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 /* 23 * Copyright 2006 Sun Microsystems, Inc. All rights reserved. 24 * Use is subject to license terms. 25 */ 26 27 #ifndef _AO_H 28 #define _AO_H 29 30 #pragma ident "%Z%%M% %I% %E% SMI" 31 32 #include <sys/types.h> 33 #include <sys/mc.h> 34 #include <sys/mca_amd.h> 35 #include <sys/cpu_module_impl.h> 36 #include <sys/nvpair.h> 37 #include <sys/cyclic.h> 38 #include <sys/errorq.h> 39 #include <sys/kobj.h> 40 #include <sys/fm/util.h> 41 42 #ifdef __cplusplus 43 extern "C" { 44 #endif 45 46 #define AO_MCA_MAX_ERRORS 10 47 48 typedef struct ao_data ao_data_t; 49 50 typedef struct ao_bank_regs { 51 uint32_t abr_status; 52 uint32_t abr_addr; 53 } ao_bank_regs_t; 54 55 extern ao_bank_regs_t ao_bank_regs[AMD_MCA_BANK_COUNT]; 56 57 /* 58 * Rather than using torturous conditionals, we match errors using a table of 59 * ao_error_disp_t's. The members in the ao_error_disp_t are matched against 60 * the value of MCi_STATUS, with a successful match indicating that the given 61 * error occurred. 62 * 63 * While aed_stat_code will match most of the status code bits, a few of the 64 * status code fields are either/or, and are treated separately so as to 65 * minimize the number of ao_error_disp_t structures that must be created. 66 * For example, the dc.tag_par error can have r4 values drd or dwr. Rather 67 * than creating two ao_error_disp_t's, we use the separate aed_stat_r4_bits 68 * field to indicate both AO_MCA_R4_BIT_DRD and AO_MCA_R4_BIT_DWD. As the 69 * matching r4 values are drawn from aed_stat_r4_bits, we don't use the r4 70 * bits in aed_stat_code for matching. Similar reasoning lies behind the 71 * creation of the pp and ii fields. 72 */ 73 #define AO_AED_PANIC_NEVER 0x00 74 #define AO_AED_PANIC_IFMCE 0x01 75 #define AO_AED_PANIC_ALWAYS 0x80 76 77 #define AO_AED_F_CORRECTABLE 0x01 78 #define AO_AED_F_LOFAULT_OK 0x02 79 80 typedef struct ao_error_disp { 81 const char *aed_class; /* ereport class for use if match */ 82 uint64_t aed_ereport_members; /* ereport contents flags if match */ 83 uint64_t aed_stat_mask; /* status msr bits for match */ 84 uint64_t aed_stat_mask_res; /* status mask result for match */ 85 uint16_t aed_stat_code; /* status code for match */ 86 uint8_t aed_stat_extcode; /* extended status code for match */ 87 uint8_t aed_stat_pp_bits:4; /* AO_MCA_PP_BIT_* for pp matching */ 88 uint8_t aed_stat_ii_bits:4; /* AO_MCA_II_BIT_* for ii matching */ 89 uint16_t aed_stat_r4_bits; /* AO_MCA_R4_BIT_* for r4 matching */ 90 uint8_t aed_panic_when; /* extra conditions for panic */ 91 uint8_t aed_flags; /* AO_AED_F_* */ 92 } ao_error_disp_t; 93 94 /* 95 * The poller has two parts. First is the omni cyclic, which runs on all 96 * CPUs, and which polls the error MSRs at some fixed (long) interval. This 97 * cyclic will run on all machines, all the time, and thus must have minimal 98 * runtime impact. The second portion of the poller is manually-initiated, and 99 * is used by the error injector/synthesizer to request an immediate poll of the 100 * error state registers. 101 * 102 * With this number of moving parts, it is essential that we have some sort of 103 * audit log for post-mortem analysis. A circular array of trace buffers 104 * (ao_mca_poll_trace_t structures) is kept to record this activity. Whenever 105 * an event occurs that is of interest to the poller, an entry is made in 106 * the trace array describing that event. 107 */ 108 #define AO_MPT_WHAT_CYC_ERR 0 /* cyclic-induced poll */ 109 #define AO_MPT_WHAT_POKE_ERR 1 /* manually-induced poll */ 110 #define AO_MPT_WHAT_UNFAULTING 2 /* discarded error state */ 111 112 typedef struct ao_mca_poll_trace { 113 hrtime_t mpt_when; /* timestamp of event */ 114 uint8_t mpt_what; /* AO_MPT_WHAT_* (which event?) */ 115 uint8_t mpt_nerr; /* number of errors discovered */ 116 uint16_t mpt_pad1; 117 uint32_t mpt_pad2; 118 } ao_mca_poll_trace_t; 119 120 /* 121 * Processor error state is saved in logout areas. There are three separate 122 * logout areas, each used for a different purpose. The logout areas are stored 123 * in an array (ao_mca_logout), indexed by the AO_MCA_LOGOUT_* macros. 124 * 125 * The save areas are: 126 * 127 * 1. Exception handler MSR save - Written to by the initial portion of the #mc 128 * handler. Read from by the main body of the exception handler. 129 * 130 * 3. Poller MSR save - Used by the poller to store error state MSR values. 131 * While this logout area doesn't necessarily have to live in the ao_mca_t, 132 * it does so to enhance observability. 133 * 134 * The logout areas contain both global error state (acl_ip, acl_timestamp, 135 * etc.), as well as a bank array. The bank array contains one ao_bank_logout_t 136 * per error reporting bank. 137 */ 138 139 typedef struct ao_bank_logout { 140 uint64_t abl_status; /* Saved MCi_STATUS register */ 141 uint64_t abl_addr; /* Saved MCi_ADDR register */ 142 } ao_bank_logout_t; 143 144 #define AO_ACL_F_PRIV 0x1 /* #mc in kernel mode (else user) */ 145 #define AO_ACL_F_FATAL 0x2 /* logout detected fatal error(s) */ 146 147 typedef struct ao_cpu_logout { 148 ao_data_t *acl_ao; /* pointer to per-cpu ao_data_t */ 149 uintptr_t acl_ip; /* instruction pointer if #mc trap */ 150 uint64_t acl_timestamp; /* gethrtime() at time of logout */ 151 uint64_t acl_mcg_status; /* MCG_STATUS register value */ 152 ao_bank_logout_t acl_banks[AMD_MCA_BANK_COUNT]; /* bank state saves */ 153 pc_t acl_stack[FM_STK_DEPTH]; /* saved stack trace (if any) */ 154 int acl_stackdepth; /* saved stack trace depth */ 155 uint_t acl_flags; /* flags (see AO_ACL_F_* above) */ 156 } ao_cpu_logout_t; 157 158 /* Index for ao_mca_logout, below */ 159 #define AO_MCA_LOGOUT_EXCEPTION 0 160 #define AO_MCA_LOGOUT_POLLER 1 161 #define AO_MCA_LOGOUT_NUM 2 162 163 #define AO_MCA_F_UNFAULTING 0x1 /* CPU exiting faulted state */ 164 165 /* 166 * We store config as inherited from the BIOS to assist in troubleshooting. 167 */ 168 typedef struct ao_bios_cfg { 169 uint64_t bcfg_bank_ctl[AMD_MCA_BANK_COUNT]; 170 uint64_t bcfg_bank_mask[AMD_MCA_BANK_COUNT]; 171 uint32_t bcfg_nb_cfg; 172 } ao_bios_cfg_t; 173 174 /* 175 * The master data structure used to hold MCA-related state. 176 */ 177 typedef struct ao_mca { 178 ao_bios_cfg_t ao_mca_bios_cfg; /* Bank and NB config before our init */ 179 ao_cpu_logout_t ao_mca_logout[AO_MCA_LOGOUT_NUM]; /* save areas */ 180 kmutex_t ao_mca_poll_lock; /* keep pollers from colliding */ 181 ao_mca_poll_trace_t *ao_mca_poll_trace; /* trace buffers for this cpu */ 182 uint_t ao_mca_poll_curtrace; /* most recently-filled trace buffer */ 183 uint_t ao_mca_flags; /* AO_MCA_F_* */ 184 } ao_mca_t; 185 186 /* 187 * Per-CPU state 188 */ 189 struct ao_data { 190 ao_mca_t ao_mca; /* MCA state for this CPU */ 191 cpu_t *ao_cpu; /* link to CPU's cpu_t */ 192 const cmi_mc_ops_t *ao_mc_ops; /* memory controller ops */ 193 void *ao_mc_data; /* argument for memory controller ops */ 194 }; 195 196 #ifdef _KERNEL 197 198 struct regs; 199 200 extern errorq_t *ao_mca_queue; 201 extern const cmi_ops_t _cmi_ops; 202 203 extern void ao_faulted_enter(void *); 204 extern void ao_faulted_exit(void *); 205 extern int ao_scrubber_enable(void *, uint64_t, uint64_t); 206 207 extern void ao_mca_post_init(void *); 208 extern void ao_mca_init(void *); 209 extern int ao_mca_trap(void *, struct regs *); 210 extern int ao_mca_inject(void *, cmi_mca_regs_t *, uint_t); 211 extern void ao_mca_poke(void *); 212 extern void ao_mca_poll_init(ao_mca_t *); 213 extern void ao_mca_poll_start(void); 214 215 extern int ao_mca_logout(ao_cpu_logout_t *, struct regs *, int *); 216 extern void ao_mca_drain(void *, const void *, const errorq_elem_t *); 217 extern nvlist_t *ao_fmri_create(ao_data_t *, nv_alloc_t *); 218 219 extern void ao_mc_register(void *, const cmi_mc_ops_t *, void *); 220 extern const struct cmi_mc_ops *ao_mc_getops(void *); 221 extern int ao_mc_patounum(ao_data_t *, uint64_t, uint32_t, int, mc_unum_t *); 222 extern int ao_mc_unumtopa(ao_data_t *, mc_unum_t *, nvlist_t *, uint64_t *); 223 224 extern void ao_pcicfg_write(uint_t, uint_t, uint_t, uint32_t); 225 extern uint32_t ao_pcicfg_read(uint_t, uint_t, uint_t); 226 227 #endif /* _KERNEL */ 228 229 #ifdef __cplusplus 230 } 231 #endif 232 233 #endif /* _AO_H */ 234