4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2012, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lnet/klnds/o2iblnd/o2iblnd_modparams.c
34 * Author: Eric Barton <eric@bartonsoftware.com>
39 static int service = 987;
40 module_param(service, int, 0444);
41 MODULE_PARM_DESC(service, "service number (within RDMA_PS_TCP)");
44 module_param(cksum, int, 0644);
45 MODULE_PARM_DESC(cksum, "set non-zero to enable message (not RDMA) checksums");
47 static int timeout = 50;
48 module_param(timeout, int, 0644);
49 MODULE_PARM_DESC(timeout, "timeout (seconds)");
52 * Number of threads in each scheduler pool which is percpt,
53 * we will estimate reasonable value based on CPUs if it's set to zero.
56 module_param(nscheds, int, 0444);
57 MODULE_PARM_DESC(nscheds, "number of threads in each scheduler pool");
59 /* NB: this value is shared by all CPTs, it can grow at runtime */
61 module_param(ntx, int, 0444);
62 MODULE_PARM_DESC(ntx, "# of message descriptors allocated for each pool");
64 /* NB: this value is shared by all CPTs */
65 static int credits = 256;
66 module_param(credits, int, 0444);
67 MODULE_PARM_DESC(credits, "# concurrent sends");
69 static int peer_credits = 8;
70 module_param(peer_credits, int, 0444);
71 MODULE_PARM_DESC(peer_credits, "# concurrent sends to 1 peer");
73 static int peer_credits_hiw;
74 module_param(peer_credits_hiw, int, 0444);
75 MODULE_PARM_DESC(peer_credits_hiw, "when eagerly to return credits");
77 static int peer_buffer_credits;
78 module_param(peer_buffer_credits, int, 0444);
79 MODULE_PARM_DESC(peer_buffer_credits, "# per-peer router buffer credits");
81 static int peer_timeout = 180;
82 module_param(peer_timeout, int, 0444);
83 MODULE_PARM_DESC(peer_timeout, "Seconds without aliveness news to declare peer dead (<=0 to disable)");
85 static char *ipif_name = "ib0";
86 module_param(ipif_name, charp, 0444);
87 MODULE_PARM_DESC(ipif_name, "IPoIB interface name");
89 static int retry_count = 5;
90 module_param(retry_count, int, 0644);
91 MODULE_PARM_DESC(retry_count, "Retransmissions when no ACK received");
93 static int rnr_retry_count = 6;
94 module_param(rnr_retry_count, int, 0644);
95 MODULE_PARM_DESC(rnr_retry_count, "RNR retransmissions");
97 static int keepalive = 100;
98 module_param(keepalive, int, 0644);
99 MODULE_PARM_DESC(keepalive, "Idle time in seconds before sending a keepalive");
102 module_param(ib_mtu, int, 0444);
103 MODULE_PARM_DESC(ib_mtu, "IB MTU 256/512/1024/2048/4096");
105 static int concurrent_sends;
106 module_param(concurrent_sends, int, 0444);
107 MODULE_PARM_DESC(concurrent_sends, "send work-queue sizing");
109 #define IBLND_DEFAULT_MAP_ON_DEMAND IBLND_MAX_RDMA_FRAGS
110 static int map_on_demand = IBLND_DEFAULT_MAP_ON_DEMAND;
111 module_param(map_on_demand, int, 0444);
112 MODULE_PARM_DESC(map_on_demand, "map on demand");
114 /* NB: this value is shared by all CPTs, it can grow at runtime */
115 static int fmr_pool_size = 512;
116 module_param(fmr_pool_size, int, 0444);
117 MODULE_PARM_DESC(fmr_pool_size, "size of fmr pool on each CPT (>= ntx / 4)");
119 /* NB: this value is shared by all CPTs, it can grow at runtime */
120 static int fmr_flush_trigger = 384;
121 module_param(fmr_flush_trigger, int, 0444);
122 MODULE_PARM_DESC(fmr_flush_trigger, "# dirty FMRs that triggers pool flush");
124 static int fmr_cache = 1;
125 module_param(fmr_cache, int, 0444);
126 MODULE_PARM_DESC(fmr_cache, "non-zero to enable FMR caching");
129 * 0: disable failover
130 * 1: enable failover if necessary
131 * 2: force to failover (for debug)
133 static int dev_failover;
134 module_param(dev_failover, int, 0444);
135 MODULE_PARM_DESC(dev_failover, "HCA failover for bonding (0 off, 1 on, other values reserved)");
137 static int require_privileged_port;
138 module_param(require_privileged_port, int, 0644);
139 MODULE_PARM_DESC(require_privileged_port, "require privileged port when accepting connection");
141 static int use_privileged_port = 1;
142 module_param(use_privileged_port, int, 0644);
143 MODULE_PARM_DESC(use_privileged_port, "use privileged port when initiating connection");
145 struct kib_tunables kiblnd_tunables = {
146 .kib_dev_failover = &dev_failover,
147 .kib_service = &service,
149 .kib_timeout = &timeout,
150 .kib_keepalive = &keepalive,
152 .kib_default_ipif = &ipif_name,
153 .kib_retry_count = &retry_count,
154 .kib_rnr_retry_count = &rnr_retry_count,
155 .kib_ib_mtu = &ib_mtu,
156 .kib_require_priv_port = &require_privileged_port,
157 .kib_use_priv_port = &use_privileged_port,
158 .kib_nscheds = &nscheds
161 static struct lnet_ioctl_config_o2iblnd_tunables default_tunables;
163 /* # messages/RDMAs in-flight */
164 int kiblnd_msg_queue_size(int version, struct lnet_ni *ni)
166 if (version == IBLND_MSG_VERSION_1)
167 return IBLND_MSG_QUEUE_SIZE_V1;
169 return ni->ni_peertxcredits;
174 int kiblnd_tunables_setup(struct lnet_ni *ni)
176 struct lnet_ioctl_config_o2iblnd_tunables *tunables;
179 * if there was no tunables specified, setup the tunables to be
182 if (!ni->ni_lnd_tunables) {
183 LIBCFS_ALLOC(ni->ni_lnd_tunables,
184 sizeof(*ni->ni_lnd_tunables));
185 if (!ni->ni_lnd_tunables)
188 memcpy(&ni->ni_lnd_tunables->lt_tun_u.lt_o2ib,
189 &default_tunables, sizeof(*tunables));
191 tunables = &ni->ni_lnd_tunables->lt_tun_u.lt_o2ib;
193 /* Current API version */
194 tunables->lnd_version = 0;
196 if (kiblnd_translate_mtu(*kiblnd_tunables.kib_ib_mtu) < 0) {
197 CERROR("Invalid ib_mtu %d, expected 256/512/1024/2048/4096\n",
198 *kiblnd_tunables.kib_ib_mtu);
202 if (!ni->ni_peertimeout)
203 ni->ni_peertimeout = peer_timeout;
205 if (!ni->ni_maxtxcredits)
206 ni->ni_maxtxcredits = credits;
208 if (!ni->ni_peertxcredits)
209 ni->ni_peertxcredits = peer_credits;
211 if (!ni->ni_peerrtrcredits)
212 ni->ni_peerrtrcredits = peer_buffer_credits;
214 if (ni->ni_peertxcredits < IBLND_CREDITS_DEFAULT)
215 ni->ni_peertxcredits = IBLND_CREDITS_DEFAULT;
217 if (ni->ni_peertxcredits > IBLND_CREDITS_MAX)
218 ni->ni_peertxcredits = IBLND_CREDITS_MAX;
220 if (ni->ni_peertxcredits > credits)
221 ni->ni_peertxcredits = credits;
223 if (!tunables->lnd_peercredits_hiw)
224 tunables->lnd_peercredits_hiw = peer_credits_hiw;
226 if (tunables->lnd_peercredits_hiw < ni->ni_peertxcredits / 2)
227 tunables->lnd_peercredits_hiw = ni->ni_peertxcredits / 2;
229 if (tunables->lnd_peercredits_hiw >= ni->ni_peertxcredits)
230 tunables->lnd_peercredits_hiw = ni->ni_peertxcredits - 1;
232 if (tunables->lnd_map_on_demand <= 0 ||
233 tunables->lnd_map_on_demand > IBLND_MAX_RDMA_FRAGS) {
234 /* Use the default */
235 CWARN("Invalid map_on_demand (%d), expects 1 - %d. Using default of %d\n",
236 tunables->lnd_map_on_demand,
237 IBLND_MAX_RDMA_FRAGS, IBLND_DEFAULT_MAP_ON_DEMAND);
238 tunables->lnd_map_on_demand = IBLND_DEFAULT_MAP_ON_DEMAND;
241 if (tunables->lnd_map_on_demand == 1) {
242 /* don't make sense to create map if only one fragment */
243 tunables->lnd_map_on_demand = 2;
246 if (!tunables->lnd_concurrent_sends) {
247 if (tunables->lnd_map_on_demand > 0 &&
248 tunables->lnd_map_on_demand <= IBLND_MAX_RDMA_FRAGS / 8) {
249 tunables->lnd_concurrent_sends =
250 ni->ni_peertxcredits * 2;
252 tunables->lnd_concurrent_sends = ni->ni_peertxcredits;
256 if (tunables->lnd_concurrent_sends > ni->ni_peertxcredits * 2)
257 tunables->lnd_concurrent_sends = ni->ni_peertxcredits * 2;
259 if (tunables->lnd_concurrent_sends < ni->ni_peertxcredits / 2)
260 tunables->lnd_concurrent_sends = ni->ni_peertxcredits / 2;
262 if (tunables->lnd_concurrent_sends < ni->ni_peertxcredits) {
263 CWARN("Concurrent sends %d is lower than message queue size: %d, performance may drop slightly.\n",
264 tunables->lnd_concurrent_sends, ni->ni_peertxcredits);
267 if (!tunables->lnd_fmr_pool_size)
268 tunables->lnd_fmr_pool_size = fmr_pool_size;
269 if (!tunables->lnd_fmr_flush_trigger)
270 tunables->lnd_fmr_flush_trigger = fmr_flush_trigger;
271 if (!tunables->lnd_fmr_cache)
272 tunables->lnd_fmr_cache = fmr_cache;
277 void kiblnd_tunables_init(void)
279 default_tunables.lnd_version = 0;
280 default_tunables.lnd_peercredits_hiw = peer_credits_hiw,
281 default_tunables.lnd_map_on_demand = map_on_demand;
282 default_tunables.lnd_concurrent_sends = concurrent_sends;
283 default_tunables.lnd_fmr_pool_size = fmr_pool_size;
284 default_tunables.lnd_fmr_flush_trigger = fmr_flush_trigger;
285 default_tunables.lnd_fmr_cache = fmr_cache;