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 static int map_on_demand;
110 module_param(map_on_demand, int, 0444);
111 MODULE_PARM_DESC(map_on_demand, "map on demand");
113 /* NB: this value is shared by all CPTs, it can grow at runtime */
114 static int fmr_pool_size = 512;
115 module_param(fmr_pool_size, int, 0444);
116 MODULE_PARM_DESC(fmr_pool_size, "size of fmr pool on each CPT (>= ntx / 4)");
118 /* NB: this value is shared by all CPTs, it can grow at runtime */
119 static int fmr_flush_trigger = 384;
120 module_param(fmr_flush_trigger, int, 0444);
121 MODULE_PARM_DESC(fmr_flush_trigger, "# dirty FMRs that triggers pool flush");
123 static int fmr_cache = 1;
124 module_param(fmr_cache, int, 0444);
125 MODULE_PARM_DESC(fmr_cache, "non-zero to enable FMR caching");
128 * 0: disable failover
129 * 1: enable failover if necessary
130 * 2: force to failover (for debug)
132 static int dev_failover;
133 module_param(dev_failover, int, 0444);
134 MODULE_PARM_DESC(dev_failover, "HCA failover for bonding (0 off, 1 on, other values reserved)");
136 static int require_privileged_port;
137 module_param(require_privileged_port, int, 0644);
138 MODULE_PARM_DESC(require_privileged_port, "require privileged port when accepting connection");
140 static int use_privileged_port = 1;
141 module_param(use_privileged_port, int, 0644);
142 MODULE_PARM_DESC(use_privileged_port, "use privileged port when initiating connection");
144 struct kib_tunables kiblnd_tunables = {
145 .kib_dev_failover = &dev_failover,
146 .kib_service = &service,
148 .kib_timeout = &timeout,
149 .kib_keepalive = &keepalive,
151 .kib_default_ipif = &ipif_name,
152 .kib_retry_count = &retry_count,
153 .kib_rnr_retry_count = &rnr_retry_count,
154 .kib_ib_mtu = &ib_mtu,
155 .kib_require_priv_port = &require_privileged_port,
156 .kib_use_priv_port = &use_privileged_port,
157 .kib_nscheds = &nscheds
160 static struct lnet_ioctl_config_o2iblnd_tunables default_tunables;
162 /* # messages/RDMAs in-flight */
163 int kiblnd_msg_queue_size(int version, lnet_ni_t *ni)
165 if (version == IBLND_MSG_VERSION_1)
166 return IBLND_MSG_QUEUE_SIZE_V1;
168 return ni->ni_peertxcredits;
173 int kiblnd_tunables_setup(struct lnet_ni *ni)
175 struct lnet_ioctl_config_o2iblnd_tunables *tunables;
178 * if there was no tunables specified, setup the tunables to be
181 if (!ni->ni_lnd_tunables) {
182 LIBCFS_ALLOC(ni->ni_lnd_tunables,
183 sizeof(*ni->ni_lnd_tunables));
184 if (!ni->ni_lnd_tunables)
187 memcpy(&ni->ni_lnd_tunables->lt_tun_u.lt_o2ib,
188 &default_tunables, sizeof(*tunables));
190 tunables = &ni->ni_lnd_tunables->lt_tun_u.lt_o2ib;
192 /* Current API version */
193 tunables->lnd_version = 0;
195 if (kiblnd_translate_mtu(*kiblnd_tunables.kib_ib_mtu) < 0) {
196 CERROR("Invalid ib_mtu %d, expected 256/512/1024/2048/4096\n",
197 *kiblnd_tunables.kib_ib_mtu);
201 if (!ni->ni_peertimeout)
202 ni->ni_peertimeout = peer_timeout;
204 if (!ni->ni_maxtxcredits)
205 ni->ni_maxtxcredits = credits;
207 if (!ni->ni_peertxcredits)
208 ni->ni_peertxcredits = peer_credits;
210 if (!ni->ni_peerrtrcredits)
211 ni->ni_peerrtrcredits = peer_buffer_credits;
213 if (ni->ni_peertxcredits < IBLND_CREDITS_DEFAULT)
214 ni->ni_peertxcredits = IBLND_CREDITS_DEFAULT;
216 if (ni->ni_peertxcredits > IBLND_CREDITS_MAX)
217 ni->ni_peertxcredits = IBLND_CREDITS_MAX;
219 if (ni->ni_peertxcredits > credits)
220 ni->ni_peertxcredits = credits;
222 if (!tunables->lnd_peercredits_hiw)
223 tunables->lnd_peercredits_hiw = peer_credits_hiw;
225 if (tunables->lnd_peercredits_hiw < ni->ni_peertxcredits / 2)
226 tunables->lnd_peercredits_hiw = ni->ni_peertxcredits / 2;
228 if (tunables->lnd_peercredits_hiw >= ni->ni_peertxcredits)
229 tunables->lnd_peercredits_hiw = ni->ni_peertxcredits - 1;
231 if (tunables->lnd_map_on_demand < 0 ||
232 tunables->lnd_map_on_demand > IBLND_MAX_RDMA_FRAGS) {
233 /* disable map-on-demand */
234 tunables->lnd_map_on_demand = 0;
237 if (tunables->lnd_map_on_demand == 1) {
238 /* don't make sense to create map if only one fragment */
239 tunables->lnd_map_on_demand = 2;
242 if (!tunables->lnd_concurrent_sends) {
243 if (tunables->lnd_map_on_demand > 0 &&
244 tunables->lnd_map_on_demand <= IBLND_MAX_RDMA_FRAGS / 8) {
245 tunables->lnd_concurrent_sends =
246 ni->ni_peertxcredits * 2;
248 tunables->lnd_concurrent_sends = ni->ni_peertxcredits;
252 if (tunables->lnd_concurrent_sends > ni->ni_peertxcredits * 2)
253 tunables->lnd_concurrent_sends = ni->ni_peertxcredits * 2;
255 if (tunables->lnd_concurrent_sends < ni->ni_peertxcredits / 2)
256 tunables->lnd_concurrent_sends = ni->ni_peertxcredits / 2;
258 if (tunables->lnd_concurrent_sends < ni->ni_peertxcredits) {
259 CWARN("Concurrent sends %d is lower than message queue size: %d, performance may drop slightly.\n",
260 tunables->lnd_concurrent_sends, ni->ni_peertxcredits);
263 if (!tunables->lnd_fmr_pool_size)
264 tunables->lnd_fmr_pool_size = fmr_pool_size;
265 if (!tunables->lnd_fmr_flush_trigger)
266 tunables->lnd_fmr_flush_trigger = fmr_flush_trigger;
267 if (!tunables->lnd_fmr_cache)
268 tunables->lnd_fmr_cache = fmr_cache;
273 void kiblnd_tunables_init(void)
275 default_tunables.lnd_version = 0;
276 default_tunables.lnd_peercredits_hiw = peer_credits_hiw,
277 default_tunables.lnd_map_on_demand = map_on_demand;
278 default_tunables.lnd_concurrent_sends = concurrent_sends;
279 default_tunables.lnd_fmr_pool_size = fmr_pool_size;
280 default_tunables.lnd_fmr_flush_trigger = fmr_flush_trigger;
281 default_tunables.lnd_fmr_cache = fmr_cache;