xref: /titanic_41/usr/src/uts/common/sys/rds.h (revision c6698ca1e2450f37cc5ee4cdded3e5472088ba41)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
23  */
24 
25 /*
26  * Copyright (c) 2008 Oracle.  All rights reserved.
27  *
28  * This software is available to you under a choice of one of two
29  * licenses.  You may choose to be licensed under the terms of the GNU
30  * General Public License (GPL) Version 2, available from the file
31  * COPYING in the main directory of this source tree, or the
32  * OpenIB.org BSD license below:
33  *
34  *     Redistribution and use in source and binary forms, with or
35  *     without modification, are permitted provided that the following
36  *     conditions are met:
37  *
38  *      - Redistributions of source code must retain the above
39  *        copyright notice, this list of conditions and the following
40  *        disclaimer.
41  *
42  *      - Redistributions in binary form must reproduce the above
43  *        copyright notice, this list of conditions and the following
44  *        disclaimer in the documentation and/or other materials
45  *        provided with the distribution.
46  *
47  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
48  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
49  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
50  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
51  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
52  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
53  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
54  * SOFTWARE.
55  *
56  */
57 
58 /*
59  * Include this file if the application uses rdsv3 sockets.
60  */
61 
62 /*
63  * This file contains definitions from the ofed rds.h and rds_rdma.h
64  * header file.
65  */
66 #ifndef _RDSV3_RDS_H
67 #define	_RDSV3_RDS_H
68 
69 #include <sys/types.h>
70 
71 #ifdef __cplusplus
72 extern "C" {
73 #endif
74 
75 #define	RDS_IB_ABI_VERSION		0x301
76 
77 #define	AF_RDS				AF_INET_OFFLOAD
78 #define	PF_RDS				AF_INET_OFFLOAD
79 
80 #define	SOL_RDS				272
81 
82 /*
83  * setsockopt/getsockopt for SOL_RDS
84  */
85 #define	RDSV3_CANCEL_SENT_TO		1
86 #define	RDSV3_GET_MR			2
87 #define	RDSV3_FREE_MR			3
88 /* deprecated: RDS_BARRIER 4 */
89 #define	RDSV3_RECVERR			5
90 #define	RDSV3_CONG_MONITOR		6
91 
92 /*
93  * Control message types for SOL_RDS.
94  *
95  * RDS_CMSG_RDMA_ARGS (sendmsg)
96  *	Request a RDMA transfer to/from the specified
97  *	memory ranges.
98  *	The cmsg_data is a struct rdsv3_rdma_args.
99  * RDS_CMSG_RDMA_DEST (recvmsg, sendmsg)
100  *	Kernel informs application about intended
101  *	source/destination of a RDMA transfer
102  * RDS_CMSG_RDMA_MAP (sendmsg)
103  *	Application asks kernel to map the given
104  *	memory range into a IB MR, and send the
105  *	R_Key along in an RDS extension header.
106  *	The cmsg_data is a struct rdsv3_get_mr_args,
107  *	the same as for the GET_MR setsockopt.
108  * RDS_CMSG_RDMA_STATUS (recvmsg)
109  *	Returns the status of a completed RDMA operation.
110  */
111 #define	RDSV3_CMSG_RDMA_ARGS		1
112 #define	RDSV3_CMSG_RDMA_DEST		2
113 #define	RDSV3_CMSG_RDMA_MAP		3
114 #define	RDSV3_CMSG_RDMA_STATUS		4
115 #define	RDSV3_CMSG_CONG_UPDATE		5
116 
117 /*
118  * RDMA related types
119  */
120 
121 /*
122  * This encapsulates a remote memory location.
123  * In the current implementation, it contains the R_Key
124  * of the remote memory region, and the offset into it
125  * (so that the application does not have to worry about
126  * alignment).
127  */
128 typedef uint64_t	rdsv3_rdma_cookie_t;
129 
130 struct rdsv3_iovec {
131 	uint64_t	addr;
132 	uint64_t	bytes;
133 };
134 
135 struct rdsv3_get_mr_args {
136 	struct rdsv3_iovec vec;
137 	uint64_t	cookie_addr;
138 	uint64_t	flags;
139 };
140 
141 struct rdsv3_free_mr_args {
142 	rdsv3_rdma_cookie_t cookie;
143 	uint64_t	flags;
144 };
145 
146 struct rdsv3_rdma_args {
147 	rdsv3_rdma_cookie_t cookie;
148 	struct rdsv3_iovec remote_vec;
149 	uint64_t	local_vec_addr;
150 	uint64_t	nr_local;
151 	uint64_t	flags;
152 	uint64_t	user_token;
153 };
154 
155 struct rdsv3_rdma_notify {
156 	uint64_t	user_token;
157 	int32_t		status;
158 };
159 
160 #define	RDSV3_RDMA_SUCCESS	0
161 #define	RDSV3_RDMA_REMOTE_ERROR	1
162 #define	RDSV3_RDMA_CANCELED	2
163 #define	RDSV3_RDMA_DROPPED	3
164 #define	RDSV3_RDMA_OTHER_ERROR	4
165 
166 /*
167  * Common set of flags for all RDMA related structs
168  */
169 #define	RDSV3_RDMA_READWRITE	0x0001
170 #define	RDSV3_RDMA_FENCE	0x0002	/* use FENCE for immediate send */
171 #define	RDSV3_RDMA_INVALIDATE	0x0004	/* invalidate R_Key after freeing MR */
172 #define	RDSV3_RDMA_USE_ONCE	0x0008	/* free MR after use */
173 #define	RDSV3_RDMA_DONTWAIT	0x0010	/* Don't wait in SET_BARRIER */
174 #define	RDSV3_RDMA_NOTIFY_ME	0x0020	/* Notify when operation completes */
175 
176 /*
177  * Congestion monitoring.
178  * Congestion control in RDS happens at the host connection
179  * level by exchanging a bitmap marking congested ports.
180  * By default, a process sleeping in poll() is always woken
181  * up when the congestion map is updated.
182  * With explicit monitoring, an application can have more
183  * fine-grained control.
184  * The application installs a 64bit mask value in the socket,
185  * where each bit corresponds to a group of ports.
186  * When a congestion update arrives, RDS checks the set of
187  * ports that are now uncongested against the list bit mask
188  * installed in the socket, and if they overlap, we queue a
189  * cong_notification on the socket.
190  *
191  * To install the congestion monitor bitmask, use RDS_CONG_MONITOR
192  * with the 64bit mask.
193  * Congestion updates are received via RDS_CMSG_CONG_UPDATE
194  * control messages.
195  *
196  * The correspondence between bits and ports is
197  *	1 << (portnum % 64)
198  */
199 #define	RDSV3_CONG_MONITOR_SIZE	64
200 #define	RDSV3_CONG_MONITOR_BIT(port)	\
201 	(((unsigned int) port) % RDSV3_CONG_MONITOR_SIZE)
202 #define	RDSV3_CONG_MONITOR_MASK(port) (1ULL << RDSV3_CONG_MONITOR_BIT(port))
203 
204 /* rds-info related */
205 
206 #define	RDSV3_INFO_FIRST		10000
207 #define	RDSV3_INFO_COUNTERS		10000
208 #define	RDSV3_INFO_CONNECTIONS		10001
209 /* 10002 aka RDS_INFO_FLOWS is deprecated */
210 #define	RDSV3_INFO_SEND_MESSAGES	10003
211 #define	RDSV3_INFO_RETRANS_MESSAGES	10004
212 #define	RDSV3_INFO_RECV_MESSAGES	10005
213 #define	RDSV3_INFO_SOCKETS		10006
214 #define	RDSV3_INFO_TCP_SOCKETS		10007
215 #define	RDSV3_INFO_IB_CONNECTIONS	10008
216 #define	RDSV3_INFO_CONNECTION_STATS	10009
217 #define	RDSV3_INFO_IWARP_CONNECTIONS	10010
218 #define	RDSV3_INFO_LAST			10010
219 
220 #ifndef __lock_lint
221 #pragma pack(1)
222 struct rdsv3_info_counter {
223 	uint8_t	name[32];
224 	uint64_t	value;
225 } __attribute__((packed));
226 #pragma pack()
227 #else
228 struct rdsv3_info_counter {
229 	uint8_t	name[32];
230 	uint64_t	value;
231 };
232 #endif
233 
234 #define	RDSV3_INFO_CONNECTION_FLAG_SENDING	0x01
235 #define	RDSV3_INFO_CONNECTION_FLAG_CONNECTING	0x02
236 #define	RDSV3_INFO_CONNECTION_FLAG_CONNECTED	0x04
237 
238 #ifndef __lock_lint
239 #pragma pack(1)
240 struct rdsv3_info_connection {
241 	uint64_t	next_tx_seq;
242 	uint64_t	next_rx_seq;
243 	uint32_t	laddr;			/* network order */
244 	uint32_t	faddr;			/* network order */
245 	uint8_t		transport[15];		/* null term ascii */
246 	uint8_t		flags;
247 } __attribute__((packed));
248 #pragma pack()
249 #else
250 struct rdsv3_info_connection {
251 	uint64_t	next_tx_seq;
252 	uint64_t	next_rx_seq;
253 	uint32_t	laddr;			/* network order */
254 	uint32_t	faddr;			/* network order */
255 	uint8_t		transport[15];		/* null term ascii */
256 	uint8_t		flags;
257 };
258 #endif
259 
260 #ifndef __lock_lint
261 #pragma pack(1)
262 struct rdsv3_info_flow {
263 	uint32_t	laddr;			/* network order */
264 	uint32_t	faddr;			/* network order */
265 	uint32_t	bytes;
266 	uint16_t	lport;			/* network order */
267 	uint16_t	fport;			/* network order */
268 } __attribute__((packed));
269 #pragma pack()
270 #else
271 struct rdsv3_info_flow {
272 	uint32_t	laddr;			/* network order */
273 	uint32_t	faddr;			/* network order */
274 	uint32_t	bytes;
275 	uint16_t	lport;			/* network order */
276 	uint16_t	fport;			/* network order */
277 };
278 #endif
279 
280 #define	RDSV3_INFO_MESSAGE_FLAG_ACK		0x01
281 #define	RDSV3_INFO_MESSAGE_FLAG_FAST_ACK		0x02
282 
283 #ifndef __lock_lint
284 #pragma pack(1)
285 struct rdsv3_info_message {
286 	uint64_t	seq;
287 	uint32_t	len;
288 	uint32_t	laddr;			/* network order */
289 	uint32_t	faddr;			/* network order */
290 	uint16_t	lport;			/* network order */
291 	uint16_t	fport;			/* network order */
292 	uint8_t		flags;
293 } __attribute__((packed));
294 #pragma pack()
295 #else
296 struct rdsv3_info_message {
297 	uint64_t	seq;
298 	uint32_t	len;
299 	uint32_t	laddr;			/* network order */
300 	uint32_t	faddr;			/* network order */
301 	uint16_t	lport;			/* network order */
302 	uint16_t	fport;			/* network order */
303 	uint8_t		flags;
304 };
305 #endif
306 
307 #ifndef __lock_lint
308 #pragma pack(1)
309 struct rdsv3_info_socket {
310 	uint32_t	sndbuf;
311 	uint32_t	bound_addr;		/* network order */
312 	uint32_t	connected_addr;		/* network order */
313 	uint16_t	bound_port;		/* network order */
314 	uint16_t	connected_port;		/* network order */
315 	uint32_t	rcvbuf;
316 	uint64_t	inum;
317 } __attribute__((packed));
318 #pragma pack()
319 #else
320 struct rdsv3_info_socket {
321 	uint32_t	sndbuf;
322 	uint32_t	bound_addr;		/* network order */
323 	uint32_t	connected_addr;		/* network order */
324 	uint16_t	bound_port;		/* network order */
325 	uint16_t	connected_port;		/* network order */
326 	uint32_t	rcvbuf;
327 	uint64_t	inum;
328 };
329 #endif
330 
331 #ifndef __lock_lint
332 #pragma pack(1)
333 struct rdsv3_info_socket_v1 {
334 	uint32_t	sndbuf;
335 	uint32_t	bound_addr;		/* network order */
336 	uint32_t	connected_addr;		/* network order */
337 	uint16_t	bound_port;		/* network order */
338 	uint16_t	connected_port;		/* network order */
339 	uint32_t	rcvbuf;
340 } __attribute__((packed));
341 #pragma pack()
342 #else
343 struct rdsv3_info_socket_v1 {
344 	uint32_t	sndbuf;
345 	uint32_t	bound_addr;		/* network order */
346 	uint32_t	connected_addr;		/* network order */
347 	uint16_t	bound_port;		/* network order */
348 	uint16_t	connected_port;		/* network order */
349 	uint32_t	rcvbuf;
350 };
351 #endif
352 
353 #define	RDS_IB_GID_LEN	16
354 struct rdsv3_info_rdma_connection {
355 	uint32_t	src_addr;		/* network order */
356 	uint32_t	dst_addr;		/* network order */
357 	uint8_t		src_gid[RDS_IB_GID_LEN];
358 	uint8_t		dst_gid[RDS_IB_GID_LEN];
359 
360 	uint32_t	max_send_wr;
361 	uint32_t	max_recv_wr;
362 	uint32_t	max_send_sge;
363 	uint32_t	rdma_mr_max;
364 	uint32_t	rdma_mr_size;
365 };
366 
367 #define	rdsv3_info_ib_connection rdsv3_info_rdma_connection
368 #define	rdma_fmr_max rdma_mr_max
369 #define	rdma_fmr_size rdma_mr_size
370 
371 #ifdef	__cplusplus
372 }
373 #endif
374 
375 #endif /* _RDSV3_RDS_H */
376