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 #include <linux/kvm_host.h>
25 #if defined(__x86_64__) || defined(__i386__)
26 #include <asm/hypervisor.h>
29 static int sock_timeout;
30 module_param(sock_timeout, int, 0644);
31 MODULE_PARM_DESC(sock_timeout, "dead socket timeout (seconds)");
33 static int credits = DEFAULT_CREDITS;
34 module_param(credits, int, 0444);
35 MODULE_PARM_DESC(credits, "# concurrent sends");
37 static int peer_credits = DEFAULT_PEER_CREDITS;
38 module_param(peer_credits, int, 0444);
39 MODULE_PARM_DESC(peer_credits, "# concurrent sends to 1 peer");
41 static int peer_buffer_credits;
42 module_param(peer_buffer_credits, int, 0444);
43 MODULE_PARM_DESC(peer_buffer_credits, "# per-peer router buffer credits");
45 static int peer_timeout = DEFAULT_PEER_TIMEOUT;
46 module_param(peer_timeout, int, 0444);
47 MODULE_PARM_DESC(peer_timeout, "Seconds without aliveness news to declare peer dead (<=0 to disable)");
49 /* Number of daemons in each thread pool which is percpt,
50 * we will estimate reasonable value based on CPUs if it's not set. */
51 static unsigned int nscheds;
52 module_param(nscheds, int, 0444);
53 MODULE_PARM_DESC(nscheds, "# scheduler daemons in each pool while starting");
55 static int nconnds = 4;
56 module_param(nconnds, int, 0444);
57 MODULE_PARM_DESC(nconnds, "# connection daemons while starting");
59 static int nconnds_max = 64;
60 module_param(nconnds_max, int, 0444);
61 MODULE_PARM_DESC(nconnds_max, "max # connection daemons");
63 static int min_reconnectms = 1000;
64 module_param(min_reconnectms, int, 0644);
65 MODULE_PARM_DESC(min_reconnectms, "min connection retry interval (mS)");
67 static int max_reconnectms = 60000;
68 module_param(max_reconnectms, int, 0644);
69 MODULE_PARM_DESC(max_reconnectms, "max connection retry interval (mS)");
72 module_param(eager_ack, int, 0644);
73 MODULE_PARM_DESC(eager_ack, "send tcp ack packets eagerly");
75 static int typed_conns = 1;
76 module_param(typed_conns, int, 0444);
77 MODULE_PARM_DESC(typed_conns, "use different sockets for bulk");
79 static int min_bulk = (1<<10);
80 module_param(min_bulk, int, 0644);
81 MODULE_PARM_DESC(min_bulk, "smallest 'large' message");
83 # define DEFAULT_BUFFER_SIZE 0
84 static int tx_buffer_size = DEFAULT_BUFFER_SIZE;
85 module_param(tx_buffer_size, int, 0644);
86 MODULE_PARM_DESC(tx_buffer_size, "socket tx buffer size (0 for system default)");
88 static int rx_buffer_size = DEFAULT_BUFFER_SIZE;
89 module_param(rx_buffer_size, int, 0644);
90 MODULE_PARM_DESC(rx_buffer_size, "socket rx buffer size (0 for system default)");
93 module_param(nagle, int, 0644);
94 MODULE_PARM_DESC(nagle, "enable NAGLE?");
96 static int round_robin = 1;
97 module_param(round_robin, int, 0644);
98 MODULE_PARM_DESC(round_robin, "Round robin for multiple interfaces");
100 static int keepalive = 30;
101 module_param(keepalive, int, 0644);
102 MODULE_PARM_DESC(keepalive, "# seconds before send keepalive");
104 static int keepalive_idle = 30;
105 module_param(keepalive_idle, int, 0644);
106 MODULE_PARM_DESC(keepalive_idle, "# idle seconds before probe");
108 #define DEFAULT_KEEPALIVE_COUNT 5
109 static int keepalive_count = DEFAULT_KEEPALIVE_COUNT;
110 module_param(keepalive_count, int, 0644);
111 MODULE_PARM_DESC(keepalive_count, "# missed probes == dead");
113 static int keepalive_intvl = 5;
114 module_param(keepalive_intvl, int, 0644);
115 MODULE_PARM_DESC(keepalive_intvl, "seconds between probes");
117 static int enable_csum = 0;
118 module_param(enable_csum, int, 0644);
119 MODULE_PARM_DESC(enable_csum, "enable check sum");
121 static int inject_csum_error = 0;
122 module_param(inject_csum_error, int, 0644);
123 MODULE_PARM_DESC(inject_csum_error, "set non-zero to inject a checksum error");
125 static int enable_irq_affinity = 0;
126 module_param(enable_irq_affinity, int, 0644);
127 MODULE_PARM_DESC(enable_irq_affinity, "enable IRQ affinity");
129 static int nonblk_zcack = 1;
130 module_param(nonblk_zcack, int, 0644);
131 MODULE_PARM_DESC(nonblk_zcack, "always send ZC-ACK on non-blocking connection");
133 static unsigned int zc_min_payload = (16 << 10);
134 module_param(zc_min_payload, int, 0644);
135 MODULE_PARM_DESC(zc_min_payload, "minimum payload size to zero copy");
137 static unsigned int zc_recv = 0;
138 module_param(zc_recv, int, 0644);
139 MODULE_PARM_DESC(zc_recv, "enable ZC recv for Chelsio driver");
141 static unsigned int zc_recv_min_nfrags = 16;
142 module_param(zc_recv_min_nfrags, int, 0644);
143 MODULE_PARM_DESC(zc_recv_min_nfrags, "minimum # of fragments to enable ZC recv");
145 static unsigned int conns_per_peer = 1;
146 module_param(conns_per_peer, uint, 0444);
147 MODULE_PARM_DESC(conns_per_peer, "number of connections per peer");
149 #ifdef SOCKNAL_BACKOFF
150 static int backoff_init = 3;
151 module_param(backoff_init, int, 0644);
152 MODULE_PARM_DESC(backoff_init, "seconds for initial tcp backoff");
154 static int backoff_max = 3;
155 module_param(backoff_max, int, 0644);
156 MODULE_PARM_DESC(backoff_max, "seconds for maximum tcp backoff");
159 #if SOCKNAL_VERSION_DEBUG
160 static int protocol = 3;
161 module_param(protocol, int, 0644);
162 MODULE_PARM_DESC(protocol, "protocol version");
165 static inline bool is_native_host(void)
167 #ifdef HAVE_HYPERVISOR_IS_TYPE
168 return hypervisor_is_type(X86_HYPER_NATIVE);
169 #elif defined(__x86_64__) || defined(__i386__)
170 return x86_hyper == NULL;
176 struct ksock_tunables ksocknal_tunables;
178 int ksocknal_tunables_init(void)
180 /* initialize ksocknal_tunables structure */
181 ksocknal_tunables.ksnd_timeout = &sock_timeout;
182 ksocknal_tunables.ksnd_nscheds = &nscheds;
183 ksocknal_tunables.ksnd_nconnds = &nconnds;
184 ksocknal_tunables.ksnd_nconnds_max = &nconnds_max;
185 ksocknal_tunables.ksnd_min_reconnectms = &min_reconnectms;
186 ksocknal_tunables.ksnd_max_reconnectms = &max_reconnectms;
187 ksocknal_tunables.ksnd_eager_ack = &eager_ack;
188 ksocknal_tunables.ksnd_typed_conns = &typed_conns;
189 ksocknal_tunables.ksnd_min_bulk = &min_bulk;
190 ksocknal_tunables.ksnd_tx_buffer_size = &tx_buffer_size;
191 ksocknal_tunables.ksnd_rx_buffer_size = &rx_buffer_size;
192 ksocknal_tunables.ksnd_nagle = &nagle;
193 ksocknal_tunables.ksnd_round_robin = &round_robin;
194 ksocknal_tunables.ksnd_keepalive = &keepalive;
195 ksocknal_tunables.ksnd_keepalive_idle = &keepalive_idle;
196 ksocknal_tunables.ksnd_keepalive_count = &keepalive_count;
197 ksocknal_tunables.ksnd_keepalive_intvl = &keepalive_intvl;
198 ksocknal_tunables.ksnd_credits = &credits;
199 ksocknal_tunables.ksnd_peertxcredits = &peer_credits;
200 ksocknal_tunables.ksnd_peerrtrcredits = &peer_buffer_credits;
201 ksocknal_tunables.ksnd_peertimeout = &peer_timeout;
202 ksocknal_tunables.ksnd_enable_csum = &enable_csum;
203 ksocknal_tunables.ksnd_inject_csum_error = &inject_csum_error;
204 ksocknal_tunables.ksnd_nonblk_zcack = &nonblk_zcack;
205 ksocknal_tunables.ksnd_zc_min_payload = &zc_min_payload;
206 ksocknal_tunables.ksnd_zc_recv = &zc_recv;
207 ksocknal_tunables.ksnd_zc_recv_min_nfrags = &zc_recv_min_nfrags;
208 if (conns_per_peer > ((1 << SOCKNAL_CONN_COUNT_MAX_BITS)-1)) {
209 CWARN("socklnd conns_per_peer is capped at %u.\n",
210 (1 << SOCKNAL_CONN_COUNT_MAX_BITS)-1);
212 ksocknal_tunables.ksnd_conns_per_peer = &conns_per_peer;
214 if (enable_irq_affinity) {
215 CWARN("irq_affinity is removed from socklnd because modern "
216 "computer always has fast CPUs and more cores than "
217 "# NICs, although you still can set irq_affinity by "
218 "another way, please check manual for details.\n");
220 ksocknal_tunables.ksnd_irq_affinity = &enable_irq_affinity;
222 #ifdef SOCKNAL_BACKOFF
223 ksocknal_tunables.ksnd_backoff_init = &backoff_init;
224 ksocknal_tunables.ksnd_backoff_max = &backoff_max;
227 #if SOCKNAL_VERSION_DEBUG
228 ksocknal_tunables.ksnd_protocol = &protocol;
231 if (*ksocknal_tunables.ksnd_zc_min_payload < (2 << 10))
232 *ksocknal_tunables.ksnd_zc_min_payload = (2 << 10);
234 /* When on a hypervisor set the minimum zero copy size
235 * above the maximum payload size
237 if (!is_native_host())
238 *ksocknal_tunables.ksnd_zc_min_payload = (16 << 20) + 1;