2 * Copyright (C) 1995 Linus Torvalds
3 * Copyright (C) 2001,2002 Andi Kleen, SuSE Labs.
6 #include <linux/signal.h>
7 #include <linux/sched.h>
8 #include <linux/kernel.h>
9 #include <linux/errno.h>
10 #include <linux/string.h>
11 #include <linux/types.h>
12 #include <linux/ptrace.h>
13 #include <linux/mmiotrace.h>
14 #include <linux/mman.h>
16 #include <linux/smp.h>
17 #include <linux/interrupt.h>
18 #include <linux/init.h>
19 #include <linux/tty.h>
20 #include <linux/vt_kern.h> /* For unblank_screen() */
21 #include <linux/compiler.h>
22 #include <linux/highmem.h>
23 #include <linux/bootmem.h> /* for max_low_pfn */
24 #include <linux/vmalloc.h>
25 #include <linux/module.h>
26 #include <linux/kprobes.h>
27 #include <linux/uaccess.h>
28 #include <linux/kdebug.h>
29 #include <linux/magic.h>
31 #include <asm/system.h>
33 #include <asm/segment.h>
34 #include <asm/pgalloc.h>
36 #include <asm/tlbflush.h>
37 #include <asm/proto.h>
38 #include <asm-generic/sections.h>
39 #include <asm/traps.h>
42 * Page fault error code bits
43 * bit 0 == 0 means no page found, 1 means protection fault
44 * bit 1 == 0 means read, 1 means write
45 * bit 2 == 0 means kernel, 1 means user-mode
46 * bit 3 == 1 means use of reserved bit detected
47 * bit 4 == 1 means fault was an instruction fetch
49 #define PF_PROT (1<<0)
50 #define PF_WRITE (1<<1)
51 #define PF_USER (1<<2)
52 #define PF_RSVD (1<<3)
53 #define PF_INSTR (1<<4)
55 static inline int kmmio_fault(struct pt_regs
*regs
, unsigned long addr
)
57 #ifdef CONFIG_MMIOTRACE
58 if (unlikely(is_kmmio_active()))
59 if (kmmio_handler(regs
, addr
) == 1)
65 static inline int notify_page_fault(struct pt_regs
*regs
)
70 /* kprobe_running() needs smp_processor_id() */
71 if (!user_mode_vm(regs
)) {
73 if (kprobe_running() && kprobe_fault_handler(regs
, 14))
86 * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch.
87 * Check that here and ignore it.
90 * Sometimes the CPU reports invalid exceptions on prefetch.
91 * Check that here and ignore it.
93 * Opcode checker based on code by Richard Brunner
95 static int is_prefetch(struct pt_regs
*regs
, unsigned long error_code
,
101 unsigned char *max_instr
;
104 * If it was a exec (instruction fetch) fault on NX page, then
105 * do not ignore the fault:
107 if (error_code
& PF_INSTR
)
110 instr
= (unsigned char *)convert_ip_to_linear(current
, regs
);
111 max_instr
= instr
+ 15;
113 if (user_mode(regs
) && instr
>= (unsigned char *)TASK_SIZE
)
116 while (scan_more
&& instr
< max_instr
) {
117 unsigned char opcode
;
118 unsigned char instr_hi
;
119 unsigned char instr_lo
;
121 if (probe_kernel_address(instr
, opcode
))
124 instr_hi
= opcode
& 0xf0;
125 instr_lo
= opcode
& 0x0f;
132 * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes.
133 * In X86_64 long mode, the CPU will signal invalid
134 * opcode if some of these prefixes are present so
135 * X86_64 will never get here anyway
137 scan_more
= ((instr_lo
& 7) == 0x6);
142 * In AMD64 long mode 0x40..0x4F are valid REX prefixes
143 * Need to figure out under what instruction mode the
144 * instruction was issued. Could check the LDT for lm,
145 * but for now it's good enough to assume that long
146 * mode only uses well known segments or kernel.
148 scan_more
= (!user_mode(regs
)) || (regs
->cs
== __USER_CS
);
152 /* 0x64 thru 0x67 are valid prefixes in all modes. */
153 scan_more
= (instr_lo
& 0xC) == 0x4;
156 /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */
157 scan_more
= !instr_lo
|| (instr_lo
>>1) == 1;
160 /* Prefetch instruction is 0x0F0D or 0x0F18 */
163 if (probe_kernel_address(instr
, opcode
))
165 prefetch
= (instr_lo
== 0xF) &&
166 (opcode
== 0x0D || opcode
== 0x18);
176 static void force_sig_info_fault(int si_signo
, int si_code
,
177 unsigned long address
, struct task_struct
*tsk
)
181 info
.si_signo
= si_signo
;
183 info
.si_code
= si_code
;
184 info
.si_addr
= (void __user
*)address
;
185 force_sig_info(si_signo
, &info
, tsk
);
189 static int bad_address(void *p
)
192 return probe_kernel_address((unsigned long *)p
, dummy
);
196 static void dump_pagetable(unsigned long address
)
199 __typeof__(pte_val(__pte(0))) page
;
202 page
= ((__typeof__(page
) *) __va(page
))[address
>> PGDIR_SHIFT
];
203 #ifdef CONFIG_X86_PAE
204 printk("*pdpt = %016Lx ", page
);
205 if ((page
>> PAGE_SHIFT
) < max_low_pfn
206 && page
& _PAGE_PRESENT
) {
208 page
= ((__typeof__(page
) *) __va(page
))[(address
>> PMD_SHIFT
)
209 & (PTRS_PER_PMD
- 1)];
210 printk(KERN_CONT
"*pde = %016Lx ", page
);
214 printk("*pde = %08lx ", page
);
218 * We must not directly access the pte in the highpte
219 * case if the page table is located in highmem.
220 * And let's rather not kmap-atomic the pte, just in case
221 * it's allocated already.
223 if ((page
>> PAGE_SHIFT
) < max_low_pfn
224 && (page
& _PAGE_PRESENT
)
225 && !(page
& _PAGE_PSE
)) {
227 page
= ((__typeof__(page
) *) __va(page
))[(address
>> PAGE_SHIFT
)
228 & (PTRS_PER_PTE
- 1)];
229 printk("*pte = %0*Lx ", sizeof(page
)*2, (u64
)page
);
233 #else /* CONFIG_X86_64 */
239 pgd
= (pgd_t
*)read_cr3();
241 pgd
= __va((unsigned long)pgd
& PHYSICAL_PAGE_MASK
);
242 pgd
+= pgd_index(address
);
243 if (bad_address(pgd
)) goto bad
;
244 printk("PGD %lx ", pgd_val(*pgd
));
245 if (!pgd_present(*pgd
)) goto ret
;
247 pud
= pud_offset(pgd
, address
);
248 if (bad_address(pud
)) goto bad
;
249 printk("PUD %lx ", pud_val(*pud
));
250 if (!pud_present(*pud
) || pud_large(*pud
))
253 pmd
= pmd_offset(pud
, address
);
254 if (bad_address(pmd
)) goto bad
;
255 printk("PMD %lx ", pmd_val(*pmd
));
256 if (!pmd_present(*pmd
) || pmd_large(*pmd
)) goto ret
;
258 pte
= pte_offset_kernel(pmd
, address
);
259 if (bad_address(pte
)) goto bad
;
260 printk("PTE %lx", pte_val(*pte
));
270 static inline pmd_t
*vmalloc_sync_one(pgd_t
*pgd
, unsigned long address
)
272 unsigned index
= pgd_index(address
);
278 pgd_k
= init_mm
.pgd
+ index
;
280 if (!pgd_present(*pgd_k
))
284 * set_pgd(pgd, *pgd_k); here would be useless on PAE
285 * and redundant with the set_pmd() on non-PAE. As would
289 pud
= pud_offset(pgd
, address
);
290 pud_k
= pud_offset(pgd_k
, address
);
291 if (!pud_present(*pud_k
))
294 pmd
= pmd_offset(pud
, address
);
295 pmd_k
= pmd_offset(pud_k
, address
);
296 if (!pmd_present(*pmd_k
))
298 if (!pmd_present(*pmd
)) {
299 set_pmd(pmd
, *pmd_k
);
300 arch_flush_lazy_mmu_mode();
302 BUG_ON(pmd_page(*pmd
) != pmd_page(*pmd_k
));
308 static const char errata93_warning
[] =
309 KERN_ERR
"******* Your BIOS seems to not contain a fix for K8 errata #93\n"
310 KERN_ERR
"******* Working around it, but it may cause SEGVs or burn power.\n"
311 KERN_ERR
"******* Please consider a BIOS update.\n"
312 KERN_ERR
"******* Disabling USB legacy in the BIOS may also help.\n";
315 /* Workaround for K8 erratum #93 & buggy BIOS.
316 BIOS SMM functions are required to use a specific workaround
317 to avoid corruption of the 64bit RIP register on C stepping K8.
318 A lot of BIOS that didn't get tested properly miss this.
319 The OS sees this as a page fault with the upper 32bits of RIP cleared.
320 Try to work around it here.
321 Note we only handle faults in kernel here.
322 Does nothing for X86_32
324 static int is_errata93(struct pt_regs
*regs
, unsigned long address
)
328 if (address
!= regs
->ip
)
330 if ((address
>> 32) != 0)
332 address
|= 0xffffffffUL
<< 32;
333 if ((address
>= (u64
)_stext
&& address
<= (u64
)_etext
) ||
334 (address
>= MODULES_VADDR
&& address
<= MODULES_END
)) {
336 printk(errata93_warning
);
347 * Work around K8 erratum #100 K8 in compat mode occasionally jumps to illegal
348 * addresses >4GB. We catch this in the page fault handler because these
349 * addresses are not reachable. Just detect this case and return. Any code
350 * segment in LDT is compatibility mode.
352 static int is_errata100(struct pt_regs
*regs
, unsigned long address
)
355 if ((regs
->cs
== __USER32_CS
|| (regs
->cs
& (1<<2))) &&
362 static int is_f00f_bug(struct pt_regs
*regs
, unsigned long address
)
364 #ifdef CONFIG_X86_F00F_BUG
367 * Pentium F0 0F C7 C8 bug workaround.
369 if (boot_cpu_data
.f00f_bug
) {
370 nr
= (address
- idt_descr
.address
) >> 3;
373 do_invalid_op(regs
, 0);
381 static void show_fault_oops(struct pt_regs
*regs
, unsigned long error_code
,
382 unsigned long address
)
385 if (!oops_may_print())
389 #ifdef CONFIG_X86_PAE
390 if (error_code
& PF_INSTR
) {
392 pte_t
*pte
= lookup_address(address
, &level
);
394 if (pte
&& pte_present(*pte
) && !pte_exec(*pte
))
395 printk(KERN_CRIT
"kernel tried to execute "
396 "NX-protected page - exploit attempt? "
397 "(uid: %d)\n", current_uid());
401 printk(KERN_ALERT
"BUG: unable to handle kernel ");
402 if (address
< PAGE_SIZE
)
403 printk(KERN_CONT
"NULL pointer dereference");
405 printk(KERN_CONT
"paging request");
406 printk(KERN_CONT
" at %p\n", (void *) address
);
407 printk(KERN_ALERT
"IP:");
408 printk_address(regs
->ip
, 1);
409 dump_pagetable(address
);
413 static noinline
void pgtable_bad(struct pt_regs
*regs
,
414 unsigned long error_code
, unsigned long address
)
416 unsigned long flags
= oops_begin();
418 struct task_struct
*tsk
= current
;
420 printk(KERN_ALERT
"%s: Corrupted page table at address %lx\n",
422 dump_pagetable(address
);
423 tsk
->thread
.cr2
= address
;
424 tsk
->thread
.trap_no
= 14;
425 tsk
->thread
.error_code
= error_code
;
426 if (__die("Bad pagetable", regs
, error_code
))
428 oops_end(flags
, regs
, sig
);
432 static noinline
void no_context(struct pt_regs
*regs
,
433 unsigned long error_code
, unsigned long address
)
435 struct task_struct
*tsk
= current
;
436 unsigned long *stackend
;
443 /* Are we prepared to handle this kernel fault? */
444 if (fixup_exception(regs
))
449 * Valid to do another page fault here, because if this fault
450 * had been triggered by is_prefetch fixup_exception would have
454 * Hall of shame of CPU/BIOS bugs.
456 if (is_prefetch(regs
, error_code
, address
))
459 if (is_errata93(regs
, address
))
463 * Oops. The kernel tried to access some bad page. We'll have to
464 * terminate things with extreme prejudice.
469 flags
= oops_begin();
472 show_fault_oops(regs
, error_code
, address
);
474 stackend
= end_of_stack(tsk
);
475 if (*stackend
!= STACK_END_MAGIC
)
476 printk(KERN_ALERT
"Thread overran stack, or stack corrupted\n");
478 tsk
->thread
.cr2
= address
;
479 tsk
->thread
.trap_no
= 14;
480 tsk
->thread
.error_code
= error_code
;
483 die("Oops", regs
, error_code
);
488 if (__die("Oops", regs
, error_code
))
490 /* Executive summary in case the body of the oops scrolled away */
491 printk(KERN_EMERG
"CR2: %016lx\n", address
);
492 oops_end(flags
, regs
, sig
);
496 static void __bad_area_nosemaphore(struct pt_regs
*regs
,
497 unsigned long error_code
, unsigned long address
,
500 struct task_struct
*tsk
= current
;
502 /* User mode accesses just cause a SIGSEGV */
503 if (error_code
& PF_USER
) {
505 * It's possible to have interrupts off here.
510 * Valid to do another page fault here because this one came
513 if (is_prefetch(regs
, error_code
, address
))
516 if (is_errata100(regs
, address
))
519 if (show_unhandled_signals
&& unhandled_signal(tsk
, SIGSEGV
) &&
520 printk_ratelimit()) {
522 "%s%s[%d]: segfault at %lx ip %p sp %p error %lx",
523 task_pid_nr(tsk
) > 1 ? KERN_INFO
: KERN_EMERG
,
524 tsk
->comm
, task_pid_nr(tsk
), address
,
525 (void *) regs
->ip
, (void *) regs
->sp
, error_code
);
526 print_vma_addr(" in ", regs
->ip
);
530 tsk
->thread
.cr2
= address
;
531 /* Kernel addresses are always protection faults */
532 tsk
->thread
.error_code
= error_code
| (address
>= TASK_SIZE
);
533 tsk
->thread
.trap_no
= 14;
534 force_sig_info_fault(SIGSEGV
, si_code
, address
, tsk
);
538 if (is_f00f_bug(regs
, address
))
541 no_context(regs
, error_code
, address
);
544 static noinline
void bad_area_nosemaphore(struct pt_regs
*regs
,
545 unsigned long error_code
, unsigned long address
)
547 __bad_area_nosemaphore(regs
, error_code
, address
, SEGV_MAPERR
);
550 static void __bad_area(struct pt_regs
*regs
,
551 unsigned long error_code
, unsigned long address
,
554 struct mm_struct
*mm
= current
->mm
;
557 * Something tried to access memory that isn't in our memory map..
558 * Fix it, but check if it's kernel or user first..
560 up_read(&mm
->mmap_sem
);
562 __bad_area_nosemaphore(regs
, error_code
, address
, si_code
);
565 static noinline
void bad_area(struct pt_regs
*regs
,
566 unsigned long error_code
, unsigned long address
)
568 __bad_area(regs
, error_code
, address
, SEGV_MAPERR
);
571 static noinline
void bad_area_access_error(struct pt_regs
*regs
,
572 unsigned long error_code
, unsigned long address
)
574 __bad_area(regs
, error_code
, address
, SEGV_ACCERR
);
577 /* TODO: fixup for "mm-invoke-oom-killer-from-page-fault.patch" */
578 static void out_of_memory(struct pt_regs
*regs
,
579 unsigned long error_code
, unsigned long address
)
582 * We ran out of memory, call the OOM killer, and return the userspace
583 * (which will retry the fault, or kill us if we got oom-killed).
585 up_read(¤t
->mm
->mmap_sem
);
586 pagefault_out_of_memory();
589 static void do_sigbus(struct pt_regs
*regs
,
590 unsigned long error_code
, unsigned long address
)
592 struct task_struct
*tsk
= current
;
593 struct mm_struct
*mm
= tsk
->mm
;
595 up_read(&mm
->mmap_sem
);
597 /* Kernel mode? Handle exceptions or die */
598 if (!(error_code
& PF_USER
))
599 no_context(regs
, error_code
, address
);
601 /* User space => ok to do another page fault */
602 if (is_prefetch(regs
, error_code
, address
))
605 tsk
->thread
.cr2
= address
;
606 tsk
->thread
.error_code
= error_code
;
607 tsk
->thread
.trap_no
= 14;
608 force_sig_info_fault(SIGBUS
, BUS_ADRERR
, address
, tsk
);
611 static noinline
void mm_fault_error(struct pt_regs
*regs
,
612 unsigned long error_code
, unsigned long address
, unsigned int fault
)
614 if (fault
& VM_FAULT_OOM
)
615 out_of_memory(regs
, error_code
, address
);
616 else if (fault
& VM_FAULT_SIGBUS
)
617 do_sigbus(regs
, error_code
, address
);
622 static int spurious_fault_check(unsigned long error_code
, pte_t
*pte
)
624 if ((error_code
& PF_WRITE
) && !pte_write(*pte
))
626 if ((error_code
& PF_INSTR
) && !pte_exec(*pte
))
633 * Handle a spurious fault caused by a stale TLB entry. This allows
634 * us to lazily refresh the TLB when increasing the permissions of a
635 * kernel page (RO -> RW or NX -> X). Doing it eagerly is very
636 * expensive since that implies doing a full cross-processor TLB
637 * flush, even if no stale TLB entries exist on other processors.
638 * There are no security implications to leaving a stale TLB when
639 * increasing the permissions on a page.
641 static noinline
int spurious_fault(unsigned long error_code
,
642 unsigned long address
)
649 /* Reserved-bit violation or user access to kernel space? */
650 if (error_code
& (PF_USER
| PF_RSVD
))
653 pgd
= init_mm
.pgd
+ pgd_index(address
);
654 if (!pgd_present(*pgd
))
657 pud
= pud_offset(pgd
, address
);
658 if (!pud_present(*pud
))
662 return spurious_fault_check(error_code
, (pte_t
*) pud
);
664 pmd
= pmd_offset(pud
, address
);
665 if (!pmd_present(*pmd
))
669 return spurious_fault_check(error_code
, (pte_t
*) pmd
);
671 pte
= pte_offset_kernel(pmd
, address
);
672 if (!pte_present(*pte
))
675 return spurious_fault_check(error_code
, pte
);
680 * Handle a fault on the vmalloc or module mapping area
683 * Handle a fault on the vmalloc area
685 * This assumes no large pages in there.
687 static noinline
int vmalloc_fault(unsigned long address
)
690 unsigned long pgd_paddr
;
694 /* Make sure we are in vmalloc area */
695 if (!(address
>= VMALLOC_START
&& address
< VMALLOC_END
))
699 * Synchronize this task's top level page-table
700 * with the 'reference' page table.
702 * Do _not_ use "current" here. We might be inside
703 * an interrupt in the middle of a task switch..
705 pgd_paddr
= read_cr3();
706 pmd_k
= vmalloc_sync_one(__va(pgd_paddr
), address
);
709 pte_k
= pte_offset_kernel(pmd_k
, address
);
710 if (!pte_present(*pte_k
))
714 pgd_t
*pgd
, *pgd_ref
;
715 pud_t
*pud
, *pud_ref
;
716 pmd_t
*pmd
, *pmd_ref
;
717 pte_t
*pte
, *pte_ref
;
719 /* Make sure we are in vmalloc area */
720 if (!(address
>= VMALLOC_START
&& address
< VMALLOC_END
))
723 /* Copy kernel mappings over when needed. This can also
724 happen within a race in page table update. In the later
727 pgd
= pgd_offset(current
->active_mm
, address
);
728 pgd_ref
= pgd_offset_k(address
);
729 if (pgd_none(*pgd_ref
))
732 set_pgd(pgd
, *pgd_ref
);
734 BUG_ON(pgd_page_vaddr(*pgd
) != pgd_page_vaddr(*pgd_ref
));
736 /* Below here mismatches are bugs because these lower tables
739 pud
= pud_offset(pgd
, address
);
740 pud_ref
= pud_offset(pgd_ref
, address
);
741 if (pud_none(*pud_ref
))
743 if (pud_none(*pud
) || pud_page_vaddr(*pud
) != pud_page_vaddr(*pud_ref
))
745 pmd
= pmd_offset(pud
, address
);
746 pmd_ref
= pmd_offset(pud_ref
, address
);
747 if (pmd_none(*pmd_ref
))
749 if (pmd_none(*pmd
) || pmd_page(*pmd
) != pmd_page(*pmd_ref
))
751 pte_ref
= pte_offset_kernel(pmd_ref
, address
);
752 if (!pte_present(*pte_ref
))
754 pte
= pte_offset_kernel(pmd
, address
);
755 /* Don't use pte_page here, because the mappings can point
756 outside mem_map, and the NUMA hash lookup cannot handle
758 if (!pte_present(*pte
) || pte_pfn(*pte
) != pte_pfn(*pte_ref
))
764 int show_unhandled_signals
= 1;
766 static inline int access_error(unsigned long error_code
, int write
,
767 struct vm_area_struct
*vma
)
770 /* write, present and write, not present */
771 if (unlikely(!(vma
->vm_flags
& VM_WRITE
)))
773 } else if (unlikely(error_code
& PF_PROT
)) {
777 /* read, not present */
778 if (unlikely(!(vma
->vm_flags
& (VM_READ
| VM_EXEC
| VM_WRITE
))))
785 static int fault_in_kernel_space(unsigned long address
)
788 return address
>= TASK_SIZE
;
789 #else /* !CONFIG_X86_32 */
790 return address
>= TASK_SIZE64
;
791 #endif /* CONFIG_X86_32 */
795 * This routine handles page faults. It determines the address,
796 * and the problem, and then passes it off to one of the appropriate
802 void __kprobes
do_page_fault(struct pt_regs
*regs
, unsigned long error_code
)
804 unsigned long address
;
805 struct task_struct
*tsk
;
806 struct mm_struct
*mm
;
807 struct vm_area_struct
*vma
;
813 prefetchw(&mm
->mmap_sem
);
815 /* get the address */
816 address
= read_cr2();
818 if (unlikely(kmmio_fault(regs
, address
)))
822 * We fault-in kernel-space virtual memory on-demand. The
823 * 'reference' page table is init_mm.pgd.
825 * NOTE! We MUST NOT take any locks for this case. We may
826 * be in an interrupt or a critical region, and should
827 * only copy the information from the master page table,
830 * This verifies that the fault happens in kernel space
831 * (error_code & 4) == 0, and that the fault was not a
832 * protection error (error_code & 9) == 0.
834 if (unlikely(fault_in_kernel_space(address
))) {
835 if (!(error_code
& (PF_RSVD
|PF_USER
|PF_PROT
)) &&
836 vmalloc_fault(address
) >= 0)
839 /* Can handle a stale RO->RW TLB */
840 if (spurious_fault(error_code
, address
))
843 /* kprobes don't want to hook the spurious faults. */
844 if (notify_page_fault(regs
))
847 * Don't take the mm semaphore here. If we fixup a prefetch
848 * fault we could otherwise deadlock.
850 bad_area_nosemaphore(regs
, error_code
, address
);
854 /* kprobes don't want to hook the spurious faults. */
855 if (unlikely(notify_page_fault(regs
)))
858 * It's safe to allow irq's after cr2 has been saved and the
859 * vmalloc fault has been handled.
861 * User-mode registers count as a user access even for any
862 * potential system fault or CPU buglet.
864 if (user_mode_vm(regs
)) {
866 error_code
|= PF_USER
;
867 } else if (regs
->flags
& X86_EFLAGS_IF
)
871 if (unlikely(error_code
& PF_RSVD
))
872 pgtable_bad(regs
, error_code
, address
);
876 * If we're in an interrupt, have no user context or are running in an
877 * atomic region then we must not take the fault.
879 if (unlikely(in_atomic() || !mm
)) {
880 bad_area_nosemaphore(regs
, error_code
, address
);
885 * When running in the kernel we expect faults to occur only to
886 * addresses in user space. All other faults represent errors in the
887 * kernel and should generate an OOPS. Unfortunately, in the case of an
888 * erroneous fault occurring in a code path which already holds mmap_sem
889 * we will deadlock attempting to validate the fault against the
890 * address space. Luckily the kernel only validly references user
891 * space from well defined areas of code, which are listed in the
894 * As the vast majority of faults will be valid we will only perform
895 * the source reference check when there is a possibility of a deadlock.
896 * Attempt to lock the address space, if we cannot we then validate the
897 * source. If this is invalid we can skip the address space check,
898 * thus avoiding the deadlock.
900 if (unlikely(!down_read_trylock(&mm
->mmap_sem
))) {
901 if ((error_code
& PF_USER
) == 0 &&
902 !search_exception_tables(regs
->ip
)) {
903 bad_area_nosemaphore(regs
, error_code
, address
);
906 down_read(&mm
->mmap_sem
);
909 * The above down_read_trylock() might have succeeded in which
910 * case we'll have missed the might_sleep() from down_read().
915 vma
= find_vma(mm
, address
);
916 if (unlikely(!vma
)) {
917 bad_area(regs
, error_code
, address
);
920 if (likely(vma
->vm_start
<= address
))
922 if (unlikely(!(vma
->vm_flags
& VM_GROWSDOWN
))) {
923 bad_area(regs
, error_code
, address
);
926 if (error_code
& PF_USER
) {
928 * Accessing the stack below %sp is always a bug.
929 * The large cushion allows instructions like enter
930 * and pusha to work. ("enter $65535,$31" pushes
931 * 32 pointers and then decrements %sp by 65535.)
933 if (unlikely(address
+ 65536 + 32 * sizeof(unsigned long) < regs
->sp
)) {
934 bad_area(regs
, error_code
, address
);
938 if (unlikely(expand_stack(vma
, address
))) {
939 bad_area(regs
, error_code
, address
);
944 * Ok, we have a good vm_area for this memory access, so
948 write
= error_code
& PF_WRITE
;
949 if (unlikely(access_error(error_code
, write
, vma
))) {
950 bad_area_access_error(regs
, error_code
, address
);
955 * If for any reason at all we couldn't handle the fault,
956 * make sure we exit gracefully rather than endlessly redo
959 fault
= handle_mm_fault(mm
, vma
, address
, write
);
960 if (unlikely(fault
& VM_FAULT_ERROR
)) {
961 mm_fault_error(regs
, error_code
, address
, fault
);
964 if (fault
& VM_FAULT_MAJOR
)
971 * Did it hit the DOS screen memory VA from vm86 mode?
973 if (v8086_mode(regs
)) {
974 unsigned long bit
= (address
- 0xA0000) >> PAGE_SHIFT
;
976 tsk
->thread
.screen_bitmap
|= 1 << bit
;
979 up_read(&mm
->mmap_sem
);
982 DEFINE_SPINLOCK(pgd_lock
);
985 void vmalloc_sync_all(void)
987 unsigned long address
;
990 if (SHARED_KERNEL_PMD
)
993 for (address
= VMALLOC_START
& PMD_MASK
;
994 address
>= TASK_SIZE
&& address
< FIXADDR_TOP
;
995 address
+= PMD_SIZE
) {
999 spin_lock_irqsave(&pgd_lock
, flags
);
1000 list_for_each_entry(page
, &pgd_list
, lru
) {
1001 if (!vmalloc_sync_one(page_address(page
),
1005 spin_unlock_irqrestore(&pgd_lock
, flags
);
1007 #else /* CONFIG_X86_64 */
1008 for (address
= VMALLOC_START
& PGDIR_MASK
; address
<= VMALLOC_END
;
1009 address
+= PGDIR_SIZE
) {
1010 const pgd_t
*pgd_ref
= pgd_offset_k(address
);
1011 unsigned long flags
;
1014 if (pgd_none(*pgd_ref
))
1016 spin_lock_irqsave(&pgd_lock
, flags
);
1017 list_for_each_entry(page
, &pgd_list
, lru
) {
1019 pgd
= (pgd_t
*)page_address(page
) + pgd_index(address
);
1021 set_pgd(pgd
, *pgd_ref
);
1023 BUG_ON(pgd_page_vaddr(*pgd
) != pgd_page_vaddr(*pgd_ref
));
1025 spin_unlock_irqrestore(&pgd_lock
, flags
);