1 /*- 2 * BSD LICENSE 3 * 4 * Copyright (c) Intel Corporation. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 11 * * Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * * Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in 15 * the documentation and/or other materials provided with the 16 * distribution. 17 * * Neither the name of Intel Corporation nor the names of its 18 * contributors may be used to endorse or promote products derived 19 * from this software without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 32 */ 33 34 #ifndef SPDK_VHOST_INTERNAL_H 35 #define SPDK_VHOST_INTERNAL_H 36 #include <linux/virtio_config.h> 37 38 #include "spdk/stdinc.h" 39 40 #include <rte_vhost.h> 41 42 #include "spdk_internal/vhost_user.h" 43 #include "spdk/log.h" 44 #include "spdk/util.h" 45 #include "spdk/rpc.h" 46 #include "spdk/config.h" 47 48 #define SPDK_VHOST_MAX_VQUEUES 256 49 #define SPDK_VHOST_MAX_VQ_SIZE 1024 50 51 #define SPDK_VHOST_SCSI_CTRLR_MAX_DEVS 8 52 53 #define SPDK_VHOST_IOVS_MAX 129 54 55 #define SPDK_VHOST_VQ_MAX_SUBMISSIONS 32 56 57 /* 58 * Rate at which stats are checked for interrupt coalescing. 59 */ 60 #define SPDK_VHOST_STATS_CHECK_INTERVAL_MS 10 61 /* 62 * Default threshold at which interrupts start to be coalesced. 63 */ 64 #define SPDK_VHOST_VQ_IOPS_COALESCING_THRESHOLD 60000 65 66 /* 67 * Currently coalescing is not used by default. 68 * Setting this to value > 0 here or by RPC will enable coalescing. 69 */ 70 #define SPDK_VHOST_COALESCING_DELAY_BASE_US 0 71 72 #define SPDK_VHOST_FEATURES ((1ULL << VHOST_F_LOG_ALL) | \ 73 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES) | \ 74 (1ULL << VIRTIO_F_VERSION_1) | \ 75 (1ULL << VIRTIO_F_NOTIFY_ON_EMPTY) | \ 76 (1ULL << VIRTIO_RING_F_EVENT_IDX) | \ 77 (1ULL << VIRTIO_RING_F_INDIRECT_DESC) | \ 78 (1ULL << VIRTIO_F_RING_PACKED)) 79 80 #define SPDK_VHOST_DISABLED_FEATURES ((1ULL << VIRTIO_RING_F_EVENT_IDX) | \ 81 (1ULL << VIRTIO_F_NOTIFY_ON_EMPTY)) 82 83 #define VRING_DESC_F_AVAIL (1ULL << VRING_PACKED_DESC_F_AVAIL) 84 #define VRING_DESC_F_USED (1ULL << VRING_PACKED_DESC_F_USED) 85 #define VRING_DESC_F_AVAIL_USED (VRING_DESC_F_AVAIL | VRING_DESC_F_USED) 86 87 typedef struct rte_vhost_resubmit_desc spdk_vhost_resubmit_desc; 88 typedef struct rte_vhost_resubmit_info spdk_vhost_resubmit_info; 89 90 struct spdk_vhost_virtqueue { 91 struct rte_vhost_vring vring; 92 struct rte_vhost_ring_inflight vring_inflight; 93 uint16_t last_avail_idx; 94 uint16_t last_used_idx; 95 96 struct { 97 /* To mark a descriptor as available in packed ring 98 * Equal to avail_wrap_counter in spec. 99 */ 100 uint8_t avail_phase : 1; 101 /* To mark a descriptor as used in packed ring 102 * Equal to used_wrap_counter in spec. 103 */ 104 uint8_t used_phase : 1; 105 uint8_t padding : 5; 106 bool packed_ring : 1; 107 } packed; 108 109 void *tasks; 110 111 /* Request count from last stats check */ 112 uint32_t req_cnt; 113 114 /* Request count from last event */ 115 uint16_t used_req_cnt; 116 117 /* How long interrupt is delayed */ 118 uint32_t irq_delay_time; 119 120 /* Next time when we need to send event */ 121 uint64_t next_event_time; 122 123 /* Associated vhost_virtqueue in the virtio device's virtqueue list */ 124 uint32_t vring_idx; 125 126 struct spdk_vhost_session *vsession; 127 128 struct spdk_interrupt *intr; 129 } __attribute((aligned(SPDK_CACHE_LINE_SIZE))); 130 131 struct spdk_vhost_session { 132 struct spdk_vhost_dev *vdev; 133 134 /* rte_vhost connection ID. */ 135 int vid; 136 137 /* Unique session ID. */ 138 uint64_t id; 139 /* Unique session name. */ 140 char *name; 141 142 bool initialized; 143 bool started; 144 bool needs_restart; 145 bool forced_polling; 146 bool interrupt_mode; 147 148 struct rte_vhost_memory *mem; 149 150 int task_cnt; 151 152 uint16_t max_queues; 153 154 uint64_t negotiated_features; 155 156 /* Local copy of device coalescing settings. */ 157 uint32_t coalescing_delay_time_base; 158 uint32_t coalescing_io_rate_threshold; 159 160 /* Next time when stats for event coalescing will be checked. */ 161 uint64_t next_stats_check_time; 162 163 /* Interval used for event coalescing checking. */ 164 uint64_t stats_check_interval; 165 166 struct spdk_vhost_virtqueue virtqueue[SPDK_VHOST_MAX_VQUEUES]; 167 168 TAILQ_ENTRY(spdk_vhost_session) tailq; 169 }; 170 171 struct spdk_vhost_dev { 172 char *name; 173 char *path; 174 175 struct spdk_thread *thread; 176 bool registered; 177 178 uint64_t virtio_features; 179 uint64_t disabled_features; 180 uint64_t protocol_features; 181 182 const struct spdk_vhost_dev_backend *backend; 183 184 /* Saved orginal values used to setup coalescing to avoid integer 185 * rounding issues during save/load config. 186 */ 187 uint32_t coalescing_delay_us; 188 uint32_t coalescing_iops_threshold; 189 190 /* Current connections to the device */ 191 TAILQ_HEAD(, spdk_vhost_session) vsessions; 192 193 /* Increment-only session counter */ 194 uint64_t vsessions_num; 195 196 /* Number of started and actively polled sessions */ 197 uint32_t active_session_num; 198 199 /* Number of pending asynchronous operations */ 200 uint32_t pending_async_op_num; 201 202 TAILQ_ENTRY(spdk_vhost_dev) tailq; 203 }; 204 205 /** 206 * \param vdev vhost device. 207 * \param vsession vhost session. 208 * \param arg user-provided parameter. 209 * 210 * \return negative values will break the foreach call, meaning 211 * the function won't be called again. Return codes zero and 212 * positive don't have any effect. 213 */ 214 typedef int (*spdk_vhost_session_fn)(struct spdk_vhost_dev *vdev, 215 struct spdk_vhost_session *vsession, 216 void *arg); 217 218 /** 219 * \param vdev vhost device. 220 * \param arg user-provided parameter. 221 */ 222 typedef void (*spdk_vhost_dev_fn)(struct spdk_vhost_dev *vdev, void *arg); 223 224 struct spdk_vhost_dev_backend { 225 /** 226 * Size of additional per-session context data 227 * allocated whenever a new client connects. 228 */ 229 size_t session_ctx_size; 230 231 int (*start_session)(struct spdk_vhost_session *vsession); 232 int (*stop_session)(struct spdk_vhost_session *vsession); 233 234 int (*vhost_get_config)(struct spdk_vhost_dev *vdev, uint8_t *config, uint32_t len); 235 int (*vhost_set_config)(struct spdk_vhost_dev *vdev, uint8_t *config, 236 uint32_t offset, uint32_t size, uint32_t flags); 237 238 void (*dump_info_json)(struct spdk_vhost_dev *vdev, struct spdk_json_write_ctx *w); 239 void (*write_config_json)(struct spdk_vhost_dev *vdev, struct spdk_json_write_ctx *w); 240 int (*remove_device)(struct spdk_vhost_dev *vdev); 241 }; 242 243 void *vhost_gpa_to_vva(struct spdk_vhost_session *vsession, uint64_t addr, uint64_t len); 244 245 uint16_t vhost_vq_avail_ring_get(struct spdk_vhost_virtqueue *vq, uint16_t *reqs, 246 uint16_t reqs_len); 247 248 /** 249 * Get a virtio split descriptor at given index in given virtqueue. 250 * The descriptor will provide access to the entire descriptor 251 * chain. The subsequent descriptors are accesible via 252 * \c spdk_vhost_vring_desc_get_next. 253 * \param vsession vhost session 254 * \param vq virtqueue 255 * \param req_idx descriptor index 256 * \param desc pointer to be set to the descriptor 257 * \param desc_table descriptor table to be used with 258 * \c spdk_vhost_vring_desc_get_next. This might be either 259 * default virtqueue descriptor table or per-chain indirect 260 * table. 261 * \param desc_table_size size of the *desc_table* 262 * \return 0 on success, -1 if given index is invalid. 263 * If -1 is returned, the content of params is undefined. 264 */ 265 int vhost_vq_get_desc(struct spdk_vhost_session *vsession, struct spdk_vhost_virtqueue *vq, 266 uint16_t req_idx, struct vring_desc **desc, struct vring_desc **desc_table, 267 uint32_t *desc_table_size); 268 269 /** 270 * Get a virtio packed descriptor at given index in given virtqueue. 271 * The descriptor will provide access to the entire descriptor 272 * chain. The subsequent descriptors are accesible via 273 * \c vhost_vring_packed_desc_get_next. 274 * \param vsession vhost session 275 * \param vq virtqueue 276 * \param req_idx descriptor index 277 * \param desc pointer to be set to the descriptor 278 * \param desc_table descriptor table to be used with 279 * \c spdk_vhost_vring_desc_get_next. This might be either 280 * \c NULL or per-chain indirect table. 281 * \param desc_table_size size of the *desc_table* 282 * \return 0 on success, -1 if given index is invalid. 283 * If -1 is returned, the content of params is undefined. 284 */ 285 int vhost_vq_get_desc_packed(struct spdk_vhost_session *vsession, 286 struct spdk_vhost_virtqueue *virtqueue, 287 uint16_t req_idx, struct vring_packed_desc **desc, 288 struct vring_packed_desc **desc_table, uint32_t *desc_table_size); 289 290 /** 291 * Send IRQ/call client (if pending) for \c vq. 292 * \param vsession vhost session 293 * \param vq virtqueue 294 * \return 295 * 0 - if no interrupt was signalled 296 * 1 - if interrupt was signalled 297 */ 298 int vhost_vq_used_signal(struct spdk_vhost_session *vsession, struct spdk_vhost_virtqueue *vq); 299 300 301 /** 302 * Send IRQs for all queues that need to be signaled. 303 * \param vsession vhost session 304 * \param vq virtqueue 305 */ 306 void vhost_session_used_signal(struct spdk_vhost_session *vsession); 307 308 /** 309 * Send IRQs for the queue that need to be signaled. 310 * \param vq virtqueue 311 */ 312 void vhost_session_vq_used_signal(struct spdk_vhost_virtqueue *virtqueue); 313 314 void vhost_vq_used_ring_enqueue(struct spdk_vhost_session *vsession, 315 struct spdk_vhost_virtqueue *vq, 316 uint16_t id, uint32_t len); 317 318 /** 319 * Enqueue the entry to the used ring when device complete the request. 320 * \param vsession vhost session 321 * \param vq virtqueue 322 * \req_idx descriptor index. It's the first index of this descriptor chain. 323 * \num_descs descriptor count. It's the count of the number of buffers in the chain. 324 * \buffer_id descriptor buffer ID. 325 * \length device write length. Specify the length of the buffer that has been initialized 326 * (written to) by the device 327 */ 328 void vhost_vq_packed_ring_enqueue(struct spdk_vhost_session *vsession, 329 struct spdk_vhost_virtqueue *virtqueue, 330 uint16_t num_descs, uint16_t buffer_id, 331 uint32_t length); 332 333 /** 334 * Get subsequent descriptor from given table. 335 * \param desc current descriptor, will be set to the 336 * next descriptor (NULL in case this is the last 337 * descriptor in the chain or the next desc is invalid) 338 * \param desc_table descriptor table 339 * \param desc_table_size size of the *desc_table* 340 * \return 0 on success, -1 if given index is invalid 341 * The *desc* param will be set regardless of the 342 * return value. 343 */ 344 int vhost_vring_desc_get_next(struct vring_desc **desc, 345 struct vring_desc *desc_table, uint32_t desc_table_size); 346 static inline bool 347 vhost_vring_desc_is_wr(struct vring_desc *cur_desc) 348 { 349 return !!(cur_desc->flags & VRING_DESC_F_WRITE); 350 } 351 352 int vhost_vring_desc_to_iov(struct spdk_vhost_session *vsession, struct iovec *iov, 353 uint16_t *iov_index, const struct vring_desc *desc); 354 355 bool vhost_vq_packed_ring_is_avail(struct spdk_vhost_virtqueue *virtqueue); 356 357 /** 358 * Get subsequent descriptor from vq or desc table. 359 * \param desc current descriptor, will be set to the 360 * next descriptor (NULL in case this is the last 361 * descriptor in the chain or the next desc is invalid) 362 * \req_idx index of current desc, will be set to the next 363 * index. If desc_table != NULL the req_idx is the the vring index 364 * or the req_idx is the desc_table index. 365 * \param desc_table descriptor table 366 * \param desc_table_size size of the *desc_table* 367 * \return 0 on success, -1 if given index is invalid 368 * The *desc* param will be set regardless of the 369 * return value. 370 */ 371 int vhost_vring_packed_desc_get_next(struct vring_packed_desc **desc, uint16_t *req_idx, 372 struct spdk_vhost_virtqueue *vq, 373 struct vring_packed_desc *desc_table, 374 uint32_t desc_table_size); 375 376 bool vhost_vring_packed_desc_is_wr(struct vring_packed_desc *cur_desc); 377 378 int vhost_vring_packed_desc_to_iov(struct spdk_vhost_session *vsession, struct iovec *iov, 379 uint16_t *iov_index, const struct vring_packed_desc *desc); 380 381 uint16_t vhost_vring_packed_desc_get_buffer_id(struct spdk_vhost_virtqueue *vq, uint16_t req_idx, 382 uint16_t *num_descs); 383 384 static inline bool __attribute__((always_inline)) 385 vhost_dev_has_feature(struct spdk_vhost_session *vsession, unsigned feature_id) 386 { 387 return vsession->negotiated_features & (1ULL << feature_id); 388 } 389 390 int vhost_dev_register(struct spdk_vhost_dev *vdev, const char *name, const char *mask_str, 391 const struct spdk_vhost_dev_backend *backend); 392 int vhost_dev_unregister(struct spdk_vhost_dev *vdev); 393 394 void vhost_dump_info_json(struct spdk_vhost_dev *vdev, struct spdk_json_write_ctx *w); 395 396 /* 397 * Vhost callbacks for vhost_device_ops interface 398 */ 399 400 int vhost_new_connection_cb(int vid, const char *ifname); 401 int vhost_start_device_cb(int vid); 402 int vhost_stop_device_cb(int vid); 403 int vhost_destroy_connection_cb(int vid); 404 405 /* 406 * Memory registration functions used in start/stop device callbacks 407 */ 408 void vhost_session_mem_register(struct rte_vhost_memory *mem); 409 void vhost_session_mem_unregister(struct rte_vhost_memory *mem); 410 411 /* 412 * Call a function for each session of the provided vhost device. 413 * The function will be called one-by-one on each session's thread. 414 * 415 * \param vdev vhost device 416 * \param fn function to call on each session's thread 417 * \param cpl_fn function to be called at the end of the iteration on 418 * the vhost management thread. 419 * Optional, can be NULL. 420 * \param arg additional argument to the both callbacks 421 */ 422 void vhost_dev_foreach_session(struct spdk_vhost_dev *dev, 423 spdk_vhost_session_fn fn, 424 spdk_vhost_dev_fn cpl_fn, 425 void *arg); 426 427 /** 428 * Call a function on the provided lcore and block until either 429 * spdk_vhost_session_start_done() or spdk_vhost_session_stop_done() 430 * is called. 431 * 432 * This must be called under the global vhost mutex, which this function 433 * will unlock for the time it's waiting. It's meant to be called only 434 * from start/stop session callbacks. 435 * 436 * \param vsession vhost session 437 * \param cb_fn the function to call. The void *arg parameter in cb_fn 438 * is always NULL. 439 * \param timeout_sec timeout in seconds. This function will still 440 * block after the timeout expires, but will print the provided errmsg. 441 * \param errmsg error message to print once the timeout expires 442 * \return return the code passed to spdk_vhost_session_event_done(). 443 */ 444 int vhost_session_send_event(struct spdk_vhost_session *vsession, 445 spdk_vhost_session_fn cb_fn, unsigned timeout_sec, 446 const char *errmsg); 447 448 /** 449 * Finish a blocking spdk_vhost_session_send_event() call and finally 450 * start the session. This must be called on the target lcore, which 451 * will now receive all session-related messages (e.g. from 452 * spdk_vhost_dev_foreach_session()). 453 * 454 * Must be called under the global vhost lock. 455 * 456 * \param vsession vhost session 457 * \param response return code 458 */ 459 void vhost_session_start_done(struct spdk_vhost_session *vsession, int response); 460 461 /** 462 * Finish a blocking spdk_vhost_session_send_event() call and finally 463 * stop the session. This must be called on the session's lcore which 464 * used to receive all session-related messages (e.g. from 465 * spdk_vhost_dev_foreach_session()). After this call, the session- 466 * related messages will be once again processed by any arbitrary thread. 467 * 468 * Must be called under the global vhost lock. 469 * 470 * Must be called under the global vhost mutex. 471 * 472 * \param vsession vhost session 473 * \param response return code 474 */ 475 void vhost_session_stop_done(struct spdk_vhost_session *vsession, int response); 476 477 struct spdk_vhost_session *vhost_session_find_by_vid(int vid); 478 void vhost_session_install_rte_compat_hooks(struct spdk_vhost_session *vsession); 479 int vhost_register_unix_socket(const char *path, const char *ctrl_name, 480 uint64_t virtio_features, uint64_t disabled_features, uint64_t protocol_features); 481 int vhost_driver_unregister(const char *path); 482 int vhost_get_mem_table(int vid, struct rte_vhost_memory **mem); 483 int vhost_get_negotiated_features(int vid, uint64_t *negotiated_features); 484 485 int remove_vhost_controller(struct spdk_vhost_dev *vdev); 486 487 #endif /* SPDK_VHOST_INTERNAL_H */ 488