1/* 2 * Copyright (c) 2005-2006 Network Appliance, Inc. 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 BSD-type 8 * license below: 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 14 * Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 17 * Redistributions in binary form must reproduce the above 18 * copyright notice, this list of conditions and the following 19 * disclaimer in the documentation and/or other materials provided 20 * with the distribution. 21 * 22 * Neither the name of the Network Appliance, Inc. nor the names of 23 * its contributors may be used to endorse or promote products 24 * derived from this software without specific prior written 25 * permission. 26 * 27 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 28 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 29 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 30 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 31 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 32 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 33 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 34 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 35 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 36 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 37 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 38 * 39 * Author: Tom Tucker <tom@opengridcomputing.com> 40 */ 41 42#ifndef SVC_RDMA_H 43#define SVC_RDMA_H 44#include <linux/sunrpc/xdr.h> 45#include <linux/sunrpc/svcsock.h> 46#include <linux/sunrpc/rpc_rdma.h> 47#include <rdma/ib_verbs.h> 48#include <rdma/rdma_cm.h> 49#define SVCRDMA_DEBUG 50 51/* RPC/RDMA parameters and stats */ 52extern unsigned int svcrdma_ord; 53extern unsigned int svcrdma_max_requests; 54extern unsigned int svcrdma_max_req_size; 55 56extern atomic_t rdma_stat_recv; 57extern atomic_t rdma_stat_read; 58extern atomic_t rdma_stat_write; 59extern atomic_t rdma_stat_sq_starve; 60extern atomic_t rdma_stat_rq_starve; 61extern atomic_t rdma_stat_rq_poll; 62extern atomic_t rdma_stat_rq_prod; 63extern atomic_t rdma_stat_sq_poll; 64extern atomic_t rdma_stat_sq_prod; 65 66/* 67 * Contexts are built when an RDMA request is created and are a 68 * record of the resources that can be recovered when the request 69 * completes. 70 */ 71struct svc_rdma_op_ctxt { 72 struct svc_rdma_op_ctxt *read_hdr; 73 struct svc_rdma_fastreg_mr *frmr; 74 int hdr_count; 75 struct xdr_buf arg; 76 struct list_head dto_q; 77 enum ib_wr_opcode wr_op; 78 enum ib_wc_status wc_status; 79 u32 byte_len; 80 u32 position; 81 struct svcxprt_rdma *xprt; 82 unsigned long flags; 83 enum dma_data_direction direction; 84 int count; 85 struct ib_sge sge[RPCSVC_MAXPAGES]; 86 struct page *pages[RPCSVC_MAXPAGES]; 87}; 88 89/* 90 * NFS_ requests are mapped on the client side by the chunk lists in 91 * the RPCRDMA header. During the fetching of the RPC from the client 92 * and the writing of the reply to the client, the memory in the 93 * client and the memory in the server must be mapped as contiguous 94 * vaddr/len for access by the hardware. These data strucures keep 95 * these mappings. 96 * 97 * For an RDMA_WRITE, the 'sge' maps the RPC REPLY. For RDMA_READ, the 98 * 'sge' in the svc_rdma_req_map maps the server side RPC reply and the 99 * 'ch' field maps the read-list of the RPCRDMA header to the 'sge' 100 * mapping of the reply. 101 */ 102struct svc_rdma_chunk_sge { 103 int start; /* sge no for this chunk */ 104 int count; /* sge count for this chunk */ 105}; 106struct svc_rdma_fastreg_mr { 107 struct ib_mr *mr; 108 void *kva; 109 struct ib_fast_reg_page_list *page_list; 110 int page_list_len; 111 unsigned long access_flags; 112 unsigned long map_len; 113 enum dma_data_direction direction; 114 struct list_head frmr_list; 115}; 116struct svc_rdma_req_map { 117 unsigned long count; 118 union { 119 struct kvec sge[RPCSVC_MAXPAGES]; 120 struct svc_rdma_chunk_sge ch[RPCSVC_MAXPAGES]; 121 unsigned long lkey[RPCSVC_MAXPAGES]; 122 }; 123}; 124#define RDMACTXT_F_LAST_CTXT 2 125 126#define SVCRDMA_DEVCAP_FAST_REG 1 /* fast mr registration */ 127#define SVCRDMA_DEVCAP_READ_W_INV 2 /* read w/ invalidate */ 128 129struct svcxprt_rdma { 130 struct svc_xprt sc_xprt; /* SVC transport structure */ 131 struct rdma_cm_id *sc_cm_id; /* RDMA connection id */ 132 struct list_head sc_accept_q; /* Conn. waiting accept */ 133 int sc_ord; /* RDMA read limit */ 134 int sc_max_sge; 135 136 int sc_sq_depth; /* Depth of SQ */ 137 atomic_t sc_sq_count; /* Number of SQ WR on queue */ 138 139 int sc_max_requests; /* Depth of RQ */ 140 int sc_max_req_size; /* Size of each RQ WR buf */ 141 142 struct ib_pd *sc_pd; 143 144 atomic_t sc_dma_used; 145 atomic_t sc_ctxt_used; 146 struct list_head sc_rq_dto_q; 147 spinlock_t sc_rq_dto_lock; 148 struct ib_qp *sc_qp; 149 struct ib_cq *sc_rq_cq; 150 struct ib_cq *sc_sq_cq; 151 struct ib_mr *sc_phys_mr; /* MR for server memory */ 152 int (*sc_reader)(struct svcxprt_rdma *, 153 struct svc_rqst *, 154 struct svc_rdma_op_ctxt *, 155 int *, u32 *, u32, u32, u64, bool); 156 u32 sc_dev_caps; /* distilled device caps */ 157 u32 sc_dma_lkey; /* local dma key */ 158 unsigned int sc_frmr_pg_list_len; 159 struct list_head sc_frmr_q; 160 spinlock_t sc_frmr_q_lock; 161 162 spinlock_t sc_lock; /* transport lock */ 163 164 wait_queue_head_t sc_send_wait; /* SQ exhaustion waitlist */ 165 unsigned long sc_flags; 166 struct list_head sc_dto_q; /* DTO tasklet I/O pending Q */ 167 struct list_head sc_read_complete_q; 168 struct work_struct sc_work; 169}; 170/* sc_flags */ 171#define RDMAXPRT_RQ_PENDING 1 172#define RDMAXPRT_SQ_PENDING 2 173#define RDMAXPRT_CONN_PENDING 3 174 175#define RPCRDMA_LISTEN_BACKLOG 10 176/* The default ORD value is based on two outstanding full-size writes with a 177 * page size of 4k, or 32k * 2 ops / 4k = 16 outstanding RDMA_READ. */ 178#define RPCRDMA_ORD (64/4) 179#define RPCRDMA_SQ_DEPTH_MULT 8 180#define RPCRDMA_MAX_REQUESTS 32 181#define RPCRDMA_MAX_REQ_SIZE 4096 182 183/* svc_rdma_marshal.c */ 184extern int svc_rdma_xdr_decode_req(struct rpcrdma_msg **, struct svc_rqst *); 185extern int svc_rdma_xdr_decode_deferred_req(struct svc_rqst *); 186extern int svc_rdma_xdr_encode_error(struct svcxprt_rdma *, 187 struct rpcrdma_msg *, 188 enum rpcrdma_errcode, u32 *); 189extern void svc_rdma_xdr_encode_write_list(struct rpcrdma_msg *, int); 190extern void svc_rdma_xdr_encode_reply_array(struct rpcrdma_write_array *, int); 191extern void svc_rdma_xdr_encode_array_chunk(struct rpcrdma_write_array *, int, 192 __be32, __be64, u32); 193extern void svc_rdma_xdr_encode_reply_header(struct svcxprt_rdma *, 194 struct rpcrdma_msg *, 195 struct rpcrdma_msg *, 196 enum rpcrdma_proc); 197extern int svc_rdma_xdr_get_reply_hdr_len(struct rpcrdma_msg *); 198 199/* svc_rdma_recvfrom.c */ 200extern int svc_rdma_recvfrom(struct svc_rqst *); 201extern int rdma_read_chunk_lcl(struct svcxprt_rdma *, struct svc_rqst *, 202 struct svc_rdma_op_ctxt *, int *, u32 *, 203 u32, u32, u64, bool); 204extern int rdma_read_chunk_frmr(struct svcxprt_rdma *, struct svc_rqst *, 205 struct svc_rdma_op_ctxt *, int *, u32 *, 206 u32, u32, u64, bool); 207 208/* svc_rdma_sendto.c */ 209extern int svc_rdma_sendto(struct svc_rqst *); 210 211/* svc_rdma_transport.c */ 212extern int svc_rdma_send(struct svcxprt_rdma *, struct ib_send_wr *); 213extern void svc_rdma_send_error(struct svcxprt_rdma *, struct rpcrdma_msg *, 214 enum rpcrdma_errcode); 215struct page *svc_rdma_get_page(void); 216extern int svc_rdma_post_recv(struct svcxprt_rdma *); 217extern int svc_rdma_create_listen(struct svc_serv *, int, struct sockaddr *); 218extern struct svc_rdma_op_ctxt *svc_rdma_get_context(struct svcxprt_rdma *); 219extern void svc_rdma_put_context(struct svc_rdma_op_ctxt *, int); 220extern void svc_rdma_unmap_dma(struct svc_rdma_op_ctxt *ctxt); 221extern struct svc_rdma_req_map *svc_rdma_get_req_map(void); 222extern void svc_rdma_put_req_map(struct svc_rdma_req_map *); 223extern int svc_rdma_fastreg(struct svcxprt_rdma *, struct svc_rdma_fastreg_mr *); 224extern struct svc_rdma_fastreg_mr *svc_rdma_get_frmr(struct svcxprt_rdma *); 225extern void svc_rdma_put_frmr(struct svcxprt_rdma *, 226 struct svc_rdma_fastreg_mr *); 227extern void svc_sq_reap(struct svcxprt_rdma *); 228extern void svc_rq_reap(struct svcxprt_rdma *); 229extern struct svc_xprt_class svc_rdma_class; 230extern void svc_rdma_prep_reply_hdr(struct svc_rqst *); 231 232/* svc_rdma.c */ 233extern int svc_rdma_init(void); 234extern void svc_rdma_cleanup(void); 235 236/* 237 * Returns the address of the first read chunk or <nul> if no read chunk is 238 * present 239 */ 240static inline struct rpcrdma_read_chunk * 241svc_rdma_get_read_chunk(struct rpcrdma_msg *rmsgp) 242{ 243 struct rpcrdma_read_chunk *ch = 244 (struct rpcrdma_read_chunk *)&rmsgp->rm_body.rm_chunks[0]; 245 246 if (ch->rc_discrim == 0) 247 return NULL; 248 249 return ch; 250} 251 252/* 253 * Returns the address of the first read write array element or <nul> if no 254 * write array list is present 255 */ 256static inline struct rpcrdma_write_array * 257svc_rdma_get_write_array(struct rpcrdma_msg *rmsgp) 258{ 259 if (rmsgp->rm_body.rm_chunks[0] != 0 260 || rmsgp->rm_body.rm_chunks[1] == 0) 261 return NULL; 262 263 return (struct rpcrdma_write_array *)&rmsgp->rm_body.rm_chunks[1]; 264} 265 266/* 267 * Returns the address of the first reply array element or <nul> if no 268 * reply array is present 269 */ 270static inline struct rpcrdma_write_array * 271svc_rdma_get_reply_array(struct rpcrdma_msg *rmsgp) 272{ 273 struct rpcrdma_read_chunk *rch; 274 struct rpcrdma_write_array *wr_ary; 275 struct rpcrdma_write_array *rp_ary; 276 277 /* XXX: Need to fix when reply list may occur with read-list and/or 278 * write list */ 279 if (rmsgp->rm_body.rm_chunks[0] != 0 || 280 rmsgp->rm_body.rm_chunks[1] != 0) 281 return NULL; 282 283 rch = svc_rdma_get_read_chunk(rmsgp); 284 if (rch) { 285 while (rch->rc_discrim) 286 rch++; 287 288 /* The reply list follows an empty write array located 289 * at 'rc_position' here. The reply array is at rc_target. 290 */ 291 rp_ary = (struct rpcrdma_write_array *)&rch->rc_target; 292 293 goto found_it; 294 } 295 296 wr_ary = svc_rdma_get_write_array(rmsgp); 297 if (wr_ary) { 298 rp_ary = (struct rpcrdma_write_array *) 299 &wr_ary-> 300 wc_array[ntohl(wr_ary->wc_nchunks)].wc_target.rs_length; 301 302 goto found_it; 303 } 304 305 /* No read list, no write list */ 306 rp_ary = (struct rpcrdma_write_array *) 307 &rmsgp->rm_body.rm_chunks[2]; 308 309 found_it: 310 if (rp_ary->wc_discrim == 0) 311 return NULL; 312 313 return rp_ary; 314} 315#endif 316