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, 2016, 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)");
51 /* Number of threads in each scheduler pool which is percpt,
52 * we will estimate reasonable value based on CPUs if it's set to zero. */
54 module_param(nscheds, int, 0444);
55 MODULE_PARM_DESC(nscheds, "number of threads in each scheduler pool");
57 /* NB: this value is shared by all CPTs, it can grow at runtime */
59 module_param(ntx, int, 0444);
60 MODULE_PARM_DESC(ntx, "# of message descriptors allocated for each pool");
62 /* NB: this value is shared by all CPTs */
63 static int credits = 256;
64 module_param(credits, int, 0444);
65 MODULE_PARM_DESC(credits, "# concurrent sends");
67 static int peer_credits = 8;
68 module_param(peer_credits, int, 0444);
69 MODULE_PARM_DESC(peer_credits, "# concurrent sends to 1 peer");
71 static int peer_credits_hiw = 0;
72 module_param(peer_credits_hiw, int, 0444);
73 MODULE_PARM_DESC(peer_credits_hiw, "when eagerly to return credits");
75 static int peer_buffer_credits = 0;
76 module_param(peer_buffer_credits, int, 0444);
77 MODULE_PARM_DESC(peer_buffer_credits, "# per-peer router buffer credits");
79 static int peer_timeout = 180;
80 module_param(peer_timeout, int, 0444);
81 MODULE_PARM_DESC(peer_timeout, "Seconds without aliveness news to declare peer dead (<=0 to disable)");
83 static char *ipif_name = "ib0";
84 module_param(ipif_name, charp, 0444);
85 MODULE_PARM_DESC(ipif_name, "IPoIB interface name");
87 static int retry_count = 5;
88 module_param(retry_count, int, 0644);
89 MODULE_PARM_DESC(retry_count, "Retransmissions when no ACK received");
91 static int rnr_retry_count = 6;
92 module_param(rnr_retry_count, int, 0644);
93 MODULE_PARM_DESC(rnr_retry_count, "RNR retransmissions");
95 static int keepalive = 100;
96 module_param(keepalive, int, 0644);
97 MODULE_PARM_DESC(keepalive, "Idle time in seconds before sending a keepalive");
100 module_param(ib_mtu, int, 0444);
101 MODULE_PARM_DESC(ib_mtu, "IB MTU 256/512/1024/2048/4096");
103 static int concurrent_sends;
104 module_param(concurrent_sends, int, 0444);
105 MODULE_PARM_DESC(concurrent_sends, "send work-queue sizing");
107 #ifdef HAVE_IB_GET_DMA_MR
108 #define IBLND_DEFAULT_MAP_ON_DEMAND 0
109 #define IBLND_MIN_MAP_ON_DEMAND 0
111 #define IBLND_DEFAULT_MAP_ON_DEMAND IBLND_MAX_RDMA_FRAGS
112 #define IBLND_MIN_MAP_ON_DEMAND 1
114 static int map_on_demand = IBLND_DEFAULT_MAP_ON_DEMAND;
115 module_param(map_on_demand, int, 0444);
116 MODULE_PARM_DESC(map_on_demand, "map on demand");
118 /* NB: this value is shared by all CPTs, it can grow at runtime */
119 static int fmr_pool_size = 512;
120 module_param(fmr_pool_size, int, 0444);
121 MODULE_PARM_DESC(fmr_pool_size, "size of fmr pool on each CPT (>= ntx / 4)");
123 /* NB: this value is shared by all CPTs, it can grow at runtime */
124 static int fmr_flush_trigger = 384;
125 module_param(fmr_flush_trigger, int, 0444);
126 MODULE_PARM_DESC(fmr_flush_trigger, "# dirty FMRs that triggers pool flush");
128 static int fmr_cache = 1;
129 module_param(fmr_cache, int, 0444);
130 MODULE_PARM_DESC(fmr_cache, "non-zero to enable FMR caching");
133 * 0: disable failover
134 * 1: enable failover if necessary
135 * 2: force to failover (for debug)
137 static int dev_failover = 0;
138 module_param(dev_failover, int, 0444);
139 MODULE_PARM_DESC(dev_failover, "HCA failover for bonding (0 off, 1 on, other values reserved)");
141 static int require_privileged_port;
142 module_param(require_privileged_port, int, 0644);
143 MODULE_PARM_DESC(require_privileged_port, "require privileged port when accepting connection");
145 static int use_privileged_port = 1;
146 module_param(use_privileged_port, int, 0644);
147 MODULE_PARM_DESC(use_privileged_port, "use privileged port when initiating connection");
149 kib_tunables_t kiblnd_tunables = {
150 .kib_dev_failover = &dev_failover,
151 .kib_service = &service,
153 .kib_timeout = &timeout,
154 .kib_keepalive = &keepalive,
156 .kib_default_ipif = &ipif_name,
157 .kib_retry_count = &retry_count,
158 .kib_rnr_retry_count = &rnr_retry_count,
159 .kib_ib_mtu = &ib_mtu,
160 .kib_require_priv_port = &require_privileged_port,
161 .kib_use_priv_port = &use_privileged_port,
162 .kib_nscheds = &nscheds
165 static struct lnet_ioctl_config_o2iblnd_tunables default_tunables;
167 /* # messages/RDMAs in-flight */
169 kiblnd_msg_queue_size(int version, struct lnet_ni *ni)
171 if (version == IBLND_MSG_VERSION_1)
172 return IBLND_MSG_QUEUE_SIZE_V1;
174 return ni->ni_net->net_tunables.lct_peer_tx_credits;
180 kiblnd_tunables_setup(struct lnet_ni *ni)
182 struct lnet_ioctl_config_o2iblnd_tunables *tunables;
183 struct lnet_ioctl_config_lnd_cmn_tunables *net_tunables;
186 * if there was no tunables specified, setup the tunables to be
189 if (!ni->ni_lnd_tunables_set)
190 memcpy(&ni->ni_lnd_tunables.lnd_tun_u.lnd_o2ib,
191 &default_tunables, sizeof(*tunables));
193 tunables = &ni->ni_lnd_tunables.lnd_tun_u.lnd_o2ib;
195 /* Current API version */
196 tunables->lnd_version = 0;
198 if (kiblnd_translate_mtu(*kiblnd_tunables.kib_ib_mtu) < 0) {
199 CERROR("Invalid ib_mtu %d, expected 256/512/1024/2048/4096\n",
200 *kiblnd_tunables.kib_ib_mtu);
204 net_tunables = &ni->ni_net->net_tunables;
206 if (net_tunables->lct_peer_timeout == -1)
207 net_tunables->lct_peer_timeout = peer_timeout;
209 if (net_tunables->lct_max_tx_credits == -1)
210 net_tunables->lct_max_tx_credits = credits;
212 if (net_tunables->lct_peer_tx_credits == -1)
213 net_tunables->lct_peer_tx_credits = peer_credits;
215 if (net_tunables->lct_peer_rtr_credits == -1)
216 net_tunables->lct_peer_rtr_credits = peer_buffer_credits;
218 if (net_tunables->lct_peer_tx_credits < IBLND_CREDITS_DEFAULT)
219 net_tunables->lct_peer_tx_credits = IBLND_CREDITS_DEFAULT;
221 if (net_tunables->lct_peer_tx_credits > IBLND_CREDITS_MAX)
222 net_tunables->lct_peer_tx_credits = IBLND_CREDITS_MAX;
224 if (net_tunables->lct_peer_tx_credits >
225 net_tunables->lct_max_tx_credits)
226 net_tunables->lct_peer_tx_credits =
227 net_tunables->lct_max_tx_credits;
229 if (!tunables->lnd_peercredits_hiw)
230 tunables->lnd_peercredits_hiw = peer_credits_hiw;
232 if (tunables->lnd_peercredits_hiw < net_tunables->lct_peer_tx_credits / 2)
233 tunables->lnd_peercredits_hiw = net_tunables->lct_peer_tx_credits / 2;
235 if (tunables->lnd_peercredits_hiw >= net_tunables->lct_peer_tx_credits)
236 tunables->lnd_peercredits_hiw = net_tunables->lct_peer_tx_credits - 1;
238 if (tunables->lnd_map_on_demand < IBLND_MIN_MAP_ON_DEMAND ||
239 tunables->lnd_map_on_demand > IBLND_MAX_RDMA_FRAGS) {
240 /* Use the default */
241 CWARN("Invalid map_on_demand (%d), expects %d - %d. Using default of %d\n",
242 tunables->lnd_map_on_demand, IBLND_MIN_MAP_ON_DEMAND,
243 IBLND_MAX_RDMA_FRAGS, IBLND_DEFAULT_MAP_ON_DEMAND);
244 tunables->lnd_map_on_demand = IBLND_DEFAULT_MAP_ON_DEMAND;
247 if (tunables->lnd_map_on_demand == 1) {
248 /* don't make sense to create map if only one fragment */
249 tunables->lnd_map_on_demand = 2;
252 if (tunables->lnd_concurrent_sends == 0) {
253 if (tunables->lnd_map_on_demand > 0 &&
254 tunables->lnd_map_on_demand <= IBLND_MAX_RDMA_FRAGS / 8) {
255 tunables->lnd_concurrent_sends =
256 net_tunables->lct_peer_tx_credits * 2;
258 tunables->lnd_concurrent_sends =
259 net_tunables->lct_peer_tx_credits;
263 if (tunables->lnd_concurrent_sends > net_tunables->lct_peer_tx_credits * 2)
264 tunables->lnd_concurrent_sends = net_tunables->lct_peer_tx_credits * 2;
266 if (tunables->lnd_concurrent_sends < net_tunables->lct_peer_tx_credits / 2)
267 tunables->lnd_concurrent_sends = net_tunables->lct_peer_tx_credits / 2;
269 if (tunables->lnd_concurrent_sends < net_tunables->lct_peer_tx_credits) {
270 CWARN("Concurrent sends %d is lower than message "
271 "queue size: %d, performance may drop slightly.\n",
272 tunables->lnd_concurrent_sends,
273 net_tunables->lct_peer_tx_credits);
276 if (!tunables->lnd_fmr_pool_size)
277 tunables->lnd_fmr_pool_size = fmr_pool_size;
278 if (!tunables->lnd_fmr_flush_trigger)
279 tunables->lnd_fmr_flush_trigger = fmr_flush_trigger;
280 if (!tunables->lnd_fmr_cache)
281 tunables->lnd_fmr_cache = fmr_cache;
287 kiblnd_tunables_init(void)
289 default_tunables.lnd_version = 0;
290 default_tunables.lnd_peercredits_hiw = peer_credits_hiw,
291 default_tunables.lnd_map_on_demand = map_on_demand;
292 default_tunables.lnd_concurrent_sends = concurrent_sends;
293 default_tunables.lnd_fmr_pool_size = fmr_pool_size;
294 default_tunables.lnd_fmr_flush_trigger = fmr_flush_trigger;
295 default_tunables.lnd_fmr_cache = fmr_cache;