2 * Copyright (c) 2003-2007 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.
40 #ifndef _LINUX_SUNRPC_XPRT_RDMA_H
41 #define _LINUX_SUNRPC_XPRT_RDMA_H
43 #include <linux/wait.h> /* wait_queue_head_t, etc */
44 #include <linux/spinlock.h> /* spinlock_t, etc */
45 #include <asm/atomic.h> /* atomic_t, etc */
47 #include <rdma/rdma_cm.h> /* RDMA connection api */
48 #include <rdma/ib_verbs.h> /* RDMA verbs api */
50 #include <linux/sunrpc/clnt.h> /* rpc_xprt */
51 #include <linux/sunrpc/rpc_rdma.h> /* RPC/RDMA protocol */
52 #include <linux/sunrpc/xprtrdma.h> /* xprt parameters */
55 * Interface Adapter -- one per transport instance
58 struct rdma_cm_id *ri_id;
60 struct ib_mr *ri_bind_mem;
61 struct completion ri_done;
63 enum rpcrdma_memreg ri_memreg_strategy;
67 * RDMA Endpoint -- one per transport instance
74 struct rpcrdma_ia *rep_ia;
76 struct ib_qp_init_attr rep_attr;
77 wait_queue_head_t rep_connect_wait;
78 struct ib_sge rep_pad; /* holds zeroed pad */
79 struct ib_mr *rep_pad_mr; /* holds zeroed pad */
80 void (*rep_func)(struct rpcrdma_ep *);
81 struct rpc_xprt *rep_xprt; /* for rep_func */
82 struct rdma_conn_param rep_remote_cma;
83 struct sockaddr_storage rep_remote_addr;
86 #define INIT_CQCOUNT(ep) atomic_set(&(ep)->rep_cqcount, (ep)->rep_cqinit)
87 #define DECR_CQCOUNT(ep) atomic_sub_return(1, &(ep)->rep_cqcount)
90 * struct rpcrdma_rep -- this structure encapsulates state required to recv
91 * and complete a reply, asychronously. It needs several pieces of
93 * o recv buffer (posted to provider)
94 * o ib_sge (also donated to provider)
95 * o status of reply (length, success or not)
96 * o bookkeeping state to get run by tasklet (list, etc)
98 * These are allocated during initialization, per-transport instance;
99 * however, the tasklet execution list itself is global, as it should
100 * always be pretty short.
102 * N of these are associated with a transport instance, and stored in
103 * struct rpcrdma_buffer. N is the max number of outstanding requests.
106 /* temporary static scatter/gather max */
107 #define RPCRDMA_MAX_DATA_SEGS (8) /* max scatter/gather */
108 #define RPCRDMA_MAX_SEGS (RPCRDMA_MAX_DATA_SEGS + 2) /* head+tail = 2 */
109 #define MAX_RPCRDMAHDR (\
110 /* max supported RPC/RDMA header */ \
111 sizeof(struct rpcrdma_msg) + (2 * sizeof(u32)) + \
112 (sizeof(struct rpcrdma_read_chunk) * RPCRDMA_MAX_SEGS) + sizeof(u32))
114 struct rpcrdma_buffer;
117 unsigned int rr_len; /* actual received reply length */
118 struct rpcrdma_buffer *rr_buffer; /* home base for this structure */
119 struct rpc_xprt *rr_xprt; /* needed for request/reply matching */
120 void (*rr_func)(struct rpcrdma_rep *);/* called by tasklet in softint */
121 struct list_head rr_list; /* tasklet list */
122 wait_queue_head_t rr_unbind; /* optional unbind wait */
123 struct ib_sge rr_iov; /* for posting */
124 struct ib_mr *rr_handle; /* handle for mem in rr_iov */
125 char rr_base[MAX_RPCRDMAHDR]; /* minimal inline receive buffer */
129 * struct rpcrdma_req -- structure central to the request/reply sequence.
131 * N of these are associated with a transport instance, and stored in
132 * struct rpcrdma_buffer. N is the max number of outstanding requests.
134 * It includes pre-registered buffer memory for send AND recv.
135 * The recv buffer, however, is not owned by this structure, and
136 * is "donated" to the hardware when a recv is posted. When a
137 * reply is handled, the recv buffer used is given back to the
138 * struct rpcrdma_req associated with the request.
140 * In addition to the basic memory, this structure includes an array
141 * of iovs for send operations. The reason is that the iovs passed to
142 * ib_post_{send,recv} must not be modified until the work request
146 * o RPCRDMA_MAX_SEGS is the max number of addressible chunk elements we
147 * marshal. The number needed varies depending on the iov lists that
148 * are passed to us, the memory registration mode we are in, and if
149 * physical addressing is used, the layout.
152 struct rpcrdma_mr_seg { /* chunk descriptors */
153 union { /* chunk memory handles */
154 struct ib_mr *rl_mr; /* if registered directly */
155 struct rpcrdma_mw { /* if registered from region */
160 struct list_head mw_list;
163 u64 mr_base; /* registration result */
164 u32 mr_rkey; /* registration result */
165 u32 mr_len; /* length of chunk or segment */
166 int mr_nsegs; /* number of segments in chunk or 0 */
167 enum dma_data_direction mr_dir; /* segment mapping direction */
168 dma_addr_t mr_dma; /* segment mapping address */
169 size_t mr_dmalen; /* segment mapping length */
170 struct page *mr_page; /* owning page, if any */
171 char *mr_offset; /* kva if no page, else offset */
175 size_t rl_size; /* actual length of buffer */
176 unsigned int rl_niovs; /* 0, 2 or 4 */
177 unsigned int rl_nchunks; /* non-zero if chunks */
178 struct rpcrdma_buffer *rl_buffer; /* home base for this structure */
179 struct rpcrdma_rep *rl_reply;/* holder for reply buffer */
180 struct rpcrdma_mr_seg rl_segments[RPCRDMA_MAX_SEGS];/* chunk segments */
181 struct ib_sge rl_send_iov[4]; /* for active requests */
182 struct ib_sge rl_iov; /* for posting */
183 struct ib_mr *rl_handle; /* handle for mem in rl_iov */
184 char rl_base[MAX_RPCRDMAHDR]; /* start of actual buffer */
185 __u32 rl_xdr_buf[0]; /* start of returned rpc rq_buffer */
187 #define rpcr_to_rdmar(r) \
188 container_of((r)->rq_buffer, struct rpcrdma_req, rl_xdr_buf[0])
191 * struct rpcrdma_buffer -- holds list/queue of pre-registered memory for
192 * inline requests/replies, and client/server credits.
194 * One of these is associated with a transport instance
196 struct rpcrdma_buffer {
197 spinlock_t rb_lock; /* protects indexes */
198 atomic_t rb_credits; /* most recent server credits */
199 unsigned long rb_cwndscale; /* cached framework rpc_cwndscale */
200 int rb_max_requests;/* client max requests */
201 struct list_head rb_mws; /* optional memory windows/fmrs */
203 struct rpcrdma_req **rb_send_bufs;
205 struct rpcrdma_rep **rb_recv_bufs;
208 #define rdmab_to_ia(b) (&container_of((b), struct rpcrdma_xprt, rx_buf)->rx_ia)
211 * Internal structure for transport instance creation. This
212 * exists primarily for modularity.
214 * This data should be set with mount options
216 struct rpcrdma_create_data_internal {
217 struct sockaddr_storage addr; /* RDMA server address */
218 unsigned int max_requests; /* max requests (slots) in flight */
219 unsigned int rsize; /* mount rsize - max read hdr+data */
220 unsigned int wsize; /* mount wsize - max write hdr+data */
221 unsigned int inline_rsize; /* max non-rdma read data payload */
222 unsigned int inline_wsize; /* max non-rdma write data payload */
223 unsigned int padding; /* non-rdma write header padding */
226 #define RPCRDMA_INLINE_READ_THRESHOLD(rq) \
227 (rpcx_to_rdmad(rq->rq_task->tk_xprt).inline_rsize)
229 #define RPCRDMA_INLINE_WRITE_THRESHOLD(rq)\
230 (rpcx_to_rdmad(rq->rq_task->tk_xprt).inline_wsize)
232 #define RPCRDMA_INLINE_PAD_VALUE(rq)\
233 rpcx_to_rdmad(rq->rq_task->tk_xprt).padding
236 * Statistics for RPCRDMA
238 struct rpcrdma_stats {
239 unsigned long read_chunk_count;
240 unsigned long write_chunk_count;
241 unsigned long reply_chunk_count;
243 unsigned long long total_rdma_request;
244 unsigned long long total_rdma_reply;
246 unsigned long long pullup_copy_count;
247 unsigned long long fixup_copy_count;
248 unsigned long hardway_register_count;
249 unsigned long failed_marshal_count;
250 unsigned long bad_reply_count;
254 * RPCRDMA transport -- encapsulates the structures above for
255 * integration with RPC.
257 * The contained structures are embedded, not pointers,
258 * for convenience. This structure need not be visible externally.
260 * It is allocated and initialized during mount, and released
263 struct rpcrdma_xprt {
264 struct rpc_xprt xprt;
265 struct rpcrdma_ia rx_ia;
266 struct rpcrdma_ep rx_ep;
267 struct rpcrdma_buffer rx_buf;
268 struct rpcrdma_create_data_internal rx_data;
269 struct delayed_work rdma_connect;
270 struct rpcrdma_stats rx_stats;
273 #define rpcx_to_rdmax(x) container_of(x, struct rpcrdma_xprt, xprt)
274 #define rpcx_to_rdmad(x) (rpcx_to_rdmax(x)->rx_data)
277 * Interface Adapter calls - xprtrdma/verbs.c
279 int rpcrdma_ia_open(struct rpcrdma_xprt *, struct sockaddr *, int);
280 void rpcrdma_ia_close(struct rpcrdma_ia *);
283 * Endpoint calls - xprtrdma/verbs.c
285 int rpcrdma_ep_create(struct rpcrdma_ep *, struct rpcrdma_ia *,
286 struct rpcrdma_create_data_internal *);
287 int rpcrdma_ep_destroy(struct rpcrdma_ep *, struct rpcrdma_ia *);
288 int rpcrdma_ep_connect(struct rpcrdma_ep *, struct rpcrdma_ia *);
289 int rpcrdma_ep_disconnect(struct rpcrdma_ep *, struct rpcrdma_ia *);
291 int rpcrdma_ep_post(struct rpcrdma_ia *, struct rpcrdma_ep *,
292 struct rpcrdma_req *);
293 int rpcrdma_ep_post_recv(struct rpcrdma_ia *, struct rpcrdma_ep *,
294 struct rpcrdma_rep *);
297 * Buffer calls - xprtrdma/verbs.c
299 int rpcrdma_buffer_create(struct rpcrdma_buffer *, struct rpcrdma_ep *,
301 struct rpcrdma_create_data_internal *);
302 void rpcrdma_buffer_destroy(struct rpcrdma_buffer *);
304 struct rpcrdma_req *rpcrdma_buffer_get(struct rpcrdma_buffer *);
305 void rpcrdma_buffer_put(struct rpcrdma_req *);
306 void rpcrdma_recv_buffer_get(struct rpcrdma_req *);
307 void rpcrdma_recv_buffer_put(struct rpcrdma_rep *);
309 int rpcrdma_register_internal(struct rpcrdma_ia *, void *, int,
310 struct ib_mr **, struct ib_sge *);
311 int rpcrdma_deregister_internal(struct rpcrdma_ia *,
312 struct ib_mr *, struct ib_sge *);
314 int rpcrdma_register_external(struct rpcrdma_mr_seg *,
315 int, int, struct rpcrdma_xprt *);
316 int rpcrdma_deregister_external(struct rpcrdma_mr_seg *,
317 struct rpcrdma_xprt *, void *);
320 * RPC/RDMA connection management calls - xprtrdma/rpc_rdma.c
322 void rpcrdma_conn_func(struct rpcrdma_ep *);
323 void rpcrdma_reply_handler(struct rpcrdma_rep *);
326 * RPC/RDMA protocol calls - xprtrdma/rpc_rdma.c
328 int rpcrdma_marshal_req(struct rpc_rqst *);
330 #endif /* _LINUX_SUNRPC_XPRT_RDMA_H */