xref: /openbmc/linux/arch/arm64/mm/fault.c (revision 93032e31)
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/extable.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 #include <linux/preempt.h>
33 
34 #include <asm/bug.h>
35 #include <asm/cpufeature.h>
36 #include <asm/exception.h>
37 #include <asm/debug-monitors.h>
38 #include <asm/esr.h>
39 #include <asm/sysreg.h>
40 #include <asm/system_misc.h>
41 #include <asm/pgtable.h>
42 #include <asm/tlbflush.h>
43 
44 static const char *fault_name(unsigned int esr);
45 
46 #ifdef CONFIG_KPROBES
47 static inline int notify_page_fault(struct pt_regs *regs, unsigned int esr)
48 {
49 	int ret = 0;
50 
51 	/* kprobe_running() needs smp_processor_id() */
52 	if (!user_mode(regs)) {
53 		preempt_disable();
54 		if (kprobe_running() && kprobe_fault_handler(regs, esr))
55 			ret = 1;
56 		preempt_enable();
57 	}
58 
59 	return ret;
60 }
61 #else
62 static inline int notify_page_fault(struct pt_regs *regs, unsigned int esr)
63 {
64 	return 0;
65 }
66 #endif
67 
68 /*
69  * Dump out the page tables associated with 'addr' in mm 'mm'.
70  */
71 void show_pte(struct mm_struct *mm, unsigned long addr)
72 {
73 	pgd_t *pgd;
74 
75 	if (!mm)
76 		mm = &init_mm;
77 
78 	pr_alert("pgd = %p\n", mm->pgd);
79 	pgd = pgd_offset(mm, addr);
80 	pr_alert("[%08lx] *pgd=%016llx", addr, pgd_val(*pgd));
81 
82 	do {
83 		pud_t *pud;
84 		pmd_t *pmd;
85 		pte_t *pte;
86 
87 		if (pgd_none(*pgd) || pgd_bad(*pgd))
88 			break;
89 
90 		pud = pud_offset(pgd, addr);
91 		printk(", *pud=%016llx", pud_val(*pud));
92 		if (pud_none(*pud) || pud_bad(*pud))
93 			break;
94 
95 		pmd = pmd_offset(pud, addr);
96 		printk(", *pmd=%016llx", pmd_val(*pmd));
97 		if (pmd_none(*pmd) || pmd_bad(*pmd))
98 			break;
99 
100 		pte = pte_offset_map(pmd, addr);
101 		printk(", *pte=%016llx", pte_val(*pte));
102 		pte_unmap(pte);
103 	} while(0);
104 
105 	printk("\n");
106 }
107 
108 #ifdef CONFIG_ARM64_HW_AFDBM
109 /*
110  * This function sets the access flags (dirty, accessed), as well as write
111  * permission, and only to a more permissive setting.
112  *
113  * It needs to cope with hardware update of the accessed/dirty state by other
114  * agents in the system and can safely skip the __sync_icache_dcache() call as,
115  * like set_pte_at(), the PTE is never changed from no-exec to exec here.
116  *
117  * Returns whether or not the PTE actually changed.
118  */
119 int ptep_set_access_flags(struct vm_area_struct *vma,
120 			  unsigned long address, pte_t *ptep,
121 			  pte_t entry, int dirty)
122 {
123 	pteval_t old_pteval;
124 	unsigned int tmp;
125 
126 	if (pte_same(*ptep, entry))
127 		return 0;
128 
129 	/* only preserve the access flags and write permission */
130 	pte_val(entry) &= PTE_AF | PTE_WRITE | PTE_DIRTY;
131 
132 	/*
133 	 * PTE_RDONLY is cleared by default in the asm below, so set it in
134 	 * back if necessary (read-only or clean PTE).
135 	 */
136 	if (!pte_write(entry) || !pte_sw_dirty(entry))
137 		pte_val(entry) |= PTE_RDONLY;
138 
139 	/*
140 	 * Setting the flags must be done atomically to avoid racing with the
141 	 * hardware update of the access/dirty state.
142 	 */
143 	asm volatile("//	ptep_set_access_flags\n"
144 	"	prfm	pstl1strm, %2\n"
145 	"1:	ldxr	%0, %2\n"
146 	"	and	%0, %0, %3		// clear PTE_RDONLY\n"
147 	"	orr	%0, %0, %4		// set flags\n"
148 	"	stxr	%w1, %0, %2\n"
149 	"	cbnz	%w1, 1b\n"
150 	: "=&r" (old_pteval), "=&r" (tmp), "+Q" (pte_val(*ptep))
151 	: "L" (~PTE_RDONLY), "r" (pte_val(entry)));
152 
153 	flush_tlb_fix_spurious_fault(vma, address);
154 	return 1;
155 }
156 #endif
157 
158 static bool is_el1_instruction_abort(unsigned int esr)
159 {
160 	return ESR_ELx_EC(esr) == ESR_ELx_EC_IABT_CUR;
161 }
162 
163 /*
164  * The kernel tried to access some page that wasn't present.
165  */
166 static void __do_kernel_fault(struct mm_struct *mm, unsigned long addr,
167 			      unsigned int esr, struct pt_regs *regs)
168 {
169 	/*
170 	 * Are we prepared to handle this kernel fault?
171 	 * We are almost certainly not prepared to handle instruction faults.
172 	 */
173 	if (!is_el1_instruction_abort(esr) && fixup_exception(regs))
174 		return;
175 
176 	/*
177 	 * No handler, we'll have to terminate things with extreme prejudice.
178 	 */
179 	bust_spinlocks(1);
180 	pr_alert("Unable to handle kernel %s at virtual address %08lx\n",
181 		 (addr < PAGE_SIZE) ? "NULL pointer dereference" :
182 		 "paging request", addr);
183 
184 	show_pte(mm, addr);
185 	die("Oops", regs, esr);
186 	bust_spinlocks(0);
187 	do_exit(SIGKILL);
188 }
189 
190 /*
191  * Something tried to access memory that isn't in our memory map. User mode
192  * accesses just cause a SIGSEGV
193  */
194 static void __do_user_fault(struct task_struct *tsk, unsigned long addr,
195 			    unsigned int esr, unsigned int sig, int code,
196 			    struct pt_regs *regs)
197 {
198 	struct siginfo si;
199 
200 	if (unhandled_signal(tsk, sig) && show_unhandled_signals_ratelimited()) {
201 		pr_info("%s[%d]: unhandled %s (%d) at 0x%08lx, esr 0x%03x\n",
202 			tsk->comm, task_pid_nr(tsk), fault_name(esr), sig,
203 			addr, esr);
204 		show_pte(tsk->mm, addr);
205 		show_regs(regs);
206 	}
207 
208 	tsk->thread.fault_address = addr;
209 	tsk->thread.fault_code = esr;
210 	si.si_signo = sig;
211 	si.si_errno = 0;
212 	si.si_code = code;
213 	si.si_addr = (void __user *)addr;
214 	force_sig_info(sig, &si, tsk);
215 }
216 
217 static void do_bad_area(unsigned long addr, unsigned int esr, struct pt_regs *regs)
218 {
219 	struct task_struct *tsk = current;
220 	struct mm_struct *mm = tsk->active_mm;
221 
222 	/*
223 	 * If we are in kernel mode at this point, we have no context to
224 	 * handle this fault with.
225 	 */
226 	if (user_mode(regs))
227 		__do_user_fault(tsk, addr, esr, SIGSEGV, SEGV_MAPERR, regs);
228 	else
229 		__do_kernel_fault(mm, addr, esr, regs);
230 }
231 
232 #define VM_FAULT_BADMAP		0x010000
233 #define VM_FAULT_BADACCESS	0x020000
234 
235 static int __do_page_fault(struct mm_struct *mm, unsigned long addr,
236 			   unsigned int mm_flags, unsigned long vm_flags,
237 			   struct task_struct *tsk)
238 {
239 	struct vm_area_struct *vma;
240 	int fault;
241 
242 	vma = find_vma(mm, addr);
243 	fault = VM_FAULT_BADMAP;
244 	if (unlikely(!vma))
245 		goto out;
246 	if (unlikely(vma->vm_start > addr))
247 		goto check_stack;
248 
249 	/*
250 	 * Ok, we have a good vm_area for this memory access, so we can handle
251 	 * it.
252 	 */
253 good_area:
254 	/*
255 	 * Check that the permissions on the VMA allow for the fault which
256 	 * occurred.
257 	 */
258 	if (!(vma->vm_flags & vm_flags)) {
259 		fault = VM_FAULT_BADACCESS;
260 		goto out;
261 	}
262 
263 	return handle_mm_fault(vma, addr & PAGE_MASK, mm_flags);
264 
265 check_stack:
266 	if (vma->vm_flags & VM_GROWSDOWN && !expand_stack(vma, addr))
267 		goto good_area;
268 out:
269 	return fault;
270 }
271 
272 static inline bool is_permission_fault(unsigned int esr)
273 {
274 	unsigned int ec       = ESR_ELx_EC(esr);
275 	unsigned int fsc_type = esr & ESR_ELx_FSC_TYPE;
276 
277 	return (ec == ESR_ELx_EC_DABT_CUR && fsc_type == ESR_ELx_FSC_PERM) ||
278 	       (ec == ESR_ELx_EC_IABT_CUR && fsc_type == ESR_ELx_FSC_PERM);
279 }
280 
281 static bool is_el0_instruction_abort(unsigned int esr)
282 {
283 	return ESR_ELx_EC(esr) == ESR_ELx_EC_IABT_LOW;
284 }
285 
286 static int __kprobes do_page_fault(unsigned long addr, unsigned int esr,
287 				   struct pt_regs *regs)
288 {
289 	struct task_struct *tsk;
290 	struct mm_struct *mm;
291 	int fault, sig, code;
292 	unsigned long vm_flags = VM_READ | VM_WRITE;
293 	unsigned int mm_flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
294 
295 	if (notify_page_fault(regs, esr))
296 		return 0;
297 
298 	tsk = current;
299 	mm  = tsk->mm;
300 
301 	/*
302 	 * If we're in an interrupt or have no user context, we must not take
303 	 * the fault.
304 	 */
305 	if (faulthandler_disabled() || !mm)
306 		goto no_context;
307 
308 	if (user_mode(regs))
309 		mm_flags |= FAULT_FLAG_USER;
310 
311 	if (is_el0_instruction_abort(esr)) {
312 		vm_flags = VM_EXEC;
313 	} else if ((esr & ESR_ELx_WNR) && !(esr & ESR_ELx_CM)) {
314 		vm_flags = VM_WRITE;
315 		mm_flags |= FAULT_FLAG_WRITE;
316 	}
317 
318 	if (is_permission_fault(esr) && (addr < USER_DS)) {
319 		/* regs->orig_addr_limit may be 0 if we entered from EL0 */
320 		if (regs->orig_addr_limit == KERNEL_DS)
321 			die("Accessing user space memory with fs=KERNEL_DS", regs, esr);
322 
323 		if (is_el1_instruction_abort(esr))
324 			die("Attempting to execute userspace memory", regs, esr);
325 
326 		if (!search_exception_tables(regs->pc))
327 			die("Accessing user space memory outside uaccess.h routines", regs, esr);
328 	}
329 
330 	/*
331 	 * As per x86, we may deadlock here. However, since the kernel only
332 	 * validly references user space from well defined areas of the code,
333 	 * we can bug out early if this is from code which shouldn't.
334 	 */
335 	if (!down_read_trylock(&mm->mmap_sem)) {
336 		if (!user_mode(regs) && !search_exception_tables(regs->pc))
337 			goto no_context;
338 retry:
339 		down_read(&mm->mmap_sem);
340 	} else {
341 		/*
342 		 * The above down_read_trylock() might have succeeded in which
343 		 * case, we'll have missed the might_sleep() from down_read().
344 		 */
345 		might_sleep();
346 #ifdef CONFIG_DEBUG_VM
347 		if (!user_mode(regs) && !search_exception_tables(regs->pc))
348 			goto no_context;
349 #endif
350 	}
351 
352 	fault = __do_page_fault(mm, addr, mm_flags, vm_flags, tsk);
353 
354 	/*
355 	 * If we need to retry but a fatal signal is pending, handle the
356 	 * signal first. We do not need to release the mmap_sem because it
357 	 * would already be released in __lock_page_or_retry in mm/filemap.c.
358 	 */
359 	if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
360 		return 0;
361 
362 	/*
363 	 * Major/minor page fault accounting is only done on the initial
364 	 * attempt. If we go through a retry, it is extremely likely that the
365 	 * page will be found in page cache at that point.
366 	 */
367 
368 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
369 	if (mm_flags & FAULT_FLAG_ALLOW_RETRY) {
370 		if (fault & VM_FAULT_MAJOR) {
371 			tsk->maj_flt++;
372 			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs,
373 				      addr);
374 		} else {
375 			tsk->min_flt++;
376 			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs,
377 				      addr);
378 		}
379 		if (fault & VM_FAULT_RETRY) {
380 			/*
381 			 * Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk of
382 			 * starvation.
383 			 */
384 			mm_flags &= ~FAULT_FLAG_ALLOW_RETRY;
385 			mm_flags |= FAULT_FLAG_TRIED;
386 			goto retry;
387 		}
388 	}
389 
390 	up_read(&mm->mmap_sem);
391 
392 	/*
393 	 * Handle the "normal" case first - VM_FAULT_MAJOR
394 	 */
395 	if (likely(!(fault & (VM_FAULT_ERROR | VM_FAULT_BADMAP |
396 			      VM_FAULT_BADACCESS))))
397 		return 0;
398 
399 	/*
400 	 * If we are in kernel mode at this point, we have no context to
401 	 * handle this fault with.
402 	 */
403 	if (!user_mode(regs))
404 		goto no_context;
405 
406 	if (fault & VM_FAULT_OOM) {
407 		/*
408 		 * We ran out of memory, call the OOM killer, and return to
409 		 * userspace (which will retry the fault, or kill us if we got
410 		 * oom-killed).
411 		 */
412 		pagefault_out_of_memory();
413 		return 0;
414 	}
415 
416 	if (fault & VM_FAULT_SIGBUS) {
417 		/*
418 		 * We had some memory, but were unable to successfully fix up
419 		 * this page fault.
420 		 */
421 		sig = SIGBUS;
422 		code = BUS_ADRERR;
423 	} else {
424 		/*
425 		 * Something tried to access memory that isn't in our memory
426 		 * map.
427 		 */
428 		sig = SIGSEGV;
429 		code = fault == VM_FAULT_BADACCESS ?
430 			SEGV_ACCERR : SEGV_MAPERR;
431 	}
432 
433 	__do_user_fault(tsk, addr, esr, sig, code, regs);
434 	return 0;
435 
436 no_context:
437 	__do_kernel_fault(mm, addr, esr, regs);
438 	return 0;
439 }
440 
441 /*
442  * First Level Translation Fault Handler
443  *
444  * We enter here because the first level page table doesn't contain a valid
445  * entry for the address.
446  *
447  * If the address is in kernel space (>= TASK_SIZE), then we are probably
448  * faulting in the vmalloc() area.
449  *
450  * If the init_task's first level page tables contains the relevant entry, we
451  * copy the it to this task.  If not, we send the process a signal, fixup the
452  * exception, or oops the kernel.
453  *
454  * NOTE! We MUST NOT take any locks for this case. We may be in an interrupt
455  * or a critical region, and should only copy the information from the master
456  * page table, nothing more.
457  */
458 static int __kprobes do_translation_fault(unsigned long addr,
459 					  unsigned int esr,
460 					  struct pt_regs *regs)
461 {
462 	if (addr < TASK_SIZE)
463 		return do_page_fault(addr, esr, regs);
464 
465 	do_bad_area(addr, esr, regs);
466 	return 0;
467 }
468 
469 static int do_alignment_fault(unsigned long addr, unsigned int esr,
470 			      struct pt_regs *regs)
471 {
472 	do_bad_area(addr, esr, regs);
473 	return 0;
474 }
475 
476 /*
477  * This abort handler always returns "fault".
478  */
479 static int do_bad(unsigned long addr, unsigned int esr, struct pt_regs *regs)
480 {
481 	return 1;
482 }
483 
484 static const struct fault_info {
485 	int	(*fn)(unsigned long addr, unsigned int esr, struct pt_regs *regs);
486 	int	sig;
487 	int	code;
488 	const char *name;
489 } fault_info[] = {
490 	{ do_bad,		SIGBUS,  0,		"ttbr address size fault"	},
491 	{ do_bad,		SIGBUS,  0,		"level 1 address size fault"	},
492 	{ do_bad,		SIGBUS,  0,		"level 2 address size fault"	},
493 	{ do_bad,		SIGBUS,  0,		"level 3 address size fault"	},
494 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 0 translation fault"	},
495 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 1 translation fault"	},
496 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"level 2 translation fault"	},
497 	{ do_page_fault,	SIGSEGV, SEGV_MAPERR,	"level 3 translation fault"	},
498 	{ do_bad,		SIGBUS,  0,		"unknown 8"			},
499 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 1 access flag fault"	},
500 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 2 access flag fault"	},
501 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 3 access flag fault"	},
502 	{ do_bad,		SIGBUS,  0,		"unknown 12"			},
503 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 1 permission fault"	},
504 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 2 permission fault"	},
505 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"level 3 permission fault"	},
506 	{ do_bad,		SIGBUS,  0,		"synchronous external abort"	},
507 	{ do_bad,		SIGBUS,  0,		"unknown 17"			},
508 	{ do_bad,		SIGBUS,  0,		"unknown 18"			},
509 	{ do_bad,		SIGBUS,  0,		"unknown 19"			},
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 abort (translation table walk)" },
514 	{ do_bad,		SIGBUS,  0,		"synchronous parity error"	},
515 	{ do_bad,		SIGBUS,  0,		"unknown 25"			},
516 	{ do_bad,		SIGBUS,  0,		"unknown 26"			},
517 	{ do_bad,		SIGBUS,  0,		"unknown 27"			},
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,		"synchronous parity error (translation table walk)" },
522 	{ do_bad,		SIGBUS,  0,		"unknown 32"			},
523 	{ do_alignment_fault,	SIGBUS,  BUS_ADRALN,	"alignment fault"		},
524 	{ do_bad,		SIGBUS,  0,		"unknown 34"			},
525 	{ do_bad,		SIGBUS,  0,		"unknown 35"			},
526 	{ do_bad,		SIGBUS,  0,		"unknown 36"			},
527 	{ do_bad,		SIGBUS,  0,		"unknown 37"			},
528 	{ do_bad,		SIGBUS,  0,		"unknown 38"			},
529 	{ do_bad,		SIGBUS,  0,		"unknown 39"			},
530 	{ do_bad,		SIGBUS,  0,		"unknown 40"			},
531 	{ do_bad,		SIGBUS,  0,		"unknown 41"			},
532 	{ do_bad,		SIGBUS,  0,		"unknown 42"			},
533 	{ do_bad,		SIGBUS,  0,		"unknown 43"			},
534 	{ do_bad,		SIGBUS,  0,		"unknown 44"			},
535 	{ do_bad,		SIGBUS,  0,		"unknown 45"			},
536 	{ do_bad,		SIGBUS,  0,		"unknown 46"			},
537 	{ do_bad,		SIGBUS,  0,		"unknown 47"			},
538 	{ do_bad,		SIGBUS,  0,		"TLB conflict abort"		},
539 	{ do_bad,		SIGBUS,  0,		"unknown 49"			},
540 	{ do_bad,		SIGBUS,  0,		"unknown 50"			},
541 	{ do_bad,		SIGBUS,  0,		"unknown 51"			},
542 	{ do_bad,		SIGBUS,  0,		"implementation fault (lockdown abort)" },
543 	{ do_bad,		SIGBUS,  0,		"implementation fault (unsupported exclusive)" },
544 	{ do_bad,		SIGBUS,  0,		"unknown 54"			},
545 	{ do_bad,		SIGBUS,  0,		"unknown 55"			},
546 	{ do_bad,		SIGBUS,  0,		"unknown 56"			},
547 	{ do_bad,		SIGBUS,  0,		"unknown 57"			},
548 	{ do_bad,		SIGBUS,  0,		"unknown 58" 			},
549 	{ do_bad,		SIGBUS,  0,		"unknown 59"			},
550 	{ do_bad,		SIGBUS,  0,		"unknown 60"			},
551 	{ do_bad,		SIGBUS,  0,		"section domain fault"		},
552 	{ do_bad,		SIGBUS,  0,		"page domain fault"		},
553 	{ do_bad,		SIGBUS,  0,		"unknown 63"			},
554 };
555 
556 static const char *fault_name(unsigned int esr)
557 {
558 	const struct fault_info *inf = fault_info + (esr & 63);
559 	return inf->name;
560 }
561 
562 /*
563  * Dispatch a data abort to the relevant handler.
564  */
565 asmlinkage void __exception do_mem_abort(unsigned long addr, unsigned int esr,
566 					 struct pt_regs *regs)
567 {
568 	const struct fault_info *inf = fault_info + (esr & 63);
569 	struct siginfo info;
570 
571 	if (!inf->fn(addr, esr, regs))
572 		return;
573 
574 	pr_alert("Unhandled fault: %s (0x%08x) at 0x%016lx\n",
575 		 inf->name, esr, addr);
576 
577 	info.si_signo = inf->sig;
578 	info.si_errno = 0;
579 	info.si_code  = inf->code;
580 	info.si_addr  = (void __user *)addr;
581 	arm64_notify_die("", regs, &info, esr);
582 }
583 
584 /*
585  * Handle stack alignment exceptions.
586  */
587 asmlinkage void __exception do_sp_pc_abort(unsigned long addr,
588 					   unsigned int esr,
589 					   struct pt_regs *regs)
590 {
591 	struct siginfo info;
592 	struct task_struct *tsk = current;
593 
594 	if (show_unhandled_signals && unhandled_signal(tsk, SIGBUS))
595 		pr_info_ratelimited("%s[%d]: %s exception: pc=%p sp=%p\n",
596 				    tsk->comm, task_pid_nr(tsk),
597 				    esr_get_class_string(esr), (void *)regs->pc,
598 				    (void *)regs->sp);
599 
600 	info.si_signo = SIGBUS;
601 	info.si_errno = 0;
602 	info.si_code  = BUS_ADRALN;
603 	info.si_addr  = (void __user *)addr;
604 	arm64_notify_die("Oops - SP/PC alignment exception", regs, &info, esr);
605 }
606 
607 int __init early_brk64(unsigned long addr, unsigned int esr,
608 		       struct pt_regs *regs);
609 
610 /*
611  * __refdata because early_brk64 is __init, but the reference to it is
612  * clobbered at arch_initcall time.
613  * See traps.c and debug-monitors.c:debug_traps_init().
614  */
615 static struct fault_info __refdata debug_fault_info[] = {
616 	{ do_bad,	SIGTRAP,	TRAP_HWBKPT,	"hardware breakpoint"	},
617 	{ do_bad,	SIGTRAP,	TRAP_HWBKPT,	"hardware single-step"	},
618 	{ do_bad,	SIGTRAP,	TRAP_HWBKPT,	"hardware watchpoint"	},
619 	{ do_bad,	SIGBUS,		0,		"unknown 3"		},
620 	{ do_bad,	SIGTRAP,	TRAP_BRKPT,	"aarch32 BKPT"		},
621 	{ do_bad,	SIGTRAP,	0,		"aarch32 vector catch"	},
622 	{ early_brk64,	SIGTRAP,	TRAP_BRKPT,	"aarch64 BRK"		},
623 	{ do_bad,	SIGBUS,		0,		"unknown 7"		},
624 };
625 
626 void __init hook_debug_fault_code(int nr,
627 				  int (*fn)(unsigned long, unsigned int, struct pt_regs *),
628 				  int sig, int code, const char *name)
629 {
630 	BUG_ON(nr < 0 || nr >= ARRAY_SIZE(debug_fault_info));
631 
632 	debug_fault_info[nr].fn		= fn;
633 	debug_fault_info[nr].sig	= sig;
634 	debug_fault_info[nr].code	= code;
635 	debug_fault_info[nr].name	= name;
636 }
637 
638 asmlinkage int __exception do_debug_exception(unsigned long addr,
639 					      unsigned int esr,
640 					      struct pt_regs *regs)
641 {
642 	const struct fault_info *inf = debug_fault_info + DBG_ESR_EVT(esr);
643 	struct siginfo info;
644 	int rv;
645 
646 	/*
647 	 * Tell lockdep we disabled irqs in entry.S. Do nothing if they were
648 	 * already disabled to preserve the last enabled/disabled addresses.
649 	 */
650 	if (interrupts_enabled(regs))
651 		trace_hardirqs_off();
652 
653 	if (!inf->fn(addr, esr, regs)) {
654 		rv = 1;
655 	} else {
656 		pr_alert("Unhandled debug exception: %s (0x%08x) at 0x%016lx\n",
657 			 inf->name, esr, addr);
658 
659 		info.si_signo = inf->sig;
660 		info.si_errno = 0;
661 		info.si_code  = inf->code;
662 		info.si_addr  = (void __user *)addr;
663 		arm64_notify_die("", regs, &info, 0);
664 		rv = 0;
665 	}
666 
667 	if (interrupts_enabled(regs))
668 		trace_hardirqs_on();
669 
670 	return rv;
671 }
672 NOKPROBE_SYMBOL(do_debug_exception);
673 
674 #ifdef CONFIG_ARM64_PAN
675 int cpu_enable_pan(void *__unused)
676 {
677 	/*
678 	 * We modify PSTATE. This won't work from irq context as the PSTATE
679 	 * is discarded once we return from the exception.
680 	 */
681 	WARN_ON_ONCE(in_interrupt());
682 
683 	config_sctlr_el1(SCTLR_EL1_SPAN, 0);
684 	asm(SET_PSTATE_PAN(1));
685 	return 0;
686 }
687 #endif /* CONFIG_ARM64_PAN */
688 
689 #ifdef CONFIG_ARM64_UAO
690 /*
691  * Kernel threads have fs=KERNEL_DS by default, and don't need to call
692  * set_fs(), devtmpfs in particular relies on this behaviour.
693  * We need to enable the feature at runtime (instead of adding it to
694  * PSR_MODE_EL1h) as the feature may not be implemented by the cpu.
695  */
696 int cpu_enable_uao(void *__unused)
697 {
698 	asm(SET_PSTATE_UAO(1));
699 	return 0;
700 }
701 #endif /* CONFIG_ARM64_UAO */
702