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; 109 110 kill_task = 0; 111 zero = 0; 112 113 if (!mci.gr) { 114 /* 115 * General purpose registers couldn't be restored and have 116 * unknown contents. Stop system or terminate process. 117 */ 118 if (!umode) 119 s390_handle_damage(); 120 kill_task = 1; 121 } 122 /* Validate control registers */ 123 if (!mci.cr) { 124 /* 125 * Control registers have unknown contents. 126 * Can't recover and therefore stopping machine. 127 */ 128 s390_handle_damage(); 129 } else { 130 asm volatile( 131 " lctlg 0,15,0(%0)\n" 132 " ptlb\n" 133 : : "a" (&S390_lowcore.cregs_save_area) : "memory"); 134 } 135 if (!mci.fp) { 136 /* 137 * Floating point registers can't be restored. If the 138 * kernel currently uses floating point registers the 139 * system is stopped. If the process has its floating 140 * pointer registers loaded it is terminated. 141 * Otherwise just revalidate the registers. 142 */ 143 if (S390_lowcore.fpu_flags & KERNEL_VXR_V0V7) 144 s390_handle_damage(); 145 if (!test_cpu_flag(CIF_FPU)) 146 kill_task = 1; 147 } 148 fpt_save_area = &S390_lowcore.floating_pt_save_area; 149 if (!mci.fc) { 150 /* 151 * Floating point control register can't be restored. 152 * If the kernel currently uses the floating pointer 153 * registers and needs the FPC register the system is 154 * stopped. If the process has its floating pointer 155 * registers loaded it is terminated. Otherwiese the 156 * FPC is just revalidated. 157 */ 158 if (S390_lowcore.fpu_flags & KERNEL_FPC) 159 s390_handle_damage(); 160 asm volatile("lfpc %0" : : "Q" (zero)); 161 if (!test_cpu_flag(CIF_FPU)) 162 kill_task = 1; 163 } else { 164 asm volatile("lfpc %0" 165 : : "Q" (S390_lowcore.fpt_creg_save_area)); 166 } 167 168 if (!MACHINE_HAS_VX) { 169 /* Validate floating point registers */ 170 asm volatile( 171 " ld 0,0(%0)\n" 172 " ld 1,8(%0)\n" 173 " ld 2,16(%0)\n" 174 " ld 3,24(%0)\n" 175 " ld 4,32(%0)\n" 176 " ld 5,40(%0)\n" 177 " ld 6,48(%0)\n" 178 " ld 7,56(%0)\n" 179 " ld 8,64(%0)\n" 180 " ld 9,72(%0)\n" 181 " ld 10,80(%0)\n" 182 " ld 11,88(%0)\n" 183 " ld 12,96(%0)\n" 184 " ld 13,104(%0)\n" 185 " ld 14,112(%0)\n" 186 " ld 15,120(%0)\n" 187 : : "a" (fpt_save_area) : "memory"); 188 } else { 189 /* Validate vector registers */ 190 union ctlreg0 cr0; 191 192 if (!mci.vr) { 193 /* 194 * Vector registers can't be restored. If the kernel 195 * currently uses vector registers the system is 196 * stopped. If the process has its vector registers 197 * loaded it is terminated. Otherwise just revalidate 198 * the registers. 199 */ 200 if (S390_lowcore.fpu_flags & KERNEL_VXR) 201 s390_handle_damage(); 202 if (!test_cpu_flag(CIF_FPU)) 203 kill_task = 1; 204 } 205 cr0.val = S390_lowcore.cregs_save_area[0]; 206 cr0.afp = cr0.vx = 1; 207 __ctl_load(cr0.val, 0, 0); 208 asm volatile( 209 " la 1,%0\n" 210 " .word 0xe70f,0x1000,0x0036\n" /* vlm 0,15,0(1) */ 211 " .word 0xe70f,0x1100,0x0c36\n" /* vlm 16,31,256(1) */ 212 : : "Q" (*(struct vx_array *) 213 &S390_lowcore.vector_save_area) : "1"); 214 __ctl_load(S390_lowcore.cregs_save_area[0], 0, 0); 215 } 216 /* Validate access registers */ 217 asm volatile( 218 " lam 0,15,0(%0)" 219 : : "a" (&S390_lowcore.access_regs_save_area)); 220 if (!mci.ar) { 221 /* 222 * Access registers have unknown contents. 223 * Terminating task. 224 */ 225 kill_task = 1; 226 } 227 /* 228 * We don't even try to validate the TOD register, since we simply 229 * can't write something sensible into that register. 230 */ 231 /* 232 * See if we can validate the TOD programmable register with its 233 * old contents (should be zero) otherwise set it to zero. 234 */ 235 if (!mci.pr) 236 asm volatile( 237 " sr 0,0\n" 238 " sckpf" 239 : : : "0", "cc"); 240 else 241 asm volatile( 242 " l 0,%0\n" 243 " sckpf" 244 : : "Q" (S390_lowcore.tod_progreg_save_area) 245 : "0", "cc"); 246 /* Validate clock comparator register */ 247 set_clock_comparator(S390_lowcore.clock_comparator); 248 /* Check if old PSW is valid */ 249 if (!mci.wp) 250 /* 251 * Can't tell if we come from user or kernel mode 252 * -> stopping machine. 253 */ 254 s390_handle_damage(); 255 256 if (!mci.ms || !mci.pm || !mci.ia) 257 kill_task = 1; 258 259 return kill_task; 260 } 261 262 #define MAX_IPD_COUNT 29 263 #define MAX_IPD_TIME (5 * 60 * USEC_PER_SEC) /* 5 minutes */ 264 265 #define ED_STP_ISLAND 6 /* External damage STP island check */ 266 #define ED_STP_SYNC 7 /* External damage STP sync check */ 267 268 /* 269 * machine check handler. 270 */ 271 void notrace s390_do_machine_check(struct pt_regs *regs) 272 { 273 static int ipd_count; 274 static DEFINE_SPINLOCK(ipd_lock); 275 static unsigned long long last_ipd; 276 struct mcck_struct *mcck; 277 unsigned long long tmp; 278 union mci mci; 279 280 nmi_enter(); 281 inc_irq_stat(NMI_NMI); 282 mci.val = S390_lowcore.mcck_interruption_code; 283 mcck = this_cpu_ptr(&cpu_mcck); 284 285 if (mci.sd) { 286 /* System damage -> stopping machine */ 287 s390_handle_damage(); 288 } 289 if (mci.pd) { 290 if (mci.b) { 291 /* Processing backup -> verify if we can survive this */ 292 u64 z_mcic, o_mcic, t_mcic; 293 z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<29); 294 o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 | 295 1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 | 296 1ULL<<30 | 1ULL<<21 | 1ULL<<20 | 1ULL<<17 | 297 1ULL<<16); 298 t_mcic = mci.val; 299 300 if (((t_mcic & z_mcic) != 0) || 301 ((t_mcic & o_mcic) != o_mcic)) { 302 s390_handle_damage(); 303 } 304 305 /* 306 * Nullifying exigent condition, therefore we might 307 * retry this instruction. 308 */ 309 spin_lock(&ipd_lock); 310 tmp = get_tod_clock(); 311 if (((tmp - last_ipd) >> 12) < MAX_IPD_TIME) 312 ipd_count++; 313 else 314 ipd_count = 1; 315 last_ipd = tmp; 316 if (ipd_count == MAX_IPD_COUNT) 317 s390_handle_damage(); 318 spin_unlock(&ipd_lock); 319 } else { 320 /* Processing damage -> stopping machine */ 321 s390_handle_damage(); 322 } 323 } 324 if (s390_validate_registers(mci, user_mode(regs))) { 325 /* 326 * Couldn't restore all register contents for the 327 * user space process -> mark task for termination. 328 */ 329 mcck->kill_task = 1; 330 mcck->mcck_code = mci.val; 331 set_cpu_flag(CIF_MCCK_PENDING); 332 } 333 if (mci.cd) { 334 /* Timing facility damage */ 335 s390_handle_damage(); 336 } 337 if (mci.ed && mci.ec) { 338 /* External damage */ 339 if (S390_lowcore.external_damage_code & (1U << ED_STP_SYNC)) 340 mcck->stp_queue |= stp_sync_check(); 341 if (S390_lowcore.external_damage_code & (1U << ED_STP_ISLAND)) 342 mcck->stp_queue |= stp_island_check(); 343 if (mcck->stp_queue) 344 set_cpu_flag(CIF_MCCK_PENDING); 345 } 346 if (mci.se) 347 /* Storage error uncorrected */ 348 s390_handle_damage(); 349 if (mci.ke) 350 /* Storage key-error uncorrected */ 351 s390_handle_damage(); 352 if (mci.ds && mci.fa) 353 /* Storage degradation */ 354 s390_handle_damage(); 355 if (mci.cp) { 356 /* Channel report word pending */ 357 mcck->channel_report = 1; 358 set_cpu_flag(CIF_MCCK_PENDING); 359 } 360 if (mci.w) { 361 /* Warning pending */ 362 mcck->warning = 1; 363 set_cpu_flag(CIF_MCCK_PENDING); 364 } 365 nmi_exit(); 366 } 367 368 static int __init machine_check_init(void) 369 { 370 ctl_set_bit(14, 25); /* enable external damage MCH */ 371 ctl_set_bit(14, 27); /* enable system recovery MCH */ 372 ctl_set_bit(14, 24); /* enable warning MCH */ 373 return 0; 374 } 375 early_initcall(machine_check_init); 376