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/vmalloc.h>
22 #include <linux/module.h>
23 #include <linux/kprobes.h>
24 #include <linux/uaccess.h>
25 #include <linux/kdebug.h>
27 #include <asm/system.h>
28 #include <asm/pgalloc.h>
30 #include <asm/tlbflush.h>
31 #include <asm/proto.h>
32 #include <asm-generic/sections.h>
35 * Page fault error code bits
36 * bit 0 == 0 means no page found, 1 means protection fault
37 * bit 1 == 0 means read, 1 means write
38 * bit 2 == 0 means kernel, 1 means user-mode
39 * bit 3 == 1 means use of reserved bit detected
40 * bit 4 == 1 means fault was an instruction fetch
42 #define PF_PROT (1<<0)
43 #define PF_WRITE (1<<1)
44 #define PF_USER (1<<2)
45 #define PF_RSVD (1<<3)
46 #define PF_INSTR (1<<4)
48 static inline int notify_page_fault(struct pt_regs
*regs
)
53 /* kprobe_running() needs smp_processor_id() */
55 if (!user_mode_vm(regs
)) {
57 if (!user_mode(regs
)) {
60 if (kprobe_running() && kprobe_fault_handler(regs
, 14))
73 * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch.
74 * Check that here and ignore it.
77 * Sometimes the CPU reports invalid exceptions on prefetch.
78 * Check that here and ignore it.
80 * Opcode checker based on code by Richard Brunner
82 static int is_prefetch(struct pt_regs
*regs
, unsigned long addr
,
83 unsigned long error_code
)
88 unsigned char *max_instr
;
91 if (unlikely(boot_cpu_data
.x86_vendor
== X86_VENDOR_AMD
&&
92 boot_cpu_data
.x86
>= 6)) {
93 /* Catch an obscure case of prefetch inside an NX page. */
94 if (nx_enabled
&& (error_code
& PF_INSTR
))
100 /* If it was a exec fault ignore */
101 if (error_code
& PF_INSTR
)
105 instr
= (unsigned char *)convert_ip_to_linear(current
, regs
);
106 max_instr
= instr
+ 15;
108 if (user_mode(regs
) && instr
>= (unsigned char *)TASK_SIZE
)
111 while (scan_more
&& instr
< max_instr
) {
112 unsigned char opcode
;
113 unsigned char instr_hi
;
114 unsigned char instr_lo
;
116 if (probe_kernel_address(instr
, opcode
))
119 instr_hi
= opcode
& 0xf0;
120 instr_lo
= opcode
& 0x0f;
127 * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes.
128 * In X86_64 long mode, the CPU will signal invalid
129 * opcode if some of these prefixes are present so
130 * X86_64 will never get here anyway
132 scan_more
= ((instr_lo
& 7) == 0x6);
137 * In AMD64 long mode 0x40..0x4F are valid REX prefixes
138 * Need to figure out under what instruction mode the
139 * instruction was issued. Could check the LDT for lm,
140 * but for now it's good enough to assume that long
141 * mode only uses well known segments or kernel.
143 scan_more
= (!user_mode(regs
)) || (regs
->cs
== __USER_CS
);
147 /* 0x64 thru 0x67 are valid prefixes in all modes. */
148 scan_more
= (instr_lo
& 0xC) == 0x4;
151 /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */
152 scan_more
= !instr_lo
|| (instr_lo
>>1) == 1;
155 /* Prefetch instruction is 0x0F0D or 0x0F18 */
158 if (probe_kernel_address(instr
, opcode
))
160 prefetch
= (instr_lo
== 0xF) &&
161 (opcode
== 0x0D || opcode
== 0x18);
171 static void force_sig_info_fault(int si_signo
, int si_code
,
172 unsigned long address
, struct task_struct
*tsk
)
176 info
.si_signo
= si_signo
;
178 info
.si_code
= si_code
;
179 info
.si_addr
= (void __user
*)address
;
180 force_sig_info(si_signo
, &info
, tsk
);
184 static int bad_address(void *p
)
187 return probe_kernel_address((unsigned long *)p
, dummy
);
191 void dump_pagetable(unsigned long address
)
194 __typeof__(pte_val(__pte(0))) page
;
197 page
= ((__typeof__(page
) *) __va(page
))[address
>> PGDIR_SHIFT
];
198 #ifdef CONFIG_X86_PAE
199 printk("*pdpt = %016Lx ", page
);
200 if ((page
>> PAGE_SHIFT
) < max_low_pfn
201 && page
& _PAGE_PRESENT
) {
203 page
= ((__typeof__(page
) *) __va(page
))[(address
>> PMD_SHIFT
)
204 & (PTRS_PER_PMD
- 1)];
205 printk(KERN_CONT
"*pde = %016Lx ", page
);
209 printk("*pde = %08lx ", page
);
213 * We must not directly access the pte in the highpte
214 * case if the page table is located in highmem.
215 * And let's rather not kmap-atomic the pte, just in case
216 * it's allocated already.
218 if ((page
>> PAGE_SHIFT
) < max_low_pfn
219 && (page
& _PAGE_PRESENT
)
220 && !(page
& _PAGE_PSE
)) {
222 page
= ((__typeof__(page
) *) __va(page
))[(address
>> PAGE_SHIFT
)
223 & (PTRS_PER_PTE
- 1)];
224 printk("*pte = %0*Lx ", sizeof(page
)*2, (u64
)page
);
228 #else /* CONFIG_X86_64 */
234 pgd
= (pgd_t
*)read_cr3();
236 pgd
= __va((unsigned long)pgd
& PHYSICAL_PAGE_MASK
);
237 pgd
+= pgd_index(address
);
238 if (bad_address(pgd
)) goto bad
;
239 printk("PGD %lx ", pgd_val(*pgd
));
240 if (!pgd_present(*pgd
)) goto ret
;
242 pud
= pud_offset(pgd
, address
);
243 if (bad_address(pud
)) goto bad
;
244 printk("PUD %lx ", pud_val(*pud
));
245 if (!pud_present(*pud
)) goto ret
;
247 pmd
= pmd_offset(pud
, address
);
248 if (bad_address(pmd
)) goto bad
;
249 printk("PMD %lx ", pmd_val(*pmd
));
250 if (!pmd_present(*pmd
) || pmd_large(*pmd
)) goto ret
;
252 pte
= pte_offset_kernel(pmd
, address
);
253 if (bad_address(pte
)) goto bad
;
254 printk("PTE %lx", pte_val(*pte
));
264 static inline pmd_t
*vmalloc_sync_one(pgd_t
*pgd
, unsigned long address
)
266 unsigned index
= pgd_index(address
);
272 pgd_k
= init_mm
.pgd
+ index
;
274 if (!pgd_present(*pgd_k
))
278 * set_pgd(pgd, *pgd_k); here would be useless on PAE
279 * and redundant with the set_pmd() on non-PAE. As would
283 pud
= pud_offset(pgd
, address
);
284 pud_k
= pud_offset(pgd_k
, address
);
285 if (!pud_present(*pud_k
))
288 pmd
= pmd_offset(pud
, address
);
289 pmd_k
= pmd_offset(pud_k
, address
);
290 if (!pmd_present(*pmd_k
))
292 if (!pmd_present(*pmd
)) {
293 set_pmd(pmd
, *pmd_k
);
294 arch_flush_lazy_mmu_mode();
296 BUG_ON(pmd_page(*pmd
) != pmd_page(*pmd_k
));
302 static const char errata93_warning
[] =
303 KERN_ERR
"******* Your BIOS seems to not contain a fix for K8 errata #93\n"
304 KERN_ERR
"******* Working around it, but it may cause SEGVs or burn power.\n"
305 KERN_ERR
"******* Please consider a BIOS update.\n"
306 KERN_ERR
"******* Disabling USB legacy in the BIOS may also help.\n";
309 /* Workaround for K8 erratum #93 & buggy BIOS.
310 BIOS SMM functions are required to use a specific workaround
311 to avoid corruption of the 64bit RIP register on C stepping K8.
312 A lot of BIOS that didn't get tested properly miss this.
313 The OS sees this as a page fault with the upper 32bits of RIP cleared.
314 Try to work around it here.
315 Note we only handle faults in kernel here.
316 Does nothing for X86_32
318 static int is_errata93(struct pt_regs
*regs
, unsigned long address
)
322 if (address
!= regs
->ip
)
324 if ((address
>> 32) != 0)
326 address
|= 0xffffffffUL
<< 32;
327 if ((address
>= (u64
)_stext
&& address
<= (u64
)_etext
) ||
328 (address
>= MODULES_VADDR
&& address
<= MODULES_END
)) {
330 printk(errata93_warning
);
341 * Work around K8 erratum #100 K8 in compat mode occasionally jumps to illegal
342 * addresses >4GB. We catch this in the page fault handler because these
343 * addresses are not reachable. Just detect this case and return. Any code
344 * segment in LDT is compatibility mode.
346 static int is_errata100(struct pt_regs
*regs
, unsigned long address
)
349 if ((regs
->cs
== __USER32_CS
|| (regs
->cs
& (1<<2))) &&
356 void do_invalid_op(struct pt_regs
*, unsigned long);
358 static int is_f00f_bug(struct pt_regs
*regs
, unsigned long address
)
360 #ifdef CONFIG_X86_F00F_BUG
363 * Pentium F0 0F C7 C8 bug workaround.
365 if (boot_cpu_data
.f00f_bug
) {
366 nr
= (address
- idt_descr
.address
) >> 3;
369 do_invalid_op(regs
, 0);
377 static void show_fault_oops(struct pt_regs
*regs
, unsigned long error_code
,
378 unsigned long address
)
381 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
);
395 printk(KERN_ALERT
"BUG: unable to handle kernel ");
396 if (address
< PAGE_SIZE
)
397 printk(KERN_CONT
"NULL pointer dereference");
399 printk(KERN_CONT
"paging request");
400 printk(KERN_CONT
" at %08lx\n", address
);
402 printk(KERN_ALERT
"IP:");
403 printk_address(regs
->ip
, 1);
404 dump_pagetable(address
);
405 #else /* CONFIG_X86_64 */
406 printk(KERN_ALERT
"BUG: unable to handle kernel ");
407 if (address
< PAGE_SIZE
)
408 printk(KERN_CONT
"NULL pointer dereference");
410 printk(KERN_CONT
"paging request");
411 printk(KERN_CONT
" at %016lx\n", address
);
413 printk(KERN_ALERT
"IP:");
414 printk_address(regs
->ip
, 1);
415 dump_pagetable(address
);
420 static noinline
void pgtable_bad(unsigned long address
, struct pt_regs
*regs
,
421 unsigned long error_code
)
423 unsigned long flags
= oops_begin();
424 struct task_struct
*tsk
;
426 printk(KERN_ALERT
"%s: Corrupted page table at address %lx\n",
427 current
->comm
, address
);
428 dump_pagetable(address
);
430 tsk
->thread
.cr2
= address
;
431 tsk
->thread
.trap_no
= 14;
432 tsk
->thread
.error_code
= error_code
;
433 if (__die("Bad pagetable", regs
, error_code
))
435 oops_end(flags
, regs
, SIGKILL
);
441 * Handle a fault on the vmalloc or module mapping area
444 * Handle a fault on the vmalloc area
446 * This assumes no large pages in there.
448 static int vmalloc_fault(unsigned long address
)
451 unsigned long pgd_paddr
;
455 * Synchronize this task's top level page-table
456 * with the 'reference' page table.
458 * Do _not_ use "current" here. We might be inside
459 * an interrupt in the middle of a task switch..
461 pgd_paddr
= read_cr3();
462 pmd_k
= vmalloc_sync_one(__va(pgd_paddr
), address
);
465 pte_k
= pte_offset_kernel(pmd_k
, address
);
466 if (!pte_present(*pte_k
))
470 pgd_t
*pgd
, *pgd_ref
;
471 pud_t
*pud
, *pud_ref
;
472 pmd_t
*pmd
, *pmd_ref
;
473 pte_t
*pte
, *pte_ref
;
475 /* Copy kernel mappings over when needed. This can also
476 happen within a race in page table update. In the later
479 pgd
= pgd_offset(current
->mm
?: &init_mm
, address
);
480 pgd_ref
= pgd_offset_k(address
);
481 if (pgd_none(*pgd_ref
))
484 set_pgd(pgd
, *pgd_ref
);
486 BUG_ON(pgd_page_vaddr(*pgd
) != pgd_page_vaddr(*pgd_ref
));
488 /* Below here mismatches are bugs because these lower tables
491 pud
= pud_offset(pgd
, address
);
492 pud_ref
= pud_offset(pgd_ref
, address
);
493 if (pud_none(*pud_ref
))
495 if (pud_none(*pud
) || pud_page_vaddr(*pud
) != pud_page_vaddr(*pud_ref
))
497 pmd
= pmd_offset(pud
, address
);
498 pmd_ref
= pmd_offset(pud_ref
, address
);
499 if (pmd_none(*pmd_ref
))
501 if (pmd_none(*pmd
) || pmd_page(*pmd
) != pmd_page(*pmd_ref
))
503 pte_ref
= pte_offset_kernel(pmd_ref
, address
);
504 if (!pte_present(*pte_ref
))
506 pte
= pte_offset_kernel(pmd
, address
);
507 /* Don't use pte_page here, because the mappings can point
508 outside mem_map, and the NUMA hash lookup cannot handle
510 if (!pte_present(*pte
) || pte_pfn(*pte
) != pte_pfn(*pte_ref
))
516 int show_unhandled_signals
= 1;
519 * This routine handles page faults. It determines the address,
520 * and the problem, and then passes it off to one of the appropriate
526 void __kprobes
do_page_fault(struct pt_regs
*regs
, unsigned long error_code
)
528 struct task_struct
*tsk
;
529 struct mm_struct
*mm
;
530 struct vm_area_struct
*vma
;
531 unsigned long address
;
539 * We can fault from pretty much anywhere, with unknown IRQ state.
541 trace_hardirqs_fixup();
545 prefetchw(&mm
->mmap_sem
);
547 /* get the address */
548 address
= read_cr2();
550 si_code
= SEGV_MAPERR
;
552 if (notify_page_fault(regs
))
556 * We fault-in kernel-space virtual memory on-demand. The
557 * 'reference' page table is init_mm.pgd.
559 * NOTE! We MUST NOT take any locks for this case. We may
560 * be in an interrupt or a critical region, and should
561 * only copy the information from the master page table,
564 * This verifies that the fault happens in kernel space
565 * (error_code & 4) == 0, and that the fault was not a
566 * protection error (error_code & 9) == 0.
569 if (unlikely(address
>= TASK_SIZE
)) {
570 if (!(error_code
& (PF_RSVD
|PF_USER
|PF_PROT
)) &&
571 vmalloc_fault(address
) >= 0)
574 * Don't take the mm semaphore here. If we fixup a prefetch
575 * fault we could otherwise deadlock.
577 goto bad_area_nosemaphore
;
580 /* It's safe to allow irq's after cr2 has been saved and the vmalloc
581 fault has been handled. */
582 if (regs
->flags
& (X86_EFLAGS_IF
|VM_MASK
))
586 * If we're in an interrupt, have no user context or are running in an
587 * atomic region then we must not take the fault.
589 if (in_atomic() || !mm
)
590 goto bad_area_nosemaphore
;
591 #else /* CONFIG_X86_64 */
592 if (unlikely(address
>= TASK_SIZE64
)) {
594 * Don't check for the module range here: its PML4
595 * is always initialized because it's shared with the main
596 * kernel text. Only vmalloc may need PML4 syncups.
598 if (!(error_code
& (PF_RSVD
|PF_USER
|PF_PROT
)) &&
599 ((address
>= VMALLOC_START
&& address
< VMALLOC_END
))) {
600 if (vmalloc_fault(address
) >= 0)
604 * Don't take the mm semaphore here. If we fixup a prefetch
605 * fault we could otherwise deadlock.
607 goto bad_area_nosemaphore
;
609 if (likely(regs
->flags
& X86_EFLAGS_IF
))
612 if (unlikely(error_code
& PF_RSVD
))
613 pgtable_bad(address
, regs
, error_code
);
616 * If we're in an interrupt, have no user context or are running in an
617 * atomic region then we must not take the fault.
619 if (unlikely(in_atomic() || !mm
))
620 goto bad_area_nosemaphore
;
623 * User-mode registers count as a user access even for any
624 * potential system fault or CPU buglet.
626 if (user_mode_vm(regs
))
627 error_code
|= PF_USER
;
630 /* When running in the kernel we expect faults to occur only to
631 * addresses in user space. All other faults represent errors in the
632 * kernel and should generate an OOPS. Unfortunately, in the case of an
633 * erroneous fault occurring in a code path which already holds mmap_sem
634 * we will deadlock attempting to validate the fault against the
635 * address space. Luckily the kernel only validly references user
636 * space from well defined areas of code, which are listed in the
639 * As the vast majority of faults will be valid we will only perform
640 * the source reference check when there is a possibility of a deadlock.
641 * Attempt to lock the address space, if we cannot we then validate the
642 * source. If this is invalid we can skip the address space check,
643 * thus avoiding the deadlock.
645 if (!down_read_trylock(&mm
->mmap_sem
)) {
646 if ((error_code
& PF_USER
) == 0 &&
647 !search_exception_tables(regs
->ip
))
648 goto bad_area_nosemaphore
;
649 down_read(&mm
->mmap_sem
);
652 vma
= find_vma(mm
, address
);
656 if (vma
->vm_start
<= address
)
658 if (likely(vma
->vm_start
<= address
))
661 if (!(vma
->vm_flags
& VM_GROWSDOWN
))
663 if (error_code
& PF_USER
) {
665 * Accessing the stack below %sp is always a bug.
666 * The large cushion allows instructions like enter
667 * and pusha to work. ("enter $65535,$31" pushes
668 * 32 pointers and then decrements %sp by 65535.)
670 if (address
+ 65536 + 32 * sizeof(unsigned long) < regs
->sp
)
673 if (expand_stack(vma
, address
))
676 * Ok, we have a good vm_area for this memory access, so
680 si_code
= SEGV_ACCERR
;
682 switch (error_code
& (PF_PROT
|PF_WRITE
)) {
683 default: /* 3: write, present */
685 case PF_WRITE
: /* write, not present */
686 if (!(vma
->vm_flags
& VM_WRITE
))
690 case PF_PROT
: /* read, present */
692 case 0: /* read, not present */
693 if (!(vma
->vm_flags
& (VM_READ
| VM_EXEC
| VM_WRITE
)))
701 * If for any reason at all we couldn't handle the fault,
702 * make sure we exit gracefully rather than endlessly redo
705 fault
= handle_mm_fault(mm
, vma
, address
, write
);
706 if (unlikely(fault
& VM_FAULT_ERROR
)) {
707 if (fault
& VM_FAULT_OOM
)
709 else if (fault
& VM_FAULT_SIGBUS
)
713 if (fault
& VM_FAULT_MAJOR
)
720 * Did it hit the DOS screen memory VA from vm86 mode?
722 if (v8086_mode(regs
)) {
723 unsigned long bit
= (address
- 0xA0000) >> PAGE_SHIFT
;
725 tsk
->thread
.screen_bitmap
|= 1 << bit
;
728 up_read(&mm
->mmap_sem
);
732 * Something tried to access memory that isn't in our memory map..
733 * Fix it, but check if it's kernel or user first..
736 up_read(&mm
->mmap_sem
);
738 bad_area_nosemaphore
:
739 /* User mode accesses just cause a SIGSEGV */
740 if (error_code
& PF_USER
) {
742 * It's possible to have interrupts off here.
747 * Valid to do another page fault here because this one came
750 if (is_prefetch(regs
, address
, error_code
))
753 if (is_errata100(regs
, address
))
756 if (show_unhandled_signals
&& unhandled_signal(tsk
, SIGSEGV
) &&
757 printk_ratelimit()) {
760 "%s%s[%d]: segfault at %lx ip %08lx sp %08lx error %lx",
762 "%s%s[%d]: segfault at %lx ip %lx sp %lx error %lx",
764 task_pid_nr(tsk
) > 1 ? KERN_INFO
: KERN_EMERG
,
765 tsk
->comm
, task_pid_nr(tsk
), address
, regs
->ip
,
766 regs
->sp
, error_code
);
767 print_vma_addr(" in ", regs
->ip
);
771 tsk
->thread
.cr2
= address
;
772 /* Kernel addresses are always protection faults */
773 tsk
->thread
.error_code
= error_code
| (address
>= TASK_SIZE
);
774 tsk
->thread
.trap_no
= 14;
775 force_sig_info_fault(SIGSEGV
, si_code
, address
, tsk
);
779 if (is_f00f_bug(regs
, address
))
783 /* Are we prepared to handle this kernel fault? */
784 if (fixup_exception(regs
))
789 * Valid to do another page fault here, because if this fault
790 * had been triggered by is_prefetch fixup_exception would have
794 * Hall of shame of CPU/BIOS bugs.
796 if (is_prefetch(regs
, address
, error_code
))
799 if (is_errata93(regs
, address
))
803 * Oops. The kernel tried to access some bad page. We'll have to
804 * terminate things with extreme prejudice.
809 show_fault_oops(regs
, error_code
, address
);
811 tsk
->thread
.cr2
= address
;
812 tsk
->thread
.trap_no
= 14;
813 tsk
->thread
.error_code
= error_code
;
814 die("Oops", regs
, error_code
);
817 #else /* CONFIG_X86_64 */
818 flags
= oops_begin();
820 show_fault_oops(regs
, error_code
, address
);
822 tsk
->thread
.cr2
= address
;
823 tsk
->thread
.trap_no
= 14;
824 tsk
->thread
.error_code
= error_code
;
825 if (__die("Oops", regs
, error_code
))
827 /* Executive summary in case the body of the oops scrolled away */
828 printk(KERN_EMERG
"CR2: %016lx\n", address
);
829 oops_end(flags
, regs
, SIGKILL
);
833 * We ran out of memory, or some other thing happened to us that made
834 * us unable to handle the page fault gracefully.
837 up_read(&mm
->mmap_sem
);
839 if (is_global_init(tsk
)) {
841 down_read(&mm
->mmap_sem
);
845 if (is_global_init(current
)) {
850 printk("VM: killing process %s\n", tsk
->comm
);
851 if (error_code
& PF_USER
)
852 do_group_exit(SIGKILL
);
856 up_read(&mm
->mmap_sem
);
858 /* Kernel mode? Handle exceptions or die */
859 if (!(error_code
& PF_USER
))
862 /* User space => ok to do another page fault */
863 if (is_prefetch(regs
, address
, error_code
))
866 tsk
->thread
.cr2
= address
;
867 tsk
->thread
.error_code
= error_code
;
868 tsk
->thread
.trap_no
= 14;
869 force_sig_info_fault(SIGBUS
, BUS_ADRERR
, address
, tsk
);
873 DEFINE_SPINLOCK(pgd_lock
);
877 void vmalloc_sync_all(void)
881 * Note that races in the updates of insync and start aren't
882 * problematic: insync can only get set bits added, and updates to
883 * start are only improving performance (without affecting correctness
886 static DECLARE_BITMAP(insync
, PTRS_PER_PGD
);
887 static unsigned long start
= TASK_SIZE
;
888 unsigned long address
;
890 if (SHARED_KERNEL_PMD
)
893 BUILD_BUG_ON(TASK_SIZE
& ~PGDIR_MASK
);
894 for (address
= start
; address
>= TASK_SIZE
; address
+= PGDIR_SIZE
) {
895 if (!test_bit(pgd_index(address
), insync
)) {
899 spin_lock_irqsave(&pgd_lock
, flags
);
900 for (page
= pgd_list
; page
; page
=
901 (struct page
*)page
->index
)
902 if (!vmalloc_sync_one(page_address(page
),
904 BUG_ON(page
!= pgd_list
);
907 spin_unlock_irqrestore(&pgd_lock
, flags
);
909 set_bit(pgd_index(address
), insync
);
911 if (address
== start
&& test_bit(pgd_index(address
), insync
))
912 start
= address
+ PGDIR_SIZE
;
914 #else /* CONFIG_X86_64 */
916 * Note that races in the updates of insync and start aren't
917 * problematic: insync can only get set bits added, and updates to
918 * start are only improving performance (without affecting correctness
921 static DECLARE_BITMAP(insync
, PTRS_PER_PGD
);
922 static unsigned long start
= VMALLOC_START
& PGDIR_MASK
;
923 unsigned long address
;
925 for (address
= start
; address
<= VMALLOC_END
; address
+= PGDIR_SIZE
) {
926 if (!test_bit(pgd_index(address
), insync
)) {
927 const pgd_t
*pgd_ref
= pgd_offset_k(address
);
930 if (pgd_none(*pgd_ref
))
932 spin_lock(&pgd_lock
);
933 list_for_each_entry(page
, &pgd_list
, lru
) {
935 pgd
= (pgd_t
*)page_address(page
) + pgd_index(address
);
937 set_pgd(pgd
, *pgd_ref
);
939 BUG_ON(pgd_page_vaddr(*pgd
) != pgd_page_vaddr(*pgd_ref
));
941 spin_unlock(&pgd_lock
);
942 set_bit(pgd_index(address
), insync
);
944 if (address
== start
)
945 start
= address
+ PGDIR_SIZE
;
947 /* Check that there is no need to do the same for the modules area. */
948 BUILD_BUG_ON(!(MODULES_VADDR
> __START_KERNEL
));
949 BUILD_BUG_ON(!(((MODULES_END
- 1) & PGDIR_MASK
) ==
950 (__START_KERNEL
& PGDIR_MASK
)));