xref: /openbmc/linux/arch/arm/mm/fault.c (revision a06c488d)
1 /*
2  *  linux/arch/arm/mm/fault.c
3  *
4  *  Copyright (C) 1995  Linus Torvalds
5  *  Modifications for ARM processor (c) 1995-2004 Russell King
6  *
7  * This program is free software; you can redistribute it and/or modify
8  * it under the terms of the GNU General Public License version 2 as
9  * published by the Free Software Foundation.
10  */
11 #include <linux/module.h>
12 #include <linux/signal.h>
13 #include <linux/mm.h>
14 #include <linux/hardirq.h>
15 #include <linux/init.h>
16 #include <linux/kprobes.h>
17 #include <linux/uaccess.h>
18 #include <linux/page-flags.h>
19 #include <linux/sched.h>
20 #include <linux/highmem.h>
21 #include <linux/perf_event.h>
22 
23 #include <asm/exception.h>
24 #include <asm/pgtable.h>
25 #include <asm/system_misc.h>
26 #include <asm/system_info.h>
27 #include <asm/tlbflush.h>
28 
29 #include "fault.h"
30 
31 #ifdef CONFIG_MMU
32 
33 #ifdef CONFIG_KPROBES
34 static inline int notify_page_fault(struct pt_regs *regs, unsigned int fsr)
35 {
36 	int ret = 0;
37 
38 	if (!user_mode(regs)) {
39 		/* kprobe_running() needs smp_processor_id() */
40 		preempt_disable();
41 		if (kprobe_running() && kprobe_fault_handler(regs, fsr))
42 			ret = 1;
43 		preempt_enable();
44 	}
45 
46 	return ret;
47 }
48 #else
49 static inline int notify_page_fault(struct pt_regs *regs, unsigned int fsr)
50 {
51 	return 0;
52 }
53 #endif
54 
55 /*
56  * This is useful to dump out the page tables associated with
57  * 'addr' in mm 'mm'.
58  */
59 void show_pte(struct mm_struct *mm, unsigned long addr)
60 {
61 	pgd_t *pgd;
62 
63 	if (!mm)
64 		mm = &init_mm;
65 
66 	pr_alert("pgd = %p\n", mm->pgd);
67 	pgd = pgd_offset(mm, addr);
68 	pr_alert("[%08lx] *pgd=%08llx",
69 			addr, (long long)pgd_val(*pgd));
70 
71 	do {
72 		pud_t *pud;
73 		pmd_t *pmd;
74 		pte_t *pte;
75 
76 		if (pgd_none(*pgd))
77 			break;
78 
79 		if (pgd_bad(*pgd)) {
80 			pr_cont("(bad)");
81 			break;
82 		}
83 
84 		pud = pud_offset(pgd, addr);
85 		if (PTRS_PER_PUD != 1)
86 			pr_cont(", *pud=%08llx", (long long)pud_val(*pud));
87 
88 		if (pud_none(*pud))
89 			break;
90 
91 		if (pud_bad(*pud)) {
92 			pr_cont("(bad)");
93 			break;
94 		}
95 
96 		pmd = pmd_offset(pud, addr);
97 		if (PTRS_PER_PMD != 1)
98 			pr_cont(", *pmd=%08llx", (long long)pmd_val(*pmd));
99 
100 		if (pmd_none(*pmd))
101 			break;
102 
103 		if (pmd_bad(*pmd)) {
104 			pr_cont("(bad)");
105 			break;
106 		}
107 
108 		/* We must not map this if we have highmem enabled */
109 		if (PageHighMem(pfn_to_page(pmd_val(*pmd) >> PAGE_SHIFT)))
110 			break;
111 
112 		pte = pte_offset_map(pmd, addr);
113 		pr_cont(", *pte=%08llx", (long long)pte_val(*pte));
114 #ifndef CONFIG_ARM_LPAE
115 		pr_cont(", *ppte=%08llx",
116 		       (long long)pte_val(pte[PTE_HWTABLE_PTRS]));
117 #endif
118 		pte_unmap(pte);
119 	} while(0);
120 
121 	pr_cont("\n");
122 }
123 #else					/* CONFIG_MMU */
124 void show_pte(struct mm_struct *mm, unsigned long addr)
125 { }
126 #endif					/* CONFIG_MMU */
127 
128 /*
129  * Oops.  The kernel tried to access some page that wasn't present.
130  */
131 static void
132 __do_kernel_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
133 		  struct pt_regs *regs)
134 {
135 	/*
136 	 * Are we prepared to handle this kernel fault?
137 	 */
138 	if (fixup_exception(regs))
139 		return;
140 
141 	/*
142 	 * No handler, we'll have to terminate things with extreme prejudice.
143 	 */
144 	bust_spinlocks(1);
145 	pr_alert("Unable to handle kernel %s at virtual address %08lx\n",
146 		 (addr < PAGE_SIZE) ? "NULL pointer dereference" :
147 		 "paging request", addr);
148 
149 	show_pte(mm, addr);
150 	die("Oops", regs, fsr);
151 	bust_spinlocks(0);
152 	do_exit(SIGKILL);
153 }
154 
155 /*
156  * Something tried to access memory that isn't in our memory map..
157  * User mode accesses just cause a SIGSEGV
158  */
159 static void
160 __do_user_fault(struct task_struct *tsk, unsigned long addr,
161 		unsigned int fsr, unsigned int sig, int code,
162 		struct pt_regs *regs)
163 {
164 	struct siginfo si;
165 
166 #ifdef CONFIG_DEBUG_USER
167 	if (((user_debug & UDBG_SEGV) && (sig == SIGSEGV)) ||
168 	    ((user_debug & UDBG_BUS)  && (sig == SIGBUS))) {
169 		printk(KERN_DEBUG "%s: unhandled page fault (%d) at 0x%08lx, code 0x%03x\n",
170 		       tsk->comm, sig, addr, fsr);
171 		show_pte(tsk->mm, addr);
172 		show_regs(regs);
173 	}
174 #endif
175 
176 	tsk->thread.address = addr;
177 	tsk->thread.error_code = fsr;
178 	tsk->thread.trap_no = 14;
179 	si.si_signo = sig;
180 	si.si_errno = 0;
181 	si.si_code = code;
182 	si.si_addr = (void __user *)addr;
183 	force_sig_info(sig, &si, tsk);
184 }
185 
186 void do_bad_area(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
187 {
188 	struct task_struct *tsk = current;
189 	struct mm_struct *mm = tsk->active_mm;
190 
191 	/*
192 	 * If we are in kernel mode at this point, we
193 	 * have no context to handle this fault with.
194 	 */
195 	if (user_mode(regs))
196 		__do_user_fault(tsk, addr, fsr, SIGSEGV, SEGV_MAPERR, regs);
197 	else
198 		__do_kernel_fault(mm, addr, fsr, regs);
199 }
200 
201 #ifdef CONFIG_MMU
202 #define VM_FAULT_BADMAP		0x010000
203 #define VM_FAULT_BADACCESS	0x020000
204 
205 /*
206  * Check that the permissions on the VMA allow for the fault which occurred.
207  * If we encountered a write fault, we must have write permission, otherwise
208  * we allow any permission.
209  */
210 static inline bool access_error(unsigned int fsr, struct vm_area_struct *vma)
211 {
212 	unsigned int mask = VM_READ | VM_WRITE | VM_EXEC;
213 
214 	if (fsr & FSR_WRITE)
215 		mask = VM_WRITE;
216 	if (fsr & FSR_LNX_PF)
217 		mask = VM_EXEC;
218 
219 	return vma->vm_flags & mask ? false : true;
220 }
221 
222 static int __kprobes
223 __do_page_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
224 		unsigned int flags, struct task_struct *tsk)
225 {
226 	struct vm_area_struct *vma;
227 	int fault;
228 
229 	vma = find_vma(mm, addr);
230 	fault = VM_FAULT_BADMAP;
231 	if (unlikely(!vma))
232 		goto out;
233 	if (unlikely(vma->vm_start > addr))
234 		goto check_stack;
235 
236 	/*
237 	 * Ok, we have a good vm_area for this
238 	 * memory access, so we can handle it.
239 	 */
240 good_area:
241 	if (access_error(fsr, vma)) {
242 		fault = VM_FAULT_BADACCESS;
243 		goto out;
244 	}
245 
246 	return handle_mm_fault(mm, vma, addr & PAGE_MASK, flags);
247 
248 check_stack:
249 	/* Don't allow expansion below FIRST_USER_ADDRESS */
250 	if (vma->vm_flags & VM_GROWSDOWN &&
251 	    addr >= FIRST_USER_ADDRESS && !expand_stack(vma, addr))
252 		goto good_area;
253 out:
254 	return fault;
255 }
256 
257 static int __kprobes
258 do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
259 {
260 	struct task_struct *tsk;
261 	struct mm_struct *mm;
262 	int fault, sig, code;
263 	unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
264 
265 	if (notify_page_fault(regs, fsr))
266 		return 0;
267 
268 	tsk = current;
269 	mm  = tsk->mm;
270 
271 	/* Enable interrupts if they were enabled in the parent context. */
272 	if (interrupts_enabled(regs))
273 		local_irq_enable();
274 
275 	/*
276 	 * If we're in an interrupt or have no user
277 	 * context, we must not take the fault..
278 	 */
279 	if (faulthandler_disabled() || !mm)
280 		goto no_context;
281 
282 	if (user_mode(regs))
283 		flags |= FAULT_FLAG_USER;
284 	if (fsr & FSR_WRITE)
285 		flags |= FAULT_FLAG_WRITE;
286 
287 	/*
288 	 * As per x86, we may deadlock here.  However, since the kernel only
289 	 * validly references user space from well defined areas of the code,
290 	 * we can bug out early if this is from code which shouldn't.
291 	 */
292 	if (!down_read_trylock(&mm->mmap_sem)) {
293 		if (!user_mode(regs) && !search_exception_tables(regs->ARM_pc))
294 			goto no_context;
295 retry:
296 		down_read(&mm->mmap_sem);
297 	} else {
298 		/*
299 		 * The above down_read_trylock() might have succeeded in
300 		 * which case, we'll have missed the might_sleep() from
301 		 * down_read()
302 		 */
303 		might_sleep();
304 #ifdef CONFIG_DEBUG_VM
305 		if (!user_mode(regs) &&
306 		    !search_exception_tables(regs->ARM_pc))
307 			goto no_context;
308 #endif
309 	}
310 
311 	fault = __do_page_fault(mm, addr, fsr, flags, tsk);
312 
313 	/* If we need to retry but a fatal signal is pending, handle the
314 	 * signal first. We do not need to release the mmap_sem because
315 	 * it would already be released in __lock_page_or_retry in
316 	 * mm/filemap.c. */
317 	if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
318 		return 0;
319 
320 	/*
321 	 * Major/minor page fault accounting is only done on the
322 	 * initial attempt. If we go through a retry, it is extremely
323 	 * likely that the page will be found in page cache at that point.
324 	 */
325 
326 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
327 	if (!(fault & VM_FAULT_ERROR) && flags & FAULT_FLAG_ALLOW_RETRY) {
328 		if (fault & VM_FAULT_MAJOR) {
329 			tsk->maj_flt++;
330 			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
331 					regs, addr);
332 		} else {
333 			tsk->min_flt++;
334 			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
335 					regs, addr);
336 		}
337 		if (fault & VM_FAULT_RETRY) {
338 			/* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
339 			* of starvation. */
340 			flags &= ~FAULT_FLAG_ALLOW_RETRY;
341 			flags |= FAULT_FLAG_TRIED;
342 			goto retry;
343 		}
344 	}
345 
346 	up_read(&mm->mmap_sem);
347 
348 	/*
349 	 * Handle the "normal" case first - VM_FAULT_MAJOR / VM_FAULT_MINOR
350 	 */
351 	if (likely(!(fault & (VM_FAULT_ERROR | VM_FAULT_BADMAP | VM_FAULT_BADACCESS))))
352 		return 0;
353 
354 	/*
355 	 * If we are in kernel mode at this point, we
356 	 * have no context to handle this fault with.
357 	 */
358 	if (!user_mode(regs))
359 		goto no_context;
360 
361 	if (fault & VM_FAULT_OOM) {
362 		/*
363 		 * We ran out of memory, call the OOM killer, and return to
364 		 * userspace (which will retry the fault, or kill us if we
365 		 * got oom-killed)
366 		 */
367 		pagefault_out_of_memory();
368 		return 0;
369 	}
370 
371 	if (fault & VM_FAULT_SIGBUS) {
372 		/*
373 		 * We had some memory, but were unable to
374 		 * successfully fix up this page fault.
375 		 */
376 		sig = SIGBUS;
377 		code = BUS_ADRERR;
378 	} else {
379 		/*
380 		 * Something tried to access memory that
381 		 * isn't in our memory map..
382 		 */
383 		sig = SIGSEGV;
384 		code = fault == VM_FAULT_BADACCESS ?
385 			SEGV_ACCERR : SEGV_MAPERR;
386 	}
387 
388 	__do_user_fault(tsk, addr, fsr, sig, code, regs);
389 	return 0;
390 
391 no_context:
392 	__do_kernel_fault(mm, addr, fsr, regs);
393 	return 0;
394 }
395 #else					/* CONFIG_MMU */
396 static int
397 do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
398 {
399 	return 0;
400 }
401 #endif					/* CONFIG_MMU */
402 
403 /*
404  * First Level Translation Fault Handler
405  *
406  * We enter here because the first level page table doesn't contain
407  * a valid entry for the address.
408  *
409  * If the address is in kernel space (>= TASK_SIZE), then we are
410  * probably faulting in the vmalloc() area.
411  *
412  * If the init_task's first level page tables contains the relevant
413  * entry, we copy the it to this task.  If not, we send the process
414  * a signal, fixup the exception, or oops the kernel.
415  *
416  * NOTE! We MUST NOT take any locks for this case. We may be in an
417  * interrupt or a critical region, and should only copy the information
418  * from the master page table, nothing more.
419  */
420 #ifdef CONFIG_MMU
421 static int __kprobes
422 do_translation_fault(unsigned long addr, unsigned int fsr,
423 		     struct pt_regs *regs)
424 {
425 	unsigned int index;
426 	pgd_t *pgd, *pgd_k;
427 	pud_t *pud, *pud_k;
428 	pmd_t *pmd, *pmd_k;
429 
430 	if (addr < TASK_SIZE)
431 		return do_page_fault(addr, fsr, regs);
432 
433 	if (user_mode(regs))
434 		goto bad_area;
435 
436 	index = pgd_index(addr);
437 
438 	pgd = cpu_get_pgd() + index;
439 	pgd_k = init_mm.pgd + index;
440 
441 	if (pgd_none(*pgd_k))
442 		goto bad_area;
443 	if (!pgd_present(*pgd))
444 		set_pgd(pgd, *pgd_k);
445 
446 	pud = pud_offset(pgd, addr);
447 	pud_k = pud_offset(pgd_k, addr);
448 
449 	if (pud_none(*pud_k))
450 		goto bad_area;
451 	if (!pud_present(*pud))
452 		set_pud(pud, *pud_k);
453 
454 	pmd = pmd_offset(pud, addr);
455 	pmd_k = pmd_offset(pud_k, addr);
456 
457 #ifdef CONFIG_ARM_LPAE
458 	/*
459 	 * Only one hardware entry per PMD with LPAE.
460 	 */
461 	index = 0;
462 #else
463 	/*
464 	 * On ARM one Linux PGD entry contains two hardware entries (see page
465 	 * tables layout in pgtable.h). We normally guarantee that we always
466 	 * fill both L1 entries. But create_mapping() doesn't follow the rule.
467 	 * It can create inidividual L1 entries, so here we have to call
468 	 * pmd_none() check for the entry really corresponded to address, not
469 	 * for the first of pair.
470 	 */
471 	index = (addr >> SECTION_SHIFT) & 1;
472 #endif
473 	if (pmd_none(pmd_k[index]))
474 		goto bad_area;
475 
476 	copy_pmd(pmd, pmd_k);
477 	return 0;
478 
479 bad_area:
480 	do_bad_area(addr, fsr, regs);
481 	return 0;
482 }
483 #else					/* CONFIG_MMU */
484 static int
485 do_translation_fault(unsigned long addr, unsigned int fsr,
486 		     struct pt_regs *regs)
487 {
488 	return 0;
489 }
490 #endif					/* CONFIG_MMU */
491 
492 /*
493  * Some section permission faults need to be handled gracefully.
494  * They can happen due to a __{get,put}_user during an oops.
495  */
496 #ifndef CONFIG_ARM_LPAE
497 static int
498 do_sect_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
499 {
500 	do_bad_area(addr, fsr, regs);
501 	return 0;
502 }
503 #endif /* CONFIG_ARM_LPAE */
504 
505 /*
506  * This abort handler always returns "fault".
507  */
508 static int
509 do_bad(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
510 {
511 	return 1;
512 }
513 
514 struct fsr_info {
515 	int	(*fn)(unsigned long addr, unsigned int fsr, struct pt_regs *regs);
516 	int	sig;
517 	int	code;
518 	const char *name;
519 };
520 
521 /* FSR definition */
522 #ifdef CONFIG_ARM_LPAE
523 #include "fsr-3level.c"
524 #else
525 #include "fsr-2level.c"
526 #endif
527 
528 void __init
529 hook_fault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
530 		int sig, int code, const char *name)
531 {
532 	if (nr < 0 || nr >= ARRAY_SIZE(fsr_info))
533 		BUG();
534 
535 	fsr_info[nr].fn   = fn;
536 	fsr_info[nr].sig  = sig;
537 	fsr_info[nr].code = code;
538 	fsr_info[nr].name = name;
539 }
540 
541 /*
542  * Dispatch a data abort to the relevant handler.
543  */
544 asmlinkage void __exception
545 do_DataAbort(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
546 {
547 	const struct fsr_info *inf = fsr_info + fsr_fs(fsr);
548 	struct siginfo info;
549 
550 	if (!inf->fn(addr, fsr & ~FSR_LNX_PF, regs))
551 		return;
552 
553 	pr_alert("Unhandled fault: %s (0x%03x) at 0x%08lx\n",
554 		inf->name, fsr, addr);
555 	show_pte(current->mm, addr);
556 
557 	info.si_signo = inf->sig;
558 	info.si_errno = 0;
559 	info.si_code  = inf->code;
560 	info.si_addr  = (void __user *)addr;
561 	arm_notify_die("", regs, &info, fsr, 0);
562 }
563 
564 void __init
565 hook_ifault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
566 		 int sig, int code, const char *name)
567 {
568 	if (nr < 0 || nr >= ARRAY_SIZE(ifsr_info))
569 		BUG();
570 
571 	ifsr_info[nr].fn   = fn;
572 	ifsr_info[nr].sig  = sig;
573 	ifsr_info[nr].code = code;
574 	ifsr_info[nr].name = name;
575 }
576 
577 asmlinkage void __exception
578 do_PrefetchAbort(unsigned long addr, unsigned int ifsr, struct pt_regs *regs)
579 {
580 	const struct fsr_info *inf = ifsr_info + fsr_fs(ifsr);
581 	struct siginfo info;
582 
583 	if (!inf->fn(addr, ifsr | FSR_LNX_PF, regs))
584 		return;
585 
586 	pr_alert("Unhandled prefetch abort: %s (0x%03x) at 0x%08lx\n",
587 		inf->name, ifsr, addr);
588 
589 	info.si_signo = inf->sig;
590 	info.si_errno = 0;
591 	info.si_code  = inf->code;
592 	info.si_addr  = (void __user *)addr;
593 	arm_notify_die("", regs, &info, ifsr, 0);
594 }
595 
596 /*
597  * Abort handler to be used only during first unmasking of asynchronous aborts
598  * on the boot CPU. This makes sure that the machine will not die if the
599  * firmware/bootloader left an imprecise abort pending for us to trip over.
600  */
601 static int __init early_abort_handler(unsigned long addr, unsigned int fsr,
602 				      struct pt_regs *regs)
603 {
604 	pr_warn("Hit pending asynchronous external abort (FSR=0x%08x) during "
605 		"first unmask, this is most likely caused by a "
606 		"firmware/bootloader bug.\n", fsr);
607 
608 	return 0;
609 }
610 
611 void __init early_abt_enable(void)
612 {
613 	fsr_info[22].fn = early_abort_handler;
614 	local_abt_enable();
615 	fsr_info[22].fn = do_bad;
616 }
617 
618 #ifndef CONFIG_ARM_LPAE
619 static int __init exceptions_init(void)
620 {
621 	if (cpu_architecture() >= CPU_ARCH_ARMv6) {
622 		hook_fault_code(4, do_translation_fault, SIGSEGV, SEGV_MAPERR,
623 				"I-cache maintenance fault");
624 	}
625 
626 	if (cpu_architecture() >= CPU_ARCH_ARMv7) {
627 		/*
628 		 * TODO: Access flag faults introduced in ARMv6K.
629 		 * Runtime check for 'K' extension is needed
630 		 */
631 		hook_fault_code(3, do_bad, SIGSEGV, SEGV_MAPERR,
632 				"section access flag fault");
633 		hook_fault_code(6, do_bad, SIGSEGV, SEGV_MAPERR,
634 				"section access flag fault");
635 	}
636 
637 	return 0;
638 }
639 
640 arch_initcall(exceptions_init);
641 #endif
642