xref: /openbmc/linux/arch/arm64/mm/fault.c (revision bc5aa3a0)
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 static bool is_el1_instruction_abort(unsigned int esr)
157 {
158 	return ESR_ELx_EC(esr) == ESR_ELx_EC_IABT_CUR;
159 }
160 
161 /*
162  * The kernel tried to access some page that wasn't present.
163  */
164 static void __do_kernel_fault(struct mm_struct *mm, unsigned long addr,
165 			      unsigned int esr, struct pt_regs *regs)
166 {
167 	/*
168 	 * Are we prepared to handle this kernel fault?
169 	 * We are almost certainly not prepared to handle instruction faults.
170 	 */
171 	if (!is_el1_instruction_abort(esr) && fixup_exception(regs))
172 		return;
173 
174 	/*
175 	 * No handler, we'll have to terminate things with extreme prejudice.
176 	 */
177 	bust_spinlocks(1);
178 	pr_alert("Unable to handle kernel %s at virtual address %08lx\n",
179 		 (addr < PAGE_SIZE) ? "NULL pointer dereference" :
180 		 "paging request", addr);
181 
182 	show_pte(mm, addr);
183 	die("Oops", regs, esr);
184 	bust_spinlocks(0);
185 	do_exit(SIGKILL);
186 }
187 
188 /*
189  * Something tried to access memory that isn't in our memory map. User mode
190  * accesses just cause a SIGSEGV
191  */
192 static void __do_user_fault(struct task_struct *tsk, unsigned long addr,
193 			    unsigned int esr, unsigned int sig, int code,
194 			    struct pt_regs *regs)
195 {
196 	struct siginfo si;
197 
198 	if (unhandled_signal(tsk, sig) && show_unhandled_signals_ratelimited()) {
199 		pr_info("%s[%d]: unhandled %s (%d) at 0x%08lx, esr 0x%03x\n",
200 			tsk->comm, task_pid_nr(tsk), fault_name(esr), sig,
201 			addr, esr);
202 		show_pte(tsk->mm, addr);
203 		show_regs(regs);
204 	}
205 
206 	tsk->thread.fault_address = addr;
207 	tsk->thread.fault_code = esr;
208 	si.si_signo = sig;
209 	si.si_errno = 0;
210 	si.si_code = code;
211 	si.si_addr = (void __user *)addr;
212 	force_sig_info(sig, &si, tsk);
213 }
214 
215 static void do_bad_area(unsigned long addr, unsigned int esr, struct pt_regs *regs)
216 {
217 	struct task_struct *tsk = current;
218 	struct mm_struct *mm = tsk->active_mm;
219 
220 	/*
221 	 * If we are in kernel mode at this point, we have no context to
222 	 * handle this fault with.
223 	 */
224 	if (user_mode(regs))
225 		__do_user_fault(tsk, addr, esr, SIGSEGV, SEGV_MAPERR, regs);
226 	else
227 		__do_kernel_fault(mm, addr, esr, regs);
228 }
229 
230 #define VM_FAULT_BADMAP		0x010000
231 #define VM_FAULT_BADACCESS	0x020000
232 
233 static int __do_page_fault(struct mm_struct *mm, unsigned long addr,
234 			   unsigned int mm_flags, unsigned long vm_flags,
235 			   struct task_struct *tsk)
236 {
237 	struct vm_area_struct *vma;
238 	int fault;
239 
240 	vma = find_vma(mm, addr);
241 	fault = VM_FAULT_BADMAP;
242 	if (unlikely(!vma))
243 		goto out;
244 	if (unlikely(vma->vm_start > addr))
245 		goto check_stack;
246 
247 	/*
248 	 * Ok, we have a good vm_area for this memory access, so we can handle
249 	 * it.
250 	 */
251 good_area:
252 	/*
253 	 * Check that the permissions on the VMA allow for the fault which
254 	 * occurred. If we encountered a write or exec fault, we must have
255 	 * appropriate permissions, otherwise we allow any permission.
256 	 */
257 	if (!(vma->vm_flags & vm_flags)) {
258 		fault = VM_FAULT_BADACCESS;
259 		goto out;
260 	}
261 
262 	return handle_mm_fault(vma, addr & PAGE_MASK, mm_flags);
263 
264 check_stack:
265 	if (vma->vm_flags & VM_GROWSDOWN && !expand_stack(vma, addr))
266 		goto good_area;
267 out:
268 	return fault;
269 }
270 
271 static inline bool is_permission_fault(unsigned int esr)
272 {
273 	unsigned int ec       = ESR_ELx_EC(esr);
274 	unsigned int fsc_type = esr & ESR_ELx_FSC_TYPE;
275 
276 	return (ec == ESR_ELx_EC_DABT_CUR && fsc_type == ESR_ELx_FSC_PERM) ||
277 	       (ec == ESR_ELx_EC_IABT_CUR && fsc_type == ESR_ELx_FSC_PERM);
278 }
279 
280 static bool is_el0_instruction_abort(unsigned int esr)
281 {
282 	return ESR_ELx_EC(esr) == ESR_ELx_EC_IABT_LOW;
283 }
284 
285 static int __kprobes do_page_fault(unsigned long addr, unsigned int esr,
286 				   struct pt_regs *regs)
287 {
288 	struct task_struct *tsk;
289 	struct mm_struct *mm;
290 	int fault, sig, code;
291 	unsigned long vm_flags = VM_READ | VM_WRITE | VM_EXEC;
292 	unsigned int mm_flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
293 
294 	if (notify_page_fault(regs, esr))
295 		return 0;
296 
297 	tsk = current;
298 	mm  = tsk->mm;
299 
300 	/*
301 	 * If we're in an interrupt or have no user context, we must not take
302 	 * the fault.
303 	 */
304 	if (faulthandler_disabled() || !mm)
305 		goto no_context;
306 
307 	if (user_mode(regs))
308 		mm_flags |= FAULT_FLAG_USER;
309 
310 	if (is_el0_instruction_abort(esr)) {
311 		vm_flags = VM_EXEC;
312 	} else if ((esr & ESR_ELx_WNR) && !(esr & ESR_ELx_CM)) {
313 		vm_flags = VM_WRITE;
314 		mm_flags |= FAULT_FLAG_WRITE;
315 	}
316 
317 	if (is_permission_fault(esr) && (addr < USER_DS)) {
318 		/* regs->orig_addr_limit may be 0 if we entered from EL0 */
319 		if (regs->orig_addr_limit == KERNEL_DS)
320 			die("Accessing user space memory with fs=KERNEL_DS", regs, esr);
321 
322 		if (is_el1_instruction_abort(esr))
323 			die("Attempting to execute userspace memory", regs, esr);
324 
325 		if (!search_exception_tables(regs->pc))
326 			die("Accessing user space memory outside uaccess.h routines", regs, esr);
327 	}
328 
329 	/*
330 	 * As per x86, we may deadlock here. However, since the kernel only
331 	 * validly references user space from well defined areas of the code,
332 	 * we can bug out early if this is from code which shouldn't.
333 	 */
334 	if (!down_read_trylock(&mm->mmap_sem)) {
335 		if (!user_mode(regs) && !search_exception_tables(regs->pc))
336 			goto no_context;
337 retry:
338 		down_read(&mm->mmap_sem);
339 	} else {
340 		/*
341 		 * The above down_read_trylock() might have succeeded in which
342 		 * case, we'll have missed the might_sleep() from down_read().
343 		 */
344 		might_sleep();
345 #ifdef CONFIG_DEBUG_VM
346 		if (!user_mode(regs) && !search_exception_tables(regs->pc))
347 			goto no_context;
348 #endif
349 	}
350 
351 	fault = __do_page_fault(mm, addr, mm_flags, vm_flags, tsk);
352 
353 	/*
354 	 * If we need to retry but a fatal signal is pending, handle the
355 	 * signal first. We do not need to release the mmap_sem because it
356 	 * would already be released in __lock_page_or_retry in mm/filemap.c.
357 	 */
358 	if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
359 		return 0;
360 
361 	/*
362 	 * Major/minor page fault accounting is only done on the initial
363 	 * attempt. If we go through a retry, it is extremely likely that the
364 	 * page will be found in page cache at that point.
365 	 */
366 
367 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
368 	if (mm_flags & FAULT_FLAG_ALLOW_RETRY) {
369 		if (fault & VM_FAULT_MAJOR) {
370 			tsk->maj_flt++;
371 			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs,
372 				      addr);
373 		} else {
374 			tsk->min_flt++;
375 			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs,
376 				      addr);
377 		}
378 		if (fault & VM_FAULT_RETRY) {
379 			/*
380 			 * Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk of
381 			 * starvation.
382 			 */
383 			mm_flags &= ~FAULT_FLAG_ALLOW_RETRY;
384 			mm_flags |= FAULT_FLAG_TRIED;
385 			goto retry;
386 		}
387 	}
388 
389 	up_read(&mm->mmap_sem);
390 
391 	/*
392 	 * Handle the "normal" case first - VM_FAULT_MAJOR
393 	 */
394 	if (likely(!(fault & (VM_FAULT_ERROR | VM_FAULT_BADMAP |
395 			      VM_FAULT_BADACCESS))))
396 		return 0;
397 
398 	/*
399 	 * If we are in kernel mode at this point, we have no context to
400 	 * handle this fault with.
401 	 */
402 	if (!user_mode(regs))
403 		goto no_context;
404 
405 	if (fault & VM_FAULT_OOM) {
406 		/*
407 		 * We ran out of memory, call the OOM killer, and return to
408 		 * userspace (which will retry the fault, or kill us if we got
409 		 * oom-killed).
410 		 */
411 		pagefault_out_of_memory();
412 		return 0;
413 	}
414 
415 	if (fault & VM_FAULT_SIGBUS) {
416 		/*
417 		 * We had some memory, but were unable to successfully fix up
418 		 * this page fault.
419 		 */
420 		sig = SIGBUS;
421 		code = BUS_ADRERR;
422 	} else {
423 		/*
424 		 * Something tried to access memory that isn't in our memory
425 		 * map.
426 		 */
427 		sig = SIGSEGV;
428 		code = fault == VM_FAULT_BADACCESS ?
429 			SEGV_ACCERR : SEGV_MAPERR;
430 	}
431 
432 	__do_user_fault(tsk, addr, esr, sig, code, regs);
433 	return 0;
434 
435 no_context:
436 	__do_kernel_fault(mm, addr, esr, regs);
437 	return 0;
438 }
439 
440 /*
441  * First Level Translation Fault Handler
442  *
443  * We enter here because the first level page table doesn't contain a valid
444  * entry for the address.
445  *
446  * If the address is in kernel space (>= TASK_SIZE), then we are probably
447  * faulting in the vmalloc() area.
448  *
449  * If the init_task's first level page tables contains the relevant entry, we
450  * copy the it to this task.  If not, we send the process a signal, fixup the
451  * exception, or oops the kernel.
452  *
453  * NOTE! We MUST NOT take any locks for this case. We may be in an interrupt
454  * or a critical region, and should only copy the information from the master
455  * page table, nothing more.
456  */
457 static int __kprobes do_translation_fault(unsigned long addr,
458 					  unsigned int esr,
459 					  struct pt_regs *regs)
460 {
461 	if (addr < TASK_SIZE)
462 		return do_page_fault(addr, esr, regs);
463 
464 	do_bad_area(addr, esr, regs);
465 	return 0;
466 }
467 
468 static int do_alignment_fault(unsigned long addr, unsigned int esr,
469 			      struct pt_regs *regs)
470 {
471 	do_bad_area(addr, esr, regs);
472 	return 0;
473 }
474 
475 /*
476  * This abort handler always returns "fault".
477  */
478 static int do_bad(unsigned long addr, unsigned int esr, struct pt_regs *regs)
479 {
480 	return 1;
481 }
482 
483 static const struct fault_info {
484 	int	(*fn)(unsigned long addr, unsigned int esr, struct pt_regs *regs);
485 	int	sig;
486 	int	code;
487 	const char *name;
488 } fault_info[] = {
489 	{ do_bad,		SIGBUS,  0,		"ttbr address size fault"	},
490 	{ do_bad,		SIGBUS,  0,		"level 1 address size fault"	},
491 	{ do_bad,		SIGBUS,  0,		"level 2 address size fault"	},
492 	{ do_bad,		SIGBUS,  0,		"level 3 address size fault"	},
493 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 0 translation fault"	},
494 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 1 translation fault"	},
495 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 2 translation fault"	},
496 	{ do_page_fault,	SIGSEGV, SEGV_MAPERR,	"level 3 translation fault"	},
497 	{ do_bad,		SIGBUS,  0,		"unknown 8"			},
498 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 1 access flag fault"	},
499 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 2 access flag fault"	},
500 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 3 access flag fault"	},
501 	{ do_bad,		SIGBUS,  0,		"unknown 12"			},
502 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 1 permission fault"	},
503 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 2 permission fault"	},
504 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 3 permission fault"	},
505 	{ do_bad,		SIGBUS,  0,		"synchronous external abort"	},
506 	{ do_bad,		SIGBUS,  0,		"unknown 17"			},
507 	{ do_bad,		SIGBUS,  0,		"unknown 18"			},
508 	{ do_bad,		SIGBUS,  0,		"unknown 19"			},
509 	{ do_bad,		SIGBUS,  0,		"synchronous abort (translation table walk)" },
510 	{ do_bad,		SIGBUS,  0,		"synchronous abort (translation table walk)" },
511 	{ do_bad,		SIGBUS,  0,		"synchronous abort (translation table walk)" },
512 	{ do_bad,		SIGBUS,  0,		"synchronous abort (translation table walk)" },
513 	{ do_bad,		SIGBUS,  0,		"synchronous parity error"	},
514 	{ do_bad,		SIGBUS,  0,		"unknown 25"			},
515 	{ do_bad,		SIGBUS,  0,		"unknown 26"			},
516 	{ do_bad,		SIGBUS,  0,		"unknown 27"			},
517 	{ do_bad,		SIGBUS,  0,		"synchronous parity error (translation table walk)" },
518 	{ do_bad,		SIGBUS,  0,		"synchronous parity error (translation table walk)" },
519 	{ do_bad,		SIGBUS,  0,		"synchronous parity error (translation table walk)" },
520 	{ do_bad,		SIGBUS,  0,		"synchronous parity error (translation table walk)" },
521 	{ do_bad,		SIGBUS,  0,		"unknown 32"			},
522 	{ do_alignment_fault,	SIGBUS,  BUS_ADRALN,	"alignment fault"		},
523 	{ do_bad,		SIGBUS,  0,		"unknown 34"			},
524 	{ do_bad,		SIGBUS,  0,		"unknown 35"			},
525 	{ do_bad,		SIGBUS,  0,		"unknown 36"			},
526 	{ do_bad,		SIGBUS,  0,		"unknown 37"			},
527 	{ do_bad,		SIGBUS,  0,		"unknown 38"			},
528 	{ do_bad,		SIGBUS,  0,		"unknown 39"			},
529 	{ do_bad,		SIGBUS,  0,		"unknown 40"			},
530 	{ do_bad,		SIGBUS,  0,		"unknown 41"			},
531 	{ do_bad,		SIGBUS,  0,		"unknown 42"			},
532 	{ do_bad,		SIGBUS,  0,		"unknown 43"			},
533 	{ do_bad,		SIGBUS,  0,		"unknown 44"			},
534 	{ do_bad,		SIGBUS,  0,		"unknown 45"			},
535 	{ do_bad,		SIGBUS,  0,		"unknown 46"			},
536 	{ do_bad,		SIGBUS,  0,		"unknown 47"			},
537 	{ do_bad,		SIGBUS,  0,		"TLB conflict abort"		},
538 	{ do_bad,		SIGBUS,  0,		"unknown 49"			},
539 	{ do_bad,		SIGBUS,  0,		"unknown 50"			},
540 	{ do_bad,		SIGBUS,  0,		"unknown 51"			},
541 	{ do_bad,		SIGBUS,  0,		"implementation fault (lockdown abort)" },
542 	{ do_bad,		SIGBUS,  0,		"implementation fault (unsupported exclusive)" },
543 	{ do_bad,		SIGBUS,  0,		"unknown 54"			},
544 	{ do_bad,		SIGBUS,  0,		"unknown 55"			},
545 	{ do_bad,		SIGBUS,  0,		"unknown 56"			},
546 	{ do_bad,		SIGBUS,  0,		"unknown 57"			},
547 	{ do_bad,		SIGBUS,  0,		"unknown 58" 			},
548 	{ do_bad,		SIGBUS,  0,		"unknown 59"			},
549 	{ do_bad,		SIGBUS,  0,		"unknown 60"			},
550 	{ do_bad,		SIGBUS,  0,		"section domain fault"		},
551 	{ do_bad,		SIGBUS,  0,		"page domain fault"		},
552 	{ do_bad,		SIGBUS,  0,		"unknown 63"			},
553 };
554 
555 static const char *fault_name(unsigned int esr)
556 {
557 	const struct fault_info *inf = fault_info + (esr & 63);
558 	return inf->name;
559 }
560 
561 /*
562  * Dispatch a data abort to the relevant handler.
563  */
564 asmlinkage void __exception do_mem_abort(unsigned long addr, unsigned int esr,
565 					 struct pt_regs *regs)
566 {
567 	const struct fault_info *inf = fault_info + (esr & 63);
568 	struct siginfo info;
569 
570 	if (!inf->fn(addr, esr, regs))
571 		return;
572 
573 	pr_alert("Unhandled fault: %s (0x%08x) at 0x%016lx\n",
574 		 inf->name, esr, addr);
575 
576 	info.si_signo = inf->sig;
577 	info.si_errno = 0;
578 	info.si_code  = inf->code;
579 	info.si_addr  = (void __user *)addr;
580 	arm64_notify_die("", regs, &info, esr);
581 }
582 
583 /*
584  * Handle stack alignment exceptions.
585  */
586 asmlinkage void __exception do_sp_pc_abort(unsigned long addr,
587 					   unsigned int esr,
588 					   struct pt_regs *regs)
589 {
590 	struct siginfo info;
591 	struct task_struct *tsk = current;
592 
593 	if (show_unhandled_signals && unhandled_signal(tsk, SIGBUS))
594 		pr_info_ratelimited("%s[%d]: %s exception: pc=%p sp=%p\n",
595 				    tsk->comm, task_pid_nr(tsk),
596 				    esr_get_class_string(esr), (void *)regs->pc,
597 				    (void *)regs->sp);
598 
599 	info.si_signo = SIGBUS;
600 	info.si_errno = 0;
601 	info.si_code  = BUS_ADRALN;
602 	info.si_addr  = (void __user *)addr;
603 	arm64_notify_die("Oops - SP/PC alignment exception", regs, &info, esr);
604 }
605 
606 int __init early_brk64(unsigned long addr, unsigned int esr,
607 		       struct pt_regs *regs);
608 
609 /*
610  * __refdata because early_brk64 is __init, but the reference to it is
611  * clobbered at arch_initcall time.
612  * See traps.c and debug-monitors.c:debug_traps_init().
613  */
614 static struct fault_info __refdata debug_fault_info[] = {
615 	{ do_bad,	SIGTRAP,	TRAP_HWBKPT,	"hardware breakpoint"	},
616 	{ do_bad,	SIGTRAP,	TRAP_HWBKPT,	"hardware single-step"	},
617 	{ do_bad,	SIGTRAP,	TRAP_HWBKPT,	"hardware watchpoint"	},
618 	{ do_bad,	SIGBUS,		0,		"unknown 3"		},
619 	{ do_bad,	SIGTRAP,	TRAP_BRKPT,	"aarch32 BKPT"		},
620 	{ do_bad,	SIGTRAP,	0,		"aarch32 vector catch"	},
621 	{ early_brk64,	SIGTRAP,	TRAP_BRKPT,	"aarch64 BRK"		},
622 	{ do_bad,	SIGBUS,		0,		"unknown 7"		},
623 };
624 
625 void __init hook_debug_fault_code(int nr,
626 				  int (*fn)(unsigned long, unsigned int, struct pt_regs *),
627 				  int sig, int code, const char *name)
628 {
629 	BUG_ON(nr < 0 || nr >= ARRAY_SIZE(debug_fault_info));
630 
631 	debug_fault_info[nr].fn		= fn;
632 	debug_fault_info[nr].sig	= sig;
633 	debug_fault_info[nr].code	= code;
634 	debug_fault_info[nr].name	= name;
635 }
636 
637 asmlinkage int __exception do_debug_exception(unsigned long addr,
638 					      unsigned int esr,
639 					      struct pt_regs *regs)
640 {
641 	const struct fault_info *inf = debug_fault_info + DBG_ESR_EVT(esr);
642 	struct siginfo info;
643 	int rv;
644 
645 	/*
646 	 * Tell lockdep we disabled irqs in entry.S. Do nothing if they were
647 	 * already disabled to preserve the last enabled/disabled addresses.
648 	 */
649 	if (interrupts_enabled(regs))
650 		trace_hardirqs_off();
651 
652 	if (!inf->fn(addr, esr, regs)) {
653 		rv = 1;
654 	} else {
655 		pr_alert("Unhandled debug exception: %s (0x%08x) at 0x%016lx\n",
656 			 inf->name, esr, addr);
657 
658 		info.si_signo = inf->sig;
659 		info.si_errno = 0;
660 		info.si_code  = inf->code;
661 		info.si_addr  = (void __user *)addr;
662 		arm64_notify_die("", regs, &info, 0);
663 		rv = 0;
664 	}
665 
666 	if (interrupts_enabled(regs))
667 		trace_hardirqs_on();
668 
669 	return rv;
670 }
671 NOKPROBE_SYMBOL(do_debug_exception);
672 
673 #ifdef CONFIG_ARM64_PAN
674 void cpu_enable_pan(void *__unused)
675 {
676 	config_sctlr_el1(SCTLR_EL1_SPAN, 0);
677 }
678 #endif /* CONFIG_ARM64_PAN */
679 
680 #ifdef CONFIG_ARM64_UAO
681 /*
682  * Kernel threads have fs=KERNEL_DS by default, and don't need to call
683  * set_fs(), devtmpfs in particular relies on this behaviour.
684  * We need to enable the feature at runtime (instead of adding it to
685  * PSR_MODE_EL1h) as the feature may not be implemented by the cpu.
686  */
687 void cpu_enable_uao(void *__unused)
688 {
689 	asm(SET_PSTATE_UAO(1));
690 }
691 #endif /* CONFIG_ARM64_UAO */
692