xref: /illumos-gate/usr/src/uts/sun4u/os/ecc.c (revision db874c57ae335a07060499f1492b0d0e2593e26c)
17c478bd9Sstevel@tonic-gate /*
27c478bd9Sstevel@tonic-gate  * CDDL HEADER START
37c478bd9Sstevel@tonic-gate  *
47c478bd9Sstevel@tonic-gate  * The contents of this file are subject to the terms of the
57c478bd9Sstevel@tonic-gate  * Common Development and Distribution License, Version 1.0 only
67c478bd9Sstevel@tonic-gate  * (the "License").  You may not use this file except in compliance
77c478bd9Sstevel@tonic-gate  * with the License.
87c478bd9Sstevel@tonic-gate  *
97c478bd9Sstevel@tonic-gate  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
107c478bd9Sstevel@tonic-gate  * or http://www.opensolaris.org/os/licensing.
117c478bd9Sstevel@tonic-gate  * See the License for the specific language governing permissions
127c478bd9Sstevel@tonic-gate  * and limitations under the License.
137c478bd9Sstevel@tonic-gate  *
147c478bd9Sstevel@tonic-gate  * When distributing Covered Code, include this CDDL HEADER in each
157c478bd9Sstevel@tonic-gate  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
167c478bd9Sstevel@tonic-gate  * If applicable, add the following below this CDDL HEADER, with the
177c478bd9Sstevel@tonic-gate  * fields enclosed by brackets "[]" replaced with your own identifying
187c478bd9Sstevel@tonic-gate  * information: Portions Copyright [yyyy] [name of copyright owner]
197c478bd9Sstevel@tonic-gate  *
207c478bd9Sstevel@tonic-gate  * CDDL HEADER END
217c478bd9Sstevel@tonic-gate  */
227c478bd9Sstevel@tonic-gate /*
237c478bd9Sstevel@tonic-gate  * Copyright 2005 Sun Microsystems, Inc.  All rights reserved.
247c478bd9Sstevel@tonic-gate  * Use is subject to license terms.
257c478bd9Sstevel@tonic-gate  */
267c478bd9Sstevel@tonic-gate 
277c478bd9Sstevel@tonic-gate #pragma ident	"%Z%%M%	%I%	%E% SMI"
287c478bd9Sstevel@tonic-gate 
297c478bd9Sstevel@tonic-gate #include <sys/types.h>
307c478bd9Sstevel@tonic-gate #include <sys/conf.h>
317c478bd9Sstevel@tonic-gate #include <sys/ddi.h>
327c478bd9Sstevel@tonic-gate #include <sys/systm.h>
337c478bd9Sstevel@tonic-gate #include <sys/sysmacros.h>
347c478bd9Sstevel@tonic-gate #include <sys/param.h>
357c478bd9Sstevel@tonic-gate #include <sys/mutex.h>
367c478bd9Sstevel@tonic-gate #include <sys/kmem.h>
377c478bd9Sstevel@tonic-gate #include <sys/machparam.h>
387c478bd9Sstevel@tonic-gate #include <sys/machsystm.h>
397c478bd9Sstevel@tonic-gate #include <sys/machthread.h>
407c478bd9Sstevel@tonic-gate #include <sys/cpu.h>
417c478bd9Sstevel@tonic-gate #include <sys/cpuvar.h>
427c478bd9Sstevel@tonic-gate #include <vm/page.h>
437c478bd9Sstevel@tonic-gate #include <vm/hat.h>
447c478bd9Sstevel@tonic-gate #include <vm/seg.h>
457c478bd9Sstevel@tonic-gate #include <vm/seg_kmem.h>
467c478bd9Sstevel@tonic-gate #include <sys/vmsystm.h>
477c478bd9Sstevel@tonic-gate #include <sys/vmem.h>
487c478bd9Sstevel@tonic-gate #include <sys/mman.h>
497c478bd9Sstevel@tonic-gate #include <sys/cmn_err.h>
507c478bd9Sstevel@tonic-gate #include <sys/time.h>
517c478bd9Sstevel@tonic-gate #include <sys/async.h>
527c478bd9Sstevel@tonic-gate #include <sys/spl.h>
537c478bd9Sstevel@tonic-gate #include <sys/trap.h>
547c478bd9Sstevel@tonic-gate #include <sys/machtrap.h>
557c478bd9Sstevel@tonic-gate #include <sys/promif.h>
567c478bd9Sstevel@tonic-gate #include <sys/prom_plat.h>
577c478bd9Sstevel@tonic-gate #include <sys/debug.h>
587c478bd9Sstevel@tonic-gate #include <sys/x_call.h>
597c478bd9Sstevel@tonic-gate #include <sys/membar.h>
607c478bd9Sstevel@tonic-gate #include <sys/ivintr.h>
617c478bd9Sstevel@tonic-gate #include <sys/cred.h>
627c478bd9Sstevel@tonic-gate #include <sys/cpu_module.h>
637c478bd9Sstevel@tonic-gate #include <sys/ontrap.h>
647c478bd9Sstevel@tonic-gate #include <sys/sdt.h>
657c478bd9Sstevel@tonic-gate #include <sys/errorq.h>
667c478bd9Sstevel@tonic-gate 
677c478bd9Sstevel@tonic-gate #define	MAX_CE_FLTS	10
687c478bd9Sstevel@tonic-gate #define	MAX_ASYNC_FLTS	6
697c478bd9Sstevel@tonic-gate 
707c478bd9Sstevel@tonic-gate errorq_t *ue_queue;			/* queue of uncorrectable errors */
717c478bd9Sstevel@tonic-gate errorq_t *ce_queue;			/* queue of correctable errors */
727c478bd9Sstevel@tonic-gate 
737c478bd9Sstevel@tonic-gate /*
747c478bd9Sstevel@tonic-gate  * ce_verbose_memory - covers CEs in DIMMs
757c478bd9Sstevel@tonic-gate  * ce_verbose_other - covers "others" (ecache, IO, etc.)
767c478bd9Sstevel@tonic-gate  *
777c478bd9Sstevel@tonic-gate  * If the value is 0, nothing is logged.
787c478bd9Sstevel@tonic-gate  * If the value is 1, the error is logged to the log file, but not console.
797c478bd9Sstevel@tonic-gate  * If the value is 2, the error is logged to the log file and console.
807c478bd9Sstevel@tonic-gate  */
817c478bd9Sstevel@tonic-gate int	ce_verbose_memory = 1;
827c478bd9Sstevel@tonic-gate int	ce_verbose_other = 1;
837c478bd9Sstevel@tonic-gate 
847c478bd9Sstevel@tonic-gate int	ce_show_data = 0;
857c478bd9Sstevel@tonic-gate int	ce_debug = 0;
867c478bd9Sstevel@tonic-gate int	ue_debug = 0;
877c478bd9Sstevel@tonic-gate int	reset_debug = 0;
887c478bd9Sstevel@tonic-gate 
897c478bd9Sstevel@tonic-gate /*
907c478bd9Sstevel@tonic-gate  * Tunables for controlling the handling of asynchronous faults (AFTs). Setting
917c478bd9Sstevel@tonic-gate  * these to non-default values on a non-DEBUG kernel is NOT supported.
927c478bd9Sstevel@tonic-gate  */
937c478bd9Sstevel@tonic-gate int	aft_verbose = 0;	/* log AFT messages > 1 to log only */
947c478bd9Sstevel@tonic-gate int	aft_panic = 0;		/* panic (not reboot) on fatal usermode AFLT */
957c478bd9Sstevel@tonic-gate int	aft_testfatal = 0;	/* force all AFTs to panic immediately */
967c478bd9Sstevel@tonic-gate 
977c478bd9Sstevel@tonic-gate /*
987c478bd9Sstevel@tonic-gate  * Panic_* variables specific to the AFT code.  These are used to record
997c478bd9Sstevel@tonic-gate  * information that the platform-specific code will need once we panic.
1007c478bd9Sstevel@tonic-gate  */
1017c478bd9Sstevel@tonic-gate struct async_flt panic_aflt;
1027c478bd9Sstevel@tonic-gate 
1037c478bd9Sstevel@tonic-gate /*
1047c478bd9Sstevel@tonic-gate  * Defined in bus_func.c but initialised in error_init
1057c478bd9Sstevel@tonic-gate  */
1067c478bd9Sstevel@tonic-gate extern kmutex_t bfd_lock;
1077c478bd9Sstevel@tonic-gate 
1087c478bd9Sstevel@tonic-gate /*
1097c478bd9Sstevel@tonic-gate  * Common bus driver async error logging routine.  This routine can be shared
1107c478bd9Sstevel@tonic-gate  * by all sun4u CPUs (unlike cpu_async_log_err) because we are assuming that
1117c478bd9Sstevel@tonic-gate  * if an i/o bus error required a panic, the error interrupt handler will
1127c478bd9Sstevel@tonic-gate  * enqueue the error and call panic itself.
1137c478bd9Sstevel@tonic-gate  */
1147c478bd9Sstevel@tonic-gate void
1157c478bd9Sstevel@tonic-gate bus_async_log_err(struct async_flt *aflt)
1167c478bd9Sstevel@tonic-gate {
1177c478bd9Sstevel@tonic-gate 	char unum[UNUM_NAMLEN];
1187c478bd9Sstevel@tonic-gate 	int len;
1197c478bd9Sstevel@tonic-gate 
1207c478bd9Sstevel@tonic-gate 	/*
1217c478bd9Sstevel@tonic-gate 	 * Call back into the processor specific routine
1227c478bd9Sstevel@tonic-gate 	 * to check for cpu related errors that may
1237c478bd9Sstevel@tonic-gate 	 * have resulted in this error. (E.g. copyout trap)
1247c478bd9Sstevel@tonic-gate 	 */
1257c478bd9Sstevel@tonic-gate 	if (aflt->flt_in_memory)
1267c478bd9Sstevel@tonic-gate 		cpu_check_allcpus(aflt);
1277c478bd9Sstevel@tonic-gate 
1287c478bd9Sstevel@tonic-gate 	/*
1297c478bd9Sstevel@tonic-gate 	 * Note that aflt->flt_stat is not the CPU afsr.
1307c478bd9Sstevel@tonic-gate 	 */
1317c478bd9Sstevel@tonic-gate 	(void) cpu_get_mem_unum_aflt(AFLT_STAT_INVALID, aflt,
1327c478bd9Sstevel@tonic-gate 		    unum, UNUM_NAMLEN, &len);
1337c478bd9Sstevel@tonic-gate 	aflt->flt_func(aflt, unum);
1347c478bd9Sstevel@tonic-gate }
1357c478bd9Sstevel@tonic-gate 
1367c478bd9Sstevel@tonic-gate /*
1377c478bd9Sstevel@tonic-gate  * ecc_cpu_call called from bus drain functions to run cpu
1387c478bd9Sstevel@tonic-gate  * specific functions to check other cpus and get the unum.
1397c478bd9Sstevel@tonic-gate  */
1407c478bd9Sstevel@tonic-gate void
1417c478bd9Sstevel@tonic-gate ecc_cpu_call(struct async_flt *ecc, char *unum, int err_type)
1427c478bd9Sstevel@tonic-gate {
1437c478bd9Sstevel@tonic-gate 	int len;
1447c478bd9Sstevel@tonic-gate 
1457c478bd9Sstevel@tonic-gate 	/*
1467c478bd9Sstevel@tonic-gate 	 * Call back into the processor
1477c478bd9Sstevel@tonic-gate 	 * specific routine to check for cpu related errors
1487c478bd9Sstevel@tonic-gate 	 * that may have resulted in this error.
1497c478bd9Sstevel@tonic-gate 	 * (E.g. copyout trap)
1507c478bd9Sstevel@tonic-gate 	 */
1517c478bd9Sstevel@tonic-gate 	if (ecc->flt_in_memory)
1527c478bd9Sstevel@tonic-gate 		cpu_check_allcpus(ecc);
1537c478bd9Sstevel@tonic-gate 
1547c478bd9Sstevel@tonic-gate 	(void) cpu_get_mem_unum(AFLT_STAT_VALID, ecc->flt_synd,
1557c478bd9Sstevel@tonic-gate 					(uint64_t)-1, ecc->flt_addr,
1567c478bd9Sstevel@tonic-gate 					ecc->flt_bus_id, ecc->flt_in_memory,
1577c478bd9Sstevel@tonic-gate 					ecc->flt_status, unum,
1587c478bd9Sstevel@tonic-gate 					UNUM_NAMLEN, &len);
1597c478bd9Sstevel@tonic-gate 
1607c478bd9Sstevel@tonic-gate 	if (err_type == ECC_IO_CE)
1617c478bd9Sstevel@tonic-gate 		cpu_ce_count_unum(ecc, len, unum);
1627c478bd9Sstevel@tonic-gate }
1637c478bd9Sstevel@tonic-gate 
1647c478bd9Sstevel@tonic-gate /*
1657c478bd9Sstevel@tonic-gate  * Handler to process a fatal error.  This routine can be called from a
1667c478bd9Sstevel@tonic-gate  * softint, called from trap()'s AST handling, or called from the panic flow.
1677c478bd9Sstevel@tonic-gate  */
1687c478bd9Sstevel@tonic-gate /*ARGSUSED*/
1697c478bd9Sstevel@tonic-gate static void
1707c478bd9Sstevel@tonic-gate ue_drain(void *ignored, struct async_flt *aflt, errorq_elem_t *eqep)
1717c478bd9Sstevel@tonic-gate {
1727c478bd9Sstevel@tonic-gate 	cpu_ue_log_err(aflt);
1737c478bd9Sstevel@tonic-gate }
1747c478bd9Sstevel@tonic-gate 
1757c478bd9Sstevel@tonic-gate /*
1767c478bd9Sstevel@tonic-gate  * Handler to process a correctable error.  This routine can be called from a
1777c478bd9Sstevel@tonic-gate  * softint.  We just call the CPU module's logging routine.
1787c478bd9Sstevel@tonic-gate  */
1797c478bd9Sstevel@tonic-gate /*ARGSUSED*/
1807c478bd9Sstevel@tonic-gate static void
1817c478bd9Sstevel@tonic-gate ce_drain(void *ignored, struct async_flt *aflt, errorq_elem_t *eqep)
1827c478bd9Sstevel@tonic-gate {
1837c478bd9Sstevel@tonic-gate 	cpu_ce_log_err(aflt, eqep);
1847c478bd9Sstevel@tonic-gate }
1857c478bd9Sstevel@tonic-gate 
1867c478bd9Sstevel@tonic-gate /*
1877c478bd9Sstevel@tonic-gate  * Scrub a non-fatal correctable ecc error.
1887c478bd9Sstevel@tonic-gate  */
1897c478bd9Sstevel@tonic-gate void
1907c478bd9Sstevel@tonic-gate ce_scrub(struct async_flt *aflt)
1917c478bd9Sstevel@tonic-gate {
1927c478bd9Sstevel@tonic-gate 	if (aflt->flt_in_memory)
1937c478bd9Sstevel@tonic-gate 		cpu_ce_scrub_mem_err(aflt, B_FALSE);
1947c478bd9Sstevel@tonic-gate }
1957c478bd9Sstevel@tonic-gate 
1967c478bd9Sstevel@tonic-gate /*
1977c478bd9Sstevel@tonic-gate  * Allocate error queue sizes based on max_ncpus.  max_ncpus is set just
1987c478bd9Sstevel@tonic-gate  * after ncpunode has been determined.  ncpus is set in start_other_cpus
1997c478bd9Sstevel@tonic-gate  * which is called after error_init() but may change dynamically.
2007c478bd9Sstevel@tonic-gate  */
2017c478bd9Sstevel@tonic-gate void
2027c478bd9Sstevel@tonic-gate error_init(void)
2037c478bd9Sstevel@tonic-gate {
2047c478bd9Sstevel@tonic-gate 	char tmp_name[MAXSYSNAME];
205fa9e4066Sahrens 	pnode_t node;
2067c478bd9Sstevel@tonic-gate 	size_t size = cpu_aflt_size();
2077c478bd9Sstevel@tonic-gate 
2087c478bd9Sstevel@tonic-gate 	/*
2097c478bd9Sstevel@tonic-gate 	 * Initialize the correctable and uncorrectable error queues.
2107c478bd9Sstevel@tonic-gate 	 */
2117c478bd9Sstevel@tonic-gate 	ue_queue = errorq_create("ue_queue", (errorq_func_t)ue_drain, NULL,
2127c478bd9Sstevel@tonic-gate 	    MAX_ASYNC_FLTS * (max_ncpus + 1), size, PIL_2, ERRORQ_VITAL);
2137c478bd9Sstevel@tonic-gate 
2147c478bd9Sstevel@tonic-gate 	ce_queue = errorq_create("ce_queue", (errorq_func_t)ce_drain, NULL,
2157c478bd9Sstevel@tonic-gate 	    MAX_CE_FLTS * (max_ncpus + 1), size, PIL_1, 0);
2167c478bd9Sstevel@tonic-gate 
2177c478bd9Sstevel@tonic-gate 	if (ue_queue == NULL || ce_queue == NULL)
2187c478bd9Sstevel@tonic-gate 		panic("failed to create required system error queue");
2197c478bd9Sstevel@tonic-gate 
2207c478bd9Sstevel@tonic-gate 	/*
2217c478bd9Sstevel@tonic-gate 	 * Initialize the busfunc list mutex.  This must be a PIL_15 spin lock
2227c478bd9Sstevel@tonic-gate 	 * because we will need to acquire it from cpu_async_error().
2237c478bd9Sstevel@tonic-gate 	 */
2247c478bd9Sstevel@tonic-gate 	mutex_init(&bfd_lock, NULL, MUTEX_SPIN, (void *)PIL_15);
2257c478bd9Sstevel@tonic-gate 
2267c478bd9Sstevel@tonic-gate 	node = prom_rootnode();
2277c478bd9Sstevel@tonic-gate 	if ((node == OBP_NONODE) || (node == OBP_BADNODE)) {
2287c478bd9Sstevel@tonic-gate 		cmn_err(CE_CONT, "error_init: node 0x%x\n", (uint_t)node);
2297c478bd9Sstevel@tonic-gate 		return;
2307c478bd9Sstevel@tonic-gate 	}
2317c478bd9Sstevel@tonic-gate 
2327c478bd9Sstevel@tonic-gate 	if (((size = prom_getproplen(node, "reset-reason")) != -1) &&
2337c478bd9Sstevel@tonic-gate 	    (size <= MAXSYSNAME) &&
2347c478bd9Sstevel@tonic-gate 	    (prom_getprop(node, "reset-reason", tmp_name) != -1)) {
2357c478bd9Sstevel@tonic-gate 		if (reset_debug) {
2367c478bd9Sstevel@tonic-gate 			cmn_err(CE_CONT, "System booting after %s\n", tmp_name);
2377c478bd9Sstevel@tonic-gate 		} else if (strncmp(tmp_name, "FATAL", 5) == 0) {
2387c478bd9Sstevel@tonic-gate 			cmn_err(CE_CONT,
2397c478bd9Sstevel@tonic-gate 			    "System booting after fatal error %s\n", tmp_name);
2407c478bd9Sstevel@tonic-gate 		}
2417c478bd9Sstevel@tonic-gate 	}
2427c478bd9Sstevel@tonic-gate 
2437c478bd9Sstevel@tonic-gate 	if (&cpu_error_init) {
2447c478bd9Sstevel@tonic-gate 		cpu_error_init((MAX_ASYNC_FLTS + MAX_CE_FLTS) *
2457c478bd9Sstevel@tonic-gate 		    (max_ncpus + 1));
2467c478bd9Sstevel@tonic-gate 	}
2477c478bd9Sstevel@tonic-gate }
2487c478bd9Sstevel@tonic-gate 
2497c478bd9Sstevel@tonic-gate /*
250*db874c57Selowe  * Flags for ecc_page_zero DTrace probe since ecc_page_zero() is called
251*db874c57Selowe  * as a softint handler.
2527c478bd9Sstevel@tonic-gate  */
2537c478bd9Sstevel@tonic-gate #define	PAGE_ZERO_SUCCESS	0
2547c478bd9Sstevel@tonic-gate #define	PAGE_ZERO_FAIL_NOLOCK	1
2557c478bd9Sstevel@tonic-gate #define	PAGE_ZERO_FAIL_ONTRAP	2
2567c478bd9Sstevel@tonic-gate 
2577c478bd9Sstevel@tonic-gate void
2587c478bd9Sstevel@tonic-gate ecc_page_zero(void *arg)
2597c478bd9Sstevel@tonic-gate {
2607c478bd9Sstevel@tonic-gate 	uint64_t pa = (uint64_t)arg;
2617c478bd9Sstevel@tonic-gate 	int ret, success_flag;
262*db874c57Selowe 	page_t *pp = page_numtopp_nolock(mmu_btop(pa));
2637c478bd9Sstevel@tonic-gate 
264*db874c57Selowe 	if (page_retire_check(pa, NULL) != 0)
2657c478bd9Sstevel@tonic-gate 		return;
2667c478bd9Sstevel@tonic-gate 
2677c478bd9Sstevel@tonic-gate 	/*
2687c478bd9Sstevel@tonic-gate 	 * Must hold a lock on the page before calling pagezero()
2697c478bd9Sstevel@tonic-gate 	 *
2707c478bd9Sstevel@tonic-gate 	 * This will only fail if someone has or wants an exclusive lock on
2717c478bd9Sstevel@tonic-gate 	 * the page.  Since it's a retired page, this shouldn't happen.
2727c478bd9Sstevel@tonic-gate 	 */
2737c478bd9Sstevel@tonic-gate 	ret = page_lock(pp, SE_SHARED, (kmutex_t *)NULL, P_NO_RECLAIM);
2747c478bd9Sstevel@tonic-gate 
2757c478bd9Sstevel@tonic-gate 	if (ret > 0) {
2767c478bd9Sstevel@tonic-gate 		on_trap_data_t otd;
2777c478bd9Sstevel@tonic-gate 
2787c478bd9Sstevel@tonic-gate 		/*
2797c478bd9Sstevel@tonic-gate 		 * Protect pagezero() from async faults
2807c478bd9Sstevel@tonic-gate 		 */
2817c478bd9Sstevel@tonic-gate 		if (!on_trap(&otd, OT_DATA_EC)) {
2827c478bd9Sstevel@tonic-gate 			pagezero(pp, 0, PAGESIZE);
2837c478bd9Sstevel@tonic-gate 			success_flag = PAGE_ZERO_SUCCESS;
2847c478bd9Sstevel@tonic-gate 		} else {
2857c478bd9Sstevel@tonic-gate 			success_flag = PAGE_ZERO_FAIL_ONTRAP;
2867c478bd9Sstevel@tonic-gate 		}
2877c478bd9Sstevel@tonic-gate 		no_trap();
2887c478bd9Sstevel@tonic-gate 		page_unlock(pp);
2897c478bd9Sstevel@tonic-gate 	} else {
2907c478bd9Sstevel@tonic-gate 		success_flag = PAGE_ZERO_FAIL_NOLOCK;
2917c478bd9Sstevel@tonic-gate 	}
2927c478bd9Sstevel@tonic-gate 	DTRACE_PROBE2(page_zero_result, int, success_flag, uint64_t, pa);
2937c478bd9Sstevel@tonic-gate }
294