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; 202 203 /* Statistics */ 204 struct xenvif_stats stats; 205 }; 206 207 enum state_bit_shift { 208 /* This bit marks that the vif is connected */ 209 VIF_STATUS_CONNECTED, 210 }; 211 212 struct xenvif_mcast_addr { 213 struct list_head entry; 214 struct rcu_head rcu; 215 u8 addr[6]; 216 }; 217 218 #define XEN_NETBK_MCAST_MAX 64 219 220 #define XEN_NETBK_MAX_HASH_KEY_SIZE 40 221 #define XEN_NETBK_MAX_HASH_MAPPING_SIZE 128 222 #define XEN_NETBK_HASH_TAG_SIZE 40 223 224 struct xenvif_hash_cache_entry { 225 struct list_head link; 226 struct rcu_head rcu; 227 u8 tag[XEN_NETBK_HASH_TAG_SIZE]; 228 unsigned int len; 229 u32 val; 230 int seq; 231 }; 232 233 struct xenvif_hash_cache { 234 spinlock_t lock; 235 struct list_head list; 236 unsigned int count; 237 atomic_t seq; 238 }; 239 240 struct xenvif_hash { 241 unsigned int alg; 242 u32 flags; 243 u8 key[XEN_NETBK_MAX_HASH_KEY_SIZE]; 244 u32 mapping[XEN_NETBK_MAX_HASH_MAPPING_SIZE]; 245 unsigned int size; 246 struct xenvif_hash_cache cache; 247 }; 248 249 struct xenvif { 250 /* Unique identifier for this interface. */ 251 domid_t domid; 252 unsigned int handle; 253 254 u8 fe_dev_addr[6]; 255 struct list_head fe_mcast_addr; 256 unsigned int fe_mcast_count; 257 258 /* Frontend feature information. */ 259 int gso_mask; 260 261 u8 can_sg:1; 262 u8 ip_csum:1; 263 u8 ipv6_csum:1; 264 u8 multicast_control:1; 265 266 /* Is this interface disabled? True when backend discovers 267 * frontend is rogue. 268 */ 269 bool disabled; 270 unsigned long status; 271 unsigned long drain_timeout; 272 unsigned long stall_timeout; 273 274 /* Queues */ 275 struct xenvif_queue *queues; 276 unsigned int num_queues; /* active queues, resource allocated */ 277 unsigned int stalled_queues; 278 279 struct xenvif_hash hash; 280 281 struct xenbus_watch credit_watch; 282 struct xenbus_watch mcast_ctrl_watch; 283 284 spinlock_t lock; 285 286 #ifdef CONFIG_DEBUG_FS 287 struct dentry *xenvif_dbg_root; 288 #endif 289 290 struct xen_netif_ctrl_back_ring ctrl; 291 unsigned int ctrl_irq; 292 293 /* Miscellaneous private stuff. */ 294 struct net_device *dev; 295 }; 296 297 struct xenvif_rx_cb { 298 unsigned long expires; 299 int meta_slots_used; 300 }; 301 302 #define XENVIF_RX_CB(skb) ((struct xenvif_rx_cb *)(skb)->cb) 303 304 static inline struct xenbus_device *xenvif_to_xenbus_device(struct xenvif *vif) 305 { 306 return to_xenbus_device(vif->dev->dev.parent); 307 } 308 309 void xenvif_tx_credit_callback(unsigned long data); 310 311 struct xenvif *xenvif_alloc(struct device *parent, 312 domid_t domid, 313 unsigned int handle); 314 315 int xenvif_init_queue(struct xenvif_queue *queue); 316 void xenvif_deinit_queue(struct xenvif_queue *queue); 317 318 int xenvif_connect_data(struct xenvif_queue *queue, 319 unsigned long tx_ring_ref, 320 unsigned long rx_ring_ref, 321 unsigned int tx_evtchn, 322 unsigned int rx_evtchn); 323 void xenvif_disconnect_data(struct xenvif *vif); 324 int xenvif_connect_ctrl(struct xenvif *vif, grant_ref_t ring_ref, 325 unsigned int evtchn); 326 void xenvif_disconnect_ctrl(struct xenvif *vif); 327 void xenvif_free(struct xenvif *vif); 328 329 int xenvif_xenbus_init(void); 330 void xenvif_xenbus_fini(void); 331 332 int xenvif_schedulable(struct xenvif *vif); 333 334 int xenvif_queue_stopped(struct xenvif_queue *queue); 335 void xenvif_wake_queue(struct xenvif_queue *queue); 336 337 /* (Un)Map communication rings. */ 338 void xenvif_unmap_frontend_data_rings(struct xenvif_queue *queue); 339 int xenvif_map_frontend_data_rings(struct xenvif_queue *queue, 340 grant_ref_t tx_ring_ref, 341 grant_ref_t rx_ring_ref); 342 343 /* Check for SKBs from frontend and schedule backend processing */ 344 void xenvif_napi_schedule_or_enable_events(struct xenvif_queue *queue); 345 346 /* Prevent the device from generating any further traffic. */ 347 void xenvif_carrier_off(struct xenvif *vif); 348 349 int xenvif_tx_action(struct xenvif_queue *queue, int budget); 350 351 int xenvif_kthread_guest_rx(void *data); 352 void xenvif_kick_thread(struct xenvif_queue *queue); 353 354 int xenvif_dealloc_kthread(void *data); 355 356 irqreturn_t xenvif_ctrl_irq_fn(int irq, void *data); 357 358 void xenvif_rx_action(struct xenvif_queue *queue); 359 void xenvif_rx_queue_tail(struct xenvif_queue *queue, struct sk_buff *skb); 360 361 void xenvif_carrier_on(struct xenvif *vif); 362 363 /* Callback from stack when TX packet can be released */ 364 void xenvif_zerocopy_callback(struct ubuf_info *ubuf, bool zerocopy_success); 365 366 /* Unmap a pending page and release it back to the guest */ 367 void xenvif_idx_unmap(struct xenvif_queue *queue, u16 pending_idx); 368 369 static inline pending_ring_idx_t nr_pending_reqs(struct xenvif_queue *queue) 370 { 371 return MAX_PENDING_REQS - 372 queue->pending_prod + queue->pending_cons; 373 } 374 375 irqreturn_t xenvif_interrupt(int irq, void *dev_id); 376 377 extern bool separate_tx_rx_irq; 378 379 extern unsigned int rx_drain_timeout_msecs; 380 extern unsigned int rx_stall_timeout_msecs; 381 extern unsigned int xenvif_max_queues; 382 extern unsigned int xenvif_hash_cache_size; 383 384 #ifdef CONFIG_DEBUG_FS 385 extern struct dentry *xen_netback_dbg_root; 386 #endif 387 388 void xenvif_skb_zerocopy_prepare(struct xenvif_queue *queue, 389 struct sk_buff *skb); 390 void xenvif_skb_zerocopy_complete(struct xenvif_queue *queue); 391 392 /* Multicast control */ 393 bool xenvif_mcast_match(struct xenvif *vif, const u8 *addr); 394 void xenvif_mcast_addr_list_free(struct xenvif *vif); 395 396 /* Hash */ 397 void xenvif_init_hash(struct xenvif *vif); 398 void xenvif_deinit_hash(struct xenvif *vif); 399 400 u32 xenvif_set_hash_alg(struct xenvif *vif, u32 alg); 401 u32 xenvif_get_hash_flags(struct xenvif *vif, u32 *flags); 402 u32 xenvif_set_hash_flags(struct xenvif *vif, u32 flags); 403 u32 xenvif_set_hash_key(struct xenvif *vif, u32 gref, u32 len); 404 u32 xenvif_set_hash_mapping_size(struct xenvif *vif, u32 size); 405 u32 xenvif_set_hash_mapping(struct xenvif *vif, u32 gref, u32 len, 406 u32 off); 407 408 void xenvif_set_skb_hash(struct xenvif *vif, struct sk_buff *skb); 409 410 #ifdef CONFIG_DEBUG_FS 411 void xenvif_dump_hash_info(struct xenvif *vif, struct seq_file *m); 412 #endif 413 414 #endif /* __XEN_NETBACK__COMMON_H__ */ 415