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/memblock.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#include <linux/efi.h> /* efi_recover_from_page_fault()*/
20#include <linux/mm_types.h>
21
22#include <asm/cpufeature.h> /* boot_cpu_has, ... */
23#include <asm/traps.h> /* dotraplinkage, ... */
24#include <asm/pgalloc.h> /* pgd_*(), ... */
25#include <asm/fixmap.h> /* VSYSCALL_ADDR */
26#include <asm/vsyscall.h> /* emulate_vsyscall */
27#include <asm/vm86.h> /* struct vm86 */
28#include <asm/mmu_context.h> /* vma_pkey() */
29#include <asm/efi.h> /* efi_recover_from_page_fault()*/
30#include <asm/desc.h> /* store_idt(), ... */
31
32#define CREATE_TRACE_POINTS
33#include <asm/trace/exceptions.h>
34
35/*
36 * Returns 0 if mmiotrace is disabled, or if the fault is not
37 * handled by mmiotrace:
38 */
39static nokprobe_inline int
40kmmio_fault(struct pt_regs *regs, unsigned long addr)
41{
42 if (unlikely(is_kmmio_active()))
43 if (kmmio_handler(regs, addr) == 1)
44 return -1;
45 return 0;
46}
47
48static nokprobe_inline int kprobes_fault(struct pt_regs *regs)
49{
50 if (!kprobes_built_in())
51 return 0;
52 if (user_mode(regs))
53 return 0;
54 /*
55 * To be potentially processing a kprobe fault and to be allowed to call
56 * kprobe_running(), we have to be non-preemptible.
57 */
58 if (preemptible())
59 return 0;
60 if (!kprobe_running())
61 return 0;
62 return kprobe_fault_handler(regs, X86_TRAP_PF);
63}
64
65/*
66 * Prefetch quirks:
67 *
68 * 32-bit mode:
69 *
70 * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch.
71 * Check that here and ignore it.
72 *
73 * 64-bit mode:
74 *
75 * Sometimes the CPU reports invalid exceptions on prefetch.
76 * Check that here and ignore it.
77 *
78 * Opcode checker based on code by Richard Brunner.
79 */
80static inline int
81check_prefetch_opcode(struct pt_regs *regs, unsigned char *instr,
82 unsigned char opcode, int *prefetch)
83{
84 unsigned char instr_hi = opcode & 0xf0;
85 unsigned char instr_lo = opcode & 0x0f;
86
87 switch (instr_hi) {
88 case 0x20:
89 case 0x30:
90 /*
91 * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes.
92 * In X86_64 long mode, the CPU will signal invalid
93 * opcode if some of these prefixes are present so
94 * X86_64 will never get here anyway
95 */
96 return ((instr_lo & 7) == 0x6);
97#ifdef CONFIG_X86_64
98 case 0x40:
99 /*
100 * In AMD64 long mode 0x40..0x4F are valid REX prefixes
101 * Need to figure out under what instruction mode the
102 * instruction was issued. Could check the LDT for lm,
103 * but for now it's good enough to assume that long
104 * mode only uses well known segments or kernel.
105 */
106 return (!user_mode(regs) || user_64bit_mode(regs));
107#endif
108 case 0x60:
109 /* 0x64 thru 0x67 are valid prefixes in all modes. */
110 return (instr_lo & 0xC) == 0x4;
111 case 0xF0:
112 /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */
113 return !instr_lo || (instr_lo>>1) == 1;
114 case 0x00:
115 /* Prefetch instruction is 0x0F0D or 0x0F18 */
116 if (probe_kernel_address(instr, opcode))
117 return 0;
118
119 *prefetch = (instr_lo == 0xF) &&
120 (opcode == 0x0D || opcode == 0x18);
121 return 0;
122 default:
123 return 0;
124 }
125}
126
127static int
128is_prefetch(struct pt_regs *regs, unsigned long error_code, unsigned long addr)
129{
130 unsigned char *max_instr;
131 unsigned char *instr;
132 int prefetch = 0;
133
134 /*
135 * If it was a exec (instruction fetch) fault on NX page, then
136 * do not ignore the fault:
137 */
138 if (error_code & X86_PF_INSTR)
139 return 0;
140
141 instr = (void *)convert_ip_to_linear(current, regs);
142 max_instr = instr + 15;
143
144 if (user_mode(regs) && instr >= (unsigned char *)TASK_SIZE_MAX)
145 return 0;
146
147 while (instr < max_instr) {
148 unsigned char opcode;
149
150 if (probe_kernel_address(instr, opcode))
151 break;
152
153 instr++;
154
155 if (!check_prefetch_opcode(regs, instr, opcode, &prefetch))
156 break;
157 }
158 return prefetch;
159}
160
161DEFINE_SPINLOCK(pgd_lock);
162LIST_HEAD(pgd_list);
163
164#ifdef CONFIG_X86_32
165static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address)
166{
167 unsigned index = pgd_index(address);
168 pgd_t *pgd_k;
169 p4d_t *p4d, *p4d_k;
170 pud_t *pud, *pud_k;
171 pmd_t *pmd, *pmd_k;
172
173 pgd += index;
174 pgd_k = init_mm.pgd + index;
175
176 if (!pgd_present(*pgd_k))
177 return NULL;
178
179 /*
180 * set_pgd(pgd, *pgd_k); here would be useless on PAE
181 * and redundant with the set_pmd() on non-PAE. As would
182 * set_p4d/set_pud.
183 */
184 p4d = p4d_offset(pgd, address);
185 p4d_k = p4d_offset(pgd_k, address);
186 if (!p4d_present(*p4d_k))
187 return NULL;
188
189 pud = pud_offset(p4d, address);
190 pud_k = pud_offset(p4d_k, address);
191 if (!pud_present(*pud_k))
192 return NULL;
193
194 pmd = pmd_offset(pud, address);
195 pmd_k = pmd_offset(pud_k, address);
196 if (!pmd_present(*pmd_k))
197 return NULL;
198
199 if (!pmd_present(*pmd))
200 set_pmd(pmd, *pmd_k);
201 else
202 BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k));
203
204 return pmd_k;
205}
206
207void vmalloc_sync_all(void)
208{
209 unsigned long address;
210
211 if (SHARED_KERNEL_PMD)
212 return;
213
214 for (address = VMALLOC_START & PMD_MASK;
215 address >= TASK_SIZE_MAX && address < FIXADDR_TOP;
216 address += PMD_SIZE) {
217 struct page *page;
218
219 spin_lock(&pgd_lock);
220 list_for_each_entry(page, &pgd_list, lru) {
221 spinlock_t *pgt_lock;
222 pmd_t *ret;
223
224 /* the pgt_lock only for Xen */
225 pgt_lock = &pgd_page_get_mm(page)->page_table_lock;
226
227 spin_lock(pgt_lock);
228 ret = vmalloc_sync_one(page_address(page), address);
229 spin_unlock(pgt_lock);
230
231 if (!ret)
232 break;
233 }
234 spin_unlock(&pgd_lock);
235 }
236}
237
238/*
239 * 32-bit:
240 *
241 * Handle a fault on the vmalloc or module mapping area
242 */
243static noinline int vmalloc_fault(unsigned long address)
244{
245 unsigned long pgd_paddr;
246 pmd_t *pmd_k;
247 pte_t *pte_k;
248
249 /* Make sure we are in vmalloc area: */
250 if (!(address >= VMALLOC_START && address < VMALLOC_END))
251 return -1;
252
253 /*
254 * Synchronize this task's top level page-table
255 * with the 'reference' page table.
256 *
257 * Do _not_ use "current" here. We might be inside
258 * an interrupt in the middle of a task switch..
259 */
260 pgd_paddr = read_cr3_pa();
261 pmd_k = vmalloc_sync_one(__va(pgd_paddr), address);
262 if (!pmd_k)
263 return -1;
264
265 if (pmd_large(*pmd_k))
266 return 0;
267
268 pte_k = pte_offset_kernel(pmd_k, address);
269 if (!pte_present(*pte_k))
270 return -1;
271
272 return 0;
273}
274NOKPROBE_SYMBOL(vmalloc_fault);
275
276/*
277 * Did it hit the DOS screen memory VA from vm86 mode?
278 */
279static inline void
280check_v8086_mode(struct pt_regs *regs, unsigned long address,
281 struct task_struct *tsk)
282{
283#ifdef CONFIG_VM86
284 unsigned long bit;
285
286 if (!v8086_mode(regs) || !tsk->thread.vm86)
287 return;
288
289 bit = (address - 0xA0000) >> PAGE_SHIFT;
290 if (bit < 32)
291 tsk->thread.vm86->screen_bitmap |= 1 << bit;
292#endif
293}
294
295static bool low_pfn(unsigned long pfn)
296{
297 return pfn < max_low_pfn;
298}
299
300static void dump_pagetable(unsigned long address)
301{
302 pgd_t *base = __va(read_cr3_pa());
303 pgd_t *pgd = &base[pgd_index(address)];
304 p4d_t *p4d;
305 pud_t *pud;
306 pmd_t *pmd;
307 pte_t *pte;
308
309#ifdef CONFIG_X86_PAE
310 pr_info("*pdpt = %016Lx ", pgd_val(*pgd));
311 if (!low_pfn(pgd_val(*pgd) >> PAGE_SHIFT) || !pgd_present(*pgd))
312 goto out;
313#define pr_pde pr_cont
314#else
315#define pr_pde pr_info
316#endif
317 p4d = p4d_offset(pgd, address);
318 pud = pud_offset(p4d, address);
319 pmd = pmd_offset(pud, address);
320 pr_pde("*pde = %0*Lx ", sizeof(*pmd) * 2, (u64)pmd_val(*pmd));
321#undef pr_pde
322
323 /*
324 * We must not directly access the pte in the highpte
325 * case if the page table is located in highmem.
326 * And let's rather not kmap-atomic the pte, just in case
327 * it's allocated already:
328 */
329 if (!low_pfn(pmd_pfn(*pmd)) || !pmd_present(*pmd) || pmd_large(*pmd))
330 goto out;
331
332 pte = pte_offset_kernel(pmd, address);
333 pr_cont("*pte = %0*Lx ", sizeof(*pte) * 2, (u64)pte_val(*pte));
334out:
335 pr_cont("\n");
336}
337
338#else /* CONFIG_X86_64: */
339
340void vmalloc_sync_all(void)
341{
342 sync_global_pgds(VMALLOC_START & PGDIR_MASK, VMALLOC_END);
343}
344
345/*
346 * 64-bit:
347 *
348 * Handle a fault on the vmalloc area
349 */
350static noinline int vmalloc_fault(unsigned long address)
351{
352 pgd_t *pgd, *pgd_k;
353 p4d_t *p4d, *p4d_k;
354 pud_t *pud;
355 pmd_t *pmd;
356 pte_t *pte;
357
358 /* Make sure we are in vmalloc area: */
359 if (!(address >= VMALLOC_START && address < VMALLOC_END))
360 return -1;
361
362 WARN_ON_ONCE(in_nmi());
363
364 /*
365 * Copy kernel mappings over when needed. This can also
366 * happen within a race in page table update. In the later
367 * case just flush:
368 */
369 pgd = (pgd_t *)__va(read_cr3_pa()) + pgd_index(address);
370 pgd_k = pgd_offset_k(address);
371 if (pgd_none(*pgd_k))
372 return -1;
373
374 if (pgtable_l5_enabled()) {
375 if (pgd_none(*pgd)) {
376 set_pgd(pgd, *pgd_k);
377 arch_flush_lazy_mmu_mode();
378 } else {
379 BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_k));
380 }
381 }
382
383 /* With 4-level paging, copying happens on the p4d level. */
384 p4d = p4d_offset(pgd, address);
385 p4d_k = p4d_offset(pgd_k, address);
386 if (p4d_none(*p4d_k))
387 return -1;
388
389 if (p4d_none(*p4d) && !pgtable_l5_enabled()) {
390 set_p4d(p4d, *p4d_k);
391 arch_flush_lazy_mmu_mode();
392 } else {
393 BUG_ON(p4d_pfn(*p4d) != p4d_pfn(*p4d_k));
394 }
395
396 BUILD_BUG_ON(CONFIG_PGTABLE_LEVELS < 4);
397
398 pud = pud_offset(p4d, address);
399 if (pud_none(*pud))
400 return -1;
401
402 if (pud_large(*pud))
403 return 0;
404
405 pmd = pmd_offset(pud, address);
406 if (pmd_none(*pmd))
407 return -1;
408
409 if (pmd_large(*pmd))
410 return 0;
411
412 pte = pte_offset_kernel(pmd, address);
413 if (!pte_present(*pte))
414 return -1;
415
416 return 0;
417}
418NOKPROBE_SYMBOL(vmalloc_fault);
419
420#ifdef CONFIG_CPU_SUP_AMD
421static const char errata93_warning[] =
422KERN_ERR
423"******* Your BIOS seems to not contain a fix for K8 errata #93\n"
424"******* Working around it, but it may cause SEGVs or burn power.\n"
425"******* Please consider a BIOS update.\n"
426"******* Disabling USB legacy in the BIOS may also help.\n";
427#endif
428
429/*
430 * No vm86 mode in 64-bit mode:
431 */
432static inline void
433check_v8086_mode(struct pt_regs *regs, unsigned long address,
434 struct task_struct *tsk)
435{
436}
437
438static int bad_address(void *p)
439{
440 unsigned long dummy;
441
442 return probe_kernel_address((unsigned long *)p, dummy);
443}
444
445static void dump_pagetable(unsigned long address)
446{
447 pgd_t *base = __va(read_cr3_pa());
448 pgd_t *pgd = base + pgd_index(address);
449 p4d_t *p4d;
450 pud_t *pud;
451 pmd_t *pmd;
452 pte_t *pte;
453
454 if (bad_address(pgd))
455 goto bad;
456
457 pr_info("PGD %lx ", pgd_val(*pgd));
458
459 if (!pgd_present(*pgd))
460 goto out;
461
462 p4d = p4d_offset(pgd, address);
463 if (bad_address(p4d))
464 goto bad;
465
466 pr_cont("P4D %lx ", p4d_val(*p4d));
467 if (!p4d_present(*p4d) || p4d_large(*p4d))
468 goto out;
469
470 pud = pud_offset(p4d, address);
471 if (bad_address(pud))
472 goto bad;
473
474 pr_cont("PUD %lx ", pud_val(*pud));
475 if (!pud_present(*pud) || pud_large(*pud))
476 goto out;
477
478 pmd = pmd_offset(pud, address);
479 if (bad_address(pmd))
480 goto bad;
481
482 pr_cont("PMD %lx ", pmd_val(*pmd));
483 if (!pmd_present(*pmd) || pmd_large(*pmd))
484 goto out;
485
486 pte = pte_offset_kernel(pmd, address);
487 if (bad_address(pte))
488 goto bad;
489
490 pr_cont("PTE %lx", pte_val(*pte));
491out:
492 pr_cont("\n");
493 return;
494bad:
495 pr_info("BAD\n");
496}
497
498#endif /* CONFIG_X86_64 */
499
500/*
501 * Workaround for K8 erratum #93 & buggy BIOS.
502 *
503 * BIOS SMM functions are required to use a specific workaround
504 * to avoid corruption of the 64bit RIP register on C stepping K8.
505 *
506 * A lot of BIOS that didn't get tested properly miss this.
507 *
508 * The OS sees this as a page fault with the upper 32bits of RIP cleared.
509 * Try to work around it here.
510 *
511 * Note we only handle faults in kernel here.
512 * Does nothing on 32-bit.
513 */
514static int is_errata93(struct pt_regs *regs, unsigned long address)
515{
516#if defined(CONFIG_X86_64) && defined(CONFIG_CPU_SUP_AMD)
517 if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD
518 || boot_cpu_data.x86 != 0xf)
519 return 0;
520
521 if (address != regs->ip)
522 return 0;
523
524 if ((address >> 32) != 0)
525 return 0;
526
527 address |= 0xffffffffUL << 32;
528 if ((address >= (u64)_stext && address <= (u64)_etext) ||
529 (address >= MODULES_VADDR && address <= MODULES_END)) {
530 printk_once(errata93_warning);
531 regs->ip = address;
532 return 1;
533 }
534#endif
535 return 0;
536}
537
538/*
539 * Work around K8 erratum #100 K8 in compat mode occasionally jumps
540 * to illegal addresses >4GB.
541 *
542 * We catch this in the page fault handler because these addresses
543 * are not reachable. Just detect this case and return. Any code
544 * segment in LDT is compatibility mode.
545 */
546static int is_errata100(struct pt_regs *regs, unsigned long address)
547{
548#ifdef CONFIG_X86_64
549 if ((regs->cs == __USER32_CS || (regs->cs & (1<<2))) && (address >> 32))
550 return 1;
551#endif
552 return 0;
553}
554
555static int is_f00f_bug(struct pt_regs *regs, unsigned long address)
556{
557#ifdef CONFIG_X86_F00F_BUG
558 unsigned long nr;
559
560 /*
561 * Pentium F0 0F C7 C8 bug workaround:
562 */
563 if (boot_cpu_has_bug(X86_BUG_F00F)) {
564 nr = (address - idt_descr.address) >> 3;
565
566 if (nr == 6) {
567 do_invalid_op(regs, 0);
568 return 1;
569 }
570 }
571#endif
572 return 0;
573}
574
575static void show_ldttss(const struct desc_ptr *gdt, const char *name, u16 index)
576{
577 u32 offset = (index >> 3) * sizeof(struct desc_struct);
578 unsigned long addr;
579 struct ldttss_desc desc;
580
581 if (index == 0) {
582 pr_alert("%s: NULL\n", name);
583 return;
584 }
585
586 if (offset + sizeof(struct ldttss_desc) >= gdt->size) {
587 pr_alert("%s: 0x%hx -- out of bounds\n", name, index);
588 return;
589 }
590
591 if (probe_kernel_read(&desc, (void *)(gdt->address + offset),
592 sizeof(struct ldttss_desc))) {
593 pr_alert("%s: 0x%hx -- GDT entry is not readable\n",
594 name, index);
595 return;
596 }
597
598 addr = desc.base0 | (desc.base1 << 16) | ((unsigned long)desc.base2 << 24);
599#ifdef CONFIG_X86_64
600 addr |= ((u64)desc.base3 << 32);
601#endif
602 pr_alert("%s: 0x%hx -- base=0x%lx limit=0x%x\n",
603 name, index, addr, (desc.limit0 | (desc.limit1 << 16)));
604}
605
606/*
607 * This helper function transforms the #PF error_code bits into
608 * "[PROT] [USER]" type of descriptive, almost human-readable error strings:
609 */
610static void err_str_append(unsigned long error_code, char *buf, unsigned long mask, const char *txt)
611{
612 if (error_code & mask) {
613 if (buf[0])
614 strcat(buf, " ");
615 strcat(buf, txt);
616 }
617}
618
619static void
620show_fault_oops(struct pt_regs *regs, unsigned long error_code, unsigned long address)
621{
622 char err_txt[64];
623
624 if (!oops_may_print())
625 return;
626
627 if (error_code & X86_PF_INSTR) {
628 unsigned int level;
629 pgd_t *pgd;
630 pte_t *pte;
631
632 pgd = __va(read_cr3_pa());
633 pgd += pgd_index(address);
634
635 pte = lookup_address_in_pgd(pgd, address, &level);
636
637 if (pte && pte_present(*pte) && !pte_exec(*pte))
638 pr_crit("kernel tried to execute NX-protected page - exploit attempt? (uid: %d)\n",
639 from_kuid(&init_user_ns, current_uid()));
640 if (pte && pte_present(*pte) && pte_exec(*pte) &&
641 (pgd_flags(*pgd) & _PAGE_USER) &&
642 (__read_cr4() & X86_CR4_SMEP))
643 pr_crit("unable to execute userspace code (SMEP?) (uid: %d)\n",
644 from_kuid(&init_user_ns, current_uid()));
645 }
646
647 pr_alert("BUG: unable to handle kernel %s at %px\n",
648 address < PAGE_SIZE ? "NULL pointer dereference" : "paging request",
649 (void *)address);
650
651 err_txt[0] = 0;
652
653 /*
654 * Note: length of these appended strings including the separation space and the
655 * zero delimiter must fit into err_txt[].
656 */
657 err_str_append(error_code, err_txt, X86_PF_PROT, "[PROT]" );
658 err_str_append(error_code, err_txt, X86_PF_WRITE, "[WRITE]");
659 err_str_append(error_code, err_txt, X86_PF_USER, "[USER]" );
660 err_str_append(error_code, err_txt, X86_PF_RSVD, "[RSVD]" );
661 err_str_append(error_code, err_txt, X86_PF_INSTR, "[INSTR]");
662 err_str_append(error_code, err_txt, X86_PF_PK, "[PK]" );
663
664 pr_alert("#PF error: %s\n", error_code ? err_txt : "[normal kernel read fault]");
665
666 if (!(error_code & X86_PF_USER) && user_mode(regs)) {
667 struct desc_ptr idt, gdt;
668 u16 ldtr, tr;
669
670 pr_alert("This was a system access from user code\n");
671
672 /*
673 * This can happen for quite a few reasons. The more obvious
674 * ones are faults accessing the GDT, or LDT. Perhaps
675 * surprisingly, if the CPU tries to deliver a benign or
676 * contributory exception from user code and gets a page fault
677 * during delivery, the page fault can be delivered as though
678 * it originated directly from user code. This could happen
679 * due to wrong permissions on the IDT, GDT, LDT, TSS, or
680 * kernel or IST stack.
681 */
682 store_idt(&idt);
683
684 /* Usable even on Xen PV -- it's just slow. */
685 native_store_gdt(&gdt);
686
687 pr_alert("IDT: 0x%lx (limit=0x%hx) GDT: 0x%lx (limit=0x%hx)\n",
688 idt.address, idt.size, gdt.address, gdt.size);
689
690 store_ldt(ldtr);
691 show_ldttss(&gdt, "LDTR", ldtr);
692
693 store_tr(tr);
694 show_ldttss(&gdt, "TR", tr);
695 }
696
697 dump_pagetable(address);
698}
699
700static noinline void
701pgtable_bad(struct pt_regs *regs, unsigned long error_code,
702 unsigned long address)
703{
704 struct task_struct *tsk;
705 unsigned long flags;
706 int sig;
707
708 flags = oops_begin();
709 tsk = current;
710 sig = SIGKILL;
711
712 printk(KERN_ALERT "%s: Corrupted page table at address %lx\n",
713 tsk->comm, address);
714 dump_pagetable(address);
715
716 if (__die("Bad pagetable", regs, error_code))
717 sig = 0;
718
719 oops_end(flags, regs, sig);
720}
721
722static void set_signal_archinfo(unsigned long address,
723 unsigned long error_code)
724{
725 struct task_struct *tsk = current;
726
727 /*
728 * To avoid leaking information about the kernel page
729 * table layout, pretend that user-mode accesses to
730 * kernel addresses are always protection faults.
731 */
732 if (address >= TASK_SIZE_MAX)
733 error_code |= X86_PF_PROT;
734
735 tsk->thread.trap_nr = X86_TRAP_PF;
736 tsk->thread.error_code = error_code | X86_PF_USER;
737 tsk->thread.cr2 = address;
738}
739
740static noinline void
741no_context(struct pt_regs *regs, unsigned long error_code,
742 unsigned long address, int signal, int si_code)
743{
744 struct task_struct *tsk = current;
745 unsigned long flags;
746 int sig;
747
748 if (user_mode(regs)) {
749 /*
750 * This is an implicit supervisor-mode access from user
751 * mode. Bypass all the kernel-mode recovery code and just
752 * OOPS.
753 */
754 goto oops;
755 }
756
757 /* Are we prepared to handle this kernel fault? */
758 if (fixup_exception(regs, X86_TRAP_PF, error_code, address)) {
759 /*
760 * Any interrupt that takes a fault gets the fixup. This makes
761 * the below recursive fault logic only apply to a faults from
762 * task context.
763 */
764 if (in_interrupt())
765 return;
766
767 /*
768 * Per the above we're !in_interrupt(), aka. task context.
769 *
770 * In this case we need to make sure we're not recursively
771 * faulting through the emulate_vsyscall() logic.
772 */
773 if (current->thread.sig_on_uaccess_err && signal) {
774 set_signal_archinfo(address, error_code);
775
776 /* XXX: hwpoison faults will set the wrong code. */
777 force_sig_fault(signal, si_code, (void __user *)address,
778 tsk);
779 }
780
781 /*
782 * Barring that, we can do the fixup and be happy.
783 */
784 return;
785 }
786
787#ifdef CONFIG_VMAP_STACK
788 /*
789 * Stack overflow? During boot, we can fault near the initial
790 * stack in the direct map, but that's not an overflow -- check
791 * that we're in vmalloc space to avoid this.
792 */
793 if (is_vmalloc_addr((void *)address) &&
794 (((unsigned long)tsk->stack - 1 - address < PAGE_SIZE) ||
795 address - ((unsigned long)tsk->stack + THREAD_SIZE) < PAGE_SIZE)) {
796 unsigned long stack = this_cpu_read(orig_ist.ist[DOUBLEFAULT_STACK]) - sizeof(void *);
797 /*
798 * We're likely to be running with very little stack space
799 * left. It's plausible that we'd hit this condition but
800 * double-fault even before we get this far, in which case
801 * we're fine: the double-fault handler will deal with it.
802 *
803 * We don't want to make it all the way into the oops code
804 * and then double-fault, though, because we're likely to
805 * break the console driver and lose most of the stack dump.
806 */
807 asm volatile ("movq %[stack], %%rsp\n\t"
808 "call handle_stack_overflow\n\t"
809 "1: jmp 1b"
810 : ASM_CALL_CONSTRAINT
811 : "D" ("kernel stack overflow (page fault)"),
812 "S" (regs), "d" (address),
813 [stack] "rm" (stack));
814 unreachable();
815 }
816#endif
817
818 /*
819 * 32-bit:
820 *
821 * Valid to do another page fault here, because if this fault
822 * had been triggered by is_prefetch fixup_exception would have
823 * handled it.
824 *
825 * 64-bit:
826 *
827 * Hall of shame of CPU/BIOS bugs.
828 */
829 if (is_prefetch(regs, error_code, address))
830 return;
831
832 if (is_errata93(regs, address))
833 return;
834
835 /*
836 * Buggy firmware could access regions which might page fault, try to
837 * recover from such faults.
838 */
839 if (IS_ENABLED(CONFIG_EFI))
840 efi_recover_from_page_fault(address);
841
842oops:
843 /*
844 * Oops. The kernel tried to access some bad page. We'll have to
845 * terminate things with extreme prejudice:
846 */
847 flags = oops_begin();
848
849 show_fault_oops(regs, error_code, address);
850
851 if (task_stack_end_corrupted(tsk))
852 printk(KERN_EMERG "Thread overran stack, or stack corrupted\n");
853
854 sig = SIGKILL;
855 if (__die("Oops", regs, error_code))
856 sig = 0;
857
858 /* Executive summary in case the body of the oops scrolled away */
859 printk(KERN_DEFAULT "CR2: %016lx\n", address);
860
861 oops_end(flags, regs, sig);
862}
863
864/*
865 * Print out info about fatal segfaults, if the show_unhandled_signals
866 * sysctl is set:
867 */
868static inline void
869show_signal_msg(struct pt_regs *regs, unsigned long error_code,
870 unsigned long address, struct task_struct *tsk)
871{
872 const char *loglvl = task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG;
873
874 if (!unhandled_signal(tsk, SIGSEGV))
875 return;
876
877 if (!printk_ratelimit())
878 return;
879
880 printk("%s%s[%d]: segfault at %lx ip %px sp %px error %lx",
881 loglvl, tsk->comm, task_pid_nr(tsk), address,
882 (void *)regs->ip, (void *)regs->sp, error_code);
883
884 print_vma_addr(KERN_CONT " in ", regs->ip);
885
886 printk(KERN_CONT "\n");
887
888 show_opcodes(regs, loglvl);
889}
890
891/*
892 * The (legacy) vsyscall page is the long page in the kernel portion
893 * of the address space that has user-accessible permissions.
894 */
895static bool is_vsyscall_vaddr(unsigned long vaddr)
896{
897 return unlikely((vaddr & PAGE_MASK) == VSYSCALL_ADDR);
898}
899
900static void
901__bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
902 unsigned long address, u32 pkey, int si_code)
903{
904 struct task_struct *tsk = current;
905
906 /* User mode accesses just cause a SIGSEGV */
907 if (user_mode(regs) && (error_code & X86_PF_USER)) {
908 /*
909 * It's possible to have interrupts off here:
910 */
911 local_irq_enable();
912
913 /*
914 * Valid to do another page fault here because this one came
915 * from user space:
916 */
917 if (is_prefetch(regs, error_code, address))
918 return;
919
920 if (is_errata100(regs, address))
921 return;
922
923 /*
924 * To avoid leaking information about the kernel page table
925 * layout, pretend that user-mode accesses to kernel addresses
926 * are always protection faults.
927 */
928 if (address >= TASK_SIZE_MAX)
929 error_code |= X86_PF_PROT;
930
931 if (likely(show_unhandled_signals))
932 show_signal_msg(regs, error_code, address, tsk);
933
934 set_signal_archinfo(address, error_code);
935
936 if (si_code == SEGV_PKUERR)
937 force_sig_pkuerr((void __user *)address, pkey);
938
939 force_sig_fault(SIGSEGV, si_code, (void __user *)address, tsk);
940
941 return;
942 }
943
944 if (is_f00f_bug(regs, address))
945 return;
946
947 no_context(regs, error_code, address, SIGSEGV, si_code);
948}
949
950static noinline void
951bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
952 unsigned long address)
953{
954 __bad_area_nosemaphore(regs, error_code, address, 0, SEGV_MAPERR);
955}
956
957static void
958__bad_area(struct pt_regs *regs, unsigned long error_code,
959 unsigned long address, u32 pkey, int si_code)
960{
961 struct mm_struct *mm = current->mm;
962 /*
963 * Something tried to access memory that isn't in our memory map..
964 * Fix it, but check if it's kernel or user first..
965 */
966 up_read(&mm->mmap_sem);
967
968 __bad_area_nosemaphore(regs, error_code, address, pkey, si_code);
969}
970
971static noinline void
972bad_area(struct pt_regs *regs, unsigned long error_code, unsigned long address)
973{
974 __bad_area(regs, error_code, address, 0, SEGV_MAPERR);
975}
976
977static inline bool bad_area_access_from_pkeys(unsigned long error_code,
978 struct vm_area_struct *vma)
979{
980 /* This code is always called on the current mm */
981 bool foreign = false;
982
983 if (!boot_cpu_has(X86_FEATURE_OSPKE))
984 return false;
985 if (error_code & X86_PF_PK)
986 return true;
987 /* this checks permission keys on the VMA: */
988 if (!arch_vma_access_permitted(vma, (error_code & X86_PF_WRITE),
989 (error_code & X86_PF_INSTR), foreign))
990 return true;
991 return false;
992}
993
994static noinline void
995bad_area_access_error(struct pt_regs *regs, unsigned long error_code,
996 unsigned long address, struct vm_area_struct *vma)
997{
998 /*
999 * This OSPKE check is not strictly necessary at runtime.
1000 * But, doing it this way allows compiler optimizations
1001 * if pkeys are compiled out.
1002 */
1003 if (bad_area_access_from_pkeys(error_code, vma)) {
1004 /*
1005 * A protection key fault means that the PKRU value did not allow
1006 * access to some PTE. Userspace can figure out what PKRU was
1007 * from the XSAVE state. This function captures the pkey from
1008 * the vma and passes it to userspace so userspace can discover
1009 * which protection key was set on the PTE.
1010 *
1011 * If we get here, we know that the hardware signaled a X86_PF_PK
1012 * fault and that there was a VMA once we got in the fault
1013 * handler. It does *not* guarantee that the VMA we find here
1014 * was the one that we faulted on.
1015 *
1016 * 1. T1 : mprotect_key(foo, PAGE_SIZE, pkey=4);
1017 * 2. T1 : set PKRU to deny access to pkey=4, touches page
1018 * 3. T1 : faults...
1019 * 4. T2: mprotect_key(foo, PAGE_SIZE, pkey=5);
1020 * 5. T1 : enters fault handler, takes mmap_sem, etc...
1021 * 6. T1 : reaches here, sees vma_pkey(vma)=5, when we really
1022 * faulted on a pte with its pkey=4.
1023 */
1024 u32 pkey = vma_pkey(vma);
1025
1026 __bad_area(regs, error_code, address, pkey, SEGV_PKUERR);
1027 } else {
1028 __bad_area(regs, error_code, address, 0, SEGV_ACCERR);
1029 }
1030}
1031
1032static void
1033do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address,
1034 vm_fault_t fault)
1035{
1036 struct task_struct *tsk = current;
1037
1038 /* Kernel mode? Handle exceptions or die: */
1039 if (!(error_code & X86_PF_USER)) {
1040 no_context(regs, error_code, address, SIGBUS, BUS_ADRERR);
1041 return;
1042 }
1043
1044 /* User-space => ok to do another page fault: */
1045 if (is_prefetch(regs, error_code, address))
1046 return;
1047
1048 set_signal_archinfo(address, error_code);
1049
1050#ifdef CONFIG_MEMORY_FAILURE
1051 if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) {
1052 unsigned lsb = 0;
1053
1054 pr_err(
1055 "MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n",
1056 tsk->comm, tsk->pid, address);
1057 if (fault & VM_FAULT_HWPOISON_LARGE)
1058 lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault));
1059 if (fault & VM_FAULT_HWPOISON)
1060 lsb = PAGE_SHIFT;
1061 force_sig_mceerr(BUS_MCEERR_AR, (void __user *)address, lsb, tsk);
1062 return;
1063 }
1064#endif
1065 force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)address, tsk);
1066}
1067
1068static noinline void
1069mm_fault_error(struct pt_regs *regs, unsigned long error_code,
1070 unsigned long address, vm_fault_t fault)
1071{
1072 if (fatal_signal_pending(current) && !(error_code & X86_PF_USER)) {
1073 no_context(regs, error_code, address, 0, 0);
1074 return;
1075 }
1076
1077 if (fault & VM_FAULT_OOM) {
1078 /* Kernel mode? Handle exceptions or die: */
1079 if (!(error_code & X86_PF_USER)) {
1080 no_context(regs, error_code, address,
1081 SIGSEGV, SEGV_MAPERR);
1082 return;
1083 }
1084
1085 /*
1086 * We ran out of memory, call the OOM killer, and return the
1087 * userspace (which will retry the fault, or kill us if we got
1088 * oom-killed):
1089 */
1090 pagefault_out_of_memory();
1091 } else {
1092 if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON|
1093 VM_FAULT_HWPOISON_LARGE))
1094 do_sigbus(regs, error_code, address, fault);
1095 else if (fault & VM_FAULT_SIGSEGV)
1096 bad_area_nosemaphore(regs, error_code, address);
1097 else
1098 BUG();
1099 }
1100}
1101
1102static int spurious_kernel_fault_check(unsigned long error_code, pte_t *pte)
1103{
1104 if ((error_code & X86_PF_WRITE) && !pte_write(*pte))
1105 return 0;
1106
1107 if ((error_code & X86_PF_INSTR) && !pte_exec(*pte))
1108 return 0;
1109
1110 return 1;
1111}
1112
1113/*
1114 * Handle a spurious fault caused by a stale TLB entry.
1115 *
1116 * This allows us to lazily refresh the TLB when increasing the
1117 * permissions of a kernel page (RO -> RW or NX -> X). Doing it
1118 * eagerly is very expensive since that implies doing a full
1119 * cross-processor TLB flush, even if no stale TLB entries exist
1120 * on other processors.
1121 *
1122 * Spurious faults may only occur if the TLB contains an entry with
1123 * fewer permission than the page table entry. Non-present (P = 0)
1124 * and reserved bit (R = 1) faults are never spurious.
1125 *
1126 * There are no security implications to leaving a stale TLB when
1127 * increasing the permissions on a page.
1128 *
1129 * Returns non-zero if a spurious fault was handled, zero otherwise.
1130 *
1131 * See Intel Developer's Manual Vol 3 Section 4.10.4.3, bullet 3
1132 * (Optional Invalidation).
1133 */
1134static noinline int
1135spurious_kernel_fault(unsigned long error_code, unsigned long address)
1136{
1137 pgd_t *pgd;
1138 p4d_t *p4d;
1139 pud_t *pud;
1140 pmd_t *pmd;
1141 pte_t *pte;
1142 int ret;
1143
1144 /*
1145 * Only writes to RO or instruction fetches from NX may cause
1146 * spurious faults.
1147 *
1148 * These could be from user or supervisor accesses but the TLB
1149 * is only lazily flushed after a kernel mapping protection
1150 * change, so user accesses are not expected to cause spurious
1151 * faults.
1152 */
1153 if (error_code != (X86_PF_WRITE | X86_PF_PROT) &&
1154 error_code != (X86_PF_INSTR | X86_PF_PROT))
1155 return 0;
1156
1157 pgd = init_mm.pgd + pgd_index(address);
1158 if (!pgd_present(*pgd))
1159 return 0;
1160
1161 p4d = p4d_offset(pgd, address);
1162 if (!p4d_present(*p4d))
1163 return 0;
1164
1165 if (p4d_large(*p4d))
1166 return spurious_kernel_fault_check(error_code, (pte_t *) p4d);
1167
1168 pud = pud_offset(p4d, address);
1169 if (!pud_present(*pud))
1170 return 0;
1171
1172 if (pud_large(*pud))
1173 return spurious_kernel_fault_check(error_code, (pte_t *) pud);
1174
1175 pmd = pmd_offset(pud, address);
1176 if (!pmd_present(*pmd))
1177 return 0;
1178
1179 if (pmd_large(*pmd))
1180 return spurious_kernel_fault_check(error_code, (pte_t *) pmd);
1181
1182 pte = pte_offset_kernel(pmd, address);
1183 if (!pte_present(*pte))
1184 return 0;
1185
1186 ret = spurious_kernel_fault_check(error_code, pte);
1187 if (!ret)
1188 return 0;
1189
1190 /*
1191 * Make sure we have permissions in PMD.
1192 * If not, then there's a bug in the page tables:
1193 */
1194 ret = spurious_kernel_fault_check(error_code, (pte_t *) pmd);
1195 WARN_ONCE(!ret, "PMD has incorrect permission bits\n");
1196
1197 return ret;
1198}
1199NOKPROBE_SYMBOL(spurious_kernel_fault);
1200
1201int show_unhandled_signals = 1;
1202
1203static inline int
1204access_error(unsigned long error_code, struct vm_area_struct *vma)
1205{
1206 /* This is only called for the current mm, so: */
1207 bool foreign = false;
1208
1209 /*
1210 * Read or write was blocked by protection keys. This is
1211 * always an unconditional error and can never result in
1212 * a follow-up action to resolve the fault, like a COW.
1213 */
1214 if (error_code & X86_PF_PK)
1215 return 1;
1216
1217 /*
1218 * Make sure to check the VMA so that we do not perform
1219 * faults just to hit a X86_PF_PK as soon as we fill in a
1220 * page.
1221 */
1222 if (!arch_vma_access_permitted(vma, (error_code & X86_PF_WRITE),
1223 (error_code & X86_PF_INSTR), foreign))
1224 return 1;
1225
1226 if (error_code & X86_PF_WRITE) {
1227 /* write, present and write, not present: */
1228 if (unlikely(!(vma->vm_flags & VM_WRITE)))
1229 return 1;
1230 return 0;
1231 }
1232
1233 /* read, present: */
1234 if (unlikely(error_code & X86_PF_PROT))
1235 return 1;
1236
1237 /* read, not present: */
1238 if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE))))
1239 return 1;
1240
1241 return 0;
1242}
1243
1244static int fault_in_kernel_space(unsigned long address)
1245{
1246 /*
1247 * On 64-bit systems, the vsyscall page is at an address above
1248 * TASK_SIZE_MAX, but is not considered part of the kernel
1249 * address space.
1250 */
1251 if (IS_ENABLED(CONFIG_X86_64) && is_vsyscall_vaddr(address))
1252 return false;
1253
1254 return address >= TASK_SIZE_MAX;
1255}
1256
1257/*
1258 * Called for all faults where 'address' is part of the kernel address
1259 * space. Might get called for faults that originate from *code* that
1260 * ran in userspace or the kernel.
1261 */
1262static void
1263do_kern_addr_fault(struct pt_regs *regs, unsigned long hw_error_code,
1264 unsigned long address)
1265{
1266 /*
1267 * Protection keys exceptions only happen on user pages. We
1268 * have no user pages in the kernel portion of the address
1269 * space, so do not expect them here.
1270 */
1271 WARN_ON_ONCE(hw_error_code & X86_PF_PK);
1272
1273 /*
1274 * We can fault-in kernel-space virtual memory on-demand. The
1275 * 'reference' page table is init_mm.pgd.
1276 *
1277 * NOTE! We MUST NOT take any locks for this case. We may
1278 * be in an interrupt or a critical region, and should
1279 * only copy the information from the master page table,
1280 * nothing more.
1281 *
1282 * Before doing this on-demand faulting, ensure that the
1283 * fault is not any of the following:
1284 * 1. A fault on a PTE with a reserved bit set.
1285 * 2. A fault caused by a user-mode access. (Do not demand-
1286 * fault kernel memory due to user-mode accesses).
1287 * 3. A fault caused by a page-level protection violation.
1288 * (A demand fault would be on a non-present page which
1289 * would have X86_PF_PROT==0).
1290 */
1291 if (!(hw_error_code & (X86_PF_RSVD | X86_PF_USER | X86_PF_PROT))) {
1292 if (vmalloc_fault(address) >= 0)
1293 return;
1294 }
1295
1296 /* Was the fault spurious, caused by lazy TLB invalidation? */
1297 if (spurious_kernel_fault(hw_error_code, address))
1298 return;
1299
1300 /* kprobes don't want to hook the spurious faults: */
1301 if (kprobes_fault(regs))
1302 return;
1303
1304 /*
1305 * Note, despite being a "bad area", there are quite a few
1306 * acceptable reasons to get here, such as erratum fixups
1307 * and handling kernel code that can fault, like get_user().
1308 *
1309 * Don't take the mm semaphore here. If we fixup a prefetch
1310 * fault we could otherwise deadlock:
1311 */
1312 bad_area_nosemaphore(regs, hw_error_code, address);
1313}
1314NOKPROBE_SYMBOL(do_kern_addr_fault);
1315
1316/* Handle faults in the user portion of the address space */
1317static inline
1318void do_user_addr_fault(struct pt_regs *regs,
1319 unsigned long hw_error_code,
1320 unsigned long address)
1321{
1322 struct vm_area_struct *vma;
1323 struct task_struct *tsk;
1324 struct mm_struct *mm;
1325 vm_fault_t fault, major = 0;
1326 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
1327
1328 tsk = current;
1329 mm = tsk->mm;
1330
1331 /* kprobes don't want to hook the spurious faults: */
1332 if (unlikely(kprobes_fault(regs)))
1333 return;
1334
1335 /*
1336 * Reserved bits are never expected to be set on
1337 * entries in the user portion of the page tables.
1338 */
1339 if (unlikely(hw_error_code & X86_PF_RSVD))
1340 pgtable_bad(regs, hw_error_code, address);
1341
1342 /*
1343 * If SMAP is on, check for invalid kernel (supervisor) access to user
1344 * pages in the user address space. The odd case here is WRUSS,
1345 * which, according to the preliminary documentation, does not respect
1346 * SMAP and will have the USER bit set so, in all cases, SMAP
1347 * enforcement appears to be consistent with the USER bit.
1348 */
1349 if (unlikely(cpu_feature_enabled(X86_FEATURE_SMAP) &&
1350 !(hw_error_code & X86_PF_USER) &&
1351 !(regs->flags & X86_EFLAGS_AC)))
1352 {
1353 bad_area_nosemaphore(regs, hw_error_code, address);
1354 return;
1355 }
1356
1357 /*
1358 * If we're in an interrupt, have no user context or are running
1359 * in a region with pagefaults disabled then we must not take the fault
1360 */
1361 if (unlikely(faulthandler_disabled() || !mm)) {
1362 bad_area_nosemaphore(regs, hw_error_code, address);
1363 return;
1364 }
1365
1366 /*
1367 * It's safe to allow irq's after cr2 has been saved and the
1368 * vmalloc fault has been handled.
1369 *
1370 * User-mode registers count as a user access even for any
1371 * potential system fault or CPU buglet:
1372 */
1373 if (user_mode(regs)) {
1374 local_irq_enable();
1375 flags |= FAULT_FLAG_USER;
1376 } else {
1377 if (regs->flags & X86_EFLAGS_IF)
1378 local_irq_enable();
1379 }
1380
1381 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
1382
1383 if (hw_error_code & X86_PF_WRITE)
1384 flags |= FAULT_FLAG_WRITE;
1385 if (hw_error_code & X86_PF_INSTR)
1386 flags |= FAULT_FLAG_INSTRUCTION;
1387
1388#ifdef CONFIG_X86_64
1389 /*
1390 * Instruction fetch faults in the vsyscall page might need
1391 * emulation. The vsyscall page is at a high address
1392 * (>PAGE_OFFSET), but is considered to be part of the user
1393 * address space.
1394 *
1395 * The vsyscall page does not have a "real" VMA, so do this
1396 * emulation before we go searching for VMAs.
1397 */
1398 if ((hw_error_code & X86_PF_INSTR) && is_vsyscall_vaddr(address)) {
1399 if (emulate_vsyscall(regs, address))
1400 return;
1401 }
1402#endif
1403
1404 /*
1405 * Kernel-mode access to the user address space should only occur
1406 * on well-defined single instructions listed in the exception
1407 * tables. But, an erroneous kernel fault occurring outside one of
1408 * those areas which also holds mmap_sem might deadlock attempting
1409 * to validate the fault against the address space.
1410 *
1411 * Only do the expensive exception table search when we might be at
1412 * risk of a deadlock. This happens if we
1413 * 1. Failed to acquire mmap_sem, and
1414 * 2. The access did not originate in userspace.
1415 */
1416 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
1417 if (!user_mode(regs) && !search_exception_tables(regs->ip)) {
1418 /*
1419 * Fault from code in kernel from
1420 * which we do not expect faults.
1421 */
1422 bad_area_nosemaphore(regs, hw_error_code, address);
1423 return;
1424 }
1425retry:
1426 down_read(&mm->mmap_sem);
1427 } else {
1428 /*
1429 * The above down_read_trylock() might have succeeded in
1430 * which case we'll have missed the might_sleep() from
1431 * down_read():
1432 */
1433 might_sleep();
1434 }
1435
1436 vma = find_vma(mm, address);
1437 if (unlikely(!vma)) {
1438 bad_area(regs, hw_error_code, address);
1439 return;
1440 }
1441 if (likely(vma->vm_start <= address))
1442 goto good_area;
1443 if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) {
1444 bad_area(regs, hw_error_code, address);
1445 return;
1446 }
1447 if (unlikely(expand_stack(vma, address))) {
1448 bad_area(regs, hw_error_code, address);
1449 return;
1450 }
1451
1452 /*
1453 * Ok, we have a good vm_area for this memory access, so
1454 * we can handle it..
1455 */
1456good_area:
1457 if (unlikely(access_error(hw_error_code, vma))) {
1458 bad_area_access_error(regs, hw_error_code, address, vma);
1459 return;
1460 }
1461
1462 /*
1463 * If for any reason at all we couldn't handle the fault,
1464 * make sure we exit gracefully rather than endlessly redo
1465 * the fault. Since we never set FAULT_FLAG_RETRY_NOWAIT, if
1466 * we get VM_FAULT_RETRY back, the mmap_sem has been unlocked.
1467 *
1468 * Note that handle_userfault() may also release and reacquire mmap_sem
1469 * (and not return with VM_FAULT_RETRY), when returning to userland to
1470 * repeat the page fault later with a VM_FAULT_NOPAGE retval
1471 * (potentially after handling any pending signal during the return to
1472 * userland). The return to userland is identified whenever
1473 * FAULT_FLAG_USER|FAULT_FLAG_KILLABLE are both set in flags.
1474 */
1475 fault = handle_mm_fault(vma, address, flags);
1476 major |= fault & VM_FAULT_MAJOR;
1477
1478 /*
1479 * If we need to retry the mmap_sem has already been released,
1480 * and if there is a fatal signal pending there is no guarantee
1481 * that we made any progress. Handle this case first.
1482 */
1483 if (unlikely(fault & VM_FAULT_RETRY)) {
1484 /* Retry at most once */
1485 if (flags & FAULT_FLAG_ALLOW_RETRY) {
1486 flags &= ~FAULT_FLAG_ALLOW_RETRY;
1487 flags |= FAULT_FLAG_TRIED;
1488 if (!fatal_signal_pending(tsk))
1489 goto retry;
1490 }
1491
1492 /* User mode? Just return to handle the fatal exception */
1493 if (flags & FAULT_FLAG_USER)
1494 return;
1495
1496 /* Not returning to user mode? Handle exceptions or die: */
1497 no_context(regs, hw_error_code, address, SIGBUS, BUS_ADRERR);
1498 return;
1499 }
1500
1501 up_read(&mm->mmap_sem);
1502 if (unlikely(fault & VM_FAULT_ERROR)) {
1503 mm_fault_error(regs, hw_error_code, address, fault);
1504 return;
1505 }
1506
1507 /*
1508 * Major/minor page fault accounting. If any of the events
1509 * returned VM_FAULT_MAJOR, we account it as a major fault.
1510 */
1511 if (major) {
1512 tsk->maj_flt++;
1513 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, address);
1514 } else {
1515 tsk->min_flt++;
1516 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, address);
1517 }
1518
1519 check_v8086_mode(regs, address, tsk);
1520}
1521NOKPROBE_SYMBOL(do_user_addr_fault);
1522
1523/*
1524 * This routine handles page faults. It determines the address,
1525 * and the problem, and then passes it off to one of the appropriate
1526 * routines.
1527 */
1528static noinline void
1529__do_page_fault(struct pt_regs *regs, unsigned long hw_error_code,
1530 unsigned long address)
1531{
1532 prefetchw(&current->mm->mmap_sem);
1533
1534 if (unlikely(kmmio_fault(regs, address)))
1535 return;
1536
1537 /* Was the fault on kernel-controlled part of the address space? */
1538 if (unlikely(fault_in_kernel_space(address)))
1539 do_kern_addr_fault(regs, hw_error_code, address);
1540 else
1541 do_user_addr_fault(regs, hw_error_code, address);
1542}
1543NOKPROBE_SYMBOL(__do_page_fault);
1544
1545static nokprobe_inline void
1546trace_page_fault_entries(unsigned long address, struct pt_regs *regs,
1547 unsigned long error_code)
1548{
1549 if (user_mode(regs))
1550 trace_page_fault_user(address, regs, error_code);
1551 else
1552 trace_page_fault_kernel(address, regs, error_code);
1553}
1554
1555/*
1556 * We must have this function blacklisted from kprobes, tagged with notrace
1557 * and call read_cr2() before calling anything else. To avoid calling any
1558 * kind of tracing machinery before we've observed the CR2 value.
1559 *
1560 * exception_{enter,exit}() contains all sorts of tracepoints.
1561 */
1562dotraplinkage void notrace
1563do_page_fault(struct pt_regs *regs, unsigned long error_code)
1564{
1565 unsigned long address = read_cr2(); /* Get the faulting address */
1566 enum ctx_state prev_state;
1567
1568 prev_state = exception_enter();
1569 if (trace_pagefault_enabled())
1570 trace_page_fault_entries(address, regs, error_code);
1571
1572 __do_page_fault(regs, error_code, address);
1573 exception_exit(prev_state);
1574}
1575NOKPROBE_SYMBOL(do_page_fault);
1576