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_bc_requests;
55 extern unsigned int svcrdma_max_req_size;
57 extern atomic_t rdma_stat_recv;
58 extern atomic_t rdma_stat_read;
59 extern atomic_t rdma_stat_write;
60 extern atomic_t rdma_stat_sq_starve;
61 extern atomic_t rdma_stat_rq_starve;
62 extern atomic_t rdma_stat_rq_poll;
63 extern atomic_t rdma_stat_rq_prod;
64 extern atomic_t rdma_stat_sq_poll;
65 extern atomic_t rdma_stat_sq_prod;
68 * Contexts are built when an RDMA request is created and are a
69 * record of the resources that can be recovered when the request
72 struct svc_rdma_op_ctxt {
73 struct list_head free;
74 struct svc_rdma_op_ctxt *read_hdr;
75 struct svc_rdma_fastreg_mr *frmr;
79 struct ib_cqe reg_cqe;
80 struct ib_cqe inv_cqe;
81 struct list_head dto_q;
82 enum ib_wc_status wc_status;
85 struct svcxprt_rdma *xprt;
87 enum dma_data_direction direction;
89 unsigned int mapped_sges;
90 struct ib_sge sge[RPCSVC_MAXPAGES];
91 struct page *pages[RPCSVC_MAXPAGES];
95 * NFS_ requests are mapped on the client side by the chunk lists in
96 * the RPCRDMA header. During the fetching of the RPC from the client
97 * and the writing of the reply to the client, the memory in the
98 * client and the memory in the server must be mapped as contiguous
99 * vaddr/len for access by the hardware. These data strucures keep
102 * For an RDMA_WRITE, the 'sge' maps the RPC REPLY. For RDMA_READ, the
103 * 'sge' in the svc_rdma_req_map maps the server side RPC reply and the
104 * 'ch' field maps the read-list of the RPCRDMA header to the 'sge'
105 * mapping of the reply.
107 struct svc_rdma_chunk_sge {
108 int start; /* sge no for this chunk */
109 int count; /* sge count for this chunk */
111 struct svc_rdma_fastreg_mr {
113 struct scatterlist *sg;
115 unsigned long access_flags;
116 enum dma_data_direction direction;
117 struct list_head frmr_list;
119 struct svc_rdma_req_map {
120 struct list_head free;
123 struct kvec sge[RPCSVC_MAXPAGES];
124 struct svc_rdma_chunk_sge ch[RPCSVC_MAXPAGES];
125 unsigned long lkey[RPCSVC_MAXPAGES];
128 #define RDMACTXT_F_LAST_CTXT 2
130 #define SVCRDMA_DEVCAP_FAST_REG 1 /* fast mr registration */
131 #define SVCRDMA_DEVCAP_READ_W_INV 2 /* read w/ invalidate */
133 struct svcxprt_rdma {
134 struct svc_xprt sc_xprt; /* SVC transport structure */
135 struct rdma_cm_id *sc_cm_id; /* RDMA connection id */
136 struct list_head sc_accept_q; /* Conn. waiting accept */
137 int sc_ord; /* RDMA read limit */
139 int sc_max_sge_rd; /* max sge for read target */
140 bool sc_snd_w_inv; /* OK to use Send With Invalidate */
142 atomic_t sc_sq_count; /* Number of SQ WR on queue */
143 unsigned int sc_sq_depth; /* Depth of SQ */
144 unsigned int sc_rq_depth; /* Depth of RQ */
145 u32 sc_max_requests; /* Forward credits */
146 u32 sc_max_bc_requests;/* Backward credits */
147 int sc_max_req_size; /* Size of each RQ WR buf */
151 atomic_t sc_dma_used;
152 spinlock_t sc_ctxt_lock;
153 struct list_head sc_ctxts;
155 spinlock_t sc_map_lock;
156 struct list_head sc_maps;
158 struct list_head sc_rq_dto_q;
159 spinlock_t sc_rq_dto_lock;
161 struct ib_cq *sc_rq_cq;
162 struct ib_cq *sc_sq_cq;
163 int (*sc_reader)(struct svcxprt_rdma *,
165 struct svc_rdma_op_ctxt *,
166 int *, u32 *, u32, u32, u64, bool);
167 u32 sc_dev_caps; /* distilled device caps */
168 unsigned int sc_frmr_pg_list_len;
169 struct list_head sc_frmr_q;
170 spinlock_t sc_frmr_q_lock;
172 spinlock_t sc_lock; /* transport lock */
174 wait_queue_head_t sc_send_wait; /* SQ exhaustion waitlist */
175 unsigned long sc_flags;
176 struct list_head sc_dto_q; /* DTO tasklet I/O pending Q */
177 struct list_head sc_read_complete_q;
178 struct work_struct sc_work;
181 #define RDMAXPRT_CONN_PENDING 3
183 #define RPCRDMA_LISTEN_BACKLOG 10
184 /* The default ORD value is based on two outstanding full-size writes with a
185 * page size of 4k, or 32k * 2 ops / 4k = 16 outstanding RDMA_READ. */
186 #define RPCRDMA_ORD (64/4)
187 #define RPCRDMA_SQ_DEPTH_MULT 8
188 #define RPCRDMA_MAX_REQUESTS 32
189 #define RPCRDMA_MAX_REQ_SIZE 4096
191 /* Typical ULP usage of BC requests is NFSv4.1 backchannel. Our
192 * current NFSv4.1 implementation supports one backchannel slot.
194 #define RPCRDMA_MAX_BC_REQUESTS 2
196 #define RPCSVC_MAXPAYLOAD_RDMA RPCSVC_MAXPAYLOAD
198 /* Track DMA maps for this transport and context */
199 static inline void svc_rdma_count_mappings(struct svcxprt_rdma *rdma,
200 struct svc_rdma_op_ctxt *ctxt)
203 atomic_inc(&rdma->sc_dma_used);
206 /* svc_rdma_backchannel.c */
207 extern int svc_rdma_handle_bc_reply(struct rpc_xprt *xprt,
208 struct rpcrdma_msg *rmsgp,
209 struct xdr_buf *rcvbuf);
211 /* svc_rdma_marshal.c */
212 extern int svc_rdma_xdr_decode_req(struct xdr_buf *);
213 extern int svc_rdma_xdr_encode_error(struct svcxprt_rdma *,
214 struct rpcrdma_msg *,
215 enum rpcrdma_errcode, __be32 *);
216 extern void svc_rdma_xdr_encode_write_list(struct rpcrdma_msg *, int);
217 extern void svc_rdma_xdr_encode_reply_array(struct rpcrdma_write_array *, int);
218 extern void svc_rdma_xdr_encode_array_chunk(struct rpcrdma_write_array *, int,
219 __be32, __be64, u32);
220 extern void svc_rdma_xdr_encode_reply_header(struct svcxprt_rdma *,
221 struct rpcrdma_msg *,
222 struct rpcrdma_msg *,
224 extern int svc_rdma_xdr_get_reply_hdr_len(struct rpcrdma_msg *);
226 /* svc_rdma_recvfrom.c */
227 extern int svc_rdma_recvfrom(struct svc_rqst *);
228 extern int rdma_read_chunk_lcl(struct svcxprt_rdma *, struct svc_rqst *,
229 struct svc_rdma_op_ctxt *, int *, u32 *,
230 u32, u32, u64, bool);
231 extern int rdma_read_chunk_frmr(struct svcxprt_rdma *, struct svc_rqst *,
232 struct svc_rdma_op_ctxt *, int *, u32 *,
233 u32, u32, u64, bool);
235 /* svc_rdma_sendto.c */
236 extern int svc_rdma_map_xdr(struct svcxprt_rdma *, struct xdr_buf *,
237 struct svc_rdma_req_map *, bool);
238 extern int svc_rdma_sendto(struct svc_rqst *);
239 extern struct rpcrdma_read_chunk *
240 svc_rdma_get_read_chunk(struct rpcrdma_msg *);
241 extern void svc_rdma_send_error(struct svcxprt_rdma *, struct rpcrdma_msg *,
244 /* svc_rdma_transport.c */
245 extern void svc_rdma_wc_send(struct ib_cq *, struct ib_wc *);
246 extern void svc_rdma_wc_write(struct ib_cq *, struct ib_wc *);
247 extern void svc_rdma_wc_reg(struct ib_cq *, struct ib_wc *);
248 extern void svc_rdma_wc_read(struct ib_cq *, struct ib_wc *);
249 extern void svc_rdma_wc_inv(struct ib_cq *, struct ib_wc *);
250 extern int svc_rdma_send(struct svcxprt_rdma *, struct ib_send_wr *);
251 extern int svc_rdma_post_recv(struct svcxprt_rdma *, gfp_t);
252 extern int svc_rdma_repost_recv(struct svcxprt_rdma *, gfp_t);
253 extern int svc_rdma_create_listen(struct svc_serv *, int, struct sockaddr *);
254 extern struct svc_rdma_op_ctxt *svc_rdma_get_context(struct svcxprt_rdma *);
255 extern void svc_rdma_put_context(struct svc_rdma_op_ctxt *, int);
256 extern void svc_rdma_unmap_dma(struct svc_rdma_op_ctxt *ctxt);
257 extern struct svc_rdma_req_map *svc_rdma_get_req_map(struct svcxprt_rdma *);
258 extern void svc_rdma_put_req_map(struct svcxprt_rdma *,
259 struct svc_rdma_req_map *);
260 extern struct svc_rdma_fastreg_mr *svc_rdma_get_frmr(struct svcxprt_rdma *);
261 extern void svc_rdma_put_frmr(struct svcxprt_rdma *,
262 struct svc_rdma_fastreg_mr *);
263 extern void svc_sq_reap(struct svcxprt_rdma *);
264 extern void svc_rq_reap(struct svcxprt_rdma *);
265 extern void svc_rdma_prep_reply_hdr(struct svc_rqst *);
267 extern struct svc_xprt_class svc_rdma_class;
268 #ifdef CONFIG_SUNRPC_BACKCHANNEL
269 extern struct svc_xprt_class svc_rdma_bc_class;
273 extern struct workqueue_struct *svc_rdma_wq;
274 extern int svc_rdma_init(void);
275 extern void svc_rdma_cleanup(void);