aboutsummaryrefslogtreecommitdiffstats
path: root/arch/x86/mm/fault_64.c
diff options
context:
space:
mode:
authorHarvey Harrison <harvey.harrison@gmail.com>2008-01-30 07:34:11 -0500
committerIngo Molnar <mingo@elte.hu>2008-01-30 07:34:11 -0500
commitc61e211d9989e4c112d3d58db12ad58f9016a3c8 (patch)
tree17ac1e9220b9becda9bee059b4aa0bb129a56e1f /arch/x86/mm/fault_64.c
parentf8c2ee224d8397364835204c6c0130d08c2e644c (diff)
x86: unify fault_32|64.c
Unify includes in moved fault.c. Modify Makefiles to pick up unified file. Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com> Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Diffstat (limited to 'arch/x86/mm/fault_64.c')
-rw-r--r--arch/x86/mm/fault_64.c952
1 files changed, 0 insertions, 952 deletions
diff --git a/arch/x86/mm/fault_64.c b/arch/x86/mm/fault_64.c
deleted file mode 100644
index 0902719388bc..000000000000
--- a/arch/x86/mm/fault_64.c
+++ /dev/null
@@ -1,952 +0,0 @@
1/*
2 * Copyright (C) 1995 Linus Torvalds
3 * Copyright (C) 2001,2002 Andi Kleen, SuSE Labs.
4 */
5
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>
14#include <linux/mm.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>
26
27#include <asm/system.h>
28#include <asm/pgalloc.h>
29#include <asm/smp.h>
30#include <asm/tlbflush.h>
31#include <asm/proto.h>
32#include <asm-generic/sections.h>
33
34/*
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
41 */
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)
47
48static inline int notify_page_fault(struct pt_regs *regs)
49{
50#ifdef CONFIG_KPROBES
51 int ret = 0;
52
53 /* kprobe_running() needs smp_processor_id() */
54#ifdef CONFIG_X86_32
55 if (!user_mode_vm(regs)) {
56#else
57 if (!user_mode(regs)) {
58#endif
59 preempt_disable();
60 if (kprobe_running() && kprobe_fault_handler(regs, 14))
61 ret = 1;
62 preempt_enable();
63 }
64
65 return ret;
66#else
67 return 0;
68#endif
69}
70
71/*
72 * X86_32
73 * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch.
74 * Check that here and ignore it.
75 *
76 * X86_64
77 * Sometimes the CPU reports invalid exceptions on prefetch.
78 * Check that here and ignore it.
79 *
80 * Opcode checker based on code by Richard Brunner
81 */
82static int is_prefetch(struct pt_regs *regs, unsigned long addr,
83 unsigned long error_code)
84{
85 unsigned char *instr;
86 int scan_more = 1;
87 int prefetch = 0;
88 unsigned char *max_instr;
89
90#ifdef CONFIG_X86_32
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))
95 return 0;
96 } else {
97 return 0;
98 }
99#else
100 /* If it was a exec fault ignore */
101 if (error_code & PF_INSTR)
102 return 0;
103#endif
104
105 instr = (unsigned char *)convert_ip_to_linear(current, regs);
106 max_instr = instr + 15;
107
108 if (user_mode(regs) && instr >= (unsigned char *)TASK_SIZE)
109 return 0;
110
111 while (scan_more && instr < max_instr) {
112 unsigned char opcode;
113 unsigned char instr_hi;
114 unsigned char instr_lo;
115
116 if (probe_kernel_address(instr, opcode))
117 break;
118
119 instr_hi = opcode & 0xf0;
120 instr_lo = opcode & 0x0f;
121 instr++;
122
123 switch (instr_hi) {
124 case 0x20:
125 case 0x30:
126 /*
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
131 */
132 scan_more = ((instr_lo & 7) == 0x6);
133 break;
134#ifdef CONFIG_X86_64
135 case 0x40:
136 /*
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.
142 */
143 scan_more = (!user_mode(regs)) || (regs->cs == __USER_CS);
144 break;
145#endif
146 case 0x60:
147 /* 0x64 thru 0x67 are valid prefixes in all modes. */
148 scan_more = (instr_lo & 0xC) == 0x4;
149 break;
150 case 0xF0:
151 /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */
152 scan_more = !instr_lo || (instr_lo>>1) == 1;
153 break;
154 case 0x00:
155 /* Prefetch instruction is 0x0F0D or 0x0F18 */
156 scan_more = 0;
157
158 if (probe_kernel_address(instr, opcode))
159 break;
160 prefetch = (instr_lo == 0xF) &&
161 (opcode == 0x0D || opcode == 0x18);
162 break;
163 default:
164 scan_more = 0;
165 break;
166 }
167 }
168 return prefetch;
169}
170
171static void force_sig_info_fault(int si_signo, int si_code,
172 unsigned long address, struct task_struct *tsk)
173{
174 siginfo_t info;
175
176 info.si_signo = si_signo;
177 info.si_errno = 0;
178 info.si_code = si_code;
179 info.si_addr = (void __user *)address;
180 force_sig_info(si_signo, &info, tsk);
181}
182
183#ifdef CONFIG_X86_64
184static int bad_address(void *p)
185{
186 unsigned long dummy;
187 return probe_kernel_address((unsigned long *)p, dummy);
188}
189#endif
190
191void dump_pagetable(unsigned long address)
192{
193#ifdef CONFIG_X86_32
194 __typeof__(pte_val(__pte(0))) page;
195
196 page = read_cr3();
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) {
202 page &= PAGE_MASK;
203 page = ((__typeof__(page) *) __va(page))[(address >> PMD_SHIFT)
204 & (PTRS_PER_PMD - 1)];
205 printk(KERN_CONT "*pde = %016Lx ", page);
206 page &= ~_PAGE_NX;
207 }
208#else
209 printk("*pde = %08lx ", page);
210#endif
211
212 /*
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.
217 */
218 if ((page >> PAGE_SHIFT) < max_low_pfn
219 && (page & _PAGE_PRESENT)
220 && !(page & _PAGE_PSE)) {
221 page &= PAGE_MASK;
222 page = ((__typeof__(page) *) __va(page))[(address >> PAGE_SHIFT)
223 & (PTRS_PER_PTE - 1)];
224 printk("*pte = %0*Lx ", sizeof(page)*2, (u64)page);
225 }
226
227 printk("\n");
228#else /* CONFIG_X86_64 */
229 pgd_t *pgd;
230 pud_t *pud;
231 pmd_t *pmd;
232 pte_t *pte;
233
234 pgd = (pgd_t *)read_cr3();
235
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;
241
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;
246
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;
251
252 pte = pte_offset_kernel(pmd, address);
253 if (bad_address(pte)) goto bad;
254 printk("PTE %lx", pte_val(*pte));
255ret:
256 printk("\n");
257 return;
258bad:
259 printk("BAD\n");
260#endif
261}
262
263#ifdef CONFIG_X86_32
264static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address)
265{
266 unsigned index = pgd_index(address);
267 pgd_t *pgd_k;
268 pud_t *pud, *pud_k;
269 pmd_t *pmd, *pmd_k;
270
271 pgd += index;
272 pgd_k = init_mm.pgd + index;
273
274 if (!pgd_present(*pgd_k))
275 return NULL;
276
277 /*
278 * set_pgd(pgd, *pgd_k); here would be useless on PAE
279 * and redundant with the set_pmd() on non-PAE. As would
280 * set_pud.
281 */
282
283 pud = pud_offset(pgd, address);
284 pud_k = pud_offset(pgd_k, address);
285 if (!pud_present(*pud_k))
286 return NULL;
287
288 pmd = pmd_offset(pud, address);
289 pmd_k = pmd_offset(pud_k, address);
290 if (!pmd_present(*pmd_k))
291 return NULL;
292 if (!pmd_present(*pmd)) {
293 set_pmd(pmd, *pmd_k);
294 arch_flush_lazy_mmu_mode();
295 } else
296 BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k));
297 return pmd_k;
298}
299#endif
300
301#ifdef CONFIG_X86_64
302static const char errata93_warning[] =
303KERN_ERR "******* Your BIOS seems to not contain a fix for K8 errata #93\n"
304KERN_ERR "******* Working around it, but it may cause SEGVs or burn power.\n"
305KERN_ERR "******* Please consider a BIOS update.\n"
306KERN_ERR "******* Disabling USB legacy in the BIOS may also help.\n";
307#endif
308
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
317 */
318static int is_errata93(struct pt_regs *regs, unsigned long address)
319{
320#ifdef CONFIG_X86_64
321 static int warned;
322 if (address != regs->ip)
323 return 0;
324 if ((address >> 32) != 0)
325 return 0;
326 address |= 0xffffffffUL << 32;
327 if ((address >= (u64)_stext && address <= (u64)_etext) ||
328 (address >= MODULES_VADDR && address <= MODULES_END)) {
329 if (!warned) {
330 printk(errata93_warning);
331 warned = 1;
332 }
333 regs->ip = address;
334 return 1;
335 }
336#endif
337 return 0;
338}
339
340/*
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.
345 */
346static int is_errata100(struct pt_regs *regs, unsigned long address)
347{
348#ifdef CONFIG_X86_64
349 if ((regs->cs == __USER32_CS || (regs->cs & (1<<2))) &&
350 (address >> 32))
351 return 1;
352#endif
353 return 0;
354}
355
356void do_invalid_op(struct pt_regs *, unsigned long);
357
358static int is_f00f_bug(struct pt_regs *regs, unsigned long address)
359{
360#ifdef CONFIG_X86_F00F_BUG
361 unsigned long nr;
362 /*
363 * Pentium F0 0F C7 C8 bug workaround.
364 */
365 if (boot_cpu_data.f00f_bug) {
366 nr = (address - idt_descr.address) >> 3;
367
368 if (nr == 6) {
369 do_invalid_op(regs, 0);
370 return 1;
371 }
372 }
373#endif
374 return 0;
375}
376
377static void show_fault_oops(struct pt_regs *regs, unsigned long error_code,
378 unsigned long address)
379{
380#ifdef CONFIG_X86_32
381 if (!oops_may_print())
382 return;
383
384#ifdef CONFIG_X86_PAE
385 if (error_code & PF_INSTR) {
386 int level;
387 pte_t *pte = lookup_address(address, &level);
388
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);
393 }
394#endif
395 printk(KERN_ALERT "BUG: unable to handle kernel ");
396 if (address < PAGE_SIZE)
397 printk(KERN_CONT "NULL pointer dereference");
398 else
399 printk(KERN_CONT "paging request");
400 printk(KERN_CONT " at %08lx\n", address);
401
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");
409 else
410 printk(KERN_CONT "paging request");
411 printk(KERN_CONT " at %016lx\n", address);
412
413 printk(KERN_ALERT "IP:");
414 printk_address(regs->ip, 1);
415 dump_pagetable(address);
416#endif
417}
418
419#ifdef CONFIG_X86_64
420static noinline void pgtable_bad(unsigned long address, struct pt_regs *regs,
421 unsigned long error_code)
422{
423 unsigned long flags = oops_begin();
424 struct task_struct *tsk;
425
426 printk(KERN_ALERT "%s: Corrupted page table at address %lx\n",
427 current->comm, address);
428 dump_pagetable(address);
429 tsk = current;
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))
434 regs = NULL;
435 oops_end(flags, regs, SIGKILL);
436}
437#endif
438
439/*
440 * X86_32
441 * Handle a fault on the vmalloc or module mapping area
442 *
443 * X86_64
444 * Handle a fault on the vmalloc area
445 *
446 * This assumes no large pages in there.
447 */
448static int vmalloc_fault(unsigned long address)
449{
450#ifdef CONFIG_X86_32
451 unsigned long pgd_paddr;
452 pmd_t *pmd_k;
453 pte_t *pte_k;
454 /*
455 * Synchronize this task's top level page-table
456 * with the 'reference' page table.
457 *
458 * Do _not_ use "current" here. We might be inside
459 * an interrupt in the middle of a task switch..
460 */
461 pgd_paddr = read_cr3();
462 pmd_k = vmalloc_sync_one(__va(pgd_paddr), address);
463 if (!pmd_k)
464 return -1;
465 pte_k = pte_offset_kernel(pmd_k, address);
466 if (!pte_present(*pte_k))
467 return -1;
468 return 0;
469#else
470 pgd_t *pgd, *pgd_ref;
471 pud_t *pud, *pud_ref;
472 pmd_t *pmd, *pmd_ref;
473 pte_t *pte, *pte_ref;
474
475 /* Copy kernel mappings over when needed. This can also
476 happen within a race in page table update. In the later
477 case just flush. */
478
479 pgd = pgd_offset(current->mm ?: &init_mm, address);
480 pgd_ref = pgd_offset_k(address);
481 if (pgd_none(*pgd_ref))
482 return -1;
483 if (pgd_none(*pgd))
484 set_pgd(pgd, *pgd_ref);
485 else
486 BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref));
487
488 /* Below here mismatches are bugs because these lower tables
489 are shared */
490
491 pud = pud_offset(pgd, address);
492 pud_ref = pud_offset(pgd_ref, address);
493 if (pud_none(*pud_ref))
494 return -1;
495 if (pud_none(*pud) || pud_page_vaddr(*pud) != pud_page_vaddr(*pud_ref))
496 BUG();
497 pmd = pmd_offset(pud, address);
498 pmd_ref = pmd_offset(pud_ref, address);
499 if (pmd_none(*pmd_ref))
500 return -1;
501 if (pmd_none(*pmd) || pmd_page(*pmd) != pmd_page(*pmd_ref))
502 BUG();
503 pte_ref = pte_offset_kernel(pmd_ref, address);
504 if (!pte_present(*pte_ref))
505 return -1;
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
509 that. */
510 if (!pte_present(*pte) || pte_pfn(*pte) != pte_pfn(*pte_ref))
511 BUG();
512 return 0;
513#endif
514}
515
516int show_unhandled_signals = 1;
517
518/*
519 * This routine handles page faults. It determines the address,
520 * and the problem, and then passes it off to one of the appropriate
521 * routines.
522 */
523#ifdef CONFIG_X86_64
524asmlinkage
525#endif
526void __kprobes do_page_fault(struct pt_regs *regs, unsigned long error_code)
527{
528 struct task_struct *tsk;
529 struct mm_struct *mm;
530 struct vm_area_struct *vma;
531 unsigned long address;
532 int write, si_code;
533 int fault;
534#ifdef CONFIG_X86_64
535 unsigned long flags;
536#endif
537
538 /*
539 * We can fault from pretty much anywhere, with unknown IRQ state.
540 */
541 trace_hardirqs_fixup();
542
543 tsk = current;
544 mm = tsk->mm;
545 prefetchw(&mm->mmap_sem);
546
547 /* get the address */
548 address = read_cr2();
549
550 si_code = SEGV_MAPERR;
551
552 if (notify_page_fault(regs))
553 return;
554
555 /*
556 * We fault-in kernel-space virtual memory on-demand. The
557 * 'reference' page table is init_mm.pgd.
558 *
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,
562 * nothing more.
563 *
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.
567 */
568#ifdef CONFIG_X86_32
569 if (unlikely(address >= TASK_SIZE)) {
570 if (!(error_code & (PF_RSVD|PF_USER|PF_PROT)) &&
571 vmalloc_fault(address) >= 0)
572 return;
573 /*
574 * Don't take the mm semaphore here. If we fixup a prefetch
575 * fault we could otherwise deadlock.
576 */
577 goto bad_area_nosemaphore;
578 }
579
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))
583 local_irq_enable();
584
585 /*
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.
588 */
589 if (in_atomic() || !mm)
590 goto bad_area_nosemaphore;
591#else /* CONFIG_X86_64 */
592 if (unlikely(address >= TASK_SIZE64)) {
593 /*
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.
597 */
598 if (!(error_code & (PF_RSVD|PF_USER|PF_PROT)) &&
599 ((address >= VMALLOC_START && address < VMALLOC_END))) {
600 if (vmalloc_fault(address) >= 0)
601 return;
602 }
603 /*
604 * Don't take the mm semaphore here. If we fixup a prefetch
605 * fault we could otherwise deadlock.
606 */
607 goto bad_area_nosemaphore;
608 }
609 if (likely(regs->flags & X86_EFLAGS_IF))
610 local_irq_enable();
611
612 if (unlikely(error_code & PF_RSVD))
613 pgtable_bad(address, regs, error_code);
614
615 /*
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.
618 */
619 if (unlikely(in_atomic() || !mm))
620 goto bad_area_nosemaphore;
621
622 /*
623 * User-mode registers count as a user access even for any
624 * potential system fault or CPU buglet.
625 */
626 if (user_mode_vm(regs))
627 error_code |= PF_USER;
628again:
629#endif
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
637 * exceptions table.
638 *
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.
644 */
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);
650 }
651
652 vma = find_vma(mm, address);
653 if (!vma)
654 goto bad_area;
655#ifdef CONFIG_X86_32
656 if (vma->vm_start <= address)
657#else
658 if (likely(vma->vm_start <= address))
659#endif
660 goto good_area;
661 if (!(vma->vm_flags & VM_GROWSDOWN))
662 goto bad_area;
663 if (error_code & PF_USER) {
664 /*
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.)
669 */
670 if (address + 65536 + 32 * sizeof(unsigned long) < regs->sp)
671 goto bad_area;
672 }
673 if (expand_stack(vma, address))
674 goto bad_area;
675/*
676 * Ok, we have a good vm_area for this memory access, so
677 * we can handle it..
678 */
679good_area:
680 si_code = SEGV_ACCERR;
681 write = 0;
682 switch (error_code & (PF_PROT|PF_WRITE)) {
683 default: /* 3: write, present */
684 /* fall through */
685 case PF_WRITE: /* write, not present */
686 if (!(vma->vm_flags & VM_WRITE))
687 goto bad_area;
688 write++;
689 break;
690 case PF_PROT: /* read, present */
691 goto bad_area;
692 case 0: /* read, not present */
693 if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))
694 goto bad_area;
695 }
696
697#ifdef CONFIG_X86_32
698survive:
699#endif
700 /*
701 * If for any reason at all we couldn't handle the fault,
702 * make sure we exit gracefully rather than endlessly redo
703 * the fault.
704 */
705 fault = handle_mm_fault(mm, vma, address, write);
706 if (unlikely(fault & VM_FAULT_ERROR)) {
707 if (fault & VM_FAULT_OOM)
708 goto out_of_memory;
709 else if (fault & VM_FAULT_SIGBUS)
710 goto do_sigbus;
711 BUG();
712 }
713 if (fault & VM_FAULT_MAJOR)
714 tsk->maj_flt++;
715 else
716 tsk->min_flt++;
717
718#ifdef CONFIG_X86_32
719 /*
720 * Did it hit the DOS screen memory VA from vm86 mode?
721 */
722 if (v8086_mode(regs)) {
723 unsigned long bit = (address - 0xA0000) >> PAGE_SHIFT;
724 if (bit < 32)
725 tsk->thread.screen_bitmap |= 1 << bit;
726 }
727#endif
728 up_read(&mm->mmap_sem);
729 return;
730
731/*
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..
734 */
735bad_area:
736 up_read(&mm->mmap_sem);
737
738bad_area_nosemaphore:
739 /* User mode accesses just cause a SIGSEGV */
740 if (error_code & PF_USER) {
741 /*
742 * It's possible to have interrupts off here.
743 */
744 local_irq_enable();
745
746 /*
747 * Valid to do another page fault here because this one came
748 * from user space.
749 */
750 if (is_prefetch(regs, address, error_code))
751 return;
752
753 if (is_errata100(regs, address))
754 return;
755
756 if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
757 printk_ratelimit()) {
758 printk(
759#ifdef CONFIG_X86_32
760 "%s%s[%d]: segfault at %lx ip %08lx sp %08lx error %lx",
761#else
762 "%s%s[%d]: segfault at %lx ip %lx sp %lx error %lx",
763#endif
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);
768 printk("\n");
769 }
770
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);
776 return;
777 }
778
779 if (is_f00f_bug(regs, address))
780 return;
781
782no_context:
783 /* Are we prepared to handle this kernel fault? */
784 if (fixup_exception(regs))
785 return;
786
787 /*
788 * X86_32
789 * Valid to do another page fault here, because if this fault
790 * had been triggered by is_prefetch fixup_exception would have
791 * handled it.
792 *
793 * X86_64
794 * Hall of shame of CPU/BIOS bugs.
795 */
796 if (is_prefetch(regs, address, error_code))
797 return;
798
799 if (is_errata93(regs, address))
800 return;
801
802/*
803 * Oops. The kernel tried to access some bad page. We'll have to
804 * terminate things with extreme prejudice.
805 */
806#ifdef CONFIG_X86_32
807 bust_spinlocks(1);
808
809 show_fault_oops(regs, error_code, address);
810
811 tsk->thread.cr2 = address;
812 tsk->thread.trap_no = 14;
813 tsk->thread.error_code = error_code;
814 die("Oops", regs, error_code);
815 bust_spinlocks(0);
816 do_exit(SIGKILL);
817#else /* CONFIG_X86_64 */
818 flags = oops_begin();
819
820 show_fault_oops(regs, error_code, address);
821
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))
826 regs = NULL;
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);
830#endif
831
832/*
833 * We ran out of memory, or some other thing happened to us that made
834 * us unable to handle the page fault gracefully.
835 */
836out_of_memory:
837 up_read(&mm->mmap_sem);
838#ifdef CONFIG_X86_32
839 if (is_global_init(tsk)) {
840 yield();
841 down_read(&mm->mmap_sem);
842 goto survive;
843 }
844#else
845 if (is_global_init(current)) {
846 yield();
847 goto again;
848 }
849#endif
850 printk("VM: killing process %s\n", tsk->comm);
851 if (error_code & PF_USER)
852 do_group_exit(SIGKILL);
853 goto no_context;
854
855do_sigbus:
856 up_read(&mm->mmap_sem);
857
858 /* Kernel mode? Handle exceptions or die */
859 if (!(error_code & PF_USER))
860 goto no_context;
861#ifdef CONFIG_X86_32
862 /* User space => ok to do another page fault */
863 if (is_prefetch(regs, address, error_code))
864 return;
865#endif
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);
870}
871
872#ifdef CONFIG_X86_64
873DEFINE_SPINLOCK(pgd_lock);
874LIST_HEAD(pgd_list);
875#endif
876
877void vmalloc_sync_all(void)
878{
879#ifdef CONFIG_X86_32
880 /*
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
884 * if undone).
885 */
886 static DECLARE_BITMAP(insync, PTRS_PER_PGD);
887 static unsigned long start = TASK_SIZE;
888 unsigned long address;
889
890 if (SHARED_KERNEL_PMD)
891 return;
892
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)) {
896 unsigned long flags;
897 struct page *page;
898
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),
903 address)) {
904 BUG_ON(page != pgd_list);
905 break;
906 }
907 spin_unlock_irqrestore(&pgd_lock, flags);
908 if (!page)
909 set_bit(pgd_index(address), insync);
910 }
911 if (address == start && test_bit(pgd_index(address), insync))
912 start = address + PGDIR_SIZE;
913 }
914#else /* CONFIG_X86_64 */
915 /*
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
919 * if undone).
920 */
921 static DECLARE_BITMAP(insync, PTRS_PER_PGD);
922 static unsigned long start = VMALLOC_START & PGDIR_MASK;
923 unsigned long address;
924
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);
928 struct page *page;
929
930 if (pgd_none(*pgd_ref))
931 continue;
932 spin_lock(&pgd_lock);
933 list_for_each_entry(page, &pgd_list, lru) {
934 pgd_t *pgd;
935 pgd = (pgd_t *)page_address(page) + pgd_index(address);
936 if (pgd_none(*pgd))
937 set_pgd(pgd, *pgd_ref);
938 else
939 BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref));
940 }
941 spin_unlock(&pgd_lock);
942 set_bit(pgd_index(address), insync);
943 }
944 if (address == start)
945 start = address + PGDIR_SIZE;
946 }
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)));
951#endif
952}
generated by cgit v1.2.2 (git 2.25.0) at 2024-11-28 08:58:24 -0500