blob: bae9351e9473872214d6bd9870776a2c48680909 [file] [log] [blame]
Borislav Petkovb70ef012009-06-25 19:32:38 +02001#include <linux/module.h>
2#include "edac_mce_amd.h"
Doug Thompsonb52401c2009-05-06 17:57:20 +02003
Borislav Petkov549d0422009-07-24 13:51:42 +02004static bool report_gart_errors;
Borislav Petkovb69b29d2009-07-27 16:21:14 +02005static void (*nb_bus_decoder)(int node_id, struct err_regs *regs);
Borislav Petkov549d0422009-07-24 13:51:42 +02006
7void amd_report_gart_errors(bool v)
8{
9 report_gart_errors = v;
10}
11EXPORT_SYMBOL_GPL(amd_report_gart_errors);
12
Borislav Petkovb69b29d2009-07-27 16:21:14 +020013void amd_register_ecc_decoder(void (*f)(int, struct err_regs *))
Borislav Petkov549d0422009-07-24 13:51:42 +020014{
15 nb_bus_decoder = f;
16}
17EXPORT_SYMBOL_GPL(amd_register_ecc_decoder);
18
Borislav Petkovb69b29d2009-07-27 16:21:14 +020019void amd_unregister_ecc_decoder(void (*f)(int, struct err_regs *))
Borislav Petkov549d0422009-07-24 13:51:42 +020020{
21 if (nb_bus_decoder) {
22 WARN_ON(nb_bus_decoder != f);
23
24 nb_bus_decoder = NULL;
25 }
26}
27EXPORT_SYMBOL_GPL(amd_unregister_ecc_decoder);
28
Doug Thompsonb52401c2009-05-06 17:57:20 +020029/*
30 * string representation for the different MCA reported error types, see F3x48
31 * or MSR0000_0411.
32 */
33const char *tt_msgs[] = { /* transaction type */
34 "instruction",
35 "data",
36 "generic",
37 "reserved"
38};
Borislav Petkovb70ef012009-06-25 19:32:38 +020039EXPORT_SYMBOL_GPL(tt_msgs);
Doug Thompsonb52401c2009-05-06 17:57:20 +020040
41const char *ll_msgs[] = { /* cache level */
42 "L0",
43 "L1",
44 "L2",
45 "L3/generic"
46};
Borislav Petkovb70ef012009-06-25 19:32:38 +020047EXPORT_SYMBOL_GPL(ll_msgs);
Doug Thompsonb52401c2009-05-06 17:57:20 +020048
49const char *rrrr_msgs[] = {
50 "generic",
51 "generic read",
52 "generic write",
53 "data read",
54 "data write",
55 "inst fetch",
56 "prefetch",
57 "evict",
58 "snoop",
59 "reserved RRRR= 9",
60 "reserved RRRR= 10",
61 "reserved RRRR= 11",
62 "reserved RRRR= 12",
63 "reserved RRRR= 13",
64 "reserved RRRR= 14",
65 "reserved RRRR= 15"
66};
Borislav Petkovb70ef012009-06-25 19:32:38 +020067EXPORT_SYMBOL_GPL(rrrr_msgs);
Doug Thompsonb52401c2009-05-06 17:57:20 +020068
69const char *pp_msgs[] = { /* participating processor */
70 "local node originated (SRC)",
71 "local node responded to request (RES)",
72 "local node observed as 3rd party (OBS)",
73 "generic"
74};
Borislav Petkovb70ef012009-06-25 19:32:38 +020075EXPORT_SYMBOL_GPL(pp_msgs);
Doug Thompsonb52401c2009-05-06 17:57:20 +020076
77const char *to_msgs[] = {
78 "no timeout",
79 "timed out"
80};
Borislav Petkovb70ef012009-06-25 19:32:38 +020081EXPORT_SYMBOL_GPL(to_msgs);
Doug Thompsonb52401c2009-05-06 17:57:20 +020082
83const char *ii_msgs[] = { /* memory or i/o */
84 "mem access",
85 "reserved",
86 "i/o access",
87 "generic"
88};
Borislav Petkovb70ef012009-06-25 19:32:38 +020089EXPORT_SYMBOL_GPL(ii_msgs);
Doug Thompsonb52401c2009-05-06 17:57:20 +020090
Borislav Petkov1c43f2e2009-07-23 15:47:51 +020091/*
92 * Map the 4 or 5 (family-specific) bits of Extended Error code to the
93 * string table.
94 */
95const char *ext_msgs[] = {
96 "K8 ECC error", /* 0_0000b */
97 "CRC error on link", /* 0_0001b */
98 "Sync error packets on link", /* 0_0010b */
99 "Master Abort during link operation", /* 0_0011b */
100 "Target Abort during link operation", /* 0_0100b */
101 "Invalid GART PTE entry during table walk", /* 0_0101b */
102 "Unsupported atomic RMW command received", /* 0_0110b */
103 "WDT error: NB transaction timeout", /* 0_0111b */
104 "ECC/ChipKill ECC error", /* 0_1000b */
105 "SVM DEV Error", /* 0_1001b */
106 "Link Data error", /* 0_1010b */
107 "Link/L3/Probe Filter Protocol error", /* 0_1011b */
108 "NB Internal Arrays Parity error", /* 0_1100b */
109 "DRAM Address/Control Parity error", /* 0_1101b */
110 "Link Transmission error", /* 0_1110b */
111 "GART/DEV Table Walk Data error" /* 0_1111b */
112 "Res 0x100 error", /* 1_0000b */
113 "Res 0x101 error", /* 1_0001b */
114 "Res 0x102 error", /* 1_0010b */
115 "Res 0x103 error", /* 1_0011b */
116 "Res 0x104 error", /* 1_0100b */
117 "Res 0x105 error", /* 1_0101b */
118 "Res 0x106 error", /* 1_0110b */
119 "Res 0x107 error", /* 1_0111b */
120 "Res 0x108 error", /* 1_1000b */
121 "Res 0x109 error", /* 1_1001b */
122 "Res 0x10A error", /* 1_1010b */
123 "Res 0x10B error", /* 1_1011b */
124 "ECC error in L3 Cache Data", /* 1_1100b */
125 "L3 Cache Tag error", /* 1_1101b */
126 "L3 Cache LRU Parity error", /* 1_1110b */
127 "Probe Filter error" /* 1_1111b */
Doug Thompsonb52401c2009-05-06 17:57:20 +0200128};
Borislav Petkovb70ef012009-06-25 19:32:38 +0200129EXPORT_SYMBOL_GPL(ext_msgs);
Borislav Petkov549d0422009-07-24 13:51:42 +0200130
Borislav Petkov51966242009-07-28 13:50:43 +0200131static void amd_decode_dc_mce(u64 mc0_status)
132{
133 u32 ec = mc0_status & 0xffff;
134 u32 xec = (mc0_status >> 16) & 0xf;
135
Borislav Petkovf4347552010-05-15 13:51:57 +0200136 pr_emerg("Data Cache Error");
Borislav Petkov51966242009-07-28 13:50:43 +0200137
138 if (xec == 1 && TLB_ERROR(ec))
139 pr_cont(": %s TLB multimatch.\n", LL_MSG(ec));
140 else if (xec == 0) {
141 if (mc0_status & (1ULL << 40))
142 pr_cont(" during Data Scrub.\n");
143 else if (TLB_ERROR(ec))
144 pr_cont(": %s TLB parity error.\n", LL_MSG(ec));
145 else if (MEM_ERROR(ec)) {
146 u8 ll = ec & 0x3;
147 u8 tt = (ec >> 2) & 0x3;
148 u8 rrrr = (ec >> 4) & 0xf;
149
150 /* see F10h BKDG (31116), Table 92. */
151 if (ll == 0x1) {
152 if (tt != 0x1)
153 goto wrong_dc_mce;
154
155 pr_cont(": Data/Tag %s error.\n", RRRR_MSG(ec));
156
157 } else if (ll == 0x2 && rrrr == 0x3)
158 pr_cont(" during L1 linefill from L2.\n");
159 else
160 goto wrong_dc_mce;
161 } else if (BUS_ERROR(ec) && boot_cpu_data.x86 == 0xf)
162 pr_cont(" during system linefill.\n");
163 else
164 goto wrong_dc_mce;
165 } else
166 goto wrong_dc_mce;
167
168 return;
169
170wrong_dc_mce:
171 pr_warning("Corrupted DC MCE info?\n");
172}
173
Borislav Petkovab5535e2009-07-28 14:06:26 +0200174static void amd_decode_ic_mce(u64 mc1_status)
175{
176 u32 ec = mc1_status & 0xffff;
177 u32 xec = (mc1_status >> 16) & 0xf;
178
Borislav Petkovf4347552010-05-15 13:51:57 +0200179 pr_emerg("Instruction Cache Error");
Borislav Petkovab5535e2009-07-28 14:06:26 +0200180
181 if (xec == 1 && TLB_ERROR(ec))
182 pr_cont(": %s TLB multimatch.\n", LL_MSG(ec));
183 else if (xec == 0) {
184 if (TLB_ERROR(ec))
185 pr_cont(": %s TLB Parity error.\n", LL_MSG(ec));
186 else if (BUS_ERROR(ec)) {
187 if (boot_cpu_data.x86 == 0xf &&
188 (mc1_status & (1ULL << 58)))
189 pr_cont(" during system linefill.\n");
190 else
191 pr_cont(" during attempted NB data read.\n");
192 } else if (MEM_ERROR(ec)) {
193 u8 ll = ec & 0x3;
194 u8 rrrr = (ec >> 4) & 0xf;
195
196 if (ll == 0x2)
197 pr_cont(" during a linefill from L2.\n");
198 else if (ll == 0x1) {
199
200 switch (rrrr) {
201 case 0x5:
202 pr_cont(": Parity error during "
203 "data load.\n");
204 break;
205
206 case 0x7:
207 pr_cont(": Copyback Parity/Victim"
208 " error.\n");
209 break;
210
211 case 0x8:
212 pr_cont(": Tag Snoop error.\n");
213 break;
214
215 default:
216 goto wrong_ic_mce;
217 break;
218 }
219 }
220 } else
221 goto wrong_ic_mce;
222 } else
223 goto wrong_ic_mce;
224
225 return;
226
227wrong_ic_mce:
228 pr_warning("Corrupted IC MCE info?\n");
229}
230
Borislav Petkov56cad2d2009-07-28 14:14:24 +0200231static void amd_decode_bu_mce(u64 mc2_status)
232{
233 u32 ec = mc2_status & 0xffff;
234 u32 xec = (mc2_status >> 16) & 0xf;
235
Borislav Petkovf4347552010-05-15 13:51:57 +0200236 pr_emerg("Bus Unit Error");
Borislav Petkov56cad2d2009-07-28 14:14:24 +0200237
238 if (xec == 0x1)
239 pr_cont(" in the write data buffers.\n");
240 else if (xec == 0x3)
241 pr_cont(" in the victim data buffers.\n");
242 else if (xec == 0x2 && MEM_ERROR(ec))
243 pr_cont(": %s error in the L2 cache tags.\n", RRRR_MSG(ec));
244 else if (xec == 0x0) {
245 if (TLB_ERROR(ec))
246 pr_cont(": %s error in a Page Descriptor Cache or "
247 "Guest TLB.\n", TT_MSG(ec));
248 else if (BUS_ERROR(ec))
249 pr_cont(": %s/ECC error in data read from NB: %s.\n",
250 RRRR_MSG(ec), PP_MSG(ec));
251 else if (MEM_ERROR(ec)) {
252 u8 rrrr = (ec >> 4) & 0xf;
253
254 if (rrrr >= 0x7)
255 pr_cont(": %s error during data copyback.\n",
256 RRRR_MSG(ec));
257 else if (rrrr <= 0x1)
258 pr_cont(": %s parity/ECC error during data "
259 "access from L2.\n", RRRR_MSG(ec));
260 else
261 goto wrong_bu_mce;
262 } else
263 goto wrong_bu_mce;
264 } else
265 goto wrong_bu_mce;
266
267 return;
268
269wrong_bu_mce:
270 pr_warning("Corrupted BU MCE info?\n");
271}
272
Borislav Petkovf9350ef2009-07-28 14:17:30 +0200273static void amd_decode_ls_mce(u64 mc3_status)
274{
275 u32 ec = mc3_status & 0xffff;
276 u32 xec = (mc3_status >> 16) & 0xf;
277
Borislav Petkovf4347552010-05-15 13:51:57 +0200278 pr_emerg("Load Store Error");
Borislav Petkovf9350ef2009-07-28 14:17:30 +0200279
280 if (xec == 0x0) {
281 u8 rrrr = (ec >> 4) & 0xf;
282
283 if (!BUS_ERROR(ec) || (rrrr != 0x3 && rrrr != 0x4))
284 goto wrong_ls_mce;
285
286 pr_cont(" during %s.\n", RRRR_MSG(ec));
287 }
288 return;
289
290wrong_ls_mce:
291 pr_warning("Corrupted LS MCE info?\n");
292}
293
Borislav Petkov549d0422009-07-24 13:51:42 +0200294void amd_decode_nb_mce(int node_id, struct err_regs *regs, int handle_errors)
295{
Borislav Petkov549d0422009-07-24 13:51:42 +0200296 u32 ec = ERROR_CODE(regs->nbsl);
Borislav Petkov549d0422009-07-24 13:51:42 +0200297
298 if (!handle_errors)
299 return;
300
Borislav Petkov256f7272009-12-15 16:03:53 +0100301 /*
302 * GART TLB error reporting is disabled by default. Bail out early.
303 */
304 if (TLB_ERROR(ec) && !report_gart_errors)
305 return;
306
Borislav Petkovf4347552010-05-15 13:51:57 +0200307 pr_emerg("Northbridge Error, node %d", node_id);
Borislav Petkov549d0422009-07-24 13:51:42 +0200308
309 /*
310 * F10h, revD can disable ErrCpu[3:0] so check that first and also the
311 * value encoding has changed so interpret those differently
312 */
313 if ((boot_cpu_data.x86 == 0x10) &&
Borislav Petkovcec79242009-10-27 19:12:02 +0100314 (boot_cpu_data.x86_model > 7)) {
Borislav Petkov549d0422009-07-24 13:51:42 +0200315 if (regs->nbsh & K8_NBSH_ERR_CPU_VAL)
316 pr_cont(", core: %u\n", (u8)(regs->nbsh & 0xf));
317 } else {
Borislav Petkov5b89d2f2010-03-09 20:38:48 +0100318 u8 assoc_cpus = regs->nbsh & 0xf;
319
320 if (assoc_cpus > 0)
321 pr_cont(", core: %d", fls(assoc_cpus) - 1);
322
323 pr_cont("\n");
Borislav Petkov549d0422009-07-24 13:51:42 +0200324 }
325
Borislav Petkov35d824b2010-04-30 15:19:02 +0200326 pr_emerg("%s.\n", EXT_ERR_MSG(regs->nbsl));
Borislav Petkovd93cc222009-07-28 10:56:15 +0200327
328 if (BUS_ERROR(ec) && nb_bus_decoder)
329 nb_bus_decoder(node_id, regs);
330}
331EXPORT_SYMBOL_GPL(amd_decode_nb_mce);
332
Borislav Petkov53bd5fe2009-07-28 14:20:46 +0200333static void amd_decode_fr_mce(u64 mc5_status)
334{
335 /* we have only one error signature so match all fields at once. */
336 if ((mc5_status & 0xffff) == 0x0f0f)
337 pr_emerg(" FR Error: CPU Watchdog timer expire.\n");
338 else
339 pr_warning("Corrupted FR MCE info?\n");
340}
341
Borislav Petkovd93cc222009-07-28 10:56:15 +0200342static inline void amd_decode_err_code(unsigned int ec)
343{
Borislav Petkov549d0422009-07-24 13:51:42 +0200344 if (TLB_ERROR(ec)) {
Borislav Petkovf4347552010-05-15 13:51:57 +0200345 pr_emerg("Transaction: %s, Cache Level %s\n",
Borislav Petkov549d0422009-07-24 13:51:42 +0200346 TT_MSG(ec), LL_MSG(ec));
347 } else if (MEM_ERROR(ec)) {
Borislav Petkovf4347552010-05-15 13:51:57 +0200348 pr_emerg("Transaction: %s, Type: %s, Cache Level: %s",
Borislav Petkov549d0422009-07-24 13:51:42 +0200349 RRRR_MSG(ec), TT_MSG(ec), LL_MSG(ec));
350 } else if (BUS_ERROR(ec)) {
Borislav Petkovf4347552010-05-15 13:51:57 +0200351 pr_emerg("Transaction type: %s(%s), %s, Cache Level: %s, "
Borislav Petkovd93cc222009-07-28 10:56:15 +0200352 "Participating Processor: %s\n",
353 RRRR_MSG(ec), II_MSG(ec), TO_MSG(ec), LL_MSG(ec),
354 PP_MSG(ec));
355 } else
356 pr_warning("Huh? Unknown MCE error 0x%x\n", ec);
Borislav Petkov549d0422009-07-24 13:51:42 +0200357}
Borislav Petkov549d0422009-07-24 13:51:42 +0200358
Borislav Petkovfb253192009-10-07 13:20:38 +0200359static int amd_decode_mce(struct notifier_block *nb, unsigned long val,
360 void *data)
Borislav Petkov549d0422009-07-24 13:51:42 +0200361{
Borislav Petkovfb253192009-10-07 13:20:38 +0200362 struct mce *m = (struct mce *)data;
Borislav Petkov549d0422009-07-24 13:51:42 +0200363 struct err_regs regs;
Borislav Petkovb69b29d2009-07-27 16:21:14 +0200364 int node, ecc;
Borislav Petkov549d0422009-07-24 13:51:42 +0200365
Borislav Petkovd93cc222009-07-28 10:56:15 +0200366 pr_emerg("MC%d_STATUS: ", m->bank);
Borislav Petkov549d0422009-07-24 13:51:42 +0200367
Borislav Petkovd93cc222009-07-28 10:56:15 +0200368 pr_cont("%sorrected error, report: %s, MiscV: %svalid, "
Borislav Petkovb69b29d2009-07-27 16:21:14 +0200369 "CPU context corrupt: %s",
370 ((m->status & MCI_STATUS_UC) ? "Unc" : "C"),
371 ((m->status & MCI_STATUS_EN) ? "yes" : "no"),
372 ((m->status & MCI_STATUS_MISCV) ? "" : "in"),
373 ((m->status & MCI_STATUS_PCC) ? "yes" : "no"));
Borislav Petkov549d0422009-07-24 13:51:42 +0200374
Borislav Petkovb69b29d2009-07-27 16:21:14 +0200375 /* do the two bits[14:13] together */
Borislav Petkov35d824b2010-04-30 15:19:02 +0200376 ecc = (m->status >> 45) & 0x3;
Borislav Petkovb69b29d2009-07-27 16:21:14 +0200377 if (ecc)
378 pr_cont(", %sECC Error", ((ecc == 2) ? "C" : "U"));
379
380 pr_cont("\n");
381
Borislav Petkov51966242009-07-28 13:50:43 +0200382 switch (m->bank) {
383 case 0:
384 amd_decode_dc_mce(m->status);
385 break;
Borislav Petkovd93cc222009-07-28 10:56:15 +0200386
Borislav Petkovab5535e2009-07-28 14:06:26 +0200387 case 1:
388 amd_decode_ic_mce(m->status);
389 break;
390
Borislav Petkov56cad2d2009-07-28 14:14:24 +0200391 case 2:
392 amd_decode_bu_mce(m->status);
393 break;
394
Borislav Petkovf9350ef2009-07-28 14:17:30 +0200395 case 3:
396 amd_decode_ls_mce(m->status);
397 break;
398
Borislav Petkov51966242009-07-28 13:50:43 +0200399 case 4:
Borislav Petkovb69b29d2009-07-27 16:21:14 +0200400 regs.nbsl = (u32) m->status;
401 regs.nbsh = (u32)(m->status >> 32);
402 regs.nbeal = (u32) m->addr;
403 regs.nbeah = (u32)(m->addr >> 32);
Andreas Herrmann6a812692009-09-16 11:33:40 +0200404 node = amd_get_nb_id(m->extcpu);
Borislav Petkovb69b29d2009-07-27 16:21:14 +0200405
406 amd_decode_nb_mce(node, &regs, 1);
Borislav Petkov51966242009-07-28 13:50:43 +0200407 break;
408
Borislav Petkov53bd5fe2009-07-28 14:20:46 +0200409 case 5:
410 amd_decode_fr_mce(m->status);
411 break;
412
Borislav Petkov51966242009-07-28 13:50:43 +0200413 default:
414 break;
Borislav Petkovb69b29d2009-07-27 16:21:14 +0200415 }
Borislav Petkov51966242009-07-28 13:50:43 +0200416
417 amd_decode_err_code(m->status & 0xffff);
Borislav Petkovfb253192009-10-07 13:20:38 +0200418
419 return NOTIFY_STOP;
Borislav Petkov549d0422009-07-24 13:51:42 +0200420}
Ingo Molnarf436f8b2009-10-01 16:14:32 +0200421
Borislav Petkovfb253192009-10-07 13:20:38 +0200422static struct notifier_block amd_mce_dec_nb = {
423 .notifier_call = amd_decode_mce,
424};
425
Ingo Molnarf436f8b2009-10-01 16:14:32 +0200426static int __init mce_amd_init(void)
427{
428 /*
429 * We can decode MCEs for Opteron and later CPUs:
430 */
431 if ((boot_cpu_data.x86_vendor == X86_VENDOR_AMD) &&
Borislav Petkovfb253192009-10-07 13:20:38 +0200432 (boot_cpu_data.x86 >= 0xf))
433 atomic_notifier_chain_register(&x86_mce_decoder_chain, &amd_mce_dec_nb);
Ingo Molnarf436f8b2009-10-01 16:14:32 +0200434
435 return 0;
436}
437early_initcall(mce_amd_init);
Borislav Petkov0d18b2e2009-10-02 15:31:48 +0200438
439#ifdef MODULE
440static void __exit mce_amd_exit(void)
441{
Borislav Petkovfb253192009-10-07 13:20:38 +0200442 atomic_notifier_chain_unregister(&x86_mce_decoder_chain, &amd_mce_dec_nb);
Borislav Petkov0d18b2e2009-10-02 15:31:48 +0200443}
444
445MODULE_DESCRIPTION("AMD MCE decoder");
446MODULE_ALIAS("edac-mce-amd");
447MODULE_LICENSE("GPL");
448module_exit(mce_amd_exit);
449#endif