1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 /* 22 * Copyright 2009 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26 /* 27 * FMD Dynamic Reconfiguration (DR) Event Handling 28 * 29 * Fault manager scheme plug-ins must track characteristics of individual 30 * pieces of hardware. As these components can be added or removed by a DR 31 * operation, we need to provide a means by which plug-ins can determine when 32 * they need to re-examine the current configuration. We provide a simple 33 * mechanism whereby this task can be implemented using lazy evaluation: a 34 * simple 64-bit generation counter is maintained and incremented on *any* DR. 35 * Schemes can store the generation number in scheme-specific data structures, 36 * and then revalidate their contents if the current generation number has 37 * changed since the resource information was cached. This method saves time, 38 * avoids the complexity of direct participation in DR, avoids the need for 39 * resource-specific processing of DR events, and is relatively easy to port 40 * to other systems that support dynamic reconfiguration. 41 * 42 * The dr generation is only incremented in response to hardware changes. Since 43 * ASRUs can be in any scheme, including the device scheme, we must also be 44 * aware of software configuration changes which may affect the resource cache. 45 * In addition, we take a snapshot of the topology whenever a reconfiguration 46 * event occurs and notify any modules of the change. 47 */ 48 49 #include <sys/types.h> 50 #include <sys/sunddi.h> 51 #include <sys/sysevent/dr.h> 52 #include <sys/sysevent/eventdefs.h> 53 54 #include <stdio.h> 55 #include <string.h> 56 #include <unistd.h> 57 #include <libsysevent.h> 58 59 #undef MUTEX_HELD 60 #undef RW_READ_HELD 61 #undef RW_WRITE_HELD 62 63 #include <fmd_asru.h> 64 #include <fmd_error.h> 65 #include <fmd_event.h> 66 #include <fmd_fmri.h> 67 #include <fmd_module.h> 68 #include <fmd_subr.h> 69 #include <fmd_topo.h> 70 #include <fmd.h> 71 72 void 73 fmd_dr_event(sysevent_t *sep) 74 { 75 uint64_t gen; 76 fmd_event_t *e; 77 const char *class = sysevent_get_class_name(sep); 78 const char *subclass = sysevent_get_subclass_name(sep); 79 hrtime_t evtime; 80 fmd_topo_t *ftp, *prev; 81 boolean_t update_topo = B_FALSE; 82 83 if (strcmp(class, EC_DR) == 0) { 84 if (strcmp(subclass, ESC_DR_AP_STATE_CHANGE) != 0 && 85 strcmp(subclass, ESC_DR_TARGET_STATE_CHANGE) != 0) 86 return; 87 88 /* 89 * The DR generation is only changed in response to DR events. 90 */ 91 update_topo = B_TRUE; 92 93 (void) pthread_mutex_lock(&fmd.d_stats_lock); 94 gen = fmd.d_stats->ds_dr_gen.fmds_value.ui64++; 95 (void) pthread_mutex_unlock(&fmd.d_stats_lock); 96 97 TRACE((FMD_DBG_XPRT, "dr event %p, gen=%llu", 98 (void *)sep, gen)); 99 } else if (strcmp(class, EC_DEVFS) == 0) { 100 /* 101 * A devfs configuration event can change the topology, 102 * as disk nodes only exist when the device is configured. 103 */ 104 update_topo = B_TRUE; 105 } else if (strcmp(class, EC_PLATFORM) == 0) { 106 if (strcmp(subclass, ESC_PLATFORM_SP_RESET) == 0) { 107 /* 108 * Since we rely on the SP to enumerate fans, 109 * power-supplies and sensors/leds, it would be prudent 110 * to take a new snapshot if the SP resets. 111 */ 112 update_topo = B_TRUE; 113 } 114 } else if (strcmp(class, EC_ZFS) == 0) { 115 /* 116 * These events can change the resource cache. 117 */ 118 if (strcmp(subclass, ESC_ZFS_VDEV_CLEAR) != 0 && 119 strcmp(subclass, ESC_ZFS_VDEV_REMOVE) != 0 && 120 strcmp(subclass, ESC_ZFS_POOL_DESTROY) != 0) 121 return; 122 } else if (strcmp(class, EC_DEV_ADD) == 0 || 123 strcmp(class, EC_DEV_REMOVE) == 0) { 124 if (strcmp(subclass, ESC_DISK) != 0) 125 return; 126 127 update_topo = B_TRUE; 128 } 129 130 /* 131 * Take a topo snapshot and notify modules of the change. Picking an 132 * accurate time here is difficult. On one hand, we have the timestamp 133 * of the underlying sysevent, indicating when the reconfiguration event 134 * occurred. On the other hand, we are taking the topo snapshot 135 * asynchronously, and hence the timestamp of the snapshot is the 136 * current time. Pretending this topo snapshot was valid at the time 137 * the sysevent was posted seems wrong, so we instead opt for the 138 * current time as an upper bound on the snapshot validity. 139 * 140 * Along these lines, we keep track of the last time we dispatched a 141 * topo snapshot. If the sysevent occurred before the last topo 142 * snapshot, then don't bother dispatching another topo change event. 143 * We've already indicated (to the best of our ability) the change in 144 * topology. This prevents endless topo snapshots in response to a 145 * flurry of sysevents. 146 */ 147 sysevent_get_time(sep, &evtime); 148 prev = fmd_topo_hold(); 149 if (evtime <= prev->ft_time_begin && 150 fmd.d_clockops == &fmd_timeops_native) { 151 fmd_topo_rele(prev); 152 return; 153 } 154 fmd_topo_rele(prev); 155 156 if (update_topo) 157 fmd_topo_update(); 158 159 ftp = fmd_topo_hold(); 160 e = fmd_event_create(FMD_EVT_TOPO, ftp->ft_time_end, NULL, ftp); 161 fmd_modhash_dispatch(fmd.d_mod_hash, e); 162 } 163