1 /* SPDX-License-Identifier: GPL-2.0 OR BSD-3-Clause */ 2 /* 3 * Copyright (c) 2005-2006 Network Appliance, Inc. All rights reserved. 4 * 5 * This software is available to you under a choice of one of two 6 * licenses. You may choose to be licensed under the terms of the GNU 7 * General Public License (GPL) Version 2, available from the file 8 * COPYING in the main directory of this source tree, or the BSD-type 9 * license below: 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 15 * Redistributions of source code must retain the above copyright 16 * notice, this list of conditions and the following 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 provided 21 * with the distribution. 22 * 23 * Neither the name of the Network Appliance, Inc. nor the names of 24 * its contributors may be used to endorse or promote products 25 * derived from this software without specific prior written 26 * permission. 27 * 28 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 29 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 30 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 31 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 32 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 33 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 34 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 35 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 36 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 37 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 38 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 39 * 40 * Author: Tom Tucker <tom@opengridcomputing.com> 41 */ 42 43 #ifndef SVC_RDMA_H 44 #define SVC_RDMA_H 45 #include <linux/llist.h> 46 #include <linux/sunrpc/xdr.h> 47 #include <linux/sunrpc/svcsock.h> 48 #include <linux/sunrpc/rpc_rdma.h> 49 #include <linux/sunrpc/rpc_rdma_cid.h> 50 #include <linux/sunrpc/svc_rdma_pcl.h> 51 #include <linux/sunrpc/rdma_rn.h> 52 53 #include <linux/percpu_counter.h> 54 #include <rdma/ib_verbs.h> 55 #include <rdma/rdma_cm.h> 56 57 /* Default and maximum inline threshold sizes */ 58 enum { 59 RPCRDMA_PULLUP_THRESH = RPCRDMA_V1_DEF_INLINE_SIZE >> 1, 60 RPCRDMA_DEF_INLINE_THRESH = 4096, 61 RPCRDMA_MAX_INLINE_THRESH = 65536 62 }; 63 64 /* RPC/RDMA parameters and stats */ 65 extern unsigned int svcrdma_ord; 66 extern unsigned int svcrdma_max_requests; 67 extern unsigned int svcrdma_max_bc_requests; 68 extern unsigned int svcrdma_max_req_size; 69 extern struct workqueue_struct *svcrdma_wq; 70 71 extern struct percpu_counter svcrdma_stat_read; 72 extern struct percpu_counter svcrdma_stat_recv; 73 extern struct percpu_counter svcrdma_stat_sq_starve; 74 extern struct percpu_counter svcrdma_stat_write; 75 76 struct svcxprt_rdma { 77 struct svc_xprt sc_xprt; /* SVC transport structure */ 78 struct rdma_cm_id *sc_cm_id; /* RDMA connection id */ 79 struct list_head sc_accept_q; /* Conn. waiting accept */ 80 struct rpcrdma_notification sc_rn; /* removal notification */ 81 int sc_ord; /* RDMA read limit */ 82 int sc_max_send_sges; 83 bool sc_snd_w_inv; /* OK to use Send With Invalidate */ 84 85 atomic_t sc_sq_avail; /* SQEs ready to be consumed */ 86 unsigned int sc_sq_depth; /* Depth of SQ */ 87 __be32 sc_fc_credits; /* Forward credits */ 88 u32 sc_max_requests; /* Max requests */ 89 u32 sc_max_bc_requests;/* Backward credits */ 90 int sc_max_req_size; /* Size of each RQ WR buf */ 91 u8 sc_port_num; 92 93 struct ib_pd *sc_pd; 94 95 spinlock_t sc_send_lock; 96 struct llist_head sc_send_ctxts; 97 spinlock_t sc_rw_ctxt_lock; 98 struct llist_head sc_rw_ctxts; 99 100 u32 sc_pending_recvs; 101 u32 sc_recv_batch; 102 struct list_head sc_rq_dto_q; 103 struct list_head sc_read_complete_q; 104 spinlock_t sc_rq_dto_lock; 105 struct ib_qp *sc_qp; 106 struct ib_cq *sc_rq_cq; 107 struct ib_cq *sc_sq_cq; 108 109 spinlock_t sc_lock; /* transport lock */ 110 111 wait_queue_head_t sc_send_wait; /* SQ exhaustion waitlist */ 112 unsigned long sc_flags; 113 struct work_struct sc_work; 114 115 struct llist_head sc_recv_ctxts; 116 117 atomic_t sc_completion_ids; 118 }; 119 /* sc_flags */ 120 #define RDMAXPRT_CONN_PENDING 3 121 122 static inline struct svcxprt_rdma *svc_rdma_rqst_rdma(struct svc_rqst *rqstp) 123 { 124 struct svc_xprt *xprt = rqstp->rq_xprt; 125 126 return container_of(xprt, struct svcxprt_rdma, sc_xprt); 127 } 128 129 /* 130 * Default connection parameters 131 */ 132 enum { 133 RPCRDMA_LISTEN_BACKLOG = 10, 134 RPCRDMA_MAX_REQUESTS = 64, 135 RPCRDMA_MAX_BC_REQUESTS = 2, 136 }; 137 138 #define RPCSVC_MAXPAYLOAD_RDMA RPCSVC_MAXPAYLOAD 139 140 /** 141 * svc_rdma_send_cid_init - Initialize a Receive Queue completion ID 142 * @rdma: controlling transport 143 * @cid: completion ID to initialize 144 */ 145 static inline void svc_rdma_recv_cid_init(struct svcxprt_rdma *rdma, 146 struct rpc_rdma_cid *cid) 147 { 148 cid->ci_queue_id = rdma->sc_rq_cq->res.id; 149 cid->ci_completion_id = atomic_inc_return(&rdma->sc_completion_ids); 150 } 151 152 /** 153 * svc_rdma_send_cid_init - Initialize a Send Queue completion ID 154 * @rdma: controlling transport 155 * @cid: completion ID to initialize 156 */ 157 static inline void svc_rdma_send_cid_init(struct svcxprt_rdma *rdma, 158 struct rpc_rdma_cid *cid) 159 { 160 cid->ci_queue_id = rdma->sc_sq_cq->res.id; 161 cid->ci_completion_id = atomic_inc_return(&rdma->sc_completion_ids); 162 } 163 164 /* 165 * A chunk context tracks all I/O for moving one Read or Write 166 * chunk. This is a set of rdma_rw's that handle data movement 167 * for all segments of one chunk. 168 */ 169 struct svc_rdma_chunk_ctxt { 170 struct rpc_rdma_cid cc_cid; 171 struct ib_cqe cc_cqe; 172 struct list_head cc_rwctxts; 173 ktime_t cc_posttime; 174 int cc_sqecount; 175 }; 176 177 struct svc_rdma_recv_ctxt { 178 struct llist_node rc_node; 179 struct list_head rc_list; 180 struct ib_recv_wr rc_recv_wr; 181 struct ib_cqe rc_cqe; 182 struct rpc_rdma_cid rc_cid; 183 struct ib_sge rc_recv_sge; 184 void *rc_recv_buf; 185 struct xdr_stream rc_stream; 186 u32 rc_byte_len; 187 u32 rc_inv_rkey; 188 __be32 rc_msgtype; 189 190 /* State for pulling a Read chunk */ 191 unsigned int rc_pageoff; 192 unsigned int rc_curpage; 193 unsigned int rc_readbytes; 194 struct xdr_buf rc_saved_arg; 195 struct svc_rdma_chunk_ctxt rc_cc; 196 197 struct svc_rdma_pcl rc_call_pcl; 198 199 struct svc_rdma_pcl rc_read_pcl; 200 struct svc_rdma_chunk *rc_cur_result_payload; 201 struct svc_rdma_pcl rc_write_pcl; 202 struct svc_rdma_pcl rc_reply_pcl; 203 204 unsigned int rc_page_count; 205 unsigned long rc_maxpages; 206 struct page *rc_pages[] __counted_by(rc_maxpages); 207 }; 208 209 /* 210 * State for sending a Write chunk. 211 * - Tracks progress of writing one chunk over all its segments 212 * - Stores arguments for the SGL constructor functions 213 */ 214 struct svc_rdma_write_info { 215 struct svcxprt_rdma *wi_rdma; 216 217 const struct svc_rdma_chunk *wi_chunk; 218 219 /* write state of this chunk */ 220 unsigned int wi_seg_off; 221 unsigned int wi_seg_no; 222 223 /* SGL constructor arguments */ 224 const struct xdr_buf *wi_xdr; 225 unsigned char *wi_base; 226 unsigned int wi_next_off; 227 228 struct svc_rdma_chunk_ctxt wi_cc; 229 struct work_struct wi_work; 230 }; 231 232 struct svc_rdma_send_ctxt { 233 struct llist_node sc_node; 234 struct rpc_rdma_cid sc_cid; 235 struct work_struct sc_work; 236 237 struct svcxprt_rdma *sc_rdma; 238 struct ib_send_wr sc_send_wr; 239 struct ib_send_wr *sc_wr_chain; 240 int sc_sqecount; 241 struct ib_cqe sc_cqe; 242 struct xdr_buf sc_hdrbuf; 243 struct xdr_stream sc_stream; 244 struct svc_rdma_write_info sc_reply_info; 245 void *sc_xprt_buf; 246 int sc_page_count; 247 int sc_cur_sge_no; 248 unsigned long sc_maxpages; 249 struct page **sc_pages; 250 struct ib_sge sc_sges[]; 251 }; 252 253 /* svc_rdma_backchannel.c */ 254 extern void svc_rdma_handle_bc_reply(struct svc_rqst *rqstp, 255 struct svc_rdma_recv_ctxt *rctxt); 256 257 /* svc_rdma_recvfrom.c */ 258 extern void svc_rdma_recv_ctxts_destroy(struct svcxprt_rdma *rdma); 259 extern bool svc_rdma_post_recvs(struct svcxprt_rdma *rdma); 260 extern struct svc_rdma_recv_ctxt * 261 svc_rdma_recv_ctxt_get(struct svcxprt_rdma *rdma); 262 extern void svc_rdma_recv_ctxt_put(struct svcxprt_rdma *rdma, 263 struct svc_rdma_recv_ctxt *ctxt); 264 extern void svc_rdma_flush_recv_queues(struct svcxprt_rdma *rdma); 265 extern void svc_rdma_release_ctxt(struct svc_xprt *xprt, void *ctxt); 266 extern int svc_rdma_recvfrom(struct svc_rqst *); 267 268 /* svc_rdma_rw.c */ 269 extern void svc_rdma_cc_init(struct svcxprt_rdma *rdma, 270 struct svc_rdma_chunk_ctxt *cc); 271 extern void svc_rdma_destroy_rw_ctxts(struct svcxprt_rdma *rdma); 272 extern void svc_rdma_cc_init(struct svcxprt_rdma *rdma, 273 struct svc_rdma_chunk_ctxt *cc); 274 extern void svc_rdma_cc_release(struct svcxprt_rdma *rdma, 275 struct svc_rdma_chunk_ctxt *cc, 276 enum dma_data_direction dir); 277 extern void svc_rdma_reply_chunk_release(struct svcxprt_rdma *rdma, 278 struct svc_rdma_send_ctxt *ctxt); 279 extern int svc_rdma_send_write_list(struct svcxprt_rdma *rdma, 280 const struct svc_rdma_recv_ctxt *rctxt, 281 const struct xdr_buf *xdr); 282 extern int svc_rdma_prepare_reply_chunk(struct svcxprt_rdma *rdma, 283 const struct svc_rdma_pcl *write_pcl, 284 const struct svc_rdma_pcl *reply_pcl, 285 struct svc_rdma_send_ctxt *sctxt, 286 const struct xdr_buf *xdr); 287 extern int svc_rdma_process_read_list(struct svcxprt_rdma *rdma, 288 struct svc_rqst *rqstp, 289 struct svc_rdma_recv_ctxt *head); 290 291 /* svc_rdma_sendto.c */ 292 extern void svc_rdma_send_ctxts_destroy(struct svcxprt_rdma *rdma); 293 extern struct svc_rdma_send_ctxt * 294 svc_rdma_send_ctxt_get(struct svcxprt_rdma *rdma); 295 extern void svc_rdma_send_ctxt_put(struct svcxprt_rdma *rdma, 296 struct svc_rdma_send_ctxt *ctxt); 297 extern int svc_rdma_post_send(struct svcxprt_rdma *rdma, 298 struct svc_rdma_send_ctxt *ctxt); 299 extern int svc_rdma_map_reply_msg(struct svcxprt_rdma *rdma, 300 struct svc_rdma_send_ctxt *sctxt, 301 const struct svc_rdma_pcl *write_pcl, 302 const struct svc_rdma_pcl *reply_pcl, 303 const struct xdr_buf *xdr); 304 extern void svc_rdma_send_error_msg(struct svcxprt_rdma *rdma, 305 struct svc_rdma_send_ctxt *sctxt, 306 struct svc_rdma_recv_ctxt *rctxt, 307 int status); 308 extern void svc_rdma_wake_send_waiters(struct svcxprt_rdma *rdma, int avail); 309 extern int svc_rdma_sendto(struct svc_rqst *); 310 extern int svc_rdma_result_payload(struct svc_rqst *rqstp, unsigned int offset, 311 unsigned int length); 312 313 /* svc_rdma_transport.c */ 314 extern struct svc_xprt_class svc_rdma_class; 315 #ifdef CONFIG_SUNRPC_BACKCHANNEL 316 extern struct svc_xprt_class svc_rdma_bc_class; 317 #endif 318 319 /* svc_rdma.c */ 320 extern int svc_rdma_init(void); 321 extern void svc_rdma_cleanup(void); 322 323 #endif 324