xref: /openbmc/linux/arch/x86/mm/fault.c (revision 80ade22c)
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/fixmap.h>			/* VSYSCALL_ADDR		*/
25 #include <asm/vsyscall.h>		/* emulate_vsyscall		*/
26 #include <asm/vm86.h>			/* struct vm86			*/
27 #include <asm/mmu_context.h>		/* vma_pkey()			*/
28 #include <asm/efi.h>			/* efi_recover_from_page_fault()*/
29 #include <asm/desc.h>			/* store_idt(), ...		*/
30 #include <asm/cpu_entry_area.h>		/* exception stack		*/
31 #include <asm/pgtable_areas.h>		/* VMALLOC_START, ...		*/
32 #include <asm/kvm_para.h>		/* kvm_handle_async_pf		*/
33 
34 #define CREATE_TRACE_POINTS
35 #include <asm/trace/exceptions.h>
36 
37 /*
38  * Returns 0 if mmiotrace is disabled, or if the fault is not
39  * handled by mmiotrace:
40  */
41 static nokprobe_inline int
42 kmmio_fault(struct pt_regs *regs, unsigned long addr)
43 {
44 	if (unlikely(is_kmmio_active()))
45 		if (kmmio_handler(regs, addr) == 1)
46 			return -1;
47 	return 0;
48 }
49 
50 /*
51  * Prefetch quirks:
52  *
53  * 32-bit mode:
54  *
55  *   Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch.
56  *   Check that here and ignore it.
57  *
58  * 64-bit mode:
59  *
60  *   Sometimes the CPU reports invalid exceptions on prefetch.
61  *   Check that here and ignore it.
62  *
63  * Opcode checker based on code by Richard Brunner.
64  */
65 static inline int
66 check_prefetch_opcode(struct pt_regs *regs, unsigned char *instr,
67 		      unsigned char opcode, int *prefetch)
68 {
69 	unsigned char instr_hi = opcode & 0xf0;
70 	unsigned char instr_lo = opcode & 0x0f;
71 
72 	switch (instr_hi) {
73 	case 0x20:
74 	case 0x30:
75 		/*
76 		 * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes.
77 		 * In X86_64 long mode, the CPU will signal invalid
78 		 * opcode if some of these prefixes are present so
79 		 * X86_64 will never get here anyway
80 		 */
81 		return ((instr_lo & 7) == 0x6);
82 #ifdef CONFIG_X86_64
83 	case 0x40:
84 		/*
85 		 * In AMD64 long mode 0x40..0x4F are valid REX prefixes
86 		 * Need to figure out under what instruction mode the
87 		 * instruction was issued. Could check the LDT for lm,
88 		 * but for now it's good enough to assume that long
89 		 * mode only uses well known segments or kernel.
90 		 */
91 		return (!user_mode(regs) || user_64bit_mode(regs));
92 #endif
93 	case 0x60:
94 		/* 0x64 thru 0x67 are valid prefixes in all modes. */
95 		return (instr_lo & 0xC) == 0x4;
96 	case 0xF0:
97 		/* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */
98 		return !instr_lo || (instr_lo>>1) == 1;
99 	case 0x00:
100 		/* Prefetch instruction is 0x0F0D or 0x0F18 */
101 		if (get_kernel_nofault(opcode, instr))
102 			return 0;
103 
104 		*prefetch = (instr_lo == 0xF) &&
105 			(opcode == 0x0D || opcode == 0x18);
106 		return 0;
107 	default:
108 		return 0;
109 	}
110 }
111 
112 static int
113 is_prefetch(struct pt_regs *regs, unsigned long error_code, unsigned long addr)
114 {
115 	unsigned char *max_instr;
116 	unsigned char *instr;
117 	int prefetch = 0;
118 
119 	/*
120 	 * If it was a exec (instruction fetch) fault on NX page, then
121 	 * do not ignore the fault:
122 	 */
123 	if (error_code & X86_PF_INSTR)
124 		return 0;
125 
126 	instr = (void *)convert_ip_to_linear(current, regs);
127 	max_instr = instr + 15;
128 
129 	if (user_mode(regs) && instr >= (unsigned char *)TASK_SIZE_MAX)
130 		return 0;
131 
132 	while (instr < max_instr) {
133 		unsigned char opcode;
134 
135 		if (get_kernel_nofault(opcode, instr))
136 			break;
137 
138 		instr++;
139 
140 		if (!check_prefetch_opcode(regs, instr, opcode, &prefetch))
141 			break;
142 	}
143 	return prefetch;
144 }
145 
146 DEFINE_SPINLOCK(pgd_lock);
147 LIST_HEAD(pgd_list);
148 
149 #ifdef CONFIG_X86_32
150 static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address)
151 {
152 	unsigned index = pgd_index(address);
153 	pgd_t *pgd_k;
154 	p4d_t *p4d, *p4d_k;
155 	pud_t *pud, *pud_k;
156 	pmd_t *pmd, *pmd_k;
157 
158 	pgd += index;
159 	pgd_k = init_mm.pgd + index;
160 
161 	if (!pgd_present(*pgd_k))
162 		return NULL;
163 
164 	/*
165 	 * set_pgd(pgd, *pgd_k); here would be useless on PAE
166 	 * and redundant with the set_pmd() on non-PAE. As would
167 	 * set_p4d/set_pud.
168 	 */
169 	p4d = p4d_offset(pgd, address);
170 	p4d_k = p4d_offset(pgd_k, address);
171 	if (!p4d_present(*p4d_k))
172 		return NULL;
173 
174 	pud = pud_offset(p4d, address);
175 	pud_k = pud_offset(p4d_k, address);
176 	if (!pud_present(*pud_k))
177 		return NULL;
178 
179 	pmd = pmd_offset(pud, address);
180 	pmd_k = pmd_offset(pud_k, address);
181 
182 	if (pmd_present(*pmd) != pmd_present(*pmd_k))
183 		set_pmd(pmd, *pmd_k);
184 
185 	if (!pmd_present(*pmd_k))
186 		return NULL;
187 	else
188 		BUG_ON(pmd_pfn(*pmd) != pmd_pfn(*pmd_k));
189 
190 	return pmd_k;
191 }
192 
193 /*
194  *   Handle a fault on the vmalloc or module mapping area
195  *
196  *   This is needed because there is a race condition between the time
197  *   when the vmalloc mapping code updates the PMD to the point in time
198  *   where it synchronizes this update with the other page-tables in the
199  *   system.
200  *
201  *   In this race window another thread/CPU can map an area on the same
202  *   PMD, finds it already present and does not synchronize it with the
203  *   rest of the system yet. As a result v[mz]alloc might return areas
204  *   which are not mapped in every page-table in the system, causing an
205  *   unhandled page-fault when they are accessed.
206  */
207 static noinline int vmalloc_fault(unsigned long address)
208 {
209 	unsigned long pgd_paddr;
210 	pmd_t *pmd_k;
211 	pte_t *pte_k;
212 
213 	/* Make sure we are in vmalloc area: */
214 	if (!(address >= VMALLOC_START && address < VMALLOC_END))
215 		return -1;
216 
217 	/*
218 	 * Synchronize this task's top level page-table
219 	 * with the 'reference' page table.
220 	 *
221 	 * Do _not_ use "current" here. We might be inside
222 	 * an interrupt in the middle of a task switch..
223 	 */
224 	pgd_paddr = read_cr3_pa();
225 	pmd_k = vmalloc_sync_one(__va(pgd_paddr), address);
226 	if (!pmd_k)
227 		return -1;
228 
229 	if (pmd_large(*pmd_k))
230 		return 0;
231 
232 	pte_k = pte_offset_kernel(pmd_k, address);
233 	if (!pte_present(*pte_k))
234 		return -1;
235 
236 	return 0;
237 }
238 NOKPROBE_SYMBOL(vmalloc_fault);
239 
240 void arch_sync_kernel_mappings(unsigned long start, unsigned long end)
241 {
242 	unsigned long addr;
243 
244 	for (addr = start & PMD_MASK;
245 	     addr >= TASK_SIZE_MAX && addr < VMALLOC_END;
246 	     addr += PMD_SIZE) {
247 		struct page *page;
248 
249 		spin_lock(&pgd_lock);
250 		list_for_each_entry(page, &pgd_list, lru) {
251 			spinlock_t *pgt_lock;
252 
253 			/* the pgt_lock only for Xen */
254 			pgt_lock = &pgd_page_get_mm(page)->page_table_lock;
255 
256 			spin_lock(pgt_lock);
257 			vmalloc_sync_one(page_address(page), addr);
258 			spin_unlock(pgt_lock);
259 		}
260 		spin_unlock(&pgd_lock);
261 	}
262 }
263 
264 /*
265  * Did it hit the DOS screen memory VA from vm86 mode?
266  */
267 static inline void
268 check_v8086_mode(struct pt_regs *regs, unsigned long address,
269 		 struct task_struct *tsk)
270 {
271 #ifdef CONFIG_VM86
272 	unsigned long bit;
273 
274 	if (!v8086_mode(regs) || !tsk->thread.vm86)
275 		return;
276 
277 	bit = (address - 0xA0000) >> PAGE_SHIFT;
278 	if (bit < 32)
279 		tsk->thread.vm86->screen_bitmap |= 1 << bit;
280 #endif
281 }
282 
283 static bool low_pfn(unsigned long pfn)
284 {
285 	return pfn < max_low_pfn;
286 }
287 
288 static void dump_pagetable(unsigned long address)
289 {
290 	pgd_t *base = __va(read_cr3_pa());
291 	pgd_t *pgd = &base[pgd_index(address)];
292 	p4d_t *p4d;
293 	pud_t *pud;
294 	pmd_t *pmd;
295 	pte_t *pte;
296 
297 #ifdef CONFIG_X86_PAE
298 	pr_info("*pdpt = %016Lx ", pgd_val(*pgd));
299 	if (!low_pfn(pgd_val(*pgd) >> PAGE_SHIFT) || !pgd_present(*pgd))
300 		goto out;
301 #define pr_pde pr_cont
302 #else
303 #define pr_pde pr_info
304 #endif
305 	p4d = p4d_offset(pgd, address);
306 	pud = pud_offset(p4d, address);
307 	pmd = pmd_offset(pud, address);
308 	pr_pde("*pde = %0*Lx ", sizeof(*pmd) * 2, (u64)pmd_val(*pmd));
309 #undef pr_pde
310 
311 	/*
312 	 * We must not directly access the pte in the highpte
313 	 * case if the page table is located in highmem.
314 	 * And let's rather not kmap-atomic the pte, just in case
315 	 * it's allocated already:
316 	 */
317 	if (!low_pfn(pmd_pfn(*pmd)) || !pmd_present(*pmd) || pmd_large(*pmd))
318 		goto out;
319 
320 	pte = pte_offset_kernel(pmd, address);
321 	pr_cont("*pte = %0*Lx ", sizeof(*pte) * 2, (u64)pte_val(*pte));
322 out:
323 	pr_cont("\n");
324 }
325 
326 #else /* CONFIG_X86_64: */
327 
328 #ifdef CONFIG_CPU_SUP_AMD
329 static const char errata93_warning[] =
330 KERN_ERR
331 "******* Your BIOS seems to not contain a fix for K8 errata #93\n"
332 "******* Working around it, but it may cause SEGVs or burn power.\n"
333 "******* Please consider a BIOS update.\n"
334 "******* Disabling USB legacy in the BIOS may also help.\n";
335 #endif
336 
337 /*
338  * No vm86 mode in 64-bit mode:
339  */
340 static inline void
341 check_v8086_mode(struct pt_regs *regs, unsigned long address,
342 		 struct task_struct *tsk)
343 {
344 }
345 
346 static int bad_address(void *p)
347 {
348 	unsigned long dummy;
349 
350 	return get_kernel_nofault(dummy, (unsigned long *)p);
351 }
352 
353 static void dump_pagetable(unsigned long address)
354 {
355 	pgd_t *base = __va(read_cr3_pa());
356 	pgd_t *pgd = base + pgd_index(address);
357 	p4d_t *p4d;
358 	pud_t *pud;
359 	pmd_t *pmd;
360 	pte_t *pte;
361 
362 	if (bad_address(pgd))
363 		goto bad;
364 
365 	pr_info("PGD %lx ", pgd_val(*pgd));
366 
367 	if (!pgd_present(*pgd))
368 		goto out;
369 
370 	p4d = p4d_offset(pgd, address);
371 	if (bad_address(p4d))
372 		goto bad;
373 
374 	pr_cont("P4D %lx ", p4d_val(*p4d));
375 	if (!p4d_present(*p4d) || p4d_large(*p4d))
376 		goto out;
377 
378 	pud = pud_offset(p4d, address);
379 	if (bad_address(pud))
380 		goto bad;
381 
382 	pr_cont("PUD %lx ", pud_val(*pud));
383 	if (!pud_present(*pud) || pud_large(*pud))
384 		goto out;
385 
386 	pmd = pmd_offset(pud, address);
387 	if (bad_address(pmd))
388 		goto bad;
389 
390 	pr_cont("PMD %lx ", pmd_val(*pmd));
391 	if (!pmd_present(*pmd) || pmd_large(*pmd))
392 		goto out;
393 
394 	pte = pte_offset_kernel(pmd, address);
395 	if (bad_address(pte))
396 		goto bad;
397 
398 	pr_cont("PTE %lx", pte_val(*pte));
399 out:
400 	pr_cont("\n");
401 	return;
402 bad:
403 	pr_info("BAD\n");
404 }
405 
406 #endif /* CONFIG_X86_64 */
407 
408 /*
409  * Workaround for K8 erratum #93 & buggy BIOS.
410  *
411  * BIOS SMM functions are required to use a specific workaround
412  * to avoid corruption of the 64bit RIP register on C stepping K8.
413  *
414  * A lot of BIOS that didn't get tested properly miss this.
415  *
416  * The OS sees this as a page fault with the upper 32bits of RIP cleared.
417  * Try to work around it here.
418  *
419  * Note we only handle faults in kernel here.
420  * Does nothing on 32-bit.
421  */
422 static int is_errata93(struct pt_regs *regs, unsigned long address)
423 {
424 #if defined(CONFIG_X86_64) && defined(CONFIG_CPU_SUP_AMD)
425 	if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD
426 	    || boot_cpu_data.x86 != 0xf)
427 		return 0;
428 
429 	if (address != regs->ip)
430 		return 0;
431 
432 	if ((address >> 32) != 0)
433 		return 0;
434 
435 	address |= 0xffffffffUL << 32;
436 	if ((address >= (u64)_stext && address <= (u64)_etext) ||
437 	    (address >= MODULES_VADDR && address <= MODULES_END)) {
438 		printk_once(errata93_warning);
439 		regs->ip = address;
440 		return 1;
441 	}
442 #endif
443 	return 0;
444 }
445 
446 /*
447  * Work around K8 erratum #100 K8 in compat mode occasionally jumps
448  * to illegal addresses >4GB.
449  *
450  * We catch this in the page fault handler because these addresses
451  * are not reachable. Just detect this case and return.  Any code
452  * segment in LDT is compatibility mode.
453  */
454 static int is_errata100(struct pt_regs *regs, unsigned long address)
455 {
456 #ifdef CONFIG_X86_64
457 	if ((regs->cs == __USER32_CS || (regs->cs & (1<<2))) && (address >> 32))
458 		return 1;
459 #endif
460 	return 0;
461 }
462 
463 /* Pentium F0 0F C7 C8 bug workaround: */
464 static int is_f00f_bug(struct pt_regs *regs, unsigned long address)
465 {
466 #ifdef CONFIG_X86_F00F_BUG
467 	if (boot_cpu_has_bug(X86_BUG_F00F) && idt_is_f00f_address(address)) {
468 		handle_invalid_op(regs);
469 		return 1;
470 	}
471 #endif
472 	return 0;
473 }
474 
475 static void show_ldttss(const struct desc_ptr *gdt, const char *name, u16 index)
476 {
477 	u32 offset = (index >> 3) * sizeof(struct desc_struct);
478 	unsigned long addr;
479 	struct ldttss_desc desc;
480 
481 	if (index == 0) {
482 		pr_alert("%s: NULL\n", name);
483 		return;
484 	}
485 
486 	if (offset + sizeof(struct ldttss_desc) >= gdt->size) {
487 		pr_alert("%s: 0x%hx -- out of bounds\n", name, index);
488 		return;
489 	}
490 
491 	if (copy_from_kernel_nofault(&desc, (void *)(gdt->address + offset),
492 			      sizeof(struct ldttss_desc))) {
493 		pr_alert("%s: 0x%hx -- GDT entry is not readable\n",
494 			 name, index);
495 		return;
496 	}
497 
498 	addr = desc.base0 | (desc.base1 << 16) | ((unsigned long)desc.base2 << 24);
499 #ifdef CONFIG_X86_64
500 	addr |= ((u64)desc.base3 << 32);
501 #endif
502 	pr_alert("%s: 0x%hx -- base=0x%lx limit=0x%x\n",
503 		 name, index, addr, (desc.limit0 | (desc.limit1 << 16)));
504 }
505 
506 static void
507 show_fault_oops(struct pt_regs *regs, unsigned long error_code, unsigned long address)
508 {
509 	if (!oops_may_print())
510 		return;
511 
512 	if (error_code & X86_PF_INSTR) {
513 		unsigned int level;
514 		pgd_t *pgd;
515 		pte_t *pte;
516 
517 		pgd = __va(read_cr3_pa());
518 		pgd += pgd_index(address);
519 
520 		pte = lookup_address_in_pgd(pgd, address, &level);
521 
522 		if (pte && pte_present(*pte) && !pte_exec(*pte))
523 			pr_crit("kernel tried to execute NX-protected page - exploit attempt? (uid: %d)\n",
524 				from_kuid(&init_user_ns, current_uid()));
525 		if (pte && pte_present(*pte) && pte_exec(*pte) &&
526 				(pgd_flags(*pgd) & _PAGE_USER) &&
527 				(__read_cr4() & X86_CR4_SMEP))
528 			pr_crit("unable to execute userspace code (SMEP?) (uid: %d)\n",
529 				from_kuid(&init_user_ns, current_uid()));
530 	}
531 
532 	if (address < PAGE_SIZE && !user_mode(regs))
533 		pr_alert("BUG: kernel NULL pointer dereference, address: %px\n",
534 			(void *)address);
535 	else
536 		pr_alert("BUG: unable to handle page fault for address: %px\n",
537 			(void *)address);
538 
539 	pr_alert("#PF: %s %s in %s mode\n",
540 		 (error_code & X86_PF_USER)  ? "user" : "supervisor",
541 		 (error_code & X86_PF_INSTR) ? "instruction fetch" :
542 		 (error_code & X86_PF_WRITE) ? "write access" :
543 					       "read access",
544 			     user_mode(regs) ? "user" : "kernel");
545 	pr_alert("#PF: error_code(0x%04lx) - %s\n", error_code,
546 		 !(error_code & X86_PF_PROT) ? "not-present page" :
547 		 (error_code & X86_PF_RSVD)  ? "reserved bit violation" :
548 		 (error_code & X86_PF_PK)    ? "protection keys violation" :
549 					       "permissions violation");
550 
551 	if (!(error_code & X86_PF_USER) && user_mode(regs)) {
552 		struct desc_ptr idt, gdt;
553 		u16 ldtr, tr;
554 
555 		/*
556 		 * This can happen for quite a few reasons.  The more obvious
557 		 * ones are faults accessing the GDT, or LDT.  Perhaps
558 		 * surprisingly, if the CPU tries to deliver a benign or
559 		 * contributory exception from user code and gets a page fault
560 		 * during delivery, the page fault can be delivered as though
561 		 * it originated directly from user code.  This could happen
562 		 * due to wrong permissions on the IDT, GDT, LDT, TSS, or
563 		 * kernel or IST stack.
564 		 */
565 		store_idt(&idt);
566 
567 		/* Usable even on Xen PV -- it's just slow. */
568 		native_store_gdt(&gdt);
569 
570 		pr_alert("IDT: 0x%lx (limit=0x%hx) GDT: 0x%lx (limit=0x%hx)\n",
571 			 idt.address, idt.size, gdt.address, gdt.size);
572 
573 		store_ldt(ldtr);
574 		show_ldttss(&gdt, "LDTR", ldtr);
575 
576 		store_tr(tr);
577 		show_ldttss(&gdt, "TR", tr);
578 	}
579 
580 	dump_pagetable(address);
581 }
582 
583 static noinline void
584 pgtable_bad(struct pt_regs *regs, unsigned long error_code,
585 	    unsigned long address)
586 {
587 	struct task_struct *tsk;
588 	unsigned long flags;
589 	int sig;
590 
591 	flags = oops_begin();
592 	tsk = current;
593 	sig = SIGKILL;
594 
595 	printk(KERN_ALERT "%s: Corrupted page table at address %lx\n",
596 	       tsk->comm, address);
597 	dump_pagetable(address);
598 
599 	if (__die("Bad pagetable", regs, error_code))
600 		sig = 0;
601 
602 	oops_end(flags, regs, sig);
603 }
604 
605 static void set_signal_archinfo(unsigned long address,
606 				unsigned long error_code)
607 {
608 	struct task_struct *tsk = current;
609 
610 	/*
611 	 * To avoid leaking information about the kernel page
612 	 * table layout, pretend that user-mode accesses to
613 	 * kernel addresses are always protection faults.
614 	 *
615 	 * NB: This means that failed vsyscalls with vsyscall=none
616 	 * will have the PROT bit.  This doesn't leak any
617 	 * information and does not appear to cause any problems.
618 	 */
619 	if (address >= TASK_SIZE_MAX)
620 		error_code |= X86_PF_PROT;
621 
622 	tsk->thread.trap_nr = X86_TRAP_PF;
623 	tsk->thread.error_code = error_code | X86_PF_USER;
624 	tsk->thread.cr2 = address;
625 }
626 
627 static noinline void
628 no_context(struct pt_regs *regs, unsigned long error_code,
629 	   unsigned long address, int signal, int si_code)
630 {
631 	struct task_struct *tsk = current;
632 	unsigned long flags;
633 	int sig;
634 
635 	if (user_mode(regs)) {
636 		/*
637 		 * This is an implicit supervisor-mode access from user
638 		 * mode.  Bypass all the kernel-mode recovery code and just
639 		 * OOPS.
640 		 */
641 		goto oops;
642 	}
643 
644 	/* Are we prepared to handle this kernel fault? */
645 	if (fixup_exception(regs, X86_TRAP_PF, error_code, address)) {
646 		/*
647 		 * Any interrupt that takes a fault gets the fixup. This makes
648 		 * the below recursive fault logic only apply to a faults from
649 		 * task context.
650 		 */
651 		if (in_interrupt())
652 			return;
653 
654 		/*
655 		 * Per the above we're !in_interrupt(), aka. task context.
656 		 *
657 		 * In this case we need to make sure we're not recursively
658 		 * faulting through the emulate_vsyscall() logic.
659 		 */
660 		if (current->thread.sig_on_uaccess_err && signal) {
661 			set_signal_archinfo(address, error_code);
662 
663 			/* XXX: hwpoison faults will set the wrong code. */
664 			force_sig_fault(signal, si_code, (void __user *)address);
665 		}
666 
667 		/*
668 		 * Barring that, we can do the fixup and be happy.
669 		 */
670 		return;
671 	}
672 
673 #ifdef CONFIG_VMAP_STACK
674 	/*
675 	 * Stack overflow?  During boot, we can fault near the initial
676 	 * stack in the direct map, but that's not an overflow -- check
677 	 * that we're in vmalloc space to avoid this.
678 	 */
679 	if (is_vmalloc_addr((void *)address) &&
680 	    (((unsigned long)tsk->stack - 1 - address < PAGE_SIZE) ||
681 	     address - ((unsigned long)tsk->stack + THREAD_SIZE) < PAGE_SIZE)) {
682 		unsigned long stack = __this_cpu_ist_top_va(DF) - sizeof(void *);
683 		/*
684 		 * We're likely to be running with very little stack space
685 		 * left.  It's plausible that we'd hit this condition but
686 		 * double-fault even before we get this far, in which case
687 		 * we're fine: the double-fault handler will deal with it.
688 		 *
689 		 * We don't want to make it all the way into the oops code
690 		 * and then double-fault, though, because we're likely to
691 		 * break the console driver and lose most of the stack dump.
692 		 */
693 		asm volatile ("movq %[stack], %%rsp\n\t"
694 			      "call handle_stack_overflow\n\t"
695 			      "1: jmp 1b"
696 			      : ASM_CALL_CONSTRAINT
697 			      : "D" ("kernel stack overflow (page fault)"),
698 				"S" (regs), "d" (address),
699 				[stack] "rm" (stack));
700 		unreachable();
701 	}
702 #endif
703 
704 	/*
705 	 * 32-bit:
706 	 *
707 	 *   Valid to do another page fault here, because if this fault
708 	 *   had been triggered by is_prefetch fixup_exception would have
709 	 *   handled it.
710 	 *
711 	 * 64-bit:
712 	 *
713 	 *   Hall of shame of CPU/BIOS bugs.
714 	 */
715 	if (is_prefetch(regs, error_code, address))
716 		return;
717 
718 	if (is_errata93(regs, address))
719 		return;
720 
721 	/*
722 	 * Buggy firmware could access regions which might page fault, try to
723 	 * recover from such faults.
724 	 */
725 	if (IS_ENABLED(CONFIG_EFI))
726 		efi_recover_from_page_fault(address);
727 
728 oops:
729 	/*
730 	 * Oops. The kernel tried to access some bad page. We'll have to
731 	 * terminate things with extreme prejudice:
732 	 */
733 	flags = oops_begin();
734 
735 	show_fault_oops(regs, error_code, address);
736 
737 	if (task_stack_end_corrupted(tsk))
738 		printk(KERN_EMERG "Thread overran stack, or stack corrupted\n");
739 
740 	sig = SIGKILL;
741 	if (__die("Oops", regs, error_code))
742 		sig = 0;
743 
744 	/* Executive summary in case the body of the oops scrolled away */
745 	printk(KERN_DEFAULT "CR2: %016lx\n", address);
746 
747 	oops_end(flags, regs, sig);
748 }
749 
750 /*
751  * Print out info about fatal segfaults, if the show_unhandled_signals
752  * sysctl is set:
753  */
754 static inline void
755 show_signal_msg(struct pt_regs *regs, unsigned long error_code,
756 		unsigned long address, struct task_struct *tsk)
757 {
758 	const char *loglvl = task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG;
759 
760 	if (!unhandled_signal(tsk, SIGSEGV))
761 		return;
762 
763 	if (!printk_ratelimit())
764 		return;
765 
766 	printk("%s%s[%d]: segfault at %lx ip %px sp %px error %lx",
767 		loglvl, tsk->comm, task_pid_nr(tsk), address,
768 		(void *)regs->ip, (void *)regs->sp, error_code);
769 
770 	print_vma_addr(KERN_CONT " in ", regs->ip);
771 
772 	printk(KERN_CONT "\n");
773 
774 	show_opcodes(regs, loglvl);
775 }
776 
777 /*
778  * The (legacy) vsyscall page is the long page in the kernel portion
779  * of the address space that has user-accessible permissions.
780  */
781 static bool is_vsyscall_vaddr(unsigned long vaddr)
782 {
783 	return unlikely((vaddr & PAGE_MASK) == VSYSCALL_ADDR);
784 }
785 
786 static void
787 __bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
788 		       unsigned long address, u32 pkey, int si_code)
789 {
790 	struct task_struct *tsk = current;
791 
792 	/* User mode accesses just cause a SIGSEGV */
793 	if (user_mode(regs) && (error_code & X86_PF_USER)) {
794 		/*
795 		 * It's possible to have interrupts off here:
796 		 */
797 		local_irq_enable();
798 
799 		/*
800 		 * Valid to do another page fault here because this one came
801 		 * from user space:
802 		 */
803 		if (is_prefetch(regs, error_code, address))
804 			return;
805 
806 		if (is_errata100(regs, address))
807 			return;
808 
809 		/*
810 		 * To avoid leaking information about the kernel page table
811 		 * layout, pretend that user-mode accesses to kernel addresses
812 		 * are always protection faults.
813 		 */
814 		if (address >= TASK_SIZE_MAX)
815 			error_code |= X86_PF_PROT;
816 
817 		if (likely(show_unhandled_signals))
818 			show_signal_msg(regs, error_code, address, tsk);
819 
820 		set_signal_archinfo(address, error_code);
821 
822 		if (si_code == SEGV_PKUERR)
823 			force_sig_pkuerr((void __user *)address, pkey);
824 
825 		force_sig_fault(SIGSEGV, si_code, (void __user *)address);
826 
827 		local_irq_disable();
828 
829 		return;
830 	}
831 
832 	if (is_f00f_bug(regs, address))
833 		return;
834 
835 	no_context(regs, error_code, address, SIGSEGV, si_code);
836 }
837 
838 static noinline void
839 bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
840 		     unsigned long address)
841 {
842 	__bad_area_nosemaphore(regs, error_code, address, 0, SEGV_MAPERR);
843 }
844 
845 static void
846 __bad_area(struct pt_regs *regs, unsigned long error_code,
847 	   unsigned long address, u32 pkey, int si_code)
848 {
849 	struct mm_struct *mm = current->mm;
850 	/*
851 	 * Something tried to access memory that isn't in our memory map..
852 	 * Fix it, but check if it's kernel or user first..
853 	 */
854 	mmap_read_unlock(mm);
855 
856 	__bad_area_nosemaphore(regs, error_code, address, pkey, si_code);
857 }
858 
859 static noinline void
860 bad_area(struct pt_regs *regs, unsigned long error_code, unsigned long address)
861 {
862 	__bad_area(regs, error_code, address, 0, SEGV_MAPERR);
863 }
864 
865 static inline bool bad_area_access_from_pkeys(unsigned long error_code,
866 		struct vm_area_struct *vma)
867 {
868 	/* This code is always called on the current mm */
869 	bool foreign = false;
870 
871 	if (!boot_cpu_has(X86_FEATURE_OSPKE))
872 		return false;
873 	if (error_code & X86_PF_PK)
874 		return true;
875 	/* this checks permission keys on the VMA: */
876 	if (!arch_vma_access_permitted(vma, (error_code & X86_PF_WRITE),
877 				       (error_code & X86_PF_INSTR), foreign))
878 		return true;
879 	return false;
880 }
881 
882 static noinline void
883 bad_area_access_error(struct pt_regs *regs, unsigned long error_code,
884 		      unsigned long address, struct vm_area_struct *vma)
885 {
886 	/*
887 	 * This OSPKE check is not strictly necessary at runtime.
888 	 * But, doing it this way allows compiler optimizations
889 	 * if pkeys are compiled out.
890 	 */
891 	if (bad_area_access_from_pkeys(error_code, vma)) {
892 		/*
893 		 * A protection key fault means that the PKRU value did not allow
894 		 * access to some PTE.  Userspace can figure out what PKRU was
895 		 * from the XSAVE state.  This function captures the pkey from
896 		 * the vma and passes it to userspace so userspace can discover
897 		 * which protection key was set on the PTE.
898 		 *
899 		 * If we get here, we know that the hardware signaled a X86_PF_PK
900 		 * fault and that there was a VMA once we got in the fault
901 		 * handler.  It does *not* guarantee that the VMA we find here
902 		 * was the one that we faulted on.
903 		 *
904 		 * 1. T1   : mprotect_key(foo, PAGE_SIZE, pkey=4);
905 		 * 2. T1   : set PKRU to deny access to pkey=4, touches page
906 		 * 3. T1   : faults...
907 		 * 4.    T2: mprotect_key(foo, PAGE_SIZE, pkey=5);
908 		 * 5. T1   : enters fault handler, takes mmap_lock, etc...
909 		 * 6. T1   : reaches here, sees vma_pkey(vma)=5, when we really
910 		 *	     faulted on a pte with its pkey=4.
911 		 */
912 		u32 pkey = vma_pkey(vma);
913 
914 		__bad_area(regs, error_code, address, pkey, SEGV_PKUERR);
915 	} else {
916 		__bad_area(regs, error_code, address, 0, SEGV_ACCERR);
917 	}
918 }
919 
920 static void
921 do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address,
922 	  vm_fault_t fault)
923 {
924 	/* Kernel mode? Handle exceptions or die: */
925 	if (!(error_code & X86_PF_USER)) {
926 		no_context(regs, error_code, address, SIGBUS, BUS_ADRERR);
927 		return;
928 	}
929 
930 	/* User-space => ok to do another page fault: */
931 	if (is_prefetch(regs, error_code, address))
932 		return;
933 
934 	set_signal_archinfo(address, error_code);
935 
936 #ifdef CONFIG_MEMORY_FAILURE
937 	if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) {
938 		struct task_struct *tsk = current;
939 		unsigned lsb = 0;
940 
941 		pr_err(
942 	"MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n",
943 			tsk->comm, tsk->pid, address);
944 		if (fault & VM_FAULT_HWPOISON_LARGE)
945 			lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault));
946 		if (fault & VM_FAULT_HWPOISON)
947 			lsb = PAGE_SHIFT;
948 		force_sig_mceerr(BUS_MCEERR_AR, (void __user *)address, lsb);
949 		return;
950 	}
951 #endif
952 	force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)address);
953 }
954 
955 static noinline void
956 mm_fault_error(struct pt_regs *regs, unsigned long error_code,
957 	       unsigned long address, vm_fault_t fault)
958 {
959 	if (fatal_signal_pending(current) && !(error_code & X86_PF_USER)) {
960 		no_context(regs, error_code, address, 0, 0);
961 		return;
962 	}
963 
964 	if (fault & VM_FAULT_OOM) {
965 		/* Kernel mode? Handle exceptions or die: */
966 		if (!(error_code & X86_PF_USER)) {
967 			no_context(regs, error_code, address,
968 				   SIGSEGV, SEGV_MAPERR);
969 			return;
970 		}
971 
972 		/*
973 		 * We ran out of memory, call the OOM killer, and return the
974 		 * userspace (which will retry the fault, or kill us if we got
975 		 * oom-killed):
976 		 */
977 		pagefault_out_of_memory();
978 	} else {
979 		if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON|
980 			     VM_FAULT_HWPOISON_LARGE))
981 			do_sigbus(regs, error_code, address, fault);
982 		else if (fault & VM_FAULT_SIGSEGV)
983 			bad_area_nosemaphore(regs, error_code, address);
984 		else
985 			BUG();
986 	}
987 }
988 
989 static int spurious_kernel_fault_check(unsigned long error_code, pte_t *pte)
990 {
991 	if ((error_code & X86_PF_WRITE) && !pte_write(*pte))
992 		return 0;
993 
994 	if ((error_code & X86_PF_INSTR) && !pte_exec(*pte))
995 		return 0;
996 
997 	return 1;
998 }
999 
1000 /*
1001  * Handle a spurious fault caused by a stale TLB entry.
1002  *
1003  * This allows us to lazily refresh the TLB when increasing the
1004  * permissions of a kernel page (RO -> RW or NX -> X).  Doing it
1005  * eagerly is very expensive since that implies doing a full
1006  * cross-processor TLB flush, even if no stale TLB entries exist
1007  * on other processors.
1008  *
1009  * Spurious faults may only occur if the TLB contains an entry with
1010  * fewer permission than the page table entry.  Non-present (P = 0)
1011  * and reserved bit (R = 1) faults are never spurious.
1012  *
1013  * There are no security implications to leaving a stale TLB when
1014  * increasing the permissions on a page.
1015  *
1016  * Returns non-zero if a spurious fault was handled, zero otherwise.
1017  *
1018  * See Intel Developer's Manual Vol 3 Section 4.10.4.3, bullet 3
1019  * (Optional Invalidation).
1020  */
1021 static noinline int
1022 spurious_kernel_fault(unsigned long error_code, unsigned long address)
1023 {
1024 	pgd_t *pgd;
1025 	p4d_t *p4d;
1026 	pud_t *pud;
1027 	pmd_t *pmd;
1028 	pte_t *pte;
1029 	int ret;
1030 
1031 	/*
1032 	 * Only writes to RO or instruction fetches from NX may cause
1033 	 * spurious faults.
1034 	 *
1035 	 * These could be from user or supervisor accesses but the TLB
1036 	 * is only lazily flushed after a kernel mapping protection
1037 	 * change, so user accesses are not expected to cause spurious
1038 	 * faults.
1039 	 */
1040 	if (error_code != (X86_PF_WRITE | X86_PF_PROT) &&
1041 	    error_code != (X86_PF_INSTR | X86_PF_PROT))
1042 		return 0;
1043 
1044 	pgd = init_mm.pgd + pgd_index(address);
1045 	if (!pgd_present(*pgd))
1046 		return 0;
1047 
1048 	p4d = p4d_offset(pgd, address);
1049 	if (!p4d_present(*p4d))
1050 		return 0;
1051 
1052 	if (p4d_large(*p4d))
1053 		return spurious_kernel_fault_check(error_code, (pte_t *) p4d);
1054 
1055 	pud = pud_offset(p4d, address);
1056 	if (!pud_present(*pud))
1057 		return 0;
1058 
1059 	if (pud_large(*pud))
1060 		return spurious_kernel_fault_check(error_code, (pte_t *) pud);
1061 
1062 	pmd = pmd_offset(pud, address);
1063 	if (!pmd_present(*pmd))
1064 		return 0;
1065 
1066 	if (pmd_large(*pmd))
1067 		return spurious_kernel_fault_check(error_code, (pte_t *) pmd);
1068 
1069 	pte = pte_offset_kernel(pmd, address);
1070 	if (!pte_present(*pte))
1071 		return 0;
1072 
1073 	ret = spurious_kernel_fault_check(error_code, pte);
1074 	if (!ret)
1075 		return 0;
1076 
1077 	/*
1078 	 * Make sure we have permissions in PMD.
1079 	 * If not, then there's a bug in the page tables:
1080 	 */
1081 	ret = spurious_kernel_fault_check(error_code, (pte_t *) pmd);
1082 	WARN_ONCE(!ret, "PMD has incorrect permission bits\n");
1083 
1084 	return ret;
1085 }
1086 NOKPROBE_SYMBOL(spurious_kernel_fault);
1087 
1088 int show_unhandled_signals = 1;
1089 
1090 static inline int
1091 access_error(unsigned long error_code, struct vm_area_struct *vma)
1092 {
1093 	/* This is only called for the current mm, so: */
1094 	bool foreign = false;
1095 
1096 	/*
1097 	 * Read or write was blocked by protection keys.  This is
1098 	 * always an unconditional error and can never result in
1099 	 * a follow-up action to resolve the fault, like a COW.
1100 	 */
1101 	if (error_code & X86_PF_PK)
1102 		return 1;
1103 
1104 	/*
1105 	 * Make sure to check the VMA so that we do not perform
1106 	 * faults just to hit a X86_PF_PK as soon as we fill in a
1107 	 * page.
1108 	 */
1109 	if (!arch_vma_access_permitted(vma, (error_code & X86_PF_WRITE),
1110 				       (error_code & X86_PF_INSTR), foreign))
1111 		return 1;
1112 
1113 	if (error_code & X86_PF_WRITE) {
1114 		/* write, present and write, not present: */
1115 		if (unlikely(!(vma->vm_flags & VM_WRITE)))
1116 			return 1;
1117 		return 0;
1118 	}
1119 
1120 	/* read, present: */
1121 	if (unlikely(error_code & X86_PF_PROT))
1122 		return 1;
1123 
1124 	/* read, not present: */
1125 	if (unlikely(!vma_is_accessible(vma)))
1126 		return 1;
1127 
1128 	return 0;
1129 }
1130 
1131 bool fault_in_kernel_space(unsigned long address)
1132 {
1133 	/*
1134 	 * On 64-bit systems, the vsyscall page is at an address above
1135 	 * TASK_SIZE_MAX, but is not considered part of the kernel
1136 	 * address space.
1137 	 */
1138 	if (IS_ENABLED(CONFIG_X86_64) && is_vsyscall_vaddr(address))
1139 		return false;
1140 
1141 	return address >= TASK_SIZE_MAX;
1142 }
1143 
1144 /*
1145  * Called for all faults where 'address' is part of the kernel address
1146  * space.  Might get called for faults that originate from *code* that
1147  * ran in userspace or the kernel.
1148  */
1149 static void
1150 do_kern_addr_fault(struct pt_regs *regs, unsigned long hw_error_code,
1151 		   unsigned long address)
1152 {
1153 	/*
1154 	 * Protection keys exceptions only happen on user pages.  We
1155 	 * have no user pages in the kernel portion of the address
1156 	 * space, so do not expect them here.
1157 	 */
1158 	WARN_ON_ONCE(hw_error_code & X86_PF_PK);
1159 
1160 #ifdef CONFIG_X86_32
1161 	/*
1162 	 * We can fault-in kernel-space virtual memory on-demand. The
1163 	 * 'reference' page table is init_mm.pgd.
1164 	 *
1165 	 * NOTE! We MUST NOT take any locks for this case. We may
1166 	 * be in an interrupt or a critical region, and should
1167 	 * only copy the information from the master page table,
1168 	 * nothing more.
1169 	 *
1170 	 * Before doing this on-demand faulting, ensure that the
1171 	 * fault is not any of the following:
1172 	 * 1. A fault on a PTE with a reserved bit set.
1173 	 * 2. A fault caused by a user-mode access.  (Do not demand-
1174 	 *    fault kernel memory due to user-mode accesses).
1175 	 * 3. A fault caused by a page-level protection violation.
1176 	 *    (A demand fault would be on a non-present page which
1177 	 *     would have X86_PF_PROT==0).
1178 	 *
1179 	 * This is only needed to close a race condition on x86-32 in
1180 	 * the vmalloc mapping/unmapping code. See the comment above
1181 	 * vmalloc_fault() for details. On x86-64 the race does not
1182 	 * exist as the vmalloc mappings don't need to be synchronized
1183 	 * there.
1184 	 */
1185 	if (!(hw_error_code & (X86_PF_RSVD | X86_PF_USER | X86_PF_PROT))) {
1186 		if (vmalloc_fault(address) >= 0)
1187 			return;
1188 	}
1189 #endif
1190 
1191 	/* Was the fault spurious, caused by lazy TLB invalidation? */
1192 	if (spurious_kernel_fault(hw_error_code, address))
1193 		return;
1194 
1195 	/* kprobes don't want to hook the spurious faults: */
1196 	if (kprobe_page_fault(regs, X86_TRAP_PF))
1197 		return;
1198 
1199 	/*
1200 	 * Note, despite being a "bad area", there are quite a few
1201 	 * acceptable reasons to get here, such as erratum fixups
1202 	 * and handling kernel code that can fault, like get_user().
1203 	 *
1204 	 * Don't take the mm semaphore here. If we fixup a prefetch
1205 	 * fault we could otherwise deadlock:
1206 	 */
1207 	bad_area_nosemaphore(regs, hw_error_code, address);
1208 }
1209 NOKPROBE_SYMBOL(do_kern_addr_fault);
1210 
1211 /* Handle faults in the user portion of the address space */
1212 static inline
1213 void do_user_addr_fault(struct pt_regs *regs,
1214 			unsigned long hw_error_code,
1215 			unsigned long address)
1216 {
1217 	struct vm_area_struct *vma;
1218 	struct task_struct *tsk;
1219 	struct mm_struct *mm;
1220 	vm_fault_t fault;
1221 	unsigned int flags = FAULT_FLAG_DEFAULT;
1222 
1223 	tsk = current;
1224 	mm = tsk->mm;
1225 
1226 	/* kprobes don't want to hook the spurious faults: */
1227 	if (unlikely(kprobe_page_fault(regs, X86_TRAP_PF)))
1228 		return;
1229 
1230 	/*
1231 	 * Reserved bits are never expected to be set on
1232 	 * entries in the user portion of the page tables.
1233 	 */
1234 	if (unlikely(hw_error_code & X86_PF_RSVD))
1235 		pgtable_bad(regs, hw_error_code, address);
1236 
1237 	/*
1238 	 * If SMAP is on, check for invalid kernel (supervisor) access to user
1239 	 * pages in the user address space.  The odd case here is WRUSS,
1240 	 * which, according to the preliminary documentation, does not respect
1241 	 * SMAP and will have the USER bit set so, in all cases, SMAP
1242 	 * enforcement appears to be consistent with the USER bit.
1243 	 */
1244 	if (unlikely(cpu_feature_enabled(X86_FEATURE_SMAP) &&
1245 		     !(hw_error_code & X86_PF_USER) &&
1246 		     !(regs->flags & X86_EFLAGS_AC)))
1247 	{
1248 		bad_area_nosemaphore(regs, hw_error_code, address);
1249 		return;
1250 	}
1251 
1252 	/*
1253 	 * If we're in an interrupt, have no user context or are running
1254 	 * in a region with pagefaults disabled then we must not take the fault
1255 	 */
1256 	if (unlikely(faulthandler_disabled() || !mm)) {
1257 		bad_area_nosemaphore(regs, hw_error_code, address);
1258 		return;
1259 	}
1260 
1261 	/*
1262 	 * It's safe to allow irq's after cr2 has been saved and the
1263 	 * vmalloc fault has been handled.
1264 	 *
1265 	 * User-mode registers count as a user access even for any
1266 	 * potential system fault or CPU buglet:
1267 	 */
1268 	if (user_mode(regs)) {
1269 		local_irq_enable();
1270 		flags |= FAULT_FLAG_USER;
1271 	} else {
1272 		if (regs->flags & X86_EFLAGS_IF)
1273 			local_irq_enable();
1274 	}
1275 
1276 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
1277 
1278 	if (hw_error_code & X86_PF_WRITE)
1279 		flags |= FAULT_FLAG_WRITE;
1280 	if (hw_error_code & X86_PF_INSTR)
1281 		flags |= FAULT_FLAG_INSTRUCTION;
1282 
1283 #ifdef CONFIG_X86_64
1284 	/*
1285 	 * Faults in the vsyscall page might need emulation.  The
1286 	 * vsyscall page is at a high address (>PAGE_OFFSET), but is
1287 	 * considered to be part of the user address space.
1288 	 *
1289 	 * The vsyscall page does not have a "real" VMA, so do this
1290 	 * emulation before we go searching for VMAs.
1291 	 *
1292 	 * PKRU never rejects instruction fetches, so we don't need
1293 	 * to consider the PF_PK bit.
1294 	 */
1295 	if (is_vsyscall_vaddr(address)) {
1296 		if (emulate_vsyscall(hw_error_code, regs, address))
1297 			return;
1298 	}
1299 #endif
1300 
1301 	/*
1302 	 * Kernel-mode access to the user address space should only occur
1303 	 * on well-defined single instructions listed in the exception
1304 	 * tables.  But, an erroneous kernel fault occurring outside one of
1305 	 * those areas which also holds mmap_lock might deadlock attempting
1306 	 * to validate the fault against the address space.
1307 	 *
1308 	 * Only do the expensive exception table search when we might be at
1309 	 * risk of a deadlock.  This happens if we
1310 	 * 1. Failed to acquire mmap_lock, and
1311 	 * 2. The access did not originate in userspace.
1312 	 */
1313 	if (unlikely(!mmap_read_trylock(mm))) {
1314 		if (!user_mode(regs) && !search_exception_tables(regs->ip)) {
1315 			/*
1316 			 * Fault from code in kernel from
1317 			 * which we do not expect faults.
1318 			 */
1319 			bad_area_nosemaphore(regs, hw_error_code, address);
1320 			return;
1321 		}
1322 retry:
1323 		mmap_read_lock(mm);
1324 	} else {
1325 		/*
1326 		 * The above down_read_trylock() might have succeeded in
1327 		 * which case we'll have missed the might_sleep() from
1328 		 * down_read():
1329 		 */
1330 		might_sleep();
1331 	}
1332 
1333 	vma = find_vma(mm, address);
1334 	if (unlikely(!vma)) {
1335 		bad_area(regs, hw_error_code, address);
1336 		return;
1337 	}
1338 	if (likely(vma->vm_start <= address))
1339 		goto good_area;
1340 	if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) {
1341 		bad_area(regs, hw_error_code, address);
1342 		return;
1343 	}
1344 	if (unlikely(expand_stack(vma, address))) {
1345 		bad_area(regs, hw_error_code, address);
1346 		return;
1347 	}
1348 
1349 	/*
1350 	 * Ok, we have a good vm_area for this memory access, so
1351 	 * we can handle it..
1352 	 */
1353 good_area:
1354 	if (unlikely(access_error(hw_error_code, vma))) {
1355 		bad_area_access_error(regs, hw_error_code, address, vma);
1356 		return;
1357 	}
1358 
1359 	/*
1360 	 * If for any reason at all we couldn't handle the fault,
1361 	 * make sure we exit gracefully rather than endlessly redo
1362 	 * the fault.  Since we never set FAULT_FLAG_RETRY_NOWAIT, if
1363 	 * we get VM_FAULT_RETRY back, the mmap_lock has been unlocked.
1364 	 *
1365 	 * Note that handle_userfault() may also release and reacquire mmap_lock
1366 	 * (and not return with VM_FAULT_RETRY), when returning to userland to
1367 	 * repeat the page fault later with a VM_FAULT_NOPAGE retval
1368 	 * (potentially after handling any pending signal during the return to
1369 	 * userland). The return to userland is identified whenever
1370 	 * FAULT_FLAG_USER|FAULT_FLAG_KILLABLE are both set in flags.
1371 	 */
1372 	fault = handle_mm_fault(vma, address, flags, regs);
1373 
1374 	/* Quick path to respond to signals */
1375 	if (fault_signal_pending(fault, regs)) {
1376 		if (!user_mode(regs))
1377 			no_context(regs, hw_error_code, address, SIGBUS,
1378 				   BUS_ADRERR);
1379 		return;
1380 	}
1381 
1382 	/*
1383 	 * If we need to retry the mmap_lock has already been released,
1384 	 * and if there is a fatal signal pending there is no guarantee
1385 	 * that we made any progress. Handle this case first.
1386 	 */
1387 	if (unlikely((fault & VM_FAULT_RETRY) &&
1388 		     (flags & FAULT_FLAG_ALLOW_RETRY))) {
1389 		flags |= FAULT_FLAG_TRIED;
1390 		goto retry;
1391 	}
1392 
1393 	mmap_read_unlock(mm);
1394 	if (unlikely(fault & VM_FAULT_ERROR)) {
1395 		mm_fault_error(regs, hw_error_code, address, fault);
1396 		return;
1397 	}
1398 
1399 	check_v8086_mode(regs, address, tsk);
1400 }
1401 NOKPROBE_SYMBOL(do_user_addr_fault);
1402 
1403 static __always_inline void
1404 trace_page_fault_entries(struct pt_regs *regs, unsigned long error_code,
1405 			 unsigned long address)
1406 {
1407 	if (!trace_pagefault_enabled())
1408 		return;
1409 
1410 	if (user_mode(regs))
1411 		trace_page_fault_user(address, regs, error_code);
1412 	else
1413 		trace_page_fault_kernel(address, regs, error_code);
1414 }
1415 
1416 static __always_inline void
1417 handle_page_fault(struct pt_regs *regs, unsigned long error_code,
1418 			      unsigned long address)
1419 {
1420 	trace_page_fault_entries(regs, error_code, address);
1421 
1422 	if (unlikely(kmmio_fault(regs, address)))
1423 		return;
1424 
1425 	/* Was the fault on kernel-controlled part of the address space? */
1426 	if (unlikely(fault_in_kernel_space(address))) {
1427 		do_kern_addr_fault(regs, error_code, address);
1428 	} else {
1429 		do_user_addr_fault(regs, error_code, address);
1430 		/*
1431 		 * User address page fault handling might have reenabled
1432 		 * interrupts. Fixing up all potential exit points of
1433 		 * do_user_addr_fault() and its leaf functions is just not
1434 		 * doable w/o creating an unholy mess or turning the code
1435 		 * upside down.
1436 		 */
1437 		local_irq_disable();
1438 	}
1439 }
1440 
1441 DEFINE_IDTENTRY_RAW_ERRORCODE(exc_page_fault)
1442 {
1443 	unsigned long address = read_cr2();
1444 	irqentry_state_t state;
1445 
1446 	prefetchw(&current->mm->mmap_lock);
1447 
1448 	/*
1449 	 * KVM uses #PF vector to deliver 'page not present' events to guests
1450 	 * (asynchronous page fault mechanism). The event happens when a
1451 	 * userspace task is trying to access some valid (from guest's point of
1452 	 * view) memory which is not currently mapped by the host (e.g. the
1453 	 * memory is swapped out). Note, the corresponding "page ready" event
1454 	 * which is injected when the memory becomes available, is delived via
1455 	 * an interrupt mechanism and not a #PF exception
1456 	 * (see arch/x86/kernel/kvm.c: sysvec_kvm_asyncpf_interrupt()).
1457 	 *
1458 	 * We are relying on the interrupted context being sane (valid RSP,
1459 	 * relevant locks not held, etc.), which is fine as long as the
1460 	 * interrupted context had IF=1.  We are also relying on the KVM
1461 	 * async pf type field and CR2 being read consistently instead of
1462 	 * getting values from real and async page faults mixed up.
1463 	 *
1464 	 * Fingers crossed.
1465 	 *
1466 	 * The async #PF handling code takes care of idtentry handling
1467 	 * itself.
1468 	 */
1469 	if (kvm_handle_async_pf(regs, (u32)address))
1470 		return;
1471 
1472 	/*
1473 	 * Entry handling for valid #PF from kernel mode is slightly
1474 	 * different: RCU is already watching and rcu_irq_enter() must not
1475 	 * be invoked because a kernel fault on a user space address might
1476 	 * sleep.
1477 	 *
1478 	 * In case the fault hit a RCU idle region the conditional entry
1479 	 * code reenabled RCU to avoid subsequent wreckage which helps
1480 	 * debugability.
1481 	 */
1482 	state = irqentry_enter(regs);
1483 
1484 	instrumentation_begin();
1485 	handle_page_fault(regs, error_code, address);
1486 	instrumentation_end();
1487 
1488 	irqentry_exit(regs, state);
1489 }
1490