1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *  linux/arch/arm/mm/fault.c
4  *
5  *  Copyright (C) 1995  Linus Torvalds
6  *  Modifications for ARM processor (c) 1995-2004 Russell King
7  */
8 #include <linux/extable.h>
9 #include <linux/signal.h>
10 #include <linux/mm.h>
11 #include <linux/hardirq.h>
12 #include <linux/init.h>
13 #include <linux/kprobes.h>
14 #include <linux/uaccess.h>
15 #include <linux/page-flags.h>
16 #include <linux/sched/signal.h>
17 #include <linux/sched/debug.h>
18 #include <linux/highmem.h>
19 #include <linux/perf_event.h>
20 #include <linux/kfence.h>
21 
22 #include <asm/system_misc.h>
23 #include <asm/system_info.h>
24 #include <asm/tlbflush.h>
25 
26 #include "fault.h"
27 
28 #ifdef CONFIG_MMU
29 
30 /*
31  * This is useful to dump out the page tables associated with
32  * 'addr' in mm 'mm'.
33  */
show_pte(const char * lvl,struct mm_struct * mm,unsigned long addr)34 void show_pte(const char *lvl, struct mm_struct *mm, unsigned long addr)
35 {
36 	pgd_t *pgd;
37 
38 	if (!mm)
39 		mm = &init_mm;
40 
41 	pgd = pgd_offset(mm, addr);
42 	printk("%s[%08lx] *pgd=%08llx", lvl, addr, (long long)pgd_val(*pgd));
43 
44 	do {
45 		p4d_t *p4d;
46 		pud_t *pud;
47 		pmd_t *pmd;
48 		pte_t *pte;
49 
50 		p4d = p4d_offset(pgd, addr);
51 		if (p4d_none(*p4d))
52 			break;
53 
54 		if (p4d_bad(*p4d)) {
55 			pr_cont("(bad)");
56 			break;
57 		}
58 
59 		pud = pud_offset(p4d, addr);
60 		if (PTRS_PER_PUD != 1)
61 			pr_cont(", *pud=%08llx", (long long)pud_val(*pud));
62 
63 		if (pud_none(*pud))
64 			break;
65 
66 		if (pud_bad(*pud)) {
67 			pr_cont("(bad)");
68 			break;
69 		}
70 
71 		pmd = pmd_offset(pud, addr);
72 		if (PTRS_PER_PMD != 1)
73 			pr_cont(", *pmd=%08llx", (long long)pmd_val(*pmd));
74 
75 		if (pmd_none(*pmd))
76 			break;
77 
78 		if (pmd_bad(*pmd)) {
79 			pr_cont("(bad)");
80 			break;
81 		}
82 
83 		/* We must not map this if we have highmem enabled */
84 		if (PageHighMem(pfn_to_page(pmd_val(*pmd) >> PAGE_SHIFT)))
85 			break;
86 
87 		pte = pte_offset_map(pmd, addr);
88 		pr_cont(", *pte=%08llx", (long long)pte_val(*pte));
89 #ifndef CONFIG_ARM_LPAE
90 		pr_cont(", *ppte=%08llx",
91 		       (long long)pte_val(pte[PTE_HWTABLE_PTRS]));
92 #endif
93 		pte_unmap(pte);
94 	} while(0);
95 
96 	pr_cont("\n");
97 }
98 #else					/* CONFIG_MMU */
show_pte(const char * lvl,struct mm_struct * mm,unsigned long addr)99 void show_pte(const char *lvl, struct mm_struct *mm, unsigned long addr)
100 { }
101 #endif					/* CONFIG_MMU */
102 
is_write_fault(unsigned int fsr)103 static inline bool is_write_fault(unsigned int fsr)
104 {
105 	return (fsr & FSR_WRITE) && !(fsr & FSR_CM);
106 }
107 
is_translation_fault(unsigned int fsr)108 static inline bool is_translation_fault(unsigned int fsr)
109 {
110 	int fs = fsr_fs(fsr);
111 #ifdef CONFIG_ARM_LPAE
112 	if ((fs & FS_MMU_NOLL_MASK) == FS_TRANS_NOLL)
113 		return true;
114 #else
115 	if (fs == FS_L1_TRANS || fs == FS_L2_TRANS)
116 		return true;
117 #endif
118 	return false;
119 }
120 
die_kernel_fault(const char * msg,struct mm_struct * mm,unsigned long addr,unsigned int fsr,struct pt_regs * regs)121 static void die_kernel_fault(const char *msg, struct mm_struct *mm,
122 			     unsigned long addr, unsigned int fsr,
123 			     struct pt_regs *regs)
124 {
125 	bust_spinlocks(1);
126 	pr_alert("8<--- cut here ---\n");
127 	pr_alert("Unable to handle kernel %s at virtual address %08lx\n",
128 		 msg, addr);
129 
130 	show_pte(KERN_ALERT, mm, addr);
131 	die("Oops", regs, fsr);
132 	bust_spinlocks(0);
133 	make_task_dead(SIGKILL);
134 }
135 
136 /*
137  * Oops.  The kernel tried to access some page that wasn't present.
138  */
139 static void
__do_kernel_fault(struct mm_struct * mm,unsigned long addr,unsigned int fsr,struct pt_regs * regs)140 __do_kernel_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
141 		  struct pt_regs *regs)
142 {
143 	const char *msg;
144 	/*
145 	 * Are we prepared to handle this kernel fault?
146 	 */
147 	if (fixup_exception(regs))
148 		return;
149 
150 	/*
151 	 * No handler, we'll have to terminate things with extreme prejudice.
152 	 */
153 	if (addr < PAGE_SIZE) {
154 		msg = "NULL pointer dereference";
155 	} else {
156 		if (is_translation_fault(fsr) &&
157 		    kfence_handle_page_fault(addr, is_write_fault(fsr), regs))
158 			return;
159 
160 		msg = "paging request";
161 	}
162 
163 	die_kernel_fault(msg, mm, addr, fsr, regs);
164 }
165 
166 /*
167  * Something tried to access memory that isn't in our memory map..
168  * User mode accesses just cause a SIGSEGV
169  */
170 static void
__do_user_fault(unsigned long addr,unsigned int fsr,unsigned int sig,int code,struct pt_regs * regs)171 __do_user_fault(unsigned long addr, unsigned int fsr, unsigned int sig,
172 		int code, struct pt_regs *regs)
173 {
174 	struct task_struct *tsk = current;
175 
176 	if (addr > TASK_SIZE)
177 		harden_branch_predictor();
178 
179 #ifdef CONFIG_DEBUG_USER
180 	if (((user_debug & UDBG_SEGV) && (sig == SIGSEGV)) ||
181 	    ((user_debug & UDBG_BUS)  && (sig == SIGBUS))) {
182 		pr_err("8<--- cut here ---\n");
183 		pr_err("%s: unhandled page fault (%d) at 0x%08lx, code 0x%03x\n",
184 		       tsk->comm, sig, addr, fsr);
185 		show_pte(KERN_ERR, tsk->mm, addr);
186 		show_regs(regs);
187 	}
188 #endif
189 #ifndef CONFIG_KUSER_HELPERS
190 	if ((sig == SIGSEGV) && ((addr & PAGE_MASK) == 0xffff0000))
191 		printk_ratelimited(KERN_DEBUG
192 				   "%s: CONFIG_KUSER_HELPERS disabled at 0x%08lx\n",
193 				   tsk->comm, addr);
194 #endif
195 
196 	tsk->thread.address = addr;
197 	tsk->thread.error_code = fsr;
198 	tsk->thread.trap_no = 14;
199 	force_sig_fault(sig, code, (void __user *)addr);
200 }
201 
do_bad_area(unsigned long addr,unsigned int fsr,struct pt_regs * regs)202 void do_bad_area(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
203 {
204 	struct task_struct *tsk = current;
205 	struct mm_struct *mm = tsk->active_mm;
206 
207 	/*
208 	 * If we are in kernel mode at this point, we
209 	 * have no context to handle this fault with.
210 	 */
211 	if (user_mode(regs))
212 		__do_user_fault(addr, fsr, SIGSEGV, SEGV_MAPERR, regs);
213 	else
214 		__do_kernel_fault(mm, addr, fsr, regs);
215 }
216 
217 #ifdef CONFIG_MMU
218 #define VM_FAULT_BADMAP		((__force vm_fault_t)0x010000)
219 #define VM_FAULT_BADACCESS	((__force vm_fault_t)0x020000)
220 
is_permission_fault(unsigned int fsr)221 static inline bool is_permission_fault(unsigned int fsr)
222 {
223 	int fs = fsr_fs(fsr);
224 #ifdef CONFIG_ARM_LPAE
225 	if ((fs & FS_MMU_NOLL_MASK) == FS_PERM_NOLL)
226 		return true;
227 #else
228 	if (fs == FS_L1_PERM || fs == FS_L2_PERM)
229 		return true;
230 #endif
231 	return false;
232 }
233 
234 static vm_fault_t __kprobes
__do_page_fault(struct mm_struct * mm,unsigned long addr,unsigned int flags,unsigned long vma_flags,struct pt_regs * regs)235 __do_page_fault(struct mm_struct *mm, unsigned long addr, unsigned int flags,
236 		unsigned long vma_flags, struct pt_regs *regs)
237 {
238 	struct vm_area_struct *vma = find_vma(mm, addr);
239 	if (unlikely(!vma))
240 		return VM_FAULT_BADMAP;
241 
242 	if (unlikely(vma->vm_start > addr)) {
243 		if (!(vma->vm_flags & VM_GROWSDOWN))
244 			return VM_FAULT_BADMAP;
245 		if (addr < FIRST_USER_ADDRESS)
246 			return VM_FAULT_BADMAP;
247 		if (expand_stack(vma, addr))
248 			return VM_FAULT_BADMAP;
249 	}
250 
251 	/*
252 	 * ok, we have a good vm_area for this memory access, check the
253 	 * permissions on the VMA allow for the fault which occurred.
254 	 */
255 	if (!(vma->vm_flags & vma_flags))
256 		return VM_FAULT_BADACCESS;
257 
258 	return handle_mm_fault(vma, addr & PAGE_MASK, flags, regs);
259 }
260 
261 static int __kprobes
do_page_fault(unsigned long addr,unsigned int fsr,struct pt_regs * regs)262 do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
263 {
264 	struct mm_struct *mm = current->mm;
265 	int sig, code;
266 	vm_fault_t fault;
267 	unsigned int flags = FAULT_FLAG_DEFAULT;
268 	unsigned long vm_flags = VM_ACCESS_FLAGS;
269 
270 	if (kprobe_page_fault(regs, fsr))
271 		return 0;
272 
273 
274 	/* Enable interrupts if they were enabled in the parent context. */
275 	if (interrupts_enabled(regs))
276 		local_irq_enable();
277 
278 	/*
279 	 * If we're in an interrupt or have no user
280 	 * context, we must not take the fault..
281 	 */
282 	if (faulthandler_disabled() || !mm)
283 		goto no_context;
284 
285 	if (user_mode(regs))
286 		flags |= FAULT_FLAG_USER;
287 
288 	if (is_write_fault(fsr)) {
289 		flags |= FAULT_FLAG_WRITE;
290 		vm_flags = VM_WRITE;
291 	}
292 
293 	if (fsr & FSR_LNX_PF) {
294 		vm_flags = VM_EXEC;
295 
296 		if (is_permission_fault(fsr) && !user_mode(regs))
297 			die_kernel_fault("execution of memory",
298 					 mm, addr, fsr, regs);
299 	}
300 
301 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
302 
303 	/*
304 	 * As per x86, we may deadlock here.  However, since the kernel only
305 	 * validly references user space from well defined areas of the code,
306 	 * we can bug out early if this is from code which shouldn't.
307 	 */
308 	if (!mmap_read_trylock(mm)) {
309 		if (!user_mode(regs) && !search_exception_tables(regs->ARM_pc))
310 			goto no_context;
311 retry:
312 		mmap_read_lock(mm);
313 	} else {
314 		/*
315 		 * The above down_read_trylock() might have succeeded in
316 		 * which case, we'll have missed the might_sleep() from
317 		 * down_read()
318 		 */
319 		might_sleep();
320 #ifdef CONFIG_DEBUG_VM
321 		if (!user_mode(regs) &&
322 		    !search_exception_tables(regs->ARM_pc))
323 			goto no_context;
324 #endif
325 	}
326 
327 	fault = __do_page_fault(mm, addr, flags, vm_flags, regs);
328 
329 	/* If we need to retry but a fatal signal is pending, handle the
330 	 * signal first. We do not need to release the mmap_lock because
331 	 * it would already be released in __lock_page_or_retry in
332 	 * mm/filemap.c. */
333 	if (fault_signal_pending(fault, regs)) {
334 		if (!user_mode(regs))
335 			goto no_context;
336 		return 0;
337 	}
338 
339 	/* The fault is fully completed (including releasing mmap lock) */
340 	if (fault & VM_FAULT_COMPLETED)
341 		return 0;
342 
343 	if (!(fault & VM_FAULT_ERROR)) {
344 		if (fault & VM_FAULT_RETRY) {
345 			flags |= FAULT_FLAG_TRIED;
346 			goto retry;
347 		}
348 	}
349 
350 	mmap_read_unlock(mm);
351 
352 	/*
353 	 * Handle the "normal" case first - VM_FAULT_MAJOR
354 	 */
355 	if (likely(!(fault & (VM_FAULT_ERROR | VM_FAULT_BADMAP | VM_FAULT_BADACCESS))))
356 		return 0;
357 
358 	/*
359 	 * If we are in kernel mode at this point, we
360 	 * have no context to handle this fault with.
361 	 */
362 	if (!user_mode(regs))
363 		goto no_context;
364 
365 	if (fault & VM_FAULT_OOM) {
366 		/*
367 		 * We ran out of memory, call the OOM killer, and return to
368 		 * userspace (which will retry the fault, or kill us if we
369 		 * got oom-killed)
370 		 */
371 		pagefault_out_of_memory();
372 		return 0;
373 	}
374 
375 	if (fault & VM_FAULT_SIGBUS) {
376 		/*
377 		 * We had some memory, but were unable to
378 		 * successfully fix up this page fault.
379 		 */
380 		sig = SIGBUS;
381 		code = BUS_ADRERR;
382 	} else {
383 		/*
384 		 * Something tried to access memory that
385 		 * isn't in our memory map..
386 		 */
387 		sig = SIGSEGV;
388 		code = fault == VM_FAULT_BADACCESS ?
389 			SEGV_ACCERR : SEGV_MAPERR;
390 	}
391 
392 	__do_user_fault(addr, fsr, sig, code, regs);
393 	return 0;
394 
395 no_context:
396 	__do_kernel_fault(mm, addr, fsr, regs);
397 	return 0;
398 }
399 #else					/* CONFIG_MMU */
400 static int
do_page_fault(unsigned long addr,unsigned int fsr,struct pt_regs * regs)401 do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
402 {
403 	return 0;
404 }
405 #endif					/* CONFIG_MMU */
406 
407 /*
408  * First Level Translation Fault Handler
409  *
410  * We enter here because the first level page table doesn't contain
411  * a valid entry for the address.
412  *
413  * If the address is in kernel space (>= TASK_SIZE), then we are
414  * probably faulting in the vmalloc() area.
415  *
416  * If the init_task's first level page tables contains the relevant
417  * entry, we copy the it to this task.  If not, we send the process
418  * a signal, fixup the exception, or oops the kernel.
419  *
420  * NOTE! We MUST NOT take any locks for this case. We may be in an
421  * interrupt or a critical region, and should only copy the information
422  * from the master page table, nothing more.
423  */
424 #ifdef CONFIG_MMU
425 static int __kprobes
do_translation_fault(unsigned long addr,unsigned int fsr,struct pt_regs * regs)426 do_translation_fault(unsigned long addr, unsigned int fsr,
427 		     struct pt_regs *regs)
428 {
429 	unsigned int index;
430 	pgd_t *pgd, *pgd_k;
431 	p4d_t *p4d, *p4d_k;
432 	pud_t *pud, *pud_k;
433 	pmd_t *pmd, *pmd_k;
434 
435 	if (addr < TASK_SIZE)
436 		return do_page_fault(addr, fsr, regs);
437 
438 	if (user_mode(regs))
439 		goto bad_area;
440 
441 	index = pgd_index(addr);
442 
443 	pgd = cpu_get_pgd() + index;
444 	pgd_k = init_mm.pgd + index;
445 
446 	p4d = p4d_offset(pgd, addr);
447 	p4d_k = p4d_offset(pgd_k, addr);
448 
449 	if (p4d_none(*p4d_k))
450 		goto bad_area;
451 	if (!p4d_present(*p4d))
452 		set_p4d(p4d, *p4d_k);
453 
454 	pud = pud_offset(p4d, addr);
455 	pud_k = pud_offset(p4d_k, addr);
456 
457 	if (pud_none(*pud_k))
458 		goto bad_area;
459 	if (!pud_present(*pud))
460 		set_pud(pud, *pud_k);
461 
462 	pmd = pmd_offset(pud, addr);
463 	pmd_k = pmd_offset(pud_k, addr);
464 
465 #ifdef CONFIG_ARM_LPAE
466 	/*
467 	 * Only one hardware entry per PMD with LPAE.
468 	 */
469 	index = 0;
470 #else
471 	/*
472 	 * On ARM one Linux PGD entry contains two hardware entries (see page
473 	 * tables layout in pgtable.h). We normally guarantee that we always
474 	 * fill both L1 entries. But create_mapping() doesn't follow the rule.
475 	 * It can create inidividual L1 entries, so here we have to call
476 	 * pmd_none() check for the entry really corresponded to address, not
477 	 * for the first of pair.
478 	 */
479 	index = (addr >> SECTION_SHIFT) & 1;
480 #endif
481 	if (pmd_none(pmd_k[index]))
482 		goto bad_area;
483 
484 	copy_pmd(pmd, pmd_k);
485 	return 0;
486 
487 bad_area:
488 	do_bad_area(addr, fsr, regs);
489 	return 0;
490 }
491 #else					/* CONFIG_MMU */
492 static int
do_translation_fault(unsigned long addr,unsigned int fsr,struct pt_regs * regs)493 do_translation_fault(unsigned long addr, unsigned int fsr,
494 		     struct pt_regs *regs)
495 {
496 	return 0;
497 }
498 #endif					/* CONFIG_MMU */
499 
500 /*
501  * Some section permission faults need to be handled gracefully.
502  * They can happen due to a __{get,put}_user during an oops.
503  */
504 #ifndef CONFIG_ARM_LPAE
505 static int
do_sect_fault(unsigned long addr,unsigned int fsr,struct pt_regs * regs)506 do_sect_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
507 {
508 	do_bad_area(addr, fsr, regs);
509 	return 0;
510 }
511 #endif /* CONFIG_ARM_LPAE */
512 
513 /*
514  * This abort handler always returns "fault".
515  */
516 static int
do_bad(unsigned long addr,unsigned int fsr,struct pt_regs * regs)517 do_bad(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
518 {
519 	return 1;
520 }
521 
522 struct fsr_info {
523 	int	(*fn)(unsigned long addr, unsigned int fsr, struct pt_regs *regs);
524 	int	sig;
525 	int	code;
526 	const char *name;
527 };
528 
529 /* FSR definition */
530 #ifdef CONFIG_ARM_LPAE
531 #include "fsr-3level.c"
532 #else
533 #include "fsr-2level.c"
534 #endif
535 
536 void __init
hook_fault_code(int nr,int (* fn)(unsigned long,unsigned int,struct pt_regs *),int sig,int code,const char * name)537 hook_fault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
538 		int sig, int code, const char *name)
539 {
540 	if (nr < 0 || nr >= ARRAY_SIZE(fsr_info))
541 		BUG();
542 
543 	fsr_info[nr].fn   = fn;
544 	fsr_info[nr].sig  = sig;
545 	fsr_info[nr].code = code;
546 	fsr_info[nr].name = name;
547 }
548 
549 /*
550  * Dispatch a data abort to the relevant handler.
551  */
552 asmlinkage void
do_DataAbort(unsigned long addr,unsigned int fsr,struct pt_regs * regs)553 do_DataAbort(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
554 {
555 	const struct fsr_info *inf = fsr_info + fsr_fs(fsr);
556 
557 	if (!inf->fn(addr, fsr & ~FSR_LNX_PF, regs))
558 		return;
559 
560 	pr_alert("8<--- cut here ---\n");
561 	pr_alert("Unhandled fault: %s (0x%03x) at 0x%08lx\n",
562 		inf->name, fsr, addr);
563 	show_pte(KERN_ALERT, current->mm, addr);
564 
565 	arm_notify_die("", regs, inf->sig, inf->code, (void __user *)addr,
566 		       fsr, 0);
567 }
568 
569 void __init
hook_ifault_code(int nr,int (* fn)(unsigned long,unsigned int,struct pt_regs *),int sig,int code,const char * name)570 hook_ifault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
571 		 int sig, int code, const char *name)
572 {
573 	if (nr < 0 || nr >= ARRAY_SIZE(ifsr_info))
574 		BUG();
575 
576 	ifsr_info[nr].fn   = fn;
577 	ifsr_info[nr].sig  = sig;
578 	ifsr_info[nr].code = code;
579 	ifsr_info[nr].name = name;
580 }
581 
582 asmlinkage void
do_PrefetchAbort(unsigned long addr,unsigned int ifsr,struct pt_regs * regs)583 do_PrefetchAbort(unsigned long addr, unsigned int ifsr, struct pt_regs *regs)
584 {
585 	const struct fsr_info *inf = ifsr_info + fsr_fs(ifsr);
586 
587 	if (!inf->fn(addr, ifsr | FSR_LNX_PF, regs))
588 		return;
589 
590 	pr_alert("Unhandled prefetch abort: %s (0x%03x) at 0x%08lx\n",
591 		inf->name, ifsr, addr);
592 
593 	arm_notify_die("", regs, inf->sig, inf->code, (void __user *)addr,
594 		       ifsr, 0);
595 }
596 
597 /*
598  * Abort handler to be used only during first unmasking of asynchronous aborts
599  * on the boot CPU. This makes sure that the machine will not die if the
600  * firmware/bootloader left an imprecise abort pending for us to trip over.
601  */
early_abort_handler(unsigned long addr,unsigned int fsr,struct pt_regs * regs)602 static int __init early_abort_handler(unsigned long addr, unsigned int fsr,
603 				      struct pt_regs *regs)
604 {
605 	pr_warn("Hit pending asynchronous external abort (FSR=0x%08x) during "
606 		"first unmask, this is most likely caused by a "
607 		"firmware/bootloader bug.\n", fsr);
608 
609 	return 0;
610 }
611 
early_abt_enable(void)612 void __init early_abt_enable(void)
613 {
614 	fsr_info[FSR_FS_AEA].fn = early_abort_handler;
615 	local_abt_enable();
616 	fsr_info[FSR_FS_AEA].fn = do_bad;
617 }
618 
619 #ifndef CONFIG_ARM_LPAE
exceptions_init(void)620 static int __init exceptions_init(void)
621 {
622 	if (cpu_architecture() >= CPU_ARCH_ARMv6) {
623 		hook_fault_code(4, do_translation_fault, SIGSEGV, SEGV_MAPERR,
624 				"I-cache maintenance fault");
625 	}
626 
627 	if (cpu_architecture() >= CPU_ARCH_ARMv7) {
628 		/*
629 		 * TODO: Access flag faults introduced in ARMv6K.
630 		 * Runtime check for 'K' extension is needed
631 		 */
632 		hook_fault_code(3, do_bad, SIGSEGV, SEGV_MAPERR,
633 				"section access flag fault");
634 		hook_fault_code(6, do_bad, SIGSEGV, SEGV_MAPERR,
635 				"section access flag fault");
636 	}
637 
638 	return 0;
639 }
640 
641 arch_initcall(exceptions_init);
642 #endif
643