xref: /illumos-gate/usr/src/uts/sun4u/os/ecc.c (revision 4fc2445a)
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 2006 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 <sys/types.h>
30 #include <sys/conf.h>
31 #include <sys/ddi.h>
32 #include <sys/systm.h>
33 #include <sys/sysmacros.h>
34 #include <sys/param.h>
35 #include <sys/mutex.h>
36 #include <sys/kmem.h>
37 #include <sys/machparam.h>
38 #include <sys/machsystm.h>
39 #include <sys/machthread.h>
40 #include <sys/cpu.h>
41 #include <sys/cpuvar.h>
42 #include <vm/page.h>
43 #include <vm/hat.h>
44 #include <vm/seg.h>
45 #include <vm/seg_kmem.h>
46 #include <sys/vmsystm.h>
47 #include <sys/vmem.h>
48 #include <sys/mman.h>
49 #include <sys/cmn_err.h>
50 #include <sys/time.h>
51 #include <sys/async.h>
52 #include <sys/spl.h>
53 #include <sys/trap.h>
54 #include <sys/machtrap.h>
55 #include <sys/promif.h>
56 #include <sys/prom_plat.h>
57 #include <sys/debug.h>
58 #include <sys/x_call.h>
59 #include <sys/membar.h>
60 #include <sys/ivintr.h>
61 #include <sys/cred.h>
62 #include <sys/cpu_module.h>
63 #include <sys/ontrap.h>
64 #include <sys/sdt.h>
65 #include <sys/errorq.h>
66 
67 #define	MAX_CE_FLTS	10
68 #define	MAX_ASYNC_FLTS	6
69 
70 errorq_t *ue_queue;			/* queue of uncorrectable errors */
71 errorq_t *ce_queue;			/* queue of correctable errors */
72 
73 /*
74  * ce_verbose_memory - covers CEs in DIMMs
75  * ce_verbose_other - covers "others" (ecache, IO, etc.)
76  *
77  * If the value is 0, nothing is logged.
78  * If the value is 1, the error is logged to the log file, but not console.
79  * If the value is 2, the error is logged to the log file and console.
80  */
81 int	ce_verbose_memory = 1;
82 int	ce_verbose_other = 1;
83 
84 int	ce_show_data = 0;
85 int	ce_debug = 0;
86 int	ue_debug = 0;
87 int	reset_debug = 0;
88 
89 /*
90  * Tunables for controlling the handling of asynchronous faults (AFTs). Setting
91  * these to non-default values on a non-DEBUG kernel is NOT supported.
92  */
93 int	aft_verbose = 0;	/* log AFT messages > 1 to log only */
94 int	aft_panic = 0;		/* panic (not reboot) on fatal usermode AFLT */
95 int	aft_testfatal = 0;	/* force all AFTs to panic immediately */
96 
97 /*
98  * Panic_* variables specific to the AFT code.  These are used to record
99  * information that the platform-specific code will need once we panic.
100  */
101 struct async_flt panic_aflt;
102 
103 /*
104  * Defined in bus_func.c but initialised in error_init
105  */
106 extern kmutex_t bfd_lock;
107 
108 /*
109  * Common bus driver async error logging routine.  This routine can be shared
110  * by all sun4u CPUs (unlike cpu_async_log_err) because we are assuming that
111  * if an i/o bus error required a panic, the error interrupt handler will
112  * enqueue the error and call panic itself.
113  */
114 void
bus_async_log_err(struct async_flt * aflt)115 bus_async_log_err(struct async_flt *aflt)
116 {
117 	char unum[UNUM_NAMLEN];
118 	int len;
119 
120 	/*
121 	 * Call back into the processor specific routine
122 	 * to check for cpu related errors that may
123 	 * have resulted in this error. (E.g. copyout trap)
124 	 */
125 	if (aflt->flt_in_memory)
126 		cpu_check_allcpus(aflt);
127 
128 	/*
129 	 * Note that aflt->flt_stat is not the CPU afsr.
130 	 */
131 	(void) cpu_get_mem_unum_aflt(AFLT_STAT_INVALID, aflt,
132 		    unum, UNUM_NAMLEN, &len);
133 	aflt->flt_func(aflt, unum);
134 }
135 
136 /*
137  * ecc_cpu_call called from bus drain functions to run cpu
138  * specific functions to check other cpus and get the unum.
139  */
140 void
ecc_cpu_call(struct async_flt * ecc,char * unum,int err_type)141 ecc_cpu_call(struct async_flt *ecc, char *unum, int err_type)
142 {
143 	int len;
144 
145 	/*
146 	 * Call back into the processor
147 	 * specific routine to check for cpu related errors
148 	 * that may have resulted in this error.
149 	 * (E.g. copyout trap)
150 	 */
151 	if (ecc->flt_in_memory)
152 		cpu_check_allcpus(ecc);
153 
154 	(void) cpu_get_mem_unum(AFLT_STAT_VALID, ecc->flt_synd,
155 					(uint64_t)-1, ecc->flt_addr,
156 					ecc->flt_bus_id, ecc->flt_in_memory,
157 					ecc->flt_status, unum,
158 					UNUM_NAMLEN, &len);
159 
160 	if (err_type == ECC_IO_CE)
161 		cpu_ce_count_unum(ecc, len, unum);
162 }
163 
164 /*
165  * Handler to process a fatal error.  This routine can be called from a
166  * softint, called from trap()'s AST handling, or called from the panic flow.
167  */
168 /*ARGSUSED*/
169 static void
ue_drain(void * ignored,struct async_flt * aflt,errorq_elem_t * eqep)170 ue_drain(void *ignored, struct async_flt *aflt, errorq_elem_t *eqep)
171 {
172 	cpu_ue_log_err(aflt);
173 }
174 
175 /*
176  * Handler to process a correctable error.  This routine can be called from a
177  * softint.  We just call the CPU module's logging routine.
178  */
179 /*ARGSUSED*/
180 static void
ce_drain(void * ignored,struct async_flt * aflt,errorq_elem_t * eqep)181 ce_drain(void *ignored, struct async_flt *aflt, errorq_elem_t *eqep)
182 {
183 	cpu_ce_log_err(aflt, eqep);
184 }
185 
186 /*
187  * Scrub a non-fatal correctable ecc error.
188  */
189 void
ce_scrub(struct async_flt * aflt)190 ce_scrub(struct async_flt *aflt)
191 {
192 	if (aflt->flt_in_memory)
193 		cpu_ce_scrub_mem_err(aflt, B_FALSE);
194 }
195 
196 /*
197  * Allocate error queue sizes based on max_ncpus.  max_ncpus is set just
198  * after ncpunode has been determined.  ncpus is set in start_other_cpus
199  * which is called after error_init() but may change dynamically.
200  */
201 void
error_init(void)202 error_init(void)
203 {
204 	char tmp_name[MAXSYSNAME];
205 	pnode_t node;
206 	size_t size = cpu_aflt_size();
207 
208 	/*
209 	 * Initialize the correctable and uncorrectable error queues.
210 	 */
211 	ue_queue = errorq_create("ue_queue", (errorq_func_t)ue_drain, NULL,
212 	    MAX_ASYNC_FLTS * (max_ncpus + 1), size, PIL_2, ERRORQ_VITAL);
213 
214 	ce_queue = errorq_create("ce_queue", (errorq_func_t)ce_drain, NULL,
215 	    MAX_CE_FLTS * (max_ncpus + 1), size, PIL_1, 0);
216 
217 	if (ue_queue == NULL || ce_queue == NULL)
218 		panic("failed to create required system error queue");
219 
220 	/*
221 	 * Initialize the busfunc list mutex.  This must be a PIL_15 spin lock
222 	 * because we will need to acquire it from cpu_async_error().
223 	 */
224 	mutex_init(&bfd_lock, NULL, MUTEX_SPIN, (void *)PIL_15);
225 
226 	node = prom_rootnode();
227 	if ((node == OBP_NONODE) || (node == OBP_BADNODE)) {
228 		cmn_err(CE_CONT, "error_init: node 0x%x\n", (uint_t)node);
229 		return;
230 	}
231 
232 	if (((size = prom_getproplen(node, "reset-reason")) != -1) &&
233 	    (size <= MAXSYSNAME) &&
234 	    (prom_getprop(node, "reset-reason", tmp_name) != -1)) {
235 		if (reset_debug) {
236 			cmn_err(CE_CONT, "System booting after %s\n", tmp_name);
237 		} else if (strncmp(tmp_name, "FATAL", 5) == 0) {
238 			cmn_err(CE_CONT,
239 			    "System booting after fatal error %s\n", tmp_name);
240 		}
241 	}
242 
243 	if (&cpu_error_init) {
244 		cpu_error_init((MAX_ASYNC_FLTS + MAX_CE_FLTS) *
245 		    (max_ncpus + 1));
246 	}
247 }
248 
249 /*
250  * Flags for ecc_page_zero DTrace probe since ecc_page_zero() is called
251  * as a softint handler.
252  */
253 #define	PAGE_ZERO_SUCCESS	0
254 #define	PAGE_ZERO_FAIL_NOLOCK	1
255 #define	PAGE_ZERO_FAIL_ONTRAP	2
256 
257 void
ecc_page_zero(void * arg)258 ecc_page_zero(void *arg)
259 {
260 	uint64_t pa = (uint64_t)arg;
261 	int ret, success_flag;
262 	page_t *pp = page_numtopp_nolock(mmu_btop(pa));
263 
264 	if (page_retire_check(pa, NULL) != 0)
265 		return;
266 
267 	/*
268 	 * Must hold a lock on the page before calling pagezero()
269 	 *
270 	 * This will only fail if someone has or wants an exclusive lock on
271 	 * the page.  Since it's a retired page, this shouldn't happen.
272 	 */
273 	ret = page_lock_es(pp, SE_SHARED, (kmutex_t *)NULL,
274 	    P_NO_RECLAIM, SE_RETIRED);
275 
276 	if (ret > 0) {
277 		on_trap_data_t otd;
278 
279 		/*
280 		 * Protect pagezero() from async faults
281 		 */
282 		if (!on_trap(&otd, OT_DATA_EC)) {
283 			pagezero(pp, 0, PAGESIZE);
284 			success_flag = PAGE_ZERO_SUCCESS;
285 		} else {
286 			success_flag = PAGE_ZERO_FAIL_ONTRAP;
287 		}
288 		no_trap();
289 		page_unlock(pp);
290 	} else {
291 		success_flag = PAGE_ZERO_FAIL_NOLOCK;
292 	}
293 	DTRACE_PROBE2(page_zero_result, int, success_flag, uint64_t, pa);
294 }
295