1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org) 4 * Copyright 2007-2010 Freescale Semiconductor, Inc. 5 * 6 * Modified by Cort Dougan (cort@cs.nmt.edu) 7 * and Paul Mackerras (paulus@samba.org) 8 */ 9 10 /* 11 * This file handles the architecture-dependent parts of hardware exceptions 12 */ 13 14 #include <linux/errno.h> 15 #include <linux/sched.h> 16 #include <linux/sched/debug.h> 17 #include <linux/kernel.h> 18 #include <linux/mm.h> 19 #include <linux/pkeys.h> 20 #include <linux/stddef.h> 21 #include <linux/unistd.h> 22 #include <linux/ptrace.h> 23 #include <linux/user.h> 24 #include <linux/interrupt.h> 25 #include <linux/init.h> 26 #include <linux/extable.h> 27 #include <linux/module.h> /* print_modules */ 28 #include <linux/prctl.h> 29 #include <linux/delay.h> 30 #include <linux/kprobes.h> 31 #include <linux/kexec.h> 32 #include <linux/backlight.h> 33 #include <linux/bug.h> 34 #include <linux/kdebug.h> 35 #include <linux/ratelimit.h> 36 #include <linux/context_tracking.h> 37 #include <linux/smp.h> 38 #include <linux/console.h> 39 #include <linux/kmsg_dump.h> 40 #include <linux/debugfs.h> 41 42 #include <asm/emulated_ops.h> 43 #include <linux/uaccess.h> 44 #include <asm/interrupt.h> 45 #include <asm/io.h> 46 #include <asm/machdep.h> 47 #include <asm/rtas.h> 48 #include <asm/pmc.h> 49 #include <asm/reg.h> 50 #ifdef CONFIG_PMAC_BACKLIGHT 51 #include <asm/backlight.h> 52 #endif 53 #ifdef CONFIG_PPC64 54 #include <asm/firmware.h> 55 #include <asm/processor.h> 56 #endif 57 #include <asm/kexec.h> 58 #include <asm/ppc-opcode.h> 59 #include <asm/rio.h> 60 #include <asm/fadump.h> 61 #include <asm/switch_to.h> 62 #include <asm/tm.h> 63 #include <asm/debug.h> 64 #include <asm/asm-prototypes.h> 65 #include <asm/hmi.h> 66 #include <sysdev/fsl_pci.h> 67 #include <asm/kprobes.h> 68 #include <asm/stacktrace.h> 69 #include <asm/nmi.h> 70 #include <asm/disassemble.h> 71 72 #if defined(CONFIG_DEBUGGER) || defined(CONFIG_KEXEC_CORE) 73 int (*__debugger)(struct pt_regs *regs) __read_mostly; 74 int (*__debugger_ipi)(struct pt_regs *regs) __read_mostly; 75 int (*__debugger_bpt)(struct pt_regs *regs) __read_mostly; 76 int (*__debugger_sstep)(struct pt_regs *regs) __read_mostly; 77 int (*__debugger_iabr_match)(struct pt_regs *regs) __read_mostly; 78 int (*__debugger_break_match)(struct pt_regs *regs) __read_mostly; 79 int (*__debugger_fault_handler)(struct pt_regs *regs) __read_mostly; 80 81 EXPORT_SYMBOL(__debugger); 82 EXPORT_SYMBOL(__debugger_ipi); 83 EXPORT_SYMBOL(__debugger_bpt); 84 EXPORT_SYMBOL(__debugger_sstep); 85 EXPORT_SYMBOL(__debugger_iabr_match); 86 EXPORT_SYMBOL(__debugger_break_match); 87 EXPORT_SYMBOL(__debugger_fault_handler); 88 #endif 89 90 /* Transactional Memory trap debug */ 91 #ifdef TM_DEBUG_SW 92 #define TM_DEBUG(x...) printk(KERN_INFO x) 93 #else 94 #define TM_DEBUG(x...) do { } while(0) 95 #endif 96 97 static const char *signame(int signr) 98 { 99 switch (signr) { 100 case SIGBUS: return "bus error"; 101 case SIGFPE: return "floating point exception"; 102 case SIGILL: return "illegal instruction"; 103 case SIGSEGV: return "segfault"; 104 case SIGTRAP: return "unhandled trap"; 105 } 106 107 return "unknown signal"; 108 } 109 110 /* 111 * Trap & Exception support 112 */ 113 114 #ifdef CONFIG_PMAC_BACKLIGHT 115 static void pmac_backlight_unblank(void) 116 { 117 mutex_lock(&pmac_backlight_mutex); 118 if (pmac_backlight) { 119 struct backlight_properties *props; 120 121 props = &pmac_backlight->props; 122 props->brightness = props->max_brightness; 123 props->power = FB_BLANK_UNBLANK; 124 backlight_update_status(pmac_backlight); 125 } 126 mutex_unlock(&pmac_backlight_mutex); 127 } 128 #else 129 static inline void pmac_backlight_unblank(void) { } 130 #endif 131 132 /* 133 * If oops/die is expected to crash the machine, return true here. 134 * 135 * This should not be expected to be 100% accurate, there may be 136 * notifiers registered or other unexpected conditions that may bring 137 * down the kernel. Or if the current process in the kernel is holding 138 * locks or has other critical state, the kernel may become effectively 139 * unusable anyway. 140 */ 141 bool die_will_crash(void) 142 { 143 if (should_fadump_crash()) 144 return true; 145 if (kexec_should_crash(current)) 146 return true; 147 if (in_interrupt() || panic_on_oops || 148 !current->pid || is_global_init(current)) 149 return true; 150 151 return false; 152 } 153 154 static arch_spinlock_t die_lock = __ARCH_SPIN_LOCK_UNLOCKED; 155 static int die_owner = -1; 156 static unsigned int die_nest_count; 157 static int die_counter; 158 159 extern void panic_flush_kmsg_start(void) 160 { 161 /* 162 * These are mostly taken from kernel/panic.c, but tries to do 163 * relatively minimal work. Don't use delay functions (TB may 164 * be broken), don't crash dump (need to set a firmware log), 165 * don't run notifiers. We do want to get some information to 166 * Linux console. 167 */ 168 console_verbose(); 169 bust_spinlocks(1); 170 } 171 172 extern void panic_flush_kmsg_end(void) 173 { 174 kmsg_dump(KMSG_DUMP_PANIC); 175 bust_spinlocks(0); 176 debug_locks_off(); 177 console_flush_on_panic(CONSOLE_FLUSH_PENDING); 178 } 179 180 static unsigned long oops_begin(struct pt_regs *regs) 181 { 182 int cpu; 183 unsigned long flags; 184 185 oops_enter(); 186 187 /* racy, but better than risking deadlock. */ 188 raw_local_irq_save(flags); 189 cpu = smp_processor_id(); 190 if (!arch_spin_trylock(&die_lock)) { 191 if (cpu == die_owner) 192 /* nested oops. should stop eventually */; 193 else 194 arch_spin_lock(&die_lock); 195 } 196 die_nest_count++; 197 die_owner = cpu; 198 console_verbose(); 199 bust_spinlocks(1); 200 if (machine_is(powermac)) 201 pmac_backlight_unblank(); 202 return flags; 203 } 204 NOKPROBE_SYMBOL(oops_begin); 205 206 static void oops_end(unsigned long flags, struct pt_regs *regs, 207 int signr) 208 { 209 bust_spinlocks(0); 210 add_taint(TAINT_DIE, LOCKDEP_NOW_UNRELIABLE); 211 die_nest_count--; 212 oops_exit(); 213 printk("\n"); 214 if (!die_nest_count) { 215 /* Nest count reaches zero, release the lock. */ 216 die_owner = -1; 217 arch_spin_unlock(&die_lock); 218 } 219 raw_local_irq_restore(flags); 220 221 /* 222 * system_reset_excption handles debugger, crash dump, panic, for 0x100 223 */ 224 if (TRAP(regs) == INTERRUPT_SYSTEM_RESET) 225 return; 226 227 crash_fadump(regs, "die oops"); 228 229 if (kexec_should_crash(current)) 230 crash_kexec(regs); 231 232 if (!signr) 233 return; 234 235 /* 236 * While our oops output is serialised by a spinlock, output 237 * from panic() called below can race and corrupt it. If we 238 * know we are going to panic, delay for 1 second so we have a 239 * chance to get clean backtraces from all CPUs that are oopsing. 240 */ 241 if (in_interrupt() || panic_on_oops || !current->pid || 242 is_global_init(current)) { 243 mdelay(MSEC_PER_SEC); 244 } 245 246 if (panic_on_oops) 247 panic("Fatal exception"); 248 do_exit(signr); 249 } 250 NOKPROBE_SYMBOL(oops_end); 251 252 static char *get_mmu_str(void) 253 { 254 if (early_radix_enabled()) 255 return " MMU=Radix"; 256 if (early_mmu_has_feature(MMU_FTR_HPTE_TABLE)) 257 return " MMU=Hash"; 258 return ""; 259 } 260 261 static int __die(const char *str, struct pt_regs *regs, long err) 262 { 263 printk("Oops: %s, sig: %ld [#%d]\n", str, err, ++die_counter); 264 265 printk("%s PAGE_SIZE=%luK%s%s%s%s%s%s %s\n", 266 IS_ENABLED(CONFIG_CPU_LITTLE_ENDIAN) ? "LE" : "BE", 267 PAGE_SIZE / 1024, get_mmu_str(), 268 IS_ENABLED(CONFIG_PREEMPT) ? " PREEMPT" : "", 269 IS_ENABLED(CONFIG_SMP) ? " SMP" : "", 270 IS_ENABLED(CONFIG_SMP) ? (" NR_CPUS=" __stringify(NR_CPUS)) : "", 271 debug_pagealloc_enabled() ? " DEBUG_PAGEALLOC" : "", 272 IS_ENABLED(CONFIG_NUMA) ? " NUMA" : "", 273 ppc_md.name ? ppc_md.name : ""); 274 275 if (notify_die(DIE_OOPS, str, regs, err, 255, SIGSEGV) == NOTIFY_STOP) 276 return 1; 277 278 print_modules(); 279 show_regs(regs); 280 281 return 0; 282 } 283 NOKPROBE_SYMBOL(__die); 284 285 void die(const char *str, struct pt_regs *regs, long err) 286 { 287 unsigned long flags; 288 289 /* 290 * system_reset_excption handles debugger, crash dump, panic, for 0x100 291 */ 292 if (TRAP(regs) != INTERRUPT_SYSTEM_RESET) { 293 if (debugger(regs)) 294 return; 295 } 296 297 flags = oops_begin(regs); 298 if (__die(str, regs, err)) 299 err = 0; 300 oops_end(flags, regs, err); 301 } 302 NOKPROBE_SYMBOL(die); 303 304 void user_single_step_report(struct pt_regs *regs) 305 { 306 force_sig_fault(SIGTRAP, TRAP_TRACE, (void __user *)regs->nip); 307 } 308 309 static void show_signal_msg(int signr, struct pt_regs *regs, int code, 310 unsigned long addr) 311 { 312 static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL, 313 DEFAULT_RATELIMIT_BURST); 314 315 if (!show_unhandled_signals) 316 return; 317 318 if (!unhandled_signal(current, signr)) 319 return; 320 321 if (!__ratelimit(&rs)) 322 return; 323 324 pr_info("%s[%d]: %s (%d) at %lx nip %lx lr %lx code %x", 325 current->comm, current->pid, signame(signr), signr, 326 addr, regs->nip, regs->link, code); 327 328 print_vma_addr(KERN_CONT " in ", regs->nip); 329 330 pr_cont("\n"); 331 332 show_user_instructions(regs); 333 } 334 335 static bool exception_common(int signr, struct pt_regs *regs, int code, 336 unsigned long addr) 337 { 338 if (!user_mode(regs)) { 339 die("Exception in kernel mode", regs, signr); 340 return false; 341 } 342 343 show_signal_msg(signr, regs, code, addr); 344 345 if (arch_irqs_disabled()) 346 interrupt_cond_local_irq_enable(regs); 347 348 current->thread.trap_nr = code; 349 350 return true; 351 } 352 353 void _exception_pkey(struct pt_regs *regs, unsigned long addr, int key) 354 { 355 if (!exception_common(SIGSEGV, regs, SEGV_PKUERR, addr)) 356 return; 357 358 force_sig_pkuerr((void __user *) addr, key); 359 } 360 361 void _exception(int signr, struct pt_regs *regs, int code, unsigned long addr) 362 { 363 if (!exception_common(signr, regs, code, addr)) 364 return; 365 366 force_sig_fault(signr, code, (void __user *)addr); 367 } 368 369 /* 370 * The interrupt architecture has a quirk in that the HV interrupts excluding 371 * the NMIs (0x100 and 0x200) do not clear MSR[RI] at entry. The first thing 372 * that an interrupt handler must do is save off a GPR into a scratch register, 373 * and all interrupts on POWERNV (HV=1) use the HSPRG1 register as scratch. 374 * Therefore an NMI can clobber an HV interrupt's live HSPRG1 without noticing 375 * that it is non-reentrant, which leads to random data corruption. 376 * 377 * The solution is for NMI interrupts in HV mode to check if they originated 378 * from these critical HV interrupt regions. If so, then mark them not 379 * recoverable. 380 * 381 * An alternative would be for HV NMIs to use SPRG for scratch to avoid the 382 * HSPRG1 clobber, however this would cause guest SPRG to be clobbered. Linux 383 * guests should always have MSR[RI]=0 when its scratch SPRG is in use, so 384 * that would work. However any other guest OS that may have the SPRG live 385 * and MSR[RI]=1 could encounter silent corruption. 386 * 387 * Builds that do not support KVM could take this second option to increase 388 * the recoverability of NMIs. 389 */ 390 void hv_nmi_check_nonrecoverable(struct pt_regs *regs) 391 { 392 #ifdef CONFIG_PPC_POWERNV 393 unsigned long kbase = (unsigned long)_stext; 394 unsigned long nip = regs->nip; 395 396 if (!(regs->msr & MSR_RI)) 397 return; 398 if (!(regs->msr & MSR_HV)) 399 return; 400 if (regs->msr & MSR_PR) 401 return; 402 403 /* 404 * Now test if the interrupt has hit a range that may be using 405 * HSPRG1 without having RI=0 (i.e., an HSRR interrupt). The 406 * problem ranges all run un-relocated. Test real and virt modes 407 * at the same time by dropping the high bit of the nip (virt mode 408 * entry points still have the +0x4000 offset). 409 */ 410 nip &= ~0xc000000000000000ULL; 411 if ((nip >= 0x500 && nip < 0x600) || (nip >= 0x4500 && nip < 0x4600)) 412 goto nonrecoverable; 413 if ((nip >= 0x980 && nip < 0xa00) || (nip >= 0x4980 && nip < 0x4a00)) 414 goto nonrecoverable; 415 if ((nip >= 0xe00 && nip < 0xec0) || (nip >= 0x4e00 && nip < 0x4ec0)) 416 goto nonrecoverable; 417 if ((nip >= 0xf80 && nip < 0xfa0) || (nip >= 0x4f80 && nip < 0x4fa0)) 418 goto nonrecoverable; 419 420 /* Trampoline code runs un-relocated so subtract kbase. */ 421 if (nip >= (unsigned long)(start_real_trampolines - kbase) && 422 nip < (unsigned long)(end_real_trampolines - kbase)) 423 goto nonrecoverable; 424 if (nip >= (unsigned long)(start_virt_trampolines - kbase) && 425 nip < (unsigned long)(end_virt_trampolines - kbase)) 426 goto nonrecoverable; 427 return; 428 429 nonrecoverable: 430 regs_set_unrecoverable(regs); 431 #endif 432 } 433 DEFINE_INTERRUPT_HANDLER_NMI(system_reset_exception) 434 { 435 unsigned long hsrr0, hsrr1; 436 bool saved_hsrrs = false; 437 438 /* 439 * System reset can interrupt code where HSRRs are live and MSR[RI]=1. 440 * The system reset interrupt itself may clobber HSRRs (e.g., to call 441 * OPAL), so save them here and restore them before returning. 442 * 443 * Machine checks don't need to save HSRRs, as the real mode handler 444 * is careful to avoid them, and the regular handler is not delivered 445 * as an NMI. 446 */ 447 if (cpu_has_feature(CPU_FTR_HVMODE)) { 448 hsrr0 = mfspr(SPRN_HSRR0); 449 hsrr1 = mfspr(SPRN_HSRR1); 450 saved_hsrrs = true; 451 } 452 453 hv_nmi_check_nonrecoverable(regs); 454 455 __this_cpu_inc(irq_stat.sreset_irqs); 456 457 /* See if any machine dependent calls */ 458 if (ppc_md.system_reset_exception) { 459 if (ppc_md.system_reset_exception(regs)) 460 goto out; 461 } 462 463 if (debugger(regs)) 464 goto out; 465 466 kmsg_dump(KMSG_DUMP_OOPS); 467 /* 468 * A system reset is a request to dump, so we always send 469 * it through the crashdump code (if fadump or kdump are 470 * registered). 471 */ 472 crash_fadump(regs, "System Reset"); 473 474 crash_kexec(regs); 475 476 /* 477 * We aren't the primary crash CPU. We need to send it 478 * to a holding pattern to avoid it ending up in the panic 479 * code. 480 */ 481 crash_kexec_secondary(regs); 482 483 /* 484 * No debugger or crash dump registered, print logs then 485 * panic. 486 */ 487 die("System Reset", regs, SIGABRT); 488 489 mdelay(2*MSEC_PER_SEC); /* Wait a little while for others to print */ 490 add_taint(TAINT_DIE, LOCKDEP_NOW_UNRELIABLE); 491 nmi_panic(regs, "System Reset"); 492 493 out: 494 #ifdef CONFIG_PPC_BOOK3S_64 495 BUG_ON(get_paca()->in_nmi == 0); 496 if (get_paca()->in_nmi > 1) 497 die("Unrecoverable nested System Reset", regs, SIGABRT); 498 #endif 499 /* Must die if the interrupt is not recoverable */ 500 if (regs_is_unrecoverable(regs)) { 501 /* For the reason explained in die_mce, nmi_exit before die */ 502 nmi_exit(); 503 die("Unrecoverable System Reset", regs, SIGABRT); 504 } 505 506 if (saved_hsrrs) { 507 mtspr(SPRN_HSRR0, hsrr0); 508 mtspr(SPRN_HSRR1, hsrr1); 509 } 510 511 /* What should we do here? We could issue a shutdown or hard reset. */ 512 513 return 0; 514 } 515 516 /* 517 * I/O accesses can cause machine checks on powermacs. 518 * Check if the NIP corresponds to the address of a sync 519 * instruction for which there is an entry in the exception 520 * table. 521 * -- paulus. 522 */ 523 static inline int check_io_access(struct pt_regs *regs) 524 { 525 #ifdef CONFIG_PPC32 526 unsigned long msr = regs->msr; 527 const struct exception_table_entry *entry; 528 unsigned int *nip = (unsigned int *)regs->nip; 529 530 if (((msr & 0xffff0000) == 0 || (msr & (0x80000 | 0x40000))) 531 && (entry = search_exception_tables(regs->nip)) != NULL) { 532 /* 533 * Check that it's a sync instruction, or somewhere 534 * in the twi; isync; nop sequence that inb/inw/inl uses. 535 * As the address is in the exception table 536 * we should be able to read the instr there. 537 * For the debug message, we look at the preceding 538 * load or store. 539 */ 540 if (*nip == PPC_RAW_NOP()) 541 nip -= 2; 542 else if (*nip == PPC_RAW_ISYNC()) 543 --nip; 544 if (*nip == PPC_RAW_SYNC() || get_op(*nip) == OP_TRAP) { 545 unsigned int rb; 546 547 --nip; 548 rb = (*nip >> 11) & 0x1f; 549 printk(KERN_DEBUG "%s bad port %lx at %p\n", 550 (*nip & 0x100)? "OUT to": "IN from", 551 regs->gpr[rb] - _IO_BASE, nip); 552 regs_set_recoverable(regs); 553 regs_set_return_ip(regs, extable_fixup(entry)); 554 return 1; 555 } 556 } 557 #endif /* CONFIG_PPC32 */ 558 return 0; 559 } 560 561 #ifdef CONFIG_PPC_ADV_DEBUG_REGS 562 /* On 4xx, the reason for the machine check or program exception 563 is in the ESR. */ 564 #define get_reason(regs) ((regs)->esr) 565 #define REASON_FP ESR_FP 566 #define REASON_ILLEGAL (ESR_PIL | ESR_PUO) 567 #define REASON_PRIVILEGED ESR_PPR 568 #define REASON_TRAP ESR_PTR 569 #define REASON_PREFIXED 0 570 #define REASON_BOUNDARY 0 571 572 /* single-step stuff */ 573 #define single_stepping(regs) (current->thread.debug.dbcr0 & DBCR0_IC) 574 #define clear_single_step(regs) (current->thread.debug.dbcr0 &= ~DBCR0_IC) 575 #define clear_br_trace(regs) do {} while(0) 576 #else 577 /* On non-4xx, the reason for the machine check or program 578 exception is in the MSR. */ 579 #define get_reason(regs) ((regs)->msr) 580 #define REASON_TM SRR1_PROGTM 581 #define REASON_FP SRR1_PROGFPE 582 #define REASON_ILLEGAL SRR1_PROGILL 583 #define REASON_PRIVILEGED SRR1_PROGPRIV 584 #define REASON_TRAP SRR1_PROGTRAP 585 #define REASON_PREFIXED SRR1_PREFIXED 586 #define REASON_BOUNDARY SRR1_BOUNDARY 587 588 #define single_stepping(regs) ((regs)->msr & MSR_SE) 589 #define clear_single_step(regs) (regs_set_return_msr((regs), (regs)->msr & ~MSR_SE)) 590 #define clear_br_trace(regs) (regs_set_return_msr((regs), (regs)->msr & ~MSR_BE)) 591 #endif 592 593 #define inst_length(reason) (((reason) & REASON_PREFIXED) ? 8 : 4) 594 595 #if defined(CONFIG_E500) 596 int machine_check_e500mc(struct pt_regs *regs) 597 { 598 unsigned long mcsr = mfspr(SPRN_MCSR); 599 unsigned long pvr = mfspr(SPRN_PVR); 600 unsigned long reason = mcsr; 601 int recoverable = 1; 602 603 if (reason & MCSR_LD) { 604 recoverable = fsl_rio_mcheck_exception(regs); 605 if (recoverable == 1) 606 goto silent_out; 607 } 608 609 printk("Machine check in kernel mode.\n"); 610 printk("Caused by (from MCSR=%lx): ", reason); 611 612 if (reason & MCSR_MCP) 613 pr_cont("Machine Check Signal\n"); 614 615 if (reason & MCSR_ICPERR) { 616 pr_cont("Instruction Cache Parity Error\n"); 617 618 /* 619 * This is recoverable by invalidating the i-cache. 620 */ 621 mtspr(SPRN_L1CSR1, mfspr(SPRN_L1CSR1) | L1CSR1_ICFI); 622 while (mfspr(SPRN_L1CSR1) & L1CSR1_ICFI) 623 ; 624 625 /* 626 * This will generally be accompanied by an instruction 627 * fetch error report -- only treat MCSR_IF as fatal 628 * if it wasn't due to an L1 parity error. 629 */ 630 reason &= ~MCSR_IF; 631 } 632 633 if (reason & MCSR_DCPERR_MC) { 634 pr_cont("Data Cache Parity Error\n"); 635 636 /* 637 * In write shadow mode we auto-recover from the error, but it 638 * may still get logged and cause a machine check. We should 639 * only treat the non-write shadow case as non-recoverable. 640 */ 641 /* On e6500 core, L1 DCWS (Data cache write shadow mode) bit 642 * is not implemented but L1 data cache always runs in write 643 * shadow mode. Hence on data cache parity errors HW will 644 * automatically invalidate the L1 Data Cache. 645 */ 646 if (PVR_VER(pvr) != PVR_VER_E6500) { 647 if (!(mfspr(SPRN_L1CSR2) & L1CSR2_DCWS)) 648 recoverable = 0; 649 } 650 } 651 652 if (reason & MCSR_L2MMU_MHIT) { 653 pr_cont("Hit on multiple TLB entries\n"); 654 recoverable = 0; 655 } 656 657 if (reason & MCSR_NMI) 658 pr_cont("Non-maskable interrupt\n"); 659 660 if (reason & MCSR_IF) { 661 pr_cont("Instruction Fetch Error Report\n"); 662 recoverable = 0; 663 } 664 665 if (reason & MCSR_LD) { 666 pr_cont("Load Error Report\n"); 667 recoverable = 0; 668 } 669 670 if (reason & MCSR_ST) { 671 pr_cont("Store Error Report\n"); 672 recoverable = 0; 673 } 674 675 if (reason & MCSR_LDG) { 676 pr_cont("Guarded Load Error Report\n"); 677 recoverable = 0; 678 } 679 680 if (reason & MCSR_TLBSYNC) 681 pr_cont("Simultaneous tlbsync operations\n"); 682 683 if (reason & MCSR_BSL2_ERR) { 684 pr_cont("Level 2 Cache Error\n"); 685 recoverable = 0; 686 } 687 688 if (reason & MCSR_MAV) { 689 u64 addr; 690 691 addr = mfspr(SPRN_MCAR); 692 addr |= (u64)mfspr(SPRN_MCARU) << 32; 693 694 pr_cont("Machine Check %s Address: %#llx\n", 695 reason & MCSR_MEA ? "Effective" : "Physical", addr); 696 } 697 698 silent_out: 699 mtspr(SPRN_MCSR, mcsr); 700 return mfspr(SPRN_MCSR) == 0 && recoverable; 701 } 702 703 int machine_check_e500(struct pt_regs *regs) 704 { 705 unsigned long reason = mfspr(SPRN_MCSR); 706 707 if (reason & MCSR_BUS_RBERR) { 708 if (fsl_rio_mcheck_exception(regs)) 709 return 1; 710 if (fsl_pci_mcheck_exception(regs)) 711 return 1; 712 } 713 714 printk("Machine check in kernel mode.\n"); 715 printk("Caused by (from MCSR=%lx): ", reason); 716 717 if (reason & MCSR_MCP) 718 pr_cont("Machine Check Signal\n"); 719 if (reason & MCSR_ICPERR) 720 pr_cont("Instruction Cache Parity Error\n"); 721 if (reason & MCSR_DCP_PERR) 722 pr_cont("Data Cache Push Parity Error\n"); 723 if (reason & MCSR_DCPERR) 724 pr_cont("Data Cache Parity Error\n"); 725 if (reason & MCSR_BUS_IAERR) 726 pr_cont("Bus - Instruction Address Error\n"); 727 if (reason & MCSR_BUS_RAERR) 728 pr_cont("Bus - Read Address Error\n"); 729 if (reason & MCSR_BUS_WAERR) 730 pr_cont("Bus - Write Address Error\n"); 731 if (reason & MCSR_BUS_IBERR) 732 pr_cont("Bus - Instruction Data Error\n"); 733 if (reason & MCSR_BUS_RBERR) 734 pr_cont("Bus - Read Data Bus Error\n"); 735 if (reason & MCSR_BUS_WBERR) 736 pr_cont("Bus - Write Data Bus Error\n"); 737 if (reason & MCSR_BUS_IPERR) 738 pr_cont("Bus - Instruction Parity Error\n"); 739 if (reason & MCSR_BUS_RPERR) 740 pr_cont("Bus - Read Parity Error\n"); 741 742 return 0; 743 } 744 745 int machine_check_generic(struct pt_regs *regs) 746 { 747 return 0; 748 } 749 #elif defined(CONFIG_PPC32) 750 int machine_check_generic(struct pt_regs *regs) 751 { 752 unsigned long reason = regs->msr; 753 754 printk("Machine check in kernel mode.\n"); 755 printk("Caused by (from SRR1=%lx): ", reason); 756 switch (reason & 0x601F0000) { 757 case 0x80000: 758 pr_cont("Machine check signal\n"); 759 break; 760 case 0x40000: 761 case 0x140000: /* 7450 MSS error and TEA */ 762 pr_cont("Transfer error ack signal\n"); 763 break; 764 case 0x20000: 765 pr_cont("Data parity error signal\n"); 766 break; 767 case 0x10000: 768 pr_cont("Address parity error signal\n"); 769 break; 770 case 0x20000000: 771 pr_cont("L1 Data Cache error\n"); 772 break; 773 case 0x40000000: 774 pr_cont("L1 Instruction Cache error\n"); 775 break; 776 case 0x00100000: 777 pr_cont("L2 data cache parity error\n"); 778 break; 779 default: 780 pr_cont("Unknown values in msr\n"); 781 } 782 return 0; 783 } 784 #endif /* everything else */ 785 786 void die_mce(const char *str, struct pt_regs *regs, long err) 787 { 788 /* 789 * The machine check wants to kill the interrupted context, but 790 * do_exit() checks for in_interrupt() and panics in that case, so 791 * exit the irq/nmi before calling die. 792 */ 793 if (IS_ENABLED(CONFIG_PPC_BOOK3S_64)) 794 irq_exit(); 795 else 796 nmi_exit(); 797 die(str, regs, err); 798 } 799 800 /* 801 * BOOK3S_64 does not call this handler as a non-maskable interrupt 802 * (it uses its own early real-mode handler to handle the MCE proper 803 * and then raises irq_work to call this handler when interrupts are 804 * enabled). 805 */ 806 #ifdef CONFIG_PPC_BOOK3S_64 807 DEFINE_INTERRUPT_HANDLER_ASYNC(machine_check_exception) 808 #else 809 DEFINE_INTERRUPT_HANDLER_NMI(machine_check_exception) 810 #endif 811 { 812 int recover = 0; 813 814 __this_cpu_inc(irq_stat.mce_exceptions); 815 816 add_taint(TAINT_MACHINE_CHECK, LOCKDEP_NOW_UNRELIABLE); 817 818 /* See if any machine dependent calls. In theory, we would want 819 * to call the CPU first, and call the ppc_md. one if the CPU 820 * one returns a positive number. However there is existing code 821 * that assumes the board gets a first chance, so let's keep it 822 * that way for now and fix things later. --BenH. 823 */ 824 if (ppc_md.machine_check_exception) 825 recover = ppc_md.machine_check_exception(regs); 826 else if (cur_cpu_spec->machine_check) 827 recover = cur_cpu_spec->machine_check(regs); 828 829 if (recover > 0) 830 goto bail; 831 832 if (debugger_fault_handler(regs)) 833 goto bail; 834 835 if (check_io_access(regs)) 836 goto bail; 837 838 die_mce("Machine check", regs, SIGBUS); 839 840 bail: 841 /* Must die if the interrupt is not recoverable */ 842 if (regs_is_unrecoverable(regs)) 843 die_mce("Unrecoverable Machine check", regs, SIGBUS); 844 845 #ifdef CONFIG_PPC_BOOK3S_64 846 return; 847 #else 848 return 0; 849 #endif 850 } 851 852 DEFINE_INTERRUPT_HANDLER(SMIException) /* async? */ 853 { 854 die("System Management Interrupt", regs, SIGABRT); 855 } 856 857 #ifdef CONFIG_VSX 858 static void p9_hmi_special_emu(struct pt_regs *regs) 859 { 860 unsigned int ra, rb, t, i, sel, instr, rc; 861 const void __user *addr; 862 u8 vbuf[16] __aligned(16), *vdst; 863 unsigned long ea, msr, msr_mask; 864 bool swap; 865 866 if (__get_user(instr, (unsigned int __user *)regs->nip)) 867 return; 868 869 /* 870 * lxvb16x opcode: 0x7c0006d8 871 * lxvd2x opcode: 0x7c000698 872 * lxvh8x opcode: 0x7c000658 873 * lxvw4x opcode: 0x7c000618 874 */ 875 if ((instr & 0xfc00073e) != 0x7c000618) { 876 pr_devel("HMI vec emu: not vector CI %i:%s[%d] nip=%016lx" 877 " instr=%08x\n", 878 smp_processor_id(), current->comm, current->pid, 879 regs->nip, instr); 880 return; 881 } 882 883 /* Grab vector registers into the task struct */ 884 msr = regs->msr; /* Grab msr before we flush the bits */ 885 flush_vsx_to_thread(current); 886 enable_kernel_altivec(); 887 888 /* 889 * Is userspace running with a different endian (this is rare but 890 * not impossible) 891 */ 892 swap = (msr & MSR_LE) != (MSR_KERNEL & MSR_LE); 893 894 /* Decode the instruction */ 895 ra = (instr >> 16) & 0x1f; 896 rb = (instr >> 11) & 0x1f; 897 t = (instr >> 21) & 0x1f; 898 if (instr & 1) 899 vdst = (u8 *)¤t->thread.vr_state.vr[t]; 900 else 901 vdst = (u8 *)¤t->thread.fp_state.fpr[t][0]; 902 903 /* Grab the vector address */ 904 ea = regs->gpr[rb] + (ra ? regs->gpr[ra] : 0); 905 if (is_32bit_task()) 906 ea &= 0xfffffffful; 907 addr = (__force const void __user *)ea; 908 909 /* Check it */ 910 if (!access_ok(addr, 16)) { 911 pr_devel("HMI vec emu: bad access %i:%s[%d] nip=%016lx" 912 " instr=%08x addr=%016lx\n", 913 smp_processor_id(), current->comm, current->pid, 914 regs->nip, instr, (unsigned long)addr); 915 return; 916 } 917 918 /* Read the vector */ 919 rc = 0; 920 if ((unsigned long)addr & 0xfUL) 921 /* unaligned case */ 922 rc = __copy_from_user_inatomic(vbuf, addr, 16); 923 else 924 __get_user_atomic_128_aligned(vbuf, addr, rc); 925 if (rc) { 926 pr_devel("HMI vec emu: page fault %i:%s[%d] nip=%016lx" 927 " instr=%08x addr=%016lx\n", 928 smp_processor_id(), current->comm, current->pid, 929 regs->nip, instr, (unsigned long)addr); 930 return; 931 } 932 933 pr_devel("HMI vec emu: emulated vector CI %i:%s[%d] nip=%016lx" 934 " instr=%08x addr=%016lx\n", 935 smp_processor_id(), current->comm, current->pid, regs->nip, 936 instr, (unsigned long) addr); 937 938 /* Grab instruction "selector" */ 939 sel = (instr >> 6) & 3; 940 941 /* 942 * Check to make sure the facility is actually enabled. This 943 * could happen if we get a false positive hit. 944 * 945 * lxvd2x/lxvw4x always check MSR VSX sel = 0,2 946 * lxvh8x/lxvb16x check MSR VSX or VEC depending on VSR used sel = 1,3 947 */ 948 msr_mask = MSR_VSX; 949 if ((sel & 1) && (instr & 1)) /* lxvh8x & lxvb16x + VSR >= 32 */ 950 msr_mask = MSR_VEC; 951 if (!(msr & msr_mask)) { 952 pr_devel("HMI vec emu: MSR fac clear %i:%s[%d] nip=%016lx" 953 " instr=%08x msr:%016lx\n", 954 smp_processor_id(), current->comm, current->pid, 955 regs->nip, instr, msr); 956 return; 957 } 958 959 /* Do logging here before we modify sel based on endian */ 960 switch (sel) { 961 case 0: /* lxvw4x */ 962 PPC_WARN_EMULATED(lxvw4x, regs); 963 break; 964 case 1: /* lxvh8x */ 965 PPC_WARN_EMULATED(lxvh8x, regs); 966 break; 967 case 2: /* lxvd2x */ 968 PPC_WARN_EMULATED(lxvd2x, regs); 969 break; 970 case 3: /* lxvb16x */ 971 PPC_WARN_EMULATED(lxvb16x, regs); 972 break; 973 } 974 975 #ifdef __LITTLE_ENDIAN__ 976 /* 977 * An LE kernel stores the vector in the task struct as an LE 978 * byte array (effectively swapping both the components and 979 * the content of the components). Those instructions expect 980 * the components to remain in ascending address order, so we 981 * swap them back. 982 * 983 * If we are running a BE user space, the expectation is that 984 * of a simple memcpy, so forcing the emulation to look like 985 * a lxvb16x should do the trick. 986 */ 987 if (swap) 988 sel = 3; 989 990 switch (sel) { 991 case 0: /* lxvw4x */ 992 for (i = 0; i < 4; i++) 993 ((u32 *)vdst)[i] = ((u32 *)vbuf)[3-i]; 994 break; 995 case 1: /* lxvh8x */ 996 for (i = 0; i < 8; i++) 997 ((u16 *)vdst)[i] = ((u16 *)vbuf)[7-i]; 998 break; 999 case 2: /* lxvd2x */ 1000 for (i = 0; i < 2; i++) 1001 ((u64 *)vdst)[i] = ((u64 *)vbuf)[1-i]; 1002 break; 1003 case 3: /* lxvb16x */ 1004 for (i = 0; i < 16; i++) 1005 vdst[i] = vbuf[15-i]; 1006 break; 1007 } 1008 #else /* __LITTLE_ENDIAN__ */ 1009 /* On a big endian kernel, a BE userspace only needs a memcpy */ 1010 if (!swap) 1011 sel = 3; 1012 1013 /* Otherwise, we need to swap the content of the components */ 1014 switch (sel) { 1015 case 0: /* lxvw4x */ 1016 for (i = 0; i < 4; i++) 1017 ((u32 *)vdst)[i] = cpu_to_le32(((u32 *)vbuf)[i]); 1018 break; 1019 case 1: /* lxvh8x */ 1020 for (i = 0; i < 8; i++) 1021 ((u16 *)vdst)[i] = cpu_to_le16(((u16 *)vbuf)[i]); 1022 break; 1023 case 2: /* lxvd2x */ 1024 for (i = 0; i < 2; i++) 1025 ((u64 *)vdst)[i] = cpu_to_le64(((u64 *)vbuf)[i]); 1026 break; 1027 case 3: /* lxvb16x */ 1028 memcpy(vdst, vbuf, 16); 1029 break; 1030 } 1031 #endif /* !__LITTLE_ENDIAN__ */ 1032 1033 /* Go to next instruction */ 1034 regs_add_return_ip(regs, 4); 1035 } 1036 #endif /* CONFIG_VSX */ 1037 1038 DEFINE_INTERRUPT_HANDLER_ASYNC(handle_hmi_exception) 1039 { 1040 struct pt_regs *old_regs; 1041 1042 old_regs = set_irq_regs(regs); 1043 1044 #ifdef CONFIG_VSX 1045 /* Real mode flagged P9 special emu is needed */ 1046 if (local_paca->hmi_p9_special_emu) { 1047 local_paca->hmi_p9_special_emu = 0; 1048 1049 /* 1050 * We don't want to take page faults while doing the 1051 * emulation, we just replay the instruction if necessary. 1052 */ 1053 pagefault_disable(); 1054 p9_hmi_special_emu(regs); 1055 pagefault_enable(); 1056 } 1057 #endif /* CONFIG_VSX */ 1058 1059 if (ppc_md.handle_hmi_exception) 1060 ppc_md.handle_hmi_exception(regs); 1061 1062 set_irq_regs(old_regs); 1063 } 1064 1065 DEFINE_INTERRUPT_HANDLER(unknown_exception) 1066 { 1067 printk("Bad trap at PC: %lx, SR: %lx, vector=%lx\n", 1068 regs->nip, regs->msr, regs->trap); 1069 1070 _exception(SIGTRAP, regs, TRAP_UNK, 0); 1071 } 1072 1073 DEFINE_INTERRUPT_HANDLER_ASYNC(unknown_async_exception) 1074 { 1075 printk("Bad trap at PC: %lx, SR: %lx, vector=%lx\n", 1076 regs->nip, regs->msr, regs->trap); 1077 1078 _exception(SIGTRAP, regs, TRAP_UNK, 0); 1079 } 1080 1081 DEFINE_INTERRUPT_HANDLER_NMI(unknown_nmi_exception) 1082 { 1083 printk("Bad trap at PC: %lx, SR: %lx, vector=%lx\n", 1084 regs->nip, regs->msr, regs->trap); 1085 1086 _exception(SIGTRAP, regs, TRAP_UNK, 0); 1087 1088 return 0; 1089 } 1090 1091 DEFINE_INTERRUPT_HANDLER(instruction_breakpoint_exception) 1092 { 1093 if (notify_die(DIE_IABR_MATCH, "iabr_match", regs, 5, 1094 5, SIGTRAP) == NOTIFY_STOP) 1095 return; 1096 if (debugger_iabr_match(regs)) 1097 return; 1098 _exception(SIGTRAP, regs, TRAP_BRKPT, regs->nip); 1099 } 1100 1101 DEFINE_INTERRUPT_HANDLER(RunModeException) 1102 { 1103 _exception(SIGTRAP, regs, TRAP_UNK, 0); 1104 } 1105 1106 static void __single_step_exception(struct pt_regs *regs) 1107 { 1108 clear_single_step(regs); 1109 clear_br_trace(regs); 1110 1111 if (kprobe_post_handler(regs)) 1112 return; 1113 1114 if (notify_die(DIE_SSTEP, "single_step", regs, 5, 1115 5, SIGTRAP) == NOTIFY_STOP) 1116 return; 1117 if (debugger_sstep(regs)) 1118 return; 1119 1120 _exception(SIGTRAP, regs, TRAP_TRACE, regs->nip); 1121 } 1122 1123 DEFINE_INTERRUPT_HANDLER(single_step_exception) 1124 { 1125 __single_step_exception(regs); 1126 } 1127 1128 /* 1129 * After we have successfully emulated an instruction, we have to 1130 * check if the instruction was being single-stepped, and if so, 1131 * pretend we got a single-step exception. This was pointed out 1132 * by Kumar Gala. -- paulus 1133 */ 1134 static void emulate_single_step(struct pt_regs *regs) 1135 { 1136 if (single_stepping(regs)) 1137 __single_step_exception(regs); 1138 } 1139 1140 static inline int __parse_fpscr(unsigned long fpscr) 1141 { 1142 int ret = FPE_FLTUNK; 1143 1144 /* Invalid operation */ 1145 if ((fpscr & FPSCR_VE) && (fpscr & FPSCR_VX)) 1146 ret = FPE_FLTINV; 1147 1148 /* Overflow */ 1149 else if ((fpscr & FPSCR_OE) && (fpscr & FPSCR_OX)) 1150 ret = FPE_FLTOVF; 1151 1152 /* Underflow */ 1153 else if ((fpscr & FPSCR_UE) && (fpscr & FPSCR_UX)) 1154 ret = FPE_FLTUND; 1155 1156 /* Divide by zero */ 1157 else if ((fpscr & FPSCR_ZE) && (fpscr & FPSCR_ZX)) 1158 ret = FPE_FLTDIV; 1159 1160 /* Inexact result */ 1161 else if ((fpscr & FPSCR_XE) && (fpscr & FPSCR_XX)) 1162 ret = FPE_FLTRES; 1163 1164 return ret; 1165 } 1166 1167 static void parse_fpe(struct pt_regs *regs) 1168 { 1169 int code = 0; 1170 1171 flush_fp_to_thread(current); 1172 1173 #ifdef CONFIG_PPC_FPU_REGS 1174 code = __parse_fpscr(current->thread.fp_state.fpscr); 1175 #endif 1176 1177 _exception(SIGFPE, regs, code, regs->nip); 1178 } 1179 1180 /* 1181 * Illegal instruction emulation support. Originally written to 1182 * provide the PVR to user applications using the mfspr rd, PVR. 1183 * Return non-zero if we can't emulate, or -EFAULT if the associated 1184 * memory access caused an access fault. Return zero on success. 1185 * 1186 * There are a couple of ways to do this, either "decode" the instruction 1187 * or directly match lots of bits. In this case, matching lots of 1188 * bits is faster and easier. 1189 * 1190 */ 1191 static int emulate_string_inst(struct pt_regs *regs, u32 instword) 1192 { 1193 u8 rT = (instword >> 21) & 0x1f; 1194 u8 rA = (instword >> 16) & 0x1f; 1195 u8 NB_RB = (instword >> 11) & 0x1f; 1196 u32 num_bytes; 1197 unsigned long EA; 1198 int pos = 0; 1199 1200 /* Early out if we are an invalid form of lswx */ 1201 if ((instword & PPC_INST_STRING_MASK) == PPC_INST_LSWX) 1202 if ((rT == rA) || (rT == NB_RB)) 1203 return -EINVAL; 1204 1205 EA = (rA == 0) ? 0 : regs->gpr[rA]; 1206 1207 switch (instword & PPC_INST_STRING_MASK) { 1208 case PPC_INST_LSWX: 1209 case PPC_INST_STSWX: 1210 EA += NB_RB; 1211 num_bytes = regs->xer & 0x7f; 1212 break; 1213 case PPC_INST_LSWI: 1214 case PPC_INST_STSWI: 1215 num_bytes = (NB_RB == 0) ? 32 : NB_RB; 1216 break; 1217 default: 1218 return -EINVAL; 1219 } 1220 1221 while (num_bytes != 0) 1222 { 1223 u8 val; 1224 u32 shift = 8 * (3 - (pos & 0x3)); 1225 1226 /* if process is 32-bit, clear upper 32 bits of EA */ 1227 if ((regs->msr & MSR_64BIT) == 0) 1228 EA &= 0xFFFFFFFF; 1229 1230 switch ((instword & PPC_INST_STRING_MASK)) { 1231 case PPC_INST_LSWX: 1232 case PPC_INST_LSWI: 1233 if (get_user(val, (u8 __user *)EA)) 1234 return -EFAULT; 1235 /* first time updating this reg, 1236 * zero it out */ 1237 if (pos == 0) 1238 regs->gpr[rT] = 0; 1239 regs->gpr[rT] |= val << shift; 1240 break; 1241 case PPC_INST_STSWI: 1242 case PPC_INST_STSWX: 1243 val = regs->gpr[rT] >> shift; 1244 if (put_user(val, (u8 __user *)EA)) 1245 return -EFAULT; 1246 break; 1247 } 1248 /* move EA to next address */ 1249 EA += 1; 1250 num_bytes--; 1251 1252 /* manage our position within the register */ 1253 if (++pos == 4) { 1254 pos = 0; 1255 if (++rT == 32) 1256 rT = 0; 1257 } 1258 } 1259 1260 return 0; 1261 } 1262 1263 static int emulate_popcntb_inst(struct pt_regs *regs, u32 instword) 1264 { 1265 u32 ra,rs; 1266 unsigned long tmp; 1267 1268 ra = (instword >> 16) & 0x1f; 1269 rs = (instword >> 21) & 0x1f; 1270 1271 tmp = regs->gpr[rs]; 1272 tmp = tmp - ((tmp >> 1) & 0x5555555555555555ULL); 1273 tmp = (tmp & 0x3333333333333333ULL) + ((tmp >> 2) & 0x3333333333333333ULL); 1274 tmp = (tmp + (tmp >> 4)) & 0x0f0f0f0f0f0f0f0fULL; 1275 regs->gpr[ra] = tmp; 1276 1277 return 0; 1278 } 1279 1280 static int emulate_isel(struct pt_regs *regs, u32 instword) 1281 { 1282 u8 rT = (instword >> 21) & 0x1f; 1283 u8 rA = (instword >> 16) & 0x1f; 1284 u8 rB = (instword >> 11) & 0x1f; 1285 u8 BC = (instword >> 6) & 0x1f; 1286 u8 bit; 1287 unsigned long tmp; 1288 1289 tmp = (rA == 0) ? 0 : regs->gpr[rA]; 1290 bit = (regs->ccr >> (31 - BC)) & 0x1; 1291 1292 regs->gpr[rT] = bit ? tmp : regs->gpr[rB]; 1293 1294 return 0; 1295 } 1296 1297 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM 1298 static inline bool tm_abort_check(struct pt_regs *regs, int cause) 1299 { 1300 /* If we're emulating a load/store in an active transaction, we cannot 1301 * emulate it as the kernel operates in transaction suspended context. 1302 * We need to abort the transaction. This creates a persistent TM 1303 * abort so tell the user what caused it with a new code. 1304 */ 1305 if (MSR_TM_TRANSACTIONAL(regs->msr)) { 1306 tm_enable(); 1307 tm_abort(cause); 1308 return true; 1309 } 1310 return false; 1311 } 1312 #else 1313 static inline bool tm_abort_check(struct pt_regs *regs, int reason) 1314 { 1315 return false; 1316 } 1317 #endif 1318 1319 static int emulate_instruction(struct pt_regs *regs) 1320 { 1321 u32 instword; 1322 u32 rd; 1323 1324 if (!user_mode(regs)) 1325 return -EINVAL; 1326 1327 if (get_user(instword, (u32 __user *)(regs->nip))) 1328 return -EFAULT; 1329 1330 /* Emulate the mfspr rD, PVR. */ 1331 if ((instword & PPC_INST_MFSPR_PVR_MASK) == PPC_INST_MFSPR_PVR) { 1332 PPC_WARN_EMULATED(mfpvr, regs); 1333 rd = (instword >> 21) & 0x1f; 1334 regs->gpr[rd] = mfspr(SPRN_PVR); 1335 return 0; 1336 } 1337 1338 /* Emulating the dcba insn is just a no-op. */ 1339 if ((instword & PPC_INST_DCBA_MASK) == PPC_INST_DCBA) { 1340 PPC_WARN_EMULATED(dcba, regs); 1341 return 0; 1342 } 1343 1344 /* Emulate the mcrxr insn. */ 1345 if ((instword & PPC_INST_MCRXR_MASK) == PPC_INST_MCRXR) { 1346 int shift = (instword >> 21) & 0x1c; 1347 unsigned long msk = 0xf0000000UL >> shift; 1348 1349 PPC_WARN_EMULATED(mcrxr, regs); 1350 regs->ccr = (regs->ccr & ~msk) | ((regs->xer >> shift) & msk); 1351 regs->xer &= ~0xf0000000UL; 1352 return 0; 1353 } 1354 1355 /* Emulate load/store string insn. */ 1356 if ((instword & PPC_INST_STRING_GEN_MASK) == PPC_INST_STRING) { 1357 if (tm_abort_check(regs, 1358 TM_CAUSE_EMULATE | TM_CAUSE_PERSISTENT)) 1359 return -EINVAL; 1360 PPC_WARN_EMULATED(string, regs); 1361 return emulate_string_inst(regs, instword); 1362 } 1363 1364 /* Emulate the popcntb (Population Count Bytes) instruction. */ 1365 if ((instword & PPC_INST_POPCNTB_MASK) == PPC_INST_POPCNTB) { 1366 PPC_WARN_EMULATED(popcntb, regs); 1367 return emulate_popcntb_inst(regs, instword); 1368 } 1369 1370 /* Emulate isel (Integer Select) instruction */ 1371 if ((instword & PPC_INST_ISEL_MASK) == PPC_INST_ISEL) { 1372 PPC_WARN_EMULATED(isel, regs); 1373 return emulate_isel(regs, instword); 1374 } 1375 1376 /* Emulate sync instruction variants */ 1377 if ((instword & PPC_INST_SYNC_MASK) == PPC_INST_SYNC) { 1378 PPC_WARN_EMULATED(sync, regs); 1379 asm volatile("sync"); 1380 return 0; 1381 } 1382 1383 #ifdef CONFIG_PPC64 1384 /* Emulate the mfspr rD, DSCR. */ 1385 if ((((instword & PPC_INST_MFSPR_DSCR_USER_MASK) == 1386 PPC_INST_MFSPR_DSCR_USER) || 1387 ((instword & PPC_INST_MFSPR_DSCR_MASK) == 1388 PPC_INST_MFSPR_DSCR)) && 1389 cpu_has_feature(CPU_FTR_DSCR)) { 1390 PPC_WARN_EMULATED(mfdscr, regs); 1391 rd = (instword >> 21) & 0x1f; 1392 regs->gpr[rd] = mfspr(SPRN_DSCR); 1393 return 0; 1394 } 1395 /* Emulate the mtspr DSCR, rD. */ 1396 if ((((instword & PPC_INST_MTSPR_DSCR_USER_MASK) == 1397 PPC_INST_MTSPR_DSCR_USER) || 1398 ((instword & PPC_INST_MTSPR_DSCR_MASK) == 1399 PPC_INST_MTSPR_DSCR)) && 1400 cpu_has_feature(CPU_FTR_DSCR)) { 1401 PPC_WARN_EMULATED(mtdscr, regs); 1402 rd = (instword >> 21) & 0x1f; 1403 current->thread.dscr = regs->gpr[rd]; 1404 current->thread.dscr_inherit = 1; 1405 mtspr(SPRN_DSCR, current->thread.dscr); 1406 return 0; 1407 } 1408 #endif 1409 1410 return -EINVAL; 1411 } 1412 1413 int is_valid_bugaddr(unsigned long addr) 1414 { 1415 return is_kernel_addr(addr); 1416 } 1417 1418 #ifdef CONFIG_MATH_EMULATION 1419 static int emulate_math(struct pt_regs *regs) 1420 { 1421 int ret; 1422 1423 ret = do_mathemu(regs); 1424 if (ret >= 0) 1425 PPC_WARN_EMULATED(math, regs); 1426 1427 switch (ret) { 1428 case 0: 1429 emulate_single_step(regs); 1430 return 0; 1431 case 1: { 1432 int code = 0; 1433 code = __parse_fpscr(current->thread.fp_state.fpscr); 1434 _exception(SIGFPE, regs, code, regs->nip); 1435 return 0; 1436 } 1437 case -EFAULT: 1438 _exception(SIGSEGV, regs, SEGV_MAPERR, regs->nip); 1439 return 0; 1440 } 1441 1442 return -1; 1443 } 1444 #else 1445 static inline int emulate_math(struct pt_regs *regs) { return -1; } 1446 #endif 1447 1448 static void do_program_check(struct pt_regs *regs) 1449 { 1450 unsigned int reason = get_reason(regs); 1451 1452 /* We can now get here via a FP Unavailable exception if the core 1453 * has no FPU, in that case the reason flags will be 0 */ 1454 1455 if (reason & REASON_FP) { 1456 /* IEEE FP exception */ 1457 parse_fpe(regs); 1458 return; 1459 } 1460 if (reason & REASON_TRAP) { 1461 unsigned long bugaddr; 1462 /* Debugger is first in line to stop recursive faults in 1463 * rcu_lock, notify_die, or atomic_notifier_call_chain */ 1464 if (debugger_bpt(regs)) 1465 return; 1466 1467 if (kprobe_handler(regs)) 1468 return; 1469 1470 /* trap exception */ 1471 if (notify_die(DIE_BPT, "breakpoint", regs, 5, 5, SIGTRAP) 1472 == NOTIFY_STOP) 1473 return; 1474 1475 bugaddr = regs->nip; 1476 /* 1477 * Fixup bugaddr for BUG_ON() in real mode 1478 */ 1479 if (!is_kernel_addr(bugaddr) && !(regs->msr & MSR_IR)) 1480 bugaddr += PAGE_OFFSET; 1481 1482 if (!(regs->msr & MSR_PR) && /* not user-mode */ 1483 report_bug(bugaddr, regs) == BUG_TRAP_TYPE_WARN) { 1484 const struct exception_table_entry *entry; 1485 1486 entry = search_exception_tables(bugaddr); 1487 if (entry) { 1488 regs_set_return_ip(regs, extable_fixup(entry) + regs->nip - bugaddr); 1489 return; 1490 } 1491 } 1492 _exception(SIGTRAP, regs, TRAP_BRKPT, regs->nip); 1493 return; 1494 } 1495 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM 1496 if (reason & REASON_TM) { 1497 /* This is a TM "Bad Thing Exception" program check. 1498 * This occurs when: 1499 * - An rfid/hrfid/mtmsrd attempts to cause an illegal 1500 * transition in TM states. 1501 * - A trechkpt is attempted when transactional. 1502 * - A treclaim is attempted when non transactional. 1503 * - A tend is illegally attempted. 1504 * - writing a TM SPR when transactional. 1505 * 1506 * If usermode caused this, it's done something illegal and 1507 * gets a SIGILL slap on the wrist. We call it an illegal 1508 * operand to distinguish from the instruction just being bad 1509 * (e.g. executing a 'tend' on a CPU without TM!); it's an 1510 * illegal /placement/ of a valid instruction. 1511 */ 1512 if (user_mode(regs)) { 1513 _exception(SIGILL, regs, ILL_ILLOPN, regs->nip); 1514 return; 1515 } else { 1516 printk(KERN_EMERG "Unexpected TM Bad Thing exception " 1517 "at %lx (msr 0x%lx) tm_scratch=%llx\n", 1518 regs->nip, regs->msr, get_paca()->tm_scratch); 1519 die("Unrecoverable exception", regs, SIGABRT); 1520 } 1521 } 1522 #endif 1523 1524 /* 1525 * If we took the program check in the kernel skip down to sending a 1526 * SIGILL. The subsequent cases all relate to emulating instructions 1527 * which we should only do for userspace. We also do not want to enable 1528 * interrupts for kernel faults because that might lead to further 1529 * faults, and loose the context of the original exception. 1530 */ 1531 if (!user_mode(regs)) 1532 goto sigill; 1533 1534 interrupt_cond_local_irq_enable(regs); 1535 1536 /* (reason & REASON_ILLEGAL) would be the obvious thing here, 1537 * but there seems to be a hardware bug on the 405GP (RevD) 1538 * that means ESR is sometimes set incorrectly - either to 1539 * ESR_DST (!?) or 0. In the process of chasing this with the 1540 * hardware people - not sure if it can happen on any illegal 1541 * instruction or only on FP instructions, whether there is a 1542 * pattern to occurrences etc. -dgibson 31/Mar/2003 1543 */ 1544 if (!emulate_math(regs)) 1545 return; 1546 1547 /* Try to emulate it if we should. */ 1548 if (reason & (REASON_ILLEGAL | REASON_PRIVILEGED)) { 1549 switch (emulate_instruction(regs)) { 1550 case 0: 1551 regs_add_return_ip(regs, 4); 1552 emulate_single_step(regs); 1553 return; 1554 case -EFAULT: 1555 _exception(SIGSEGV, regs, SEGV_MAPERR, regs->nip); 1556 return; 1557 } 1558 } 1559 1560 sigill: 1561 if (reason & REASON_PRIVILEGED) 1562 _exception(SIGILL, regs, ILL_PRVOPC, regs->nip); 1563 else 1564 _exception(SIGILL, regs, ILL_ILLOPC, regs->nip); 1565 1566 } 1567 1568 DEFINE_INTERRUPT_HANDLER(program_check_exception) 1569 { 1570 do_program_check(regs); 1571 } 1572 1573 /* 1574 * This occurs when running in hypervisor mode on POWER6 or later 1575 * and an illegal instruction is encountered. 1576 */ 1577 DEFINE_INTERRUPT_HANDLER(emulation_assist_interrupt) 1578 { 1579 regs_set_return_msr(regs, regs->msr | REASON_ILLEGAL); 1580 do_program_check(regs); 1581 } 1582 1583 DEFINE_INTERRUPT_HANDLER(alignment_exception) 1584 { 1585 int sig, code, fixed = 0; 1586 unsigned long reason; 1587 1588 interrupt_cond_local_irq_enable(regs); 1589 1590 reason = get_reason(regs); 1591 if (reason & REASON_BOUNDARY) { 1592 sig = SIGBUS; 1593 code = BUS_ADRALN; 1594 goto bad; 1595 } 1596 1597 if (tm_abort_check(regs, TM_CAUSE_ALIGNMENT | TM_CAUSE_PERSISTENT)) 1598 return; 1599 1600 /* we don't implement logging of alignment exceptions */ 1601 if (!(current->thread.align_ctl & PR_UNALIGN_SIGBUS)) 1602 fixed = fix_alignment(regs); 1603 1604 if (fixed == 1) { 1605 /* skip over emulated instruction */ 1606 regs_add_return_ip(regs, inst_length(reason)); 1607 emulate_single_step(regs); 1608 return; 1609 } 1610 1611 /* Operand address was bad */ 1612 if (fixed == -EFAULT) { 1613 sig = SIGSEGV; 1614 code = SEGV_ACCERR; 1615 } else { 1616 sig = SIGBUS; 1617 code = BUS_ADRALN; 1618 } 1619 bad: 1620 if (user_mode(regs)) 1621 _exception(sig, regs, code, regs->dar); 1622 else 1623 bad_page_fault(regs, sig); 1624 } 1625 1626 DEFINE_INTERRUPT_HANDLER(stack_overflow_exception) 1627 { 1628 die("Kernel stack overflow", regs, SIGSEGV); 1629 } 1630 1631 DEFINE_INTERRUPT_HANDLER(kernel_fp_unavailable_exception) 1632 { 1633 printk(KERN_EMERG "Unrecoverable FP Unavailable Exception " 1634 "%lx at %lx\n", regs->trap, regs->nip); 1635 die("Unrecoverable FP Unavailable Exception", regs, SIGABRT); 1636 } 1637 1638 DEFINE_INTERRUPT_HANDLER(altivec_unavailable_exception) 1639 { 1640 if (user_mode(regs)) { 1641 /* A user program has executed an altivec instruction, 1642 but this kernel doesn't support altivec. */ 1643 _exception(SIGILL, regs, ILL_ILLOPC, regs->nip); 1644 return; 1645 } 1646 1647 printk(KERN_EMERG "Unrecoverable VMX/Altivec Unavailable Exception " 1648 "%lx at %lx\n", regs->trap, regs->nip); 1649 die("Unrecoverable VMX/Altivec Unavailable Exception", regs, SIGABRT); 1650 } 1651 1652 DEFINE_INTERRUPT_HANDLER(vsx_unavailable_exception) 1653 { 1654 if (user_mode(regs)) { 1655 /* A user program has executed an vsx instruction, 1656 but this kernel doesn't support vsx. */ 1657 _exception(SIGILL, regs, ILL_ILLOPC, regs->nip); 1658 return; 1659 } 1660 1661 printk(KERN_EMERG "Unrecoverable VSX Unavailable Exception " 1662 "%lx at %lx\n", regs->trap, regs->nip); 1663 die("Unrecoverable VSX Unavailable Exception", regs, SIGABRT); 1664 } 1665 1666 #ifdef CONFIG_PPC64 1667 static void tm_unavailable(struct pt_regs *regs) 1668 { 1669 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM 1670 if (user_mode(regs)) { 1671 current->thread.load_tm++; 1672 regs_set_return_msr(regs, regs->msr | MSR_TM); 1673 tm_enable(); 1674 tm_restore_sprs(¤t->thread); 1675 return; 1676 } 1677 #endif 1678 pr_emerg("Unrecoverable TM Unavailable Exception " 1679 "%lx at %lx\n", regs->trap, regs->nip); 1680 die("Unrecoverable TM Unavailable Exception", regs, SIGABRT); 1681 } 1682 1683 DEFINE_INTERRUPT_HANDLER(facility_unavailable_exception) 1684 { 1685 static char *facility_strings[] = { 1686 [FSCR_FP_LG] = "FPU", 1687 [FSCR_VECVSX_LG] = "VMX/VSX", 1688 [FSCR_DSCR_LG] = "DSCR", 1689 [FSCR_PM_LG] = "PMU SPRs", 1690 [FSCR_BHRB_LG] = "BHRB", 1691 [FSCR_TM_LG] = "TM", 1692 [FSCR_EBB_LG] = "EBB", 1693 [FSCR_TAR_LG] = "TAR", 1694 [FSCR_MSGP_LG] = "MSGP", 1695 [FSCR_SCV_LG] = "SCV", 1696 [FSCR_PREFIX_LG] = "PREFIX", 1697 }; 1698 char *facility = "unknown"; 1699 u64 value; 1700 u32 instword, rd; 1701 u8 status; 1702 bool hv; 1703 1704 hv = (TRAP(regs) == INTERRUPT_H_FAC_UNAVAIL); 1705 if (hv) 1706 value = mfspr(SPRN_HFSCR); 1707 else 1708 value = mfspr(SPRN_FSCR); 1709 1710 status = value >> 56; 1711 if ((hv || status >= 2) && 1712 (status < ARRAY_SIZE(facility_strings)) && 1713 facility_strings[status]) 1714 facility = facility_strings[status]; 1715 1716 /* We should not have taken this interrupt in kernel */ 1717 if (!user_mode(regs)) { 1718 pr_emerg("Facility '%s' unavailable (%d) exception in kernel mode at %lx\n", 1719 facility, status, regs->nip); 1720 die("Unexpected facility unavailable exception", regs, SIGABRT); 1721 } 1722 1723 interrupt_cond_local_irq_enable(regs); 1724 1725 if (status == FSCR_DSCR_LG) { 1726 /* 1727 * User is accessing the DSCR register using the problem 1728 * state only SPR number (0x03) either through a mfspr or 1729 * a mtspr instruction. If it is a write attempt through 1730 * a mtspr, then we set the inherit bit. This also allows 1731 * the user to write or read the register directly in the 1732 * future by setting via the FSCR DSCR bit. But in case it 1733 * is a read DSCR attempt through a mfspr instruction, we 1734 * just emulate the instruction instead. This code path will 1735 * always emulate all the mfspr instructions till the user 1736 * has attempted at least one mtspr instruction. This way it 1737 * preserves the same behaviour when the user is accessing 1738 * the DSCR through privilege level only SPR number (0x11) 1739 * which is emulated through illegal instruction exception. 1740 * We always leave HFSCR DSCR set. 1741 */ 1742 if (get_user(instword, (u32 __user *)(regs->nip))) { 1743 pr_err("Failed to fetch the user instruction\n"); 1744 return; 1745 } 1746 1747 /* Write into DSCR (mtspr 0x03, RS) */ 1748 if ((instword & PPC_INST_MTSPR_DSCR_USER_MASK) 1749 == PPC_INST_MTSPR_DSCR_USER) { 1750 rd = (instword >> 21) & 0x1f; 1751 current->thread.dscr = regs->gpr[rd]; 1752 current->thread.dscr_inherit = 1; 1753 current->thread.fscr |= FSCR_DSCR; 1754 mtspr(SPRN_FSCR, current->thread.fscr); 1755 } 1756 1757 /* Read from DSCR (mfspr RT, 0x03) */ 1758 if ((instword & PPC_INST_MFSPR_DSCR_USER_MASK) 1759 == PPC_INST_MFSPR_DSCR_USER) { 1760 if (emulate_instruction(regs)) { 1761 pr_err("DSCR based mfspr emulation failed\n"); 1762 return; 1763 } 1764 regs_add_return_ip(regs, 4); 1765 emulate_single_step(regs); 1766 } 1767 return; 1768 } 1769 1770 if (status == FSCR_TM_LG) { 1771 /* 1772 * If we're here then the hardware is TM aware because it 1773 * generated an exception with FSRM_TM set. 1774 * 1775 * If cpu_has_feature(CPU_FTR_TM) is false, then either firmware 1776 * told us not to do TM, or the kernel is not built with TM 1777 * support. 1778 * 1779 * If both of those things are true, then userspace can spam the 1780 * console by triggering the printk() below just by continually 1781 * doing tbegin (or any TM instruction). So in that case just 1782 * send the process a SIGILL immediately. 1783 */ 1784 if (!cpu_has_feature(CPU_FTR_TM)) 1785 goto out; 1786 1787 tm_unavailable(regs); 1788 return; 1789 } 1790 1791 pr_err_ratelimited("%sFacility '%s' unavailable (%d), exception at 0x%lx, MSR=%lx\n", 1792 hv ? "Hypervisor " : "", facility, status, regs->nip, regs->msr); 1793 1794 out: 1795 _exception(SIGILL, regs, ILL_ILLOPC, regs->nip); 1796 } 1797 #endif 1798 1799 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM 1800 1801 DEFINE_INTERRUPT_HANDLER(fp_unavailable_tm) 1802 { 1803 /* Note: This does not handle any kind of FP laziness. */ 1804 1805 TM_DEBUG("FP Unavailable trap whilst transactional at 0x%lx, MSR=%lx\n", 1806 regs->nip, regs->msr); 1807 1808 /* We can only have got here if the task started using FP after 1809 * beginning the transaction. So, the transactional regs are just a 1810 * copy of the checkpointed ones. But, we still need to recheckpoint 1811 * as we're enabling FP for the process; it will return, abort the 1812 * transaction, and probably retry but now with FP enabled. So the 1813 * checkpointed FP registers need to be loaded. 1814 */ 1815 tm_reclaim_current(TM_CAUSE_FAC_UNAV); 1816 1817 /* 1818 * Reclaim initially saved out bogus (lazy) FPRs to ckfp_state, and 1819 * then it was overwrite by the thr->fp_state by tm_reclaim_thread(). 1820 * 1821 * At this point, ck{fp,vr}_state contains the exact values we want to 1822 * recheckpoint. 1823 */ 1824 1825 /* Enable FP for the task: */ 1826 current->thread.load_fp = 1; 1827 1828 /* 1829 * Recheckpoint all the checkpointed ckpt, ck{fp, vr}_state registers. 1830 */ 1831 tm_recheckpoint(¤t->thread); 1832 } 1833 1834 DEFINE_INTERRUPT_HANDLER(altivec_unavailable_tm) 1835 { 1836 /* See the comments in fp_unavailable_tm(). This function operates 1837 * the same way. 1838 */ 1839 1840 TM_DEBUG("Vector Unavailable trap whilst transactional at 0x%lx," 1841 "MSR=%lx\n", 1842 regs->nip, regs->msr); 1843 tm_reclaim_current(TM_CAUSE_FAC_UNAV); 1844 current->thread.load_vec = 1; 1845 tm_recheckpoint(¤t->thread); 1846 current->thread.used_vr = 1; 1847 } 1848 1849 DEFINE_INTERRUPT_HANDLER(vsx_unavailable_tm) 1850 { 1851 /* See the comments in fp_unavailable_tm(). This works similarly, 1852 * though we're loading both FP and VEC registers in here. 1853 * 1854 * If FP isn't in use, load FP regs. If VEC isn't in use, load VEC 1855 * regs. Either way, set MSR_VSX. 1856 */ 1857 1858 TM_DEBUG("VSX Unavailable trap whilst transactional at 0x%lx," 1859 "MSR=%lx\n", 1860 regs->nip, regs->msr); 1861 1862 current->thread.used_vsr = 1; 1863 1864 /* This reclaims FP and/or VR regs if they're already enabled */ 1865 tm_reclaim_current(TM_CAUSE_FAC_UNAV); 1866 1867 current->thread.load_vec = 1; 1868 current->thread.load_fp = 1; 1869 1870 tm_recheckpoint(¤t->thread); 1871 } 1872 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */ 1873 1874 #ifdef CONFIG_PPC64 1875 DECLARE_INTERRUPT_HANDLER_NMI(performance_monitor_exception_nmi); 1876 DEFINE_INTERRUPT_HANDLER_NMI(performance_monitor_exception_nmi) 1877 { 1878 __this_cpu_inc(irq_stat.pmu_irqs); 1879 1880 perf_irq(regs); 1881 1882 return 0; 1883 } 1884 #endif 1885 1886 DECLARE_INTERRUPT_HANDLER_ASYNC(performance_monitor_exception_async); 1887 DEFINE_INTERRUPT_HANDLER_ASYNC(performance_monitor_exception_async) 1888 { 1889 __this_cpu_inc(irq_stat.pmu_irqs); 1890 1891 perf_irq(regs); 1892 } 1893 1894 DEFINE_INTERRUPT_HANDLER_RAW(performance_monitor_exception) 1895 { 1896 /* 1897 * On 64-bit, if perf interrupts hit in a local_irq_disable 1898 * (soft-masked) region, we consider them as NMIs. This is required to 1899 * prevent hash faults on user addresses when reading callchains (and 1900 * looks better from an irq tracing perspective). 1901 */ 1902 if (IS_ENABLED(CONFIG_PPC64) && unlikely(arch_irq_disabled_regs(regs))) 1903 performance_monitor_exception_nmi(regs); 1904 else 1905 performance_monitor_exception_async(regs); 1906 1907 return 0; 1908 } 1909 1910 #ifdef CONFIG_PPC_ADV_DEBUG_REGS 1911 static void handle_debug(struct pt_regs *regs, unsigned long debug_status) 1912 { 1913 int changed = 0; 1914 /* 1915 * Determine the cause of the debug event, clear the 1916 * event flags and send a trap to the handler. Torez 1917 */ 1918 if (debug_status & (DBSR_DAC1R | DBSR_DAC1W)) { 1919 dbcr_dac(current) &= ~(DBCR_DAC1R | DBCR_DAC1W); 1920 #ifdef CONFIG_PPC_ADV_DEBUG_DAC_RANGE 1921 current->thread.debug.dbcr2 &= ~DBCR2_DAC12MODE; 1922 #endif 1923 do_send_trap(regs, mfspr(SPRN_DAC1), debug_status, 1924 5); 1925 changed |= 0x01; 1926 } else if (debug_status & (DBSR_DAC2R | DBSR_DAC2W)) { 1927 dbcr_dac(current) &= ~(DBCR_DAC2R | DBCR_DAC2W); 1928 do_send_trap(regs, mfspr(SPRN_DAC2), debug_status, 1929 6); 1930 changed |= 0x01; 1931 } else if (debug_status & DBSR_IAC1) { 1932 current->thread.debug.dbcr0 &= ~DBCR0_IAC1; 1933 dbcr_iac_range(current) &= ~DBCR_IAC12MODE; 1934 do_send_trap(regs, mfspr(SPRN_IAC1), debug_status, 1935 1); 1936 changed |= 0x01; 1937 } else if (debug_status & DBSR_IAC2) { 1938 current->thread.debug.dbcr0 &= ~DBCR0_IAC2; 1939 do_send_trap(regs, mfspr(SPRN_IAC2), debug_status, 1940 2); 1941 changed |= 0x01; 1942 } else if (debug_status & DBSR_IAC3) { 1943 current->thread.debug.dbcr0 &= ~DBCR0_IAC3; 1944 dbcr_iac_range(current) &= ~DBCR_IAC34MODE; 1945 do_send_trap(regs, mfspr(SPRN_IAC3), debug_status, 1946 3); 1947 changed |= 0x01; 1948 } else if (debug_status & DBSR_IAC4) { 1949 current->thread.debug.dbcr0 &= ~DBCR0_IAC4; 1950 do_send_trap(regs, mfspr(SPRN_IAC4), debug_status, 1951 4); 1952 changed |= 0x01; 1953 } 1954 /* 1955 * At the point this routine was called, the MSR(DE) was turned off. 1956 * Check all other debug flags and see if that bit needs to be turned 1957 * back on or not. 1958 */ 1959 if (DBCR_ACTIVE_EVENTS(current->thread.debug.dbcr0, 1960 current->thread.debug.dbcr1)) 1961 regs_set_return_msr(regs, regs->msr | MSR_DE); 1962 else 1963 /* Make sure the IDM flag is off */ 1964 current->thread.debug.dbcr0 &= ~DBCR0_IDM; 1965 1966 if (changed & 0x01) 1967 mtspr(SPRN_DBCR0, current->thread.debug.dbcr0); 1968 } 1969 1970 DEFINE_INTERRUPT_HANDLER(DebugException) 1971 { 1972 unsigned long debug_status = regs->dsisr; 1973 1974 current->thread.debug.dbsr = debug_status; 1975 1976 /* Hack alert: On BookE, Branch Taken stops on the branch itself, while 1977 * on server, it stops on the target of the branch. In order to simulate 1978 * the server behaviour, we thus restart right away with a single step 1979 * instead of stopping here when hitting a BT 1980 */ 1981 if (debug_status & DBSR_BT) { 1982 regs_set_return_msr(regs, regs->msr & ~MSR_DE); 1983 1984 /* Disable BT */ 1985 mtspr(SPRN_DBCR0, mfspr(SPRN_DBCR0) & ~DBCR0_BT); 1986 /* Clear the BT event */ 1987 mtspr(SPRN_DBSR, DBSR_BT); 1988 1989 /* Do the single step trick only when coming from userspace */ 1990 if (user_mode(regs)) { 1991 current->thread.debug.dbcr0 &= ~DBCR0_BT; 1992 current->thread.debug.dbcr0 |= DBCR0_IDM | DBCR0_IC; 1993 regs_set_return_msr(regs, regs->msr | MSR_DE); 1994 return; 1995 } 1996 1997 if (kprobe_post_handler(regs)) 1998 return; 1999 2000 if (notify_die(DIE_SSTEP, "block_step", regs, 5, 2001 5, SIGTRAP) == NOTIFY_STOP) { 2002 return; 2003 } 2004 if (debugger_sstep(regs)) 2005 return; 2006 } else if (debug_status & DBSR_IC) { /* Instruction complete */ 2007 regs_set_return_msr(regs, regs->msr & ~MSR_DE); 2008 2009 /* Disable instruction completion */ 2010 mtspr(SPRN_DBCR0, mfspr(SPRN_DBCR0) & ~DBCR0_IC); 2011 /* Clear the instruction completion event */ 2012 mtspr(SPRN_DBSR, DBSR_IC); 2013 2014 if (kprobe_post_handler(regs)) 2015 return; 2016 2017 if (notify_die(DIE_SSTEP, "single_step", regs, 5, 2018 5, SIGTRAP) == NOTIFY_STOP) { 2019 return; 2020 } 2021 2022 if (debugger_sstep(regs)) 2023 return; 2024 2025 if (user_mode(regs)) { 2026 current->thread.debug.dbcr0 &= ~DBCR0_IC; 2027 if (DBCR_ACTIVE_EVENTS(current->thread.debug.dbcr0, 2028 current->thread.debug.dbcr1)) 2029 regs_set_return_msr(regs, regs->msr | MSR_DE); 2030 else 2031 /* Make sure the IDM bit is off */ 2032 current->thread.debug.dbcr0 &= ~DBCR0_IDM; 2033 } 2034 2035 _exception(SIGTRAP, regs, TRAP_TRACE, regs->nip); 2036 } else 2037 handle_debug(regs, debug_status); 2038 } 2039 #endif /* CONFIG_PPC_ADV_DEBUG_REGS */ 2040 2041 #ifdef CONFIG_ALTIVEC 2042 DEFINE_INTERRUPT_HANDLER(altivec_assist_exception) 2043 { 2044 int err; 2045 2046 if (!user_mode(regs)) { 2047 printk(KERN_EMERG "VMX/Altivec assist exception in kernel mode" 2048 " at %lx\n", regs->nip); 2049 die("Kernel VMX/Altivec assist exception", regs, SIGILL); 2050 } 2051 2052 flush_altivec_to_thread(current); 2053 2054 PPC_WARN_EMULATED(altivec, regs); 2055 err = emulate_altivec(regs); 2056 if (err == 0) { 2057 regs_add_return_ip(regs, 4); /* skip emulated instruction */ 2058 emulate_single_step(regs); 2059 return; 2060 } 2061 2062 if (err == -EFAULT) { 2063 /* got an error reading the instruction */ 2064 _exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip); 2065 } else { 2066 /* didn't recognize the instruction */ 2067 /* XXX quick hack for now: set the non-Java bit in the VSCR */ 2068 printk_ratelimited(KERN_ERR "Unrecognized altivec instruction " 2069 "in %s at %lx\n", current->comm, regs->nip); 2070 current->thread.vr_state.vscr.u[3] |= 0x10000; 2071 } 2072 } 2073 #endif /* CONFIG_ALTIVEC */ 2074 2075 #ifdef CONFIG_FSL_BOOKE 2076 DEFINE_INTERRUPT_HANDLER(CacheLockingException) 2077 { 2078 unsigned long error_code = regs->dsisr; 2079 2080 /* We treat cache locking instructions from the user 2081 * as priv ops, in the future we could try to do 2082 * something smarter 2083 */ 2084 if (error_code & (ESR_DLK|ESR_ILK)) 2085 _exception(SIGILL, regs, ILL_PRVOPC, regs->nip); 2086 return; 2087 } 2088 #endif /* CONFIG_FSL_BOOKE */ 2089 2090 #ifdef CONFIG_SPE 2091 DEFINE_INTERRUPT_HANDLER(SPEFloatingPointException) 2092 { 2093 extern int do_spe_mathemu(struct pt_regs *regs); 2094 unsigned long spefscr; 2095 int fpexc_mode; 2096 int code = FPE_FLTUNK; 2097 int err; 2098 2099 interrupt_cond_local_irq_enable(regs); 2100 2101 flush_spe_to_thread(current); 2102 2103 spefscr = current->thread.spefscr; 2104 fpexc_mode = current->thread.fpexc_mode; 2105 2106 if ((spefscr & SPEFSCR_FOVF) && (fpexc_mode & PR_FP_EXC_OVF)) { 2107 code = FPE_FLTOVF; 2108 } 2109 else if ((spefscr & SPEFSCR_FUNF) && (fpexc_mode & PR_FP_EXC_UND)) { 2110 code = FPE_FLTUND; 2111 } 2112 else if ((spefscr & SPEFSCR_FDBZ) && (fpexc_mode & PR_FP_EXC_DIV)) 2113 code = FPE_FLTDIV; 2114 else if ((spefscr & SPEFSCR_FINV) && (fpexc_mode & PR_FP_EXC_INV)) { 2115 code = FPE_FLTINV; 2116 } 2117 else if ((spefscr & (SPEFSCR_FG | SPEFSCR_FX)) && (fpexc_mode & PR_FP_EXC_RES)) 2118 code = FPE_FLTRES; 2119 2120 err = do_spe_mathemu(regs); 2121 if (err == 0) { 2122 regs_add_return_ip(regs, 4); /* skip emulated instruction */ 2123 emulate_single_step(regs); 2124 return; 2125 } 2126 2127 if (err == -EFAULT) { 2128 /* got an error reading the instruction */ 2129 _exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip); 2130 } else if (err == -EINVAL) { 2131 /* didn't recognize the instruction */ 2132 printk(KERN_ERR "unrecognized spe instruction " 2133 "in %s at %lx\n", current->comm, regs->nip); 2134 } else { 2135 _exception(SIGFPE, regs, code, regs->nip); 2136 } 2137 2138 return; 2139 } 2140 2141 DEFINE_INTERRUPT_HANDLER(SPEFloatingPointRoundException) 2142 { 2143 extern int speround_handler(struct pt_regs *regs); 2144 int err; 2145 2146 interrupt_cond_local_irq_enable(regs); 2147 2148 preempt_disable(); 2149 if (regs->msr & MSR_SPE) 2150 giveup_spe(current); 2151 preempt_enable(); 2152 2153 regs_add_return_ip(regs, -4); 2154 err = speround_handler(regs); 2155 if (err == 0) { 2156 regs_add_return_ip(regs, 4); /* skip emulated instruction */ 2157 emulate_single_step(regs); 2158 return; 2159 } 2160 2161 if (err == -EFAULT) { 2162 /* got an error reading the instruction */ 2163 _exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip); 2164 } else if (err == -EINVAL) { 2165 /* didn't recognize the instruction */ 2166 printk(KERN_ERR "unrecognized spe instruction " 2167 "in %s at %lx\n", current->comm, regs->nip); 2168 } else { 2169 _exception(SIGFPE, regs, FPE_FLTUNK, regs->nip); 2170 return; 2171 } 2172 } 2173 #endif 2174 2175 /* 2176 * We enter here if we get an unrecoverable exception, that is, one 2177 * that happened at a point where the RI (recoverable interrupt) bit 2178 * in the MSR is 0. This indicates that SRR0/1 are live, and that 2179 * we therefore lost state by taking this exception. 2180 */ 2181 void __noreturn unrecoverable_exception(struct pt_regs *regs) 2182 { 2183 pr_emerg("Unrecoverable exception %lx at %lx (msr=%lx)\n", 2184 regs->trap, regs->nip, regs->msr); 2185 die("Unrecoverable exception", regs, SIGABRT); 2186 /* die() should not return */ 2187 for (;;) 2188 ; 2189 } 2190 2191 #if defined(CONFIG_BOOKE_WDT) || defined(CONFIG_40x) 2192 /* 2193 * Default handler for a Watchdog exception, 2194 * spins until a reboot occurs 2195 */ 2196 void __attribute__ ((weak)) WatchdogHandler(struct pt_regs *regs) 2197 { 2198 /* Generic WatchdogHandler, implement your own */ 2199 mtspr(SPRN_TCR, mfspr(SPRN_TCR)&(~TCR_WIE)); 2200 return; 2201 } 2202 2203 DEFINE_INTERRUPT_HANDLER_NMI(WatchdogException) 2204 { 2205 printk (KERN_EMERG "PowerPC Book-E Watchdog Exception\n"); 2206 WatchdogHandler(regs); 2207 return 0; 2208 } 2209 #endif 2210 2211 /* 2212 * We enter here if we discover during exception entry that we are 2213 * running in supervisor mode with a userspace value in the stack pointer. 2214 */ 2215 DEFINE_INTERRUPT_HANDLER(kernel_bad_stack) 2216 { 2217 printk(KERN_EMERG "Bad kernel stack pointer %lx at %lx\n", 2218 regs->gpr[1], regs->nip); 2219 die("Bad kernel stack pointer", regs, SIGABRT); 2220 } 2221 2222 #ifdef CONFIG_PPC_EMULATED_STATS 2223 2224 #define WARN_EMULATED_SETUP(type) .type = { .name = #type } 2225 2226 struct ppc_emulated ppc_emulated = { 2227 #ifdef CONFIG_ALTIVEC 2228 WARN_EMULATED_SETUP(altivec), 2229 #endif 2230 WARN_EMULATED_SETUP(dcba), 2231 WARN_EMULATED_SETUP(dcbz), 2232 WARN_EMULATED_SETUP(fp_pair), 2233 WARN_EMULATED_SETUP(isel), 2234 WARN_EMULATED_SETUP(mcrxr), 2235 WARN_EMULATED_SETUP(mfpvr), 2236 WARN_EMULATED_SETUP(multiple), 2237 WARN_EMULATED_SETUP(popcntb), 2238 WARN_EMULATED_SETUP(spe), 2239 WARN_EMULATED_SETUP(string), 2240 WARN_EMULATED_SETUP(sync), 2241 WARN_EMULATED_SETUP(unaligned), 2242 #ifdef CONFIG_MATH_EMULATION 2243 WARN_EMULATED_SETUP(math), 2244 #endif 2245 #ifdef CONFIG_VSX 2246 WARN_EMULATED_SETUP(vsx), 2247 #endif 2248 #ifdef CONFIG_PPC64 2249 WARN_EMULATED_SETUP(mfdscr), 2250 WARN_EMULATED_SETUP(mtdscr), 2251 WARN_EMULATED_SETUP(lq_stq), 2252 WARN_EMULATED_SETUP(lxvw4x), 2253 WARN_EMULATED_SETUP(lxvh8x), 2254 WARN_EMULATED_SETUP(lxvd2x), 2255 WARN_EMULATED_SETUP(lxvb16x), 2256 #endif 2257 }; 2258 2259 u32 ppc_warn_emulated; 2260 2261 void ppc_warn_emulated_print(const char *type) 2262 { 2263 pr_warn_ratelimited("%s used emulated %s instruction\n", current->comm, 2264 type); 2265 } 2266 2267 static int __init ppc_warn_emulated_init(void) 2268 { 2269 struct dentry *dir; 2270 unsigned int i; 2271 struct ppc_emulated_entry *entries = (void *)&ppc_emulated; 2272 2273 dir = debugfs_create_dir("emulated_instructions", 2274 arch_debugfs_dir); 2275 2276 debugfs_create_u32("do_warn", 0644, dir, &ppc_warn_emulated); 2277 2278 for (i = 0; i < sizeof(ppc_emulated)/sizeof(*entries); i++) 2279 debugfs_create_u32(entries[i].name, 0644, dir, 2280 (u32 *)&entries[i].val.counter); 2281 2282 return 0; 2283 } 2284 2285 device_initcall(ppc_warn_emulated_init); 2286 2287 #endif /* CONFIG_PPC_EMULATED_STATS */ 2288