xref: /linux/arch/arm/mm/fault.c (revision 56ef27e3abe6d6453b1f4f6127041f3a65d7cbc9)
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  */
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 		if (!pte)
89 			break;
90 
91 		pr_cont(", *pte=%08llx", (long long)pte_val(*pte));
92 #ifndef CONFIG_ARM_LPAE
93 		pr_cont(", *ppte=%08llx",
94 		       (long long)pte_val(pte[PTE_HWTABLE_PTRS]));
95 #endif
96 		pte_unmap(pte);
97 	} while(0);
98 
99 	pr_cont("\n");
100 }
101 #else					/* CONFIG_MMU */
102 void show_pte(const char *lvl, struct mm_struct *mm, unsigned long addr)
103 { }
104 #endif					/* CONFIG_MMU */
105 
106 static inline bool is_write_fault(unsigned int fsr)
107 {
108 	return (fsr & FSR_WRITE) && !(fsr & FSR_CM);
109 }
110 
111 static inline bool is_translation_fault(unsigned int fsr)
112 {
113 	int fs = fsr_fs(fsr);
114 #ifdef CONFIG_ARM_LPAE
115 	if ((fs & FS_MMU_NOLL_MASK) == FS_TRANS_NOLL)
116 		return true;
117 #else
118 	if (fs == FS_L1_TRANS || fs == FS_L2_TRANS)
119 		return true;
120 #endif
121 	return false;
122 }
123 
124 static void die_kernel_fault(const char *msg, struct mm_struct *mm,
125 			     unsigned long addr, unsigned int fsr,
126 			     struct pt_regs *regs)
127 {
128 	bust_spinlocks(1);
129 	pr_alert("8<--- cut here ---\n");
130 	pr_alert("Unable to handle kernel %s at virtual address %08lx when %s\n",
131 		 msg, addr, fsr & FSR_LNX_PF ? "execute" :
132 		 fsr & FSR_WRITE ? "write" : "read");
133 
134 	show_pte(KERN_ALERT, mm, addr);
135 	die("Oops", regs, fsr);
136 	bust_spinlocks(0);
137 	make_task_dead(SIGKILL);
138 }
139 
140 /*
141  * Oops.  The kernel tried to access some page that wasn't present.
142  */
143 static void
144 __do_kernel_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
145 		  struct pt_regs *regs)
146 {
147 	const char *msg;
148 	/*
149 	 * Are we prepared to handle this kernel fault?
150 	 */
151 	if (fixup_exception(regs))
152 		return;
153 
154 	/*
155 	 * No handler, we'll have to terminate things with extreme prejudice.
156 	 */
157 	if (addr < PAGE_SIZE) {
158 		msg = "NULL pointer dereference";
159 	} else {
160 		if (is_translation_fault(fsr) &&
161 		    kfence_handle_page_fault(addr, is_write_fault(fsr), regs))
162 			return;
163 
164 		msg = "paging request";
165 	}
166 
167 	die_kernel_fault(msg, mm, addr, fsr, regs);
168 }
169 
170 /*
171  * Something tried to access memory that isn't in our memory map..
172  * User mode accesses just cause a SIGSEGV
173  */
174 static void
175 __do_user_fault(unsigned long addr, unsigned int fsr, unsigned int sig,
176 		int code, struct pt_regs *regs)
177 {
178 	struct task_struct *tsk = current;
179 
180 	if (addr > TASK_SIZE)
181 		harden_branch_predictor();
182 
183 #ifdef CONFIG_DEBUG_USER
184 	if (((user_debug & UDBG_SEGV) && (sig == SIGSEGV)) ||
185 	    ((user_debug & UDBG_BUS)  && (sig == SIGBUS))) {
186 		pr_err("8<--- cut here ---\n");
187 		pr_err("%s: unhandled page fault (%d) at 0x%08lx, code 0x%03x\n",
188 		       tsk->comm, sig, addr, fsr);
189 		show_pte(KERN_ERR, tsk->mm, addr);
190 		show_regs(regs);
191 	}
192 #endif
193 #ifndef CONFIG_KUSER_HELPERS
194 	if ((sig == SIGSEGV) && ((addr & PAGE_MASK) == 0xffff0000))
195 		printk_ratelimited(KERN_DEBUG
196 				   "%s: CONFIG_KUSER_HELPERS disabled at 0x%08lx\n",
197 				   tsk->comm, addr);
198 #endif
199 
200 	tsk->thread.address = addr;
201 	tsk->thread.error_code = fsr;
202 	tsk->thread.trap_no = 14;
203 	force_sig_fault(sig, code, (void __user *)addr);
204 }
205 
206 void do_bad_area(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
207 {
208 	struct task_struct *tsk = current;
209 	struct mm_struct *mm = tsk->active_mm;
210 
211 	/*
212 	 * If we are in kernel mode at this point, we
213 	 * have no context to handle this fault with.
214 	 */
215 	if (user_mode(regs))
216 		__do_user_fault(addr, fsr, SIGSEGV, SEGV_MAPERR, regs);
217 	else
218 		__do_kernel_fault(mm, addr, fsr, regs);
219 }
220 
221 #ifdef CONFIG_MMU
222 #define VM_FAULT_BADMAP		((__force vm_fault_t)0x010000)
223 #define VM_FAULT_BADACCESS	((__force vm_fault_t)0x020000)
224 
225 static inline bool is_permission_fault(unsigned int fsr)
226 {
227 	int fs = fsr_fs(fsr);
228 #ifdef CONFIG_ARM_LPAE
229 	if ((fs & FS_MMU_NOLL_MASK) == FS_PERM_NOLL)
230 		return true;
231 #else
232 	if (fs == FS_L1_PERM || fs == FS_L2_PERM)
233 		return true;
234 #endif
235 	return false;
236 }
237 
238 static int __kprobes
239 do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
240 {
241 	struct mm_struct *mm = current->mm;
242 	struct vm_area_struct *vma;
243 	int sig, code;
244 	vm_fault_t fault;
245 	unsigned int flags = FAULT_FLAG_DEFAULT;
246 	unsigned long vm_flags = VM_ACCESS_FLAGS;
247 
248 	if (kprobe_page_fault(regs, fsr))
249 		return 0;
250 
251 
252 	/* Enable interrupts if they were enabled in the parent context. */
253 	if (interrupts_enabled(regs))
254 		local_irq_enable();
255 
256 	/*
257 	 * If we're in an interrupt or have no user
258 	 * context, we must not take the fault..
259 	 */
260 	if (faulthandler_disabled() || !mm)
261 		goto no_context;
262 
263 	if (user_mode(regs))
264 		flags |= FAULT_FLAG_USER;
265 
266 	if (is_write_fault(fsr)) {
267 		flags |= FAULT_FLAG_WRITE;
268 		vm_flags = VM_WRITE;
269 	}
270 
271 	if (fsr & FSR_LNX_PF) {
272 		vm_flags = VM_EXEC;
273 
274 		if (is_permission_fault(fsr) && !user_mode(regs))
275 			die_kernel_fault("execution of memory",
276 					 mm, addr, fsr, regs);
277 	}
278 
279 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
280 
281 	if (!(flags & FAULT_FLAG_USER))
282 		goto lock_mmap;
283 
284 	vma = lock_vma_under_rcu(mm, addr);
285 	if (!vma)
286 		goto lock_mmap;
287 
288 	if (!(vma->vm_flags & vm_flags)) {
289 		vma_end_read(vma);
290 		goto lock_mmap;
291 	}
292 	fault = handle_mm_fault(vma, addr, flags | FAULT_FLAG_VMA_LOCK, regs);
293 	if (!(fault & (VM_FAULT_RETRY | VM_FAULT_COMPLETED)))
294 		vma_end_read(vma);
295 
296 	if (!(fault & VM_FAULT_RETRY)) {
297 		count_vm_vma_lock_event(VMA_LOCK_SUCCESS);
298 		goto done;
299 	}
300 	count_vm_vma_lock_event(VMA_LOCK_RETRY);
301 	if (fault & VM_FAULT_MAJOR)
302 		flags |= FAULT_FLAG_TRIED;
303 
304 	/* Quick path to respond to signals */
305 	if (fault_signal_pending(fault, regs)) {
306 		if (!user_mode(regs))
307 			goto no_context;
308 		return 0;
309 	}
310 lock_mmap:
311 
312 retry:
313 	vma = lock_mm_and_find_vma(mm, addr, regs);
314 	if (unlikely(!vma)) {
315 		fault = VM_FAULT_BADMAP;
316 		goto bad_area;
317 	}
318 
319 	/*
320 	 * ok, we have a good vm_area for this memory access, check the
321 	 * permissions on the VMA allow for the fault which occurred.
322 	 */
323 	if (!(vma->vm_flags & vm_flags))
324 		fault = VM_FAULT_BADACCESS;
325 	else
326 		fault = handle_mm_fault(vma, addr & PAGE_MASK, flags, regs);
327 
328 	/* If we need to retry but a fatal signal is pending, handle the
329 	 * signal first. We do not need to release the mmap_lock because
330 	 * it would already be released in __lock_page_or_retry in
331 	 * mm/filemap.c. */
332 	if (fault_signal_pending(fault, regs)) {
333 		if (!user_mode(regs))
334 			goto no_context;
335 		return 0;
336 	}
337 
338 	/* The fault is fully completed (including releasing mmap lock) */
339 	if (fault & VM_FAULT_COMPLETED)
340 		return 0;
341 
342 	if (!(fault & VM_FAULT_ERROR)) {
343 		if (fault & VM_FAULT_RETRY) {
344 			flags |= FAULT_FLAG_TRIED;
345 			goto retry;
346 		}
347 	}
348 
349 	mmap_read_unlock(mm);
350 done:
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 bad_area:
359 	/*
360 	 * If we are in kernel mode at this point, we
361 	 * have no context to handle this fault with.
362 	 */
363 	if (!user_mode(regs))
364 		goto no_context;
365 
366 	if (fault & VM_FAULT_OOM) {
367 		/*
368 		 * We ran out of memory, call the OOM killer, and return to
369 		 * userspace (which will retry the fault, or kill us if we
370 		 * got oom-killed)
371 		 */
372 		pagefault_out_of_memory();
373 		return 0;
374 	}
375 
376 	if (fault & VM_FAULT_SIGBUS) {
377 		/*
378 		 * We had some memory, but were unable to
379 		 * successfully fix up this page fault.
380 		 */
381 		sig = SIGBUS;
382 		code = BUS_ADRERR;
383 	} else {
384 		/*
385 		 * Something tried to access memory that
386 		 * isn't in our memory map..
387 		 */
388 		sig = SIGSEGV;
389 		code = fault == VM_FAULT_BADACCESS ?
390 			SEGV_ACCERR : SEGV_MAPERR;
391 	}
392 
393 	__do_user_fault(addr, fsr, sig, code, regs);
394 	return 0;
395 
396 no_context:
397 	__do_kernel_fault(mm, addr, fsr, regs);
398 	return 0;
399 }
400 #else					/* CONFIG_MMU */
401 static int
402 do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
403 {
404 	return 0;
405 }
406 #endif					/* CONFIG_MMU */
407 
408 /*
409  * First Level Translation Fault Handler
410  *
411  * We enter here because the first level page table doesn't contain
412  * a valid entry for the address.
413  *
414  * If the address is in kernel space (>= TASK_SIZE), then we are
415  * probably faulting in the vmalloc() area.
416  *
417  * If the init_task's first level page tables contains the relevant
418  * entry, we copy the it to this task.  If not, we send the process
419  * a signal, fixup the exception, or oops the kernel.
420  *
421  * NOTE! We MUST NOT take any locks for this case. We may be in an
422  * interrupt or a critical region, and should only copy the information
423  * from the master page table, nothing more.
424  */
425 #ifdef CONFIG_MMU
426 static int __kprobes
427 do_translation_fault(unsigned long addr, unsigned int fsr,
428 		     struct pt_regs *regs)
429 {
430 	unsigned int index;
431 	pgd_t *pgd, *pgd_k;
432 	p4d_t *p4d, *p4d_k;
433 	pud_t *pud, *pud_k;
434 	pmd_t *pmd, *pmd_k;
435 
436 	if (addr < TASK_SIZE)
437 		return do_page_fault(addr, fsr, regs);
438 
439 	if (user_mode(regs))
440 		goto bad_area;
441 
442 	index = pgd_index(addr);
443 
444 	pgd = cpu_get_pgd() + index;
445 	pgd_k = init_mm.pgd + index;
446 
447 	p4d = p4d_offset(pgd, addr);
448 	p4d_k = p4d_offset(pgd_k, addr);
449 
450 	if (p4d_none(*p4d_k))
451 		goto bad_area;
452 	if (!p4d_present(*p4d))
453 		set_p4d(p4d, *p4d_k);
454 
455 	pud = pud_offset(p4d, addr);
456 	pud_k = pud_offset(p4d_k, addr);
457 
458 	if (pud_none(*pud_k))
459 		goto bad_area;
460 	if (!pud_present(*pud))
461 		set_pud(pud, *pud_k);
462 
463 	pmd = pmd_offset(pud, addr);
464 	pmd_k = pmd_offset(pud_k, addr);
465 
466 #ifdef CONFIG_ARM_LPAE
467 	/*
468 	 * Only one hardware entry per PMD with LPAE.
469 	 */
470 	index = 0;
471 #else
472 	/*
473 	 * On ARM one Linux PGD entry contains two hardware entries (see page
474 	 * tables layout in pgtable.h). We normally guarantee that we always
475 	 * fill both L1 entries. But create_mapping() doesn't follow the rule.
476 	 * It can create inidividual L1 entries, so here we have to call
477 	 * pmd_none() check for the entry really corresponded to address, not
478 	 * for the first of pair.
479 	 */
480 	index = (addr >> SECTION_SHIFT) & 1;
481 #endif
482 	if (pmd_none(pmd_k[index]))
483 		goto bad_area;
484 
485 	copy_pmd(pmd, pmd_k);
486 	return 0;
487 
488 bad_area:
489 	do_bad_area(addr, fsr, regs);
490 	return 0;
491 }
492 #else					/* CONFIG_MMU */
493 static int
494 do_translation_fault(unsigned long addr, unsigned int fsr,
495 		     struct pt_regs *regs)
496 {
497 	return 0;
498 }
499 #endif					/* CONFIG_MMU */
500 
501 /*
502  * Some section permission faults need to be handled gracefully.
503  * They can happen due to a __{get,put}_user during an oops.
504  */
505 #ifndef CONFIG_ARM_LPAE
506 static int
507 do_sect_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
508 {
509 	do_bad_area(addr, fsr, regs);
510 	return 0;
511 }
512 #endif /* CONFIG_ARM_LPAE */
513 
514 /*
515  * This abort handler always returns "fault".
516  */
517 static int
518 do_bad(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
519 {
520 	return 1;
521 }
522 
523 struct fsr_info {
524 	int	(*fn)(unsigned long addr, unsigned int fsr, struct pt_regs *regs);
525 	int	sig;
526 	int	code;
527 	const char *name;
528 };
529 
530 /* FSR definition */
531 #ifdef CONFIG_ARM_LPAE
532 #include "fsr-3level.c"
533 #else
534 #include "fsr-2level.c"
535 #endif
536 
537 void __init
538 hook_fault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
539 		int sig, int code, const char *name)
540 {
541 	if (nr < 0 || nr >= ARRAY_SIZE(fsr_info))
542 		BUG();
543 
544 	fsr_info[nr].fn   = fn;
545 	fsr_info[nr].sig  = sig;
546 	fsr_info[nr].code = code;
547 	fsr_info[nr].name = name;
548 }
549 
550 /*
551  * Dispatch a data abort to the relevant handler.
552  */
553 asmlinkage void
554 do_DataAbort(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
555 {
556 	const struct fsr_info *inf = fsr_info + fsr_fs(fsr);
557 
558 	if (!inf->fn(addr, fsr & ~FSR_LNX_PF, regs))
559 		return;
560 
561 	pr_alert("8<--- cut here ---\n");
562 	pr_alert("Unhandled fault: %s (0x%03x) at 0x%08lx\n",
563 		inf->name, fsr, addr);
564 	show_pte(KERN_ALERT, current->mm, addr);
565 
566 	arm_notify_die("", regs, inf->sig, inf->code, (void __user *)addr,
567 		       fsr, 0);
568 }
569 
570 void __init
571 hook_ifault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
572 		 int sig, int code, const char *name)
573 {
574 	if (nr < 0 || nr >= ARRAY_SIZE(ifsr_info))
575 		BUG();
576 
577 	ifsr_info[nr].fn   = fn;
578 	ifsr_info[nr].sig  = sig;
579 	ifsr_info[nr].code = code;
580 	ifsr_info[nr].name = name;
581 }
582 
583 asmlinkage void
584 do_PrefetchAbort(unsigned long addr, unsigned int ifsr, struct pt_regs *regs)
585 {
586 	const struct fsr_info *inf = ifsr_info + fsr_fs(ifsr);
587 
588 	if (!inf->fn(addr, ifsr | FSR_LNX_PF, regs))
589 		return;
590 
591 	pr_alert("Unhandled prefetch abort: %s (0x%03x) at 0x%08lx\n",
592 		inf->name, ifsr, addr);
593 
594 	arm_notify_die("", regs, inf->sig, inf->code, (void __user *)addr,
595 		       ifsr, 0);
596 }
597 
598 /*
599  * Abort handler to be used only during first unmasking of asynchronous aborts
600  * on the boot CPU. This makes sure that the machine will not die if the
601  * firmware/bootloader left an imprecise abort pending for us to trip over.
602  */
603 static int __init early_abort_handler(unsigned long addr, unsigned int fsr,
604 				      struct pt_regs *regs)
605 {
606 	pr_warn("Hit pending asynchronous external abort (FSR=0x%08x) during "
607 		"first unmask, this is most likely caused by a "
608 		"firmware/bootloader bug.\n", fsr);
609 
610 	return 0;
611 }
612 
613 void __init early_abt_enable(void)
614 {
615 	fsr_info[FSR_FS_AEA].fn = early_abort_handler;
616 	local_abt_enable();
617 	fsr_info[FSR_FS_AEA].fn = do_bad;
618 }
619 
620 #ifndef CONFIG_ARM_LPAE
621 static int __init exceptions_init(void)
622 {
623 	if (cpu_architecture() >= CPU_ARCH_ARMv6) {
624 		hook_fault_code(4, do_translation_fault, SIGSEGV, SEGV_MAPERR,
625 				"I-cache maintenance fault");
626 	}
627 
628 	if (cpu_architecture() >= CPU_ARCH_ARMv7) {
629 		/*
630 		 * TODO: Access flag faults introduced in ARMv6K.
631 		 * Runtime check for 'K' extension is needed
632 		 */
633 		hook_fault_code(3, do_bad, SIGSEGV, SEGV_MAPERR,
634 				"section access flag fault");
635 		hook_fault_code(6, do_bad, SIGSEGV, SEGV_MAPERR,
636 				"section access flag fault");
637 	}
638 
639 	return 0;
640 }
641 
642 arch_initcall(exceptions_init);
643 #endif
644