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_mi_hdl_t f_mapped; 64 }; 65 66 struct rdsv3_ib_incoming { 67 struct list ii_frags; 68 struct rdsv3_incoming ii_inc; 69 }; 70 71 struct rdsv3_ib_connect_private { 72 /* Add new fields at the end, and don't permute existing fields. */ 73 uint32_be_t dp_saddr; 74 uint32_be_t dp_daddr; 75 uint8_t dp_protocol_major; 76 uint8_t dp_protocol_minor; 77 uint16_be_t dp_protocol_minor_mask; /* bitmask */ 78 uint32_be_t dp_reserved1; 79 uint32_be_t dp_ack_seq; 80 uint32_be_t dp_credit; /* non-zero enables flow ctl */ 81 }; 82 83 struct rdsv3_ib_send_work { 84 struct rdsv3_message *s_rm; 85 struct rdsv3_rdma_op *s_op; 86 ibt_wrc_opcode_t s_opcode; 87 unsigned long s_queued; 88 }; 89 90 struct rdsv3_ib_recv_work { 91 struct rdsv3_ib_incoming *r_ibinc; 92 struct rdsv3_page_frag *r_frag; 93 ibt_all_wr_t r_wr; 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 struct rdsv3_ib_device *rds_ibdev; 111 struct rdsv3_connection *conn; 112 113 /* alphabet soup, IBTA style */ 114 struct rdma_cm_id *i_cm_id; 115 struct ib_pd *i_pd; 116 struct rdsv3_hdrs_mr *i_mr; 117 struct ib_cq *i_send_cq; 118 struct ib_cq *i_recv_cq; 119 120 /* tx */ 121 struct rdsv3_ib_work_ring i_send_ring; 122 struct rdsv3_message *i_rm; 123 struct rdsv3_header *i_send_hdrs; 124 uint64_t i_send_hdrs_dma; 125 struct rdsv3_ib_send_work *i_sends; 126 ibt_send_wr_t *i_send_wrs; 127 128 /* rx */ 129 ddi_taskq_t *i_recv_tasklet; 130 struct mutex i_recv_mutex; 131 struct rdsv3_ib_work_ring i_recv_ring; 132 struct rdsv3_ib_incoming *i_ibinc; 133 uint32_t i_recv_data_rem; 134 struct rdsv3_header *i_recv_hdrs; 135 uint64_t i_recv_hdrs_dma; 136 struct rdsv3_ib_recv_work *i_recvs; 137 struct rdsv3_page_frag i_frag; 138 uint64_t i_ack_recv; /* last ACK received */ 139 processorid_t i_recv_tasklet_cpuid; 140 /* CPU to which the tasklet taskq should be bound */ 141 142 /* sending acks */ 143 unsigned long i_ack_flags; 144 #ifndef KERNEL_HAS_ATOMIC64 145 kmutex_t i_ack_lock; /* protect i_ack_next */ 146 uint64_t i_ack_next; /* next ACK to send */ 147 #else 148 atomic64_t i_ack_next; /* next ACK to send */ 149 #endif 150 struct rdsv3_header *i_ack; 151 ibt_send_wr_t i_ack_wr; 152 ibt_wr_ds_t i_ack_sge; 153 uint64_t i_ack_dma; 154 unsigned long i_ack_queued; 155 156 /* 157 * Flow control related information 158 * 159 * Our algorithm uses a pair variables that we need to access 160 * atomically - one for the send credits, and one posted 161 * recv credits we need to transfer to remote. 162 * Rather than protect them using a slow spinlock, we put both into 163 * a single atomic_t and update it using cmpxchg 164 */ 165 atomic_t i_credits; 166 167 /* Protocol version specific information */ 168 unsigned int i_flowctl:1; /* enable/disable flow ctl */ 169 170 /* Batched completions */ 171 unsigned int i_unsignaled_wrs; 172 long i_unsignaled_bytes; 173 }; 174 175 /* This assumes that atomic_t is at least 32 bits */ 176 #define IB_GET_SEND_CREDITS(v) ((v) & 0xffff) 177 #define IB_GET_POST_CREDITS(v) ((v) >> 16) 178 #define IB_SET_SEND_CREDITS(v) ((v) & 0xffff) 179 #define IB_SET_POST_CREDITS(v) ((v) << 16) 180 181 struct rdsv3_ib_ipaddr { 182 struct list_node list; 183 uint32_be_t ipaddr; 184 }; 185 186 struct rdsv3_ib_device { 187 struct list_node list; 188 struct list ipaddr_list; 189 struct list conn_list; 190 ib_device_t *dev; 191 struct ib_pd *pd; 192 ibt_lkey_t local_dma_lkey; 193 struct rds_ib_mr_pool *mr_pool; 194 unsigned int fmr_max_remaps; 195 unsigned int max_fmrs; 196 unsigned int fmr_message_size; 197 int max_sge; 198 unsigned int max_wrs; 199 ibt_fmr_pool_hdl_t fmr_pool_hdl; 200 kmutex_t spinlock; /* protect the above */ 201 ibt_hca_attr_t hca_attr; 202 }; 203 204 /* bits for i_ack_flags */ 205 #define IB_ACK_IN_FLIGHT 0 206 #define IB_ACK_REQUESTED 1 207 208 /* Magic WR_ID for ACKs */ 209 #define RDSV3_IB_ACK_WR_ID (~(uint64_t)0) 210 211 struct rdsv3_ib_statistics { 212 uint64_t s_ib_connect_raced; 213 uint64_t s_ib_listen_closed_stale; 214 uint64_t s_ib_tx_cq_call; 215 uint64_t s_ib_tx_cq_event; 216 uint64_t s_ib_tx_ring_full; 217 uint64_t s_ib_tx_throttle; 218 uint64_t s_ib_tx_sg_mapping_failure; 219 uint64_t s_ib_tx_stalled; 220 uint64_t s_ib_tx_credit_updates; 221 uint64_t s_ib_rx_cq_call; 222 uint64_t s_ib_rx_cq_event; 223 uint64_t s_ib_rx_ring_empty; 224 uint64_t s_ib_rx_refill_from_cq; 225 uint64_t s_ib_rx_refill_from_thread; 226 uint64_t s_ib_rx_alloc_limit; 227 uint64_t s_ib_rx_credit_updates; 228 uint64_t s_ib_ack_sent; 229 uint64_t s_ib_ack_send_failure; 230 uint64_t s_ib_ack_send_delayed; 231 uint64_t s_ib_ack_send_piggybacked; 232 uint64_t s_ib_ack_received; 233 uint64_t s_ib_rdma_mr_alloc; 234 uint64_t s_ib_rdma_mr_free; 235 uint64_t s_ib_rdma_mr_used; 236 uint64_t s_ib_rdma_mr_pool_flush; 237 uint64_t s_ib_rdma_mr_pool_wait; 238 uint64_t s_ib_rdma_mr_pool_depleted; 239 }; 240 241 extern struct rdsv3_workqueue_struct_s *rds_ib_wq; 242 243 /* ib.c */ 244 extern struct rdsv3_transport rdsv3_ib_transport; 245 extern void rdsv3_ib_add_one(ib_device_t *device); 246 extern void rdsv3_ib_remove_one(ib_device_t *device); 247 extern struct ib_client rdsv3_ib_client; 248 249 extern unsigned int fmr_pool_size; 250 extern unsigned int fmr_message_size; 251 extern unsigned int rdsv3_ib_retry_count; 252 253 extern kmutex_t ib_nodev_conns_lock; 254 extern struct list ib_nodev_conns; 255 256 /* ib_cm.c */ 257 int rdsv3_ib_conn_alloc(struct rdsv3_connection *conn, int gfp); 258 void rdsv3_ib_conn_free(void *arg); 259 int rdsv3_ib_conn_connect(struct rdsv3_connection *conn); 260 void rdsv3_ib_conn_shutdown(struct rdsv3_connection *conn); 261 void rdsv3_conn_drop(struct rdsv3_connection *conn); 262 int rdsv3_ib_cm_handle_connect(struct rdma_cm_id *cm_id, 263 struct rdma_cm_event *event); 264 int rdsv3_ib_cm_initiate_connect(struct rdma_cm_id *cm_id); 265 void rdsv3_ib_cm_connect_complete(struct rdsv3_connection *conn, 266 struct rdma_cm_event *event); 267 268 /* ib_rdma.c */ 269 int rdsv3_ib_update_ipaddr(struct rdsv3_ib_device *rds_ibdev, 270 uint32_be_t ipaddr); 271 void rdsv3_ib_add_conn(struct rdsv3_ib_device *rds_ibdev, 272 struct rdsv3_connection *conn); 273 void rdsv3_ib_remove_conn(struct rdsv3_ib_device *rds_ibdev, 274 struct rdsv3_connection *conn); 275 void __rdsv3_ib_destroy_conns(struct list *list, kmutex_t *list_lock); 276 static inline void rdsv3_ib_destroy_nodev_conns(void) 277 { 278 __rdsv3_ib_destroy_conns(&ib_nodev_conns, &ib_nodev_conns_lock); 279 } 280 static inline void rdsv3_ib_destroy_conns(struct rdsv3_ib_device *rds_ibdev) 281 { 282 __rdsv3_ib_destroy_conns(&rds_ibdev->conn_list, &rds_ibdev->spinlock); 283 } 284 285 int rdsv3_ib_create_mr_pool(struct rdsv3_ib_device *); 286 void rdsv3_ib_destroy_mr_pool(struct rdsv3_ib_device *); 287 void rdsv3_ib_get_mr_info(struct rdsv3_ib_device *rds_ibdev, 288 struct rdsv3_info_rdma_connection *iinfo); 289 void *rdsv3_ib_get_mr(struct rdsv3_iovec *args, unsigned long nents, 290 struct rdsv3_sock *rs, uint32_t *key_ret); 291 void rdsv3_ib_sync_mr(void *trans_private, int dir); 292 void rdsv3_ib_free_mr(void *trans_private, int invalidate); 293 void rdsv3_ib_flush_mrs(void); 294 295 /* ib_recv.c */ 296 int rdsv3_ib_recv_init(void); 297 void rdsv3_ib_recv_exit(void); 298 int rdsv3_ib_recv(struct rdsv3_connection *conn); 299 int rdsv3_ib_recv_refill(struct rdsv3_connection *conn, int kptr_gfp, 300 int page_gfp, int prefill); 301 void rdsv3_ib_inc_purge(struct rdsv3_incoming *inc); 302 void rdsv3_ib_inc_free(struct rdsv3_incoming *inc); 303 int rdsv3_ib_inc_copy_to_user(struct rdsv3_incoming *inc, uio_t *uiop, 304 size_t size); 305 void rdsv3_ib_recv_cq_comp_handler(struct ib_cq *cq, void *context); 306 void rdsv3_ib_recv_tasklet_fn(void *data); 307 void rdsv3_ib_recv_init_ring(struct rdsv3_ib_connection *ic); 308 void rdsv3_ib_recv_clear_ring(struct rdsv3_ib_connection *ic); 309 void rdsv3_ib_recv_init_ack(struct rdsv3_ib_connection *ic); 310 void rdsv3_ib_attempt_ack(struct rdsv3_ib_connection *ic); 311 void rdsv3_ib_ack_send_complete(struct rdsv3_ib_connection *ic); 312 uint64_t rdsv3_ib_piggyb_ack(struct rdsv3_ib_connection *ic); 313 314 /* ib_ring.c */ 315 void rdsv3_ib_ring_init(struct rdsv3_ib_work_ring *ring, uint32_t nr); 316 void rdsv3_ib_ring_resize(struct rdsv3_ib_work_ring *ring, uint32_t nr); 317 uint32_t rdsv3_ib_ring_alloc(struct rdsv3_ib_work_ring *ring, uint32_t val, 318 uint32_t *pos); 319 void rdsv3_ib_ring_free(struct rdsv3_ib_work_ring *ring, uint32_t val); 320 void rdsv3_ib_ring_unalloc(struct rdsv3_ib_work_ring *ring, uint32_t val); 321 int rdsv3_ib_ring_empty(struct rdsv3_ib_work_ring *ring); 322 int rdsv3_ib_ring_low(struct rdsv3_ib_work_ring *ring); 323 uint32_t rdsv3_ib_ring_oldest(struct rdsv3_ib_work_ring *ring); 324 uint32_t rdsv3_ib_ring_completed(struct rdsv3_ib_work_ring *ring, 325 uint32_t wr_id, uint32_t oldest); 326 extern rdsv3_wait_queue_t rdsv3_ib_ring_empty_wait; 327 328 /* ib_send.c */ 329 void rdsv3_ib_xmit_complete(struct rdsv3_connection *conn); 330 int rdsv3_ib_xmit(struct rdsv3_connection *conn, struct rdsv3_message *rm, 331 unsigned int hdr_off, unsigned int sg, unsigned int off); 332 void rdsv3_ib_send_cq_comp_handler(struct ib_cq *cq, void *context); 333 void rdsv3_ib_send_init_ring(struct rdsv3_ib_connection *ic); 334 void rdsv3_ib_send_clear_ring(struct rdsv3_ib_connection *ic); 335 int rdsv3_ib_xmit_rdma(struct rdsv3_connection *conn, struct rdsv3_rdma_op *op); 336 void rdsv3_ib_send_add_credits(struct rdsv3_connection *conn, 337 unsigned int credits); 338 void rdsv3_ib_advertise_credits(struct rdsv3_connection *conn, 339 unsigned int posted); 340 int rdsv3_ib_send_grab_credits(struct rdsv3_ib_connection *ic, uint32_t wanted, 341 uint32_t *adv_credits, int need_posted, int max_posted); 342 343 /* ib_stats.c */ 344 RDSV3_DECLARE_PER_CPU(struct rdsv3_ib_statistics, rdsv3_ib_stats); 345 #define rdsv3_ib_stats_inc(member) rdsv3_stats_inc_which(rdsv3_ib_stats, member) 346 unsigned int rdsv3_ib_stats_info_copy(struct rdsv3_info_iterator *iter, 347 unsigned int avail); 348 349 /* ib_sysctl.c */ 350 int rdsv3_ib_sysctl_init(void); 351 void rdsv3_ib_sysctl_exit(void); 352 extern unsigned long rdsv3_ib_sysctl_max_send_wr; 353 extern unsigned long rdsv3_ib_sysctl_max_recv_wr; 354 extern unsigned long rdsv3_ib_sysctl_max_unsig_wrs; 355 extern unsigned long rdsv3_ib_sysctl_max_unsig_bytes; 356 extern unsigned long rdsv3_ib_sysctl_max_recv_allocation; 357 extern unsigned int rdsv3_ib_sysctl_flow_control; 358 359 #endif /* _RDSV3_IB_H */ 360