1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Machine check handler 4 * 5 * Copyright IBM Corp. 2000, 2009 6 * Author(s): Ingo Adlung <adlung@de.ibm.com>, 7 * Martin Schwidefsky <schwidefsky@de.ibm.com>, 8 * Cornelia Huck <cornelia.huck@de.ibm.com>, 9 * Heiko Carstens <heiko.carstens@de.ibm.com>, 10 */ 11 12 #include <linux/kernel_stat.h> 13 #include <linux/init.h> 14 #include <linux/errno.h> 15 #include <linux/hardirq.h> 16 #include <linux/log2.h> 17 #include <linux/kprobes.h> 18 #include <linux/kmemleak.h> 19 #include <linux/time.h> 20 #include <linux/module.h> 21 #include <linux/sched/signal.h> 22 23 #include <linux/export.h> 24 #include <asm/lowcore.h> 25 #include <asm/smp.h> 26 #include <asm/stp.h> 27 #include <asm/cputime.h> 28 #include <asm/nmi.h> 29 #include <asm/crw.h> 30 #include <asm/switch_to.h> 31 #include <asm/ctl_reg.h> 32 #include <asm/asm-offsets.h> 33 #include <linux/kvm_host.h> 34 35 struct mcck_struct { 36 unsigned int kill_task : 1; 37 unsigned int channel_report : 1; 38 unsigned int warning : 1; 39 unsigned int stp_queue : 1; 40 unsigned long mcck_code; 41 }; 42 43 static DEFINE_PER_CPU(struct mcck_struct, cpu_mcck); 44 static struct kmem_cache *mcesa_cache; 45 static unsigned long mcesa_origin_lc; 46 47 static inline int nmi_needs_mcesa(void) 48 { 49 return MACHINE_HAS_VX || MACHINE_HAS_GS; 50 } 51 52 static inline unsigned long nmi_get_mcesa_size(void) 53 { 54 if (MACHINE_HAS_GS) 55 return MCESA_MAX_SIZE; 56 return MCESA_MIN_SIZE; 57 } 58 59 /* 60 * The initial machine check extended save area for the boot CPU. 61 * It will be replaced by nmi_init() with an allocated structure. 62 * The structure is required for machine check happening early in 63 * the boot process. 64 */ 65 static struct mcesa boot_mcesa __initdata __aligned(MCESA_MAX_SIZE); 66 67 void __init nmi_alloc_boot_cpu(struct lowcore *lc) 68 { 69 if (!nmi_needs_mcesa()) 70 return; 71 lc->mcesad = (unsigned long) &boot_mcesa; 72 if (MACHINE_HAS_GS) 73 lc->mcesad |= ilog2(MCESA_MAX_SIZE); 74 } 75 76 static int __init nmi_init(void) 77 { 78 unsigned long origin, cr0, size; 79 80 if (!nmi_needs_mcesa()) 81 return 0; 82 size = nmi_get_mcesa_size(); 83 if (size > MCESA_MIN_SIZE) 84 mcesa_origin_lc = ilog2(size); 85 /* create slab cache for the machine-check-extended-save-areas */ 86 mcesa_cache = kmem_cache_create("nmi_save_areas", size, size, 0, NULL); 87 if (!mcesa_cache) 88 panic("Couldn't create nmi save area cache"); 89 origin = (unsigned long) kmem_cache_alloc(mcesa_cache, GFP_KERNEL); 90 if (!origin) 91 panic("Couldn't allocate nmi save area"); 92 /* The pointer is stored with mcesa_bits ORed in */ 93 kmemleak_not_leak((void *) origin); 94 __ctl_store(cr0, 0, 0); 95 __ctl_clear_bit(0, 28); /* disable lowcore protection */ 96 /* Replace boot_mcesa on the boot CPU */ 97 S390_lowcore.mcesad = origin | mcesa_origin_lc; 98 __ctl_load(cr0, 0, 0); 99 return 0; 100 } 101 early_initcall(nmi_init); 102 103 int nmi_alloc_per_cpu(struct lowcore *lc) 104 { 105 unsigned long origin; 106 107 if (!nmi_needs_mcesa()) 108 return 0; 109 origin = (unsigned long) kmem_cache_alloc(mcesa_cache, GFP_KERNEL); 110 if (!origin) 111 return -ENOMEM; 112 /* The pointer is stored with mcesa_bits ORed in */ 113 kmemleak_not_leak((void *) origin); 114 lc->mcesad = origin | mcesa_origin_lc; 115 return 0; 116 } 117 118 void nmi_free_per_cpu(struct lowcore *lc) 119 { 120 if (!nmi_needs_mcesa()) 121 return; 122 kmem_cache_free(mcesa_cache, (void *)(lc->mcesad & MCESA_ORIGIN_MASK)); 123 } 124 125 static notrace void s390_handle_damage(void) 126 { 127 smp_emergency_stop(); 128 disabled_wait(); 129 while (1); 130 } 131 NOKPROBE_SYMBOL(s390_handle_damage); 132 133 /* 134 * Main machine check handler function. Will be called with interrupts disabled 135 * and machine checks enabled. 136 */ 137 void __s390_handle_mcck(void) 138 { 139 struct mcck_struct mcck; 140 141 /* 142 * Disable machine checks and get the current state of accumulated 143 * machine checks. Afterwards delete the old state and enable machine 144 * checks again. 145 */ 146 local_mcck_disable(); 147 mcck = *this_cpu_ptr(&cpu_mcck); 148 memset(this_cpu_ptr(&cpu_mcck), 0, sizeof(mcck)); 149 local_mcck_enable(); 150 151 if (mcck.channel_report) 152 crw_handle_channel_report(); 153 /* 154 * A warning may remain for a prolonged period on the bare iron. 155 * (actually until the machine is powered off, or the problem is gone) 156 * So we just stop listening for the WARNING MCH and avoid continuously 157 * being interrupted. One caveat is however, that we must do this per 158 * processor and cannot use the smp version of ctl_clear_bit(). 159 * On VM we only get one interrupt per virtally presented machinecheck. 160 * Though one suffices, we may get one interrupt per (virtual) cpu. 161 */ 162 if (mcck.warning) { /* WARNING pending ? */ 163 static int mchchk_wng_posted = 0; 164 165 /* Use single cpu clear, as we cannot handle smp here. */ 166 __ctl_clear_bit(14, 24); /* Disable WARNING MCH */ 167 if (xchg(&mchchk_wng_posted, 1) == 0) 168 kill_cad_pid(SIGPWR, 1); 169 } 170 if (mcck.stp_queue) 171 stp_queue_work(); 172 if (mcck.kill_task) { 173 local_irq_enable(); 174 printk(KERN_EMERG "mcck: Terminating task because of machine " 175 "malfunction (code 0x%016lx).\n", mcck.mcck_code); 176 printk(KERN_EMERG "mcck: task: %s, pid: %d.\n", 177 current->comm, current->pid); 178 do_exit(SIGSEGV); 179 } 180 } 181 182 void noinstr s390_handle_mcck(void) 183 { 184 trace_hardirqs_off(); 185 __s390_handle_mcck(); 186 trace_hardirqs_on(); 187 } 188 /* 189 * returns 0 if all required registers are available 190 * returns 1 otherwise 191 */ 192 static int notrace s390_check_registers(union mci mci, int umode) 193 { 194 union ctlreg2 cr2; 195 int kill_task; 196 197 kill_task = 0; 198 199 if (!mci.gr) { 200 /* 201 * General purpose registers couldn't be restored and have 202 * unknown contents. Stop system or terminate process. 203 */ 204 if (!umode) 205 s390_handle_damage(); 206 kill_task = 1; 207 } 208 /* Check control registers */ 209 if (!mci.cr) { 210 /* 211 * Control registers have unknown contents. 212 * Can't recover and therefore stopping machine. 213 */ 214 s390_handle_damage(); 215 } 216 if (!mci.fp) { 217 /* 218 * Floating point registers can't be restored. If the 219 * kernel currently uses floating point registers the 220 * system is stopped. If the process has its floating 221 * pointer registers loaded it is terminated. 222 */ 223 if (S390_lowcore.fpu_flags & KERNEL_VXR_V0V7) 224 s390_handle_damage(); 225 if (!test_cpu_flag(CIF_FPU)) 226 kill_task = 1; 227 } 228 if (!mci.fc) { 229 /* 230 * Floating point control register can't be restored. 231 * If the kernel currently uses the floating pointer 232 * registers and needs the FPC register the system is 233 * stopped. If the process has its floating pointer 234 * registers loaded it is terminated. 235 */ 236 if (S390_lowcore.fpu_flags & KERNEL_FPC) 237 s390_handle_damage(); 238 if (!test_cpu_flag(CIF_FPU)) 239 kill_task = 1; 240 } 241 242 if (MACHINE_HAS_VX) { 243 if (!mci.vr) { 244 /* 245 * Vector registers can't be restored. If the kernel 246 * currently uses vector registers the system is 247 * stopped. If the process has its vector registers 248 * loaded it is terminated. 249 */ 250 if (S390_lowcore.fpu_flags & KERNEL_VXR) 251 s390_handle_damage(); 252 if (!test_cpu_flag(CIF_FPU)) 253 kill_task = 1; 254 } 255 } 256 /* Check if access registers are valid */ 257 if (!mci.ar) { 258 /* 259 * Access registers have unknown contents. 260 * Terminating task. 261 */ 262 kill_task = 1; 263 } 264 /* Check guarded storage registers */ 265 cr2.val = S390_lowcore.cregs_save_area[2]; 266 if (cr2.gse) { 267 if (!mci.gs) { 268 /* 269 * Guarded storage register can't be restored and 270 * the current processes uses guarded storage. 271 * It has to be terminated. 272 */ 273 kill_task = 1; 274 } 275 } 276 /* Check if old PSW is valid */ 277 if (!mci.wp) { 278 /* 279 * Can't tell if we come from user or kernel mode 280 * -> stopping machine. 281 */ 282 s390_handle_damage(); 283 } 284 /* Check for invalid kernel instruction address */ 285 if (!mci.ia && !umode) { 286 /* 287 * The instruction address got lost while running 288 * in the kernel -> stopping machine. 289 */ 290 s390_handle_damage(); 291 } 292 293 if (!mci.ms || !mci.pm || !mci.ia) 294 kill_task = 1; 295 296 return kill_task; 297 } 298 NOKPROBE_SYMBOL(s390_check_registers); 299 300 /* 301 * Backup the guest's machine check info to its description block 302 */ 303 static void notrace s390_backup_mcck_info(struct pt_regs *regs) 304 { 305 struct mcck_volatile_info *mcck_backup; 306 struct sie_page *sie_page; 307 308 /* r14 contains the sie block, which was set in sie64a */ 309 struct kvm_s390_sie_block *sie_block = 310 (struct kvm_s390_sie_block *) regs->gprs[14]; 311 312 if (sie_block == NULL) 313 /* Something's seriously wrong, stop system. */ 314 s390_handle_damage(); 315 316 sie_page = container_of(sie_block, struct sie_page, sie_block); 317 mcck_backup = &sie_page->mcck_info; 318 mcck_backup->mcic = S390_lowcore.mcck_interruption_code & 319 ~(MCCK_CODE_CP | MCCK_CODE_EXT_DAMAGE); 320 mcck_backup->ext_damage_code = S390_lowcore.external_damage_code; 321 mcck_backup->failing_storage_address 322 = S390_lowcore.failing_storage_address; 323 } 324 NOKPROBE_SYMBOL(s390_backup_mcck_info); 325 326 #define MAX_IPD_COUNT 29 327 #define MAX_IPD_TIME (5 * 60 * USEC_PER_SEC) /* 5 minutes */ 328 329 #define ED_STP_ISLAND 6 /* External damage STP island check */ 330 #define ED_STP_SYNC 7 /* External damage STP sync check */ 331 332 #define MCCK_CODE_NO_GUEST (MCCK_CODE_CP | MCCK_CODE_EXT_DAMAGE) 333 334 /* 335 * machine check handler. 336 */ 337 int notrace s390_do_machine_check(struct pt_regs *regs) 338 { 339 static int ipd_count; 340 static DEFINE_SPINLOCK(ipd_lock); 341 static unsigned long long last_ipd; 342 struct mcck_struct *mcck; 343 unsigned long long tmp; 344 union mci mci; 345 unsigned long mcck_dam_code; 346 int mcck_pending = 0; 347 348 nmi_enter(); 349 350 if (user_mode(regs)) 351 update_timer_mcck(); 352 inc_irq_stat(NMI_NMI); 353 mci.val = S390_lowcore.mcck_interruption_code; 354 mcck = this_cpu_ptr(&cpu_mcck); 355 356 if (mci.sd) { 357 /* System damage -> stopping machine */ 358 s390_handle_damage(); 359 } 360 361 /* 362 * Reinject the instruction processing damages' machine checks 363 * including Delayed Access Exception into the guest 364 * instead of damaging the host if they happen in the guest. 365 */ 366 if (mci.pd && !test_cpu_flag(CIF_MCCK_GUEST)) { 367 if (mci.b) { 368 /* Processing backup -> verify if we can survive this */ 369 u64 z_mcic, o_mcic, t_mcic; 370 z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<29); 371 o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 | 372 1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 | 373 1ULL<<30 | 1ULL<<21 | 1ULL<<20 | 1ULL<<17 | 374 1ULL<<16); 375 t_mcic = mci.val; 376 377 if (((t_mcic & z_mcic) != 0) || 378 ((t_mcic & o_mcic) != o_mcic)) { 379 s390_handle_damage(); 380 } 381 382 /* 383 * Nullifying exigent condition, therefore we might 384 * retry this instruction. 385 */ 386 spin_lock(&ipd_lock); 387 tmp = get_tod_clock(); 388 if (((tmp - last_ipd) >> 12) < MAX_IPD_TIME) 389 ipd_count++; 390 else 391 ipd_count = 1; 392 last_ipd = tmp; 393 if (ipd_count == MAX_IPD_COUNT) 394 s390_handle_damage(); 395 spin_unlock(&ipd_lock); 396 } else { 397 /* Processing damage -> stopping machine */ 398 s390_handle_damage(); 399 } 400 } 401 if (s390_check_registers(mci, user_mode(regs))) { 402 /* 403 * Couldn't restore all register contents for the 404 * user space process -> mark task for termination. 405 */ 406 mcck->kill_task = 1; 407 mcck->mcck_code = mci.val; 408 mcck_pending = 1; 409 } 410 411 /* 412 * Backup the machine check's info if it happens when the guest 413 * is running. 414 */ 415 if (test_cpu_flag(CIF_MCCK_GUEST)) 416 s390_backup_mcck_info(regs); 417 418 if (mci.cd) { 419 /* Timing facility damage */ 420 s390_handle_damage(); 421 } 422 if (mci.ed && mci.ec) { 423 /* External damage */ 424 if (S390_lowcore.external_damage_code & (1U << ED_STP_SYNC)) 425 mcck->stp_queue |= stp_sync_check(); 426 if (S390_lowcore.external_damage_code & (1U << ED_STP_ISLAND)) 427 mcck->stp_queue |= stp_island_check(); 428 mcck_pending = 1; 429 } 430 431 /* 432 * Reinject storage related machine checks into the guest if they 433 * happen when the guest is running. 434 */ 435 if (!test_cpu_flag(CIF_MCCK_GUEST)) { 436 if (mci.se) 437 /* Storage error uncorrected */ 438 s390_handle_damage(); 439 if (mci.ke) 440 /* Storage key-error uncorrected */ 441 s390_handle_damage(); 442 if (mci.ds && mci.fa) 443 /* Storage degradation */ 444 s390_handle_damage(); 445 } 446 if (mci.cp) { 447 /* Channel report word pending */ 448 mcck->channel_report = 1; 449 mcck_pending = 1; 450 } 451 if (mci.w) { 452 /* Warning pending */ 453 mcck->warning = 1; 454 mcck_pending = 1; 455 } 456 457 /* 458 * If there are only Channel Report Pending and External Damage 459 * machine checks, they will not be reinjected into the guest 460 * because they refer to host conditions only. 461 */ 462 mcck_dam_code = (mci.val & MCIC_SUBCLASS_MASK); 463 if (test_cpu_flag(CIF_MCCK_GUEST) && 464 (mcck_dam_code & MCCK_CODE_NO_GUEST) != mcck_dam_code) { 465 /* Set exit reason code for host's later handling */ 466 *((long *)(regs->gprs[15] + __SF_SIE_REASON)) = -EINTR; 467 } 468 clear_cpu_flag(CIF_MCCK_GUEST); 469 470 if (user_mode(regs) && mcck_pending) { 471 nmi_exit(); 472 return 1; 473 } 474 475 if (mcck_pending) 476 schedule_mcck_handler(); 477 478 nmi_exit(); 479 return 0; 480 } 481 NOKPROBE_SYMBOL(s390_do_machine_check); 482 483 static int __init machine_check_init(void) 484 { 485 ctl_set_bit(14, 25); /* enable external damage MCH */ 486 ctl_set_bit(14, 27); /* enable system recovery MCH */ 487 ctl_set_bit(14, 24); /* enable warning MCH */ 488 return 0; 489 } 490 early_initcall(machine_check_init); 491