Linux kernel & device driver programming

Cross-Referenced Linux and Device Driver Code

[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ]
Version: [ 2.6.11.8 ] [ 2.6.25 ] [ 2.6.25.8 ] [ 2.6.31.13 ] Architecture: [ i386 ]
  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/highmem.h>
 22 #include <linux/bootmem.h>              /* for max_low_pfn */
 23 #include <linux/vmalloc.h>
 24 #include <linux/module.h>
 25 #include <linux/kprobes.h>
 26 #include <linux/uaccess.h>
 27 #include <linux/kdebug.h>
 28 #include <linux/ftrace.h>
 29 
 30 #include <asm/system.h>
 31 #include <asm/desc.h>
 32 #include <asm/segment.h>
 33 #include <asm/pgalloc.h>
 34 #include <asm/smp.h>
 35 #include <asm/tlbflush.h>
 36 #include <asm/proto.h>
 37 #include <asm-generic/sections.h>
 38 
 39 /*
 40  * Page fault error code bits
 41  *      bit 0 == 0 means no page found, 1 means protection fault
 42  *      bit 1 == 0 means read, 1 means write
 43  *      bit 2 == 0 means kernel, 1 means user-mode
 44  *      bit 3 == 1 means use of reserved bit detected
 45  *      bit 4 == 1 means fault was an instruction fetch
 46  */
 47 #define PF_PROT         (1<<0)
 48 #define PF_WRITE        (1<<1)
 49 #define PF_USER         (1<<2)
 50 #define PF_RSVD         (1<<3)
 51 #define PF_INSTR        (1<<4)
 52 
 53 static inline int notify_page_fault(struct pt_regs *regs)
 54 {
 55 #ifdef CONFIG_KPROBES
 56         int ret = 0;
 57 
 58         /* kprobe_running() needs smp_processor_id() */
 59 #ifdef CONFIG_X86_32
 60         if (!user_mode_vm(regs)) {
 61 #else
 62         if (!user_mode(regs)) {
 63 #endif
 64                 preempt_disable();
 65                 if (kprobe_running() && kprobe_fault_handler(regs, 14))
 66                         ret = 1;
 67                 preempt_enable();
 68         }
 69 
 70         return ret;
 71 #else
 72         return 0;
 73 #endif
 74 }
 75 
 76 /*
 77  * X86_32
 78  * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch.
 79  * Check that here and ignore it.
 80  *
 81  * X86_64
 82  * Sometimes the CPU reports invalid exceptions on prefetch.
 83  * Check that here and ignore it.
 84  *
 85  * Opcode checker based on code by Richard Brunner
 86  */
 87 static int is_prefetch(struct pt_regs *regs, unsigned long addr,
 88                        unsigned long error_code)
 89 {
 90         unsigned char *instr;
 91         int scan_more = 1;
 92         int prefetch = 0;
 93         unsigned char *max_instr;
 94 
 95         /*
 96          * If it was a exec (instruction fetch) fault on NX page, then
 97          * do not ignore the fault:
 98          */
 99         if (error_code & PF_INSTR)
100                 return 0;
101 
102         instr = (unsigned char *)convert_ip_to_linear(current, regs);
103         max_instr = instr + 15;
104 
105         if (user_mode(regs) && instr >= (unsigned char *)TASK_SIZE)
106                 return 0;
107 
108         while (scan_more && instr < max_instr) {
109                 unsigned char opcode;
110                 unsigned char instr_hi;
111                 unsigned char instr_lo;
112 
113                 if (probe_kernel_address(instr, opcode))
114                         break;
115 
116                 instr_hi = opcode & 0xf0;
117                 instr_lo = opcode & 0x0f;
118                 instr++;
119 
120                 switch (instr_hi) {
121                 case 0x20:
122                 case 0x30:
123                         /*
124                          * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes.
125                          * In X86_64 long mode, the CPU will signal invalid
126                          * opcode if some of these prefixes are present so
127                          * X86_64 will never get here anyway
128                          */
129                         scan_more = ((instr_lo & 7) == 0x6);
130                         break;
131 #ifdef CONFIG_X86_64
132                 case 0x40:
133                         /*
134                          * In AMD64 long mode 0x40..0x4F are valid REX prefixes
135                          * Need to figure out under what instruction mode the
136                          * instruction was issued. Could check the LDT for lm,
137                          * but for now it's good enough to assume that long
138                          * mode only uses well known segments or kernel.
139                          */
140                         scan_more = (!user_mode(regs)) || (regs->cs == __USER_CS);
141                         break;
142 #endif
143                 case 0x60:
144                         /* 0x64 thru 0x67 are valid prefixes in all modes. */
145                         scan_more = (instr_lo & 0xC) == 0x4;
146                         break;
147                 case 0xF0:
148                         /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */
149                         scan_more = !instr_lo || (instr_lo>>1) == 1;
150                         break;
151                 case 0x00:
152                         /* Prefetch instruction is 0x0F0D or 0x0F18 */
153                         scan_more = 0;
154 
155                         if (probe_kernel_address(instr, opcode))
156                                 break;
157                         prefetch = (instr_lo == 0xF) &&
158                                 (opcode == 0x0D || opcode == 0x18);
159                         break;
160                 default:
161                         scan_more = 0;
162                         break;
163                 }
164         }
165         return prefetch;
166 }
167 
168 static void force_sig_info_fault(int si_signo, int si_code,
169         unsigned long address, struct task_struct *tsk)
170 {
171         siginfo_t info;
172 
173         info.si_signo = si_signo;
174         info.si_errno = 0;
175         info.si_code = si_code;
176         info.si_addr = (void __user *)address;
177         force_sig_info(si_signo, &info, tsk);
178 }
179 
180 #ifdef CONFIG_X86_64
181 static int bad_address(void *p)
182 {
183         unsigned long dummy;
184         return probe_kernel_address((unsigned long *)p, dummy);
185 }
186 #endif
187 
188 static void dump_pagetable(unsigned long address)
189 {
190 #ifdef CONFIG_X86_32
191         __typeof__(pte_val(__pte(0))) page;
192 
193         page = read_cr3();
194         page = ((__typeof__(page) *) __va(page))[address >> PGDIR_SHIFT];
195 #ifdef CONFIG_X86_PAE
196         printk("*pdpt = %016Lx ", page);
197         if ((page >> PAGE_SHIFT) < max_low_pfn
198             && page & _PAGE_PRESENT) {
199                 page &= PAGE_MASK;
200                 page = ((__typeof__(page) *) __va(page))[(address >> PMD_SHIFT)
201                                                          & (PTRS_PER_PMD - 1)];
202                 printk(KERN_CONT "*pde = %016Lx ", page);
203                 page &= ~_PAGE_NX;
204         }
205 #else
206         printk("*pde = %08lx ", page);
207 #endif
208 
209         /*
210          * We must not directly access the pte in the highpte
211          * case if the page table is located in highmem.
212          * And let's rather not kmap-atomic the pte, just in case
213          * it's allocated already.
214          */
215         if ((page >> PAGE_SHIFT) < max_low_pfn
216             && (page & _PAGE_PRESENT)
217             && !(page & _PAGE_PSE)) {
218                 page &= PAGE_MASK;
219                 page = ((__typeof__(page) *) __va(page))[(address >> PAGE_SHIFT)
220                                                          & (PTRS_PER_PTE - 1)];
221                 printk("*pte = %0*Lx ", sizeof(page)*2, (u64)page);
222         }
223 
224         printk("\n");
225 #else /* CONFIG_X86_64 */
226         pgd_t *pgd;
227         pud_t *pud;
228         pmd_t *pmd;
229         pte_t *pte;
230 
231         pgd = (pgd_t *)read_cr3();
232 
233         pgd = __va((unsigned long)pgd & PHYSICAL_PAGE_MASK);
234         pgd += pgd_index(address);
235         if (bad_address(pgd)) goto bad;
236         printk("PGD %lx ", pgd_val(*pgd));
237         if (!pgd_present(*pgd)) goto ret;
238 
239         pud = pud_offset(pgd, address);
240         if (bad_address(pud)) goto bad;
241         printk("PUD %lx ", pud_val(*pud));
242         if (!pud_present(*pud) || pud_large(*pud))
243                 goto ret;
244 
245         pmd = pmd_offset(pud, address);
246         if (bad_address(pmd)) goto bad;
247         printk("PMD %lx ", pmd_val(*pmd));
248         if (!pmd_present(*pmd) || pmd_large(*pmd)) goto ret;
249 
250         pte = pte_offset_kernel(pmd, address);
251         if (bad_address(pte)) goto bad;
252         printk("PTE %lx", pte_val(*pte));
253 ret:
254         printk("\n");
255         return;
256 bad:
257         printk("BAD\n");
258 #endif
259 }
260 
261 #ifdef CONFIG_X86_32
262 static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address)
263 {
264         unsigned index = pgd_index(address);
265         pgd_t *pgd_k;
266         pud_t *pud, *pud_k;
267         pmd_t *pmd, *pmd_k;
268 
269         pgd += index;
270         pgd_k = init_mm.pgd + index;
271 
272         if (!pgd_present(*pgd_k))
273                 return NULL;
274 
275         /*
276          * set_pgd(pgd, *pgd_k); here would be useless on PAE
277          * and redundant with the set_pmd() on non-PAE. As would
278          * set_pud.
279          */
280 
281         pud = pud_offset(pgd, address);
282         pud_k = pud_offset(pgd_k, address);
283         if (!pud_present(*pud_k))
284                 return NULL;
285 
286         pmd = pmd_offset(pud, address);
287         pmd_k = pmd_offset(pud_k, address);
288         if (!pmd_present(*pmd_k))
289                 return NULL;
290         if (!pmd_present(*pmd)) {
291                 set_pmd(pmd, *pmd_k);
292                 arch_flush_lazy_mmu_mode();
293         } else
294                 BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k));
295         return pmd_k;
296 }
297 #endif
298 
299 #ifdef CONFIG_X86_64
300 static const char errata93_warning[] =
301 KERN_ERR "******* Your BIOS seems to not contain a fix for K8 errata #93\n"
302 KERN_ERR "******* Working around it, but it may cause SEGVs or burn power.\n"
303 KERN_ERR "******* Please consider a BIOS update.\n"
304 KERN_ERR "******* Disabling USB legacy in the BIOS may also help.\n";
305 #endif
306 
307 /* Workaround for K8 erratum #93 & buggy BIOS.
308    BIOS SMM functions are required to use a specific workaround
309    to avoid corruption of the 64bit RIP register on C stepping K8.
310    A lot of BIOS that didn't get tested properly miss this.
311    The OS sees this as a page fault with the upper 32bits of RIP cleared.
312    Try to work around it here.
313    Note we only handle faults in kernel here.
314    Does nothing for X86_32
315  */
316 static int is_errata93(struct pt_regs *regs, unsigned long address)
317 {
318 #ifdef CONFIG_X86_64
319         static int warned;
320         if (address != regs->ip)
321                 return 0;
322         if ((address >> 32) != 0)
323                 return 0;
324         address |= 0xffffffffUL << 32;
325         if ((address >= (u64)_stext && address <= (u64)_etext) ||
326             (address >= MODULES_VADDR && address <= MODULES_END)) {
327                 if (!warned) {
328                         printk(errata93_warning);
329                         warned = 1;
330                 }
331                 regs->ip = address;
332                 return 1;
333         }
334 #endif
335         return 0;
336 }
337 
338 /*
339  * Work around K8 erratum #100 K8 in compat mode occasionally jumps to illegal
340  * addresses >4GB.  We catch this in the page fault handler because these
341  * addresses are not reachable. Just detect this case and return.  Any code
342  * segment in LDT is compatibility mode.
343  */
344 static int is_errata100(struct pt_regs *regs, unsigned long address)
345 {
346 #ifdef CONFIG_X86_64
347         if ((regs->cs == __USER32_CS || (regs->cs & (1<<2))) &&
348             (address >> 32))
349                 return 1;
350 #endif
351         return 0;
352 }
353 
354 void do_invalid_op(struct pt_regs *, unsigned long);
355 
356 static int is_f00f_bug(struct pt_regs *regs, unsigned long address)
357 {
358 #ifdef CONFIG_X86_F00F_BUG
359         unsigned long nr;
360         /*
361          * Pentium F0 0F C7 C8 bug workaround.
362          */
363         if (boot_cpu_data.f00f_bug) {
364                 nr = (address - idt_descr.address) >> 3;
365 
366                 if (nr == 6) {
367                         zap_rt_locks();
368                         do_invalid_op(regs, 0);
369                         return 1;
370                 }
371         }
372 #endif
373         return 0;
374 }
375 
376 static void show_fault_oops(struct pt_regs *regs, unsigned long error_code,
377                             unsigned long address)
378 {
379 #ifdef CONFIG_X86_32
380         if (!oops_may_print())
381                 return;
382 #endif
383 
384 #ifdef CONFIG_X86_PAE
385         if (error_code & PF_INSTR) {
386                 unsigned 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 
396         printk(KERN_ALERT "BUG: unable to handle kernel ");
397         if (address < PAGE_SIZE)
398                 printk(KERN_CONT "NULL pointer dereference");
399         else
400                 printk(KERN_CONT "paging request");
401 #ifdef CONFIG_X86_32
402         printk(KERN_CONT " at %08lx\n", address);
403 #else
404         printk(KERN_CONT " at %016lx\n", address);
405 #endif
406         printk(KERN_ALERT "IP:");
407         printk_address(regs->ip, 1);
408         dump_pagetable(address);
409 }
410 
411 #ifdef CONFIG_X86_64
412 static noinline void pgtable_bad(unsigned long address, struct pt_regs *regs,
413                                  unsigned long error_code)
414 {
415         unsigned long flags = oops_begin();
416         struct task_struct *tsk;
417 
418         printk(KERN_ALERT "%s: Corrupted page table at address %lx\n",
419                current->comm, address);
420         dump_pagetable(address);
421         tsk = current;
422         tsk->thread.cr2 = address;
423         tsk->thread.trap_no = 14;
424         tsk->thread.error_code = error_code;
425         if (__die("Bad pagetable", regs, error_code))
426                 regs = NULL;
427         oops_end(flags, regs, SIGKILL);
428 }
429 #endif
430 
431 static int spurious_fault_check(unsigned long error_code, pte_t *pte)
432 {
433         if ((error_code & PF_WRITE) && !pte_write(*pte))
434                 return 0;
435         if ((error_code & PF_INSTR) && !pte_exec(*pte))
436                 return 0;
437 
438         return 1;
439 }
440 
441 /*
442  * Handle a spurious fault caused by a stale TLB entry.  This allows
443  * us to lazily refresh the TLB when increasing the permissions of a
444  * kernel page (RO -> RW or NX -> X).  Doing it eagerly is very
445  * expensive since that implies doing a full cross-processor TLB
446  * flush, even if no stale TLB entries exist on other processors.
447  * There are no security implications to leaving a stale TLB when
448  * increasing the permissions on a page.
449  */
450 static int spurious_fault(unsigned long address,
451                           unsigned long error_code)
452 {
453         pgd_t *pgd;
454         pud_t *pud;
455         pmd_t *pmd;
456         pte_t *pte;
457 
458         /* Reserved-bit violation or user access to kernel space? */
459         if (error_code & (PF_USER | PF_RSVD))
460                 return 0;
461 
462         pgd = init_mm.pgd + pgd_index(address);
463         if (!pgd_present(*pgd))
464                 return 0;
465 
466         pud = pud_offset(pgd, address);
467         if (!pud_present(*pud))
468                 return 0;
469 
470         if (pud_large(*pud))
471                 return spurious_fault_check(error_code, (pte_t *) pud);
472 
473         pmd = pmd_offset(pud, address);
474         if (!pmd_present(*pmd))
475                 return 0;
476 
477         if (pmd_large(*pmd))
478                 return spurious_fault_check(error_code, (pte_t *) pmd);
479 
480         pte = pte_offset_kernel(pmd, address);
481         if (!pte_present(*pte))
482                 return 0;
483 
484         return spurious_fault_check(error_code, pte);
485 }
486 
487 /*
488  * X86_32
489  * Handle a fault on the vmalloc or module mapping area
490  *
491  * X86_64
492  * Handle a fault on the vmalloc area
493  *
494  * This assumes no large pages in there.
495  */
496 static int vmalloc_fault(unsigned long address)
497 {
498 #ifdef CONFIG_X86_32
499         unsigned long pgd_paddr;
500         pmd_t *pmd_k;
501         pte_t *pte_k;
502         /*
503          * Synchronize this task's top level page-table
504          * with the 'reference' page table.
505          *
506          * Do _not_ use "current" here. We might be inside
507          * an interrupt in the middle of a task switch..
508          */
509         pgd_paddr = read_cr3();
510         pmd_k = vmalloc_sync_one(__va(pgd_paddr), address);
511         if (!pmd_k)
512                 return -1;
513         pte_k = pte_offset_kernel(pmd_k, address);
514         if (!pte_present(*pte_k))
515                 return -1;
516         return 0;
517 #else
518         pgd_t *pgd, *pgd_ref;
519         pud_t *pud, *pud_ref;
520         pmd_t *pmd, *pmd_ref;
521         pte_t *pte, *pte_ref;
522 
523         /* Make sure we are in vmalloc area */
524         if (!(address >= VMALLOC_START && address < VMALLOC_END))
525                 return -1;
526 
527         /* Copy kernel mappings over when needed. This can also
528            happen within a race in page table update. In the later
529            case just flush. */
530 
531         pgd = pgd_offset(current->mm ?: &init_mm, address);
532         pgd_ref = pgd_offset_k(address);
533         if (pgd_none(*pgd_ref))
534                 return -1;
535         if (pgd_none(*pgd))
536                 set_pgd(pgd, *pgd_ref);
537         else
538                 BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref));
539 
540         /* Below here mismatches are bugs because these lower tables
541            are shared */
542 
543         pud = pud_offset(pgd, address);
544         pud_ref = pud_offset(pgd_ref, address);
545         if (pud_none(*pud_ref))
546                 return -1;
547         if (pud_none(*pud) || pud_page_vaddr(*pud) != pud_page_vaddr(*pud_ref))
548                 BUG();
549         pmd = pmd_offset(pud, address);
550         pmd_ref = pmd_offset(pud_ref, address);
551         if (pmd_none(*pmd_ref))
552                 return -1;
553         if (pmd_none(*pmd) || pmd_page(*pmd) != pmd_page(*pmd_ref))
554                 BUG();
555         pte_ref = pte_offset_kernel(pmd_ref, address);
556         if (!pte_present(*pte_ref))
557                 return -1;
558         pte = pte_offset_kernel(pmd, address);
559         /* Don't use pte_page here, because the mappings can point
560            outside mem_map, and the NUMA hash lookup cannot handle
561            that. */
562         if (!pte_present(*pte) || pte_pfn(*pte) != pte_pfn(*pte_ref))
563                 BUG();
564         return 0;
565 #endif
566 }
567 
568 int show_unhandled_signals = 1;
569 
570 /*
571  * This routine handles page faults.  It determines the address,
572  * and the problem, and then passes it off to one of the appropriate
573  * routines.
574  */
575 #ifdef CONFIG_X86_64
576 asmlinkage
577 #endif
578 void __kprobes do_page_fault(struct pt_regs *regs, unsigned long error_code)
579 {
580         struct task_struct *tsk;
581         struct mm_struct *mm;
582         struct vm_area_struct *vma;
583         unsigned long address;
584         int write, si_code;
585         int fault;
586 #ifdef CONFIG_X86_64
587         unsigned long flags;
588 #endif
589 
590         /*
591          * We can fault from pretty much anywhere, with unknown IRQ state.
592          */
593         trace_hardirqs_fixup();
594 
595         tsk = current;
596         mm = tsk->mm;
597         prefetchw(&mm->mmap_sem);
598 
599         /* get the address */
600         address = read_cr2();
601 
602         ftrace_event_fault(regs->ip, error_code, address);
603 
604         si_code = SEGV_MAPERR;
605 
606         if (notify_page_fault(regs))
607                 return;
608 
609         /*
610          * We fault-in kernel-space virtual memory on-demand. The
611          * 'reference' page table is init_mm.pgd.
612          *
613          * NOTE! We MUST NOT take any locks for this case. We may
614          * be in an interrupt or a critical region, and should
615          * only copy the information from the master page table,
616          * nothing more.
617          *
618          * This verifies that the fault happens in kernel space
619          * (error_code & 4) == 0, and that the fault was not a
620          * protection error (error_code & 9) == 0.
621          */
622 #ifdef CONFIG_X86_32
623         if (unlikely(address >= TASK_SIZE)) {
624 #else
625         if (unlikely(address >= TASK_SIZE64)) {
626 #endif
627                 if (!(error_code & (PF_RSVD|PF_USER|PF_PROT)) &&
628                     vmalloc_fault(address) >= 0)
629                         return;
630 
631                 /* Can handle a stale RO->RW TLB */
632                 if (spurious_fault(address, error_code))
633                         return;
634 
635                 /*
636                  * Don't take the mm semaphore here. If we fixup a prefetch
637                  * fault we could otherwise deadlock.
638                  */
639                 goto bad_area_nosemaphore;
640         }
641 
642 
643 #ifdef CONFIG_X86_32
644         /* It's safe to allow irq's after cr2 has been saved and the vmalloc
645            fault has been handled. */
646         if (regs->flags & (X86_EFLAGS_IF|VM_MASK))
647                 local_irq_enable();
648 
649         /*
650          * If we're in an interrupt, have no user context or are running in an
651          * atomic region then we must not take the fault.
652          */
653         if (unlikely(in_atomic() || !mm || current->pagefault_disabled))
654                 goto bad_area_nosemaphore;
655 #else /* CONFIG_X86_64 */
656         if (likely(regs->flags & X86_EFLAGS_IF))
657                 local_irq_enable();
658 
659         if (unlikely(error_code & PF_RSVD))
660                 pgtable_bad(address, regs, error_code);
661 
662         /*
663          * If we're in an interrupt, have no user context or are running in an
664          * atomic region then we must not take the fault.
665          */
666         if (unlikely(in_atomic() || !mm))
667                 goto bad_area_nosemaphore;
668 
669         /*
670          * User-mode registers count as a user access even for any
671          * potential system fault or CPU buglet.
672          */
673         if (user_mode_vm(regs))
674                 error_code |= PF_USER;
675 again:
676 #endif
677         /* When running in the kernel we expect faults to occur only to
678          * addresses in user space.  All other faults represent errors in the
679          * kernel and should generate an OOPS.  Unfortunately, in the case of an
680          * erroneous fault occurring in a code path which already holds mmap_sem
681          * we will deadlock attempting to validate the fault against the
682          * address space.  Luckily the kernel only validly references user
683          * space from well defined areas of code, which are listed in the
684          * exceptions table.
685          *
686          * As the vast majority of faults will be valid we will only perform
687          * the source reference check when there is a possibility of a deadlock.
688          * Attempt to lock the address space, if we cannot we then validate the
689          * source.  If this is invalid we can skip the address space check,
690          * thus avoiding the deadlock.
691          */
692         if (!down_read_trylock(&mm->mmap_sem)) {
693                 if ((error_code & PF_USER) == 0 &&
694                     !search_exception_tables(regs->ip))
695                         goto bad_area_nosemaphore;
696                 down_read(&mm->mmap_sem);
697         }
698 
699         vma = find_vma(mm, address);
700         if (!vma)
701                 goto bad_area;
702         if (vma->vm_start <= address)
703                 goto good_area;
704         if (!(vma->vm_flags & VM_GROWSDOWN))
705                 goto bad_area;
706         if (error_code & PF_USER) {
707                 /*
708                  * Accessing the stack below %sp is always a bug.
709                  * The large cushion allows instructions like enter
710                  * and pusha to work.  ("enter $65535,$31" pushes
711                  * 32 pointers and then decrements %sp by 65535.)
712                  */
713                 if (address + 65536 + 32 * sizeof(unsigned long) < regs->sp)
714                         goto bad_area;
715         }
716         if (expand_stack(vma, address))
717                 goto bad_area;
718 /*
719  * Ok, we have a good vm_area for this memory access, so
720  * we can handle it..
721  */
722 good_area:
723         si_code = SEGV_ACCERR;
724         write = 0;
725         switch (error_code & (PF_PROT|PF_WRITE)) {
726         default:        /* 3: write, present */
727                 /* fall through */
728         case PF_WRITE:          /* write, not present */
729                 if (!(vma->vm_flags & VM_WRITE))
730                         goto bad_area;
731                 write++;
732                 break;
733         case PF_PROT:           /* read, present */
734                 goto bad_area;
735         case 0:                 /* read, not present */
736                 if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))
737                         goto bad_area;
738         }
739 
740 #ifdef CONFIG_X86_32
741 survive:
742 #endif
743         /*
744          * If for any reason at all we couldn't handle the fault,
745          * make sure we exit gracefully rather than endlessly redo
746          * the fault.
747          */
748         fault = handle_mm_fault(mm, vma, address, write);
749         if (unlikely(fault & VM_FAULT_ERROR)) {
750                 if (fault & VM_FAULT_OOM)
751                         goto out_of_memory;
752                 else if (fault & VM_FAULT_SIGBUS)
753                         goto do_sigbus;
754                 BUG();
755         }
756         if (fault & VM_FAULT_MAJOR)
757                 tsk->maj_flt++;
758         else
759                 tsk->min_flt++;
760 
761 #ifdef CONFIG_X86_32
762         /*
763          * Did it hit the DOS screen memory VA from vm86 mode?
764          */
765         if (v8086_mode(regs)) {
766                 unsigned long bit = (address - 0xA0000) >> PAGE_SHIFT;
767                 if (bit < 32)
768                         tsk->thread.screen_bitmap |= 1 << bit;
769         }
770 #endif
771         up_read(&mm->mmap_sem);
772         return;
773 
774 /*
775  * Something tried to access memory that isn't in our memory map..
776  * Fix it, but check if it's kernel or user first..
777  */
778 bad_area:
779         up_read(&mm->mmap_sem);
780 
781 bad_area_nosemaphore:
782         /* User mode accesses just cause a SIGSEGV */
783         if (error_code & PF_USER) {
784                 /*
785                  * It's possible to have interrupts off here.
786                  */
787                 local_irq_enable();
788 
789                 /*
790                  * Valid to do another page fault here because this one came
791                  * from user space.
792                  */
793                 if (is_prefetch(regs, address, error_code))
794                         return;
795 
796                 if (is_errata100(regs, address))
797                         return;
798 
799                 if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
800                     printk_ratelimit()) {
801                         printk(
802 #ifdef CONFIG_X86_32
803                         "%s%s[%d]: segfault at %lx ip %08lx sp %08lx error %lx",
804 #else
805                         "%s%s[%d]: segfault at %lx ip %lx sp %lx error %lx",
806 #endif
807                         task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG,
808                         tsk->comm, task_pid_nr(tsk), address, regs->ip,
809                         regs->sp, error_code);
810                         print_vma_addr(" in ", regs->ip);
811                         printk("\n");
812                 }
813 
814                 tsk->thread.cr2 = address;
815                 /* Kernel addresses are always protection faults */
816                 tsk->thread.error_code = error_code | (address >= TASK_SIZE);
817                 tsk->thread.trap_no = 14;
818                 force_sig_info_fault(SIGSEGV, si_code, address, tsk);
819                 return;
820         }
821 
822         if (is_f00f_bug(regs, address))
823                 return;
824 
825 no_context:
826         /* Are we prepared to handle this kernel fault?  */
827         if (fixup_exception(regs))
828                 return;
829 
830         /*
831          * X86_32
832          * Valid to do another page fault here, because if this fault
833          * had been triggered by is_prefetch fixup_exception would have
834          * handled it.
835          *
836          * X86_64
837          * Hall of shame of CPU/BIOS bugs.
838          */
839         if (is_prefetch(regs, address, error_code))
840                 return;
841 
842         if (is_errata93(regs, address))
843                 return;
844 
845 /*
846  * Oops. The kernel tried to access some bad page. We'll have to
847  * terminate things with extreme prejudice.
848  */
849 #ifdef CONFIG_X86_32
850         bust_spinlocks(1);
851 #else
852         flags = oops_begin();
853 #endif
854 
855         show_fault_oops(regs, error_code, address);
856 
857         tsk->thread.cr2 = address;
858         tsk->thread.trap_no = 14;
859         tsk->thread.error_code = error_code;
860 
861 #ifdef CONFIG_X86_32
862         die("Oops", regs, error_code);
863         bust_spinlocks(0);
864         do_exit(SIGKILL);
865 #else
866         if (__die("Oops", regs, error_code))
867                 regs = NULL;
868         /* Executive summary in case the body of the oops scrolled away */
869         printk(KERN_EMERG "CR2: %016lx\n", address);
870         oops_end(flags, regs, SIGKILL);
871 #endif
872 
873 /*
874  * We ran out of memory, or some other thing happened to us that made
875  * us unable to handle the page fault gracefully.
876  */
877 out_of_memory:
878         up_read(&mm->mmap_sem);
879         if (is_global_init(tsk)) {
880                 yield();
881 #ifdef CONFIG_X86_32
882                 down_read(&mm->mmap_sem);
883                 goto survive;
884 #else
885                 goto again;
886 #endif
887         }
888 
889         printk("VM: killing process %s\n", tsk->comm);
890         if (error_code & PF_USER)
891                 do_group_exit(SIGKILL);
892         goto no_context;
893 
894 do_sigbus:
895         up_read(&mm->mmap_sem);
896 
897         /* Kernel mode? Handle exceptions or die */
898         if (!(error_code & PF_USER))
899                 goto no_context;
900 #ifdef CONFIG_X86_32
901         /* User space => ok to do another page fault */
902         if (is_prefetch(regs, address, error_code))
903                 return;
904 #endif
905         tsk->thread.cr2 = address;
906         tsk->thread.error_code = error_code;
907         tsk->thread.trap_no = 14;
908         force_sig_info_fault(SIGBUS, BUS_ADRERR, address, tsk);
909 }
910 
911 DEFINE_SPINLOCK(pgd_lock);
912 LIST_HEAD(pgd_list);
913 
914 void vmalloc_sync_all(void)
915 {
916 #ifdef CONFIG_X86_32
917         /*
918          * Note that races in the updates of insync and start aren't
919          * problematic: insync can only get set bits added, and updates to
920          * start are only improving performance (without affecting correctness
921          * if undone).
922          */
923         static DECLARE_BITMAP(insync, PTRS_PER_PGD);
924         static unsigned long start = TASK_SIZE;
925         unsigned long address;
926 
927         if (SHARED_KERNEL_PMD)
928                 return;
929 
930         BUILD_BUG_ON(TASK_SIZE & ~PGDIR_MASK);
931         for (address = start; address >= TASK_SIZE; address += PGDIR_SIZE) {
932                 if (!test_bit(pgd_index(address), insync)) {
933                         unsigned long flags;
934                         struct page *page;
935 
936                         spin_lock_irqsave(&pgd_lock, flags);
937                         list_for_each_entry(page, &pgd_list, lru) {
938                                 if (!vmalloc_sync_one(page_address(page),
939                                                       address))
940                                         break;
941                         }
942                         spin_unlock_irqrestore(&pgd_lock, flags);
943                         if (!page)
944                                 set_bit(pgd_index(address), insync);
945                 }
946                 if (address == start && test_bit(pgd_index(address), insync))
947                         start = address + PGDIR_SIZE;
948         }
949 #else /* CONFIG_X86_64 */
950         /*
951          * Note that races in the updates of insync and start aren't
952          * problematic: insync can only get set bits added, and updates to
953          * start are only improving performance (without affecting correctness
954          * if undone).
955          */
956         static DECLARE_BITMAP(insync, PTRS_PER_PGD);
957         static unsigned long start = VMALLOC_START & PGDIR_MASK;
958         unsigned long address;
959 
960         for (address = start; address <= VMALLOC_END; address += PGDIR_SIZE) {
961                 if (!test_bit(pgd_index(address), insync)) {
962                         const pgd_t *pgd_ref = pgd_offset_k(address);
963                         unsigned long flags;
964                         struct page *page;
965 
966                         if (pgd_none(*pgd_ref))
967                                 continue;
968                         spin_lock_irqsave(&pgd_lock, flags);
969                         list_for_each_entry(page, &pgd_list, lru) {
970                                 pgd_t *pgd;
971                                 pgd = (pgd_t *)page_address(page) + pgd_index(address);
972                                 if (pgd_none(*pgd))
973                                         set_pgd(pgd, *pgd_ref);
974                                 else
975                                         BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref));
976                         }
977                         spin_unlock_irqrestore(&pgd_lock, flags);
978                         set_bit(pgd_index(address), insync);
979                 }
980                 if (address == start)
981                         start = address + PGDIR_SIZE;
982         }
983         /* Check that there is no need to do the same for the modules area. */
984         BUILD_BUG_ON(!(MODULES_VADDR > __START_KERNEL));
985         BUILD_BUG_ON(!(((MODULES_END - 1) & PGDIR_MASK) ==
986                                 (__START_KERNEL & PGDIR_MASK)));
987 #endif
988 }
989 
  This page was automatically generated by the LXR engine.