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/mman.h>
15 #include <linux/smp.h>
16 #include <linux/interrupt.h>
17 #include <linux/init.h>
18 #include <linux/tty.h>
19 #include <linux/vt_kern.h> /* For unblank_screen() */
20 #include <linux/compiler.h>
21 #include <linux/highmem.h>
22 #include <linux/bootmem.h> /* for max_low_pfn */
23 #include <linux/vmalloc.h>
24 #include <linux/module.h>
25 #include <linux/kprobes.h>
26 #include <linux/uaccess.h>
27 #include <linux/kdebug.h>
29 #include <asm/system.h>
31 #include <asm/segment.h>
32 #include <asm/pgalloc.h>
34 #include <asm/tlbflush.h>
35 #include <asm/proto.h>
36 #include <asm-generic/sections.h>
39 * Page fault error code bits
40 * bit 0 == 0 means no page found, 1 means protection fault
41 * bit 1 == 0 means read, 1 means write
42 * bit 2 == 0 means kernel, 1 means user-mode
43 * bit 3 == 1 means use of reserved bit detected
44 * bit 4 == 1 means fault was an instruction fetch
46 #define PF_PROT (1<<0)
47 #define PF_WRITE (1<<1)
48 #define PF_USER (1<<2)
49 #define PF_RSVD (1<<3)
50 #define PF_INSTR (1<<4)
52 static inline int notify_page_fault(struct pt_regs
*regs
)
57 /* kprobe_running() needs smp_processor_id() */
59 if (!user_mode_vm(regs
)) {
61 if (!user_mode(regs
)) {
64 if (kprobe_running() && kprobe_fault_handler(regs
, 14))
77 * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch.
78 * Check that here and ignore it.
81 * Sometimes the CPU reports invalid exceptions on prefetch.
82 * Check that here and ignore it.
84 * Opcode checker based on code by Richard Brunner
86 static int is_prefetch(struct pt_regs
*regs
, unsigned long addr
,
87 unsigned long error_code
)
92 unsigned char *max_instr
;
95 if (!(__supported_pte_mask
& _PAGE_NX
))
99 /* If it was a exec fault on NX page, ignore */
100 if (error_code
& PF_INSTR
)
103 instr
= (unsigned char *)convert_ip_to_linear(current
, regs
);
104 max_instr
= instr
+ 15;
106 if (user_mode(regs
) && instr
>= (unsigned char *)TASK_SIZE
)
109 while (scan_more
&& instr
< max_instr
) {
110 unsigned char opcode
;
111 unsigned char instr_hi
;
112 unsigned char instr_lo
;
114 if (probe_kernel_address(instr
, opcode
))
117 instr_hi
= opcode
& 0xf0;
118 instr_lo
= opcode
& 0x0f;
125 * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes.
126 * In X86_64 long mode, the CPU will signal invalid
127 * opcode if some of these prefixes are present so
128 * X86_64 will never get here anyway
130 scan_more
= ((instr_lo
& 7) == 0x6);
135 * In AMD64 long mode 0x40..0x4F are valid REX prefixes
136 * Need to figure out under what instruction mode the
137 * instruction was issued. Could check the LDT for lm,
138 * but for now it's good enough to assume that long
139 * mode only uses well known segments or kernel.
141 scan_more
= (!user_mode(regs
)) || (regs
->cs
== __USER_CS
);
145 /* 0x64 thru 0x67 are valid prefixes in all modes. */
146 scan_more
= (instr_lo
& 0xC) == 0x4;
149 /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */
150 scan_more
= !instr_lo
|| (instr_lo
>>1) == 1;
153 /* Prefetch instruction is 0x0F0D or 0x0F18 */
156 if (probe_kernel_address(instr
, opcode
))
158 prefetch
= (instr_lo
== 0xF) &&
159 (opcode
== 0x0D || opcode
== 0x18);
169 static void force_sig_info_fault(int si_signo
, int si_code
,
170 unsigned long address
, struct task_struct
*tsk
)
174 info
.si_signo
= si_signo
;
176 info
.si_code
= si_code
;
177 info
.si_addr
= (void __user
*)address
;
178 force_sig_info(si_signo
, &info
, tsk
);
182 static int bad_address(void *p
)
185 return probe_kernel_address((unsigned long *)p
, dummy
);
189 void dump_pagetable(unsigned long address
)
192 __typeof__(pte_val(__pte(0))) page
;
195 page
= ((__typeof__(page
) *) __va(page
))[address
>> PGDIR_SHIFT
];
196 #ifdef CONFIG_X86_PAE
197 printk("*pdpt = %016Lx ", page
);
198 if ((page
>> PAGE_SHIFT
) < max_low_pfn
199 && page
& _PAGE_PRESENT
) {
201 page
= ((__typeof__(page
) *) __va(page
))[(address
>> PMD_SHIFT
)
202 & (PTRS_PER_PMD
- 1)];
203 printk(KERN_CONT
"*pde = %016Lx ", page
);
207 printk("*pde = %08lx ", page
);
211 * We must not directly access the pte in the highpte
212 * case if the page table is located in highmem.
213 * And let's rather not kmap-atomic the pte, just in case
214 * it's allocated already.
216 if ((page
>> PAGE_SHIFT
) < max_low_pfn
217 && (page
& _PAGE_PRESENT
)
218 && !(page
& _PAGE_PSE
)) {
220 page
= ((__typeof__(page
) *) __va(page
))[(address
>> PAGE_SHIFT
)
221 & (PTRS_PER_PTE
- 1)];
222 printk("*pte = %0*Lx ", sizeof(page
)*2, (u64
)page
);
226 #else /* CONFIG_X86_64 */
232 pgd
= (pgd_t
*)read_cr3();
234 pgd
= __va((unsigned long)pgd
& PHYSICAL_PAGE_MASK
);
235 pgd
+= pgd_index(address
);
236 if (bad_address(pgd
)) goto bad
;
237 printk("PGD %lx ", pgd_val(*pgd
));
238 if (!pgd_present(*pgd
)) goto ret
;
240 pud
= pud_offset(pgd
, address
);
241 if (bad_address(pud
)) goto bad
;
242 printk("PUD %lx ", pud_val(*pud
));
243 if (!pud_present(*pud
) || pud_large(*pud
))
246 pmd
= pmd_offset(pud
, address
);
247 if (bad_address(pmd
)) goto bad
;
248 printk("PMD %lx ", pmd_val(*pmd
));
249 if (!pmd_present(*pmd
) || pmd_large(*pmd
)) goto ret
;
251 pte
= pte_offset_kernel(pmd
, address
);
252 if (bad_address(pte
)) goto bad
;
253 printk("PTE %lx", pte_val(*pte
));
263 static inline pmd_t
*vmalloc_sync_one(pgd_t
*pgd
, unsigned long address
)
265 unsigned index
= pgd_index(address
);
271 pgd_k
= init_mm
.pgd
+ index
;
273 if (!pgd_present(*pgd_k
))
277 * set_pgd(pgd, *pgd_k); here would be useless on PAE
278 * and redundant with the set_pmd() on non-PAE. As would
282 pud
= pud_offset(pgd
, address
);
283 pud_k
= pud_offset(pgd_k
, address
);
284 if (!pud_present(*pud_k
))
287 pmd
= pmd_offset(pud
, address
);
288 pmd_k
= pmd_offset(pud_k
, address
);
289 if (!pmd_present(*pmd_k
))
291 if (!pmd_present(*pmd
)) {
292 set_pmd(pmd
, *pmd_k
);
293 arch_flush_lazy_mmu_mode();
295 BUG_ON(pmd_page(*pmd
) != pmd_page(*pmd_k
));
301 static const char errata93_warning
[] =
302 KERN_ERR
"******* Your BIOS seems to not contain a fix for K8 errata #93\n"
303 KERN_ERR
"******* Working around it, but it may cause SEGVs or burn power.\n"
304 KERN_ERR
"******* Please consider a BIOS update.\n"
305 KERN_ERR
"******* Disabling USB legacy in the BIOS may also help.\n";
308 /* Workaround for K8 erratum #93 & buggy BIOS.
309 BIOS SMM functions are required to use a specific workaround
310 to avoid corruption of the 64bit RIP register on C stepping K8.
311 A lot of BIOS that didn't get tested properly miss this.
312 The OS sees this as a page fault with the upper 32bits of RIP cleared.
313 Try to work around it here.
314 Note we only handle faults in kernel here.
315 Does nothing for X86_32
317 static int is_errata93(struct pt_regs
*regs
, unsigned long address
)
321 if (address
!= regs
->ip
)
323 if ((address
>> 32) != 0)
325 address
|= 0xffffffffUL
<< 32;
326 if ((address
>= (u64
)_stext
&& address
<= (u64
)_etext
) ||
327 (address
>= MODULES_VADDR
&& address
<= MODULES_END
)) {
329 printk(errata93_warning
);
340 * Work around K8 erratum #100 K8 in compat mode occasionally jumps to illegal
341 * addresses >4GB. We catch this in the page fault handler because these
342 * addresses are not reachable. Just detect this case and return. Any code
343 * segment in LDT is compatibility mode.
345 static int is_errata100(struct pt_regs
*regs
, unsigned long address
)
348 if ((regs
->cs
== __USER32_CS
|| (regs
->cs
& (1<<2))) &&
355 void do_invalid_op(struct pt_regs
*, unsigned long);
357 static int is_f00f_bug(struct pt_regs
*regs
, unsigned long address
)
359 #ifdef CONFIG_X86_F00F_BUG
362 * Pentium F0 0F C7 C8 bug workaround.
364 if (boot_cpu_data
.f00f_bug
) {
365 nr
= (address
- idt_descr
.address
) >> 3;
368 do_invalid_op(regs
, 0);
376 static void show_fault_oops(struct pt_regs
*regs
, unsigned long error_code
,
377 unsigned long address
)
380 if (!oops_may_print())
384 #ifdef CONFIG_X86_PAE
385 if (error_code
& PF_INSTR
) {
387 pte_t
*pte
= lookup_address(address
, &level
);
389 if (pte
&& pte_present(*pte
) && !pte_exec(*pte
))
390 printk(KERN_CRIT
"kernel tried to execute "
391 "NX-protected page - exploit attempt? "
392 "(uid: %d)\n", current
->uid
);
396 printk(KERN_ALERT
"BUG: unable to handle kernel ");
397 if (address
< PAGE_SIZE
)
398 printk(KERN_CONT
"NULL pointer dereference");
400 printk(KERN_CONT
"paging request");
402 printk(KERN_CONT
" at %08lx\n", address
);
404 printk(KERN_CONT
" at %016lx\n", address
);
406 printk(KERN_ALERT
"IP:");
407 printk_address(regs
->ip
, 1);
408 dump_pagetable(address
);
412 static noinline
void pgtable_bad(unsigned long address
, struct pt_regs
*regs
,
413 unsigned long error_code
)
415 unsigned long flags
= oops_begin();
416 struct task_struct
*tsk
;
418 printk(KERN_ALERT
"%s: Corrupted page table at address %lx\n",
419 current
->comm
, address
);
420 dump_pagetable(address
);
422 tsk
->thread
.cr2
= address
;
423 tsk
->thread
.trap_no
= 14;
424 tsk
->thread
.error_code
= error_code
;
425 if (__die("Bad pagetable", regs
, error_code
))
427 oops_end(flags
, regs
, SIGKILL
);
432 * Handle a spurious fault caused by a stale TLB entry. This allows
433 * us to lazily refresh the TLB when increasing the permissions of a
434 * kernel page (RO -> RW or NX -> X). Doing it eagerly is very
435 * expensive since that implies doing a full cross-processor TLB
436 * flush, even if no stale TLB entries exist on other processors.
437 * There are no security implications to leaving a stale TLB when
438 * increasing the permissions on a page.
440 static int spurious_fault(unsigned long address
,
441 unsigned long error_code
)
448 /* Reserved-bit violation or user access to kernel space? */
449 if (error_code
& (PF_USER
| PF_RSVD
))
452 pgd
= init_mm
.pgd
+ pgd_index(address
);
453 if (!pgd_present(*pgd
))
456 pud
= pud_offset(pgd
, address
);
457 if (!pud_present(*pud
))
460 pmd
= pmd_offset(pud
, address
);
461 if (!pmd_present(*pmd
))
464 pte
= pte_offset_kernel(pmd
, address
);
465 if (!pte_present(*pte
))
468 if ((error_code
& PF_WRITE
) && !pte_write(*pte
))
470 if ((error_code
& PF_INSTR
) && !pte_exec(*pte
))
478 * Handle a fault on the vmalloc or module mapping area
481 * Handle a fault on the vmalloc area
483 * This assumes no large pages in there.
485 static int vmalloc_fault(unsigned long address
)
488 unsigned long pgd_paddr
;
492 * Synchronize this task's top level page-table
493 * with the 'reference' page table.
495 * Do _not_ use "current" here. We might be inside
496 * an interrupt in the middle of a task switch..
498 pgd_paddr
= read_cr3();
499 pmd_k
= vmalloc_sync_one(__va(pgd_paddr
), address
);
502 pte_k
= pte_offset_kernel(pmd_k
, address
);
503 if (!pte_present(*pte_k
))
507 pgd_t
*pgd
, *pgd_ref
;
508 pud_t
*pud
, *pud_ref
;
509 pmd_t
*pmd
, *pmd_ref
;
510 pte_t
*pte
, *pte_ref
;
512 /* Make sure we are in vmalloc area */
513 if (!(address
>= VMALLOC_START
&& address
< VMALLOC_END
))
516 /* Copy kernel mappings over when needed. This can also
517 happen within a race in page table update. In the later
520 pgd
= pgd_offset(current
->mm
?: &init_mm
, address
);
521 pgd_ref
= pgd_offset_k(address
);
522 if (pgd_none(*pgd_ref
))
525 set_pgd(pgd
, *pgd_ref
);
527 BUG_ON(pgd_page_vaddr(*pgd
) != pgd_page_vaddr(*pgd_ref
));
529 /* Below here mismatches are bugs because these lower tables
532 pud
= pud_offset(pgd
, address
);
533 pud_ref
= pud_offset(pgd_ref
, address
);
534 if (pud_none(*pud_ref
))
536 if (pud_none(*pud
) || pud_page_vaddr(*pud
) != pud_page_vaddr(*pud_ref
))
538 pmd
= pmd_offset(pud
, address
);
539 pmd_ref
= pmd_offset(pud_ref
, address
);
540 if (pmd_none(*pmd_ref
))
542 if (pmd_none(*pmd
) || pmd_page(*pmd
) != pmd_page(*pmd_ref
))
544 pte_ref
= pte_offset_kernel(pmd_ref
, address
);
545 if (!pte_present(*pte_ref
))
547 pte
= pte_offset_kernel(pmd
, address
);
548 /* Don't use pte_page here, because the mappings can point
549 outside mem_map, and the NUMA hash lookup cannot handle
551 if (!pte_present(*pte
) || pte_pfn(*pte
) != pte_pfn(*pte_ref
))
557 int show_unhandled_signals
= 1;
560 * This routine handles page faults. It determines the address,
561 * and the problem, and then passes it off to one of the appropriate
567 void __kprobes
do_page_fault(struct pt_regs
*regs
, unsigned long error_code
)
569 struct task_struct
*tsk
;
570 struct mm_struct
*mm
;
571 struct vm_area_struct
*vma
;
572 unsigned long address
;
580 * We can fault from pretty much anywhere, with unknown IRQ state.
582 trace_hardirqs_fixup();
586 prefetchw(&mm
->mmap_sem
);
588 /* get the address */
589 address
= read_cr2();
591 si_code
= SEGV_MAPERR
;
593 if (notify_page_fault(regs
))
597 * We fault-in kernel-space virtual memory on-demand. The
598 * 'reference' page table is init_mm.pgd.
600 * NOTE! We MUST NOT take any locks for this case. We may
601 * be in an interrupt or a critical region, and should
602 * only copy the information from the master page table,
605 * This verifies that the fault happens in kernel space
606 * (error_code & 4) == 0, and that the fault was not a
607 * protection error (error_code & 9) == 0.
610 if (unlikely(address
>= TASK_SIZE
)) {
612 if (unlikely(address
>= TASK_SIZE64
)) {
614 if (!(error_code
& (PF_RSVD
|PF_USER
|PF_PROT
)) &&
615 vmalloc_fault(address
) >= 0)
618 /* Can handle a stale RO->RW TLB */
619 if (spurious_fault(address
, error_code
))
623 * Don't take the mm semaphore here. If we fixup a prefetch
624 * fault we could otherwise deadlock.
626 goto bad_area_nosemaphore
;
631 /* It's safe to allow irq's after cr2 has been saved and the vmalloc
632 fault has been handled. */
633 if (regs
->flags
& (X86_EFLAGS_IF
|VM_MASK
))
637 * If we're in an interrupt, have no user context or are running in an
638 * atomic region then we must not take the fault.
640 if (in_atomic() || !mm
)
641 goto bad_area_nosemaphore
;
642 #else /* CONFIG_X86_64 */
643 if (likely(regs
->flags
& X86_EFLAGS_IF
))
646 if (unlikely(error_code
& PF_RSVD
))
647 pgtable_bad(address
, regs
, error_code
);
650 * If we're in an interrupt, have no user context or are running in an
651 * atomic region then we must not take the fault.
653 if (unlikely(in_atomic() || !mm
))
654 goto bad_area_nosemaphore
;
657 * User-mode registers count as a user access even for any
658 * potential system fault or CPU buglet.
660 if (user_mode_vm(regs
))
661 error_code
|= PF_USER
;
664 /* When running in the kernel we expect faults to occur only to
665 * addresses in user space. All other faults represent errors in the
666 * kernel and should generate an OOPS. Unfortunately, in the case of an
667 * erroneous fault occurring in a code path which already holds mmap_sem
668 * we will deadlock attempting to validate the fault against the
669 * address space. Luckily the kernel only validly references user
670 * space from well defined areas of code, which are listed in the
673 * As the vast majority of faults will be valid we will only perform
674 * the source reference check when there is a possibility of a deadlock.
675 * Attempt to lock the address space, if we cannot we then validate the
676 * source. If this is invalid we can skip the address space check,
677 * thus avoiding the deadlock.
679 if (!down_read_trylock(&mm
->mmap_sem
)) {
680 if ((error_code
& PF_USER
) == 0 &&
681 !search_exception_tables(regs
->ip
))
682 goto bad_area_nosemaphore
;
683 down_read(&mm
->mmap_sem
);
686 vma
= find_vma(mm
, address
);
689 if (vma
->vm_start
<= address
)
691 if (!(vma
->vm_flags
& VM_GROWSDOWN
))
693 if (error_code
& PF_USER
) {
695 * Accessing the stack below %sp is always a bug.
696 * The large cushion allows instructions like enter
697 * and pusha to work. ("enter $65535,$31" pushes
698 * 32 pointers and then decrements %sp by 65535.)
700 if (address
+ 65536 + 32 * sizeof(unsigned long) < regs
->sp
)
703 if (expand_stack(vma
, address
))
706 * Ok, we have a good vm_area for this memory access, so
710 si_code
= SEGV_ACCERR
;
712 switch (error_code
& (PF_PROT
|PF_WRITE
)) {
713 default: /* 3: write, present */
715 case PF_WRITE
: /* write, not present */
716 if (!(vma
->vm_flags
& VM_WRITE
))
720 case PF_PROT
: /* read, present */
722 case 0: /* read, not present */
723 if (!(vma
->vm_flags
& (VM_READ
| VM_EXEC
| VM_WRITE
)))
731 * If for any reason at all we couldn't handle the fault,
732 * make sure we exit gracefully rather than endlessly redo
735 fault
= handle_mm_fault(mm
, vma
, address
, write
);
736 if (unlikely(fault
& VM_FAULT_ERROR
)) {
737 if (fault
& VM_FAULT_OOM
)
739 else if (fault
& VM_FAULT_SIGBUS
)
743 if (fault
& VM_FAULT_MAJOR
)
750 * Did it hit the DOS screen memory VA from vm86 mode?
752 if (v8086_mode(regs
)) {
753 unsigned long bit
= (address
- 0xA0000) >> PAGE_SHIFT
;
755 tsk
->thread
.screen_bitmap
|= 1 << bit
;
758 up_read(&mm
->mmap_sem
);
762 * Something tried to access memory that isn't in our memory map..
763 * Fix it, but check if it's kernel or user first..
766 up_read(&mm
->mmap_sem
);
768 bad_area_nosemaphore
:
769 /* User mode accesses just cause a SIGSEGV */
770 if (error_code
& PF_USER
) {
772 * It's possible to have interrupts off here.
777 * Valid to do another page fault here because this one came
780 if (is_prefetch(regs
, address
, error_code
))
783 if (is_errata100(regs
, address
))
786 if (show_unhandled_signals
&& unhandled_signal(tsk
, SIGSEGV
) &&
787 printk_ratelimit()) {
790 "%s%s[%d]: segfault at %lx ip %08lx sp %08lx error %lx",
792 "%s%s[%d]: segfault at %lx ip %lx sp %lx error %lx",
794 task_pid_nr(tsk
) > 1 ? KERN_INFO
: KERN_EMERG
,
795 tsk
->comm
, task_pid_nr(tsk
), address
, regs
->ip
,
796 regs
->sp
, error_code
);
797 print_vma_addr(" in ", regs
->ip
);
801 tsk
->thread
.cr2
= address
;
802 /* Kernel addresses are always protection faults */
803 tsk
->thread
.error_code
= error_code
| (address
>= TASK_SIZE
);
804 tsk
->thread
.trap_no
= 14;
805 force_sig_info_fault(SIGSEGV
, si_code
, address
, tsk
);
809 if (is_f00f_bug(regs
, address
))
813 /* Are we prepared to handle this kernel fault? */
814 if (fixup_exception(regs
))
819 * Valid to do another page fault here, because if this fault
820 * had been triggered by is_prefetch fixup_exception would have
824 * Hall of shame of CPU/BIOS bugs.
826 if (is_prefetch(regs
, address
, error_code
))
829 if (is_errata93(regs
, address
))
833 * Oops. The kernel tried to access some bad page. We'll have to
834 * terminate things with extreme prejudice.
839 flags
= oops_begin();
842 show_fault_oops(regs
, error_code
, address
);
844 tsk
->thread
.cr2
= address
;
845 tsk
->thread
.trap_no
= 14;
846 tsk
->thread
.error_code
= error_code
;
849 die("Oops", regs
, error_code
);
853 if (__die("Oops", regs
, error_code
))
855 /* Executive summary in case the body of the oops scrolled away */
856 printk(KERN_EMERG
"CR2: %016lx\n", address
);
857 oops_end(flags
, regs
, SIGKILL
);
861 * We ran out of memory, or some other thing happened to us that made
862 * us unable to handle the page fault gracefully.
865 up_read(&mm
->mmap_sem
);
866 if (is_global_init(tsk
)) {
869 down_read(&mm
->mmap_sem
);
876 printk("VM: killing process %s\n", tsk
->comm
);
877 if (error_code
& PF_USER
)
878 do_group_exit(SIGKILL
);
882 up_read(&mm
->mmap_sem
);
884 /* Kernel mode? Handle exceptions or die */
885 if (!(error_code
& PF_USER
))
888 /* User space => ok to do another page fault */
889 if (is_prefetch(regs
, address
, error_code
))
892 tsk
->thread
.cr2
= address
;
893 tsk
->thread
.error_code
= error_code
;
894 tsk
->thread
.trap_no
= 14;
895 force_sig_info_fault(SIGBUS
, BUS_ADRERR
, address
, tsk
);
898 DEFINE_SPINLOCK(pgd_lock
);
901 void vmalloc_sync_all(void)
905 * Note that races in the updates of insync and start aren't
906 * problematic: insync can only get set bits added, and updates to
907 * start are only improving performance (without affecting correctness
910 static DECLARE_BITMAP(insync
, PTRS_PER_PGD
);
911 static unsigned long start
= TASK_SIZE
;
912 unsigned long address
;
914 if (SHARED_KERNEL_PMD
)
917 BUILD_BUG_ON(TASK_SIZE
& ~PGDIR_MASK
);
918 for (address
= start
; address
>= TASK_SIZE
; address
+= PGDIR_SIZE
) {
919 if (!test_bit(pgd_index(address
), insync
)) {
923 spin_lock_irqsave(&pgd_lock
, flags
);
924 list_for_each_entry(page
, &pgd_list
, lru
) {
925 if (!vmalloc_sync_one(page_address(page
),
929 spin_unlock_irqrestore(&pgd_lock
, flags
);
931 set_bit(pgd_index(address
), insync
);
933 if (address
== start
&& test_bit(pgd_index(address
), insync
))
934 start
= address
+ PGDIR_SIZE
;
936 #else /* CONFIG_X86_64 */
938 * Note that races in the updates of insync and start aren't
939 * problematic: insync can only get set bits added, and updates to
940 * start are only improving performance (without affecting correctness
943 static DECLARE_BITMAP(insync
, PTRS_PER_PGD
);
944 static unsigned long start
= VMALLOC_START
& PGDIR_MASK
;
945 unsigned long address
;
947 for (address
= start
; address
<= VMALLOC_END
; address
+= PGDIR_SIZE
) {
948 if (!test_bit(pgd_index(address
), insync
)) {
949 const pgd_t
*pgd_ref
= pgd_offset_k(address
);
952 if (pgd_none(*pgd_ref
))
954 spin_lock(&pgd_lock
);
955 list_for_each_entry(page
, &pgd_list
, lru
) {
957 pgd
= (pgd_t
*)page_address(page
) + pgd_index(address
);
959 set_pgd(pgd
, *pgd_ref
);
961 BUG_ON(pgd_page_vaddr(*pgd
) != pgd_page_vaddr(*pgd_ref
));
963 spin_unlock(&pgd_lock
);
964 set_bit(pgd_index(address
), insync
);
966 if (address
== start
)
967 start
= address
+ PGDIR_SIZE
;
969 /* Check that there is no need to do the same for the modules area. */
970 BUILD_BUG_ON(!(MODULES_VADDR
> __START_KERNEL
));
971 BUILD_BUG_ON(!(((MODULES_END
- 1) & PGDIR_MASK
) ==
972 (__START_KERNEL
& PGDIR_MASK
)));