xref: /titanic_50/usr/src/cmd/fm/modules/sun4u/USII-io-diagnosis/iod_main.c (revision d485aa23b5e424dd136afdf657683389f93f72d6)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License, Version 1.0 only
6  * (the "License").  You may not use this file except in compliance
7  * with the License.
8  *
9  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10  * or http://www.opensolaris.org/os/licensing.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When distributing Covered Code, include this CDDL HEADER in each
15  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16  * If applicable, add the following below this CDDL HEADER, with the
17  * fields enclosed by brackets "[]" replaced with your own identifying
18  * information: Portions Copyright [yyyy] [name of copyright owner]
19  *
20  * CDDL HEADER END
21  */
22 /*
23  * Copyright 2005 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #pragma ident	"%Z%%M%	%I%	%E% SMI"
28 
29 #include <stdarg.h>
30 #include <sys/param.h>
31 #include <errno.h>
32 #include <string.h>
33 #include <unistd.h>
34 #include <strings.h>
35 #include <kstat.h>
36 #include <fm/fmd_api.h>
37 #include <sys/fm/protocol.h>
38 #include <sys/async.h>
39 
40 #define	IOD_STAT_BUMP(name)		iod.iod_stats->name.fmds_value.ui64++
41 
42 /*
43  * USII-io-diagnosis:
44  *
45  * This diagnosis engine consumes the Psycho based IO UEs and CEs
46  * which are generated due to DMA Reads or Writes.
47  *
48  * When a UE is received it will generate the appropriate fault
49  * (defect.ultraSPARC-II.memory.nodiag). This is because we do not plan
50  * to currently add greater diagnosis capabilities for USII systems.
51  *
52  * When a CE is received we allow the legacy in kernel SERD do it's job
53  * and we just bump a counter here. The legacy SERD algorithm will print
54  * out a message and retire pages when the SERD threshold is reached.
55  *
56  */
57 static void iod_recv(fmd_hdl_t *, fmd_event_t *, nvlist_t *, const char *);
58 
59 typedef struct iod_stat {
60 	fmd_stat_t ue;	/* # of UEs received */
61 	fmd_stat_t ce;	/* # of CEs received */
62 } iod_stat_t;
63 
64 typedef struct iod {
65 	iod_stat_t *iod_stats;	/* Module statistics */
66 } iod_t;
67 
68 static const fmd_hdl_ops_t fmd_ops = {
69 	iod_recv,	/* fmdo_recv */
70 	NULL,		/* fmdo_timeout */
71 	NULL,		/* fmdo_close */
72 	NULL,		/* fmdo_stats */
73 	NULL		/* fmdo_gc */
74 };
75 
76 static const fmd_hdl_info_t fmd_info = {
77 	"UltraSPARC-II I/O Diagnosis", IOD_VERSION, &fmd_ops, NULL
78 };
79 
80 static const iod_stat_t iod_stat = {
81 	{ "ue", FMD_TYPE_UINT64, "number of received IO UEs" },
82 	{ "ce", FMD_TYPE_UINT64, "number of received IO CEs" }
83 };
84 
85 iod_t iod;
86 
87 /*ARGSUSED*/
88 static void
iod_recv(fmd_hdl_t * hdl,fmd_event_t * ep,nvlist_t * nvl,const char * class)89 iod_recv(fmd_hdl_t *hdl, fmd_event_t *ep, nvlist_t *nvl, const char *class)
90 {
91 	fmd_case_t *cp;
92 	nvlist_t *fault;
93 	char flt[] = "defect.ultraSPARC-II.memory.nodiag";
94 
95 	if (fmd_nvl_class_match(hdl, nvl, "*ue")) {
96 		IOD_STAT_BUMP(ue);
97 		cp = fmd_case_open(hdl, NULL);
98 		fault = fmd_nvl_create_fault(hdl, flt, 100, NULL, NULL, NULL);
99 		fmd_case_add_ereport(hdl, cp, ep);
100 		fmd_case_add_suspect(hdl, cp, fault);
101 		fmd_case_solve(hdl, cp);
102 	} else if (fmd_nvl_class_match(hdl, nvl, "*ce")) {
103 		IOD_STAT_BUMP(ce);
104 	}
105 }
106 
107 int
iod_cpu_check_support(void)108 iod_cpu_check_support(void)
109 {
110 	kstat_named_t *kn;
111 	kstat_ctl_t *kc;
112 	kstat_t *ksp;
113 	int i;
114 
115 	if ((kc = kstat_open()) == NULL)
116 		return (0);
117 
118 	for (ksp = kc->kc_chain; ksp != NULL; ksp = ksp->ks_next) {
119 		if (strcmp(ksp->ks_module, "cpu_info") != 0)
120 			continue;
121 
122 		if (kstat_read(kc, ksp, NULL) == -1) {
123 			(void) kstat_close(kc);
124 			return (0);
125 		}
126 
127 		for (kn = ksp->ks_data, i = 0; i < ksp->ks_ndata; i++, kn++) {
128 			if (strcmp(kn->name, "implementation") != 0)
129 				continue;
130 
131 			if (strncmp(KSTAT_NAMED_STR_PTR(kn), "UltraSPARC-II",
132 			    sizeof ("UltraSPARC-II") - 1) == 0 &&
133 			    strncmp(KSTAT_NAMED_STR_PTR(kn), "UltraSPARC-III",
134 			    sizeof ("UltraSPARC-III") - 1) != 0) {
135 				(void) kstat_close(kc);
136 				return (1);
137 			}
138 		}
139 	}
140 
141 	(void) kstat_close(kc);
142 	return (0);
143 }
144 
145 void
_fmd_init(fmd_hdl_t * hdl)146 _fmd_init(fmd_hdl_t *hdl)
147 {
148 	if (fmd_hdl_register(hdl, FMD_API_VERSION, &fmd_info) != 0)
149 		return; /* error in configuration file or fmd_info */
150 
151 	if (!iod_cpu_check_support()) {
152 		fmd_hdl_debug(hdl, "no supported CPUs found");
153 		fmd_hdl_unregister(hdl);
154 		return;
155 	}
156 
157 	fmd_hdl_subscribe(hdl, "ereport.io.psy.ecc.drue");
158 	fmd_hdl_subscribe(hdl, "ereport.io.psy.ecc.s-drue");
159 	fmd_hdl_subscribe(hdl, "ereport.io.psy.ecc.dwue");
160 	fmd_hdl_subscribe(hdl, "ereport.io.psy.ecc.s-dwue");
161 	fmd_hdl_subscribe(hdl, "ereport.io.psy.ecc.drce");
162 	fmd_hdl_subscribe(hdl, "ereport.io.psy.ecc.s-drce");
163 	fmd_hdl_subscribe(hdl, "ereport.io.psy.ecc.dwce");
164 	fmd_hdl_subscribe(hdl, "ereport.io.psy.ecc.s-dwce");
165 
166 	bzero(&iod, sizeof (iod_t));
167 
168 	iod.iod_stats = (iod_stat_t *)fmd_stat_create(hdl, FMD_STAT_NOALLOC,
169 	    sizeof (iod_stat) / sizeof (fmd_stat_t), (fmd_stat_t *)&iod_stat);
170 }
171