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 (1ULL << VIRTIO_F_ANY_LAYOUT)) 80 81 #define SPDK_VHOST_DISABLED_FEATURES ((1ULL << VIRTIO_RING_F_EVENT_IDX) | \ 82 (1ULL << VIRTIO_F_NOTIFY_ON_EMPTY)) 83 84 #define VRING_DESC_F_AVAIL (1ULL << VRING_PACKED_DESC_F_AVAIL) 85 #define VRING_DESC_F_USED (1ULL << VRING_PACKED_DESC_F_USED) 86 #define VRING_DESC_F_AVAIL_USED (VRING_DESC_F_AVAIL | VRING_DESC_F_USED) 87 88 typedef struct rte_vhost_resubmit_desc spdk_vhost_resubmit_desc; 89 typedef struct rte_vhost_resubmit_info spdk_vhost_resubmit_info; 90 typedef struct rte_vhost_inflight_desc_packed spdk_vhost_inflight_desc; 91 92 struct spdk_vhost_virtqueue { 93 struct rte_vhost_vring vring; 94 struct rte_vhost_ring_inflight vring_inflight; 95 uint16_t last_avail_idx; 96 uint16_t last_used_idx; 97 98 struct { 99 /* To mark a descriptor as available in packed ring 100 * Equal to avail_wrap_counter in spec. 101 */ 102 uint8_t avail_phase : 1; 103 /* To mark a descriptor as used in packed ring 104 * Equal to used_wrap_counter in spec. 105 */ 106 uint8_t used_phase : 1; 107 uint8_t padding : 5; 108 bool packed_ring : 1; 109 } packed; 110 111 void *tasks; 112 113 /* Request count from last stats check */ 114 uint32_t req_cnt; 115 116 /* Request count from last event */ 117 uint16_t used_req_cnt; 118 119 /* How long interrupt is delayed */ 120 uint32_t irq_delay_time; 121 122 /* Next time when we need to send event */ 123 uint64_t next_event_time; 124 125 /* Associated vhost_virtqueue in the virtio device's virtqueue list */ 126 uint32_t vring_idx; 127 128 struct spdk_vhost_session *vsession; 129 130 struct spdk_interrupt *intr; 131 } __attribute((aligned(SPDK_CACHE_LINE_SIZE))); 132 133 struct spdk_vhost_session { 134 struct spdk_vhost_dev *vdev; 135 136 /* rte_vhost connection ID. */ 137 int vid; 138 139 /* Unique session ID. */ 140 uint64_t id; 141 /* Unique session name. */ 142 char *name; 143 144 bool initialized; 145 bool started; 146 bool needs_restart; 147 bool forced_polling; 148 bool interrupt_mode; 149 150 struct rte_vhost_memory *mem; 151 152 int task_cnt; 153 154 uint16_t max_queues; 155 156 uint64_t negotiated_features; 157 158 /* Local copy of device coalescing settings. */ 159 uint32_t coalescing_delay_time_base; 160 uint32_t coalescing_io_rate_threshold; 161 162 /* Next time when stats for event coalescing will be checked. */ 163 uint64_t next_stats_check_time; 164 165 /* Interval used for event coalescing checking. */ 166 uint64_t stats_check_interval; 167 168 /* Session's stop poller will only try limited times to destroy the session. */ 169 uint32_t stop_retry_count; 170 171 struct spdk_vhost_virtqueue virtqueue[SPDK_VHOST_MAX_VQUEUES]; 172 173 TAILQ_ENTRY(spdk_vhost_session) tailq; 174 }; 175 176 struct spdk_vhost_user_dev { 177 struct spdk_vhost_dev *vdev; 178 179 const struct spdk_vhost_user_dev_backend *user_backend; 180 181 /* Saved original values used to setup coalescing to avoid integer 182 * rounding issues during save/load config. 183 */ 184 uint32_t coalescing_delay_us; 185 uint32_t coalescing_iops_threshold; 186 187 /* Current connections to the device */ 188 TAILQ_HEAD(, spdk_vhost_session) vsessions; 189 190 /* Increment-only session counter */ 191 uint64_t vsessions_num; 192 193 /* Number of started and actively polled sessions */ 194 uint32_t active_session_num; 195 196 /* Number of pending asynchronous operations */ 197 uint32_t pending_async_op_num; 198 }; 199 200 struct spdk_vhost_dev { 201 char *name; 202 char *path; 203 204 struct spdk_thread *thread; 205 bool registered; 206 207 uint64_t virtio_features; 208 uint64_t disabled_features; 209 uint64_t protocol_features; 210 bool packed_ring_recovery; 211 212 const struct spdk_vhost_dev_backend *backend; 213 214 /* Context passed from transport */ 215 void *ctxt; 216 217 TAILQ_ENTRY(spdk_vhost_dev) tailq; 218 }; 219 220 /** 221 * \param vdev vhost device. 222 * \param vsession vhost session. 223 * \param arg user-provided parameter. 224 * 225 * \return negative values will break the foreach call, meaning 226 * the function won't be called again. Return codes zero and 227 * positive don't have any effect. 228 */ 229 typedef int (*spdk_vhost_session_fn)(struct spdk_vhost_dev *vdev, 230 struct spdk_vhost_session *vsession, 231 void *arg); 232 233 /** 234 * \param vdev vhost device. 235 * \param arg user-provided parameter. 236 */ 237 typedef void (*spdk_vhost_dev_fn)(struct spdk_vhost_dev *vdev, void *arg); 238 239 struct spdk_vhost_user_dev_backend { 240 /** 241 * Size of additional per-session context data 242 * allocated whenever a new client connects. 243 */ 244 size_t session_ctx_size; 245 246 int (*start_session)(struct spdk_vhost_session *vsession); 247 int (*stop_session)(struct spdk_vhost_session *vsession); 248 }; 249 250 struct spdk_vhost_dev_backend { 251 int (*vhost_get_config)(struct spdk_vhost_dev *vdev, uint8_t *config, uint32_t len); 252 int (*vhost_set_config)(struct spdk_vhost_dev *vdev, uint8_t *config, 253 uint32_t offset, uint32_t size, uint32_t flags); 254 255 void (*dump_info_json)(struct spdk_vhost_dev *vdev, struct spdk_json_write_ctx *w); 256 void (*write_config_json)(struct spdk_vhost_dev *vdev, struct spdk_json_write_ctx *w); 257 int (*remove_device)(struct spdk_vhost_dev *vdev); 258 }; 259 260 void *vhost_gpa_to_vva(struct spdk_vhost_session *vsession, uint64_t addr, uint64_t len); 261 262 uint16_t vhost_vq_avail_ring_get(struct spdk_vhost_virtqueue *vq, uint16_t *reqs, 263 uint16_t reqs_len); 264 265 /** 266 * Get a virtio split descriptor at given index in given virtqueue. 267 * The descriptor will provide access to the entire descriptor 268 * chain. The subsequent descriptors are accessible via 269 * \c spdk_vhost_vring_desc_get_next. 270 * \param vsession vhost session 271 * \param vq virtqueue 272 * \param req_idx descriptor index 273 * \param desc pointer to be set to the descriptor 274 * \param desc_table descriptor table to be used with 275 * \c spdk_vhost_vring_desc_get_next. This might be either 276 * default virtqueue descriptor table or per-chain indirect 277 * table. 278 * \param desc_table_size size of the *desc_table* 279 * \return 0 on success, -1 if given index is invalid. 280 * If -1 is returned, the content of params is undefined. 281 */ 282 int vhost_vq_get_desc(struct spdk_vhost_session *vsession, struct spdk_vhost_virtqueue *vq, 283 uint16_t req_idx, struct vring_desc **desc, struct vring_desc **desc_table, 284 uint32_t *desc_table_size); 285 286 /** 287 * Get a virtio packed descriptor at given index in given virtqueue. 288 * The descriptor will provide access to the entire descriptor 289 * chain. The subsequent descriptors are accessible via 290 * \c vhost_vring_packed_desc_get_next. 291 * \param vsession vhost session 292 * \param vq virtqueue 293 * \param req_idx descriptor index 294 * \param desc pointer to be set to the descriptor 295 * \param desc_table descriptor table to be used with 296 * \c spdk_vhost_vring_desc_get_next. This might be either 297 * \c NULL or per-chain indirect table. 298 * \param desc_table_size size of the *desc_table* 299 * \return 0 on success, -1 if given index is invalid. 300 * If -1 is returned, the content of params is undefined. 301 */ 302 int vhost_vq_get_desc_packed(struct spdk_vhost_session *vsession, 303 struct spdk_vhost_virtqueue *virtqueue, 304 uint16_t req_idx, struct vring_packed_desc **desc, 305 struct vring_packed_desc **desc_table, uint32_t *desc_table_size); 306 307 int vhost_inflight_queue_get_desc(struct spdk_vhost_session *vsession, 308 spdk_vhost_inflight_desc *desc_array, 309 uint16_t req_idx, spdk_vhost_inflight_desc **desc, 310 struct vring_packed_desc **desc_table, uint32_t *desc_table_size); 311 312 /** 313 * Send IRQ/call client (if pending) for \c vq. 314 * \param vsession vhost session 315 * \param vq virtqueue 316 * \return 317 * 0 - if no interrupt was signalled 318 * 1 - if interrupt was signalled 319 */ 320 int vhost_vq_used_signal(struct spdk_vhost_session *vsession, struct spdk_vhost_virtqueue *vq); 321 322 323 /** 324 * Send IRQs for all queues that need to be signaled. 325 * \param vsession vhost session 326 * \param vq virtqueue 327 */ 328 void vhost_session_used_signal(struct spdk_vhost_session *vsession); 329 330 /** 331 * Send IRQs for the queue that need to be signaled. 332 * \param vq virtqueue 333 */ 334 void vhost_session_vq_used_signal(struct spdk_vhost_virtqueue *virtqueue); 335 336 void vhost_vq_used_ring_enqueue(struct spdk_vhost_session *vsession, 337 struct spdk_vhost_virtqueue *vq, 338 uint16_t id, uint32_t len); 339 340 /** 341 * Enqueue the entry to the used ring when device complete the request. 342 * \param vsession vhost session 343 * \param vq virtqueue 344 * \req_idx descriptor index. It's the first index of this descriptor chain. 345 * \num_descs descriptor count. It's the count of the number of buffers in the chain. 346 * \buffer_id descriptor buffer ID. 347 * \length device write length. Specify the length of the buffer that has been initialized 348 * (written to) by the device 349 * \inflight_head the head idx of this IO inflight desc chain. 350 */ 351 void vhost_vq_packed_ring_enqueue(struct spdk_vhost_session *vsession, 352 struct spdk_vhost_virtqueue *virtqueue, 353 uint16_t num_descs, uint16_t buffer_id, 354 uint32_t length, uint16_t inflight_head); 355 356 /** 357 * Get subsequent descriptor from given table. 358 * \param desc current descriptor, will be set to the 359 * next descriptor (NULL in case this is the last 360 * descriptor in the chain or the next desc is invalid) 361 * \param desc_table descriptor table 362 * \param desc_table_size size of the *desc_table* 363 * \return 0 on success, -1 if given index is invalid 364 * The *desc* param will be set regardless of the 365 * return value. 366 */ 367 int vhost_vring_desc_get_next(struct vring_desc **desc, 368 struct vring_desc *desc_table, uint32_t desc_table_size); 369 static inline bool 370 vhost_vring_desc_is_wr(struct vring_desc *cur_desc) 371 { 372 return !!(cur_desc->flags & VRING_DESC_F_WRITE); 373 } 374 375 int vhost_vring_desc_to_iov(struct spdk_vhost_session *vsession, struct iovec *iov, 376 uint16_t *iov_index, const struct vring_desc *desc); 377 378 bool vhost_vq_packed_ring_is_avail(struct spdk_vhost_virtqueue *virtqueue); 379 380 /** 381 * Get subsequent descriptor from vq or desc table. 382 * \param desc current descriptor, will be set to the 383 * next descriptor (NULL in case this is the last 384 * descriptor in the chain or the next desc is invalid) 385 * \req_idx index of current desc, will be set to the next 386 * index. If desc_table != NULL the req_idx is the the vring index 387 * or the req_idx is the desc_table index. 388 * \param desc_table descriptor table 389 * \param desc_table_size size of the *desc_table* 390 * \return 0 on success, -1 if given index is invalid 391 * The *desc* param will be set regardless of the 392 * return value. 393 */ 394 int vhost_vring_packed_desc_get_next(struct vring_packed_desc **desc, uint16_t *req_idx, 395 struct spdk_vhost_virtqueue *vq, 396 struct vring_packed_desc *desc_table, 397 uint32_t desc_table_size); 398 399 bool vhost_vring_packed_desc_is_wr(struct vring_packed_desc *cur_desc); 400 401 int vhost_vring_packed_desc_to_iov(struct spdk_vhost_session *vsession, struct iovec *iov, 402 uint16_t *iov_index, const struct vring_packed_desc *desc); 403 404 bool vhost_vring_inflight_desc_is_wr(spdk_vhost_inflight_desc *cur_desc); 405 406 int vhost_vring_inflight_desc_to_iov(struct spdk_vhost_session *vsession, struct iovec *iov, 407 uint16_t *iov_index, const spdk_vhost_inflight_desc *desc); 408 409 uint16_t vhost_vring_packed_desc_get_buffer_id(struct spdk_vhost_virtqueue *vq, uint16_t req_idx, 410 uint16_t *num_descs); 411 412 static inline bool __attribute__((always_inline)) 413 vhost_dev_has_feature(struct spdk_vhost_session *vsession, unsigned feature_id) 414 { 415 return vsession->negotiated_features & (1ULL << feature_id); 416 } 417 418 int vhost_dev_register(struct spdk_vhost_dev *vdev, const char *name, const char *mask_str, 419 const struct spdk_vhost_dev_backend *backend, 420 const struct spdk_vhost_user_dev_backend *user_backend); 421 int vhost_dev_unregister(struct spdk_vhost_dev *vdev); 422 423 void vhost_dump_info_json(struct spdk_vhost_dev *vdev, struct spdk_json_write_ctx *w); 424 425 /* 426 * Set vhost session to run in interrupt or poll mode 427 */ 428 void vhost_user_session_set_interrupt_mode(struct spdk_vhost_session *vsession, 429 bool interrupt_mode); 430 431 /* 432 * Memory registration functions used in start/stop device callbacks 433 */ 434 void vhost_session_mem_register(struct rte_vhost_memory *mem); 435 void vhost_session_mem_unregister(struct rte_vhost_memory *mem); 436 437 /* 438 * Call a function for each session of the provided vhost device. 439 * The function will be called one-by-one on each session's thread. 440 * 441 * \param vdev vhost device 442 * \param fn function to call on each session's thread 443 * \param cpl_fn function to be called at the end of the iteration on 444 * the vhost management thread. 445 * Optional, can be NULL. 446 * \param arg additional argument to the both callbacks 447 */ 448 void vhost_user_dev_foreach_session(struct spdk_vhost_dev *dev, 449 spdk_vhost_session_fn fn, 450 spdk_vhost_dev_fn cpl_fn, 451 void *arg); 452 453 /** 454 * Call a function on the provided lcore and block until either 455 * vhost_user_session_start_done() or vhost_user_session_stop_done() 456 * is called. 457 * 458 * This must be called under the global vhost mutex, which this function 459 * will unlock for the time it's waiting. It's meant to be called only 460 * from start/stop session callbacks. 461 * 462 * \param vsession vhost session 463 * \param cb_fn the function to call. The void *arg parameter in cb_fn 464 * is always NULL. 465 * \param timeout_sec timeout in seconds. This function will still 466 * block after the timeout expires, but will print the provided errmsg. 467 * \param errmsg error message to print once the timeout expires 468 * \return return the code passed to spdk_vhost_session_event_done(). 469 */ 470 int vhost_user_session_send_event(struct spdk_vhost_session *vsession, 471 spdk_vhost_session_fn cb_fn, unsigned timeout_sec, 472 const char *errmsg); 473 474 /** 475 * Finish a blocking spdk_vhost_user_session_send_event() call and finally 476 * start the session. This must be called on the target lcore, which 477 * will now receive all session-related messages (e.g. from 478 * vhost_user_dev_foreach_session()). 479 * 480 * Must be called under the global vhost lock. 481 * 482 * \param vsession vhost session 483 * \param response return code 484 */ 485 void vhost_user_session_start_done(struct spdk_vhost_session *vsession, int response); 486 487 /** 488 * Finish a blocking spdk_vhost_user_session_send_event() call and finally 489 * stop the session. This must be called on the session's lcore which 490 * used to receive all session-related messages (e.g. from 491 * vhost_user_dev_foreach_session()). After this call, the session- 492 * related messages will be once again processed by any arbitrary thread. 493 * 494 * Must be called under the global vhost lock. 495 * 496 * Must be called under the global vhost mutex. 497 * 498 * \param vsession vhost session 499 * \param response return code 500 */ 501 void vhost_user_session_stop_done(struct spdk_vhost_session *vsession, int response); 502 503 struct spdk_vhost_session *vhost_session_find_by_vid(int vid); 504 void vhost_session_install_rte_compat_hooks(struct spdk_vhost_session *vsession); 505 int vhost_register_unix_socket(const char *path, const char *ctrl_name, 506 uint64_t virtio_features, uint64_t disabled_features, uint64_t protocol_features); 507 int vhost_driver_unregister(const char *path); 508 int vhost_get_mem_table(int vid, struct rte_vhost_memory **mem); 509 int vhost_get_negotiated_features(int vid, uint64_t *negotiated_features); 510 511 int remove_vhost_controller(struct spdk_vhost_dev *vdev); 512 513 struct spdk_io_channel *vhost_blk_get_io_channel(struct spdk_vhost_dev *vdev); 514 void vhost_blk_put_io_channel(struct spdk_io_channel *ch); 515 516 /* Function calls from vhost.c to rte_vhost_user.c, 517 * shall removed once virtio transport abstraction is complete. */ 518 int vhost_user_session_set_coalescing(struct spdk_vhost_dev *dev, 519 struct spdk_vhost_session *vsession, void *ctx); 520 int vhost_user_dev_set_coalescing(struct spdk_vhost_user_dev *user_dev, uint32_t delay_base_us, 521 uint32_t iops_threshold); 522 int vhost_user_dev_register(struct spdk_vhost_dev *vdev, const char *name, 523 struct spdk_cpuset *cpumask, const struct spdk_vhost_user_dev_backend *user_backend); 524 int vhost_user_dev_unregister(struct spdk_vhost_dev *vdev); 525 int vhost_user_init(void); 526 typedef void (*vhost_fini_cb)(void *ctx); 527 void vhost_user_fini(vhost_fini_cb vhost_cb); 528 529 #endif /* SPDK_VHOST_INTERNAL_H */ 530