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