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