xref: /openbmc/linux/arch/x86/mm/fault.c (revision 727dede0)
1 /*
2  *  Copyright (C) 1995  Linus Torvalds
3  *  Copyright (C) 2001, 2002 Andi Kleen, SuSE Labs.
4  *  Copyright (C) 2008-2009, Red Hat Inc., Ingo Molnar
5  */
6 #include <linux/sched.h>		/* test_thread_flag(), ...	*/
7 #include <linux/sched/task_stack.h>	/* task_stack_*(), ...		*/
8 #include <linux/kdebug.h>		/* oops_begin/end, ...		*/
9 #include <linux/extable.h>		/* search_exception_tables	*/
10 #include <linux/bootmem.h>		/* max_low_pfn			*/
11 #include <linux/kprobes.h>		/* NOKPROBE_SYMBOL, ...		*/
12 #include <linux/mmiotrace.h>		/* kmmio_handler, ...		*/
13 #include <linux/perf_event.h>		/* perf_sw_event		*/
14 #include <linux/hugetlb.h>		/* hstate_index_to_shift	*/
15 #include <linux/prefetch.h>		/* prefetchw			*/
16 #include <linux/context_tracking.h>	/* exception_enter(), ...	*/
17 #include <linux/uaccess.h>		/* faulthandler_disabled()	*/
18 
19 #include <asm/cpufeature.h>		/* boot_cpu_has, ...		*/
20 #include <asm/traps.h>			/* dotraplinkage, ...		*/
21 #include <asm/pgalloc.h>		/* pgd_*(), ...			*/
22 #include <asm/kmemcheck.h>		/* kmemcheck_*(), ...		*/
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  * Page fault error code bits:
33  *
34  *   bit 0 ==	 0: no page found	1: protection fault
35  *   bit 1 ==	 0: read access		1: write access
36  *   bit 2 ==	 0: kernel-mode access	1: user-mode access
37  *   bit 3 ==				1: use of reserved bit detected
38  *   bit 4 ==				1: fault was an instruction fetch
39  *   bit 5 ==				1: protection keys block access
40  */
41 enum x86_pf_error_code {
42 
43 	PF_PROT		=		1 << 0,
44 	PF_WRITE	=		1 << 1,
45 	PF_USER		=		1 << 2,
46 	PF_RSVD		=		1 << 3,
47 	PF_INSTR	=		1 << 4,
48 	PF_PK		=		1 << 5,
49 };
50 
51 /*
52  * Returns 0 if mmiotrace is disabled, or if the fault is not
53  * handled by mmiotrace:
54  */
55 static nokprobe_inline int
56 kmmio_fault(struct pt_regs *regs, unsigned long addr)
57 {
58 	if (unlikely(is_kmmio_active()))
59 		if (kmmio_handler(regs, addr) == 1)
60 			return -1;
61 	return 0;
62 }
63 
64 static nokprobe_inline int kprobes_fault(struct pt_regs *regs)
65 {
66 	int ret = 0;
67 
68 	/* kprobe_running() needs smp_processor_id() */
69 	if (kprobes_built_in() && !user_mode(regs)) {
70 		preempt_disable();
71 		if (kprobe_running() && kprobe_fault_handler(regs, 14))
72 			ret = 1;
73 		preempt_enable();
74 	}
75 
76 	return ret;
77 }
78 
79 /*
80  * Prefetch quirks:
81  *
82  * 32-bit mode:
83  *
84  *   Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch.
85  *   Check that here and ignore it.
86  *
87  * 64-bit mode:
88  *
89  *   Sometimes the CPU reports invalid exceptions on prefetch.
90  *   Check that here and ignore it.
91  *
92  * Opcode checker based on code by Richard Brunner.
93  */
94 static inline int
95 check_prefetch_opcode(struct pt_regs *regs, unsigned char *instr,
96 		      unsigned char opcode, int *prefetch)
97 {
98 	unsigned char instr_hi = opcode & 0xf0;
99 	unsigned char instr_lo = opcode & 0x0f;
100 
101 	switch (instr_hi) {
102 	case 0x20:
103 	case 0x30:
104 		/*
105 		 * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes.
106 		 * In X86_64 long mode, the CPU will signal invalid
107 		 * opcode if some of these prefixes are present so
108 		 * X86_64 will never get here anyway
109 		 */
110 		return ((instr_lo & 7) == 0x6);
111 #ifdef CONFIG_X86_64
112 	case 0x40:
113 		/*
114 		 * In AMD64 long mode 0x40..0x4F are valid REX prefixes
115 		 * Need to figure out under what instruction mode the
116 		 * instruction was issued. Could check the LDT for lm,
117 		 * but for now it's good enough to assume that long
118 		 * mode only uses well known segments or kernel.
119 		 */
120 		return (!user_mode(regs) || user_64bit_mode(regs));
121 #endif
122 	case 0x60:
123 		/* 0x64 thru 0x67 are valid prefixes in all modes. */
124 		return (instr_lo & 0xC) == 0x4;
125 	case 0xF0:
126 		/* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */
127 		return !instr_lo || (instr_lo>>1) == 1;
128 	case 0x00:
129 		/* Prefetch instruction is 0x0F0D or 0x0F18 */
130 		if (probe_kernel_address(instr, opcode))
131 			return 0;
132 
133 		*prefetch = (instr_lo == 0xF) &&
134 			(opcode == 0x0D || opcode == 0x18);
135 		return 0;
136 	default:
137 		return 0;
138 	}
139 }
140 
141 static int
142 is_prefetch(struct pt_regs *regs, unsigned long error_code, unsigned long addr)
143 {
144 	unsigned char *max_instr;
145 	unsigned char *instr;
146 	int prefetch = 0;
147 
148 	/*
149 	 * If it was a exec (instruction fetch) fault on NX page, then
150 	 * do not ignore the fault:
151 	 */
152 	if (error_code & PF_INSTR)
153 		return 0;
154 
155 	instr = (void *)convert_ip_to_linear(current, regs);
156 	max_instr = instr + 15;
157 
158 	if (user_mode(regs) && instr >= (unsigned char *)TASK_SIZE_MAX)
159 		return 0;
160 
161 	while (instr < max_instr) {
162 		unsigned char opcode;
163 
164 		if (probe_kernel_address(instr, opcode))
165 			break;
166 
167 		instr++;
168 
169 		if (!check_prefetch_opcode(regs, instr, opcode, &prefetch))
170 			break;
171 	}
172 	return prefetch;
173 }
174 
175 /*
176  * A protection key fault means that the PKRU value did not allow
177  * access to some PTE.  Userspace can figure out what PKRU was
178  * from the XSAVE state, and this function fills out a field in
179  * siginfo so userspace can discover which protection key was set
180  * on the PTE.
181  *
182  * If we get here, we know that the hardware signaled a PF_PK
183  * fault and that there was a VMA once we got in the fault
184  * handler.  It does *not* guarantee that the VMA we find here
185  * was the one that we faulted on.
186  *
187  * 1. T1   : mprotect_key(foo, PAGE_SIZE, pkey=4);
188  * 2. T1   : set PKRU to deny access to pkey=4, touches page
189  * 3. T1   : faults...
190  * 4.    T2: mprotect_key(foo, PAGE_SIZE, pkey=5);
191  * 5. T1   : enters fault handler, takes mmap_sem, etc...
192  * 6. T1   : reaches here, sees vma_pkey(vma)=5, when we really
193  *	     faulted on a pte with its pkey=4.
194  */
195 static void fill_sig_info_pkey(int si_code, siginfo_t *info, u32 *pkey)
196 {
197 	/* This is effectively an #ifdef */
198 	if (!boot_cpu_has(X86_FEATURE_OSPKE))
199 		return;
200 
201 	/* Fault not from Protection Keys: nothing to do */
202 	if (si_code != SEGV_PKUERR)
203 		return;
204 	/*
205 	 * force_sig_info_fault() is called from a number of
206 	 * contexts, some of which have a VMA and some of which
207 	 * do not.  The PF_PK handing happens after we have a
208 	 * valid VMA, so we should never reach this without a
209 	 * valid VMA.
210 	 */
211 	if (!pkey) {
212 		WARN_ONCE(1, "PKU fault with no VMA passed in");
213 		info->si_pkey = 0;
214 		return;
215 	}
216 	/*
217 	 * si_pkey should be thought of as a strong hint, but not
218 	 * absolutely guranteed to be 100% accurate because of
219 	 * the race explained above.
220 	 */
221 	info->si_pkey = *pkey;
222 }
223 
224 static void
225 force_sig_info_fault(int si_signo, int si_code, unsigned long address,
226 		     struct task_struct *tsk, u32 *pkey, int fault)
227 {
228 	unsigned lsb = 0;
229 	siginfo_t info;
230 
231 	info.si_signo	= si_signo;
232 	info.si_errno	= 0;
233 	info.si_code	= si_code;
234 	info.si_addr	= (void __user *)address;
235 	if (fault & VM_FAULT_HWPOISON_LARGE)
236 		lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault));
237 	if (fault & VM_FAULT_HWPOISON)
238 		lsb = PAGE_SHIFT;
239 	info.si_addr_lsb = lsb;
240 
241 	fill_sig_info_pkey(si_code, &info, pkey);
242 
243 	force_sig_info(si_signo, &info, tsk);
244 }
245 
246 DEFINE_SPINLOCK(pgd_lock);
247 LIST_HEAD(pgd_list);
248 
249 #ifdef CONFIG_X86_32
250 static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address)
251 {
252 	unsigned index = pgd_index(address);
253 	pgd_t *pgd_k;
254 	p4d_t *p4d, *p4d_k;
255 	pud_t *pud, *pud_k;
256 	pmd_t *pmd, *pmd_k;
257 
258 	pgd += index;
259 	pgd_k = init_mm.pgd + index;
260 
261 	if (!pgd_present(*pgd_k))
262 		return NULL;
263 
264 	/*
265 	 * set_pgd(pgd, *pgd_k); here would be useless on PAE
266 	 * and redundant with the set_pmd() on non-PAE. As would
267 	 * set_p4d/set_pud.
268 	 */
269 	p4d = p4d_offset(pgd, address);
270 	p4d_k = p4d_offset(pgd_k, address);
271 	if (!p4d_present(*p4d_k))
272 		return NULL;
273 
274 	pud = pud_offset(p4d, address);
275 	pud_k = pud_offset(p4d_k, address);
276 	if (!pud_present(*pud_k))
277 		return NULL;
278 
279 	pmd = pmd_offset(pud, address);
280 	pmd_k = pmd_offset(pud_k, address);
281 	if (!pmd_present(*pmd_k))
282 		return NULL;
283 
284 	if (!pmd_present(*pmd))
285 		set_pmd(pmd, *pmd_k);
286 	else
287 		BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k));
288 
289 	return pmd_k;
290 }
291 
292 void vmalloc_sync_all(void)
293 {
294 	unsigned long address;
295 
296 	if (SHARED_KERNEL_PMD)
297 		return;
298 
299 	for (address = VMALLOC_START & PMD_MASK;
300 	     address >= TASK_SIZE_MAX && address < FIXADDR_TOP;
301 	     address += PMD_SIZE) {
302 		struct page *page;
303 
304 		spin_lock(&pgd_lock);
305 		list_for_each_entry(page, &pgd_list, lru) {
306 			spinlock_t *pgt_lock;
307 			pmd_t *ret;
308 
309 			/* the pgt_lock only for Xen */
310 			pgt_lock = &pgd_page_get_mm(page)->page_table_lock;
311 
312 			spin_lock(pgt_lock);
313 			ret = vmalloc_sync_one(page_address(page), address);
314 			spin_unlock(pgt_lock);
315 
316 			if (!ret)
317 				break;
318 		}
319 		spin_unlock(&pgd_lock);
320 	}
321 }
322 
323 /*
324  * 32-bit:
325  *
326  *   Handle a fault on the vmalloc or module mapping area
327  */
328 static noinline int vmalloc_fault(unsigned long address)
329 {
330 	unsigned long pgd_paddr;
331 	pmd_t *pmd_k;
332 	pte_t *pte_k;
333 
334 	/* Make sure we are in vmalloc area: */
335 	if (!(address >= VMALLOC_START && address < VMALLOC_END))
336 		return -1;
337 
338 	WARN_ON_ONCE(in_nmi());
339 
340 	/*
341 	 * Synchronize this task's top level page-table
342 	 * with the 'reference' page table.
343 	 *
344 	 * Do _not_ use "current" here. We might be inside
345 	 * an interrupt in the middle of a task switch..
346 	 */
347 	pgd_paddr = read_cr3_pa();
348 	pmd_k = vmalloc_sync_one(__va(pgd_paddr), address);
349 	if (!pmd_k)
350 		return -1;
351 
352 	if (pmd_huge(*pmd_k))
353 		return 0;
354 
355 	pte_k = pte_offset_kernel(pmd_k, address);
356 	if (!pte_present(*pte_k))
357 		return -1;
358 
359 	return 0;
360 }
361 NOKPROBE_SYMBOL(vmalloc_fault);
362 
363 /*
364  * Did it hit the DOS screen memory VA from vm86 mode?
365  */
366 static inline void
367 check_v8086_mode(struct pt_regs *regs, unsigned long address,
368 		 struct task_struct *tsk)
369 {
370 #ifdef CONFIG_VM86
371 	unsigned long bit;
372 
373 	if (!v8086_mode(regs) || !tsk->thread.vm86)
374 		return;
375 
376 	bit = (address - 0xA0000) >> PAGE_SHIFT;
377 	if (bit < 32)
378 		tsk->thread.vm86->screen_bitmap |= 1 << bit;
379 #endif
380 }
381 
382 static bool low_pfn(unsigned long pfn)
383 {
384 	return pfn < max_low_pfn;
385 }
386 
387 static void dump_pagetable(unsigned long address)
388 {
389 	pgd_t *base = __va(read_cr3_pa());
390 	pgd_t *pgd = &base[pgd_index(address)];
391 	p4d_t *p4d;
392 	pud_t *pud;
393 	pmd_t *pmd;
394 	pte_t *pte;
395 
396 #ifdef CONFIG_X86_PAE
397 	pr_info("*pdpt = %016Lx ", pgd_val(*pgd));
398 	if (!low_pfn(pgd_val(*pgd) >> PAGE_SHIFT) || !pgd_present(*pgd))
399 		goto out;
400 #define pr_pde pr_cont
401 #else
402 #define pr_pde pr_info
403 #endif
404 	p4d = p4d_offset(pgd, address);
405 	pud = pud_offset(p4d, address);
406 	pmd = pmd_offset(pud, address);
407 	pr_pde("*pde = %0*Lx ", sizeof(*pmd) * 2, (u64)pmd_val(*pmd));
408 #undef pr_pde
409 
410 	/*
411 	 * We must not directly access the pte in the highpte
412 	 * case if the page table is located in highmem.
413 	 * And let's rather not kmap-atomic the pte, just in case
414 	 * it's allocated already:
415 	 */
416 	if (!low_pfn(pmd_pfn(*pmd)) || !pmd_present(*pmd) || pmd_large(*pmd))
417 		goto out;
418 
419 	pte = pte_offset_kernel(pmd, address);
420 	pr_cont("*pte = %0*Lx ", sizeof(*pte) * 2, (u64)pte_val(*pte));
421 out:
422 	pr_cont("\n");
423 }
424 
425 #else /* CONFIG_X86_64: */
426 
427 void vmalloc_sync_all(void)
428 {
429 	sync_global_pgds(VMALLOC_START & PGDIR_MASK, VMALLOC_END);
430 }
431 
432 /*
433  * 64-bit:
434  *
435  *   Handle a fault on the vmalloc area
436  */
437 static noinline int vmalloc_fault(unsigned long address)
438 {
439 	pgd_t *pgd, *pgd_ref;
440 	p4d_t *p4d, *p4d_ref;
441 	pud_t *pud, *pud_ref;
442 	pmd_t *pmd, *pmd_ref;
443 	pte_t *pte, *pte_ref;
444 
445 	/* Make sure we are in vmalloc area: */
446 	if (!(address >= VMALLOC_START && address < VMALLOC_END))
447 		return -1;
448 
449 	WARN_ON_ONCE(in_nmi());
450 
451 	/*
452 	 * Copy kernel mappings over when needed. This can also
453 	 * happen within a race in page table update. In the later
454 	 * case just flush:
455 	 */
456 	pgd = (pgd_t *)__va(read_cr3_pa()) + pgd_index(address);
457 	pgd_ref = pgd_offset_k(address);
458 	if (pgd_none(*pgd_ref))
459 		return -1;
460 
461 	if (pgd_none(*pgd)) {
462 		set_pgd(pgd, *pgd_ref);
463 		arch_flush_lazy_mmu_mode();
464 	} else if (CONFIG_PGTABLE_LEVELS > 4) {
465 		/*
466 		 * With folded p4d, pgd_none() is always false, so the pgd may
467 		 * point to an empty page table entry and pgd_page_vaddr()
468 		 * will return garbage.
469 		 *
470 		 * We will do the correct sanity check on the p4d level.
471 		 */
472 		BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref));
473 	}
474 
475 	/* With 4-level paging, copying happens on the p4d level. */
476 	p4d = p4d_offset(pgd, address);
477 	p4d_ref = p4d_offset(pgd_ref, address);
478 	if (p4d_none(*p4d_ref))
479 		return -1;
480 
481 	if (p4d_none(*p4d)) {
482 		set_p4d(p4d, *p4d_ref);
483 		arch_flush_lazy_mmu_mode();
484 	} else {
485 		BUG_ON(p4d_pfn(*p4d) != p4d_pfn(*p4d_ref));
486 	}
487 
488 	/*
489 	 * Below here mismatches are bugs because these lower tables
490 	 * are shared:
491 	 */
492 
493 	pud = pud_offset(p4d, address);
494 	pud_ref = pud_offset(p4d_ref, address);
495 	if (pud_none(*pud_ref))
496 		return -1;
497 
498 	if (pud_none(*pud) || pud_pfn(*pud) != pud_pfn(*pud_ref))
499 		BUG();
500 
501 	if (pud_huge(*pud))
502 		return 0;
503 
504 	pmd = pmd_offset(pud, address);
505 	pmd_ref = pmd_offset(pud_ref, address);
506 	if (pmd_none(*pmd_ref))
507 		return -1;
508 
509 	if (pmd_none(*pmd) || pmd_pfn(*pmd) != pmd_pfn(*pmd_ref))
510 		BUG();
511 
512 	if (pmd_huge(*pmd))
513 		return 0;
514 
515 	pte_ref = pte_offset_kernel(pmd_ref, address);
516 	if (!pte_present(*pte_ref))
517 		return -1;
518 
519 	pte = pte_offset_kernel(pmd, address);
520 
521 	/*
522 	 * Don't use pte_page here, because the mappings can point
523 	 * outside mem_map, and the NUMA hash lookup cannot handle
524 	 * that:
525 	 */
526 	if (!pte_present(*pte) || pte_pfn(*pte) != pte_pfn(*pte_ref))
527 		BUG();
528 
529 	return 0;
530 }
531 NOKPROBE_SYMBOL(vmalloc_fault);
532 
533 #ifdef CONFIG_CPU_SUP_AMD
534 static const char errata93_warning[] =
535 KERN_ERR
536 "******* Your BIOS seems to not contain a fix for K8 errata #93\n"
537 "******* Working around it, but it may cause SEGVs or burn power.\n"
538 "******* Please consider a BIOS update.\n"
539 "******* Disabling USB legacy in the BIOS may also help.\n";
540 #endif
541 
542 /*
543  * No vm86 mode in 64-bit mode:
544  */
545 static inline void
546 check_v8086_mode(struct pt_regs *regs, unsigned long address,
547 		 struct task_struct *tsk)
548 {
549 }
550 
551 static int bad_address(void *p)
552 {
553 	unsigned long dummy;
554 
555 	return probe_kernel_address((unsigned long *)p, dummy);
556 }
557 
558 static void dump_pagetable(unsigned long address)
559 {
560 	pgd_t *base = __va(read_cr3_pa());
561 	pgd_t *pgd = base + pgd_index(address);
562 	p4d_t *p4d;
563 	pud_t *pud;
564 	pmd_t *pmd;
565 	pte_t *pte;
566 
567 	if (bad_address(pgd))
568 		goto bad;
569 
570 	pr_info("PGD %lx ", pgd_val(*pgd));
571 
572 	if (!pgd_present(*pgd))
573 		goto out;
574 
575 	p4d = p4d_offset(pgd, address);
576 	if (bad_address(p4d))
577 		goto bad;
578 
579 	pr_cont("P4D %lx ", p4d_val(*p4d));
580 	if (!p4d_present(*p4d) || p4d_large(*p4d))
581 		goto out;
582 
583 	pud = pud_offset(p4d, address);
584 	if (bad_address(pud))
585 		goto bad;
586 
587 	pr_cont("PUD %lx ", pud_val(*pud));
588 	if (!pud_present(*pud) || pud_large(*pud))
589 		goto out;
590 
591 	pmd = pmd_offset(pud, address);
592 	if (bad_address(pmd))
593 		goto bad;
594 
595 	pr_cont("PMD %lx ", pmd_val(*pmd));
596 	if (!pmd_present(*pmd) || pmd_large(*pmd))
597 		goto out;
598 
599 	pte = pte_offset_kernel(pmd, address);
600 	if (bad_address(pte))
601 		goto bad;
602 
603 	pr_cont("PTE %lx", pte_val(*pte));
604 out:
605 	pr_cont("\n");
606 	return;
607 bad:
608 	pr_info("BAD\n");
609 }
610 
611 #endif /* CONFIG_X86_64 */
612 
613 /*
614  * Workaround for K8 erratum #93 & buggy BIOS.
615  *
616  * BIOS SMM functions are required to use a specific workaround
617  * to avoid corruption of the 64bit RIP register on C stepping K8.
618  *
619  * A lot of BIOS that didn't get tested properly miss this.
620  *
621  * The OS sees this as a page fault with the upper 32bits of RIP cleared.
622  * Try to work around it here.
623  *
624  * Note we only handle faults in kernel here.
625  * Does nothing on 32-bit.
626  */
627 static int is_errata93(struct pt_regs *regs, unsigned long address)
628 {
629 #if defined(CONFIG_X86_64) && defined(CONFIG_CPU_SUP_AMD)
630 	if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD
631 	    || boot_cpu_data.x86 != 0xf)
632 		return 0;
633 
634 	if (address != regs->ip)
635 		return 0;
636 
637 	if ((address >> 32) != 0)
638 		return 0;
639 
640 	address |= 0xffffffffUL << 32;
641 	if ((address >= (u64)_stext && address <= (u64)_etext) ||
642 	    (address >= MODULES_VADDR && address <= MODULES_END)) {
643 		printk_once(errata93_warning);
644 		regs->ip = address;
645 		return 1;
646 	}
647 #endif
648 	return 0;
649 }
650 
651 /*
652  * Work around K8 erratum #100 K8 in compat mode occasionally jumps
653  * to illegal addresses >4GB.
654  *
655  * We catch this in the page fault handler because these addresses
656  * are not reachable. Just detect this case and return.  Any code
657  * segment in LDT is compatibility mode.
658  */
659 static int is_errata100(struct pt_regs *regs, unsigned long address)
660 {
661 #ifdef CONFIG_X86_64
662 	if ((regs->cs == __USER32_CS || (regs->cs & (1<<2))) && (address >> 32))
663 		return 1;
664 #endif
665 	return 0;
666 }
667 
668 static int is_f00f_bug(struct pt_regs *regs, unsigned long address)
669 {
670 #ifdef CONFIG_X86_F00F_BUG
671 	unsigned long nr;
672 
673 	/*
674 	 * Pentium F0 0F C7 C8 bug workaround:
675 	 */
676 	if (boot_cpu_has_bug(X86_BUG_F00F)) {
677 		nr = (address - idt_descr.address) >> 3;
678 
679 		if (nr == 6) {
680 			do_invalid_op(regs, 0);
681 			return 1;
682 		}
683 	}
684 #endif
685 	return 0;
686 }
687 
688 static const char nx_warning[] = KERN_CRIT
689 "kernel tried to execute NX-protected page - exploit attempt? (uid: %d)\n";
690 static const char smep_warning[] = KERN_CRIT
691 "unable to execute userspace code (SMEP?) (uid: %d)\n";
692 
693 static void
694 show_fault_oops(struct pt_regs *regs, unsigned long error_code,
695 		unsigned long address)
696 {
697 	if (!oops_may_print())
698 		return;
699 
700 	if (error_code & PF_INSTR) {
701 		unsigned int level;
702 		pgd_t *pgd;
703 		pte_t *pte;
704 
705 		pgd = __va(read_cr3_pa());
706 		pgd += pgd_index(address);
707 
708 		pte = lookup_address_in_pgd(pgd, address, &level);
709 
710 		if (pte && pte_present(*pte) && !pte_exec(*pte))
711 			printk(nx_warning, from_kuid(&init_user_ns, current_uid()));
712 		if (pte && pte_present(*pte) && pte_exec(*pte) &&
713 				(pgd_flags(*pgd) & _PAGE_USER) &&
714 				(__read_cr4() & X86_CR4_SMEP))
715 			printk(smep_warning, from_kuid(&init_user_ns, current_uid()));
716 	}
717 
718 	printk(KERN_ALERT "BUG: unable to handle kernel ");
719 	if (address < PAGE_SIZE)
720 		printk(KERN_CONT "NULL pointer dereference");
721 	else
722 		printk(KERN_CONT "paging request");
723 
724 	printk(KERN_CONT " at %p\n", (void *) address);
725 	printk(KERN_ALERT "IP: %pS\n", (void *)regs->ip);
726 
727 	dump_pagetable(address);
728 }
729 
730 static noinline void
731 pgtable_bad(struct pt_regs *regs, unsigned long error_code,
732 	    unsigned long address)
733 {
734 	struct task_struct *tsk;
735 	unsigned long flags;
736 	int sig;
737 
738 	flags = oops_begin();
739 	tsk = current;
740 	sig = SIGKILL;
741 
742 	printk(KERN_ALERT "%s: Corrupted page table at address %lx\n",
743 	       tsk->comm, address);
744 	dump_pagetable(address);
745 
746 	tsk->thread.cr2		= address;
747 	tsk->thread.trap_nr	= X86_TRAP_PF;
748 	tsk->thread.error_code	= error_code;
749 
750 	if (__die("Bad pagetable", regs, error_code))
751 		sig = 0;
752 
753 	oops_end(flags, regs, sig);
754 }
755 
756 static noinline void
757 no_context(struct pt_regs *regs, unsigned long error_code,
758 	   unsigned long address, int signal, int si_code)
759 {
760 	struct task_struct *tsk = current;
761 	unsigned long flags;
762 	int sig;
763 
764 	/* Are we prepared to handle this kernel fault? */
765 	if (fixup_exception(regs, X86_TRAP_PF)) {
766 		/*
767 		 * Any interrupt that takes a fault gets the fixup. This makes
768 		 * the below recursive fault logic only apply to a faults from
769 		 * task context.
770 		 */
771 		if (in_interrupt())
772 			return;
773 
774 		/*
775 		 * Per the above we're !in_interrupt(), aka. task context.
776 		 *
777 		 * In this case we need to make sure we're not recursively
778 		 * faulting through the emulate_vsyscall() logic.
779 		 */
780 		if (current->thread.sig_on_uaccess_err && signal) {
781 			tsk->thread.trap_nr = X86_TRAP_PF;
782 			tsk->thread.error_code = error_code | PF_USER;
783 			tsk->thread.cr2 = address;
784 
785 			/* XXX: hwpoison faults will set the wrong code. */
786 			force_sig_info_fault(signal, si_code, address,
787 					     tsk, NULL, 0);
788 		}
789 
790 		/*
791 		 * Barring that, we can do the fixup and be happy.
792 		 */
793 		return;
794 	}
795 
796 #ifdef CONFIG_VMAP_STACK
797 	/*
798 	 * Stack overflow?  During boot, we can fault near the initial
799 	 * stack in the direct map, but that's not an overflow -- check
800 	 * that we're in vmalloc space to avoid this.
801 	 */
802 	if (is_vmalloc_addr((void *)address) &&
803 	    (((unsigned long)tsk->stack - 1 - address < PAGE_SIZE) ||
804 	     address - ((unsigned long)tsk->stack + THREAD_SIZE) < PAGE_SIZE)) {
805 		unsigned long stack = this_cpu_read(orig_ist.ist[DOUBLEFAULT_STACK]) - sizeof(void *);
806 		/*
807 		 * We're likely to be running with very little stack space
808 		 * left.  It's plausible that we'd hit this condition but
809 		 * double-fault even before we get this far, in which case
810 		 * we're fine: the double-fault handler will deal with it.
811 		 *
812 		 * We don't want to make it all the way into the oops code
813 		 * and then double-fault, though, because we're likely to
814 		 * break the console driver and lose most of the stack dump.
815 		 */
816 		asm volatile ("movq %[stack], %%rsp\n\t"
817 			      "call handle_stack_overflow\n\t"
818 			      "1: jmp 1b"
819 			      : ASM_CALL_CONSTRAINT
820 			      : "D" ("kernel stack overflow (page fault)"),
821 				"S" (regs), "d" (address),
822 				[stack] "rm" (stack));
823 		unreachable();
824 	}
825 #endif
826 
827 	/*
828 	 * 32-bit:
829 	 *
830 	 *   Valid to do another page fault here, because if this fault
831 	 *   had been triggered by is_prefetch fixup_exception would have
832 	 *   handled it.
833 	 *
834 	 * 64-bit:
835 	 *
836 	 *   Hall of shame of CPU/BIOS bugs.
837 	 */
838 	if (is_prefetch(regs, error_code, address))
839 		return;
840 
841 	if (is_errata93(regs, address))
842 		return;
843 
844 	/*
845 	 * Oops. The kernel tried to access some bad page. We'll have to
846 	 * terminate things with extreme prejudice:
847 	 */
848 	flags = oops_begin();
849 
850 	show_fault_oops(regs, error_code, address);
851 
852 	if (task_stack_end_corrupted(tsk))
853 		printk(KERN_EMERG "Thread overran stack, or stack corrupted\n");
854 
855 	tsk->thread.cr2		= address;
856 	tsk->thread.trap_nr	= X86_TRAP_PF;
857 	tsk->thread.error_code	= error_code;
858 
859 	sig = SIGKILL;
860 	if (__die("Oops", regs, error_code))
861 		sig = 0;
862 
863 	/* Executive summary in case the body of the oops scrolled away */
864 	printk(KERN_DEFAULT "CR2: %016lx\n", address);
865 
866 	oops_end(flags, regs, sig);
867 }
868 
869 /*
870  * Print out info about fatal segfaults, if the show_unhandled_signals
871  * sysctl is set:
872  */
873 static inline void
874 show_signal_msg(struct pt_regs *regs, unsigned long error_code,
875 		unsigned long address, struct task_struct *tsk)
876 {
877 	if (!unhandled_signal(tsk, SIGSEGV))
878 		return;
879 
880 	if (!printk_ratelimit())
881 		return;
882 
883 	printk("%s%s[%d]: segfault at %lx ip %p sp %p error %lx",
884 		task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG,
885 		tsk->comm, task_pid_nr(tsk), address,
886 		(void *)regs->ip, (void *)regs->sp, error_code);
887 
888 	print_vma_addr(KERN_CONT " in ", regs->ip);
889 
890 	printk(KERN_CONT "\n");
891 }
892 
893 static void
894 __bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
895 		       unsigned long address, u32 *pkey, int si_code)
896 {
897 	struct task_struct *tsk = current;
898 
899 	/* User mode accesses just cause a SIGSEGV */
900 	if (error_code & PF_USER) {
901 		/*
902 		 * It's possible to have interrupts off here:
903 		 */
904 		local_irq_enable();
905 
906 		/*
907 		 * Valid to do another page fault here because this one came
908 		 * from user space:
909 		 */
910 		if (is_prefetch(regs, error_code, address))
911 			return;
912 
913 		if (is_errata100(regs, address))
914 			return;
915 
916 #ifdef CONFIG_X86_64
917 		/*
918 		 * Instruction fetch faults in the vsyscall page might need
919 		 * emulation.
920 		 */
921 		if (unlikely((error_code & PF_INSTR) &&
922 			     ((address & ~0xfff) == VSYSCALL_ADDR))) {
923 			if (emulate_vsyscall(regs, address))
924 				return;
925 		}
926 #endif
927 
928 		/*
929 		 * To avoid leaking information about the kernel page table
930 		 * layout, pretend that user-mode accesses to kernel addresses
931 		 * are always protection faults.
932 		 */
933 		if (address >= TASK_SIZE_MAX)
934 			error_code |= PF_PROT;
935 
936 		if (likely(show_unhandled_signals))
937 			show_signal_msg(regs, error_code, address, tsk);
938 
939 		tsk->thread.cr2		= address;
940 		tsk->thread.error_code	= error_code;
941 		tsk->thread.trap_nr	= X86_TRAP_PF;
942 
943 		force_sig_info_fault(SIGSEGV, si_code, address, tsk, pkey, 0);
944 
945 		return;
946 	}
947 
948 	if (is_f00f_bug(regs, address))
949 		return;
950 
951 	no_context(regs, error_code, address, SIGSEGV, si_code);
952 }
953 
954 static noinline void
955 bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
956 		     unsigned long address, u32 *pkey)
957 {
958 	__bad_area_nosemaphore(regs, error_code, address, pkey, SEGV_MAPERR);
959 }
960 
961 static void
962 __bad_area(struct pt_regs *regs, unsigned long error_code,
963 	   unsigned long address,  struct vm_area_struct *vma, int si_code)
964 {
965 	struct mm_struct *mm = current->mm;
966 	u32 pkey;
967 
968 	if (vma)
969 		pkey = vma_pkey(vma);
970 
971 	/*
972 	 * Something tried to access memory that isn't in our memory map..
973 	 * Fix it, but check if it's kernel or user first..
974 	 */
975 	up_read(&mm->mmap_sem);
976 
977 	__bad_area_nosemaphore(regs, error_code, address,
978 			       (vma) ? &pkey : NULL, si_code);
979 }
980 
981 static noinline void
982 bad_area(struct pt_regs *regs, unsigned long error_code, unsigned long address)
983 {
984 	__bad_area(regs, error_code, address, NULL, SEGV_MAPERR);
985 }
986 
987 static inline bool bad_area_access_from_pkeys(unsigned long error_code,
988 		struct vm_area_struct *vma)
989 {
990 	/* This code is always called on the current mm */
991 	bool foreign = false;
992 
993 	if (!boot_cpu_has(X86_FEATURE_OSPKE))
994 		return false;
995 	if (error_code & PF_PK)
996 		return true;
997 	/* this checks permission keys on the VMA: */
998 	if (!arch_vma_access_permitted(vma, (error_code & PF_WRITE),
999 				(error_code & PF_INSTR), foreign))
1000 		return true;
1001 	return false;
1002 }
1003 
1004 static noinline void
1005 bad_area_access_error(struct pt_regs *regs, unsigned long error_code,
1006 		      unsigned long address, struct vm_area_struct *vma)
1007 {
1008 	/*
1009 	 * This OSPKE check is not strictly necessary at runtime.
1010 	 * But, doing it this way allows compiler optimizations
1011 	 * if pkeys are compiled out.
1012 	 */
1013 	if (bad_area_access_from_pkeys(error_code, vma))
1014 		__bad_area(regs, error_code, address, vma, SEGV_PKUERR);
1015 	else
1016 		__bad_area(regs, error_code, address, vma, SEGV_ACCERR);
1017 }
1018 
1019 static void
1020 do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address,
1021 	  u32 *pkey, unsigned int fault)
1022 {
1023 	struct task_struct *tsk = current;
1024 	int code = BUS_ADRERR;
1025 
1026 	/* Kernel mode? Handle exceptions or die: */
1027 	if (!(error_code & PF_USER)) {
1028 		no_context(regs, error_code, address, SIGBUS, BUS_ADRERR);
1029 		return;
1030 	}
1031 
1032 	/* User-space => ok to do another page fault: */
1033 	if (is_prefetch(regs, error_code, address))
1034 		return;
1035 
1036 	tsk->thread.cr2		= address;
1037 	tsk->thread.error_code	= error_code;
1038 	tsk->thread.trap_nr	= X86_TRAP_PF;
1039 
1040 #ifdef CONFIG_MEMORY_FAILURE
1041 	if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) {
1042 		printk(KERN_ERR
1043 	"MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n",
1044 			tsk->comm, tsk->pid, address);
1045 		code = BUS_MCEERR_AR;
1046 	}
1047 #endif
1048 	force_sig_info_fault(SIGBUS, code, address, tsk, pkey, fault);
1049 }
1050 
1051 static noinline void
1052 mm_fault_error(struct pt_regs *regs, unsigned long error_code,
1053 	       unsigned long address, u32 *pkey, unsigned int fault)
1054 {
1055 	if (fatal_signal_pending(current) && !(error_code & PF_USER)) {
1056 		no_context(regs, error_code, address, 0, 0);
1057 		return;
1058 	}
1059 
1060 	if (fault & VM_FAULT_OOM) {
1061 		/* Kernel mode? Handle exceptions or die: */
1062 		if (!(error_code & PF_USER)) {
1063 			no_context(regs, error_code, address,
1064 				   SIGSEGV, SEGV_MAPERR);
1065 			return;
1066 		}
1067 
1068 		/*
1069 		 * We ran out of memory, call the OOM killer, and return the
1070 		 * userspace (which will retry the fault, or kill us if we got
1071 		 * oom-killed):
1072 		 */
1073 		pagefault_out_of_memory();
1074 	} else {
1075 		if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON|
1076 			     VM_FAULT_HWPOISON_LARGE))
1077 			do_sigbus(regs, error_code, address, pkey, fault);
1078 		else if (fault & VM_FAULT_SIGSEGV)
1079 			bad_area_nosemaphore(regs, error_code, address, pkey);
1080 		else
1081 			BUG();
1082 	}
1083 }
1084 
1085 static int spurious_fault_check(unsigned long error_code, pte_t *pte)
1086 {
1087 	if ((error_code & PF_WRITE) && !pte_write(*pte))
1088 		return 0;
1089 
1090 	if ((error_code & PF_INSTR) && !pte_exec(*pte))
1091 		return 0;
1092 	/*
1093 	 * Note: We do not do lazy flushing on protection key
1094 	 * changes, so no spurious fault will ever set PF_PK.
1095 	 */
1096 	if ((error_code & PF_PK))
1097 		return 1;
1098 
1099 	return 1;
1100 }
1101 
1102 /*
1103  * Handle a spurious fault caused by a stale TLB entry.
1104  *
1105  * This allows us to lazily refresh the TLB when increasing the
1106  * permissions of a kernel page (RO -> RW or NX -> X).  Doing it
1107  * eagerly is very expensive since that implies doing a full
1108  * cross-processor TLB flush, even if no stale TLB entries exist
1109  * on other processors.
1110  *
1111  * Spurious faults may only occur if the TLB contains an entry with
1112  * fewer permission than the page table entry.  Non-present (P = 0)
1113  * and reserved bit (R = 1) faults are never spurious.
1114  *
1115  * There are no security implications to leaving a stale TLB when
1116  * increasing the permissions on a page.
1117  *
1118  * Returns non-zero if a spurious fault was handled, zero otherwise.
1119  *
1120  * See Intel Developer's Manual Vol 3 Section 4.10.4.3, bullet 3
1121  * (Optional Invalidation).
1122  */
1123 static noinline int
1124 spurious_fault(unsigned long error_code, unsigned long address)
1125 {
1126 	pgd_t *pgd;
1127 	p4d_t *p4d;
1128 	pud_t *pud;
1129 	pmd_t *pmd;
1130 	pte_t *pte;
1131 	int ret;
1132 
1133 	/*
1134 	 * Only writes to RO or instruction fetches from NX may cause
1135 	 * spurious faults.
1136 	 *
1137 	 * These could be from user or supervisor accesses but the TLB
1138 	 * is only lazily flushed after a kernel mapping protection
1139 	 * change, so user accesses are not expected to cause spurious
1140 	 * faults.
1141 	 */
1142 	if (error_code != (PF_WRITE | PF_PROT)
1143 	    && error_code != (PF_INSTR | PF_PROT))
1144 		return 0;
1145 
1146 	pgd = init_mm.pgd + pgd_index(address);
1147 	if (!pgd_present(*pgd))
1148 		return 0;
1149 
1150 	p4d = p4d_offset(pgd, address);
1151 	if (!p4d_present(*p4d))
1152 		return 0;
1153 
1154 	if (p4d_large(*p4d))
1155 		return spurious_fault_check(error_code, (pte_t *) p4d);
1156 
1157 	pud = pud_offset(p4d, address);
1158 	if (!pud_present(*pud))
1159 		return 0;
1160 
1161 	if (pud_large(*pud))
1162 		return spurious_fault_check(error_code, (pte_t *) pud);
1163 
1164 	pmd = pmd_offset(pud, address);
1165 	if (!pmd_present(*pmd))
1166 		return 0;
1167 
1168 	if (pmd_large(*pmd))
1169 		return spurious_fault_check(error_code, (pte_t *) pmd);
1170 
1171 	pte = pte_offset_kernel(pmd, address);
1172 	if (!pte_present(*pte))
1173 		return 0;
1174 
1175 	ret = spurious_fault_check(error_code, pte);
1176 	if (!ret)
1177 		return 0;
1178 
1179 	/*
1180 	 * Make sure we have permissions in PMD.
1181 	 * If not, then there's a bug in the page tables:
1182 	 */
1183 	ret = spurious_fault_check(error_code, (pte_t *) pmd);
1184 	WARN_ONCE(!ret, "PMD has incorrect permission bits\n");
1185 
1186 	return ret;
1187 }
1188 NOKPROBE_SYMBOL(spurious_fault);
1189 
1190 int show_unhandled_signals = 1;
1191 
1192 static inline int
1193 access_error(unsigned long error_code, struct vm_area_struct *vma)
1194 {
1195 	/* This is only called for the current mm, so: */
1196 	bool foreign = false;
1197 
1198 	/*
1199 	 * Read or write was blocked by protection keys.  This is
1200 	 * always an unconditional error and can never result in
1201 	 * a follow-up action to resolve the fault, like a COW.
1202 	 */
1203 	if (error_code & PF_PK)
1204 		return 1;
1205 
1206 	/*
1207 	 * Make sure to check the VMA so that we do not perform
1208 	 * faults just to hit a PF_PK as soon as we fill in a
1209 	 * page.
1210 	 */
1211 	if (!arch_vma_access_permitted(vma, (error_code & PF_WRITE),
1212 				(error_code & PF_INSTR), foreign))
1213 		return 1;
1214 
1215 	if (error_code & PF_WRITE) {
1216 		/* write, present and write, not present: */
1217 		if (unlikely(!(vma->vm_flags & VM_WRITE)))
1218 			return 1;
1219 		return 0;
1220 	}
1221 
1222 	/* read, present: */
1223 	if (unlikely(error_code & PF_PROT))
1224 		return 1;
1225 
1226 	/* read, not present: */
1227 	if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE))))
1228 		return 1;
1229 
1230 	return 0;
1231 }
1232 
1233 static int fault_in_kernel_space(unsigned long address)
1234 {
1235 	return address >= TASK_SIZE_MAX;
1236 }
1237 
1238 static inline bool smap_violation(int error_code, struct pt_regs *regs)
1239 {
1240 	if (!IS_ENABLED(CONFIG_X86_SMAP))
1241 		return false;
1242 
1243 	if (!static_cpu_has(X86_FEATURE_SMAP))
1244 		return false;
1245 
1246 	if (error_code & PF_USER)
1247 		return false;
1248 
1249 	if (!user_mode(regs) && (regs->flags & X86_EFLAGS_AC))
1250 		return false;
1251 
1252 	return true;
1253 }
1254 
1255 /*
1256  * This routine handles page faults.  It determines the address,
1257  * and the problem, and then passes it off to one of the appropriate
1258  * routines.
1259  */
1260 static noinline void
1261 __do_page_fault(struct pt_regs *regs, unsigned long error_code,
1262 		unsigned long address)
1263 {
1264 	struct vm_area_struct *vma;
1265 	struct task_struct *tsk;
1266 	struct mm_struct *mm;
1267 	int fault, major = 0;
1268 	unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
1269 	u32 pkey;
1270 
1271 	tsk = current;
1272 	mm = tsk->mm;
1273 
1274 	/*
1275 	 * Detect and handle instructions that would cause a page fault for
1276 	 * both a tracked kernel page and a userspace page.
1277 	 */
1278 	if (kmemcheck_active(regs))
1279 		kmemcheck_hide(regs);
1280 	prefetchw(&mm->mmap_sem);
1281 
1282 	if (unlikely(kmmio_fault(regs, address)))
1283 		return;
1284 
1285 	/*
1286 	 * We fault-in kernel-space virtual memory on-demand. The
1287 	 * 'reference' page table is init_mm.pgd.
1288 	 *
1289 	 * NOTE! We MUST NOT take any locks for this case. We may
1290 	 * be in an interrupt or a critical region, and should
1291 	 * only copy the information from the master page table,
1292 	 * nothing more.
1293 	 *
1294 	 * This verifies that the fault happens in kernel space
1295 	 * (error_code & 4) == 0, and that the fault was not a
1296 	 * protection error (error_code & 9) == 0.
1297 	 */
1298 	if (unlikely(fault_in_kernel_space(address))) {
1299 		if (!(error_code & (PF_RSVD | PF_USER | PF_PROT))) {
1300 			if (vmalloc_fault(address) >= 0)
1301 				return;
1302 
1303 			if (kmemcheck_fault(regs, address, error_code))
1304 				return;
1305 		}
1306 
1307 		/* Can handle a stale RO->RW TLB: */
1308 		if (spurious_fault(error_code, address))
1309 			return;
1310 
1311 		/* kprobes don't want to hook the spurious faults: */
1312 		if (kprobes_fault(regs))
1313 			return;
1314 		/*
1315 		 * Don't take the mm semaphore here. If we fixup a prefetch
1316 		 * fault we could otherwise deadlock:
1317 		 */
1318 		bad_area_nosemaphore(regs, error_code, address, NULL);
1319 
1320 		return;
1321 	}
1322 
1323 	/* kprobes don't want to hook the spurious faults: */
1324 	if (unlikely(kprobes_fault(regs)))
1325 		return;
1326 
1327 	if (unlikely(error_code & PF_RSVD))
1328 		pgtable_bad(regs, error_code, address);
1329 
1330 	if (unlikely(smap_violation(error_code, regs))) {
1331 		bad_area_nosemaphore(regs, error_code, address, NULL);
1332 		return;
1333 	}
1334 
1335 	/*
1336 	 * If we're in an interrupt, have no user context or are running
1337 	 * in a region with pagefaults disabled then we must not take the fault
1338 	 */
1339 	if (unlikely(faulthandler_disabled() || !mm)) {
1340 		bad_area_nosemaphore(regs, error_code, address, NULL);
1341 		return;
1342 	}
1343 
1344 	/*
1345 	 * It's safe to allow irq's after cr2 has been saved and the
1346 	 * vmalloc fault has been handled.
1347 	 *
1348 	 * User-mode registers count as a user access even for any
1349 	 * potential system fault or CPU buglet:
1350 	 */
1351 	if (user_mode(regs)) {
1352 		local_irq_enable();
1353 		error_code |= PF_USER;
1354 		flags |= FAULT_FLAG_USER;
1355 	} else {
1356 		if (regs->flags & X86_EFLAGS_IF)
1357 			local_irq_enable();
1358 	}
1359 
1360 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
1361 
1362 	if (error_code & PF_WRITE)
1363 		flags |= FAULT_FLAG_WRITE;
1364 	if (error_code & PF_INSTR)
1365 		flags |= FAULT_FLAG_INSTRUCTION;
1366 
1367 	/*
1368 	 * When running in the kernel we expect faults to occur only to
1369 	 * addresses in user space.  All other faults represent errors in
1370 	 * the kernel and should generate an OOPS.  Unfortunately, in the
1371 	 * case of an erroneous fault occurring in a code path which already
1372 	 * holds mmap_sem we will deadlock attempting to validate the fault
1373 	 * against the address space.  Luckily the kernel only validly
1374 	 * references user space from well defined areas of code, which are
1375 	 * listed in the exceptions table.
1376 	 *
1377 	 * As the vast majority of faults will be valid we will only perform
1378 	 * the source reference check when there is a possibility of a
1379 	 * deadlock. Attempt to lock the address space, if we cannot we then
1380 	 * validate the source. If this is invalid we can skip the address
1381 	 * space check, thus avoiding the deadlock:
1382 	 */
1383 	if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
1384 		if ((error_code & PF_USER) == 0 &&
1385 		    !search_exception_tables(regs->ip)) {
1386 			bad_area_nosemaphore(regs, error_code, address, NULL);
1387 			return;
1388 		}
1389 retry:
1390 		down_read(&mm->mmap_sem);
1391 	} else {
1392 		/*
1393 		 * The above down_read_trylock() might have succeeded in
1394 		 * which case we'll have missed the might_sleep() from
1395 		 * down_read():
1396 		 */
1397 		might_sleep();
1398 	}
1399 
1400 	vma = find_vma(mm, address);
1401 	if (unlikely(!vma)) {
1402 		bad_area(regs, error_code, address);
1403 		return;
1404 	}
1405 	if (likely(vma->vm_start <= address))
1406 		goto good_area;
1407 	if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) {
1408 		bad_area(regs, error_code, address);
1409 		return;
1410 	}
1411 	if (error_code & PF_USER) {
1412 		/*
1413 		 * Accessing the stack below %sp is always a bug.
1414 		 * The large cushion allows instructions like enter
1415 		 * and pusha to work. ("enter $65535, $31" pushes
1416 		 * 32 pointers and then decrements %sp by 65535.)
1417 		 */
1418 		if (unlikely(address + 65536 + 32 * sizeof(unsigned long) < regs->sp)) {
1419 			bad_area(regs, error_code, address);
1420 			return;
1421 		}
1422 	}
1423 	if (unlikely(expand_stack(vma, address))) {
1424 		bad_area(regs, error_code, address);
1425 		return;
1426 	}
1427 
1428 	/*
1429 	 * Ok, we have a good vm_area for this memory access, so
1430 	 * we can handle it..
1431 	 */
1432 good_area:
1433 	if (unlikely(access_error(error_code, vma))) {
1434 		bad_area_access_error(regs, error_code, address, vma);
1435 		return;
1436 	}
1437 
1438 	/*
1439 	 * If for any reason at all we couldn't handle the fault,
1440 	 * make sure we exit gracefully rather than endlessly redo
1441 	 * the fault.  Since we never set FAULT_FLAG_RETRY_NOWAIT, if
1442 	 * we get VM_FAULT_RETRY back, the mmap_sem has been unlocked.
1443 	 */
1444 	fault = handle_mm_fault(vma, address, flags);
1445 	major |= fault & VM_FAULT_MAJOR;
1446 
1447 	/*
1448 	 * If we need to retry the mmap_sem has already been released,
1449 	 * and if there is a fatal signal pending there is no guarantee
1450 	 * that we made any progress. Handle this case first.
1451 	 */
1452 	if (unlikely(fault & VM_FAULT_RETRY)) {
1453 		/* Retry at most once */
1454 		if (flags & FAULT_FLAG_ALLOW_RETRY) {
1455 			flags &= ~FAULT_FLAG_ALLOW_RETRY;
1456 			flags |= FAULT_FLAG_TRIED;
1457 			if (!fatal_signal_pending(tsk))
1458 				goto retry;
1459 		}
1460 
1461 		/* User mode? Just return to handle the fatal exception */
1462 		if (flags & FAULT_FLAG_USER)
1463 			return;
1464 
1465 		/* Not returning to user mode? Handle exceptions or die: */
1466 		no_context(regs, error_code, address, SIGBUS, BUS_ADRERR);
1467 		return;
1468 	}
1469 
1470 	pkey = vma_pkey(vma);
1471 	up_read(&mm->mmap_sem);
1472 	if (unlikely(fault & VM_FAULT_ERROR)) {
1473 		mm_fault_error(regs, error_code, address, &pkey, fault);
1474 		return;
1475 	}
1476 
1477 	/*
1478 	 * Major/minor page fault accounting. If any of the events
1479 	 * returned VM_FAULT_MAJOR, we account it as a major fault.
1480 	 */
1481 	if (major) {
1482 		tsk->maj_flt++;
1483 		perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, address);
1484 	} else {
1485 		tsk->min_flt++;
1486 		perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, address);
1487 	}
1488 
1489 	check_v8086_mode(regs, address, tsk);
1490 }
1491 NOKPROBE_SYMBOL(__do_page_fault);
1492 
1493 static nokprobe_inline void
1494 trace_page_fault_entries(unsigned long address, struct pt_regs *regs,
1495 			 unsigned long error_code)
1496 {
1497 	if (user_mode(regs))
1498 		trace_page_fault_user(address, regs, error_code);
1499 	else
1500 		trace_page_fault_kernel(address, regs, error_code);
1501 }
1502 
1503 /*
1504  * We must have this function blacklisted from kprobes, tagged with notrace
1505  * and call read_cr2() before calling anything else. To avoid calling any
1506  * kind of tracing machinery before we've observed the CR2 value.
1507  *
1508  * exception_{enter,exit}() contains all sorts of tracepoints.
1509  */
1510 dotraplinkage void notrace
1511 do_page_fault(struct pt_regs *regs, unsigned long error_code)
1512 {
1513 	unsigned long address = read_cr2(); /* Get the faulting address */
1514 	enum ctx_state prev_state;
1515 
1516 	prev_state = exception_enter();
1517 	if (trace_pagefault_enabled())
1518 		trace_page_fault_entries(address, regs, error_code);
1519 
1520 	__do_page_fault(regs, error_code, address);
1521 	exception_exit(prev_state);
1522 }
1523 NOKPROBE_SYMBOL(do_page_fault);
1524