Home | History | Annotate | Line # | Download | only in dev
memerr.c revision 1.10
      1 /*	$NetBSD: memerr.c,v 1.10 1998/02/05 04:56:43 gwr 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. All advertising materials mentioning features or use of this software
     25  *    must display the following acknowledgement:
     26  *	This product includes software developed by the University of
     27  *	California, Berkeley and its contributors.
     28  * 4. Neither the name of the University nor the names of its contributors
     29  *    may be used to endorse or promote products derived from this software
     30  *    without specific prior written permission.
     31  *
     32  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     33  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     34  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     35  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     36  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     37  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     38  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     39  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     40  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     41  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     42  * SUCH DAMAGE.
     43  *
     44  *	@(#)memreg.c	8.1 (Berkeley) 6/11/93
     45  */
     46 
     47 #include <sys/param.h>
     48 #include <sys/systm.h>
     49 #include <sys/device.h>
     50 
     51 #include <machine/autoconf.h>
     52 #include <machine/cpu.h>
     53 #include <machine/idprom.h>
     54 #include <machine/pte.h>
     55 
     56 #include <sun3/sun3/machdep.h>
     57 #include <sun3/dev/memerr.h>
     58 /* #include <sun3/dev/eccreg.h> - not yet */
     59 
     60 #define	ME_PRI	7	/* Interrupt level (NMI) */
     61 
     62 extern unsigned char cpu_machine_id;
     63 
     64 enum memerr_type { ME_PAR = 0, ME_ECC = 1 };
     65 
     66 struct memerr_softc {
     67 	struct device sc_dev;
     68 	struct memerr *sc_reg;
     69 	enum memerr_type sc_type;
     70 	char *sc_typename;	/* "Parity" or "ECC" */
     71 	char *sc_csrbits;	/* how to print csr bits */
     72 	/* XXX: counters? */
     73 };
     74 
     75 static int  memerr_match __P((struct device *, struct cfdata *, void *));
     76 static void memerr_attach __P((struct device *, struct device *, void *));
     77 static int  memerr_interrupt __P((void *));
     78 static void memerr_correctable __P((struct memerr_softc *));
     79 
     80 struct cfattach memerr_ca = {
     81 	sizeof(struct memerr_softc), memerr_match, memerr_attach
     82 };
     83 
     84 static int
     85 memerr_match(parent, cf, args)
     86     struct device *parent;
     87     struct cfdata *cf;
     88     void *args;
     89 {
     90 	struct confargs *ca = args;
     91 
     92 	/* This driver only supports one unit. */
     93 	if (cf->cf_unit != 0)
     94 		return (0);
     95 
     96 	/* Make sure there is something there... */
     97 	if (bus_peek(ca->ca_bustype, ca->ca_paddr, 1) == -1)
     98 		return (0);
     99 
    100 	/* Default interrupt priority. */
    101 	if (ca->ca_intpri == -1)
    102 		ca->ca_intpri = ME_PRI;
    103 
    104 	return (1);
    105 }
    106 
    107 static void
    108 memerr_attach(parent, self, args)
    109 	struct device *parent;
    110 	struct device *self;
    111 	void *args;
    112 {
    113 	struct memerr_softc *sc = (void *)self;
    114 	struct confargs *ca = args;
    115 	struct memerr *mer;
    116 
    117 	/*
    118 	 * Which type of memory subsystem do we have?
    119 	 */
    120 	switch (cpu_machine_id) {
    121 	case SUN3_MACH_160:		/* XXX: correct? */
    122 	case SUN3_MACH_260:
    123 	case SUN3X_MACH_470:
    124 		sc->sc_type = ME_ECC;
    125 		sc->sc_typename = "ECC";
    126 		sc->sc_csrbits = ME_ECC_STR;
    127 		break;
    128 
    129 	default:
    130 		sc->sc_type = ME_PAR;
    131 		sc->sc_typename = "Parity";
    132 		sc->sc_csrbits = ME_PAR_STR;
    133 		break;
    134 	}
    135 	printf(": (%s memory)\n", sc->sc_typename);
    136 
    137 	mer = bus_mapin(ca->ca_bustype, ca->ca_paddr, sizeof(*mer));
    138 	if (mer == NULL)
    139 		panic("memerr: can not map register");
    140 	sc->sc_reg = mer;
    141 
    142 	/* Install interrupt handler. */
    143 	isr_add_autovect(memerr_interrupt,
    144 		(void *)sc, ca->ca_intpri);
    145 
    146 	/* Enable error interrupt (and checking). */
    147 	if (sc->sc_type == ME_PAR)
    148 		mer->me_csr = ME_CSR_IENA | ME_PAR_CHECK;
    149 	else {
    150 		/*
    151 		 * XXX:  Some day, figure out how to decode
    152 		 * correctable errors and set ME_ECC_CE_ENA
    153 		 * here so we can log them...
    154 		 */
    155 		mer->me_csr = ME_CSR_IENA; /* | ME_ECC_CE_ENA */
    156 	}
    157 }
    158 
    159 /*****************************************************************
    160  * Functions for ECC memory
    161  *****************************************************************/
    162 
    163 static int
    164 memerr_interrupt(arg)
    165 	void *arg;
    166 {
    167 	struct memerr_softc *sc = arg;
    168 	volatile struct memerr *me = sc->sc_reg;
    169 	u_char csr, ctx;
    170 	u_int pa, va;
    171 	int pte;
    172 	char bits[64];
    173 
    174 	csr = me->me_csr;
    175 	if ((csr & ME_CSR_IPEND) == 0)
    176 		return (0);
    177 
    178 	va = me->me_vaddr;
    179  	ctx = (va >> 28) & 0xF;
    180 	va &= 0x0FFFffff;
    181 	pte = get_pte(va);
    182 	pa = PG_PA(pte);
    183 
    184 	printf("\nMemory error on %s cycle!\n",
    185 		(ctx & 8) ? "DVMA" : "CPU");
    186 	printf(" ctx=%d, vaddr=0x%x, paddr=0x%x\n",
    187 		   (ctx & 7), va, pa);
    188 	printf(" csr=%s\n", bitmask_snprintf(csr, sc->sc_csrbits,
    189 	    bits, sizeof(bits)));
    190 
    191 	/*
    192 	 * If we have parity-checked memory, there is
    193 	 * not much to be done.  Any error is fatal.
    194 	 */
    195 	if (sc->sc_type == ME_PAR) {
    196 		if (csr & ME_PAR_EMASK) {
    197 			/* Parity errors are fatal. */
    198 			goto die;
    199 		}
    200 		/* The IPEND bit was set, but no error bits. */
    201 		goto noerror;
    202 	}
    203 
    204 	/*
    205 	 * We have ECC memory.  More complicated...
    206 	 */
    207 	if (csr & (ME_ECC_WBTMO | ME_ECC_WBERR)) {
    208 		printf(" write-back failed, pte=0x%x\n", pte);
    209 		goto die;
    210 	}
    211 	if (csr & ME_ECC_UE) {
    212 		printf(" uncorrectable ECC error\n");
    213 		goto die;
    214 	}
    215 	if (csr & ME_ECC_CE) {
    216 		/* Just log this and continue. */
    217 		memerr_correctable(sc);
    218 		goto recover;
    219 	}
    220 	/* The IPEND bit was set, but no error bits. */
    221 	goto noerror;
    222 
    223 die:
    224 	panic("all bets are off...");
    225 
    226 noerror:
    227 	printf("memerr: no error bits set?\n");
    228 
    229 recover:
    230 	/* Clear the error by writing the address register. */
    231 	me->me_vaddr = 0;
    232 	return (1);
    233 }
    234 
    235 /*
    236  * Announce (and log) a correctable ECC error.
    237  * Need to look at the ECC syndrome register on
    238  * the memory board that caused the error...
    239  */
    240 void
    241 memerr_correctable(sc)
    242 	struct memerr_softc *sc;
    243 {
    244 	/* XXX: Not yet... */
    245 }
    246