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/time.h> 16 #include <linux/module.h> 17 #include <linux/sched/signal.h> 18 19 #include <linux/export.h> 20 #include <asm/lowcore.h> 21 #include <asm/smp.h> 22 #include <asm/stp.h> 23 #include <asm/cputime.h> 24 #include <asm/nmi.h> 25 #include <asm/crw.h> 26 #include <asm/switch_to.h> 27 #include <asm/ctl_reg.h> 28 #include <asm/asm-offsets.h> 29 #include <linux/kvm_host.h> 30 31 struct mcck_struct { 32 unsigned int kill_task : 1; 33 unsigned int channel_report : 1; 34 unsigned int warning : 1; 35 unsigned int stp_queue : 1; 36 unsigned long mcck_code; 37 }; 38 39 static DEFINE_PER_CPU(struct mcck_struct, cpu_mcck); 40 41 static void s390_handle_damage(void) 42 { 43 smp_send_stop(); 44 disabled_wait((unsigned long) __builtin_return_address(0)); 45 while (1); 46 } 47 48 /* 49 * Main machine check handler function. Will be called with interrupts enabled 50 * or disabled and machine checks enabled or disabled. 51 */ 52 void s390_handle_mcck(void) 53 { 54 unsigned long flags; 55 struct mcck_struct mcck; 56 57 /* 58 * Disable machine checks and get the current state of accumulated 59 * machine checks. Afterwards delete the old state and enable machine 60 * checks again. 61 */ 62 local_irq_save(flags); 63 local_mcck_disable(); 64 mcck = *this_cpu_ptr(&cpu_mcck); 65 memset(this_cpu_ptr(&cpu_mcck), 0, sizeof(mcck)); 66 clear_cpu_flag(CIF_MCCK_PENDING); 67 local_mcck_enable(); 68 local_irq_restore(flags); 69 70 if (mcck.channel_report) 71 crw_handle_channel_report(); 72 /* 73 * A warning may remain for a prolonged period on the bare iron. 74 * (actually until the machine is powered off, or the problem is gone) 75 * So we just stop listening for the WARNING MCH and avoid continuously 76 * being interrupted. One caveat is however, that we must do this per 77 * processor and cannot use the smp version of ctl_clear_bit(). 78 * On VM we only get one interrupt per virtally presented machinecheck. 79 * Though one suffices, we may get one interrupt per (virtual) cpu. 80 */ 81 if (mcck.warning) { /* WARNING pending ? */ 82 static int mchchk_wng_posted = 0; 83 84 /* Use single cpu clear, as we cannot handle smp here. */ 85 __ctl_clear_bit(14, 24); /* Disable WARNING MCH */ 86 if (xchg(&mchchk_wng_posted, 1) == 0) 87 kill_cad_pid(SIGPWR, 1); 88 } 89 if (mcck.stp_queue) 90 stp_queue_work(); 91 if (mcck.kill_task) { 92 local_irq_enable(); 93 printk(KERN_EMERG "mcck: Terminating task because of machine " 94 "malfunction (code 0x%016lx).\n", mcck.mcck_code); 95 printk(KERN_EMERG "mcck: task: %s, pid: %d.\n", 96 current->comm, current->pid); 97 do_exit(SIGSEGV); 98 } 99 } 100 EXPORT_SYMBOL_GPL(s390_handle_mcck); 101 102 /* 103 * returns 0 if all registers could be validated 104 * returns 1 otherwise 105 */ 106 static int notrace s390_validate_registers(union mci mci, int umode) 107 { 108 int kill_task; 109 u64 zero; 110 void *fpt_save_area; 111 struct mcesa *mcesa; 112 113 kill_task = 0; 114 zero = 0; 115 116 if (!mci.gr) { 117 /* 118 * General purpose registers couldn't be restored and have 119 * unknown contents. Stop system or terminate process. 120 */ 121 if (!umode) 122 s390_handle_damage(); 123 kill_task = 1; 124 } 125 /* Validate control registers */ 126 if (!mci.cr) { 127 /* 128 * Control registers have unknown contents. 129 * Can't recover and therefore stopping machine. 130 */ 131 s390_handle_damage(); 132 } else { 133 asm volatile( 134 " lctlg 0,15,0(%0)\n" 135 " ptlb\n" 136 : : "a" (&S390_lowcore.cregs_save_area) : "memory"); 137 } 138 if (!mci.fp) { 139 /* 140 * Floating point registers can't be restored. If the 141 * kernel currently uses floating point registers the 142 * system is stopped. If the process has its floating 143 * pointer registers loaded it is terminated. 144 * Otherwise just revalidate the registers. 145 */ 146 if (S390_lowcore.fpu_flags & KERNEL_VXR_V0V7) 147 s390_handle_damage(); 148 if (!test_cpu_flag(CIF_FPU)) 149 kill_task = 1; 150 } 151 fpt_save_area = &S390_lowcore.floating_pt_save_area; 152 if (!mci.fc) { 153 /* 154 * Floating point control register can't be restored. 155 * If the kernel currently uses the floating pointer 156 * registers and needs the FPC register the system is 157 * stopped. If the process has its floating pointer 158 * registers loaded it is terminated. Otherwiese the 159 * FPC is just revalidated. 160 */ 161 if (S390_lowcore.fpu_flags & KERNEL_FPC) 162 s390_handle_damage(); 163 asm volatile("lfpc %0" : : "Q" (zero)); 164 if (!test_cpu_flag(CIF_FPU)) 165 kill_task = 1; 166 } else { 167 asm volatile("lfpc %0" 168 : : "Q" (S390_lowcore.fpt_creg_save_area)); 169 } 170 171 mcesa = (struct mcesa *)(S390_lowcore.mcesad & MCESA_ORIGIN_MASK); 172 if (!MACHINE_HAS_VX) { 173 /* Validate floating point registers */ 174 asm volatile( 175 " ld 0,0(%0)\n" 176 " ld 1,8(%0)\n" 177 " ld 2,16(%0)\n" 178 " ld 3,24(%0)\n" 179 " ld 4,32(%0)\n" 180 " ld 5,40(%0)\n" 181 " ld 6,48(%0)\n" 182 " ld 7,56(%0)\n" 183 " ld 8,64(%0)\n" 184 " ld 9,72(%0)\n" 185 " ld 10,80(%0)\n" 186 " ld 11,88(%0)\n" 187 " ld 12,96(%0)\n" 188 " ld 13,104(%0)\n" 189 " ld 14,112(%0)\n" 190 " ld 15,120(%0)\n" 191 : : "a" (fpt_save_area) : "memory"); 192 } else { 193 /* Validate vector registers */ 194 union ctlreg0 cr0; 195 196 if (!mci.vr) { 197 /* 198 * Vector registers can't be restored. If the kernel 199 * currently uses vector registers the system is 200 * stopped. If the process has its vector registers 201 * loaded it is terminated. Otherwise just revalidate 202 * the registers. 203 */ 204 if (S390_lowcore.fpu_flags & KERNEL_VXR) 205 s390_handle_damage(); 206 if (!test_cpu_flag(CIF_FPU)) 207 kill_task = 1; 208 } 209 cr0.val = S390_lowcore.cregs_save_area[0]; 210 cr0.afp = cr0.vx = 1; 211 __ctl_load(cr0.val, 0, 0); 212 asm volatile( 213 " la 1,%0\n" 214 " .word 0xe70f,0x1000,0x0036\n" /* vlm 0,15,0(1) */ 215 " .word 0xe70f,0x1100,0x0c36\n" /* vlm 16,31,256(1) */ 216 : : "Q" (*(struct vx_array *) mcesa->vector_save_area) 217 : "1"); 218 __ctl_load(S390_lowcore.cregs_save_area[0], 0, 0); 219 } 220 /* Validate access registers */ 221 asm volatile( 222 " lam 0,15,0(%0)" 223 : : "a" (&S390_lowcore.access_regs_save_area)); 224 if (!mci.ar) { 225 /* 226 * Access registers have unknown contents. 227 * Terminating task. 228 */ 229 kill_task = 1; 230 } 231 /* Validate guarded storage registers */ 232 if (MACHINE_HAS_GS && (S390_lowcore.cregs_save_area[2] & (1UL << 4))) { 233 if (!mci.gs) 234 /* 235 * Guarded storage register can't be restored and 236 * the current processes uses guarded storage. 237 * It has to be terminated. 238 */ 239 kill_task = 1; 240 else 241 load_gs_cb((struct gs_cb *) 242 mcesa->guarded_storage_save_area); 243 } 244 /* 245 * We don't even try to validate the TOD register, since we simply 246 * can't write something sensible into that register. 247 */ 248 /* 249 * See if we can validate the TOD programmable register with its 250 * old contents (should be zero) otherwise set it to zero. 251 */ 252 if (!mci.pr) 253 asm volatile( 254 " sr 0,0\n" 255 " sckpf" 256 : : : "0", "cc"); 257 else 258 asm volatile( 259 " l 0,%0\n" 260 " sckpf" 261 : : "Q" (S390_lowcore.tod_progreg_save_area) 262 : "0", "cc"); 263 /* Validate clock comparator register */ 264 set_clock_comparator(S390_lowcore.clock_comparator); 265 /* Check if old PSW is valid */ 266 if (!mci.wp) 267 /* 268 * Can't tell if we come from user or kernel mode 269 * -> stopping machine. 270 */ 271 s390_handle_damage(); 272 273 if (!mci.ms || !mci.pm || !mci.ia) 274 kill_task = 1; 275 276 return kill_task; 277 } 278 279 /* 280 * Backup the guest's machine check info to its description block 281 */ 282 static void notrace s390_backup_mcck_info(struct pt_regs *regs) 283 { 284 struct mcck_volatile_info *mcck_backup; 285 struct sie_page *sie_page; 286 287 /* r14 contains the sie block, which was set in sie64a */ 288 struct kvm_s390_sie_block *sie_block = 289 (struct kvm_s390_sie_block *) regs->gprs[14]; 290 291 if (sie_block == NULL) 292 /* Something's seriously wrong, stop system. */ 293 s390_handle_damage(); 294 295 sie_page = container_of(sie_block, struct sie_page, sie_block); 296 mcck_backup = &sie_page->mcck_info; 297 mcck_backup->mcic = S390_lowcore.mcck_interruption_code & 298 ~(MCCK_CODE_CP | MCCK_CODE_EXT_DAMAGE); 299 mcck_backup->ext_damage_code = S390_lowcore.external_damage_code; 300 mcck_backup->failing_storage_address 301 = S390_lowcore.failing_storage_address; 302 } 303 304 #define MAX_IPD_COUNT 29 305 #define MAX_IPD_TIME (5 * 60 * USEC_PER_SEC) /* 5 minutes */ 306 307 #define ED_STP_ISLAND 6 /* External damage STP island check */ 308 #define ED_STP_SYNC 7 /* External damage STP sync check */ 309 310 #define MCCK_CODE_NO_GUEST (MCCK_CODE_CP | MCCK_CODE_EXT_DAMAGE) 311 312 /* 313 * machine check handler. 314 */ 315 void notrace s390_do_machine_check(struct pt_regs *regs) 316 { 317 static int ipd_count; 318 static DEFINE_SPINLOCK(ipd_lock); 319 static unsigned long long last_ipd; 320 struct mcck_struct *mcck; 321 unsigned long long tmp; 322 union mci mci; 323 unsigned long mcck_dam_code; 324 325 nmi_enter(); 326 inc_irq_stat(NMI_NMI); 327 mci.val = S390_lowcore.mcck_interruption_code; 328 mcck = this_cpu_ptr(&cpu_mcck); 329 330 if (mci.sd) { 331 /* System damage -> stopping machine */ 332 s390_handle_damage(); 333 } 334 335 /* 336 * Reinject the instruction processing damages' machine checks 337 * including Delayed Access Exception into the guest 338 * instead of damaging the host if they happen in the guest. 339 */ 340 if (mci.pd && !test_cpu_flag(CIF_MCCK_GUEST)) { 341 if (mci.b) { 342 /* Processing backup -> verify if we can survive this */ 343 u64 z_mcic, o_mcic, t_mcic; 344 z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<29); 345 o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 | 346 1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 | 347 1ULL<<30 | 1ULL<<21 | 1ULL<<20 | 1ULL<<17 | 348 1ULL<<16); 349 t_mcic = mci.val; 350 351 if (((t_mcic & z_mcic) != 0) || 352 ((t_mcic & o_mcic) != o_mcic)) { 353 s390_handle_damage(); 354 } 355 356 /* 357 * Nullifying exigent condition, therefore we might 358 * retry this instruction. 359 */ 360 spin_lock(&ipd_lock); 361 tmp = get_tod_clock(); 362 if (((tmp - last_ipd) >> 12) < MAX_IPD_TIME) 363 ipd_count++; 364 else 365 ipd_count = 1; 366 last_ipd = tmp; 367 if (ipd_count == MAX_IPD_COUNT) 368 s390_handle_damage(); 369 spin_unlock(&ipd_lock); 370 } else { 371 /* Processing damage -> stopping machine */ 372 s390_handle_damage(); 373 } 374 } 375 if (s390_validate_registers(mci, user_mode(regs))) { 376 /* 377 * Couldn't restore all register contents for the 378 * user space process -> mark task for termination. 379 */ 380 mcck->kill_task = 1; 381 mcck->mcck_code = mci.val; 382 set_cpu_flag(CIF_MCCK_PENDING); 383 } 384 385 /* 386 * Backup the machine check's info if it happens when the guest 387 * is running. 388 */ 389 if (test_cpu_flag(CIF_MCCK_GUEST)) 390 s390_backup_mcck_info(regs); 391 392 if (mci.cd) { 393 /* Timing facility damage */ 394 s390_handle_damage(); 395 } 396 if (mci.ed && mci.ec) { 397 /* External damage */ 398 if (S390_lowcore.external_damage_code & (1U << ED_STP_SYNC)) 399 mcck->stp_queue |= stp_sync_check(); 400 if (S390_lowcore.external_damage_code & (1U << ED_STP_ISLAND)) 401 mcck->stp_queue |= stp_island_check(); 402 if (mcck->stp_queue) 403 set_cpu_flag(CIF_MCCK_PENDING); 404 } 405 406 /* 407 * Reinject storage related machine checks into the guest if they 408 * happen when the guest is running. 409 */ 410 if (!test_cpu_flag(CIF_MCCK_GUEST)) { 411 if (mci.se) 412 /* Storage error uncorrected */ 413 s390_handle_damage(); 414 if (mci.ke) 415 /* Storage key-error uncorrected */ 416 s390_handle_damage(); 417 if (mci.ds && mci.fa) 418 /* Storage degradation */ 419 s390_handle_damage(); 420 } 421 if (mci.cp) { 422 /* Channel report word pending */ 423 mcck->channel_report = 1; 424 set_cpu_flag(CIF_MCCK_PENDING); 425 } 426 if (mci.w) { 427 /* Warning pending */ 428 mcck->warning = 1; 429 set_cpu_flag(CIF_MCCK_PENDING); 430 } 431 432 /* 433 * If there are only Channel Report Pending and External Damage 434 * machine checks, they will not be reinjected into the guest 435 * because they refer to host conditions only. 436 */ 437 mcck_dam_code = (mci.val & MCIC_SUBCLASS_MASK); 438 if (test_cpu_flag(CIF_MCCK_GUEST) && 439 (mcck_dam_code & MCCK_CODE_NO_GUEST) != mcck_dam_code) { 440 /* Set exit reason code for host's later handling */ 441 *((long *)(regs->gprs[15] + __SF_SIE_REASON)) = -EINTR; 442 } 443 clear_cpu_flag(CIF_MCCK_GUEST); 444 nmi_exit(); 445 } 446 447 static int __init machine_check_init(void) 448 { 449 ctl_set_bit(14, 25); /* enable external damage MCH */ 450 ctl_set_bit(14, 27); /* enable system recovery MCH */ 451 ctl_set_bit(14, 24); /* enable warning MCH */ 452 return 0; 453 } 454 early_initcall(machine_check_init); 455