Linux Audio

Check our new training course

Loading...
v6.13.7
  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
 30bool copy_from_kernel_nofault_allowed(const void *unsafe_src, size_t size)
 
 31{
 32	unsigned long addr = (unsigned long)unsafe_src;
 
 
 
 
 
 
 
 
 33
 34	return addr >= TASK_SIZE && ULONG_MAX - addr >= size;
 
 
 
 
 
 35}
 
 36
 37/*
 38 * This is useful to dump out the page tables associated with
 39 * 'addr' in mm 'mm'.
 40 */
 41void show_pte(const char *lvl, struct mm_struct *mm, unsigned long addr)
 42{
 43	pgd_t *pgd;
 44
 45	if (!mm)
 46		mm = &init_mm;
 47
 
 48	pgd = pgd_offset(mm, addr);
 49	printk("%s[%08lx] *pgd=%08llx", lvl, addr, (long long)pgd_val(*pgd));
 
 50
 51	do {
 52		p4d_t *p4d;
 53		pud_t *pud;
 54		pmd_t *pmd;
 55		pte_t *pte;
 56
 57		p4d = p4d_offset(pgd, addr);
 58		if (p4d_none(*p4d))
 59			break;
 60
 61		if (p4d_bad(*p4d)) {
 62			pr_cont("(bad)");
 63			break;
 64		}
 65
 66		pud = pud_offset(p4d, addr);
 67		if (PTRS_PER_PUD != 1)
 68			pr_cont(", *pud=%08llx", (long long)pud_val(*pud));
 69
 70		if (pud_none(*pud))
 71			break;
 72
 73		if (pud_bad(*pud)) {
 74			pr_cont("(bad)");
 75			break;
 76		}
 77
 78		pmd = pmd_offset(pud, addr);
 79		if (PTRS_PER_PMD != 1)
 80			pr_cont(", *pmd=%08llx", (long long)pmd_val(*pmd));
 81
 82		if (pmd_none(*pmd))
 83			break;
 84
 85		if (pmd_bad(*pmd)) {
 86			pr_cont("(bad)");
 87			break;
 88		}
 89
 90		/* We must not map this if we have highmem enabled */
 91		if (PageHighMem(pfn_to_page(pmd_val(*pmd) >> PAGE_SHIFT)))
 92			break;
 93
 94		pte = pte_offset_map(pmd, addr);
 95		if (!pte)
 96			break;
 97
 98		pr_cont(", *pte=%08llx", (long long)pte_val(*pte));
 99#ifndef CONFIG_ARM_LPAE
100		pr_cont(", *ppte=%08llx",
101		       (long long)pte_val(pte[PTE_HWTABLE_PTRS]));
102#endif
103		pte_unmap(pte);
104	} while(0);
105
106	pr_cont("\n");
107}
108#else					/* CONFIG_MMU */
109void show_pte(const char *lvl, struct mm_struct *mm, unsigned long addr)
110{ }
111#endif					/* CONFIG_MMU */
112
113static inline bool is_write_fault(unsigned int fsr)
114{
115	return (fsr & FSR_WRITE) && !(fsr & FSR_CM);
116}
117
118static inline bool is_translation_fault(unsigned int fsr)
119{
120	int fs = fsr_fs(fsr);
121#ifdef CONFIG_ARM_LPAE
122	if ((fs & FS_MMU_NOLL_MASK) == FS_TRANS_NOLL)
123		return true;
124#else
125	if (fs == FS_L1_TRANS || fs == FS_L2_TRANS)
126		return true;
127#endif
128	return false;
129}
130
131static void die_kernel_fault(const char *msg, struct mm_struct *mm,
132			     unsigned long addr, unsigned int fsr,
133			     struct pt_regs *regs)
134{
135	bust_spinlocks(1);
136	pr_alert("8<--- cut here ---\n");
137	pr_alert("Unable to handle kernel %s at virtual address %08lx when %s\n",
138		 msg, addr, fsr & FSR_LNX_PF ? "execute" :
139		 fsr & FSR_WRITE ? "write" : "read");
140
141	show_pte(KERN_ALERT, mm, addr);
142	die("Oops", regs, fsr);
143	bust_spinlocks(0);
144	make_task_dead(SIGKILL);
145}
146
147/*
148 * Oops.  The kernel tried to access some page that wasn't present.
149 */
150static void
151__do_kernel_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
152		  struct pt_regs *regs)
153{
154	const char *msg;
155	/*
156	 * Are we prepared to handle this kernel fault?
157	 */
158	if (fixup_exception(regs))
159		return;
160
161	/*
162	 * No handler, we'll have to terminate things with extreme prejudice.
163	 */
164	if (addr < PAGE_SIZE) {
165		msg = "NULL pointer dereference";
166	} else {
167		if (is_translation_fault(fsr) &&
168		    kfence_handle_page_fault(addr, is_write_fault(fsr), regs))
169			return;
170
171		msg = "paging request";
172	}
173
174	die_kernel_fault(msg, mm, addr, fsr, regs);
 
 
 
175}
176
177/*
178 * Something tried to access memory that isn't in our memory map..
179 * User mode accesses just cause a SIGSEGV
180 */
181static void
182__do_user_fault(unsigned long addr, unsigned int fsr, unsigned int sig,
183		int code, struct pt_regs *regs)
 
184{
185	struct task_struct *tsk = current;
186
187	if (addr > TASK_SIZE)
188		harden_branch_predictor();
189
190#ifdef CONFIG_DEBUG_USER
191	if (((user_debug & UDBG_SEGV) && (sig == SIGSEGV)) ||
192	    ((user_debug & UDBG_BUS)  && (sig == SIGBUS))) {
193		pr_err("8<--- cut here ---\n");
194		pr_err("%s: unhandled page fault (%d) at 0x%08lx, code 0x%03x\n",
195		       tsk->comm, sig, addr, fsr);
196		show_pte(KERN_ERR, tsk->mm, addr);
197		show_regs(regs);
198	}
199#endif
200#ifndef CONFIG_KUSER_HELPERS
201	if ((sig == SIGSEGV) && ((addr & PAGE_MASK) == 0xffff0000))
202		printk_ratelimited(KERN_DEBUG
203				   "%s: CONFIG_KUSER_HELPERS disabled at 0x%08lx\n",
204				   tsk->comm, addr);
205#endif
206
207	tsk->thread.address = addr;
208	tsk->thread.error_code = fsr;
209	tsk->thread.trap_no = 14;
210	force_sig_fault(sig, code, (void __user *)addr);
 
 
 
 
211}
212
213void do_bad_area(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
214{
215	struct task_struct *tsk = current;
216	struct mm_struct *mm = tsk->active_mm;
217
218	/*
219	 * If we are in kernel mode at this point, we
220	 * have no context to handle this fault with.
221	 */
222	if (user_mode(regs))
223		__do_user_fault(addr, fsr, SIGSEGV, SEGV_MAPERR, regs);
224	else
225		__do_kernel_fault(mm, addr, fsr, regs);
226}
227
228#ifdef CONFIG_MMU
229static inline bool is_permission_fault(unsigned int fsr)
 
 
 
 
 
 
 
 
230{
231	int fs = fsr_fs(fsr);
232#ifdef CONFIG_ARM_LPAE
233	if ((fs & FS_MMU_NOLL_MASK) == FS_PERM_NOLL)
234		return true;
235#else
236	if (fs == FS_L1_PERM || fs == FS_L2_PERM)
237		return true;
238#endif
239	return false;
240}
241
242#ifdef CONFIG_CPU_TTBR0_PAN
243static inline bool ttbr0_usermode_access_allowed(struct pt_regs *regs)
 
244{
245	struct svc_pt_regs *svcregs;
 
246
247	/* If we are in user mode: permission granted */
248	if (user_mode(regs))
249		return true;
 
 
 
250
251	/* uaccess state saved above pt_regs on SVC exception entry */
252	svcregs = to_svc_pt_regs(regs);
 
 
 
 
 
 
 
253
254	return !(svcregs->ttbcr & TTBCR_EPD0);
255}
256#else
257static inline bool ttbr0_usermode_access_allowed(struct pt_regs *regs)
258{
259	return true;
 
 
 
260}
261#endif
262
263static int __kprobes
264do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
265{
266	struct mm_struct *mm = current->mm;
267	struct vm_area_struct *vma;
268	int sig, code;
269	vm_fault_t fault;
270	unsigned int flags = FAULT_FLAG_DEFAULT;
271	unsigned long vm_flags = VM_ACCESS_FLAGS;
272
273	if (kprobe_page_fault(regs, fsr))
274		return 0;
275
 
 
276
277	/* Enable interrupts if they were enabled in the parent context. */
278	if (interrupts_enabled(regs))
279		local_irq_enable();
280
281	/*
282	 * If we're in an interrupt or have no user
283	 * context, we must not take the fault..
284	 */
285	if (faulthandler_disabled() || !mm)
286		goto no_context;
287
288	if (user_mode(regs))
289		flags |= FAULT_FLAG_USER;
290
291	if (is_write_fault(fsr)) {
292		flags |= FAULT_FLAG_WRITE;
293		vm_flags = VM_WRITE;
294	}
295
296	if (fsr & FSR_LNX_PF) {
297		vm_flags = VM_EXEC;
298
299		if (is_permission_fault(fsr) && !user_mode(regs))
300			die_kernel_fault("execution of memory",
301					 mm, addr, fsr, regs);
302	}
303
304	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
305
306	/*
307	 * Privileged access aborts with CONFIG_CPU_TTBR0_PAN enabled are
308	 * routed via the translation fault mechanism. Check whether uaccess
309	 * is disabled while in kernel mode.
310	 */
311	if (!ttbr0_usermode_access_allowed(regs))
312		goto no_context;
313
314	if (!(flags & FAULT_FLAG_USER))
315		goto lock_mmap;
316
317	vma = lock_vma_under_rcu(mm, addr);
318	if (!vma)
319		goto lock_mmap;
320
321	if (!(vma->vm_flags & vm_flags)) {
322		vma_end_read(vma);
323		count_vm_vma_lock_event(VMA_LOCK_SUCCESS);
324		fault = 0;
325		code = SEGV_ACCERR;
326		goto bad_area;
327	}
328	fault = handle_mm_fault(vma, addr, flags | FAULT_FLAG_VMA_LOCK, regs);
329	if (!(fault & (VM_FAULT_RETRY | VM_FAULT_COMPLETED)))
330		vma_end_read(vma);
331
332	if (!(fault & VM_FAULT_RETRY)) {
333		count_vm_vma_lock_event(VMA_LOCK_SUCCESS);
334		goto done;
335	}
336	count_vm_vma_lock_event(VMA_LOCK_RETRY);
337	if (fault & VM_FAULT_MAJOR)
338		flags |= FAULT_FLAG_TRIED;
339
340	/* Quick path to respond to signals */
341	if (fault_signal_pending(fault, regs)) {
342		if (!user_mode(regs))
343			goto no_context;
344		return 0;
345	}
346lock_mmap:
347
348retry:
349	vma = lock_mm_and_find_vma(mm, addr, regs);
350	if (unlikely(!vma)) {
351		fault = 0;
352		code = SEGV_MAPERR;
353		goto bad_area;
354	}
355
356	/*
357	 * ok, we have a good vm_area for this memory access, check the
358	 * permissions on the VMA allow for the fault which occurred.
359	 */
360	if (!(vma->vm_flags & vm_flags)) {
361		mmap_read_unlock(mm);
362		fault = 0;
363		code = SEGV_ACCERR;
364		goto bad_area;
365	}
366
367	fault = handle_mm_fault(vma, addr & PAGE_MASK, flags, regs);
368
369	/* If we need to retry but a fatal signal is pending, handle the
370	 * signal first. We do not need to release the mmap_lock because
371	 * it would already be released in __lock_page_or_retry in
372	 * mm/filemap.c. */
373	if (fault_signal_pending(fault, regs)) {
374		if (!user_mode(regs))
375			goto no_context;
376		return 0;
377	}
378
379	/* The fault is fully completed (including releasing mmap lock) */
380	if (fault & VM_FAULT_COMPLETED)
381		return 0;
 
 
382
383	if (!(fault & VM_FAULT_ERROR)) {
 
 
 
 
 
 
 
 
 
 
384		if (fault & VM_FAULT_RETRY) {
385			flags |= FAULT_FLAG_TRIED;
 
 
386			goto retry;
387		}
388	}
389
390	mmap_read_unlock(mm);
391done:
392
393	/* Handle the "normal" case first */
394	if (likely(!(fault & VM_FAULT_ERROR)))
395		return 0;
396
397	code = SEGV_MAPERR;
398bad_area:
399	/*
400	 * If we are in kernel mode at this point, we
401	 * have no context to 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
410		 * got 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
419		 * successfully fix up this page fault.
420		 */
421		sig = SIGBUS;
422		code = BUS_ADRERR;
423	} else {
424		/*
425		 * Something tried to access memory that
426		 * isn't in our memory map..
427		 */
428		sig = SIGSEGV;
 
 
429	}
430
431	__do_user_fault(addr, fsr, sig, code, regs);
432	return 0;
433
434no_context:
435	__do_kernel_fault(mm, addr, fsr, regs);
436	return 0;
437}
438#else					/* CONFIG_MMU */
439static int
440do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
441{
442	return 0;
443}
444#endif					/* CONFIG_MMU */
445
446/*
447 * First Level Translation Fault Handler
448 *
449 * We enter here because the first level page table doesn't contain
450 * a valid entry for the address.
451 *
452 * If the address is in kernel space (>= TASK_SIZE), then we are
453 * probably faulting in the vmalloc() area.
454 *
455 * If the init_task's first level page tables contains the relevant
456 * entry, we copy the it to this task.  If not, we send the process
457 * a signal, fixup the exception, or oops the kernel.
458 *
459 * NOTE! We MUST NOT take any locks for this case. We may be in an
460 * interrupt or a critical region, and should only copy the information
461 * from the master page table, nothing more.
462 */
463#ifdef CONFIG_MMU
464static int __kprobes
465do_translation_fault(unsigned long addr, unsigned int fsr,
466		     struct pt_regs *regs)
467{
468	unsigned int index;
469	pgd_t *pgd, *pgd_k;
470	p4d_t *p4d, *p4d_k;
471	pud_t *pud, *pud_k;
472	pmd_t *pmd, *pmd_k;
473
474	if (addr < TASK_SIZE)
475		return do_page_fault(addr, fsr, regs);
476
477	if (user_mode(regs))
478		goto bad_area;
479
480	index = pgd_index(addr);
481
482	pgd = cpu_get_pgd() + index;
483	pgd_k = init_mm.pgd + index;
484
485	p4d = p4d_offset(pgd, addr);
486	p4d_k = p4d_offset(pgd_k, addr);
487
488	if (p4d_none(*p4d_k))
489		goto bad_area;
490	if (!p4d_present(*p4d))
491		set_p4d(p4d, *p4d_k);
492
493	pud = pud_offset(p4d, addr);
494	pud_k = pud_offset(p4d_k, addr);
495
496	if (pud_none(*pud_k))
497		goto bad_area;
498	if (!pud_present(*pud))
499		set_pud(pud, *pud_k);
500
501	pmd = pmd_offset(pud, addr);
502	pmd_k = pmd_offset(pud_k, addr);
503
504#ifdef CONFIG_ARM_LPAE
505	/*
506	 * Only one hardware entry per PMD with LPAE.
507	 */
508	index = 0;
509#else
510	/*
511	 * On ARM one Linux PGD entry contains two hardware entries (see page
512	 * tables layout in pgtable.h). We normally guarantee that we always
513	 * fill both L1 entries. But create_mapping() doesn't follow the rule.
514	 * It can create inidividual L1 entries, so here we have to call
515	 * pmd_none() check for the entry really corresponded to address, not
516	 * for the first of pair.
517	 */
518	index = (addr >> SECTION_SHIFT) & 1;
519#endif
520	if (pmd_none(pmd_k[index]))
521		goto bad_area;
522
523	copy_pmd(pmd, pmd_k);
524	return 0;
525
526bad_area:
527	do_bad_area(addr, fsr, regs);
528	return 0;
529}
530#else					/* CONFIG_MMU */
531static int
532do_translation_fault(unsigned long addr, unsigned int fsr,
533		     struct pt_regs *regs)
534{
535	return 0;
536}
537#endif					/* CONFIG_MMU */
538
539/*
540 * Some section permission faults need to be handled gracefully.
541 * They can happen due to a __{get,put}_user during an oops.
542 */
543#ifndef CONFIG_ARM_LPAE
544static int
545do_sect_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
546{
547	do_bad_area(addr, fsr, regs);
548	return 0;
549}
550#endif /* CONFIG_ARM_LPAE */
551
552/*
553 * This abort handler always returns "fault".
554 */
555static int
556do_bad(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
557{
558	return 1;
559}
560
561struct fsr_info {
562	int	(*fn)(unsigned long addr, unsigned int fsr, struct pt_regs *regs);
563	int	sig;
564	int	code;
565	const char *name;
566};
567
568/* FSR definition */
569#ifdef CONFIG_ARM_LPAE
570#include "fsr-3level.c"
571#else
572#include "fsr-2level.c"
573#endif
574
575void __init
576hook_fault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
577		int sig, int code, const char *name)
578{
579	if (nr < 0 || nr >= ARRAY_SIZE(fsr_info))
580		BUG();
581
582	fsr_info[nr].fn   = fn;
583	fsr_info[nr].sig  = sig;
584	fsr_info[nr].code = code;
585	fsr_info[nr].name = name;
586}
587
588/*
589 * Dispatch a data abort to the relevant handler.
590 */
591asmlinkage void
592do_DataAbort(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
593{
594	const struct fsr_info *inf = fsr_info + fsr_fs(fsr);
 
595
596	if (!inf->fn(addr, fsr & ~FSR_LNX_PF, regs))
597		return;
598
599	pr_alert("8<--- cut here ---\n");
600	pr_alert("Unhandled fault: %s (0x%03x) at 0x%08lx\n",
601		inf->name, fsr, addr);
602	show_pte(KERN_ALERT, current->mm, addr);
603
604	arm_notify_die("", regs, inf->sig, inf->code, (void __user *)addr,
605		       fsr, 0);
 
 
 
606}
607
608void __init
609hook_ifault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
610		 int sig, int code, const char *name)
611{
612	if (nr < 0 || nr >= ARRAY_SIZE(ifsr_info))
613		BUG();
614
615	ifsr_info[nr].fn   = fn;
616	ifsr_info[nr].sig  = sig;
617	ifsr_info[nr].code = code;
618	ifsr_info[nr].name = name;
619}
620
621asmlinkage void
622do_PrefetchAbort(unsigned long addr, unsigned int ifsr, struct pt_regs *regs)
623{
624	const struct fsr_info *inf = ifsr_info + fsr_fs(ifsr);
 
625
626	if (!inf->fn(addr, ifsr | FSR_LNX_PF, regs))
627		return;
628
629	pr_alert("8<--- cut here ---\n");
630	pr_alert("Unhandled prefetch abort: %s (0x%03x) at 0x%08lx\n",
631		inf->name, ifsr, addr);
632
633	arm_notify_die("", regs, inf->sig, inf->code, (void __user *)addr,
634		       ifsr, 0);
635}
636
637/*
638 * Abort handler to be used only during first unmasking of asynchronous aborts
639 * on the boot CPU. This makes sure that the machine will not die if the
640 * firmware/bootloader left an imprecise abort pending for us to trip over.
641 */
642static int __init early_abort_handler(unsigned long addr, unsigned int fsr,
643				      struct pt_regs *regs)
644{
645	pr_warn("Hit pending asynchronous external abort (FSR=0x%08x) during "
646		"first unmask, this is most likely caused by a "
647		"firmware/bootloader bug.\n", fsr);
648
649	return 0;
650}
651
652void __init early_abt_enable(void)
653{
654	fsr_info[FSR_FS_AEA].fn = early_abort_handler;
655	local_abt_enable();
656	fsr_info[FSR_FS_AEA].fn = do_bad;
657}
658
659#ifndef CONFIG_ARM_LPAE
660static int __init exceptions_init(void)
661{
662	if (cpu_architecture() >= CPU_ARCH_ARMv6) {
663		hook_fault_code(4, do_translation_fault, SIGSEGV, SEGV_MAPERR,
664				"I-cache maintenance fault");
665	}
666
667	if (cpu_architecture() >= CPU_ARCH_ARMv7) {
668		/*
669		 * TODO: Access flag faults introduced in ARMv6K.
670		 * Runtime check for 'K' extension is needed
671		 */
672		hook_fault_code(3, do_bad, SIGSEGV, SEGV_MAPERR,
673				"section access flag fault");
674		hook_fault_code(6, do_bad, SIGSEGV, SEGV_MAPERR,
675				"section access flag fault");
676	}
677
678	return 0;
679}
680
681arch_initcall(exceptions_init);
682#endif
v3.5.6
 
  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	int write = fsr & FSR_WRITE;
265	unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE |
266				(write ? FAULT_FLAG_WRITE : 0);
267
268	if (notify_page_fault(regs, fsr))
269		return 0;
270
271	tsk = current;
272	mm  = tsk->mm;
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 (in_atomic() || !mm)
283		goto no_context;
284
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
285	/*
286	 * As per x86, we may deadlock here.  However, since the kernel only
287	 * validly references user space from well defined areas of the code,
288	 * we can bug out early if this is from code which shouldn't.
289	 */
290	if (!down_read_trylock(&mm->mmap_sem)) {
291		if (!user_mode(regs) && !search_exception_tables(regs->ARM_pc))
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
292			goto no_context;
 
 
 
 
293retry:
294		down_read(&mm->mmap_sem);
295	} else {
296		/*
297		 * The above down_read_trylock() might have succeeded in
298		 * which case, we'll have missed the might_sleep() from
299		 * down_read()
300		 */
301		might_sleep();
302#ifdef CONFIG_DEBUG_VM
303		if (!user_mode(regs) &&
304		    !search_exception_tables(regs->ARM_pc))
305			goto no_context;
306#endif
 
 
 
307	}
308
309	fault = __do_page_fault(mm, addr, fsr, flags, tsk);
310
311	/* If we need to retry but a fatal signal is pending, handle the
312	 * signal first. We do not need to release the mmap_sem because
313	 * it would already be released in __lock_page_or_retry in
314	 * mm/filemap.c. */
315	if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
 
 
316		return 0;
 
317
318	/*
319	 * Major/minor page fault accounting is only done on the
320	 * initial attempt. If we go through a retry, it is extremely
321	 * likely that the page will be found in page cache at that point.
322	 */
323
324	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
325	if (!(fault & VM_FAULT_ERROR) && flags & FAULT_FLAG_ALLOW_RETRY) {
326		if (fault & VM_FAULT_MAJOR) {
327			tsk->maj_flt++;
328			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
329					regs, addr);
330		} else {
331			tsk->min_flt++;
332			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
333					regs, addr);
334		}
335		if (fault & VM_FAULT_RETRY) {
336			/* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
337			* of starvation. */
338			flags &= ~FAULT_FLAG_ALLOW_RETRY;
339			goto retry;
340		}
341	}
342
343	up_read(&mm->mmap_sem);
 
 
 
 
 
344
 
 
345	/*
346	 * Handle the "normal" case first - VM_FAULT_MAJOR / VM_FAULT_MINOR
 
347	 */
348	if (likely(!(fault & (VM_FAULT_ERROR | VM_FAULT_BADMAP | VM_FAULT_BADACCESS))))
349		return 0;
350
351	if (fault & VM_FAULT_OOM) {
352		/*
353		 * We ran out of memory, call the OOM killer, and return to
354		 * userspace (which will retry the fault, or kill us if we
355		 * got oom-killed)
356		 */
357		pagefault_out_of_memory();
358		return 0;
359	}
360
361	/*
362	 * If we are in kernel mode at this point, we
363	 * have no context to handle this fault with.
364	 */
365	if (!user_mode(regs))
366		goto no_context;
367
368	if (fault & VM_FAULT_SIGBUS) {
369		/*
370		 * We had some memory, but were unable to
371		 * successfully fix up this page fault.
372		 */
373		sig = SIGBUS;
374		code = BUS_ADRERR;
375	} else {
376		/*
377		 * Something tried to access memory that
378		 * isn't in our memory map..
379		 */
380		sig = SIGSEGV;
381		code = fault == VM_FAULT_BADACCESS ?
382			SEGV_ACCERR : SEGV_MAPERR;
383	}
384
385	__do_user_fault(tsk, addr, fsr, sig, code, regs);
386	return 0;
387
388no_context:
389	__do_kernel_fault(mm, addr, fsr, regs);
390	return 0;
391}
392#else					/* CONFIG_MMU */
393static int
394do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
395{
396	return 0;
397}
398#endif					/* CONFIG_MMU */
399
400/*
401 * First Level Translation Fault Handler
402 *
403 * We enter here because the first level page table doesn't contain
404 * a valid entry for the address.
405 *
406 * If the address is in kernel space (>= TASK_SIZE), then we are
407 * probably faulting in the vmalloc() area.
408 *
409 * If the init_task's first level page tables contains the relevant
410 * entry, we copy the it to this task.  If not, we send the process
411 * a signal, fixup the exception, or oops the kernel.
412 *
413 * NOTE! We MUST NOT take any locks for this case. We may be in an
414 * interrupt or a critical region, and should only copy the information
415 * from the master page table, nothing more.
416 */
417#ifdef CONFIG_MMU
418static int __kprobes
419do_translation_fault(unsigned long addr, unsigned int fsr,
420		     struct pt_regs *regs)
421{
422	unsigned int index;
423	pgd_t *pgd, *pgd_k;
 
424	pud_t *pud, *pud_k;
425	pmd_t *pmd, *pmd_k;
426
427	if (addr < TASK_SIZE)
428		return do_page_fault(addr, fsr, regs);
429
430	if (user_mode(regs))
431		goto bad_area;
432
433	index = pgd_index(addr);
434
435	pgd = cpu_get_pgd() + index;
436	pgd_k = init_mm.pgd + index;
437
438	if (pgd_none(*pgd_k))
 
 
 
439		goto bad_area;
440	if (!pgd_present(*pgd))
441		set_pgd(pgd, *pgd_k);
442
443	pud = pud_offset(pgd, addr);
444	pud_k = pud_offset(pgd_k, addr);
445
446	if (pud_none(*pud_k))
447		goto bad_area;
448	if (!pud_present(*pud))
449		set_pud(pud, *pud_k);
450
451	pmd = pmd_offset(pud, addr);
452	pmd_k = pmd_offset(pud_k, addr);
453
454#ifdef CONFIG_ARM_LPAE
455	/*
456	 * Only one hardware entry per PMD with LPAE.
457	 */
458	index = 0;
459#else
460	/*
461	 * On ARM one Linux PGD entry contains two hardware entries (see page
462	 * tables layout in pgtable.h). We normally guarantee that we always
463	 * fill both L1 entries. But create_mapping() doesn't follow the rule.
464	 * It can create inidividual L1 entries, so here we have to call
465	 * pmd_none() check for the entry really corresponded to address, not
466	 * for the first of pair.
467	 */
468	index = (addr >> SECTION_SHIFT) & 1;
469#endif
470	if (pmd_none(pmd_k[index]))
471		goto bad_area;
472
473	copy_pmd(pmd, pmd_k);
474	return 0;
475
476bad_area:
477	do_bad_area(addr, fsr, regs);
478	return 0;
479}
480#else					/* CONFIG_MMU */
481static int
482do_translation_fault(unsigned long addr, unsigned int fsr,
483		     struct pt_regs *regs)
484{
485	return 0;
486}
487#endif					/* CONFIG_MMU */
488
489/*
490 * Some section permission faults need to be handled gracefully.
491 * They can happen due to a __{get,put}_user during an oops.
492 */
 
493static int
494do_sect_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
495{
496	do_bad_area(addr, fsr, regs);
497	return 0;
498}
 
499
500/*
501 * This abort handler always returns "fault".
502 */
503static int
504do_bad(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
505{
506	return 1;
507}
508
509struct fsr_info {
510	int	(*fn)(unsigned long addr, unsigned int fsr, struct pt_regs *regs);
511	int	sig;
512	int	code;
513	const char *name;
514};
515
516/* FSR definition */
517#ifdef CONFIG_ARM_LPAE
518#include "fsr-3level.c"
519#else
520#include "fsr-2level.c"
521#endif
522
523void __init
524hook_fault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
525		int sig, int code, const char *name)
526{
527	if (nr < 0 || nr >= ARRAY_SIZE(fsr_info))
528		BUG();
529
530	fsr_info[nr].fn   = fn;
531	fsr_info[nr].sig  = sig;
532	fsr_info[nr].code = code;
533	fsr_info[nr].name = name;
534}
535
536/*
537 * Dispatch a data abort to the relevant handler.
538 */
539asmlinkage void __exception
540do_DataAbort(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
541{
542	const struct fsr_info *inf = fsr_info + fsr_fs(fsr);
543	struct siginfo info;
544
545	if (!inf->fn(addr, fsr & ~FSR_LNX_PF, regs))
546		return;
547
548	printk(KERN_ALERT "Unhandled fault: %s (0x%03x) at 0x%08lx\n",
 
549		inf->name, fsr, addr);
 
550
551	info.si_signo = inf->sig;
552	info.si_errno = 0;
553	info.si_code  = inf->code;
554	info.si_addr  = (void __user *)addr;
555	arm_notify_die("", regs, &info, fsr, 0);
556}
557
558void __init
559hook_ifault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
560		 int sig, int code, const char *name)
561{
562	if (nr < 0 || nr >= ARRAY_SIZE(ifsr_info))
563		BUG();
564
565	ifsr_info[nr].fn   = fn;
566	ifsr_info[nr].sig  = sig;
567	ifsr_info[nr].code = code;
568	ifsr_info[nr].name = name;
569}
570
571asmlinkage void __exception
572do_PrefetchAbort(unsigned long addr, unsigned int ifsr, struct pt_regs *regs)
573{
574	const struct fsr_info *inf = ifsr_info + fsr_fs(ifsr);
575	struct siginfo info;
576
577	if (!inf->fn(addr, ifsr | FSR_LNX_PF, regs))
578		return;
579
580	printk(KERN_ALERT "Unhandled prefetch abort: %s (0x%03x) at 0x%08lx\n",
 
581		inf->name, ifsr, addr);
582
583	info.si_signo = inf->sig;
584	info.si_errno = 0;
585	info.si_code  = inf->code;
586	info.si_addr  = (void __user *)addr;
587	arm_notify_die("", regs, &info, ifsr, 0);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
588}
589
590#ifndef CONFIG_ARM_LPAE
591static int __init exceptions_init(void)
592{
593	if (cpu_architecture() >= CPU_ARCH_ARMv6) {
594		hook_fault_code(4, do_translation_fault, SIGSEGV, SEGV_MAPERR,
595				"I-cache maintenance fault");
596	}
597
598	if (cpu_architecture() >= CPU_ARCH_ARMv7) {
599		/*
600		 * TODO: Access flag faults introduced in ARMv6K.
601		 * Runtime check for 'K' extension is needed
602		 */
603		hook_fault_code(3, do_bad, SIGSEGV, SEGV_MAPERR,
604				"section access flag fault");
605		hook_fault_code(6, do_bad, SIGSEGV, SEGV_MAPERR,
606				"section access flag fault");
607	}
608
609	return 0;
610}
611
612arch_initcall(exceptions_init);
613#endif