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