Commit | Line | Data |
---|---|---|
b70ef010 BP |
1 | #include <linux/module.h> |
2 | #include "edac_mce_amd.h" | |
b52401ce | 3 | |
549d042d | 4 | static bool report_gart_errors; |
7cfd4a87 | 5 | static void (*nb_bus_decoder)(int node_id, struct mce *m, u32 nbcfg); |
549d042d BP |
6 | |
7 | void amd_report_gart_errors(bool v) | |
8 | { | |
9 | report_gart_errors = v; | |
10 | } | |
11 | EXPORT_SYMBOL_GPL(amd_report_gart_errors); | |
12 | ||
7cfd4a87 | 13 | void amd_register_ecc_decoder(void (*f)(int, struct mce *, u32)) |
549d042d BP |
14 | { |
15 | nb_bus_decoder = f; | |
16 | } | |
17 | EXPORT_SYMBOL_GPL(amd_register_ecc_decoder); | |
18 | ||
7cfd4a87 | 19 | void amd_unregister_ecc_decoder(void (*f)(int, struct mce *, u32)) |
549d042d BP |
20 | { |
21 | if (nb_bus_decoder) { | |
22 | WARN_ON(nb_bus_decoder != f); | |
23 | ||
24 | nb_bus_decoder = NULL; | |
25 | } | |
26 | } | |
27 | EXPORT_SYMBOL_GPL(amd_unregister_ecc_decoder); | |
28 | ||
b52401ce DT |
29 | /* |
30 | * string representation for the different MCA reported error types, see F3x48 | |
31 | * or MSR0000_0411. | |
32 | */ | |
6337583d BP |
33 | |
34 | /* transaction type */ | |
35 | const char *tt_msgs[] = { "INSN", "DATA", "GEN", "RESV" }; | |
b70ef010 | 36 | EXPORT_SYMBOL_GPL(tt_msgs); |
b52401ce | 37 | |
6337583d BP |
38 | /* cache level */ |
39 | const char *ll_msgs[] = { "RESV", "L1", "L2", "L3/GEN" }; | |
b70ef010 | 40 | EXPORT_SYMBOL_GPL(ll_msgs); |
b52401ce | 41 | |
6337583d | 42 | /* memory transaction type */ |
b52401ce | 43 | const char *rrrr_msgs[] = { |
6337583d | 44 | "GEN", "RD", "WR", "DRD", "DWR", "IRD", "PRF", "EV", "SNP" |
b52401ce | 45 | }; |
b70ef010 | 46 | EXPORT_SYMBOL_GPL(rrrr_msgs); |
b52401ce | 47 | |
6337583d BP |
48 | /* participating processor */ |
49 | const char *pp_msgs[] = { "SRC", "RES", "OBS", "GEN" }; | |
b70ef010 | 50 | EXPORT_SYMBOL_GPL(pp_msgs); |
b52401ce | 51 | |
6337583d BP |
52 | /* request timeout */ |
53 | const char *to_msgs[] = { "no timeout", "timed out" }; | |
b70ef010 | 54 | EXPORT_SYMBOL_GPL(to_msgs); |
b52401ce | 55 | |
6337583d BP |
56 | /* memory or i/o */ |
57 | const char *ii_msgs[] = { "MEM", "RESV", "IO", "GEN" }; | |
b70ef010 | 58 | EXPORT_SYMBOL_GPL(ii_msgs); |
b52401ce | 59 | |
1c43f2e2 BP |
60 | /* |
61 | * Map the 4 or 5 (family-specific) bits of Extended Error code to the | |
62 | * string table. | |
63 | */ | |
64 | const char *ext_msgs[] = { | |
65 | "K8 ECC error", /* 0_0000b */ | |
66 | "CRC error on link", /* 0_0001b */ | |
67 | "Sync error packets on link", /* 0_0010b */ | |
68 | "Master Abort during link operation", /* 0_0011b */ | |
69 | "Target Abort during link operation", /* 0_0100b */ | |
70 | "Invalid GART PTE entry during table walk", /* 0_0101b */ | |
71 | "Unsupported atomic RMW command received", /* 0_0110b */ | |
72 | "WDT error: NB transaction timeout", /* 0_0111b */ | |
73 | "ECC/ChipKill ECC error", /* 0_1000b */ | |
74 | "SVM DEV Error", /* 0_1001b */ | |
75 | "Link Data error", /* 0_1010b */ | |
76 | "Link/L3/Probe Filter Protocol error", /* 0_1011b */ | |
77 | "NB Internal Arrays Parity error", /* 0_1100b */ | |
78 | "DRAM Address/Control Parity error", /* 0_1101b */ | |
79 | "Link Transmission error", /* 0_1110b */ | |
80 | "GART/DEV Table Walk Data error" /* 0_1111b */ | |
81 | "Res 0x100 error", /* 1_0000b */ | |
82 | "Res 0x101 error", /* 1_0001b */ | |
83 | "Res 0x102 error", /* 1_0010b */ | |
84 | "Res 0x103 error", /* 1_0011b */ | |
85 | "Res 0x104 error", /* 1_0100b */ | |
86 | "Res 0x105 error", /* 1_0101b */ | |
87 | "Res 0x106 error", /* 1_0110b */ | |
88 | "Res 0x107 error", /* 1_0111b */ | |
89 | "Res 0x108 error", /* 1_1000b */ | |
90 | "Res 0x109 error", /* 1_1001b */ | |
91 | "Res 0x10A error", /* 1_1010b */ | |
92 | "Res 0x10B error", /* 1_1011b */ | |
93 | "ECC error in L3 Cache Data", /* 1_1100b */ | |
94 | "L3 Cache Tag error", /* 1_1101b */ | |
95 | "L3 Cache LRU Parity error", /* 1_1110b */ | |
96 | "Probe Filter error" /* 1_1111b */ | |
b52401ce | 97 | }; |
b70ef010 | 98 | EXPORT_SYMBOL_GPL(ext_msgs); |
549d042d | 99 | |
7cfd4a87 | 100 | static void amd_decode_dc_mce(struct mce *m) |
51966241 | 101 | { |
7cfd4a87 BP |
102 | u32 ec = m->status & 0xffff; |
103 | u32 xec = (m->status >> 16) & 0xf; | |
51966241 | 104 | |
c9f281fd | 105 | pr_emerg(HW_ERR "Data Cache Error: "); |
51966241 BP |
106 | |
107 | if (xec == 1 && TLB_ERROR(ec)) | |
108 | pr_cont(": %s TLB multimatch.\n", LL_MSG(ec)); | |
109 | else if (xec == 0) { | |
7cfd4a87 | 110 | if (m->status & (1ULL << 40)) |
51966241 BP |
111 | pr_cont(" during Data Scrub.\n"); |
112 | else if (TLB_ERROR(ec)) | |
113 | pr_cont(": %s TLB parity error.\n", LL_MSG(ec)); | |
114 | else if (MEM_ERROR(ec)) { | |
115 | u8 ll = ec & 0x3; | |
116 | u8 tt = (ec >> 2) & 0x3; | |
117 | u8 rrrr = (ec >> 4) & 0xf; | |
118 | ||
119 | /* see F10h BKDG (31116), Table 92. */ | |
120 | if (ll == 0x1) { | |
121 | if (tt != 0x1) | |
122 | goto wrong_dc_mce; | |
123 | ||
124 | pr_cont(": Data/Tag %s error.\n", RRRR_MSG(ec)); | |
125 | ||
126 | } else if (ll == 0x2 && rrrr == 0x3) | |
127 | pr_cont(" during L1 linefill from L2.\n"); | |
128 | else | |
129 | goto wrong_dc_mce; | |
130 | } else if (BUS_ERROR(ec) && boot_cpu_data.x86 == 0xf) | |
131 | pr_cont(" during system linefill.\n"); | |
132 | else | |
133 | goto wrong_dc_mce; | |
134 | } else | |
135 | goto wrong_dc_mce; | |
136 | ||
137 | return; | |
138 | ||
139 | wrong_dc_mce: | |
c9f281fd | 140 | pr_emerg(HW_ERR "Corrupted DC MCE info?\n"); |
51966241 BP |
141 | } |
142 | ||
7cfd4a87 | 143 | static void amd_decode_ic_mce(struct mce *m) |
ab5535e7 | 144 | { |
7cfd4a87 BP |
145 | u32 ec = m->status & 0xffff; |
146 | u32 xec = (m->status >> 16) & 0xf; | |
ab5535e7 | 147 | |
c9f281fd | 148 | pr_emerg(HW_ERR "Instruction Cache Error"); |
ab5535e7 BP |
149 | |
150 | if (xec == 1 && TLB_ERROR(ec)) | |
151 | pr_cont(": %s TLB multimatch.\n", LL_MSG(ec)); | |
152 | else if (xec == 0) { | |
153 | if (TLB_ERROR(ec)) | |
154 | pr_cont(": %s TLB Parity error.\n", LL_MSG(ec)); | |
155 | else if (BUS_ERROR(ec)) { | |
156 | if (boot_cpu_data.x86 == 0xf && | |
7cfd4a87 | 157 | (m->status & BIT(58))) |
ab5535e7 BP |
158 | pr_cont(" during system linefill.\n"); |
159 | else | |
160 | pr_cont(" during attempted NB data read.\n"); | |
161 | } else if (MEM_ERROR(ec)) { | |
162 | u8 ll = ec & 0x3; | |
163 | u8 rrrr = (ec >> 4) & 0xf; | |
164 | ||
165 | if (ll == 0x2) | |
166 | pr_cont(" during a linefill from L2.\n"); | |
167 | else if (ll == 0x1) { | |
168 | ||
169 | switch (rrrr) { | |
170 | case 0x5: | |
171 | pr_cont(": Parity error during " | |
172 | "data load.\n"); | |
173 | break; | |
174 | ||
175 | case 0x7: | |
176 | pr_cont(": Copyback Parity/Victim" | |
177 | " error.\n"); | |
178 | break; | |
179 | ||
180 | case 0x8: | |
181 | pr_cont(": Tag Snoop error.\n"); | |
182 | break; | |
183 | ||
184 | default: | |
185 | goto wrong_ic_mce; | |
186 | break; | |
187 | } | |
188 | } | |
189 | } else | |
190 | goto wrong_ic_mce; | |
191 | } else | |
192 | goto wrong_ic_mce; | |
193 | ||
194 | return; | |
195 | ||
196 | wrong_ic_mce: | |
c9f281fd | 197 | pr_emerg(HW_ERR "Corrupted IC MCE info?\n"); |
ab5535e7 BP |
198 | } |
199 | ||
7cfd4a87 | 200 | static void amd_decode_bu_mce(struct mce *m) |
56cad2d6 | 201 | { |
7cfd4a87 BP |
202 | u32 ec = m->status & 0xffff; |
203 | u32 xec = (m->status >> 16) & 0xf; | |
56cad2d6 | 204 | |
c9f281fd | 205 | pr_emerg(HW_ERR "Bus Unit Error"); |
56cad2d6 BP |
206 | |
207 | if (xec == 0x1) | |
208 | pr_cont(" in the write data buffers.\n"); | |
209 | else if (xec == 0x3) | |
210 | pr_cont(" in the victim data buffers.\n"); | |
211 | else if (xec == 0x2 && MEM_ERROR(ec)) | |
212 | pr_cont(": %s error in the L2 cache tags.\n", RRRR_MSG(ec)); | |
213 | else if (xec == 0x0) { | |
214 | if (TLB_ERROR(ec)) | |
215 | pr_cont(": %s error in a Page Descriptor Cache or " | |
216 | "Guest TLB.\n", TT_MSG(ec)); | |
217 | else if (BUS_ERROR(ec)) | |
218 | pr_cont(": %s/ECC error in data read from NB: %s.\n", | |
219 | RRRR_MSG(ec), PP_MSG(ec)); | |
220 | else if (MEM_ERROR(ec)) { | |
221 | u8 rrrr = (ec >> 4) & 0xf; | |
222 | ||
223 | if (rrrr >= 0x7) | |
224 | pr_cont(": %s error during data copyback.\n", | |
225 | RRRR_MSG(ec)); | |
226 | else if (rrrr <= 0x1) | |
227 | pr_cont(": %s parity/ECC error during data " | |
228 | "access from L2.\n", RRRR_MSG(ec)); | |
229 | else | |
230 | goto wrong_bu_mce; | |
231 | } else | |
232 | goto wrong_bu_mce; | |
233 | } else | |
234 | goto wrong_bu_mce; | |
235 | ||
236 | return; | |
237 | ||
238 | wrong_bu_mce: | |
c9f281fd | 239 | pr_emerg(HW_ERR "Corrupted BU MCE info?\n"); |
56cad2d6 BP |
240 | } |
241 | ||
7cfd4a87 | 242 | static void amd_decode_ls_mce(struct mce *m) |
f9350efd | 243 | { |
7cfd4a87 BP |
244 | u32 ec = m->status & 0xffff; |
245 | u32 xec = (m->status >> 16) & 0xf; | |
f9350efd | 246 | |
c9f281fd | 247 | pr_emerg(HW_ERR "Load Store Error"); |
f9350efd BP |
248 | |
249 | if (xec == 0x0) { | |
250 | u8 rrrr = (ec >> 4) & 0xf; | |
251 | ||
252 | if (!BUS_ERROR(ec) || (rrrr != 0x3 && rrrr != 0x4)) | |
253 | goto wrong_ls_mce; | |
254 | ||
255 | pr_cont(" during %s.\n", RRRR_MSG(ec)); | |
256 | } | |
257 | return; | |
258 | ||
259 | wrong_ls_mce: | |
c9f281fd | 260 | pr_emerg(HW_ERR "Corrupted LS MCE info?\n"); |
f9350efd BP |
261 | } |
262 | ||
7cfd4a87 | 263 | void amd_decode_nb_mce(int node_id, struct mce *m, u32 nbcfg) |
549d042d | 264 | { |
7cfd4a87 BP |
265 | u32 ec = m->status & 0xffff; |
266 | u32 nbsh = (u32)(m->status >> 32); | |
267 | u32 nbsl = (u32)m->status; | |
549d042d | 268 | |
256f7276 BP |
269 | /* |
270 | * GART TLB error reporting is disabled by default. Bail out early. | |
271 | */ | |
272 | if (TLB_ERROR(ec) && !report_gart_errors) | |
273 | return; | |
274 | ||
c9f281fd | 275 | pr_emerg(HW_ERR "Northbridge Error, node %d", node_id); |
549d042d BP |
276 | |
277 | /* | |
278 | * F10h, revD can disable ErrCpu[3:0] so check that first and also the | |
279 | * value encoding has changed so interpret those differently | |
280 | */ | |
281 | if ((boot_cpu_data.x86 == 0x10) && | |
cec7924f | 282 | (boot_cpu_data.x86_model > 7)) { |
7cfd4a87 BP |
283 | if (nbsh & K8_NBSH_ERR_CPU_VAL) |
284 | pr_cont(", core: %u\n", (u8)(nbsh & 0xf)); | |
549d042d | 285 | } else { |
7cfd4a87 | 286 | u8 assoc_cpus = nbsh & 0xf; |
5b89d2f9 BP |
287 | |
288 | if (assoc_cpus > 0) | |
289 | pr_cont(", core: %d", fls(assoc_cpus) - 1); | |
290 | ||
291 | pr_cont("\n"); | |
549d042d BP |
292 | } |
293 | ||
7cfd4a87 | 294 | pr_emerg(HW_ERR "%s.\n", EXT_ERR_MSG(nbsl)); |
d93cc222 BP |
295 | |
296 | if (BUS_ERROR(ec) && nb_bus_decoder) | |
7cfd4a87 | 297 | nb_bus_decoder(node_id, m, nbcfg); |
d93cc222 BP |
298 | } |
299 | EXPORT_SYMBOL_GPL(amd_decode_nb_mce); | |
300 | ||
7cfd4a87 | 301 | static void amd_decode_fr_mce(struct mce *m) |
53bd5fed BP |
302 | { |
303 | /* we have only one error signature so match all fields at once. */ | |
7cfd4a87 | 304 | if ((m->status & 0xffff) == 0x0f0f) |
c9f281fd | 305 | pr_emerg(HW_ERR " FR Error: CPU Watchdog timer expire.\n"); |
53bd5fed | 306 | else |
c9f281fd | 307 | pr_emerg(HW_ERR "Corrupted FR MCE info?\n"); |
53bd5fed BP |
308 | } |
309 | ||
6337583d | 310 | static inline void amd_decode_err_code(u16 ec) |
d93cc222 | 311 | { |
549d042d | 312 | if (TLB_ERROR(ec)) { |
c9f281fd | 313 | pr_emerg(HW_ERR "Transaction: %s, Cache Level: %s\n", |
549d042d BP |
314 | TT_MSG(ec), LL_MSG(ec)); |
315 | } else if (MEM_ERROR(ec)) { | |
6337583d | 316 | pr_emerg(HW_ERR "Transaction: %s, Type: %s, Cache Level: %s\n", |
549d042d BP |
317 | RRRR_MSG(ec), TT_MSG(ec), LL_MSG(ec)); |
318 | } else if (BUS_ERROR(ec)) { | |
6337583d | 319 | pr_emerg(HW_ERR "Transaction: %s (%s), %s, Cache Level: %s, " |
d93cc222 BP |
320 | "Participating Processor: %s\n", |
321 | RRRR_MSG(ec), II_MSG(ec), TO_MSG(ec), LL_MSG(ec), | |
322 | PP_MSG(ec)); | |
323 | } else | |
c9f281fd | 324 | pr_emerg(HW_ERR "Huh? Unknown MCE error 0x%x\n", ec); |
549d042d | 325 | } |
549d042d | 326 | |
9cdeb404 | 327 | int amd_decode_mce(struct notifier_block *nb, unsigned long val, void *data) |
549d042d | 328 | { |
fb253195 | 329 | struct mce *m = (struct mce *)data; |
b69b29de | 330 | int node, ecc; |
549d042d | 331 | |
c9f281fd | 332 | pr_emerg(HW_ERR "MC%d_STATUS: ", m->bank); |
549d042d | 333 | |
37b7370a | 334 | pr_cont("%sorrected error, other errors lost: %s, " |
b69b29de BP |
335 | "CPU context corrupt: %s", |
336 | ((m->status & MCI_STATUS_UC) ? "Unc" : "C"), | |
37b7370a | 337 | ((m->status & MCI_STATUS_OVER) ? "yes" : "no"), |
b69b29de | 338 | ((m->status & MCI_STATUS_PCC) ? "yes" : "no")); |
549d042d | 339 | |
b69b29de | 340 | /* do the two bits[14:13] together */ |
35d824b2 | 341 | ecc = (m->status >> 45) & 0x3; |
b69b29de BP |
342 | if (ecc) |
343 | pr_cont(", %sECC Error", ((ecc == 2) ? "C" : "U")); | |
344 | ||
345 | pr_cont("\n"); | |
346 | ||
51966241 BP |
347 | switch (m->bank) { |
348 | case 0: | |
7cfd4a87 | 349 | amd_decode_dc_mce(m); |
51966241 | 350 | break; |
d93cc222 | 351 | |
ab5535e7 | 352 | case 1: |
7cfd4a87 | 353 | amd_decode_ic_mce(m); |
ab5535e7 BP |
354 | break; |
355 | ||
56cad2d6 | 356 | case 2: |
7cfd4a87 | 357 | amd_decode_bu_mce(m); |
56cad2d6 BP |
358 | break; |
359 | ||
f9350efd | 360 | case 3: |
7cfd4a87 | 361 | amd_decode_ls_mce(m); |
f9350efd BP |
362 | break; |
363 | ||
51966241 | 364 | case 4: |
7cfd4a87 BP |
365 | node = amd_get_nb_id(m->extcpu); |
366 | amd_decode_nb_mce(node, m, 0); | |
51966241 BP |
367 | break; |
368 | ||
53bd5fed | 369 | case 5: |
7cfd4a87 | 370 | amd_decode_fr_mce(m); |
53bd5fed BP |
371 | break; |
372 | ||
51966241 BP |
373 | default: |
374 | break; | |
b69b29de | 375 | } |
51966241 BP |
376 | |
377 | amd_decode_err_code(m->status & 0xffff); | |
fb253195 BP |
378 | |
379 | return NOTIFY_STOP; | |
549d042d | 380 | } |
9cdeb404 | 381 | EXPORT_SYMBOL_GPL(amd_decode_mce); |
f436f8bb | 382 | |
fb253195 BP |
383 | static struct notifier_block amd_mce_dec_nb = { |
384 | .notifier_call = amd_decode_mce, | |
385 | }; | |
386 | ||
f436f8bb IM |
387 | static int __init mce_amd_init(void) |
388 | { | |
389 | /* | |
e045c291 | 390 | * We can decode MCEs for K8, F10h and F11h CPUs: |
f436f8bb | 391 | */ |
e045c291 BP |
392 | if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD) |
393 | return 0; | |
394 | ||
395 | if (boot_cpu_data.x86 < 0xf || boot_cpu_data.x86 > 0x11) | |
396 | return 0; | |
397 | ||
398 | atomic_notifier_chain_register(&x86_mce_decoder_chain, &amd_mce_dec_nb); | |
f436f8bb IM |
399 | |
400 | return 0; | |
401 | } | |
402 | early_initcall(mce_amd_init); | |
0d18b2e3 BP |
403 | |
404 | #ifdef MODULE | |
405 | static void __exit mce_amd_exit(void) | |
406 | { | |
fb253195 | 407 | atomic_notifier_chain_unregister(&x86_mce_decoder_chain, &amd_mce_dec_nb); |
0d18b2e3 BP |
408 | } |
409 | ||
410 | MODULE_DESCRIPTION("AMD MCE decoder"); | |
411 | MODULE_ALIAS("edac-mce-amd"); | |
412 | MODULE_LICENSE("GPL"); | |
413 | module_exit(mce_amd_exit); | |
414 | #endif |