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 {
58 static inline int kmmio_fault(struct pt_regs *regs, unsigned long addr)
60 #ifdef CONFIG_MMIOTRACE
61 if (unlikely(is_kmmio_active()))
62 if (kmmio_handler(regs, addr) == 1)
68 static inline int notify_page_fault(struct pt_regs *regs)
73 /* kprobe_running() needs smp_processor_id() */
74 if (!user_mode_vm(regs)) {
76 if (kprobe_running() && kprobe_fault_handler(regs, 14))
92 * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch.
93 * Check that here and ignore it.
97 * Sometimes the CPU reports invalid exceptions on prefetch.
98 * Check that here and ignore it.
100 * Opcode checker based on code by Richard Brunner.
103 check_prefetch_opcode(struct pt_regs *regs, unsigned char *instr,
104 unsigned char opcode, int *prefetch)
106 unsigned char instr_hi = opcode & 0xf0;
107 unsigned char instr_lo = opcode & 0x0f;
113 * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes.
114 * In X86_64 long mode, the CPU will signal invalid
115 * opcode if some of these prefixes are present so
116 * X86_64 will never get here anyway
118 return ((instr_lo & 7) == 0x6);
122 * In AMD64 long mode 0x40..0x4F are valid REX prefixes
123 * Need to figure out under what instruction mode the
124 * instruction was issued. Could check the LDT for lm,
125 * but for now it's good enough to assume that long
126 * mode only uses well known segments or kernel.
128 return (!user_mode(regs)) || (regs->cs == __USER_CS);
131 /* 0x64 thru 0x67 are valid prefixes in all modes. */
132 return (instr_lo & 0xC) == 0x4;
134 /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */
135 return !instr_lo || (instr_lo>>1) == 1;
137 /* Prefetch instruction is 0x0F0D or 0x0F18 */
138 if (probe_kernel_address(instr, opcode))
141 *prefetch = (instr_lo == 0xF) &&
142 (opcode == 0x0D || opcode == 0x18);
150 is_prefetch(struct pt_regs *regs, unsigned long error_code, unsigned long addr)
152 unsigned char *max_instr;
153 unsigned char *instr;
157 * If it was a exec (instruction fetch) fault on NX page, then
158 * do not ignore the fault:
160 if (error_code & PF_INSTR)
163 instr = (void *)convert_ip_to_linear(current, regs);
164 max_instr = instr + 15;
166 if (user_mode(regs) && instr >= (unsigned char *)TASK_SIZE)
169 while (instr < max_instr) {
170 unsigned char opcode;
172 if (probe_kernel_address(instr, opcode))
177 if (!check_prefetch_opcode(regs, instr, opcode, &prefetch))
184 force_sig_info_fault(int si_signo, int si_code, unsigned long address,
185 struct task_struct *tsk)
189 info.si_signo = si_signo;
191 info.si_code = si_code;
192 info.si_addr = (void __user *)address;
194 force_sig_info(si_signo, &info, tsk);
198 static int bad_address(void *p)
202 return probe_kernel_address((unsigned long *)p, dummy);
206 static void dump_pagetable(unsigned long address)
209 __typeof__(pte_val(__pte(0))) page;
212 page = ((__typeof__(page) *) __va(page))[address >> PGDIR_SHIFT];
214 #ifdef CONFIG_X86_PAE
215 printk("*pdpt = %016Lx ", page);
216 if ((page >> PAGE_SHIFT) < max_low_pfn
217 && page & _PAGE_PRESENT) {
219 page = ((__typeof__(page) *) __va(page))[(address >> PMD_SHIFT)
220 & (PTRS_PER_PMD - 1)];
221 printk(KERN_CONT "*pde = %016Lx ", page);
225 printk("*pde = %08lx ", page);
229 * We must not directly access the pte in the highpte
230 * case if the page table is located in highmem.
231 * And let's rather not kmap-atomic the pte, just in case
232 * it's allocated already:
234 if ((page >> PAGE_SHIFT) < max_low_pfn
235 && (page & _PAGE_PRESENT)
236 && !(page & _PAGE_PSE)) {
239 page = ((__typeof__(page) *) __va(page))[(address >> PAGE_SHIFT)
240 & (PTRS_PER_PTE - 1)];
241 printk("*pte = %0*Lx ", sizeof(page)*2, (u64)page);
245 #else /* CONFIG_X86_64 */
251 pgd = (pgd_t *)read_cr3();
253 pgd = __va((unsigned long)pgd & PHYSICAL_PAGE_MASK);
255 pgd += pgd_index(address);
256 if (bad_address(pgd))
259 printk("PGD %lx ", pgd_val(*pgd));
261 if (!pgd_present(*pgd))
264 pud = pud_offset(pgd, address);
265 if (bad_address(pud))
268 printk("PUD %lx ", pud_val(*pud));
269 if (!pud_present(*pud) || pud_large(*pud))
272 pmd = pmd_offset(pud, address);
273 if (bad_address(pmd))
276 printk("PMD %lx ", pmd_val(*pmd));
277 if (!pmd_present(*pmd) || pmd_large(*pmd))
280 pte = pte_offset_kernel(pmd, address);
281 if (bad_address(pte))
284 printk("PTE %lx", pte_val(*pte));
294 static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address)
296 unsigned index = pgd_index(address);
302 pgd_k = init_mm.pgd + index;
304 if (!pgd_present(*pgd_k))
308 * set_pgd(pgd, *pgd_k); here would be useless on PAE
309 * and redundant with the set_pmd() on non-PAE. As would
312 pud = pud_offset(pgd, address);
313 pud_k = pud_offset(pgd_k, address);
314 if (!pud_present(*pud_k))
317 pmd = pmd_offset(pud, address);
318 pmd_k = pmd_offset(pud_k, address);
319 if (!pmd_present(*pmd_k))
322 if (!pmd_present(*pmd)) {
323 set_pmd(pmd, *pmd_k);
324 arch_flush_lazy_mmu_mode();
326 BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k));
333 * Did it hit the DOS screen memory VA from vm86 mode?
336 check_v8086_mode(struct pt_regs *regs, unsigned long address,
337 struct task_struct *tsk)
341 if (!v8086_mode(regs))
344 bit = (address - 0xA0000) >> PAGE_SHIFT;
346 tsk->thread.screen_bitmap |= 1 << bit;
349 #else /* CONFIG_X86_64: */
351 static const char errata93_warning[] =
352 KERN_ERR "******* Your BIOS seems to not contain a fix for K8 errata #93\n"
353 KERN_ERR "******* Working around it, but it may cause SEGVs or burn power.\n"
354 KERN_ERR "******* Please consider a BIOS update.\n"
355 KERN_ERR "******* Disabling USB legacy in the BIOS may also help.\n";
358 * No vm86 mode in 64-bit mode:
361 check_v8086_mode(struct pt_regs *regs, unsigned long address,
362 struct task_struct *tsk)
369 * Workaround for K8 erratum #93 & buggy BIOS.
371 * BIOS SMM functions are required to use a specific workaround
372 * to avoid corruption of the 64bit RIP register on C stepping K8.
374 * A lot of BIOS that didn't get tested properly miss this.
376 * The OS sees this as a page fault with the upper 32bits of RIP cleared.
377 * Try to work around it here.
379 * Note we only handle faults in kernel here.
380 * Does nothing on 32-bit.
382 static int is_errata93(struct pt_regs *regs, unsigned long address)
387 if (address != regs->ip)
390 if ((address >> 32) != 0)
393 address |= 0xffffffffUL << 32;
394 if ((address >= (u64)_stext && address <= (u64)_etext) ||
395 (address >= MODULES_VADDR && address <= MODULES_END)) {
397 printk(errata93_warning);
408 * Work around K8 erratum #100 K8 in compat mode occasionally jumps
409 * to illegal addresses >4GB.
411 * We catch this in the page fault handler because these addresses
412 * are not reachable. Just detect this case and return. Any code
413 * segment in LDT is compatibility mode.
415 static int is_errata100(struct pt_regs *regs, unsigned long address)
418 if ((regs->cs == __USER32_CS || (regs->cs & (1<<2))) && (address >> 32))
424 static int is_f00f_bug(struct pt_regs *regs, unsigned long address)
426 #ifdef CONFIG_X86_F00F_BUG
430 * Pentium F0 0F C7 C8 bug workaround:
432 if (boot_cpu_data.f00f_bug) {
433 nr = (address - idt_descr.address) >> 3;
436 do_invalid_op(regs, 0);
445 show_fault_oops(struct pt_regs *regs, unsigned long error_code,
446 unsigned long address)
449 if (!oops_may_print())
453 #ifdef CONFIG_X86_PAE
454 if (error_code & PF_INSTR) {
457 pte_t *pte = lookup_address(address, &level);
459 if (pte && pte_present(*pte) && !pte_exec(*pte)) {
460 printk(KERN_CRIT "kernel tried to execute "
461 "NX-protected page - exploit attempt? "
462 "(uid: %d)\n", current_uid());
467 printk(KERN_ALERT "BUG: unable to handle kernel ");
468 if (address < PAGE_SIZE)
469 printk(KERN_CONT "NULL pointer dereference");
471 printk(KERN_CONT "paging request");
473 printk(KERN_CONT " at %p\n", (void *) address);
474 printk(KERN_ALERT "IP:");
475 printk_address(regs->ip, 1);
477 dump_pagetable(address);
482 pgtable_bad(struct pt_regs *regs, unsigned long error_code,
483 unsigned long address)
485 struct task_struct *tsk;
489 flags = oops_begin();
493 printk(KERN_ALERT "%s: Corrupted page table at address %lx\n",
495 dump_pagetable(address);
497 tsk->thread.cr2 = address;
498 tsk->thread.trap_no = 14;
499 tsk->thread.error_code = error_code;
501 if (__die("Bad pagetable", regs, error_code))
504 oops_end(flags, regs, sig);
509 no_context(struct pt_regs *regs, unsigned long error_code,
510 unsigned long address)
512 struct task_struct *tsk = current;
513 unsigned long *stackend;
520 /* Are we prepared to handle this kernel fault? */
521 if (fixup_exception(regs))
527 * Valid to do another page fault here, because if this fault
528 * had been triggered by is_prefetch fixup_exception would have
533 * Hall of shame of CPU/BIOS bugs.
535 if (is_prefetch(regs, error_code, address))
538 if (is_errata93(regs, address))
542 * Oops. The kernel tried to access some bad page. We'll have to
543 * terminate things with extreme prejudice:
548 flags = oops_begin();
551 show_fault_oops(regs, error_code, address);
553 stackend = end_of_stack(tsk);
554 if (*stackend != STACK_END_MAGIC)
555 printk(KERN_ALERT "Thread overran stack, or stack corrupted\n");
557 tsk->thread.cr2 = address;
558 tsk->thread.trap_no = 14;
559 tsk->thread.error_code = error_code;
562 die("Oops", regs, error_code);
567 if (__die("Oops", regs, error_code))
570 /* Executive summary in case the body of the oops scrolled away */
571 printk(KERN_EMERG "CR2: %016lx\n", address);
573 oops_end(flags, regs, sig);
578 * Print out info about fatal segfaults, if the show_unhandled_signals
582 show_signal_msg(struct pt_regs *regs, unsigned long error_code,
583 unsigned long address, struct task_struct *tsk)
585 if (!unhandled_signal(tsk, SIGSEGV))
588 if (!printk_ratelimit())
591 printk(KERN_CONT "%s%s[%d]: segfault at %lx ip %p sp %p error %lx",
592 task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG,
593 tsk->comm, task_pid_nr(tsk), address,
594 (void *)regs->ip, (void *)regs->sp, error_code);
596 print_vma_addr(KERN_CONT " in ", regs->ip);
598 printk(KERN_CONT "\n");
602 __bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
603 unsigned long address, int si_code)
605 struct task_struct *tsk = current;
607 /* User mode accesses just cause a SIGSEGV */
608 if (error_code & PF_USER) {
610 * It's possible to have interrupts off here:
615 * Valid to do another page fault here because this one came
618 if (is_prefetch(regs, error_code, address))
621 if (is_errata100(regs, address))
624 if (unlikely(show_unhandled_signals))
625 show_signal_msg(regs, error_code, address, tsk);
627 /* Kernel addresses are always protection faults: */
628 tsk->thread.cr2 = address;
629 tsk->thread.error_code = error_code | (address >= TASK_SIZE);
630 tsk->thread.trap_no = 14;
632 force_sig_info_fault(SIGSEGV, si_code, address, tsk);
637 if (is_f00f_bug(regs, address))
640 no_context(regs, error_code, address);
644 bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
645 unsigned long address)
647 __bad_area_nosemaphore(regs, error_code, address, SEGV_MAPERR);
651 __bad_area(struct pt_regs *regs, unsigned long error_code,
652 unsigned long address, int si_code)
654 struct mm_struct *mm = current->mm;
657 * Something tried to access memory that isn't in our memory map..
658 * Fix it, but check if it's kernel or user first..
660 up_read(&mm->mmap_sem);
662 __bad_area_nosemaphore(regs, error_code, address, si_code);
666 bad_area(struct pt_regs *regs, unsigned long error_code, unsigned long address)
668 __bad_area(regs, error_code, address, SEGV_MAPERR);
672 bad_area_access_error(struct pt_regs *regs, unsigned long error_code,
673 unsigned long address)
675 __bad_area(regs, error_code, address, SEGV_ACCERR);
678 /* TODO: fixup for "mm-invoke-oom-killer-from-page-fault.patch" */
680 out_of_memory(struct pt_regs *regs, unsigned long error_code,
681 unsigned long address)
684 * We ran out of memory, call the OOM killer, and return the userspace
685 * (which will retry the fault, or kill us if we got oom-killed):
687 up_read(¤t->mm->mmap_sem);
689 pagefault_out_of_memory();
693 do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address)
695 struct task_struct *tsk = current;
696 struct mm_struct *mm = tsk->mm;
698 up_read(&mm->mmap_sem);
700 /* Kernel mode? Handle exceptions or die: */
701 if (!(error_code & PF_USER))
702 no_context(regs, error_code, address);
705 /* User space => ok to do another page fault: */
706 if (is_prefetch(regs, error_code, address))
710 tsk->thread.cr2 = address;
711 tsk->thread.error_code = error_code;
712 tsk->thread.trap_no = 14;
714 force_sig_info_fault(SIGBUS, BUS_ADRERR, address, tsk);
718 mm_fault_error(struct pt_regs *regs, unsigned long error_code,
719 unsigned long address, unsigned int fault)
721 if (fault & VM_FAULT_OOM) {
722 out_of_memory(regs, error_code, address);
724 if (fault & VM_FAULT_SIGBUS)
725 do_sigbus(regs, error_code, address);
731 static int spurious_fault_check(unsigned long error_code, pte_t *pte)
733 if ((error_code & PF_WRITE) && !pte_write(*pte))
736 if ((error_code & PF_INSTR) && !pte_exec(*pte))
743 * Handle a spurious fault caused by a stale TLB entry.
745 * This allows us to lazily refresh the TLB when increasing the
746 * permissions of a kernel page (RO -> RW or NX -> X). Doing it
747 * eagerly is very expensive since that implies doing a full
748 * cross-processor TLB flush, even if no stale TLB entries exist
749 * on other processors.
751 * There are no security implications to leaving a stale TLB when
752 * increasing the permissions on a page.
755 spurious_fault(unsigned long error_code, unsigned long address)
763 /* Reserved-bit violation or user access to kernel space? */
764 if (error_code & (PF_USER | PF_RSVD))
767 pgd = init_mm.pgd + pgd_index(address);
768 if (!pgd_present(*pgd))
771 pud = pud_offset(pgd, address);
772 if (!pud_present(*pud))
776 return spurious_fault_check(error_code, (pte_t *) pud);
778 pmd = pmd_offset(pud, address);
779 if (!pmd_present(*pmd))
783 return spurious_fault_check(error_code, (pte_t *) pmd);
785 pte = pte_offset_kernel(pmd, address);
786 if (!pte_present(*pte))
789 ret = spurious_fault_check(error_code, pte);
794 * Make sure we have permissions in PMD.
795 * If not, then there's a bug in the page tables:
797 ret = spurious_fault_check(error_code, (pte_t *) pmd);
798 WARN_ONCE(!ret, "PMD has incorrect permission bits\n");
806 * Handle a fault on the vmalloc or module mapping area
810 * Handle a fault on the vmalloc area
812 * This assumes no large pages in there.
814 static noinline int vmalloc_fault(unsigned long address)
817 unsigned long pgd_paddr;
821 /* Make sure we are in vmalloc area: */
822 if (!(address >= VMALLOC_START && address < VMALLOC_END))
826 * Synchronize this task's top level page-table
827 * with the 'reference' page table.
829 * Do _not_ use "current" here. We might be inside
830 * an interrupt in the middle of a task switch..
832 pgd_paddr = read_cr3();
833 pmd_k = vmalloc_sync_one(__va(pgd_paddr), address);
837 pte_k = pte_offset_kernel(pmd_k, address);
838 if (!pte_present(*pte_k))
843 pgd_t *pgd, *pgd_ref;
844 pud_t *pud, *pud_ref;
845 pmd_t *pmd, *pmd_ref;
846 pte_t *pte, *pte_ref;
848 /* Make sure we are in vmalloc area: */
849 if (!(address >= VMALLOC_START && address < VMALLOC_END))
853 * Copy kernel mappings over when needed. This can also
854 * happen within a race in page table update. In the later
857 pgd = pgd_offset(current->active_mm, address);
858 pgd_ref = pgd_offset_k(address);
859 if (pgd_none(*pgd_ref))
863 set_pgd(pgd, *pgd_ref);
865 BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref));
868 * Below here mismatches are bugs because these lower tables
872 pud = pud_offset(pgd, address);
873 pud_ref = pud_offset(pgd_ref, address);
874 if (pud_none(*pud_ref))
877 if (pud_none(*pud) || pud_page_vaddr(*pud) != pud_page_vaddr(*pud_ref))
880 pmd = pmd_offset(pud, address);
881 pmd_ref = pmd_offset(pud_ref, address);
882 if (pmd_none(*pmd_ref))
885 if (pmd_none(*pmd) || pmd_page(*pmd) != pmd_page(*pmd_ref))
888 pte_ref = pte_offset_kernel(pmd_ref, address);
889 if (!pte_present(*pte_ref))
892 pte = pte_offset_kernel(pmd, address);
895 * Don't use pte_page here, because the mappings can point
896 * outside mem_map, and the NUMA hash lookup cannot handle
899 if (!pte_present(*pte) || pte_pfn(*pte) != pte_pfn(*pte_ref))
906 int show_unhandled_signals = 1;
909 access_error(unsigned long error_code, int write, struct vm_area_struct *vma)
912 /* write, present and write, not present: */
913 if (unlikely(!(vma->vm_flags & VM_WRITE)))
919 if (unlikely(error_code & PF_PROT))
922 /* read, not present: */
923 if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE))))
929 static int fault_in_kernel_space(unsigned long address)
932 return address >= TASK_SIZE;
934 return address >= TASK_SIZE64;
939 * This routine handles page faults. It determines the address,
940 * and the problem, and then passes it off to one of the appropriate
946 void __kprobes do_page_fault(struct pt_regs *regs, unsigned long error_code)
948 struct vm_area_struct *vma;
949 struct task_struct *tsk;
950 unsigned long address;
951 struct mm_struct *mm;
958 prefetchw(&mm->mmap_sem);
960 /* Get the faulting address: */
961 address = read_cr2();
963 if (unlikely(kmmio_fault(regs, address)))
967 * We fault-in kernel-space virtual memory on-demand. The
968 * 'reference' page table is init_mm.pgd.
970 * NOTE! We MUST NOT take any locks for this case. We may
971 * be in an interrupt or a critical region, and should
972 * only copy the information from the master page table,
975 * This verifies that the fault happens in kernel space
976 * (error_code & 4) == 0, and that the fault was not a
977 * protection error (error_code & 9) == 0.
979 if (unlikely(fault_in_kernel_space(address))) {
980 if (!(error_code & (PF_RSVD|PF_USER|PF_PROT)) &&
981 vmalloc_fault(address) >= 0)
984 /* Can handle a stale RO->RW TLB: */
985 if (spurious_fault(error_code, address))
988 /* kprobes don't want to hook the spurious faults: */
989 if (notify_page_fault(regs))
992 * Don't take the mm semaphore here. If we fixup a prefetch
993 * fault we could otherwise deadlock:
995 bad_area_nosemaphore(regs, error_code, address);
1000 /* kprobes don't want to hook the spurious faults: */
1001 if (unlikely(notify_page_fault(regs)))
1004 * It's safe to allow irq's after cr2 has been saved and the
1005 * vmalloc fault has been handled.
1007 * User-mode registers count as a user access even for any
1008 * potential system fault or CPU buglet:
1010 if (user_mode_vm(regs)) {
1012 error_code |= PF_USER;
1014 if (regs->flags & X86_EFLAGS_IF)
1018 #ifdef CONFIG_X86_64
1019 if (unlikely(error_code & PF_RSVD))
1020 pgtable_bad(regs, error_code, address);
1024 * If we're in an interrupt, have no user context or are running
1025 * in an atomic region then we must not take the fault:
1027 if (unlikely(in_atomic() || !mm)) {
1028 bad_area_nosemaphore(regs, error_code, address);
1033 * When running in the kernel we expect faults to occur only to
1034 * addresses in user space. All other faults represent errors in
1035 * the kernel and should generate an OOPS. Unfortunately, in the
1036 * case of an erroneous fault occurring in a code path which already
1037 * holds mmap_sem we will deadlock attempting to validate the fault
1038 * against the address space. Luckily the kernel only validly
1039 * references user space from well defined areas of code, which are
1040 * listed in the exceptions table.
1042 * As the vast majority of faults will be valid we will only perform
1043 * the source reference check when there is a possibility of a
1044 * deadlock. Attempt to lock the address space, if we cannot we then
1045 * validate the source. If this is invalid we can skip the address
1046 * space check, thus avoiding the deadlock:
1048 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
1049 if ((error_code & PF_USER) == 0 &&
1050 !search_exception_tables(regs->ip)) {
1051 bad_area_nosemaphore(regs, error_code, address);
1054 down_read(&mm->mmap_sem);
1057 * The above down_read_trylock() might have succeeded in
1058 * which case we'll have missed the might_sleep() from
1064 vma = find_vma(mm, address);
1065 if (unlikely(!vma)) {
1066 bad_area(regs, error_code, address);
1069 if (likely(vma->vm_start <= address))
1071 if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) {
1072 bad_area(regs, error_code, address);
1075 if (error_code & PF_USER) {
1077 * Accessing the stack below %sp is always a bug.
1078 * The large cushion allows instructions like enter
1079 * and pusha to work. ("enter $65535, $31" pushes
1080 * 32 pointers and then decrements %sp by 65535.)
1082 if (unlikely(address + 65536 + 32 * sizeof(unsigned long) < regs->sp)) {
1083 bad_area(regs, error_code, address);
1087 if (unlikely(expand_stack(vma, address))) {
1088 bad_area(regs, error_code, address);
1093 * Ok, we have a good vm_area for this memory access, so
1094 * we can handle it..
1097 write = error_code & PF_WRITE;
1099 if (unlikely(access_error(error_code, write, vma))) {
1100 bad_area_access_error(regs, error_code, address);
1105 * If for any reason at all we couldn't handle the fault,
1106 * make sure we exit gracefully rather than endlessly redo
1109 fault = handle_mm_fault(mm, vma, address, write);
1111 if (unlikely(fault & VM_FAULT_ERROR)) {
1112 mm_fault_error(regs, error_code, address, fault);
1116 if (fault & VM_FAULT_MAJOR)
1121 check_v8086_mode(regs, address, tsk);
1123 up_read(&mm->mmap_sem);
1126 DEFINE_SPINLOCK(pgd_lock);
1127 LIST_HEAD(pgd_list);
1129 void vmalloc_sync_all(void)
1131 unsigned long address;
1133 #ifdef CONFIG_X86_32
1134 if (SHARED_KERNEL_PMD)
1137 for (address = VMALLOC_START & PMD_MASK;
1138 address >= TASK_SIZE && address < FIXADDR_TOP;
1139 address += PMD_SIZE) {
1141 unsigned long flags;
1144 spin_lock_irqsave(&pgd_lock, flags);
1145 list_for_each_entry(page, &pgd_list, lru) {
1146 if (!vmalloc_sync_one(page_address(page), address))
1149 spin_unlock_irqrestore(&pgd_lock, flags);
1151 #else /* CONFIG_X86_64 */
1152 for (address = VMALLOC_START & PGDIR_MASK; address <= VMALLOC_END;
1153 address += PGDIR_SIZE) {
1155 const pgd_t *pgd_ref = pgd_offset_k(address);
1156 unsigned long flags;
1159 if (pgd_none(*pgd_ref))
1162 spin_lock_irqsave(&pgd_lock, flags);
1163 list_for_each_entry(page, &pgd_list, lru) {
1165 pgd = (pgd_t *)page_address(page) + pgd_index(address);
1167 set_pgd(pgd, *pgd_ref);
1169 BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref));
1171 spin_unlock_irqrestore(&pgd_lock, flags);