xref: /openbmc/linux/drivers/net/xen-netback/common.h (revision 4a3a9904)
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 
48 typedef unsigned int pending_ring_idx_t;
49 #define INVALID_PENDING_RING_IDX (~0U)
50 
51 struct pending_tx_info {
52 	struct xen_netif_tx_request req; /* tx request */
53 	/* Callback data for released SKBs. The callback is always
54 	 * xenvif_zerocopy_callback, desc contains the pending_idx, which is
55 	 * also an index in pending_tx_info array. It is initialized in
56 	 * xenvif_alloc and it never changes.
57 	 * skb_shinfo(skb)->destructor_arg points to the first mapped slot's
58 	 * callback_struct in this array of struct pending_tx_info's, then ctx
59 	 * to the next, or NULL if there is no more slot for this skb.
60 	 * ubuf_to_vif is a helper which finds the struct xenvif from a pointer
61 	 * to this field.
62 	 */
63 	struct ubuf_info callback_struct;
64 };
65 
66 #define XEN_NETIF_TX_RING_SIZE __CONST_RING_SIZE(xen_netif_tx, PAGE_SIZE)
67 #define XEN_NETIF_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, PAGE_SIZE)
68 
69 struct xenvif_rx_meta {
70 	int id;
71 	int size;
72 	int gso_type;
73 	int gso_size;
74 };
75 
76 #define GSO_BIT(type) \
77 	(1 << XEN_NETIF_GSO_TYPE_ ## type)
78 
79 /* Discriminate from any valid pending_idx value. */
80 #define INVALID_PENDING_IDX 0xFFFF
81 
82 #define MAX_BUFFER_OFFSET PAGE_SIZE
83 
84 #define MAX_PENDING_REQS XEN_NETIF_TX_RING_SIZE
85 
86 /* It's possible for an skb to have a maximal number of frags
87  * but still be less than MAX_BUFFER_OFFSET in size. Thus the
88  * worst-case number of copy operations is MAX_SKB_FRAGS per
89  * ring slot.
90  */
91 #define MAX_GRANT_COPY_OPS (MAX_SKB_FRAGS * XEN_NETIF_RX_RING_SIZE)
92 
93 #define NETBACK_INVALID_HANDLE -1
94 
95 /* To avoid confusion, we define XEN_NETBK_LEGACY_SLOTS_MAX indicating
96  * the maximum slots a valid packet can use. Now this value is defined
97  * to be XEN_NETIF_NR_SLOTS_MIN, which is supposed to be supported by
98  * all backend.
99  */
100 #define XEN_NETBK_LEGACY_SLOTS_MAX XEN_NETIF_NR_SLOTS_MIN
101 
102 /* Queue name is interface name with "-qNNN" appended */
103 #define QUEUE_NAME_SIZE (IFNAMSIZ + 5)
104 
105 /* IRQ name is queue name with "-tx" or "-rx" appended */
106 #define IRQ_NAME_SIZE (QUEUE_NAME_SIZE + 3)
107 
108 struct xenvif;
109 
110 struct xenvif_stats {
111 	/* Stats fields to be updated per-queue.
112 	 * A subset of struct net_device_stats that contains only the
113 	 * fields that are updated in netback.c for each queue.
114 	 */
115 	unsigned int rx_bytes;
116 	unsigned int rx_packets;
117 	unsigned int tx_bytes;
118 	unsigned int tx_packets;
119 
120 	/* Additional stats used by xenvif */
121 	unsigned long rx_gso_checksum_fixup;
122 	unsigned long tx_zerocopy_sent;
123 	unsigned long tx_zerocopy_success;
124 	unsigned long tx_zerocopy_fail;
125 	unsigned long tx_frag_overflow;
126 };
127 
128 struct xenvif_queue { /* Per-queue data for xenvif */
129 	unsigned int id; /* Queue ID, 0-based */
130 	char name[QUEUE_NAME_SIZE]; /* DEVNAME-qN */
131 	struct xenvif *vif; /* Parent VIF */
132 
133 	/* Use NAPI for guest TX */
134 	struct napi_struct napi;
135 	/* When feature-split-event-channels = 0, tx_irq = rx_irq. */
136 	unsigned int tx_irq;
137 	/* Only used when feature-split-event-channels = 1 */
138 	char tx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-tx */
139 	struct xen_netif_tx_back_ring tx;
140 	struct sk_buff_head tx_queue;
141 	struct page *mmap_pages[MAX_PENDING_REQS];
142 	pending_ring_idx_t pending_prod;
143 	pending_ring_idx_t pending_cons;
144 	u16 pending_ring[MAX_PENDING_REQS];
145 	struct pending_tx_info pending_tx_info[MAX_PENDING_REQS];
146 	grant_handle_t grant_tx_handle[MAX_PENDING_REQS];
147 
148 	struct gnttab_copy tx_copy_ops[MAX_PENDING_REQS];
149 	struct gnttab_map_grant_ref tx_map_ops[MAX_PENDING_REQS];
150 	struct gnttab_unmap_grant_ref tx_unmap_ops[MAX_PENDING_REQS];
151 	/* passed to gnttab_[un]map_refs with pages under (un)mapping */
152 	struct page *pages_to_map[MAX_PENDING_REQS];
153 	struct page *pages_to_unmap[MAX_PENDING_REQS];
154 
155 	/* This prevents zerocopy callbacks  to race over dealloc_ring */
156 	spinlock_t callback_lock;
157 	/* This prevents dealloc thread and NAPI instance to race over response
158 	 * creation and pending_ring in xenvif_idx_release. In xenvif_tx_err
159 	 * it only protect response creation
160 	 */
161 	spinlock_t response_lock;
162 	pending_ring_idx_t dealloc_prod;
163 	pending_ring_idx_t dealloc_cons;
164 	u16 dealloc_ring[MAX_PENDING_REQS];
165 	struct task_struct *dealloc_task;
166 	wait_queue_head_t dealloc_wq;
167 
168 	/* Use kthread for guest RX */
169 	struct task_struct *task;
170 	wait_queue_head_t wq;
171 	/* When feature-split-event-channels = 0, tx_irq = rx_irq. */
172 	unsigned int rx_irq;
173 	/* Only used when feature-split-event-channels = 1 */
174 	char rx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-rx */
175 	struct xen_netif_rx_back_ring rx;
176 	struct sk_buff_head rx_queue;
177 	RING_IDX rx_last_skb_slots;
178 	bool rx_queue_purge;
179 
180 	struct timer_list wake_queue;
181 
182 	struct gnttab_copy grant_copy_op[MAX_GRANT_COPY_OPS];
183 
184 	/* We create one meta structure per ring request we consume, so
185 	 * the maximum number is the same as the ring size.
186 	 */
187 	struct xenvif_rx_meta meta[XEN_NETIF_RX_RING_SIZE];
188 
189 	/* Transmit shaping: allow 'credit_bytes' every 'credit_usec'. */
190 	unsigned long   credit_bytes;
191 	unsigned long   credit_usec;
192 	unsigned long   remaining_credit;
193 	struct timer_list credit_timeout;
194 	u64 credit_window_start;
195 
196 	/* Statistics */
197 	struct xenvif_stats stats;
198 };
199 
200 struct xenvif {
201 	/* Unique identifier for this interface. */
202 	domid_t          domid;
203 	unsigned int     handle;
204 
205 	u8               fe_dev_addr[6];
206 
207 	/* Frontend feature information. */
208 	int gso_mask;
209 	int gso_prefix_mask;
210 
211 	u8 can_sg:1;
212 	u8 ip_csum:1;
213 	u8 ipv6_csum:1;
214 
215 	/* Internal feature information. */
216 	u8 can_queue:1;	    /* can queue packets for receiver? */
217 
218 	/* Is this interface disabled? True when backend discovers
219 	 * frontend is rogue.
220 	 */
221 	bool disabled;
222 
223 	/* Queues */
224 	struct xenvif_queue *queues;
225 	unsigned int num_queues; /* active queues, resource allocated */
226 
227 	/* Miscellaneous private stuff. */
228 	struct net_device *dev;
229 };
230 
231 static inline struct xenbus_device *xenvif_to_xenbus_device(struct xenvif *vif)
232 {
233 	return to_xenbus_device(vif->dev->dev.parent);
234 }
235 
236 struct xenvif *xenvif_alloc(struct device *parent,
237 			    domid_t domid,
238 			    unsigned int handle);
239 
240 int xenvif_init_queue(struct xenvif_queue *queue);
241 void xenvif_deinit_queue(struct xenvif_queue *queue);
242 
243 int xenvif_connect(struct xenvif_queue *queue, unsigned long tx_ring_ref,
244 		   unsigned long rx_ring_ref, unsigned int tx_evtchn,
245 		   unsigned int rx_evtchn);
246 void xenvif_disconnect(struct xenvif *vif);
247 void xenvif_free(struct xenvif *vif);
248 
249 int xenvif_xenbus_init(void);
250 void xenvif_xenbus_fini(void);
251 
252 int xenvif_schedulable(struct xenvif *vif);
253 
254 int xenvif_must_stop_queue(struct xenvif_queue *queue);
255 
256 int xenvif_queue_stopped(struct xenvif_queue *queue);
257 void xenvif_wake_queue(struct xenvif_queue *queue);
258 
259 /* (Un)Map communication rings. */
260 void xenvif_unmap_frontend_rings(struct xenvif_queue *queue);
261 int xenvif_map_frontend_rings(struct xenvif_queue *queue,
262 			      grant_ref_t tx_ring_ref,
263 			      grant_ref_t rx_ring_ref);
264 
265 /* Check for SKBs from frontend and schedule backend processing */
266 void xenvif_napi_schedule_or_enable_events(struct xenvif_queue *queue);
267 
268 /* Prevent the device from generating any further traffic. */
269 void xenvif_carrier_off(struct xenvif *vif);
270 
271 int xenvif_tx_action(struct xenvif_queue *queue, int budget);
272 
273 int xenvif_kthread_guest_rx(void *data);
274 void xenvif_kick_thread(struct xenvif_queue *queue);
275 
276 int xenvif_dealloc_kthread(void *data);
277 
278 /* Determine whether the needed number of slots (req) are available,
279  * and set req_event if not.
280  */
281 bool xenvif_rx_ring_slots_available(struct xenvif_queue *queue, int needed);
282 
283 void xenvif_carrier_on(struct xenvif *vif);
284 
285 /* Callback from stack when TX packet can be released */
286 void xenvif_zerocopy_callback(struct ubuf_info *ubuf, bool zerocopy_success);
287 
288 /* Unmap a pending page and release it back to the guest */
289 void xenvif_idx_unmap(struct xenvif_queue *queue, u16 pending_idx);
290 
291 static inline pending_ring_idx_t nr_pending_reqs(struct xenvif_queue *queue)
292 {
293 	return MAX_PENDING_REQS -
294 		queue->pending_prod + queue->pending_cons;
295 }
296 
297 /* Callback from stack when TX packet can be released */
298 void xenvif_zerocopy_callback(struct ubuf_info *ubuf, bool zerocopy_success);
299 
300 extern bool separate_tx_rx_irq;
301 
302 extern unsigned int rx_drain_timeout_msecs;
303 extern unsigned int rx_drain_timeout_jiffies;
304 extern unsigned int xenvif_max_queues;
305 
306 #endif /* __XEN_NETBACK__COMMON_H__ */
307