X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lnet%2Flnet%2Fapi-ni.c;h=0a0b8a901ae016182cd3d69fc886e6f463cb4412;hb=bfb38a047e0fc12e68a7aa3a2d2ec7a6e9d696ed;hp=61f881f32827c76a5e83be4e182036f32230e185;hpb=e5c35288861cf8ca4aecc30747b61069f9d80bb5;p=fs%2Flustre-release.git diff --git a/lnet/lnet/api-ni.c b/lnet/lnet/api-ni.c index 61f881f..0a0b8a9 100644 --- a/lnet/lnet/api-ni.c +++ b/lnet/lnet/api-ni.c @@ -1,6 +1,4 @@ -/* -*- 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. @@ -26,8 +24,10 @@ * GPL HEADER END */ /* - * Copyright 2008 Sun Microsystems, Inc. All rights reserved + * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. + * + * Copyright (c) 2011, Whamcloud, Inc. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -44,6 +44,7 @@ #endif lnet_t the_lnet; /* THE state of the network */ +EXPORT_SYMBOL(the_lnet); #ifdef __KERNEL__ @@ -91,10 +92,10 @@ lnet_get_networks(void) void lnet_init_locks(void) { - spin_lock_init (&the_lnet.ln_lock); - cfs_waitq_init (&the_lnet.ln_waitq); - init_mutex(&the_lnet.ln_lnd_mutex); - init_mutex(&the_lnet.ln_api_mutex); + cfs_spin_lock_init(&the_lnet.ln_eq_wait_lock); + cfs_waitq_init(&the_lnet.ln_eq_waitq); + cfs_mutex_init(&the_lnet.ln_lnd_mutex); + cfs_mutex_init(&the_lnet.ln_api_mutex); } void @@ -123,7 +124,7 @@ lnet_get_networks (void) int len; int nob; int rc; - struct list_head *tmp; + cfs_list_t *tmp; #ifdef NOT_YET if (networks != NULL && ip2nets != NULL) { @@ -137,8 +138,8 @@ lnet_get_networks (void) return (rc == 0) ? networks : NULL; } #else - ip2nets = NULL; - rc = 0; + SET_BUT_UNUSED(ip2nets); + SET_BUT_UNUSED(rc); #endif if (networks != NULL) return networks; @@ -150,8 +151,8 @@ lnet_get_networks (void) *str = 0; sep = ""; - list_for_each (tmp, &the_lnet.ln_lnds) { - lnd_t *lnd = list_entry(tmp, lnd_t, lnd_list); + cfs_list_for_each (tmp, &the_lnet.ln_lnds) { + lnd_t *lnd = cfs_list_entry(tmp, lnd_t, lnd_list); nob = snprintf(str, len, "%s%s", sep, libcfs_lnd2str(lnd->lnd_type)); @@ -173,39 +174,75 @@ lnet_get_networks (void) void lnet_init_locks(void) { - the_lnet.ln_lock = 0; - the_lnet.ln_lnd_mutex = 0; - the_lnet.ln_api_mutex = 0; + the_lnet.ln_eq_wait_lock = 0; + the_lnet.ln_lnd_mutex = 0; + the_lnet.ln_api_mutex = 0; } void lnet_fini_locks(void) { - LASSERT (the_lnet.ln_api_mutex == 0); - LASSERT (the_lnet.ln_lnd_mutex == 0); - LASSERT (the_lnet.ln_lock == 0); + LASSERT(the_lnet.ln_api_mutex == 0); + LASSERT(the_lnet.ln_lnd_mutex == 0); + LASSERT(the_lnet.ln_eq_wait_lock == 0); } # else void lnet_init_locks(void) { - pthread_cond_init(&the_lnet.ln_cond, NULL); - pthread_mutex_init(&the_lnet.ln_lock, NULL); - pthread_mutex_init(&the_lnet.ln_lnd_mutex, NULL); - pthread_mutex_init(&the_lnet.ln_api_mutex, NULL); + pthread_cond_init(&the_lnet.ln_eq_cond, NULL); + pthread_mutex_init(&the_lnet.ln_eq_wait_lock, NULL); + pthread_mutex_init(&the_lnet.ln_lnd_mutex, NULL); + pthread_mutex_init(&the_lnet.ln_api_mutex, NULL); } void lnet_fini_locks(void) { - pthread_mutex_destroy(&the_lnet.ln_api_mutex); - pthread_mutex_destroy(&the_lnet.ln_lnd_mutex); - pthread_mutex_destroy(&the_lnet.ln_lock); - pthread_cond_destroy(&the_lnet.ln_cond); + pthread_mutex_destroy(&the_lnet.ln_api_mutex); + pthread_mutex_destroy(&the_lnet.ln_lnd_mutex); + pthread_mutex_destroy(&the_lnet.ln_eq_wait_lock); + pthread_cond_destroy(&the_lnet.ln_eq_cond); } # endif #endif +static int +lnet_create_locks(void) +{ + lnet_init_locks(); + + the_lnet.ln_res_lock = cfs_percpt_lock_alloc(lnet_cpt_table()); + if (the_lnet.ln_res_lock == NULL) + goto failed; + + the_lnet.ln_net_lock = cfs_percpt_lock_alloc(lnet_cpt_table()); + if (the_lnet.ln_net_lock == NULL) + goto failed; + + return 0; + + failed: + lnet_fini_locks(); + return -ENOMEM; +} + +static void +lnet_destroy_locks(void) +{ + if (the_lnet.ln_res_lock != NULL) { + cfs_percpt_lock_free(the_lnet.ln_res_lock); + the_lnet.ln_res_lock = NULL; + } + + if (the_lnet.ln_net_lock != NULL) { + cfs_percpt_lock_free(the_lnet.ln_net_lock); + the_lnet.ln_net_lock = NULL; + } + + lnet_fini_locks(); +} + void lnet_assert_wire_constants (void) { /* Wire protocol assertions generated by 'wirecheck' @@ -303,11 +340,11 @@ lnd_t * lnet_find_lnd_by_type (int type) { lnd_t *lnd; - struct list_head *tmp; + cfs_list_t *tmp; /* holding lnd mutex */ - list_for_each (tmp, &the_lnet.ln_lnds) { - lnd = list_entry(tmp, lnd_t, lnd_list); + cfs_list_for_each (tmp, &the_lnet.ln_lnds) { + lnd = cfs_list_entry(tmp, lnd_t, lnd_list); if ((int)lnd->lnd_type == type) return lnd; @@ -319,49 +356,81 @@ lnet_find_lnd_by_type (int type) void lnet_register_lnd (lnd_t *lnd) { - LNET_MUTEX_DOWN(&the_lnet.ln_lnd_mutex); + LNET_MUTEX_LOCK(&the_lnet.ln_lnd_mutex); LASSERT (the_lnet.ln_init); LASSERT (libcfs_isknown_lnd(lnd->lnd_type)); LASSERT (lnet_find_lnd_by_type(lnd->lnd_type) == NULL); - list_add_tail (&lnd->lnd_list, &the_lnet.ln_lnds); + cfs_list_add_tail (&lnd->lnd_list, &the_lnet.ln_lnds); lnd->lnd_refcount = 0; CDEBUG(D_NET, "%s LND registered\n", libcfs_lnd2str(lnd->lnd_type)); - LNET_MUTEX_UP(&the_lnet.ln_lnd_mutex); + LNET_MUTEX_UNLOCK(&the_lnet.ln_lnd_mutex); } +EXPORT_SYMBOL(lnet_register_lnd); void lnet_unregister_lnd (lnd_t *lnd) { - LNET_MUTEX_DOWN(&the_lnet.ln_lnd_mutex); + LNET_MUTEX_LOCK(&the_lnet.ln_lnd_mutex); LASSERT (the_lnet.ln_init); LASSERT (lnet_find_lnd_by_type(lnd->lnd_type) == lnd); LASSERT (lnd->lnd_refcount == 0); - list_del (&lnd->lnd_list); + cfs_list_del (&lnd->lnd_list); CDEBUG(D_NET, "%s LND unregistered\n", libcfs_lnd2str(lnd->lnd_type)); - LNET_MUTEX_UP(&the_lnet.ln_lnd_mutex); + LNET_MUTEX_UNLOCK(&the_lnet.ln_lnd_mutex); } +EXPORT_SYMBOL(lnet_unregister_lnd); -#ifndef LNET_USE_LIB_FREELIST - -int -lnet_descriptor_setup (void) +void +lnet_counters_get(lnet_counters_t *counters) { - return 0; + lnet_counters_t *ctr; + int i; + + memset(counters, 0, sizeof(*counters)); + + lnet_net_lock(LNET_LOCK_EX); + + cfs_percpt_for_each(ctr, i, the_lnet.ln_counters) { + counters->msgs_max += ctr->msgs_max; + counters->msgs_alloc += ctr->msgs_alloc; + counters->errors += ctr->errors; + counters->send_count += ctr->send_count; + counters->recv_count += ctr->recv_count; + counters->route_count += ctr->route_count; + counters->drop_length += ctr->drop_length; + counters->send_length += ctr->send_length; + counters->recv_length += ctr->recv_length; + counters->route_length += ctr->route_length; + counters->drop_length += ctr->drop_length; + + } + lnet_net_unlock(LNET_LOCK_EX); } +EXPORT_SYMBOL(lnet_counters_get); void -lnet_descriptor_cleanup (void) +lnet_counters_reset(void) { + lnet_counters_t *counters; + int i; + + lnet_net_lock(LNET_LOCK_EX); + + cfs_percpt_for_each(counters, i, the_lnet.ln_counters) + memset(counters, 0, sizeof(lnet_counters_t)); + + lnet_net_unlock(LNET_LOCK_EX); } +EXPORT_SYMBOL(lnet_counters_reset); -#else +#ifdef LNET_USE_LIB_FREELIST int lnet_freelist_init (lnet_freelist_t *fl, int n, int size) @@ -384,7 +453,7 @@ lnet_freelist_init (lnet_freelist_t *fl, int n, int size) do { memset (space, 0, size); - list_add ((struct list_head *)space, &fl->fl_list); + cfs_list_add ((cfs_list_t *)space, &fl->fl_list); space += size; } while (--n != 0); @@ -394,7 +463,7 @@ lnet_freelist_init (lnet_freelist_t *fl, int n, int size) void lnet_freelist_fini (lnet_freelist_t *fl) { - struct list_head *el; + cfs_list_t *el; int count; if (fl->fl_nobjs == 0) @@ -410,48 +479,7 @@ lnet_freelist_fini (lnet_freelist_t *fl) memset (fl, 0, sizeof (*fl)); } -int -lnet_descriptor_setup (void) -{ - /* NB on failure caller must still call lnet_descriptor_cleanup */ - /* ****** */ - int rc; - - memset (&the_lnet.ln_free_mes, 0, sizeof (the_lnet.ln_free_mes)); - memset (&the_lnet.ln_free_msgs, 0, sizeof (the_lnet.ln_free_msgs)); - memset (&the_lnet.ln_free_mds, 0, sizeof (the_lnet.ln_free_mds)); - memset (&the_lnet.ln_free_eqs, 0, sizeof (the_lnet.ln_free_eqs)); - - rc = lnet_freelist_init(&the_lnet.ln_free_mes, - MAX_MES, sizeof (lnet_me_t)); - if (rc != 0) - return (rc); - - rc = lnet_freelist_init(&the_lnet.ln_free_msgs, - MAX_MSGS, sizeof (lnet_msg_t)); - if (rc != 0) - return (rc); - - rc = lnet_freelist_init(&the_lnet.ln_free_mds, - MAX_MDS, sizeof (lnet_libmd_t)); - if (rc != 0) - return (rc); - - rc = lnet_freelist_init(&the_lnet.ln_free_eqs, - MAX_EQS, sizeof (lnet_eq_t)); - return (rc); -} - -void -lnet_descriptor_cleanup (void) -{ - lnet_freelist_fini (&the_lnet.ln_free_mes); - lnet_freelist_fini (&the_lnet.ln_free_msgs); - lnet_freelist_fini (&the_lnet.ln_free_mds); - lnet_freelist_fini (&the_lnet.ln_free_eqs); -} - -#endif +#endif /* LNET_USE_LIB_FREELIST */ __u64 lnet_create_interface_cookie (void) @@ -466,7 +494,7 @@ lnet_create_interface_cookie (void) int rc = gettimeofday (&tv, NULL); LASSERT (rc == 0); #else - do_gettimeofday(&tv); + cfs_gettimeofday(&tv); #endif cookie = tv.tv_sec; cookie *= 1000000; @@ -474,132 +502,184 @@ lnet_create_interface_cookie (void) return cookie; } -int -lnet_setup_handle_hash (void) +static char * +lnet_res_type2str(int type) { - int i; + switch (type) { + default: + LBUG(); + case LNET_COOKIE_TYPE_MD: + return "MD"; + case LNET_COOKIE_TYPE_ME: + return "ME"; + case LNET_COOKIE_TYPE_EQ: + return "EQ"; + } +} - /* Arbitrary choice of hash table size */ -#ifdef __KERNEL__ - the_lnet.ln_lh_hash_size = CFS_PAGE_SIZE / sizeof (struct list_head); -#else - the_lnet.ln_lh_hash_size = (MAX_MES + MAX_MDS + MAX_EQS)/4; -#endif - LIBCFS_ALLOC(the_lnet.ln_lh_hash_table, - the_lnet.ln_lh_hash_size * sizeof (struct list_head)); - if (the_lnet.ln_lh_hash_table == NULL) - return (-ENOMEM); +void +lnet_res_container_cleanup(struct lnet_res_container *rec) +{ + int count = 0; - for (i = 0; i < the_lnet.ln_lh_hash_size; i++) - CFS_INIT_LIST_HEAD (&the_lnet.ln_lh_hash_table[i]); + if (rec->rec_type == 0) /* not set yet, it's uninitialized */ + return; - the_lnet.ln_next_object_cookie = LNET_COOKIE_TYPES; + while (!cfs_list_empty(&rec->rec_active)) { + cfs_list_t *e = rec->rec_active.next; - return (0); -} + cfs_list_del_init(e); + if (rec->rec_type == LNET_COOKIE_TYPE_EQ) { + lnet_eq_free(cfs_list_entry(e, lnet_eq_t, eq_list)); -void -lnet_cleanup_handle_hash (void) -{ - if (the_lnet.ln_lh_hash_table == NULL) - return; + } else if (rec->rec_type == LNET_COOKIE_TYPE_MD) { + lnet_md_free(cfs_list_entry(e, lnet_libmd_t, md_list)); - LIBCFS_FREE(the_lnet.ln_lh_hash_table, - the_lnet.ln_lh_hash_size * sizeof (struct list_head)); -} + } else { /* NB: Active MEs should be attached on portals */ + LBUG(); + } + count++; + } -lnet_libhandle_t * -lnet_lookup_cookie (__u64 cookie, int type) -{ - /* ALWAYS called with LNET_LOCK held */ - struct list_head *list; - struct list_head *el; - unsigned int hash; + if (count > 0) { + /* Found alive MD/ME/EQ, user really should unlink/free + * all of them before finalize LNet, but if someone didn't, + * we have to recycle garbage for him */ + CERROR("%d active elements on exit of %s container\n", + count, lnet_res_type2str(rec->rec_type)); + } - if ((cookie & (LNET_COOKIE_TYPES - 1)) != type) - return (NULL); +#ifdef LNET_USE_LIB_FREELIST + lnet_freelist_fini(&rec->rec_freelist); +#endif + if (rec->rec_lh_hash != NULL) { + LIBCFS_FREE(rec->rec_lh_hash, + LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0])); + rec->rec_lh_hash = NULL; + } - hash = ((unsigned int)cookie) % the_lnet.ln_lh_hash_size; - list = &the_lnet.ln_lh_hash_table[hash]; + rec->rec_type = 0; /* mark it as finalized */ +} - list_for_each (el, list) { - lnet_libhandle_t *lh = list_entry (el, lnet_libhandle_t, - lh_hash_chain); +int +lnet_res_container_setup(struct lnet_res_container *rec, + int cpt, int type, int objnum, int objsz) +{ + int rc = 0; + int i; - if (lh->lh_cookie == cookie) - return (lh); - } + LASSERT(rec->rec_type == 0); + + rec->rec_type = type; + CFS_INIT_LIST_HEAD(&rec->rec_active); - return (NULL); +#ifdef LNET_USE_LIB_FREELIST + memset(&rec->rec_freelist, 0, sizeof(rec->rec_freelist)); + rc = lnet_freelist_init(&rec->rec_freelist, objnum, objsz); + if (rc != 0) + goto out; +#endif + rec->rec_lh_cookie = (cpt << LNET_COOKIE_TYPE_BITS) | type; + + /* Arbitrary choice of hash table size */ + LIBCFS_CPT_ALLOC(rec->rec_lh_hash, lnet_cpt_table(), cpt, + LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0])); + if (rec->rec_lh_hash == NULL) { + rc = -ENOMEM; + goto out; + } + + for (i = 0; i < LNET_LH_HASH_SIZE; i++) + CFS_INIT_LIST_HEAD(&rec->rec_lh_hash[i]); + + return 0; + +out: + CERROR("Failed to setup %s resource container\n", + lnet_res_type2str(type)); + lnet_res_container_cleanup(rec); + return rc; } -void -lnet_initialise_handle (lnet_libhandle_t *lh, int type) +static void +lnet_res_containers_destroy(struct lnet_res_container **recs) { - /* ALWAYS called with LNET_LOCK held */ - unsigned int hash; + struct lnet_res_container *rec; + int i; - LASSERT (type >= 0 && type < LNET_COOKIE_TYPES); - lh->lh_cookie = the_lnet.ln_next_object_cookie | type; - the_lnet.ln_next_object_cookie += LNET_COOKIE_TYPES; + cfs_percpt_for_each(rec, i, recs) + lnet_res_container_cleanup(rec); - hash = ((unsigned int)lh->lh_cookie) % the_lnet.ln_lh_hash_size; - list_add (&lh->lh_hash_chain, &the_lnet.ln_lh_hash_table[hash]); + cfs_percpt_free(recs); } -void -lnet_invalidate_handle (lnet_libhandle_t *lh) +static struct lnet_res_container ** +lnet_res_containers_create(int type, int objnum, int objsz) { - /* ALWAYS called with LNET_LOCK held */ - list_del (&lh->lh_hash_chain); + struct lnet_res_container **recs; + struct lnet_res_container *rec; + int rc; + int i; + + recs = cfs_percpt_alloc(lnet_cpt_table(), sizeof(*rec)); + if (recs == NULL) { + CERROR("Failed to allocate %s resource containers\n", + lnet_res_type2str(type)); + return NULL; + } + + cfs_percpt_for_each(rec, i, recs) { + rc = lnet_res_container_setup(rec, i, type, objnum, objsz); + if (rc != 0) { + lnet_res_containers_destroy(recs); + return NULL; + } + } + + return recs; } -int -lnet_init_finalizers(void) +lnet_libhandle_t * +lnet_res_lh_lookup(struct lnet_res_container *rec, __u64 cookie) { -#ifdef __KERNEL__ - int i; + /* ALWAYS called with lnet_res_lock held */ + cfs_list_t *head; + lnet_libhandle_t *lh; + unsigned int hash; - the_lnet.ln_nfinalizers = (int) num_online_cpus(); + if ((cookie & LNET_COOKIE_MASK) != rec->rec_type) + return NULL; - LIBCFS_ALLOC(the_lnet.ln_finalizers, - the_lnet.ln_nfinalizers * - sizeof(*the_lnet.ln_finalizers)); - if (the_lnet.ln_finalizers == NULL) { - CERROR("Can't allocate ln_finalizers\n"); - return -ENOMEM; - } + hash = cookie >> (LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS); + head = &rec->rec_lh_hash[hash & LNET_LH_HASH_MASK]; - for (i = 0; i < the_lnet.ln_nfinalizers; i++) - the_lnet.ln_finalizers[i] = NULL; -#else - the_lnet.ln_finalizing = 0; -#endif + cfs_list_for_each_entry(lh, head, lh_hash_chain) { + if (lh->lh_cookie == cookie) + return lh; + } - CFS_INIT_LIST_HEAD(&the_lnet.ln_finalizeq); - return 0; + return NULL; } void -lnet_fini_finalizers(void) +lnet_res_lh_initialize(struct lnet_res_container *rec, lnet_libhandle_t *lh) { -#ifdef __KERNEL__ - int i; + /* ALWAYS called with lnet_res_lock held */ + unsigned int ibits = LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS; + unsigned int hash; - for (i = 0; i < the_lnet.ln_nfinalizers; i++) - LASSERT (the_lnet.ln_finalizers[i] == NULL); + lh->lh_cookie = rec->rec_lh_cookie; + rec->rec_lh_cookie += 1 << ibits; - LIBCFS_FREE(the_lnet.ln_finalizers, - the_lnet.ln_nfinalizers * - sizeof(*the_lnet.ln_finalizers)); -#else - LASSERT (!the_lnet.ln_finalizing); -#endif - LASSERT (list_empty(&the_lnet.ln_finalizeq)); + hash = (lh->lh_cookie >> ibits) & LNET_LH_HASH_MASK; + + cfs_list_add(&lh->lh_hash_chain, &rec->rec_lh_hash[hash]); } #ifndef __KERNEL__ -/* Temporary workaround to allow uOSS and test programs force server +/** + * Reserved API - do not use. + * Temporary workaround to allow uOSS and test programs force server * mode in userspace. See comments near ln_server_mode_flag in * lnet/lib-types.h */ @@ -609,12 +689,14 @@ lnet_server_mode() { } #endif +int lnet_unprepare(void); + int lnet_prepare(lnet_pid_t requested_pid) { - /* Prepare to bring up the network */ - int rc = 0; - int i; + /* Prepare to bring up the network */ + struct lnet_res_container **recs; + int rc = 0; LASSERT (the_lnet.ln_refcount == 0); @@ -638,223 +720,293 @@ lnet_prepare(lnet_pid_t requested_pid) } #endif - rc = lnet_descriptor_setup(); - if (rc != 0) - goto failed0; + CFS_INIT_LIST_HEAD(&the_lnet.ln_test_peers); + CFS_INIT_LIST_HEAD(&the_lnet.ln_nis); + CFS_INIT_LIST_HEAD(&the_lnet.ln_nis_cpt); + CFS_INIT_LIST_HEAD(&the_lnet.ln_nis_zombie); + CFS_INIT_LIST_HEAD(&the_lnet.ln_remote_nets); + CFS_INIT_LIST_HEAD(&the_lnet.ln_routers); - memset(&the_lnet.ln_counters, 0, - sizeof(the_lnet.ln_counters)); + the_lnet.ln_interface_cookie = lnet_create_interface_cookie(); - CFS_INIT_LIST_HEAD (&the_lnet.ln_active_msgs); - CFS_INIT_LIST_HEAD (&the_lnet.ln_active_mds); - CFS_INIT_LIST_HEAD (&the_lnet.ln_active_eqs); - CFS_INIT_LIST_HEAD (&the_lnet.ln_test_peers); - CFS_INIT_LIST_HEAD (&the_lnet.ln_nis); - CFS_INIT_LIST_HEAD (&the_lnet.ln_zombie_nis); - CFS_INIT_LIST_HEAD (&the_lnet.ln_remote_nets); - CFS_INIT_LIST_HEAD (&the_lnet.ln_routers); + the_lnet.ln_counters = cfs_percpt_alloc(lnet_cpt_table(), + sizeof(lnet_counters_t)); + if (the_lnet.ln_counters == NULL) { + CERROR("Failed to allocate counters for LNet\n"); + rc = -ENOMEM; + goto failed; + } - the_lnet.ln_interface_cookie = lnet_create_interface_cookie(); + rc = lnet_peer_tables_create(); + if (rc != 0) + goto failed; - lnet_init_rtrpools(); + rc = lnet_msg_containers_create(); + if (rc != 0) + goto failed; - rc = lnet_setup_handle_hash (); - if (rc != 0) - goto failed0; + rc = lnet_res_container_setup(&the_lnet.ln_eq_container, 0, + LNET_COOKIE_TYPE_EQ, LNET_FL_MAX_EQS, + sizeof(lnet_eq_t)); + if (rc != 0) + goto failed; - rc = lnet_create_peer_table(); - if (rc != 0) - goto failed1; + recs = lnet_res_containers_create(LNET_COOKIE_TYPE_ME, LNET_FL_MAX_MES, + sizeof(lnet_me_t)); + if (recs == NULL) + goto failed; - rc = lnet_init_finalizers(); - if (rc != 0) - goto failed2; + the_lnet.ln_me_containers = recs; - the_lnet.ln_nportals = MAX_PORTALS; - LIBCFS_ALLOC(the_lnet.ln_portals, - the_lnet.ln_nportals * - sizeof(*the_lnet.ln_portals)); - if (the_lnet.ln_portals == NULL) { - rc = -ENOMEM; - goto failed3; - } + recs = lnet_res_containers_create(LNET_COOKIE_TYPE_MD, LNET_FL_MAX_MDS, + sizeof(lnet_libmd_t)); + if (recs == NULL) + goto failed; - for (i = 0; i < the_lnet.ln_nportals; i++) { - CFS_INIT_LIST_HEAD(&(the_lnet.ln_portals[i].ptl_ml)); - CFS_INIT_LIST_HEAD(&(the_lnet.ln_portals[i].ptl_msgq)); - the_lnet.ln_portals[i].ptl_options = 0; - } + the_lnet.ln_md_containers = recs; - return 0; + rc = lnet_portals_create(); + if (rc != 0) { + CERROR("Failed to create portals for LNet: %d\n", rc); + goto failed; + } - failed3: - lnet_fini_finalizers(); - failed2: - lnet_destroy_peer_table(); - failed1: - lnet_cleanup_handle_hash(); - failed0: - lnet_descriptor_cleanup(); - return rc; + return 0; + + failed: + lnet_unprepare(); + return rc; } int lnet_unprepare (void) { - int idx; - /* NB no LNET_LOCK since this is the last reference. All LND instances * have shut down already, so it is safe to unlink and free all * descriptors, even those that appear committed to a network op (eg MD * with non-zero pending count) */ - lnet_fail_nid(LNET_NID_ANY, 0); + lnet_fail_nid(LNET_NID_ANY, 0); - LASSERT (list_empty(&the_lnet.ln_test_peers)); - LASSERT (the_lnet.ln_refcount == 0); - LASSERT (list_empty(&the_lnet.ln_nis)); - LASSERT (list_empty(&the_lnet.ln_zombie_nis)); - LASSERT (the_lnet.ln_nzombie_nis == 0); + LASSERT(the_lnet.ln_refcount == 0); + LASSERT(cfs_list_empty(&the_lnet.ln_test_peers)); + LASSERT(cfs_list_empty(&the_lnet.ln_nis)); + LASSERT(cfs_list_empty(&the_lnet.ln_nis_cpt)); + LASSERT(cfs_list_empty(&the_lnet.ln_nis_zombie)); - for (idx = 0; idx < the_lnet.ln_nportals; idx++) { - LASSERT (list_empty(&the_lnet.ln_portals[idx].ptl_msgq)); + lnet_portals_destroy(); - while (!list_empty (&the_lnet.ln_portals[idx].ptl_ml)) { - lnet_me_t *me = list_entry (the_lnet.ln_portals[idx].ptl_ml.next, - lnet_me_t, me_list); + if (the_lnet.ln_md_containers != NULL) { + lnet_res_containers_destroy(the_lnet.ln_md_containers); + the_lnet.ln_md_containers = NULL; + } - CERROR ("Active me %p on exit\n", me); - list_del (&me->me_list); - lnet_me_free (me); - } - } + if (the_lnet.ln_me_containers != NULL) { + lnet_res_containers_destroy(the_lnet.ln_me_containers); + the_lnet.ln_me_containers = NULL; + } - while (!list_empty (&the_lnet.ln_active_mds)) { - lnet_libmd_t *md = list_entry (the_lnet.ln_active_mds.next, - lnet_libmd_t, md_list); + lnet_res_container_cleanup(&the_lnet.ln_eq_container); - CERROR ("Active md %p on exit\n", md); - list_del_init (&md->md_list); - lnet_md_free (md); - } + lnet_msg_containers_destroy(); + lnet_peer_tables_destroy(); + lnet_rtrpools_free(); - while (!list_empty (&the_lnet.ln_active_eqs)) { - lnet_eq_t *eq = list_entry (the_lnet.ln_active_eqs.next, - lnet_eq_t, eq_list); + if (the_lnet.ln_counters != NULL) { + cfs_percpt_free(the_lnet.ln_counters); + the_lnet.ln_counters = NULL; + } - CERROR ("Active eq %p on exit\n", eq); - list_del (&eq->eq_list); - lnet_eq_free (eq); - } + return 0; +} - while (!list_empty (&the_lnet.ln_active_msgs)) { - lnet_msg_t *msg = list_entry (the_lnet.ln_active_msgs.next, - lnet_msg_t, msg_activelist); +lnet_ni_t * +lnet_net2ni_locked(__u32 net, int cpt) +{ + cfs_list_t *tmp; + lnet_ni_t *ni; - CERROR ("Active msg %p on exit\n", msg); - LASSERT (msg->msg_onactivelist); - msg->msg_onactivelist = 0; - list_del (&msg->msg_activelist); - lnet_msg_free (msg); - } + LASSERT(cpt != LNET_LOCK_EX); - LIBCFS_FREE(the_lnet.ln_portals, - the_lnet.ln_nportals * sizeof(*the_lnet.ln_portals)); + cfs_list_for_each(tmp, &the_lnet.ln_nis) { + ni = cfs_list_entry(tmp, lnet_ni_t, ni_list); - lnet_free_rtrpools(); - lnet_fini_finalizers(); - lnet_destroy_peer_table(); - lnet_cleanup_handle_hash(); - lnet_descriptor_cleanup(); + if (LNET_NIDNET(ni->ni_nid) == net) { + lnet_ni_addref_locked(ni, cpt); + return ni; + } + } - return (0); + return NULL; } -lnet_ni_t * -lnet_net2ni_locked (__u32 net) +lnet_ni_t * +lnet_net2ni(__u32 net) { - struct list_head *tmp; - lnet_ni_t *ni; + lnet_ni_t *ni; - list_for_each (tmp, &the_lnet.ln_nis) { - ni = list_entry(tmp, lnet_ni_t, ni_list); + lnet_net_lock(0); + ni = lnet_net2ni_locked(net, 0); + lnet_net_unlock(0); - if (LNET_NIDNET(ni->ni_nid) == net) { - lnet_ni_addref_locked(ni); - return ni; - } - } + return ni; +} +EXPORT_SYMBOL(lnet_net2ni); - return NULL; +static unsigned int +lnet_nid_cpt_hash(lnet_nid_t nid, unsigned int number) +{ + __u64 key = nid; + unsigned int val; + + LASSERT(number >= 1 && number <= LNET_CPT_NUMBER); + + if (number == 1) + return 0; + + val = cfs_hash_long(key, LNET_CPT_BITS); + /* NB: LNET_CP_NUMBER doesn't have to be PO2 */ + if (val < number) + return val; + + return (unsigned int)(key + val + (val >> 1)) % number; } int -lnet_islocalnet (__u32 net) +lnet_cpt_of_nid_locked(lnet_nid_t nid) { - lnet_ni_t *ni; + struct lnet_ni *ni; + + /* must called with hold of lnet_net_lock */ + if (LNET_CPT_NUMBER == 1) + return 0; /* the only one */ - LNET_LOCK(); - ni = lnet_net2ni_locked(net); - if (ni != NULL) - lnet_ni_decref_locked(ni); - LNET_UNLOCK(); + /* take lnet_net_lock(any) would be OK */ + if (!cfs_list_empty(&the_lnet.ln_nis_cpt)) { + cfs_list_for_each_entry(ni, &the_lnet.ln_nis_cpt, ni_cptlist) { + if (LNET_NIDNET(ni->ni_nid) != LNET_NIDNET(nid)) + continue; - return ni != NULL; + LASSERT(ni->ni_cpts != NULL); + return ni->ni_cpts[lnet_nid_cpt_hash + (nid, ni->ni_ncpts)]; + } + } + + return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER); +} + +int +lnet_cpt_of_nid(lnet_nid_t nid) +{ + int cpt; + int cpt2; + + if (LNET_CPT_NUMBER == 1) + return 0; /* the only one */ + + if (cfs_list_empty(&the_lnet.ln_nis_cpt)) + return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER); + + cpt = lnet_net_lock_current(); + cpt2 = lnet_cpt_of_nid_locked(nid); + lnet_net_unlock(cpt); + + return cpt2; +} +EXPORT_SYMBOL(lnet_cpt_of_nid); + +int +lnet_islocalnet(__u32 net) +{ + struct lnet_ni *ni; + int cpt; + + cpt = lnet_net_lock_current(); + + ni = lnet_net2ni_locked(net, cpt); + if (ni != NULL) + lnet_ni_decref_locked(ni, cpt); + + lnet_net_unlock(cpt); + + return ni != NULL; } lnet_ni_t * -lnet_nid2ni_locked (lnet_nid_t nid) +lnet_nid2ni_locked(lnet_nid_t nid, int cpt) { - struct list_head *tmp; - lnet_ni_t *ni; + struct lnet_ni *ni; + cfs_list_t *tmp; - list_for_each (tmp, &the_lnet.ln_nis) { - ni = list_entry(tmp, lnet_ni_t, ni_list); + LASSERT(cpt != LNET_LOCK_EX); - if (ni->ni_nid == nid) { - lnet_ni_addref_locked(ni); - return ni; - } - } + cfs_list_for_each(tmp, &the_lnet.ln_nis) { + ni = cfs_list_entry(tmp, lnet_ni_t, ni_list); - return NULL; + if (ni->ni_nid == nid) { + lnet_ni_addref_locked(ni, cpt); + return ni; + } + } + + return NULL; } int -lnet_islocalnid (lnet_nid_t nid) +lnet_islocalnid(lnet_nid_t nid) { - lnet_ni_t *ni; + struct lnet_ni *ni; + int cpt; - LNET_LOCK(); - ni = lnet_nid2ni_locked(nid); - if (ni != NULL) - lnet_ni_decref_locked(ni); - LNET_UNLOCK(); + cpt = lnet_net_lock_current(); + ni = lnet_nid2ni_locked(nid, cpt); + if (ni != NULL) + lnet_ni_decref_locked(ni, cpt); + lnet_net_unlock(cpt); - return ni != NULL; + return ni != NULL; } int lnet_count_acceptor_nis (void) { - /* Return the # of NIs that need the acceptor. */ - int count = 0; + /* Return the # of NIs that need the acceptor. */ + int count = 0; #if defined(__KERNEL__) || defined(HAVE_LIBPTHREAD) - struct list_head *tmp; - lnet_ni_t *ni; + cfs_list_t *tmp; + struct lnet_ni *ni; + int cpt; - LNET_LOCK(); - list_for_each (tmp, &the_lnet.ln_nis) { - ni = list_entry(tmp, lnet_ni_t, ni_list); + cpt = lnet_net_lock_current(); + cfs_list_for_each(tmp, &the_lnet.ln_nis) { + ni = cfs_list_entry(tmp, lnet_ni_t, ni_list); - if (ni->ni_lnd->lnd_accept != NULL) - count++; - } + if (ni->ni_lnd->lnd_accept != NULL) + count++; + } - LNET_UNLOCK(); + lnet_net_unlock(cpt); #endif /* defined(__KERNEL__) || defined(HAVE_LIBPTHREAD) */ - return count; + return count; +} + +static int +lnet_ni_tq_credits(lnet_ni_t *ni) +{ + int credits; + + LASSERT(ni->ni_ncpts >= 1); + + if (ni->ni_ncpts == 1) + return ni->ni_maxtxcredits; + + credits = ni->ni_maxtxcredits / ni->ni_ncpts; + credits = max(credits, 8 * ni->ni_peertxcredits); + credits = min(credits, ni->ni_maxtxcredits); + + return credits; } void @@ -867,38 +1019,41 @@ lnet_shutdown_lndnis (void) /* NB called holding the global mutex */ /* All quiet on the API front */ - LASSERT (!the_lnet.ln_shutdown); - LASSERT (the_lnet.ln_refcount == 0); - LASSERT (list_empty(&the_lnet.ln_zombie_nis)); - LASSERT (the_lnet.ln_nzombie_nis == 0); - LASSERT (list_empty(&the_lnet.ln_remote_nets)); - - LNET_LOCK(); - the_lnet.ln_shutdown = 1; /* flag shutdown */ - - /* Unlink NIs from the global table */ - while (!list_empty(&the_lnet.ln_nis)) { - ni = list_entry(the_lnet.ln_nis.next, - lnet_ni_t, ni_list); - list_del (&ni->ni_list); - - the_lnet.ln_nzombie_nis++; - lnet_ni_decref_locked(ni); /* drop apini's ref */ - } - - /* Drop the cached eqwait NI. */ - if (the_lnet.ln_eqwaitni != NULL) { - lnet_ni_decref_locked(the_lnet.ln_eqwaitni); - the_lnet.ln_eqwaitni = NULL; - } - - /* Drop the cached loopback NI. */ - if (the_lnet.ln_loni != NULL) { - lnet_ni_decref_locked(the_lnet.ln_loni); - the_lnet.ln_loni = NULL; - } - - LNET_UNLOCK(); + LASSERT(!the_lnet.ln_shutdown); + LASSERT(the_lnet.ln_refcount == 0); + LASSERT(cfs_list_empty(&the_lnet.ln_nis_zombie)); + LASSERT(cfs_list_empty(&the_lnet.ln_remote_nets)); + + lnet_net_lock(LNET_LOCK_EX); + the_lnet.ln_shutdown = 1; /* flag shutdown */ + + /* Unlink NIs from the global table */ + while (!cfs_list_empty(&the_lnet.ln_nis)) { + ni = cfs_list_entry(the_lnet.ln_nis.next, + lnet_ni_t, ni_list); + /* move it to zombie list and nobody can find it anymore */ + cfs_list_move(&ni->ni_list, &the_lnet.ln_nis_zombie); + lnet_ni_decref_locked(ni, 0); /* drop ln_nis' ref */ + + if (!cfs_list_empty(&ni->ni_cptlist)) { + cfs_list_del_init(&ni->ni_cptlist); + lnet_ni_decref_locked(ni, 0); + } + } + + /* Drop the cached eqwait NI. */ + if (the_lnet.ln_eq_waitni != NULL) { + lnet_ni_decref_locked(the_lnet.ln_eq_waitni, 0); + the_lnet.ln_eq_waitni = NULL; + } + + /* Drop the cached loopback NI. */ + if (the_lnet.ln_loni != NULL) { + lnet_ni_decref_locked(the_lnet.ln_loni, 0); + the_lnet.ln_loni = NULL; + } + + lnet_net_unlock(LNET_LOCK_EX); /* Clear lazy portals and drop delayed messages which hold refs * on their lnet_msg_t::msg_rxpeer */ @@ -907,34 +1062,46 @@ lnet_shutdown_lndnis (void) /* Clear the peer table and wait for all peers to go (they hold refs on * their NIs) */ - lnet_clear_peer_table(); - - LNET_LOCK(); - /* Now wait for the NI's I just nuked to show up on apini_zombie_nis - * and shut them down in guaranteed thread context */ - i = 2; - while (the_lnet.ln_nzombie_nis != 0) { - - while (list_empty(&the_lnet.ln_zombie_nis)) { - LNET_UNLOCK(); - ++i; - if ((i & (-i)) == i) - CDEBUG(D_WARNING,"Waiting for %d zombie NIs\n", - the_lnet.ln_nzombie_nis); - cfs_pause(cfs_time_seconds(1)); - LNET_LOCK(); - } - - ni = list_entry(the_lnet.ln_zombie_nis.next, - lnet_ni_t, ni_list); - list_del(&ni->ni_list); - ni->ni_lnd->lnd_refcount--; - - LNET_UNLOCK(); + lnet_peer_tables_cleanup(); + + lnet_net_lock(LNET_LOCK_EX); + /* Now wait for the NI's I just nuked to show up on ln_zombie_nis + * and shut them down in guaranteed thread context */ + i = 2; + while (!cfs_list_empty(&the_lnet.ln_nis_zombie)) { + int *ref; + int j; + + ni = cfs_list_entry(the_lnet.ln_nis_zombie.next, + lnet_ni_t, ni_list); + cfs_list_del_init(&ni->ni_list); + cfs_percpt_for_each(ref, j, ni->ni_refs) { + if (*ref == 0) + continue; + /* still busy, add it back to zombie list */ + cfs_list_add(&ni->ni_list, &the_lnet.ln_nis_zombie); + break; + } + + while (!cfs_list_empty(&ni->ni_list)) { + lnet_net_unlock(LNET_LOCK_EX); + ++i; + if ((i & (-i)) == i) { + CDEBUG(D_WARNING, + "Waiting for zombie LNI %s\n", + libcfs_nid2str(ni->ni_nid)); + } + cfs_pause(cfs_time_seconds(1)); + lnet_net_lock(LNET_LOCK_EX); + continue; + } + + ni->ni_lnd->lnd_refcount--; + lnet_net_unlock(LNET_LOCK_EX); islo = ni->ni_lnd->lnd_type == LOLND; - LASSERT (!in_interrupt ()); + LASSERT (!cfs_in_interrupt ()); (ni->ni_lnd->lnd_shutdown)(ni); /* can't deref lnd anymore now; it might have unregistered @@ -944,28 +1111,28 @@ lnet_shutdown_lndnis (void) CDEBUG(D_LNI, "Removed LNI %s\n", libcfs_nid2str(ni->ni_nid)); - LIBCFS_FREE(ni, sizeof(*ni)); + lnet_ni_free(ni); + lnet_net_lock(LNET_LOCK_EX); + } - LNET_LOCK(); - the_lnet.ln_nzombie_nis--; - } - - the_lnet.ln_shutdown = 0; - LNET_UNLOCK(); + the_lnet.ln_shutdown = 0; + lnet_net_unlock(LNET_LOCK_EX); - if (the_lnet.ln_network_tokens != NULL) { - LIBCFS_FREE(the_lnet.ln_network_tokens, - the_lnet.ln_network_tokens_nob); - the_lnet.ln_network_tokens = NULL; - } + if (the_lnet.ln_network_tokens != NULL) { + LIBCFS_FREE(the_lnet.ln_network_tokens, + the_lnet.ln_network_tokens_nob); + the_lnet.ln_network_tokens = NULL; + } } int lnet_startup_lndnis (void) { - lnd_t *lnd; - lnet_ni_t *ni; - struct list_head nilist; + lnd_t *lnd; + struct lnet_ni *ni; + struct lnet_tx_queue *tq; + cfs_list_t nilist; + int i; int rc = 0; int lnd_type; int nicount = 0; @@ -980,71 +1147,89 @@ lnet_startup_lndnis (void) if (rc != 0) goto failed; - while (!list_empty(&nilist)) { - ni = list_entry(nilist.next, lnet_ni_t, ni_list); + while (!cfs_list_empty(&nilist)) { + ni = cfs_list_entry(nilist.next, lnet_ni_t, ni_list); lnd_type = LNET_NETTYP(LNET_NIDNET(ni->ni_nid)); LASSERT (libcfs_isknown_lnd(lnd_type)); - LNET_MUTEX_DOWN(&the_lnet.ln_lnd_mutex); + if (lnd_type == CIBLND || + lnd_type == OPENIBLND || + lnd_type == IIBLND || + lnd_type == VIBLND) { + CERROR("LND %s obsoleted\n", + libcfs_lnd2str(lnd_type)); + goto failed; + } + + LNET_MUTEX_LOCK(&the_lnet.ln_lnd_mutex); lnd = lnet_find_lnd_by_type(lnd_type); #ifdef __KERNEL__ if (lnd == NULL) { - LNET_MUTEX_UP(&the_lnet.ln_lnd_mutex); - rc = request_module(libcfs_lnd2modname(lnd_type)); - LNET_MUTEX_DOWN(&the_lnet.ln_lnd_mutex); + LNET_MUTEX_UNLOCK(&the_lnet.ln_lnd_mutex); + rc = cfs_request_module("%s", + libcfs_lnd2modname(lnd_type)); + LNET_MUTEX_LOCK(&the_lnet.ln_lnd_mutex); lnd = lnet_find_lnd_by_type(lnd_type); if (lnd == NULL) { - LNET_MUTEX_UP(&the_lnet.ln_lnd_mutex); + LNET_MUTEX_UNLOCK(&the_lnet.ln_lnd_mutex); CERROR("Can't load LND %s, module %s, rc=%d\n", libcfs_lnd2str(lnd_type), libcfs_lnd2modname(lnd_type), rc); -#ifndef CONFIG_KMOD +#ifndef HAVE_MODULE_LOADING_SUPPORT LCONSOLE_ERROR_MSG(0x104, "Your kernel must be " - "compiled with CONFIG_KMOD set for " - "automatic module loading."); + "compiled with kernel module " + "loading support."); #endif goto failed; } } #else if (lnd == NULL) { - LNET_MUTEX_UP(&the_lnet.ln_lnd_mutex); + LNET_MUTEX_UNLOCK(&the_lnet.ln_lnd_mutex); CERROR("LND %s not supported\n", libcfs_lnd2str(lnd_type)); goto failed; } #endif - ni->ni_refcount = 1; - - LNET_LOCK(); - lnd->lnd_refcount++; - LNET_UNLOCK(); + lnet_net_lock(LNET_LOCK_EX); + lnd->lnd_refcount++; + lnet_net_unlock(LNET_LOCK_EX); ni->ni_lnd = lnd; rc = (lnd->lnd_startup)(ni); - LNET_MUTEX_UP(&the_lnet.ln_lnd_mutex); + LNET_MUTEX_UNLOCK(&the_lnet.ln_lnd_mutex); if (rc != 0) { LCONSOLE_ERROR_MSG(0x105, "Error %d starting up LNI %s" "\n", rc, libcfs_lnd2str(lnd->lnd_type)); - LNET_LOCK(); - lnd->lnd_refcount--; - LNET_UNLOCK(); + lnet_net_lock(LNET_LOCK_EX); + lnd->lnd_refcount--; + lnet_net_unlock(LNET_LOCK_EX); goto failed; } - list_del(&ni->ni_list); + LASSERT (ni->ni_peertimeout <= 0 || lnd->lnd_query != NULL); - LNET_LOCK(); - list_add_tail(&ni->ni_list, &the_lnet.ln_nis); - LNET_UNLOCK(); + cfs_list_del(&ni->ni_list); + + lnet_net_lock(LNET_LOCK_EX); + /* refcount for ln_nis */ + lnet_ni_addref_locked(ni, 0); + cfs_list_add_tail(&ni->ni_list, &the_lnet.ln_nis); + if (ni->ni_cpts != NULL) { + cfs_list_add_tail(&ni->ni_cptlist, + &the_lnet.ln_nis_cpt); + lnet_ni_addref_locked(ni, 0); + } + + lnet_net_unlock(LNET_LOCK_EX); if (lnd->lnd_type == LOLND) { lnet_ni_addref(ni); @@ -1055,10 +1240,10 @@ lnet_startup_lndnis (void) #ifndef __KERNEL__ if (lnd->lnd_wait != NULL) { - if (the_lnet.ln_eqwaitni == NULL) { - lnet_ni_addref(ni); - the_lnet.ln_eqwaitni = ni; - } + if (the_lnet.ln_eq_waitni == NULL) { + lnet_ni_addref(ni); + the_lnet.ln_eq_waitni = ni; + } } else { # ifndef HAVE_LIBPTHREAD LCONSOLE_ERROR_MSG(0x106, "LND %s not supported in a " @@ -1077,17 +1262,22 @@ lnet_startup_lndnis (void) goto failed; } - ni->ni_txcredits = ni->ni_mintxcredits = ni->ni_maxtxcredits; + cfs_percpt_for_each(tq, i, ni->ni_tx_queues) { + tq->tq_credits_min = + tq->tq_credits_max = + tq->tq_credits = lnet_ni_tq_credits(ni); + } - CDEBUG(D_LNI, "Added LNI %s [%d/%d]\n", - libcfs_nid2str(ni->ni_nid), - ni->ni_peertxcredits, ni->ni_txcredits); + CDEBUG(D_LNI, "Added LNI %s [%d/%d/%d/%d]\n", + libcfs_nid2str(ni->ni_nid), ni->ni_peertxcredits, + lnet_ni_tq_credits(ni) * LNET_CPT_NUMBER, + ni->ni_peerrtrcredits, ni->ni_peertimeout); - nicount++; - } + nicount++; + } - if (the_lnet.ln_eqwaitni != NULL && nicount > 1) { - lnd_type = the_lnet.ln_eqwaitni->ni_lnd->lnd_type; + if (the_lnet.ln_eq_waitni != NULL && nicount > 1) { + lnd_type = the_lnet.ln_eq_waitni->ni_lnd->lnd_type; LCONSOLE_ERROR_MSG(0x109, "LND %s can only run single-network" "\n", libcfs_lnd2str(lnd_type)); @@ -1099,27 +1289,63 @@ lnet_startup_lndnis (void) failed: lnet_shutdown_lndnis(); - while (!list_empty(&nilist)) { - ni = list_entry(nilist.next, lnet_ni_t, ni_list); - list_del(&ni->ni_list); - LIBCFS_FREE(ni, sizeof(*ni)); - } + while (!cfs_list_empty(&nilist)) { + ni = cfs_list_entry(nilist.next, lnet_ni_t, ni_list); + cfs_list_del(&ni->ni_list); + lnet_ni_free(ni); + } - return -ENETDOWN; + return -ENETDOWN; } +/** + * Initialize LNet library. + * + * Only userspace program needs to call this function - it's automatically + * called in the kernel at module loading time. Caller has to call LNetFini() + * after a call to LNetInit(), if and only if the latter returned 0. It must + * be called exactly once. + * + * \return 0 on success, and -ve on failures. + */ int LNetInit(void) { - lnet_assert_wire_constants (); - LASSERT (!the_lnet.ln_init); - - memset(&the_lnet, 0, sizeof(the_lnet)); - - lnet_init_locks(); - CFS_INIT_LIST_HEAD(&the_lnet.ln_lnds); - the_lnet.ln_refcount = 0; - the_lnet.ln_init = 1; + int rc; + + lnet_assert_wire_constants(); + LASSERT(!the_lnet.ln_init); + + memset(&the_lnet, 0, sizeof(the_lnet)); + + /* refer to global cfs_cpt_table for now */ + the_lnet.ln_cpt_table = cfs_cpt_table; + the_lnet.ln_cpt_number = cfs_cpt_number(cfs_cpt_table); + + LASSERT(the_lnet.ln_cpt_number > 0); + if (the_lnet.ln_cpt_number > LNET_CPT_MAX) { + /* we are under risk of consuming all lh_cookie */ + CERROR("Can't have %d CPTs for LNet (max allowed is %d), " + "please change setting of CPT-table and retry\n", + the_lnet.ln_cpt_number, LNET_CPT_MAX); + return -1; + } + + while ((1 << the_lnet.ln_cpt_bits) < the_lnet.ln_cpt_number) + the_lnet.ln_cpt_bits++; + + rc = lnet_create_locks(); + if (rc != 0) { + CERROR("Can't create LNet global locks: %d\n", rc); + return -1; + } + + the_lnet.ln_refcount = 0; + the_lnet.ln_init = 1; + LNetInvalidateHandle(&the_lnet.ln_rc_eqh); + CFS_INIT_LIST_HEAD(&the_lnet.ln_lnds); + CFS_INIT_LIST_HEAD(&the_lnet.ln_rcd_zombie); + CFS_INIT_LIST_HEAD(&the_lnet.ln_rcd_deathrow); #ifdef __KERNEL__ /* All LNDs apart from the LOLND are in separate modules. They @@ -1128,9 +1354,6 @@ LNetInit(void) #else /* Register LNDs * NB the order here determines default 'networks=' order */ -# ifdef CRAY_XT3 - LNET_REGISTER_ULND(the_ptllnd); -# endif # ifdef HAVE_LIBPTHREAD LNET_REGISTER_ULND(the_tcplnd); # endif @@ -1138,28 +1361,55 @@ LNetInit(void) lnet_register_lnd(&the_lolnd); return 0; } +EXPORT_SYMBOL(LNetInit); +/** + * Finalize LNet library. + * + * Only userspace program needs to call this function. It can be called + * at most once. + * + * \pre LNetInit() called with success. + * \pre All LNet users called LNetNIFini() for matching LNetNIInit() calls. + */ void LNetFini(void) { - LASSERT (the_lnet.ln_init); - LASSERT (the_lnet.ln_refcount == 0); + LASSERT(the_lnet.ln_init); + LASSERT(the_lnet.ln_refcount == 0); - while (!list_empty(&the_lnet.ln_lnds)) - lnet_unregister_lnd(list_entry(the_lnet.ln_lnds.next, - lnd_t, lnd_list)); - lnet_fini_locks(); + while (!cfs_list_empty(&the_lnet.ln_lnds)) + lnet_unregister_lnd(cfs_list_entry(the_lnet.ln_lnds.next, + lnd_t, lnd_list)); + lnet_destroy_locks(); - the_lnet.ln_init = 0; + the_lnet.ln_init = 0; } +EXPORT_SYMBOL(LNetFini); +/** + * Set LNet PID and start LNet interfaces, routing, and forwarding. + * + * Userspace program should call this after a successful call to LNetInit(). + * Users must call this function at least once before any other functions. + * For each successful call there must be a corresponding call to + * LNetNIFini(). For subsequent calls to LNetNIInit(), \a requested_pid is + * ignored. + * + * The PID used by LNet may be different from the one requested. + * See LNetGetId(). + * + * \param requested_pid PID requested by the caller. + * + * \return >= 0 on success, and < 0 error code on failures. + */ int LNetNIInit(lnet_pid_t requested_pid) { int im_a_router = 0; int rc; - LNET_MUTEX_DOWN(&the_lnet.ln_api_mutex); + LNET_MUTEX_LOCK(&the_lnet.ln_api_mutex); LASSERT (the_lnet.ln_init); CDEBUG(D_OTHER, "refs %d\n", the_lnet.ln_refcount); @@ -1169,6 +1419,8 @@ LNetNIInit(lnet_pid_t requested_pid) goto out; } + lnet_get_tunables(); + if (requested_pid == LNET_PID_ANY) { /* Don't instantiate LNET just for me */ rc = -ENETDOWN; @@ -1191,7 +1443,7 @@ LNetNIInit(lnet_pid_t requested_pid) if (rc != 0) goto failed2; - rc = lnet_alloc_rtrpools(im_a_router); + rc = lnet_rtrpools_alloc(im_a_router); if (rc != 0) goto failed2; @@ -1202,11 +1454,13 @@ LNetNIInit(lnet_pid_t requested_pid) the_lnet.ln_refcount = 1; /* Now I may use my own API functions... */ - rc = lnet_router_checker_start(); + /* NB router checker needs the_lnet.ln_ping_info in + * lnet_router_checker -> lnet_update_ni_status_locked */ + rc = lnet_ping_target_init(); if (rc != 0) goto failed3; - rc = lnet_ping_target_init(); + rc = lnet_router_checker_start(); if (rc != 0) goto failed4; @@ -1214,7 +1468,7 @@ LNetNIInit(lnet_pid_t requested_pid) goto out; failed4: - lnet_router_checker_stop(); + lnet_ping_target_fini(); failed3: the_lnet.ln_refcount = 0; lnet_acceptor_stop(); @@ -1226,14 +1480,24 @@ LNetNIInit(lnet_pid_t requested_pid) failed0: LASSERT (rc < 0); out: - LNET_MUTEX_UP(&the_lnet.ln_api_mutex); + LNET_MUTEX_UNLOCK(&the_lnet.ln_api_mutex); return rc; } +EXPORT_SYMBOL(LNetNIInit); +/** + * Stop LNet interfaces, routing, and forwarding. + * + * Users must call this function once for each successful call to LNetNIInit(). + * Once the LNetNIFini() operation has been started, the results of pending + * API operations are undefined. + * + * \return always 0 for current implementation. + */ int LNetNIFini() { - LNET_MUTEX_DOWN(&the_lnet.ln_api_mutex); + LNET_MUTEX_LOCK(&the_lnet.ln_api_mutex); LASSERT (the_lnet.ln_init); LASSERT (the_lnet.ln_refcount > 0); @@ -1244,8 +1508,8 @@ LNetNIFini() LASSERT (!the_lnet.ln_niinit_self); lnet_proc_fini(); - lnet_ping_target_fini(); lnet_router_checker_stop(); + lnet_ping_target_fini(); /* Teardown fns that use my own API functions BEFORE here */ the_lnet.ln_refcount = 0; @@ -1256,10 +1520,24 @@ LNetNIFini() lnet_unprepare(); } - LNET_MUTEX_UP(&the_lnet.ln_api_mutex); + LNET_MUTEX_UNLOCK(&the_lnet.ln_api_mutex); return 0; } +EXPORT_SYMBOL(LNetNIFini); +/** + * This is an ugly hack to export IOC_LIBCFS_DEBUG_PEER and + * IOC_LIBCFS_PORTALS_COMPATIBILITY commands to users, by tweaking the LNet + * internal ioctl handler. + * + * IOC_LIBCFS_PORTALS_COMPATIBILITY is now deprecated, don't use it. + * + * \param cmd IOC_LIBCFS_DEBUG_PEER to print debugging data about a peer. + * The data will be printed to system console. Don't use it excessively. + * \param arg A pointer to lnet_process_id_t, process ID of the peer. + * + * \return Always return 0 when called by users directly (i.e., not via ioctl). + */ int LNetCtl(unsigned int cmd, void *arg) { @@ -1294,7 +1572,9 @@ LNetCtl(unsigned int cmd, void *arg) &data->ioc_nid, &data->ioc_flags); case IOC_LIBCFS_NOTIFY_ROUTER: return lnet_notify(NULL, data->ioc_nid, data->ioc_flags, - (time_t)data->ioc_u64[0]); + cfs_time_current() - + cfs_time_seconds(cfs_time_current_sec() - + (time_t)data->ioc_u64[0])); case IOC_LIBCFS_PORTALS_COMPATIBILITY: /* This can be removed once lustre stops calling it */ @@ -1309,9 +1589,9 @@ LNetCtl(unsigned int cmd, void *arg) return 0; case IOC_LIBCFS_TESTPROTOCOMPAT: - LNET_LOCK(); - the_lnet.ln_testprotocompat = data->ioc_flags; - LNET_UNLOCK(); + lnet_net_lock(LNET_LOCK_EX); + the_lnet.ln_testprotocompat = data->ioc_flags; + lnet_net_unlock(LNET_LOCK_EX); return 0; case IOC_LIBCFS_PING: @@ -1363,24 +1643,37 @@ LNetCtl(unsigned int cmd, void *arg) } /* not reached */ } +EXPORT_SYMBOL(LNetCtl); +/** + * Retrieve the lnet_process_id_t ID of LNet interface at \a index. Note that + * all interfaces share a same PID, as requested by LNetNIInit(). + * + * \param index Index of the interface to look up. + * \param id On successful return, this location will hold the + * lnet_process_id_t ID of the interface. + * + * \retval 0 If an interface exists at \a index. + * \retval -ENOENT If no interface has been found. + */ int LNetGetId(unsigned int index, lnet_process_id_t *id) { - lnet_ni_t *ni; - struct list_head *tmp; - int rc = -ENOENT; + struct lnet_ni *ni; + cfs_list_t *tmp; + int cpt; + int rc = -ENOENT; - LASSERT (the_lnet.ln_init); - LASSERT (the_lnet.ln_refcount > 0); + LASSERT(the_lnet.ln_init); + LASSERT(the_lnet.ln_refcount > 0); - LNET_LOCK(); + cpt = lnet_net_lock_current(); - list_for_each(tmp, &the_lnet.ln_nis) { + cfs_list_for_each(tmp, &the_lnet.ln_nis) { if (index-- != 0) continue; - ni = list_entry(tmp, lnet_ni_t, ni_list); + ni = cfs_list_entry(tmp, lnet_ni_t, ni_list); id->nid = ni->ni_nid; id->pid = the_lnet.ln_pid; @@ -1388,29 +1681,32 @@ LNetGetId(unsigned int index, lnet_process_id_t *id) break; } - LNET_UNLOCK(); - - return rc; + lnet_net_unlock(cpt); + return rc; } +EXPORT_SYMBOL(LNetGetId); +/** + * Print a string representation of handle \a h into buffer \a str of + * \a len bytes. + */ void LNetSnprintHandle(char *str, int len, lnet_handle_any_t h) { snprintf(str, len, LPX64, h.cookie); } +EXPORT_SYMBOL(LNetSnprintHandle); - -int -lnet_ping_target_init(void) +static int +lnet_create_ping_info(void) { - lnet_handle_me_t meh; - lnet_process_id_t id; - lnet_md_t md = {0}; - int rc; - int rc2; + int i; int n; + int rc; unsigned int infosz; - int i; + lnet_ni_t *ni; + lnet_process_id_t id; + lnet_ping_info_t *pinfo; for (n = 0; ; n++) { rc = LNetGetId(n, &id); @@ -1420,23 +1716,80 @@ lnet_ping_target_init(void) LASSERT (rc == 0); } - infosz = offsetof(lnet_ping_info_t, pi_nid[n]); - LIBCFS_ALLOC(the_lnet.ln_ping_info, infosz); - if (the_lnet.ln_ping_info == NULL) { + infosz = offsetof(lnet_ping_info_t, pi_ni[n]); + LIBCFS_ALLOC(pinfo, infosz); + if (pinfo == NULL) { CERROR("Can't allocate ping info[%d]\n", n); return -ENOMEM; } - the_lnet.ln_ping_info->pi_magic = LNET_PROTO_PING_MAGIC; - the_lnet.ln_ping_info->pi_version = LNET_PROTO_PING_VERSION; - the_lnet.ln_ping_info->pi_pid = the_lnet.ln_pid; - the_lnet.ln_ping_info->pi_nnids = n; + pinfo->pi_nnis = n; + pinfo->pi_pid = the_lnet.ln_pid; + pinfo->pi_magic = LNET_PROTO_PING_MAGIC; + pinfo->pi_features = LNET_PING_FEAT_NI_STATUS; for (i = 0; i < n; i++) { + lnet_ni_status_t *ns = &pinfo->pi_ni[i]; + rc = LNetGetId(i, &id); LASSERT (rc == 0); - the_lnet.ln_ping_info->pi_nid[i] = id.nid; - } + + ns->ns_nid = id.nid; + ns->ns_status = LNET_NI_STATUS_UP; + + lnet_net_lock(0); + + ni = lnet_nid2ni_locked(id.nid, 0); + LASSERT(ni != NULL); + + lnet_ni_lock(ni); + LASSERT(ni->ni_status == NULL); + ni->ni_status = ns; + lnet_ni_unlock(ni); + + lnet_ni_decref_locked(ni, 0); + lnet_net_unlock(0); + } + + the_lnet.ln_ping_info = pinfo; + return 0; +} + +static void +lnet_destroy_ping_info(void) +{ + struct lnet_ni *ni; + + lnet_net_lock(0); + + cfs_list_for_each_entry(ni, &the_lnet.ln_nis, ni_list) { + lnet_ni_lock(ni); + ni->ni_status = NULL; + lnet_ni_unlock(ni); + } + + lnet_net_unlock(0); + + LIBCFS_FREE(the_lnet.ln_ping_info, + offsetof(lnet_ping_info_t, + pi_ni[the_lnet.ln_ping_info->pi_nnis])); + the_lnet.ln_ping_info = NULL; + return; +} + +int +lnet_ping_target_init(void) +{ + lnet_md_t md = {0}; + lnet_handle_me_t meh; + lnet_process_id_t id; + int rc; + int rc2; + int infosz; + + rc = lnet_create_ping_info(); + if (rc != 0) + return rc; /* We can have a tiny EQ since we only need to see the unlink event on * teardown, which by definition is the last one! */ @@ -1460,6 +1813,8 @@ lnet_ping_target_init(void) } /* initialize md content */ + infosz = offsetof(lnet_ping_info_t, + pi_ni[the_lnet.ln_ping_info->pi_nnis]); md.start = the_lnet.ln_ping_info; md.length = infosz; md.threshold = LNET_MD_THRESH_INF; @@ -1486,8 +1841,7 @@ lnet_ping_target_init(void) rc2 = LNetEQFree(the_lnet.ln_ping_target_eq); LASSERT (rc2 == 0); failed_0: - LIBCFS_FREE(the_lnet.ln_ping_info, infosz); - + lnet_destroy_ping_info(); return rc; } @@ -1524,11 +1878,7 @@ lnet_ping_target_fini(void) rc = LNetEQFree(the_lnet.ln_ping_target_eq); LASSERT (rc == 0); - - LIBCFS_FREE(the_lnet.ln_ping_info, - offsetof(lnet_ping_info_t, - pi_nid[the_lnet.ln_ping_info->pi_nnids])); - + lnet_destroy_ping_info(); cfs_restore_sigs(blocked); } @@ -1543,7 +1893,7 @@ lnet_ping (lnet_process_id_t id, int timeout_ms, lnet_process_id_t *ids, int n_i int unlinked = 0; int replied = 0; const int a_long_time = 60000; /* mS */ - int infosz = offsetof(lnet_ping_info_t, pi_nid[n_ids]); + int infosz = offsetof(lnet_ping_info_t, pi_ni[n_ids]); lnet_ping_info_t *info; lnet_process_id_t tmpid; int i; @@ -1637,7 +1987,6 @@ lnet_ping (lnet_process_id_t id, int timeout_ms, lnet_process_id_t *ids, int n_i CWARN("ping %s: late network completion\n", libcfs_id2str(id)); } - } else if (event.type == LNET_EVENT_REPLY) { replied = 1; rc = event.mlength; @@ -1666,38 +2015,31 @@ lnet_ping (lnet_process_id_t id, int timeout_ms, lnet_process_id_t *ids, int n_i } if (info->pi_magic == __swab32(LNET_PROTO_PING_MAGIC)) { - /* NB I might be swabbing garbage until I check below, but it - * doesn't matter */ - __swab32s(&info->pi_version); - __swab32s(&info->pi_pid); - __swab32s(&info->pi_nnids); - for (i = 0; i < (int)info->pi_nnids && i < (int)n_ids; i++) - __swab64s(&info->pi_nid[i]); - + lnet_swap_pinginfo(info); } else if (info->pi_magic != LNET_PROTO_PING_MAGIC) { CERROR("%s: Unexpected magic %08x\n", libcfs_id2str(id), info->pi_magic); goto out_1; } - if (info->pi_version != LNET_PROTO_PING_VERSION) { - CERROR("%s: Unexpected version 0x%x\n", - libcfs_id2str(id), info->pi_version); - goto out_1; - } + if ((info->pi_features & LNET_PING_FEAT_NI_STATUS) == 0) { + CERROR("%s: ping w/o NI status: 0x%x\n", + libcfs_id2str(id), info->pi_features); + goto out_1; + } - if (nob < (int)offsetof(lnet_ping_info_t, pi_nid[0])) { + if (nob < offsetof(lnet_ping_info_t, pi_ni[0])) { CERROR("%s: Short reply %d(%d min)\n", libcfs_id2str(id), - nob, (int)offsetof(lnet_ping_info_t, pi_nid[0])); + nob, (int)offsetof(lnet_ping_info_t, pi_ni[0])); goto out_1; } - if ((int) info->pi_nnids < n_ids) - n_ids = info->pi_nnids; + if (info->pi_nnis < n_ids) + n_ids = info->pi_nnis; - if (nob < (int)offsetof(lnet_ping_info_t, pi_nid[n_ids])) { + if (nob < offsetof(lnet_ping_info_t, pi_ni[n_ids])) { CERROR("%s: Short reply %d(%d expected)\n", libcfs_id2str(id), - nob, (int)offsetof(lnet_ping_info_t, pi_nid[n_ids])); + nob, (int)offsetof(lnet_ping_info_t, pi_ni[n_ids])); goto out_1; } @@ -1705,15 +2047,15 @@ lnet_ping (lnet_process_id_t id, int timeout_ms, lnet_process_id_t *ids, int n_i for (i = 0; i < n_ids; i++) { tmpid.pid = info->pi_pid; - tmpid.nid = info->pi_nid[i]; + tmpid.nid = info->pi_ni[i].ns_nid; #ifdef __KERNEL__ - if (copy_to_user(&ids[i], &tmpid, sizeof(tmpid))) + if (cfs_copy_to_user(&ids[i], &tmpid, sizeof(tmpid))) goto out_1; #else ids[i] = tmpid; #endif } - rc = info->pi_nnids; + rc = info->pi_nnis; out_1: rc2 = LNetEQFree(eqh);