2 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
4 * Copyright (c) 2011, 2012, Intel Corporation.
6 * Author: Eric Barton <eric@bartonsoftware.com>
8 * Portals is free software; you can redistribute it and/or
9 * modify it under the terms of version 2 of the GNU General Public
10 * License as published by the Free Software Foundation.
12 * Portals is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with Portals; if not, write to the Free Software
19 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
24 static int sock_timeout = 50;
25 CFS_MODULE_PARM(sock_timeout, "i", int, 0644,
26 "dead socket timeout (seconds)");
28 static int credits = 256;
29 CFS_MODULE_PARM(credits, "i", int, 0444,
30 "# concurrent sends");
32 static int peer_credits = 8;
33 CFS_MODULE_PARM(peer_credits, "i", int, 0444,
34 "# concurrent sends to 1 peer");
36 static int peer_buffer_credits = 0;
37 CFS_MODULE_PARM(peer_buffer_credits, "i", int, 0444,
38 "# per-peer router buffer credits");
40 static int peer_timeout = 180;
41 CFS_MODULE_PARM(peer_timeout, "i", int, 0444,
42 "Seconds without aliveness news to declare peer dead (<=0 to disable)");
44 /* Number of daemons in each thread pool which is percpt,
45 * we will estimate reasonable value based on CPUs if it's not set. */
46 static unsigned int nscheds;
47 CFS_MODULE_PARM(nscheds, "i", int, 0444,
48 "# scheduler daemons in each pool while starting");
50 static int nconnds = 4;
51 CFS_MODULE_PARM(nconnds, "i", int, 0444,
52 "# connection daemons while starting");
54 static int nconnds_max = 64;
55 CFS_MODULE_PARM(nconnds_max, "i", int, 0444,
56 "max # connection daemons");
58 static int min_reconnectms = 1000;
59 CFS_MODULE_PARM(min_reconnectms, "i", int, 0644,
60 "min connection retry interval (mS)");
62 static int max_reconnectms = 60000;
63 CFS_MODULE_PARM(max_reconnectms, "i", int, 0644,
64 "max connection retry interval (mS)");
66 #if defined(__APPLE__) && !defined(__DARWIN8__)
67 # define DEFAULT_EAGER_ACK 1
69 # define DEFAULT_EAGER_ACK 0
71 static int eager_ack = DEFAULT_EAGER_ACK;
72 CFS_MODULE_PARM(eager_ack, "i", int, 0644,
73 "send tcp ack packets eagerly");
75 static int typed_conns = 1;
76 CFS_MODULE_PARM(typed_conns, "i", int, 0444,
77 "use different sockets for bulk");
79 static int min_bulk = (1<<10);
80 CFS_MODULE_PARM(min_bulk, "i", int, 0644,
81 "smallest 'large' message");
85 # define DEFAULT_BUFFER_SIZE (224*1024)
87 # define DEFAULT_BUFFER_SIZE (1152 * 1024)
90 # define DEFAULT_BUFFER_SIZE 0
92 static int tx_buffer_size = DEFAULT_BUFFER_SIZE;
93 CFS_MODULE_PARM(tx_buffer_size, "i", int, 0644,
94 "socket tx buffer size (0 for system default)");
96 static int rx_buffer_size = DEFAULT_BUFFER_SIZE;
97 CFS_MODULE_PARM(rx_buffer_size, "i", int, 0644,
98 "socket rx buffer size (0 for system default)");
100 static int nagle = 0;
101 CFS_MODULE_PARM(nagle, "i", int, 0644,
104 static int round_robin = 1;
105 CFS_MODULE_PARM(round_robin, "i", int, 0644,
106 "Round robin for multiple interfaces");
108 static int keepalive = 30;
109 CFS_MODULE_PARM(keepalive, "i", int, 0644,
110 "# seconds before send keepalive");
112 static int keepalive_idle = 30;
113 CFS_MODULE_PARM(keepalive_idle, "i", int, 0644,
114 "# idle seconds before probe");
116 #define DEFAULT_KEEPALIVE_COUNT 5
117 static int keepalive_count = DEFAULT_KEEPALIVE_COUNT;
118 CFS_MODULE_PARM(keepalive_count, "i", int, 0644,
119 "# missed probes == dead");
121 static int keepalive_intvl = 5;
122 CFS_MODULE_PARM(keepalive_intvl, "i", int, 0644,
123 "seconds between probes");
125 static int enable_csum = 0;
126 CFS_MODULE_PARM(enable_csum, "i", int, 0644,
129 static int inject_csum_error = 0;
130 CFS_MODULE_PARM(inject_csum_error, "i", int, 0644,
131 "set non-zero to inject a checksum error");
133 static int enable_irq_affinity = 0;
134 CFS_MODULE_PARM(enable_irq_affinity, "i", int, 0644,
135 "enable IRQ affinity");
138 static int nonblk_zcack = 1;
139 CFS_MODULE_PARM(nonblk_zcack, "i", int, 0644,
140 "always send ZC-ACK on non-blocking connection");
142 static unsigned int zc_min_payload = (16 << 10);
143 CFS_MODULE_PARM(zc_min_payload, "i", int, 0644,
144 "minimum payload size to zero copy");
146 static unsigned int zc_recv = 0;
147 CFS_MODULE_PARM(zc_recv, "i", int, 0644,
148 "enable ZC recv for Chelsio driver");
150 static unsigned int zc_recv_min_nfrags = 16;
151 CFS_MODULE_PARM(zc_recv_min_nfrags, "i", int, 0644,
152 "minimum # of fragments to enable ZC recv");
154 #ifdef SOCKNAL_BACKOFF
155 static int backoff_init = 3;
156 CFS_MODULE_PARM(backoff_init, "i", int, 0644,
157 "seconds for initial tcp backoff");
159 static int backoff_max = 3;
160 CFS_MODULE_PARM(backoff_max, "i", int, 0644,
161 "seconds for maximum tcp backoff");
164 #if SOCKNAL_VERSION_DEBUG
165 static int protocol = 3;
166 CFS_MODULE_PARM(protocol, "i", int, 0644,
170 ksock_tunables_t ksocknal_tunables;
172 int ksocknal_tunables_init(void)
175 /* initialize ksocknal_tunables structure */
176 ksocknal_tunables.ksnd_timeout = &sock_timeout;
177 ksocknal_tunables.ksnd_nscheds = &nscheds;
178 ksocknal_tunables.ksnd_nconnds = &nconnds;
179 ksocknal_tunables.ksnd_nconnds_max = &nconnds_max;
180 ksocknal_tunables.ksnd_min_reconnectms = &min_reconnectms;
181 ksocknal_tunables.ksnd_max_reconnectms = &max_reconnectms;
182 ksocknal_tunables.ksnd_eager_ack = &eager_ack;
183 ksocknal_tunables.ksnd_typed_conns = &typed_conns;
184 ksocknal_tunables.ksnd_min_bulk = &min_bulk;
185 ksocknal_tunables.ksnd_tx_buffer_size = &tx_buffer_size;
186 ksocknal_tunables.ksnd_rx_buffer_size = &rx_buffer_size;
187 ksocknal_tunables.ksnd_nagle = &nagle;
188 ksocknal_tunables.ksnd_round_robin = &round_robin;
189 ksocknal_tunables.ksnd_keepalive = &keepalive;
190 ksocknal_tunables.ksnd_keepalive_idle = &keepalive_idle;
191 ksocknal_tunables.ksnd_keepalive_count = &keepalive_count;
192 ksocknal_tunables.ksnd_keepalive_intvl = &keepalive_intvl;
193 ksocknal_tunables.ksnd_credits = &credits;
194 ksocknal_tunables.ksnd_peertxcredits = &peer_credits;
195 ksocknal_tunables.ksnd_peerrtrcredits = &peer_buffer_credits;
196 ksocknal_tunables.ksnd_peertimeout = &peer_timeout;
197 ksocknal_tunables.ksnd_enable_csum = &enable_csum;
198 ksocknal_tunables.ksnd_inject_csum_error = &inject_csum_error;
199 ksocknal_tunables.ksnd_nonblk_zcack = &nonblk_zcack;
200 ksocknal_tunables.ksnd_zc_min_payload = &zc_min_payload;
201 ksocknal_tunables.ksnd_zc_recv = &zc_recv;
202 ksocknal_tunables.ksnd_zc_recv_min_nfrags = &zc_recv_min_nfrags;
205 if (enable_irq_affinity) {
206 CWARN("irq_affinity is removed from socklnd because modern "
207 "computer always has fast CPUs and more cores than "
208 "# NICs, although you still can set irq_affinity by "
209 "another way, please check manual for details.\n");
211 ksocknal_tunables.ksnd_irq_affinity = &enable_irq_affinity;
214 #ifdef SOCKNAL_BACKOFF
215 ksocknal_tunables.ksnd_backoff_init = &backoff_init;
216 ksocknal_tunables.ksnd_backoff_max = &backoff_max;
219 #if SOCKNAL_VERSION_DEBUG
220 ksocknal_tunables.ksnd_protocol = &protocol;
223 #if defined(CONFIG_SYSCTL) && !CFS_SYSFS_MODULE_PARM
224 ksocknal_tunables.ksnd_sysctl = NULL;
227 if (*ksocknal_tunables.ksnd_zc_min_payload < (2 << 10))
228 *ksocknal_tunables.ksnd_zc_min_payload = (2 << 10);
230 /* initialize platform-sepcific tunables */
231 return ksocknal_lib_tunables_init();
234 void ksocknal_tunables_fini(void)
236 ksocknal_lib_tunables_fini();