xref: /openbmc/linux/arch/x86/kernel/cpu/mce/severity.c (revision be709d48)
1 /*
2  * MCE grading rules.
3  * Copyright 2008, 2009 Intel Corporation.
4  *
5  * This program is free software; you can redistribute it and/or
6  * modify it under the terms of the GNU General Public License
7  * as published by the Free Software Foundation; version 2
8  * of the License.
9  *
10  * Author: Andi Kleen
11  */
12 #include <linux/kernel.h>
13 #include <linux/seq_file.h>
14 #include <linux/init.h>
15 #include <linux/debugfs.h>
16 #include <asm/mce.h>
17 #include <linux/uaccess.h>
18 
19 #include "internal.h"
20 
21 /*
22  * Grade an mce by severity. In general the most severe ones are processed
23  * first. Since there are quite a lot of combinations test the bits in a
24  * table-driven way. The rules are simply processed in order, first
25  * match wins.
26  *
27  * Note this is only used for machine check exceptions, the corrected
28  * errors use much simpler rules. The exceptions still check for the corrected
29  * errors, but only to leave them alone for the CMCI handler (except for
30  * panic situations)
31  */
32 
33 enum context { IN_KERNEL = 1, IN_USER = 2, IN_KERNEL_RECOV = 3 };
34 enum ser { SER_REQUIRED = 1, NO_SER = 2 };
35 enum exception { EXCP_CONTEXT = 1, NO_EXCP = 2 };
36 
37 static struct severity {
38 	u64 mask;
39 	u64 result;
40 	unsigned char sev;
41 	unsigned char mcgmask;
42 	unsigned char mcgres;
43 	unsigned char ser;
44 	unsigned char context;
45 	unsigned char excp;
46 	unsigned char covered;
47 	char *msg;
48 } severities[] = {
49 #define MCESEV(s, m, c...) { .sev = MCE_ ## s ## _SEVERITY, .msg = m, ## c }
50 #define  KERNEL		.context = IN_KERNEL
51 #define  USER		.context = IN_USER
52 #define  KERNEL_RECOV	.context = IN_KERNEL_RECOV
53 #define  SER		.ser = SER_REQUIRED
54 #define  NOSER		.ser = NO_SER
55 #define  EXCP		.excp = EXCP_CONTEXT
56 #define  NOEXCP		.excp = NO_EXCP
57 #define  BITCLR(x)	.mask = x, .result = 0
58 #define  BITSET(x)	.mask = x, .result = x
59 #define  MCGMASK(x, y)	.mcgmask = x, .mcgres = y
60 #define  MASK(x, y)	.mask = x, .result = y
61 #define MCI_UC_S (MCI_STATUS_UC|MCI_STATUS_S)
62 #define MCI_UC_AR (MCI_STATUS_UC|MCI_STATUS_AR)
63 #define MCI_UC_SAR (MCI_STATUS_UC|MCI_STATUS_S|MCI_STATUS_AR)
64 #define	MCI_ADDR (MCI_STATUS_ADDRV|MCI_STATUS_MISCV)
65 
66 	MCESEV(
67 		NO, "Invalid",
68 		BITCLR(MCI_STATUS_VAL)
69 		),
70 	MCESEV(
71 		NO, "Not enabled",
72 		EXCP, BITCLR(MCI_STATUS_EN)
73 		),
74 	MCESEV(
75 		PANIC, "Processor context corrupt",
76 		BITSET(MCI_STATUS_PCC)
77 		),
78 	/* When MCIP is not set something is very confused */
79 	MCESEV(
80 		PANIC, "MCIP not set in MCA handler",
81 		EXCP, MCGMASK(MCG_STATUS_MCIP, 0)
82 		),
83 	/* Neither return not error IP -- no chance to recover -> PANIC */
84 	MCESEV(
85 		PANIC, "Neither restart nor error IP",
86 		EXCP, MCGMASK(MCG_STATUS_RIPV|MCG_STATUS_EIPV, 0)
87 		),
88 	MCESEV(
89 		PANIC, "In kernel and no restart IP",
90 		EXCP, KERNEL, MCGMASK(MCG_STATUS_RIPV, 0)
91 		),
92 	MCESEV(
93 		PANIC, "In kernel and no restart IP",
94 		EXCP, KERNEL_RECOV, MCGMASK(MCG_STATUS_RIPV, 0)
95 		),
96 	MCESEV(
97 		DEFERRED, "Deferred error",
98 		NOSER, MASK(MCI_STATUS_UC|MCI_STATUS_DEFERRED|MCI_STATUS_POISON, MCI_STATUS_DEFERRED)
99 		),
100 	MCESEV(
101 		KEEP, "Corrected error",
102 		NOSER, BITCLR(MCI_STATUS_UC)
103 		),
104 
105 	/*
106 	 * known AO MCACODs reported via MCE or CMC:
107 	 *
108 	 * SRAO could be signaled either via a machine check exception or
109 	 * CMCI with the corresponding bit S 1 or 0. So we don't need to
110 	 * check bit S for SRAO.
111 	 */
112 	MCESEV(
113 		AO, "Action optional: memory scrubbing error",
114 		SER, MASK(MCI_STATUS_OVER|MCI_UC_AR|MCACOD_SCRUBMSK, MCI_STATUS_UC|MCACOD_SCRUB)
115 		),
116 	MCESEV(
117 		AO, "Action optional: last level cache writeback error",
118 		SER, MASK(MCI_STATUS_OVER|MCI_UC_AR|MCACOD, MCI_STATUS_UC|MCACOD_L3WB)
119 		),
120 
121 	/* ignore OVER for UCNA */
122 	MCESEV(
123 		UCNA, "Uncorrected no action required",
124 		SER, MASK(MCI_UC_SAR, MCI_STATUS_UC)
125 		),
126 	MCESEV(
127 		PANIC, "Illegal combination (UCNA with AR=1)",
128 		SER,
129 		MASK(MCI_STATUS_OVER|MCI_UC_SAR, MCI_STATUS_UC|MCI_STATUS_AR)
130 		),
131 	MCESEV(
132 		KEEP, "Non signalled machine check",
133 		SER, BITCLR(MCI_STATUS_S)
134 		),
135 
136 	MCESEV(
137 		PANIC, "Action required with lost events",
138 		SER, BITSET(MCI_STATUS_OVER|MCI_UC_SAR)
139 		),
140 
141 	/* known AR MCACODs: */
142 #ifdef	CONFIG_MEMORY_FAILURE
143 	MCESEV(
144 		KEEP, "Action required but unaffected thread is continuable",
145 		SER, MASK(MCI_STATUS_OVER|MCI_UC_SAR|MCI_ADDR, MCI_UC_SAR|MCI_ADDR),
146 		MCGMASK(MCG_STATUS_RIPV|MCG_STATUS_EIPV, MCG_STATUS_RIPV)
147 		),
148 	MCESEV(
149 		AR, "Action required: data load in error recoverable area of kernel",
150 		SER, MASK(MCI_STATUS_OVER|MCI_UC_SAR|MCI_ADDR|MCACOD, MCI_UC_SAR|MCI_ADDR|MCACOD_DATA),
151 		KERNEL_RECOV
152 		),
153 	MCESEV(
154 		AR, "Action required: data load error in a user process",
155 		SER, MASK(MCI_STATUS_OVER|MCI_UC_SAR|MCI_ADDR|MCACOD, MCI_UC_SAR|MCI_ADDR|MCACOD_DATA),
156 		USER
157 		),
158 	MCESEV(
159 		AR, "Action required: instruction fetch error in a user process",
160 		SER, MASK(MCI_STATUS_OVER|MCI_UC_SAR|MCI_ADDR|MCACOD, MCI_UC_SAR|MCI_ADDR|MCACOD_INSTR),
161 		USER
162 		),
163 	MCESEV(
164 		PANIC, "Data load in unrecoverable area of kernel",
165 		SER, MASK(MCI_STATUS_OVER|MCI_UC_SAR|MCI_ADDR|MCACOD, MCI_UC_SAR|MCI_ADDR|MCACOD_DATA),
166 		KERNEL
167 		),
168 	MCESEV(
169 		PANIC, "Instruction fetch error in kernel",
170 		SER, MASK(MCI_STATUS_OVER|MCI_UC_SAR|MCI_ADDR|MCACOD, MCI_UC_SAR|MCI_ADDR|MCACOD_INSTR),
171 		KERNEL
172 		),
173 #endif
174 	MCESEV(
175 		PANIC, "Action required: unknown MCACOD",
176 		SER, MASK(MCI_STATUS_OVER|MCI_UC_SAR, MCI_UC_SAR)
177 		),
178 
179 	MCESEV(
180 		SOME, "Action optional: unknown MCACOD",
181 		SER, MASK(MCI_STATUS_OVER|MCI_UC_SAR, MCI_UC_S)
182 		),
183 	MCESEV(
184 		SOME, "Action optional with lost events",
185 		SER, MASK(MCI_STATUS_OVER|MCI_UC_SAR, MCI_STATUS_OVER|MCI_UC_S)
186 		),
187 
188 	MCESEV(
189 		PANIC, "Overflowed uncorrected",
190 		BITSET(MCI_STATUS_OVER|MCI_STATUS_UC)
191 		),
192 	MCESEV(
193 		UC, "Uncorrected",
194 		BITSET(MCI_STATUS_UC)
195 		),
196 	MCESEV(
197 		SOME, "No match",
198 		BITSET(0)
199 		)	/* always matches. keep at end */
200 };
201 
202 #define mc_recoverable(mcg) (((mcg) & (MCG_STATUS_RIPV|MCG_STATUS_EIPV)) == \
203 				(MCG_STATUS_RIPV|MCG_STATUS_EIPV))
204 
205 /*
206  * If mcgstatus indicated that ip/cs on the stack were
207  * no good, then "m->cs" will be zero and we will have
208  * to assume the worst case (IN_KERNEL) as we actually
209  * have no idea what we were executing when the machine
210  * check hit.
211  * If we do have a good "m->cs" (or a faked one in the
212  * case we were executing in VM86 mode) we can use it to
213  * distinguish an exception taken in user from from one
214  * taken in the kernel.
215  */
216 static int error_context(struct mce *m)
217 {
218 	if ((m->cs & 3) == 3)
219 		return IN_USER;
220 	if (mc_recoverable(m->mcgstatus) && ex_has_fault_handler(m->ip))
221 		return IN_KERNEL_RECOV;
222 	return IN_KERNEL;
223 }
224 
225 static int mce_severity_amd_smca(struct mce *m, enum context err_ctx)
226 {
227 	u32 addr = MSR_AMD64_SMCA_MCx_CONFIG(m->bank);
228 	u32 low, high;
229 
230 	/*
231 	 * We need to look at the following bits:
232 	 * - "succor" bit (data poisoning support), and
233 	 * - TCC bit (Task Context Corrupt)
234 	 * in MCi_STATUS to determine error severity.
235 	 */
236 	if (!mce_flags.succor)
237 		return MCE_PANIC_SEVERITY;
238 
239 	if (rdmsr_safe(addr, &low, &high))
240 		return MCE_PANIC_SEVERITY;
241 
242 	/* TCC (Task context corrupt). If set and if IN_KERNEL, panic. */
243 	if ((low & MCI_CONFIG_MCAX) &&
244 	    (m->status & MCI_STATUS_TCC) &&
245 	    (err_ctx == IN_KERNEL))
246 		return MCE_PANIC_SEVERITY;
247 
248 	 /* ...otherwise invoke hwpoison handler. */
249 	return MCE_AR_SEVERITY;
250 }
251 
252 /*
253  * See AMD Error Scope Hierarchy table in a newer BKDG. For example
254  * 49125_15h_Models_30h-3Fh_BKDG.pdf, section "RAS Features"
255  */
256 static int mce_severity_amd(struct mce *m, int tolerant, char **msg, bool is_excp)
257 {
258 	enum context ctx = error_context(m);
259 
260 	/* Processor Context Corrupt, no need to fumble too much, die! */
261 	if (m->status & MCI_STATUS_PCC)
262 		return MCE_PANIC_SEVERITY;
263 
264 	if (m->status & MCI_STATUS_UC) {
265 
266 		if (ctx == IN_KERNEL)
267 			return MCE_PANIC_SEVERITY;
268 
269 		/*
270 		 * On older systems where overflow_recov flag is not present, we
271 		 * should simply panic if an error overflow occurs. If
272 		 * overflow_recov flag is present and set, then software can try
273 		 * to at least kill process to prolong system operation.
274 		 */
275 		if (mce_flags.overflow_recov) {
276 			if (mce_flags.smca)
277 				return mce_severity_amd_smca(m, ctx);
278 
279 			/* kill current process */
280 			return MCE_AR_SEVERITY;
281 		} else {
282 			/* at least one error was not logged */
283 			if (m->status & MCI_STATUS_OVER)
284 				return MCE_PANIC_SEVERITY;
285 		}
286 
287 		/*
288 		 * For any other case, return MCE_UC_SEVERITY so that we log the
289 		 * error and exit #MC handler.
290 		 */
291 		return MCE_UC_SEVERITY;
292 	}
293 
294 	/*
295 	 * deferred error: poll handler catches these and adds to mce_ring so
296 	 * memory-failure can take recovery actions.
297 	 */
298 	if (m->status & MCI_STATUS_DEFERRED)
299 		return MCE_DEFERRED_SEVERITY;
300 
301 	/*
302 	 * corrected error: poll handler catches these and passes responsibility
303 	 * of decoding the error to EDAC
304 	 */
305 	return MCE_KEEP_SEVERITY;
306 }
307 
308 static int mce_severity_intel(struct mce *m, int tolerant, char **msg, bool is_excp)
309 {
310 	enum exception excp = (is_excp ? EXCP_CONTEXT : NO_EXCP);
311 	enum context ctx = error_context(m);
312 	struct severity *s;
313 
314 	for (s = severities;; s++) {
315 		if ((m->status & s->mask) != s->result)
316 			continue;
317 		if ((m->mcgstatus & s->mcgmask) != s->mcgres)
318 			continue;
319 		if (s->ser == SER_REQUIRED && !mca_cfg.ser)
320 			continue;
321 		if (s->ser == NO_SER && mca_cfg.ser)
322 			continue;
323 		if (s->context && ctx != s->context)
324 			continue;
325 		if (s->excp && excp != s->excp)
326 			continue;
327 		if (msg)
328 			*msg = s->msg;
329 		s->covered = 1;
330 		if (s->sev >= MCE_UC_SEVERITY && ctx == IN_KERNEL) {
331 			if (tolerant < 1)
332 				return MCE_PANIC_SEVERITY;
333 		}
334 		return s->sev;
335 	}
336 }
337 
338 /* Default to mce_severity_intel */
339 int (*mce_severity)(struct mce *m, int tolerant, char **msg, bool is_excp) =
340 		    mce_severity_intel;
341 
342 void __init mcheck_vendor_init_severity(void)
343 {
344 	if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD ||
345 	    boot_cpu_data.x86_vendor == X86_VENDOR_HYGON)
346 		mce_severity = mce_severity_amd;
347 }
348 
349 #ifdef CONFIG_DEBUG_FS
350 static void *s_start(struct seq_file *f, loff_t *pos)
351 {
352 	if (*pos >= ARRAY_SIZE(severities))
353 		return NULL;
354 	return &severities[*pos];
355 }
356 
357 static void *s_next(struct seq_file *f, void *data, loff_t *pos)
358 {
359 	if (++(*pos) >= ARRAY_SIZE(severities))
360 		return NULL;
361 	return &severities[*pos];
362 }
363 
364 static void s_stop(struct seq_file *f, void *data)
365 {
366 }
367 
368 static int s_show(struct seq_file *f, void *data)
369 {
370 	struct severity *ser = data;
371 	seq_printf(f, "%d\t%s\n", ser->covered, ser->msg);
372 	return 0;
373 }
374 
375 static const struct seq_operations severities_seq_ops = {
376 	.start	= s_start,
377 	.next	= s_next,
378 	.stop	= s_stop,
379 	.show	= s_show,
380 };
381 
382 static int severities_coverage_open(struct inode *inode, struct file *file)
383 {
384 	return seq_open(file, &severities_seq_ops);
385 }
386 
387 static ssize_t severities_coverage_write(struct file *file,
388 					 const char __user *ubuf,
389 					 size_t count, loff_t *ppos)
390 {
391 	int i;
392 	for (i = 0; i < ARRAY_SIZE(severities); i++)
393 		severities[i].covered = 0;
394 	return count;
395 }
396 
397 static const struct file_operations severities_coverage_fops = {
398 	.open		= severities_coverage_open,
399 	.release	= seq_release,
400 	.read		= seq_read,
401 	.write		= severities_coverage_write,
402 	.llseek		= seq_lseek,
403 };
404 
405 static int __init severities_debugfs_init(void)
406 {
407 	struct dentry *dmce, *fsev;
408 
409 	dmce = mce_get_debugfs_dir();
410 	if (!dmce)
411 		goto err_out;
412 
413 	fsev = debugfs_create_file("severities-coverage", 0444, dmce, NULL,
414 				   &severities_coverage_fops);
415 	if (!fsev)
416 		goto err_out;
417 
418 	return 0;
419 
420 err_out:
421 	return -ENOMEM;
422 }
423 late_initcall(severities_debugfs_init);
424 #endif /* CONFIG_DEBUG_FS */
425