projects / deliverable / linux.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
Encode huge page size for VM_FAULT_HWPOISON errors
[deliverable/linux.git] / arch / x86 / mm / fault.c
... / ...
CommitLineData
1/*
2 * Copyright (C) 1995 Linus Torvalds
3 * Copyright (C) 2001, 2002 Andi Kleen, SuSE Labs.
4 * Copyright (C) 2008-2009, Red Hat Inc., Ingo Molnar
5 */
6#include <linux/magic.h> /* STACK_END_MAGIC */
7#include <linux/sched.h> /* test_thread_flag(), ... */
8#include <linux/kdebug.h> /* oops_begin/end, ... */
9#include <linux/module.h> /* search_exception_table */
10#include <linux/bootmem.h> /* max_low_pfn */
11#include <linux/kprobes.h> /* __kprobes, ... */
12#include <linux/mmiotrace.h> /* kmmio_handler, ... */
13#include <linux/perf_event.h> /* perf_sw_event */
14
15#include <asm/traps.h> /* dotraplinkage, ... */
16#include <asm/pgalloc.h> /* pgd_*(), ... */
17#include <asm/kmemcheck.h> /* kmemcheck_*(), ... */
18
19/*
20 * Page fault error code bits:
21 *
22 * bit 0 == 0: no page found 1: protection fault
23 * bit 1 == 0: read access 1: write access
24 * bit 2 == 0: kernel-mode access 1: user-mode access
25 * bit 3 == 1: use of reserved bit detected
26 * bit 4 == 1: fault was an instruction fetch
27 */
28enum x86_pf_error_code {
29
30 PF_PROT = 1 << 0,
31 PF_WRITE = 1 << 1,
32 PF_USER = 1 << 2,
33 PF_RSVD = 1 << 3,
34 PF_INSTR = 1 << 4,
35};
36
37/*
38 * Returns 0 if mmiotrace is disabled, or if the fault is not
39 * handled by mmiotrace:
40 */
41static inline int __kprobes
42kmmio_fault(struct pt_regs *regs, unsigned long addr)
43{
44 if (unlikely(is_kmmio_active()))
45 if (kmmio_handler(regs, addr) == 1)
46 return -1;
47 return 0;
48}
49
50static inline int __kprobes notify_page_fault(struct pt_regs *regs)
51{
52 int ret = 0;
53
54 /* kprobe_running() needs smp_processor_id() */
55 if (kprobes_built_in() && !user_mode_vm(regs)) {
56 preempt_disable();
57 if (kprobe_running() && kprobe_fault_handler(regs, 14))
58 ret = 1;
59 preempt_enable();
60 }
61
62 return ret;
63}
64
65/*
66 * Prefetch quirks:
67 *
68 * 32-bit mode:
69 *
70 * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch.
71 * Check that here and ignore it.
72 *
73 * 64-bit mode:
74 *
75 * Sometimes the CPU reports invalid exceptions on prefetch.
76 * Check that here and ignore it.
77 *
78 * Opcode checker based on code by Richard Brunner.
79 */
80static inline int
81check_prefetch_opcode(struct pt_regs *regs, unsigned char *instr,
82 unsigned char opcode, int *prefetch)
83{
84 unsigned char instr_hi = opcode & 0xf0;
85 unsigned char instr_lo = opcode & 0x0f;
86
87 switch (instr_hi) {
88 case 0x20:
89 case 0x30:
90 /*
91 * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes.
92 * In X86_64 long mode, the CPU will signal invalid
93 * opcode if some of these prefixes are present so
94 * X86_64 will never get here anyway
95 */
96 return ((instr_lo & 7) == 0x6);
97#ifdef CONFIG_X86_64
98 case 0x40:
99 /*
100 * In AMD64 long mode 0x40..0x4F are valid REX prefixes
101 * Need to figure out under what instruction mode the
102 * instruction was issued. Could check the LDT for lm,
103 * but for now it's good enough to assume that long
104 * mode only uses well known segments or kernel.
105 */
106 return (!user_mode(regs)) || (regs->cs == __USER_CS);
107#endif
108 case 0x60:
109 /* 0x64 thru 0x67 are valid prefixes in all modes. */
110 return (instr_lo & 0xC) == 0x4;
111 case 0xF0:
112 /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */
113 return !instr_lo || (instr_lo>>1) == 1;
114 case 0x00:
115 /* Prefetch instruction is 0x0F0D or 0x0F18 */
116 if (probe_kernel_address(instr, opcode))
117 return 0;
118
119 *prefetch = (instr_lo == 0xF) &&
120 (opcode == 0x0D || opcode == 0x18);
121 return 0;
122 default:
123 return 0;
124 }
125}
126
127static int
128is_prefetch(struct pt_regs *regs, unsigned long error_code, unsigned long addr)
129{
130 unsigned char *max_instr;
131 unsigned char *instr;
132 int prefetch = 0;
133
134 /*
135 * If it was a exec (instruction fetch) fault on NX page, then
136 * do not ignore the fault:
137 */
138 if (error_code & PF_INSTR)
139 return 0;
140
141 instr = (void *)convert_ip_to_linear(current, regs);
142 max_instr = instr + 15;
143
144 if (user_mode(regs) && instr >= (unsigned char *)TASK_SIZE)
145 return 0;
146
147 while (instr < max_instr) {
148 unsigned char opcode;
149
150 if (probe_kernel_address(instr, opcode))
151 break;
152
153 instr++;
154
155 if (!check_prefetch_opcode(regs, instr, opcode, &prefetch))
156 break;
157 }
158 return prefetch;
159}
160
161static void
162force_sig_info_fault(int si_signo, int si_code, unsigned long address,
163 struct task_struct *tsk)
164{
165 siginfo_t info;
166
167 info.si_signo = si_signo;
168 info.si_errno = 0;
169 info.si_code = si_code;
170 info.si_addr = (void __user *)address;
171 info.si_addr_lsb = si_code == BUS_MCEERR_AR ? PAGE_SHIFT : 0;
172
173 force_sig_info(si_signo, &info, tsk);
174}
175
176DEFINE_SPINLOCK(pgd_lock);
177LIST_HEAD(pgd_list);
178
179#ifdef CONFIG_X86_32
180static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address)
181{
182 unsigned index = pgd_index(address);
183 pgd_t *pgd_k;
184 pud_t *pud, *pud_k;
185 pmd_t *pmd, *pmd_k;
186
187 pgd += index;
188 pgd_k = init_mm.pgd + index;
189
190 if (!pgd_present(*pgd_k))
191 return NULL;
192
193 /*
194 * set_pgd(pgd, *pgd_k); here would be useless on PAE
195 * and redundant with the set_pmd() on non-PAE. As would
196 * set_pud.
197 */
198 pud = pud_offset(pgd, address);
199 pud_k = pud_offset(pgd_k, address);
200 if (!pud_present(*pud_k))
201 return NULL;
202
203 pmd = pmd_offset(pud, address);
204 pmd_k = pmd_offset(pud_k, address);
205 if (!pmd_present(*pmd_k))
206 return NULL;
207
208 if (!pmd_present(*pmd))
209 set_pmd(pmd, *pmd_k);
210 else
211 BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k));
212
213 return pmd_k;
214}
215
216void vmalloc_sync_all(void)
217{
218 unsigned long address;
219
220 if (SHARED_KERNEL_PMD)
221 return;
222
223 for (address = VMALLOC_START & PMD_MASK;
224 address >= TASK_SIZE && address < FIXADDR_TOP;
225 address += PMD_SIZE) {
226
227 unsigned long flags;
228 struct page *page;
229
230 spin_lock_irqsave(&pgd_lock, flags);
231 list_for_each_entry(page, &pgd_list, lru) {
232 if (!vmalloc_sync_one(page_address(page), address))
233 break;
234 }
235 spin_unlock_irqrestore(&pgd_lock, flags);
236 }
237}
238
239/*
240 * 32-bit:
241 *
242 * Handle a fault on the vmalloc or module mapping area
243 */
244static noinline __kprobes int vmalloc_fault(unsigned long address)
245{
246 unsigned long pgd_paddr;
247 pmd_t *pmd_k;
248 pte_t *pte_k;
249
250 /* Make sure we are in vmalloc area: */
251 if (!(address >= VMALLOC_START && address < VMALLOC_END))
252 return -1;
253
254 /*
255 * Synchronize this task's top level page-table
256 * with the 'reference' page table.
257 *
258 * Do _not_ use "current" here. We might be inside
259 * an interrupt in the middle of a task switch..
260 */
261 pgd_paddr = read_cr3();
262 pmd_k = vmalloc_sync_one(__va(pgd_paddr), address);
263 if (!pmd_k)
264 return -1;
265
266 pte_k = pte_offset_kernel(pmd_k, address);
267 if (!pte_present(*pte_k))
268 return -1;
269
270 return 0;
271}
272
273/*
274 * Did it hit the DOS screen memory VA from vm86 mode?
275 */
276static inline void
277check_v8086_mode(struct pt_regs *regs, unsigned long address,
278 struct task_struct *tsk)
279{
280 unsigned long bit;
281
282 if (!v8086_mode(regs))
283 return;
284
285 bit = (address - 0xA0000) >> PAGE_SHIFT;
286 if (bit < 32)
287 tsk->thread.screen_bitmap |= 1 << bit;
288}
289
290static bool low_pfn(unsigned long pfn)
291{
292 return pfn < max_low_pfn;
293}
294
295static void dump_pagetable(unsigned long address)
296{
297 pgd_t *base = __va(read_cr3());
298 pgd_t *pgd = &base[pgd_index(address)];
299 pmd_t *pmd;
300 pte_t *pte;
301
302#ifdef CONFIG_X86_PAE
303 printk("*pdpt = %016Lx ", pgd_val(*pgd));
304 if (!low_pfn(pgd_val(*pgd) >> PAGE_SHIFT) || !pgd_present(*pgd))
305 goto out;
306#endif
307 pmd = pmd_offset(pud_offset(pgd, address), address);
308 printk(KERN_CONT "*pde = %0*Lx ", sizeof(*pmd) * 2, (u64)pmd_val(*pmd));
309
310 /*
311 * We must not directly access the pte in the highpte
312 * case if the page table is located in highmem.
313 * And let's rather not kmap-atomic the pte, just in case
314 * it's allocated already:
315 */
316 if (!low_pfn(pmd_pfn(*pmd)) || !pmd_present(*pmd) || pmd_large(*pmd))
317 goto out;
318
319 pte = pte_offset_kernel(pmd, address);
320 printk("*pte = %0*Lx ", sizeof(*pte) * 2, (u64)pte_val(*pte));
321out:
322 printk("\n");
323}
324
325#else /* CONFIG_X86_64: */
326
327void vmalloc_sync_all(void)
328{
329 unsigned long address;
330
331 for (address = VMALLOC_START & PGDIR_MASK; address <= VMALLOC_END;
332 address += PGDIR_SIZE) {
333
334 const pgd_t *pgd_ref = pgd_offset_k(address);
335 unsigned long flags;
336 struct page *page;
337
338 if (pgd_none(*pgd_ref))
339 continue;
340
341 spin_lock_irqsave(&pgd_lock, flags);
342 list_for_each_entry(page, &pgd_list, lru) {
343 pgd_t *pgd;
344 pgd = (pgd_t *)page_address(page) + pgd_index(address);
345 if (pgd_none(*pgd))
346 set_pgd(pgd, *pgd_ref);
347 else
348 BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref));
349 }
350 spin_unlock_irqrestore(&pgd_lock, flags);
351 }
352}
353
354/*
355 * 64-bit:
356 *
357 * Handle a fault on the vmalloc area
358 *
359 * This assumes no large pages in there.
360 */
361static noinline __kprobes int vmalloc_fault(unsigned long address)
362{
363 pgd_t *pgd, *pgd_ref;
364 pud_t *pud, *pud_ref;
365 pmd_t *pmd, *pmd_ref;
366 pte_t *pte, *pte_ref;
367
368 /* Make sure we are in vmalloc area: */
369 if (!(address >= VMALLOC_START && address < VMALLOC_END))
370 return -1;
371
372 /*
373 * Copy kernel mappings over when needed. This can also
374 * happen within a race in page table update. In the later
375 * case just flush:
376 */
377 pgd = pgd_offset(current->active_mm, address);
378 pgd_ref = pgd_offset_k(address);
379 if (pgd_none(*pgd_ref))
380 return -1;
381
382 if (pgd_none(*pgd))
383 set_pgd(pgd, *pgd_ref);
384 else
385 BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref));
386
387 /*
388 * Below here mismatches are bugs because these lower tables
389 * are shared:
390 */
391
392 pud = pud_offset(pgd, address);
393 pud_ref = pud_offset(pgd_ref, address);
394 if (pud_none(*pud_ref))
395 return -1;
396
397 if (pud_none(*pud) || pud_page_vaddr(*pud) != pud_page_vaddr(*pud_ref))
398 BUG();
399
400 pmd = pmd_offset(pud, address);
401 pmd_ref = pmd_offset(pud_ref, address);
402 if (pmd_none(*pmd_ref))
403 return -1;
404
405 if (pmd_none(*pmd) || pmd_page(*pmd) != pmd_page(*pmd_ref))
406 BUG();
407
408 pte_ref = pte_offset_kernel(pmd_ref, address);
409 if (!pte_present(*pte_ref))
410 return -1;
411
412 pte = pte_offset_kernel(pmd, address);
413
414 /*
415 * Don't use pte_page here, because the mappings can point
416 * outside mem_map, and the NUMA hash lookup cannot handle
417 * that:
418 */
419 if (!pte_present(*pte) || pte_pfn(*pte) != pte_pfn(*pte_ref))
420 BUG();
421
422 return 0;
423}
424
425static const char errata93_warning[] =
426KERN_ERR
427"******* Your BIOS seems to not contain a fix for K8 errata #93\n"
428"******* Working around it, but it may cause SEGVs or burn power.\n"
429"******* Please consider a BIOS update.\n"
430"******* Disabling USB legacy in the BIOS may also help.\n";
431
432/*
433 * No vm86 mode in 64-bit mode:
434 */
435static inline void
436check_v8086_mode(struct pt_regs *regs, unsigned long address,
437 struct task_struct *tsk)
438{
439}
440
441static int bad_address(void *p)
442{
443 unsigned long dummy;
444
445 return probe_kernel_address((unsigned long *)p, dummy);
446}
447
448static void dump_pagetable(unsigned long address)
449{
450 pgd_t *base = __va(read_cr3() & PHYSICAL_PAGE_MASK);
451 pgd_t *pgd = base + pgd_index(address);
452 pud_t *pud;
453 pmd_t *pmd;
454 pte_t *pte;
455
456 if (bad_address(pgd))
457 goto bad;
458
459 printk("PGD %lx ", pgd_val(*pgd));
460
461 if (!pgd_present(*pgd))
462 goto out;
463
464 pud = pud_offset(pgd, address);
465 if (bad_address(pud))
466 goto bad;
467
468 printk("PUD %lx ", pud_val(*pud));
469 if (!pud_present(*pud) || pud_large(*pud))
470 goto out;
471
472 pmd = pmd_offset(pud, address);
473 if (bad_address(pmd))
474 goto bad;
475
476 printk("PMD %lx ", pmd_val(*pmd));
477 if (!pmd_present(*pmd) || pmd_large(*pmd))
478 goto out;
479
480 pte = pte_offset_kernel(pmd, address);
481 if (bad_address(pte))
482 goto bad;
483
484 printk("PTE %lx", pte_val(*pte));
485out:
486 printk("\n");
487 return;
488bad:
489 printk("BAD\n");
490}
491
492#endif /* CONFIG_X86_64 */
493
494/*
495 * Workaround for K8 erratum #93 & buggy BIOS.
496 *
497 * BIOS SMM functions are required to use a specific workaround
498 * to avoid corruption of the 64bit RIP register on C stepping K8.
499 *
500 * A lot of BIOS that didn't get tested properly miss this.
501 *
502 * The OS sees this as a page fault with the upper 32bits of RIP cleared.
503 * Try to work around it here.
504 *
505 * Note we only handle faults in kernel here.
506 * Does nothing on 32-bit.
507 */
508static int is_errata93(struct pt_regs *regs, unsigned long address)
509{
510#ifdef CONFIG_X86_64
511 if (address != regs->ip)
512 return 0;
513
514 if ((address >> 32) != 0)
515 return 0;
516
517 address |= 0xffffffffUL << 32;
518 if ((address >= (u64)_stext && address <= (u64)_etext) ||
519 (address >= MODULES_VADDR && address <= MODULES_END)) {
520 printk_once(errata93_warning);
521 regs->ip = address;
522 return 1;
523 }
524#endif
525 return 0;
526}
527
528/*
529 * Work around K8 erratum #100 K8 in compat mode occasionally jumps
530 * to illegal addresses >4GB.
531 *
532 * We catch this in the page fault handler because these addresses
533 * are not reachable. Just detect this case and return. Any code
534 * segment in LDT is compatibility mode.
535 */
536static int is_errata100(struct pt_regs *regs, unsigned long address)
537{
538#ifdef CONFIG_X86_64
539 if ((regs->cs == __USER32_CS || (regs->cs & (1<<2))) && (address >> 32))
540 return 1;
541#endif
542 return 0;
543}
544
545static int is_f00f_bug(struct pt_regs *regs, unsigned long address)
546{
547#ifdef CONFIG_X86_F00F_BUG
548 unsigned long nr;
549
550 /*
551 * Pentium F0 0F C7 C8 bug workaround:
552 */
553 if (boot_cpu_data.f00f_bug) {
554 nr = (address - idt_descr.address) >> 3;
555
556 if (nr == 6) {
557 do_invalid_op(regs, 0);
558 return 1;
559 }
560 }
561#endif
562 return 0;
563}
564
565static const char nx_warning[] = KERN_CRIT
566"kernel tried to execute NX-protected page - exploit attempt? (uid: %d)\n";
567
568static void
569show_fault_oops(struct pt_regs *regs, unsigned long error_code,
570 unsigned long address)
571{
572 if (!oops_may_print())
573 return;
574
575 if (error_code & PF_INSTR) {
576 unsigned int level;
577
578 pte_t *pte = lookup_address(address, &level);
579
580 if (pte && pte_present(*pte) && !pte_exec(*pte))
581 printk(nx_warning, current_uid());
582 }
583
584 printk(KERN_ALERT "BUG: unable to handle kernel ");
585 if (address < PAGE_SIZE)
586 printk(KERN_CONT "NULL pointer dereference");
587 else
588 printk(KERN_CONT "paging request");
589
590 printk(KERN_CONT " at %p\n", (void *) address);
591 printk(KERN_ALERT "IP:");
592 printk_address(regs->ip, 1);
593
594 dump_pagetable(address);
595}
596
597static noinline void
598pgtable_bad(struct pt_regs *regs, unsigned long error_code,
599 unsigned long address)
600{
601 struct task_struct *tsk;
602 unsigned long flags;
603 int sig;
604
605 flags = oops_begin();
606 tsk = current;
607 sig = SIGKILL;
608
609 printk(KERN_ALERT "%s: Corrupted page table at address %lx\n",
610 tsk->comm, address);
611 dump_pagetable(address);
612
613 tsk->thread.cr2 = address;
614 tsk->thread.trap_no = 14;
615 tsk->thread.error_code = error_code;
616
617 if (__die("Bad pagetable", regs, error_code))
618 sig = 0;
619
620 oops_end(flags, regs, sig);
621}
622
623static noinline void
624no_context(struct pt_regs *regs, unsigned long error_code,
625 unsigned long address)
626{
627 struct task_struct *tsk = current;
628 unsigned long *stackend;
629 unsigned long flags;
630 int sig;
631
632 /* Are we prepared to handle this kernel fault? */
633 if (fixup_exception(regs))
634 return;
635
636 /*
637 * 32-bit:
638 *
639 * Valid to do another page fault here, because if this fault
640 * had been triggered by is_prefetch fixup_exception would have
641 * handled it.
642 *
643 * 64-bit:
644 *
645 * Hall of shame of CPU/BIOS bugs.
646 */
647 if (is_prefetch(regs, error_code, address))
648 return;
649
650 if (is_errata93(regs, address))
651 return;
652
653 /*
654 * Oops. The kernel tried to access some bad page. We'll have to
655 * terminate things with extreme prejudice:
656 */
657 flags = oops_begin();
658
659 show_fault_oops(regs, error_code, address);
660
661 stackend = end_of_stack(tsk);
662 if (tsk != &init_task && *stackend != STACK_END_MAGIC)
663 printk(KERN_ALERT "Thread overran stack, or stack corrupted\n");
664
665 tsk->thread.cr2 = address;
666 tsk->thread.trap_no = 14;
667 tsk->thread.error_code = error_code;
668
669 sig = SIGKILL;
670 if (__die("Oops", regs, error_code))
671 sig = 0;
672
673 /* Executive summary in case the body of the oops scrolled away */
674 printk(KERN_EMERG "CR2: %016lx\n", address);
675
676 oops_end(flags, regs, sig);
677}
678
679/*
680 * Print out info about fatal segfaults, if the show_unhandled_signals
681 * sysctl is set:
682 */
683static inline void
684show_signal_msg(struct pt_regs *regs, unsigned long error_code,
685 unsigned long address, struct task_struct *tsk)
686{
687 if (!unhandled_signal(tsk, SIGSEGV))
688 return;
689
690 if (!printk_ratelimit())
691 return;
692
693 printk("%s%s[%d]: segfault at %lx ip %p sp %p error %lx",
694 task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG,
695 tsk->comm, task_pid_nr(tsk), address,
696 (void *)regs->ip, (void *)regs->sp, error_code);
697
698 print_vma_addr(KERN_CONT " in ", regs->ip);
699
700 printk(KERN_CONT "\n");
701}
702
703static void
704__bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
705 unsigned long address, int si_code)
706{
707 struct task_struct *tsk = current;
708
709 /* User mode accesses just cause a SIGSEGV */
710 if (error_code & PF_USER) {
711 /*
712 * It's possible to have interrupts off here:
713 */
714 local_irq_enable();
715
716 /*
717 * Valid to do another page fault here because this one came
718 * from user space:
719 */
720 if (is_prefetch(regs, error_code, address))
721 return;
722
723 if (is_errata100(regs, address))
724 return;
725
726 if (unlikely(show_unhandled_signals))
727 show_signal_msg(regs, error_code, address, tsk);
728
729 /* Kernel addresses are always protection faults: */
730 tsk->thread.cr2 = address;
731 tsk->thread.error_code = error_code | (address >= TASK_SIZE);
732 tsk->thread.trap_no = 14;
733
734 force_sig_info_fault(SIGSEGV, si_code, address, tsk);
735
736 return;
737 }
738
739 if (is_f00f_bug(regs, address))
740 return;
741
742 no_context(regs, error_code, address);
743}
744
745static noinline void
746bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
747 unsigned long address)
748{
749 __bad_area_nosemaphore(regs, error_code, address, SEGV_MAPERR);
750}
751
752static void
753__bad_area(struct pt_regs *regs, unsigned long error_code,
754 unsigned long address, int si_code)
755{
756 struct mm_struct *mm = current->mm;
757
758 /*
759 * Something tried to access memory that isn't in our memory map..
760 * Fix it, but check if it's kernel or user first..
761 */
762 up_read(&mm->mmap_sem);
763
764 __bad_area_nosemaphore(regs, error_code, address, si_code);
765}
766
767static noinline void
768bad_area(struct pt_regs *regs, unsigned long error_code, unsigned long address)
769{
770 __bad_area(regs, error_code, address, SEGV_MAPERR);
771}
772
773static noinline void
774bad_area_access_error(struct pt_regs *regs, unsigned long error_code,
775 unsigned long address)
776{
777 __bad_area(regs, error_code, address, SEGV_ACCERR);
778}
779
780/* TODO: fixup for "mm-invoke-oom-killer-from-page-fault.patch" */
781static void
782out_of_memory(struct pt_regs *regs, unsigned long error_code,
783 unsigned long address)
784{
785 /*
786 * We ran out of memory, call the OOM killer, and return the userspace
787 * (which will retry the fault, or kill us if we got oom-killed):
788 */
789 up_read(&current->mm->mmap_sem);
790
791 pagefault_out_of_memory();
792}
793
794static void
795do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address,
796 unsigned int fault)
797{
798 struct task_struct *tsk = current;
799 struct mm_struct *mm = tsk->mm;
800 int code = BUS_ADRERR;
801
802 up_read(&mm->mmap_sem);
803
804 /* Kernel mode? Handle exceptions or die: */
805 if (!(error_code & PF_USER)) {
806 no_context(regs, error_code, address);
807 return;
808 }
809
810 /* User-space => ok to do another page fault: */
811 if (is_prefetch(regs, error_code, address))
812 return;
813
814 tsk->thread.cr2 = address;
815 tsk->thread.error_code = error_code;
816 tsk->thread.trap_no = 14;
817
818#ifdef CONFIG_MEMORY_FAILURE
819 if (fault & VM_FAULT_HWPOISON) {
820 printk(KERN_ERR
821 "MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n",
822 tsk->comm, tsk->pid, address);
823 code = BUS_MCEERR_AR;
824 }
825#endif
826 force_sig_info_fault(SIGBUS, code, address, tsk);
827}
828
829static noinline void
830mm_fault_error(struct pt_regs *regs, unsigned long error_code,
831 unsigned long address, unsigned int fault)
832{
833 if (fault & VM_FAULT_OOM) {
834 out_of_memory(regs, error_code, address);
835 } else {
836 if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON))
837 do_sigbus(regs, error_code, address, fault);
838 else
839 BUG();
840 }
841}
842
843static int spurious_fault_check(unsigned long error_code, pte_t *pte)
844{
845 if ((error_code & PF_WRITE) && !pte_write(*pte))
846 return 0;
847
848 if ((error_code & PF_INSTR) && !pte_exec(*pte))
849 return 0;
850
851 return 1;
852}
853
854/*
855 * Handle a spurious fault caused by a stale TLB entry.
856 *
857 * This allows us to lazily refresh the TLB when increasing the
858 * permissions of a kernel page (RO -> RW or NX -> X). Doing it
859 * eagerly is very expensive since that implies doing a full
860 * cross-processor TLB flush, even if no stale TLB entries exist
861 * on other processors.
862 *
863 * There are no security implications to leaving a stale TLB when
864 * increasing the permissions on a page.
865 */
866static noinline __kprobes int
867spurious_fault(unsigned long error_code, unsigned long address)
868{
869 pgd_t *pgd;
870 pud_t *pud;
871 pmd_t *pmd;
872 pte_t *pte;
873 int ret;
874
875 /* Reserved-bit violation or user access to kernel space? */
876 if (error_code & (PF_USER | PF_RSVD))
877 return 0;
878
879 pgd = init_mm.pgd + pgd_index(address);
880 if (!pgd_present(*pgd))
881 return 0;
882
883 pud = pud_offset(pgd, address);
884 if (!pud_present(*pud))
885 return 0;
886
887 if (pud_large(*pud))
888 return spurious_fault_check(error_code, (pte_t *) pud);
889
890 pmd = pmd_offset(pud, address);
891 if (!pmd_present(*pmd))
892 return 0;
893
894 if (pmd_large(*pmd))
895 return spurious_fault_check(error_code, (pte_t *) pmd);
896
897 pte = pte_offset_kernel(pmd, address);
898 if (!pte_present(*pte))
899 return 0;
900
901 ret = spurious_fault_check(error_code, pte);
902 if (!ret)
903 return 0;
904
905 /*
906 * Make sure we have permissions in PMD.
907 * If not, then there's a bug in the page tables:
908 */
909 ret = spurious_fault_check(error_code, (pte_t *) pmd);
910 WARN_ONCE(!ret, "PMD has incorrect permission bits\n");
911
912 return ret;
913}
914
915int show_unhandled_signals = 1;
916
917static inline int
918access_error(unsigned long error_code, int write, struct vm_area_struct *vma)
919{
920 if (write) {
921 /* write, present and write, not present: */
922 if (unlikely(!(vma->vm_flags & VM_WRITE)))
923 return 1;
924 return 0;
925 }
926
927 /* read, present: */
928 if (unlikely(error_code & PF_PROT))
929 return 1;
930
931 /* read, not present: */
932 if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE))))
933 return 1;
934
935 return 0;
936}
937
938static int fault_in_kernel_space(unsigned long address)
939{
940 return address >= TASK_SIZE_MAX;
941}
942
943/*
944 * This routine handles page faults. It determines the address,
945 * and the problem, and then passes it off to one of the appropriate
946 * routines.
947 */
948dotraplinkage void __kprobes
949do_page_fault(struct pt_regs *regs, unsigned long error_code)
950{
951 struct vm_area_struct *vma;
952 struct task_struct *tsk;
953 unsigned long address;
954 struct mm_struct *mm;
955 int write;
956 int fault;
957
958 tsk = current;
959 mm = tsk->mm;
960
961 /* Get the faulting address: */
962 address = read_cr2();
963
964 /*
965 * Detect and handle instructions that would cause a page fault for
966 * both a tracked kernel page and a userspace page.
967 */
968 if (kmemcheck_active(regs))
969 kmemcheck_hide(regs);
970 prefetchw(&mm->mmap_sem);
971
972 if (unlikely(kmmio_fault(regs, address)))
973 return;
974
975 /*
976 * We fault-in kernel-space virtual memory on-demand. The
977 * 'reference' page table is init_mm.pgd.
978 *
979 * NOTE! We MUST NOT take any locks for this case. We may
980 * be in an interrupt or a critical region, and should
981 * only copy the information from the master page table,
982 * nothing more.
983 *
984 * This verifies that the fault happens in kernel space
985 * (error_code & 4) == 0, and that the fault was not a
986 * protection error (error_code & 9) == 0.
987 */
988 if (unlikely(fault_in_kernel_space(address))) {
989 if (!(error_code & (PF_RSVD | PF_USER | PF_PROT))) {
990 if (vmalloc_fault(address) >= 0)
991 return;
992
993 if (kmemcheck_fault(regs, address, error_code))
994 return;
995 }
996
997 /* Can handle a stale RO->RW TLB: */
998 if (spurious_fault(error_code, address))
999 return;
1000
1001 /* kprobes don't want to hook the spurious faults: */
1002 if (notify_page_fault(regs))
1003 return;
1004 /*
1005 * Don't take the mm semaphore here. If we fixup a prefetch
1006 * fault we could otherwise deadlock:
1007 */
1008 bad_area_nosemaphore(regs, error_code, address);
1009
1010 return;
1011 }
1012
1013 /* kprobes don't want to hook the spurious faults: */
1014 if (unlikely(notify_page_fault(regs)))
1015 return;
1016 /*
1017 * It's safe to allow irq's after cr2 has been saved and the
1018 * vmalloc fault has been handled.
1019 *
1020 * User-mode registers count as a user access even for any
1021 * potential system fault or CPU buglet:
1022 */
1023 if (user_mode_vm(regs)) {
1024 local_irq_enable();
1025 error_code |= PF_USER;
1026 } else {
1027 if (regs->flags & X86_EFLAGS_IF)
1028 local_irq_enable();
1029 }
1030
1031 if (unlikely(error_code & PF_RSVD))
1032 pgtable_bad(regs, error_code, address);
1033
1034 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, 0, regs, address);
1035
1036 /*
1037 * If we're in an interrupt, have no user context or are running
1038 * in an atomic region then we must not take the fault:
1039 */
1040 if (unlikely(in_atomic() || !mm)) {
1041 bad_area_nosemaphore(regs, error_code, address);
1042 return;
1043 }
1044
1045 /*
1046 * When running in the kernel we expect faults to occur only to
1047 * addresses in user space. All other faults represent errors in
1048 * the kernel and should generate an OOPS. Unfortunately, in the
1049 * case of an erroneous fault occurring in a code path which already
1050 * holds mmap_sem we will deadlock attempting to validate the fault
1051 * against the address space. Luckily the kernel only validly
1052 * references user space from well defined areas of code, which are
1053 * listed in the exceptions table.
1054 *
1055 * As the vast majority of faults will be valid we will only perform
1056 * the source reference check when there is a possibility of a
1057 * deadlock. Attempt to lock the address space, if we cannot we then
1058 * validate the source. If this is invalid we can skip the address
1059 * space check, thus avoiding the deadlock:
1060 */
1061 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
1062 if ((error_code & PF_USER) == 0 &&
1063 !search_exception_tables(regs->ip)) {
1064 bad_area_nosemaphore(regs, error_code, address);
1065 return;
1066 }
1067 down_read(&mm->mmap_sem);
1068 } else {
1069 /*
1070 * The above down_read_trylock() might have succeeded in
1071 * which case we'll have missed the might_sleep() from
1072 * down_read():
1073 */
1074 might_sleep();
1075 }
1076
1077 vma = find_vma(mm, address);
1078 if (unlikely(!vma)) {
1079 bad_area(regs, error_code, address);
1080 return;
1081 }
1082 if (likely(vma->vm_start <= address))
1083 goto good_area;
1084 if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) {
1085 bad_area(regs, error_code, address);
1086 return;
1087 }
1088 if (error_code & PF_USER) {
1089 /*
1090 * Accessing the stack below %sp is always a bug.
1091 * The large cushion allows instructions like enter
1092 * and pusha to work. ("enter $65535, $31" pushes
1093 * 32 pointers and then decrements %sp by 65535.)
1094 */
1095 if (unlikely(address + 65536 + 32 * sizeof(unsigned long) < regs->sp)) {
1096 bad_area(regs, error_code, address);
1097 return;
1098 }
1099 }
1100 if (unlikely(expand_stack(vma, address))) {
1101 bad_area(regs, error_code, address);
1102 return;
1103 }
1104
1105 /*
1106 * Ok, we have a good vm_area for this memory access, so
1107 * we can handle it..
1108 */
1109good_area:
1110 write = error_code & PF_WRITE;
1111
1112 if (unlikely(access_error(error_code, write, vma))) {
1113 bad_area_access_error(regs, error_code, address);
1114 return;
1115 }
1116
1117 /*
1118 * If for any reason at all we couldn't handle the fault,
1119 * make sure we exit gracefully rather than endlessly redo
1120 * the fault:
1121 */
1122 fault = handle_mm_fault(mm, vma, address, write ? FAULT_FLAG_WRITE : 0);
1123
1124 if (unlikely(fault & VM_FAULT_ERROR)) {
1125 mm_fault_error(regs, error_code, address, fault);
1126 return;
1127 }
1128
1129 if (fault & VM_FAULT_MAJOR) {
1130 tsk->maj_flt++;
1131 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, 0,
1132 regs, address);
1133 } else {
1134 tsk->min_flt++;
1135 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, 0,
1136 regs, address);
1137 }
1138
1139 check_v8086_mode(regs, address, tsk);
1140
1141 up_read(&mm->mmap_sem);
1142}
Linux kernel - Deliverables
This page took 0.030677 seconds and 5 git commands to generate.