X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lnet%2Fklnds%2Fo2iblnd%2Fo2iblnd_modparams.c;h=5a14fb24c7c60e7b258b6dc9f4139e3b0f24149d;hp=fa063e562b4cead2238be1a5d5926bafe37c3bc1;hb=HEAD;hpb=d96a9248708d4da02728c9976a9a90ba29bd2bc0 diff --git a/lnet/klnds/o2iblnd/o2iblnd_modparams.c b/lnet/klnds/o2iblnd/o2iblnd_modparams.c index fa063e5..ffb68e5 100644 --- a/lnet/klnds/o2iblnd/o2iblnd_modparams.c +++ b/lnet/klnds/o2iblnd/o2iblnd_modparams.c @@ -15,11 +15,7 @@ * * You should have received a copy of the GNU General Public License * version 2 along with this program; If not, see - * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf - * - * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, - * CA 95054 USA or visit www.sun.com if you need additional information or - * have any questions. + * http://www.gnu.org/licenses/gpl-2.0.html * * GPL HEADER END */ @@ -27,11 +23,10 @@ * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * - * Copyright (c) 2012, Intel Corporation. + * Copyright (c) 2012, 2017, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ - * Lustre is a trademark of Sun Microsystems, Inc. * * lnet/klnds/o2iblnd/o2iblnd_modparams.c * @@ -40,6 +35,8 @@ #include "o2iblnd.h" +#define CURRENT_LND_VERSION 1 + static int service = 987; module_param(service, int, 0444); MODULE_PARM_DESC(service, "service number (within RDMA_PS_TCP)"); @@ -48,7 +45,7 @@ static int cksum = 0; module_param(cksum, int, 0644); MODULE_PARM_DESC(cksum, "set non-zero to enable message (not RDMA) checksums"); -static int timeout = 50; +static int timeout; module_param(timeout, int, 0644); MODULE_PARM_DESC(timeout, "timeout (seconds)"); @@ -58,17 +55,21 @@ static int nscheds; module_param(nscheds, int, 0444); MODULE_PARM_DESC(nscheds, "number of threads in each scheduler pool"); +static unsigned int conns_per_peer = 1; +module_param(conns_per_peer, uint, 0444); +MODULE_PARM_DESC(conns_per_peer, "number of connections per peer"); + /* NB: this value is shared by all CPTs, it can grow at runtime */ static int ntx = 512; module_param(ntx, int, 0444); MODULE_PARM_DESC(ntx, "# of message descriptors allocated for each pool"); /* NB: this value is shared by all CPTs */ -static int credits = 256; +static int credits = DEFAULT_CREDITS; module_param(credits, int, 0444); MODULE_PARM_DESC(credits, "# concurrent sends"); -static int peer_credits = 8; +static int peer_credits = DEFAULT_PEER_CREDITS; module_param(peer_credits, int, 0444); MODULE_PARM_DESC(peer_credits, "# concurrent sends to 1 peer"); @@ -80,7 +81,7 @@ static int peer_buffer_credits = 0; module_param(peer_buffer_credits, int, 0444); MODULE_PARM_DESC(peer_buffer_credits, "# per-peer router buffer credits"); -static int peer_timeout = 180; +static int peer_timeout = DEFAULT_PEER_TIMEOUT; module_param(peer_timeout, int, 0444); MODULE_PARM_DESC(peer_timeout, "Seconds without aliveness news to declare peer dead (<=0 to disable)"); @@ -90,7 +91,7 @@ MODULE_PARM_DESC(ipif_name, "IPoIB interface name"); static int retry_count = 5; module_param(retry_count, int, 0644); -MODULE_PARM_DESC(retry_count, "Retransmissions when no ACK received"); +MODULE_PARM_DESC(retry_count, "Number of times to retry connection operations"); static int rnr_retry_count = 6; module_param(rnr_retry_count, int, 0644); @@ -108,9 +109,45 @@ static int concurrent_sends; module_param(concurrent_sends, int, 0444); MODULE_PARM_DESC(concurrent_sends, "send work-queue sizing"); -static int map_on_demand; +static int use_fastreg_gaps; +module_param(use_fastreg_gaps, int, 0444); +MODULE_PARM_DESC(use_fastreg_gaps, "Enable discontiguous fastreg fragment support. Expect performance drop"); + +/* + * map_on_demand is a flag used to determine if we can use FMR or FastReg. + * This is applicable for kernels which support global memory regions. For + * later kernels this flag is always enabled, since we will always either + * use FMR or FastReg + * For kernels which support global memory regions map_on_demand defaults + * to 0 which means we will be using global memory regions exclusively. + * If it is set to a value other than 0, then we will behave as follows: + * 1. Always default the number of fragments to IBLND_MAX_RDMA_FRAGS + * 2. Create FMR/FastReg pools + * 3. Negotiate the supported number of fragments per connection + * 4. Attempt to transmit using global memory regions only if + * map-on-demand is not turned on, otherwise use FMR or FastReg + * 5. In case of transmitting tx with GAPS over FMR we will need to + * transmit it with multiple fragments. Look at the comments in + * kiblnd_fmr_map_tx() for an explanation of the behavior. + * + * For later kernels we default map_on_demand to 1 and not allow + * it to be set to 0, since there is no longer support for global memory + * regions. Behavior: + * 1. Default the number of fragments to IBLND_MAX_RDMA_FRAGS + * 2. Create FMR/FastReg pools + * 3. Negotiate the supported number of fragments per connection + * 4. Look at the comments in kiblnd_fmr_map_tx() for an explanation of + * the behavior when transmit with GAPS verses contiguous. + */ + +#ifdef HAVE_OFED_IB_GET_DMA_MR +#define MOD_STR "map on demand" +#else +#define MOD_STR "map on demand (obsolete)" +#endif +static int map_on_demand = 1; module_param(map_on_demand, int, 0444); -MODULE_PARM_DESC(map_on_demand, "map on demand"); +MODULE_PARM_DESC(map_on_demand, MOD_STR); /* NB: this value is shared by all CPTs, it can grow at runtime */ static int fmr_pool_size = 512; @@ -143,346 +180,163 @@ static int use_privileged_port = 1; module_param(use_privileged_port, int, 0644); MODULE_PARM_DESC(use_privileged_port, "use privileged port when initiating connection"); -kib_tunables_t kiblnd_tunables = { +static unsigned int wrq_sge = 2; +module_param(wrq_sge, uint, 0444); +MODULE_PARM_DESC(wrq_sge, "# scatter/gather element per work request"); + +static int tos = -1; +static int param_set_tos(const char *val, cfs_kernel_param_arg_t *kp); +#ifdef HAVE_KERNEL_PARAM_OPS +static const struct kernel_param_ops param_ops_tos = { + .set = param_set_tos, + .get = param_get_int, +}; + +#define param_check_tos(name, p) \ + __param_check(name, p, int) +module_param(tos, tos, 0444); +#else +module_param_call(tos, param_set_tos, param_get_int, &tos, 0444); +#endif +MODULE_PARM_DESC(tos, "Set the type of service (=-1 to disable)"); + +struct kib_tunables kiblnd_tunables = { .kib_dev_failover = &dev_failover, .kib_service = &service, .kib_cksum = &cksum, .kib_timeout = &timeout, .kib_keepalive = &keepalive, - .kib_ntx = &ntx, .kib_default_ipif = &ipif_name, .kib_retry_count = &retry_count, .kib_rnr_retry_count = &rnr_retry_count, .kib_ib_mtu = &ib_mtu, .kib_require_priv_port = &require_privileged_port, .kib_use_priv_port = &use_privileged_port, - .kib_nscheds = &nscheds -}; - -static struct lnet_ioctl_config_o2iblnd_tunables default_tunables; - -#if defined(CONFIG_SYSCTL) && !CFS_SYSFS_MODULE_PARM - -static char ipif_basename_space[32]; - -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 = "dev_failover", - .data = &dev_failover, - .maxlen = sizeof(int), - .mode = 0444, - .proc_handler = &proc_dointvec - }, - { 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 } + .kib_nscheds = &nscheds, + .kib_wrq_sge = &wrq_sge, + .kib_use_fastreg_gaps = &use_fastreg_gaps, }; -void -kiblnd_initstrtunable(char *space, char *str, int size) -{ - strncpy(space, str, size); - space[size-1] = 0; -} +struct lnet_ioctl_config_o2iblnd_tunables kib_default_tunables; -static void -kiblnd_sysctl_init (void) +static int param_set_tos(const char *val, cfs_kernel_param_arg_t *kp) { - kiblnd_initstrtunable(ipif_basename_space, ipif_name, - sizeof(ipif_basename_space)); + int rc, t; - 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 (!val) + return -EINVAL; -static void -kiblnd_sysctl_fini (void) -{ - if (kiblnd_tunables.kib_sysctl != NULL) - unregister_sysctl_table(kiblnd_tunables.kib_sysctl); -} + rc = kstrtoint(val, 0, &t); + if (rc) + return rc; -#else + if (t < -1 || t > 0xff) + return -ERANGE; -static void -kiblnd_sysctl_init (void) -{ -} + *((int *)kp->arg) = t; -static void -kiblnd_sysctl_fini (void) -{ + return 0; } -#endif /* # messages/RDMAs in-flight */ int -kiblnd_msg_queue_size(int version, lnet_ni_t *ni) +kiblnd_msg_queue_size(int version, struct lnet_ni *ni) { if (version == IBLND_MSG_VERSION_1) return IBLND_MSG_QUEUE_SIZE_V1; else if (ni) - return ni->ni_peertxcredits; + return ni->ni_net->net_tunables.lct_peer_tx_credits; else return peer_credits; } int -kiblnd_tunables_setup(lnet_ni_t *ni) +kiblnd_tunables_setup(struct lnet_ni *ni) { struct lnet_ioctl_config_o2iblnd_tunables *tunables; + struct lnet_ioctl_config_lnd_cmn_tunables *net_tunables; /* * if there was no tunables specified, setup the tunables to be * defaulted */ - if (!ni->ni_lnd_tunables) { - LIBCFS_ALLOC(ni->ni_lnd_tunables, - sizeof(*ni->ni_lnd_tunables)); - if (!ni->ni_lnd_tunables) - return -ENOMEM; - - memcpy(&ni->ni_lnd_tunables->lt_tun_u.lt_o2ib, - &default_tunables, sizeof(*tunables)); - } - tunables = &ni->ni_lnd_tunables->lt_tun_u.lt_o2ib; + if (!ni->ni_lnd_tunables_set) + memcpy(&ni->ni_lnd_tunables.lnd_tun_u.lnd_o2ib, + &kib_default_tunables, sizeof(*tunables)); + + tunables = &ni->ni_lnd_tunables.lnd_tun_u.lnd_o2ib; /* Current API version */ - tunables->lnd_version = 0; + tunables->lnd_version = CURRENT_LND_VERSION; - if (kiblnd_translate_mtu(*kiblnd_tunables.kib_ib_mtu) < 0) { + if (*kiblnd_tunables.kib_ib_mtu && + ib_mtu_enum_to_int(ib_mtu_int_to_enum(*kiblnd_tunables.kib_ib_mtu)) != + *kiblnd_tunables.kib_ib_mtu) { CERROR("Invalid ib_mtu %d, expected 256/512/1024/2048/4096\n", *kiblnd_tunables.kib_ib_mtu); return -EINVAL; } - if (!ni->ni_peertimeout) - ni->ni_peertimeout = peer_timeout; + net_tunables = &ni->ni_net->net_tunables; - if (!ni->ni_maxtxcredits) - ni->ni_maxtxcredits = credits; + if (net_tunables->lct_peer_timeout == -1) + net_tunables->lct_peer_timeout = peer_timeout; - if (!ni->ni_peertxcredits) - ni->ni_peertxcredits = peer_credits; + if (net_tunables->lct_max_tx_credits == -1) + net_tunables->lct_max_tx_credits = credits; - if (!ni->ni_peerrtrcredits) - ni->ni_peerrtrcredits = peer_buffer_credits; + if (net_tunables->lct_peer_tx_credits == -1) + net_tunables->lct_peer_tx_credits = peer_credits; - if (ni->ni_peertxcredits < IBLND_CREDITS_DEFAULT) - ni->ni_peertxcredits = IBLND_CREDITS_DEFAULT; + if (net_tunables->lct_peer_rtr_credits == -1) + net_tunables->lct_peer_rtr_credits = peer_buffer_credits; - if (ni->ni_peertxcredits > IBLND_CREDITS_MAX) - ni->ni_peertxcredits = IBLND_CREDITS_MAX; + if (net_tunables->lct_peer_tx_credits < IBLND_CREDITS_DEFAULT) + net_tunables->lct_peer_tx_credits = IBLND_CREDITS_DEFAULT; - if (ni->ni_peertxcredits > credits) - ni->ni_peertxcredits = credits; + if (net_tunables->lct_peer_tx_credits > IBLND_CREDITS_MAX) + net_tunables->lct_peer_tx_credits = IBLND_CREDITS_MAX; - if (!tunables->lnd_peercredits_hiw) - tunables->lnd_peercredits_hiw = peer_credits_hiw; + if (net_tunables->lct_peer_tx_credits > + net_tunables->lct_max_tx_credits) + net_tunables->lct_peer_tx_credits = + net_tunables->lct_max_tx_credits; - if (tunables->lnd_peercredits_hiw < ni->ni_peertxcredits / 2) - tunables->lnd_peercredits_hiw = ni->ni_peertxcredits / 2; + if (tunables->lnd_map_on_demand == UINT_MAX) + tunables->lnd_map_on_demand = map_on_demand; - if (tunables->lnd_peercredits_hiw >= ni->ni_peertxcredits) - tunables->lnd_peercredits_hiw = ni->ni_peertxcredits - 1; +#ifndef HAVE_OFED_IB_GET_DMA_MR + /* + * For kernels which do not support global memory regions, always + * enable map_on_demand + */ + if (tunables->lnd_map_on_demand == 0) + tunables->lnd_map_on_demand = 1; +#endif - if (tunables->lnd_map_on_demand < 0 || - tunables->lnd_map_on_demand > IBLND_MAX_RDMA_FRAGS) { - /* disable map-on-demand */ - tunables->lnd_map_on_demand = 0; - } + if (!tunables->lnd_peercredits_hiw) + tunables->lnd_peercredits_hiw = peer_credits_hiw; - if (tunables->lnd_map_on_demand == 1) { - /* don't make sense to create map if only one fragment */ - tunables->lnd_map_on_demand = 2; - } + if (tunables->lnd_peercredits_hiw < net_tunables->lct_peer_tx_credits / 2) + tunables->lnd_peercredits_hiw = net_tunables->lct_peer_tx_credits / 2; - if (tunables->lnd_concurrent_sends == 0) { - if (tunables->lnd_map_on_demand > 0 && - tunables->lnd_map_on_demand <= IBLND_MAX_RDMA_FRAGS / 8) { - tunables->lnd_concurrent_sends = - ni->ni_peertxcredits * 2; - } else { - tunables->lnd_concurrent_sends = ni->ni_peertxcredits; - } - } + if (tunables->lnd_peercredits_hiw >= net_tunables->lct_peer_tx_credits) + tunables->lnd_peercredits_hiw = net_tunables->lct_peer_tx_credits - 1; + + if (tunables->lnd_concurrent_sends == 0) + tunables->lnd_concurrent_sends = net_tunables->lct_peer_tx_credits; - if (tunables->lnd_concurrent_sends > ni->ni_peertxcredits * 2) - tunables->lnd_concurrent_sends = ni->ni_peertxcredits * 2; + if (tunables->lnd_concurrent_sends > net_tunables->lct_peer_tx_credits * 2) + tunables->lnd_concurrent_sends = net_tunables->lct_peer_tx_credits * 2; - if (tunables->lnd_concurrent_sends < ni->ni_peertxcredits / 2) - tunables->lnd_concurrent_sends = ni->ni_peertxcredits / 2; + if (tunables->lnd_concurrent_sends < net_tunables->lct_peer_tx_credits / 2) + tunables->lnd_concurrent_sends = net_tunables->lct_peer_tx_credits / 2; - if (tunables->lnd_concurrent_sends < ni->ni_peertxcredits) { + if (tunables->lnd_concurrent_sends < net_tunables->lct_peer_tx_credits) { CWARN("Concurrent sends %d is lower than message " "queue size: %d, performance may drop slightly.\n", - tunables->lnd_concurrent_sends, ni->ni_peertxcredits); + tunables->lnd_concurrent_sends, + net_tunables->lct_peer_tx_credits); } if (!tunables->lnd_fmr_pool_size) @@ -491,6 +345,15 @@ kiblnd_tunables_setup(lnet_ni_t *ni) tunables->lnd_fmr_flush_trigger = fmr_flush_trigger; if (!tunables->lnd_fmr_cache) tunables->lnd_fmr_cache = fmr_cache; + if (!tunables->lnd_ntx) + tunables->lnd_ntx = ntx; + if (!tunables->lnd_conns_per_peer) + tunables->lnd_conns_per_peer = (conns_per_peer) ? + conns_per_peer : 1; + if (tunables->lnd_tos < 0) + tunables->lnd_tos = tos; + + tunables->lnd_timeout = kiblnd_timeout(); return 0; } @@ -498,20 +361,15 @@ kiblnd_tunables_setup(lnet_ni_t *ni) int kiblnd_tunables_init(void) { - default_tunables.lnd_version = 0; - default_tunables.lnd_peercredits_hiw = peer_credits_hiw, - default_tunables.lnd_map_on_demand = map_on_demand; - default_tunables.lnd_concurrent_sends = concurrent_sends; - default_tunables.lnd_fmr_pool_size = fmr_pool_size; - default_tunables.lnd_fmr_flush_trigger = fmr_flush_trigger; - default_tunables.lnd_fmr_cache = fmr_cache; - - kiblnd_sysctl_init(); + kib_default_tunables.lnd_version = CURRENT_LND_VERSION; + kib_default_tunables.lnd_peercredits_hiw = peer_credits_hiw; + kib_default_tunables.lnd_map_on_demand = map_on_demand; + kib_default_tunables.lnd_concurrent_sends = concurrent_sends; + kib_default_tunables.lnd_fmr_pool_size = fmr_pool_size; + kib_default_tunables.lnd_fmr_flush_trigger = fmr_flush_trigger; + kib_default_tunables.lnd_fmr_cache = fmr_cache; + kib_default_tunables.lnd_ntx = ntx; + kib_default_tunables.lnd_conns_per_peer = conns_per_peer; + kib_default_tunables.lnd_tos = tos; return 0; } - -void -kiblnd_tunables_fini(void) -{ - kiblnd_sysctl_fini(); -}