Linux Audio

Check our new training course

Loading...
v3.15
 
  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
 34static 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
 49static 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 */
 59void show_pte(struct mm_struct *mm, unsigned long addr)
 60{
 61	pgd_t *pgd;
 62
 63	if (!mm)
 64		mm = &init_mm;
 65
 66	printk(KERN_ALERT "pgd = %p\n", mm->pgd);
 67	pgd = pgd_offset(mm, addr);
 68	printk(KERN_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			printk("(bad)");
 81			break;
 82		}
 83
 84		pud = pud_offset(pgd, addr);
 85		if (PTRS_PER_PUD != 1)
 86			printk(", *pud=%08llx", (long long)pud_val(*pud));
 87
 88		if (pud_none(*pud))
 89			break;
 90
 91		if (pud_bad(*pud)) {
 92			printk("(bad)");
 93			break;
 94		}
 95
 96		pmd = pmd_offset(pud, addr);
 97		if (PTRS_PER_PMD != 1)
 98			printk(", *pmd=%08llx", (long long)pmd_val(*pmd));
 99
100		if (pmd_none(*pmd))
101			break;
102
103		if (pmd_bad(*pmd)) {
104			printk("(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		printk(", *pte=%08llx", (long long)pte_val(*pte));
114#ifndef CONFIG_ARM_LPAE
115		printk(", *ppte=%08llx",
116		       (long long)pte_val(pte[PTE_HWTABLE_PTRS]));
117#endif
118		pte_unmap(pte);
119	} while(0);
120
121	printk("\n");
122}
123#else					/* CONFIG_MMU */
124void 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 */
131static 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	printk(KERN_ALERT
146		"Unable to handle kernel %s at virtual address %08lx\n",
147		(addr < PAGE_SIZE) ? "NULL pointer dereference" :
148		"paging request", addr);
149
150	show_pte(mm, addr);
151	die("Oops", regs, fsr);
152	bust_spinlocks(0);
153	do_exit(SIGKILL);
154}
155
156/*
157 * Something tried to access memory that isn't in our memory map..
158 * User mode accesses just cause a SIGSEGV
159 */
160static void
161__do_user_fault(struct task_struct *tsk, unsigned long addr,
162		unsigned int fsr, unsigned int sig, int code,
163		struct pt_regs *regs)
164{
165	struct siginfo si;
 
 
 
166
167#ifdef CONFIG_DEBUG_USER
168	if (((user_debug & UDBG_SEGV) && (sig == SIGSEGV)) ||
169	    ((user_debug & UDBG_BUS)  && (sig == SIGBUS))) {
170		printk(KERN_DEBUG "%s: unhandled page fault (%d) at 0x%08lx, code 0x%03x\n",
 
171		       tsk->comm, sig, addr, fsr);
172		show_pte(tsk->mm, addr);
173		show_regs(regs);
174	}
175#endif
 
 
 
 
 
 
176
177	tsk->thread.address = addr;
178	tsk->thread.error_code = fsr;
179	tsk->thread.trap_no = 14;
180	si.si_signo = sig;
181	si.si_errno = 0;
182	si.si_code = code;
183	si.si_addr = (void __user *)addr;
184	force_sig_info(sig, &si, tsk);
185}
186
187void do_bad_area(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
188{
189	struct task_struct *tsk = current;
190	struct mm_struct *mm = tsk->active_mm;
191
192	/*
193	 * If we are in kernel mode at this point, we
194	 * have no context to handle this fault with.
195	 */
196	if (user_mode(regs))
197		__do_user_fault(tsk, addr, fsr, SIGSEGV, SEGV_MAPERR, regs);
198	else
199		__do_kernel_fault(mm, addr, fsr, regs);
200}
201
202#ifdef CONFIG_MMU
203#define VM_FAULT_BADMAP		0x010000
204#define VM_FAULT_BADACCESS	0x020000
205
206/*
207 * Check that the permissions on the VMA allow for the fault which occurred.
208 * If we encountered a write fault, we must have write permission, otherwise
209 * we allow any permission.
210 */
211static inline bool access_error(unsigned int fsr, struct vm_area_struct *vma)
212{
213	unsigned int mask = VM_READ | VM_WRITE | VM_EXEC;
214
215	if (fsr & FSR_WRITE)
216		mask = VM_WRITE;
217	if (fsr & FSR_LNX_PF)
218		mask = VM_EXEC;
219
220	return vma->vm_flags & mask ? false : true;
221}
222
223static int __kprobes
224__do_page_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
225		unsigned int flags, struct task_struct *tsk)
 
226{
227	struct vm_area_struct *vma;
228	int fault;
229
230	vma = find_vma(mm, addr);
231	fault = VM_FAULT_BADMAP;
232	if (unlikely(!vma))
233		goto out;
234	if (unlikely(vma->vm_start > addr))
235		goto check_stack;
236
237	/*
238	 * Ok, we have a good vm_area for this
239	 * memory access, so we can handle it.
240	 */
241good_area:
242	if (access_error(fsr, vma)) {
243		fault = VM_FAULT_BADACCESS;
244		goto out;
245	}
246
247	return handle_mm_fault(mm, vma, addr & PAGE_MASK, flags);
248
249check_stack:
250	/* Don't allow expansion below FIRST_USER_ADDRESS */
251	if (vma->vm_flags & VM_GROWSDOWN &&
252	    addr >= FIRST_USER_ADDRESS && !expand_stack(vma, addr))
253		goto good_area;
254out:
255	return fault;
256}
257
258static int __kprobes
259do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
260{
261	struct task_struct *tsk;
262	struct mm_struct *mm;
263	int fault, sig, code;
264	unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
 
265
266	if (notify_page_fault(regs, fsr))
267		return 0;
268
269	tsk = current;
270	mm  = tsk->mm;
271
272	/* Enable interrupts if they were enabled in the parent context. */
273	if (interrupts_enabled(regs))
274		local_irq_enable();
275
276	/*
277	 * If we're in an interrupt or have no user
278	 * context, we must not take the fault..
279	 */
280	if (in_atomic() || !mm)
281		goto no_context;
282
283	if (user_mode(regs))
284		flags |= FAULT_FLAG_USER;
285	if (fsr & FSR_WRITE)
286		flags |= FAULT_FLAG_WRITE;
287
 
 
288	/*
289	 * As per x86, we may deadlock here.  However, since the kernel only
290	 * validly references user space from well defined areas of the code,
291	 * we can bug out early if this is from code which shouldn't.
292	 */
293	if (!down_read_trylock(&mm->mmap_sem)) {
294		if (!user_mode(regs) && !search_exception_tables(regs->ARM_pc))
295			goto no_context;
296retry:
297		down_read(&mm->mmap_sem);
298	} else {
299		/*
300		 * The above down_read_trylock() might have succeeded in
301		 * which case, we'll have missed the might_sleep() from
302		 * down_read()
303		 */
304		might_sleep();
305#ifdef CONFIG_DEBUG_VM
306		if (!user_mode(regs) &&
307		    !search_exception_tables(regs->ARM_pc))
308			goto no_context;
309#endif
310	}
311
312	fault = __do_page_fault(mm, addr, fsr, flags, tsk);
313
314	/* If we need to retry but a fatal signal is pending, handle the
315	 * signal first. We do not need to release the mmap_sem because
316	 * it would already be released in __lock_page_or_retry in
317	 * mm/filemap.c. */
318	if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
 
 
319		return 0;
 
320
321	/*
322	 * Major/minor page fault accounting is only done on the
323	 * initial attempt. If we go through a retry, it is extremely
324	 * likely that the page will be found in page cache at that point.
325	 */
326
327	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
328	if (!(fault & VM_FAULT_ERROR) && flags & FAULT_FLAG_ALLOW_RETRY) {
329		if (fault & VM_FAULT_MAJOR) {
330			tsk->maj_flt++;
331			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
332					regs, addr);
333		} else {
334			tsk->min_flt++;
335			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
336					regs, addr);
337		}
338		if (fault & VM_FAULT_RETRY) {
339			/* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
340			* of starvation. */
341			flags &= ~FAULT_FLAG_ALLOW_RETRY;
342			flags |= FAULT_FLAG_TRIED;
343			goto retry;
344		}
345	}
346
347	up_read(&mm->mmap_sem);
348
349	/*
350	 * Handle the "normal" case first - VM_FAULT_MAJOR / VM_FAULT_MINOR
351	 */
352	if (likely(!(fault & (VM_FAULT_ERROR | VM_FAULT_BADMAP | VM_FAULT_BADACCESS))))
353		return 0;
354
355	/*
356	 * If we are in kernel mode at this point, we
357	 * have no context to handle this fault with.
358	 */
359	if (!user_mode(regs))
360		goto no_context;
361
362	if (fault & VM_FAULT_OOM) {
363		/*
364		 * We ran out of memory, call the OOM killer, and return to
365		 * userspace (which will retry the fault, or kill us if we
366		 * got oom-killed)
367		 */
368		pagefault_out_of_memory();
369		return 0;
370	}
371
372	if (fault & VM_FAULT_SIGBUS) {
373		/*
374		 * We had some memory, but were unable to
375		 * successfully fix up this page fault.
376		 */
377		sig = SIGBUS;
378		code = BUS_ADRERR;
379	} else {
380		/*
381		 * Something tried to access memory that
382		 * isn't in our memory map..
383		 */
384		sig = SIGSEGV;
385		code = fault == VM_FAULT_BADACCESS ?
386			SEGV_ACCERR : SEGV_MAPERR;
387	}
388
389	__do_user_fault(tsk, addr, fsr, sig, code, regs);
390	return 0;
391
392no_context:
393	__do_kernel_fault(mm, addr, fsr, regs);
394	return 0;
395}
396#else					/* CONFIG_MMU */
397static int
398do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
399{
400	return 0;
401}
402#endif					/* CONFIG_MMU */
403
404/*
405 * First Level Translation Fault Handler
406 *
407 * We enter here because the first level page table doesn't contain
408 * a valid entry for the address.
409 *
410 * If the address is in kernel space (>= TASK_SIZE), then we are
411 * probably faulting in the vmalloc() area.
412 *
413 * If the init_task's first level page tables contains the relevant
414 * entry, we copy the it to this task.  If not, we send the process
415 * a signal, fixup the exception, or oops the kernel.
416 *
417 * NOTE! We MUST NOT take any locks for this case. We may be in an
418 * interrupt or a critical region, and should only copy the information
419 * from the master page table, nothing more.
420 */
421#ifdef CONFIG_MMU
422static int __kprobes
423do_translation_fault(unsigned long addr, unsigned int fsr,
424		     struct pt_regs *regs)
425{
426	unsigned int index;
427	pgd_t *pgd, *pgd_k;
 
428	pud_t *pud, *pud_k;
429	pmd_t *pmd, *pmd_k;
430
431	if (addr < TASK_SIZE)
432		return do_page_fault(addr, fsr, regs);
433
434	if (user_mode(regs))
435		goto bad_area;
436
437	index = pgd_index(addr);
438
439	pgd = cpu_get_pgd() + index;
440	pgd_k = init_mm.pgd + index;
441
442	if (pgd_none(*pgd_k))
 
 
 
443		goto bad_area;
444	if (!pgd_present(*pgd))
445		set_pgd(pgd, *pgd_k);
446
447	pud = pud_offset(pgd, addr);
448	pud_k = pud_offset(pgd_k, addr);
449
450	if (pud_none(*pud_k))
451		goto bad_area;
452	if (!pud_present(*pud))
453		set_pud(pud, *pud_k);
454
455	pmd = pmd_offset(pud, addr);
456	pmd_k = pmd_offset(pud_k, addr);
457
458#ifdef CONFIG_ARM_LPAE
459	/*
460	 * Only one hardware entry per PMD with LPAE.
461	 */
462	index = 0;
463#else
464	/*
465	 * On ARM one Linux PGD entry contains two hardware entries (see page
466	 * tables layout in pgtable.h). We normally guarantee that we always
467	 * fill both L1 entries. But create_mapping() doesn't follow the rule.
468	 * It can create inidividual L1 entries, so here we have to call
469	 * pmd_none() check for the entry really corresponded to address, not
470	 * for the first of pair.
471	 */
472	index = (addr >> SECTION_SHIFT) & 1;
473#endif
474	if (pmd_none(pmd_k[index]))
475		goto bad_area;
476
477	copy_pmd(pmd, pmd_k);
478	return 0;
479
480bad_area:
481	do_bad_area(addr, fsr, regs);
482	return 0;
483}
484#else					/* CONFIG_MMU */
485static int
486do_translation_fault(unsigned long addr, unsigned int fsr,
487		     struct pt_regs *regs)
488{
489	return 0;
490}
491#endif					/* CONFIG_MMU */
492
493/*
494 * Some section permission faults need to be handled gracefully.
495 * They can happen due to a __{get,put}_user during an oops.
496 */
497#ifndef CONFIG_ARM_LPAE
498static int
499do_sect_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
500{
501	do_bad_area(addr, fsr, regs);
502	return 0;
503}
504#endif /* CONFIG_ARM_LPAE */
505
506/*
507 * This abort handler always returns "fault".
508 */
509static int
510do_bad(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
511{
512	return 1;
513}
514
515struct fsr_info {
516	int	(*fn)(unsigned long addr, unsigned int fsr, struct pt_regs *regs);
517	int	sig;
518	int	code;
519	const char *name;
520};
521
522/* FSR definition */
523#ifdef CONFIG_ARM_LPAE
524#include "fsr-3level.c"
525#else
526#include "fsr-2level.c"
527#endif
528
529void __init
530hook_fault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
531		int sig, int code, const char *name)
532{
533	if (nr < 0 || nr >= ARRAY_SIZE(fsr_info))
534		BUG();
535
536	fsr_info[nr].fn   = fn;
537	fsr_info[nr].sig  = sig;
538	fsr_info[nr].code = code;
539	fsr_info[nr].name = name;
540}
541
542/*
543 * Dispatch a data abort to the relevant handler.
544 */
545asmlinkage void __exception
546do_DataAbort(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
547{
548	const struct fsr_info *inf = fsr_info + fsr_fs(fsr);
549	struct siginfo info;
550
551	if (!inf->fn(addr, fsr & ~FSR_LNX_PF, regs))
552		return;
553
554	printk(KERN_ALERT "Unhandled fault: %s (0x%03x) at 0x%08lx\n",
 
555		inf->name, fsr, 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
564void __init
565hook_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
577asmlinkage void __exception
578do_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	printk(KERN_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#ifndef CONFIG_ARM_LPAE
597static int __init exceptions_init(void)
598{
599	if (cpu_architecture() >= CPU_ARCH_ARMv6) {
600		hook_fault_code(4, do_translation_fault, SIGSEGV, SEGV_MAPERR,
601				"I-cache maintenance fault");
602	}
603
604	if (cpu_architecture() >= CPU_ARCH_ARMv7) {
605		/*
606		 * TODO: Access flag faults introduced in ARMv6K.
607		 * Runtime check for 'K' extension is needed
608		 */
609		hook_fault_code(3, do_bad, SIGSEGV, SEGV_MAPERR,
610				"section access flag fault");
611		hook_fault_code(6, do_bad, SIGSEGV, SEGV_MAPERR,
612				"section access flag fault");
613	}
614
615	return 0;
616}
617
618arch_initcall(exceptions_init);
619#endif
v5.9
  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
 
 
 21#include <asm/system_misc.h>
 22#include <asm/system_info.h>
 23#include <asm/tlbflush.h>
 24
 25#include "fault.h"
 26
 27#ifdef CONFIG_MMU
 28
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 29/*
 30 * This is useful to dump out the page tables associated with
 31 * 'addr' in mm 'mm'.
 32 */
 33void show_pte(const char *lvl, struct mm_struct *mm, unsigned long addr)
 34{
 35	pgd_t *pgd;
 36
 37	if (!mm)
 38		mm = &init_mm;
 39
 40	printk("%spgd = %p\n", lvl, mm->pgd);
 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 */
 99void show_pte(const char *lvl, struct mm_struct *mm, unsigned long addr)
100{ }
101#endif					/* CONFIG_MMU */
102
103/*
104 * Oops.  The kernel tried to access some page that wasn't present.
105 */
106static void
107__do_kernel_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
108		  struct pt_regs *regs)
109{
110	/*
111	 * Are we prepared to handle this kernel fault?
112	 */
113	if (fixup_exception(regs))
114		return;
115
116	/*
117	 * No handler, we'll have to terminate things with extreme prejudice.
118	 */
119	bust_spinlocks(1);
120	pr_alert("8<--- cut here ---\n");
121	pr_alert("Unable to handle kernel %s at virtual address %08lx\n",
122		 (addr < PAGE_SIZE) ? "NULL pointer dereference" :
123		 "paging request", addr);
124
125	show_pte(KERN_ALERT, mm, addr);
126	die("Oops", regs, fsr);
127	bust_spinlocks(0);
128	do_exit(SIGKILL);
129}
130
131/*
132 * Something tried to access memory that isn't in our memory map..
133 * User mode accesses just cause a SIGSEGV
134 */
135static void
136__do_user_fault(unsigned long addr, unsigned int fsr, unsigned int sig,
137		int code, struct pt_regs *regs)
 
138{
139	struct task_struct *tsk = current;
140
141	if (addr > TASK_SIZE)
142		harden_branch_predictor();
143
144#ifdef CONFIG_DEBUG_USER
145	if (((user_debug & UDBG_SEGV) && (sig == SIGSEGV)) ||
146	    ((user_debug & UDBG_BUS)  && (sig == SIGBUS))) {
147		pr_err("8<--- cut here ---\n");
148		pr_err("%s: unhandled page fault (%d) at 0x%08lx, code 0x%03x\n",
149		       tsk->comm, sig, addr, fsr);
150		show_pte(KERN_ERR, tsk->mm, addr);
151		show_regs(regs);
152	}
153#endif
154#ifndef CONFIG_KUSER_HELPERS
155	if ((sig == SIGSEGV) && ((addr & PAGE_MASK) == 0xffff0000))
156		printk_ratelimited(KERN_DEBUG
157				   "%s: CONFIG_KUSER_HELPERS disabled at 0x%08lx\n",
158				   tsk->comm, addr);
159#endif
160
161	tsk->thread.address = addr;
162	tsk->thread.error_code = fsr;
163	tsk->thread.trap_no = 14;
164	force_sig_fault(sig, code, (void __user *)addr);
 
 
 
 
165}
166
167void do_bad_area(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
168{
169	struct task_struct *tsk = current;
170	struct mm_struct *mm = tsk->active_mm;
171
172	/*
173	 * If we are in kernel mode at this point, we
174	 * have no context to handle this fault with.
175	 */
176	if (user_mode(regs))
177		__do_user_fault(addr, fsr, SIGSEGV, SEGV_MAPERR, regs);
178	else
179		__do_kernel_fault(mm, addr, fsr, regs);
180}
181
182#ifdef CONFIG_MMU
183#define VM_FAULT_BADMAP		0x010000
184#define VM_FAULT_BADACCESS	0x020000
185
186/*
187 * Check that the permissions on the VMA allow for the fault which occurred.
188 * If we encountered a write fault, we must have write permission, otherwise
189 * we allow any permission.
190 */
191static inline bool access_error(unsigned int fsr, struct vm_area_struct *vma)
192{
193	unsigned int mask = VM_ACCESS_FLAGS;
194
195	if ((fsr & FSR_WRITE) && !(fsr & FSR_CM))
196		mask = VM_WRITE;
197	if (fsr & FSR_LNX_PF)
198		mask = VM_EXEC;
199
200	return vma->vm_flags & mask ? false : true;
201}
202
203static vm_fault_t __kprobes
204__do_page_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
205		unsigned int flags, struct task_struct *tsk,
206		struct pt_regs *regs)
207{
208	struct vm_area_struct *vma;
209	vm_fault_t fault;
210
211	vma = find_vma(mm, addr);
212	fault = VM_FAULT_BADMAP;
213	if (unlikely(!vma))
214		goto out;
215	if (unlikely(vma->vm_start > addr))
216		goto check_stack;
217
218	/*
219	 * Ok, we have a good vm_area for this
220	 * memory access, so we can handle it.
221	 */
222good_area:
223	if (access_error(fsr, vma)) {
224		fault = VM_FAULT_BADACCESS;
225		goto out;
226	}
227
228	return handle_mm_fault(vma, addr & PAGE_MASK, flags, regs);
229
230check_stack:
231	/* Don't allow expansion below FIRST_USER_ADDRESS */
232	if (vma->vm_flags & VM_GROWSDOWN &&
233	    addr >= FIRST_USER_ADDRESS && !expand_stack(vma, addr))
234		goto good_area;
235out:
236	return fault;
237}
238
239static int __kprobes
240do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
241{
242	struct task_struct *tsk;
243	struct mm_struct *mm;
244	int sig, code;
245	vm_fault_t fault;
246	unsigned int flags = FAULT_FLAG_DEFAULT;
247
248	if (kprobe_page_fault(regs, fsr))
249		return 0;
250
251	tsk = current;
252	mm  = tsk->mm;
253
254	/* Enable interrupts if they were enabled in the parent context. */
255	if (interrupts_enabled(regs))
256		local_irq_enable();
257
258	/*
259	 * If we're in an interrupt or have no user
260	 * context, we must not take the fault..
261	 */
262	if (faulthandler_disabled() || !mm)
263		goto no_context;
264
265	if (user_mode(regs))
266		flags |= FAULT_FLAG_USER;
267	if ((fsr & FSR_WRITE) && !(fsr & FSR_CM))
268		flags |= FAULT_FLAG_WRITE;
269
270	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
271
272	/*
273	 * As per x86, we may deadlock here.  However, since the kernel only
274	 * validly references user space from well defined areas of the code,
275	 * we can bug out early if this is from code which shouldn't.
276	 */
277	if (!mmap_read_trylock(mm)) {
278		if (!user_mode(regs) && !search_exception_tables(regs->ARM_pc))
279			goto no_context;
280retry:
281		mmap_read_lock(mm);
282	} else {
283		/*
284		 * The above down_read_trylock() might have succeeded in
285		 * which case, we'll have missed the might_sleep() from
286		 * down_read()
287		 */
288		might_sleep();
289#ifdef CONFIG_DEBUG_VM
290		if (!user_mode(regs) &&
291		    !search_exception_tables(regs->ARM_pc))
292			goto no_context;
293#endif
294	}
295
296	fault = __do_page_fault(mm, addr, fsr, flags, tsk, regs);
297
298	/* If we need to retry but a fatal signal is pending, handle the
299	 * signal first. We do not need to release the mmap_lock because
300	 * it would already be released in __lock_page_or_retry in
301	 * mm/filemap.c. */
302	if (fault_signal_pending(fault, regs)) {
303		if (!user_mode(regs))
304			goto no_context;
305		return 0;
306	}
307
 
 
 
 
 
 
 
308	if (!(fault & VM_FAULT_ERROR) && flags & FAULT_FLAG_ALLOW_RETRY) {
 
 
 
 
 
 
 
 
 
309		if (fault & VM_FAULT_RETRY) {
 
 
 
310			flags |= FAULT_FLAG_TRIED;
311			goto retry;
312		}
313	}
314
315	mmap_read_unlock(mm);
316
317	/*
318	 * Handle the "normal" case first - VM_FAULT_MAJOR
319	 */
320	if (likely(!(fault & (VM_FAULT_ERROR | VM_FAULT_BADMAP | VM_FAULT_BADACCESS))))
321		return 0;
322
323	/*
324	 * If we are in kernel mode at this point, we
325	 * have no context to handle this fault with.
326	 */
327	if (!user_mode(regs))
328		goto no_context;
329
330	if (fault & VM_FAULT_OOM) {
331		/*
332		 * We ran out of memory, call the OOM killer, and return to
333		 * userspace (which will retry the fault, or kill us if we
334		 * got oom-killed)
335		 */
336		pagefault_out_of_memory();
337		return 0;
338	}
339
340	if (fault & VM_FAULT_SIGBUS) {
341		/*
342		 * We had some memory, but were unable to
343		 * successfully fix up this page fault.
344		 */
345		sig = SIGBUS;
346		code = BUS_ADRERR;
347	} else {
348		/*
349		 * Something tried to access memory that
350		 * isn't in our memory map..
351		 */
352		sig = SIGSEGV;
353		code = fault == VM_FAULT_BADACCESS ?
354			SEGV_ACCERR : SEGV_MAPERR;
355	}
356
357	__do_user_fault(addr, fsr, sig, code, regs);
358	return 0;
359
360no_context:
361	__do_kernel_fault(mm, addr, fsr, regs);
362	return 0;
363}
364#else					/* CONFIG_MMU */
365static int
366do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
367{
368	return 0;
369}
370#endif					/* CONFIG_MMU */
371
372/*
373 * First Level Translation Fault Handler
374 *
375 * We enter here because the first level page table doesn't contain
376 * a valid entry for the address.
377 *
378 * If the address is in kernel space (>= TASK_SIZE), then we are
379 * probably faulting in the vmalloc() area.
380 *
381 * If the init_task's first level page tables contains the relevant
382 * entry, we copy the it to this task.  If not, we send the process
383 * a signal, fixup the exception, or oops the kernel.
384 *
385 * NOTE! We MUST NOT take any locks for this case. We may be in an
386 * interrupt or a critical region, and should only copy the information
387 * from the master page table, nothing more.
388 */
389#ifdef CONFIG_MMU
390static int __kprobes
391do_translation_fault(unsigned long addr, unsigned int fsr,
392		     struct pt_regs *regs)
393{
394	unsigned int index;
395	pgd_t *pgd, *pgd_k;
396	p4d_t *p4d, *p4d_k;
397	pud_t *pud, *pud_k;
398	pmd_t *pmd, *pmd_k;
399
400	if (addr < TASK_SIZE)
401		return do_page_fault(addr, fsr, regs);
402
403	if (user_mode(regs))
404		goto bad_area;
405
406	index = pgd_index(addr);
407
408	pgd = cpu_get_pgd() + index;
409	pgd_k = init_mm.pgd + index;
410
411	p4d = p4d_offset(pgd, addr);
412	p4d_k = p4d_offset(pgd_k, addr);
413
414	if (p4d_none(*p4d_k))
415		goto bad_area;
416	if (!p4d_present(*p4d))
417		set_p4d(p4d, *p4d_k);
418
419	pud = pud_offset(p4d, addr);
420	pud_k = pud_offset(p4d_k, addr);
421
422	if (pud_none(*pud_k))
423		goto bad_area;
424	if (!pud_present(*pud))
425		set_pud(pud, *pud_k);
426
427	pmd = pmd_offset(pud, addr);
428	pmd_k = pmd_offset(pud_k, addr);
429
430#ifdef CONFIG_ARM_LPAE
431	/*
432	 * Only one hardware entry per PMD with LPAE.
433	 */
434	index = 0;
435#else
436	/*
437	 * On ARM one Linux PGD entry contains two hardware entries (see page
438	 * tables layout in pgtable.h). We normally guarantee that we always
439	 * fill both L1 entries. But create_mapping() doesn't follow the rule.
440	 * It can create inidividual L1 entries, so here we have to call
441	 * pmd_none() check for the entry really corresponded to address, not
442	 * for the first of pair.
443	 */
444	index = (addr >> SECTION_SHIFT) & 1;
445#endif
446	if (pmd_none(pmd_k[index]))
447		goto bad_area;
448
449	copy_pmd(pmd, pmd_k);
450	return 0;
451
452bad_area:
453	do_bad_area(addr, fsr, regs);
454	return 0;
455}
456#else					/* CONFIG_MMU */
457static int
458do_translation_fault(unsigned long addr, unsigned int fsr,
459		     struct pt_regs *regs)
460{
461	return 0;
462}
463#endif					/* CONFIG_MMU */
464
465/*
466 * Some section permission faults need to be handled gracefully.
467 * They can happen due to a __{get,put}_user during an oops.
468 */
469#ifndef CONFIG_ARM_LPAE
470static int
471do_sect_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
472{
473	do_bad_area(addr, fsr, regs);
474	return 0;
475}
476#endif /* CONFIG_ARM_LPAE */
477
478/*
479 * This abort handler always returns "fault".
480 */
481static int
482do_bad(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
483{
484	return 1;
485}
486
487struct fsr_info {
488	int	(*fn)(unsigned long addr, unsigned int fsr, struct pt_regs *regs);
489	int	sig;
490	int	code;
491	const char *name;
492};
493
494/* FSR definition */
495#ifdef CONFIG_ARM_LPAE
496#include "fsr-3level.c"
497#else
498#include "fsr-2level.c"
499#endif
500
501void __init
502hook_fault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
503		int sig, int code, const char *name)
504{
505	if (nr < 0 || nr >= ARRAY_SIZE(fsr_info))
506		BUG();
507
508	fsr_info[nr].fn   = fn;
509	fsr_info[nr].sig  = sig;
510	fsr_info[nr].code = code;
511	fsr_info[nr].name = name;
512}
513
514/*
515 * Dispatch a data abort to the relevant handler.
516 */
517asmlinkage void
518do_DataAbort(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
519{
520	const struct fsr_info *inf = fsr_info + fsr_fs(fsr);
 
521
522	if (!inf->fn(addr, fsr & ~FSR_LNX_PF, regs))
523		return;
524
525	pr_alert("8<--- cut here ---\n");
526	pr_alert("Unhandled fault: %s (0x%03x) at 0x%08lx\n",
527		inf->name, fsr, addr);
528	show_pte(KERN_ALERT, current->mm, addr);
529
530	arm_notify_die("", regs, inf->sig, inf->code, (void __user *)addr,
531		       fsr, 0);
 
 
 
532}
533
534void __init
535hook_ifault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
536		 int sig, int code, const char *name)
537{
538	if (nr < 0 || nr >= ARRAY_SIZE(ifsr_info))
539		BUG();
540
541	ifsr_info[nr].fn   = fn;
542	ifsr_info[nr].sig  = sig;
543	ifsr_info[nr].code = code;
544	ifsr_info[nr].name = name;
545}
546
547asmlinkage void
548do_PrefetchAbort(unsigned long addr, unsigned int ifsr, struct pt_regs *regs)
549{
550	const struct fsr_info *inf = ifsr_info + fsr_fs(ifsr);
 
551
552	if (!inf->fn(addr, ifsr | FSR_LNX_PF, regs))
553		return;
554
555	pr_alert("Unhandled prefetch abort: %s (0x%03x) at 0x%08lx\n",
556		inf->name, ifsr, addr);
557
558	arm_notify_die("", regs, inf->sig, inf->code, (void __user *)addr,
559		       ifsr, 0);
560}
561
562/*
563 * Abort handler to be used only during first unmasking of asynchronous aborts
564 * on the boot CPU. This makes sure that the machine will not die if the
565 * firmware/bootloader left an imprecise abort pending for us to trip over.
566 */
567static int __init early_abort_handler(unsigned long addr, unsigned int fsr,
568				      struct pt_regs *regs)
569{
570	pr_warn("Hit pending asynchronous external abort (FSR=0x%08x) during "
571		"first unmask, this is most likely caused by a "
572		"firmware/bootloader bug.\n", fsr);
573
574	return 0;
575}
576
577void __init early_abt_enable(void)
578{
579	fsr_info[FSR_FS_AEA].fn = early_abort_handler;
580	local_abt_enable();
581	fsr_info[FSR_FS_AEA].fn = do_bad;
582}
583
584#ifndef CONFIG_ARM_LPAE
585static int __init exceptions_init(void)
586{
587	if (cpu_architecture() >= CPU_ARCH_ARMv6) {
588		hook_fault_code(4, do_translation_fault, SIGSEGV, SEGV_MAPERR,
589				"I-cache maintenance fault");
590	}
591
592	if (cpu_architecture() >= CPU_ARCH_ARMv7) {
593		/*
594		 * TODO: Access flag faults introduced in ARMv6K.
595		 * Runtime check for 'K' extension is needed
596		 */
597		hook_fault_code(3, do_bad, SIGSEGV, SEGV_MAPERR,
598				"section access flag fault");
599		hook_fault_code(6, do_bad, SIGSEGV, SEGV_MAPERR,
600				"section access flag fault");
601	}
602
603	return 0;
604}
605
606arch_initcall(exceptions_init);
607#endif