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/kallsyms.h>
13 #include <linux/kprobes.h>
14 #include <linux/spinlock.h>
15 #include <linux/uaccess.h>
16 #include <linux/hardirq.h>
17 #include <linux/kdebug.h>
18 #include <linux/module.h>
19 #include <linux/kexec.h>
20 #include <linux/delay.h>
21 #include <linux/efi.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 #include <linux/ubsan.h>
31 #include <linux/cfi.h>
32
33 #include <asm/atomic.h>
34 #include <asm/bug.h>
35 #include <asm/cpufeature.h>
36 #include <asm/daifflags.h>
37 #include <asm/debug-monitors.h>
38 #include <asm/efi.h>
39 #include <asm/esr.h>
40 #include <asm/exception.h>
41 #include <asm/extable.h>
42 #include <asm/insn.h>
43 #include <asm/kprobes.h>
44 #include <asm/patching.h>
45 #include <asm/traps.h>
46 #include <asm/smp.h>
47 #include <asm/stack_pointer.h>
48 #include <asm/stacktrace.h>
49 #include <asm/system_misc.h>
50 #include <asm/sysreg.h>
51
__check_eq(unsigned long pstate)52 static bool __kprobes __check_eq(unsigned long pstate)
53 {
54 return (pstate & PSR_Z_BIT) != 0;
55 }
56
__check_ne(unsigned long pstate)57 static bool __kprobes __check_ne(unsigned long pstate)
58 {
59 return (pstate & PSR_Z_BIT) == 0;
60 }
61
__check_cs(unsigned long pstate)62 static bool __kprobes __check_cs(unsigned long pstate)
63 {
64 return (pstate & PSR_C_BIT) != 0;
65 }
66
__check_cc(unsigned long pstate)67 static bool __kprobes __check_cc(unsigned long pstate)
68 {
69 return (pstate & PSR_C_BIT) == 0;
70 }
71
__check_mi(unsigned long pstate)72 static bool __kprobes __check_mi(unsigned long pstate)
73 {
74 return (pstate & PSR_N_BIT) != 0;
75 }
76
__check_pl(unsigned long pstate)77 static bool __kprobes __check_pl(unsigned long pstate)
78 {
79 return (pstate & PSR_N_BIT) == 0;
80 }
81
__check_vs(unsigned long pstate)82 static bool __kprobes __check_vs(unsigned long pstate)
83 {
84 return (pstate & PSR_V_BIT) != 0;
85 }
86
__check_vc(unsigned long pstate)87 static bool __kprobes __check_vc(unsigned long pstate)
88 {
89 return (pstate & PSR_V_BIT) == 0;
90 }
91
__check_hi(unsigned long pstate)92 static bool __kprobes __check_hi(unsigned long pstate)
93 {
94 pstate &= ~(pstate >> 1); /* PSR_C_BIT &= ~PSR_Z_BIT */
95 return (pstate & PSR_C_BIT) != 0;
96 }
97
__check_ls(unsigned long pstate)98 static bool __kprobes __check_ls(unsigned long pstate)
99 {
100 pstate &= ~(pstate >> 1); /* PSR_C_BIT &= ~PSR_Z_BIT */
101 return (pstate & PSR_C_BIT) == 0;
102 }
103
__check_ge(unsigned long pstate)104 static bool __kprobes __check_ge(unsigned long pstate)
105 {
106 pstate ^= (pstate << 3); /* PSR_N_BIT ^= PSR_V_BIT */
107 return (pstate & PSR_N_BIT) == 0;
108 }
109
__check_lt(unsigned long pstate)110 static bool __kprobes __check_lt(unsigned long pstate)
111 {
112 pstate ^= (pstate << 3); /* PSR_N_BIT ^= PSR_V_BIT */
113 return (pstate & PSR_N_BIT) != 0;
114 }
115
__check_gt(unsigned long pstate)116 static bool __kprobes __check_gt(unsigned long pstate)
117 {
118 /*PSR_N_BIT ^= PSR_V_BIT */
119 unsigned long temp = pstate ^ (pstate << 3);
120
121 temp |= (pstate << 1); /*PSR_N_BIT |= PSR_Z_BIT */
122 return (temp & PSR_N_BIT) == 0;
123 }
124
__check_le(unsigned long pstate)125 static bool __kprobes __check_le(unsigned long pstate)
126 {
127 /*PSR_N_BIT ^= PSR_V_BIT */
128 unsigned long temp = pstate ^ (pstate << 3);
129
130 temp |= (pstate << 1); /*PSR_N_BIT |= PSR_Z_BIT */
131 return (temp & PSR_N_BIT) != 0;
132 }
133
__check_al(unsigned long pstate)134 static bool __kprobes __check_al(unsigned long pstate)
135 {
136 return true;
137 }
138
139 /*
140 * Note that the ARMv8 ARM calls condition code 0b1111 "nv", but states that
141 * it behaves identically to 0b1110 ("al").
142 */
143 pstate_check_t * const aarch32_opcode_cond_checks[16] = {
144 __check_eq, __check_ne, __check_cs, __check_cc,
145 __check_mi, __check_pl, __check_vs, __check_vc,
146 __check_hi, __check_ls, __check_ge, __check_lt,
147 __check_gt, __check_le, __check_al, __check_al
148 };
149
150 int show_unhandled_signals = 0;
151
dump_kernel_instr(const char * lvl,struct pt_regs * regs)152 static void dump_kernel_instr(const char *lvl, struct pt_regs *regs)
153 {
154 unsigned long addr = instruction_pointer(regs);
155 char str[sizeof("00000000 ") * 5 + 2 + 1], *p = str;
156 int i;
157
158 if (user_mode(regs))
159 return;
160
161 for (i = -4; i < 1; i++) {
162 unsigned int val, bad;
163
164 bad = aarch64_insn_read(&((u32 *)addr)[i], &val);
165
166 if (!bad)
167 p += sprintf(p, i == 0 ? "(%08x) " : "%08x ", val);
168 else
169 p += sprintf(p, i == 0 ? "(????????) " : "???????? ");
170 }
171
172 printk("%sCode: %s\n", lvl, str);
173 }
174
175 #ifdef CONFIG_PREEMPT
176 #define S_PREEMPT " PREEMPT"
177 #elif defined(CONFIG_PREEMPT_RT)
178 #define S_PREEMPT " PREEMPT_RT"
179 #else
180 #define S_PREEMPT ""
181 #endif
182
183 #define S_SMP " SMP"
184
__die(const char * str,long err,struct pt_regs * regs)185 static int __die(const char *str, long err, struct pt_regs *regs)
186 {
187 static int die_counter;
188 int ret;
189
190 pr_emerg("Internal error: %s: %016lx [#%d]" S_PREEMPT S_SMP "\n",
191 str, err, ++die_counter);
192
193 /* trap and error numbers are mostly meaningless on ARM */
194 ret = notify_die(DIE_OOPS, str, regs, err, 0, SIGSEGV);
195 if (ret == NOTIFY_STOP)
196 return ret;
197
198 print_modules();
199 show_regs(regs);
200
201 dump_kernel_instr(KERN_EMERG, regs);
202
203 return ret;
204 }
205
206 static DEFINE_RAW_SPINLOCK(die_lock);
207
208 /*
209 * This function is protected against re-entrancy.
210 */
die(const char * str,struct pt_regs * regs,long err)211 void die(const char *str, struct pt_regs *regs, long err)
212 {
213 int ret;
214 unsigned long flags;
215
216 raw_spin_lock_irqsave(&die_lock, flags);
217
218 oops_enter();
219
220 console_verbose();
221 bust_spinlocks(1);
222 ret = __die(str, err, regs);
223
224 if (regs && kexec_should_crash(current))
225 crash_kexec(regs);
226
227 bust_spinlocks(0);
228 add_taint(TAINT_DIE, LOCKDEP_NOW_UNRELIABLE);
229 oops_exit();
230
231 if (in_interrupt())
232 panic("%s: Fatal exception in interrupt", str);
233 if (panic_on_oops)
234 panic("%s: Fatal exception", str);
235
236 raw_spin_unlock_irqrestore(&die_lock, flags);
237
238 if (ret != NOTIFY_STOP)
239 make_task_dead(SIGSEGV);
240 }
241
arm64_show_signal(int signo,const char * str)242 static void arm64_show_signal(int signo, const char *str)
243 {
244 static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,
245 DEFAULT_RATELIMIT_BURST);
246 struct task_struct *tsk = current;
247 unsigned long esr = tsk->thread.fault_code;
248 struct pt_regs *regs = task_pt_regs(tsk);
249
250 /* Leave if the signal won't be shown */
251 if (!show_unhandled_signals ||
252 !unhandled_signal(tsk, signo) ||
253 !__ratelimit(&rs))
254 return;
255
256 pr_info("%s[%d]: unhandled exception: ", tsk->comm, task_pid_nr(tsk));
257 if (esr)
258 pr_cont("%s, ESR 0x%016lx, ", esr_get_class_string(esr), esr);
259
260 pr_cont("%s", str);
261 print_vma_addr(KERN_CONT " in ", regs->pc);
262 pr_cont("\n");
263 __show_regs(regs);
264 }
265
arm64_force_sig_fault(int signo,int code,unsigned long far,const char * str)266 void arm64_force_sig_fault(int signo, int code, unsigned long far,
267 const char *str)
268 {
269 arm64_show_signal(signo, str);
270 if (signo == SIGKILL)
271 force_sig(SIGKILL);
272 else
273 force_sig_fault(signo, code, (void __user *)far);
274 }
275
arm64_force_sig_mceerr(int code,unsigned long far,short lsb,const char * str)276 void arm64_force_sig_mceerr(int code, unsigned long far, short lsb,
277 const char *str)
278 {
279 arm64_show_signal(SIGBUS, str);
280 force_sig_mceerr(code, (void __user *)far, lsb);
281 }
282
arm64_force_sig_ptrace_errno_trap(int errno,unsigned long far,const char * str)283 void arm64_force_sig_ptrace_errno_trap(int errno, unsigned long far,
284 const char *str)
285 {
286 arm64_show_signal(SIGTRAP, str);
287 force_sig_ptrace_errno_trap(errno, (void __user *)far);
288 }
289
arm64_notify_die(const char * str,struct pt_regs * regs,int signo,int sicode,unsigned long far,unsigned long err)290 void arm64_notify_die(const char *str, struct pt_regs *regs,
291 int signo, int sicode, unsigned long far,
292 unsigned long err)
293 {
294 if (user_mode(regs)) {
295 WARN_ON(regs != current_pt_regs());
296 current->thread.fault_address = 0;
297 current->thread.fault_code = err;
298
299 arm64_force_sig_fault(signo, sicode, far, str);
300 } else {
301 die(str, regs, err);
302 }
303 }
304
305 #ifdef CONFIG_COMPAT
306 #define PSTATE_IT_1_0_SHIFT 25
307 #define PSTATE_IT_1_0_MASK (0x3 << PSTATE_IT_1_0_SHIFT)
308 #define PSTATE_IT_7_2_SHIFT 10
309 #define PSTATE_IT_7_2_MASK (0x3f << PSTATE_IT_7_2_SHIFT)
310
compat_get_it_state(struct pt_regs * regs)311 static u32 compat_get_it_state(struct pt_regs *regs)
312 {
313 u32 it, pstate = regs->pstate;
314
315 it = (pstate & PSTATE_IT_1_0_MASK) >> PSTATE_IT_1_0_SHIFT;
316 it |= ((pstate & PSTATE_IT_7_2_MASK) >> PSTATE_IT_7_2_SHIFT) << 2;
317
318 return it;
319 }
320
compat_set_it_state(struct pt_regs * regs,u32 it)321 static void compat_set_it_state(struct pt_regs *regs, u32 it)
322 {
323 u32 pstate_it;
324
325 pstate_it = (it << PSTATE_IT_1_0_SHIFT) & PSTATE_IT_1_0_MASK;
326 pstate_it |= ((it >> 2) << PSTATE_IT_7_2_SHIFT) & PSTATE_IT_7_2_MASK;
327
328 regs->pstate &= ~PSR_AA32_IT_MASK;
329 regs->pstate |= pstate_it;
330 }
331
advance_itstate(struct pt_regs * regs)332 static void advance_itstate(struct pt_regs *regs)
333 {
334 u32 it;
335
336 /* ARM mode */
337 if (!(regs->pstate & PSR_AA32_T_BIT) ||
338 !(regs->pstate & PSR_AA32_IT_MASK))
339 return;
340
341 it = compat_get_it_state(regs);
342
343 /*
344 * If this is the last instruction of the block, wipe the IT
345 * state. Otherwise advance it.
346 */
347 if (!(it & 7))
348 it = 0;
349 else
350 it = (it & 0xe0) | ((it << 1) & 0x1f);
351
352 compat_set_it_state(regs, it);
353 }
354 #else
advance_itstate(struct pt_regs * regs)355 static void advance_itstate(struct pt_regs *regs)
356 {
357 }
358 #endif
359
arm64_skip_faulting_instruction(struct pt_regs * regs,unsigned long size)360 void arm64_skip_faulting_instruction(struct pt_regs *regs, unsigned long size)
361 {
362 regs->pc += size;
363
364 /*
365 * If we were single stepping, we want to get the step exception after
366 * we return from the trap.
367 */
368 if (user_mode(regs))
369 user_fastforward_single_step(current);
370
371 if (compat_user_mode(regs))
372 advance_itstate(regs);
373 else
374 regs->pstate &= ~PSR_BTYPE_MASK;
375 }
376
user_insn_read(struct pt_regs * regs,u32 * insnp)377 static int user_insn_read(struct pt_regs *regs, u32 *insnp)
378 {
379 u32 instr;
380 unsigned long pc = instruction_pointer(regs);
381
382 if (compat_thumb_mode(regs)) {
383 /* 16-bit Thumb instruction */
384 __le16 instr_le;
385 if (get_user(instr_le, (__le16 __user *)pc))
386 return -EFAULT;
387 instr = le16_to_cpu(instr_le);
388 if (aarch32_insn_is_wide(instr)) {
389 u32 instr2;
390
391 if (get_user(instr_le, (__le16 __user *)(pc + 2)))
392 return -EFAULT;
393 instr2 = le16_to_cpu(instr_le);
394 instr = (instr << 16) | instr2;
395 }
396 } else {
397 /* 32-bit ARM instruction */
398 __le32 instr_le;
399 if (get_user(instr_le, (__le32 __user *)pc))
400 return -EFAULT;
401 instr = le32_to_cpu(instr_le);
402 }
403
404 *insnp = instr;
405 return 0;
406 }
407
force_signal_inject(int signal,int code,unsigned long address,unsigned long err)408 void force_signal_inject(int signal, int code, unsigned long address, unsigned long err)
409 {
410 const char *desc;
411 struct pt_regs *regs = current_pt_regs();
412
413 if (WARN_ON(!user_mode(regs)))
414 return;
415
416 switch (signal) {
417 case SIGILL:
418 desc = "undefined instruction";
419 break;
420 case SIGSEGV:
421 desc = "illegal memory access";
422 break;
423 default:
424 desc = "unknown or unrecoverable error";
425 break;
426 }
427
428 /* Force signals we don't understand to SIGKILL */
429 if (WARN_ON(signal != SIGKILL &&
430 siginfo_layout(signal, code) != SIL_FAULT)) {
431 signal = SIGKILL;
432 }
433
434 arm64_notify_die(desc, regs, signal, code, address, err);
435 }
436
437 /*
438 * Set up process info to signal segmentation fault - called on access error.
439 */
arm64_notify_segfault(unsigned long addr)440 void arm64_notify_segfault(unsigned long addr)
441 {
442 int code;
443
444 mmap_read_lock(current->mm);
445 if (find_vma(current->mm, untagged_addr(addr)) == NULL)
446 code = SEGV_MAPERR;
447 else
448 code = SEGV_ACCERR;
449 mmap_read_unlock(current->mm);
450
451 force_signal_inject(SIGSEGV, code, addr, 0);
452 }
453
do_el0_undef(struct pt_regs * regs,unsigned long esr)454 void do_el0_undef(struct pt_regs *regs, unsigned long esr)
455 {
456 u32 insn;
457
458 /* check for AArch32 breakpoint instructions */
459 if (!aarch32_break_handler(regs))
460 return;
461
462 if (user_insn_read(regs, &insn))
463 goto out_err;
464
465 if (try_emulate_mrs(regs, insn))
466 return;
467
468 if (try_emulate_armv8_deprecated(regs, insn))
469 return;
470
471 out_err:
472 force_signal_inject(SIGILL, ILL_ILLOPC, regs->pc, 0);
473 }
474
do_el1_undef(struct pt_regs * regs,unsigned long esr)475 void do_el1_undef(struct pt_regs *regs, unsigned long esr)
476 {
477 u32 insn;
478
479 if (aarch64_insn_read((void *)regs->pc, &insn))
480 goto out_err;
481
482 if (try_emulate_el1_ssbs(regs, insn))
483 return;
484
485 out_err:
486 die("Oops - Undefined instruction", regs, esr);
487 }
488
do_el0_bti(struct pt_regs * regs)489 void do_el0_bti(struct pt_regs *regs)
490 {
491 force_signal_inject(SIGILL, ILL_ILLOPC, regs->pc, 0);
492 }
493
do_el1_bti(struct pt_regs * regs,unsigned long esr)494 void do_el1_bti(struct pt_regs *regs, unsigned long esr)
495 {
496 if (efi_runtime_fixup_exception(regs, "BTI violation")) {
497 regs->pstate &= ~PSR_BTYPE_MASK;
498 return;
499 }
500 die("Oops - BTI", regs, esr);
501 }
502
do_el0_fpac(struct pt_regs * regs,unsigned long esr)503 void do_el0_fpac(struct pt_regs *regs, unsigned long esr)
504 {
505 force_signal_inject(SIGILL, ILL_ILLOPN, regs->pc, esr);
506 }
507
do_el1_fpac(struct pt_regs * regs,unsigned long esr)508 void do_el1_fpac(struct pt_regs *regs, unsigned long esr)
509 {
510 /*
511 * Unexpected FPAC exception in the kernel: kill the task before it
512 * does any more harm.
513 */
514 die("Oops - FPAC", regs, esr);
515 }
516
do_el0_mops(struct pt_regs * regs,unsigned long esr)517 void do_el0_mops(struct pt_regs *regs, unsigned long esr)
518 {
519 arm64_mops_reset_regs(®s->user_regs, esr);
520
521 /*
522 * If single stepping then finish the step before executing the
523 * prologue instruction.
524 */
525 user_fastforward_single_step(current);
526 }
527
528 #define __user_cache_maint(insn, address, res) \
529 if (address >= TASK_SIZE_MAX) { \
530 res = -EFAULT; \
531 } else { \
532 uaccess_ttbr0_enable(); \
533 asm volatile ( \
534 "1: " insn ", %1\n" \
535 " mov %w0, #0\n" \
536 "2:\n" \
537 _ASM_EXTABLE_UACCESS_ERR(1b, 2b, %w0) \
538 : "=r" (res) \
539 : "r" (address)); \
540 uaccess_ttbr0_disable(); \
541 }
542
user_cache_maint_handler(unsigned long esr,struct pt_regs * regs)543 static void user_cache_maint_handler(unsigned long esr, struct pt_regs *regs)
544 {
545 unsigned long tagged_address, address;
546 int rt = ESR_ELx_SYS64_ISS_RT(esr);
547 int crm = (esr & ESR_ELx_SYS64_ISS_CRM_MASK) >> ESR_ELx_SYS64_ISS_CRM_SHIFT;
548 int ret = 0;
549
550 tagged_address = pt_regs_read_reg(regs, rt);
551 address = untagged_addr(tagged_address);
552
553 switch (crm) {
554 case ESR_ELx_SYS64_ISS_CRM_DC_CVAU: /* DC CVAU, gets promoted */
555 __user_cache_maint("dc civac", address, ret);
556 break;
557 case ESR_ELx_SYS64_ISS_CRM_DC_CVAC: /* DC CVAC, gets promoted */
558 __user_cache_maint("dc civac", address, ret);
559 break;
560 case ESR_ELx_SYS64_ISS_CRM_DC_CVADP: /* DC CVADP */
561 __user_cache_maint("sys 3, c7, c13, 1", address, ret);
562 break;
563 case ESR_ELx_SYS64_ISS_CRM_DC_CVAP: /* DC CVAP */
564 __user_cache_maint("sys 3, c7, c12, 1", address, ret);
565 break;
566 case ESR_ELx_SYS64_ISS_CRM_DC_CIVAC: /* DC CIVAC */
567 __user_cache_maint("dc civac", address, ret);
568 break;
569 case ESR_ELx_SYS64_ISS_CRM_IC_IVAU: /* IC IVAU */
570 __user_cache_maint("ic ivau", address, ret);
571 break;
572 default:
573 force_signal_inject(SIGILL, ILL_ILLOPC, regs->pc, 0);
574 return;
575 }
576
577 if (ret)
578 arm64_notify_segfault(tagged_address);
579 else
580 arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
581 }
582
ctr_read_handler(unsigned long esr,struct pt_regs * regs)583 static void ctr_read_handler(unsigned long esr, struct pt_regs *regs)
584 {
585 int rt = ESR_ELx_SYS64_ISS_RT(esr);
586 unsigned long val = arm64_ftr_reg_user_value(&arm64_ftr_reg_ctrel0);
587
588 if (cpus_have_final_cap(ARM64_WORKAROUND_1542419)) {
589 /* Hide DIC so that we can trap the unnecessary maintenance...*/
590 val &= ~BIT(CTR_EL0_DIC_SHIFT);
591
592 /* ... and fake IminLine to reduce the number of traps. */
593 val &= ~CTR_EL0_IminLine_MASK;
594 val |= (PAGE_SHIFT - 2) & CTR_EL0_IminLine_MASK;
595 }
596
597 pt_regs_write_reg(regs, rt, val);
598
599 arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
600 }
601
cntvct_read_handler(unsigned long esr,struct pt_regs * regs)602 static void cntvct_read_handler(unsigned long esr, struct pt_regs *regs)
603 {
604 int rt = ESR_ELx_SYS64_ISS_RT(esr);
605
606 pt_regs_write_reg(regs, rt, arch_timer_read_counter());
607 arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
608 }
609
cntfrq_read_handler(unsigned long esr,struct pt_regs * regs)610 static void cntfrq_read_handler(unsigned long esr, struct pt_regs *regs)
611 {
612 int rt = ESR_ELx_SYS64_ISS_RT(esr);
613
614 pt_regs_write_reg(regs, rt, arch_timer_get_rate());
615 arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
616 }
617
mrs_handler(unsigned long esr,struct pt_regs * regs)618 static void mrs_handler(unsigned long esr, struct pt_regs *regs)
619 {
620 u32 sysreg, rt;
621
622 rt = ESR_ELx_SYS64_ISS_RT(esr);
623 sysreg = esr_sys64_to_sysreg(esr);
624
625 if (do_emulate_mrs(regs, sysreg, rt) != 0)
626 force_signal_inject(SIGILL, ILL_ILLOPC, regs->pc, 0);
627 }
628
wfi_handler(unsigned long esr,struct pt_regs * regs)629 static void wfi_handler(unsigned long esr, struct pt_regs *regs)
630 {
631 arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
632 }
633
634 struct sys64_hook {
635 unsigned long esr_mask;
636 unsigned long esr_val;
637 void (*handler)(unsigned long esr, struct pt_regs *regs);
638 };
639
640 static const struct sys64_hook sys64_hooks[] = {
641 {
642 .esr_mask = ESR_ELx_SYS64_ISS_EL0_CACHE_OP_MASK,
643 .esr_val = ESR_ELx_SYS64_ISS_EL0_CACHE_OP_VAL,
644 .handler = user_cache_maint_handler,
645 },
646 {
647 /* Trap read access to CTR_EL0 */
648 .esr_mask = ESR_ELx_SYS64_ISS_SYS_OP_MASK,
649 .esr_val = ESR_ELx_SYS64_ISS_SYS_CTR_READ,
650 .handler = ctr_read_handler,
651 },
652 {
653 /* Trap read access to CNTVCT_EL0 */
654 .esr_mask = ESR_ELx_SYS64_ISS_SYS_OP_MASK,
655 .esr_val = ESR_ELx_SYS64_ISS_SYS_CNTVCT,
656 .handler = cntvct_read_handler,
657 },
658 {
659 /* Trap read access to CNTVCTSS_EL0 */
660 .esr_mask = ESR_ELx_SYS64_ISS_SYS_OP_MASK,
661 .esr_val = ESR_ELx_SYS64_ISS_SYS_CNTVCTSS,
662 .handler = cntvct_read_handler,
663 },
664 {
665 /* Trap read access to CNTFRQ_EL0 */
666 .esr_mask = ESR_ELx_SYS64_ISS_SYS_OP_MASK,
667 .esr_val = ESR_ELx_SYS64_ISS_SYS_CNTFRQ,
668 .handler = cntfrq_read_handler,
669 },
670 {
671 /* Trap read access to CPUID registers */
672 .esr_mask = ESR_ELx_SYS64_ISS_SYS_MRS_OP_MASK,
673 .esr_val = ESR_ELx_SYS64_ISS_SYS_MRS_OP_VAL,
674 .handler = mrs_handler,
675 },
676 {
677 /* Trap WFI instructions executed in userspace */
678 .esr_mask = ESR_ELx_WFx_MASK,
679 .esr_val = ESR_ELx_WFx_WFI_VAL,
680 .handler = wfi_handler,
681 },
682 {},
683 };
684
685 #ifdef CONFIG_COMPAT
cp15_cond_valid(unsigned long esr,struct pt_regs * regs)686 static bool cp15_cond_valid(unsigned long esr, struct pt_regs *regs)
687 {
688 int cond;
689
690 /* Only a T32 instruction can trap without CV being set */
691 if (!(esr & ESR_ELx_CV)) {
692 u32 it;
693
694 it = compat_get_it_state(regs);
695 if (!it)
696 return true;
697
698 cond = it >> 4;
699 } else {
700 cond = (esr & ESR_ELx_COND_MASK) >> ESR_ELx_COND_SHIFT;
701 }
702
703 return aarch32_opcode_cond_checks[cond](regs->pstate);
704 }
705
compat_cntfrq_read_handler(unsigned long esr,struct pt_regs * regs)706 static void compat_cntfrq_read_handler(unsigned long esr, struct pt_regs *regs)
707 {
708 int reg = (esr & ESR_ELx_CP15_32_ISS_RT_MASK) >> ESR_ELx_CP15_32_ISS_RT_SHIFT;
709
710 pt_regs_write_reg(regs, reg, arch_timer_get_rate());
711 arm64_skip_faulting_instruction(regs, 4);
712 }
713
714 static const struct sys64_hook cp15_32_hooks[] = {
715 {
716 .esr_mask = ESR_ELx_CP15_32_ISS_SYS_MASK,
717 .esr_val = ESR_ELx_CP15_32_ISS_SYS_CNTFRQ,
718 .handler = compat_cntfrq_read_handler,
719 },
720 {},
721 };
722
compat_cntvct_read_handler(unsigned long esr,struct pt_regs * regs)723 static void compat_cntvct_read_handler(unsigned long esr, struct pt_regs *regs)
724 {
725 int rt = (esr & ESR_ELx_CP15_64_ISS_RT_MASK) >> ESR_ELx_CP15_64_ISS_RT_SHIFT;
726 int rt2 = (esr & ESR_ELx_CP15_64_ISS_RT2_MASK) >> ESR_ELx_CP15_64_ISS_RT2_SHIFT;
727 u64 val = arch_timer_read_counter();
728
729 pt_regs_write_reg(regs, rt, lower_32_bits(val));
730 pt_regs_write_reg(regs, rt2, upper_32_bits(val));
731 arm64_skip_faulting_instruction(regs, 4);
732 }
733
734 static const struct sys64_hook cp15_64_hooks[] = {
735 {
736 .esr_mask = ESR_ELx_CP15_64_ISS_SYS_MASK,
737 .esr_val = ESR_ELx_CP15_64_ISS_SYS_CNTVCT,
738 .handler = compat_cntvct_read_handler,
739 },
740 {
741 .esr_mask = ESR_ELx_CP15_64_ISS_SYS_MASK,
742 .esr_val = ESR_ELx_CP15_64_ISS_SYS_CNTVCTSS,
743 .handler = compat_cntvct_read_handler,
744 },
745 {},
746 };
747
do_el0_cp15(unsigned long esr,struct pt_regs * regs)748 void do_el0_cp15(unsigned long esr, struct pt_regs *regs)
749 {
750 const struct sys64_hook *hook, *hook_base;
751
752 if (!cp15_cond_valid(esr, regs)) {
753 /*
754 * There is no T16 variant of a CP access, so we
755 * always advance PC by 4 bytes.
756 */
757 arm64_skip_faulting_instruction(regs, 4);
758 return;
759 }
760
761 switch (ESR_ELx_EC(esr)) {
762 case ESR_ELx_EC_CP15_32:
763 hook_base = cp15_32_hooks;
764 break;
765 case ESR_ELx_EC_CP15_64:
766 hook_base = cp15_64_hooks;
767 break;
768 default:
769 do_el0_undef(regs, esr);
770 return;
771 }
772
773 for (hook = hook_base; hook->handler; hook++)
774 if ((hook->esr_mask & esr) == hook->esr_val) {
775 hook->handler(esr, regs);
776 return;
777 }
778
779 /*
780 * New cp15 instructions may previously have been undefined at
781 * EL0. Fall back to our usual undefined instruction handler
782 * so that we handle these consistently.
783 */
784 do_el0_undef(regs, esr);
785 }
786 #endif
787
do_el0_sys(unsigned long esr,struct pt_regs * regs)788 void do_el0_sys(unsigned long esr, struct pt_regs *regs)
789 {
790 const struct sys64_hook *hook;
791
792 for (hook = sys64_hooks; hook->handler; hook++)
793 if ((hook->esr_mask & esr) == hook->esr_val) {
794 hook->handler(esr, regs);
795 return;
796 }
797
798 /*
799 * New SYS instructions may previously have been undefined at EL0. Fall
800 * back to our usual undefined instruction handler so that we handle
801 * these consistently.
802 */
803 do_el0_undef(regs, esr);
804 }
805
806 static const char *esr_class_str[] = {
807 [0 ... ESR_ELx_EC_MAX] = "UNRECOGNIZED EC",
808 [ESR_ELx_EC_UNKNOWN] = "Unknown/Uncategorized",
809 [ESR_ELx_EC_WFx] = "WFI/WFE",
810 [ESR_ELx_EC_CP15_32] = "CP15 MCR/MRC",
811 [ESR_ELx_EC_CP15_64] = "CP15 MCRR/MRRC",
812 [ESR_ELx_EC_CP14_MR] = "CP14 MCR/MRC",
813 [ESR_ELx_EC_CP14_LS] = "CP14 LDC/STC",
814 [ESR_ELx_EC_FP_ASIMD] = "ASIMD",
815 [ESR_ELx_EC_CP10_ID] = "CP10 MRC/VMRS",
816 [ESR_ELx_EC_PAC] = "PAC",
817 [ESR_ELx_EC_CP14_64] = "CP14 MCRR/MRRC",
818 [ESR_ELx_EC_BTI] = "BTI",
819 [ESR_ELx_EC_ILL] = "PSTATE.IL",
820 [ESR_ELx_EC_SVC32] = "SVC (AArch32)",
821 [ESR_ELx_EC_HVC32] = "HVC (AArch32)",
822 [ESR_ELx_EC_SMC32] = "SMC (AArch32)",
823 [ESR_ELx_EC_SVC64] = "SVC (AArch64)",
824 [ESR_ELx_EC_HVC64] = "HVC (AArch64)",
825 [ESR_ELx_EC_SMC64] = "SMC (AArch64)",
826 [ESR_ELx_EC_SYS64] = "MSR/MRS (AArch64)",
827 [ESR_ELx_EC_SVE] = "SVE",
828 [ESR_ELx_EC_ERET] = "ERET/ERETAA/ERETAB",
829 [ESR_ELx_EC_FPAC] = "FPAC",
830 [ESR_ELx_EC_SME] = "SME",
831 [ESR_ELx_EC_IMP_DEF] = "EL3 IMP DEF",
832 [ESR_ELx_EC_IABT_LOW] = "IABT (lower EL)",
833 [ESR_ELx_EC_IABT_CUR] = "IABT (current EL)",
834 [ESR_ELx_EC_PC_ALIGN] = "PC Alignment",
835 [ESR_ELx_EC_DABT_LOW] = "DABT (lower EL)",
836 [ESR_ELx_EC_DABT_CUR] = "DABT (current EL)",
837 [ESR_ELx_EC_SP_ALIGN] = "SP Alignment",
838 [ESR_ELx_EC_MOPS] = "MOPS",
839 [ESR_ELx_EC_FP_EXC32] = "FP (AArch32)",
840 [ESR_ELx_EC_FP_EXC64] = "FP (AArch64)",
841 [ESR_ELx_EC_SERROR] = "SError",
842 [ESR_ELx_EC_BREAKPT_LOW] = "Breakpoint (lower EL)",
843 [ESR_ELx_EC_BREAKPT_CUR] = "Breakpoint (current EL)",
844 [ESR_ELx_EC_SOFTSTP_LOW] = "Software Step (lower EL)",
845 [ESR_ELx_EC_SOFTSTP_CUR] = "Software Step (current EL)",
846 [ESR_ELx_EC_WATCHPT_LOW] = "Watchpoint (lower EL)",
847 [ESR_ELx_EC_WATCHPT_CUR] = "Watchpoint (current EL)",
848 [ESR_ELx_EC_BKPT32] = "BKPT (AArch32)",
849 [ESR_ELx_EC_VECTOR32] = "Vector catch (AArch32)",
850 [ESR_ELx_EC_BRK64] = "BRK (AArch64)",
851 };
852
esr_get_class_string(unsigned long esr)853 const char *esr_get_class_string(unsigned long esr)
854 {
855 return esr_class_str[ESR_ELx_EC(esr)];
856 }
857
858 /*
859 * bad_el0_sync handles unexpected, but potentially recoverable synchronous
860 * exceptions taken from EL0.
861 */
bad_el0_sync(struct pt_regs * regs,int reason,unsigned long esr)862 void bad_el0_sync(struct pt_regs *regs, int reason, unsigned long esr)
863 {
864 unsigned long pc = instruction_pointer(regs);
865
866 current->thread.fault_address = 0;
867 current->thread.fault_code = esr;
868
869 arm64_force_sig_fault(SIGILL, ILL_ILLOPC, pc,
870 "Bad EL0 synchronous exception");
871 }
872
873 #ifdef CONFIG_VMAP_STACK
874
875 DEFINE_PER_CPU(unsigned long [OVERFLOW_STACK_SIZE/sizeof(long)], overflow_stack)
876 __aligned(16);
877
panic_bad_stack(struct pt_regs * regs,unsigned long esr,unsigned long far)878 void __noreturn panic_bad_stack(struct pt_regs *regs, unsigned long esr, unsigned long far)
879 {
880 unsigned long tsk_stk = (unsigned long)current->stack;
881 unsigned long irq_stk = (unsigned long)this_cpu_read(irq_stack_ptr);
882 unsigned long ovf_stk = (unsigned long)this_cpu_ptr(overflow_stack);
883
884 console_verbose();
885 pr_emerg("Insufficient stack space to handle exception!");
886
887 pr_emerg("ESR: 0x%016lx -- %s\n", esr, esr_get_class_string(esr));
888 pr_emerg("FAR: 0x%016lx\n", far);
889
890 pr_emerg("Task stack: [0x%016lx..0x%016lx]\n",
891 tsk_stk, tsk_stk + THREAD_SIZE);
892 pr_emerg("IRQ stack: [0x%016lx..0x%016lx]\n",
893 irq_stk, irq_stk + IRQ_STACK_SIZE);
894 pr_emerg("Overflow stack: [0x%016lx..0x%016lx]\n",
895 ovf_stk, ovf_stk + OVERFLOW_STACK_SIZE);
896
897 __show_regs(regs);
898
899 /*
900 * We use nmi_panic to limit the potential for recusive overflows, and
901 * to get a better stack trace.
902 */
903 nmi_panic(NULL, "kernel stack overflow");
904 cpu_park_loop();
905 }
906 #endif
907
arm64_serror_panic(struct pt_regs * regs,unsigned long esr)908 void __noreturn arm64_serror_panic(struct pt_regs *regs, unsigned long esr)
909 {
910 console_verbose();
911
912 pr_crit("SError Interrupt on CPU%d, code 0x%016lx -- %s\n",
913 smp_processor_id(), esr, esr_get_class_string(esr));
914 if (regs)
915 __show_regs(regs);
916
917 nmi_panic(regs, "Asynchronous SError Interrupt");
918
919 cpu_park_loop();
920 }
921
arm64_is_fatal_ras_serror(struct pt_regs * regs,unsigned long esr)922 bool arm64_is_fatal_ras_serror(struct pt_regs *regs, unsigned long esr)
923 {
924 unsigned long aet = arm64_ras_serror_get_severity(esr);
925
926 switch (aet) {
927 case ESR_ELx_AET_CE: /* corrected error */
928 case ESR_ELx_AET_UEO: /* restartable, not yet consumed */
929 /*
930 * The CPU can make progress. We may take UEO again as
931 * a more severe error.
932 */
933 return false;
934
935 case ESR_ELx_AET_UEU: /* Uncorrected Unrecoverable */
936 case ESR_ELx_AET_UER: /* Uncorrected Recoverable */
937 /*
938 * The CPU can't make progress. The exception may have
939 * been imprecise.
940 *
941 * Neoverse-N1 #1349291 means a non-KVM SError reported as
942 * Unrecoverable should be treated as Uncontainable. We
943 * call arm64_serror_panic() in both cases.
944 */
945 return true;
946
947 case ESR_ELx_AET_UC: /* Uncontainable or Uncategorized error */
948 default:
949 /* Error has been silently propagated */
950 arm64_serror_panic(regs, esr);
951 }
952 }
953
do_serror(struct pt_regs * regs,unsigned long esr)954 void do_serror(struct pt_regs *regs, unsigned long esr)
955 {
956 /* non-RAS errors are not containable */
957 if (!arm64_is_ras_serror(esr) || arm64_is_fatal_ras_serror(regs, esr))
958 arm64_serror_panic(regs, esr);
959 }
960
961 /* GENERIC_BUG traps */
962 #ifdef CONFIG_GENERIC_BUG
is_valid_bugaddr(unsigned long addr)963 int is_valid_bugaddr(unsigned long addr)
964 {
965 /*
966 * bug_handler() only called for BRK #BUG_BRK_IMM.
967 * So the answer is trivial -- any spurious instances with no
968 * bug table entry will be rejected by report_bug() and passed
969 * back to the debug-monitors code and handled as a fatal
970 * unexpected debug exception.
971 */
972 return 1;
973 }
974 #endif
975
bug_handler(struct pt_regs * regs,unsigned long esr)976 static int bug_handler(struct pt_regs *regs, unsigned long esr)
977 {
978 switch (report_bug(regs->pc, regs)) {
979 case BUG_TRAP_TYPE_BUG:
980 die("Oops - BUG", regs, esr);
981 break;
982
983 case BUG_TRAP_TYPE_WARN:
984 break;
985
986 default:
987 /* unknown/unrecognised bug trap type */
988 return DBG_HOOK_ERROR;
989 }
990
991 /* If thread survives, skip over the BUG instruction and continue: */
992 arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
993 return DBG_HOOK_HANDLED;
994 }
995
996 static struct break_hook bug_break_hook = {
997 .fn = bug_handler,
998 .imm = BUG_BRK_IMM,
999 };
1000
1001 #ifdef CONFIG_CFI_CLANG
cfi_handler(struct pt_regs * regs,unsigned long esr)1002 static int cfi_handler(struct pt_regs *regs, unsigned long esr)
1003 {
1004 unsigned long target;
1005 u32 type;
1006
1007 target = pt_regs_read_reg(regs, FIELD_GET(CFI_BRK_IMM_TARGET, esr));
1008 type = (u32)pt_regs_read_reg(regs, FIELD_GET(CFI_BRK_IMM_TYPE, esr));
1009
1010 switch (report_cfi_failure(regs, regs->pc, &target, type)) {
1011 case BUG_TRAP_TYPE_BUG:
1012 die("Oops - CFI", regs, esr);
1013 break;
1014
1015 case BUG_TRAP_TYPE_WARN:
1016 break;
1017
1018 default:
1019 return DBG_HOOK_ERROR;
1020 }
1021
1022 arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
1023 return DBG_HOOK_HANDLED;
1024 }
1025
1026 static struct break_hook cfi_break_hook = {
1027 .fn = cfi_handler,
1028 .imm = CFI_BRK_IMM_BASE,
1029 .mask = CFI_BRK_IMM_MASK,
1030 };
1031 #endif /* CONFIG_CFI_CLANG */
1032
reserved_fault_handler(struct pt_regs * regs,unsigned long esr)1033 static int reserved_fault_handler(struct pt_regs *regs, unsigned long esr)
1034 {
1035 pr_err("%s generated an invalid instruction at %pS!\n",
1036 "Kernel text patching",
1037 (void *)instruction_pointer(regs));
1038
1039 /* We cannot handle this */
1040 return DBG_HOOK_ERROR;
1041 }
1042
1043 static struct break_hook fault_break_hook = {
1044 .fn = reserved_fault_handler,
1045 .imm = FAULT_BRK_IMM,
1046 };
1047
1048 #ifdef CONFIG_KASAN_SW_TAGS
1049
1050 #define KASAN_ESR_RECOVER 0x20
1051 #define KASAN_ESR_WRITE 0x10
1052 #define KASAN_ESR_SIZE_MASK 0x0f
1053 #define KASAN_ESR_SIZE(esr) (1 << ((esr) & KASAN_ESR_SIZE_MASK))
1054
kasan_handler(struct pt_regs * regs,unsigned long esr)1055 static int kasan_handler(struct pt_regs *regs, unsigned long esr)
1056 {
1057 bool recover = esr & KASAN_ESR_RECOVER;
1058 bool write = esr & KASAN_ESR_WRITE;
1059 size_t size = KASAN_ESR_SIZE(esr);
1060 void *addr = (void *)regs->regs[0];
1061 u64 pc = regs->pc;
1062
1063 kasan_report(addr, size, write, pc);
1064
1065 /*
1066 * The instrumentation allows to control whether we can proceed after
1067 * a crash was detected. This is done by passing the -recover flag to
1068 * the compiler. Disabling recovery allows to generate more compact
1069 * code.
1070 *
1071 * Unfortunately disabling recovery doesn't work for the kernel right
1072 * now. KASAN reporting is disabled in some contexts (for example when
1073 * the allocator accesses slab object metadata; this is controlled by
1074 * current->kasan_depth). All these accesses are detected by the tool,
1075 * even though the reports for them are not printed.
1076 *
1077 * This is something that might be fixed at some point in the future.
1078 */
1079 if (!recover)
1080 die("Oops - KASAN", regs, esr);
1081
1082 /* If thread survives, skip over the brk instruction and continue: */
1083 arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
1084 return DBG_HOOK_HANDLED;
1085 }
1086
1087 static struct break_hook kasan_break_hook = {
1088 .fn = kasan_handler,
1089 .imm = KASAN_BRK_IMM,
1090 .mask = KASAN_BRK_MASK,
1091 };
1092 #endif
1093
1094 #ifdef CONFIG_UBSAN_TRAP
ubsan_handler(struct pt_regs * regs,unsigned long esr)1095 static int ubsan_handler(struct pt_regs *regs, unsigned long esr)
1096 {
1097 die(report_ubsan_failure(regs, esr & UBSAN_BRK_MASK), regs, esr);
1098 return DBG_HOOK_HANDLED;
1099 }
1100
1101 static struct break_hook ubsan_break_hook = {
1102 .fn = ubsan_handler,
1103 .imm = UBSAN_BRK_IMM,
1104 .mask = UBSAN_BRK_MASK,
1105 };
1106 #endif
1107
1108 #define esr_comment(esr) ((esr) & ESR_ELx_BRK64_ISS_COMMENT_MASK)
1109
1110 /*
1111 * Initial handler for AArch64 BRK exceptions
1112 * This handler only used until debug_traps_init().
1113 */
early_brk64(unsigned long addr,unsigned long esr,struct pt_regs * regs)1114 int __init early_brk64(unsigned long addr, unsigned long esr,
1115 struct pt_regs *regs)
1116 {
1117 #ifdef CONFIG_CFI_CLANG
1118 if ((esr_comment(esr) & ~CFI_BRK_IMM_MASK) == CFI_BRK_IMM_BASE)
1119 return cfi_handler(regs, esr) != DBG_HOOK_HANDLED;
1120 #endif
1121 #ifdef CONFIG_KASAN_SW_TAGS
1122 if ((esr_comment(esr) & ~KASAN_BRK_MASK) == KASAN_BRK_IMM)
1123 return kasan_handler(regs, esr) != DBG_HOOK_HANDLED;
1124 #endif
1125 #ifdef CONFIG_UBSAN_TRAP
1126 if ((esr_comment(esr) & ~UBSAN_BRK_MASK) == UBSAN_BRK_IMM)
1127 return ubsan_handler(regs, esr) != DBG_HOOK_HANDLED;
1128 #endif
1129 return bug_handler(regs, esr) != DBG_HOOK_HANDLED;
1130 }
1131
trap_init(void)1132 void __init trap_init(void)
1133 {
1134 register_kernel_break_hook(&bug_break_hook);
1135 #ifdef CONFIG_CFI_CLANG
1136 register_kernel_break_hook(&cfi_break_hook);
1137 #endif
1138 register_kernel_break_hook(&fault_break_hook);
1139 #ifdef CONFIG_KASAN_SW_TAGS
1140 register_kernel_break_hook(&kasan_break_hook);
1141 #endif
1142 #ifdef CONFIG_UBSAN_TRAP
1143 register_kernel_break_hook(&ubsan_break_hook);
1144 #endif
1145 debug_traps_init();
1146 }
1147