-/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
- * vim:expandtab:shiftwidth=8:tabstop=8:
- *
+/*
* GPL HEADER START
*
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
* GPL HEADER END
*/
/*
- * Copyright 2008 Sun Microsystems, Inc. All rights reserved
+ * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
+ *
+ * Copyright (c) 2012, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
CFS_MODULE_PARM(timeout, "i", int, 0644,
"timeout (seconds)");
-static int ntx = 256;
+/* Number of threads in each scheduler pool which is percpt,
+ * we will estimate reasonable value based on CPUs if it's set to zero. */
+static int nscheds;
+CFS_MODULE_PARM(nscheds, "i", int, 0444,
+ "number of threads in each scheduler pool");
+
+/* NB: this value is shared by all CPTs, it can grow at runtime */
+static int ntx = 512;
CFS_MODULE_PARM(ntx, "i", int, 0444,
- "# of message descriptors");
+ "# of message descriptors allocated for each pool");
-static int credits = 64;
+/* NB: this value is shared by all CPTs */
+static int credits = 256;
CFS_MODULE_PARM(credits, "i", int, 0444,
"# concurrent sends");
CFS_MODULE_PARM(peer_credits, "i", int, 0444,
"# concurrent sends to 1 peer");
+static int peer_credits_hiw = 0;
+CFS_MODULE_PARM(peer_credits_hiw, "i", int, 0444,
+ "when eagerly to return credits");
+
+static int peer_buffer_credits = 0;
+CFS_MODULE_PARM(peer_buffer_credits, "i", int, 0444,
+ "# per-peer router buffer credits");
+
+static int peer_timeout = 180;
+CFS_MODULE_PARM(peer_timeout, "i", int, 0444,
+ "Seconds without aliveness news to declare peer dead (<=0 to disable)");
+
static char *ipif_name = "ib0";
CFS_MODULE_PARM(ipif_name, "s", charp, 0444,
"IPoIB interface name");
CFS_MODULE_PARM(ib_mtu, "i", int, 0444,
"IB MTU 256/512/1024/2048/4096");
-#if IBLND_MAP_ON_DEMAND
-static int concurrent_sends = IBLND_RX_MSGS;
-#else
-static int concurrent_sends = IBLND_MSG_QUEUE_SIZE;
-#endif
+static int concurrent_sends = 0;
CFS_MODULE_PARM(concurrent_sends, "i", int, 0444,
"send work-queue sizing");
-#if IBLND_MAP_ON_DEMAND
+static int map_on_demand = 0;
+CFS_MODULE_PARM(map_on_demand, "i", int, 0444,
+ "map on demand");
+
+/* NB: this value is shared by all CPTs, it can grow at runtime */
static int fmr_pool_size = 512;
CFS_MODULE_PARM(fmr_pool_size, "i", int, 0444,
- "size of the fmr pool (>= ntx)");
+ "size of fmr pool on each CPT (>= ntx / 4)");
+/* NB: this value is shared by all CPTs, it can grow at runtime */
static int fmr_flush_trigger = 384;
CFS_MODULE_PARM(fmr_flush_trigger, "i", int, 0444,
- "# dirty FMRs that triggers pool flush");
+ "# dirty FMRs that triggers pool flush");
static int fmr_cache = 1;
CFS_MODULE_PARM(fmr_cache, "i", int, 0444,
- "non-zero to enable FMR caching");
-#endif
+ "non-zero to enable FMR caching");
+
+/* NB: this value is shared by all CPTs, it can grow at runtime */
+static int pmr_pool_size = 512;
+CFS_MODULE_PARM(pmr_pool_size, "i", int, 0444,
+ "size of MR cache pmr pool on each CPT");
+
+/*
+ * 0: disable failover
+ * 1: enable failover if necessary
+ * 2: force to failover (for debug)
+ */
+static int dev_failover = 0;
+CFS_MODULE_PARM(dev_failover, "i", int, 0444,
+ "HCA failover for bonding (0 off, 1 on, other values reserved)");
+
+
+static int require_privileged_port = 0;
+CFS_MODULE_PARM(require_privileged_port, "i", int, 0644,
+ "require privileged port when accepting connection");
+
+static int use_privileged_port = 1;
+CFS_MODULE_PARM(use_privileged_port, "i", int, 0644,
+ "use privileged port when initiating connection");
kib_tunables_t kiblnd_tunables = {
+ .kib_dev_failover = &dev_failover,
.kib_service = &service,
.kib_cksum = &cksum,
.kib_timeout = &timeout,
.kib_keepalive = &keepalive,
.kib_ntx = &ntx,
.kib_credits = &credits,
- .kib_peercredits = &peer_credits,
+ .kib_peertxcredits = &peer_credits,
+ .kib_peercredits_hiw = &peer_credits_hiw,
+ .kib_peerrtrcredits = &peer_buffer_credits,
+ .kib_peertimeout = &peer_timeout,
.kib_default_ipif = &ipif_name,
.kib_retry_count = &retry_count,
.kib_rnr_retry_count = &rnr_retry_count,
.kib_concurrent_sends = &concurrent_sends,
.kib_ib_mtu = &ib_mtu,
-#if IBLND_MAP_ON_DEMAND
+ .kib_map_on_demand = &map_on_demand,
.kib_fmr_pool_size = &fmr_pool_size,
.kib_fmr_flush_trigger = &fmr_flush_trigger,
.kib_fmr_cache = &fmr_cache,
-#endif
+ .kib_pmr_pool_size = &pmr_pool_size,
+ .kib_require_priv_port = &require_privileged_port,
+ .kib_use_priv_port = &use_privileged_port,
+ .kib_nscheds = &nscheds
};
#if defined(CONFIG_SYSCTL) && !CFS_SYSFS_MODULE_PARM
static char ipif_basename_space[32];
-#ifndef HAVE_SYSCTL_UNNUMBERED
-
-enum {
- O2IBLND_SERVICE = 1,
- O2IBLND_CKSUM,
- O2IBLND_TIMEOUT,
- O2IBLND_NTX,
- O2IBLND_CREDITS,
- O2IBLND_PEER_CREDITS,
- O2IBLND_IPIF_BASENAME,
- O2IBLND_RETRY_COUNT,
- O2IBLND_RNR_RETRY_COUNT,
- O2IBLND_KEEPALIVE,
- O2IBLND_CONCURRENT_SENDS,
- O2IBLND_IB_MTU,
- O2IBLND_FMR_POOL_SIZE,
- O2IBLND_FMR_FLUSH_TRIGGER,
- O2IBLND_FMR_CACHE
+static struct ctl_table kiblnd_ctl_table[] = {
+ {
+ INIT_CTL_NAME
+ .procname = "service",
+ .data = &service,
+ .maxlen = sizeof(int),
+ .mode = 0444,
+ .proc_handler = &proc_dointvec
+ },
+ {
+ INIT_CTL_NAME
+ .procname = "cksum",
+ .data = &cksum,
+ .maxlen = sizeof(int),
+ .mode = 0644,
+ .proc_handler = &proc_dointvec
+ },
+ {
+ INIT_CTL_NAME
+ .procname = "timeout",
+ .data = &timeout,
+ .maxlen = sizeof(int),
+ .mode = 0644,
+ .proc_handler = &proc_dointvec
+ },
+ {
+ INIT_CTL_NAME
+ .procname = "ntx",
+ .data = &ntx,
+ .maxlen = sizeof(int),
+ .mode = 0444,
+ .proc_handler = &proc_dointvec
+ },
+ {
+ INIT_CTL_NAME
+ .procname = "credits",
+ .data = &credits,
+ .maxlen = sizeof(int),
+ .mode = 0444,
+ .proc_handler = &proc_dointvec
+ },
+ {
+ INIT_CTL_NAME
+ .procname = "peer_credits",
+ .data = &peer_credits,
+ .maxlen = sizeof(int),
+ .mode = 0444,
+ .proc_handler = &proc_dointvec
+ },
+ {
+ INIT_CTL_NAME
+ .procname = "peer_credits_hiw",
+ .data = &peer_credits_hiw,
+ .maxlen = sizeof(int),
+ .mode = 0444,
+ .proc_handler = &proc_dointvec
+ },
+ {
+ INIT_CTL_NAME
+ .procname = "peer_buffer_credits",
+ .data = &peer_buffer_credits,
+ .maxlen = sizeof(int),
+ .mode = 0444,
+ .proc_handler = &proc_dointvec
+ },
+ {
+ INIT_CTL_NAME
+ .procname = "peer_timeout",
+ .data = &peer_timeout,
+ .maxlen = sizeof(int),
+ .mode = 0444,
+ .proc_handler = &proc_dointvec
+ },
+ {
+ INIT_CTL_NAME
+ .procname = "ipif_name",
+ .data = ipif_basename_space,
+ .maxlen = sizeof(ipif_basename_space),
+ .mode = 0444,
+ .proc_handler = &proc_dostring
+ },
+ {
+ INIT_CTL_NAME
+ .procname = "retry_count",
+ .data = &retry_count,
+ .maxlen = sizeof(int),
+ .mode = 0644,
+ .proc_handler = &proc_dointvec
+ },
+ {
+ INIT_CTL_NAME
+ .procname = "rnr_retry_count",
+ .data = &rnr_retry_count,
+ .maxlen = sizeof(int),
+ .mode = 0644,
+ .proc_handler = &proc_dointvec
+ },
+ {
+ INIT_CTL_NAME
+ .procname = "keepalive",
+ .data = &keepalive,
+ .maxlen = sizeof(int),
+ .mode = 0644,
+ .proc_handler = &proc_dointvec
+ },
+ {
+ INIT_CTL_NAME
+ .procname = "concurrent_sends",
+ .data = &concurrent_sends,
+ .maxlen = sizeof(int),
+ .mode = 0444,
+ .proc_handler = &proc_dointvec
+ },
+ {
+ INIT_CTL_NAME
+ .procname = "ib_mtu",
+ .data = &ib_mtu,
+ .maxlen = sizeof(int),
+ .mode = 0444,
+ .proc_handler = &proc_dointvec
+ },
+ {
+ INIT_CTL_NAME
+ .procname = "map_on_demand",
+ .data = &map_on_demand,
+ .maxlen = sizeof(int),
+ .mode = 0444,
+ .proc_handler = &proc_dointvec
+ },
+ {
+ INIT_CTL_NAME
+ .procname = "fmr_pool_size",
+ .data = &fmr_pool_size,
+ .maxlen = sizeof(int),
+ .mode = 0444,
+ .proc_handler = &proc_dointvec
+ },
+ {
+ INIT_CTL_NAME
+ .procname = "fmr_flush_trigger",
+ .data = &fmr_flush_trigger,
+ .maxlen = sizeof(int),
+ .mode = 0444,
+ .proc_handler = &proc_dointvec
+ },
+ {
+ INIT_CTL_NAME
+ .procname = "fmr_cache",
+ .data = &fmr_cache,
+ .maxlen = sizeof(int),
+ .mode = 0444,
+ .proc_handler = &proc_dointvec
+ },
+ {
+ INIT_CTL_NAME
+ .procname = "pmr_pool_size",
+ .data = &pmr_pool_size,
+ .maxlen = sizeof(int),
+ .mode = 0444,
+ .proc_handler = &proc_dointvec
+ },
+ {
+ INIT_CTL_NAME
+ .procname = "dev_failover",
+ .data = &dev_failover,
+ .maxlen = sizeof(int),
+ .mode = 0444,
+ .proc_handler = &proc_dointvec
+ },
+ { 0 }
};
-#else
-#define O2IBLND_SERVICE CTL_UNNUMBERED
-#define O2IBLND_CKSUM CTL_UNNUMBERED
-#define O2IBLND_TIMEOUT CTL_UNNUMBERED
-#define O2IBLND_NTX CTL_UNNUMBERED
-#define O2IBLND_CREDITS CTL_UNNUMBERED
-#define O2IBLND_PEER_CREDITS CTL_UNNUMBERED
-#define O2IBLND_IPIF_BASENAME CTL_UNNUMBERED
-#define O2IBLND_RETRY_COUNT CTL_UNNUMBERED
-#define O2IBLND_RNR_RETRY_COUNT CTL_UNNUMBERED
-#define O2IBLND_KEEPALIVE CTL_UNNUMBERED
-#define O2IBLND_CONCURRENT_SENDS CTL_UNNUMBERED
-#define O2IBLND_IB_MTU CTL_UNNUMBERED
-#define O2IBLND_FMR_POOL_SIZE CTL_UNNUMBERED
-#define O2IBLND_FMR_FLUSH_TRIGGER CTL_UNNUMBERED
-#define O2IBLND_FMR_CACHE CTL_UNNUMBERED
-
-#endif
-
-static cfs_sysctl_table_t kiblnd_ctl_table[] = {
- {
- .ctl_name = O2IBLND_SERVICE,
- .procname = "service",
- .data = &service,
- .maxlen = sizeof(int),
- .mode = 0444,
- .proc_handler = &proc_dointvec
- },
- {
- .ctl_name = O2IBLND_CKSUM,
- .procname = "cksum",
- .data = &cksum,
- .maxlen = sizeof(int),
- .mode = 0644,
- .proc_handler = &proc_dointvec
- },
- {
- .ctl_name = O2IBLND_TIMEOUT,
- .procname = "timeout",
- .data = &timeout,
- .maxlen = sizeof(int),
- .mode = 0644,
- .proc_handler = &proc_dointvec
- },
- {
- .ctl_name = O2IBLND_NTX,
- .procname = "ntx",
- .data = &ntx,
- .maxlen = sizeof(int),
- .mode = 0444,
- .proc_handler = &proc_dointvec
- },
- {
- .ctl_name = O2IBLND_CREDITS,
- .procname = "credits",
- .data = &credits,
- .maxlen = sizeof(int),
- .mode = 0444,
- .proc_handler = &proc_dointvec
- },
- {
- .ctl_name = O2IBLND_PEER_CREDITS,
- .procname = "peer_credits",
- .data = &peer_credits,
- .maxlen = sizeof(int),
- .mode = 0444,
- .proc_handler = &proc_dointvec
- },
- {
- .ctl_name = O2IBLND_IPIF_BASENAME,
- .procname = "ipif_name",
- .data = ipif_basename_space,
- .maxlen = sizeof(ipif_basename_space),
- .mode = 0444,
- .proc_handler = &proc_dostring
- },
- {
- .ctl_name = O2IBLND_RETRY_COUNT,
- .procname = "retry_count",
- .data = &retry_count,
- .maxlen = sizeof(int),
- .mode = 0644,
- .proc_handler = &proc_dointvec
- },
- {
- .ctl_name = O2IBLND_RNR_RETRY_COUNT,
- .procname = "rnr_retry_count",
- .data = &rnr_retry_count,
- .maxlen = sizeof(int),
- .mode = 0644,
- .proc_handler = &proc_dointvec
- },
- {
- .ctl_name = O2IBLND_KEEPALIVE,
- .procname = "keepalive",
- .data = &keepalive,
- .maxlen = sizeof(int),
- .mode = 0644,
- .proc_handler = &proc_dointvec
- },
- {
- .ctl_name = O2IBLND_CONCURRENT_SENDS,
- .procname = "concurrent_sends",
- .data = &concurrent_sends,
- .maxlen = sizeof(int),
- .mode = 0444,
- .proc_handler = &proc_dointvec
- },
- {
- .ctl_name = O2IBLND_IB_MTU,
- .procname = "ib_mtu",
- .data = &ib_mtu,
- .maxlen = sizeof(int),
- .mode = 0444,
- .proc_handler = &proc_dointvec
- },
-#if IBLND_MAP_ON_DEMAND
- {
- .ctl_name = O2IBLND_FMR_POOL_SIZE,
- .procname = "fmr_pool_size",
- .data = &fmr_pool_size,
- .maxlen = sizeof(int),
- .mode = 0444,
- .proc_handler = &proc_dointvec
- },
- {
- .ctl_name = O2IBLND_FMR_FLUSH_TRIGGER,
- .procname = "fmr_flush_trigger",
- .data = &fmr_flush_trigger,
- .maxlen = sizeof(int),
- .mode = 0444,
- .proc_handler = &proc_dointvec
- },
- {
- .ctl_name = O2IBLND_FMR_CACHE,
- .procname = "fmr_cache",
- .data = &fmr_cache,
- .maxlen = sizeof(int),
- .mode = 0444,
- .proc_handler = &proc_dointvec
- },
-#endif
- {0}
-};
-
-static cfs_sysctl_table_t kiblnd_top_ctl_table[] = {
- {
- .ctl_name = CTL_O2IBLND,
- .procname = "o2iblnd",
- .data = NULL,
- .maxlen = 0,
- .mode = 0555,
- .child = kiblnd_ctl_table
- },
- {0}
+static struct ctl_table kiblnd_top_ctl_table[] = {
+ {
+ INIT_CTL_NAME
+ .procname = "o2iblnd",
+ .data = NULL,
+ .maxlen = 0,
+ .mode = 0555,
+ .child = kiblnd_ctl_table
+ },
+ { 0 }
};
void
void
kiblnd_sysctl_init (void)
{
- kiblnd_initstrtunable(ipif_basename_space, ipif_name,
- sizeof(ipif_basename_space));
+ kiblnd_initstrtunable(ipif_basename_space, ipif_name,
+ sizeof(ipif_basename_space));
- kiblnd_tunables.kib_sysctl =
- cfs_register_sysctl_table(kiblnd_top_ctl_table, 0);
+ kiblnd_tunables.kib_sysctl =
+ register_sysctl_table(kiblnd_top_ctl_table);
- if (kiblnd_tunables.kib_sysctl == NULL)
- CWARN("Can't setup /proc tunables\n");
+ if (kiblnd_tunables.kib_sysctl == NULL)
+ CWARN("Can't setup /proc tunables\n");
}
void
kiblnd_sysctl_fini (void)
{
- if (kiblnd_tunables.kib_sysctl != NULL)
- cfs_unregister_sysctl_table(kiblnd_tunables.kib_sysctl);
+ if (kiblnd_tunables.kib_sysctl != NULL)
+ unregister_sysctl_table(kiblnd_tunables.kib_sysctl);
}
#else
int
kiblnd_tunables_init (void)
{
- if (*kiblnd_tunables.kib_credits > *kiblnd_tunables.kib_ntx) {
- CERROR("Can't set credits(%d) > ntx(%d)\n",
- *kiblnd_tunables.kib_credits,
- *kiblnd_tunables.kib_ntx);
- return -EINVAL;
- }
-
- if (*kiblnd_tunables.kib_ib_mtu != 0 &&
- *kiblnd_tunables.kib_ib_mtu != 256 &&
- *kiblnd_tunables.kib_ib_mtu != 512 &&
- *kiblnd_tunables.kib_ib_mtu != 1024 &&
- *kiblnd_tunables.kib_ib_mtu != 2048 &&
- *kiblnd_tunables.kib_ib_mtu != 4096) {
+ if (kiblnd_translate_mtu(*kiblnd_tunables.kib_ib_mtu) < 0) {
CERROR("Invalid ib_mtu %d, expected 256/512/1024/2048/4096\n",
*kiblnd_tunables.kib_ib_mtu);
return -EINVAL;
}
- if (*kiblnd_tunables.kib_concurrent_sends > IBLND_RX_MSGS)
- *kiblnd_tunables.kib_concurrent_sends = IBLND_RX_MSGS;
- if (*kiblnd_tunables.kib_concurrent_sends < IBLND_MSG_QUEUE_SIZE / 2)
- *kiblnd_tunables.kib_concurrent_sends = IBLND_MSG_QUEUE_SIZE / 2;
+ if (*kiblnd_tunables.kib_peertxcredits < IBLND_CREDITS_DEFAULT)
+ *kiblnd_tunables.kib_peertxcredits = IBLND_CREDITS_DEFAULT;
+
+ if (*kiblnd_tunables.kib_peertxcredits > IBLND_CREDITS_MAX)
+ *kiblnd_tunables.kib_peertxcredits = IBLND_CREDITS_MAX;
+
+ if (*kiblnd_tunables.kib_peertxcredits > *kiblnd_tunables.kib_credits)
+ *kiblnd_tunables.kib_peertxcredits = *kiblnd_tunables.kib_credits;
+
+ if (*kiblnd_tunables.kib_peercredits_hiw < *kiblnd_tunables.kib_peertxcredits / 2)
+ *kiblnd_tunables.kib_peercredits_hiw = *kiblnd_tunables.kib_peertxcredits / 2;
+
+ if (*kiblnd_tunables.kib_peercredits_hiw >= *kiblnd_tunables.kib_peertxcredits)
+ *kiblnd_tunables.kib_peercredits_hiw = *kiblnd_tunables.kib_peertxcredits - 1;
+
+ if (*kiblnd_tunables.kib_map_on_demand < 0 ||
+ *kiblnd_tunables.kib_map_on_demand > IBLND_MAX_RDMA_FRAGS)
+ *kiblnd_tunables.kib_map_on_demand = 0; /* disable map-on-demand */
+
+ if (*kiblnd_tunables.kib_map_on_demand == 1)
+ *kiblnd_tunables.kib_map_on_demand = 2; /* don't make sense to create map if only one fragment */
+
+ if (*kiblnd_tunables.kib_concurrent_sends == 0) {
+ if (*kiblnd_tunables.kib_map_on_demand > 0 &&
+ *kiblnd_tunables.kib_map_on_demand <= IBLND_MAX_RDMA_FRAGS / 8)
+ *kiblnd_tunables.kib_concurrent_sends = (*kiblnd_tunables.kib_peertxcredits) * 2;
+ else
+ *kiblnd_tunables.kib_concurrent_sends = (*kiblnd_tunables.kib_peertxcredits);
+ }
+
+ if (*kiblnd_tunables.kib_concurrent_sends > *kiblnd_tunables.kib_peertxcredits * 2)
+ *kiblnd_tunables.kib_concurrent_sends = *kiblnd_tunables.kib_peertxcredits * 2;
+
+ if (*kiblnd_tunables.kib_concurrent_sends < *kiblnd_tunables.kib_peertxcredits / 2)
+ *kiblnd_tunables.kib_concurrent_sends = *kiblnd_tunables.kib_peertxcredits / 2;
- if (*kiblnd_tunables.kib_concurrent_sends < IBLND_MSG_QUEUE_SIZE) {
+ if (*kiblnd_tunables.kib_concurrent_sends < *kiblnd_tunables.kib_peertxcredits) {
CWARN("Concurrent sends %d is lower than message queue size: %d, "
"performance may drop slightly.\n",
- *kiblnd_tunables.kib_concurrent_sends, IBLND_MSG_QUEUE_SIZE);
+ *kiblnd_tunables.kib_concurrent_sends, *kiblnd_tunables.kib_peertxcredits);
}
kiblnd_sysctl_init();
git://git.whamcloud.com / fs / lustre-release.git / blobdiff