2 * Copyright (C) 1995 Linus Torvalds
3 * Copyright (C) 2001, 2002 Andi Kleen, SuSE Labs.
4 * Copyright (C) 2008-2009, Red Hat Inc., Ingo Molnar
6 #include <linux/magic.h> /* STACK_END_MAGIC */
7 #include <linux/sched.h> /* test_thread_flag(), ... */
8 #include <linux/kdebug.h> /* oops_begin/end, ... */
9 #include <linux/module.h> /* search_exception_table */
10 #include <linux/bootmem.h> /* max_low_pfn */
11 #include <linux/kprobes.h> /* __kprobes, ... */
12 #include <linux/mmiotrace.h> /* kmmio_handler, ... */
13 #include <linux/perf_counter.h> /* perf_swcounter_event */
15 #include <asm/traps.h> /* dotraplinkage, ... */
16 #include <asm/pgalloc.h> /* pgd_*(), ... */
17 #include <asm/kmemcheck.h> /* kmemcheck_*(), ... */
20 * Page fault error code bits:
22 * bit 0 == 0: no page found 1: protection fault
23 * bit 1 == 0: read access 1: write access
24 * bit 2 == 0: kernel-mode access 1: user-mode access
25 * bit 3 == 1: use of reserved bit detected
26 * bit 4 == 1: fault was an instruction fetch
28 enum x86_pf_error_code
{
38 * Returns 0 if mmiotrace is disabled, or if the fault is not
39 * handled by mmiotrace:
41 static inline int __kprobes
42 kmmio_fault(struct pt_regs
*regs
, unsigned long addr
)
44 if (unlikely(is_kmmio_active()))
45 if (kmmio_handler(regs
, addr
) == 1)
50 static inline int __kprobes
notify_page_fault(struct pt_regs
*regs
)
54 /* kprobe_running() needs smp_processor_id() */
55 if (kprobes_built_in() && !user_mode_vm(regs
)) {
57 if (kprobe_running() && kprobe_fault_handler(regs
, 14))
70 * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch.
71 * Check that here and ignore it.
75 * Sometimes the CPU reports invalid exceptions on prefetch.
76 * Check that here and ignore it.
78 * Opcode checker based on code by Richard Brunner.
81 check_prefetch_opcode(struct pt_regs
*regs
, unsigned char *instr
,
82 unsigned char opcode
, int *prefetch
)
84 unsigned char instr_hi
= opcode
& 0xf0;
85 unsigned char instr_lo
= opcode
& 0x0f;
91 * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes.
92 * In X86_64 long mode, the CPU will signal invalid
93 * opcode if some of these prefixes are present so
94 * X86_64 will never get here anyway
96 return ((instr_lo
& 7) == 0x6);
100 * In AMD64 long mode 0x40..0x4F are valid REX prefixes
101 * Need to figure out under what instruction mode the
102 * instruction was issued. Could check the LDT for lm,
103 * but for now it's good enough to assume that long
104 * mode only uses well known segments or kernel.
106 return (!user_mode(regs
)) || (regs
->cs
== __USER_CS
);
109 /* 0x64 thru 0x67 are valid prefixes in all modes. */
110 return (instr_lo
& 0xC) == 0x4;
112 /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */
113 return !instr_lo
|| (instr_lo
>>1) == 1;
115 /* Prefetch instruction is 0x0F0D or 0x0F18 */
116 if (probe_kernel_address(instr
, opcode
))
119 *prefetch
= (instr_lo
== 0xF) &&
120 (opcode
== 0x0D || opcode
== 0x18);
128 is_prefetch(struct pt_regs
*regs
, unsigned long error_code
, unsigned long addr
)
130 unsigned char *max_instr
;
131 unsigned char *instr
;
135 * If it was a exec (instruction fetch) fault on NX page, then
136 * do not ignore the fault:
138 if (error_code
& PF_INSTR
)
141 instr
= (void *)convert_ip_to_linear(current
, regs
);
142 max_instr
= instr
+ 15;
144 if (user_mode(regs
) && instr
>= (unsigned char *)TASK_SIZE
)
147 while (instr
< max_instr
) {
148 unsigned char opcode
;
150 if (probe_kernel_address(instr
, opcode
))
155 if (!check_prefetch_opcode(regs
, instr
, opcode
, &prefetch
))
162 force_sig_info_fault(int si_signo
, int si_code
, unsigned long address
,
163 struct task_struct
*tsk
)
167 info
.si_signo
= si_signo
;
169 info
.si_code
= si_code
;
170 info
.si_addr
= (void __user
*)address
;
172 force_sig_info(si_signo
, &info
, tsk
);
175 DEFINE_SPINLOCK(pgd_lock
);
179 static inline pmd_t
*vmalloc_sync_one(pgd_t
*pgd
, unsigned long address
)
181 unsigned index
= pgd_index(address
);
187 pgd_k
= init_mm
.pgd
+ index
;
189 if (!pgd_present(*pgd_k
))
193 * set_pgd(pgd, *pgd_k); here would be useless on PAE
194 * and redundant with the set_pmd() on non-PAE. As would
197 pud
= pud_offset(pgd
, address
);
198 pud_k
= pud_offset(pgd_k
, address
);
199 if (!pud_present(*pud_k
))
202 pmd
= pmd_offset(pud
, address
);
203 pmd_k
= pmd_offset(pud_k
, address
);
204 if (!pmd_present(*pmd_k
))
207 if (!pmd_present(*pmd
))
208 set_pmd(pmd
, *pmd_k
);
210 BUG_ON(pmd_page(*pmd
) != pmd_page(*pmd_k
));
215 void vmalloc_sync_all(void)
217 unsigned long address
;
219 if (SHARED_KERNEL_PMD
)
222 for (address
= VMALLOC_START
& PMD_MASK
;
223 address
>= TASK_SIZE
&& address
< FIXADDR_TOP
;
224 address
+= PMD_SIZE
) {
229 spin_lock_irqsave(&pgd_lock
, flags
);
230 list_for_each_entry(page
, &pgd_list
, lru
) {
231 if (!vmalloc_sync_one(page_address(page
), address
))
234 spin_unlock_irqrestore(&pgd_lock
, flags
);
241 * Handle a fault on the vmalloc or module mapping area
243 static noinline __kprobes
int vmalloc_fault(unsigned long address
)
245 unsigned long pgd_paddr
;
249 /* Make sure we are in vmalloc area: */
250 if (!(address
>= VMALLOC_START
&& address
< VMALLOC_END
))
254 * Synchronize this task's top level page-table
255 * with the 'reference' page table.
257 * Do _not_ use "current" here. We might be inside
258 * an interrupt in the middle of a task switch..
260 pgd_paddr
= read_cr3();
261 pmd_k
= vmalloc_sync_one(__va(pgd_paddr
), address
);
265 pte_k
= pte_offset_kernel(pmd_k
, address
);
266 if (!pte_present(*pte_k
))
273 * Did it hit the DOS screen memory VA from vm86 mode?
276 check_v8086_mode(struct pt_regs
*regs
, unsigned long address
,
277 struct task_struct
*tsk
)
281 if (!v8086_mode(regs
))
284 bit
= (address
- 0xA0000) >> PAGE_SHIFT
;
286 tsk
->thread
.screen_bitmap
|= 1 << bit
;
289 static void dump_pagetable(unsigned long address
)
291 __typeof__(pte_val(__pte(0))) page
;
294 page
= ((__typeof__(page
) *) __va(page
))[address
>> PGDIR_SHIFT
];
296 #ifdef CONFIG_X86_PAE
297 printk("*pdpt = %016Lx ", page
);
298 if ((page
>> PAGE_SHIFT
) < max_low_pfn
299 && page
& _PAGE_PRESENT
) {
301 page
= ((__typeof__(page
) *) __va(page
))[(address
>> PMD_SHIFT
)
302 & (PTRS_PER_PMD
- 1)];
303 printk(KERN_CONT
"*pde = %016Lx ", page
);
307 printk("*pde = %08lx ", page
);
311 * We must not directly access the pte in the highpte
312 * case if the page table is located in highmem.
313 * And let's rather not kmap-atomic the pte, just in case
314 * it's allocated already:
316 if ((page
>> PAGE_SHIFT
) < max_low_pfn
317 && (page
& _PAGE_PRESENT
)
318 && !(page
& _PAGE_PSE
)) {
321 page
= ((__typeof__(page
) *) __va(page
))[(address
>> PAGE_SHIFT
)
322 & (PTRS_PER_PTE
- 1)];
323 printk("*pte = %0*Lx ", sizeof(page
)*2, (u64
)page
);
329 #else /* CONFIG_X86_64: */
331 void vmalloc_sync_all(void)
333 unsigned long address
;
335 for (address
= VMALLOC_START
& PGDIR_MASK
; address
<= VMALLOC_END
;
336 address
+= PGDIR_SIZE
) {
338 const pgd_t
*pgd_ref
= pgd_offset_k(address
);
342 if (pgd_none(*pgd_ref
))
345 spin_lock_irqsave(&pgd_lock
, flags
);
346 list_for_each_entry(page
, &pgd_list
, lru
) {
348 pgd
= (pgd_t
*)page_address(page
) + pgd_index(address
);
350 set_pgd(pgd
, *pgd_ref
);
352 BUG_ON(pgd_page_vaddr(*pgd
) != pgd_page_vaddr(*pgd_ref
));
354 spin_unlock_irqrestore(&pgd_lock
, flags
);
361 * Handle a fault on the vmalloc area
363 * This assumes no large pages in there.
365 static noinline __kprobes
int vmalloc_fault(unsigned long address
)
367 pgd_t
*pgd
, *pgd_ref
;
368 pud_t
*pud
, *pud_ref
;
369 pmd_t
*pmd
, *pmd_ref
;
370 pte_t
*pte
, *pte_ref
;
372 /* Make sure we are in vmalloc area: */
373 if (!(address
>= VMALLOC_START
&& address
< VMALLOC_END
))
377 * Copy kernel mappings over when needed. This can also
378 * happen within a race in page table update. In the later
381 pgd
= pgd_offset(current
->active_mm
, address
);
382 pgd_ref
= pgd_offset_k(address
);
383 if (pgd_none(*pgd_ref
))
387 set_pgd(pgd
, *pgd_ref
);
389 BUG_ON(pgd_page_vaddr(*pgd
) != pgd_page_vaddr(*pgd_ref
));
392 * Below here mismatches are bugs because these lower tables
396 pud
= pud_offset(pgd
, address
);
397 pud_ref
= pud_offset(pgd_ref
, address
);
398 if (pud_none(*pud_ref
))
401 if (pud_none(*pud
) || pud_page_vaddr(*pud
) != pud_page_vaddr(*pud_ref
))
404 pmd
= pmd_offset(pud
, address
);
405 pmd_ref
= pmd_offset(pud_ref
, address
);
406 if (pmd_none(*pmd_ref
))
409 if (pmd_none(*pmd
) || pmd_page(*pmd
) != pmd_page(*pmd_ref
))
412 pte_ref
= pte_offset_kernel(pmd_ref
, address
);
413 if (!pte_present(*pte_ref
))
416 pte
= pte_offset_kernel(pmd
, address
);
419 * Don't use pte_page here, because the mappings can point
420 * outside mem_map, and the NUMA hash lookup cannot handle
423 if (!pte_present(*pte
) || pte_pfn(*pte
) != pte_pfn(*pte_ref
))
429 static const char errata93_warning
[] =
431 "******* Your BIOS seems to not contain a fix for K8 errata #93\n"
432 "******* Working around it, but it may cause SEGVs or burn power.\n"
433 "******* Please consider a BIOS update.\n"
434 "******* Disabling USB legacy in the BIOS may also help.\n";
437 * No vm86 mode in 64-bit mode:
440 check_v8086_mode(struct pt_regs
*regs
, unsigned long address
,
441 struct task_struct
*tsk
)
445 static int bad_address(void *p
)
449 return probe_kernel_address((unsigned long *)p
, dummy
);
452 static void dump_pagetable(unsigned long address
)
459 pgd
= (pgd_t
*)read_cr3();
461 pgd
= __va((unsigned long)pgd
& PHYSICAL_PAGE_MASK
);
463 pgd
+= pgd_index(address
);
464 if (bad_address(pgd
))
467 printk("PGD %lx ", pgd_val(*pgd
));
469 if (!pgd_present(*pgd
))
472 pud
= pud_offset(pgd
, address
);
473 if (bad_address(pud
))
476 printk("PUD %lx ", pud_val(*pud
));
477 if (!pud_present(*pud
) || pud_large(*pud
))
480 pmd
= pmd_offset(pud
, address
);
481 if (bad_address(pmd
))
484 printk("PMD %lx ", pmd_val(*pmd
));
485 if (!pmd_present(*pmd
) || pmd_large(*pmd
))
488 pte
= pte_offset_kernel(pmd
, address
);
489 if (bad_address(pte
))
492 printk("PTE %lx", pte_val(*pte
));
500 #endif /* CONFIG_X86_64 */
503 * Workaround for K8 erratum #93 & buggy BIOS.
505 * BIOS SMM functions are required to use a specific workaround
506 * to avoid corruption of the 64bit RIP register on C stepping K8.
508 * A lot of BIOS that didn't get tested properly miss this.
510 * The OS sees this as a page fault with the upper 32bits of RIP cleared.
511 * Try to work around it here.
513 * Note we only handle faults in kernel here.
514 * Does nothing on 32-bit.
516 static int is_errata93(struct pt_regs
*regs
, unsigned long address
)
519 if (address
!= regs
->ip
)
522 if ((address
>> 32) != 0)
525 address
|= 0xffffffffUL
<< 32;
526 if ((address
>= (u64
)_stext
&& address
<= (u64
)_etext
) ||
527 (address
>= MODULES_VADDR
&& address
<= MODULES_END
)) {
528 printk_once(errata93_warning
);
537 * Work around K8 erratum #100 K8 in compat mode occasionally jumps
538 * to illegal addresses >4GB.
540 * We catch this in the page fault handler because these addresses
541 * are not reachable. Just detect this case and return. Any code
542 * segment in LDT is compatibility mode.
544 static int is_errata100(struct pt_regs
*regs
, unsigned long address
)
547 if ((regs
->cs
== __USER32_CS
|| (regs
->cs
& (1<<2))) && (address
>> 32))
553 static int is_f00f_bug(struct pt_regs
*regs
, unsigned long address
)
555 #ifdef CONFIG_X86_F00F_BUG
559 * Pentium F0 0F C7 C8 bug workaround:
561 if (boot_cpu_data
.f00f_bug
) {
562 nr
= (address
- idt_descr
.address
) >> 3;
565 do_invalid_op(regs
, 0);
573 static const char nx_warning
[] = KERN_CRIT
574 "kernel tried to execute NX-protected page - exploit attempt? (uid: %d)\n";
577 show_fault_oops(struct pt_regs
*regs
, unsigned long error_code
,
578 unsigned long address
)
580 if (!oops_may_print())
583 if (error_code
& PF_INSTR
) {
586 pte_t
*pte
= lookup_address(address
, &level
);
588 if (pte
&& pte_present(*pte
) && !pte_exec(*pte
))
589 printk(nx_warning
, current_uid());
592 printk(KERN_ALERT
"BUG: unable to handle kernel ");
593 if (address
< PAGE_SIZE
)
594 printk(KERN_CONT
"NULL pointer dereference");
596 printk(KERN_CONT
"paging request");
598 printk(KERN_CONT
" at %p\n", (void *) address
);
599 printk(KERN_ALERT
"IP:");
600 printk_address(regs
->ip
, 1);
602 dump_pagetable(address
);
606 pgtable_bad(struct pt_regs
*regs
, unsigned long error_code
,
607 unsigned long address
)
609 struct task_struct
*tsk
;
613 flags
= oops_begin();
617 printk(KERN_ALERT
"%s: Corrupted page table at address %lx\n",
619 dump_pagetable(address
);
621 tsk
->thread
.cr2
= address
;
622 tsk
->thread
.trap_no
= 14;
623 tsk
->thread
.error_code
= error_code
;
625 if (__die("Bad pagetable", regs
, error_code
))
628 oops_end(flags
, regs
, sig
);
632 no_context(struct pt_regs
*regs
, unsigned long error_code
,
633 unsigned long address
)
635 struct task_struct
*tsk
= current
;
636 unsigned long *stackend
;
640 /* Are we prepared to handle this kernel fault? */
641 if (fixup_exception(regs
))
647 * Valid to do another page fault here, because if this fault
648 * had been triggered by is_prefetch fixup_exception would have
653 * Hall of shame of CPU/BIOS bugs.
655 if (is_prefetch(regs
, error_code
, address
))
658 if (is_errata93(regs
, address
))
662 * Oops. The kernel tried to access some bad page. We'll have to
663 * terminate things with extreme prejudice:
665 flags
= oops_begin();
667 show_fault_oops(regs
, error_code
, address
);
669 stackend
= end_of_stack(tsk
);
670 if (*stackend
!= STACK_END_MAGIC
)
671 printk(KERN_ALERT
"Thread overran stack, or stack corrupted\n");
673 tsk
->thread
.cr2
= address
;
674 tsk
->thread
.trap_no
= 14;
675 tsk
->thread
.error_code
= error_code
;
678 if (__die("Oops", regs
, error_code
))
681 /* Executive summary in case the body of the oops scrolled away */
682 printk(KERN_EMERG
"CR2: %016lx\n", address
);
684 oops_end(flags
, regs
, sig
);
688 * Print out info about fatal segfaults, if the show_unhandled_signals
692 show_signal_msg(struct pt_regs
*regs
, unsigned long error_code
,
693 unsigned long address
, struct task_struct
*tsk
)
695 if (!unhandled_signal(tsk
, SIGSEGV
))
698 if (!printk_ratelimit())
701 printk("%s%s[%d]: segfault at %lx ip %p sp %p error %lx",
702 task_pid_nr(tsk
) > 1 ? KERN_INFO
: KERN_EMERG
,
703 tsk
->comm
, task_pid_nr(tsk
), address
,
704 (void *)regs
->ip
, (void *)regs
->sp
, error_code
);
706 print_vma_addr(KERN_CONT
" in ", regs
->ip
);
708 printk(KERN_CONT
"\n");
712 __bad_area_nosemaphore(struct pt_regs
*regs
, unsigned long error_code
,
713 unsigned long address
, int si_code
)
715 struct task_struct
*tsk
= current
;
717 /* User mode accesses just cause a SIGSEGV */
718 if (error_code
& PF_USER
) {
720 * It's possible to have interrupts off here:
725 * Valid to do another page fault here because this one came
728 if (is_prefetch(regs
, error_code
, address
))
731 if (is_errata100(regs
, address
))
734 if (unlikely(show_unhandled_signals
))
735 show_signal_msg(regs
, error_code
, address
, tsk
);
737 /* Kernel addresses are always protection faults: */
738 tsk
->thread
.cr2
= address
;
739 tsk
->thread
.error_code
= error_code
| (address
>= TASK_SIZE
);
740 tsk
->thread
.trap_no
= 14;
742 force_sig_info_fault(SIGSEGV
, si_code
, address
, tsk
);
747 if (is_f00f_bug(regs
, address
))
750 no_context(regs
, error_code
, address
);
754 bad_area_nosemaphore(struct pt_regs
*regs
, unsigned long error_code
,
755 unsigned long address
)
757 __bad_area_nosemaphore(regs
, error_code
, address
, SEGV_MAPERR
);
761 __bad_area(struct pt_regs
*regs
, unsigned long error_code
,
762 unsigned long address
, int si_code
)
764 struct mm_struct
*mm
= current
->mm
;
767 * Something tried to access memory that isn't in our memory map..
768 * Fix it, but check if it's kernel or user first..
770 up_read(&mm
->mmap_sem
);
772 __bad_area_nosemaphore(regs
, error_code
, address
, si_code
);
776 bad_area(struct pt_regs
*regs
, unsigned long error_code
, unsigned long address
)
778 __bad_area(regs
, error_code
, address
, SEGV_MAPERR
);
782 bad_area_access_error(struct pt_regs
*regs
, unsigned long error_code
,
783 unsigned long address
)
785 __bad_area(regs
, error_code
, address
, SEGV_ACCERR
);
788 /* TODO: fixup for "mm-invoke-oom-killer-from-page-fault.patch" */
790 out_of_memory(struct pt_regs
*regs
, unsigned long error_code
,
791 unsigned long address
)
794 * We ran out of memory, call the OOM killer, and return the userspace
795 * (which will retry the fault, or kill us if we got oom-killed):
797 up_read(¤t
->mm
->mmap_sem
);
799 pagefault_out_of_memory();
803 do_sigbus(struct pt_regs
*regs
, unsigned long error_code
, unsigned long address
)
805 struct task_struct
*tsk
= current
;
806 struct mm_struct
*mm
= tsk
->mm
;
808 up_read(&mm
->mmap_sem
);
810 /* Kernel mode? Handle exceptions or die: */
811 if (!(error_code
& PF_USER
))
812 no_context(regs
, error_code
, address
);
814 /* User-space => ok to do another page fault: */
815 if (is_prefetch(regs
, error_code
, address
))
818 tsk
->thread
.cr2
= address
;
819 tsk
->thread
.error_code
= error_code
;
820 tsk
->thread
.trap_no
= 14;
822 force_sig_info_fault(SIGBUS
, BUS_ADRERR
, address
, tsk
);
826 mm_fault_error(struct pt_regs
*regs
, unsigned long error_code
,
827 unsigned long address
, unsigned int fault
)
829 if (fault
& VM_FAULT_OOM
) {
830 out_of_memory(regs
, error_code
, address
);
832 if (fault
& VM_FAULT_SIGBUS
)
833 do_sigbus(regs
, error_code
, address
);
839 static int spurious_fault_check(unsigned long error_code
, pte_t
*pte
)
841 if ((error_code
& PF_WRITE
) && !pte_write(*pte
))
844 if ((error_code
& PF_INSTR
) && !pte_exec(*pte
))
851 * Handle a spurious fault caused by a stale TLB entry.
853 * This allows us to lazily refresh the TLB when increasing the
854 * permissions of a kernel page (RO -> RW or NX -> X). Doing it
855 * eagerly is very expensive since that implies doing a full
856 * cross-processor TLB flush, even if no stale TLB entries exist
857 * on other processors.
859 * There are no security implications to leaving a stale TLB when
860 * increasing the permissions on a page.
862 static noinline __kprobes
int
863 spurious_fault(unsigned long error_code
, unsigned long address
)
871 /* Reserved-bit violation or user access to kernel space? */
872 if (error_code
& (PF_USER
| PF_RSVD
))
875 pgd
= init_mm
.pgd
+ pgd_index(address
);
876 if (!pgd_present(*pgd
))
879 pud
= pud_offset(pgd
, address
);
880 if (!pud_present(*pud
))
884 return spurious_fault_check(error_code
, (pte_t
*) pud
);
886 pmd
= pmd_offset(pud
, address
);
887 if (!pmd_present(*pmd
))
891 return spurious_fault_check(error_code
, (pte_t
*) pmd
);
893 pte
= pte_offset_kernel(pmd
, address
);
894 if (!pte_present(*pte
))
897 ret
= spurious_fault_check(error_code
, pte
);
902 * Make sure we have permissions in PMD.
903 * If not, then there's a bug in the page tables:
905 ret
= spurious_fault_check(error_code
, (pte_t
*) pmd
);
906 WARN_ONCE(!ret
, "PMD has incorrect permission bits\n");
911 int show_unhandled_signals
= 1;
914 access_error(unsigned long error_code
, int write
, struct vm_area_struct
*vma
)
917 /* write, present and write, not present: */
918 if (unlikely(!(vma
->vm_flags
& VM_WRITE
)))
924 if (unlikely(error_code
& PF_PROT
))
927 /* read, not present: */
928 if (unlikely(!(vma
->vm_flags
& (VM_READ
| VM_EXEC
| VM_WRITE
))))
934 static int fault_in_kernel_space(unsigned long address
)
936 return address
>= TASK_SIZE_MAX
;
940 * This routine handles page faults. It determines the address,
941 * and the problem, and then passes it off to one of the appropriate
944 dotraplinkage
void __kprobes
945 do_page_fault(struct pt_regs
*regs
, unsigned long error_code
)
947 struct vm_area_struct
*vma
;
948 struct task_struct
*tsk
;
949 unsigned long address
;
950 struct mm_struct
*mm
;
957 /* Get the faulting address: */
958 address
= read_cr2();
961 * Detect and handle instructions that would cause a page fault for
962 * both a tracked kernel page and a userspace page.
964 if (kmemcheck_active(regs
))
965 kmemcheck_hide(regs
);
966 prefetchw(&mm
->mmap_sem
);
968 if (unlikely(kmmio_fault(regs
, address
)))
972 * We fault-in kernel-space virtual memory on-demand. The
973 * 'reference' page table is init_mm.pgd.
975 * NOTE! We MUST NOT take any locks for this case. We may
976 * be in an interrupt or a critical region, and should
977 * only copy the information from the master page table,
980 * This verifies that the fault happens in kernel space
981 * (error_code & 4) == 0, and that the fault was not a
982 * protection error (error_code & 9) == 0.
984 if (unlikely(fault_in_kernel_space(address
))) {
985 if (!(error_code
& (PF_RSVD
| PF_USER
| PF_PROT
))) {
986 if (vmalloc_fault(address
) >= 0)
989 if (kmemcheck_fault(regs
, address
, error_code
))
993 /* Can handle a stale RO->RW TLB: */
994 if (spurious_fault(error_code
, address
))
997 /* kprobes don't want to hook the spurious faults: */
998 if (notify_page_fault(regs
))
1001 * Don't take the mm semaphore here. If we fixup a prefetch
1002 * fault we could otherwise deadlock:
1004 bad_area_nosemaphore(regs
, error_code
, address
);
1009 /* kprobes don't want to hook the spurious faults: */
1010 if (unlikely(notify_page_fault(regs
)))
1013 * It's safe to allow irq's after cr2 has been saved and the
1014 * vmalloc fault has been handled.
1016 * User-mode registers count as a user access even for any
1017 * potential system fault or CPU buglet:
1019 if (user_mode_vm(regs
)) {
1021 error_code
|= PF_USER
;
1023 if (regs
->flags
& X86_EFLAGS_IF
)
1027 if (unlikely(error_code
& PF_RSVD
))
1028 pgtable_bad(regs
, error_code
, address
);
1030 perf_swcounter_event(PERF_COUNT_SW_PAGE_FAULTS
, 1, 0, regs
, address
);
1033 * If we're in an interrupt, have no user context or are running
1034 * in an atomic region then we must not take the fault:
1036 if (unlikely(in_atomic() || !mm
)) {
1037 bad_area_nosemaphore(regs
, error_code
, address
);
1042 * When running in the kernel we expect faults to occur only to
1043 * addresses in user space. All other faults represent errors in
1044 * the kernel and should generate an OOPS. Unfortunately, in the
1045 * case of an erroneous fault occurring in a code path which already
1046 * holds mmap_sem we will deadlock attempting to validate the fault
1047 * against the address space. Luckily the kernel only validly
1048 * references user space from well defined areas of code, which are
1049 * listed in the exceptions table.
1051 * As the vast majority of faults will be valid we will only perform
1052 * the source reference check when there is a possibility of a
1053 * deadlock. Attempt to lock the address space, if we cannot we then
1054 * validate the source. If this is invalid we can skip the address
1055 * space check, thus avoiding the deadlock:
1057 if (unlikely(!down_read_trylock(&mm
->mmap_sem
))) {
1058 if ((error_code
& PF_USER
) == 0 &&
1059 !search_exception_tables(regs
->ip
)) {
1060 bad_area_nosemaphore(regs
, error_code
, address
);
1063 down_read(&mm
->mmap_sem
);
1066 * The above down_read_trylock() might have succeeded in
1067 * which case we'll have missed the might_sleep() from
1073 vma
= find_vma(mm
, address
);
1074 if (unlikely(!vma
)) {
1075 bad_area(regs
, error_code
, address
);
1078 if (likely(vma
->vm_start
<= address
))
1080 if (unlikely(!(vma
->vm_flags
& VM_GROWSDOWN
))) {
1081 bad_area(regs
, error_code
, address
);
1084 if (error_code
& PF_USER
) {
1086 * Accessing the stack below %sp is always a bug.
1087 * The large cushion allows instructions like enter
1088 * and pusha to work. ("enter $65535, $31" pushes
1089 * 32 pointers and then decrements %sp by 65535.)
1091 if (unlikely(address
+ 65536 + 32 * sizeof(unsigned long) < regs
->sp
)) {
1092 bad_area(regs
, error_code
, address
);
1096 if (unlikely(expand_stack(vma
, address
))) {
1097 bad_area(regs
, error_code
, address
);
1102 * Ok, we have a good vm_area for this memory access, so
1103 * we can handle it..
1106 write
= error_code
& PF_WRITE
;
1108 if (unlikely(access_error(error_code
, write
, vma
))) {
1109 bad_area_access_error(regs
, error_code
, address
);
1114 * If for any reason at all we couldn't handle the fault,
1115 * make sure we exit gracefully rather than endlessly redo
1118 fault
= handle_mm_fault(mm
, vma
, address
, write
? FAULT_FLAG_WRITE
: 0);
1120 if (unlikely(fault
& VM_FAULT_ERROR
)) {
1121 mm_fault_error(regs
, error_code
, address
, fault
);
1125 if (fault
& VM_FAULT_MAJOR
) {
1127 perf_swcounter_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ
, 1, 0,
1131 perf_swcounter_event(PERF_COUNT_SW_PAGE_FAULTS_MIN
, 1, 0,
1135 check_v8086_mode(regs
, address
, tsk
);
1137 up_read(&mm
->mmap_sem
);