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