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 /* supported values for SO_RDS_TRANSPORT */ 56 #define RDS_TRANS_IB 0 57 #define RDS_TRANS_IWARP 1 58 #define RDS_TRANS_TCP 2 59 #define RDS_TRANS_COUNT 3 60 #define RDS_TRANS_NONE (~0) 61 62 /* 63 * Control message types for SOL_RDS. 64 * 65 * CMSG_RDMA_ARGS (sendmsg) 66 * Request a RDMA transfer to/from the specified 67 * memory ranges. 68 * The cmsg_data is a struct rds_rdma_args. 69 * RDS_CMSG_RDMA_DEST (recvmsg, sendmsg) 70 * Kernel informs application about intended 71 * source/destination of a RDMA transfer 72 * RDS_CMSG_RDMA_MAP (sendmsg) 73 * Application asks kernel to map the given 74 * memory range into a IB MR, and send the 75 * R_Key along in an RDS extension header. 76 * The cmsg_data is a struct rds_get_mr_args, 77 * the same as for the GET_MR setsockopt. 78 * RDS_CMSG_RDMA_STATUS (recvmsg) 79 * Returns the status of a completed RDMA operation. 80 */ 81 #define RDS_CMSG_RDMA_ARGS 1 82 #define RDS_CMSG_RDMA_DEST 2 83 #define RDS_CMSG_RDMA_MAP 3 84 #define RDS_CMSG_RDMA_STATUS 4 85 #define RDS_CMSG_CONG_UPDATE 5 86 #define RDS_CMSG_ATOMIC_FADD 6 87 #define RDS_CMSG_ATOMIC_CSWP 7 88 #define RDS_CMSG_MASKED_ATOMIC_FADD 8 89 #define RDS_CMSG_MASKED_ATOMIC_CSWP 9 90 91 #define RDS_INFO_FIRST 10000 92 #define RDS_INFO_COUNTERS 10000 93 #define RDS_INFO_CONNECTIONS 10001 94 /* 10002 aka RDS_INFO_FLOWS is deprecated */ 95 #define RDS_INFO_SEND_MESSAGES 10003 96 #define RDS_INFO_RETRANS_MESSAGES 10004 97 #define RDS_INFO_RECV_MESSAGES 10005 98 #define RDS_INFO_SOCKETS 10006 99 #define RDS_INFO_TCP_SOCKETS 10007 100 #define RDS_INFO_IB_CONNECTIONS 10008 101 #define RDS_INFO_CONNECTION_STATS 10009 102 #define RDS_INFO_IWARP_CONNECTIONS 10010 103 #define RDS_INFO_LAST 10010 104 105 struct rds_info_counter { 106 uint8_t name[32]; 107 uint64_t value; 108 } __attribute__((packed)); 109 110 #define RDS_INFO_CONNECTION_FLAG_SENDING 0x01 111 #define RDS_INFO_CONNECTION_FLAG_CONNECTING 0x02 112 #define RDS_INFO_CONNECTION_FLAG_CONNECTED 0x04 113 114 #define TRANSNAMSIZ 16 115 116 struct rds_info_connection { 117 uint64_t next_tx_seq; 118 uint64_t next_rx_seq; 119 __be32 laddr; 120 __be32 faddr; 121 uint8_t transport[TRANSNAMSIZ]; /* null term ascii */ 122 uint8_t flags; 123 } __attribute__((packed)); 124 125 #define RDS_INFO_MESSAGE_FLAG_ACK 0x01 126 #define RDS_INFO_MESSAGE_FLAG_FAST_ACK 0x02 127 128 struct rds_info_message { 129 uint64_t seq; 130 uint32_t len; 131 __be32 laddr; 132 __be32 faddr; 133 __be16 lport; 134 __be16 fport; 135 uint8_t flags; 136 } __attribute__((packed)); 137 138 struct rds_info_socket { 139 uint32_t sndbuf; 140 __be32 bound_addr; 141 __be32 connected_addr; 142 __be16 bound_port; 143 __be16 connected_port; 144 uint32_t rcvbuf; 145 uint64_t inum; 146 } __attribute__((packed)); 147 148 struct rds_info_tcp_socket { 149 __be32 local_addr; 150 __be16 local_port; 151 __be32 peer_addr; 152 __be16 peer_port; 153 uint64_t hdr_rem; 154 uint64_t data_rem; 155 uint32_t last_sent_nxt; 156 uint32_t last_expected_una; 157 uint32_t last_seen_una; 158 } __attribute__((packed)); 159 160 #define RDS_IB_GID_LEN 16 161 struct rds_info_rdma_connection { 162 __be32 src_addr; 163 __be32 dst_addr; 164 uint8_t src_gid[RDS_IB_GID_LEN]; 165 uint8_t dst_gid[RDS_IB_GID_LEN]; 166 167 uint32_t max_send_wr; 168 uint32_t max_recv_wr; 169 uint32_t max_send_sge; 170 uint32_t rdma_mr_max; 171 uint32_t rdma_mr_size; 172 }; 173 174 /* 175 * Congestion monitoring. 176 * Congestion control in RDS happens at the host connection 177 * level by exchanging a bitmap marking congested ports. 178 * By default, a process sleeping in poll() is always woken 179 * up when the congestion map is updated. 180 * With explicit monitoring, an application can have more 181 * fine-grained control. 182 * The application installs a 64bit mask value in the socket, 183 * where each bit corresponds to a group of ports. 184 * When a congestion update arrives, RDS checks the set of 185 * ports that are now uncongested against the list bit mask 186 * installed in the socket, and if they overlap, we queue a 187 * cong_notification on the socket. 188 * 189 * To install the congestion monitor bitmask, use RDS_CONG_MONITOR 190 * with the 64bit mask. 191 * Congestion updates are received via RDS_CMSG_CONG_UPDATE 192 * control messages. 193 * 194 * The correspondence between bits and ports is 195 * 1 << (portnum % 64) 196 */ 197 #define RDS_CONG_MONITOR_SIZE 64 198 #define RDS_CONG_MONITOR_BIT(port) (((unsigned int) port) % RDS_CONG_MONITOR_SIZE) 199 #define RDS_CONG_MONITOR_MASK(port) (1ULL << RDS_CONG_MONITOR_BIT(port)) 200 201 /* 202 * RDMA related types 203 */ 204 205 /* 206 * This encapsulates a remote memory location. 207 * In the current implementation, it contains the R_Key 208 * of the remote memory region, and the offset into it 209 * (so that the application does not have to worry about 210 * alignment). 211 */ 212 typedef uint64_t rds_rdma_cookie_t; 213 214 struct rds_iovec { 215 uint64_t addr; 216 uint64_t bytes; 217 }; 218 219 struct rds_get_mr_args { 220 struct rds_iovec vec; 221 uint64_t cookie_addr; 222 uint64_t flags; 223 }; 224 225 struct rds_get_mr_for_dest_args { 226 struct sockaddr_storage dest_addr; 227 struct rds_iovec vec; 228 uint64_t cookie_addr; 229 uint64_t flags; 230 }; 231 232 struct rds_free_mr_args { 233 rds_rdma_cookie_t cookie; 234 uint64_t flags; 235 }; 236 237 struct rds_rdma_args { 238 rds_rdma_cookie_t cookie; 239 struct rds_iovec remote_vec; 240 uint64_t local_vec_addr; 241 uint64_t nr_local; 242 uint64_t flags; 243 uint64_t user_token; 244 }; 245 246 struct rds_atomic_args { 247 rds_rdma_cookie_t cookie; 248 uint64_t local_addr; 249 uint64_t remote_addr; 250 union { 251 struct { 252 uint64_t compare; 253 uint64_t swap; 254 } cswp; 255 struct { 256 uint64_t add; 257 } fadd; 258 struct { 259 uint64_t compare; 260 uint64_t swap; 261 uint64_t compare_mask; 262 uint64_t swap_mask; 263 } m_cswp; 264 struct { 265 uint64_t add; 266 uint64_t nocarry_mask; 267 } m_fadd; 268 }; 269 uint64_t flags; 270 uint64_t user_token; 271 }; 272 273 struct rds_rdma_notify { 274 uint64_t user_token; 275 int32_t status; 276 }; 277 278 #define RDS_RDMA_SUCCESS 0 279 #define RDS_RDMA_REMOTE_ERROR 1 280 #define RDS_RDMA_CANCELED 2 281 #define RDS_RDMA_DROPPED 3 282 #define RDS_RDMA_OTHER_ERROR 4 283 284 /* 285 * Common set of flags for all RDMA related structs 286 */ 287 #define RDS_RDMA_READWRITE 0x0001 288 #define RDS_RDMA_FENCE 0x0002 /* use FENCE for immediate send */ 289 #define RDS_RDMA_INVALIDATE 0x0004 /* invalidate R_Key after freeing MR */ 290 #define RDS_RDMA_USE_ONCE 0x0008 /* free MR after use */ 291 #define RDS_RDMA_DONTWAIT 0x0010 /* Don't wait in SET_BARRIER */ 292 #define RDS_RDMA_NOTIFY_ME 0x0020 /* Notify when operation completes */ 293 #define RDS_RDMA_SILENT 0x0040 /* Do not interrupt remote */ 294 295 #endif /* IB_RDS_H */ 296