arch/arm64/kernel/traps.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Based on arch/arm/kernel/traps.c
   4  *
   5  * Copyright (C) 1995-2009 Russell King
   6  * Copyright (C) 2012 ARM Ltd.
   7  */
   8
   9 #include <linux/bug.h>
  10 #include <linux/context_tracking.h>
  11 #include <linux/signal.h>
  12 #include <linux/personality.h>
  13 #include <linux/kallsyms.h>
  14 #include <linux/kprobes.h>
  15 #include <linux/spinlock.h>
  16 #include <linux/uaccess.h>
  17 #include <linux/hardirq.h>
  18 #include <linux/kdebug.h>
  19 #include <linux/module.h>
  20 #include <linux/kexec.h>
  21 #include <linux/delay.h>
  22 #include <linux/init.h>
  23 #include <linux/sched/signal.h>
  24 #include <linux/sched/debug.h>
  25 #include <linux/sched/task_stack.h>
  26 #include <linux/sizes.h>
  27 #include <linux/syscalls.h>
  28 #include <linux/mm_types.h>
  29 #include <linux/kasan.h>
  30
  31 #include <asm/atomic.h>
  32 #include <asm/bug.h>
  33 #include <asm/cpufeature.h>
  34 #include <asm/daifflags.h>
  35 #include <asm/debug-monitors.h>
  36 #include <asm/esr.h>
  37 #include <asm/extable.h>
  38 #include <asm/insn.h>
  39 #include <asm/kprobes.h>
  40 #include <asm/traps.h>
  41 #include <asm/smp.h>
  42 #include <asm/stack_pointer.h>
  43 #include <asm/stacktrace.h>
  44 #include <asm/exception.h>
  45 #include <asm/system_misc.h>
  46 #include <asm/sysreg.h>
  47
  48 static const char *handler[]= {
  49         "Synchronous Abort",
  50         "IRQ",
  51         "FIQ",
  52         "Error"
  53 };
  54
  55 int show_unhandled_signals = 0;
  56
  57 static void dump_kernel_instr(const char *lvl, struct pt_regs *regs)
  58 {
  59         unsigned long addr = instruction_pointer(regs);
  60         char str[sizeof("00000000 ") * 5 + 2 + 1], *p = str;
  61         int i;
  62
  63         if (user_mode(regs))
  64                 return;
  65
  66         for (i = -4; i < 1; i++) {
  67                 unsigned int val, bad;
  68
  69                 bad = aarch64_insn_read(&((u32 *)addr)[i], &val);
  70
  71                 if (!bad)
  72                         p += sprintf(p, i == 0 ? "(%08x) " : "%08x ", val);
  73                 else {
  74                         p += sprintf(p, "bad PC value");
  75                         break;
  76                 }
  77         }
  78
  79         printk("%sCode: %s\n", lvl, str);
  80 }
  81
  82 #ifdef CONFIG_PREEMPT
  83 #define S_PREEMPT " PREEMPT"
  84 #elif defined(CONFIG_PREEMPT_RT)
  85 #define S_PREEMPT " PREEMPT_RT"
  86 #else
  87 #define S_PREEMPT ""
  88 #endif
  89
  90 #define S_SMP " SMP"
  91
  92 static int __die(const char *str, int err, struct pt_regs *regs)
  93 {
  94         static int die_counter;
  95         int ret;
  96
  97         pr_emerg("Internal error: %s: %x [#%d]" S_PREEMPT S_SMP "\n",
  98                  str, err, ++die_counter);
  99
 100         /* trap and error numbers are mostly meaningless on ARM */
 101         ret = notify_die(DIE_OOPS, str, regs, err, 0, SIGSEGV);
 102         if (ret == NOTIFY_STOP)
 103                 return ret;
 104
 105         print_modules();
 106         show_regs(regs);
 107
 108         dump_kernel_instr(KERN_EMERG, regs);
 109
 110         return ret;
 111 }
 112
 113 static DEFINE_RAW_SPINLOCK(die_lock);
 114
 115 /*
 116  * This function is protected against re-entrancy.
 117  */
 118 void die(const char *str, struct pt_regs *regs, int err)
 119 {
 120         int ret;
 121         unsigned long flags;
 122
 123         raw_spin_lock_irqsave(&die_lock, flags);
 124
 125         oops_enter();
 126
 127         console_verbose();
 128         bust_spinlocks(1);
 129         ret = __die(str, err, regs);
 130
 131         if (regs && kexec_should_crash(current))
 132                 crash_kexec(regs);
 133
 134         bust_spinlocks(0);
 135         add_taint(TAINT_DIE, LOCKDEP_NOW_UNRELIABLE);
 136         oops_exit();
 137
 138         if (in_interrupt())
 139                 panic("%s: Fatal exception in interrupt", str);
 140         if (panic_on_oops)
 141                 panic("%s: Fatal exception", str);
 142
 143         raw_spin_unlock_irqrestore(&die_lock, flags);
 144
 145         if (ret != NOTIFY_STOP)
 146                 do_exit(SIGSEGV);
 147 }
 148
 149 static void arm64_show_signal(int signo, const char *str)
 150 {
 151         static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,
 152                                       DEFAULT_RATELIMIT_BURST);
 153         struct task_struct *tsk = current;
 154         unsigned int esr = tsk->thread.fault_code;
 155         struct pt_regs *regs = task_pt_regs(tsk);
 156
 157         /* Leave if the signal won't be shown */
 158         if (!show_unhandled_signals ||
 159             !unhandled_signal(tsk, signo) ||
 160             !__ratelimit(&rs))
 161                 return;
 162
 163         pr_info("%s[%d]: unhandled exception: ", tsk->comm, task_pid_nr(tsk));
 164         if (esr)
 165                 pr_cont("%s, ESR 0x%08x, ", esr_get_class_string(esr), esr);
 166
 167         pr_cont("%s", str);
 168         print_vma_addr(KERN_CONT " in ", regs->pc);
 169         pr_cont("\n");
 170         __show_regs(regs);
 171 }
 172
 173 void arm64_force_sig_fault(int signo, int code, void __user *addr,
 174                            const char *str)
 175 {
 176         arm64_show_signal(signo, str);
 177         if (signo == SIGKILL)
 178                 force_sig(SIGKILL);
 179         else
 180                 force_sig_fault(signo, code, addr);
 181 }
 182
 183 void arm64_force_sig_mceerr(int code, void __user *addr, short lsb,
 184                             const char *str)
 185 {
 186         arm64_show_signal(SIGBUS, str);
 187         force_sig_mceerr(code, addr, lsb);
 188 }
 189
 190 void arm64_force_sig_ptrace_errno_trap(int errno, void __user *addr,
 191                                        const char *str)
 192 {
 193         arm64_show_signal(SIGTRAP, str);
 194         force_sig_ptrace_errno_trap(errno, addr);
 195 }
 196
 197 void arm64_notify_die(const char *str, struct pt_regs *regs,
 198                       int signo, int sicode, void __user *addr,
 199                       int err)
 200 {
 201         if (user_mode(regs)) {
 202                 WARN_ON(regs != current_pt_regs());
 203                 current->thread.fault_address = 0;
 204                 current->thread.fault_code = err;
 205
 206                 arm64_force_sig_fault(signo, sicode, addr, str);
 207         } else {
 208                 die(str, regs, err);
 209         }
 210 }
 211
 212 #ifdef CONFIG_COMPAT
 213 #define PSTATE_IT_1_0_SHIFT     25
 214 #define PSTATE_IT_1_0_MASK      (0x3 << PSTATE_IT_1_0_SHIFT)
 215 #define PSTATE_IT_7_2_SHIFT     10
 216 #define PSTATE_IT_7_2_MASK      (0x3f << PSTATE_IT_7_2_SHIFT)
 217
 218 static u32 compat_get_it_state(struct pt_regs *regs)
 219 {
 220         u32 it, pstate = regs->pstate;
 221
 222         it  = (pstate & PSTATE_IT_1_0_MASK) >> PSTATE_IT_1_0_SHIFT;
 223         it |= ((pstate & PSTATE_IT_7_2_MASK) >> PSTATE_IT_7_2_SHIFT) << 2;
 224
 225         return it;
 226 }
 227
 228 static void compat_set_it_state(struct pt_regs *regs, u32 it)
 229 {
 230         u32 pstate_it;
 231
 232         pstate_it  = (it << PSTATE_IT_1_0_SHIFT) & PSTATE_IT_1_0_MASK;
 233         pstate_it |= ((it >> 2) << PSTATE_IT_7_2_SHIFT) & PSTATE_IT_7_2_MASK;
 234
 235         regs->pstate &= ~PSR_AA32_IT_MASK;
 236         regs->pstate |= pstate_it;
 237 }
 238
 239 static void advance_itstate(struct pt_regs *regs)
 240 {
 241         u32 it;
 242
 243         /* ARM mode */
 244         if (!(regs->pstate & PSR_AA32_T_BIT) ||
 245             !(regs->pstate & PSR_AA32_IT_MASK))
 246                 return;
 247
 248         it  = compat_get_it_state(regs);
 249
 250         /*
 251          * If this is the last instruction of the block, wipe the IT
 252          * state. Otherwise advance it.
 253          */
 254         if (!(it & 7))
 255                 it = 0;
 256         else
 257                 it = (it & 0xe0) | ((it << 1) & 0x1f);
 258
 259         compat_set_it_state(regs, it);
 260 }
 261 #else
 262 static void advance_itstate(struct pt_regs *regs)
 263 {
 264 }
 265 #endif
 266
 267 void arm64_skip_faulting_instruction(struct pt_regs *regs, unsigned long size)
 268 {
 269         regs->pc += size;
 270
 271         /*
 272          * If we were single stepping, we want to get the step exception after
 273          * we return from the trap.
 274          */
 275         if (user_mode(regs))
 276                 user_fastforward_single_step(current);
 277
 278         if (compat_user_mode(regs))
 279                 advance_itstate(regs);
 280         else
 281                 regs->pstate &= ~PSR_BTYPE_MASK;
 282 }
 283
 284 static LIST_HEAD(undef_hook);
 285 static DEFINE_RAW_SPINLOCK(undef_lock);
 286
 287 void register_undef_hook(struct undef_hook *hook)
 288 {
 289         unsigned long flags;
 290
 291         raw_spin_lock_irqsave(&undef_lock, flags);
 292         list_add(&hook->node, &undef_hook);
 293         raw_spin_unlock_irqrestore(&undef_lock, flags);
 294 }
 295
 296 void unregister_undef_hook(struct undef_hook *hook)
 297 {
 298         unsigned long flags;
 299
 300         raw_spin_lock_irqsave(&undef_lock, flags);
 301         list_del(&hook->node);
 302         raw_spin_unlock_irqrestore(&undef_lock, flags);
 303 }
 304
 305 static int call_undef_hook(struct pt_regs *regs)
 306 {
 307         struct undef_hook *hook;
 308         unsigned long flags;
 309         u32 instr;
 310         int (*fn)(struct pt_regs *regs, u32 instr) = NULL;
 311         void __user *pc = (void __user *)instruction_pointer(regs);
 312
 313         if (!user_mode(regs)) {
 314                 __le32 instr_le;
 315                 if (get_kernel_nofault(instr_le, (__force __le32 *)pc))
 316                         goto exit;
 317                 instr = le32_to_cpu(instr_le);
 318         } else if (compat_thumb_mode(regs)) {
 319                 /* 16-bit Thumb instruction */
 320                 __le16 instr_le;
 321                 if (get_user(instr_le, (__le16 __user *)pc))
 322                         goto exit;
 323                 instr = le16_to_cpu(instr_le);
 324                 if (aarch32_insn_is_wide(instr)) {
 325                         u32 instr2;
 326
 327                         if (get_user(instr_le, (__le16 __user *)(pc + 2)))
 328                                 goto exit;
 329                         instr2 = le16_to_cpu(instr_le);
 330                         instr = (instr << 16) | instr2;
 331                 }
 332         } else {
 333                 /* 32-bit ARM instruction */
 334                 __le32 instr_le;
 335                 if (get_user(instr_le, (__le32 __user *)pc))
 336                         goto exit;
 337                 instr = le32_to_cpu(instr_le);
 338         }
 339
 340         raw_spin_lock_irqsave(&undef_lock, flags);
 341         list_for_each_entry(hook, &undef_hook, node)
 342                 if ((instr & hook->instr_mask) == hook->instr_val &&
 343                         (regs->pstate & hook->pstate_mask) == hook->pstate_val)
 344                         fn = hook->fn;
 345
 346         raw_spin_unlock_irqrestore(&undef_lock, flags);
 347 exit:
 348         return fn ? fn(regs, instr) : 1;
 349 }
 350
 351 void force_signal_inject(int signal, int code, unsigned long address, unsigned int err)
 352 {
 353         const char *desc;
 354         struct pt_regs *regs = current_pt_regs();
 355
 356         if (WARN_ON(!user_mode(regs)))
 357                 return;
 358
 359         switch (signal) {
 360         case SIGILL:
 361                 desc = "undefined instruction";
 362                 break;
 363         case SIGSEGV:
 364                 desc = "illegal memory access";
 365                 break;
 366         default:
 367                 desc = "unknown or unrecoverable error";
 368                 break;
 369         }
 370
 371         /* Force signals we don't understand to SIGKILL */
 372         if (WARN_ON(signal != SIGKILL &&
 373                     siginfo_layout(signal, code) != SIL_FAULT)) {
 374                 signal = SIGKILL;
 375         }
 376
 377         arm64_notify_die(desc, regs, signal, code, (void __user *)address, err);
 378 }
 379
 380 /*
 381  * Set up process info to signal segmentation fault - called on access error.
 382  */
 383 void arm64_notify_segfault(unsigned long addr)
 384 {
 385         int code;
 386
 387         mmap_read_lock(current->mm);
 388         if (find_vma(current->mm, addr) == NULL)
 389                 code = SEGV_MAPERR;
 390         else
 391                 code = SEGV_ACCERR;
 392         mmap_read_unlock(current->mm);
 393
 394         force_signal_inject(SIGSEGV, code, addr, 0);
 395 }
 396
 397 void do_undefinstr(struct pt_regs *regs)
 398 {
 399         /* check for AArch32 breakpoint instructions */
 400         if (!aarch32_break_handler(regs))
 401                 return;
 402
 403         if (call_undef_hook(regs) == 0)
 404                 return;
 405
 406         BUG_ON(!user_mode(regs));
 407         force_signal_inject(SIGILL, ILL_ILLOPC, regs->pc, 0);
 408 }
 409 NOKPROBE_SYMBOL(do_undefinstr);
 410
 411 void do_bti(struct pt_regs *regs)
 412 {
 413         BUG_ON(!user_mode(regs));
 414         force_signal_inject(SIGILL, ILL_ILLOPC, regs->pc, 0);
 415 }
 416 NOKPROBE_SYMBOL(do_bti);
 417
 418 void do_ptrauth_fault(struct pt_regs *regs, unsigned int esr)
 419 {
 420         /*
 421          * Unexpected FPAC exception or pointer authentication failure in
 422          * the kernel: kill the task before it does any more harm.
 423          */
 424         BUG_ON(!user_mode(regs));
 425         force_signal_inject(SIGILL, ILL_ILLOPN, regs->pc, esr);
 426 }
 427 NOKPROBE_SYMBOL(do_ptrauth_fault);
 428
 429 #define __user_cache_maint(insn, address, res)                  \
 430         if (address >= user_addr_max()) {                       \
 431                 res = -EFAULT;                                  \
 432         } else {                                                \
 433                 uaccess_ttbr0_enable();                         \
 434                 asm volatile (                                  \
 435                         "1:     " insn ", %1\n"                 \
 436                         "       mov     %w0, #0\n"              \
 437                         "2:\n"                                  \
 438                         "       .pushsection .fixup,\"ax\"\n"   \
 439                         "       .align  2\n"                    \
 440                         "3:     mov     %w0, %w2\n"             \
 441                         "       b       2b\n"                   \
 442                         "       .popsection\n"                  \
 443                         _ASM_EXTABLE(1b, 3b)                    \
 444                         : "=r" (res)                            \
 445                         : "r" (address), "i" (-EFAULT));        \
 446                 uaccess_ttbr0_disable();                        \
 447         }
 448
 449 static void user_cache_maint_handler(unsigned int esr, struct pt_regs *regs)
 450 {
 451         unsigned long address;
 452         int rt = ESR_ELx_SYS64_ISS_RT(esr);
 453         int crm = (esr & ESR_ELx_SYS64_ISS_CRM_MASK) >> ESR_ELx_SYS64_ISS_CRM_SHIFT;
 454         int ret = 0;
 455
 456         address = untagged_addr(pt_regs_read_reg(regs, rt));
 457
 458         switch (crm) {
 459         case ESR_ELx_SYS64_ISS_CRM_DC_CVAU:     /* DC CVAU, gets promoted */
 460                 __user_cache_maint("dc civac", address, ret);
 461                 break;
 462         case ESR_ELx_SYS64_ISS_CRM_DC_CVAC:     /* DC CVAC, gets promoted */
 463                 __user_cache_maint("dc civac", address, ret);
 464                 break;
 465         case ESR_ELx_SYS64_ISS_CRM_DC_CVADP:    /* DC CVADP */
 466                 __user_cache_maint("sys 3, c7, c13, 1", address, ret);
 467                 break;
 468         case ESR_ELx_SYS64_ISS_CRM_DC_CVAP:     /* DC CVAP */
 469                 __user_cache_maint("sys 3, c7, c12, 1", address, ret);
 470                 break;
 471         case ESR_ELx_SYS64_ISS_CRM_DC_CIVAC:    /* DC CIVAC */
 472                 __user_cache_maint("dc civac", address, ret);
 473                 break;
 474         case ESR_ELx_SYS64_ISS_CRM_IC_IVAU:     /* IC IVAU */
 475                 __user_cache_maint("ic ivau", address, ret);
 476                 break;
 477         default:
 478                 force_signal_inject(SIGILL, ILL_ILLOPC, regs->pc, 0);
 479                 return;
 480         }
 481
 482         if (ret)
 483                 arm64_notify_segfault(address);
 484         else
 485                 arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
 486 }
 487
 488 static void ctr_read_handler(unsigned int esr, struct pt_regs *regs)
 489 {
 490         int rt = ESR_ELx_SYS64_ISS_RT(esr);
 491         unsigned long val = arm64_ftr_reg_user_value(&arm64_ftr_reg_ctrel0);
 492
 493         if (cpus_have_const_cap(ARM64_WORKAROUND_1542419)) {
 494                 /* Hide DIC so that we can trap the unnecessary maintenance...*/
 495                 val &= ~BIT(CTR_DIC_SHIFT);
 496
 497                 /* ... and fake IminLine to reduce the number of traps. */
 498                 val &= ~CTR_IMINLINE_MASK;
 499                 val |= (PAGE_SHIFT - 2) & CTR_IMINLINE_MASK;
 500         }
 501
 502         pt_regs_write_reg(regs, rt, val);
 503
 504         arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
 505 }
 506
 507 static void cntvct_read_handler(unsigned int esr, struct pt_regs *regs)
 508 {
 509         int rt = ESR_ELx_SYS64_ISS_RT(esr);
 510
 511         pt_regs_write_reg(regs, rt, arch_timer_read_counter());
 512         arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
 513 }
 514
 515 static void cntfrq_read_handler(unsigned int esr, struct pt_regs *regs)
 516 {
 517         int rt = ESR_ELx_SYS64_ISS_RT(esr);
 518
 519         pt_regs_write_reg(regs, rt, arch_timer_get_rate());
 520         arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
 521 }
 522
 523 static void mrs_handler(unsigned int esr, struct pt_regs *regs)
 524 {
 525         u32 sysreg, rt;
 526
 527         rt = ESR_ELx_SYS64_ISS_RT(esr);
 528         sysreg = esr_sys64_to_sysreg(esr);
 529
 530         if (do_emulate_mrs(regs, sysreg, rt) != 0)
 531                 force_signal_inject(SIGILL, ILL_ILLOPC, regs->pc, 0);
 532 }
 533
 534 static void wfi_handler(unsigned int esr, struct pt_regs *regs)
 535 {
 536         arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
 537 }
 538
 539 struct sys64_hook {
 540         unsigned int esr_mask;
 541         unsigned int esr_val;
 542         void (*handler)(unsigned int esr, struct pt_regs *regs);
 543 };
 544
 545 static const struct sys64_hook sys64_hooks[] = {
 546         {
 547                 .esr_mask = ESR_ELx_SYS64_ISS_EL0_CACHE_OP_MASK,
 548                 .esr_val = ESR_ELx_SYS64_ISS_EL0_CACHE_OP_VAL,
 549                 .handler = user_cache_maint_handler,
 550         },
 551         {
 552                 /* Trap read access to CTR_EL0 */
 553                 .esr_mask = ESR_ELx_SYS64_ISS_SYS_OP_MASK,
 554                 .esr_val = ESR_ELx_SYS64_ISS_SYS_CTR_READ,
 555                 .handler = ctr_read_handler,
 556         },
 557         {
 558                 /* Trap read access to CNTVCT_EL0 */
 559                 .esr_mask = ESR_ELx_SYS64_ISS_SYS_OP_MASK,
 560                 .esr_val = ESR_ELx_SYS64_ISS_SYS_CNTVCT,
 561                 .handler = cntvct_read_handler,
 562         },
 563         {
 564                 /* Trap read access to CNTFRQ_EL0 */
 565                 .esr_mask = ESR_ELx_SYS64_ISS_SYS_OP_MASK,
 566                 .esr_val = ESR_ELx_SYS64_ISS_SYS_CNTFRQ,
 567                 .handler = cntfrq_read_handler,
 568         },
 569         {
 570                 /* Trap read access to CPUID registers */
 571                 .esr_mask = ESR_ELx_SYS64_ISS_SYS_MRS_OP_MASK,
 572                 .esr_val = ESR_ELx_SYS64_ISS_SYS_MRS_OP_VAL,
 573                 .handler = mrs_handler,
 574         },
 575         {
 576                 /* Trap WFI instructions executed in userspace */
 577                 .esr_mask = ESR_ELx_WFx_MASK,
 578                 .esr_val = ESR_ELx_WFx_WFI_VAL,
 579                 .handler = wfi_handler,
 580         },
 581         {},
 582 };
 583
 584 #ifdef CONFIG_COMPAT
 585 static bool cp15_cond_valid(unsigned int esr, struct pt_regs *regs)
 586 {
 587         int cond;
 588
 589         /* Only a T32 instruction can trap without CV being set */
 590         if (!(esr & ESR_ELx_CV)) {
 591                 u32 it;
 592
 593                 it = compat_get_it_state(regs);
 594                 if (!it)
 595                         return true;
 596
 597                 cond = it >> 4;
 598         } else {
 599                 cond = (esr & ESR_ELx_COND_MASK) >> ESR_ELx_COND_SHIFT;
 600         }
 601
 602         return aarch32_opcode_cond_checks[cond](regs->pstate);
 603 }
 604
 605 static void compat_cntfrq_read_handler(unsigned int esr, struct pt_regs *regs)
 606 {
 607         int reg = (esr & ESR_ELx_CP15_32_ISS_RT_MASK) >> ESR_ELx_CP15_32_ISS_RT_SHIFT;
 608
 609         pt_regs_write_reg(regs, reg, arch_timer_get_rate());
 610         arm64_skip_faulting_instruction(regs, 4);
 611 }
 612
 613 static const struct sys64_hook cp15_32_hooks[] = {
 614         {
 615                 .esr_mask = ESR_ELx_CP15_32_ISS_SYS_MASK,
 616                 .esr_val = ESR_ELx_CP15_32_ISS_SYS_CNTFRQ,
 617                 .handler = compat_cntfrq_read_handler,
 618         },
 619         {},
 620 };
 621
 622 static void compat_cntvct_read_handler(unsigned int esr, struct pt_regs *regs)
 623 {
 624         int rt = (esr & ESR_ELx_CP15_64_ISS_RT_MASK) >> ESR_ELx_CP15_64_ISS_RT_SHIFT;
 625         int rt2 = (esr & ESR_ELx_CP15_64_ISS_RT2_MASK) >> ESR_ELx_CP15_64_ISS_RT2_SHIFT;
 626         u64 val = arch_timer_read_counter();
 627
 628         pt_regs_write_reg(regs, rt, lower_32_bits(val));
 629         pt_regs_write_reg(regs, rt2, upper_32_bits(val));
 630         arm64_skip_faulting_instruction(regs, 4);
 631 }
 632
 633 static const struct sys64_hook cp15_64_hooks[] = {
 634         {
 635                 .esr_mask = ESR_ELx_CP15_64_ISS_SYS_MASK,
 636                 .esr_val = ESR_ELx_CP15_64_ISS_SYS_CNTVCT,
 637                 .handler = compat_cntvct_read_handler,
 638         },
 639         {},
 640 };
 641
 642 void do_cp15instr(unsigned int esr, struct pt_regs *regs)
 643 {
 644         const struct sys64_hook *hook, *hook_base;
 645
 646         if (!cp15_cond_valid(esr, regs)) {
 647                 /*
 648                  * There is no T16 variant of a CP access, so we
 649                  * always advance PC by 4 bytes.
 650                  */
 651                 arm64_skip_faulting_instruction(regs, 4);
 652                 return;
 653         }
 654
 655         switch (ESR_ELx_EC(esr)) {
 656         case ESR_ELx_EC_CP15_32:
 657                 hook_base = cp15_32_hooks;
 658                 break;
 659         case ESR_ELx_EC_CP15_64:
 660                 hook_base = cp15_64_hooks;
 661                 break;
 662         default:
 663                 do_undefinstr(regs);
 664                 return;
 665         }
 666
 667         for (hook = hook_base; hook->handler; hook++)
 668                 if ((hook->esr_mask & esr) == hook->esr_val) {
 669                         hook->handler(esr, regs);
 670                         return;
 671                 }
 672
 673         /*
 674          * New cp15 instructions may previously have been undefined at
 675          * EL0. Fall back to our usual undefined instruction handler
 676          * so that we handle these consistently.
 677          */
 678         do_undefinstr(regs);
 679 }
 680 NOKPROBE_SYMBOL(do_cp15instr);
 681 #endif
 682
 683 void do_sysinstr(unsigned int esr, struct pt_regs *regs)
 684 {
 685         const struct sys64_hook *hook;
 686
 687         for (hook = sys64_hooks; hook->handler; hook++)
 688                 if ((hook->esr_mask & esr) == hook->esr_val) {
 689                         hook->handler(esr, regs);
 690                         return;
 691                 }
 692
 693         /*
 694          * New SYS instructions may previously have been undefined at EL0. Fall
 695          * back to our usual undefined instruction handler so that we handle
 696          * these consistently.
 697          */
 698         do_undefinstr(regs);
 699 }
 700 NOKPROBE_SYMBOL(do_sysinstr);
 701
 702 static const char *esr_class_str[] = {
 703         [0 ... ESR_ELx_EC_MAX]          = "UNRECOGNIZED EC",
 704         [ESR_ELx_EC_UNKNOWN]            = "Unknown/Uncategorized",
 705         [ESR_ELx_EC_WFx]                = "WFI/WFE",
 706         [ESR_ELx_EC_CP15_32]            = "CP15 MCR/MRC",
 707         [ESR_ELx_EC_CP15_64]            = "CP15 MCRR/MRRC",
 708         [ESR_ELx_EC_CP14_MR]            = "CP14 MCR/MRC",
 709         [ESR_ELx_EC_CP14_LS]            = "CP14 LDC/STC",
 710         [ESR_ELx_EC_FP_ASIMD]           = "ASIMD",
 711         [ESR_ELx_EC_CP10_ID]            = "CP10 MRC/VMRS",
 712         [ESR_ELx_EC_PAC]                = "PAC",
 713         [ESR_ELx_EC_CP14_64]            = "CP14 MCRR/MRRC",
 714         [ESR_ELx_EC_BTI]                = "BTI",
 715         [ESR_ELx_EC_ILL]                = "PSTATE.IL",
 716         [ESR_ELx_EC_SVC32]              = "SVC (AArch32)",
 717         [ESR_ELx_EC_HVC32]              = "HVC (AArch32)",
 718         [ESR_ELx_EC_SMC32]              = "SMC (AArch32)",
 719         [ESR_ELx_EC_SVC64]              = "SVC (AArch64)",
 720         [ESR_ELx_EC_HVC64]              = "HVC (AArch64)",
 721         [ESR_ELx_EC_SMC64]              = "SMC (AArch64)",
 722         [ESR_ELx_EC_SYS64]              = "MSR/MRS (AArch64)",
 723         [ESR_ELx_EC_SVE]                = "SVE",
 724         [ESR_ELx_EC_ERET]               = "ERET/ERETAA/ERETAB",
 725         [ESR_ELx_EC_FPAC]               = "FPAC",
 726         [ESR_ELx_EC_IMP_DEF]            = "EL3 IMP DEF",
 727         [ESR_ELx_EC_IABT_LOW]           = "IABT (lower EL)",
 728         [ESR_ELx_EC_IABT_CUR]           = "IABT (current EL)",
 729         [ESR_ELx_EC_PC_ALIGN]           = "PC Alignment",
 730         [ESR_ELx_EC_DABT_LOW]           = "DABT (lower EL)",
 731         [ESR_ELx_EC_DABT_CUR]           = "DABT (current EL)",
 732         [ESR_ELx_EC_SP_ALIGN]           = "SP Alignment",
 733         [ESR_ELx_EC_FP_EXC32]           = "FP (AArch32)",
 734         [ESR_ELx_EC_FP_EXC64]           = "FP (AArch64)",
 735         [ESR_ELx_EC_SERROR]             = "SError",
 736         [ESR_ELx_EC_BREAKPT_LOW]        = "Breakpoint (lower EL)",
 737         [ESR_ELx_EC_BREAKPT_CUR]        = "Breakpoint (current EL)",
 738         [ESR_ELx_EC_SOFTSTP_LOW]        = "Software Step (lower EL)",
 739         [ESR_ELx_EC_SOFTSTP_CUR]        = "Software Step (current EL)",
 740         [ESR_ELx_EC_WATCHPT_LOW]        = "Watchpoint (lower EL)",
 741         [ESR_ELx_EC_WATCHPT_CUR]        = "Watchpoint (current EL)",
 742         [ESR_ELx_EC_BKPT32]             = "BKPT (AArch32)",
 743         [ESR_ELx_EC_VECTOR32]           = "Vector catch (AArch32)",
 744         [ESR_ELx_EC_BRK64]              = "BRK (AArch64)",
 745 };
 746
 747 const char *esr_get_class_string(u32 esr)
 748 {
 749         return esr_class_str[ESR_ELx_EC(esr)];
 750 }
 751
 752 /*
 753  * bad_mode handles the impossible case in the exception vector. This is always
 754  * fatal.
 755  */
 756 asmlinkage void bad_mode(struct pt_regs *regs, int reason, unsigned int esr)
 757 {
 758         console_verbose();
 759
 760         pr_crit("Bad mode in %s handler detected on CPU%d, code 0x%08x -- %s\n",
 761                 handler[reason], smp_processor_id(), esr,
 762                 esr_get_class_string(esr));
 763
 764         __show_regs(regs);
 765         local_daif_mask();
 766         panic("bad mode");
 767 }
 768
 769 /*
 770  * bad_el0_sync handles unexpected, but potentially recoverable synchronous
 771  * exceptions taken from EL0. Unlike bad_mode, this returns.
 772  */
 773 void bad_el0_sync(struct pt_regs *regs, int reason, unsigned int esr)
 774 {
 775         void __user *pc = (void __user *)instruction_pointer(regs);
 776
 777         current->thread.fault_address = 0;
 778         current->thread.fault_code = esr;
 779
 780         arm64_force_sig_fault(SIGILL, ILL_ILLOPC, pc,
 781                               "Bad EL0 synchronous exception");
 782 }
 783
 784 #ifdef CONFIG_VMAP_STACK
 785
 786 DEFINE_PER_CPU(unsigned long [OVERFLOW_STACK_SIZE/sizeof(long)], overflow_stack)
 787         __aligned(16);
 788
 789 asmlinkage void handle_bad_stack(struct pt_regs *regs)
 790 {
 791         unsigned long tsk_stk = (unsigned long)current->stack;
 792         unsigned long irq_stk = (unsigned long)this_cpu_read(irq_stack_ptr);
 793         unsigned long ovf_stk = (unsigned long)this_cpu_ptr(overflow_stack);
 794         unsigned int esr = read_sysreg(esr_el1);
 795         unsigned long far = read_sysreg(far_el1);
 796
 797         console_verbose();
 798         pr_emerg("Insufficient stack space to handle exception!");
 799
 800         pr_emerg("ESR: 0x%08x -- %s\n", esr, esr_get_class_string(esr));
 801         pr_emerg("FAR: 0x%016lx\n", far);
 802
 803         pr_emerg("Task stack:     [0x%016lx..0x%016lx]\n",
 804                  tsk_stk, tsk_stk + THREAD_SIZE);
 805         pr_emerg("IRQ stack:      [0x%016lx..0x%016lx]\n",
 806                  irq_stk, irq_stk + IRQ_STACK_SIZE);
 807         pr_emerg("Overflow stack: [0x%016lx..0x%016lx]\n",
 808                  ovf_stk, ovf_stk + OVERFLOW_STACK_SIZE);
 809
 810         __show_regs(regs);
 811
 812         /*
 813          * We use nmi_panic to limit the potential for recusive overflows, and
 814          * to get a better stack trace.
 815          */
 816         nmi_panic(NULL, "kernel stack overflow");
 817         cpu_park_loop();
 818 }
 819 #endif
 820
 821 void __noreturn arm64_serror_panic(struct pt_regs *regs, u32 esr)
 822 {
 823         console_verbose();
 824
 825         pr_crit("SError Interrupt on CPU%d, code 0x%08x -- %s\n",
 826                 smp_processor_id(), esr, esr_get_class_string(esr));
 827         if (regs)
 828                 __show_regs(regs);
 829
 830         nmi_panic(regs, "Asynchronous SError Interrupt");
 831
 832         cpu_park_loop();
 833         unreachable();
 834 }
 835
 836 bool arm64_is_fatal_ras_serror(struct pt_regs *regs, unsigned int esr)
 837 {
 838         u32 aet = arm64_ras_serror_get_severity(esr);
 839
 840         switch (aet) {
 841         case ESR_ELx_AET_CE:    /* corrected error */
 842         case ESR_ELx_AET_UEO:   /* restartable, not yet consumed */
 843                 /*
 844                  * The CPU can make progress. We may take UEO again as
 845                  * a more severe error.
 846                  */
 847                 return false;
 848
 849         case ESR_ELx_AET_UEU:   /* Uncorrected Unrecoverable */
 850         case ESR_ELx_AET_UER:   /* Uncorrected Recoverable */
 851                 /*
 852                  * The CPU can't make progress. The exception may have
 853                  * been imprecise.
 854                  *
 855                  * Neoverse-N1 #1349291 means a non-KVM SError reported as
 856                  * Unrecoverable should be treated as Uncontainable. We
 857                  * call arm64_serror_panic() in both cases.
 858                  */
 859                 return true;
 860
 861         case ESR_ELx_AET_UC:    /* Uncontainable or Uncategorized error */
 862         default:
 863                 /* Error has been silently propagated */
 864                 arm64_serror_panic(regs, esr);
 865         }
 866 }
 867
 868 asmlinkage void do_serror(struct pt_regs *regs, unsigned int esr)
 869 {
 870         nmi_enter();
 871
 872         /* non-RAS errors are not containable */
 873         if (!arm64_is_ras_serror(esr) || arm64_is_fatal_ras_serror(regs, esr))
 874                 arm64_serror_panic(regs, esr);
 875
 876         nmi_exit();
 877 }
 878
 879 asmlinkage void enter_from_user_mode(void)
 880 {
 881         CT_WARN_ON(ct_state() != CONTEXT_USER);
 882         user_exit_irqoff();
 883 }
 884 NOKPROBE_SYMBOL(enter_from_user_mode);
 885
 886 /* GENERIC_BUG traps */
 887
 888 int is_valid_bugaddr(unsigned long addr)
 889 {
 890         /*
 891          * bug_handler() only called for BRK #BUG_BRK_IMM.
 892          * So the answer is trivial -- any spurious instances with no
 893          * bug table entry will be rejected by report_bug() and passed
 894          * back to the debug-monitors code and handled as a fatal
 895          * unexpected debug exception.
 896          */
 897         return 1;
 898 }
 899
 900 static int bug_handler(struct pt_regs *regs, unsigned int esr)
 901 {
 902         switch (report_bug(regs->pc, regs)) {
 903         case BUG_TRAP_TYPE_BUG:
 904                 die("Oops - BUG", regs, 0);
 905                 break;
 906
 907         case BUG_TRAP_TYPE_WARN:
 908                 break;
 909
 910         default:
 911                 /* unknown/unrecognised bug trap type */
 912                 return DBG_HOOK_ERROR;
 913         }
 914
 915         /* If thread survives, skip over the BUG instruction and continue: */
 916         arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
 917         return DBG_HOOK_HANDLED;
 918 }
 919
 920 static struct break_hook bug_break_hook = {
 921         .fn = bug_handler,
 922         .imm = BUG_BRK_IMM,
 923 };
 924
 925 static int reserved_fault_handler(struct pt_regs *regs, unsigned int esr)
 926 {
 927         pr_err("%s generated an invalid instruction at %pS!\n",
 928                 in_bpf_jit(regs) ? "BPF JIT" : "Kernel text patching",
 929                 (void *)instruction_pointer(regs));
 930
 931         /* We cannot handle this */
 932         return DBG_HOOK_ERROR;
 933 }
 934
 935 static struct break_hook fault_break_hook = {
 936         .fn = reserved_fault_handler,
 937         .imm = FAULT_BRK_IMM,
 938 };
 939
 940 #ifdef CONFIG_KASAN_SW_TAGS
 941
 942 #define KASAN_ESR_RECOVER       0x20
 943 #define KASAN_ESR_WRITE 0x10
 944 #define KASAN_ESR_SIZE_MASK     0x0f
 945 #define KASAN_ESR_SIZE(esr)     (1 << ((esr) & KASAN_ESR_SIZE_MASK))
 946
 947 static int kasan_handler(struct pt_regs *regs, unsigned int esr)
 948 {
 949         bool recover = esr & KASAN_ESR_RECOVER;
 950         bool write = esr & KASAN_ESR_WRITE;
 951         size_t size = KASAN_ESR_SIZE(esr);
 952         u64 addr = regs->regs[0];
 953         u64 pc = regs->pc;
 954
 955         kasan_report(addr, size, write, pc);
 956
 957         /*
 958          * The instrumentation allows to control whether we can proceed after
 959          * a crash was detected. This is done by passing the -recover flag to
 960          * the compiler. Disabling recovery allows to generate more compact
 961          * code.
 962          *
 963          * Unfortunately disabling recovery doesn't work for the kernel right
 964          * now. KASAN reporting is disabled in some contexts (for example when
 965          * the allocator accesses slab object metadata; this is controlled by
 966          * current->kasan_depth). All these accesses are detected by the tool,
 967          * even though the reports for them are not printed.
 968          *
 969          * This is something that might be fixed at some point in the future.
 970          */
 971         if (!recover)
 972                 die("Oops - KASAN", regs, 0);
 973
 974         /* If thread survives, skip over the brk instruction and continue: */
 975         arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
 976         return DBG_HOOK_HANDLED;
 977 }
 978
 979 static struct break_hook kasan_break_hook = {
 980         .fn     = kasan_handler,
 981         .imm    = KASAN_BRK_IMM,
 982         .mask   = KASAN_BRK_MASK,
 983 };
 984 #endif
 985
 986 /*
 987  * Initial handler for AArch64 BRK exceptions
 988  * This handler only used until debug_traps_init().
 989  */
 990 int __init early_brk64(unsigned long addr, unsigned int esr,
 991                 struct pt_regs *regs)
 992 {
 993 #ifdef CONFIG_KASAN_SW_TAGS
 994         unsigned int comment = esr & ESR_ELx_BRK64_ISS_COMMENT_MASK;
 995
 996         if ((comment & ~KASAN_BRK_MASK) == KASAN_BRK_IMM)
 997                 return kasan_handler(regs, esr) != DBG_HOOK_HANDLED;
 998 #endif
 999         return bug_handler(regs, esr) != DBG_HOOK_HANDLED;
1000 }
1001
1002 void __init trap_init(void)
1003 {
1004         register_kernel_break_hook(&bug_break_hook);
1005         register_kernel_break_hook(&fault_break_hook);
1006 #ifdef CONFIG_KASAN_SW_TAGS
1007         register_kernel_break_hook(&kasan_break_hook);
1008 #endif
1009         debug_traps_init();
1010 }