2 * Copyright (C) 1995 Linus Torvalds
5 #include <linux/signal.h>
6 #include <linux/sched.h>
7 #include <linux/kernel.h>
8 #include <linux/errno.h>
9 #include <linux/string.h>
10 #include <linux/types.h>
11 #include <linux/ptrace.h>
12 #include <linux/mman.h>
14 #include <linux/smp.h>
15 #include <linux/interrupt.h>
16 #include <linux/init.h>
17 #include <linux/tty.h>
18 #include <linux/vt_kern.h> /* For unblank_screen() */
19 #include <linux/highmem.h>
20 #include <linux/bootmem.h> /* for max_low_pfn */
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>
29 #include <asm/segment.h>
32 * Page fault error code bits
33 * bit 0 == 0 means no page found, 1 means protection fault
34 * bit 1 == 0 means read, 1 means write
35 * bit 2 == 0 means kernel, 1 means user-mode
36 * bit 3 == 1 means use of reserved bit detected
37 * bit 4 == 1 means fault was an instruction fetch
39 #define PF_PROT (1<<0)
40 #define PF_WRITE (1<<1)
41 #define PF_USER (1<<2)
42 #define PF_RSVD (1<<3)
43 #define PF_INSTR (1<<4)
45 static inline int notify_page_fault(struct pt_regs *regs)
50 /* kprobe_running() needs smp_processor_id() */
51 if (!user_mode_vm(regs)) {
53 if (kprobe_running() && kprobe_fault_handler(regs, 14))
66 * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch.
67 * Check that here and ignore it.
70 * Sometimes the CPU reports invalid exceptions on prefetch.
71 * Check that here and ignore it.
73 * Opcode checker based on code by Richard Brunner
75 static int is_prefetch(struct pt_regs *regs, unsigned long addr,
76 unsigned long error_code)
81 unsigned char *max_instr;
85 if (unlikely(boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
86 boot_cpu_data.x86 >= 6)) {
87 /* Catch an obscure case of prefetch inside an NX page. */
88 if (nx_enabled && (error_code & PF_INSTR))
93 instr = (unsigned char *)get_segment_eip(regs, &limit);
95 /* If it was a exec fault ignore */
96 if (error_code & PF_INSTR)
98 instr = (unsigned char __user *)convert_rip_to_linear(current, regs);
101 max_instr = instr + 15;
104 if (user_mode(regs) && instr >= (unsigned char *)TASK_SIZE)
108 while (scan_more && instr < max_instr) {
109 unsigned char opcode;
110 unsigned char instr_hi;
111 unsigned char instr_lo;
114 if (instr > (unsigned char *)limit)
117 if (probe_kernel_address(instr, opcode))
120 instr_hi = opcode & 0xf0;
121 instr_lo = opcode & 0x0f;
128 * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes.
129 * In X86_64 long mode, the CPU will signal invalid
130 * opcode if some of these prefixes are present so
131 * X86_64 will never get here anyway
133 scan_more = ((instr_lo & 7) == 0x6);
138 * In AMD64 long mode 0x40..0x4F are valid REX prefixes
139 * Need to figure out under what instruction mode the
140 * instruction was issued. Could check the LDT for lm,
141 * but for now it's good enough to assume that long
142 * mode only uses well known segments or kernel.
144 scan_more = (!user_mode(regs)) || (regs->cs == __USER_CS);
148 /* 0x64 thru 0x67 are valid prefixes in all modes. */
149 scan_more = (instr_lo & 0xC) == 0x4;
152 /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */
153 scan_more = !instr_lo || (instr_lo>>1) == 1;
156 /* Prefetch instruction is 0x0F0D or 0x0F18 */
159 if (instr > (unsigned char *)limit)
162 if (probe_kernel_address(instr, opcode))
164 prefetch = (instr_lo == 0xF) &&
165 (opcode == 0x0D || opcode == 0x18);
175 static void force_sig_info_fault(int si_signo, int si_code,
176 unsigned long address, struct task_struct *tsk)
180 info.si_signo = si_signo;
182 info.si_code = si_code;
183 info.si_addr = (void __user *)address;
184 force_sig_info(si_signo, &info, tsk);
187 void do_invalid_op(struct pt_regs *, unsigned long);
189 static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address)
191 unsigned index = pgd_index(address);
197 pgd_k = init_mm.pgd + index;
199 if (!pgd_present(*pgd_k))
203 * set_pgd(pgd, *pgd_k); here would be useless on PAE
204 * and redundant with the set_pmd() on non-PAE. As would
208 pud = pud_offset(pgd, address);
209 pud_k = pud_offset(pgd_k, address);
210 if (!pud_present(*pud_k))
213 pmd = pmd_offset(pud, address);
214 pmd_k = pmd_offset(pud_k, address);
215 if (!pmd_present(*pmd_k))
217 if (!pmd_present(*pmd)) {
218 set_pmd(pmd, *pmd_k);
219 arch_flush_lazy_mmu_mode();
221 BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k));
226 static const char errata93_warning[] =
227 KERN_ERR "******* Your BIOS seems to not contain a fix for K8 errata #93\n"
228 KERN_ERR "******* Working around it, but it may cause SEGVs or burn power.\n"
229 KERN_ERR "******* Please consider a BIOS update.\n"
230 KERN_ERR "******* Disabling USB legacy in the BIOS may also help.\n";
232 /* Workaround for K8 erratum #93 & buggy BIOS.
233 BIOS SMM functions are required to use a specific workaround
234 to avoid corruption of the 64bit RIP register on C stepping K8.
235 A lot of BIOS that didn't get tested properly miss this.
236 The OS sees this as a page fault with the upper 32bits of RIP cleared.
237 Try to work around it here.
238 Note we only handle faults in kernel here. */
240 static int is_errata93(struct pt_regs *regs, unsigned long address)
243 if (address != regs->ip)
245 if ((address >> 32) != 0)
247 address |= 0xffffffffUL << 32;
248 if ((address >= (u64)_stext && address <= (u64)_etext) ||
249 (address >= MODULES_VADDR && address <= MODULES_END)) {
251 printk(errata93_warning);
262 * Handle a fault on the vmalloc or module mapping area
264 * This assumes no large pages in there.
266 static inline int vmalloc_fault(unsigned long address)
268 unsigned long pgd_paddr;
272 * Synchronize this task's top level page-table
273 * with the 'reference' page table.
275 * Do _not_ use "current" here. We might be inside
276 * an interrupt in the middle of a task switch..
278 pgd_paddr = read_cr3();
279 pmd_k = vmalloc_sync_one(__va(pgd_paddr), address);
282 pte_k = pte_offset_kernel(pmd_k, address);
283 if (!pte_present(*pte_k))
288 int show_unhandled_signals = 1;
291 * This routine handles page faults. It determines the address,
292 * and the problem, and then passes it off to one of the appropriate
295 void __kprobes do_page_fault(struct pt_regs *regs, unsigned long error_code)
297 struct task_struct *tsk;
298 struct mm_struct *mm;
299 struct vm_area_struct *vma;
300 unsigned long address;
305 * We can fault from pretty much anywhere, with unknown IRQ state.
307 trace_hardirqs_fixup();
309 /* get the address */
310 address = read_cr2();
314 si_code = SEGV_MAPERR;
317 * We fault-in kernel-space virtual memory on-demand. The
318 * 'reference' page table is init_mm.pgd.
320 * NOTE! We MUST NOT take any locks for this case. We may
321 * be in an interrupt or a critical region, and should
322 * only copy the information from the master page table,
325 * This verifies that the fault happens in kernel space
326 * (error_code & 4) == 0, and that the fault was not a
327 * protection error (error_code & 9) == 0.
329 if (unlikely(address >= TASK_SIZE)) {
330 if (!(error_code & (PF_RSVD|PF_USER|PF_PROT)) &&
331 vmalloc_fault(address) >= 0)
333 if (notify_page_fault(regs))
336 * Don't take the mm semaphore here. If we fixup a prefetch
337 * fault we could otherwise deadlock.
339 goto bad_area_nosemaphore;
342 if (notify_page_fault(regs))
345 /* It's safe to allow irq's after cr2 has been saved and the vmalloc
346 fault has been handled. */
347 if (regs->flags & (X86_EFLAGS_IF|VM_MASK))
353 * If we're in an interrupt, have no user context or are running in an
354 * atomic region then we must not take the fault.
356 if (in_atomic() || !mm)
357 goto bad_area_nosemaphore;
359 /* When running in the kernel we expect faults to occur only to
360 * addresses in user space. All other faults represent errors in the
361 * kernel and should generate an OOPS. Unfortunately, in the case of an
362 * erroneous fault occurring in a code path which already holds mmap_sem
363 * we will deadlock attempting to validate the fault against the
364 * address space. Luckily the kernel only validly references user
365 * space from well defined areas of code, which are listed in the
368 * As the vast majority of faults will be valid we will only perform
369 * the source reference check when there is a possibility of a deadlock.
370 * Attempt to lock the address space, if we cannot we then validate the
371 * source. If this is invalid we can skip the address space check,
372 * thus avoiding the deadlock.
374 if (!down_read_trylock(&mm->mmap_sem)) {
375 if ((error_code & PF_USER) == 0 &&
376 !search_exception_tables(regs->ip))
377 goto bad_area_nosemaphore;
378 down_read(&mm->mmap_sem);
381 vma = find_vma(mm, address);
384 if (vma->vm_start <= address)
386 if (!(vma->vm_flags & VM_GROWSDOWN))
388 if (error_code & PF_USER) {
390 * Accessing the stack below %sp is always a bug.
391 * The large cushion allows instructions like enter
392 * and pusha to work. ("enter $65535,$31" pushes
393 * 32 pointers and then decrements %sp by 65535.)
395 if (address + 65536 + 32 * sizeof(unsigned long) < regs->sp)
398 if (expand_stack(vma, address))
401 * Ok, we have a good vm_area for this memory access, so
405 si_code = SEGV_ACCERR;
407 switch (error_code & (PF_PROT|PF_WRITE)) {
408 default: /* 3: write, present */
410 case PF_WRITE: /* write, not present */
411 if (!(vma->vm_flags & VM_WRITE))
415 case PF_PROT: /* read, present */
417 case 0: /* read, not present */
418 if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))
424 * If for any reason at all we couldn't handle the fault,
425 * make sure we exit gracefully rather than endlessly redo
428 fault = handle_mm_fault(mm, vma, address, write);
429 if (unlikely(fault & VM_FAULT_ERROR)) {
430 if (fault & VM_FAULT_OOM)
432 else if (fault & VM_FAULT_SIGBUS)
436 if (fault & VM_FAULT_MAJOR)
442 * Did it hit the DOS screen memory VA from vm86 mode?
444 if (regs->flags & VM_MASK) {
445 unsigned long bit = (address - 0xA0000) >> PAGE_SHIFT;
447 tsk->thread.screen_bitmap |= 1 << bit;
449 up_read(&mm->mmap_sem);
453 * Something tried to access memory that isn't in our memory map..
454 * Fix it, but check if it's kernel or user first..
457 up_read(&mm->mmap_sem);
459 bad_area_nosemaphore:
460 /* User mode accesses just cause a SIGSEGV */
461 if (error_code & PF_USER) {
463 * It's possible to have interrupts off here.
468 * Valid to do another page fault here because this one came
471 if (is_prefetch(regs, address, error_code))
474 if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
475 printk_ratelimit()) {
476 printk("%s%s[%d]: segfault at %08lx ip %08lx "
477 "sp %08lx error %lx\n",
478 task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG,
479 tsk->comm, task_pid_nr(tsk), address, regs->ip,
480 regs->sp, error_code);
482 tsk->thread.cr2 = address;
483 /* Kernel addresses are always protection faults */
484 tsk->thread.error_code = error_code | (address >= TASK_SIZE);
485 tsk->thread.trap_no = 14;
486 force_sig_info_fault(SIGSEGV, si_code, address, tsk);
490 #ifdef CONFIG_X86_F00F_BUG
492 * Pentium F0 0F C7 C8 bug workaround.
494 if (boot_cpu_data.f00f_bug) {
497 nr = (address - idt_descr.address) >> 3;
500 do_invalid_op(regs, 0);
507 /* Are we prepared to handle this kernel fault? */
508 if (fixup_exception(regs))
512 * Valid to do another page fault here, because if this fault
513 * had been triggered by is_prefetch fixup_exception would have
516 if (is_prefetch(regs, address, error_code))
520 * Oops. The kernel tried to access some bad page. We'll have to
521 * terminate things with extreme prejudice.
526 if (oops_may_print()) {
527 __typeof__(pte_val(__pte(0))) page;
529 #ifdef CONFIG_X86_PAE
530 if (error_code & PF_INSTR) {
531 pte_t *pte = lookup_address(address);
533 if (pte && pte_present(*pte) && !pte_exec_kernel(*pte))
534 printk(KERN_CRIT "kernel tried to execute "
535 "NX-protected page - exploit attempt? "
536 "(uid: %d)\n", current->uid);
539 if (address < PAGE_SIZE)
540 printk(KERN_ALERT "BUG: unable to handle kernel NULL "
541 "pointer dereference");
543 printk(KERN_ALERT "BUG: unable to handle kernel paging"
545 printk(" at virtual address %08lx\n", address);
546 printk(KERN_ALERT "printing ip: %08lx ", regs->ip);
549 page = ((__typeof__(page) *) __va(page))[address >> PGDIR_SHIFT];
550 #ifdef CONFIG_X86_PAE
551 printk("*pdpt = %016Lx ", page);
552 if ((page >> PAGE_SHIFT) < max_low_pfn
553 && page & _PAGE_PRESENT) {
555 page = ((__typeof__(page) *) __va(page))[(address >> PMD_SHIFT)
556 & (PTRS_PER_PMD - 1)];
557 printk(KERN_CONT "*pde = %016Lx ", page);
561 printk("*pde = %08lx ", page);
565 * We must not directly access the pte in the highpte
566 * case if the page table is located in highmem.
567 * And let's rather not kmap-atomic the pte, just in case
568 * it's allocated already.
570 if ((page >> PAGE_SHIFT) < max_low_pfn
571 && (page & _PAGE_PRESENT)
572 && !(page & _PAGE_PSE)) {
574 page = ((__typeof__(page) *) __va(page))[(address >> PAGE_SHIFT)
575 & (PTRS_PER_PTE - 1)];
576 printk("*pte = %0*Lx ", sizeof(page)*2, (u64)page);
582 tsk->thread.cr2 = address;
583 tsk->thread.trap_no = 14;
584 tsk->thread.error_code = error_code;
585 die("Oops", regs, error_code);
590 * We ran out of memory, or some other thing happened to us that made
591 * us unable to handle the page fault gracefully.
594 up_read(&mm->mmap_sem);
595 if (is_global_init(tsk)) {
597 down_read(&mm->mmap_sem);
600 printk("VM: killing process %s\n", tsk->comm);
601 if (error_code & PF_USER)
602 do_group_exit(SIGKILL);
606 up_read(&mm->mmap_sem);
608 /* Kernel mode? Handle exceptions or die */
609 if (!(error_code & PF_USER))
612 /* User space => ok to do another page fault */
613 if (is_prefetch(regs, address, error_code))
616 tsk->thread.cr2 = address;
617 tsk->thread.error_code = error_code;
618 tsk->thread.trap_no = 14;
619 force_sig_info_fault(SIGBUS, BUS_ADRERR, address, tsk);
622 void vmalloc_sync_all(void)
625 * Note that races in the updates of insync and start aren't
626 * problematic: insync can only get set bits added, and updates to
627 * start are only improving performance (without affecting correctness
630 static DECLARE_BITMAP(insync, PTRS_PER_PGD);
631 static unsigned long start = TASK_SIZE;
632 unsigned long address;
634 if (SHARED_KERNEL_PMD)
637 BUILD_BUG_ON(TASK_SIZE & ~PGDIR_MASK);
638 for (address = start; address >= TASK_SIZE; address += PGDIR_SIZE) {
639 if (!test_bit(pgd_index(address), insync)) {
643 spin_lock_irqsave(&pgd_lock, flags);
644 for (page = pgd_list; page; page =
645 (struct page *)page->index)
646 if (!vmalloc_sync_one(page_address(page),
648 BUG_ON(page != pgd_list);
651 spin_unlock_irqrestore(&pgd_lock, flags);
653 set_bit(pgd_index(address), insync);
655 if (address == start && test_bit(pgd_index(address), insync))
656 start = address + PGDIR_SIZE;