1 /* 2 * Copyright (c) 2008 Oracle. All rights reserved. 3 * 4 * This software is available to you under a choice of one of two 5 * licenses. You may choose to be licensed under the terms of the GNU 6 * General Public License (GPL) Version 2, available from the file 7 * COPYING in the main directory of this source tree, or the 8 * OpenIB.org BSD license below: 9 * 10 * Redistribution and use in source and binary forms, with or 11 * without modification, are permitted provided that the following 12 * conditions are met: 13 * 14 * - Redistributions of source code must retain the above 15 * copyright notice, this list of conditions and the following 16 * disclaimer. 17 * 18 * - Redistributions in binary form must reproduce the above 19 * copyright notice, this list of conditions and the following 20 * disclaimer in the documentation and/or other materials 21 * provided with the distribution. 22 * 23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS 27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN 28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN 29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 30 * SOFTWARE. 31 * 32 */ 33 34 #ifndef _LINUX_RDS_H 35 #define _LINUX_RDS_H 36 37 #include <linux/types.h> 38 39 #define RDS_IB_ABI_VERSION 0x301 40 41 #define SOL_RDS 276 42 43 /* 44 * setsockopt/getsockopt for SOL_RDS 45 */ 46 #define RDS_CANCEL_SENT_TO 1 47 #define RDS_GET_MR 2 48 #define RDS_FREE_MR 3 49 /* deprecated: RDS_BARRIER 4 */ 50 #define RDS_RECVERR 5 51 #define RDS_CONG_MONITOR 6 52 #define RDS_GET_MR_FOR_DEST 7 53 #define SO_RDS_TRANSPORT 8 54 55 /* Socket option to tap receive path latency 56 * SO_RDS: SO_RDS_MSG_RXPATH_LATENCY 57 * Format used struct rds_rx_trace_so 58 */ 59 #define SO_RDS_MSG_RXPATH_LATENCY 10 60 61 62 /* supported values for SO_RDS_TRANSPORT */ 63 #define RDS_TRANS_IB 0 64 #define RDS_TRANS_IWARP 1 65 #define RDS_TRANS_TCP 2 66 #define RDS_TRANS_COUNT 3 67 #define RDS_TRANS_NONE (~0) 68 69 /* 70 * Control message types for SOL_RDS. 71 * 72 * CMSG_RDMA_ARGS (sendmsg) 73 * Request a RDMA transfer to/from the specified 74 * memory ranges. 75 * The cmsg_data is a struct rds_rdma_args. 76 * RDS_CMSG_RDMA_DEST (recvmsg, sendmsg) 77 * Kernel informs application about intended 78 * source/destination of a RDMA transfer 79 * RDS_CMSG_RDMA_MAP (sendmsg) 80 * Application asks kernel to map the given 81 * memory range into a IB MR, and send the 82 * R_Key along in an RDS extension header. 83 * The cmsg_data is a struct rds_get_mr_args, 84 * the same as for the GET_MR setsockopt. 85 * RDS_CMSG_RDMA_STATUS (recvmsg) 86 * Returns the status of a completed RDMA operation. 87 * RDS_CMSG_RXPATH_LATENCY(recvmsg) 88 * Returns rds message latencies in various stages of receive 89 * path in nS. Its set per socket using SO_RDS_MSG_RXPATH_LATENCY 90 * socket option. Legitimate points are defined in 91 * enum rds_message_rxpath_latency. More points can be added in 92 * future. CSMG format is struct rds_cmsg_rx_trace. 93 */ 94 #define RDS_CMSG_RDMA_ARGS 1 95 #define RDS_CMSG_RDMA_DEST 2 96 #define RDS_CMSG_RDMA_MAP 3 97 #define RDS_CMSG_RDMA_STATUS 4 98 #define RDS_CMSG_CONG_UPDATE 5 99 #define RDS_CMSG_ATOMIC_FADD 6 100 #define RDS_CMSG_ATOMIC_CSWP 7 101 #define RDS_CMSG_MASKED_ATOMIC_FADD 8 102 #define RDS_CMSG_MASKED_ATOMIC_CSWP 9 103 #define RDS_CMSG_RXPATH_LATENCY 11 104 105 #define RDS_INFO_FIRST 10000 106 #define RDS_INFO_COUNTERS 10000 107 #define RDS_INFO_CONNECTIONS 10001 108 /* 10002 aka RDS_INFO_FLOWS is deprecated */ 109 #define RDS_INFO_SEND_MESSAGES 10003 110 #define RDS_INFO_RETRANS_MESSAGES 10004 111 #define RDS_INFO_RECV_MESSAGES 10005 112 #define RDS_INFO_SOCKETS 10006 113 #define RDS_INFO_TCP_SOCKETS 10007 114 #define RDS_INFO_IB_CONNECTIONS 10008 115 #define RDS_INFO_CONNECTION_STATS 10009 116 #define RDS_INFO_IWARP_CONNECTIONS 10010 117 #define RDS_INFO_LAST 10010 118 119 struct rds_info_counter { 120 uint8_t name[32]; 121 uint64_t value; 122 } __attribute__((packed)); 123 124 #define RDS_INFO_CONNECTION_FLAG_SENDING 0x01 125 #define RDS_INFO_CONNECTION_FLAG_CONNECTING 0x02 126 #define RDS_INFO_CONNECTION_FLAG_CONNECTED 0x04 127 128 #define TRANSNAMSIZ 16 129 130 struct rds_info_connection { 131 uint64_t next_tx_seq; 132 uint64_t next_rx_seq; 133 __be32 laddr; 134 __be32 faddr; 135 uint8_t transport[TRANSNAMSIZ]; /* null term ascii */ 136 uint8_t flags; 137 } __attribute__((packed)); 138 139 #define RDS_INFO_MESSAGE_FLAG_ACK 0x01 140 #define RDS_INFO_MESSAGE_FLAG_FAST_ACK 0x02 141 142 struct rds_info_message { 143 uint64_t seq; 144 uint32_t len; 145 __be32 laddr; 146 __be32 faddr; 147 __be16 lport; 148 __be16 fport; 149 uint8_t flags; 150 } __attribute__((packed)); 151 152 struct rds_info_socket { 153 uint32_t sndbuf; 154 __be32 bound_addr; 155 __be32 connected_addr; 156 __be16 bound_port; 157 __be16 connected_port; 158 uint32_t rcvbuf; 159 uint64_t inum; 160 } __attribute__((packed)); 161 162 struct rds_info_tcp_socket { 163 __be32 local_addr; 164 __be16 local_port; 165 __be32 peer_addr; 166 __be16 peer_port; 167 uint64_t hdr_rem; 168 uint64_t data_rem; 169 uint32_t last_sent_nxt; 170 uint32_t last_expected_una; 171 uint32_t last_seen_una; 172 } __attribute__((packed)); 173 174 #define RDS_IB_GID_LEN 16 175 struct rds_info_rdma_connection { 176 __be32 src_addr; 177 __be32 dst_addr; 178 uint8_t src_gid[RDS_IB_GID_LEN]; 179 uint8_t dst_gid[RDS_IB_GID_LEN]; 180 181 uint32_t max_send_wr; 182 uint32_t max_recv_wr; 183 uint32_t max_send_sge; 184 uint32_t rdma_mr_max; 185 uint32_t rdma_mr_size; 186 }; 187 188 /* RDS message Receive Path Latency points */ 189 enum rds_message_rxpath_latency { 190 RDS_MSG_RX_HDR_TO_DGRAM_START = 0, 191 RDS_MSG_RX_DGRAM_REASSEMBLE, 192 RDS_MSG_RX_DGRAM_DELIVERED, 193 RDS_MSG_RX_DGRAM_TRACE_MAX 194 }; 195 196 struct rds_rx_trace_so { 197 u8 rx_traces; 198 u8 rx_trace_pos[RDS_MSG_RX_DGRAM_TRACE_MAX]; 199 }; 200 201 struct rds_cmsg_rx_trace { 202 u8 rx_traces; 203 u8 rx_trace_pos[RDS_MSG_RX_DGRAM_TRACE_MAX]; 204 u64 rx_trace[RDS_MSG_RX_DGRAM_TRACE_MAX]; 205 }; 206 207 /* 208 * Congestion monitoring. 209 * Congestion control in RDS happens at the host connection 210 * level by exchanging a bitmap marking congested ports. 211 * By default, a process sleeping in poll() is always woken 212 * up when the congestion map is updated. 213 * With explicit monitoring, an application can have more 214 * fine-grained control. 215 * The application installs a 64bit mask value in the socket, 216 * where each bit corresponds to a group of ports. 217 * When a congestion update arrives, RDS checks the set of 218 * ports that are now uncongested against the list bit mask 219 * installed in the socket, and if they overlap, we queue a 220 * cong_notification on the socket. 221 * 222 * To install the congestion monitor bitmask, use RDS_CONG_MONITOR 223 * with the 64bit mask. 224 * Congestion updates are received via RDS_CMSG_CONG_UPDATE 225 * control messages. 226 * 227 * The correspondence between bits and ports is 228 * 1 << (portnum % 64) 229 */ 230 #define RDS_CONG_MONITOR_SIZE 64 231 #define RDS_CONG_MONITOR_BIT(port) (((unsigned int) port) % RDS_CONG_MONITOR_SIZE) 232 #define RDS_CONG_MONITOR_MASK(port) (1ULL << RDS_CONG_MONITOR_BIT(port)) 233 234 /* 235 * RDMA related types 236 */ 237 238 /* 239 * This encapsulates a remote memory location. 240 * In the current implementation, it contains the R_Key 241 * of the remote memory region, and the offset into it 242 * (so that the application does not have to worry about 243 * alignment). 244 */ 245 typedef uint64_t rds_rdma_cookie_t; 246 247 struct rds_iovec { 248 uint64_t addr; 249 uint64_t bytes; 250 }; 251 252 struct rds_get_mr_args { 253 struct rds_iovec vec; 254 uint64_t cookie_addr; 255 uint64_t flags; 256 }; 257 258 struct rds_get_mr_for_dest_args { 259 struct sockaddr_storage dest_addr; 260 struct rds_iovec vec; 261 uint64_t cookie_addr; 262 uint64_t flags; 263 }; 264 265 struct rds_free_mr_args { 266 rds_rdma_cookie_t cookie; 267 uint64_t flags; 268 }; 269 270 struct rds_rdma_args { 271 rds_rdma_cookie_t cookie; 272 struct rds_iovec remote_vec; 273 uint64_t local_vec_addr; 274 uint64_t nr_local; 275 uint64_t flags; 276 uint64_t user_token; 277 }; 278 279 struct rds_atomic_args { 280 rds_rdma_cookie_t cookie; 281 uint64_t local_addr; 282 uint64_t remote_addr; 283 union { 284 struct { 285 uint64_t compare; 286 uint64_t swap; 287 } cswp; 288 struct { 289 uint64_t add; 290 } fadd; 291 struct { 292 uint64_t compare; 293 uint64_t swap; 294 uint64_t compare_mask; 295 uint64_t swap_mask; 296 } m_cswp; 297 struct { 298 uint64_t add; 299 uint64_t nocarry_mask; 300 } m_fadd; 301 }; 302 uint64_t flags; 303 uint64_t user_token; 304 }; 305 306 struct rds_rdma_notify { 307 uint64_t user_token; 308 int32_t status; 309 }; 310 311 #define RDS_RDMA_SUCCESS 0 312 #define RDS_RDMA_REMOTE_ERROR 1 313 #define RDS_RDMA_CANCELED 2 314 #define RDS_RDMA_DROPPED 3 315 #define RDS_RDMA_OTHER_ERROR 4 316 317 /* 318 * Common set of flags for all RDMA related structs 319 */ 320 #define RDS_RDMA_READWRITE 0x0001 321 #define RDS_RDMA_FENCE 0x0002 /* use FENCE for immediate send */ 322 #define RDS_RDMA_INVALIDATE 0x0004 /* invalidate R_Key after freeing MR */ 323 #define RDS_RDMA_USE_ONCE 0x0008 /* free MR after use */ 324 #define RDS_RDMA_DONTWAIT 0x0010 /* Don't wait in SET_BARRIER */ 325 #define RDS_RDMA_NOTIFY_ME 0x0020 /* Notify when operation completes */ 326 #define RDS_RDMA_SILENT 0x0040 /* Do not interrupt remote */ 327 328 #endif /* IB_RDS_H */ 329