Home | History | Annotate | Line # | Download | only in dev
memerr.c revision 1.15
      1 /*	$NetBSD: memerr.c,v 1.15 2003/07/15 03:36:15 lukem 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/cdefs.h>
     48 __KERNEL_RCSID(0, "$NetBSD: memerr.c,v 1.15 2003/07/15 03:36:15 lukem Exp $");
     49 
     50 #include <sys/param.h>
     51 #include <sys/systm.h>
     52 #include <sys/device.h>
     53 
     54 #include <machine/autoconf.h>
     55 #include <machine/cpu.h>
     56 #include <machine/idprom.h>
     57 #include <machine/pte.h>
     58 
     59 #include <sun3/sun3/machdep.h>
     60 #include <sun3/dev/memerr.h>
     61 /* #include <sun3/dev/eccreg.h> - not yet */
     62 
     63 #define	ME_PRI	7	/* Interrupt level (NMI) */
     64 
     65 extern unsigned char cpu_machine_id;
     66 
     67 enum memerr_type { ME_PAR = 0, ME_ECC = 1 };
     68 
     69 struct memerr_softc {
     70 	struct device sc_dev;
     71 	struct memerr *sc_reg;
     72 	enum memerr_type sc_type;
     73 	char *sc_typename;	/* "Parity" or "ECC" */
     74 	char *sc_csrbits;	/* how to print csr bits */
     75 	/* XXX: counters? */
     76 };
     77 
     78 static int  memerr_match __P((struct device *, struct cfdata *, void *));
     79 static void memerr_attach __P((struct device *, struct device *, void *));
     80 static int  memerr_interrupt __P((void *));
     81 static void memerr_correctable __P((struct memerr_softc *));
     82 
     83 CFATTACH_DECL(memerr, sizeof(struct memerr_softc),
     84     memerr_match, memerr_attach, NULL, NULL);
     85 
     86 int memerr_attached;
     87 
     88 static int
     89 memerr_match(parent, cf, args)
     90 	struct device *parent;
     91 	struct cfdata *cf;
     92 	void *args;
     93 {
     94 	struct confargs *ca = args;
     95 
     96 	/* This driver only supports one instance. */
     97 	if (memerr_attached)
     98 		return (0);
     99 
    100 	/* Make sure there is something there... */
    101 	if (bus_peek(ca->ca_bustype, ca->ca_paddr, 1) == -1)
    102 		return (0);
    103 
    104 	/* Default interrupt priority. */
    105 	if (ca->ca_intpri == -1)
    106 		ca->ca_intpri = ME_PRI;
    107 
    108 	return (1);
    109 }
    110 
    111 static void
    112 memerr_attach(parent, self, args)
    113 	struct device *parent;
    114 	struct device *self;
    115 	void *args;
    116 {
    117 	struct memerr_softc *sc = (void *)self;
    118 	struct confargs *ca = args;
    119 	struct memerr *mer;
    120 
    121 	/*
    122 	 * Which type of memory subsystem do we have?
    123 	 */
    124 	switch (cpu_machine_id) {
    125 	case SUN3_MACH_160:		/* XXX: correct? */
    126 	case SUN3_MACH_260:
    127 	case SUN3X_MACH_470:
    128 		sc->sc_type = ME_ECC;
    129 		sc->sc_typename = "ECC";
    130 		sc->sc_csrbits = ME_ECC_STR;
    131 		break;
    132 
    133 	default:
    134 		sc->sc_type = ME_PAR;
    135 		sc->sc_typename = "Parity";
    136 		sc->sc_csrbits = ME_PAR_STR;
    137 		break;
    138 	}
    139 	printf(": (%s memory)\n", sc->sc_typename);
    140 
    141 	mer = bus_mapin(ca->ca_bustype, ca->ca_paddr, sizeof(*mer));
    142 	if (mer == NULL)
    143 		panic("memerr: can not map register");
    144 	sc->sc_reg = mer;
    145 
    146 	/* Install interrupt handler. */
    147 	isr_add_autovect(memerr_interrupt, sc, ca->ca_intpri);
    148 
    149 	/* Enable error interrupt (and checking). */
    150 	if (sc->sc_type == ME_PAR)
    151 		mer->me_csr = ME_CSR_IENA | ME_PAR_CHECK;
    152 	else {
    153 		/*
    154 		 * XXX:  Some day, figure out how to decode
    155 		 * correctable errors and set ME_ECC_CE_ENA
    156 		 * here so we can log them...
    157 		 */
    158 		mer->me_csr = ME_CSR_IENA; /* | ME_ECC_CE_ENA */
    159 	}
    160 	memerr_attached = 1;
    161 }
    162 
    163 /*****************************************************************
    164  * Functions for ECC memory
    165  *****************************************************************/
    166 
    167 static int
    168 memerr_interrupt(arg)
    169 	void *arg;
    170 {
    171 	struct memerr_softc *sc = arg;
    172 	volatile struct memerr *me = sc->sc_reg;
    173 	u_char csr, ctx;
    174 	u_int pa, va;
    175 	int pte;
    176 	char bits[64];
    177 
    178 	csr = me->me_csr;
    179 	if ((csr & ME_CSR_IPEND) == 0)
    180 		return (0);
    181 
    182 	va = me->me_vaddr;
    183  	ctx = (va >> 28) & 0xF;
    184 	va &= 0x0FFFffff;
    185 	pte = get_pte(va);
    186 	pa = PG_PA(pte);
    187 
    188 	printf("\nMemory error on %s cycle!\n",
    189 		(ctx & 8) ? "DVMA" : "CPU");
    190 	printf(" ctx=%d, vaddr=0x%x, paddr=0x%x\n",
    191 		   (ctx & 7), va, pa);
    192 	printf(" csr=%s\n", bitmask_snprintf(csr, sc->sc_csrbits,
    193 	    bits, sizeof(bits)));
    194 
    195 	/*
    196 	 * If we have parity-checked memory, there is
    197 	 * not much to be done.  Any error is fatal.
    198 	 */
    199 	if (sc->sc_type == ME_PAR) {
    200 		if (csr & ME_PAR_EMASK) {
    201 			/* Parity errors are fatal. */
    202 			goto die;
    203 		}
    204 		/* The IPEND bit was set, but no error bits. */
    205 		goto noerror;
    206 	}
    207 
    208 	/*
    209 	 * We have ECC memory.  More complicated...
    210 	 */
    211 	if (csr & (ME_ECC_WBTMO | ME_ECC_WBERR)) {
    212 		printf(" write-back failed, pte=0x%x\n", pte);
    213 		goto die;
    214 	}
    215 	if (csr & ME_ECC_UE) {
    216 		printf(" uncorrectable ECC error\n");
    217 		goto die;
    218 	}
    219 	if (csr & ME_ECC_CE) {
    220 		/* Just log this and continue. */
    221 		memerr_correctable(sc);
    222 		goto recover;
    223 	}
    224 	/* The IPEND bit was set, but no error bits. */
    225 	goto noerror;
    226 
    227 die:
    228 	panic("all bets are off...");
    229 
    230 noerror:
    231 	printf("memerr: no error bits set?\n");
    232 
    233 recover:
    234 	/* Clear the error by writing the address register. */
    235 	me->me_vaddr = 0;
    236 	return (1);
    237 }
    238 
    239 /*
    240  * Announce (and log) a correctable ECC error.
    241  * Need to look at the ECC syndrome register on
    242  * the memory board that caused the error...
    243  */
    244 void
    245 memerr_correctable(sc)
    246 	struct memerr_softc *sc;
    247 {
    248 	/* XXX: Not yet... */
    249 }
    250