~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/arch/x86/mm/fault.c

Version: ~ [ linux-5.2-rc4 ] ~ [ linux-5.1.9 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.50 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.125 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.181 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.181 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.68 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.39.4 ] ~ [ linux-2.6.38.8 ] ~ [ linux-2.6.37.6 ] ~ [ linux-2.6.36.4 ] ~ [ linux-2.6.35.14 ] ~ [ linux-2.6.34.15 ] ~ [ linux-2.6.33.20 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 // SPDX-License-Identifier: GPL-2.0
  2 /*
  3  *  Copyright (C) 1995  Linus Torvalds
  4  *  Copyright (C) 2001, 2002 Andi Kleen, SuSE Labs.
  5  *  Copyright (C) 2008-2009, Red Hat Inc., Ingo Molnar
  6  */
  7 #include <linux/sched.h>                /* test_thread_flag(), ...      */
  8 #include <linux/sched/task_stack.h>     /* task_stack_*(), ...          */
  9 #include <linux/kdebug.h>               /* oops_begin/end, ...          */
 10 #include <linux/extable.h>              /* search_exception_tables      */
 11 #include <linux/bootmem.h>              /* max_low_pfn                  */
 12 #include <linux/kprobes.h>              /* NOKPROBE_SYMBOL, ...         */
 13 #include <linux/mmiotrace.h>            /* kmmio_handler, ...           */
 14 #include <linux/perf_event.h>           /* perf_sw_event                */
 15 #include <linux/hugetlb.h>              /* hstate_index_to_shift        */
 16 #include <linux/prefetch.h>             /* prefetchw                    */
 17 #include <linux/context_tracking.h>     /* exception_enter(), ...       */
 18 #include <linux/uaccess.h>              /* faulthandler_disabled()      */
 19 
 20 #include <asm/cpufeature.h>             /* boot_cpu_has, ...            */
 21 #include <asm/traps.h>                  /* dotraplinkage, ...           */
 22 #include <asm/pgalloc.h>                /* pgd_*(), ...                 */
 23 #include <asm/fixmap.h>                 /* VSYSCALL_ADDR                */
 24 #include <asm/vsyscall.h>               /* emulate_vsyscall             */
 25 #include <asm/vm86.h>                   /* struct vm86                  */
 26 #include <asm/mmu_context.h>            /* vma_pkey()                   */
 27 
 28 #define CREATE_TRACE_POINTS
 29 #include <asm/trace/exceptions.h>
 30 
 31 /*
 32  * Returns 0 if mmiotrace is disabled, or if the fault is not
 33  * handled by mmiotrace:
 34  */
 35 static nokprobe_inline int
 36 kmmio_fault(struct pt_regs *regs, unsigned long addr)
 37 {
 38         if (unlikely(is_kmmio_active()))
 39                 if (kmmio_handler(regs, addr) == 1)
 40                         return -1;
 41         return 0;
 42 }
 43 
 44 static nokprobe_inline int kprobes_fault(struct pt_regs *regs)
 45 {
 46         int ret = 0;
 47 
 48         /* kprobe_running() needs smp_processor_id() */
 49         if (kprobes_built_in() && !user_mode(regs)) {
 50                 preempt_disable();
 51                 if (kprobe_running() && kprobe_fault_handler(regs, 14))
 52                         ret = 1;
 53                 preempt_enable();
 54         }
 55 
 56         return ret;
 57 }
 58 
 59 /*
 60  * Prefetch quirks:
 61  *
 62  * 32-bit mode:
 63  *
 64  *   Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch.
 65  *   Check that here and ignore it.
 66  *
 67  * 64-bit mode:
 68  *
 69  *   Sometimes the CPU reports invalid exceptions on prefetch.
 70  *   Check that here and ignore it.
 71  *
 72  * Opcode checker based on code by Richard Brunner.
 73  */
 74 static inline int
 75 check_prefetch_opcode(struct pt_regs *regs, unsigned char *instr,
 76                       unsigned char opcode, int *prefetch)
 77 {
 78         unsigned char instr_hi = opcode & 0xf0;
 79         unsigned char instr_lo = opcode & 0x0f;
 80 
 81         switch (instr_hi) {
 82         case 0x20:
 83         case 0x30:
 84                 /*
 85                  * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes.
 86                  * In X86_64 long mode, the CPU will signal invalid
 87                  * opcode if some of these prefixes are present so
 88                  * X86_64 will never get here anyway
 89                  */
 90                 return ((instr_lo & 7) == 0x6);
 91 #ifdef CONFIG_X86_64
 92         case 0x40:
 93                 /*
 94                  * In AMD64 long mode 0x40..0x4F are valid REX prefixes
 95                  * Need to figure out under what instruction mode the
 96                  * instruction was issued. Could check the LDT for lm,
 97                  * but for now it's good enough to assume that long
 98                  * mode only uses well known segments or kernel.
 99                  */
100                 return (!user_mode(regs) || user_64bit_mode(regs));
101 #endif
102         case 0x60:
103                 /* 0x64 thru 0x67 are valid prefixes in all modes. */
104                 return (instr_lo & 0xC) == 0x4;
105         case 0xF0:
106                 /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */
107                 return !instr_lo || (instr_lo>>1) == 1;
108         case 0x00:
109                 /* Prefetch instruction is 0x0F0D or 0x0F18 */
110                 if (probe_kernel_address(instr, opcode))
111                         return 0;
112 
113                 *prefetch = (instr_lo == 0xF) &&
114                         (opcode == 0x0D || opcode == 0x18);
115                 return 0;
116         default:
117                 return 0;
118         }
119 }
120 
121 static int
122 is_prefetch(struct pt_regs *regs, unsigned long error_code, unsigned long addr)
123 {
124         unsigned char *max_instr;
125         unsigned char *instr;
126         int prefetch = 0;
127 
128         /*
129          * If it was a exec (instruction fetch) fault on NX page, then
130          * do not ignore the fault:
131          */
132         if (error_code & X86_PF_INSTR)
133                 return 0;
134 
135         instr = (void *)convert_ip_to_linear(current, regs);
136         max_instr = instr + 15;
137 
138         if (user_mode(regs) && instr >= (unsigned char *)TASK_SIZE_MAX)
139                 return 0;
140 
141         while (instr < max_instr) {
142                 unsigned char opcode;
143 
144                 if (probe_kernel_address(instr, opcode))
145                         break;
146 
147                 instr++;
148 
149                 if (!check_prefetch_opcode(regs, instr, opcode, &prefetch))
150                         break;
151         }
152         return prefetch;
153 }
154 
155 /*
156  * A protection key fault means that the PKRU value did not allow
157  * access to some PTE.  Userspace can figure out what PKRU was
158  * from the XSAVE state, and this function fills out a field in
159  * siginfo so userspace can discover which protection key was set
160  * on the PTE.
161  *
162  * If we get here, we know that the hardware signaled a X86_PF_PK
163  * fault and that there was a VMA once we got in the fault
164  * handler.  It does *not* guarantee that the VMA we find here
165  * was the one that we faulted on.
166  *
167  * 1. T1   : mprotect_key(foo, PAGE_SIZE, pkey=4);
168  * 2. T1   : set PKRU to deny access to pkey=4, touches page
169  * 3. T1   : faults...
170  * 4.    T2: mprotect_key(foo, PAGE_SIZE, pkey=5);
171  * 5. T1   : enters fault handler, takes mmap_sem, etc...
172  * 6. T1   : reaches here, sees vma_pkey(vma)=5, when we really
173  *           faulted on a pte with its pkey=4.
174  */
175 static void fill_sig_info_pkey(int si_signo, int si_code, siginfo_t *info,
176                 u32 *pkey)
177 {
178         /* This is effectively an #ifdef */
179         if (!boot_cpu_has(X86_FEATURE_OSPKE))
180                 return;
181 
182         /* Fault not from Protection Keys: nothing to do */
183         if ((si_code != SEGV_PKUERR) || (si_signo != SIGSEGV))
184                 return;
185         /*
186          * force_sig_info_fault() is called from a number of
187          * contexts, some of which have a VMA and some of which
188          * do not.  The X86_PF_PK handing happens after we have a
189          * valid VMA, so we should never reach this without a
190          * valid VMA.
191          */
192         if (!pkey) {
193                 WARN_ONCE(1, "PKU fault with no VMA passed in");
194                 info->si_pkey = 0;
195                 return;
196         }
197         /*
198          * si_pkey should be thought of as a strong hint, but not
199          * absolutely guranteed to be 100% accurate because of
200          * the race explained above.
201          */
202         info->si_pkey = *pkey;
203 }
204 
205 static void
206 force_sig_info_fault(int si_signo, int si_code, unsigned long address,
207                      struct task_struct *tsk, u32 *pkey, int fault)
208 {
209         unsigned lsb = 0;
210         siginfo_t info;
211 
212         info.si_signo   = si_signo;
213         info.si_errno   = 0;
214         info.si_code    = si_code;
215         info.si_addr    = (void __user *)address;
216         if (fault & VM_FAULT_HWPOISON_LARGE)
217                 lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault)); 
218         if (fault & VM_FAULT_HWPOISON)
219                 lsb = PAGE_SHIFT;
220         info.si_addr_lsb = lsb;
221 
222         fill_sig_info_pkey(si_signo, si_code, &info, pkey);
223 
224         force_sig_info(si_signo, &info, tsk);
225 }
226 
227 DEFINE_SPINLOCK(pgd_lock);
228 LIST_HEAD(pgd_list);
229 
230 #ifdef CONFIG_X86_32
231 static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address)
232 {
233         unsigned index = pgd_index(address);
234         pgd_t *pgd_k;
235         p4d_t *p4d, *p4d_k;
236         pud_t *pud, *pud_k;
237         pmd_t *pmd, *pmd_k;
238 
239         pgd += index;
240         pgd_k = init_mm.pgd + index;
241 
242         if (!pgd_present(*pgd_k))
243                 return NULL;
244 
245         /*
246          * set_pgd(pgd, *pgd_k); here would be useless on PAE
247          * and redundant with the set_pmd() on non-PAE. As would
248          * set_p4d/set_pud.
249          */
250         p4d = p4d_offset(pgd, address);
251         p4d_k = p4d_offset(pgd_k, address);
252         if (!p4d_present(*p4d_k))
253                 return NULL;
254 
255         pud = pud_offset(p4d, address);
256         pud_k = pud_offset(p4d_k, address);
257         if (!pud_present(*pud_k))
258                 return NULL;
259 
260         pmd = pmd_offset(pud, address);
261         pmd_k = pmd_offset(pud_k, address);
262         if (!pmd_present(*pmd_k))
263                 return NULL;
264 
265         if (!pmd_present(*pmd))
266                 set_pmd(pmd, *pmd_k);
267         else
268                 BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k));
269 
270         return pmd_k;
271 }
272 
273 void vmalloc_sync_all(void)
274 {
275         unsigned long address;
276 
277         if (SHARED_KERNEL_PMD)
278                 return;
279 
280         for (address = VMALLOC_START & PMD_MASK;
281              address >= TASK_SIZE_MAX && address < FIXADDR_TOP;
282              address += PMD_SIZE) {
283                 struct page *page;
284 
285                 spin_lock(&pgd_lock);
286                 list_for_each_entry(page, &pgd_list, lru) {
287                         spinlock_t *pgt_lock;
288                         pmd_t *ret;
289 
290                         /* the pgt_lock only for Xen */
291                         pgt_lock = &pgd_page_get_mm(page)->page_table_lock;
292 
293                         spin_lock(pgt_lock);
294                         ret = vmalloc_sync_one(page_address(page), address);
295                         spin_unlock(pgt_lock);
296 
297                         if (!ret)
298                                 break;
299                 }
300                 spin_unlock(&pgd_lock);
301         }
302 }
303 
304 /*
305  * 32-bit:
306  *
307  *   Handle a fault on the vmalloc or module mapping area
308  */
309 static noinline int vmalloc_fault(unsigned long address)
310 {
311         unsigned long pgd_paddr;
312         pmd_t *pmd_k;
313         pte_t *pte_k;
314 
315         /* Make sure we are in vmalloc area: */
316         if (!(address >= VMALLOC_START && address < VMALLOC_END))
317                 return -1;
318 
319         WARN_ON_ONCE(in_nmi());
320 
321         /*
322          * Synchronize this task's top level page-table
323          * with the 'reference' page table.
324          *
325          * Do _not_ use "current" here. We might be inside
326          * an interrupt in the middle of a task switch..
327          */
328         pgd_paddr = read_cr3_pa();
329         pmd_k = vmalloc_sync_one(__va(pgd_paddr), address);
330         if (!pmd_k)
331                 return -1;
332 
333         if (pmd_large(*pmd_k))
334                 return 0;
335 
336         pte_k = pte_offset_kernel(pmd_k, address);
337         if (!pte_present(*pte_k))
338                 return -1;
339 
340         return 0;
341 }
342 NOKPROBE_SYMBOL(vmalloc_fault);
343 
344 /*
345  * Did it hit the DOS screen memory VA from vm86 mode?
346  */
347 static inline void
348 check_v8086_mode(struct pt_regs *regs, unsigned long address,
349                  struct task_struct *tsk)
350 {
351 #ifdef CONFIG_VM86
352         unsigned long bit;
353 
354         if (!v8086_mode(regs) || !tsk->thread.vm86)
355                 return;
356 
357         bit = (address - 0xA0000) >> PAGE_SHIFT;
358         if (bit < 32)
359                 tsk->thread.vm86->screen_bitmap |= 1 << bit;
360 #endif
361 }
362 
363 static bool low_pfn(unsigned long pfn)
364 {
365         return pfn < max_low_pfn;
366 }
367 
368 static void dump_pagetable(unsigned long address)
369 {
370         pgd_t *base = __va(read_cr3_pa());
371         pgd_t *pgd = &base[pgd_index(address)];
372         p4d_t *p4d;
373         pud_t *pud;
374         pmd_t *pmd;
375         pte_t *pte;
376 
377 #ifdef CONFIG_X86_PAE
378         pr_info("*pdpt = %016Lx ", pgd_val(*pgd));
379         if (!low_pfn(pgd_val(*pgd) >> PAGE_SHIFT) || !pgd_present(*pgd))
380                 goto out;
381 #define pr_pde pr_cont
382 #else
383 #define pr_pde pr_info
384 #endif
385         p4d = p4d_offset(pgd, address);
386         pud = pud_offset(p4d, address);
387         pmd = pmd_offset(pud, address);
388         pr_pde("*pde = %0*Lx ", sizeof(*pmd) * 2, (u64)pmd_val(*pmd));
389 #undef pr_pde
390 
391         /*
392          * We must not directly access the pte in the highpte
393          * case if the page table is located in highmem.
394          * And let's rather not kmap-atomic the pte, just in case
395          * it's allocated already:
396          */
397         if (!low_pfn(pmd_pfn(*pmd)) || !pmd_present(*pmd) || pmd_large(*pmd))
398                 goto out;
399 
400         pte = pte_offset_kernel(pmd, address);
401         pr_cont("*pte = %0*Lx ", sizeof(*pte) * 2, (u64)pte_val(*pte));
402 out:
403         pr_cont("\n");
404 }
405 
406 #else /* CONFIG_X86_64: */
407 
408 void vmalloc_sync_all(void)
409 {
410         sync_global_pgds(VMALLOC_START & PGDIR_MASK, VMALLOC_END);
411 }
412 
413 /*
414  * 64-bit:
415  *
416  *   Handle a fault on the vmalloc area
417  */
418 static noinline int vmalloc_fault(unsigned long address)
419 {
420         pgd_t *pgd, *pgd_ref;
421         p4d_t *p4d, *p4d_ref;
422         pud_t *pud, *pud_ref;
423         pmd_t *pmd, *pmd_ref;
424         pte_t *pte, *pte_ref;
425 
426         /* Make sure we are in vmalloc area: */
427         if (!(address >= VMALLOC_START && address < VMALLOC_END))
428                 return -1;
429 
430         WARN_ON_ONCE(in_nmi());
431 
432         /*
433          * Copy kernel mappings over when needed. This can also
434          * happen within a race in page table update. In the later
435          * case just flush:
436          */
437         pgd = (pgd_t *)__va(read_cr3_pa()) + pgd_index(address);
438         pgd_ref = pgd_offset_k(address);
439         if (pgd_none(*pgd_ref))
440                 return -1;
441 
442         if (CONFIG_PGTABLE_LEVELS > 4) {
443                 if (pgd_none(*pgd)) {
444                         set_pgd(pgd, *pgd_ref);
445                         arch_flush_lazy_mmu_mode();
446                 } else {
447                         BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref));
448                 }
449         }
450 
451         /* With 4-level paging, copying happens on the p4d level. */
452         p4d = p4d_offset(pgd, address);
453         p4d_ref = p4d_offset(pgd_ref, address);
454         if (p4d_none(*p4d_ref))
455                 return -1;
456 
457         if (p4d_none(*p4d) && CONFIG_PGTABLE_LEVELS == 4) {
458                 set_p4d(p4d, *p4d_ref);
459                 arch_flush_lazy_mmu_mode();
460         } else {
461                 BUG_ON(p4d_pfn(*p4d) != p4d_pfn(*p4d_ref));
462         }
463 
464         /*
465          * Below here mismatches are bugs because these lower tables
466          * are shared:
467          */
468         BUILD_BUG_ON(CONFIG_PGTABLE_LEVELS < 4);
469 
470         pud = pud_offset(p4d, address);
471         pud_ref = pud_offset(p4d_ref, address);
472         if (pud_none(*pud_ref))
473                 return -1;
474 
475         if (pud_none(*pud) || pud_pfn(*pud) != pud_pfn(*pud_ref))
476                 BUG();
477 
478         if (pud_large(*pud))
479                 return 0;
480 
481         pmd = pmd_offset(pud, address);
482         pmd_ref = pmd_offset(pud_ref, address);
483         if (pmd_none(*pmd_ref))
484                 return -1;
485 
486         if (pmd_none(*pmd) || pmd_pfn(*pmd) != pmd_pfn(*pmd_ref))
487                 BUG();
488 
489         if (pmd_large(*pmd))
490                 return 0;
491 
492         pte_ref = pte_offset_kernel(pmd_ref, address);
493         if (!pte_present(*pte_ref))
494                 return -1;
495 
496         pte = pte_offset_kernel(pmd, address);
497 
498         /*
499          * Don't use pte_page here, because the mappings can point
500          * outside mem_map, and the NUMA hash lookup cannot handle
501          * that:
502          */
503         if (!pte_present(*pte) || pte_pfn(*pte) != pte_pfn(*pte_ref))
504                 BUG();
505 
506         return 0;
507 }
508 NOKPROBE_SYMBOL(vmalloc_fault);
509 
510 #ifdef CONFIG_CPU_SUP_AMD
511 static const char errata93_warning[] =
512 KERN_ERR 
513 "******* Your BIOS seems to not contain a fix for K8 errata #93\n"
514 "******* Working around it, but it may cause SEGVs or burn power.\n"
515 "******* Please consider a BIOS update.\n"
516 "******* Disabling USB legacy in the BIOS may also help.\n";
517 #endif
518 
519 /*
520  * No vm86 mode in 64-bit mode:
521  */
522 static inline void
523 check_v8086_mode(struct pt_regs *regs, unsigned long address,
524                  struct task_struct *tsk)
525 {
526 }
527 
528 static int bad_address(void *p)
529 {
530         unsigned long dummy;
531 
532         return probe_kernel_address((unsigned long *)p, dummy);
533 }
534 
535 static void dump_pagetable(unsigned long address)
536 {
537         pgd_t *base = __va(read_cr3_pa());
538         pgd_t *pgd = base + pgd_index(address);
539         p4d_t *p4d;
540         pud_t *pud;
541         pmd_t *pmd;
542         pte_t *pte;
543 
544         if (bad_address(pgd))
545                 goto bad;
546 
547         pr_info("PGD %lx ", pgd_val(*pgd));
548 
549         if (!pgd_present(*pgd))
550                 goto out;
551 
552         p4d = p4d_offset(pgd, address);
553         if (bad_address(p4d))
554                 goto bad;
555 
556         pr_cont("P4D %lx ", p4d_val(*p4d));
557         if (!p4d_present(*p4d) || p4d_large(*p4d))
558                 goto out;
559 
560         pud = pud_offset(p4d, address);
561         if (bad_address(pud))
562                 goto bad;
563 
564         pr_cont("PUD %lx ", pud_val(*pud));
565         if (!pud_present(*pud) || pud_large(*pud))
566                 goto out;
567 
568         pmd = pmd_offset(pud, address);
569         if (bad_address(pmd))
570                 goto bad;
571 
572         pr_cont("PMD %lx ", pmd_val(*pmd));
573         if (!pmd_present(*pmd) || pmd_large(*pmd))
574                 goto out;
575 
576         pte = pte_offset_kernel(pmd, address);
577         if (bad_address(pte))
578                 goto bad;
579 
580         pr_cont("PTE %lx", pte_val(*pte));
581 out:
582         pr_cont("\n");
583         return;
584 bad:
585         pr_info("BAD\n");
586 }
587 
588 #endif /* CONFIG_X86_64 */
589 
590 /*
591  * Workaround for K8 erratum #93 & buggy BIOS.
592  *
593  * BIOS SMM functions are required to use a specific workaround
594  * to avoid corruption of the 64bit RIP register on C stepping K8.
595  *
596  * A lot of BIOS that didn't get tested properly miss this.
597  *
598  * The OS sees this as a page fault with the upper 32bits of RIP cleared.
599  * Try to work around it here.
600  *
601  * Note we only handle faults in kernel here.
602  * Does nothing on 32-bit.
603  */
604 static int is_errata93(struct pt_regs *regs, unsigned long address)
605 {
606 #if defined(CONFIG_X86_64) && defined(CONFIG_CPU_SUP_AMD)
607         if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD
608             || boot_cpu_data.x86 != 0xf)
609                 return 0;
610 
611         if (address != regs->ip)
612                 return 0;
613 
614         if ((address >> 32) != 0)
615                 return 0;
616 
617         address |= 0xffffffffUL << 32;
618         if ((address >= (u64)_stext && address <= (u64)_etext) ||
619             (address >= MODULES_VADDR && address <= MODULES_END)) {
620                 printk_once(errata93_warning);
621                 regs->ip = address;
622                 return 1;
623         }
624 #endif
625         return 0;
626 }
627 
628 /*
629  * Work around K8 erratum #100 K8 in compat mode occasionally jumps
630  * to illegal addresses >4GB.
631  *
632  * We catch this in the page fault handler because these addresses
633  * are not reachable. Just detect this case and return.  Any code
634  * segment in LDT is compatibility mode.
635  */
636 static int is_errata100(struct pt_regs *regs, unsigned long address)
637 {
638 #ifdef CONFIG_X86_64
639         if ((regs->cs == __USER32_CS || (regs->cs & (1<<2))) && (address >> 32))
640                 return 1;
641 #endif
642         return 0;
643 }
644 
645 static int is_f00f_bug(struct pt_regs *regs, unsigned long address)
646 {
647 #ifdef CONFIG_X86_F00F_BUG
648         unsigned long nr;
649 
650         /*
651          * Pentium F0 0F C7 C8 bug workaround:
652          */
653         if (boot_cpu_has_bug(X86_BUG_F00F)) {
654                 nr = (address - idt_descr.address) >> 3;
655 
656                 if (nr == 6) {
657                         do_invalid_op(regs, 0);
658                         return 1;
659                 }
660         }
661 #endif
662         return 0;
663 }
664 
665 static const char nx_warning[] = KERN_CRIT
666 "kernel tried to execute NX-protected page - exploit attempt? (uid: %d)\n";
667 static const char smep_warning[] = KERN_CRIT
668 "unable to execute userspace code (SMEP?) (uid: %d)\n";
669 
670 static void
671 show_fault_oops(struct pt_regs *regs, unsigned long error_code,
672                 unsigned long address)
673 {
674         if (!oops_may_print())
675                 return;
676 
677         if (error_code & X86_PF_INSTR) {
678                 unsigned int level;
679                 pgd_t *pgd;
680                 pte_t *pte;
681 
682                 pgd = __va(read_cr3_pa());
683                 pgd += pgd_index(address);
684 
685                 pte = lookup_address_in_pgd(pgd, address, &level);
686 
687                 if (pte && pte_present(*pte) && !pte_exec(*pte))
688                         printk(nx_warning, from_kuid(&init_user_ns, current_uid()));
689                 if (pte && pte_present(*pte) && pte_exec(*pte) &&
690                                 (pgd_flags(*pgd) & _PAGE_USER) &&
691                                 (__read_cr4() & X86_CR4_SMEP))
692                         printk(smep_warning, from_kuid(&init_user_ns, current_uid()));
693         }
694 
695         printk(KERN_ALERT "BUG: unable to handle kernel ");
696         if (address < PAGE_SIZE)
697                 printk(KERN_CONT "NULL pointer dereference");
698         else
699                 printk(KERN_CONT "paging request");
700 
701         printk(KERN_CONT " at %px\n", (void *) address);
702         printk(KERN_ALERT "IP: %pS\n", (void *)regs->ip);
703 
704         dump_pagetable(address);
705 }
706 
707 static noinline void
708 pgtable_bad(struct pt_regs *regs, unsigned long error_code,
709             unsigned long address)
710 {
711         struct task_struct *tsk;
712         unsigned long flags;
713         int sig;
714 
715         flags = oops_begin();
716         tsk = current;
717         sig = SIGKILL;
718 
719         printk(KERN_ALERT "%s: Corrupted page table at address %lx\n",
720                tsk->comm, address);
721         dump_pagetable(address);
722 
723         tsk->thread.cr2         = address;
724         tsk->thread.trap_nr     = X86_TRAP_PF;
725         tsk->thread.error_code  = error_code;
726 
727         if (__die("Bad pagetable", regs, error_code))
728                 sig = 0;
729 
730         oops_end(flags, regs, sig);
731 }
732 
733 static noinline void
734 no_context(struct pt_regs *regs, unsigned long error_code,
735            unsigned long address, int signal, int si_code)
736 {
737         struct task_struct *tsk = current;
738         unsigned long flags;
739         int sig;
740 
741         /* Are we prepared to handle this kernel fault? */
742         if (fixup_exception(regs, X86_TRAP_PF)) {
743                 /*
744                  * Any interrupt that takes a fault gets the fixup. This makes
745                  * the below recursive fault logic only apply to a faults from
746                  * task context.
747                  */
748                 if (in_interrupt())
749                         return;
750 
751                 /*
752                  * Per the above we're !in_interrupt(), aka. task context.
753                  *
754                  * In this case we need to make sure we're not recursively
755                  * faulting through the emulate_vsyscall() logic.
756                  */
757                 if (current->thread.sig_on_uaccess_err && signal) {
758                         tsk->thread.trap_nr = X86_TRAP_PF;
759                         tsk->thread.error_code = error_code | X86_PF_USER;
760                         tsk->thread.cr2 = address;
761 
762                         /* XXX: hwpoison faults will set the wrong code. */
763                         force_sig_info_fault(signal, si_code, address,
764                                              tsk, NULL, 0);
765                 }
766 
767                 /*
768                  * Barring that, we can do the fixup and be happy.
769                  */
770                 return;
771         }
772 
773 #ifdef CONFIG_VMAP_STACK
774         /*
775          * Stack overflow?  During boot, we can fault near the initial
776          * stack in the direct map, but that's not an overflow -- check
777          * that we're in vmalloc space to avoid this.
778          */
779         if (is_vmalloc_addr((void *)address) &&
780             (((unsigned long)tsk->stack - 1 - address < PAGE_SIZE) ||
781              address - ((unsigned long)tsk->stack + THREAD_SIZE) < PAGE_SIZE)) {
782                 unsigned long stack = this_cpu_read(orig_ist.ist[DOUBLEFAULT_STACK]) - sizeof(void *);
783                 /*
784                  * We're likely to be running with very little stack space
785                  * left.  It's plausible that we'd hit this condition but
786                  * double-fault even before we get this far, in which case
787                  * we're fine: the double-fault handler will deal with it.
788                  *
789                  * We don't want to make it all the way into the oops code
790                  * and then double-fault, though, because we're likely to
791                  * break the console driver and lose most of the stack dump.
792                  */
793                 asm volatile ("movq %[stack], %%rsp\n\t"
794                               "call handle_stack_overflow\n\t"
795                               "1: jmp 1b"
796                               : ASM_CALL_CONSTRAINT
797                               : "D" ("kernel stack overflow (page fault)"),
798                                 "S" (regs), "d" (address),
799                                 [stack] "rm" (stack));
800                 unreachable();
801         }
802 #endif
803 
804         /*
805          * 32-bit:
806          *
807          *   Valid to do another page fault here, because if this fault
808          *   had been triggered by is_prefetch fixup_exception would have
809          *   handled it.
810          *
811          * 64-bit:
812          *
813          *   Hall of shame of CPU/BIOS bugs.
814          */
815         if (is_prefetch(regs, error_code, address))
816                 return;
817 
818         if (is_errata93(regs, address))
819                 return;
820 
821         /*
822          * Oops. The kernel tried to access some bad page. We'll have to
823          * terminate things with extreme prejudice:
824          */
825         flags = oops_begin();
826 
827         show_fault_oops(regs, error_code, address);
828 
829         if (task_stack_end_corrupted(tsk))
830                 printk(KERN_EMERG "Thread overran stack, or stack corrupted\n");
831 
832         tsk->thread.cr2         = address;
833         tsk->thread.trap_nr     = X86_TRAP_PF;
834         tsk->thread.error_code  = error_code;
835 
836         sig = SIGKILL;
837         if (__die("Oops", regs, error_code))
838                 sig = 0;
839 
840         /* Executive summary in case the body of the oops scrolled away */
841         printk(KERN_DEFAULT "CR2: %016lx\n", address);
842 
843         oops_end(flags, regs, sig);
844 }
845 
846 /*
847  * Print out info about fatal segfaults, if the show_unhandled_signals
848  * sysctl is set:
849  */
850 static inline void
851 show_signal_msg(struct pt_regs *regs, unsigned long error_code,
852                 unsigned long address, struct task_struct *tsk)
853 {
854         if (!unhandled_signal(tsk, SIGSEGV))
855                 return;
856 
857         if (!printk_ratelimit())
858                 return;
859 
860         printk("%s%s[%d]: segfault at %lx ip %px sp %px error %lx",
861                 task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG,
862                 tsk->comm, task_pid_nr(tsk), address,
863                 (void *)regs->ip, (void *)regs->sp, error_code);
864 
865         print_vma_addr(KERN_CONT " in ", regs->ip);
866 
867         printk(KERN_CONT "\n");
868 }
869 
870 static void
871 __bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
872                        unsigned long address, u32 *pkey, int si_code)
873 {
874         struct task_struct *tsk = current;
875 
876         /* User mode accesses just cause a SIGSEGV */
877         if (error_code & X86_PF_USER) {
878                 /*
879                  * It's possible to have interrupts off here:
880                  */
881                 local_irq_enable();
882 
883                 /*
884                  * Valid to do another page fault here because this one came
885                  * from user space:
886                  */
887                 if (is_prefetch(regs, error_code, address))
888                         return;
889 
890                 if (is_errata100(regs, address))
891                         return;
892 
893 #ifdef CONFIG_X86_64
894                 /*
895                  * Instruction fetch faults in the vsyscall page might need
896                  * emulation.
897                  */
898                 if (unlikely((error_code & X86_PF_INSTR) &&
899                              ((address & ~0xfff) == VSYSCALL_ADDR))) {
900                         if (emulate_vsyscall(regs, address))
901                                 return;
902                 }
903 #endif
904 
905                 /*
906                  * To avoid leaking information about the kernel page table
907                  * layout, pretend that user-mode accesses to kernel addresses
908                  * are always protection faults.
909                  */
910                 if (address >= TASK_SIZE_MAX)
911                         error_code |= X86_PF_PROT;
912 
913                 if (likely(show_unhandled_signals))
914                         show_signal_msg(regs, error_code, address, tsk);
915 
916                 tsk->thread.cr2         = address;
917                 tsk->thread.error_code  = error_code;
918                 tsk->thread.trap_nr     = X86_TRAP_PF;
919 
920                 force_sig_info_fault(SIGSEGV, si_code, address, tsk, pkey, 0);
921 
922                 return;
923         }
924 
925         if (is_f00f_bug(regs, address))
926                 return;
927 
928         no_context(regs, error_code, address, SIGSEGV, si_code);
929 }
930 
931 static noinline void
932 bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
933                      unsigned long address, u32 *pkey)
934 {
935         __bad_area_nosemaphore(regs, error_code, address, pkey, SEGV_MAPERR);
936 }
937 
938 static void
939 __bad_area(struct pt_regs *regs, unsigned long error_code,
940            unsigned long address,  struct vm_area_struct *vma, int si_code)
941 {
942         struct mm_struct *mm = current->mm;
943         u32 pkey;
944 
945         if (vma)
946                 pkey = vma_pkey(vma);
947 
948         /*
949          * Something tried to access memory that isn't in our memory map..
950          * Fix it, but check if it's kernel or user first..
951          */
952         up_read(&mm->mmap_sem);
953 
954         __bad_area_nosemaphore(regs, error_code, address,
955                                (vma) ? &pkey : NULL, si_code);
956 }
957 
958 static noinline void
959 bad_area(struct pt_regs *regs, unsigned long error_code, unsigned long address)
960 {
961         __bad_area(regs, error_code, address, NULL, SEGV_MAPERR);
962 }
963 
964 static inline bool bad_area_access_from_pkeys(unsigned long error_code,
965                 struct vm_area_struct *vma)
966 {
967         /* This code is always called on the current mm */
968         bool foreign = false;
969 
970         if (!boot_cpu_has(X86_FEATURE_OSPKE))
971                 return false;
972         if (error_code & X86_PF_PK)
973                 return true;
974         /* this checks permission keys on the VMA: */
975         if (!arch_vma_access_permitted(vma, (error_code & X86_PF_WRITE),
976                                        (error_code & X86_PF_INSTR), foreign))
977                 return true;
978         return false;
979 }
980 
981 static noinline void
982 bad_area_access_error(struct pt_regs *regs, unsigned long error_code,
983                       unsigned long address, struct vm_area_struct *vma)
984 {
985         /*
986          * This OSPKE check is not strictly necessary at runtime.
987          * But, doing it this way allows compiler optimizations
988          * if pkeys are compiled out.
989          */
990         if (bad_area_access_from_pkeys(error_code, vma))
991                 __bad_area(regs, error_code, address, vma, SEGV_PKUERR);
992         else
993                 __bad_area(regs, error_code, address, vma, SEGV_ACCERR);
994 }
995 
996 static void
997 do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address,
998           u32 *pkey, unsigned int fault)
999 {
1000         struct task_struct *tsk = current;
1001         int code = BUS_ADRERR;
1002 
1003         /* Kernel mode? Handle exceptions or die: */
1004         if (!(error_code & X86_PF_USER)) {
1005                 no_context(regs, error_code, address, SIGBUS, BUS_ADRERR);
1006                 return;
1007         }
1008 
1009         /* User-space => ok to do another page fault: */
1010         if (is_prefetch(regs, error_code, address))
1011                 return;
1012 
1013         tsk->thread.cr2         = address;
1014         tsk->thread.error_code  = error_code;
1015         tsk->thread.trap_nr     = X86_TRAP_PF;
1016 
1017 #ifdef CONFIG_MEMORY_FAILURE
1018         if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) {
1019                 printk(KERN_ERR
1020         "MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n",
1021                         tsk->comm, tsk->pid, address);
1022                 code = BUS_MCEERR_AR;
1023         }
1024 #endif
1025         force_sig_info_fault(SIGBUS, code, address, tsk, pkey, fault);
1026 }
1027 
1028 static noinline void
1029 mm_fault_error(struct pt_regs *regs, unsigned long error_code,
1030                unsigned long address, u32 *pkey, unsigned int fault)
1031 {
1032         if (fatal_signal_pending(current) && !(error_code & X86_PF_USER)) {
1033                 no_context(regs, error_code, address, 0, 0);
1034                 return;
1035         }
1036 
1037         if (fault & VM_FAULT_OOM) {
1038                 /* Kernel mode? Handle exceptions or die: */
1039                 if (!(error_code & X86_PF_USER)) {
1040                         no_context(regs, error_code, address,
1041                                    SIGSEGV, SEGV_MAPERR);
1042                         return;
1043                 }
1044 
1045                 /*
1046                  * We ran out of memory, call the OOM killer, and return the
1047                  * userspace (which will retry the fault, or kill us if we got
1048                  * oom-killed):
1049                  */
1050                 pagefault_out_of_memory();
1051         } else {
1052                 if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON|
1053                              VM_FAULT_HWPOISON_LARGE))
1054                         do_sigbus(regs, error_code, address, pkey, fault);
1055                 else if (fault & VM_FAULT_SIGSEGV)
1056                         bad_area_nosemaphore(regs, error_code, address, pkey);
1057                 else
1058                         BUG();
1059         }
1060 }
1061 
1062 static int spurious_fault_check(unsigned long error_code, pte_t *pte)
1063 {
1064         if ((error_code & X86_PF_WRITE) && !pte_write(*pte))
1065                 return 0;
1066 
1067         if ((error_code & X86_PF_INSTR) && !pte_exec(*pte))
1068                 return 0;
1069         /*
1070          * Note: We do not do lazy flushing on protection key
1071          * changes, so no spurious fault will ever set X86_PF_PK.
1072          */
1073         if ((error_code & X86_PF_PK))
1074                 return 1;
1075 
1076         return 1;
1077 }
1078 
1079 /*
1080  * Handle a spurious fault caused by a stale TLB entry.
1081  *
1082  * This allows us to lazily refresh the TLB when increasing the
1083  * permissions of a kernel page (RO -> RW or NX -> X).  Doing it
1084  * eagerly is very expensive since that implies doing a full
1085  * cross-processor TLB flush, even if no stale TLB entries exist
1086  * on other processors.
1087  *
1088  * Spurious faults may only occur if the TLB contains an entry with
1089  * fewer permission than the page table entry.  Non-present (P = 0)
1090  * and reserved bit (R = 1) faults are never spurious.
1091  *
1092  * There are no security implications to leaving a stale TLB when
1093  * increasing the permissions on a page.
1094  *
1095  * Returns non-zero if a spurious fault was handled, zero otherwise.
1096  *
1097  * See Intel Developer's Manual Vol 3 Section 4.10.4.3, bullet 3
1098  * (Optional Invalidation).
1099  */
1100 static noinline int
1101 spurious_fault(unsigned long error_code, unsigned long address)
1102 {
1103         pgd_t *pgd;
1104         p4d_t *p4d;
1105         pud_t *pud;
1106         pmd_t *pmd;
1107         pte_t *pte;
1108         int ret;
1109 
1110         /*
1111          * Only writes to RO or instruction fetches from NX may cause
1112          * spurious faults.
1113          *
1114          * These could be from user or supervisor accesses but the TLB
1115          * is only lazily flushed after a kernel mapping protection
1116          * change, so user accesses are not expected to cause spurious
1117          * faults.
1118          */
1119         if (error_code != (X86_PF_WRITE | X86_PF_PROT) &&
1120             error_code != (X86_PF_INSTR | X86_PF_PROT))
1121                 return 0;
1122 
1123         pgd = init_mm.pgd + pgd_index(address);
1124         if (!pgd_present(*pgd))
1125                 return 0;
1126 
1127         p4d = p4d_offset(pgd, address);
1128         if (!p4d_present(*p4d))
1129                 return 0;
1130 
1131         if (p4d_large(*p4d))
1132                 return spurious_fault_check(error_code, (pte_t *) p4d);
1133 
1134         pud = pud_offset(p4d, address);
1135         if (!pud_present(*pud))
1136                 return 0;
1137 
1138         if (pud_large(*pud))
1139                 return spurious_fault_check(error_code, (pte_t *) pud);
1140 
1141         pmd = pmd_offset(pud, address);
1142         if (!pmd_present(*pmd))
1143                 return 0;
1144 
1145         if (pmd_large(*pmd))
1146                 return spurious_fault_check(error_code, (pte_t *) pmd);
1147 
1148         pte = pte_offset_kernel(pmd, address);
1149         if (!pte_present(*pte))
1150                 return 0;
1151 
1152         ret = spurious_fault_check(error_code, pte);
1153         if (!ret)
1154                 return 0;
1155 
1156         /*
1157          * Make sure we have permissions in PMD.
1158          * If not, then there's a bug in the page tables:
1159          */
1160         ret = spurious_fault_check(error_code, (pte_t *) pmd);
1161         WARN_ONCE(!ret, "PMD has incorrect permission bits\n");
1162 
1163         return ret;
1164 }
1165 NOKPROBE_SYMBOL(spurious_fault);
1166 
1167 int show_unhandled_signals = 1;
1168 
1169 static inline int
1170 access_error(unsigned long error_code, struct vm_area_struct *vma)
1171 {
1172         /* This is only called for the current mm, so: */
1173         bool foreign = false;
1174 
1175         /*
1176          * Read or write was blocked by protection keys.  This is
1177          * always an unconditional error and can never result in
1178          * a follow-up action to resolve the fault, like a COW.
1179          */
1180         if (error_code & X86_PF_PK)
1181                 return 1;
1182 
1183         /*
1184          * Make sure to check the VMA so that we do not perform
1185          * faults just to hit a X86_PF_PK as soon as we fill in a
1186          * page.
1187          */
1188         if (!arch_vma_access_permitted(vma, (error_code & X86_PF_WRITE),
1189                                        (error_code & X86_PF_INSTR), foreign))
1190                 return 1;
1191 
1192         if (error_code & X86_PF_WRITE) {
1193                 /* write, present and write, not present: */
1194                 if (unlikely(!(vma->vm_flags & VM_WRITE)))
1195                         return 1;
1196                 return 0;
1197         }
1198 
1199         /* read, present: */
1200         if (unlikely(error_code & X86_PF_PROT))
1201                 return 1;
1202 
1203         /* read, not present: */
1204         if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE))))
1205                 return 1;
1206 
1207         return 0;
1208 }
1209 
1210 static int fault_in_kernel_space(unsigned long address)
1211 {
1212         return address >= TASK_SIZE_MAX;
1213 }
1214 
1215 static inline bool smap_violation(int error_code, struct pt_regs *regs)
1216 {
1217         if (!IS_ENABLED(CONFIG_X86_SMAP))
1218                 return false;
1219 
1220         if (!static_cpu_has(X86_FEATURE_SMAP))
1221                 return false;
1222 
1223         if (error_code & X86_PF_USER)
1224                 return false;
1225 
1226         if (!user_mode(regs) && (regs->flags & X86_EFLAGS_AC))
1227                 return false;
1228 
1229         return true;
1230 }
1231 
1232 /*
1233  * This routine handles page faults.  It determines the address,
1234  * and the problem, and then passes it off to one of the appropriate
1235  * routines.
1236  */
1237 static noinline void
1238 __do_page_fault(struct pt_regs *regs, unsigned long error_code,
1239                 unsigned long address)
1240 {
1241         struct vm_area_struct *vma;
1242         struct task_struct *tsk;
1243         struct mm_struct *mm;
1244         int fault, major = 0;
1245         unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
1246         u32 pkey;
1247 
1248         tsk = current;
1249         mm = tsk->mm;
1250 
1251         prefetchw(&mm->mmap_sem);
1252 
1253         if (unlikely(kmmio_fault(regs, address)))
1254                 return;
1255 
1256         /*
1257          * We fault-in kernel-space virtual memory on-demand. The
1258          * 'reference' page table is init_mm.pgd.
1259          *
1260          * NOTE! We MUST NOT take any locks for this case. We may
1261          * be in an interrupt or a critical region, and should
1262          * only copy the information from the master page table,
1263          * nothing more.
1264          *
1265          * This verifies that the fault happens in kernel space
1266          * (error_code & 4) == 0, and that the fault was not a
1267          * protection error (error_code & 9) == 0.
1268          */
1269         if (unlikely(fault_in_kernel_space(address))) {
1270                 if (!(error_code & (X86_PF_RSVD | X86_PF_USER | X86_PF_PROT))) {
1271                         if (vmalloc_fault(address) >= 0)
1272                                 return;
1273                 }
1274 
1275                 /* Can handle a stale RO->RW TLB: */
1276                 if (spurious_fault(error_code, address))
1277                         return;
1278 
1279                 /* kprobes don't want to hook the spurious faults: */
1280                 if (kprobes_fault(regs))
1281                         return;
1282                 /*
1283                  * Don't take the mm semaphore here. If we fixup a prefetch
1284                  * fault we could otherwise deadlock:
1285                  */
1286                 bad_area_nosemaphore(regs, error_code, address, NULL);
1287 
1288                 return;
1289         }
1290 
1291         /* kprobes don't want to hook the spurious faults: */
1292         if (unlikely(kprobes_fault(regs)))
1293                 return;
1294 
1295         if (unlikely(error_code & X86_PF_RSVD))
1296                 pgtable_bad(regs, error_code, address);
1297 
1298         if (unlikely(smap_violation(error_code, regs))) {
1299                 bad_area_nosemaphore(regs, error_code, address, NULL);
1300                 return;
1301         }
1302 
1303         /*
1304          * If we're in an interrupt, have no user context or are running
1305          * in a region with pagefaults disabled then we must not take the fault
1306          */
1307         if (unlikely(faulthandler_disabled() || !mm)) {
1308                 bad_area_nosemaphore(regs, error_code, address, NULL);
1309                 return;
1310         }
1311 
1312         /*
1313          * It's safe to allow irq's after cr2 has been saved and the
1314          * vmalloc fault has been handled.
1315          *
1316          * User-mode registers count as a user access even for any
1317          * potential system fault or CPU buglet:
1318          */
1319         if (user_mode(regs)) {
1320                 local_irq_enable();
1321                 error_code |= X86_PF_USER;
1322                 flags |= FAULT_FLAG_USER;
1323         } else {
1324                 if (regs->flags & X86_EFLAGS_IF)
1325                         local_irq_enable();
1326         }
1327 
1328         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
1329 
1330         if (error_code & X86_PF_WRITE)
1331                 flags |= FAULT_FLAG_WRITE;
1332         if (error_code & X86_PF_INSTR)
1333                 flags |= FAULT_FLAG_INSTRUCTION;
1334 
1335         /*
1336          * When running in the kernel we expect faults to occur only to
1337          * addresses in user space.  All other faults represent errors in
1338          * the kernel and should generate an OOPS.  Unfortunately, in the
1339          * case of an erroneous fault occurring in a code path which already
1340          * holds mmap_sem we will deadlock attempting to validate the fault
1341          * against the address space.  Luckily the kernel only validly
1342          * references user space from well defined areas of code, which are
1343          * listed in the exceptions table.
1344          *
1345          * As the vast majority of faults will be valid we will only perform
1346          * the source reference check when there is a possibility of a
1347          * deadlock. Attempt to lock the address space, if we cannot we then
1348          * validate the source. If this is invalid we can skip the address
1349          * space check, thus avoiding the deadlock:
1350          */
1351         if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
1352                 if (!(error_code & X86_PF_USER) &&
1353                     !search_exception_tables(regs->ip)) {
1354                         bad_area_nosemaphore(regs, error_code, address, NULL);
1355                         return;
1356                 }
1357 retry:
1358                 down_read(&mm->mmap_sem);
1359         } else {
1360                 /*
1361                  * The above down_read_trylock() might have succeeded in
1362                  * which case we'll have missed the might_sleep() from
1363                  * down_read():
1364                  */
1365                 might_sleep();
1366         }
1367 
1368         vma = find_vma(mm, address);
1369         if (unlikely(!vma)) {
1370                 bad_area(regs, error_code, address);
1371                 return;
1372         }
1373         if (likely(vma->vm_start <= address))
1374                 goto good_area;
1375         if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) {
1376                 bad_area(regs, error_code, address);
1377                 return;
1378         }
1379         if (error_code & X86_PF_USER) {
1380                 /*
1381                  * Accessing the stack below %sp is always a bug.
1382                  * The large cushion allows instructions like enter
1383                  * and pusha to work. ("enter $65535, $31" pushes
1384                  * 32 pointers and then decrements %sp by 65535.)
1385                  */
1386                 if (unlikely(address + 65536 + 32 * sizeof(unsigned long) < regs->sp)) {
1387                         bad_area(regs, error_code, address);
1388                         return;
1389                 }
1390         }
1391         if (unlikely(expand_stack(vma, address))) {
1392                 bad_area(regs, error_code, address);
1393                 return;
1394         }
1395 
1396         /*
1397          * Ok, we have a good vm_area for this memory access, so
1398          * we can handle it..
1399          */
1400 good_area:
1401         if (unlikely(access_error(error_code, vma))) {
1402                 bad_area_access_error(regs, error_code, address, vma);
1403                 return;
1404         }
1405 
1406         /*
1407          * If for any reason at all we couldn't handle the fault,
1408          * make sure we exit gracefully rather than endlessly redo
1409          * the fault.  Since we never set FAULT_FLAG_RETRY_NOWAIT, if
1410          * we get VM_FAULT_RETRY back, the mmap_sem has been unlocked.
1411          *
1412          * Note that handle_userfault() may also release and reacquire mmap_sem
1413          * (and not return with VM_FAULT_RETRY), when returning to userland to
1414          * repeat the page fault later with a VM_FAULT_NOPAGE retval
1415          * (potentially after handling any pending signal during the return to
1416          * userland). The return to userland is identified whenever
1417          * FAULT_FLAG_USER|FAULT_FLAG_KILLABLE are both set in flags.
1418          * Thus we have to be careful about not touching vma after handling the
1419          * fault, so we read the pkey beforehand.
1420          */
1421         pkey = vma_pkey(vma);
1422         fault = handle_mm_fault(vma, address, flags);
1423         major |= fault & VM_FAULT_MAJOR;
1424 
1425         /*
1426          * If we need to retry the mmap_sem has already been released,
1427          * and if there is a fatal signal pending there is no guarantee
1428          * that we made any progress. Handle this case first.
1429          */
1430         if (unlikely(fault & VM_FAULT_RETRY)) {
1431                 /* Retry at most once */
1432                 if (flags & FAULT_FLAG_ALLOW_RETRY) {
1433                         flags &= ~FAULT_FLAG_ALLOW_RETRY;
1434                         flags |= FAULT_FLAG_TRIED;
1435                         if (!fatal_signal_pending(tsk))
1436                                 goto retry;
1437                 }
1438 
1439                 /* User mode? Just return to handle the fatal exception */
1440                 if (flags & FAULT_FLAG_USER)
1441                         return;
1442 
1443                 /* Not returning to user mode? Handle exceptions or die: */
1444                 no_context(regs, error_code, address, SIGBUS, BUS_ADRERR);
1445                 return;
1446         }
1447 
1448         up_read(&mm->mmap_sem);
1449         if (unlikely(fault & VM_FAULT_ERROR)) {
1450                 mm_fault_error(regs, error_code, address, &pkey, fault);
1451                 return;
1452         }
1453 
1454         /*
1455          * Major/minor page fault accounting. If any of the events
1456          * returned VM_FAULT_MAJOR, we account it as a major fault.
1457          */
1458         if (major) {
1459                 tsk->maj_flt++;
1460                 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, address);
1461         } else {
1462                 tsk->min_flt++;
1463                 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, address);
1464         }
1465 
1466         check_v8086_mode(regs, address, tsk);
1467 }
1468 NOKPROBE_SYMBOL(__do_page_fault);
1469 
1470 static nokprobe_inline void
1471 trace_page_fault_entries(unsigned long address, struct pt_regs *regs,
1472                          unsigned long error_code)
1473 {
1474         if (user_mode(regs))
1475                 trace_page_fault_user(address, regs, error_code);
1476         else
1477                 trace_page_fault_kernel(address, regs, error_code);
1478 }
1479 
1480 /*
1481  * We must have this function blacklisted from kprobes, tagged with notrace
1482  * and call read_cr2() before calling anything else. To avoid calling any
1483  * kind of tracing machinery before we've observed the CR2 value.
1484  *
1485  * exception_{enter,exit}() contains all sorts of tracepoints.
1486  */
1487 dotraplinkage void notrace
1488 do_page_fault(struct pt_regs *regs, unsigned long error_code)
1489 {
1490         unsigned long address = read_cr2(); /* Get the faulting address */
1491         enum ctx_state prev_state;
1492 
1493         prev_state = exception_enter();
1494         if (trace_pagefault_enabled())
1495                 trace_page_fault_entries(address, regs, error_code);
1496 
1497         __do_page_fault(regs, error_code, address);
1498         exception_exit(prev_state);
1499 }
1500 NOKPROBE_SYMBOL(do_page_fault);
1501 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp