2 * Copyright (c) 2005-2006 Network Appliance, Inc. All rights reserved.
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
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
14 * Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
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.
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
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.
39 * Author: Tom Tucker <tom@opengridcomputing.com>
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>
51 /* RPC/RDMA parameters and stats */
52 extern unsigned int svcrdma_ord;
53 extern unsigned int svcrdma_max_requests;
54 extern unsigned int svcrdma_max_req_size;
56 extern atomic_t rdma_stat_recv;
57 extern atomic_t rdma_stat_read;
58 extern atomic_t rdma_stat_write;
59 extern atomic_t rdma_stat_sq_starve;
60 extern atomic_t rdma_stat_rq_starve;
61 extern atomic_t rdma_stat_rq_poll;
62 extern atomic_t rdma_stat_rq_prod;
63 extern atomic_t rdma_stat_sq_poll;
64 extern atomic_t rdma_stat_sq_prod;
66 #define RPCRDMA_VERSION 1
69 * Contexts are built when an RDMA request is created and are a
70 * record of the resources that can be recovered when the request
73 struct svc_rdma_op_ctxt {
74 struct svc_rdma_op_ctxt *next;
76 struct list_head dto_q;
77 enum ib_wr_opcode wr_op;
78 enum ib_wc_status wc_status;
80 struct svcxprt_rdma *xprt;
82 enum dma_data_direction direction;
84 struct ib_sge sge[RPCSVC_MAXPAGES];
85 struct page *pages[RPCSVC_MAXPAGES];
88 #define RDMACTXT_F_LAST_CTXT 2
91 struct svc_xprt sc_xprt; /* SVC transport structure */
92 struct rdma_cm_id *sc_cm_id; /* RDMA connection id */
93 struct list_head sc_accept_q; /* Conn. waiting accept */
94 int sc_ord; /* RDMA read limit */
95 wait_queue_head_t sc_read_wait;
98 int sc_sq_depth; /* Depth of SQ */
99 atomic_t sc_sq_count; /* Number of SQ WR on queue */
101 int sc_max_requests; /* Depth of RQ */
102 int sc_max_req_size; /* Size of each RQ WR buf */
106 struct svc_rdma_op_ctxt *sc_ctxt_head;
110 spinlock_t sc_ctxt_lock;
111 struct list_head sc_rq_dto_q;
112 spinlock_t sc_rq_dto_lock;
114 struct ib_cq *sc_rq_cq;
115 struct ib_cq *sc_sq_cq;
116 struct ib_mr *sc_phys_mr; /* MR for server memory */
118 spinlock_t sc_lock; /* transport lock */
120 wait_queue_head_t sc_send_wait; /* SQ exhaustion waitlist */
121 unsigned long sc_flags;
122 struct list_head sc_dto_q; /* DTO tasklet I/O pending Q */
123 struct list_head sc_read_complete_q;
124 spinlock_t sc_read_complete_lock;
127 #define RDMAXPRT_RQ_PENDING 1
128 #define RDMAXPRT_SQ_PENDING 2
129 #define RDMAXPRT_CONN_PENDING 3
131 #define RPCRDMA_LISTEN_BACKLOG 10
132 /* The default ORD value is based on two outstanding full-size writes with a
133 * page size of 4k, or 32k * 2 ops / 4k = 16 outstanding RDMA_READ. */
134 #define RPCRDMA_ORD (64/4)
135 #define RPCRDMA_SQ_DEPTH_MULT 8
136 #define RPCRDMA_MAX_THREADS 16
137 #define RPCRDMA_MAX_REQUESTS 16
138 #define RPCRDMA_MAX_REQ_SIZE 4096
140 /* svc_rdma_marshal.c */
141 extern void svc_rdma_rcl_chunk_counts(struct rpcrdma_read_chunk *,
143 extern int svc_rdma_xdr_decode_req(struct rpcrdma_msg **, struct svc_rqst *);
144 extern int svc_rdma_xdr_decode_deferred_req(struct svc_rqst *);
145 extern int svc_rdma_xdr_encode_error(struct svcxprt_rdma *,
146 struct rpcrdma_msg *,
147 enum rpcrdma_errcode, u32 *);
148 extern void svc_rdma_xdr_encode_write_list(struct rpcrdma_msg *, int);
149 extern void svc_rdma_xdr_encode_reply_array(struct rpcrdma_write_array *, int);
150 extern void svc_rdma_xdr_encode_array_chunk(struct rpcrdma_write_array *, int,
152 extern void svc_rdma_xdr_encode_reply_header(struct svcxprt_rdma *,
153 struct rpcrdma_msg *,
154 struct rpcrdma_msg *,
156 extern int svc_rdma_xdr_get_reply_hdr_len(struct rpcrdma_msg *);
158 /* svc_rdma_recvfrom.c */
159 extern int svc_rdma_recvfrom(struct svc_rqst *);
161 /* svc_rdma_sendto.c */
162 extern int svc_rdma_sendto(struct svc_rqst *);
164 /* svc_rdma_transport.c */
165 extern int svc_rdma_send(struct svcxprt_rdma *, struct ib_send_wr *);
166 extern int svc_rdma_send_error(struct svcxprt_rdma *, struct rpcrdma_msg *,
167 enum rpcrdma_errcode);
168 struct page *svc_rdma_get_page(void);
169 extern int svc_rdma_post_recv(struct svcxprt_rdma *);
170 extern int svc_rdma_create_listen(struct svc_serv *, int, struct sockaddr *);
171 extern struct svc_rdma_op_ctxt *svc_rdma_get_context(struct svcxprt_rdma *);
172 extern void svc_rdma_put_context(struct svc_rdma_op_ctxt *, int);
173 extern void svc_sq_reap(struct svcxprt_rdma *);
174 extern void svc_rq_reap(struct svcxprt_rdma *);
175 extern struct svc_xprt_class svc_rdma_class;
176 extern void svc_rdma_prep_reply_hdr(struct svc_rqst *);
179 extern int svc_rdma_init(void);
180 extern void svc_rdma_cleanup(void);
183 * Returns the address of the first read chunk or <nul> if no read chunk is
186 static inline struct rpcrdma_read_chunk *
187 svc_rdma_get_read_chunk(struct rpcrdma_msg *rmsgp)
189 struct rpcrdma_read_chunk *ch =
190 (struct rpcrdma_read_chunk *)&rmsgp->rm_body.rm_chunks[0];
192 if (ch->rc_discrim == 0)
199 * Returns the address of the first read write array element or <nul> if no
200 * write array list is present
202 static inline struct rpcrdma_write_array *
203 svc_rdma_get_write_array(struct rpcrdma_msg *rmsgp)
205 if (rmsgp->rm_body.rm_chunks[0] != 0
206 || rmsgp->rm_body.rm_chunks[1] == 0)
209 return (struct rpcrdma_write_array *)&rmsgp->rm_body.rm_chunks[1];
213 * Returns the address of the first reply array element or <nul> if no
214 * reply array is present
216 static inline struct rpcrdma_write_array *
217 svc_rdma_get_reply_array(struct rpcrdma_msg *rmsgp)
219 struct rpcrdma_read_chunk *rch;
220 struct rpcrdma_write_array *wr_ary;
221 struct rpcrdma_write_array *rp_ary;
223 /* XXX: Need to fix when reply list may occur with read-list and/or
225 if (rmsgp->rm_body.rm_chunks[0] != 0 ||
226 rmsgp->rm_body.rm_chunks[1] != 0)
229 rch = svc_rdma_get_read_chunk(rmsgp);
231 while (rch->rc_discrim)
234 /* The reply list follows an empty write array located
235 * at 'rc_position' here. The reply array is at rc_target.
237 rp_ary = (struct rpcrdma_write_array *)&rch->rc_target;
242 wr_ary = svc_rdma_get_write_array(rmsgp);
244 rp_ary = (struct rpcrdma_write_array *)
246 wc_array[wr_ary->wc_nchunks].wc_target.rs_length;
251 /* No read list, no write list */
252 rp_ary = (struct rpcrdma_write_array *)
253 &rmsgp->rm_body.rm_chunks[2];
256 if (rp_ary->wc_discrim == 0)