xref: /linux-6.15/arch/arm64/kernel/traps.c (revision bff8f413)
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 call_undef_hook(struct pt_regs *regs)
398 {
399 	struct undef_hook *hook;
400 	unsigned long flags;
401 	u32 instr;
402 	int (*fn)(struct pt_regs *regs, u32 instr) = NULL;
403 	unsigned long pc = instruction_pointer(regs);
404 
405 	if (compat_thumb_mode(regs)) {
406 		/* 16-bit Thumb instruction */
407 		__le16 instr_le;
408 		if (get_user(instr_le, (__le16 __user *)pc))
409 			goto exit;
410 		instr = le16_to_cpu(instr_le);
411 		if (aarch32_insn_is_wide(instr)) {
412 			u32 instr2;
413 
414 			if (get_user(instr_le, (__le16 __user *)(pc + 2)))
415 				goto exit;
416 			instr2 = le16_to_cpu(instr_le);
417 			instr = (instr << 16) | instr2;
418 		}
419 	} else {
420 		/* 32-bit ARM instruction */
421 		__le32 instr_le;
422 		if (get_user(instr_le, (__le32 __user *)pc))
423 			goto exit;
424 		instr = le32_to_cpu(instr_le);
425 	}
426 
427 	raw_spin_lock_irqsave(&undef_lock, flags);
428 	list_for_each_entry(hook, &undef_hook, node)
429 		if ((instr & hook->instr_mask) == hook->instr_val &&
430 			(regs->pstate & hook->pstate_mask) == hook->pstate_val)
431 			fn = hook->fn;
432 
433 	raw_spin_unlock_irqrestore(&undef_lock, flags);
434 exit:
435 	return fn ? fn(regs, instr) : 1;
436 }
437 
438 void force_signal_inject(int signal, int code, unsigned long address, unsigned long err)
439 {
440 	const char *desc;
441 	struct pt_regs *regs = current_pt_regs();
442 
443 	if (WARN_ON(!user_mode(regs)))
444 		return;
445 
446 	switch (signal) {
447 	case SIGILL:
448 		desc = "undefined instruction";
449 		break;
450 	case SIGSEGV:
451 		desc = "illegal memory access";
452 		break;
453 	default:
454 		desc = "unknown or unrecoverable error";
455 		break;
456 	}
457 
458 	/* Force signals we don't understand to SIGKILL */
459 	if (WARN_ON(signal != SIGKILL &&
460 		    siginfo_layout(signal, code) != SIL_FAULT)) {
461 		signal = SIGKILL;
462 	}
463 
464 	arm64_notify_die(desc, regs, signal, code, address, err);
465 }
466 
467 /*
468  * Set up process info to signal segmentation fault - called on access error.
469  */
470 void arm64_notify_segfault(unsigned long addr)
471 {
472 	int code;
473 
474 	mmap_read_lock(current->mm);
475 	if (find_vma(current->mm, untagged_addr(addr)) == NULL)
476 		code = SEGV_MAPERR;
477 	else
478 		code = SEGV_ACCERR;
479 	mmap_read_unlock(current->mm);
480 
481 	force_signal_inject(SIGSEGV, code, addr, 0);
482 }
483 
484 void do_el0_undef(struct pt_regs *regs, unsigned long esr)
485 {
486 	/* check for AArch32 breakpoint instructions */
487 	if (!aarch32_break_handler(regs))
488 		return;
489 
490 	if (call_undef_hook(regs) == 0)
491 		return;
492 
493 	force_signal_inject(SIGILL, ILL_ILLOPC, regs->pc, 0);
494 }
495 
496 void do_el1_undef(struct pt_regs *regs, unsigned long esr)
497 {
498 	u32 insn;
499 
500 	if (aarch64_insn_read((void *)regs->pc, &insn))
501 		goto out_err;
502 
503 	if (try_emulate_el1_ssbs(regs, insn))
504 		return;
505 
506 out_err:
507 	die("Oops - Undefined instruction", regs, esr);
508 }
509 
510 void do_el0_bti(struct pt_regs *regs)
511 {
512 	force_signal_inject(SIGILL, ILL_ILLOPC, regs->pc, 0);
513 }
514 
515 void do_el1_bti(struct pt_regs *regs, unsigned long esr)
516 {
517 	die("Oops - BTI", regs, esr);
518 }
519 
520 void do_el0_fpac(struct pt_regs *regs, unsigned long esr)
521 {
522 	force_signal_inject(SIGILL, ILL_ILLOPN, regs->pc, esr);
523 }
524 
525 void do_el1_fpac(struct pt_regs *regs, unsigned long esr)
526 {
527 	/*
528 	 * Unexpected FPAC exception in the kernel: kill the task before it
529 	 * does any more harm.
530 	 */
531 	die("Oops - FPAC", regs, esr);
532 }
533 
534 #define __user_cache_maint(insn, address, res)			\
535 	if (address >= TASK_SIZE_MAX) {				\
536 		res = -EFAULT;					\
537 	} else {						\
538 		uaccess_ttbr0_enable();				\
539 		asm volatile (					\
540 			"1:	" insn ", %1\n"			\
541 			"	mov	%w0, #0\n"		\
542 			"2:\n"					\
543 			_ASM_EXTABLE_UACCESS_ERR(1b, 2b, %w0)	\
544 			: "=r" (res)				\
545 			: "r" (address));			\
546 		uaccess_ttbr0_disable();			\
547 	}
548 
549 static void user_cache_maint_handler(unsigned long esr, struct pt_regs *regs)
550 {
551 	unsigned long tagged_address, address;
552 	int rt = ESR_ELx_SYS64_ISS_RT(esr);
553 	int crm = (esr & ESR_ELx_SYS64_ISS_CRM_MASK) >> ESR_ELx_SYS64_ISS_CRM_SHIFT;
554 	int ret = 0;
555 
556 	tagged_address = pt_regs_read_reg(regs, rt);
557 	address = untagged_addr(tagged_address);
558 
559 	switch (crm) {
560 	case ESR_ELx_SYS64_ISS_CRM_DC_CVAU:	/* DC CVAU, gets promoted */
561 		__user_cache_maint("dc civac", address, ret);
562 		break;
563 	case ESR_ELx_SYS64_ISS_CRM_DC_CVAC:	/* DC CVAC, gets promoted */
564 		__user_cache_maint("dc civac", address, ret);
565 		break;
566 	case ESR_ELx_SYS64_ISS_CRM_DC_CVADP:	/* DC CVADP */
567 		__user_cache_maint("sys 3, c7, c13, 1", address, ret);
568 		break;
569 	case ESR_ELx_SYS64_ISS_CRM_DC_CVAP:	/* DC CVAP */
570 		__user_cache_maint("sys 3, c7, c12, 1", address, ret);
571 		break;
572 	case ESR_ELx_SYS64_ISS_CRM_DC_CIVAC:	/* DC CIVAC */
573 		__user_cache_maint("dc civac", address, ret);
574 		break;
575 	case ESR_ELx_SYS64_ISS_CRM_IC_IVAU:	/* IC IVAU */
576 		__user_cache_maint("ic ivau", address, ret);
577 		break;
578 	default:
579 		force_signal_inject(SIGILL, ILL_ILLOPC, regs->pc, 0);
580 		return;
581 	}
582 
583 	if (ret)
584 		arm64_notify_segfault(tagged_address);
585 	else
586 		arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
587 }
588 
589 static void ctr_read_handler(unsigned long esr, struct pt_regs *regs)
590 {
591 	int rt = ESR_ELx_SYS64_ISS_RT(esr);
592 	unsigned long val = arm64_ftr_reg_user_value(&arm64_ftr_reg_ctrel0);
593 
594 	if (cpus_have_const_cap(ARM64_WORKAROUND_1542419)) {
595 		/* Hide DIC so that we can trap the unnecessary maintenance...*/
596 		val &= ~BIT(CTR_EL0_DIC_SHIFT);
597 
598 		/* ... and fake IminLine to reduce the number of traps. */
599 		val &= ~CTR_EL0_IminLine_MASK;
600 		val |= (PAGE_SHIFT - 2) & CTR_EL0_IminLine_MASK;
601 	}
602 
603 	pt_regs_write_reg(regs, rt, val);
604 
605 	arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
606 }
607 
608 static void cntvct_read_handler(unsigned long esr, struct pt_regs *regs)
609 {
610 	int rt = ESR_ELx_SYS64_ISS_RT(esr);
611 
612 	pt_regs_write_reg(regs, rt, arch_timer_read_counter());
613 	arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
614 }
615 
616 static void cntfrq_read_handler(unsigned long esr, struct pt_regs *regs)
617 {
618 	int rt = ESR_ELx_SYS64_ISS_RT(esr);
619 
620 	pt_regs_write_reg(regs, rt, arch_timer_get_rate());
621 	arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
622 }
623 
624 static void mrs_handler(unsigned long esr, struct pt_regs *regs)
625 {
626 	u32 sysreg, rt;
627 
628 	rt = ESR_ELx_SYS64_ISS_RT(esr);
629 	sysreg = esr_sys64_to_sysreg(esr);
630 
631 	if (do_emulate_mrs(regs, sysreg, rt) != 0)
632 		force_signal_inject(SIGILL, ILL_ILLOPC, regs->pc, 0);
633 }
634 
635 static void wfi_handler(unsigned long esr, struct pt_regs *regs)
636 {
637 	arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
638 }
639 
640 struct sys64_hook {
641 	unsigned long esr_mask;
642 	unsigned long esr_val;
643 	void (*handler)(unsigned long esr, struct pt_regs *regs);
644 };
645 
646 static const struct sys64_hook sys64_hooks[] = {
647 	{
648 		.esr_mask = ESR_ELx_SYS64_ISS_EL0_CACHE_OP_MASK,
649 		.esr_val = ESR_ELx_SYS64_ISS_EL0_CACHE_OP_VAL,
650 		.handler = user_cache_maint_handler,
651 	},
652 	{
653 		/* Trap read access to CTR_EL0 */
654 		.esr_mask = ESR_ELx_SYS64_ISS_SYS_OP_MASK,
655 		.esr_val = ESR_ELx_SYS64_ISS_SYS_CTR_READ,
656 		.handler = ctr_read_handler,
657 	},
658 	{
659 		/* Trap read access to CNTVCT_EL0 */
660 		.esr_mask = ESR_ELx_SYS64_ISS_SYS_OP_MASK,
661 		.esr_val = ESR_ELx_SYS64_ISS_SYS_CNTVCT,
662 		.handler = cntvct_read_handler,
663 	},
664 	{
665 		/* Trap read access to CNTVCTSS_EL0 */
666 		.esr_mask = ESR_ELx_SYS64_ISS_SYS_OP_MASK,
667 		.esr_val = ESR_ELx_SYS64_ISS_SYS_CNTVCTSS,
668 		.handler = cntvct_read_handler,
669 	},
670 	{
671 		/* Trap read access to CNTFRQ_EL0 */
672 		.esr_mask = ESR_ELx_SYS64_ISS_SYS_OP_MASK,
673 		.esr_val = ESR_ELx_SYS64_ISS_SYS_CNTFRQ,
674 		.handler = cntfrq_read_handler,
675 	},
676 	{
677 		/* Trap read access to CPUID registers */
678 		.esr_mask = ESR_ELx_SYS64_ISS_SYS_MRS_OP_MASK,
679 		.esr_val = ESR_ELx_SYS64_ISS_SYS_MRS_OP_VAL,
680 		.handler = mrs_handler,
681 	},
682 	{
683 		/* Trap WFI instructions executed in userspace */
684 		.esr_mask = ESR_ELx_WFx_MASK,
685 		.esr_val = ESR_ELx_WFx_WFI_VAL,
686 		.handler = wfi_handler,
687 	},
688 	{},
689 };
690 
691 #ifdef CONFIG_COMPAT
692 static bool cp15_cond_valid(unsigned long esr, struct pt_regs *regs)
693 {
694 	int cond;
695 
696 	/* Only a T32 instruction can trap without CV being set */
697 	if (!(esr & ESR_ELx_CV)) {
698 		u32 it;
699 
700 		it = compat_get_it_state(regs);
701 		if (!it)
702 			return true;
703 
704 		cond = it >> 4;
705 	} else {
706 		cond = (esr & ESR_ELx_COND_MASK) >> ESR_ELx_COND_SHIFT;
707 	}
708 
709 	return aarch32_opcode_cond_checks[cond](regs->pstate);
710 }
711 
712 static void compat_cntfrq_read_handler(unsigned long esr, struct pt_regs *regs)
713 {
714 	int reg = (esr & ESR_ELx_CP15_32_ISS_RT_MASK) >> ESR_ELx_CP15_32_ISS_RT_SHIFT;
715 
716 	pt_regs_write_reg(regs, reg, arch_timer_get_rate());
717 	arm64_skip_faulting_instruction(regs, 4);
718 }
719 
720 static const struct sys64_hook cp15_32_hooks[] = {
721 	{
722 		.esr_mask = ESR_ELx_CP15_32_ISS_SYS_MASK,
723 		.esr_val = ESR_ELx_CP15_32_ISS_SYS_CNTFRQ,
724 		.handler = compat_cntfrq_read_handler,
725 	},
726 	{},
727 };
728 
729 static void compat_cntvct_read_handler(unsigned long esr, struct pt_regs *regs)
730 {
731 	int rt = (esr & ESR_ELx_CP15_64_ISS_RT_MASK) >> ESR_ELx_CP15_64_ISS_RT_SHIFT;
732 	int rt2 = (esr & ESR_ELx_CP15_64_ISS_RT2_MASK) >> ESR_ELx_CP15_64_ISS_RT2_SHIFT;
733 	u64 val = arch_timer_read_counter();
734 
735 	pt_regs_write_reg(regs, rt, lower_32_bits(val));
736 	pt_regs_write_reg(regs, rt2, upper_32_bits(val));
737 	arm64_skip_faulting_instruction(regs, 4);
738 }
739 
740 static const struct sys64_hook cp15_64_hooks[] = {
741 	{
742 		.esr_mask = ESR_ELx_CP15_64_ISS_SYS_MASK,
743 		.esr_val = ESR_ELx_CP15_64_ISS_SYS_CNTVCT,
744 		.handler = compat_cntvct_read_handler,
745 	},
746 	{
747 		.esr_mask = ESR_ELx_CP15_64_ISS_SYS_MASK,
748 		.esr_val = ESR_ELx_CP15_64_ISS_SYS_CNTVCTSS,
749 		.handler = compat_cntvct_read_handler,
750 	},
751 	{},
752 };
753 
754 void do_el0_cp15(unsigned long esr, struct pt_regs *regs)
755 {
756 	const struct sys64_hook *hook, *hook_base;
757 
758 	if (!cp15_cond_valid(esr, regs)) {
759 		/*
760 		 * There is no T16 variant of a CP access, so we
761 		 * always advance PC by 4 bytes.
762 		 */
763 		arm64_skip_faulting_instruction(regs, 4);
764 		return;
765 	}
766 
767 	switch (ESR_ELx_EC(esr)) {
768 	case ESR_ELx_EC_CP15_32:
769 		hook_base = cp15_32_hooks;
770 		break;
771 	case ESR_ELx_EC_CP15_64:
772 		hook_base = cp15_64_hooks;
773 		break;
774 	default:
775 		do_el0_undef(regs, esr);
776 		return;
777 	}
778 
779 	for (hook = hook_base; hook->handler; hook++)
780 		if ((hook->esr_mask & esr) == hook->esr_val) {
781 			hook->handler(esr, regs);
782 			return;
783 		}
784 
785 	/*
786 	 * New cp15 instructions may previously have been undefined at
787 	 * EL0. Fall back to our usual undefined instruction handler
788 	 * so that we handle these consistently.
789 	 */
790 	do_el0_undef(regs, esr);
791 }
792 #endif
793 
794 void do_el0_sys(unsigned long esr, struct pt_regs *regs)
795 {
796 	const struct sys64_hook *hook;
797 
798 	for (hook = sys64_hooks; hook->handler; hook++)
799 		if ((hook->esr_mask & esr) == hook->esr_val) {
800 			hook->handler(esr, regs);
801 			return;
802 		}
803 
804 	/*
805 	 * New SYS instructions may previously have been undefined at EL0. Fall
806 	 * back to our usual undefined instruction handler so that we handle
807 	 * these consistently.
808 	 */
809 	do_el0_undef(regs, esr);
810 }
811 
812 static const char *esr_class_str[] = {
813 	[0 ... ESR_ELx_EC_MAX]		= "UNRECOGNIZED EC",
814 	[ESR_ELx_EC_UNKNOWN]		= "Unknown/Uncategorized",
815 	[ESR_ELx_EC_WFx]		= "WFI/WFE",
816 	[ESR_ELx_EC_CP15_32]		= "CP15 MCR/MRC",
817 	[ESR_ELx_EC_CP15_64]		= "CP15 MCRR/MRRC",
818 	[ESR_ELx_EC_CP14_MR]		= "CP14 MCR/MRC",
819 	[ESR_ELx_EC_CP14_LS]		= "CP14 LDC/STC",
820 	[ESR_ELx_EC_FP_ASIMD]		= "ASIMD",
821 	[ESR_ELx_EC_CP10_ID]		= "CP10 MRC/VMRS",
822 	[ESR_ELx_EC_PAC]		= "PAC",
823 	[ESR_ELx_EC_CP14_64]		= "CP14 MCRR/MRRC",
824 	[ESR_ELx_EC_BTI]		= "BTI",
825 	[ESR_ELx_EC_ILL]		= "PSTATE.IL",
826 	[ESR_ELx_EC_SVC32]		= "SVC (AArch32)",
827 	[ESR_ELx_EC_HVC32]		= "HVC (AArch32)",
828 	[ESR_ELx_EC_SMC32]		= "SMC (AArch32)",
829 	[ESR_ELx_EC_SVC64]		= "SVC (AArch64)",
830 	[ESR_ELx_EC_HVC64]		= "HVC (AArch64)",
831 	[ESR_ELx_EC_SMC64]		= "SMC (AArch64)",
832 	[ESR_ELx_EC_SYS64]		= "MSR/MRS (AArch64)",
833 	[ESR_ELx_EC_SVE]		= "SVE",
834 	[ESR_ELx_EC_ERET]		= "ERET/ERETAA/ERETAB",
835 	[ESR_ELx_EC_FPAC]		= "FPAC",
836 	[ESR_ELx_EC_SME]		= "SME",
837 	[ESR_ELx_EC_IMP_DEF]		= "EL3 IMP DEF",
838 	[ESR_ELx_EC_IABT_LOW]		= "IABT (lower EL)",
839 	[ESR_ELx_EC_IABT_CUR]		= "IABT (current EL)",
840 	[ESR_ELx_EC_PC_ALIGN]		= "PC Alignment",
841 	[ESR_ELx_EC_DABT_LOW]		= "DABT (lower EL)",
842 	[ESR_ELx_EC_DABT_CUR]		= "DABT (current EL)",
843 	[ESR_ELx_EC_SP_ALIGN]		= "SP Alignment",
844 	[ESR_ELx_EC_FP_EXC32]		= "FP (AArch32)",
845 	[ESR_ELx_EC_FP_EXC64]		= "FP (AArch64)",
846 	[ESR_ELx_EC_SERROR]		= "SError",
847 	[ESR_ELx_EC_BREAKPT_LOW]	= "Breakpoint (lower EL)",
848 	[ESR_ELx_EC_BREAKPT_CUR]	= "Breakpoint (current EL)",
849 	[ESR_ELx_EC_SOFTSTP_LOW]	= "Software Step (lower EL)",
850 	[ESR_ELx_EC_SOFTSTP_CUR]	= "Software Step (current EL)",
851 	[ESR_ELx_EC_WATCHPT_LOW]	= "Watchpoint (lower EL)",
852 	[ESR_ELx_EC_WATCHPT_CUR]	= "Watchpoint (current EL)",
853 	[ESR_ELx_EC_BKPT32]		= "BKPT (AArch32)",
854 	[ESR_ELx_EC_VECTOR32]		= "Vector catch (AArch32)",
855 	[ESR_ELx_EC_BRK64]		= "BRK (AArch64)",
856 };
857 
858 const char *esr_get_class_string(unsigned long esr)
859 {
860 	return esr_class_str[ESR_ELx_EC(esr)];
861 }
862 
863 /*
864  * bad_el0_sync handles unexpected, but potentially recoverable synchronous
865  * exceptions taken from EL0.
866  */
867 void bad_el0_sync(struct pt_regs *regs, int reason, unsigned long esr)
868 {
869 	unsigned long pc = instruction_pointer(regs);
870 
871 	current->thread.fault_address = 0;
872 	current->thread.fault_code = esr;
873 
874 	arm64_force_sig_fault(SIGILL, ILL_ILLOPC, pc,
875 			      "Bad EL0 synchronous exception");
876 }
877 
878 #ifdef CONFIG_VMAP_STACK
879 
880 DEFINE_PER_CPU(unsigned long [OVERFLOW_STACK_SIZE/sizeof(long)], overflow_stack)
881 	__aligned(16);
882 
883 void panic_bad_stack(struct pt_regs *regs, unsigned long esr, unsigned long far)
884 {
885 	unsigned long tsk_stk = (unsigned long)current->stack;
886 	unsigned long irq_stk = (unsigned long)this_cpu_read(irq_stack_ptr);
887 	unsigned long ovf_stk = (unsigned long)this_cpu_ptr(overflow_stack);
888 
889 	console_verbose();
890 	pr_emerg("Insufficient stack space to handle exception!");
891 
892 	pr_emerg("ESR: 0x%016lx -- %s\n", esr, esr_get_class_string(esr));
893 	pr_emerg("FAR: 0x%016lx\n", far);
894 
895 	pr_emerg("Task stack:     [0x%016lx..0x%016lx]\n",
896 		 tsk_stk, tsk_stk + THREAD_SIZE);
897 	pr_emerg("IRQ stack:      [0x%016lx..0x%016lx]\n",
898 		 irq_stk, irq_stk + IRQ_STACK_SIZE);
899 	pr_emerg("Overflow stack: [0x%016lx..0x%016lx]\n",
900 		 ovf_stk, ovf_stk + OVERFLOW_STACK_SIZE);
901 
902 	__show_regs(regs);
903 
904 	/*
905 	 * We use nmi_panic to limit the potential for recusive overflows, and
906 	 * to get a better stack trace.
907 	 */
908 	nmi_panic(NULL, "kernel stack overflow");
909 	cpu_park_loop();
910 }
911 #endif
912 
913 void __noreturn arm64_serror_panic(struct pt_regs *regs, unsigned long esr)
914 {
915 	console_verbose();
916 
917 	pr_crit("SError Interrupt on CPU%d, code 0x%016lx -- %s\n",
918 		smp_processor_id(), esr, esr_get_class_string(esr));
919 	if (regs)
920 		__show_regs(regs);
921 
922 	nmi_panic(regs, "Asynchronous SError Interrupt");
923 
924 	cpu_park_loop();
925 	unreachable();
926 }
927 
928 bool arm64_is_fatal_ras_serror(struct pt_regs *regs, unsigned long esr)
929 {
930 	unsigned long aet = arm64_ras_serror_get_severity(esr);
931 
932 	switch (aet) {
933 	case ESR_ELx_AET_CE:	/* corrected error */
934 	case ESR_ELx_AET_UEO:	/* restartable, not yet consumed */
935 		/*
936 		 * The CPU can make progress. We may take UEO again as
937 		 * a more severe error.
938 		 */
939 		return false;
940 
941 	case ESR_ELx_AET_UEU:	/* Uncorrected Unrecoverable */
942 	case ESR_ELx_AET_UER:	/* Uncorrected Recoverable */
943 		/*
944 		 * The CPU can't make progress. The exception may have
945 		 * been imprecise.
946 		 *
947 		 * Neoverse-N1 #1349291 means a non-KVM SError reported as
948 		 * Unrecoverable should be treated as Uncontainable. We
949 		 * call arm64_serror_panic() in both cases.
950 		 */
951 		return true;
952 
953 	case ESR_ELx_AET_UC:	/* Uncontainable or Uncategorized error */
954 	default:
955 		/* Error has been silently propagated */
956 		arm64_serror_panic(regs, esr);
957 	}
958 }
959 
960 void do_serror(struct pt_regs *regs, unsigned long esr)
961 {
962 	/* non-RAS errors are not containable */
963 	if (!arm64_is_ras_serror(esr) || arm64_is_fatal_ras_serror(regs, esr))
964 		arm64_serror_panic(regs, esr);
965 }
966 
967 /* GENERIC_BUG traps */
968 
969 int is_valid_bugaddr(unsigned long addr)
970 {
971 	/*
972 	 * bug_handler() only called for BRK #BUG_BRK_IMM.
973 	 * So the answer is trivial -- any spurious instances with no
974 	 * bug table entry will be rejected by report_bug() and passed
975 	 * back to the debug-monitors code and handled as a fatal
976 	 * unexpected debug exception.
977 	 */
978 	return 1;
979 }
980 
981 static int bug_handler(struct pt_regs *regs, unsigned long esr)
982 {
983 	switch (report_bug(regs->pc, regs)) {
984 	case BUG_TRAP_TYPE_BUG:
985 		die("Oops - BUG", regs, esr);
986 		break;
987 
988 	case BUG_TRAP_TYPE_WARN:
989 		break;
990 
991 	default:
992 		/* unknown/unrecognised bug trap type */
993 		return DBG_HOOK_ERROR;
994 	}
995 
996 	/* If thread survives, skip over the BUG instruction and continue: */
997 	arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
998 	return DBG_HOOK_HANDLED;
999 }
1000 
1001 static struct break_hook bug_break_hook = {
1002 	.fn = bug_handler,
1003 	.imm = BUG_BRK_IMM,
1004 };
1005 
1006 #ifdef CONFIG_CFI_CLANG
1007 static int cfi_handler(struct pt_regs *regs, unsigned long esr)
1008 {
1009 	unsigned long target;
1010 	u32 type;
1011 
1012 	target = pt_regs_read_reg(regs, FIELD_GET(CFI_BRK_IMM_TARGET, esr));
1013 	type = (u32)pt_regs_read_reg(regs, FIELD_GET(CFI_BRK_IMM_TYPE, esr));
1014 
1015 	switch (report_cfi_failure(regs, regs->pc, &target, type)) {
1016 	case BUG_TRAP_TYPE_BUG:
1017 		die("Oops - CFI", regs, 0);
1018 		break;
1019 
1020 	case BUG_TRAP_TYPE_WARN:
1021 		break;
1022 
1023 	default:
1024 		return DBG_HOOK_ERROR;
1025 	}
1026 
1027 	arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
1028 	return DBG_HOOK_HANDLED;
1029 }
1030 
1031 static struct break_hook cfi_break_hook = {
1032 	.fn = cfi_handler,
1033 	.imm = CFI_BRK_IMM_BASE,
1034 	.mask = CFI_BRK_IMM_MASK,
1035 };
1036 #endif /* CONFIG_CFI_CLANG */
1037 
1038 static int reserved_fault_handler(struct pt_regs *regs, unsigned long esr)
1039 {
1040 	pr_err("%s generated an invalid instruction at %pS!\n",
1041 		"Kernel text patching",
1042 		(void *)instruction_pointer(regs));
1043 
1044 	/* We cannot handle this */
1045 	return DBG_HOOK_ERROR;
1046 }
1047 
1048 static struct break_hook fault_break_hook = {
1049 	.fn = reserved_fault_handler,
1050 	.imm = FAULT_BRK_IMM,
1051 };
1052 
1053 #ifdef CONFIG_KASAN_SW_TAGS
1054 
1055 #define KASAN_ESR_RECOVER	0x20
1056 #define KASAN_ESR_WRITE	0x10
1057 #define KASAN_ESR_SIZE_MASK	0x0f
1058 #define KASAN_ESR_SIZE(esr)	(1 << ((esr) & KASAN_ESR_SIZE_MASK))
1059 
1060 static int kasan_handler(struct pt_regs *regs, unsigned long esr)
1061 {
1062 	bool recover = esr & KASAN_ESR_RECOVER;
1063 	bool write = esr & KASAN_ESR_WRITE;
1064 	size_t size = KASAN_ESR_SIZE(esr);
1065 	u64 addr = regs->regs[0];
1066 	u64 pc = regs->pc;
1067 
1068 	kasan_report(addr, size, write, pc);
1069 
1070 	/*
1071 	 * The instrumentation allows to control whether we can proceed after
1072 	 * a crash was detected. This is done by passing the -recover flag to
1073 	 * the compiler. Disabling recovery allows to generate more compact
1074 	 * code.
1075 	 *
1076 	 * Unfortunately disabling recovery doesn't work for the kernel right
1077 	 * now. KASAN reporting is disabled in some contexts (for example when
1078 	 * the allocator accesses slab object metadata; this is controlled by
1079 	 * current->kasan_depth). All these accesses are detected by the tool,
1080 	 * even though the reports for them are not printed.
1081 	 *
1082 	 * This is something that might be fixed at some point in the future.
1083 	 */
1084 	if (!recover)
1085 		die("Oops - KASAN", regs, esr);
1086 
1087 	/* If thread survives, skip over the brk instruction and continue: */
1088 	arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
1089 	return DBG_HOOK_HANDLED;
1090 }
1091 
1092 static struct break_hook kasan_break_hook = {
1093 	.fn	= kasan_handler,
1094 	.imm	= KASAN_BRK_IMM,
1095 	.mask	= KASAN_BRK_MASK,
1096 };
1097 #endif
1098 
1099 
1100 #define esr_comment(esr) ((esr) & ESR_ELx_BRK64_ISS_COMMENT_MASK)
1101 
1102 /*
1103  * Initial handler for AArch64 BRK exceptions
1104  * This handler only used until debug_traps_init().
1105  */
1106 int __init early_brk64(unsigned long addr, unsigned long esr,
1107 		struct pt_regs *regs)
1108 {
1109 #ifdef CONFIG_CFI_CLANG
1110 	if ((esr_comment(esr) & ~CFI_BRK_IMM_MASK) == CFI_BRK_IMM_BASE)
1111 		return cfi_handler(regs, esr) != DBG_HOOK_HANDLED;
1112 #endif
1113 #ifdef CONFIG_KASAN_SW_TAGS
1114 	if ((esr_comment(esr) & ~KASAN_BRK_MASK) == KASAN_BRK_IMM)
1115 		return kasan_handler(regs, esr) != DBG_HOOK_HANDLED;
1116 #endif
1117 	return bug_handler(regs, esr) != DBG_HOOK_HANDLED;
1118 }
1119 
1120 void __init trap_init(void)
1121 {
1122 	register_kernel_break_hook(&bug_break_hook);
1123 #ifdef CONFIG_CFI_CLANG
1124 	register_kernel_break_hook(&cfi_break_hook);
1125 #endif
1126 	register_kernel_break_hook(&fault_break_hook);
1127 #ifdef CONFIG_KASAN_SW_TAGS
1128 	register_kernel_break_hook(&kasan_break_hook);
1129 #endif
1130 	debug_traps_init();
1131 }
1132