10Sstevel@tonic-gate /* 20Sstevel@tonic-gate * CDDL HEADER START 30Sstevel@tonic-gate * 40Sstevel@tonic-gate * The contents of this file are subject to the terms of the 52181Sayznaga * Common Development and Distribution License (the "License"). 62181Sayznaga * You may not use this file except in compliance with the License. 70Sstevel@tonic-gate * 80Sstevel@tonic-gate * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 90Sstevel@tonic-gate * or http://www.opensolaris.org/os/licensing. 100Sstevel@tonic-gate * See the License for the specific language governing permissions 110Sstevel@tonic-gate * and limitations under the License. 120Sstevel@tonic-gate * 130Sstevel@tonic-gate * When distributing Covered Code, include this CDDL HEADER in each 140Sstevel@tonic-gate * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 150Sstevel@tonic-gate * If applicable, add the following below this CDDL HEADER, with the 160Sstevel@tonic-gate * fields enclosed by brackets "[]" replaced with your own identifying 170Sstevel@tonic-gate * information: Portions Copyright [yyyy] [name of copyright owner] 180Sstevel@tonic-gate * 190Sstevel@tonic-gate * CDDL HEADER END 200Sstevel@tonic-gate */ 210Sstevel@tonic-gate /* 221280Srf157361 * Copyright 2006 Sun Microsystems, Inc. All rights reserved. 230Sstevel@tonic-gate * Use is subject to license terms. 240Sstevel@tonic-gate */ 250Sstevel@tonic-gate 260Sstevel@tonic-gate #pragma ident "%Z%%M% %I% %E% SMI" 270Sstevel@tonic-gate 280Sstevel@tonic-gate #include <sys/types.h> 290Sstevel@tonic-gate #include <sys/machsystm.h> 30*3199Sep32863 #include <sys/sysmacros.h> 310Sstevel@tonic-gate #include <sys/cpuvar.h> 320Sstevel@tonic-gate #include <sys/async.h> 330Sstevel@tonic-gate #include <sys/ontrap.h> 340Sstevel@tonic-gate #include <sys/ddifm.h> 350Sstevel@tonic-gate #include <sys/hypervisor_api.h> 360Sstevel@tonic-gate #include <sys/errorq.h> 370Sstevel@tonic-gate #include <sys/promif.h> 380Sstevel@tonic-gate #include <sys/prom_plat.h> 390Sstevel@tonic-gate #include <sys/x_call.h> 400Sstevel@tonic-gate #include <sys/error.h> 410Sstevel@tonic-gate #include <sys/fm/util.h> 42541Srf157361 #include <sys/ivintr.h> 433156Sgirish #include <sys/archsystm.h> 440Sstevel@tonic-gate 450Sstevel@tonic-gate #define MAX_CE_FLTS 10 460Sstevel@tonic-gate #define MAX_ASYNC_FLTS 6 470Sstevel@tonic-gate 480Sstevel@tonic-gate errorq_t *ue_queue; /* queue of uncorrectable errors */ 490Sstevel@tonic-gate errorq_t *ce_queue; /* queue of correctable errors */ 500Sstevel@tonic-gate 510Sstevel@tonic-gate /* 520Sstevel@tonic-gate * Being used by memory test driver. 530Sstevel@tonic-gate * ce_verbose_memory - covers CEs in DIMMs 540Sstevel@tonic-gate * ce_verbose_other - covers "others" (ecache, IO, etc.) 550Sstevel@tonic-gate * 560Sstevel@tonic-gate * If the value is 0, nothing is logged. 570Sstevel@tonic-gate * If the value is 1, the error is logged to the log file, but not console. 580Sstevel@tonic-gate * If the value is 2, the error is logged to the log file and console. 590Sstevel@tonic-gate */ 600Sstevel@tonic-gate int ce_verbose_memory = 1; 610Sstevel@tonic-gate int ce_verbose_other = 1; 620Sstevel@tonic-gate 630Sstevel@tonic-gate int ce_show_data = 0; 640Sstevel@tonic-gate int ce_debug = 0; 650Sstevel@tonic-gate int ue_debug = 0; 660Sstevel@tonic-gate int reset_debug = 0; 670Sstevel@tonic-gate 680Sstevel@tonic-gate /* 690Sstevel@tonic-gate * Tunables for controlling the handling of asynchronous faults (AFTs). Setting 700Sstevel@tonic-gate * these to non-default values on a non-DEBUG kernel is NOT supported. 710Sstevel@tonic-gate */ 720Sstevel@tonic-gate int aft_verbose = 0; /* log AFT messages > 1 to log only */ 730Sstevel@tonic-gate int aft_panic = 0; /* panic (not reboot) on fatal usermode AFLT */ 740Sstevel@tonic-gate int aft_testfatal = 0; /* force all AFTs to panic immediately */ 750Sstevel@tonic-gate 760Sstevel@tonic-gate /* 77541Srf157361 * Used for vbsc hostshutdown (power-off buton) 78541Srf157361 */ 79541Srf157361 int err_shutdown_triggered = 0; /* only once */ 802973Sgovinda uint64_t err_shutdown_inum = 0; /* used to pull the trigger */ 81541Srf157361 82541Srf157361 /* 830Sstevel@tonic-gate * Defined in bus_func.c but initialised in error_init 840Sstevel@tonic-gate */ 850Sstevel@tonic-gate extern kmutex_t bfd_lock; 860Sstevel@tonic-gate 870Sstevel@tonic-gate static uint32_t rq_overflow_count = 0; /* counter for rq overflow */ 880Sstevel@tonic-gate 890Sstevel@tonic-gate static void cpu_queue_one_event(errh_async_flt_t *); 900Sstevel@tonic-gate static uint32_t count_entries_on_queue(uint64_t, uint64_t, uint32_t); 91917Selowe static void errh_page_retire(errh_async_flt_t *, uchar_t); 920Sstevel@tonic-gate static int errh_error_protected(struct regs *, struct async_flt *, int *); 930Sstevel@tonic-gate static void errh_rq_full(struct async_flt *); 940Sstevel@tonic-gate static void ue_drain(void *, struct async_flt *, errorq_elem_t *); 950Sstevel@tonic-gate static void ce_drain(void *, struct async_flt *, errorq_elem_t *); 963156Sgirish static void errh_handle_attr(errh_async_flt_t *); 973156Sgirish static void errh_handle_asr(errh_async_flt_t *); 980Sstevel@tonic-gate 990Sstevel@tonic-gate /*ARGSUSED*/ 1000Sstevel@tonic-gate void 1010Sstevel@tonic-gate process_resumable_error(struct regs *rp, uint32_t head_offset, 1020Sstevel@tonic-gate uint32_t tail_offset) 1030Sstevel@tonic-gate { 1040Sstevel@tonic-gate struct machcpu *mcpup; 1050Sstevel@tonic-gate struct async_flt *aflt; 1060Sstevel@tonic-gate errh_async_flt_t errh_flt; 1070Sstevel@tonic-gate errh_er_t *head_va; 1080Sstevel@tonic-gate 1090Sstevel@tonic-gate mcpup = &(CPU->cpu_m); 1100Sstevel@tonic-gate 1110Sstevel@tonic-gate while (head_offset != tail_offset) { 1120Sstevel@tonic-gate /* kernel buffer starts right after the resumable queue */ 1130Sstevel@tonic-gate head_va = (errh_er_t *)(mcpup->cpu_rq_va + head_offset + 1140Sstevel@tonic-gate CPU_RQ_SIZE); 1150Sstevel@tonic-gate /* Copy the error report to local buffer */ 1160Sstevel@tonic-gate bzero(&errh_flt, sizeof (errh_async_flt_t)); 1170Sstevel@tonic-gate bcopy((char *)head_va, &(errh_flt.errh_er), 1180Sstevel@tonic-gate sizeof (errh_er_t)); 1190Sstevel@tonic-gate 1200Sstevel@tonic-gate /* Increment the queue head */ 1210Sstevel@tonic-gate head_offset += Q_ENTRY_SIZE; 1220Sstevel@tonic-gate /* Wrap around */ 1230Sstevel@tonic-gate head_offset &= (CPU_RQ_SIZE - 1); 1240Sstevel@tonic-gate 1250Sstevel@tonic-gate /* set error handle to zero so it can hold new error report */ 1260Sstevel@tonic-gate head_va->ehdl = 0; 1270Sstevel@tonic-gate 1280Sstevel@tonic-gate switch (errh_flt.errh_er.desc) { 1290Sstevel@tonic-gate case ERRH_DESC_UCOR_RE: 1303156Sgirish /* 1313156Sgirish * Check error attribute, handle individual error 1323156Sgirish * if it is needed. 1333156Sgirish */ 1343156Sgirish errh_handle_attr(&errh_flt); 1350Sstevel@tonic-gate break; 1360Sstevel@tonic-gate 137541Srf157361 case ERRH_DESC_WARN_RE: 138541Srf157361 /* 139541Srf157361 * Power-off requested, but handle it one time only. 140541Srf157361 */ 141541Srf157361 if (!err_shutdown_triggered) { 142541Srf157361 setsoftint(err_shutdown_inum); 143541Srf157361 ++err_shutdown_triggered; 144541Srf157361 } 145541Srf157361 continue; 146541Srf157361 1470Sstevel@tonic-gate default: 1480Sstevel@tonic-gate cmn_err(CE_WARN, "Error Descriptor 0x%llx " 1490Sstevel@tonic-gate " invalid in resumable error handler", 1500Sstevel@tonic-gate (long long) errh_flt.errh_er.desc); 1510Sstevel@tonic-gate continue; 1520Sstevel@tonic-gate } 1530Sstevel@tonic-gate 1540Sstevel@tonic-gate aflt = (struct async_flt *)&(errh_flt.cmn_asyncflt); 1550Sstevel@tonic-gate aflt->flt_id = gethrtime(); 1560Sstevel@tonic-gate aflt->flt_bus_id = getprocessorid(); 1570Sstevel@tonic-gate aflt->flt_class = CPU_FAULT; 1580Sstevel@tonic-gate aflt->flt_prot = AFLT_PROT_NONE; 1590Sstevel@tonic-gate aflt->flt_priv = (((errh_flt.errh_er.attr & ERRH_MODE_MASK) 1600Sstevel@tonic-gate >> ERRH_MODE_SHIFT) == ERRH_MODE_PRIV); 1610Sstevel@tonic-gate 1620Sstevel@tonic-gate if (errh_flt.errh_er.attr & ERRH_ATTR_CPU) 1630Sstevel@tonic-gate /* If it is an error on other cpu */ 1640Sstevel@tonic-gate aflt->flt_panic = 1; 1650Sstevel@tonic-gate else 1660Sstevel@tonic-gate aflt->flt_panic = 0; 1670Sstevel@tonic-gate 1680Sstevel@tonic-gate /* 1690Sstevel@tonic-gate * Handle resumable queue full case. 1700Sstevel@tonic-gate */ 1710Sstevel@tonic-gate if (errh_flt.errh_er.attr & ERRH_ATTR_RQF) { 1720Sstevel@tonic-gate (void) errh_rq_full(aflt); 1730Sstevel@tonic-gate } 1740Sstevel@tonic-gate 1750Sstevel@tonic-gate /* 1760Sstevel@tonic-gate * Queue the error on ce or ue queue depend on flt_panic. 1770Sstevel@tonic-gate * Even if flt_panic is set, the code still keep processing 1780Sstevel@tonic-gate * the rest element on rq until the panic starts. 1790Sstevel@tonic-gate */ 1800Sstevel@tonic-gate (void) cpu_queue_one_event(&errh_flt); 1810Sstevel@tonic-gate 1820Sstevel@tonic-gate /* 1830Sstevel@tonic-gate * Panic here if aflt->flt_panic has been set. 1840Sstevel@tonic-gate * Enqueued errors will be logged as part of the panic flow. 1850Sstevel@tonic-gate */ 1860Sstevel@tonic-gate if (aflt->flt_panic) { 1870Sstevel@tonic-gate fm_panic("Unrecoverable error on another CPU"); 1880Sstevel@tonic-gate } 1890Sstevel@tonic-gate } 1900Sstevel@tonic-gate } 1910Sstevel@tonic-gate 1920Sstevel@tonic-gate void 1931457Swh94709 process_nonresumable_error(struct regs *rp, uint64_t flags, 1940Sstevel@tonic-gate uint32_t head_offset, uint32_t tail_offset) 1950Sstevel@tonic-gate { 1960Sstevel@tonic-gate struct machcpu *mcpup; 1970Sstevel@tonic-gate struct async_flt *aflt; 1980Sstevel@tonic-gate errh_async_flt_t errh_flt; 1990Sstevel@tonic-gate errh_er_t *head_va; 2000Sstevel@tonic-gate int trampolined = 0; 2010Sstevel@tonic-gate int expected = DDI_FM_ERR_UNEXPECTED; 2020Sstevel@tonic-gate uint64_t exec_mode; 2031457Swh94709 uint8_t u_spill_fill; 2040Sstevel@tonic-gate 2050Sstevel@tonic-gate mcpup = &(CPU->cpu_m); 2060Sstevel@tonic-gate 2070Sstevel@tonic-gate while (head_offset != tail_offset) { 2080Sstevel@tonic-gate /* kernel buffer starts right after the nonresumable queue */ 2090Sstevel@tonic-gate head_va = (errh_er_t *)(mcpup->cpu_nrq_va + head_offset + 2100Sstevel@tonic-gate CPU_NRQ_SIZE); 2110Sstevel@tonic-gate 2120Sstevel@tonic-gate /* Copy the error report to local buffer */ 2130Sstevel@tonic-gate bzero(&errh_flt, sizeof (errh_async_flt_t)); 2140Sstevel@tonic-gate 2150Sstevel@tonic-gate bcopy((char *)head_va, &(errh_flt.errh_er), 2160Sstevel@tonic-gate sizeof (errh_er_t)); 2170Sstevel@tonic-gate 2180Sstevel@tonic-gate /* Increment the queue head */ 2190Sstevel@tonic-gate head_offset += Q_ENTRY_SIZE; 2200Sstevel@tonic-gate /* Wrap around */ 2210Sstevel@tonic-gate head_offset &= (CPU_NRQ_SIZE - 1); 2220Sstevel@tonic-gate 2230Sstevel@tonic-gate /* set error handle to zero so it can hold new error report */ 2240Sstevel@tonic-gate head_va->ehdl = 0; 2250Sstevel@tonic-gate 2260Sstevel@tonic-gate aflt = (struct async_flt *)&(errh_flt.cmn_asyncflt); 2270Sstevel@tonic-gate 2280Sstevel@tonic-gate trampolined = 0; 2290Sstevel@tonic-gate 2300Sstevel@tonic-gate if (errh_flt.errh_er.attr & ERRH_ATTR_PIO) 2310Sstevel@tonic-gate aflt->flt_class = BUS_FAULT; 2320Sstevel@tonic-gate else 2330Sstevel@tonic-gate aflt->flt_class = CPU_FAULT; 2340Sstevel@tonic-gate 2350Sstevel@tonic-gate aflt->flt_id = gethrtime(); 2360Sstevel@tonic-gate aflt->flt_bus_id = getprocessorid(); 2370Sstevel@tonic-gate aflt->flt_pc = (caddr_t)rp->r_pc; 2380Sstevel@tonic-gate exec_mode = (errh_flt.errh_er.attr & ERRH_MODE_MASK) 2390Sstevel@tonic-gate >> ERRH_MODE_SHIFT; 2400Sstevel@tonic-gate aflt->flt_priv = (exec_mode == ERRH_MODE_PRIV || 2410Sstevel@tonic-gate exec_mode == ERRH_MODE_UNKNOWN); 2420Sstevel@tonic-gate aflt->flt_prot = AFLT_PROT_NONE; 2431457Swh94709 aflt->flt_tl = (uchar_t)(flags & ERRH_TL_MASK); 2440Sstevel@tonic-gate aflt->flt_panic = ((aflt->flt_tl != 0) || 2450Sstevel@tonic-gate (aft_testfatal != 0)); 2460Sstevel@tonic-gate 2471457Swh94709 /* 2481457Swh94709 * For the first error packet on the queue, check if it 2491457Swh94709 * happened in user fill/spill trap. 2501457Swh94709 */ 2511457Swh94709 if (flags & ERRH_U_SPILL_FILL) { 2521457Swh94709 u_spill_fill = 1; 2531457Swh94709 /* clear the user fill/spill flag in flags */ 2541457Swh94709 flags = (uint64_t)aflt->flt_tl; 2551457Swh94709 } else 2561457Swh94709 u_spill_fill = 0; 2571457Swh94709 2580Sstevel@tonic-gate switch (errh_flt.errh_er.desc) { 2590Sstevel@tonic-gate case ERRH_DESC_PR_NRE: 2601457Swh94709 if (u_spill_fill) { 2611457Swh94709 aflt->flt_panic = 0; 2621457Swh94709 break; 2631457Swh94709 } 2640Sstevel@tonic-gate /* 2650Sstevel@tonic-gate * Fall through, precise fault also need to check 2660Sstevel@tonic-gate * to see if it was protected. 2670Sstevel@tonic-gate */ 2681457Swh94709 /*FALLTHRU*/ 2690Sstevel@tonic-gate 2700Sstevel@tonic-gate case ERRH_DESC_DEF_NRE: 2710Sstevel@tonic-gate /* 2720Sstevel@tonic-gate * If the trap occurred in privileged mode at TL=0, 2730Sstevel@tonic-gate * we need to check to see if we were executing 2740Sstevel@tonic-gate * in kernel under on_trap() or t_lofault 2751280Srf157361 * protection. If so, and if it was a PIO or MEM 2761280Srf157361 * error, then modify the saved registers so that 2771280Srf157361 * we return from the trap to the appropriate 2781280Srf157361 * trampoline routine. 2790Sstevel@tonic-gate */ 2801280Srf157361 if (aflt->flt_priv == 1 && aflt->flt_tl == 0 && 2811280Srf157361 ((errh_flt.errh_er.attr & ERRH_ATTR_PIO) || 2821280Srf157361 (errh_flt.errh_er.attr & ERRH_ATTR_MEM))) { 2830Sstevel@tonic-gate trampolined = 2840Sstevel@tonic-gate errh_error_protected(rp, aflt, &expected); 2851280Srf157361 } 2860Sstevel@tonic-gate 2870Sstevel@tonic-gate if (!aflt->flt_priv || aflt->flt_prot == 2880Sstevel@tonic-gate AFLT_PROT_COPY) { 2890Sstevel@tonic-gate aflt->flt_panic |= aft_panic; 2900Sstevel@tonic-gate } else if (!trampolined && 2912477Srf157361 (aflt->flt_class != BUS_FAULT)) { 2920Sstevel@tonic-gate aflt->flt_panic = 1; 2930Sstevel@tonic-gate } 2940Sstevel@tonic-gate 2950Sstevel@tonic-gate /* 2963156Sgirish * Check error attribute, handle individual error 2973156Sgirish * if it is needed. 2983156Sgirish */ 2993156Sgirish errh_handle_attr(&errh_flt); 3003156Sgirish 3013156Sgirish /* 3020Sstevel@tonic-gate * If PIO error, we need to query the bus nexus 3030Sstevel@tonic-gate * for fatal errors. 3040Sstevel@tonic-gate */ 3050Sstevel@tonic-gate if (aflt->flt_class == BUS_FAULT) { 3060Sstevel@tonic-gate aflt->flt_addr = errh_flt.errh_er.ra; 3070Sstevel@tonic-gate errh_cpu_run_bus_error_handlers(aflt, 3080Sstevel@tonic-gate expected); 3090Sstevel@tonic-gate } 3100Sstevel@tonic-gate 3110Sstevel@tonic-gate break; 3120Sstevel@tonic-gate 3130Sstevel@tonic-gate default: 3142218Swh94709 cmn_err(CE_WARN, "Panic - Error Descriptor 0x%llx " 3152218Swh94709 " invalid in non-resumable error handler", 3160Sstevel@tonic-gate (long long) errh_flt.errh_er.desc); 3172218Swh94709 aflt->flt_panic = 1; 3182218Swh94709 break; 3190Sstevel@tonic-gate } 3200Sstevel@tonic-gate 3210Sstevel@tonic-gate /* 3220Sstevel@tonic-gate * Queue the error report for further processing. If 3230Sstevel@tonic-gate * flt_panic is set, code still process other errors 3240Sstevel@tonic-gate * in the queue until the panic routine stops the 3250Sstevel@tonic-gate * kernel. 3260Sstevel@tonic-gate */ 3270Sstevel@tonic-gate (void) cpu_queue_one_event(&errh_flt); 3280Sstevel@tonic-gate 3290Sstevel@tonic-gate /* 3300Sstevel@tonic-gate * Panic here if aflt->flt_panic has been set. 3310Sstevel@tonic-gate * Enqueued errors will be logged as part of the panic flow. 3320Sstevel@tonic-gate */ 3330Sstevel@tonic-gate if (aflt->flt_panic) { 3340Sstevel@tonic-gate fm_panic("Unrecoverable hardware error"); 3350Sstevel@tonic-gate } 3360Sstevel@tonic-gate 3370Sstevel@tonic-gate /* 338917Selowe * Call page_retire() to handle memory errors. 3390Sstevel@tonic-gate */ 3400Sstevel@tonic-gate if (errh_flt.errh_er.attr & ERRH_ATTR_MEM) 341917Selowe errh_page_retire(&errh_flt, PR_UE); 3420Sstevel@tonic-gate 3430Sstevel@tonic-gate /* 3441457Swh94709 * If we queued an error and the it was in user mode, or 3451457Swh94709 * protected by t_lofault, or user_spill_fill is set, we 3460Sstevel@tonic-gate * set AST flag so the queue will be drained before 3470Sstevel@tonic-gate * returning to user mode. 3480Sstevel@tonic-gate */ 3491457Swh94709 if (!aflt->flt_priv || aflt->flt_prot == AFLT_PROT_COPY || 3501457Swh94709 u_spill_fill) { 3510Sstevel@tonic-gate int pcb_flag = 0; 3520Sstevel@tonic-gate 3530Sstevel@tonic-gate if (aflt->flt_class == CPU_FAULT) 3540Sstevel@tonic-gate pcb_flag |= ASYNC_HWERR; 3550Sstevel@tonic-gate else if (aflt->flt_class == BUS_FAULT) 3560Sstevel@tonic-gate pcb_flag |= ASYNC_BERR; 3570Sstevel@tonic-gate 3580Sstevel@tonic-gate ttolwp(curthread)->lwp_pcb.pcb_flags |= pcb_flag; 3590Sstevel@tonic-gate aston(curthread); 3600Sstevel@tonic-gate } 3610Sstevel@tonic-gate } 3620Sstevel@tonic-gate } 3630Sstevel@tonic-gate 3640Sstevel@tonic-gate /* 3650Sstevel@tonic-gate * For PIO errors, this routine calls nexus driver's error 3660Sstevel@tonic-gate * callback routines. If the callback routine returns fatal, and 3670Sstevel@tonic-gate * we are in kernel or unknow mode without any error protection, 3680Sstevel@tonic-gate * we need to turn on the panic flag. 3690Sstevel@tonic-gate */ 3700Sstevel@tonic-gate void 3710Sstevel@tonic-gate errh_cpu_run_bus_error_handlers(struct async_flt *aflt, int expected) 3720Sstevel@tonic-gate { 3730Sstevel@tonic-gate int status; 3740Sstevel@tonic-gate ddi_fm_error_t de; 3750Sstevel@tonic-gate 3760Sstevel@tonic-gate bzero(&de, sizeof (ddi_fm_error_t)); 3770Sstevel@tonic-gate 3780Sstevel@tonic-gate de.fme_version = DDI_FME_VERSION; 3790Sstevel@tonic-gate de.fme_ena = fm_ena_generate(aflt->flt_id, FM_ENA_FMT1); 3800Sstevel@tonic-gate de.fme_flag = expected; 3810Sstevel@tonic-gate de.fme_bus_specific = (void *)aflt->flt_addr; 3820Sstevel@tonic-gate status = ndi_fm_handler_dispatch(ddi_root_node(), NULL, &de); 3830Sstevel@tonic-gate 3840Sstevel@tonic-gate /* 3850Sstevel@tonic-gate * If error is protected, it will jump to proper routine 3860Sstevel@tonic-gate * to handle the handle; if it is in user level, we just 3870Sstevel@tonic-gate * kill the user process; if the driver thinks the error is 3880Sstevel@tonic-gate * not fatal, we can drive on. If none of above are true, 3890Sstevel@tonic-gate * we panic 3900Sstevel@tonic-gate */ 3910Sstevel@tonic-gate if ((aflt->flt_prot == AFLT_PROT_NONE) && (aflt->flt_priv == 1) && 3920Sstevel@tonic-gate (status == DDI_FM_FATAL)) 3930Sstevel@tonic-gate aflt->flt_panic = 1; 3940Sstevel@tonic-gate } 3950Sstevel@tonic-gate 3960Sstevel@tonic-gate /* 3970Sstevel@tonic-gate * This routine checks to see if we are under any error protection when 3980Sstevel@tonic-gate * the error happens. If we are under error protection, we unwind to 3990Sstevel@tonic-gate * the protection and indicate fault. 4000Sstevel@tonic-gate */ 4010Sstevel@tonic-gate static int 4020Sstevel@tonic-gate errh_error_protected(struct regs *rp, struct async_flt *aflt, int *expected) 4030Sstevel@tonic-gate { 4040Sstevel@tonic-gate int trampolined = 0; 4050Sstevel@tonic-gate ddi_acc_hdl_t *hp; 4060Sstevel@tonic-gate 4070Sstevel@tonic-gate if (curthread->t_ontrap != NULL) { 4080Sstevel@tonic-gate on_trap_data_t *otp = curthread->t_ontrap; 4090Sstevel@tonic-gate 4100Sstevel@tonic-gate if (otp->ot_prot & OT_DATA_EC) { 4110Sstevel@tonic-gate aflt->flt_prot = AFLT_PROT_EC; 4120Sstevel@tonic-gate otp->ot_trap |= OT_DATA_EC; 4130Sstevel@tonic-gate rp->r_pc = otp->ot_trampoline; 4140Sstevel@tonic-gate rp->r_npc = rp->r_pc +4; 4150Sstevel@tonic-gate trampolined = 1; 4160Sstevel@tonic-gate } 4170Sstevel@tonic-gate 4180Sstevel@tonic-gate if (otp->ot_prot & OT_DATA_ACCESS) { 4190Sstevel@tonic-gate aflt->flt_prot = AFLT_PROT_ACCESS; 4200Sstevel@tonic-gate otp->ot_trap |= OT_DATA_ACCESS; 4210Sstevel@tonic-gate rp->r_pc = otp->ot_trampoline; 4220Sstevel@tonic-gate rp->r_npc = rp->r_pc + 4; 4230Sstevel@tonic-gate trampolined = 1; 4240Sstevel@tonic-gate /* 4250Sstevel@tonic-gate * for peek and caut_gets 4260Sstevel@tonic-gate * errors are expected 4270Sstevel@tonic-gate */ 4280Sstevel@tonic-gate hp = (ddi_acc_hdl_t *)otp->ot_handle; 4290Sstevel@tonic-gate if (!hp) 4300Sstevel@tonic-gate *expected = DDI_FM_ERR_PEEK; 4310Sstevel@tonic-gate else if (hp->ah_acc.devacc_attr_access == 4320Sstevel@tonic-gate DDI_CAUTIOUS_ACC) 4330Sstevel@tonic-gate *expected = DDI_FM_ERR_EXPECTED; 4340Sstevel@tonic-gate } 4350Sstevel@tonic-gate } else if (curthread->t_lofault) { 4360Sstevel@tonic-gate aflt->flt_prot = AFLT_PROT_COPY; 4370Sstevel@tonic-gate rp->r_g1 = EFAULT; 4380Sstevel@tonic-gate rp->r_pc = curthread->t_lofault; 4390Sstevel@tonic-gate rp->r_npc = rp->r_pc + 4; 4400Sstevel@tonic-gate trampolined = 1; 4410Sstevel@tonic-gate } 4420Sstevel@tonic-gate 4430Sstevel@tonic-gate return (trampolined); 4440Sstevel@tonic-gate } 4450Sstevel@tonic-gate 4460Sstevel@tonic-gate /* 4470Sstevel@tonic-gate * Queue one event. 4480Sstevel@tonic-gate */ 4490Sstevel@tonic-gate static void 4500Sstevel@tonic-gate cpu_queue_one_event(errh_async_flt_t *errh_fltp) 4510Sstevel@tonic-gate { 4520Sstevel@tonic-gate struct async_flt *aflt = (struct async_flt *)errh_fltp; 4530Sstevel@tonic-gate errorq_t *eqp; 4540Sstevel@tonic-gate 4550Sstevel@tonic-gate if (aflt->flt_panic) 4560Sstevel@tonic-gate eqp = ue_queue; 4570Sstevel@tonic-gate else 4580Sstevel@tonic-gate eqp = ce_queue; 4590Sstevel@tonic-gate 4600Sstevel@tonic-gate errorq_dispatch(eqp, errh_fltp, sizeof (errh_async_flt_t), 4610Sstevel@tonic-gate aflt->flt_panic); 4620Sstevel@tonic-gate } 4630Sstevel@tonic-gate 4640Sstevel@tonic-gate /* 4650Sstevel@tonic-gate * The cpu_async_log_err() function is called by the ce/ue_drain() function to 4660Sstevel@tonic-gate * handle logging for CPU events that are dequeued. As such, it can be invoked 4670Sstevel@tonic-gate * from softint context, from AST processing in the trap() flow, or from the 4680Sstevel@tonic-gate * panic flow. We decode the CPU-specific data, and log appropriate messages. 4690Sstevel@tonic-gate */ 4700Sstevel@tonic-gate void 4710Sstevel@tonic-gate cpu_async_log_err(void *flt) 4720Sstevel@tonic-gate { 4730Sstevel@tonic-gate errh_async_flt_t *errh_fltp = (errh_async_flt_t *)flt; 4740Sstevel@tonic-gate errh_er_t *errh_erp = (errh_er_t *)&errh_fltp->errh_er; 4750Sstevel@tonic-gate 4760Sstevel@tonic-gate switch (errh_erp->desc) { 4770Sstevel@tonic-gate case ERRH_DESC_UCOR_RE: 4780Sstevel@tonic-gate if (errh_erp->attr & ERRH_ATTR_MEM) { 4790Sstevel@tonic-gate /* 480917Selowe * Turn on the PR_UE flag. The page will be 4810Sstevel@tonic-gate * scrubbed when it is freed. 4820Sstevel@tonic-gate */ 483917Selowe errh_page_retire(errh_fltp, PR_UE); 4840Sstevel@tonic-gate } 4850Sstevel@tonic-gate 4860Sstevel@tonic-gate break; 4870Sstevel@tonic-gate 4880Sstevel@tonic-gate case ERRH_DESC_PR_NRE: 4890Sstevel@tonic-gate case ERRH_DESC_DEF_NRE: 4900Sstevel@tonic-gate if (errh_erp->attr & ERRH_ATTR_MEM) { 4910Sstevel@tonic-gate /* 4920Sstevel@tonic-gate * For non-resumable memory error, retire 4930Sstevel@tonic-gate * the page here. 4940Sstevel@tonic-gate */ 495917Selowe errh_page_retire(errh_fltp, PR_UE); 496639Swh94709 497639Swh94709 /* 498639Swh94709 * If we are going to panic, scrub the page first 499639Swh94709 */ 500639Swh94709 if (errh_fltp->cmn_asyncflt.flt_panic) 501639Swh94709 mem_scrub(errh_fltp->errh_er.ra, 502639Swh94709 errh_fltp->errh_er.sz); 5030Sstevel@tonic-gate } 5040Sstevel@tonic-gate break; 5050Sstevel@tonic-gate 5060Sstevel@tonic-gate default: 5070Sstevel@tonic-gate break; 5080Sstevel@tonic-gate } 5090Sstevel@tonic-gate } 5100Sstevel@tonic-gate 5110Sstevel@tonic-gate /* 5120Sstevel@tonic-gate * Called from ce_drain(). 5130Sstevel@tonic-gate */ 5140Sstevel@tonic-gate void 5150Sstevel@tonic-gate cpu_ce_log_err(struct async_flt *aflt) 5160Sstevel@tonic-gate { 5170Sstevel@tonic-gate switch (aflt->flt_class) { 5180Sstevel@tonic-gate case CPU_FAULT: 5190Sstevel@tonic-gate cpu_async_log_err(aflt); 5200Sstevel@tonic-gate break; 5210Sstevel@tonic-gate 5220Sstevel@tonic-gate case BUS_FAULT: 5230Sstevel@tonic-gate cpu_async_log_err(aflt); 5240Sstevel@tonic-gate break; 5250Sstevel@tonic-gate 5260Sstevel@tonic-gate default: 5270Sstevel@tonic-gate break; 5280Sstevel@tonic-gate } 5290Sstevel@tonic-gate } 5300Sstevel@tonic-gate 5310Sstevel@tonic-gate /* 5320Sstevel@tonic-gate * Called from ue_drain(). 5330Sstevel@tonic-gate */ 5340Sstevel@tonic-gate void 5350Sstevel@tonic-gate cpu_ue_log_err(struct async_flt *aflt) 5360Sstevel@tonic-gate { 5370Sstevel@tonic-gate switch (aflt->flt_class) { 5380Sstevel@tonic-gate case CPU_FAULT: 5390Sstevel@tonic-gate cpu_async_log_err(aflt); 5400Sstevel@tonic-gate break; 5410Sstevel@tonic-gate 5420Sstevel@tonic-gate case BUS_FAULT: 5430Sstevel@tonic-gate cpu_async_log_err(aflt); 5440Sstevel@tonic-gate break; 5450Sstevel@tonic-gate 5460Sstevel@tonic-gate default: 5470Sstevel@tonic-gate break; 5480Sstevel@tonic-gate } 5490Sstevel@tonic-gate } 5500Sstevel@tonic-gate 5510Sstevel@tonic-gate /* 5520Sstevel@tonic-gate * Turn on flag on the error memory region. 5530Sstevel@tonic-gate */ 5540Sstevel@tonic-gate static void 555917Selowe errh_page_retire(errh_async_flt_t *errh_fltp, uchar_t flag) 5560Sstevel@tonic-gate { 5570Sstevel@tonic-gate uint64_t flt_real_addr_start = errh_fltp->errh_er.ra; 5580Sstevel@tonic-gate uint64_t flt_real_addr_end = flt_real_addr_start + 5590Sstevel@tonic-gate errh_fltp->errh_er.sz - 1; 5600Sstevel@tonic-gate int64_t current_addr; 5610Sstevel@tonic-gate 5620Sstevel@tonic-gate if (errh_fltp->errh_er.sz == 0) 5630Sstevel@tonic-gate return; 5640Sstevel@tonic-gate 5650Sstevel@tonic-gate for (current_addr = flt_real_addr_start; 5660Sstevel@tonic-gate current_addr < flt_real_addr_end; current_addr += MMU_PAGESIZE) { 567917Selowe (void) page_retire(current_addr, flag); 5680Sstevel@tonic-gate } 5690Sstevel@tonic-gate } 5700Sstevel@tonic-gate 5710Sstevel@tonic-gate void 5720Sstevel@tonic-gate mem_scrub(uint64_t paddr, uint64_t len) 5730Sstevel@tonic-gate { 5740Sstevel@tonic-gate uint64_t pa, length, scrubbed_len; 5750Sstevel@tonic-gate 5760Sstevel@tonic-gate pa = paddr; 5770Sstevel@tonic-gate length = len; 5780Sstevel@tonic-gate scrubbed_len = 0; 5790Sstevel@tonic-gate 580639Swh94709 while (length > 0) { 581639Swh94709 if (hv_mem_scrub(pa, length, &scrubbed_len) != H_EOK) 5820Sstevel@tonic-gate break; 5830Sstevel@tonic-gate 5840Sstevel@tonic-gate pa += scrubbed_len; 5850Sstevel@tonic-gate length -= scrubbed_len; 5860Sstevel@tonic-gate } 5870Sstevel@tonic-gate } 5880Sstevel@tonic-gate 5891457Swh94709 /* 590*3199Sep32863 * Call hypervisor to flush the memory region. 591*3199Sep32863 * Both va and len must be MMU_PAGESIZE aligned. 592*3199Sep32863 * Returns the total number of bytes flushed. 5931457Swh94709 */ 594*3199Sep32863 uint64_t 5950Sstevel@tonic-gate mem_sync(caddr_t va, size_t len) 5960Sstevel@tonic-gate { 5970Sstevel@tonic-gate uint64_t pa, length, flushed; 598*3199Sep32863 uint64_t chunk_len = MMU_PAGESIZE; 599*3199Sep32863 uint64_t total_flushed = 0; 6000Sstevel@tonic-gate 601*3199Sep32863 if (((uint64_t)va | (uint64_t)len) & MMU_PAGEOFFSET) 602*3199Sep32863 return (total_flushed); 6030Sstevel@tonic-gate 604*3199Sep32863 while (len > 0) { 605*3199Sep32863 pa = va_to_pa((caddr_t)va); 606*3199Sep32863 if (pa == (uint64_t)-1) 607*3199Sep32863 return (total_flushed); 6080Sstevel@tonic-gate 609*3199Sep32863 length = chunk_len; 610*3199Sep32863 flushed = 0; 6111457Swh94709 612*3199Sep32863 while (length > 0) { 613*3199Sep32863 if (hv_mem_sync(pa, length, &flushed) != H_EOK) 614*3199Sep32863 return (total_flushed); 6150Sstevel@tonic-gate 616*3199Sep32863 pa += flushed; 617*3199Sep32863 length -= flushed; 618*3199Sep32863 total_flushed += flushed; 619*3199Sep32863 } 6200Sstevel@tonic-gate 621*3199Sep32863 va += chunk_len; 622*3199Sep32863 len -= chunk_len; 6230Sstevel@tonic-gate } 624*3199Sep32863 625*3199Sep32863 return (total_flushed); 6260Sstevel@tonic-gate } 6270Sstevel@tonic-gate 6280Sstevel@tonic-gate /* 6290Sstevel@tonic-gate * If resumable queue is full, we need to check if any cpu is in 6300Sstevel@tonic-gate * error state. If not, we drive on. If yes, we need to panic. The 6310Sstevel@tonic-gate * hypervisor call hv_cpu_state() is being used for checking the 6320Sstevel@tonic-gate * cpu state. 6330Sstevel@tonic-gate */ 6340Sstevel@tonic-gate static void 6350Sstevel@tonic-gate errh_rq_full(struct async_flt *afltp) 6360Sstevel@tonic-gate { 6370Sstevel@tonic-gate processorid_t who; 6380Sstevel@tonic-gate uint64_t cpu_state; 6390Sstevel@tonic-gate uint64_t retval; 6400Sstevel@tonic-gate 6410Sstevel@tonic-gate for (who = 0; who < NCPU; who++) 6420Sstevel@tonic-gate if (CPU_IN_SET(cpu_ready_set, who)) { 6430Sstevel@tonic-gate retval = hv_cpu_state(who, &cpu_state); 6440Sstevel@tonic-gate if (retval != H_EOK || cpu_state == CPU_STATE_ERROR) { 6450Sstevel@tonic-gate afltp->flt_panic = 1; 6460Sstevel@tonic-gate break; 6470Sstevel@tonic-gate } 6480Sstevel@tonic-gate } 6490Sstevel@tonic-gate } 6500Sstevel@tonic-gate 6510Sstevel@tonic-gate /* 6520Sstevel@tonic-gate * Return processor specific async error structure 6530Sstevel@tonic-gate * size used. 6540Sstevel@tonic-gate */ 6550Sstevel@tonic-gate int 6560Sstevel@tonic-gate cpu_aflt_size(void) 6570Sstevel@tonic-gate { 6580Sstevel@tonic-gate return (sizeof (errh_async_flt_t)); 6590Sstevel@tonic-gate } 6600Sstevel@tonic-gate 6610Sstevel@tonic-gate #define SZ_TO_ETRS_SHIFT 6 6620Sstevel@tonic-gate 6630Sstevel@tonic-gate /* 6640Sstevel@tonic-gate * Message print out when resumable queue is overflown 6650Sstevel@tonic-gate */ 6660Sstevel@tonic-gate /*ARGSUSED*/ 6670Sstevel@tonic-gate void 6680Sstevel@tonic-gate rq_overflow(struct regs *rp, uint64_t head_offset, 6690Sstevel@tonic-gate uint64_t tail_offset) 6700Sstevel@tonic-gate { 6710Sstevel@tonic-gate rq_overflow_count++; 6720Sstevel@tonic-gate } 6730Sstevel@tonic-gate 6740Sstevel@tonic-gate /* 6750Sstevel@tonic-gate * Handler to process a fatal error. This routine can be called from a 6760Sstevel@tonic-gate * softint, called from trap()'s AST handling, or called from the panic flow. 6770Sstevel@tonic-gate */ 6780Sstevel@tonic-gate /*ARGSUSED*/ 6790Sstevel@tonic-gate static void 6800Sstevel@tonic-gate ue_drain(void *ignored, struct async_flt *aflt, errorq_elem_t *eqep) 6810Sstevel@tonic-gate { 6820Sstevel@tonic-gate cpu_ue_log_err(aflt); 6830Sstevel@tonic-gate } 6840Sstevel@tonic-gate 6850Sstevel@tonic-gate /* 6860Sstevel@tonic-gate * Handler to process a correctable error. This routine can be called from a 6870Sstevel@tonic-gate * softint. We just call the CPU module's logging routine. 6880Sstevel@tonic-gate */ 6890Sstevel@tonic-gate /*ARGSUSED*/ 6900Sstevel@tonic-gate static void 6910Sstevel@tonic-gate ce_drain(void *ignored, struct async_flt *aflt, errorq_elem_t *eqep) 6920Sstevel@tonic-gate { 6930Sstevel@tonic-gate cpu_ce_log_err(aflt); 6940Sstevel@tonic-gate } 6950Sstevel@tonic-gate 6960Sstevel@tonic-gate /* 697541Srf157361 * Handler to process vbsc hostshutdown (power-off button). 698541Srf157361 */ 699541Srf157361 static int 700541Srf157361 err_shutdown_softintr() 701541Srf157361 { 702541Srf157361 cmn_err(CE_WARN, "Power-off requested, system will now shutdown."); 703541Srf157361 do_shutdown(); 704541Srf157361 705541Srf157361 /* 706541Srf157361 * just in case do_shutdown() fails 707541Srf157361 */ 708541Srf157361 (void) timeout((void(*)(void *))power_down, NULL, 100 * hz); 709541Srf157361 return (DDI_INTR_CLAIMED); 710541Srf157361 } 711541Srf157361 712541Srf157361 /* 7130Sstevel@tonic-gate * Allocate error queue sizes based on max_ncpus. max_ncpus is set just 7140Sstevel@tonic-gate * after ncpunode has been determined. ncpus is set in start_other_cpus 7150Sstevel@tonic-gate * which is called after error_init() but may change dynamically. 7160Sstevel@tonic-gate */ 7170Sstevel@tonic-gate void 7180Sstevel@tonic-gate error_init(void) 7190Sstevel@tonic-gate { 7200Sstevel@tonic-gate char tmp_name[MAXSYSNAME]; 721789Sahrens pnode_t node; 7220Sstevel@tonic-gate size_t size = cpu_aflt_size(); 7230Sstevel@tonic-gate 7240Sstevel@tonic-gate /* 7250Sstevel@tonic-gate * Initialize the correctable and uncorrectable error queues. 7260Sstevel@tonic-gate */ 7270Sstevel@tonic-gate ue_queue = errorq_create("ue_queue", (errorq_func_t)ue_drain, NULL, 7280Sstevel@tonic-gate MAX_ASYNC_FLTS * (max_ncpus + 1), size, PIL_2, ERRORQ_VITAL); 7290Sstevel@tonic-gate 7300Sstevel@tonic-gate ce_queue = errorq_create("ce_queue", (errorq_func_t)ce_drain, NULL, 7310Sstevel@tonic-gate MAX_CE_FLTS * (max_ncpus + 1), size, PIL_1, 0); 7320Sstevel@tonic-gate 7330Sstevel@tonic-gate if (ue_queue == NULL || ce_queue == NULL) 7340Sstevel@tonic-gate panic("failed to create required system error queue"); 7350Sstevel@tonic-gate 7360Sstevel@tonic-gate /* 737541Srf157361 * Setup interrupt handler for power-off button. 738541Srf157361 */ 739541Srf157361 err_shutdown_inum = add_softintr(PIL_9, 7402973Sgovinda (softintrfunc)err_shutdown_softintr, NULL, SOFTINT_ST); 741541Srf157361 742541Srf157361 /* 7430Sstevel@tonic-gate * Initialize the busfunc list mutex. This must be a PIL_15 spin lock 7440Sstevel@tonic-gate * because we will need to acquire it from cpu_async_error(). 7450Sstevel@tonic-gate */ 7460Sstevel@tonic-gate mutex_init(&bfd_lock, NULL, MUTEX_SPIN, (void *)PIL_15); 7470Sstevel@tonic-gate 7480Sstevel@tonic-gate node = prom_rootnode(); 7490Sstevel@tonic-gate if ((node == OBP_NONODE) || (node == OBP_BADNODE)) { 7500Sstevel@tonic-gate cmn_err(CE_CONT, "error_init: node 0x%x\n", (uint_t)node); 7510Sstevel@tonic-gate return; 7520Sstevel@tonic-gate } 7530Sstevel@tonic-gate 7540Sstevel@tonic-gate if (((size = prom_getproplen(node, "reset-reason")) != -1) && 7550Sstevel@tonic-gate (size <= MAXSYSNAME) && 7560Sstevel@tonic-gate (prom_getprop(node, "reset-reason", tmp_name) != -1)) { 7570Sstevel@tonic-gate if (reset_debug) { 7580Sstevel@tonic-gate cmn_err(CE_CONT, "System booting after %s\n", tmp_name); 7590Sstevel@tonic-gate } else if (strncmp(tmp_name, "FATAL", 5) == 0) { 7600Sstevel@tonic-gate cmn_err(CE_CONT, 7610Sstevel@tonic-gate "System booting after fatal error %s\n", tmp_name); 7620Sstevel@tonic-gate } 7630Sstevel@tonic-gate } 7640Sstevel@tonic-gate } 765817Swh94709 766817Swh94709 /* 767817Swh94709 * Nonresumable queue is full, panic here 768817Swh94709 */ 769817Swh94709 /*ARGSUSED*/ 770817Swh94709 void 771817Swh94709 nrq_overflow(struct regs *rp) 772817Swh94709 { 773817Swh94709 fm_panic("Nonresumable queue full"); 774817Swh94709 } 7753156Sgirish 7763156Sgirish /* 7773156Sgirish * This is the place for special error handling for individual errors. 7783156Sgirish */ 7793156Sgirish static void 7803156Sgirish errh_handle_attr(errh_async_flt_t *errh_fltp) 7813156Sgirish { 7823156Sgirish switch (errh_fltp->errh_er.attr & ~ERRH_MODE_MASK) { 7833156Sgirish case ERRH_ATTR_CPU: 7843156Sgirish case ERRH_ATTR_MEM: 7853156Sgirish case ERRH_ATTR_PIO: 7863156Sgirish case ERRH_ATTR_IRF: 7873156Sgirish case ERRH_ATTR_FRF: 7883156Sgirish case ERRH_ATTR_SHUT: 7893156Sgirish break; 7903156Sgirish 7913156Sgirish case ERRH_ATTR_ASR: 7923156Sgirish errh_handle_asr(errh_fltp); 7933156Sgirish break; 7943156Sgirish 7953156Sgirish case ERRH_ATTR_ASI: 7963156Sgirish case ERRH_ATTR_PREG: 7973156Sgirish case ERRH_ATTR_RQF: 7983156Sgirish break; 7993156Sgirish 8003156Sgirish default: 8013156Sgirish break; 8023156Sgirish } 8033156Sgirish } 8043156Sgirish 8053156Sgirish /* 8063156Sgirish * Handle ASR bit set in ATTR 8073156Sgirish */ 8083156Sgirish static void 8093156Sgirish errh_handle_asr(errh_async_flt_t *errh_fltp) 8103156Sgirish { 8113156Sgirish uint64_t current_tick; 8123156Sgirish 8133156Sgirish switch (errh_fltp->errh_er.reg) { 8143156Sgirish case ASR_REG_VALID | ASR_REG_TICK: 8153156Sgirish /* 8163156Sgirish * For Tick Compare Register error, it only happens when 8173156Sgirish * the register is being read or compared with the %tick 8183156Sgirish * register. Since we lost the contents of the register, 8193156Sgirish * we set the %tick_compr in the future. An interrupt will 8203156Sgirish * happen when %tick matches the value field of %tick_compr. 8213156Sgirish */ 8223156Sgirish current_tick = (uint64_t)gettick(); 8233156Sgirish tickcmpr_set(current_tick); 8243156Sgirish /* Do not panic */ 8253156Sgirish errh_fltp->cmn_asyncflt.flt_panic = 0; 8263156Sgirish break; 8273156Sgirish 8283156Sgirish default: 8293156Sgirish break; 8303156Sgirish } 8313156Sgirish } 832