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 */ 10 11 #include <linux/kernel_stat.h> 12 #include <linux/init.h> 13 #include <linux/errno.h> 14 #include <linux/entry-common.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 #include <linux/kvm_host.h> 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 <asm/pai.h> 34 #include <asm/vx-insn.h> 35 36 struct mcck_struct { 37 unsigned int kill_task : 1; 38 unsigned int channel_report : 1; 39 unsigned int warning : 1; 40 unsigned int stp_queue : 1; 41 unsigned long mcck_code; 42 }; 43 44 static DEFINE_PER_CPU(struct mcck_struct, cpu_mcck); 45 46 static inline int nmi_needs_mcesa(void) 47 { 48 return MACHINE_HAS_VX || MACHINE_HAS_GS; 49 } 50 51 /* 52 * The initial machine check extended save area for the boot CPU. 53 * It will be replaced on the boot CPU reinit with an allocated 54 * structure. The structure is required for machine check happening 55 * early in the boot process. 56 */ 57 static struct mcesa boot_mcesa __aligned(MCESA_MAX_SIZE); 58 59 void __init nmi_alloc_mcesa_early(u64 *mcesad) 60 { 61 if (!nmi_needs_mcesa()) 62 return; 63 *mcesad = __pa(&boot_mcesa); 64 if (MACHINE_HAS_GS) 65 *mcesad |= ilog2(MCESA_MAX_SIZE); 66 } 67 68 int nmi_alloc_mcesa(u64 *mcesad) 69 { 70 unsigned long size; 71 void *origin; 72 73 *mcesad = 0; 74 if (!nmi_needs_mcesa()) 75 return 0; 76 size = MACHINE_HAS_GS ? MCESA_MAX_SIZE : MCESA_MIN_SIZE; 77 origin = kmalloc(size, GFP_KERNEL); 78 if (!origin) 79 return -ENOMEM; 80 /* The pointer is stored with mcesa_bits ORed in */ 81 kmemleak_not_leak(origin); 82 *mcesad = __pa(origin); 83 if (MACHINE_HAS_GS) 84 *mcesad |= ilog2(MCESA_MAX_SIZE); 85 return 0; 86 } 87 88 void nmi_free_mcesa(u64 *mcesad) 89 { 90 if (!nmi_needs_mcesa()) 91 return; 92 kfree(__va(*mcesad & MCESA_ORIGIN_MASK)); 93 } 94 95 static __always_inline char *nmi_puts(char *dest, const char *src) 96 { 97 while (*src) 98 *dest++ = *src++; 99 *dest = 0; 100 return dest; 101 } 102 103 static __always_inline char *u64_to_hex(char *dest, u64 val) 104 { 105 int i, num; 106 107 for (i = 1; i <= 16; i++) { 108 num = (val >> (64 - 4 * i)) & 0xf; 109 if (num >= 10) 110 *dest++ = 'A' + num - 10; 111 else 112 *dest++ = '0' + num; 113 } 114 *dest = 0; 115 return dest; 116 } 117 118 static notrace void s390_handle_damage(void) 119 { 120 union ctlreg0 cr0, cr0_new; 121 char message[100]; 122 psw_t psw_save; 123 char *ptr; 124 125 smp_emergency_stop(); 126 diag_amode31_ops.diag308_reset(); 127 ptr = nmi_puts(message, "System stopped due to unrecoverable machine check, code: 0x"); 128 u64_to_hex(ptr, S390_lowcore.mcck_interruption_code); 129 130 /* 131 * Disable low address protection and make machine check new PSW a 132 * disabled wait PSW. Any additional machine check cannot be handled. 133 */ 134 __ctl_store(cr0.val, 0, 0); 135 cr0_new = cr0; 136 cr0_new.lap = 0; 137 __ctl_load(cr0_new.val, 0, 0); 138 psw_save = S390_lowcore.mcck_new_psw; 139 psw_bits(S390_lowcore.mcck_new_psw).io = 0; 140 psw_bits(S390_lowcore.mcck_new_psw).ext = 0; 141 psw_bits(S390_lowcore.mcck_new_psw).wait = 1; 142 sclp_emergency_printk(message); 143 144 /* 145 * Restore machine check new PSW and control register 0 to original 146 * values. This makes possible system dump analysis easier. 147 */ 148 S390_lowcore.mcck_new_psw = psw_save; 149 __ctl_load(cr0.val, 0, 0); 150 disabled_wait(); 151 while (1); 152 } 153 NOKPROBE_SYMBOL(s390_handle_damage); 154 155 /* 156 * Main machine check handler function. Will be called with interrupts disabled 157 * and machine checks enabled. 158 */ 159 void __s390_handle_mcck(void) 160 { 161 struct mcck_struct mcck; 162 163 /* 164 * Disable machine checks and get the current state of accumulated 165 * machine checks. Afterwards delete the old state and enable machine 166 * checks again. 167 */ 168 local_mcck_disable(); 169 mcck = *this_cpu_ptr(&cpu_mcck); 170 memset(this_cpu_ptr(&cpu_mcck), 0, sizeof(mcck)); 171 local_mcck_enable(); 172 173 if (mcck.channel_report) 174 crw_handle_channel_report(); 175 /* 176 * A warning may remain for a prolonged period on the bare iron. 177 * (actually until the machine is powered off, or the problem is gone) 178 * So we just stop listening for the WARNING MCH and avoid continuously 179 * being interrupted. One caveat is however, that we must do this per 180 * processor and cannot use the smp version of ctl_clear_bit(). 181 * On VM we only get one interrupt per virtally presented machinecheck. 182 * Though one suffices, we may get one interrupt per (virtual) cpu. 183 */ 184 if (mcck.warning) { /* WARNING pending ? */ 185 static int mchchk_wng_posted = 0; 186 187 /* Use single cpu clear, as we cannot handle smp here. */ 188 __ctl_clear_bit(14, 24); /* Disable WARNING MCH */ 189 if (xchg(&mchchk_wng_posted, 1) == 0) 190 kill_cad_pid(SIGPWR, 1); 191 } 192 if (mcck.stp_queue) 193 stp_queue_work(); 194 if (mcck.kill_task) { 195 local_irq_enable(); 196 printk(KERN_EMERG "mcck: Terminating task because of machine " 197 "malfunction (code 0x%016lx).\n", mcck.mcck_code); 198 printk(KERN_EMERG "mcck: task: %s, pid: %d.\n", 199 current->comm, current->pid); 200 make_task_dead(SIGSEGV); 201 } 202 } 203 204 void noinstr s390_handle_mcck(struct pt_regs *regs) 205 { 206 trace_hardirqs_off(); 207 pai_kernel_enter(regs); 208 __s390_handle_mcck(); 209 pai_kernel_exit(regs); 210 trace_hardirqs_on(); 211 } 212 /* 213 * returns 0 if register contents could be validated 214 * returns 1 otherwise 215 */ 216 static int notrace s390_validate_registers(union mci mci) 217 { 218 struct mcesa *mcesa; 219 void *fpt_save_area; 220 union ctlreg2 cr2; 221 int kill_task; 222 u64 zero; 223 224 kill_task = 0; 225 zero = 0; 226 227 if (!mci.gr || !mci.fp) 228 kill_task = 1; 229 fpt_save_area = &S390_lowcore.floating_pt_save_area; 230 if (!mci.fc) { 231 kill_task = 1; 232 asm volatile( 233 " lfpc %0\n" 234 : 235 : "Q" (zero)); 236 } else { 237 asm volatile( 238 " lfpc %0\n" 239 : 240 : "Q" (S390_lowcore.fpt_creg_save_area)); 241 } 242 243 mcesa = __va(S390_lowcore.mcesad & MCESA_ORIGIN_MASK); 244 if (!MACHINE_HAS_VX) { 245 /* Validate floating point registers */ 246 asm volatile( 247 " ld 0,0(%0)\n" 248 " ld 1,8(%0)\n" 249 " ld 2,16(%0)\n" 250 " ld 3,24(%0)\n" 251 " ld 4,32(%0)\n" 252 " ld 5,40(%0)\n" 253 " ld 6,48(%0)\n" 254 " ld 7,56(%0)\n" 255 " ld 8,64(%0)\n" 256 " ld 9,72(%0)\n" 257 " ld 10,80(%0)\n" 258 " ld 11,88(%0)\n" 259 " ld 12,96(%0)\n" 260 " ld 13,104(%0)\n" 261 " ld 14,112(%0)\n" 262 " ld 15,120(%0)\n" 263 : 264 : "a" (fpt_save_area) 265 : "memory"); 266 } else { 267 /* Validate vector registers */ 268 union ctlreg0 cr0; 269 270 /* 271 * The vector validity must only be checked if not running a 272 * KVM guest. For KVM guests the machine check is forwarded by 273 * KVM and it is the responsibility of the guest to take 274 * appropriate actions. The host vector or FPU values have been 275 * saved by KVM and will be restored by KVM. 276 */ 277 if (!mci.vr && !test_cpu_flag(CIF_MCCK_GUEST)) 278 kill_task = 1; 279 cr0.val = S390_lowcore.cregs_save_area[0]; 280 cr0.afp = cr0.vx = 1; 281 __ctl_load(cr0.val, 0, 0); 282 asm volatile( 283 " la 1,%0\n" 284 " VLM 0,15,0,1\n" 285 " VLM 16,31,256,1\n" 286 : 287 : "Q" (*(struct vx_array *)mcesa->vector_save_area) 288 : "1"); 289 __ctl_load(S390_lowcore.cregs_save_area[0], 0, 0); 290 } 291 /* Validate access registers */ 292 asm volatile( 293 " lam 0,15,0(%0)\n" 294 : 295 : "a" (&S390_lowcore.access_regs_save_area) 296 : "memory"); 297 if (!mci.ar) 298 kill_task = 1; 299 /* Validate guarded storage registers */ 300 cr2.val = S390_lowcore.cregs_save_area[2]; 301 if (cr2.gse) { 302 if (!mci.gs) { 303 /* 304 * 2 cases: 305 * - machine check in kernel or userspace 306 * - machine check while running SIE (KVM guest) 307 * For kernel or userspace the userspace values of 308 * guarded storage control can not be recreated, the 309 * process must be terminated. 310 * For SIE the guest values of guarded storage can not 311 * be recreated. This is either due to a bug or due to 312 * GS being disabled in the guest. The guest will be 313 * notified by KVM code and the guests machine check 314 * handling must take care of this. The host values 315 * are saved by KVM and are not affected. 316 */ 317 if (!test_cpu_flag(CIF_MCCK_GUEST)) 318 kill_task = 1; 319 } else { 320 load_gs_cb((struct gs_cb *)mcesa->guarded_storage_save_area); 321 } 322 } 323 /* 324 * The getcpu vdso syscall reads CPU number from the programmable 325 * field of the TOD clock. Disregard the TOD programmable register 326 * validity bit and load the CPU number into the TOD programmable 327 * field unconditionally. 328 */ 329 set_tod_programmable_field(raw_smp_processor_id()); 330 /* Validate clock comparator register */ 331 set_clock_comparator(S390_lowcore.clock_comparator); 332 333 if (!mci.ms || !mci.pm || !mci.ia) 334 kill_task = 1; 335 336 return kill_task; 337 } 338 NOKPROBE_SYMBOL(s390_validate_registers); 339 340 /* 341 * Backup the guest's machine check info to its description block 342 */ 343 static void notrace s390_backup_mcck_info(struct pt_regs *regs) 344 { 345 struct mcck_volatile_info *mcck_backup; 346 struct sie_page *sie_page; 347 348 /* r14 contains the sie block, which was set in sie64a */ 349 struct kvm_s390_sie_block *sie_block = 350 (struct kvm_s390_sie_block *) regs->gprs[14]; 351 352 if (sie_block == NULL) 353 /* Something's seriously wrong, stop system. */ 354 s390_handle_damage(); 355 356 sie_page = container_of(sie_block, struct sie_page, sie_block); 357 mcck_backup = &sie_page->mcck_info; 358 mcck_backup->mcic = S390_lowcore.mcck_interruption_code & 359 ~(MCCK_CODE_CP | MCCK_CODE_EXT_DAMAGE); 360 mcck_backup->ext_damage_code = S390_lowcore.external_damage_code; 361 mcck_backup->failing_storage_address 362 = S390_lowcore.failing_storage_address; 363 } 364 NOKPROBE_SYMBOL(s390_backup_mcck_info); 365 366 #define MAX_IPD_COUNT 29 367 #define MAX_IPD_TIME (5 * 60 * USEC_PER_SEC) /* 5 minutes */ 368 369 #define ED_STP_ISLAND 6 /* External damage STP island check */ 370 #define ED_STP_SYNC 7 /* External damage STP sync check */ 371 372 #define MCCK_CODE_NO_GUEST (MCCK_CODE_CP | MCCK_CODE_EXT_DAMAGE) 373 374 /* 375 * machine check handler. 376 */ 377 int notrace s390_do_machine_check(struct pt_regs *regs) 378 { 379 static int ipd_count; 380 static DEFINE_SPINLOCK(ipd_lock); 381 static unsigned long long last_ipd; 382 struct mcck_struct *mcck; 383 unsigned long long tmp; 384 irqentry_state_t irq_state; 385 union mci mci; 386 unsigned long mcck_dam_code; 387 int mcck_pending = 0; 388 389 irq_state = irqentry_nmi_enter(regs); 390 391 if (user_mode(regs)) 392 update_timer_mcck(); 393 inc_irq_stat(NMI_NMI); 394 mci.val = S390_lowcore.mcck_interruption_code; 395 mcck = this_cpu_ptr(&cpu_mcck); 396 397 /* 398 * Reinject the instruction processing damages' machine checks 399 * including Delayed Access Exception into the guest 400 * instead of damaging the host if they happen in the guest. 401 */ 402 if (mci.pd && !test_cpu_flag(CIF_MCCK_GUEST)) { 403 if (mci.b) { 404 /* Processing backup -> verify if we can survive this */ 405 u64 z_mcic, o_mcic, t_mcic; 406 z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<29); 407 o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 | 408 1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 | 409 1ULL<<30 | 1ULL<<21 | 1ULL<<20 | 1ULL<<17 | 410 1ULL<<16); 411 t_mcic = mci.val; 412 413 if (((t_mcic & z_mcic) != 0) || 414 ((t_mcic & o_mcic) != o_mcic)) { 415 s390_handle_damage(); 416 } 417 418 /* 419 * Nullifying exigent condition, therefore we might 420 * retry this instruction. 421 */ 422 spin_lock(&ipd_lock); 423 tmp = get_tod_clock(); 424 if (((tmp - last_ipd) >> 12) < MAX_IPD_TIME) 425 ipd_count++; 426 else 427 ipd_count = 1; 428 last_ipd = tmp; 429 if (ipd_count == MAX_IPD_COUNT) 430 s390_handle_damage(); 431 spin_unlock(&ipd_lock); 432 } else { 433 /* Processing damage -> stopping machine */ 434 s390_handle_damage(); 435 } 436 } 437 if (s390_validate_registers(mci)) { 438 if (!user_mode(regs)) 439 s390_handle_damage(); 440 /* 441 * Couldn't restore all register contents for the 442 * user space process -> mark task for termination. 443 */ 444 mcck->kill_task = 1; 445 mcck->mcck_code = mci.val; 446 mcck_pending = 1; 447 } 448 449 /* 450 * Backup the machine check's info if it happens when the guest 451 * is running. 452 */ 453 if (test_cpu_flag(CIF_MCCK_GUEST)) 454 s390_backup_mcck_info(regs); 455 456 if (mci.cd) { 457 /* Timing facility damage */ 458 s390_handle_damage(); 459 } 460 if (mci.ed && mci.ec) { 461 /* External damage */ 462 if (S390_lowcore.external_damage_code & (1U << ED_STP_SYNC)) 463 mcck->stp_queue |= stp_sync_check(); 464 if (S390_lowcore.external_damage_code & (1U << ED_STP_ISLAND)) 465 mcck->stp_queue |= stp_island_check(); 466 mcck_pending = 1; 467 } 468 /* 469 * Reinject storage related machine checks into the guest if they 470 * happen when the guest is running. 471 */ 472 if (!test_cpu_flag(CIF_MCCK_GUEST)) { 473 /* Storage error uncorrected */ 474 if (mci.se) 475 s390_handle_damage(); 476 /* Storage key-error uncorrected */ 477 if (mci.ke) 478 s390_handle_damage(); 479 /* Storage degradation */ 480 if (mci.ds && mci.fa) 481 s390_handle_damage(); 482 } 483 if (mci.cp) { 484 /* Channel report word pending */ 485 mcck->channel_report = 1; 486 mcck_pending = 1; 487 } 488 if (mci.w) { 489 /* Warning pending */ 490 mcck->warning = 1; 491 mcck_pending = 1; 492 } 493 494 /* 495 * If there are only Channel Report Pending and External Damage 496 * machine checks, they will not be reinjected into the guest 497 * because they refer to host conditions only. 498 */ 499 mcck_dam_code = (mci.val & MCIC_SUBCLASS_MASK); 500 if (test_cpu_flag(CIF_MCCK_GUEST) && 501 (mcck_dam_code & MCCK_CODE_NO_GUEST) != mcck_dam_code) { 502 /* Set exit reason code for host's later handling */ 503 *((long *)(regs->gprs[15] + __SF_SIE_REASON)) = -EINTR; 504 } 505 clear_cpu_flag(CIF_MCCK_GUEST); 506 507 if (user_mode(regs) && mcck_pending) { 508 irqentry_nmi_exit(regs, irq_state); 509 return 1; 510 } 511 512 if (mcck_pending) 513 schedule_mcck_handler(); 514 515 irqentry_nmi_exit(regs, irq_state); 516 return 0; 517 } 518 NOKPROBE_SYMBOL(s390_do_machine_check); 519 520 static int __init machine_check_init(void) 521 { 522 ctl_set_bit(14, 25); /* enable external damage MCH */ 523 ctl_set_bit(14, 27); /* enable system recovery MCH */ 524 ctl_set_bit(14, 24); /* enable warning MCH */ 525 return 0; 526 } 527 early_initcall(machine_check_init); 528