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/hardirq.h> 15 #include <linux/log2.h> 16 #include <linux/kprobes.h> 17 #include <linux/kmemleak.h> 18 #include <linux/time.h> 19 #include <linux/module.h> 20 #include <linux/sched/signal.h> 21 22 #include <linux/export.h> 23 #include <asm/lowcore.h> 24 #include <asm/smp.h> 25 #include <asm/stp.h> 26 #include <asm/cputime.h> 27 #include <asm/nmi.h> 28 #include <asm/crw.h> 29 #include <asm/switch_to.h> 30 #include <asm/ctl_reg.h> 31 #include <asm/asm-offsets.h> 32 #include <linux/kvm_host.h> 33 34 struct mcck_struct { 35 unsigned int kill_task : 1; 36 unsigned int channel_report : 1; 37 unsigned int warning : 1; 38 unsigned int stp_queue : 1; 39 unsigned long mcck_code; 40 }; 41 42 static DEFINE_PER_CPU(struct mcck_struct, cpu_mcck); 43 static struct kmem_cache *mcesa_cache; 44 static unsigned long mcesa_origin_lc; 45 46 static inline int nmi_needs_mcesa(void) 47 { 48 return MACHINE_HAS_VX || MACHINE_HAS_GS; 49 } 50 51 static inline unsigned long nmi_get_mcesa_size(void) 52 { 53 if (MACHINE_HAS_GS) 54 return MCESA_MAX_SIZE; 55 return MCESA_MIN_SIZE; 56 } 57 58 /* 59 * The initial machine check extended save area for the boot CPU. 60 * It will be replaced on the boot CPU reinit with an allocated 61 * structure. The structure is required for machine check happening 62 * early in the boot process. 63 */ 64 static struct mcesa boot_mcesa __initdata __aligned(MCESA_MAX_SIZE); 65 66 void __init nmi_alloc_mcesa_early(u64 *mcesad) 67 { 68 if (!nmi_needs_mcesa()) 69 return; 70 *mcesad = __pa(&boot_mcesa); 71 if (MACHINE_HAS_GS) 72 *mcesad |= ilog2(MCESA_MAX_SIZE); 73 } 74 75 static void __init nmi_alloc_cache(void) 76 { 77 unsigned long size; 78 79 if (!nmi_needs_mcesa()) 80 return; 81 size = nmi_get_mcesa_size(); 82 if (size > MCESA_MIN_SIZE) 83 mcesa_origin_lc = ilog2(size); 84 /* create slab cache for the machine-check-extended-save-areas */ 85 mcesa_cache = kmem_cache_create("nmi_save_areas", size, size, 0, NULL); 86 if (!mcesa_cache) 87 panic("Couldn't create nmi save area cache"); 88 } 89 90 int __ref nmi_alloc_mcesa(u64 *mcesad) 91 { 92 unsigned long origin; 93 94 *mcesad = 0; 95 if (!nmi_needs_mcesa()) 96 return 0; 97 if (!mcesa_cache) 98 nmi_alloc_cache(); 99 origin = (unsigned long) kmem_cache_alloc(mcesa_cache, GFP_KERNEL); 100 if (!origin) 101 return -ENOMEM; 102 /* The pointer is stored with mcesa_bits ORed in */ 103 kmemleak_not_leak((void *) origin); 104 *mcesad = __pa(origin) | mcesa_origin_lc; 105 return 0; 106 } 107 108 void nmi_free_mcesa(u64 *mcesad) 109 { 110 if (!nmi_needs_mcesa()) 111 return; 112 kmem_cache_free(mcesa_cache, __va(*mcesad & MCESA_ORIGIN_MASK)); 113 } 114 115 static notrace void s390_handle_damage(void) 116 { 117 smp_emergency_stop(); 118 disabled_wait(); 119 while (1); 120 } 121 NOKPROBE_SYMBOL(s390_handle_damage); 122 123 /* 124 * Main machine check handler function. Will be called with interrupts disabled 125 * and machine checks enabled. 126 */ 127 void __s390_handle_mcck(void) 128 { 129 struct mcck_struct mcck; 130 131 /* 132 * Disable machine checks and get the current state of accumulated 133 * machine checks. Afterwards delete the old state and enable machine 134 * checks again. 135 */ 136 local_mcck_disable(); 137 mcck = *this_cpu_ptr(&cpu_mcck); 138 memset(this_cpu_ptr(&cpu_mcck), 0, sizeof(mcck)); 139 local_mcck_enable(); 140 141 if (mcck.channel_report) 142 crw_handle_channel_report(); 143 /* 144 * A warning may remain for a prolonged period on the bare iron. 145 * (actually until the machine is powered off, or the problem is gone) 146 * So we just stop listening for the WARNING MCH and avoid continuously 147 * being interrupted. One caveat is however, that we must do this per 148 * processor and cannot use the smp version of ctl_clear_bit(). 149 * On VM we only get one interrupt per virtally presented machinecheck. 150 * Though one suffices, we may get one interrupt per (virtual) cpu. 151 */ 152 if (mcck.warning) { /* WARNING pending ? */ 153 static int mchchk_wng_posted = 0; 154 155 /* Use single cpu clear, as we cannot handle smp here. */ 156 __ctl_clear_bit(14, 24); /* Disable WARNING MCH */ 157 if (xchg(&mchchk_wng_posted, 1) == 0) 158 kill_cad_pid(SIGPWR, 1); 159 } 160 if (mcck.stp_queue) 161 stp_queue_work(); 162 if (mcck.kill_task) { 163 local_irq_enable(); 164 printk(KERN_EMERG "mcck: Terminating task because of machine " 165 "malfunction (code 0x%016lx).\n", mcck.mcck_code); 166 printk(KERN_EMERG "mcck: task: %s, pid: %d.\n", 167 current->comm, current->pid); 168 make_task_dead(SIGSEGV); 169 } 170 } 171 172 void noinstr s390_handle_mcck(void) 173 { 174 trace_hardirqs_off(); 175 __s390_handle_mcck(); 176 trace_hardirqs_on(); 177 } 178 /* 179 * returns 0 if all required registers are available 180 * returns 1 otherwise 181 */ 182 static int notrace s390_validate_registers(union mci mci, int umode) 183 { 184 struct mcesa *mcesa; 185 void *fpt_save_area; 186 union ctlreg2 cr2; 187 int kill_task; 188 u64 zero; 189 190 kill_task = 0; 191 zero = 0; 192 193 if (!mci.gr) { 194 /* 195 * General purpose registers couldn't be restored and have 196 * unknown contents. Stop system or terminate process. 197 */ 198 if (!umode) 199 s390_handle_damage(); 200 kill_task = 1; 201 } 202 if (!mci.fp) { 203 /* 204 * Floating point registers can't be restored. If the 205 * kernel currently uses floating point registers the 206 * system is stopped. If the process has its floating 207 * pointer registers loaded it is terminated. 208 */ 209 if (S390_lowcore.fpu_flags & KERNEL_VXR_V0V7) 210 s390_handle_damage(); 211 if (!test_cpu_flag(CIF_FPU)) 212 kill_task = 1; 213 } 214 fpt_save_area = &S390_lowcore.floating_pt_save_area; 215 if (!mci.fc) { 216 /* 217 * Floating point control register can't be restored. 218 * If the kernel currently uses the floating pointer 219 * registers and needs the FPC register the system is 220 * stopped. If the process has its floating pointer 221 * registers loaded it is terminated. Otherwise the 222 * FPC is just validated. 223 */ 224 if (S390_lowcore.fpu_flags & KERNEL_FPC) 225 s390_handle_damage(); 226 asm volatile( 227 " lfpc %0\n" 228 : 229 : "Q" (zero)); 230 if (!test_cpu_flag(CIF_FPU)) 231 kill_task = 1; 232 } else { 233 asm volatile( 234 " lfpc %0\n" 235 : 236 : "Q" (S390_lowcore.fpt_creg_save_area)); 237 } 238 239 mcesa = __va(S390_lowcore.mcesad & MCESA_ORIGIN_MASK); 240 if (!MACHINE_HAS_VX) { 241 /* Validate floating point registers */ 242 asm volatile( 243 " ld 0,0(%0)\n" 244 " ld 1,8(%0)\n" 245 " ld 2,16(%0)\n" 246 " ld 3,24(%0)\n" 247 " ld 4,32(%0)\n" 248 " ld 5,40(%0)\n" 249 " ld 6,48(%0)\n" 250 " ld 7,56(%0)\n" 251 " ld 8,64(%0)\n" 252 " ld 9,72(%0)\n" 253 " ld 10,80(%0)\n" 254 " ld 11,88(%0)\n" 255 " ld 12,96(%0)\n" 256 " ld 13,104(%0)\n" 257 " ld 14,112(%0)\n" 258 " ld 15,120(%0)\n" 259 : 260 : "a" (fpt_save_area) 261 : "memory"); 262 } else { 263 /* Validate vector registers */ 264 union ctlreg0 cr0; 265 266 /* 267 * The vector validity must only be checked if not running a 268 * KVM guest. For KVM guests the machine check is forwarded by 269 * KVM and it is the responsibility of the guest to take 270 * appropriate actions. The host vector or FPU values have been 271 * saved by KVM and will be restored by KVM. 272 */ 273 if (!mci.vr && !test_cpu_flag(CIF_MCCK_GUEST)) { 274 /* 275 * Vector registers can't be restored. If the kernel 276 * currently uses vector registers the system is 277 * stopped. If the process has its vector registers 278 * loaded it is terminated. Otherwise just validate 279 * the registers. 280 */ 281 if (S390_lowcore.fpu_flags & KERNEL_VXR) 282 s390_handle_damage(); 283 if (!test_cpu_flag(CIF_FPU)) 284 kill_task = 1; 285 } 286 cr0.val = S390_lowcore.cregs_save_area[0]; 287 cr0.afp = cr0.vx = 1; 288 __ctl_load(cr0.val, 0, 0); 289 asm volatile( 290 " la 1,%0\n" 291 " .word 0xe70f,0x1000,0x0036\n" /* vlm 0,15,0(1) */ 292 " .word 0xe70f,0x1100,0x0c36\n" /* vlm 16,31,256(1) */ 293 : 294 : "Q" (*(struct vx_array *)mcesa->vector_save_area) 295 : "1"); 296 __ctl_load(S390_lowcore.cregs_save_area[0], 0, 0); 297 } 298 /* Validate access registers */ 299 asm volatile( 300 " lam 0,15,0(%0)\n" 301 : 302 : "a" (&S390_lowcore.access_regs_save_area) 303 : "memory"); 304 if (!mci.ar) { 305 /* 306 * Access registers have unknown contents. 307 * Terminating task. 308 */ 309 kill_task = 1; 310 } 311 /* Validate guarded storage registers */ 312 cr2.val = S390_lowcore.cregs_save_area[2]; 313 if (cr2.gse) { 314 if (!mci.gs) { 315 /* 316 * 2 cases: 317 * - machine check in kernel or userspace 318 * - machine check while running SIE (KVM guest) 319 * For kernel or userspace the userspace values of 320 * guarded storage control can not be recreated, the 321 * process must be terminated. 322 * For SIE the guest values of guarded storage can not 323 * be recreated. This is either due to a bug or due to 324 * GS being disabled in the guest. The guest will be 325 * notified by KVM code and the guests machine check 326 * handling must take care of this. The host values 327 * are saved by KVM and are not affected. 328 */ 329 if (!test_cpu_flag(CIF_MCCK_GUEST)) 330 kill_task = 1; 331 } else { 332 load_gs_cb((struct gs_cb *)mcesa->guarded_storage_save_area); 333 } 334 } 335 /* 336 * The getcpu vdso syscall reads CPU number from the programmable 337 * field of the TOD clock. Disregard the TOD programmable register 338 * validity bit and load the CPU number into the TOD programmable 339 * field unconditionally. 340 */ 341 set_tod_programmable_field(raw_smp_processor_id()); 342 /* Validate clock comparator register */ 343 set_clock_comparator(S390_lowcore.clock_comparator); 344 345 if (!mci.ms || !mci.pm || !mci.ia) 346 kill_task = 1; 347 348 return kill_task; 349 } 350 NOKPROBE_SYMBOL(s390_validate_registers); 351 352 /* 353 * Backup the guest's machine check info to its description block 354 */ 355 static void notrace s390_backup_mcck_info(struct pt_regs *regs) 356 { 357 struct mcck_volatile_info *mcck_backup; 358 struct sie_page *sie_page; 359 360 /* r14 contains the sie block, which was set in sie64a */ 361 struct kvm_s390_sie_block *sie_block = 362 (struct kvm_s390_sie_block *) regs->gprs[14]; 363 364 if (sie_block == NULL) 365 /* Something's seriously wrong, stop system. */ 366 s390_handle_damage(); 367 368 sie_page = container_of(sie_block, struct sie_page, sie_block); 369 mcck_backup = &sie_page->mcck_info; 370 mcck_backup->mcic = S390_lowcore.mcck_interruption_code & 371 ~(MCCK_CODE_CP | MCCK_CODE_EXT_DAMAGE); 372 mcck_backup->ext_damage_code = S390_lowcore.external_damage_code; 373 mcck_backup->failing_storage_address 374 = S390_lowcore.failing_storage_address; 375 } 376 NOKPROBE_SYMBOL(s390_backup_mcck_info); 377 378 #define MAX_IPD_COUNT 29 379 #define MAX_IPD_TIME (5 * 60 * USEC_PER_SEC) /* 5 minutes */ 380 381 #define ED_STP_ISLAND 6 /* External damage STP island check */ 382 #define ED_STP_SYNC 7 /* External damage STP sync check */ 383 384 #define MCCK_CODE_NO_GUEST (MCCK_CODE_CP | MCCK_CODE_EXT_DAMAGE) 385 386 /* 387 * machine check handler. 388 */ 389 int notrace s390_do_machine_check(struct pt_regs *regs) 390 { 391 static int ipd_count; 392 static DEFINE_SPINLOCK(ipd_lock); 393 static unsigned long long last_ipd; 394 struct mcck_struct *mcck; 395 unsigned long long tmp; 396 union mci mci; 397 unsigned long mcck_dam_code; 398 int mcck_pending = 0; 399 400 nmi_enter(); 401 402 if (user_mode(regs)) 403 update_timer_mcck(); 404 inc_irq_stat(NMI_NMI); 405 mci.val = S390_lowcore.mcck_interruption_code; 406 mcck = this_cpu_ptr(&cpu_mcck); 407 408 /* 409 * Reinject the instruction processing damages' machine checks 410 * including Delayed Access Exception into the guest 411 * instead of damaging the host if they happen in the guest. 412 */ 413 if (mci.pd && !test_cpu_flag(CIF_MCCK_GUEST)) { 414 if (mci.b) { 415 /* Processing backup -> verify if we can survive this */ 416 u64 z_mcic, o_mcic, t_mcic; 417 z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<29); 418 o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 | 419 1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 | 420 1ULL<<30 | 1ULL<<21 | 1ULL<<20 | 1ULL<<17 | 421 1ULL<<16); 422 t_mcic = mci.val; 423 424 if (((t_mcic & z_mcic) != 0) || 425 ((t_mcic & o_mcic) != o_mcic)) { 426 s390_handle_damage(); 427 } 428 429 /* 430 * Nullifying exigent condition, therefore we might 431 * retry this instruction. 432 */ 433 spin_lock(&ipd_lock); 434 tmp = get_tod_clock(); 435 if (((tmp - last_ipd) >> 12) < MAX_IPD_TIME) 436 ipd_count++; 437 else 438 ipd_count = 1; 439 last_ipd = tmp; 440 if (ipd_count == MAX_IPD_COUNT) 441 s390_handle_damage(); 442 spin_unlock(&ipd_lock); 443 } else { 444 /* Processing damage -> stopping machine */ 445 s390_handle_damage(); 446 } 447 } 448 if (s390_validate_registers(mci, user_mode(regs))) { 449 /* 450 * Couldn't restore all register contents for the 451 * user space process -> mark task for termination. 452 */ 453 mcck->kill_task = 1; 454 mcck->mcck_code = mci.val; 455 mcck_pending = 1; 456 } 457 458 /* 459 * Backup the machine check's info if it happens when the guest 460 * is running. 461 */ 462 if (test_cpu_flag(CIF_MCCK_GUEST)) 463 s390_backup_mcck_info(regs); 464 465 if (mci.cd) { 466 /* Timing facility damage */ 467 s390_handle_damage(); 468 } 469 if (mci.ed && mci.ec) { 470 /* External damage */ 471 if (S390_lowcore.external_damage_code & (1U << ED_STP_SYNC)) 472 mcck->stp_queue |= stp_sync_check(); 473 if (S390_lowcore.external_damage_code & (1U << ED_STP_ISLAND)) 474 mcck->stp_queue |= stp_island_check(); 475 mcck_pending = 1; 476 } 477 478 if (mci.cp) { 479 /* Channel report word pending */ 480 mcck->channel_report = 1; 481 mcck_pending = 1; 482 } 483 if (mci.w) { 484 /* Warning pending */ 485 mcck->warning = 1; 486 mcck_pending = 1; 487 } 488 489 /* 490 * If there are only Channel Report Pending and External Damage 491 * machine checks, they will not be reinjected into the guest 492 * because they refer to host conditions only. 493 */ 494 mcck_dam_code = (mci.val & MCIC_SUBCLASS_MASK); 495 if (test_cpu_flag(CIF_MCCK_GUEST) && 496 (mcck_dam_code & MCCK_CODE_NO_GUEST) != mcck_dam_code) { 497 /* Set exit reason code for host's later handling */ 498 *((long *)(regs->gprs[15] + __SF_SIE_REASON)) = -EINTR; 499 } 500 clear_cpu_flag(CIF_MCCK_GUEST); 501 502 if (user_mode(regs) && mcck_pending) { 503 nmi_exit(); 504 return 1; 505 } 506 507 if (mcck_pending) 508 schedule_mcck_handler(); 509 510 nmi_exit(); 511 return 0; 512 } 513 NOKPROBE_SYMBOL(s390_do_machine_check); 514 515 static int __init machine_check_init(void) 516 { 517 ctl_set_bit(14, 25); /* enable external damage MCH */ 518 ctl_set_bit(14, 27); /* enable system recovery MCH */ 519 ctl_set_bit(14, 24); /* enable warning MCH */ 520 return 0; 521 } 522 early_initcall(machine_check_init); 523