1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * Copyright(c) 2017 Intel Corporation. All rights reserved.
4 *
5 * This code is based in part on work published here:
6 *
7 * https://github.com/IAIK/KAISER
8 *
9 * The original work was written by and signed off by for the Linux
10 * kernel by:
11 *
12 * Signed-off-by: Richard Fellner <richard.fellner@student.tugraz.at>
13 * Signed-off-by: Moritz Lipp <moritz.lipp@iaik.tugraz.at>
14 * Signed-off-by: Daniel Gruss <daniel.gruss@iaik.tugraz.at>
15 * Signed-off-by: Michael Schwarz <michael.schwarz@iaik.tugraz.at>
16 *
17 * Major changes to the original code by: Dave Hansen <dave.hansen@intel.com>
18 * Mostly rewritten by Thomas Gleixner <tglx@linutronix.de> and
19 * Andy Lutomirsky <luto@amacapital.net>
20 */
21 #include <linux/kernel.h>
22 #include <linux/errno.h>
23 #include <linux/string.h>
24 #include <linux/types.h>
25 #include <linux/bug.h>
26 #include <linux/init.h>
27 #include <linux/spinlock.h>
28 #include <linux/mm.h>
29 #include <linux/uaccess.h>
30 #include <linux/cpu.h>
31
32 #include <asm/cpufeature.h>
33 #include <asm/hypervisor.h>
34 #include <asm/vsyscall.h>
35 #include <asm/cmdline.h>
36 #include <asm/pti.h>
37 #include <asm/tlbflush.h>
38 #include <asm/desc.h>
39 #include <asm/sections.h>
40 #include <asm/set_memory.h>
41
42 #undef pr_fmt
43 #define pr_fmt(fmt) "Kernel/User page tables isolation: " fmt
44
45 /* Backporting helper */
46 #ifndef __GFP_NOTRACK
47 #define __GFP_NOTRACK 0
48 #endif
49
50 /*
51 * Define the page-table levels we clone for user-space on 32
52 * and 64 bit.
53 */
54 #ifdef CONFIG_X86_64
55 #define PTI_LEVEL_KERNEL_IMAGE PTI_CLONE_PMD
56 #else
57 #define PTI_LEVEL_KERNEL_IMAGE PTI_CLONE_PTE
58 #endif
59
pti_print_if_insecure(const char * reason)60 static void __init pti_print_if_insecure(const char *reason)
61 {
62 if (boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
63 pr_info("%s\n", reason);
64 }
65
pti_print_if_secure(const char * reason)66 static void __init pti_print_if_secure(const char *reason)
67 {
68 if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
69 pr_info("%s\n", reason);
70 }
71
72 /* Assume mode is auto unless overridden via cmdline below. */
73 static enum pti_mode {
74 PTI_AUTO = 0,
75 PTI_FORCE_OFF,
76 PTI_FORCE_ON
77 } pti_mode;
78
pti_check_boottime_disable(void)79 void __init pti_check_boottime_disable(void)
80 {
81 if (hypervisor_is_type(X86_HYPER_XEN_PV)) {
82 pti_mode = PTI_FORCE_OFF;
83 pti_print_if_insecure("disabled on XEN PV.");
84 return;
85 }
86
87 if (cpu_mitigations_off())
88 pti_mode = PTI_FORCE_OFF;
89 if (pti_mode == PTI_FORCE_OFF) {
90 pti_print_if_insecure("disabled on command line.");
91 return;
92 }
93
94 if (pti_mode == PTI_FORCE_ON)
95 pti_print_if_secure("force enabled on command line.");
96
97 if (pti_mode == PTI_AUTO && !boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
98 return;
99
100 setup_force_cpu_cap(X86_FEATURE_PTI);
101
102 if (cpu_feature_enabled(X86_FEATURE_INVLPGB)) {
103 pr_debug("PTI enabled, disabling INVLPGB\n");
104 setup_clear_cpu_cap(X86_FEATURE_INVLPGB);
105 }
106 }
107
pti_parse_cmdline(char * arg)108 static int __init pti_parse_cmdline(char *arg)
109 {
110 if (!strcmp(arg, "off"))
111 pti_mode = PTI_FORCE_OFF;
112 else if (!strcmp(arg, "on"))
113 pti_mode = PTI_FORCE_ON;
114 else if (!strcmp(arg, "auto"))
115 pti_mode = PTI_AUTO;
116 else
117 return -EINVAL;
118 return 0;
119 }
120 early_param("pti", pti_parse_cmdline);
121
pti_parse_cmdline_nopti(char * arg)122 static int __init pti_parse_cmdline_nopti(char *arg)
123 {
124 pti_mode = PTI_FORCE_OFF;
125 return 0;
126 }
127 early_param("nopti", pti_parse_cmdline_nopti);
128
__pti_set_user_pgtbl(pgd_t * pgdp,pgd_t pgd)129 pgd_t __pti_set_user_pgtbl(pgd_t *pgdp, pgd_t pgd)
130 {
131 /*
132 * Changes to the high (kernel) portion of the kernelmode page
133 * tables are not automatically propagated to the usermode tables.
134 *
135 * Users should keep in mind that, unlike the kernelmode tables,
136 * there is no vmalloc_fault equivalent for the usermode tables.
137 * Top-level entries added to init_mm's usermode pgd after boot
138 * will not be automatically propagated to other mms.
139 */
140 if (!pgdp_maps_userspace(pgdp) || (pgd.pgd & _PAGE_NOPTISHADOW))
141 return pgd;
142
143 /*
144 * The user page tables get the full PGD, accessible from
145 * userspace:
146 */
147 kernel_to_user_pgdp(pgdp)->pgd = pgd.pgd;
148
149 /*
150 * If this is normal user memory, make it NX in the kernel
151 * pagetables so that, if we somehow screw up and return to
152 * usermode with the kernel CR3 loaded, we'll get a page fault
153 * instead of allowing user code to execute with the wrong CR3.
154 *
155 * As exceptions, we don't set NX if:
156 * - _PAGE_USER is not set. This could be an executable
157 * EFI runtime mapping or something similar, and the kernel
158 * may execute from it
159 * - we don't have NX support
160 * - we're clearing the PGD (i.e. the new pgd is not present).
161 */
162 if ((pgd.pgd & (_PAGE_USER|_PAGE_PRESENT)) == (_PAGE_USER|_PAGE_PRESENT) &&
163 (__supported_pte_mask & _PAGE_NX))
164 pgd.pgd |= _PAGE_NX;
165
166 /* return the copy of the PGD we want the kernel to use: */
167 return pgd;
168 }
169
170 /*
171 * Walk the user copy of the page tables (optionally) trying to allocate
172 * page table pages on the way down.
173 *
174 * Returns a pointer to a P4D on success, or NULL on failure.
175 */
pti_user_pagetable_walk_p4d(unsigned long address)176 static p4d_t *pti_user_pagetable_walk_p4d(unsigned long address)
177 {
178 pgd_t *pgd = kernel_to_user_pgdp(pgd_offset_k(address));
179 gfp_t gfp = (GFP_KERNEL | __GFP_NOTRACK | __GFP_ZERO);
180
181 if (address < PAGE_OFFSET) {
182 WARN_ONCE(1, "attempt to walk user address\n");
183 return NULL;
184 }
185
186 if (pgd_none(*pgd)) {
187 unsigned long new_p4d_page = __get_free_page(gfp);
188 if (WARN_ON_ONCE(!new_p4d_page))
189 return NULL;
190
191 set_pgd(pgd, __pgd(_KERNPG_TABLE | __pa(new_p4d_page)));
192 }
193 BUILD_BUG_ON(pgd_leaf(*pgd));
194
195 return p4d_offset(pgd, address);
196 }
197
198 /*
199 * Walk the user copy of the page tables (optionally) trying to allocate
200 * page table pages on the way down.
201 *
202 * Returns a pointer to a PMD on success, or NULL on failure.
203 */
pti_user_pagetable_walk_pmd(unsigned long address)204 static pmd_t *pti_user_pagetable_walk_pmd(unsigned long address)
205 {
206 gfp_t gfp = (GFP_KERNEL | __GFP_NOTRACK | __GFP_ZERO);
207 p4d_t *p4d;
208 pud_t *pud;
209
210 p4d = pti_user_pagetable_walk_p4d(address);
211 if (!p4d)
212 return NULL;
213
214 BUILD_BUG_ON(p4d_leaf(*p4d));
215 if (p4d_none(*p4d)) {
216 unsigned long new_pud_page = __get_free_page(gfp);
217 if (WARN_ON_ONCE(!new_pud_page))
218 return NULL;
219
220 set_p4d(p4d, __p4d(_KERNPG_TABLE | __pa(new_pud_page)));
221 }
222
223 pud = pud_offset(p4d, address);
224 /* The user page tables do not use large mappings: */
225 if (pud_leaf(*pud)) {
226 WARN_ON(1);
227 return NULL;
228 }
229 if (pud_none(*pud)) {
230 unsigned long new_pmd_page = __get_free_page(gfp);
231 if (WARN_ON_ONCE(!new_pmd_page))
232 return NULL;
233
234 set_pud(pud, __pud(_KERNPG_TABLE | __pa(new_pmd_page)));
235 }
236
237 return pmd_offset(pud, address);
238 }
239
240 /*
241 * Walk the shadow copy of the page tables (optionally) trying to allocate
242 * page table pages on the way down. Does not support large pages.
243 *
244 * Note: this is only used when mapping *new* kernel data into the
245 * user/shadow page tables. It is never used for userspace data.
246 *
247 * Returns a pointer to a PTE on success, or NULL on failure.
248 */
pti_user_pagetable_walk_pte(unsigned long address,bool late_text)249 static pte_t *pti_user_pagetable_walk_pte(unsigned long address, bool late_text)
250 {
251 gfp_t gfp = (GFP_KERNEL | __GFP_NOTRACK | __GFP_ZERO);
252 pmd_t *pmd;
253 pte_t *pte;
254
255 pmd = pti_user_pagetable_walk_pmd(address);
256 if (!pmd)
257 return NULL;
258
259 /* Large PMD mapping found */
260 if (pmd_leaf(*pmd)) {
261 /* Clear the PMD if we hit a large mapping from the first round */
262 if (late_text) {
263 set_pmd(pmd, __pmd(0));
264 } else {
265 WARN_ON_ONCE(1);
266 return NULL;
267 }
268 }
269
270 if (pmd_none(*pmd)) {
271 unsigned long new_pte_page = __get_free_page(gfp);
272 if (!new_pte_page)
273 return NULL;
274
275 set_pmd(pmd, __pmd(_KERNPG_TABLE | __pa(new_pte_page)));
276 }
277
278 pte = pte_offset_kernel(pmd, address);
279 if (pte_flags(*pte) & _PAGE_USER) {
280 WARN_ONCE(1, "attempt to walk to user pte\n");
281 return NULL;
282 }
283 return pte;
284 }
285
286 #ifdef CONFIG_X86_VSYSCALL_EMULATION
pti_setup_vsyscall(void)287 static void __init pti_setup_vsyscall(void)
288 {
289 pte_t *pte, *target_pte;
290 unsigned int level;
291
292 pte = lookup_address(VSYSCALL_ADDR, &level);
293 if (!pte || WARN_ON(level != PG_LEVEL_4K) || pte_none(*pte))
294 return;
295
296 target_pte = pti_user_pagetable_walk_pte(VSYSCALL_ADDR, false);
297 if (WARN_ON(!target_pte))
298 return;
299
300 *target_pte = *pte;
301 set_vsyscall_pgtable_user_bits(kernel_to_user_pgdp(swapper_pg_dir));
302 }
303 #else
pti_setup_vsyscall(void)304 static void __init pti_setup_vsyscall(void) { }
305 #endif
306
307 enum pti_clone_level {
308 PTI_CLONE_PMD,
309 PTI_CLONE_PTE,
310 };
311
312 static void
pti_clone_pgtable(unsigned long start,unsigned long end,enum pti_clone_level level,bool late_text)313 pti_clone_pgtable(unsigned long start, unsigned long end,
314 enum pti_clone_level level, bool late_text)
315 {
316 unsigned long addr;
317
318 /*
319 * Clone the populated PMDs which cover start to end. These PMD areas
320 * can have holes.
321 */
322 for (addr = start; addr < end;) {
323 pte_t *pte, *target_pte;
324 pmd_t *pmd, *target_pmd;
325 pgd_t *pgd;
326 p4d_t *p4d;
327 pud_t *pud;
328
329 /* Overflow check */
330 if (addr < start)
331 break;
332
333 pgd = pgd_offset_k(addr);
334 if (WARN_ON(pgd_none(*pgd)))
335 return;
336 p4d = p4d_offset(pgd, addr);
337 if (WARN_ON(p4d_none(*p4d)))
338 return;
339
340 pud = pud_offset(p4d, addr);
341 if (pud_none(*pud)) {
342 WARN_ON_ONCE(addr & ~PUD_MASK);
343 addr = round_up(addr + 1, PUD_SIZE);
344 continue;
345 }
346
347 pmd = pmd_offset(pud, addr);
348 if (pmd_none(*pmd)) {
349 WARN_ON_ONCE(addr & ~PMD_MASK);
350 addr = round_up(addr + 1, PMD_SIZE);
351 continue;
352 }
353
354 if (pmd_leaf(*pmd) || level == PTI_CLONE_PMD) {
355 target_pmd = pti_user_pagetable_walk_pmd(addr);
356 if (WARN_ON(!target_pmd))
357 return;
358
359 /*
360 * Only clone present PMDs. This ensures only setting
361 * _PAGE_GLOBAL on present PMDs. This should only be
362 * called on well-known addresses anyway, so a non-
363 * present PMD would be a surprise.
364 */
365 if (WARN_ON(!(pmd_flags(*pmd) & _PAGE_PRESENT)))
366 return;
367
368 /*
369 * Setting 'target_pmd' below creates a mapping in both
370 * the user and kernel page tables. It is effectively
371 * global, so set it as global in both copies. Note:
372 * the X86_FEATURE_PGE check is not _required_ because
373 * the CPU ignores _PAGE_GLOBAL when PGE is not
374 * supported. The check keeps consistency with
375 * code that only set this bit when supported.
376 */
377 if (boot_cpu_has(X86_FEATURE_PGE))
378 *pmd = pmd_set_flags(*pmd, _PAGE_GLOBAL);
379
380 /*
381 * Copy the PMD. That is, the kernelmode and usermode
382 * tables will share the last-level page tables of this
383 * address range
384 */
385 *target_pmd = *pmd;
386
387 addr = round_up(addr + 1, PMD_SIZE);
388
389 } else if (level == PTI_CLONE_PTE) {
390
391 /* Walk the page-table down to the pte level */
392 pte = pte_offset_kernel(pmd, addr);
393 if (pte_none(*pte)) {
394 addr = round_up(addr + 1, PAGE_SIZE);
395 continue;
396 }
397
398 /* Only clone present PTEs */
399 if (WARN_ON(!(pte_flags(*pte) & _PAGE_PRESENT)))
400 return;
401
402 /* Allocate PTE in the user page-table */
403 target_pte = pti_user_pagetable_walk_pte(addr, late_text);
404 if (WARN_ON(!target_pte))
405 return;
406
407 /* Set GLOBAL bit in both PTEs */
408 if (boot_cpu_has(X86_FEATURE_PGE))
409 *pte = pte_set_flags(*pte, _PAGE_GLOBAL);
410
411 /* Clone the PTE */
412 *target_pte = *pte;
413
414 addr = round_up(addr + 1, PAGE_SIZE);
415
416 } else {
417 BUG();
418 }
419 }
420 }
421
422 #ifdef CONFIG_X86_64
423 /*
424 * Clone a single p4d (i.e. a top-level entry on 4-level systems and a
425 * next-level entry on 5-level systems.
426 */
pti_clone_p4d(unsigned long addr)427 static void __init pti_clone_p4d(unsigned long addr)
428 {
429 p4d_t *kernel_p4d, *user_p4d;
430 pgd_t *kernel_pgd;
431
432 user_p4d = pti_user_pagetable_walk_p4d(addr);
433 if (!user_p4d)
434 return;
435
436 kernel_pgd = pgd_offset_k(addr);
437 kernel_p4d = p4d_offset(kernel_pgd, addr);
438 *user_p4d = *kernel_p4d;
439 }
440
441 /*
442 * Clone the CPU_ENTRY_AREA and associated data into the user space visible
443 * page table.
444 */
pti_clone_user_shared(void)445 static void __init pti_clone_user_shared(void)
446 {
447 unsigned int cpu;
448
449 pti_clone_p4d(CPU_ENTRY_AREA_BASE);
450
451 for_each_possible_cpu(cpu) {
452 /*
453 * The SYSCALL64 entry code needs one word of scratch space
454 * in which to spill a register. It lives in the sp2 slot
455 * of the CPU's TSS.
456 *
457 * This is done for all possible CPUs during boot to ensure
458 * that it's propagated to all mms.
459 */
460
461 unsigned long va = (unsigned long)&per_cpu(cpu_tss_rw, cpu);
462 phys_addr_t pa = per_cpu_ptr_to_phys((void *)va);
463 pte_t *target_pte;
464
465 target_pte = pti_user_pagetable_walk_pte(va, false);
466 if (WARN_ON(!target_pte))
467 return;
468
469 *target_pte = pfn_pte(pa >> PAGE_SHIFT, PAGE_KERNEL);
470 }
471 }
472
473 #else /* CONFIG_X86_64 */
474
475 /*
476 * On 32 bit PAE systems with 1GB of Kernel address space there is only
477 * one pgd/p4d for the whole kernel. Cloning that would map the whole
478 * address space into the user page-tables, making PTI useless. So clone
479 * the page-table on the PMD level to prevent that.
480 */
pti_clone_user_shared(void)481 static void __init pti_clone_user_shared(void)
482 {
483 unsigned long start, end;
484
485 start = CPU_ENTRY_AREA_BASE;
486 end = start + (PAGE_SIZE * CPU_ENTRY_AREA_PAGES);
487
488 pti_clone_pgtable(start, end, PTI_CLONE_PMD, false);
489 }
490 #endif /* CONFIG_X86_64 */
491
492 /*
493 * Clone the ESPFIX P4D into the user space visible page table
494 */
pti_setup_espfix64(void)495 static void __init pti_setup_espfix64(void)
496 {
497 #ifdef CONFIG_X86_ESPFIX64
498 pti_clone_p4d(ESPFIX_BASE_ADDR);
499 #endif
500 }
501
502 /*
503 * Clone the populated PMDs of the entry text and force it RO.
504 */
pti_clone_entry_text(bool late)505 static void pti_clone_entry_text(bool late)
506 {
507 pti_clone_pgtable((unsigned long) __entry_text_start,
508 (unsigned long) __entry_text_end,
509 PTI_LEVEL_KERNEL_IMAGE, late);
510 }
511
512 /*
513 * Global pages and PCIDs are both ways to make kernel TLB entries
514 * live longer, reduce TLB misses and improve kernel performance.
515 * But, leaving all kernel text Global makes it potentially accessible
516 * to Meltdown-style attacks which make it trivial to find gadgets or
517 * defeat KASLR.
518 *
519 * Only use global pages when it is really worth it.
520 */
pti_kernel_image_global_ok(void)521 static inline bool pti_kernel_image_global_ok(void)
522 {
523 /*
524 * Systems with PCIDs get little benefit from global
525 * kernel text and are not worth the downsides.
526 */
527 if (cpu_feature_enabled(X86_FEATURE_PCID))
528 return false;
529
530 /*
531 * Only do global kernel image for pti=auto. Do the most
532 * secure thing (not global) if pti=on specified.
533 */
534 if (pti_mode != PTI_AUTO)
535 return false;
536
537 /*
538 * K8 may not tolerate the cleared _PAGE_RW on the userspace
539 * global kernel image pages. Do the safe thing (disable
540 * global kernel image). This is unlikely to ever be
541 * noticed because PTI is disabled by default on AMD CPUs.
542 */
543 if (boot_cpu_has(X86_FEATURE_K8))
544 return false;
545
546 /*
547 * RANDSTRUCT derives its hardening benefits from the
548 * attacker's lack of knowledge about the layout of kernel
549 * data structures. Keep the kernel image non-global in
550 * cases where RANDSTRUCT is in use to help keep the layout a
551 * secret.
552 */
553 if (IS_ENABLED(CONFIG_RANDSTRUCT))
554 return false;
555
556 return true;
557 }
558
559 /*
560 * For some configurations, map all of kernel text into the user page
561 * tables. This reduces TLB misses, especially on non-PCID systems.
562 */
pti_clone_kernel_text(void)563 static void pti_clone_kernel_text(void)
564 {
565 /*
566 * rodata is part of the kernel image and is normally
567 * readable on the filesystem or on the web. But, do not
568 * clone the areas past rodata, they might contain secrets.
569 */
570 unsigned long start = PFN_ALIGN(_text);
571 unsigned long end_clone = (unsigned long)__end_rodata_aligned;
572 unsigned long end_global = PFN_ALIGN((unsigned long)_etext);
573
574 if (!pti_kernel_image_global_ok())
575 return;
576
577 pr_debug("mapping partial kernel image into user address space\n");
578
579 /*
580 * Note that this will undo _some_ of the work that
581 * pti_set_kernel_image_nonglobal() did to clear the
582 * global bit.
583 */
584 pti_clone_pgtable(start, end_clone, PTI_LEVEL_KERNEL_IMAGE, false);
585
586 /*
587 * pti_clone_pgtable() will set the global bit in any PMDs
588 * that it clones, but we also need to get any PTEs in
589 * the last level for areas that are not huge-page-aligned.
590 */
591
592 /* Set the global bit for normal non-__init kernel text: */
593 set_memory_global(start, (end_global - start) >> PAGE_SHIFT);
594 }
595
pti_set_kernel_image_nonglobal(void)596 static void pti_set_kernel_image_nonglobal(void)
597 {
598 /*
599 * The identity map is created with PMDs, regardless of the
600 * actual length of the kernel. We need to clear
601 * _PAGE_GLOBAL up to a PMD boundary, not just to the end
602 * of the image.
603 */
604 unsigned long start = PFN_ALIGN(_text);
605 unsigned long end = ALIGN((unsigned long)_end, PMD_SIZE);
606
607 /*
608 * This clears _PAGE_GLOBAL from the entire kernel image.
609 * pti_clone_kernel_text() map put _PAGE_GLOBAL back for
610 * areas that are mapped to userspace.
611 */
612 set_memory_nonglobal(start, (end - start) >> PAGE_SHIFT);
613 }
614
615 /*
616 * Initialize kernel page table isolation
617 */
pti_init(void)618 void __init pti_init(void)
619 {
620 if (!boot_cpu_has(X86_FEATURE_PTI))
621 return;
622
623 pr_info("enabled\n");
624
625 #ifdef CONFIG_X86_32
626 /*
627 * We check for X86_FEATURE_PCID here. But the init-code will
628 * clear the feature flag on 32 bit because the feature is not
629 * supported on 32 bit anyway. To print the warning we need to
630 * check with cpuid directly again.
631 */
632 if (cpuid_ecx(0x1) & BIT(17)) {
633 /* Use printk to work around pr_fmt() */
634 printk(KERN_WARNING "\n");
635 printk(KERN_WARNING "************************************************************\n");
636 printk(KERN_WARNING "** WARNING! WARNING! WARNING! WARNING! WARNING! WARNING! **\n");
637 printk(KERN_WARNING "** **\n");
638 printk(KERN_WARNING "** You are using 32-bit PTI on a 64-bit PCID-capable CPU. **\n");
639 printk(KERN_WARNING "** Your performance will increase dramatically if you **\n");
640 printk(KERN_WARNING "** switch to a 64-bit kernel! **\n");
641 printk(KERN_WARNING "** **\n");
642 printk(KERN_WARNING "** WARNING! WARNING! WARNING! WARNING! WARNING! WARNING! **\n");
643 printk(KERN_WARNING "************************************************************\n");
644 }
645 #endif
646
647 pti_clone_user_shared();
648
649 /* Undo all global bits from the init pagetables in head_64.S: */
650 pti_set_kernel_image_nonglobal();
651
652 /* Replace some of the global bits just for shared entry text: */
653 /*
654 * This is very early in boot. Device and Late initcalls can do
655 * modprobe before free_initmem() and mark_readonly(). This
656 * pti_clone_entry_text() allows those user-mode-helpers to function,
657 * but notably the text is still RW.
658 */
659 pti_clone_entry_text(false);
660 pti_setup_espfix64();
661 pti_setup_vsyscall();
662 }
663
664 /*
665 * Finalize the kernel mappings in the userspace page-table. Some of the
666 * mappings for the kernel image might have changed since pti_init()
667 * cloned them. This is because parts of the kernel image have been
668 * mapped RO and/or NX. These changes need to be cloned again to the
669 * userspace page-table.
670 */
pti_finalize(void)671 void pti_finalize(void)
672 {
673 if (!boot_cpu_has(X86_FEATURE_PTI))
674 return;
675 /*
676 * This is after free_initmem() (all initcalls are done) and we've done
677 * mark_readonly(). Text is now NX which might've split some PMDs
678 * relative to the early clone.
679 */
680 pti_clone_entry_text(true);
681 pti_clone_kernel_text();
682
683 debug_checkwx_user();
684 }
685