xref: /linux-6.15/arch/arm64/kernel/traps.c (revision dbfbd87e)
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/init.h>
22 #include <linux/sched/signal.h>
23 #include <linux/sched/debug.h>
24 #include <linux/sched/task_stack.h>
25 #include <linux/sizes.h>
26 #include <linux/syscalls.h>
27 #include <linux/mm_types.h>
28 #include <linux/kasan.h>
29 #include <linux/cfi.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/exception.h>
38 #include <asm/extable.h>
39 #include <asm/insn.h>
40 #include <asm/kprobes.h>
41 #include <asm/patching.h>
42 #include <asm/traps.h>
43 #include <asm/smp.h>
44 #include <asm/stack_pointer.h>
45 #include <asm/stacktrace.h>
46 #include <asm/system_misc.h>
47 #include <asm/sysreg.h>
48 
49 static bool __kprobes __check_eq(unsigned long pstate)
50 {
51 	return (pstate & PSR_Z_BIT) != 0;
52 }
53 
54 static bool __kprobes __check_ne(unsigned long pstate)
55 {
56 	return (pstate & PSR_Z_BIT) == 0;
57 }
58 
59 static bool __kprobes __check_cs(unsigned long pstate)
60 {
61 	return (pstate & PSR_C_BIT) != 0;
62 }
63 
64 static bool __kprobes __check_cc(unsigned long pstate)
65 {
66 	return (pstate & PSR_C_BIT) == 0;
67 }
68 
69 static bool __kprobes __check_mi(unsigned long pstate)
70 {
71 	return (pstate & PSR_N_BIT) != 0;
72 }
73 
74 static bool __kprobes __check_pl(unsigned long pstate)
75 {
76 	return (pstate & PSR_N_BIT) == 0;
77 }
78 
79 static bool __kprobes __check_vs(unsigned long pstate)
80 {
81 	return (pstate & PSR_V_BIT) != 0;
82 }
83 
84 static bool __kprobes __check_vc(unsigned long pstate)
85 {
86 	return (pstate & PSR_V_BIT) == 0;
87 }
88 
89 static bool __kprobes __check_hi(unsigned long pstate)
90 {
91 	pstate &= ~(pstate >> 1);	/* PSR_C_BIT &= ~PSR_Z_BIT */
92 	return (pstate & PSR_C_BIT) != 0;
93 }
94 
95 static bool __kprobes __check_ls(unsigned long pstate)
96 {
97 	pstate &= ~(pstate >> 1);	/* PSR_C_BIT &= ~PSR_Z_BIT */
98 	return (pstate & PSR_C_BIT) == 0;
99 }
100 
101 static bool __kprobes __check_ge(unsigned long pstate)
102 {
103 	pstate ^= (pstate << 3);	/* PSR_N_BIT ^= PSR_V_BIT */
104 	return (pstate & PSR_N_BIT) == 0;
105 }
106 
107 static bool __kprobes __check_lt(unsigned long pstate)
108 {
109 	pstate ^= (pstate << 3);	/* PSR_N_BIT ^= PSR_V_BIT */
110 	return (pstate & PSR_N_BIT) != 0;
111 }
112 
113 static bool __kprobes __check_gt(unsigned long pstate)
114 {
115 	/*PSR_N_BIT ^= PSR_V_BIT */
116 	unsigned long temp = pstate ^ (pstate << 3);
117 
118 	temp |= (pstate << 1);	/*PSR_N_BIT |= PSR_Z_BIT */
119 	return (temp & PSR_N_BIT) == 0;
120 }
121 
122 static bool __kprobes __check_le(unsigned long pstate)
123 {
124 	/*PSR_N_BIT ^= PSR_V_BIT */
125 	unsigned long temp = pstate ^ (pstate << 3);
126 
127 	temp |= (pstate << 1);	/*PSR_N_BIT |= PSR_Z_BIT */
128 	return (temp & PSR_N_BIT) != 0;
129 }
130 
131 static bool __kprobes __check_al(unsigned long pstate)
132 {
133 	return true;
134 }
135 
136 /*
137  * Note that the ARMv8 ARM calls condition code 0b1111 "nv", but states that
138  * it behaves identically to 0b1110 ("al").
139  */
140 pstate_check_t * const aarch32_opcode_cond_checks[16] = {
141 	__check_eq, __check_ne, __check_cs, __check_cc,
142 	__check_mi, __check_pl, __check_vs, __check_vc,
143 	__check_hi, __check_ls, __check_ge, __check_lt,
144 	__check_gt, __check_le, __check_al, __check_al
145 };
146 
147 int show_unhandled_signals = 0;
148 
149 static void dump_kernel_instr(const char *lvl, struct pt_regs *regs)
150 {
151 	unsigned long addr = instruction_pointer(regs);
152 	char str[sizeof("00000000 ") * 5 + 2 + 1], *p = str;
153 	int i;
154 
155 	if (user_mode(regs))
156 		return;
157 
158 	for (i = -4; i < 1; i++) {
159 		unsigned int val, bad;
160 
161 		bad = aarch64_insn_read(&((u32 *)addr)[i], &val);
162 
163 		if (!bad)
164 			p += sprintf(p, i == 0 ? "(%08x) " : "%08x ", val);
165 		else {
166 			p += sprintf(p, "bad PC value");
167 			break;
168 		}
169 	}
170 
171 	printk("%sCode: %s\n", lvl, str);
172 }
173 
174 #ifdef CONFIG_PREEMPT
175 #define S_PREEMPT " PREEMPT"
176 #elif defined(CONFIG_PREEMPT_RT)
177 #define S_PREEMPT " PREEMPT_RT"
178 #else
179 #define S_PREEMPT ""
180 #endif
181 
182 #define S_SMP " SMP"
183 
184 static int __die(const char *str, long err, struct pt_regs *regs)
185 {
186 	static int die_counter;
187 	int ret;
188 
189 	pr_emerg("Internal error: %s: %016lx [#%d]" S_PREEMPT S_SMP "\n",
190 		 str, err, ++die_counter);
191 
192 	/* trap and error numbers are mostly meaningless on ARM */
193 	ret = notify_die(DIE_OOPS, str, regs, err, 0, SIGSEGV);
194 	if (ret == NOTIFY_STOP)
195 		return ret;
196 
197 	print_modules();
198 	show_regs(regs);
199 
200 	dump_kernel_instr(KERN_EMERG, regs);
201 
202 	return ret;
203 }
204 
205 static DEFINE_RAW_SPINLOCK(die_lock);
206 
207 /*
208  * This function is protected against re-entrancy.
209  */
210 void die(const char *str, struct pt_regs *regs, long err)
211 {
212 	int ret;
213 	unsigned long flags;
214 
215 	raw_spin_lock_irqsave(&die_lock, flags);
216 
217 	oops_enter();
218 
219 	console_verbose();
220 	bust_spinlocks(1);
221 	ret = __die(str, err, regs);
222 
223 	if (regs && kexec_should_crash(current))
224 		crash_kexec(regs);
225 
226 	bust_spinlocks(0);
227 	add_taint(TAINT_DIE, LOCKDEP_NOW_UNRELIABLE);
228 	oops_exit();
229 
230 	if (in_interrupt())
231 		panic("%s: Fatal exception in interrupt", str);
232 	if (panic_on_oops)
233 		panic("%s: Fatal exception", str);
234 
235 	raw_spin_unlock_irqrestore(&die_lock, flags);
236 
237 	if (ret != NOTIFY_STOP)
238 		make_task_dead(SIGSEGV);
239 }
240 
241 static void arm64_show_signal(int signo, const char *str)
242 {
243 	static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,
244 				      DEFAULT_RATELIMIT_BURST);
245 	struct task_struct *tsk = current;
246 	unsigned long esr = tsk->thread.fault_code;
247 	struct pt_regs *regs = task_pt_regs(tsk);
248 
249 	/* Leave if the signal won't be shown */
250 	if (!show_unhandled_signals ||
251 	    !unhandled_signal(tsk, signo) ||
252 	    !__ratelimit(&rs))
253 		return;
254 
255 	pr_info("%s[%d]: unhandled exception: ", tsk->comm, task_pid_nr(tsk));
256 	if (esr)
257 		pr_cont("%s, ESR 0x%016lx, ", esr_get_class_string(esr), esr);
258 
259 	pr_cont("%s", str);
260 	print_vma_addr(KERN_CONT " in ", regs->pc);
261 	pr_cont("\n");
262 	__show_regs(regs);
263 }
264 
265 void arm64_force_sig_fault(int signo, int code, unsigned long far,
266 			   const char *str)
267 {
268 	arm64_show_signal(signo, str);
269 	if (signo == SIGKILL)
270 		force_sig(SIGKILL);
271 	else
272 		force_sig_fault(signo, code, (void __user *)far);
273 }
274 
275 void arm64_force_sig_mceerr(int code, unsigned long far, short lsb,
276 			    const char *str)
277 {
278 	arm64_show_signal(SIGBUS, str);
279 	force_sig_mceerr(code, (void __user *)far, lsb);
280 }
281 
282 void arm64_force_sig_ptrace_errno_trap(int errno, unsigned long far,
283 				       const char *str)
284 {
285 	arm64_show_signal(SIGTRAP, str);
286 	force_sig_ptrace_errno_trap(errno, (void __user *)far);
287 }
288 
289 void arm64_notify_die(const char *str, struct pt_regs *regs,
290 		      int signo, int sicode, unsigned long far,
291 		      unsigned long err)
292 {
293 	if (user_mode(regs)) {
294 		WARN_ON(regs != current_pt_regs());
295 		current->thread.fault_address = 0;
296 		current->thread.fault_code = err;
297 
298 		arm64_force_sig_fault(signo, sicode, far, str);
299 	} else {
300 		die(str, regs, err);
301 	}
302 }
303 
304 #ifdef CONFIG_COMPAT
305 #define PSTATE_IT_1_0_SHIFT	25
306 #define PSTATE_IT_1_0_MASK	(0x3 << PSTATE_IT_1_0_SHIFT)
307 #define PSTATE_IT_7_2_SHIFT	10
308 #define PSTATE_IT_7_2_MASK	(0x3f << PSTATE_IT_7_2_SHIFT)
309 
310 static u32 compat_get_it_state(struct pt_regs *regs)
311 {
312 	u32 it, pstate = regs->pstate;
313 
314 	it  = (pstate & PSTATE_IT_1_0_MASK) >> PSTATE_IT_1_0_SHIFT;
315 	it |= ((pstate & PSTATE_IT_7_2_MASK) >> PSTATE_IT_7_2_SHIFT) << 2;
316 
317 	return it;
318 }
319 
320 static void compat_set_it_state(struct pt_regs *regs, u32 it)
321 {
322 	u32 pstate_it;
323 
324 	pstate_it  = (it << PSTATE_IT_1_0_SHIFT) & PSTATE_IT_1_0_MASK;
325 	pstate_it |= ((it >> 2) << PSTATE_IT_7_2_SHIFT) & PSTATE_IT_7_2_MASK;
326 
327 	regs->pstate &= ~PSR_AA32_IT_MASK;
328 	regs->pstate |= pstate_it;
329 }
330 
331 static void advance_itstate(struct pt_regs *regs)
332 {
333 	u32 it;
334 
335 	/* ARM mode */
336 	if (!(regs->pstate & PSR_AA32_T_BIT) ||
337 	    !(regs->pstate & PSR_AA32_IT_MASK))
338 		return;
339 
340 	it  = compat_get_it_state(regs);
341 
342 	/*
343 	 * If this is the last instruction of the block, wipe the IT
344 	 * state. Otherwise advance it.
345 	 */
346 	if (!(it & 7))
347 		it = 0;
348 	else
349 		it = (it & 0xe0) | ((it << 1) & 0x1f);
350 
351 	compat_set_it_state(regs, it);
352 }
353 #else
354 static void advance_itstate(struct pt_regs *regs)
355 {
356 }
357 #endif
358 
359 void arm64_skip_faulting_instruction(struct pt_regs *regs, unsigned long size)
360 {
361 	regs->pc += size;
362 
363 	/*
364 	 * If we were single stepping, we want to get the step exception after
365 	 * we return from the trap.
366 	 */
367 	if (user_mode(regs))
368 		user_fastforward_single_step(current);
369 
370 	if (compat_user_mode(regs))
371 		advance_itstate(regs);
372 	else
373 		regs->pstate &= ~PSR_BTYPE_MASK;
374 }
375 
376 static LIST_HEAD(undef_hook);
377 static DEFINE_RAW_SPINLOCK(undef_lock);
378 
379 void register_undef_hook(struct undef_hook *hook)
380 {
381 	unsigned long flags;
382 
383 	raw_spin_lock_irqsave(&undef_lock, flags);
384 	list_add(&hook->node, &undef_hook);
385 	raw_spin_unlock_irqrestore(&undef_lock, flags);
386 }
387 
388 void unregister_undef_hook(struct undef_hook *hook)
389 {
390 	unsigned long flags;
391 
392 	raw_spin_lock_irqsave(&undef_lock, flags);
393 	list_del(&hook->node);
394 	raw_spin_unlock_irqrestore(&undef_lock, flags);
395 }
396 
397 static int user_insn_read(struct pt_regs *regs, u32 *insnp)
398 {
399 	u32 instr;
400 	unsigned long pc = instruction_pointer(regs);
401 
402 	if (compat_thumb_mode(regs)) {
403 		/* 16-bit Thumb instruction */
404 		__le16 instr_le;
405 		if (get_user(instr_le, (__le16 __user *)pc))
406 			return -EFAULT;
407 		instr = le16_to_cpu(instr_le);
408 		if (aarch32_insn_is_wide(instr)) {
409 			u32 instr2;
410 
411 			if (get_user(instr_le, (__le16 __user *)(pc + 2)))
412 				return -EFAULT;
413 			instr2 = le16_to_cpu(instr_le);
414 			instr = (instr << 16) | instr2;
415 		}
416 	} else {
417 		/* 32-bit ARM instruction */
418 		__le32 instr_le;
419 		if (get_user(instr_le, (__le32 __user *)pc))
420 			return -EFAULT;
421 		instr = le32_to_cpu(instr_le);
422 	}
423 
424 	*insnp = instr;
425 	return 0;
426 }
427 
428 static int call_undef_hook(struct pt_regs *regs, u32 instr)
429 {
430 	struct undef_hook *hook;
431 	unsigned long flags;
432 	int (*fn)(struct pt_regs *regs, u32 instr) = NULL;
433 
434 	raw_spin_lock_irqsave(&undef_lock, flags);
435 	list_for_each_entry(hook, &undef_hook, node)
436 		if ((instr & hook->instr_mask) == hook->instr_val &&
437 			(regs->pstate & hook->pstate_mask) == hook->pstate_val)
438 			fn = hook->fn;
439 
440 	raw_spin_unlock_irqrestore(&undef_lock, flags);
441 
442 	return fn ? fn(regs, instr) : 1;
443 }
444 
445 void force_signal_inject(int signal, int code, unsigned long address, unsigned long err)
446 {
447 	const char *desc;
448 	struct pt_regs *regs = current_pt_regs();
449 
450 	if (WARN_ON(!user_mode(regs)))
451 		return;
452 
453 	switch (signal) {
454 	case SIGILL:
455 		desc = "undefined instruction";
456 		break;
457 	case SIGSEGV:
458 		desc = "illegal memory access";
459 		break;
460 	default:
461 		desc = "unknown or unrecoverable error";
462 		break;
463 	}
464 
465 	/* Force signals we don't understand to SIGKILL */
466 	if (WARN_ON(signal != SIGKILL &&
467 		    siginfo_layout(signal, code) != SIL_FAULT)) {
468 		signal = SIGKILL;
469 	}
470 
471 	arm64_notify_die(desc, regs, signal, code, address, err);
472 }
473 
474 /*
475  * Set up process info to signal segmentation fault - called on access error.
476  */
477 void arm64_notify_segfault(unsigned long addr)
478 {
479 	int code;
480 
481 	mmap_read_lock(current->mm);
482 	if (find_vma(current->mm, untagged_addr(addr)) == NULL)
483 		code = SEGV_MAPERR;
484 	else
485 		code = SEGV_ACCERR;
486 	mmap_read_unlock(current->mm);
487 
488 	force_signal_inject(SIGSEGV, code, addr, 0);
489 }
490 
491 void do_el0_undef(struct pt_regs *regs, unsigned long esr)
492 {
493 	u32 insn;
494 
495 	/* check for AArch32 breakpoint instructions */
496 	if (!aarch32_break_handler(regs))
497 		return;
498 
499 	if (user_insn_read(regs, &insn))
500 		goto out_err;
501 
502 	if (call_undef_hook(regs, insn) == 0)
503 		return;
504 
505 out_err:
506 	force_signal_inject(SIGILL, ILL_ILLOPC, regs->pc, 0);
507 }
508 
509 void do_el1_undef(struct pt_regs *regs, unsigned long esr)
510 {
511 	u32 insn;
512 
513 	if (aarch64_insn_read((void *)regs->pc, &insn))
514 		goto out_err;
515 
516 	if (try_emulate_el1_ssbs(regs, insn))
517 		return;
518 
519 out_err:
520 	die("Oops - Undefined instruction", regs, esr);
521 }
522 
523 void do_el0_bti(struct pt_regs *regs)
524 {
525 	force_signal_inject(SIGILL, ILL_ILLOPC, regs->pc, 0);
526 }
527 
528 void do_el1_bti(struct pt_regs *regs, unsigned long esr)
529 {
530 	die("Oops - BTI", regs, esr);
531 }
532 
533 void do_el0_fpac(struct pt_regs *regs, unsigned long esr)
534 {
535 	force_signal_inject(SIGILL, ILL_ILLOPN, regs->pc, esr);
536 }
537 
538 void do_el1_fpac(struct pt_regs *regs, unsigned long esr)
539 {
540 	/*
541 	 * Unexpected FPAC exception in the kernel: kill the task before it
542 	 * does any more harm.
543 	 */
544 	die("Oops - FPAC", regs, esr);
545 }
546 
547 #define __user_cache_maint(insn, address, res)			\
548 	if (address >= TASK_SIZE_MAX) {				\
549 		res = -EFAULT;					\
550 	} else {						\
551 		uaccess_ttbr0_enable();				\
552 		asm volatile (					\
553 			"1:	" insn ", %1\n"			\
554 			"	mov	%w0, #0\n"		\
555 			"2:\n"					\
556 			_ASM_EXTABLE_UACCESS_ERR(1b, 2b, %w0)	\
557 			: "=r" (res)				\
558 			: "r" (address));			\
559 		uaccess_ttbr0_disable();			\
560 	}
561 
562 static void user_cache_maint_handler(unsigned long esr, struct pt_regs *regs)
563 {
564 	unsigned long tagged_address, address;
565 	int rt = ESR_ELx_SYS64_ISS_RT(esr);
566 	int crm = (esr & ESR_ELx_SYS64_ISS_CRM_MASK) >> ESR_ELx_SYS64_ISS_CRM_SHIFT;
567 	int ret = 0;
568 
569 	tagged_address = pt_regs_read_reg(regs, rt);
570 	address = untagged_addr(tagged_address);
571 
572 	switch (crm) {
573 	case ESR_ELx_SYS64_ISS_CRM_DC_CVAU:	/* DC CVAU, gets promoted */
574 		__user_cache_maint("dc civac", address, ret);
575 		break;
576 	case ESR_ELx_SYS64_ISS_CRM_DC_CVAC:	/* DC CVAC, gets promoted */
577 		__user_cache_maint("dc civac", address, ret);
578 		break;
579 	case ESR_ELx_SYS64_ISS_CRM_DC_CVADP:	/* DC CVADP */
580 		__user_cache_maint("sys 3, c7, c13, 1", address, ret);
581 		break;
582 	case ESR_ELx_SYS64_ISS_CRM_DC_CVAP:	/* DC CVAP */
583 		__user_cache_maint("sys 3, c7, c12, 1", address, ret);
584 		break;
585 	case ESR_ELx_SYS64_ISS_CRM_DC_CIVAC:	/* DC CIVAC */
586 		__user_cache_maint("dc civac", address, ret);
587 		break;
588 	case ESR_ELx_SYS64_ISS_CRM_IC_IVAU:	/* IC IVAU */
589 		__user_cache_maint("ic ivau", address, ret);
590 		break;
591 	default:
592 		force_signal_inject(SIGILL, ILL_ILLOPC, regs->pc, 0);
593 		return;
594 	}
595 
596 	if (ret)
597 		arm64_notify_segfault(tagged_address);
598 	else
599 		arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
600 }
601 
602 static void ctr_read_handler(unsigned long esr, struct pt_regs *regs)
603 {
604 	int rt = ESR_ELx_SYS64_ISS_RT(esr);
605 	unsigned long val = arm64_ftr_reg_user_value(&arm64_ftr_reg_ctrel0);
606 
607 	if (cpus_have_const_cap(ARM64_WORKAROUND_1542419)) {
608 		/* Hide DIC so that we can trap the unnecessary maintenance...*/
609 		val &= ~BIT(CTR_EL0_DIC_SHIFT);
610 
611 		/* ... and fake IminLine to reduce the number of traps. */
612 		val &= ~CTR_EL0_IminLine_MASK;
613 		val |= (PAGE_SHIFT - 2) & CTR_EL0_IminLine_MASK;
614 	}
615 
616 	pt_regs_write_reg(regs, rt, val);
617 
618 	arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
619 }
620 
621 static void cntvct_read_handler(unsigned long esr, struct pt_regs *regs)
622 {
623 	int rt = ESR_ELx_SYS64_ISS_RT(esr);
624 
625 	pt_regs_write_reg(regs, rt, arch_timer_read_counter());
626 	arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
627 }
628 
629 static void cntfrq_read_handler(unsigned long esr, struct pt_regs *regs)
630 {
631 	int rt = ESR_ELx_SYS64_ISS_RT(esr);
632 
633 	pt_regs_write_reg(regs, rt, arch_timer_get_rate());
634 	arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
635 }
636 
637 static void mrs_handler(unsigned long esr, struct pt_regs *regs)
638 {
639 	u32 sysreg, rt;
640 
641 	rt = ESR_ELx_SYS64_ISS_RT(esr);
642 	sysreg = esr_sys64_to_sysreg(esr);
643 
644 	if (do_emulate_mrs(regs, sysreg, rt) != 0)
645 		force_signal_inject(SIGILL, ILL_ILLOPC, regs->pc, 0);
646 }
647 
648 static void wfi_handler(unsigned long esr, struct pt_regs *regs)
649 {
650 	arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
651 }
652 
653 struct sys64_hook {
654 	unsigned long esr_mask;
655 	unsigned long esr_val;
656 	void (*handler)(unsigned long esr, struct pt_regs *regs);
657 };
658 
659 static const struct sys64_hook sys64_hooks[] = {
660 	{
661 		.esr_mask = ESR_ELx_SYS64_ISS_EL0_CACHE_OP_MASK,
662 		.esr_val = ESR_ELx_SYS64_ISS_EL0_CACHE_OP_VAL,
663 		.handler = user_cache_maint_handler,
664 	},
665 	{
666 		/* Trap read access to CTR_EL0 */
667 		.esr_mask = ESR_ELx_SYS64_ISS_SYS_OP_MASK,
668 		.esr_val = ESR_ELx_SYS64_ISS_SYS_CTR_READ,
669 		.handler = ctr_read_handler,
670 	},
671 	{
672 		/* Trap read access to CNTVCT_EL0 */
673 		.esr_mask = ESR_ELx_SYS64_ISS_SYS_OP_MASK,
674 		.esr_val = ESR_ELx_SYS64_ISS_SYS_CNTVCT,
675 		.handler = cntvct_read_handler,
676 	},
677 	{
678 		/* Trap read access to CNTVCTSS_EL0 */
679 		.esr_mask = ESR_ELx_SYS64_ISS_SYS_OP_MASK,
680 		.esr_val = ESR_ELx_SYS64_ISS_SYS_CNTVCTSS,
681 		.handler = cntvct_read_handler,
682 	},
683 	{
684 		/* Trap read access to CNTFRQ_EL0 */
685 		.esr_mask = ESR_ELx_SYS64_ISS_SYS_OP_MASK,
686 		.esr_val = ESR_ELx_SYS64_ISS_SYS_CNTFRQ,
687 		.handler = cntfrq_read_handler,
688 	},
689 	{
690 		/* Trap read access to CPUID registers */
691 		.esr_mask = ESR_ELx_SYS64_ISS_SYS_MRS_OP_MASK,
692 		.esr_val = ESR_ELx_SYS64_ISS_SYS_MRS_OP_VAL,
693 		.handler = mrs_handler,
694 	},
695 	{
696 		/* Trap WFI instructions executed in userspace */
697 		.esr_mask = ESR_ELx_WFx_MASK,
698 		.esr_val = ESR_ELx_WFx_WFI_VAL,
699 		.handler = wfi_handler,
700 	},
701 	{},
702 };
703 
704 #ifdef CONFIG_COMPAT
705 static bool cp15_cond_valid(unsigned long esr, struct pt_regs *regs)
706 {
707 	int cond;
708 
709 	/* Only a T32 instruction can trap without CV being set */
710 	if (!(esr & ESR_ELx_CV)) {
711 		u32 it;
712 
713 		it = compat_get_it_state(regs);
714 		if (!it)
715 			return true;
716 
717 		cond = it >> 4;
718 	} else {
719 		cond = (esr & ESR_ELx_COND_MASK) >> ESR_ELx_COND_SHIFT;
720 	}
721 
722 	return aarch32_opcode_cond_checks[cond](regs->pstate);
723 }
724 
725 static void compat_cntfrq_read_handler(unsigned long esr, struct pt_regs *regs)
726 {
727 	int reg = (esr & ESR_ELx_CP15_32_ISS_RT_MASK) >> ESR_ELx_CP15_32_ISS_RT_SHIFT;
728 
729 	pt_regs_write_reg(regs, reg, arch_timer_get_rate());
730 	arm64_skip_faulting_instruction(regs, 4);
731 }
732 
733 static const struct sys64_hook cp15_32_hooks[] = {
734 	{
735 		.esr_mask = ESR_ELx_CP15_32_ISS_SYS_MASK,
736 		.esr_val = ESR_ELx_CP15_32_ISS_SYS_CNTFRQ,
737 		.handler = compat_cntfrq_read_handler,
738 	},
739 	{},
740 };
741 
742 static void compat_cntvct_read_handler(unsigned long esr, struct pt_regs *regs)
743 {
744 	int rt = (esr & ESR_ELx_CP15_64_ISS_RT_MASK) >> ESR_ELx_CP15_64_ISS_RT_SHIFT;
745 	int rt2 = (esr & ESR_ELx_CP15_64_ISS_RT2_MASK) >> ESR_ELx_CP15_64_ISS_RT2_SHIFT;
746 	u64 val = arch_timer_read_counter();
747 
748 	pt_regs_write_reg(regs, rt, lower_32_bits(val));
749 	pt_regs_write_reg(regs, rt2, upper_32_bits(val));
750 	arm64_skip_faulting_instruction(regs, 4);
751 }
752 
753 static const struct sys64_hook cp15_64_hooks[] = {
754 	{
755 		.esr_mask = ESR_ELx_CP15_64_ISS_SYS_MASK,
756 		.esr_val = ESR_ELx_CP15_64_ISS_SYS_CNTVCT,
757 		.handler = compat_cntvct_read_handler,
758 	},
759 	{
760 		.esr_mask = ESR_ELx_CP15_64_ISS_SYS_MASK,
761 		.esr_val = ESR_ELx_CP15_64_ISS_SYS_CNTVCTSS,
762 		.handler = compat_cntvct_read_handler,
763 	},
764 	{},
765 };
766 
767 void do_el0_cp15(unsigned long esr, struct pt_regs *regs)
768 {
769 	const struct sys64_hook *hook, *hook_base;
770 
771 	if (!cp15_cond_valid(esr, regs)) {
772 		/*
773 		 * There is no T16 variant of a CP access, so we
774 		 * always advance PC by 4 bytes.
775 		 */
776 		arm64_skip_faulting_instruction(regs, 4);
777 		return;
778 	}
779 
780 	switch (ESR_ELx_EC(esr)) {
781 	case ESR_ELx_EC_CP15_32:
782 		hook_base = cp15_32_hooks;
783 		break;
784 	case ESR_ELx_EC_CP15_64:
785 		hook_base = cp15_64_hooks;
786 		break;
787 	default:
788 		do_el0_undef(regs, esr);
789 		return;
790 	}
791 
792 	for (hook = hook_base; hook->handler; hook++)
793 		if ((hook->esr_mask & esr) == hook->esr_val) {
794 			hook->handler(esr, regs);
795 			return;
796 		}
797 
798 	/*
799 	 * New cp15 instructions may previously have been undefined at
800 	 * EL0. Fall back to our usual undefined instruction handler
801 	 * so that we handle these consistently.
802 	 */
803 	do_el0_undef(regs, esr);
804 }
805 #endif
806 
807 void do_el0_sys(unsigned long esr, struct pt_regs *regs)
808 {
809 	const struct sys64_hook *hook;
810 
811 	for (hook = sys64_hooks; hook->handler; hook++)
812 		if ((hook->esr_mask & esr) == hook->esr_val) {
813 			hook->handler(esr, regs);
814 			return;
815 		}
816 
817 	/*
818 	 * New SYS instructions may previously have been undefined at EL0. Fall
819 	 * back to our usual undefined instruction handler so that we handle
820 	 * these consistently.
821 	 */
822 	do_el0_undef(regs, esr);
823 }
824 
825 static const char *esr_class_str[] = {
826 	[0 ... ESR_ELx_EC_MAX]		= "UNRECOGNIZED EC",
827 	[ESR_ELx_EC_UNKNOWN]		= "Unknown/Uncategorized",
828 	[ESR_ELx_EC_WFx]		= "WFI/WFE",
829 	[ESR_ELx_EC_CP15_32]		= "CP15 MCR/MRC",
830 	[ESR_ELx_EC_CP15_64]		= "CP15 MCRR/MRRC",
831 	[ESR_ELx_EC_CP14_MR]		= "CP14 MCR/MRC",
832 	[ESR_ELx_EC_CP14_LS]		= "CP14 LDC/STC",
833 	[ESR_ELx_EC_FP_ASIMD]		= "ASIMD",
834 	[ESR_ELx_EC_CP10_ID]		= "CP10 MRC/VMRS",
835 	[ESR_ELx_EC_PAC]		= "PAC",
836 	[ESR_ELx_EC_CP14_64]		= "CP14 MCRR/MRRC",
837 	[ESR_ELx_EC_BTI]		= "BTI",
838 	[ESR_ELx_EC_ILL]		= "PSTATE.IL",
839 	[ESR_ELx_EC_SVC32]		= "SVC (AArch32)",
840 	[ESR_ELx_EC_HVC32]		= "HVC (AArch32)",
841 	[ESR_ELx_EC_SMC32]		= "SMC (AArch32)",
842 	[ESR_ELx_EC_SVC64]		= "SVC (AArch64)",
843 	[ESR_ELx_EC_HVC64]		= "HVC (AArch64)",
844 	[ESR_ELx_EC_SMC64]		= "SMC (AArch64)",
845 	[ESR_ELx_EC_SYS64]		= "MSR/MRS (AArch64)",
846 	[ESR_ELx_EC_SVE]		= "SVE",
847 	[ESR_ELx_EC_ERET]		= "ERET/ERETAA/ERETAB",
848 	[ESR_ELx_EC_FPAC]		= "FPAC",
849 	[ESR_ELx_EC_SME]		= "SME",
850 	[ESR_ELx_EC_IMP_DEF]		= "EL3 IMP DEF",
851 	[ESR_ELx_EC_IABT_LOW]		= "IABT (lower EL)",
852 	[ESR_ELx_EC_IABT_CUR]		= "IABT (current EL)",
853 	[ESR_ELx_EC_PC_ALIGN]		= "PC Alignment",
854 	[ESR_ELx_EC_DABT_LOW]		= "DABT (lower EL)",
855 	[ESR_ELx_EC_DABT_CUR]		= "DABT (current EL)",
856 	[ESR_ELx_EC_SP_ALIGN]		= "SP Alignment",
857 	[ESR_ELx_EC_FP_EXC32]		= "FP (AArch32)",
858 	[ESR_ELx_EC_FP_EXC64]		= "FP (AArch64)",
859 	[ESR_ELx_EC_SERROR]		= "SError",
860 	[ESR_ELx_EC_BREAKPT_LOW]	= "Breakpoint (lower EL)",
861 	[ESR_ELx_EC_BREAKPT_CUR]	= "Breakpoint (current EL)",
862 	[ESR_ELx_EC_SOFTSTP_LOW]	= "Software Step (lower EL)",
863 	[ESR_ELx_EC_SOFTSTP_CUR]	= "Software Step (current EL)",
864 	[ESR_ELx_EC_WATCHPT_LOW]	= "Watchpoint (lower EL)",
865 	[ESR_ELx_EC_WATCHPT_CUR]	= "Watchpoint (current EL)",
866 	[ESR_ELx_EC_BKPT32]		= "BKPT (AArch32)",
867 	[ESR_ELx_EC_VECTOR32]		= "Vector catch (AArch32)",
868 	[ESR_ELx_EC_BRK64]		= "BRK (AArch64)",
869 };
870 
871 const char *esr_get_class_string(unsigned long esr)
872 {
873 	return esr_class_str[ESR_ELx_EC(esr)];
874 }
875 
876 /*
877  * bad_el0_sync handles unexpected, but potentially recoverable synchronous
878  * exceptions taken from EL0.
879  */
880 void bad_el0_sync(struct pt_regs *regs, int reason, unsigned long esr)
881 {
882 	unsigned long pc = instruction_pointer(regs);
883 
884 	current->thread.fault_address = 0;
885 	current->thread.fault_code = esr;
886 
887 	arm64_force_sig_fault(SIGILL, ILL_ILLOPC, pc,
888 			      "Bad EL0 synchronous exception");
889 }
890 
891 #ifdef CONFIG_VMAP_STACK
892 
893 DEFINE_PER_CPU(unsigned long [OVERFLOW_STACK_SIZE/sizeof(long)], overflow_stack)
894 	__aligned(16);
895 
896 void panic_bad_stack(struct pt_regs *regs, unsigned long esr, unsigned long far)
897 {
898 	unsigned long tsk_stk = (unsigned long)current->stack;
899 	unsigned long irq_stk = (unsigned long)this_cpu_read(irq_stack_ptr);
900 	unsigned long ovf_stk = (unsigned long)this_cpu_ptr(overflow_stack);
901 
902 	console_verbose();
903 	pr_emerg("Insufficient stack space to handle exception!");
904 
905 	pr_emerg("ESR: 0x%016lx -- %s\n", esr, esr_get_class_string(esr));
906 	pr_emerg("FAR: 0x%016lx\n", far);
907 
908 	pr_emerg("Task stack:     [0x%016lx..0x%016lx]\n",
909 		 tsk_stk, tsk_stk + THREAD_SIZE);
910 	pr_emerg("IRQ stack:      [0x%016lx..0x%016lx]\n",
911 		 irq_stk, irq_stk + IRQ_STACK_SIZE);
912 	pr_emerg("Overflow stack: [0x%016lx..0x%016lx]\n",
913 		 ovf_stk, ovf_stk + OVERFLOW_STACK_SIZE);
914 
915 	__show_regs(regs);
916 
917 	/*
918 	 * We use nmi_panic to limit the potential for recusive overflows, and
919 	 * to get a better stack trace.
920 	 */
921 	nmi_panic(NULL, "kernel stack overflow");
922 	cpu_park_loop();
923 }
924 #endif
925 
926 void __noreturn arm64_serror_panic(struct pt_regs *regs, unsigned long esr)
927 {
928 	console_verbose();
929 
930 	pr_crit("SError Interrupt on CPU%d, code 0x%016lx -- %s\n",
931 		smp_processor_id(), esr, esr_get_class_string(esr));
932 	if (regs)
933 		__show_regs(regs);
934 
935 	nmi_panic(regs, "Asynchronous SError Interrupt");
936 
937 	cpu_park_loop();
938 	unreachable();
939 }
940 
941 bool arm64_is_fatal_ras_serror(struct pt_regs *regs, unsigned long esr)
942 {
943 	unsigned long aet = arm64_ras_serror_get_severity(esr);
944 
945 	switch (aet) {
946 	case ESR_ELx_AET_CE:	/* corrected error */
947 	case ESR_ELx_AET_UEO:	/* restartable, not yet consumed */
948 		/*
949 		 * The CPU can make progress. We may take UEO again as
950 		 * a more severe error.
951 		 */
952 		return false;
953 
954 	case ESR_ELx_AET_UEU:	/* Uncorrected Unrecoverable */
955 	case ESR_ELx_AET_UER:	/* Uncorrected Recoverable */
956 		/*
957 		 * The CPU can't make progress. The exception may have
958 		 * been imprecise.
959 		 *
960 		 * Neoverse-N1 #1349291 means a non-KVM SError reported as
961 		 * Unrecoverable should be treated as Uncontainable. We
962 		 * call arm64_serror_panic() in both cases.
963 		 */
964 		return true;
965 
966 	case ESR_ELx_AET_UC:	/* Uncontainable or Uncategorized error */
967 	default:
968 		/* Error has been silently propagated */
969 		arm64_serror_panic(regs, esr);
970 	}
971 }
972 
973 void do_serror(struct pt_regs *regs, unsigned long esr)
974 {
975 	/* non-RAS errors are not containable */
976 	if (!arm64_is_ras_serror(esr) || arm64_is_fatal_ras_serror(regs, esr))
977 		arm64_serror_panic(regs, esr);
978 }
979 
980 /* GENERIC_BUG traps */
981 
982 int is_valid_bugaddr(unsigned long addr)
983 {
984 	/*
985 	 * bug_handler() only called for BRK #BUG_BRK_IMM.
986 	 * So the answer is trivial -- any spurious instances with no
987 	 * bug table entry will be rejected by report_bug() and passed
988 	 * back to the debug-monitors code and handled as a fatal
989 	 * unexpected debug exception.
990 	 */
991 	return 1;
992 }
993 
994 static int bug_handler(struct pt_regs *regs, unsigned long esr)
995 {
996 	switch (report_bug(regs->pc, regs)) {
997 	case BUG_TRAP_TYPE_BUG:
998 		die("Oops - BUG", regs, esr);
999 		break;
1000 
1001 	case BUG_TRAP_TYPE_WARN:
1002 		break;
1003 
1004 	default:
1005 		/* unknown/unrecognised bug trap type */
1006 		return DBG_HOOK_ERROR;
1007 	}
1008 
1009 	/* If thread survives, skip over the BUG instruction and continue: */
1010 	arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
1011 	return DBG_HOOK_HANDLED;
1012 }
1013 
1014 static struct break_hook bug_break_hook = {
1015 	.fn = bug_handler,
1016 	.imm = BUG_BRK_IMM,
1017 };
1018 
1019 #ifdef CONFIG_CFI_CLANG
1020 static int cfi_handler(struct pt_regs *regs, unsigned long esr)
1021 {
1022 	unsigned long target;
1023 	u32 type;
1024 
1025 	target = pt_regs_read_reg(regs, FIELD_GET(CFI_BRK_IMM_TARGET, esr));
1026 	type = (u32)pt_regs_read_reg(regs, FIELD_GET(CFI_BRK_IMM_TYPE, esr));
1027 
1028 	switch (report_cfi_failure(regs, regs->pc, &target, type)) {
1029 	case BUG_TRAP_TYPE_BUG:
1030 		die("Oops - CFI", regs, 0);
1031 		break;
1032 
1033 	case BUG_TRAP_TYPE_WARN:
1034 		break;
1035 
1036 	default:
1037 		return DBG_HOOK_ERROR;
1038 	}
1039 
1040 	arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
1041 	return DBG_HOOK_HANDLED;
1042 }
1043 
1044 static struct break_hook cfi_break_hook = {
1045 	.fn = cfi_handler,
1046 	.imm = CFI_BRK_IMM_BASE,
1047 	.mask = CFI_BRK_IMM_MASK,
1048 };
1049 #endif /* CONFIG_CFI_CLANG */
1050 
1051 static int reserved_fault_handler(struct pt_regs *regs, unsigned long esr)
1052 {
1053 	pr_err("%s generated an invalid instruction at %pS!\n",
1054 		"Kernel text patching",
1055 		(void *)instruction_pointer(regs));
1056 
1057 	/* We cannot handle this */
1058 	return DBG_HOOK_ERROR;
1059 }
1060 
1061 static struct break_hook fault_break_hook = {
1062 	.fn = reserved_fault_handler,
1063 	.imm = FAULT_BRK_IMM,
1064 };
1065 
1066 #ifdef CONFIG_KASAN_SW_TAGS
1067 
1068 #define KASAN_ESR_RECOVER	0x20
1069 #define KASAN_ESR_WRITE	0x10
1070 #define KASAN_ESR_SIZE_MASK	0x0f
1071 #define KASAN_ESR_SIZE(esr)	(1 << ((esr) & KASAN_ESR_SIZE_MASK))
1072 
1073 static int kasan_handler(struct pt_regs *regs, unsigned long esr)
1074 {
1075 	bool recover = esr & KASAN_ESR_RECOVER;
1076 	bool write = esr & KASAN_ESR_WRITE;
1077 	size_t size = KASAN_ESR_SIZE(esr);
1078 	u64 addr = regs->regs[0];
1079 	u64 pc = regs->pc;
1080 
1081 	kasan_report(addr, size, write, pc);
1082 
1083 	/*
1084 	 * The instrumentation allows to control whether we can proceed after
1085 	 * a crash was detected. This is done by passing the -recover flag to
1086 	 * the compiler. Disabling recovery allows to generate more compact
1087 	 * code.
1088 	 *
1089 	 * Unfortunately disabling recovery doesn't work for the kernel right
1090 	 * now. KASAN reporting is disabled in some contexts (for example when
1091 	 * the allocator accesses slab object metadata; this is controlled by
1092 	 * current->kasan_depth). All these accesses are detected by the tool,
1093 	 * even though the reports for them are not printed.
1094 	 *
1095 	 * This is something that might be fixed at some point in the future.
1096 	 */
1097 	if (!recover)
1098 		die("Oops - KASAN", regs, esr);
1099 
1100 	/* If thread survives, skip over the brk instruction and continue: */
1101 	arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
1102 	return DBG_HOOK_HANDLED;
1103 }
1104 
1105 static struct break_hook kasan_break_hook = {
1106 	.fn	= kasan_handler,
1107 	.imm	= KASAN_BRK_IMM,
1108 	.mask	= KASAN_BRK_MASK,
1109 };
1110 #endif
1111 
1112 
1113 #define esr_comment(esr) ((esr) & ESR_ELx_BRK64_ISS_COMMENT_MASK)
1114 
1115 /*
1116  * Initial handler for AArch64 BRK exceptions
1117  * This handler only used until debug_traps_init().
1118  */
1119 int __init early_brk64(unsigned long addr, unsigned long esr,
1120 		struct pt_regs *regs)
1121 {
1122 #ifdef CONFIG_CFI_CLANG
1123 	if ((esr_comment(esr) & ~CFI_BRK_IMM_MASK) == CFI_BRK_IMM_BASE)
1124 		return cfi_handler(regs, esr) != DBG_HOOK_HANDLED;
1125 #endif
1126 #ifdef CONFIG_KASAN_SW_TAGS
1127 	if ((esr_comment(esr) & ~KASAN_BRK_MASK) == KASAN_BRK_IMM)
1128 		return kasan_handler(regs, esr) != DBG_HOOK_HANDLED;
1129 #endif
1130 	return bug_handler(regs, esr) != DBG_HOOK_HANDLED;
1131 }
1132 
1133 void __init trap_init(void)
1134 {
1135 	register_kernel_break_hook(&bug_break_hook);
1136 #ifdef CONFIG_CFI_CLANG
1137 	register_kernel_break_hook(&cfi_break_hook);
1138 #endif
1139 	register_kernel_break_hook(&fault_break_hook);
1140 #ifdef CONFIG_KASAN_SW_TAGS
1141 	register_kernel_break_hook(&kasan_break_hook);
1142 #endif
1143 	debug_traps_init();
1144 }
1145