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 <linux/debugfs.h> 48 49 typedef unsigned int pending_ring_idx_t; 50 #define INVALID_PENDING_RING_IDX (~0U) 51 52 struct pending_tx_info { 53 struct xen_netif_tx_request req; /* tx request */ 54 /* Callback data for released SKBs. The callback is always 55 * xenvif_zerocopy_callback, desc contains the pending_idx, which is 56 * also an index in pending_tx_info array. It is initialized in 57 * xenvif_alloc and it never changes. 58 * skb_shinfo(skb)->destructor_arg points to the first mapped slot's 59 * callback_struct in this array of struct pending_tx_info's, then ctx 60 * to the next, or NULL if there is no more slot for this skb. 61 * ubuf_to_vif is a helper which finds the struct xenvif from a pointer 62 * to this field. 63 */ 64 struct ubuf_info callback_struct; 65 }; 66 67 #define XEN_NETIF_TX_RING_SIZE __CONST_RING_SIZE(xen_netif_tx, PAGE_SIZE) 68 #define XEN_NETIF_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, PAGE_SIZE) 69 70 struct xenvif_rx_meta { 71 int id; 72 int size; 73 int gso_type; 74 int gso_size; 75 }; 76 77 #define GSO_BIT(type) \ 78 (1 << XEN_NETIF_GSO_TYPE_ ## type) 79 80 /* Discriminate from any valid pending_idx value. */ 81 #define INVALID_PENDING_IDX 0xFFFF 82 83 #define MAX_BUFFER_OFFSET PAGE_SIZE 84 85 #define MAX_PENDING_REQS XEN_NETIF_TX_RING_SIZE 86 87 /* It's possible for an skb to have a maximal number of frags 88 * but still be less than MAX_BUFFER_OFFSET in size. Thus the 89 * worst-case number of copy operations is MAX_SKB_FRAGS per 90 * ring slot. 91 */ 92 #define MAX_GRANT_COPY_OPS (MAX_SKB_FRAGS * XEN_NETIF_RX_RING_SIZE) 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 unsigned int rx_bytes; 117 unsigned int rx_packets; 118 unsigned int tx_bytes; 119 unsigned int 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 struct xenvif_queue { /* Per-queue data for xenvif */ 130 unsigned int id; /* Queue ID, 0-based */ 131 char name[QUEUE_NAME_SIZE]; /* DEVNAME-qN */ 132 struct xenvif *vif; /* Parent VIF */ 133 134 /* Use NAPI for guest TX */ 135 struct napi_struct napi; 136 /* When feature-split-event-channels = 0, tx_irq = rx_irq. */ 137 unsigned int tx_irq; 138 /* Only used when feature-split-event-channels = 1 */ 139 char tx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-tx */ 140 struct xen_netif_tx_back_ring tx; 141 struct sk_buff_head tx_queue; 142 struct page *mmap_pages[MAX_PENDING_REQS]; 143 pending_ring_idx_t pending_prod; 144 pending_ring_idx_t pending_cons; 145 u16 pending_ring[MAX_PENDING_REQS]; 146 struct pending_tx_info pending_tx_info[MAX_PENDING_REQS]; 147 grant_handle_t grant_tx_handle[MAX_PENDING_REQS]; 148 149 struct gnttab_copy tx_copy_ops[MAX_PENDING_REQS]; 150 struct gnttab_map_grant_ref tx_map_ops[MAX_PENDING_REQS]; 151 struct gnttab_unmap_grant_ref tx_unmap_ops[MAX_PENDING_REQS]; 152 /* passed to gnttab_[un]map_refs with pages under (un)mapping */ 153 struct page *pages_to_map[MAX_PENDING_REQS]; 154 struct page *pages_to_unmap[MAX_PENDING_REQS]; 155 156 /* This prevents zerocopy callbacks to race over dealloc_ring */ 157 spinlock_t callback_lock; 158 /* This prevents dealloc thread and NAPI instance to race over response 159 * creation and pending_ring in xenvif_idx_release. In xenvif_tx_err 160 * it only protect response creation 161 */ 162 spinlock_t response_lock; 163 pending_ring_idx_t dealloc_prod; 164 pending_ring_idx_t dealloc_cons; 165 u16 dealloc_ring[MAX_PENDING_REQS]; 166 struct task_struct *dealloc_task; 167 wait_queue_head_t dealloc_wq; 168 atomic_t inflight_packets; 169 170 /* Use kthread for guest RX */ 171 struct task_struct *task; 172 wait_queue_head_t wq; 173 /* When feature-split-event-channels = 0, tx_irq = rx_irq. */ 174 unsigned int rx_irq; 175 /* Only used when feature-split-event-channels = 1 */ 176 char rx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-rx */ 177 struct xen_netif_rx_back_ring rx; 178 struct sk_buff_head rx_queue; 179 180 unsigned int rx_queue_max; 181 unsigned int rx_queue_len; 182 unsigned long last_rx_time; 183 bool stalled; 184 185 struct gnttab_copy grant_copy_op[MAX_GRANT_COPY_OPS]; 186 187 /* We create one meta structure per ring request we consume, so 188 * the maximum number is the same as the ring size. 189 */ 190 struct xenvif_rx_meta meta[XEN_NETIF_RX_RING_SIZE]; 191 192 /* Transmit shaping: allow 'credit_bytes' every 'credit_usec'. */ 193 unsigned long credit_bytes; 194 unsigned long credit_usec; 195 unsigned long remaining_credit; 196 struct timer_list credit_timeout; 197 u64 credit_window_start; 198 199 /* Statistics */ 200 struct xenvif_stats stats; 201 }; 202 203 /* Maximum number of Rx slots a to-guest packet may use, including the 204 * slot needed for GSO meta-data. 205 */ 206 #define XEN_NETBK_RX_SLOTS_MAX (MAX_SKB_FRAGS + 1) 207 208 enum state_bit_shift { 209 /* This bit marks that the vif is connected */ 210 VIF_STATUS_CONNECTED, 211 }; 212 213 struct xenvif_mcast_addr { 214 struct list_head entry; 215 struct rcu_head rcu; 216 u8 addr[6]; 217 }; 218 219 #define XEN_NETBK_MCAST_MAX 64 220 221 struct xenvif { 222 /* Unique identifier for this interface. */ 223 domid_t domid; 224 unsigned int handle; 225 226 u8 fe_dev_addr[6]; 227 struct list_head fe_mcast_addr; 228 unsigned int fe_mcast_count; 229 230 /* Frontend feature information. */ 231 int gso_mask; 232 int gso_prefix_mask; 233 234 u8 can_sg:1; 235 u8 ip_csum:1; 236 u8 ipv6_csum:1; 237 u8 multicast_control:1; 238 239 /* Is this interface disabled? True when backend discovers 240 * frontend is rogue. 241 */ 242 bool disabled; 243 unsigned long status; 244 unsigned long drain_timeout; 245 unsigned long stall_timeout; 246 247 /* Queues */ 248 struct xenvif_queue *queues; 249 unsigned int num_queues; /* active queues, resource allocated */ 250 unsigned int stalled_queues; 251 252 struct xenbus_watch credit_watch; 253 254 spinlock_t lock; 255 256 #ifdef CONFIG_DEBUG_FS 257 struct dentry *xenvif_dbg_root; 258 #endif 259 260 /* Miscellaneous private stuff. */ 261 struct net_device *dev; 262 }; 263 264 struct xenvif_rx_cb { 265 unsigned long expires; 266 int meta_slots_used; 267 }; 268 269 #define XENVIF_RX_CB(skb) ((struct xenvif_rx_cb *)(skb)->cb) 270 271 static inline struct xenbus_device *xenvif_to_xenbus_device(struct xenvif *vif) 272 { 273 return to_xenbus_device(vif->dev->dev.parent); 274 } 275 276 void xenvif_tx_credit_callback(unsigned long data); 277 278 struct xenvif *xenvif_alloc(struct device *parent, 279 domid_t domid, 280 unsigned int handle); 281 282 int xenvif_init_queue(struct xenvif_queue *queue); 283 void xenvif_deinit_queue(struct xenvif_queue *queue); 284 285 int xenvif_connect(struct xenvif_queue *queue, unsigned long tx_ring_ref, 286 unsigned long rx_ring_ref, unsigned int tx_evtchn, 287 unsigned int rx_evtchn); 288 void xenvif_disconnect(struct xenvif *vif); 289 void xenvif_free(struct xenvif *vif); 290 291 int xenvif_xenbus_init(void); 292 void xenvif_xenbus_fini(void); 293 294 int xenvif_schedulable(struct xenvif *vif); 295 296 int xenvif_queue_stopped(struct xenvif_queue *queue); 297 void xenvif_wake_queue(struct xenvif_queue *queue); 298 299 /* (Un)Map communication rings. */ 300 void xenvif_unmap_frontend_rings(struct xenvif_queue *queue); 301 int xenvif_map_frontend_rings(struct xenvif_queue *queue, 302 grant_ref_t tx_ring_ref, 303 grant_ref_t rx_ring_ref); 304 305 /* Check for SKBs from frontend and schedule backend processing */ 306 void xenvif_napi_schedule_or_enable_events(struct xenvif_queue *queue); 307 308 /* Prevent the device from generating any further traffic. */ 309 void xenvif_carrier_off(struct xenvif *vif); 310 311 int xenvif_tx_action(struct xenvif_queue *queue, int budget); 312 313 int xenvif_kthread_guest_rx(void *data); 314 void xenvif_kick_thread(struct xenvif_queue *queue); 315 316 int xenvif_dealloc_kthread(void *data); 317 318 void xenvif_rx_queue_tail(struct xenvif_queue *queue, struct sk_buff *skb); 319 320 /* Determine whether the needed number of slots (req) are available, 321 * and set req_event if not. 322 */ 323 bool xenvif_rx_ring_slots_available(struct xenvif_queue *queue, int needed); 324 325 void xenvif_carrier_on(struct xenvif *vif); 326 327 /* Callback from stack when TX packet can be released */ 328 void xenvif_zerocopy_callback(struct ubuf_info *ubuf, bool zerocopy_success); 329 330 /* Unmap a pending page and release it back to the guest */ 331 void xenvif_idx_unmap(struct xenvif_queue *queue, u16 pending_idx); 332 333 static inline pending_ring_idx_t nr_pending_reqs(struct xenvif_queue *queue) 334 { 335 return MAX_PENDING_REQS - 336 queue->pending_prod + queue->pending_cons; 337 } 338 339 irqreturn_t xenvif_interrupt(int irq, void *dev_id); 340 341 extern bool separate_tx_rx_irq; 342 343 extern unsigned int rx_drain_timeout_msecs; 344 extern unsigned int rx_stall_timeout_msecs; 345 extern unsigned int xenvif_max_queues; 346 347 #ifdef CONFIG_DEBUG_FS 348 extern struct dentry *xen_netback_dbg_root; 349 #endif 350 351 void xenvif_skb_zerocopy_prepare(struct xenvif_queue *queue, 352 struct sk_buff *skb); 353 void xenvif_skb_zerocopy_complete(struct xenvif_queue *queue); 354 355 /* Multicast control */ 356 bool xenvif_mcast_match(struct xenvif *vif, const u8 *addr); 357 void xenvif_mcast_addr_list_free(struct xenvif *vif); 358 359 #endif /* __XEN_NETBACK__COMMON_H__ */ 360