xref: /openbmc/linux/arch/arm64/kernel/traps.c (revision 3c8c1539)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Based on arch/arm/kernel/traps.c
4  *
5  * Copyright (C) 1995-2009 Russell King
6  * Copyright (C) 2012 ARM Ltd.
7  */
8 
9 #include <linux/bug.h>
10 #include <linux/context_tracking.h>
11 #include <linux/signal.h>
12 #include <linux/personality.h>
13 #include <linux/kallsyms.h>
14 #include <linux/kprobes.h>
15 #include <linux/spinlock.h>
16 #include <linux/uaccess.h>
17 #include <linux/hardirq.h>
18 #include <linux/kdebug.h>
19 #include <linux/module.h>
20 #include <linux/kexec.h>
21 #include <linux/delay.h>
22 #include <linux/init.h>
23 #include <linux/sched/signal.h>
24 #include <linux/sched/debug.h>
25 #include <linux/sched/task_stack.h>
26 #include <linux/sizes.h>
27 #include <linux/syscalls.h>
28 #include <linux/mm_types.h>
29 #include <linux/kasan.h>
30 
31 #include <asm/atomic.h>
32 #include <asm/bug.h>
33 #include <asm/cpufeature.h>
34 #include <asm/daifflags.h>
35 #include <asm/debug-monitors.h>
36 #include <asm/esr.h>
37 #include <asm/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, int err, struct pt_regs *regs)
185 {
186 	static int die_counter;
187 	int ret;
188 
189 	pr_emerg("Internal error: %s: %x [#%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, int 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 		do_exit(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 int 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%08x, ", 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 		      int 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 	void __user *pc = (void __user *)instruction_pointer(regs);
404 
405 	if (!user_mode(regs)) {
406 		__le32 instr_le;
407 		if (get_kernel_nofault(instr_le, (__force __le32 *)pc))
408 			goto exit;
409 		instr = le32_to_cpu(instr_le);
410 	} else if (compat_thumb_mode(regs)) {
411 		/* 16-bit Thumb instruction */
412 		__le16 instr_le;
413 		if (get_user(instr_le, (__le16 __user *)pc))
414 			goto exit;
415 		instr = le16_to_cpu(instr_le);
416 		if (aarch32_insn_is_wide(instr)) {
417 			u32 instr2;
418 
419 			if (get_user(instr_le, (__le16 __user *)(pc + 2)))
420 				goto exit;
421 			instr2 = le16_to_cpu(instr_le);
422 			instr = (instr << 16) | instr2;
423 		}
424 	} else {
425 		/* 32-bit ARM instruction */
426 		__le32 instr_le;
427 		if (get_user(instr_le, (__le32 __user *)pc))
428 			goto exit;
429 		instr = le32_to_cpu(instr_le);
430 	}
431 
432 	raw_spin_lock_irqsave(&undef_lock, flags);
433 	list_for_each_entry(hook, &undef_hook, node)
434 		if ((instr & hook->instr_mask) == hook->instr_val &&
435 			(regs->pstate & hook->pstate_mask) == hook->pstate_val)
436 			fn = hook->fn;
437 
438 	raw_spin_unlock_irqrestore(&undef_lock, flags);
439 exit:
440 	return fn ? fn(regs, instr) : 1;
441 }
442 
443 void force_signal_inject(int signal, int code, unsigned long address, unsigned int err)
444 {
445 	const char *desc;
446 	struct pt_regs *regs = current_pt_regs();
447 
448 	if (WARN_ON(!user_mode(regs)))
449 		return;
450 
451 	switch (signal) {
452 	case SIGILL:
453 		desc = "undefined instruction";
454 		break;
455 	case SIGSEGV:
456 		desc = "illegal memory access";
457 		break;
458 	default:
459 		desc = "unknown or unrecoverable error";
460 		break;
461 	}
462 
463 	/* Force signals we don't understand to SIGKILL */
464 	if (WARN_ON(signal != SIGKILL &&
465 		    siginfo_layout(signal, code) != SIL_FAULT)) {
466 		signal = SIGKILL;
467 	}
468 
469 	arm64_notify_die(desc, regs, signal, code, address, err);
470 }
471 
472 /*
473  * Set up process info to signal segmentation fault - called on access error.
474  */
475 void arm64_notify_segfault(unsigned long addr)
476 {
477 	int code;
478 
479 	mmap_read_lock(current->mm);
480 	if (find_vma(current->mm, untagged_addr(addr)) == NULL)
481 		code = SEGV_MAPERR;
482 	else
483 		code = SEGV_ACCERR;
484 	mmap_read_unlock(current->mm);
485 
486 	force_signal_inject(SIGSEGV, code, addr, 0);
487 }
488 
489 void do_undefinstr(struct pt_regs *regs)
490 {
491 	/* check for AArch32 breakpoint instructions */
492 	if (!aarch32_break_handler(regs))
493 		return;
494 
495 	if (call_undef_hook(regs) == 0)
496 		return;
497 
498 	BUG_ON(!user_mode(regs));
499 	force_signal_inject(SIGILL, ILL_ILLOPC, regs->pc, 0);
500 }
501 NOKPROBE_SYMBOL(do_undefinstr);
502 
503 void do_bti(struct pt_regs *regs)
504 {
505 	BUG_ON(!user_mode(regs));
506 	force_signal_inject(SIGILL, ILL_ILLOPC, regs->pc, 0);
507 }
508 NOKPROBE_SYMBOL(do_bti);
509 
510 void do_ptrauth_fault(struct pt_regs *regs, unsigned int esr)
511 {
512 	/*
513 	 * Unexpected FPAC exception or pointer authentication failure in
514 	 * the kernel: kill the task before it does any more harm.
515 	 */
516 	BUG_ON(!user_mode(regs));
517 	force_signal_inject(SIGILL, ILL_ILLOPN, regs->pc, esr);
518 }
519 NOKPROBE_SYMBOL(do_ptrauth_fault);
520 
521 #define __user_cache_maint(insn, address, res)			\
522 	if (address >= user_addr_max()) {			\
523 		res = -EFAULT;					\
524 	} else {						\
525 		uaccess_ttbr0_enable();				\
526 		asm volatile (					\
527 			"1:	" insn ", %1\n"			\
528 			"	mov	%w0, #0\n"		\
529 			"2:\n"					\
530 			"	.pushsection .fixup,\"ax\"\n"	\
531 			"	.align	2\n"			\
532 			"3:	mov	%w0, %w2\n"		\
533 			"	b	2b\n"			\
534 			"	.popsection\n"			\
535 			_ASM_EXTABLE(1b, 3b)			\
536 			: "=r" (res)				\
537 			: "r" (address), "i" (-EFAULT));	\
538 		uaccess_ttbr0_disable();			\
539 	}
540 
541 static void user_cache_maint_handler(unsigned int esr, struct pt_regs *regs)
542 {
543 	unsigned long tagged_address, address;
544 	int rt = ESR_ELx_SYS64_ISS_RT(esr);
545 	int crm = (esr & ESR_ELx_SYS64_ISS_CRM_MASK) >> ESR_ELx_SYS64_ISS_CRM_SHIFT;
546 	int ret = 0;
547 
548 	tagged_address = pt_regs_read_reg(regs, rt);
549 	address = untagged_addr(tagged_address);
550 
551 	switch (crm) {
552 	case ESR_ELx_SYS64_ISS_CRM_DC_CVAU:	/* DC CVAU, gets promoted */
553 		__user_cache_maint("dc civac", address, ret);
554 		break;
555 	case ESR_ELx_SYS64_ISS_CRM_DC_CVAC:	/* DC CVAC, gets promoted */
556 		__user_cache_maint("dc civac", address, ret);
557 		break;
558 	case ESR_ELx_SYS64_ISS_CRM_DC_CVADP:	/* DC CVADP */
559 		__user_cache_maint("sys 3, c7, c13, 1", address, ret);
560 		break;
561 	case ESR_ELx_SYS64_ISS_CRM_DC_CVAP:	/* DC CVAP */
562 		__user_cache_maint("sys 3, c7, c12, 1", address, ret);
563 		break;
564 	case ESR_ELx_SYS64_ISS_CRM_DC_CIVAC:	/* DC CIVAC */
565 		__user_cache_maint("dc civac", address, ret);
566 		break;
567 	case ESR_ELx_SYS64_ISS_CRM_IC_IVAU:	/* IC IVAU */
568 		__user_cache_maint("ic ivau", address, ret);
569 		break;
570 	default:
571 		force_signal_inject(SIGILL, ILL_ILLOPC, regs->pc, 0);
572 		return;
573 	}
574 
575 	if (ret)
576 		arm64_notify_segfault(tagged_address);
577 	else
578 		arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
579 }
580 
581 static void ctr_read_handler(unsigned int esr, struct pt_regs *regs)
582 {
583 	int rt = ESR_ELx_SYS64_ISS_RT(esr);
584 	unsigned long val = arm64_ftr_reg_user_value(&arm64_ftr_reg_ctrel0);
585 
586 	if (cpus_have_const_cap(ARM64_WORKAROUND_1542419)) {
587 		/* Hide DIC so that we can trap the unnecessary maintenance...*/
588 		val &= ~BIT(CTR_DIC_SHIFT);
589 
590 		/* ... and fake IminLine to reduce the number of traps. */
591 		val &= ~CTR_IMINLINE_MASK;
592 		val |= (PAGE_SHIFT - 2) & CTR_IMINLINE_MASK;
593 	}
594 
595 	pt_regs_write_reg(regs, rt, val);
596 
597 	arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
598 }
599 
600 static void cntvct_read_handler(unsigned int esr, struct pt_regs *regs)
601 {
602 	int rt = ESR_ELx_SYS64_ISS_RT(esr);
603 
604 	pt_regs_write_reg(regs, rt, arch_timer_read_counter());
605 	arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
606 }
607 
608 static void cntfrq_read_handler(unsigned int 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_get_rate());
613 	arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
614 }
615 
616 static void mrs_handler(unsigned int esr, struct pt_regs *regs)
617 {
618 	u32 sysreg, rt;
619 
620 	rt = ESR_ELx_SYS64_ISS_RT(esr);
621 	sysreg = esr_sys64_to_sysreg(esr);
622 
623 	if (do_emulate_mrs(regs, sysreg, rt) != 0)
624 		force_signal_inject(SIGILL, ILL_ILLOPC, regs->pc, 0);
625 }
626 
627 static void wfi_handler(unsigned int esr, struct pt_regs *regs)
628 {
629 	arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
630 }
631 
632 struct sys64_hook {
633 	unsigned int esr_mask;
634 	unsigned int esr_val;
635 	void (*handler)(unsigned int esr, struct pt_regs *regs);
636 };
637 
638 static const struct sys64_hook sys64_hooks[] = {
639 	{
640 		.esr_mask = ESR_ELx_SYS64_ISS_EL0_CACHE_OP_MASK,
641 		.esr_val = ESR_ELx_SYS64_ISS_EL0_CACHE_OP_VAL,
642 		.handler = user_cache_maint_handler,
643 	},
644 	{
645 		/* Trap read access to CTR_EL0 */
646 		.esr_mask = ESR_ELx_SYS64_ISS_SYS_OP_MASK,
647 		.esr_val = ESR_ELx_SYS64_ISS_SYS_CTR_READ,
648 		.handler = ctr_read_handler,
649 	},
650 	{
651 		/* Trap read access to CNTVCT_EL0 */
652 		.esr_mask = ESR_ELx_SYS64_ISS_SYS_OP_MASK,
653 		.esr_val = ESR_ELx_SYS64_ISS_SYS_CNTVCT,
654 		.handler = cntvct_read_handler,
655 	},
656 	{
657 		/* Trap read access to CNTFRQ_EL0 */
658 		.esr_mask = ESR_ELx_SYS64_ISS_SYS_OP_MASK,
659 		.esr_val = ESR_ELx_SYS64_ISS_SYS_CNTFRQ,
660 		.handler = cntfrq_read_handler,
661 	},
662 	{
663 		/* Trap read access to CPUID registers */
664 		.esr_mask = ESR_ELx_SYS64_ISS_SYS_MRS_OP_MASK,
665 		.esr_val = ESR_ELx_SYS64_ISS_SYS_MRS_OP_VAL,
666 		.handler = mrs_handler,
667 	},
668 	{
669 		/* Trap WFI instructions executed in userspace */
670 		.esr_mask = ESR_ELx_WFx_MASK,
671 		.esr_val = ESR_ELx_WFx_WFI_VAL,
672 		.handler = wfi_handler,
673 	},
674 	{},
675 };
676 
677 #ifdef CONFIG_COMPAT
678 static bool cp15_cond_valid(unsigned int esr, struct pt_regs *regs)
679 {
680 	int cond;
681 
682 	/* Only a T32 instruction can trap without CV being set */
683 	if (!(esr & ESR_ELx_CV)) {
684 		u32 it;
685 
686 		it = compat_get_it_state(regs);
687 		if (!it)
688 			return true;
689 
690 		cond = it >> 4;
691 	} else {
692 		cond = (esr & ESR_ELx_COND_MASK) >> ESR_ELx_COND_SHIFT;
693 	}
694 
695 	return aarch32_opcode_cond_checks[cond](regs->pstate);
696 }
697 
698 static void compat_cntfrq_read_handler(unsigned int esr, struct pt_regs *regs)
699 {
700 	int reg = (esr & ESR_ELx_CP15_32_ISS_RT_MASK) >> ESR_ELx_CP15_32_ISS_RT_SHIFT;
701 
702 	pt_regs_write_reg(regs, reg, arch_timer_get_rate());
703 	arm64_skip_faulting_instruction(regs, 4);
704 }
705 
706 static const struct sys64_hook cp15_32_hooks[] = {
707 	{
708 		.esr_mask = ESR_ELx_CP15_32_ISS_SYS_MASK,
709 		.esr_val = ESR_ELx_CP15_32_ISS_SYS_CNTFRQ,
710 		.handler = compat_cntfrq_read_handler,
711 	},
712 	{},
713 };
714 
715 static void compat_cntvct_read_handler(unsigned int esr, struct pt_regs *regs)
716 {
717 	int rt = (esr & ESR_ELx_CP15_64_ISS_RT_MASK) >> ESR_ELx_CP15_64_ISS_RT_SHIFT;
718 	int rt2 = (esr & ESR_ELx_CP15_64_ISS_RT2_MASK) >> ESR_ELx_CP15_64_ISS_RT2_SHIFT;
719 	u64 val = arch_timer_read_counter();
720 
721 	pt_regs_write_reg(regs, rt, lower_32_bits(val));
722 	pt_regs_write_reg(regs, rt2, upper_32_bits(val));
723 	arm64_skip_faulting_instruction(regs, 4);
724 }
725 
726 static const struct sys64_hook cp15_64_hooks[] = {
727 	{
728 		.esr_mask = ESR_ELx_CP15_64_ISS_SYS_MASK,
729 		.esr_val = ESR_ELx_CP15_64_ISS_SYS_CNTVCT,
730 		.handler = compat_cntvct_read_handler,
731 	},
732 	{},
733 };
734 
735 void do_cp15instr(unsigned int esr, struct pt_regs *regs)
736 {
737 	const struct sys64_hook *hook, *hook_base;
738 
739 	if (!cp15_cond_valid(esr, regs)) {
740 		/*
741 		 * There is no T16 variant of a CP access, so we
742 		 * always advance PC by 4 bytes.
743 		 */
744 		arm64_skip_faulting_instruction(regs, 4);
745 		return;
746 	}
747 
748 	switch (ESR_ELx_EC(esr)) {
749 	case ESR_ELx_EC_CP15_32:
750 		hook_base = cp15_32_hooks;
751 		break;
752 	case ESR_ELx_EC_CP15_64:
753 		hook_base = cp15_64_hooks;
754 		break;
755 	default:
756 		do_undefinstr(regs);
757 		return;
758 	}
759 
760 	for (hook = hook_base; hook->handler; hook++)
761 		if ((hook->esr_mask & esr) == hook->esr_val) {
762 			hook->handler(esr, regs);
763 			return;
764 		}
765 
766 	/*
767 	 * New cp15 instructions may previously have been undefined at
768 	 * EL0. Fall back to our usual undefined instruction handler
769 	 * so that we handle these consistently.
770 	 */
771 	do_undefinstr(regs);
772 }
773 NOKPROBE_SYMBOL(do_cp15instr);
774 #endif
775 
776 void do_sysinstr(unsigned int esr, struct pt_regs *regs)
777 {
778 	const struct sys64_hook *hook;
779 
780 	for (hook = sys64_hooks; hook->handler; hook++)
781 		if ((hook->esr_mask & esr) == hook->esr_val) {
782 			hook->handler(esr, regs);
783 			return;
784 		}
785 
786 	/*
787 	 * New SYS instructions may previously have been undefined at EL0. Fall
788 	 * back to our usual undefined instruction handler so that we handle
789 	 * these consistently.
790 	 */
791 	do_undefinstr(regs);
792 }
793 NOKPROBE_SYMBOL(do_sysinstr);
794 
795 static const char *esr_class_str[] = {
796 	[0 ... ESR_ELx_EC_MAX]		= "UNRECOGNIZED EC",
797 	[ESR_ELx_EC_UNKNOWN]		= "Unknown/Uncategorized",
798 	[ESR_ELx_EC_WFx]		= "WFI/WFE",
799 	[ESR_ELx_EC_CP15_32]		= "CP15 MCR/MRC",
800 	[ESR_ELx_EC_CP15_64]		= "CP15 MCRR/MRRC",
801 	[ESR_ELx_EC_CP14_MR]		= "CP14 MCR/MRC",
802 	[ESR_ELx_EC_CP14_LS]		= "CP14 LDC/STC",
803 	[ESR_ELx_EC_FP_ASIMD]		= "ASIMD",
804 	[ESR_ELx_EC_CP10_ID]		= "CP10 MRC/VMRS",
805 	[ESR_ELx_EC_PAC]		= "PAC",
806 	[ESR_ELx_EC_CP14_64]		= "CP14 MCRR/MRRC",
807 	[ESR_ELx_EC_BTI]		= "BTI",
808 	[ESR_ELx_EC_ILL]		= "PSTATE.IL",
809 	[ESR_ELx_EC_SVC32]		= "SVC (AArch32)",
810 	[ESR_ELx_EC_HVC32]		= "HVC (AArch32)",
811 	[ESR_ELx_EC_SMC32]		= "SMC (AArch32)",
812 	[ESR_ELx_EC_SVC64]		= "SVC (AArch64)",
813 	[ESR_ELx_EC_HVC64]		= "HVC (AArch64)",
814 	[ESR_ELx_EC_SMC64]		= "SMC (AArch64)",
815 	[ESR_ELx_EC_SYS64]		= "MSR/MRS (AArch64)",
816 	[ESR_ELx_EC_SVE]		= "SVE",
817 	[ESR_ELx_EC_ERET]		= "ERET/ERETAA/ERETAB",
818 	[ESR_ELx_EC_FPAC]		= "FPAC",
819 	[ESR_ELx_EC_IMP_DEF]		= "EL3 IMP DEF",
820 	[ESR_ELx_EC_IABT_LOW]		= "IABT (lower EL)",
821 	[ESR_ELx_EC_IABT_CUR]		= "IABT (current EL)",
822 	[ESR_ELx_EC_PC_ALIGN]		= "PC Alignment",
823 	[ESR_ELx_EC_DABT_LOW]		= "DABT (lower EL)",
824 	[ESR_ELx_EC_DABT_CUR]		= "DABT (current EL)",
825 	[ESR_ELx_EC_SP_ALIGN]		= "SP Alignment",
826 	[ESR_ELx_EC_FP_EXC32]		= "FP (AArch32)",
827 	[ESR_ELx_EC_FP_EXC64]		= "FP (AArch64)",
828 	[ESR_ELx_EC_SERROR]		= "SError",
829 	[ESR_ELx_EC_BREAKPT_LOW]	= "Breakpoint (lower EL)",
830 	[ESR_ELx_EC_BREAKPT_CUR]	= "Breakpoint (current EL)",
831 	[ESR_ELx_EC_SOFTSTP_LOW]	= "Software Step (lower EL)",
832 	[ESR_ELx_EC_SOFTSTP_CUR]	= "Software Step (current EL)",
833 	[ESR_ELx_EC_WATCHPT_LOW]	= "Watchpoint (lower EL)",
834 	[ESR_ELx_EC_WATCHPT_CUR]	= "Watchpoint (current EL)",
835 	[ESR_ELx_EC_BKPT32]		= "BKPT (AArch32)",
836 	[ESR_ELx_EC_VECTOR32]		= "Vector catch (AArch32)",
837 	[ESR_ELx_EC_BRK64]		= "BRK (AArch64)",
838 };
839 
840 const char *esr_get_class_string(u32 esr)
841 {
842 	return esr_class_str[ESR_ELx_EC(esr)];
843 }
844 
845 /*
846  * bad_el0_sync handles unexpected, but potentially recoverable synchronous
847  * exceptions taken from EL0.
848  */
849 void bad_el0_sync(struct pt_regs *regs, int reason, unsigned int esr)
850 {
851 	unsigned long pc = instruction_pointer(regs);
852 
853 	current->thread.fault_address = 0;
854 	current->thread.fault_code = esr;
855 
856 	arm64_force_sig_fault(SIGILL, ILL_ILLOPC, pc,
857 			      "Bad EL0 synchronous exception");
858 }
859 
860 #ifdef CONFIG_VMAP_STACK
861 
862 DEFINE_PER_CPU(unsigned long [OVERFLOW_STACK_SIZE/sizeof(long)], overflow_stack)
863 	__aligned(16);
864 
865 void panic_bad_stack(struct pt_regs *regs, unsigned int esr, unsigned long far)
866 {
867 	unsigned long tsk_stk = (unsigned long)current->stack;
868 	unsigned long irq_stk = (unsigned long)this_cpu_read(irq_stack_ptr);
869 	unsigned long ovf_stk = (unsigned long)this_cpu_ptr(overflow_stack);
870 
871 	console_verbose();
872 	pr_emerg("Insufficient stack space to handle exception!");
873 
874 	pr_emerg("ESR: 0x%08x -- %s\n", esr, esr_get_class_string(esr));
875 	pr_emerg("FAR: 0x%016lx\n", far);
876 
877 	pr_emerg("Task stack:     [0x%016lx..0x%016lx]\n",
878 		 tsk_stk, tsk_stk + THREAD_SIZE);
879 	pr_emerg("IRQ stack:      [0x%016lx..0x%016lx]\n",
880 		 irq_stk, irq_stk + IRQ_STACK_SIZE);
881 	pr_emerg("Overflow stack: [0x%016lx..0x%016lx]\n",
882 		 ovf_stk, ovf_stk + OVERFLOW_STACK_SIZE);
883 
884 	__show_regs(regs);
885 
886 	/*
887 	 * We use nmi_panic to limit the potential for recusive overflows, and
888 	 * to get a better stack trace.
889 	 */
890 	nmi_panic(NULL, "kernel stack overflow");
891 	cpu_park_loop();
892 }
893 #endif
894 
895 void __noreturn arm64_serror_panic(struct pt_regs *regs, u32 esr)
896 {
897 	console_verbose();
898 
899 	pr_crit("SError Interrupt on CPU%d, code 0x%08x -- %s\n",
900 		smp_processor_id(), esr, esr_get_class_string(esr));
901 	if (regs)
902 		__show_regs(regs);
903 
904 	nmi_panic(regs, "Asynchronous SError Interrupt");
905 
906 	cpu_park_loop();
907 	unreachable();
908 }
909 
910 bool arm64_is_fatal_ras_serror(struct pt_regs *regs, unsigned int esr)
911 {
912 	u32 aet = arm64_ras_serror_get_severity(esr);
913 
914 	switch (aet) {
915 	case ESR_ELx_AET_CE:	/* corrected error */
916 	case ESR_ELx_AET_UEO:	/* restartable, not yet consumed */
917 		/*
918 		 * The CPU can make progress. We may take UEO again as
919 		 * a more severe error.
920 		 */
921 		return false;
922 
923 	case ESR_ELx_AET_UEU:	/* Uncorrected Unrecoverable */
924 	case ESR_ELx_AET_UER:	/* Uncorrected Recoverable */
925 		/*
926 		 * The CPU can't make progress. The exception may have
927 		 * been imprecise.
928 		 *
929 		 * Neoverse-N1 #1349291 means a non-KVM SError reported as
930 		 * Unrecoverable should be treated as Uncontainable. We
931 		 * call arm64_serror_panic() in both cases.
932 		 */
933 		return true;
934 
935 	case ESR_ELx_AET_UC:	/* Uncontainable or Uncategorized error */
936 	default:
937 		/* Error has been silently propagated */
938 		arm64_serror_panic(regs, esr);
939 	}
940 }
941 
942 void do_serror(struct pt_regs *regs, unsigned int esr)
943 {
944 	/* non-RAS errors are not containable */
945 	if (!arm64_is_ras_serror(esr) || arm64_is_fatal_ras_serror(regs, esr))
946 		arm64_serror_panic(regs, esr);
947 }
948 
949 /* GENERIC_BUG traps */
950 
951 int is_valid_bugaddr(unsigned long addr)
952 {
953 	/*
954 	 * bug_handler() only called for BRK #BUG_BRK_IMM.
955 	 * So the answer is trivial -- any spurious instances with no
956 	 * bug table entry will be rejected by report_bug() and passed
957 	 * back to the debug-monitors code and handled as a fatal
958 	 * unexpected debug exception.
959 	 */
960 	return 1;
961 }
962 
963 static int bug_handler(struct pt_regs *regs, unsigned int esr)
964 {
965 	switch (report_bug(regs->pc, regs)) {
966 	case BUG_TRAP_TYPE_BUG:
967 		die("Oops - BUG", regs, 0);
968 		break;
969 
970 	case BUG_TRAP_TYPE_WARN:
971 		break;
972 
973 	default:
974 		/* unknown/unrecognised bug trap type */
975 		return DBG_HOOK_ERROR;
976 	}
977 
978 	/* If thread survives, skip over the BUG instruction and continue: */
979 	arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
980 	return DBG_HOOK_HANDLED;
981 }
982 
983 static struct break_hook bug_break_hook = {
984 	.fn = bug_handler,
985 	.imm = BUG_BRK_IMM,
986 };
987 
988 static int reserved_fault_handler(struct pt_regs *regs, unsigned int esr)
989 {
990 	pr_err("%s generated an invalid instruction at %pS!\n",
991 		in_bpf_jit(regs) ? "BPF JIT" : "Kernel text patching",
992 		(void *)instruction_pointer(regs));
993 
994 	/* We cannot handle this */
995 	return DBG_HOOK_ERROR;
996 }
997 
998 static struct break_hook fault_break_hook = {
999 	.fn = reserved_fault_handler,
1000 	.imm = FAULT_BRK_IMM,
1001 };
1002 
1003 #ifdef CONFIG_KASAN_SW_TAGS
1004 
1005 #define KASAN_ESR_RECOVER	0x20
1006 #define KASAN_ESR_WRITE	0x10
1007 #define KASAN_ESR_SIZE_MASK	0x0f
1008 #define KASAN_ESR_SIZE(esr)	(1 << ((esr) & KASAN_ESR_SIZE_MASK))
1009 
1010 static int kasan_handler(struct pt_regs *regs, unsigned int esr)
1011 {
1012 	bool recover = esr & KASAN_ESR_RECOVER;
1013 	bool write = esr & KASAN_ESR_WRITE;
1014 	size_t size = KASAN_ESR_SIZE(esr);
1015 	u64 addr = regs->regs[0];
1016 	u64 pc = regs->pc;
1017 
1018 	kasan_report(addr, size, write, pc);
1019 
1020 	/*
1021 	 * The instrumentation allows to control whether we can proceed after
1022 	 * a crash was detected. This is done by passing the -recover flag to
1023 	 * the compiler. Disabling recovery allows to generate more compact
1024 	 * code.
1025 	 *
1026 	 * Unfortunately disabling recovery doesn't work for the kernel right
1027 	 * now. KASAN reporting is disabled in some contexts (for example when
1028 	 * the allocator accesses slab object metadata; this is controlled by
1029 	 * current->kasan_depth). All these accesses are detected by the tool,
1030 	 * even though the reports for them are not printed.
1031 	 *
1032 	 * This is something that might be fixed at some point in the future.
1033 	 */
1034 	if (!recover)
1035 		die("Oops - KASAN", regs, 0);
1036 
1037 	/* If thread survives, skip over the brk instruction and continue: */
1038 	arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
1039 	return DBG_HOOK_HANDLED;
1040 }
1041 
1042 static struct break_hook kasan_break_hook = {
1043 	.fn	= kasan_handler,
1044 	.imm	= KASAN_BRK_IMM,
1045 	.mask	= KASAN_BRK_MASK,
1046 };
1047 #endif
1048 
1049 /*
1050  * Initial handler for AArch64 BRK exceptions
1051  * This handler only used until debug_traps_init().
1052  */
1053 int __init early_brk64(unsigned long addr, unsigned int esr,
1054 		struct pt_regs *regs)
1055 {
1056 #ifdef CONFIG_KASAN_SW_TAGS
1057 	unsigned int comment = esr & ESR_ELx_BRK64_ISS_COMMENT_MASK;
1058 
1059 	if ((comment & ~KASAN_BRK_MASK) == KASAN_BRK_IMM)
1060 		return kasan_handler(regs, esr) != DBG_HOOK_HANDLED;
1061 #endif
1062 	return bug_handler(regs, esr) != DBG_HOOK_HANDLED;
1063 }
1064 
1065 void __init trap_init(void)
1066 {
1067 	register_kernel_break_hook(&bug_break_hook);
1068 	register_kernel_break_hook(&fault_break_hook);
1069 #ifdef CONFIG_KASAN_SW_TAGS
1070 	register_kernel_break_hook(&kasan_break_hook);
1071 #endif
1072 	debug_traps_init();
1073 }
1074