1/*
2 * Copyright (C) 2009 Fen Systems Ltd <mbrown@fensystems.co.uk>.
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 *
9 *   Redistributions of source code must retain the above copyright
10 *   notice, this list of conditions and the following disclaimer.
11 *
12 *   Redistributions in binary form must reproduce the above copyright
13 *   notice, this list of conditions and the following disclaimer in
14 *   the documentation and/or other materials provided with the
15 *   distribution.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
18 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
19 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
20 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
21 * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
22 * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
23 * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
24 * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
26 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
27 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
28 * OF THE POSSIBILITY OF SUCH DAMAGE.
29 */
30
31FILE_LICENCE ( BSD2 );
32
33#include <stdlib.h>
34#include <string.h>
35#include <errno.h>
36#include <gpxe/iobuf.h>
37#include <gpxe/xfer.h>
38#include <gpxe/process.h>
39#include <gpxe/infiniband.h>
40#include <gpxe/ib_cm.h>
41#include <gpxe/ib_cmrc.h>
42
43/**
44 * @file
45 *
46 * Infiniband Communication-managed Reliable Connections
47 *
48 */
49
50/** CMRC number of send WQEs
51 *
52 * This is a policy decision.
53 */
54#define IB_CMRC_NUM_SEND_WQES 4
55
56/** CMRC number of receive WQEs
57 *
58 * This is a policy decision.
59 */
60#define IB_CMRC_NUM_RECV_WQES 2
61
62/** CMRC number of completion queue entries
63 *
64 * This is a policy decision
65 */
66#define IB_CMRC_NUM_CQES 8
67
68/** An Infiniband Communication-Managed Reliable Connection */
69struct ib_cmrc_connection {
70	/** Reference count */
71	struct refcnt refcnt;
72	/** Data transfer interface */
73	struct xfer_interface xfer;
74	/** Infiniband device */
75	struct ib_device *ibdev;
76	/** Completion queue */
77	struct ib_completion_queue *cq;
78	/** Queue pair */
79	struct ib_queue_pair *qp;
80	/** Connection */
81	struct ib_connection *conn;
82	/** Destination GID */
83	struct ib_gid dgid;
84	/** Service ID */
85	struct ib_gid_half service_id;
86	/** QP is connected */
87	int connected;
88	/** Shutdown process */
89	struct process shutdown;
90};
91
92/**
93 * Shut down CMRC connection gracefully
94 *
95 * @v process		Process
96 *
97 * The Infiniband data structures are not reference-counted or
98 * guarded.  It is therefore unsafe to shut them down while we may be
99 * in the middle of a callback from the Infiniband stack (e.g. in a
100 * receive completion handler).
101 *
102 * This shutdown process will run some time after the call to
103 * ib_cmrc_close(), after control has returned out of the Infiniband
104 * core, and will shut down the Infiniband interfaces cleanly.
105 *
106 * The shutdown process holds an implicit reference on the CMRC
107 * connection, ensuring that the structure is not freed before the
108 * shutdown process has run.
109 */
110static void ib_cmrc_shutdown ( struct process *process ) {
111	struct ib_cmrc_connection *cmrc =
112		container_of ( process, struct ib_cmrc_connection, shutdown );
113
114	DBGC ( cmrc, "CMRC %p shutting down\n", cmrc );
115
116	/* Shut down Infiniband interface */
117	ib_destroy_conn ( cmrc->ibdev, cmrc->qp, cmrc->conn );
118	ib_destroy_qp ( cmrc->ibdev, cmrc->qp );
119	ib_destroy_cq ( cmrc->ibdev, cmrc->cq );
120	ib_close ( cmrc->ibdev );
121
122	/* Remove process from run queue */
123	process_del ( &cmrc->shutdown );
124
125	/* Drop the remaining reference */
126	ref_put ( &cmrc->refcnt );
127}
128
129/**
130 * Close CMRC connection
131 *
132 * @v cmrc		Communication-Managed Reliable Connection
133 * @v rc		Reason for close
134 */
135static void ib_cmrc_close ( struct ib_cmrc_connection *cmrc, int rc ) {
136
137	/* Close data transfer interface */
138	xfer_nullify ( &cmrc->xfer );
139	xfer_close ( &cmrc->xfer, rc );
140
141	/* Schedule shutdown process */
142	process_add ( &cmrc->shutdown );
143}
144
145/**
146 * Handle change of CMRC connection status
147 *
148 * @v ibdev		Infiniband device
149 * @v qp		Queue pair
150 * @v conn		Connection
151 * @v rc_cm		Connection status code
152 * @v private_data	Private data, if available
153 * @v private_data_len	Length of private data
154 */
155static void ib_cmrc_changed ( struct ib_device *ibdev __unused,
156			      struct ib_queue_pair *qp,
157			      struct ib_connection *conn __unused, int rc_cm,
158			      void *private_data, size_t private_data_len ) {
159	struct ib_cmrc_connection *cmrc = ib_qp_get_ownerdata ( qp );
160	int rc_xfer;
161
162	/* Record connection status */
163	if ( rc_cm == 0 ) {
164		DBGC ( cmrc, "CMRC %p connected\n", cmrc );
165		cmrc->connected = 1;
166	} else {
167		DBGC ( cmrc, "CMRC %p disconnected: %s\n",
168		       cmrc, strerror ( rc_cm ) );
169		cmrc->connected = 0;
170	}
171
172	/* Pass up any private data */
173	DBGC2 ( cmrc, "CMRC %p received private data:\n", cmrc );
174	DBGC2_HDA ( cmrc, 0, private_data, private_data_len );
175	if ( private_data &&
176	     ( rc_xfer = xfer_deliver_raw ( &cmrc->xfer, private_data,
177					    private_data_len ) ) != 0 ) {
178		DBGC ( cmrc, "CMRC %p could not deliver private data: %s\n",
179		       cmrc, strerror ( rc_xfer ) );
180		ib_cmrc_close ( cmrc, rc_xfer );
181		return;
182	}
183
184	/* If we are disconnected, close the upper connection */
185	if ( rc_cm != 0 ) {
186		ib_cmrc_close ( cmrc, rc_cm );
187		return;
188	}
189}
190
191/** CMRC connection operations */
192static struct ib_connection_operations ib_cmrc_conn_op = {
193	.changed = ib_cmrc_changed,
194};
195
196/**
197 * Handle CMRC send completion
198 *
199 * @v ibdev		Infiniband device
200 * @v qp		Queue pair
201 * @v iobuf		I/O buffer
202 * @v rc		Completion status code
203 */
204static void ib_cmrc_complete_send ( struct ib_device *ibdev __unused,
205				    struct ib_queue_pair *qp,
206				    struct io_buffer *iobuf, int rc ) {
207	struct ib_cmrc_connection *cmrc = ib_qp_get_ownerdata ( qp );
208
209	/* Free the completed I/O buffer */
210	free_iob ( iobuf );
211
212	/* Close the connection on any send errors */
213	if ( rc != 0 ) {
214		DBGC ( cmrc, "CMRC %p send error: %s\n",
215		       cmrc, strerror ( rc ) );
216		ib_cmrc_close ( cmrc, rc );
217		return;
218	}
219}
220
221/**
222 * Handle CMRC receive completion
223 *
224 * @v ibdev		Infiniband device
225 * @v qp		Queue pair
226 * @v av		Address vector, or NULL
227 * @v iobuf		I/O buffer
228 * @v rc		Completion status code
229 */
230static void ib_cmrc_complete_recv ( struct ib_device *ibdev __unused,
231				    struct ib_queue_pair *qp,
232				    struct ib_address_vector *av __unused,
233				    struct io_buffer *iobuf, int rc ) {
234	struct ib_cmrc_connection *cmrc = ib_qp_get_ownerdata ( qp );
235
236	/* Close the connection on any receive errors */
237	if ( rc != 0 ) {
238		DBGC ( cmrc, "CMRC %p receive error: %s\n",
239		       cmrc, strerror ( rc ) );
240		free_iob ( iobuf );
241		ib_cmrc_close ( cmrc, rc );
242		return;
243	}
244
245	DBGC2 ( cmrc, "CMRC %p received:\n", cmrc );
246	DBGC2_HDA ( cmrc, 0, iobuf->data, iob_len ( iobuf ) );
247
248	/* Pass up data */
249	if ( ( rc = xfer_deliver_iob ( &cmrc->xfer, iobuf ) ) != 0 ) {
250		DBGC ( cmrc, "CMRC %p could not deliver data: %s\n",
251		       cmrc, strerror ( rc ) );
252		ib_cmrc_close ( cmrc, rc );
253		return;
254	}
255}
256
257/** Infiniband CMRC completion operations */
258static struct ib_completion_queue_operations ib_cmrc_completion_ops = {
259	.complete_send = ib_cmrc_complete_send,
260	.complete_recv = ib_cmrc_complete_recv,
261};
262
263/**
264 * Send data via CMRC
265 *
266 * @v xfer		Data transfer interface
267 * @v iobuf		Datagram I/O buffer
268 * @v meta		Data transfer metadata
269 * @ret rc		Return status code
270 */
271static int ib_cmrc_xfer_deliver_iob ( struct xfer_interface *xfer,
272				      struct io_buffer *iobuf,
273				      struct xfer_metadata *meta __unused ) {
274	struct ib_cmrc_connection *cmrc =
275		container_of ( xfer, struct ib_cmrc_connection, xfer );
276	int rc;
277
278	/* If no connection has yet been attempted, send this datagram
279	 * as the CM REQ private data.  Otherwise, send it via the QP.
280	 */
281	if ( ! cmrc->connected ) {
282
283		/* Abort if we have already sent a CM connection request */
284		if ( cmrc->conn ) {
285			DBGC ( cmrc, "CMRC %p attempt to send before "
286			       "connection is complete\n", cmrc );
287			rc = -EIO;
288			goto out;
289		}
290
291		/* Send via CM connection request */
292		cmrc->conn = ib_create_conn ( cmrc->ibdev, cmrc->qp,
293					      &cmrc->dgid, &cmrc->service_id,
294					      iobuf->data, iob_len ( iobuf ),
295					      &ib_cmrc_conn_op );
296		if ( ! cmrc->conn ) {
297			DBGC ( cmrc, "CMRC %p could not connect\n", cmrc );
298			rc = -ENOMEM;
299			goto out;
300		}
301
302	} else {
303
304		/* Send via QP */
305		if ( ( rc = ib_post_send ( cmrc->ibdev, cmrc->qp, NULL,
306					   iob_disown ( iobuf ) ) ) != 0 ) {
307			DBGC ( cmrc, "CMRC %p could not send: %s\n",
308			       cmrc, strerror ( rc ) );
309			goto out;
310		}
311
312	}
313	return 0;
314
315 out:
316	/* Free the I/O buffer if necessary */
317	free_iob ( iobuf );
318
319	/* Close the connection on any errors */
320	if ( rc != 0 )
321		ib_cmrc_close ( cmrc, rc );
322
323	return rc;
324}
325
326/**
327 * Check CMRC flow control window
328 *
329 * @v xfer		Data transfer interface
330 * @ret len		Length of window
331 */
332static size_t ib_cmrc_xfer_window ( struct xfer_interface *xfer ) {
333	struct ib_cmrc_connection *cmrc =
334		container_of ( xfer, struct ib_cmrc_connection, xfer );
335
336	/* We indicate a window only when we are successfully
337	 * connected.
338	 */
339	return ( cmrc->connected ? IB_MAX_PAYLOAD_SIZE : 0 );
340}
341
342/**
343 * Close CMRC data-transfer interface
344 *
345 * @v xfer		Data transfer interface
346 * @v rc		Reason for close
347 */
348static void ib_cmrc_xfer_close ( struct xfer_interface *xfer, int rc ) {
349	struct ib_cmrc_connection *cmrc =
350		container_of ( xfer, struct ib_cmrc_connection, xfer );
351
352	DBGC ( cmrc, "CMRC %p closed: %s\n", cmrc, strerror ( rc ) );
353	ib_cmrc_close ( cmrc, rc );
354}
355
356/** CMRC data transfer interface operations */
357static struct xfer_interface_operations ib_cmrc_xfer_operations = {
358	.close		= ib_cmrc_xfer_close,
359	.vredirect	= ignore_xfer_vredirect,
360	.window		= ib_cmrc_xfer_window,
361	.alloc_iob	= default_xfer_alloc_iob,
362	.deliver_iob	= ib_cmrc_xfer_deliver_iob,
363	.deliver_raw	= xfer_deliver_as_iob,
364};
365
366/**
367 * Open CMRC connection
368 *
369 * @v xfer		Data transfer interface
370 * @v ibdev		Infiniband device
371 * @v dgid		Destination GID
372 * @v service_id	Service ID
373 * @ret rc		Returns status code
374 */
375int ib_cmrc_open ( struct xfer_interface *xfer, struct ib_device *ibdev,
376		   struct ib_gid *dgid, struct ib_gid_half *service_id ) {
377	struct ib_cmrc_connection *cmrc;
378	int rc;
379
380	/* Allocate and initialise structure */
381	cmrc = zalloc ( sizeof ( *cmrc ) );
382	if ( ! cmrc ) {
383		rc = -ENOMEM;
384		goto err_alloc;
385	}
386	xfer_init ( &cmrc->xfer, &ib_cmrc_xfer_operations, &cmrc->refcnt );
387	cmrc->ibdev = ibdev;
388	memcpy ( &cmrc->dgid, dgid, sizeof ( cmrc->dgid ) );
389	memcpy ( &cmrc->service_id, service_id, sizeof ( cmrc->service_id ) );
390	process_init_stopped ( &cmrc->shutdown, ib_cmrc_shutdown,
391			       &cmrc->refcnt );
392
393	/* Open Infiniband device */
394	if ( ( rc = ib_open ( ibdev ) ) != 0 ) {
395		DBGC ( cmrc, "CMRC %p could not open device: %s\n",
396		       cmrc, strerror ( rc ) );
397		goto err_open;
398	}
399
400	/* Create completion queue */
401	cmrc->cq = ib_create_cq ( ibdev, IB_CMRC_NUM_CQES,
402				  &ib_cmrc_completion_ops );
403	if ( ! cmrc->cq ) {
404		DBGC ( cmrc, "CMRC %p could not create completion queue\n",
405		       cmrc );
406		rc = -ENOMEM;
407		goto err_create_cq;
408	}
409
410	/* Create queue pair */
411	cmrc->qp = ib_create_qp ( ibdev, IB_QPT_RC, IB_CMRC_NUM_SEND_WQES,
412				  cmrc->cq, IB_CMRC_NUM_RECV_WQES, cmrc->cq );
413	if ( ! cmrc->qp ) {
414		DBGC ( cmrc, "CMRC %p could not create queue pair\n", cmrc );
415		rc = -ENOMEM;
416		goto err_create_qp;
417	}
418	ib_qp_set_ownerdata ( cmrc->qp, cmrc );
419	DBGC ( cmrc, "CMRC %p using QPN %lx\n", cmrc, cmrc->qp->qpn );
420
421	/* Attach to parent interface, transfer reference (implicitly)
422	 * to our shutdown process, and return.
423	 */
424	xfer_plug_plug ( &cmrc->xfer, xfer );
425	return 0;
426
427	ib_destroy_qp ( ibdev, cmrc->qp );
428 err_create_qp:
429	ib_destroy_cq ( ibdev, cmrc->cq );
430 err_create_cq:
431	ib_close ( ibdev );
432 err_open:
433	ref_put ( &cmrc->refcnt );
434 err_alloc:
435	return rc;
436}
437