MAPPING_SIZE];
261 	unsigned int size;
262 	struct xenvif_hash_cache cache;
263 };
264 
265 struct backend_info {
266 	struct xenbus_device *dev;
267 	struct xenvif *vif;
268 
269 	/* This is the state that will be reflected in xenstore when any
270 	 * active hotplug script completes.
271 	 */
272 	enum xenbus_state state;
273 
274 	enum xenbus_state frontend_state;
275 	struct xenbus_watch hotplug_status_watch;
276 	u8 have_hotplug_status_watch:1;
277 
278 	const char *hotplug_script;
279 };
280 
281 struct xenvif {
282 	/* Unique identifier for this interface. */
283 	domid_t          domid;
284 	unsigned int     handle;
285 
286 	u8               fe_dev_addr[6];
287 	struct list_head fe_mcast_addr;
288 	unsigned int     fe_mcast_count;
289 
290 	/* Frontend feature information. */
291 	int gso_mask;
292 
293 	u8 can_sg:1;
294 	u8 ip_csum:1;
295 	u8 ipv6_csum:1;
296 	u8 multicast_control:1;
297 
298 	/* headroom requested by xen-netfront */
299 	u16 xdp_headroom;
300 
301 	/* Is this interface disabled? True when backend discovers
302 	 * frontend is rogue.
303 	 */
304 	bool disabled;
305 	unsigned long status;
306 	unsigned long drain_timeout;
307 	unsigned long stall_timeout;
308 
309 	/* Queues */
310 	struct xenvif_queue *queues;
311 	unsigned int num_queues; /* active queues, resource allocated */
312 	unsigned int stalled_queues;
313 
314 	struct xenvif_hash hash;
315 
316 	struct xenbus_watch credit_watch;
317 	struct xenbus_watch mcast_ctrl_watch;
318 
319 	struct backend_info *be;
320 
321 	spinlock_t lock;
322 
323 #ifdef CONFIG_DEBUG_FS
324 	struct dentry *xenvif_dbg_root;
325 #endif
326 
327 	struct xen_netif_ctrl_back_ring ctrl;
328 	unsigned int ctrl_irq;
329 
330 	/* Miscellaneous private stuff. */
331 	struct net_device *dev;
332 };
333 
334 struct xenvif_rx_cb {
335 	unsigned long expires;
336 	int meta_slots_used;
337 };
338 
339 #define XENVIF_RX_CB(skb) ((struct xenvif_rx_cb *)(skb)->cb)
340 
341 static inline struct xenbus_device *xenvif_to_xenbus_device(struct xenvif *vif)
342 {
343 	return to_xenbus_device(vif->dev->dev.parent);
344 }
345 
346 void xenvif_tx_credit_callback(struct timer_list *t);
347 
348 struct xenvif *xenvif_alloc(struct device *parent,
349 			    domid_t domid,
350 			    unsigned int handle);
351 
352 int xenvif_init_queue(struct xenvif_queue *queue);
353 void xenvif_deinit_queue(struct xenvif_queue *queue);
354 
355 int xenvif_connect_data(struct xenvif_queue *queue,
356 			unsigned long tx_ring_ref,
357 			unsigned long rx_ring_ref,
358 			unsigned int tx_evtchn,
359 			unsigned int rx_evtchn);
360 void xenvif_disconnect_data(struct xenvif *vif);
361 int xenvif_connect_ctrl(struct xenvif *vif, grant_ref_t ring_ref,
362 			unsigned int evtchn);
363 void xenvif_disconnect_ctrl(struct xenvif *vif);
364 void xenvif_free(struct xenvif *vif);
365 
366 int xenvif_xenbus_init(void);
367 void xenvif_xenbus_fini(void);
368 
369 int xenvif_schedulable(struct xenvif *vif);
370 
371 int xenvif_queue_stopped(struct xenvif_queue *queue);
372 void xenvif_wake_queue(struct xenvif_queue *queue);
373 
374 /* (Un)Map communication rings. */
375 void xenvif_unmap_frontend_data_rings(struct xenvif_queue *queue);
376 int xenvif_map_frontend_data_rings(struct xenvif_queue *queue,
377 				   grant_ref_t tx_ring_ref,
378 				   grant_ref_t rx_ring_ref);
379 
380 /* Check for SKBs from frontend and schedule backend processing */
381 void xenvif_napi_schedule_or_enable_events(struct xenvif_queue *queue);
382 
383 /* Prevent the device from generating any further traffic. */
384 void xenvif_carrier_off(struct xenvif *vif);
385 
386 int xenvif_tx_action(struct xenvif_queue *queue, int budget);
387 
388 int xenvif_kthread_guest_rx(void *data);
389 void xenvif_kick_thread(struct xenvif_queue *queue);
390 
391 int xenvif_dealloc_kthread(void *data);
392 
393 irqreturn_t xenvif_ctrl_irq_fn(int irq, void *data);
394 
395 bool xenvif_have_rx_work(struct xenvif_queue *queue, bool test_kthread);
396 void xenvif_rx_action(struct xenvif_queue *queue);
397 void xenvif_rx_queue_tail(struct xenvif_queue *queue, struct sk_buff *skb);
398 
399 void xenvif_carrier_on(struct xenvif *vif);
400 
401 /* Callback from stack when TX packet can be released */
402 void xenvif_zerocopy_callback(struct ubuf_info *ubuf, bool zerocopy_success);
403 
404 /* Unmap a pending page and release it back to the guest */
405 void xenvif_idx_unmap(struct xenvif_queue *queue, u16 pending_idx);
406 
407 static inline pending_ring_idx_t nr_pending_reqs(struct xenvif_queue *queue)
408 {
409 	return MAX_PENDING_REQS -
410 		queue->pending_prod + queue->pending_cons;
411 }
412 
413 irqreturn_t xenvif_interrupt(int irq, void *dev_id);
414 
415 extern bool separate_tx_rx_irq;
416 extern bool provides_xdp_headroom;
417 
418 extern unsigned int rx_drain_timeout_msecs;
419 extern unsigned int rx_stall_timeout_msecs;
420 extern unsigned int xenvif_max_queues;
421 extern unsigned int xenvif_hash_cache_size;
422 
423 #ifdef CONFIG_DEBUG_FS
424 extern struct dentry *xen_netback_dbg_root;
425 #endif
426 
427 void xenvif_skb_zerocopy_prepare(struct xenvif_queue *queue,
428 				 struct sk_buff *skb);
429 void xenvif_skb_zerocopy_complete(struct xenvif_queue *queue);
430 
431 /* Multicast control */
432 bool xenvif_mcast_match(struct xenvif *vif, const u8 *addr);
433 void xenvif_mcast_addr_list_free(struct xenvif *vif);
434 
435 /* Hash */
436 void xenvif_init_hash(struct xenvif *vif);
437 void xenvif_deinit_hash(struct xenvif *vif);
438 
439 u32 xenvif_set_hash_alg(struct xenvif *vif, u32 alg);
440 u32 xenvif_get_hash_flags(struct xenvif *vif, u32 *flags);
441 u32 xenvif_set_hash_flags(struct xenvif *vif, u32 flags);
442 u32 xenvif_set_hash_key(struct xenvif *vif, u32 gref, u32 len);
443 u32 xenvif_set_hash_mapping_size(struct xenvif *vif, u32 size);
444 u32 xenvif_set_hash_mapping(struct xenvif *vif, u32 gref, u32 len,
445 			    u32 off);
446 
447 void xenvif_set_skb_hash(struct xenvif *vif, struct sk_buff *skb);
448 
449 #ifdef CONFIG_DEBUG_FS
450 void xenvif_dump_hash_info(struct xenvif *vif, struct seq_file *m);
451 #endif
452 
453 #endif /* __XEN_NETBACK__COMMON_H__ */
454