mce_power.c (26476B)
1// SPDX-License-Identifier: GPL-2.0-or-later 2/* 3 * Machine check exception handling CPU-side for power7 and power8 4 * 5 * Copyright 2013 IBM Corporation 6 * Author: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com> 7 */ 8 9#undef DEBUG 10#define pr_fmt(fmt) "mce_power: " fmt 11 12#include <linux/types.h> 13#include <linux/ptrace.h> 14#include <linux/extable.h> 15#include <linux/pgtable.h> 16#include <asm/mmu.h> 17#include <asm/mce.h> 18#include <asm/machdep.h> 19#include <asm/pte-walk.h> 20#include <asm/sstep.h> 21#include <asm/exception-64s.h> 22#include <asm/extable.h> 23#include <asm/inst.h> 24 25/* 26 * Convert an address related to an mm to a PFN. NOTE: we are in real 27 * mode, we could potentially race with page table updates. 28 */ 29unsigned long addr_to_pfn(struct pt_regs *regs, unsigned long addr) 30{ 31 pte_t *ptep, pte; 32 unsigned int shift; 33 unsigned long pfn, flags; 34 struct mm_struct *mm; 35 36 if (user_mode(regs)) 37 mm = current->mm; 38 else 39 mm = &init_mm; 40 41 local_irq_save(flags); 42 ptep = __find_linux_pte(mm->pgd, addr, NULL, &shift); 43 if (!ptep) { 44 pfn = ULONG_MAX; 45 goto out; 46 } 47 pte = READ_ONCE(*ptep); 48 49 if (!pte_present(pte) || pte_special(pte)) { 50 pfn = ULONG_MAX; 51 goto out; 52 } 53 54 if (shift <= PAGE_SHIFT) 55 pfn = pte_pfn(pte); 56 else { 57 unsigned long rpnmask = (1ul << shift) - PAGE_SIZE; 58 pfn = pte_pfn(__pte(pte_val(pte) | (addr & rpnmask))); 59 } 60out: 61 local_irq_restore(flags); 62 return pfn; 63} 64 65static bool mce_in_guest(void) 66{ 67#ifdef CONFIG_KVM_BOOK3S_HANDLER 68 /* 69 * If machine check is hit when in guest context or low level KVM 70 * code, avoid looking up any translations or making any attempts 71 * to recover, just record the event and pass to KVM. 72 */ 73 if (get_paca()->kvm_hstate.in_guest) 74 return true; 75#endif 76 return false; 77} 78 79/* flush SLBs and reload */ 80#ifdef CONFIG_PPC_64S_HASH_MMU 81void flush_and_reload_slb(void) 82{ 83 if (early_radix_enabled()) 84 return; 85 86 /* Invalidate all SLBs */ 87 slb_flush_all_realmode(); 88 89 /* 90 * This probably shouldn't happen, but it may be possible it's 91 * called in early boot before SLB shadows are allocated. 92 */ 93 if (!get_slb_shadow()) 94 return; 95 96 slb_restore_bolted_realmode(); 97} 98#endif 99 100void flush_erat(void) 101{ 102#ifdef CONFIG_PPC_64S_HASH_MMU 103 if (!early_cpu_has_feature(CPU_FTR_ARCH_300)) { 104 flush_and_reload_slb(); 105 return; 106 } 107#endif 108 asm volatile(PPC_ISA_3_0_INVALIDATE_ERAT : : :"memory"); 109} 110 111#define MCE_FLUSH_SLB 1 112#define MCE_FLUSH_TLB 2 113#define MCE_FLUSH_ERAT 3 114 115static int mce_flush(int what) 116{ 117#ifdef CONFIG_PPC_64S_HASH_MMU 118 if (what == MCE_FLUSH_SLB) { 119 flush_and_reload_slb(); 120 return 1; 121 } 122#endif 123 if (what == MCE_FLUSH_ERAT) { 124 flush_erat(); 125 return 1; 126 } 127 if (what == MCE_FLUSH_TLB) { 128 tlbiel_all(); 129 return 1; 130 } 131 132 return 0; 133} 134 135#define SRR1_MC_LOADSTORE(srr1) ((srr1) & PPC_BIT(42)) 136 137struct mce_ierror_table { 138 unsigned long srr1_mask; 139 unsigned long srr1_value; 140 bool nip_valid; /* nip is a valid indicator of faulting address */ 141 unsigned int error_type; 142 unsigned int error_subtype; 143 unsigned int error_class; 144 unsigned int initiator; 145 unsigned int severity; 146 bool sync_error; 147}; 148 149static const struct mce_ierror_table mce_p7_ierror_table[] = { 150{ 0x00000000001c0000, 0x0000000000040000, true, 151 MCE_ERROR_TYPE_UE, MCE_UE_ERROR_IFETCH, MCE_ECLASS_HARDWARE, 152 MCE_INITIATOR_CPU, MCE_SEV_SEVERE, true }, 153{ 0x00000000001c0000, 0x0000000000080000, true, 154 MCE_ERROR_TYPE_SLB, MCE_SLB_ERROR_PARITY, MCE_ECLASS_HARD_INDETERMINATE, 155 MCE_INITIATOR_CPU, MCE_SEV_SEVERE, true }, 156{ 0x00000000001c0000, 0x00000000000c0000, true, 157 MCE_ERROR_TYPE_SLB, MCE_SLB_ERROR_MULTIHIT, MCE_ECLASS_SOFT_INDETERMINATE, 158 MCE_INITIATOR_CPU, MCE_SEV_WARNING, true }, 159{ 0x00000000001c0000, 0x0000000000100000, true, 160 MCE_ERROR_TYPE_SLB, MCE_SLB_ERROR_INDETERMINATE, /* BOTH */ 161 MCE_ECLASS_SOFT_INDETERMINATE, 162 MCE_INITIATOR_CPU, MCE_SEV_WARNING, true }, 163{ 0x00000000001c0000, 0x0000000000140000, true, 164 MCE_ERROR_TYPE_TLB, MCE_TLB_ERROR_MULTIHIT, MCE_ECLASS_SOFT_INDETERMINATE, 165 MCE_INITIATOR_CPU, MCE_SEV_WARNING, true }, 166{ 0x00000000001c0000, 0x0000000000180000, true, 167 MCE_ERROR_TYPE_UE, MCE_UE_ERROR_PAGE_TABLE_WALK_IFETCH, MCE_ECLASS_HARDWARE, 168 MCE_INITIATOR_CPU, MCE_SEV_SEVERE, true }, 169{ 0x00000000001c0000, 0x00000000001c0000, true, 170 MCE_ERROR_TYPE_UE, MCE_UE_ERROR_IFETCH, MCE_ECLASS_HARDWARE, 171 MCE_INITIATOR_CPU, MCE_SEV_SEVERE, true }, 172{ 0, 0, 0, 0, 0, 0, 0 } }; 173 174static const struct mce_ierror_table mce_p8_ierror_table[] = { 175{ 0x00000000081c0000, 0x0000000000040000, true, 176 MCE_ERROR_TYPE_UE, MCE_UE_ERROR_IFETCH, MCE_ECLASS_HARDWARE, 177 MCE_INITIATOR_CPU, MCE_SEV_SEVERE, true }, 178{ 0x00000000081c0000, 0x0000000000080000, true, 179 MCE_ERROR_TYPE_SLB, MCE_SLB_ERROR_PARITY, MCE_ECLASS_HARD_INDETERMINATE, 180 MCE_INITIATOR_CPU, MCE_SEV_SEVERE, true }, 181{ 0x00000000081c0000, 0x00000000000c0000, true, 182 MCE_ERROR_TYPE_SLB, MCE_SLB_ERROR_MULTIHIT, MCE_ECLASS_SOFT_INDETERMINATE, 183 MCE_INITIATOR_CPU, MCE_SEV_WARNING, true }, 184{ 0x00000000081c0000, 0x0000000000100000, true, 185 MCE_ERROR_TYPE_ERAT, MCE_ERAT_ERROR_MULTIHIT, MCE_ECLASS_SOFT_INDETERMINATE, 186 MCE_INITIATOR_CPU, MCE_SEV_WARNING, true }, 187{ 0x00000000081c0000, 0x0000000000140000, true, 188 MCE_ERROR_TYPE_TLB, MCE_TLB_ERROR_MULTIHIT, MCE_ECLASS_SOFT_INDETERMINATE, 189 MCE_INITIATOR_CPU, MCE_SEV_WARNING, true }, 190{ 0x00000000081c0000, 0x0000000000180000, true, 191 MCE_ERROR_TYPE_UE, MCE_UE_ERROR_PAGE_TABLE_WALK_IFETCH, 192 MCE_ECLASS_HARDWARE, 193 MCE_INITIATOR_CPU, MCE_SEV_SEVERE, true }, 194{ 0x00000000081c0000, 0x00000000001c0000, true, 195 MCE_ERROR_TYPE_UE, MCE_UE_ERROR_IFETCH, MCE_ECLASS_HARDWARE, 196 MCE_INITIATOR_CPU, MCE_SEV_SEVERE, true }, 197{ 0x00000000081c0000, 0x0000000008000000, true, 198 MCE_ERROR_TYPE_LINK, MCE_LINK_ERROR_IFETCH_TIMEOUT, MCE_ECLASS_HARDWARE, 199 MCE_INITIATOR_CPU, MCE_SEV_SEVERE, true }, 200{ 0x00000000081c0000, 0x0000000008040000, true, 201 MCE_ERROR_TYPE_LINK,MCE_LINK_ERROR_PAGE_TABLE_WALK_IFETCH_TIMEOUT, 202 MCE_ECLASS_HARDWARE, 203 MCE_INITIATOR_CPU, MCE_SEV_SEVERE, true }, 204{ 0, 0, 0, 0, 0, 0, 0 } }; 205 206static const struct mce_ierror_table mce_p9_ierror_table[] = { 207{ 0x00000000081c0000, 0x0000000000040000, true, 208 MCE_ERROR_TYPE_UE, MCE_UE_ERROR_IFETCH, MCE_ECLASS_HARDWARE, 209 MCE_INITIATOR_CPU, MCE_SEV_SEVERE, true }, 210{ 0x00000000081c0000, 0x0000000000080000, true, 211 MCE_ERROR_TYPE_SLB, MCE_SLB_ERROR_PARITY, MCE_ECLASS_HARD_INDETERMINATE, 212 MCE_INITIATOR_CPU, MCE_SEV_SEVERE, true }, 213{ 0x00000000081c0000, 0x00000000000c0000, true, 214 MCE_ERROR_TYPE_SLB, MCE_SLB_ERROR_MULTIHIT, MCE_ECLASS_SOFT_INDETERMINATE, 215 MCE_INITIATOR_CPU, MCE_SEV_WARNING, true }, 216{ 0x00000000081c0000, 0x0000000000100000, true, 217 MCE_ERROR_TYPE_ERAT, MCE_ERAT_ERROR_MULTIHIT, MCE_ECLASS_SOFT_INDETERMINATE, 218 MCE_INITIATOR_CPU, MCE_SEV_WARNING, true }, 219{ 0x00000000081c0000, 0x0000000000140000, true, 220 MCE_ERROR_TYPE_TLB, MCE_TLB_ERROR_MULTIHIT, MCE_ECLASS_SOFT_INDETERMINATE, 221 MCE_INITIATOR_CPU, MCE_SEV_WARNING, true }, 222{ 0x00000000081c0000, 0x0000000000180000, true, 223 MCE_ERROR_TYPE_UE, MCE_UE_ERROR_PAGE_TABLE_WALK_IFETCH, MCE_ECLASS_HARDWARE, 224 MCE_INITIATOR_CPU, MCE_SEV_SEVERE, true }, 225{ 0x00000000081c0000, 0x00000000001c0000, true, 226 MCE_ERROR_TYPE_RA, MCE_RA_ERROR_IFETCH_FOREIGN, MCE_ECLASS_SOFTWARE, 227 MCE_INITIATOR_CPU, MCE_SEV_SEVERE, true }, 228{ 0x00000000081c0000, 0x0000000008000000, true, 229 MCE_ERROR_TYPE_LINK, MCE_LINK_ERROR_IFETCH_TIMEOUT, MCE_ECLASS_HARDWARE, 230 MCE_INITIATOR_CPU, MCE_SEV_SEVERE, true }, 231{ 0x00000000081c0000, 0x0000000008040000, true, 232 MCE_ERROR_TYPE_LINK,MCE_LINK_ERROR_PAGE_TABLE_WALK_IFETCH_TIMEOUT, 233 MCE_ECLASS_HARDWARE, 234 MCE_INITIATOR_CPU, MCE_SEV_SEVERE, true }, 235{ 0x00000000081c0000, 0x00000000080c0000, true, 236 MCE_ERROR_TYPE_RA, MCE_RA_ERROR_IFETCH, MCE_ECLASS_SOFTWARE, 237 MCE_INITIATOR_CPU, MCE_SEV_SEVERE, true }, 238{ 0x00000000081c0000, 0x0000000008100000, true, 239 MCE_ERROR_TYPE_RA, MCE_RA_ERROR_PAGE_TABLE_WALK_IFETCH, MCE_ECLASS_SOFTWARE, 240 MCE_INITIATOR_CPU, MCE_SEV_SEVERE, true }, 241{ 0x00000000081c0000, 0x0000000008140000, false, 242 MCE_ERROR_TYPE_RA, MCE_RA_ERROR_STORE, MCE_ECLASS_HARDWARE, 243 MCE_INITIATOR_CPU, MCE_SEV_FATAL, false }, /* ASYNC is fatal */ 244{ 0x00000000081c0000, 0x0000000008180000, false, 245 MCE_ERROR_TYPE_LINK,MCE_LINK_ERROR_STORE_TIMEOUT, 246 MCE_INITIATOR_CPU, MCE_SEV_FATAL, false }, /* ASYNC is fatal */ 247{ 0x00000000081c0000, 0x00000000081c0000, true, MCE_ECLASS_HARDWARE, 248 MCE_ERROR_TYPE_RA, MCE_RA_ERROR_PAGE_TABLE_WALK_IFETCH_FOREIGN, 249 MCE_INITIATOR_CPU, MCE_SEV_SEVERE, true }, 250{ 0, 0, 0, 0, 0, 0, 0 } }; 251 252static const struct mce_ierror_table mce_p10_ierror_table[] = { 253{ 0x00000000081c0000, 0x0000000000040000, true, 254 MCE_ERROR_TYPE_UE, MCE_UE_ERROR_IFETCH, MCE_ECLASS_HARDWARE, 255 MCE_INITIATOR_CPU, MCE_SEV_SEVERE, true }, 256{ 0x00000000081c0000, 0x0000000000080000, true, 257 MCE_ERROR_TYPE_SLB, MCE_SLB_ERROR_PARITY, MCE_ECLASS_HARD_INDETERMINATE, 258 MCE_INITIATOR_CPU, MCE_SEV_SEVERE, true }, 259{ 0x00000000081c0000, 0x00000000000c0000, true, 260 MCE_ERROR_TYPE_SLB, MCE_SLB_ERROR_MULTIHIT, MCE_ECLASS_SOFT_INDETERMINATE, 261 MCE_INITIATOR_CPU, MCE_SEV_WARNING, true }, 262{ 0x00000000081c0000, 0x0000000000100000, true, 263 MCE_ERROR_TYPE_ERAT, MCE_ERAT_ERROR_MULTIHIT, MCE_ECLASS_SOFT_INDETERMINATE, 264 MCE_INITIATOR_CPU, MCE_SEV_WARNING, true }, 265{ 0x00000000081c0000, 0x0000000000140000, true, 266 MCE_ERROR_TYPE_TLB, MCE_TLB_ERROR_MULTIHIT, MCE_ECLASS_SOFT_INDETERMINATE, 267 MCE_INITIATOR_CPU, MCE_SEV_WARNING, true }, 268{ 0x00000000081c0000, 0x0000000000180000, true, 269 MCE_ERROR_TYPE_UE, MCE_UE_ERROR_PAGE_TABLE_WALK_IFETCH, MCE_ECLASS_HARDWARE, 270 MCE_INITIATOR_CPU, MCE_SEV_SEVERE, true }, 271{ 0x00000000081c0000, 0x00000000001c0000, true, 272 MCE_ERROR_TYPE_RA, MCE_RA_ERROR_IFETCH_FOREIGN, MCE_ECLASS_SOFTWARE, 273 MCE_INITIATOR_CPU, MCE_SEV_SEVERE, true }, 274{ 0x00000000081c0000, 0x0000000008080000, true, 275 MCE_ERROR_TYPE_USER,MCE_USER_ERROR_SCV, MCE_ECLASS_SOFTWARE, 276 MCE_INITIATOR_CPU, MCE_SEV_WARNING, true }, 277{ 0x00000000081c0000, 0x00000000080c0000, true, 278 MCE_ERROR_TYPE_RA, MCE_RA_ERROR_IFETCH, MCE_ECLASS_SOFTWARE, 279 MCE_INITIATOR_CPU, MCE_SEV_SEVERE, true }, 280{ 0x00000000081c0000, 0x0000000008100000, true, 281 MCE_ERROR_TYPE_RA, MCE_RA_ERROR_PAGE_TABLE_WALK_IFETCH, MCE_ECLASS_SOFTWARE, 282 MCE_INITIATOR_CPU, MCE_SEV_SEVERE, true }, 283{ 0x00000000081c0000, 0x0000000008140000, false, 284 MCE_ERROR_TYPE_RA, MCE_RA_ERROR_STORE, MCE_ECLASS_HARDWARE, 285 MCE_INITIATOR_CPU, MCE_SEV_FATAL, false }, /* ASYNC is fatal */ 286{ 0x00000000081c0000, 0x00000000081c0000, true, MCE_ECLASS_HARDWARE, 287 MCE_ERROR_TYPE_RA, MCE_RA_ERROR_PAGE_TABLE_WALK_IFETCH_FOREIGN, 288 MCE_INITIATOR_CPU, MCE_SEV_SEVERE, true }, 289{ 0, 0, 0, 0, 0, 0, 0 } }; 290 291struct mce_derror_table { 292 unsigned long dsisr_value; 293 bool dar_valid; /* dar is a valid indicator of faulting address */ 294 unsigned int error_type; 295 unsigned int error_subtype; 296 unsigned int error_class; 297 unsigned int initiator; 298 unsigned int severity; 299 bool sync_error; 300}; 301 302static const struct mce_derror_table mce_p7_derror_table[] = { 303{ 0x00008000, false, 304 MCE_ERROR_TYPE_UE, MCE_UE_ERROR_LOAD_STORE, MCE_ECLASS_HARDWARE, 305 MCE_INITIATOR_CPU, MCE_SEV_SEVERE, true }, 306{ 0x00004000, true, 307 MCE_ERROR_TYPE_UE, MCE_UE_ERROR_PAGE_TABLE_WALK_LOAD_STORE, 308 MCE_ECLASS_HARDWARE, 309 MCE_INITIATOR_CPU, MCE_SEV_SEVERE, true }, 310{ 0x00000800, true, 311 MCE_ERROR_TYPE_ERAT, MCE_ERAT_ERROR_MULTIHIT, MCE_ECLASS_SOFT_INDETERMINATE, 312 MCE_INITIATOR_CPU, MCE_SEV_WARNING, true }, 313{ 0x00000400, true, 314 MCE_ERROR_TYPE_TLB, MCE_TLB_ERROR_MULTIHIT, MCE_ECLASS_SOFT_INDETERMINATE, 315 MCE_INITIATOR_CPU, MCE_SEV_WARNING, true }, 316{ 0x00000080, true, 317 MCE_ERROR_TYPE_SLB, MCE_SLB_ERROR_MULTIHIT, MCE_ECLASS_SOFT_INDETERMINATE, 318 MCE_INITIATOR_CPU, MCE_SEV_WARNING, true }, 319{ 0x00000100, true, 320 MCE_ERROR_TYPE_SLB, MCE_SLB_ERROR_PARITY, MCE_ECLASS_HARD_INDETERMINATE, 321 MCE_INITIATOR_CPU, MCE_SEV_SEVERE, true }, 322{ 0x00000040, true, 323 MCE_ERROR_TYPE_SLB, MCE_SLB_ERROR_INDETERMINATE, /* BOTH */ 324 MCE_ECLASS_HARD_INDETERMINATE, 325 MCE_INITIATOR_CPU, MCE_SEV_WARNING, true }, 326{ 0, false, 0, 0, 0, 0, 0 } }; 327 328static const struct mce_derror_table mce_p8_derror_table[] = { 329{ 0x00008000, false, 330 MCE_ERROR_TYPE_UE, MCE_UE_ERROR_LOAD_STORE, MCE_ECLASS_HARDWARE, 331 MCE_INITIATOR_CPU, MCE_SEV_SEVERE, true }, 332{ 0x00004000, true, 333 MCE_ERROR_TYPE_UE, MCE_UE_ERROR_PAGE_TABLE_WALK_LOAD_STORE, 334 MCE_ECLASS_HARDWARE, 335 MCE_INITIATOR_CPU, MCE_SEV_SEVERE, true }, 336{ 0x00002000, true, 337 MCE_ERROR_TYPE_LINK, MCE_LINK_ERROR_LOAD_TIMEOUT, MCE_ECLASS_HARDWARE, 338 MCE_INITIATOR_CPU, MCE_SEV_SEVERE, true }, 339{ 0x00001000, true, 340 MCE_ERROR_TYPE_LINK, MCE_LINK_ERROR_PAGE_TABLE_WALK_LOAD_STORE_TIMEOUT, 341 MCE_ECLASS_HARDWARE, 342 MCE_INITIATOR_CPU, MCE_SEV_SEVERE, true }, 343{ 0x00000800, true, 344 MCE_ERROR_TYPE_ERAT, MCE_ERAT_ERROR_MULTIHIT, MCE_ECLASS_SOFT_INDETERMINATE, 345 MCE_INITIATOR_CPU, MCE_SEV_WARNING, true }, 346{ 0x00000400, true, 347 MCE_ERROR_TYPE_TLB, MCE_TLB_ERROR_MULTIHIT, MCE_ECLASS_SOFT_INDETERMINATE, 348 MCE_INITIATOR_CPU, MCE_SEV_WARNING, true }, 349{ 0x00000200, true, 350 MCE_ERROR_TYPE_ERAT, MCE_ERAT_ERROR_MULTIHIT, /* SECONDARY ERAT */ 351 MCE_ECLASS_SOFT_INDETERMINATE, 352 MCE_INITIATOR_CPU, MCE_SEV_WARNING, true }, 353{ 0x00000080, true, 354 MCE_ERROR_TYPE_SLB, MCE_SLB_ERROR_MULTIHIT, /* Before PARITY */ 355 MCE_ECLASS_SOFT_INDETERMINATE, 356 MCE_INITIATOR_CPU, MCE_SEV_WARNING, true }, 357{ 0x00000100, true, 358 MCE_ERROR_TYPE_SLB, MCE_SLB_ERROR_PARITY, MCE_ECLASS_HARD_INDETERMINATE, 359 MCE_INITIATOR_CPU, MCE_SEV_SEVERE, true }, 360{ 0, false, 0, 0, 0, 0, 0 } }; 361 362static const struct mce_derror_table mce_p9_derror_table[] = { 363{ 0x00008000, false, 364 MCE_ERROR_TYPE_UE, MCE_UE_ERROR_LOAD_STORE, MCE_ECLASS_HARDWARE, 365 MCE_INITIATOR_CPU, MCE_SEV_SEVERE, true }, 366{ 0x00004000, true, 367 MCE_ERROR_TYPE_UE, MCE_UE_ERROR_PAGE_TABLE_WALK_LOAD_STORE, 368 MCE_ECLASS_HARDWARE, 369 MCE_INITIATOR_CPU, MCE_SEV_SEVERE, true }, 370{ 0x00002000, true, 371 MCE_ERROR_TYPE_LINK, MCE_LINK_ERROR_LOAD_TIMEOUT, MCE_ECLASS_HARDWARE, 372 MCE_INITIATOR_CPU, MCE_SEV_SEVERE, true }, 373{ 0x00001000, true, 374 MCE_ERROR_TYPE_LINK, MCE_LINK_ERROR_PAGE_TABLE_WALK_LOAD_STORE_TIMEOUT, 375 MCE_ECLASS_HARDWARE, 376 MCE_INITIATOR_CPU, MCE_SEV_SEVERE, true }, 377{ 0x00000800, true, 378 MCE_ERROR_TYPE_ERAT, MCE_ERAT_ERROR_MULTIHIT, MCE_ECLASS_SOFT_INDETERMINATE, 379 MCE_INITIATOR_CPU, MCE_SEV_WARNING, true }, 380{ 0x00000400, true, 381 MCE_ERROR_TYPE_TLB, MCE_TLB_ERROR_MULTIHIT, MCE_ECLASS_SOFT_INDETERMINATE, 382 MCE_INITIATOR_CPU, MCE_SEV_WARNING, true }, 383{ 0x00000200, false, 384 MCE_ERROR_TYPE_USER, MCE_USER_ERROR_TLBIE, MCE_ECLASS_SOFTWARE, 385 MCE_INITIATOR_CPU, MCE_SEV_WARNING, true }, 386{ 0x00000080, true, 387 MCE_ERROR_TYPE_SLB, MCE_SLB_ERROR_MULTIHIT, /* Before PARITY */ 388 MCE_ECLASS_SOFT_INDETERMINATE, 389 MCE_INITIATOR_CPU, MCE_SEV_WARNING, true }, 390{ 0x00000100, true, 391 MCE_ERROR_TYPE_SLB, MCE_SLB_ERROR_PARITY, MCE_ECLASS_HARD_INDETERMINATE, 392 MCE_INITIATOR_CPU, MCE_SEV_SEVERE, true }, 393{ 0x00000040, true, 394 MCE_ERROR_TYPE_RA, MCE_RA_ERROR_LOAD, MCE_ECLASS_HARDWARE, 395 MCE_INITIATOR_CPU, MCE_SEV_SEVERE, true }, 396{ 0x00000020, false, 397 MCE_ERROR_TYPE_RA, MCE_RA_ERROR_PAGE_TABLE_WALK_LOAD_STORE, 398 MCE_ECLASS_HARDWARE, 399 MCE_INITIATOR_CPU, MCE_SEV_SEVERE, true }, 400{ 0x00000010, false, 401 MCE_ERROR_TYPE_RA, MCE_RA_ERROR_PAGE_TABLE_WALK_LOAD_STORE_FOREIGN, 402 MCE_ECLASS_HARDWARE, 403 MCE_INITIATOR_CPU, MCE_SEV_SEVERE, true }, 404{ 0x00000008, false, 405 MCE_ERROR_TYPE_RA, MCE_RA_ERROR_LOAD_STORE_FOREIGN, MCE_ECLASS_HARDWARE, 406 MCE_INITIATOR_CPU, MCE_SEV_SEVERE, true }, 407{ 0, false, 0, 0, 0, 0, 0 } }; 408 409static const struct mce_derror_table mce_p10_derror_table[] = { 410{ 0x00008000, false, 411 MCE_ERROR_TYPE_UE, MCE_UE_ERROR_LOAD_STORE, MCE_ECLASS_HARDWARE, 412 MCE_INITIATOR_CPU, MCE_SEV_SEVERE, true }, 413{ 0x00004000, true, 414 MCE_ERROR_TYPE_UE, MCE_UE_ERROR_PAGE_TABLE_WALK_LOAD_STORE, 415 MCE_ECLASS_HARDWARE, 416 MCE_INITIATOR_CPU, MCE_SEV_SEVERE, true }, 417{ 0x00000800, true, 418 MCE_ERROR_TYPE_ERAT, MCE_ERAT_ERROR_MULTIHIT, MCE_ECLASS_SOFT_INDETERMINATE, 419 MCE_INITIATOR_CPU, MCE_SEV_WARNING, true }, 420{ 0x00000400, true, 421 MCE_ERROR_TYPE_TLB, MCE_TLB_ERROR_MULTIHIT, MCE_ECLASS_SOFT_INDETERMINATE, 422 MCE_INITIATOR_CPU, MCE_SEV_WARNING, true }, 423{ 0x00000200, false, 424 MCE_ERROR_TYPE_USER, MCE_USER_ERROR_TLBIE, MCE_ECLASS_SOFTWARE, 425 MCE_INITIATOR_CPU, MCE_SEV_WARNING, true }, 426{ 0x00000080, true, 427 MCE_ERROR_TYPE_SLB, MCE_SLB_ERROR_MULTIHIT, /* Before PARITY */ 428 MCE_ECLASS_SOFT_INDETERMINATE, 429 MCE_INITIATOR_CPU, MCE_SEV_WARNING, true }, 430{ 0x00000100, true, 431 MCE_ERROR_TYPE_SLB, MCE_SLB_ERROR_PARITY, MCE_ECLASS_HARD_INDETERMINATE, 432 MCE_INITIATOR_CPU, MCE_SEV_SEVERE, true }, 433{ 0x00000040, true, 434 MCE_ERROR_TYPE_RA, MCE_RA_ERROR_LOAD, MCE_ECLASS_HARDWARE, 435 MCE_INITIATOR_CPU, MCE_SEV_SEVERE, true }, 436{ 0x00000020, false, 437 MCE_ERROR_TYPE_RA, MCE_RA_ERROR_PAGE_TABLE_WALK_LOAD_STORE, 438 MCE_ECLASS_HARDWARE, 439 MCE_INITIATOR_CPU, MCE_SEV_SEVERE, true }, 440{ 0x00000010, false, 441 MCE_ERROR_TYPE_RA, MCE_RA_ERROR_PAGE_TABLE_WALK_LOAD_STORE_FOREIGN, 442 MCE_ECLASS_HARDWARE, 443 MCE_INITIATOR_CPU, MCE_SEV_SEVERE, true }, 444{ 0x00000008, false, 445 MCE_ERROR_TYPE_RA, MCE_RA_ERROR_LOAD_STORE_FOREIGN, MCE_ECLASS_HARDWARE, 446 MCE_INITIATOR_CPU, MCE_SEV_SEVERE, true }, 447{ 0, false, 0, 0, 0, 0, 0 } }; 448 449static int mce_find_instr_ea_and_phys(struct pt_regs *regs, uint64_t *addr, 450 uint64_t *phys_addr) 451{ 452 /* 453 * Carefully look at the NIP to determine 454 * the instruction to analyse. Reading the NIP 455 * in real-mode is tricky and can lead to recursive 456 * faults 457 */ 458 ppc_inst_t instr; 459 unsigned long pfn, instr_addr; 460 struct instruction_op op; 461 struct pt_regs tmp = *regs; 462 463 pfn = addr_to_pfn(regs, regs->nip); 464 if (pfn != ULONG_MAX) { 465 instr_addr = (pfn << PAGE_SHIFT) + (regs->nip & ~PAGE_MASK); 466 instr = ppc_inst_read((u32 *)instr_addr); 467 if (!analyse_instr(&op, &tmp, instr)) { 468 pfn = addr_to_pfn(regs, op.ea); 469 *addr = op.ea; 470 *phys_addr = (pfn << PAGE_SHIFT); 471 return 0; 472 } 473 /* 474 * analyse_instr() might fail if the instruction 475 * is not a load/store, although this is unexpected 476 * for load/store errors or if we got the NIP 477 * wrong 478 */ 479 } 480 *addr = 0; 481 return -1; 482} 483 484static int mce_handle_ierror(struct pt_regs *regs, unsigned long srr1, 485 const struct mce_ierror_table table[], 486 struct mce_error_info *mce_err, uint64_t *addr, 487 uint64_t *phys_addr) 488{ 489 int handled = 0; 490 int i; 491 492 *addr = 0; 493 494 for (i = 0; table[i].srr1_mask; i++) { 495 if ((srr1 & table[i].srr1_mask) != table[i].srr1_value) 496 continue; 497 498 if (!mce_in_guest()) { 499 /* attempt to correct the error */ 500 switch (table[i].error_type) { 501 case MCE_ERROR_TYPE_SLB: 502#ifdef CONFIG_PPC_64S_HASH_MMU 503 if (local_paca->in_mce == 1) 504 slb_save_contents(local_paca->mce_faulty_slbs); 505#endif 506 handled = mce_flush(MCE_FLUSH_SLB); 507 break; 508 case MCE_ERROR_TYPE_ERAT: 509 handled = mce_flush(MCE_FLUSH_ERAT); 510 break; 511 case MCE_ERROR_TYPE_TLB: 512 handled = mce_flush(MCE_FLUSH_TLB); 513 break; 514 } 515 } 516 517 /* now fill in mce_error_info */ 518 mce_err->error_type = table[i].error_type; 519 mce_err->error_class = table[i].error_class; 520 switch (table[i].error_type) { 521 case MCE_ERROR_TYPE_UE: 522 mce_err->u.ue_error_type = table[i].error_subtype; 523 break; 524 case MCE_ERROR_TYPE_SLB: 525 mce_err->u.slb_error_type = table[i].error_subtype; 526 break; 527 case MCE_ERROR_TYPE_ERAT: 528 mce_err->u.erat_error_type = table[i].error_subtype; 529 break; 530 case MCE_ERROR_TYPE_TLB: 531 mce_err->u.tlb_error_type = table[i].error_subtype; 532 break; 533 case MCE_ERROR_TYPE_USER: 534 mce_err->u.user_error_type = table[i].error_subtype; 535 break; 536 case MCE_ERROR_TYPE_RA: 537 mce_err->u.ra_error_type = table[i].error_subtype; 538 break; 539 case MCE_ERROR_TYPE_LINK: 540 mce_err->u.link_error_type = table[i].error_subtype; 541 break; 542 } 543 mce_err->sync_error = table[i].sync_error; 544 mce_err->severity = table[i].severity; 545 mce_err->initiator = table[i].initiator; 546 if (table[i].nip_valid && !mce_in_guest()) { 547 *addr = regs->nip; 548 if (mce_err->sync_error && 549 table[i].error_type == MCE_ERROR_TYPE_UE) { 550 unsigned long pfn; 551 552 if (get_paca()->in_mce < MAX_MCE_DEPTH) { 553 pfn = addr_to_pfn(regs, regs->nip); 554 if (pfn != ULONG_MAX) { 555 *phys_addr = 556 (pfn << PAGE_SHIFT); 557 } 558 } 559 } 560 } 561 return handled; 562 } 563 564 mce_err->error_type = MCE_ERROR_TYPE_UNKNOWN; 565 mce_err->error_class = MCE_ECLASS_UNKNOWN; 566 mce_err->severity = MCE_SEV_SEVERE; 567 mce_err->initiator = MCE_INITIATOR_CPU; 568 mce_err->sync_error = true; 569 570 return 0; 571} 572 573static int mce_handle_derror(struct pt_regs *regs, 574 const struct mce_derror_table table[], 575 struct mce_error_info *mce_err, uint64_t *addr, 576 uint64_t *phys_addr) 577{ 578 uint64_t dsisr = regs->dsisr; 579 int handled = 0; 580 int found = 0; 581 int i; 582 583 *addr = 0; 584 585 for (i = 0; table[i].dsisr_value; i++) { 586 if (!(dsisr & table[i].dsisr_value)) 587 continue; 588 589 if (!mce_in_guest()) { 590 /* attempt to correct the error */ 591 switch (table[i].error_type) { 592 case MCE_ERROR_TYPE_SLB: 593#ifdef CONFIG_PPC_64S_HASH_MMU 594 if (local_paca->in_mce == 1) 595 slb_save_contents(local_paca->mce_faulty_slbs); 596#endif 597 if (mce_flush(MCE_FLUSH_SLB)) 598 handled = 1; 599 break; 600 case MCE_ERROR_TYPE_ERAT: 601 if (mce_flush(MCE_FLUSH_ERAT)) 602 handled = 1; 603 break; 604 case MCE_ERROR_TYPE_TLB: 605 if (mce_flush(MCE_FLUSH_TLB)) 606 handled = 1; 607 break; 608 } 609 } 610 611 /* 612 * Attempt to handle multiple conditions, but only return 613 * one. Ensure uncorrectable errors are first in the table 614 * to match. 615 */ 616 if (found) 617 continue; 618 619 /* now fill in mce_error_info */ 620 mce_err->error_type = table[i].error_type; 621 mce_err->error_class = table[i].error_class; 622 switch (table[i].error_type) { 623 case MCE_ERROR_TYPE_UE: 624 mce_err->u.ue_error_type = table[i].error_subtype; 625 break; 626 case MCE_ERROR_TYPE_SLB: 627 mce_err->u.slb_error_type = table[i].error_subtype; 628 break; 629 case MCE_ERROR_TYPE_ERAT: 630 mce_err->u.erat_error_type = table[i].error_subtype; 631 break; 632 case MCE_ERROR_TYPE_TLB: 633 mce_err->u.tlb_error_type = table[i].error_subtype; 634 break; 635 case MCE_ERROR_TYPE_USER: 636 mce_err->u.user_error_type = table[i].error_subtype; 637 break; 638 case MCE_ERROR_TYPE_RA: 639 mce_err->u.ra_error_type = table[i].error_subtype; 640 break; 641 case MCE_ERROR_TYPE_LINK: 642 mce_err->u.link_error_type = table[i].error_subtype; 643 break; 644 } 645 mce_err->sync_error = table[i].sync_error; 646 mce_err->severity = table[i].severity; 647 mce_err->initiator = table[i].initiator; 648 if (table[i].dar_valid) 649 *addr = regs->dar; 650 else if (mce_err->sync_error && !mce_in_guest() && 651 table[i].error_type == MCE_ERROR_TYPE_UE) { 652 /* 653 * We do a maximum of 4 nested MCE calls, see 654 * kernel/exception-64s.h 655 */ 656 if (get_paca()->in_mce < MAX_MCE_DEPTH) 657 mce_find_instr_ea_and_phys(regs, addr, 658 phys_addr); 659 } 660 found = 1; 661 } 662 663 if (found) 664 return handled; 665 666 mce_err->error_type = MCE_ERROR_TYPE_UNKNOWN; 667 mce_err->error_class = MCE_ECLASS_UNKNOWN; 668 mce_err->severity = MCE_SEV_SEVERE; 669 mce_err->initiator = MCE_INITIATOR_CPU; 670 mce_err->sync_error = true; 671 672 return 0; 673} 674 675static long mce_handle_ue_error(struct pt_regs *regs, 676 struct mce_error_info *mce_err) 677{ 678 if (mce_in_guest()) 679 return 0; 680 681 mce_common_process_ue(regs, mce_err); 682 if (mce_err->ignore_event) 683 return 1; 684 685 /* 686 * On specific SCOM read via MMIO we may get a machine check 687 * exception with SRR0 pointing inside opal. If that is the 688 * case OPAL may have recovery address to re-read SCOM data in 689 * different way and hence we can recover from this MC. 690 */ 691 692 if (ppc_md.mce_check_early_recovery) { 693 if (ppc_md.mce_check_early_recovery(regs)) 694 return 1; 695 } 696 697 return 0; 698} 699 700static long mce_handle_error(struct pt_regs *regs, 701 unsigned long srr1, 702 const struct mce_derror_table dtable[], 703 const struct mce_ierror_table itable[]) 704{ 705 struct mce_error_info mce_err = { 0 }; 706 uint64_t addr, phys_addr = ULONG_MAX; 707 long handled; 708 709 if (SRR1_MC_LOADSTORE(srr1)) 710 handled = mce_handle_derror(regs, dtable, &mce_err, &addr, 711 &phys_addr); 712 else 713 handled = mce_handle_ierror(regs, srr1, itable, &mce_err, &addr, 714 &phys_addr); 715 716 if (!handled && mce_err.error_type == MCE_ERROR_TYPE_UE) 717 handled = mce_handle_ue_error(regs, &mce_err); 718 719 save_mce_event(regs, handled, &mce_err, regs->nip, addr, phys_addr); 720 721 return handled; 722} 723 724long __machine_check_early_realmode_p7(struct pt_regs *regs) 725{ 726 /* P7 DD1 leaves top bits of DSISR undefined */ 727 regs->dsisr &= 0x0000ffff; 728 729 return mce_handle_error(regs, regs->msr, 730 mce_p7_derror_table, mce_p7_ierror_table); 731} 732 733long __machine_check_early_realmode_p8(struct pt_regs *regs) 734{ 735 return mce_handle_error(regs, regs->msr, 736 mce_p8_derror_table, mce_p8_ierror_table); 737} 738 739long __machine_check_early_realmode_p9(struct pt_regs *regs) 740{ 741 unsigned long srr1 = regs->msr; 742 743 /* 744 * On POWER9 DD2.1 and below, it's possible to get a machine check 745 * caused by a paste instruction where only DSISR bit 25 is set. This 746 * will result in the MCE handler seeing an unknown event and the kernel 747 * crashing. An MCE that occurs like this is spurious, so we don't need 748 * to do anything in terms of servicing it. If there is something that 749 * needs to be serviced, the CPU will raise the MCE again with the 750 * correct DSISR so that it can be serviced properly. So detect this 751 * case and mark it as handled. 752 */ 753 if (SRR1_MC_LOADSTORE(regs->msr) && regs->dsisr == 0x02000000) 754 return 1; 755 756 /* 757 * Async machine check due to bad real address from store or foreign 758 * link time out comes with the load/store bit (PPC bit 42) set in 759 * SRR1, but the cause comes in SRR1 not DSISR. Clear bit 42 so we're 760 * directed to the ierror table so it will find the cause (which 761 * describes it correctly as a store error). 762 */ 763 if (SRR1_MC_LOADSTORE(srr1) && 764 ((srr1 & 0x081c0000) == 0x08140000 || 765 (srr1 & 0x081c0000) == 0x08180000)) { 766 srr1 &= ~PPC_BIT(42); 767 } 768 769 return mce_handle_error(regs, srr1, 770 mce_p9_derror_table, mce_p9_ierror_table); 771} 772 773long __machine_check_early_realmode_p10(struct pt_regs *regs) 774{ 775 unsigned long srr1 = regs->msr; 776 777 /* 778 * Async machine check due to bad real address from store comes with 779 * the load/store bit (PPC bit 42) set in SRR1, but the cause comes in 780 * SRR1 not DSISR. Clear bit 42 so we're directed to the ierror table 781 * so it will find the cause (which describes it correctly as a store 782 * error). 783 */ 784 if (SRR1_MC_LOADSTORE(srr1) && 785 (srr1 & 0x081c0000) == 0x08140000) { 786 srr1 &= ~PPC_BIT(42); 787 } 788 789 return mce_handle_error(regs, srr1, 790 mce_p10_derror_table, mce_p10_ierror_table); 791}