xref: /linux/arch/arm64/mm/fault.c (revision 2dd0e8d2d2a157dbc83295a78336c2217110f2f8)
1 /*
2  * Based on arch/arm/mm/fault.c
3  *
4  * Copyright (C) 1995  Linus Torvalds
5  * Copyright (C) 1995-2004 Russell King
6  * Copyright (C) 2012 ARM Ltd.
7  *
8  * This program is free software; you can redistribute it and/or modify
9  * it under the terms of the GNU General Public License version 2 as
10  * published by the Free Software Foundation.
11  *
12  * This program is distributed in the hope that it will be useful,
13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
15  * GNU General Public License for more details.
16  *
17  * You should have received a copy of the GNU General Public License
18  * along with this program.  If not, see <http://www.gnu.org/licenses/>.
19  */
20 
21 #include <linux/module.h>
22 #include <linux/signal.h>
23 #include <linux/mm.h>
24 #include <linux/hardirq.h>
25 #include <linux/init.h>
26 #include <linux/kprobes.h>
27 #include <linux/uaccess.h>
28 #include <linux/page-flags.h>
29 #include <linux/sched.h>
30 #include <linux/highmem.h>
31 #include <linux/perf_event.h>
32 
33 #include <asm/cpufeature.h>
34 #include <asm/exception.h>
35 #include <asm/debug-monitors.h>
36 #include <asm/esr.h>
37 #include <asm/sysreg.h>
38 #include <asm/system_misc.h>
39 #include <asm/pgtable.h>
40 #include <asm/tlbflush.h>
41 
42 static const char *fault_name(unsigned int esr);
43 
44 #ifdef CONFIG_KPROBES
45 static inline int notify_page_fault(struct pt_regs *regs, unsigned int esr)
46 {
47 	int ret = 0;
48 
49 	/* kprobe_running() needs smp_processor_id() */
50 	if (!user_mode(regs)) {
51 		preempt_disable();
52 		if (kprobe_running() && kprobe_fault_handler(regs, esr))
53 			ret = 1;
54 		preempt_enable();
55 	}
56 
57 	return ret;
58 }
59 #else
60 static inline int notify_page_fault(struct pt_regs *regs, unsigned int esr)
61 {
62 	return 0;
63 }
64 #endif
65 
66 /*
67  * Dump out the page tables associated with 'addr' in mm 'mm'.
68  */
69 void show_pte(struct mm_struct *mm, unsigned long addr)
70 {
71 	pgd_t *pgd;
72 
73 	if (!mm)
74 		mm = &init_mm;
75 
76 	pr_alert("pgd = %p\n", mm->pgd);
77 	pgd = pgd_offset(mm, addr);
78 	pr_alert("[%08lx] *pgd=%016llx", addr, pgd_val(*pgd));
79 
80 	do {
81 		pud_t *pud;
82 		pmd_t *pmd;
83 		pte_t *pte;
84 
85 		if (pgd_none(*pgd) || pgd_bad(*pgd))
86 			break;
87 
88 		pud = pud_offset(pgd, addr);
89 		printk(", *pud=%016llx", pud_val(*pud));
90 		if (pud_none(*pud) || pud_bad(*pud))
91 			break;
92 
93 		pmd = pmd_offset(pud, addr);
94 		printk(", *pmd=%016llx", pmd_val(*pmd));
95 		if (pmd_none(*pmd) || pmd_bad(*pmd))
96 			break;
97 
98 		pte = pte_offset_map(pmd, addr);
99 		printk(", *pte=%016llx", pte_val(*pte));
100 		pte_unmap(pte);
101 	} while(0);
102 
103 	printk("\n");
104 }
105 
106 #ifdef CONFIG_ARM64_HW_AFDBM
107 /*
108  * This function sets the access flags (dirty, accessed), as well as write
109  * permission, and only to a more permissive setting.
110  *
111  * It needs to cope with hardware update of the accessed/dirty state by other
112  * agents in the system and can safely skip the __sync_icache_dcache() call as,
113  * like set_pte_at(), the PTE is never changed from no-exec to exec here.
114  *
115  * Returns whether or not the PTE actually changed.
116  */
117 int ptep_set_access_flags(struct vm_area_struct *vma,
118 			  unsigned long address, pte_t *ptep,
119 			  pte_t entry, int dirty)
120 {
121 	pteval_t old_pteval;
122 	unsigned int tmp;
123 
124 	if (pte_same(*ptep, entry))
125 		return 0;
126 
127 	/* only preserve the access flags and write permission */
128 	pte_val(entry) &= PTE_AF | PTE_WRITE | PTE_DIRTY;
129 
130 	/*
131 	 * PTE_RDONLY is cleared by default in the asm below, so set it in
132 	 * back if necessary (read-only or clean PTE).
133 	 */
134 	if (!pte_write(entry) || !pte_sw_dirty(entry))
135 		pte_val(entry) |= PTE_RDONLY;
136 
137 	/*
138 	 * Setting the flags must be done atomically to avoid racing with the
139 	 * hardware update of the access/dirty state.
140 	 */
141 	asm volatile("//	ptep_set_access_flags\n"
142 	"	prfm	pstl1strm, %2\n"
143 	"1:	ldxr	%0, %2\n"
144 	"	and	%0, %0, %3		// clear PTE_RDONLY\n"
145 	"	orr	%0, %0, %4		// set flags\n"
146 	"	stxr	%w1, %0, %2\n"
147 	"	cbnz	%w1, 1b\n"
148 	: "=&r" (old_pteval), "=&r" (tmp), "+Q" (pte_val(*ptep))
149 	: "L" (~PTE_RDONLY), "r" (pte_val(entry)));
150 
151 	flush_tlb_fix_spurious_fault(vma, address);
152 	return 1;
153 }
154 #endif
155 
156 /*
157  * The kernel tried to access some page that wasn't present.
158  */
159 static void __do_kernel_fault(struct mm_struct *mm, unsigned long addr,
160 			      unsigned int esr, struct pt_regs *regs)
161 {
162 	/*
163 	 * Are we prepared to handle this kernel fault?
164 	 */
165 	if (fixup_exception(regs))
166 		return;
167 
168 	/*
169 	 * No handler, we'll have to terminate things with extreme prejudice.
170 	 */
171 	bust_spinlocks(1);
172 	pr_alert("Unable to handle kernel %s at virtual address %08lx\n",
173 		 (addr < PAGE_SIZE) ? "NULL pointer dereference" :
174 		 "paging request", addr);
175 
176 	show_pte(mm, addr);
177 	die("Oops", regs, esr);
178 	bust_spinlocks(0);
179 	do_exit(SIGKILL);
180 }
181 
182 /*
183  * Something tried to access memory that isn't in our memory map. User mode
184  * accesses just cause a SIGSEGV
185  */
186 static void __do_user_fault(struct task_struct *tsk, unsigned long addr,
187 			    unsigned int esr, unsigned int sig, int code,
188 			    struct pt_regs *regs)
189 {
190 	struct siginfo si;
191 
192 	if (unhandled_signal(tsk, sig) && show_unhandled_signals_ratelimited()) {
193 		pr_info("%s[%d]: unhandled %s (%d) at 0x%08lx, esr 0x%03x\n",
194 			tsk->comm, task_pid_nr(tsk), fault_name(esr), sig,
195 			addr, esr);
196 		show_pte(tsk->mm, addr);
197 		show_regs(regs);
198 	}
199 
200 	tsk->thread.fault_address = addr;
201 	tsk->thread.fault_code = esr;
202 	si.si_signo = sig;
203 	si.si_errno = 0;
204 	si.si_code = code;
205 	si.si_addr = (void __user *)addr;
206 	force_sig_info(sig, &si, tsk);
207 }
208 
209 static void do_bad_area(unsigned long addr, unsigned int esr, struct pt_regs *regs)
210 {
211 	struct task_struct *tsk = current;
212 	struct mm_struct *mm = tsk->active_mm;
213 
214 	/*
215 	 * If we are in kernel mode at this point, we have no context to
216 	 * handle this fault with.
217 	 */
218 	if (user_mode(regs))
219 		__do_user_fault(tsk, addr, esr, SIGSEGV, SEGV_MAPERR, regs);
220 	else
221 		__do_kernel_fault(mm, addr, esr, regs);
222 }
223 
224 #define VM_FAULT_BADMAP		0x010000
225 #define VM_FAULT_BADACCESS	0x020000
226 
227 #define ESR_LNX_EXEC		(1 << 24)
228 
229 static int __do_page_fault(struct mm_struct *mm, unsigned long addr,
230 			   unsigned int mm_flags, unsigned long vm_flags,
231 			   struct task_struct *tsk)
232 {
233 	struct vm_area_struct *vma;
234 	int fault;
235 
236 	vma = find_vma(mm, addr);
237 	fault = VM_FAULT_BADMAP;
238 	if (unlikely(!vma))
239 		goto out;
240 	if (unlikely(vma->vm_start > addr))
241 		goto check_stack;
242 
243 	/*
244 	 * Ok, we have a good vm_area for this memory access, so we can handle
245 	 * it.
246 	 */
247 good_area:
248 	/*
249 	 * Check that the permissions on the VMA allow for the fault which
250 	 * occurred. If we encountered a write or exec fault, we must have
251 	 * appropriate permissions, otherwise we allow any permission.
252 	 */
253 	if (!(vma->vm_flags & vm_flags)) {
254 		fault = VM_FAULT_BADACCESS;
255 		goto out;
256 	}
257 
258 	return handle_mm_fault(mm, vma, addr & PAGE_MASK, mm_flags);
259 
260 check_stack:
261 	if (vma->vm_flags & VM_GROWSDOWN && !expand_stack(vma, addr))
262 		goto good_area;
263 out:
264 	return fault;
265 }
266 
267 static inline int permission_fault(unsigned int esr)
268 {
269 	unsigned int ec       = (esr & ESR_ELx_EC_MASK) >> ESR_ELx_EC_SHIFT;
270 	unsigned int fsc_type = esr & ESR_ELx_FSC_TYPE;
271 
272 	return (ec == ESR_ELx_EC_DABT_CUR && fsc_type == ESR_ELx_FSC_PERM);
273 }
274 
275 static int __kprobes do_page_fault(unsigned long addr, unsigned int esr,
276 				   struct pt_regs *regs)
277 {
278 	struct task_struct *tsk;
279 	struct mm_struct *mm;
280 	int fault, sig, code;
281 	unsigned long vm_flags = VM_READ | VM_WRITE | VM_EXEC;
282 	unsigned int mm_flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
283 
284 	if (notify_page_fault(regs, esr))
285 		return 0;
286 
287 	tsk = current;
288 	mm  = tsk->mm;
289 
290 	/*
291 	 * If we're in an interrupt or have no user context, we must not take
292 	 * the fault.
293 	 */
294 	if (faulthandler_disabled() || !mm)
295 		goto no_context;
296 
297 	if (user_mode(regs))
298 		mm_flags |= FAULT_FLAG_USER;
299 
300 	if (esr & ESR_LNX_EXEC) {
301 		vm_flags = VM_EXEC;
302 	} else if ((esr & ESR_ELx_WNR) && !(esr & ESR_ELx_CM)) {
303 		vm_flags = VM_WRITE;
304 		mm_flags |= FAULT_FLAG_WRITE;
305 	}
306 
307 	if (permission_fault(esr) && (addr < USER_DS)) {
308 		if (get_fs() == KERNEL_DS)
309 			die("Accessing user space memory with fs=KERNEL_DS", regs, esr);
310 
311 		if (!search_exception_tables(regs->pc))
312 			die("Accessing user space memory outside uaccess.h routines", regs, esr);
313 	}
314 
315 	/*
316 	 * As per x86, we may deadlock here. However, since the kernel only
317 	 * validly references user space from well defined areas of the code,
318 	 * we can bug out early if this is from code which shouldn't.
319 	 */
320 	if (!down_read_trylock(&mm->mmap_sem)) {
321 		if (!user_mode(regs) && !search_exception_tables(regs->pc))
322 			goto no_context;
323 retry:
324 		down_read(&mm->mmap_sem);
325 	} else {
326 		/*
327 		 * The above down_read_trylock() might have succeeded in which
328 		 * case, we'll have missed the might_sleep() from down_read().
329 		 */
330 		might_sleep();
331 #ifdef CONFIG_DEBUG_VM
332 		if (!user_mode(regs) && !search_exception_tables(regs->pc))
333 			goto no_context;
334 #endif
335 	}
336 
337 	fault = __do_page_fault(mm, addr, mm_flags, vm_flags, tsk);
338 
339 	/*
340 	 * If we need to retry but a fatal signal is pending, handle the
341 	 * signal first. We do not need to release the mmap_sem because it
342 	 * would already be released in __lock_page_or_retry in mm/filemap.c.
343 	 */
344 	if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
345 		return 0;
346 
347 	/*
348 	 * Major/minor page fault accounting is only done on the initial
349 	 * attempt. If we go through a retry, it is extremely likely that the
350 	 * page will be found in page cache at that point.
351 	 */
352 
353 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
354 	if (mm_flags & FAULT_FLAG_ALLOW_RETRY) {
355 		if (fault & VM_FAULT_MAJOR) {
356 			tsk->maj_flt++;
357 			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs,
358 				      addr);
359 		} else {
360 			tsk->min_flt++;
361 			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs,
362 				      addr);
363 		}
364 		if (fault & VM_FAULT_RETRY) {
365 			/*
366 			 * Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk of
367 			 * starvation.
368 			 */
369 			mm_flags &= ~FAULT_FLAG_ALLOW_RETRY;
370 			mm_flags |= FAULT_FLAG_TRIED;
371 			goto retry;
372 		}
373 	}
374 
375 	up_read(&mm->mmap_sem);
376 
377 	/*
378 	 * Handle the "normal" case first - VM_FAULT_MAJOR
379 	 */
380 	if (likely(!(fault & (VM_FAULT_ERROR | VM_FAULT_BADMAP |
381 			      VM_FAULT_BADACCESS))))
382 		return 0;
383 
384 	/*
385 	 * If we are in kernel mode at this point, we have no context to
386 	 * handle this fault with.
387 	 */
388 	if (!user_mode(regs))
389 		goto no_context;
390 
391 	if (fault & VM_FAULT_OOM) {
392 		/*
393 		 * We ran out of memory, call the OOM killer, and return to
394 		 * userspace (which will retry the fault, or kill us if we got
395 		 * oom-killed).
396 		 */
397 		pagefault_out_of_memory();
398 		return 0;
399 	}
400 
401 	if (fault & VM_FAULT_SIGBUS) {
402 		/*
403 		 * We had some memory, but were unable to successfully fix up
404 		 * this page fault.
405 		 */
406 		sig = SIGBUS;
407 		code = BUS_ADRERR;
408 	} else {
409 		/*
410 		 * Something tried to access memory that isn't in our memory
411 		 * map.
412 		 */
413 		sig = SIGSEGV;
414 		code = fault == VM_FAULT_BADACCESS ?
415 			SEGV_ACCERR : SEGV_MAPERR;
416 	}
417 
418 	__do_user_fault(tsk, addr, esr, sig, code, regs);
419 	return 0;
420 
421 no_context:
422 	__do_kernel_fault(mm, addr, esr, regs);
423 	return 0;
424 }
425 
426 /*
427  * First Level Translation Fault Handler
428  *
429  * We enter here because the first level page table doesn't contain a valid
430  * entry for the address.
431  *
432  * If the address is in kernel space (>= TASK_SIZE), then we are probably
433  * faulting in the vmalloc() area.
434  *
435  * If the init_task's first level page tables contains the relevant entry, we
436  * copy the it to this task.  If not, we send the process a signal, fixup the
437  * exception, or oops the kernel.
438  *
439  * NOTE! We MUST NOT take any locks for this case. We may be in an interrupt
440  * or a critical region, and should only copy the information from the master
441  * page table, nothing more.
442  */
443 static int __kprobes do_translation_fault(unsigned long addr,
444 					  unsigned int esr,
445 					  struct pt_regs *regs)
446 {
447 	if (addr < TASK_SIZE)
448 		return do_page_fault(addr, esr, regs);
449 
450 	do_bad_area(addr, esr, regs);
451 	return 0;
452 }
453 
454 static int do_alignment_fault(unsigned long addr, unsigned int esr,
455 			      struct pt_regs *regs)
456 {
457 	do_bad_area(addr, esr, regs);
458 	return 0;
459 }
460 
461 /*
462  * This abort handler always returns "fault".
463  */
464 static int do_bad(unsigned long addr, unsigned int esr, struct pt_regs *regs)
465 {
466 	return 1;
467 }
468 
469 static const struct fault_info {
470 	int	(*fn)(unsigned long addr, unsigned int esr, struct pt_regs *regs);
471 	int	sig;
472 	int	code;
473 	const char *name;
474 } fault_info[] = {
475 	{ do_bad,		SIGBUS,  0,		"ttbr address size fault"	},
476 	{ do_bad,		SIGBUS,  0,		"level 1 address size fault"	},
477 	{ do_bad,		SIGBUS,  0,		"level 2 address size fault"	},
478 	{ do_bad,		SIGBUS,  0,		"level 3 address size fault"	},
479 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 0 translation fault"	},
480 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 1 translation fault"	},
481 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 2 translation fault"	},
482 	{ do_page_fault,	SIGSEGV, SEGV_MAPERR,	"level 3 translation fault"	},
483 	{ do_bad,		SIGBUS,  0,		"unknown 8"			},
484 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 1 access flag fault"	},
485 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 2 access flag fault"	},
486 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 3 access flag fault"	},
487 	{ do_bad,		SIGBUS,  0,		"unknown 12"			},
488 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 1 permission fault"	},
489 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 2 permission fault"	},
490 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 3 permission fault"	},
491 	{ do_bad,		SIGBUS,  0,		"synchronous external abort"	},
492 	{ do_bad,		SIGBUS,  0,		"unknown 17"			},
493 	{ do_bad,		SIGBUS,  0,		"unknown 18"			},
494 	{ do_bad,		SIGBUS,  0,		"unknown 19"			},
495 	{ do_bad,		SIGBUS,  0,		"synchronous abort (translation table walk)" },
496 	{ do_bad,		SIGBUS,  0,		"synchronous abort (translation table walk)" },
497 	{ do_bad,		SIGBUS,  0,		"synchronous abort (translation table walk)" },
498 	{ do_bad,		SIGBUS,  0,		"synchronous abort (translation table walk)" },
499 	{ do_bad,		SIGBUS,  0,		"synchronous parity error"	},
500 	{ do_bad,		SIGBUS,  0,		"unknown 25"			},
501 	{ do_bad,		SIGBUS,  0,		"unknown 26"			},
502 	{ do_bad,		SIGBUS,  0,		"unknown 27"			},
503 	{ do_bad,		SIGBUS,  0,		"synchronous parity error (translation table walk)" },
504 	{ do_bad,		SIGBUS,  0,		"synchronous parity error (translation table walk)" },
505 	{ do_bad,		SIGBUS,  0,		"synchronous parity error (translation table walk)" },
506 	{ do_bad,		SIGBUS,  0,		"synchronous parity error (translation table walk)" },
507 	{ do_bad,		SIGBUS,  0,		"unknown 32"			},
508 	{ do_alignment_fault,	SIGBUS,  BUS_ADRALN,	"alignment fault"		},
509 	{ do_bad,		SIGBUS,  0,		"unknown 34"			},
510 	{ do_bad,		SIGBUS,  0,		"unknown 35"			},
511 	{ do_bad,		SIGBUS,  0,		"unknown 36"			},
512 	{ do_bad,		SIGBUS,  0,		"unknown 37"			},
513 	{ do_bad,		SIGBUS,  0,		"unknown 38"			},
514 	{ do_bad,		SIGBUS,  0,		"unknown 39"			},
515 	{ do_bad,		SIGBUS,  0,		"unknown 40"			},
516 	{ do_bad,		SIGBUS,  0,		"unknown 41"			},
517 	{ do_bad,		SIGBUS,  0,		"unknown 42"			},
518 	{ do_bad,		SIGBUS,  0,		"unknown 43"			},
519 	{ do_bad,		SIGBUS,  0,		"unknown 44"			},
520 	{ do_bad,		SIGBUS,  0,		"unknown 45"			},
521 	{ do_bad,		SIGBUS,  0,		"unknown 46"			},
522 	{ do_bad,		SIGBUS,  0,		"unknown 47"			},
523 	{ do_bad,		SIGBUS,  0,		"TLB conflict abort"		},
524 	{ do_bad,		SIGBUS,  0,		"unknown 49"			},
525 	{ do_bad,		SIGBUS,  0,		"unknown 50"			},
526 	{ do_bad,		SIGBUS,  0,		"unknown 51"			},
527 	{ do_bad,		SIGBUS,  0,		"implementation fault (lockdown abort)" },
528 	{ do_bad,		SIGBUS,  0,		"implementation fault (unsupported exclusive)" },
529 	{ do_bad,		SIGBUS,  0,		"unknown 54"			},
530 	{ do_bad,		SIGBUS,  0,		"unknown 55"			},
531 	{ do_bad,		SIGBUS,  0,		"unknown 56"			},
532 	{ do_bad,		SIGBUS,  0,		"unknown 57"			},
533 	{ do_bad,		SIGBUS,  0,		"unknown 58" 			},
534 	{ do_bad,		SIGBUS,  0,		"unknown 59"			},
535 	{ do_bad,		SIGBUS,  0,		"unknown 60"			},
536 	{ do_bad,		SIGBUS,  0,		"section domain fault"		},
537 	{ do_bad,		SIGBUS,  0,		"page domain fault"		},
538 	{ do_bad,		SIGBUS,  0,		"unknown 63"			},
539 };
540 
541 static const char *fault_name(unsigned int esr)
542 {
543 	const struct fault_info *inf = fault_info + (esr & 63);
544 	return inf->name;
545 }
546 
547 /*
548  * Dispatch a data abort to the relevant handler.
549  */
550 asmlinkage void __exception do_mem_abort(unsigned long addr, unsigned int esr,
551 					 struct pt_regs *regs)
552 {
553 	const struct fault_info *inf = fault_info + (esr & 63);
554 	struct siginfo info;
555 
556 	if (!inf->fn(addr, esr, regs))
557 		return;
558 
559 	pr_alert("Unhandled fault: %s (0x%08x) at 0x%016lx\n",
560 		 inf->name, esr, addr);
561 
562 	info.si_signo = inf->sig;
563 	info.si_errno = 0;
564 	info.si_code  = inf->code;
565 	info.si_addr  = (void __user *)addr;
566 	arm64_notify_die("", regs, &info, esr);
567 }
568 
569 /*
570  * Handle stack alignment exceptions.
571  */
572 asmlinkage void __exception do_sp_pc_abort(unsigned long addr,
573 					   unsigned int esr,
574 					   struct pt_regs *regs)
575 {
576 	struct siginfo info;
577 	struct task_struct *tsk = current;
578 
579 	if (show_unhandled_signals && unhandled_signal(tsk, SIGBUS))
580 		pr_info_ratelimited("%s[%d]: %s exception: pc=%p sp=%p\n",
581 				    tsk->comm, task_pid_nr(tsk),
582 				    esr_get_class_string(esr), (void *)regs->pc,
583 				    (void *)regs->sp);
584 
585 	info.si_signo = SIGBUS;
586 	info.si_errno = 0;
587 	info.si_code  = BUS_ADRALN;
588 	info.si_addr  = (void __user *)addr;
589 	arm64_notify_die("Oops - SP/PC alignment exception", regs, &info, esr);
590 }
591 
592 int __init early_brk64(unsigned long addr, unsigned int esr,
593 		       struct pt_regs *regs);
594 
595 /*
596  * __refdata because early_brk64 is __init, but the reference to it is
597  * clobbered at arch_initcall time.
598  * See traps.c and debug-monitors.c:debug_traps_init().
599  */
600 static struct fault_info __refdata debug_fault_info[] = {
601 	{ do_bad,	SIGTRAP,	TRAP_HWBKPT,	"hardware breakpoint"	},
602 	{ do_bad,	SIGTRAP,	TRAP_HWBKPT,	"hardware single-step"	},
603 	{ do_bad,	SIGTRAP,	TRAP_HWBKPT,	"hardware watchpoint"	},
604 	{ do_bad,	SIGBUS,		0,		"unknown 3"		},
605 	{ do_bad,	SIGTRAP,	TRAP_BRKPT,	"aarch32 BKPT"		},
606 	{ do_bad,	SIGTRAP,	0,		"aarch32 vector catch"	},
607 	{ early_brk64,	SIGTRAP,	TRAP_BRKPT,	"aarch64 BRK"		},
608 	{ do_bad,	SIGBUS,		0,		"unknown 7"		},
609 };
610 
611 void __init hook_debug_fault_code(int nr,
612 				  int (*fn)(unsigned long, unsigned int, struct pt_regs *),
613 				  int sig, int code, const char *name)
614 {
615 	BUG_ON(nr < 0 || nr >= ARRAY_SIZE(debug_fault_info));
616 
617 	debug_fault_info[nr].fn		= fn;
618 	debug_fault_info[nr].sig	= sig;
619 	debug_fault_info[nr].code	= code;
620 	debug_fault_info[nr].name	= name;
621 }
622 
623 asmlinkage int __exception do_debug_exception(unsigned long addr,
624 					      unsigned int esr,
625 					      struct pt_regs *regs)
626 {
627 	const struct fault_info *inf = debug_fault_info + DBG_ESR_EVT(esr);
628 	struct siginfo info;
629 	int rv;
630 
631 	/*
632 	 * Tell lockdep we disabled irqs in entry.S. Do nothing if they were
633 	 * already disabled to preserve the last enabled/disabled addresses.
634 	 */
635 	if (interrupts_enabled(regs))
636 		trace_hardirqs_off();
637 
638 	if (!inf->fn(addr, esr, regs)) {
639 		rv = 1;
640 	} else {
641 		pr_alert("Unhandled debug exception: %s (0x%08x) at 0x%016lx\n",
642 			 inf->name, esr, addr);
643 
644 		info.si_signo = inf->sig;
645 		info.si_errno = 0;
646 		info.si_code  = inf->code;
647 		info.si_addr  = (void __user *)addr;
648 		arm64_notify_die("", regs, &info, 0);
649 		rv = 0;
650 	}
651 
652 	if (interrupts_enabled(regs))
653 		trace_hardirqs_on();
654 
655 	return rv;
656 }
657 NOKPROBE_SYMBOL(do_debug_exception);
658 
659 #ifdef CONFIG_ARM64_PAN
660 void cpu_enable_pan(void *__unused)
661 {
662 	config_sctlr_el1(SCTLR_EL1_SPAN, 0);
663 }
664 #endif /* CONFIG_ARM64_PAN */
665 
666 #ifdef CONFIG_ARM64_UAO
667 /*
668  * Kernel threads have fs=KERNEL_DS by default, and don't need to call
669  * set_fs(), devtmpfs in particular relies on this behaviour.
670  * We need to enable the feature at runtime (instead of adding it to
671  * PSR_MODE_EL1h) as the feature may not be implemented by the cpu.
672  */
673 void cpu_enable_uao(void *__unused)
674 {
675 	asm(SET_PSTATE_UAO(1));
676 }
677 #endif /* CONFIG_ARM64_UAO */
678