1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 /* 22 * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved. 23 */ 24 25 #ifndef _RDSV3_IB_H 26 #define _RDSV3_IB_H 27 28 #include <sys/rds.h> 29 30 #include <sys/ib/clients/rdsv3/rdsv3.h> 31 #include <sys/ib/clients/rdsv3/rdma_transport.h> 32 33 #define RDSV3_FMR_SIZE 256 34 #define RDSV3_FMR_POOL_SIZE (12 * 1024) 35 36 #define RDSV3_IB_SEND_WRS 64 37 38 #define RDSV3_IB_MAX_SGE 8 39 #define RDSV3_IB_RECV_SGE 2 40 41 #define RDSV3_IB_DEFAULT_RECV_WR 1024 42 #define RDSV3_IB_DEFAULT_SEND_WR 256 43 44 #define RDSV3_IB_DEFAULT_RETRY_COUNT 2 45 46 /* minor versions supported */ 47 #define RDSV3_IB_SUPPORTED_PROTOCOLS 0x00000003 48 49 extern struct list rdsv3_ib_devices; 50 51 /* 52 * IB posts RDSV3_FRAG_SIZE fragments of pages to the receive queues to 53 * try and minimize the amount of memory tied up both the device and 54 * socket receive queues. 55 */ 56 /* page offset of the final full frag that fits in the page */ 57 #define RDSV3_PAGE_LAST_OFF \ 58 (((PAGE_SIZE / RDSV3_FRAG_SIZE) - 1) * RDSV3_FRAG_SIZE) 59 struct rdsv3_page_frag { 60 struct list_node f_item; 61 caddr_t f_page; 62 unsigned long f_offset; 63 ibt_wr_ds_t f_sge; 64 ibt_mi_hdl_t f_mapped; 65 }; 66 67 struct rdsv3_ib_incoming { 68 struct list ii_frags; 69 struct rdsv3_incoming ii_inc; 70 }; 71 72 struct rdsv3_ib_connect_private { 73 /* Add new fields at the end, and don't permute existing fields. */ 74 uint32_be_t dp_saddr; 75 uint32_be_t dp_daddr; 76 uint8_t dp_protocol_major; 77 uint8_t dp_protocol_minor; 78 uint16_be_t dp_protocol_minor_mask; /* bitmask */ 79 uint32_be_t dp_reserved1; 80 uint32_be_t dp_ack_seq; 81 uint32_be_t dp_credit; /* non-zero enables flow ctl */ 82 }; 83 84 struct rdsv3_ib_send_work { 85 struct rdsv3_message *s_rm; 86 struct rdsv3_rdma_op *s_op; 87 ibt_wrc_opcode_t s_opcode; 88 unsigned long s_queued; 89 }; 90 91 struct rdsv3_ib_recv_work { 92 struct rdsv3_ib_incoming *r_ibinc; 93 struct rdsv3_page_frag *r_frag; 94 ibt_wr_ds_t r_sge[2]; 95 }; 96 97 struct rdsv3_ib_work_ring { 98 uint32_t w_nr; 99 uint32_t w_alloc_ptr; 100 uint32_t w_alloc_ctr; 101 uint32_t w_free_ptr; 102 atomic_t w_free_ctr; 103 }; 104 105 struct rdsv3_ib_device; 106 107 struct rdsv3_ib_connection { 108 109 struct list_node ib_node; 110 boolean_t i_on_dev_list; 111 struct rdsv3_ib_device *rds_ibdev; 112 struct rdsv3_connection *conn; 113 114 /* alphabet soup, IBTA style */ 115 struct rdma_cm_id *i_cm_id; 116 struct ib_pd *i_pd; 117 struct rdsv3_hdrs_mr *i_mr; 118 struct ib_cq *i_send_cq; 119 struct ib_cq *i_recv_cq; 120 121 /* tx */ 122 struct rdsv3_ib_work_ring i_send_ring; 123 struct rdsv3_message *i_rm; 124 struct rdsv3_header *i_send_hdrs; 125 uint64_t i_send_hdrs_dma; 126 struct rdsv3_ib_send_work *i_sends; 127 ibt_send_wr_t *i_send_wrs; 128 129 /* rx */ 130 ddi_taskq_t *i_recv_tasklet; 131 struct mutex i_recv_mutex; 132 struct rdsv3_ib_work_ring i_recv_ring; 133 struct rdsv3_ib_incoming *i_ibinc; 134 uint32_t i_recv_data_rem; 135 struct rdsv3_header *i_recv_hdrs; 136 uint64_t i_recv_hdrs_dma; 137 struct rdsv3_ib_recv_work *i_recvs; 138 ibt_recv_wr_t *i_recv_wrs; 139 struct rdsv3_page_frag i_frag; 140 uint64_t i_ack_recv; /* last ACK received */ 141 processorid_t i_recv_tasklet_cpuid; 142 /* CPU to which the tasklet taskq should be bound */ 143 144 /* sending acks */ 145 unsigned long i_ack_flags; 146 #ifdef KERNEL_HAS_ATOMIC64 147 atomic64_t i_ack_next; /* next ACK to send */ 148 #else 149 kmutex_t i_ack_lock; /* protect i_ack_next */ 150 uint64_t i_ack_next; /* next ACK to send */ 151 #endif 152 struct rdsv3_header *i_ack; 153 ibt_send_wr_t i_ack_wr; 154 ibt_wr_ds_t i_ack_sge; 155 uint64_t i_ack_dma; 156 unsigned long i_ack_queued; 157 158 /* 159 * Flow control related information 160 * 161 * Our algorithm uses a pair variables that we need to access 162 * atomically - one for the send credits, and one posted 163 * recv credits we need to transfer to remote. 164 * Rather than protect them using a slow spinlock, we put both into 165 * a single atomic_t and update it using cmpxchg 166 */ 167 atomic_t i_credits; 168 169 /* Protocol version specific information */ 170 unsigned int i_flowctl:1; /* enable/disable flow ctl */ 171 172 /* Batched completions */ 173 unsigned int i_unsignaled_wrs; 174 long i_unsignaled_bytes; 175 }; 176 177 /* This assumes that atomic_t is at least 32 bits */ 178 #define IB_GET_SEND_CREDITS(v) ((v) & 0xffff) 179 #define IB_GET_POST_CREDITS(v) ((v) >> 16) 180 #define IB_SET_SEND_CREDITS(v) ((v) & 0xffff) 181 #define IB_SET_POST_CREDITS(v) ((v) << 16) 182 183 struct rdsv3_ib_ipaddr { 184 struct list_node list; 185 uint32_be_t ipaddr; 186 }; 187 188 struct rdsv3_ib_device { 189 struct list_node list; 190 struct list ipaddr_list; 191 struct list conn_list; 192 ib_device_t *dev; 193 struct ib_pd *pd; 194 struct kmem_cache *ib_frag_slab; 195 struct rds_ib_mr_pool *mr_pool; 196 unsigned int fmr_max_remaps; 197 unsigned int max_fmrs; 198 unsigned int fmr_message_size; 199 int max_sge; 200 unsigned int max_wrs; 201 ibt_fmr_pool_hdl_t fmr_pool_hdl; 202 kmutex_t spinlock; /* protect the above */ 203 ibt_hca_attr_t hca_attr; 204 }; 205 206 /* bits for i_ack_flags */ 207 #define IB_ACK_IN_FLIGHT 0 208 #define IB_ACK_REQUESTED 1 209 210 /* Magic WR_ID for ACKs */ 211 #define RDSV3_IB_ACK_WR_ID (~(uint64_t)0) 212 213 struct rdsv3_ib_statistics { 214 uint64_t s_ib_connect_raced; 215 uint64_t s_ib_listen_closed_stale; 216 uint64_t s_ib_tx_cq_call; 217 uint64_t s_ib_tx_cq_event; 218 uint64_t s_ib_tx_ring_full; 219 uint64_t s_ib_tx_throttle; 220 uint64_t s_ib_tx_sg_mapping_failure; 221 uint64_t s_ib_tx_stalled; 222 uint64_t s_ib_tx_credit_updates; 223 uint64_t s_ib_rx_cq_call; 224 uint64_t s_ib_rx_cq_event; 225 uint64_t s_ib_rx_ring_empty; 226 uint64_t s_ib_rx_refill_from_cq; 227 uint64_t s_ib_rx_refill_from_thread; 228 uint64_t s_ib_rx_alloc_limit; 229 uint64_t s_ib_rx_credit_updates; 230 uint64_t s_ib_ack_sent; 231 uint64_t s_ib_ack_send_failure; 232 uint64_t s_ib_ack_send_delayed; 233 uint64_t s_ib_ack_send_piggybacked; 234 uint64_t s_ib_ack_received; 235 uint64_t s_ib_rdma_mr_alloc; 236 uint64_t s_ib_rdma_mr_free; 237 uint64_t s_ib_rdma_mr_used; 238 uint64_t s_ib_rdma_mr_pool_flush; 239 uint64_t s_ib_rdma_mr_pool_wait; 240 uint64_t s_ib_rdma_mr_pool_depleted; 241 }; 242 243 extern struct rdsv3_workqueue_struct_s *rds_ib_wq; 244 245 /* ib.c */ 246 extern struct rdsv3_transport rdsv3_ib_transport; 247 extern void rdsv3_ib_add_one(ib_device_t *device); 248 extern void rdsv3_ib_remove_one(ib_device_t *device); 249 extern struct ib_client rdsv3_ib_client; 250 251 extern unsigned int fmr_pool_size; 252 extern unsigned int fmr_message_size; 253 extern unsigned int rdsv3_ib_retry_count; 254 255 extern kmutex_t ib_nodev_conns_lock; 256 extern struct list ib_nodev_conns; 257 258 /* ib_cm.c */ 259 int rdsv3_ib_conn_alloc(struct rdsv3_connection *conn, int gfp); 260 void rdsv3_ib_conn_free(void *arg); 261 int rdsv3_ib_conn_connect(struct rdsv3_connection *conn); 262 void rdsv3_ib_conn_shutdown(struct rdsv3_connection *conn); 263 void rdsv3_conn_drop(struct rdsv3_connection *conn); 264 int rdsv3_ib_cm_handle_connect(struct rdma_cm_id *cm_id, 265 struct rdma_cm_event *event); 266 int rdsv3_ib_cm_initiate_connect(struct rdma_cm_id *cm_id); 267 void rdsv3_ib_cm_connect_complete(struct rdsv3_connection *conn, 268 struct rdma_cm_event *event); 269 270 /* ib_rdma.c */ 271 int rdsv3_ib_update_ipaddr(struct rdsv3_ib_device *rds_ibdev, 272 uint32_be_t ipaddr); 273 void rdsv3_ib_add_conn(struct rdsv3_ib_device *rds_ibdev, 274 struct rdsv3_connection *conn); 275 void rdsv3_ib_remove_conn(struct rdsv3_ib_device *rds_ibdev, 276 struct rdsv3_connection *conn); 277 void __rdsv3_ib_destroy_conns(struct list *list, kmutex_t *list_lock); 278 static inline void rdsv3_ib_destroy_nodev_conns(void) 279 { 280 __rdsv3_ib_destroy_conns(&ib_nodev_conns, &ib_nodev_conns_lock); 281 } 282 static inline void rdsv3_ib_destroy_conns(struct rdsv3_ib_device *rds_ibdev) 283 { 284 __rdsv3_ib_destroy_conns(&rds_ibdev->conn_list, &rds_ibdev->spinlock); 285 } 286 287 int rdsv3_ib_create_mr_pool(struct rdsv3_ib_device *); 288 void rdsv3_ib_destroy_mr_pool(struct rdsv3_ib_device *); 289 void rdsv3_ib_get_mr_info(struct rdsv3_ib_device *rds_ibdev, 290 struct rdsv3_info_rdma_connection *iinfo); 291 void *rdsv3_ib_get_mr(struct rdsv3_iovec *args, unsigned long nents, 292 struct rdsv3_sock *rs, uint32_t *key_ret); 293 void rdsv3_ib_sync_mr(void *trans_private, int dir); 294 void rdsv3_ib_free_mr(void *trans_private, int invalidate); 295 void rdsv3_ib_flush_mrs(void); 296 297 /* ib_recv.c */ 298 int rdsv3_ib_recv_init(void); 299 void rdsv3_ib_recv_exit(void); 300 int rdsv3_ib_recv(struct rdsv3_connection *conn); 301 int rdsv3_ib_recv_refill(struct rdsv3_connection *conn, int kmflags, 302 int prefill); 303 void rdsv3_ib_inc_purge(struct rdsv3_incoming *inc); 304 void rdsv3_ib_inc_free(struct rdsv3_incoming *inc); 305 int rdsv3_ib_inc_copy_to_user(struct rdsv3_incoming *inc, uio_t *uiop, 306 size_t size); 307 void rdsv3_ib_recv_cq_comp_handler(struct ib_cq *cq, void *context); 308 void rdsv3_ib_recv_tasklet_fn(void *data); 309 void rdsv3_ib_recv_init_ring(struct rdsv3_ib_connection *ic); 310 void rdsv3_ib_recv_clear_ring(struct rdsv3_ib_connection *ic); 311 void rdsv3_ib_recv_init_ack(struct rdsv3_ib_connection *ic); 312 void rdsv3_ib_attempt_ack(struct rdsv3_ib_connection *ic); 313 void rdsv3_ib_ack_send_complete(struct rdsv3_ib_connection *ic); 314 uint64_t rdsv3_ib_piggyb_ack(struct rdsv3_ib_connection *ic); 315 316 /* ib_ring.c */ 317 void rdsv3_ib_ring_init(struct rdsv3_ib_work_ring *ring, uint32_t nr); 318 void rdsv3_ib_ring_resize(struct rdsv3_ib_work_ring *ring, uint32_t nr); 319 uint32_t rdsv3_ib_ring_alloc(struct rdsv3_ib_work_ring *ring, uint32_t val, 320 uint32_t *pos); 321 void rdsv3_ib_ring_free(struct rdsv3_ib_work_ring *ring, uint32_t val); 322 void rdsv3_ib_ring_unalloc(struct rdsv3_ib_work_ring *ring, uint32_t val); 323 int rdsv3_ib_ring_empty(struct rdsv3_ib_work_ring *ring); 324 int rdsv3_ib_ring_low(struct rdsv3_ib_work_ring *ring); 325 uint32_t rdsv3_ib_ring_oldest(struct rdsv3_ib_work_ring *ring); 326 uint32_t rdsv3_ib_ring_completed(struct rdsv3_ib_work_ring *ring, 327 uint32_t wr_id, uint32_t oldest); 328 extern rdsv3_wait_queue_t rdsv3_ib_ring_empty_wait; 329 330 /* ib_send.c */ 331 void rdsv3_ib_xmit_complete(struct rdsv3_connection *conn); 332 int rdsv3_ib_xmit(struct rdsv3_connection *conn, struct rdsv3_message *rm, 333 unsigned int hdr_off, unsigned int sg, unsigned int off); 334 void rdsv3_ib_send_cq_comp_handler(struct ib_cq *cq, void *context); 335 void rdsv3_ib_send_init_ring(struct rdsv3_ib_connection *ic); 336 void rdsv3_ib_send_clear_ring(struct rdsv3_ib_connection *ic); 337 int rdsv3_ib_xmit_rdma(struct rdsv3_connection *conn, struct rdsv3_rdma_op *op); 338 void rdsv3_ib_send_add_credits(struct rdsv3_connection *conn, 339 unsigned int credits); 340 void rdsv3_ib_advertise_credits(struct rdsv3_connection *conn, 341 unsigned int posted); 342 int rdsv3_ib_send_grab_credits(struct rdsv3_ib_connection *ic, uint32_t wanted, 343 uint32_t *adv_credits, int need_posted); 344 345 /* ib_stats.c */ 346 RDSV3_DECLARE_PER_CPU(struct rdsv3_ib_statistics, rdsv3_ib_stats); 347 #define rdsv3_ib_stats_inc(member) rdsv3_stats_inc_which(rdsv3_ib_stats, member) 348 unsigned int rdsv3_ib_stats_info_copy(struct rdsv3_info_iterator *iter, 349 unsigned int avail); 350 351 /* ib_sysctl.c */ 352 int rdsv3_ib_sysctl_init(void); 353 void rdsv3_ib_sysctl_exit(void); 354 extern unsigned long rdsv3_ib_sysctl_max_send_wr; 355 extern unsigned long rdsv3_ib_sysctl_max_recv_wr; 356 extern unsigned long rdsv3_ib_sysctl_max_unsig_wrs; 357 extern unsigned long rdsv3_ib_sysctl_max_unsig_bytes; 358 extern unsigned long rdsv3_ib_sysctl_max_recv_allocation; 359 extern unsigned int rdsv3_ib_sysctl_flow_control; 360 361 #endif /* _RDSV3_IB_H */ 362