Home | History | Annotate | Line # | Download | only in dev
      1 /*	$NetBSD: memerr.c,v 1.23 2024/12/20 23:52:00 tsutsui Exp $ */
      2 
      3 /*
      4  * Copyright (c) 1992, 1993
      5  *	The Regents of the University of California.  All rights reserved.
      6  *
      7  * This software was developed by the Computer Systems Engineering group
      8  * at Lawrence Berkeley Laboratory under DARPA contract BG 91-66 and
      9  * contributed to Berkeley.
     10  *
     11  * All advertising materials mentioning features or use of this software
     12  * must display the following acknowledgement:
     13  *	This product includes software developed by the University of
     14  *	California, Lawrence Berkeley Laboratory.
     15  *
     16  * Redistribution and use in source and binary forms, with or without
     17  * modification, are permitted provided that the following conditions
     18  * are met:
     19  * 1. Redistributions of source code must retain the above copyright
     20  *    notice, this list of conditions and the following disclaimer.
     21  * 2. Redistributions in binary form must reproduce the above copyright
     22  *    notice, this list of conditions and the following disclaimer in the
     23  *    documentation and/or other materials provided with the distribution.
     24  * 3. Neither the name of the University nor the names of its contributors
     25  *    may be used to endorse or promote products derived from this software
     26  *    without specific prior written permission.
     27  *
     28  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     29  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     30  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     31  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     32  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     33  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     34  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     35  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     36  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     37  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     38  * SUCH DAMAGE.
     39  *
     40  *	@(#)memreg.c	8.1 (Berkeley) 6/11/93
     41  */
     42 
     43 #include <sys/cdefs.h>
     44 __KERNEL_RCSID(0, "$NetBSD: memerr.c,v 1.23 2024/12/20 23:52:00 tsutsui Exp $");
     45 
     46 #include <sys/param.h>
     47 #include <sys/systm.h>
     48 #include <sys/device.h>
     49 
     50 #include <machine/autoconf.h>
     51 #include <machine/cpu.h>
     52 #include <machine/idprom.h>
     53 #include <machine/pte.h>
     54 
     55 #include <sun3/sun3/machdep.h>
     56 #include <sun3/dev/memerr.h>
     57 /* #include <sun3/dev/eccreg.h> - not yet */
     58 
     59 #define	ME_PRI	7	/* Interrupt level (NMI) */
     60 
     61 enum memerr_type { ME_PAR = 0, ME_ECC = 1 };
     62 
     63 struct memerr_softc {
     64 	device_t sc_dev;
     65 	struct memerr *sc_reg;
     66 	enum memerr_type sc_type;
     67 	const char *sc_typename;	/* "Parity" or "ECC" */
     68 	const char *sc_csrbits;		/* how to print csr bits */
     69 	/* XXX: counters? */
     70 };
     71 
     72 static int  memerr_match(device_t, cfdata_t, void *);
     73 static void memerr_attach(device_t, device_t, void *);
     74 static int  memerr_interrupt(void *);
     75 static void memerr_correctable(struct memerr_softc *);
     76 
     77 CFATTACH_DECL_NEW(memerr, sizeof(struct memerr_softc),
     78     memerr_match, memerr_attach, NULL, NULL);
     79 
     80 static int memerr_attached;
     81 
     82 static int
     83 memerr_match(device_t parent, cfdata_t cf, void *args)
     84 {
     85 	struct confargs *ca = args;
     86 
     87 	/* This driver only supports one instance. */
     88 	if (memerr_attached)
     89 		return 0;
     90 
     91 	/* Make sure there is something there... */
     92 	if (bus_peek(ca->ca_bustype, ca->ca_paddr, 1) == -1)
     93 		return 0;
     94 
     95 	/* Default interrupt priority. */
     96 	if (ca->ca_intpri == -1)
     97 		ca->ca_intpri = ME_PRI;
     98 
     99 	return 1;
    100 }
    101 
    102 static void
    103 memerr_attach(device_t parent, device_t self, void *args)
    104 {
    105 	struct memerr_softc *sc = device_private(self);
    106 	struct confargs *ca = args;
    107 	struct memerr *mer;
    108 
    109 	sc->sc_dev = self;
    110 
    111 	/*
    112 	 * Which type of memory subsystem do we have?
    113 	 */
    114 	switch (cpu_machine_id) {
    115 	case ID_SUN3_160:		/* XXX: correct? */
    116 	case ID_SUN3_260:
    117 	case ID_SUN3X_470:
    118 		sc->sc_type = ME_ECC;
    119 		sc->sc_typename = "ECC";
    120 		sc->sc_csrbits = ME_ECC_STR;
    121 		break;
    122 
    123 	default:
    124 		sc->sc_type = ME_PAR;
    125 		sc->sc_typename = "Parity";
    126 		sc->sc_csrbits = ME_PAR_STR;
    127 		break;
    128 	}
    129 	aprint_normal(": (%s memory)\n", sc->sc_typename);
    130 
    131 	mer = bus_mapin(ca->ca_bustype, ca->ca_paddr, sizeof(*mer));
    132 	if (mer == NULL)
    133 		panic("%s: can not map register", device_xname(self));
    134 	sc->sc_reg = mer;
    135 
    136 	/* Install interrupt handler. */
    137 	isr_add_autovect(memerr_interrupt, sc, ca->ca_intpri);
    138 
    139 	/* Enable error interrupt (and checking). */
    140 	if (sc->sc_type == ME_PAR)
    141 		mer->me_csr = ME_CSR_IENA | ME_PAR_CHECK;
    142 	else {
    143 		/*
    144 		 * XXX:  Some day, figure out how to decode
    145 		 * correctable errors and set ME_ECC_CE_ENA
    146 		 * here so we can log them...
    147 		 */
    148 		mer->me_csr = ME_CSR_IENA; /* | ME_ECC_CE_ENA */
    149 	}
    150 	memerr_attached = 1;
    151 }
    152 
    153 /*****************************************************************
    154  * Functions for ECC memory
    155  *****************************************************************/
    156 
    157 static int
    158 memerr_interrupt(void *arg)
    159 {
    160 	struct memerr_softc *sc = arg;
    161 	volatile struct memerr *me = sc->sc_reg;
    162 	uint8_t csr, ctx;
    163 	u_int pa, va;
    164 	int pte;
    165 	uint8_t bits[64];
    166 
    167 	csr = me->me_csr;
    168 	if ((csr & ME_CSR_IPEND) == 0)
    169 		return 0;
    170 
    171 	va = me->me_vaddr;
    172  	ctx = (va >> 28) & 0xF;
    173 	va &= 0x0FFFffff;
    174 	pte = get_pte(va);
    175 	pa = PG_PA(pte);
    176 
    177 	printf("\nMemory error on %s cycle!\n",
    178 	    (ctx & 8) ? "DVMA" : "CPU");
    179 	printf(" ctx=%d, vaddr=0x%x, paddr=0x%x\n",
    180 	    (ctx & 7), va, pa);
    181 	snprintb(bits, sizeof(bits), sc->sc_csrbits, csr);
    182 	printf(" csr=%s\n", bits);
    183 
    184 	/*
    185 	 * If we have parity-checked memory, there is
    186 	 * not much to be done.  Any error is fatal.
    187 	 */
    188 	if (sc->sc_type == ME_PAR) {
    189 		if (csr & ME_PAR_EMASK) {
    190 			/* Parity errors are fatal. */
    191 			goto die;
    192 		}
    193 		/* The IPEND bit was set, but no error bits. */
    194 		goto noerror;
    195 	}
    196 
    197 	/*
    198 	 * We have ECC memory.  More complicated...
    199 	 */
    200 	if (csr & (ME_ECC_WBTMO | ME_ECC_WBERR)) {
    201 		printf(" write-back failed, pte=0x%x\n", pte);
    202 		goto die;
    203 	}
    204 	if (csr & ME_ECC_UE) {
    205 		printf(" uncorrectable ECC error\n");
    206 		goto die;
    207 	}
    208 	if (csr & ME_ECC_CE) {
    209 		/* Just log this and continue. */
    210 		memerr_correctable(sc);
    211 		goto recover;
    212 	}
    213 	/* The IPEND bit was set, but no error bits. */
    214 	goto noerror;
    215 
    216 die:
    217 	panic("all bets are off...");
    218 
    219 noerror:
    220 	printf("memerr: no error bits set?\n");
    221 
    222 recover:
    223 	/* Clear the error by writing the address register. */
    224 	me->me_vaddr = 0;
    225 	return 1;
    226 }
    227 
    228 /*
    229  * Announce (and log) a correctable ECC error.
    230  * Need to look at the ECC syndrome register on
    231  * the memory board that caused the error...
    232  */
    233 void
    234 memerr_correctable(struct memerr_softc *sc)
    235 {
    236 
    237 	/* XXX: Not yet... */
    238 }
    239