2 * Copyright (C) 1995 Linus Torvalds
3 * Copyright (C) 2001, 2002 Andi Kleen, SuSE Labs.
5 #include <linux/interrupt.h>
6 #include <linux/mmiotrace.h>
7 #include <linux/bootmem.h>
8 #include <linux/compiler.h>
9 #include <linux/highmem.h>
10 #include <linux/kprobes.h>
11 #include <linux/uaccess.h>
12 #include <linux/vmalloc.h>
13 #include <linux/vt_kern.h>
14 #include <linux/signal.h>
15 #include <linux/kernel.h>
16 #include <linux/ptrace.h>
17 #include <linux/string.h>
18 #include <linux/module.h>
19 #include <linux/kdebug.h>
20 #include <linux/errno.h>
21 #include <linux/magic.h>
22 #include <linux/sched.h>
23 #include <linux/types.h>
24 #include <linux/init.h>
25 #include <linux/mman.h>
26 #include <linux/tty.h>
27 #include <linux/smp.h>
30 #include <asm-generic/sections.h>
32 #include <asm/tlbflush.h>
33 #include <asm/pgalloc.h>
34 #include <asm/segment.h>
35 #include <asm/system.h>
36 #include <asm/proto.h>
37 #include <asm/traps.h>
41 * Page fault error code bits:
43 * bit 0 == 0: no page found 1: protection fault
44 * bit 1 == 0: read access 1: write access
45 * bit 2 == 0: kernel-mode access 1: user-mode access
46 * bit 3 == 1: use of reserved bit detected
47 * bit 4 == 1: fault was an instruction fetch
49 enum x86_pf_error_code
{
59 * (returns 0 if mmiotrace is disabled)
61 static inline int kmmio_fault(struct pt_regs
*regs
, unsigned long addr
)
63 if (unlikely(is_kmmio_active()))
64 if (kmmio_handler(regs
, addr
) == 1)
69 static inline int notify_page_fault(struct pt_regs
*regs
)
73 /* kprobe_running() needs smp_processor_id() */
74 if (kprobes_built_in() && !user_mode_vm(regs
)) {
76 if (kprobe_running() && kprobe_fault_handler(regs
, 14))
89 * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch.
90 * Check that here and ignore it.
94 * Sometimes the CPU reports invalid exceptions on prefetch.
95 * Check that here and ignore it.
97 * Opcode checker based on code by Richard Brunner.
100 check_prefetch_opcode(struct pt_regs
*regs
, unsigned char *instr
,
101 unsigned char opcode
, int *prefetch
)
103 unsigned char instr_hi
= opcode
& 0xf0;
104 unsigned char instr_lo
= opcode
& 0x0f;
110 * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes.
111 * In X86_64 long mode, the CPU will signal invalid
112 * opcode if some of these prefixes are present so
113 * X86_64 will never get here anyway
115 return ((instr_lo
& 7) == 0x6);
119 * In AMD64 long mode 0x40..0x4F are valid REX prefixes
120 * Need to figure out under what instruction mode the
121 * instruction was issued. Could check the LDT for lm,
122 * but for now it's good enough to assume that long
123 * mode only uses well known segments or kernel.
125 return (!user_mode(regs
)) || (regs
->cs
== __USER_CS
);
128 /* 0x64 thru 0x67 are valid prefixes in all modes. */
129 return (instr_lo
& 0xC) == 0x4;
131 /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */
132 return !instr_lo
|| (instr_lo
>>1) == 1;
134 /* Prefetch instruction is 0x0F0D or 0x0F18 */
135 if (probe_kernel_address(instr
, opcode
))
138 *prefetch
= (instr_lo
== 0xF) &&
139 (opcode
== 0x0D || opcode
== 0x18);
147 is_prefetch(struct pt_regs
*regs
, unsigned long error_code
, unsigned long addr
)
149 unsigned char *max_instr
;
150 unsigned char *instr
;
154 * If it was a exec (instruction fetch) fault on NX page, then
155 * do not ignore the fault:
157 if (error_code
& PF_INSTR
)
160 instr
= (void *)convert_ip_to_linear(current
, regs
);
161 max_instr
= instr
+ 15;
163 if (user_mode(regs
) && instr
>= (unsigned char *)TASK_SIZE
)
166 while (instr
< max_instr
) {
167 unsigned char opcode
;
169 if (probe_kernel_address(instr
, opcode
))
174 if (!check_prefetch_opcode(regs
, instr
, opcode
, &prefetch
))
181 force_sig_info_fault(int si_signo
, int si_code
, unsigned long address
,
182 struct task_struct
*tsk
)
186 info
.si_signo
= si_signo
;
188 info
.si_code
= si_code
;
189 info
.si_addr
= (void __user
*)address
;
191 force_sig_info(si_signo
, &info
, tsk
);
194 DEFINE_SPINLOCK(pgd_lock
);
198 static inline pmd_t
*vmalloc_sync_one(pgd_t
*pgd
, unsigned long address
)
200 unsigned index
= pgd_index(address
);
206 pgd_k
= init_mm
.pgd
+ index
;
208 if (!pgd_present(*pgd_k
))
212 * set_pgd(pgd, *pgd_k); here would be useless on PAE
213 * and redundant with the set_pmd() on non-PAE. As would
216 pud
= pud_offset(pgd
, address
);
217 pud_k
= pud_offset(pgd_k
, address
);
218 if (!pud_present(*pud_k
))
221 pmd
= pmd_offset(pud
, address
);
222 pmd_k
= pmd_offset(pud_k
, address
);
223 if (!pmd_present(*pmd_k
))
226 if (!pmd_present(*pmd
)) {
227 set_pmd(pmd
, *pmd_k
);
228 arch_flush_lazy_mmu_mode();
230 BUG_ON(pmd_page(*pmd
) != pmd_page(*pmd_k
));
236 void vmalloc_sync_all(void)
238 unsigned long address
;
240 if (SHARED_KERNEL_PMD
)
243 for (address
= VMALLOC_START
& PMD_MASK
;
244 address
>= TASK_SIZE
&& address
< FIXADDR_TOP
;
245 address
+= PMD_SIZE
) {
250 spin_lock_irqsave(&pgd_lock
, flags
);
251 list_for_each_entry(page
, &pgd_list
, lru
) {
252 if (!vmalloc_sync_one(page_address(page
), address
))
255 spin_unlock_irqrestore(&pgd_lock
, flags
);
262 * Handle a fault on the vmalloc or module mapping area
264 static noinline
int vmalloc_fault(unsigned long address
)
266 unsigned long pgd_paddr
;
270 /* Make sure we are in vmalloc area: */
271 if (!(address
>= VMALLOC_START
&& address
< VMALLOC_END
))
275 * Synchronize this task's top level page-table
276 * with the 'reference' page table.
278 * Do _not_ use "current" here. We might be inside
279 * an interrupt in the middle of a task switch..
281 pgd_paddr
= read_cr3();
282 pmd_k
= vmalloc_sync_one(__va(pgd_paddr
), address
);
286 pte_k
= pte_offset_kernel(pmd_k
, address
);
287 if (!pte_present(*pte_k
))
294 * Did it hit the DOS screen memory VA from vm86 mode?
297 check_v8086_mode(struct pt_regs
*regs
, unsigned long address
,
298 struct task_struct
*tsk
)
302 if (!v8086_mode(regs
))
305 bit
= (address
- 0xA0000) >> PAGE_SHIFT
;
307 tsk
->thread
.screen_bitmap
|= 1 << bit
;
310 static void dump_pagetable(unsigned long address
)
312 __typeof__(pte_val(__pte(0))) page
;
315 page
= ((__typeof__(page
) *) __va(page
))[address
>> PGDIR_SHIFT
];
317 #ifdef CONFIG_X86_PAE
318 printk("*pdpt = %016Lx ", page
);
319 if ((page
>> PAGE_SHIFT
) < max_low_pfn
320 && page
& _PAGE_PRESENT
) {
322 page
= ((__typeof__(page
) *) __va(page
))[(address
>> PMD_SHIFT
)
323 & (PTRS_PER_PMD
- 1)];
324 printk(KERN_CONT
"*pde = %016Lx ", page
);
328 printk("*pde = %08lx ", page
);
332 * We must not directly access the pte in the highpte
333 * case if the page table is located in highmem.
334 * And let's rather not kmap-atomic the pte, just in case
335 * it's allocated already:
337 if ((page
>> PAGE_SHIFT
) < max_low_pfn
338 && (page
& _PAGE_PRESENT
)
339 && !(page
& _PAGE_PSE
)) {
342 page
= ((__typeof__(page
) *) __va(page
))[(address
>> PAGE_SHIFT
)
343 & (PTRS_PER_PTE
- 1)];
344 printk("*pte = %0*Lx ", sizeof(page
)*2, (u64
)page
);
350 #else /* CONFIG_X86_64: */
352 void vmalloc_sync_all(void)
354 unsigned long address
;
356 for (address
= VMALLOC_START
& PGDIR_MASK
; address
<= VMALLOC_END
;
357 address
+= PGDIR_SIZE
) {
359 const pgd_t
*pgd_ref
= pgd_offset_k(address
);
363 if (pgd_none(*pgd_ref
))
366 spin_lock_irqsave(&pgd_lock
, flags
);
367 list_for_each_entry(page
, &pgd_list
, lru
) {
369 pgd
= (pgd_t
*)page_address(page
) + pgd_index(address
);
371 set_pgd(pgd
, *pgd_ref
);
373 BUG_ON(pgd_page_vaddr(*pgd
) != pgd_page_vaddr(*pgd_ref
));
375 spin_unlock_irqrestore(&pgd_lock
, flags
);
382 * Handle a fault on the vmalloc area
384 * This assumes no large pages in there.
386 static noinline
int vmalloc_fault(unsigned long address
)
388 pgd_t
*pgd
, *pgd_ref
;
389 pud_t
*pud
, *pud_ref
;
390 pmd_t
*pmd
, *pmd_ref
;
391 pte_t
*pte
, *pte_ref
;
393 /* Make sure we are in vmalloc area: */
394 if (!(address
>= VMALLOC_START
&& address
< VMALLOC_END
))
398 * Copy kernel mappings over when needed. This can also
399 * happen within a race in page table update. In the later
402 pgd
= pgd_offset(current
->active_mm
, address
);
403 pgd_ref
= pgd_offset_k(address
);
404 if (pgd_none(*pgd_ref
))
408 set_pgd(pgd
, *pgd_ref
);
410 BUG_ON(pgd_page_vaddr(*pgd
) != pgd_page_vaddr(*pgd_ref
));
413 * Below here mismatches are bugs because these lower tables
417 pud
= pud_offset(pgd
, address
);
418 pud_ref
= pud_offset(pgd_ref
, address
);
419 if (pud_none(*pud_ref
))
422 if (pud_none(*pud
) || pud_page_vaddr(*pud
) != pud_page_vaddr(*pud_ref
))
425 pmd
= pmd_offset(pud
, address
);
426 pmd_ref
= pmd_offset(pud_ref
, address
);
427 if (pmd_none(*pmd_ref
))
430 if (pmd_none(*pmd
) || pmd_page(*pmd
) != pmd_page(*pmd_ref
))
433 pte_ref
= pte_offset_kernel(pmd_ref
, address
);
434 if (!pte_present(*pte_ref
))
437 pte
= pte_offset_kernel(pmd
, address
);
440 * Don't use pte_page here, because the mappings can point
441 * outside mem_map, and the NUMA hash lookup cannot handle
444 if (!pte_present(*pte
) || pte_pfn(*pte
) != pte_pfn(*pte_ref
))
450 static const char errata93_warning
[] =
451 KERN_ERR
"******* Your BIOS seems to not contain a fix for K8 errata #93\n"
452 KERN_ERR
"******* Working around it, but it may cause SEGVs or burn power.\n"
453 KERN_ERR
"******* Please consider a BIOS update.\n"
454 KERN_ERR
"******* Disabling USB legacy in the BIOS may also help.\n";
457 * No vm86 mode in 64-bit mode:
460 check_v8086_mode(struct pt_regs
*regs
, unsigned long address
,
461 struct task_struct
*tsk
)
465 static int bad_address(void *p
)
469 return probe_kernel_address((unsigned long *)p
, dummy
);
472 static void dump_pagetable(unsigned long address
)
479 pgd
= (pgd_t
*)read_cr3();
481 pgd
= __va((unsigned long)pgd
& PHYSICAL_PAGE_MASK
);
483 pgd
+= pgd_index(address
);
484 if (bad_address(pgd
))
487 printk("PGD %lx ", pgd_val(*pgd
));
489 if (!pgd_present(*pgd
))
492 pud
= pud_offset(pgd
, address
);
493 if (bad_address(pud
))
496 printk("PUD %lx ", pud_val(*pud
));
497 if (!pud_present(*pud
) || pud_large(*pud
))
500 pmd
= pmd_offset(pud
, address
);
501 if (bad_address(pmd
))
504 printk("PMD %lx ", pmd_val(*pmd
));
505 if (!pmd_present(*pmd
) || pmd_large(*pmd
))
508 pte
= pte_offset_kernel(pmd
, address
);
509 if (bad_address(pte
))
512 printk("PTE %lx", pte_val(*pte
));
520 #endif /* CONFIG_X86_64 */
523 * Workaround for K8 erratum #93 & buggy BIOS.
525 * BIOS SMM functions are required to use a specific workaround
526 * to avoid corruption of the 64bit RIP register on C stepping K8.
528 * A lot of BIOS that didn't get tested properly miss this.
530 * The OS sees this as a page fault with the upper 32bits of RIP cleared.
531 * Try to work around it here.
533 * Note we only handle faults in kernel here.
534 * Does nothing on 32-bit.
536 static int is_errata93(struct pt_regs
*regs
, unsigned long address
)
541 if (address
!= regs
->ip
)
544 if ((address
>> 32) != 0)
547 address
|= 0xffffffffUL
<< 32;
548 if ((address
>= (u64
)_stext
&& address
<= (u64
)_etext
) ||
549 (address
>= MODULES_VADDR
&& address
<= MODULES_END
)) {
551 printk(errata93_warning
);
562 * Work around K8 erratum #100 K8 in compat mode occasionally jumps
563 * to illegal addresses >4GB.
565 * We catch this in the page fault handler because these addresses
566 * are not reachable. Just detect this case and return. Any code
567 * segment in LDT is compatibility mode.
569 static int is_errata100(struct pt_regs
*regs
, unsigned long address
)
572 if ((regs
->cs
== __USER32_CS
|| (regs
->cs
& (1<<2))) && (address
>> 32))
578 static int is_f00f_bug(struct pt_regs
*regs
, unsigned long address
)
580 #ifdef CONFIG_X86_F00F_BUG
584 * Pentium F0 0F C7 C8 bug workaround:
586 if (boot_cpu_data
.f00f_bug
) {
587 nr
= (address
- idt_descr
.address
) >> 3;
590 do_invalid_op(regs
, 0);
598 static const char nx_warning
[] = KERN_CRIT
599 "kernel tried to execute NX-protected page - exploit attempt? (uid: %d)\n";
602 show_fault_oops(struct pt_regs
*regs
, unsigned long error_code
,
603 unsigned long address
)
605 if (!oops_may_print())
608 if (error_code
& PF_INSTR
) {
611 pte_t
*pte
= lookup_address(address
, &level
);
613 if (pte
&& pte_present(*pte
) && !pte_exec(*pte
))
614 printk(nx_warning
, current_uid());
617 printk(KERN_ALERT
"BUG: unable to handle kernel ");
618 if (address
< PAGE_SIZE
)
619 printk(KERN_CONT
"NULL pointer dereference");
621 printk(KERN_CONT
"paging request");
623 printk(KERN_CONT
" at %p\n", (void *) address
);
624 printk(KERN_ALERT
"IP:");
625 printk_address(regs
->ip
, 1);
627 dump_pagetable(address
);
631 pgtable_bad(struct pt_regs
*regs
, unsigned long error_code
,
632 unsigned long address
)
634 struct task_struct
*tsk
;
638 flags
= oops_begin();
642 printk(KERN_ALERT
"%s: Corrupted page table at address %lx\n",
644 dump_pagetable(address
);
646 tsk
->thread
.cr2
= address
;
647 tsk
->thread
.trap_no
= 14;
648 tsk
->thread
.error_code
= error_code
;
650 if (__die("Bad pagetable", regs
, error_code
))
653 oops_end(flags
, regs
, sig
);
657 no_context(struct pt_regs
*regs
, unsigned long error_code
,
658 unsigned long address
)
660 struct task_struct
*tsk
= current
;
661 unsigned long *stackend
;
668 /* Are we prepared to handle this kernel fault? */
669 if (fixup_exception(regs
))
675 * Valid to do another page fault here, because if this fault
676 * had been triggered by is_prefetch fixup_exception would have
681 * Hall of shame of CPU/BIOS bugs.
683 if (is_prefetch(regs
, error_code
, address
))
686 if (is_errata93(regs
, address
))
690 * Oops. The kernel tried to access some bad page. We'll have to
691 * terminate things with extreme prejudice:
696 flags
= oops_begin();
699 show_fault_oops(regs
, error_code
, address
);
701 stackend
= end_of_stack(tsk
);
702 if (*stackend
!= STACK_END_MAGIC
)
703 printk(KERN_ALERT
"Thread overran stack, or stack corrupted\n");
705 tsk
->thread
.cr2
= address
;
706 tsk
->thread
.trap_no
= 14;
707 tsk
->thread
.error_code
= error_code
;
710 die("Oops", regs
, error_code
);
715 if (__die("Oops", regs
, error_code
))
718 /* Executive summary in case the body of the oops scrolled away */
719 printk(KERN_EMERG
"CR2: %016lx\n", address
);
721 oops_end(flags
, regs
, sig
);
726 * Print out info about fatal segfaults, if the show_unhandled_signals
730 show_signal_msg(struct pt_regs
*regs
, unsigned long error_code
,
731 unsigned long address
, struct task_struct
*tsk
)
733 if (!unhandled_signal(tsk
, SIGSEGV
))
736 if (!printk_ratelimit())
739 printk(KERN_CONT
"%s%s[%d]: segfault at %lx ip %p sp %p error %lx",
740 task_pid_nr(tsk
) > 1 ? KERN_INFO
: KERN_EMERG
,
741 tsk
->comm
, task_pid_nr(tsk
), address
,
742 (void *)regs
->ip
, (void *)regs
->sp
, error_code
);
744 print_vma_addr(KERN_CONT
" in ", regs
->ip
);
746 printk(KERN_CONT
"\n");
750 __bad_area_nosemaphore(struct pt_regs
*regs
, unsigned long error_code
,
751 unsigned long address
, int si_code
)
753 struct task_struct
*tsk
= current
;
755 /* User mode accesses just cause a SIGSEGV */
756 if (error_code
& PF_USER
) {
758 * It's possible to have interrupts off here:
763 * Valid to do another page fault here because this one came
766 if (is_prefetch(regs
, error_code
, address
))
769 if (is_errata100(regs
, address
))
772 if (unlikely(show_unhandled_signals
))
773 show_signal_msg(regs
, error_code
, address
, tsk
);
775 /* Kernel addresses are always protection faults: */
776 tsk
->thread
.cr2
= address
;
777 tsk
->thread
.error_code
= error_code
| (address
>= TASK_SIZE
);
778 tsk
->thread
.trap_no
= 14;
780 force_sig_info_fault(SIGSEGV
, si_code
, address
, tsk
);
785 if (is_f00f_bug(regs
, address
))
788 no_context(regs
, error_code
, address
);
792 bad_area_nosemaphore(struct pt_regs
*regs
, unsigned long error_code
,
793 unsigned long address
)
795 __bad_area_nosemaphore(regs
, error_code
, address
, SEGV_MAPERR
);
799 __bad_area(struct pt_regs
*regs
, unsigned long error_code
,
800 unsigned long address
, int si_code
)
802 struct mm_struct
*mm
= current
->mm
;
805 * Something tried to access memory that isn't in our memory map..
806 * Fix it, but check if it's kernel or user first..
808 up_read(&mm
->mmap_sem
);
810 __bad_area_nosemaphore(regs
, error_code
, address
, si_code
);
814 bad_area(struct pt_regs
*regs
, unsigned long error_code
, unsigned long address
)
816 __bad_area(regs
, error_code
, address
, SEGV_MAPERR
);
820 bad_area_access_error(struct pt_regs
*regs
, unsigned long error_code
,
821 unsigned long address
)
823 __bad_area(regs
, error_code
, address
, SEGV_ACCERR
);
826 /* TODO: fixup for "mm-invoke-oom-killer-from-page-fault.patch" */
828 out_of_memory(struct pt_regs
*regs
, unsigned long error_code
,
829 unsigned long address
)
832 * We ran out of memory, call the OOM killer, and return the userspace
833 * (which will retry the fault, or kill us if we got oom-killed):
835 up_read(¤t
->mm
->mmap_sem
);
837 pagefault_out_of_memory();
841 do_sigbus(struct pt_regs
*regs
, unsigned long error_code
, unsigned long address
)
843 struct task_struct
*tsk
= current
;
844 struct mm_struct
*mm
= tsk
->mm
;
846 up_read(&mm
->mmap_sem
);
848 /* Kernel mode? Handle exceptions or die: */
849 if (!(error_code
& PF_USER
))
850 no_context(regs
, error_code
, address
);
853 /* User space => ok to do another page fault: */
854 if (is_prefetch(regs
, error_code
, address
))
858 tsk
->thread
.cr2
= address
;
859 tsk
->thread
.error_code
= error_code
;
860 tsk
->thread
.trap_no
= 14;
862 force_sig_info_fault(SIGBUS
, BUS_ADRERR
, address
, tsk
);
866 mm_fault_error(struct pt_regs
*regs
, unsigned long error_code
,
867 unsigned long address
, unsigned int fault
)
869 if (fault
& VM_FAULT_OOM
) {
870 out_of_memory(regs
, error_code
, address
);
872 if (fault
& VM_FAULT_SIGBUS
)
873 do_sigbus(regs
, error_code
, address
);
879 static int spurious_fault_check(unsigned long error_code
, pte_t
*pte
)
881 if ((error_code
& PF_WRITE
) && !pte_write(*pte
))
884 if ((error_code
& PF_INSTR
) && !pte_exec(*pte
))
891 * Handle a spurious fault caused by a stale TLB entry.
893 * This allows us to lazily refresh the TLB when increasing the
894 * permissions of a kernel page (RO -> RW or NX -> X). Doing it
895 * eagerly is very expensive since that implies doing a full
896 * cross-processor TLB flush, even if no stale TLB entries exist
897 * on other processors.
899 * There are no security implications to leaving a stale TLB when
900 * increasing the permissions on a page.
903 spurious_fault(unsigned long error_code
, unsigned long address
)
911 /* Reserved-bit violation or user access to kernel space? */
912 if (error_code
& (PF_USER
| PF_RSVD
))
915 pgd
= init_mm
.pgd
+ pgd_index(address
);
916 if (!pgd_present(*pgd
))
919 pud
= pud_offset(pgd
, address
);
920 if (!pud_present(*pud
))
924 return spurious_fault_check(error_code
, (pte_t
*) pud
);
926 pmd
= pmd_offset(pud
, address
);
927 if (!pmd_present(*pmd
))
931 return spurious_fault_check(error_code
, (pte_t
*) pmd
);
933 pte
= pte_offset_kernel(pmd
, address
);
934 if (!pte_present(*pte
))
937 ret
= spurious_fault_check(error_code
, pte
);
942 * Make sure we have permissions in PMD.
943 * If not, then there's a bug in the page tables:
945 ret
= spurious_fault_check(error_code
, (pte_t
*) pmd
);
946 WARN_ONCE(!ret
, "PMD has incorrect permission bits\n");
951 int show_unhandled_signals
= 1;
954 access_error(unsigned long error_code
, int write
, struct vm_area_struct
*vma
)
957 /* write, present and write, not present: */
958 if (unlikely(!(vma
->vm_flags
& VM_WRITE
)))
964 if (unlikely(error_code
& PF_PROT
))
967 /* read, not present: */
968 if (unlikely(!(vma
->vm_flags
& (VM_READ
| VM_EXEC
| VM_WRITE
))))
974 static int fault_in_kernel_space(unsigned long address
)
977 return address
>= TASK_SIZE
;
979 return address
>= TASK_SIZE64
;
984 * This routine handles page faults. It determines the address,
985 * and the problem, and then passes it off to one of the appropriate
991 void __kprobes
do_page_fault(struct pt_regs
*regs
, unsigned long error_code
)
993 struct vm_area_struct
*vma
;
994 struct task_struct
*tsk
;
995 unsigned long address
;
996 struct mm_struct
*mm
;
1003 prefetchw(&mm
->mmap_sem
);
1005 /* Get the faulting address: */
1006 address
= read_cr2();
1008 if (unlikely(kmmio_fault(regs
, address
)))
1012 * We fault-in kernel-space virtual memory on-demand. The
1013 * 'reference' page table is init_mm.pgd.
1015 * NOTE! We MUST NOT take any locks for this case. We may
1016 * be in an interrupt or a critical region, and should
1017 * only copy the information from the master page table,
1020 * This verifies that the fault happens in kernel space
1021 * (error_code & 4) == 0, and that the fault was not a
1022 * protection error (error_code & 9) == 0.
1024 if (unlikely(fault_in_kernel_space(address
))) {
1025 if (!(error_code
& (PF_RSVD
|PF_USER
|PF_PROT
)) &&
1026 vmalloc_fault(address
) >= 0)
1029 /* Can handle a stale RO->RW TLB: */
1030 if (spurious_fault(error_code
, address
))
1033 /* kprobes don't want to hook the spurious faults: */
1034 if (notify_page_fault(regs
))
1037 * Don't take the mm semaphore here. If we fixup a prefetch
1038 * fault we could otherwise deadlock:
1040 bad_area_nosemaphore(regs
, error_code
, address
);
1045 /* kprobes don't want to hook the spurious faults: */
1046 if (unlikely(notify_page_fault(regs
)))
1049 * It's safe to allow irq's after cr2 has been saved and the
1050 * vmalloc fault has been handled.
1052 * User-mode registers count as a user access even for any
1053 * potential system fault or CPU buglet:
1055 if (user_mode_vm(regs
)) {
1057 error_code
|= PF_USER
;
1059 if (regs
->flags
& X86_EFLAGS_IF
)
1063 if (unlikely(error_code
& PF_RSVD
))
1064 pgtable_bad(regs
, error_code
, address
);
1067 * If we're in an interrupt, have no user context or are running
1068 * in an atomic region then we must not take the fault:
1070 if (unlikely(in_atomic() || !mm
)) {
1071 bad_area_nosemaphore(regs
, error_code
, address
);
1076 * When running in the kernel we expect faults to occur only to
1077 * addresses in user space. All other faults represent errors in
1078 * the kernel and should generate an OOPS. Unfortunately, in the
1079 * case of an erroneous fault occurring in a code path which already
1080 * holds mmap_sem we will deadlock attempting to validate the fault
1081 * against the address space. Luckily the kernel only validly
1082 * references user space from well defined areas of code, which are
1083 * listed in the exceptions table.
1085 * As the vast majority of faults will be valid we will only perform
1086 * the source reference check when there is a possibility of a
1087 * deadlock. Attempt to lock the address space, if we cannot we then
1088 * validate the source. If this is invalid we can skip the address
1089 * space check, thus avoiding the deadlock:
1091 if (unlikely(!down_read_trylock(&mm
->mmap_sem
))) {
1092 if ((error_code
& PF_USER
) == 0 &&
1093 !search_exception_tables(regs
->ip
)) {
1094 bad_area_nosemaphore(regs
, error_code
, address
);
1097 down_read(&mm
->mmap_sem
);
1100 * The above down_read_trylock() might have succeeded in
1101 * which case we'll have missed the might_sleep() from
1107 vma
= find_vma(mm
, address
);
1108 if (unlikely(!vma
)) {
1109 bad_area(regs
, error_code
, address
);
1112 if (likely(vma
->vm_start
<= address
))
1114 if (unlikely(!(vma
->vm_flags
& VM_GROWSDOWN
))) {
1115 bad_area(regs
, error_code
, address
);
1118 if (error_code
& PF_USER
) {
1120 * Accessing the stack below %sp is always a bug.
1121 * The large cushion allows instructions like enter
1122 * and pusha to work. ("enter $65535, $31" pushes
1123 * 32 pointers and then decrements %sp by 65535.)
1125 if (unlikely(address
+ 65536 + 32 * sizeof(unsigned long) < regs
->sp
)) {
1126 bad_area(regs
, error_code
, address
);
1130 if (unlikely(expand_stack(vma
, address
))) {
1131 bad_area(regs
, error_code
, address
);
1136 * Ok, we have a good vm_area for this memory access, so
1137 * we can handle it..
1140 write
= error_code
& PF_WRITE
;
1142 if (unlikely(access_error(error_code
, write
, vma
))) {
1143 bad_area_access_error(regs
, error_code
, address
);
1148 * If for any reason at all we couldn't handle the fault,
1149 * make sure we exit gracefully rather than endlessly redo
1152 fault
= handle_mm_fault(mm
, vma
, address
, write
);
1154 if (unlikely(fault
& VM_FAULT_ERROR
)) {
1155 mm_fault_error(regs
, error_code
, address
, fault
);
1159 if (fault
& VM_FAULT_MAJOR
)
1164 check_v8086_mode(regs
, address
, tsk
);
1166 up_read(&mm
->mmap_sem
);