blob: 0a19d2626686361e0eef19bbdcf2ef253a200060 [file] [log] [blame]
Borislav Petkovb70ef012009-06-25 19:32:38 +02001#include <linux/module.h>
Borislav Petkov888ab8e2010-08-18 15:11:35 +02002#include <linux/slab.h>
3
Borislav Petkov47ca08a2010-09-27 15:30:39 +02004#include "mce_amd.h"
Doug Thompsonb52401ce2009-05-06 17:57:20 +02005
Borislav Petkov888ab8e2010-08-18 15:11:35 +02006static struct amd_decoder_ops *fam_ops;
7
Borislav Petkov2be64bf2010-09-17 19:11:47 +02008static u8 xec_mask = 0xf;
Borislav Petkov5ce88f62010-08-31 18:28:08 +02009static u8 nb_err_cpumask = 0xf;
10
Borislav Petkov549d0422009-07-24 13:51:42 +020011static bool report_gart_errors;
Borislav Petkov7cfd4a82010-09-01 14:45:20 +020012static void (*nb_bus_decoder)(int node_id, struct mce *m, u32 nbcfg);
Borislav Petkov549d0422009-07-24 13:51:42 +020013
14void amd_report_gart_errors(bool v)
15{
16 report_gart_errors = v;
17}
18EXPORT_SYMBOL_GPL(amd_report_gart_errors);
19
Borislav Petkov7cfd4a82010-09-01 14:45:20 +020020void amd_register_ecc_decoder(void (*f)(int, struct mce *, u32))
Borislav Petkov549d0422009-07-24 13:51:42 +020021{
22 nb_bus_decoder = f;
23}
24EXPORT_SYMBOL_GPL(amd_register_ecc_decoder);
25
Borislav Petkov7cfd4a82010-09-01 14:45:20 +020026void amd_unregister_ecc_decoder(void (*f)(int, struct mce *, u32))
Borislav Petkov549d0422009-07-24 13:51:42 +020027{
28 if (nb_bus_decoder) {
29 WARN_ON(nb_bus_decoder != f);
30
31 nb_bus_decoder = NULL;
32 }
33}
34EXPORT_SYMBOL_GPL(amd_unregister_ecc_decoder);
35
Doug Thompsonb52401ce2009-05-06 17:57:20 +020036/*
37 * string representation for the different MCA reported error types, see F3x48
38 * or MSR0000_0411.
39 */
Borislav Petkov63375832010-09-06 18:13:39 +020040
41/* transaction type */
42const char *tt_msgs[] = { "INSN", "DATA", "GEN", "RESV" };
Borislav Petkovb70ef012009-06-25 19:32:38 +020043EXPORT_SYMBOL_GPL(tt_msgs);
Doug Thompsonb52401ce2009-05-06 17:57:20 +020044
Borislav Petkov63375832010-09-06 18:13:39 +020045/* cache level */
46const char *ll_msgs[] = { "RESV", "L1", "L2", "L3/GEN" };
Borislav Petkovb70ef012009-06-25 19:32:38 +020047EXPORT_SYMBOL_GPL(ll_msgs);
Doug Thompsonb52401ce2009-05-06 17:57:20 +020048
Borislav Petkov63375832010-09-06 18:13:39 +020049/* memory transaction type */
Doug Thompsonb52401ce2009-05-06 17:57:20 +020050const char *rrrr_msgs[] = {
Borislav Petkov63375832010-09-06 18:13:39 +020051 "GEN", "RD", "WR", "DRD", "DWR", "IRD", "PRF", "EV", "SNP"
Doug Thompsonb52401ce2009-05-06 17:57:20 +020052};
Borislav Petkovb70ef012009-06-25 19:32:38 +020053EXPORT_SYMBOL_GPL(rrrr_msgs);
Doug Thompsonb52401ce2009-05-06 17:57:20 +020054
Borislav Petkov63375832010-09-06 18:13:39 +020055/* participating processor */
56const char *pp_msgs[] = { "SRC", "RES", "OBS", "GEN" };
Borislav Petkovb70ef012009-06-25 19:32:38 +020057EXPORT_SYMBOL_GPL(pp_msgs);
Doug Thompsonb52401ce2009-05-06 17:57:20 +020058
Borislav Petkov63375832010-09-06 18:13:39 +020059/* request timeout */
60const char *to_msgs[] = { "no timeout", "timed out" };
Borislav Petkovb70ef012009-06-25 19:32:38 +020061EXPORT_SYMBOL_GPL(to_msgs);
Doug Thompsonb52401ce2009-05-06 17:57:20 +020062
Borislav Petkov63375832010-09-06 18:13:39 +020063/* memory or i/o */
64const char *ii_msgs[] = { "MEM", "RESV", "IO", "GEN" };
Borislav Petkovb70ef012009-06-25 19:32:38 +020065EXPORT_SYMBOL_GPL(ii_msgs);
Doug Thompsonb52401ce2009-05-06 17:57:20 +020066
Borislav Petkov5ce88f62010-08-31 18:28:08 +020067static const char *f10h_nb_mce_desc[] = {
68 "HT link data error",
69 "Protocol error (link, L3, probe filter, etc.)",
70 "Parity error in NB-internal arrays",
71 "Link Retry due to IO link transmission error",
72 "L3 ECC data cache error",
73 "ECC error in L3 cache tag",
74 "L3 LRU parity bits error",
75 "ECC Error in the Probe Filter directory"
Doug Thompsonb52401ce2009-05-06 17:57:20 +020076};
Borislav Petkov549d0422009-07-24 13:51:42 +020077
Borislav Petkov86039cd2010-11-08 15:03:35 +010078static const char * const f15h_ic_mce_desc[] = {
79 "UC during a demand linefill from L2",
80 "Parity error during data load from IC",
81 "Parity error for IC valid bit",
82 "Main tag parity error",
83 "Parity error in prediction queue",
84 "PFB data/address parity error",
85 "Parity error in the branch status reg",
86 "PFB promotion address error",
87 "Tag error during probe/victimization",
88 "Parity error for IC probe tag valid bit",
89 "PFB non-cacheable bit parity error",
90 "PFB valid bit parity error", /* xec = 0xd */
91 "patch RAM", /* xec = 010 */
92 "uop queue",
93 "insn buffer",
94 "predecode buffer",
95 "fetch address FIFO"
96};
97
Borislav Petkov70fdb492010-09-21 20:45:10 +020098static const char * const f15h_cu_mce_desc[] = {
99 "Fill ECC error on data fills", /* xec = 0x4 */
100 "Fill parity error on insn fills",
101 "Prefetcher request FIFO parity error",
102 "PRQ address parity error",
103 "PRQ data parity error",
104 "WCC Tag ECC error",
105 "WCC Data ECC error",
106 "WCB Data parity error",
107 "VB Data/ECC error",
108 "L2 Tag ECC error", /* xec = 0x10 */
109 "Hard L2 Tag ECC error",
110 "Multiple hits on L2 tag",
111 "XAB parity error",
112 "PRB address parity error"
113};
114
Borislav Petkov8259a7e2010-09-22 15:28:59 +0200115static const char * const fr_ex_mce_desc[] = {
116 "CPU Watchdog timer expire",
117 "Wakeup array dest tag",
118 "AG payload array",
119 "EX payload array",
120 "IDRF array",
121 "Retire dispatch queue",
122 "Mapper checkpoint array",
123 "Physical register file EX0 port",
124 "Physical register file EX1 port",
125 "Physical register file AG0 port",
126 "Physical register file AG1 port",
127 "Flag register file",
128 "DE correctable error could not be corrected"
129};
130
Borislav Petkov25a4f8b2010-09-17 19:22:34 +0200131static bool f12h_dc_mce(u16 ec, u8 xec)
Borislav Petkov888ab8e2010-08-18 15:11:35 +0200132{
Borislav Petkov888ab8e2010-08-18 15:11:35 +0200133 bool ret = false;
134
Borislav Petkov888ab8e2010-08-18 15:11:35 +0200135 if (MEM_ERROR(ec)) {
Borislav Petkov62452882010-09-22 16:08:37 +0200136 u8 ll = LL(ec);
Borislav Petkov888ab8e2010-08-18 15:11:35 +0200137 ret = true;
138
139 if (ll == LL_L2)
140 pr_cont("during L1 linefill from L2.\n");
141 else if (ll == LL_L1)
Borislav Petkov62452882010-09-22 16:08:37 +0200142 pr_cont("Data/Tag %s error.\n", R4_MSG(ec));
Borislav Petkov888ab8e2010-08-18 15:11:35 +0200143 else
144 ret = false;
145 }
146 return ret;
147}
148
Borislav Petkov25a4f8b2010-09-17 19:22:34 +0200149static bool f10h_dc_mce(u16 ec, u8 xec)
Borislav Petkov9be0bb12010-09-16 15:08:14 +0200150{
Borislav Petkov62452882010-09-22 16:08:37 +0200151 if (R4(ec) == R4_GEN && LL(ec) == LL_L1) {
Borislav Petkov9be0bb12010-09-16 15:08:14 +0200152 pr_cont("during data scrub.\n");
153 return true;
154 }
Borislav Petkov25a4f8b2010-09-17 19:22:34 +0200155 return f12h_dc_mce(ec, xec);
Borislav Petkov9be0bb12010-09-16 15:08:14 +0200156}
157
Borislav Petkov25a4f8b2010-09-17 19:22:34 +0200158static bool k8_dc_mce(u16 ec, u8 xec)
Borislav Petkov888ab8e2010-08-18 15:11:35 +0200159{
160 if (BUS_ERROR(ec)) {
161 pr_cont("during system linefill.\n");
162 return true;
163 }
164
Borislav Petkov25a4f8b2010-09-17 19:22:34 +0200165 return f10h_dc_mce(ec, xec);
Borislav Petkov888ab8e2010-08-18 15:11:35 +0200166}
167
Borislav Petkov25a4f8b2010-09-17 19:22:34 +0200168static bool f14h_dc_mce(u16 ec, u8 xec)
Borislav Petkov888ab8e2010-08-18 15:11:35 +0200169{
Borislav Petkov62452882010-09-22 16:08:37 +0200170 u8 r4 = R4(ec);
Borislav Petkov888ab8e2010-08-18 15:11:35 +0200171 bool ret = true;
172
173 if (MEM_ERROR(ec)) {
174
Borislav Petkov62452882010-09-22 16:08:37 +0200175 if (TT(ec) != TT_DATA || LL(ec) != LL_L1)
Borislav Petkov888ab8e2010-08-18 15:11:35 +0200176 return false;
177
178 switch (r4) {
179 case R4_DRD:
180 case R4_DWR:
181 pr_cont("Data/Tag parity error due to %s.\n",
182 (r4 == R4_DRD ? "load/hw prf" : "store"));
183 break;
184 case R4_EVICT:
185 pr_cont("Copyback parity error on a tag miss.\n");
186 break;
187 case R4_SNOOP:
188 pr_cont("Tag parity error during snoop.\n");
189 break;
190 default:
191 ret = false;
192 }
193 } else if (BUS_ERROR(ec)) {
194
Borislav Petkov62452882010-09-22 16:08:37 +0200195 if ((II(ec) != II_MEM && II(ec) != II_IO) || LL(ec) != LL_LG)
Borislav Petkov888ab8e2010-08-18 15:11:35 +0200196 return false;
197
198 pr_cont("System read data error on a ");
199
200 switch (r4) {
201 case R4_RD:
202 pr_cont("TLB reload.\n");
203 break;
204 case R4_DWR:
205 pr_cont("store.\n");
206 break;
207 case R4_DRD:
208 pr_cont("load.\n");
209 break;
210 default:
211 ret = false;
212 }
213 } else {
214 ret = false;
215 }
216
217 return ret;
218}
219
Borislav Petkov25a4f8b2010-09-17 19:22:34 +0200220static bool f15h_dc_mce(u16 ec, u8 xec)
221{
222 bool ret = true;
223
224 if (MEM_ERROR(ec)) {
225
226 switch (xec) {
227 case 0x0:
228 pr_cont("Data Array access error.\n");
229 break;
230
231 case 0x1:
232 pr_cont("UC error during a linefill from L2/NB.\n");
233 break;
234
235 case 0x2:
236 case 0x11:
237 pr_cont("STQ access error.\n");
238 break;
239
240 case 0x3:
241 pr_cont("SCB access error.\n");
242 break;
243
244 case 0x10:
245 pr_cont("Tag error.\n");
246 break;
247
248 case 0x12:
249 pr_cont("LDQ access error.\n");
250 break;
251
252 default:
253 ret = false;
254 }
255 } else if (BUS_ERROR(ec)) {
256
257 if (!xec)
258 pr_cont("during system linefill.\n");
259 else
260 pr_cont(" Internal %s condition.\n",
261 ((xec == 1) ? "livelock" : "deadlock"));
262 } else
263 ret = false;
264
265 return ret;
266}
267
Borislav Petkov7cfd4a82010-09-01 14:45:20 +0200268static void amd_decode_dc_mce(struct mce *m)
Borislav Petkov51966242009-07-28 13:50:43 +0200269{
Borislav Petkov62452882010-09-22 16:08:37 +0200270 u16 ec = EC(m->status);
271 u8 xec = XEC(m->status, xec_mask);
Borislav Petkov51966242009-07-28 13:50:43 +0200272
Borislav Petkovc9f281f2010-08-18 18:21:42 +0200273 pr_emerg(HW_ERR "Data Cache Error: ");
Borislav Petkov51966242009-07-28 13:50:43 +0200274
Borislav Petkov888ab8e2010-08-18 15:11:35 +0200275 /* TLB error signatures are the same across families */
276 if (TLB_ERROR(ec)) {
Borislav Petkov62452882010-09-22 16:08:37 +0200277 if (TT(ec) == TT_DATA) {
Borislav Petkov888ab8e2010-08-18 15:11:35 +0200278 pr_cont("%s TLB %s.\n", LL_MSG(ec),
Borislav Petkov25a4f8b2010-09-17 19:22:34 +0200279 ((xec == 2) ? "locked miss"
280 : (xec ? "multimatch" : "parity")));
Borislav Petkov888ab8e2010-08-18 15:11:35 +0200281 return;
282 }
Borislav Petkov25a4f8b2010-09-17 19:22:34 +0200283 } else if (fam_ops->dc_mce(ec, xec))
284 ;
285 else
286 pr_emerg(HW_ERR "Corrupted DC MCE info?\n");
Borislav Petkov51966242009-07-28 13:50:43 +0200287}
288
Borislav Petkov86039cd2010-11-08 15:03:35 +0100289static bool k8_ic_mce(u16 ec, u8 xec)
Borislav Petkovdd53bce2010-08-26 19:05:49 +0200290{
Borislav Petkov62452882010-09-22 16:08:37 +0200291 u8 ll = LL(ec);
Borislav Petkovdd53bce2010-08-26 19:05:49 +0200292 bool ret = true;
293
294 if (!MEM_ERROR(ec))
295 return false;
296
297 if (ll == 0x2)
298 pr_cont("during a linefill from L2.\n");
299 else if (ll == 0x1) {
Borislav Petkov62452882010-09-22 16:08:37 +0200300 switch (R4(ec)) {
Borislav Petkovdd53bce2010-08-26 19:05:49 +0200301 case R4_IRD:
302 pr_cont("Parity error during data load.\n");
303 break;
304
305 case R4_EVICT:
306 pr_cont("Copyback Parity/Victim error.\n");
307 break;
308
309 case R4_SNOOP:
310 pr_cont("Tag Snoop error.\n");
311 break;
312
313 default:
314 ret = false;
315 break;
316 }
317 } else
318 ret = false;
319
320 return ret;
321}
322
Borislav Petkov86039cd2010-11-08 15:03:35 +0100323static bool f14h_ic_mce(u16 ec, u8 xec)
Borislav Petkovdd53bce2010-08-26 19:05:49 +0200324{
Borislav Petkov62452882010-09-22 16:08:37 +0200325 u8 r4 = R4(ec);
Borislav Petkovdd53bce2010-08-26 19:05:49 +0200326 bool ret = true;
327
328 if (MEM_ERROR(ec)) {
Borislav Petkov62452882010-09-22 16:08:37 +0200329 if (TT(ec) != 0 || LL(ec) != 1)
Borislav Petkovdd53bce2010-08-26 19:05:49 +0200330 ret = false;
331
332 if (r4 == R4_IRD)
333 pr_cont("Data/tag array parity error for a tag hit.\n");
334 else if (r4 == R4_SNOOP)
335 pr_cont("Tag error during snoop/victimization.\n");
336 else
337 ret = false;
338 }
339 return ret;
340}
341
Borislav Petkov86039cd2010-11-08 15:03:35 +0100342static bool f15h_ic_mce(u16 ec, u8 xec)
343{
344 bool ret = true;
345
346 if (!MEM_ERROR(ec))
347 return false;
348
349 switch (xec) {
350 case 0x0 ... 0xa:
351 pr_cont("%s.\n", f15h_ic_mce_desc[xec]);
352 break;
353
354 case 0xd:
355 pr_cont("%s.\n", f15h_ic_mce_desc[xec-2]);
356 break;
357
358 case 0x10 ... 0x14:
359 pr_cont("Decoder %s parity error.\n", f15h_ic_mce_desc[xec-4]);
360 break;
361
362 default:
363 ret = false;
364 }
365 return ret;
366}
367
Borislav Petkov7cfd4a82010-09-01 14:45:20 +0200368static void amd_decode_ic_mce(struct mce *m)
Borislav Petkovab5535e2009-07-28 14:06:26 +0200369{
Borislav Petkov62452882010-09-22 16:08:37 +0200370 u16 ec = EC(m->status);
371 u8 xec = XEC(m->status, xec_mask);
Borislav Petkovab5535e2009-07-28 14:06:26 +0200372
Borislav Petkovdd53bce2010-08-26 19:05:49 +0200373 pr_emerg(HW_ERR "Instruction Cache Error: ");
Borislav Petkovab5535e2009-07-28 14:06:26 +0200374
Borislav Petkovdd53bce2010-08-26 19:05:49 +0200375 if (TLB_ERROR(ec))
376 pr_cont("%s TLB %s.\n", LL_MSG(ec),
377 (xec ? "multimatch" : "parity error"));
378 else if (BUS_ERROR(ec)) {
Borislav Petkov525906b2010-10-15 15:27:02 +0200379 bool k8 = (boot_cpu_data.x86 == 0xf && (m->status & BIT_64(58)));
Borislav Petkovab5535e2009-07-28 14:06:26 +0200380
Borislav Petkovdd53bce2010-08-26 19:05:49 +0200381 pr_cont("during %s.\n", (k8 ? "system linefill" : "NB data read"));
Borislav Petkov86039cd2010-11-08 15:03:35 +0100382 } else if (fam_ops->ic_mce(ec, xec))
Borislav Petkovdd53bce2010-08-26 19:05:49 +0200383 ;
384 else
385 pr_emerg(HW_ERR "Corrupted IC MCE info?\n");
Borislav Petkovab5535e2009-07-28 14:06:26 +0200386}
387
Borislav Petkov7cfd4a82010-09-01 14:45:20 +0200388static void amd_decode_bu_mce(struct mce *m)
Borislav Petkov56cad2d2009-07-28 14:14:24 +0200389{
Borislav Petkov62452882010-09-22 16:08:37 +0200390 u16 ec = EC(m->status);
391 u8 xec = XEC(m->status, xec_mask);
Borislav Petkov56cad2d2009-07-28 14:14:24 +0200392
Borislav Petkovc9f281f2010-08-18 18:21:42 +0200393 pr_emerg(HW_ERR "Bus Unit Error");
Borislav Petkov56cad2d2009-07-28 14:14:24 +0200394
395 if (xec == 0x1)
396 pr_cont(" in the write data buffers.\n");
397 else if (xec == 0x3)
398 pr_cont(" in the victim data buffers.\n");
399 else if (xec == 0x2 && MEM_ERROR(ec))
Borislav Petkov62452882010-09-22 16:08:37 +0200400 pr_cont(": %s error in the L2 cache tags.\n", R4_MSG(ec));
Borislav Petkov56cad2d2009-07-28 14:14:24 +0200401 else if (xec == 0x0) {
402 if (TLB_ERROR(ec))
403 pr_cont(": %s error in a Page Descriptor Cache or "
404 "Guest TLB.\n", TT_MSG(ec));
405 else if (BUS_ERROR(ec))
406 pr_cont(": %s/ECC error in data read from NB: %s.\n",
Borislav Petkov62452882010-09-22 16:08:37 +0200407 R4_MSG(ec), PP_MSG(ec));
Borislav Petkov56cad2d2009-07-28 14:14:24 +0200408 else if (MEM_ERROR(ec)) {
Borislav Petkov62452882010-09-22 16:08:37 +0200409 u8 r4 = R4(ec);
Borislav Petkov56cad2d2009-07-28 14:14:24 +0200410
Borislav Petkov62452882010-09-22 16:08:37 +0200411 if (r4 >= 0x7)
Borislav Petkov56cad2d2009-07-28 14:14:24 +0200412 pr_cont(": %s error during data copyback.\n",
Borislav Petkov62452882010-09-22 16:08:37 +0200413 R4_MSG(ec));
414 else if (r4 <= 0x1)
Borislav Petkov56cad2d2009-07-28 14:14:24 +0200415 pr_cont(": %s parity/ECC error during data "
Borislav Petkov62452882010-09-22 16:08:37 +0200416 "access from L2.\n", R4_MSG(ec));
Borislav Petkov56cad2d2009-07-28 14:14:24 +0200417 else
418 goto wrong_bu_mce;
419 } else
420 goto wrong_bu_mce;
421 } else
422 goto wrong_bu_mce;
423
424 return;
425
426wrong_bu_mce:
Borislav Petkovc9f281f2010-08-18 18:21:42 +0200427 pr_emerg(HW_ERR "Corrupted BU MCE info?\n");
Borislav Petkov56cad2d2009-07-28 14:14:24 +0200428}
429
Borislav Petkov70fdb492010-09-21 20:45:10 +0200430static void amd_decode_cu_mce(struct mce *m)
431{
Borislav Petkov62452882010-09-22 16:08:37 +0200432 u16 ec = EC(m->status);
433 u8 xec = XEC(m->status, xec_mask);
Borislav Petkov70fdb492010-09-21 20:45:10 +0200434
435 pr_emerg(HW_ERR "Combined Unit Error: ");
436
437 if (TLB_ERROR(ec)) {
438 if (xec == 0x0)
439 pr_cont("Data parity TLB read error.\n");
440 else if (xec == 0x1)
441 pr_cont("Poison data provided for TLB fill.\n");
442 else
443 goto wrong_cu_mce;
444 } else if (BUS_ERROR(ec)) {
445 if (xec > 2)
446 goto wrong_cu_mce;
447
448 pr_cont("Error during attempted NB data read.\n");
449 } else if (MEM_ERROR(ec)) {
450 switch (xec) {
451 case 0x4 ... 0xc:
452 pr_cont("%s.\n", f15h_cu_mce_desc[xec - 0x4]);
453 break;
454
455 case 0x10 ... 0x14:
456 pr_cont("%s.\n", f15h_cu_mce_desc[xec - 0x7]);
457 break;
458
459 default:
460 goto wrong_cu_mce;
461 }
462 }
463
464 return;
465
466wrong_cu_mce:
467 pr_emerg(HW_ERR "Corrupted CU MCE info?\n");
468}
469
Borislav Petkov7cfd4a82010-09-01 14:45:20 +0200470static void amd_decode_ls_mce(struct mce *m)
Borislav Petkovf9350ef2009-07-28 14:17:30 +0200471{
Borislav Petkov62452882010-09-22 16:08:37 +0200472 u16 ec = EC(m->status);
473 u8 xec = XEC(m->status, xec_mask);
Borislav Petkovded50622010-08-27 17:03:34 +0200474
Borislav Petkovb18434c2010-09-22 11:53:32 +0200475 if (boot_cpu_data.x86 >= 0x14) {
Borislav Petkovded50622010-08-27 17:03:34 +0200476 pr_emerg("You shouldn't be seeing an LS MCE on this cpu family,"
477 " please report on LKML.\n");
478 return;
479 }
Borislav Petkovf9350ef2009-07-28 14:17:30 +0200480
Borislav Petkovc9f281f2010-08-18 18:21:42 +0200481 pr_emerg(HW_ERR "Load Store Error");
Borislav Petkovf9350ef2009-07-28 14:17:30 +0200482
483 if (xec == 0x0) {
Borislav Petkov62452882010-09-22 16:08:37 +0200484 u8 r4 = R4(ec);
Borislav Petkovf9350ef2009-07-28 14:17:30 +0200485
Borislav Petkovded50622010-08-27 17:03:34 +0200486 if (!BUS_ERROR(ec) || (r4 != R4_DRD && r4 != R4_DWR))
Borislav Petkovf9350ef2009-07-28 14:17:30 +0200487 goto wrong_ls_mce;
488
Borislav Petkov62452882010-09-22 16:08:37 +0200489 pr_cont(" during %s.\n", R4_MSG(ec));
Borislav Petkovded50622010-08-27 17:03:34 +0200490 } else
491 goto wrong_ls_mce;
492
Borislav Petkovf9350ef2009-07-28 14:17:30 +0200493 return;
494
495wrong_ls_mce:
Borislav Petkovc9f281f2010-08-18 18:21:42 +0200496 pr_emerg(HW_ERR "Corrupted LS MCE info?\n");
Borislav Petkovf9350ef2009-07-28 14:17:30 +0200497}
498
Borislav Petkov5ce88f62010-08-31 18:28:08 +0200499static bool k8_nb_mce(u16 ec, u8 xec)
500{
501 bool ret = true;
502
503 switch (xec) {
504 case 0x1:
505 pr_cont("CRC error detected on HT link.\n");
506 break;
507
508 case 0x5:
509 pr_cont("Invalid GART PTE entry during GART table walk.\n");
510 break;
511
512 case 0x6:
513 pr_cont("Unsupported atomic RMW received from an IO link.\n");
514 break;
515
516 case 0x0:
517 case 0x8:
Borislav Petkovf0157b32010-10-05 19:07:16 +0200518 if (boot_cpu_data.x86 == 0x11)
519 return false;
520
Borislav Petkov5ce88f62010-08-31 18:28:08 +0200521 pr_cont("DRAM ECC error detected on the NB.\n");
522 break;
523
524 case 0xd:
525 pr_cont("Parity error on the DRAM addr/ctl signals.\n");
526 break;
527
528 default:
529 ret = false;
530 break;
531 }
532
533 return ret;
534}
535
536static bool f10h_nb_mce(u16 ec, u8 xec)
537{
538 bool ret = true;
539 u8 offset = 0;
540
541 if (k8_nb_mce(ec, xec))
542 return true;
543
544 switch(xec) {
545 case 0xa ... 0xc:
546 offset = 10;
547 break;
548
549 case 0xe:
550 offset = 11;
551 break;
552
553 case 0xf:
554 if (TLB_ERROR(ec))
555 pr_cont("GART Table Walk data error.\n");
556 else if (BUS_ERROR(ec))
557 pr_cont("DMA Exclusion Vector Table Walk error.\n");
558 else
559 ret = false;
560
561 goto out;
562 break;
563
Borislav Petkov05cd6672010-09-22 15:06:24 +0200564 case 0x19:
565 if (boot_cpu_data.x86 == 0x15)
566 pr_cont("Compute Unit Data Error.\n");
567 else
568 ret = false;
569
570 goto out;
571 break;
572
Borislav Petkov5ce88f62010-08-31 18:28:08 +0200573 case 0x1c ... 0x1f:
574 offset = 24;
575 break;
576
577 default:
578 ret = false;
579
580 goto out;
581 break;
582 }
583
584 pr_cont("%s.\n", f10h_nb_mce_desc[xec - offset]);
585
586out:
587 return ret;
588}
589
Borislav Petkovcb9d5ec2010-09-16 17:36:12 +0200590static bool nb_noop_mce(u16 ec, u8 xec)
Borislav Petkov5ce88f62010-08-31 18:28:08 +0200591{
592 return false;
593}
594
Borislav Petkov7cfd4a82010-09-01 14:45:20 +0200595void amd_decode_nb_mce(int node_id, struct mce *m, u32 nbcfg)
Borislav Petkov549d0422009-07-24 13:51:42 +0200596{
Borislav Petkov62452882010-09-22 16:08:37 +0200597 u16 ec = EC(m->status);
598 u8 xec = XEC(m->status, 0x1f);
Borislav Petkov7cfd4a82010-09-01 14:45:20 +0200599 u32 nbsh = (u32)(m->status >> 32);
Borislav Petkov549d0422009-07-24 13:51:42 +0200600
Borislav Petkov5ce88f62010-08-31 18:28:08 +0200601 pr_emerg(HW_ERR "Northbridge Error, node %d: ", node_id);
Borislav Petkov549d0422009-07-24 13:51:42 +0200602
603 /*
604 * F10h, revD can disable ErrCpu[3:0] so check that first and also the
605 * value encoding has changed so interpret those differently
606 */
607 if ((boot_cpu_data.x86 == 0x10) &&
Borislav Petkovcec79242009-10-27 19:12:02 +0100608 (boot_cpu_data.x86_model > 7)) {
Borislav Petkov7cfd4a82010-09-01 14:45:20 +0200609 if (nbsh & K8_NBSH_ERR_CPU_VAL)
Borislav Petkov5ce88f62010-08-31 18:28:08 +0200610 pr_cont(", core: %u", (u8)(nbsh & nb_err_cpumask));
Borislav Petkov549d0422009-07-24 13:51:42 +0200611 } else {
Borislav Petkov5ce88f62010-08-31 18:28:08 +0200612 u8 assoc_cpus = nbsh & nb_err_cpumask;
Borislav Petkov5b89d2f2010-03-09 20:38:48 +0100613
614 if (assoc_cpus > 0)
615 pr_cont(", core: %d", fls(assoc_cpus) - 1);
Borislav Petkov549d0422009-07-24 13:51:42 +0200616 }
617
Borislav Petkov5ce88f62010-08-31 18:28:08 +0200618 switch (xec) {
619 case 0x2:
620 pr_cont("Sync error (sync packets on HT link detected).\n");
621 return;
Borislav Petkovd93cc222009-07-28 10:56:15 +0200622
Borislav Petkov5ce88f62010-08-31 18:28:08 +0200623 case 0x3:
624 pr_cont("HT Master abort.\n");
625 return;
626
627 case 0x4:
628 pr_cont("HT Target abort.\n");
629 return;
630
631 case 0x7:
632 pr_cont("NB Watchdog timeout.\n");
633 return;
634
635 case 0x9:
636 pr_cont("SVM DMA Exclusion Vector error.\n");
637 return;
638
639 default:
640 break;
641 }
642
643 if (!fam_ops->nb_mce(ec, xec))
644 goto wrong_nb_mce;
645
646 if (boot_cpu_data.x86 == 0xf || boot_cpu_data.x86 == 0x10)
647 if ((xec == 0x8 || xec == 0x0) && nb_bus_decoder)
648 nb_bus_decoder(node_id, m, nbcfg);
649
650 return;
651
652wrong_nb_mce:
653 pr_emerg(HW_ERR "Corrupted NB MCE info?\n");
Borislav Petkovd93cc222009-07-28 10:56:15 +0200654}
655EXPORT_SYMBOL_GPL(amd_decode_nb_mce);
656
Borislav Petkov7cfd4a82010-09-01 14:45:20 +0200657static void amd_decode_fr_mce(struct mce *m)
Borislav Petkov53bd5fe2009-07-28 14:20:46 +0200658{
Borislav Petkov8259a7e2010-09-22 15:28:59 +0200659 struct cpuinfo_x86 *c = &boot_cpu_data;
Borislav Petkov62452882010-09-22 16:08:37 +0200660 u8 xec = XEC(m->status, xec_mask);
Borislav Petkov8259a7e2010-09-22 15:28:59 +0200661
662 if (c->x86 == 0xf || c->x86 == 0x11)
Borislav Petkovfe4ea262010-08-31 18:38:24 +0200663 goto wrong_fr_mce;
664
Borislav Petkov8259a7e2010-09-22 15:28:59 +0200665 if (c->x86 != 0x15 && xec != 0x0)
666 goto wrong_fr_mce;
667
668 pr_emerg(HW_ERR "%s Error: ",
669 (c->x86 == 0x15 ? "Execution Unit" : "FIROB"));
670
671 if (xec == 0x0 || xec == 0xc)
672 pr_cont("%s.\n", fr_ex_mce_desc[xec]);
673 else if (xec < 0xd)
674 pr_cont("%s parity error.\n", fr_ex_mce_desc[xec]);
675 else
676 goto wrong_fr_mce;
677
678 return;
Borislav Petkovfe4ea262010-08-31 18:38:24 +0200679
680wrong_fr_mce:
681 pr_emerg(HW_ERR "Corrupted FR MCE info?\n");
Borislav Petkov53bd5fe2009-07-28 14:20:46 +0200682}
683
Borislav Petkovb8f85c42010-09-22 15:37:58 +0200684static void amd_decode_fp_mce(struct mce *m)
685{
Borislav Petkov62452882010-09-22 16:08:37 +0200686 u8 xec = XEC(m->status, xec_mask);
Borislav Petkovb8f85c42010-09-22 15:37:58 +0200687
688 pr_emerg(HW_ERR "Floating Point Unit Error: ");
689
690 switch (xec) {
691 case 0x1:
692 pr_cont("Free List");
693 break;
694
695 case 0x2:
696 pr_cont("Physical Register File");
697 break;
698
699 case 0x3:
700 pr_cont("Retire Queue");
701 break;
702
703 case 0x4:
704 pr_cont("Scheduler table");
705 break;
706
707 case 0x5:
708 pr_cont("Status Register File");
709 break;
710
711 default:
712 goto wrong_fp_mce;
713 break;
714 }
715
716 pr_cont(" parity error.\n");
717
718 return;
719
720wrong_fp_mce:
721 pr_emerg(HW_ERR "Corrupted FP MCE info?\n");
722}
723
Borislav Petkov63375832010-09-06 18:13:39 +0200724static inline void amd_decode_err_code(u16 ec)
Borislav Petkovd93cc222009-07-28 10:56:15 +0200725{
Borislav Petkovfa7ae8c2010-09-22 17:42:27 +0200726
727 pr_emerg(HW_ERR "cache level: %s", LL_MSG(ec));
728
729 if (BUS_ERROR(ec))
730 pr_cont(", mem/io: %s", II_MSG(ec));
731 else
732 pr_cont(", tx: %s", TT_MSG(ec));
733
734 if (MEM_ERROR(ec) || BUS_ERROR(ec)) {
735 pr_cont(", mem-tx: %s", R4_MSG(ec));
736
737 if (BUS_ERROR(ec))
738 pr_cont(", part-proc: %s (%s)", PP_MSG(ec), TO_MSG(ec));
739 }
740
741 pr_cont("\n");
Borislav Petkov549d0422009-07-24 13:51:42 +0200742}
Borislav Petkov549d0422009-07-24 13:51:42 +0200743
Borislav Petkov5ce88f62010-08-31 18:28:08 +0200744/*
745 * Filter out unwanted MCE signatures here.
746 */
747static bool amd_filter_mce(struct mce *m)
748{
749 u8 xec = (m->status >> 16) & 0x1f;
750
751 /*
752 * NB GART TLB error reporting is disabled by default.
753 */
754 if (m->bank == 4 && xec == 0x5 && !report_gart_errors)
755 return true;
756
757 return false;
758}
759
Borislav Petkov9cdeb402010-09-02 18:33:24 +0200760int amd_decode_mce(struct notifier_block *nb, unsigned long val, void *data)
Borislav Petkov549d0422009-07-24 13:51:42 +0200761{
Borislav Petkovfb253192009-10-07 13:20:38 +0200762 struct mce *m = (struct mce *)data;
Borislav Petkovfa7ae8c2010-09-22 17:42:27 +0200763 struct cpuinfo_x86 *c = &boot_cpu_data;
Borislav Petkovb69b29d2009-07-27 16:21:14 +0200764 int node, ecc;
Borislav Petkov549d0422009-07-24 13:51:42 +0200765
Borislav Petkov5ce88f62010-08-31 18:28:08 +0200766 if (amd_filter_mce(m))
767 return NOTIFY_STOP;
768
Borislav Petkovfa7ae8c2010-09-22 17:42:27 +0200769 pr_emerg(HW_ERR "MC%d_STATUS[%s|%s|%s|%s|%s",
770 m->bank,
771 ((m->status & MCI_STATUS_OVER) ? "Over" : "-"),
772 ((m->status & MCI_STATUS_UC) ? "UE" : "CE"),
773 ((m->status & MCI_STATUS_MISCV) ? "MiscV" : "-"),
774 ((m->status & MCI_STATUS_PCC) ? "PCC" : "-"),
775 ((m->status & MCI_STATUS_ADDRV) ? "AddrV" : "-"));
Borislav Petkov549d0422009-07-24 13:51:42 +0200776
Borislav Petkovfa7ae8c2010-09-22 17:42:27 +0200777 if (c->x86 == 0x15)
778 pr_cont("|%s|%s",
779 ((m->status & BIT(44)) ? "Deferred" : "-"),
780 ((m->status & BIT(43)) ? "Poison" : "-"));
Borislav Petkov549d0422009-07-24 13:51:42 +0200781
Borislav Petkovb69b29d2009-07-27 16:21:14 +0200782 /* do the two bits[14:13] together */
Borislav Petkov35d824b2010-04-30 15:19:02 +0200783 ecc = (m->status >> 45) & 0x3;
Borislav Petkovb69b29d2009-07-27 16:21:14 +0200784 if (ecc)
Borislav Petkovfa7ae8c2010-09-22 17:42:27 +0200785 pr_cont("|%sECC", ((ecc == 2) ? "C" : "U"));
Borislav Petkovb69b29d2009-07-27 16:21:14 +0200786
Borislav Petkovfa7ae8c2010-09-22 17:42:27 +0200787 pr_cont("]: 0x%016llx\n", m->status);
788
Borislav Petkovb69b29d2009-07-27 16:21:14 +0200789
Borislav Petkov51966242009-07-28 13:50:43 +0200790 switch (m->bank) {
791 case 0:
Borislav Petkov7cfd4a82010-09-01 14:45:20 +0200792 amd_decode_dc_mce(m);
Borislav Petkov51966242009-07-28 13:50:43 +0200793 break;
Borislav Petkovd93cc222009-07-28 10:56:15 +0200794
Borislav Petkovab5535e2009-07-28 14:06:26 +0200795 case 1:
Borislav Petkov7cfd4a82010-09-01 14:45:20 +0200796 amd_decode_ic_mce(m);
Borislav Petkovab5535e2009-07-28 14:06:26 +0200797 break;
798
Borislav Petkov56cad2d2009-07-28 14:14:24 +0200799 case 2:
Borislav Petkovfa7ae8c2010-09-22 17:42:27 +0200800 if (c->x86 == 0x15)
Borislav Petkov70fdb492010-09-21 20:45:10 +0200801 amd_decode_cu_mce(m);
802 else
803 amd_decode_bu_mce(m);
Borislav Petkov56cad2d2009-07-28 14:14:24 +0200804 break;
805
Borislav Petkovf9350ef2009-07-28 14:17:30 +0200806 case 3:
Borislav Petkov7cfd4a82010-09-01 14:45:20 +0200807 amd_decode_ls_mce(m);
Borislav Petkovf9350ef2009-07-28 14:17:30 +0200808 break;
809
Borislav Petkov51966242009-07-28 13:50:43 +0200810 case 4:
Borislav Petkov7cfd4a82010-09-01 14:45:20 +0200811 node = amd_get_nb_id(m->extcpu);
812 amd_decode_nb_mce(node, m, 0);
Borislav Petkov51966242009-07-28 13:50:43 +0200813 break;
814
Borislav Petkov53bd5fe2009-07-28 14:20:46 +0200815 case 5:
Borislav Petkov7cfd4a82010-09-01 14:45:20 +0200816 amd_decode_fr_mce(m);
Borislav Petkov53bd5fe2009-07-28 14:20:46 +0200817 break;
818
Borislav Petkovb8f85c42010-09-22 15:37:58 +0200819 case 6:
820 amd_decode_fp_mce(m);
821 break;
822
Borislav Petkov51966242009-07-28 13:50:43 +0200823 default:
824 break;
Borislav Petkovb69b29d2009-07-27 16:21:14 +0200825 }
Borislav Petkov51966242009-07-28 13:50:43 +0200826
827 amd_decode_err_code(m->status & 0xffff);
Borislav Petkovfb253192009-10-07 13:20:38 +0200828
829 return NOTIFY_STOP;
Borislav Petkov549d0422009-07-24 13:51:42 +0200830}
Borislav Petkov9cdeb402010-09-02 18:33:24 +0200831EXPORT_SYMBOL_GPL(amd_decode_mce);
Ingo Molnarf436f8b2009-10-01 16:14:32 +0200832
Borislav Petkovfb253192009-10-07 13:20:38 +0200833static struct notifier_block amd_mce_dec_nb = {
834 .notifier_call = amd_decode_mce,
835};
836
Ingo Molnarf436f8b2009-10-01 16:14:32 +0200837static int __init mce_amd_init(void)
838{
Borislav Petkove045c292010-08-06 18:55:45 +0200839 if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD)
840 return 0;
841
Borislav Petkovfda75612010-09-22 16:12:03 +0200842 if ((boot_cpu_data.x86 < 0xf || boot_cpu_data.x86 > 0x12) &&
Borislav Petkov9530d602010-09-06 15:05:45 +0200843 (boot_cpu_data.x86 != 0x14 || boot_cpu_data.x86_model > 0xf))
Borislav Petkove045c292010-08-06 18:55:45 +0200844 return 0;
845
Borislav Petkov888ab8e2010-08-18 15:11:35 +0200846 fam_ops = kzalloc(sizeof(struct amd_decoder_ops), GFP_KERNEL);
847 if (!fam_ops)
848 return -ENOMEM;
849
850 switch (boot_cpu_data.x86) {
851 case 0xf:
852 fam_ops->dc_mce = k8_dc_mce;
Borislav Petkovdd53bce2010-08-26 19:05:49 +0200853 fam_ops->ic_mce = k8_ic_mce;
Borislav Petkov5ce88f62010-08-31 18:28:08 +0200854 fam_ops->nb_mce = k8_nb_mce;
Borislav Petkov888ab8e2010-08-18 15:11:35 +0200855 break;
856
857 case 0x10:
858 fam_ops->dc_mce = f10h_dc_mce;
Borislav Petkovdd53bce2010-08-26 19:05:49 +0200859 fam_ops->ic_mce = k8_ic_mce;
Borislav Petkov5ce88f62010-08-31 18:28:08 +0200860 fam_ops->nb_mce = f10h_nb_mce;
Borislav Petkov888ab8e2010-08-18 15:11:35 +0200861 break;
862
Borislav Petkovf0157b32010-10-05 19:07:16 +0200863 case 0x11:
864 fam_ops->dc_mce = k8_dc_mce;
865 fam_ops->ic_mce = k8_ic_mce;
866 fam_ops->nb_mce = f10h_nb_mce;
867 break;
868
Borislav Petkov9be0bb12010-09-16 15:08:14 +0200869 case 0x12:
870 fam_ops->dc_mce = f12h_dc_mce;
Borislav Petkove7281eb32010-09-16 16:45:22 +0200871 fam_ops->ic_mce = k8_ic_mce;
Borislav Petkovcb9d5ec2010-09-16 17:36:12 +0200872 fam_ops->nb_mce = nb_noop_mce;
Borislav Petkov9be0bb12010-09-16 15:08:14 +0200873 break;
874
Borislav Petkov888ab8e2010-08-18 15:11:35 +0200875 case 0x14:
Borislav Petkov5ce88f62010-08-31 18:28:08 +0200876 nb_err_cpumask = 0x3;
Borislav Petkov888ab8e2010-08-18 15:11:35 +0200877 fam_ops->dc_mce = f14h_dc_mce;
Borislav Petkovdd53bce2010-08-26 19:05:49 +0200878 fam_ops->ic_mce = f14h_ic_mce;
Borislav Petkovcb9d5ec2010-09-16 17:36:12 +0200879 fam_ops->nb_mce = nb_noop_mce;
Borislav Petkov888ab8e2010-08-18 15:11:35 +0200880 break;
881
Borislav Petkov2be64bf2010-09-17 19:11:47 +0200882 case 0x15:
883 xec_mask = 0x1f;
Borislav Petkov25a4f8b2010-09-17 19:22:34 +0200884 fam_ops->dc_mce = f15h_dc_mce;
Borislav Petkov86039cd2010-11-08 15:03:35 +0100885 fam_ops->ic_mce = f15h_ic_mce;
Borislav Petkov05cd6672010-09-22 15:06:24 +0200886 fam_ops->nb_mce = f10h_nb_mce;
Borislav Petkov2be64bf2010-09-17 19:11:47 +0200887 break;
888
Borislav Petkov888ab8e2010-08-18 15:11:35 +0200889 default:
890 printk(KERN_WARNING "Huh? What family is that: %d?!\n",
891 boot_cpu_data.x86);
892 kfree(fam_ops);
893 return -EINVAL;
894 }
895
Borislav Petkov9530d602010-09-06 15:05:45 +0200896 pr_info("MCE: In-kernel MCE decoding enabled.\n");
897
Borislav Petkove045c292010-08-06 18:55:45 +0200898 atomic_notifier_chain_register(&x86_mce_decoder_chain, &amd_mce_dec_nb);
Ingo Molnarf436f8b2009-10-01 16:14:32 +0200899
900 return 0;
901}
902early_initcall(mce_amd_init);
Borislav Petkov0d18b2e2009-10-02 15:31:48 +0200903
904#ifdef MODULE
905static void __exit mce_amd_exit(void)
906{
Borislav Petkovfb253192009-10-07 13:20:38 +0200907 atomic_notifier_chain_unregister(&x86_mce_decoder_chain, &amd_mce_dec_nb);
Borislav Petkov888ab8e2010-08-18 15:11:35 +0200908 kfree(fam_ops);
Borislav Petkov0d18b2e2009-10-02 15:31:48 +0200909}
910
911MODULE_DESCRIPTION("AMD MCE decoder");
912MODULE_ALIAS("edac-mce-amd");
913MODULE_LICENSE("GPL");
914module_exit(mce_amd_exit);
915#endif