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 /* Callback data for released SKBs. The callback is always 56 * xenvif_zerocopy_callback, desc contains the pending_idx, which is 57 * also an index in pending_tx_info array. It is initialized in 58 * xenvif_alloc and it never changes. 59 * skb_shinfo(skb)->destructor_arg points to the first mapped slot's 60 * callback_struct in this array of struct pending_tx_info's, then ctx 61 * to the next, or NULL if there is no more slot for this skb. 62 * ubuf_to_vif is a helper which finds the struct xenvif from a pointer 63 * to this field. 64 */ 65 struct ubuf_info callback_struct; 66 }; 67 68 #define XEN_NETIF_TX_RING_SIZE __CONST_RING_SIZE(xen_netif_tx, XEN_PAGE_SIZE) 69 #define XEN_NETIF_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, XEN_PAGE_SIZE) 70 71 struct xenvif_rx_meta { 72 int id; 73 int size; 74 int gso_type; 75 int gso_size; 76 }; 77 78 #define GSO_BIT(type) \ 79 (1 << XEN_NETIF_GSO_TYPE_ ## type) 80 81 /* Discriminate from any valid pending_idx value. */ 82 #define INVALID_PENDING_IDX 0xFFFF 83 84 #define MAX_BUFFER_OFFSET XEN_PAGE_SIZE 85 86 #define MAX_PENDING_REQS XEN_NETIF_TX_RING_SIZE 87 88 /* The maximum number of frags is derived from the size of a grant (same 89 * as a Xen page size for now). 90 */ 91 #define MAX_XEN_SKB_FRAGS (65536 / XEN_PAGE_SIZE + 1) 92 93 /* It's possible for an skb to have a maximal number of frags 94 * but still be less than MAX_BUFFER_OFFSET in size. Thus the 95 * worst-case number of copy operations is MAX_XEN_SKB_FRAGS per 96 * ring slot. 97 */ 98 #define MAX_GRANT_COPY_OPS (MAX_XEN_SKB_FRAGS * XEN_NETIF_RX_RING_SIZE) 99 100 #define NETBACK_INVALID_HANDLE -1 101 102 /* To avoid confusion, we define XEN_NETBK_LEGACY_SLOTS_MAX indicating 103 * the maximum slots a valid packet can use. Now this value is defined 104 * to be XEN_NETIF_NR_SLOTS_MIN, which is supposed to be supported by 105 * all backend. 106 */ 107 #define XEN_NETBK_LEGACY_SLOTS_MAX XEN_NETIF_NR_SLOTS_MIN 108 109 /* Queue name is interface name with "-qNNN" appended */ 110 #define QUEUE_NAME_SIZE (IFNAMSIZ + 5) 111 112 /* IRQ name is queue name with "-tx" or "-rx" appended */ 113 #define IRQ_NAME_SIZE (QUEUE_NAME_SIZE + 3) 114 115 struct xenvif; 116 117 struct xenvif_stats { 118 /* Stats fields to be updated per-queue. 119 * A subset of struct net_device_stats that contains only the 120 * fields that are updated in netback.c for each queue. 121 */ 122 unsigned int rx_bytes; 123 unsigned int rx_packets; 124 unsigned int tx_bytes; 125 unsigned int tx_packets; 126 127 /* Additional stats used by xenvif */ 128 unsigned long rx_gso_checksum_fixup; 129 unsigned long tx_zerocopy_sent; 130 unsigned long tx_zerocopy_success; 131 unsigned long tx_zerocopy_fail; 132 unsigned long tx_frag_overflow; 133 }; 134 135 struct xenvif_queue { /* Per-queue data for xenvif */ 136 unsigned int id; /* Queue ID, 0-based */ 137 char name[QUEUE_NAME_SIZE]; /* DEVNAME-qN */ 138 struct xenvif *vif; /* Parent VIF */ 139 140 /* Use NAPI for guest TX */ 141 struct napi_struct napi; 142 /* When feature-split-event-channels = 0, tx_irq = rx_irq. */ 143 unsigned int tx_irq; 144 /* Only used when feature-split-event-channels = 1 */ 145 char tx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-tx */ 146 struct xen_netif_tx_back_ring tx; 147 struct sk_buff_head tx_queue; 148 struct page *mmap_pages[MAX_PENDING_REQS]; 149 pending_ring_idx_t pending_prod; 150 pending_ring_idx_t pending_cons; 151 u16 pending_ring[MAX_PENDING_REQS]; 152 struct pending_tx_info pending_tx_info[MAX_PENDING_REQS]; 153 grant_handle_t grant_tx_handle[MAX_PENDING_REQS]; 154 155 struct gnttab_copy tx_copy_ops[MAX_PENDING_REQS]; 156 struct gnttab_map_grant_ref tx_map_ops[MAX_PENDING_REQS]; 157 struct gnttab_unmap_grant_ref tx_unmap_ops[MAX_PENDING_REQS]; 158 /* passed to gnttab_[un]map_refs with pages under (un)mapping */ 159 struct page *pages_to_map[MAX_PENDING_REQS]; 160 struct page *pages_to_unmap[MAX_PENDING_REQS]; 161 162 /* This prevents zerocopy callbacks to race over dealloc_ring */ 163 spinlock_t callback_lock; 164 /* This prevents dealloc thread and NAPI instance to race over response 165 * creation and pending_ring in xenvif_idx_release. In xenvif_tx_err 166 * it only protect response creation 167 */ 168 spinlock_t response_lock; 169 pending_ring_idx_t dealloc_prod; 170 pending_ring_idx_t dealloc_cons; 171 u16 dealloc_ring[MAX_PENDING_REQS]; 172 struct task_struct *dealloc_task; 173 wait_queue_head_t dealloc_wq; 174 atomic_t inflight_packets; 175 176 /* Use kthread for guest RX */ 177 struct task_struct *task; 178 wait_queue_head_t wq; 179 /* When feature-split-event-channels = 0, tx_irq = rx_irq. */ 180 unsigned int rx_irq; 181 /* Only used when feature-split-event-channels = 1 */ 182 char rx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-rx */ 183 struct xen_netif_rx_back_ring rx; 184 struct sk_buff_head rx_queue; 185 186 unsigned int rx_queue_max; 187 unsigned int rx_queue_len; 188 unsigned long last_rx_time; 189 bool stalled; 190 191 struct gnttab_copy grant_copy_op[MAX_GRANT_COPY_OPS]; 192 193 /* We create one meta structure per ring request we consume, so 194 * the maximum number is the same as the ring size. 195 */ 196 struct xenvif_rx_meta meta[XEN_NETIF_RX_RING_SIZE]; 197 198 /* Transmit shaping: allow 'credit_bytes' every 'credit_usec'. */ 199 unsigned long credit_bytes; 200 unsigned long credit_usec; 201 unsigned long remaining_credit; 202 struct timer_list credit_timeout; 203 u64 credit_window_start; 204 205 /* Statistics */ 206 struct xenvif_stats stats; 207 }; 208 209 enum state_bit_shift { 210 /* This bit marks that the vif is connected */ 211 VIF_STATUS_CONNECTED, 212 }; 213 214 struct xenvif_mcast_addr { 215 struct list_head entry; 216 struct rcu_head rcu; 217 u8 addr[6]; 218 }; 219 220 #define XEN_NETBK_MCAST_MAX 64 221 222 struct xenvif { 223 /* Unique identifier for this interface. */ 224 domid_t domid; 225 unsigned int handle; 226 227 u8 fe_dev_addr[6]; 228 struct list_head fe_mcast_addr; 229 unsigned int fe_mcast_count; 230 231 /* Frontend feature information. */ 232 int gso_mask; 233 int gso_prefix_mask; 234 235 u8 can_sg:1; 236 u8 ip_csum:1; 237 u8 ipv6_csum:1; 238 u8 multicast_control:1; 239 240 /* Is this interface disabled? True when backend discovers 241 * frontend is rogue. 242 */ 243 bool disabled; 244 unsigned long status; 245 unsigned long drain_timeout; 246 unsigned long stall_timeout; 247 248 /* Queues */ 249 struct xenvif_queue *queues; 250 unsigned int num_queues; /* active queues, resource allocated */ 251 unsigned int stalled_queues; 252 253 struct xenbus_watch credit_watch; 254 255 spinlock_t lock; 256 257 #ifdef CONFIG_DEBUG_FS 258 struct dentry *xenvif_dbg_root; 259 #endif 260 261 /* Miscellaneous private stuff. */ 262 struct net_device *dev; 263 }; 264 265 struct xenvif_rx_cb { 266 unsigned long expires; 267 int meta_slots_used; 268 }; 269 270 #define XENVIF_RX_CB(skb) ((struct xenvif_rx_cb *)(skb)->cb) 271 272 static inline struct xenbus_device *xenvif_to_xenbus_device(struct xenvif *vif) 273 { 274 return to_xenbus_device(vif->dev->dev.parent); 275 } 276 277 void xenvif_tx_credit_callback(unsigned long data); 278 279 struct xenvif *xenvif_alloc(struct device *parent, 280 domid_t domid, 281 unsigned int handle); 282 283 int xenvif_init_queue(struct xenvif_queue *queue); 284 void xenvif_deinit_queue(struct xenvif_queue *queue); 285 286 int xenvif_connect(struct xenvif_queue *queue, unsigned long tx_ring_ref, 287 unsigned long rx_ring_ref, unsigned int tx_evtchn, 288 unsigned int rx_evtchn); 289 void xenvif_disconnect(struct xenvif *vif); 290 void xenvif_free(struct xenvif *vif); 291 292 int xenvif_xenbus_init(void); 293 void xenvif_xenbus_fini(void); 294 295 int xenvif_schedulable(struct xenvif *vif); 296 297 int xenvif_queue_stopped(struct xenvif_queue *queue); 298 void xenvif_wake_queue(struct xenvif_queue *queue); 299 300 /* (Un)Map communication rings. */ 301 void xenvif_unmap_frontend_rings(struct xenvif_queue *queue); 302 int xenvif_map_frontend_rings(struct xenvif_queue *queue, 303 grant_ref_t tx_ring_ref, 304 grant_ref_t rx_ring_ref); 305 306 /* Check for SKBs from frontend and schedule backend processing */ 307 void xenvif_napi_schedule_or_enable_events(struct xenvif_queue *queue); 308 309 /* Prevent the device from generating any further traffic. */ 310 void xenvif_carrier_off(struct xenvif *vif); 311 312 int xenvif_tx_action(struct xenvif_queue *queue, int budget); 313 314 int xenvif_kthread_guest_rx(void *data); 315 void xenvif_kick_thread(struct xenvif_queue *queue); 316 317 int xenvif_dealloc_kthread(void *data); 318 319 void xenvif_rx_queue_tail(struct xenvif_queue *queue, struct sk_buff *skb); 320 321 void xenvif_carrier_on(struct xenvif *vif); 322 323 /* Callback from stack when TX packet can be released */ 324 void xenvif_zerocopy_callback(struct ubuf_info *ubuf, bool zerocopy_success); 325 326 /* Unmap a pending page and release it back to the guest */ 327 void xenvif_idx_unmap(struct xenvif_queue *queue, u16 pending_idx); 328 329 static inline pending_ring_idx_t nr_pending_reqs(struct xenvif_queue *queue) 330 { 331 return MAX_PENDING_REQS - 332 queue->pending_prod + queue->pending_cons; 333 } 334 335 irqreturn_t xenvif_interrupt(int irq, void *dev_id); 336 337 extern bool separate_tx_rx_irq; 338 339 extern unsigned int rx_drain_timeout_msecs; 340 extern unsigned int rx_stall_timeout_msecs; 341 extern unsigned int xenvif_max_queues; 342 343 #ifdef CONFIG_DEBUG_FS 344 extern struct dentry *xen_netback_dbg_root; 345 #endif 346 347 void xenvif_skb_zerocopy_prepare(struct xenvif_queue *queue, 348 struct sk_buff *skb); 349 void xenvif_skb_zerocopy_complete(struct xenvif_queue *queue); 350 351 /* Multicast control */ 352 bool xenvif_mcast_match(struct xenvif *vif, const u8 *addr); 353 void xenvif_mcast_addr_list_free(struct xenvif *vif); 354 355 #endif /* __XEN_NETBACK__COMMON_H__ */ 356