6371709 Events with different AFARs should be treated as separate events

5 files changed, 139 insertions, 55 deletions

diff --git a/usr/src/cmd/fm/modules/sun4/cpumem-diagnosis/cmd_cpu.c b/usr/src/cmd/fm/modules/sun4/cpumem-diagnosis/cmd_cpu.c
index f240974776..8af910607c 100644
--- a/usr/src/cmd/fm/modules/sun4/cpumem-diagnosis/cmd_cpu.c
+++ b/usr/src/cmd/fm/modules/sun4/cpumem-diagnosis/cmd_cpu.c
@@ -1239,22 +1239,32 @@ cmd_xxcu_train_match(cmd_errcl_t mask)
 
 	return (0);
 }
-
+/*
+ * Search for the entry that matches the ena and the AFAR
+ * if we have a valid AFAR, otherwise just match the ENA
+ */
 cmd_xxcu_trw_t *
-cmd_trw_lookup(uint64_t ena)
+cmd_trw_lookup(uint64_t ena, uint8_t afar_status, uint64_t afar)
 {
 	int i;
 
-	for (i = 0; i < cmd.cmd_xxcu_ntrw; i++) {
+	if (afar_status == AFLT_STAT_VALID) {
+		for (i = 0; i < cmd.cmd_xxcu_ntrw; i++) {
+			if (cmd.cmd_xxcu_trw[i].trw_ena == ena &&
+			    cmd.cmd_xxcu_trw[i].trw_afar == afar)
+				return (&cmd.cmd_xxcu_trw[i]);
+		}
+	} else  {
+		for (i = 0; i < cmd.cmd_xxcu_ntrw; i++) {
 		if (cmd.cmd_xxcu_trw[i].trw_ena == ena)
 			return (&cmd.cmd_xxcu_trw[i]);
+		}
 	}
-
 	return (NULL);
 }
 
 cmd_xxcu_trw_t *
-cmd_trw_alloc(uint64_t ena)
+cmd_trw_alloc(uint64_t ena, uint64_t afar)
 {
 	int i;
 
@@ -1262,6 +1272,7 @@ cmd_trw_alloc(uint64_t ena)
 		cmd_xxcu_trw_t *trw = &cmd.cmd_xxcu_trw[i];
 		if (trw->trw_ena == NULL) {
 			trw->trw_ena = ena;
+			trw->trw_afar = afar;
 			return (trw);
 		}
 	}
diff --git a/usr/src/cmd/fm/modules/sun4/cpumem-diagnosis/cmd_cpu.h b/usr/src/cmd/fm/modules/sun4/cpumem-diagnosis/cmd_cpu.h
index c9ff6ac050..5c88d6cc4c 100644
--- a/usr/src/cmd/fm/modules/sun4/cpumem-diagnosis/cmd_cpu.h
+++ b/usr/src/cmd/fm/modules/sun4/cpumem-diagnosis/cmd_cpu.h
@@ -205,6 +205,7 @@ extern void cmd_cpu_uec_set_allmatch(fmd_hdl_t *, cmd_cpu_t *);
 
 typedef struct cmd_xxcu_trw {
 	uint64_t trw_ena;	/* the ENA for this group of ereports */
+	uint64_t trw_afar;	/* the AFAR for this group of ereports */
 	cmd_errcl_t trw_mask;	/* ereports seen thus far with this ENA */
 	uint16_t trw_cpuid;	/* CPU to which this watcher belongs */
 	uint8_t	 trw_ref;	/* number of ereports with this ENA */
@@ -212,8 +213,8 @@ typedef struct cmd_xxcu_trw {
 	uint32_t trw_pad;
 } cmd_xxcu_trw_t;
 
-extern cmd_xxcu_trw_t *cmd_trw_lookup(uint64_t);
-extern cmd_xxcu_trw_t *cmd_trw_alloc(uint64_t);
+extern cmd_xxcu_trw_t *cmd_trw_lookup(uint64_t, uint8_t, uint64_t);
+extern cmd_xxcu_trw_t *cmd_trw_alloc(uint64_t, uint64_t);
 extern void cmd_trw_restore(fmd_hdl_t *);
 extern void cmd_trw_write(fmd_hdl_t *);
 extern void cmd_trw_ref(fmd_hdl_t *, cmd_xxcu_trw_t *, cmd_errcl_t);
@@ -663,6 +664,9 @@ extern int cmd_cpu_synd_check(uint16_t);
 extern int cmd_cpu_synd_check(uint32_t);
 #endif /* sun4u */
 
+extern int cmd_afar_valid(fmd_hdl_t *hdl, nvlist_t *nvl, cmd_errcl_t,
+    uint64_t *afar);
+
 #ifdef __cplusplus
 }
 #endif
diff --git a/usr/src/cmd/fm/modules/sun4/cpumem-diagnosis/cmd_cpuerr.c b/usr/src/cmd/fm/modules/sun4/cpumem-diagnosis/cmd_cpuerr.c
index 65effd3def..4e1891898e 100644
--- a/usr/src/cmd/fm/modules/sun4/cpumem-diagnosis/cmd_cpuerr.c
+++ b/usr/src/cmd/fm/modules/sun4/cpumem-diagnosis/cmd_cpuerr.c
@@ -352,9 +352,17 @@ cmd_xxcu_resolve(fmd_hdl_t *hdl, cmd_xr_t *xr, fmd_event_t *ep,
 {
 	cmd_xxcu_trw_t *trw;
 	cmd_errcl_t cause;
+	uint64_t afar;
 
-	if ((trw = cmd_trw_lookup(xr->xr_ena)) == NULL) {
-		hdlr(hdl, xr, ep);
+
+	afar = NULL;
+
+	if (xr->xr_afar_status == AFLT_STAT_VALID)
+		afar = xr->xr_afar;
+
+	if ((trw = cmd_trw_lookup(xr->xr_ena,
+		xr->xr_afar_status, afar)) == NULL) {
+		fmd_hdl_debug(hdl, "cmd_trw_lookup: Not found\n");
 		return;
 	}
 
@@ -414,7 +422,9 @@ cmd_xxcu_initial(fmd_hdl_t *hdl, fmd_event_t *ep, nvlist_t *nvl,
 	cmd_cpu_t *cpu;
 	cmd_xr_t *xr;
 	uint64_t ena;
+	uint64_t afar;
 	uint8_t level = clcode & CMD_ERRCL_LEVEL_EXTRACT;
+	uint8_t	afar_status;
 
 	clcode &= CMD_ERRCL_LEVEL_MASK; /* keep level bits out of train masks */
 
@@ -428,11 +438,21 @@ cmd_xxcu_initial(fmd_hdl_t *hdl, fmd_event_t *ep, nvlist_t *nvl,
 
 	(void) nvlist_lookup_uint64(nvl, FM_EREPORT_ENA, &ena);
 
+	if (cmd_afar_valid(hdl, nvl, clcode, &afar) != 0) {
+		afar_status = AFLT_STAT_INVALID;
+		afar = NULL;
+	} else {
+		afar_status = AFLT_STAT_VALID;
+	}
+
 	fmd_hdl_debug(hdl, "scheduling %s (%llx) for redelivery\n",
 	    class, clcode);
+	fmd_hdl_debug(hdl, "looking up ena %llx,afar %llx with\n", ena, afar);
+
+	fmd_hdl_debug(hdl, "afar status of %02x\n", afar_status);
 
-	if ((trw = cmd_trw_lookup(ena)) == NULL) {
-		if ((trw = cmd_trw_alloc(ena)) == NULL) {
+	if ((trw = cmd_trw_lookup(ena, afar_status, afar)) == NULL) {
+		if ((trw = cmd_trw_alloc(ena, afar)) == NULL) {
 			fmd_hdl_debug(hdl, "failed to get new trw\n");
 			goto redeliver;
 		}
diff --git a/usr/src/cmd/fm/modules/sun4u/cpumem-diagnosis/cmd_cpu_arch.c b/usr/src/cmd/fm/modules/sun4u/cpumem-diagnosis/cmd_cpu_arch.c
index 9f07d1b878..7ae7fae7b9 100644
--- a/usr/src/cmd/fm/modules/sun4u/cpumem-diagnosis/cmd_cpu_arch.c
+++ b/usr/src/cmd/fm/modules/sun4u/cpumem-diagnosis/cmd_cpu_arch.c
@@ -2,9 +2,8 @@
  * CDDL HEADER START
  *
  * The contents of this file are subject to the terms of the
- * Common Development and Distribution License, Version 1.0 only
- * (the "License").  You may not use this file except in compliance
- * with the License.
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
  *
  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
  * or http://www.opensolaris.org/os/licensing.
@@ -20,7 +19,7 @@
  * CDDL HEADER END
  */
 /*
- * Copyright 2005 Sun Microsystems, Inc.  All rights reserved.
+ * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
  */
 
@@ -79,3 +78,20 @@ cmd_cpu_synd_check(uint16_t synd)
 	else
 		return (0);
 }
+int
+cmd_afar_valid(fmd_hdl_t *hdl, nvlist_t *nvl, cmd_errcl_t clcode,
+    uint64_t *afar)
+{
+	uint8_t afar_status;
+
+	if (nvlist_lookup_uint8(nvl,
+	    FM_EREPORT_PAYLOAD_NAME_AFAR_STATUS, &afar_status) == 0) {
+		if (afar_status == AFLT_STAT_VALID) {
+			(void) nvlist_lookup_uint64(nvl,
+			    FM_EREPORT_PAYLOAD_NAME_AFAR, afar);
+			return (0);
+		} else
+			return (-1);
+	}
+	return (-1);
+}
diff --git a/usr/src/cmd/fm/modules/sun4v/cpumem-diagnosis/cmd_cpu_arch.c b/usr/src/cmd/fm/modules/sun4v/cpumem-diagnosis/cmd_cpu_arch.c
index 6fa9378c69..1e2361c482 100644
--- a/usr/src/cmd/fm/modules/sun4v/cpumem-diagnosis/cmd_cpu_arch.c
+++ b/usr/src/cmd/fm/modules/sun4v/cpumem-diagnosis/cmd_cpu_arch.c
@@ -2,9 +2,8 @@
  * CDDL HEADER START
  *
  * The contents of this file are subject to the terms of the
- * Common Development and Distribution License, Version 1.0 only
- * (the "License").  You may not use this file except in compliance
- * with the License.
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
  *
  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
  * or http://www.opensolaris.org/os/licensing.
@@ -20,7 +19,7 @@
  * CDDL HEADER END
  */
 /*
- * Copyright 2005 Sun Microsystems, Inc.  All rights reserved.
+ * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
  */
 
@@ -45,6 +44,8 @@
 #include <sys/fm/cpu/UltraSPARC-T1.h>
 #include <sys/niagararegs.h>
 
+int cmd_afsr_check(fmd_hdl_t *,  uint64_t, cmd_errcl_t, uint8_t *);
+
 int
 cmd_xr_fill(fmd_hdl_t *hdl, nvlist_t *nvl, cmd_xr_t *xr, cmd_errcl_t clcode)
 {
@@ -60,6 +61,42 @@ cmd_xr_fill(fmd_hdl_t *hdl, nvlist_t *nvl, cmd_xr_t *xr, cmd_errcl_t clcode)
 	    &xr->xr_synd) != 0)
 		return (-1);
 
+	if (cmd_afsr_check(hdl, niagara_l2_afsr, clcode,
+	    &xr->xr_synd_status) != 0)
+		return (-1);
+
+	xr->xr_afar_status = xr->xr_synd_status;
+	return (0);
+}
+
+int
+cmd_cpu_synd_check(uint32_t synd)
+{
+	int i;
+
+	/*
+	 * Niagara L2 fetches from a memory location containing a UE
+	 * are given a poison syndrome in one or more 7 bit subsyndromes
+	 * each covering one of 4 4 byte checkwords.
+	 *
+	 * 0 is an invalid syndrome because it denotes no error, but
+	 * is associated with an ereport -- meaning there WAS an error.
+	 */
+	if (synd == 0)
+		return (-1);
+
+	for (i = 0; i < 4; i++) {
+		if (((synd >> i*NI_L2_POISON_SYND_SIZE) &
+		    NI_L2_POISON_SYND_MASK) == NI_L2_POISON_SYND_FROM_DAU)
+			return (-1);
+	}
+	return (0);
+}
+
+int
+cmd_afsr_check(fmd_hdl_t *hdl, uint64_t afsr,
+    cmd_errcl_t clcode, uint8_t *stat_val)
+{
 	/*
 	 * Set Niagara afar and synd validity.
 	 * For a given set of error registers, the payload value is valid iff
@@ -69,65 +106,61 @@ cmd_xr_fill(fmd_hdl_t *hdl, nvlist_t *nvl, cmd_xr_t *xr, cmd_errcl_t clcode)
 	switch (clcode) {
 	case CMD_ERRCL_LDAU:
 	case CMD_ERRCL_LDSU:
-		xr->xr_synd_status =
-		    ((niagara_l2_afsr & NI_L2AFSR_P02) == 0) ?
-		    AFLT_STAT_VALID : AFLT_STAT_INVALID;
+		*stat_val =
+			((afsr & NI_L2AFSR_P02) == 0) ?
+			AFLT_STAT_VALID: AFLT_STAT_INVALID;
 		break;
 	case CMD_ERRCL_LDWU:
-		xr->xr_synd_status =
-		    ((niagara_l2_afsr & NI_L2AFSR_P03) == 0) ?
-		    AFLT_STAT_VALID : AFLT_STAT_INVALID;
+		*stat_val =
+			((afsr & NI_L2AFSR_P03) == 0) ?
+			AFLT_STAT_VALID : AFLT_STAT_INVALID;
 		break;
 	case CMD_ERRCL_LDRU:
-		xr->xr_synd_status =
-		    ((niagara_l2_afsr & NI_L2AFSR_P04) == 0) ?
-		    AFLT_STAT_VALID : AFLT_STAT_INVALID;
+		*stat_val =
+			((afsr & NI_L2AFSR_P04) == 0) ?
+			AFLT_STAT_VALID : AFLT_STAT_INVALID;
 		break;
 	case CMD_ERRCL_LDAC:
 	case CMD_ERRCL_LDSC:
-		xr->xr_synd_status =
-		    ((niagara_l2_afsr & NI_L2AFSR_P07) == 0) ?
-		    AFLT_STAT_VALID : AFLT_STAT_INVALID;
+		*stat_val =
+			((afsr & NI_L2AFSR_P07) == 0) ?
+			AFLT_STAT_VALID : AFLT_STAT_INVALID;
 		break;
 	case CMD_ERRCL_LDWC:
-		xr->xr_synd_status =
-		    ((niagara_l2_afsr & NI_L2AFSR_P08) == 0) ?
-		    AFLT_STAT_VALID : AFLT_STAT_INVALID;
+		*stat_val =
+			((afsr & NI_L2AFSR_P08) == 0) ?
+			AFLT_STAT_VALID : AFLT_STAT_INVALID;
 		break;
 	case CMD_ERRCL_LDRC:
-		xr->xr_synd_status =
-		    ((niagara_l2_afsr & NI_L2AFSR_P09) == 0) ?
-		    AFLT_STAT_VALID : AFLT_STAT_INVALID;
+		*stat_val =
+			((afsr & NI_L2AFSR_P09) == 0) ?
+			AFLT_STAT_VALID : AFLT_STAT_INVALID;
 		break;
 	default:
 		fmd_hdl_debug(hdl, "Niagara unrecognized l2cache error\n");
-		xr->xr_synd_status = 0;
 		return (-1);
 	}
-	xr->xr_afar_status = xr->xr_synd_status;
 	return (0);
 }
 
+
 int
-cmd_cpu_synd_check(uint32_t synd)
+cmd_afar_valid(fmd_hdl_t *hdl, nvlist_t *nvl, cmd_errcl_t clcode,
+    uint64_t *afar)
 {
-	int i;
+	uint64_t niagara_l2_afsr = 0;
+	uint8_t stat_val;
+	if (nvlist_lookup_uint64(nvl, FM_EREPORT_PAYLOAD_NAME_L2_AFSR,
+	    &niagara_l2_afsr) != 0)
+		return (-1);
 
-	/*
-	 * Niagara L2 fetches from a memory location containing a UE
-	 * are given a poison syndrome in one or more 7 bit subsyndromes
-	 * each covering one of 4 4 byte checkwords.
-	 *
-	 * 0 is an invalid syndrome because it denotes no error, but
-	 * is associated with an ereport -- meaning there WAS an error.
-	 */
-	if (synd == 0)
+	if (cmd_afsr_check(hdl, niagara_l2_afsr, clcode, &stat_val) != 0)
 		return (-1);
 
-	for (i = 0; i < 4; i++) {
-		if (((synd >> i*NI_L2_POISON_SYND_SIZE) &
-		    NI_L2_POISON_SYND_MASK) == NI_L2_POISON_SYND_FROM_DAU)
-			return (-1);
+	if (stat_val == AFLT_STAT_VALID) {
+		if (nvlist_lookup_uint64(nvl,
+		    FM_EREPORT_PAYLOAD_NAME_L2_REAL_AFAR, afar) == 0)
+			return (0);
 	}
-	return (0);
+	return (-1);
 }