tjh.dev
Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
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#define RPCRDMA_VERSION 1
67
68/*
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
71 * completes.
72 */
73struct svc_rdma_op_ctxt {
74 struct svc_rdma_op_ctxt *read_hdr;
75 struct svc_rdma_fastreg_mr *frmr;
76 int hdr_count;
77 struct xdr_buf arg;
78 struct list_head dto_q;
79 enum ib_wr_opcode wr_op;
80 enum ib_wc_status wc_status;
81 u32 byte_len;
82 struct svcxprt_rdma *xprt;
83 unsigned long flags;
84 enum dma_data_direction direction;
85 int count;
86 struct ib_sge sge[RPCSVC_MAXPAGES];
87 struct page *pages[RPCSVC_MAXPAGES];
88};
89
90/*
91 * NFS_ requests are mapped on the client side by the chunk lists in
92 * the RPCRDMA header. During the fetching of the RPC from the client
93 * and the writing of the reply to the client, the memory in the
94 * client and the memory in the server must be mapped as contiguous
95 * vaddr/len for access by the hardware. These data strucures keep
96 * these mappings.
97 *
98 * For an RDMA_WRITE, the 'sge' maps the RPC REPLY. For RDMA_READ, the
99 * 'sge' in the svc_rdma_req_map maps the server side RPC reply and the
100 * 'ch' field maps the read-list of the RPCRDMA header to the 'sge'
101 * mapping of the reply.
102 */
103struct svc_rdma_chunk_sge {
104 int start; /* sge no for this chunk */
105 int count; /* sge count for this chunk */
106};
107struct svc_rdma_fastreg_mr {
108 struct ib_mr *mr;
109 void *kva;
110 struct ib_fast_reg_page_list *page_list;
111 int page_list_len;
112 unsigned long access_flags;
113 unsigned long map_len;
114 enum dma_data_direction direction;
115 struct list_head frmr_list;
116};
117struct svc_rdma_req_map {
118 struct svc_rdma_fastreg_mr *frmr;
119 unsigned long count;
120 union {
121 struct kvec sge[RPCSVC_MAXPAGES];
122 struct svc_rdma_chunk_sge ch[RPCSVC_MAXPAGES];
123 };
124};
125#define RDMACTXT_F_FAST_UNREG 1
126#define RDMACTXT_F_LAST_CTXT 2
127
128#define SVCRDMA_DEVCAP_FAST_REG 1 /* fast mr registration */
129#define SVCRDMA_DEVCAP_READ_W_INV 2 /* read w/ invalidate */
130
131struct svcxprt_rdma {
132 struct svc_xprt sc_xprt; /* SVC transport structure */
133 struct rdma_cm_id *sc_cm_id; /* RDMA connection id */
134 struct list_head sc_accept_q; /* Conn. waiting accept */
135 int sc_ord; /* RDMA read limit */
136 int sc_max_sge;
137
138 int sc_sq_depth; /* Depth of SQ */
139 atomic_t sc_sq_count; /* Number of SQ WR on queue */
140
141 int sc_max_requests; /* Depth of RQ */
142 int sc_max_req_size; /* Size of each RQ WR buf */
143
144 struct ib_pd *sc_pd;
145
146 atomic_t sc_dma_used;
147 atomic_t sc_ctxt_used;
148 struct list_head sc_rq_dto_q;
149 spinlock_t sc_rq_dto_lock;
150 struct ib_qp *sc_qp;
151 struct ib_cq *sc_rq_cq;
152 struct ib_cq *sc_sq_cq;
153 struct ib_mr *sc_phys_mr; /* MR for server memory */
154 u32 sc_dev_caps; /* distilled device caps */
155 u32 sc_dma_lkey; /* local dma key */
156 unsigned int sc_frmr_pg_list_len;
157 struct list_head sc_frmr_q;
158 spinlock_t sc_frmr_q_lock;
159
160 spinlock_t sc_lock; /* transport lock */
161
162 wait_queue_head_t sc_send_wait; /* SQ exhaustion waitlist */
163 unsigned long sc_flags;
164 struct list_head sc_dto_q; /* DTO tasklet I/O pending Q */
165 struct list_head sc_read_complete_q;
166 struct work_struct sc_work;
167};
168/* sc_flags */
169#define RDMAXPRT_RQ_PENDING 1
170#define RDMAXPRT_SQ_PENDING 2
171#define RDMAXPRT_CONN_PENDING 3
172
173#define RPCRDMA_LISTEN_BACKLOG 10
174/* The default ORD value is based on two outstanding full-size writes with a
175 * page size of 4k, or 32k * 2 ops / 4k = 16 outstanding RDMA_READ. */
176#define RPCRDMA_ORD (64/4)
177#define RPCRDMA_SQ_DEPTH_MULT 8
178#define RPCRDMA_MAX_THREADS 16
179#define RPCRDMA_MAX_REQUESTS 16
180#define RPCRDMA_MAX_REQ_SIZE 4096
181
182/* svc_rdma_marshal.c */
183extern void svc_rdma_rcl_chunk_counts(struct rpcrdma_read_chunk *,
184 int *, int *);
185extern int svc_rdma_xdr_decode_req(struct rpcrdma_msg **, struct svc_rqst *);
186extern int svc_rdma_xdr_decode_deferred_req(struct svc_rqst *);
187extern int svc_rdma_xdr_encode_error(struct svcxprt_rdma *,
188 struct rpcrdma_msg *,
189 enum rpcrdma_errcode, u32 *);
190extern void svc_rdma_xdr_encode_write_list(struct rpcrdma_msg *, int);
191extern void svc_rdma_xdr_encode_reply_array(struct rpcrdma_write_array *, int);
192extern void svc_rdma_xdr_encode_array_chunk(struct rpcrdma_write_array *, int,
193 __be32, __be64, u32);
194extern void svc_rdma_xdr_encode_reply_header(struct svcxprt_rdma *,
195 struct rpcrdma_msg *,
196 struct rpcrdma_msg *,
197 enum rpcrdma_proc);
198extern int svc_rdma_xdr_get_reply_hdr_len(struct rpcrdma_msg *);
199
200/* svc_rdma_recvfrom.c */
201extern int svc_rdma_recvfrom(struct svc_rqst *);
202
203/* svc_rdma_sendto.c */
204extern int svc_rdma_sendto(struct svc_rqst *);
205
206/* svc_rdma_transport.c */
207extern int svc_rdma_send(struct svcxprt_rdma *, struct ib_send_wr *);
208extern void svc_rdma_send_error(struct svcxprt_rdma *, struct rpcrdma_msg *,
209 enum rpcrdma_errcode);
210struct page *svc_rdma_get_page(void);
211extern int svc_rdma_post_recv(struct svcxprt_rdma *);
212extern int svc_rdma_create_listen(struct svc_serv *, int, struct sockaddr *);
213extern struct svc_rdma_op_ctxt *svc_rdma_get_context(struct svcxprt_rdma *);
214extern void svc_rdma_put_context(struct svc_rdma_op_ctxt *, int);
215extern void svc_rdma_unmap_dma(struct svc_rdma_op_ctxt *ctxt);
216extern struct svc_rdma_req_map *svc_rdma_get_req_map(void);
217extern void svc_rdma_put_req_map(struct svc_rdma_req_map *);
218extern int svc_rdma_fastreg(struct svcxprt_rdma *, struct svc_rdma_fastreg_mr *);
219extern struct svc_rdma_fastreg_mr *svc_rdma_get_frmr(struct svcxprt_rdma *);
220extern void svc_rdma_put_frmr(struct svcxprt_rdma *,
221 struct svc_rdma_fastreg_mr *);
222extern void svc_sq_reap(struct svcxprt_rdma *);
223extern void svc_rq_reap(struct svcxprt_rdma *);
224extern struct svc_xprt_class svc_rdma_class;
225extern void svc_rdma_prep_reply_hdr(struct svc_rqst *);
226
227/* svc_rdma.c */
228extern int svc_rdma_init(void);
229extern void svc_rdma_cleanup(void);
230
231/*
232 * Returns the address of the first read chunk or <nul> if no read chunk is
233 * present
234 */
235static inline struct rpcrdma_read_chunk *
236svc_rdma_get_read_chunk(struct rpcrdma_msg *rmsgp)
237{
238 struct rpcrdma_read_chunk *ch =
239 (struct rpcrdma_read_chunk *)&rmsgp->rm_body.rm_chunks[0];
240
241 if (ch->rc_discrim == 0)
242 return NULL;
243
244 return ch;
245}
246
247/*
248 * Returns the address of the first read write array element or <nul> if no
249 * write array list is present
250 */
251static inline struct rpcrdma_write_array *
252svc_rdma_get_write_array(struct rpcrdma_msg *rmsgp)
253{
254 if (rmsgp->rm_body.rm_chunks[0] != 0
255 || rmsgp->rm_body.rm_chunks[1] == 0)
256 return NULL;
257
258 return (struct rpcrdma_write_array *)&rmsgp->rm_body.rm_chunks[1];
259}
260
261/*
262 * Returns the address of the first reply array element or <nul> if no
263 * reply array is present
264 */
265static inline struct rpcrdma_write_array *
266svc_rdma_get_reply_array(struct rpcrdma_msg *rmsgp)
267{
268 struct rpcrdma_read_chunk *rch;
269 struct rpcrdma_write_array *wr_ary;
270 struct rpcrdma_write_array *rp_ary;
271
272 /* XXX: Need to fix when reply list may occur with read-list and/or
273 * write list */
274 if (rmsgp->rm_body.rm_chunks[0] != 0 ||
275 rmsgp->rm_body.rm_chunks[1] != 0)
276 return NULL;
277
278 rch = svc_rdma_get_read_chunk(rmsgp);
279 if (rch) {
280 while (rch->rc_discrim)
281 rch++;
282
283 /* The reply list follows an empty write array located
284 * at 'rc_position' here. The reply array is at rc_target.
285 */
286 rp_ary = (struct rpcrdma_write_array *)&rch->rc_target;
287
288 goto found_it;
289 }
290
291 wr_ary = svc_rdma_get_write_array(rmsgp);
292 if (wr_ary) {
293 rp_ary = (struct rpcrdma_write_array *)
294 &wr_ary->
295 wc_array[ntohl(wr_ary->wc_nchunks)].wc_target.rs_length;
296
297 goto found_it;
298 }
299
300 /* No read list, no write list */
301 rp_ary = (struct rpcrdma_write_array *)
302 &rmsgp->rm_body.rm_chunks[2];
303
304 found_it:
305 if (rp_ary->wc_discrim == 0)
306 return NULL;
307
308 return rp_ary;
309}
310#endif