xref: /openbmc/linux/arch/arm64/kernel/traps.c (revision ecfb9f40)
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 int user_insn_read(struct pt_regs *regs, u32 *insnp)
377 {
378 	u32 instr;
379 	unsigned long pc = instruction_pointer(regs);
380 
381 	if (compat_thumb_mode(regs)) {
382 		/* 16-bit Thumb instruction */
383 		__le16 instr_le;
384 		if (get_user(instr_le, (__le16 __user *)pc))
385 			return -EFAULT;
386 		instr = le16_to_cpu(instr_le);
387 		if (aarch32_insn_is_wide(instr)) {
388 			u32 instr2;
389 
390 			if (get_user(instr_le, (__le16 __user *)(pc + 2)))
391 				return -EFAULT;
392 			instr2 = le16_to_cpu(instr_le);
393 			instr = (instr << 16) | instr2;
394 		}
395 	} else {
396 		/* 32-bit ARM instruction */
397 		__le32 instr_le;
398 		if (get_user(instr_le, (__le32 __user *)pc))
399 			return -EFAULT;
400 		instr = le32_to_cpu(instr_le);
401 	}
402 
403 	*insnp = instr;
404 	return 0;
405 }
406 
407 void force_signal_inject(int signal, int code, unsigned long address, unsigned long err)
408 {
409 	const char *desc;
410 	struct pt_regs *regs = current_pt_regs();
411 
412 	if (WARN_ON(!user_mode(regs)))
413 		return;
414 
415 	switch (signal) {
416 	case SIGILL:
417 		desc = "undefined instruction";
418 		break;
419 	case SIGSEGV:
420 		desc = "illegal memory access";
421 		break;
422 	default:
423 		desc = "unknown or unrecoverable error";
424 		break;
425 	}
426 
427 	/* Force signals we don't understand to SIGKILL */
428 	if (WARN_ON(signal != SIGKILL &&
429 		    siginfo_layout(signal, code) != SIL_FAULT)) {
430 		signal = SIGKILL;
431 	}
432 
433 	arm64_notify_die(desc, regs, signal, code, address, err);
434 }
435 
436 /*
437  * Set up process info to signal segmentation fault - called on access error.
438  */
439 void arm64_notify_segfault(unsigned long addr)
440 {
441 	int code;
442 
443 	mmap_read_lock(current->mm);
444 	if (find_vma(current->mm, untagged_addr(addr)) == NULL)
445 		code = SEGV_MAPERR;
446 	else
447 		code = SEGV_ACCERR;
448 	mmap_read_unlock(current->mm);
449 
450 	force_signal_inject(SIGSEGV, code, addr, 0);
451 }
452 
453 void do_el0_undef(struct pt_regs *regs, unsigned long esr)
454 {
455 	u32 insn;
456 
457 	/* check for AArch32 breakpoint instructions */
458 	if (!aarch32_break_handler(regs))
459 		return;
460 
461 	if (user_insn_read(regs, &insn))
462 		goto out_err;
463 
464 	if (try_emulate_mrs(regs, insn))
465 		return;
466 
467 	if (try_emulate_armv8_deprecated(regs, insn))
468 		return;
469 
470 out_err:
471 	force_signal_inject(SIGILL, ILL_ILLOPC, regs->pc, 0);
472 }
473 
474 void do_el1_undef(struct pt_regs *regs, unsigned long esr)
475 {
476 	u32 insn;
477 
478 	if (aarch64_insn_read((void *)regs->pc, &insn))
479 		goto out_err;
480 
481 	if (try_emulate_el1_ssbs(regs, insn))
482 		return;
483 
484 out_err:
485 	die("Oops - Undefined instruction", regs, esr);
486 }
487 
488 void do_el0_bti(struct pt_regs *regs)
489 {
490 	force_signal_inject(SIGILL, ILL_ILLOPC, regs->pc, 0);
491 }
492 
493 void do_el1_bti(struct pt_regs *regs, unsigned long esr)
494 {
495 	die("Oops - BTI", regs, esr);
496 }
497 
498 void do_el0_fpac(struct pt_regs *regs, unsigned long esr)
499 {
500 	force_signal_inject(SIGILL, ILL_ILLOPN, regs->pc, esr);
501 }
502 
503 void do_el1_fpac(struct pt_regs *regs, unsigned long esr)
504 {
505 	/*
506 	 * Unexpected FPAC exception in the kernel: kill the task before it
507 	 * does any more harm.
508 	 */
509 	die("Oops - FPAC", regs, esr);
510 }
511 
512 #define __user_cache_maint(insn, address, res)			\
513 	if (address >= TASK_SIZE_MAX) {				\
514 		res = -EFAULT;					\
515 	} else {						\
516 		uaccess_ttbr0_enable();				\
517 		asm volatile (					\
518 			"1:	" insn ", %1\n"			\
519 			"	mov	%w0, #0\n"		\
520 			"2:\n"					\
521 			_ASM_EXTABLE_UACCESS_ERR(1b, 2b, %w0)	\
522 			: "=r" (res)				\
523 			: "r" (address));			\
524 		uaccess_ttbr0_disable();			\
525 	}
526 
527 static void user_cache_maint_handler(unsigned long esr, struct pt_regs *regs)
528 {
529 	unsigned long tagged_address, address;
530 	int rt = ESR_ELx_SYS64_ISS_RT(esr);
531 	int crm = (esr & ESR_ELx_SYS64_ISS_CRM_MASK) >> ESR_ELx_SYS64_ISS_CRM_SHIFT;
532 	int ret = 0;
533 
534 	tagged_address = pt_regs_read_reg(regs, rt);
535 	address = untagged_addr(tagged_address);
536 
537 	switch (crm) {
538 	case ESR_ELx_SYS64_ISS_CRM_DC_CVAU:	/* DC CVAU, gets promoted */
539 		__user_cache_maint("dc civac", address, ret);
540 		break;
541 	case ESR_ELx_SYS64_ISS_CRM_DC_CVAC:	/* DC CVAC, gets promoted */
542 		__user_cache_maint("dc civac", address, ret);
543 		break;
544 	case ESR_ELx_SYS64_ISS_CRM_DC_CVADP:	/* DC CVADP */
545 		__user_cache_maint("sys 3, c7, c13, 1", address, ret);
546 		break;
547 	case ESR_ELx_SYS64_ISS_CRM_DC_CVAP:	/* DC CVAP */
548 		__user_cache_maint("sys 3, c7, c12, 1", address, ret);
549 		break;
550 	case ESR_ELx_SYS64_ISS_CRM_DC_CIVAC:	/* DC CIVAC */
551 		__user_cache_maint("dc civac", address, ret);
552 		break;
553 	case ESR_ELx_SYS64_ISS_CRM_IC_IVAU:	/* IC IVAU */
554 		__user_cache_maint("ic ivau", address, ret);
555 		break;
556 	default:
557 		force_signal_inject(SIGILL, ILL_ILLOPC, regs->pc, 0);
558 		return;
559 	}
560 
561 	if (ret)
562 		arm64_notify_segfault(tagged_address);
563 	else
564 		arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
565 }
566 
567 static void ctr_read_handler(unsigned long esr, struct pt_regs *regs)
568 {
569 	int rt = ESR_ELx_SYS64_ISS_RT(esr);
570 	unsigned long val = arm64_ftr_reg_user_value(&arm64_ftr_reg_ctrel0);
571 
572 	if (cpus_have_const_cap(ARM64_WORKAROUND_1542419)) {
573 		/* Hide DIC so that we can trap the unnecessary maintenance...*/
574 		val &= ~BIT(CTR_EL0_DIC_SHIFT);
575 
576 		/* ... and fake IminLine to reduce the number of traps. */
577 		val &= ~CTR_EL0_IminLine_MASK;
578 		val |= (PAGE_SHIFT - 2) & CTR_EL0_IminLine_MASK;
579 	}
580 
581 	pt_regs_write_reg(regs, rt, val);
582 
583 	arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
584 }
585 
586 static void cntvct_read_handler(unsigned long esr, struct pt_regs *regs)
587 {
588 	int rt = ESR_ELx_SYS64_ISS_RT(esr);
589 
590 	pt_regs_write_reg(regs, rt, arch_timer_read_counter());
591 	arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
592 }
593 
594 static void cntfrq_read_handler(unsigned long esr, struct pt_regs *regs)
595 {
596 	int rt = ESR_ELx_SYS64_ISS_RT(esr);
597 
598 	pt_regs_write_reg(regs, rt, arch_timer_get_rate());
599 	arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
600 }
601 
602 static void mrs_handler(unsigned long esr, struct pt_regs *regs)
603 {
604 	u32 sysreg, rt;
605 
606 	rt = ESR_ELx_SYS64_ISS_RT(esr);
607 	sysreg = esr_sys64_to_sysreg(esr);
608 
609 	if (do_emulate_mrs(regs, sysreg, rt) != 0)
610 		force_signal_inject(SIGILL, ILL_ILLOPC, regs->pc, 0);
611 }
612 
613 static void wfi_handler(unsigned long esr, struct pt_regs *regs)
614 {
615 	arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
616 }
617 
618 struct sys64_hook {
619 	unsigned long esr_mask;
620 	unsigned long esr_val;
621 	void (*handler)(unsigned long esr, struct pt_regs *regs);
622 };
623 
624 static const struct sys64_hook sys64_hooks[] = {
625 	{
626 		.esr_mask = ESR_ELx_SYS64_ISS_EL0_CACHE_OP_MASK,
627 		.esr_val = ESR_ELx_SYS64_ISS_EL0_CACHE_OP_VAL,
628 		.handler = user_cache_maint_handler,
629 	},
630 	{
631 		/* Trap read access to CTR_EL0 */
632 		.esr_mask = ESR_ELx_SYS64_ISS_SYS_OP_MASK,
633 		.esr_val = ESR_ELx_SYS64_ISS_SYS_CTR_READ,
634 		.handler = ctr_read_handler,
635 	},
636 	{
637 		/* Trap read access to CNTVCT_EL0 */
638 		.esr_mask = ESR_ELx_SYS64_ISS_SYS_OP_MASK,
639 		.esr_val = ESR_ELx_SYS64_ISS_SYS_CNTVCT,
640 		.handler = cntvct_read_handler,
641 	},
642 	{
643 		/* Trap read access to CNTVCTSS_EL0 */
644 		.esr_mask = ESR_ELx_SYS64_ISS_SYS_OP_MASK,
645 		.esr_val = ESR_ELx_SYS64_ISS_SYS_CNTVCTSS,
646 		.handler = cntvct_read_handler,
647 	},
648 	{
649 		/* Trap read access to CNTFRQ_EL0 */
650 		.esr_mask = ESR_ELx_SYS64_ISS_SYS_OP_MASK,
651 		.esr_val = ESR_ELx_SYS64_ISS_SYS_CNTFRQ,
652 		.handler = cntfrq_read_handler,
653 	},
654 	{
655 		/* Trap read access to CPUID registers */
656 		.esr_mask = ESR_ELx_SYS64_ISS_SYS_MRS_OP_MASK,
657 		.esr_val = ESR_ELx_SYS64_ISS_SYS_MRS_OP_VAL,
658 		.handler = mrs_handler,
659 	},
660 	{
661 		/* Trap WFI instructions executed in userspace */
662 		.esr_mask = ESR_ELx_WFx_MASK,
663 		.esr_val = ESR_ELx_WFx_WFI_VAL,
664 		.handler = wfi_handler,
665 	},
666 	{},
667 };
668 
669 #ifdef CONFIG_COMPAT
670 static bool cp15_cond_valid(unsigned long esr, struct pt_regs *regs)
671 {
672 	int cond;
673 
674 	/* Only a T32 instruction can trap without CV being set */
675 	if (!(esr & ESR_ELx_CV)) {
676 		u32 it;
677 
678 		it = compat_get_it_state(regs);
679 		if (!it)
680 			return true;
681 
682 		cond = it >> 4;
683 	} else {
684 		cond = (esr & ESR_ELx_COND_MASK) >> ESR_ELx_COND_SHIFT;
685 	}
686 
687 	return aarch32_opcode_cond_checks[cond](regs->pstate);
688 }
689 
690 static void compat_cntfrq_read_handler(unsigned long esr, struct pt_regs *regs)
691 {
692 	int reg = (esr & ESR_ELx_CP15_32_ISS_RT_MASK) >> ESR_ELx_CP15_32_ISS_RT_SHIFT;
693 
694 	pt_regs_write_reg(regs, reg, arch_timer_get_rate());
695 	arm64_skip_faulting_instruction(regs, 4);
696 }
697 
698 static const struct sys64_hook cp15_32_hooks[] = {
699 	{
700 		.esr_mask = ESR_ELx_CP15_32_ISS_SYS_MASK,
701 		.esr_val = ESR_ELx_CP15_32_ISS_SYS_CNTFRQ,
702 		.handler = compat_cntfrq_read_handler,
703 	},
704 	{},
705 };
706 
707 static void compat_cntvct_read_handler(unsigned long esr, struct pt_regs *regs)
708 {
709 	int rt = (esr & ESR_ELx_CP15_64_ISS_RT_MASK) >> ESR_ELx_CP15_64_ISS_RT_SHIFT;
710 	int rt2 = (esr & ESR_ELx_CP15_64_ISS_RT2_MASK) >> ESR_ELx_CP15_64_ISS_RT2_SHIFT;
711 	u64 val = arch_timer_read_counter();
712 
713 	pt_regs_write_reg(regs, rt, lower_32_bits(val));
714 	pt_regs_write_reg(regs, rt2, upper_32_bits(val));
715 	arm64_skip_faulting_instruction(regs, 4);
716 }
717 
718 static const struct sys64_hook cp15_64_hooks[] = {
719 	{
720 		.esr_mask = ESR_ELx_CP15_64_ISS_SYS_MASK,
721 		.esr_val = ESR_ELx_CP15_64_ISS_SYS_CNTVCT,
722 		.handler = compat_cntvct_read_handler,
723 	},
724 	{
725 		.esr_mask = ESR_ELx_CP15_64_ISS_SYS_MASK,
726 		.esr_val = ESR_ELx_CP15_64_ISS_SYS_CNTVCTSS,
727 		.handler = compat_cntvct_read_handler,
728 	},
729 	{},
730 };
731 
732 void do_el0_cp15(unsigned long esr, struct pt_regs *regs)
733 {
734 	const struct sys64_hook *hook, *hook_base;
735 
736 	if (!cp15_cond_valid(esr, regs)) {
737 		/*
738 		 * There is no T16 variant of a CP access, so we
739 		 * always advance PC by 4 bytes.
740 		 */
741 		arm64_skip_faulting_instruction(regs, 4);
742 		return;
743 	}
744 
745 	switch (ESR_ELx_EC(esr)) {
746 	case ESR_ELx_EC_CP15_32:
747 		hook_base = cp15_32_hooks;
748 		break;
749 	case ESR_ELx_EC_CP15_64:
750 		hook_base = cp15_64_hooks;
751 		break;
752 	default:
753 		do_el0_undef(regs, esr);
754 		return;
755 	}
756 
757 	for (hook = hook_base; hook->handler; hook++)
758 		if ((hook->esr_mask & esr) == hook->esr_val) {
759 			hook->handler(esr, regs);
760 			return;
761 		}
762 
763 	/*
764 	 * New cp15 instructions may previously have been undefined at
765 	 * EL0. Fall back to our usual undefined instruction handler
766 	 * so that we handle these consistently.
767 	 */
768 	do_el0_undef(regs, esr);
769 }
770 #endif
771 
772 void do_el0_sys(unsigned long esr, struct pt_regs *regs)
773 {
774 	const struct sys64_hook *hook;
775 
776 	for (hook = sys64_hooks; hook->handler; hook++)
777 		if ((hook->esr_mask & esr) == hook->esr_val) {
778 			hook->handler(esr, regs);
779 			return;
780 		}
781 
782 	/*
783 	 * New SYS instructions may previously have been undefined at EL0. Fall
784 	 * back to our usual undefined instruction handler so that we handle
785 	 * these consistently.
786 	 */
787 	do_el0_undef(regs, esr);
788 }
789 
790 static const char *esr_class_str[] = {
791 	[0 ... ESR_ELx_EC_MAX]		= "UNRECOGNIZED EC",
792 	[ESR_ELx_EC_UNKNOWN]		= "Unknown/Uncategorized",
793 	[ESR_ELx_EC_WFx]		= "WFI/WFE",
794 	[ESR_ELx_EC_CP15_32]		= "CP15 MCR/MRC",
795 	[ESR_ELx_EC_CP15_64]		= "CP15 MCRR/MRRC",
796 	[ESR_ELx_EC_CP14_MR]		= "CP14 MCR/MRC",
797 	[ESR_ELx_EC_CP14_LS]		= "CP14 LDC/STC",
798 	[ESR_ELx_EC_FP_ASIMD]		= "ASIMD",
799 	[ESR_ELx_EC_CP10_ID]		= "CP10 MRC/VMRS",
800 	[ESR_ELx_EC_PAC]		= "PAC",
801 	[ESR_ELx_EC_CP14_64]		= "CP14 MCRR/MRRC",
802 	[ESR_ELx_EC_BTI]		= "BTI",
803 	[ESR_ELx_EC_ILL]		= "PSTATE.IL",
804 	[ESR_ELx_EC_SVC32]		= "SVC (AArch32)",
805 	[ESR_ELx_EC_HVC32]		= "HVC (AArch32)",
806 	[ESR_ELx_EC_SMC32]		= "SMC (AArch32)",
807 	[ESR_ELx_EC_SVC64]		= "SVC (AArch64)",
808 	[ESR_ELx_EC_HVC64]		= "HVC (AArch64)",
809 	[ESR_ELx_EC_SMC64]		= "SMC (AArch64)",
810 	[ESR_ELx_EC_SYS64]		= "MSR/MRS (AArch64)",
811 	[ESR_ELx_EC_SVE]		= "SVE",
812 	[ESR_ELx_EC_ERET]		= "ERET/ERETAA/ERETAB",
813 	[ESR_ELx_EC_FPAC]		= "FPAC",
814 	[ESR_ELx_EC_SME]		= "SME",
815 	[ESR_ELx_EC_IMP_DEF]		= "EL3 IMP DEF",
816 	[ESR_ELx_EC_IABT_LOW]		= "IABT (lower EL)",
817 	[ESR_ELx_EC_IABT_CUR]		= "IABT (current EL)",
818 	[ESR_ELx_EC_PC_ALIGN]		= "PC Alignment",
819 	[ESR_ELx_EC_DABT_LOW]		= "DABT (lower EL)",
820 	[ESR_ELx_EC_DABT_CUR]		= "DABT (current EL)",
821 	[ESR_ELx_EC_SP_ALIGN]		= "SP Alignment",
822 	[ESR_ELx_EC_FP_EXC32]		= "FP (AArch32)",
823 	[ESR_ELx_EC_FP_EXC64]		= "FP (AArch64)",
824 	[ESR_ELx_EC_SERROR]		= "SError",
825 	[ESR_ELx_EC_BREAKPT_LOW]	= "Breakpoint (lower EL)",
826 	[ESR_ELx_EC_BREAKPT_CUR]	= "Breakpoint (current EL)",
827 	[ESR_ELx_EC_SOFTSTP_LOW]	= "Software Step (lower EL)",
828 	[ESR_ELx_EC_SOFTSTP_CUR]	= "Software Step (current EL)",
829 	[ESR_ELx_EC_WATCHPT_LOW]	= "Watchpoint (lower EL)",
830 	[ESR_ELx_EC_WATCHPT_CUR]	= "Watchpoint (current EL)",
831 	[ESR_ELx_EC_BKPT32]		= "BKPT (AArch32)",
832 	[ESR_ELx_EC_VECTOR32]		= "Vector catch (AArch32)",
833 	[ESR_ELx_EC_BRK64]		= "BRK (AArch64)",
834 };
835 
836 const char *esr_get_class_string(unsigned long esr)
837 {
838 	return esr_class_str[ESR_ELx_EC(esr)];
839 }
840 
841 /*
842  * bad_el0_sync handles unexpected, but potentially recoverable synchronous
843  * exceptions taken from EL0.
844  */
845 void bad_el0_sync(struct pt_regs *regs, int reason, unsigned long esr)
846 {
847 	unsigned long pc = instruction_pointer(regs);
848 
849 	current->thread.fault_address = 0;
850 	current->thread.fault_code = esr;
851 
852 	arm64_force_sig_fault(SIGILL, ILL_ILLOPC, pc,
853 			      "Bad EL0 synchronous exception");
854 }
855 
856 #ifdef CONFIG_VMAP_STACK
857 
858 DEFINE_PER_CPU(unsigned long [OVERFLOW_STACK_SIZE/sizeof(long)], overflow_stack)
859 	__aligned(16);
860 
861 void panic_bad_stack(struct pt_regs *regs, unsigned long esr, unsigned long far)
862 {
863 	unsigned long tsk_stk = (unsigned long)current->stack;
864 	unsigned long irq_stk = (unsigned long)this_cpu_read(irq_stack_ptr);
865 	unsigned long ovf_stk = (unsigned long)this_cpu_ptr(overflow_stack);
866 
867 	console_verbose();
868 	pr_emerg("Insufficient stack space to handle exception!");
869 
870 	pr_emerg("ESR: 0x%016lx -- %s\n", esr, esr_get_class_string(esr));
871 	pr_emerg("FAR: 0x%016lx\n", far);
872 
873 	pr_emerg("Task stack:     [0x%016lx..0x%016lx]\n",
874 		 tsk_stk, tsk_stk + THREAD_SIZE);
875 	pr_emerg("IRQ stack:      [0x%016lx..0x%016lx]\n",
876 		 irq_stk, irq_stk + IRQ_STACK_SIZE);
877 	pr_emerg("Overflow stack: [0x%016lx..0x%016lx]\n",
878 		 ovf_stk, ovf_stk + OVERFLOW_STACK_SIZE);
879 
880 	__show_regs(regs);
881 
882 	/*
883 	 * We use nmi_panic to limit the potential for recusive overflows, and
884 	 * to get a better stack trace.
885 	 */
886 	nmi_panic(NULL, "kernel stack overflow");
887 	cpu_park_loop();
888 }
889 #endif
890 
891 void __noreturn arm64_serror_panic(struct pt_regs *regs, unsigned long esr)
892 {
893 	console_verbose();
894 
895 	pr_crit("SError Interrupt on CPU%d, code 0x%016lx -- %s\n",
896 		smp_processor_id(), esr, esr_get_class_string(esr));
897 	if (regs)
898 		__show_regs(regs);
899 
900 	nmi_panic(regs, "Asynchronous SError Interrupt");
901 
902 	cpu_park_loop();
903 	unreachable();
904 }
905 
906 bool arm64_is_fatal_ras_serror(struct pt_regs *regs, unsigned long esr)
907 {
908 	unsigned long aet = arm64_ras_serror_get_severity(esr);
909 
910 	switch (aet) {
911 	case ESR_ELx_AET_CE:	/* corrected error */
912 	case ESR_ELx_AET_UEO:	/* restartable, not yet consumed */
913 		/*
914 		 * The CPU can make progress. We may take UEO again as
915 		 * a more severe error.
916 		 */
917 		return false;
918 
919 	case ESR_ELx_AET_UEU:	/* Uncorrected Unrecoverable */
920 	case ESR_ELx_AET_UER:	/* Uncorrected Recoverable */
921 		/*
922 		 * The CPU can't make progress. The exception may have
923 		 * been imprecise.
924 		 *
925 		 * Neoverse-N1 #1349291 means a non-KVM SError reported as
926 		 * Unrecoverable should be treated as Uncontainable. We
927 		 * call arm64_serror_panic() in both cases.
928 		 */
929 		return true;
930 
931 	case ESR_ELx_AET_UC:	/* Uncontainable or Uncategorized error */
932 	default:
933 		/* Error has been silently propagated */
934 		arm64_serror_panic(regs, esr);
935 	}
936 }
937 
938 void do_serror(struct pt_regs *regs, unsigned long esr)
939 {
940 	/* non-RAS errors are not containable */
941 	if (!arm64_is_ras_serror(esr) || arm64_is_fatal_ras_serror(regs, esr))
942 		arm64_serror_panic(regs, esr);
943 }
944 
945 /* GENERIC_BUG traps */
946 
947 int is_valid_bugaddr(unsigned long addr)
948 {
949 	/*
950 	 * bug_handler() only called for BRK #BUG_BRK_IMM.
951 	 * So the answer is trivial -- any spurious instances with no
952 	 * bug table entry will be rejected by report_bug() and passed
953 	 * back to the debug-monitors code and handled as a fatal
954 	 * unexpected debug exception.
955 	 */
956 	return 1;
957 }
958 
959 static int bug_handler(struct pt_regs *regs, unsigned long esr)
960 {
961 	switch (report_bug(regs->pc, regs)) {
962 	case BUG_TRAP_TYPE_BUG:
963 		die("Oops - BUG", regs, esr);
964 		break;
965 
966 	case BUG_TRAP_TYPE_WARN:
967 		break;
968 
969 	default:
970 		/* unknown/unrecognised bug trap type */
971 		return DBG_HOOK_ERROR;
972 	}
973 
974 	/* If thread survives, skip over the BUG instruction and continue: */
975 	arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
976 	return DBG_HOOK_HANDLED;
977 }
978 
979 static struct break_hook bug_break_hook = {
980 	.fn = bug_handler,
981 	.imm = BUG_BRK_IMM,
982 };
983 
984 #ifdef CONFIG_CFI_CLANG
985 static int cfi_handler(struct pt_regs *regs, unsigned long esr)
986 {
987 	unsigned long target;
988 	u32 type;
989 
990 	target = pt_regs_read_reg(regs, FIELD_GET(CFI_BRK_IMM_TARGET, esr));
991 	type = (u32)pt_regs_read_reg(regs, FIELD_GET(CFI_BRK_IMM_TYPE, esr));
992 
993 	switch (report_cfi_failure(regs, regs->pc, &target, type)) {
994 	case BUG_TRAP_TYPE_BUG:
995 		die("Oops - CFI", regs, 0);
996 		break;
997 
998 	case BUG_TRAP_TYPE_WARN:
999 		break;
1000 
1001 	default:
1002 		return DBG_HOOK_ERROR;
1003 	}
1004 
1005 	arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
1006 	return DBG_HOOK_HANDLED;
1007 }
1008 
1009 static struct break_hook cfi_break_hook = {
1010 	.fn = cfi_handler,
1011 	.imm = CFI_BRK_IMM_BASE,
1012 	.mask = CFI_BRK_IMM_MASK,
1013 };
1014 #endif /* CONFIG_CFI_CLANG */
1015 
1016 static int reserved_fault_handler(struct pt_regs *regs, unsigned long esr)
1017 {
1018 	pr_err("%s generated an invalid instruction at %pS!\n",
1019 		"Kernel text patching",
1020 		(void *)instruction_pointer(regs));
1021 
1022 	/* We cannot handle this */
1023 	return DBG_HOOK_ERROR;
1024 }
1025 
1026 static struct break_hook fault_break_hook = {
1027 	.fn = reserved_fault_handler,
1028 	.imm = FAULT_BRK_IMM,
1029 };
1030 
1031 #ifdef CONFIG_KASAN_SW_TAGS
1032 
1033 #define KASAN_ESR_RECOVER	0x20
1034 #define KASAN_ESR_WRITE	0x10
1035 #define KASAN_ESR_SIZE_MASK	0x0f
1036 #define KASAN_ESR_SIZE(esr)	(1 << ((esr) & KASAN_ESR_SIZE_MASK))
1037 
1038 static int kasan_handler(struct pt_regs *regs, unsigned long esr)
1039 {
1040 	bool recover = esr & KASAN_ESR_RECOVER;
1041 	bool write = esr & KASAN_ESR_WRITE;
1042 	size_t size = KASAN_ESR_SIZE(esr);
1043 	u64 addr = regs->regs[0];
1044 	u64 pc = regs->pc;
1045 
1046 	kasan_report(addr, size, write, pc);
1047 
1048 	/*
1049 	 * The instrumentation allows to control whether we can proceed after
1050 	 * a crash was detected. This is done by passing the -recover flag to
1051 	 * the compiler. Disabling recovery allows to generate more compact
1052 	 * code.
1053 	 *
1054 	 * Unfortunately disabling recovery doesn't work for the kernel right
1055 	 * now. KASAN reporting is disabled in some contexts (for example when
1056 	 * the allocator accesses slab object metadata; this is controlled by
1057 	 * current->kasan_depth). All these accesses are detected by the tool,
1058 	 * even though the reports for them are not printed.
1059 	 *
1060 	 * This is something that might be fixed at some point in the future.
1061 	 */
1062 	if (!recover)
1063 		die("Oops - KASAN", regs, esr);
1064 
1065 	/* If thread survives, skip over the brk instruction and continue: */
1066 	arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
1067 	return DBG_HOOK_HANDLED;
1068 }
1069 
1070 static struct break_hook kasan_break_hook = {
1071 	.fn	= kasan_handler,
1072 	.imm	= KASAN_BRK_IMM,
1073 	.mask	= KASAN_BRK_MASK,
1074 };
1075 #endif
1076 
1077 
1078 #define esr_comment(esr) ((esr) & ESR_ELx_BRK64_ISS_COMMENT_MASK)
1079 
1080 /*
1081  * Initial handler for AArch64 BRK exceptions
1082  * This handler only used until debug_traps_init().
1083  */
1084 int __init early_brk64(unsigned long addr, unsigned long esr,
1085 		struct pt_regs *regs)
1086 {
1087 #ifdef CONFIG_CFI_CLANG
1088 	if ((esr_comment(esr) & ~CFI_BRK_IMM_MASK) == CFI_BRK_IMM_BASE)
1089 		return cfi_handler(regs, esr) != DBG_HOOK_HANDLED;
1090 #endif
1091 #ifdef CONFIG_KASAN_SW_TAGS
1092 	if ((esr_comment(esr) & ~KASAN_BRK_MASK) == KASAN_BRK_IMM)
1093 		return kasan_handler(regs, esr) != DBG_HOOK_HANDLED;
1094 #endif
1095 	return bug_handler(regs, esr) != DBG_HOOK_HANDLED;
1096 }
1097 
1098 void __init trap_init(void)
1099 {
1100 	register_kernel_break_hook(&bug_break_hook);
1101 #ifdef CONFIG_CFI_CLANG
1102 	register_kernel_break_hook(&cfi_break_hook);
1103 #endif
1104 	register_kernel_break_hook(&fault_break_hook);
1105 #ifdef CONFIG_KASAN_SW_TAGS
1106 	register_kernel_break_hook(&kasan_break_hook);
1107 #endif
1108 	debug_traps_init();
1109 }
1110