Home | History | Annotate | Line # | Download | only in dev
memerr.c revision 1.7
      1 /*	$NetBSD: memerr.c,v 1.7 1996/12/17 21:10:50 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/control.h>
     53 #include <machine/cpu.h>
     54 #include <machine/idprom.h>
     55 #include <machine/obio.h>
     56 #include <machine/pte.h>
     57 
     58 #include <sun3/dev/memerr.h>
     59 /* #include <sun3/dev/eccreg.h> - not yet */
     60 
     61 #define	ME_PRI	7	/* Interrupt level (NMI) */
     62 
     63 extern unsigned char cpu_machine_id;
     64 
     65 enum memerr_type { ME_PAR = 0, ME_ECC = 1 };
     66 
     67 struct memerr_softc {
     68 	struct device sc_dev;
     69 	struct memerr *sc_reg;
     70 	enum memerr_type sc_type;
     71 	char *sc_typename;	/* "Parity" or "ECC" */
     72 	char *sc_csrbits;	/* how to print csr bits */
     73 	/* XXX: counters? */
     74 };
     75 
     76 static int  memerr_match __P((struct device *, struct cfdata *, void *));
     77 static void memerr_attach __P((struct device *, struct device *, void *));
     78 static int  memerr_interrupt __P((void *));
     79 static void memerr_correctable __P((struct memerr_softc *));
     80 
     81 struct cfattach memerr_ca = {
     82 	sizeof(struct memerr_softc), memerr_match, memerr_attach
     83 };
     84 
     85 struct cfdriver memerr_cd = {
     86 	NULL, "memerr", DV_DULL
     87 };
     88 
     89 
     90 static int
     91 memerr_match(parent, cf, args)
     92     struct device *parent;
     93     struct cfdata *cf;
     94     void *args;
     95 {
     96 	struct confargs *ca = args;
     97 
     98 	/* This driver only supports one unit. */
     99 	if (cf->cf_unit != 0)
    100 		return (0);
    101 
    102 	/* The peek returns -1 on bus error. */
    103 	if (bus_peek(ca->ca_bustype, ca->ca_paddr, 1) == -1)
    104 		return (0);
    105 
    106 	/* Default interrupt priority. */
    107 	if (ca->ca_intpri == -1)
    108 		ca->ca_intpri = ME_PRI;
    109 
    110 	return (1);
    111 }
    112 
    113 static void
    114 memerr_attach(parent, self, args)
    115 	struct device *parent;
    116 	struct device *self;
    117 	void *args;
    118 {
    119 	struct memerr_softc *sc = (void *)self;
    120 	struct confargs *ca = args;
    121 	struct memerr *mer;
    122 
    123 	/*
    124 	 * Which type of memory subsystem do we have?
    125 	 */
    126 	switch (cpu_machine_id) {
    127 	case SUN3_MACH_160:		/* XXX: correct? */
    128 	case SUN3_MACH_260:
    129 		sc->sc_type = ME_ECC;
    130 		sc->sc_typename = "ECC";
    131 		sc->sc_csrbits = ME_ECC_STR;
    132 		break;
    133 
    134 	default:
    135 		sc->sc_type = ME_PAR;
    136 		sc->sc_typename = "Parity";
    137 		sc->sc_csrbits = ME_PAR_STR;
    138 		break;
    139 	}
    140 	printf(": (%s memory)\n", sc->sc_typename);
    141 
    142 	mer = (struct memerr *)
    143 	    obio_alloc(ca->ca_paddr, sizeof(*mer));
    144 	if (mer == NULL)
    145 		panic("memerr: can not map register");
    146 	sc->sc_reg = mer;
    147 
    148 	/* Install interrupt handler. */
    149 	isr_add_autovect(memerr_interrupt,
    150 		(void *)sc, ca->ca_intpri);
    151 
    152 	/* Enable error interrupt (and checking). */
    153 	if (sc->sc_type == ME_PAR)
    154 		mer->me_csr = ME_CSR_IENA | ME_PAR_CHECK;
    155 	else {
    156 		/*
    157 		 * XXX:  Some day, figure out how to decode
    158 		 * correctable errors and set ME_ECC_CE_ENA
    159 		 * here so we can log them...
    160 		 */
    161 		mer->me_csr = ME_CSR_IENA; /* | ME_ECC_CE_ENA */
    162 	}
    163 }
    164 
    165 /*****************************************************************
    166  * Functions for ECC memory
    167  *****************************************************************/
    168 
    169 static int
    170 memerr_interrupt(arg)
    171 	void *arg;
    172 {
    173 	struct memerr_softc *sc = arg;
    174 	volatile struct memerr *me = sc->sc_reg;
    175 	u_char csr, ctx;
    176 	u_int pa, va;
    177 	int pte;
    178 	char bits[64];
    179 
    180 	csr = me->me_csr;
    181 	if ((csr & ME_CSR_IPEND) == 0)
    182 		return (0);
    183 
    184 	va = me->me_vaddr;
    185  	ctx = (va >> 28) & 0xF;
    186 	va &= 0x0FFFffff;
    187 	pte = get_pte(va);
    188 	pa = PG_PA(pte);
    189 
    190 	printf("\nMemory error on %s cycle!\n",
    191 		(ctx & 8) ? "DVMA" : "CPU");
    192 	printf(" ctx=%d, vaddr=0x%x, paddr=0x%x\n",
    193 		   (ctx & 7), va, pa);
    194 	printf(" csr=%s\n", bitmask_snprintf(csr, sc->sc_csrbits,
    195 	    bits, sizeof(bits)));
    196 
    197 	/*
    198 	 * If we have parity-checked memory, there is
    199 	 * not much to be done.  Any error is fatal.
    200 	 */
    201 	if (sc->sc_type == ME_PAR) {
    202 		if (csr & ME_PAR_EMASK) {
    203 			/* Parity errors are fatal. */
    204 			goto die;
    205 		}
    206 		/* The IPEND bit was set, but no error bits. */
    207 		goto noerror;
    208 	}
    209 
    210 	/*
    211 	 * We have ECC memory.  More complicated...
    212 	 */
    213 	if (csr & (ME_ECC_WBTMO | ME_ECC_WBERR)) {
    214 		printf(" write-back failed, pte=0x%x\n", pte);
    215 		goto die;
    216 	}
    217 	if (csr & ME_ECC_UE) {
    218 		printf(" uncorrectable ECC error\n");
    219 		goto die;
    220 	}
    221 	if (csr & ME_ECC_CE) {
    222 		/* Just log this and continue. */
    223 		memerr_correctable(sc);
    224 		goto recover;
    225 	}
    226 	/* The IPEND bit was set, but no error bits. */
    227 	goto noerror;
    228 
    229 die:
    230 	panic("all bets are off...");
    231 
    232 noerror:
    233 	printf("memerr: no error bits set?\n");
    234 
    235 recover:
    236 	/* Clear the error by writing the address register. */
    237 	me->me_vaddr = 0;
    238 	return (1);
    239 }
    240 
    241 /*
    242  * Announce (and log) a correctable ECC error.
    243  * Need to look at the ECC syndrome register on
    244  * the memory board that caused the error...
    245  */
    246 void
    247 memerr_correctable(sc)
    248 	struct memerr_softc *sc;
    249 {
    250 	/* XXX: Not yet... */
    251 }
    252