xref: /openbmc/linux/arch/arm64/kernel/traps.c (revision 3a83e4e6)
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/insn.h>
38 #include <asm/kprobes.h>
39 #include <asm/traps.h>
40 #include <asm/smp.h>
41 #include <asm/stack_pointer.h>
42 #include <asm/stacktrace.h>
43 #include <asm/exception.h>
44 #include <asm/system_misc.h>
45 #include <asm/sysreg.h>
46 
47 static const char *handler[]= {
48 	"Synchronous Abort",
49 	"IRQ",
50 	"FIQ",
51 	"Error"
52 };
53 
54 int show_unhandled_signals = 0;
55 
56 static void dump_backtrace_entry(unsigned long where, const char *loglvl)
57 {
58 	printk("%s %pS\n", loglvl, (void *)where);
59 }
60 
61 static void dump_kernel_instr(const char *lvl, struct pt_regs *regs)
62 {
63 	unsigned long addr = instruction_pointer(regs);
64 	char str[sizeof("00000000 ") * 5 + 2 + 1], *p = str;
65 	int i;
66 
67 	if (user_mode(regs))
68 		return;
69 
70 	for (i = -4; i < 1; i++) {
71 		unsigned int val, bad;
72 
73 		bad = aarch64_insn_read(&((u32 *)addr)[i], &val);
74 
75 		if (!bad)
76 			p += sprintf(p, i == 0 ? "(%08x) " : "%08x ", val);
77 		else {
78 			p += sprintf(p, "bad PC value");
79 			break;
80 		}
81 	}
82 
83 	printk("%sCode: %s\n", lvl, str);
84 }
85 
86 void dump_backtrace(struct pt_regs *regs, struct task_struct *tsk,
87 		    const char *loglvl)
88 {
89 	struct stackframe frame;
90 	int skip = 0;
91 
92 	pr_debug("%s(regs = %p tsk = %p)\n", __func__, regs, tsk);
93 
94 	if (regs) {
95 		if (user_mode(regs))
96 			return;
97 		skip = 1;
98 	}
99 
100 	if (!tsk)
101 		tsk = current;
102 
103 	if (!try_get_task_stack(tsk))
104 		return;
105 
106 	if (tsk == current) {
107 		start_backtrace(&frame,
108 				(unsigned long)__builtin_frame_address(0),
109 				(unsigned long)dump_backtrace);
110 	} else {
111 		/*
112 		 * task blocked in __switch_to
113 		 */
114 		start_backtrace(&frame,
115 				thread_saved_fp(tsk),
116 				thread_saved_pc(tsk));
117 	}
118 
119 	printk("%sCall trace:\n", loglvl);
120 	do {
121 		/* skip until specified stack frame */
122 		if (!skip) {
123 			dump_backtrace_entry(frame.pc, loglvl);
124 		} else if (frame.fp == regs->regs[29]) {
125 			skip = 0;
126 			/*
127 			 * Mostly, this is the case where this function is
128 			 * called in panic/abort. As exception handler's
129 			 * stack frame does not contain the corresponding pc
130 			 * at which an exception has taken place, use regs->pc
131 			 * instead.
132 			 */
133 			dump_backtrace_entry(regs->pc, loglvl);
134 		}
135 	} while (!unwind_frame(tsk, &frame));
136 
137 	put_task_stack(tsk);
138 }
139 
140 void show_stack(struct task_struct *tsk, unsigned long *sp, const char *loglvl)
141 {
142 	dump_backtrace(NULL, tsk, loglvl);
143 	barrier();
144 }
145 
146 #ifdef CONFIG_PREEMPT
147 #define S_PREEMPT " PREEMPT"
148 #elif defined(CONFIG_PREEMPT_RT)
149 #define S_PREEMPT " PREEMPT_RT"
150 #else
151 #define S_PREEMPT ""
152 #endif
153 
154 #define S_SMP " SMP"
155 
156 static int __die(const char *str, int err, struct pt_regs *regs)
157 {
158 	static int die_counter;
159 	int ret;
160 
161 	pr_emerg("Internal error: %s: %x [#%d]" S_PREEMPT S_SMP "\n",
162 		 str, err, ++die_counter);
163 
164 	/* trap and error numbers are mostly meaningless on ARM */
165 	ret = notify_die(DIE_OOPS, str, regs, err, 0, SIGSEGV);
166 	if (ret == NOTIFY_STOP)
167 		return ret;
168 
169 	print_modules();
170 	show_regs(regs);
171 
172 	dump_kernel_instr(KERN_EMERG, regs);
173 
174 	return ret;
175 }
176 
177 static DEFINE_RAW_SPINLOCK(die_lock);
178 
179 /*
180  * This function is protected against re-entrancy.
181  */
182 void die(const char *str, struct pt_regs *regs, int err)
183 {
184 	int ret;
185 	unsigned long flags;
186 
187 	raw_spin_lock_irqsave(&die_lock, flags);
188 
189 	oops_enter();
190 
191 	console_verbose();
192 	bust_spinlocks(1);
193 	ret = __die(str, err, regs);
194 
195 	if (regs && kexec_should_crash(current))
196 		crash_kexec(regs);
197 
198 	bust_spinlocks(0);
199 	add_taint(TAINT_DIE, LOCKDEP_NOW_UNRELIABLE);
200 	oops_exit();
201 
202 	if (in_interrupt())
203 		panic("Fatal exception in interrupt");
204 	if (panic_on_oops)
205 		panic("Fatal exception");
206 
207 	raw_spin_unlock_irqrestore(&die_lock, flags);
208 
209 	if (ret != NOTIFY_STOP)
210 		do_exit(SIGSEGV);
211 }
212 
213 static void arm64_show_signal(int signo, const char *str)
214 {
215 	static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,
216 				      DEFAULT_RATELIMIT_BURST);
217 	struct task_struct *tsk = current;
218 	unsigned int esr = tsk->thread.fault_code;
219 	struct pt_regs *regs = task_pt_regs(tsk);
220 
221 	/* Leave if the signal won't be shown */
222 	if (!show_unhandled_signals ||
223 	    !unhandled_signal(tsk, signo) ||
224 	    !__ratelimit(&rs))
225 		return;
226 
227 	pr_info("%s[%d]: unhandled exception: ", tsk->comm, task_pid_nr(tsk));
228 	if (esr)
229 		pr_cont("%s, ESR 0x%08x, ", esr_get_class_string(esr), esr);
230 
231 	pr_cont("%s", str);
232 	print_vma_addr(KERN_CONT " in ", regs->pc);
233 	pr_cont("\n");
234 	__show_regs(regs);
235 }
236 
237 void arm64_force_sig_fault(int signo, int code, void __user *addr,
238 			   const char *str)
239 {
240 	arm64_show_signal(signo, str);
241 	if (signo == SIGKILL)
242 		force_sig(SIGKILL);
243 	else
244 		force_sig_fault(signo, code, addr);
245 }
246 
247 void arm64_force_sig_mceerr(int code, void __user *addr, short lsb,
248 			    const char *str)
249 {
250 	arm64_show_signal(SIGBUS, str);
251 	force_sig_mceerr(code, addr, lsb);
252 }
253 
254 void arm64_force_sig_ptrace_errno_trap(int errno, void __user *addr,
255 				       const char *str)
256 {
257 	arm64_show_signal(SIGTRAP, str);
258 	force_sig_ptrace_errno_trap(errno, addr);
259 }
260 
261 void arm64_notify_die(const char *str, struct pt_regs *regs,
262 		      int signo, int sicode, void __user *addr,
263 		      int err)
264 {
265 	if (user_mode(regs)) {
266 		WARN_ON(regs != current_pt_regs());
267 		current->thread.fault_address = 0;
268 		current->thread.fault_code = err;
269 
270 		arm64_force_sig_fault(signo, sicode, addr, str);
271 	} else {
272 		die(str, regs, err);
273 	}
274 }
275 
276 #ifdef CONFIG_COMPAT
277 #define PSTATE_IT_1_0_SHIFT	25
278 #define PSTATE_IT_1_0_MASK	(0x3 << PSTATE_IT_1_0_SHIFT)
279 #define PSTATE_IT_7_2_SHIFT	10
280 #define PSTATE_IT_7_2_MASK	(0x3f << PSTATE_IT_7_2_SHIFT)
281 
282 static u32 compat_get_it_state(struct pt_regs *regs)
283 {
284 	u32 it, pstate = regs->pstate;
285 
286 	it  = (pstate & PSTATE_IT_1_0_MASK) >> PSTATE_IT_1_0_SHIFT;
287 	it |= ((pstate & PSTATE_IT_7_2_MASK) >> PSTATE_IT_7_2_SHIFT) << 2;
288 
289 	return it;
290 }
291 
292 static void compat_set_it_state(struct pt_regs *regs, u32 it)
293 {
294 	u32 pstate_it;
295 
296 	pstate_it  = (it << PSTATE_IT_1_0_SHIFT) & PSTATE_IT_1_0_MASK;
297 	pstate_it |= ((it >> 2) << PSTATE_IT_7_2_SHIFT) & PSTATE_IT_7_2_MASK;
298 
299 	regs->pstate &= ~PSR_AA32_IT_MASK;
300 	regs->pstate |= pstate_it;
301 }
302 
303 static void advance_itstate(struct pt_regs *regs)
304 {
305 	u32 it;
306 
307 	/* ARM mode */
308 	if (!(regs->pstate & PSR_AA32_T_BIT) ||
309 	    !(regs->pstate & PSR_AA32_IT_MASK))
310 		return;
311 
312 	it  = compat_get_it_state(regs);
313 
314 	/*
315 	 * If this is the last instruction of the block, wipe the IT
316 	 * state. Otherwise advance it.
317 	 */
318 	if (!(it & 7))
319 		it = 0;
320 	else
321 		it = (it & 0xe0) | ((it << 1) & 0x1f);
322 
323 	compat_set_it_state(regs, it);
324 }
325 #else
326 static void advance_itstate(struct pt_regs *regs)
327 {
328 }
329 #endif
330 
331 void arm64_skip_faulting_instruction(struct pt_regs *regs, unsigned long size)
332 {
333 	regs->pc += size;
334 
335 	/*
336 	 * If we were single stepping, we want to get the step exception after
337 	 * we return from the trap.
338 	 */
339 	if (user_mode(regs))
340 		user_fastforward_single_step(current);
341 
342 	if (compat_user_mode(regs))
343 		advance_itstate(regs);
344 	else
345 		regs->pstate &= ~PSR_BTYPE_MASK;
346 }
347 
348 static LIST_HEAD(undef_hook);
349 static DEFINE_RAW_SPINLOCK(undef_lock);
350 
351 void register_undef_hook(struct undef_hook *hook)
352 {
353 	unsigned long flags;
354 
355 	raw_spin_lock_irqsave(&undef_lock, flags);
356 	list_add(&hook->node, &undef_hook);
357 	raw_spin_unlock_irqrestore(&undef_lock, flags);
358 }
359 
360 void unregister_undef_hook(struct undef_hook *hook)
361 {
362 	unsigned long flags;
363 
364 	raw_spin_lock_irqsave(&undef_lock, flags);
365 	list_del(&hook->node);
366 	raw_spin_unlock_irqrestore(&undef_lock, flags);
367 }
368 
369 static int call_undef_hook(struct pt_regs *regs)
370 {
371 	struct undef_hook *hook;
372 	unsigned long flags;
373 	u32 instr;
374 	int (*fn)(struct pt_regs *regs, u32 instr) = NULL;
375 	void __user *pc = (void __user *)instruction_pointer(regs);
376 
377 	if (!user_mode(regs)) {
378 		__le32 instr_le;
379 		if (get_kernel_nofault(instr_le, (__force __le32 *)pc))
380 			goto exit;
381 		instr = le32_to_cpu(instr_le);
382 	} else if (compat_thumb_mode(regs)) {
383 		/* 16-bit Thumb instruction */
384 		__le16 instr_le;
385 		if (get_user(instr_le, (__le16 __user *)pc))
386 			goto exit;
387 		instr = le16_to_cpu(instr_le);
388 		if (aarch32_insn_is_wide(instr)) {
389 			u32 instr2;
390 
391 			if (get_user(instr_le, (__le16 __user *)(pc + 2)))
392 				goto exit;
393 			instr2 = le16_to_cpu(instr_le);
394 			instr = (instr << 16) | instr2;
395 		}
396 	} else {
397 		/* 32-bit ARM instruction */
398 		__le32 instr_le;
399 		if (get_user(instr_le, (__le32 __user *)pc))
400 			goto exit;
401 		instr = le32_to_cpu(instr_le);
402 	}
403 
404 	raw_spin_lock_irqsave(&undef_lock, flags);
405 	list_for_each_entry(hook, &undef_hook, node)
406 		if ((instr & hook->instr_mask) == hook->instr_val &&
407 			(regs->pstate & hook->pstate_mask) == hook->pstate_val)
408 			fn = hook->fn;
409 
410 	raw_spin_unlock_irqrestore(&undef_lock, flags);
411 exit:
412 	return fn ? fn(regs, instr) : 1;
413 }
414 
415 void force_signal_inject(int signal, int code, unsigned long address)
416 {
417 	const char *desc;
418 	struct pt_regs *regs = current_pt_regs();
419 
420 	if (WARN_ON(!user_mode(regs)))
421 		return;
422 
423 	switch (signal) {
424 	case SIGILL:
425 		desc = "undefined instruction";
426 		break;
427 	case SIGSEGV:
428 		desc = "illegal memory access";
429 		break;
430 	default:
431 		desc = "unknown or unrecoverable error";
432 		break;
433 	}
434 
435 	/* Force signals we don't understand to SIGKILL */
436 	if (WARN_ON(signal != SIGKILL &&
437 		    siginfo_layout(signal, code) != SIL_FAULT)) {
438 		signal = SIGKILL;
439 	}
440 
441 	arm64_notify_die(desc, regs, signal, code, (void __user *)address, 0);
442 }
443 
444 /*
445  * Set up process info to signal segmentation fault - called on access error.
446  */
447 void arm64_notify_segfault(unsigned long addr)
448 {
449 	int code;
450 
451 	mmap_read_lock(current->mm);
452 	if (find_vma(current->mm, addr) == NULL)
453 		code = SEGV_MAPERR;
454 	else
455 		code = SEGV_ACCERR;
456 	mmap_read_unlock(current->mm);
457 
458 	force_signal_inject(SIGSEGV, code, addr);
459 }
460 
461 void do_undefinstr(struct pt_regs *regs)
462 {
463 	/* check for AArch32 breakpoint instructions */
464 	if (!aarch32_break_handler(regs))
465 		return;
466 
467 	if (call_undef_hook(regs) == 0)
468 		return;
469 
470 	BUG_ON(!user_mode(regs));
471 	force_signal_inject(SIGILL, ILL_ILLOPC, regs->pc);
472 }
473 NOKPROBE_SYMBOL(do_undefinstr);
474 
475 void do_bti(struct pt_regs *regs)
476 {
477 	BUG_ON(!user_mode(regs));
478 	force_signal_inject(SIGILL, ILL_ILLOPC, regs->pc);
479 }
480 NOKPROBE_SYMBOL(do_bti);
481 
482 #define __user_cache_maint(insn, address, res)			\
483 	if (address >= user_addr_max()) {			\
484 		res = -EFAULT;					\
485 	} else {						\
486 		uaccess_ttbr0_enable();				\
487 		asm volatile (					\
488 			"1:	" insn ", %1\n"			\
489 			"	mov	%w0, #0\n"		\
490 			"2:\n"					\
491 			"	.pushsection .fixup,\"ax\"\n"	\
492 			"	.align	2\n"			\
493 			"3:	mov	%w0, %w2\n"		\
494 			"	b	2b\n"			\
495 			"	.popsection\n"			\
496 			_ASM_EXTABLE(1b, 3b)			\
497 			: "=r" (res)				\
498 			: "r" (address), "i" (-EFAULT));	\
499 		uaccess_ttbr0_disable();			\
500 	}
501 
502 static void user_cache_maint_handler(unsigned int esr, struct pt_regs *regs)
503 {
504 	unsigned long address;
505 	int rt = ESR_ELx_SYS64_ISS_RT(esr);
506 	int crm = (esr & ESR_ELx_SYS64_ISS_CRM_MASK) >> ESR_ELx_SYS64_ISS_CRM_SHIFT;
507 	int ret = 0;
508 
509 	address = untagged_addr(pt_regs_read_reg(regs, rt));
510 
511 	switch (crm) {
512 	case ESR_ELx_SYS64_ISS_CRM_DC_CVAU:	/* DC CVAU, gets promoted */
513 		__user_cache_maint("dc civac", address, ret);
514 		break;
515 	case ESR_ELx_SYS64_ISS_CRM_DC_CVAC:	/* DC CVAC, gets promoted */
516 		__user_cache_maint("dc civac", address, ret);
517 		break;
518 	case ESR_ELx_SYS64_ISS_CRM_DC_CVADP:	/* DC CVADP */
519 		__user_cache_maint("sys 3, c7, c13, 1", address, ret);
520 		break;
521 	case ESR_ELx_SYS64_ISS_CRM_DC_CVAP:	/* DC CVAP */
522 		__user_cache_maint("sys 3, c7, c12, 1", address, ret);
523 		break;
524 	case ESR_ELx_SYS64_ISS_CRM_DC_CIVAC:	/* DC CIVAC */
525 		__user_cache_maint("dc civac", address, ret);
526 		break;
527 	case ESR_ELx_SYS64_ISS_CRM_IC_IVAU:	/* IC IVAU */
528 		__user_cache_maint("ic ivau", address, ret);
529 		break;
530 	default:
531 		force_signal_inject(SIGILL, ILL_ILLOPC, regs->pc);
532 		return;
533 	}
534 
535 	if (ret)
536 		arm64_notify_segfault(address);
537 	else
538 		arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
539 }
540 
541 static void ctr_read_handler(unsigned int esr, struct pt_regs *regs)
542 {
543 	int rt = ESR_ELx_SYS64_ISS_RT(esr);
544 	unsigned long val = arm64_ftr_reg_user_value(&arm64_ftr_reg_ctrel0);
545 
546 	if (cpus_have_const_cap(ARM64_WORKAROUND_1542419)) {
547 		/* Hide DIC so that we can trap the unnecessary maintenance...*/
548 		val &= ~BIT(CTR_DIC_SHIFT);
549 
550 		/* ... and fake IminLine to reduce the number of traps. */
551 		val &= ~CTR_IMINLINE_MASK;
552 		val |= (PAGE_SHIFT - 2) & CTR_IMINLINE_MASK;
553 	}
554 
555 	pt_regs_write_reg(regs, rt, val);
556 
557 	arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
558 }
559 
560 static void cntvct_read_handler(unsigned int esr, struct pt_regs *regs)
561 {
562 	int rt = ESR_ELx_SYS64_ISS_RT(esr);
563 
564 	pt_regs_write_reg(regs, rt, arch_timer_read_counter());
565 	arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
566 }
567 
568 static void cntfrq_read_handler(unsigned int esr, struct pt_regs *regs)
569 {
570 	int rt = ESR_ELx_SYS64_ISS_RT(esr);
571 
572 	pt_regs_write_reg(regs, rt, arch_timer_get_rate());
573 	arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
574 }
575 
576 static void mrs_handler(unsigned int esr, struct pt_regs *regs)
577 {
578 	u32 sysreg, rt;
579 
580 	rt = ESR_ELx_SYS64_ISS_RT(esr);
581 	sysreg = esr_sys64_to_sysreg(esr);
582 
583 	if (do_emulate_mrs(regs, sysreg, rt) != 0)
584 		force_signal_inject(SIGILL, ILL_ILLOPC, regs->pc);
585 }
586 
587 static void wfi_handler(unsigned int esr, struct pt_regs *regs)
588 {
589 	arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
590 }
591 
592 struct sys64_hook {
593 	unsigned int esr_mask;
594 	unsigned int esr_val;
595 	void (*handler)(unsigned int esr, struct pt_regs *regs);
596 };
597 
598 static const struct sys64_hook sys64_hooks[] = {
599 	{
600 		.esr_mask = ESR_ELx_SYS64_ISS_EL0_CACHE_OP_MASK,
601 		.esr_val = ESR_ELx_SYS64_ISS_EL0_CACHE_OP_VAL,
602 		.handler = user_cache_maint_handler,
603 	},
604 	{
605 		/* Trap read access to CTR_EL0 */
606 		.esr_mask = ESR_ELx_SYS64_ISS_SYS_OP_MASK,
607 		.esr_val = ESR_ELx_SYS64_ISS_SYS_CTR_READ,
608 		.handler = ctr_read_handler,
609 	},
610 	{
611 		/* Trap read access to CNTVCT_EL0 */
612 		.esr_mask = ESR_ELx_SYS64_ISS_SYS_OP_MASK,
613 		.esr_val = ESR_ELx_SYS64_ISS_SYS_CNTVCT,
614 		.handler = cntvct_read_handler,
615 	},
616 	{
617 		/* Trap read access to CNTFRQ_EL0 */
618 		.esr_mask = ESR_ELx_SYS64_ISS_SYS_OP_MASK,
619 		.esr_val = ESR_ELx_SYS64_ISS_SYS_CNTFRQ,
620 		.handler = cntfrq_read_handler,
621 	},
622 	{
623 		/* Trap read access to CPUID registers */
624 		.esr_mask = ESR_ELx_SYS64_ISS_SYS_MRS_OP_MASK,
625 		.esr_val = ESR_ELx_SYS64_ISS_SYS_MRS_OP_VAL,
626 		.handler = mrs_handler,
627 	},
628 	{
629 		/* Trap WFI instructions executed in userspace */
630 		.esr_mask = ESR_ELx_WFx_MASK,
631 		.esr_val = ESR_ELx_WFx_WFI_VAL,
632 		.handler = wfi_handler,
633 	},
634 	{},
635 };
636 
637 #ifdef CONFIG_COMPAT
638 static bool cp15_cond_valid(unsigned int esr, struct pt_regs *regs)
639 {
640 	int cond;
641 
642 	/* Only a T32 instruction can trap without CV being set */
643 	if (!(esr & ESR_ELx_CV)) {
644 		u32 it;
645 
646 		it = compat_get_it_state(regs);
647 		if (!it)
648 			return true;
649 
650 		cond = it >> 4;
651 	} else {
652 		cond = (esr & ESR_ELx_COND_MASK) >> ESR_ELx_COND_SHIFT;
653 	}
654 
655 	return aarch32_opcode_cond_checks[cond](regs->pstate);
656 }
657 
658 static void compat_cntfrq_read_handler(unsigned int esr, struct pt_regs *regs)
659 {
660 	int reg = (esr & ESR_ELx_CP15_32_ISS_RT_MASK) >> ESR_ELx_CP15_32_ISS_RT_SHIFT;
661 
662 	pt_regs_write_reg(regs, reg, arch_timer_get_rate());
663 	arm64_skip_faulting_instruction(regs, 4);
664 }
665 
666 static const struct sys64_hook cp15_32_hooks[] = {
667 	{
668 		.esr_mask = ESR_ELx_CP15_32_ISS_SYS_MASK,
669 		.esr_val = ESR_ELx_CP15_32_ISS_SYS_CNTFRQ,
670 		.handler = compat_cntfrq_read_handler,
671 	},
672 	{},
673 };
674 
675 static void compat_cntvct_read_handler(unsigned int esr, struct pt_regs *regs)
676 {
677 	int rt = (esr & ESR_ELx_CP15_64_ISS_RT_MASK) >> ESR_ELx_CP15_64_ISS_RT_SHIFT;
678 	int rt2 = (esr & ESR_ELx_CP15_64_ISS_RT2_MASK) >> ESR_ELx_CP15_64_ISS_RT2_SHIFT;
679 	u64 val = arch_timer_read_counter();
680 
681 	pt_regs_write_reg(regs, rt, lower_32_bits(val));
682 	pt_regs_write_reg(regs, rt2, upper_32_bits(val));
683 	arm64_skip_faulting_instruction(regs, 4);
684 }
685 
686 static const struct sys64_hook cp15_64_hooks[] = {
687 	{
688 		.esr_mask = ESR_ELx_CP15_64_ISS_SYS_MASK,
689 		.esr_val = ESR_ELx_CP15_64_ISS_SYS_CNTVCT,
690 		.handler = compat_cntvct_read_handler,
691 	},
692 	{},
693 };
694 
695 void do_cp15instr(unsigned int esr, struct pt_regs *regs)
696 {
697 	const struct sys64_hook *hook, *hook_base;
698 
699 	if (!cp15_cond_valid(esr, regs)) {
700 		/*
701 		 * There is no T16 variant of a CP access, so we
702 		 * always advance PC by 4 bytes.
703 		 */
704 		arm64_skip_faulting_instruction(regs, 4);
705 		return;
706 	}
707 
708 	switch (ESR_ELx_EC(esr)) {
709 	case ESR_ELx_EC_CP15_32:
710 		hook_base = cp15_32_hooks;
711 		break;
712 	case ESR_ELx_EC_CP15_64:
713 		hook_base = cp15_64_hooks;
714 		break;
715 	default:
716 		do_undefinstr(regs);
717 		return;
718 	}
719 
720 	for (hook = hook_base; hook->handler; hook++)
721 		if ((hook->esr_mask & esr) == hook->esr_val) {
722 			hook->handler(esr, regs);
723 			return;
724 		}
725 
726 	/*
727 	 * New cp15 instructions may previously have been undefined at
728 	 * EL0. Fall back to our usual undefined instruction handler
729 	 * so that we handle these consistently.
730 	 */
731 	do_undefinstr(regs);
732 }
733 NOKPROBE_SYMBOL(do_cp15instr);
734 #endif
735 
736 void do_sysinstr(unsigned int esr, struct pt_regs *regs)
737 {
738 	const struct sys64_hook *hook;
739 
740 	for (hook = sys64_hooks; hook->handler; hook++)
741 		if ((hook->esr_mask & esr) == hook->esr_val) {
742 			hook->handler(esr, regs);
743 			return;
744 		}
745 
746 	/*
747 	 * New SYS instructions may previously have been undefined at EL0. Fall
748 	 * back to our usual undefined instruction handler so that we handle
749 	 * these consistently.
750 	 */
751 	do_undefinstr(regs);
752 }
753 NOKPROBE_SYMBOL(do_sysinstr);
754 
755 static const char *esr_class_str[] = {
756 	[0 ... ESR_ELx_EC_MAX]		= "UNRECOGNIZED EC",
757 	[ESR_ELx_EC_UNKNOWN]		= "Unknown/Uncategorized",
758 	[ESR_ELx_EC_WFx]		= "WFI/WFE",
759 	[ESR_ELx_EC_CP15_32]		= "CP15 MCR/MRC",
760 	[ESR_ELx_EC_CP15_64]		= "CP15 MCRR/MRRC",
761 	[ESR_ELx_EC_CP14_MR]		= "CP14 MCR/MRC",
762 	[ESR_ELx_EC_CP14_LS]		= "CP14 LDC/STC",
763 	[ESR_ELx_EC_FP_ASIMD]		= "ASIMD",
764 	[ESR_ELx_EC_CP10_ID]		= "CP10 MRC/VMRS",
765 	[ESR_ELx_EC_PAC]		= "PAC",
766 	[ESR_ELx_EC_CP14_64]		= "CP14 MCRR/MRRC",
767 	[ESR_ELx_EC_BTI]		= "BTI",
768 	[ESR_ELx_EC_ILL]		= "PSTATE.IL",
769 	[ESR_ELx_EC_SVC32]		= "SVC (AArch32)",
770 	[ESR_ELx_EC_HVC32]		= "HVC (AArch32)",
771 	[ESR_ELx_EC_SMC32]		= "SMC (AArch32)",
772 	[ESR_ELx_EC_SVC64]		= "SVC (AArch64)",
773 	[ESR_ELx_EC_HVC64]		= "HVC (AArch64)",
774 	[ESR_ELx_EC_SMC64]		= "SMC (AArch64)",
775 	[ESR_ELx_EC_SYS64]		= "MSR/MRS (AArch64)",
776 	[ESR_ELx_EC_SVE]		= "SVE",
777 	[ESR_ELx_EC_ERET]		= "ERET/ERETAA/ERETAB",
778 	[ESR_ELx_EC_IMP_DEF]		= "EL3 IMP DEF",
779 	[ESR_ELx_EC_IABT_LOW]		= "IABT (lower EL)",
780 	[ESR_ELx_EC_IABT_CUR]		= "IABT (current EL)",
781 	[ESR_ELx_EC_PC_ALIGN]		= "PC Alignment",
782 	[ESR_ELx_EC_DABT_LOW]		= "DABT (lower EL)",
783 	[ESR_ELx_EC_DABT_CUR]		= "DABT (current EL)",
784 	[ESR_ELx_EC_SP_ALIGN]		= "SP Alignment",
785 	[ESR_ELx_EC_FP_EXC32]		= "FP (AArch32)",
786 	[ESR_ELx_EC_FP_EXC64]		= "FP (AArch64)",
787 	[ESR_ELx_EC_SERROR]		= "SError",
788 	[ESR_ELx_EC_BREAKPT_LOW]	= "Breakpoint (lower EL)",
789 	[ESR_ELx_EC_BREAKPT_CUR]	= "Breakpoint (current EL)",
790 	[ESR_ELx_EC_SOFTSTP_LOW]	= "Software Step (lower EL)",
791 	[ESR_ELx_EC_SOFTSTP_CUR]	= "Software Step (current EL)",
792 	[ESR_ELx_EC_WATCHPT_LOW]	= "Watchpoint (lower EL)",
793 	[ESR_ELx_EC_WATCHPT_CUR]	= "Watchpoint (current EL)",
794 	[ESR_ELx_EC_BKPT32]		= "BKPT (AArch32)",
795 	[ESR_ELx_EC_VECTOR32]		= "Vector catch (AArch32)",
796 	[ESR_ELx_EC_BRK64]		= "BRK (AArch64)",
797 };
798 
799 const char *esr_get_class_string(u32 esr)
800 {
801 	return esr_class_str[ESR_ELx_EC(esr)];
802 }
803 
804 /*
805  * bad_mode handles the impossible case in the exception vector. This is always
806  * fatal.
807  */
808 asmlinkage void bad_mode(struct pt_regs *regs, int reason, unsigned int esr)
809 {
810 	console_verbose();
811 
812 	pr_crit("Bad mode in %s handler detected on CPU%d, code 0x%08x -- %s\n",
813 		handler[reason], smp_processor_id(), esr,
814 		esr_get_class_string(esr));
815 
816 	__show_regs(regs);
817 	local_daif_mask();
818 	panic("bad mode");
819 }
820 
821 /*
822  * bad_el0_sync handles unexpected, but potentially recoverable synchronous
823  * exceptions taken from EL0. Unlike bad_mode, this returns.
824  */
825 void bad_el0_sync(struct pt_regs *regs, int reason, unsigned int esr)
826 {
827 	void __user *pc = (void __user *)instruction_pointer(regs);
828 
829 	current->thread.fault_address = 0;
830 	current->thread.fault_code = esr;
831 
832 	arm64_force_sig_fault(SIGILL, ILL_ILLOPC, pc,
833 			      "Bad EL0 synchronous exception");
834 }
835 
836 #ifdef CONFIG_VMAP_STACK
837 
838 DEFINE_PER_CPU(unsigned long [OVERFLOW_STACK_SIZE/sizeof(long)], overflow_stack)
839 	__aligned(16);
840 
841 asmlinkage void handle_bad_stack(struct pt_regs *regs)
842 {
843 	unsigned long tsk_stk = (unsigned long)current->stack;
844 	unsigned long irq_stk = (unsigned long)this_cpu_read(irq_stack_ptr);
845 	unsigned long ovf_stk = (unsigned long)this_cpu_ptr(overflow_stack);
846 	unsigned int esr = read_sysreg(esr_el1);
847 	unsigned long far = read_sysreg(far_el1);
848 
849 	console_verbose();
850 	pr_emerg("Insufficient stack space to handle exception!");
851 
852 	pr_emerg("ESR: 0x%08x -- %s\n", esr, esr_get_class_string(esr));
853 	pr_emerg("FAR: 0x%016lx\n", far);
854 
855 	pr_emerg("Task stack:     [0x%016lx..0x%016lx]\n",
856 		 tsk_stk, tsk_stk + THREAD_SIZE);
857 	pr_emerg("IRQ stack:      [0x%016lx..0x%016lx]\n",
858 		 irq_stk, irq_stk + IRQ_STACK_SIZE);
859 	pr_emerg("Overflow stack: [0x%016lx..0x%016lx]\n",
860 		 ovf_stk, ovf_stk + OVERFLOW_STACK_SIZE);
861 
862 	__show_regs(regs);
863 
864 	/*
865 	 * We use nmi_panic to limit the potential for recusive overflows, and
866 	 * to get a better stack trace.
867 	 */
868 	nmi_panic(NULL, "kernel stack overflow");
869 	cpu_park_loop();
870 }
871 #endif
872 
873 void __noreturn arm64_serror_panic(struct pt_regs *regs, u32 esr)
874 {
875 	console_verbose();
876 
877 	pr_crit("SError Interrupt on CPU%d, code 0x%08x -- %s\n",
878 		smp_processor_id(), esr, esr_get_class_string(esr));
879 	if (regs)
880 		__show_regs(regs);
881 
882 	nmi_panic(regs, "Asynchronous SError Interrupt");
883 
884 	cpu_park_loop();
885 	unreachable();
886 }
887 
888 bool arm64_is_fatal_ras_serror(struct pt_regs *regs, unsigned int esr)
889 {
890 	u32 aet = arm64_ras_serror_get_severity(esr);
891 
892 	switch (aet) {
893 	case ESR_ELx_AET_CE:	/* corrected error */
894 	case ESR_ELx_AET_UEO:	/* restartable, not yet consumed */
895 		/*
896 		 * The CPU can make progress. We may take UEO again as
897 		 * a more severe error.
898 		 */
899 		return false;
900 
901 	case ESR_ELx_AET_UEU:	/* Uncorrected Unrecoverable */
902 	case ESR_ELx_AET_UER:	/* Uncorrected Recoverable */
903 		/*
904 		 * The CPU can't make progress. The exception may have
905 		 * been imprecise.
906 		 *
907 		 * Neoverse-N1 #1349291 means a non-KVM SError reported as
908 		 * Unrecoverable should be treated as Uncontainable. We
909 		 * call arm64_serror_panic() in both cases.
910 		 */
911 		return true;
912 
913 	case ESR_ELx_AET_UC:	/* Uncontainable or Uncategorized error */
914 	default:
915 		/* Error has been silently propagated */
916 		arm64_serror_panic(regs, esr);
917 	}
918 }
919 
920 asmlinkage void do_serror(struct pt_regs *regs, unsigned int esr)
921 {
922 	nmi_enter();
923 
924 	/* non-RAS errors are not containable */
925 	if (!arm64_is_ras_serror(esr) || arm64_is_fatal_ras_serror(regs, esr))
926 		arm64_serror_panic(regs, esr);
927 
928 	nmi_exit();
929 }
930 
931 asmlinkage void enter_from_user_mode(void)
932 {
933 	CT_WARN_ON(ct_state() != CONTEXT_USER);
934 	user_exit_irqoff();
935 }
936 NOKPROBE_SYMBOL(enter_from_user_mode);
937 
938 void __pte_error(const char *file, int line, unsigned long val)
939 {
940 	pr_err("%s:%d: bad pte %016lx.\n", file, line, val);
941 }
942 
943 void __pmd_error(const char *file, int line, unsigned long val)
944 {
945 	pr_err("%s:%d: bad pmd %016lx.\n", file, line, val);
946 }
947 
948 void __pud_error(const char *file, int line, unsigned long val)
949 {
950 	pr_err("%s:%d: bad pud %016lx.\n", file, line, val);
951 }
952 
953 void __pgd_error(const char *file, int line, unsigned long val)
954 {
955 	pr_err("%s:%d: bad pgd %016lx.\n", file, line, val);
956 }
957 
958 /* GENERIC_BUG traps */
959 
960 int is_valid_bugaddr(unsigned long addr)
961 {
962 	/*
963 	 * bug_handler() only called for BRK #BUG_BRK_IMM.
964 	 * So the answer is trivial -- any spurious instances with no
965 	 * bug table entry will be rejected by report_bug() and passed
966 	 * back to the debug-monitors code and handled as a fatal
967 	 * unexpected debug exception.
968 	 */
969 	return 1;
970 }
971 
972 static int bug_handler(struct pt_regs *regs, unsigned int esr)
973 {
974 	switch (report_bug(regs->pc, regs)) {
975 	case BUG_TRAP_TYPE_BUG:
976 		die("Oops - BUG", regs, 0);
977 		break;
978 
979 	case BUG_TRAP_TYPE_WARN:
980 		break;
981 
982 	default:
983 		/* unknown/unrecognised bug trap type */
984 		return DBG_HOOK_ERROR;
985 	}
986 
987 	/* If thread survives, skip over the BUG instruction and continue: */
988 	arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
989 	return DBG_HOOK_HANDLED;
990 }
991 
992 static struct break_hook bug_break_hook = {
993 	.fn = bug_handler,
994 	.imm = BUG_BRK_IMM,
995 };
996 
997 #ifdef CONFIG_KASAN_SW_TAGS
998 
999 #define KASAN_ESR_RECOVER	0x20
1000 #define KASAN_ESR_WRITE	0x10
1001 #define KASAN_ESR_SIZE_MASK	0x0f
1002 #define KASAN_ESR_SIZE(esr)	(1 << ((esr) & KASAN_ESR_SIZE_MASK))
1003 
1004 static int kasan_handler(struct pt_regs *regs, unsigned int esr)
1005 {
1006 	bool recover = esr & KASAN_ESR_RECOVER;
1007 	bool write = esr & KASAN_ESR_WRITE;
1008 	size_t size = KASAN_ESR_SIZE(esr);
1009 	u64 addr = regs->regs[0];
1010 	u64 pc = regs->pc;
1011 
1012 	kasan_report(addr, size, write, pc);
1013 
1014 	/*
1015 	 * The instrumentation allows to control whether we can proceed after
1016 	 * a crash was detected. This is done by passing the -recover flag to
1017 	 * the compiler. Disabling recovery allows to generate more compact
1018 	 * code.
1019 	 *
1020 	 * Unfortunately disabling recovery doesn't work for the kernel right
1021 	 * now. KASAN reporting is disabled in some contexts (for example when
1022 	 * the allocator accesses slab object metadata; this is controlled by
1023 	 * current->kasan_depth). All these accesses are detected by the tool,
1024 	 * even though the reports for them are not printed.
1025 	 *
1026 	 * This is something that might be fixed at some point in the future.
1027 	 */
1028 	if (!recover)
1029 		die("Oops - KASAN", regs, 0);
1030 
1031 	/* If thread survives, skip over the brk instruction and continue: */
1032 	arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
1033 	return DBG_HOOK_HANDLED;
1034 }
1035 
1036 static struct break_hook kasan_break_hook = {
1037 	.fn	= kasan_handler,
1038 	.imm	= KASAN_BRK_IMM,
1039 	.mask	= KASAN_BRK_MASK,
1040 };
1041 #endif
1042 
1043 /*
1044  * Initial handler for AArch64 BRK exceptions
1045  * This handler only used until debug_traps_init().
1046  */
1047 int __init early_brk64(unsigned long addr, unsigned int esr,
1048 		struct pt_regs *regs)
1049 {
1050 #ifdef CONFIG_KASAN_SW_TAGS
1051 	unsigned int comment = esr & ESR_ELx_BRK64_ISS_COMMENT_MASK;
1052 
1053 	if ((comment & ~KASAN_BRK_MASK) == KASAN_BRK_IMM)
1054 		return kasan_handler(regs, esr) != DBG_HOOK_HANDLED;
1055 #endif
1056 	return bug_handler(regs, esr) != DBG_HOOK_HANDLED;
1057 }
1058 
1059 void __init trap_init(void)
1060 {
1061 	register_kernel_break_hook(&bug_break_hook);
1062 #ifdef CONFIG_KASAN_SW_TAGS
1063 	register_kernel_break_hook(&kasan_break_hook);
1064 #endif
1065 	debug_traps_init();
1066 }
1067