xref: /openbmc/linux/arch/arm64/kernel/traps.c (revision 2985bed6)
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)
57 {
58 	printk(" %pS\n", (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 {
88 	struct stackframe frame;
89 	int skip = 0;
90 
91 	pr_debug("%s(regs = %p tsk = %p)\n", __func__, regs, tsk);
92 
93 	if (regs) {
94 		if (user_mode(regs))
95 			return;
96 		skip = 1;
97 	}
98 
99 	if (!tsk)
100 		tsk = current;
101 
102 	if (!try_get_task_stack(tsk))
103 		return;
104 
105 	if (tsk == current) {
106 		start_backtrace(&frame,
107 				(unsigned long)__builtin_frame_address(0),
108 				(unsigned long)dump_backtrace);
109 	} else {
110 		/*
111 		 * task blocked in __switch_to
112 		 */
113 		start_backtrace(&frame,
114 				thread_saved_fp(tsk),
115 				thread_saved_pc(tsk));
116 	}
117 
118 	printk("Call trace:\n");
119 	do {
120 		/* skip until specified stack frame */
121 		if (!skip) {
122 			dump_backtrace_entry(frame.pc);
123 		} else if (frame.fp == regs->regs[29]) {
124 			skip = 0;
125 			/*
126 			 * Mostly, this is the case where this function is
127 			 * called in panic/abort. As exception handler's
128 			 * stack frame does not contain the corresponding pc
129 			 * at which an exception has taken place, use regs->pc
130 			 * instead.
131 			 */
132 			dump_backtrace_entry(regs->pc);
133 		}
134 	} while (!unwind_frame(tsk, &frame));
135 
136 	put_task_stack(tsk);
137 }
138 
139 void show_stack(struct task_struct *tsk, unsigned long *sp)
140 {
141 	dump_backtrace(NULL, tsk);
142 	barrier();
143 }
144 
145 #ifdef CONFIG_PREEMPT
146 #define S_PREEMPT " PREEMPT"
147 #elif defined(CONFIG_PREEMPT_RT)
148 #define S_PREEMPT " PREEMPT_RT"
149 #else
150 #define S_PREEMPT ""
151 #endif
152 
153 #define S_SMP " SMP"
154 
155 static int __die(const char *str, int err, struct pt_regs *regs)
156 {
157 	static int die_counter;
158 	int ret;
159 
160 	pr_emerg("Internal error: %s: %x [#%d]" S_PREEMPT S_SMP "\n",
161 		 str, err, ++die_counter);
162 
163 	/* trap and error numbers are mostly meaningless on ARM */
164 	ret = notify_die(DIE_OOPS, str, regs, err, 0, SIGSEGV);
165 	if (ret == NOTIFY_STOP)
166 		return ret;
167 
168 	print_modules();
169 	show_regs(regs);
170 
171 	dump_kernel_instr(KERN_EMERG, regs);
172 
173 	return ret;
174 }
175 
176 static DEFINE_RAW_SPINLOCK(die_lock);
177 
178 /*
179  * This function is protected against re-entrancy.
180  */
181 void die(const char *str, struct pt_regs *regs, int err)
182 {
183 	int ret;
184 	unsigned long flags;
185 
186 	raw_spin_lock_irqsave(&die_lock, flags);
187 
188 	oops_enter();
189 
190 	console_verbose();
191 	bust_spinlocks(1);
192 	ret = __die(str, err, regs);
193 
194 	if (regs && kexec_should_crash(current))
195 		crash_kexec(regs);
196 
197 	bust_spinlocks(0);
198 	add_taint(TAINT_DIE, LOCKDEP_NOW_UNRELIABLE);
199 	oops_exit();
200 
201 	if (in_interrupt())
202 		panic("Fatal exception in interrupt");
203 	if (panic_on_oops)
204 		panic("Fatal exception");
205 
206 	raw_spin_unlock_irqrestore(&die_lock, flags);
207 
208 	if (ret != NOTIFY_STOP)
209 		do_exit(SIGSEGV);
210 }
211 
212 static void arm64_show_signal(int signo, const char *str)
213 {
214 	static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,
215 				      DEFAULT_RATELIMIT_BURST);
216 	struct task_struct *tsk = current;
217 	unsigned int esr = tsk->thread.fault_code;
218 	struct pt_regs *regs = task_pt_regs(tsk);
219 
220 	/* Leave if the signal won't be shown */
221 	if (!show_unhandled_signals ||
222 	    !unhandled_signal(tsk, signo) ||
223 	    !__ratelimit(&rs))
224 		return;
225 
226 	pr_info("%s[%d]: unhandled exception: ", tsk->comm, task_pid_nr(tsk));
227 	if (esr)
228 		pr_cont("%s, ESR 0x%08x, ", esr_get_class_string(esr), esr);
229 
230 	pr_cont("%s", str);
231 	print_vma_addr(KERN_CONT " in ", regs->pc);
232 	pr_cont("\n");
233 	__show_regs(regs);
234 }
235 
236 void arm64_force_sig_fault(int signo, int code, void __user *addr,
237 			   const char *str)
238 {
239 	arm64_show_signal(signo, str);
240 	if (signo == SIGKILL)
241 		force_sig(SIGKILL);
242 	else
243 		force_sig_fault(signo, code, addr);
244 }
245 
246 void arm64_force_sig_mceerr(int code, void __user *addr, short lsb,
247 			    const char *str)
248 {
249 	arm64_show_signal(SIGBUS, str);
250 	force_sig_mceerr(code, addr, lsb);
251 }
252 
253 void arm64_force_sig_ptrace_errno_trap(int errno, void __user *addr,
254 				       const char *str)
255 {
256 	arm64_show_signal(SIGTRAP, str);
257 	force_sig_ptrace_errno_trap(errno, addr);
258 }
259 
260 void arm64_notify_die(const char *str, struct pt_regs *regs,
261 		      int signo, int sicode, void __user *addr,
262 		      int err)
263 {
264 	if (user_mode(regs)) {
265 		WARN_ON(regs != current_pt_regs());
266 		current->thread.fault_address = 0;
267 		current->thread.fault_code = err;
268 
269 		arm64_force_sig_fault(signo, sicode, addr, str);
270 	} else {
271 		die(str, regs, err);
272 	}
273 }
274 
275 void arm64_skip_faulting_instruction(struct pt_regs *regs, unsigned long size)
276 {
277 	regs->pc += size;
278 
279 	/*
280 	 * If we were single stepping, we want to get the step exception after
281 	 * we return from the trap.
282 	 */
283 	if (user_mode(regs))
284 		user_fastforward_single_step(current);
285 }
286 
287 static LIST_HEAD(undef_hook);
288 static DEFINE_RAW_SPINLOCK(undef_lock);
289 
290 void register_undef_hook(struct undef_hook *hook)
291 {
292 	unsigned long flags;
293 
294 	raw_spin_lock_irqsave(&undef_lock, flags);
295 	list_add(&hook->node, &undef_hook);
296 	raw_spin_unlock_irqrestore(&undef_lock, flags);
297 }
298 
299 void unregister_undef_hook(struct undef_hook *hook)
300 {
301 	unsigned long flags;
302 
303 	raw_spin_lock_irqsave(&undef_lock, flags);
304 	list_del(&hook->node);
305 	raw_spin_unlock_irqrestore(&undef_lock, flags);
306 }
307 
308 static int call_undef_hook(struct pt_regs *regs)
309 {
310 	struct undef_hook *hook;
311 	unsigned long flags;
312 	u32 instr;
313 	int (*fn)(struct pt_regs *regs, u32 instr) = NULL;
314 	void __user *pc = (void __user *)instruction_pointer(regs);
315 
316 	if (!user_mode(regs)) {
317 		__le32 instr_le;
318 		if (probe_kernel_address((__force __le32 *)pc, instr_le))
319 			goto exit;
320 		instr = le32_to_cpu(instr_le);
321 	} else if (compat_thumb_mode(regs)) {
322 		/* 16-bit Thumb instruction */
323 		__le16 instr_le;
324 		if (get_user(instr_le, (__le16 __user *)pc))
325 			goto exit;
326 		instr = le16_to_cpu(instr_le);
327 		if (aarch32_insn_is_wide(instr)) {
328 			u32 instr2;
329 
330 			if (get_user(instr_le, (__le16 __user *)(pc + 2)))
331 				goto exit;
332 			instr2 = le16_to_cpu(instr_le);
333 			instr = (instr << 16) | instr2;
334 		}
335 	} else {
336 		/* 32-bit ARM instruction */
337 		__le32 instr_le;
338 		if (get_user(instr_le, (__le32 __user *)pc))
339 			goto exit;
340 		instr = le32_to_cpu(instr_le);
341 	}
342 
343 	raw_spin_lock_irqsave(&undef_lock, flags);
344 	list_for_each_entry(hook, &undef_hook, node)
345 		if ((instr & hook->instr_mask) == hook->instr_val &&
346 			(regs->pstate & hook->pstate_mask) == hook->pstate_val)
347 			fn = hook->fn;
348 
349 	raw_spin_unlock_irqrestore(&undef_lock, flags);
350 exit:
351 	return fn ? fn(regs, instr) : 1;
352 }
353 
354 void force_signal_inject(int signal, int code, unsigned long address)
355 {
356 	const char *desc;
357 	struct pt_regs *regs = current_pt_regs();
358 
359 	if (WARN_ON(!user_mode(regs)))
360 		return;
361 
362 	switch (signal) {
363 	case SIGILL:
364 		desc = "undefined instruction";
365 		break;
366 	case SIGSEGV:
367 		desc = "illegal memory access";
368 		break;
369 	default:
370 		desc = "unknown or unrecoverable error";
371 		break;
372 	}
373 
374 	/* Force signals we don't understand to SIGKILL */
375 	if (WARN_ON(signal != SIGKILL &&
376 		    siginfo_layout(signal, code) != SIL_FAULT)) {
377 		signal = SIGKILL;
378 	}
379 
380 	arm64_notify_die(desc, regs, signal, code, (void __user *)address, 0);
381 }
382 
383 /*
384  * Set up process info to signal segmentation fault - called on access error.
385  */
386 void arm64_notify_segfault(unsigned long addr)
387 {
388 	int code;
389 
390 	down_read(&current->mm->mmap_sem);
391 	if (find_vma(current->mm, addr) == NULL)
392 		code = SEGV_MAPERR;
393 	else
394 		code = SEGV_ACCERR;
395 	up_read(&current->mm->mmap_sem);
396 
397 	force_signal_inject(SIGSEGV, code, addr);
398 }
399 
400 void do_undefinstr(struct pt_regs *regs)
401 {
402 	/* check for AArch32 breakpoint instructions */
403 	if (!aarch32_break_handler(regs))
404 		return;
405 
406 	if (call_undef_hook(regs) == 0)
407 		return;
408 
409 	BUG_ON(!user_mode(regs));
410 	force_signal_inject(SIGILL, ILL_ILLOPC, regs->pc);
411 }
412 NOKPROBE_SYMBOL(do_undefinstr);
413 
414 #define __user_cache_maint(insn, address, res)			\
415 	if (address >= user_addr_max()) {			\
416 		res = -EFAULT;					\
417 	} else {						\
418 		uaccess_ttbr0_enable();				\
419 		asm volatile (					\
420 			"1:	" insn ", %1\n"			\
421 			"	mov	%w0, #0\n"		\
422 			"2:\n"					\
423 			"	.pushsection .fixup,\"ax\"\n"	\
424 			"	.align	2\n"			\
425 			"3:	mov	%w0, %w2\n"		\
426 			"	b	2b\n"			\
427 			"	.popsection\n"			\
428 			_ASM_EXTABLE(1b, 3b)			\
429 			: "=r" (res)				\
430 			: "r" (address), "i" (-EFAULT));	\
431 		uaccess_ttbr0_disable();			\
432 	}
433 
434 static void user_cache_maint_handler(unsigned int esr, struct pt_regs *regs)
435 {
436 	unsigned long address;
437 	int rt = ESR_ELx_SYS64_ISS_RT(esr);
438 	int crm = (esr & ESR_ELx_SYS64_ISS_CRM_MASK) >> ESR_ELx_SYS64_ISS_CRM_SHIFT;
439 	int ret = 0;
440 
441 	address = untagged_addr(pt_regs_read_reg(regs, rt));
442 
443 	switch (crm) {
444 	case ESR_ELx_SYS64_ISS_CRM_DC_CVAU:	/* DC CVAU, gets promoted */
445 		__user_cache_maint("dc civac", address, ret);
446 		break;
447 	case ESR_ELx_SYS64_ISS_CRM_DC_CVAC:	/* DC CVAC, gets promoted */
448 		__user_cache_maint("dc civac", address, ret);
449 		break;
450 	case ESR_ELx_SYS64_ISS_CRM_DC_CVADP:	/* DC CVADP */
451 		__user_cache_maint("sys 3, c7, c13, 1", address, ret);
452 		break;
453 	case ESR_ELx_SYS64_ISS_CRM_DC_CVAP:	/* DC CVAP */
454 		__user_cache_maint("sys 3, c7, c12, 1", address, ret);
455 		break;
456 	case ESR_ELx_SYS64_ISS_CRM_DC_CIVAC:	/* DC CIVAC */
457 		__user_cache_maint("dc civac", address, ret);
458 		break;
459 	case ESR_ELx_SYS64_ISS_CRM_IC_IVAU:	/* IC IVAU */
460 		__user_cache_maint("ic ivau", address, ret);
461 		break;
462 	default:
463 		force_signal_inject(SIGILL, ILL_ILLOPC, regs->pc);
464 		return;
465 	}
466 
467 	if (ret)
468 		arm64_notify_segfault(address);
469 	else
470 		arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
471 }
472 
473 static void ctr_read_handler(unsigned int esr, struct pt_regs *regs)
474 {
475 	int rt = ESR_ELx_SYS64_ISS_RT(esr);
476 	unsigned long val = arm64_ftr_reg_user_value(&arm64_ftr_reg_ctrel0);
477 
478 	if (cpus_have_const_cap(ARM64_WORKAROUND_1542419)) {
479 		/* Hide DIC so that we can trap the unnecessary maintenance...*/
480 		val &= ~BIT(CTR_DIC_SHIFT);
481 
482 		/* ... and fake IminLine to reduce the number of traps. */
483 		val &= ~CTR_IMINLINE_MASK;
484 		val |= (PAGE_SHIFT - 2) & CTR_IMINLINE_MASK;
485 	}
486 
487 	pt_regs_write_reg(regs, rt, val);
488 
489 	arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
490 }
491 
492 static void cntvct_read_handler(unsigned int esr, struct pt_regs *regs)
493 {
494 	int rt = ESR_ELx_SYS64_ISS_RT(esr);
495 
496 	pt_regs_write_reg(regs, rt, arch_timer_read_counter());
497 	arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
498 }
499 
500 static void cntfrq_read_handler(unsigned int esr, struct pt_regs *regs)
501 {
502 	int rt = ESR_ELx_SYS64_ISS_RT(esr);
503 
504 	pt_regs_write_reg(regs, rt, arch_timer_get_rate());
505 	arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
506 }
507 
508 static void mrs_handler(unsigned int esr, struct pt_regs *regs)
509 {
510 	u32 sysreg, rt;
511 
512 	rt = ESR_ELx_SYS64_ISS_RT(esr);
513 	sysreg = esr_sys64_to_sysreg(esr);
514 
515 	if (do_emulate_mrs(regs, sysreg, rt) != 0)
516 		force_signal_inject(SIGILL, ILL_ILLOPC, regs->pc);
517 }
518 
519 static void wfi_handler(unsigned int esr, struct pt_regs *regs)
520 {
521 	arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
522 }
523 
524 struct sys64_hook {
525 	unsigned int esr_mask;
526 	unsigned int esr_val;
527 	void (*handler)(unsigned int esr, struct pt_regs *regs);
528 };
529 
530 static const struct sys64_hook sys64_hooks[] = {
531 	{
532 		.esr_mask = ESR_ELx_SYS64_ISS_EL0_CACHE_OP_MASK,
533 		.esr_val = ESR_ELx_SYS64_ISS_EL0_CACHE_OP_VAL,
534 		.handler = user_cache_maint_handler,
535 	},
536 	{
537 		/* Trap read access to CTR_EL0 */
538 		.esr_mask = ESR_ELx_SYS64_ISS_SYS_OP_MASK,
539 		.esr_val = ESR_ELx_SYS64_ISS_SYS_CTR_READ,
540 		.handler = ctr_read_handler,
541 	},
542 	{
543 		/* Trap read access to CNTVCT_EL0 */
544 		.esr_mask = ESR_ELx_SYS64_ISS_SYS_OP_MASK,
545 		.esr_val = ESR_ELx_SYS64_ISS_SYS_CNTVCT,
546 		.handler = cntvct_read_handler,
547 	},
548 	{
549 		/* Trap read access to CNTFRQ_EL0 */
550 		.esr_mask = ESR_ELx_SYS64_ISS_SYS_OP_MASK,
551 		.esr_val = ESR_ELx_SYS64_ISS_SYS_CNTFRQ,
552 		.handler = cntfrq_read_handler,
553 	},
554 	{
555 		/* Trap read access to CPUID registers */
556 		.esr_mask = ESR_ELx_SYS64_ISS_SYS_MRS_OP_MASK,
557 		.esr_val = ESR_ELx_SYS64_ISS_SYS_MRS_OP_VAL,
558 		.handler = mrs_handler,
559 	},
560 	{
561 		/* Trap WFI instructions executed in userspace */
562 		.esr_mask = ESR_ELx_WFx_MASK,
563 		.esr_val = ESR_ELx_WFx_WFI_VAL,
564 		.handler = wfi_handler,
565 	},
566 	{},
567 };
568 
569 
570 #ifdef CONFIG_COMPAT
571 #define PSTATE_IT_1_0_SHIFT	25
572 #define PSTATE_IT_1_0_MASK	(0x3 << PSTATE_IT_1_0_SHIFT)
573 #define PSTATE_IT_7_2_SHIFT	10
574 #define PSTATE_IT_7_2_MASK	(0x3f << PSTATE_IT_7_2_SHIFT)
575 
576 static u32 compat_get_it_state(struct pt_regs *regs)
577 {
578 	u32 it, pstate = regs->pstate;
579 
580 	it  = (pstate & PSTATE_IT_1_0_MASK) >> PSTATE_IT_1_0_SHIFT;
581 	it |= ((pstate & PSTATE_IT_7_2_MASK) >> PSTATE_IT_7_2_SHIFT) << 2;
582 
583 	return it;
584 }
585 
586 static void compat_set_it_state(struct pt_regs *regs, u32 it)
587 {
588 	u32 pstate_it;
589 
590 	pstate_it  = (it << PSTATE_IT_1_0_SHIFT) & PSTATE_IT_1_0_MASK;
591 	pstate_it |= ((it >> 2) << PSTATE_IT_7_2_SHIFT) & PSTATE_IT_7_2_MASK;
592 
593 	regs->pstate &= ~PSR_AA32_IT_MASK;
594 	regs->pstate |= pstate_it;
595 }
596 
597 static bool cp15_cond_valid(unsigned int esr, struct pt_regs *regs)
598 {
599 	int cond;
600 
601 	/* Only a T32 instruction can trap without CV being set */
602 	if (!(esr & ESR_ELx_CV)) {
603 		u32 it;
604 
605 		it = compat_get_it_state(regs);
606 		if (!it)
607 			return true;
608 
609 		cond = it >> 4;
610 	} else {
611 		cond = (esr & ESR_ELx_COND_MASK) >> ESR_ELx_COND_SHIFT;
612 	}
613 
614 	return aarch32_opcode_cond_checks[cond](regs->pstate);
615 }
616 
617 static void advance_itstate(struct pt_regs *regs)
618 {
619 	u32 it;
620 
621 	/* ARM mode */
622 	if (!(regs->pstate & PSR_AA32_T_BIT) ||
623 	    !(regs->pstate & PSR_AA32_IT_MASK))
624 		return;
625 
626 	it  = compat_get_it_state(regs);
627 
628 	/*
629 	 * If this is the last instruction of the block, wipe the IT
630 	 * state. Otherwise advance it.
631 	 */
632 	if (!(it & 7))
633 		it = 0;
634 	else
635 		it = (it & 0xe0) | ((it << 1) & 0x1f);
636 
637 	compat_set_it_state(regs, it);
638 }
639 
640 static void arm64_compat_skip_faulting_instruction(struct pt_regs *regs,
641 						   unsigned int sz)
642 {
643 	advance_itstate(regs);
644 	arm64_skip_faulting_instruction(regs, sz);
645 }
646 
647 static void compat_cntfrq_read_handler(unsigned int esr, struct pt_regs *regs)
648 {
649 	int reg = (esr & ESR_ELx_CP15_32_ISS_RT_MASK) >> ESR_ELx_CP15_32_ISS_RT_SHIFT;
650 
651 	pt_regs_write_reg(regs, reg, arch_timer_get_rate());
652 	arm64_compat_skip_faulting_instruction(regs, 4);
653 }
654 
655 static const struct sys64_hook cp15_32_hooks[] = {
656 	{
657 		.esr_mask = ESR_ELx_CP15_32_ISS_SYS_MASK,
658 		.esr_val = ESR_ELx_CP15_32_ISS_SYS_CNTFRQ,
659 		.handler = compat_cntfrq_read_handler,
660 	},
661 	{},
662 };
663 
664 static void compat_cntvct_read_handler(unsigned int esr, struct pt_regs *regs)
665 {
666 	int rt = (esr & ESR_ELx_CP15_64_ISS_RT_MASK) >> ESR_ELx_CP15_64_ISS_RT_SHIFT;
667 	int rt2 = (esr & ESR_ELx_CP15_64_ISS_RT2_MASK) >> ESR_ELx_CP15_64_ISS_RT2_SHIFT;
668 	u64 val = arch_timer_read_counter();
669 
670 	pt_regs_write_reg(regs, rt, lower_32_bits(val));
671 	pt_regs_write_reg(regs, rt2, upper_32_bits(val));
672 	arm64_compat_skip_faulting_instruction(regs, 4);
673 }
674 
675 static const struct sys64_hook cp15_64_hooks[] = {
676 	{
677 		.esr_mask = ESR_ELx_CP15_64_ISS_SYS_MASK,
678 		.esr_val = ESR_ELx_CP15_64_ISS_SYS_CNTVCT,
679 		.handler = compat_cntvct_read_handler,
680 	},
681 	{},
682 };
683 
684 void do_cp15instr(unsigned int esr, struct pt_regs *regs)
685 {
686 	const struct sys64_hook *hook, *hook_base;
687 
688 	if (!cp15_cond_valid(esr, regs)) {
689 		/*
690 		 * There is no T16 variant of a CP access, so we
691 		 * always advance PC by 4 bytes.
692 		 */
693 		arm64_compat_skip_faulting_instruction(regs, 4);
694 		return;
695 	}
696 
697 	switch (ESR_ELx_EC(esr)) {
698 	case ESR_ELx_EC_CP15_32:
699 		hook_base = cp15_32_hooks;
700 		break;
701 	case ESR_ELx_EC_CP15_64:
702 		hook_base = cp15_64_hooks;
703 		break;
704 	default:
705 		do_undefinstr(regs);
706 		return;
707 	}
708 
709 	for (hook = hook_base; hook->handler; hook++)
710 		if ((hook->esr_mask & esr) == hook->esr_val) {
711 			hook->handler(esr, regs);
712 			return;
713 		}
714 
715 	/*
716 	 * New cp15 instructions may previously have been undefined at
717 	 * EL0. Fall back to our usual undefined instruction handler
718 	 * so that we handle these consistently.
719 	 */
720 	do_undefinstr(regs);
721 }
722 NOKPROBE_SYMBOL(do_cp15instr);
723 #endif
724 
725 void do_sysinstr(unsigned int esr, struct pt_regs *regs)
726 {
727 	const struct sys64_hook *hook;
728 
729 	for (hook = sys64_hooks; hook->handler; hook++)
730 		if ((hook->esr_mask & esr) == hook->esr_val) {
731 			hook->handler(esr, regs);
732 			return;
733 		}
734 
735 	/*
736 	 * New SYS instructions may previously have been undefined at EL0. Fall
737 	 * back to our usual undefined instruction handler so that we handle
738 	 * these consistently.
739 	 */
740 	do_undefinstr(regs);
741 }
742 NOKPROBE_SYMBOL(do_sysinstr);
743 
744 static const char *esr_class_str[] = {
745 	[0 ... ESR_ELx_EC_MAX]		= "UNRECOGNIZED EC",
746 	[ESR_ELx_EC_UNKNOWN]		= "Unknown/Uncategorized",
747 	[ESR_ELx_EC_WFx]		= "WFI/WFE",
748 	[ESR_ELx_EC_CP15_32]		= "CP15 MCR/MRC",
749 	[ESR_ELx_EC_CP15_64]		= "CP15 MCRR/MRRC",
750 	[ESR_ELx_EC_CP14_MR]		= "CP14 MCR/MRC",
751 	[ESR_ELx_EC_CP14_LS]		= "CP14 LDC/STC",
752 	[ESR_ELx_EC_FP_ASIMD]		= "ASIMD",
753 	[ESR_ELx_EC_CP10_ID]		= "CP10 MRC/VMRS",
754 	[ESR_ELx_EC_PAC]		= "PAC",
755 	[ESR_ELx_EC_CP14_64]		= "CP14 MCRR/MRRC",
756 	[ESR_ELx_EC_ILL]		= "PSTATE.IL",
757 	[ESR_ELx_EC_SVC32]		= "SVC (AArch32)",
758 	[ESR_ELx_EC_HVC32]		= "HVC (AArch32)",
759 	[ESR_ELx_EC_SMC32]		= "SMC (AArch32)",
760 	[ESR_ELx_EC_SVC64]		= "SVC (AArch64)",
761 	[ESR_ELx_EC_HVC64]		= "HVC (AArch64)",
762 	[ESR_ELx_EC_SMC64]		= "SMC (AArch64)",
763 	[ESR_ELx_EC_SYS64]		= "MSR/MRS (AArch64)",
764 	[ESR_ELx_EC_SVE]		= "SVE",
765 	[ESR_ELx_EC_ERET]		= "ERET/ERETAA/ERETAB",
766 	[ESR_ELx_EC_IMP_DEF]		= "EL3 IMP DEF",
767 	[ESR_ELx_EC_IABT_LOW]		= "IABT (lower EL)",
768 	[ESR_ELx_EC_IABT_CUR]		= "IABT (current EL)",
769 	[ESR_ELx_EC_PC_ALIGN]		= "PC Alignment",
770 	[ESR_ELx_EC_DABT_LOW]		= "DABT (lower EL)",
771 	[ESR_ELx_EC_DABT_CUR]		= "DABT (current EL)",
772 	[ESR_ELx_EC_SP_ALIGN]		= "SP Alignment",
773 	[ESR_ELx_EC_FP_EXC32]		= "FP (AArch32)",
774 	[ESR_ELx_EC_FP_EXC64]		= "FP (AArch64)",
775 	[ESR_ELx_EC_SERROR]		= "SError",
776 	[ESR_ELx_EC_BREAKPT_LOW]	= "Breakpoint (lower EL)",
777 	[ESR_ELx_EC_BREAKPT_CUR]	= "Breakpoint (current EL)",
778 	[ESR_ELx_EC_SOFTSTP_LOW]	= "Software Step (lower EL)",
779 	[ESR_ELx_EC_SOFTSTP_CUR]	= "Software Step (current EL)",
780 	[ESR_ELx_EC_WATCHPT_LOW]	= "Watchpoint (lower EL)",
781 	[ESR_ELx_EC_WATCHPT_CUR]	= "Watchpoint (current EL)",
782 	[ESR_ELx_EC_BKPT32]		= "BKPT (AArch32)",
783 	[ESR_ELx_EC_VECTOR32]		= "Vector catch (AArch32)",
784 	[ESR_ELx_EC_BRK64]		= "BRK (AArch64)",
785 };
786 
787 const char *esr_get_class_string(u32 esr)
788 {
789 	return esr_class_str[ESR_ELx_EC(esr)];
790 }
791 
792 /*
793  * bad_mode handles the impossible case in the exception vector. This is always
794  * fatal.
795  */
796 asmlinkage void bad_mode(struct pt_regs *regs, int reason, unsigned int esr)
797 {
798 	console_verbose();
799 
800 	pr_crit("Bad mode in %s handler detected on CPU%d, code 0x%08x -- %s\n",
801 		handler[reason], smp_processor_id(), esr,
802 		esr_get_class_string(esr));
803 
804 	local_daif_mask();
805 	panic("bad mode");
806 }
807 
808 /*
809  * bad_el0_sync handles unexpected, but potentially recoverable synchronous
810  * exceptions taken from EL0. Unlike bad_mode, this returns.
811  */
812 void bad_el0_sync(struct pt_regs *regs, int reason, unsigned int esr)
813 {
814 	void __user *pc = (void __user *)instruction_pointer(regs);
815 
816 	current->thread.fault_address = 0;
817 	current->thread.fault_code = esr;
818 
819 	arm64_force_sig_fault(SIGILL, ILL_ILLOPC, pc,
820 			      "Bad EL0 synchronous exception");
821 }
822 
823 #ifdef CONFIG_VMAP_STACK
824 
825 DEFINE_PER_CPU(unsigned long [OVERFLOW_STACK_SIZE/sizeof(long)], overflow_stack)
826 	__aligned(16);
827 
828 asmlinkage void handle_bad_stack(struct pt_regs *regs)
829 {
830 	unsigned long tsk_stk = (unsigned long)current->stack;
831 	unsigned long irq_stk = (unsigned long)this_cpu_read(irq_stack_ptr);
832 	unsigned long ovf_stk = (unsigned long)this_cpu_ptr(overflow_stack);
833 	unsigned int esr = read_sysreg(esr_el1);
834 	unsigned long far = read_sysreg(far_el1);
835 
836 	console_verbose();
837 	pr_emerg("Insufficient stack space to handle exception!");
838 
839 	pr_emerg("ESR: 0x%08x -- %s\n", esr, esr_get_class_string(esr));
840 	pr_emerg("FAR: 0x%016lx\n", far);
841 
842 	pr_emerg("Task stack:     [0x%016lx..0x%016lx]\n",
843 		 tsk_stk, tsk_stk + THREAD_SIZE);
844 	pr_emerg("IRQ stack:      [0x%016lx..0x%016lx]\n",
845 		 irq_stk, irq_stk + THREAD_SIZE);
846 	pr_emerg("Overflow stack: [0x%016lx..0x%016lx]\n",
847 		 ovf_stk, ovf_stk + OVERFLOW_STACK_SIZE);
848 
849 	__show_regs(regs);
850 
851 	/*
852 	 * We use nmi_panic to limit the potential for recusive overflows, and
853 	 * to get a better stack trace.
854 	 */
855 	nmi_panic(NULL, "kernel stack overflow");
856 	cpu_park_loop();
857 }
858 #endif
859 
860 void __noreturn arm64_serror_panic(struct pt_regs *regs, u32 esr)
861 {
862 	console_verbose();
863 
864 	pr_crit("SError Interrupt on CPU%d, code 0x%08x -- %s\n",
865 		smp_processor_id(), esr, esr_get_class_string(esr));
866 	if (regs)
867 		__show_regs(regs);
868 
869 	nmi_panic(regs, "Asynchronous SError Interrupt");
870 
871 	cpu_park_loop();
872 	unreachable();
873 }
874 
875 bool arm64_is_fatal_ras_serror(struct pt_regs *regs, unsigned int esr)
876 {
877 	u32 aet = arm64_ras_serror_get_severity(esr);
878 
879 	switch (aet) {
880 	case ESR_ELx_AET_CE:	/* corrected error */
881 	case ESR_ELx_AET_UEO:	/* restartable, not yet consumed */
882 		/*
883 		 * The CPU can make progress. We may take UEO again as
884 		 * a more severe error.
885 		 */
886 		return false;
887 
888 	case ESR_ELx_AET_UEU:	/* Uncorrected Unrecoverable */
889 	case ESR_ELx_AET_UER:	/* Uncorrected Recoverable */
890 		/*
891 		 * The CPU can't make progress. The exception may have
892 		 * been imprecise.
893 		 *
894 		 * Neoverse-N1 #1349291 means a non-KVM SError reported as
895 		 * Unrecoverable should be treated as Uncontainable. We
896 		 * call arm64_serror_panic() in both cases.
897 		 */
898 		return true;
899 
900 	case ESR_ELx_AET_UC:	/* Uncontainable or Uncategorized error */
901 	default:
902 		/* Error has been silently propagated */
903 		arm64_serror_panic(regs, esr);
904 	}
905 }
906 
907 asmlinkage void do_serror(struct pt_regs *regs, unsigned int esr)
908 {
909 	const bool was_in_nmi = in_nmi();
910 
911 	if (!was_in_nmi)
912 		nmi_enter();
913 
914 	/* non-RAS errors are not containable */
915 	if (!arm64_is_ras_serror(esr) || arm64_is_fatal_ras_serror(regs, esr))
916 		arm64_serror_panic(regs, esr);
917 
918 	if (!was_in_nmi)
919 		nmi_exit();
920 }
921 
922 asmlinkage void enter_from_user_mode(void)
923 {
924 	CT_WARN_ON(ct_state() != CONTEXT_USER);
925 	user_exit_irqoff();
926 }
927 NOKPROBE_SYMBOL(enter_from_user_mode);
928 
929 void __pte_error(const char *file, int line, unsigned long val)
930 {
931 	pr_err("%s:%d: bad pte %016lx.\n", file, line, val);
932 }
933 
934 void __pmd_error(const char *file, int line, unsigned long val)
935 {
936 	pr_err("%s:%d: bad pmd %016lx.\n", file, line, val);
937 }
938 
939 void __pud_error(const char *file, int line, unsigned long val)
940 {
941 	pr_err("%s:%d: bad pud %016lx.\n", file, line, val);
942 }
943 
944 void __pgd_error(const char *file, int line, unsigned long val)
945 {
946 	pr_err("%s:%d: bad pgd %016lx.\n", file, line, val);
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 #ifdef CONFIG_KASAN_SW_TAGS
989 
990 #define KASAN_ESR_RECOVER	0x20
991 #define KASAN_ESR_WRITE	0x10
992 #define KASAN_ESR_SIZE_MASK	0x0f
993 #define KASAN_ESR_SIZE(esr)	(1 << ((esr) & KASAN_ESR_SIZE_MASK))
994 
995 static int kasan_handler(struct pt_regs *regs, unsigned int esr)
996 {
997 	bool recover = esr & KASAN_ESR_RECOVER;
998 	bool write = esr & KASAN_ESR_WRITE;
999 	size_t size = KASAN_ESR_SIZE(esr);
1000 	u64 addr = regs->regs[0];
1001 	u64 pc = regs->pc;
1002 
1003 	kasan_report(addr, size, write, pc);
1004 
1005 	/*
1006 	 * The instrumentation allows to control whether we can proceed after
1007 	 * a crash was detected. This is done by passing the -recover flag to
1008 	 * the compiler. Disabling recovery allows to generate more compact
1009 	 * code.
1010 	 *
1011 	 * Unfortunately disabling recovery doesn't work for the kernel right
1012 	 * now. KASAN reporting is disabled in some contexts (for example when
1013 	 * the allocator accesses slab object metadata; this is controlled by
1014 	 * current->kasan_depth). All these accesses are detected by the tool,
1015 	 * even though the reports for them are not printed.
1016 	 *
1017 	 * This is something that might be fixed at some point in the future.
1018 	 */
1019 	if (!recover)
1020 		die("Oops - KASAN", regs, 0);
1021 
1022 	/* If thread survives, skip over the brk instruction and continue: */
1023 	arm64_skip_faulting_instruction(regs, AARCH64_INSN_SIZE);
1024 	return DBG_HOOK_HANDLED;
1025 }
1026 
1027 static struct break_hook kasan_break_hook = {
1028 	.fn	= kasan_handler,
1029 	.imm	= KASAN_BRK_IMM,
1030 	.mask	= KASAN_BRK_MASK,
1031 };
1032 #endif
1033 
1034 /*
1035  * Initial handler for AArch64 BRK exceptions
1036  * This handler only used until debug_traps_init().
1037  */
1038 int __init early_brk64(unsigned long addr, unsigned int esr,
1039 		struct pt_regs *regs)
1040 {
1041 #ifdef CONFIG_KASAN_SW_TAGS
1042 	unsigned int comment = esr & ESR_ELx_BRK64_ISS_COMMENT_MASK;
1043 
1044 	if ((comment & ~KASAN_BRK_MASK) == KASAN_BRK_IMM)
1045 		return kasan_handler(regs, esr) != DBG_HOOK_HANDLED;
1046 #endif
1047 	return bug_handler(regs, esr) != DBG_HOOK_HANDLED;
1048 }
1049 
1050 /* This registration must happen early, before debug_traps_init(). */
1051 void __init trap_init(void)
1052 {
1053 	register_kernel_break_hook(&bug_break_hook);
1054 #ifdef CONFIG_KASAN_SW_TAGS
1055 	register_kernel_break_hook(&kasan_break_hook);
1056 #endif
1057 }
1058